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