projects/linux·blob v5.0-rc3 · refs

1759 lines · 44.2 KiB

1/*
2 * linux/kernel/exit.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/mm.h>
8#include <linux/slab.h>
9#include <linux/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/init_task.h>
55#include <linux/perf_event.h>
56#include <trace/events/sched.h>
57#include <linux/hw_breakpoint.h>
58#include <linux/oom.h>
59#include <linux/writeback.h>
60#include <linux/shm.h>
61#include <linux/kcov.h>
62#include <linux/random.h>
63#include <linux/rcuwait.h>
64#include <linux/compat.h>
65
66#include <linux/uaccess.h>
67#include <asm/unistd.h>
68#include <asm/pgtable.h>
69#include <asm/mmu_context.h>
70
71static 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_TGID);
77 detach_pid(p, PIDTYPE_PGID);
78 detach_pid(p, PIDTYPE_SID);
79
80 list_del_rcu(&p->tasks);
81 list_del_init(&p->sibling);
82 __this_cpu_dec(process_counts);
83 }
84 list_del_rcu(&p->thread_group);
85 list_del_rcu(&p->thread_node);
86}
87
88/*
89 * This function expects the tasklist_lock write-locked.
90 */
91static void __exit_signal(struct task_struct *tsk)
92{
93 struct signal_struct *sig = tsk->signal;
94 bool group_dead = thread_group_leader(tsk);
95 struct sighand_struct *sighand;
96 struct tty_struct *uninitialized_var(tty);
97 u64 utime, stime;
98
99 sighand = rcu_dereference_check(tsk->sighand,
100 lockdep_tasklist_lock_is_held());
101 spin_lock(&sighand->siglock);
102
103#ifdef CONFIG_POSIX_TIMERS
104 posix_cpu_timers_exit(tsk);
105 if (group_dead) {
106 posix_cpu_timers_exit_group(tsk);
107 } else {
108 /*
109 * This can only happen if the caller is de_thread().
110 * FIXME: this is the temporary hack, we should teach
111 * posix-cpu-timers to handle this case correctly.
112 */
113 if (unlikely(has_group_leader_pid(tsk)))
114 posix_cpu_timers_exit_group(tsk);
115 }
116#endif
117
118 if (group_dead) {
119 tty = sig->tty;
120 sig->tty = NULL;
121 } else {
122 /*
123 * If there is any task waiting for the group exit
124 * then notify it:
125 */
126 if (sig->notify_count > 0 && !--sig->notify_count)
127 wake_up_process(sig->group_exit_task);
128
129 if (tsk == sig->curr_target)
130 sig->curr_target = next_thread(tsk);
131 }
132
133 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
134 sizeof(unsigned long long));
135
136 /*
137 * Accumulate here the counters for all threads as they die. We could
138 * skip the group leader because it is the last user of signal_struct,
139 * but we want to avoid the race with thread_group_cputime() which can
140 * see the empty ->thread_head list.
141 */
142 task_cputime(tsk, &utime, &stime);
143 write_seqlock(&sig->stats_lock);
144 sig->utime += utime;
145 sig->stime += stime;
146 sig->gtime += task_gtime(tsk);
147 sig->min_flt += tsk->min_flt;
148 sig->maj_flt += tsk->maj_flt;
149 sig->nvcsw += tsk->nvcsw;
150 sig->nivcsw += tsk->nivcsw;
151 sig->inblock += task_io_get_inblock(tsk);
152 sig->oublock += task_io_get_oublock(tsk);
153 task_io_accounting_add(&sig->ioac, &tsk->ioac);
154 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
155 sig->nr_threads--;
156 __unhash_process(tsk, group_dead);
157 write_sequnlock(&sig->stats_lock);
158
159 /*
160 * Do this under ->siglock, we can race with another thread
161 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
162 */
163 flush_sigqueue(&tsk->pending);
164 tsk->sighand = NULL;
165 spin_unlock(&sighand->siglock);
166
167 __cleanup_sighand(sighand);
168 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
169 if (group_dead) {
170 flush_sigqueue(&sig->shared_pending);
171 tty_kref_put(tty);
172 }
173}
174
175static void delayed_put_task_struct(struct rcu_head *rhp)
176{
177 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
178
179 perf_event_delayed_put(tsk);
180 trace_sched_process_free(tsk);
181 put_task_struct(tsk);
182}
183
184
185void release_task(struct task_struct *p)
186{
187 struct task_struct *leader;
188 int zap_leader;
189repeat:
190 /* don't need to get the RCU readlock here - the process is dead and
191 * can't be modifying its own credentials. But shut RCU-lockdep up */
192 rcu_read_lock();
193 atomic_dec(&__task_cred(p)->user->processes);
194 rcu_read_unlock();
195
196 proc_flush_task(p);
197
198 write_lock_irq(&tasklist_lock);
199 ptrace_release_task(p);
200 __exit_signal(p);
201
202 /*
203 * If we are the last non-leader member of the thread
204 * group, and the leader is zombie, then notify the
205 * group leader's parent process. (if it wants notification.)
206 */
207 zap_leader = 0;
208 leader = p->group_leader;
209 if (leader != p && thread_group_empty(leader)
210 && leader->exit_state == EXIT_ZOMBIE) {
211 /*
212 * If we were the last child thread and the leader has
213 * exited already, and the leader's parent ignores SIGCHLD,
214 * then we are the one who should release the leader.
215 */
216 zap_leader = do_notify_parent(leader, leader->exit_signal);
217 if (zap_leader)
218 leader->exit_state = EXIT_DEAD;
219 }
220
221 write_unlock_irq(&tasklist_lock);
222 release_thread(p);
223 call_rcu(&p->rcu, delayed_put_task_struct);
224
225 p = leader;
226 if (unlikely(zap_leader))
227 goto repeat;
228}
229
230/*
231 * Note that if this function returns a valid task_struct pointer (!NULL)
232 * task->usage must remain >0 for the duration of the RCU critical section.
233 */
234struct task_struct *task_rcu_dereference(struct task_struct **ptask)
235{
236 struct sighand_struct *sighand;
237 struct task_struct *task;
238
239 /*
240 * We need to verify that release_task() was not called and thus
241 * delayed_put_task_struct() can't run and drop the last reference
242 * before rcu_read_unlock(). We check task->sighand != NULL,
243 * but we can read the already freed and reused memory.
244 */
245retry:
246 task = rcu_dereference(*ptask);
247 if (!task)
248 return NULL;
249
250 probe_kernel_address(&task->sighand, sighand);
251
252 /*
253 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
254 * was already freed we can not miss the preceding update of this
255 * pointer.
256 */
257 smp_rmb();
258 if (unlikely(task != READ_ONCE(*ptask)))
259 goto retry;
260
261 /*
262 * We've re-checked that "task == *ptask", now we have two different
263 * cases:
264 *
265 * 1. This is actually the same task/task_struct. In this case
266 * sighand != NULL tells us it is still alive.
267 *
268 * 2. This is another task which got the same memory for task_struct.
269 * We can't know this of course, and we can not trust
270 * sighand != NULL.
271 *
272 * In this case we actually return a random value, but this is
273 * correct.
274 *
275 * If we return NULL - we can pretend that we actually noticed that
276 * *ptask was updated when the previous task has exited. Or pretend
277 * that probe_slab_address(&sighand) reads NULL.
278 *
279 * If we return the new task (because sighand is not NULL for any
280 * reason) - this is fine too. This (new) task can't go away before
281 * another gp pass.
282 *
283 * And note: We could even eliminate the false positive if re-read
284 * task->sighand once again to avoid the falsely NULL. But this case
285 * is very unlikely so we don't care.
286 */
287 if (!sighand)
288 return NULL;
289
290 return task;
291}
292
293void rcuwait_wake_up(struct rcuwait *w)
294{
295 struct task_struct *task;
296
297 rcu_read_lock();
298
299 /*
300 * Order condition vs @task, such that everything prior to the load
301 * of @task is visible. This is the condition as to why the user called
302 * rcuwait_trywake() in the first place. Pairs with set_current_state()
303 * barrier (A) in rcuwait_wait_event().
304 *
305 * WAIT WAKE
306 * [S] tsk = current [S] cond = true
307 * MB (A) MB (B)
308 * [L] cond [L] tsk
309 */
310 smp_rmb(); /* (B) */
311
312 /*
313 * Avoid using task_rcu_dereference() magic as long as we are careful,
314 * see comment in rcuwait_wait_event() regarding ->exit_state.
315 */
316 task = rcu_dereference(w->task);
317 if (task)
318 wake_up_process(task);
319 rcu_read_unlock();
320}
321
322/*
323 * Determine if a process group is "orphaned", according to the POSIX
324 * definition in 2.2.2.52. Orphaned process groups are not to be affected
325 * by terminal-generated stop signals. Newly orphaned process groups are
326 * to receive a SIGHUP and a SIGCONT.
327 *
328 * "I ask you, have you ever known what it is to be an orphan?"
329 */
330static int will_become_orphaned_pgrp(struct pid *pgrp,
331 struct task_struct *ignored_task)
332{
333 struct task_struct *p;
334
335 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
336 if ((p == ignored_task) ||
337 (p->exit_state && thread_group_empty(p)) ||
338 is_global_init(p->real_parent))
339 continue;
340
341 if (task_pgrp(p->real_parent) != pgrp &&
342 task_session(p->real_parent) == task_session(p))
343 return 0;
344 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
345
346 return 1;
347}
348
349int is_current_pgrp_orphaned(void)
350{
351 int retval;
352
353 read_lock(&tasklist_lock);
354 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
355 read_unlock(&tasklist_lock);
356
357 return retval;
358}
359
360static bool has_stopped_jobs(struct pid *pgrp)
361{
362 struct task_struct *p;
363
364 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
365 if (p->signal->flags & SIGNAL_STOP_STOPPED)
366 return true;
367 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
368
369 return false;
370}
371
372/*
373 * Check to see if any process groups have become orphaned as
374 * a result of our exiting, and if they have any stopped jobs,
375 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
376 */
377static void
378kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
379{
380 struct pid *pgrp = task_pgrp(tsk);
381 struct task_struct *ignored_task = tsk;
382
383 if (!parent)
384 /* exit: our father is in a different pgrp than
385 * we are and we were the only connection outside.
386 */
387 parent = tsk->real_parent;
388 else
389 /* reparent: our child is in a different pgrp than
390 * we are, and it was the only connection outside.
391 */
392 ignored_task = NULL;
393
394 if (task_pgrp(parent) != pgrp &&
395 task_session(parent) == task_session(tsk) &&
396 will_become_orphaned_pgrp(pgrp, ignored_task) &&
397 has_stopped_jobs(pgrp)) {
398 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
399 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
400 }
401}
402
403#ifdef CONFIG_MEMCG
404/*
405 * A task is exiting. If it owned this mm, find a new owner for the mm.
406 */
407void mm_update_next_owner(struct mm_struct *mm)
408{
409 struct task_struct *c, *g, *p = current;
410
411retry:
412 /*
413 * If the exiting or execing task is not the owner, it's
414 * someone else's problem.
415 */
416 if (mm->owner != p)
417 return;
418 /*
419 * The current owner is exiting/execing and there are no other
420 * candidates. Do not leave the mm pointing to a possibly
421 * freed task structure.
422 */
423 if (atomic_read(&mm->mm_users) <= 1) {
424 mm->owner = NULL;
425 return;
426 }
427
428 read_lock(&tasklist_lock);
429 /*
430 * Search in the children
431 */
432 list_for_each_entry(c, &p->children, sibling) {
433 if (c->mm == mm)
434 goto assign_new_owner;
435 }
436
437 /*
438 * Search in the siblings
439 */
440 list_for_each_entry(c, &p->real_parent->children, sibling) {
441 if (c->mm == mm)
442 goto assign_new_owner;
443 }
444
445 /*
446 * Search through everything else, we should not get here often.
447 */
448 for_each_process(g) {
449 if (g->flags & PF_KTHREAD)
450 continue;
451 for_each_thread(g, c) {
452 if (c->mm == mm)
453 goto assign_new_owner;
454 if (c->mm)
455 break;
456 }
457 }
458 read_unlock(&tasklist_lock);
459 /*
460 * We found no owner yet mm_users > 1: this implies that we are
461 * most likely racing with swapoff (try_to_unuse()) or /proc or
462 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
463 */
464 mm->owner = NULL;
465 return;
466
467assign_new_owner:
468 BUG_ON(c == p);
469 get_task_struct(c);
470 /*
471 * The task_lock protects c->mm from changing.
472 * We always want mm->owner->mm == mm
473 */
474 task_lock(c);
475 /*
476 * Delay read_unlock() till we have the task_lock()
477 * to ensure that c does not slip away underneath us
478 */
479 read_unlock(&tasklist_lock);
480 if (c->mm != mm) {
481 task_unlock(c);
482 put_task_struct(c);
483 goto retry;
484 }
485 mm->owner = c;
486 task_unlock(c);
487 put_task_struct(c);
488}
489#endif /* CONFIG_MEMCG */
490
491/*
492 * Turn us into a lazy TLB process if we
493 * aren't already..
494 */
495static void exit_mm(void)
496{
497 struct mm_struct *mm = current->mm;
498 struct core_state *core_state;
499
500 mm_release(current, mm);
501 if (!mm)
502 return;
503 sync_mm_rss(mm);
504 /*
505 * Serialize with any possible pending coredump.
506 * We must hold mmap_sem around checking core_state
507 * and clearing tsk->mm. The core-inducing thread
508 * will increment ->nr_threads for each thread in the
509 * group with ->mm != NULL.
510 */
511 down_read(&mm->mmap_sem);
512 core_state = mm->core_state;
513 if (core_state) {
514 struct core_thread self;
515
516 up_read(&mm->mmap_sem);
517
518 self.task = current;
519 self.next = xchg(&core_state->dumper.next, &self);
520 /*
521 * Implies mb(), the result of xchg() must be visible
522 * to core_state->dumper.
523 */
524 if (atomic_dec_and_test(&core_state->nr_threads))
525 complete(&core_state->startup);
526
527 for (;;) {
528 set_current_state(TASK_UNINTERRUPTIBLE);
529 if (!self.task) /* see coredump_finish() */
530 break;
531 freezable_schedule();
532 }
533 __set_current_state(TASK_RUNNING);
534 down_read(&mm->mmap_sem);
535 }
536 mmgrab(mm);
537 BUG_ON(mm != current->active_mm);
538 /* more a memory barrier than a real lock */
539 task_lock(current);
540 current->mm = NULL;
541 up_read(&mm->mmap_sem);
542 enter_lazy_tlb(mm, current);
543 task_unlock(current);
544 mm_update_next_owner(mm);
545 mmput(mm);
546 if (test_thread_flag(TIF_MEMDIE))
547 exit_oom_victim();
548}
549
550static struct task_struct *find_alive_thread(struct task_struct *p)
551{
552 struct task_struct *t;
553
554 for_each_thread(p, t) {
555 if (!(t->flags & PF_EXITING))
556 return t;
557 }
558 return NULL;
559}
560
561static struct task_struct *find_child_reaper(struct task_struct *father)
562 __releases(&tasklist_lock)
563 __acquires(&tasklist_lock)
564{
565 struct pid_namespace *pid_ns = task_active_pid_ns(father);
566 struct task_struct *reaper = pid_ns->child_reaper;
567
568 if (likely(reaper != father))
569 return reaper;
570
571 reaper = find_alive_thread(father);
572 if (reaper) {
573 pid_ns->child_reaper = reaper;
574 return reaper;
575 }
576
577 write_unlock_irq(&tasklist_lock);
578 if (unlikely(pid_ns == &init_pid_ns)) {
579 panic("Attempted to kill init! exitcode=0x%08x\n",
580 father->signal->group_exit_code ?: father->exit_code);
581 }
582 zap_pid_ns_processes(pid_ns);
583 write_lock_irq(&tasklist_lock);
584
585 return father;
586}
587
588/*
589 * When we die, we re-parent all our children, and try to:
590 * 1. give them to another thread in our thread group, if such a member exists
591 * 2. give it to the first ancestor process which prctl'd itself as a
592 * child_subreaper for its children (like a service manager)
593 * 3. give it to the init process (PID 1) in our pid namespace
594 */
595static struct task_struct *find_new_reaper(struct task_struct *father,
596 struct task_struct *child_reaper)
597{
598 struct task_struct *thread, *reaper;
599
600 thread = find_alive_thread(father);
601 if (thread)
602 return thread;
603
604 if (father->signal->has_child_subreaper) {
605 unsigned int ns_level = task_pid(father)->level;
606 /*
607 * Find the first ->is_child_subreaper ancestor in our pid_ns.
608 * We can't check reaper != child_reaper to ensure we do not
609 * cross the namespaces, the exiting parent could be injected
610 * by setns() + fork().
611 * We check pid->level, this is slightly more efficient than
612 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
613 */
614 for (reaper = father->real_parent;
615 task_pid(reaper)->level == ns_level;
616 reaper = reaper->real_parent) {
617 if (reaper == &init_task)
618 break;
619 if (!reaper->signal->is_child_subreaper)
620 continue;
621 thread = find_alive_thread(reaper);
622 if (thread)
623 return thread;
624 }
625 }
626
627 return child_reaper;
628}
629
630/*
631* Any that need to be release_task'd are put on the @dead list.
632 */
633static void reparent_leader(struct task_struct *father, struct task_struct *p,
634 struct list_head *dead)
635{
636 if (unlikely(p->exit_state == EXIT_DEAD))
637 return;
638
639 /* We don't want people slaying init. */
640 p->exit_signal = SIGCHLD;
641
642 /* If it has exited notify the new parent about this child's death. */
643 if (!p->ptrace &&
644 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
645 if (do_notify_parent(p, p->exit_signal)) {
646 p->exit_state = EXIT_DEAD;
647 list_add(&p->ptrace_entry, dead);
648 }
649 }
650
651 kill_orphaned_pgrp(p, father);
652}
653
654/*
655 * This does two things:
656 *
657 * A. Make init inherit all the child processes
658 * B. Check to see if any process groups have become orphaned
659 * as a result of our exiting, and if they have any stopped
660 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
661 */
662static void forget_original_parent(struct task_struct *father,
663 struct list_head *dead)
664{
665 struct task_struct *p, *t, *reaper;
666
667 if (unlikely(!list_empty(&father->ptraced)))
668 exit_ptrace(father, dead);
669
670 /* Can drop and reacquire tasklist_lock */
671 reaper = find_child_reaper(father);
672 if (list_empty(&father->children))
673 return;
674
675 reaper = find_new_reaper(father, reaper);
676 list_for_each_entry(p, &father->children, sibling) {
677 for_each_thread(p, t) {
678 t->real_parent = reaper;
679 BUG_ON((!t->ptrace) != (t->parent == father));
680 if (likely(!t->ptrace))
681 t->parent = t->real_parent;
682 if (t->pdeath_signal)
683 group_send_sig_info(t->pdeath_signal,
684 SEND_SIG_NOINFO, t,
685 PIDTYPE_TGID);
686 }
687 /*
688 * If this is a threaded reparent there is no need to
689 * notify anyone anything has happened.
690 */
691 if (!same_thread_group(reaper, father))
692 reparent_leader(father, p, dead);
693 }
694 list_splice_tail_init(&father->children, &reaper->children);
695}
696
697/*
698 * Send signals to all our closest relatives so that they know
699 * to properly mourn us..
700 */
701static void exit_notify(struct task_struct *tsk, int group_dead)
702{
703 bool autoreap;
704 struct task_struct *p, *n;
705 LIST_HEAD(dead);
706
707 write_lock_irq(&tasklist_lock);
708 forget_original_parent(tsk, &dead);
709
710 if (group_dead)
711 kill_orphaned_pgrp(tsk->group_leader, NULL);
712
713 if (unlikely(tsk->ptrace)) {
714 int sig = thread_group_leader(tsk) &&
715 thread_group_empty(tsk) &&
716 !ptrace_reparented(tsk) ?
717 tsk->exit_signal : SIGCHLD;
718 autoreap = do_notify_parent(tsk, sig);
719 } else if (thread_group_leader(tsk)) {
720 autoreap = thread_group_empty(tsk) &&
721 do_notify_parent(tsk, tsk->exit_signal);
722 } else {
723 autoreap = true;
724 }
725
726 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
727 if (tsk->exit_state == EXIT_DEAD)
728 list_add(&tsk->ptrace_entry, &dead);
729
730 /* mt-exec, de_thread() is waiting for group leader */
731 if (unlikely(tsk->signal->notify_count < 0))
732 wake_up_process(tsk->signal->group_exit_task);
733 write_unlock_irq(&tasklist_lock);
734
735 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
736 list_del_init(&p->ptrace_entry);
737 release_task(p);
738 }
739}
740
741#ifdef CONFIG_DEBUG_STACK_USAGE
742static void check_stack_usage(void)
743{
744 static DEFINE_SPINLOCK(low_water_lock);
745 static int lowest_to_date = THREAD_SIZE;
746 unsigned long free;
747
748 free = stack_not_used(current);
749
750 if (free >= lowest_to_date)
751 return;
752
753 spin_lock(&low_water_lock);
754 if (free < lowest_to_date) {
755 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
756 current->comm, task_pid_nr(current), free);
757 lowest_to_date = free;
758 }
759 spin_unlock(&low_water_lock);
760}
761#else
762static inline void check_stack_usage(void) {}
763#endif
764
765void __noreturn do_exit(long code)
766{
767 struct task_struct *tsk = current;
768 int group_dead;
769
770 profile_task_exit(tsk);
771 kcov_task_exit(tsk);
772
773 WARN_ON(blk_needs_flush_plug(tsk));
774
775 if (unlikely(in_interrupt()))
776 panic("Aiee, killing interrupt handler!");
777 if (unlikely(!tsk->pid))
778 panic("Attempted to kill the idle task!");
779
780 /*
781 * If do_exit is called because this processes oopsed, it's possible
782 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
783 * continuing. Amongst other possible reasons, this is to prevent
784 * mm_release()->clear_child_tid() from writing to a user-controlled
785 * kernel address.
786 */
787 set_fs(USER_DS);
788
789 ptrace_event(PTRACE_EVENT_EXIT, code);
790
791 validate_creds_for_do_exit(tsk);
792
793 /*
794 * We're taking recursive faults here in do_exit. Safest is to just
795 * leave this task alone and wait for reboot.
796 */
797 if (unlikely(tsk->flags & PF_EXITING)) {
798 pr_alert("Fixing recursive fault but reboot is needed!\n");
799 /*
800 * We can do this unlocked here. The futex code uses
801 * this flag just to verify whether the pi state
802 * cleanup has been done or not. In the worst case it
803 * loops once more. We pretend that the cleanup was
804 * done as there is no way to return. Either the
805 * OWNER_DIED bit is set by now or we push the blocked
806 * task into the wait for ever nirwana as well.
807 */
808 tsk->flags |= PF_EXITPIDONE;
809 set_current_state(TASK_UNINTERRUPTIBLE);
810 schedule();
811 }
812
813 exit_signals(tsk); /* sets PF_EXITING */
814 /*
815 * Ensure that all new tsk->pi_lock acquisitions must observe
816 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
817 */
818 smp_mb();
819 /*
820 * Ensure that we must observe the pi_state in exit_mm() ->
821 * mm_release() -> exit_pi_state_list().
822 */
823 raw_spin_lock_irq(&tsk->pi_lock);
824 raw_spin_unlock_irq(&tsk->pi_lock);
825
826 if (unlikely(in_atomic())) {
827 pr_info("note: %s[%d] exited with preempt_count %d\n",
828 current->comm, task_pid_nr(current),
829 preempt_count());
830 preempt_count_set(PREEMPT_ENABLED);
831 }
832
833 /* sync mm's RSS info before statistics gathering */
834 if (tsk->mm)
835 sync_mm_rss(tsk->mm);
836 acct_update_integrals(tsk);
837 group_dead = atomic_dec_and_test(&tsk->signal->live);
838 if (group_dead) {
839#ifdef CONFIG_POSIX_TIMERS
840 hrtimer_cancel(&tsk->signal->real_timer);
841 exit_itimers(tsk->signal);
842#endif
843 if (tsk->mm)
844 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
845 }
846 acct_collect(code, group_dead);
847 if (group_dead)
848 tty_audit_exit();
849 audit_free(tsk);
850
851 tsk->exit_code = code;
852 taskstats_exit(tsk, group_dead);
853
854 exit_mm();
855
856 if (group_dead)
857 acct_process();
858 trace_sched_process_exit(tsk);
859
860 exit_sem(tsk);
861 exit_shm(tsk);
862 exit_files(tsk);
863 exit_fs(tsk);
864 if (group_dead)
865 disassociate_ctty(1);
866 exit_task_namespaces(tsk);
867 exit_task_work(tsk);
868 exit_thread(tsk);
869 exit_umh(tsk);
870
871 /*
872 * Flush inherited counters to the parent - before the parent
873 * gets woken up by child-exit notifications.
874 *
875 * because of cgroup mode, must be called before cgroup_exit()
876 */
877 perf_event_exit_task(tsk);
878
879 sched_autogroup_exit_task(tsk);
880 cgroup_exit(tsk);
881
882 /*
883 * FIXME: do that only when needed, using sched_exit tracepoint
884 */
885 flush_ptrace_hw_breakpoint(tsk);
886
887 exit_tasks_rcu_start();
888 exit_notify(tsk, group_dead);
889 proc_exit_connector(tsk);
890 mpol_put_task_policy(tsk);
891#ifdef CONFIG_FUTEX
892 if (unlikely(current->pi_state_cache))
893 kfree(current->pi_state_cache);
894#endif
895 /*
896 * Make sure we are holding no locks:
897 */
898 debug_check_no_locks_held();
899 /*
900 * We can do this unlocked here. The futex code uses this flag
901 * just to verify whether the pi state cleanup has been done
902 * or not. In the worst case it loops once more.
903 */
904 tsk->flags |= PF_EXITPIDONE;
905
906 if (tsk->io_context)
907 exit_io_context(tsk);
908
909 if (tsk->splice_pipe)
910 free_pipe_info(tsk->splice_pipe);
911
912 if (tsk->task_frag.page)
913 put_page(tsk->task_frag.page);
914
915 validate_creds_for_do_exit(tsk);
916
917 check_stack_usage();
918 preempt_disable();
919 if (tsk->nr_dirtied)
920 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
921 exit_rcu();
922 exit_tasks_rcu_finish();
923
924 lockdep_free_task(tsk);
925 do_task_dead();
926}
927EXPORT_SYMBOL_GPL(do_exit);
928
929void complete_and_exit(struct completion *comp, long code)
930{
931 if (comp)
932 complete(comp);
933
934 do_exit(code);
935}
936EXPORT_SYMBOL(complete_and_exit);
937
938SYSCALL_DEFINE1(exit, int, error_code)
939{
940 do_exit((error_code&0xff)<<8);
941}
942
943/*
944 * Take down every thread in the group. This is called by fatal signals
945 * as well as by sys_exit_group (below).
946 */
947void
948do_group_exit(int exit_code)
949{
950 struct signal_struct *sig = current->signal;
951
952 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
953
954 if (signal_group_exit(sig))
955 exit_code = sig->group_exit_code;
956 else if (!thread_group_empty(current)) {
957 struct sighand_struct *const sighand = current->sighand;
958
959 spin_lock_irq(&sighand->siglock);
960 if (signal_group_exit(sig))
961 /* Another thread got here before we took the lock. */
962 exit_code = sig->group_exit_code;
963 else {
964 sig->group_exit_code = exit_code;
965 sig->flags = SIGNAL_GROUP_EXIT;
966 zap_other_threads(current);
967 }
968 spin_unlock_irq(&sighand->siglock);
969 }
970
971 do_exit(exit_code);
972 /* NOTREACHED */
973}
974
975/*
976 * this kills every thread in the thread group. Note that any externally
977 * wait4()-ing process will get the correct exit code - even if this
978 * thread is not the thread group leader.
979 */
980SYSCALL_DEFINE1(exit_group, int, error_code)
981{
982 do_group_exit((error_code & 0xff) << 8);
983 /* NOTREACHED */
984 return 0;
985}
986
987struct waitid_info {
988 pid_t pid;
989 uid_t uid;
990 int status;
991 int cause;
992};
993
994struct wait_opts {
995 enum pid_type wo_type;
996 int wo_flags;
997 struct pid *wo_pid;
998
999 struct waitid_info *wo_info;
1000 int wo_stat;
1001 struct rusage *wo_rusage;
1002
1003 wait_queue_entry_t child_wait;
1004 int notask_error;
1005};
1006
1007static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1008{
1009 return wo->wo_type == PIDTYPE_MAX ||
1010 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1011}
1012
1013static int
1014eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1015{
1016 if (!eligible_pid(wo, p))
1017 return 0;
1018
1019 /*
1020 * Wait for all children (clone and not) if __WALL is set or
1021 * if it is traced by us.
1022 */
1023 if (ptrace || (wo->wo_flags & __WALL))
1024 return 1;
1025
1026 /*
1027 * Otherwise, wait for clone children *only* if __WCLONE is set;
1028 * otherwise, wait for non-clone children *only*.
1029 *
1030 * Note: a "clone" child here is one that reports to its parent
1031 * using a signal other than SIGCHLD, or a non-leader thread which
1032 * we can only see if it is traced by us.
1033 */
1034 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1035 return 0;
1036
1037 return 1;
1038}
1039
1040/*
1041 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1042 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1043 * the lock and this task is uninteresting. If we return nonzero, we have
1044 * released the lock and the system call should return.
1045 */
1046static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1047{
1048 int state, status;
1049 pid_t pid = task_pid_vnr(p);
1050 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1051 struct waitid_info *infop;
1052
1053 if (!likely(wo->wo_flags & WEXITED))
1054 return 0;
1055
1056 if (unlikely(wo->wo_flags & WNOWAIT)) {
1057 status = p->exit_code;
1058 get_task_struct(p);
1059 read_unlock(&tasklist_lock);
1060 sched_annotate_sleep();
1061 if (wo->wo_rusage)
1062 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1063 put_task_struct(p);
1064 goto out_info;
1065 }
1066 /*
1067 * Move the task's state to DEAD/TRACE, only one thread can do this.
1068 */
1069 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1070 EXIT_TRACE : EXIT_DEAD;
1071 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1072 return 0;
1073 /*
1074 * We own this thread, nobody else can reap it.
1075 */
1076 read_unlock(&tasklist_lock);
1077 sched_annotate_sleep();
1078
1079 /*
1080 * Check thread_group_leader() to exclude the traced sub-threads.
1081 */
1082 if (state == EXIT_DEAD && thread_group_leader(p)) {
1083 struct signal_struct *sig = p->signal;
1084 struct signal_struct *psig = current->signal;
1085 unsigned long maxrss;
1086 u64 tgutime, tgstime;
1087
1088 /*
1089 * The resource counters for the group leader are in its
1090 * own task_struct. Those for dead threads in the group
1091 * are in its signal_struct, as are those for the child
1092 * processes it has previously reaped. All these
1093 * accumulate in the parent's signal_struct c* fields.
1094 *
1095 * We don't bother to take a lock here to protect these
1096 * p->signal fields because the whole thread group is dead
1097 * and nobody can change them.
1098 *
1099 * psig->stats_lock also protects us from our sub-theads
1100 * which can reap other children at the same time. Until
1101 * we change k_getrusage()-like users to rely on this lock
1102 * we have to take ->siglock as well.
1103 *
1104 * We use thread_group_cputime_adjusted() to get times for
1105 * the thread group, which consolidates times for all threads
1106 * in the group including the group leader.
1107 */
1108 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1109 spin_lock_irq(&current->sighand->siglock);
1110 write_seqlock(&psig->stats_lock);
1111 psig->cutime += tgutime + sig->cutime;
1112 psig->cstime += tgstime + sig->cstime;
1113 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1114 psig->cmin_flt +=
1115 p->min_flt + sig->min_flt + sig->cmin_flt;
1116 psig->cmaj_flt +=
1117 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1118 psig->cnvcsw +=
1119 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1120 psig->cnivcsw +=
1121 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1122 psig->cinblock +=
1123 task_io_get_inblock(p) +
1124 sig->inblock + sig->cinblock;
1125 psig->coublock +=
1126 task_io_get_oublock(p) +
1127 sig->oublock + sig->coublock;
1128 maxrss = max(sig->maxrss, sig->cmaxrss);
1129 if (psig->cmaxrss < maxrss)
1130 psig->cmaxrss = maxrss;
1131 task_io_accounting_add(&psig->ioac, &p->ioac);
1132 task_io_accounting_add(&psig->ioac, &sig->ioac);
1133 write_sequnlock(&psig->stats_lock);
1134 spin_unlock_irq(&current->sighand->siglock);
1135 }
1136
1137 if (wo->wo_rusage)
1138 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1139 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1140 ? p->signal->group_exit_code : p->exit_code;
1141 wo->wo_stat = status;
1142
1143 if (state == EXIT_TRACE) {
1144 write_lock_irq(&tasklist_lock);
1145 /* We dropped tasklist, ptracer could die and untrace */
1146 ptrace_unlink(p);
1147
1148 /* If parent wants a zombie, don't release it now */
1149 state = EXIT_ZOMBIE;
1150 if (do_notify_parent(p, p->exit_signal))
1151 state = EXIT_DEAD;
1152 p->exit_state = state;
1153 write_unlock_irq(&tasklist_lock);
1154 }
1155 if (state == EXIT_DEAD)
1156 release_task(p);
1157
1158out_info:
1159 infop = wo->wo_info;
1160 if (infop) {
1161 if ((status & 0x7f) == 0) {
1162 infop->cause = CLD_EXITED;
1163 infop->status = status >> 8;
1164 } else {
1165 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1166 infop->status = status & 0x7f;
1167 }
1168 infop->pid = pid;
1169 infop->uid = uid;
1170 }
1171
1172 return pid;
1173}
1174
1175static int *task_stopped_code(struct task_struct *p, bool ptrace)
1176{
1177 if (ptrace) {
1178 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1179 return &p->exit_code;
1180 } else {
1181 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1182 return &p->signal->group_exit_code;
1183 }
1184 return NULL;
1185}
1186
1187/**
1188 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1189 * @wo: wait options
1190 * @ptrace: is the wait for ptrace
1191 * @p: task to wait for
1192 *
1193 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1194 *
1195 * CONTEXT:
1196 * read_lock(&tasklist_lock), which is released if return value is
1197 * non-zero. Also, grabs and releases @p->sighand->siglock.
1198 *
1199 * RETURNS:
1200 * 0 if wait condition didn't exist and search for other wait conditions
1201 * should continue. Non-zero return, -errno on failure and @p's pid on
1202 * success, implies that tasklist_lock is released and wait condition
1203 * search should terminate.
1204 */
1205static int wait_task_stopped(struct wait_opts *wo,
1206 int ptrace, struct task_struct *p)
1207{
1208 struct waitid_info *infop;
1209 int exit_code, *p_code, why;
1210 uid_t uid = 0; /* unneeded, required by compiler */
1211 pid_t pid;
1212
1213 /*
1214 * Traditionally we see ptrace'd stopped tasks regardless of options.
1215 */
1216 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1217 return 0;
1218
1219 if (!task_stopped_code(p, ptrace))
1220 return 0;
1221
1222 exit_code = 0;
1223 spin_lock_irq(&p->sighand->siglock);
1224
1225 p_code = task_stopped_code(p, ptrace);
1226 if (unlikely(!p_code))
1227 goto unlock_sig;
1228
1229 exit_code = *p_code;
1230 if (!exit_code)
1231 goto unlock_sig;
1232
1233 if (!unlikely(wo->wo_flags & WNOWAIT))
1234 *p_code = 0;
1235
1236 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1237unlock_sig:
1238 spin_unlock_irq(&p->sighand->siglock);
1239 if (!exit_code)
1240 return 0;
1241
1242 /*
1243 * Now we are pretty sure this task is interesting.
1244 * Make sure it doesn't get reaped out from under us while we
1245 * give up the lock and then examine it below. We don't want to
1246 * keep holding onto the tasklist_lock while we call getrusage and
1247 * possibly take page faults for user memory.
1248 */
1249 get_task_struct(p);
1250 pid = task_pid_vnr(p);
1251 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1252 read_unlock(&tasklist_lock);
1253 sched_annotate_sleep();
1254 if (wo->wo_rusage)
1255 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1256 put_task_struct(p);
1257
1258 if (likely(!(wo->wo_flags & WNOWAIT)))
1259 wo->wo_stat = (exit_code << 8) | 0x7f;
1260
1261 infop = wo->wo_info;
1262 if (infop) {
1263 infop->cause = why;
1264 infop->status = exit_code;
1265 infop->pid = pid;
1266 infop->uid = uid;
1267 }
1268 return pid;
1269}
1270
1271/*
1272 * Handle do_wait work for one task in a live, non-stopped state.
1273 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1274 * the lock and this task is uninteresting. If we return nonzero, we have
1275 * released the lock and the system call should return.
1276 */
1277static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1278{
1279 struct waitid_info *infop;
1280 pid_t pid;
1281 uid_t uid;
1282
1283 if (!unlikely(wo->wo_flags & WCONTINUED))
1284 return 0;
1285
1286 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1287 return 0;
1288
1289 spin_lock_irq(&p->sighand->siglock);
1290 /* Re-check with the lock held. */
1291 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1292 spin_unlock_irq(&p->sighand->siglock);
1293 return 0;
1294 }
1295 if (!unlikely(wo->wo_flags & WNOWAIT))
1296 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1297 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1298 spin_unlock_irq(&p->sighand->siglock);
1299
1300 pid = task_pid_vnr(p);
1301 get_task_struct(p);
1302 read_unlock(&tasklist_lock);
1303 sched_annotate_sleep();
1304 if (wo->wo_rusage)
1305 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1306 put_task_struct(p);
1307
1308 infop = wo->wo_info;
1309 if (!infop) {
1310 wo->wo_stat = 0xffff;
1311 } else {
1312 infop->cause = CLD_CONTINUED;
1313 infop->pid = pid;
1314 infop->uid = uid;
1315 infop->status = SIGCONT;
1316 }
1317 return pid;
1318}
1319
1320/*
1321 * Consider @p for a wait by @parent.
1322 *
1323 * -ECHILD should be in ->notask_error before the first call.
1324 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1325 * Returns zero if the search for a child should continue;
1326 * then ->notask_error is 0 if @p is an eligible child,
1327 * or still -ECHILD.
1328 */
1329static int wait_consider_task(struct wait_opts *wo, int ptrace,
1330 struct task_struct *p)
1331{
1332 /*
1333 * We can race with wait_task_zombie() from another thread.
1334 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1335 * can't confuse the checks below.
1336 */
1337 int exit_state = READ_ONCE(p->exit_state);
1338 int ret;
1339
1340 if (unlikely(exit_state == EXIT_DEAD))
1341 return 0;
1342
1343 ret = eligible_child(wo, ptrace, p);
1344 if (!ret)
1345 return ret;
1346
1347 if (unlikely(exit_state == EXIT_TRACE)) {
1348 /*
1349 * ptrace == 0 means we are the natural parent. In this case
1350 * we should clear notask_error, debugger will notify us.
1351 */
1352 if (likely(!ptrace))
1353 wo->notask_error = 0;
1354 return 0;
1355 }
1356
1357 if (likely(!ptrace) && unlikely(p->ptrace)) {
1358 /*
1359 * If it is traced by its real parent's group, just pretend
1360 * the caller is ptrace_do_wait() and reap this child if it
1361 * is zombie.
1362 *
1363 * This also hides group stop state from real parent; otherwise
1364 * a single stop can be reported twice as group and ptrace stop.
1365 * If a ptracer wants to distinguish these two events for its
1366 * own children it should create a separate process which takes
1367 * the role of real parent.
1368 */
1369 if (!ptrace_reparented(p))
1370 ptrace = 1;
1371 }
1372
1373 /* slay zombie? */
1374 if (exit_state == EXIT_ZOMBIE) {
1375 /* we don't reap group leaders with subthreads */
1376 if (!delay_group_leader(p)) {
1377 /*
1378 * A zombie ptracee is only visible to its ptracer.
1379 * Notification and reaping will be cascaded to the
1380 * real parent when the ptracer detaches.
1381 */
1382 if (unlikely(ptrace) || likely(!p->ptrace))
1383 return wait_task_zombie(wo, p);
1384 }
1385
1386 /*
1387 * Allow access to stopped/continued state via zombie by
1388 * falling through. Clearing of notask_error is complex.
1389 *
1390 * When !@ptrace:
1391 *
1392 * If WEXITED is set, notask_error should naturally be
1393 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1394 * so, if there are live subthreads, there are events to
1395 * wait for. If all subthreads are dead, it's still safe
1396 * to clear - this function will be called again in finite
1397 * amount time once all the subthreads are released and
1398 * will then return without clearing.
1399 *
1400 * When @ptrace:
1401 *
1402 * Stopped state is per-task and thus can't change once the
1403 * target task dies. Only continued and exited can happen.
1404 * Clear notask_error if WCONTINUED | WEXITED.
1405 */
1406 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1407 wo->notask_error = 0;
1408 } else {
1409 /*
1410 * @p is alive and it's gonna stop, continue or exit, so
1411 * there always is something to wait for.
1412 */
1413 wo->notask_error = 0;
1414 }
1415
1416 /*
1417 * Wait for stopped. Depending on @ptrace, different stopped state
1418 * is used and the two don't interact with each other.
1419 */
1420 ret = wait_task_stopped(wo, ptrace, p);
1421 if (ret)
1422 return ret;
1423
1424 /*
1425 * Wait for continued. There's only one continued state and the
1426 * ptracer can consume it which can confuse the real parent. Don't
1427 * use WCONTINUED from ptracer. You don't need or want it.
1428 */
1429 return wait_task_continued(wo, p);
1430}
1431
1432/*
1433 * Do the work of do_wait() for one thread in the group, @tsk.
1434 *
1435 * -ECHILD should be in ->notask_error before the first call.
1436 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1437 * Returns zero if the search for a child should continue; then
1438 * ->notask_error is 0 if there were any eligible children,
1439 * or still -ECHILD.
1440 */
1441static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1442{
1443 struct task_struct *p;
1444
1445 list_for_each_entry(p, &tsk->children, sibling) {
1446 int ret = wait_consider_task(wo, 0, p);
1447
1448 if (ret)
1449 return ret;
1450 }
1451
1452 return 0;
1453}
1454
1455static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1456{
1457 struct task_struct *p;
1458
1459 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1460 int ret = wait_consider_task(wo, 1, p);
1461
1462 if (ret)
1463 return ret;
1464 }
1465
1466 return 0;
1467}
1468
1469static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1470 int sync, void *key)
1471{
1472 struct wait_opts *wo = container_of(wait, struct wait_opts,
1473 child_wait);
1474 struct task_struct *p = key;
1475
1476 if (!eligible_pid(wo, p))
1477 return 0;
1478
1479 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1480 return 0;
1481
1482 return default_wake_function(wait, mode, sync, key);
1483}
1484
1485void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1486{
1487 __wake_up_sync_key(&parent->signal->wait_chldexit,
1488 TASK_INTERRUPTIBLE, 1, p);
1489}
1490
1491static long do_wait(struct wait_opts *wo)
1492{
1493 struct task_struct *tsk;
1494 int retval;
1495
1496 trace_sched_process_wait(wo->wo_pid);
1497
1498 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1499 wo->child_wait.private = current;
1500 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1501repeat:
1502 /*
1503 * If there is nothing that can match our criteria, just get out.
1504 * We will clear ->notask_error to zero if we see any child that
1505 * might later match our criteria, even if we are not able to reap
1506 * it yet.
1507 */
1508 wo->notask_error = -ECHILD;
1509 if ((wo->wo_type < PIDTYPE_MAX) &&
1510 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1511 goto notask;
1512
1513 set_current_state(TASK_INTERRUPTIBLE);
1514 read_lock(&tasklist_lock);
1515 tsk = current;
1516 do {
1517 retval = do_wait_thread(wo, tsk);
1518 if (retval)
1519 goto end;
1520
1521 retval = ptrace_do_wait(wo, tsk);
1522 if (retval)
1523 goto end;
1524
1525 if (wo->wo_flags & __WNOTHREAD)
1526 break;
1527 } while_each_thread(current, tsk);
1528 read_unlock(&tasklist_lock);
1529
1530notask:
1531 retval = wo->notask_error;
1532 if (!retval && !(wo->wo_flags & WNOHANG)) {
1533 retval = -ERESTARTSYS;
1534 if (!signal_pending(current)) {
1535 schedule();
1536 goto repeat;
1537 }
1538 }
1539end:
1540 __set_current_state(TASK_RUNNING);
1541 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1542 return retval;
1543}
1544
1545static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1546 int options, struct rusage *ru)
1547{
1548 struct wait_opts wo;
1549 struct pid *pid = NULL;
1550 enum pid_type type;
1551 long ret;
1552
1553 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1554 __WNOTHREAD|__WCLONE|__WALL))
1555 return -EINVAL;
1556 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1557 return -EINVAL;
1558
1559 switch (which) {
1560 case P_ALL:
1561 type = PIDTYPE_MAX;
1562 break;
1563 case P_PID:
1564 type = PIDTYPE_PID;
1565 if (upid <= 0)
1566 return -EINVAL;
1567 break;
1568 case P_PGID:
1569 type = PIDTYPE_PGID;
1570 if (upid <= 0)
1571 return -EINVAL;
1572 break;
1573 default:
1574 return -EINVAL;
1575 }
1576
1577 if (type < PIDTYPE_MAX)
1578 pid = find_get_pid(upid);
1579
1580 wo.wo_type = type;
1581 wo.wo_pid = pid;
1582 wo.wo_flags = options;
1583 wo.wo_info = infop;
1584 wo.wo_rusage = ru;
1585 ret = do_wait(&wo);
1586
1587 put_pid(pid);
1588 return ret;
1589}
1590
1591SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1592 infop, int, options, struct rusage __user *, ru)
1593{
1594 struct rusage r;
1595 struct waitid_info info = {.status = 0};
1596 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1597 int signo = 0;
1598
1599 if (err > 0) {
1600 signo = SIGCHLD;
1601 err = 0;
1602 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1603 return -EFAULT;
1604 }
1605 if (!infop)
1606 return err;
1607
1608 if (!user_access_begin(infop, sizeof(*infop)))
1609 return -EFAULT;
1610
1611 unsafe_put_user(signo, &infop->si_signo, Efault);
1612 unsafe_put_user(0, &infop->si_errno, Efault);
1613 unsafe_put_user(info.cause, &infop->si_code, Efault);
1614 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1615 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1616 unsafe_put_user(info.status, &infop->si_status, Efault);
1617 user_access_end();
1618 return err;
1619Efault:
1620 user_access_end();
1621 return -EFAULT;
1622}
1623
1624long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1625 struct rusage *ru)
1626{
1627 struct wait_opts wo;
1628 struct pid *pid = NULL;
1629 enum pid_type type;
1630 long ret;
1631
1632 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1633 __WNOTHREAD|__WCLONE|__WALL))
1634 return -EINVAL;
1635
1636 /* -INT_MIN is not defined */
1637 if (upid == INT_MIN)
1638 return -ESRCH;
1639
1640 if (upid == -1)
1641 type = PIDTYPE_MAX;
1642 else if (upid < 0) {
1643 type = PIDTYPE_PGID;
1644 pid = find_get_pid(-upid);
1645 } else if (upid == 0) {
1646 type = PIDTYPE_PGID;
1647 pid = get_task_pid(current, PIDTYPE_PGID);
1648 } else /* upid > 0 */ {
1649 type = PIDTYPE_PID;
1650 pid = find_get_pid(upid);
1651 }
1652
1653 wo.wo_type = type;
1654 wo.wo_pid = pid;
1655 wo.wo_flags = options | WEXITED;
1656 wo.wo_info = NULL;
1657 wo.wo_stat = 0;
1658 wo.wo_rusage = ru;
1659 ret = do_wait(&wo);
1660 put_pid(pid);
1661 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1662 ret = -EFAULT;
1663
1664 return ret;
1665}
1666
1667SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1668 int, options, struct rusage __user *, ru)
1669{
1670 struct rusage r;
1671 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1672
1673 if (err > 0) {
1674 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1675 return -EFAULT;
1676 }
1677 return err;
1678}
1679
1680#ifdef __ARCH_WANT_SYS_WAITPID
1681
1682/*
1683 * sys_waitpid() remains for compatibility. waitpid() should be
1684 * implemented by calling sys_wait4() from libc.a.
1685 */
1686SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1687{
1688 return kernel_wait4(pid, stat_addr, options, NULL);
1689}
1690
1691#endif
1692
1693#ifdef CONFIG_COMPAT
1694COMPAT_SYSCALL_DEFINE4(wait4,
1695 compat_pid_t, pid,
1696 compat_uint_t __user *, stat_addr,
1697 int, options,
1698 struct compat_rusage __user *, ru)
1699{
1700 struct rusage r;
1701 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1702 if (err > 0) {
1703 if (ru && put_compat_rusage(&r, ru))
1704 return -EFAULT;
1705 }
1706 return err;
1707}
1708
1709COMPAT_SYSCALL_DEFINE5(waitid,
1710 int, which, compat_pid_t, pid,
1711 struct compat_siginfo __user *, infop, int, options,
1712 struct compat_rusage __user *, uru)
1713{
1714 struct rusage ru;
1715 struct waitid_info info = {.status = 0};
1716 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1717 int signo = 0;
1718 if (err > 0) {
1719 signo = SIGCHLD;
1720 err = 0;
1721 if (uru) {
1722 /* kernel_waitid() overwrites everything in ru */
1723 if (COMPAT_USE_64BIT_TIME)
1724 err = copy_to_user(uru, &ru, sizeof(ru));
1725 else
1726 err = put_compat_rusage(&ru, uru);
1727 if (err)
1728 return -EFAULT;
1729 }
1730 }
1731
1732 if (!infop)
1733 return err;
1734
1735 if (!user_access_begin(infop, sizeof(*infop)))
1736 return -EFAULT;
1737
1738 unsafe_put_user(signo, &infop->si_signo, Efault);
1739 unsafe_put_user(0, &infop->si_errno, Efault);
1740 unsafe_put_user(info.cause, &infop->si_code, Efault);
1741 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1742 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1743 unsafe_put_user(info.status, &infop->si_status, Efault);
1744 user_access_end();
1745 return err;
1746Efault:
1747 user_access_end();
1748 return -EFAULT;
1749}
1750#endif
1751
1752__weak void abort(void)
1753{
1754 BUG();
1755
1756 /* if that doesn't kill us, halt */
1757 panic("Oops failed to kill thread");
1758}
1759EXPORT_SYMBOL(abort);