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/config.h>
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/mman.h>
  29 #include <linux/a.out.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/key.h>
  38 #include <linux/personality.h>
  39 #include <linux/binfmts.h>
  40 #include <linux/swap.h>
  41 #include <linux/utsname.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/acct.h>
  51 #include <linux/vs_memory.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[65] = "core";
  62 /* The maximal length of core_pattern is also specified in sysctl.c */
  63
  64 static struct linux_binfmt *formats;
  65 static DEFINE_RWLOCK(binfmt_lock);
  66
  67 int register_binfmt(struct linux_binfmt * fmt)
  68 {
  69         struct linux_binfmt ** tmp = &formats;
  70
  71         if (!fmt)
  72                 return -EINVAL;
  73         if (fmt->next)
  74                 return -EBUSY;
  75         write_lock(&binfmt_lock);
  76         while (*tmp) {
  77                 if (fmt == *tmp) {
  78                         write_unlock(&binfmt_lock);
  79                         return -EBUSY;
  80                 }
  81                 tmp = &(*tmp)->next;
  82         }
  83         fmt->next = formats;
  84         formats = fmt;
  85         write_unlock(&binfmt_lock);
  86         return 0;
  87 }
  88
  89 EXPORT_SYMBOL(register_binfmt);
  90
  91 int unregister_binfmt(struct linux_binfmt * fmt)
  92 {
  93         struct linux_binfmt ** tmp = &formats;
  94
  95         write_lock(&binfmt_lock);
  96         while (*tmp) {
  97                 if (fmt == *tmp) {
  98                         *tmp = fmt->next;
  99                         write_unlock(&binfmt_lock);
 100                         return 0;
 101                 }
 102                 tmp = &(*tmp)->next;
 103         }
 104         write_unlock(&binfmt_lock);
 105         return -EINVAL;
 106 }
 107
 108 EXPORT_SYMBOL(unregister_binfmt);
 109
 110 static inline void put_binfmt(struct linux_binfmt * fmt)
 111 {
 112         module_put(fmt->module);
 113 }
 114
 115 /*
 116  * Note that a shared library must be both readable and executable due to
 117  * security reasons.
 118  *
 119  * Also note that we take the address to load from from the file itself.
 120  */
 121 asmlinkage long sys_uselib(const char __user * library)
 122 {
 123         struct file * file;
 124         struct nameidata nd;
 125         int error;
 126
 127         nd.intent.open.flags = FMODE_READ;
 128         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 129         if (error)
 130                 goto out;
 131
 132         error = -EINVAL;
 133         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 134                 goto exit;
 135
 136         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 137         if (error)
 138                 goto exit;
 139
 140         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 141         error = PTR_ERR(file);
 142         if (IS_ERR(file))
 143                 goto out;
 144
 145         error = -ENOEXEC;
 146         if(file->f_op) {
 147                 struct linux_binfmt * fmt;
 148
 149                 read_lock(&binfmt_lock);
 150                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 151                         if (!fmt->load_shlib)
 152                                 continue;
 153                         if (!try_module_get(fmt->module))
 154                                 continue;
 155                         read_unlock(&binfmt_lock);
 156                         error = fmt->load_shlib(file);
 157                         read_lock(&binfmt_lock);
 158                         put_binfmt(fmt);
 159                         if (error != -ENOEXEC)
 160                                 break;
 161                 }
 162                 read_unlock(&binfmt_lock);
 163         }
 164         fput(file);
 165 out:
 166         return error;
 167 exit:
 168         path_release(&nd);
 169         goto out;
 170 }
 171
 172 /*
 173  * count() counts the number of strings in array ARGV.
 174  */
 175 static int count(char __user * __user * argv, int max)
 176 {
 177         int i = 0;
 178
 179         if (argv != NULL) {
 180                 for (;;) {
 181                         char __user * p;
 182
 183                         if (get_user(p, argv))
 184                                 return -EFAULT;
 185                         if (!p)
 186                                 break;
 187                         argv++;
 188                         if(++i > max)
 189                                 return -E2BIG;
 190                         cond_resched();
 191                 }
 192         }
 193         return i;
 194 }
 195
 196 /*
 197  * 'copy_strings()' copies argument/environment strings from user
 198  * memory to free pages in kernel mem. These are in a format ready
 199  * to be put directly into the top of new user memory.
 200  */
 201 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 202 {
 203         struct page *kmapped_page = NULL;
 204         char *kaddr = NULL;
 205         int ret;
 206
 207         while (argc-- > 0) {
 208                 char __user *str;
 209                 int len;
 210                 unsigned long pos;
 211
 212                 if (get_user(str, argv+argc) ||
 213                                 !(len = strnlen_user(str, bprm->p))) {
 214                         ret = -EFAULT;
 215                         goto out;
 216                 }
 217
 218                 if (bprm->p < len)  {
 219                         ret = -E2BIG;
 220                         goto out;
 221                 }
 222
 223                 bprm->p -= len;
 224                 /* XXX: add architecture specific overflow check here. */
 225                 pos = bprm->p;
 226
 227                 while (len > 0) {
 228                         int i, new, err;
 229                         int offset, bytes_to_copy;
 230                         struct page *page;
 231
 232                         offset = pos % PAGE_SIZE;
 233                         i = pos/PAGE_SIZE;
 234                         page = bprm->page[i];
 235                         new = 0;
 236                         if (!page) {
 237                                 page = alloc_page(GFP_HIGHUSER);
 238                                 bprm->page[i] = page;
 239                                 if (!page) {
 240                                         ret = -ENOMEM;
 241                                         goto out;
 242                                 }
 243                                 new = 1;
 244                         }
 245
 246                         if (page != kmapped_page) {
 247                                 if (kmapped_page)
 248                                         kunmap(kmapped_page);
 249                                 kmapped_page = page;
 250                                 kaddr = kmap(kmapped_page);
 251                         }
 252                         if (new && offset)
 253                                 memset(kaddr, 0, offset);
 254                         bytes_to_copy = PAGE_SIZE - offset;
 255                         if (bytes_to_copy > len) {
 256                                 bytes_to_copy = len;
 257                                 if (new)
 258                                         memset(kaddr+offset+len, 0,
 259                                                 PAGE_SIZE-offset-len);
 260                         }
 261                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 262                         if (err) {
 263                                 ret = -EFAULT;
 264                                 goto out;
 265                         }
 266
 267                         pos += bytes_to_copy;
 268                         str += bytes_to_copy;
 269                         len -= bytes_to_copy;
 270                 }
 271         }
 272         ret = 0;
 273 out:
 274         if (kmapped_page)
 275                 kunmap(kmapped_page);
 276         return ret;
 277 }
 278
 279 /*
 280  * Like copy_strings, but get argv and its values from kernel memory.
 281  */
 282 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 283 {
 284         int r;
 285         mm_segment_t oldfs = get_fs();
 286         set_fs(KERNEL_DS);
 287         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 288         set_fs(oldfs);
 289         return r;
 290 }
 291
 292 EXPORT_SYMBOL(copy_strings_kernel);
 293
 294 #ifdef CONFIG_MMU
 295 /*
 296  * This routine is used to map in a page into an address space: needed by
 297  * execve() for the initial stack and environment pages.
 298  *
 299  * vma->vm_mm->mmap_sem is held for writing.
 300  */
 301 void install_arg_page(struct vm_area_struct *vma,
 302                         struct page *page, unsigned long address)
 303 {
 304         struct mm_struct *mm = vma->vm_mm;
 305         pgd_t * pgd;
 306         pud_t * pud;
 307         pmd_t * pmd;
 308         pte_t * pte;
 309
 310         if (unlikely(anon_vma_prepare(vma)))
 311                 goto out_sig;
 312
 313         flush_dcache_page(page);
 314         pgd = pgd_offset(mm, address);
 315
 316         spin_lock(&mm->page_table_lock);
 317         pud = pud_alloc(mm, pgd, address);
 318         if (!pud)
 319                 goto out;
 320         pmd = pmd_alloc(mm, pud, address);
 321         if (!pmd)
 322                 goto out;
 323         pte = pte_alloc_map(mm, pmd, address);
 324         if (!pte)
 325                 goto out;
 326         if (!pte_none(*pte)) {
 327                 pte_unmap(pte);
 328                 goto out;
 329         }
 330         vx_rsspages_inc(mm);
 331         lru_cache_add_active(page);
 332         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
 333                                         page, vma->vm_page_prot))));
 334         page_add_anon_rmap(page, vma, address);
 335         pte_unmap(pte);
 336         spin_unlock(&mm->page_table_lock);
 337
 338         /* no need for flush_tlb */
 339         return;
 340 out:
 341         spin_unlock(&mm->page_table_lock);
 342 out_sig:
 343         __free_page(page);
 344         force_sig(SIGKILL, current);
 345 }
 346
 347 #define EXTRA_STACK_VM_PAGES    20      /* random */
 348
 349 int setup_arg_pages(struct linux_binprm *bprm,
 350                     unsigned long stack_top,
 351                     int executable_stack)
 352 {
 353         unsigned long stack_base;
 354         struct vm_area_struct *mpnt;
 355         struct mm_struct *mm = current->mm;
 356         int i, ret;
 357         long arg_size;
 358
 359 #ifdef CONFIG_STACK_GROWSUP
 360         /* Move the argument and environment strings to the bottom of the
 361          * stack space.
 362          */
 363         int offset, j;
 364         char *to, *from;
 365
 366         /* Start by shifting all the pages down */
 367         i = 0;
 368         for (j = 0; j < MAX_ARG_PAGES; j++) {
 369                 struct page *page = bprm->page[j];
 370                 if (!page)
 371                         continue;
 372                 bprm->page[i++] = page;
 373         }
 374
 375         /* Now move them within their pages */
 376         offset = bprm->p % PAGE_SIZE;
 377         to = kmap(bprm->page[0]);
 378         for (j = 1; j < i; j++) {
 379                 memmove(to, to + offset, PAGE_SIZE - offset);
 380                 from = kmap(bprm->page[j]);
 381                 memcpy(to + PAGE_SIZE - offset, from, offset);
 382                 kunmap(bprm->page[j - 1]);
 383                 to = from;
 384         }
 385         memmove(to, to + offset, PAGE_SIZE - offset);
 386         kunmap(bprm->page[j - 1]);
 387
 388         /* Limit stack size to 1GB */
 389         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 390         if (stack_base > (1 << 30))
 391                 stack_base = 1 << 30;
 392         stack_base = PAGE_ALIGN(stack_top - stack_base);
 393
 394         /* Adjust bprm->p to point to the end of the strings. */
 395         bprm->p = stack_base + PAGE_SIZE * i - offset;
 396
 397         mm->arg_start = stack_base;
 398         arg_size = i << PAGE_SHIFT;
 399
 400         /* zero pages that were copied above */
 401         while (i < MAX_ARG_PAGES)
 402                 bprm->page[i++] = NULL;
 403 #else
 404         stack_base = stack_top - MAX_ARG_PAGES * PAGE_SIZE;
 405         bprm->p += stack_base;
 406         mm->arg_start = bprm->p;
 407         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 408 #endif
 409
 410         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 411
 412         if (bprm->loader)
 413                 bprm->loader += stack_base;
 414         bprm->exec += stack_base;
 415
 416         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 417         if (!mpnt)
 418                 return -ENOMEM;
 419
 420         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT) ||
 421                 !vx_vmpages_avail(mm, arg_size >> PAGE_SHIFT)) {
 422                 kmem_cache_free(vm_area_cachep, mpnt);
 423                 return -ENOMEM;
 424         }
 425
 426         memset(mpnt, 0, sizeof(*mpnt));
 427
 428         down_write(&mm->mmap_sem);
 429         {
 430                 mpnt->vm_mm = mm;
 431 #ifdef CONFIG_STACK_GROWSUP
 432                 mpnt->vm_start = stack_base;
 433                 mpnt->vm_end = stack_base + arg_size;
 434 #else
 435                 mpnt->vm_end = stack_top;
 436                 mpnt->vm_start = mpnt->vm_end - arg_size;
 437 #endif
 438                 /* Adjust stack execute permissions; explicitly enable
 439                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 440                  * and leave alone (arch default) otherwise. */
 441                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 442                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 443                 else if (executable_stack == EXSTACK_DISABLE_X)
 444                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 445                 else
 446                         mpnt->vm_flags = VM_STACK_FLAGS;
 447                 mpnt->vm_flags |= mm->def_flags;
 448                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 449                 if ((ret = insert_vm_struct(mm, mpnt))) {
 450                         up_write(&mm->mmap_sem);
 451                         kmem_cache_free(vm_area_cachep, mpnt);
 452                         return ret;
 453                 }
 454                 vx_vmpages_sub(mm, mm->total_vm - vma_pages(mpnt));
 455                 mm->stack_vm = mm->total_vm;
 456         }
 457
 458         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 459                 struct page *page = bprm->page[i];
 460                 if (page) {
 461                         bprm->page[i] = NULL;
 462                         install_arg_page(mpnt, page, stack_base);
 463                 }
 464                 stack_base += PAGE_SIZE;
 465         }
 466         up_write(&mm->mmap_sem);
 467
 468         return 0;
 469 }
 470
 471 EXPORT_SYMBOL(setup_arg_pages);
 472
 473 #define free_arg_pages(bprm) do { } while (0)
 474
 475 #else
 476
 477 static inline void free_arg_pages(struct linux_binprm *bprm)
 478 {
 479         int i;
 480
 481         for (i = 0; i < MAX_ARG_PAGES; i++) {
 482                 if (bprm->page[i])
 483                         __free_page(bprm->page[i]);
 484                 bprm->page[i] = NULL;
 485         }
 486 }
 487
 488 #endif /* CONFIG_MMU */
 489
 490 struct file *open_exec(const char *name)
 491 {
 492         struct nameidata nd;
 493         int err;
 494         struct file *file;
 495
 496         nd.intent.open.flags = FMODE_READ;
 497         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 498         file = ERR_PTR(err);
 499
 500         if (!err) {
 501                 struct inode *inode = nd.dentry->d_inode;
 502                 file = ERR_PTR(-EACCES);
 503                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 504                     S_ISREG(inode->i_mode)) {
 505                         int err = permission(inode, MAY_EXEC, &nd);
 506                         if (!err && !(inode->i_mode & 0111))
 507                                 err = -EACCES;
 508                         file = ERR_PTR(err);
 509                         if (!err) {
 510                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 511                                 if (!IS_ERR(file)) {
 512                                         err = deny_write_access(file);
 513                                         if (err) {
 514                                                 fput(file);
 515                                                 file = ERR_PTR(err);
 516                                         }
 517                                 }
 518 out:
 519                                 return file;
 520                         }
 521                 }
 522                 path_release(&nd);
 523         }
 524         goto out;
 525 }
 526
 527 EXPORT_SYMBOL(open_exec);
 528
 529 int kernel_read(struct file *file, unsigned long offset,
 530         char *addr, unsigned long count)
 531 {
 532         mm_segment_t old_fs;
 533         loff_t pos = offset;
 534         int result;
 535
 536         old_fs = get_fs();
 537         set_fs(get_ds());
 538         /* The cast to a user pointer is valid due to the set_fs() */
 539         result = vfs_read(file, (void __user *)addr, count, &pos);
 540         set_fs(old_fs);
 541         return result;
 542 }
 543
 544 EXPORT_SYMBOL(kernel_read);
 545
 546 static int exec_mmap(struct mm_struct *mm)
 547 {
 548         struct task_struct *tsk;
 549         struct mm_struct * old_mm, *active_mm;
 550
 551         /* Notify parent that we're no longer interested in the old VM */
 552         tsk = current;
 553         old_mm = current->mm;
 554         mm_release(tsk, old_mm);
 555
 556         if (old_mm) {
 557                 /*
 558                  * Make sure that if there is a core dump in progress
 559                  * for the old mm, we get out and die instead of going
 560                  * through with the exec.  We must hold mmap_sem around
 561                  * checking core_waiters and changing tsk->mm.  The
 562                  * core-inducing thread will increment core_waiters for
 563                  * each thread whose ->mm == old_mm.
 564                  */
 565                 down_read(&old_mm->mmap_sem);
 566                 if (unlikely(old_mm->core_waiters)) {
 567                         up_read(&old_mm->mmap_sem);
 568                         return -EINTR;
 569                 }
 570         }
 571         task_lock(tsk);
 572         active_mm = tsk->active_mm;
 573         tsk->mm = mm;
 574         tsk->active_mm = mm;
 575         activate_mm(active_mm, mm);
 576         task_unlock(tsk);
 577         arch_pick_mmap_layout(mm);
 578         if (old_mm) {
 579                 up_read(&old_mm->mmap_sem);
 580                 if (active_mm != old_mm) BUG();
 581                 mmput(old_mm);
 582                 return 0;
 583         }
 584         mmdrop(active_mm);
 585         return 0;
 586 }
 587
 588 /*
 589  * This function makes sure the current process has its own signal table,
 590  * so that flush_signal_handlers can later reset the handlers without
 591  * disturbing other processes.  (Other processes might share the signal
 592  * table via the CLONE_SIGHAND option to clone().)
 593  */
 594 static inline int de_thread(struct task_struct *tsk)
 595 {
 596         struct signal_struct *sig = tsk->signal;
 597         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 598         spinlock_t *lock = &oldsighand->siglock;
 599         int count;
 600
 601         /*
 602          * If we don't share sighandlers, then we aren't sharing anything
 603          * and we can just re-use it all.
 604          */
 605         if (atomic_read(&oldsighand->count) <= 1) {
 606                 BUG_ON(atomic_read(&sig->count) != 1);
 607                 exit_itimers(sig);
 608                 return 0;
 609         }
 610
 611         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 612         if (!newsighand)
 613                 return -ENOMEM;
 614
 615         if (thread_group_empty(current))
 616                 goto no_thread_group;
 617
 618         /*
 619          * Kill all other threads in the thread group.
 620          * We must hold tasklist_lock to call zap_other_threads.
 621          */
 622         read_lock(&tasklist_lock);
 623         spin_lock_irq(lock);
 624         if (sig->flags & SIGNAL_GROUP_EXIT) {
 625                 /*
 626                  * Another group action in progress, just
 627                  * return so that the signal is processed.
 628                  */
 629                 spin_unlock_irq(lock);
 630                 read_unlock(&tasklist_lock);
 631                 kmem_cache_free(sighand_cachep, newsighand);
 632                 return -EAGAIN;
 633         }
 634         zap_other_threads(current);
 635         read_unlock(&tasklist_lock);
 636
 637         /*
 638          * Account for the thread group leader hanging around:
 639          */
 640         count = 2;
 641         if (thread_group_leader(current))
 642                 count = 1;
 643         while (atomic_read(&sig->count) > count) {
 644                 sig->group_exit_task = current;
 645                 sig->notify_count = count;
 646                 __set_current_state(TASK_UNINTERRUPTIBLE);
 647                 spin_unlock_irq(lock);
 648                 schedule();
 649                 spin_lock_irq(lock);
 650         }
 651         sig->group_exit_task = NULL;
 652         sig->notify_count = 0;
 653         spin_unlock_irq(lock);
 654
 655         /*
 656          * At this point all other threads have exited, all we have to
 657          * do is to wait for the thread group leader to become inactive,
 658          * and to assume its PID:
 659          */
 660         if (!thread_group_leader(current)) {
 661                 struct task_struct *leader = current->group_leader, *parent;
 662                 struct dentry *proc_dentry1, *proc_dentry2;
 663                 unsigned long exit_state, ptrace;
 664
 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                 while (leader->exit_state != EXIT_ZOMBIE)
 671                         yield();
 672
 673                 spin_lock(&leader->proc_lock);
 674                 spin_lock(&current->proc_lock);
 675                 proc_dentry1 = proc_pid_unhash(current);
 676                 proc_dentry2 = proc_pid_unhash(leader);
 677                 write_lock_irq(&tasklist_lock);
 678
 679                 if (leader->tgid != current->tgid)
 680                         BUG();
 681                 if (current->pid == current->tgid)
 682                         BUG();
 683                 /*
 684                  * An exec() starts a new thread group with the
 685                  * TGID of the previous thread group. Rehash the
 686                  * two threads with a switched PID, and release
 687                  * the former thread group leader:
 688                  */
 689                 ptrace = leader->ptrace;
 690                 parent = leader->parent;
 691                 if (unlikely(ptrace) && unlikely(parent == current)) {
 692                         /*
 693                          * Joker was ptracing his own group leader,
 694                          * and now he wants to be his own parent!
 695                          * We can't have that.
 696                          */
 697                         ptrace = 0;
 698                 }
 699
 700                 ptrace_unlink(current);
 701                 ptrace_unlink(leader);
 702                 remove_parent(current);
 703                 remove_parent(leader);
 704
 705                 switch_exec_pids(leader, current);
 706
 707                 current->parent = current->real_parent = leader->real_parent;
 708                 leader->parent = leader->real_parent = child_reaper;
 709                 current->group_leader = current;
 710                 leader->group_leader = leader;
 711
 712                 add_parent(current, current->parent);
 713                 add_parent(leader, leader->parent);
 714                 if (ptrace) {
 715                         current->ptrace = ptrace;
 716                         __ptrace_link(current, parent);
 717                 }
 718
 719                 list_del(&current->tasks);
 720                 list_add_tail(&current->tasks, &init_task.tasks);
 721                 current->exit_signal = SIGCHLD;
 722                 exit_state = leader->exit_state;
 723
 724                 write_unlock_irq(&tasklist_lock);
 725                 spin_unlock(&leader->proc_lock);
 726                 spin_unlock(&current->proc_lock);
 727                 proc_pid_flush(proc_dentry1);
 728                 proc_pid_flush(proc_dentry2);
 729
 730                 if (exit_state != EXIT_ZOMBIE)
 731                         BUG();
 732                 release_task(leader);
 733         }
 734
 735         /*
 736          * Now there are really no other threads at all,
 737          * so it's safe to stop telling them to kill themselves.
 738          */
 739         sig->flags = 0;
 740
 741 no_thread_group:
 742         BUG_ON(atomic_read(&sig->count) != 1);
 743         exit_itimers(sig);
 744
 745         if (atomic_read(&oldsighand->count) == 1) {
 746                 /*
 747                  * Now that we nuked the rest of the thread group,
 748                  * it turns out we are not sharing sighand any more either.
 749                  * So we can just keep it.
 750                  */
 751                 kmem_cache_free(sighand_cachep, newsighand);
 752         } else {
 753                 /*
 754                  * Move our state over to newsighand and switch it in.
 755                  */
 756                 spin_lock_init(&newsighand->siglock);
 757                 atomic_set(&newsighand->count, 1);
 758                 memcpy(newsighand->action, oldsighand->action,
 759                        sizeof(newsighand->action));
 760
 761                 write_lock_irq(&tasklist_lock);
 762                 spin_lock(&oldsighand->siglock);
 763                 spin_lock(&newsighand->siglock);
 764
 765                 current->sighand = newsighand;
 766                 recalc_sigpending();
 767
 768                 spin_unlock(&newsighand->siglock);
 769                 spin_unlock(&oldsighand->siglock);
 770                 write_unlock_irq(&tasklist_lock);
 771
 772                 if (atomic_dec_and_test(&oldsighand->count))
 773                         kmem_cache_free(sighand_cachep, oldsighand);
 774         }
 775
 776         if (!thread_group_empty(current))
 777                 BUG();
 778         if (!thread_group_leader(current))
 779                 BUG();
 780         return 0;
 781 }
 782
 783 /*
 784  * These functions flushes out all traces of the currently running executable
 785  * so that a new one can be started
 786  */
 787
 788 static inline void flush_old_files(struct files_struct * files)
 789 {
 790         long j = -1;
 791
 792         spin_lock(&files->file_lock);
 793         for (;;) {
 794                 unsigned long set, i;
 795
 796                 j++;
 797                 i = j * __NFDBITS;
 798                 if (i >= files->max_fds || i >= files->max_fdset)
 799                         break;
 800                 set = files->close_on_exec->fds_bits[j];
 801                 if (!set)
 802                         continue;
 803                 files->close_on_exec->fds_bits[j] = 0;
 804                 spin_unlock(&files->file_lock);
 805                 for ( ; set ; i++,set >>= 1) {
 806                         if (set & 1) {
 807                                 sys_close(i);
 808                         }
 809                 }
 810                 spin_lock(&files->file_lock);
 811
 812         }
 813         spin_unlock(&files->file_lock);
 814 }
 815
 816 void get_task_comm(char *buf, struct task_struct *tsk)
 817 {
 818         /* buf must be at least sizeof(tsk->comm) in size */
 819         task_lock(tsk);
 820         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 821         task_unlock(tsk);
 822 }
 823
 824 void set_task_comm(struct task_struct *tsk, char *buf)
 825 {
 826         task_lock(tsk);
 827         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 828         task_unlock(tsk);
 829 }
 830
 831 int flush_old_exec(struct linux_binprm * bprm)
 832 {
 833         char * name;
 834         int i, ch, retval;
 835         struct files_struct *files;
 836         char tcomm[sizeof(current->comm)];
 837
 838         /*
 839          * Make sure we have a private signal table and that
 840          * we are unassociated from the previous thread group.
 841          */
 842         retval = de_thread(current);
 843         if (retval)
 844                 goto out;
 845
 846         /*
 847          * Make sure we have private file handles. Ask the
 848          * fork helper to do the work for us and the exit
 849          * helper to do the cleanup of the old one.
 850          */
 851         files = current->files;         /* refcounted so safe to hold */
 852         retval = unshare_files();
 853         if (retval)
 854                 goto out;
 855         /*
 856          * Release all of the old mmap stuff
 857          */
 858         retval = exec_mmap(bprm->mm);
 859         if (retval)
 860                 goto mmap_failed;
 861
 862         bprm->mm = NULL;                /* We're using it now */
 863
 864         /* This is the point of no return */
 865         steal_locks(files);
 866         put_files_struct(files);
 867
 868         current->sas_ss_sp = current->sas_ss_size = 0;
 869
 870         if (current->euid == current->uid && current->egid == current->gid)
 871                 current->mm->dumpable = 1;
 872         name = bprm->filename;
 873         for (i=0; (ch = *(name++)) != '\0';) {
 874                 if (ch == '/')
 875                         i = 0;
 876                 else
 877                         if (i < (sizeof(tcomm) - 1))
 878                                 tcomm[i++] = ch;
 879         }
 880         tcomm[i] = '\0';
 881         set_task_comm(current, tcomm);
 882
 883         flush_thread();
 884
 885         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 886             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
 887             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 888                 suid_keys(current);
 889                 current->mm->dumpable = 0;
 890         }
 891
 892         /* An exec changes our domain. We are no longer part of the thread
 893            group */
 894
 895         current->self_exec_id++;
 896
 897         flush_signal_handlers(current, 0);
 898         flush_old_files(current->files);
 899
 900         return 0;
 901
 902 mmap_failed:
 903         put_files_struct(current->files);
 904         current->files = files;
 905 out:
 906         return retval;
 907 }
 908
 909 EXPORT_SYMBOL(flush_old_exec);
 910
 911 /*
 912  * Fill the binprm structure from the inode.
 913  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 914  */
 915 int prepare_binprm(struct linux_binprm *bprm)
 916 {
 917         int mode;
 918         struct inode * inode = bprm->file->f_dentry->d_inode;
 919         int retval;
 920
 921         mode = inode->i_mode;
 922         /*
 923          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 924          * generic_permission lets a non-executable through
 925          */
 926         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 927                 return -EACCES;
 928         if (bprm->file->f_op == NULL)
 929                 return -EACCES;
 930
 931         bprm->e_uid = current->euid;
 932         bprm->e_gid = current->egid;
 933
 934         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 935                 /* Set-uid? */
 936                 if (mode & S_ISUID) {
 937                         current->personality &= ~PER_CLEAR_ON_SETID;
 938                         bprm->e_uid = inode->i_uid;
 939                 }
 940
 941                 /* Set-gid? */
 942                 /*
 943                  * If setgid is set but no group execute bit then this
 944                  * is a candidate for mandatory locking, not a setgid
 945                  * executable.
 946                  */
 947                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 948                         current->personality &= ~PER_CLEAR_ON_SETID;
 949                         bprm->e_gid = inode->i_gid;
 950                 }
 951         }
 952
 953         /* fill in binprm security blob */
 954         retval = security_bprm_set(bprm);
 955         if (retval)
 956                 return retval;
 957
 958         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 959         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 960 }
 961
 962 EXPORT_SYMBOL(prepare_binprm);
 963
 964 static inline int unsafe_exec(struct task_struct *p)
 965 {
 966         int unsafe = 0;
 967         if (p->ptrace & PT_PTRACED) {
 968                 if (p->ptrace & PT_PTRACE_CAP)
 969                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 970                 else
 971                         unsafe |= LSM_UNSAFE_PTRACE;
 972         }
 973         if (atomic_read(&p->fs->count) > 1 ||
 974             atomic_read(&p->files->count) > 1 ||
 975             atomic_read(&p->sighand->count) > 1)
 976                 unsafe |= LSM_UNSAFE_SHARE;
 977
 978         return unsafe;
 979 }
 980
 981 void compute_creds(struct linux_binprm *bprm)
 982 {
 983         int unsafe;
 984
 985         if (bprm->e_uid != current->uid)
 986                 suid_keys(current);
 987         exec_keys(current);
 988
 989         task_lock(current);
 990         unsafe = unsafe_exec(current);
 991         security_bprm_apply_creds(bprm, unsafe);
 992         task_unlock(current);
 993         security_bprm_post_apply_creds(bprm);
 994 }
 995
 996 EXPORT_SYMBOL(compute_creds);
 997
 998 void remove_arg_zero(struct linux_binprm *bprm)
 999 {
1000         if (bprm->argc) {
1001                 unsigned long offset;
1002                 char * kaddr;
1003                 struct page *page;
1004
1005                 offset = bprm->p % PAGE_SIZE;
1006                 goto inside;
1007
1008                 while (bprm->p++, *(kaddr+offset++)) {
1009                         if (offset != PAGE_SIZE)
1010                                 continue;
1011                         offset = 0;
1012                         kunmap_atomic(kaddr, KM_USER0);
1013 inside:
1014                         page = bprm->page[bprm->p/PAGE_SIZE];
1015                         kaddr = kmap_atomic(page, KM_USER0);
1016                 }
1017                 kunmap_atomic(kaddr, KM_USER0);
1018                 bprm->argc--;
1019         }
1020 }
1021
1022 EXPORT_SYMBOL(remove_arg_zero);
1023
1024 /*
1025  * cycle the list of binary formats handler, until one recognizes the image
1026  */
1027 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1028 {
1029         int try,retval;
1030         struct linux_binfmt *fmt;
1031 #ifdef __alpha__
1032         /* handle /sbin/loader.. */
1033         {
1034             struct exec * eh = (struct exec *) bprm->buf;
1035
1036             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1037                 (eh->fh.f_flags & 0x3000) == 0x3000)
1038             {
1039                 struct file * file;
1040                 unsigned long loader;
1041
1042                 allow_write_access(bprm->file);
1043                 fput(bprm->file);
1044                 bprm->file = NULL;
1045
1046                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1047
1048                 file = open_exec("/sbin/loader");
1049                 retval = PTR_ERR(file);
1050                 if (IS_ERR(file))
1051                         return retval;
1052
1053                 /* Remember if the application is TASO.  */
1054                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1055
1056                 bprm->file = file;
1057                 bprm->loader = loader;
1058                 retval = prepare_binprm(bprm);
1059                 if (retval<0)
1060                         return retval;
1061                 /* should call search_binary_handler recursively here,
1062                    but it does not matter */
1063             }
1064         }
1065 #endif
1066         retval = security_bprm_check(bprm);
1067         if (retval)
1068                 return retval;
1069
1070         /* kernel module loader fixup */
1071         /* so we don't try to load run modprobe in kernel space. */
1072         set_fs(USER_DS);
1073         retval = -ENOENT;
1074         for (try=0; try<2; try++) {
1075                 read_lock(&binfmt_lock);
1076                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1077                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1078                         if (!fn)
1079                                 continue;
1080                         if (!try_module_get(fmt->module))
1081                                 continue;
1082                         read_unlock(&binfmt_lock);
1083                         retval = fn(bprm, regs);
1084                         if (retval >= 0) {
1085                                 put_binfmt(fmt);
1086                                 allow_write_access(bprm->file);
1087                                 if (bprm->file)
1088                                         fput(bprm->file);
1089                                 bprm->file = NULL;
1090                                 current->did_exec = 1;
1091                                 return retval;
1092                         }
1093                         read_lock(&binfmt_lock);
1094                         put_binfmt(fmt);
1095                         if (retval != -ENOEXEC || bprm->mm == NULL)
1096                                 break;
1097                         if (!bprm->file) {
1098                                 read_unlock(&binfmt_lock);
1099                                 return retval;
1100                         }
1101                 }
1102                 read_unlock(&binfmt_lock);
1103                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1104                         break;
1105 #ifdef CONFIG_KMOD
1106                 }else{
1107 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1108                         if (printable(bprm->buf[0]) &&
1109                             printable(bprm->buf[1]) &&
1110                             printable(bprm->buf[2]) &&
1111                             printable(bprm->buf[3]))
1112                                 break; /* -ENOEXEC */
1113                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1114 #endif
1115                 }
1116         }
1117         return retval;
1118 }
1119
1120 EXPORT_SYMBOL(search_binary_handler);
1121
1122 /*
1123  * sys_execve() executes a new program.
1124  */
1125 int do_execve(char * filename,
1126         char __user *__user *argv,
1127         char __user *__user *envp,
1128         struct pt_regs * regs)
1129 {
1130         struct linux_binprm *bprm;
1131         struct file *file;
1132         int retval;
1133         int i;
1134
1135         retval = -ENOMEM;
1136         bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1137         if (!bprm)
1138                 goto out_ret;
1139         memset(bprm, 0, sizeof(*bprm));
1140
1141         file = open_exec(filename);
1142         retval = PTR_ERR(file);
1143         if (IS_ERR(file))
1144                 goto out_kfree;
1145
1146         sched_exec();
1147
1148         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1149
1150         bprm->file = file;
1151         bprm->filename = filename;
1152         bprm->interp = filename;
1153         bprm->mm = mm_alloc();
1154         retval = -ENOMEM;
1155         if (!bprm->mm)
1156                 goto out_file;
1157
1158         retval = init_new_context(current, bprm->mm);
1159         if (retval < 0)
1160                 goto out_mm;
1161
1162         bprm->argc = count(argv, bprm->p / sizeof(void *));
1163         if ((retval = bprm->argc) < 0)
1164                 goto out_mm;
1165
1166         bprm->envc = count(envp, bprm->p / sizeof(void *));
1167         if ((retval = bprm->envc) < 0)
1168                 goto out_mm;
1169
1170         retval = security_bprm_alloc(bprm);
1171         if (retval)
1172                 goto out;
1173
1174         retval = prepare_binprm(bprm);
1175         if (retval < 0)
1176                 goto out;
1177
1178         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1179         if (retval < 0)
1180                 goto out;
1181
1182         bprm->exec = bprm->p;
1183         retval = copy_strings(bprm->envc, envp, bprm);
1184         if (retval < 0)
1185                 goto out;
1186
1187         retval = copy_strings(bprm->argc, argv, bprm);
1188         if (retval < 0)
1189                 goto out;
1190
1191         retval = search_binary_handler(bprm,regs);
1192         if (retval >= 0) {
1193                 free_arg_pages(bprm);
1194
1195                 /* execve success */
1196                 security_bprm_free(bprm);
1197                 acct_update_integrals();
1198                 update_mem_hiwater();
1199                 kfree(bprm);
1200                 return retval;
1201         }
1202
1203 out:
1204         /* Something went wrong, return the inode and free the argument pages*/
1205         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1206                 struct page * page = bprm->page[i];
1207                 if (page)
1208                         __free_page(page);
1209         }
1210
1211         if (bprm->security)
1212                 security_bprm_free(bprm);
1213
1214 out_mm:
1215         if (bprm->mm)
1216                 mmdrop(bprm->mm);
1217
1218 out_file:
1219         if (bprm->file) {
1220                 allow_write_access(bprm->file);
1221                 fput(bprm->file);
1222         }
1223
1224 out_kfree:
1225         kfree(bprm);
1226
1227 out_ret:
1228         return retval;
1229 }
1230
1231 int set_binfmt(struct linux_binfmt *new)
1232 {
1233         struct linux_binfmt *old = current->binfmt;
1234
1235         if (new) {
1236                 if (!try_module_get(new->module))
1237                         return -1;
1238         }
1239         current->binfmt = new;
1240         if (old)
1241                 module_put(old->module);
1242         return 0;
1243 }
1244
1245 EXPORT_SYMBOL(set_binfmt);
1246
1247 #define CORENAME_MAX_SIZE 64
1248
1249 /* format_corename will inspect the pattern parameter, and output a
1250  * name into corename, which must have space for at least
1251  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1252  */
1253 static void format_corename(char *corename, const char *pattern, long signr)
1254 {
1255         const char *pat_ptr = pattern;
1256         char *out_ptr = corename;
1257         char *const out_end = corename + CORENAME_MAX_SIZE;
1258         int rc;
1259         int pid_in_pattern = 0;
1260
1261         /* Repeat as long as we have more pattern to process and more output
1262            space */
1263         while (*pat_ptr) {
1264                 if (*pat_ptr != '%') {
1265                         if (out_ptr == out_end)
1266                                 goto out;
1267                         *out_ptr++ = *pat_ptr++;
1268                 } else {
1269                         switch (*++pat_ptr) {
1270                         case 0:
1271                                 goto out;
1272                         /* Double percent, output one percent */
1273                         case '%':
1274                                 if (out_ptr == out_end)
1275                                         goto out;
1276                                 *out_ptr++ = '%';
1277                                 break;
1278                         /* pid */
1279                         case 'p':
1280                                 pid_in_pattern = 1;
1281                                 rc = snprintf(out_ptr, out_end - out_ptr,
1282                                               "%d", current->tgid);
1283                                 if (rc > out_end - out_ptr)
1284                                         goto out;
1285                                 out_ptr += rc;
1286                                 break;
1287                         /* uid */
1288                         case 'u':
1289                                 rc = snprintf(out_ptr, out_end - out_ptr,
1290                                               "%d", current->uid);
1291                                 if (rc > out_end - out_ptr)
1292                                         goto out;
1293                                 out_ptr += rc;
1294                                 break;
1295                         /* gid */
1296                         case 'g':
1297                                 rc = snprintf(out_ptr, out_end - out_ptr,
1298                                               "%d", current->gid);
1299                                 if (rc > out_end - out_ptr)
1300                                         goto out;
1301                                 out_ptr += rc;
1302                                 break;
1303                         /* signal that caused the coredump */
1304                         case 's':
1305                                 rc = snprintf(out_ptr, out_end - out_ptr,
1306                                               "%ld", signr);
1307                                 if (rc > out_end - out_ptr)
1308                                         goto out;
1309                                 out_ptr += rc;
1310                                 break;
1311                         /* UNIX time of coredump */
1312                         case 't': {
1313                                 struct timeval tv;
1314                                 do_gettimeofday(&tv);
1315                                 rc = snprintf(out_ptr, out_end - out_ptr,
1316                                               "%lu", tv.tv_sec);
1317                                 if (rc > out_end - out_ptr)
1318                                         goto out;
1319                                 out_ptr += rc;
1320                                 break;
1321                         }
1322                         /* hostname */
1323                         case 'h':
1324                                 down_read(&uts_sem);
1325                                 rc = snprintf(out_ptr, out_end - out_ptr,
1326                                               "%s", system_utsname.nodename);
1327                                 up_read(&uts_sem);
1328                                 if (rc > out_end - out_ptr)
1329                                         goto out;
1330                                 out_ptr += rc;
1331                                 break;
1332                         /* executable */
1333                         case 'e':
1334                                 rc = snprintf(out_ptr, out_end - out_ptr,
1335                                               "%s", current->comm);
1336                                 if (rc > out_end - out_ptr)
1337                                         goto out;
1338                                 out_ptr += rc;
1339                                 break;
1340                         default:
1341                                 break;
1342                         }
1343                         ++pat_ptr;
1344                 }
1345         }
1346         /* Backward compatibility with core_uses_pid:
1347          *
1348          * If core_pattern does not include a %p (as is the default)
1349          * and core_uses_pid is set, then .%pid will be appended to
1350          * the filename */
1351         if (!pid_in_pattern
1352             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1353                 rc = snprintf(out_ptr, out_end - out_ptr,
1354                               ".%d", current->tgid);
1355                 if (rc > out_end - out_ptr)
1356                         goto out;
1357                 out_ptr += rc;
1358         }
1359       out:
1360         *out_ptr = 0;
1361 }
1362
1363 static void zap_threads (struct mm_struct *mm)
1364 {
1365         struct task_struct *g, *p;
1366         struct task_struct *tsk = current;
1367         struct completion *vfork_done = tsk->vfork_done;
1368         int traced = 0;
1369
1370         /*
1371          * Make sure nobody is waiting for us to release the VM,
1372          * otherwise we can deadlock when we wait on each other
1373          */
1374         if (vfork_done) {
1375                 tsk->vfork_done = NULL;
1376                 complete(vfork_done);
1377         }
1378
1379         read_lock(&tasklist_lock);
1380         do_each_thread(g,p)
1381                 if (mm == p->mm && p != tsk) {
1382                         force_sig_specific(SIGKILL, p);
1383                         mm->core_waiters++;
1384                         if (unlikely(p->ptrace) &&
1385                             unlikely(p->parent->mm == mm))
1386                                 traced = 1;
1387                 }
1388         while_each_thread(g,p);
1389
1390         read_unlock(&tasklist_lock);
1391
1392         if (unlikely(traced)) {
1393                 /*
1394                  * We are zapping a thread and the thread it ptraces.
1395                  * If the tracee went into a ptrace stop for exit tracing,
1396                  * we could deadlock since the tracer is waiting for this
1397                  * coredump to finish.  Detach them so they can both die.
1398                  */
1399                 write_lock_irq(&tasklist_lock);
1400                 do_each_thread(g,p) {
1401                         if (mm == p->mm && p != tsk &&
1402                             p->ptrace && p->parent->mm == mm) {
1403                                 __ptrace_unlink(p);
1404                         }
1405                 } while_each_thread(g,p);
1406                 write_unlock_irq(&tasklist_lock);
1407         }
1408 }
1409
1410 static void coredump_wait(struct mm_struct *mm)
1411 {
1412         DECLARE_COMPLETION(startup_done);
1413
1414         mm->core_waiters++; /* let other threads block */
1415         mm->core_startup_done = &startup_done;
1416
1417         /* give other threads a chance to run: */
1418         yield();
1419
1420         zap_threads(mm);
1421         if (--mm->core_waiters) {
1422                 up_write(&mm->mmap_sem);
1423                 wait_for_completion(&startup_done);
1424         } else
1425                 up_write(&mm->mmap_sem);
1426         BUG_ON(mm->core_waiters);
1427 }
1428
1429 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1430 {
1431         char corename[CORENAME_MAX_SIZE + 1];
1432         struct mm_struct *mm = current->mm;
1433         struct linux_binfmt * binfmt;
1434         struct inode * inode;
1435         struct file * file;
1436         int retval = 0;
1437
1438         binfmt = current->binfmt;
1439         if (!binfmt || !binfmt->core_dump)
1440                 goto fail;
1441         down_write(&mm->mmap_sem);
1442         if (!mm->dumpable) {
1443                 up_write(&mm->mmap_sem);
1444                 goto fail;
1445         }
1446         mm->dumpable = 0;
1447         init_completion(&mm->core_done);
1448         spin_lock_irq(&current->sighand->siglock);
1449         current->signal->flags = SIGNAL_GROUP_EXIT;
1450         current->signal->group_exit_code = exit_code;
1451         spin_unlock_irq(&current->sighand->siglock);
1452         coredump_wait(mm);
1453
1454         /*
1455          * Clear any false indication of pending signals that might
1456          * be seen by the filesystem code called to write the core file.
1457          */
1458         current->signal->group_stop_count = 0;
1459         clear_thread_flag(TIF_SIGPENDING);
1460
1461         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1462                 goto fail_unlock;
1463
1464         /*
1465          * lock_kernel() because format_corename() is controlled by sysctl, which
1466          * uses lock_kernel()
1467          */
1468         lock_kernel();
1469         format_corename(corename, core_pattern, signr);
1470         unlock_kernel();
1471         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1472         if (IS_ERR(file))
1473                 goto fail_unlock;
1474         inode = file->f_dentry->d_inode;
1475         if (inode->i_nlink > 1)
1476                 goto close_fail;        /* multiple links - don't dump */
1477         if (d_unhashed(file->f_dentry))
1478                 goto close_fail;
1479
1480         if (!S_ISREG(inode->i_mode))
1481                 goto close_fail;
1482         if (!file->f_op)
1483                 goto close_fail;
1484         if (!file->f_op->write)
1485                 goto close_fail;
1486         if (do_truncate(file->f_dentry, 0) != 0)
1487                 goto close_fail;
1488
1489         retval = binfmt->core_dump(signr, regs, file);
1490
1491         if (retval)
1492                 current->signal->group_exit_code |= 0x80;
1493 close_fail:
1494         filp_close(file, NULL);
1495 fail_unlock:
1496         complete_all(&mm->core_done);
1497 fail:
1498         return retval;
1499 }