kernel/exit.c

   1 /*
   2  *  linux/kernel/exit.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/mm.h>
   9 #include <linux/slab.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/smp_lock.h>
  12 #include <linux/module.h>
  13 #include <linux/capability.h>
  14 #include <linux/completion.h>
  15 #include <linux/personality.h>
  16 #include <linux/tty.h>
  17 #include <linux/namespace.h>
  18 #include <linux/key.h>
  19 #include <linux/security.h>
  20 #include <linux/cpu.h>
  21 #include <linux/acct.h>
  22 #include <linux/file.h>
  23 #include <linux/binfmts.h>
  24 #include <linux/ptrace.h>
  25 #include <linux/profile.h>
  26 #include <linux/mount.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/mempolicy.h>
  29 #include <linux/cpuset.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/signal.h>
  32 #include <linux/cn_proc.h>
  33 #include <linux/mutex.h>
  34 #include <linux/futex.h>
  35 #include <linux/compat.h>
  36
  37 #include <asm/uaccess.h>
  38 #include <asm/unistd.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/mmu_context.h>
  41
  42 extern void sem_exit (void);
  43 extern struct task_struct *child_reaper;
  44
  45 int getrusage(struct task_struct *, int, struct rusage __user *);
  46
  47 static void exit_mm(struct task_struct * tsk);
  48
  49 static void __unhash_process(struct task_struct *p)
  50 {
  51         nr_threads--;
  52         detach_pid(p, PIDTYPE_PID);
  53         detach_pid(p, PIDTYPE_TGID);
  54         if (thread_group_leader(p)) {
  55                 detach_pid(p, PIDTYPE_PGID);
  56                 detach_pid(p, PIDTYPE_SID);
  57
  58                 list_del_init(&p->tasks);
  59                 __get_cpu_var(process_counts)--;
  60         }
  61
  62         remove_parent(p);
  63 }
  64
  65 void release_task(struct task_struct * p)
  66 {
  67         int zap_leader;
  68         task_t *leader;
  69         struct dentry *proc_dentry;
  70
  71 repeat:
  72         atomic_dec(&p->user->processes);
  73         spin_lock(&p->proc_lock);
  74         proc_dentry = proc_pid_unhash(p);
  75         write_lock_irq(&tasklist_lock);
  76         ptrace_unlink(p);
  77         BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
  78         __exit_signal(p);
  79         /*
  80          * Note that the fastpath in sys_times depends on __exit_signal having
  81          * updated the counters before a task is removed from the tasklist of
  82          * the process by __unhash_process.
  83          */
  84         __unhash_process(p);
  85
  86         /*
  87          * If we are the last non-leader member of the thread
  88          * group, and the leader is zombie, then notify the
  89          * group leader's parent process. (if it wants notification.)
  90          */
  91         zap_leader = 0;
  92         leader = p->group_leader;
  93         if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
  94                 BUG_ON(leader->exit_signal == -1);
  95                 do_notify_parent(leader, leader->exit_signal);
  96                 /*
  97                  * If we were the last child thread and the leader has
  98                  * exited already, and the leader's parent ignores SIGCHLD,
  99                  * then we are the one who should release the leader.
 100                  *
 101                  * do_notify_parent() will have marked it self-reaping in
 102                  * that case.
 103                  */
 104                 zap_leader = (leader->exit_signal == -1);
 105         }
 106
 107         sched_exit(p);
 108         write_unlock_irq(&tasklist_lock);
 109         spin_unlock(&p->proc_lock);
 110         proc_pid_flush(proc_dentry);
 111         release_thread(p);
 112         put_task_struct(p);
 113
 114         p = leader;
 115         if (unlikely(zap_leader))
 116                 goto repeat;
 117 }
 118
 119 /*
 120  * This checks not only the pgrp, but falls back on the pid if no
 121  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
 122  * without this...
 123  */
 124 int session_of_pgrp(int pgrp)
 125 {
 126         struct task_struct *p;
 127         int sid = -1;
 128
 129         read_lock(&tasklist_lock);
 130         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 131                 if (p->signal->session > 0) {
 132                         sid = p->signal->session;
 133                         goto out;
 134                 }
 135         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 136         p = find_task_by_pid(pgrp);
 137         if (p)
 138                 sid = p->signal->session;
 139 out:
 140         read_unlock(&tasklist_lock);
 141
 142         return sid;
 143 }
 144
 145 /*
 146  * Determine if a process group is "orphaned", according to the POSIX
 147  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 148  * by terminal-generated stop signals.  Newly orphaned process groups are
 149  * to receive a SIGHUP and a SIGCONT.
 150  *
 151  * "I ask you, have you ever known what it is to be an orphan?"
 152  */
 153 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 154 {
 155         struct task_struct *p;
 156         int ret = 1;
 157
 158         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 159                 if (p == ignored_task
 160                                 || p->exit_state
 161                                 || p->real_parent->pid == 1)
 162                         continue;
 163                 if (process_group(p->real_parent) != pgrp
 164                             && p->real_parent->signal->session == p->signal->session) {
 165                         ret = 0;
 166                         break;
 167                 }
 168         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 169         return ret;     /* (sighing) "Often!" */
 170 }
 171
 172 int is_orphaned_pgrp(int pgrp)
 173 {
 174         int retval;
 175
 176         read_lock(&tasklist_lock);
 177         retval = will_become_orphaned_pgrp(pgrp, NULL);
 178         read_unlock(&tasklist_lock);
 179
 180         return retval;
 181 }
 182
 183 static int has_stopped_jobs(int pgrp)
 184 {
 185         int retval = 0;
 186         struct task_struct *p;
 187
 188         do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
 189                 if (p->state != TASK_STOPPED)
 190                         continue;
 191
 192                 /* If p is stopped by a debugger on a signal that won't
 193                    stop it, then don't count p as stopped.  This isn't
 194                    perfect but it's a good approximation.  */
 195                 if (unlikely (p->ptrace)
 196                     && p->exit_code != SIGSTOP
 197                     && p->exit_code != SIGTSTP
 198                     && p->exit_code != SIGTTOU
 199                     && p->exit_code != SIGTTIN)
 200                         continue;
 201
 202                 retval = 1;
 203                 break;
 204         } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
 205         return retval;
 206 }
 207
 208 /**
 209  * reparent_to_init - Reparent the calling kernel thread to the init task.
 210  *
 211  * If a kernel thread is launched as a result of a system call, or if
 212  * it ever exits, it should generally reparent itself to init so that
 213  * it is correctly cleaned up on exit.
 214  *
 215  * The various task state such as scheduling policy and priority may have
 216  * been inherited from a user process, so we reset them to sane values here.
 217  *
 218  * NOTE that reparent_to_init() gives the caller full capabilities.
 219  */
 220 static void reparent_to_init(void)
 221 {
 222         write_lock_irq(&tasklist_lock);
 223
 224         ptrace_unlink(current);
 225         /* Reparent to init */
 226         remove_parent(current);
 227         current->parent = child_reaper;
 228         current->real_parent = child_reaper;
 229         add_parent(current);
 230
 231         /* Set the exit signal to SIGCHLD so we signal init on exit */
 232         current->exit_signal = SIGCHLD;
 233
 234         if ((current->policy == SCHED_NORMAL ||
 235                         current->policy == SCHED_BATCH)
 236                                 && (task_nice(current) < 0))
 237                 set_user_nice(current, 0);
 238         /* cpus_allowed? */
 239         /* rt_priority? */
 240         /* signals? */
 241         security_task_reparent_to_init(current);
 242         memcpy(current->signal->rlim, init_task.signal->rlim,
 243                sizeof(current->signal->rlim));
 244         atomic_inc(&(INIT_USER->__count));
 245         write_unlock_irq(&tasklist_lock);
 246         switch_uid(INIT_USER);
 247 }
 248
 249 void __set_special_pids(pid_t session, pid_t pgrp)
 250 {
 251         struct task_struct *curr = current->group_leader;
 252
 253         if (curr->signal->session != session) {
 254                 detach_pid(curr, PIDTYPE_SID);
 255                 curr->signal->session = session;
 256                 attach_pid(curr, PIDTYPE_SID, session);
 257         }
 258         if (process_group(curr) != pgrp) {
 259                 detach_pid(curr, PIDTYPE_PGID);
 260                 curr->signal->pgrp = pgrp;
 261                 attach_pid(curr, PIDTYPE_PGID, pgrp);
 262         }
 263 }
 264
 265 void set_special_pids(pid_t session, pid_t pgrp)
 266 {
 267         write_lock_irq(&tasklist_lock);
 268         __set_special_pids(session, pgrp);
 269         write_unlock_irq(&tasklist_lock);
 270 }
 271
 272 /*
 273  * Let kernel threads use this to say that they
 274  * allow a certain signal (since daemonize() will
 275  * have disabled all of them by default).
 276  */
 277 int allow_signal(int sig)
 278 {
 279         if (!valid_signal(sig) || sig < 1)
 280                 return -EINVAL;
 281
 282         spin_lock_irq(&current->sighand->siglock);
 283         sigdelset(&current->blocked, sig);
 284         if (!current->mm) {
 285                 /* Kernel threads handle their own signals.
 286                    Let the signal code know it'll be handled, so
 287                    that they don't get converted to SIGKILL or
 288                    just silently dropped */
 289                 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
 290         }
 291         recalc_sigpending();
 292         spin_unlock_irq(&current->sighand->siglock);
 293         return 0;
 294 }
 295
 296 EXPORT_SYMBOL(allow_signal);
 297
 298 int disallow_signal(int sig)
 299 {
 300         if (!valid_signal(sig) || sig < 1)
 301                 return -EINVAL;
 302
 303         spin_lock_irq(&current->sighand->siglock);
 304         sigaddset(&current->blocked, sig);
 305         recalc_sigpending();
 306         spin_unlock_irq(&current->sighand->siglock);
 307         return 0;
 308 }
 309
 310 EXPORT_SYMBOL(disallow_signal);
 311
 312 /*
 313  *      Put all the gunge required to become a kernel thread without
 314  *      attached user resources in one place where it belongs.
 315  */
 316
 317 void daemonize(const char *name, ...)
 318 {
 319         va_list args;
 320         struct fs_struct *fs;
 321         sigset_t blocked;
 322
 323         va_start(args, name);
 324         vsnprintf(current->comm, sizeof(current->comm), name, args);
 325         va_end(args);
 326
 327         /*
 328          * If we were started as result of loading a module, close all of the
 329          * user space pages.  We don't need them, and if we didn't close them
 330          * they would be locked into memory.
 331          */
 332         exit_mm(current);
 333
 334         set_special_pids(1, 1);
 335         mutex_lock(&tty_mutex);
 336         current->signal->tty = NULL;
 337         mutex_unlock(&tty_mutex);
 338
 339         /* Block and flush all signals */
 340         sigfillset(&blocked);
 341         sigprocmask(SIG_BLOCK, &blocked, NULL);
 342         flush_signals(current);
 343
 344         /* Become as one with the init task */
 345
 346         exit_fs(current);       /* current->fs->count--; */
 347         fs = init_task.fs;
 348         current->fs = fs;
 349         atomic_inc(&fs->count);
 350         exit_namespace(current);
 351         current->namespace = init_task.namespace;
 352         get_namespace(current->namespace);
 353         exit_files(current);
 354         current->files = init_task.files;
 355         atomic_inc(&current->files->count);
 356
 357         reparent_to_init();
 358 }
 359
 360 EXPORT_SYMBOL(daemonize);
 361
 362 static void close_files(struct files_struct * files)
 363 {
 364         int i, j;
 365         struct fdtable *fdt;
 366
 367         j = 0;
 368
 369         /*
 370          * It is safe to dereference the fd table without RCU or
 371          * ->file_lock because this is the last reference to the
 372          * files structure.
 373          */
 374         fdt = files_fdtable(files);
 375         for (;;) {
 376                 unsigned long set;
 377                 i = j * __NFDBITS;
 378                 if (i >= fdt->max_fdset || i >= fdt->max_fds)
 379                         break;
 380                 set = fdt->open_fds->fds_bits[j++];
 381                 while (set) {
 382                         if (set & 1) {
 383                                 struct file * file = xchg(&fdt->fd[i], NULL);
 384                                 if (file)
 385                                         filp_close(file, files);
 386                         }
 387                         i++;
 388                         set >>= 1;
 389                 }
 390         }
 391 }
 392
 393 struct files_struct *get_files_struct(struct task_struct *task)
 394 {
 395         struct files_struct *files;
 396
 397         task_lock(task);
 398         files = task->files;
 399         if (files)
 400                 atomic_inc(&files->count);
 401         task_unlock(task);
 402
 403         return files;
 404 }
 405
 406 void fastcall put_files_struct(struct files_struct *files)
 407 {
 408         struct fdtable *fdt;
 409
 410         if (atomic_dec_and_test(&files->count)) {
 411                 close_files(files);
 412                 /*
 413                  * Free the fd and fdset arrays if we expanded them.
 414                  * If the fdtable was embedded, pass files for freeing
 415                  * at the end of the RCU grace period. Otherwise,
 416                  * you can free files immediately.
 417                  */
 418                 fdt = files_fdtable(files);
 419                 if (fdt == &files->fdtab)
 420                         fdt->free_files = files;
 421                 else
 422                         kmem_cache_free(files_cachep, files);
 423                 free_fdtable(fdt);
 424         }
 425 }
 426
 427 EXPORT_SYMBOL(put_files_struct);
 428
 429 static inline void __exit_files(struct task_struct *tsk)
 430 {
 431         struct files_struct * files = tsk->files;
 432
 433         if (files) {
 434                 task_lock(tsk);
 435                 tsk->files = NULL;
 436                 task_unlock(tsk);
 437                 put_files_struct(files);
 438         }
 439 }
 440
 441 void exit_files(struct task_struct *tsk)
 442 {
 443         __exit_files(tsk);
 444 }
 445
 446 static inline void __put_fs_struct(struct fs_struct *fs)
 447 {
 448         /* No need to hold fs->lock if we are killing it */
 449         if (atomic_dec_and_test(&fs->count)) {
 450                 dput(fs->root);
 451                 mntput(fs->rootmnt);
 452                 dput(fs->pwd);
 453                 mntput(fs->pwdmnt);
 454                 if (fs->altroot) {
 455                         dput(fs->altroot);
 456                         mntput(fs->altrootmnt);
 457                 }
 458                 kmem_cache_free(fs_cachep, fs);
 459         }
 460 }
 461
 462 void put_fs_struct(struct fs_struct *fs)
 463 {
 464         __put_fs_struct(fs);
 465 }
 466
 467 static inline void __exit_fs(struct task_struct *tsk)
 468 {
 469         struct fs_struct * fs = tsk->fs;
 470
 471         if (fs) {
 472                 task_lock(tsk);
 473                 tsk->fs = NULL;
 474                 task_unlock(tsk);
 475                 __put_fs_struct(fs);
 476         }
 477 }
 478
 479 void exit_fs(struct task_struct *tsk)
 480 {
 481         __exit_fs(tsk);
 482 }
 483
 484 EXPORT_SYMBOL_GPL(exit_fs);
 485
 486 /*
 487  * Turn us into a lazy TLB process if we
 488  * aren't already..
 489  */
 490 static void exit_mm(struct task_struct * tsk)
 491 {
 492         struct mm_struct *mm = tsk->mm;
 493
 494         mm_release(tsk, mm);
 495         if (!mm)
 496                 return;
 497         /*
 498          * Serialize with any possible pending coredump.
 499          * We must hold mmap_sem around checking core_waiters
 500          * and clearing tsk->mm.  The core-inducing thread
 501          * will increment core_waiters for each thread in the
 502          * group with ->mm != NULL.
 503          */
 504         down_read(&mm->mmap_sem);
 505         if (mm->core_waiters) {
 506                 up_read(&mm->mmap_sem);
 507                 down_write(&mm->mmap_sem);
 508                 if (!--mm->core_waiters)
 509                         complete(mm->core_startup_done);
 510                 up_write(&mm->mmap_sem);
 511
 512                 wait_for_completion(&mm->core_done);
 513                 down_read(&mm->mmap_sem);
 514         }
 515         atomic_inc(&mm->mm_count);
 516         if (mm != tsk->active_mm) BUG();
 517         /* more a memory barrier than a real lock */
 518         task_lock(tsk);
 519         tsk->mm = NULL;
 520         up_read(&mm->mmap_sem);
 521         enter_lazy_tlb(mm, current);
 522         task_unlock(tsk);
 523         mmput(mm);
 524 }
 525
 526 static inline void choose_new_parent(task_t *p, task_t *reaper)
 527 {
 528         /*
 529          * Make sure we're not reparenting to ourselves and that
 530          * the parent is not a zombie.
 531          */
 532         BUG_ON(p == reaper || reaper->exit_state);
 533         p->real_parent = reaper;
 534 }
 535
 536 static void reparent_thread(task_t *p, task_t *father, int traced)
 537 {
 538         /* We don't want people slaying init.  */
 539         if (p->exit_signal != -1)
 540                 p->exit_signal = SIGCHLD;
 541
 542         if (p->pdeath_signal)
 543                 /* We already hold the tasklist_lock here.  */
 544                 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 545
 546         /* Move the child from its dying parent to the new one.  */
 547         if (unlikely(traced)) {
 548                 /* Preserve ptrace links if someone else is tracing this child.  */
 549                 list_del_init(&p->ptrace_list);
 550                 if (p->parent != p->real_parent)
 551                         list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
 552         } else {
 553                 /* If this child is being traced, then we're the one tracing it
 554                  * anyway, so let go of it.
 555                  */
 556                 p->ptrace = 0;
 557                 remove_parent(p);
 558                 p->parent = p->real_parent;
 559                 add_parent(p);
 560
 561                 /* If we'd notified the old parent about this child's death,
 562                  * also notify the new parent.
 563                  */
 564                 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 565                     thread_group_empty(p))
 566                         do_notify_parent(p, p->exit_signal);
 567                 else if (p->state == TASK_TRACED) {
 568                         /*
 569                          * If it was at a trace stop, turn it into
 570                          * a normal stop since it's no longer being
 571                          * traced.
 572                          */
 573                         ptrace_untrace(p);
 574                 }
 575         }
 576
 577         /*
 578          * process group orphan check
 579          * Case ii: Our child is in a different pgrp
 580          * than we are, and it was the only connection
 581          * outside, so the child pgrp is now orphaned.
 582          */
 583         if ((process_group(p) != process_group(father)) &&
 584             (p->signal->session == father->signal->session)) {
 585                 int pgrp = process_group(p);
 586
 587                 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
 588                         __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
 589                         __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
 590                 }
 591         }
 592 }
 593
 594 /*
 595  * When we die, we re-parent all our children.
 596  * Try to give them to another thread in our thread
 597  * group, and if no such member exists, give it to
 598  * the global child reaper process (ie "init")
 599  */
 600 static void forget_original_parent(struct task_struct * father,
 601                                           struct list_head *to_release)
 602 {
 603         struct task_struct *p, *reaper = father;
 604         struct list_head *_p, *_n;
 605
 606         do {
 607                 reaper = next_thread(reaper);
 608                 if (reaper == father) {
 609                         reaper = child_reaper;
 610                         break;
 611                 }
 612         } while (reaper->exit_state);
 613
 614         /*
 615          * There are only two places where our children can be:
 616          *
 617          * - in our child list
 618          * - in our ptraced child list
 619          *
 620          * Search them and reparent children.
 621          */
 622         list_for_each_safe(_p, _n, &father->children) {
 623                 int ptrace;
 624                 p = list_entry(_p,struct task_struct,sibling);
 625
 626                 ptrace = p->ptrace;
 627
 628                 /* if father isn't the real parent, then ptrace must be enabled */
 629                 BUG_ON(father != p->real_parent && !ptrace);
 630
 631                 if (father == p->real_parent) {
 632                         /* reparent with a reaper, real father it's us */
 633                         choose_new_parent(p, reaper);
 634                         reparent_thread(p, father, 0);
 635                 } else {
 636                         /* reparent ptraced task to its real parent */
 637                         __ptrace_unlink (p);
 638                         if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
 639                             thread_group_empty(p))
 640                                 do_notify_parent(p, p->exit_signal);
 641                 }
 642
 643                 /*
 644                  * if the ptraced child is a zombie with exit_signal == -1
 645                  * we must collect it before we exit, or it will remain
 646                  * zombie forever since we prevented it from self-reap itself
 647                  * while it was being traced by us, to be able to see it in wait4.
 648                  */
 649                 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
 650                         list_add(&p->ptrace_list, to_release);
 651         }
 652         list_for_each_safe(_p, _n, &father->ptrace_children) {
 653                 p = list_entry(_p,struct task_struct,ptrace_list);
 654                 choose_new_parent(p, reaper);
 655                 reparent_thread(p, father, 1);
 656         }
 657 }
 658
 659 /*
 660  * Send signals to all our closest relatives so that they know
 661  * to properly mourn us..
 662  */
 663 static void exit_notify(struct task_struct *tsk)
 664 {
 665         int state;
 666         struct task_struct *t;
 667         struct list_head ptrace_dead, *_p, *_n;
 668
 669         if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
 670             && !thread_group_empty(tsk)) {
 671                 /*
 672                  * This occurs when there was a race between our exit
 673                  * syscall and a group signal choosing us as the one to
 674                  * wake up.  It could be that we are the only thread
 675                  * alerted to check for pending signals, but another thread
 676                  * should be woken now to take the signal since we will not.
 677                  * Now we'll wake all the threads in the group just to make
 678                  * sure someone gets all the pending signals.
 679                  */
 680                 read_lock(&tasklist_lock);
 681                 spin_lock_irq(&tsk->sighand->siglock);
 682                 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
 683                         if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
 684                                 recalc_sigpending_tsk(t);
 685                                 if (signal_pending(t))
 686                                         signal_wake_up(t, 0);
 687                         }
 688                 spin_unlock_irq(&tsk->sighand->siglock);
 689                 read_unlock(&tasklist_lock);
 690         }
 691
 692         write_lock_irq(&tasklist_lock);
 693
 694         /*
 695          * This does two things:
 696          *
 697          * A.  Make init inherit all the child processes
 698          * B.  Check to see if any process groups have become orphaned
 699          *      as a result of our exiting, and if they have any stopped
 700          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 701          */
 702
 703         INIT_LIST_HEAD(&ptrace_dead);
 704         forget_original_parent(tsk, &ptrace_dead);
 705         BUG_ON(!list_empty(&tsk->children));
 706         BUG_ON(!list_empty(&tsk->ptrace_children));
 707
 708         /*
 709          * Check to see if any process groups have become orphaned
 710          * as a result of our exiting, and if they have any stopped
 711          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 712          *
 713          * Case i: Our father is in a different pgrp than we are
 714          * and we were the only connection outside, so our pgrp
 715          * is about to become orphaned.
 716          */
 717
 718         t = tsk->real_parent;
 719
 720         if ((process_group(t) != process_group(tsk)) &&
 721             (t->signal->session == tsk->signal->session) &&
 722             will_become_orphaned_pgrp(process_group(tsk), tsk) &&
 723             has_stopped_jobs(process_group(tsk))) {
 724                 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
 725                 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
 726         }
 727
 728         /* Let father know we died
 729          *
 730          * Thread signals are configurable, but you aren't going to use
 731          * that to send signals to arbitary processes.
 732          * That stops right now.
 733          *
 734          * If the parent exec id doesn't match the exec id we saved
 735          * when we started then we know the parent has changed security
 736          * domain.
 737          *
 738          * If our self_exec id doesn't match our parent_exec_id then
 739          * we have changed execution domain as these two values started
 740          * the same after a fork.
 741          *
 742          */
 743
 744         if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
 745             ( tsk->parent_exec_id != t->self_exec_id  ||
 746               tsk->self_exec_id != tsk->parent_exec_id)
 747             && !capable(CAP_KILL))
 748                 tsk->exit_signal = SIGCHLD;
 749
 750
 751         /* If something other than our normal parent is ptracing us, then
 752          * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
 753          * only has special meaning to our real parent.
 754          */
 755         if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
 756                 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
 757                 do_notify_parent(tsk, signal);
 758         } else if (tsk->ptrace) {
 759                 do_notify_parent(tsk, SIGCHLD);
 760         }
 761
 762         state = EXIT_ZOMBIE;
 763         if (tsk->exit_signal == -1 &&
 764             (likely(tsk->ptrace == 0) ||
 765              unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
 766                 state = EXIT_DEAD;
 767         tsk->exit_state = state;
 768
 769         write_unlock_irq(&tasklist_lock);
 770
 771         list_for_each_safe(_p, _n, &ptrace_dead) {
 772                 list_del_init(_p);
 773                 t = list_entry(_p,struct task_struct,ptrace_list);
 774                 release_task(t);
 775         }
 776
 777         /* If the process is dead, release it - nobody will wait for it */
 778         if (state == EXIT_DEAD)
 779                 release_task(tsk);
 780 }
 781
 782 fastcall NORET_TYPE void do_exit(long code)
 783 {
 784         struct task_struct *tsk = current;
 785         int group_dead;
 786
 787         profile_task_exit(tsk);
 788
 789         WARN_ON(atomic_read(&tsk->fs_excl));
 790
 791         if (unlikely(in_interrupt()))
 792                 panic("Aiee, killing interrupt handler!");
 793         if (unlikely(!tsk->pid))
 794                 panic("Attempted to kill the idle task!");
 795         if (unlikely(tsk == child_reaper))
 796                 panic("Attempted to kill init!");
 797
 798         if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
 799                 current->ptrace_message = code;
 800                 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
 801         }
 802
 803         /*
 804          * We're taking recursive faults here in do_exit. Safest is to just
 805          * leave this task alone and wait for reboot.
 806          */
 807         if (unlikely(tsk->flags & PF_EXITING)) {
 808                 printk(KERN_ALERT
 809                         "Fixing recursive fault but reboot is needed!\n");
 810                 if (tsk->io_context)
 811                         exit_io_context();
 812                 set_current_state(TASK_UNINTERRUPTIBLE);
 813                 schedule();
 814         }
 815
 816         tsk->flags |= PF_EXITING;
 817
 818         /*
 819          * Make sure we don't try to process any timer firings
 820          * while we are already exiting.
 821          */
 822         tsk->it_virt_expires = cputime_zero;
 823         tsk->it_prof_expires = cputime_zero;
 824         tsk->it_sched_expires = 0;
 825
 826         if (unlikely(in_atomic()))
 827                 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
 828                                 current->comm, current->pid,
 829                                 preempt_count());
 830
 831         acct_update_integrals(tsk);
 832         if (tsk->mm) {
 833                 update_hiwater_rss(tsk->mm);
 834                 update_hiwater_vm(tsk->mm);
 835         }
 836         group_dead = atomic_dec_and_test(&tsk->signal->live);
 837         if (group_dead) {
 838                 hrtimer_cancel(&tsk->signal->real_timer);
 839                 exit_itimers(tsk->signal);
 840                 acct_process(code);
 841         }
 842         if (unlikely(tsk->robust_list))
 843                 exit_robust_list(tsk);
 844 #ifdef CONFIG_COMPAT
 845         if (unlikely(tsk->compat_robust_list))
 846                 compat_exit_robust_list(tsk);
 847 #endif
 848         exit_mm(tsk);
 849
 850         exit_sem(tsk);
 851         __exit_files(tsk);
 852         __exit_fs(tsk);
 853         exit_namespace(tsk);
 854         exit_thread();
 855         cpuset_exit(tsk);
 856         exit_keys(tsk);
 857
 858         if (group_dead && tsk->signal->leader)
 859                 disassociate_ctty(1);
 860
 861         module_put(task_thread_info(tsk)->exec_domain->module);
 862         if (tsk->binfmt)
 863                 module_put(tsk->binfmt->module);
 864
 865         tsk->exit_code = code;
 866         proc_exit_connector(tsk);
 867         exit_notify(tsk);
 868 #ifdef CONFIG_NUMA
 869         mpol_free(tsk->mempolicy);
 870         tsk->mempolicy = NULL;
 871 #endif
 872         /*
 873          * If DEBUG_MUTEXES is on, make sure we are holding no locks:
 874          */
 875         mutex_debug_check_no_locks_held(tsk);
 876
 877         if (tsk->io_context)
 878                 exit_io_context();
 879
 880         /* PF_DEAD causes final put_task_struct after we schedule. */
 881         preempt_disable();
 882         BUG_ON(tsk->flags & PF_DEAD);
 883         tsk->flags |= PF_DEAD;
 884
 885         schedule();
 886         BUG();
 887         /* Avoid "noreturn function does return".  */
 888         for (;;) ;
 889 }
 890
 891 EXPORT_SYMBOL_GPL(do_exit);
 892
 893 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 894 {
 895         if (comp)
 896                 complete(comp);
 897
 898         do_exit(code);
 899 }
 900
 901 EXPORT_SYMBOL(complete_and_exit);
 902
 903 asmlinkage long sys_exit(int error_code)
 904 {
 905         do_exit((error_code&0xff)<<8);
 906 }
 907
 908 task_t fastcall *next_thread(const task_t *p)
 909 {
 910         return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
 911 }
 912
 913 EXPORT_SYMBOL(next_thread);
 914
 915 /*
 916  * Take down every thread in the group.  This is called by fatal signals
 917  * as well as by sys_exit_group (below).
 918  */
 919 NORET_TYPE void
 920 do_group_exit(int exit_code)
 921 {
 922         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
 923
 924         if (current->signal->flags & SIGNAL_GROUP_EXIT)
 925                 exit_code = current->signal->group_exit_code;
 926         else if (!thread_group_empty(current)) {
 927                 struct signal_struct *const sig = current->signal;
 928                 struct sighand_struct *const sighand = current->sighand;
 929                 read_lock(&tasklist_lock);
 930                 spin_lock_irq(&sighand->siglock);
 931                 if (sig->flags & SIGNAL_GROUP_EXIT)
 932                         /* Another thread got here before we took the lock.  */
 933                         exit_code = sig->group_exit_code;
 934                 else {
 935                         sig->group_exit_code = exit_code;
 936                         zap_other_threads(current);
 937                 }
 938                 spin_unlock_irq(&sighand->siglock);
 939                 read_unlock(&tasklist_lock);
 940         }
 941
 942         do_exit(exit_code);
 943         /* NOTREACHED */
 944 }
 945
 946 /*
 947  * this kills every thread in the thread group. Note that any externally
 948  * wait4()-ing process will get the correct exit code - even if this
 949  * thread is not the thread group leader.
 950  */
 951 asmlinkage void sys_exit_group(int error_code)
 952 {
 953         do_group_exit((error_code & 0xff) << 8);
 954 }
 955
 956 static int eligible_child(pid_t pid, int options, task_t *p)
 957 {
 958         if (pid > 0) {
 959                 if (p->pid != pid)
 960                         return 0;
 961         } else if (!pid) {
 962                 if (process_group(p) != process_group(current))
 963                         return 0;
 964         } else if (pid != -1) {
 965                 if (process_group(p) != -pid)
 966                         return 0;
 967         }
 968
 969         /*
 970          * Do not consider detached threads that are
 971          * not ptraced:
 972          */
 973         if (p->exit_signal == -1 && !p->ptrace)
 974                 return 0;
 975
 976         /* Wait for all children (clone and not) if __WALL is set;
 977          * otherwise, wait for clone children *only* if __WCLONE is
 978          * set; otherwise, wait for non-clone children *only*.  (Note:
 979          * A "clone" child here is one that reports to its parent
 980          * using a signal other than SIGCHLD.) */
 981         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 982             && !(options & __WALL))
 983                 return 0;
 984         /*
 985          * Do not consider thread group leaders that are
 986          * in a non-empty thread group:
 987          */
 988         if (current->tgid != p->tgid && delay_group_leader(p))
 989                 return 2;
 990
 991         if (security_task_wait(p))
 992                 return 0;
 993
 994         return 1;
 995 }
 996
 997 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 998                                int why, int status,
 999                                struct siginfo __user *infop,
1000                                struct rusage __user *rusagep)
1001 {
1002         int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1003         put_task_struct(p);
1004         if (!retval)
1005                 retval = put_user(SIGCHLD, &infop->si_signo);
1006         if (!retval)
1007                 retval = put_user(0, &infop->si_errno);
1008         if (!retval)
1009                 retval = put_user((short)why, &infop->si_code);
1010         if (!retval)
1011                 retval = put_user(pid, &infop->si_pid);
1012         if (!retval)
1013                 retval = put_user(uid, &infop->si_uid);
1014         if (!retval)
1015                 retval = put_user(status, &infop->si_status);
1016         if (!retval)
1017                 retval = pid;
1018         return retval;
1019 }
1020
1021 /*
1022  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1023  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1024  * the lock and this task is uninteresting.  If we return nonzero, we have
1025  * released the lock and the system call should return.
1026  */
1027 static int wait_task_zombie(task_t *p, int noreap,
1028                             struct siginfo __user *infop,
1029                             int __user *stat_addr, struct rusage __user *ru)
1030 {
1031         unsigned long state;
1032         int retval;
1033         int status;
1034
1035         if (unlikely(noreap)) {
1036                 pid_t pid = p->pid;
1037                 uid_t uid = p->uid;
1038                 int exit_code = p->exit_code;
1039                 int why, status;
1040
1041                 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1042                         return 0;
1043                 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1044                         return 0;
1045                 get_task_struct(p);
1046                 read_unlock(&tasklist_lock);
1047                 if ((exit_code & 0x7f) == 0) {
1048                         why = CLD_EXITED;
1049                         status = exit_code >> 8;
1050                 } else {
1051                         why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1052                         status = exit_code & 0x7f;
1053                 }
1054                 return wait_noreap_copyout(p, pid, uid, why,
1055                                            status, infop, ru);
1056         }
1057
1058         /*
1059          * Try to move the task's state to DEAD
1060          * only one thread is allowed to do this:
1061          */
1062         state = xchg(&p->exit_state, EXIT_DEAD);
1063         if (state != EXIT_ZOMBIE) {
1064                 BUG_ON(state != EXIT_DEAD);
1065                 return 0;
1066         }
1067         if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1068                 /*
1069                  * This can only happen in a race with a ptraced thread
1070                  * dying on another processor.
1071                  */
1072                 return 0;
1073         }
1074
1075         if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1076                 struct signal_struct *psig;
1077                 struct signal_struct *sig;
1078
1079                 /*
1080                  * The resource counters for the group leader are in its
1081                  * own task_struct.  Those for dead threads in the group
1082                  * are in its signal_struct, as are those for the child
1083                  * processes it has previously reaped.  All these
1084                  * accumulate in the parent's signal_struct c* fields.
1085                  *
1086                  * We don't bother to take a lock here to protect these
1087                  * p->signal fields, because they are only touched by
1088                  * __exit_signal, which runs with tasklist_lock
1089                  * write-locked anyway, and so is excluded here.  We do
1090                  * need to protect the access to p->parent->signal fields,
1091                  * as other threads in the parent group can be right
1092                  * here reaping other children at the same time.
1093                  */
1094                 spin_lock_irq(&p->parent->sighand->siglock);
1095                 psig = p->parent->signal;
1096                 sig = p->signal;
1097                 psig->cutime =
1098                         cputime_add(psig->cutime,
1099                         cputime_add(p->utime,
1100                         cputime_add(sig->utime,
1101                                     sig->cutime)));
1102                 psig->cstime =
1103                         cputime_add(psig->cstime,
1104                         cputime_add(p->stime,
1105                         cputime_add(sig->stime,
1106                                     sig->cstime)));
1107                 psig->cmin_flt +=
1108                         p->min_flt + sig->min_flt + sig->cmin_flt;
1109                 psig->cmaj_flt +=
1110                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1111                 psig->cnvcsw +=
1112                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1113                 psig->cnivcsw +=
1114                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1115                 spin_unlock_irq(&p->parent->sighand->siglock);
1116         }
1117
1118         /*
1119          * Now we are sure this task is interesting, and no other
1120          * thread can reap it because we set its state to EXIT_DEAD.
1121          */
1122         read_unlock(&tasklist_lock);
1123
1124         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1125         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1126                 ? p->signal->group_exit_code : p->exit_code;
1127         if (!retval && stat_addr)
1128                 retval = put_user(status, stat_addr);
1129         if (!retval && infop)
1130                 retval = put_user(SIGCHLD, &infop->si_signo);
1131         if (!retval && infop)
1132                 retval = put_user(0, &infop->si_errno);
1133         if (!retval && infop) {
1134                 int why;
1135
1136                 if ((status & 0x7f) == 0) {
1137                         why = CLD_EXITED;
1138                         status >>= 8;
1139                 } else {
1140                         why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1141                         status &= 0x7f;
1142                 }
1143                 retval = put_user((short)why, &infop->si_code);
1144                 if (!retval)
1145                         retval = put_user(status, &infop->si_status);
1146         }
1147         if (!retval && infop)
1148                 retval = put_user(p->pid, &infop->si_pid);
1149         if (!retval && infop)
1150                 retval = put_user(p->uid, &infop->si_uid);
1151         if (retval) {
1152                 // TODO: is this safe?
1153                 p->exit_state = EXIT_ZOMBIE;
1154                 return retval;
1155         }
1156         retval = p->pid;
1157         if (p->real_parent != p->parent) {
1158                 write_lock_irq(&tasklist_lock);
1159                 /* Double-check with lock held.  */
1160                 if (p->real_parent != p->parent) {
1161                         __ptrace_unlink(p);
1162                         // TODO: is this safe?
1163                         p->exit_state = EXIT_ZOMBIE;
1164                         /*
1165                          * If this is not a detached task, notify the parent.
1166                          * If it's still not detached after that, don't release
1167                          * it now.
1168                          */
1169                         if (p->exit_signal != -1) {
1170                                 do_notify_parent(p, p->exit_signal);
1171                                 if (p->exit_signal != -1)
1172                                         p = NULL;
1173                         }
1174                 }
1175                 write_unlock_irq(&tasklist_lock);
1176         }
1177         if (p != NULL)
1178                 release_task(p);
1179         BUG_ON(!retval);
1180         return retval;
1181 }
1182
1183 /*
1184  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1185  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1186  * the lock and this task is uninteresting.  If we return nonzero, we have
1187  * released the lock and the system call should return.
1188  */
1189 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1190                              struct siginfo __user *infop,
1191                              int __user *stat_addr, struct rusage __user *ru)
1192 {
1193         int retval, exit_code;
1194
1195         if (!p->exit_code)
1196                 return 0;
1197         if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1198             p->signal && p->signal->group_stop_count > 0)
1199                 /*
1200                  * A group stop is in progress and this is the group leader.
1201                  * We won't report until all threads have stopped.
1202                  */
1203                 return 0;
1204
1205         /*
1206          * Now we are pretty sure this task is interesting.
1207          * Make sure it doesn't get reaped out from under us while we
1208          * give up the lock and then examine it below.  We don't want to
1209          * keep holding onto the tasklist_lock while we call getrusage and
1210          * possibly take page faults for user memory.
1211          */
1212         get_task_struct(p);
1213         read_unlock(&tasklist_lock);
1214
1215         if (unlikely(noreap)) {
1216                 pid_t pid = p->pid;
1217                 uid_t uid = p->uid;
1218                 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1219
1220                 exit_code = p->exit_code;
1221                 if (unlikely(!exit_code) ||
1222                     unlikely(p->state & TASK_TRACED))
1223                         goto bail_ref;
1224                 return wait_noreap_copyout(p, pid, uid,
1225                                            why, (exit_code << 8) | 0x7f,
1226                                            infop, ru);
1227         }
1228
1229         write_lock_irq(&tasklist_lock);
1230
1231         /*
1232          * This uses xchg to be atomic with the thread resuming and setting
1233          * it.  It must also be done with the write lock held to prevent a
1234          * race with the EXIT_ZOMBIE case.
1235          */
1236         exit_code = xchg(&p->exit_code, 0);
1237         if (unlikely(p->exit_state)) {
1238                 /*
1239                  * The task resumed and then died.  Let the next iteration
1240                  * catch it in EXIT_ZOMBIE.  Note that exit_code might
1241                  * already be zero here if it resumed and did _exit(0).
1242                  * The task itself is dead and won't touch exit_code again;
1243                  * other processors in this function are locked out.
1244                  */
1245                 p->exit_code = exit_code;
1246                 exit_code = 0;
1247         }
1248         if (unlikely(exit_code == 0)) {
1249                 /*
1250                  * Another thread in this function got to it first, or it
1251                  * resumed, or it resumed and then died.
1252                  */
1253                 write_unlock_irq(&tasklist_lock);
1254 bail_ref:
1255                 put_task_struct(p);
1256                 /*
1257                  * We are returning to the wait loop without having successfully
1258                  * removed the process and having released the lock. We cannot
1259                  * continue, since the "p" task pointer is potentially stale.
1260                  *
1261                  * Return -EAGAIN, and do_wait() will restart the loop from the
1262                  * beginning. Do _not_ re-acquire the lock.
1263                  */
1264                 return -EAGAIN;
1265         }
1266
1267         /* move to end of parent's list to avoid starvation */
1268         remove_parent(p);
1269         add_parent(p);
1270
1271         write_unlock_irq(&tasklist_lock);
1272
1273         retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1274         if (!retval && stat_addr)
1275                 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1276         if (!retval && infop)
1277                 retval = put_user(SIGCHLD, &infop->si_signo);
1278         if (!retval && infop)
1279                 retval = put_user(0, &infop->si_errno);
1280         if (!retval && infop)
1281                 retval = put_user((short)((p->ptrace & PT_PTRACED)
1282                                           ? CLD_TRAPPED : CLD_STOPPED),
1283                                   &infop->si_code);
1284         if (!retval && infop)
1285                 retval = put_user(exit_code, &infop->si_status);
1286         if (!retval && infop)
1287                 retval = put_user(p->pid, &infop->si_pid);
1288         if (!retval && infop)
1289                 retval = put_user(p->uid, &infop->si_uid);
1290         if (!retval)
1291                 retval = p->pid;
1292         put_task_struct(p);
1293
1294         BUG_ON(!retval);
1295         return retval;
1296 }
1297
1298 /*
1299  * Handle do_wait work for one task in a live, non-stopped state.
1300  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1301  * the lock and this task is uninteresting.  If we return nonzero, we have
1302  * released the lock and the system call should return.
1303  */
1304 static int wait_task_continued(task_t *p, int noreap,
1305                                struct siginfo __user *infop,
1306                                int __user *stat_addr, struct rusage __user *ru)
1307 {
1308         int retval;
1309         pid_t pid;
1310         uid_t uid;
1311
1312         if (unlikely(!p->signal))
1313                 return 0;
1314
1315         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1316                 return 0;
1317
1318         spin_lock_irq(&p->sighand->siglock);
1319         /* Re-check with the lock held.  */
1320         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1321                 spin_unlock_irq(&p->sighand->siglock);
1322                 return 0;
1323         }
1324         if (!noreap)
1325                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1326         spin_unlock_irq(&p->sighand->siglock);
1327
1328         pid = p->pid;
1329         uid = p->uid;
1330         get_task_struct(p);
1331         read_unlock(&tasklist_lock);
1332
1333         if (!infop) {
1334                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1335                 put_task_struct(p);
1336                 if (!retval && stat_addr)
1337                         retval = put_user(0xffff, stat_addr);
1338                 if (!retval)
1339                         retval = p->pid;
1340         } else {
1341                 retval = wait_noreap_copyout(p, pid, uid,
1342                                              CLD_CONTINUED, SIGCONT,
1343                                              infop, ru);
1344                 BUG_ON(retval == 0);
1345         }
1346
1347         return retval;
1348 }
1349
1350
1351 static inline int my_ptrace_child(struct task_struct *p)
1352 {
1353         if (!(p->ptrace & PT_PTRACED))
1354                 return 0;
1355         if (!(p->ptrace & PT_ATTACHED))
1356                 return 1;
1357         /*
1358          * This child was PTRACE_ATTACH'd.  We should be seeing it only if
1359          * we are the attacher.  If we are the real parent, this is a race
1360          * inside ptrace_attach.  It is waiting for the tasklist_lock,
1361          * which we have to switch the parent links, but has already set
1362          * the flags in p->ptrace.
1363          */
1364         return (p->parent != p->real_parent);
1365 }
1366
1367 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1368                     int __user *stat_addr, struct rusage __user *ru)
1369 {
1370         DECLARE_WAITQUEUE(wait, current);
1371         struct task_struct *tsk;
1372         int flag, retval;
1373
1374         add_wait_queue(&current->signal->wait_chldexit,&wait);
1375 repeat:
1376         /*
1377          * We will set this flag if we see any child that might later
1378          * match our criteria, even if we are not able to reap it yet.
1379          */
1380         flag = 0;
1381         current->state = TASK_INTERRUPTIBLE;
1382         read_lock(&tasklist_lock);
1383         tsk = current;
1384         do {
1385                 struct task_struct *p;
1386                 struct list_head *_p;
1387                 int ret;
1388
1389                 list_for_each(_p,&tsk->children) {
1390                         p = list_entry(_p,struct task_struct,sibling);
1391
1392                         ret = eligible_child(pid, options, p);
1393                         if (!ret)
1394                                 continue;
1395
1396                         switch (p->state) {
1397                         case TASK_TRACED:
1398                                 /*
1399                                  * When we hit the race with PTRACE_ATTACH,
1400                                  * we will not report this child.  But the
1401                                  * race means it has not yet been moved to
1402                                  * our ptrace_children list, so we need to
1403                                  * set the flag here to avoid a spurious ECHILD
1404                                  * when the race happens with the only child.
1405                                  */
1406                                 flag = 1;
1407                                 if (!my_ptrace_child(p))
1408                                         continue;
1409                                 /*FALLTHROUGH*/
1410                         case TASK_STOPPED:
1411                                 /*
1412                                  * It's stopped now, so it might later
1413                                  * continue, exit, or stop again.
1414                                  */
1415                                 flag = 1;
1416                                 if (!(options & WUNTRACED) &&
1417                                     !my_ptrace_child(p))
1418                                         continue;
1419                                 retval = wait_task_stopped(p, ret == 2,
1420                                                            (options & WNOWAIT),
1421                                                            infop,
1422                                                            stat_addr, ru);
1423                                 if (retval == -EAGAIN)
1424                                         goto repeat;
1425                                 if (retval != 0) /* He released the lock.  */
1426                                         goto end;
1427                                 break;
1428                         default:
1429                         // case EXIT_DEAD:
1430                                 if (p->exit_state == EXIT_DEAD)
1431                                         continue;
1432                         // case EXIT_ZOMBIE:
1433                                 if (p->exit_state == EXIT_ZOMBIE) {
1434                                         /*
1435                                          * Eligible but we cannot release
1436                                          * it yet:
1437                                          */
1438                                         if (ret == 2)
1439                                                 goto check_continued;
1440                                         if (!likely(options & WEXITED))
1441                                                 continue;
1442                                         retval = wait_task_zombie(
1443                                                 p, (options & WNOWAIT),
1444                                                 infop, stat_addr, ru);
1445                                         /* He released the lock.  */
1446                                         if (retval != 0)
1447                                                 goto end;
1448                                         break;
1449                                 }
1450 check_continued:
1451                                 /*
1452                                  * It's running now, so it might later
1453                                  * exit, stop, or stop and then continue.
1454                                  */
1455                                 flag = 1;
1456                                 if (!unlikely(options & WCONTINUED))
1457                                         continue;
1458                                 retval = wait_task_continued(
1459                                         p, (options & WNOWAIT),
1460                                         infop, stat_addr, ru);
1461                                 if (retval != 0) /* He released the lock.  */
1462                                         goto end;
1463                                 break;
1464                         }
1465                 }
1466                 if (!flag) {
1467                         list_for_each(_p, &tsk->ptrace_children) {
1468                                 p = list_entry(_p, struct task_struct,
1469                                                 ptrace_list);
1470                                 if (!eligible_child(pid, options, p))
1471                                         continue;
1472                                 flag = 1;
1473                                 break;
1474                         }
1475                 }
1476                 if (options & __WNOTHREAD)
1477                         break;
1478                 tsk = next_thread(tsk);
1479                 if (tsk->signal != current->signal)
1480                         BUG();
1481         } while (tsk != current);
1482
1483         read_unlock(&tasklist_lock);
1484         if (flag) {
1485                 retval = 0;
1486                 if (options & WNOHANG)
1487                         goto end;
1488                 retval = -ERESTARTSYS;
1489                 if (signal_pending(current))
1490                         goto end;
1491                 schedule();
1492                 goto repeat;
1493         }
1494         retval = -ECHILD;
1495 end:
1496         current->state = TASK_RUNNING;
1497         remove_wait_queue(&current->signal->wait_chldexit,&wait);
1498         if (infop) {
1499                 if (retval > 0)
1500                 retval = 0;
1501                 else {
1502                         /*
1503                          * For a WNOHANG return, clear out all the fields
1504                          * we would set so the user can easily tell the
1505                          * difference.
1506                          */
1507                         if (!retval)
1508                                 retval = put_user(0, &infop->si_signo);
1509                         if (!retval)
1510                                 retval = put_user(0, &infop->si_errno);
1511                         if (!retval)
1512                                 retval = put_user(0, &infop->si_code);
1513                         if (!retval)
1514                                 retval = put_user(0, &infop->si_pid);
1515                         if (!retval)
1516                                 retval = put_user(0, &infop->si_uid);
1517                         if (!retval)
1518                                 retval = put_user(0, &infop->si_status);
1519                 }
1520         }
1521         return retval;
1522 }
1523
1524 asmlinkage long sys_waitid(int which, pid_t pid,
1525                            struct siginfo __user *infop, int options,
1526                            struct rusage __user *ru)
1527 {
1528         long ret;
1529
1530         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1531                 return -EINVAL;
1532         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1533                 return -EINVAL;
1534
1535         switch (which) {
1536         case P_ALL:
1537                 pid = -1;
1538                 break;
1539         case P_PID:
1540                 if (pid <= 0)
1541                         return -EINVAL;
1542                 break;
1543         case P_PGID:
1544                 if (pid <= 0)
1545                         return -EINVAL;
1546                 pid = -pid;
1547                 break;
1548         default:
1549                 return -EINVAL;
1550         }
1551
1552         ret = do_wait(pid, options, infop, NULL, ru);
1553
1554         /* avoid REGPARM breakage on x86: */
1555         prevent_tail_call(ret);
1556         return ret;
1557 }
1558
1559 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1560                           int options, struct rusage __user *ru)
1561 {
1562         long ret;
1563
1564         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1565                         __WNOTHREAD|__WCLONE|__WALL))
1566                 return -EINVAL;
1567         ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1568
1569         /* avoid REGPARM breakage on x86: */
1570         prevent_tail_call(ret);
1571         return ret;
1572 }
1573
1574 #ifdef __ARCH_WANT_SYS_WAITPID
1575
1576 /*
1577  * sys_waitpid() remains for compatibility. waitpid() should be
1578  * implemented by calling sys_wait4() from libc.a.
1579  */
1580 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1581 {
1582         return sys_wait4(pid, stat_addr, options, NULL);
1583 }
1584
1585 #endif