1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/kernel/exit.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8 #include <linux/mm.h>
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/task_io_accounting_ops.h>
52 #include <linux/blkdev.h>
53 #include <linux/task_work.h>
54 #include <linux/fs_struct.h>
55 #include <linux/init_task.h>
56 #include <linux/perf_event.h>
57 #include <trace/events/sched.h>
58 #include <linux/hw_breakpoint.h>
59 #include <linux/oom.h>
60 #include <linux/writeback.h>
61 #include <linux/shm.h>
62 #include <linux/kcov.h>
63 #include <linux/kmsan.h>
64 #include <linux/random.h>
65 #include <linux/rcuwait.h>
66 #include <linux/compat.h>
67 #include <linux/io_uring.h>
68 #include <linux/kprobes.h>
69 #include <linux/rethook.h>
70 #include <linux/sysfs.h>
71 #include <linux/user_events.h>
72 #include <linux/uaccess.h>
73
74 #include <uapi/linux/wait.h>
75
76 #include <asm/unistd.h>
77 #include <asm/mmu_context.h>
78
79 #include "exit.h"
80
81 /*
82 * The default value should be high enough to not crash a system that randomly
83 * crashes its kernel from time to time, but low enough to at least not permit
84 * overflowing 32-bit refcounts or the ldsem writer count.
85 */
86 static unsigned int oops_limit = 10000;
87
88 #ifdef CONFIG_SYSCTL
89 static struct ctl_table kern_exit_table[] = {
90 {
91 .procname = "oops_limit",
92 .data = &oops_limit,
93 .maxlen = sizeof(oops_limit),
94 .mode = 0644,
95 .proc_handler = proc_douintvec,
96 },
97 };
98
kernel_exit_sysctls_init(void)99 static __init int kernel_exit_sysctls_init(void)
100 {
101 register_sysctl_init("kernel", kern_exit_table);
102 return 0;
103 }
104 late_initcall(kernel_exit_sysctls_init);
105 #endif
106
107 static atomic_t oops_count = ATOMIC_INIT(0);
108
109 #ifdef CONFIG_SYSFS
oops_count_show(struct kobject * kobj,struct kobj_attribute * attr,char * page)110 static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
111 char *page)
112 {
113 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
114 }
115
116 static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
117
kernel_exit_sysfs_init(void)118 static __init int kernel_exit_sysfs_init(void)
119 {
120 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
121 return 0;
122 }
123 late_initcall(kernel_exit_sysfs_init);
124 #endif
125
__unhash_process(struct task_struct * p,bool group_dead)126 static void __unhash_process(struct task_struct *p, bool group_dead)
127 {
128 nr_threads--;
129 detach_pid(p, PIDTYPE_PID);
130 if (group_dead) {
131 detach_pid(p, PIDTYPE_TGID);
132 detach_pid(p, PIDTYPE_PGID);
133 detach_pid(p, PIDTYPE_SID);
134
135 list_del_rcu(&p->tasks);
136 list_del_init(&p->sibling);
137 __this_cpu_dec(process_counts);
138 }
139 list_del_rcu(&p->thread_node);
140 }
141
142 /*
143 * This function expects the tasklist_lock write-locked.
144 */
__exit_signal(struct task_struct * tsk)145 static void __exit_signal(struct task_struct *tsk)
146 {
147 struct signal_struct *sig = tsk->signal;
148 bool group_dead = thread_group_leader(tsk);
149 struct sighand_struct *sighand;
150 struct tty_struct *tty;
151 u64 utime, stime;
152
153 sighand = rcu_dereference_check(tsk->sighand,
154 lockdep_tasklist_lock_is_held());
155 spin_lock(&sighand->siglock);
156
157 #ifdef CONFIG_POSIX_TIMERS
158 posix_cpu_timers_exit(tsk);
159 if (group_dead)
160 posix_cpu_timers_exit_group(tsk);
161 #endif
162
163 if (group_dead) {
164 tty = sig->tty;
165 sig->tty = NULL;
166 } else {
167 /*
168 * If there is any task waiting for the group exit
169 * then notify it:
170 */
171 if (sig->notify_count > 0 && !--sig->notify_count)
172 wake_up_process(sig->group_exec_task);
173
174 if (tsk == sig->curr_target)
175 sig->curr_target = next_thread(tsk);
176 }
177
178 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
179 sizeof(unsigned long long));
180
181 /*
182 * Accumulate here the counters for all threads as they die. We could
183 * skip the group leader because it is the last user of signal_struct,
184 * but we want to avoid the race with thread_group_cputime() which can
185 * see the empty ->thread_head list.
186 */
187 task_cputime(tsk, &utime, &stime);
188 write_seqlock(&sig->stats_lock);
189 sig->utime += utime;
190 sig->stime += stime;
191 sig->gtime += task_gtime(tsk);
192 sig->min_flt += tsk->min_flt;
193 sig->maj_flt += tsk->maj_flt;
194 sig->nvcsw += tsk->nvcsw;
195 sig->nivcsw += tsk->nivcsw;
196 sig->inblock += task_io_get_inblock(tsk);
197 sig->oublock += task_io_get_oublock(tsk);
198 task_io_accounting_add(&sig->ioac, &tsk->ioac);
199 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
200 sig->nr_threads--;
201 __unhash_process(tsk, group_dead);
202 write_sequnlock(&sig->stats_lock);
203
204 /*
205 * Do this under ->siglock, we can race with another thread
206 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
207 */
208 flush_sigqueue(&tsk->pending);
209 tsk->sighand = NULL;
210 spin_unlock(&sighand->siglock);
211
212 __cleanup_sighand(sighand);
213 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
214 if (group_dead) {
215 flush_sigqueue(&sig->shared_pending);
216 tty_kref_put(tty);
217 }
218 }
219
delayed_put_task_struct(struct rcu_head * rhp)220 static void delayed_put_task_struct(struct rcu_head *rhp)
221 {
222 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
223
224 kprobe_flush_task(tsk);
225 rethook_flush_task(tsk);
226 perf_event_delayed_put(tsk);
227 trace_sched_process_free(tsk);
228 put_task_struct(tsk);
229 }
230
put_task_struct_rcu_user(struct task_struct * task)231 void put_task_struct_rcu_user(struct task_struct *task)
232 {
233 if (refcount_dec_and_test(&task->rcu_users))
234 call_rcu(&task->rcu, delayed_put_task_struct);
235 }
236
release_thread(struct task_struct * dead_task)237 void __weak release_thread(struct task_struct *dead_task)
238 {
239 }
240
release_task(struct task_struct * p)241 void release_task(struct task_struct *p)
242 {
243 struct task_struct *leader;
244 struct pid *thread_pid;
245 int zap_leader;
246 repeat:
247 /* don't need to get the RCU readlock here - the process is dead and
248 * can't be modifying its own credentials. But shut RCU-lockdep up */
249 rcu_read_lock();
250 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
251 rcu_read_unlock();
252
253 cgroup_release(p);
254
255 write_lock_irq(&tasklist_lock);
256 ptrace_release_task(p);
257 thread_pid = get_pid(p->thread_pid);
258 __exit_signal(p);
259
260 /*
261 * If we are the last non-leader member of the thread
262 * group, and the leader is zombie, then notify the
263 * group leader's parent process. (if it wants notification.)
264 */
265 zap_leader = 0;
266 leader = p->group_leader;
267 if (leader != p && thread_group_empty(leader)
268 && leader->exit_state == EXIT_ZOMBIE) {
269 /*
270 * If we were the last child thread and the leader has
271 * exited already, and the leader's parent ignores SIGCHLD,
272 * then we are the one who should release the leader.
273 */
274 zap_leader = do_notify_parent(leader, leader->exit_signal);
275 if (zap_leader)
276 leader->exit_state = EXIT_DEAD;
277 }
278
279 write_unlock_irq(&tasklist_lock);
280 seccomp_filter_release(p);
281 proc_flush_pid(thread_pid);
282 put_pid(thread_pid);
283 release_thread(p);
284 put_task_struct_rcu_user(p);
285
286 p = leader;
287 if (unlikely(zap_leader))
288 goto repeat;
289 }
290
rcuwait_wake_up(struct rcuwait * w)291 int rcuwait_wake_up(struct rcuwait *w)
292 {
293 int ret = 0;
294 struct task_struct *task;
295
296 rcu_read_lock();
297
298 /*
299 * Order condition vs @task, such that everything prior to the load
300 * of @task is visible. This is the condition as to why the user called
301 * rcuwait_wake() in the first place. Pairs with set_current_state()
302 * barrier (A) in rcuwait_wait_event().
303 *
304 * WAIT WAKE
305 * [S] tsk = current [S] cond = true
306 * MB (A) MB (B)
307 * [L] cond [L] tsk
308 */
309 smp_mb(); /* (B) */
310
311 task = rcu_dereference(w->task);
312 if (task)
313 ret = wake_up_process(task);
314 rcu_read_unlock();
315
316 return ret;
317 }
318 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
319
320 /*
321 * Determine if a process group is "orphaned", according to the POSIX
322 * definition in 2.2.2.52. Orphaned process groups are not to be affected
323 * by terminal-generated stop signals. Newly orphaned process groups are
324 * to receive a SIGHUP and a SIGCONT.
325 *
326 * "I ask you, have you ever known what it is to be an orphan?"
327 */
will_become_orphaned_pgrp(struct pid * pgrp,struct task_struct * ignored_task)328 static int will_become_orphaned_pgrp(struct pid *pgrp,
329 struct task_struct *ignored_task)
330 {
331 struct task_struct *p;
332
333 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
334 if ((p == ignored_task) ||
335 (p->exit_state && thread_group_empty(p)) ||
336 is_global_init(p->real_parent))
337 continue;
338
339 if (task_pgrp(p->real_parent) != pgrp &&
340 task_session(p->real_parent) == task_session(p))
341 return 0;
342 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
343
344 return 1;
345 }
346
is_current_pgrp_orphaned(void)347 int is_current_pgrp_orphaned(void)
348 {
349 int retval;
350
351 read_lock(&tasklist_lock);
352 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
353 read_unlock(&tasklist_lock);
354
355 return retval;
356 }
357
has_stopped_jobs(struct pid * pgrp)358 static bool has_stopped_jobs(struct pid *pgrp)
359 {
360 struct task_struct *p;
361
362 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
363 if (p->signal->flags & SIGNAL_STOP_STOPPED)
364 return true;
365 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
366
367 return false;
368 }
369
370 /*
371 * Check to see if any process groups have become orphaned as
372 * a result of our exiting, and if they have any stopped jobs,
373 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
374 */
375 static void
kill_orphaned_pgrp(struct task_struct * tsk,struct task_struct * parent)376 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
377 {
378 struct pid *pgrp = task_pgrp(tsk);
379 struct task_struct *ignored_task = tsk;
380
381 if (!parent)
382 /* exit: our father is in a different pgrp than
383 * we are and we were the only connection outside.
384 */
385 parent = tsk->real_parent;
386 else
387 /* reparent: our child is in a different pgrp than
388 * we are, and it was the only connection outside.
389 */
390 ignored_task = NULL;
391
392 if (task_pgrp(parent) != pgrp &&
393 task_session(parent) == task_session(tsk) &&
394 will_become_orphaned_pgrp(pgrp, ignored_task) &&
395 has_stopped_jobs(pgrp)) {
396 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
397 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
398 }
399 }
400
coredump_task_exit(struct task_struct * tsk)401 static void coredump_task_exit(struct task_struct *tsk)
402 {
403 struct core_state *core_state;
404
405 /*
406 * Serialize with any possible pending coredump.
407 * We must hold siglock around checking core_state
408 * and setting PF_POSTCOREDUMP. The core-inducing thread
409 * will increment ->nr_threads for each thread in the
410 * group without PF_POSTCOREDUMP set.
411 */
412 spin_lock_irq(&tsk->sighand->siglock);
413 tsk->flags |= PF_POSTCOREDUMP;
414 core_state = tsk->signal->core_state;
415 spin_unlock_irq(&tsk->sighand->siglock);
416 if (core_state) {
417 struct core_thread self;
418
419 self.task = current;
420 if (self.task->flags & PF_SIGNALED)
421 self.next = xchg(&core_state->dumper.next, &self);
422 else
423 self.task = NULL;
424 /*
425 * Implies mb(), the result of xchg() must be visible
426 * to core_state->dumper.
427 */
428 if (atomic_dec_and_test(&core_state->nr_threads))
429 complete(&core_state->startup);
430
431 for (;;) {
432 set_current_state(TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
433 if (!self.task) /* see coredump_finish() */
434 break;
435 schedule();
436 }
437 __set_current_state(TASK_RUNNING);
438 }
439 }
440
441 #ifdef CONFIG_MEMCG
442 /*
443 * A task is exiting. If it owned this mm, find a new owner for the mm.
444 */
mm_update_next_owner(struct mm_struct * mm)445 void mm_update_next_owner(struct mm_struct *mm)
446 {
447 struct task_struct *c, *g, *p = current;
448
449 retry:
450 /*
451 * If the exiting or execing task is not the owner, it's
452 * someone else's problem.
453 */
454 if (mm->owner != p)
455 return;
456 /*
457 * The current owner is exiting/execing and there are no other
458 * candidates. Do not leave the mm pointing to a possibly
459 * freed task structure.
460 */
461 if (atomic_read(&mm->mm_users) <= 1) {
462 WRITE_ONCE(mm->owner, NULL);
463 return;
464 }
465
466 read_lock(&tasklist_lock);
467 /*
468 * Search in the children
469 */
470 list_for_each_entry(c, &p->children, sibling) {
471 if (c->mm == mm)
472 goto assign_new_owner;
473 }
474
475 /*
476 * Search in the siblings
477 */
478 list_for_each_entry(c, &p->real_parent->children, sibling) {
479 if (c->mm == mm)
480 goto assign_new_owner;
481 }
482
483 /*
484 * Search through everything else, we should not get here often.
485 */
486 for_each_process(g) {
487 if (g->flags & PF_KTHREAD)
488 continue;
489 for_each_thread(g, c) {
490 if (c->mm == mm)
491 goto assign_new_owner;
492 if (c->mm)
493 break;
494 }
495 }
496 read_unlock(&tasklist_lock);
497 /*
498 * We found no owner yet mm_users > 1: this implies that we are
499 * most likely racing with swapoff (try_to_unuse()) or /proc or
500 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
501 */
502 WRITE_ONCE(mm->owner, NULL);
503 return;
504
505 assign_new_owner:
506 BUG_ON(c == p);
507 get_task_struct(c);
508 /*
509 * The task_lock protects c->mm from changing.
510 * We always want mm->owner->mm == mm
511 */
512 task_lock(c);
513 /*
514 * Delay read_unlock() till we have the task_lock()
515 * to ensure that c does not slip away underneath us
516 */
517 read_unlock(&tasklist_lock);
518 if (c->mm != mm) {
519 task_unlock(c);
520 put_task_struct(c);
521 goto retry;
522 }
523 WRITE_ONCE(mm->owner, c);
524 lru_gen_migrate_mm(mm);
525 task_unlock(c);
526 put_task_struct(c);
527 }
528 #endif /* CONFIG_MEMCG */
529
530 /*
531 * Turn us into a lazy TLB process if we
532 * aren't already..
533 */
exit_mm(void)534 static void exit_mm(void)
535 {
536 struct mm_struct *mm = current->mm;
537
538 exit_mm_release(current, mm);
539 if (!mm)
540 return;
541 mmap_read_lock(mm);
542 mmgrab_lazy_tlb(mm);
543 BUG_ON(mm != current->active_mm);
544 /* more a memory barrier than a real lock */
545 task_lock(current);
546 /*
547 * When a thread stops operating on an address space, the loop
548 * in membarrier_private_expedited() may not observe that
549 * tsk->mm, and the loop in membarrier_global_expedited() may
550 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
551 * rq->membarrier_state, so those would not issue an IPI.
552 * Membarrier requires a memory barrier after accessing
553 * user-space memory, before clearing tsk->mm or the
554 * rq->membarrier_state.
555 */
556 smp_mb__after_spinlock();
557 local_irq_disable();
558 current->mm = NULL;
559 membarrier_update_current_mm(NULL);
560 enter_lazy_tlb(mm, current);
561 local_irq_enable();
562 task_unlock(current);
563 mmap_read_unlock(mm);
564 mm_update_next_owner(mm);
565 mmput(mm);
566 if (test_thread_flag(TIF_MEMDIE))
567 exit_oom_victim();
568 }
569
find_alive_thread(struct task_struct * p)570 static struct task_struct *find_alive_thread(struct task_struct *p)
571 {
572 struct task_struct *t;
573
574 for_each_thread(p, t) {
575 if (!(t->flags & PF_EXITING))
576 return t;
577 }
578 return NULL;
579 }
580
find_child_reaper(struct task_struct * father,struct list_head * dead)581 static struct task_struct *find_child_reaper(struct task_struct *father,
582 struct list_head *dead)
583 __releases(&tasklist_lock)
584 __acquires(&tasklist_lock)
585 {
586 struct pid_namespace *pid_ns = task_active_pid_ns(father);
587 struct task_struct *reaper = pid_ns->child_reaper;
588 struct task_struct *p, *n;
589
590 if (likely(reaper != father))
591 return reaper;
592
593 reaper = find_alive_thread(father);
594 if (reaper) {
595 pid_ns->child_reaper = reaper;
596 return reaper;
597 }
598
599 write_unlock_irq(&tasklist_lock);
600
601 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
602 list_del_init(&p->ptrace_entry);
603 release_task(p);
604 }
605
606 zap_pid_ns_processes(pid_ns);
607 write_lock_irq(&tasklist_lock);
608
609 return father;
610 }
611
612 /*
613 * When we die, we re-parent all our children, and try to:
614 * 1. give them to another thread in our thread group, if such a member exists
615 * 2. give it to the first ancestor process which prctl'd itself as a
616 * child_subreaper for its children (like a service manager)
617 * 3. give it to the init process (PID 1) in our pid namespace
618 */
find_new_reaper(struct task_struct * father,struct task_struct * child_reaper)619 static struct task_struct *find_new_reaper(struct task_struct *father,
620 struct task_struct *child_reaper)
621 {
622 struct task_struct *thread, *reaper;
623
624 thread = find_alive_thread(father);
625 if (thread)
626 return thread;
627
628 if (father->signal->has_child_subreaper) {
629 unsigned int ns_level = task_pid(father)->level;
630 /*
631 * Find the first ->is_child_subreaper ancestor in our pid_ns.
632 * We can't check reaper != child_reaper to ensure we do not
633 * cross the namespaces, the exiting parent could be injected
634 * by setns() + fork().
635 * We check pid->level, this is slightly more efficient than
636 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
637 */
638 for (reaper = father->real_parent;
639 task_pid(reaper)->level == ns_level;
640 reaper = reaper->real_parent) {
641 if (reaper == &init_task)
642 break;
643 if (!reaper->signal->is_child_subreaper)
644 continue;
645 thread = find_alive_thread(reaper);
646 if (thread)
647 return thread;
648 }
649 }
650
651 return child_reaper;
652 }
653
654 /*
655 * Any that need to be release_task'd are put on the @dead list.
656 */
reparent_leader(struct task_struct * father,struct task_struct * p,struct list_head * dead)657 static void reparent_leader(struct task_struct *father, struct task_struct *p,
658 struct list_head *dead)
659 {
660 if (unlikely(p->exit_state == EXIT_DEAD))
661 return;
662
663 /* We don't want people slaying init. */
664 p->exit_signal = SIGCHLD;
665
666 /* If it has exited notify the new parent about this child's death. */
667 if (!p->ptrace &&
668 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
669 if (do_notify_parent(p, p->exit_signal)) {
670 p->exit_state = EXIT_DEAD;
671 list_add(&p->ptrace_entry, dead);
672 }
673 }
674
675 kill_orphaned_pgrp(p, father);
676 }
677
678 /*
679 * This does two things:
680 *
681 * A. Make init inherit all the child processes
682 * B. Check to see if any process groups have become orphaned
683 * as a result of our exiting, and if they have any stopped
684 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
685 */
forget_original_parent(struct task_struct * father,struct list_head * dead)686 static void forget_original_parent(struct task_struct *father,
687 struct list_head *dead)
688 {
689 struct task_struct *p, *t, *reaper;
690
691 if (unlikely(!list_empty(&father->ptraced)))
692 exit_ptrace(father, dead);
693
694 /* Can drop and reacquire tasklist_lock */
695 reaper = find_child_reaper(father, dead);
696 if (list_empty(&father->children))
697 return;
698
699 reaper = find_new_reaper(father, reaper);
700 list_for_each_entry(p, &father->children, sibling) {
701 for_each_thread(p, t) {
702 RCU_INIT_POINTER(t->real_parent, reaper);
703 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
704 if (likely(!t->ptrace))
705 t->parent = t->real_parent;
706 if (t->pdeath_signal)
707 group_send_sig_info(t->pdeath_signal,
708 SEND_SIG_NOINFO, t,
709 PIDTYPE_TGID);
710 }
711 /*
712 * If this is a threaded reparent there is no need to
713 * notify anyone anything has happened.
714 */
715 if (!same_thread_group(reaper, father))
716 reparent_leader(father, p, dead);
717 }
718 list_splice_tail_init(&father->children, &reaper->children);
719 }
720
721 /*
722 * Send signals to all our closest relatives so that they know
723 * to properly mourn us..
724 */
exit_notify(struct task_struct * tsk,int group_dead)725 static void exit_notify(struct task_struct *tsk, int group_dead)
726 {
727 bool autoreap;
728 struct task_struct *p, *n;
729 LIST_HEAD(dead);
730
731 write_lock_irq(&tasklist_lock);
732 forget_original_parent(tsk, &dead);
733
734 if (group_dead)
735 kill_orphaned_pgrp(tsk->group_leader, NULL);
736
737 tsk->exit_state = EXIT_ZOMBIE;
738 /*
739 * sub-thread or delay_group_leader(), wake up the
740 * PIDFD_THREAD waiters.
741 */
742 if (!thread_group_empty(tsk))
743 do_notify_pidfd(tsk);
744
745 if (unlikely(tsk->ptrace)) {
746 int sig = thread_group_leader(tsk) &&
747 thread_group_empty(tsk) &&
748 !ptrace_reparented(tsk) ?
749 tsk->exit_signal : SIGCHLD;
750 autoreap = do_notify_parent(tsk, sig);
751 } else if (thread_group_leader(tsk)) {
752 autoreap = thread_group_empty(tsk) &&
753 do_notify_parent(tsk, tsk->exit_signal);
754 } else {
755 autoreap = true;
756 }
757
758 if (autoreap) {
759 tsk->exit_state = EXIT_DEAD;
760 list_add(&tsk->ptrace_entry, &dead);
761 }
762
763 /* mt-exec, de_thread() is waiting for group leader */
764 if (unlikely(tsk->signal->notify_count < 0))
765 wake_up_process(tsk->signal->group_exec_task);
766 write_unlock_irq(&tasklist_lock);
767
768 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
769 list_del_init(&p->ptrace_entry);
770 release_task(p);
771 }
772 }
773
774 #ifdef CONFIG_DEBUG_STACK_USAGE
check_stack_usage(void)775 static void check_stack_usage(void)
776 {
777 static DEFINE_SPINLOCK(low_water_lock);
778 static int lowest_to_date = THREAD_SIZE;
779 unsigned long free;
780
781 free = stack_not_used(current);
782
783 if (free >= lowest_to_date)
784 return;
785
786 spin_lock(&low_water_lock);
787 if (free < lowest_to_date) {
788 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
789 current->comm, task_pid_nr(current), free);
790 lowest_to_date = free;
791 }
792 spin_unlock(&low_water_lock);
793 }
794 #else
check_stack_usage(void)795 static inline void check_stack_usage(void) {}
796 #endif
797
synchronize_group_exit(struct task_struct * tsk,long code)798 static void synchronize_group_exit(struct task_struct *tsk, long code)
799 {
800 struct sighand_struct *sighand = tsk->sighand;
801 struct signal_struct *signal = tsk->signal;
802
803 spin_lock_irq(&sighand->siglock);
804 signal->quick_threads--;
805 if ((signal->quick_threads == 0) &&
806 !(signal->flags & SIGNAL_GROUP_EXIT)) {
807 signal->flags = SIGNAL_GROUP_EXIT;
808 signal->group_exit_code = code;
809 signal->group_stop_count = 0;
810 }
811 spin_unlock_irq(&sighand->siglock);
812 }
813
do_exit(long code)814 void __noreturn do_exit(long code)
815 {
816 struct task_struct *tsk = current;
817 int group_dead;
818
819 WARN_ON(irqs_disabled());
820
821 synchronize_group_exit(tsk, code);
822
823 WARN_ON(tsk->plug);
824
825 kcov_task_exit(tsk);
826 kmsan_task_exit(tsk);
827
828 coredump_task_exit(tsk);
829 ptrace_event(PTRACE_EVENT_EXIT, code);
830 user_events_exit(tsk);
831
832 io_uring_files_cancel();
833 exit_signals(tsk); /* sets PF_EXITING */
834
835 acct_update_integrals(tsk);
836 group_dead = atomic_dec_and_test(&tsk->signal->live);
837 if (group_dead) {
838 /*
839 * If the last thread of global init has exited, panic
840 * immediately to get a useable coredump.
841 */
842 if (unlikely(is_global_init(tsk)))
843 panic("Attempted to kill init! exitcode=0x%08x\n",
844 tsk->signal->group_exit_code ?: (int)code);
845
846 #ifdef CONFIG_POSIX_TIMERS
847 hrtimer_cancel(&tsk->signal->real_timer);
848 exit_itimers(tsk);
849 #endif
850 if (tsk->mm)
851 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
852 }
853 acct_collect(code, group_dead);
854 if (group_dead)
855 tty_audit_exit();
856 audit_free(tsk);
857
858 tsk->exit_code = code;
859 taskstats_exit(tsk, group_dead);
860
861 exit_mm();
862
863 if (group_dead)
864 acct_process();
865 trace_sched_process_exit(tsk);
866
867 exit_sem(tsk);
868 exit_shm(tsk);
869 exit_files(tsk);
870 exit_fs(tsk);
871 if (group_dead)
872 disassociate_ctty(1);
873 exit_task_namespaces(tsk);
874 exit_task_work(tsk);
875 exit_thread(tsk);
876
877 /*
878 * Flush inherited counters to the parent - before the parent
879 * gets woken up by child-exit notifications.
880 *
881 * because of cgroup mode, must be called before cgroup_exit()
882 */
883 perf_event_exit_task(tsk);
884
885 sched_autogroup_exit_task(tsk);
886 cgroup_exit(tsk);
887
888 /*
889 * FIXME: do that only when needed, using sched_exit tracepoint
890 */
891 flush_ptrace_hw_breakpoint(tsk);
892
893 exit_tasks_rcu_start();
894 exit_notify(tsk, group_dead);
895 proc_exit_connector(tsk);
896 mpol_put_task_policy(tsk);
897 #ifdef CONFIG_FUTEX
898 if (unlikely(current->pi_state_cache))
899 kfree(current->pi_state_cache);
900 #endif
901 /*
902 * Make sure we are holding no locks:
903 */
904 debug_check_no_locks_held();
905
906 if (tsk->io_context)
907 exit_io_context(tsk);
908
909 if (tsk->splice_pipe)
910 free_pipe_info(tsk->splice_pipe);
911
912 if (tsk->task_frag.page)
913 put_page(tsk->task_frag.page);
914
915 exit_task_stack_account(tsk);
916
917 check_stack_usage();
918 preempt_disable();
919 if (tsk->nr_dirtied)
920 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
921 exit_rcu();
922 exit_tasks_rcu_finish();
923
924 lockdep_free_task(tsk);
925 do_task_dead();
926 }
927
make_task_dead(int signr)928 void __noreturn make_task_dead(int signr)
929 {
930 /*
931 * Take the task off the cpu after something catastrophic has
932 * happened.
933 *
934 * We can get here from a kernel oops, sometimes with preemption off.
935 * Start by checking for critical errors.
936 * Then fix up important state like USER_DS and preemption.
937 * Then do everything else.
938 */
939 struct task_struct *tsk = current;
940 unsigned int limit;
941
942 if (unlikely(in_interrupt()))
943 panic("Aiee, killing interrupt handler!");
944 if (unlikely(!tsk->pid))
945 panic("Attempted to kill the idle task!");
946
947 if (unlikely(irqs_disabled())) {
948 pr_info("note: %s[%d] exited with irqs disabled\n",
949 current->comm, task_pid_nr(current));
950 local_irq_enable();
951 }
952 if (unlikely(in_atomic())) {
953 pr_info("note: %s[%d] exited with preempt_count %d\n",
954 current->comm, task_pid_nr(current),
955 preempt_count());
956 preempt_count_set(PREEMPT_ENABLED);
957 }
958
959 /*
960 * Every time the system oopses, if the oops happens while a reference
961 * to an object was held, the reference leaks.
962 * If the oops doesn't also leak memory, repeated oopsing can cause
963 * reference counters to wrap around (if they're not using refcount_t).
964 * This means that repeated oopsing can make unexploitable-looking bugs
965 * exploitable through repeated oopsing.
966 * To make sure this can't happen, place an upper bound on how often the
967 * kernel may oops without panic().
968 */
969 limit = READ_ONCE(oops_limit);
970 if (atomic_inc_return(&oops_count) >= limit && limit)
971 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
972
973 /*
974 * We're taking recursive faults here in make_task_dead. Safest is to just
975 * leave this task alone and wait for reboot.
976 */
977 if (unlikely(tsk->flags & PF_EXITING)) {
978 pr_alert("Fixing recursive fault but reboot is needed!\n");
979 futex_exit_recursive(tsk);
980 tsk->exit_state = EXIT_DEAD;
981 refcount_inc(&tsk->rcu_users);
982 do_task_dead();
983 }
984
985 do_exit(signr);
986 }
987
SYSCALL_DEFINE1(exit,int,error_code)988 SYSCALL_DEFINE1(exit, int, error_code)
989 {
990 do_exit((error_code&0xff)<<8);
991 }
992
993 /*
994 * Take down every thread in the group. This is called by fatal signals
995 * as well as by sys_exit_group (below).
996 */
997 void __noreturn
do_group_exit(int exit_code)998 do_group_exit(int exit_code)
999 {
1000 struct signal_struct *sig = current->signal;
1001
1002 if (sig->flags & SIGNAL_GROUP_EXIT)
1003 exit_code = sig->group_exit_code;
1004 else if (sig->group_exec_task)
1005 exit_code = 0;
1006 else {
1007 struct sighand_struct *const sighand = current->sighand;
1008
1009 spin_lock_irq(&sighand->siglock);
1010 if (sig->flags & SIGNAL_GROUP_EXIT)
1011 /* Another thread got here before we took the lock. */
1012 exit_code = sig->group_exit_code;
1013 else if (sig->group_exec_task)
1014 exit_code = 0;
1015 else {
1016 sig->group_exit_code = exit_code;
1017 sig->flags = SIGNAL_GROUP_EXIT;
1018 zap_other_threads(current);
1019 }
1020 spin_unlock_irq(&sighand->siglock);
1021 }
1022
1023 do_exit(exit_code);
1024 /* NOTREACHED */
1025 }
1026
1027 /*
1028 * this kills every thread in the thread group. Note that any externally
1029 * wait4()-ing process will get the correct exit code - even if this
1030 * thread is not the thread group leader.
1031 */
SYSCALL_DEFINE1(exit_group,int,error_code)1032 SYSCALL_DEFINE1(exit_group, int, error_code)
1033 {
1034 do_group_exit((error_code & 0xff) << 8);
1035 /* NOTREACHED */
1036 return 0;
1037 }
1038
eligible_pid(struct wait_opts * wo,struct task_struct * p)1039 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1040 {
1041 return wo->wo_type == PIDTYPE_MAX ||
1042 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1043 }
1044
1045 static int
eligible_child(struct wait_opts * wo,bool ptrace,struct task_struct * p)1046 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1047 {
1048 if (!eligible_pid(wo, p))
1049 return 0;
1050
1051 /*
1052 * Wait for all children (clone and not) if __WALL is set or
1053 * if it is traced by us.
1054 */
1055 if (ptrace || (wo->wo_flags & __WALL))
1056 return 1;
1057
1058 /*
1059 * Otherwise, wait for clone children *only* if __WCLONE is set;
1060 * otherwise, wait for non-clone children *only*.
1061 *
1062 * Note: a "clone" child here is one that reports to its parent
1063 * using a signal other than SIGCHLD, or a non-leader thread which
1064 * we can only see if it is traced by us.
1065 */
1066 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1067 return 0;
1068
1069 return 1;
1070 }
1071
1072 /*
1073 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1074 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1075 * the lock and this task is uninteresting. If we return nonzero, we have
1076 * released the lock and the system call should return.
1077 */
wait_task_zombie(struct wait_opts * wo,struct task_struct * p)1078 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1079 {
1080 int state, status;
1081 pid_t pid = task_pid_vnr(p);
1082 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1083 struct waitid_info *infop;
1084
1085 if (!likely(wo->wo_flags & WEXITED))
1086 return 0;
1087
1088 if (unlikely(wo->wo_flags & WNOWAIT)) {
1089 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1090 ? p->signal->group_exit_code : p->exit_code;
1091 get_task_struct(p);
1092 read_unlock(&tasklist_lock);
1093 sched_annotate_sleep();
1094 if (wo->wo_rusage)
1095 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1096 put_task_struct(p);
1097 goto out_info;
1098 }
1099 /*
1100 * Move the task's state to DEAD/TRACE, only one thread can do this.
1101 */
1102 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1103 EXIT_TRACE : EXIT_DEAD;
1104 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1105 return 0;
1106 /*
1107 * We own this thread, nobody else can reap it.
1108 */
1109 read_unlock(&tasklist_lock);
1110 sched_annotate_sleep();
1111
1112 /*
1113 * Check thread_group_leader() to exclude the traced sub-threads.
1114 */
1115 if (state == EXIT_DEAD && thread_group_leader(p)) {
1116 struct signal_struct *sig = p->signal;
1117 struct signal_struct *psig = current->signal;
1118 unsigned long maxrss;
1119 u64 tgutime, tgstime;
1120
1121 /*
1122 * The resource counters for the group leader are in its
1123 * own task_struct. Those for dead threads in the group
1124 * are in its signal_struct, as are those for the child
1125 * processes it has previously reaped. All these
1126 * accumulate in the parent's signal_struct c* fields.
1127 *
1128 * We don't bother to take a lock here to protect these
1129 * p->signal fields because the whole thread group is dead
1130 * and nobody can change them.
1131 *
1132 * psig->stats_lock also protects us from our sub-threads
1133 * which can reap other children at the same time.
1134 *
1135 * We use thread_group_cputime_adjusted() to get times for
1136 * the thread group, which consolidates times for all threads
1137 * in the group including the group leader.
1138 */
1139 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1140 write_seqlock_irq(&psig->stats_lock);
1141 psig->cutime += tgutime + sig->cutime;
1142 psig->cstime += tgstime + sig->cstime;
1143 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1144 psig->cmin_flt +=
1145 p->min_flt + sig->min_flt + sig->cmin_flt;
1146 psig->cmaj_flt +=
1147 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1148 psig->cnvcsw +=
1149 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1150 psig->cnivcsw +=
1151 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1152 psig->cinblock +=
1153 task_io_get_inblock(p) +
1154 sig->inblock + sig->cinblock;
1155 psig->coublock +=
1156 task_io_get_oublock(p) +
1157 sig->oublock + sig->coublock;
1158 maxrss = max(sig->maxrss, sig->cmaxrss);
1159 if (psig->cmaxrss < maxrss)
1160 psig->cmaxrss = maxrss;
1161 task_io_accounting_add(&psig->ioac, &p->ioac);
1162 task_io_accounting_add(&psig->ioac, &sig->ioac);
1163 write_sequnlock_irq(&psig->stats_lock);
1164 }
1165
1166 if (wo->wo_rusage)
1167 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1168 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1169 ? p->signal->group_exit_code : p->exit_code;
1170 wo->wo_stat = status;
1171
1172 if (state == EXIT_TRACE) {
1173 write_lock_irq(&tasklist_lock);
1174 /* We dropped tasklist, ptracer could die and untrace */
1175 ptrace_unlink(p);
1176
1177 /* If parent wants a zombie, don't release it now */
1178 state = EXIT_ZOMBIE;
1179 if (do_notify_parent(p, p->exit_signal))
1180 state = EXIT_DEAD;
1181 p->exit_state = state;
1182 write_unlock_irq(&tasklist_lock);
1183 }
1184 if (state == EXIT_DEAD)
1185 release_task(p);
1186
1187 out_info:
1188 infop = wo->wo_info;
1189 if (infop) {
1190 if ((status & 0x7f) == 0) {
1191 infop->cause = CLD_EXITED;
1192 infop->status = status >> 8;
1193 } else {
1194 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1195 infop->status = status & 0x7f;
1196 }
1197 infop->pid = pid;
1198 infop->uid = uid;
1199 }
1200
1201 return pid;
1202 }
1203
task_stopped_code(struct task_struct * p,bool ptrace)1204 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1205 {
1206 if (ptrace) {
1207 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1208 return &p->exit_code;
1209 } else {
1210 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1211 return &p->signal->group_exit_code;
1212 }
1213 return NULL;
1214 }
1215
1216 /**
1217 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1218 * @wo: wait options
1219 * @ptrace: is the wait for ptrace
1220 * @p: task to wait for
1221 *
1222 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1223 *
1224 * CONTEXT:
1225 * read_lock(&tasklist_lock), which is released if return value is
1226 * non-zero. Also, grabs and releases @p->sighand->siglock.
1227 *
1228 * RETURNS:
1229 * 0 if wait condition didn't exist and search for other wait conditions
1230 * should continue. Non-zero return, -errno on failure and @p's pid on
1231 * success, implies that tasklist_lock is released and wait condition
1232 * search should terminate.
1233 */
wait_task_stopped(struct wait_opts * wo,int ptrace,struct task_struct * p)1234 static int wait_task_stopped(struct wait_opts *wo,
1235 int ptrace, struct task_struct *p)
1236 {
1237 struct waitid_info *infop;
1238 int exit_code, *p_code, why;
1239 uid_t uid = 0; /* unneeded, required by compiler */
1240 pid_t pid;
1241
1242 /*
1243 * Traditionally we see ptrace'd stopped tasks regardless of options.
1244 */
1245 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1246 return 0;
1247
1248 if (!task_stopped_code(p, ptrace))
1249 return 0;
1250
1251 exit_code = 0;
1252 spin_lock_irq(&p->sighand->siglock);
1253
1254 p_code = task_stopped_code(p, ptrace);
1255 if (unlikely(!p_code))
1256 goto unlock_sig;
1257
1258 exit_code = *p_code;
1259 if (!exit_code)
1260 goto unlock_sig;
1261
1262 if (!unlikely(wo->wo_flags & WNOWAIT))
1263 *p_code = 0;
1264
1265 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1266 unlock_sig:
1267 spin_unlock_irq(&p->sighand->siglock);
1268 if (!exit_code)
1269 return 0;
1270
1271 /*
1272 * Now we are pretty sure this task is interesting.
1273 * Make sure it doesn't get reaped out from under us while we
1274 * give up the lock and then examine it below. We don't want to
1275 * keep holding onto the tasklist_lock while we call getrusage and
1276 * possibly take page faults for user memory.
1277 */
1278 get_task_struct(p);
1279 pid = task_pid_vnr(p);
1280 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1281 read_unlock(&tasklist_lock);
1282 sched_annotate_sleep();
1283 if (wo->wo_rusage)
1284 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1285 put_task_struct(p);
1286
1287 if (likely(!(wo->wo_flags & WNOWAIT)))
1288 wo->wo_stat = (exit_code << 8) | 0x7f;
1289
1290 infop = wo->wo_info;
1291 if (infop) {
1292 infop->cause = why;
1293 infop->status = exit_code;
1294 infop->pid = pid;
1295 infop->uid = uid;
1296 }
1297 return pid;
1298 }
1299
1300 /*
1301 * Handle do_wait work for one task in a live, non-stopped state.
1302 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1303 * the lock and this task is uninteresting. If we return nonzero, we have
1304 * released the lock and the system call should return.
1305 */
wait_task_continued(struct wait_opts * wo,struct task_struct * p)1306 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1307 {
1308 struct waitid_info *infop;
1309 pid_t pid;
1310 uid_t uid;
1311
1312 if (!unlikely(wo->wo_flags & WCONTINUED))
1313 return 0;
1314
1315 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1316 return 0;
1317
1318 spin_lock_irq(&p->sighand->siglock);
1319 /* Re-check with the lock held. */
1320 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1321 spin_unlock_irq(&p->sighand->siglock);
1322 return 0;
1323 }
1324 if (!unlikely(wo->wo_flags & WNOWAIT))
1325 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1326 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1327 spin_unlock_irq(&p->sighand->siglock);
1328
1329 pid = task_pid_vnr(p);
1330 get_task_struct(p);
1331 read_unlock(&tasklist_lock);
1332 sched_annotate_sleep();
1333 if (wo->wo_rusage)
1334 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1335 put_task_struct(p);
1336
1337 infop = wo->wo_info;
1338 if (!infop) {
1339 wo->wo_stat = 0xffff;
1340 } else {
1341 infop->cause = CLD_CONTINUED;
1342 infop->pid = pid;
1343 infop->uid = uid;
1344 infop->status = SIGCONT;
1345 }
1346 return pid;
1347 }
1348
1349 /*
1350 * Consider @p for a wait by @parent.
1351 *
1352 * -ECHILD should be in ->notask_error before the first call.
1353 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1354 * Returns zero if the search for a child should continue;
1355 * then ->notask_error is 0 if @p is an eligible child,
1356 * or still -ECHILD.
1357 */
wait_consider_task(struct wait_opts * wo,int ptrace,struct task_struct * p)1358 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1359 struct task_struct *p)
1360 {
1361 /*
1362 * We can race with wait_task_zombie() from another thread.
1363 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1364 * can't confuse the checks below.
1365 */
1366 int exit_state = READ_ONCE(p->exit_state);
1367 int ret;
1368
1369 if (unlikely(exit_state == EXIT_DEAD))
1370 return 0;
1371
1372 ret = eligible_child(wo, ptrace, p);
1373 if (!ret)
1374 return ret;
1375
1376 if (unlikely(exit_state == EXIT_TRACE)) {
1377 /*
1378 * ptrace == 0 means we are the natural parent. In this case
1379 * we should clear notask_error, debugger will notify us.
1380 */
1381 if (likely(!ptrace))
1382 wo->notask_error = 0;
1383 return 0;
1384 }
1385
1386 if (likely(!ptrace) && unlikely(p->ptrace)) {
1387 /*
1388 * If it is traced by its real parent's group, just pretend
1389 * the caller is ptrace_do_wait() and reap this child if it
1390 * is zombie.
1391 *
1392 * This also hides group stop state from real parent; otherwise
1393 * a single stop can be reported twice as group and ptrace stop.
1394 * If a ptracer wants to distinguish these two events for its
1395 * own children it should create a separate process which takes
1396 * the role of real parent.
1397 */
1398 if (!ptrace_reparented(p))
1399 ptrace = 1;
1400 }
1401
1402 /* slay zombie? */
1403 if (exit_state == EXIT_ZOMBIE) {
1404 /* we don't reap group leaders with subthreads */
1405 if (!delay_group_leader(p)) {
1406 /*
1407 * A zombie ptracee is only visible to its ptracer.
1408 * Notification and reaping will be cascaded to the
1409 * real parent when the ptracer detaches.
1410 */
1411 if (unlikely(ptrace) || likely(!p->ptrace))
1412 return wait_task_zombie(wo, p);
1413 }
1414
1415 /*
1416 * Allow access to stopped/continued state via zombie by
1417 * falling through. Clearing of notask_error is complex.
1418 *
1419 * When !@ptrace:
1420 *
1421 * If WEXITED is set, notask_error should naturally be
1422 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1423 * so, if there are live subthreads, there are events to
1424 * wait for. If all subthreads are dead, it's still safe
1425 * to clear - this function will be called again in finite
1426 * amount time once all the subthreads are released and
1427 * will then return without clearing.
1428 *
1429 * When @ptrace:
1430 *
1431 * Stopped state is per-task and thus can't change once the
1432 * target task dies. Only continued and exited can happen.
1433 * Clear notask_error if WCONTINUED | WEXITED.
1434 */
1435 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1436 wo->notask_error = 0;
1437 } else {
1438 /*
1439 * @p is alive and it's gonna stop, continue or exit, so
1440 * there always is something to wait for.
1441 */
1442 wo->notask_error = 0;
1443 }
1444
1445 /*
1446 * Wait for stopped. Depending on @ptrace, different stopped state
1447 * is used and the two don't interact with each other.
1448 */
1449 ret = wait_task_stopped(wo, ptrace, p);
1450 if (ret)
1451 return ret;
1452
1453 /*
1454 * Wait for continued. There's only one continued state and the
1455 * ptracer can consume it which can confuse the real parent. Don't
1456 * use WCONTINUED from ptracer. You don't need or want it.
1457 */
1458 return wait_task_continued(wo, p);
1459 }
1460
1461 /*
1462 * Do the work of do_wait() for one thread in the group, @tsk.
1463 *
1464 * -ECHILD should be in ->notask_error before the first call.
1465 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1466 * Returns zero if the search for a child should continue; then
1467 * ->notask_error is 0 if there were any eligible children,
1468 * or still -ECHILD.
1469 */
do_wait_thread(struct wait_opts * wo,struct task_struct * tsk)1470 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1471 {
1472 struct task_struct *p;
1473
1474 list_for_each_entry(p, &tsk->children, sibling) {
1475 int ret = wait_consider_task(wo, 0, p);
1476
1477 if (ret)
1478 return ret;
1479 }
1480
1481 return 0;
1482 }
1483
ptrace_do_wait(struct wait_opts * wo,struct task_struct * tsk)1484 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1485 {
1486 struct task_struct *p;
1487
1488 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1489 int ret = wait_consider_task(wo, 1, p);
1490
1491 if (ret)
1492 return ret;
1493 }
1494
1495 return 0;
1496 }
1497
pid_child_should_wake(struct wait_opts * wo,struct task_struct * p)1498 bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1499 {
1500 if (!eligible_pid(wo, p))
1501 return false;
1502
1503 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1504 return false;
1505
1506 return true;
1507 }
1508
child_wait_callback(wait_queue_entry_t * wait,unsigned mode,int sync,void * key)1509 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1510 int sync, void *key)
1511 {
1512 struct wait_opts *wo = container_of(wait, struct wait_opts,
1513 child_wait);
1514 struct task_struct *p = key;
1515
1516 if (pid_child_should_wake(wo, p))
1517 return default_wake_function(wait, mode, sync, key);
1518
1519 return 0;
1520 }
1521
__wake_up_parent(struct task_struct * p,struct task_struct * parent)1522 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1523 {
1524 __wake_up_sync_key(&parent->signal->wait_chldexit,
1525 TASK_INTERRUPTIBLE, p);
1526 }
1527
is_effectively_child(struct wait_opts * wo,bool ptrace,struct task_struct * target)1528 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1529 struct task_struct *target)
1530 {
1531 struct task_struct *parent =
1532 !ptrace ? target->real_parent : target->parent;
1533
1534 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1535 same_thread_group(current, parent));
1536 }
1537
1538 /*
1539 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1540 * and tracee lists to find the target task.
1541 */
do_wait_pid(struct wait_opts * wo)1542 static int do_wait_pid(struct wait_opts *wo)
1543 {
1544 bool ptrace;
1545 struct task_struct *target;
1546 int retval;
1547
1548 ptrace = false;
1549 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1550 if (target && is_effectively_child(wo, ptrace, target)) {
1551 retval = wait_consider_task(wo, ptrace, target);
1552 if (retval)
1553 return retval;
1554 }
1555
1556 ptrace = true;
1557 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1558 if (target && target->ptrace &&
1559 is_effectively_child(wo, ptrace, target)) {
1560 retval = wait_consider_task(wo, ptrace, target);
1561 if (retval)
1562 return retval;
1563 }
1564
1565 return 0;
1566 }
1567
__do_wait(struct wait_opts * wo)1568 long __do_wait(struct wait_opts *wo)
1569 {
1570 long retval;
1571
1572 /*
1573 * If there is nothing that can match our criteria, just get out.
1574 * We will clear ->notask_error to zero if we see any child that
1575 * might later match our criteria, even if we are not able to reap
1576 * it yet.
1577 */
1578 wo->notask_error = -ECHILD;
1579 if ((wo->wo_type < PIDTYPE_MAX) &&
1580 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1581 goto notask;
1582
1583 read_lock(&tasklist_lock);
1584
1585 if (wo->wo_type == PIDTYPE_PID) {
1586 retval = do_wait_pid(wo);
1587 if (retval)
1588 return retval;
1589 } else {
1590 struct task_struct *tsk = current;
1591
1592 do {
1593 retval = do_wait_thread(wo, tsk);
1594 if (retval)
1595 return retval;
1596
1597 retval = ptrace_do_wait(wo, tsk);
1598 if (retval)
1599 return retval;
1600
1601 if (wo->wo_flags & __WNOTHREAD)
1602 break;
1603 } while_each_thread(current, tsk);
1604 }
1605 read_unlock(&tasklist_lock);
1606
1607 notask:
1608 retval = wo->notask_error;
1609 if (!retval && !(wo->wo_flags & WNOHANG))
1610 return -ERESTARTSYS;
1611
1612 return retval;
1613 }
1614
do_wait(struct wait_opts * wo)1615 static long do_wait(struct wait_opts *wo)
1616 {
1617 int retval;
1618
1619 trace_sched_process_wait(wo->wo_pid);
1620
1621 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1622 wo->child_wait.private = current;
1623 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1624
1625 do {
1626 set_current_state(TASK_INTERRUPTIBLE);
1627 retval = __do_wait(wo);
1628 if (retval != -ERESTARTSYS)
1629 break;
1630 if (signal_pending(current))
1631 break;
1632 schedule();
1633 } while (1);
1634
1635 __set_current_state(TASK_RUNNING);
1636 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1637 return retval;
1638 }
1639
kernel_waitid_prepare(struct wait_opts * wo,int which,pid_t upid,struct waitid_info * infop,int options,struct rusage * ru)1640 int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1641 struct waitid_info *infop, int options,
1642 struct rusage *ru)
1643 {
1644 unsigned int f_flags = 0;
1645 struct pid *pid = NULL;
1646 enum pid_type type;
1647
1648 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1649 __WNOTHREAD|__WCLONE|__WALL))
1650 return -EINVAL;
1651 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1652 return -EINVAL;
1653
1654 switch (which) {
1655 case P_ALL:
1656 type = PIDTYPE_MAX;
1657 break;
1658 case P_PID:
1659 type = PIDTYPE_PID;
1660 if (upid <= 0)
1661 return -EINVAL;
1662
1663 pid = find_get_pid(upid);
1664 break;
1665 case P_PGID:
1666 type = PIDTYPE_PGID;
1667 if (upid < 0)
1668 return -EINVAL;
1669
1670 if (upid)
1671 pid = find_get_pid(upid);
1672 else
1673 pid = get_task_pid(current, PIDTYPE_PGID);
1674 break;
1675 case P_PIDFD:
1676 type = PIDTYPE_PID;
1677 if (upid < 0)
1678 return -EINVAL;
1679
1680 pid = pidfd_get_pid(upid, &f_flags);
1681 if (IS_ERR(pid))
1682 return PTR_ERR(pid);
1683
1684 break;
1685 default:
1686 return -EINVAL;
1687 }
1688
1689 wo->wo_type = type;
1690 wo->wo_pid = pid;
1691 wo->wo_flags = options;
1692 wo->wo_info = infop;
1693 wo->wo_rusage = ru;
1694 if (f_flags & O_NONBLOCK)
1695 wo->wo_flags |= WNOHANG;
1696
1697 return 0;
1698 }
1699
kernel_waitid(int which,pid_t upid,struct waitid_info * infop,int options,struct rusage * ru)1700 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1701 int options, struct rusage *ru)
1702 {
1703 struct wait_opts wo;
1704 long ret;
1705
1706 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1707 if (ret)
1708 return ret;
1709
1710 ret = do_wait(&wo);
1711 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1712 ret = -EAGAIN;
1713
1714 put_pid(wo.wo_pid);
1715 return ret;
1716 }
1717
SYSCALL_DEFINE5(waitid,int,which,pid_t,upid,struct siginfo __user *,infop,int,options,struct rusage __user *,ru)1718 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1719 infop, int, options, struct rusage __user *, ru)
1720 {
1721 struct rusage r;
1722 struct waitid_info info = {.status = 0};
1723 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1724 int signo = 0;
1725
1726 if (err > 0) {
1727 signo = SIGCHLD;
1728 err = 0;
1729 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1730 return -EFAULT;
1731 }
1732 if (!infop)
1733 return err;
1734
1735 if (!user_write_access_begin(infop, sizeof(*infop)))
1736 return -EFAULT;
1737
1738 unsafe_put_user(signo, &infop->si_signo, Efault);
1739 unsafe_put_user(0, &infop->si_errno, Efault);
1740 unsafe_put_user(info.cause, &infop->si_code, Efault);
1741 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1742 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1743 unsafe_put_user(info.status, &infop->si_status, Efault);
1744 user_write_access_end();
1745 return err;
1746 Efault:
1747 user_write_access_end();
1748 return -EFAULT;
1749 }
1750
kernel_wait4(pid_t upid,int __user * stat_addr,int options,struct rusage * ru)1751 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1752 struct rusage *ru)
1753 {
1754 struct wait_opts wo;
1755 struct pid *pid = NULL;
1756 enum pid_type type;
1757 long ret;
1758
1759 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1760 __WNOTHREAD|__WCLONE|__WALL))
1761 return -EINVAL;
1762
1763 /* -INT_MIN is not defined */
1764 if (upid == INT_MIN)
1765 return -ESRCH;
1766
1767 if (upid == -1)
1768 type = PIDTYPE_MAX;
1769 else if (upid < 0) {
1770 type = PIDTYPE_PGID;
1771 pid = find_get_pid(-upid);
1772 } else if (upid == 0) {
1773 type = PIDTYPE_PGID;
1774 pid = get_task_pid(current, PIDTYPE_PGID);
1775 } else /* upid > 0 */ {
1776 type = PIDTYPE_PID;
1777 pid = find_get_pid(upid);
1778 }
1779
1780 wo.wo_type = type;
1781 wo.wo_pid = pid;
1782 wo.wo_flags = options | WEXITED;
1783 wo.wo_info = NULL;
1784 wo.wo_stat = 0;
1785 wo.wo_rusage = ru;
1786 ret = do_wait(&wo);
1787 put_pid(pid);
1788 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1789 ret = -EFAULT;
1790
1791 return ret;
1792 }
1793
kernel_wait(pid_t pid,int * stat)1794 int kernel_wait(pid_t pid, int *stat)
1795 {
1796 struct wait_opts wo = {
1797 .wo_type = PIDTYPE_PID,
1798 .wo_pid = find_get_pid(pid),
1799 .wo_flags = WEXITED,
1800 };
1801 int ret;
1802
1803 ret = do_wait(&wo);
1804 if (ret > 0 && wo.wo_stat)
1805 *stat = wo.wo_stat;
1806 put_pid(wo.wo_pid);
1807 return ret;
1808 }
1809
SYSCALL_DEFINE4(wait4,pid_t,upid,int __user *,stat_addr,int,options,struct rusage __user *,ru)1810 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1811 int, options, struct rusage __user *, ru)
1812 {
1813 struct rusage r;
1814 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1815
1816 if (err > 0) {
1817 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1818 return -EFAULT;
1819 }
1820 return err;
1821 }
1822
1823 #ifdef __ARCH_WANT_SYS_WAITPID
1824
1825 /*
1826 * sys_waitpid() remains for compatibility. waitpid() should be
1827 * implemented by calling sys_wait4() from libc.a.
1828 */
SYSCALL_DEFINE3(waitpid,pid_t,pid,int __user *,stat_addr,int,options)1829 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1830 {
1831 return kernel_wait4(pid, stat_addr, options, NULL);
1832 }
1833
1834 #endif
1835
1836 #ifdef CONFIG_COMPAT
COMPAT_SYSCALL_DEFINE4(wait4,compat_pid_t,pid,compat_uint_t __user *,stat_addr,int,options,struct compat_rusage __user *,ru)1837 COMPAT_SYSCALL_DEFINE4(wait4,
1838 compat_pid_t, pid,
1839 compat_uint_t __user *, stat_addr,
1840 int, options,
1841 struct compat_rusage __user *, ru)
1842 {
1843 struct rusage r;
1844 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1845 if (err > 0) {
1846 if (ru && put_compat_rusage(&r, ru))
1847 return -EFAULT;
1848 }
1849 return err;
1850 }
1851
COMPAT_SYSCALL_DEFINE5(waitid,int,which,compat_pid_t,pid,struct compat_siginfo __user *,infop,int,options,struct compat_rusage __user *,uru)1852 COMPAT_SYSCALL_DEFINE5(waitid,
1853 int, which, compat_pid_t, pid,
1854 struct compat_siginfo __user *, infop, int, options,
1855 struct compat_rusage __user *, uru)
1856 {
1857 struct rusage ru;
1858 struct waitid_info info = {.status = 0};
1859 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1860 int signo = 0;
1861 if (err > 0) {
1862 signo = SIGCHLD;
1863 err = 0;
1864 if (uru) {
1865 /* kernel_waitid() overwrites everything in ru */
1866 if (COMPAT_USE_64BIT_TIME)
1867 err = copy_to_user(uru, &ru, sizeof(ru));
1868 else
1869 err = put_compat_rusage(&ru, uru);
1870 if (err)
1871 return -EFAULT;
1872 }
1873 }
1874
1875 if (!infop)
1876 return err;
1877
1878 if (!user_write_access_begin(infop, sizeof(*infop)))
1879 return -EFAULT;
1880
1881 unsafe_put_user(signo, &infop->si_signo, Efault);
1882 unsafe_put_user(0, &infop->si_errno, Efault);
1883 unsafe_put_user(info.cause, &infop->si_code, Efault);
1884 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1885 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1886 unsafe_put_user(info.status, &infop->si_status, Efault);
1887 user_write_access_end();
1888 return err;
1889 Efault:
1890 user_write_access_end();
1891 return -EFAULT;
1892 }
1893 #endif
1894
1895 /*
1896 * This needs to be __function_aligned as GCC implicitly makes any
1897 * implementation of abort() cold and drops alignment specified by
1898 * -falign-functions=N.
1899 *
1900 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1901 */
abort(void)1902 __weak __function_aligned void abort(void)
1903 {
1904 BUG();
1905
1906 /* if that doesn't kill us, halt */
1907 panic("Oops failed to kill thread");
1908 }
1909 EXPORT_SYMBOL(abort);
1910