kernel/exit.c

   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
  71 static 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  */
  91 static 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
 166 static 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
 175 void 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
 181 void release_task(struct task_struct *p)
 182 {
 183         struct task_struct *leader;
 184         struct pid *thread_pid;
 185         int zap_leader;
 186 repeat:
 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
 231 int 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 }
 258 EXPORT_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  */
 268 static 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
 287 int 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
 298 static 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  */
 315 static void
 316 kill_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  */
 345 void mm_update_next_owner(struct mm_struct *mm)
 346 {
 347         struct task_struct *c, *g, *p = current;
 348
 349 retry:
 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
 405 assign_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  */
 433 static 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                 if (self.task->flags & PF_SIGNALED)
 458                         self.next = xchg(&core_state->dumper.next, &self);
 459                 else
 460                         self.task = NULL;
 461                 /*
 462                  * Implies mb(), the result of xchg() must be visible
 463                  * to core_state->dumper.
 464                  */
 465                 if (atomic_dec_and_test(&core_state->nr_threads))
 466                         complete(&core_state->startup);
 467
 468                 for (;;) {
 469                         set_current_state(TASK_UNINTERRUPTIBLE);
 470                         if (!self.task) /* see coredump_finish() */
 471                                 break;
 472                         freezable_schedule();
 473                 }
 474                 __set_current_state(TASK_RUNNING);
 475                 mmap_read_lock(mm);
 476         }
 477         mmgrab(mm);
 478         BUG_ON(mm != current->active_mm);
 479         /* more a memory barrier than a real lock */
 480         task_lock(current);
 481         /*
 482          * When a thread stops operating on an address space, the loop
 483          * in membarrier_private_expedited() may not observe that
 484          * tsk->mm, and the loop in membarrier_global_expedited() may
 485          * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
 486          * rq->membarrier_state, so those would not issue an IPI.
 487          * Membarrier requires a memory barrier after accessing
 488          * user-space memory, before clearing tsk->mm or the
 489          * rq->membarrier_state.
 490          */
 491         smp_mb__after_spinlock();
 492         local_irq_disable();
 493         current->mm = NULL;
 494         membarrier_update_current_mm(NULL);
 495         enter_lazy_tlb(mm, current);
 496         local_irq_enable();
 497         task_unlock(current);
 498         mmap_read_unlock(mm);
 499         mm_update_next_owner(mm);
 500         mmput(mm);
 501         if (test_thread_flag(TIF_MEMDIE))
 502                 exit_oom_victim();
 503 }
 504
 505 static struct task_struct *find_alive_thread(struct task_struct *p)
 506 {
 507         struct task_struct *t;
 508
 509         for_each_thread(p, t) {
 510                 if (!(t->flags & PF_EXITING))
 511                         return t;
 512         }
 513         return NULL;
 514 }
 515
 516 static struct task_struct *find_child_reaper(struct task_struct *father,
 517                                                 struct list_head *dead)
 518         __releases(&tasklist_lock)
 519         __acquires(&tasklist_lock)
 520 {
 521         struct pid_namespace *pid_ns = task_active_pid_ns(father);
 522         struct task_struct *reaper = pid_ns->child_reaper;
 523         struct task_struct *p, *n;
 524
 525         if (likely(reaper != father))
 526                 return reaper;
 527
 528         reaper = find_alive_thread(father);
 529         if (reaper) {
 530                 pid_ns->child_reaper = reaper;
 531                 return reaper;
 532         }
 533
 534         write_unlock_irq(&tasklist_lock);
 535
 536         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
 537                 list_del_init(&p->ptrace_entry);
 538                 release_task(p);
 539         }
 540
 541         zap_pid_ns_processes(pid_ns);
 542         write_lock_irq(&tasklist_lock);
 543
 544         return father;
 545 }
 546
 547 /*
 548  * When we die, we re-parent all our children, and try to:
 549  * 1. give them to another thread in our thread group, if such a member exists
 550  * 2. give it to the first ancestor process which prctl'd itself as a
 551  *    child_subreaper for its children (like a service manager)
 552  * 3. give it to the init process (PID 1) in our pid namespace
 553  */
 554 static struct task_struct *find_new_reaper(struct task_struct *father,
 555                                            struct task_struct *child_reaper)
 556 {
 557         struct task_struct *thread, *reaper;
 558
 559         thread = find_alive_thread(father);
 560         if (thread)
 561                 return thread;
 562
 563         if (father->signal->has_child_subreaper) {
 564                 unsigned int ns_level = task_pid(father)->level;
 565                 /*
 566                  * Find the first ->is_child_subreaper ancestor in our pid_ns.
 567                  * We can't check reaper != child_reaper to ensure we do not
 568                  * cross the namespaces, the exiting parent could be injected
 569                  * by setns() + fork().
 570                  * We check pid->level, this is slightly more efficient than
 571                  * task_active_pid_ns(reaper) != task_active_pid_ns(father).
 572                  */
 573                 for (reaper = father->real_parent;
 574                      task_pid(reaper)->level == ns_level;
 575                      reaper = reaper->real_parent) {
 576                         if (reaper == &init_task)
 577                                 break;
 578                         if (!reaper->signal->is_child_subreaper)
 579                                 continue;
 580                         thread = find_alive_thread(reaper);
 581                         if (thread)
 582                                 return thread;
 583                 }
 584         }
 585
 586         return child_reaper;
 587 }
 588
 589 /*
 590 * Any that need to be release_task'd are put on the @dead list.
 591  */
 592 static void reparent_leader(struct task_struct *father, struct task_struct *p,
 593                                 struct list_head *dead)
 594 {
 595         if (unlikely(p->exit_state == EXIT_DEAD))
 596                 return;
 597
 598         /* We don't want people slaying init. */
 599         p->exit_signal = SIGCHLD;
 600
 601         /* If it has exited notify the new parent about this child's death. */
 602         if (!p->ptrace &&
 603             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
 604                 if (do_notify_parent(p, p->exit_signal)) {
 605                         p->exit_state = EXIT_DEAD;
 606                         list_add(&p->ptrace_entry, dead);
 607                 }
 608         }
 609
 610         kill_orphaned_pgrp(p, father);
 611 }
 612
 613 /*
 614  * This does two things:
 615  *
 616  * A.  Make init inherit all the child processes
 617  * B.  Check to see if any process groups have become orphaned
 618  *      as a result of our exiting, and if they have any stopped
 619  *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 620  */
 621 static void forget_original_parent(struct task_struct *father,
 622                                         struct list_head *dead)
 623 {
 624         struct task_struct *p, *t, *reaper;
 625
 626         if (unlikely(!list_empty(&father->ptraced)))
 627                 exit_ptrace(father, dead);
 628
 629         /* Can drop and reacquire tasklist_lock */
 630         reaper = find_child_reaper(father, dead);
 631         if (list_empty(&father->children))
 632                 return;
 633
 634         reaper = find_new_reaper(father, reaper);
 635         list_for_each_entry(p, &father->children, sibling) {
 636                 for_each_thread(p, t) {
 637                         RCU_INIT_POINTER(t->real_parent, reaper);
 638                         BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
 639                         if (likely(!t->ptrace))
 640                                 t->parent = t->real_parent;
 641                         if (t->pdeath_signal)
 642                                 group_send_sig_info(t->pdeath_signal,
 643                                                     SEND_SIG_NOINFO, t,
 644                                                     PIDTYPE_TGID);
 645                 }
 646                 /*
 647                  * If this is a threaded reparent there is no need to
 648                  * notify anyone anything has happened.
 649                  */
 650                 if (!same_thread_group(reaper, father))
 651                         reparent_leader(father, p, dead);
 652         }
 653         list_splice_tail_init(&father->children, &reaper->children);
 654 }
 655
 656 /*
 657  * Send signals to all our closest relatives so that they know
 658  * to properly mourn us..
 659  */
 660 static void exit_notify(struct task_struct *tsk, int group_dead)
 661 {
 662         bool autoreap;
 663         struct task_struct *p, *n;
 664         LIST_HEAD(dead);
 665
 666         write_lock_irq(&tasklist_lock);
 667         forget_original_parent(tsk, &dead);
 668
 669         if (group_dead)
 670                 kill_orphaned_pgrp(tsk->group_leader, NULL);
 671
 672         tsk->exit_state = EXIT_ZOMBIE;
 673         if (unlikely(tsk->ptrace)) {
 674                 int sig = thread_group_leader(tsk) &&
 675                                 thread_group_empty(tsk) &&
 676                                 !ptrace_reparented(tsk) ?
 677                         tsk->exit_signal : SIGCHLD;
 678                 autoreap = do_notify_parent(tsk, sig);
 679         } else if (thread_group_leader(tsk)) {
 680                 autoreap = thread_group_empty(tsk) &&
 681                         do_notify_parent(tsk, tsk->exit_signal);
 682         } else {
 683                 autoreap = true;
 684         }
 685
 686         if (autoreap) {
 687                 tsk->exit_state = EXIT_DEAD;
 688                 list_add(&tsk->ptrace_entry, &dead);
 689         }
 690
 691         /* mt-exec, de_thread() is waiting for group leader */
 692         if (unlikely(tsk->signal->notify_count < 0))
 693                 wake_up_process(tsk->signal->group_exit_task);
 694         write_unlock_irq(&tasklist_lock);
 695
 696         list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
 697                 list_del_init(&p->ptrace_entry);
 698                 release_task(p);
 699         }
 700 }
 701
 702 #ifdef CONFIG_DEBUG_STACK_USAGE
 703 static void check_stack_usage(void)
 704 {
 705         static DEFINE_SPINLOCK(low_water_lock);
 706         static int lowest_to_date = THREAD_SIZE;
 707         unsigned long free;
 708
 709         free = stack_not_used(current);
 710
 711         if (free >= lowest_to_date)
 712                 return;
 713
 714         spin_lock(&low_water_lock);
 715         if (free < lowest_to_date) {
 716                 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
 717                         current->comm, task_pid_nr(current), free);
 718                 lowest_to_date = free;
 719         }
 720         spin_unlock(&low_water_lock);
 721 }
 722 #else
 723 static inline void check_stack_usage(void) {}
 724 #endif
 725
 726 void __noreturn do_exit(long code)
 727 {
 728         struct task_struct *tsk = current;
 729         int group_dead;
 730
 731         /*
 732          * We can get here from a kernel oops, sometimes with preemption off.
 733          * Start by checking for critical errors.
 734          * Then fix up important state like USER_DS and preemption.
 735          * Then do everything else.
 736          */
 737
 738         WARN_ON(blk_needs_flush_plug(tsk));
 739
 740         if (unlikely(in_interrupt()))
 741                 panic("Aiee, killing interrupt handler!");
 742         if (unlikely(!tsk->pid))
 743                 panic("Attempted to kill the idle task!");
 744
 745         /*
 746          * If do_exit is called because this processes oopsed, it's possible
 747          * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
 748          * continuing. Amongst other possible reasons, this is to prevent
 749          * mm_release()->clear_child_tid() from writing to a user-controlled
 750          * kernel address.
 751          */
 752         force_uaccess_begin();
 753
 754         if (unlikely(in_atomic())) {
 755                 pr_info("note: %s[%d] exited with preempt_count %d\n",
 756                         current->comm, task_pid_nr(current),
 757                         preempt_count());
 758                 preempt_count_set(PREEMPT_ENABLED);
 759         }
 760
 761         profile_task_exit(tsk);
 762         kcov_task_exit(tsk);
 763
 764         ptrace_event(PTRACE_EVENT_EXIT, code);
 765
 766         validate_creds_for_do_exit(tsk);
 767
 768         /*
 769          * We're taking recursive faults here in do_exit. Safest is to just
 770          * leave this task alone and wait for reboot.
 771          */
 772         if (unlikely(tsk->flags & PF_EXITING)) {
 773                 pr_alert("Fixing recursive fault but reboot is needed!\n");
 774                 futex_exit_recursive(tsk);
 775                 set_current_state(TASK_UNINTERRUPTIBLE);
 776                 schedule();
 777         }
 778
 779         exit_signals(tsk);  /* sets PF_EXITING */
 780
 781         /* sync mm's RSS info before statistics gathering */
 782         if (tsk->mm)
 783                 sync_mm_rss(tsk->mm);
 784         acct_update_integrals(tsk);
 785         group_dead = atomic_dec_and_test(&tsk->signal->live);
 786         if (group_dead) {
 787                 /*
 788                  * If the last thread of global init has exited, panic
 789                  * immediately to get a useable coredump.
 790                  */
 791                 if (unlikely(is_global_init(tsk)))
 792                         panic("Attempted to kill init! exitcode=0x%08x\n",
 793                                 tsk->signal->group_exit_code ?: (int)code);
 794
 795 #ifdef CONFIG_POSIX_TIMERS
 796                 hrtimer_cancel(&tsk->signal->real_timer);
 797                 exit_itimers(tsk->signal);
 798 #endif
 799                 if (tsk->mm)
 800                         setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
 801         }
 802         acct_collect(code, group_dead);
 803         if (group_dead)
 804                 tty_audit_exit();
 805         audit_free(tsk);
 806
 807         tsk->exit_code = code;
 808         taskstats_exit(tsk, group_dead);
 809
 810         exit_mm();
 811
 812         if (group_dead)
 813                 acct_process();
 814         trace_sched_process_exit(tsk);
 815
 816         exit_sem(tsk);
 817         exit_shm(tsk);
 818         exit_files(tsk);
 819         exit_fs(tsk);
 820         if (group_dead)
 821                 disassociate_ctty(1);
 822         exit_task_namespaces(tsk);
 823         exit_task_work(tsk);
 824         exit_thread(tsk);
 825
 826         /*
 827          * Flush inherited counters to the parent - before the parent
 828          * gets woken up by child-exit notifications.
 829          *
 830          * because of cgroup mode, must be called before cgroup_exit()
 831          */
 832         perf_event_exit_task(tsk);
 833
 834         sched_autogroup_exit_task(tsk);
 835         cgroup_exit(tsk);
 836
 837         /*
 838          * FIXME: do that only when needed, using sched_exit tracepoint
 839          */
 840         flush_ptrace_hw_breakpoint(tsk);
 841
 842         exit_tasks_rcu_start();
 843         exit_notify(tsk, group_dead);
 844         proc_exit_connector(tsk);
 845         mpol_put_task_policy(tsk);
 846 #ifdef CONFIG_FUTEX
 847         if (unlikely(current->pi_state_cache))
 848                 kfree(current->pi_state_cache);
 849 #endif
 850         /*
 851          * Make sure we are holding no locks:
 852          */
 853         debug_check_no_locks_held();
 854
 855         if (tsk->io_context)
 856                 exit_io_context(tsk);
 857
 858         if (tsk->splice_pipe)
 859                 free_pipe_info(tsk->splice_pipe);
 860
 861         if (tsk->task_frag.page)
 862                 put_page(tsk->task_frag.page);
 863
 864         validate_creds_for_do_exit(tsk);
 865
 866         check_stack_usage();
 867         preempt_disable();
 868         if (tsk->nr_dirtied)
 869                 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
 870         exit_rcu();
 871         exit_tasks_rcu_finish();
 872
 873         lockdep_free_task(tsk);
 874         do_task_dead();
 875 }
 876 EXPORT_SYMBOL_GPL(do_exit);
 877
 878 void complete_and_exit(struct completion *comp, long code)
 879 {
 880         if (comp)
 881                 complete(comp);
 882
 883         do_exit(code);
 884 }
 885 EXPORT_SYMBOL(complete_and_exit);
 886
 887 SYSCALL_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  */
 896 void
 897 do_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
 908                 spin_lock_irq(&sighand->siglock);
 909                 if (signal_group_exit(sig))
 910                         /* Another thread got here before we took the lock.  */
 911                         exit_code = sig->group_exit_code;
 912                 else {
 913                         sig->group_exit_code = exit_code;
 914                         sig->flags = SIGNAL_GROUP_EXIT;
 915                         zap_other_threads(current);
 916                 }
 917                 spin_unlock_irq(&sighand->siglock);
 918         }
 919
 920         do_exit(exit_code);
 921         /* NOTREACHED */
 922 }
 923
 924 /*
 925  * this kills every thread in the thread group. Note that any externally
 926  * wait4()-ing process will get the correct exit code - even if this
 927  * thread is not the thread group leader.
 928  */
 929 SYSCALL_DEFINE1(exit_group, int, error_code)
 930 {
 931         do_group_exit((error_code & 0xff) << 8);
 932         /* NOTREACHED */
 933         return 0;
 934 }
 935
 936 struct waitid_info {
 937         pid_t pid;
 938         uid_t uid;
 939         int status;
 940         int cause;
 941 };
 942
 943 struct wait_opts {
 944         enum pid_type           wo_type;
 945         int                     wo_flags;
 946         struct pid              *wo_pid;
 947
 948         struct waitid_info      *wo_info;
 949         int                     wo_stat;
 950         struct rusage           *wo_rusage;
 951
 952         wait_queue_entry_t              child_wait;
 953         int                     notask_error;
 954 };
 955
 956 static 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
 962 static int
 963 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
 964 {
 965         if (!eligible_pid(wo, p))
 966                 return 0;
 967
 968         /*
 969          * Wait for all children (clone and not) if __WALL is set or
 970          * if it is traced by us.
 971          */
 972         if (ptrace || (wo->wo_flags & __WALL))
 973                 return 1;
 974
 975         /*
 976          * Otherwise, wait for clone children *only* if __WCLONE is set;
 977          * otherwise, wait for non-clone children *only*.
 978          *
 979          * Note: a "clone" child here is one that reports to its parent
 980          * using a signal other than SIGCHLD, or a non-leader thread which
 981          * we can only see if it is traced by us.
 982          */
 983         if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
 984                 return 0;
 985
 986         return 1;
 987 }
 988
 989 /*
 990  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
 991  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
 992  * the lock and this task is uninteresting.  If we return nonzero, we have
 993  * released the lock and the system call should return.
 994  */
 995 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
 996 {
 997         int state, status;
 998         pid_t pid = task_pid_vnr(p);
 999         uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1000         struct waitid_info *infop;
1001
1002         if (!likely(wo->wo_flags & WEXITED))
1003                 return 0;
1004
1005         if (unlikely(wo->wo_flags & WNOWAIT)) {
1006                 status = p->exit_code;
1007                 get_task_struct(p);
1008                 read_unlock(&tasklist_lock);
1009                 sched_annotate_sleep();
1010                 if (wo->wo_rusage)
1011                         getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1012                 put_task_struct(p);
1013                 goto out_info;
1014         }
1015         /*
1016          * Move the task's state to DEAD/TRACE, only one thread can do this.
1017          */
1018         state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1019                 EXIT_TRACE : EXIT_DEAD;
1020         if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1021                 return 0;
1022         /*
1023          * We own this thread, nobody else can reap it.
1024          */
1025         read_unlock(&tasklist_lock);
1026         sched_annotate_sleep();
1027
1028         /*
1029          * Check thread_group_leader() to exclude the traced sub-threads.
1030          */
1031         if (state == EXIT_DEAD && thread_group_leader(p)) {
1032                 struct signal_struct *sig = p->signal;
1033                 struct signal_struct *psig = current->signal;
1034                 unsigned long maxrss;
1035                 u64 tgutime, tgstime;
1036
1037                 /*
1038                  * The resource counters for the group leader are in its
1039                  * own task_struct.  Those for dead threads in the group
1040                  * are in its signal_struct, as are those for the child
1041                  * processes it has previously reaped.  All these
1042                  * accumulate in the parent's signal_struct c* fields.
1043                  *
1044                  * We don't bother to take a lock here to protect these
1045                  * p->signal fields because the whole thread group is dead
1046                  * and nobody can change them.
1047                  *
1048                  * psig->stats_lock also protects us from our sub-theads
1049                  * which can reap other children at the same time. Until
1050                  * we change k_getrusage()-like users to rely on this lock
1051                  * we have to take ->siglock as well.
1052                  *
1053                  * We use thread_group_cputime_adjusted() to get times for
1054                  * the thread group, which consolidates times for all threads
1055                  * in the group including the group leader.
1056                  */
1057                 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1058                 spin_lock_irq(&current->sighand->siglock);
1059                 write_seqlock(&psig->stats_lock);
1060                 psig->cutime += tgutime + sig->cutime;
1061                 psig->cstime += tgstime + sig->cstime;
1062                 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1063                 psig->cmin_flt +=
1064                         p->min_flt + sig->min_flt + sig->cmin_flt;
1065                 psig->cmaj_flt +=
1066                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1067                 psig->cnvcsw +=
1068                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1069                 psig->cnivcsw +=
1070                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1071                 psig->cinblock +=
1072                         task_io_get_inblock(p) +
1073                         sig->inblock + sig->cinblock;
1074                 psig->coublock +=
1075                         task_io_get_oublock(p) +
1076                         sig->oublock + sig->coublock;
1077                 maxrss = max(sig->maxrss, sig->cmaxrss);
1078                 if (psig->cmaxrss < maxrss)
1079                         psig->cmaxrss = maxrss;
1080                 task_io_accounting_add(&psig->ioac, &p->ioac);
1081                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1082                 write_sequnlock(&psig->stats_lock);
1083                 spin_unlock_irq(&current->sighand->siglock);
1084         }
1085
1086         if (wo->wo_rusage)
1087                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1088         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1089                 ? p->signal->group_exit_code : p->exit_code;
1090         wo->wo_stat = status;
1091
1092         if (state == EXIT_TRACE) {
1093                 write_lock_irq(&tasklist_lock);
1094                 /* We dropped tasklist, ptracer could die and untrace */
1095                 ptrace_unlink(p);
1096
1097                 /* If parent wants a zombie, don't release it now */
1098                 state = EXIT_ZOMBIE;
1099                 if (do_notify_parent(p, p->exit_signal))
1100                         state = EXIT_DEAD;
1101                 p->exit_state = state;
1102                 write_unlock_irq(&tasklist_lock);
1103         }
1104         if (state == EXIT_DEAD)
1105                 release_task(p);
1106
1107 out_info:
1108         infop = wo->wo_info;
1109         if (infop) {
1110                 if ((status & 0x7f) == 0) {
1111                         infop->cause = CLD_EXITED;
1112                         infop->status = status >> 8;
1113                 } else {
1114                         infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1115                         infop->status = status & 0x7f;
1116                 }
1117                 infop->pid = pid;
1118                 infop->uid = uid;
1119         }
1120
1121         return pid;
1122 }
1123
1124 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1125 {
1126         if (ptrace) {
1127                 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1128                         return &p->exit_code;
1129         } else {
1130                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1131                         return &p->signal->group_exit_code;
1132         }
1133         return NULL;
1134 }
1135
1136 /**
1137  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1138  * @wo: wait options
1139  * @ptrace: is the wait for ptrace
1140  * @p: task to wait for
1141  *
1142  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1143  *
1144  * CONTEXT:
1145  * read_lock(&tasklist_lock), which is released if return value is
1146  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1147  *
1148  * RETURNS:
1149  * 0 if wait condition didn't exist and search for other wait conditions
1150  * should continue.  Non-zero return, -errno on failure and @p's pid on
1151  * success, implies that tasklist_lock is released and wait condition
1152  * search should terminate.
1153  */
1154 static int wait_task_stopped(struct wait_opts *wo,
1155                                 int ptrace, struct task_struct *p)
1156 {
1157         struct waitid_info *infop;
1158         int exit_code, *p_code, why;
1159         uid_t uid = 0; /* unneeded, required by compiler */
1160         pid_t pid;
1161
1162         /*
1163          * Traditionally we see ptrace'd stopped tasks regardless of options.
1164          */
1165         if (!ptrace && !(wo->wo_flags & WUNTRACED))
1166                 return 0;
1167
1168         if (!task_stopped_code(p, ptrace))
1169                 return 0;
1170
1171         exit_code = 0;
1172         spin_lock_irq(&p->sighand->siglock);
1173
1174         p_code = task_stopped_code(p, ptrace);
1175         if (unlikely(!p_code))
1176                 goto unlock_sig;
1177
1178         exit_code = *p_code;
1179         if (!exit_code)
1180                 goto unlock_sig;
1181
1182         if (!unlikely(wo->wo_flags & WNOWAIT))
1183                 *p_code = 0;
1184
1185         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1186 unlock_sig:
1187         spin_unlock_irq(&p->sighand->siglock);
1188         if (!exit_code)
1189                 return 0;
1190
1191         /*
1192          * Now we are pretty sure this task is interesting.
1193          * Make sure it doesn't get reaped out from under us while we
1194          * give up the lock and then examine it below.  We don't want to
1195          * keep holding onto the tasklist_lock while we call getrusage and
1196          * possibly take page faults for user memory.
1197          */
1198         get_task_struct(p);
1199         pid = task_pid_vnr(p);
1200         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1201         read_unlock(&tasklist_lock);
1202         sched_annotate_sleep();
1203         if (wo->wo_rusage)
1204                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1205         put_task_struct(p);
1206
1207         if (likely(!(wo->wo_flags & WNOWAIT)))
1208                 wo->wo_stat = (exit_code << 8) | 0x7f;
1209
1210         infop = wo->wo_info;
1211         if (infop) {
1212                 infop->cause = why;
1213                 infop->status = exit_code;
1214                 infop->pid = pid;
1215                 infop->uid = uid;
1216         }
1217         return pid;
1218 }
1219
1220 /*
1221  * Handle do_wait work for one task in a live, non-stopped state.
1222  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1223  * the lock and this task is uninteresting.  If we return nonzero, we have
1224  * released the lock and the system call should return.
1225  */
1226 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1227 {
1228         struct waitid_info *infop;
1229         pid_t pid;
1230         uid_t uid;
1231
1232         if (!unlikely(wo->wo_flags & WCONTINUED))
1233                 return 0;
1234
1235         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1236                 return 0;
1237
1238         spin_lock_irq(&p->sighand->siglock);
1239         /* Re-check with the lock held.  */
1240         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1241                 spin_unlock_irq(&p->sighand->siglock);
1242                 return 0;
1243         }
1244         if (!unlikely(wo->wo_flags & WNOWAIT))
1245                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1246         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1247         spin_unlock_irq(&p->sighand->siglock);
1248
1249         pid = task_pid_vnr(p);
1250         get_task_struct(p);
1251         read_unlock(&tasklist_lock);
1252         sched_annotate_sleep();
1253         if (wo->wo_rusage)
1254                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1255         put_task_struct(p);
1256
1257         infop = wo->wo_info;
1258         if (!infop) {
1259                 wo->wo_stat = 0xffff;
1260         } else {
1261                 infop->cause = CLD_CONTINUED;
1262                 infop->pid = pid;
1263                 infop->uid = uid;
1264                 infop->status = SIGCONT;
1265         }
1266         return pid;
1267 }
1268
1269 /*
1270  * Consider @p for a wait by @parent.
1271  *
1272  * -ECHILD should be in ->notask_error before the first call.
1273  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1274  * Returns zero if the search for a child should continue;
1275  * then ->notask_error is 0 if @p is an eligible child,
1276  * or still -ECHILD.
1277  */
1278 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1279                                 struct task_struct *p)
1280 {
1281         /*
1282          * We can race with wait_task_zombie() from another thread.
1283          * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1284          * can't confuse the checks below.
1285          */
1286         int exit_state = READ_ONCE(p->exit_state);
1287         int ret;
1288
1289         if (unlikely(exit_state == EXIT_DEAD))
1290                 return 0;
1291
1292         ret = eligible_child(wo, ptrace, p);
1293         if (!ret)
1294                 return ret;
1295
1296         if (unlikely(exit_state == EXIT_TRACE)) {
1297                 /*
1298                  * ptrace == 0 means we are the natural parent. In this case
1299                  * we should clear notask_error, debugger will notify us.
1300                  */
1301                 if (likely(!ptrace))
1302                         wo->notask_error = 0;
1303                 return 0;
1304         }
1305
1306         if (likely(!ptrace) && unlikely(p->ptrace)) {
1307                 /*
1308                  * If it is traced by its real parent's group, just pretend
1309                  * the caller is ptrace_do_wait() and reap this child if it
1310                  * is zombie.
1311                  *
1312                  * This also hides group stop state from real parent; otherwise
1313                  * a single stop can be reported twice as group and ptrace stop.
1314                  * If a ptracer wants to distinguish these two events for its
1315                  * own children it should create a separate process which takes
1316                  * the role of real parent.
1317                  */
1318                 if (!ptrace_reparented(p))
1319                         ptrace = 1;
1320         }
1321
1322         /* slay zombie? */
1323         if (exit_state == EXIT_ZOMBIE) {
1324                 /* we don't reap group leaders with subthreads */
1325                 if (!delay_group_leader(p)) {
1326                         /*
1327                          * A zombie ptracee is only visible to its ptracer.
1328                          * Notification and reaping will be cascaded to the
1329                          * real parent when the ptracer detaches.
1330                          */
1331                         if (unlikely(ptrace) || likely(!p->ptrace))
1332                                 return wait_task_zombie(wo, p);
1333                 }
1334
1335                 /*
1336                  * Allow access to stopped/continued state via zombie by
1337                  * falling through.  Clearing of notask_error is complex.
1338                  *
1339                  * When !@ptrace:
1340                  *
1341                  * If WEXITED is set, notask_error should naturally be
1342                  * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1343                  * so, if there are live subthreads, there are events to
1344                  * wait for.  If all subthreads are dead, it's still safe
1345                  * to clear - this function will be called again in finite
1346                  * amount time once all the subthreads are released and
1347                  * will then return without clearing.
1348                  *
1349                  * When @ptrace:
1350                  *
1351                  * Stopped state is per-task and thus can't change once the
1352                  * target task dies.  Only continued and exited can happen.
1353                  * Clear notask_error if WCONTINUED | WEXITED.
1354                  */
1355                 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1356                         wo->notask_error = 0;
1357         } else {
1358                 /*
1359                  * @p is alive and it's gonna stop, continue or exit, so
1360                  * there always is something to wait for.
1361                  */
1362                 wo->notask_error = 0;
1363         }
1364
1365         /*
1366          * Wait for stopped.  Depending on @ptrace, different stopped state
1367          * is used and the two don't interact with each other.
1368          */
1369         ret = wait_task_stopped(wo, ptrace, p);
1370         if (ret)
1371                 return ret;
1372
1373         /*
1374          * Wait for continued.  There's only one continued state and the
1375          * ptracer can consume it which can confuse the real parent.  Don't
1376          * use WCONTINUED from ptracer.  You don't need or want it.
1377          */
1378         return wait_task_continued(wo, p);
1379 }
1380
1381 /*
1382  * Do the work of do_wait() for one thread in the group, @tsk.
1383  *
1384  * -ECHILD should be in ->notask_error before the first call.
1385  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1386  * Returns zero if the search for a child should continue; then
1387  * ->notask_error is 0 if there were any eligible children,
1388  * or still -ECHILD.
1389  */
1390 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1391 {
1392         struct task_struct *p;
1393
1394         list_for_each_entry(p, &tsk->children, sibling) {
1395                 int ret = wait_consider_task(wo, 0, p);
1396
1397                 if (ret)
1398                         return ret;
1399         }
1400
1401         return 0;
1402 }
1403
1404 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1405 {
1406         struct task_struct *p;
1407
1408         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1409                 int ret = wait_consider_task(wo, 1, p);
1410
1411                 if (ret)
1412                         return ret;
1413         }
1414
1415         return 0;
1416 }
1417
1418 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1419                                 int sync, void *key)
1420 {
1421         struct wait_opts *wo = container_of(wait, struct wait_opts,
1422                                                 child_wait);
1423         struct task_struct *p = key;
1424
1425         if (!eligible_pid(wo, p))
1426                 return 0;
1427
1428         if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1429                 return 0;
1430
1431         return default_wake_function(wait, mode, sync, key);
1432 }
1433
1434 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1435 {
1436         __wake_up_sync_key(&parent->signal->wait_chldexit,
1437                            TASK_INTERRUPTIBLE, p);
1438 }
1439
1440 static long do_wait(struct wait_opts *wo)
1441 {
1442         struct task_struct *tsk;
1443         int retval;
1444
1445         trace_sched_process_wait(wo->wo_pid);
1446
1447         init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1448         wo->child_wait.private = current;
1449         add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1450 repeat:
1451         /*
1452          * If there is nothing that can match our criteria, just get out.
1453          * We will clear ->notask_error to zero if we see any child that
1454          * might later match our criteria, even if we are not able to reap
1455          * it yet.
1456          */
1457         wo->notask_error = -ECHILD;
1458         if ((wo->wo_type < PIDTYPE_MAX) &&
1459            (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1460                 goto notask;
1461
1462         set_current_state(TASK_INTERRUPTIBLE);
1463         read_lock(&tasklist_lock);
1464         tsk = current;
1465         do {
1466                 retval = do_wait_thread(wo, tsk);
1467                 if (retval)
1468                         goto end;
1469
1470                 retval = ptrace_do_wait(wo, tsk);
1471                 if (retval)
1472                         goto end;
1473
1474                 if (wo->wo_flags & __WNOTHREAD)
1475                         break;
1476         } while_each_thread(current, tsk);
1477         read_unlock(&tasklist_lock);
1478
1479 notask:
1480         retval = wo->notask_error;
1481         if (!retval && !(wo->wo_flags & WNOHANG)) {
1482                 retval = -ERESTARTSYS;
1483                 if (!signal_pending(current)) {
1484                         schedule();
1485                         goto repeat;
1486                 }
1487         }
1488 end:
1489         __set_current_state(TASK_RUNNING);
1490         remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1491         return retval;
1492 }
1493
1494 static 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         unsigned int f_flags = 0;
1502
1503         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1504                         __WNOTHREAD|__WCLONE|__WALL))
1505                 return -EINVAL;
1506         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1507                 return -EINVAL;
1508
1509         switch (which) {
1510         case P_ALL:
1511                 type = PIDTYPE_MAX;
1512                 break;
1513         case P_PID:
1514                 type = PIDTYPE_PID;
1515                 if (upid <= 0)
1516                         return -EINVAL;
1517
1518                 pid = find_get_pid(upid);
1519                 break;
1520         case P_PGID:
1521                 type = PIDTYPE_PGID;
1522                 if (upid < 0)
1523                         return -EINVAL;
1524
1525                 if (upid)
1526                         pid = find_get_pid(upid);
1527                 else
1528                         pid = get_task_pid(current, PIDTYPE_PGID);
1529                 break;
1530         case P_PIDFD:
1531                 type = PIDTYPE_PID;
1532                 if (upid < 0)
1533                         return -EINVAL;
1534
1535                 pid = pidfd_get_pid(upid, &f_flags);
1536                 if (IS_ERR(pid))
1537                         return PTR_ERR(pid);
1538
1539                 break;
1540         default:
1541                 return -EINVAL;
1542         }
1543
1544         wo.wo_type      = type;
1545         wo.wo_pid       = pid;
1546         wo.wo_flags     = options;
1547         wo.wo_info      = infop;
1548         wo.wo_rusage    = ru;
1549         if (f_flags & O_NONBLOCK)
1550                 wo.wo_flags |= WNOHANG;
1551
1552         ret = do_wait(&wo);
1553         if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1554                 ret = -EAGAIN;
1555
1556         put_pid(pid);
1557         return ret;
1558 }
1559
1560 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1561                 infop, int, options, struct rusage __user *, ru)
1562 {
1563         struct rusage r;
1564         struct waitid_info info = {.status = 0};
1565         long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1566         int signo = 0;
1567
1568         if (err > 0) {
1569                 signo = SIGCHLD;
1570                 err = 0;
1571                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1572                         return -EFAULT;
1573         }
1574         if (!infop)
1575                 return err;
1576
1577         if (!user_write_access_begin(infop, sizeof(*infop)))
1578                 return -EFAULT;
1579
1580         unsafe_put_user(signo, &infop->si_signo, Efault);
1581         unsafe_put_user(0, &infop->si_errno, Efault);
1582         unsafe_put_user(info.cause, &infop->si_code, Efault);
1583         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1584         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1585         unsafe_put_user(info.status, &infop->si_status, Efault);
1586         user_write_access_end();
1587         return err;
1588 Efault:
1589         user_write_access_end();
1590         return -EFAULT;
1591 }
1592
1593 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1594                   struct rusage *ru)
1595 {
1596         struct wait_opts wo;
1597         struct pid *pid = NULL;
1598         enum pid_type type;
1599         long ret;
1600
1601         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1602                         __WNOTHREAD|__WCLONE|__WALL))
1603                 return -EINVAL;
1604
1605         /* -INT_MIN is not defined */
1606         if (upid == INT_MIN)
1607                 return -ESRCH;
1608
1609         if (upid == -1)
1610                 type = PIDTYPE_MAX;
1611         else if (upid < 0) {
1612                 type = PIDTYPE_PGID;
1613                 pid = find_get_pid(-upid);
1614         } else if (upid == 0) {
1615                 type = PIDTYPE_PGID;
1616                 pid = get_task_pid(current, PIDTYPE_PGID);
1617         } else /* upid > 0 */ {
1618                 type = PIDTYPE_PID;
1619                 pid = find_get_pid(upid);
1620         }
1621
1622         wo.wo_type      = type;
1623         wo.wo_pid       = pid;
1624         wo.wo_flags     = options | WEXITED;
1625         wo.wo_info      = NULL;
1626         wo.wo_stat      = 0;
1627         wo.wo_rusage    = ru;
1628         ret = do_wait(&wo);
1629         put_pid(pid);
1630         if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1631                 ret = -EFAULT;
1632
1633         return ret;
1634 }
1635
1636 int kernel_wait(pid_t pid, int *stat)
1637 {
1638         struct wait_opts wo = {
1639                 .wo_type        = PIDTYPE_PID,
1640                 .wo_pid         = find_get_pid(pid),
1641                 .wo_flags       = WEXITED,
1642         };
1643         int ret;
1644
1645         ret = do_wait(&wo);
1646         if (ret > 0 && wo.wo_stat)
1647                 *stat = wo.wo_stat;
1648         put_pid(wo.wo_pid);
1649         return ret;
1650 }
1651
1652 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1653                 int, options, struct rusage __user *, ru)
1654 {
1655         struct rusage r;
1656         long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1657
1658         if (err > 0) {
1659                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1660                         return -EFAULT;
1661         }
1662         return err;
1663 }
1664
1665 #ifdef __ARCH_WANT_SYS_WAITPID
1666
1667 /*
1668  * sys_waitpid() remains for compatibility. waitpid() should be
1669  * implemented by calling sys_wait4() from libc.a.
1670  */
1671 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1672 {
1673         return kernel_wait4(pid, stat_addr, options, NULL);
1674 }
1675
1676 #endif
1677
1678 #ifdef CONFIG_COMPAT
1679 COMPAT_SYSCALL_DEFINE4(wait4,
1680         compat_pid_t, pid,
1681         compat_uint_t __user *, stat_addr,
1682         int, options,
1683         struct compat_rusage __user *, ru)
1684 {
1685         struct rusage r;
1686         long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1687         if (err > 0) {
1688                 if (ru && put_compat_rusage(&r, ru))
1689                         return -EFAULT;
1690         }
1691         return err;
1692 }
1693
1694 COMPAT_SYSCALL_DEFINE5(waitid,
1695                 int, which, compat_pid_t, pid,
1696                 struct compat_siginfo __user *, infop, int, options,
1697                 struct compat_rusage __user *, uru)
1698 {
1699         struct rusage ru;
1700         struct waitid_info info = {.status = 0};
1701         long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1702         int signo = 0;
1703         if (err > 0) {
1704                 signo = SIGCHLD;
1705                 err = 0;
1706                 if (uru) {
1707                         /* kernel_waitid() overwrites everything in ru */
1708                         if (COMPAT_USE_64BIT_TIME)
1709                                 err = copy_to_user(uru, &ru, sizeof(ru));
1710                         else
1711                                 err = put_compat_rusage(&ru, uru);
1712                         if (err)
1713                                 return -EFAULT;
1714                 }
1715         }
1716
1717         if (!infop)
1718                 return err;
1719
1720         if (!user_write_access_begin(infop, sizeof(*infop)))
1721                 return -EFAULT;
1722
1723         unsafe_put_user(signo, &infop->si_signo, Efault);
1724         unsafe_put_user(0, &infop->si_errno, Efault);
1725         unsafe_put_user(info.cause, &infop->si_code, Efault);
1726         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1727         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1728         unsafe_put_user(info.status, &infop->si_status, Efault);
1729         user_write_access_end();
1730         return err;
1731 Efault:
1732         user_write_access_end();
1733         return -EFAULT;
1734 }
1735 #endif
1736
1737 /**
1738  * thread_group_exited - check that a thread group has exited
1739  * @pid: tgid of thread group to be checked.
1740  *
1741  * Test if the thread group represented by tgid has exited (all
1742  * threads are zombies, dead or completely gone).
1743  *
1744  * Return: true if the thread group has exited. false otherwise.
1745  */
1746 bool thread_group_exited(struct pid *pid)
1747 {
1748         struct task_struct *task;
1749         bool exited;
1750
1751         rcu_read_lock();
1752         task = pid_task(pid, PIDTYPE_PID);
1753         exited = !task ||
1754                 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1755         rcu_read_unlock();
1756
1757         return exited;
1758 }
1759 EXPORT_SYMBOL(thread_group_exited);
1760
1761 __weak void abort(void)
1762 {
1763         BUG();
1764
1765         /* if that doesn't kill us, halt */
1766         panic("Oops failed to kill thread");
1767 }
1768 EXPORT_SYMBOL(abort);