projects/linux·blob v6.10 · refs

1911 lines · 47.3 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 seccomp_filter_release(p);
281 proc_flush_pid(thread_pid);
282 put_pid(thread_pid);
283 release_thread(p);
284 put_task_struct_rcu_user(p);
285
286 p = leader;
287 if (unlikely(zap_leader))
288 goto repeat;
289}
290
291int rcuwait_wake_up(struct rcuwait *w)
292{
293 int ret = 0;
294 struct task_struct *task;
295
296 rcu_read_lock();
297
298 /*
299 * Order condition vs @task, such that everything prior to the load
300 * of @task is visible. This is the condition as to why the user called
301 * rcuwait_wake() in the first place. Pairs with set_current_state()
302 * barrier (A) in rcuwait_wait_event().
303 *
304 * WAIT WAKE
305 * [S] tsk = current [S] cond = true
306 * MB (A) MB (B)
307 * [L] cond [L] tsk
308 */
309 smp_mb(); /* (B) */
310
311 task = rcu_dereference(w->task);
312 if (task)
313 ret = wake_up_process(task);
314 rcu_read_unlock();
315
316 return ret;
317}
318EXPORT_SYMBOL_GPL(rcuwait_wake_up);
319
320/*
321 * Determine if a process group is "orphaned", according to the POSIX
322 * definition in 2.2.2.52. Orphaned process groups are not to be affected
323 * by terminal-generated stop signals. Newly orphaned process groups are
324 * to receive a SIGHUP and a SIGCONT.
325 *
326 * "I ask you, have you ever known what it is to be an orphan?"
327 */
328static int will_become_orphaned_pgrp(struct pid *pgrp,
329 struct task_struct *ignored_task)
330{
331 struct task_struct *p;
332
333 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
334 if ((p == ignored_task) ||
335 (p->exit_state && thread_group_empty(p)) ||
336 is_global_init(p->real_parent))
337 continue;
338
339 if (task_pgrp(p->real_parent) != pgrp &&
340 task_session(p->real_parent) == task_session(p))
341 return 0;
342 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
343
344 return 1;
345}
346
347int is_current_pgrp_orphaned(void)
348{
349 int retval;
350
351 read_lock(&tasklist_lock);
352 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
353 read_unlock(&tasklist_lock);
354
355 return retval;
356}
357
358static bool has_stopped_jobs(struct pid *pgrp)
359{
360 struct task_struct *p;
361
362 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
363 if (p->signal->flags & SIGNAL_STOP_STOPPED)
364 return true;
365 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
366
367 return false;
368}
369
370/*
371 * Check to see if any process groups have become orphaned as
372 * a result of our exiting, and if they have any stopped jobs,
373 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
374 */
375static void
376kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
377{
378 struct pid *pgrp = task_pgrp(tsk);
379 struct task_struct *ignored_task = tsk;
380
381 if (!parent)
382 /* exit: our father is in a different pgrp than
383 * we are and we were the only connection outside.
384 */
385 parent = tsk->real_parent;
386 else
387 /* reparent: our child is in a different pgrp than
388 * we are, and it was the only connection outside.
389 */
390 ignored_task = NULL;
391
392 if (task_pgrp(parent) != pgrp &&
393 task_session(parent) == task_session(tsk) &&
394 will_become_orphaned_pgrp(pgrp, ignored_task) &&
395 has_stopped_jobs(pgrp)) {
396 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
397 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
398 }
399}
400
401static void coredump_task_exit(struct task_struct *tsk)
402{
403 struct core_state *core_state;
404
405 /*
406 * Serialize with any possible pending coredump.
407 * We must hold siglock around checking core_state
408 * and setting PF_POSTCOREDUMP. The core-inducing thread
409 * will increment ->nr_threads for each thread in the
410 * group without PF_POSTCOREDUMP set.
411 */
412 spin_lock_irq(&tsk->sighand->siglock);
413 tsk->flags |= PF_POSTCOREDUMP;
414 core_state = tsk->signal->core_state;
415 spin_unlock_irq(&tsk->sighand->siglock);
416 if (core_state) {
417 struct core_thread self;
418
419 self.task = current;
420 if (self.task->flags & PF_SIGNALED)
421 self.next = xchg(&core_state->dumper.next, &self);
422 else
423 self.task = NULL;
424 /*
425 * Implies mb(), the result of xchg() must be visible
426 * to core_state->dumper.
427 */
428 if (atomic_dec_and_test(&core_state->nr_threads))
429 complete(&core_state->startup);
430
431 for (;;) {
432 set_current_state(TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
433 if (!self.task) /* see coredump_finish() */
434 break;
435 schedule();
436 }
437 __set_current_state(TASK_RUNNING);
438 }
439}
440
441#ifdef CONFIG_MEMCG
442/*
443 * A task is exiting. If it owned this mm, find a new owner for the mm.
444 */
445void mm_update_next_owner(struct mm_struct *mm)
446{
447 struct task_struct *c, *g, *p = current;
448
449retry:
450 /*
451 * If the exiting or execing task is not the owner, it's
452 * someone else's problem.
453 */
454 if (mm->owner != p)
455 return;
456 /*
457 * The current owner is exiting/execing and there are no other
458 * candidates. Do not leave the mm pointing to a possibly
459 * freed task structure.
460 */
461 if (atomic_read(&mm->mm_users) <= 1) {
462 WRITE_ONCE(mm->owner, NULL);
463 return;
464 }
465
466 read_lock(&tasklist_lock);
467 /*
468 * Search in the children
469 */
470 list_for_each_entry(c, &p->children, sibling) {
471 if (c->mm == mm)
472 goto assign_new_owner;
473 }
474
475 /*
476 * Search in the siblings
477 */
478 list_for_each_entry(c, &p->real_parent->children, sibling) {
479 if (c->mm == mm)
480 goto assign_new_owner;
481 }
482
483 /*
484 * Search through everything else, we should not get here often.
485 */
486 for_each_process(g) {
487 if (atomic_read(&mm->mm_users) <= 1)
488 break;
489 if (g->flags & PF_KTHREAD)
490 continue;
491 for_each_thread(g, c) {
492 if (c->mm == mm)
493 goto assign_new_owner;
494 if (c->mm)
495 break;
496 }
497 }
498 read_unlock(&tasklist_lock);
499 /*
500 * We found no owner yet mm_users > 1: this implies that we are
501 * most likely racing with swapoff (try_to_unuse()) or /proc or
502 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
503 */
504 WRITE_ONCE(mm->owner, NULL);
505 return;
506
507assign_new_owner:
508 BUG_ON(c == p);
509 get_task_struct(c);
510 /*
511 * The task_lock protects c->mm from changing.
512 * We always want mm->owner->mm == mm
513 */
514 task_lock(c);
515 /*
516 * Delay read_unlock() till we have the task_lock()
517 * to ensure that c does not slip away underneath us
518 */
519 read_unlock(&tasklist_lock);
520 if (c->mm != mm) {
521 task_unlock(c);
522 put_task_struct(c);
523 goto retry;
524 }
525 WRITE_ONCE(mm->owner, c);
526 lru_gen_migrate_mm(mm);
527 task_unlock(c);
528 put_task_struct(c);
529}
530#endif /* CONFIG_MEMCG */
531
532/*
533 * Turn us into a lazy TLB process if we
534 * aren't already..
535 */
536static void exit_mm(void)
537{
538 struct mm_struct *mm = current->mm;
539
540 exit_mm_release(current, mm);
541 if (!mm)
542 return;
543 mmap_read_lock(mm);
544 mmgrab_lazy_tlb(mm);
545 BUG_ON(mm != current->active_mm);
546 /* more a memory barrier than a real lock */
547 task_lock(current);
548 /*
549 * When a thread stops operating on an address space, the loop
550 * in membarrier_private_expedited() may not observe that
551 * tsk->mm, and the loop in membarrier_global_expedited() may
552 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
553 * rq->membarrier_state, so those would not issue an IPI.
554 * Membarrier requires a memory barrier after accessing
555 * user-space memory, before clearing tsk->mm or the
556 * rq->membarrier_state.
557 */
558 smp_mb__after_spinlock();
559 local_irq_disable();
560 current->mm = NULL;
561 membarrier_update_current_mm(NULL);
562 enter_lazy_tlb(mm, current);
563 local_irq_enable();
564 task_unlock(current);
565 mmap_read_unlock(mm);
566 mm_update_next_owner(mm);
567 mmput(mm);
568 if (test_thread_flag(TIF_MEMDIE))
569 exit_oom_victim();
570}
571
572static struct task_struct *find_alive_thread(struct task_struct *p)
573{
574 struct task_struct *t;
575
576 for_each_thread(p, t) {
577 if (!(t->flags & PF_EXITING))
578 return t;
579 }
580 return NULL;
581}
582
583static struct task_struct *find_child_reaper(struct task_struct *father,
584 struct list_head *dead)
585 __releases(&tasklist_lock)
586 __acquires(&tasklist_lock)
587{
588 struct pid_namespace *pid_ns = task_active_pid_ns(father);
589 struct task_struct *reaper = pid_ns->child_reaper;
590 struct task_struct *p, *n;
591
592 if (likely(reaper != father))
593 return reaper;
594
595 reaper = find_alive_thread(father);
596 if (reaper) {
597 pid_ns->child_reaper = reaper;
598 return reaper;
599 }
600
601 write_unlock_irq(&tasklist_lock);
602
603 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
604 list_del_init(&p->ptrace_entry);
605 release_task(p);
606 }
607
608 zap_pid_ns_processes(pid_ns);
609 write_lock_irq(&tasklist_lock);
610
611 return father;
612}
613
614/*
615 * When we die, we re-parent all our children, and try to:
616 * 1. give them to another thread in our thread group, if such a member exists
617 * 2. give it to the first ancestor process which prctl'd itself as a
618 * child_subreaper for its children (like a service manager)
619 * 3. give it to the init process (PID 1) in our pid namespace
620 */
621static struct task_struct *find_new_reaper(struct task_struct *father,
622 struct task_struct *child_reaper)
623{
624 struct task_struct *thread, *reaper;
625
626 thread = find_alive_thread(father);
627 if (thread)
628 return thread;
629
630 if (father->signal->has_child_subreaper) {
631 unsigned int ns_level = task_pid(father)->level;
632 /*
633 * Find the first ->is_child_subreaper ancestor in our pid_ns.
634 * We can't check reaper != child_reaper to ensure we do not
635 * cross the namespaces, the exiting parent could be injected
636 * by setns() + fork().
637 * We check pid->level, this is slightly more efficient than
638 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
639 */
640 for (reaper = father->real_parent;
641 task_pid(reaper)->level == ns_level;
642 reaper = reaper->real_parent) {
643 if (reaper == &init_task)
644 break;
645 if (!reaper->signal->is_child_subreaper)
646 continue;
647 thread = find_alive_thread(reaper);
648 if (thread)
649 return thread;
650 }
651 }
652
653 return child_reaper;
654}
655
656/*
657* Any that need to be release_task'd are put on the @dead list.
658 */
659static void reparent_leader(struct task_struct *father, struct task_struct *p,
660 struct list_head *dead)
661{
662 if (unlikely(p->exit_state == EXIT_DEAD))
663 return;
664
665 /* We don't want people slaying init. */
666 p->exit_signal = SIGCHLD;
667
668 /* If it has exited notify the new parent about this child's death. */
669 if (!p->ptrace &&
670 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
671 if (do_notify_parent(p, p->exit_signal)) {
672 p->exit_state = EXIT_DEAD;
673 list_add(&p->ptrace_entry, dead);
674 }
675 }
676
677 kill_orphaned_pgrp(p, father);
678}
679
680/*
681 * This does two things:
682 *
683 * A. Make init inherit all the child processes
684 * B. Check to see if any process groups have become orphaned
685 * as a result of our exiting, and if they have any stopped
686 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
687 */
688static void forget_original_parent(struct task_struct *father,
689 struct list_head *dead)
690{
691 struct task_struct *p, *t, *reaper;
692
693 if (unlikely(!list_empty(&father->ptraced)))
694 exit_ptrace(father, dead);
695
696 /* Can drop and reacquire tasklist_lock */
697 reaper = find_child_reaper(father, dead);
698 if (list_empty(&father->children))
699 return;
700
701 reaper = find_new_reaper(father, reaper);
702 list_for_each_entry(p, &father->children, sibling) {
703 for_each_thread(p, t) {
704 RCU_INIT_POINTER(t->real_parent, reaper);
705 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
706 if (likely(!t->ptrace))
707 t->parent = t->real_parent;
708 if (t->pdeath_signal)
709 group_send_sig_info(t->pdeath_signal,
710 SEND_SIG_NOINFO, t,
711 PIDTYPE_TGID);
712 }
713 /*
714 * If this is a threaded reparent there is no need to
715 * notify anyone anything has happened.
716 */
717 if (!same_thread_group(reaper, father))
718 reparent_leader(father, p, dead);
719 }
720 list_splice_tail_init(&father->children, &reaper->children);
721}
722
723/*
724 * Send signals to all our closest relatives so that they know
725 * to properly mourn us..
726 */
727static void exit_notify(struct task_struct *tsk, int group_dead)
728{
729 bool autoreap;
730 struct task_struct *p, *n;
731 LIST_HEAD(dead);
732
733 write_lock_irq(&tasklist_lock);
734 forget_original_parent(tsk, &dead);
735
736 if (group_dead)
737 kill_orphaned_pgrp(tsk->group_leader, NULL);
738
739 tsk->exit_state = EXIT_ZOMBIE;
740 /*
741 * sub-thread or delay_group_leader(), wake up the
742 * PIDFD_THREAD waiters.
743 */
744 if (!thread_group_empty(tsk))
745 do_notify_pidfd(tsk);
746
747 if (unlikely(tsk->ptrace)) {
748 int sig = thread_group_leader(tsk) &&
749 thread_group_empty(tsk) &&
750 !ptrace_reparented(tsk) ?
751 tsk->exit_signal : SIGCHLD;
752 autoreap = do_notify_parent(tsk, sig);
753 } else if (thread_group_leader(tsk)) {
754 autoreap = thread_group_empty(tsk) &&
755 do_notify_parent(tsk, tsk->exit_signal);
756 } else {
757 autoreap = true;
758 }
759
760 if (autoreap) {
761 tsk->exit_state = EXIT_DEAD;
762 list_add(&tsk->ptrace_entry, &dead);
763 }
764
765 /* mt-exec, de_thread() is waiting for group leader */
766 if (unlikely(tsk->signal->notify_count < 0))
767 wake_up_process(tsk->signal->group_exec_task);
768 write_unlock_irq(&tasklist_lock);
769
770 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
771 list_del_init(&p->ptrace_entry);
772 release_task(p);
773 }
774}
775
776#ifdef CONFIG_DEBUG_STACK_USAGE
777static void check_stack_usage(void)
778{
779 static DEFINE_SPINLOCK(low_water_lock);
780 static int lowest_to_date = THREAD_SIZE;
781 unsigned long free;
782
783 free = stack_not_used(current);
784
785 if (free >= lowest_to_date)
786 return;
787
788 spin_lock(&low_water_lock);
789 if (free < lowest_to_date) {
790 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
791 current->comm, task_pid_nr(current), free);
792 lowest_to_date = free;
793 }
794 spin_unlock(&low_water_lock);
795}
796#else
797static inline void check_stack_usage(void) {}
798#endif
799
800static void synchronize_group_exit(struct task_struct *tsk, long code)
801{
802 struct sighand_struct *sighand = tsk->sighand;
803 struct signal_struct *signal = tsk->signal;
804
805 spin_lock_irq(&sighand->siglock);
806 signal->quick_threads--;
807 if ((signal->quick_threads == 0) &&
808 !(signal->flags & SIGNAL_GROUP_EXIT)) {
809 signal->flags = SIGNAL_GROUP_EXIT;
810 signal->group_exit_code = code;
811 signal->group_stop_count = 0;
812 }
813 spin_unlock_irq(&sighand->siglock);
814}
815
816void __noreturn do_exit(long code)
817{
818 struct task_struct *tsk = current;
819 int group_dead;
820
821 WARN_ON(irqs_disabled());
822
823 synchronize_group_exit(tsk, code);
824
825 WARN_ON(tsk->plug);
826
827 kcov_task_exit(tsk);
828 kmsan_task_exit(tsk);
829
830 coredump_task_exit(tsk);
831 ptrace_event(PTRACE_EVENT_EXIT, code);
832 user_events_exit(tsk);
833
834 io_uring_files_cancel();
835 exit_signals(tsk); /* sets PF_EXITING */
836
837 acct_update_integrals(tsk);
838 group_dead = atomic_dec_and_test(&tsk->signal->live);
839 if (group_dead) {
840 /*
841 * If the last thread of global init has exited, panic
842 * immediately to get a useable coredump.
843 */
844 if (unlikely(is_global_init(tsk)))
845 panic("Attempted to kill init! exitcode=0x%08x\n",
846 tsk->signal->group_exit_code ?: (int)code);
847
848#ifdef CONFIG_POSIX_TIMERS
849 hrtimer_cancel(&tsk->signal->real_timer);
850 exit_itimers(tsk);
851#endif
852 if (tsk->mm)
853 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
854 }
855 acct_collect(code, group_dead);
856 if (group_dead)
857 tty_audit_exit();
858 audit_free(tsk);
859
860 tsk->exit_code = code;
861 taskstats_exit(tsk, group_dead);
862
863 exit_mm();
864
865 if (group_dead)
866 acct_process();
867 trace_sched_process_exit(tsk);
868
869 exit_sem(tsk);
870 exit_shm(tsk);
871 exit_files(tsk);
872 exit_fs(tsk);
873 if (group_dead)
874 disassociate_ctty(1);
875 exit_task_namespaces(tsk);
876 exit_task_work(tsk);
877 exit_thread(tsk);
878
879 /*
880 * Flush inherited counters to the parent - before the parent
881 * gets woken up by child-exit notifications.
882 *
883 * because of cgroup mode, must be called before cgroup_exit()
884 */
885 perf_event_exit_task(tsk);
886
887 sched_autogroup_exit_task(tsk);
888 cgroup_exit(tsk);
889
890 /*
891 * FIXME: do that only when needed, using sched_exit tracepoint
892 */
893 flush_ptrace_hw_breakpoint(tsk);
894
895 exit_tasks_rcu_start();
896 exit_notify(tsk, group_dead);
897 proc_exit_connector(tsk);
898 mpol_put_task_policy(tsk);
899#ifdef CONFIG_FUTEX
900 if (unlikely(current->pi_state_cache))
901 kfree(current->pi_state_cache);
902#endif
903 /*
904 * Make sure we are holding no locks:
905 */
906 debug_check_no_locks_held();
907
908 if (tsk->io_context)
909 exit_io_context(tsk);
910
911 if (tsk->splice_pipe)
912 free_pipe_info(tsk->splice_pipe);
913
914 if (tsk->task_frag.page)
915 put_page(tsk->task_frag.page);
916
917 exit_task_stack_account(tsk);
918
919 check_stack_usage();
920 preempt_disable();
921 if (tsk->nr_dirtied)
922 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
923 exit_rcu();
924 exit_tasks_rcu_finish();
925
926 lockdep_free_task(tsk);
927 do_task_dead();
928}
929
930void __noreturn make_task_dead(int signr)
931{
932 /*
933 * Take the task off the cpu after something catastrophic has
934 * happened.
935 *
936 * We can get here from a kernel oops, sometimes with preemption off.
937 * Start by checking for critical errors.
938 * Then fix up important state like USER_DS and preemption.
939 * Then do everything else.
940 */
941 struct task_struct *tsk = current;
942 unsigned int limit;
943
944 if (unlikely(in_interrupt()))
945 panic("Aiee, killing interrupt handler!");
946 if (unlikely(!tsk->pid))
947 panic("Attempted to kill the idle task!");
948
949 if (unlikely(irqs_disabled())) {
950 pr_info("note: %s[%d] exited with irqs disabled\n",
951 current->comm, task_pid_nr(current));
952 local_irq_enable();
953 }
954 if (unlikely(in_atomic())) {
955 pr_info("note: %s[%d] exited with preempt_count %d\n",
956 current->comm, task_pid_nr(current),
957 preempt_count());
958 preempt_count_set(PREEMPT_ENABLED);
959 }
960
961 /*
962 * Every time the system oopses, if the oops happens while a reference
963 * to an object was held, the reference leaks.
964 * If the oops doesn't also leak memory, repeated oopsing can cause
965 * reference counters to wrap around (if they're not using refcount_t).
966 * This means that repeated oopsing can make unexploitable-looking bugs
967 * exploitable through repeated oopsing.
968 * To make sure this can't happen, place an upper bound on how often the
969 * kernel may oops without panic().
970 */
971 limit = READ_ONCE(oops_limit);
972 if (atomic_inc_return(&oops_count) >= limit && limit)
973 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
974
975 /*
976 * We're taking recursive faults here in make_task_dead. Safest is to just
977 * leave this task alone and wait for reboot.
978 */
979 if (unlikely(tsk->flags & PF_EXITING)) {
980 pr_alert("Fixing recursive fault but reboot is needed!\n");
981 futex_exit_recursive(tsk);
982 tsk->exit_state = EXIT_DEAD;
983 refcount_inc(&tsk->rcu_users);
984 do_task_dead();
985 }
986
987 do_exit(signr);
988}
989
990SYSCALL_DEFINE1(exit, int, error_code)
991{
992 do_exit((error_code&0xff)<<8);
993}
994
995/*
996 * Take down every thread in the group. This is called by fatal signals
997 * as well as by sys_exit_group (below).
998 */
999void __noreturn
1000do_group_exit(int exit_code)
1001{
1002 struct signal_struct *sig = current->signal;
1003
1004 if (sig->flags & SIGNAL_GROUP_EXIT)
1005 exit_code = sig->group_exit_code;
1006 else if (sig->group_exec_task)
1007 exit_code = 0;
1008 else {
1009 struct sighand_struct *const sighand = current->sighand;
1010
1011 spin_lock_irq(&sighand->siglock);
1012 if (sig->flags & SIGNAL_GROUP_EXIT)
1013 /* Another thread got here before we took the lock. */
1014 exit_code = sig->group_exit_code;
1015 else if (sig->group_exec_task)
1016 exit_code = 0;
1017 else {
1018 sig->group_exit_code = exit_code;
1019 sig->flags = SIGNAL_GROUP_EXIT;
1020 zap_other_threads(current);
1021 }
1022 spin_unlock_irq(&sighand->siglock);
1023 }
1024
1025 do_exit(exit_code);
1026 /* NOTREACHED */
1027}
1028
1029/*
1030 * this kills every thread in the thread group. Note that any externally
1031 * wait4()-ing process will get the correct exit code - even if this
1032 * thread is not the thread group leader.
1033 */
1034SYSCALL_DEFINE1(exit_group, int, error_code)
1035{
1036 do_group_exit((error_code & 0xff) << 8);
1037 /* NOTREACHED */
1038 return 0;
1039}
1040
1041static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1042{
1043 return wo->wo_type == PIDTYPE_MAX ||
1044 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1045}
1046
1047static int
1048eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1049{
1050 if (!eligible_pid(wo, p))
1051 return 0;
1052
1053 /*
1054 * Wait for all children (clone and not) if __WALL is set or
1055 * if it is traced by us.
1056 */
1057 if (ptrace || (wo->wo_flags & __WALL))
1058 return 1;
1059
1060 /*
1061 * Otherwise, wait for clone children *only* if __WCLONE is set;
1062 * otherwise, wait for non-clone children *only*.
1063 *
1064 * Note: a "clone" child here is one that reports to its parent
1065 * using a signal other than SIGCHLD, or a non-leader thread which
1066 * we can only see if it is traced by us.
1067 */
1068 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1069 return 0;
1070
1071 return 1;
1072}
1073
1074/*
1075 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1076 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1077 * the lock and this task is uninteresting. If we return nonzero, we have
1078 * released the lock and the system call should return.
1079 */
1080static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1081{
1082 int state, status;
1083 pid_t pid = task_pid_vnr(p);
1084 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1085 struct waitid_info *infop;
1086
1087 if (!likely(wo->wo_flags & WEXITED))
1088 return 0;
1089
1090 if (unlikely(wo->wo_flags & WNOWAIT)) {
1091 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1092 ? p->signal->group_exit_code : p->exit_code;
1093 get_task_struct(p);
1094 read_unlock(&tasklist_lock);
1095 sched_annotate_sleep();
1096 if (wo->wo_rusage)
1097 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1098 put_task_struct(p);
1099 goto out_info;
1100 }
1101 /*
1102 * Move the task's state to DEAD/TRACE, only one thread can do this.
1103 */
1104 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1105 EXIT_TRACE : EXIT_DEAD;
1106 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1107 return 0;
1108 /*
1109 * We own this thread, nobody else can reap it.
1110 */
1111 read_unlock(&tasklist_lock);
1112 sched_annotate_sleep();
1113
1114 /*
1115 * Check thread_group_leader() to exclude the traced sub-threads.
1116 */
1117 if (state == EXIT_DEAD && thread_group_leader(p)) {
1118 struct signal_struct *sig = p->signal;
1119 struct signal_struct *psig = current->signal;
1120 unsigned long maxrss;
1121 u64 tgutime, tgstime;
1122
1123 /*
1124 * The resource counters for the group leader are in its
1125 * own task_struct. Those for dead threads in the group
1126 * are in its signal_struct, as are those for the child
1127 * processes it has previously reaped. All these
1128 * accumulate in the parent's signal_struct c* fields.
1129 *
1130 * We don't bother to take a lock here to protect these
1131 * p->signal fields because the whole thread group is dead
1132 * and nobody can change them.
1133 *
1134 * psig->stats_lock also protects us from our sub-threads
1135 * which can reap other children at the same time.
1136 *
1137 * We use thread_group_cputime_adjusted() to get times for
1138 * the thread group, which consolidates times for all threads
1139 * in the group including the group leader.
1140 */
1141 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1142 write_seqlock_irq(&psig->stats_lock);
1143 psig->cutime += tgutime + sig->cutime;
1144 psig->cstime += tgstime + sig->cstime;
1145 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1146 psig->cmin_flt +=
1147 p->min_flt + sig->min_flt + sig->cmin_flt;
1148 psig->cmaj_flt +=
1149 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1150 psig->cnvcsw +=
1151 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1152 psig->cnivcsw +=
1153 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1154 psig->cinblock +=
1155 task_io_get_inblock(p) +
1156 sig->inblock + sig->cinblock;
1157 psig->coublock +=
1158 task_io_get_oublock(p) +
1159 sig->oublock + sig->coublock;
1160 maxrss = max(sig->maxrss, sig->cmaxrss);
1161 if (psig->cmaxrss < maxrss)
1162 psig->cmaxrss = maxrss;
1163 task_io_accounting_add(&psig->ioac, &p->ioac);
1164 task_io_accounting_add(&psig->ioac, &sig->ioac);
1165 write_sequnlock_irq(&psig->stats_lock);
1166 }
1167
1168 if (wo->wo_rusage)
1169 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1170 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1171 ? p->signal->group_exit_code : p->exit_code;
1172 wo->wo_stat = status;
1173
1174 if (state == EXIT_TRACE) {
1175 write_lock_irq(&tasklist_lock);
1176 /* We dropped tasklist, ptracer could die and untrace */
1177 ptrace_unlink(p);
1178
1179 /* If parent wants a zombie, don't release it now */
1180 state = EXIT_ZOMBIE;
1181 if (do_notify_parent(p, p->exit_signal))
1182 state = EXIT_DEAD;
1183 p->exit_state = state;
1184 write_unlock_irq(&tasklist_lock);
1185 }
1186 if (state == EXIT_DEAD)
1187 release_task(p);
1188
1189out_info:
1190 infop = wo->wo_info;
1191 if (infop) {
1192 if ((status & 0x7f) == 0) {
1193 infop->cause = CLD_EXITED;
1194 infop->status = status >> 8;
1195 } else {
1196 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1197 infop->status = status & 0x7f;
1198 }
1199 infop->pid = pid;
1200 infop->uid = uid;
1201 }
1202
1203 return pid;
1204}
1205
1206static int *task_stopped_code(struct task_struct *p, bool ptrace)
1207{
1208 if (ptrace) {
1209 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1210 return &p->exit_code;
1211 } else {
1212 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1213 return &p->signal->group_exit_code;
1214 }
1215 return NULL;
1216}
1217
1218/**
1219 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1220 * @wo: wait options
1221 * @ptrace: is the wait for ptrace
1222 * @p: task to wait for
1223 *
1224 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1225 *
1226 * CONTEXT:
1227 * read_lock(&tasklist_lock), which is released if return value is
1228 * non-zero. Also, grabs and releases @p->sighand->siglock.
1229 *
1230 * RETURNS:
1231 * 0 if wait condition didn't exist and search for other wait conditions
1232 * should continue. Non-zero return, -errno on failure and @p's pid on
1233 * success, implies that tasklist_lock is released and wait condition
1234 * search should terminate.
1235 */
1236static int wait_task_stopped(struct wait_opts *wo,
1237 int ptrace, struct task_struct *p)
1238{
1239 struct waitid_info *infop;
1240 int exit_code, *p_code, why;
1241 uid_t uid = 0; /* unneeded, required by compiler */
1242 pid_t pid;
1243
1244 /*
1245 * Traditionally we see ptrace'd stopped tasks regardless of options.
1246 */
1247 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1248 return 0;
1249
1250 if (!task_stopped_code(p, ptrace))
1251 return 0;
1252
1253 exit_code = 0;
1254 spin_lock_irq(&p->sighand->siglock);
1255
1256 p_code = task_stopped_code(p, ptrace);
1257 if (unlikely(!p_code))
1258 goto unlock_sig;
1259
1260 exit_code = *p_code;
1261 if (!exit_code)
1262 goto unlock_sig;
1263
1264 if (!unlikely(wo->wo_flags & WNOWAIT))
1265 *p_code = 0;
1266
1267 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1268unlock_sig:
1269 spin_unlock_irq(&p->sighand->siglock);
1270 if (!exit_code)
1271 return 0;
1272
1273 /*
1274 * Now we are pretty sure this task is interesting.
1275 * Make sure it doesn't get reaped out from under us while we
1276 * give up the lock and then examine it below. We don't want to
1277 * keep holding onto the tasklist_lock while we call getrusage and
1278 * possibly take page faults for user memory.
1279 */
1280 get_task_struct(p);
1281 pid = task_pid_vnr(p);
1282 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1283 read_unlock(&tasklist_lock);
1284 sched_annotate_sleep();
1285 if (wo->wo_rusage)
1286 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1287 put_task_struct(p);
1288
1289 if (likely(!(wo->wo_flags & WNOWAIT)))
1290 wo->wo_stat = (exit_code << 8) | 0x7f;
1291
1292 infop = wo->wo_info;
1293 if (infop) {
1294 infop->cause = why;
1295 infop->status = exit_code;
1296 infop->pid = pid;
1297 infop->uid = uid;
1298 }
1299 return pid;
1300}
1301
1302/*
1303 * Handle do_wait work for one task in a live, non-stopped state.
1304 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1305 * the lock and this task is uninteresting. If we return nonzero, we have
1306 * released the lock and the system call should return.
1307 */
1308static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1309{
1310 struct waitid_info *infop;
1311 pid_t pid;
1312 uid_t uid;
1313
1314 if (!unlikely(wo->wo_flags & WCONTINUED))
1315 return 0;
1316
1317 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1318 return 0;
1319
1320 spin_lock_irq(&p->sighand->siglock);
1321 /* Re-check with the lock held. */
1322 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1323 spin_unlock_irq(&p->sighand->siglock);
1324 return 0;
1325 }
1326 if (!unlikely(wo->wo_flags & WNOWAIT))
1327 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1328 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1329 spin_unlock_irq(&p->sighand->siglock);
1330
1331 pid = task_pid_vnr(p);
1332 get_task_struct(p);
1333 read_unlock(&tasklist_lock);
1334 sched_annotate_sleep();
1335 if (wo->wo_rusage)
1336 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1337 put_task_struct(p);
1338
1339 infop = wo->wo_info;
1340 if (!infop) {
1341 wo->wo_stat = 0xffff;
1342 } else {
1343 infop->cause = CLD_CONTINUED;
1344 infop->pid = pid;
1345 infop->uid = uid;
1346 infop->status = SIGCONT;
1347 }
1348 return pid;
1349}
1350
1351/*
1352 * Consider @p for a wait by @parent.
1353 *
1354 * -ECHILD should be in ->notask_error before the first call.
1355 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1356 * Returns zero if the search for a child should continue;
1357 * then ->notask_error is 0 if @p is an eligible child,
1358 * or still -ECHILD.
1359 */
1360static int wait_consider_task(struct wait_opts *wo, int ptrace,
1361 struct task_struct *p)
1362{
1363 /*
1364 * We can race with wait_task_zombie() from another thread.
1365 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1366 * can't confuse the checks below.
1367 */
1368 int exit_state = READ_ONCE(p->exit_state);
1369 int ret;
1370
1371 if (unlikely(exit_state == EXIT_DEAD))
1372 return 0;
1373
1374 ret = eligible_child(wo, ptrace, p);
1375 if (!ret)
1376 return ret;
1377
1378 if (unlikely(exit_state == EXIT_TRACE)) {
1379 /*
1380 * ptrace == 0 means we are the natural parent. In this case
1381 * we should clear notask_error, debugger will notify us.
1382 */
1383 if (likely(!ptrace))
1384 wo->notask_error = 0;
1385 return 0;
1386 }
1387
1388 if (likely(!ptrace) && unlikely(p->ptrace)) {
1389 /*
1390 * If it is traced by its real parent's group, just pretend
1391 * the caller is ptrace_do_wait() and reap this child if it
1392 * is zombie.
1393 *
1394 * This also hides group stop state from real parent; otherwise
1395 * a single stop can be reported twice as group and ptrace stop.
1396 * If a ptracer wants to distinguish these two events for its
1397 * own children it should create a separate process which takes
1398 * the role of real parent.
1399 */
1400 if (!ptrace_reparented(p))
1401 ptrace = 1;
1402 }
1403
1404 /* slay zombie? */
1405 if (exit_state == EXIT_ZOMBIE) {
1406 /* we don't reap group leaders with subthreads */
1407 if (!delay_group_leader(p)) {
1408 /*
1409 * A zombie ptracee is only visible to its ptracer.
1410 * Notification and reaping will be cascaded to the
1411 * real parent when the ptracer detaches.
1412 */
1413 if (unlikely(ptrace) || likely(!p->ptrace))
1414 return wait_task_zombie(wo, p);
1415 }
1416
1417 /*
1418 * Allow access to stopped/continued state via zombie by
1419 * falling through. Clearing of notask_error is complex.
1420 *
1421 * When !@ptrace:
1422 *
1423 * If WEXITED is set, notask_error should naturally be
1424 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1425 * so, if there are live subthreads, there are events to
1426 * wait for. If all subthreads are dead, it's still safe
1427 * to clear - this function will be called again in finite
1428 * amount time once all the subthreads are released and
1429 * will then return without clearing.
1430 *
1431 * When @ptrace:
1432 *
1433 * Stopped state is per-task and thus can't change once the
1434 * target task dies. Only continued and exited can happen.
1435 * Clear notask_error if WCONTINUED | WEXITED.
1436 */
1437 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1438 wo->notask_error = 0;
1439 } else {
1440 /*
1441 * @p is alive and it's gonna stop, continue or exit, so
1442 * there always is something to wait for.
1443 */
1444 wo->notask_error = 0;
1445 }
1446
1447 /*
1448 * Wait for stopped. Depending on @ptrace, different stopped state
1449 * is used and the two don't interact with each other.
1450 */
1451 ret = wait_task_stopped(wo, ptrace, p);
1452 if (ret)
1453 return ret;
1454
1455 /*
1456 * Wait for continued. There's only one continued state and the
1457 * ptracer can consume it which can confuse the real parent. Don't
1458 * use WCONTINUED from ptracer. You don't need or want it.
1459 */
1460 return wait_task_continued(wo, p);
1461}
1462
1463/*
1464 * Do the work of do_wait() for one thread in the group, @tsk.
1465 *
1466 * -ECHILD should be in ->notask_error before the first call.
1467 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1468 * Returns zero if the search for a child should continue; then
1469 * ->notask_error is 0 if there were any eligible children,
1470 * or still -ECHILD.
1471 */
1472static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1473{
1474 struct task_struct *p;
1475
1476 list_for_each_entry(p, &tsk->children, sibling) {
1477 int ret = wait_consider_task(wo, 0, p);
1478
1479 if (ret)
1480 return ret;
1481 }
1482
1483 return 0;
1484}
1485
1486static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1487{
1488 struct task_struct *p;
1489
1490 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1491 int ret = wait_consider_task(wo, 1, p);
1492
1493 if (ret)
1494 return ret;
1495 }
1496
1497 return 0;
1498}
1499
1500bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1501{
1502 if (!eligible_pid(wo, p))
1503 return false;
1504
1505 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1506 return false;
1507
1508 return true;
1509}
1510
1511static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1512 int sync, void *key)
1513{
1514 struct wait_opts *wo = container_of(wait, struct wait_opts,
1515 child_wait);
1516 struct task_struct *p = key;
1517
1518 if (pid_child_should_wake(wo, p))
1519 return default_wake_function(wait, mode, sync, key);
1520
1521 return 0;
1522}
1523
1524void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1525{
1526 __wake_up_sync_key(&parent->signal->wait_chldexit,
1527 TASK_INTERRUPTIBLE, p);
1528}
1529
1530static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1531 struct task_struct *target)
1532{
1533 struct task_struct *parent =
1534 !ptrace ? target->real_parent : target->parent;
1535
1536 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1537 same_thread_group(current, parent));
1538}
1539
1540/*
1541 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1542 * and tracee lists to find the target task.
1543 */
1544static int do_wait_pid(struct wait_opts *wo)
1545{
1546 bool ptrace;
1547 struct task_struct *target;
1548 int retval;
1549
1550 ptrace = false;
1551 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1552 if (target && is_effectively_child(wo, ptrace, target)) {
1553 retval = wait_consider_task(wo, ptrace, target);
1554 if (retval)
1555 return retval;
1556 }
1557
1558 ptrace = true;
1559 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1560 if (target && target->ptrace &&
1561 is_effectively_child(wo, ptrace, target)) {
1562 retval = wait_consider_task(wo, ptrace, target);
1563 if (retval)
1564 return retval;
1565 }
1566
1567 return 0;
1568}
1569
1570long __do_wait(struct wait_opts *wo)
1571{
1572 long retval;
1573
1574 /*
1575 * If there is nothing that can match our criteria, just get out.
1576 * We will clear ->notask_error to zero if we see any child that
1577 * might later match our criteria, even if we are not able to reap
1578 * it yet.
1579 */
1580 wo->notask_error = -ECHILD;
1581 if ((wo->wo_type < PIDTYPE_MAX) &&
1582 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1583 goto notask;
1584
1585 read_lock(&tasklist_lock);
1586
1587 if (wo->wo_type == PIDTYPE_PID) {
1588 retval = do_wait_pid(wo);
1589 if (retval)
1590 return retval;
1591 } else {
1592 struct task_struct *tsk = current;
1593
1594 do {
1595 retval = do_wait_thread(wo, tsk);
1596 if (retval)
1597 return retval;
1598
1599 retval = ptrace_do_wait(wo, tsk);
1600 if (retval)
1601 return retval;
1602
1603 if (wo->wo_flags & __WNOTHREAD)
1604 break;
1605 } while_each_thread(current, tsk);
1606 }
1607 read_unlock(&tasklist_lock);
1608
1609notask:
1610 retval = wo->notask_error;
1611 if (!retval && !(wo->wo_flags & WNOHANG))
1612 return -ERESTARTSYS;
1613
1614 return retval;
1615}
1616
1617static long do_wait(struct wait_opts *wo)
1618{
1619 int retval;
1620
1621 trace_sched_process_wait(wo->wo_pid);
1622
1623 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1624 wo->child_wait.private = current;
1625 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1626
1627 do {
1628 set_current_state(TASK_INTERRUPTIBLE);
1629 retval = __do_wait(wo);
1630 if (retval != -ERESTARTSYS)
1631 break;
1632 if (signal_pending(current))
1633 break;
1634 schedule();
1635 } while (1);
1636
1637 __set_current_state(TASK_RUNNING);
1638 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1639 return retval;
1640}
1641
1642int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1643 struct waitid_info *infop, int options,
1644 struct rusage *ru)
1645{
1646 unsigned int f_flags = 0;
1647 struct pid *pid = NULL;
1648 enum pid_type type;
1649
1650 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1651 __WNOTHREAD|__WCLONE|__WALL))
1652 return -EINVAL;
1653 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1654 return -EINVAL;
1655
1656 switch (which) {
1657 case P_ALL:
1658 type = PIDTYPE_MAX;
1659 break;
1660 case P_PID:
1661 type = PIDTYPE_PID;
1662 if (upid <= 0)
1663 return -EINVAL;
1664
1665 pid = find_get_pid(upid);
1666 break;
1667 case P_PGID:
1668 type = PIDTYPE_PGID;
1669 if (upid < 0)
1670 return -EINVAL;
1671
1672 if (upid)
1673 pid = find_get_pid(upid);
1674 else
1675 pid = get_task_pid(current, PIDTYPE_PGID);
1676 break;
1677 case P_PIDFD:
1678 type = PIDTYPE_PID;
1679 if (upid < 0)
1680 return -EINVAL;
1681
1682 pid = pidfd_get_pid(upid, &f_flags);
1683 if (IS_ERR(pid))
1684 return PTR_ERR(pid);
1685
1686 break;
1687 default:
1688 return -EINVAL;
1689 }
1690
1691 wo->wo_type = type;
1692 wo->wo_pid = pid;
1693 wo->wo_flags = options;
1694 wo->wo_info = infop;
1695 wo->wo_rusage = ru;
1696 if (f_flags & O_NONBLOCK)
1697 wo->wo_flags |= WNOHANG;
1698
1699 return 0;
1700}
1701
1702static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1703 int options, struct rusage *ru)
1704{
1705 struct wait_opts wo;
1706 long ret;
1707
1708 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1709 if (ret)
1710 return ret;
1711
1712 ret = do_wait(&wo);
1713 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1714 ret = -EAGAIN;
1715
1716 put_pid(wo.wo_pid);
1717 return ret;
1718}
1719
1720SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1721 infop, int, options, struct rusage __user *, ru)
1722{
1723 struct rusage r;
1724 struct waitid_info info = {.status = 0};
1725 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1726 int signo = 0;
1727
1728 if (err > 0) {
1729 signo = SIGCHLD;
1730 err = 0;
1731 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1732 return -EFAULT;
1733 }
1734 if (!infop)
1735 return err;
1736
1737 if (!user_write_access_begin(infop, sizeof(*infop)))
1738 return -EFAULT;
1739
1740 unsafe_put_user(signo, &infop->si_signo, Efault);
1741 unsafe_put_user(0, &infop->si_errno, Efault);
1742 unsafe_put_user(info.cause, &infop->si_code, Efault);
1743 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1744 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1745 unsafe_put_user(info.status, &infop->si_status, Efault);
1746 user_write_access_end();
1747 return err;
1748Efault:
1749 user_write_access_end();
1750 return -EFAULT;
1751}
1752
1753long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1754 struct rusage *ru)
1755{
1756 struct wait_opts wo;
1757 struct pid *pid = NULL;
1758 enum pid_type type;
1759 long ret;
1760
1761 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1762 __WNOTHREAD|__WCLONE|__WALL))
1763 return -EINVAL;
1764
1765 /* -INT_MIN is not defined */
1766 if (upid == INT_MIN)
1767 return -ESRCH;
1768
1769 if (upid == -1)
1770 type = PIDTYPE_MAX;
1771 else if (upid < 0) {
1772 type = PIDTYPE_PGID;
1773 pid = find_get_pid(-upid);
1774 } else if (upid == 0) {
1775 type = PIDTYPE_PGID;
1776 pid = get_task_pid(current, PIDTYPE_PGID);
1777 } else /* upid > 0 */ {
1778 type = PIDTYPE_PID;
1779 pid = find_get_pid(upid);
1780 }
1781
1782 wo.wo_type = type;
1783 wo.wo_pid = pid;
1784 wo.wo_flags = options | WEXITED;
1785 wo.wo_info = NULL;
1786 wo.wo_stat = 0;
1787 wo.wo_rusage = ru;
1788 ret = do_wait(&wo);
1789 put_pid(pid);
1790 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1791 ret = -EFAULT;
1792
1793 return ret;
1794}
1795
1796int kernel_wait(pid_t pid, int *stat)
1797{
1798 struct wait_opts wo = {
1799 .wo_type = PIDTYPE_PID,
1800 .wo_pid = find_get_pid(pid),
1801 .wo_flags = WEXITED,
1802 };
1803 int ret;
1804
1805 ret = do_wait(&wo);
1806 if (ret > 0 && wo.wo_stat)
1807 *stat = wo.wo_stat;
1808 put_pid(wo.wo_pid);
1809 return ret;
1810}
1811
1812SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1813 int, options, struct rusage __user *, ru)
1814{
1815 struct rusage r;
1816 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1817
1818 if (err > 0) {
1819 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1820 return -EFAULT;
1821 }
1822 return err;
1823}
1824
1825#ifdef __ARCH_WANT_SYS_WAITPID
1826
1827/*
1828 * sys_waitpid() remains for compatibility. waitpid() should be
1829 * implemented by calling sys_wait4() from libc.a.
1830 */
1831SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1832{
1833 return kernel_wait4(pid, stat_addr, options, NULL);
1834}
1835
1836#endif
1837
1838#ifdef CONFIG_COMPAT
1839COMPAT_SYSCALL_DEFINE4(wait4,
1840 compat_pid_t, pid,
1841 compat_uint_t __user *, stat_addr,
1842 int, options,
1843 struct compat_rusage __user *, ru)
1844{
1845 struct rusage r;
1846 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1847 if (err > 0) {
1848 if (ru && put_compat_rusage(&r, ru))
1849 return -EFAULT;
1850 }
1851 return err;
1852}
1853
1854COMPAT_SYSCALL_DEFINE5(waitid,
1855 int, which, compat_pid_t, pid,
1856 struct compat_siginfo __user *, infop, int, options,
1857 struct compat_rusage __user *, uru)
1858{
1859 struct rusage ru;
1860 struct waitid_info info = {.status = 0};
1861 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1862 int signo = 0;
1863 if (err > 0) {
1864 signo = SIGCHLD;
1865 err = 0;
1866 if (uru) {
1867 /* kernel_waitid() overwrites everything in ru */
1868 if (COMPAT_USE_64BIT_TIME)
1869 err = copy_to_user(uru, &ru, sizeof(ru));
1870 else
1871 err = put_compat_rusage(&ru, uru);
1872 if (err)
1873 return -EFAULT;
1874 }
1875 }
1876
1877 if (!infop)
1878 return err;
1879
1880 if (!user_write_access_begin(infop, sizeof(*infop)))
1881 return -EFAULT;
1882
1883 unsafe_put_user(signo, &infop->si_signo, Efault);
1884 unsafe_put_user(0, &infop->si_errno, Efault);
1885 unsafe_put_user(info.cause, &infop->si_code, Efault);
1886 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1887 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1888 unsafe_put_user(info.status, &infop->si_status, Efault);
1889 user_write_access_end();
1890 return err;
1891Efault:
1892 user_write_access_end();
1893 return -EFAULT;
1894}
1895#endif
1896
1897/*
1898 * This needs to be __function_aligned as GCC implicitly makes any
1899 * implementation of abort() cold and drops alignment specified by
1900 * -falign-functions=N.
1901 *
1902 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1903 */
1904__weak __function_aligned void abort(void)
1905{
1906 BUG();
1907
1908 /* if that doesn't kill us, halt */
1909 panic("Oops failed to kill thread");
1910}
1911EXPORT_SYMBOL(abort);