projects/linux·blob v6.7 · refs

1932 lines · 48.0 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#include "exit.h"
78
79/*
80 * The default value should be high enough to not crash a system that randomly
81 * crashes its kernel from time to time, but low enough to at least not permit
82 * overflowing 32-bit refcounts or the ldsem writer count.
83 */
84static unsigned int oops_limit = 10000;
85
86#ifdef CONFIG_SYSCTL
87static struct ctl_table kern_exit_table[] = {
88 {
89 .procname = "oops_limit",
90 .data = &oops_limit,
91 .maxlen = sizeof(oops_limit),
92 .mode = 0644,
93 .proc_handler = proc_douintvec,
94 },
95 { }
96};
97
98static __init int kernel_exit_sysctls_init(void)
99{
100 register_sysctl_init("kernel", kern_exit_table);
101 return 0;
102}
103late_initcall(kernel_exit_sysctls_init);
104#endif
105
106static atomic_t oops_count = ATOMIC_INIT(0);
107
108#ifdef CONFIG_SYSFS
109static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
110 char *page)
111{
112 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
113}
114
115static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
116
117static __init int kernel_exit_sysfs_init(void)
118{
119 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
120 return 0;
121}
122late_initcall(kernel_exit_sysfs_init);
123#endif
124
125static void __unhash_process(struct task_struct *p, bool group_dead)
126{
127 nr_threads--;
128 detach_pid(p, PIDTYPE_PID);
129 if (group_dead) {
130 detach_pid(p, PIDTYPE_TGID);
131 detach_pid(p, PIDTYPE_PGID);
132 detach_pid(p, PIDTYPE_SID);
133
134 list_del_rcu(&p->tasks);
135 list_del_init(&p->sibling);
136 __this_cpu_dec(process_counts);
137 }
138 list_del_rcu(&p->thread_node);
139}
140
141/*
142 * This function expects the tasklist_lock write-locked.
143 */
144static void __exit_signal(struct task_struct *tsk)
145{
146 struct signal_struct *sig = tsk->signal;
147 bool group_dead = thread_group_leader(tsk);
148 struct sighand_struct *sighand;
149 struct tty_struct *tty;
150 u64 utime, stime;
151
152 sighand = rcu_dereference_check(tsk->sighand,
153 lockdep_tasklist_lock_is_held());
154 spin_lock(&sighand->siglock);
155
156#ifdef CONFIG_POSIX_TIMERS
157 posix_cpu_timers_exit(tsk);
158 if (group_dead)
159 posix_cpu_timers_exit_group(tsk);
160#endif
161
162 if (group_dead) {
163 tty = sig->tty;
164 sig->tty = NULL;
165 } else {
166 /*
167 * If there is any task waiting for the group exit
168 * then notify it:
169 */
170 if (sig->notify_count > 0 && !--sig->notify_count)
171 wake_up_process(sig->group_exec_task);
172
173 if (tsk == sig->curr_target)
174 sig->curr_target = next_thread(tsk);
175 }
176
177 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
178 sizeof(unsigned long long));
179
180 /*
181 * Accumulate here the counters for all threads as they die. We could
182 * skip the group leader because it is the last user of signal_struct,
183 * but we want to avoid the race with thread_group_cputime() which can
184 * see the empty ->thread_head list.
185 */
186 task_cputime(tsk, &utime, &stime);
187 write_seqlock(&sig->stats_lock);
188 sig->utime += utime;
189 sig->stime += stime;
190 sig->gtime += task_gtime(tsk);
191 sig->min_flt += tsk->min_flt;
192 sig->maj_flt += tsk->maj_flt;
193 sig->nvcsw += tsk->nvcsw;
194 sig->nivcsw += tsk->nivcsw;
195 sig->inblock += task_io_get_inblock(tsk);
196 sig->oublock += task_io_get_oublock(tsk);
197 task_io_accounting_add(&sig->ioac, &tsk->ioac);
198 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
199 sig->nr_threads--;
200 __unhash_process(tsk, group_dead);
201 write_sequnlock(&sig->stats_lock);
202
203 /*
204 * Do this under ->siglock, we can race with another thread
205 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
206 */
207 flush_sigqueue(&tsk->pending);
208 tsk->sighand = NULL;
209 spin_unlock(&sighand->siglock);
210
211 __cleanup_sighand(sighand);
212 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
213 if (group_dead) {
214 flush_sigqueue(&sig->shared_pending);
215 tty_kref_put(tty);
216 }
217}
218
219static void delayed_put_task_struct(struct rcu_head *rhp)
220{
221 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
222
223 kprobe_flush_task(tsk);
224 rethook_flush_task(tsk);
225 perf_event_delayed_put(tsk);
226 trace_sched_process_free(tsk);
227 put_task_struct(tsk);
228}
229
230void put_task_struct_rcu_user(struct task_struct *task)
231{
232 if (refcount_dec_and_test(&task->rcu_users))
233 call_rcu(&task->rcu, delayed_put_task_struct);
234}
235
236void __weak release_thread(struct task_struct *dead_task)
237{
238}
239
240void release_task(struct task_struct *p)
241{
242 struct task_struct *leader;
243 struct pid *thread_pid;
244 int zap_leader;
245repeat:
246 /* don't need to get the RCU readlock here - the process is dead and
247 * can't be modifying its own credentials. But shut RCU-lockdep up */
248 rcu_read_lock();
249 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
250 rcu_read_unlock();
251
252 cgroup_release(p);
253
254 write_lock_irq(&tasklist_lock);
255 ptrace_release_task(p);
256 thread_pid = get_pid(p->thread_pid);
257 __exit_signal(p);
258
259 /*
260 * If we are the last non-leader member of the thread
261 * group, and the leader is zombie, then notify the
262 * group leader's parent process. (if it wants notification.)
263 */
264 zap_leader = 0;
265 leader = p->group_leader;
266 if (leader != p && thread_group_empty(leader)
267 && leader->exit_state == EXIT_ZOMBIE) {
268 /*
269 * If we were the last child thread and the leader has
270 * exited already, and the leader's parent ignores SIGCHLD,
271 * then we are the one who should release the leader.
272 */
273 zap_leader = do_notify_parent(leader, leader->exit_signal);
274 if (zap_leader)
275 leader->exit_state = EXIT_DEAD;
276 }
277
278 write_unlock_irq(&tasklist_lock);
279 seccomp_filter_release(p);
280 proc_flush_pid(thread_pid);
281 put_pid(thread_pid);
282 release_thread(p);
283 put_task_struct_rcu_user(p);
284
285 p = leader;
286 if (unlikely(zap_leader))
287 goto repeat;
288}
289
290int rcuwait_wake_up(struct rcuwait *w)
291{
292 int ret = 0;
293 struct task_struct *task;
294
295 rcu_read_lock();
296
297 /*
298 * Order condition vs @task, such that everything prior to the load
299 * of @task is visible. This is the condition as to why the user called
300 * rcuwait_wake() in the first place. Pairs with set_current_state()
301 * barrier (A) in rcuwait_wait_event().
302 *
303 * WAIT WAKE
304 * [S] tsk = current [S] cond = true
305 * MB (A) MB (B)
306 * [L] cond [L] tsk
307 */
308 smp_mb(); /* (B) */
309
310 task = rcu_dereference(w->task);
311 if (task)
312 ret = wake_up_process(task);
313 rcu_read_unlock();
314
315 return ret;
316}
317EXPORT_SYMBOL_GPL(rcuwait_wake_up);
318
319/*
320 * Determine if a process group is "orphaned", according to the POSIX
321 * definition in 2.2.2.52. Orphaned process groups are not to be affected
322 * by terminal-generated stop signals. Newly orphaned process groups are
323 * to receive a SIGHUP and a SIGCONT.
324 *
325 * "I ask you, have you ever known what it is to be an orphan?"
326 */
327static int will_become_orphaned_pgrp(struct pid *pgrp,
328 struct task_struct *ignored_task)
329{
330 struct task_struct *p;
331
332 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
333 if ((p == ignored_task) ||
334 (p->exit_state && thread_group_empty(p)) ||
335 is_global_init(p->real_parent))
336 continue;
337
338 if (task_pgrp(p->real_parent) != pgrp &&
339 task_session(p->real_parent) == task_session(p))
340 return 0;
341 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
342
343 return 1;
344}
345
346int is_current_pgrp_orphaned(void)
347{
348 int retval;
349
350 read_lock(&tasklist_lock);
351 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
352 read_unlock(&tasklist_lock);
353
354 return retval;
355}
356
357static bool has_stopped_jobs(struct pid *pgrp)
358{
359 struct task_struct *p;
360
361 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
362 if (p->signal->flags & SIGNAL_STOP_STOPPED)
363 return true;
364 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
365
366 return false;
367}
368
369/*
370 * Check to see if any process groups have become orphaned as
371 * a result of our exiting, and if they have any stopped jobs,
372 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
373 */
374static void
375kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
376{
377 struct pid *pgrp = task_pgrp(tsk);
378 struct task_struct *ignored_task = tsk;
379
380 if (!parent)
381 /* exit: our father is in a different pgrp than
382 * we are and we were the only connection outside.
383 */
384 parent = tsk->real_parent;
385 else
386 /* reparent: our child is in a different pgrp than
387 * we are, and it was the only connection outside.
388 */
389 ignored_task = NULL;
390
391 if (task_pgrp(parent) != pgrp &&
392 task_session(parent) == task_session(tsk) &&
393 will_become_orphaned_pgrp(pgrp, ignored_task) &&
394 has_stopped_jobs(pgrp)) {
395 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
396 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
397 }
398}
399
400static void coredump_task_exit(struct task_struct *tsk)
401{
402 struct core_state *core_state;
403
404 /*
405 * Serialize with any possible pending coredump.
406 * We must hold siglock around checking core_state
407 * and setting PF_POSTCOREDUMP. The core-inducing thread
408 * will increment ->nr_threads for each thread in the
409 * group without PF_POSTCOREDUMP set.
410 */
411 spin_lock_irq(&tsk->sighand->siglock);
412 tsk->flags |= PF_POSTCOREDUMP;
413 core_state = tsk->signal->core_state;
414 spin_unlock_irq(&tsk->sighand->siglock);
415
416 /* The vhost_worker does not particpate in coredumps */
417 if (core_state &&
418 ((tsk->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)) {
419 struct core_thread self;
420
421 self.task = current;
422 if (self.task->flags & PF_SIGNALED)
423 self.next = xchg(&core_state->dumper.next, &self);
424 else
425 self.task = NULL;
426 /*
427 * Implies mb(), the result of xchg() must be visible
428 * to core_state->dumper.
429 */
430 if (atomic_dec_and_test(&core_state->nr_threads))
431 complete(&core_state->startup);
432
433 for (;;) {
434 set_current_state(TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
435 if (!self.task) /* see coredump_finish() */
436 break;
437 schedule();
438 }
439 __set_current_state(TASK_RUNNING);
440 }
441}
442
443#ifdef CONFIG_MEMCG
444/*
445 * A task is exiting. If it owned this mm, find a new owner for the mm.
446 */
447void mm_update_next_owner(struct mm_struct *mm)
448{
449 struct task_struct *c, *g, *p = current;
450
451retry:
452 /*
453 * If the exiting or execing task is not the owner, it's
454 * someone else's problem.
455 */
456 if (mm->owner != p)
457 return;
458 /*
459 * The current owner is exiting/execing and there are no other
460 * candidates. Do not leave the mm pointing to a possibly
461 * freed task structure.
462 */
463 if (atomic_read(&mm->mm_users) <= 1) {
464 WRITE_ONCE(mm->owner, NULL);
465 return;
466 }
467
468 read_lock(&tasklist_lock);
469 /*
470 * Search in the children
471 */
472 list_for_each_entry(c, &p->children, sibling) {
473 if (c->mm == mm)
474 goto assign_new_owner;
475 }
476
477 /*
478 * Search in the siblings
479 */
480 list_for_each_entry(c, &p->real_parent->children, sibling) {
481 if (c->mm == mm)
482 goto assign_new_owner;
483 }
484
485 /*
486 * Search through everything else, we should not get here often.
487 */
488 for_each_process(g) {
489 if (g->flags & PF_KTHREAD)
490 continue;
491 for_each_thread(g, c) {
492 if (c->mm == mm)
493 goto assign_new_owner;
494 if (c->mm)
495 break;
496 }
497 }
498 read_unlock(&tasklist_lock);
499 /*
500 * We found no owner yet mm_users > 1: this implies that we are
501 * most likely racing with swapoff (try_to_unuse()) or /proc or
502 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
503 */
504 WRITE_ONCE(mm->owner, NULL);
505 return;
506
507assign_new_owner:
508 BUG_ON(c == p);
509 get_task_struct(c);
510 /*
511 * The task_lock protects c->mm from changing.
512 * We always want mm->owner->mm == mm
513 */
514 task_lock(c);
515 /*
516 * Delay read_unlock() till we have the task_lock()
517 * to ensure that c does not slip away underneath us
518 */
519 read_unlock(&tasklist_lock);
520 if (c->mm != mm) {
521 task_unlock(c);
522 put_task_struct(c);
523 goto retry;
524 }
525 WRITE_ONCE(mm->owner, c);
526 lru_gen_migrate_mm(mm);
527 task_unlock(c);
528 put_task_struct(c);
529}
530#endif /* CONFIG_MEMCG */
531
532/*
533 * Turn us into a lazy TLB process if we
534 * aren't already..
535 */
536static void exit_mm(void)
537{
538 struct mm_struct *mm = current->mm;
539
540 exit_mm_release(current, mm);
541 if (!mm)
542 return;
543 mmap_read_lock(mm);
544 mmgrab_lazy_tlb(mm);
545 BUG_ON(mm != current->active_mm);
546 /* more a memory barrier than a real lock */
547 task_lock(current);
548 /*
549 * When a thread stops operating on an address space, the loop
550 * in membarrier_private_expedited() may not observe that
551 * tsk->mm, and the loop in membarrier_global_expedited() may
552 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
553 * rq->membarrier_state, so those would not issue an IPI.
554 * Membarrier requires a memory barrier after accessing
555 * user-space memory, before clearing tsk->mm or the
556 * rq->membarrier_state.
557 */
558 smp_mb__after_spinlock();
559 local_irq_disable();
560 current->mm = NULL;
561 membarrier_update_current_mm(NULL);
562 enter_lazy_tlb(mm, current);
563 local_irq_enable();
564 task_unlock(current);
565 mmap_read_unlock(mm);
566 mm_update_next_owner(mm);
567 mmput(mm);
568 if (test_thread_flag(TIF_MEMDIE))
569 exit_oom_victim();
570}
571
572static struct task_struct *find_alive_thread(struct task_struct *p)
573{
574 struct task_struct *t;
575
576 for_each_thread(p, t) {
577 if (!(t->flags & PF_EXITING))
578 return t;
579 }
580 return NULL;
581}
582
583static struct task_struct *find_child_reaper(struct task_struct *father,
584 struct list_head *dead)
585 __releases(&tasklist_lock)
586 __acquires(&tasklist_lock)
587{
588 struct pid_namespace *pid_ns = task_active_pid_ns(father);
589 struct task_struct *reaper = pid_ns->child_reaper;
590 struct task_struct *p, *n;
591
592 if (likely(reaper != father))
593 return reaper;
594
595 reaper = find_alive_thread(father);
596 if (reaper) {
597 pid_ns->child_reaper = reaper;
598 return reaper;
599 }
600
601 write_unlock_irq(&tasklist_lock);
602
603 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
604 list_del_init(&p->ptrace_entry);
605 release_task(p);
606 }
607
608 zap_pid_ns_processes(pid_ns);
609 write_lock_irq(&tasklist_lock);
610
611 return father;
612}
613
614/*
615 * When we die, we re-parent all our children, and try to:
616 * 1. give them to another thread in our thread group, if such a member exists
617 * 2. give it to the first ancestor process which prctl'd itself as a
618 * child_subreaper for its children (like a service manager)
619 * 3. give it to the init process (PID 1) in our pid namespace
620 */
621static struct task_struct *find_new_reaper(struct task_struct *father,
622 struct task_struct *child_reaper)
623{
624 struct task_struct *thread, *reaper;
625
626 thread = find_alive_thread(father);
627 if (thread)
628 return thread;
629
630 if (father->signal->has_child_subreaper) {
631 unsigned int ns_level = task_pid(father)->level;
632 /*
633 * Find the first ->is_child_subreaper ancestor in our pid_ns.
634 * We can't check reaper != child_reaper to ensure we do not
635 * cross the namespaces, the exiting parent could be injected
636 * by setns() + fork().
637 * We check pid->level, this is slightly more efficient than
638 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
639 */
640 for (reaper = father->real_parent;
641 task_pid(reaper)->level == ns_level;
642 reaper = reaper->real_parent) {
643 if (reaper == &init_task)
644 break;
645 if (!reaper->signal->is_child_subreaper)
646 continue;
647 thread = find_alive_thread(reaper);
648 if (thread)
649 return thread;
650 }
651 }
652
653 return child_reaper;
654}
655
656/*
657* Any that need to be release_task'd are put on the @dead list.
658 */
659static void reparent_leader(struct task_struct *father, struct task_struct *p,
660 struct list_head *dead)
661{
662 if (unlikely(p->exit_state == EXIT_DEAD))
663 return;
664
665 /* We don't want people slaying init. */
666 p->exit_signal = SIGCHLD;
667
668 /* If it has exited notify the new parent about this child's death. */
669 if (!p->ptrace &&
670 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
671 if (do_notify_parent(p, p->exit_signal)) {
672 p->exit_state = EXIT_DEAD;
673 list_add(&p->ptrace_entry, dead);
674 }
675 }
676
677 kill_orphaned_pgrp(p, father);
678}
679
680/*
681 * This does two things:
682 *
683 * A. Make init inherit all the child processes
684 * B. Check to see if any process groups have become orphaned
685 * as a result of our exiting, and if they have any stopped
686 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
687 */
688static void forget_original_parent(struct task_struct *father,
689 struct list_head *dead)
690{
691 struct task_struct *p, *t, *reaper;
692
693 if (unlikely(!list_empty(&father->ptraced)))
694 exit_ptrace(father, dead);
695
696 /* Can drop and reacquire tasklist_lock */
697 reaper = find_child_reaper(father, dead);
698 if (list_empty(&father->children))
699 return;
700
701 reaper = find_new_reaper(father, reaper);
702 list_for_each_entry(p, &father->children, sibling) {
703 for_each_thread(p, t) {
704 RCU_INIT_POINTER(t->real_parent, reaper);
705 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
706 if (likely(!t->ptrace))
707 t->parent = t->real_parent;
708 if (t->pdeath_signal)
709 group_send_sig_info(t->pdeath_signal,
710 SEND_SIG_NOINFO, t,
711 PIDTYPE_TGID);
712 }
713 /*
714 * If this is a threaded reparent there is no need to
715 * notify anyone anything has happened.
716 */
717 if (!same_thread_group(reaper, father))
718 reparent_leader(father, p, dead);
719 }
720 list_splice_tail_init(&father->children, &reaper->children);
721}
722
723/*
724 * Send signals to all our closest relatives so that they know
725 * to properly mourn us..
726 */
727static void exit_notify(struct task_struct *tsk, int group_dead)
728{
729 bool autoreap;
730 struct task_struct *p, *n;
731 LIST_HEAD(dead);
732
733 write_lock_irq(&tasklist_lock);
734 forget_original_parent(tsk, &dead);
735
736 if (group_dead)
737 kill_orphaned_pgrp(tsk->group_leader, NULL);
738
739 tsk->exit_state = EXIT_ZOMBIE;
740 if (unlikely(tsk->ptrace)) {
741 int sig = thread_group_leader(tsk) &&
742 thread_group_empty(tsk) &&
743 !ptrace_reparented(tsk) ?
744 tsk->exit_signal : SIGCHLD;
745 autoreap = do_notify_parent(tsk, sig);
746 } else if (thread_group_leader(tsk)) {
747 autoreap = thread_group_empty(tsk) &&
748 do_notify_parent(tsk, tsk->exit_signal);
749 } else {
750 autoreap = true;
751 }
752
753 if (autoreap) {
754 tsk->exit_state = EXIT_DEAD;
755 list_add(&tsk->ptrace_entry, &dead);
756 }
757
758 /* mt-exec, de_thread() is waiting for group leader */
759 if (unlikely(tsk->signal->notify_count < 0))
760 wake_up_process(tsk->signal->group_exec_task);
761 write_unlock_irq(&tasklist_lock);
762
763 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
764 list_del_init(&p->ptrace_entry);
765 release_task(p);
766 }
767}
768
769#ifdef CONFIG_DEBUG_STACK_USAGE
770static void check_stack_usage(void)
771{
772 static DEFINE_SPINLOCK(low_water_lock);
773 static int lowest_to_date = THREAD_SIZE;
774 unsigned long free;
775
776 free = stack_not_used(current);
777
778 if (free >= lowest_to_date)
779 return;
780
781 spin_lock(&low_water_lock);
782 if (free < lowest_to_date) {
783 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
784 current->comm, task_pid_nr(current), free);
785 lowest_to_date = free;
786 }
787 spin_unlock(&low_water_lock);
788}
789#else
790static inline void check_stack_usage(void) {}
791#endif
792
793static void synchronize_group_exit(struct task_struct *tsk, long code)
794{
795 struct sighand_struct *sighand = tsk->sighand;
796 struct signal_struct *signal = tsk->signal;
797
798 spin_lock_irq(&sighand->siglock);
799 signal->quick_threads--;
800 if ((signal->quick_threads == 0) &&
801 !(signal->flags & SIGNAL_GROUP_EXIT)) {
802 signal->flags = SIGNAL_GROUP_EXIT;
803 signal->group_exit_code = code;
804 signal->group_stop_count = 0;
805 }
806 spin_unlock_irq(&sighand->siglock);
807}
808
809void __noreturn do_exit(long code)
810{
811 struct task_struct *tsk = current;
812 int group_dead;
813
814 WARN_ON(irqs_disabled());
815
816 synchronize_group_exit(tsk, code);
817
818 WARN_ON(tsk->plug);
819
820 kcov_task_exit(tsk);
821 kmsan_task_exit(tsk);
822
823 coredump_task_exit(tsk);
824 ptrace_event(PTRACE_EVENT_EXIT, code);
825 user_events_exit(tsk);
826
827 io_uring_files_cancel();
828 exit_signals(tsk); /* sets PF_EXITING */
829
830 acct_update_integrals(tsk);
831 group_dead = atomic_dec_and_test(&tsk->signal->live);
832 if (group_dead) {
833 /*
834 * If the last thread of global init has exited, panic
835 * immediately to get a useable coredump.
836 */
837 if (unlikely(is_global_init(tsk)))
838 panic("Attempted to kill init! exitcode=0x%08x\n",
839 tsk->signal->group_exit_code ?: (int)code);
840
841#ifdef CONFIG_POSIX_TIMERS
842 hrtimer_cancel(&tsk->signal->real_timer);
843 exit_itimers(tsk);
844#endif
845 if (tsk->mm)
846 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
847 }
848 acct_collect(code, group_dead);
849 if (group_dead)
850 tty_audit_exit();
851 audit_free(tsk);
852
853 tsk->exit_code = code;
854 taskstats_exit(tsk, group_dead);
855
856 exit_mm();
857
858 if (group_dead)
859 acct_process();
860 trace_sched_process_exit(tsk);
861
862 exit_sem(tsk);
863 exit_shm(tsk);
864 exit_files(tsk);
865 exit_fs(tsk);
866 if (group_dead)
867 disassociate_ctty(1);
868 exit_task_namespaces(tsk);
869 exit_task_work(tsk);
870 exit_thread(tsk);
871
872 /*
873 * Flush inherited counters to the parent - before the parent
874 * gets woken up by child-exit notifications.
875 *
876 * because of cgroup mode, must be called before cgroup_exit()
877 */
878 perf_event_exit_task(tsk);
879
880 sched_autogroup_exit_task(tsk);
881 cgroup_exit(tsk);
882
883 /*
884 * FIXME: do that only when needed, using sched_exit tracepoint
885 */
886 flush_ptrace_hw_breakpoint(tsk);
887
888 exit_tasks_rcu_start();
889 exit_notify(tsk, group_dead);
890 proc_exit_connector(tsk);
891 mpol_put_task_policy(tsk);
892#ifdef CONFIG_FUTEX
893 if (unlikely(current->pi_state_cache))
894 kfree(current->pi_state_cache);
895#endif
896 /*
897 * Make sure we are holding no locks:
898 */
899 debug_check_no_locks_held();
900
901 if (tsk->io_context)
902 exit_io_context(tsk);
903
904 if (tsk->splice_pipe)
905 free_pipe_info(tsk->splice_pipe);
906
907 if (tsk->task_frag.page)
908 put_page(tsk->task_frag.page);
909
910 exit_task_stack_account(tsk);
911
912 check_stack_usage();
913 preempt_disable();
914 if (tsk->nr_dirtied)
915 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
916 exit_rcu();
917 exit_tasks_rcu_finish();
918
919 lockdep_free_task(tsk);
920 do_task_dead();
921}
922
923void __noreturn make_task_dead(int signr)
924{
925 /*
926 * Take the task off the cpu after something catastrophic has
927 * happened.
928 *
929 * We can get here from a kernel oops, sometimes with preemption off.
930 * Start by checking for critical errors.
931 * Then fix up important state like USER_DS and preemption.
932 * Then do everything else.
933 */
934 struct task_struct *tsk = current;
935 unsigned int limit;
936
937 if (unlikely(in_interrupt()))
938 panic("Aiee, killing interrupt handler!");
939 if (unlikely(!tsk->pid))
940 panic("Attempted to kill the idle task!");
941
942 if (unlikely(irqs_disabled())) {
943 pr_info("note: %s[%d] exited with irqs disabled\n",
944 current->comm, task_pid_nr(current));
945 local_irq_enable();
946 }
947 if (unlikely(in_atomic())) {
948 pr_info("note: %s[%d] exited with preempt_count %d\n",
949 current->comm, task_pid_nr(current),
950 preempt_count());
951 preempt_count_set(PREEMPT_ENABLED);
952 }
953
954 /*
955 * Every time the system oopses, if the oops happens while a reference
956 * to an object was held, the reference leaks.
957 * If the oops doesn't also leak memory, repeated oopsing can cause
958 * reference counters to wrap around (if they're not using refcount_t).
959 * This means that repeated oopsing can make unexploitable-looking bugs
960 * exploitable through repeated oopsing.
961 * To make sure this can't happen, place an upper bound on how often the
962 * kernel may oops without panic().
963 */
964 limit = READ_ONCE(oops_limit);
965 if (atomic_inc_return(&oops_count) >= limit && limit)
966 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
967
968 /*
969 * We're taking recursive faults here in make_task_dead. Safest is to just
970 * leave this task alone and wait for reboot.
971 */
972 if (unlikely(tsk->flags & PF_EXITING)) {
973 pr_alert("Fixing recursive fault but reboot is needed!\n");
974 futex_exit_recursive(tsk);
975 tsk->exit_state = EXIT_DEAD;
976 refcount_inc(&tsk->rcu_users);
977 do_task_dead();
978 }
979
980 do_exit(signr);
981}
982
983SYSCALL_DEFINE1(exit, int, error_code)
984{
985 do_exit((error_code&0xff)<<8);
986}
987
988/*
989 * Take down every thread in the group. This is called by fatal signals
990 * as well as by sys_exit_group (below).
991 */
992void __noreturn
993do_group_exit(int exit_code)
994{
995 struct signal_struct *sig = current->signal;
996
997 if (sig->flags & SIGNAL_GROUP_EXIT)
998 exit_code = sig->group_exit_code;
999 else if (sig->group_exec_task)
1000 exit_code = 0;
1001 else {
1002 struct sighand_struct *const sighand = current->sighand;
1003
1004 spin_lock_irq(&sighand->siglock);
1005 if (sig->flags & SIGNAL_GROUP_EXIT)
1006 /* Another thread got here before we took the lock. */
1007 exit_code = sig->group_exit_code;
1008 else if (sig->group_exec_task)
1009 exit_code = 0;
1010 else {
1011 sig->group_exit_code = exit_code;
1012 sig->flags = SIGNAL_GROUP_EXIT;
1013 zap_other_threads(current);
1014 }
1015 spin_unlock_irq(&sighand->siglock);
1016 }
1017
1018 do_exit(exit_code);
1019 /* NOTREACHED */
1020}
1021
1022/*
1023 * this kills every thread in the thread group. Note that any externally
1024 * wait4()-ing process will get the correct exit code - even if this
1025 * thread is not the thread group leader.
1026 */
1027SYSCALL_DEFINE1(exit_group, int, error_code)
1028{
1029 do_group_exit((error_code & 0xff) << 8);
1030 /* NOTREACHED */
1031 return 0;
1032}
1033
1034static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1035{
1036 return wo->wo_type == PIDTYPE_MAX ||
1037 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1038}
1039
1040static int
1041eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1042{
1043 if (!eligible_pid(wo, p))
1044 return 0;
1045
1046 /*
1047 * Wait for all children (clone and not) if __WALL is set or
1048 * if it is traced by us.
1049 */
1050 if (ptrace || (wo->wo_flags & __WALL))
1051 return 1;
1052
1053 /*
1054 * Otherwise, wait for clone children *only* if __WCLONE is set;
1055 * otherwise, wait for non-clone children *only*.
1056 *
1057 * Note: a "clone" child here is one that reports to its parent
1058 * using a signal other than SIGCHLD, or a non-leader thread which
1059 * we can only see if it is traced by us.
1060 */
1061 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1062 return 0;
1063
1064 return 1;
1065}
1066
1067/*
1068 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1069 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1070 * the lock and this task is uninteresting. If we return nonzero, we have
1071 * released the lock and the system call should return.
1072 */
1073static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1074{
1075 int state, status;
1076 pid_t pid = task_pid_vnr(p);
1077 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1078 struct waitid_info *infop;
1079
1080 if (!likely(wo->wo_flags & WEXITED))
1081 return 0;
1082
1083 if (unlikely(wo->wo_flags & WNOWAIT)) {
1084 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1085 ? p->signal->group_exit_code : p->exit_code;
1086 get_task_struct(p);
1087 read_unlock(&tasklist_lock);
1088 sched_annotate_sleep();
1089 if (wo->wo_rusage)
1090 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1091 put_task_struct(p);
1092 goto out_info;
1093 }
1094 /*
1095 * Move the task's state to DEAD/TRACE, only one thread can do this.
1096 */
1097 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1098 EXIT_TRACE : EXIT_DEAD;
1099 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1100 return 0;
1101 /*
1102 * We own this thread, nobody else can reap it.
1103 */
1104 read_unlock(&tasklist_lock);
1105 sched_annotate_sleep();
1106
1107 /*
1108 * Check thread_group_leader() to exclude the traced sub-threads.
1109 */
1110 if (state == EXIT_DEAD && thread_group_leader(p)) {
1111 struct signal_struct *sig = p->signal;
1112 struct signal_struct *psig = current->signal;
1113 unsigned long maxrss;
1114 u64 tgutime, tgstime;
1115
1116 /*
1117 * The resource counters for the group leader are in its
1118 * own task_struct. Those for dead threads in the group
1119 * are in its signal_struct, as are those for the child
1120 * processes it has previously reaped. All these
1121 * accumulate in the parent's signal_struct c* fields.
1122 *
1123 * We don't bother to take a lock here to protect these
1124 * p->signal fields because the whole thread group is dead
1125 * and nobody can change them.
1126 *
1127 * psig->stats_lock also protects us from our sub-threads
1128 * which can reap other children at the same time. Until
1129 * we change k_getrusage()-like users to rely on this lock
1130 * we have to take ->siglock as well.
1131 *
1132 * We use thread_group_cputime_adjusted() to get times for
1133 * the thread group, which consolidates times for all threads
1134 * in the group including the group leader.
1135 */
1136 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1137 spin_lock_irq(&current->sighand->siglock);
1138 write_seqlock(&psig->stats_lock);
1139 psig->cutime += tgutime + sig->cutime;
1140 psig->cstime += tgstime + sig->cstime;
1141 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1142 psig->cmin_flt +=
1143 p->min_flt + sig->min_flt + sig->cmin_flt;
1144 psig->cmaj_flt +=
1145 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1146 psig->cnvcsw +=
1147 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1148 psig->cnivcsw +=
1149 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1150 psig->cinblock +=
1151 task_io_get_inblock(p) +
1152 sig->inblock + sig->cinblock;
1153 psig->coublock +=
1154 task_io_get_oublock(p) +
1155 sig->oublock + sig->coublock;
1156 maxrss = max(sig->maxrss, sig->cmaxrss);
1157 if (psig->cmaxrss < maxrss)
1158 psig->cmaxrss = maxrss;
1159 task_io_accounting_add(&psig->ioac, &p->ioac);
1160 task_io_accounting_add(&psig->ioac, &sig->ioac);
1161 write_sequnlock(&psig->stats_lock);
1162 spin_unlock_irq(&current->sighand->siglock);
1163 }
1164
1165 if (wo->wo_rusage)
1166 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1167 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1168 ? p->signal->group_exit_code : p->exit_code;
1169 wo->wo_stat = status;
1170
1171 if (state == EXIT_TRACE) {
1172 write_lock_irq(&tasklist_lock);
1173 /* We dropped tasklist, ptracer could die and untrace */
1174 ptrace_unlink(p);
1175
1176 /* If parent wants a zombie, don't release it now */
1177 state = EXIT_ZOMBIE;
1178 if (do_notify_parent(p, p->exit_signal))
1179 state = EXIT_DEAD;
1180 p->exit_state = state;
1181 write_unlock_irq(&tasklist_lock);
1182 }
1183 if (state == EXIT_DEAD)
1184 release_task(p);
1185
1186out_info:
1187 infop = wo->wo_info;
1188 if (infop) {
1189 if ((status & 0x7f) == 0) {
1190 infop->cause = CLD_EXITED;
1191 infop->status = status >> 8;
1192 } else {
1193 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1194 infop->status = status & 0x7f;
1195 }
1196 infop->pid = pid;
1197 infop->uid = uid;
1198 }
1199
1200 return pid;
1201}
1202
1203static int *task_stopped_code(struct task_struct *p, bool ptrace)
1204{
1205 if (ptrace) {
1206 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1207 return &p->exit_code;
1208 } else {
1209 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1210 return &p->signal->group_exit_code;
1211 }
1212 return NULL;
1213}
1214
1215/**
1216 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1217 * @wo: wait options
1218 * @ptrace: is the wait for ptrace
1219 * @p: task to wait for
1220 *
1221 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1222 *
1223 * CONTEXT:
1224 * read_lock(&tasklist_lock), which is released if return value is
1225 * non-zero. Also, grabs and releases @p->sighand->siglock.
1226 *
1227 * RETURNS:
1228 * 0 if wait condition didn't exist and search for other wait conditions
1229 * should continue. Non-zero return, -errno on failure and @p's pid on
1230 * success, implies that tasklist_lock is released and wait condition
1231 * search should terminate.
1232 */
1233static int wait_task_stopped(struct wait_opts *wo,
1234 int ptrace, struct task_struct *p)
1235{
1236 struct waitid_info *infop;
1237 int exit_code, *p_code, why;
1238 uid_t uid = 0; /* unneeded, required by compiler */
1239 pid_t pid;
1240
1241 /*
1242 * Traditionally we see ptrace'd stopped tasks regardless of options.
1243 */
1244 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1245 return 0;
1246
1247 if (!task_stopped_code(p, ptrace))
1248 return 0;
1249
1250 exit_code = 0;
1251 spin_lock_irq(&p->sighand->siglock);
1252
1253 p_code = task_stopped_code(p, ptrace);
1254 if (unlikely(!p_code))
1255 goto unlock_sig;
1256
1257 exit_code = *p_code;
1258 if (!exit_code)
1259 goto unlock_sig;
1260
1261 if (!unlikely(wo->wo_flags & WNOWAIT))
1262 *p_code = 0;
1263
1264 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1265unlock_sig:
1266 spin_unlock_irq(&p->sighand->siglock);
1267 if (!exit_code)
1268 return 0;
1269
1270 /*
1271 * Now we are pretty sure this task is interesting.
1272 * Make sure it doesn't get reaped out from under us while we
1273 * give up the lock and then examine it below. We don't want to
1274 * keep holding onto the tasklist_lock while we call getrusage and
1275 * possibly take page faults for user memory.
1276 */
1277 get_task_struct(p);
1278 pid = task_pid_vnr(p);
1279 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1280 read_unlock(&tasklist_lock);
1281 sched_annotate_sleep();
1282 if (wo->wo_rusage)
1283 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1284 put_task_struct(p);
1285
1286 if (likely(!(wo->wo_flags & WNOWAIT)))
1287 wo->wo_stat = (exit_code << 8) | 0x7f;
1288
1289 infop = wo->wo_info;
1290 if (infop) {
1291 infop->cause = why;
1292 infop->status = exit_code;
1293 infop->pid = pid;
1294 infop->uid = uid;
1295 }
1296 return pid;
1297}
1298
1299/*
1300 * Handle do_wait work for one task in a live, non-stopped state.
1301 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1302 * the lock and this task is uninteresting. If we return nonzero, we have
1303 * released the lock and the system call should return.
1304 */
1305static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1306{
1307 struct waitid_info *infop;
1308 pid_t pid;
1309 uid_t uid;
1310
1311 if (!unlikely(wo->wo_flags & WCONTINUED))
1312 return 0;
1313
1314 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1315 return 0;
1316
1317 spin_lock_irq(&p->sighand->siglock);
1318 /* Re-check with the lock held. */
1319 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1320 spin_unlock_irq(&p->sighand->siglock);
1321 return 0;
1322 }
1323 if (!unlikely(wo->wo_flags & WNOWAIT))
1324 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1325 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1326 spin_unlock_irq(&p->sighand->siglock);
1327
1328 pid = task_pid_vnr(p);
1329 get_task_struct(p);
1330 read_unlock(&tasklist_lock);
1331 sched_annotate_sleep();
1332 if (wo->wo_rusage)
1333 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1334 put_task_struct(p);
1335
1336 infop = wo->wo_info;
1337 if (!infop) {
1338 wo->wo_stat = 0xffff;
1339 } else {
1340 infop->cause = CLD_CONTINUED;
1341 infop->pid = pid;
1342 infop->uid = uid;
1343 infop->status = SIGCONT;
1344 }
1345 return pid;
1346}
1347
1348/*
1349 * Consider @p for a wait by @parent.
1350 *
1351 * -ECHILD should be in ->notask_error before the first call.
1352 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1353 * Returns zero if the search for a child should continue;
1354 * then ->notask_error is 0 if @p is an eligible child,
1355 * or still -ECHILD.
1356 */
1357static int wait_consider_task(struct wait_opts *wo, int ptrace,
1358 struct task_struct *p)
1359{
1360 /*
1361 * We can race with wait_task_zombie() from another thread.
1362 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1363 * can't confuse the checks below.
1364 */
1365 int exit_state = READ_ONCE(p->exit_state);
1366 int ret;
1367
1368 if (unlikely(exit_state == EXIT_DEAD))
1369 return 0;
1370
1371 ret = eligible_child(wo, ptrace, p);
1372 if (!ret)
1373 return ret;
1374
1375 if (unlikely(exit_state == EXIT_TRACE)) {
1376 /*
1377 * ptrace == 0 means we are the natural parent. In this case
1378 * we should clear notask_error, debugger will notify us.
1379 */
1380 if (likely(!ptrace))
1381 wo->notask_error = 0;
1382 return 0;
1383 }
1384
1385 if (likely(!ptrace) && unlikely(p->ptrace)) {
1386 /*
1387 * If it is traced by its real parent's group, just pretend
1388 * the caller is ptrace_do_wait() and reap this child if it
1389 * is zombie.
1390 *
1391 * This also hides group stop state from real parent; otherwise
1392 * a single stop can be reported twice as group and ptrace stop.
1393 * If a ptracer wants to distinguish these two events for its
1394 * own children it should create a separate process which takes
1395 * the role of real parent.
1396 */
1397 if (!ptrace_reparented(p))
1398 ptrace = 1;
1399 }
1400
1401 /* slay zombie? */
1402 if (exit_state == EXIT_ZOMBIE) {
1403 /* we don't reap group leaders with subthreads */
1404 if (!delay_group_leader(p)) {
1405 /*
1406 * A zombie ptracee is only visible to its ptracer.
1407 * Notification and reaping will be cascaded to the
1408 * real parent when the ptracer detaches.
1409 */
1410 if (unlikely(ptrace) || likely(!p->ptrace))
1411 return wait_task_zombie(wo, p);
1412 }
1413
1414 /*
1415 * Allow access to stopped/continued state via zombie by
1416 * falling through. Clearing of notask_error is complex.
1417 *
1418 * When !@ptrace:
1419 *
1420 * If WEXITED is set, notask_error should naturally be
1421 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1422 * so, if there are live subthreads, there are events to
1423 * wait for. If all subthreads are dead, it's still safe
1424 * to clear - this function will be called again in finite
1425 * amount time once all the subthreads are released and
1426 * will then return without clearing.
1427 *
1428 * When @ptrace:
1429 *
1430 * Stopped state is per-task and thus can't change once the
1431 * target task dies. Only continued and exited can happen.
1432 * Clear notask_error if WCONTINUED | WEXITED.
1433 */
1434 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1435 wo->notask_error = 0;
1436 } else {
1437 /*
1438 * @p is alive and it's gonna stop, continue or exit, so
1439 * there always is something to wait for.
1440 */
1441 wo->notask_error = 0;
1442 }
1443
1444 /*
1445 * Wait for stopped. Depending on @ptrace, different stopped state
1446 * is used and the two don't interact with each other.
1447 */
1448 ret = wait_task_stopped(wo, ptrace, p);
1449 if (ret)
1450 return ret;
1451
1452 /*
1453 * Wait for continued. There's only one continued state and the
1454 * ptracer can consume it which can confuse the real parent. Don't
1455 * use WCONTINUED from ptracer. You don't need or want it.
1456 */
1457 return wait_task_continued(wo, p);
1458}
1459
1460/*
1461 * Do the work of do_wait() for one thread in the group, @tsk.
1462 *
1463 * -ECHILD should be in ->notask_error before the first call.
1464 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1465 * Returns zero if the search for a child should continue; then
1466 * ->notask_error is 0 if there were any eligible children,
1467 * or still -ECHILD.
1468 */
1469static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1470{
1471 struct task_struct *p;
1472
1473 list_for_each_entry(p, &tsk->children, sibling) {
1474 int ret = wait_consider_task(wo, 0, p);
1475
1476 if (ret)
1477 return ret;
1478 }
1479
1480 return 0;
1481}
1482
1483static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1484{
1485 struct task_struct *p;
1486
1487 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1488 int ret = wait_consider_task(wo, 1, p);
1489
1490 if (ret)
1491 return ret;
1492 }
1493
1494 return 0;
1495}
1496
1497bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1498{
1499 if (!eligible_pid(wo, p))
1500 return false;
1501
1502 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1503 return false;
1504
1505 return true;
1506}
1507
1508static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1509 int sync, void *key)
1510{
1511 struct wait_opts *wo = container_of(wait, struct wait_opts,
1512 child_wait);
1513 struct task_struct *p = key;
1514
1515 if (pid_child_should_wake(wo, p))
1516 return default_wake_function(wait, mode, sync, key);
1517
1518 return 0;
1519}
1520
1521void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1522{
1523 __wake_up_sync_key(&parent->signal->wait_chldexit,
1524 TASK_INTERRUPTIBLE, p);
1525}
1526
1527static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1528 struct task_struct *target)
1529{
1530 struct task_struct *parent =
1531 !ptrace ? target->real_parent : target->parent;
1532
1533 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1534 same_thread_group(current, parent));
1535}
1536
1537/*
1538 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1539 * and tracee lists to find the target task.
1540 */
1541static int do_wait_pid(struct wait_opts *wo)
1542{
1543 bool ptrace;
1544 struct task_struct *target;
1545 int retval;
1546
1547 ptrace = false;
1548 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1549 if (target && is_effectively_child(wo, ptrace, target)) {
1550 retval = wait_consider_task(wo, ptrace, target);
1551 if (retval)
1552 return retval;
1553 }
1554
1555 ptrace = true;
1556 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1557 if (target && target->ptrace &&
1558 is_effectively_child(wo, ptrace, target)) {
1559 retval = wait_consider_task(wo, ptrace, target);
1560 if (retval)
1561 return retval;
1562 }
1563
1564 return 0;
1565}
1566
1567long __do_wait(struct wait_opts *wo)
1568{
1569 long retval;
1570
1571 /*
1572 * If there is nothing that can match our criteria, just get out.
1573 * We will clear ->notask_error to zero if we see any child that
1574 * might later match our criteria, even if we are not able to reap
1575 * it yet.
1576 */
1577 wo->notask_error = -ECHILD;
1578 if ((wo->wo_type < PIDTYPE_MAX) &&
1579 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1580 goto notask;
1581
1582 read_lock(&tasklist_lock);
1583
1584 if (wo->wo_type == PIDTYPE_PID) {
1585 retval = do_wait_pid(wo);
1586 if (retval)
1587 return retval;
1588 } else {
1589 struct task_struct *tsk = current;
1590
1591 do {
1592 retval = do_wait_thread(wo, tsk);
1593 if (retval)
1594 return retval;
1595
1596 retval = ptrace_do_wait(wo, tsk);
1597 if (retval)
1598 return retval;
1599
1600 if (wo->wo_flags & __WNOTHREAD)
1601 break;
1602 } while_each_thread(current, tsk);
1603 }
1604 read_unlock(&tasklist_lock);
1605
1606notask:
1607 retval = wo->notask_error;
1608 if (!retval && !(wo->wo_flags & WNOHANG))
1609 return -ERESTARTSYS;
1610
1611 return retval;
1612}
1613
1614static long do_wait(struct wait_opts *wo)
1615{
1616 int retval;
1617
1618 trace_sched_process_wait(wo->wo_pid);
1619
1620 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1621 wo->child_wait.private = current;
1622 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1623
1624 do {
1625 set_current_state(TASK_INTERRUPTIBLE);
1626 retval = __do_wait(wo);
1627 if (retval != -ERESTARTSYS)
1628 break;
1629 if (signal_pending(current))
1630 break;
1631 schedule();
1632 } while (1);
1633
1634 __set_current_state(TASK_RUNNING);
1635 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1636 return retval;
1637}
1638
1639int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1640 struct waitid_info *infop, int options,
1641 struct rusage *ru)
1642{
1643 unsigned int f_flags = 0;
1644 struct pid *pid = NULL;
1645 enum pid_type type;
1646
1647 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1648 __WNOTHREAD|__WCLONE|__WALL))
1649 return -EINVAL;
1650 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1651 return -EINVAL;
1652
1653 switch (which) {
1654 case P_ALL:
1655 type = PIDTYPE_MAX;
1656 break;
1657 case P_PID:
1658 type = PIDTYPE_PID;
1659 if (upid <= 0)
1660 return -EINVAL;
1661
1662 pid = find_get_pid(upid);
1663 break;
1664 case P_PGID:
1665 type = PIDTYPE_PGID;
1666 if (upid < 0)
1667 return -EINVAL;
1668
1669 if (upid)
1670 pid = find_get_pid(upid);
1671 else
1672 pid = get_task_pid(current, PIDTYPE_PGID);
1673 break;
1674 case P_PIDFD:
1675 type = PIDTYPE_PID;
1676 if (upid < 0)
1677 return -EINVAL;
1678
1679 pid = pidfd_get_pid(upid, &f_flags);
1680 if (IS_ERR(pid))
1681 return PTR_ERR(pid);
1682
1683 break;
1684 default:
1685 return -EINVAL;
1686 }
1687
1688 wo->wo_type = type;
1689 wo->wo_pid = pid;
1690 wo->wo_flags = options;
1691 wo->wo_info = infop;
1692 wo->wo_rusage = ru;
1693 if (f_flags & O_NONBLOCK)
1694 wo->wo_flags |= WNOHANG;
1695
1696 return 0;
1697}
1698
1699static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1700 int options, struct rusage *ru)
1701{
1702 struct wait_opts wo;
1703 long ret;
1704
1705 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1706 if (ret)
1707 return ret;
1708
1709 ret = do_wait(&wo);
1710 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1711 ret = -EAGAIN;
1712
1713 put_pid(wo.wo_pid);
1714 return ret;
1715}
1716
1717SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1718 infop, int, options, struct rusage __user *, ru)
1719{
1720 struct rusage r;
1721 struct waitid_info info = {.status = 0};
1722 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1723 int signo = 0;
1724
1725 if (err > 0) {
1726 signo = SIGCHLD;
1727 err = 0;
1728 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1729 return -EFAULT;
1730 }
1731 if (!infop)
1732 return err;
1733
1734 if (!user_write_access_begin(infop, sizeof(*infop)))
1735 return -EFAULT;
1736
1737 unsafe_put_user(signo, &infop->si_signo, Efault);
1738 unsafe_put_user(0, &infop->si_errno, Efault);
1739 unsafe_put_user(info.cause, &infop->si_code, Efault);
1740 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1741 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1742 unsafe_put_user(info.status, &infop->si_status, Efault);
1743 user_write_access_end();
1744 return err;
1745Efault:
1746 user_write_access_end();
1747 return -EFAULT;
1748}
1749
1750long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1751 struct rusage *ru)
1752{
1753 struct wait_opts wo;
1754 struct pid *pid = NULL;
1755 enum pid_type type;
1756 long ret;
1757
1758 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1759 __WNOTHREAD|__WCLONE|__WALL))
1760 return -EINVAL;
1761
1762 /* -INT_MIN is not defined */
1763 if (upid == INT_MIN)
1764 return -ESRCH;
1765
1766 if (upid == -1)
1767 type = PIDTYPE_MAX;
1768 else if (upid < 0) {
1769 type = PIDTYPE_PGID;
1770 pid = find_get_pid(-upid);
1771 } else if (upid == 0) {
1772 type = PIDTYPE_PGID;
1773 pid = get_task_pid(current, PIDTYPE_PGID);
1774 } else /* upid > 0 */ {
1775 type = PIDTYPE_PID;
1776 pid = find_get_pid(upid);
1777 }
1778
1779 wo.wo_type = type;
1780 wo.wo_pid = pid;
1781 wo.wo_flags = options | WEXITED;
1782 wo.wo_info = NULL;
1783 wo.wo_stat = 0;
1784 wo.wo_rusage = ru;
1785 ret = do_wait(&wo);
1786 put_pid(pid);
1787 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1788 ret = -EFAULT;
1789
1790 return ret;
1791}
1792
1793int kernel_wait(pid_t pid, int *stat)
1794{
1795 struct wait_opts wo = {
1796 .wo_type = PIDTYPE_PID,
1797 .wo_pid = find_get_pid(pid),
1798 .wo_flags = WEXITED,
1799 };
1800 int ret;
1801
1802 ret = do_wait(&wo);
1803 if (ret > 0 && wo.wo_stat)
1804 *stat = wo.wo_stat;
1805 put_pid(wo.wo_pid);
1806 return ret;
1807}
1808
1809SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1810 int, options, struct rusage __user *, ru)
1811{
1812 struct rusage r;
1813 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1814
1815 if (err > 0) {
1816 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1817 return -EFAULT;
1818 }
1819 return err;
1820}
1821
1822#ifdef __ARCH_WANT_SYS_WAITPID
1823
1824/*
1825 * sys_waitpid() remains for compatibility. waitpid() should be
1826 * implemented by calling sys_wait4() from libc.a.
1827 */
1828SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1829{
1830 return kernel_wait4(pid, stat_addr, options, NULL);
1831}
1832
1833#endif
1834
1835#ifdef CONFIG_COMPAT
1836COMPAT_SYSCALL_DEFINE4(wait4,
1837 compat_pid_t, pid,
1838 compat_uint_t __user *, stat_addr,
1839 int, options,
1840 struct compat_rusage __user *, ru)
1841{
1842 struct rusage r;
1843 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1844 if (err > 0) {
1845 if (ru && put_compat_rusage(&r, ru))
1846 return -EFAULT;
1847 }
1848 return err;
1849}
1850
1851COMPAT_SYSCALL_DEFINE5(waitid,
1852 int, which, compat_pid_t, pid,
1853 struct compat_siginfo __user *, infop, int, options,
1854 struct compat_rusage __user *, uru)
1855{
1856 struct rusage ru;
1857 struct waitid_info info = {.status = 0};
1858 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1859 int signo = 0;
1860 if (err > 0) {
1861 signo = SIGCHLD;
1862 err = 0;
1863 if (uru) {
1864 /* kernel_waitid() overwrites everything in ru */
1865 if (COMPAT_USE_64BIT_TIME)
1866 err = copy_to_user(uru, &ru, sizeof(ru));
1867 else
1868 err = put_compat_rusage(&ru, uru);
1869 if (err)
1870 return -EFAULT;
1871 }
1872 }
1873
1874 if (!infop)
1875 return err;
1876
1877 if (!user_write_access_begin(infop, sizeof(*infop)))
1878 return -EFAULT;
1879
1880 unsafe_put_user(signo, &infop->si_signo, Efault);
1881 unsafe_put_user(0, &infop->si_errno, Efault);
1882 unsafe_put_user(info.cause, &infop->si_code, Efault);
1883 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1884 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1885 unsafe_put_user(info.status, &infop->si_status, Efault);
1886 user_write_access_end();
1887 return err;
1888Efault:
1889 user_write_access_end();
1890 return -EFAULT;
1891}
1892#endif
1893
1894/**
1895 * thread_group_exited - check that a thread group has exited
1896 * @pid: tgid of thread group to be checked.
1897 *
1898 * Test if the thread group represented by tgid has exited (all
1899 * threads are zombies, dead or completely gone).
1900 *
1901 * Return: true if the thread group has exited. false otherwise.
1902 */
1903bool thread_group_exited(struct pid *pid)
1904{
1905 struct task_struct *task;
1906 bool exited;
1907
1908 rcu_read_lock();
1909 task = pid_task(pid, PIDTYPE_PID);
1910 exited = !task ||
1911 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1912 rcu_read_unlock();
1913
1914 return exited;
1915}
1916EXPORT_SYMBOL(thread_group_exited);
1917
1918/*
1919 * This needs to be __function_aligned as GCC implicitly makes any
1920 * implementation of abort() cold and drops alignment specified by
1921 * -falign-functions=N.
1922 *
1923 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1924 */
1925__weak __function_aligned void abort(void)
1926{
1927 BUG();
1928
1929 /* if that doesn't kill us, halt */
1930 panic("Oops failed to kill thread");
1931}
1932EXPORT_SYMBOL(abort);