SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/mman.h>
  28 #include <linux/a.out.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/init.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/highmem.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/key.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/pid_namespace.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53
  54 #include <asm/uaccess.h>
  55 #include <asm/mmu_context.h>
  56
  57 #ifdef CONFIG_KMOD
  58 #include <linux/kmod.h>
  59 #endif
  60
  61 int core_uses_pid;
  62 char core_pattern[128] = "core";
  63 int suid_dumpable = 0;
  64
  65 EXPORT_SYMBOL(suid_dumpable);
  66 /* The maximal length of core_pattern is also specified in sysctl.c */
  67
  68 static struct linux_binfmt *formats;
  69 static DEFINE_RWLOCK(binfmt_lock);
  70
  71 int register_binfmt(struct linux_binfmt * fmt)
  72 {
  73         struct linux_binfmt ** tmp = &formats;
  74
  75         if (!fmt)
  76                 return -EINVAL;
  77         if (fmt->next)
  78                 return -EBUSY;
  79         write_lock(&binfmt_lock);
  80         while (*tmp) {
  81                 if (fmt == *tmp) {
  82                         write_unlock(&binfmt_lock);
  83                         return -EBUSY;
  84                 }
  85                 tmp = &(*tmp)->next;
  86         }
  87         fmt->next = formats;
  88         formats = fmt;
  89         write_unlock(&binfmt_lock);
  90         return 0;
  91 }
  92
  93 EXPORT_SYMBOL(register_binfmt);
  94
  95 int unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         struct linux_binfmt ** tmp = &formats;
  98
  99         write_lock(&binfmt_lock);
 100         while (*tmp) {
 101                 if (fmt == *tmp) {
 102                         *tmp = fmt->next;
 103                         write_unlock(&binfmt_lock);
 104                         return 0;
 105                 }
 106                 tmp = &(*tmp)->next;
 107         }
 108         write_unlock(&binfmt_lock);
 109         return -EINVAL;
 110 }
 111
 112 EXPORT_SYMBOL(unregister_binfmt);
 113
 114 static inline void put_binfmt(struct linux_binfmt * fmt)
 115 {
 116         module_put(fmt->module);
 117 }
 118
 119 /*
 120  * Note that a shared library must be both readable and executable due to
 121  * security reasons.
 122  *
 123  * Also note that we take the address to load from from the file itself.
 124  */
 125 asmlinkage long sys_uselib(const char __user * library)
 126 {
 127         struct file * file;
 128         struct nameidata nd;
 129         int error;
 130
 131         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 132         if (error)
 133                 goto out;
 134
 135         error = -EINVAL;
 136         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 137                 goto exit;
 138
 139         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 140         if (error)
 141                 goto exit;
 142
 143         file = nameidata_to_filp(&nd, O_RDONLY);
 144         error = PTR_ERR(file);
 145         if (IS_ERR(file))
 146                 goto out;
 147
 148         error = -ENOEXEC;
 149         if(file->f_op) {
 150                 struct linux_binfmt * fmt;
 151
 152                 read_lock(&binfmt_lock);
 153                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 154                         if (!fmt->load_shlib)
 155                                 continue;
 156                         if (!try_module_get(fmt->module))
 157                                 continue;
 158                         read_unlock(&binfmt_lock);
 159                         error = fmt->load_shlib(file);
 160                         read_lock(&binfmt_lock);
 161                         put_binfmt(fmt);
 162                         if (error != -ENOEXEC)
 163                                 break;
 164                 }
 165                 read_unlock(&binfmt_lock);
 166         }
 167         fput(file);
 168 out:
 169         return error;
 170 exit:
 171         release_open_intent(&nd);
 172         path_release(&nd);
 173         goto out;
 174 }
 175
 176 /*
 177  * count() counts the number of strings in array ARGV.
 178  */
 179 static int count(char __user * __user * argv, int max)
 180 {
 181         int i = 0;
 182
 183         if (argv != NULL) {
 184                 for (;;) {
 185                         char __user * p;
 186
 187                         if (get_user(p, argv))
 188                                 return -EFAULT;
 189                         if (!p)
 190                                 break;
 191                         argv++;
 192                         if(++i > max)
 193                                 return -E2BIG;
 194                         cond_resched();
 195                 }
 196         }
 197         return i;
 198 }
 199
 200 /*
 201  * 'copy_strings()' copies argument/environment strings from user
 202  * memory to free pages in kernel mem. These are in a format ready
 203  * to be put directly into the top of new user memory.
 204  */
 205 static int copy_strings(int argc, char __user * __user * argv,
 206                         struct linux_binprm *bprm)
 207 {
 208         struct page *kmapped_page = NULL;
 209         char *kaddr = NULL;
 210         int ret;
 211
 212         while (argc-- > 0) {
 213                 char __user *str;
 214                 int len;
 215                 unsigned long pos;
 216
 217                 if (get_user(str, argv+argc) ||
 218                                 !(len = strnlen_user(str, bprm->p))) {
 219                         ret = -EFAULT;
 220                         goto out;
 221                 }
 222
 223                 if (bprm->p < len)  {
 224                         ret = -E2BIG;
 225                         goto out;
 226                 }
 227
 228                 bprm->p -= len;
 229                 /* XXX: add architecture specific overflow check here. */
 230                 pos = bprm->p;
 231
 232                 while (len > 0) {
 233                         int i, new, err;
 234                         int offset, bytes_to_copy;
 235                         struct page *page;
 236
 237                         offset = pos % PAGE_SIZE;
 238                         i = pos/PAGE_SIZE;
 239                         page = bprm->page[i];
 240                         new = 0;
 241                         if (!page) {
 242                                 page = alloc_page(GFP_HIGHUSER);
 243                                 bprm->page[i] = page;
 244                                 if (!page) {
 245                                         ret = -ENOMEM;
 246                                         goto out;
 247                                 }
 248                                 new = 1;
 249                         }
 250
 251                         if (page != kmapped_page) {
 252                                 if (kmapped_page)
 253                                         kunmap(kmapped_page);
 254                                 kmapped_page = page;
 255                                 kaddr = kmap(kmapped_page);
 256                         }
 257                         if (new && offset)
 258                                 memset(kaddr, 0, offset);
 259                         bytes_to_copy = PAGE_SIZE - offset;
 260                         if (bytes_to_copy > len) {
 261                                 bytes_to_copy = len;
 262                                 if (new)
 263                                         memset(kaddr+offset+len, 0,
 264                                                 PAGE_SIZE-offset-len);
 265                         }
 266                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 267                         if (err) {
 268                                 ret = -EFAULT;
 269                                 goto out;
 270                         }
 271
 272                         pos += bytes_to_copy;
 273                         str += bytes_to_copy;
 274                         len -= bytes_to_copy;
 275                 }
 276         }
 277         ret = 0;
 278 out:
 279         if (kmapped_page)
 280                 kunmap(kmapped_page);
 281         return ret;
 282 }
 283
 284 /*
 285  * Like copy_strings, but get argv and its values from kernel memory.
 286  */
 287 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 288 {
 289         int r;
 290         mm_segment_t oldfs = get_fs();
 291         set_fs(KERNEL_DS);
 292         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 293         set_fs(oldfs);
 294         return r;
 295 }
 296
 297 EXPORT_SYMBOL(copy_strings_kernel);
 298
 299 #ifdef CONFIG_MMU
 300 /*
 301  * This routine is used to map in a page into an address space: needed by
 302  * execve() for the initial stack and environment pages.
 303  *
 304  * vma->vm_mm->mmap_sem is held for writing.
 305  */
 306 void install_arg_page(struct vm_area_struct *vma,
 307                         struct page *page, unsigned long address)
 308 {
 309         struct mm_struct *mm = vma->vm_mm;
 310         pte_t * pte;
 311         spinlock_t *ptl;
 312
 313         if (unlikely(anon_vma_prepare(vma)))
 314                 goto out;
 315
 316         flush_dcache_page(page);
 317         pte = get_locked_pte(mm, address, &ptl);
 318         if (!pte)
 319                 goto out;
 320         if (!pte_none(*pte)) {
 321                 pte_unmap_unlock(pte, ptl);
 322                 goto out;
 323         }
 324         inc_mm_counter(mm, anon_rss);
 325         lru_cache_add_active(page);
 326         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
 327                                         page, vma->vm_page_prot))));
 328         page_add_new_anon_rmap(page, vma, address);
 329         pte_unmap_unlock(pte, ptl);
 330
 331         /* no need for flush_tlb */
 332         return;
 333 out:
 334         __free_page(page);
 335         force_sig(SIGKILL, current);
 336 }
 337
 338 #define EXTRA_STACK_VM_PAGES    20      /* random */
 339
 340 int setup_arg_pages(struct linux_binprm *bprm,
 341                     unsigned long stack_top,
 342                     int executable_stack)
 343 {
 344         unsigned long stack_base;
 345         struct vm_area_struct *mpnt;
 346         struct mm_struct *mm = current->mm;
 347         int i, ret;
 348         long arg_size;
 349
 350 #ifdef CONFIG_STACK_GROWSUP
 351         /* Move the argument and environment strings to the bottom of the
 352          * stack space.
 353          */
 354         int offset, j;
 355         char *to, *from;
 356
 357         /* Start by shifting all the pages down */
 358         i = 0;
 359         for (j = 0; j < MAX_ARG_PAGES; j++) {
 360                 struct page *page = bprm->page[j];
 361                 if (!page)
 362                         continue;
 363                 bprm->page[i++] = page;
 364         }
 365
 366         /* Now move them within their pages */
 367         offset = bprm->p % PAGE_SIZE;
 368         to = kmap(bprm->page[0]);
 369         for (j = 1; j < i; j++) {
 370                 memmove(to, to + offset, PAGE_SIZE - offset);
 371                 from = kmap(bprm->page[j]);
 372                 memcpy(to + PAGE_SIZE - offset, from, offset);
 373                 kunmap(bprm->page[j - 1]);
 374                 to = from;
 375         }
 376         memmove(to, to + offset, PAGE_SIZE - offset);
 377         kunmap(bprm->page[j - 1]);
 378
 379         /* Limit stack size to 1GB */
 380         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 381         if (stack_base > (1 << 30))
 382                 stack_base = 1 << 30;
 383         stack_base = PAGE_ALIGN(stack_top - stack_base);
 384
 385         /* Adjust bprm->p to point to the end of the strings. */
 386         bprm->p = stack_base + PAGE_SIZE * i - offset;
 387
 388         mm->arg_start = stack_base;
 389         arg_size = i << PAGE_SHIFT;
 390
 391         /* zero pages that were copied above */
 392         while (i < MAX_ARG_PAGES)
 393                 bprm->page[i++] = NULL;
 394 #else
 395         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
 396         stack_base = PAGE_ALIGN(stack_base);
 397         bprm->p += stack_base;
 398         mm->arg_start = bprm->p;
 399         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 400 #endif
 401
 402         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 403
 404         if (bprm->loader)
 405                 bprm->loader += stack_base;
 406         bprm->exec += stack_base;
 407
 408         mpnt = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 409         if (!mpnt)
 410                 return -ENOMEM;
 411
 412         down_write(&mm->mmap_sem);
 413         {
 414                 mpnt->vm_mm = mm;
 415 #ifdef CONFIG_STACK_GROWSUP
 416                 mpnt->vm_start = stack_base;
 417                 mpnt->vm_end = stack_base + arg_size;
 418 #else
 419                 mpnt->vm_end = stack_top;
 420                 mpnt->vm_start = mpnt->vm_end - arg_size;
 421 #endif
 422                 /* Adjust stack execute permissions; explicitly enable
 423                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 424                  * and leave alone (arch default) otherwise. */
 425                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 426                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 427                 else if (executable_stack == EXSTACK_DISABLE_X)
 428                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 429                 else
 430                         mpnt->vm_flags = VM_STACK_FLAGS;
 431                 mpnt->vm_flags |= mm->def_flags;
 432                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 433                 if ((ret = insert_vm_struct(mm, mpnt))) {
 434                         up_write(&mm->mmap_sem);
 435                         kmem_cache_free(vm_area_cachep, mpnt);
 436                         return ret;
 437                 }
 438                 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 439         }
 440
 441         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 442                 struct page *page = bprm->page[i];
 443                 if (page) {
 444                         bprm->page[i] = NULL;
 445                         install_arg_page(mpnt, page, stack_base);
 446                 }
 447                 stack_base += PAGE_SIZE;
 448         }
 449         up_write(&mm->mmap_sem);
 450
 451         return 0;
 452 }
 453
 454 EXPORT_SYMBOL(setup_arg_pages);
 455
 456 #define free_arg_pages(bprm) do { } while (0)
 457
 458 #else
 459
 460 static inline void free_arg_pages(struct linux_binprm *bprm)
 461 {
 462         int i;
 463
 464         for (i = 0; i < MAX_ARG_PAGES; i++) {
 465                 if (bprm->page[i])
 466                         __free_page(bprm->page[i]);
 467                 bprm->page[i] = NULL;
 468         }
 469 }
 470
 471 #endif /* CONFIG_MMU */
 472
 473 struct file *open_exec(const char *name)
 474 {
 475         struct nameidata nd;
 476         int err;
 477         struct file *file;
 478
 479         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 480         file = ERR_PTR(err);
 481
 482         if (!err) {
 483                 struct inode *inode = nd.dentry->d_inode;
 484                 file = ERR_PTR(-EACCES);
 485                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 486                     S_ISREG(inode->i_mode)) {
 487                         int err = vfs_permission(&nd, MAY_EXEC);
 488                         file = ERR_PTR(err);
 489                         if (!err) {
 490                                 file = nameidata_to_filp(&nd, O_RDONLY);
 491                                 if (!IS_ERR(file)) {
 492                                         err = deny_write_access(file);
 493                                         if (err) {
 494                                                 fput(file);
 495                                                 file = ERR_PTR(err);
 496                                         }
 497                                 }
 498 out:
 499                                 return file;
 500                         }
 501                 }
 502                 release_open_intent(&nd);
 503                 path_release(&nd);
 504         }
 505         goto out;
 506 }
 507
 508 EXPORT_SYMBOL(open_exec);
 509
 510 int kernel_read(struct file *file, unsigned long offset,
 511         char *addr, unsigned long count)
 512 {
 513         mm_segment_t old_fs;
 514         loff_t pos = offset;
 515         int result;
 516
 517         old_fs = get_fs();
 518         set_fs(get_ds());
 519         /* The cast to a user pointer is valid due to the set_fs() */
 520         result = vfs_read(file, (void __user *)addr, count, &pos);
 521         set_fs(old_fs);
 522         return result;
 523 }
 524
 525 EXPORT_SYMBOL(kernel_read);
 526
 527 static int exec_mmap(struct mm_struct *mm)
 528 {
 529         struct task_struct *tsk;
 530         struct mm_struct * old_mm, *active_mm;
 531
 532         /* Notify parent that we're no longer interested in the old VM */
 533         tsk = current;
 534         old_mm = current->mm;
 535         mm_release(tsk, old_mm);
 536
 537         if (old_mm) {
 538                 /*
 539                  * Make sure that if there is a core dump in progress
 540                  * for the old mm, we get out and die instead of going
 541                  * through with the exec.  We must hold mmap_sem around
 542                  * checking core_waiters and changing tsk->mm.  The
 543                  * core-inducing thread will increment core_waiters for
 544                  * each thread whose ->mm == old_mm.
 545                  */
 546                 down_read(&old_mm->mmap_sem);
 547                 if (unlikely(old_mm->core_waiters)) {
 548                         up_read(&old_mm->mmap_sem);
 549                         return -EINTR;
 550                 }
 551         }
 552         task_lock(tsk);
 553         active_mm = tsk->active_mm;
 554         tsk->mm = mm;
 555         tsk->active_mm = mm;
 556         activate_mm(active_mm, mm);
 557         task_unlock(tsk);
 558         arch_pick_mmap_layout(mm);
 559         if (old_mm) {
 560                 up_read(&old_mm->mmap_sem);
 561                 BUG_ON(active_mm != old_mm);
 562                 mmput(old_mm);
 563                 return 0;
 564         }
 565         mmdrop(active_mm);
 566         return 0;
 567 }
 568
 569 /*
 570  * This function makes sure the current process has its own signal table,
 571  * so that flush_signal_handlers can later reset the handlers without
 572  * disturbing other processes.  (Other processes might share the signal
 573  * table via the CLONE_SIGHAND option to clone().)
 574  */
 575 static int de_thread(struct task_struct *tsk)
 576 {
 577         struct signal_struct *sig = tsk->signal;
 578         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 579         spinlock_t *lock = &oldsighand->siglock;
 580         struct task_struct *leader = NULL;
 581         int count;
 582
 583         /*
 584          * If we don't share sighandlers, then we aren't sharing anything
 585          * and we can just re-use it all.
 586          */
 587         if (atomic_read(&oldsighand->count) <= 1) {
 588                 BUG_ON(atomic_read(&sig->count) != 1);
 589                 exit_itimers(sig);
 590                 return 0;
 591         }
 592
 593         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 594         if (!newsighand)
 595                 return -ENOMEM;
 596
 597         if (thread_group_empty(tsk))
 598                 goto no_thread_group;
 599
 600         /*
 601          * Kill all other threads in the thread group.
 602          * We must hold tasklist_lock to call zap_other_threads.
 603          */
 604         read_lock(&tasklist_lock);
 605         spin_lock_irq(lock);
 606         if (sig->flags & SIGNAL_GROUP_EXIT) {
 607                 /*
 608                  * Another group action in progress, just
 609                  * return so that the signal is processed.
 610                  */
 611                 spin_unlock_irq(lock);
 612                 read_unlock(&tasklist_lock);
 613                 kmem_cache_free(sighand_cachep, newsighand);
 614                 return -EAGAIN;
 615         }
 616
 617         /*
 618          * child_reaper ignores SIGKILL, change it now.
 619          * Reparenting needs write_lock on tasklist_lock,
 620          * so it is safe to do it under read_lock.
 621          */
 622         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 623                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 624
 625         zap_other_threads(tsk);
 626         read_unlock(&tasklist_lock);
 627
 628         /*
 629          * Account for the thread group leader hanging around:
 630          */
 631         count = 1;
 632         if (!thread_group_leader(tsk)) {
 633                 count = 2;
 634                 /*
 635                  * The SIGALRM timer survives the exec, but needs to point
 636                  * at us as the new group leader now.  We have a race with
 637                  * a timer firing now getting the old leader, so we need to
 638                  * synchronize with any firing (by calling del_timer_sync)
 639                  * before we can safely let the old group leader die.
 640                  */
 641                 sig->tsk = tsk;
 642                 spin_unlock_irq(lock);
 643                 if (hrtimer_cancel(&sig->real_timer))
 644                         hrtimer_restart(&sig->real_timer);
 645                 spin_lock_irq(lock);
 646         }
 647         while (atomic_read(&sig->count) > count) {
 648                 sig->group_exit_task = tsk;
 649                 sig->notify_count = count;
 650                 __set_current_state(TASK_UNINTERRUPTIBLE);
 651                 spin_unlock_irq(lock);
 652                 schedule();
 653                 spin_lock_irq(lock);
 654         }
 655         sig->group_exit_task = NULL;
 656         sig->notify_count = 0;
 657         spin_unlock_irq(lock);
 658
 659         /*
 660          * At this point all other threads have exited, all we have to
 661          * do is to wait for the thread group leader to become inactive,
 662          * and to assume its PID:
 663          */
 664         if (!thread_group_leader(tsk)) {
 665                 /*
 666                  * Wait for the thread group leader to be a zombie.
 667                  * It should already be zombie at this point, most
 668                  * of the time.
 669                  */
 670                 leader = tsk->group_leader;
 671                 while (leader->exit_state != EXIT_ZOMBIE)
 672                         yield();
 673
 674                 /*
 675                  * The only record we have of the real-time age of a
 676                  * process, regardless of execs it's done, is start_time.
 677                  * All the past CPU time is accumulated in signal_struct
 678                  * from sister threads now dead.  But in this non-leader
 679                  * exec, nothing survives from the original leader thread,
 680                  * whose birth marks the true age of this process now.
 681                  * When we take on its identity by switching to its PID, we
 682                  * also take its birthdate (always earlier than our own).
 683                  */
 684                 tsk->start_time = leader->start_time;
 685
 686                 write_lock_irq(&tasklist_lock);
 687
 688                 BUG_ON(leader->tgid != tsk->tgid);
 689                 BUG_ON(tsk->pid == tsk->tgid);
 690                 /*
 691                  * An exec() starts a new thread group with the
 692                  * TGID of the previous thread group. Rehash the
 693                  * two threads with a switched PID, and release
 694                  * the former thread group leader:
 695                  */
 696
 697                 /* Become a process group leader with the old leader's pid.
 698                  * The old leader becomes a thread of the this thread group.
 699                  * Note: The old leader also uses this pid until release_task
 700                  *       is called.  Odd but simple and correct.
 701                  */
 702                 detach_pid(tsk, PIDTYPE_PID);
 703                 tsk->pid = leader->pid;
 704                 attach_pid(tsk, PIDTYPE_PID,  tsk->pid);
 705                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 706                 transfer_pid(leader, tsk, PIDTYPE_SID);
 707                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 708
 709                 tsk->group_leader = tsk;
 710                 leader->group_leader = tsk;
 711
 712                 tsk->exit_signal = SIGCHLD;
 713
 714                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 715                 leader->exit_state = EXIT_DEAD;
 716
 717                 write_unlock_irq(&tasklist_lock);
 718         }
 719
 720         /*
 721          * There may be one thread left which is just exiting,
 722          * but it's safe to stop telling the group to kill themselves.
 723          */
 724         sig->flags = 0;
 725
 726 no_thread_group:
 727         exit_itimers(sig);
 728         if (leader)
 729                 release_task(leader);
 730
 731         BUG_ON(atomic_read(&sig->count) != 1);
 732
 733         if (atomic_read(&oldsighand->count) == 1) {
 734                 /*
 735                  * Now that we nuked the rest of the thread group,
 736                  * it turns out we are not sharing sighand any more either.
 737                  * So we can just keep it.
 738                  */
 739                 kmem_cache_free(sighand_cachep, newsighand);
 740         } else {
 741                 /*
 742                  * Move our state over to newsighand and switch it in.
 743                  */
 744                 atomic_set(&newsighand->count, 1);
 745                 memcpy(newsighand->action, oldsighand->action,
 746                        sizeof(newsighand->action));
 747
 748                 write_lock_irq(&tasklist_lock);
 749                 spin_lock(&oldsighand->siglock);
 750                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 751
 752                 rcu_assign_pointer(tsk->sighand, newsighand);
 753                 recalc_sigpending();
 754
 755                 spin_unlock(&newsighand->siglock);
 756                 spin_unlock(&oldsighand->siglock);
 757                 write_unlock_irq(&tasklist_lock);
 758
 759                 if (atomic_dec_and_test(&oldsighand->count))
 760                         kmem_cache_free(sighand_cachep, oldsighand);
 761         }
 762
 763         BUG_ON(!thread_group_leader(tsk));
 764         return 0;
 765 }
 766
 767 /*
 768  * These functions flushes out all traces of the currently running executable
 769  * so that a new one can be started
 770  */
 771
 772 static void flush_old_files(struct files_struct * files)
 773 {
 774         long j = -1;
 775         struct fdtable *fdt;
 776
 777         spin_lock(&files->file_lock);
 778         for (;;) {
 779                 unsigned long set, i;
 780
 781                 j++;
 782                 i = j * __NFDBITS;
 783                 fdt = files_fdtable(files);
 784                 if (i >= fdt->max_fds)
 785                         break;
 786                 set = fdt->close_on_exec->fds_bits[j];
 787                 if (!set)
 788                         continue;
 789                 fdt->close_on_exec->fds_bits[j] = 0;
 790                 spin_unlock(&files->file_lock);
 791                 for ( ; set ; i++,set >>= 1) {
 792                         if (set & 1) {
 793                                 sys_close(i);
 794                         }
 795                 }
 796                 spin_lock(&files->file_lock);
 797
 798         }
 799         spin_unlock(&files->file_lock);
 800 }
 801
 802 void get_task_comm(char *buf, struct task_struct *tsk)
 803 {
 804         /* buf must be at least sizeof(tsk->comm) in size */
 805         task_lock(tsk);
 806         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 807         task_unlock(tsk);
 808 }
 809
 810 void set_task_comm(struct task_struct *tsk, char *buf)
 811 {
 812         task_lock(tsk);
 813         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 814         task_unlock(tsk);
 815 }
 816
 817 int flush_old_exec(struct linux_binprm * bprm)
 818 {
 819         char * name;
 820         int i, ch, retval;
 821         struct files_struct *files;
 822         char tcomm[sizeof(current->comm)];
 823
 824         /*
 825          * Make sure we have a private signal table and that
 826          * we are unassociated from the previous thread group.
 827          */
 828         retval = de_thread(current);
 829         if (retval)
 830                 goto out;
 831
 832         /*
 833          * Make sure we have private file handles. Ask the
 834          * fork helper to do the work for us and the exit
 835          * helper to do the cleanup of the old one.
 836          */
 837         files = current->files;         /* refcounted so safe to hold */
 838         retval = unshare_files();
 839         if (retval)
 840                 goto out;
 841         /*
 842          * Release all of the old mmap stuff
 843          */
 844         retval = exec_mmap(bprm->mm);
 845         if (retval)
 846                 goto mmap_failed;
 847
 848         bprm->mm = NULL;                /* We're using it now */
 849
 850         /* This is the point of no return */
 851         put_files_struct(files);
 852
 853         current->sas_ss_sp = current->sas_ss_size = 0;
 854
 855         if (current->euid == current->uid && current->egid == current->gid)
 856                 current->mm->dumpable = 1;
 857         else
 858                 current->mm->dumpable = suid_dumpable;
 859
 860         name = bprm->filename;
 861
 862         /* Copies the binary name from after last slash */
 863         for (i=0; (ch = *(name++)) != '\0';) {
 864                 if (ch == '/')
 865                         i = 0; /* overwrite what we wrote */
 866                 else
 867                         if (i < (sizeof(tcomm) - 1))
 868                                 tcomm[i++] = ch;
 869         }
 870         tcomm[i] = '\0';
 871         set_task_comm(current, tcomm);
 872
 873         current->flags &= ~PF_RANDOMIZE;
 874         flush_thread();
 875
 876         /* Set the new mm task size. We have to do that late because it may
 877          * depend on TIF_32BIT which is only updated in flush_thread() on
 878          * some architectures like powerpc
 879          */
 880         current->mm->task_size = TASK_SIZE;
 881
 882         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 883             file_permission(bprm->file, MAY_READ) ||
 884             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 885                 suid_keys(current);
 886                 current->mm->dumpable = suid_dumpable;
 887         }
 888
 889         /* An exec changes our domain. We are no longer part of the thread
 890            group */
 891
 892         current->self_exec_id++;
 893
 894         flush_signal_handlers(current, 0);
 895         flush_old_files(current->files);
 896
 897         return 0;
 898
 899 mmap_failed:
 900         reset_files_struct(current, files);
 901 out:
 902         return retval;
 903 }
 904
 905 EXPORT_SYMBOL(flush_old_exec);
 906
 907 /*
 908  * Fill the binprm structure from the inode.
 909  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 910  */
 911 int prepare_binprm(struct linux_binprm *bprm)
 912 {
 913         int mode;
 914         struct inode * inode = bprm->file->f_path.dentry->d_inode;
 915         int retval;
 916
 917         mode = inode->i_mode;
 918         if (bprm->file->f_op == NULL)
 919                 return -EACCES;
 920
 921         bprm->e_uid = current->euid;
 922         bprm->e_gid = current->egid;
 923
 924         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
 925                 /* Set-uid? */
 926                 if (mode & S_ISUID) {
 927                         current->personality &= ~PER_CLEAR_ON_SETID;
 928                         bprm->e_uid = inode->i_uid;
 929                 }
 930
 931                 /* Set-gid? */
 932                 /*
 933                  * If setgid is set but no group execute bit then this
 934                  * is a candidate for mandatory locking, not a setgid
 935                  * executable.
 936                  */
 937                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 938                         current->personality &= ~PER_CLEAR_ON_SETID;
 939                         bprm->e_gid = inode->i_gid;
 940                 }
 941         }
 942
 943         /* fill in binprm security blob */
 944         retval = security_bprm_set(bprm);
 945         if (retval)
 946                 return retval;
 947
 948         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 949         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 950 }
 951
 952 EXPORT_SYMBOL(prepare_binprm);
 953
 954 static int unsafe_exec(struct task_struct *p)
 955 {
 956         int unsafe = 0;
 957         if (p->ptrace & PT_PTRACED) {
 958                 if (p->ptrace & PT_PTRACE_CAP)
 959                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 960                 else
 961                         unsafe |= LSM_UNSAFE_PTRACE;
 962         }
 963         if (atomic_read(&p->fs->count) > 1 ||
 964             atomic_read(&p->files->count) > 1 ||
 965             atomic_read(&p->sighand->count) > 1)
 966                 unsafe |= LSM_UNSAFE_SHARE;
 967
 968         return unsafe;
 969 }
 970
 971 void compute_creds(struct linux_binprm *bprm)
 972 {
 973         int unsafe;
 974
 975         if (bprm->e_uid != current->uid)
 976                 suid_keys(current);
 977         exec_keys(current);
 978
 979         task_lock(current);
 980         unsafe = unsafe_exec(current);
 981         security_bprm_apply_creds(bprm, unsafe);
 982         task_unlock(current);
 983         security_bprm_post_apply_creds(bprm);
 984 }
 985 EXPORT_SYMBOL(compute_creds);
 986
 987 /*
 988  * Arguments are '\0' separated strings found at the location bprm->p
 989  * points to; chop off the first by relocating brpm->p to right after
 990  * the first '\0' encountered.
 991  */
 992 void remove_arg_zero(struct linux_binprm *bprm)
 993 {
 994         if (bprm->argc) {
 995                 char ch;
 996
 997                 do {
 998                         unsigned long offset;
 999                         unsigned long index;
1000                         char *kaddr;
1001                         struct page *page;
1002
1003                         offset = bprm->p & ~PAGE_MASK;
1004                         index = bprm->p >> PAGE_SHIFT;
1005
1006                         page = bprm->page[index];
1007                         kaddr = kmap_atomic(page, KM_USER0);
1008
1009                         /* run through page until we reach end or find NUL */
1010                         do {
1011                                 ch = *(kaddr + offset);
1012
1013                                 /* discard that character... */
1014                                 bprm->p++;
1015                                 offset++;
1016                         } while (offset < PAGE_SIZE && ch != '\0');
1017
1018                         kunmap_atomic(kaddr, KM_USER0);
1019
1020                         /* free the old page */
1021                         if (offset == PAGE_SIZE) {
1022                                 __free_page(page);
1023                                 bprm->page[index] = NULL;
1024                         }
1025                 } while (ch != '\0');
1026
1027                 bprm->argc--;
1028         }
1029 }
1030 EXPORT_SYMBOL(remove_arg_zero);
1031
1032 /*
1033  * cycle the list of binary formats handler, until one recognizes the image
1034  */
1035 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1036 {
1037         int try,retval;
1038         struct linux_binfmt *fmt;
1039 #ifdef __alpha__
1040         /* handle /sbin/loader.. */
1041         {
1042             struct exec * eh = (struct exec *) bprm->buf;
1043
1044             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1045                 (eh->fh.f_flags & 0x3000) == 0x3000)
1046             {
1047                 struct file * file;
1048                 unsigned long loader;
1049
1050                 allow_write_access(bprm->file);
1051                 fput(bprm->file);
1052                 bprm->file = NULL;
1053
1054                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1055
1056                 file = open_exec("/sbin/loader");
1057                 retval = PTR_ERR(file);
1058                 if (IS_ERR(file))
1059                         return retval;
1060
1061                 /* Remember if the application is TASO.  */
1062                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1063
1064                 bprm->file = file;
1065                 bprm->loader = loader;
1066                 retval = prepare_binprm(bprm);
1067                 if (retval<0)
1068                         return retval;
1069                 /* should call search_binary_handler recursively here,
1070                    but it does not matter */
1071             }
1072         }
1073 #endif
1074         retval = security_bprm_check(bprm);
1075         if (retval)
1076                 return retval;
1077
1078         /* kernel module loader fixup */
1079         /* so we don't try to load run modprobe in kernel space. */
1080         set_fs(USER_DS);
1081
1082         retval = audit_bprm(bprm);
1083         if (retval)
1084                 return retval;
1085
1086         retval = -ENOENT;
1087         for (try=0; try<2; try++) {
1088                 read_lock(&binfmt_lock);
1089                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1090                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1091                         if (!fn)
1092                                 continue;
1093                         if (!try_module_get(fmt->module))
1094                                 continue;
1095                         read_unlock(&binfmt_lock);
1096                         retval = fn(bprm, regs);
1097                         if (retval >= 0) {
1098                                 put_binfmt(fmt);
1099                                 allow_write_access(bprm->file);
1100                                 if (bprm->file)
1101                                         fput(bprm->file);
1102                                 bprm->file = NULL;
1103                                 current->did_exec = 1;
1104                                 proc_exec_connector(current);
1105                                 return retval;
1106                         }
1107                         read_lock(&binfmt_lock);
1108                         put_binfmt(fmt);
1109                         if (retval != -ENOEXEC || bprm->mm == NULL)
1110                                 break;
1111                         if (!bprm->file) {
1112                                 read_unlock(&binfmt_lock);
1113                                 return retval;
1114                         }
1115                 }
1116                 read_unlock(&binfmt_lock);
1117                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1118                         break;
1119 #ifdef CONFIG_KMOD
1120                 }else{
1121 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1122                         if (printable(bprm->buf[0]) &&
1123                             printable(bprm->buf[1]) &&
1124                             printable(bprm->buf[2]) &&
1125                             printable(bprm->buf[3]))
1126                                 break; /* -ENOEXEC */
1127                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1128 #endif
1129                 }
1130         }
1131         return retval;
1132 }
1133
1134 EXPORT_SYMBOL(search_binary_handler);
1135
1136 /*
1137  * sys_execve() executes a new program.
1138  */
1139 int do_execve(char * filename,
1140         char __user *__user *argv,
1141         char __user *__user *envp,
1142         struct pt_regs * regs)
1143 {
1144         struct linux_binprm *bprm;
1145         struct file *file;
1146         int retval;
1147         int i;
1148
1149         retval = -ENOMEM;
1150         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1151         if (!bprm)
1152                 goto out_ret;
1153
1154         file = open_exec(filename);
1155         retval = PTR_ERR(file);
1156         if (IS_ERR(file))
1157                 goto out_kfree;
1158
1159         sched_exec();
1160
1161         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1162
1163         bprm->file = file;
1164         bprm->filename = filename;
1165         bprm->interp = filename;
1166         bprm->mm = mm_alloc();
1167         retval = -ENOMEM;
1168         if (!bprm->mm)
1169                 goto out_file;
1170
1171         retval = init_new_context(current, bprm->mm);
1172         if (retval < 0)
1173                 goto out_mm;
1174
1175         bprm->argc = count(argv, bprm->p / sizeof(void *));
1176         if ((retval = bprm->argc) < 0)
1177                 goto out_mm;
1178
1179         bprm->envc = count(envp, bprm->p / sizeof(void *));
1180         if ((retval = bprm->envc) < 0)
1181                 goto out_mm;
1182
1183         retval = security_bprm_alloc(bprm);
1184         if (retval)
1185                 goto out;
1186
1187         retval = prepare_binprm(bprm);
1188         if (retval < 0)
1189                 goto out;
1190
1191         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1192         if (retval < 0)
1193                 goto out;
1194
1195         bprm->exec = bprm->p;
1196         retval = copy_strings(bprm->envc, envp, bprm);
1197         if (retval < 0)
1198                 goto out;
1199
1200         retval = copy_strings(bprm->argc, argv, bprm);
1201         if (retval < 0)
1202                 goto out;
1203
1204         retval = search_binary_handler(bprm,regs);
1205         if (retval >= 0) {
1206                 free_arg_pages(bprm);
1207
1208                 /* execve success */
1209                 security_bprm_free(bprm);
1210                 acct_update_integrals(current);
1211                 kfree(bprm);
1212                 return retval;
1213         }
1214
1215 out:
1216         /* Something went wrong, return the inode and free the argument pages*/
1217         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1218                 struct page * page = bprm->page[i];
1219                 if (page)
1220                         __free_page(page);
1221         }
1222
1223         if (bprm->security)
1224                 security_bprm_free(bprm);
1225
1226 out_mm:
1227         if (bprm->mm)
1228                 mmdrop(bprm->mm);
1229
1230 out_file:
1231         if (bprm->file) {
1232                 allow_write_access(bprm->file);
1233                 fput(bprm->file);
1234         }
1235
1236 out_kfree:
1237         kfree(bprm);
1238
1239 out_ret:
1240         return retval;
1241 }
1242
1243 int set_binfmt(struct linux_binfmt *new)
1244 {
1245         struct linux_binfmt *old = current->binfmt;
1246
1247         if (new) {
1248                 if (!try_module_get(new->module))
1249                         return -1;
1250         }
1251         current->binfmt = new;
1252         if (old)
1253                 module_put(old->module);
1254         return 0;
1255 }
1256
1257 EXPORT_SYMBOL(set_binfmt);
1258
1259 #define CORENAME_MAX_SIZE 64
1260
1261 /* format_corename will inspect the pattern parameter, and output a
1262  * name into corename, which must have space for at least
1263  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1264  */
1265 static int format_corename(char *corename, const char *pattern, long signr)
1266 {
1267         const char *pat_ptr = pattern;
1268         char *out_ptr = corename;
1269         char *const out_end = corename + CORENAME_MAX_SIZE;
1270         int rc;
1271         int pid_in_pattern = 0;
1272         int ispipe = 0;
1273
1274         if (*pattern == '|')
1275                 ispipe = 1;
1276
1277         /* Repeat as long as we have more pattern to process and more output
1278            space */
1279         while (*pat_ptr) {
1280                 if (*pat_ptr != '%') {
1281                         if (out_ptr == out_end)
1282                                 goto out;
1283                         *out_ptr++ = *pat_ptr++;
1284                 } else {
1285                         switch (*++pat_ptr) {
1286                         case 0:
1287                                 goto out;
1288                         /* Double percent, output one percent */
1289                         case '%':
1290                                 if (out_ptr == out_end)
1291                                         goto out;
1292                                 *out_ptr++ = '%';
1293                                 break;
1294                         /* pid */
1295                         case 'p':
1296                                 pid_in_pattern = 1;
1297                                 rc = snprintf(out_ptr, out_end - out_ptr,
1298                                               "%d", current->tgid);
1299                                 if (rc > out_end - out_ptr)
1300                                         goto out;
1301                                 out_ptr += rc;
1302                                 break;
1303                         /* uid */
1304                         case 'u':
1305                                 rc = snprintf(out_ptr, out_end - out_ptr,
1306                                               "%d", current->uid);
1307                                 if (rc > out_end - out_ptr)
1308                                         goto out;
1309                                 out_ptr += rc;
1310                                 break;
1311                         /* gid */
1312                         case 'g':
1313                                 rc = snprintf(out_ptr, out_end - out_ptr,
1314                                               "%d", current->gid);
1315                                 if (rc > out_end - out_ptr)
1316                                         goto out;
1317                                 out_ptr += rc;
1318                                 break;
1319                         /* signal that caused the coredump */
1320                         case 's':
1321                                 rc = snprintf(out_ptr, out_end - out_ptr,
1322                                               "%ld", signr);
1323                                 if (rc > out_end - out_ptr)
1324                                         goto out;
1325                                 out_ptr += rc;
1326                                 break;
1327                         /* UNIX time of coredump */
1328                         case 't': {
1329                                 struct timeval tv;
1330                                 do_gettimeofday(&tv);
1331                                 rc = snprintf(out_ptr, out_end - out_ptr,
1332                                               "%lu", tv.tv_sec);
1333                                 if (rc > out_end - out_ptr)
1334                                         goto out;
1335                                 out_ptr += rc;
1336                                 break;
1337                         }
1338                         /* hostname */
1339                         case 'h':
1340                                 down_read(&uts_sem);
1341                                 rc = snprintf(out_ptr, out_end - out_ptr,
1342                                               "%s", utsname()->nodename);
1343                                 up_read(&uts_sem);
1344                                 if (rc > out_end - out_ptr)
1345                                         goto out;
1346                                 out_ptr += rc;
1347                                 break;
1348                         /* executable */
1349                         case 'e':
1350                                 rc = snprintf(out_ptr, out_end - out_ptr,
1351                                               "%s", current->comm);
1352                                 if (rc > out_end - out_ptr)
1353                                         goto out;
1354                                 out_ptr += rc;
1355                                 break;
1356                         default:
1357                                 break;
1358                         }
1359                         ++pat_ptr;
1360                 }
1361         }
1362         /* Backward compatibility with core_uses_pid:
1363          *
1364          * If core_pattern does not include a %p (as is the default)
1365          * and core_uses_pid is set, then .%pid will be appended to
1366          * the filename. Do not do this for piped commands. */
1367         if (!ispipe && !pid_in_pattern
1368             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1369                 rc = snprintf(out_ptr, out_end - out_ptr,
1370                               ".%d", current->tgid);
1371                 if (rc > out_end - out_ptr)
1372                         goto out;
1373                 out_ptr += rc;
1374         }
1375 out:
1376         *out_ptr = 0;
1377         return ispipe;
1378 }
1379
1380 static void zap_process(struct task_struct *start)
1381 {
1382         struct task_struct *t;
1383
1384         start->signal->flags = SIGNAL_GROUP_EXIT;
1385         start->signal->group_stop_count = 0;
1386
1387         t = start;
1388         do {
1389                 if (t != current && t->mm) {
1390                         t->mm->core_waiters++;
1391                         sigaddset(&t->pending.signal, SIGKILL);
1392                         signal_wake_up(t, 1);
1393                 }
1394         } while ((t = next_thread(t)) != start);
1395 }
1396
1397 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1398                                 int exit_code)
1399 {
1400         struct task_struct *g, *p;
1401         unsigned long flags;
1402         int err = -EAGAIN;
1403
1404         spin_lock_irq(&tsk->sighand->siglock);
1405         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1406                 tsk->signal->group_exit_code = exit_code;
1407                 zap_process(tsk);
1408                 err = 0;
1409         }
1410         spin_unlock_irq(&tsk->sighand->siglock);
1411         if (err)
1412                 return err;
1413
1414         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1415                 goto done;
1416
1417         rcu_read_lock();
1418         for_each_process(g) {
1419                 if (g == tsk->group_leader)
1420                         continue;
1421
1422                 p = g;
1423                 do {
1424                         if (p->mm) {
1425                                 if (p->mm == mm) {
1426                                         /*
1427                                          * p->sighand can't disappear, but
1428                                          * may be changed by de_thread()
1429                                          */
1430                                         lock_task_sighand(p, &flags);
1431                                         zap_process(p);
1432                                         unlock_task_sighand(p, &flags);
1433                                 }
1434                                 break;
1435                         }
1436                 } while ((p = next_thread(p)) != g);
1437         }
1438         rcu_read_unlock();
1439 done:
1440         return mm->core_waiters;
1441 }
1442
1443 static int coredump_wait(int exit_code)
1444 {
1445         struct task_struct *tsk = current;
1446         struct mm_struct *mm = tsk->mm;
1447         struct completion startup_done;
1448         struct completion *vfork_done;
1449         int core_waiters;
1450
1451         init_completion(&mm->core_done);
1452         init_completion(&startup_done);
1453         mm->core_startup_done = &startup_done;
1454
1455         core_waiters = zap_threads(tsk, mm, exit_code);
1456         up_write(&mm->mmap_sem);
1457
1458         if (unlikely(core_waiters < 0))
1459                 goto fail;
1460
1461         /*
1462          * Make sure nobody is waiting for us to release the VM,
1463          * otherwise we can deadlock when we wait on each other
1464          */
1465         vfork_done = tsk->vfork_done;
1466         if (vfork_done) {
1467                 tsk->vfork_done = NULL;
1468                 complete(vfork_done);
1469         }
1470
1471         if (core_waiters)
1472                 wait_for_completion(&startup_done);
1473 fail:
1474         BUG_ON(mm->core_waiters);
1475         return core_waiters;
1476 }
1477
1478 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1479 {
1480         char corename[CORENAME_MAX_SIZE + 1];
1481         struct mm_struct *mm = current->mm;
1482         struct linux_binfmt * binfmt;
1483         struct inode * inode;
1484         struct file * file;
1485         int retval = 0;
1486         int fsuid = current->fsuid;
1487         int flag = 0;
1488         int ispipe = 0;
1489
1490         binfmt = current->binfmt;
1491         if (!binfmt || !binfmt->core_dump)
1492                 goto fail;
1493         down_write(&mm->mmap_sem);
1494         if (!mm->dumpable) {
1495                 up_write(&mm->mmap_sem);
1496                 goto fail;
1497         }
1498
1499         /*
1500          *      We cannot trust fsuid as being the "true" uid of the
1501          *      process nor do we know its entire history. We only know it
1502          *      was tainted so we dump it as root in mode 2.
1503          */
1504         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1505                 flag = O_EXCL;          /* Stop rewrite attacks */
1506                 current->fsuid = 0;     /* Dump root private */
1507         }
1508         mm->dumpable = 0;
1509
1510         retval = coredump_wait(exit_code);
1511         if (retval < 0)
1512                 goto fail;
1513
1514         /*
1515          * Clear any false indication of pending signals that might
1516          * be seen by the filesystem code called to write the core file.
1517          */
1518         clear_thread_flag(TIF_SIGPENDING);
1519
1520         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1521                 goto fail_unlock;
1522
1523         /*
1524          * lock_kernel() because format_corename() is controlled by sysctl, which
1525          * uses lock_kernel()
1526          */
1527         lock_kernel();
1528         ispipe = format_corename(corename, core_pattern, signr);
1529         unlock_kernel();
1530         if (ispipe) {
1531                 /* SIGPIPE can happen, but it's just never processed */
1532                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1533                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1534                                corename);
1535                         goto fail_unlock;
1536                 }
1537         } else
1538                 file = filp_open(corename,
1539                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1540                                  0600);
1541         if (IS_ERR(file))
1542                 goto fail_unlock;
1543         inode = file->f_path.dentry->d_inode;
1544         if (inode->i_nlink > 1)
1545                 goto close_fail;        /* multiple links - don't dump */
1546         if (!ispipe && d_unhashed(file->f_path.dentry))
1547                 goto close_fail;
1548
1549         /* AK: actually i see no reason to not allow this for named pipes etc.,
1550            but keep the previous behaviour for now. */
1551         if (!ispipe && !S_ISREG(inode->i_mode))
1552                 goto close_fail;
1553         if (!file->f_op)
1554                 goto close_fail;
1555         if (!file->f_op->write)
1556                 goto close_fail;
1557         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1558                 goto close_fail;
1559
1560         retval = binfmt->core_dump(signr, regs, file);
1561
1562         if (retval)
1563                 current->signal->group_exit_code |= 0x80;
1564 close_fail:
1565         filp_close(file, NULL);
1566 fail_unlock:
1567         current->fsuid = fsuid;
1568         complete_all(&mm->core_done);
1569 fail:
1570         return retval;
1571 }