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/slab.h>
   9 #include <linux/interrupt.h>
  10 #include <linux/smp_lock.h>
  11 #include <linux/module.h>
  12 #include <linux/completion.h>
  13 #include <linux/personality.h>
  14 #include <linux/tty.h>
  15 #include <linux/namespace.h>
  16 #ifdef CONFIG_BSD_PROCESS_ACCT
  17 #include <linux/acct.h>
  18 #endif
  19
  20 #include <asm/uaccess.h>
  21 #include <asm/pgtable.h>
  22 #include <asm/mmu_context.h>
  23
  24 extern void sem_exit (void);
  25 extern struct task_struct *child_reaper;
  26
  27 int getrusage(struct task_struct *, int, struct rusage *);
  28
  29 static void release_task(struct task_struct * p)
  30 {
  31         if (p != current) {
  32 #ifdef CONFIG_SMP
  33                 /*
  34                  * Wait to make sure the process isn't on the
  35                  * runqueue (active on some other CPU still)
  36                  */
  37                 for (;;) {
  38                         task_lock(p);
  39                         if (!task_has_cpu(p))
  40                                 break;
  41                         task_unlock(p);
  42                         do {
  43                                 cpu_relax();
  44                                 barrier();
  45                         } while (task_has_cpu(p));
  46                 }
  47                 task_unlock(p);
  48 #endif
  49                 atomic_dec(&p->user->processes);
  50                 free_uid(p->user);
  51                 unhash_process(p);
  52
  53                 release_thread(p);
  54                 current->cmin_flt += p->min_flt + p->cmin_flt;
  55                 current->cmaj_flt += p->maj_flt + p->cmaj_flt;
  56                 current->cnswap += p->nswap + p->cnswap;
  57                 /*
  58                  * Potentially available timeslices are retrieved
  59                  * here - this way the parent does not get penalized
  60                  * for creating too many processes.
  61                  *
  62                  * (this cannot be used to artificially 'generate'
  63                  * timeslices, because any timeslice recovered here
  64                  * was given away by the parent in the first place.)
  65                  */
  66                 current->counter += p->counter;
  67                 if (current->counter >= MAX_COUNTER)
  68                         current->counter = MAX_COUNTER;
  69                 p->pid = 0;
  70                 free_task_struct(p);
  71         } else {
  72                 printk("task releasing itself\n");
  73         }
  74 }
  75
  76 /*
  77  * This checks not only the pgrp, but falls back on the pid if no
  78  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
  79  * without this...
  80  */
  81 int session_of_pgrp(int pgrp)
  82 {
  83         struct task_struct *p;
  84         int fallback;
  85
  86         fallback = -1;
  87         read_lock(&tasklist_lock);
  88         for_each_task(p) {
  89                 if (p->session <= 0)
  90                         continue;
  91                 if (p->pgrp == pgrp) {
  92                         fallback = p->session;
  93                         break;
  94                 }
  95                 if (p->pid == pgrp)
  96                         fallback = p->session;
  97         }
  98         read_unlock(&tasklist_lock);
  99         return fallback;
 100 }
 101
 102 /*
 103  * Determine if a process group is "orphaned", according to the POSIX
 104  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
 105  * by terminal-generated stop signals.  Newly orphaned process groups are
 106  * to receive a SIGHUP and a SIGCONT.
 107  *
 108  * "I ask you, have you ever known what it is to be an orphan?"
 109  */
 110 static int will_become_orphaned_pgrp(int pgrp, struct task_struct * ignored_task)
 111 {
 112         struct task_struct *p;
 113
 114         read_lock(&tasklist_lock);
 115         for_each_task(p) {
 116                 if ((p == ignored_task) || (p->pgrp != pgrp) ||
 117                     (p->state == TASK_ZOMBIE) ||
 118                     (p->p_pptr->pid == 1))
 119                         continue;
 120                 if ((p->p_pptr->pgrp != pgrp) &&
 121                     (p->p_pptr->session == p->session)) {
 122                         read_unlock(&tasklist_lock);
 123                         return 0;
 124                 }
 125         }
 126         read_unlock(&tasklist_lock);
 127         return 1;       /* (sighing) "Often!" */
 128 }
 129
 130 int is_orphaned_pgrp(int pgrp)
 131 {
 132         return will_become_orphaned_pgrp(pgrp, 0);
 133 }
 134
 135 static inline int has_stopped_jobs(int pgrp)
 136 {
 137         int retval = 0;
 138         struct task_struct * p;
 139
 140         read_lock(&tasklist_lock);
 141         for_each_task(p) {
 142                 if (p->pgrp != pgrp)
 143                         continue;
 144                 if (p->state != TASK_STOPPED)
 145                         continue;
 146                 retval = 1;
 147                 break;
 148         }
 149         read_unlock(&tasklist_lock);
 150         return retval;
 151 }
 152
 153 /*
 154  * When we die, we re-parent all our children.
 155  * Try to give them to another thread in our thread
 156  * group, and if no such member exists, give it to
 157  * the global child reaper process (ie "init")
 158  */
 159 static inline void forget_original_parent(struct task_struct * father)
 160 {
 161         struct task_struct * p;
 162
 163         read_lock(&tasklist_lock);
 164
 165         for_each_task(p) {
 166                 if (p->p_opptr == father) {
 167                         /* We dont want people slaying init */
 168                         p->exit_signal = SIGCHLD;
 169                         p->self_exec_id++;
 170
 171                         /* Make sure we're not reparenting to ourselves */
 172                         p->p_opptr = child_reaper;
 173
 174                         if (p->pdeath_signal) send_sig(p->pdeath_signal, p, 0);
 175                 }
 176         }
 177         read_unlock(&tasklist_lock);
 178 }
 179
 180 static inline void close_files(struct files_struct * files)
 181 {
 182         int i, j;
 183
 184         j = 0;
 185         for (;;) {
 186                 unsigned long set;
 187                 i = j * __NFDBITS;
 188                 if (i >= files->max_fdset || i >= files->max_fds)
 189                         break;
 190                 set = files->open_fds->fds_bits[j++];
 191                 while (set) {
 192                         if (set & 1) {
 193                                 struct file * file = xchg(&files->fd[i], NULL);
 194                                 if (file)
 195                                         filp_close(file, files);
 196                         }
 197                         i++;
 198                         set >>= 1;
 199                 }
 200         }
 201 }
 202
 203 void put_files_struct(struct files_struct *files)
 204 {
 205         if (atomic_dec_and_test(&files->count)) {
 206                 close_files(files);
 207                 /*
 208                  * Free the fd and fdset arrays if we expanded them.
 209                  */
 210                 if (files->fd != &files->fd_array[0])
 211                         free_fd_array(files->fd, files->max_fds);
 212                 if (files->max_fdset > __FD_SETSIZE) {
 213                         free_fdset(files->open_fds, files->max_fdset);
 214                         free_fdset(files->close_on_exec, files->max_fdset);
 215                 }
 216                 kmem_cache_free(files_cachep, files);
 217         }
 218 }
 219
 220 static inline void __exit_files(struct task_struct *tsk)
 221 {
 222         struct files_struct * files = tsk->files;
 223
 224         if (files) {
 225                 task_lock(tsk);
 226                 tsk->files = NULL;
 227                 task_unlock(tsk);
 228                 put_files_struct(files);
 229         }
 230 }
 231
 232 void exit_files(struct task_struct *tsk)
 233 {
 234         __exit_files(tsk);
 235 }
 236
 237 static inline void __put_fs_struct(struct fs_struct *fs)
 238 {
 239         /* No need to hold fs->lock if we are killing it */
 240         if (atomic_dec_and_test(&fs->count)) {
 241                 dput(fs->root);
 242                 mntput(fs->rootmnt);
 243                 dput(fs->pwd);
 244                 mntput(fs->pwdmnt);
 245                 if (fs->altroot) {
 246                         dput(fs->altroot);
 247                         mntput(fs->altrootmnt);
 248                 }
 249                 kmem_cache_free(fs_cachep, fs);
 250         }
 251 }
 252
 253 void put_fs_struct(struct fs_struct *fs)
 254 {
 255         __put_fs_struct(fs);
 256 }
 257
 258 static inline void __exit_fs(struct task_struct *tsk)
 259 {
 260         struct fs_struct * fs = tsk->fs;
 261
 262         if (fs) {
 263                 task_lock(tsk);
 264                 tsk->fs = NULL;
 265                 task_unlock(tsk);
 266                 __put_fs_struct(fs);
 267         }
 268 }
 269
 270 void exit_fs(struct task_struct *tsk)
 271 {
 272         __exit_fs(tsk);
 273 }
 274
 275 /*
 276  * We can use these to temporarily drop into
 277  * "lazy TLB" mode and back.
 278  */
 279 struct mm_struct * start_lazy_tlb(void)
 280 {
 281         struct mm_struct *mm = current->mm;
 282         current->mm = NULL;
 283         /* active_mm is still 'mm' */
 284         atomic_inc(&mm->mm_count);
 285         enter_lazy_tlb(mm, current, smp_processor_id());
 286         return mm;
 287 }
 288
 289 void end_lazy_tlb(struct mm_struct *mm)
 290 {
 291         struct mm_struct *active_mm = current->active_mm;
 292
 293         current->mm = mm;
 294         if (mm != active_mm) {
 295                 current->active_mm = mm;
 296                 activate_mm(active_mm, mm);
 297         }
 298         mmdrop(active_mm);
 299 }
 300
 301 /*
 302  * Turn us into a lazy TLB process if we
 303  * aren't already..
 304  */
 305 static inline void __exit_mm(struct task_struct * tsk)
 306 {
 307         struct mm_struct * mm = tsk->mm;
 308
 309         mm_release();
 310         if (mm) {
 311                 atomic_inc(&mm->mm_count);
 312                 BUG_ON(mm != tsk->active_mm);
 313                 /* more a memory barrier than a real lock */
 314                 task_lock(tsk);
 315                 tsk->mm = NULL;
 316                 task_unlock(tsk);
 317                 enter_lazy_tlb(mm, current, smp_processor_id());
 318                 mmput(mm);
 319         }
 320 }
 321
 322 void exit_mm(struct task_struct *tsk)
 323 {
 324         __exit_mm(tsk);
 325 }
 326
 327 /*
 328  * Send signals to all our closest relatives so that they know
 329  * to properly mourn us..
 330  */
 331 static void exit_notify(void)
 332 {
 333         struct task_struct * p, *t;
 334
 335         forget_original_parent(current);
 336         /*
 337          * Check to see if any process groups have become orphaned
 338          * as a result of our exiting, and if they have any stopped
 339          * jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 340          *
 341          * Case i: Our father is in a different pgrp than we are
 342          * and we were the only connection outside, so our pgrp
 343          * is about to become orphaned.
 344          */
 345
 346         t = current->p_pptr;
 347
 348         if ((t->pgrp != current->pgrp) &&
 349             (t->session == current->session) &&
 350             will_become_orphaned_pgrp(current->pgrp, current) &&
 351             has_stopped_jobs(current->pgrp)) {
 352                 kill_pg(current->pgrp,SIGHUP,1);
 353                 kill_pg(current->pgrp,SIGCONT,1);
 354         }
 355
 356         /* Let father know we died
 357          *
 358          * Thread signals are configurable, but you aren't going to use
 359          * that to send signals to arbitary processes.
 360          * That stops right now.
 361          *
 362          * If the parent exec id doesn't match the exec id we saved
 363          * when we started then we know the parent has changed security
 364          * domain.
 365          *
 366          * If our self_exec id doesn't match our parent_exec_id then
 367          * we have changed execution domain as these two values started
 368          * the same after a fork.
 369          *
 370          */
 371
 372         if(current->exit_signal != SIGCHLD &&
 373             ( current->parent_exec_id != t->self_exec_id  ||
 374               current->self_exec_id != current->parent_exec_id)
 375             && !capable(CAP_KILL))
 376                 current->exit_signal = SIGCHLD;
 377
 378
 379         /*
 380          * This loop does two things:
 381          *
 382          * A.  Make init inherit all the child processes
 383          * B.  Check to see if any process groups have become orphaned
 384          *      as a result of our exiting, and if they have any stopped
 385          *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
 386          */
 387
 388         write_lock_irq(&tasklist_lock);
 389         current->state = TASK_ZOMBIE;
 390         do_notify_parent(current, current->exit_signal);
 391         while (current->p_cptr != NULL) {
 392                 p = current->p_cptr;
 393                 current->p_cptr = p->p_osptr;
 394                 p->p_ysptr = NULL;
 395                 p->ptrace = 0;
 396
 397                 p->p_pptr = p->p_opptr;
 398                 p->p_osptr = p->p_pptr->p_cptr;
 399                 if (p->p_osptr)
 400                         p->p_osptr->p_ysptr = p;
 401                 p->p_pptr->p_cptr = p;
 402                 if (p->state == TASK_ZOMBIE)
 403                         do_notify_parent(p, p->exit_signal);
 404                 /*
 405                  * process group orphan check
 406                  * Case ii: Our child is in a different pgrp
 407                  * than we are, and it was the only connection
 408                  * outside, so the child pgrp is now orphaned.
 409                  */
 410                 if ((p->pgrp != current->pgrp) &&
 411                     (p->session == current->session)) {
 412                         int pgrp = p->pgrp;
 413
 414                         write_unlock_irq(&tasklist_lock);
 415                         if (is_orphaned_pgrp(pgrp) && has_stopped_jobs(pgrp)) {
 416                                 kill_pg(pgrp,SIGHUP,1);
 417                                 kill_pg(pgrp,SIGCONT,1);
 418                         }
 419                         write_lock_irq(&tasklist_lock);
 420                 }
 421         }
 422         write_unlock_irq(&tasklist_lock);
 423 }
 424
 425 NORET_TYPE void do_exit(long code)
 426 {
 427         struct task_struct *tsk = current;
 428
 429         if (in_interrupt())
 430                 panic("Aiee, killing interrupt handler!");
 431         if (!tsk->pid)
 432                 panic("Attempted to kill the idle task!");
 433         if (tsk->pid == 1)
 434                 panic("Attempted to kill init!");
 435         tsk->flags |= PF_EXITING;
 436         del_timer_sync(&tsk->real_timer);
 437
 438 fake_volatile:
 439 #ifdef CONFIG_BSD_PROCESS_ACCT
 440         acct_process(code);
 441 #endif
 442         __exit_mm(tsk);
 443
 444         lock_kernel();
 445         sem_exit();
 446         __exit_files(tsk);
 447         __exit_fs(tsk);
 448         exit_namespace(tsk);
 449         exit_sighand(tsk);
 450         exit_thread();
 451
 452         if (current->leader)
 453                 disassociate_ctty(1);
 454
 455         put_exec_domain(tsk->exec_domain);
 456         if (tsk->binfmt && tsk->binfmt->module)
 457                 __MOD_DEC_USE_COUNT(tsk->binfmt->module);
 458
 459         tsk->exit_code = code;
 460         exit_notify();
 461         schedule();
 462         BUG();
 463 /*
 464  * In order to get rid of the "volatile function does return" message
 465  * I did this little loop that confuses gcc to think do_exit really
 466  * is volatile. In fact it's schedule() that is volatile in some
 467  * circumstances: when current->state = ZOMBIE, schedule() never
 468  * returns.
 469  *
 470  * In fact the natural way to do all this is to have the label and the
 471  * goto right after each other, but I put the fake_volatile label at
 472  * the start of the function just in case something /really/ bad
 473  * happens, and the schedule returns. This way we can try again. I'm
 474  * not paranoid: it's just that everybody is out to get me.
 475  */
 476         goto fake_volatile;
 477 }
 478
 479 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 480 {
 481         if (comp)
 482                 complete(comp);
 483
 484         do_exit(code);
 485 }
 486
 487 asmlinkage long sys_exit(int error_code)
 488 {
 489         do_exit((error_code&0xff)<<8);
 490 }
 491
 492 asmlinkage long sys_wait4(pid_t pid,unsigned int * stat_addr, int options, struct rusage * ru)
 493 {
 494         int flag, retval;
 495         DECLARE_WAITQUEUE(wait, current);
 496         struct task_struct *tsk;
 497
 498         if (options & ~(WNOHANG|WUNTRACED|__WNOTHREAD|__WCLONE|__WALL))
 499                 return -EINVAL;
 500
 501         add_wait_queue(&current->wait_chldexit,&wait);
 502 repeat:
 503         flag = 0;
 504         current->state = TASK_INTERRUPTIBLE;
 505         read_lock(&tasklist_lock);
 506         tsk = current;
 507         do {
 508                 struct task_struct *p;
 509                 for (p = tsk->p_cptr ; p ; p = p->p_osptr) {
 510                         if (pid>0) {
 511                                 if (p->pid != pid)
 512                                         continue;
 513                         } else if (!pid) {
 514                                 if (p->pgrp != current->pgrp)
 515                                         continue;
 516                         } else if (pid != -1) {
 517                                 if (p->pgrp != -pid)
 518                                         continue;
 519                         }
 520                         /* Wait for all children (clone and not) if __WALL is set;
 521                          * otherwise, wait for clone children *only* if __WCLONE is
 522                          * set; otherwise, wait for non-clone children *only*.  (Note:
 523                          * A "clone" child here is one that reports to its parent
 524                          * using a signal other than SIGCHLD.) */
 525                         if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
 526                             && !(options & __WALL))
 527                                 continue;
 528                         flag = 1;
 529                         switch (p->state) {
 530                         case TASK_STOPPED:
 531                                 if (!p->exit_code)
 532                                         continue;
 533                                 if (!(options & WUNTRACED) && !(p->ptrace & PT_PTRACED))
 534                                         continue;
 535                                 read_unlock(&tasklist_lock);
 536                                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
 537                                 if (!retval && stat_addr)
 538                                         retval = put_user((p->exit_code << 8) | 0x7f, stat_addr);
 539                                 if (!retval) {
 540                                         p->exit_code = 0;
 541                                         retval = p->pid;
 542                                 }
 543                                 goto end_wait4;
 544                         case TASK_ZOMBIE:
 545                                 current->times.tms_cutime += p->times.tms_utime + p->times.tms_cutime;
 546                                 current->times.tms_cstime += p->times.tms_stime + p->times.tms_cstime;
 547                                 read_unlock(&tasklist_lock);
 548                                 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
 549                                 if (!retval && stat_addr)
 550                                         retval = put_user(p->exit_code, stat_addr);
 551                                 if (retval)
 552                                         goto end_wait4;
 553                                 retval = p->pid;
 554                                 if (p->p_opptr != p->p_pptr) {
 555                                         write_lock_irq(&tasklist_lock);
 556                                         REMOVE_LINKS(p);
 557                                         p->p_pptr = p->p_opptr;
 558                                         SET_LINKS(p);
 559                                         do_notify_parent(p, SIGCHLD);
 560                                         write_unlock_irq(&tasklist_lock);
 561                                 } else
 562                                         release_task(p);
 563                                 goto end_wait4;
 564                         default:
 565                                 continue;
 566                         }
 567                 }
 568                 if (options & __WNOTHREAD)
 569                         break;
 570                 tsk = next_thread(tsk);
 571         } while (tsk != current);
 572         read_unlock(&tasklist_lock);
 573         if (flag) {
 574                 retval = 0;
 575                 if (options & WNOHANG)
 576                         goto end_wait4;
 577                 retval = -ERESTARTSYS;
 578                 if (signal_pending(current))
 579                         goto end_wait4;
 580                 schedule();
 581                 goto repeat;
 582         }
 583         retval = -ECHILD;
 584 end_wait4:
 585         current->state = TASK_RUNNING;
 586         remove_wait_queue(&current->wait_chldexit,&wait);
 587         return retval;
 588 }
 589
 590 #if !defined(__alpha__) && !defined(__ia64__)
 591
 592 /*
 593  * sys_waitpid() remains for compatibility. waitpid() should be
 594  * implemented by calling sys_wait4() from libc.a.
 595  */
 596 asmlinkage long sys_waitpid(pid_t pid,unsigned int * stat_addr, int options)
 597 {
 598         return sys_wait4(pid, stat_addr, options, NULL);
 599 }
 600
 601 #endif