projects/linux·blob 4563243e · refs

1974 lines · 48.6 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/exit.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/mm.h>
9#include <linux/slab.h>
10#include <linux/sched/autogroup.h>
11#include <linux/sched/mm.h>
12#include <linux/sched/stat.h>
13#include <linux/sched/task.h>
14#include <linux/sched/task_stack.h>
15#include <linux/sched/cputime.h>
16#include <linux/interrupt.h>
17#include <linux/module.h>
18#include <linux/capability.h>
19#include <linux/completion.h>
20#include <linux/personality.h>
21#include <linux/tty.h>
22#include <linux/iocontext.h>
23#include <linux/key.h>
24#include <linux/cpu.h>
25#include <linux/acct.h>
26#include <linux/tsacct_kern.h>
27#include <linux/file.h>
28#include <linux/fdtable.h>
29#include <linux/freezer.h>
30#include <linux/binfmts.h>
31#include <linux/nsproxy.h>
32#include <linux/pid_namespace.h>
33#include <linux/ptrace.h>
34#include <linux/profile.h>
35#include <linux/mount.h>
36#include <linux/proc_fs.h>
37#include <linux/kthread.h>
38#include <linux/mempolicy.h>
39#include <linux/taskstats_kern.h>
40#include <linux/delayacct.h>
41#include <linux/cgroup.h>
42#include <linux/syscalls.h>
43#include <linux/signal.h>
44#include <linux/posix-timers.h>
45#include <linux/cn_proc.h>
46#include <linux/mutex.h>
47#include <linux/futex.h>
48#include <linux/pipe_fs_i.h>
49#include <linux/audit.h> /* for audit_free() */
50#include <linux/resource.h>
51#include <linux/task_io_accounting_ops.h>
52#include <linux/blkdev.h>
53#include <linux/task_work.h>
54#include <linux/fs_struct.h>
55#include <linux/init_task.h>
56#include <linux/perf_event.h>
57#include <trace/events/sched.h>
58#include <linux/hw_breakpoint.h>
59#include <linux/oom.h>
60#include <linux/writeback.h>
61#include <linux/shm.h>
62#include <linux/kcov.h>
63#include <linux/kmsan.h>
64#include <linux/random.h>
65#include <linux/rcuwait.h>
66#include <linux/compat.h>
67#include <linux/io_uring.h>
68#include <linux/kprobes.h>
69#include <linux/rethook.h>
70#include <linux/sysfs.h>
71#include <linux/user_events.h>
72#include <linux/uaccess.h>
73
74#include <uapi/linux/wait.h>
75
76#include <asm/unistd.h>
77#include <asm/mmu_context.h>
78
79#include "exit.h"
80
81/*
82 * The default value should be high enough to not crash a system that randomly
83 * crashes its kernel from time to time, but low enough to at least not permit
84 * overflowing 32-bit refcounts or the ldsem writer count.
85 */
86static unsigned int oops_limit = 10000;
87
88#ifdef CONFIG_SYSCTL
89static struct ctl_table kern_exit_table[] = {
90 {
91 .procname = "oops_limit",
92 .data = &oops_limit,
93 .maxlen = sizeof(oops_limit),
94 .mode = 0644,
95 .proc_handler = proc_douintvec,
96 },
97};
98
99static __init int kernel_exit_sysctls_init(void)
100{
101 register_sysctl_init("kernel", kern_exit_table);
102 return 0;
103}
104late_initcall(kernel_exit_sysctls_init);
105#endif
106
107static atomic_t oops_count = ATOMIC_INIT(0);
108
109#ifdef CONFIG_SYSFS
110static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
111 char *page)
112{
113 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
114}
115
116static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
117
118static __init int kernel_exit_sysfs_init(void)
119{
120 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
121 return 0;
122}
123late_initcall(kernel_exit_sysfs_init);
124#endif
125
126static void __unhash_process(struct task_struct *p, bool group_dead)
127{
128 nr_threads--;
129 detach_pid(p, PIDTYPE_PID);
130 if (group_dead) {
131 detach_pid(p, PIDTYPE_TGID);
132 detach_pid(p, PIDTYPE_PGID);
133 detach_pid(p, PIDTYPE_SID);
134
135 list_del_rcu(&p->tasks);
136 list_del_init(&p->sibling);
137 __this_cpu_dec(process_counts);
138 }
139 list_del_rcu(&p->thread_node);
140}
141
142/*
143 * This function expects the tasklist_lock write-locked.
144 */
145static void __exit_signal(struct task_struct *tsk)
146{
147 struct signal_struct *sig = tsk->signal;
148 bool group_dead = thread_group_leader(tsk);
149 struct sighand_struct *sighand;
150 struct tty_struct *tty;
151 u64 utime, stime;
152
153 sighand = rcu_dereference_check(tsk->sighand,
154 lockdep_tasklist_lock_is_held());
155 spin_lock(&sighand->siglock);
156
157#ifdef CONFIG_POSIX_TIMERS
158 posix_cpu_timers_exit(tsk);
159 if (group_dead)
160 posix_cpu_timers_exit_group(tsk);
161#endif
162
163 if (group_dead) {
164 tty = sig->tty;
165 sig->tty = NULL;
166 } else {
167 /*
168 * If there is any task waiting for the group exit
169 * then notify it:
170 */
171 if (sig->notify_count > 0 && !--sig->notify_count)
172 wake_up_process(sig->group_exec_task);
173
174 if (tsk == sig->curr_target)
175 sig->curr_target = next_thread(tsk);
176 }
177
178 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
179 sizeof(unsigned long long));
180
181 /*
182 * Accumulate here the counters for all threads as they die. We could
183 * skip the group leader because it is the last user of signal_struct,
184 * but we want to avoid the race with thread_group_cputime() which can
185 * see the empty ->thread_head list.
186 */
187 task_cputime(tsk, &utime, &stime);
188 write_seqlock(&sig->stats_lock);
189 sig->utime += utime;
190 sig->stime += stime;
191 sig->gtime += task_gtime(tsk);
192 sig->min_flt += tsk->min_flt;
193 sig->maj_flt += tsk->maj_flt;
194 sig->nvcsw += tsk->nvcsw;
195 sig->nivcsw += tsk->nivcsw;
196 sig->inblock += task_io_get_inblock(tsk);
197 sig->oublock += task_io_get_oublock(tsk);
198 task_io_accounting_add(&sig->ioac, &tsk->ioac);
199 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
200 sig->nr_threads--;
201 __unhash_process(tsk, group_dead);
202 write_sequnlock(&sig->stats_lock);
203
204 /*
205 * Do this under ->siglock, we can race with another thread
206 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
207 */
208 flush_sigqueue(&tsk->pending);
209 tsk->sighand = NULL;
210 spin_unlock(&sighand->siglock);
211
212 __cleanup_sighand(sighand);
213 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
214 if (group_dead) {
215 flush_sigqueue(&sig->shared_pending);
216 tty_kref_put(tty);
217 }
218}
219
220static void delayed_put_task_struct(struct rcu_head *rhp)
221{
222 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
223
224 kprobe_flush_task(tsk);
225 rethook_flush_task(tsk);
226 perf_event_delayed_put(tsk);
227 trace_sched_process_free(tsk);
228 put_task_struct(tsk);
229}
230
231void put_task_struct_rcu_user(struct task_struct *task)
232{
233 if (refcount_dec_and_test(&task->rcu_users))
234 call_rcu(&task->rcu, delayed_put_task_struct);
235}
236
237void __weak release_thread(struct task_struct *dead_task)
238{
239}
240
241void release_task(struct task_struct *p)
242{
243 struct task_struct *leader;
244 struct pid *thread_pid;
245 int zap_leader;
246repeat:
247 /* don't need to get the RCU readlock here - the process is dead and
248 * can't be modifying its own credentials. But shut RCU-lockdep up */
249 rcu_read_lock();
250 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
251 rcu_read_unlock();
252
253 cgroup_release(p);
254
255 write_lock_irq(&tasklist_lock);
256 ptrace_release_task(p);
257 thread_pid = get_pid(p->thread_pid);
258 __exit_signal(p);
259
260 /*
261 * If we are the last non-leader member of the thread
262 * group, and the leader is zombie, then notify the
263 * group leader's parent process. (if it wants notification.)
264 */
265 zap_leader = 0;
266 leader = p->group_leader;
267 if (leader != p && thread_group_empty(leader)
268 && leader->exit_state == EXIT_ZOMBIE) {
269 /*
270 * If we were the last child thread and the leader has
271 * exited already, and the leader's parent ignores SIGCHLD,
272 * then we are the one who should release the leader.
273 */
274 zap_leader = do_notify_parent(leader, leader->exit_signal);
275 if (zap_leader)
276 leader->exit_state = EXIT_DEAD;
277 }
278
279 write_unlock_irq(&tasklist_lock);
280 proc_flush_pid(thread_pid);
281 put_pid(thread_pid);
282 release_thread(p);
283 put_task_struct_rcu_user(p);
284
285 p = leader;
286 if (unlikely(zap_leader))
287 goto repeat;
288}
289
290int rcuwait_wake_up(struct rcuwait *w)
291{
292 int ret = 0;
293 struct task_struct *task;
294
295 rcu_read_lock();
296
297 /*
298 * Order condition vs @task, such that everything prior to the load
299 * of @task is visible. This is the condition as to why the user called
300 * rcuwait_wake() in the first place. Pairs with set_current_state()
301 * barrier (A) in rcuwait_wait_event().
302 *
303 * WAIT WAKE
304 * [S] tsk = current [S] cond = true
305 * MB (A) MB (B)
306 * [L] cond [L] tsk
307 */
308 smp_mb(); /* (B) */
309
310 task = rcu_dereference(w->task);
311 if (task)
312 ret = wake_up_process(task);
313 rcu_read_unlock();
314
315 return ret;
316}
317EXPORT_SYMBOL_GPL(rcuwait_wake_up);
318
319/*
320 * Determine if a process group is "orphaned", according to the POSIX
321 * definition in 2.2.2.52. Orphaned process groups are not to be affected
322 * by terminal-generated stop signals. Newly orphaned process groups are
323 * to receive a SIGHUP and a SIGCONT.
324 *
325 * "I ask you, have you ever known what it is to be an orphan?"
326 */
327static int will_become_orphaned_pgrp(struct pid *pgrp,
328 struct task_struct *ignored_task)
329{
330 struct task_struct *p;
331
332 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
333 if ((p == ignored_task) ||
334 (p->exit_state && thread_group_empty(p)) ||
335 is_global_init(p->real_parent))
336 continue;
337
338 if (task_pgrp(p->real_parent) != pgrp &&
339 task_session(p->real_parent) == task_session(p))
340 return 0;
341 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
342
343 return 1;
344}
345
346int is_current_pgrp_orphaned(void)
347{
348 int retval;
349
350 read_lock(&tasklist_lock);
351 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
352 read_unlock(&tasklist_lock);
353
354 return retval;
355}
356
357static bool has_stopped_jobs(struct pid *pgrp)
358{
359 struct task_struct *p;
360
361 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
362 if (p->signal->flags & SIGNAL_STOP_STOPPED)
363 return true;
364 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
365
366 return false;
367}
368
369/*
370 * Check to see if any process groups have become orphaned as
371 * a result of our exiting, and if they have any stopped jobs,
372 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
373 */
374static void
375kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
376{
377 struct pid *pgrp = task_pgrp(tsk);
378 struct task_struct *ignored_task = tsk;
379
380 if (!parent)
381 /* exit: our father is in a different pgrp than
382 * we are and we were the only connection outside.
383 */
384 parent = tsk->real_parent;
385 else
386 /* reparent: our child is in a different pgrp than
387 * we are, and it was the only connection outside.
388 */
389 ignored_task = NULL;
390
391 if (task_pgrp(parent) != pgrp &&
392 task_session(parent) == task_session(tsk) &&
393 will_become_orphaned_pgrp(pgrp, ignored_task) &&
394 has_stopped_jobs(pgrp)) {
395 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
396 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
397 }
398}
399
400static void coredump_task_exit(struct task_struct *tsk)
401{
402 struct core_state *core_state;
403
404 /*
405 * Serialize with any possible pending coredump.
406 * We must hold siglock around checking core_state
407 * and setting PF_POSTCOREDUMP. The core-inducing thread
408 * will increment ->nr_threads for each thread in the
409 * group without PF_POSTCOREDUMP set.
410 */
411 spin_lock_irq(&tsk->sighand->siglock);
412 tsk->flags |= PF_POSTCOREDUMP;
413 core_state = tsk->signal->core_state;
414 spin_unlock_irq(&tsk->sighand->siglock);
415 if (core_state) {
416 struct core_thread self;
417
418 self.task = current;
419 if (self.task->flags & PF_SIGNALED)
420 self.next = xchg(&core_state->dumper.next, &self);
421 else
422 self.task = NULL;
423 /*
424 * Implies mb(), the result of xchg() must be visible
425 * to core_state->dumper.
426 */
427 if (atomic_dec_and_test(&core_state->nr_threads))
428 complete(&core_state->startup);
429
430 for (;;) {
431 set_current_state(TASK_IDLE|TASK_FREEZABLE);
432 if (!self.task) /* see coredump_finish() */
433 break;
434 schedule();
435 }
436 __set_current_state(TASK_RUNNING);
437 }
438}
439
440#ifdef CONFIG_MEMCG
441/* drops tasklist_lock if succeeds */
442static bool __try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm)
443{
444 bool ret = false;
445
446 task_lock(tsk);
447 if (likely(tsk->mm == mm)) {
448 /* tsk can't pass exit_mm/exec_mmap and exit */
449 read_unlock(&tasklist_lock);
450 WRITE_ONCE(mm->owner, tsk);
451 lru_gen_migrate_mm(mm);
452 ret = true;
453 }
454 task_unlock(tsk);
455 return ret;
456}
457
458static bool try_to_set_owner(struct task_struct *g, struct mm_struct *mm)
459{
460 struct task_struct *t;
461
462 for_each_thread(g, t) {
463 struct mm_struct *t_mm = READ_ONCE(t->mm);
464 if (t_mm == mm) {
465 if (__try_to_set_owner(t, mm))
466 return true;
467 } else if (t_mm)
468 break;
469 }
470
471 return false;
472}
473
474/*
475 * A task is exiting. If it owned this mm, find a new owner for the mm.
476 */
477void mm_update_next_owner(struct mm_struct *mm)
478{
479 struct task_struct *g, *p = current;
480
481 /*
482 * If the exiting or execing task is not the owner, it's
483 * someone else's problem.
484 */
485 if (mm->owner != p)
486 return;
487 /*
488 * The current owner is exiting/execing and there are no other
489 * candidates. Do not leave the mm pointing to a possibly
490 * freed task structure.
491 */
492 if (atomic_read(&mm->mm_users) <= 1) {
493 WRITE_ONCE(mm->owner, NULL);
494 return;
495 }
496
497 read_lock(&tasklist_lock);
498 /*
499 * Search in the children
500 */
501 list_for_each_entry(g, &p->children, sibling) {
502 if (try_to_set_owner(g, mm))
503 goto ret;
504 }
505 /*
506 * Search in the siblings
507 */
508 list_for_each_entry(g, &p->real_parent->children, sibling) {
509 if (try_to_set_owner(g, mm))
510 goto ret;
511 }
512 /*
513 * Search through everything else, we should not get here often.
514 */
515 for_each_process(g) {
516 if (atomic_read(&mm->mm_users) <= 1)
517 break;
518 if (g->flags & PF_KTHREAD)
519 continue;
520 if (try_to_set_owner(g, mm))
521 goto ret;
522 }
523 read_unlock(&tasklist_lock);
524 /*
525 * We found no owner yet mm_users > 1: this implies that we are
526 * most likely racing with swapoff (try_to_unuse()) or /proc or
527 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
528 */
529 WRITE_ONCE(mm->owner, NULL);
530 ret:
531 return;
532
533}
534#endif /* CONFIG_MEMCG */
535
536/*
537 * Turn us into a lazy TLB process if we
538 * aren't already..
539 */
540static void exit_mm(void)
541{
542 struct mm_struct *mm = current->mm;
543
544 exit_mm_release(current, mm);
545 if (!mm)
546 return;
547 mmap_read_lock(mm);
548 mmgrab_lazy_tlb(mm);
549 BUG_ON(mm != current->active_mm);
550 /* more a memory barrier than a real lock */
551 task_lock(current);
552 /*
553 * When a thread stops operating on an address space, the loop
554 * in membarrier_private_expedited() may not observe that
555 * tsk->mm, and the loop in membarrier_global_expedited() may
556 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
557 * rq->membarrier_state, so those would not issue an IPI.
558 * Membarrier requires a memory barrier after accessing
559 * user-space memory, before clearing tsk->mm or the
560 * rq->membarrier_state.
561 */
562 smp_mb__after_spinlock();
563 local_irq_disable();
564 current->mm = NULL;
565 membarrier_update_current_mm(NULL);
566 enter_lazy_tlb(mm, current);
567 local_irq_enable();
568 task_unlock(current);
569 mmap_read_unlock(mm);
570 mm_update_next_owner(mm);
571 mmput(mm);
572 if (test_thread_flag(TIF_MEMDIE))
573 exit_oom_victim();
574}
575
576static struct task_struct *find_alive_thread(struct task_struct *p)
577{
578 struct task_struct *t;
579
580 for_each_thread(p, t) {
581 if (!(t->flags & PF_EXITING))
582 return t;
583 }
584 return NULL;
585}
586
587static struct task_struct *find_child_reaper(struct task_struct *father,
588 struct list_head *dead)
589 __releases(&tasklist_lock)
590 __acquires(&tasklist_lock)
591{
592 struct pid_namespace *pid_ns = task_active_pid_ns(father);
593 struct task_struct *reaper = pid_ns->child_reaper;
594 struct task_struct *p, *n;
595
596 if (likely(reaper != father))
597 return reaper;
598
599 reaper = find_alive_thread(father);
600 if (reaper) {
601 pid_ns->child_reaper = reaper;
602 return reaper;
603 }
604
605 write_unlock_irq(&tasklist_lock);
606
607 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
608 list_del_init(&p->ptrace_entry);
609 release_task(p);
610 }
611
612 zap_pid_ns_processes(pid_ns);
613 write_lock_irq(&tasklist_lock);
614
615 return father;
616}
617
618/*
619 * When we die, we re-parent all our children, and try to:
620 * 1. give them to another thread in our thread group, if such a member exists
621 * 2. give it to the first ancestor process which prctl'd itself as a
622 * child_subreaper for its children (like a service manager)
623 * 3. give it to the init process (PID 1) in our pid namespace
624 */
625static struct task_struct *find_new_reaper(struct task_struct *father,
626 struct task_struct *child_reaper)
627{
628 struct task_struct *thread, *reaper;
629
630 thread = find_alive_thread(father);
631 if (thread)
632 return thread;
633
634 if (father->signal->has_child_subreaper) {
635 unsigned int ns_level = task_pid(father)->level;
636 /*
637 * Find the first ->is_child_subreaper ancestor in our pid_ns.
638 * We can't check reaper != child_reaper to ensure we do not
639 * cross the namespaces, the exiting parent could be injected
640 * by setns() + fork().
641 * We check pid->level, this is slightly more efficient than
642 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
643 */
644 for (reaper = father->real_parent;
645 task_pid(reaper)->level == ns_level;
646 reaper = reaper->real_parent) {
647 if (reaper == &init_task)
648 break;
649 if (!reaper->signal->is_child_subreaper)
650 continue;
651 thread = find_alive_thread(reaper);
652 if (thread)
653 return thread;
654 }
655 }
656
657 return child_reaper;
658}
659
660/*
661* Any that need to be release_task'd are put on the @dead list.
662 */
663static void reparent_leader(struct task_struct *father, struct task_struct *p,
664 struct list_head *dead)
665{
666 if (unlikely(p->exit_state == EXIT_DEAD))
667 return;
668
669 /* We don't want people slaying init. */
670 p->exit_signal = SIGCHLD;
671
672 /* If it has exited notify the new parent about this child's death. */
673 if (!p->ptrace &&
674 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
675 if (do_notify_parent(p, p->exit_signal)) {
676 p->exit_state = EXIT_DEAD;
677 list_add(&p->ptrace_entry, dead);
678 }
679 }
680
681 kill_orphaned_pgrp(p, father);
682}
683
684/*
685 * This does two things:
686 *
687 * A. Make init inherit all the child processes
688 * B. Check to see if any process groups have become orphaned
689 * as a result of our exiting, and if they have any stopped
690 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
691 */
692static void forget_original_parent(struct task_struct *father,
693 struct list_head *dead)
694{
695 struct task_struct *p, *t, *reaper;
696
697 if (unlikely(!list_empty(&father->ptraced)))
698 exit_ptrace(father, dead);
699
700 /* Can drop and reacquire tasklist_lock */
701 reaper = find_child_reaper(father, dead);
702 if (list_empty(&father->children))
703 return;
704
705 reaper = find_new_reaper(father, reaper);
706 list_for_each_entry(p, &father->children, sibling) {
707 for_each_thread(p, t) {
708 RCU_INIT_POINTER(t->real_parent, reaper);
709 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
710 if (likely(!t->ptrace))
711 t->parent = t->real_parent;
712 if (t->pdeath_signal)
713 group_send_sig_info(t->pdeath_signal,
714 SEND_SIG_NOINFO, t,
715 PIDTYPE_TGID);
716 }
717 /*
718 * If this is a threaded reparent there is no need to
719 * notify anyone anything has happened.
720 */
721 if (!same_thread_group(reaper, father))
722 reparent_leader(father, p, dead);
723 }
724 list_splice_tail_init(&father->children, &reaper->children);
725}
726
727/*
728 * Send signals to all our closest relatives so that they know
729 * to properly mourn us..
730 */
731static void exit_notify(struct task_struct *tsk, int group_dead)
732{
733 bool autoreap;
734 struct task_struct *p, *n;
735 LIST_HEAD(dead);
736
737 write_lock_irq(&tasklist_lock);
738 forget_original_parent(tsk, &dead);
739
740 if (group_dead)
741 kill_orphaned_pgrp(tsk->group_leader, NULL);
742
743 tsk->exit_state = EXIT_ZOMBIE;
744 /*
745 * sub-thread or delay_group_leader(), wake up the
746 * PIDFD_THREAD waiters.
747 */
748 if (!thread_group_empty(tsk))
749 do_notify_pidfd(tsk);
750
751 if (unlikely(tsk->ptrace)) {
752 int sig = thread_group_leader(tsk) &&
753 thread_group_empty(tsk) &&
754 !ptrace_reparented(tsk) ?
755 tsk->exit_signal : SIGCHLD;
756 autoreap = do_notify_parent(tsk, sig);
757 } else if (thread_group_leader(tsk)) {
758 autoreap = thread_group_empty(tsk) &&
759 do_notify_parent(tsk, tsk->exit_signal);
760 } else {
761 autoreap = true;
762 }
763
764 if (autoreap) {
765 tsk->exit_state = EXIT_DEAD;
766 list_add(&tsk->ptrace_entry, &dead);
767 }
768
769 /* mt-exec, de_thread() is waiting for group leader */
770 if (unlikely(tsk->signal->notify_count < 0))
771 wake_up_process(tsk->signal->group_exec_task);
772 write_unlock_irq(&tasklist_lock);
773
774 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
775 list_del_init(&p->ptrace_entry);
776 release_task(p);
777 }
778}
779
780#ifdef CONFIG_DEBUG_STACK_USAGE
781unsigned long stack_not_used(struct task_struct *p)
782{
783 unsigned long *n = end_of_stack(p);
784
785 do { /* Skip over canary */
786# ifdef CONFIG_STACK_GROWSUP
787 n--;
788# else
789 n++;
790# endif
791 } while (!*n);
792
793# ifdef CONFIG_STACK_GROWSUP
794 return (unsigned long)end_of_stack(p) - (unsigned long)n;
795# else
796 return (unsigned long)n - (unsigned long)end_of_stack(p);
797# endif
798}
799
800/* Count the maximum pages reached in kernel stacks */
801static inline void kstack_histogram(unsigned long used_stack)
802{
803#ifdef CONFIG_VM_EVENT_COUNTERS
804 if (used_stack <= 1024)
805 count_vm_event(KSTACK_1K);
806#if THREAD_SIZE > 1024
807 else if (used_stack <= 2048)
808 count_vm_event(KSTACK_2K);
809#endif
810#if THREAD_SIZE > 2048
811 else if (used_stack <= 4096)
812 count_vm_event(KSTACK_4K);
813#endif
814#if THREAD_SIZE > 4096
815 else if (used_stack <= 8192)
816 count_vm_event(KSTACK_8K);
817#endif
818#if THREAD_SIZE > 8192
819 else if (used_stack <= 16384)
820 count_vm_event(KSTACK_16K);
821#endif
822#if THREAD_SIZE > 16384
823 else if (used_stack <= 32768)
824 count_vm_event(KSTACK_32K);
825#endif
826#if THREAD_SIZE > 32768
827 else if (used_stack <= 65536)
828 count_vm_event(KSTACK_64K);
829#endif
830#if THREAD_SIZE > 65536
831 else
832 count_vm_event(KSTACK_REST);
833#endif
834#endif /* CONFIG_VM_EVENT_COUNTERS */
835}
836
837static void check_stack_usage(void)
838{
839 static DEFINE_SPINLOCK(low_water_lock);
840 static int lowest_to_date = THREAD_SIZE;
841 unsigned long free;
842
843 free = stack_not_used(current);
844 kstack_histogram(THREAD_SIZE - free);
845
846 if (free >= lowest_to_date)
847 return;
848
849 spin_lock(&low_water_lock);
850 if (free < lowest_to_date) {
851 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
852 current->comm, task_pid_nr(current), free);
853 lowest_to_date = free;
854 }
855 spin_unlock(&low_water_lock);
856}
857#else
858static inline void check_stack_usage(void) {}
859#endif
860
861static void synchronize_group_exit(struct task_struct *tsk, long code)
862{
863 struct sighand_struct *sighand = tsk->sighand;
864 struct signal_struct *signal = tsk->signal;
865
866 spin_lock_irq(&sighand->siglock);
867 signal->quick_threads--;
868 if ((signal->quick_threads == 0) &&
869 !(signal->flags & SIGNAL_GROUP_EXIT)) {
870 signal->flags = SIGNAL_GROUP_EXIT;
871 signal->group_exit_code = code;
872 signal->group_stop_count = 0;
873 }
874 spin_unlock_irq(&sighand->siglock);
875}
876
877void __noreturn do_exit(long code)
878{
879 struct task_struct *tsk = current;
880 int group_dead;
881
882 WARN_ON(irqs_disabled());
883
884 synchronize_group_exit(tsk, code);
885
886 WARN_ON(tsk->plug);
887
888 kcov_task_exit(tsk);
889 kmsan_task_exit(tsk);
890
891 coredump_task_exit(tsk);
892 ptrace_event(PTRACE_EVENT_EXIT, code);
893 user_events_exit(tsk);
894
895 io_uring_files_cancel();
896 exit_signals(tsk); /* sets PF_EXITING */
897
898 seccomp_filter_release(tsk);
899
900 acct_update_integrals(tsk);
901 group_dead = atomic_dec_and_test(&tsk->signal->live);
902 if (group_dead) {
903 /*
904 * If the last thread of global init has exited, panic
905 * immediately to get a useable coredump.
906 */
907 if (unlikely(is_global_init(tsk)))
908 panic("Attempted to kill init! exitcode=0x%08x\n",
909 tsk->signal->group_exit_code ?: (int)code);
910
911#ifdef CONFIG_POSIX_TIMERS
912 hrtimer_cancel(&tsk->signal->real_timer);
913 exit_itimers(tsk);
914#endif
915 if (tsk->mm)
916 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
917 }
918 acct_collect(code, group_dead);
919 if (group_dead)
920 tty_audit_exit();
921 audit_free(tsk);
922
923 tsk->exit_code = code;
924 taskstats_exit(tsk, group_dead);
925
926 exit_mm();
927
928 if (group_dead)
929 acct_process();
930 trace_sched_process_exit(tsk);
931
932 exit_sem(tsk);
933 exit_shm(tsk);
934 exit_files(tsk);
935 exit_fs(tsk);
936 if (group_dead)
937 disassociate_ctty(1);
938 exit_task_namespaces(tsk);
939 exit_task_work(tsk);
940 exit_thread(tsk);
941
942 /*
943 * Flush inherited counters to the parent - before the parent
944 * gets woken up by child-exit notifications.
945 *
946 * because of cgroup mode, must be called before cgroup_exit()
947 */
948 perf_event_exit_task(tsk);
949
950 sched_autogroup_exit_task(tsk);
951 cgroup_exit(tsk);
952
953 /*
954 * FIXME: do that only when needed, using sched_exit tracepoint
955 */
956 flush_ptrace_hw_breakpoint(tsk);
957
958 exit_tasks_rcu_start();
959 exit_notify(tsk, group_dead);
960 proc_exit_connector(tsk);
961 mpol_put_task_policy(tsk);
962#ifdef CONFIG_FUTEX
963 if (unlikely(current->pi_state_cache))
964 kfree(current->pi_state_cache);
965#endif
966 /*
967 * Make sure we are holding no locks:
968 */
969 debug_check_no_locks_held();
970
971 if (tsk->io_context)
972 exit_io_context(tsk);
973
974 if (tsk->splice_pipe)
975 free_pipe_info(tsk->splice_pipe);
976
977 if (tsk->task_frag.page)
978 put_page(tsk->task_frag.page);
979
980 exit_task_stack_account(tsk);
981
982 check_stack_usage();
983 preempt_disable();
984 if (tsk->nr_dirtied)
985 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
986 exit_rcu();
987 exit_tasks_rcu_finish();
988
989 lockdep_free_task(tsk);
990 do_task_dead();
991}
992
993void __noreturn make_task_dead(int signr)
994{
995 /*
996 * Take the task off the cpu after something catastrophic has
997 * happened.
998 *
999 * We can get here from a kernel oops, sometimes with preemption off.
1000 * Start by checking for critical errors.
1001 * Then fix up important state like USER_DS and preemption.
1002 * Then do everything else.
1003 */
1004 struct task_struct *tsk = current;
1005 unsigned int limit;
1006
1007 if (unlikely(in_interrupt()))
1008 panic("Aiee, killing interrupt handler!");
1009 if (unlikely(!tsk->pid))
1010 panic("Attempted to kill the idle task!");
1011
1012 if (unlikely(irqs_disabled())) {
1013 pr_info("note: %s[%d] exited with irqs disabled\n",
1014 current->comm, task_pid_nr(current));
1015 local_irq_enable();
1016 }
1017 if (unlikely(in_atomic())) {
1018 pr_info("note: %s[%d] exited with preempt_count %d\n",
1019 current->comm, task_pid_nr(current),
1020 preempt_count());
1021 preempt_count_set(PREEMPT_ENABLED);
1022 }
1023
1024 /*
1025 * Every time the system oopses, if the oops happens while a reference
1026 * to an object was held, the reference leaks.
1027 * If the oops doesn't also leak memory, repeated oopsing can cause
1028 * reference counters to wrap around (if they're not using refcount_t).
1029 * This means that repeated oopsing can make unexploitable-looking bugs
1030 * exploitable through repeated oopsing.
1031 * To make sure this can't happen, place an upper bound on how often the
1032 * kernel may oops without panic().
1033 */
1034 limit = READ_ONCE(oops_limit);
1035 if (atomic_inc_return(&oops_count) >= limit && limit)
1036 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
1037
1038 /*
1039 * We're taking recursive faults here in make_task_dead. Safest is to just
1040 * leave this task alone and wait for reboot.
1041 */
1042 if (unlikely(tsk->flags & PF_EXITING)) {
1043 pr_alert("Fixing recursive fault but reboot is needed!\n");
1044 futex_exit_recursive(tsk);
1045 tsk->exit_state = EXIT_DEAD;
1046 refcount_inc(&tsk->rcu_users);
1047 do_task_dead();
1048 }
1049
1050 do_exit(signr);
1051}
1052
1053SYSCALL_DEFINE1(exit, int, error_code)
1054{
1055 do_exit((error_code&0xff)<<8);
1056}
1057
1058/*
1059 * Take down every thread in the group. This is called by fatal signals
1060 * as well as by sys_exit_group (below).
1061 */
1062void __noreturn
1063do_group_exit(int exit_code)
1064{
1065 struct signal_struct *sig = current->signal;
1066
1067 if (sig->flags & SIGNAL_GROUP_EXIT)
1068 exit_code = sig->group_exit_code;
1069 else if (sig->group_exec_task)
1070 exit_code = 0;
1071 else {
1072 struct sighand_struct *const sighand = current->sighand;
1073
1074 spin_lock_irq(&sighand->siglock);
1075 if (sig->flags & SIGNAL_GROUP_EXIT)
1076 /* Another thread got here before we took the lock. */
1077 exit_code = sig->group_exit_code;
1078 else if (sig->group_exec_task)
1079 exit_code = 0;
1080 else {
1081 sig->group_exit_code = exit_code;
1082 sig->flags = SIGNAL_GROUP_EXIT;
1083 zap_other_threads(current);
1084 }
1085 spin_unlock_irq(&sighand->siglock);
1086 }
1087
1088 do_exit(exit_code);
1089 /* NOTREACHED */
1090}
1091
1092/*
1093 * this kills every thread in the thread group. Note that any externally
1094 * wait4()-ing process will get the correct exit code - even if this
1095 * thread is not the thread group leader.
1096 */
1097SYSCALL_DEFINE1(exit_group, int, error_code)
1098{
1099 do_group_exit((error_code & 0xff) << 8);
1100 /* NOTREACHED */
1101 return 0;
1102}
1103
1104static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1105{
1106 return wo->wo_type == PIDTYPE_MAX ||
1107 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1108}
1109
1110static int
1111eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1112{
1113 if (!eligible_pid(wo, p))
1114 return 0;
1115
1116 /*
1117 * Wait for all children (clone and not) if __WALL is set or
1118 * if it is traced by us.
1119 */
1120 if (ptrace || (wo->wo_flags & __WALL))
1121 return 1;
1122
1123 /*
1124 * Otherwise, wait for clone children *only* if __WCLONE is set;
1125 * otherwise, wait for non-clone children *only*.
1126 *
1127 * Note: a "clone" child here is one that reports to its parent
1128 * using a signal other than SIGCHLD, or a non-leader thread which
1129 * we can only see if it is traced by us.
1130 */
1131 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1132 return 0;
1133
1134 return 1;
1135}
1136
1137/*
1138 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1139 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1140 * the lock and this task is uninteresting. If we return nonzero, we have
1141 * released the lock and the system call should return.
1142 */
1143static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1144{
1145 int state, status;
1146 pid_t pid = task_pid_vnr(p);
1147 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1148 struct waitid_info *infop;
1149
1150 if (!likely(wo->wo_flags & WEXITED))
1151 return 0;
1152
1153 if (unlikely(wo->wo_flags & WNOWAIT)) {
1154 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1155 ? p->signal->group_exit_code : p->exit_code;
1156 get_task_struct(p);
1157 read_unlock(&tasklist_lock);
1158 sched_annotate_sleep();
1159 if (wo->wo_rusage)
1160 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1161 put_task_struct(p);
1162 goto out_info;
1163 }
1164 /*
1165 * Move the task's state to DEAD/TRACE, only one thread can do this.
1166 */
1167 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1168 EXIT_TRACE : EXIT_DEAD;
1169 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1170 return 0;
1171 /*
1172 * We own this thread, nobody else can reap it.
1173 */
1174 read_unlock(&tasklist_lock);
1175 sched_annotate_sleep();
1176
1177 /*
1178 * Check thread_group_leader() to exclude the traced sub-threads.
1179 */
1180 if (state == EXIT_DEAD && thread_group_leader(p)) {
1181 struct signal_struct *sig = p->signal;
1182 struct signal_struct *psig = current->signal;
1183 unsigned long maxrss;
1184 u64 tgutime, tgstime;
1185
1186 /*
1187 * The resource counters for the group leader are in its
1188 * own task_struct. Those for dead threads in the group
1189 * are in its signal_struct, as are those for the child
1190 * processes it has previously reaped. All these
1191 * accumulate in the parent's signal_struct c* fields.
1192 *
1193 * We don't bother to take a lock here to protect these
1194 * p->signal fields because the whole thread group is dead
1195 * and nobody can change them.
1196 *
1197 * psig->stats_lock also protects us from our sub-threads
1198 * which can reap other children at the same time.
1199 *
1200 * We use thread_group_cputime_adjusted() to get times for
1201 * the thread group, which consolidates times for all threads
1202 * in the group including the group leader.
1203 */
1204 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1205 write_seqlock_irq(&psig->stats_lock);
1206 psig->cutime += tgutime + sig->cutime;
1207 psig->cstime += tgstime + sig->cstime;
1208 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1209 psig->cmin_flt +=
1210 p->min_flt + sig->min_flt + sig->cmin_flt;
1211 psig->cmaj_flt +=
1212 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1213 psig->cnvcsw +=
1214 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1215 psig->cnivcsw +=
1216 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1217 psig->cinblock +=
1218 task_io_get_inblock(p) +
1219 sig->inblock + sig->cinblock;
1220 psig->coublock +=
1221 task_io_get_oublock(p) +
1222 sig->oublock + sig->coublock;
1223 maxrss = max(sig->maxrss, sig->cmaxrss);
1224 if (psig->cmaxrss < maxrss)
1225 psig->cmaxrss = maxrss;
1226 task_io_accounting_add(&psig->ioac, &p->ioac);
1227 task_io_accounting_add(&psig->ioac, &sig->ioac);
1228 write_sequnlock_irq(&psig->stats_lock);
1229 }
1230
1231 if (wo->wo_rusage)
1232 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1233 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1234 ? p->signal->group_exit_code : p->exit_code;
1235 wo->wo_stat = status;
1236
1237 if (state == EXIT_TRACE) {
1238 write_lock_irq(&tasklist_lock);
1239 /* We dropped tasklist, ptracer could die and untrace */
1240 ptrace_unlink(p);
1241
1242 /* If parent wants a zombie, don't release it now */
1243 state = EXIT_ZOMBIE;
1244 if (do_notify_parent(p, p->exit_signal))
1245 state = EXIT_DEAD;
1246 p->exit_state = state;
1247 write_unlock_irq(&tasklist_lock);
1248 }
1249 if (state == EXIT_DEAD)
1250 release_task(p);
1251
1252out_info:
1253 infop = wo->wo_info;
1254 if (infop) {
1255 if ((status & 0x7f) == 0) {
1256 infop->cause = CLD_EXITED;
1257 infop->status = status >> 8;
1258 } else {
1259 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1260 infop->status = status & 0x7f;
1261 }
1262 infop->pid = pid;
1263 infop->uid = uid;
1264 }
1265
1266 return pid;
1267}
1268
1269static int *task_stopped_code(struct task_struct *p, bool ptrace)
1270{
1271 if (ptrace) {
1272 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1273 return &p->exit_code;
1274 } else {
1275 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1276 return &p->signal->group_exit_code;
1277 }
1278 return NULL;
1279}
1280
1281/**
1282 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1283 * @wo: wait options
1284 * @ptrace: is the wait for ptrace
1285 * @p: task to wait for
1286 *
1287 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1288 *
1289 * CONTEXT:
1290 * read_lock(&tasklist_lock), which is released if return value is
1291 * non-zero. Also, grabs and releases @p->sighand->siglock.
1292 *
1293 * RETURNS:
1294 * 0 if wait condition didn't exist and search for other wait conditions
1295 * should continue. Non-zero return, -errno on failure and @p's pid on
1296 * success, implies that tasklist_lock is released and wait condition
1297 * search should terminate.
1298 */
1299static int wait_task_stopped(struct wait_opts *wo,
1300 int ptrace, struct task_struct *p)
1301{
1302 struct waitid_info *infop;
1303 int exit_code, *p_code, why;
1304 uid_t uid = 0; /* unneeded, required by compiler */
1305 pid_t pid;
1306
1307 /*
1308 * Traditionally we see ptrace'd stopped tasks regardless of options.
1309 */
1310 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1311 return 0;
1312
1313 if (!task_stopped_code(p, ptrace))
1314 return 0;
1315
1316 exit_code = 0;
1317 spin_lock_irq(&p->sighand->siglock);
1318
1319 p_code = task_stopped_code(p, ptrace);
1320 if (unlikely(!p_code))
1321 goto unlock_sig;
1322
1323 exit_code = *p_code;
1324 if (!exit_code)
1325 goto unlock_sig;
1326
1327 if (!unlikely(wo->wo_flags & WNOWAIT))
1328 *p_code = 0;
1329
1330 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1331unlock_sig:
1332 spin_unlock_irq(&p->sighand->siglock);
1333 if (!exit_code)
1334 return 0;
1335
1336 /*
1337 * Now we are pretty sure this task is interesting.
1338 * Make sure it doesn't get reaped out from under us while we
1339 * give up the lock and then examine it below. We don't want to
1340 * keep holding onto the tasklist_lock while we call getrusage and
1341 * possibly take page faults for user memory.
1342 */
1343 get_task_struct(p);
1344 pid = task_pid_vnr(p);
1345 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1346 read_unlock(&tasklist_lock);
1347 sched_annotate_sleep();
1348 if (wo->wo_rusage)
1349 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1350 put_task_struct(p);
1351
1352 if (likely(!(wo->wo_flags & WNOWAIT)))
1353 wo->wo_stat = (exit_code << 8) | 0x7f;
1354
1355 infop = wo->wo_info;
1356 if (infop) {
1357 infop->cause = why;
1358 infop->status = exit_code;
1359 infop->pid = pid;
1360 infop->uid = uid;
1361 }
1362 return pid;
1363}
1364
1365/*
1366 * Handle do_wait work for one task in a live, non-stopped state.
1367 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1368 * the lock and this task is uninteresting. If we return nonzero, we have
1369 * released the lock and the system call should return.
1370 */
1371static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1372{
1373 struct waitid_info *infop;
1374 pid_t pid;
1375 uid_t uid;
1376
1377 if (!unlikely(wo->wo_flags & WCONTINUED))
1378 return 0;
1379
1380 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1381 return 0;
1382
1383 spin_lock_irq(&p->sighand->siglock);
1384 /* Re-check with the lock held. */
1385 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1386 spin_unlock_irq(&p->sighand->siglock);
1387 return 0;
1388 }
1389 if (!unlikely(wo->wo_flags & WNOWAIT))
1390 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1391 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1392 spin_unlock_irq(&p->sighand->siglock);
1393
1394 pid = task_pid_vnr(p);
1395 get_task_struct(p);
1396 read_unlock(&tasklist_lock);
1397 sched_annotate_sleep();
1398 if (wo->wo_rusage)
1399 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1400 put_task_struct(p);
1401
1402 infop = wo->wo_info;
1403 if (!infop) {
1404 wo->wo_stat = 0xffff;
1405 } else {
1406 infop->cause = CLD_CONTINUED;
1407 infop->pid = pid;
1408 infop->uid = uid;
1409 infop->status = SIGCONT;
1410 }
1411 return pid;
1412}
1413
1414/*
1415 * Consider @p for a wait by @parent.
1416 *
1417 * -ECHILD should be in ->notask_error before the first call.
1418 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1419 * Returns zero if the search for a child should continue;
1420 * then ->notask_error is 0 if @p is an eligible child,
1421 * or still -ECHILD.
1422 */
1423static int wait_consider_task(struct wait_opts *wo, int ptrace,
1424 struct task_struct *p)
1425{
1426 /*
1427 * We can race with wait_task_zombie() from another thread.
1428 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1429 * can't confuse the checks below.
1430 */
1431 int exit_state = READ_ONCE(p->exit_state);
1432 int ret;
1433
1434 if (unlikely(exit_state == EXIT_DEAD))
1435 return 0;
1436
1437 ret = eligible_child(wo, ptrace, p);
1438 if (!ret)
1439 return ret;
1440
1441 if (unlikely(exit_state == EXIT_TRACE)) {
1442 /*
1443 * ptrace == 0 means we are the natural parent. In this case
1444 * we should clear notask_error, debugger will notify us.
1445 */
1446 if (likely(!ptrace))
1447 wo->notask_error = 0;
1448 return 0;
1449 }
1450
1451 if (likely(!ptrace) && unlikely(p->ptrace)) {
1452 /*
1453 * If it is traced by its real parent's group, just pretend
1454 * the caller is ptrace_do_wait() and reap this child if it
1455 * is zombie.
1456 *
1457 * This also hides group stop state from real parent; otherwise
1458 * a single stop can be reported twice as group and ptrace stop.
1459 * If a ptracer wants to distinguish these two events for its
1460 * own children it should create a separate process which takes
1461 * the role of real parent.
1462 */
1463 if (!ptrace_reparented(p))
1464 ptrace = 1;
1465 }
1466
1467 /* slay zombie? */
1468 if (exit_state == EXIT_ZOMBIE) {
1469 /* we don't reap group leaders with subthreads */
1470 if (!delay_group_leader(p)) {
1471 /*
1472 * A zombie ptracee is only visible to its ptracer.
1473 * Notification and reaping will be cascaded to the
1474 * real parent when the ptracer detaches.
1475 */
1476 if (unlikely(ptrace) || likely(!p->ptrace))
1477 return wait_task_zombie(wo, p);
1478 }
1479
1480 /*
1481 * Allow access to stopped/continued state via zombie by
1482 * falling through. Clearing of notask_error is complex.
1483 *
1484 * When !@ptrace:
1485 *
1486 * If WEXITED is set, notask_error should naturally be
1487 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1488 * so, if there are live subthreads, there are events to
1489 * wait for. If all subthreads are dead, it's still safe
1490 * to clear - this function will be called again in finite
1491 * amount time once all the subthreads are released and
1492 * will then return without clearing.
1493 *
1494 * When @ptrace:
1495 *
1496 * Stopped state is per-task and thus can't change once the
1497 * target task dies. Only continued and exited can happen.
1498 * Clear notask_error if WCONTINUED | WEXITED.
1499 */
1500 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1501 wo->notask_error = 0;
1502 } else {
1503 /*
1504 * @p is alive and it's gonna stop, continue or exit, so
1505 * there always is something to wait for.
1506 */
1507 wo->notask_error = 0;
1508 }
1509
1510 /*
1511 * Wait for stopped. Depending on @ptrace, different stopped state
1512 * is used and the two don't interact with each other.
1513 */
1514 ret = wait_task_stopped(wo, ptrace, p);
1515 if (ret)
1516 return ret;
1517
1518 /*
1519 * Wait for continued. There's only one continued state and the
1520 * ptracer can consume it which can confuse the real parent. Don't
1521 * use WCONTINUED from ptracer. You don't need or want it.
1522 */
1523 return wait_task_continued(wo, p);
1524}
1525
1526/*
1527 * Do the work of do_wait() for one thread in the group, @tsk.
1528 *
1529 * -ECHILD should be in ->notask_error before the first call.
1530 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1531 * Returns zero if the search for a child should continue; then
1532 * ->notask_error is 0 if there were any eligible children,
1533 * or still -ECHILD.
1534 */
1535static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1536{
1537 struct task_struct *p;
1538
1539 list_for_each_entry(p, &tsk->children, sibling) {
1540 int ret = wait_consider_task(wo, 0, p);
1541
1542 if (ret)
1543 return ret;
1544 }
1545
1546 return 0;
1547}
1548
1549static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1550{
1551 struct task_struct *p;
1552
1553 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1554 int ret = wait_consider_task(wo, 1, p);
1555
1556 if (ret)
1557 return ret;
1558 }
1559
1560 return 0;
1561}
1562
1563bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1564{
1565 if (!eligible_pid(wo, p))
1566 return false;
1567
1568 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1569 return false;
1570
1571 return true;
1572}
1573
1574static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1575 int sync, void *key)
1576{
1577 struct wait_opts *wo = container_of(wait, struct wait_opts,
1578 child_wait);
1579 struct task_struct *p = key;
1580
1581 if (pid_child_should_wake(wo, p))
1582 return default_wake_function(wait, mode, sync, key);
1583
1584 return 0;
1585}
1586
1587void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1588{
1589 __wake_up_sync_key(&parent->signal->wait_chldexit,
1590 TASK_INTERRUPTIBLE, p);
1591}
1592
1593static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1594 struct task_struct *target)
1595{
1596 struct task_struct *parent =
1597 !ptrace ? target->real_parent : target->parent;
1598
1599 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1600 same_thread_group(current, parent));
1601}
1602
1603/*
1604 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1605 * and tracee lists to find the target task.
1606 */
1607static int do_wait_pid(struct wait_opts *wo)
1608{
1609 bool ptrace;
1610 struct task_struct *target;
1611 int retval;
1612
1613 ptrace = false;
1614 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1615 if (target && is_effectively_child(wo, ptrace, target)) {
1616 retval = wait_consider_task(wo, ptrace, target);
1617 if (retval)
1618 return retval;
1619 }
1620
1621 ptrace = true;
1622 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1623 if (target && target->ptrace &&
1624 is_effectively_child(wo, ptrace, target)) {
1625 retval = wait_consider_task(wo, ptrace, target);
1626 if (retval)
1627 return retval;
1628 }
1629
1630 return 0;
1631}
1632
1633long __do_wait(struct wait_opts *wo)
1634{
1635 long retval;
1636
1637 /*
1638 * If there is nothing that can match our criteria, just get out.
1639 * We will clear ->notask_error to zero if we see any child that
1640 * might later match our criteria, even if we are not able to reap
1641 * it yet.
1642 */
1643 wo->notask_error = -ECHILD;
1644 if ((wo->wo_type < PIDTYPE_MAX) &&
1645 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1646 goto notask;
1647
1648 read_lock(&tasklist_lock);
1649
1650 if (wo->wo_type == PIDTYPE_PID) {
1651 retval = do_wait_pid(wo);
1652 if (retval)
1653 return retval;
1654 } else {
1655 struct task_struct *tsk = current;
1656
1657 do {
1658 retval = do_wait_thread(wo, tsk);
1659 if (retval)
1660 return retval;
1661
1662 retval = ptrace_do_wait(wo, tsk);
1663 if (retval)
1664 return retval;
1665
1666 if (wo->wo_flags & __WNOTHREAD)
1667 break;
1668 } while_each_thread(current, tsk);
1669 }
1670 read_unlock(&tasklist_lock);
1671
1672notask:
1673 retval = wo->notask_error;
1674 if (!retval && !(wo->wo_flags & WNOHANG))
1675 return -ERESTARTSYS;
1676
1677 return retval;
1678}
1679
1680static long do_wait(struct wait_opts *wo)
1681{
1682 int retval;
1683
1684 trace_sched_process_wait(wo->wo_pid);
1685
1686 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1687 wo->child_wait.private = current;
1688 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1689
1690 do {
1691 set_current_state(TASK_INTERRUPTIBLE);
1692 retval = __do_wait(wo);
1693 if (retval != -ERESTARTSYS)
1694 break;
1695 if (signal_pending(current))
1696 break;
1697 schedule();
1698 } while (1);
1699
1700 __set_current_state(TASK_RUNNING);
1701 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1702 return retval;
1703}
1704
1705int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1706 struct waitid_info *infop, int options,
1707 struct rusage *ru)
1708{
1709 unsigned int f_flags = 0;
1710 struct pid *pid = NULL;
1711 enum pid_type type;
1712
1713 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1714 __WNOTHREAD|__WCLONE|__WALL))
1715 return -EINVAL;
1716 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1717 return -EINVAL;
1718
1719 switch (which) {
1720 case P_ALL:
1721 type = PIDTYPE_MAX;
1722 break;
1723 case P_PID:
1724 type = PIDTYPE_PID;
1725 if (upid <= 0)
1726 return -EINVAL;
1727
1728 pid = find_get_pid(upid);
1729 break;
1730 case P_PGID:
1731 type = PIDTYPE_PGID;
1732 if (upid < 0)
1733 return -EINVAL;
1734
1735 if (upid)
1736 pid = find_get_pid(upid);
1737 else
1738 pid = get_task_pid(current, PIDTYPE_PGID);
1739 break;
1740 case P_PIDFD:
1741 type = PIDTYPE_PID;
1742 if (upid < 0)
1743 return -EINVAL;
1744
1745 pid = pidfd_get_pid(upid, &f_flags);
1746 if (IS_ERR(pid))
1747 return PTR_ERR(pid);
1748
1749 break;
1750 default:
1751 return -EINVAL;
1752 }
1753
1754 wo->wo_type = type;
1755 wo->wo_pid = pid;
1756 wo->wo_flags = options;
1757 wo->wo_info = infop;
1758 wo->wo_rusage = ru;
1759 if (f_flags & O_NONBLOCK)
1760 wo->wo_flags |= WNOHANG;
1761
1762 return 0;
1763}
1764
1765static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1766 int options, struct rusage *ru)
1767{
1768 struct wait_opts wo;
1769 long ret;
1770
1771 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1772 if (ret)
1773 return ret;
1774
1775 ret = do_wait(&wo);
1776 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1777 ret = -EAGAIN;
1778
1779 put_pid(wo.wo_pid);
1780 return ret;
1781}
1782
1783SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1784 infop, int, options, struct rusage __user *, ru)
1785{
1786 struct rusage r;
1787 struct waitid_info info = {.status = 0};
1788 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1789 int signo = 0;
1790
1791 if (err > 0) {
1792 signo = SIGCHLD;
1793 err = 0;
1794 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1795 return -EFAULT;
1796 }
1797 if (!infop)
1798 return err;
1799
1800 if (!user_write_access_begin(infop, sizeof(*infop)))
1801 return -EFAULT;
1802
1803 unsafe_put_user(signo, &infop->si_signo, Efault);
1804 unsafe_put_user(0, &infop->si_errno, Efault);
1805 unsafe_put_user(info.cause, &infop->si_code, Efault);
1806 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1807 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1808 unsafe_put_user(info.status, &infop->si_status, Efault);
1809 user_write_access_end();
1810 return err;
1811Efault:
1812 user_write_access_end();
1813 return -EFAULT;
1814}
1815
1816long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1817 struct rusage *ru)
1818{
1819 struct wait_opts wo;
1820 struct pid *pid = NULL;
1821 enum pid_type type;
1822 long ret;
1823
1824 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1825 __WNOTHREAD|__WCLONE|__WALL))
1826 return -EINVAL;
1827
1828 /* -INT_MIN is not defined */
1829 if (upid == INT_MIN)
1830 return -ESRCH;
1831
1832 if (upid == -1)
1833 type = PIDTYPE_MAX;
1834 else if (upid < 0) {
1835 type = PIDTYPE_PGID;
1836 pid = find_get_pid(-upid);
1837 } else if (upid == 0) {
1838 type = PIDTYPE_PGID;
1839 pid = get_task_pid(current, PIDTYPE_PGID);
1840 } else /* upid > 0 */ {
1841 type = PIDTYPE_PID;
1842 pid = find_get_pid(upid);
1843 }
1844
1845 wo.wo_type = type;
1846 wo.wo_pid = pid;
1847 wo.wo_flags = options | WEXITED;
1848 wo.wo_info = NULL;
1849 wo.wo_stat = 0;
1850 wo.wo_rusage = ru;
1851 ret = do_wait(&wo);
1852 put_pid(pid);
1853 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1854 ret = -EFAULT;
1855
1856 return ret;
1857}
1858
1859int kernel_wait(pid_t pid, int *stat)
1860{
1861 struct wait_opts wo = {
1862 .wo_type = PIDTYPE_PID,
1863 .wo_pid = find_get_pid(pid),
1864 .wo_flags = WEXITED,
1865 };
1866 int ret;
1867
1868 ret = do_wait(&wo);
1869 if (ret > 0 && wo.wo_stat)
1870 *stat = wo.wo_stat;
1871 put_pid(wo.wo_pid);
1872 return ret;
1873}
1874
1875SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1876 int, options, struct rusage __user *, ru)
1877{
1878 struct rusage r;
1879 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1880
1881 if (err > 0) {
1882 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1883 return -EFAULT;
1884 }
1885 return err;
1886}
1887
1888#ifdef __ARCH_WANT_SYS_WAITPID
1889
1890/*
1891 * sys_waitpid() remains for compatibility. waitpid() should be
1892 * implemented by calling sys_wait4() from libc.a.
1893 */
1894SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1895{
1896 return kernel_wait4(pid, stat_addr, options, NULL);
1897}
1898
1899#endif
1900
1901#ifdef CONFIG_COMPAT
1902COMPAT_SYSCALL_DEFINE4(wait4,
1903 compat_pid_t, pid,
1904 compat_uint_t __user *, stat_addr,
1905 int, options,
1906 struct compat_rusage __user *, ru)
1907{
1908 struct rusage r;
1909 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1910 if (err > 0) {
1911 if (ru && put_compat_rusage(&r, ru))
1912 return -EFAULT;
1913 }
1914 return err;
1915}
1916
1917COMPAT_SYSCALL_DEFINE5(waitid,
1918 int, which, compat_pid_t, pid,
1919 struct compat_siginfo __user *, infop, int, options,
1920 struct compat_rusage __user *, uru)
1921{
1922 struct rusage ru;
1923 struct waitid_info info = {.status = 0};
1924 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1925 int signo = 0;
1926 if (err > 0) {
1927 signo = SIGCHLD;
1928 err = 0;
1929 if (uru) {
1930 /* kernel_waitid() overwrites everything in ru */
1931 if (COMPAT_USE_64BIT_TIME)
1932 err = copy_to_user(uru, &ru, sizeof(ru));
1933 else
1934 err = put_compat_rusage(&ru, uru);
1935 if (err)
1936 return -EFAULT;
1937 }
1938 }
1939
1940 if (!infop)
1941 return err;
1942
1943 if (!user_write_access_begin(infop, sizeof(*infop)))
1944 return -EFAULT;
1945
1946 unsafe_put_user(signo, &infop->si_signo, Efault);
1947 unsafe_put_user(0, &infop->si_errno, Efault);
1948 unsafe_put_user(info.cause, &infop->si_code, Efault);
1949 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1950 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1951 unsafe_put_user(info.status, &infop->si_status, Efault);
1952 user_write_access_end();
1953 return err;
1954Efault:
1955 user_write_access_end();
1956 return -EFAULT;
1957}
1958#endif
1959
1960/*
1961 * This needs to be __function_aligned as GCC implicitly makes any
1962 * implementation of abort() cold and drops alignment specified by
1963 * -falign-functions=N.
1964 *
1965 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1966 */
1967__weak __function_aligned void abort(void)
1968{
1969 BUG();
1970
1971 /* if that doesn't kill us, halt */
1972 panic("Oops failed to kill thread");
1973}
1974EXPORT_SYMBOL(abort);