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

TOMOYO Linux Cross Reference
Linux/kernel/exit.c

Version: ~ [ linux-5.3-rc5 ] ~ [ linux-5.2.9 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.67 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.139 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ 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-3.16.72 ] ~ [ 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/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/iocontext.h>
 16 #include <linux/key.h>
 17 #include <linux/security.h>
 18 #include <linux/cpu.h>
 19 #include <linux/acct.h>
 20 #include <linux/tsacct_kern.h>
 21 #include <linux/file.h>
 22 #include <linux/fdtable.h>
 23 #include <linux/freezer.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/cgroup.h>
 36 #include <linux/syscalls.h>
 37 #include <linux/signal.h>
 38 #include <linux/posix-timers.h>
 39 #include <linux/cn_proc.h>
 40 #include <linux/mutex.h>
 41 #include <linux/futex.h>
 42 #include <linux/pipe_fs_i.h>
 43 #include <linux/audit.h> /* for audit_free() */
 44 #include <linux/resource.h>
 45 #include <linux/blkdev.h>
 46 #include <linux/task_io_accounting_ops.h>
 47 #include <linux/tracehook.h>
 48 #include <linux/fs_struct.h>
 49 #include <linux/init_task.h>
 50 #include <linux/perf_event.h>
 51 #include <trace/events/sched.h>
 52 #include <linux/hw_breakpoint.h>
 53 #include <linux/oom.h>
 54 #include <linux/writeback.h>
 55 #include <linux/shm.h>
 56 #include <linux/kcov.h>
 57 
 58 #include <asm/uaccess.h>
 59 #include <asm/unistd.h>
 60 #include <asm/pgtable.h>
 61 #include <asm/mmu_context.h>
 62 
 63 static void __unhash_process(struct task_struct *p, bool group_dead)
 64 {
 65         nr_threads--;
 66         detach_pid(p, PIDTYPE_PID);
 67         if (group_dead) {
 68                 detach_pid(p, PIDTYPE_PGID);
 69                 detach_pid(p, PIDTYPE_SID);
 70 
 71                 list_del_rcu(&p->tasks);
 72                 list_del_init(&p->sibling);
 73                 __this_cpu_dec(process_counts);
 74         }
 75         list_del_rcu(&p->thread_group);
 76         list_del_rcu(&p->thread_node);
 77 }
 78 
 79 /*
 80  * This function expects the tasklist_lock write-locked.
 81  */
 82 static void __exit_signal(struct task_struct *tsk)
 83 {
 84         struct signal_struct *sig = tsk->signal;
 85         bool group_dead = thread_group_leader(tsk);
 86         struct sighand_struct *sighand;
 87         struct tty_struct *uninitialized_var(tty);
 88         cputime_t utime, stime;
 89 
 90         sighand = rcu_dereference_check(tsk->sighand,
 91                                         lockdep_tasklist_lock_is_held());
 92         spin_lock(&sighand->siglock);
 93 
 94         posix_cpu_timers_exit(tsk);
 95         if (group_dead) {
 96                 posix_cpu_timers_exit_group(tsk);
 97                 tty = sig->tty;
 98                 sig->tty = NULL;
 99         } else {
100                 /*
101                  * This can only happen if the caller is de_thread().
102                  * FIXME: this is the temporary hack, we should teach
103                  * posix-cpu-timers to handle this case correctly.
104                  */
105                 if (unlikely(has_group_leader_pid(tsk)))
106                         posix_cpu_timers_exit_group(tsk);
107 
108                 /*
109                  * If there is any task waiting for the group exit
110                  * then notify it:
111                  */
112                 if (sig->notify_count > 0 && !--sig->notify_count)
113                         wake_up_process(sig->group_exit_task);
114 
115                 if (tsk == sig->curr_target)
116                         sig->curr_target = next_thread(tsk);
117         }
118 
119         /*
120          * Accumulate here the counters for all threads as they die. We could
121          * skip the group leader because it is the last user of signal_struct,
122          * but we want to avoid the race with thread_group_cputime() which can
123          * see the empty ->thread_head list.
124          */
125         task_cputime(tsk, &utime, &stime);
126         write_seqlock(&sig->stats_lock);
127         sig->utime += utime;
128         sig->stime += stime;
129         sig->gtime += task_gtime(tsk);
130         sig->min_flt += tsk->min_flt;
131         sig->maj_flt += tsk->maj_flt;
132         sig->nvcsw += tsk->nvcsw;
133         sig->nivcsw += tsk->nivcsw;
134         sig->inblock += task_io_get_inblock(tsk);
135         sig->oublock += task_io_get_oublock(tsk);
136         task_io_accounting_add(&sig->ioac, &tsk->ioac);
137         sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
138         sig->nr_threads--;
139         __unhash_process(tsk, group_dead);
140         write_sequnlock(&sig->stats_lock);
141 
142         /*
143          * Do this under ->siglock, we can race with another thread
144          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
145          */
146         flush_sigqueue(&tsk->pending);
147         tsk->sighand = NULL;
148         spin_unlock(&sighand->siglock);
149 
150         __cleanup_sighand(sighand);
151         clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
152         if (group_dead) {
153                 flush_sigqueue(&sig->shared_pending);
154                 tty_kref_put(tty);
155         }
156 }
157 
158 static void delayed_put_task_struct(struct rcu_head *rhp)
159 {
160         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161 
162         perf_event_delayed_put(tsk);
163         trace_sched_process_free(tsk);
164         put_task_struct(tsk);
165 }
166 
167 
168 void release_task(struct task_struct *p)
169 {
170         struct task_struct *leader;
171         int zap_leader;
172 repeat:
173         /* don't need to get the RCU readlock here - the process is dead and
174          * can't be modifying its own credentials. But shut RCU-lockdep up */
175         rcu_read_lock();
176         atomic_dec(&__task_cred(p)->user->processes);
177         rcu_read_unlock();
178 
179         proc_flush_task(p);
180 
181         write_lock_irq(&tasklist_lock);
182         ptrace_release_task(p);
183         __exit_signal(p);
184 
185         /*
186          * If we are the last non-leader member of the thread
187          * group, and the leader is zombie, then notify the
188          * group leader's parent process. (if it wants notification.)
189          */
190         zap_leader = 0;
191         leader = p->group_leader;
192         if (leader != p && thread_group_empty(leader)
193                         && leader->exit_state == EXIT_ZOMBIE) {
194                 /*
195                  * If we were the last child thread and the leader has
196                  * exited already, and the leader's parent ignores SIGCHLD,
197                  * then we are the one who should release the leader.
198                  */
199                 zap_leader = do_notify_parent(leader, leader->exit_signal);
200                 if (zap_leader)
201                         leader->exit_state = EXIT_DEAD;
202         }
203 
204         write_unlock_irq(&tasklist_lock);
205         release_thread(p);
206         call_rcu(&p->rcu, delayed_put_task_struct);
207 
208         p = leader;
209         if (unlikely(zap_leader))
210                 goto repeat;
211 }
212 
213 /*
214  * Note that if this function returns a valid task_struct pointer (!NULL)
215  * task->usage must remain >0 for the duration of the RCU critical section.
216  */
217 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
218 {
219         struct sighand_struct *sighand;
220         struct task_struct *task;
221 
222         /*
223          * We need to verify that release_task() was not called and thus
224          * delayed_put_task_struct() can't run and drop the last reference
225          * before rcu_read_unlock(). We check task->sighand != NULL,
226          * but we can read the already freed and reused memory.
227          */
228 retry:
229         task = rcu_dereference(*ptask);
230         if (!task)
231                 return NULL;
232 
233         probe_kernel_address(&task->sighand, sighand);
234 
235         /*
236          * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
237          * was already freed we can not miss the preceding update of this
238          * pointer.
239          */
240         smp_rmb();
241         if (unlikely(task != READ_ONCE(*ptask)))
242                 goto retry;
243 
244         /*
245          * We've re-checked that "task == *ptask", now we have two different
246          * cases:
247          *
248          * 1. This is actually the same task/task_struct. In this case
249          *    sighand != NULL tells us it is still alive.
250          *
251          * 2. This is another task which got the same memory for task_struct.
252          *    We can't know this of course, and we can not trust
253          *    sighand != NULL.
254          *
255          *    In this case we actually return a random value, but this is
256          *    correct.
257          *
258          *    If we return NULL - we can pretend that we actually noticed that
259          *    *ptask was updated when the previous task has exited. Or pretend
260          *    that probe_slab_address(&sighand) reads NULL.
261          *
262          *    If we return the new task (because sighand is not NULL for any
263          *    reason) - this is fine too. This (new) task can't go away before
264          *    another gp pass.
265          *
266          *    And note: We could even eliminate the false positive if re-read
267          *    task->sighand once again to avoid the falsely NULL. But this case
268          *    is very unlikely so we don't care.
269          */
270         if (!sighand)
271                 return NULL;
272 
273         return task;
274 }
275 
276 struct task_struct *try_get_task_struct(struct task_struct **ptask)
277 {
278         struct task_struct *task;
279 
280         rcu_read_lock();
281         task = task_rcu_dereference(ptask);
282         if (task)
283                 get_task_struct(task);
284         rcu_read_unlock();
285 
286         return task;
287 }
288 
289 /*
290  * Determine if a process group is "orphaned", according to the POSIX
291  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
292  * by terminal-generated stop signals.  Newly orphaned process groups are
293  * to receive a SIGHUP and a SIGCONT.
294  *
295  * "I ask you, have you ever known what it is to be an orphan?"
296  */
297 static int will_become_orphaned_pgrp(struct pid *pgrp,
298                                         struct task_struct *ignored_task)
299 {
300         struct task_struct *p;
301 
302         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303                 if ((p == ignored_task) ||
304                     (p->exit_state && thread_group_empty(p)) ||
305                     is_global_init(p->real_parent))
306                         continue;
307 
308                 if (task_pgrp(p->real_parent) != pgrp &&
309                     task_session(p->real_parent) == task_session(p))
310                         return 0;
311         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
312 
313         return 1;
314 }
315 
316 int is_current_pgrp_orphaned(void)
317 {
318         int retval;
319 
320         read_lock(&tasklist_lock);
321         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
322         read_unlock(&tasklist_lock);
323 
324         return retval;
325 }
326 
327 static bool has_stopped_jobs(struct pid *pgrp)
328 {
329         struct task_struct *p;
330 
331         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
332                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
333                         return true;
334         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
335 
336         return false;
337 }
338 
339 /*
340  * Check to see if any process groups have become orphaned as
341  * a result of our exiting, and if they have any stopped jobs,
342  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
343  */
344 static void
345 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
346 {
347         struct pid *pgrp = task_pgrp(tsk);
348         struct task_struct *ignored_task = tsk;
349 
350         if (!parent)
351                 /* exit: our father is in a different pgrp than
352                  * we are and we were the only connection outside.
353                  */
354                 parent = tsk->real_parent;
355         else
356                 /* reparent: our child is in a different pgrp than
357                  * we are, and it was the only connection outside.
358                  */
359                 ignored_task = NULL;
360 
361         if (task_pgrp(parent) != pgrp &&
362             task_session(parent) == task_session(tsk) &&
363             will_become_orphaned_pgrp(pgrp, ignored_task) &&
364             has_stopped_jobs(pgrp)) {
365                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
366                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
367         }
368 }
369 
370 #ifdef CONFIG_MEMCG
371 /*
372  * A task is exiting.   If it owned this mm, find a new owner for the mm.
373  */
374 void mm_update_next_owner(struct mm_struct *mm)
375 {
376         struct task_struct *c, *g, *p = current;
377 
378 retry:
379         /*
380          * If the exiting or execing task is not the owner, it's
381          * someone else's problem.
382          */
383         if (mm->owner != p)
384                 return;
385         /*
386          * The current owner is exiting/execing and there are no other
387          * candidates.  Do not leave the mm pointing to a possibly
388          * freed task structure.
389          */
390         if (atomic_read(&mm->mm_users) <= 1) {
391                 mm->owner = NULL;
392                 return;
393         }
394 
395         read_lock(&tasklist_lock);
396         /*
397          * Search in the children
398          */
399         list_for_each_entry(c, &p->children, sibling) {
400                 if (c->mm == mm)
401                         goto assign_new_owner;
402         }
403 
404         /*
405          * Search in the siblings
406          */
407         list_for_each_entry(c, &p->real_parent->children, sibling) {
408                 if (c->mm == mm)
409                         goto assign_new_owner;
410         }
411 
412         /*
413          * Search through everything else, we should not get here often.
414          */
415         for_each_process(g) {
416                 if (g->flags & PF_KTHREAD)
417                         continue;
418                 for_each_thread(g, c) {
419                         if (c->mm == mm)
420                                 goto assign_new_owner;
421                         if (c->mm)
422                                 break;
423                 }
424         }
425         read_unlock(&tasklist_lock);
426         /*
427          * We found no owner yet mm_users > 1: this implies that we are
428          * most likely racing with swapoff (try_to_unuse()) or /proc or
429          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
430          */
431         mm->owner = NULL;
432         return;
433 
434 assign_new_owner:
435         BUG_ON(c == p);
436         get_task_struct(c);
437         /*
438          * The task_lock protects c->mm from changing.
439          * We always want mm->owner->mm == mm
440          */
441         task_lock(c);
442         /*
443          * Delay read_unlock() till we have the task_lock()
444          * to ensure that c does not slip away underneath us
445          */
446         read_unlock(&tasklist_lock);
447         if (c->mm != mm) {
448                 task_unlock(c);
449                 put_task_struct(c);
450                 goto retry;
451         }
452         mm->owner = c;
453         task_unlock(c);
454         put_task_struct(c);
455 }
456 #endif /* CONFIG_MEMCG */
457 
458 /*
459  * Turn us into a lazy TLB process if we
460  * aren't already..
461  */
462 static void exit_mm(struct task_struct *tsk)
463 {
464         struct mm_struct *mm = tsk->mm;
465         struct core_state *core_state;
466 
467         mm_release(tsk, mm);
468         if (!mm)
469                 return;
470         sync_mm_rss(mm);
471         /*
472          * Serialize with any possible pending coredump.
473          * We must hold mmap_sem around checking core_state
474          * and clearing tsk->mm.  The core-inducing thread
475          * will increment ->nr_threads for each thread in the
476          * group with ->mm != NULL.
477          */
478         down_read(&mm->mmap_sem);
479         core_state = mm->core_state;
480         if (core_state) {
481                 struct core_thread self;
482 
483                 up_read(&mm->mmap_sem);
484 
485                 self.task = tsk;
486                 self.next = xchg(&core_state->dumper.next, &self);
487                 /*
488                  * Implies mb(), the result of xchg() must be visible
489                  * to core_state->dumper.
490                  */
491                 if (atomic_dec_and_test(&core_state->nr_threads))
492                         complete(&core_state->startup);
493 
494                 for (;;) {
495                         set_task_state(tsk, TASK_UNINTERRUPTIBLE);
496                         if (!self.task) /* see coredump_finish() */
497                                 break;
498                         freezable_schedule();
499                 }
500                 __set_task_state(tsk, TASK_RUNNING);
501                 down_read(&mm->mmap_sem);
502         }
503         atomic_inc(&mm->mm_count);
504         BUG_ON(mm != tsk->active_mm);
505         /* more a memory barrier than a real lock */
506         task_lock(tsk);
507         tsk->mm = NULL;
508         up_read(&mm->mmap_sem);
509         enter_lazy_tlb(mm, current);
510         task_unlock(tsk);
511         mm_update_next_owner(mm);
512         mmput(mm);
513         if (test_thread_flag(TIF_MEMDIE))
514                 exit_oom_victim(tsk);
515 }
516 
517 static struct task_struct *find_alive_thread(struct task_struct *p)
518 {
519         struct task_struct *t;
520 
521         for_each_thread(p, t) {
522                 if (!(t->flags & PF_EXITING))
523                         return t;
524         }
525         return NULL;
526 }
527 
528 static struct task_struct *find_child_reaper(struct task_struct *father)
529         __releases(&tasklist_lock)
530         __acquires(&tasklist_lock)
531 {
532         struct pid_namespace *pid_ns = task_active_pid_ns(father);
533         struct task_struct *reaper = pid_ns->child_reaper;
534 
535         if (likely(reaper != father))
536                 return reaper;
537 
538         reaper = find_alive_thread(father);
539         if (reaper) {
540                 pid_ns->child_reaper = reaper;
541                 return reaper;
542         }
543 
544         write_unlock_irq(&tasklist_lock);
545         if (unlikely(pid_ns == &init_pid_ns)) {
546                 panic("Attempted to kill init! exitcode=0x%08x\n",
547                         father->signal->group_exit_code ?: father->exit_code);
548         }
549         zap_pid_ns_processes(pid_ns);
550         write_lock_irq(&tasklist_lock);
551 
552         return father;
553 }
554 
555 /*
556  * When we die, we re-parent all our children, and try to:
557  * 1. give them to another thread in our thread group, if such a member exists
558  * 2. give it to the first ancestor process which prctl'd itself as a
559  *    child_subreaper for its children (like a service manager)
560  * 3. give it to the init process (PID 1) in our pid namespace
561  */
562 static struct task_struct *find_new_reaper(struct task_struct *father,
563                                            struct task_struct *child_reaper)
564 {
565         struct task_struct *thread, *reaper;
566 
567         thread = find_alive_thread(father);
568         if (thread)
569                 return thread;
570 
571         if (father->signal->has_child_subreaper) {
572                 /*
573                  * Find the first ->is_child_subreaper ancestor in our pid_ns.
574                  * We start from father to ensure we can not look into another
575                  * namespace, this is safe because all its threads are dead.
576                  */
577                 for (reaper = father;
578                      !same_thread_group(reaper, child_reaper);
579                      reaper = reaper->real_parent) {
580                         /* call_usermodehelper() descendants need this check */
581                         if (reaper == &init_task)
582                                 break;
583                         if (!reaper->signal->is_child_subreaper)
584                                 continue;
585                         thread = find_alive_thread(reaper);
586                         if (thread)
587                                 return thread;
588                 }
589         }
590 
591         return child_reaper;
592 }
593 
594 /*
595 * Any that need to be release_task'd are put on the @dead list.
596  */
597 static void reparent_leader(struct task_struct *father, struct task_struct *p,
598                                 struct list_head *dead)
599 {
600         if (unlikely(p->exit_state == EXIT_DEAD))
601                 return;
602 
603         /* We don't want people slaying init. */
604         p->exit_signal = SIGCHLD;
605 
606         /* If it has exited notify the new parent about this child's death. */
607         if (!p->ptrace &&
608             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
609                 if (do_notify_parent(p, p->exit_signal)) {
610                         p->exit_state = EXIT_DEAD;
611                         list_add(&p->ptrace_entry, dead);
612                 }
613         }
614 
615         kill_orphaned_pgrp(p, father);
616 }
617 
618 /*
619  * This does two things:
620  *
621  * A.  Make init inherit all the child processes
622  * B.  Check to see if any process groups have become orphaned
623  *      as a result of our exiting, and if they have any stopped
624  *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
625  */
626 static void forget_original_parent(struct task_struct *father,
627                                         struct list_head *dead)
628 {
629         struct task_struct *p, *t, *reaper;
630 
631         if (unlikely(!list_empty(&father->ptraced)))
632                 exit_ptrace(father, dead);
633 
634         /* Can drop and reacquire tasklist_lock */
635         reaper = find_child_reaper(father);
636         if (list_empty(&father->children))
637                 return;
638 
639         reaper = find_new_reaper(father, reaper);
640         list_for_each_entry(p, &father->children, sibling) {
641                 for_each_thread(p, t) {
642                         t->real_parent = reaper;
643                         BUG_ON((!t->ptrace) != (t->parent == father));
644                         if (likely(!t->ptrace))
645                                 t->parent = t->real_parent;
646                         if (t->pdeath_signal)
647                                 group_send_sig_info(t->pdeath_signal,
648                                                     SEND_SIG_NOINFO, t);
649                 }
650                 /*
651                  * If this is a threaded reparent there is no need to
652                  * notify anyone anything has happened.
653                  */
654                 if (!same_thread_group(reaper, father))
655                         reparent_leader(father, p, dead);
656         }
657         list_splice_tail_init(&father->children, &reaper->children);
658 }
659 
660 /*
661  * Send signals to all our closest relatives so that they know
662  * to properly mourn us..
663  */
664 static void exit_notify(struct task_struct *tsk, int group_dead)
665 {
666         bool autoreap;
667         struct task_struct *p, *n;
668         LIST_HEAD(dead);
669 
670         write_lock_irq(&tasklist_lock);
671         forget_original_parent(tsk, &dead);
672 
673         if (group_dead)
674                 kill_orphaned_pgrp(tsk->group_leader, NULL);
675 
676         if (unlikely(tsk->ptrace)) {
677                 int sig = thread_group_leader(tsk) &&
678                                 thread_group_empty(tsk) &&
679                                 !ptrace_reparented(tsk) ?
680                         tsk->exit_signal : SIGCHLD;
681                 autoreap = do_notify_parent(tsk, sig);
682         } else if (thread_group_leader(tsk)) {
683                 autoreap = thread_group_empty(tsk) &&
684                         do_notify_parent(tsk, tsk->exit_signal);
685         } else {
686                 autoreap = true;
687         }
688 
689         tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
690         if (tsk->exit_state == EXIT_DEAD)
691                 list_add(&tsk->ptrace_entry, &dead);
692 
693         /* mt-exec, de_thread() is waiting for group leader */
694         if (unlikely(tsk->signal->notify_count < 0))
695                 wake_up_process(tsk->signal->group_exit_task);
696         write_unlock_irq(&tasklist_lock);
697 
698         list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
699                 list_del_init(&p->ptrace_entry);
700                 release_task(p);
701         }
702 }
703 
704 #ifdef CONFIG_DEBUG_STACK_USAGE
705 static void check_stack_usage(void)
706 {
707         static DEFINE_SPINLOCK(low_water_lock);
708         static int lowest_to_date = THREAD_SIZE;
709         unsigned long free;
710 
711         free = stack_not_used(current);
712 
713         if (free >= lowest_to_date)
714                 return;
715 
716         spin_lock(&low_water_lock);
717         if (free < lowest_to_date) {
718                 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
719                         current->comm, task_pid_nr(current), free);
720                 lowest_to_date = free;
721         }
722         spin_unlock(&low_water_lock);
723 }
724 #else
725 static inline void check_stack_usage(void) {}
726 #endif
727 
728 void do_exit(long code)
729 {
730         struct task_struct *tsk = current;
731         int group_dead;
732         TASKS_RCU(int tasks_rcu_i);
733 
734         profile_task_exit(tsk);
735         kcov_task_exit(tsk);
736 
737         WARN_ON(blk_needs_flush_plug(tsk));
738 
739         if (unlikely(in_interrupt()))
740                 panic("Aiee, killing interrupt handler!");
741         if (unlikely(!tsk->pid))
742                 panic("Attempted to kill the idle task!");
743 
744         /*
745          * If do_exit is called because this processes oopsed, it's possible
746          * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
747          * continuing. Amongst other possible reasons, this is to prevent
748          * mm_release()->clear_child_tid() from writing to a user-controlled
749          * kernel address.
750          */
751         set_fs(USER_DS);
752 
753         ptrace_event(PTRACE_EVENT_EXIT, code);
754 
755         validate_creds_for_do_exit(tsk);
756 
757         /*
758          * We're taking recursive faults here in do_exit. Safest is to just
759          * leave this task alone and wait for reboot.
760          */
761         if (unlikely(tsk->flags & PF_EXITING)) {
762                 pr_alert("Fixing recursive fault but reboot is needed!\n");
763                 /*
764                  * We can do this unlocked here. The futex code uses
765                  * this flag just to verify whether the pi state
766                  * cleanup has been done or not. In the worst case it
767                  * loops once more. We pretend that the cleanup was
768                  * done as there is no way to return. Either the
769                  * OWNER_DIED bit is set by now or we push the blocked
770                  * task into the wait for ever nirwana as well.
771                  */
772                 tsk->flags |= PF_EXITPIDONE;
773                 set_current_state(TASK_UNINTERRUPTIBLE);
774                 schedule();
775         }
776 
777         exit_signals(tsk);  /* sets PF_EXITING */
778         /*
779          * Ensure that all new tsk->pi_lock acquisitions must observe
780          * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
781          */
782         smp_mb();
783         /*
784          * Ensure that we must observe the pi_state in exit_mm() ->
785          * mm_release() -> exit_pi_state_list().
786          */
787         raw_spin_unlock_wait(&tsk->pi_lock);
788 
789         if (unlikely(in_atomic())) {
790                 pr_info("note: %s[%d] exited with preempt_count %d\n",
791                         current->comm, task_pid_nr(current),
792                         preempt_count());
793                 preempt_count_set(PREEMPT_ENABLED);
794         }
795 
796         /* sync mm's RSS info before statistics gathering */
797         if (tsk->mm)
798                 sync_mm_rss(tsk->mm);
799         acct_update_integrals(tsk);
800         group_dead = atomic_dec_and_test(&tsk->signal->live);
801         if (group_dead) {
802                 hrtimer_cancel(&tsk->signal->real_timer);
803                 exit_itimers(tsk->signal);
804                 if (tsk->mm)
805                         setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
806         }
807         acct_collect(code, group_dead);
808         if (group_dead)
809                 tty_audit_exit();
810         audit_free(tsk);
811 
812         tsk->exit_code = code;
813         taskstats_exit(tsk, group_dead);
814 
815         exit_mm(tsk);
816 
817         if (group_dead)
818                 acct_process();
819         trace_sched_process_exit(tsk);
820 
821         exit_sem(tsk);
822         exit_shm(tsk);
823         exit_files(tsk);
824         exit_fs(tsk);
825         if (group_dead)
826                 disassociate_ctty(1);
827         exit_task_namespaces(tsk);
828         exit_task_work(tsk);
829         exit_thread(tsk);
830 
831         /*
832          * Flush inherited counters to the parent - before the parent
833          * gets woken up by child-exit notifications.
834          *
835          * because of cgroup mode, must be called before cgroup_exit()
836          */
837         perf_event_exit_task(tsk);
838 
839         cgroup_exit(tsk);
840 
841         /*
842          * FIXME: do that only when needed, using sched_exit tracepoint
843          */
844         flush_ptrace_hw_breakpoint(tsk);
845 
846         TASKS_RCU(preempt_disable());
847         TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
848         TASKS_RCU(preempt_enable());
849         exit_notify(tsk, group_dead);
850         proc_exit_connector(tsk);
851         mpol_put_task_policy(tsk);
852 #ifdef CONFIG_FUTEX
853         if (unlikely(current->pi_state_cache))
854                 kfree(current->pi_state_cache);
855 #endif
856         /*
857          * Make sure we are holding no locks:
858          */
859         debug_check_no_locks_held();
860         /*
861          * We can do this unlocked here. The futex code uses this flag
862          * just to verify whether the pi state cleanup has been done
863          * or not. In the worst case it loops once more.
864          */
865         tsk->flags |= PF_EXITPIDONE;
866 
867         if (tsk->io_context)
868                 exit_io_context(tsk);
869 
870         if (tsk->splice_pipe)
871                 free_pipe_info(tsk->splice_pipe);
872 
873         if (tsk->task_frag.page)
874                 put_page(tsk->task_frag.page);
875 
876         validate_creds_for_do_exit(tsk);
877 
878         check_stack_usage();
879         preempt_disable();
880         if (tsk->nr_dirtied)
881                 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
882         exit_rcu();
883         TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
884 
885         /*
886          * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
887          * when the following two conditions become true.
888          *   - There is race condition of mmap_sem (It is acquired by
889          *     exit_mm()), and
890          *   - SMI occurs before setting TASK_RUNINNG.
891          *     (or hypervisor of virtual machine switches to other guest)
892          *  As a result, we may become TASK_RUNNING after becoming TASK_DEAD
893          *
894          * To avoid it, we have to wait for releasing tsk->pi_lock which
895          * is held by try_to_wake_up()
896          */
897         smp_mb();
898         raw_spin_unlock_wait(&tsk->pi_lock);
899 
900         /* causes final put_task_struct in finish_task_switch(). */
901         tsk->state = TASK_DEAD;
902         tsk->flags |= PF_NOFREEZE;      /* tell freezer to ignore us */
903         schedule();
904         BUG();
905         /* Avoid "noreturn function does return".  */
906         for (;;)
907                 cpu_relax();    /* For when BUG is null */
908 }
909 EXPORT_SYMBOL_GPL(do_exit);
910 
911 void complete_and_exit(struct completion *comp, long code)
912 {
913         if (comp)
914                 complete(comp);
915 
916         do_exit(code);
917 }
918 EXPORT_SYMBOL(complete_and_exit);
919 
920 SYSCALL_DEFINE1(exit, int, error_code)
921 {
922         do_exit((error_code&0xff)<<8);
923 }
924 
925 /*
926  * Take down every thread in the group.  This is called by fatal signals
927  * as well as by sys_exit_group (below).
928  */
929 void
930 do_group_exit(int exit_code)
931 {
932         struct signal_struct *sig = current->signal;
933 
934         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
935 
936         if (signal_group_exit(sig))
937                 exit_code = sig->group_exit_code;
938         else if (!thread_group_empty(current)) {
939                 struct sighand_struct *const sighand = current->sighand;
940 
941                 spin_lock_irq(&sighand->siglock);
942                 if (signal_group_exit(sig))
943                         /* Another thread got here before we took the lock.  */
944                         exit_code = sig->group_exit_code;
945                 else {
946                         sig->group_exit_code = exit_code;
947                         sig->flags = SIGNAL_GROUP_EXIT;
948                         zap_other_threads(current);
949                 }
950                 spin_unlock_irq(&sighand->siglock);
951         }
952 
953         do_exit(exit_code);
954         /* NOTREACHED */
955 }
956 
957 /*
958  * this kills every thread in the thread group. Note that any externally
959  * wait4()-ing process will get the correct exit code - even if this
960  * thread is not the thread group leader.
961  */
962 SYSCALL_DEFINE1(exit_group, int, error_code)
963 {
964         do_group_exit((error_code & 0xff) << 8);
965         /* NOTREACHED */
966         return 0;
967 }
968 
969 struct wait_opts {
970         enum pid_type           wo_type;
971         int                     wo_flags;
972         struct pid              *wo_pid;
973 
974         struct siginfo __user   *wo_info;
975         int __user              *wo_stat;
976         struct rusage __user    *wo_rusage;
977 
978         wait_queue_t            child_wait;
979         int                     notask_error;
980 };
981 
982 static inline
983 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
984 {
985         if (type != PIDTYPE_PID)
986                 task = task->group_leader;
987         return task->pids[type].pid;
988 }
989 
990 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
991 {
992         return  wo->wo_type == PIDTYPE_MAX ||
993                 task_pid_type(p, wo->wo_type) == wo->wo_pid;
994 }
995 
996 static int
997 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
998 {
999         if (!eligible_pid(wo, p))
1000                 return 0;
1001 
1002         /*
1003          * Wait for all children (clone and not) if __WALL is set or
1004          * if it is traced by us.
1005          */
1006         if (ptrace || (wo->wo_flags & __WALL))
1007                 return 1;
1008 
1009         /*
1010          * Otherwise, wait for clone children *only* if __WCLONE is set;
1011          * otherwise, wait for non-clone children *only*.
1012          *
1013          * Note: a "clone" child here is one that reports to its parent
1014          * using a signal other than SIGCHLD, or a non-leader thread which
1015          * we can only see if it is traced by us.
1016          */
1017         if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1018                 return 0;
1019 
1020         return 1;
1021 }
1022 
1023 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1024                                 pid_t pid, uid_t uid, int why, int status)
1025 {
1026         struct siginfo __user *infop;
1027         int retval = wo->wo_rusage
1028                 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1029 
1030         put_task_struct(p);
1031         infop = wo->wo_info;
1032         if (infop) {
1033                 if (!retval)
1034                         retval = put_user(SIGCHLD, &infop->si_signo);
1035                 if (!retval)
1036                         retval = put_user(0, &infop->si_errno);
1037                 if (!retval)
1038                         retval = put_user((short)why, &infop->si_code);
1039                 if (!retval)
1040                         retval = put_user(pid, &infop->si_pid);
1041                 if (!retval)
1042                         retval = put_user(uid, &infop->si_uid);
1043                 if (!retval)
1044                         retval = put_user(status, &infop->si_status);
1045         }
1046         if (!retval)
1047                 retval = pid;
1048         return retval;
1049 }
1050 
1051 /*
1052  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1053  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1054  * the lock and this task is uninteresting.  If we return nonzero, we have
1055  * released the lock and the system call should return.
1056  */
1057 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1058 {
1059         int state, retval, status;
1060         pid_t pid = task_pid_vnr(p);
1061         uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1062         struct siginfo __user *infop;
1063 
1064         if (!likely(wo->wo_flags & WEXITED))
1065                 return 0;
1066 
1067         if (unlikely(wo->wo_flags & WNOWAIT)) {
1068                 int exit_code = p->exit_code;
1069                 int why;
1070 
1071                 get_task_struct(p);
1072                 read_unlock(&tasklist_lock);
1073                 sched_annotate_sleep();
1074 
1075                 if ((exit_code & 0x7f) == 0) {
1076                         why = CLD_EXITED;
1077                         status = exit_code >> 8;
1078                 } else {
1079                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1080                         status = exit_code & 0x7f;
1081                 }
1082                 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1083         }
1084         /*
1085          * Move the task's state to DEAD/TRACE, only one thread can do this.
1086          */
1087         state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1088                 EXIT_TRACE : EXIT_DEAD;
1089         if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1090                 return 0;
1091         /*
1092          * We own this thread, nobody else can reap it.
1093          */
1094         read_unlock(&tasklist_lock);
1095         sched_annotate_sleep();
1096 
1097         /*
1098          * Check thread_group_leader() to exclude the traced sub-threads.
1099          */
1100         if (state == EXIT_DEAD && thread_group_leader(p)) {
1101                 struct signal_struct *sig = p->signal;
1102                 struct signal_struct *psig = current->signal;
1103                 unsigned long maxrss;
1104                 cputime_t tgutime, tgstime;
1105 
1106                 /*
1107                  * The resource counters for the group leader are in its
1108                  * own task_struct.  Those for dead threads in the group
1109                  * are in its signal_struct, as are those for the child
1110                  * processes it has previously reaped.  All these
1111                  * accumulate in the parent's signal_struct c* fields.
1112                  *
1113                  * We don't bother to take a lock here to protect these
1114                  * p->signal fields because the whole thread group is dead
1115                  * and nobody can change them.
1116                  *
1117                  * psig->stats_lock also protects us from our sub-theads
1118                  * which can reap other children at the same time. Until
1119                  * we change k_getrusage()-like users to rely on this lock
1120                  * we have to take ->siglock as well.
1121                  *
1122                  * We use thread_group_cputime_adjusted() to get times for
1123                  * the thread group, which consolidates times for all threads
1124                  * in the group including the group leader.
1125                  */
1126                 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1127                 spin_lock_irq(&current->sighand->siglock);
1128                 write_seqlock(&psig->stats_lock);
1129                 psig->cutime += tgutime + sig->cutime;
1130                 psig->cstime += tgstime + sig->cstime;
1131                 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1132                 psig->cmin_flt +=
1133                         p->min_flt + sig->min_flt + sig->cmin_flt;
1134                 psig->cmaj_flt +=
1135                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1136                 psig->cnvcsw +=
1137                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1138                 psig->cnivcsw +=
1139                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1140                 psig->cinblock +=
1141                         task_io_get_inblock(p) +
1142                         sig->inblock + sig->cinblock;
1143                 psig->coublock +=
1144                         task_io_get_oublock(p) +
1145                         sig->oublock + sig->coublock;
1146                 maxrss = max(sig->maxrss, sig->cmaxrss);
1147                 if (psig->cmaxrss < maxrss)
1148                         psig->cmaxrss = maxrss;
1149                 task_io_accounting_add(&psig->ioac, &p->ioac);
1150                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1151                 write_sequnlock(&psig->stats_lock);
1152                 spin_unlock_irq(&current->sighand->siglock);
1153         }
1154 
1155         retval = wo->wo_rusage
1156                 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1157         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1158                 ? p->signal->group_exit_code : p->exit_code;
1159         if (!retval && wo->wo_stat)
1160                 retval = put_user(status, wo->wo_stat);
1161 
1162         infop = wo->wo_info;
1163         if (!retval && infop)
1164                 retval = put_user(SIGCHLD, &infop->si_signo);
1165         if (!retval && infop)
1166                 retval = put_user(0, &infop->si_errno);
1167         if (!retval && infop) {
1168                 int why;
1169 
1170                 if ((status & 0x7f) == 0) {
1171                         why = CLD_EXITED;
1172                         status >>= 8;
1173                 } else {
1174                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1175                         status &= 0x7f;
1176                 }
1177                 retval = put_user((short)why, &infop->si_code);
1178                 if (!retval)
1179                         retval = put_user(status, &infop->si_status);
1180         }
1181         if (!retval && infop)
1182                 retval = put_user(pid, &infop->si_pid);
1183         if (!retval && infop)
1184                 retval = put_user(uid, &infop->si_uid);
1185         if (!retval)
1186                 retval = pid;
1187 
1188         if (state == EXIT_TRACE) {
1189                 write_lock_irq(&tasklist_lock);
1190                 /* We dropped tasklist, ptracer could die and untrace */
1191                 ptrace_unlink(p);
1192 
1193                 /* If parent wants a zombie, don't release it now */
1194                 state = EXIT_ZOMBIE;
1195                 if (do_notify_parent(p, p->exit_signal))
1196                         state = EXIT_DEAD;
1197                 p->exit_state = state;
1198                 write_unlock_irq(&tasklist_lock);
1199         }
1200         if (state == EXIT_DEAD)
1201                 release_task(p);
1202 
1203         return retval;
1204 }
1205 
1206 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1207 {
1208         if (ptrace) {
1209                 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1210                         return &p->exit_code;
1211         } else {
1212                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1213                         return &p->signal->group_exit_code;
1214         }
1215         return NULL;
1216 }
1217 
1218 /**
1219  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1220  * @wo: wait options
1221  * @ptrace: is the wait for ptrace
1222  * @p: task to wait for
1223  *
1224  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1225  *
1226  * CONTEXT:
1227  * read_lock(&tasklist_lock), which is released if return value is
1228  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1229  *
1230  * RETURNS:
1231  * 0 if wait condition didn't exist and search for other wait conditions
1232  * should continue.  Non-zero return, -errno on failure and @p's pid on
1233  * success, implies that tasklist_lock is released and wait condition
1234  * search should terminate.
1235  */
1236 static int wait_task_stopped(struct wait_opts *wo,
1237                                 int ptrace, struct task_struct *p)
1238 {
1239         struct siginfo __user *infop;
1240         int retval, exit_code, *p_code, why;
1241         uid_t uid = 0; /* unneeded, required by compiler */
1242         pid_t pid;
1243 
1244         /*
1245          * Traditionally we see ptrace'd stopped tasks regardless of options.
1246          */
1247         if (!ptrace && !(wo->wo_flags & WUNTRACED))
1248                 return 0;
1249 
1250         if (!task_stopped_code(p, ptrace))
1251                 return 0;
1252 
1253         exit_code = 0;
1254         spin_lock_irq(&p->sighand->siglock);
1255 
1256         p_code = task_stopped_code(p, ptrace);
1257         if (unlikely(!p_code))
1258                 goto unlock_sig;
1259 
1260         exit_code = *p_code;
1261         if (!exit_code)
1262                 goto unlock_sig;
1263 
1264         if (!unlikely(wo->wo_flags & WNOWAIT))
1265                 *p_code = 0;
1266 
1267         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1268 unlock_sig:
1269         spin_unlock_irq(&p->sighand->siglock);
1270         if (!exit_code)
1271                 return 0;
1272 
1273         /*
1274          * Now we are pretty sure this task is interesting.
1275          * Make sure it doesn't get reaped out from under us while we
1276          * give up the lock and then examine it below.  We don't want to
1277          * keep holding onto the tasklist_lock while we call getrusage and
1278          * possibly take page faults for user memory.
1279          */
1280         get_task_struct(p);
1281         pid = task_pid_vnr(p);
1282         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1283         read_unlock(&tasklist_lock);
1284         sched_annotate_sleep();
1285 
1286         if (unlikely(wo->wo_flags & WNOWAIT))
1287                 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1288 
1289         retval = wo->wo_rusage
1290                 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1291         if (!retval && wo->wo_stat)
1292                 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1293 
1294         infop = wo->wo_info;
1295         if (!retval && infop)
1296                 retval = put_user(SIGCHLD, &infop->si_signo);
1297         if (!retval && infop)
1298                 retval = put_user(0, &infop->si_errno);
1299         if (!retval && infop)
1300                 retval = put_user((short)why, &infop->si_code);
1301         if (!retval && infop)
1302                 retval = put_user(exit_code, &infop->si_status);
1303         if (!retval && infop)
1304                 retval = put_user(pid, &infop->si_pid);
1305         if (!retval && infop)
1306                 retval = put_user(uid, &infop->si_uid);
1307         if (!retval)
1308                 retval = pid;
1309         put_task_struct(p);
1310 
1311         BUG_ON(!retval);
1312         return retval;
1313 }
1314 
1315 /*
1316  * Handle do_wait work for one task in a live, non-stopped state.
1317  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1318  * the lock and this task is uninteresting.  If we return nonzero, we have
1319  * released the lock and the system call should return.
1320  */
1321 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1322 {
1323         int retval;
1324         pid_t pid;
1325         uid_t uid;
1326 
1327         if (!unlikely(wo->wo_flags & WCONTINUED))
1328                 return 0;
1329 
1330         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1331                 return 0;
1332 
1333         spin_lock_irq(&p->sighand->siglock);
1334         /* Re-check with the lock held.  */
1335         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1336                 spin_unlock_irq(&p->sighand->siglock);
1337                 return 0;
1338         }
1339         if (!unlikely(wo->wo_flags & WNOWAIT))
1340                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1341         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1342         spin_unlock_irq(&p->sighand->siglock);
1343 
1344         pid = task_pid_vnr(p);
1345         get_task_struct(p);
1346         read_unlock(&tasklist_lock);
1347         sched_annotate_sleep();
1348 
1349         if (!wo->wo_info) {
1350                 retval = wo->wo_rusage
1351                         ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1352                 put_task_struct(p);
1353                 if (!retval && wo->wo_stat)
1354                         retval = put_user(0xffff, wo->wo_stat);
1355                 if (!retval)
1356                         retval = pid;
1357         } else {
1358                 retval = wait_noreap_copyout(wo, p, pid, uid,
1359                                              CLD_CONTINUED, SIGCONT);
1360                 BUG_ON(retval == 0);
1361         }
1362 
1363         return retval;
1364 }
1365 
1366 /*
1367  * Consider @p for a wait by @parent.
1368  *
1369  * -ECHILD should be in ->notask_error before the first call.
1370  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1371  * Returns zero if the search for a child should continue;
1372  * then ->notask_error is 0 if @p is an eligible child,
1373  * or another error from security_task_wait(), or still -ECHILD.
1374  */
1375 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1376                                 struct task_struct *p)
1377 {
1378         /*
1379          * We can race with wait_task_zombie() from another thread.
1380          * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1381          * can't confuse the checks below.
1382          */
1383         int exit_state = ACCESS_ONCE(p->exit_state);
1384         int ret;
1385 
1386         if (unlikely(exit_state == EXIT_DEAD))
1387                 return 0;
1388 
1389         ret = eligible_child(wo, ptrace, p);
1390         if (!ret)
1391                 return ret;
1392 
1393         ret = security_task_wait(p);
1394         if (unlikely(ret < 0)) {
1395                 /*
1396                  * If we have not yet seen any eligible child,
1397                  * then let this error code replace -ECHILD.
1398                  * A permission error will give the user a clue
1399                  * to look for security policy problems, rather
1400                  * than for mysterious wait bugs.
1401                  */
1402                 if (wo->notask_error)
1403                         wo->notask_error = ret;
1404                 return 0;
1405         }
1406 
1407         if (unlikely(exit_state == EXIT_TRACE)) {
1408                 /*
1409                  * ptrace == 0 means we are the natural parent. In this case
1410                  * we should clear notask_error, debugger will notify us.
1411                  */
1412                 if (likely(!ptrace))
1413                         wo->notask_error = 0;
1414                 return 0;
1415         }
1416 
1417         if (likely(!ptrace) && unlikely(p->ptrace)) {
1418                 /*
1419                  * If it is traced by its real parent's group, just pretend
1420                  * the caller is ptrace_do_wait() and reap this child if it
1421                  * is zombie.
1422                  *
1423                  * This also hides group stop state from real parent; otherwise
1424                  * a single stop can be reported twice as group and ptrace stop.
1425                  * If a ptracer wants to distinguish these two events for its
1426                  * own children it should create a separate process which takes
1427                  * the role of real parent.
1428                  */
1429                 if (!ptrace_reparented(p))
1430                         ptrace = 1;
1431         }
1432 
1433         /* slay zombie? */
1434         if (exit_state == EXIT_ZOMBIE) {
1435                 /* we don't reap group leaders with subthreads */
1436                 if (!delay_group_leader(p)) {
1437                         /*
1438                          * A zombie ptracee is only visible to its ptracer.
1439                          * Notification and reaping will be cascaded to the
1440                          * real parent when the ptracer detaches.
1441                          */
1442                         if (unlikely(ptrace) || likely(!p->ptrace))
1443                                 return wait_task_zombie(wo, p);
1444                 }
1445 
1446                 /*
1447                  * Allow access to stopped/continued state via zombie by
1448                  * falling through.  Clearing of notask_error is complex.
1449                  *
1450                  * When !@ptrace:
1451                  *
1452                  * If WEXITED is set, notask_error should naturally be
1453                  * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1454                  * so, if there are live subthreads, there are events to
1455                  * wait for.  If all subthreads are dead, it's still safe
1456                  * to clear - this function will be called again in finite
1457                  * amount time once all the subthreads are released and
1458                  * will then return without clearing.
1459                  *
1460                  * When @ptrace:
1461                  *
1462                  * Stopped state is per-task and thus can't change once the
1463                  * target task dies.  Only continued and exited can happen.
1464                  * Clear notask_error if WCONTINUED | WEXITED.
1465                  */
1466                 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1467                         wo->notask_error = 0;
1468         } else {
1469                 /*
1470                  * @p is alive and it's gonna stop, continue or exit, so
1471                  * there always is something to wait for.
1472                  */
1473                 wo->notask_error = 0;
1474         }
1475 
1476         /*
1477          * Wait for stopped.  Depending on @ptrace, different stopped state
1478          * is used and the two don't interact with each other.
1479          */
1480         ret = wait_task_stopped(wo, ptrace, p);
1481         if (ret)
1482                 return ret;
1483 
1484         /*
1485          * Wait for continued.  There's only one continued state and the
1486          * ptracer can consume it which can confuse the real parent.  Don't
1487          * use WCONTINUED from ptracer.  You don't need or want it.
1488          */
1489         return wait_task_continued(wo, p);
1490 }
1491 
1492 /*
1493  * Do the work of do_wait() for one thread in the group, @tsk.
1494  *
1495  * -ECHILD should be in ->notask_error before the first call.
1496  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1497  * Returns zero if the search for a child should continue; then
1498  * ->notask_error is 0 if there were any eligible children,
1499  * or another error from security_task_wait(), or still -ECHILD.
1500  */
1501 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1502 {
1503         struct task_struct *p;
1504 
1505         list_for_each_entry(p, &tsk->children, sibling) {
1506                 int ret = wait_consider_task(wo, 0, p);
1507 
1508                 if (ret)
1509                         return ret;
1510         }
1511 
1512         return 0;
1513 }
1514 
1515 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1516 {
1517         struct task_struct *p;
1518 
1519         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1520                 int ret = wait_consider_task(wo, 1, p);
1521 
1522                 if (ret)
1523                         return ret;
1524         }
1525 
1526         return 0;
1527 }
1528 
1529 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1530                                 int sync, void *key)
1531 {
1532         struct wait_opts *wo = container_of(wait, struct wait_opts,
1533                                                 child_wait);
1534         struct task_struct *p = key;
1535 
1536         if (!eligible_pid(wo, p))
1537                 return 0;
1538 
1539         if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1540                 return 0;
1541 
1542         return default_wake_function(wait, mode, sync, key);
1543 }
1544 
1545 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1546 {
1547         __wake_up_sync_key(&parent->signal->wait_chldexit,
1548                                 TASK_INTERRUPTIBLE, 1, p);
1549 }
1550 
1551 static long do_wait(struct wait_opts *wo)
1552 {
1553         struct task_struct *tsk;
1554         int retval;
1555 
1556         trace_sched_process_wait(wo->wo_pid);
1557 
1558         init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1559         wo->child_wait.private = current;
1560         add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1561 repeat:
1562         /*
1563          * If there is nothing that can match our criteria, just get out.
1564          * We will clear ->notask_error to zero if we see any child that
1565          * might later match our criteria, even if we are not able to reap
1566          * it yet.
1567          */
1568         wo->notask_error = -ECHILD;
1569         if ((wo->wo_type < PIDTYPE_MAX) &&
1570            (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1571                 goto notask;
1572 
1573         set_current_state(TASK_INTERRUPTIBLE);
1574         read_lock(&tasklist_lock);
1575         tsk = current;
1576         do {
1577                 retval = do_wait_thread(wo, tsk);
1578                 if (retval)
1579                         goto end;
1580 
1581                 retval = ptrace_do_wait(wo, tsk);
1582                 if (retval)
1583                         goto end;
1584 
1585                 if (wo->wo_flags & __WNOTHREAD)
1586                         break;
1587         } while_each_thread(current, tsk);
1588         read_unlock(&tasklist_lock);
1589 
1590 notask:
1591         retval = wo->notask_error;
1592         if (!retval && !(wo->wo_flags & WNOHANG)) {
1593                 retval = -ERESTARTSYS;
1594                 if (!signal_pending(current)) {
1595                         schedule();
1596                         goto repeat;
1597                 }
1598         }
1599 end:
1600         __set_current_state(TASK_RUNNING);
1601         remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1602         return retval;
1603 }
1604 
1605 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1606                 infop, int, options, struct rusage __user *, ru)
1607 {
1608         struct wait_opts wo;
1609         struct pid *pid = NULL;
1610         enum pid_type type;
1611         long ret;
1612 
1613         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1614                         __WNOTHREAD|__WCLONE|__WALL))
1615                 return -EINVAL;
1616         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1617                 return -EINVAL;
1618 
1619         switch (which) {
1620         case P_ALL:
1621                 type = PIDTYPE_MAX;
1622                 break;
1623         case P_PID:
1624                 type = PIDTYPE_PID;
1625                 if (upid <= 0)
1626                         return -EINVAL;
1627                 break;
1628         case P_PGID:
1629                 type = PIDTYPE_PGID;
1630                 if (upid <= 0)
1631                         return -EINVAL;
1632                 break;
1633         default:
1634                 return -EINVAL;
1635         }
1636 
1637         if (type < PIDTYPE_MAX)
1638                 pid = find_get_pid(upid);
1639 
1640         wo.wo_type      = type;
1641         wo.wo_pid       = pid;
1642         wo.wo_flags     = options;
1643         wo.wo_info      = infop;
1644         wo.wo_stat      = NULL;
1645         wo.wo_rusage    = ru;
1646         ret = do_wait(&wo);
1647 
1648         if (ret > 0) {
1649                 ret = 0;
1650         } else if (infop) {
1651                 /*
1652                  * For a WNOHANG return, clear out all the fields
1653                  * we would set so the user can easily tell the
1654                  * difference.
1655                  */
1656                 if (!ret)
1657                         ret = put_user(0, &infop->si_signo);
1658                 if (!ret)
1659                         ret = put_user(0, &infop->si_errno);
1660                 if (!ret)
1661                         ret = put_user(0, &infop->si_code);
1662                 if (!ret)
1663                         ret = put_user(0, &infop->si_pid);
1664                 if (!ret)
1665                         ret = put_user(0, &infop->si_uid);
1666                 if (!ret)
1667                         ret = put_user(0, &infop->si_status);
1668         }
1669 
1670         put_pid(pid);
1671         return ret;
1672 }
1673 
1674 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1675                 int, options, struct rusage __user *, ru)
1676 {
1677         struct wait_opts wo;
1678         struct pid *pid = NULL;
1679         enum pid_type type;
1680         long ret;
1681 
1682         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1683                         __WNOTHREAD|__WCLONE|__WALL))
1684                 return -EINVAL;
1685 
1686         if (upid == -1)
1687                 type = PIDTYPE_MAX;
1688         else if (upid < 0) {
1689                 type = PIDTYPE_PGID;
1690                 pid = find_get_pid(-upid);
1691         } else if (upid == 0) {
1692                 type = PIDTYPE_PGID;
1693                 pid = get_task_pid(current, PIDTYPE_PGID);
1694         } else /* upid > 0 */ {
1695                 type = PIDTYPE_PID;
1696                 pid = find_get_pid(upid);
1697         }
1698 
1699         wo.wo_type      = type;
1700         wo.wo_pid       = pid;
1701         wo.wo_flags     = options | WEXITED;
1702         wo.wo_info      = NULL;
1703         wo.wo_stat      = stat_addr;
1704         wo.wo_rusage    = ru;
1705         ret = do_wait(&wo);
1706         put_pid(pid);
1707 
1708         return ret;
1709 }
1710 
1711 #ifdef __ARCH_WANT_SYS_WAITPID
1712 
1713 /*
1714  * sys_waitpid() remains for compatibility. waitpid() should be
1715  * implemented by calling sys_wait4() from libc.a.
1716  */
1717 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1718 {
1719         return sys_wait4(pid, stat_addr, options, NULL);
1720 }
1721 
1722 #endif
1723 

~ [ 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