개념 설명 전체
kernel/exit.c · v6.18.37
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/freezer.h>
29 #include <linux/binfmts.h>
30 #include <linux/nsproxy.h>
31 #include <linux/pid_namespace.h>
32 #include <linux/ptrace.h>
33 #include <linux/profile.h>
34 #include <linux/mount.h>
35 #include <linux/proc_fs.h>
36 #include <linux/kthread.h>
37 #include <linux/mempolicy.h>
38 #include <linux/taskstats_kern.h>
39 #include <linux/delayacct.h>
40 #include <linux/cgroup.h>
41 #include <linux/syscalls.h>
42 #include <linux/signal.h>
43 #include <linux/posix-timers.h>
44 #include <linux/cn_proc.h>
45 #include <linux/mutex.h>
46 #include <linux/futex.h>
47 #include <linux/pipe_fs_i.h>
48 #include <linux/audit.h> /* for audit_free() */
49 #include <linux/resource.h>
50 #include <linux/task_io_accounting_ops.h>
51 #include <linux/blkdev.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 */
87 static unsigned int oops_limit = 10000;
88
89 #ifdef CONFIG_SYSCTL
90 static 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
100 static __init int kernel_exit_sysctls_init(void)
101 {
102 register_sysctl_init("kernel", kern_exit_table);
103 return 0;
104 }
105 late_initcall(kernel_exit_sysctls_init);
106 #endif
107
108 static atomic_t oops_count = ATOMIC_INIT(0);
109
110 #ifdef CONFIG_SYSFS
111 static 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
117 static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
118
119 static __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 }
124 late_initcall(kernel_exit_sysfs_init);
125 #endif
126
127 /*
128 * For things release_task() would like to do *after* tasklist_lock is released.
129 */
130 struct release_task_post {
131 struct pid *pids[PIDTYPE_MAX];
132 };
133
134 static 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 */
159 static 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 tsk->sighand = NULL;
216 spin_unlock(&sighand->siglock);
217
218 __cleanup_sighand(sighand);
219 if (group_dead)
220 tty_kref_put(tty);
221 }
222
223 static void delayed_put_task_struct(struct rcu_head *rhp)
224 {
225 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
226
227 kprobe_flush_task(tsk);
228 rethook_flush_task(tsk);
229 perf_event_delayed_put(tsk);
230 trace_sched_process_free(tsk);
231 put_task_struct(tsk);
232 }
233
234 void put_task_struct_rcu_user(struct task_struct *task)
235 {
236 if (refcount_dec_and_test(&task->rcu_users))
237 call_rcu(&task->rcu, delayed_put_task_struct);
238 }
239
240 void __weak release_thread(struct task_struct *dead_task)
241 {
242 }
243
244 void release_task(struct task_struct *p)
245 {
246 struct release_task_post post;
247 struct task_struct *leader;
248 struct pid *thread_pid;
249 int zap_leader;
250 repeat:
251 memset(&post, 0, sizeof(post));
252
253 /* don't need to get the RCU readlock here - the process is dead and
254 * can't be modifying its own credentials. But shut RCU-lockdep up */
255 rcu_read_lock();
256 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
257 rcu_read_unlock();
258
259 pidfs_exit(p);
260 cgroup_release(p);
261
262 /* Retrieve @thread_pid before __unhash_process() may set it to NULL. */
263 thread_pid = task_pid(p);
264
265 write_lock_irq(&tasklist_lock);
266 ptrace_release_task(p);
267 __exit_signal(&post, p);
268
269 /*
270 * If we are the last non-leader member of the thread
271 * group, and the leader is zombie, then notify the
272 * group leader's parent process. (if it wants notification.)
273 */
274 zap_leader = 0;
275 leader = p->group_leader;
276 if (leader != p && thread_group_empty(leader)
277 && leader->exit_state == EXIT_ZOMBIE) {
278 /* for pidfs_exit() and do_notify_parent() */
279 if (leader->signal->flags & SIGNAL_GROUP_EXIT)
280 leader->exit_code = leader->signal->group_exit_code;
281 /*
282 * If we were the last child thread and the leader has
283 * exited already, and the leader's parent ignores SIGCHLD,
284 * then we are the one who should release the leader.
285 */
286 zap_leader = do_notify_parent(leader, leader->exit_signal);
287 if (zap_leader)
288 leader->exit_state = EXIT_DEAD;
289 }
290
291 write_unlock_irq(&tasklist_lock);
292 /* @thread_pid can't go away until free_pids() below */
293 proc_flush_pid(thread_pid);
294 add_device_randomness(&p->se.sum_exec_runtime,
295 sizeof(p->se.sum_exec_runtime));
296 free_pids(post.pids);
297 release_thread(p);
298 /*
299 * This task was already removed from the process/thread/pid lists
300 * and lock_task_sighand(p) can't succeed. Nobody else can touch
301 * ->pending or, if group dead, signal->shared_pending. We can call
302 * flush_sigqueue() lockless.
303 */
304 flush_sigqueue(&p->pending);
305 if (thread_group_leader(p))
306 flush_sigqueue(&p->signal->shared_pending);
307
308 put_task_struct_rcu_user(p);
309
310 p = leader;
311 if (unlikely(zap_leader))
312 goto repeat;
313 }
314
315 int rcuwait_wake_up(struct rcuwait *w)
316 {
317 int ret = 0;
318 struct task_struct *task;
319
320 rcu_read_lock();
321
322 /*
323 * Order condition vs @task, such that everything prior to the load
324 * of @task is visible. This is the condition as to why the user called
325 * rcuwait_wake() in the first place. Pairs with set_current_state()
326 * barrier (A) in rcuwait_wait_event().
327 *
328 * WAIT WAKE
329 * [S] tsk = current [S] cond = true
330 * MB (A) MB (B)
331 * [L] cond [L] tsk
332 */
333 smp_mb(); /* (B) */
334
335 task = rcu_dereference(w->task);
336 if (task)
337 ret = wake_up_process(task);
338 rcu_read_unlock();
339
340 return ret;
341 }
342 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
343
344 /*
345 * Determine if a process group is "orphaned", according to the POSIX
346 * definition in 2.2.2.52. Orphaned process groups are not to be affected
347 * by terminal-generated stop signals. Newly orphaned process groups are
348 * to receive a SIGHUP and a SIGCONT.
349 *
350 * "I ask you, have you ever known what it is to be an orphan?"
351 */
352 static int will_become_orphaned_pgrp(struct pid *pgrp,
353 struct task_struct *ignored_task)
354 {
355 struct task_struct *p;
356
357 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
358 if ((p == ignored_task) ||
359 (p->exit_state && thread_group_empty(p)) ||
360 is_global_init(p->real_parent))
361 continue;
362
363 if (task_pgrp(p->real_parent) != pgrp &&
364 task_session(p->real_parent) == task_session(p))
365 return 0;
366 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
367
368 return 1;
369 }
370
371 int is_current_pgrp_orphaned(void)
372 {
373 int retval;
374
375 read_lock(&tasklist_lock);
376 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
377 read_unlock(&tasklist_lock);
378
379 return retval;
380 }
381
382 static bool has_stopped_jobs(struct pid *pgrp)
383 {
384 struct task_struct *p;
385
386 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
387 if (p->signal->flags & SIGNAL_STOP_STOPPED)
388 return true;
389 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
390
391 return false;
392 }
393
394 /*
395 * Check to see if any process groups have become orphaned as
396 * a result of our exiting, and if they have any stopped jobs,
397 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
398 */
399 static void
400 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
401 {
402 struct pid *pgrp = task_pgrp(tsk);
403 struct task_struct *ignored_task = tsk;
404
405 if (!parent)
406 /* exit: our father is in a different pgrp than
407 * we are and we were the only connection outside.
408 */
409 parent = tsk->real_parent;
410 else
411 /* reparent: our child is in a different pgrp than
412 * we are, and it was the only connection outside.
413 */
414 ignored_task = NULL;
415
416 if (task_pgrp(parent) != pgrp &&
417 task_session(parent) == task_session(tsk) &&
418 will_become_orphaned_pgrp(pgrp, ignored_task) &&
419 has_stopped_jobs(pgrp)) {
420 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
421 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
422 }
423 }
424
425 static void coredump_task_exit(struct task_struct *tsk,
426 struct core_state *core_state)
427 {
428 struct core_thread self;
429
430 self.task = tsk;
431 if (self.task->flags & PF_SIGNALED)
432 self.next = xchg(&core_state->dumper.next, &self);
433 else
434 self.task = NULL;
435 /*
436 * Implies mb(), the result of xchg() must be visible
437 * to core_state->dumper.
438 */
439 if (atomic_dec_and_test(&core_state->nr_threads))
440 complete(&core_state->startup);
441
442 for (;;) {
443 set_current_state(TASK_IDLE|TASK_FREEZABLE);
444 if (!self.task) /* see coredump_finish() */
445 break;
446 schedule();
447 }
448 __set_current_state(TASK_RUNNING);
449 }
450
451 #ifdef CONFIG_MEMCG
452 /* drops tasklist_lock if succeeds */
453 static bool __try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm)
454 {
455 bool ret = false;
456
457 task_lock(tsk);
458 if (likely(tsk->mm == mm)) {
459 /* tsk can't pass exit_mm/exec_mmap and exit */
460 read_unlock(&tasklist_lock);
461 WRITE_ONCE(mm->owner, tsk);
462 lru_gen_migrate_mm(mm);
463 ret = true;
464 }
465 task_unlock(tsk);
466 return ret;
467 }
468
469 static bool try_to_set_owner(struct task_struct *g, struct mm_struct *mm)
470 {
471 struct task_struct *t;
472
473 for_each_thread(g, t) {
474 struct mm_struct *t_mm = READ_ONCE(t->mm);
475 if (t_mm == mm) {
476 if (__try_to_set_owner(t, mm))
477 return true;
478 } else if (t_mm)
479 break;
480 }
481
482 return false;
483 }
484
485 /*
486 * A task is exiting. If it owned this mm, find a new owner for the mm.
487 */
488 void mm_update_next_owner(struct mm_struct *mm)
489 {
490 struct task_struct *g, *p = current;
491
492 /*
493 * If the exiting or execing task is not the owner, it's
494 * someone else's problem.
495 */
496 if (mm->owner != p)
497 return;
498 /*
499 * The current owner is exiting/execing and there are no other
500 * candidates. Do not leave the mm pointing to a possibly
501 * freed task structure.
502 */
503 if (atomic_read(&mm->mm_users) <= 1) {
504 WRITE_ONCE(mm->owner, NULL);
505 return;
506 }
507
508 read_lock(&tasklist_lock);
509 /*
510 * Search in the children
511 */
512 list_for_each_entry(g, &p->children, sibling) {
513 if (try_to_set_owner(g, mm))
514 goto ret;
515 }
516 /*
517 * Search in the siblings
518 */
519 list_for_each_entry(g, &p->real_parent->children, sibling) {
520 if (try_to_set_owner(g, mm))
521 goto ret;
522 }
523 /*
524 * Search through everything else, we should not get here often.
525 */
526 for_each_process(g) {
527 if (atomic_read(&mm->mm_users) <= 1)
528 break;
529 if (g->flags & PF_KTHREAD)
530 continue;
531 if (try_to_set_owner(g, mm))
532 goto ret;
533 }
534 read_unlock(&tasklist_lock);
535 /*
536 * We found no owner yet mm_users > 1: this implies that we are
537 * most likely racing with swapoff (try_to_unuse()) or /proc or
538 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
539 */
540 WRITE_ONCE(mm->owner, NULL);
541 ret:
542 return;
543
544 }
545 #endif /* CONFIG_MEMCG */
546
547 /*
548 * Turn us into a lazy TLB process if we
549 * aren't already..
550 */
551 static void exit_mm(void)
552 {
553 struct mm_struct *mm = current->mm;
554
555 exit_mm_release(current, mm);
556 if (!mm)
557 return;
558 mmap_read_lock(mm);
559 mmgrab_lazy_tlb(mm);
560 BUG_ON(mm != current->active_mm);
561 /* more a memory barrier than a real lock */
562 task_lock(current);
563 /*
564 * When a thread stops operating on an address space, the loop
565 * in membarrier_private_expedited() may not observe that
566 * tsk->mm, and the loop in membarrier_global_expedited() may
567 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
568 * rq->membarrier_state, so those would not issue an IPI.
569 * Membarrier requires a memory barrier after accessing
570 * user-space memory, before clearing tsk->mm or the
571 * rq->membarrier_state.
572 */
573 smp_mb__after_spinlock();
574 local_irq_disable();
575 current->user_dumpable = (get_dumpable(mm) == SUID_DUMP_USER);
576 current->mm = NULL;
577 membarrier_update_current_mm(NULL);
578 enter_lazy_tlb(mm, current);
579 local_irq_enable();
580 task_unlock(current);
581 mmap_read_unlock(mm);
582 mm_update_next_owner(mm);
583 mmput(mm);
584 if (test_thread_flag(TIF_MEMDIE))
585 exit_oom_victim();
586 }
587
588 static struct task_struct *find_alive_thread(struct task_struct *p)
589 {
590 struct task_struct *t;
591
592 for_each_thread(p, t) {
593 if (!(t->flags & PF_EXITING))
594 return t;
595 }
596 return NULL;
597 }
598
599 static struct task_struct *find_child_reaper(struct task_struct *father,
600 struct list_head *dead)
601 __releases(&tasklist_lock)
602 __acquires(&tasklist_lock)
603 {
604 struct pid_namespace *pid_ns = task_active_pid_ns(father);
605 struct task_struct *reaper = pid_ns->child_reaper;
606 struct task_struct *p, *n;
607
608 if (likely(reaper != father))
609 return reaper;
610
611 reaper = find_alive_thread(father);
612 if (reaper) {
613 pid_ns->child_reaper = reaper;
614 return reaper;
615 }
616
617 write_unlock_irq(&tasklist_lock);
618
619 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
620 list_del_init(&p->ptrace_entry);
621 release_task(p);
622 }
623
624 zap_pid_ns_processes(pid_ns);
625 write_lock_irq(&tasklist_lock);
626
627 return father;
628 }
629
630 /*
631 * When we die, we re-parent all our children, and try to:
632 * 1. give them to another thread in our thread group, if such a member exists
633 * 2. give it to the first ancestor process which prctl'd itself as a
634 * child_subreaper for its children (like a service manager)
635 * 3. give it to the init process (PID 1) in our pid namespace
636 */
637 static struct task_struct *find_new_reaper(struct task_struct *father,
638 struct task_struct *child_reaper)
639 {
640 struct task_struct *thread, *reaper;
641
642 thread = find_alive_thread(father);
643 if (thread)
644 return thread;
645
646 if (father->signal->has_child_subreaper) {
647 unsigned int ns_level = task_pid(father)->level;
648 /*
649 * Find the first ->is_child_subreaper ancestor in our pid_ns.
650 * We can't check reaper != child_reaper to ensure we do not
651 * cross the namespaces, the exiting parent could be injected
652 * by setns() + fork().
653 * We check pid->level, this is slightly more efficient than
654 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
655 */
656 for (reaper = father->real_parent;
657 task_pid(reaper)->level == ns_level;
658 reaper = reaper->real_parent) {
659 if (reaper == &init_task)
660 break;
661 if (!reaper->signal->is_child_subreaper)
662 continue;
663 thread = find_alive_thread(reaper);
664 if (thread)
665 return thread;
666 }
667 }
668
669 return child_reaper;
670 }
671
672 /*
673 * Any that need to be release_task'd are put on the @dead list.
674 */
675 static void reparent_leader(struct task_struct *father, struct task_struct *p,
676 struct list_head *dead)
677 {
678 if (unlikely(p->exit_state == EXIT_DEAD))
679 return;
680
681 /* We don't want people slaying init. */
682 p->exit_signal = SIGCHLD;
683
684 /* If it has exited notify the new parent about this child's death. */
685 if (!p->ptrace &&
686 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
687 if (do_notify_parent(p, p->exit_signal)) {
688 p->exit_state = EXIT_DEAD;
689 list_add(&p->ptrace_entry, dead);
690 }
691 }
692
693 kill_orphaned_pgrp(p, father);
694 }
695
696 /*
697 * Make init inherit all the child processes
698 */
699 static void forget_original_parent(struct task_struct *father,
700 struct list_head *dead)
701 {
702 struct task_struct *p, *t, *reaper;
703
704 if (unlikely(!list_empty(&father->ptraced)))
705 exit_ptrace(father, dead);
706
707 /* Can drop and reacquire tasklist_lock */
708 reaper = find_child_reaper(father, dead);
709 if (list_empty(&father->children))
710 return;
711
712 reaper = find_new_reaper(father, reaper);
713 list_for_each_entry(p, &father->children, sibling) {
714 for_each_thread(p, t) {
715 RCU_INIT_POINTER(t->real_parent, reaper);
716 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
717 if (likely(!t->ptrace))
718 t->parent = t->real_parent;
719 if (t->pdeath_signal)
720 group_send_sig_info(t->pdeath_signal,
721 SEND_SIG_NOINFO, t,
722 PIDTYPE_TGID);
723 }
724 /*
725 * If this is a threaded reparent there is no need to
726 * notify anyone anything has happened.
727 */
728 if (!same_thread_group(reaper, father))
729 reparent_leader(father, p, dead);
730 }
731 list_splice_tail_init(&father->children, &reaper->children);
732 }
733
734 /*
735 * Send signals to all our closest relatives so that they know
736 * to properly mourn us..
737 */
738 static void exit_notify(struct task_struct *tsk, int group_dead)
739 {
740 bool autoreap;
741 struct task_struct *p, *n;
742 LIST_HEAD(dead);
743
744 write_lock_irq(&tasklist_lock);
745 forget_original_parent(tsk, &dead);
746
747 if (group_dead)
748 kill_orphaned_pgrp(tsk->group_leader, NULL);
749
750 tsk->exit_state = EXIT_ZOMBIE;
751
752 if (unlikely(tsk->ptrace)) {
753 int sig = thread_group_leader(tsk) &&
754 thread_group_empty(tsk) &&
755 !ptrace_reparented(tsk) ?
756 tsk->exit_signal : SIGCHLD;
757 autoreap = do_notify_parent(tsk, sig);
758 } else if (thread_group_leader(tsk)) {
759 autoreap = thread_group_empty(tsk) &&
760 do_notify_parent(tsk, tsk->exit_signal);
761 } else {
762 autoreap = true;
763 /* untraced sub-thread */
764 do_notify_pidfd(tsk);
765 }
766
767 if (autoreap) {
768 tsk->exit_state = EXIT_DEAD;
769 list_add(&tsk->ptrace_entry, &dead);
770 }
771
772 /* mt-exec, de_thread() is waiting for group leader */
773 if (unlikely(tsk->signal->notify_count < 0))
774 wake_up_process(tsk->signal->group_exec_task);
775 write_unlock_irq(&tasklist_lock);
776
777 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
778 list_del_init(&p->ptrace_entry);
779 release_task(p);
780 }
781 }
782
783 #ifdef CONFIG_DEBUG_STACK_USAGE
784 #ifdef CONFIG_STACK_GROWSUP
785 unsigned long stack_not_used(struct task_struct *p)
786 {
787 unsigned long *n = end_of_stack(p);
788
789 do { /* Skip over canary */
790 n--;
791 } while (!*n);
792
793 return (unsigned long)end_of_stack(p) - (unsigned long)n;
794 }
795 #else /* !CONFIG_STACK_GROWSUP */
796 unsigned long stack_not_used(struct task_struct *p)
797 {
798 unsigned long *n = end_of_stack(p);
799
800 do { /* Skip over canary */
801 n++;
802 } while (!*n);
803
804 return (unsigned long)n - (unsigned long)end_of_stack(p);
805 }
806 #endif /* CONFIG_STACK_GROWSUP */
807
808 /* Count the maximum pages reached in kernel stacks */
809 static inline void kstack_histogram(unsigned long used_stack)
810 {
811 #ifdef CONFIG_VM_EVENT_COUNTERS
812 if (used_stack <= 1024)
813 count_vm_event(KSTACK_1K);
814 #if THREAD_SIZE > 1024
815 else if (used_stack <= 2048)
816 count_vm_event(KSTACK_2K);
817 #endif
818 #if THREAD_SIZE > 2048
819 else if (used_stack <= 4096)
820 count_vm_event(KSTACK_4K);
821 #endif
822 #if THREAD_SIZE > 4096
823 else if (used_stack <= 8192)
824 count_vm_event(KSTACK_8K);
825 #endif
826 #if THREAD_SIZE > 8192
827 else if (used_stack <= 16384)
828 count_vm_event(KSTACK_16K);
829 #endif
830 #if THREAD_SIZE > 16384
831 else if (used_stack <= 32768)
832 count_vm_event(KSTACK_32K);
833 #endif
834 #if THREAD_SIZE > 32768
835 else if (used_stack <= 65536)
836 count_vm_event(KSTACK_64K);
837 #endif
838 #if THREAD_SIZE > 65536
839 else
840 count_vm_event(KSTACK_REST);
841 #endif
842 #endif /* CONFIG_VM_EVENT_COUNTERS */
843 }
844
845 static void check_stack_usage(void)
846 {
847 static DEFINE_SPINLOCK(low_water_lock);
848 static int lowest_to_date = THREAD_SIZE;
849 unsigned long free;
850
851 free = stack_not_used(current);
852 kstack_histogram(THREAD_SIZE - free);
853
854 if (free >= lowest_to_date)
855 return;
856
857 spin_lock(&low_water_lock);
858 if (free < lowest_to_date) {
859 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
860 current->comm, task_pid_nr(current), free);
861 lowest_to_date = free;
862 }
863 spin_unlock(&low_water_lock);
864 }
865 #else /* !CONFIG_DEBUG_STACK_USAGE */
866 static inline void check_stack_usage(void) {}
867 #endif /* CONFIG_DEBUG_STACK_USAGE */
868
869 static void synchronize_group_exit(struct task_struct *tsk, long code)
870 {
871 struct sighand_struct *sighand = tsk->sighand;
872 struct signal_struct *signal = tsk->signal;
873 struct core_state *core_state;
874
875 spin_lock_irq(&sighand->siglock);
876 signal->quick_threads--;
877 if ((signal->quick_threads == 0) &&
878 !(signal->flags & SIGNAL_GROUP_EXIT)) {
879 signal->flags = SIGNAL_GROUP_EXIT;
880 signal->group_exit_code = code;
881 signal->group_stop_count = 0;
882 }
883 /*
884 * Serialize with any possible pending coredump.
885 * We must hold siglock around checking core_state
886 * and setting PF_POSTCOREDUMP. The core-inducing thread
887 * will increment ->nr_threads for each thread in the
888 * group without PF_POSTCOREDUMP set.
889 */
890 tsk->flags |= PF_POSTCOREDUMP;
891 core_state = signal->core_state;
892 spin_unlock_irq(&sighand->siglock);
893
894 if (unlikely(core_state))
895 coredump_task_exit(tsk, core_state);
896 }
897
898 void __noreturn do_exit(long code)
899 {
900 struct task_struct *tsk = current;
901 struct kthread *kthread;
902 int group_dead;
903
904 WARN_ON(irqs_disabled());
905 WARN_ON(tsk->plug);
906
907 kthread = tsk_is_kthread(tsk);
908 if (unlikely(kthread))
909 kthread_do_exit(kthread, code);
910
911 kcov_task_exit(tsk);
912 kmsan_task_exit(tsk);
913
914 synchronize_group_exit(tsk, code);
915 ptrace_event(PTRACE_EVENT_EXIT, code);
916 user_events_exit(tsk);
917
918 io_uring_files_cancel();
919 exit_signals(tsk); /* sets PF_EXITING */
920
921 seccomp_filter_release(tsk);
922
923 acct_update_integrals(tsk);
924 group_dead = atomic_dec_and_test(&tsk->signal->live);
925 if (group_dead) {
926 /*
927 * If the last thread of global init has exited, panic
928 * immediately to get a useable coredump.
929 */
930 if (unlikely(is_global_init(tsk)))
931 panic("Attempted to kill init! exitcode=0x%08x\n",
932 tsk->signal->group_exit_code ?: (int)code);
933
934 #ifdef CONFIG_POSIX_TIMERS
935 hrtimer_cancel(&tsk->signal->real_timer);
936 exit_itimers(tsk);
937 #endif
938 if (tsk->mm)
939 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
940 }
941 acct_collect(code, group_dead);
942 if (group_dead)
943 tty_audit_exit();
944 audit_free(tsk);
945
946 tsk->exit_code = code;
947 taskstats_exit(tsk, group_dead);
948 unwind_deferred_task_exit(tsk);
949 trace_sched_process_exit(tsk, group_dead);
950
951 /*
952 * Since sampling can touch ->mm, make sure to stop everything before we
953 * tear it down.
954 *
955 * Also flushes inherited counters to the parent - before the parent
956 * gets woken up by child-exit notifications.
957 */
958 perf_event_exit_task(tsk);
959
960 exit_mm();
961
962 if (group_dead)
963 acct_process();
964
965 exit_sem(tsk);
966 exit_shm(tsk);
967 exit_files(tsk);
968 exit_fs(tsk);
969 if (group_dead)
970 disassociate_ctty(1);
971 exit_task_namespaces(tsk);
972 exit_task_work(tsk);
973 exit_thread(tsk);
974
975 sched_autogroup_exit_task(tsk);
976 cgroup_exit(tsk);
977
978 /*
979 * FIXME: do that only when needed, using sched_exit tracepoint
980 */
981 flush_ptrace_hw_breakpoint(tsk);
982
983 exit_tasks_rcu_start();
984 exit_notify(tsk, group_dead);
985 proc_exit_connector(tsk);
986 mpol_put_task_policy(tsk);
987 #ifdef CONFIG_FUTEX
988 if (unlikely(current->pi_state_cache))
989 kfree(current->pi_state_cache);
990 #endif
991 /*
992 * Make sure we are holding no locks:
993 */
994 debug_check_no_locks_held();
995
996 if (tsk->io_context)
997 exit_io_context(tsk);
998
999 if (tsk->splice_pipe)
1000 free_pipe_info(tsk->splice_pipe);
1001
1002 if (tsk->task_frag.page)
1003 put_page(tsk->task_frag.page);
1004
1005 exit_task_stack_account(tsk);
1006
1007 check_stack_usage();
1008 preempt_disable();
1009 if (tsk->nr_dirtied)
1010 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
1011 exit_rcu();
1012 exit_tasks_rcu_finish();
1013
1014 lockdep_free_task(tsk);
1015 do_task_dead();
1016 }
1017 EXPORT_SYMBOL(do_exit);
1018
1019 void __noreturn make_task_dead(int signr)
1020 {
1021 /*
1022 * Take the task off the cpu after something catastrophic has
1023 * happened.
1024 *
1025 * We can get here from a kernel oops, sometimes with preemption off.
1026 * Start by checking for critical errors.
1027 * Then fix up important state like USER_DS and preemption.
1028 * Then do everything else.
1029 */
1030 struct task_struct *tsk = current;
1031 unsigned int limit;
1032
1033 if (unlikely(in_interrupt()))
1034 panic("Aiee, killing interrupt handler!");
1035 if (unlikely(!tsk->pid))
1036 panic("Attempted to kill the idle task!");
1037
1038 if (unlikely(irqs_disabled())) {
1039 pr_info("note: %s[%d] exited with irqs disabled\n",
1040 current->comm, task_pid_nr(current));
1041 local_irq_enable();
1042 }
1043 if (unlikely(in_atomic())) {
1044 pr_info("note: %s[%d] exited with preempt_count %d\n",
1045 current->comm, task_pid_nr(current),
1046 preempt_count());
1047 preempt_count_set(PREEMPT_ENABLED);
1048 }
1049
1050 /*
1051 * Every time the system oopses, if the oops happens while a reference
1052 * to an object was held, the reference leaks.
1053 * If the oops doesn't also leak memory, repeated oopsing can cause
1054 * reference counters to wrap around (if they're not using refcount_t).
1055 * This means that repeated oopsing can make unexploitable-looking bugs
1056 * exploitable through repeated oopsing.
1057 * To make sure this can't happen, place an upper bound on how often the
1058 * kernel may oops without panic().
1059 */
1060 limit = READ_ONCE(oops_limit);
1061 if (atomic_inc_return(&oops_count) >= limit && limit)
1062 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
1063
1064 /*
1065 * We're taking recursive faults here in make_task_dead. Safest is to just
1066 * leave this task alone and wait for reboot.
1067 */
1068 if (unlikely(tsk->flags & PF_EXITING)) {
1069 pr_alert("Fixing recursive fault but reboot is needed!\n");
1070 futex_exit_recursive(tsk);
1071 tsk->exit_state = EXIT_DEAD;
1072 refcount_inc(&tsk->rcu_users);
1073 preempt_disable();
1074 do_task_dead();
1075 }
1076
1077 do_exit(signr);
1078 }
1079
1080 SYSCALL_DEFINE1(exit, int, error_code)
1081 {
1082 do_exit((error_code&0xff)<<8);
1083 }
1084
1085 /*
1086 * Take down every thread in the group. This is called by fatal signals
1087 * as well as by sys_exit_group (below).
1088 */
1089 void __noreturn
1090 do_group_exit(int exit_code)
1091 {
1092 struct signal_struct *sig = current->signal;
1093
1094 if (sig->flags & SIGNAL_GROUP_EXIT)
1095 exit_code = sig->group_exit_code;
1096 else if (sig->group_exec_task)
1097 exit_code = 0;
1098 else {
1099 struct sighand_struct *const sighand = current->sighand;
1100
1101 spin_lock_irq(&sighand->siglock);
1102 if (sig->flags & SIGNAL_GROUP_EXIT)
1103 /* Another thread got here before we took the lock. */
1104 exit_code = sig->group_exit_code;
1105 else if (sig->group_exec_task)
1106 exit_code = 0;
1107 else {
1108 sig->group_exit_code = exit_code;
1109 sig->flags = SIGNAL_GROUP_EXIT;
1110 zap_other_threads(current);
1111 }
1112 spin_unlock_irq(&sighand->siglock);
1113 }
1114
1115 do_exit(exit_code);
1116 /* NOTREACHED */
1117 }
1118
1119 /*
1120 * this kills every thread in the thread group. Note that any externally
1121 * wait4()-ing process will get the correct exit code - even if this
1122 * thread is not the thread group leader.
1123 */
1124 SYSCALL_DEFINE1(exit_group, int, error_code)
1125 {
1126 do_group_exit((error_code & 0xff) << 8);
1127 /* NOTREACHED */
1128 return 0;
1129 }
1130
1131 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1132 {
1133 return wo->wo_type == PIDTYPE_MAX ||
1134 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1135 }
1136
1137 static int
1138 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1139 {
1140 if (!eligible_pid(wo, p))
1141 return 0;
1142
1143 /*
1144 * Wait for all children (clone and not) if __WALL is set or
1145 * if it is traced by us.
1146 */
1147 if (ptrace || (wo->wo_flags & __WALL))
1148 return 1;
1149
1150 /*
1151 * Otherwise, wait for clone children *only* if __WCLONE is set;
1152 * otherwise, wait for non-clone children *only*.
1153 *
1154 * Note: a "clone" child here is one that reports to its parent
1155 * using a signal other than SIGCHLD, or a non-leader thread which
1156 * we can only see if it is traced by us.
1157 */
1158 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1159 return 0;
1160
1161 return 1;
1162 }
1163
1164 /*
1165 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1166 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1167 * the lock and this task is uninteresting. If we return nonzero, we have
1168 * released the lock and the system call should return.
1169 */
1170 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1171 {
1172 int state, status;
1173 pid_t pid = task_pid_vnr(p);
1174 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1175 struct waitid_info *infop;
1176
1177 if (!likely(wo->wo_flags & WEXITED))
1178 return 0;
1179
1180 if (unlikely(wo->wo_flags & WNOWAIT)) {
1181 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1182 ? p->signal->group_exit_code : p->exit_code;
1183 get_task_struct(p);
1184 read_unlock(&tasklist_lock);
1185 sched_annotate_sleep();
1186 if (wo->wo_rusage)
1187 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1188 put_task_struct(p);
1189 goto out_info;
1190 }
1191 /*
1192 * Move the task's state to DEAD/TRACE, only one thread can do this.
1193 */
1194 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1195 EXIT_TRACE : EXIT_DEAD;
1196 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1197 return 0;
1198 /*
1199 * We own this thread, nobody else can reap it.
1200 */
1201 read_unlock(&tasklist_lock);
1202 sched_annotate_sleep();
1203
1204 /*
1205 * Check thread_group_leader() to exclude the traced sub-threads.
1206 */
1207 if (state == EXIT_DEAD && thread_group_leader(p)) {
1208 struct signal_struct *sig = p->signal;
1209 struct signal_struct *psig = current->signal;
1210 unsigned long maxrss;
1211 u64 tgutime, tgstime;
1212
1213 /*
1214 * The resource counters for the group leader are in its
1215 * own task_struct. Those for dead threads in the group
1216 * are in its signal_struct, as are those for the child
1217 * processes it has previously reaped. All these
1218 * accumulate in the parent's signal_struct c* fields.
1219 *
1220 * We don't bother to take a lock here to protect these
1221 * p->signal fields because the whole thread group is dead
1222 * and nobody can change them.
1223 *
1224 * psig->stats_lock also protects us from our sub-threads
1225 * which can reap other children at the same time.
1226 *
1227 * We use thread_group_cputime_adjusted() to get times for
1228 * the thread group, which consolidates times for all threads
1229 * in the group including the group leader.
1230 */
1231 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1232 write_seqlock_irq(&psig->stats_lock);
1233 psig->cutime += tgutime + sig->cutime;
1234 psig->cstime += tgstime + sig->cstime;
1235 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1236 psig->cmin_flt +=
1237 p->min_flt + sig->min_flt + sig->cmin_flt;
1238 psig->cmaj_flt +=
1239 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1240 psig->cnvcsw +=
1241 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1242 psig->cnivcsw +=
1243 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1244 psig->cinblock +=
1245 task_io_get_inblock(p) +
1246 sig->inblock + sig->cinblock;
1247 psig->coublock +=
1248 task_io_get_oublock(p) +
1249 sig->oublock + sig->coublock;
1250 maxrss = max(sig->maxrss, sig->cmaxrss);
1251 if (psig->cmaxrss < maxrss)
1252 psig->cmaxrss = maxrss;
1253 task_io_accounting_add(&psig->ioac, &p->ioac);
1254 task_io_accounting_add(&psig->ioac, &sig->ioac);
1255 write_sequnlock_irq(&psig->stats_lock);
1256 }
1257
1258 if (wo->wo_rusage)
1259 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1260 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1261 ? p->signal->group_exit_code : p->exit_code;
1262 wo->wo_stat = status;
1263
1264 if (state == EXIT_TRACE) {
1265 write_lock_irq(&tasklist_lock);
1266 /* We dropped tasklist, ptracer could die and untrace */
1267 ptrace_unlink(p);
1268
1269 /* If parent wants a zombie, don't release it now */
1270 state = EXIT_ZOMBIE;
1271 if (do_notify_parent(p, p->exit_signal))
1272 state = EXIT_DEAD;
1273 p->exit_state = state;
1274 write_unlock_irq(&tasklist_lock);
1275 }
1276 if (state == EXIT_DEAD)
1277 release_task(p);
1278
1279 out_info:
1280 infop = wo->wo_info;
1281 if (infop) {
1282 if ((status & 0x7f) == 0) {
1283 infop->cause = CLD_EXITED;
1284 infop->status = status >> 8;
1285 } else {
1286 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1287 infop->status = status & 0x7f;
1288 }
1289 infop->pid = pid;
1290 infop->uid = uid;
1291 }
1292
1293 return pid;
1294 }
1295
1296 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1297 {
1298 if (ptrace) {
1299 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1300 return &p->exit_code;
1301 } else {
1302 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1303 return &p->signal->group_exit_code;
1304 }
1305 return NULL;
1306 }
1307
1308 /**
1309 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1310 * @wo: wait options
1311 * @ptrace: is the wait for ptrace
1312 * @p: task to wait for
1313 *
1314 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1315 *
1316 * CONTEXT:
1317 * read_lock(&tasklist_lock), which is released if return value is
1318 * non-zero. Also, grabs and releases @p->sighand->siglock.
1319 *
1320 * RETURNS:
1321 * 0 if wait condition didn't exist and search for other wait conditions
1322 * should continue. Non-zero return, -errno on failure and @p's pid on
1323 * success, implies that tasklist_lock is released and wait condition
1324 * search should terminate.
1325 */
1326 static int wait_task_stopped(struct wait_opts *wo,
1327 int ptrace, struct task_struct *p)
1328 {
1329 struct waitid_info *infop;
1330 int exit_code, *p_code, why;
1331 uid_t uid = 0; /* unneeded, required by compiler */
1332 pid_t pid;
1333
1334 /*
1335 * Traditionally we see ptrace'd stopped tasks regardless of options.
1336 */
1337 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1338 return 0;
1339
1340 if (!task_stopped_code(p, ptrace))
1341 return 0;
1342
1343 exit_code = 0;
1344 spin_lock_irq(&p->sighand->siglock);
1345
1346 p_code = task_stopped_code(p, ptrace);
1347 if (unlikely(!p_code))
1348 goto unlock_sig;
1349
1350 exit_code = *p_code;
1351 if (!exit_code)
1352 goto unlock_sig;
1353
1354 if (!unlikely(wo->wo_flags & WNOWAIT))
1355 *p_code = 0;
1356
1357 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1358 unlock_sig:
1359 spin_unlock_irq(&p->sighand->siglock);
1360 if (!exit_code)
1361 return 0;
1362
1363 /*
1364 * Now we are pretty sure this task is interesting.
1365 * Make sure it doesn't get reaped out from under us while we
1366 * give up the lock and then examine it below. We don't want to
1367 * keep holding onto the tasklist_lock while we call getrusage and
1368 * possibly take page faults for user memory.
1369 */
1370 get_task_struct(p);
1371 pid = task_pid_vnr(p);
1372 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1373 read_unlock(&tasklist_lock);
1374 sched_annotate_sleep();
1375 if (wo->wo_rusage)
1376 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1377 put_task_struct(p);
1378
1379 if (likely(!(wo->wo_flags & WNOWAIT)))
1380 wo->wo_stat = (exit_code << 8) | 0x7f;
1381
1382 infop = wo->wo_info;
1383 if (infop) {
1384 infop->cause = why;
1385 infop->status = exit_code;
1386 infop->pid = pid;
1387 infop->uid = uid;
1388 }
1389 return pid;
1390 }
1391
1392 /*
1393 * Handle do_wait work for one task in a live, non-stopped state.
1394 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1395 * the lock and this task is uninteresting. If we return nonzero, we have
1396 * released the lock and the system call should return.
1397 */
1398 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1399 {
1400 struct waitid_info *infop;
1401 pid_t pid;
1402 uid_t uid;
1403
1404 if (!unlikely(wo->wo_flags & WCONTINUED))
1405 return 0;
1406
1407 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1408 return 0;
1409
1410 spin_lock_irq(&p->sighand->siglock);
1411 /* Re-check with the lock held. */
1412 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1413 spin_unlock_irq(&p->sighand->siglock);
1414 return 0;
1415 }
1416 if (!unlikely(wo->wo_flags & WNOWAIT))
1417 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1418 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1419 spin_unlock_irq(&p->sighand->siglock);
1420
1421 pid = task_pid_vnr(p);
1422 get_task_struct(p);
1423 read_unlock(&tasklist_lock);
1424 sched_annotate_sleep();
1425 if (wo->wo_rusage)
1426 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1427 put_task_struct(p);
1428
1429 infop = wo->wo_info;
1430 if (!infop) {
1431 wo->wo_stat = 0xffff;
1432 } else {
1433 infop->cause = CLD_CONTINUED;
1434 infop->pid = pid;
1435 infop->uid = uid;
1436 infop->status = SIGCONT;
1437 }
1438 return pid;
1439 }
1440
1441 /*
1442 * Consider @p for a wait by @parent.
1443 *
1444 * -ECHILD should be in ->notask_error before the first call.
1445 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1446 * Returns zero if the search for a child should continue;
1447 * then ->notask_error is 0 if @p is an eligible child,
1448 * or still -ECHILD.
1449 */
1450 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1451 struct task_struct *p)
1452 {
1453 /*
1454 * We can race with wait_task_zombie() from another thread.
1455 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1456 * can't confuse the checks below.
1457 */
1458 int exit_state = READ_ONCE(p->exit_state);
1459 int ret;
1460
1461 if (unlikely(exit_state == EXIT_DEAD))
1462 return 0;
1463
1464 ret = eligible_child(wo, ptrace, p);
1465 if (!ret)
1466 return ret;
1467
1468 if (unlikely(exit_state == EXIT_TRACE)) {
1469 /*
1470 * ptrace == 0 means we are the natural parent. In this case
1471 * we should clear notask_error, debugger will notify us.
1472 */
1473 if (likely(!ptrace))
1474 wo->notask_error = 0;
1475 return 0;
1476 }
1477
1478 if (likely(!ptrace) && unlikely(p->ptrace)) {
1479 /*
1480 * If it is traced by its real parent's group, just pretend
1481 * the caller is ptrace_do_wait() and reap this child if it
1482 * is zombie.
1483 *
1484 * This also hides group stop state from real parent; otherwise
1485 * a single stop can be reported twice as group and ptrace stop.
1486 * If a ptracer wants to distinguish these two events for its
1487 * own children it should create a separate process which takes
1488 * the role of real parent.
1489 */
1490 if (!ptrace_reparented(p))
1491 ptrace = 1;
1492 }
1493
1494 /* slay zombie? */
1495 if (exit_state == EXIT_ZOMBIE) {
1496 /* we don't reap group leaders with subthreads */
1497 if (!delay_group_leader(p)) {
1498 /*
1499 * A zombie ptracee is only visible to its ptracer.
1500 * Notification and reaping will be cascaded to the
1501 * real parent when the ptracer detaches.
1502 */
1503 if (unlikely(ptrace) || likely(!p->ptrace))
1504 return wait_task_zombie(wo, p);
1505 }
1506
1507 /*
1508 * Allow access to stopped/continued state via zombie by
1509 * falling through. Clearing of notask_error is complex.
1510 *
1511 * When !@ptrace:
1512 *
1513 * If WEXITED is set, notask_error should naturally be
1514 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1515 * so, if there are live subthreads, there are events to
1516 * wait for. If all subthreads are dead, it's still safe
1517 * to clear - this function will be called again in finite
1518 * amount time once all the subthreads are released and
1519 * will then return without clearing.
1520 *
1521 * When @ptrace:
1522 *
1523 * Stopped state is per-task and thus can't change once the
1524 * target task dies. Only continued and exited can happen.
1525 * Clear notask_error if WCONTINUED | WEXITED.
1526 */
1527 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1528 wo->notask_error = 0;
1529 } else {
1530 /*
1531 * @p is alive and it's gonna stop, continue or exit, so
1532 * there always is something to wait for.
1533 */
1534 wo->notask_error = 0;
1535 }
1536
1537 /*
1538 * Wait for stopped. Depending on @ptrace, different stopped state
1539 * is used and the two don't interact with each other.
1540 */
1541 ret = wait_task_stopped(wo, ptrace, p);
1542 if (ret)
1543 return ret;
1544
1545 /*
1546 * Wait for continued. There's only one continued state and the
1547 * ptracer can consume it which can confuse the real parent. Don't
1548 * use WCONTINUED from ptracer. You don't need or want it.
1549 */
1550 return wait_task_continued(wo, p);
1551 }
1552
1553 /*
1554 * Do the work of do_wait() for one thread in the group, @tsk.
1555 *
1556 * -ECHILD should be in ->notask_error before the first call.
1557 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1558 * Returns zero if the search for a child should continue; then
1559 * ->notask_error is 0 if there were any eligible children,
1560 * or still -ECHILD.
1561 */
1562 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1563 {
1564 struct task_struct *p;
1565
1566 list_for_each_entry(p, &tsk->children, sibling) {
1567 int ret = wait_consider_task(wo, 0, p);
1568
1569 if (ret)
1570 return ret;
1571 }
1572
1573 return 0;
1574 }
1575
1576 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1577 {
1578 struct task_struct *p;
1579
1580 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1581 int ret = wait_consider_task(wo, 1, p);
1582
1583 if (ret)
1584 return ret;
1585 }
1586
1587 return 0;
1588 }
1589
1590 bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1591 {
1592 if (!eligible_pid(wo, p))
1593 return false;
1594
1595 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1596 return false;
1597
1598 return true;
1599 }
1600
1601 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1602 int sync, void *key)
1603 {
1604 struct wait_opts *wo = container_of(wait, struct wait_opts,
1605 child_wait);
1606 struct task_struct *p = key;
1607
1608 if (pid_child_should_wake(wo, p))
1609 return default_wake_function(wait, mode, sync, key);
1610
1611 return 0;
1612 }
1613
1614 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1615 {
1616 __wake_up_sync_key(&parent->signal->wait_chldexit,
1617 TASK_INTERRUPTIBLE, p);
1618 }
1619
1620 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1621 struct task_struct *target)
1622 {
1623 struct task_struct *parent =
1624 !ptrace ? target->real_parent : target->parent;
1625
1626 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1627 same_thread_group(current, parent));
1628 }
1629
1630 /*
1631 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1632 * and tracee lists to find the target task.
1633 */
1634 static int do_wait_pid(struct wait_opts *wo)
1635 {
1636 bool ptrace;
1637 struct task_struct *target;
1638 int retval;
1639
1640 ptrace = false;
1641 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1642 if (target && is_effectively_child(wo, ptrace, target)) {
1643 retval = wait_consider_task(wo, ptrace, target);
1644 if (retval)
1645 return retval;
1646 }
1647
1648 ptrace = true;
1649 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1650 if (target && target->ptrace &&
1651 is_effectively_child(wo, ptrace, target)) {
1652 retval = wait_consider_task(wo, ptrace, target);
1653 if (retval)
1654 return retval;
1655 }
1656
1657 return 0;
1658 }
1659
1660 long __do_wait(struct wait_opts *wo)
1661 {
1662 long retval;
1663
1664 /*
1665 * If there is nothing that can match our criteria, just get out.
1666 * We will clear ->notask_error to zero if we see any child that
1667 * might later match our criteria, even if we are not able to reap
1668 * it yet.
1669 */
1670 wo->notask_error = -ECHILD;
1671 if ((wo->wo_type < PIDTYPE_MAX) &&
1672 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1673 goto notask;
1674
1675 read_lock(&tasklist_lock);
1676
1677 if (wo->wo_type == PIDTYPE_PID) {
1678 retval = do_wait_pid(wo);
1679 if (retval)
1680 return retval;
1681 } else {
1682 struct task_struct *tsk = current;
1683
1684 do {
1685 retval = do_wait_thread(wo, tsk);
1686 if (retval)
1687 return retval;
1688
1689 retval = ptrace_do_wait(wo, tsk);
1690 if (retval)
1691 return retval;
1692
1693 if (wo->wo_flags & __WNOTHREAD)
1694 break;
1695 } while_each_thread(current, tsk);
1696 }
1697 read_unlock(&tasklist_lock);
1698
1699 notask:
1700 retval = wo->notask_error;
1701 if (!retval && !(wo->wo_flags & WNOHANG))
1702 return -ERESTARTSYS;
1703
1704 return retval;
1705 }
1706
1707 static long do_wait(struct wait_opts *wo)
1708 {
1709 int retval;
1710
1711 trace_sched_process_wait(wo->wo_pid);
1712
1713 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1714 wo->child_wait.private = current;
1715 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1716
1717 do {
1718 set_current_state(TASK_INTERRUPTIBLE);
1719 retval = __do_wait(wo);
1720 if (retval != -ERESTARTSYS)
1721 break;
1722 if (signal_pending(current))
1723 break;
1724 schedule();
1725 } while (1);
1726
1727 __set_current_state(TASK_RUNNING);
1728 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1729 return retval;
1730 }
1731
1732 int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1733 struct waitid_info *infop, int options,
1734 struct rusage *ru)
1735 {
1736 unsigned int f_flags = 0;
1737 struct pid *pid = NULL;
1738 enum pid_type type;
1739
1740 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1741 __WNOTHREAD|__WCLONE|__WALL))
1742 return -EINVAL;
1743 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1744 return -EINVAL;
1745
1746 switch (which) {
1747 case P_ALL:
1748 type = PIDTYPE_MAX;
1749 break;
1750 case P_PID:
1751 type = PIDTYPE_PID;
1752 if (upid <= 0)
1753 return -EINVAL;
1754
1755 pid = find_get_pid(upid);
1756 break;
1757 case P_PGID:
1758 type = PIDTYPE_PGID;
1759 if (upid < 0)
1760 return -EINVAL;
1761
1762 if (upid)
1763 pid = find_get_pid(upid);
1764 else
1765 pid = get_task_pid(current, PIDTYPE_PGID);
1766 break;
1767 case P_PIDFD:
1768 type = PIDTYPE_PID;
1769 if (upid < 0)
1770 return -EINVAL;
1771
1772 pid = pidfd_get_pid(upid, &f_flags);
1773 if (IS_ERR(pid))
1774 return PTR_ERR(pid);
1775
1776 break;
1777 default:
1778 return -EINVAL;
1779 }
1780
1781 wo->wo_type = type;
1782 wo->wo_pid = pid;
1783 wo->wo_flags = options;
1784 wo->wo_info = infop;
1785 wo->wo_rusage = ru;
1786 if (f_flags & O_NONBLOCK)
1787 wo->wo_flags |= WNOHANG;
1788
1789 return 0;
1790 }
1791
1792 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1793 int options, struct rusage *ru)
1794 {
1795 struct wait_opts wo;
1796 long ret;
1797
1798 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1799 if (ret)
1800 return ret;
1801
1802 ret = do_wait(&wo);
1803 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1804 ret = -EAGAIN;
1805
1806 put_pid(wo.wo_pid);
1807 return ret;
1808 }
1809
1810 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1811 infop, int, options, struct rusage __user *, ru)
1812 {
1813 struct rusage r;
1814 struct waitid_info info = {.status = 0};
1815 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1816 int signo = 0;
1817
1818 if (err > 0) {
1819 signo = SIGCHLD;
1820 err = 0;
1821 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1822 return -EFAULT;
1823 }
1824 if (!infop)
1825 return err;
1826
1827 if (!user_write_access_begin(infop, sizeof(*infop)))
1828 return -EFAULT;
1829
1830 unsafe_put_user(signo, &infop->si_signo, Efault);
1831 unsafe_put_user(0, &infop->si_errno, Efault);
1832 unsafe_put_user(info.cause, &infop->si_code, Efault);
1833 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1834 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1835 unsafe_put_user(info.status, &infop->si_status, Efault);
1836 user_write_access_end();
1837 return err;
1838 Efault:
1839 user_write_access_end();
1840 return -EFAULT;
1841 }
1842
1843 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1844 struct rusage *ru)
1845 {
1846 struct wait_opts wo;
1847 struct pid *pid = NULL;
1848 enum pid_type type;
1849 long ret;
1850
1851 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1852 __WNOTHREAD|__WCLONE|__WALL))
1853 return -EINVAL;
1854
1855 /* -INT_MIN is not defined */
1856 if (upid == INT_MIN)
1857 return -ESRCH;
1858
1859 if (upid == -1)
1860 type = PIDTYPE_MAX;
1861 else if (upid < 0) {
1862 type = PIDTYPE_PGID;
1863 pid = find_get_pid(-upid);
1864 } else if (upid == 0) {
1865 type = PIDTYPE_PGID;
1866 pid = get_task_pid(current, PIDTYPE_PGID);
1867 } else /* upid > 0 */ {
1868 type = PIDTYPE_PID;
1869 pid = find_get_pid(upid);
1870 }
1871
1872 wo.wo_type = type;
1873 wo.wo_pid = pid;
1874 wo.wo_flags = options | WEXITED;
1875 wo.wo_info = NULL;
1876 wo.wo_stat = 0;
1877 wo.wo_rusage = ru;
1878 ret = do_wait(&wo);
1879 put_pid(pid);
1880 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1881 ret = -EFAULT;
1882
1883 return ret;
1884 }
1885
1886 int kernel_wait(pid_t pid, int *stat)
1887 {
1888 struct wait_opts wo = {
1889 .wo_type = PIDTYPE_PID,
1890 .wo_pid = find_get_pid(pid),
1891 .wo_flags = WEXITED,
1892 };
1893 int ret;
1894
1895 ret = do_wait(&wo);
1896 if (ret > 0 && wo.wo_stat)
1897 *stat = wo.wo_stat;
1898 put_pid(wo.wo_pid);
1899 return ret;
1900 }
1901
1902 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1903 int, options, struct rusage __user *, ru)
1904 {
1905 struct rusage r;
1906 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1907
1908 if (err > 0) {
1909 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1910 return -EFAULT;
1911 }
1912 return err;
1913 }
1914
1915 #ifdef __ARCH_WANT_SYS_WAITPID
1916
1917 /*
1918 * sys_waitpid() remains for compatibility. waitpid() should be
1919 * implemented by calling sys_wait4() from libc.a.
1920 */
1921 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1922 {
1923 return kernel_wait4(pid, stat_addr, options, NULL);
1924 }
1925
1926 #endif
1927
1928 #ifdef CONFIG_COMPAT
1929 COMPAT_SYSCALL_DEFINE4(wait4,
1930 compat_pid_t, pid,
1931 compat_uint_t __user *, stat_addr,
1932 int, options,
1933 struct compat_rusage __user *, ru)
1934 {
1935 struct rusage r;
1936 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1937 if (err > 0) {
1938 if (ru && put_compat_rusage(&r, ru))
1939 return -EFAULT;
1940 }
1941 return err;
1942 }
1943
1944 COMPAT_SYSCALL_DEFINE5(waitid,
1945 int, which, compat_pid_t, pid,
1946 struct compat_siginfo __user *, infop, int, options,
1947 struct compat_rusage __user *, uru)
1948 {
1949 struct rusage ru;
1950 struct waitid_info info = {.status = 0};
1951 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1952 int signo = 0;
1953 if (err > 0) {
1954 signo = SIGCHLD;
1955 err = 0;
1956 if (uru) {
1957 /* kernel_waitid() overwrites everything in ru */
1958 if (COMPAT_USE_64BIT_TIME)
1959 err = copy_to_user(uru, &ru, sizeof(ru));
1960 else
1961 err = put_compat_rusage(&ru, uru);
1962 if (err)
1963 return -EFAULT;
1964 }
1965 }
1966
1967 if (!infop)
1968 return err;
1969
1970 if (!user_write_access_begin(infop, sizeof(*infop)))
1971 return -EFAULT;
1972
1973 unsafe_put_user(signo, &infop->si_signo, Efault);
1974 unsafe_put_user(0, &infop->si_errno, Efault);
1975 unsafe_put_user(info.cause, &infop->si_code, Efault);
1976 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1977 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1978 unsafe_put_user(info.status, &infop->si_status, Efault);
1979 user_write_access_end();
1980 return err;
1981 Efault:
1982 user_write_access_end();
1983 return -EFAULT;
1984 }
1985 #endif
1986
1987 /*
1988 * This needs to be __function_aligned as GCC implicitly makes any
1989 * implementation of abort() cold and drops alignment specified by
1990 * -falign-functions=N.
1991 *
1992 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1993 */
1994 __weak __function_aligned void abort(void)
1995 {
1996 BUG();
1997
1998 /* if that doesn't kill us, halt */
1999 panic("Oops failed to kill thread");
2000 }
2001 EXPORT_SYMBOL(abort);