projects/linux·blob v7.2 · refs

2038 lines · 50.6 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 /*
216 * Ensure that all preceeding state is visible. Pairs with
217 * the smp_acquire__after_ctrl_dep() in the sighand == NULL
218 * path of lock_task_sighand().
219 */
220 smp_store_release(&tsk->sighand, NULL);
221 spin_unlock(&sighand->siglock);
222
223 __cleanup_sighand(sighand);
224 if (group_dead)
225 tty_kref_put(tty);
226}
227
228static void delayed_put_task_struct(struct rcu_head *rhp)
229{
230 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
231
232 kprobe_flush_task(tsk);
233 rethook_flush_task(tsk);
234 perf_event_delayed_put(tsk);
235 trace_sched_process_free(tsk);
236 put_task_struct(tsk);
237}
238
239void put_task_struct_rcu_user(struct task_struct *task)
240{
241 if (refcount_dec_and_test(&task->rcu_users))
242 call_rcu(&task->rcu, delayed_put_task_struct);
243}
244
245void __weak release_thread(struct task_struct *dead_task)
246{
247}
248
249void release_task(struct task_struct *p)
250{
251 struct release_task_post post;
252 struct task_struct *leader;
253 struct pid *thread_pid;
254 int zap_leader;
255repeat:
256 memset(&post, 0, sizeof(post));
257
258 /* don't need to get the RCU readlock here - the process is dead and
259 * can't be modifying its own credentials. */
260 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
261
262 pidfs_exit(p);
263 cgroup_task_release(p);
264
265 /* Retrieve @thread_pid before __unhash_process() may set it to NULL. */
266 thread_pid = task_pid(p);
267
268 write_lock_irq(&tasklist_lock);
269 ptrace_release_task(p);
270 __exit_signal(&post, p);
271
272 /*
273 * If we are the last non-leader member of the thread
274 * group, and the leader is zombie, then notify the
275 * group leader's parent process. (if it wants notification.)
276 */
277 zap_leader = 0;
278 leader = p->group_leader;
279 if (leader != p && thread_group_empty(leader)
280 && leader->exit_state == EXIT_ZOMBIE) {
281 /* for pidfs_exit() and do_notify_parent() */
282 if (leader->signal->flags & SIGNAL_GROUP_EXIT)
283 leader->exit_code = leader->signal->group_exit_code;
284 /*
285 * If we were the last child thread and the leader has
286 * exited already, and the leader's parent ignores SIGCHLD,
287 * then we are the one who should release the leader.
288 */
289 zap_leader = do_notify_parent(leader, leader->exit_signal);
290 if (zap_leader)
291 leader->exit_state = EXIT_DEAD;
292 }
293
294 write_unlock_irq(&tasklist_lock);
295 /* @thread_pid can't go away until free_pids() below */
296 proc_flush_pid(thread_pid);
297 exit_cred_namespaces(p);
298 add_device_randomness(&p->se.sum_exec_runtime,
299 sizeof(p->se.sum_exec_runtime));
300 free_pids(post.pids);
301 release_thread(p);
302 /*
303 * This task was already removed from the process/thread/pid lists
304 * and lock_task_sighand(p) can't succeed. Nobody else can touch
305 * ->pending or, if group dead, signal->shared_pending. We can call
306 * flush_sigqueue() lockless.
307 */
308 flush_sigqueue(&p->pending);
309 if (thread_group_leader(p))
310 flush_sigqueue(&p->signal->shared_pending);
311
312 put_task_struct_rcu_user(p);
313
314 p = leader;
315 if (unlikely(zap_leader))
316 goto repeat;
317}
318
319int rcuwait_wake_up(struct rcuwait *w)
320{
321 int ret = 0;
322 struct task_struct *task;
323
324 rcu_read_lock();
325
326 /*
327 * Order condition vs @task, such that everything prior to the load
328 * of @task is visible. This is the condition as to why the user called
329 * rcuwait_wake() in the first place. Pairs with set_current_state()
330 * barrier (A) in rcuwait_wait_event().
331 *
332 * WAIT WAKE
333 * [S] tsk = current [S] cond = true
334 * MB (A) MB (B)
335 * [L] cond [L] tsk
336 */
337 smp_mb(); /* (B) */
338
339 task = rcu_dereference(w->task);
340 if (task)
341 ret = wake_up_process(task);
342 rcu_read_unlock();
343
344 return ret;
345}
346EXPORT_SYMBOL_GPL(rcuwait_wake_up);
347
348/*
349 * Determine if a process group is "orphaned", according to the POSIX
350 * definition in 2.2.2.52. Orphaned process groups are not to be affected
351 * by terminal-generated stop signals. Newly orphaned process groups are
352 * to receive a SIGHUP and a SIGCONT.
353 *
354 * "I ask you, have you ever known what it is to be an orphan?"
355 */
356static int will_become_orphaned_pgrp(struct pid *pgrp,
357 struct task_struct *ignored_task)
358{
359 struct task_struct *p;
360
361 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
362 if ((p == ignored_task) ||
363 (p->exit_state && thread_group_empty(p)) ||
364 is_global_init(p->real_parent))
365 continue;
366
367 if (task_pgrp(p->real_parent) != pgrp &&
368 task_session(p->real_parent) == task_session(p))
369 return 0;
370 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
371
372 return 1;
373}
374
375int is_current_pgrp_orphaned(void)
376{
377 int retval;
378
379 read_lock(&tasklist_lock);
380 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
381 read_unlock(&tasklist_lock);
382
383 return retval;
384}
385
386static bool has_stopped_jobs(struct pid *pgrp)
387{
388 struct task_struct *p;
389
390 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
391 if (p->signal->flags & SIGNAL_STOP_STOPPED)
392 return true;
393 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
394
395 return false;
396}
397
398/*
399 * Check to see if any process groups have become orphaned as
400 * a result of our exiting, and if they have any stopped jobs,
401 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
402 */
403static void
404kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
405{
406 struct pid *pgrp = task_pgrp(tsk);
407 struct task_struct *ignored_task = tsk;
408
409 if (!parent)
410 /* exit: our father is in a different pgrp than
411 * we are and we were the only connection outside.
412 */
413 parent = tsk->real_parent;
414 else
415 /* reparent: our child is in a different pgrp than
416 * we are, and it was the only connection outside.
417 */
418 ignored_task = NULL;
419
420 if (task_pgrp(parent) != pgrp &&
421 task_session(parent) == task_session(tsk) &&
422 will_become_orphaned_pgrp(pgrp, ignored_task) &&
423 has_stopped_jobs(pgrp)) {
424 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
425 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
426 }
427}
428
429static void coredump_task_exit(struct task_struct *tsk,
430 struct core_state *core_state)
431{
432 struct core_thread self;
433
434 self.task = tsk;
435 if (self.task->flags & PF_SIGNALED)
436 self.next = xchg(&core_state->dumper.next, &self);
437 else
438 self.task = NULL;
439 /*
440 * Implies mb(), the result of xchg() must be visible
441 * to core_state->dumper.
442 */
443 if (atomic_dec_and_test(&core_state->nr_threads))
444 complete(&core_state->startup);
445
446 for (;;) {
447 set_current_state(TASK_IDLE|TASK_FREEZABLE);
448 if (!self.task) /* see coredump_finish() */
449 break;
450 schedule();
451 }
452 __set_current_state(TASK_RUNNING);
453}
454
455#ifdef CONFIG_MEMCG
456/* drops tasklist_lock if succeeds */
457static bool __try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm)
458{
459 bool ret = false;
460
461 task_lock(tsk);
462 if (likely(tsk->mm == mm)) {
463 /* tsk can't pass exit_mm/exec_mmap and exit */
464 read_unlock(&tasklist_lock);
465 WRITE_ONCE(mm->owner, tsk);
466 lru_gen_migrate_mm(mm);
467 ret = true;
468 }
469 task_unlock(tsk);
470 return ret;
471}
472
473static bool try_to_set_owner(struct task_struct *g, struct mm_struct *mm)
474{
475 struct task_struct *t;
476
477 for_each_thread(g, t) {
478 struct mm_struct *t_mm = READ_ONCE(t->mm);
479 if (t_mm == mm) {
480 if (__try_to_set_owner(t, mm))
481 return true;
482 } else if (t_mm)
483 break;
484 }
485
486 return false;
487}
488
489/*
490 * A task is exiting. If it owned this mm, find a new owner for the mm.
491 */
492void mm_update_next_owner(struct mm_struct *mm)
493{
494 struct task_struct *g, *p = current;
495
496 /*
497 * If the exiting or execing task is not the owner, it's
498 * someone else's problem.
499 */
500 if (mm->owner != p)
501 return;
502 /*
503 * The current owner is exiting/execing and there are no other
504 * candidates. Do not leave the mm pointing to a possibly
505 * freed task structure.
506 */
507 if (atomic_read(&mm->mm_users) <= 1) {
508 WRITE_ONCE(mm->owner, NULL);
509 return;
510 }
511
512 read_lock(&tasklist_lock);
513 /*
514 * Search in the children
515 */
516 list_for_each_entry(g, &p->children, sibling) {
517 if (try_to_set_owner(g, mm))
518 goto ret;
519 }
520 /*
521 * Search in the siblings
522 */
523 list_for_each_entry(g, &p->real_parent->children, sibling) {
524 if (try_to_set_owner(g, mm))
525 goto ret;
526 }
527 /*
528 * Search through everything else, we should not get here often.
529 */
530 for_each_process(g) {
531 if (atomic_read(&mm->mm_users) <= 1)
532 break;
533 if (g->flags & PF_KTHREAD)
534 continue;
535 if (try_to_set_owner(g, mm))
536 goto ret;
537 }
538 read_unlock(&tasklist_lock);
539 /*
540 * We found no owner yet mm_users > 1: this implies that we are
541 * most likely racing with swapoff (try_to_unuse()) or /proc or
542 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
543 */
544 WRITE_ONCE(mm->owner, NULL);
545 ret:
546 return;
547
548}
549#endif /* CONFIG_MEMCG */
550
551#if defined(CONFIG_SCHED_CACHE) && defined(CONFIG_NUMA_BALANCING)
552/*
553 * Subtract the memory footprint of the current task from
554 * mm.
555 */
556static void exit_mm_sched_cache(struct mm_struct *mm)
557{
558 unsigned long fp, sub;
559
560 if (!current->total_numa_faults)
561 return;
562 /*
563 * No lock protection due to performance considerations.
564 * Make sure mm->sc_stat.footprint does not become
565 * negative.
566 */
567 fp = READ_ONCE(mm->sc_stat.footprint);
568 sub = min(fp, current->total_numa_faults);
569 WRITE_ONCE(mm->sc_stat.footprint, fp - sub);
570}
571#else
572static inline void exit_mm_sched_cache(struct mm_struct *mm)
573{
574}
575#endif /* CONFIG_SCHED_CACHE CONFIG_NUMA_BALANCING */
576
577/*
578 * Turn us into a lazy TLB process if we
579 * aren't already..
580 */
581static void exit_mm(void)
582{
583 struct mm_struct *mm = current->mm;
584
585 exit_mm_release(current, mm);
586 if (!mm)
587 return;
588
589 exit_mm_sched_cache(mm);
590
591 mmap_read_lock(mm);
592 mmgrab_lazy_tlb(mm);
593 BUG_ON(mm != current->active_mm);
594 /* more a memory barrier than a real lock */
595 task_lock(current);
596 /*
597 * When a thread stops operating on an address space, the loop
598 * in membarrier_private_expedited() may not observe that
599 * tsk->mm, and the loop in membarrier_global_expedited() may
600 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
601 * rq->membarrier_state, so those would not issue an IPI.
602 * Membarrier requires a memory barrier after accessing
603 * user-space memory, before clearing tsk->mm or the
604 * rq->membarrier_state.
605 */
606 smp_mb__after_spinlock();
607 local_irq_disable();
608 current->mm = NULL;
609 membarrier_update_current_mm(NULL);
610 enter_lazy_tlb(mm, current);
611 local_irq_enable();
612 task_unlock(current);
613 mmap_read_unlock(mm);
614 mm_update_next_owner(mm);
615 mmput(mm);
616 if (test_thread_flag(TIF_MEMDIE))
617 exit_oom_victim();
618}
619
620static struct task_struct *find_alive_thread(struct task_struct *p)
621{
622 struct task_struct *t;
623
624 for_each_thread(p, t) {
625 if (!(t->flags & PF_EXITING))
626 return t;
627 }
628 return NULL;
629}
630
631static struct task_struct *find_child_reaper(struct task_struct *father,
632 struct list_head *dead)
633 __releases(&tasklist_lock)
634 __acquires(&tasklist_lock)
635{
636 struct pid_namespace *pid_ns = task_active_pid_ns(father);
637 struct task_struct *reaper = pid_ns->child_reaper;
638 struct task_struct *p, *n;
639
640 if (likely(reaper != father))
641 return reaper;
642
643 reaper = find_alive_thread(father);
644 if (reaper) {
645 ASSERT_EXCLUSIVE_WRITER(pid_ns->child_reaper);
646 WRITE_ONCE(pid_ns->child_reaper, reaper);
647 return reaper;
648 }
649
650 write_unlock_irq(&tasklist_lock);
651
652 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
653 list_del_init(&p->ptrace_entry);
654 release_task(p);
655 }
656
657 zap_pid_ns_processes(pid_ns);
658 write_lock_irq(&tasklist_lock);
659
660 return father;
661}
662
663/*
664 * When we die, we re-parent all our children, and try to:
665 * 1. give them to another thread in our thread group, if such a member exists
666 * 2. give it to the first ancestor process which prctl'd itself as a
667 * child_subreaper for its children (like a service manager)
668 * 3. give it to the init process (PID 1) in our pid namespace
669 */
670static struct task_struct *find_new_reaper(struct task_struct *father,
671 struct task_struct *child_reaper)
672{
673 struct task_struct *thread, *reaper;
674
675 thread = find_alive_thread(father);
676 if (thread)
677 return thread;
678
679 if (father->signal->has_child_subreaper) {
680 unsigned int ns_level = task_pid(father)->level;
681 /*
682 * Find the first ->is_child_subreaper ancestor in our pid_ns.
683 * We can't check reaper != child_reaper to ensure we do not
684 * cross the namespaces, the exiting parent could be injected
685 * by setns() + fork().
686 * We check pid->level, this is slightly more efficient than
687 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
688 */
689 for (reaper = father->real_parent;
690 task_pid(reaper)->level == ns_level;
691 reaper = reaper->real_parent) {
692 if (reaper == &init_task)
693 break;
694 if (!reaper->signal->is_child_subreaper)
695 continue;
696 thread = find_alive_thread(reaper);
697 if (thread)
698 return thread;
699 }
700 }
701
702 return child_reaper;
703}
704
705/*
706* Any that need to be release_task'd are put on the @dead list.
707 */
708static void reparent_leader(struct task_struct *father, struct task_struct *p,
709 struct list_head *dead)
710{
711 if (unlikely(p->exit_state == EXIT_DEAD))
712 return;
713
714 /* We don't want people slaying init. */
715 p->exit_signal = SIGCHLD;
716
717 /* If it has exited notify the new parent about this child's death. */
718 if (!p->ptrace &&
719 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
720 if (do_notify_parent(p, p->exit_signal)) {
721 p->exit_state = EXIT_DEAD;
722 list_add(&p->ptrace_entry, dead);
723 }
724 }
725
726 kill_orphaned_pgrp(p, father);
727}
728
729/*
730 * Make init inherit all the child processes
731 */
732static void forget_original_parent(struct task_struct *father,
733 struct list_head *dead)
734{
735 struct task_struct *p, *t, *reaper;
736
737 if (unlikely(!list_empty(&father->ptraced)))
738 exit_ptrace(father, dead);
739
740 /* Can drop and reacquire tasklist_lock */
741 reaper = find_child_reaper(father, dead);
742 if (list_empty(&father->children))
743 return;
744
745 reaper = find_new_reaper(father, reaper);
746 list_for_each_entry(p, &father->children, sibling) {
747 for_each_thread(p, t) {
748 RCU_INIT_POINTER(t->real_parent, reaper);
749 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
750 if (likely(!t->ptrace))
751 t->parent = t->real_parent;
752 if (t->pdeath_signal)
753 group_send_sig_info(t->pdeath_signal,
754 SEND_SIG_NOINFO, t,
755 PIDTYPE_TGID);
756 }
757 /*
758 * If this is a threaded reparent there is no need to
759 * notify anyone anything has happened.
760 */
761 if (!same_thread_group(reaper, father))
762 reparent_leader(father, p, dead);
763 }
764 list_splice_tail_init(&father->children, &reaper->children);
765}
766
767/*
768 * Send signals to all our closest relatives so that they know
769 * to properly mourn us..
770 */
771static void exit_notify(struct task_struct *tsk, int group_dead)
772{
773 bool autoreap;
774 struct task_struct *p, *n;
775 LIST_HEAD(dead);
776
777 write_lock_irq(&tasklist_lock);
778 forget_original_parent(tsk, &dead);
779
780 if (group_dead)
781 kill_orphaned_pgrp(tsk->group_leader, NULL);
782
783 tsk->exit_state = EXIT_ZOMBIE;
784
785 if (unlikely(tsk->ptrace)) {
786 int sig = thread_group_empty(tsk) && !ptrace_reparented(tsk)
787 ? tsk->exit_signal : SIGCHLD;
788 autoreap = do_notify_parent(tsk, sig);
789 } else if (thread_group_leader(tsk)) {
790 autoreap = thread_group_empty(tsk) &&
791 do_notify_parent(tsk, tsk->exit_signal);
792 } else {
793 autoreap = true;
794 /* untraced sub-thread */
795 do_notify_pidfd(tsk);
796 }
797
798 if (autoreap) {
799 tsk->exit_state = EXIT_DEAD;
800 list_add(&tsk->ptrace_entry, &dead);
801 }
802
803 /* mt-exec, de_thread() is waiting for group leader */
804 if (unlikely(tsk->signal->notify_count < 0))
805 wake_up_process(tsk->signal->group_exec_task);
806 write_unlock_irq(&tasklist_lock);
807
808 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
809 list_del_init(&p->ptrace_entry);
810 release_task(p);
811 }
812}
813
814#ifdef CONFIG_DEBUG_STACK_USAGE
815#ifdef CONFIG_STACK_GROWSUP
816unsigned long stack_not_used(struct task_struct *p)
817{
818 unsigned long *n = end_of_stack(p);
819
820 do { /* Skip over canary */
821 n--;
822 } while (!*n);
823
824 return (unsigned long)end_of_stack(p) - (unsigned long)n;
825}
826#else /* !CONFIG_STACK_GROWSUP */
827unsigned long stack_not_used(struct task_struct *p)
828{
829 unsigned long *n = end_of_stack(p);
830
831 do { /* Skip over canary */
832 n++;
833 } while (!*n);
834
835 return (unsigned long)n - (unsigned long)end_of_stack(p);
836}
837#endif /* CONFIG_STACK_GROWSUP */
838
839/* Count the maximum pages reached in kernel stacks */
840static inline void kstack_histogram(unsigned long used_stack)
841{
842#ifdef CONFIG_VM_EVENT_COUNTERS
843 if (used_stack <= 1024)
844 count_vm_event(KSTACK_1K);
845#if THREAD_SIZE > 1024
846 else if (used_stack <= 2048)
847 count_vm_event(KSTACK_2K);
848#endif
849#if THREAD_SIZE > 2048
850 else if (used_stack <= 4096)
851 count_vm_event(KSTACK_4K);
852#endif
853#if THREAD_SIZE > 4096
854 else if (used_stack <= 8192)
855 count_vm_event(KSTACK_8K);
856#endif
857#if THREAD_SIZE > 8192
858 else if (used_stack <= 16384)
859 count_vm_event(KSTACK_16K);
860#endif
861#if THREAD_SIZE > 16384
862 else if (used_stack <= 32768)
863 count_vm_event(KSTACK_32K);
864#endif
865#if THREAD_SIZE > 32768
866 else if (used_stack <= 65536)
867 count_vm_event(KSTACK_64K);
868#endif
869#if THREAD_SIZE > 65536
870 else
871 count_vm_event(KSTACK_REST);
872#endif
873#endif /* CONFIG_VM_EVENT_COUNTERS */
874}
875
876static void check_stack_usage(void)
877{
878 static DEFINE_SPINLOCK(low_water_lock);
879 static int lowest_to_date = THREAD_SIZE;
880 unsigned long free;
881
882 free = stack_not_used(current);
883 kstack_histogram(THREAD_SIZE - free);
884
885 if (free >= lowest_to_date)
886 return;
887
888 spin_lock(&low_water_lock);
889 if (free < lowest_to_date) {
890 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
891 current->comm, task_pid_nr(current), free);
892 lowest_to_date = free;
893 }
894 spin_unlock(&low_water_lock);
895}
896#else /* !CONFIG_DEBUG_STACK_USAGE */
897static inline void check_stack_usage(void) {}
898#endif /* CONFIG_DEBUG_STACK_USAGE */
899
900static void synchronize_group_exit(struct task_struct *tsk, long code)
901{
902 struct sighand_struct *sighand = tsk->sighand;
903 struct signal_struct *signal = tsk->signal;
904 struct core_state *core_state;
905
906 spin_lock_irq(&sighand->siglock);
907 signal->quick_threads--;
908 if ((signal->quick_threads == 0) &&
909 !(signal->flags & SIGNAL_GROUP_EXIT)) {
910 signal->flags = SIGNAL_GROUP_EXIT;
911 signal->group_exit_code = code;
912 signal->group_stop_count = 0;
913 }
914 /*
915 * Serialize with any possible pending coredump.
916 * We must hold siglock around checking core_state
917 * and setting PF_POSTCOREDUMP. The core-inducing thread
918 * will increment ->nr_threads for each thread in the
919 * group without PF_POSTCOREDUMP set.
920 */
921 tsk->flags |= PF_POSTCOREDUMP;
922 core_state = signal->core_state;
923 spin_unlock_irq(&sighand->siglock);
924
925 if (unlikely(core_state))
926 coredump_task_exit(tsk, core_state);
927}
928
929void __noreturn do_exit(long code)
930{
931 struct task_struct *tsk = current;
932 struct kthread *kthread;
933 int group_dead;
934
935 WARN_ON(irqs_disabled());
936 WARN_ON(tsk->plug);
937
938 kthread = tsk_is_kthread(tsk);
939 if (unlikely(kthread))
940 kthread_do_exit(kthread, code);
941
942 kcov_task_exit(tsk);
943 kmsan_task_exit(tsk);
944
945 synchronize_group_exit(tsk, code);
946 ptrace_event(PTRACE_EVENT_EXIT, code);
947 user_events_exit(tsk);
948
949 io_uring_files_cancel();
950 sched_mm_cid_exit(tsk);
951 exit_signals(tsk); /* sets PF_EXITING */
952
953 seccomp_filter_release(tsk);
954
955 acct_update_integrals(tsk);
956 group_dead = atomic_dec_and_test(&tsk->signal->live);
957 if (group_dead) {
958 /*
959 * If the last thread of global init has exited, panic
960 * immediately to get a useable coredump.
961 */
962 if (unlikely(is_global_init(tsk)))
963 panic("Attempted to kill init! exitcode=0x%08x\n",
964 tsk->signal->group_exit_code ?: (int)code);
965
966#ifdef CONFIG_POSIX_TIMERS
967 hrtimer_cancel(&tsk->signal->real_timer);
968 exit_itimers(tsk);
969#endif
970 if (tsk->mm)
971 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
972 }
973 acct_collect(code, group_dead);
974 if (group_dead)
975 tty_audit_exit();
976 audit_free(tsk);
977
978 tsk->exit_code = code;
979 taskstats_exit(tsk, group_dead);
980 trace_sched_process_exit(tsk, group_dead);
981
982 /*
983 * Since sampling can touch ->mm, make sure to stop everything before we
984 * tear it down.
985 *
986 * Also flushes inherited counters to the parent - before the parent
987 * gets woken up by child-exit notifications.
988 */
989 perf_event_exit_task(tsk);
990 /*
991 * PF_EXITING (above) ensures unwind_deferred_request() will no
992 * longer add new unwinds. While exit_mm() (below) will destroy the
993 * abaility to do unwinds. So flush any pending unwinds here.
994 */
995 unwind_deferred_task_exit(tsk);
996
997 exit_mm();
998
999 if (group_dead)
1000 acct_process();
1001
1002 exit_sem(tsk);
1003 exit_shm(tsk);
1004 exit_files(tsk);
1005 exit_fs(tsk);
1006 if (group_dead)
1007 disassociate_ctty(1);
1008 exit_nsproxy_namespaces(tsk);
1009 exit_task_work(tsk);
1010 exit_thread(tsk);
1011
1012 sched_autogroup_exit_task(tsk);
1013 cgroup_task_exit(tsk);
1014
1015 /*
1016 * FIXME: do that only when needed, using sched_exit tracepoint
1017 */
1018 flush_ptrace_hw_breakpoint(tsk);
1019
1020 exit_tasks_rcu_start();
1021 exit_notify(tsk, group_dead);
1022 proc_exit_connector(tsk);
1023 mpol_put_task_policy(tsk);
1024#ifdef CONFIG_FUTEX
1025 if (unlikely(current->futex.pi_state_cache))
1026 kfree(current->futex.pi_state_cache);
1027#endif
1028 /*
1029 * Make sure we are holding no locks:
1030 */
1031 debug_check_no_locks_held();
1032
1033 if (tsk->io_context)
1034 exit_io_context(tsk);
1035
1036 if (tsk->splice_pipe)
1037 free_pipe_info(tsk->splice_pipe);
1038
1039 if (tsk->task_frag.page)
1040 put_page(tsk->task_frag.page);
1041
1042 exit_task_stack_account(tsk);
1043
1044 check_stack_usage();
1045 preempt_disable();
1046 if (tsk->nr_dirtied)
1047 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
1048 exit_rcu();
1049 exit_tasks_rcu_finish();
1050
1051 lockdep_free_task(tsk);
1052 do_task_dead();
1053}
1054EXPORT_SYMBOL(do_exit);
1055
1056void __noreturn make_task_dead(int signr)
1057{
1058 /*
1059 * Take the task off the cpu after something catastrophic has
1060 * happened.
1061 *
1062 * We can get here from a kernel oops, sometimes with preemption off.
1063 * Start by checking for critical errors.
1064 * Then fix up important state like USER_DS and preemption.
1065 * Then do everything else.
1066 */
1067 struct task_struct *tsk = current;
1068 unsigned int limit;
1069
1070 if (unlikely(in_interrupt()))
1071 panic("Aiee, killing interrupt handler!");
1072 if (unlikely(!tsk->pid))
1073 panic("Attempted to kill the idle task!");
1074
1075 if (unlikely(irqs_disabled())) {
1076 pr_info("note: %s[%d] exited with irqs disabled\n",
1077 current->comm, task_pid_nr(current));
1078 local_irq_enable();
1079 }
1080 if (unlikely(in_atomic())) {
1081 pr_info("note: %s[%d] exited with preempt_count %d\n",
1082 current->comm, task_pid_nr(current),
1083 preempt_count());
1084 preempt_count_set(PREEMPT_ENABLED);
1085 }
1086
1087 /*
1088 * Every time the system oopses, if the oops happens while a reference
1089 * to an object was held, the reference leaks.
1090 * If the oops doesn't also leak memory, repeated oopsing can cause
1091 * reference counters to wrap around (if they're not using refcount_t).
1092 * This means that repeated oopsing can make unexploitable-looking bugs
1093 * exploitable through repeated oopsing.
1094 * To make sure this can't happen, place an upper bound on how often the
1095 * kernel may oops without panic().
1096 */
1097 limit = READ_ONCE(oops_limit);
1098 if (atomic_inc_return(&oops_count) >= limit && limit)
1099 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
1100
1101 /*
1102 * We're taking recursive faults here in make_task_dead. Safest is to just
1103 * leave this task alone and wait for reboot.
1104 */
1105 if (unlikely(tsk->flags & PF_EXITING)) {
1106 pr_alert("Fixing recursive fault but reboot is needed!\n");
1107 futex_exit_recursive(tsk);
1108 tsk->exit_state = EXIT_DEAD;
1109 refcount_inc(&tsk->rcu_users);
1110 preempt_disable();
1111 do_task_dead();
1112 }
1113
1114 do_exit(signr);
1115}
1116
1117SYSCALL_DEFINE1(exit, int, error_code)
1118{
1119 do_exit((error_code&0xff)<<8);
1120}
1121
1122/*
1123 * Take down every thread in the group. This is called by fatal signals
1124 * as well as by sys_exit_group (below).
1125 */
1126void __noreturn
1127do_group_exit(int exit_code)
1128{
1129 struct signal_struct *sig = current->signal;
1130
1131 if (sig->flags & SIGNAL_GROUP_EXIT)
1132 exit_code = sig->group_exit_code;
1133 else if (sig->group_exec_task)
1134 exit_code = 0;
1135 else {
1136 struct sighand_struct *const sighand = current->sighand;
1137
1138 spin_lock_irq(&sighand->siglock);
1139 if (sig->flags & SIGNAL_GROUP_EXIT)
1140 /* Another thread got here before we took the lock. */
1141 exit_code = sig->group_exit_code;
1142 else if (sig->group_exec_task)
1143 exit_code = 0;
1144 else {
1145 sig->group_exit_code = exit_code;
1146 sig->flags = SIGNAL_GROUP_EXIT;
1147 zap_other_threads(current);
1148 }
1149 spin_unlock_irq(&sighand->siglock);
1150 }
1151
1152 do_exit(exit_code);
1153 /* NOTREACHED */
1154}
1155
1156/*
1157 * this kills every thread in the thread group. Note that any externally
1158 * wait4()-ing process will get the correct exit code - even if this
1159 * thread is not the thread group leader.
1160 */
1161SYSCALL_DEFINE1(exit_group, int, error_code)
1162{
1163 do_group_exit((error_code & 0xff) << 8);
1164 /* NOTREACHED */
1165 return 0;
1166}
1167
1168static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1169{
1170 return wo->wo_type == PIDTYPE_MAX ||
1171 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1172}
1173
1174static int
1175eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1176{
1177 if (!eligible_pid(wo, p))
1178 return 0;
1179
1180 /*
1181 * Wait for all children (clone and not) if __WALL is set or
1182 * if it is traced by us.
1183 */
1184 if (ptrace || (wo->wo_flags & __WALL))
1185 return 1;
1186
1187 /*
1188 * Otherwise, wait for clone children *only* if __WCLONE is set;
1189 * otherwise, wait for non-clone children *only*.
1190 *
1191 * Note: a "clone" child here is one that reports to its parent
1192 * using a signal other than SIGCHLD, or a non-leader thread which
1193 * we can only see if it is traced by us.
1194 */
1195 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1196 return 0;
1197
1198 return 1;
1199}
1200
1201/*
1202 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1203 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1204 * the lock and this task is uninteresting. If we return nonzero, we have
1205 * released the lock and the system call should return.
1206 */
1207static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1208{
1209 int state, status;
1210 pid_t pid = task_pid_vnr(p);
1211 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1212 struct waitid_info *infop;
1213
1214 if (!likely(wo->wo_flags & WEXITED))
1215 return 0;
1216
1217 if (unlikely(wo->wo_flags & WNOWAIT)) {
1218 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1219 ? p->signal->group_exit_code : p->exit_code;
1220 get_task_struct(p);
1221 read_unlock(&tasklist_lock);
1222 sched_annotate_sleep();
1223 if (wo->wo_rusage)
1224 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1225 put_task_struct(p);
1226 goto out_info;
1227 }
1228 /*
1229 * Move the task's state to DEAD/TRACE, only one thread can do this.
1230 */
1231 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1232 EXIT_TRACE : EXIT_DEAD;
1233 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1234 return 0;
1235 /*
1236 * We own this thread, nobody else can reap it.
1237 */
1238 read_unlock(&tasklist_lock);
1239 sched_annotate_sleep();
1240
1241 /*
1242 * Check thread_group_leader() to exclude the traced sub-threads.
1243 */
1244 if (state == EXIT_DEAD && thread_group_leader(p)) {
1245 struct signal_struct *sig = p->signal;
1246 struct signal_struct *psig = current->signal;
1247 unsigned long maxrss;
1248 u64 tgutime, tgstime;
1249
1250 /*
1251 * The resource counters for the group leader are in its
1252 * own task_struct. Those for dead threads in the group
1253 * are in its signal_struct, as are those for the child
1254 * processes it has previously reaped. All these
1255 * accumulate in the parent's signal_struct c* fields.
1256 *
1257 * We don't bother to take a lock here to protect these
1258 * p->signal fields because the whole thread group is dead
1259 * and nobody can change them.
1260 *
1261 * psig->stats_lock also protects us from our sub-threads
1262 * which can reap other children at the same time.
1263 *
1264 * We use thread_group_cputime_adjusted() to get times for
1265 * the thread group, which consolidates times for all threads
1266 * in the group including the group leader.
1267 */
1268 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1269 write_seqlock_irq(&psig->stats_lock);
1270 psig->cutime += tgutime + sig->cutime;
1271 psig->cstime += tgstime + sig->cstime;
1272 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1273 psig->cmin_flt +=
1274 p->min_flt + sig->min_flt + sig->cmin_flt;
1275 psig->cmaj_flt +=
1276 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1277 psig->cnvcsw +=
1278 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1279 psig->cnivcsw +=
1280 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1281 psig->cinblock +=
1282 task_io_get_inblock(p) +
1283 sig->inblock + sig->cinblock;
1284 psig->coublock +=
1285 task_io_get_oublock(p) +
1286 sig->oublock + sig->coublock;
1287 maxrss = max(sig->maxrss, sig->cmaxrss);
1288 if (psig->cmaxrss < maxrss)
1289 psig->cmaxrss = maxrss;
1290 task_io_accounting_add(&psig->ioac, &p->ioac);
1291 task_io_accounting_add(&psig->ioac, &sig->ioac);
1292 write_sequnlock_irq(&psig->stats_lock);
1293 }
1294
1295 if (wo->wo_rusage)
1296 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1297 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1298 ? p->signal->group_exit_code : p->exit_code;
1299 wo->wo_stat = status;
1300
1301 if (state == EXIT_TRACE) {
1302 write_lock_irq(&tasklist_lock);
1303 /* We dropped tasklist, ptracer could die and untrace */
1304 ptrace_unlink(p);
1305
1306 /* If parent wants a zombie, don't release it now */
1307 state = EXIT_ZOMBIE;
1308 if (do_notify_parent(p, p->exit_signal))
1309 state = EXIT_DEAD;
1310 p->exit_state = state;
1311 write_unlock_irq(&tasklist_lock);
1312 }
1313 if (state == EXIT_DEAD)
1314 release_task(p);
1315
1316out_info:
1317 infop = wo->wo_info;
1318 if (infop) {
1319 if ((status & 0x7f) == 0) {
1320 infop->cause = CLD_EXITED;
1321 infop->status = status >> 8;
1322 } else {
1323 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1324 infop->status = status & 0x7f;
1325 }
1326 infop->pid = pid;
1327 infop->uid = uid;
1328 }
1329
1330 return pid;
1331}
1332
1333static int *task_stopped_code(struct task_struct *p, bool ptrace)
1334{
1335 if (ptrace) {
1336 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1337 return &p->exit_code;
1338 } else {
1339 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1340 return &p->signal->group_exit_code;
1341 }
1342 return NULL;
1343}
1344
1345/**
1346 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1347 * @wo: wait options
1348 * @ptrace: is the wait for ptrace
1349 * @p: task to wait for
1350 *
1351 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1352 *
1353 * CONTEXT:
1354 * read_lock(&tasklist_lock), which is released if return value is
1355 * non-zero. Also, grabs and releases @p->sighand->siglock.
1356 *
1357 * RETURNS:
1358 * 0 if wait condition didn't exist and search for other wait conditions
1359 * should continue. Non-zero return, -errno on failure and @p's pid on
1360 * success, implies that tasklist_lock is released and wait condition
1361 * search should terminate.
1362 */
1363static int wait_task_stopped(struct wait_opts *wo,
1364 int ptrace, struct task_struct *p)
1365{
1366 struct waitid_info *infop;
1367 int exit_code, *p_code, why;
1368 uid_t uid = 0; /* unneeded, required by compiler */
1369 pid_t pid;
1370
1371 /*
1372 * Traditionally we see ptrace'd stopped tasks regardless of options.
1373 */
1374 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1375 return 0;
1376
1377 if (!task_stopped_code(p, ptrace))
1378 return 0;
1379
1380 exit_code = 0;
1381 spin_lock_irq(&p->sighand->siglock);
1382
1383 p_code = task_stopped_code(p, ptrace);
1384 if (unlikely(!p_code))
1385 goto unlock_sig;
1386
1387 exit_code = *p_code;
1388 if (!exit_code)
1389 goto unlock_sig;
1390
1391 if (!unlikely(wo->wo_flags & WNOWAIT))
1392 *p_code = 0;
1393
1394 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1395unlock_sig:
1396 spin_unlock_irq(&p->sighand->siglock);
1397 if (!exit_code)
1398 return 0;
1399
1400 /*
1401 * Now we are pretty sure this task is interesting.
1402 * Make sure it doesn't get reaped out from under us while we
1403 * give up the lock and then examine it below. We don't want to
1404 * keep holding onto the tasklist_lock while we call getrusage and
1405 * possibly take page faults for user memory.
1406 */
1407 get_task_struct(p);
1408 pid = task_pid_vnr(p);
1409 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1410 read_unlock(&tasklist_lock);
1411 sched_annotate_sleep();
1412 if (wo->wo_rusage)
1413 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1414 put_task_struct(p);
1415
1416 if (likely(!(wo->wo_flags & WNOWAIT)))
1417 wo->wo_stat = (exit_code << 8) | 0x7f;
1418
1419 infop = wo->wo_info;
1420 if (infop) {
1421 infop->cause = why;
1422 infop->status = exit_code;
1423 infop->pid = pid;
1424 infop->uid = uid;
1425 }
1426 return pid;
1427}
1428
1429/*
1430 * Handle do_wait work for one task in a live, non-stopped state.
1431 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1432 * the lock and this task is uninteresting. If we return nonzero, we have
1433 * released the lock and the system call should return.
1434 */
1435static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1436{
1437 struct waitid_info *infop;
1438 pid_t pid;
1439 uid_t uid;
1440
1441 if (!unlikely(wo->wo_flags & WCONTINUED))
1442 return 0;
1443
1444 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1445 return 0;
1446
1447 spin_lock_irq(&p->sighand->siglock);
1448 /* Re-check with the lock held. */
1449 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1450 spin_unlock_irq(&p->sighand->siglock);
1451 return 0;
1452 }
1453 if (!unlikely(wo->wo_flags & WNOWAIT))
1454 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1455 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1456 spin_unlock_irq(&p->sighand->siglock);
1457
1458 pid = task_pid_vnr(p);
1459 get_task_struct(p);
1460 read_unlock(&tasklist_lock);
1461 sched_annotate_sleep();
1462 if (wo->wo_rusage)
1463 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1464 put_task_struct(p);
1465
1466 infop = wo->wo_info;
1467 if (!infop) {
1468 wo->wo_stat = 0xffff;
1469 } else {
1470 infop->cause = CLD_CONTINUED;
1471 infop->pid = pid;
1472 infop->uid = uid;
1473 infop->status = SIGCONT;
1474 }
1475 return pid;
1476}
1477
1478/*
1479 * Consider @p for a wait by @parent.
1480 *
1481 * -ECHILD should be in ->notask_error before the first call.
1482 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1483 * Returns zero if the search for a child should continue;
1484 * then ->notask_error is 0 if @p is an eligible child,
1485 * or still -ECHILD.
1486 */
1487static int wait_consider_task(struct wait_opts *wo, int ptrace,
1488 struct task_struct *p)
1489{
1490 /*
1491 * We can race with wait_task_zombie() from another thread.
1492 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1493 * can't confuse the checks below.
1494 */
1495 int exit_state = READ_ONCE(p->exit_state);
1496 int ret;
1497
1498 if (unlikely(exit_state == EXIT_DEAD))
1499 return 0;
1500
1501 ret = eligible_child(wo, ptrace, p);
1502 if (!ret)
1503 return ret;
1504
1505 if (unlikely(exit_state == EXIT_TRACE)) {
1506 /*
1507 * ptrace == 0 means we are the natural parent. In this case
1508 * we should clear notask_error, debugger will notify us.
1509 */
1510 if (likely(!ptrace))
1511 wo->notask_error = 0;
1512 return 0;
1513 }
1514
1515 if (likely(!ptrace) && unlikely(p->ptrace)) {
1516 /*
1517 * If it is traced by its real parent's group, just pretend
1518 * the caller is ptrace_do_wait() and reap this child if it
1519 * is zombie.
1520 *
1521 * This also hides group stop state from real parent; otherwise
1522 * a single stop can be reported twice as group and ptrace stop.
1523 * If a ptracer wants to distinguish these two events for its
1524 * own children it should create a separate process which takes
1525 * the role of real parent.
1526 */
1527 if (!ptrace_reparented(p))
1528 ptrace = 1;
1529 }
1530
1531 /* slay zombie? */
1532 if (exit_state == EXIT_ZOMBIE) {
1533 /* we don't reap group leaders with subthreads */
1534 if (!delay_group_leader(p)) {
1535 /*
1536 * A zombie ptracee is only visible to its ptracer.
1537 * Notification and reaping will be cascaded to the
1538 * real parent when the ptracer detaches.
1539 */
1540 if (unlikely(ptrace) || likely(!p->ptrace))
1541 return wait_task_zombie(wo, p);
1542 }
1543
1544 /*
1545 * Allow access to stopped/continued state via zombie by
1546 * falling through. Clearing of notask_error is complex.
1547 *
1548 * When !@ptrace:
1549 *
1550 * If WEXITED is set, notask_error should naturally be
1551 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1552 * so, if there are live subthreads, there are events to
1553 * wait for. If all subthreads are dead, it's still safe
1554 * to clear - this function will be called again in finite
1555 * amount time once all the subthreads are released and
1556 * will then return without clearing.
1557 *
1558 * When @ptrace:
1559 *
1560 * Stopped state is per-task and thus can't change once the
1561 * target task dies. Only continued and exited can happen.
1562 * Clear notask_error if WCONTINUED | WEXITED.
1563 */
1564 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1565 wo->notask_error = 0;
1566 } else {
1567 /*
1568 * @p is alive and it's gonna stop, continue or exit, so
1569 * there always is something to wait for.
1570 */
1571 wo->notask_error = 0;
1572 }
1573
1574 /*
1575 * Wait for stopped. Depending on @ptrace, different stopped state
1576 * is used and the two don't interact with each other.
1577 */
1578 ret = wait_task_stopped(wo, ptrace, p);
1579 if (ret)
1580 return ret;
1581
1582 /*
1583 * Wait for continued. There's only one continued state and the
1584 * ptracer can consume it which can confuse the real parent. Don't
1585 * use WCONTINUED from ptracer. You don't need or want it.
1586 */
1587 return wait_task_continued(wo, p);
1588}
1589
1590/*
1591 * Do the work of do_wait() for one thread in the group, @tsk.
1592 *
1593 * -ECHILD should be in ->notask_error before the first call.
1594 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1595 * Returns zero if the search for a child should continue; then
1596 * ->notask_error is 0 if there were any eligible children,
1597 * or still -ECHILD.
1598 */
1599static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1600{
1601 struct task_struct *p;
1602
1603 list_for_each_entry(p, &tsk->children, sibling) {
1604 int ret = wait_consider_task(wo, 0, p);
1605
1606 if (ret)
1607 return ret;
1608 }
1609
1610 return 0;
1611}
1612
1613static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1614{
1615 struct task_struct *p;
1616
1617 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1618 int ret = wait_consider_task(wo, 1, p);
1619
1620 if (ret)
1621 return ret;
1622 }
1623
1624 return 0;
1625}
1626
1627bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1628{
1629 if (!eligible_pid(wo, p))
1630 return false;
1631
1632 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1633 return false;
1634
1635 return true;
1636}
1637
1638static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1639 int sync, void *key)
1640{
1641 struct wait_opts *wo = container_of(wait, struct wait_opts,
1642 child_wait);
1643 struct task_struct *p = key;
1644
1645 if (pid_child_should_wake(wo, p))
1646 return default_wake_function(wait, mode, sync, key);
1647
1648 return 0;
1649}
1650
1651void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1652{
1653 __wake_up_sync_key(&parent->signal->wait_chldexit,
1654 TASK_INTERRUPTIBLE, p);
1655}
1656
1657static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1658 struct task_struct *target)
1659{
1660 struct task_struct *parent =
1661 !ptrace ? target->real_parent : target->parent;
1662
1663 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1664 same_thread_group(current, parent));
1665}
1666
1667/*
1668 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1669 * and tracee lists to find the target task.
1670 */
1671static int do_wait_pid(struct wait_opts *wo)
1672{
1673 bool ptrace;
1674 struct task_struct *target;
1675 int retval;
1676
1677 ptrace = false;
1678 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1679 if (target && is_effectively_child(wo, ptrace, target)) {
1680 retval = wait_consider_task(wo, ptrace, target);
1681 if (retval)
1682 return retval;
1683 }
1684
1685 ptrace = true;
1686 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1687 if (target && target->ptrace &&
1688 is_effectively_child(wo, ptrace, target)) {
1689 retval = wait_consider_task(wo, ptrace, target);
1690 if (retval)
1691 return retval;
1692 }
1693
1694 return 0;
1695}
1696
1697long __do_wait(struct wait_opts *wo)
1698{
1699 long retval;
1700
1701 /*
1702 * If there is nothing that can match our criteria, just get out.
1703 * We will clear ->notask_error to zero if we see any child that
1704 * might later match our criteria, even if we are not able to reap
1705 * it yet.
1706 */
1707 wo->notask_error = -ECHILD;
1708 if ((wo->wo_type < PIDTYPE_MAX) &&
1709 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1710 goto notask;
1711
1712 read_lock(&tasklist_lock);
1713
1714 if (wo->wo_type == PIDTYPE_PID) {
1715 retval = do_wait_pid(wo);
1716 if (retval)
1717 return retval;
1718 } else {
1719 struct task_struct *tsk = current;
1720
1721 do {
1722 retval = do_wait_thread(wo, tsk);
1723 if (retval)
1724 return retval;
1725
1726 retval = ptrace_do_wait(wo, tsk);
1727 if (retval)
1728 return retval;
1729
1730 if (wo->wo_flags & __WNOTHREAD)
1731 break;
1732 } while_each_thread(current, tsk);
1733 }
1734 read_unlock(&tasklist_lock);
1735
1736notask:
1737 retval = wo->notask_error;
1738 if (!retval && !(wo->wo_flags & WNOHANG))
1739 return -ERESTARTSYS;
1740
1741 return retval;
1742}
1743
1744static long do_wait(struct wait_opts *wo)
1745{
1746 int retval;
1747
1748 trace_sched_process_wait(wo->wo_pid);
1749
1750 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1751 wo->child_wait.private = current;
1752 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1753
1754 do {
1755 set_current_state(TASK_INTERRUPTIBLE);
1756 retval = __do_wait(wo);
1757 if (retval != -ERESTARTSYS)
1758 break;
1759 if (signal_pending(current))
1760 break;
1761 schedule();
1762 } while (1);
1763
1764 __set_current_state(TASK_RUNNING);
1765 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1766 return retval;
1767}
1768
1769int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1770 struct waitid_info *infop, int options,
1771 struct rusage *ru)
1772{
1773 unsigned int f_flags = 0;
1774 struct pid *pid = NULL;
1775 enum pid_type type;
1776
1777 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1778 __WNOTHREAD|__WCLONE|__WALL))
1779 return -EINVAL;
1780 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1781 return -EINVAL;
1782
1783 switch (which) {
1784 case P_ALL:
1785 type = PIDTYPE_MAX;
1786 break;
1787 case P_PID:
1788 type = PIDTYPE_PID;
1789 if (upid <= 0)
1790 return -EINVAL;
1791
1792 pid = find_get_pid(upid);
1793 break;
1794 case P_PGID:
1795 type = PIDTYPE_PGID;
1796 if (upid < 0)
1797 return -EINVAL;
1798
1799 if (upid)
1800 pid = find_get_pid(upid);
1801 else
1802 pid = get_task_pid(current, PIDTYPE_PGID);
1803 break;
1804 case P_PIDFD:
1805 type = PIDTYPE_PID;
1806 if (upid < 0)
1807 return -EINVAL;
1808
1809 pid = pidfd_get_pid(upid, &f_flags);
1810 if (IS_ERR(pid))
1811 return PTR_ERR(pid);
1812
1813 break;
1814 default:
1815 return -EINVAL;
1816 }
1817
1818 wo->wo_type = type;
1819 wo->wo_pid = pid;
1820 wo->wo_flags = options;
1821 wo->wo_info = infop;
1822 wo->wo_rusage = ru;
1823 if (f_flags & O_NONBLOCK)
1824 wo->wo_flags |= WNOHANG;
1825
1826 return 0;
1827}
1828
1829static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1830 int options, struct rusage *ru)
1831{
1832 struct wait_opts wo;
1833 long ret;
1834
1835 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1836 if (ret)
1837 return ret;
1838
1839 ret = do_wait(&wo);
1840 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1841 ret = -EAGAIN;
1842
1843 put_pid(wo.wo_pid);
1844 return ret;
1845}
1846
1847SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1848 infop, int, options, struct rusage __user *, ru)
1849{
1850 struct rusage r;
1851 struct waitid_info info = {.status = 0};
1852 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1853 int signo = 0;
1854
1855 if (err > 0) {
1856 signo = SIGCHLD;
1857 err = 0;
1858 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1859 return -EFAULT;
1860 }
1861 if (!infop)
1862 return err;
1863
1864 if (!user_write_access_begin(infop, sizeof(*infop)))
1865 return -EFAULT;
1866
1867 unsafe_put_user(signo, &infop->si_signo, Efault);
1868 unsafe_put_user(0, &infop->si_errno, Efault);
1869 unsafe_put_user(info.cause, &infop->si_code, Efault);
1870 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1871 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1872 unsafe_put_user(info.status, &infop->si_status, Efault);
1873 user_write_access_end();
1874 return err;
1875Efault:
1876 user_write_access_end();
1877 return -EFAULT;
1878}
1879
1880long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1881 struct rusage *ru)
1882{
1883 struct wait_opts wo;
1884 struct pid *pid = NULL;
1885 enum pid_type type;
1886 long ret;
1887
1888 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1889 __WNOTHREAD|__WCLONE|__WALL))
1890 return -EINVAL;
1891
1892 /* -INT_MIN is not defined */
1893 if (upid == INT_MIN)
1894 return -ESRCH;
1895
1896 if (upid == -1)
1897 type = PIDTYPE_MAX;
1898 else if (upid < 0) {
1899 type = PIDTYPE_PGID;
1900 pid = find_get_pid(-upid);
1901 } else if (upid == 0) {
1902 type = PIDTYPE_PGID;
1903 pid = get_task_pid(current, PIDTYPE_PGID);
1904 } else /* upid > 0 */ {
1905 type = PIDTYPE_PID;
1906 pid = find_get_pid(upid);
1907 }
1908
1909 wo.wo_type = type;
1910 wo.wo_pid = pid;
1911 wo.wo_flags = options | WEXITED;
1912 wo.wo_info = NULL;
1913 wo.wo_stat = 0;
1914 wo.wo_rusage = ru;
1915 ret = do_wait(&wo);
1916 put_pid(pid);
1917 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1918 ret = -EFAULT;
1919
1920 return ret;
1921}
1922
1923int kernel_wait(pid_t pid, int *stat)
1924{
1925 struct wait_opts wo = {
1926 .wo_type = PIDTYPE_PID,
1927 .wo_pid = find_get_pid(pid),
1928 .wo_flags = WEXITED,
1929 };
1930 int ret;
1931
1932 ret = do_wait(&wo);
1933 if (ret > 0 && wo.wo_stat)
1934 *stat = wo.wo_stat;
1935 put_pid(wo.wo_pid);
1936 return ret;
1937}
1938
1939SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1940 int, options, struct rusage __user *, ru)
1941{
1942 struct rusage r;
1943 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1944
1945 if (err > 0) {
1946 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1947 return -EFAULT;
1948 }
1949 return err;
1950}
1951
1952#ifdef __ARCH_WANT_SYS_WAITPID
1953
1954/*
1955 * sys_waitpid() remains for compatibility. waitpid() should be
1956 * implemented by calling sys_wait4() from libc.a.
1957 */
1958SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1959{
1960 return kernel_wait4(pid, stat_addr, options, NULL);
1961}
1962
1963#endif
1964
1965#ifdef CONFIG_COMPAT
1966COMPAT_SYSCALL_DEFINE4(wait4,
1967 compat_pid_t, pid,
1968 compat_uint_t __user *, stat_addr,
1969 int, options,
1970 struct compat_rusage __user *, ru)
1971{
1972 struct rusage r;
1973 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1974 if (err > 0) {
1975 if (ru && put_compat_rusage(&r, ru))
1976 return -EFAULT;
1977 }
1978 return err;
1979}
1980
1981COMPAT_SYSCALL_DEFINE5(waitid,
1982 int, which, compat_pid_t, pid,
1983 struct compat_siginfo __user *, infop, int, options,
1984 struct compat_rusage __user *, uru)
1985{
1986 struct rusage ru;
1987 struct waitid_info info = {.status = 0};
1988 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1989 int signo = 0;
1990 if (err > 0) {
1991 signo = SIGCHLD;
1992 err = 0;
1993 if (uru) {
1994 /* kernel_waitid() overwrites everything in ru */
1995 if (COMPAT_USE_64BIT_TIME)
1996 err = copy_to_user(uru, &ru, sizeof(ru));
1997 else
1998 err = put_compat_rusage(&ru, uru);
1999 if (err)
2000 return -EFAULT;
2001 }
2002 }
2003
2004 if (!infop)
2005 return err;
2006
2007 if (!user_write_access_begin(infop, sizeof(*infop)))
2008 return -EFAULT;
2009
2010 unsafe_put_user(signo, &infop->si_signo, Efault);
2011 unsafe_put_user(0, &infop->si_errno, Efault);
2012 unsafe_put_user(info.cause, &infop->si_code, Efault);
2013 unsafe_put_user(info.pid, &infop->si_pid, Efault);
2014 unsafe_put_user(info.uid, &infop->si_uid, Efault);
2015 unsafe_put_user(info.status, &infop->si_status, Efault);
2016 user_write_access_end();
2017 return err;
2018Efault:
2019 user_write_access_end();
2020 return -EFAULT;
2021}
2022#endif
2023
2024/*
2025 * This needs to be __function_aligned as GCC implicitly makes any
2026 * implementation of abort() cold and drops alignment specified by
2027 * -falign-functions=N.
2028 *
2029 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
2030 */
2031__weak __function_aligned void abort(void)
2032{
2033 BUG();
2034
2035 /* if that doesn't kill us, halt */
2036 panic("Oops failed to kill thread");
2037}
2038EXPORT_SYMBOL(abort);