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