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