Loading...
1/*
2 * linux/kernel/exit.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/mm.h>
8#include <linux/slab.h>
9#include <linux/sched/autogroup.h>
10#include <linux/sched/mm.h>
11#include <linux/sched/stat.h>
12#include <linux/sched/task.h>
13#include <linux/sched/task_stack.h>
14#include <linux/sched/cputime.h>
15#include <linux/interrupt.h>
16#include <linux/module.h>
17#include <linux/capability.h>
18#include <linux/completion.h>
19#include <linux/personality.h>
20#include <linux/tty.h>
21#include <linux/iocontext.h>
22#include <linux/key.h>
23#include <linux/cpu.h>
24#include <linux/acct.h>
25#include <linux/tsacct_kern.h>
26#include <linux/file.h>
27#include <linux/fdtable.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/blkdev.h>
51#include <linux/task_io_accounting_ops.h>
52#include <linux/tracehook.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/random.h>
63#include <linux/rcuwait.h>
64#include <linux/compat.h>
65
66#include <linux/uaccess.h>
67#include <asm/unistd.h>
68#include <asm/pgtable.h>
69#include <asm/mmu_context.h>
70
71static void __unhash_process(struct task_struct *p, bool group_dead)
72{
73 nr_threads--;
74 detach_pid(p, PIDTYPE_PID);
75 if (group_dead) {
76 detach_pid(p, PIDTYPE_PGID);
77 detach_pid(p, PIDTYPE_SID);
78
79 list_del_rcu(&p->tasks);
80 list_del_init(&p->sibling);
81 __this_cpu_dec(process_counts);
82 }
83 list_del_rcu(&p->thread_group);
84 list_del_rcu(&p->thread_node);
85}
86
87/*
88 * This function expects the tasklist_lock write-locked.
89 */
90static void __exit_signal(struct task_struct *tsk)
91{
92 struct signal_struct *sig = tsk->signal;
93 bool group_dead = thread_group_leader(tsk);
94 struct sighand_struct *sighand;
95 struct tty_struct *uninitialized_var(tty);
96 u64 utime, stime;
97
98 sighand = rcu_dereference_check(tsk->sighand,
99 lockdep_tasklist_lock_is_held());
100 spin_lock(&sighand->siglock);
101
102#ifdef CONFIG_POSIX_TIMERS
103 posix_cpu_timers_exit(tsk);
104 if (group_dead) {
105 posix_cpu_timers_exit_group(tsk);
106 } else {
107 /*
108 * This can only happen if the caller is de_thread().
109 * FIXME: this is the temporary hack, we should teach
110 * posix-cpu-timers to handle this case correctly.
111 */
112 if (unlikely(has_group_leader_pid(tsk)))
113 posix_cpu_timers_exit_group(tsk);
114 }
115#endif
116
117 if (group_dead) {
118 tty = sig->tty;
119 sig->tty = NULL;
120 } else {
121 /*
122 * If there is any task waiting for the group exit
123 * then notify it:
124 */
125 if (sig->notify_count > 0 && !--sig->notify_count)
126 wake_up_process(sig->group_exit_task);
127
128 if (tsk == sig->curr_target)
129 sig->curr_target = next_thread(tsk);
130 }
131
132 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
133 sizeof(unsigned long long));
134
135 /*
136 * Accumulate here the counters for all threads as they die. We could
137 * skip the group leader because it is the last user of signal_struct,
138 * but we want to avoid the race with thread_group_cputime() which can
139 * see the empty ->thread_head list.
140 */
141 task_cputime(tsk, &utime, &stime);
142 write_seqlock(&sig->stats_lock);
143 sig->utime += utime;
144 sig->stime += stime;
145 sig->gtime += task_gtime(tsk);
146 sig->min_flt += tsk->min_flt;
147 sig->maj_flt += tsk->maj_flt;
148 sig->nvcsw += tsk->nvcsw;
149 sig->nivcsw += tsk->nivcsw;
150 sig->inblock += task_io_get_inblock(tsk);
151 sig->oublock += task_io_get_oublock(tsk);
152 task_io_accounting_add(&sig->ioac, &tsk->ioac);
153 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
154 sig->nr_threads--;
155 __unhash_process(tsk, group_dead);
156 write_sequnlock(&sig->stats_lock);
157
158 /*
159 * Do this under ->siglock, we can race with another thread
160 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
161 */
162 flush_sigqueue(&tsk->pending);
163 tsk->sighand = NULL;
164 spin_unlock(&sighand->siglock);
165
166 __cleanup_sighand(sighand);
167 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
168 if (group_dead) {
169 flush_sigqueue(&sig->shared_pending);
170 tty_kref_put(tty);
171 }
172}
173
174static void delayed_put_task_struct(struct rcu_head *rhp)
175{
176 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
177
178 perf_event_delayed_put(tsk);
179 trace_sched_process_free(tsk);
180 put_task_struct(tsk);
181}
182
183
184void release_task(struct task_struct *p)
185{
186 struct task_struct *leader;
187 int zap_leader;
188repeat:
189 /* don't need to get the RCU readlock here - the process is dead and
190 * can't be modifying its own credentials. But shut RCU-lockdep up */
191 rcu_read_lock();
192 atomic_dec(&__task_cred(p)->user->processes);
193 rcu_read_unlock();
194
195 proc_flush_task(p);
196
197 write_lock_irq(&tasklist_lock);
198 ptrace_release_task(p);
199 __exit_signal(p);
200
201 /*
202 * If we are the last non-leader member of the thread
203 * group, and the leader is zombie, then notify the
204 * group leader's parent process. (if it wants notification.)
205 */
206 zap_leader = 0;
207 leader = p->group_leader;
208 if (leader != p && thread_group_empty(leader)
209 && leader->exit_state == EXIT_ZOMBIE) {
210 /*
211 * If we were the last child thread and the leader has
212 * exited already, and the leader's parent ignores SIGCHLD,
213 * then we are the one who should release the leader.
214 */
215 zap_leader = do_notify_parent(leader, leader->exit_signal);
216 if (zap_leader)
217 leader->exit_state = EXIT_DEAD;
218 }
219
220 write_unlock_irq(&tasklist_lock);
221 release_thread(p);
222 call_rcu(&p->rcu, delayed_put_task_struct);
223
224 p = leader;
225 if (unlikely(zap_leader))
226 goto repeat;
227}
228
229/*
230 * Note that if this function returns a valid task_struct pointer (!NULL)
231 * task->usage must remain >0 for the duration of the RCU critical section.
232 */
233struct task_struct *task_rcu_dereference(struct task_struct **ptask)
234{
235 struct sighand_struct *sighand;
236 struct task_struct *task;
237
238 /*
239 * We need to verify that release_task() was not called and thus
240 * delayed_put_task_struct() can't run and drop the last reference
241 * before rcu_read_unlock(). We check task->sighand != NULL,
242 * but we can read the already freed and reused memory.
243 */
244retry:
245 task = rcu_dereference(*ptask);
246 if (!task)
247 return NULL;
248
249 probe_kernel_address(&task->sighand, sighand);
250
251 /*
252 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
253 * was already freed we can not miss the preceding update of this
254 * pointer.
255 */
256 smp_rmb();
257 if (unlikely(task != READ_ONCE(*ptask)))
258 goto retry;
259
260 /*
261 * We've re-checked that "task == *ptask", now we have two different
262 * cases:
263 *
264 * 1. This is actually the same task/task_struct. In this case
265 * sighand != NULL tells us it is still alive.
266 *
267 * 2. This is another task which got the same memory for task_struct.
268 * We can't know this of course, and we can not trust
269 * sighand != NULL.
270 *
271 * In this case we actually return a random value, but this is
272 * correct.
273 *
274 * If we return NULL - we can pretend that we actually noticed that
275 * *ptask was updated when the previous task has exited. Or pretend
276 * that probe_slab_address(&sighand) reads NULL.
277 *
278 * If we return the new task (because sighand is not NULL for any
279 * reason) - this is fine too. This (new) task can't go away before
280 * another gp pass.
281 *
282 * And note: We could even eliminate the false positive if re-read
283 * task->sighand once again to avoid the falsely NULL. But this case
284 * is very unlikely so we don't care.
285 */
286 if (!sighand)
287 return NULL;
288
289 return task;
290}
291
292void rcuwait_wake_up(struct rcuwait *w)
293{
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_trywake() 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_rmb(); /* (B) */
310
311 /*
312 * Avoid using task_rcu_dereference() magic as long as we are careful,
313 * see comment in rcuwait_wait_event() regarding ->exit_state.
314 */
315 task = rcu_dereference(w->task);
316 if (task)
317 wake_up_process(task);
318 rcu_read_unlock();
319}
320
321/*
322 * Determine if a process group is "orphaned", according to the POSIX
323 * definition in 2.2.2.52. Orphaned process groups are not to be affected
324 * by terminal-generated stop signals. Newly orphaned process groups are
325 * to receive a SIGHUP and a SIGCONT.
326 *
327 * "I ask you, have you ever known what it is to be an orphan?"
328 */
329static int will_become_orphaned_pgrp(struct pid *pgrp,
330 struct task_struct *ignored_task)
331{
332 struct task_struct *p;
333
334 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
335 if ((p == ignored_task) ||
336 (p->exit_state && thread_group_empty(p)) ||
337 is_global_init(p->real_parent))
338 continue;
339
340 if (task_pgrp(p->real_parent) != pgrp &&
341 task_session(p->real_parent) == task_session(p))
342 return 0;
343 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
344
345 return 1;
346}
347
348int is_current_pgrp_orphaned(void)
349{
350 int retval;
351
352 read_lock(&tasklist_lock);
353 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
354 read_unlock(&tasklist_lock);
355
356 return retval;
357}
358
359static bool has_stopped_jobs(struct pid *pgrp)
360{
361 struct task_struct *p;
362
363 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
364 if (p->signal->flags & SIGNAL_STOP_STOPPED)
365 return true;
366 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
367
368 return false;
369}
370
371/*
372 * Check to see if any process groups have become orphaned as
373 * a result of our exiting, and if they have any stopped jobs,
374 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
375 */
376static void
377kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
378{
379 struct pid *pgrp = task_pgrp(tsk);
380 struct task_struct *ignored_task = tsk;
381
382 if (!parent)
383 /* exit: our father is in a different pgrp than
384 * we are and we were the only connection outside.
385 */
386 parent = tsk->real_parent;
387 else
388 /* reparent: our child is in a different pgrp than
389 * we are, and it was the only connection outside.
390 */
391 ignored_task = NULL;
392
393 if (task_pgrp(parent) != pgrp &&
394 task_session(parent) == task_session(tsk) &&
395 will_become_orphaned_pgrp(pgrp, ignored_task) &&
396 has_stopped_jobs(pgrp)) {
397 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
398 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
399 }
400}
401
402#ifdef CONFIG_MEMCG
403/*
404 * A task is exiting. If it owned this mm, find a new owner for the mm.
405 */
406void mm_update_next_owner(struct mm_struct *mm)
407{
408 struct task_struct *c, *g, *p = current;
409
410retry:
411 /*
412 * If the exiting or execing task is not the owner, it's
413 * someone else's problem.
414 */
415 if (mm->owner != p)
416 return;
417 /*
418 * The current owner is exiting/execing and there are no other
419 * candidates. Do not leave the mm pointing to a possibly
420 * freed task structure.
421 */
422 if (atomic_read(&mm->mm_users) <= 1) {
423 mm->owner = NULL;
424 return;
425 }
426
427 read_lock(&tasklist_lock);
428 /*
429 * Search in the children
430 */
431 list_for_each_entry(c, &p->children, sibling) {
432 if (c->mm == mm)
433 goto assign_new_owner;
434 }
435
436 /*
437 * Search in the siblings
438 */
439 list_for_each_entry(c, &p->real_parent->children, sibling) {
440 if (c->mm == mm)
441 goto assign_new_owner;
442 }
443
444 /*
445 * Search through everything else, we should not get here often.
446 */
447 for_each_process(g) {
448 if (g->flags & PF_KTHREAD)
449 continue;
450 for_each_thread(g, c) {
451 if (c->mm == mm)
452 goto assign_new_owner;
453 if (c->mm)
454 break;
455 }
456 }
457 read_unlock(&tasklist_lock);
458 /*
459 * We found no owner yet mm_users > 1: this implies that we are
460 * most likely racing with swapoff (try_to_unuse()) or /proc or
461 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
462 */
463 mm->owner = NULL;
464 return;
465
466assign_new_owner:
467 BUG_ON(c == p);
468 get_task_struct(c);
469 /*
470 * The task_lock protects c->mm from changing.
471 * We always want mm->owner->mm == mm
472 */
473 task_lock(c);
474 /*
475 * Delay read_unlock() till we have the task_lock()
476 * to ensure that c does not slip away underneath us
477 */
478 read_unlock(&tasklist_lock);
479 if (c->mm != mm) {
480 task_unlock(c);
481 put_task_struct(c);
482 goto retry;
483 }
484 mm->owner = c;
485 task_unlock(c);
486 put_task_struct(c);
487}
488#endif /* CONFIG_MEMCG */
489
490/*
491 * Turn us into a lazy TLB process if we
492 * aren't already..
493 */
494static void exit_mm(void)
495{
496 struct mm_struct *mm = current->mm;
497 struct core_state *core_state;
498
499 mm_release(current, mm);
500 if (!mm)
501 return;
502 sync_mm_rss(mm);
503 /*
504 * Serialize with any possible pending coredump.
505 * We must hold mmap_sem around checking core_state
506 * and clearing tsk->mm. The core-inducing thread
507 * will increment ->nr_threads for each thread in the
508 * group with ->mm != NULL.
509 */
510 down_read(&mm->mmap_sem);
511 core_state = mm->core_state;
512 if (core_state) {
513 struct core_thread self;
514
515 up_read(&mm->mmap_sem);
516
517 self.task = current;
518 self.next = xchg(&core_state->dumper.next, &self);
519 /*
520 * Implies mb(), the result of xchg() must be visible
521 * to core_state->dumper.
522 */
523 if (atomic_dec_and_test(&core_state->nr_threads))
524 complete(&core_state->startup);
525
526 for (;;) {
527 set_current_state(TASK_UNINTERRUPTIBLE);
528 if (!self.task) /* see coredump_finish() */
529 break;
530 freezable_schedule();
531 }
532 __set_current_state(TASK_RUNNING);
533 down_read(&mm->mmap_sem);
534 }
535 mmgrab(mm);
536 BUG_ON(mm != current->active_mm);
537 /* more a memory barrier than a real lock */
538 task_lock(current);
539 current->mm = NULL;
540 up_read(&mm->mmap_sem);
541 enter_lazy_tlb(mm, current);
542 task_unlock(current);
543 mm_update_next_owner(mm);
544 mmput(mm);
545 if (test_thread_flag(TIF_MEMDIE))
546 exit_oom_victim();
547}
548
549static struct task_struct *find_alive_thread(struct task_struct *p)
550{
551 struct task_struct *t;
552
553 for_each_thread(p, t) {
554 if (!(t->flags & PF_EXITING))
555 return t;
556 }
557 return NULL;
558}
559
560static struct task_struct *find_child_reaper(struct task_struct *father)
561 __releases(&tasklist_lock)
562 __acquires(&tasklist_lock)
563{
564 struct pid_namespace *pid_ns = task_active_pid_ns(father);
565 struct task_struct *reaper = pid_ns->child_reaper;
566
567 if (likely(reaper != father))
568 return reaper;
569
570 reaper = find_alive_thread(father);
571 if (reaper) {
572 pid_ns->child_reaper = reaper;
573 return reaper;
574 }
575
576 write_unlock_irq(&tasklist_lock);
577 if (unlikely(pid_ns == &init_pid_ns)) {
578 panic("Attempted to kill init! exitcode=0x%08x\n",
579 father->signal->group_exit_code ?: father->exit_code);
580 }
581 zap_pid_ns_processes(pid_ns);
582 write_lock_irq(&tasklist_lock);
583
584 return father;
585}
586
587/*
588 * When we die, we re-parent all our children, and try to:
589 * 1. give them to another thread in our thread group, if such a member exists
590 * 2. give it to the first ancestor process which prctl'd itself as a
591 * child_subreaper for its children (like a service manager)
592 * 3. give it to the init process (PID 1) in our pid namespace
593 */
594static struct task_struct *find_new_reaper(struct task_struct *father,
595 struct task_struct *child_reaper)
596{
597 struct task_struct *thread, *reaper;
598
599 thread = find_alive_thread(father);
600 if (thread)
601 return thread;
602
603 if (father->signal->has_child_subreaper) {
604 unsigned int ns_level = task_pid(father)->level;
605 /*
606 * Find the first ->is_child_subreaper ancestor in our pid_ns.
607 * We can't check reaper != child_reaper to ensure we do not
608 * cross the namespaces, the exiting parent could be injected
609 * by setns() + fork().
610 * We check pid->level, this is slightly more efficient than
611 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
612 */
613 for (reaper = father->real_parent;
614 task_pid(reaper)->level == ns_level;
615 reaper = reaper->real_parent) {
616 if (reaper == &init_task)
617 break;
618 if (!reaper->signal->is_child_subreaper)
619 continue;
620 thread = find_alive_thread(reaper);
621 if (thread)
622 return thread;
623 }
624 }
625
626 return child_reaper;
627}
628
629/*
630* Any that need to be release_task'd are put on the @dead list.
631 */
632static void reparent_leader(struct task_struct *father, struct task_struct *p,
633 struct list_head *dead)
634{
635 if (unlikely(p->exit_state == EXIT_DEAD))
636 return;
637
638 /* We don't want people slaying init. */
639 p->exit_signal = SIGCHLD;
640
641 /* If it has exited notify the new parent about this child's death. */
642 if (!p->ptrace &&
643 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
644 if (do_notify_parent(p, p->exit_signal)) {
645 p->exit_state = EXIT_DEAD;
646 list_add(&p->ptrace_entry, dead);
647 }
648 }
649
650 kill_orphaned_pgrp(p, father);
651}
652
653/*
654 * This does two things:
655 *
656 * A. Make init inherit all the child processes
657 * B. Check to see if any process groups have become orphaned
658 * as a result of our exiting, and if they have any stopped
659 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
660 */
661static void forget_original_parent(struct task_struct *father,
662 struct list_head *dead)
663{
664 struct task_struct *p, *t, *reaper;
665
666 if (unlikely(!list_empty(&father->ptraced)))
667 exit_ptrace(father, dead);
668
669 /* Can drop and reacquire tasklist_lock */
670 reaper = find_child_reaper(father);
671 if (list_empty(&father->children))
672 return;
673
674 reaper = find_new_reaper(father, reaper);
675 list_for_each_entry(p, &father->children, sibling) {
676 for_each_thread(p, t) {
677 t->real_parent = reaper;
678 BUG_ON((!t->ptrace) != (t->parent == father));
679 if (likely(!t->ptrace))
680 t->parent = t->real_parent;
681 if (t->pdeath_signal)
682 group_send_sig_info(t->pdeath_signal,
683 SEND_SIG_NOINFO, t);
684 }
685 /*
686 * If this is a threaded reparent there is no need to
687 * notify anyone anything has happened.
688 */
689 if (!same_thread_group(reaper, father))
690 reparent_leader(father, p, dead);
691 }
692 list_splice_tail_init(&father->children, &reaper->children);
693}
694
695/*
696 * Send signals to all our closest relatives so that they know
697 * to properly mourn us..
698 */
699static void exit_notify(struct task_struct *tsk, int group_dead)
700{
701 bool autoreap;
702 struct task_struct *p, *n;
703 LIST_HEAD(dead);
704
705 write_lock_irq(&tasklist_lock);
706 forget_original_parent(tsk, &dead);
707
708 if (group_dead)
709 kill_orphaned_pgrp(tsk->group_leader, NULL);
710
711 if (unlikely(tsk->ptrace)) {
712 int sig = thread_group_leader(tsk) &&
713 thread_group_empty(tsk) &&
714 !ptrace_reparented(tsk) ?
715 tsk->exit_signal : SIGCHLD;
716 autoreap = do_notify_parent(tsk, sig);
717 } else if (thread_group_leader(tsk)) {
718 autoreap = thread_group_empty(tsk) &&
719 do_notify_parent(tsk, tsk->exit_signal);
720 } else {
721 autoreap = true;
722 }
723
724 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
725 if (tsk->exit_state == EXIT_DEAD)
726 list_add(&tsk->ptrace_entry, &dead);
727
728 /* mt-exec, de_thread() is waiting for group leader */
729 if (unlikely(tsk->signal->notify_count < 0))
730 wake_up_process(tsk->signal->group_exit_task);
731 write_unlock_irq(&tasklist_lock);
732
733 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
734 list_del_init(&p->ptrace_entry);
735 release_task(p);
736 }
737}
738
739#ifdef CONFIG_DEBUG_STACK_USAGE
740static void check_stack_usage(void)
741{
742 static DEFINE_SPINLOCK(low_water_lock);
743 static int lowest_to_date = THREAD_SIZE;
744 unsigned long free;
745
746 free = stack_not_used(current);
747
748 if (free >= lowest_to_date)
749 return;
750
751 spin_lock(&low_water_lock);
752 if (free < lowest_to_date) {
753 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
754 current->comm, task_pid_nr(current), free);
755 lowest_to_date = free;
756 }
757 spin_unlock(&low_water_lock);
758}
759#else
760static inline void check_stack_usage(void) {}
761#endif
762
763void __noreturn do_exit(long code)
764{
765 struct task_struct *tsk = current;
766 int group_dead;
767
768 profile_task_exit(tsk);
769 kcov_task_exit(tsk);
770
771 WARN_ON(blk_needs_flush_plug(tsk));
772
773 if (unlikely(in_interrupt()))
774 panic("Aiee, killing interrupt handler!");
775 if (unlikely(!tsk->pid))
776 panic("Attempted to kill the idle task!");
777
778 /*
779 * If do_exit is called because this processes oopsed, it's possible
780 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
781 * continuing. Amongst other possible reasons, this is to prevent
782 * mm_release()->clear_child_tid() from writing to a user-controlled
783 * kernel address.
784 */
785 set_fs(USER_DS);
786
787 ptrace_event(PTRACE_EVENT_EXIT, code);
788
789 validate_creds_for_do_exit(tsk);
790
791 /*
792 * We're taking recursive faults here in do_exit. Safest is to just
793 * leave this task alone and wait for reboot.
794 */
795 if (unlikely(tsk->flags & PF_EXITING)) {
796 pr_alert("Fixing recursive fault but reboot is needed!\n");
797 /*
798 * We can do this unlocked here. The futex code uses
799 * this flag just to verify whether the pi state
800 * cleanup has been done or not. In the worst case it
801 * loops once more. We pretend that the cleanup was
802 * done as there is no way to return. Either the
803 * OWNER_DIED bit is set by now or we push the blocked
804 * task into the wait for ever nirwana as well.
805 */
806 tsk->flags |= PF_EXITPIDONE;
807 set_current_state(TASK_UNINTERRUPTIBLE);
808 schedule();
809 }
810
811 exit_signals(tsk); /* sets PF_EXITING */
812 /*
813 * Ensure that all new tsk->pi_lock acquisitions must observe
814 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
815 */
816 smp_mb();
817 /*
818 * Ensure that we must observe the pi_state in exit_mm() ->
819 * mm_release() -> exit_pi_state_list().
820 */
821 raw_spin_lock_irq(&tsk->pi_lock);
822 raw_spin_unlock_irq(&tsk->pi_lock);
823
824 if (unlikely(in_atomic())) {
825 pr_info("note: %s[%d] exited with preempt_count %d\n",
826 current->comm, task_pid_nr(current),
827 preempt_count());
828 preempt_count_set(PREEMPT_ENABLED);
829 }
830
831 /* sync mm's RSS info before statistics gathering */
832 if (tsk->mm)
833 sync_mm_rss(tsk->mm);
834 acct_update_integrals(tsk);
835 group_dead = atomic_dec_and_test(&tsk->signal->live);
836 if (group_dead) {
837#ifdef CONFIG_POSIX_TIMERS
838 hrtimer_cancel(&tsk->signal->real_timer);
839 exit_itimers(tsk->signal);
840#endif
841 if (tsk->mm)
842 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
843 }
844 acct_collect(code, group_dead);
845 if (group_dead)
846 tty_audit_exit();
847 audit_free(tsk);
848
849 tsk->exit_code = code;
850 taskstats_exit(tsk, group_dead);
851
852 exit_mm();
853
854 if (group_dead)
855 acct_process();
856 trace_sched_process_exit(tsk);
857
858 exit_sem(tsk);
859 exit_shm(tsk);
860 exit_files(tsk);
861 exit_fs(tsk);
862 if (group_dead)
863 disassociate_ctty(1);
864 exit_task_namespaces(tsk);
865 exit_task_work(tsk);
866 exit_thread(tsk);
867
868 /*
869 * Flush inherited counters to the parent - before the parent
870 * gets woken up by child-exit notifications.
871 *
872 * because of cgroup mode, must be called before cgroup_exit()
873 */
874 perf_event_exit_task(tsk);
875
876 sched_autogroup_exit_task(tsk);
877 cgroup_exit(tsk);
878
879 /*
880 * FIXME: do that only when needed, using sched_exit tracepoint
881 */
882 flush_ptrace_hw_breakpoint(tsk);
883
884 exit_tasks_rcu_start();
885 exit_notify(tsk, group_dead);
886 proc_exit_connector(tsk);
887 mpol_put_task_policy(tsk);
888#ifdef CONFIG_FUTEX
889 if (unlikely(current->pi_state_cache))
890 kfree(current->pi_state_cache);
891#endif
892 /*
893 * Make sure we are holding no locks:
894 */
895 debug_check_no_locks_held();
896 /*
897 * We can do this unlocked here. The futex code uses this flag
898 * just to verify whether the pi state cleanup has been done
899 * or not. In the worst case it loops once more.
900 */
901 tsk->flags |= PF_EXITPIDONE;
902
903 if (tsk->io_context)
904 exit_io_context(tsk);
905
906 if (tsk->splice_pipe)
907 free_pipe_info(tsk->splice_pipe);
908
909 if (tsk->task_frag.page)
910 put_page(tsk->task_frag.page);
911
912 validate_creds_for_do_exit(tsk);
913
914 check_stack_usage();
915 preempt_disable();
916 if (tsk->nr_dirtied)
917 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
918 exit_rcu();
919 exit_tasks_rcu_finish();
920
921 lockdep_free_task(tsk);
922 do_task_dead();
923}
924EXPORT_SYMBOL_GPL(do_exit);
925
926void complete_and_exit(struct completion *comp, long code)
927{
928 if (comp)
929 complete(comp);
930
931 do_exit(code);
932}
933EXPORT_SYMBOL(complete_and_exit);
934
935SYSCALL_DEFINE1(exit, int, error_code)
936{
937 do_exit((error_code&0xff)<<8);
938}
939
940/*
941 * Take down every thread in the group. This is called by fatal signals
942 * as well as by sys_exit_group (below).
943 */
944void
945do_group_exit(int exit_code)
946{
947 struct signal_struct *sig = current->signal;
948
949 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
950
951 if (signal_group_exit(sig))
952 exit_code = sig->group_exit_code;
953 else if (!thread_group_empty(current)) {
954 struct sighand_struct *const sighand = current->sighand;
955
956 spin_lock_irq(&sighand->siglock);
957 if (signal_group_exit(sig))
958 /* Another thread got here before we took the lock. */
959 exit_code = sig->group_exit_code;
960 else {
961 sig->group_exit_code = exit_code;
962 sig->flags = SIGNAL_GROUP_EXIT;
963 zap_other_threads(current);
964 }
965 spin_unlock_irq(&sighand->siglock);
966 }
967
968 do_exit(exit_code);
969 /* NOTREACHED */
970}
971
972/*
973 * this kills every thread in the thread group. Note that any externally
974 * wait4()-ing process will get the correct exit code - even if this
975 * thread is not the thread group leader.
976 */
977SYSCALL_DEFINE1(exit_group, int, error_code)
978{
979 do_group_exit((error_code & 0xff) << 8);
980 /* NOTREACHED */
981 return 0;
982}
983
984struct waitid_info {
985 pid_t pid;
986 uid_t uid;
987 int status;
988 int cause;
989};
990
991struct wait_opts {
992 enum pid_type wo_type;
993 int wo_flags;
994 struct pid *wo_pid;
995
996 struct waitid_info *wo_info;
997 int wo_stat;
998 struct rusage *wo_rusage;
999
1000 wait_queue_entry_t child_wait;
1001 int notask_error;
1002};
1003
1004static inline
1005struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1006{
1007 if (type != PIDTYPE_PID)
1008 task = task->group_leader;
1009 return task->pids[type].pid;
1010}
1011
1012static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1013{
1014 return wo->wo_type == PIDTYPE_MAX ||
1015 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1016}
1017
1018static int
1019eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1020{
1021 if (!eligible_pid(wo, p))
1022 return 0;
1023
1024 /*
1025 * Wait for all children (clone and not) if __WALL is set or
1026 * if it is traced by us.
1027 */
1028 if (ptrace || (wo->wo_flags & __WALL))
1029 return 1;
1030
1031 /*
1032 * Otherwise, wait for clone children *only* if __WCLONE is set;
1033 * otherwise, wait for non-clone children *only*.
1034 *
1035 * Note: a "clone" child here is one that reports to its parent
1036 * using a signal other than SIGCHLD, or a non-leader thread which
1037 * we can only see if it is traced by us.
1038 */
1039 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1040 return 0;
1041
1042 return 1;
1043}
1044
1045/*
1046 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1047 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1048 * the lock and this task is uninteresting. If we return nonzero, we have
1049 * released the lock and the system call should return.
1050 */
1051static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1052{
1053 int state, status;
1054 pid_t pid = task_pid_vnr(p);
1055 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1056 struct waitid_info *infop;
1057
1058 if (!likely(wo->wo_flags & WEXITED))
1059 return 0;
1060
1061 if (unlikely(wo->wo_flags & WNOWAIT)) {
1062 status = p->exit_code;
1063 get_task_struct(p);
1064 read_unlock(&tasklist_lock);
1065 sched_annotate_sleep();
1066 if (wo->wo_rusage)
1067 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1068 put_task_struct(p);
1069 goto out_info;
1070 }
1071 /*
1072 * Move the task's state to DEAD/TRACE, only one thread can do this.
1073 */
1074 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1075 EXIT_TRACE : EXIT_DEAD;
1076 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1077 return 0;
1078 /*
1079 * We own this thread, nobody else can reap it.
1080 */
1081 read_unlock(&tasklist_lock);
1082 sched_annotate_sleep();
1083
1084 /*
1085 * Check thread_group_leader() to exclude the traced sub-threads.
1086 */
1087 if (state == EXIT_DEAD && thread_group_leader(p)) {
1088 struct signal_struct *sig = p->signal;
1089 struct signal_struct *psig = current->signal;
1090 unsigned long maxrss;
1091 u64 tgutime, tgstime;
1092
1093 /*
1094 * The resource counters for the group leader are in its
1095 * own task_struct. Those for dead threads in the group
1096 * are in its signal_struct, as are those for the child
1097 * processes it has previously reaped. All these
1098 * accumulate in the parent's signal_struct c* fields.
1099 *
1100 * We don't bother to take a lock here to protect these
1101 * p->signal fields because the whole thread group is dead
1102 * and nobody can change them.
1103 *
1104 * psig->stats_lock also protects us from our sub-theads
1105 * which can reap other children at the same time. Until
1106 * we change k_getrusage()-like users to rely on this lock
1107 * we have to take ->siglock as well.
1108 *
1109 * We use thread_group_cputime_adjusted() to get times for
1110 * the thread group, which consolidates times for all threads
1111 * in the group including the group leader.
1112 */
1113 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1114 spin_lock_irq(¤t->sighand->siglock);
1115 write_seqlock(&psig->stats_lock);
1116 psig->cutime += tgutime + sig->cutime;
1117 psig->cstime += tgstime + sig->cstime;
1118 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1119 psig->cmin_flt +=
1120 p->min_flt + sig->min_flt + sig->cmin_flt;
1121 psig->cmaj_flt +=
1122 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1123 psig->cnvcsw +=
1124 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1125 psig->cnivcsw +=
1126 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1127 psig->cinblock +=
1128 task_io_get_inblock(p) +
1129 sig->inblock + sig->cinblock;
1130 psig->coublock +=
1131 task_io_get_oublock(p) +
1132 sig->oublock + sig->coublock;
1133 maxrss = max(sig->maxrss, sig->cmaxrss);
1134 if (psig->cmaxrss < maxrss)
1135 psig->cmaxrss = maxrss;
1136 task_io_accounting_add(&psig->ioac, &p->ioac);
1137 task_io_accounting_add(&psig->ioac, &sig->ioac);
1138 write_sequnlock(&psig->stats_lock);
1139 spin_unlock_irq(¤t->sighand->siglock);
1140 }
1141
1142 if (wo->wo_rusage)
1143 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1144 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1145 ? p->signal->group_exit_code : p->exit_code;
1146 wo->wo_stat = status;
1147
1148 if (state == EXIT_TRACE) {
1149 write_lock_irq(&tasklist_lock);
1150 /* We dropped tasklist, ptracer could die and untrace */
1151 ptrace_unlink(p);
1152
1153 /* If parent wants a zombie, don't release it now */
1154 state = EXIT_ZOMBIE;
1155 if (do_notify_parent(p, p->exit_signal))
1156 state = EXIT_DEAD;
1157 p->exit_state = state;
1158 write_unlock_irq(&tasklist_lock);
1159 }
1160 if (state == EXIT_DEAD)
1161 release_task(p);
1162
1163out_info:
1164 infop = wo->wo_info;
1165 if (infop) {
1166 if ((status & 0x7f) == 0) {
1167 infop->cause = CLD_EXITED;
1168 infop->status = status >> 8;
1169 } else {
1170 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1171 infop->status = status & 0x7f;
1172 }
1173 infop->pid = pid;
1174 infop->uid = uid;
1175 }
1176
1177 return pid;
1178}
1179
1180static int *task_stopped_code(struct task_struct *p, bool ptrace)
1181{
1182 if (ptrace) {
1183 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1184 return &p->exit_code;
1185 } else {
1186 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1187 return &p->signal->group_exit_code;
1188 }
1189 return NULL;
1190}
1191
1192/**
1193 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1194 * @wo: wait options
1195 * @ptrace: is the wait for ptrace
1196 * @p: task to wait for
1197 *
1198 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1199 *
1200 * CONTEXT:
1201 * read_lock(&tasklist_lock), which is released if return value is
1202 * non-zero. Also, grabs and releases @p->sighand->siglock.
1203 *
1204 * RETURNS:
1205 * 0 if wait condition didn't exist and search for other wait conditions
1206 * should continue. Non-zero return, -errno on failure and @p's pid on
1207 * success, implies that tasklist_lock is released and wait condition
1208 * search should terminate.
1209 */
1210static int wait_task_stopped(struct wait_opts *wo,
1211 int ptrace, struct task_struct *p)
1212{
1213 struct waitid_info *infop;
1214 int exit_code, *p_code, why;
1215 uid_t uid = 0; /* unneeded, required by compiler */
1216 pid_t pid;
1217
1218 /*
1219 * Traditionally we see ptrace'd stopped tasks regardless of options.
1220 */
1221 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1222 return 0;
1223
1224 if (!task_stopped_code(p, ptrace))
1225 return 0;
1226
1227 exit_code = 0;
1228 spin_lock_irq(&p->sighand->siglock);
1229
1230 p_code = task_stopped_code(p, ptrace);
1231 if (unlikely(!p_code))
1232 goto unlock_sig;
1233
1234 exit_code = *p_code;
1235 if (!exit_code)
1236 goto unlock_sig;
1237
1238 if (!unlikely(wo->wo_flags & WNOWAIT))
1239 *p_code = 0;
1240
1241 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1242unlock_sig:
1243 spin_unlock_irq(&p->sighand->siglock);
1244 if (!exit_code)
1245 return 0;
1246
1247 /*
1248 * Now we are pretty sure this task is interesting.
1249 * Make sure it doesn't get reaped out from under us while we
1250 * give up the lock and then examine it below. We don't want to
1251 * keep holding onto the tasklist_lock while we call getrusage and
1252 * possibly take page faults for user memory.
1253 */
1254 get_task_struct(p);
1255 pid = task_pid_vnr(p);
1256 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1257 read_unlock(&tasklist_lock);
1258 sched_annotate_sleep();
1259 if (wo->wo_rusage)
1260 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1261 put_task_struct(p);
1262
1263 if (likely(!(wo->wo_flags & WNOWAIT)))
1264 wo->wo_stat = (exit_code << 8) | 0x7f;
1265
1266 infop = wo->wo_info;
1267 if (infop) {
1268 infop->cause = why;
1269 infop->status = exit_code;
1270 infop->pid = pid;
1271 infop->uid = uid;
1272 }
1273 return pid;
1274}
1275
1276/*
1277 * Handle do_wait work for one task in a live, non-stopped state.
1278 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1279 * the lock and this task is uninteresting. If we return nonzero, we have
1280 * released the lock and the system call should return.
1281 */
1282static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1283{
1284 struct waitid_info *infop;
1285 pid_t pid;
1286 uid_t uid;
1287
1288 if (!unlikely(wo->wo_flags & WCONTINUED))
1289 return 0;
1290
1291 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1292 return 0;
1293
1294 spin_lock_irq(&p->sighand->siglock);
1295 /* Re-check with the lock held. */
1296 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1297 spin_unlock_irq(&p->sighand->siglock);
1298 return 0;
1299 }
1300 if (!unlikely(wo->wo_flags & WNOWAIT))
1301 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1302 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1303 spin_unlock_irq(&p->sighand->siglock);
1304
1305 pid = task_pid_vnr(p);
1306 get_task_struct(p);
1307 read_unlock(&tasklist_lock);
1308 sched_annotate_sleep();
1309 if (wo->wo_rusage)
1310 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1311 put_task_struct(p);
1312
1313 infop = wo->wo_info;
1314 if (!infop) {
1315 wo->wo_stat = 0xffff;
1316 } else {
1317 infop->cause = CLD_CONTINUED;
1318 infop->pid = pid;
1319 infop->uid = uid;
1320 infop->status = SIGCONT;
1321 }
1322 return pid;
1323}
1324
1325/*
1326 * Consider @p for a wait by @parent.
1327 *
1328 * -ECHILD should be in ->notask_error before the first call.
1329 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1330 * Returns zero if the search for a child should continue;
1331 * then ->notask_error is 0 if @p is an eligible child,
1332 * or still -ECHILD.
1333 */
1334static int wait_consider_task(struct wait_opts *wo, int ptrace,
1335 struct task_struct *p)
1336{
1337 /*
1338 * We can race with wait_task_zombie() from another thread.
1339 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1340 * can't confuse the checks below.
1341 */
1342 int exit_state = READ_ONCE(p->exit_state);
1343 int ret;
1344
1345 if (unlikely(exit_state == EXIT_DEAD))
1346 return 0;
1347
1348 ret = eligible_child(wo, ptrace, p);
1349 if (!ret)
1350 return ret;
1351
1352 if (unlikely(exit_state == EXIT_TRACE)) {
1353 /*
1354 * ptrace == 0 means we are the natural parent. In this case
1355 * we should clear notask_error, debugger will notify us.
1356 */
1357 if (likely(!ptrace))
1358 wo->notask_error = 0;
1359 return 0;
1360 }
1361
1362 if (likely(!ptrace) && unlikely(p->ptrace)) {
1363 /*
1364 * If it is traced by its real parent's group, just pretend
1365 * the caller is ptrace_do_wait() and reap this child if it
1366 * is zombie.
1367 *
1368 * This also hides group stop state from real parent; otherwise
1369 * a single stop can be reported twice as group and ptrace stop.
1370 * If a ptracer wants to distinguish these two events for its
1371 * own children it should create a separate process which takes
1372 * the role of real parent.
1373 */
1374 if (!ptrace_reparented(p))
1375 ptrace = 1;
1376 }
1377
1378 /* slay zombie? */
1379 if (exit_state == EXIT_ZOMBIE) {
1380 /* we don't reap group leaders with subthreads */
1381 if (!delay_group_leader(p)) {
1382 /*
1383 * A zombie ptracee is only visible to its ptracer.
1384 * Notification and reaping will be cascaded to the
1385 * real parent when the ptracer detaches.
1386 */
1387 if (unlikely(ptrace) || likely(!p->ptrace))
1388 return wait_task_zombie(wo, p);
1389 }
1390
1391 /*
1392 * Allow access to stopped/continued state via zombie by
1393 * falling through. Clearing of notask_error is complex.
1394 *
1395 * When !@ptrace:
1396 *
1397 * If WEXITED is set, notask_error should naturally be
1398 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1399 * so, if there are live subthreads, there are events to
1400 * wait for. If all subthreads are dead, it's still safe
1401 * to clear - this function will be called again in finite
1402 * amount time once all the subthreads are released and
1403 * will then return without clearing.
1404 *
1405 * When @ptrace:
1406 *
1407 * Stopped state is per-task and thus can't change once the
1408 * target task dies. Only continued and exited can happen.
1409 * Clear notask_error if WCONTINUED | WEXITED.
1410 */
1411 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1412 wo->notask_error = 0;
1413 } else {
1414 /*
1415 * @p is alive and it's gonna stop, continue or exit, so
1416 * there always is something to wait for.
1417 */
1418 wo->notask_error = 0;
1419 }
1420
1421 /*
1422 * Wait for stopped. Depending on @ptrace, different stopped state
1423 * is used and the two don't interact with each other.
1424 */
1425 ret = wait_task_stopped(wo, ptrace, p);
1426 if (ret)
1427 return ret;
1428
1429 /*
1430 * Wait for continued. There's only one continued state and the
1431 * ptracer can consume it which can confuse the real parent. Don't
1432 * use WCONTINUED from ptracer. You don't need or want it.
1433 */
1434 return wait_task_continued(wo, p);
1435}
1436
1437/*
1438 * Do the work of do_wait() for one thread in the group, @tsk.
1439 *
1440 * -ECHILD should be in ->notask_error before the first call.
1441 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1442 * Returns zero if the search for a child should continue; then
1443 * ->notask_error is 0 if there were any eligible children,
1444 * or still -ECHILD.
1445 */
1446static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1447{
1448 struct task_struct *p;
1449
1450 list_for_each_entry(p, &tsk->children, sibling) {
1451 int ret = wait_consider_task(wo, 0, p);
1452
1453 if (ret)
1454 return ret;
1455 }
1456
1457 return 0;
1458}
1459
1460static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1461{
1462 struct task_struct *p;
1463
1464 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1465 int ret = wait_consider_task(wo, 1, p);
1466
1467 if (ret)
1468 return ret;
1469 }
1470
1471 return 0;
1472}
1473
1474static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1475 int sync, void *key)
1476{
1477 struct wait_opts *wo = container_of(wait, struct wait_opts,
1478 child_wait);
1479 struct task_struct *p = key;
1480
1481 if (!eligible_pid(wo, p))
1482 return 0;
1483
1484 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1485 return 0;
1486
1487 return default_wake_function(wait, mode, sync, key);
1488}
1489
1490void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1491{
1492 __wake_up_sync_key(&parent->signal->wait_chldexit,
1493 TASK_INTERRUPTIBLE, 1, p);
1494}
1495
1496static long do_wait(struct wait_opts *wo)
1497{
1498 struct task_struct *tsk;
1499 int retval;
1500
1501 trace_sched_process_wait(wo->wo_pid);
1502
1503 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1504 wo->child_wait.private = current;
1505 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1506repeat:
1507 /*
1508 * If there is nothing that can match our criteria, just get out.
1509 * We will clear ->notask_error to zero if we see any child that
1510 * might later match our criteria, even if we are not able to reap
1511 * it yet.
1512 */
1513 wo->notask_error = -ECHILD;
1514 if ((wo->wo_type < PIDTYPE_MAX) &&
1515 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1516 goto notask;
1517
1518 set_current_state(TASK_INTERRUPTIBLE);
1519 read_lock(&tasklist_lock);
1520 tsk = current;
1521 do {
1522 retval = do_wait_thread(wo, tsk);
1523 if (retval)
1524 goto end;
1525
1526 retval = ptrace_do_wait(wo, tsk);
1527 if (retval)
1528 goto end;
1529
1530 if (wo->wo_flags & __WNOTHREAD)
1531 break;
1532 } while_each_thread(current, tsk);
1533 read_unlock(&tasklist_lock);
1534
1535notask:
1536 retval = wo->notask_error;
1537 if (!retval && !(wo->wo_flags & WNOHANG)) {
1538 retval = -ERESTARTSYS;
1539 if (!signal_pending(current)) {
1540 schedule();
1541 goto repeat;
1542 }
1543 }
1544end:
1545 __set_current_state(TASK_RUNNING);
1546 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1547 return retval;
1548}
1549
1550static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1551 int options, struct rusage *ru)
1552{
1553 struct wait_opts wo;
1554 struct pid *pid = NULL;
1555 enum pid_type type;
1556 long ret;
1557
1558 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1559 __WNOTHREAD|__WCLONE|__WALL))
1560 return -EINVAL;
1561 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1562 return -EINVAL;
1563
1564 switch (which) {
1565 case P_ALL:
1566 type = PIDTYPE_MAX;
1567 break;
1568 case P_PID:
1569 type = PIDTYPE_PID;
1570 if (upid <= 0)
1571 return -EINVAL;
1572 break;
1573 case P_PGID:
1574 type = PIDTYPE_PGID;
1575 if (upid <= 0)
1576 return -EINVAL;
1577 break;
1578 default:
1579 return -EINVAL;
1580 }
1581
1582 if (type < PIDTYPE_MAX)
1583 pid = find_get_pid(upid);
1584
1585 wo.wo_type = type;
1586 wo.wo_pid = pid;
1587 wo.wo_flags = options;
1588 wo.wo_info = infop;
1589 wo.wo_rusage = ru;
1590 ret = do_wait(&wo);
1591
1592 put_pid(pid);
1593 return ret;
1594}
1595
1596SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1597 infop, int, options, struct rusage __user *, ru)
1598{
1599 struct rusage r;
1600 struct waitid_info info = {.status = 0};
1601 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1602 int signo = 0;
1603
1604 if (err > 0) {
1605 signo = SIGCHLD;
1606 err = 0;
1607 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1608 return -EFAULT;
1609 }
1610 if (!infop)
1611 return err;
1612
1613 if (!access_ok(VERIFY_WRITE, infop, sizeof(*infop)))
1614 return -EFAULT;
1615
1616 user_access_begin();
1617 unsafe_put_user(signo, &infop->si_signo, Efault);
1618 unsafe_put_user(0, &infop->si_errno, Efault);
1619 unsafe_put_user(info.cause, &infop->si_code, Efault);
1620 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1621 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1622 unsafe_put_user(info.status, &infop->si_status, Efault);
1623 user_access_end();
1624 return err;
1625Efault:
1626 user_access_end();
1627 return -EFAULT;
1628}
1629
1630long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1631 struct rusage *ru)
1632{
1633 struct wait_opts wo;
1634 struct pid *pid = NULL;
1635 enum pid_type type;
1636 long ret;
1637
1638 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1639 __WNOTHREAD|__WCLONE|__WALL))
1640 return -EINVAL;
1641
1642 /* -INT_MIN is not defined */
1643 if (upid == INT_MIN)
1644 return -ESRCH;
1645
1646 if (upid == -1)
1647 type = PIDTYPE_MAX;
1648 else if (upid < 0) {
1649 type = PIDTYPE_PGID;
1650 pid = find_get_pid(-upid);
1651 } else if (upid == 0) {
1652 type = PIDTYPE_PGID;
1653 pid = get_task_pid(current, PIDTYPE_PGID);
1654 } else /* upid > 0 */ {
1655 type = PIDTYPE_PID;
1656 pid = find_get_pid(upid);
1657 }
1658
1659 wo.wo_type = type;
1660 wo.wo_pid = pid;
1661 wo.wo_flags = options | WEXITED;
1662 wo.wo_info = NULL;
1663 wo.wo_stat = 0;
1664 wo.wo_rusage = ru;
1665 ret = do_wait(&wo);
1666 put_pid(pid);
1667 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1668 ret = -EFAULT;
1669
1670 return ret;
1671}
1672
1673SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1674 int, options, struct rusage __user *, ru)
1675{
1676 struct rusage r;
1677 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1678
1679 if (err > 0) {
1680 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1681 return -EFAULT;
1682 }
1683 return err;
1684}
1685
1686#ifdef __ARCH_WANT_SYS_WAITPID
1687
1688/*
1689 * sys_waitpid() remains for compatibility. waitpid() should be
1690 * implemented by calling sys_wait4() from libc.a.
1691 */
1692SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1693{
1694 return kernel_wait4(pid, stat_addr, options, NULL);
1695}
1696
1697#endif
1698
1699#ifdef CONFIG_COMPAT
1700COMPAT_SYSCALL_DEFINE4(wait4,
1701 compat_pid_t, pid,
1702 compat_uint_t __user *, stat_addr,
1703 int, options,
1704 struct compat_rusage __user *, ru)
1705{
1706 struct rusage r;
1707 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1708 if (err > 0) {
1709 if (ru && put_compat_rusage(&r, ru))
1710 return -EFAULT;
1711 }
1712 return err;
1713}
1714
1715COMPAT_SYSCALL_DEFINE5(waitid,
1716 int, which, compat_pid_t, pid,
1717 struct compat_siginfo __user *, infop, int, options,
1718 struct compat_rusage __user *, uru)
1719{
1720 struct rusage ru;
1721 struct waitid_info info = {.status = 0};
1722 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1723 int signo = 0;
1724 if (err > 0) {
1725 signo = SIGCHLD;
1726 err = 0;
1727 if (uru) {
1728 /* kernel_waitid() overwrites everything in ru */
1729 if (COMPAT_USE_64BIT_TIME)
1730 err = copy_to_user(uru, &ru, sizeof(ru));
1731 else
1732 err = put_compat_rusage(&ru, uru);
1733 if (err)
1734 return -EFAULT;
1735 }
1736 }
1737
1738 if (!infop)
1739 return err;
1740
1741 if (!access_ok(VERIFY_WRITE, infop, sizeof(*infop)))
1742 return -EFAULT;
1743
1744 user_access_begin();
1745 unsafe_put_user(signo, &infop->si_signo, Efault);
1746 unsafe_put_user(0, &infop->si_errno, Efault);
1747 unsafe_put_user(info.cause, &infop->si_code, Efault);
1748 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1749 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1750 unsafe_put_user(info.status, &infop->si_status, Efault);
1751 user_access_end();
1752 return err;
1753Efault:
1754 user_access_end();
1755 return -EFAULT;
1756}
1757#endif
1758
1759__weak void abort(void)
1760{
1761 BUG();
1762
1763 /* if that doesn't kill us, halt */
1764 panic("Oops failed to kill thread");
1765}
1766EXPORT_SYMBOL(abort);
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/uaccess.h>
72
73#include <uapi/linux/wait.h>
74
75#include <asm/unistd.h>
76#include <asm/mmu_context.h>
77
78#include "exit.h"
79
80/*
81 * The default value should be high enough to not crash a system that randomly
82 * crashes its kernel from time to time, but low enough to at least not permit
83 * overflowing 32-bit refcounts or the ldsem writer count.
84 */
85static unsigned int oops_limit = 10000;
86
87#ifdef CONFIG_SYSCTL
88static struct ctl_table kern_exit_table[] = {
89 {
90 .procname = "oops_limit",
91 .data = &oops_limit,
92 .maxlen = sizeof(oops_limit),
93 .mode = 0644,
94 .proc_handler = proc_douintvec,
95 },
96};
97
98static __init int kernel_exit_sysctls_init(void)
99{
100 register_sysctl_init("kernel", kern_exit_table);
101 return 0;
102}
103late_initcall(kernel_exit_sysctls_init);
104#endif
105
106static atomic_t oops_count = ATOMIC_INIT(0);
107
108#ifdef CONFIG_SYSFS
109static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
110 char *page)
111{
112 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
113}
114
115static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
116
117static __init int kernel_exit_sysfs_init(void)
118{
119 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
120 return 0;
121}
122late_initcall(kernel_exit_sysfs_init);
123#endif
124
125static void __unhash_process(struct task_struct *p, bool group_dead)
126{
127 nr_threads--;
128 detach_pid(p, PIDTYPE_PID);
129 if (group_dead) {
130 detach_pid(p, PIDTYPE_TGID);
131 detach_pid(p, PIDTYPE_PGID);
132 detach_pid(p, PIDTYPE_SID);
133
134 list_del_rcu(&p->tasks);
135 list_del_init(&p->sibling);
136 __this_cpu_dec(process_counts);
137 }
138 list_del_rcu(&p->thread_node);
139}
140
141/*
142 * This function expects the tasklist_lock write-locked.
143 */
144static void __exit_signal(struct task_struct *tsk)
145{
146 struct signal_struct *sig = tsk->signal;
147 bool group_dead = thread_group_leader(tsk);
148 struct sighand_struct *sighand;
149 struct tty_struct *tty;
150 u64 utime, stime;
151
152 sighand = rcu_dereference_check(tsk->sighand,
153 lockdep_tasklist_lock_is_held());
154 spin_lock(&sighand->siglock);
155
156#ifdef CONFIG_POSIX_TIMERS
157 posix_cpu_timers_exit(tsk);
158 if (group_dead)
159 posix_cpu_timers_exit_group(tsk);
160#endif
161
162 if (group_dead) {
163 tty = sig->tty;
164 sig->tty = NULL;
165 } else {
166 /*
167 * If there is any task waiting for the group exit
168 * then notify it:
169 */
170 if (sig->notify_count > 0 && !--sig->notify_count)
171 wake_up_process(sig->group_exec_task);
172
173 if (tsk == sig->curr_target)
174 sig->curr_target = next_thread(tsk);
175 }
176
177 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
178 sizeof(unsigned long long));
179
180 /*
181 * Accumulate here the counters for all threads as they die. We could
182 * skip the group leader because it is the last user of signal_struct,
183 * but we want to avoid the race with thread_group_cputime() which can
184 * see the empty ->thread_head list.
185 */
186 task_cputime(tsk, &utime, &stime);
187 write_seqlock(&sig->stats_lock);
188 sig->utime += utime;
189 sig->stime += stime;
190 sig->gtime += task_gtime(tsk);
191 sig->min_flt += tsk->min_flt;
192 sig->maj_flt += tsk->maj_flt;
193 sig->nvcsw += tsk->nvcsw;
194 sig->nivcsw += tsk->nivcsw;
195 sig->inblock += task_io_get_inblock(tsk);
196 sig->oublock += task_io_get_oublock(tsk);
197 task_io_accounting_add(&sig->ioac, &tsk->ioac);
198 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
199 sig->nr_threads--;
200 __unhash_process(tsk, group_dead);
201 write_sequnlock(&sig->stats_lock);
202
203 /*
204 * Do this under ->siglock, we can race with another thread
205 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
206 */
207 flush_sigqueue(&tsk->pending);
208 tsk->sighand = NULL;
209 spin_unlock(&sighand->siglock);
210
211 __cleanup_sighand(sighand);
212 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
213 if (group_dead) {
214 flush_sigqueue(&sig->shared_pending);
215 tty_kref_put(tty);
216 }
217}
218
219static void delayed_put_task_struct(struct rcu_head *rhp)
220{
221 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
222
223 kprobe_flush_task(tsk);
224 rethook_flush_task(tsk);
225 perf_event_delayed_put(tsk);
226 trace_sched_process_free(tsk);
227 put_task_struct(tsk);
228}
229
230void put_task_struct_rcu_user(struct task_struct *task)
231{
232 if (refcount_dec_and_test(&task->rcu_users))
233 call_rcu(&task->rcu, delayed_put_task_struct);
234}
235
236void __weak release_thread(struct task_struct *dead_task)
237{
238}
239
240void release_task(struct task_struct *p)
241{
242 struct task_struct *leader;
243 struct pid *thread_pid;
244 int zap_leader;
245repeat:
246 /* don't need to get the RCU readlock here - the process is dead and
247 * can't be modifying its own credentials. But shut RCU-lockdep up */
248 rcu_read_lock();
249 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
250 rcu_read_unlock();
251
252 cgroup_release(p);
253
254 write_lock_irq(&tasklist_lock);
255 ptrace_release_task(p);
256 thread_pid = get_pid(p->thread_pid);
257 __exit_signal(p);
258
259 /*
260 * If we are the last non-leader member of the thread
261 * group, and the leader is zombie, then notify the
262 * group leader's parent process. (if it wants notification.)
263 */
264 zap_leader = 0;
265 leader = p->group_leader;
266 if (leader != p && thread_group_empty(leader)
267 && leader->exit_state == EXIT_ZOMBIE) {
268 /*
269 * If we were the last child thread and the leader has
270 * exited already, and the leader's parent ignores SIGCHLD,
271 * then we are the one who should release the leader.
272 */
273 zap_leader = do_notify_parent(leader, leader->exit_signal);
274 if (zap_leader)
275 leader->exit_state = EXIT_DEAD;
276 }
277
278 write_unlock_irq(&tasklist_lock);
279 proc_flush_pid(thread_pid);
280 put_pid(thread_pid);
281 release_thread(p);
282 put_task_struct_rcu_user(p);
283
284 p = leader;
285 if (unlikely(zap_leader))
286 goto repeat;
287}
288
289int rcuwait_wake_up(struct rcuwait *w)
290{
291 int ret = 0;
292 struct task_struct *task;
293
294 rcu_read_lock();
295
296 /*
297 * Order condition vs @task, such that everything prior to the load
298 * of @task is visible. This is the condition as to why the user called
299 * rcuwait_wake() in the first place. Pairs with set_current_state()
300 * barrier (A) in rcuwait_wait_event().
301 *
302 * WAIT WAKE
303 * [S] tsk = current [S] cond = true
304 * MB (A) MB (B)
305 * [L] cond [L] tsk
306 */
307 smp_mb(); /* (B) */
308
309 task = rcu_dereference(w->task);
310 if (task)
311 ret = wake_up_process(task);
312 rcu_read_unlock();
313
314 return ret;
315}
316EXPORT_SYMBOL_GPL(rcuwait_wake_up);
317
318/*
319 * Determine if a process group is "orphaned", according to the POSIX
320 * definition in 2.2.2.52. Orphaned process groups are not to be affected
321 * by terminal-generated stop signals. Newly orphaned process groups are
322 * to receive a SIGHUP and a SIGCONT.
323 *
324 * "I ask you, have you ever known what it is to be an orphan?"
325 */
326static int will_become_orphaned_pgrp(struct pid *pgrp,
327 struct task_struct *ignored_task)
328{
329 struct task_struct *p;
330
331 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
332 if ((p == ignored_task) ||
333 (p->exit_state && thread_group_empty(p)) ||
334 is_global_init(p->real_parent))
335 continue;
336
337 if (task_pgrp(p->real_parent) != pgrp &&
338 task_session(p->real_parent) == task_session(p))
339 return 0;
340 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
341
342 return 1;
343}
344
345int is_current_pgrp_orphaned(void)
346{
347 int retval;
348
349 read_lock(&tasklist_lock);
350 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
351 read_unlock(&tasklist_lock);
352
353 return retval;
354}
355
356static bool has_stopped_jobs(struct pid *pgrp)
357{
358 struct task_struct *p;
359
360 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
361 if (p->signal->flags & SIGNAL_STOP_STOPPED)
362 return true;
363 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
364
365 return false;
366}
367
368/*
369 * Check to see if any process groups have become orphaned as
370 * a result of our exiting, and if they have any stopped jobs,
371 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
372 */
373static void
374kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
375{
376 struct pid *pgrp = task_pgrp(tsk);
377 struct task_struct *ignored_task = tsk;
378
379 if (!parent)
380 /* exit: our father is in a different pgrp than
381 * we are and we were the only connection outside.
382 */
383 parent = tsk->real_parent;
384 else
385 /* reparent: our child is in a different pgrp than
386 * we are, and it was the only connection outside.
387 */
388 ignored_task = NULL;
389
390 if (task_pgrp(parent) != pgrp &&
391 task_session(parent) == task_session(tsk) &&
392 will_become_orphaned_pgrp(pgrp, ignored_task) &&
393 has_stopped_jobs(pgrp)) {
394 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
395 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
396 }
397}
398
399static void coredump_task_exit(struct task_struct *tsk)
400{
401 struct core_state *core_state;
402
403 /*
404 * Serialize with any possible pending coredump.
405 * We must hold siglock around checking core_state
406 * and setting PF_POSTCOREDUMP. The core-inducing thread
407 * will increment ->nr_threads for each thread in the
408 * group without PF_POSTCOREDUMP set.
409 */
410 spin_lock_irq(&tsk->sighand->siglock);
411 tsk->flags |= PF_POSTCOREDUMP;
412 core_state = tsk->signal->core_state;
413 spin_unlock_irq(&tsk->sighand->siglock);
414 if (core_state) {
415 struct core_thread self;
416
417 self.task = current;
418 if (self.task->flags & PF_SIGNALED)
419 self.next = xchg(&core_state->dumper.next, &self);
420 else
421 self.task = NULL;
422 /*
423 * Implies mb(), the result of xchg() must be visible
424 * to core_state->dumper.
425 */
426 if (atomic_dec_and_test(&core_state->nr_threads))
427 complete(&core_state->startup);
428
429 for (;;) {
430 set_current_state(TASK_IDLE|TASK_FREEZABLE);
431 if (!self.task) /* see coredump_finish() */
432 break;
433 schedule();
434 }
435 __set_current_state(TASK_RUNNING);
436 }
437}
438
439#ifdef CONFIG_MEMCG
440/* drops tasklist_lock if succeeds */
441static bool __try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm)
442{
443 bool ret = false;
444
445 task_lock(tsk);
446 if (likely(tsk->mm == mm)) {
447 /* tsk can't pass exit_mm/exec_mmap and exit */
448 read_unlock(&tasklist_lock);
449 WRITE_ONCE(mm->owner, tsk);
450 lru_gen_migrate_mm(mm);
451 ret = true;
452 }
453 task_unlock(tsk);
454 return ret;
455}
456
457static bool try_to_set_owner(struct task_struct *g, struct mm_struct *mm)
458{
459 struct task_struct *t;
460
461 for_each_thread(g, t) {
462 struct mm_struct *t_mm = READ_ONCE(t->mm);
463 if (t_mm == mm) {
464 if (__try_to_set_owner(t, mm))
465 return true;
466 } else if (t_mm)
467 break;
468 }
469
470 return false;
471}
472
473/*
474 * A task is exiting. If it owned this mm, find a new owner for the mm.
475 */
476void mm_update_next_owner(struct mm_struct *mm)
477{
478 struct task_struct *g, *p = current;
479
480 /*
481 * If the exiting or execing task is not the owner, it's
482 * someone else's problem.
483 */
484 if (mm->owner != p)
485 return;
486 /*
487 * The current owner is exiting/execing and there are no other
488 * candidates. Do not leave the mm pointing to a possibly
489 * freed task structure.
490 */
491 if (atomic_read(&mm->mm_users) <= 1) {
492 WRITE_ONCE(mm->owner, NULL);
493 return;
494 }
495
496 read_lock(&tasklist_lock);
497 /*
498 * Search in the children
499 */
500 list_for_each_entry(g, &p->children, sibling) {
501 if (try_to_set_owner(g, mm))
502 goto ret;
503 }
504 /*
505 * Search in the siblings
506 */
507 list_for_each_entry(g, &p->real_parent->children, sibling) {
508 if (try_to_set_owner(g, mm))
509 goto ret;
510 }
511 /*
512 * Search through everything else, we should not get here often.
513 */
514 for_each_process(g) {
515 if (atomic_read(&mm->mm_users) <= 1)
516 break;
517 if (g->flags & PF_KTHREAD)
518 continue;
519 if (try_to_set_owner(g, mm))
520 goto ret;
521 }
522 read_unlock(&tasklist_lock);
523 /*
524 * We found no owner yet mm_users > 1: this implies that we are
525 * most likely racing with swapoff (try_to_unuse()) or /proc or
526 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
527 */
528 WRITE_ONCE(mm->owner, NULL);
529 ret:
530 return;
531
532}
533#endif /* CONFIG_MEMCG */
534
535/*
536 * Turn us into a lazy TLB process if we
537 * aren't already..
538 */
539static void exit_mm(void)
540{
541 struct mm_struct *mm = current->mm;
542
543 exit_mm_release(current, mm);
544 if (!mm)
545 return;
546 mmap_read_lock(mm);
547 mmgrab_lazy_tlb(mm);
548 BUG_ON(mm != current->active_mm);
549 /* more a memory barrier than a real lock */
550 task_lock(current);
551 /*
552 * When a thread stops operating on an address space, the loop
553 * in membarrier_private_expedited() may not observe that
554 * tsk->mm, and the loop in membarrier_global_expedited() may
555 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
556 * rq->membarrier_state, so those would not issue an IPI.
557 * Membarrier requires a memory barrier after accessing
558 * user-space memory, before clearing tsk->mm or the
559 * rq->membarrier_state.
560 */
561 smp_mb__after_spinlock();
562 local_irq_disable();
563 current->mm = NULL;
564 membarrier_update_current_mm(NULL);
565 enter_lazy_tlb(mm, current);
566 local_irq_enable();
567 task_unlock(current);
568 mmap_read_unlock(mm);
569 mm_update_next_owner(mm);
570 mmput(mm);
571 if (test_thread_flag(TIF_MEMDIE))
572 exit_oom_victim();
573}
574
575static struct task_struct *find_alive_thread(struct task_struct *p)
576{
577 struct task_struct *t;
578
579 for_each_thread(p, t) {
580 if (!(t->flags & PF_EXITING))
581 return t;
582 }
583 return NULL;
584}
585
586static struct task_struct *find_child_reaper(struct task_struct *father,
587 struct list_head *dead)
588 __releases(&tasklist_lock)
589 __acquires(&tasklist_lock)
590{
591 struct pid_namespace *pid_ns = task_active_pid_ns(father);
592 struct task_struct *reaper = pid_ns->child_reaper;
593 struct task_struct *p, *n;
594
595 if (likely(reaper != father))
596 return reaper;
597
598 reaper = find_alive_thread(father);
599 if (reaper) {
600 pid_ns->child_reaper = reaper;
601 return reaper;
602 }
603
604 write_unlock_irq(&tasklist_lock);
605
606 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
607 list_del_init(&p->ptrace_entry);
608 release_task(p);
609 }
610
611 zap_pid_ns_processes(pid_ns);
612 write_lock_irq(&tasklist_lock);
613
614 return father;
615}
616
617/*
618 * When we die, we re-parent all our children, and try to:
619 * 1. give them to another thread in our thread group, if such a member exists
620 * 2. give it to the first ancestor process which prctl'd itself as a
621 * child_subreaper for its children (like a service manager)
622 * 3. give it to the init process (PID 1) in our pid namespace
623 */
624static struct task_struct *find_new_reaper(struct task_struct *father,
625 struct task_struct *child_reaper)
626{
627 struct task_struct *thread, *reaper;
628
629 thread = find_alive_thread(father);
630 if (thread)
631 return thread;
632
633 if (father->signal->has_child_subreaper) {
634 unsigned int ns_level = task_pid(father)->level;
635 /*
636 * Find the first ->is_child_subreaper ancestor in our pid_ns.
637 * We can't check reaper != child_reaper to ensure we do not
638 * cross the namespaces, the exiting parent could be injected
639 * by setns() + fork().
640 * We check pid->level, this is slightly more efficient than
641 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
642 */
643 for (reaper = father->real_parent;
644 task_pid(reaper)->level == ns_level;
645 reaper = reaper->real_parent) {
646 if (reaper == &init_task)
647 break;
648 if (!reaper->signal->is_child_subreaper)
649 continue;
650 thread = find_alive_thread(reaper);
651 if (thread)
652 return thread;
653 }
654 }
655
656 return child_reaper;
657}
658
659/*
660* Any that need to be release_task'd are put on the @dead list.
661 */
662static void reparent_leader(struct task_struct *father, struct task_struct *p,
663 struct list_head *dead)
664{
665 if (unlikely(p->exit_state == EXIT_DEAD))
666 return;
667
668 /* We don't want people slaying init. */
669 p->exit_signal = SIGCHLD;
670
671 /* If it has exited notify the new parent about this child's death. */
672 if (!p->ptrace &&
673 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
674 if (do_notify_parent(p, p->exit_signal)) {
675 p->exit_state = EXIT_DEAD;
676 list_add(&p->ptrace_entry, dead);
677 }
678 }
679
680 kill_orphaned_pgrp(p, father);
681}
682
683/*
684 * This does two things:
685 *
686 * A. Make init inherit all the child processes
687 * B. Check to see if any process groups have become orphaned
688 * as a result of our exiting, and if they have any stopped
689 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
690 */
691static void forget_original_parent(struct task_struct *father,
692 struct list_head *dead)
693{
694 struct task_struct *p, *t, *reaper;
695
696 if (unlikely(!list_empty(&father->ptraced)))
697 exit_ptrace(father, dead);
698
699 /* Can drop and reacquire tasklist_lock */
700 reaper = find_child_reaper(father, dead);
701 if (list_empty(&father->children))
702 return;
703
704 reaper = find_new_reaper(father, reaper);
705 list_for_each_entry(p, &father->children, sibling) {
706 for_each_thread(p, t) {
707 RCU_INIT_POINTER(t->real_parent, reaper);
708 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
709 if (likely(!t->ptrace))
710 t->parent = t->real_parent;
711 if (t->pdeath_signal)
712 group_send_sig_info(t->pdeath_signal,
713 SEND_SIG_NOINFO, t,
714 PIDTYPE_TGID);
715 }
716 /*
717 * If this is a threaded reparent there is no need to
718 * notify anyone anything has happened.
719 */
720 if (!same_thread_group(reaper, father))
721 reparent_leader(father, p, dead);
722 }
723 list_splice_tail_init(&father->children, &reaper->children);
724}
725
726/*
727 * Send signals to all our closest relatives so that they know
728 * to properly mourn us..
729 */
730static void exit_notify(struct task_struct *tsk, int group_dead)
731{
732 bool autoreap;
733 struct task_struct *p, *n;
734 LIST_HEAD(dead);
735
736 write_lock_irq(&tasklist_lock);
737 forget_original_parent(tsk, &dead);
738
739 if (group_dead)
740 kill_orphaned_pgrp(tsk->group_leader, NULL);
741
742 tsk->exit_state = EXIT_ZOMBIE;
743 /*
744 * sub-thread or delay_group_leader(), wake up the
745 * PIDFD_THREAD waiters.
746 */
747 if (!thread_group_empty(tsk))
748 do_notify_pidfd(tsk);
749
750 if (unlikely(tsk->ptrace)) {
751 int sig = thread_group_leader(tsk) &&
752 thread_group_empty(tsk) &&
753 !ptrace_reparented(tsk) ?
754 tsk->exit_signal : SIGCHLD;
755 autoreap = do_notify_parent(tsk, sig);
756 } else if (thread_group_leader(tsk)) {
757 autoreap = thread_group_empty(tsk) &&
758 do_notify_parent(tsk, tsk->exit_signal);
759 } else {
760 autoreap = true;
761 }
762
763 if (autoreap) {
764 tsk->exit_state = EXIT_DEAD;
765 list_add(&tsk->ptrace_entry, &dead);
766 }
767
768 /* mt-exec, de_thread() is waiting for group leader */
769 if (unlikely(tsk->signal->notify_count < 0))
770 wake_up_process(tsk->signal->group_exec_task);
771 write_unlock_irq(&tasklist_lock);
772
773 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
774 list_del_init(&p->ptrace_entry);
775 release_task(p);
776 }
777}
778
779#ifdef CONFIG_DEBUG_STACK_USAGE
780unsigned long stack_not_used(struct task_struct *p)
781{
782 unsigned long *n = end_of_stack(p);
783
784 do { /* Skip over canary */
785# ifdef CONFIG_STACK_GROWSUP
786 n--;
787# else
788 n++;
789# endif
790 } while (!*n);
791
792# ifdef CONFIG_STACK_GROWSUP
793 return (unsigned long)end_of_stack(p) - (unsigned long)n;
794# else
795 return (unsigned long)n - (unsigned long)end_of_stack(p);
796# endif
797}
798
799/* Count the maximum pages reached in kernel stacks */
800static inline void kstack_histogram(unsigned long used_stack)
801{
802#ifdef CONFIG_VM_EVENT_COUNTERS
803 if (used_stack <= 1024)
804 count_vm_event(KSTACK_1K);
805#if THREAD_SIZE > 1024
806 else if (used_stack <= 2048)
807 count_vm_event(KSTACK_2K);
808#endif
809#if THREAD_SIZE > 2048
810 else if (used_stack <= 4096)
811 count_vm_event(KSTACK_4K);
812#endif
813#if THREAD_SIZE > 4096
814 else if (used_stack <= 8192)
815 count_vm_event(KSTACK_8K);
816#endif
817#if THREAD_SIZE > 8192
818 else if (used_stack <= 16384)
819 count_vm_event(KSTACK_16K);
820#endif
821#if THREAD_SIZE > 16384
822 else if (used_stack <= 32768)
823 count_vm_event(KSTACK_32K);
824#endif
825#if THREAD_SIZE > 32768
826 else if (used_stack <= 65536)
827 count_vm_event(KSTACK_64K);
828#endif
829#if THREAD_SIZE > 65536
830 else
831 count_vm_event(KSTACK_REST);
832#endif
833#endif /* CONFIG_VM_EVENT_COUNTERS */
834}
835
836static void check_stack_usage(void)
837{
838 static DEFINE_SPINLOCK(low_water_lock);
839 static int lowest_to_date = THREAD_SIZE;
840 unsigned long free;
841
842 free = stack_not_used(current);
843 kstack_histogram(THREAD_SIZE - free);
844
845 if (free >= lowest_to_date)
846 return;
847
848 spin_lock(&low_water_lock);
849 if (free < lowest_to_date) {
850 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
851 current->comm, task_pid_nr(current), free);
852 lowest_to_date = free;
853 }
854 spin_unlock(&low_water_lock);
855}
856#else
857static inline void check_stack_usage(void) {}
858#endif
859
860static void synchronize_group_exit(struct task_struct *tsk, long code)
861{
862 struct sighand_struct *sighand = tsk->sighand;
863 struct signal_struct *signal = tsk->signal;
864
865 spin_lock_irq(&sighand->siglock);
866 signal->quick_threads--;
867 if ((signal->quick_threads == 0) &&
868 !(signal->flags & SIGNAL_GROUP_EXIT)) {
869 signal->flags = SIGNAL_GROUP_EXIT;
870 signal->group_exit_code = code;
871 signal->group_stop_count = 0;
872 }
873 spin_unlock_irq(&sighand->siglock);
874}
875
876void __noreturn do_exit(long code)
877{
878 struct task_struct *tsk = current;
879 int group_dead;
880
881 WARN_ON(irqs_disabled());
882
883 synchronize_group_exit(tsk, code);
884
885 WARN_ON(tsk->plug);
886
887 kcov_task_exit(tsk);
888 kmsan_task_exit(tsk);
889
890 coredump_task_exit(tsk);
891 ptrace_event(PTRACE_EVENT_EXIT, code);
892 user_events_exit(tsk);
893
894 io_uring_files_cancel();
895 exit_signals(tsk); /* sets PF_EXITING */
896
897 seccomp_filter_release(tsk);
898
899 acct_update_integrals(tsk);
900 group_dead = atomic_dec_and_test(&tsk->signal->live);
901 if (group_dead) {
902 /*
903 * If the last thread of global init has exited, panic
904 * immediately to get a useable coredump.
905 */
906 if (unlikely(is_global_init(tsk)))
907 panic("Attempted to kill init! exitcode=0x%08x\n",
908 tsk->signal->group_exit_code ?: (int)code);
909
910#ifdef CONFIG_POSIX_TIMERS
911 hrtimer_cancel(&tsk->signal->real_timer);
912 exit_itimers(tsk);
913#endif
914 if (tsk->mm)
915 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
916 }
917 acct_collect(code, group_dead);
918 if (group_dead)
919 tty_audit_exit();
920 audit_free(tsk);
921
922 tsk->exit_code = code;
923 taskstats_exit(tsk, group_dead);
924
925 exit_mm();
926
927 if (group_dead)
928 acct_process();
929 trace_sched_process_exit(tsk);
930
931 exit_sem(tsk);
932 exit_shm(tsk);
933 exit_files(tsk);
934 exit_fs(tsk);
935 if (group_dead)
936 disassociate_ctty(1);
937 exit_task_namespaces(tsk);
938 exit_task_work(tsk);
939 exit_thread(tsk);
940
941 /*
942 * Flush inherited counters to the parent - before the parent
943 * gets woken up by child-exit notifications.
944 *
945 * because of cgroup mode, must be called before cgroup_exit()
946 */
947 perf_event_exit_task(tsk);
948
949 sched_autogroup_exit_task(tsk);
950 cgroup_exit(tsk);
951
952 /*
953 * FIXME: do that only when needed, using sched_exit tracepoint
954 */
955 flush_ptrace_hw_breakpoint(tsk);
956
957 exit_tasks_rcu_start();
958 exit_notify(tsk, group_dead);
959 proc_exit_connector(tsk);
960 mpol_put_task_policy(tsk);
961#ifdef CONFIG_FUTEX
962 if (unlikely(current->pi_state_cache))
963 kfree(current->pi_state_cache);
964#endif
965 /*
966 * Make sure we are holding no locks:
967 */
968 debug_check_no_locks_held();
969
970 if (tsk->io_context)
971 exit_io_context(tsk);
972
973 if (tsk->splice_pipe)
974 free_pipe_info(tsk->splice_pipe);
975
976 if (tsk->task_frag.page)
977 put_page(tsk->task_frag.page);
978
979 exit_task_stack_account(tsk);
980
981 check_stack_usage();
982 preempt_disable();
983 if (tsk->nr_dirtied)
984 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
985 exit_rcu();
986 exit_tasks_rcu_finish();
987
988 lockdep_free_task(tsk);
989 do_task_dead();
990}
991
992void __noreturn make_task_dead(int signr)
993{
994 /*
995 * Take the task off the cpu after something catastrophic has
996 * happened.
997 *
998 * We can get here from a kernel oops, sometimes with preemption off.
999 * Start by checking for critical errors.
1000 * Then fix up important state like USER_DS and preemption.
1001 * Then do everything else.
1002 */
1003 struct task_struct *tsk = current;
1004 unsigned int limit;
1005
1006 if (unlikely(in_interrupt()))
1007 panic("Aiee, killing interrupt handler!");
1008 if (unlikely(!tsk->pid))
1009 panic("Attempted to kill the idle task!");
1010
1011 if (unlikely(irqs_disabled())) {
1012 pr_info("note: %s[%d] exited with irqs disabled\n",
1013 current->comm, task_pid_nr(current));
1014 local_irq_enable();
1015 }
1016 if (unlikely(in_atomic())) {
1017 pr_info("note: %s[%d] exited with preempt_count %d\n",
1018 current->comm, task_pid_nr(current),
1019 preempt_count());
1020 preempt_count_set(PREEMPT_ENABLED);
1021 }
1022
1023 /*
1024 * Every time the system oopses, if the oops happens while a reference
1025 * to an object was held, the reference leaks.
1026 * If the oops doesn't also leak memory, repeated oopsing can cause
1027 * reference counters to wrap around (if they're not using refcount_t).
1028 * This means that repeated oopsing can make unexploitable-looking bugs
1029 * exploitable through repeated oopsing.
1030 * To make sure this can't happen, place an upper bound on how often the
1031 * kernel may oops without panic().
1032 */
1033 limit = READ_ONCE(oops_limit);
1034 if (atomic_inc_return(&oops_count) >= limit && limit)
1035 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
1036
1037 /*
1038 * We're taking recursive faults here in make_task_dead. Safest is to just
1039 * leave this task alone and wait for reboot.
1040 */
1041 if (unlikely(tsk->flags & PF_EXITING)) {
1042 pr_alert("Fixing recursive fault but reboot is needed!\n");
1043 futex_exit_recursive(tsk);
1044 tsk->exit_state = EXIT_DEAD;
1045 refcount_inc(&tsk->rcu_users);
1046 do_task_dead();
1047 }
1048
1049 do_exit(signr);
1050}
1051
1052SYSCALL_DEFINE1(exit, int, error_code)
1053{
1054 do_exit((error_code&0xff)<<8);
1055}
1056
1057/*
1058 * Take down every thread in the group. This is called by fatal signals
1059 * as well as by sys_exit_group (below).
1060 */
1061void __noreturn
1062do_group_exit(int exit_code)
1063{
1064 struct signal_struct *sig = current->signal;
1065
1066 if (sig->flags & SIGNAL_GROUP_EXIT)
1067 exit_code = sig->group_exit_code;
1068 else if (sig->group_exec_task)
1069 exit_code = 0;
1070 else {
1071 struct sighand_struct *const sighand = current->sighand;
1072
1073 spin_lock_irq(&sighand->siglock);
1074 if (sig->flags & SIGNAL_GROUP_EXIT)
1075 /* Another thread got here before we took the lock. */
1076 exit_code = sig->group_exit_code;
1077 else if (sig->group_exec_task)
1078 exit_code = 0;
1079 else {
1080 sig->group_exit_code = exit_code;
1081 sig->flags = SIGNAL_GROUP_EXIT;
1082 zap_other_threads(current);
1083 }
1084 spin_unlock_irq(&sighand->siglock);
1085 }
1086
1087 do_exit(exit_code);
1088 /* NOTREACHED */
1089}
1090
1091/*
1092 * this kills every thread in the thread group. Note that any externally
1093 * wait4()-ing process will get the correct exit code - even if this
1094 * thread is not the thread group leader.
1095 */
1096SYSCALL_DEFINE1(exit_group, int, error_code)
1097{
1098 do_group_exit((error_code & 0xff) << 8);
1099 /* NOTREACHED */
1100 return 0;
1101}
1102
1103static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1104{
1105 return wo->wo_type == PIDTYPE_MAX ||
1106 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1107}
1108
1109static int
1110eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1111{
1112 if (!eligible_pid(wo, p))
1113 return 0;
1114
1115 /*
1116 * Wait for all children (clone and not) if __WALL is set or
1117 * if it is traced by us.
1118 */
1119 if (ptrace || (wo->wo_flags & __WALL))
1120 return 1;
1121
1122 /*
1123 * Otherwise, wait for clone children *only* if __WCLONE is set;
1124 * otherwise, wait for non-clone children *only*.
1125 *
1126 * Note: a "clone" child here is one that reports to its parent
1127 * using a signal other than SIGCHLD, or a non-leader thread which
1128 * we can only see if it is traced by us.
1129 */
1130 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1131 return 0;
1132
1133 return 1;
1134}
1135
1136/*
1137 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1138 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1139 * the lock and this task is uninteresting. If we return nonzero, we have
1140 * released the lock and the system call should return.
1141 */
1142static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1143{
1144 int state, status;
1145 pid_t pid = task_pid_vnr(p);
1146 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1147 struct waitid_info *infop;
1148
1149 if (!likely(wo->wo_flags & WEXITED))
1150 return 0;
1151
1152 if (unlikely(wo->wo_flags & WNOWAIT)) {
1153 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1154 ? p->signal->group_exit_code : p->exit_code;
1155 get_task_struct(p);
1156 read_unlock(&tasklist_lock);
1157 sched_annotate_sleep();
1158 if (wo->wo_rusage)
1159 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1160 put_task_struct(p);
1161 goto out_info;
1162 }
1163 /*
1164 * Move the task's state to DEAD/TRACE, only one thread can do this.
1165 */
1166 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1167 EXIT_TRACE : EXIT_DEAD;
1168 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1169 return 0;
1170 /*
1171 * We own this thread, nobody else can reap it.
1172 */
1173 read_unlock(&tasklist_lock);
1174 sched_annotate_sleep();
1175
1176 /*
1177 * Check thread_group_leader() to exclude the traced sub-threads.
1178 */
1179 if (state == EXIT_DEAD && thread_group_leader(p)) {
1180 struct signal_struct *sig = p->signal;
1181 struct signal_struct *psig = current->signal;
1182 unsigned long maxrss;
1183 u64 tgutime, tgstime;
1184
1185 /*
1186 * The resource counters for the group leader are in its
1187 * own task_struct. Those for dead threads in the group
1188 * are in its signal_struct, as are those for the child
1189 * processes it has previously reaped. All these
1190 * accumulate in the parent's signal_struct c* fields.
1191 *
1192 * We don't bother to take a lock here to protect these
1193 * p->signal fields because the whole thread group is dead
1194 * and nobody can change them.
1195 *
1196 * psig->stats_lock also protects us from our sub-threads
1197 * which can reap other children at the same time.
1198 *
1199 * We use thread_group_cputime_adjusted() to get times for
1200 * the thread group, which consolidates times for all threads
1201 * in the group including the group leader.
1202 */
1203 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1204 write_seqlock_irq(&psig->stats_lock);
1205 psig->cutime += tgutime + sig->cutime;
1206 psig->cstime += tgstime + sig->cstime;
1207 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1208 psig->cmin_flt +=
1209 p->min_flt + sig->min_flt + sig->cmin_flt;
1210 psig->cmaj_flt +=
1211 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1212 psig->cnvcsw +=
1213 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1214 psig->cnivcsw +=
1215 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1216 psig->cinblock +=
1217 task_io_get_inblock(p) +
1218 sig->inblock + sig->cinblock;
1219 psig->coublock +=
1220 task_io_get_oublock(p) +
1221 sig->oublock + sig->coublock;
1222 maxrss = max(sig->maxrss, sig->cmaxrss);
1223 if (psig->cmaxrss < maxrss)
1224 psig->cmaxrss = maxrss;
1225 task_io_accounting_add(&psig->ioac, &p->ioac);
1226 task_io_accounting_add(&psig->ioac, &sig->ioac);
1227 write_sequnlock_irq(&psig->stats_lock);
1228 }
1229
1230 if (wo->wo_rusage)
1231 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1232 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1233 ? p->signal->group_exit_code : p->exit_code;
1234 wo->wo_stat = status;
1235
1236 if (state == EXIT_TRACE) {
1237 write_lock_irq(&tasklist_lock);
1238 /* We dropped tasklist, ptracer could die and untrace */
1239 ptrace_unlink(p);
1240
1241 /* If parent wants a zombie, don't release it now */
1242 state = EXIT_ZOMBIE;
1243 if (do_notify_parent(p, p->exit_signal))
1244 state = EXIT_DEAD;
1245 p->exit_state = state;
1246 write_unlock_irq(&tasklist_lock);
1247 }
1248 if (state == EXIT_DEAD)
1249 release_task(p);
1250
1251out_info:
1252 infop = wo->wo_info;
1253 if (infop) {
1254 if ((status & 0x7f) == 0) {
1255 infop->cause = CLD_EXITED;
1256 infop->status = status >> 8;
1257 } else {
1258 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1259 infop->status = status & 0x7f;
1260 }
1261 infop->pid = pid;
1262 infop->uid = uid;
1263 }
1264
1265 return pid;
1266}
1267
1268static int *task_stopped_code(struct task_struct *p, bool ptrace)
1269{
1270 if (ptrace) {
1271 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1272 return &p->exit_code;
1273 } else {
1274 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1275 return &p->signal->group_exit_code;
1276 }
1277 return NULL;
1278}
1279
1280/**
1281 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1282 * @wo: wait options
1283 * @ptrace: is the wait for ptrace
1284 * @p: task to wait for
1285 *
1286 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1287 *
1288 * CONTEXT:
1289 * read_lock(&tasklist_lock), which is released if return value is
1290 * non-zero. Also, grabs and releases @p->sighand->siglock.
1291 *
1292 * RETURNS:
1293 * 0 if wait condition didn't exist and search for other wait conditions
1294 * should continue. Non-zero return, -errno on failure and @p's pid on
1295 * success, implies that tasklist_lock is released and wait condition
1296 * search should terminate.
1297 */
1298static int wait_task_stopped(struct wait_opts *wo,
1299 int ptrace, struct task_struct *p)
1300{
1301 struct waitid_info *infop;
1302 int exit_code, *p_code, why;
1303 uid_t uid = 0; /* unneeded, required by compiler */
1304 pid_t pid;
1305
1306 /*
1307 * Traditionally we see ptrace'd stopped tasks regardless of options.
1308 */
1309 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1310 return 0;
1311
1312 if (!task_stopped_code(p, ptrace))
1313 return 0;
1314
1315 exit_code = 0;
1316 spin_lock_irq(&p->sighand->siglock);
1317
1318 p_code = task_stopped_code(p, ptrace);
1319 if (unlikely(!p_code))
1320 goto unlock_sig;
1321
1322 exit_code = *p_code;
1323 if (!exit_code)
1324 goto unlock_sig;
1325
1326 if (!unlikely(wo->wo_flags & WNOWAIT))
1327 *p_code = 0;
1328
1329 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1330unlock_sig:
1331 spin_unlock_irq(&p->sighand->siglock);
1332 if (!exit_code)
1333 return 0;
1334
1335 /*
1336 * Now we are pretty sure this task is interesting.
1337 * Make sure it doesn't get reaped out from under us while we
1338 * give up the lock and then examine it below. We don't want to
1339 * keep holding onto the tasklist_lock while we call getrusage and
1340 * possibly take page faults for user memory.
1341 */
1342 get_task_struct(p);
1343 pid = task_pid_vnr(p);
1344 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1345 read_unlock(&tasklist_lock);
1346 sched_annotate_sleep();
1347 if (wo->wo_rusage)
1348 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1349 put_task_struct(p);
1350
1351 if (likely(!(wo->wo_flags & WNOWAIT)))
1352 wo->wo_stat = (exit_code << 8) | 0x7f;
1353
1354 infop = wo->wo_info;
1355 if (infop) {
1356 infop->cause = why;
1357 infop->status = exit_code;
1358 infop->pid = pid;
1359 infop->uid = uid;
1360 }
1361 return pid;
1362}
1363
1364/*
1365 * Handle do_wait work for one task in a live, non-stopped state.
1366 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1367 * the lock and this task is uninteresting. If we return nonzero, we have
1368 * released the lock and the system call should return.
1369 */
1370static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1371{
1372 struct waitid_info *infop;
1373 pid_t pid;
1374 uid_t uid;
1375
1376 if (!unlikely(wo->wo_flags & WCONTINUED))
1377 return 0;
1378
1379 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1380 return 0;
1381
1382 spin_lock_irq(&p->sighand->siglock);
1383 /* Re-check with the lock held. */
1384 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1385 spin_unlock_irq(&p->sighand->siglock);
1386 return 0;
1387 }
1388 if (!unlikely(wo->wo_flags & WNOWAIT))
1389 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1390 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1391 spin_unlock_irq(&p->sighand->siglock);
1392
1393 pid = task_pid_vnr(p);
1394 get_task_struct(p);
1395 read_unlock(&tasklist_lock);
1396 sched_annotate_sleep();
1397 if (wo->wo_rusage)
1398 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1399 put_task_struct(p);
1400
1401 infop = wo->wo_info;
1402 if (!infop) {
1403 wo->wo_stat = 0xffff;
1404 } else {
1405 infop->cause = CLD_CONTINUED;
1406 infop->pid = pid;
1407 infop->uid = uid;
1408 infop->status = SIGCONT;
1409 }
1410 return pid;
1411}
1412
1413/*
1414 * Consider @p for a wait by @parent.
1415 *
1416 * -ECHILD should be in ->notask_error before the first call.
1417 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1418 * Returns zero if the search for a child should continue;
1419 * then ->notask_error is 0 if @p is an eligible child,
1420 * or still -ECHILD.
1421 */
1422static int wait_consider_task(struct wait_opts *wo, int ptrace,
1423 struct task_struct *p)
1424{
1425 /*
1426 * We can race with wait_task_zombie() from another thread.
1427 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1428 * can't confuse the checks below.
1429 */
1430 int exit_state = READ_ONCE(p->exit_state);
1431 int ret;
1432
1433 if (unlikely(exit_state == EXIT_DEAD))
1434 return 0;
1435
1436 ret = eligible_child(wo, ptrace, p);
1437 if (!ret)
1438 return ret;
1439
1440 if (unlikely(exit_state == EXIT_TRACE)) {
1441 /*
1442 * ptrace == 0 means we are the natural parent. In this case
1443 * we should clear notask_error, debugger will notify us.
1444 */
1445 if (likely(!ptrace))
1446 wo->notask_error = 0;
1447 return 0;
1448 }
1449
1450 if (likely(!ptrace) && unlikely(p->ptrace)) {
1451 /*
1452 * If it is traced by its real parent's group, just pretend
1453 * the caller is ptrace_do_wait() and reap this child if it
1454 * is zombie.
1455 *
1456 * This also hides group stop state from real parent; otherwise
1457 * a single stop can be reported twice as group and ptrace stop.
1458 * If a ptracer wants to distinguish these two events for its
1459 * own children it should create a separate process which takes
1460 * the role of real parent.
1461 */
1462 if (!ptrace_reparented(p))
1463 ptrace = 1;
1464 }
1465
1466 /* slay zombie? */
1467 if (exit_state == EXIT_ZOMBIE) {
1468 /* we don't reap group leaders with subthreads */
1469 if (!delay_group_leader(p)) {
1470 /*
1471 * A zombie ptracee is only visible to its ptracer.
1472 * Notification and reaping will be cascaded to the
1473 * real parent when the ptracer detaches.
1474 */
1475 if (unlikely(ptrace) || likely(!p->ptrace))
1476 return wait_task_zombie(wo, p);
1477 }
1478
1479 /*
1480 * Allow access to stopped/continued state via zombie by
1481 * falling through. Clearing of notask_error is complex.
1482 *
1483 * When !@ptrace:
1484 *
1485 * If WEXITED is set, notask_error should naturally be
1486 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1487 * so, if there are live subthreads, there are events to
1488 * wait for. If all subthreads are dead, it's still safe
1489 * to clear - this function will be called again in finite
1490 * amount time once all the subthreads are released and
1491 * will then return without clearing.
1492 *
1493 * When @ptrace:
1494 *
1495 * Stopped state is per-task and thus can't change once the
1496 * target task dies. Only continued and exited can happen.
1497 * Clear notask_error if WCONTINUED | WEXITED.
1498 */
1499 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1500 wo->notask_error = 0;
1501 } else {
1502 /*
1503 * @p is alive and it's gonna stop, continue or exit, so
1504 * there always is something to wait for.
1505 */
1506 wo->notask_error = 0;
1507 }
1508
1509 /*
1510 * Wait for stopped. Depending on @ptrace, different stopped state
1511 * is used and the two don't interact with each other.
1512 */
1513 ret = wait_task_stopped(wo, ptrace, p);
1514 if (ret)
1515 return ret;
1516
1517 /*
1518 * Wait for continued. There's only one continued state and the
1519 * ptracer can consume it which can confuse the real parent. Don't
1520 * use WCONTINUED from ptracer. You don't need or want it.
1521 */
1522 return wait_task_continued(wo, p);
1523}
1524
1525/*
1526 * Do the work of do_wait() for one thread in the group, @tsk.
1527 *
1528 * -ECHILD should be in ->notask_error before the first call.
1529 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1530 * Returns zero if the search for a child should continue; then
1531 * ->notask_error is 0 if there were any eligible children,
1532 * or still -ECHILD.
1533 */
1534static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1535{
1536 struct task_struct *p;
1537
1538 list_for_each_entry(p, &tsk->children, sibling) {
1539 int ret = wait_consider_task(wo, 0, p);
1540
1541 if (ret)
1542 return ret;
1543 }
1544
1545 return 0;
1546}
1547
1548static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1549{
1550 struct task_struct *p;
1551
1552 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1553 int ret = wait_consider_task(wo, 1, p);
1554
1555 if (ret)
1556 return ret;
1557 }
1558
1559 return 0;
1560}
1561
1562bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1563{
1564 if (!eligible_pid(wo, p))
1565 return false;
1566
1567 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1568 return false;
1569
1570 return true;
1571}
1572
1573static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1574 int sync, void *key)
1575{
1576 struct wait_opts *wo = container_of(wait, struct wait_opts,
1577 child_wait);
1578 struct task_struct *p = key;
1579
1580 if (pid_child_should_wake(wo, p))
1581 return default_wake_function(wait, mode, sync, key);
1582
1583 return 0;
1584}
1585
1586void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1587{
1588 __wake_up_sync_key(&parent->signal->wait_chldexit,
1589 TASK_INTERRUPTIBLE, p);
1590}
1591
1592static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1593 struct task_struct *target)
1594{
1595 struct task_struct *parent =
1596 !ptrace ? target->real_parent : target->parent;
1597
1598 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1599 same_thread_group(current, parent));
1600}
1601
1602/*
1603 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1604 * and tracee lists to find the target task.
1605 */
1606static int do_wait_pid(struct wait_opts *wo)
1607{
1608 bool ptrace;
1609 struct task_struct *target;
1610 int retval;
1611
1612 ptrace = false;
1613 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1614 if (target && is_effectively_child(wo, ptrace, target)) {
1615 retval = wait_consider_task(wo, ptrace, target);
1616 if (retval)
1617 return retval;
1618 }
1619
1620 ptrace = true;
1621 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1622 if (target && target->ptrace &&
1623 is_effectively_child(wo, ptrace, target)) {
1624 retval = wait_consider_task(wo, ptrace, target);
1625 if (retval)
1626 return retval;
1627 }
1628
1629 return 0;
1630}
1631
1632long __do_wait(struct wait_opts *wo)
1633{
1634 long retval;
1635
1636 /*
1637 * If there is nothing that can match our criteria, just get out.
1638 * We will clear ->notask_error to zero if we see any child that
1639 * might later match our criteria, even if we are not able to reap
1640 * it yet.
1641 */
1642 wo->notask_error = -ECHILD;
1643 if ((wo->wo_type < PIDTYPE_MAX) &&
1644 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1645 goto notask;
1646
1647 read_lock(&tasklist_lock);
1648
1649 if (wo->wo_type == PIDTYPE_PID) {
1650 retval = do_wait_pid(wo);
1651 if (retval)
1652 return retval;
1653 } else {
1654 struct task_struct *tsk = current;
1655
1656 do {
1657 retval = do_wait_thread(wo, tsk);
1658 if (retval)
1659 return retval;
1660
1661 retval = ptrace_do_wait(wo, tsk);
1662 if (retval)
1663 return retval;
1664
1665 if (wo->wo_flags & __WNOTHREAD)
1666 break;
1667 } while_each_thread(current, tsk);
1668 }
1669 read_unlock(&tasklist_lock);
1670
1671notask:
1672 retval = wo->notask_error;
1673 if (!retval && !(wo->wo_flags & WNOHANG))
1674 return -ERESTARTSYS;
1675
1676 return retval;
1677}
1678
1679static long do_wait(struct wait_opts *wo)
1680{
1681 int retval;
1682
1683 trace_sched_process_wait(wo->wo_pid);
1684
1685 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1686 wo->child_wait.private = current;
1687 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1688
1689 do {
1690 set_current_state(TASK_INTERRUPTIBLE);
1691 retval = __do_wait(wo);
1692 if (retval != -ERESTARTSYS)
1693 break;
1694 if (signal_pending(current))
1695 break;
1696 schedule();
1697 } while (1);
1698
1699 __set_current_state(TASK_RUNNING);
1700 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1701 return retval;
1702}
1703
1704int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1705 struct waitid_info *infop, int options,
1706 struct rusage *ru)
1707{
1708 unsigned int f_flags = 0;
1709 struct pid *pid = NULL;
1710 enum pid_type type;
1711
1712 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1713 __WNOTHREAD|__WCLONE|__WALL))
1714 return -EINVAL;
1715 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1716 return -EINVAL;
1717
1718 switch (which) {
1719 case P_ALL:
1720 type = PIDTYPE_MAX;
1721 break;
1722 case P_PID:
1723 type = PIDTYPE_PID;
1724 if (upid <= 0)
1725 return -EINVAL;
1726
1727 pid = find_get_pid(upid);
1728 break;
1729 case P_PGID:
1730 type = PIDTYPE_PGID;
1731 if (upid < 0)
1732 return -EINVAL;
1733
1734 if (upid)
1735 pid = find_get_pid(upid);
1736 else
1737 pid = get_task_pid(current, PIDTYPE_PGID);
1738 break;
1739 case P_PIDFD:
1740 type = PIDTYPE_PID;
1741 if (upid < 0)
1742 return -EINVAL;
1743
1744 pid = pidfd_get_pid(upid, &f_flags);
1745 if (IS_ERR(pid))
1746 return PTR_ERR(pid);
1747
1748 break;
1749 default:
1750 return -EINVAL;
1751 }
1752
1753 wo->wo_type = type;
1754 wo->wo_pid = pid;
1755 wo->wo_flags = options;
1756 wo->wo_info = infop;
1757 wo->wo_rusage = ru;
1758 if (f_flags & O_NONBLOCK)
1759 wo->wo_flags |= WNOHANG;
1760
1761 return 0;
1762}
1763
1764static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1765 int options, struct rusage *ru)
1766{
1767 struct wait_opts wo;
1768 long ret;
1769
1770 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1771 if (ret)
1772 return ret;
1773
1774 ret = do_wait(&wo);
1775 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1776 ret = -EAGAIN;
1777
1778 put_pid(wo.wo_pid);
1779 return ret;
1780}
1781
1782SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1783 infop, int, options, struct rusage __user *, ru)
1784{
1785 struct rusage r;
1786 struct waitid_info info = {.status = 0};
1787 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1788 int signo = 0;
1789
1790 if (err > 0) {
1791 signo = SIGCHLD;
1792 err = 0;
1793 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1794 return -EFAULT;
1795 }
1796 if (!infop)
1797 return err;
1798
1799 if (!user_write_access_begin(infop, sizeof(*infop)))
1800 return -EFAULT;
1801
1802 unsafe_put_user(signo, &infop->si_signo, Efault);
1803 unsafe_put_user(0, &infop->si_errno, Efault);
1804 unsafe_put_user(info.cause, &infop->si_code, Efault);
1805 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1806 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1807 unsafe_put_user(info.status, &infop->si_status, Efault);
1808 user_write_access_end();
1809 return err;
1810Efault:
1811 user_write_access_end();
1812 return -EFAULT;
1813}
1814
1815long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1816 struct rusage *ru)
1817{
1818 struct wait_opts wo;
1819 struct pid *pid = NULL;
1820 enum pid_type type;
1821 long ret;
1822
1823 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1824 __WNOTHREAD|__WCLONE|__WALL))
1825 return -EINVAL;
1826
1827 /* -INT_MIN is not defined */
1828 if (upid == INT_MIN)
1829 return -ESRCH;
1830
1831 if (upid == -1)
1832 type = PIDTYPE_MAX;
1833 else if (upid < 0) {
1834 type = PIDTYPE_PGID;
1835 pid = find_get_pid(-upid);
1836 } else if (upid == 0) {
1837 type = PIDTYPE_PGID;
1838 pid = get_task_pid(current, PIDTYPE_PGID);
1839 } else /* upid > 0 */ {
1840 type = PIDTYPE_PID;
1841 pid = find_get_pid(upid);
1842 }
1843
1844 wo.wo_type = type;
1845 wo.wo_pid = pid;
1846 wo.wo_flags = options | WEXITED;
1847 wo.wo_info = NULL;
1848 wo.wo_stat = 0;
1849 wo.wo_rusage = ru;
1850 ret = do_wait(&wo);
1851 put_pid(pid);
1852 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1853 ret = -EFAULT;
1854
1855 return ret;
1856}
1857
1858int kernel_wait(pid_t pid, int *stat)
1859{
1860 struct wait_opts wo = {
1861 .wo_type = PIDTYPE_PID,
1862 .wo_pid = find_get_pid(pid),
1863 .wo_flags = WEXITED,
1864 };
1865 int ret;
1866
1867 ret = do_wait(&wo);
1868 if (ret > 0 && wo.wo_stat)
1869 *stat = wo.wo_stat;
1870 put_pid(wo.wo_pid);
1871 return ret;
1872}
1873
1874SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1875 int, options, struct rusage __user *, ru)
1876{
1877 struct rusage r;
1878 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1879
1880 if (err > 0) {
1881 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1882 return -EFAULT;
1883 }
1884 return err;
1885}
1886
1887#ifdef __ARCH_WANT_SYS_WAITPID
1888
1889/*
1890 * sys_waitpid() remains for compatibility. waitpid() should be
1891 * implemented by calling sys_wait4() from libc.a.
1892 */
1893SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1894{
1895 return kernel_wait4(pid, stat_addr, options, NULL);
1896}
1897
1898#endif
1899
1900#ifdef CONFIG_COMPAT
1901COMPAT_SYSCALL_DEFINE4(wait4,
1902 compat_pid_t, pid,
1903 compat_uint_t __user *, stat_addr,
1904 int, options,
1905 struct compat_rusage __user *, ru)
1906{
1907 struct rusage r;
1908 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1909 if (err > 0) {
1910 if (ru && put_compat_rusage(&r, ru))
1911 return -EFAULT;
1912 }
1913 return err;
1914}
1915
1916COMPAT_SYSCALL_DEFINE5(waitid,
1917 int, which, compat_pid_t, pid,
1918 struct compat_siginfo __user *, infop, int, options,
1919 struct compat_rusage __user *, uru)
1920{
1921 struct rusage ru;
1922 struct waitid_info info = {.status = 0};
1923 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1924 int signo = 0;
1925 if (err > 0) {
1926 signo = SIGCHLD;
1927 err = 0;
1928 if (uru) {
1929 /* kernel_waitid() overwrites everything in ru */
1930 if (COMPAT_USE_64BIT_TIME)
1931 err = copy_to_user(uru, &ru, sizeof(ru));
1932 else
1933 err = put_compat_rusage(&ru, uru);
1934 if (err)
1935 return -EFAULT;
1936 }
1937 }
1938
1939 if (!infop)
1940 return err;
1941
1942 if (!user_write_access_begin(infop, sizeof(*infop)))
1943 return -EFAULT;
1944
1945 unsafe_put_user(signo, &infop->si_signo, Efault);
1946 unsafe_put_user(0, &infop->si_errno, Efault);
1947 unsafe_put_user(info.cause, &infop->si_code, Efault);
1948 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1949 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1950 unsafe_put_user(info.status, &infop->si_status, Efault);
1951 user_write_access_end();
1952 return err;
1953Efault:
1954 user_write_access_end();
1955 return -EFAULT;
1956}
1957#endif
1958
1959/*
1960 * This needs to be __function_aligned as GCC implicitly makes any
1961 * implementation of abort() cold and drops alignment specified by
1962 * -falign-functions=N.
1963 *
1964 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1965 */
1966__weak __function_aligned void abort(void)
1967{
1968 BUG();
1969
1970 /* if that doesn't kill us, halt */
1971 panic("Oops failed to kill thread");
1972}
1973EXPORT_SYMBOL(abort);