projects/linux·blob 1e098dec · refs

1924 lines · 47.8 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
73#include <linux/uaccess.h>
74#include <asm/unistd.h>
75#include <asm/mmu_context.h>
76
77/*
78 * The default value should be high enough to not crash a system that randomly
79 * crashes its kernel from time to time, but low enough to at least not permit
80 * overflowing 32-bit refcounts or the ldsem writer count.
81 */
82static unsigned int oops_limit = 10000;
83
84#ifdef CONFIG_SYSCTL
85static struct ctl_table kern_exit_table[] = {
86 {
87 .procname = "oops_limit",
88 .data = &oops_limit,
89 .maxlen = sizeof(oops_limit),
90 .mode = 0644,
91 .proc_handler = proc_douintvec,
92 },
93 { }
94};
95
96static __init int kernel_exit_sysctls_init(void)
97{
98 register_sysctl_init("kernel", kern_exit_table);
99 return 0;
100}
101late_initcall(kernel_exit_sysctls_init);
102#endif
103
104static atomic_t oops_count = ATOMIC_INIT(0);
105
106#ifdef CONFIG_SYSFS
107static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
108 char *page)
109{
110 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
111}
112
113static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
114
115static __init int kernel_exit_sysfs_init(void)
116{
117 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
118 return 0;
119}
120late_initcall(kernel_exit_sysfs_init);
121#endif
122
123static void __unhash_process(struct task_struct *p, bool group_dead)
124{
125 nr_threads--;
126 detach_pid(p, PIDTYPE_PID);
127 if (group_dead) {
128 detach_pid(p, PIDTYPE_TGID);
129 detach_pid(p, PIDTYPE_PGID);
130 detach_pid(p, PIDTYPE_SID);
131
132 list_del_rcu(&p->tasks);
133 list_del_init(&p->sibling);
134 __this_cpu_dec(process_counts);
135 }
136 list_del_rcu(&p->thread_group);
137 list_del_rcu(&p->thread_node);
138}
139
140/*
141 * This function expects the tasklist_lock write-locked.
142 */
143static void __exit_signal(struct task_struct *tsk)
144{
145 struct signal_struct *sig = tsk->signal;
146 bool group_dead = thread_group_leader(tsk);
147 struct sighand_struct *sighand;
148 struct tty_struct *tty;
149 u64 utime, stime;
150
151 sighand = rcu_dereference_check(tsk->sighand,
152 lockdep_tasklist_lock_is_held());
153 spin_lock(&sighand->siglock);
154
155#ifdef CONFIG_POSIX_TIMERS
156 posix_cpu_timers_exit(tsk);
157 if (group_dead)
158 posix_cpu_timers_exit_group(tsk);
159#endif
160
161 if (group_dead) {
162 tty = sig->tty;
163 sig->tty = NULL;
164 } else {
165 /*
166 * If there is any task waiting for the group exit
167 * then notify it:
168 */
169 if (sig->notify_count > 0 && !--sig->notify_count)
170 wake_up_process(sig->group_exec_task);
171
172 if (tsk == sig->curr_target)
173 sig->curr_target = next_thread(tsk);
174 }
175
176 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
177 sizeof(unsigned long long));
178
179 /*
180 * Accumulate here the counters for all threads as they die. We could
181 * skip the group leader because it is the last user of signal_struct,
182 * but we want to avoid the race with thread_group_cputime() which can
183 * see the empty ->thread_head list.
184 */
185 task_cputime(tsk, &utime, &stime);
186 write_seqlock(&sig->stats_lock);
187 sig->utime += utime;
188 sig->stime += stime;
189 sig->gtime += task_gtime(tsk);
190 sig->min_flt += tsk->min_flt;
191 sig->maj_flt += tsk->maj_flt;
192 sig->nvcsw += tsk->nvcsw;
193 sig->nivcsw += tsk->nivcsw;
194 sig->inblock += task_io_get_inblock(tsk);
195 sig->oublock += task_io_get_oublock(tsk);
196 task_io_accounting_add(&sig->ioac, &tsk->ioac);
197 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
198 sig->nr_threads--;
199 __unhash_process(tsk, group_dead);
200 write_sequnlock(&sig->stats_lock);
201
202 /*
203 * Do this under ->siglock, we can race with another thread
204 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
205 */
206 flush_sigqueue(&tsk->pending);
207 tsk->sighand = NULL;
208 spin_unlock(&sighand->siglock);
209
210 __cleanup_sighand(sighand);
211 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
212 if (group_dead) {
213 flush_sigqueue(&sig->shared_pending);
214 tty_kref_put(tty);
215 }
216}
217
218static void delayed_put_task_struct(struct rcu_head *rhp)
219{
220 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
221
222 kprobe_flush_task(tsk);
223 rethook_flush_task(tsk);
224 perf_event_delayed_put(tsk);
225 trace_sched_process_free(tsk);
226 put_task_struct(tsk);
227}
228
229void put_task_struct_rcu_user(struct task_struct *task)
230{
231 if (refcount_dec_and_test(&task->rcu_users))
232 call_rcu(&task->rcu, delayed_put_task_struct);
233}
234
235void __weak release_thread(struct task_struct *dead_task)
236{
237}
238
239void release_task(struct task_struct *p)
240{
241 struct task_struct *leader;
242 struct pid *thread_pid;
243 int zap_leader;
244repeat:
245 /* don't need to get the RCU readlock here - the process is dead and
246 * can't be modifying its own credentials. But shut RCU-lockdep up */
247 rcu_read_lock();
248 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
249 rcu_read_unlock();
250
251 cgroup_release(p);
252
253 write_lock_irq(&tasklist_lock);
254 ptrace_release_task(p);
255 thread_pid = get_pid(p->thread_pid);
256 __exit_signal(p);
257
258 /*
259 * If we are the last non-leader member of the thread
260 * group, and the leader is zombie, then notify the
261 * group leader's parent process. (if it wants notification.)
262 */
263 zap_leader = 0;
264 leader = p->group_leader;
265 if (leader != p && thread_group_empty(leader)
266 && leader->exit_state == EXIT_ZOMBIE) {
267 /*
268 * If we were the last child thread and the leader has
269 * exited already, and the leader's parent ignores SIGCHLD,
270 * then we are the one who should release the leader.
271 */
272 zap_leader = do_notify_parent(leader, leader->exit_signal);
273 if (zap_leader)
274 leader->exit_state = EXIT_DEAD;
275 }
276
277 write_unlock_irq(&tasklist_lock);
278 seccomp_filter_release(p);
279 proc_flush_pid(thread_pid);
280 put_pid(thread_pid);
281 release_thread(p);
282 put_task_struct_rcu_user(p);
283
284 p = leader;
285 if (unlikely(zap_leader))
286 goto repeat;
287}
288
289int rcuwait_wake_up(struct rcuwait *w)
290{
291 int ret = 0;
292 struct task_struct *task;
293
294 rcu_read_lock();
295
296 /*
297 * Order condition vs @task, such that everything prior to the load
298 * of @task is visible. This is the condition as to why the user called
299 * rcuwait_wake() in the first place. Pairs with set_current_state()
300 * barrier (A) in rcuwait_wait_event().
301 *
302 * WAIT WAKE
303 * [S] tsk = current [S] cond = true
304 * MB (A) MB (B)
305 * [L] cond [L] tsk
306 */
307 smp_mb(); /* (B) */
308
309 task = rcu_dereference(w->task);
310 if (task)
311 ret = wake_up_process(task);
312 rcu_read_unlock();
313
314 return ret;
315}
316EXPORT_SYMBOL_GPL(rcuwait_wake_up);
317
318/*
319 * Determine if a process group is "orphaned", according to the POSIX
320 * definition in 2.2.2.52. Orphaned process groups are not to be affected
321 * by terminal-generated stop signals. Newly orphaned process groups are
322 * to receive a SIGHUP and a SIGCONT.
323 *
324 * "I ask you, have you ever known what it is to be an orphan?"
325 */
326static int will_become_orphaned_pgrp(struct pid *pgrp,
327 struct task_struct *ignored_task)
328{
329 struct task_struct *p;
330
331 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
332 if ((p == ignored_task) ||
333 (p->exit_state && thread_group_empty(p)) ||
334 is_global_init(p->real_parent))
335 continue;
336
337 if (task_pgrp(p->real_parent) != pgrp &&
338 task_session(p->real_parent) == task_session(p))
339 return 0;
340 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
341
342 return 1;
343}
344
345int is_current_pgrp_orphaned(void)
346{
347 int retval;
348
349 read_lock(&tasklist_lock);
350 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
351 read_unlock(&tasklist_lock);
352
353 return retval;
354}
355
356static bool has_stopped_jobs(struct pid *pgrp)
357{
358 struct task_struct *p;
359
360 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
361 if (p->signal->flags & SIGNAL_STOP_STOPPED)
362 return true;
363 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
364
365 return false;
366}
367
368/*
369 * Check to see if any process groups have become orphaned as
370 * a result of our exiting, and if they have any stopped jobs,
371 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
372 */
373static void
374kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
375{
376 struct pid *pgrp = task_pgrp(tsk);
377 struct task_struct *ignored_task = tsk;
378
379 if (!parent)
380 /* exit: our father is in a different pgrp than
381 * we are and we were the only connection outside.
382 */
383 parent = tsk->real_parent;
384 else
385 /* reparent: our child is in a different pgrp than
386 * we are, and it was the only connection outside.
387 */
388 ignored_task = NULL;
389
390 if (task_pgrp(parent) != pgrp &&
391 task_session(parent) == task_session(tsk) &&
392 will_become_orphaned_pgrp(pgrp, ignored_task) &&
393 has_stopped_jobs(pgrp)) {
394 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
395 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
396 }
397}
398
399static void coredump_task_exit(struct task_struct *tsk)
400{
401 struct core_state *core_state;
402
403 /*
404 * Serialize with any possible pending coredump.
405 * We must hold siglock around checking core_state
406 * and setting PF_POSTCOREDUMP. The core-inducing thread
407 * will increment ->nr_threads for each thread in the
408 * group without PF_POSTCOREDUMP set.
409 */
410 spin_lock_irq(&tsk->sighand->siglock);
411 tsk->flags |= PF_POSTCOREDUMP;
412 core_state = tsk->signal->core_state;
413 spin_unlock_irq(&tsk->sighand->siglock);
414 if (core_state) {
415 struct core_thread self;
416
417 self.task = current;
418 if (self.task->flags & PF_SIGNALED)
419 self.next = xchg(&core_state->dumper.next, &self);
420 else
421 self.task = NULL;
422 /*
423 * Implies mb(), the result of xchg() must be visible
424 * to core_state->dumper.
425 */
426 if (atomic_dec_and_test(&core_state->nr_threads))
427 complete(&core_state->startup);
428
429 for (;;) {
430 set_current_state(TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
431 if (!self.task) /* see coredump_finish() */
432 break;
433 schedule();
434 }
435 __set_current_state(TASK_RUNNING);
436 }
437}
438
439#ifdef CONFIG_MEMCG
440/*
441 * A task is exiting. If it owned this mm, find a new owner for the mm.
442 */
443void mm_update_next_owner(struct mm_struct *mm)
444{
445 struct task_struct *c, *g, *p = current;
446
447retry:
448 /*
449 * If the exiting or execing task is not the owner, it's
450 * someone else's problem.
451 */
452 if (mm->owner != p)
453 return;
454 /*
455 * The current owner is exiting/execing and there are no other
456 * candidates. Do not leave the mm pointing to a possibly
457 * freed task structure.
458 */
459 if (atomic_read(&mm->mm_users) <= 1) {
460 WRITE_ONCE(mm->owner, NULL);
461 return;
462 }
463
464 read_lock(&tasklist_lock);
465 /*
466 * Search in the children
467 */
468 list_for_each_entry(c, &p->children, sibling) {
469 if (c->mm == mm)
470 goto assign_new_owner;
471 }
472
473 /*
474 * Search in the siblings
475 */
476 list_for_each_entry(c, &p->real_parent->children, sibling) {
477 if (c->mm == mm)
478 goto assign_new_owner;
479 }
480
481 /*
482 * Search through everything else, we should not get here often.
483 */
484 for_each_process(g) {
485 if (g->flags & PF_KTHREAD)
486 continue;
487 for_each_thread(g, c) {
488 if (c->mm == mm)
489 goto assign_new_owner;
490 if (c->mm)
491 break;
492 }
493 }
494 read_unlock(&tasklist_lock);
495 /*
496 * We found no owner yet mm_users > 1: this implies that we are
497 * most likely racing with swapoff (try_to_unuse()) or /proc or
498 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
499 */
500 WRITE_ONCE(mm->owner, NULL);
501 return;
502
503assign_new_owner:
504 BUG_ON(c == p);
505 get_task_struct(c);
506 /*
507 * The task_lock protects c->mm from changing.
508 * We always want mm->owner->mm == mm
509 */
510 task_lock(c);
511 /*
512 * Delay read_unlock() till we have the task_lock()
513 * to ensure that c does not slip away underneath us
514 */
515 read_unlock(&tasklist_lock);
516 if (c->mm != mm) {
517 task_unlock(c);
518 put_task_struct(c);
519 goto retry;
520 }
521 WRITE_ONCE(mm->owner, c);
522 lru_gen_migrate_mm(mm);
523 task_unlock(c);
524 put_task_struct(c);
525}
526#endif /* CONFIG_MEMCG */
527
528/*
529 * Turn us into a lazy TLB process if we
530 * aren't already..
531 */
532static void exit_mm(void)
533{
534 struct mm_struct *mm = current->mm;
535
536 exit_mm_release(current, mm);
537 if (!mm)
538 return;
539 sync_mm_rss(mm);
540 mmap_read_lock(mm);
541 mmgrab_lazy_tlb(mm);
542 BUG_ON(mm != current->active_mm);
543 /* more a memory barrier than a real lock */
544 task_lock(current);
545 /*
546 * When a thread stops operating on an address space, the loop
547 * in membarrier_private_expedited() may not observe that
548 * tsk->mm, and the loop in membarrier_global_expedited() may
549 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
550 * rq->membarrier_state, so those would not issue an IPI.
551 * Membarrier requires a memory barrier after accessing
552 * user-space memory, before clearing tsk->mm or the
553 * rq->membarrier_state.
554 */
555 smp_mb__after_spinlock();
556 local_irq_disable();
557 current->mm = NULL;
558 membarrier_update_current_mm(NULL);
559 enter_lazy_tlb(mm, current);
560 local_irq_enable();
561 task_unlock(current);
562 mmap_read_unlock(mm);
563 mm_update_next_owner(mm);
564 mmput(mm);
565 if (test_thread_flag(TIF_MEMDIE))
566 exit_oom_victim();
567}
568
569static struct task_struct *find_alive_thread(struct task_struct *p)
570{
571 struct task_struct *t;
572
573 for_each_thread(p, t) {
574 if (!(t->flags & PF_EXITING))
575 return t;
576 }
577 return NULL;
578}
579
580static struct task_struct *find_child_reaper(struct task_struct *father,
581 struct list_head *dead)
582 __releases(&tasklist_lock)
583 __acquires(&tasklist_lock)
584{
585 struct pid_namespace *pid_ns = task_active_pid_ns(father);
586 struct task_struct *reaper = pid_ns->child_reaper;
587 struct task_struct *p, *n;
588
589 if (likely(reaper != father))
590 return reaper;
591
592 reaper = find_alive_thread(father);
593 if (reaper) {
594 pid_ns->child_reaper = reaper;
595 return reaper;
596 }
597
598 write_unlock_irq(&tasklist_lock);
599
600 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
601 list_del_init(&p->ptrace_entry);
602 release_task(p);
603 }
604
605 zap_pid_ns_processes(pid_ns);
606 write_lock_irq(&tasklist_lock);
607
608 return father;
609}
610
611/*
612 * When we die, we re-parent all our children, and try to:
613 * 1. give them to another thread in our thread group, if such a member exists
614 * 2. give it to the first ancestor process which prctl'd itself as a
615 * child_subreaper for its children (like a service manager)
616 * 3. give it to the init process (PID 1) in our pid namespace
617 */
618static struct task_struct *find_new_reaper(struct task_struct *father,
619 struct task_struct *child_reaper)
620{
621 struct task_struct *thread, *reaper;
622
623 thread = find_alive_thread(father);
624 if (thread)
625 return thread;
626
627 if (father->signal->has_child_subreaper) {
628 unsigned int ns_level = task_pid(father)->level;
629 /*
630 * Find the first ->is_child_subreaper ancestor in our pid_ns.
631 * We can't check reaper != child_reaper to ensure we do not
632 * cross the namespaces, the exiting parent could be injected
633 * by setns() + fork().
634 * We check pid->level, this is slightly more efficient than
635 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
636 */
637 for (reaper = father->real_parent;
638 task_pid(reaper)->level == ns_level;
639 reaper = reaper->real_parent) {
640 if (reaper == &init_task)
641 break;
642 if (!reaper->signal->is_child_subreaper)
643 continue;
644 thread = find_alive_thread(reaper);
645 if (thread)
646 return thread;
647 }
648 }
649
650 return child_reaper;
651}
652
653/*
654* Any that need to be release_task'd are put on the @dead list.
655 */
656static void reparent_leader(struct task_struct *father, struct task_struct *p,
657 struct list_head *dead)
658{
659 if (unlikely(p->exit_state == EXIT_DEAD))
660 return;
661
662 /* We don't want people slaying init. */
663 p->exit_signal = SIGCHLD;
664
665 /* If it has exited notify the new parent about this child's death. */
666 if (!p->ptrace &&
667 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
668 if (do_notify_parent(p, p->exit_signal)) {
669 p->exit_state = EXIT_DEAD;
670 list_add(&p->ptrace_entry, dead);
671 }
672 }
673
674 kill_orphaned_pgrp(p, father);
675}
676
677/*
678 * This does two things:
679 *
680 * A. Make init inherit all the child processes
681 * B. Check to see if any process groups have become orphaned
682 * as a result of our exiting, and if they have any stopped
683 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
684 */
685static void forget_original_parent(struct task_struct *father,
686 struct list_head *dead)
687{
688 struct task_struct *p, *t, *reaper;
689
690 if (unlikely(!list_empty(&father->ptraced)))
691 exit_ptrace(father, dead);
692
693 /* Can drop and reacquire tasklist_lock */
694 reaper = find_child_reaper(father, dead);
695 if (list_empty(&father->children))
696 return;
697
698 reaper = find_new_reaper(father, reaper);
699 list_for_each_entry(p, &father->children, sibling) {
700 for_each_thread(p, t) {
701 RCU_INIT_POINTER(t->real_parent, reaper);
702 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
703 if (likely(!t->ptrace))
704 t->parent = t->real_parent;
705 if (t->pdeath_signal)
706 group_send_sig_info(t->pdeath_signal,
707 SEND_SIG_NOINFO, t,
708 PIDTYPE_TGID);
709 }
710 /*
711 * If this is a threaded reparent there is no need to
712 * notify anyone anything has happened.
713 */
714 if (!same_thread_group(reaper, father))
715 reparent_leader(father, p, dead);
716 }
717 list_splice_tail_init(&father->children, &reaper->children);
718}
719
720/*
721 * Send signals to all our closest relatives so that they know
722 * to properly mourn us..
723 */
724static void exit_notify(struct task_struct *tsk, int group_dead)
725{
726 bool autoreap;
727 struct task_struct *p, *n;
728 LIST_HEAD(dead);
729
730 write_lock_irq(&tasklist_lock);
731 forget_original_parent(tsk, &dead);
732
733 if (group_dead)
734 kill_orphaned_pgrp(tsk->group_leader, NULL);
735
736 tsk->exit_state = EXIT_ZOMBIE;
737 if (unlikely(tsk->ptrace)) {
738 int sig = thread_group_leader(tsk) &&
739 thread_group_empty(tsk) &&
740 !ptrace_reparented(tsk) ?
741 tsk->exit_signal : SIGCHLD;
742 autoreap = do_notify_parent(tsk, sig);
743 } else if (thread_group_leader(tsk)) {
744 autoreap = thread_group_empty(tsk) &&
745 do_notify_parent(tsk, tsk->exit_signal);
746 } else {
747 autoreap = true;
748 }
749
750 if (autoreap) {
751 tsk->exit_state = EXIT_DEAD;
752 list_add(&tsk->ptrace_entry, &dead);
753 }
754
755 /* mt-exec, de_thread() is waiting for group leader */
756 if (unlikely(tsk->signal->notify_count < 0))
757 wake_up_process(tsk->signal->group_exec_task);
758 write_unlock_irq(&tasklist_lock);
759
760 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
761 list_del_init(&p->ptrace_entry);
762 release_task(p);
763 }
764}
765
766#ifdef CONFIG_DEBUG_STACK_USAGE
767static void check_stack_usage(void)
768{
769 static DEFINE_SPINLOCK(low_water_lock);
770 static int lowest_to_date = THREAD_SIZE;
771 unsigned long free;
772
773 free = stack_not_used(current);
774
775 if (free >= lowest_to_date)
776 return;
777
778 spin_lock(&low_water_lock);
779 if (free < lowest_to_date) {
780 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
781 current->comm, task_pid_nr(current), free);
782 lowest_to_date = free;
783 }
784 spin_unlock(&low_water_lock);
785}
786#else
787static inline void check_stack_usage(void) {}
788#endif
789
790static void synchronize_group_exit(struct task_struct *tsk, long code)
791{
792 struct sighand_struct *sighand = tsk->sighand;
793 struct signal_struct *signal = tsk->signal;
794
795 spin_lock_irq(&sighand->siglock);
796 signal->quick_threads--;
797 if ((signal->quick_threads == 0) &&
798 !(signal->flags & SIGNAL_GROUP_EXIT)) {
799 signal->flags = SIGNAL_GROUP_EXIT;
800 signal->group_exit_code = code;
801 signal->group_stop_count = 0;
802 }
803 spin_unlock_irq(&sighand->siglock);
804}
805
806void __noreturn do_exit(long code)
807{
808 struct task_struct *tsk = current;
809 int group_dead;
810
811 WARN_ON(irqs_disabled());
812
813 synchronize_group_exit(tsk, code);
814
815 WARN_ON(tsk->plug);
816
817 kcov_task_exit(tsk);
818 kmsan_task_exit(tsk);
819
820 coredump_task_exit(tsk);
821 ptrace_event(PTRACE_EVENT_EXIT, code);
822 user_events_exit(tsk);
823
824 validate_creds_for_do_exit(tsk);
825
826 io_uring_files_cancel();
827 exit_signals(tsk); /* sets PF_EXITING */
828
829 /* sync mm's RSS info before statistics gathering */
830 if (tsk->mm)
831 sync_mm_rss(tsk->mm);
832 acct_update_integrals(tsk);
833 group_dead = atomic_dec_and_test(&tsk->signal->live);
834 if (group_dead) {
835 /*
836 * If the last thread of global init has exited, panic
837 * immediately to get a useable coredump.
838 */
839 if (unlikely(is_global_init(tsk)))
840 panic("Attempted to kill init! exitcode=0x%08x\n",
841 tsk->signal->group_exit_code ?: (int)code);
842
843#ifdef CONFIG_POSIX_TIMERS
844 hrtimer_cancel(&tsk->signal->real_timer);
845 exit_itimers(tsk);
846#endif
847 if (tsk->mm)
848 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
849 }
850 acct_collect(code, group_dead);
851 if (group_dead)
852 tty_audit_exit();
853 audit_free(tsk);
854
855 tsk->exit_code = code;
856 taskstats_exit(tsk, group_dead);
857
858 exit_mm();
859
860 if (group_dead)
861 acct_process();
862 trace_sched_process_exit(tsk);
863
864 exit_sem(tsk);
865 exit_shm(tsk);
866 exit_files(tsk);
867 exit_fs(tsk);
868 if (group_dead)
869 disassociate_ctty(1);
870 exit_task_namespaces(tsk);
871 exit_task_work(tsk);
872 exit_thread(tsk);
873
874 /*
875 * Flush inherited counters to the parent - before the parent
876 * gets woken up by child-exit notifications.
877 *
878 * because of cgroup mode, must be called before cgroup_exit()
879 */
880 perf_event_exit_task(tsk);
881
882 sched_autogroup_exit_task(tsk);
883 cgroup_exit(tsk);
884
885 /*
886 * FIXME: do that only when needed, using sched_exit tracepoint
887 */
888 flush_ptrace_hw_breakpoint(tsk);
889
890 exit_tasks_rcu_start();
891 exit_notify(tsk, group_dead);
892 proc_exit_connector(tsk);
893 mpol_put_task_policy(tsk);
894#ifdef CONFIG_FUTEX
895 if (unlikely(current->pi_state_cache))
896 kfree(current->pi_state_cache);
897#endif
898 /*
899 * Make sure we are holding no locks:
900 */
901 debug_check_no_locks_held();
902
903 if (tsk->io_context)
904 exit_io_context(tsk);
905
906 if (tsk->splice_pipe)
907 free_pipe_info(tsk->splice_pipe);
908
909 if (tsk->task_frag.page)
910 put_page(tsk->task_frag.page);
911
912 validate_creds_for_do_exit(tsk);
913 exit_task_stack_account(tsk);
914
915 check_stack_usage();
916 preempt_disable();
917 if (tsk->nr_dirtied)
918 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
919 exit_rcu();
920 exit_tasks_rcu_finish();
921
922 lockdep_free_task(tsk);
923 do_task_dead();
924}
925
926void __noreturn make_task_dead(int signr)
927{
928 /*
929 * Take the task off the cpu after something catastrophic has
930 * happened.
931 *
932 * We can get here from a kernel oops, sometimes with preemption off.
933 * Start by checking for critical errors.
934 * Then fix up important state like USER_DS and preemption.
935 * Then do everything else.
936 */
937 struct task_struct *tsk = current;
938 unsigned int limit;
939
940 if (unlikely(in_interrupt()))
941 panic("Aiee, killing interrupt handler!");
942 if (unlikely(!tsk->pid))
943 panic("Attempted to kill the idle task!");
944
945 if (unlikely(irqs_disabled())) {
946 pr_info("note: %s[%d] exited with irqs disabled\n",
947 current->comm, task_pid_nr(current));
948 local_irq_enable();
949 }
950 if (unlikely(in_atomic())) {
951 pr_info("note: %s[%d] exited with preempt_count %d\n",
952 current->comm, task_pid_nr(current),
953 preempt_count());
954 preempt_count_set(PREEMPT_ENABLED);
955 }
956
957 /*
958 * Every time the system oopses, if the oops happens while a reference
959 * to an object was held, the reference leaks.
960 * If the oops doesn't also leak memory, repeated oopsing can cause
961 * reference counters to wrap around (if they're not using refcount_t).
962 * This means that repeated oopsing can make unexploitable-looking bugs
963 * exploitable through repeated oopsing.
964 * To make sure this can't happen, place an upper bound on how often the
965 * kernel may oops without panic().
966 */
967 limit = READ_ONCE(oops_limit);
968 if (atomic_inc_return(&oops_count) >= limit && limit)
969 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
970
971 /*
972 * We're taking recursive faults here in make_task_dead. Safest is to just
973 * leave this task alone and wait for reboot.
974 */
975 if (unlikely(tsk->flags & PF_EXITING)) {
976 pr_alert("Fixing recursive fault but reboot is needed!\n");
977 futex_exit_recursive(tsk);
978 tsk->exit_state = EXIT_DEAD;
979 refcount_inc(&tsk->rcu_users);
980 do_task_dead();
981 }
982
983 do_exit(signr);
984}
985
986SYSCALL_DEFINE1(exit, int, error_code)
987{
988 do_exit((error_code&0xff)<<8);
989}
990
991/*
992 * Take down every thread in the group. This is called by fatal signals
993 * as well as by sys_exit_group (below).
994 */
995void __noreturn
996do_group_exit(int exit_code)
997{
998 struct signal_struct *sig = current->signal;
999
1000 if (sig->flags & SIGNAL_GROUP_EXIT)
1001 exit_code = sig->group_exit_code;
1002 else if (sig->group_exec_task)
1003 exit_code = 0;
1004 else {
1005 struct sighand_struct *const sighand = current->sighand;
1006
1007 spin_lock_irq(&sighand->siglock);
1008 if (sig->flags & SIGNAL_GROUP_EXIT)
1009 /* Another thread got here before we took the lock. */
1010 exit_code = sig->group_exit_code;
1011 else if (sig->group_exec_task)
1012 exit_code = 0;
1013 else {
1014 sig->group_exit_code = exit_code;
1015 sig->flags = SIGNAL_GROUP_EXIT;
1016 zap_other_threads(current);
1017 }
1018 spin_unlock_irq(&sighand->siglock);
1019 }
1020
1021 do_exit(exit_code);
1022 /* NOTREACHED */
1023}
1024
1025/*
1026 * this kills every thread in the thread group. Note that any externally
1027 * wait4()-ing process will get the correct exit code - even if this
1028 * thread is not the thread group leader.
1029 */
1030SYSCALL_DEFINE1(exit_group, int, error_code)
1031{
1032 do_group_exit((error_code & 0xff) << 8);
1033 /* NOTREACHED */
1034 return 0;
1035}
1036
1037struct waitid_info {
1038 pid_t pid;
1039 uid_t uid;
1040 int status;
1041 int cause;
1042};
1043
1044struct wait_opts {
1045 enum pid_type wo_type;
1046 int wo_flags;
1047 struct pid *wo_pid;
1048
1049 struct waitid_info *wo_info;
1050 int wo_stat;
1051 struct rusage *wo_rusage;
1052
1053 wait_queue_entry_t child_wait;
1054 int notask_error;
1055};
1056
1057static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1058{
1059 return wo->wo_type == PIDTYPE_MAX ||
1060 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1061}
1062
1063static int
1064eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1065{
1066 if (!eligible_pid(wo, p))
1067 return 0;
1068
1069 /*
1070 * Wait for all children (clone and not) if __WALL is set or
1071 * if it is traced by us.
1072 */
1073 if (ptrace || (wo->wo_flags & __WALL))
1074 return 1;
1075
1076 /*
1077 * Otherwise, wait for clone children *only* if __WCLONE is set;
1078 * otherwise, wait for non-clone children *only*.
1079 *
1080 * Note: a "clone" child here is one that reports to its parent
1081 * using a signal other than SIGCHLD, or a non-leader thread which
1082 * we can only see if it is traced by us.
1083 */
1084 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1085 return 0;
1086
1087 return 1;
1088}
1089
1090/*
1091 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1092 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1093 * the lock and this task is uninteresting. If we return nonzero, we have
1094 * released the lock and the system call should return.
1095 */
1096static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1097{
1098 int state, status;
1099 pid_t pid = task_pid_vnr(p);
1100 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1101 struct waitid_info *infop;
1102
1103 if (!likely(wo->wo_flags & WEXITED))
1104 return 0;
1105
1106 if (unlikely(wo->wo_flags & WNOWAIT)) {
1107 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1108 ? p->signal->group_exit_code : p->exit_code;
1109 get_task_struct(p);
1110 read_unlock(&tasklist_lock);
1111 sched_annotate_sleep();
1112 if (wo->wo_rusage)
1113 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1114 put_task_struct(p);
1115 goto out_info;
1116 }
1117 /*
1118 * Move the task's state to DEAD/TRACE, only one thread can do this.
1119 */
1120 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1121 EXIT_TRACE : EXIT_DEAD;
1122 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1123 return 0;
1124 /*
1125 * We own this thread, nobody else can reap it.
1126 */
1127 read_unlock(&tasklist_lock);
1128 sched_annotate_sleep();
1129
1130 /*
1131 * Check thread_group_leader() to exclude the traced sub-threads.
1132 */
1133 if (state == EXIT_DEAD && thread_group_leader(p)) {
1134 struct signal_struct *sig = p->signal;
1135 struct signal_struct *psig = current->signal;
1136 unsigned long maxrss;
1137 u64 tgutime, tgstime;
1138
1139 /*
1140 * The resource counters for the group leader are in its
1141 * own task_struct. Those for dead threads in the group
1142 * are in its signal_struct, as are those for the child
1143 * processes it has previously reaped. All these
1144 * accumulate in the parent's signal_struct c* fields.
1145 *
1146 * We don't bother to take a lock here to protect these
1147 * p->signal fields because the whole thread group is dead
1148 * and nobody can change them.
1149 *
1150 * psig->stats_lock also protects us from our sub-threads
1151 * which can reap other children at the same time. Until
1152 * we change k_getrusage()-like users to rely on this lock
1153 * we have to take ->siglock as well.
1154 *
1155 * We use thread_group_cputime_adjusted() to get times for
1156 * the thread group, which consolidates times for all threads
1157 * in the group including the group leader.
1158 */
1159 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1160 spin_lock_irq(&current->sighand->siglock);
1161 write_seqlock(&psig->stats_lock);
1162 psig->cutime += tgutime + sig->cutime;
1163 psig->cstime += tgstime + sig->cstime;
1164 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1165 psig->cmin_flt +=
1166 p->min_flt + sig->min_flt + sig->cmin_flt;
1167 psig->cmaj_flt +=
1168 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1169 psig->cnvcsw +=
1170 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1171 psig->cnivcsw +=
1172 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1173 psig->cinblock +=
1174 task_io_get_inblock(p) +
1175 sig->inblock + sig->cinblock;
1176 psig->coublock +=
1177 task_io_get_oublock(p) +
1178 sig->oublock + sig->coublock;
1179 maxrss = max(sig->maxrss, sig->cmaxrss);
1180 if (psig->cmaxrss < maxrss)
1181 psig->cmaxrss = maxrss;
1182 task_io_accounting_add(&psig->ioac, &p->ioac);
1183 task_io_accounting_add(&psig->ioac, &sig->ioac);
1184 write_sequnlock(&psig->stats_lock);
1185 spin_unlock_irq(&current->sighand->siglock);
1186 }
1187
1188 if (wo->wo_rusage)
1189 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1190 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1191 ? p->signal->group_exit_code : p->exit_code;
1192 wo->wo_stat = status;
1193
1194 if (state == EXIT_TRACE) {
1195 write_lock_irq(&tasklist_lock);
1196 /* We dropped tasklist, ptracer could die and untrace */
1197 ptrace_unlink(p);
1198
1199 /* If parent wants a zombie, don't release it now */
1200 state = EXIT_ZOMBIE;
1201 if (do_notify_parent(p, p->exit_signal))
1202 state = EXIT_DEAD;
1203 p->exit_state = state;
1204 write_unlock_irq(&tasklist_lock);
1205 }
1206 if (state == EXIT_DEAD)
1207 release_task(p);
1208
1209out_info:
1210 infop = wo->wo_info;
1211 if (infop) {
1212 if ((status & 0x7f) == 0) {
1213 infop->cause = CLD_EXITED;
1214 infop->status = status >> 8;
1215 } else {
1216 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1217 infop->status = status & 0x7f;
1218 }
1219 infop->pid = pid;
1220 infop->uid = uid;
1221 }
1222
1223 return pid;
1224}
1225
1226static int *task_stopped_code(struct task_struct *p, bool ptrace)
1227{
1228 if (ptrace) {
1229 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1230 return &p->exit_code;
1231 } else {
1232 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1233 return &p->signal->group_exit_code;
1234 }
1235 return NULL;
1236}
1237
1238/**
1239 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1240 * @wo: wait options
1241 * @ptrace: is the wait for ptrace
1242 * @p: task to wait for
1243 *
1244 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1245 *
1246 * CONTEXT:
1247 * read_lock(&tasklist_lock), which is released if return value is
1248 * non-zero. Also, grabs and releases @p->sighand->siglock.
1249 *
1250 * RETURNS:
1251 * 0 if wait condition didn't exist and search for other wait conditions
1252 * should continue. Non-zero return, -errno on failure and @p's pid on
1253 * success, implies that tasklist_lock is released and wait condition
1254 * search should terminate.
1255 */
1256static int wait_task_stopped(struct wait_opts *wo,
1257 int ptrace, struct task_struct *p)
1258{
1259 struct waitid_info *infop;
1260 int exit_code, *p_code, why;
1261 uid_t uid = 0; /* unneeded, required by compiler */
1262 pid_t pid;
1263
1264 /*
1265 * Traditionally we see ptrace'd stopped tasks regardless of options.
1266 */
1267 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1268 return 0;
1269
1270 if (!task_stopped_code(p, ptrace))
1271 return 0;
1272
1273 exit_code = 0;
1274 spin_lock_irq(&p->sighand->siglock);
1275
1276 p_code = task_stopped_code(p, ptrace);
1277 if (unlikely(!p_code))
1278 goto unlock_sig;
1279
1280 exit_code = *p_code;
1281 if (!exit_code)
1282 goto unlock_sig;
1283
1284 if (!unlikely(wo->wo_flags & WNOWAIT))
1285 *p_code = 0;
1286
1287 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1288unlock_sig:
1289 spin_unlock_irq(&p->sighand->siglock);
1290 if (!exit_code)
1291 return 0;
1292
1293 /*
1294 * Now we are pretty sure this task is interesting.
1295 * Make sure it doesn't get reaped out from under us while we
1296 * give up the lock and then examine it below. We don't want to
1297 * keep holding onto the tasklist_lock while we call getrusage and
1298 * possibly take page faults for user memory.
1299 */
1300 get_task_struct(p);
1301 pid = task_pid_vnr(p);
1302 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1303 read_unlock(&tasklist_lock);
1304 sched_annotate_sleep();
1305 if (wo->wo_rusage)
1306 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1307 put_task_struct(p);
1308
1309 if (likely(!(wo->wo_flags & WNOWAIT)))
1310 wo->wo_stat = (exit_code << 8) | 0x7f;
1311
1312 infop = wo->wo_info;
1313 if (infop) {
1314 infop->cause = why;
1315 infop->status = exit_code;
1316 infop->pid = pid;
1317 infop->uid = uid;
1318 }
1319 return pid;
1320}
1321
1322/*
1323 * Handle do_wait work for one task in a live, non-stopped state.
1324 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1325 * the lock and this task is uninteresting. If we return nonzero, we have
1326 * released the lock and the system call should return.
1327 */
1328static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1329{
1330 struct waitid_info *infop;
1331 pid_t pid;
1332 uid_t uid;
1333
1334 if (!unlikely(wo->wo_flags & WCONTINUED))
1335 return 0;
1336
1337 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1338 return 0;
1339
1340 spin_lock_irq(&p->sighand->siglock);
1341 /* Re-check with the lock held. */
1342 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1343 spin_unlock_irq(&p->sighand->siglock);
1344 return 0;
1345 }
1346 if (!unlikely(wo->wo_flags & WNOWAIT))
1347 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1348 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1349 spin_unlock_irq(&p->sighand->siglock);
1350
1351 pid = task_pid_vnr(p);
1352 get_task_struct(p);
1353 read_unlock(&tasklist_lock);
1354 sched_annotate_sleep();
1355 if (wo->wo_rusage)
1356 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1357 put_task_struct(p);
1358
1359 infop = wo->wo_info;
1360 if (!infop) {
1361 wo->wo_stat = 0xffff;
1362 } else {
1363 infop->cause = CLD_CONTINUED;
1364 infop->pid = pid;
1365 infop->uid = uid;
1366 infop->status = SIGCONT;
1367 }
1368 return pid;
1369}
1370
1371/*
1372 * Consider @p for a wait by @parent.
1373 *
1374 * -ECHILD should be in ->notask_error before the first call.
1375 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1376 * Returns zero if the search for a child should continue;
1377 * then ->notask_error is 0 if @p is an eligible child,
1378 * or still -ECHILD.
1379 */
1380static int wait_consider_task(struct wait_opts *wo, int ptrace,
1381 struct task_struct *p)
1382{
1383 /*
1384 * We can race with wait_task_zombie() from another thread.
1385 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1386 * can't confuse the checks below.
1387 */
1388 int exit_state = READ_ONCE(p->exit_state);
1389 int ret;
1390
1391 if (unlikely(exit_state == EXIT_DEAD))
1392 return 0;
1393
1394 ret = eligible_child(wo, ptrace, p);
1395 if (!ret)
1396 return ret;
1397
1398 if (unlikely(exit_state == EXIT_TRACE)) {
1399 /*
1400 * ptrace == 0 means we are the natural parent. In this case
1401 * we should clear notask_error, debugger will notify us.
1402 */
1403 if (likely(!ptrace))
1404 wo->notask_error = 0;
1405 return 0;
1406 }
1407
1408 if (likely(!ptrace) && unlikely(p->ptrace)) {
1409 /*
1410 * If it is traced by its real parent's group, just pretend
1411 * the caller is ptrace_do_wait() and reap this child if it
1412 * is zombie.
1413 *
1414 * This also hides group stop state from real parent; otherwise
1415 * a single stop can be reported twice as group and ptrace stop.
1416 * If a ptracer wants to distinguish these two events for its
1417 * own children it should create a separate process which takes
1418 * the role of real parent.
1419 */
1420 if (!ptrace_reparented(p))
1421 ptrace = 1;
1422 }
1423
1424 /* slay zombie? */
1425 if (exit_state == EXIT_ZOMBIE) {
1426 /* we don't reap group leaders with subthreads */
1427 if (!delay_group_leader(p)) {
1428 /*
1429 * A zombie ptracee is only visible to its ptracer.
1430 * Notification and reaping will be cascaded to the
1431 * real parent when the ptracer detaches.
1432 */
1433 if (unlikely(ptrace) || likely(!p->ptrace))
1434 return wait_task_zombie(wo, p);
1435 }
1436
1437 /*
1438 * Allow access to stopped/continued state via zombie by
1439 * falling through. Clearing of notask_error is complex.
1440 *
1441 * When !@ptrace:
1442 *
1443 * If WEXITED is set, notask_error should naturally be
1444 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1445 * so, if there are live subthreads, there are events to
1446 * wait for. If all subthreads are dead, it's still safe
1447 * to clear - this function will be called again in finite
1448 * amount time once all the subthreads are released and
1449 * will then return without clearing.
1450 *
1451 * When @ptrace:
1452 *
1453 * Stopped state is per-task and thus can't change once the
1454 * target task dies. Only continued and exited can happen.
1455 * Clear notask_error if WCONTINUED | WEXITED.
1456 */
1457 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1458 wo->notask_error = 0;
1459 } else {
1460 /*
1461 * @p is alive and it's gonna stop, continue or exit, so
1462 * there always is something to wait for.
1463 */
1464 wo->notask_error = 0;
1465 }
1466
1467 /*
1468 * Wait for stopped. Depending on @ptrace, different stopped state
1469 * is used and the two don't interact with each other.
1470 */
1471 ret = wait_task_stopped(wo, ptrace, p);
1472 if (ret)
1473 return ret;
1474
1475 /*
1476 * Wait for continued. There's only one continued state and the
1477 * ptracer can consume it which can confuse the real parent. Don't
1478 * use WCONTINUED from ptracer. You don't need or want it.
1479 */
1480 return wait_task_continued(wo, p);
1481}
1482
1483/*
1484 * Do the work of do_wait() for one thread in the group, @tsk.
1485 *
1486 * -ECHILD should be in ->notask_error before the first call.
1487 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1488 * Returns zero if the search for a child should continue; then
1489 * ->notask_error is 0 if there were any eligible children,
1490 * or still -ECHILD.
1491 */
1492static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1493{
1494 struct task_struct *p;
1495
1496 list_for_each_entry(p, &tsk->children, sibling) {
1497 int ret = wait_consider_task(wo, 0, p);
1498
1499 if (ret)
1500 return ret;
1501 }
1502
1503 return 0;
1504}
1505
1506static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1507{
1508 struct task_struct *p;
1509
1510 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1511 int ret = wait_consider_task(wo, 1, p);
1512
1513 if (ret)
1514 return ret;
1515 }
1516
1517 return 0;
1518}
1519
1520static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1521 int sync, void *key)
1522{
1523 struct wait_opts *wo = container_of(wait, struct wait_opts,
1524 child_wait);
1525 struct task_struct *p = key;
1526
1527 if (!eligible_pid(wo, p))
1528 return 0;
1529
1530 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1531 return 0;
1532
1533 return default_wake_function(wait, mode, sync, key);
1534}
1535
1536void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1537{
1538 __wake_up_sync_key(&parent->signal->wait_chldexit,
1539 TASK_INTERRUPTIBLE, p);
1540}
1541
1542static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1543 struct task_struct *target)
1544{
1545 struct task_struct *parent =
1546 !ptrace ? target->real_parent : target->parent;
1547
1548 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1549 same_thread_group(current, parent));
1550}
1551
1552/*
1553 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1554 * and tracee lists to find the target task.
1555 */
1556static int do_wait_pid(struct wait_opts *wo)
1557{
1558 bool ptrace;
1559 struct task_struct *target;
1560 int retval;
1561
1562 ptrace = false;
1563 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1564 if (target && is_effectively_child(wo, ptrace, target)) {
1565 retval = wait_consider_task(wo, ptrace, target);
1566 if (retval)
1567 return retval;
1568 }
1569
1570 ptrace = true;
1571 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1572 if (target && target->ptrace &&
1573 is_effectively_child(wo, ptrace, target)) {
1574 retval = wait_consider_task(wo, ptrace, target);
1575 if (retval)
1576 return retval;
1577 }
1578
1579 return 0;
1580}
1581
1582static long do_wait(struct wait_opts *wo)
1583{
1584 int retval;
1585
1586 trace_sched_process_wait(wo->wo_pid);
1587
1588 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1589 wo->child_wait.private = current;
1590 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1591repeat:
1592 /*
1593 * If there is nothing that can match our criteria, just get out.
1594 * We will clear ->notask_error to zero if we see any child that
1595 * might later match our criteria, even if we are not able to reap
1596 * it yet.
1597 */
1598 wo->notask_error = -ECHILD;
1599 if ((wo->wo_type < PIDTYPE_MAX) &&
1600 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1601 goto notask;
1602
1603 set_current_state(TASK_INTERRUPTIBLE);
1604 read_lock(&tasklist_lock);
1605
1606 if (wo->wo_type == PIDTYPE_PID) {
1607 retval = do_wait_pid(wo);
1608 if (retval)
1609 goto end;
1610 } else {
1611 struct task_struct *tsk = current;
1612
1613 do {
1614 retval = do_wait_thread(wo, tsk);
1615 if (retval)
1616 goto end;
1617
1618 retval = ptrace_do_wait(wo, tsk);
1619 if (retval)
1620 goto end;
1621
1622 if (wo->wo_flags & __WNOTHREAD)
1623 break;
1624 } while_each_thread(current, tsk);
1625 }
1626 read_unlock(&tasklist_lock);
1627
1628notask:
1629 retval = wo->notask_error;
1630 if (!retval && !(wo->wo_flags & WNOHANG)) {
1631 retval = -ERESTARTSYS;
1632 if (!signal_pending(current)) {
1633 schedule();
1634 goto repeat;
1635 }
1636 }
1637end:
1638 __set_current_state(TASK_RUNNING);
1639 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1640 return retval;
1641}
1642
1643static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1644 int options, struct rusage *ru)
1645{
1646 struct wait_opts wo;
1647 struct pid *pid = NULL;
1648 enum pid_type type;
1649 long ret;
1650 unsigned int f_flags = 0;
1651
1652 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1653 __WNOTHREAD|__WCLONE|__WALL))
1654 return -EINVAL;
1655 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1656 return -EINVAL;
1657
1658 switch (which) {
1659 case P_ALL:
1660 type = PIDTYPE_MAX;
1661 break;
1662 case P_PID:
1663 type = PIDTYPE_PID;
1664 if (upid <= 0)
1665 return -EINVAL;
1666
1667 pid = find_get_pid(upid);
1668 break;
1669 case P_PGID:
1670 type = PIDTYPE_PGID;
1671 if (upid < 0)
1672 return -EINVAL;
1673
1674 if (upid)
1675 pid = find_get_pid(upid);
1676 else
1677 pid = get_task_pid(current, PIDTYPE_PGID);
1678 break;
1679 case P_PIDFD:
1680 type = PIDTYPE_PID;
1681 if (upid < 0)
1682 return -EINVAL;
1683
1684 pid = pidfd_get_pid(upid, &f_flags);
1685 if (IS_ERR(pid))
1686 return PTR_ERR(pid);
1687
1688 break;
1689 default:
1690 return -EINVAL;
1691 }
1692
1693 wo.wo_type = type;
1694 wo.wo_pid = pid;
1695 wo.wo_flags = options;
1696 wo.wo_info = infop;
1697 wo.wo_rusage = ru;
1698 if (f_flags & O_NONBLOCK)
1699 wo.wo_flags |= WNOHANG;
1700
1701 ret = do_wait(&wo);
1702 if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1703 ret = -EAGAIN;
1704
1705 put_pid(pid);
1706 return ret;
1707}
1708
1709SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1710 infop, int, options, struct rusage __user *, ru)
1711{
1712 struct rusage r;
1713 struct waitid_info info = {.status = 0};
1714 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1715 int signo = 0;
1716
1717 if (err > 0) {
1718 signo = SIGCHLD;
1719 err = 0;
1720 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1721 return -EFAULT;
1722 }
1723 if (!infop)
1724 return err;
1725
1726 if (!user_write_access_begin(infop, sizeof(*infop)))
1727 return -EFAULT;
1728
1729 unsafe_put_user(signo, &infop->si_signo, Efault);
1730 unsafe_put_user(0, &infop->si_errno, Efault);
1731 unsafe_put_user(info.cause, &infop->si_code, Efault);
1732 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1733 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1734 unsafe_put_user(info.status, &infop->si_status, Efault);
1735 user_write_access_end();
1736 return err;
1737Efault:
1738 user_write_access_end();
1739 return -EFAULT;
1740}
1741
1742long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1743 struct rusage *ru)
1744{
1745 struct wait_opts wo;
1746 struct pid *pid = NULL;
1747 enum pid_type type;
1748 long ret;
1749
1750 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1751 __WNOTHREAD|__WCLONE|__WALL))
1752 return -EINVAL;
1753
1754 /* -INT_MIN is not defined */
1755 if (upid == INT_MIN)
1756 return -ESRCH;
1757
1758 if (upid == -1)
1759 type = PIDTYPE_MAX;
1760 else if (upid < 0) {
1761 type = PIDTYPE_PGID;
1762 pid = find_get_pid(-upid);
1763 } else if (upid == 0) {
1764 type = PIDTYPE_PGID;
1765 pid = get_task_pid(current, PIDTYPE_PGID);
1766 } else /* upid > 0 */ {
1767 type = PIDTYPE_PID;
1768 pid = find_get_pid(upid);
1769 }
1770
1771 wo.wo_type = type;
1772 wo.wo_pid = pid;
1773 wo.wo_flags = options | WEXITED;
1774 wo.wo_info = NULL;
1775 wo.wo_stat = 0;
1776 wo.wo_rusage = ru;
1777 ret = do_wait(&wo);
1778 put_pid(pid);
1779 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1780 ret = -EFAULT;
1781
1782 return ret;
1783}
1784
1785int kernel_wait(pid_t pid, int *stat)
1786{
1787 struct wait_opts wo = {
1788 .wo_type = PIDTYPE_PID,
1789 .wo_pid = find_get_pid(pid),
1790 .wo_flags = WEXITED,
1791 };
1792 int ret;
1793
1794 ret = do_wait(&wo);
1795 if (ret > 0 && wo.wo_stat)
1796 *stat = wo.wo_stat;
1797 put_pid(wo.wo_pid);
1798 return ret;
1799}
1800
1801SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1802 int, options, struct rusage __user *, ru)
1803{
1804 struct rusage r;
1805 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1806
1807 if (err > 0) {
1808 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1809 return -EFAULT;
1810 }
1811 return err;
1812}
1813
1814#ifdef __ARCH_WANT_SYS_WAITPID
1815
1816/*
1817 * sys_waitpid() remains for compatibility. waitpid() should be
1818 * implemented by calling sys_wait4() from libc.a.
1819 */
1820SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1821{
1822 return kernel_wait4(pid, stat_addr, options, NULL);
1823}
1824
1825#endif
1826
1827#ifdef CONFIG_COMPAT
1828COMPAT_SYSCALL_DEFINE4(wait4,
1829 compat_pid_t, pid,
1830 compat_uint_t __user *, stat_addr,
1831 int, options,
1832 struct compat_rusage __user *, ru)
1833{
1834 struct rusage r;
1835 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1836 if (err > 0) {
1837 if (ru && put_compat_rusage(&r, ru))
1838 return -EFAULT;
1839 }
1840 return err;
1841}
1842
1843COMPAT_SYSCALL_DEFINE5(waitid,
1844 int, which, compat_pid_t, pid,
1845 struct compat_siginfo __user *, infop, int, options,
1846 struct compat_rusage __user *, uru)
1847{
1848 struct rusage ru;
1849 struct waitid_info info = {.status = 0};
1850 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1851 int signo = 0;
1852 if (err > 0) {
1853 signo = SIGCHLD;
1854 err = 0;
1855 if (uru) {
1856 /* kernel_waitid() overwrites everything in ru */
1857 if (COMPAT_USE_64BIT_TIME)
1858 err = copy_to_user(uru, &ru, sizeof(ru));
1859 else
1860 err = put_compat_rusage(&ru, uru);
1861 if (err)
1862 return -EFAULT;
1863 }
1864 }
1865
1866 if (!infop)
1867 return err;
1868
1869 if (!user_write_access_begin(infop, sizeof(*infop)))
1870 return -EFAULT;
1871
1872 unsafe_put_user(signo, &infop->si_signo, Efault);
1873 unsafe_put_user(0, &infop->si_errno, Efault);
1874 unsafe_put_user(info.cause, &infop->si_code, Efault);
1875 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1876 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1877 unsafe_put_user(info.status, &infop->si_status, Efault);
1878 user_write_access_end();
1879 return err;
1880Efault:
1881 user_write_access_end();
1882 return -EFAULT;
1883}
1884#endif
1885
1886/**
1887 * thread_group_exited - check that a thread group has exited
1888 * @pid: tgid of thread group to be checked.
1889 *
1890 * Test if the thread group represented by tgid has exited (all
1891 * threads are zombies, dead or completely gone).
1892 *
1893 * Return: true if the thread group has exited. false otherwise.
1894 */
1895bool thread_group_exited(struct pid *pid)
1896{
1897 struct task_struct *task;
1898 bool exited;
1899
1900 rcu_read_lock();
1901 task = pid_task(pid, PIDTYPE_PID);
1902 exited = !task ||
1903 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1904 rcu_read_unlock();
1905
1906 return exited;
1907}
1908EXPORT_SYMBOL(thread_group_exited);
1909
1910/*
1911 * This needs to be __function_aligned as GCC implicitly makes any
1912 * implementation of abort() cold and drops alignment specified by
1913 * -falign-functions=N.
1914 *
1915 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1916 */
1917__weak __function_aligned void abort(void)
1918{
1919 BUG();
1920
1921 /* if that doesn't kill us, halt */
1922 panic("Oops failed to kill thread");
1923}
1924EXPORT_SYMBOL(abort);