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