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/pgtable.h>
  70 #include <asm/mmu_context.h>
  71
  72 static void __unhash_process(struct task_struct *p, bool group_dead)
  73 {
  74         nr_threads--;
  75         detach_pid(p, PIDTYPE_PID);
  76         if (group_dead) {
  77                 detach_pid(p, PIDTYPE_TGID);
  78                 detach_pid(p, PIDTYPE_PGID);
  79                 detach_pid(p, PIDTYPE_SID);
  80
  81                 list_del_rcu(&p->tasks);
  82                 list_del_init(&p->sibling);
  83                 __this_cpu_dec(process_counts);
  84         }
  85         list_del_rcu(&p->thread_group);
  86         list_del_rcu(&p->thread_node);
  87 }
  88
  89 /*
  90  * This function expects the tasklist_lock write-locked.
  91  */
  92 static void __exit_signal(struct task_struct *tsk)
  93 {
  94         struct signal_struct *sig = tsk->signal;
  95         bool group_dead = thread_group_leader(tsk);
  96         struct sighand_struct *sighand;
  97         struct tty_struct *uninitialized_var(tty);
  98         u64 utime, stime;
  99
 100         sighand = rcu_dereference_check(tsk->sighand,
 101                                         lockdep_tasklist_lock_is_held());
 102         spin_lock(&sighand->siglock);
 103
 104 #ifdef CONFIG_POSIX_TIMERS
 105         posix_cpu_timers_exit(tsk);
 106         if (group_dead)
 107                 posix_cpu_timers_exit_group(tsk);
 108 #endif
 109
 110         if (group_dead) {
 111                 tty = sig->tty;
 112                 sig->tty = NULL;
 113         } else {
 114                 /*
 115                  * If there is any task waiting for the group exit
 116                  * then notify it:
 117                  */
 118                 if (sig->notify_count > 0 && !--sig->notify_count)
 119                         wake_up_process(sig->group_exit_task);
 120
 121                 if (tsk == sig->curr_target)
 122                         sig->curr_target = next_thread(tsk);
 123         }
 124
 125         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
 126                               sizeof(unsigned long long));
 127
 128         /*
 129          * Accumulate here the counters for all threads as they die. We could
 130          * skip the group leader because it is the last user of signal_struct,
 131          * but we want to avoid the race with thread_group_cputime() which can
 132          * see the empty ->thread_head list.
 133          */
 134         task_cputime(tsk, &utime, &stime);
 135         write_seqlock(&sig->stats_lock);
 136         sig->utime += utime;
 137         sig->stime += stime;
 138         sig->gtime += task_gtime(tsk);
 139         sig->min_flt += tsk->min_flt;
 140         sig->maj_flt += tsk->maj_flt;
 141         sig->nvcsw += tsk->nvcsw;
 142         sig->nivcsw += tsk->nivcsw;
 143         sig->inblock += task_io_get_inblock(tsk);
 144         sig->oublock += task_io_get_oublock(tsk);
 145         task_io_accounting_add(&sig->ioac, &tsk->ioac);
 146         sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
 147         sig->nr_threads--;
 148         __unhash_process(tsk, group_dead);
 149         write_sequnlock(&sig->stats_lock);
 150
 151         /*
 152          * Do this under ->siglock, we can race with another thread
 153          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
 154          */
 155         flush_sigqueue(&tsk->pending);
 156         tsk->sighand = NULL;
 157         spin_unlock(&sighand->siglock);
 158
 159         __cleanup_sighand(sighand);
 160         clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
 161         if (group_dead) {
 162                 flush_sigqueue(&sig->shared_pending);
 163                 tty_kref_put(tty);
 164         }
 165 }
 166
 167 static void delayed_put_task_struct(struct rcu_head *rhp)
 168 {
 169         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
 170
 171         perf_event_delayed_put(tsk);
 172         trace_sched_process_free(tsk);
 173         put_task_struct(tsk);
 174 }
 175
 176 void put_task_struct_rcu_user(struct task_struct *task)
 177 {
 178         if (refcount_dec_and_test(&task->rcu_users))
 179                 call_rcu(&task->rcu, delayed_put_task_struct);
 180 }
 181
 182 void release_task(struct task_struct *p)
 183 {
 184         struct task_struct *leader;
 185         struct pid *thread_pid;
 186         int zap_leader;
 187 repeat:
 188         /* don't need to get the RCU readlock here - the process is dead and
 189          * can't be modifying its own credentials. But shut RCU-lockdep up */
 190         rcu_read_lock();
 191         atomic_dec(&__task_cred(p)->user->processes);
 192         rcu_read_unlock();
 193
 194         cgroup_release(p);
 195
 196         write_lock_irq(&tasklist_lock);
 197         ptrace_release_task(p);
 198         thread_pid = get_pid(p->thread_pid);
 199         __exit_signal(p);
 200
 201         /*
 202          * If we are the last non-leader member of the thread
 203          * group, and the leader is zombie, then notify the
 204          * group leader's parent process. (if it wants notification.)
 205          */
 206         zap_leader = 0;
 207         leader = p->group_leader;
 208         if (leader != p && thread_group_empty(leader)
 209                         && leader->exit_state == EXIT_ZOMBIE) {
 210                 /*
 211                  * If we were the last child thread and the leader has
 212                  * exited already, and the leader's parent ignores SIGCHLD,
 213                  * then we are the one who should release the leader.
 214                  */
 215                 zap_leader = do_notify_parent(leader, leader->exit_signal);
 216                 if (zap_leader)
 217                         leader->exit_state = EXIT_DEAD;
 218         }
 219
 220         write_unlock_irq(&tasklist_lock);
 221         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_sem 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         down_read(&mm->mmap_sem);
 450         core_state = mm->core_state;
 451         if (core_state) {
 452                 struct core_thread self;
 453
 454                 up_read(&mm->mmap_sem);
 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                 down_read(&mm->mmap_sem);
 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         up_read(&mm->mmap_sem);
 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
 488 static 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
 499 static 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  */
 537 static 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  */
 575 static 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  */
 604 static 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  */
 643 static 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
 686 static 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
 706 static inline void check_stack_usage(void) {}
 707 #endif
 708
 709 void __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         set_fs(USER_DS);
 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         exit_umh(tsk);
 809
 810         /*
 811          * Flush inherited counters to the parent - before the parent
 812          * gets woken up by child-exit notifications.
 813          *
 814          * because of cgroup mode, must be called before cgroup_exit()
 815          */
 816         perf_event_exit_task(tsk);
 817
 818         sched_autogroup_exit_task(tsk);
 819         cgroup_exit(tsk);
 820
 821         /*
 822          * FIXME: do that only when needed, using sched_exit tracepoint
 823          */
 824         flush_ptrace_hw_breakpoint(tsk);
 825
 826         exit_tasks_rcu_start();
 827         exit_notify(tsk, group_dead);
 828         proc_exit_connector(tsk);
 829         mpol_put_task_policy(tsk);
 830 #ifdef CONFIG_FUTEX
 831         if (unlikely(current->pi_state_cache))
 832                 kfree(current->pi_state_cache);
 833 #endif
 834         /*
 835          * Make sure we are holding no locks:
 836          */
 837         debug_check_no_locks_held();
 838
 839         if (tsk->io_context)
 840                 exit_io_context(tsk);
 841
 842         if (tsk->splice_pipe)
 843                 free_pipe_info(tsk->splice_pipe);
 844
 845         if (tsk->task_frag.page)
 846                 put_page(tsk->task_frag.page);
 847
 848         validate_creds_for_do_exit(tsk);
 849
 850         check_stack_usage();
 851         preempt_disable();
 852         if (tsk->nr_dirtied)
 853                 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
 854         exit_rcu();
 855         exit_tasks_rcu_finish();
 856
 857         lockdep_free_task(tsk);
 858         do_task_dead();
 859 }
 860 EXPORT_SYMBOL_GPL(do_exit);
 861
 862 void complete_and_exit(struct completion *comp, long code)
 863 {
 864         if (comp)
 865                 complete(comp);
 866
 867         do_exit(code);
 868 }
 869 EXPORT_SYMBOL(complete_and_exit);
 870
 871 SYSCALL_DEFINE1(exit, int, error_code)
 872 {
 873         do_exit((error_code&0xff)<<8);
 874 }
 875
 876 /*
 877  * Take down every thread in the group.  This is called by fatal signals
 878  * as well as by sys_exit_group (below).
 879  */
 880 void
 881 do_group_exit(int exit_code)
 882 {
 883         struct signal_struct *sig = current->signal;
 884
 885         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 886
 887         if (signal_group_exit(sig))
 888                 exit_code = sig->group_exit_code;
 889         else if (!thread_group_empty(current)) {
 890                 struct sighand_struct *const sighand = current->sighand;
 891
 892                 spin_lock_irq(&sighand->siglock);
 893                 if (signal_group_exit(sig))
 894                         /* Another thread got here before we took the lock.  */
 895                         exit_code = sig->group_exit_code;
 896                 else {
 897                         sig->group_exit_code = exit_code;
 898                         sig->flags = SIGNAL_GROUP_EXIT;
 899                         zap_other_threads(current);
 900                 }
 901                 spin_unlock_irq(&sighand->siglock);
 902         }
 903
 904         do_exit(exit_code);
 905         /* NOTREACHED */
 906 }
 907
 908 /*
 909  * this kills every thread in the thread group. Note that any externally
 910  * wait4()-ing process will get the correct exit code - even if this
 911  * thread is not the thread group leader.
 912  */
 913 SYSCALL_DEFINE1(exit_group, int, error_code)
 914 {
 915         do_group_exit((error_code & 0xff) << 8);
 916         /* NOTREACHED */
 917         return 0;
 918 }
 919
 920 struct waitid_info {
 921         pid_t pid;
 922         uid_t uid;
 923         int status;
 924         int cause;
 925 };
 926
 927 struct wait_opts {
 928         enum pid_type           wo_type;
 929         int                     wo_flags;
 930         struct pid              *wo_pid;
 931
 932         struct waitid_info      *wo_info;
 933         int                     wo_stat;
 934         struct rusage           *wo_rusage;
 935
 936         wait_queue_entry_t              child_wait;
 937         int                     notask_error;
 938 };
 939
 940 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
 941 {
 942         return  wo->wo_type == PIDTYPE_MAX ||
 943                 task_pid_type(p, wo->wo_type) == wo->wo_pid;
 944 }
 945
 946 static int
 947 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
 948 {
 949         if (!eligible_pid(wo, p))
 950                 return 0;
 951
 952         /*
 953          * Wait for all children (clone and not) if __WALL is set or
 954          * if it is traced by us.
 955          */
 956         if (ptrace || (wo->wo_flags & __WALL))
 957                 return 1;
 958
 959         /*
 960          * Otherwise, wait for clone children *only* if __WCLONE is set;
 961          * otherwise, wait for non-clone children *only*.
 962          *
 963          * Note: a "clone" child here is one that reports to its parent
 964          * using a signal other than SIGCHLD, or a non-leader thread which
 965          * we can only see if it is traced by us.
 966          */
 967         if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
 968                 return 0;
 969
 970         return 1;
 971 }
 972
 973 /*
 974  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
 975  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
 976  * the lock and this task is uninteresting.  If we return nonzero, we have
 977  * released the lock and the system call should return.
 978  */
 979 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
 980 {
 981         int state, status;
 982         pid_t pid = task_pid_vnr(p);
 983         uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
 984         struct waitid_info *infop;
 985
 986         if (!likely(wo->wo_flags & WEXITED))
 987                 return 0;
 988
 989         if (unlikely(wo->wo_flags & WNOWAIT)) {
 990                 status = p->exit_code;
 991                 get_task_struct(p);
 992                 read_unlock(&tasklist_lock);
 993                 sched_annotate_sleep();
 994                 if (wo->wo_rusage)
 995                         getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
 996                 put_task_struct(p);
 997                 goto out_info;
 998         }
 999         /*
1000          * Move the task's state to DEAD/TRACE, only one thread can do this.
1001          */
1002         state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1003                 EXIT_TRACE : EXIT_DEAD;
1004         if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1005                 return 0;
1006         /*
1007          * We own this thread, nobody else can reap it.
1008          */
1009         read_unlock(&tasklist_lock);
1010         sched_annotate_sleep();
1011
1012         /*
1013          * Check thread_group_leader() to exclude the traced sub-threads.
1014          */
1015         if (state == EXIT_DEAD && thread_group_leader(p)) {
1016                 struct signal_struct *sig = p->signal;
1017                 struct signal_struct *psig = current->signal;
1018                 unsigned long maxrss;
1019                 u64 tgutime, tgstime;
1020
1021                 /*
1022                  * The resource counters for the group leader are in its
1023                  * own task_struct.  Those for dead threads in the group
1024                  * are in its signal_struct, as are those for the child
1025                  * processes it has previously reaped.  All these
1026                  * accumulate in the parent's signal_struct c* fields.
1027                  *
1028                  * We don't bother to take a lock here to protect these
1029                  * p->signal fields because the whole thread group is dead
1030                  * and nobody can change them.
1031                  *
1032                  * psig->stats_lock also protects us from our sub-theads
1033                  * which can reap other children at the same time. Until
1034                  * we change k_getrusage()-like users to rely on this lock
1035                  * we have to take ->siglock as well.
1036                  *
1037                  * We use thread_group_cputime_adjusted() to get times for
1038                  * the thread group, which consolidates times for all threads
1039                  * in the group including the group leader.
1040                  */
1041                 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1042                 spin_lock_irq(&current->sighand->siglock);
1043                 write_seqlock(&psig->stats_lock);
1044                 psig->cutime += tgutime + sig->cutime;
1045                 psig->cstime += tgstime + sig->cstime;
1046                 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1047                 psig->cmin_flt +=
1048                         p->min_flt + sig->min_flt + sig->cmin_flt;
1049                 psig->cmaj_flt +=
1050                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1051                 psig->cnvcsw +=
1052                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1053                 psig->cnivcsw +=
1054                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1055                 psig->cinblock +=
1056                         task_io_get_inblock(p) +
1057                         sig->inblock + sig->cinblock;
1058                 psig->coublock +=
1059                         task_io_get_oublock(p) +
1060                         sig->oublock + sig->coublock;
1061                 maxrss = max(sig->maxrss, sig->cmaxrss);
1062                 if (psig->cmaxrss < maxrss)
1063                         psig->cmaxrss = maxrss;
1064                 task_io_accounting_add(&psig->ioac, &p->ioac);
1065                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1066                 write_sequnlock(&psig->stats_lock);
1067                 spin_unlock_irq(&current->sighand->siglock);
1068         }
1069
1070         if (wo->wo_rusage)
1071                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1072         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1073                 ? p->signal->group_exit_code : p->exit_code;
1074         wo->wo_stat = status;
1075
1076         if (state == EXIT_TRACE) {
1077                 write_lock_irq(&tasklist_lock);
1078                 /* We dropped tasklist, ptracer could die and untrace */
1079                 ptrace_unlink(p);
1080
1081                 /* If parent wants a zombie, don't release it now */
1082                 state = EXIT_ZOMBIE;
1083                 if (do_notify_parent(p, p->exit_signal))
1084                         state = EXIT_DEAD;
1085                 p->exit_state = state;
1086                 write_unlock_irq(&tasklist_lock);
1087         }
1088         if (state == EXIT_DEAD)
1089                 release_task(p);
1090
1091 out_info:
1092         infop = wo->wo_info;
1093         if (infop) {
1094                 if ((status & 0x7f) == 0) {
1095                         infop->cause = CLD_EXITED;
1096                         infop->status = status >> 8;
1097                 } else {
1098                         infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1099                         infop->status = status & 0x7f;
1100                 }
1101                 infop->pid = pid;
1102                 infop->uid = uid;
1103         }
1104
1105         return pid;
1106 }
1107
1108 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1109 {
1110         if (ptrace) {
1111                 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1112                         return &p->exit_code;
1113         } else {
1114                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1115                         return &p->signal->group_exit_code;
1116         }
1117         return NULL;
1118 }
1119
1120 /**
1121  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1122  * @wo: wait options
1123  * @ptrace: is the wait for ptrace
1124  * @p: task to wait for
1125  *
1126  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1127  *
1128  * CONTEXT:
1129  * read_lock(&tasklist_lock), which is released if return value is
1130  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1131  *
1132  * RETURNS:
1133  * 0 if wait condition didn't exist and search for other wait conditions
1134  * should continue.  Non-zero return, -errno on failure and @p's pid on
1135  * success, implies that tasklist_lock is released and wait condition
1136  * search should terminate.
1137  */
1138 static int wait_task_stopped(struct wait_opts *wo,
1139                                 int ptrace, struct task_struct *p)
1140 {
1141         struct waitid_info *infop;
1142         int exit_code, *p_code, why;
1143         uid_t uid = 0; /* unneeded, required by compiler */
1144         pid_t pid;
1145
1146         /*
1147          * Traditionally we see ptrace'd stopped tasks regardless of options.
1148          */
1149         if (!ptrace && !(wo->wo_flags & WUNTRACED))
1150                 return 0;
1151
1152         if (!task_stopped_code(p, ptrace))
1153                 return 0;
1154
1155         exit_code = 0;
1156         spin_lock_irq(&p->sighand->siglock);
1157
1158         p_code = task_stopped_code(p, ptrace);
1159         if (unlikely(!p_code))
1160                 goto unlock_sig;
1161
1162         exit_code = *p_code;
1163         if (!exit_code)
1164                 goto unlock_sig;
1165
1166         if (!unlikely(wo->wo_flags & WNOWAIT))
1167                 *p_code = 0;
1168
1169         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1170 unlock_sig:
1171         spin_unlock_irq(&p->sighand->siglock);
1172         if (!exit_code)
1173                 return 0;
1174
1175         /*
1176          * Now we are pretty sure this task is interesting.
1177          * Make sure it doesn't get reaped out from under us while we
1178          * give up the lock and then examine it below.  We don't want to
1179          * keep holding onto the tasklist_lock while we call getrusage and
1180          * possibly take page faults for user memory.
1181          */
1182         get_task_struct(p);
1183         pid = task_pid_vnr(p);
1184         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1185         read_unlock(&tasklist_lock);
1186         sched_annotate_sleep();
1187         if (wo->wo_rusage)
1188                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1189         put_task_struct(p);
1190
1191         if (likely(!(wo->wo_flags & WNOWAIT)))
1192                 wo->wo_stat = (exit_code << 8) | 0x7f;
1193
1194         infop = wo->wo_info;
1195         if (infop) {
1196                 infop->cause = why;
1197                 infop->status = exit_code;
1198                 infop->pid = pid;
1199                 infop->uid = uid;
1200         }
1201         return pid;
1202 }
1203
1204 /*
1205  * Handle do_wait work for one task in a live, non-stopped state.
1206  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1207  * the lock and this task is uninteresting.  If we return nonzero, we have
1208  * released the lock and the system call should return.
1209  */
1210 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1211 {
1212         struct waitid_info *infop;
1213         pid_t pid;
1214         uid_t uid;
1215
1216         if (!unlikely(wo->wo_flags & WCONTINUED))
1217                 return 0;
1218
1219         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1220                 return 0;
1221
1222         spin_lock_irq(&p->sighand->siglock);
1223         /* Re-check with the lock held.  */
1224         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1225                 spin_unlock_irq(&p->sighand->siglock);
1226                 return 0;
1227         }
1228         if (!unlikely(wo->wo_flags & WNOWAIT))
1229                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1230         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1231         spin_unlock_irq(&p->sighand->siglock);
1232
1233         pid = task_pid_vnr(p);
1234         get_task_struct(p);
1235         read_unlock(&tasklist_lock);
1236         sched_annotate_sleep();
1237         if (wo->wo_rusage)
1238                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1239         put_task_struct(p);
1240
1241         infop = wo->wo_info;
1242         if (!infop) {
1243                 wo->wo_stat = 0xffff;
1244         } else {
1245                 infop->cause = CLD_CONTINUED;
1246                 infop->pid = pid;
1247                 infop->uid = uid;
1248                 infop->status = SIGCONT;
1249         }
1250         return pid;
1251 }
1252
1253 /*
1254  * Consider @p for a wait by @parent.
1255  *
1256  * -ECHILD should be in ->notask_error before the first call.
1257  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1258  * Returns zero if the search for a child should continue;
1259  * then ->notask_error is 0 if @p is an eligible child,
1260  * or still -ECHILD.
1261  */
1262 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1263                                 struct task_struct *p)
1264 {
1265         /*
1266          * We can race with wait_task_zombie() from another thread.
1267          * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1268          * can't confuse the checks below.
1269          */
1270         int exit_state = READ_ONCE(p->exit_state);
1271         int ret;
1272
1273         if (unlikely(exit_state == EXIT_DEAD))
1274                 return 0;
1275
1276         ret = eligible_child(wo, ptrace, p);
1277         if (!ret)
1278                 return ret;
1279
1280         if (unlikely(exit_state == EXIT_TRACE)) {
1281                 /*
1282                  * ptrace == 0 means we are the natural parent. In this case
1283                  * we should clear notask_error, debugger will notify us.
1284                  */
1285                 if (likely(!ptrace))
1286                         wo->notask_error = 0;
1287                 return 0;
1288         }
1289
1290         if (likely(!ptrace) && unlikely(p->ptrace)) {
1291                 /*
1292                  * If it is traced by its real parent's group, just pretend
1293                  * the caller is ptrace_do_wait() and reap this child if it
1294                  * is zombie.
1295                  *
1296                  * This also hides group stop state from real parent; otherwise
1297                  * a single stop can be reported twice as group and ptrace stop.
1298                  * If a ptracer wants to distinguish these two events for its
1299                  * own children it should create a separate process which takes
1300                  * the role of real parent.
1301                  */
1302                 if (!ptrace_reparented(p))
1303                         ptrace = 1;
1304         }
1305
1306         /* slay zombie? */
1307         if (exit_state == EXIT_ZOMBIE) {
1308                 /* we don't reap group leaders with subthreads */
1309                 if (!delay_group_leader(p)) {
1310                         /*
1311                          * A zombie ptracee is only visible to its ptracer.
1312                          * Notification and reaping will be cascaded to the
1313                          * real parent when the ptracer detaches.
1314                          */
1315                         if (unlikely(ptrace) || likely(!p->ptrace))
1316                                 return wait_task_zombie(wo, p);
1317                 }
1318
1319                 /*
1320                  * Allow access to stopped/continued state via zombie by
1321                  * falling through.  Clearing of notask_error is complex.
1322                  *
1323                  * When !@ptrace:
1324                  *
1325                  * If WEXITED is set, notask_error should naturally be
1326                  * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1327                  * so, if there are live subthreads, there are events to
1328                  * wait for.  If all subthreads are dead, it's still safe
1329                  * to clear - this function will be called again in finite
1330                  * amount time once all the subthreads are released and
1331                  * will then return without clearing.
1332                  *
1333                  * When @ptrace:
1334                  *
1335                  * Stopped state is per-task and thus can't change once the
1336                  * target task dies.  Only continued and exited can happen.
1337                  * Clear notask_error if WCONTINUED | WEXITED.
1338                  */
1339                 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1340                         wo->notask_error = 0;
1341         } else {
1342                 /*
1343                  * @p is alive and it's gonna stop, continue or exit, so
1344                  * there always is something to wait for.
1345                  */
1346                 wo->notask_error = 0;
1347         }
1348
1349         /*
1350          * Wait for stopped.  Depending on @ptrace, different stopped state
1351          * is used and the two don't interact with each other.
1352          */
1353         ret = wait_task_stopped(wo, ptrace, p);
1354         if (ret)
1355                 return ret;
1356
1357         /*
1358          * Wait for continued.  There's only one continued state and the
1359          * ptracer can consume it which can confuse the real parent.  Don't
1360          * use WCONTINUED from ptracer.  You don't need or want it.
1361          */
1362         return wait_task_continued(wo, p);
1363 }
1364
1365 /*
1366  * Do the work of do_wait() for one thread in the group, @tsk.
1367  *
1368  * -ECHILD should be in ->notask_error before the first call.
1369  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1370  * Returns zero if the search for a child should continue; then
1371  * ->notask_error is 0 if there were any eligible children,
1372  * or still -ECHILD.
1373  */
1374 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1375 {
1376         struct task_struct *p;
1377
1378         list_for_each_entry(p, &tsk->children, sibling) {
1379                 int ret = wait_consider_task(wo, 0, p);
1380
1381                 if (ret)
1382                         return ret;
1383         }
1384
1385         return 0;
1386 }
1387
1388 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1389 {
1390         struct task_struct *p;
1391
1392         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1393                 int ret = wait_consider_task(wo, 1, p);
1394
1395                 if (ret)
1396                         return ret;
1397         }
1398
1399         return 0;
1400 }
1401
1402 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1403                                 int sync, void *key)
1404 {
1405         struct wait_opts *wo = container_of(wait, struct wait_opts,
1406                                                 child_wait);
1407         struct task_struct *p = key;
1408
1409         if (!eligible_pid(wo, p))
1410                 return 0;
1411
1412         if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1413                 return 0;
1414
1415         return default_wake_function(wait, mode, sync, key);
1416 }
1417
1418 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1419 {
1420         __wake_up_sync_key(&parent->signal->wait_chldexit,
1421                            TASK_INTERRUPTIBLE, p);
1422 }
1423
1424 static long do_wait(struct wait_opts *wo)
1425 {
1426         struct task_struct *tsk;
1427         int retval;
1428
1429         trace_sched_process_wait(wo->wo_pid);
1430
1431         init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1432         wo->child_wait.private = current;
1433         add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1434 repeat:
1435         /*
1436          * If there is nothing that can match our criteria, just get out.
1437          * We will clear ->notask_error to zero if we see any child that
1438          * might later match our criteria, even if we are not able to reap
1439          * it yet.
1440          */
1441         wo->notask_error = -ECHILD;
1442         if ((wo->wo_type < PIDTYPE_MAX) &&
1443            (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1444                 goto notask;
1445
1446         set_current_state(TASK_INTERRUPTIBLE);
1447         read_lock(&tasklist_lock);
1448         tsk = current;
1449         do {
1450                 retval = do_wait_thread(wo, tsk);
1451                 if (retval)
1452                         goto end;
1453
1454                 retval = ptrace_do_wait(wo, tsk);
1455                 if (retval)
1456                         goto end;
1457
1458                 if (wo->wo_flags & __WNOTHREAD)
1459                         break;
1460         } while_each_thread(current, tsk);
1461         read_unlock(&tasklist_lock);
1462
1463 notask:
1464         retval = wo->notask_error;
1465         if (!retval && !(wo->wo_flags & WNOHANG)) {
1466                 retval = -ERESTARTSYS;
1467                 if (!signal_pending(current)) {
1468                         schedule();
1469                         goto repeat;
1470                 }
1471         }
1472 end:
1473         __set_current_state(TASK_RUNNING);
1474         remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1475         return retval;
1476 }
1477
1478 static struct pid *pidfd_get_pid(unsigned int fd)
1479 {
1480         struct fd f;
1481         struct pid *pid;
1482
1483         f = fdget(fd);
1484         if (!f.file)
1485                 return ERR_PTR(-EBADF);
1486
1487         pid = pidfd_pid(f.file);
1488         if (!IS_ERR(pid))
1489                 get_pid(pid);
1490
1491         fdput(f);
1492         return pid;
1493 }
1494
1495 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1496                           int options, struct rusage *ru)
1497 {
1498         struct wait_opts wo;
1499         struct pid *pid = NULL;
1500         enum pid_type type;
1501         long ret;
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);
1536                 if (IS_ERR(pid))
1537                         return PTR_ERR(pid);
1538                 break;
1539         default:
1540                 return -EINVAL;
1541         }
1542
1543         wo.wo_type      = type;
1544         wo.wo_pid       = pid;
1545         wo.wo_flags     = options;
1546         wo.wo_info      = infop;
1547         wo.wo_rusage    = ru;
1548         ret = do_wait(&wo);
1549
1550         put_pid(pid);
1551         return ret;
1552 }
1553
1554 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1555                 infop, int, options, struct rusage __user *, ru)
1556 {
1557         struct rusage r;
1558         struct waitid_info info = {.status = 0};
1559         long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1560         int signo = 0;
1561
1562         if (err > 0) {
1563                 signo = SIGCHLD;
1564                 err = 0;
1565                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1566                         return -EFAULT;
1567         }
1568         if (!infop)
1569                 return err;
1570
1571         if (!user_write_access_begin(infop, sizeof(*infop)))
1572                 return -EFAULT;
1573
1574         unsafe_put_user(signo, &infop->si_signo, Efault);
1575         unsafe_put_user(0, &infop->si_errno, Efault);
1576         unsafe_put_user(info.cause, &infop->si_code, Efault);
1577         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1578         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1579         unsafe_put_user(info.status, &infop->si_status, Efault);
1580         user_write_access_end();
1581         return err;
1582 Efault:
1583         user_write_access_end();
1584         return -EFAULT;
1585 }
1586
1587 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1588                   struct rusage *ru)
1589 {
1590         struct wait_opts wo;
1591         struct pid *pid = NULL;
1592         enum pid_type type;
1593         long ret;
1594
1595         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1596                         __WNOTHREAD|__WCLONE|__WALL))
1597                 return -EINVAL;
1598
1599         /* -INT_MIN is not defined */
1600         if (upid == INT_MIN)
1601                 return -ESRCH;
1602
1603         if (upid == -1)
1604                 type = PIDTYPE_MAX;
1605         else if (upid < 0) {
1606                 type = PIDTYPE_PGID;
1607                 pid = find_get_pid(-upid);
1608         } else if (upid == 0) {
1609                 type = PIDTYPE_PGID;
1610                 pid = get_task_pid(current, PIDTYPE_PGID);
1611         } else /* upid > 0 */ {
1612                 type = PIDTYPE_PID;
1613                 pid = find_get_pid(upid);
1614         }
1615
1616         wo.wo_type      = type;
1617         wo.wo_pid       = pid;
1618         wo.wo_flags     = options | WEXITED;
1619         wo.wo_info      = NULL;
1620         wo.wo_stat      = 0;
1621         wo.wo_rusage    = ru;
1622         ret = do_wait(&wo);
1623         put_pid(pid);
1624         if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1625                 ret = -EFAULT;
1626
1627         return ret;
1628 }
1629
1630 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1631                 int, options, struct rusage __user *, ru)
1632 {
1633         struct rusage r;
1634         long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1635
1636         if (err > 0) {
1637                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1638                         return -EFAULT;
1639         }
1640         return err;
1641 }
1642
1643 #ifdef __ARCH_WANT_SYS_WAITPID
1644
1645 /*
1646  * sys_waitpid() remains for compatibility. waitpid() should be
1647  * implemented by calling sys_wait4() from libc.a.
1648  */
1649 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1650 {
1651         return kernel_wait4(pid, stat_addr, options, NULL);
1652 }
1653
1654 #endif
1655
1656 #ifdef CONFIG_COMPAT
1657 COMPAT_SYSCALL_DEFINE4(wait4,
1658         compat_pid_t, pid,
1659         compat_uint_t __user *, stat_addr,
1660         int, options,
1661         struct compat_rusage __user *, ru)
1662 {
1663         struct rusage r;
1664         long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1665         if (err > 0) {
1666                 if (ru && put_compat_rusage(&r, ru))
1667                         return -EFAULT;
1668         }
1669         return err;
1670 }
1671
1672 COMPAT_SYSCALL_DEFINE5(waitid,
1673                 int, which, compat_pid_t, pid,
1674                 struct compat_siginfo __user *, infop, int, options,
1675                 struct compat_rusage __user *, uru)
1676 {
1677         struct rusage ru;
1678         struct waitid_info info = {.status = 0};
1679         long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1680         int signo = 0;
1681         if (err > 0) {
1682                 signo = SIGCHLD;
1683                 err = 0;
1684                 if (uru) {
1685                         /* kernel_waitid() overwrites everything in ru */
1686                         if (COMPAT_USE_64BIT_TIME)
1687                                 err = copy_to_user(uru, &ru, sizeof(ru));
1688                         else
1689                                 err = put_compat_rusage(&ru, uru);
1690                         if (err)
1691                                 return -EFAULT;
1692                 }
1693         }
1694
1695         if (!infop)
1696                 return err;
1697
1698         if (!user_write_access_begin(infop, sizeof(*infop)))
1699                 return -EFAULT;
1700
1701         unsafe_put_user(signo, &infop->si_signo, Efault);
1702         unsafe_put_user(0, &infop->si_errno, Efault);
1703         unsafe_put_user(info.cause, &infop->si_code, Efault);
1704         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1705         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1706         unsafe_put_user(info.status, &infop->si_status, Efault);
1707         user_write_access_end();
1708         return err;
1709 Efault:
1710         user_write_access_end();
1711         return -EFAULT;
1712 }
1713 #endif
1714
1715 __weak void abort(void)
1716 {
1717         BUG();
1718
1719         /* if that doesn't kill us, halt */
1720         panic("Oops failed to kill thread");
1721 }
1722 EXPORT_SYMBOL(abort);