projects/linux·blob v7.0 · refs

2004 lines · 49.7 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/exit.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/mm.h>
9#include <linux/slab.h>
10#include <linux/sched/autogroup.h>
11#include <linux/sched/mm.h>
12#include <linux/sched/stat.h>
13#include <linux/sched/task.h>
14#include <linux/sched/task_stack.h>
15#include <linux/sched/cputime.h>
16#include <linux/interrupt.h>
17#include <linux/module.h>
18#include <linux/capability.h>
19#include <linux/completion.h>
20#include <linux/personality.h>
21#include <linux/tty.h>
22#include <linux/iocontext.h>
23#include <linux/key.h>
24#include <linux/cpu.h>
25#include <linux/acct.h>
26#include <linux/tsacct_kern.h>
27#include <linux/file.h>
28#include <linux/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 */
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 tsk->sighand = NULL;
216 spin_unlock(&sighand->siglock);
217
218 __cleanup_sighand(sighand);
219 if (group_dead)
220 tty_kref_put(tty);
221}
222
223static 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
234void 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
240void __weak release_thread(struct task_struct *dead_task)
241{
242}
243
244void 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;
250repeat:
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. */
255 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
256
257 pidfs_exit(p);
258 cgroup_task_release(p);
259
260 /* Retrieve @thread_pid before __unhash_process() may set it to NULL. */
261 thread_pid = task_pid(p);
262
263 write_lock_irq(&tasklist_lock);
264 ptrace_release_task(p);
265 __exit_signal(&post, p);
266
267 /*
268 * If we are the last non-leader member of the thread
269 * group, and the leader is zombie, then notify the
270 * group leader's parent process. (if it wants notification.)
271 */
272 zap_leader = 0;
273 leader = p->group_leader;
274 if (leader != p && thread_group_empty(leader)
275 && leader->exit_state == EXIT_ZOMBIE) {
276 /* for pidfs_exit() and do_notify_parent() */
277 if (leader->signal->flags & SIGNAL_GROUP_EXIT)
278 leader->exit_code = leader->signal->group_exit_code;
279 /*
280 * If we were the last child thread and the leader has
281 * exited already, and the leader's parent ignores SIGCHLD,
282 * then we are the one who should release the leader.
283 */
284 zap_leader = do_notify_parent(leader, leader->exit_signal);
285 if (zap_leader)
286 leader->exit_state = EXIT_DEAD;
287 }
288
289 write_unlock_irq(&tasklist_lock);
290 /* @thread_pid can't go away until free_pids() below */
291 proc_flush_pid(thread_pid);
292 exit_cred_namespaces(p);
293 add_device_randomness(&p->se.sum_exec_runtime,
294 sizeof(p->se.sum_exec_runtime));
295 free_pids(post.pids);
296 release_thread(p);
297 /*
298 * This task was already removed from the process/thread/pid lists
299 * and lock_task_sighand(p) can't succeed. Nobody else can touch
300 * ->pending or, if group dead, signal->shared_pending. We can call
301 * flush_sigqueue() lockless.
302 */
303 flush_sigqueue(&p->pending);
304 if (thread_group_leader(p))
305 flush_sigqueue(&p->signal->shared_pending);
306
307 put_task_struct_rcu_user(p);
308
309 p = leader;
310 if (unlikely(zap_leader))
311 goto repeat;
312}
313
314int rcuwait_wake_up(struct rcuwait *w)
315{
316 int ret = 0;
317 struct task_struct *task;
318
319 rcu_read_lock();
320
321 /*
322 * Order condition vs @task, such that everything prior to the load
323 * of @task is visible. This is the condition as to why the user called
324 * rcuwait_wake() in the first place. Pairs with set_current_state()
325 * barrier (A) in rcuwait_wait_event().
326 *
327 * WAIT WAKE
328 * [S] tsk = current [S] cond = true
329 * MB (A) MB (B)
330 * [L] cond [L] tsk
331 */
332 smp_mb(); /* (B) */
333
334 task = rcu_dereference(w->task);
335 if (task)
336 ret = wake_up_process(task);
337 rcu_read_unlock();
338
339 return ret;
340}
341EXPORT_SYMBOL_GPL(rcuwait_wake_up);
342
343/*
344 * Determine if a process group is "orphaned", according to the POSIX
345 * definition in 2.2.2.52. Orphaned process groups are not to be affected
346 * by terminal-generated stop signals. Newly orphaned process groups are
347 * to receive a SIGHUP and a SIGCONT.
348 *
349 * "I ask you, have you ever known what it is to be an orphan?"
350 */
351static int will_become_orphaned_pgrp(struct pid *pgrp,
352 struct task_struct *ignored_task)
353{
354 struct task_struct *p;
355
356 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
357 if ((p == ignored_task) ||
358 (p->exit_state && thread_group_empty(p)) ||
359 is_global_init(p->real_parent))
360 continue;
361
362 if (task_pgrp(p->real_parent) != pgrp &&
363 task_session(p->real_parent) == task_session(p))
364 return 0;
365 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
366
367 return 1;
368}
369
370int is_current_pgrp_orphaned(void)
371{
372 int retval;
373
374 read_lock(&tasklist_lock);
375 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
376 read_unlock(&tasklist_lock);
377
378 return retval;
379}
380
381static bool has_stopped_jobs(struct pid *pgrp)
382{
383 struct task_struct *p;
384
385 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
386 if (p->signal->flags & SIGNAL_STOP_STOPPED)
387 return true;
388 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
389
390 return false;
391}
392
393/*
394 * Check to see if any process groups have become orphaned as
395 * a result of our exiting, and if they have any stopped jobs,
396 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
397 */
398static void
399kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
400{
401 struct pid *pgrp = task_pgrp(tsk);
402 struct task_struct *ignored_task = tsk;
403
404 if (!parent)
405 /* exit: our father is in a different pgrp than
406 * we are and we were the only connection outside.
407 */
408 parent = tsk->real_parent;
409 else
410 /* reparent: our child is in a different pgrp than
411 * we are, and it was the only connection outside.
412 */
413 ignored_task = NULL;
414
415 if (task_pgrp(parent) != pgrp &&
416 task_session(parent) == task_session(tsk) &&
417 will_become_orphaned_pgrp(pgrp, ignored_task) &&
418 has_stopped_jobs(pgrp)) {
419 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
420 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
421 }
422}
423
424static void coredump_task_exit(struct task_struct *tsk,
425 struct core_state *core_state)
426{
427 struct core_thread self;
428
429 self.task = tsk;
430 if (self.task->flags & PF_SIGNALED)
431 self.next = xchg(&core_state->dumper.next, &self);
432 else
433 self.task = NULL;
434 /*
435 * Implies mb(), the result of xchg() must be visible
436 * to core_state->dumper.
437 */
438 if (atomic_dec_and_test(&core_state->nr_threads))
439 complete(&core_state->startup);
440
441 for (;;) {
442 set_current_state(TASK_IDLE|TASK_FREEZABLE);
443 if (!self.task) /* see coredump_finish() */
444 break;
445 schedule();
446 }
447 __set_current_state(TASK_RUNNING);
448}
449
450#ifdef CONFIG_MEMCG
451/* drops tasklist_lock if succeeds */
452static bool __try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm)
453{
454 bool ret = false;
455
456 task_lock(tsk);
457 if (likely(tsk->mm == mm)) {
458 /* tsk can't pass exit_mm/exec_mmap and exit */
459 read_unlock(&tasklist_lock);
460 WRITE_ONCE(mm->owner, tsk);
461 lru_gen_migrate_mm(mm);
462 ret = true;
463 }
464 task_unlock(tsk);
465 return ret;
466}
467
468static bool try_to_set_owner(struct task_struct *g, struct mm_struct *mm)
469{
470 struct task_struct *t;
471
472 for_each_thread(g, t) {
473 struct mm_struct *t_mm = READ_ONCE(t->mm);
474 if (t_mm == mm) {
475 if (__try_to_set_owner(t, mm))
476 return true;
477 } else if (t_mm)
478 break;
479 }
480
481 return false;
482}
483
484/*
485 * A task is exiting. If it owned this mm, find a new owner for the mm.
486 */
487void mm_update_next_owner(struct mm_struct *mm)
488{
489 struct task_struct *g, *p = current;
490
491 /*
492 * If the exiting or execing task is not the owner, it's
493 * someone else's problem.
494 */
495 if (mm->owner != p)
496 return;
497 /*
498 * The current owner is exiting/execing and there are no other
499 * candidates. Do not leave the mm pointing to a possibly
500 * freed task structure.
501 */
502 if (atomic_read(&mm->mm_users) <= 1) {
503 WRITE_ONCE(mm->owner, NULL);
504 return;
505 }
506
507 read_lock(&tasklist_lock);
508 /*
509 * Search in the children
510 */
511 list_for_each_entry(g, &p->children, sibling) {
512 if (try_to_set_owner(g, mm))
513 goto ret;
514 }
515 /*
516 * Search in the siblings
517 */
518 list_for_each_entry(g, &p->real_parent->children, sibling) {
519 if (try_to_set_owner(g, mm))
520 goto ret;
521 }
522 /*
523 * Search through everything else, we should not get here often.
524 */
525 for_each_process(g) {
526 if (atomic_read(&mm->mm_users) <= 1)
527 break;
528 if (g->flags & PF_KTHREAD)
529 continue;
530 if (try_to_set_owner(g, mm))
531 goto ret;
532 }
533 read_unlock(&tasklist_lock);
534 /*
535 * We found no owner yet mm_users > 1: this implies that we are
536 * most likely racing with swapoff (try_to_unuse()) or /proc or
537 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
538 */
539 WRITE_ONCE(mm->owner, NULL);
540 ret:
541 return;
542
543}
544#endif /* CONFIG_MEMCG */
545
546/*
547 * Turn us into a lazy TLB process if we
548 * aren't already..
549 */
550static void exit_mm(void)
551{
552 struct mm_struct *mm = current->mm;
553
554 exit_mm_release(current, mm);
555 if (!mm)
556 return;
557 mmap_read_lock(mm);
558 mmgrab_lazy_tlb(mm);
559 BUG_ON(mm != current->active_mm);
560 /* more a memory barrier than a real lock */
561 task_lock(current);
562 /*
563 * When a thread stops operating on an address space, the loop
564 * in membarrier_private_expedited() may not observe that
565 * tsk->mm, and the loop in membarrier_global_expedited() may
566 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
567 * rq->membarrier_state, so those would not issue an IPI.
568 * Membarrier requires a memory barrier after accessing
569 * user-space memory, before clearing tsk->mm or the
570 * rq->membarrier_state.
571 */
572 smp_mb__after_spinlock();
573 local_irq_disable();
574 current->mm = NULL;
575 membarrier_update_current_mm(NULL);
576 enter_lazy_tlb(mm, current);
577 local_irq_enable();
578 task_unlock(current);
579 mmap_read_unlock(mm);
580 mm_update_next_owner(mm);
581 mmput(mm);
582 if (test_thread_flag(TIF_MEMDIE))
583 exit_oom_victim();
584}
585
586static struct task_struct *find_alive_thread(struct task_struct *p)
587{
588 struct task_struct *t;
589
590 for_each_thread(p, t) {
591 if (!(t->flags & PF_EXITING))
592 return t;
593 }
594 return NULL;
595}
596
597static struct task_struct *find_child_reaper(struct task_struct *father,
598 struct list_head *dead)
599 __releases(&tasklist_lock)
600 __acquires(&tasklist_lock)
601{
602 struct pid_namespace *pid_ns = task_active_pid_ns(father);
603 struct task_struct *reaper = pid_ns->child_reaper;
604 struct task_struct *p, *n;
605
606 if (likely(reaper != father))
607 return reaper;
608
609 reaper = find_alive_thread(father);
610 if (reaper) {
611 pid_ns->child_reaper = reaper;
612 return reaper;
613 }
614
615 write_unlock_irq(&tasklist_lock);
616
617 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
618 list_del_init(&p->ptrace_entry);
619 release_task(p);
620 }
621
622 zap_pid_ns_processes(pid_ns);
623 write_lock_irq(&tasklist_lock);
624
625 return father;
626}
627
628/*
629 * When we die, we re-parent all our children, and try to:
630 * 1. give them to another thread in our thread group, if such a member exists
631 * 2. give it to the first ancestor process which prctl'd itself as a
632 * child_subreaper for its children (like a service manager)
633 * 3. give it to the init process (PID 1) in our pid namespace
634 */
635static struct task_struct *find_new_reaper(struct task_struct *father,
636 struct task_struct *child_reaper)
637{
638 struct task_struct *thread, *reaper;
639
640 thread = find_alive_thread(father);
641 if (thread)
642 return thread;
643
644 if (father->signal->has_child_subreaper) {
645 unsigned int ns_level = task_pid(father)->level;
646 /*
647 * Find the first ->is_child_subreaper ancestor in our pid_ns.
648 * We can't check reaper != child_reaper to ensure we do not
649 * cross the namespaces, the exiting parent could be injected
650 * by setns() + fork().
651 * We check pid->level, this is slightly more efficient than
652 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
653 */
654 for (reaper = father->real_parent;
655 task_pid(reaper)->level == ns_level;
656 reaper = reaper->real_parent) {
657 if (reaper == &init_task)
658 break;
659 if (!reaper->signal->is_child_subreaper)
660 continue;
661 thread = find_alive_thread(reaper);
662 if (thread)
663 return thread;
664 }
665 }
666
667 return child_reaper;
668}
669
670/*
671* Any that need to be release_task'd are put on the @dead list.
672 */
673static void reparent_leader(struct task_struct *father, struct task_struct *p,
674 struct list_head *dead)
675{
676 if (unlikely(p->exit_state == EXIT_DEAD))
677 return;
678
679 /* We don't want people slaying init. */
680 p->exit_signal = SIGCHLD;
681
682 /* If it has exited notify the new parent about this child's death. */
683 if (!p->ptrace &&
684 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
685 if (do_notify_parent(p, p->exit_signal)) {
686 p->exit_state = EXIT_DEAD;
687 list_add(&p->ptrace_entry, dead);
688 }
689 }
690
691 kill_orphaned_pgrp(p, father);
692}
693
694/*
695 * Make init inherit all the child processes
696 */
697static void forget_original_parent(struct task_struct *father,
698 struct list_head *dead)
699{
700 struct task_struct *p, *t, *reaper;
701
702 if (unlikely(!list_empty(&father->ptraced)))
703 exit_ptrace(father, dead);
704
705 /* Can drop and reacquire tasklist_lock */
706 reaper = find_child_reaper(father, dead);
707 if (list_empty(&father->children))
708 return;
709
710 reaper = find_new_reaper(father, reaper);
711 list_for_each_entry(p, &father->children, sibling) {
712 for_each_thread(p, t) {
713 RCU_INIT_POINTER(t->real_parent, reaper);
714 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
715 if (likely(!t->ptrace))
716 t->parent = t->real_parent;
717 if (t->pdeath_signal)
718 group_send_sig_info(t->pdeath_signal,
719 SEND_SIG_NOINFO, t,
720 PIDTYPE_TGID);
721 }
722 /*
723 * If this is a threaded reparent there is no need to
724 * notify anyone anything has happened.
725 */
726 if (!same_thread_group(reaper, father))
727 reparent_leader(father, p, dead);
728 }
729 list_splice_tail_init(&father->children, &reaper->children);
730}
731
732/*
733 * Send signals to all our closest relatives so that they know
734 * to properly mourn us..
735 */
736static void exit_notify(struct task_struct *tsk, int group_dead)
737{
738 bool autoreap;
739 struct task_struct *p, *n;
740 LIST_HEAD(dead);
741
742 write_lock_irq(&tasklist_lock);
743 forget_original_parent(tsk, &dead);
744
745 if (group_dead)
746 kill_orphaned_pgrp(tsk->group_leader, NULL);
747
748 tsk->exit_state = EXIT_ZOMBIE;
749
750 if (unlikely(tsk->ptrace)) {
751 int sig = thread_group_leader(tsk) &&
752 thread_group_empty(tsk) &&
753 !ptrace_reparented(tsk) ?
754 tsk->exit_signal : SIGCHLD;
755 autoreap = do_notify_parent(tsk, sig);
756 } else if (thread_group_leader(tsk)) {
757 autoreap = thread_group_empty(tsk) &&
758 do_notify_parent(tsk, tsk->exit_signal);
759 } else {
760 autoreap = true;
761 /* untraced sub-thread */
762 do_notify_pidfd(tsk);
763 }
764
765 if (autoreap) {
766 tsk->exit_state = EXIT_DEAD;
767 list_add(&tsk->ptrace_entry, &dead);
768 }
769
770 /* mt-exec, de_thread() is waiting for group leader */
771 if (unlikely(tsk->signal->notify_count < 0))
772 wake_up_process(tsk->signal->group_exec_task);
773 write_unlock_irq(&tasklist_lock);
774
775 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
776 list_del_init(&p->ptrace_entry);
777 release_task(p);
778 }
779}
780
781#ifdef CONFIG_DEBUG_STACK_USAGE
782#ifdef CONFIG_STACK_GROWSUP
783unsigned long stack_not_used(struct task_struct *p)
784{
785 unsigned long *n = end_of_stack(p);
786
787 do { /* Skip over canary */
788 n--;
789 } while (!*n);
790
791 return (unsigned long)end_of_stack(p) - (unsigned long)n;
792}
793#else /* !CONFIG_STACK_GROWSUP */
794unsigned long stack_not_used(struct task_struct *p)
795{
796 unsigned long *n = end_of_stack(p);
797
798 do { /* Skip over canary */
799 n++;
800 } while (!*n);
801
802 return (unsigned long)n - (unsigned long)end_of_stack(p);
803}
804#endif /* CONFIG_STACK_GROWSUP */
805
806/* Count the maximum pages reached in kernel stacks */
807static inline void kstack_histogram(unsigned long used_stack)
808{
809#ifdef CONFIG_VM_EVENT_COUNTERS
810 if (used_stack <= 1024)
811 count_vm_event(KSTACK_1K);
812#if THREAD_SIZE > 1024
813 else if (used_stack <= 2048)
814 count_vm_event(KSTACK_2K);
815#endif
816#if THREAD_SIZE > 2048
817 else if (used_stack <= 4096)
818 count_vm_event(KSTACK_4K);
819#endif
820#if THREAD_SIZE > 4096
821 else if (used_stack <= 8192)
822 count_vm_event(KSTACK_8K);
823#endif
824#if THREAD_SIZE > 8192
825 else if (used_stack <= 16384)
826 count_vm_event(KSTACK_16K);
827#endif
828#if THREAD_SIZE > 16384
829 else if (used_stack <= 32768)
830 count_vm_event(KSTACK_32K);
831#endif
832#if THREAD_SIZE > 32768
833 else if (used_stack <= 65536)
834 count_vm_event(KSTACK_64K);
835#endif
836#if THREAD_SIZE > 65536
837 else
838 count_vm_event(KSTACK_REST);
839#endif
840#endif /* CONFIG_VM_EVENT_COUNTERS */
841}
842
843static void check_stack_usage(void)
844{
845 static DEFINE_SPINLOCK(low_water_lock);
846 static int lowest_to_date = THREAD_SIZE;
847 unsigned long free;
848
849 free = stack_not_used(current);
850 kstack_histogram(THREAD_SIZE - free);
851
852 if (free >= lowest_to_date)
853 return;
854
855 spin_lock(&low_water_lock);
856 if (free < lowest_to_date) {
857 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
858 current->comm, task_pid_nr(current), free);
859 lowest_to_date = free;
860 }
861 spin_unlock(&low_water_lock);
862}
863#else /* !CONFIG_DEBUG_STACK_USAGE */
864static inline void check_stack_usage(void) {}
865#endif /* CONFIG_DEBUG_STACK_USAGE */
866
867static void synchronize_group_exit(struct task_struct *tsk, long code)
868{
869 struct sighand_struct *sighand = tsk->sighand;
870 struct signal_struct *signal = tsk->signal;
871 struct core_state *core_state;
872
873 spin_lock_irq(&sighand->siglock);
874 signal->quick_threads--;
875 if ((signal->quick_threads == 0) &&
876 !(signal->flags & SIGNAL_GROUP_EXIT)) {
877 signal->flags = SIGNAL_GROUP_EXIT;
878 signal->group_exit_code = code;
879 signal->group_stop_count = 0;
880 }
881 /*
882 * Serialize with any possible pending coredump.
883 * We must hold siglock around checking core_state
884 * and setting PF_POSTCOREDUMP. The core-inducing thread
885 * will increment ->nr_threads for each thread in the
886 * group without PF_POSTCOREDUMP set.
887 */
888 tsk->flags |= PF_POSTCOREDUMP;
889 core_state = signal->core_state;
890 spin_unlock_irq(&sighand->siglock);
891
892 if (unlikely(core_state))
893 coredump_task_exit(tsk, core_state);
894}
895
896void __noreturn do_exit(long code)
897{
898 struct task_struct *tsk = current;
899 struct kthread *kthread;
900 int group_dead;
901
902 WARN_ON(irqs_disabled());
903 WARN_ON(tsk->plug);
904
905 kthread = tsk_is_kthread(tsk);
906 if (unlikely(kthread))
907 kthread_do_exit(kthread, code);
908
909 kcov_task_exit(tsk);
910 kmsan_task_exit(tsk);
911
912 synchronize_group_exit(tsk, code);
913 ptrace_event(PTRACE_EVENT_EXIT, code);
914 user_events_exit(tsk);
915
916 io_uring_files_cancel();
917 sched_mm_cid_exit(tsk);
918 exit_signals(tsk); /* sets PF_EXITING */
919
920 seccomp_filter_release(tsk);
921
922 acct_update_integrals(tsk);
923 group_dead = atomic_dec_and_test(&tsk->signal->live);
924 if (group_dead) {
925 /*
926 * If the last thread of global init has exited, panic
927 * immediately to get a useable coredump.
928 */
929 if (unlikely(is_global_init(tsk)))
930 panic("Attempted to kill init! exitcode=0x%08x\n",
931 tsk->signal->group_exit_code ?: (int)code);
932
933#ifdef CONFIG_POSIX_TIMERS
934 hrtimer_cancel(&tsk->signal->real_timer);
935 exit_itimers(tsk);
936#endif
937 if (tsk->mm)
938 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
939 }
940 acct_collect(code, group_dead);
941 if (group_dead)
942 tty_audit_exit();
943 audit_free(tsk);
944
945 tsk->exit_code = code;
946 taskstats_exit(tsk, group_dead);
947 trace_sched_process_exit(tsk, group_dead);
948
949 /*
950 * Since sampling can touch ->mm, make sure to stop everything before we
951 * tear it down.
952 *
953 * Also flushes inherited counters to the parent - before the parent
954 * gets woken up by child-exit notifications.
955 */
956 perf_event_exit_task(tsk);
957 /*
958 * PF_EXITING (above) ensures unwind_deferred_request() will no
959 * longer add new unwinds. While exit_mm() (below) will destroy the
960 * abaility to do unwinds. So flush any pending unwinds here.
961 */
962 unwind_deferred_task_exit(tsk);
963
964 exit_mm();
965
966 if (group_dead)
967 acct_process();
968
969 exit_sem(tsk);
970 exit_shm(tsk);
971 exit_files(tsk);
972 exit_fs(tsk);
973 if (group_dead)
974 disassociate_ctty(1);
975 exit_nsproxy_namespaces(tsk);
976 exit_task_work(tsk);
977 exit_thread(tsk);
978
979 sched_autogroup_exit_task(tsk);
980 cgroup_task_exit(tsk);
981
982 /*
983 * FIXME: do that only when needed, using sched_exit tracepoint
984 */
985 flush_ptrace_hw_breakpoint(tsk);
986
987 exit_tasks_rcu_start();
988 exit_notify(tsk, group_dead);
989 proc_exit_connector(tsk);
990 mpol_put_task_policy(tsk);
991#ifdef CONFIG_FUTEX
992 if (unlikely(current->pi_state_cache))
993 kfree(current->pi_state_cache);
994#endif
995 /*
996 * Make sure we are holding no locks:
997 */
998 debug_check_no_locks_held();
999
1000 if (tsk->io_context)
1001 exit_io_context(tsk);
1002
1003 if (tsk->splice_pipe)
1004 free_pipe_info(tsk->splice_pipe);
1005
1006 if (tsk->task_frag.page)
1007 put_page(tsk->task_frag.page);
1008
1009 exit_task_stack_account(tsk);
1010
1011 check_stack_usage();
1012 preempt_disable();
1013 if (tsk->nr_dirtied)
1014 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
1015 exit_rcu();
1016 exit_tasks_rcu_finish();
1017
1018 lockdep_free_task(tsk);
1019 do_task_dead();
1020}
1021EXPORT_SYMBOL(do_exit);
1022
1023void __noreturn make_task_dead(int signr)
1024{
1025 /*
1026 * Take the task off the cpu after something catastrophic has
1027 * happened.
1028 *
1029 * We can get here from a kernel oops, sometimes with preemption off.
1030 * Start by checking for critical errors.
1031 * Then fix up important state like USER_DS and preemption.
1032 * Then do everything else.
1033 */
1034 struct task_struct *tsk = current;
1035 unsigned int limit;
1036
1037 if (unlikely(in_interrupt()))
1038 panic("Aiee, killing interrupt handler!");
1039 if (unlikely(!tsk->pid))
1040 panic("Attempted to kill the idle task!");
1041
1042 if (unlikely(irqs_disabled())) {
1043 pr_info("note: %s[%d] exited with irqs disabled\n",
1044 current->comm, task_pid_nr(current));
1045 local_irq_enable();
1046 }
1047 if (unlikely(in_atomic())) {
1048 pr_info("note: %s[%d] exited with preempt_count %d\n",
1049 current->comm, task_pid_nr(current),
1050 preempt_count());
1051 preempt_count_set(PREEMPT_ENABLED);
1052 }
1053
1054 /*
1055 * Every time the system oopses, if the oops happens while a reference
1056 * to an object was held, the reference leaks.
1057 * If the oops doesn't also leak memory, repeated oopsing can cause
1058 * reference counters to wrap around (if they're not using refcount_t).
1059 * This means that repeated oopsing can make unexploitable-looking bugs
1060 * exploitable through repeated oopsing.
1061 * To make sure this can't happen, place an upper bound on how often the
1062 * kernel may oops without panic().
1063 */
1064 limit = READ_ONCE(oops_limit);
1065 if (atomic_inc_return(&oops_count) >= limit && limit)
1066 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
1067
1068 /*
1069 * We're taking recursive faults here in make_task_dead. Safest is to just
1070 * leave this task alone and wait for reboot.
1071 */
1072 if (unlikely(tsk->flags & PF_EXITING)) {
1073 pr_alert("Fixing recursive fault but reboot is needed!\n");
1074 futex_exit_recursive(tsk);
1075 tsk->exit_state = EXIT_DEAD;
1076 refcount_inc(&tsk->rcu_users);
1077 do_task_dead();
1078 }
1079
1080 do_exit(signr);
1081}
1082
1083SYSCALL_DEFINE1(exit, int, error_code)
1084{
1085 do_exit((error_code&0xff)<<8);
1086}
1087
1088/*
1089 * Take down every thread in the group. This is called by fatal signals
1090 * as well as by sys_exit_group (below).
1091 */
1092void __noreturn
1093do_group_exit(int exit_code)
1094{
1095 struct signal_struct *sig = current->signal;
1096
1097 if (sig->flags & SIGNAL_GROUP_EXIT)
1098 exit_code = sig->group_exit_code;
1099 else if (sig->group_exec_task)
1100 exit_code = 0;
1101 else {
1102 struct sighand_struct *const sighand = current->sighand;
1103
1104 spin_lock_irq(&sighand->siglock);
1105 if (sig->flags & SIGNAL_GROUP_EXIT)
1106 /* Another thread got here before we took the lock. */
1107 exit_code = sig->group_exit_code;
1108 else if (sig->group_exec_task)
1109 exit_code = 0;
1110 else {
1111 sig->group_exit_code = exit_code;
1112 sig->flags = SIGNAL_GROUP_EXIT;
1113 zap_other_threads(current);
1114 }
1115 spin_unlock_irq(&sighand->siglock);
1116 }
1117
1118 do_exit(exit_code);
1119 /* NOTREACHED */
1120}
1121
1122/*
1123 * this kills every thread in the thread group. Note that any externally
1124 * wait4()-ing process will get the correct exit code - even if this
1125 * thread is not the thread group leader.
1126 */
1127SYSCALL_DEFINE1(exit_group, int, error_code)
1128{
1129 do_group_exit((error_code & 0xff) << 8);
1130 /* NOTREACHED */
1131 return 0;
1132}
1133
1134static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1135{
1136 return wo->wo_type == PIDTYPE_MAX ||
1137 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1138}
1139
1140static int
1141eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1142{
1143 if (!eligible_pid(wo, p))
1144 return 0;
1145
1146 /*
1147 * Wait for all children (clone and not) if __WALL is set or
1148 * if it is traced by us.
1149 */
1150 if (ptrace || (wo->wo_flags & __WALL))
1151 return 1;
1152
1153 /*
1154 * Otherwise, wait for clone children *only* if __WCLONE is set;
1155 * otherwise, wait for non-clone children *only*.
1156 *
1157 * Note: a "clone" child here is one that reports to its parent
1158 * using a signal other than SIGCHLD, or a non-leader thread which
1159 * we can only see if it is traced by us.
1160 */
1161 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1162 return 0;
1163
1164 return 1;
1165}
1166
1167/*
1168 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1169 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1170 * the lock and this task is uninteresting. If we return nonzero, we have
1171 * released the lock and the system call should return.
1172 */
1173static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1174{
1175 int state, status;
1176 pid_t pid = task_pid_vnr(p);
1177 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1178 struct waitid_info *infop;
1179
1180 if (!likely(wo->wo_flags & WEXITED))
1181 return 0;
1182
1183 if (unlikely(wo->wo_flags & WNOWAIT)) {
1184 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1185 ? p->signal->group_exit_code : p->exit_code;
1186 get_task_struct(p);
1187 read_unlock(&tasklist_lock);
1188 sched_annotate_sleep();
1189 if (wo->wo_rusage)
1190 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1191 put_task_struct(p);
1192 goto out_info;
1193 }
1194 /*
1195 * Move the task's state to DEAD/TRACE, only one thread can do this.
1196 */
1197 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1198 EXIT_TRACE : EXIT_DEAD;
1199 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1200 return 0;
1201 /*
1202 * We own this thread, nobody else can reap it.
1203 */
1204 read_unlock(&tasklist_lock);
1205 sched_annotate_sleep();
1206
1207 /*
1208 * Check thread_group_leader() to exclude the traced sub-threads.
1209 */
1210 if (state == EXIT_DEAD && thread_group_leader(p)) {
1211 struct signal_struct *sig = p->signal;
1212 struct signal_struct *psig = current->signal;
1213 unsigned long maxrss;
1214 u64 tgutime, tgstime;
1215
1216 /*
1217 * The resource counters for the group leader are in its
1218 * own task_struct. Those for dead threads in the group
1219 * are in its signal_struct, as are those for the child
1220 * processes it has previously reaped. All these
1221 * accumulate in the parent's signal_struct c* fields.
1222 *
1223 * We don't bother to take a lock here to protect these
1224 * p->signal fields because the whole thread group is dead
1225 * and nobody can change them.
1226 *
1227 * psig->stats_lock also protects us from our sub-threads
1228 * which can reap other children at the same time.
1229 *
1230 * We use thread_group_cputime_adjusted() to get times for
1231 * the thread group, which consolidates times for all threads
1232 * in the group including the group leader.
1233 */
1234 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1235 write_seqlock_irq(&psig->stats_lock);
1236 psig->cutime += tgutime + sig->cutime;
1237 psig->cstime += tgstime + sig->cstime;
1238 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1239 psig->cmin_flt +=
1240 p->min_flt + sig->min_flt + sig->cmin_flt;
1241 psig->cmaj_flt +=
1242 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1243 psig->cnvcsw +=
1244 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1245 psig->cnivcsw +=
1246 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1247 psig->cinblock +=
1248 task_io_get_inblock(p) +
1249 sig->inblock + sig->cinblock;
1250 psig->coublock +=
1251 task_io_get_oublock(p) +
1252 sig->oublock + sig->coublock;
1253 maxrss = max(sig->maxrss, sig->cmaxrss);
1254 if (psig->cmaxrss < maxrss)
1255 psig->cmaxrss = maxrss;
1256 task_io_accounting_add(&psig->ioac, &p->ioac);
1257 task_io_accounting_add(&psig->ioac, &sig->ioac);
1258 write_sequnlock_irq(&psig->stats_lock);
1259 }
1260
1261 if (wo->wo_rusage)
1262 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1263 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1264 ? p->signal->group_exit_code : p->exit_code;
1265 wo->wo_stat = status;
1266
1267 if (state == EXIT_TRACE) {
1268 write_lock_irq(&tasklist_lock);
1269 /* We dropped tasklist, ptracer could die and untrace */
1270 ptrace_unlink(p);
1271
1272 /* If parent wants a zombie, don't release it now */
1273 state = EXIT_ZOMBIE;
1274 if (do_notify_parent(p, p->exit_signal))
1275 state = EXIT_DEAD;
1276 p->exit_state = state;
1277 write_unlock_irq(&tasklist_lock);
1278 }
1279 if (state == EXIT_DEAD)
1280 release_task(p);
1281
1282out_info:
1283 infop = wo->wo_info;
1284 if (infop) {
1285 if ((status & 0x7f) == 0) {
1286 infop->cause = CLD_EXITED;
1287 infop->status = status >> 8;
1288 } else {
1289 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1290 infop->status = status & 0x7f;
1291 }
1292 infop->pid = pid;
1293 infop->uid = uid;
1294 }
1295
1296 return pid;
1297}
1298
1299static int *task_stopped_code(struct task_struct *p, bool ptrace)
1300{
1301 if (ptrace) {
1302 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1303 return &p->exit_code;
1304 } else {
1305 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1306 return &p->signal->group_exit_code;
1307 }
1308 return NULL;
1309}
1310
1311/**
1312 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1313 * @wo: wait options
1314 * @ptrace: is the wait for ptrace
1315 * @p: task to wait for
1316 *
1317 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1318 *
1319 * CONTEXT:
1320 * read_lock(&tasklist_lock), which is released if return value is
1321 * non-zero. Also, grabs and releases @p->sighand->siglock.
1322 *
1323 * RETURNS:
1324 * 0 if wait condition didn't exist and search for other wait conditions
1325 * should continue. Non-zero return, -errno on failure and @p's pid on
1326 * success, implies that tasklist_lock is released and wait condition
1327 * search should terminate.
1328 */
1329static int wait_task_stopped(struct wait_opts *wo,
1330 int ptrace, struct task_struct *p)
1331{
1332 struct waitid_info *infop;
1333 int exit_code, *p_code, why;
1334 uid_t uid = 0; /* unneeded, required by compiler */
1335 pid_t pid;
1336
1337 /*
1338 * Traditionally we see ptrace'd stopped tasks regardless of options.
1339 */
1340 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1341 return 0;
1342
1343 if (!task_stopped_code(p, ptrace))
1344 return 0;
1345
1346 exit_code = 0;
1347 spin_lock_irq(&p->sighand->siglock);
1348
1349 p_code = task_stopped_code(p, ptrace);
1350 if (unlikely(!p_code))
1351 goto unlock_sig;
1352
1353 exit_code = *p_code;
1354 if (!exit_code)
1355 goto unlock_sig;
1356
1357 if (!unlikely(wo->wo_flags & WNOWAIT))
1358 *p_code = 0;
1359
1360 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1361unlock_sig:
1362 spin_unlock_irq(&p->sighand->siglock);
1363 if (!exit_code)
1364 return 0;
1365
1366 /*
1367 * Now we are pretty sure this task is interesting.
1368 * Make sure it doesn't get reaped out from under us while we
1369 * give up the lock and then examine it below. We don't want to
1370 * keep holding onto the tasklist_lock while we call getrusage and
1371 * possibly take page faults for user memory.
1372 */
1373 get_task_struct(p);
1374 pid = task_pid_vnr(p);
1375 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1376 read_unlock(&tasklist_lock);
1377 sched_annotate_sleep();
1378 if (wo->wo_rusage)
1379 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1380 put_task_struct(p);
1381
1382 if (likely(!(wo->wo_flags & WNOWAIT)))
1383 wo->wo_stat = (exit_code << 8) | 0x7f;
1384
1385 infop = wo->wo_info;
1386 if (infop) {
1387 infop->cause = why;
1388 infop->status = exit_code;
1389 infop->pid = pid;
1390 infop->uid = uid;
1391 }
1392 return pid;
1393}
1394
1395/*
1396 * Handle do_wait work for one task in a live, non-stopped state.
1397 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1398 * the lock and this task is uninteresting. If we return nonzero, we have
1399 * released the lock and the system call should return.
1400 */
1401static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1402{
1403 struct waitid_info *infop;
1404 pid_t pid;
1405 uid_t uid;
1406
1407 if (!unlikely(wo->wo_flags & WCONTINUED))
1408 return 0;
1409
1410 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1411 return 0;
1412
1413 spin_lock_irq(&p->sighand->siglock);
1414 /* Re-check with the lock held. */
1415 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1416 spin_unlock_irq(&p->sighand->siglock);
1417 return 0;
1418 }
1419 if (!unlikely(wo->wo_flags & WNOWAIT))
1420 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1421 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1422 spin_unlock_irq(&p->sighand->siglock);
1423
1424 pid = task_pid_vnr(p);
1425 get_task_struct(p);
1426 read_unlock(&tasklist_lock);
1427 sched_annotate_sleep();
1428 if (wo->wo_rusage)
1429 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1430 put_task_struct(p);
1431
1432 infop = wo->wo_info;
1433 if (!infop) {
1434 wo->wo_stat = 0xffff;
1435 } else {
1436 infop->cause = CLD_CONTINUED;
1437 infop->pid = pid;
1438 infop->uid = uid;
1439 infop->status = SIGCONT;
1440 }
1441 return pid;
1442}
1443
1444/*
1445 * Consider @p for a wait by @parent.
1446 *
1447 * -ECHILD should be in ->notask_error before the first call.
1448 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1449 * Returns zero if the search for a child should continue;
1450 * then ->notask_error is 0 if @p is an eligible child,
1451 * or still -ECHILD.
1452 */
1453static int wait_consider_task(struct wait_opts *wo, int ptrace,
1454 struct task_struct *p)
1455{
1456 /*
1457 * We can race with wait_task_zombie() from another thread.
1458 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1459 * can't confuse the checks below.
1460 */
1461 int exit_state = READ_ONCE(p->exit_state);
1462 int ret;
1463
1464 if (unlikely(exit_state == EXIT_DEAD))
1465 return 0;
1466
1467 ret = eligible_child(wo, ptrace, p);
1468 if (!ret)
1469 return ret;
1470
1471 if (unlikely(exit_state == EXIT_TRACE)) {
1472 /*
1473 * ptrace == 0 means we are the natural parent. In this case
1474 * we should clear notask_error, debugger will notify us.
1475 */
1476 if (likely(!ptrace))
1477 wo->notask_error = 0;
1478 return 0;
1479 }
1480
1481 if (likely(!ptrace) && unlikely(p->ptrace)) {
1482 /*
1483 * If it is traced by its real parent's group, just pretend
1484 * the caller is ptrace_do_wait() and reap this child if it
1485 * is zombie.
1486 *
1487 * This also hides group stop state from real parent; otherwise
1488 * a single stop can be reported twice as group and ptrace stop.
1489 * If a ptracer wants to distinguish these two events for its
1490 * own children it should create a separate process which takes
1491 * the role of real parent.
1492 */
1493 if (!ptrace_reparented(p))
1494 ptrace = 1;
1495 }
1496
1497 /* slay zombie? */
1498 if (exit_state == EXIT_ZOMBIE) {
1499 /* we don't reap group leaders with subthreads */
1500 if (!delay_group_leader(p)) {
1501 /*
1502 * A zombie ptracee is only visible to its ptracer.
1503 * Notification and reaping will be cascaded to the
1504 * real parent when the ptracer detaches.
1505 */
1506 if (unlikely(ptrace) || likely(!p->ptrace))
1507 return wait_task_zombie(wo, p);
1508 }
1509
1510 /*
1511 * Allow access to stopped/continued state via zombie by
1512 * falling through. Clearing of notask_error is complex.
1513 *
1514 * When !@ptrace:
1515 *
1516 * If WEXITED is set, notask_error should naturally be
1517 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1518 * so, if there are live subthreads, there are events to
1519 * wait for. If all subthreads are dead, it's still safe
1520 * to clear - this function will be called again in finite
1521 * amount time once all the subthreads are released and
1522 * will then return without clearing.
1523 *
1524 * When @ptrace:
1525 *
1526 * Stopped state is per-task and thus can't change once the
1527 * target task dies. Only continued and exited can happen.
1528 * Clear notask_error if WCONTINUED | WEXITED.
1529 */
1530 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1531 wo->notask_error = 0;
1532 } else {
1533 /*
1534 * @p is alive and it's gonna stop, continue or exit, so
1535 * there always is something to wait for.
1536 */
1537 wo->notask_error = 0;
1538 }
1539
1540 /*
1541 * Wait for stopped. Depending on @ptrace, different stopped state
1542 * is used and the two don't interact with each other.
1543 */
1544 ret = wait_task_stopped(wo, ptrace, p);
1545 if (ret)
1546 return ret;
1547
1548 /*
1549 * Wait for continued. There's only one continued state and the
1550 * ptracer can consume it which can confuse the real parent. Don't
1551 * use WCONTINUED from ptracer. You don't need or want it.
1552 */
1553 return wait_task_continued(wo, p);
1554}
1555
1556/*
1557 * Do the work of do_wait() for one thread in the group, @tsk.
1558 *
1559 * -ECHILD should be in ->notask_error before the first call.
1560 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1561 * Returns zero if the search for a child should continue; then
1562 * ->notask_error is 0 if there were any eligible children,
1563 * or still -ECHILD.
1564 */
1565static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1566{
1567 struct task_struct *p;
1568
1569 list_for_each_entry(p, &tsk->children, sibling) {
1570 int ret = wait_consider_task(wo, 0, p);
1571
1572 if (ret)
1573 return ret;
1574 }
1575
1576 return 0;
1577}
1578
1579static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1580{
1581 struct task_struct *p;
1582
1583 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1584 int ret = wait_consider_task(wo, 1, p);
1585
1586 if (ret)
1587 return ret;
1588 }
1589
1590 return 0;
1591}
1592
1593bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1594{
1595 if (!eligible_pid(wo, p))
1596 return false;
1597
1598 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1599 return false;
1600
1601 return true;
1602}
1603
1604static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1605 int sync, void *key)
1606{
1607 struct wait_opts *wo = container_of(wait, struct wait_opts,
1608 child_wait);
1609 struct task_struct *p = key;
1610
1611 if (pid_child_should_wake(wo, p))
1612 return default_wake_function(wait, mode, sync, key);
1613
1614 return 0;
1615}
1616
1617void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1618{
1619 __wake_up_sync_key(&parent->signal->wait_chldexit,
1620 TASK_INTERRUPTIBLE, p);
1621}
1622
1623static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1624 struct task_struct *target)
1625{
1626 struct task_struct *parent =
1627 !ptrace ? target->real_parent : target->parent;
1628
1629 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1630 same_thread_group(current, parent));
1631}
1632
1633/*
1634 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1635 * and tracee lists to find the target task.
1636 */
1637static int do_wait_pid(struct wait_opts *wo)
1638{
1639 bool ptrace;
1640 struct task_struct *target;
1641 int retval;
1642
1643 ptrace = false;
1644 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1645 if (target && is_effectively_child(wo, ptrace, target)) {
1646 retval = wait_consider_task(wo, ptrace, target);
1647 if (retval)
1648 return retval;
1649 }
1650
1651 ptrace = true;
1652 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1653 if (target && target->ptrace &&
1654 is_effectively_child(wo, ptrace, target)) {
1655 retval = wait_consider_task(wo, ptrace, target);
1656 if (retval)
1657 return retval;
1658 }
1659
1660 return 0;
1661}
1662
1663long __do_wait(struct wait_opts *wo)
1664{
1665 long retval;
1666
1667 /*
1668 * If there is nothing that can match our criteria, just get out.
1669 * We will clear ->notask_error to zero if we see any child that
1670 * might later match our criteria, even if we are not able to reap
1671 * it yet.
1672 */
1673 wo->notask_error = -ECHILD;
1674 if ((wo->wo_type < PIDTYPE_MAX) &&
1675 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1676 goto notask;
1677
1678 read_lock(&tasklist_lock);
1679
1680 if (wo->wo_type == PIDTYPE_PID) {
1681 retval = do_wait_pid(wo);
1682 if (retval)
1683 return retval;
1684 } else {
1685 struct task_struct *tsk = current;
1686
1687 do {
1688 retval = do_wait_thread(wo, tsk);
1689 if (retval)
1690 return retval;
1691
1692 retval = ptrace_do_wait(wo, tsk);
1693 if (retval)
1694 return retval;
1695
1696 if (wo->wo_flags & __WNOTHREAD)
1697 break;
1698 } while_each_thread(current, tsk);
1699 }
1700 read_unlock(&tasklist_lock);
1701
1702notask:
1703 retval = wo->notask_error;
1704 if (!retval && !(wo->wo_flags & WNOHANG))
1705 return -ERESTARTSYS;
1706
1707 return retval;
1708}
1709
1710static long do_wait(struct wait_opts *wo)
1711{
1712 int retval;
1713
1714 trace_sched_process_wait(wo->wo_pid);
1715
1716 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1717 wo->child_wait.private = current;
1718 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1719
1720 do {
1721 set_current_state(TASK_INTERRUPTIBLE);
1722 retval = __do_wait(wo);
1723 if (retval != -ERESTARTSYS)
1724 break;
1725 if (signal_pending(current))
1726 break;
1727 schedule();
1728 } while (1);
1729
1730 __set_current_state(TASK_RUNNING);
1731 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1732 return retval;
1733}
1734
1735int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1736 struct waitid_info *infop, int options,
1737 struct rusage *ru)
1738{
1739 unsigned int f_flags = 0;
1740 struct pid *pid = NULL;
1741 enum pid_type type;
1742
1743 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1744 __WNOTHREAD|__WCLONE|__WALL))
1745 return -EINVAL;
1746 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1747 return -EINVAL;
1748
1749 switch (which) {
1750 case P_ALL:
1751 type = PIDTYPE_MAX;
1752 break;
1753 case P_PID:
1754 type = PIDTYPE_PID;
1755 if (upid <= 0)
1756 return -EINVAL;
1757
1758 pid = find_get_pid(upid);
1759 break;
1760 case P_PGID:
1761 type = PIDTYPE_PGID;
1762 if (upid < 0)
1763 return -EINVAL;
1764
1765 if (upid)
1766 pid = find_get_pid(upid);
1767 else
1768 pid = get_task_pid(current, PIDTYPE_PGID);
1769 break;
1770 case P_PIDFD:
1771 type = PIDTYPE_PID;
1772 if (upid < 0)
1773 return -EINVAL;
1774
1775 pid = pidfd_get_pid(upid, &f_flags);
1776 if (IS_ERR(pid))
1777 return PTR_ERR(pid);
1778
1779 break;
1780 default:
1781 return -EINVAL;
1782 }
1783
1784 wo->wo_type = type;
1785 wo->wo_pid = pid;
1786 wo->wo_flags = options;
1787 wo->wo_info = infop;
1788 wo->wo_rusage = ru;
1789 if (f_flags & O_NONBLOCK)
1790 wo->wo_flags |= WNOHANG;
1791
1792 return 0;
1793}
1794
1795static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1796 int options, struct rusage *ru)
1797{
1798 struct wait_opts wo;
1799 long ret;
1800
1801 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1802 if (ret)
1803 return ret;
1804
1805 ret = do_wait(&wo);
1806 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1807 ret = -EAGAIN;
1808
1809 put_pid(wo.wo_pid);
1810 return ret;
1811}
1812
1813SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1814 infop, int, options, struct rusage __user *, ru)
1815{
1816 struct rusage r;
1817 struct waitid_info info = {.status = 0};
1818 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1819 int signo = 0;
1820
1821 if (err > 0) {
1822 signo = SIGCHLD;
1823 err = 0;
1824 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1825 return -EFAULT;
1826 }
1827 if (!infop)
1828 return err;
1829
1830 if (!user_write_access_begin(infop, sizeof(*infop)))
1831 return -EFAULT;
1832
1833 unsafe_put_user(signo, &infop->si_signo, Efault);
1834 unsafe_put_user(0, &infop->si_errno, Efault);
1835 unsafe_put_user(info.cause, &infop->si_code, Efault);
1836 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1837 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1838 unsafe_put_user(info.status, &infop->si_status, Efault);
1839 user_write_access_end();
1840 return err;
1841Efault:
1842 user_write_access_end();
1843 return -EFAULT;
1844}
1845
1846long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1847 struct rusage *ru)
1848{
1849 struct wait_opts wo;
1850 struct pid *pid = NULL;
1851 enum pid_type type;
1852 long ret;
1853
1854 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1855 __WNOTHREAD|__WCLONE|__WALL))
1856 return -EINVAL;
1857
1858 /* -INT_MIN is not defined */
1859 if (upid == INT_MIN)
1860 return -ESRCH;
1861
1862 if (upid == -1)
1863 type = PIDTYPE_MAX;
1864 else if (upid < 0) {
1865 type = PIDTYPE_PGID;
1866 pid = find_get_pid(-upid);
1867 } else if (upid == 0) {
1868 type = PIDTYPE_PGID;
1869 pid = get_task_pid(current, PIDTYPE_PGID);
1870 } else /* upid > 0 */ {
1871 type = PIDTYPE_PID;
1872 pid = find_get_pid(upid);
1873 }
1874
1875 wo.wo_type = type;
1876 wo.wo_pid = pid;
1877 wo.wo_flags = options | WEXITED;
1878 wo.wo_info = NULL;
1879 wo.wo_stat = 0;
1880 wo.wo_rusage = ru;
1881 ret = do_wait(&wo);
1882 put_pid(pid);
1883 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1884 ret = -EFAULT;
1885
1886 return ret;
1887}
1888
1889int kernel_wait(pid_t pid, int *stat)
1890{
1891 struct wait_opts wo = {
1892 .wo_type = PIDTYPE_PID,
1893 .wo_pid = find_get_pid(pid),
1894 .wo_flags = WEXITED,
1895 };
1896 int ret;
1897
1898 ret = do_wait(&wo);
1899 if (ret > 0 && wo.wo_stat)
1900 *stat = wo.wo_stat;
1901 put_pid(wo.wo_pid);
1902 return ret;
1903}
1904
1905SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1906 int, options, struct rusage __user *, ru)
1907{
1908 struct rusage r;
1909 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1910
1911 if (err > 0) {
1912 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1913 return -EFAULT;
1914 }
1915 return err;
1916}
1917
1918#ifdef __ARCH_WANT_SYS_WAITPID
1919
1920/*
1921 * sys_waitpid() remains for compatibility. waitpid() should be
1922 * implemented by calling sys_wait4() from libc.a.
1923 */
1924SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1925{
1926 return kernel_wait4(pid, stat_addr, options, NULL);
1927}
1928
1929#endif
1930
1931#ifdef CONFIG_COMPAT
1932COMPAT_SYSCALL_DEFINE4(wait4,
1933 compat_pid_t, pid,
1934 compat_uint_t __user *, stat_addr,
1935 int, options,
1936 struct compat_rusage __user *, ru)
1937{
1938 struct rusage r;
1939 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1940 if (err > 0) {
1941 if (ru && put_compat_rusage(&r, ru))
1942 return -EFAULT;
1943 }
1944 return err;
1945}
1946
1947COMPAT_SYSCALL_DEFINE5(waitid,
1948 int, which, compat_pid_t, pid,
1949 struct compat_siginfo __user *, infop, int, options,
1950 struct compat_rusage __user *, uru)
1951{
1952 struct rusage ru;
1953 struct waitid_info info = {.status = 0};
1954 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1955 int signo = 0;
1956 if (err > 0) {
1957 signo = SIGCHLD;
1958 err = 0;
1959 if (uru) {
1960 /* kernel_waitid() overwrites everything in ru */
1961 if (COMPAT_USE_64BIT_TIME)
1962 err = copy_to_user(uru, &ru, sizeof(ru));
1963 else
1964 err = put_compat_rusage(&ru, uru);
1965 if (err)
1966 return -EFAULT;
1967 }
1968 }
1969
1970 if (!infop)
1971 return err;
1972
1973 if (!user_write_access_begin(infop, sizeof(*infop)))
1974 return -EFAULT;
1975
1976 unsafe_put_user(signo, &infop->si_signo, Efault);
1977 unsafe_put_user(0, &infop->si_errno, Efault);
1978 unsafe_put_user(info.cause, &infop->si_code, Efault);
1979 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1980 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1981 unsafe_put_user(info.status, &infop->si_status, Efault);
1982 user_write_access_end();
1983 return err;
1984Efault:
1985 user_write_access_end();
1986 return -EFAULT;
1987}
1988#endif
1989
1990/*
1991 * This needs to be __function_aligned as GCC implicitly makes any
1992 * implementation of abort() cold and drops alignment specified by
1993 * -falign-functions=N.
1994 *
1995 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1996 */
1997__weak __function_aligned void abort(void)
1998{
1999 BUG();
2000
2001 /* if that doesn't kill us, halt */
2002 panic("Oops failed to kill thread");
2003}
2004EXPORT_SYMBOL(abort);