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/interrupt.h>
10#include <linux/module.h>
11#include <linux/capability.h>
12#include <linux/completion.h>
13#include <linux/personality.h>
14#include <linux/tty.h>
15#include <linux/iocontext.h>
16#include <linux/key.h>
17#include <linux/security.h>
18#include <linux/cpu.h>
19#include <linux/acct.h>
20#include <linux/tsacct_kern.h>
21#include <linux/file.h>
22#include <linux/fdtable.h>
23#include <linux/freezer.h>
24#include <linux/binfmts.h>
25#include <linux/nsproxy.h>
26#include <linux/pid_namespace.h>
27#include <linux/ptrace.h>
28#include <linux/profile.h>
29#include <linux/mount.h>
30#include <linux/proc_fs.h>
31#include <linux/kthread.h>
32#include <linux/mempolicy.h>
33#include <linux/taskstats_kern.h>
34#include <linux/delayacct.h>
35#include <linux/cgroup.h>
36#include <linux/syscalls.h>
37#include <linux/signal.h>
38#include <linux/posix-timers.h>
39#include <linux/cn_proc.h>
40#include <linux/mutex.h>
41#include <linux/futex.h>
42#include <linux/pipe_fs_i.h>
43#include <linux/audit.h> /* for audit_free() */
44#include <linux/resource.h>
45#include <linux/blkdev.h>
46#include <linux/task_io_accounting_ops.h>
47#include <linux/tracehook.h>
48#include <linux/fs_struct.h>
49#include <linux/init_task.h>
50#include <linux/perf_event.h>
51#include <trace/events/sched.h>
52#include <linux/hw_breakpoint.h>
53#include <linux/oom.h>
54#include <linux/writeback.h>
55#include <linux/shm.h>
56
57#include <asm/uaccess.h>
58#include <asm/unistd.h>
59#include <asm/pgtable.h>
60#include <asm/mmu_context.h>
61
62static void exit_mm(struct task_struct * tsk);
63
64static void __unhash_process(struct task_struct *p, bool group_dead)
65{
66 nr_threads--;
67 detach_pid(p, PIDTYPE_PID);
68 if (group_dead) {
69 detach_pid(p, PIDTYPE_PGID);
70 detach_pid(p, PIDTYPE_SID);
71
72 list_del_rcu(&p->tasks);
73 list_del_init(&p->sibling);
74 __this_cpu_dec(process_counts);
75 }
76 list_del_rcu(&p->thread_group);
77 list_del_rcu(&p->thread_node);
78}
79
80/*
81 * This function expects the tasklist_lock write-locked.
82 */
83static void __exit_signal(struct task_struct *tsk)
84{
85 struct signal_struct *sig = tsk->signal;
86 bool group_dead = thread_group_leader(tsk);
87 struct sighand_struct *sighand;
88 struct tty_struct *uninitialized_var(tty);
89 cputime_t utime, stime;
90
91 sighand = rcu_dereference_check(tsk->sighand,
92 lockdep_tasklist_lock_is_held());
93 spin_lock(&sighand->siglock);
94
95 posix_cpu_timers_exit(tsk);
96 if (group_dead) {
97 posix_cpu_timers_exit_group(tsk);
98 tty = sig->tty;
99 sig->tty = NULL;
100 } else {
101 /*
102 * This can only happen if the caller is de_thread().
103 * FIXME: this is the temporary hack, we should teach
104 * posix-cpu-timers to handle this case correctly.
105 */
106 if (unlikely(has_group_leader_pid(tsk)))
107 posix_cpu_timers_exit_group(tsk);
108
109 /*
110 * If there is any task waiting for the group exit
111 * then notify it:
112 */
113 if (sig->notify_count > 0 && !--sig->notify_count)
114 wake_up_process(sig->group_exit_task);
115
116 if (tsk == sig->curr_target)
117 sig->curr_target = next_thread(tsk);
118 /*
119 * Accumulate here the counters for all threads but the
120 * group leader as they die, so they can be added into
121 * the process-wide totals when those are taken.
122 * The group leader stays around as a zombie as long
123 * as there are other threads. When it gets reaped,
124 * the exit.c code will add its counts into these totals.
125 * We won't ever get here for the group leader, since it
126 * will have been the last reference on the signal_struct.
127 */
128 task_cputime(tsk, &utime, &stime);
129 sig->utime += utime;
130 sig->stime += stime;
131 sig->gtime += task_gtime(tsk);
132 sig->min_flt += tsk->min_flt;
133 sig->maj_flt += tsk->maj_flt;
134 sig->nvcsw += tsk->nvcsw;
135 sig->nivcsw += tsk->nivcsw;
136 sig->inblock += task_io_get_inblock(tsk);
137 sig->oublock += task_io_get_oublock(tsk);
138 task_io_accounting_add(&sig->ioac, &tsk->ioac);
139 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
140 }
141
142 sig->nr_threads--;
143 __unhash_process(tsk, group_dead);
144
145 /*
146 * Do this under ->siglock, we can race with another thread
147 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
148 */
149 flush_sigqueue(&tsk->pending);
150 tsk->sighand = NULL;
151 spin_unlock(&sighand->siglock);
152
153 __cleanup_sighand(sighand);
154 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
155 if (group_dead) {
156 flush_sigqueue(&sig->shared_pending);
157 tty_kref_put(tty);
158 }
159}
160
161static void delayed_put_task_struct(struct rcu_head *rhp)
162{
163 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
164
165 perf_event_delayed_put(tsk);
166 trace_sched_process_free(tsk);
167 put_task_struct(tsk);
168}
169
170
171void release_task(struct task_struct * p)
172{
173 struct task_struct *leader;
174 int zap_leader;
175repeat:
176 /* don't need to get the RCU readlock here - the process is dead and
177 * can't be modifying its own credentials. But shut RCU-lockdep up */
178 rcu_read_lock();
179 atomic_dec(&__task_cred(p)->user->processes);
180 rcu_read_unlock();
181
182 proc_flush_task(p);
183
184 write_lock_irq(&tasklist_lock);
185 ptrace_release_task(p);
186 __exit_signal(p);
187
188 /*
189 * If we are the last non-leader member of the thread
190 * group, and the leader is zombie, then notify the
191 * group leader's parent process. (if it wants notification.)
192 */
193 zap_leader = 0;
194 leader = p->group_leader;
195 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
196 /*
197 * If we were the last child thread and the leader has
198 * exited already, and the leader's parent ignores SIGCHLD,
199 * then we are the one who should release the leader.
200 */
201 zap_leader = do_notify_parent(leader, leader->exit_signal);
202 if (zap_leader)
203 leader->exit_state = EXIT_DEAD;
204 }
205
206 write_unlock_irq(&tasklist_lock);
207 release_thread(p);
208 call_rcu(&p->rcu, delayed_put_task_struct);
209
210 p = leader;
211 if (unlikely(zap_leader))
212 goto repeat;
213}
214
215/*
216 * This checks not only the pgrp, but falls back on the pid if no
217 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
218 * without this...
219 *
220 * The caller must hold rcu lock or the tasklist lock.
221 */
222struct pid *session_of_pgrp(struct pid *pgrp)
223{
224 struct task_struct *p;
225 struct pid *sid = NULL;
226
227 p = pid_task(pgrp, PIDTYPE_PGID);
228 if (p == NULL)
229 p = pid_task(pgrp, PIDTYPE_PID);
230 if (p != NULL)
231 sid = task_session(p);
232
233 return sid;
234}
235
236/*
237 * Determine if a process group is "orphaned", according to the POSIX
238 * definition in 2.2.2.52. Orphaned process groups are not to be affected
239 * by terminal-generated stop signals. Newly orphaned process groups are
240 * to receive a SIGHUP and a SIGCONT.
241 *
242 * "I ask you, have you ever known what it is to be an orphan?"
243 */
244static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
245{
246 struct task_struct *p;
247
248 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
249 if ((p == ignored_task) ||
250 (p->exit_state && thread_group_empty(p)) ||
251 is_global_init(p->real_parent))
252 continue;
253
254 if (task_pgrp(p->real_parent) != pgrp &&
255 task_session(p->real_parent) == task_session(p))
256 return 0;
257 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
258
259 return 1;
260}
261
262int is_current_pgrp_orphaned(void)
263{
264 int retval;
265
266 read_lock(&tasklist_lock);
267 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
268 read_unlock(&tasklist_lock);
269
270 return retval;
271}
272
273static bool has_stopped_jobs(struct pid *pgrp)
274{
275 struct task_struct *p;
276
277 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
278 if (p->signal->flags & SIGNAL_STOP_STOPPED)
279 return true;
280 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
281
282 return false;
283}
284
285/*
286 * Check to see if any process groups have become orphaned as
287 * a result of our exiting, and if they have any stopped jobs,
288 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
289 */
290static void
291kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
292{
293 struct pid *pgrp = task_pgrp(tsk);
294 struct task_struct *ignored_task = tsk;
295
296 if (!parent)
297 /* exit: our father is in a different pgrp than
298 * we are and we were the only connection outside.
299 */
300 parent = tsk->real_parent;
301 else
302 /* reparent: our child is in a different pgrp than
303 * we are, and it was the only connection outside.
304 */
305 ignored_task = NULL;
306
307 if (task_pgrp(parent) != pgrp &&
308 task_session(parent) == task_session(tsk) &&
309 will_become_orphaned_pgrp(pgrp, ignored_task) &&
310 has_stopped_jobs(pgrp)) {
311 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
312 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
313 }
314}
315
316/*
317 * Let kernel threads use this to say that they allow a certain signal.
318 * Must not be used if kthread was cloned with CLONE_SIGHAND.
319 */
320int allow_signal(int sig)
321{
322 if (!valid_signal(sig) || sig < 1)
323 return -EINVAL;
324
325 spin_lock_irq(¤t->sighand->siglock);
326 /* This is only needed for daemonize()'ed kthreads */
327 sigdelset(¤t->blocked, sig);
328 /*
329 * Kernel threads handle their own signals. Let the signal code
330 * know it'll be handled, so that they don't get converted to
331 * SIGKILL or just silently dropped.
332 */
333 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
334 recalc_sigpending();
335 spin_unlock_irq(¤t->sighand->siglock);
336 return 0;
337}
338
339EXPORT_SYMBOL(allow_signal);
340
341int disallow_signal(int sig)
342{
343 if (!valid_signal(sig) || sig < 1)
344 return -EINVAL;
345
346 spin_lock_irq(¤t->sighand->siglock);
347 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
348 recalc_sigpending();
349 spin_unlock_irq(¤t->sighand->siglock);
350 return 0;
351}
352
353EXPORT_SYMBOL(disallow_signal);
354
355#ifdef CONFIG_MM_OWNER
356/*
357 * A task is exiting. If it owned this mm, find a new owner for the mm.
358 */
359void mm_update_next_owner(struct mm_struct *mm)
360{
361 struct task_struct *c, *g, *p = current;
362
363retry:
364 /*
365 * If the exiting or execing task is not the owner, it's
366 * someone else's problem.
367 */
368 if (mm->owner != p)
369 return;
370 /*
371 * The current owner is exiting/execing and there are no other
372 * candidates. Do not leave the mm pointing to a possibly
373 * freed task structure.
374 */
375 if (atomic_read(&mm->mm_users) <= 1) {
376 mm->owner = NULL;
377 return;
378 }
379
380 read_lock(&tasklist_lock);
381 /*
382 * Search in the children
383 */
384 list_for_each_entry(c, &p->children, sibling) {
385 if (c->mm == mm)
386 goto assign_new_owner;
387 }
388
389 /*
390 * Search in the siblings
391 */
392 list_for_each_entry(c, &p->real_parent->children, sibling) {
393 if (c->mm == mm)
394 goto assign_new_owner;
395 }
396
397 /*
398 * Search through everything else. We should not get
399 * here often
400 */
401 do_each_thread(g, c) {
402 if (c->mm == mm)
403 goto assign_new_owner;
404 } while_each_thread(g, c);
405
406 read_unlock(&tasklist_lock);
407 /*
408 * We found no owner yet mm_users > 1: this implies that we are
409 * most likely racing with swapoff (try_to_unuse()) or /proc or
410 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
411 */
412 mm->owner = NULL;
413 return;
414
415assign_new_owner:
416 BUG_ON(c == p);
417 get_task_struct(c);
418 /*
419 * The task_lock protects c->mm from changing.
420 * We always want mm->owner->mm == mm
421 */
422 task_lock(c);
423 /*
424 * Delay read_unlock() till we have the task_lock()
425 * to ensure that c does not slip away underneath us
426 */
427 read_unlock(&tasklist_lock);
428 if (c->mm != mm) {
429 task_unlock(c);
430 put_task_struct(c);
431 goto retry;
432 }
433 mm->owner = c;
434 task_unlock(c);
435 put_task_struct(c);
436}
437#endif /* CONFIG_MM_OWNER */
438
439/*
440 * Turn us into a lazy TLB process if we
441 * aren't already..
442 */
443static void exit_mm(struct task_struct * tsk)
444{
445 struct mm_struct *mm = tsk->mm;
446 struct core_state *core_state;
447
448 mm_release(tsk, mm);
449 if (!mm)
450 return;
451 sync_mm_rss(mm);
452 /*
453 * Serialize with any possible pending coredump.
454 * We must hold mmap_sem around checking core_state
455 * and clearing tsk->mm. The core-inducing thread
456 * will increment ->nr_threads for each thread in the
457 * group with ->mm != NULL.
458 */
459 down_read(&mm->mmap_sem);
460 core_state = mm->core_state;
461 if (core_state) {
462 struct core_thread self;
463 up_read(&mm->mmap_sem);
464
465 self.task = tsk;
466 self.next = xchg(&core_state->dumper.next, &self);
467 /*
468 * Implies mb(), the result of xchg() must be visible
469 * to core_state->dumper.
470 */
471 if (atomic_dec_and_test(&core_state->nr_threads))
472 complete(&core_state->startup);
473
474 for (;;) {
475 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
476 if (!self.task) /* see coredump_finish() */
477 break;
478 freezable_schedule();
479 }
480 __set_task_state(tsk, TASK_RUNNING);
481 down_read(&mm->mmap_sem);
482 }
483 atomic_inc(&mm->mm_count);
484 BUG_ON(mm != tsk->active_mm);
485 /* more a memory barrier than a real lock */
486 task_lock(tsk);
487 tsk->mm = NULL;
488 up_read(&mm->mmap_sem);
489 enter_lazy_tlb(mm, current);
490 task_unlock(tsk);
491 mm_update_next_owner(mm);
492 mmput(mm);
493}
494
495/*
496 * When we die, we re-parent all our children, and try to:
497 * 1. give them to another thread in our thread group, if such a member exists
498 * 2. give it to the first ancestor process which prctl'd itself as a
499 * child_subreaper for its children (like a service manager)
500 * 3. give it to the init process (PID 1) in our pid namespace
501 */
502static struct task_struct *find_new_reaper(struct task_struct *father)
503 __releases(&tasklist_lock)
504 __acquires(&tasklist_lock)
505{
506 struct pid_namespace *pid_ns = task_active_pid_ns(father);
507 struct task_struct *thread;
508
509 thread = father;
510 while_each_thread(father, thread) {
511 if (thread->flags & PF_EXITING)
512 continue;
513 if (unlikely(pid_ns->child_reaper == father))
514 pid_ns->child_reaper = thread;
515 return thread;
516 }
517
518 if (unlikely(pid_ns->child_reaper == father)) {
519 write_unlock_irq(&tasklist_lock);
520 if (unlikely(pid_ns == &init_pid_ns)) {
521 panic("Attempted to kill init! exitcode=0x%08x\n",
522 father->signal->group_exit_code ?:
523 father->exit_code);
524 }
525
526 zap_pid_ns_processes(pid_ns);
527 write_lock_irq(&tasklist_lock);
528 } else if (father->signal->has_child_subreaper) {
529 struct task_struct *reaper;
530
531 /*
532 * Find the first ancestor marked as child_subreaper.
533 * Note that the code below checks same_thread_group(reaper,
534 * pid_ns->child_reaper). This is what we need to DTRT in a
535 * PID namespace. However we still need the check above, see
536 * http://marc.info/?l=linux-kernel&m=131385460420380
537 */
538 for (reaper = father->real_parent;
539 reaper != &init_task;
540 reaper = reaper->real_parent) {
541 if (same_thread_group(reaper, pid_ns->child_reaper))
542 break;
543 if (!reaper->signal->is_child_subreaper)
544 continue;
545 thread = reaper;
546 do {
547 if (!(thread->flags & PF_EXITING))
548 return reaper;
549 } while_each_thread(reaper, thread);
550 }
551 }
552
553 return pid_ns->child_reaper;
554}
555
556/*
557* Any that need to be release_task'd are put on the @dead list.
558 */
559static void reparent_leader(struct task_struct *father, struct task_struct *p,
560 struct list_head *dead)
561{
562 list_move_tail(&p->sibling, &p->real_parent->children);
563
564 if (p->exit_state == EXIT_DEAD)
565 return;
566 /*
567 * If this is a threaded reparent there is no need to
568 * notify anyone anything has happened.
569 */
570 if (same_thread_group(p->real_parent, father))
571 return;
572
573 /* We don't want people slaying init. */
574 p->exit_signal = SIGCHLD;
575
576 /* If it has exited notify the new parent about this child's death. */
577 if (!p->ptrace &&
578 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
579 if (do_notify_parent(p, p->exit_signal)) {
580 p->exit_state = EXIT_DEAD;
581 list_move_tail(&p->sibling, dead);
582 }
583 }
584
585 kill_orphaned_pgrp(p, father);
586}
587
588static void forget_original_parent(struct task_struct *father)
589{
590 struct task_struct *p, *n, *reaper;
591 LIST_HEAD(dead_children);
592
593 write_lock_irq(&tasklist_lock);
594 /*
595 * Note that exit_ptrace() and find_new_reaper() might
596 * drop tasklist_lock and reacquire it.
597 */
598 exit_ptrace(father);
599 reaper = find_new_reaper(father);
600
601 list_for_each_entry_safe(p, n, &father->children, sibling) {
602 struct task_struct *t = p;
603 do {
604 t->real_parent = reaper;
605 if (t->parent == father) {
606 BUG_ON(t->ptrace);
607 t->parent = t->real_parent;
608 }
609 if (t->pdeath_signal)
610 group_send_sig_info(t->pdeath_signal,
611 SEND_SIG_NOINFO, t);
612 } while_each_thread(p, t);
613 reparent_leader(father, p, &dead_children);
614 }
615 write_unlock_irq(&tasklist_lock);
616
617 BUG_ON(!list_empty(&father->children));
618
619 list_for_each_entry_safe(p, n, &dead_children, sibling) {
620 list_del_init(&p->sibling);
621 release_task(p);
622 }
623}
624
625/*
626 * Send signals to all our closest relatives so that they know
627 * to properly mourn us..
628 */
629static void exit_notify(struct task_struct *tsk, int group_dead)
630{
631 bool autoreap;
632
633 /*
634 * This does two things:
635 *
636 * A. Make init inherit all the child processes
637 * B. Check to see if any process groups have become orphaned
638 * as a result of our exiting, and if they have any stopped
639 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
640 */
641 forget_original_parent(tsk);
642
643 write_lock_irq(&tasklist_lock);
644 if (group_dead)
645 kill_orphaned_pgrp(tsk->group_leader, NULL);
646
647 if (unlikely(tsk->ptrace)) {
648 int sig = thread_group_leader(tsk) &&
649 thread_group_empty(tsk) &&
650 !ptrace_reparented(tsk) ?
651 tsk->exit_signal : SIGCHLD;
652 autoreap = do_notify_parent(tsk, sig);
653 } else if (thread_group_leader(tsk)) {
654 autoreap = thread_group_empty(tsk) &&
655 do_notify_parent(tsk, tsk->exit_signal);
656 } else {
657 autoreap = true;
658 }
659
660 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
661
662 /* mt-exec, de_thread() is waiting for group leader */
663 if (unlikely(tsk->signal->notify_count < 0))
664 wake_up_process(tsk->signal->group_exit_task);
665 write_unlock_irq(&tasklist_lock);
666
667 /* If the process is dead, release it - nobody will wait for it */
668 if (autoreap)
669 release_task(tsk);
670}
671
672#ifdef CONFIG_DEBUG_STACK_USAGE
673static void check_stack_usage(void)
674{
675 static DEFINE_SPINLOCK(low_water_lock);
676 static int lowest_to_date = THREAD_SIZE;
677 unsigned long free;
678
679 free = stack_not_used(current);
680
681 if (free >= lowest_to_date)
682 return;
683
684 spin_lock(&low_water_lock);
685 if (free < lowest_to_date) {
686 printk(KERN_WARNING "%s (%d) used greatest stack depth: "
687 "%lu bytes left\n",
688 current->comm, task_pid_nr(current), free);
689 lowest_to_date = free;
690 }
691 spin_unlock(&low_water_lock);
692}
693#else
694static inline void check_stack_usage(void) {}
695#endif
696
697void do_exit(long code)
698{
699 struct task_struct *tsk = current;
700 int group_dead;
701
702 profile_task_exit(tsk);
703
704 WARN_ON(blk_needs_flush_plug(tsk));
705
706 if (unlikely(in_interrupt()))
707 panic("Aiee, killing interrupt handler!");
708 if (unlikely(!tsk->pid))
709 panic("Attempted to kill the idle task!");
710
711 /*
712 * If do_exit is called because this processes oopsed, it's possible
713 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
714 * continuing. Amongst other possible reasons, this is to prevent
715 * mm_release()->clear_child_tid() from writing to a user-controlled
716 * kernel address.
717 */
718 set_fs(USER_DS);
719
720 ptrace_event(PTRACE_EVENT_EXIT, code);
721
722 validate_creds_for_do_exit(tsk);
723
724 /*
725 * We're taking recursive faults here in do_exit. Safest is to just
726 * leave this task alone and wait for reboot.
727 */
728 if (unlikely(tsk->flags & PF_EXITING)) {
729 printk(KERN_ALERT
730 "Fixing recursive fault but reboot is needed!\n");
731 /*
732 * We can do this unlocked here. The futex code uses
733 * this flag just to verify whether the pi state
734 * cleanup has been done or not. In the worst case it
735 * loops once more. We pretend that the cleanup was
736 * done as there is no way to return. Either the
737 * OWNER_DIED bit is set by now or we push the blocked
738 * task into the wait for ever nirwana as well.
739 */
740 tsk->flags |= PF_EXITPIDONE;
741 set_current_state(TASK_UNINTERRUPTIBLE);
742 schedule();
743 }
744
745 exit_signals(tsk); /* sets PF_EXITING */
746 /*
747 * tsk->flags are checked in the futex code to protect against
748 * an exiting task cleaning up the robust pi futexes.
749 */
750 smp_mb();
751 raw_spin_unlock_wait(&tsk->pi_lock);
752
753 if (unlikely(in_atomic()))
754 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
755 current->comm, task_pid_nr(current),
756 preempt_count());
757
758 acct_update_integrals(tsk);
759 /* sync mm's RSS info before statistics gathering */
760 if (tsk->mm)
761 sync_mm_rss(tsk->mm);
762 group_dead = atomic_dec_and_test(&tsk->signal->live);
763 if (group_dead) {
764 hrtimer_cancel(&tsk->signal->real_timer);
765 exit_itimers(tsk->signal);
766 if (tsk->mm)
767 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
768 }
769 acct_collect(code, group_dead);
770 if (group_dead)
771 tty_audit_exit();
772 audit_free(tsk);
773
774 tsk->exit_code = code;
775 taskstats_exit(tsk, group_dead);
776
777 exit_mm(tsk);
778
779 if (group_dead)
780 acct_process();
781 trace_sched_process_exit(tsk);
782
783 exit_sem(tsk);
784 exit_shm(tsk);
785 exit_files(tsk);
786 exit_fs(tsk);
787 if (group_dead)
788 disassociate_ctty(1);
789 exit_task_namespaces(tsk);
790 exit_task_work(tsk);
791 exit_thread();
792
793 /*
794 * Flush inherited counters to the parent - before the parent
795 * gets woken up by child-exit notifications.
796 *
797 * because of cgroup mode, must be called before cgroup_exit()
798 */
799 perf_event_exit_task(tsk);
800
801 cgroup_exit(tsk);
802
803 module_put(task_thread_info(tsk)->exec_domain->module);
804
805 /*
806 * FIXME: do that only when needed, using sched_exit tracepoint
807 */
808 flush_ptrace_hw_breakpoint(tsk);
809
810 exit_notify(tsk, group_dead);
811 proc_exit_connector(tsk);
812#ifdef CONFIG_NUMA
813 task_lock(tsk);
814 mpol_put(tsk->mempolicy);
815 tsk->mempolicy = NULL;
816 task_unlock(tsk);
817#endif
818#ifdef CONFIG_FUTEX
819 if (unlikely(current->pi_state_cache))
820 kfree(current->pi_state_cache);
821#endif
822 /*
823 * Make sure we are holding no locks:
824 */
825 debug_check_no_locks_held();
826 /*
827 * We can do this unlocked here. The futex code uses this flag
828 * just to verify whether the pi state cleanup has been done
829 * or not. In the worst case it loops once more.
830 */
831 tsk->flags |= PF_EXITPIDONE;
832
833 if (tsk->io_context)
834 exit_io_context(tsk);
835
836 if (tsk->splice_pipe)
837 free_pipe_info(tsk->splice_pipe);
838
839 if (tsk->task_frag.page)
840 put_page(tsk->task_frag.page);
841
842 validate_creds_for_do_exit(tsk);
843
844 check_stack_usage();
845 preempt_disable();
846 if (tsk->nr_dirtied)
847 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
848 exit_rcu();
849
850 /*
851 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
852 * when the following two conditions become true.
853 * - There is race condition of mmap_sem (It is acquired by
854 * exit_mm()), and
855 * - SMI occurs before setting TASK_RUNINNG.
856 * (or hypervisor of virtual machine switches to other guest)
857 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
858 *
859 * To avoid it, we have to wait for releasing tsk->pi_lock which
860 * is held by try_to_wake_up()
861 */
862 smp_mb();
863 raw_spin_unlock_wait(&tsk->pi_lock);
864
865 /* causes final put_task_struct in finish_task_switch(). */
866 tsk->state = TASK_DEAD;
867 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
868 schedule();
869 BUG();
870 /* Avoid "noreturn function does return". */
871 for (;;)
872 cpu_relax(); /* For when BUG is null */
873}
874
875EXPORT_SYMBOL_GPL(do_exit);
876
877void complete_and_exit(struct completion *comp, long code)
878{
879 if (comp)
880 complete(comp);
881
882 do_exit(code);
883}
884
885EXPORT_SYMBOL(complete_and_exit);
886
887SYSCALL_DEFINE1(exit, int, error_code)
888{
889 do_exit((error_code&0xff)<<8);
890}
891
892/*
893 * Take down every thread in the group. This is called by fatal signals
894 * as well as by sys_exit_group (below).
895 */
896void
897do_group_exit(int exit_code)
898{
899 struct signal_struct *sig = current->signal;
900
901 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
902
903 if (signal_group_exit(sig))
904 exit_code = sig->group_exit_code;
905 else if (!thread_group_empty(current)) {
906 struct sighand_struct *const sighand = current->sighand;
907 spin_lock_irq(&sighand->siglock);
908 if (signal_group_exit(sig))
909 /* Another thread got here before we took the lock. */
910 exit_code = sig->group_exit_code;
911 else {
912 sig->group_exit_code = exit_code;
913 sig->flags = SIGNAL_GROUP_EXIT;
914 zap_other_threads(current);
915 }
916 spin_unlock_irq(&sighand->siglock);
917 }
918
919 do_exit(exit_code);
920 /* NOTREACHED */
921}
922
923/*
924 * this kills every thread in the thread group. Note that any externally
925 * wait4()-ing process will get the correct exit code - even if this
926 * thread is not the thread group leader.
927 */
928SYSCALL_DEFINE1(exit_group, int, error_code)
929{
930 do_group_exit((error_code & 0xff) << 8);
931 /* NOTREACHED */
932 return 0;
933}
934
935struct wait_opts {
936 enum pid_type wo_type;
937 int wo_flags;
938 struct pid *wo_pid;
939
940 struct siginfo __user *wo_info;
941 int __user *wo_stat;
942 struct rusage __user *wo_rusage;
943
944 wait_queue_t child_wait;
945 int notask_error;
946};
947
948static inline
949struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
950{
951 if (type != PIDTYPE_PID)
952 task = task->group_leader;
953 return task->pids[type].pid;
954}
955
956static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
957{
958 return wo->wo_type == PIDTYPE_MAX ||
959 task_pid_type(p, wo->wo_type) == wo->wo_pid;
960}
961
962static int eligible_child(struct wait_opts *wo, struct task_struct *p)
963{
964 if (!eligible_pid(wo, p))
965 return 0;
966 /* Wait for all children (clone and not) if __WALL is set;
967 * otherwise, wait for clone children *only* if __WCLONE is
968 * set; otherwise, wait for non-clone children *only*. (Note:
969 * A "clone" child here is one that reports to its parent
970 * using a signal other than SIGCHLD.) */
971 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
972 && !(wo->wo_flags & __WALL))
973 return 0;
974
975 return 1;
976}
977
978static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
979 pid_t pid, uid_t uid, int why, int status)
980{
981 struct siginfo __user *infop;
982 int retval = wo->wo_rusage
983 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
984
985 put_task_struct(p);
986 infop = wo->wo_info;
987 if (infop) {
988 if (!retval)
989 retval = put_user(SIGCHLD, &infop->si_signo);
990 if (!retval)
991 retval = put_user(0, &infop->si_errno);
992 if (!retval)
993 retval = put_user((short)why, &infop->si_code);
994 if (!retval)
995 retval = put_user(pid, &infop->si_pid);
996 if (!retval)
997 retval = put_user(uid, &infop->si_uid);
998 if (!retval)
999 retval = put_user(status, &infop->si_status);
1000 }
1001 if (!retval)
1002 retval = pid;
1003 return retval;
1004}
1005
1006/*
1007 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1008 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1009 * the lock and this task is uninteresting. If we return nonzero, we have
1010 * released the lock and the system call should return.
1011 */
1012static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1013{
1014 unsigned long state;
1015 int retval, status, traced;
1016 pid_t pid = task_pid_vnr(p);
1017 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1018 struct siginfo __user *infop;
1019
1020 if (!likely(wo->wo_flags & WEXITED))
1021 return 0;
1022
1023 if (unlikely(wo->wo_flags & WNOWAIT)) {
1024 int exit_code = p->exit_code;
1025 int why;
1026
1027 get_task_struct(p);
1028 read_unlock(&tasklist_lock);
1029 if ((exit_code & 0x7f) == 0) {
1030 why = CLD_EXITED;
1031 status = exit_code >> 8;
1032 } else {
1033 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1034 status = exit_code & 0x7f;
1035 }
1036 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1037 }
1038
1039 traced = ptrace_reparented(p);
1040 /*
1041 * Move the task's state to DEAD/TRACE, only one thread can do this.
1042 */
1043 state = traced && thread_group_leader(p) ? EXIT_TRACE : EXIT_DEAD;
1044 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1045 return 0;
1046 /*
1047 * It can be ptraced but not reparented, check
1048 * thread_group_leader() to filter out sub-threads.
1049 */
1050 if (likely(!traced) && thread_group_leader(p)) {
1051 struct signal_struct *psig;
1052 struct signal_struct *sig;
1053 unsigned long maxrss;
1054 cputime_t tgutime, tgstime;
1055
1056 /*
1057 * The resource counters for the group leader are in its
1058 * own task_struct. Those for dead threads in the group
1059 * are in its signal_struct, as are those for the child
1060 * processes it has previously reaped. All these
1061 * accumulate in the parent's signal_struct c* fields.
1062 *
1063 * We don't bother to take a lock here to protect these
1064 * p->signal fields, because they are only touched by
1065 * __exit_signal, which runs with tasklist_lock
1066 * write-locked anyway, and so is excluded here. We do
1067 * need to protect the access to parent->signal fields,
1068 * as other threads in the parent group can be right
1069 * here reaping other children at the same time.
1070 *
1071 * We use thread_group_cputime_adjusted() to get times for the thread
1072 * group, which consolidates times for all threads in the
1073 * group including the group leader.
1074 */
1075 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1076 spin_lock_irq(&p->real_parent->sighand->siglock);
1077 psig = p->real_parent->signal;
1078 sig = p->signal;
1079 psig->cutime += tgutime + sig->cutime;
1080 psig->cstime += tgstime + sig->cstime;
1081 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1082 psig->cmin_flt +=
1083 p->min_flt + sig->min_flt + sig->cmin_flt;
1084 psig->cmaj_flt +=
1085 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1086 psig->cnvcsw +=
1087 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1088 psig->cnivcsw +=
1089 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1090 psig->cinblock +=
1091 task_io_get_inblock(p) +
1092 sig->inblock + sig->cinblock;
1093 psig->coublock +=
1094 task_io_get_oublock(p) +
1095 sig->oublock + sig->coublock;
1096 maxrss = max(sig->maxrss, sig->cmaxrss);
1097 if (psig->cmaxrss < maxrss)
1098 psig->cmaxrss = maxrss;
1099 task_io_accounting_add(&psig->ioac, &p->ioac);
1100 task_io_accounting_add(&psig->ioac, &sig->ioac);
1101 spin_unlock_irq(&p->real_parent->sighand->siglock);
1102 }
1103
1104 /*
1105 * Now we are sure this task is interesting, and no other
1106 * thread can reap it because we its state == DEAD/TRACE.
1107 */
1108 read_unlock(&tasklist_lock);
1109
1110 retval = wo->wo_rusage
1111 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1112 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1113 ? p->signal->group_exit_code : p->exit_code;
1114 if (!retval && wo->wo_stat)
1115 retval = put_user(status, wo->wo_stat);
1116
1117 infop = wo->wo_info;
1118 if (!retval && infop)
1119 retval = put_user(SIGCHLD, &infop->si_signo);
1120 if (!retval && infop)
1121 retval = put_user(0, &infop->si_errno);
1122 if (!retval && infop) {
1123 int why;
1124
1125 if ((status & 0x7f) == 0) {
1126 why = CLD_EXITED;
1127 status >>= 8;
1128 } else {
1129 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1130 status &= 0x7f;
1131 }
1132 retval = put_user((short)why, &infop->si_code);
1133 if (!retval)
1134 retval = put_user(status, &infop->si_status);
1135 }
1136 if (!retval && infop)
1137 retval = put_user(pid, &infop->si_pid);
1138 if (!retval && infop)
1139 retval = put_user(uid, &infop->si_uid);
1140 if (!retval)
1141 retval = pid;
1142
1143 if (state == EXIT_TRACE) {
1144 write_lock_irq(&tasklist_lock);
1145 /* We dropped tasklist, ptracer could die and untrace */
1146 ptrace_unlink(p);
1147
1148 /* If parent wants a zombie, don't release it now */
1149 state = EXIT_ZOMBIE;
1150 if (do_notify_parent(p, p->exit_signal))
1151 state = EXIT_DEAD;
1152 p->exit_state = state;
1153 write_unlock_irq(&tasklist_lock);
1154 }
1155 if (state == EXIT_DEAD)
1156 release_task(p);
1157
1158 return retval;
1159}
1160
1161static int *task_stopped_code(struct task_struct *p, bool ptrace)
1162{
1163 if (ptrace) {
1164 if (task_is_stopped_or_traced(p) &&
1165 !(p->jobctl & JOBCTL_LISTENING))
1166 return &p->exit_code;
1167 } else {
1168 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1169 return &p->signal->group_exit_code;
1170 }
1171 return NULL;
1172}
1173
1174/**
1175 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1176 * @wo: wait options
1177 * @ptrace: is the wait for ptrace
1178 * @p: task to wait for
1179 *
1180 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1181 *
1182 * CONTEXT:
1183 * read_lock(&tasklist_lock), which is released if return value is
1184 * non-zero. Also, grabs and releases @p->sighand->siglock.
1185 *
1186 * RETURNS:
1187 * 0 if wait condition didn't exist and search for other wait conditions
1188 * should continue. Non-zero return, -errno on failure and @p's pid on
1189 * success, implies that tasklist_lock is released and wait condition
1190 * search should terminate.
1191 */
1192static int wait_task_stopped(struct wait_opts *wo,
1193 int ptrace, struct task_struct *p)
1194{
1195 struct siginfo __user *infop;
1196 int retval, exit_code, *p_code, why;
1197 uid_t uid = 0; /* unneeded, required by compiler */
1198 pid_t pid;
1199
1200 /*
1201 * Traditionally we see ptrace'd stopped tasks regardless of options.
1202 */
1203 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1204 return 0;
1205
1206 if (!task_stopped_code(p, ptrace))
1207 return 0;
1208
1209 exit_code = 0;
1210 spin_lock_irq(&p->sighand->siglock);
1211
1212 p_code = task_stopped_code(p, ptrace);
1213 if (unlikely(!p_code))
1214 goto unlock_sig;
1215
1216 exit_code = *p_code;
1217 if (!exit_code)
1218 goto unlock_sig;
1219
1220 if (!unlikely(wo->wo_flags & WNOWAIT))
1221 *p_code = 0;
1222
1223 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1224unlock_sig:
1225 spin_unlock_irq(&p->sighand->siglock);
1226 if (!exit_code)
1227 return 0;
1228
1229 /*
1230 * Now we are pretty sure this task is interesting.
1231 * Make sure it doesn't get reaped out from under us while we
1232 * give up the lock and then examine it below. We don't want to
1233 * keep holding onto the tasklist_lock while we call getrusage and
1234 * possibly take page faults for user memory.
1235 */
1236 get_task_struct(p);
1237 pid = task_pid_vnr(p);
1238 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1239 read_unlock(&tasklist_lock);
1240
1241 if (unlikely(wo->wo_flags & WNOWAIT))
1242 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1243
1244 retval = wo->wo_rusage
1245 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1246 if (!retval && wo->wo_stat)
1247 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1248
1249 infop = wo->wo_info;
1250 if (!retval && infop)
1251 retval = put_user(SIGCHLD, &infop->si_signo);
1252 if (!retval && infop)
1253 retval = put_user(0, &infop->si_errno);
1254 if (!retval && infop)
1255 retval = put_user((short)why, &infop->si_code);
1256 if (!retval && infop)
1257 retval = put_user(exit_code, &infop->si_status);
1258 if (!retval && infop)
1259 retval = put_user(pid, &infop->si_pid);
1260 if (!retval && infop)
1261 retval = put_user(uid, &infop->si_uid);
1262 if (!retval)
1263 retval = pid;
1264 put_task_struct(p);
1265
1266 BUG_ON(!retval);
1267 return retval;
1268}
1269
1270/*
1271 * Handle do_wait work for one task in a live, non-stopped state.
1272 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1273 * the lock and this task is uninteresting. If we return nonzero, we have
1274 * released the lock and the system call should return.
1275 */
1276static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1277{
1278 int retval;
1279 pid_t pid;
1280 uid_t uid;
1281
1282 if (!unlikely(wo->wo_flags & WCONTINUED))
1283 return 0;
1284
1285 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1286 return 0;
1287
1288 spin_lock_irq(&p->sighand->siglock);
1289 /* Re-check with the lock held. */
1290 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1291 spin_unlock_irq(&p->sighand->siglock);
1292 return 0;
1293 }
1294 if (!unlikely(wo->wo_flags & WNOWAIT))
1295 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1296 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1297 spin_unlock_irq(&p->sighand->siglock);
1298
1299 pid = task_pid_vnr(p);
1300 get_task_struct(p);
1301 read_unlock(&tasklist_lock);
1302
1303 if (!wo->wo_info) {
1304 retval = wo->wo_rusage
1305 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1306 put_task_struct(p);
1307 if (!retval && wo->wo_stat)
1308 retval = put_user(0xffff, wo->wo_stat);
1309 if (!retval)
1310 retval = pid;
1311 } else {
1312 retval = wait_noreap_copyout(wo, p, pid, uid,
1313 CLD_CONTINUED, SIGCONT);
1314 BUG_ON(retval == 0);
1315 }
1316
1317 return retval;
1318}
1319
1320/*
1321 * Consider @p for a wait by @parent.
1322 *
1323 * -ECHILD should be in ->notask_error before the first call.
1324 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1325 * Returns zero if the search for a child should continue;
1326 * then ->notask_error is 0 if @p is an eligible child,
1327 * or another error from security_task_wait(), or still -ECHILD.
1328 */
1329static int wait_consider_task(struct wait_opts *wo, int ptrace,
1330 struct task_struct *p)
1331{
1332 int ret;
1333
1334 if (unlikely(p->exit_state == EXIT_DEAD))
1335 return 0;
1336
1337 ret = eligible_child(wo, p);
1338 if (!ret)
1339 return ret;
1340
1341 ret = security_task_wait(p);
1342 if (unlikely(ret < 0)) {
1343 /*
1344 * If we have not yet seen any eligible child,
1345 * then let this error code replace -ECHILD.
1346 * A permission error will give the user a clue
1347 * to look for security policy problems, rather
1348 * than for mysterious wait bugs.
1349 */
1350 if (wo->notask_error)
1351 wo->notask_error = ret;
1352 return 0;
1353 }
1354
1355 if (unlikely(p->exit_state == EXIT_TRACE)) {
1356 /*
1357 * ptrace == 0 means we are the natural parent. In this case
1358 * we should clear notask_error, debugger will notify us.
1359 */
1360 if (likely(!ptrace))
1361 wo->notask_error = 0;
1362 return 0;
1363 }
1364
1365 if (likely(!ptrace) && unlikely(p->ptrace)) {
1366 /*
1367 * If it is traced by its real parent's group, just pretend
1368 * the caller is ptrace_do_wait() and reap this child if it
1369 * is zombie.
1370 *
1371 * This also hides group stop state from real parent; otherwise
1372 * a single stop can be reported twice as group and ptrace stop.
1373 * If a ptracer wants to distinguish these two events for its
1374 * own children it should create a separate process which takes
1375 * the role of real parent.
1376 */
1377 if (!ptrace_reparented(p))
1378 ptrace = 1;
1379 }
1380
1381 /* slay zombie? */
1382 if (p->exit_state == EXIT_ZOMBIE) {
1383 /* we don't reap group leaders with subthreads */
1384 if (!delay_group_leader(p)) {
1385 /*
1386 * A zombie ptracee is only visible to its ptracer.
1387 * Notification and reaping will be cascaded to the
1388 * real parent when the ptracer detaches.
1389 */
1390 if (unlikely(ptrace) || likely(!p->ptrace))
1391 return wait_task_zombie(wo, p);
1392 }
1393
1394 /*
1395 * Allow access to stopped/continued state via zombie by
1396 * falling through. Clearing of notask_error is complex.
1397 *
1398 * When !@ptrace:
1399 *
1400 * If WEXITED is set, notask_error should naturally be
1401 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1402 * so, if there are live subthreads, there are events to
1403 * wait for. If all subthreads are dead, it's still safe
1404 * to clear - this function will be called again in finite
1405 * amount time once all the subthreads are released and
1406 * will then return without clearing.
1407 *
1408 * When @ptrace:
1409 *
1410 * Stopped state is per-task and thus can't change once the
1411 * target task dies. Only continued and exited can happen.
1412 * Clear notask_error if WCONTINUED | WEXITED.
1413 */
1414 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1415 wo->notask_error = 0;
1416 } else {
1417 /*
1418 * @p is alive and it's gonna stop, continue or exit, so
1419 * there always is something to wait for.
1420 */
1421 wo->notask_error = 0;
1422 }
1423
1424 /*
1425 * Wait for stopped. Depending on @ptrace, different stopped state
1426 * is used and the two don't interact with each other.
1427 */
1428 ret = wait_task_stopped(wo, ptrace, p);
1429 if (ret)
1430 return ret;
1431
1432 /*
1433 * Wait for continued. There's only one continued state and the
1434 * ptracer can consume it which can confuse the real parent. Don't
1435 * use WCONTINUED from ptracer. You don't need or want it.
1436 */
1437 return wait_task_continued(wo, p);
1438}
1439
1440/*
1441 * Do the work of do_wait() for one thread in the group, @tsk.
1442 *
1443 * -ECHILD should be in ->notask_error before the first call.
1444 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1445 * Returns zero if the search for a child should continue; then
1446 * ->notask_error is 0 if there were any eligible children,
1447 * or another error from security_task_wait(), or still -ECHILD.
1448 */
1449static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1450{
1451 struct task_struct *p;
1452
1453 list_for_each_entry(p, &tsk->children, sibling) {
1454 int ret = wait_consider_task(wo, 0, p);
1455 if (ret)
1456 return ret;
1457 }
1458
1459 return 0;
1460}
1461
1462static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1463{
1464 struct task_struct *p;
1465
1466 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1467 int ret = wait_consider_task(wo, 1, p);
1468 if (ret)
1469 return ret;
1470 }
1471
1472 return 0;
1473}
1474
1475static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1476 int sync, void *key)
1477{
1478 struct wait_opts *wo = container_of(wait, struct wait_opts,
1479 child_wait);
1480 struct task_struct *p = key;
1481
1482 if (!eligible_pid(wo, p))
1483 return 0;
1484
1485 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1486 return 0;
1487
1488 return default_wake_function(wait, mode, sync, key);
1489}
1490
1491void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1492{
1493 __wake_up_sync_key(&parent->signal->wait_chldexit,
1494 TASK_INTERRUPTIBLE, 1, p);
1495}
1496
1497static long do_wait(struct wait_opts *wo)
1498{
1499 struct task_struct *tsk;
1500 int retval;
1501
1502 trace_sched_process_wait(wo->wo_pid);
1503
1504 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1505 wo->child_wait.private = current;
1506 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1507repeat:
1508 /*
1509 * If there is nothing that can match our critiera just get out.
1510 * We will clear ->notask_error to zero if we see any child that
1511 * might later match our criteria, even if we are not able to reap
1512 * it yet.
1513 */
1514 wo->notask_error = -ECHILD;
1515 if ((wo->wo_type < PIDTYPE_MAX) &&
1516 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1517 goto notask;
1518
1519 set_current_state(TASK_INTERRUPTIBLE);
1520 read_lock(&tasklist_lock);
1521 tsk = current;
1522 do {
1523 retval = do_wait_thread(wo, tsk);
1524 if (retval)
1525 goto end;
1526
1527 retval = ptrace_do_wait(wo, tsk);
1528 if (retval)
1529 goto end;
1530
1531 if (wo->wo_flags & __WNOTHREAD)
1532 break;
1533 } while_each_thread(current, tsk);
1534 read_unlock(&tasklist_lock);
1535
1536notask:
1537 retval = wo->notask_error;
1538 if (!retval && !(wo->wo_flags & WNOHANG)) {
1539 retval = -ERESTARTSYS;
1540 if (!signal_pending(current)) {
1541 schedule();
1542 goto repeat;
1543 }
1544 }
1545end:
1546 __set_current_state(TASK_RUNNING);
1547 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1548 return retval;
1549}
1550
1551SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1552 infop, int, options, struct rusage __user *, ru)
1553{
1554 struct wait_opts wo;
1555 struct pid *pid = NULL;
1556 enum pid_type type;
1557 long ret;
1558
1559 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
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_stat = NULL;
1590 wo.wo_rusage = ru;
1591 ret = do_wait(&wo);
1592
1593 if (ret > 0) {
1594 ret = 0;
1595 } else if (infop) {
1596 /*
1597 * For a WNOHANG return, clear out all the fields
1598 * we would set so the user can easily tell the
1599 * difference.
1600 */
1601 if (!ret)
1602 ret = put_user(0, &infop->si_signo);
1603 if (!ret)
1604 ret = put_user(0, &infop->si_errno);
1605 if (!ret)
1606 ret = put_user(0, &infop->si_code);
1607 if (!ret)
1608 ret = put_user(0, &infop->si_pid);
1609 if (!ret)
1610 ret = put_user(0, &infop->si_uid);
1611 if (!ret)
1612 ret = put_user(0, &infop->si_status);
1613 }
1614
1615 put_pid(pid);
1616 return ret;
1617}
1618
1619SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1620 int, options, struct rusage __user *, ru)
1621{
1622 struct wait_opts wo;
1623 struct pid *pid = NULL;
1624 enum pid_type type;
1625 long ret;
1626
1627 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1628 __WNOTHREAD|__WCLONE|__WALL))
1629 return -EINVAL;
1630
1631 if (upid == -1)
1632 type = PIDTYPE_MAX;
1633 else if (upid < 0) {
1634 type = PIDTYPE_PGID;
1635 pid = find_get_pid(-upid);
1636 } else if (upid == 0) {
1637 type = PIDTYPE_PGID;
1638 pid = get_task_pid(current, PIDTYPE_PGID);
1639 } else /* upid > 0 */ {
1640 type = PIDTYPE_PID;
1641 pid = find_get_pid(upid);
1642 }
1643
1644 wo.wo_type = type;
1645 wo.wo_pid = pid;
1646 wo.wo_flags = options | WEXITED;
1647 wo.wo_info = NULL;
1648 wo.wo_stat = stat_addr;
1649 wo.wo_rusage = ru;
1650 ret = do_wait(&wo);
1651 put_pid(pid);
1652
1653 return ret;
1654}
1655
1656#ifdef __ARCH_WANT_SYS_WAITPID
1657
1658/*
1659 * sys_waitpid() remains for compatibility. waitpid() should be
1660 * implemented by calling sys_wait4() from libc.a.
1661 */
1662SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1663{
1664 return sys_wait4(pid, stat_addr, options, NULL);
1665}
1666
1667#endif
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/exit.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/mm.h>
9#include <linux/slab.h>
10#include <linux/sched/autogroup.h>
11#include <linux/sched/mm.h>
12#include <linux/sched/stat.h>
13#include <linux/sched/task.h>
14#include <linux/sched/task_stack.h>
15#include <linux/sched/cputime.h>
16#include <linux/interrupt.h>
17#include <linux/module.h>
18#include <linux/capability.h>
19#include <linux/completion.h>
20#include <linux/personality.h>
21#include <linux/tty.h>
22#include <linux/iocontext.h>
23#include <linux/key.h>
24#include <linux/cpu.h>
25#include <linux/acct.h>
26#include <linux/tsacct_kern.h>
27#include <linux/file.h>
28#include <linux/fdtable.h>
29#include <linux/freezer.h>
30#include <linux/binfmts.h>
31#include <linux/nsproxy.h>
32#include <linux/pid_namespace.h>
33#include <linux/ptrace.h>
34#include <linux/profile.h>
35#include <linux/mount.h>
36#include <linux/proc_fs.h>
37#include <linux/kthread.h>
38#include <linux/mempolicy.h>
39#include <linux/taskstats_kern.h>
40#include <linux/delayacct.h>
41#include <linux/cgroup.h>
42#include <linux/syscalls.h>
43#include <linux/signal.h>
44#include <linux/posix-timers.h>
45#include <linux/cn_proc.h>
46#include <linux/mutex.h>
47#include <linux/futex.h>
48#include <linux/pipe_fs_i.h>
49#include <linux/audit.h> /* for audit_free() */
50#include <linux/resource.h>
51#include <linux/blkdev.h>
52#include <linux/task_io_accounting_ops.h>
53#include <linux/tracehook.h>
54#include <linux/fs_struct.h>
55#include <linux/init_task.h>
56#include <linux/perf_event.h>
57#include <trace/events/sched.h>
58#include <linux/hw_breakpoint.h>
59#include <linux/oom.h>
60#include <linux/writeback.h>
61#include <linux/shm.h>
62#include <linux/kcov.h>
63#include <linux/random.h>
64#include <linux/rcuwait.h>
65#include <linux/compat.h>
66
67#include <linux/uaccess.h>
68#include <asm/unistd.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_TGID);
77 detach_pid(p, PIDTYPE_PGID);
78 detach_pid(p, PIDTYPE_SID);
79
80 list_del_rcu(&p->tasks);
81 list_del_init(&p->sibling);
82 __this_cpu_dec(process_counts);
83 }
84 list_del_rcu(&p->thread_group);
85 list_del_rcu(&p->thread_node);
86}
87
88/*
89 * This function expects the tasklist_lock write-locked.
90 */
91static void __exit_signal(struct task_struct *tsk)
92{
93 struct signal_struct *sig = tsk->signal;
94 bool group_dead = thread_group_leader(tsk);
95 struct sighand_struct *sighand;
96 struct tty_struct *tty;
97 u64 utime, stime;
98
99 sighand = rcu_dereference_check(tsk->sighand,
100 lockdep_tasklist_lock_is_held());
101 spin_lock(&sighand->siglock);
102
103#ifdef CONFIG_POSIX_TIMERS
104 posix_cpu_timers_exit(tsk);
105 if (group_dead)
106 posix_cpu_timers_exit_group(tsk);
107#endif
108
109 if (group_dead) {
110 tty = sig->tty;
111 sig->tty = NULL;
112 } else {
113 /*
114 * If there is any task waiting for the group exit
115 * then notify it:
116 */
117 if (sig->notify_count > 0 && !--sig->notify_count)
118 wake_up_process(sig->group_exit_task);
119
120 if (tsk == sig->curr_target)
121 sig->curr_target = next_thread(tsk);
122 }
123
124 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
125 sizeof(unsigned long long));
126
127 /*
128 * Accumulate here the counters for all threads as they die. We could
129 * skip the group leader because it is the last user of signal_struct,
130 * but we want to avoid the race with thread_group_cputime() which can
131 * see the empty ->thread_head list.
132 */
133 task_cputime(tsk, &utime, &stime);
134 write_seqlock(&sig->stats_lock);
135 sig->utime += utime;
136 sig->stime += stime;
137 sig->gtime += task_gtime(tsk);
138 sig->min_flt += tsk->min_flt;
139 sig->maj_flt += tsk->maj_flt;
140 sig->nvcsw += tsk->nvcsw;
141 sig->nivcsw += tsk->nivcsw;
142 sig->inblock += task_io_get_inblock(tsk);
143 sig->oublock += task_io_get_oublock(tsk);
144 task_io_accounting_add(&sig->ioac, &tsk->ioac);
145 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
146 sig->nr_threads--;
147 __unhash_process(tsk, group_dead);
148 write_sequnlock(&sig->stats_lock);
149
150 /*
151 * Do this under ->siglock, we can race with another thread
152 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
153 */
154 flush_sigqueue(&tsk->pending);
155 tsk->sighand = NULL;
156 spin_unlock(&sighand->siglock);
157
158 __cleanup_sighand(sighand);
159 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
160 if (group_dead) {
161 flush_sigqueue(&sig->shared_pending);
162 tty_kref_put(tty);
163 }
164}
165
166static void delayed_put_task_struct(struct rcu_head *rhp)
167{
168 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
169
170 perf_event_delayed_put(tsk);
171 trace_sched_process_free(tsk);
172 put_task_struct(tsk);
173}
174
175void put_task_struct_rcu_user(struct task_struct *task)
176{
177 if (refcount_dec_and_test(&task->rcu_users))
178 call_rcu(&task->rcu, delayed_put_task_struct);
179}
180
181void release_task(struct task_struct *p)
182{
183 struct task_struct *leader;
184 struct pid *thread_pid;
185 int zap_leader;
186repeat:
187 /* don't need to get the RCU readlock here - the process is dead and
188 * can't be modifying its own credentials. But shut RCU-lockdep up */
189 rcu_read_lock();
190 atomic_dec(&__task_cred(p)->user->processes);
191 rcu_read_unlock();
192
193 cgroup_release(p);
194
195 write_lock_irq(&tasklist_lock);
196 ptrace_release_task(p);
197 thread_pid = get_pid(p->thread_pid);
198 __exit_signal(p);
199
200 /*
201 * If we are the last non-leader member of the thread
202 * group, and the leader is zombie, then notify the
203 * group leader's parent process. (if it wants notification.)
204 */
205 zap_leader = 0;
206 leader = p->group_leader;
207 if (leader != p && thread_group_empty(leader)
208 && leader->exit_state == EXIT_ZOMBIE) {
209 /*
210 * If we were the last child thread and the leader has
211 * exited already, and the leader's parent ignores SIGCHLD,
212 * then we are the one who should release the leader.
213 */
214 zap_leader = do_notify_parent(leader, leader->exit_signal);
215 if (zap_leader)
216 leader->exit_state = EXIT_DEAD;
217 }
218
219 write_unlock_irq(&tasklist_lock);
220 seccomp_filter_release(p);
221 proc_flush_pid(thread_pid);
222 put_pid(thread_pid);
223 release_thread(p);
224 put_task_struct_rcu_user(p);
225
226 p = leader;
227 if (unlikely(zap_leader))
228 goto repeat;
229}
230
231int rcuwait_wake_up(struct rcuwait *w)
232{
233 int ret = 0;
234 struct task_struct *task;
235
236 rcu_read_lock();
237
238 /*
239 * Order condition vs @task, such that everything prior to the load
240 * of @task is visible. This is the condition as to why the user called
241 * rcuwait_wake() in the first place. Pairs with set_current_state()
242 * barrier (A) in rcuwait_wait_event().
243 *
244 * WAIT WAKE
245 * [S] tsk = current [S] cond = true
246 * MB (A) MB (B)
247 * [L] cond [L] tsk
248 */
249 smp_mb(); /* (B) */
250
251 task = rcu_dereference(w->task);
252 if (task)
253 ret = wake_up_process(task);
254 rcu_read_unlock();
255
256 return ret;
257}
258EXPORT_SYMBOL_GPL(rcuwait_wake_up);
259
260/*
261 * Determine if a process group is "orphaned", according to the POSIX
262 * definition in 2.2.2.52. Orphaned process groups are not to be affected
263 * by terminal-generated stop signals. Newly orphaned process groups are
264 * to receive a SIGHUP and a SIGCONT.
265 *
266 * "I ask you, have you ever known what it is to be an orphan?"
267 */
268static int will_become_orphaned_pgrp(struct pid *pgrp,
269 struct task_struct *ignored_task)
270{
271 struct task_struct *p;
272
273 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
274 if ((p == ignored_task) ||
275 (p->exit_state && thread_group_empty(p)) ||
276 is_global_init(p->real_parent))
277 continue;
278
279 if (task_pgrp(p->real_parent) != pgrp &&
280 task_session(p->real_parent) == task_session(p))
281 return 0;
282 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
283
284 return 1;
285}
286
287int is_current_pgrp_orphaned(void)
288{
289 int retval;
290
291 read_lock(&tasklist_lock);
292 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
293 read_unlock(&tasklist_lock);
294
295 return retval;
296}
297
298static bool has_stopped_jobs(struct pid *pgrp)
299{
300 struct task_struct *p;
301
302 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303 if (p->signal->flags & SIGNAL_STOP_STOPPED)
304 return true;
305 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
306
307 return false;
308}
309
310/*
311 * Check to see if any process groups have become orphaned as
312 * a result of our exiting, and if they have any stopped jobs,
313 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
314 */
315static void
316kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
317{
318 struct pid *pgrp = task_pgrp(tsk);
319 struct task_struct *ignored_task = tsk;
320
321 if (!parent)
322 /* exit: our father is in a different pgrp than
323 * we are and we were the only connection outside.
324 */
325 parent = tsk->real_parent;
326 else
327 /* reparent: our child is in a different pgrp than
328 * we are, and it was the only connection outside.
329 */
330 ignored_task = NULL;
331
332 if (task_pgrp(parent) != pgrp &&
333 task_session(parent) == task_session(tsk) &&
334 will_become_orphaned_pgrp(pgrp, ignored_task) &&
335 has_stopped_jobs(pgrp)) {
336 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
337 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
338 }
339}
340
341#ifdef CONFIG_MEMCG
342/*
343 * A task is exiting. If it owned this mm, find a new owner for the mm.
344 */
345void mm_update_next_owner(struct mm_struct *mm)
346{
347 struct task_struct *c, *g, *p = current;
348
349retry:
350 /*
351 * If the exiting or execing task is not the owner, it's
352 * someone else's problem.
353 */
354 if (mm->owner != p)
355 return;
356 /*
357 * The current owner is exiting/execing and there are no other
358 * candidates. Do not leave the mm pointing to a possibly
359 * freed task structure.
360 */
361 if (atomic_read(&mm->mm_users) <= 1) {
362 WRITE_ONCE(mm->owner, NULL);
363 return;
364 }
365
366 read_lock(&tasklist_lock);
367 /*
368 * Search in the children
369 */
370 list_for_each_entry(c, &p->children, sibling) {
371 if (c->mm == mm)
372 goto assign_new_owner;
373 }
374
375 /*
376 * Search in the siblings
377 */
378 list_for_each_entry(c, &p->real_parent->children, sibling) {
379 if (c->mm == mm)
380 goto assign_new_owner;
381 }
382
383 /*
384 * Search through everything else, we should not get here often.
385 */
386 for_each_process(g) {
387 if (g->flags & PF_KTHREAD)
388 continue;
389 for_each_thread(g, c) {
390 if (c->mm == mm)
391 goto assign_new_owner;
392 if (c->mm)
393 break;
394 }
395 }
396 read_unlock(&tasklist_lock);
397 /*
398 * We found no owner yet mm_users > 1: this implies that we are
399 * most likely racing with swapoff (try_to_unuse()) or /proc or
400 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
401 */
402 WRITE_ONCE(mm->owner, NULL);
403 return;
404
405assign_new_owner:
406 BUG_ON(c == p);
407 get_task_struct(c);
408 /*
409 * The task_lock protects c->mm from changing.
410 * We always want mm->owner->mm == mm
411 */
412 task_lock(c);
413 /*
414 * Delay read_unlock() till we have the task_lock()
415 * to ensure that c does not slip away underneath us
416 */
417 read_unlock(&tasklist_lock);
418 if (c->mm != mm) {
419 task_unlock(c);
420 put_task_struct(c);
421 goto retry;
422 }
423 WRITE_ONCE(mm->owner, c);
424 task_unlock(c);
425 put_task_struct(c);
426}
427#endif /* CONFIG_MEMCG */
428
429/*
430 * Turn us into a lazy TLB process if we
431 * aren't already..
432 */
433static void exit_mm(void)
434{
435 struct mm_struct *mm = current->mm;
436 struct core_state *core_state;
437
438 exit_mm_release(current, mm);
439 if (!mm)
440 return;
441 sync_mm_rss(mm);
442 /*
443 * Serialize with any possible pending coredump.
444 * We must hold mmap_lock around checking core_state
445 * and clearing tsk->mm. The core-inducing thread
446 * will increment ->nr_threads for each thread in the
447 * group with ->mm != NULL.
448 */
449 mmap_read_lock(mm);
450 core_state = mm->core_state;
451 if (core_state) {
452 struct core_thread self;
453
454 mmap_read_unlock(mm);
455
456 self.task = current;
457 self.next = xchg(&core_state->dumper.next, &self);
458 /*
459 * Implies mb(), the result of xchg() must be visible
460 * to core_state->dumper.
461 */
462 if (atomic_dec_and_test(&core_state->nr_threads))
463 complete(&core_state->startup);
464
465 for (;;) {
466 set_current_state(TASK_UNINTERRUPTIBLE);
467 if (!self.task) /* see coredump_finish() */
468 break;
469 freezable_schedule();
470 }
471 __set_current_state(TASK_RUNNING);
472 mmap_read_lock(mm);
473 }
474 mmgrab(mm);
475 BUG_ON(mm != current->active_mm);
476 /* more a memory barrier than a real lock */
477 task_lock(current);
478 current->mm = NULL;
479 mmap_read_unlock(mm);
480 enter_lazy_tlb(mm, current);
481 task_unlock(current);
482 mm_update_next_owner(mm);
483 mmput(mm);
484 if (test_thread_flag(TIF_MEMDIE))
485 exit_oom_victim();
486}
487
488static struct task_struct *find_alive_thread(struct task_struct *p)
489{
490 struct task_struct *t;
491
492 for_each_thread(p, t) {
493 if (!(t->flags & PF_EXITING))
494 return t;
495 }
496 return NULL;
497}
498
499static struct task_struct *find_child_reaper(struct task_struct *father,
500 struct list_head *dead)
501 __releases(&tasklist_lock)
502 __acquires(&tasklist_lock)
503{
504 struct pid_namespace *pid_ns = task_active_pid_ns(father);
505 struct task_struct *reaper = pid_ns->child_reaper;
506 struct task_struct *p, *n;
507
508 if (likely(reaper != father))
509 return reaper;
510
511 reaper = find_alive_thread(father);
512 if (reaper) {
513 pid_ns->child_reaper = reaper;
514 return reaper;
515 }
516
517 write_unlock_irq(&tasklist_lock);
518
519 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
520 list_del_init(&p->ptrace_entry);
521 release_task(p);
522 }
523
524 zap_pid_ns_processes(pid_ns);
525 write_lock_irq(&tasklist_lock);
526
527 return father;
528}
529
530/*
531 * When we die, we re-parent all our children, and try to:
532 * 1. give them to another thread in our thread group, if such a member exists
533 * 2. give it to the first ancestor process which prctl'd itself as a
534 * child_subreaper for its children (like a service manager)
535 * 3. give it to the init process (PID 1) in our pid namespace
536 */
537static struct task_struct *find_new_reaper(struct task_struct *father,
538 struct task_struct *child_reaper)
539{
540 struct task_struct *thread, *reaper;
541
542 thread = find_alive_thread(father);
543 if (thread)
544 return thread;
545
546 if (father->signal->has_child_subreaper) {
547 unsigned int ns_level = task_pid(father)->level;
548 /*
549 * Find the first ->is_child_subreaper ancestor in our pid_ns.
550 * We can't check reaper != child_reaper to ensure we do not
551 * cross the namespaces, the exiting parent could be injected
552 * by setns() + fork().
553 * We check pid->level, this is slightly more efficient than
554 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
555 */
556 for (reaper = father->real_parent;
557 task_pid(reaper)->level == ns_level;
558 reaper = reaper->real_parent) {
559 if (reaper == &init_task)
560 break;
561 if (!reaper->signal->is_child_subreaper)
562 continue;
563 thread = find_alive_thread(reaper);
564 if (thread)
565 return thread;
566 }
567 }
568
569 return child_reaper;
570}
571
572/*
573* Any that need to be release_task'd are put on the @dead list.
574 */
575static void reparent_leader(struct task_struct *father, struct task_struct *p,
576 struct list_head *dead)
577{
578 if (unlikely(p->exit_state == EXIT_DEAD))
579 return;
580
581 /* We don't want people slaying init. */
582 p->exit_signal = SIGCHLD;
583
584 /* If it has exited notify the new parent about this child's death. */
585 if (!p->ptrace &&
586 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
587 if (do_notify_parent(p, p->exit_signal)) {
588 p->exit_state = EXIT_DEAD;
589 list_add(&p->ptrace_entry, dead);
590 }
591 }
592
593 kill_orphaned_pgrp(p, father);
594}
595
596/*
597 * This does two things:
598 *
599 * A. Make init inherit all the child processes
600 * B. Check to see if any process groups have become orphaned
601 * as a result of our exiting, and if they have any stopped
602 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
603 */
604static void forget_original_parent(struct task_struct *father,
605 struct list_head *dead)
606{
607 struct task_struct *p, *t, *reaper;
608
609 if (unlikely(!list_empty(&father->ptraced)))
610 exit_ptrace(father, dead);
611
612 /* Can drop and reacquire tasklist_lock */
613 reaper = find_child_reaper(father, dead);
614 if (list_empty(&father->children))
615 return;
616
617 reaper = find_new_reaper(father, reaper);
618 list_for_each_entry(p, &father->children, sibling) {
619 for_each_thread(p, t) {
620 RCU_INIT_POINTER(t->real_parent, reaper);
621 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
622 if (likely(!t->ptrace))
623 t->parent = t->real_parent;
624 if (t->pdeath_signal)
625 group_send_sig_info(t->pdeath_signal,
626 SEND_SIG_NOINFO, t,
627 PIDTYPE_TGID);
628 }
629 /*
630 * If this is a threaded reparent there is no need to
631 * notify anyone anything has happened.
632 */
633 if (!same_thread_group(reaper, father))
634 reparent_leader(father, p, dead);
635 }
636 list_splice_tail_init(&father->children, &reaper->children);
637}
638
639/*
640 * Send signals to all our closest relatives so that they know
641 * to properly mourn us..
642 */
643static void exit_notify(struct task_struct *tsk, int group_dead)
644{
645 bool autoreap;
646 struct task_struct *p, *n;
647 LIST_HEAD(dead);
648
649 write_lock_irq(&tasklist_lock);
650 forget_original_parent(tsk, &dead);
651
652 if (group_dead)
653 kill_orphaned_pgrp(tsk->group_leader, NULL);
654
655 tsk->exit_state = EXIT_ZOMBIE;
656 if (unlikely(tsk->ptrace)) {
657 int sig = thread_group_leader(tsk) &&
658 thread_group_empty(tsk) &&
659 !ptrace_reparented(tsk) ?
660 tsk->exit_signal : SIGCHLD;
661 autoreap = do_notify_parent(tsk, sig);
662 } else if (thread_group_leader(tsk)) {
663 autoreap = thread_group_empty(tsk) &&
664 do_notify_parent(tsk, tsk->exit_signal);
665 } else {
666 autoreap = true;
667 }
668
669 if (autoreap) {
670 tsk->exit_state = EXIT_DEAD;
671 list_add(&tsk->ptrace_entry, &dead);
672 }
673
674 /* mt-exec, de_thread() is waiting for group leader */
675 if (unlikely(tsk->signal->notify_count < 0))
676 wake_up_process(tsk->signal->group_exit_task);
677 write_unlock_irq(&tasklist_lock);
678
679 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
680 list_del_init(&p->ptrace_entry);
681 release_task(p);
682 }
683}
684
685#ifdef CONFIG_DEBUG_STACK_USAGE
686static void check_stack_usage(void)
687{
688 static DEFINE_SPINLOCK(low_water_lock);
689 static int lowest_to_date = THREAD_SIZE;
690 unsigned long free;
691
692 free = stack_not_used(current);
693
694 if (free >= lowest_to_date)
695 return;
696
697 spin_lock(&low_water_lock);
698 if (free < lowest_to_date) {
699 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
700 current->comm, task_pid_nr(current), free);
701 lowest_to_date = free;
702 }
703 spin_unlock(&low_water_lock);
704}
705#else
706static inline void check_stack_usage(void) {}
707#endif
708
709void __noreturn do_exit(long code)
710{
711 struct task_struct *tsk = current;
712 int group_dead;
713
714 /*
715 * We can get here from a kernel oops, sometimes with preemption off.
716 * Start by checking for critical errors.
717 * Then fix up important state like USER_DS and preemption.
718 * Then do everything else.
719 */
720
721 WARN_ON(blk_needs_flush_plug(tsk));
722
723 if (unlikely(in_interrupt()))
724 panic("Aiee, killing interrupt handler!");
725 if (unlikely(!tsk->pid))
726 panic("Attempted to kill the idle task!");
727
728 /*
729 * If do_exit is called because this processes oopsed, it's possible
730 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
731 * continuing. Amongst other possible reasons, this is to prevent
732 * mm_release()->clear_child_tid() from writing to a user-controlled
733 * kernel address.
734 */
735 force_uaccess_begin();
736
737 if (unlikely(in_atomic())) {
738 pr_info("note: %s[%d] exited with preempt_count %d\n",
739 current->comm, task_pid_nr(current),
740 preempt_count());
741 preempt_count_set(PREEMPT_ENABLED);
742 }
743
744 profile_task_exit(tsk);
745 kcov_task_exit(tsk);
746
747 ptrace_event(PTRACE_EVENT_EXIT, code);
748
749 validate_creds_for_do_exit(tsk);
750
751 /*
752 * We're taking recursive faults here in do_exit. Safest is to just
753 * leave this task alone and wait for reboot.
754 */
755 if (unlikely(tsk->flags & PF_EXITING)) {
756 pr_alert("Fixing recursive fault but reboot is needed!\n");
757 futex_exit_recursive(tsk);
758 set_current_state(TASK_UNINTERRUPTIBLE);
759 schedule();
760 }
761
762 exit_signals(tsk); /* sets PF_EXITING */
763
764 /* sync mm's RSS info before statistics gathering */
765 if (tsk->mm)
766 sync_mm_rss(tsk->mm);
767 acct_update_integrals(tsk);
768 group_dead = atomic_dec_and_test(&tsk->signal->live);
769 if (group_dead) {
770 /*
771 * If the last thread of global init has exited, panic
772 * immediately to get a useable coredump.
773 */
774 if (unlikely(is_global_init(tsk)))
775 panic("Attempted to kill init! exitcode=0x%08x\n",
776 tsk->signal->group_exit_code ?: (int)code);
777
778#ifdef CONFIG_POSIX_TIMERS
779 hrtimer_cancel(&tsk->signal->real_timer);
780 exit_itimers(tsk->signal);
781#endif
782 if (tsk->mm)
783 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
784 }
785 acct_collect(code, group_dead);
786 if (group_dead)
787 tty_audit_exit();
788 audit_free(tsk);
789
790 tsk->exit_code = code;
791 taskstats_exit(tsk, group_dead);
792
793 exit_mm();
794
795 if (group_dead)
796 acct_process();
797 trace_sched_process_exit(tsk);
798
799 exit_sem(tsk);
800 exit_shm(tsk);
801 exit_files(tsk);
802 exit_fs(tsk);
803 if (group_dead)
804 disassociate_ctty(1);
805 exit_task_namespaces(tsk);
806 exit_task_work(tsk);
807 exit_thread(tsk);
808
809 /*
810 * Flush inherited counters to the parent - before the parent
811 * gets woken up by child-exit notifications.
812 *
813 * because of cgroup mode, must be called before cgroup_exit()
814 */
815 perf_event_exit_task(tsk);
816
817 sched_autogroup_exit_task(tsk);
818 cgroup_exit(tsk);
819
820 /*
821 * FIXME: do that only when needed, using sched_exit tracepoint
822 */
823 flush_ptrace_hw_breakpoint(tsk);
824
825 exit_tasks_rcu_start();
826 exit_notify(tsk, group_dead);
827 proc_exit_connector(tsk);
828 mpol_put_task_policy(tsk);
829#ifdef CONFIG_FUTEX
830 if (unlikely(current->pi_state_cache))
831 kfree(current->pi_state_cache);
832#endif
833 /*
834 * Make sure we are holding no locks:
835 */
836 debug_check_no_locks_held();
837
838 if (tsk->io_context)
839 exit_io_context(tsk);
840
841 if (tsk->splice_pipe)
842 free_pipe_info(tsk->splice_pipe);
843
844 if (tsk->task_frag.page)
845 put_page(tsk->task_frag.page);
846
847 validate_creds_for_do_exit(tsk);
848
849 check_stack_usage();
850 preempt_disable();
851 if (tsk->nr_dirtied)
852 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
853 exit_rcu();
854 exit_tasks_rcu_finish();
855
856 lockdep_free_task(tsk);
857 do_task_dead();
858}
859EXPORT_SYMBOL_GPL(do_exit);
860
861void complete_and_exit(struct completion *comp, long code)
862{
863 if (comp)
864 complete(comp);
865
866 do_exit(code);
867}
868EXPORT_SYMBOL(complete_and_exit);
869
870SYSCALL_DEFINE1(exit, int, error_code)
871{
872 do_exit((error_code&0xff)<<8);
873}
874
875/*
876 * Take down every thread in the group. This is called by fatal signals
877 * as well as by sys_exit_group (below).
878 */
879void
880do_group_exit(int exit_code)
881{
882 struct signal_struct *sig = current->signal;
883
884 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
885
886 if (signal_group_exit(sig))
887 exit_code = sig->group_exit_code;
888 else if (!thread_group_empty(current)) {
889 struct sighand_struct *const sighand = current->sighand;
890
891 spin_lock_irq(&sighand->siglock);
892 if (signal_group_exit(sig))
893 /* Another thread got here before we took the lock. */
894 exit_code = sig->group_exit_code;
895 else {
896 sig->group_exit_code = exit_code;
897 sig->flags = SIGNAL_GROUP_EXIT;
898 zap_other_threads(current);
899 }
900 spin_unlock_irq(&sighand->siglock);
901 }
902
903 do_exit(exit_code);
904 /* NOTREACHED */
905}
906
907/*
908 * this kills every thread in the thread group. Note that any externally
909 * wait4()-ing process will get the correct exit code - even if this
910 * thread is not the thread group leader.
911 */
912SYSCALL_DEFINE1(exit_group, int, error_code)
913{
914 do_group_exit((error_code & 0xff) << 8);
915 /* NOTREACHED */
916 return 0;
917}
918
919struct waitid_info {
920 pid_t pid;
921 uid_t uid;
922 int status;
923 int cause;
924};
925
926struct wait_opts {
927 enum pid_type wo_type;
928 int wo_flags;
929 struct pid *wo_pid;
930
931 struct waitid_info *wo_info;
932 int wo_stat;
933 struct rusage *wo_rusage;
934
935 wait_queue_entry_t child_wait;
936 int notask_error;
937};
938
939static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
940{
941 return wo->wo_type == PIDTYPE_MAX ||
942 task_pid_type(p, wo->wo_type) == wo->wo_pid;
943}
944
945static int
946eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
947{
948 if (!eligible_pid(wo, p))
949 return 0;
950
951 /*
952 * Wait for all children (clone and not) if __WALL is set or
953 * if it is traced by us.
954 */
955 if (ptrace || (wo->wo_flags & __WALL))
956 return 1;
957
958 /*
959 * Otherwise, wait for clone children *only* if __WCLONE is set;
960 * otherwise, wait for non-clone children *only*.
961 *
962 * Note: a "clone" child here is one that reports to its parent
963 * using a signal other than SIGCHLD, or a non-leader thread which
964 * we can only see if it is traced by us.
965 */
966 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
967 return 0;
968
969 return 1;
970}
971
972/*
973 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
974 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
975 * the lock and this task is uninteresting. If we return nonzero, we have
976 * released the lock and the system call should return.
977 */
978static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
979{
980 int state, status;
981 pid_t pid = task_pid_vnr(p);
982 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
983 struct waitid_info *infop;
984
985 if (!likely(wo->wo_flags & WEXITED))
986 return 0;
987
988 if (unlikely(wo->wo_flags & WNOWAIT)) {
989 status = p->exit_code;
990 get_task_struct(p);
991 read_unlock(&tasklist_lock);
992 sched_annotate_sleep();
993 if (wo->wo_rusage)
994 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
995 put_task_struct(p);
996 goto out_info;
997 }
998 /*
999 * Move the task's state to DEAD/TRACE, only one thread can do this.
1000 */
1001 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1002 EXIT_TRACE : EXIT_DEAD;
1003 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1004 return 0;
1005 /*
1006 * We own this thread, nobody else can reap it.
1007 */
1008 read_unlock(&tasklist_lock);
1009 sched_annotate_sleep();
1010
1011 /*
1012 * Check thread_group_leader() to exclude the traced sub-threads.
1013 */
1014 if (state == EXIT_DEAD && thread_group_leader(p)) {
1015 struct signal_struct *sig = p->signal;
1016 struct signal_struct *psig = current->signal;
1017 unsigned long maxrss;
1018 u64 tgutime, tgstime;
1019
1020 /*
1021 * The resource counters for the group leader are in its
1022 * own task_struct. Those for dead threads in the group
1023 * are in its signal_struct, as are those for the child
1024 * processes it has previously reaped. All these
1025 * accumulate in the parent's signal_struct c* fields.
1026 *
1027 * We don't bother to take a lock here to protect these
1028 * p->signal fields because the whole thread group is dead
1029 * and nobody can change them.
1030 *
1031 * psig->stats_lock also protects us from our sub-theads
1032 * which can reap other children at the same time. Until
1033 * we change k_getrusage()-like users to rely on this lock
1034 * we have to take ->siglock as well.
1035 *
1036 * We use thread_group_cputime_adjusted() to get times for
1037 * the thread group, which consolidates times for all threads
1038 * in the group including the group leader.
1039 */
1040 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1041 spin_lock_irq(¤t->sighand->siglock);
1042 write_seqlock(&psig->stats_lock);
1043 psig->cutime += tgutime + sig->cutime;
1044 psig->cstime += tgstime + sig->cstime;
1045 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1046 psig->cmin_flt +=
1047 p->min_flt + sig->min_flt + sig->cmin_flt;
1048 psig->cmaj_flt +=
1049 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1050 psig->cnvcsw +=
1051 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1052 psig->cnivcsw +=
1053 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1054 psig->cinblock +=
1055 task_io_get_inblock(p) +
1056 sig->inblock + sig->cinblock;
1057 psig->coublock +=
1058 task_io_get_oublock(p) +
1059 sig->oublock + sig->coublock;
1060 maxrss = max(sig->maxrss, sig->cmaxrss);
1061 if (psig->cmaxrss < maxrss)
1062 psig->cmaxrss = maxrss;
1063 task_io_accounting_add(&psig->ioac, &p->ioac);
1064 task_io_accounting_add(&psig->ioac, &sig->ioac);
1065 write_sequnlock(&psig->stats_lock);
1066 spin_unlock_irq(¤t->sighand->siglock);
1067 }
1068
1069 if (wo->wo_rusage)
1070 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1071 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1072 ? p->signal->group_exit_code : p->exit_code;
1073 wo->wo_stat = status;
1074
1075 if (state == EXIT_TRACE) {
1076 write_lock_irq(&tasklist_lock);
1077 /* We dropped tasklist, ptracer could die and untrace */
1078 ptrace_unlink(p);
1079
1080 /* If parent wants a zombie, don't release it now */
1081 state = EXIT_ZOMBIE;
1082 if (do_notify_parent(p, p->exit_signal))
1083 state = EXIT_DEAD;
1084 p->exit_state = state;
1085 write_unlock_irq(&tasklist_lock);
1086 }
1087 if (state == EXIT_DEAD)
1088 release_task(p);
1089
1090out_info:
1091 infop = wo->wo_info;
1092 if (infop) {
1093 if ((status & 0x7f) == 0) {
1094 infop->cause = CLD_EXITED;
1095 infop->status = status >> 8;
1096 } else {
1097 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1098 infop->status = status & 0x7f;
1099 }
1100 infop->pid = pid;
1101 infop->uid = uid;
1102 }
1103
1104 return pid;
1105}
1106
1107static int *task_stopped_code(struct task_struct *p, bool ptrace)
1108{
1109 if (ptrace) {
1110 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1111 return &p->exit_code;
1112 } else {
1113 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1114 return &p->signal->group_exit_code;
1115 }
1116 return NULL;
1117}
1118
1119/**
1120 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1121 * @wo: wait options
1122 * @ptrace: is the wait for ptrace
1123 * @p: task to wait for
1124 *
1125 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1126 *
1127 * CONTEXT:
1128 * read_lock(&tasklist_lock), which is released if return value is
1129 * non-zero. Also, grabs and releases @p->sighand->siglock.
1130 *
1131 * RETURNS:
1132 * 0 if wait condition didn't exist and search for other wait conditions
1133 * should continue. Non-zero return, -errno on failure and @p's pid on
1134 * success, implies that tasklist_lock is released and wait condition
1135 * search should terminate.
1136 */
1137static int wait_task_stopped(struct wait_opts *wo,
1138 int ptrace, struct task_struct *p)
1139{
1140 struct waitid_info *infop;
1141 int exit_code, *p_code, why;
1142 uid_t uid = 0; /* unneeded, required by compiler */
1143 pid_t pid;
1144
1145 /*
1146 * Traditionally we see ptrace'd stopped tasks regardless of options.
1147 */
1148 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1149 return 0;
1150
1151 if (!task_stopped_code(p, ptrace))
1152 return 0;
1153
1154 exit_code = 0;
1155 spin_lock_irq(&p->sighand->siglock);
1156
1157 p_code = task_stopped_code(p, ptrace);
1158 if (unlikely(!p_code))
1159 goto unlock_sig;
1160
1161 exit_code = *p_code;
1162 if (!exit_code)
1163 goto unlock_sig;
1164
1165 if (!unlikely(wo->wo_flags & WNOWAIT))
1166 *p_code = 0;
1167
1168 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1169unlock_sig:
1170 spin_unlock_irq(&p->sighand->siglock);
1171 if (!exit_code)
1172 return 0;
1173
1174 /*
1175 * Now we are pretty sure this task is interesting.
1176 * Make sure it doesn't get reaped out from under us while we
1177 * give up the lock and then examine it below. We don't want to
1178 * keep holding onto the tasklist_lock while we call getrusage and
1179 * possibly take page faults for user memory.
1180 */
1181 get_task_struct(p);
1182 pid = task_pid_vnr(p);
1183 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1184 read_unlock(&tasklist_lock);
1185 sched_annotate_sleep();
1186 if (wo->wo_rusage)
1187 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1188 put_task_struct(p);
1189
1190 if (likely(!(wo->wo_flags & WNOWAIT)))
1191 wo->wo_stat = (exit_code << 8) | 0x7f;
1192
1193 infop = wo->wo_info;
1194 if (infop) {
1195 infop->cause = why;
1196 infop->status = exit_code;
1197 infop->pid = pid;
1198 infop->uid = uid;
1199 }
1200 return pid;
1201}
1202
1203/*
1204 * Handle do_wait work for one task in a live, non-stopped state.
1205 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1206 * the lock and this task is uninteresting. If we return nonzero, we have
1207 * released the lock and the system call should return.
1208 */
1209static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1210{
1211 struct waitid_info *infop;
1212 pid_t pid;
1213 uid_t uid;
1214
1215 if (!unlikely(wo->wo_flags & WCONTINUED))
1216 return 0;
1217
1218 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1219 return 0;
1220
1221 spin_lock_irq(&p->sighand->siglock);
1222 /* Re-check with the lock held. */
1223 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1224 spin_unlock_irq(&p->sighand->siglock);
1225 return 0;
1226 }
1227 if (!unlikely(wo->wo_flags & WNOWAIT))
1228 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1229 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1230 spin_unlock_irq(&p->sighand->siglock);
1231
1232 pid = task_pid_vnr(p);
1233 get_task_struct(p);
1234 read_unlock(&tasklist_lock);
1235 sched_annotate_sleep();
1236 if (wo->wo_rusage)
1237 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1238 put_task_struct(p);
1239
1240 infop = wo->wo_info;
1241 if (!infop) {
1242 wo->wo_stat = 0xffff;
1243 } else {
1244 infop->cause = CLD_CONTINUED;
1245 infop->pid = pid;
1246 infop->uid = uid;
1247 infop->status = SIGCONT;
1248 }
1249 return pid;
1250}
1251
1252/*
1253 * Consider @p for a wait by @parent.
1254 *
1255 * -ECHILD should be in ->notask_error before the first call.
1256 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1257 * Returns zero if the search for a child should continue;
1258 * then ->notask_error is 0 if @p is an eligible child,
1259 * or still -ECHILD.
1260 */
1261static int wait_consider_task(struct wait_opts *wo, int ptrace,
1262 struct task_struct *p)
1263{
1264 /*
1265 * We can race with wait_task_zombie() from another thread.
1266 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1267 * can't confuse the checks below.
1268 */
1269 int exit_state = READ_ONCE(p->exit_state);
1270 int ret;
1271
1272 if (unlikely(exit_state == EXIT_DEAD))
1273 return 0;
1274
1275 ret = eligible_child(wo, ptrace, p);
1276 if (!ret)
1277 return ret;
1278
1279 if (unlikely(exit_state == EXIT_TRACE)) {
1280 /*
1281 * ptrace == 0 means we are the natural parent. In this case
1282 * we should clear notask_error, debugger will notify us.
1283 */
1284 if (likely(!ptrace))
1285 wo->notask_error = 0;
1286 return 0;
1287 }
1288
1289 if (likely(!ptrace) && unlikely(p->ptrace)) {
1290 /*
1291 * If it is traced by its real parent's group, just pretend
1292 * the caller is ptrace_do_wait() and reap this child if it
1293 * is zombie.
1294 *
1295 * This also hides group stop state from real parent; otherwise
1296 * a single stop can be reported twice as group and ptrace stop.
1297 * If a ptracer wants to distinguish these two events for its
1298 * own children it should create a separate process which takes
1299 * the role of real parent.
1300 */
1301 if (!ptrace_reparented(p))
1302 ptrace = 1;
1303 }
1304
1305 /* slay zombie? */
1306 if (exit_state == EXIT_ZOMBIE) {
1307 /* we don't reap group leaders with subthreads */
1308 if (!delay_group_leader(p)) {
1309 /*
1310 * A zombie ptracee is only visible to its ptracer.
1311 * Notification and reaping will be cascaded to the
1312 * real parent when the ptracer detaches.
1313 */
1314 if (unlikely(ptrace) || likely(!p->ptrace))
1315 return wait_task_zombie(wo, p);
1316 }
1317
1318 /*
1319 * Allow access to stopped/continued state via zombie by
1320 * falling through. Clearing of notask_error is complex.
1321 *
1322 * When !@ptrace:
1323 *
1324 * If WEXITED is set, notask_error should naturally be
1325 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1326 * so, if there are live subthreads, there are events to
1327 * wait for. If all subthreads are dead, it's still safe
1328 * to clear - this function will be called again in finite
1329 * amount time once all the subthreads are released and
1330 * will then return without clearing.
1331 *
1332 * When @ptrace:
1333 *
1334 * Stopped state is per-task and thus can't change once the
1335 * target task dies. Only continued and exited can happen.
1336 * Clear notask_error if WCONTINUED | WEXITED.
1337 */
1338 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1339 wo->notask_error = 0;
1340 } else {
1341 /*
1342 * @p is alive and it's gonna stop, continue or exit, so
1343 * there always is something to wait for.
1344 */
1345 wo->notask_error = 0;
1346 }
1347
1348 /*
1349 * Wait for stopped. Depending on @ptrace, different stopped state
1350 * is used and the two don't interact with each other.
1351 */
1352 ret = wait_task_stopped(wo, ptrace, p);
1353 if (ret)
1354 return ret;
1355
1356 /*
1357 * Wait for continued. There's only one continued state and the
1358 * ptracer can consume it which can confuse the real parent. Don't
1359 * use WCONTINUED from ptracer. You don't need or want it.
1360 */
1361 return wait_task_continued(wo, p);
1362}
1363
1364/*
1365 * Do the work of do_wait() for one thread in the group, @tsk.
1366 *
1367 * -ECHILD should be in ->notask_error before the first call.
1368 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1369 * Returns zero if the search for a child should continue; then
1370 * ->notask_error is 0 if there were any eligible children,
1371 * or still -ECHILD.
1372 */
1373static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1374{
1375 struct task_struct *p;
1376
1377 list_for_each_entry(p, &tsk->children, sibling) {
1378 int ret = wait_consider_task(wo, 0, p);
1379
1380 if (ret)
1381 return ret;
1382 }
1383
1384 return 0;
1385}
1386
1387static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1388{
1389 struct task_struct *p;
1390
1391 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1392 int ret = wait_consider_task(wo, 1, p);
1393
1394 if (ret)
1395 return ret;
1396 }
1397
1398 return 0;
1399}
1400
1401static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1402 int sync, void *key)
1403{
1404 struct wait_opts *wo = container_of(wait, struct wait_opts,
1405 child_wait);
1406 struct task_struct *p = key;
1407
1408 if (!eligible_pid(wo, p))
1409 return 0;
1410
1411 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1412 return 0;
1413
1414 return default_wake_function(wait, mode, sync, key);
1415}
1416
1417void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1418{
1419 __wake_up_sync_key(&parent->signal->wait_chldexit,
1420 TASK_INTERRUPTIBLE, p);
1421}
1422
1423static long do_wait(struct wait_opts *wo)
1424{
1425 struct task_struct *tsk;
1426 int retval;
1427
1428 trace_sched_process_wait(wo->wo_pid);
1429
1430 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1431 wo->child_wait.private = current;
1432 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1433repeat:
1434 /*
1435 * If there is nothing that can match our criteria, just get out.
1436 * We will clear ->notask_error to zero if we see any child that
1437 * might later match our criteria, even if we are not able to reap
1438 * it yet.
1439 */
1440 wo->notask_error = -ECHILD;
1441 if ((wo->wo_type < PIDTYPE_MAX) &&
1442 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1443 goto notask;
1444
1445 set_current_state(TASK_INTERRUPTIBLE);
1446 read_lock(&tasklist_lock);
1447 tsk = current;
1448 do {
1449 retval = do_wait_thread(wo, tsk);
1450 if (retval)
1451 goto end;
1452
1453 retval = ptrace_do_wait(wo, tsk);
1454 if (retval)
1455 goto end;
1456
1457 if (wo->wo_flags & __WNOTHREAD)
1458 break;
1459 } while_each_thread(current, tsk);
1460 read_unlock(&tasklist_lock);
1461
1462notask:
1463 retval = wo->notask_error;
1464 if (!retval && !(wo->wo_flags & WNOHANG)) {
1465 retval = -ERESTARTSYS;
1466 if (!signal_pending(current)) {
1467 schedule();
1468 goto repeat;
1469 }
1470 }
1471end:
1472 __set_current_state(TASK_RUNNING);
1473 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1474 return retval;
1475}
1476
1477static struct pid *pidfd_get_pid(unsigned int fd)
1478{
1479 struct fd f;
1480 struct pid *pid;
1481
1482 f = fdget(fd);
1483 if (!f.file)
1484 return ERR_PTR(-EBADF);
1485
1486 pid = pidfd_pid(f.file);
1487 if (!IS_ERR(pid))
1488 get_pid(pid);
1489
1490 fdput(f);
1491 return pid;
1492}
1493
1494static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1495 int options, struct rusage *ru)
1496{
1497 struct wait_opts wo;
1498 struct pid *pid = NULL;
1499 enum pid_type type;
1500 long ret;
1501
1502 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1503 __WNOTHREAD|__WCLONE|__WALL))
1504 return -EINVAL;
1505 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1506 return -EINVAL;
1507
1508 switch (which) {
1509 case P_ALL:
1510 type = PIDTYPE_MAX;
1511 break;
1512 case P_PID:
1513 type = PIDTYPE_PID;
1514 if (upid <= 0)
1515 return -EINVAL;
1516
1517 pid = find_get_pid(upid);
1518 break;
1519 case P_PGID:
1520 type = PIDTYPE_PGID;
1521 if (upid < 0)
1522 return -EINVAL;
1523
1524 if (upid)
1525 pid = find_get_pid(upid);
1526 else
1527 pid = get_task_pid(current, PIDTYPE_PGID);
1528 break;
1529 case P_PIDFD:
1530 type = PIDTYPE_PID;
1531 if (upid < 0)
1532 return -EINVAL;
1533
1534 pid = pidfd_get_pid(upid);
1535 if (IS_ERR(pid))
1536 return PTR_ERR(pid);
1537 break;
1538 default:
1539 return -EINVAL;
1540 }
1541
1542 wo.wo_type = type;
1543 wo.wo_pid = pid;
1544 wo.wo_flags = options;
1545 wo.wo_info = infop;
1546 wo.wo_rusage = ru;
1547 ret = do_wait(&wo);
1548
1549 put_pid(pid);
1550 return ret;
1551}
1552
1553SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1554 infop, int, options, struct rusage __user *, ru)
1555{
1556 struct rusage r;
1557 struct waitid_info info = {.status = 0};
1558 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1559 int signo = 0;
1560
1561 if (err > 0) {
1562 signo = SIGCHLD;
1563 err = 0;
1564 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1565 return -EFAULT;
1566 }
1567 if (!infop)
1568 return err;
1569
1570 if (!user_write_access_begin(infop, sizeof(*infop)))
1571 return -EFAULT;
1572
1573 unsafe_put_user(signo, &infop->si_signo, Efault);
1574 unsafe_put_user(0, &infop->si_errno, Efault);
1575 unsafe_put_user(info.cause, &infop->si_code, Efault);
1576 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1577 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1578 unsafe_put_user(info.status, &infop->si_status, Efault);
1579 user_write_access_end();
1580 return err;
1581Efault:
1582 user_write_access_end();
1583 return -EFAULT;
1584}
1585
1586long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1587 struct rusage *ru)
1588{
1589 struct wait_opts wo;
1590 struct pid *pid = NULL;
1591 enum pid_type type;
1592 long ret;
1593
1594 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1595 __WNOTHREAD|__WCLONE|__WALL))
1596 return -EINVAL;
1597
1598 /* -INT_MIN is not defined */
1599 if (upid == INT_MIN)
1600 return -ESRCH;
1601
1602 if (upid == -1)
1603 type = PIDTYPE_MAX;
1604 else if (upid < 0) {
1605 type = PIDTYPE_PGID;
1606 pid = find_get_pid(-upid);
1607 } else if (upid == 0) {
1608 type = PIDTYPE_PGID;
1609 pid = get_task_pid(current, PIDTYPE_PGID);
1610 } else /* upid > 0 */ {
1611 type = PIDTYPE_PID;
1612 pid = find_get_pid(upid);
1613 }
1614
1615 wo.wo_type = type;
1616 wo.wo_pid = pid;
1617 wo.wo_flags = options | WEXITED;
1618 wo.wo_info = NULL;
1619 wo.wo_stat = 0;
1620 wo.wo_rusage = ru;
1621 ret = do_wait(&wo);
1622 put_pid(pid);
1623 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1624 ret = -EFAULT;
1625
1626 return ret;
1627}
1628
1629int kernel_wait(pid_t pid, int *stat)
1630{
1631 struct wait_opts wo = {
1632 .wo_type = PIDTYPE_PID,
1633 .wo_pid = find_get_pid(pid),
1634 .wo_flags = WEXITED,
1635 };
1636 int ret;
1637
1638 ret = do_wait(&wo);
1639 if (ret > 0 && wo.wo_stat)
1640 *stat = wo.wo_stat;
1641 put_pid(wo.wo_pid);
1642 return ret;
1643}
1644
1645SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1646 int, options, struct rusage __user *, ru)
1647{
1648 struct rusage r;
1649 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1650
1651 if (err > 0) {
1652 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1653 return -EFAULT;
1654 }
1655 return err;
1656}
1657
1658#ifdef __ARCH_WANT_SYS_WAITPID
1659
1660/*
1661 * sys_waitpid() remains for compatibility. waitpid() should be
1662 * implemented by calling sys_wait4() from libc.a.
1663 */
1664SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1665{
1666 return kernel_wait4(pid, stat_addr, options, NULL);
1667}
1668
1669#endif
1670
1671#ifdef CONFIG_COMPAT
1672COMPAT_SYSCALL_DEFINE4(wait4,
1673 compat_pid_t, pid,
1674 compat_uint_t __user *, stat_addr,
1675 int, options,
1676 struct compat_rusage __user *, ru)
1677{
1678 struct rusage r;
1679 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1680 if (err > 0) {
1681 if (ru && put_compat_rusage(&r, ru))
1682 return -EFAULT;
1683 }
1684 return err;
1685}
1686
1687COMPAT_SYSCALL_DEFINE5(waitid,
1688 int, which, compat_pid_t, pid,
1689 struct compat_siginfo __user *, infop, int, options,
1690 struct compat_rusage __user *, uru)
1691{
1692 struct rusage ru;
1693 struct waitid_info info = {.status = 0};
1694 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1695 int signo = 0;
1696 if (err > 0) {
1697 signo = SIGCHLD;
1698 err = 0;
1699 if (uru) {
1700 /* kernel_waitid() overwrites everything in ru */
1701 if (COMPAT_USE_64BIT_TIME)
1702 err = copy_to_user(uru, &ru, sizeof(ru));
1703 else
1704 err = put_compat_rusage(&ru, uru);
1705 if (err)
1706 return -EFAULT;
1707 }
1708 }
1709
1710 if (!infop)
1711 return err;
1712
1713 if (!user_write_access_begin(infop, sizeof(*infop)))
1714 return -EFAULT;
1715
1716 unsafe_put_user(signo, &infop->si_signo, Efault);
1717 unsafe_put_user(0, &infop->si_errno, Efault);
1718 unsafe_put_user(info.cause, &infop->si_code, Efault);
1719 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1720 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1721 unsafe_put_user(info.status, &infop->si_status, Efault);
1722 user_write_access_end();
1723 return err;
1724Efault:
1725 user_write_access_end();
1726 return -EFAULT;
1727}
1728#endif
1729
1730/**
1731 * thread_group_exited - check that a thread group has exited
1732 * @pid: tgid of thread group to be checked.
1733 *
1734 * Test if the thread group represented by tgid has exited (all
1735 * threads are zombies, dead or completely gone).
1736 *
1737 * Return: true if the thread group has exited. false otherwise.
1738 */
1739bool thread_group_exited(struct pid *pid)
1740{
1741 struct task_struct *task;
1742 bool exited;
1743
1744 rcu_read_lock();
1745 task = pid_task(pid, PIDTYPE_PID);
1746 exited = !task ||
1747 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1748 rcu_read_unlock();
1749
1750 return exited;
1751}
1752EXPORT_SYMBOL(thread_group_exited);
1753
1754__weak void abort(void)
1755{
1756 BUG();
1757
1758 /* if that doesn't kill us, halt */
1759 panic("Oops failed to kill thread");
1760}
1761EXPORT_SYMBOL(abort);