projects/linux·blob master · refs

2016 lines · 50.2 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/wait_bit.h>
21#include <linux/personality.h>
22#include <linux/tty.h>
23#include <linux/iocontext.h>
24#include <linux/key.h>
25#include <linux/cpu.h>
26#include <linux/acct.h>
27#include <linux/tsacct_kern.h>
28#include <linux/file.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/task_work.h>
53#include <linux/fs_struct.h>
54#include <linux/init_task.h>
55#include <linux/perf_event.h>
56#include <trace/events/sched.h>
57#include <linux/hw_breakpoint.h>
58#include <linux/oom.h>
59#include <linux/writeback.h>
60#include <linux/shm.h>
61#include <linux/kcov.h>
62#include <linux/kmsan.h>
63#include <linux/random.h>
64#include <linux/rcuwait.h>
65#include <linux/compat.h>
66#include <linux/io_uring.h>
67#include <linux/kprobes.h>
68#include <linux/rethook.h>
69#include <linux/sysfs.h>
70#include <linux/user_events.h>
71#include <linux/unwind_deferred.h>
72#include <linux/uaccess.h>
73#include <linux/pidfs.h>
74
75#include <uapi/linux/wait.h>
76
77#include <asm/unistd.h>
78#include <asm/mmu_context.h>
79
80#include "exit.h"
81
82/*
83 * The default value should be high enough to not crash a system that randomly
84 * crashes its kernel from time to time, but low enough to at least not permit
85 * overflowing 32-bit refcounts or the ldsem writer count.
86 */
87static unsigned int oops_limit = 10000;
88
89#ifdef CONFIG_SYSCTL
90static const struct ctl_table kern_exit_table[] = {
91 {
92 .procname = "oops_limit",
93 .data = &oops_limit,
94 .maxlen = sizeof(oops_limit),
95 .mode = 0644,
96 .proc_handler = proc_douintvec,
97 },
98};
99
100static __init int kernel_exit_sysctls_init(void)
101{
102 register_sysctl_init("kernel", kern_exit_table);
103 return 0;
104}
105late_initcall(kernel_exit_sysctls_init);
106#endif
107
108static atomic_t oops_count = ATOMIC_INIT(0);
109
110#ifdef CONFIG_SYSFS
111static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
112 char *page)
113{
114 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
115}
116
117static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
118
119static __init int kernel_exit_sysfs_init(void)
120{
121 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
122 return 0;
123}
124late_initcall(kernel_exit_sysfs_init);
125#endif
126
127/*
128 * For things release_task() would like to do *after* tasklist_lock is released.
129 */
130struct release_task_post {
131 struct pid *pids[PIDTYPE_MAX];
132};
133
134static void __unhash_process(struct release_task_post *post, struct task_struct *p,
135 bool group_dead)
136{
137 struct pid *pid = task_pid(p);
138
139 nr_threads--;
140
141 detach_pid(post->pids, p, PIDTYPE_PID);
142 wake_up_all(&pid->wait_pidfd);
143
144 if (group_dead) {
145 detach_pid(post->pids, p, PIDTYPE_TGID);
146 detach_pid(post->pids, p, PIDTYPE_PGID);
147 detach_pid(post->pids, p, PIDTYPE_SID);
148
149 list_del_rcu(&p->tasks);
150 list_del_init(&p->sibling);
151 __this_cpu_dec(process_counts);
152 }
153 list_del_rcu(&p->thread_node);
154}
155
156/*
157 * This function expects the tasklist_lock write-locked.
158 */
159static void __exit_signal(struct release_task_post *post, struct task_struct *tsk)
160{
161 struct signal_struct *sig = tsk->signal;
162 bool group_dead = thread_group_leader(tsk);
163 struct sighand_struct *sighand;
164 struct tty_struct *tty;
165 u64 utime, stime;
166
167 sighand = rcu_dereference_check(tsk->sighand,
168 lockdep_tasklist_lock_is_held());
169 spin_lock(&sighand->siglock);
170
171#ifdef CONFIG_POSIX_TIMERS
172 posix_cpu_timers_exit(tsk);
173 if (group_dead)
174 posix_cpu_timers_exit_group(tsk);
175#endif
176
177 if (group_dead) {
178 tty = sig->tty;
179 sig->tty = NULL;
180 } else {
181 /*
182 * If there is any task waiting for the group exit
183 * then notify it:
184 */
185 if (sig->notify_count > 0 && !--sig->notify_count)
186 wake_up_process(sig->group_exec_task);
187
188 if (tsk == sig->curr_target)
189 sig->curr_target = next_thread(tsk);
190 }
191
192 /*
193 * Accumulate here the counters for all threads as they die. We could
194 * skip the group leader because it is the last user of signal_struct,
195 * but we want to avoid the race with thread_group_cputime() which can
196 * see the empty ->thread_head list.
197 */
198 task_cputime(tsk, &utime, &stime);
199 write_seqlock(&sig->stats_lock);
200 sig->utime += utime;
201 sig->stime += stime;
202 sig->gtime += task_gtime(tsk);
203 sig->min_flt += tsk->min_flt;
204 sig->maj_flt += tsk->maj_flt;
205 sig->nvcsw += tsk->nvcsw;
206 sig->nivcsw += tsk->nivcsw;
207 sig->inblock += task_io_get_inblock(tsk);
208 sig->oublock += task_io_get_oublock(tsk);
209 task_io_accounting_add(&sig->ioac, &tsk->ioac);
210 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
211 sig->nr_threads--;
212 __unhash_process(post, tsk, group_dead);
213 write_sequnlock(&sig->stats_lock);
214
215 /*
216 * Ensure that all preceeding state is visible. Pairs with
217 * the smp_acquire__after_ctrl_dep() in the sighand == NULL
218 * path of lock_task_sighand().
219 */
220 smp_store_release(&tsk->sighand, NULL);
221 spin_unlock(&sighand->siglock);
222
223 __cleanup_sighand(sighand);
224 if (group_dead)
225 tty_kref_put(tty);
226}
227
228static void delayed_put_task_struct(struct rcu_head *rhp)
229{
230 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
231
232 kprobe_flush_task(tsk);
233 rethook_flush_task(tsk);
234 perf_event_delayed_put(tsk);
235 trace_sched_process_free(tsk);
236 put_task_struct(tsk);
237}
238
239void put_task_struct_rcu_user(struct task_struct *task)
240{
241 if (refcount_dec_and_test(&task->rcu_users))
242 call_rcu(&task->rcu, delayed_put_task_struct);
243}
244
245void __weak release_thread(struct task_struct *dead_task)
246{
247}
248
249void release_task(struct task_struct *p)
250{
251 struct release_task_post post;
252 struct task_struct *leader;
253 struct pid *thread_pid;
254 int zap_leader;
255repeat:
256 memset(&post, 0, sizeof(post));
257
258 /* don't need to get the RCU readlock here - the process is dead and
259 * can't be modifying its own credentials. */
260 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
261
262 pidfs_exit(p);
263 cgroup_task_release(p);
264
265 /*
266 * Pin @thread_pid before __unhash_process() clears it. The last
267 * PIDTYPE detach can otherwise free it before proc_flush_pid().
268 */
269 thread_pid = get_pid(task_pid(p));
270
271 write_lock_irq(&tasklist_lock);
272 ptrace_release_task(p);
273 __exit_signal(&post, p);
274
275 /*
276 * If we are the last non-leader member of the thread
277 * group, and the leader is zombie, then notify the
278 * group leader's parent process. (if it wants notification.)
279 */
280 zap_leader = 0;
281 leader = p->group_leader;
282 if (leader != p && thread_group_empty(leader)
283 && leader->exit_state == EXIT_ZOMBIE) {
284 /* for pidfs_exit() and do_notify_parent() */
285 if (leader->signal->flags & SIGNAL_GROUP_EXIT)
286 leader->exit_code = leader->signal->group_exit_code;
287 /*
288 * If we were the last child thread and the leader has
289 * exited already, and the leader's parent ignores SIGCHLD,
290 * then we are the one who should release the leader.
291 */
292 zap_leader = do_notify_parent(leader, leader->exit_signal);
293 if (zap_leader)
294 leader->exit_state = EXIT_DEAD;
295 }
296
297 write_unlock_irq(&tasklist_lock);
298 proc_flush_pid(thread_pid);
299 put_pid(thread_pid);
300 exit_cred_namespaces(p);
301 add_device_randomness(&p->se.sum_exec_runtime,
302 sizeof(p->se.sum_exec_runtime));
303 free_pids(post.pids);
304 release_thread(p);
305 /*
306 * This task was already removed from the process/thread/pid lists and
307 * lock_task_sighand(p) can't succeed. If it's the group leader then
308 * flush tsk->signal->shared_pending. tsk->pending has been flushed
309 * already in exit_signals(). Nothing else can touch
310 * signal->shared_pending anymore, so flush_sigqueue() can be invoked
311 * lockless.
312 */
313 if (thread_group_leader(p))
314 flush_sigqueue(&p->signal->shared_pending);
315
316 put_task_struct_rcu_user(p);
317
318 p = leader;
319 if (unlikely(zap_leader))
320 goto repeat;
321}
322
323int rcuwait_wake_up(struct rcuwait *w)
324{
325 int ret = 0;
326 struct task_struct *task;
327
328 rcu_read_lock();
329
330 /*
331 * Order condition vs @task, such that everything prior to the load
332 * of @task is visible. This is the condition as to why the user called
333 * rcuwait_wake() in the first place. Pairs with set_current_state()
334 * barrier (A) in rcuwait_wait_event().
335 *
336 * WAIT WAKE
337 * [S] tsk = current [S] cond = true
338 * MB (A) MB (B)
339 * [L] cond [L] tsk
340 */
341 smp_mb(); /* (B) */
342
343 task = rcu_dereference(w->task);
344 if (task)
345 ret = wake_up_process(task);
346 rcu_read_unlock();
347
348 return ret;
349}
350EXPORT_SYMBOL_GPL(rcuwait_wake_up);
351
352/*
353 * Determine if a process group is "orphaned", according to the POSIX
354 * definition in 2.2.2.52. Orphaned process groups are not to be affected
355 * by terminal-generated stop signals. Newly orphaned process groups are
356 * to receive a SIGHUP and a SIGCONT.
357 *
358 * "I ask you, have you ever known what it is to be an orphan?"
359 */
360static int will_become_orphaned_pgrp(struct pid *pgrp,
361 struct task_struct *ignored_task)
362{
363 struct task_struct *p;
364
365 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
366 if ((p == ignored_task) ||
367 (p->exit_state && thread_group_empty(p)) ||
368 is_global_init(p->real_parent))
369 continue;
370
371 if (task_pgrp(p->real_parent) != pgrp &&
372 task_session(p->real_parent) == task_session(p))
373 return 0;
374 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
375
376 return 1;
377}
378
379int is_current_pgrp_orphaned(void)
380{
381 int retval;
382
383 read_lock(&tasklist_lock);
384 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
385 read_unlock(&tasklist_lock);
386
387 return retval;
388}
389
390static bool has_stopped_jobs(struct pid *pgrp)
391{
392 struct task_struct *p;
393
394 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
395 if (p->signal->flags & SIGNAL_STOP_STOPPED)
396 return true;
397 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
398
399 return false;
400}
401
402/*
403 * Check to see if any process groups have become orphaned as
404 * a result of our exiting, and if they have any stopped jobs,
405 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
406 */
407static void
408kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
409{
410 struct pid *pgrp = task_pgrp(tsk);
411 struct task_struct *ignored_task = tsk;
412
413 if (!parent)
414 /* exit: our father is in a different pgrp than
415 * we are and we were the only connection outside.
416 */
417 parent = tsk->real_parent;
418 else
419 /* reparent: our child is in a different pgrp than
420 * we are, and it was the only connection outside.
421 */
422 ignored_task = NULL;
423
424 if (task_pgrp(parent) != pgrp &&
425 task_session(parent) == task_session(tsk) &&
426 will_become_orphaned_pgrp(pgrp, ignored_task) &&
427 has_stopped_jobs(pgrp)) {
428 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
429 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
430 }
431}
432
433static void coredump_task_exit(struct task_struct *tsk,
434 struct core_state *core_state)
435{
436 struct core_thread self;
437
438 self.task = tsk;
439 if (self.task->flags & PF_SIGNALED)
440 self.next = xchg(&core_state->tasks, &self);
441 else
442 self.task = NULL;
443 /*
444 * Implies mb(), the result of xchg() must be visible
445 * to the dumper.
446 */
447 atomic_dec_and_wake_up(&core_state->threads_remaining);
448
449 for (;;) {
450 set_current_state(TASK_IDLE|TASK_FREEZABLE);
451 if (!self.task) /* see coredump_finish() */
452 break;
453 schedule();
454 }
455 __set_current_state(TASK_RUNNING);
456}
457
458#ifdef CONFIG_MEMCG
459/* drops tasklist_lock if succeeds */
460static bool __try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm)
461{
462 bool ret = false;
463
464 task_lock(tsk);
465 if (likely(tsk->mm == mm)) {
466 /* tsk can't pass exit_mm/exec_mmap and exit */
467 read_unlock(&tasklist_lock);
468 WRITE_ONCE(mm->owner, tsk);
469 lru_gen_migrate_mm(mm);
470 ret = true;
471 }
472 task_unlock(tsk);
473 return ret;
474}
475
476static bool try_to_set_owner(struct task_struct *g, struct mm_struct *mm)
477{
478 struct task_struct *t;
479
480 for_each_thread(g, t) {
481 struct mm_struct *t_mm = READ_ONCE(t->mm);
482 if (t_mm == mm) {
483 if (__try_to_set_owner(t, mm))
484 return true;
485 } else if (t_mm)
486 break;
487 }
488
489 return false;
490}
491
492/*
493 * A task is exiting. If it owned this mm, find a new owner for the mm.
494 */
495void mm_update_next_owner(struct mm_struct *mm)
496{
497 struct task_struct *g, *p = current;
498
499 /*
500 * If the exiting or execing task is not the owner, it's
501 * someone else's problem.
502 */
503 if (mm->owner != p)
504 return;
505 /*
506 * The current owner is exiting/execing and there are no other
507 * candidates. Do not leave the mm pointing to a possibly
508 * freed task structure.
509 */
510 if (atomic_read(&mm->mm_users) <= 1) {
511 WRITE_ONCE(mm->owner, NULL);
512 return;
513 }
514
515 read_lock(&tasklist_lock);
516 /*
517 * Search in the children
518 */
519 list_for_each_entry(g, &p->children, sibling) {
520 if (try_to_set_owner(g, mm))
521 goto ret;
522 }
523 /*
524 * Search in the siblings
525 */
526 list_for_each_entry(g, &p->real_parent->children, sibling) {
527 if (try_to_set_owner(g, mm))
528 goto ret;
529 }
530 /*
531 * Search through everything else, we should not get here often.
532 */
533 for_each_process(g) {
534 if (atomic_read(&mm->mm_users) <= 1)
535 break;
536 if (g->flags & PF_KTHREAD)
537 continue;
538 if (try_to_set_owner(g, mm))
539 goto ret;
540 }
541 read_unlock(&tasklist_lock);
542 /*
543 * We found no owner yet mm_users > 1: this implies that we are
544 * most likely racing with swapoff (try_to_unuse()) or /proc or
545 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
546 */
547 WRITE_ONCE(mm->owner, NULL);
548 ret:
549 return;
550
551}
552#endif /* CONFIG_MEMCG */
553
554/*
555 * Turn us into a lazy TLB process if we
556 * aren't already..
557 */
558static void exit_mm(void)
559{
560 struct mm_struct *mm = current->mm;
561
562 mm_exit_exec_release(current, mm);
563 if (!mm)
564 return;
565
566 sched_cache_exit_mm(current);
567
568 mmap_read_lock(mm);
569 mmgrab_lazy_tlb(mm);
570 BUG_ON(mm != current->active_mm);
571 /* more a memory barrier than a real lock */
572 task_lock(current);
573 /*
574 * When a thread stops operating on an address space, the loop
575 * in membarrier_private_expedited() may not observe that
576 * tsk->mm, and the loop in membarrier_global_expedited() may
577 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
578 * rq->membarrier_state, so those would not issue an IPI.
579 * Membarrier requires a memory barrier after accessing
580 * user-space memory, before clearing tsk->mm or the
581 * rq->membarrier_state.
582 */
583 smp_mb__after_spinlock();
584 local_irq_disable();
585 current->mm = NULL;
586 membarrier_update_current_mm(NULL);
587 enter_lazy_tlb(mm, current);
588 local_irq_enable();
589 task_unlock(current);
590 mmap_read_unlock(mm);
591 mm_update_next_owner(mm);
592 mmput(mm);
593 if (test_thread_flag(TIF_MEMDIE))
594 exit_oom_victim();
595}
596
597static struct task_struct *find_alive_thread(struct task_struct *p)
598{
599 struct task_struct *t;
600
601 for_each_thread(p, t) {
602 if (!(t->flags & PF_EXITING))
603 return t;
604 }
605 return NULL;
606}
607
608static struct task_struct *find_child_reaper(struct task_struct *father,
609 struct list_head *dead)
610 __releases(&tasklist_lock)
611 __acquires(&tasklist_lock)
612{
613 struct pid_namespace *pid_ns = task_active_pid_ns(father);
614 struct task_struct *reaper = pid_ns->child_reaper;
615 struct task_struct *p, *n;
616
617 if (likely(reaper != father))
618 return reaper;
619
620 reaper = find_alive_thread(father);
621 if (reaper) {
622 ASSERT_EXCLUSIVE_WRITER(pid_ns->child_reaper);
623 WRITE_ONCE(pid_ns->child_reaper, reaper);
624 return reaper;
625 }
626
627 write_unlock_irq(&tasklist_lock);
628
629 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
630 list_del_init(&p->ptrace_entry);
631 release_task(p);
632 }
633
634 zap_pid_ns_processes(pid_ns);
635 write_lock_irq(&tasklist_lock);
636
637 return father;
638}
639
640/*
641 * When we die, we re-parent all our children, and try to:
642 * 1. give them to another thread in our thread group, if such a member exists
643 * 2. give it to the first ancestor process which prctl'd itself as a
644 * child_subreaper for its children (like a service manager)
645 * 3. give it to the init process (PID 1) in our pid namespace
646 */
647static struct task_struct *find_new_reaper(struct task_struct *father,
648 struct task_struct *child_reaper)
649{
650 struct task_struct *thread, *reaper;
651
652 thread = find_alive_thread(father);
653 if (thread)
654 return thread;
655
656 if (father->signal->has_child_subreaper) {
657 unsigned int ns_level = task_pid(father)->level;
658 /*
659 * Find the first ->is_child_subreaper ancestor in our pid_ns.
660 * We can't check reaper != child_reaper to ensure we do not
661 * cross the namespaces, the exiting parent could be injected
662 * by setns() + fork().
663 * We check pid->level, this is slightly more efficient than
664 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
665 */
666 for (reaper = father->real_parent;
667 task_pid(reaper)->level == ns_level;
668 reaper = reaper->real_parent) {
669 if (reaper == &init_task)
670 break;
671 if (!reaper->signal->is_child_subreaper)
672 continue;
673 thread = find_alive_thread(reaper);
674 if (thread)
675 return thread;
676 }
677 }
678
679 return child_reaper;
680}
681
682/*
683* Any that need to be release_task'd are put on the @dead list.
684 */
685static void reparent_leader(struct task_struct *father, struct task_struct *p,
686 struct list_head *dead)
687{
688 if (unlikely(p->exit_state == EXIT_DEAD))
689 return;
690
691 /* We don't want people slaying init. */
692 p->exit_signal = SIGCHLD;
693
694 /* If it has exited notify the new parent about this child's death. */
695 if (!p->ptrace &&
696 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
697 if (do_notify_parent(p, p->exit_signal)) {
698 p->exit_state = EXIT_DEAD;
699 list_add(&p->ptrace_entry, dead);
700 }
701 }
702
703 kill_orphaned_pgrp(p, father);
704}
705
706/*
707 * Make init inherit all the child processes
708 */
709static void forget_original_parent(struct task_struct *father,
710 struct list_head *dead)
711{
712 struct task_struct *p, *t, *reaper;
713
714 if (unlikely(!list_empty(&father->ptraced)))
715 exit_ptrace(father, dead);
716
717 /* Can drop and reacquire tasklist_lock */
718 reaper = find_child_reaper(father, dead);
719 if (list_empty(&father->children))
720 return;
721
722 reaper = find_new_reaper(father, reaper);
723 list_for_each_entry(p, &father->children, sibling) {
724 for_each_thread(p, t) {
725 RCU_INIT_POINTER(t->real_parent, reaper);
726 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
727 if (likely(!t->ptrace))
728 t->parent = t->real_parent;
729 if (t->pdeath_signal)
730 group_send_sig_info(t->pdeath_signal,
731 SEND_SIG_NOINFO, t,
732 PIDTYPE_TGID);
733 }
734 /*
735 * If this is a threaded reparent there is no need to
736 * notify anyone anything has happened.
737 */
738 if (!same_thread_group(reaper, father))
739 reparent_leader(father, p, dead);
740 }
741 list_splice_tail_init(&father->children, &reaper->children);
742}
743
744/*
745 * Send signals to all our closest relatives so that they know
746 * to properly mourn us..
747 */
748static void exit_notify(struct task_struct *tsk, int group_dead)
749{
750 bool autoreap;
751 struct task_struct *p, *n;
752 LIST_HEAD(dead);
753
754 write_lock_irq(&tasklist_lock);
755 forget_original_parent(tsk, &dead);
756
757 if (group_dead)
758 kill_orphaned_pgrp(tsk->group_leader, NULL);
759
760 tsk->exit_state = EXIT_ZOMBIE;
761
762 if (unlikely(tsk->ptrace)) {
763 int sig = thread_group_empty(tsk) && !ptrace_reparented(tsk)
764 ? tsk->exit_signal : SIGCHLD;
765 autoreap = do_notify_parent(tsk, sig);
766 } else if (thread_group_leader(tsk)) {
767 autoreap = thread_group_empty(tsk) &&
768 do_notify_parent(tsk, tsk->exit_signal);
769 } else {
770 autoreap = true;
771 /* untraced sub-thread */
772 do_notify_pidfd(tsk);
773 }
774
775 if (autoreap) {
776 tsk->exit_state = EXIT_DEAD;
777 list_add(&tsk->ptrace_entry, &dead);
778 }
779
780 /* mt-exec, de_thread() is waiting for group leader */
781 if (unlikely(tsk->signal->notify_count < 0))
782 wake_up_process(tsk->signal->group_exec_task);
783 write_unlock_irq(&tasklist_lock);
784
785 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
786 list_del_init(&p->ptrace_entry);
787 release_task(p);
788 }
789}
790
791#ifdef CONFIG_DEBUG_STACK_USAGE
792#ifdef CONFIG_STACK_GROWSUP
793unsigned long stack_not_used(struct task_struct *p)
794{
795 unsigned long *n = end_of_stack(p);
796
797 do { /* Skip over canary */
798 n--;
799 } while (!*n);
800
801 return (unsigned long)end_of_stack(p) - (unsigned long)n;
802}
803#else /* !CONFIG_STACK_GROWSUP */
804unsigned long stack_not_used(struct task_struct *p)
805{
806 unsigned long *n = end_of_stack(p);
807
808 do { /* Skip over canary */
809 n++;
810 } while (!*n);
811
812 return (unsigned long)n - (unsigned long)end_of_stack(p);
813}
814#endif /* CONFIG_STACK_GROWSUP */
815
816/* Count the maximum pages reached in kernel stacks */
817static inline void kstack_histogram(unsigned long used_stack)
818{
819#ifdef CONFIG_VM_EVENT_COUNTERS
820 if (used_stack <= 1024)
821 count_vm_event(KSTACK_1K);
822#if THREAD_SIZE > 1024
823 else if (used_stack <= 2048)
824 count_vm_event(KSTACK_2K);
825#endif
826#if THREAD_SIZE > 2048
827 else if (used_stack <= 4096)
828 count_vm_event(KSTACK_4K);
829#endif
830#if THREAD_SIZE > 4096
831 else if (used_stack <= 8192)
832 count_vm_event(KSTACK_8K);
833#endif
834#if THREAD_SIZE > 8192
835 else if (used_stack <= 16384)
836 count_vm_event(KSTACK_16K);
837#endif
838#if THREAD_SIZE > 16384
839 else if (used_stack <= 32768)
840 count_vm_event(KSTACK_32K);
841#endif
842#if THREAD_SIZE > 32768
843 else if (used_stack <= 65536)
844 count_vm_event(KSTACK_64K);
845#endif
846#if THREAD_SIZE > 65536
847 else
848 count_vm_event(KSTACK_REST);
849#endif
850#endif /* CONFIG_VM_EVENT_COUNTERS */
851}
852
853static void check_stack_usage(void)
854{
855 static DEFINE_SPINLOCK(low_water_lock);
856 static int lowest_to_date = THREAD_SIZE;
857 unsigned long free;
858
859 free = stack_not_used(current);
860 kstack_histogram(THREAD_SIZE - free);
861
862 if (free >= lowest_to_date)
863 return;
864
865 spin_lock(&low_water_lock);
866 if (free < lowest_to_date) {
867 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
868 current->comm, task_pid_nr(current), free);
869 lowest_to_date = free;
870 }
871 spin_unlock(&low_water_lock);
872}
873#else /* !CONFIG_DEBUG_STACK_USAGE */
874static inline void check_stack_usage(void) {}
875#endif /* CONFIG_DEBUG_STACK_USAGE */
876
877static void synchronize_group_exit(struct task_struct *tsk, long code)
878{
879 struct sighand_struct *sighand = tsk->sighand;
880 struct signal_struct *signal = tsk->signal;
881 struct core_state *core_state;
882
883 spin_lock_irq(&sighand->siglock);
884 signal->quick_threads--;
885 if ((signal->quick_threads == 0) &&
886 !(signal->flags & SIGNAL_GROUP_EXIT)) {
887 signal->flags = SIGNAL_GROUP_EXIT;
888 signal->group_exit_code = code;
889 signal->group_stop_count = 0;
890 }
891 /*
892 * Serialize with any possible pending coredump.
893 * We must hold siglock around checking core_state
894 * and setting PF_POSTCOREDUMP. The core-inducing thread
895 * will increment ->threads_remaining for each thread in the
896 * group without PF_POSTCOREDUMP set.
897 */
898 tsk->flags |= PF_POSTCOREDUMP;
899 core_state = signal->core_state;
900 spin_unlock_irq(&sighand->siglock);
901
902 if (unlikely(core_state))
903 coredump_task_exit(tsk, core_state);
904}
905
906void __noreturn do_exit(long code)
907{
908 struct task_struct *tsk = current;
909 struct kthread *kthread;
910 int group_dead;
911
912 WARN_ON(irqs_disabled());
913 WARN_ON(tsk->plug);
914
915 kthread = tsk_is_kthread(tsk);
916 if (unlikely(kthread))
917 kthread_do_exit(kthread, code);
918
919 kcov_task_exit(tsk);
920 kmsan_task_exit(tsk);
921
922 synchronize_group_exit(tsk, code);
923 ptrace_event(PTRACE_EVENT_EXIT, code);
924 user_events_exit(tsk);
925
926 io_uring_files_cancel();
927 sched_mm_cid_exit(tsk);
928 exit_signals(tsk); /* sets PF_EXITING */
929
930 seccomp_filter_release(tsk);
931
932 acct_update_integrals(tsk);
933 group_dead = atomic_dec_and_test(&tsk->signal->live);
934 if (group_dead) {
935 /*
936 * If the last thread of global init has exited, panic
937 * immediately to get a useable coredump.
938 */
939 if (unlikely(is_global_init(tsk)))
940 panic("Attempted to kill init! exitcode=0x%08x\n",
941 tsk->signal->group_exit_code ?: (int)code);
942
943#ifdef CONFIG_POSIX_TIMERS
944 hrtimer_cancel(&tsk->signal->real_timer);
945 exit_itimers(tsk);
946#endif
947 if (tsk->mm)
948 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
949 }
950 acct_collect(code, group_dead);
951 if (group_dead)
952 tty_audit_exit();
953 audit_free(tsk);
954
955 tsk->exit_code = code;
956 taskstats_exit(tsk, group_dead);
957 trace_sched_process_exit(tsk, group_dead);
958
959 /*
960 * Since sampling can touch ->mm, make sure to stop everything before we
961 * tear it down.
962 *
963 * Also flushes inherited counters to the parent - before the parent
964 * gets woken up by child-exit notifications.
965 */
966 perf_event_exit_task(tsk);
967 /*
968 * PF_EXITING (above) ensures unwind_deferred_request() will no
969 * longer add new unwinds. While exit_mm() (below) will destroy the
970 * abaility to do unwinds. So flush any pending unwinds here.
971 */
972 unwind_deferred_task_exit(tsk);
973
974 exit_mm();
975
976 if (group_dead)
977 acct_process();
978
979 exit_sem(tsk);
980 exit_shm(tsk);
981 /* Hang the tty up before the last close of it can clear the session. */
982 if (group_dead)
983 disassociate_ctty(1);
984 exit_files(tsk);
985 exit_fs(tsk);
986 exit_nsproxy_namespaces(tsk);
987 exit_task_work(tsk);
988 exit_thread(tsk);
989
990 sched_autogroup_exit_task(tsk);
991 cgroup_task_exit(tsk);
992
993 /*
994 * FIXME: do that only when needed, using sched_exit tracepoint
995 */
996 flush_ptrace_hw_breakpoint(tsk);
997
998 exit_tasks_rcu_start();
999 exit_notify(tsk, group_dead);
1000 proc_exit_connector(tsk);
1001 mpol_put_task_policy(tsk);
1002#ifdef CONFIG_FUTEX
1003 if (unlikely(current->futex.pi_state_cache))
1004 kfree(current->futex.pi_state_cache);
1005#endif
1006 /*
1007 * Make sure we are holding no locks:
1008 */
1009 debug_check_no_locks_held();
1010
1011 if (tsk->io_context)
1012 exit_io_context(tsk);
1013
1014 if (tsk->splice_pipe)
1015 free_pipe_info(tsk->splice_pipe);
1016
1017 if (tsk->task_frag.page)
1018 put_page(tsk->task_frag.page);
1019
1020 exit_task_stack_account(tsk);
1021
1022 check_stack_usage();
1023 preempt_disable();
1024 if (tsk->nr_dirtied)
1025 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
1026 exit_rcu();
1027 exit_tasks_rcu_finish();
1028
1029 lockdep_free_task(tsk);
1030 do_task_dead();
1031}
1032EXPORT_SYMBOL(do_exit);
1033
1034void __noreturn make_task_dead(int signr)
1035{
1036 /*
1037 * Take the task off the cpu after something catastrophic has
1038 * happened.
1039 *
1040 * We can get here from a kernel oops, sometimes with preemption off.
1041 * Start by checking for critical errors.
1042 * Then fix up important state like USER_DS and preemption.
1043 * Then do everything else.
1044 */
1045 struct task_struct *tsk = current;
1046 unsigned int limit;
1047
1048 if (unlikely(in_interrupt()))
1049 panic("Aiee, killing interrupt handler!");
1050 if (unlikely(!tsk->pid))
1051 panic("Attempted to kill the idle task!");
1052
1053 if (unlikely(irqs_disabled())) {
1054 pr_info("note: %s[%d] exited with irqs disabled\n",
1055 current->comm, task_pid_nr(current));
1056 local_irq_enable();
1057 }
1058 if (unlikely(in_atomic())) {
1059 pr_info("note: %s[%d] exited with preempt_count %d\n",
1060 current->comm, task_pid_nr(current),
1061 preempt_count());
1062 preempt_count_set(PREEMPT_ENABLED);
1063 }
1064
1065 /*
1066 * Every time the system oopses, if the oops happens while a reference
1067 * to an object was held, the reference leaks.
1068 * If the oops doesn't also leak memory, repeated oopsing can cause
1069 * reference counters to wrap around (if they're not using refcount_t).
1070 * This means that repeated oopsing can make unexploitable-looking bugs
1071 * exploitable through repeated oopsing.
1072 * To make sure this can't happen, place an upper bound on how often the
1073 * kernel may oops without panic().
1074 */
1075 limit = READ_ONCE(oops_limit);
1076 if (atomic_inc_return(&oops_count) >= limit && limit)
1077 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
1078
1079 /*
1080 * We're taking recursive faults here in make_task_dead. Safest is to just
1081 * leave this task alone and wait for reboot.
1082 */
1083 if (unlikely(tsk->flags & PF_EXITING)) {
1084 pr_alert("Fixing recursive fault but reboot is needed!\n");
1085 futex_exit_recursive(tsk);
1086 tsk->exit_state = EXIT_DEAD;
1087 refcount_inc(&tsk->rcu_users);
1088 preempt_disable();
1089 do_task_dead();
1090 }
1091
1092 do_exit(signr);
1093}
1094
1095SYSCALL_DEFINE1(exit, int, error_code)
1096{
1097 do_exit((error_code & 0xff) << 8);
1098}
1099
1100/*
1101 * Take down every thread in the group. This is called by fatal signals
1102 * as well as by sys_exit_group (below).
1103 */
1104void __noreturn
1105do_group_exit(int exit_code)
1106{
1107 struct signal_struct *sig = current->signal;
1108
1109 if (sig->flags & SIGNAL_GROUP_EXIT)
1110 exit_code = sig->group_exit_code;
1111 else if (sig->group_exec_task)
1112 exit_code = 0;
1113 else {
1114 struct sighand_struct *const sighand = current->sighand;
1115
1116 spin_lock_irq(&sighand->siglock);
1117 if (sig->flags & SIGNAL_GROUP_EXIT)
1118 /* Another thread got here before we took the lock. */
1119 exit_code = sig->group_exit_code;
1120 else if (sig->group_exec_task)
1121 exit_code = 0;
1122 else {
1123 sig->group_exit_code = exit_code;
1124 sig->flags = SIGNAL_GROUP_EXIT;
1125 zap_other_threads(current);
1126 }
1127 spin_unlock_irq(&sighand->siglock);
1128 }
1129
1130 do_exit(exit_code);
1131 /* NOTREACHED */
1132}
1133
1134/*
1135 * this kills every thread in the thread group. Note that any externally
1136 * wait4()-ing process will get the correct exit code - even if this
1137 * thread is not the thread group leader.
1138 */
1139SYSCALL_DEFINE1(exit_group, int, error_code)
1140{
1141 do_group_exit((error_code & 0xff) << 8);
1142 /* NOTREACHED */
1143 return 0;
1144}
1145
1146static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1147{
1148 return wo->wo_type == PIDTYPE_MAX ||
1149 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1150}
1151
1152static int
1153eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1154{
1155 if (!eligible_pid(wo, p))
1156 return 0;
1157
1158 /*
1159 * Wait for all children (clone and not) if __WALL is set or
1160 * if it is traced by us.
1161 */
1162 if (ptrace || (wo->wo_flags & __WALL))
1163 return 1;
1164
1165 /*
1166 * Otherwise, wait for clone children *only* if __WCLONE is set;
1167 * otherwise, wait for non-clone children *only*.
1168 *
1169 * Note: a "clone" child here is one that reports to its parent
1170 * using a signal other than SIGCHLD, or a non-leader thread which
1171 * we can only see if it is traced by us.
1172 */
1173 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1174 return 0;
1175
1176 return 1;
1177}
1178
1179/*
1180 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1181 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1182 * the lock and this task is uninteresting. If we return nonzero, we have
1183 * released the lock and the system call should return.
1184 */
1185static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1186{
1187 int state, status;
1188 pid_t pid = task_pid_vnr(p);
1189 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1190 struct waitid_info *infop;
1191
1192 if (!likely(wo->wo_flags & WEXITED))
1193 return 0;
1194
1195 if (unlikely(wo->wo_flags & WNOWAIT)) {
1196 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1197 ? p->signal->group_exit_code : p->exit_code;
1198 get_task_struct(p);
1199 read_unlock(&tasklist_lock);
1200 sched_annotate_sleep();
1201 if (wo->wo_rusage)
1202 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1203 put_task_struct(p);
1204 goto out_info;
1205 }
1206 /*
1207 * Move the task's state to DEAD/TRACE, only one thread can do this.
1208 */
1209 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1210 EXIT_TRACE : EXIT_DEAD;
1211 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1212 return 0;
1213 /*
1214 * We own this thread, nobody else can reap it.
1215 */
1216 read_unlock(&tasklist_lock);
1217 sched_annotate_sleep();
1218
1219 /*
1220 * Check thread_group_leader() to exclude the traced sub-threads.
1221 */
1222 if (state == EXIT_DEAD && thread_group_leader(p)) {
1223 struct signal_struct *sig = p->signal;
1224 struct signal_struct *psig = current->signal;
1225 unsigned long maxrss;
1226 u64 tgutime, tgstime;
1227
1228 /*
1229 * The resource counters for the group leader are in its
1230 * own task_struct. Those for dead threads in the group
1231 * are in its signal_struct, as are those for the child
1232 * processes it has previously reaped. All these
1233 * accumulate in the parent's signal_struct c* fields.
1234 *
1235 * We don't bother to take a lock here to protect these
1236 * p->signal fields because the whole thread group is dead
1237 * and nobody can change them.
1238 *
1239 * psig->stats_lock also protects us from our sub-threads
1240 * which can reap other children at the same time.
1241 *
1242 * We use thread_group_cputime_adjusted() to get times for
1243 * the thread group, which consolidates times for all threads
1244 * in the group including the group leader.
1245 */
1246 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1247 write_seqlock_irq(&psig->stats_lock);
1248 psig->cutime += tgutime + sig->cutime;
1249 psig->cstime += tgstime + sig->cstime;
1250 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1251 psig->cmin_flt +=
1252 p->min_flt + sig->min_flt + sig->cmin_flt;
1253 psig->cmaj_flt +=
1254 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1255 psig->cnvcsw +=
1256 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1257 psig->cnivcsw +=
1258 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1259 psig->cinblock +=
1260 task_io_get_inblock(p) +
1261 sig->inblock + sig->cinblock;
1262 psig->coublock +=
1263 task_io_get_oublock(p) +
1264 sig->oublock + sig->coublock;
1265 maxrss = max(sig->maxrss, sig->cmaxrss);
1266 if (psig->cmaxrss < maxrss)
1267 psig->cmaxrss = maxrss;
1268 task_io_accounting_add(&psig->ioac, &p->ioac);
1269 task_io_accounting_add(&psig->ioac, &sig->ioac);
1270 write_sequnlock_irq(&psig->stats_lock);
1271 }
1272
1273 if (wo->wo_rusage)
1274 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1275 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1276 ? p->signal->group_exit_code : p->exit_code;
1277 wo->wo_stat = status;
1278
1279 if (state == EXIT_TRACE) {
1280 write_lock_irq(&tasklist_lock);
1281 /* We dropped tasklist, ptracer could die and untrace */
1282 ptrace_unlink(p);
1283
1284 /* If parent wants a zombie, don't release it now */
1285 state = EXIT_ZOMBIE;
1286 if (do_notify_parent(p, p->exit_signal))
1287 state = EXIT_DEAD;
1288 p->exit_state = state;
1289 write_unlock_irq(&tasklist_lock);
1290 }
1291 if (state == EXIT_DEAD)
1292 release_task(p);
1293
1294out_info:
1295 infop = wo->wo_info;
1296 if (infop) {
1297 if ((status & 0x7f) == 0) {
1298 infop->cause = CLD_EXITED;
1299 infop->status = status >> 8;
1300 } else {
1301 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1302 infop->status = status & 0x7f;
1303 }
1304 infop->pid = pid;
1305 infop->uid = uid;
1306 }
1307
1308 return pid;
1309}
1310
1311static int *task_stopped_code(struct task_struct *p, bool ptrace)
1312{
1313 if (ptrace) {
1314 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1315 return &p->exit_code;
1316 } else {
1317 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1318 return &p->signal->group_exit_code;
1319 }
1320 return NULL;
1321}
1322
1323/**
1324 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1325 * @wo: wait options
1326 * @ptrace: is the wait for ptrace
1327 * @p: task to wait for
1328 *
1329 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1330 *
1331 * CONTEXT:
1332 * read_lock(&tasklist_lock), which is released if return value is
1333 * non-zero. Also, grabs and releases @p->sighand->siglock.
1334 *
1335 * RETURNS:
1336 * 0 if wait condition didn't exist and search for other wait conditions
1337 * should continue. Non-zero return, -errno on failure and @p's pid on
1338 * success, implies that tasklist_lock is released and wait condition
1339 * search should terminate.
1340 */
1341static int wait_task_stopped(struct wait_opts *wo,
1342 int ptrace, struct task_struct *p)
1343{
1344 struct waitid_info *infop;
1345 int exit_code, *p_code, why;
1346 uid_t uid = 0; /* unneeded, required by compiler */
1347 pid_t pid;
1348
1349 /*
1350 * Traditionally we see ptrace'd stopped tasks regardless of options.
1351 */
1352 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1353 return 0;
1354
1355 if (!task_stopped_code(p, ptrace))
1356 return 0;
1357
1358 exit_code = 0;
1359 spin_lock_irq(&p->sighand->siglock);
1360
1361 p_code = task_stopped_code(p, ptrace);
1362 if (unlikely(!p_code))
1363 goto unlock_sig;
1364
1365 exit_code = *p_code;
1366 if (!exit_code)
1367 goto unlock_sig;
1368
1369 if (!unlikely(wo->wo_flags & WNOWAIT))
1370 *p_code = 0;
1371
1372 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1373unlock_sig:
1374 spin_unlock_irq(&p->sighand->siglock);
1375 if (!exit_code)
1376 return 0;
1377
1378 /*
1379 * Now we are pretty sure this task is interesting.
1380 * Make sure it doesn't get reaped out from under us while we
1381 * give up the lock and then examine it below. We don't want to
1382 * keep holding onto the tasklist_lock while we call getrusage and
1383 * possibly take page faults for user memory.
1384 */
1385 get_task_struct(p);
1386 pid = task_pid_vnr(p);
1387 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1388 read_unlock(&tasklist_lock);
1389 sched_annotate_sleep();
1390 if (wo->wo_rusage)
1391 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1392 put_task_struct(p);
1393
1394 if (likely(!(wo->wo_flags & WNOWAIT)))
1395 wo->wo_stat = (exit_code << 8) | 0x7f;
1396
1397 infop = wo->wo_info;
1398 if (infop) {
1399 infop->cause = why;
1400 infop->status = exit_code;
1401 infop->pid = pid;
1402 infop->uid = uid;
1403 }
1404 return pid;
1405}
1406
1407/*
1408 * Handle do_wait work for one task in a live, non-stopped state.
1409 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1410 * the lock and this task is uninteresting. If we return nonzero, we have
1411 * released the lock and the system call should return.
1412 */
1413static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1414{
1415 struct waitid_info *infop;
1416 pid_t pid;
1417 uid_t uid;
1418
1419 if (!unlikely(wo->wo_flags & WCONTINUED))
1420 return 0;
1421
1422 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1423 return 0;
1424
1425 spin_lock_irq(&p->sighand->siglock);
1426 /* Re-check with the lock held. */
1427 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1428 spin_unlock_irq(&p->sighand->siglock);
1429 return 0;
1430 }
1431 if (!unlikely(wo->wo_flags & WNOWAIT))
1432 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1433 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1434 spin_unlock_irq(&p->sighand->siglock);
1435
1436 pid = task_pid_vnr(p);
1437 get_task_struct(p);
1438 read_unlock(&tasklist_lock);
1439 sched_annotate_sleep();
1440 if (wo->wo_rusage)
1441 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1442 put_task_struct(p);
1443
1444 infop = wo->wo_info;
1445 if (!infop) {
1446 wo->wo_stat = 0xffff;
1447 } else {
1448 infop->cause = CLD_CONTINUED;
1449 infop->pid = pid;
1450 infop->uid = uid;
1451 infop->status = SIGCONT;
1452 }
1453 return pid;
1454}
1455
1456/*
1457 * Consider @p for a wait by @parent.
1458 *
1459 * -ECHILD should be in ->notask_error before the first call.
1460 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1461 * Returns zero if the search for a child should continue;
1462 * then ->notask_error is 0 if @p is an eligible child,
1463 * or still -ECHILD.
1464 */
1465static int wait_consider_task(struct wait_opts *wo, int ptrace,
1466 struct task_struct *p)
1467{
1468 /*
1469 * We can race with wait_task_zombie() from another thread.
1470 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1471 * can't confuse the checks below.
1472 */
1473 int exit_state = READ_ONCE(p->exit_state);
1474 int ret;
1475
1476 if (unlikely(exit_state == EXIT_DEAD))
1477 return 0;
1478
1479 ret = eligible_child(wo, ptrace, p);
1480 if (!ret)
1481 return ret;
1482
1483 if (unlikely(exit_state == EXIT_TRACE)) {
1484 /*
1485 * ptrace == 0 means we are the natural parent. In this case
1486 * we should clear notask_error, debugger will notify us.
1487 */
1488 if (likely(!ptrace))
1489 wo->notask_error = 0;
1490 return 0;
1491 }
1492
1493 if (likely(!ptrace) && unlikely(p->ptrace)) {
1494 /*
1495 * If it is traced by its real parent's group, just pretend
1496 * the caller is ptrace_do_wait() and reap this child if it
1497 * is zombie.
1498 *
1499 * This also hides group stop state from real parent; otherwise
1500 * a single stop can be reported twice as group and ptrace stop.
1501 * If a ptracer wants to distinguish these two events for its
1502 * own children it should create a separate process which takes
1503 * the role of real parent.
1504 */
1505 if (!ptrace_reparented(p))
1506 ptrace = 1;
1507 }
1508
1509 /* slay zombie? */
1510 if (exit_state == EXIT_ZOMBIE) {
1511 /* we don't reap group leaders with subthreads */
1512 if (!delay_group_leader(p)) {
1513 /*
1514 * A zombie ptracee is only visible to its ptracer.
1515 * Notification and reaping will be cascaded to the
1516 * real parent when the ptracer detaches.
1517 */
1518 if (unlikely(ptrace) || likely(!p->ptrace))
1519 return wait_task_zombie(wo, p);
1520 }
1521
1522 /*
1523 * Allow access to stopped/continued state via zombie by
1524 * falling through. Clearing of notask_error is complex.
1525 *
1526 * When !@ptrace:
1527 *
1528 * If WEXITED is set, notask_error should naturally be
1529 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1530 * so, if there are live subthreads, there are events to
1531 * wait for. If all subthreads are dead, it's still safe
1532 * to clear - this function will be called again in finite
1533 * amount time once all the subthreads are released and
1534 * will then return without clearing.
1535 *
1536 * When @ptrace:
1537 *
1538 * Stopped state is per-task and thus can't change once the
1539 * target task dies. Only continued and exited can happen.
1540 * Clear notask_error if WCONTINUED | WEXITED.
1541 */
1542 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1543 wo->notask_error = 0;
1544 } else {
1545 /*
1546 * @p is alive and it's gonna stop, continue or exit, so
1547 * there always is something to wait for.
1548 */
1549 wo->notask_error = 0;
1550 }
1551
1552 /*
1553 * Wait for stopped. Depending on @ptrace, different stopped state
1554 * is used and the two don't interact with each other.
1555 */
1556 ret = wait_task_stopped(wo, ptrace, p);
1557 if (ret)
1558 return ret;
1559
1560 /*
1561 * Wait for continued. There's only one continued state and the
1562 * ptracer can consume it which can confuse the real parent. Don't
1563 * use WCONTINUED from ptracer. You don't need or want it.
1564 */
1565 return wait_task_continued(wo, p);
1566}
1567
1568/*
1569 * Do the work of do_wait() for one thread in the group, @tsk.
1570 *
1571 * -ECHILD should be in ->notask_error before the first call.
1572 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1573 * Returns zero if the search for a child should continue; then
1574 * ->notask_error is 0 if there were any eligible children,
1575 * or still -ECHILD.
1576 */
1577static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1578{
1579 struct task_struct *p;
1580
1581 list_for_each_entry(p, &tsk->children, sibling) {
1582 int ret = wait_consider_task(wo, 0, p);
1583
1584 if (ret)
1585 return ret;
1586 }
1587
1588 return 0;
1589}
1590
1591static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1592{
1593 struct task_struct *p;
1594
1595 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1596 int ret = wait_consider_task(wo, 1, p);
1597
1598 if (ret)
1599 return ret;
1600 }
1601
1602 return 0;
1603}
1604
1605bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1606{
1607 if (!eligible_pid(wo, p))
1608 return false;
1609
1610 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1611 return false;
1612
1613 return true;
1614}
1615
1616static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1617 int sync, void *key)
1618{
1619 struct wait_opts *wo = container_of(wait, struct wait_opts,
1620 child_wait);
1621 struct task_struct *p = key;
1622
1623 if (pid_child_should_wake(wo, p))
1624 return default_wake_function(wait, mode, sync, key);
1625
1626 return 0;
1627}
1628
1629void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1630{
1631 __wake_up_sync_key(&parent->signal->wait_chldexit,
1632 TASK_INTERRUPTIBLE, p);
1633}
1634
1635static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1636 struct task_struct *target)
1637{
1638 struct task_struct *parent =
1639 !ptrace ? target->real_parent : target->parent;
1640
1641 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1642 same_thread_group(current, parent));
1643}
1644
1645/*
1646 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1647 * and tracee lists to find the target task.
1648 */
1649static int do_wait_pid(struct wait_opts *wo)
1650{
1651 bool ptrace;
1652 struct task_struct *target;
1653 int retval;
1654
1655 ptrace = false;
1656 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1657 if (target && is_effectively_child(wo, ptrace, target)) {
1658 retval = wait_consider_task(wo, ptrace, target);
1659 if (retval)
1660 return retval;
1661 }
1662
1663 ptrace = true;
1664 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1665 if (target && target->ptrace &&
1666 is_effectively_child(wo, ptrace, target)) {
1667 retval = wait_consider_task(wo, ptrace, target);
1668 if (retval)
1669 return retval;
1670 }
1671
1672 return 0;
1673}
1674
1675long __do_wait(struct wait_opts *wo)
1676{
1677 long retval;
1678
1679 /*
1680 * If there is nothing that can match our criteria, just get out.
1681 * We will clear ->notask_error to zero if we see any child that
1682 * might later match our criteria, even if we are not able to reap
1683 * it yet.
1684 */
1685 wo->notask_error = -ECHILD;
1686 if ((wo->wo_type < PIDTYPE_MAX) &&
1687 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1688 goto notask;
1689
1690 read_lock(&tasklist_lock);
1691
1692 if (wo->wo_type == PIDTYPE_PID) {
1693 retval = do_wait_pid(wo);
1694 if (retval)
1695 return retval;
1696 } else {
1697 struct task_struct *tsk = current;
1698
1699 do {
1700 retval = do_wait_thread(wo, tsk);
1701 if (retval)
1702 return retval;
1703
1704 retval = ptrace_do_wait(wo, tsk);
1705 if (retval)
1706 return retval;
1707
1708 if (wo->wo_flags & __WNOTHREAD)
1709 break;
1710 } while_each_thread(current, tsk);
1711 }
1712 read_unlock(&tasklist_lock);
1713
1714notask:
1715 retval = wo->notask_error;
1716 if (!retval && !(wo->wo_flags & WNOHANG))
1717 return -ERESTARTSYS;
1718
1719 return retval;
1720}
1721
1722static long do_wait(struct wait_opts *wo)
1723{
1724 int retval;
1725
1726 trace_sched_process_wait(wo->wo_pid);
1727
1728 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1729 wo->child_wait.private = current;
1730 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1731
1732 do {
1733 set_current_state(TASK_INTERRUPTIBLE | TASK_FREEZABLE);
1734 retval = __do_wait(wo);
1735 if (retval != -ERESTARTSYS)
1736 break;
1737 if (signal_pending(current))
1738 break;
1739 schedule();
1740 } while (1);
1741
1742 __set_current_state(TASK_RUNNING);
1743 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1744 return retval;
1745}
1746
1747int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1748 struct waitid_info *infop, int options,
1749 struct rusage *ru)
1750{
1751 unsigned int f_flags = 0;
1752 struct pid *pid = NULL;
1753 enum pid_type type;
1754
1755 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1756 __WNOTHREAD|__WCLONE|__WALL))
1757 return -EINVAL;
1758 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1759 return -EINVAL;
1760
1761 switch (which) {
1762 case P_ALL:
1763 type = PIDTYPE_MAX;
1764 break;
1765 case P_PID:
1766 type = PIDTYPE_PID;
1767 if (upid <= 0)
1768 return -EINVAL;
1769
1770 pid = find_get_pid(upid);
1771 break;
1772 case P_PGID:
1773 type = PIDTYPE_PGID;
1774 if (upid < 0)
1775 return -EINVAL;
1776
1777 if (upid)
1778 pid = find_get_pid(upid);
1779 else
1780 pid = get_task_pid(current, PIDTYPE_PGID);
1781 break;
1782 case P_PIDFD:
1783 type = PIDTYPE_PID;
1784 if (upid < 0)
1785 return -EINVAL;
1786
1787 pid = pidfd_get_pid(upid, &f_flags);
1788 if (IS_ERR(pid))
1789 return PTR_ERR(pid);
1790
1791 break;
1792 default:
1793 return -EINVAL;
1794 }
1795
1796 wo->wo_type = type;
1797 wo->wo_pid = pid;
1798 wo->wo_flags = options;
1799 wo->wo_info = infop;
1800 wo->wo_rusage = ru;
1801 if (f_flags & O_NONBLOCK)
1802 wo->wo_flags |= WNOHANG;
1803
1804 return 0;
1805}
1806
1807static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1808 int options, struct rusage *ru)
1809{
1810 struct wait_opts wo;
1811 long ret;
1812
1813 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1814 if (ret)
1815 return ret;
1816
1817 ret = do_wait(&wo);
1818 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1819 ret = -EAGAIN;
1820
1821 put_pid(wo.wo_pid);
1822 return ret;
1823}
1824
1825SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1826 infop, int, options, struct rusage __user *, ru)
1827{
1828 struct rusage r;
1829 struct waitid_info info = {.status = 0};
1830 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1831 int signo = 0;
1832
1833 if (err > 0) {
1834 signo = SIGCHLD;
1835 err = 0;
1836 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1837 return -EFAULT;
1838 }
1839 if (!infop)
1840 return err;
1841
1842 if (!user_write_access_begin(infop, sizeof(*infop)))
1843 return -EFAULT;
1844
1845 unsafe_put_user(signo, &infop->si_signo, Efault);
1846 unsafe_put_user(0, &infop->si_errno, Efault);
1847 unsafe_put_user(info.cause, &infop->si_code, Efault);
1848 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1849 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1850 unsafe_put_user(info.status, &infop->si_status, Efault);
1851 user_write_access_end();
1852 return err;
1853Efault:
1854 user_write_access_end();
1855 return -EFAULT;
1856}
1857
1858long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1859 struct rusage *ru)
1860{
1861 struct wait_opts wo;
1862 struct pid *pid = NULL;
1863 enum pid_type type;
1864 long ret;
1865
1866 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1867 __WNOTHREAD|__WCLONE|__WALL))
1868 return -EINVAL;
1869
1870 /* -INT_MIN is not defined */
1871 if (upid == INT_MIN)
1872 return -ESRCH;
1873
1874 if (upid == -1)
1875 type = PIDTYPE_MAX;
1876 else if (upid < 0) {
1877 type = PIDTYPE_PGID;
1878 pid = find_get_pid(-upid);
1879 } else if (upid == 0) {
1880 type = PIDTYPE_PGID;
1881 pid = get_task_pid(current, PIDTYPE_PGID);
1882 } else /* upid > 0 */ {
1883 type = PIDTYPE_PID;
1884 pid = find_get_pid(upid);
1885 }
1886
1887 wo.wo_type = type;
1888 wo.wo_pid = pid;
1889 wo.wo_flags = options | WEXITED;
1890 wo.wo_info = NULL;
1891 wo.wo_stat = 0;
1892 wo.wo_rusage = ru;
1893 ret = do_wait(&wo);
1894 put_pid(pid);
1895 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1896 ret = -EFAULT;
1897
1898 return ret;
1899}
1900
1901int kernel_wait(pid_t pid, int *stat)
1902{
1903 struct wait_opts wo = {
1904 .wo_type = PIDTYPE_PID,
1905 .wo_pid = find_get_pid(pid),
1906 .wo_flags = WEXITED,
1907 };
1908 int ret;
1909
1910 ret = do_wait(&wo);
1911 if (ret > 0 && wo.wo_stat)
1912 *stat = wo.wo_stat;
1913 put_pid(wo.wo_pid);
1914 return ret;
1915}
1916
1917SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1918 int, options, struct rusage __user *, ru)
1919{
1920 struct rusage r;
1921 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1922
1923 if (err > 0) {
1924 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1925 return -EFAULT;
1926 }
1927 return err;
1928}
1929
1930#ifdef __ARCH_WANT_SYS_WAITPID
1931
1932/*
1933 * sys_waitpid() remains for compatibility. waitpid() should be
1934 * implemented by calling sys_wait4() from libc.a.
1935 */
1936SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1937{
1938 return kernel_wait4(pid, stat_addr, options, NULL);
1939}
1940
1941#endif
1942
1943#ifdef CONFIG_COMPAT
1944COMPAT_SYSCALL_DEFINE4(wait4,
1945 compat_pid_t, pid,
1946 compat_uint_t __user *, stat_addr,
1947 int, options,
1948 struct compat_rusage __user *, ru)
1949{
1950 struct rusage r;
1951 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1952 if (err > 0) {
1953 if (ru && put_compat_rusage(&r, ru))
1954 return -EFAULT;
1955 }
1956 return err;
1957}
1958
1959COMPAT_SYSCALL_DEFINE5(waitid,
1960 int, which, compat_pid_t, pid,
1961 struct compat_siginfo __user *, infop, int, options,
1962 struct compat_rusage __user *, uru)
1963{
1964 struct rusage ru;
1965 struct waitid_info info = {.status = 0};
1966 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1967 int signo = 0;
1968 if (err > 0) {
1969 signo = SIGCHLD;
1970 err = 0;
1971 if (uru) {
1972 /* kernel_waitid() overwrites everything in ru */
1973 if (COMPAT_USE_64BIT_TIME)
1974 err = copy_to_user(uru, &ru, sizeof(ru));
1975 else
1976 err = put_compat_rusage(&ru, uru);
1977 if (err)
1978 return -EFAULT;
1979 }
1980 }
1981
1982 if (!infop)
1983 return err;
1984
1985 if (!user_write_access_begin(infop, sizeof(*infop)))
1986 return -EFAULT;
1987
1988 unsafe_put_user(signo, &infop->si_signo, Efault);
1989 unsafe_put_user(0, &infop->si_errno, Efault);
1990 unsafe_put_user(info.cause, &infop->si_code, Efault);
1991 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1992 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1993 unsafe_put_user(info.status, &infop->si_status, Efault);
1994 user_write_access_end();
1995 return err;
1996Efault:
1997 user_write_access_end();
1998 return -EFAULT;
1999}
2000#endif
2001
2002/*
2003 * This needs to be __function_aligned as GCC implicitly makes any
2004 * implementation of abort() cold and drops alignment specified by
2005 * -falign-functions=N.
2006 *
2007 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
2008 */
2009__weak __function_aligned void abort(void)
2010{
2011 BUG();
2012
2013 /* if that doesn't kill us, halt */
2014 panic("Oops failed to kill thread");
2015}
2016EXPORT_SYMBOL(abort);