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/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/ckrm.h>
  50 #include <linux/vs_memory.h>
  51 #include <linux/ckrm_mem.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 rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  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                 }
 191         }
 192         return i;
 193 }
 194
 195 /*
 196  * 'copy_strings()' copies argument/environment strings from user
 197  * memory to free pages in kernel mem. These are in a format ready
 198  * to be put directly into the top of new user memory.
 199  */
 200 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 201 {
 202         struct page *kmapped_page = NULL;
 203         char *kaddr = NULL;
 204         int ret;
 205
 206         while (argc-- > 0) {
 207                 char __user *str;
 208                 int len;
 209                 unsigned long pos;
 210
 211                 if (get_user(str, argv+argc) ||
 212                                 !(len = strnlen_user(str, bprm->p))) {
 213                         ret = -EFAULT;
 214                         goto out;
 215                 }
 216
 217                 if (bprm->p < len)  {
 218                         ret = -E2BIG;
 219                         goto out;
 220                 }
 221
 222                 bprm->p -= len;
 223                 /* XXX: add architecture specific overflow check here. */
 224                 pos = bprm->p;
 225
 226                 while (len > 0) {
 227                         int i, new, err;
 228                         int offset, bytes_to_copy;
 229                         struct page *page;
 230
 231                         offset = pos % PAGE_SIZE;
 232                         i = pos/PAGE_SIZE;
 233                         page = bprm->page[i];
 234                         new = 0;
 235                         if (!page) {
 236                                 page = alloc_page(GFP_HIGHUSER);
 237                                 bprm->page[i] = page;
 238                                 if (!page) {
 239                                         ret = -ENOMEM;
 240                                         goto out;
 241                                 }
 242                                 new = 1;
 243                         }
 244
 245                         if (page != kmapped_page) {
 246                                 if (kmapped_page)
 247                                         kunmap(kmapped_page);
 248                                 kmapped_page = page;
 249                                 kaddr = kmap(kmapped_page);
 250                         }
 251                         if (new && offset)
 252                                 memset(kaddr, 0, offset);
 253                         bytes_to_copy = PAGE_SIZE - offset;
 254                         if (bytes_to_copy > len) {
 255                                 bytes_to_copy = len;
 256                                 if (new)
 257                                         memset(kaddr+offset+len, 0,
 258                                                 PAGE_SIZE-offset-len);
 259                         }
 260                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 261                         if (err) {
 262                                 ret = -EFAULT;
 263                                 goto out;
 264                         }
 265
 266                         pos += bytes_to_copy;
 267                         str += bytes_to_copy;
 268                         len -= bytes_to_copy;
 269                 }
 270         }
 271         ret = 0;
 272 out:
 273         if (kmapped_page)
 274                 kunmap(kmapped_page);
 275         return ret;
 276 }
 277
 278 /*
 279  * Like copy_strings, but get argv and its values from kernel memory.
 280  */
 281 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 282 {
 283         int r;
 284         mm_segment_t oldfs = get_fs();
 285         set_fs(KERNEL_DS);
 286         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 287         set_fs(oldfs);
 288         return r;
 289 }
 290
 291 EXPORT_SYMBOL(copy_strings_kernel);
 292
 293 #ifdef CONFIG_MMU
 294 /*
 295  * This routine is used to map in a page into an address space: needed by
 296  * execve() for the initial stack and environment pages.
 297  *
 298  * vma->vm_mm->mmap_sem is held for writing.
 299  */
 300 void install_arg_page(struct vm_area_struct *vma,
 301                         struct page *page, unsigned long address)
 302 {
 303         struct mm_struct *mm = vma->vm_mm;
 304         pgd_t * pgd;
 305         pmd_t * pmd;
 306         pte_t * pte;
 307
 308         if (unlikely(anon_vma_prepare(vma)))
 309                 goto out_sig;
 310
 311         flush_dcache_page(page);
 312         pgd = pgd_offset(mm, address);
 313
 314         spin_lock(&mm->page_table_lock);
 315         pmd = pmd_alloc(mm, pgd, address);
 316         if (!pmd)
 317                 goto out;
 318         pte = pte_alloc_map(mm, pmd, address);
 319         if (!pte)
 320                 goto out;
 321         if (!pte_none(*pte)) {
 322                 pte_unmap(pte);
 323                 goto out;
 324         }
 325         // mm->rss++;
 326         vx_rsspages_inc(mm);
 327         lru_cache_add_active(page);
 328         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
 329                                         page, vma->vm_page_prot))));
 330         page_add_anon_rmap(page, vma, address);
 331         pte_unmap(pte);
 332         spin_unlock(&mm->page_table_lock);
 333
 334         /* no need for flush_tlb */
 335         return;
 336 out:
 337         spin_unlock(&mm->page_table_lock);
 338 out_sig:
 339         __free_page(page);
 340         force_sig(SIGKILL, current);
 341 }
 342
 343 int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
 344 {
 345         unsigned long stack_base;
 346         struct vm_area_struct *mpnt;
 347         struct mm_struct *mm = current->mm;
 348         int i;
 349         long arg_size;
 350
 351 #ifdef CONFIG_STACK_GROWSUP
 352         /* Move the argument and environment strings to the bottom of the
 353          * stack space.
 354          */
 355         int offset, j;
 356         char *to, *from;
 357
 358         /* Start by shifting all the pages down */
 359         i = 0;
 360         for (j = 0; j < MAX_ARG_PAGES; j++) {
 361                 struct page *page = bprm->page[j];
 362                 if (!page)
 363                         continue;
 364                 bprm->page[i++] = page;
 365         }
 366
 367         /* Now move them within their pages */
 368         offset = bprm->p % PAGE_SIZE;
 369         to = kmap(bprm->page[0]);
 370         for (j = 1; j < i; j++) {
 371                 memmove(to, to + offset, PAGE_SIZE - offset);
 372                 from = kmap(bprm->page[j]);
 373                 memcpy(to + PAGE_SIZE - offset, from, offset);
 374                 kunmap(bprm->page[j - 1]);
 375                 to = from;
 376         }
 377         memmove(to, to + offset, PAGE_SIZE - offset);
 378         kunmap(bprm->page[j - 1]);
 379
 380         /* Adjust bprm->p to point to the end of the strings. */
 381         bprm->p = PAGE_SIZE * i - offset;
 382
 383         /* Limit stack size to 1GB */
 384         stack_base = current->rlim[RLIMIT_STACK].rlim_max;
 385         if (stack_base > (1 << 30))
 386                 stack_base = 1 << 30;
 387         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
 388
 389         mm->arg_start = stack_base;
 390         arg_size = i << PAGE_SHIFT;
 391
 392         /* zero pages that were copied above */
 393         while (i < MAX_ARG_PAGES)
 394                 bprm->page[i++] = NULL;
 395 #else
 396 #ifdef __HAVE_ARCH_ALIGN_STACK
 397         stack_base = arch_align_stack(STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE);
 398         stack_base = PAGE_ALIGN(stack_base);
 399 #else
 400         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 401 #endif
 402         mm->arg_start = bprm->p + stack_base;
 403         arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
 404 #endif
 405
 406         bprm->p += stack_base;
 407         if (bprm->loader)
 408                 bprm->loader += stack_base;
 409         bprm->exec += stack_base;
 410
 411         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 412         if (!mpnt)
 413                 return -ENOMEM;
 414
 415         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT) ||
 416                 !vx_vmpages_avail(mm, arg_size >> PAGE_SHIFT)) {
 417                 kmem_cache_free(vm_area_cachep, mpnt);
 418                 return -ENOMEM;
 419         }
 420
 421         memset(mpnt, 0, sizeof(*mpnt));
 422
 423         down_write(&mm->mmap_sem);
 424         {
 425                 mpnt->vm_mm = mm;
 426 #ifdef CONFIG_STACK_GROWSUP
 427                 mpnt->vm_start = stack_base;
 428                 mpnt->vm_end = PAGE_MASK &
 429                         (PAGE_SIZE - 1 + (unsigned long) bprm->p);
 430 #else
 431                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 432                 mpnt->vm_end = STACK_TOP;
 433 #endif
 434                 /* Adjust stack execute permissions; explicitly enable
 435                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 436                  * and leave alone (arch default) otherwise. */
 437                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 438                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 439                 else if (executable_stack == EXSTACK_DISABLE_X)
 440                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 441                 else
 442                         mpnt->vm_flags = VM_STACK_FLAGS;
 443                 mpnt->vm_flags |= mm->def_flags;
 444                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 445                 insert_vm_struct(mm, mpnt);
 446                 // mm->total_vm = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
 447                 vx_vmpages_sub(mm, mm->total_vm -
 448                         ((mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT));
 449         }
 450
 451         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 452                 struct page *page = bprm->page[i];
 453                 if (page) {
 454                         bprm->page[i] = NULL;
 455                         install_arg_page(mpnt, page, stack_base);
 456                 }
 457                 stack_base += PAGE_SIZE;
 458         }
 459         up_write(&mm->mmap_sem);
 460
 461         return 0;
 462 }
 463
 464 EXPORT_SYMBOL(setup_arg_pages);
 465
 466 #define free_arg_pages(bprm) do { } while (0)
 467
 468 #else
 469
 470 static inline void free_arg_pages(struct linux_binprm *bprm)
 471 {
 472         int i;
 473
 474         for (i = 0; i < MAX_ARG_PAGES; i++) {
 475                 if (bprm->page[i])
 476                         __free_page(bprm->page[i]);
 477                 bprm->page[i] = NULL;
 478         }
 479 }
 480
 481 #endif /* CONFIG_MMU */
 482
 483 struct file *open_exec(const char *name)
 484 {
 485         struct nameidata nd;
 486         int err;
 487         struct file *file;
 488
 489         nd.intent.open.flags = FMODE_READ;
 490         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 491         file = ERR_PTR(err);
 492
 493         if (!err) {
 494                 struct inode *inode = nd.dentry->d_inode;
 495                 file = ERR_PTR(-EACCES);
 496                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 497                     S_ISREG(inode->i_mode)) {
 498                         int err = permission(inode, MAY_EXEC, &nd);
 499                         if (!err && !(inode->i_mode & 0111))
 500                                 err = -EACCES;
 501                         file = ERR_PTR(err);
 502                         if (!err) {
 503                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 504                                 if (!IS_ERR(file)) {
 505                                         err = deny_write_access(file);
 506                                         if (err) {
 507                                                 fput(file);
 508                                                 file = ERR_PTR(err);
 509                                         }
 510                                 }
 511 out:
 512                                 return file;
 513                         }
 514                 }
 515                 path_release(&nd);
 516         }
 517         goto out;
 518 }
 519
 520 EXPORT_SYMBOL(open_exec);
 521
 522 int kernel_read(struct file *file, unsigned long offset,
 523         char *addr, unsigned long count)
 524 {
 525         mm_segment_t old_fs;
 526         loff_t pos = offset;
 527         int result;
 528
 529         old_fs = get_fs();
 530         set_fs(get_ds());
 531         /* The cast to a user pointer is valid due to the set_fs() */
 532         result = vfs_read(file, (void __user *)addr, count, &pos);
 533         set_fs(old_fs);
 534         return result;
 535 }
 536
 537 EXPORT_SYMBOL(kernel_read);
 538
 539 static int exec_mmap(struct mm_struct *mm)
 540 {
 541         struct task_struct *tsk;
 542         struct mm_struct * old_mm, *active_mm;
 543
 544         /* Add it to the list of mm's */
 545         spin_lock(&mmlist_lock);
 546         list_add(&mm->mmlist, &init_mm.mmlist);
 547         mmlist_nr++;
 548         spin_unlock(&mmlist_lock);
 549
 550         /* Notify parent that we're no longer interested in the old VM */
 551         tsk = current;
 552         old_mm = current->mm;
 553         mm_release(tsk, old_mm);
 554
 555         task_lock(tsk);
 556         active_mm = tsk->active_mm;
 557         tsk->mm = mm;
 558         tsk->active_mm = mm;
 559         activate_mm(active_mm, mm);
 560         task_unlock(tsk);
 561         arch_pick_mmap_layout(mm);
 562 #ifdef CONFIG_CKRM_RES_MEM
 563         if (old_mm) {
 564                 spin_lock(&old_mm->peertask_lock);
 565                 list_del(&tsk->mm_peers);
 566                 ckrm_mem_evaluate_mm(old_mm);
 567                 spin_unlock(&old_mm->peertask_lock);
 568         }
 569         spin_lock(&mm->peertask_lock);
 570         list_add_tail(&tsk->mm_peers, &mm->tasklist);
 571         ckrm_mem_evaluate_mm(mm);
 572         spin_unlock(&mm->peertask_lock);
 573 #endif
 574         if (old_mm) {
 575                 if (active_mm != old_mm) BUG();
 576                 mmput(old_mm);
 577                 return 0;
 578         }
 579         mmdrop(active_mm);
 580         return 0;
 581 }
 582
 583 /*
 584  * This function makes sure the current process has its own signal table,
 585  * so that flush_signal_handlers can later reset the handlers without
 586  * disturbing other processes.  (Other processes might share the signal
 587  * table via the CLONE_SIGHAND option to clone().)
 588  */
 589 static inline int de_thread(struct task_struct *tsk)
 590 {
 591         struct signal_struct *newsig, *oldsig = tsk->signal;
 592         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 593         spinlock_t *lock = &oldsighand->siglock;
 594         int count;
 595
 596         /*
 597          * If we don't share sighandlers, then we aren't sharing anything
 598          * and we can just re-use it all.
 599          */
 600         if (atomic_read(&oldsighand->count) <= 1)
 601                 return 0;
 602
 603         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 604         if (!newsighand)
 605                 return -ENOMEM;
 606
 607         spin_lock_init(&newsighand->siglock);
 608         atomic_set(&newsighand->count, 1);
 609         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
 610
 611         /*
 612          * See if we need to allocate a new signal structure
 613          */
 614         newsig = NULL;
 615         if (atomic_read(&oldsig->count) > 1) {
 616                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 617                 if (!newsig) {
 618                         kmem_cache_free(sighand_cachep, newsighand);
 619                         return -ENOMEM;
 620                 }
 621                 atomic_set(&newsig->count, 1);
 622                 newsig->group_exit = 0;
 623                 newsig->group_exit_code = 0;
 624                 newsig->group_exit_task = NULL;
 625                 newsig->group_stop_count = 0;
 626                 newsig->curr_target = NULL;
 627                 init_sigpending(&newsig->shared_pending);
 628                 INIT_LIST_HEAD(&newsig->posix_timers);
 629
 630                 newsig->tty = oldsig->tty;
 631                 newsig->pgrp = oldsig->pgrp;
 632                 newsig->session = oldsig->session;
 633                 newsig->leader = oldsig->leader;
 634                 newsig->tty_old_pgrp = oldsig->tty_old_pgrp;
 635         }
 636
 637         if (thread_group_empty(current))
 638                 goto no_thread_group;
 639
 640         /*
 641          * Kill all other threads in the thread group.
 642          * We must hold tasklist_lock to call zap_other_threads.
 643          */
 644         read_lock(&tasklist_lock);
 645         spin_lock_irq(lock);
 646         if (oldsig->group_exit) {
 647                 /*
 648                  * Another group action in progress, just
 649                  * return so that the signal is processed.
 650                  */
 651                 spin_unlock_irq(lock);
 652                 read_unlock(&tasklist_lock);
 653                 kmem_cache_free(sighand_cachep, newsighand);
 654                 if (newsig)
 655                         kmem_cache_free(signal_cachep, newsig);
 656                 return -EAGAIN;
 657         }
 658         oldsig->group_exit = 1;
 659         zap_other_threads(current);
 660         read_unlock(&tasklist_lock);
 661
 662         /*
 663          * Account for the thread group leader hanging around:
 664          */
 665         count = 2;
 666         if (current->pid == current->tgid)
 667                 count = 1;
 668         while (atomic_read(&oldsig->count) > count) {
 669                 oldsig->group_exit_task = current;
 670                 oldsig->notify_count = count;
 671                 __set_current_state(TASK_UNINTERRUPTIBLE);
 672                 spin_unlock_irq(lock);
 673                 schedule();
 674                 spin_lock_irq(lock);
 675         }
 676         spin_unlock_irq(lock);
 677
 678         /*
 679          * At this point all other threads have exited, all we have to
 680          * do is to wait for the thread group leader to become inactive,
 681          * and to assume its PID:
 682          */
 683         if (current->pid != current->tgid) {
 684                 struct task_struct *leader = current->group_leader, *parent;
 685                 struct dentry *proc_dentry1, *proc_dentry2;
 686                 unsigned long state, ptrace;
 687
 688                 /*
 689                  * Wait for the thread group leader to be a zombie.
 690                  * It should already be zombie at this point, most
 691                  * of the time.
 692                  */
 693                 while (leader->state != TASK_ZOMBIE)
 694                         yield();
 695
 696                 spin_lock(&leader->proc_lock);
 697                 spin_lock(&current->proc_lock);
 698                 proc_dentry1 = proc_pid_unhash(current);
 699                 proc_dentry2 = proc_pid_unhash(leader);
 700                 write_lock_irq(&tasklist_lock);
 701
 702                 if (leader->tgid != current->tgid)
 703                         BUG();
 704                 if (current->pid == current->tgid)
 705                         BUG();
 706                 /*
 707                  * An exec() starts a new thread group with the
 708                  * TGID of the previous thread group. Rehash the
 709                  * two threads with a switched PID, and release
 710                  * the former thread group leader:
 711                  */
 712                 ptrace = leader->ptrace;
 713                 parent = leader->parent;
 714
 715                 ptrace_unlink(current);
 716                 ptrace_unlink(leader);
 717                 remove_parent(current);
 718                 remove_parent(leader);
 719
 720                 switch_exec_pids(leader, current);
 721
 722                 current->parent = current->real_parent = leader->real_parent;
 723                 leader->parent = leader->real_parent = child_reaper;
 724                 current->group_leader = current;
 725                 leader->group_leader = leader;
 726
 727                 add_parent(current, current->parent);
 728                 add_parent(leader, leader->parent);
 729                 if (ptrace) {
 730                         current->ptrace = ptrace;
 731                         __ptrace_link(current, parent);
 732                 }
 733
 734                 list_del(&current->tasks);
 735                 list_add_tail(&current->tasks, &init_task.tasks);
 736                 current->exit_signal = SIGCHLD;
 737                 state = leader->state;
 738
 739                 write_unlock_irq(&tasklist_lock);
 740                 spin_unlock(&leader->proc_lock);
 741                 spin_unlock(&current->proc_lock);
 742                 proc_pid_flush(proc_dentry1);
 743                 proc_pid_flush(proc_dentry2);
 744
 745                 if (state != TASK_ZOMBIE)
 746                         BUG();
 747                 release_task(leader);
 748         }
 749
 750 no_thread_group:
 751
 752         write_lock_irq(&tasklist_lock);
 753         spin_lock(&oldsighand->siglock);
 754         spin_lock(&newsighand->siglock);
 755
 756         if (current == oldsig->curr_target)
 757                 oldsig->curr_target = next_thread(current);
 758         if (newsig)
 759                 current->signal = newsig;
 760         current->sighand = newsighand;
 761         init_sigpending(&current->pending);
 762         recalc_sigpending();
 763
 764         spin_unlock(&newsighand->siglock);
 765         spin_unlock(&oldsighand->siglock);
 766         write_unlock_irq(&tasklist_lock);
 767
 768         if (newsig && atomic_dec_and_test(&oldsig->count))
 769                 kmem_cache_free(signal_cachep, oldsig);
 770
 771         if (atomic_dec_and_test(&oldsighand->count))
 772                 kmem_cache_free(sighand_cachep, oldsighand);
 773
 774         if (!thread_group_empty(current))
 775                 BUG();
 776         if (current->tgid != current->pid)
 777                 BUG();
 778         return 0;
 779 }
 780
 781 /*
 782  * These functions flushes out all traces of the currently running executable
 783  * so that a new one can be started
 784  */
 785
 786 static inline void flush_old_files(struct files_struct * files)
 787 {
 788         long j = -1;
 789
 790         spin_lock(&files->file_lock);
 791         for (;;) {
 792                 unsigned long set, i;
 793
 794                 j++;
 795                 i = j * __NFDBITS;
 796                 if (i >= files->max_fds || i >= files->max_fdset)
 797                         break;
 798                 set = files->close_on_exec->fds_bits[j];
 799                 if (!set)
 800                         continue;
 801                 files->close_on_exec->fds_bits[j] = 0;
 802                 spin_unlock(&files->file_lock);
 803                 for ( ; set ; i++,set >>= 1) {
 804                         if (set & 1) {
 805                                 sys_close(i);
 806                         }
 807                 }
 808                 spin_lock(&files->file_lock);
 809
 810         }
 811         spin_unlock(&files->file_lock);
 812 }
 813
 814 int flush_old_exec(struct linux_binprm * bprm)
 815 {
 816         char * name;
 817         int i, ch, retval;
 818         struct files_struct *files;
 819
 820         /*
 821          * Make sure we have a private signal table and that
 822          * we are unassociated from the previous thread group.
 823          */
 824         retval = de_thread(current);
 825         if (retval)
 826                 goto out;
 827
 828         /*
 829          * Make sure we have private file handles. Ask the
 830          * fork helper to do the work for us and the exit
 831          * helper to do the cleanup of the old one.
 832          */
 833         files = current->files;         /* refcounted so safe to hold */
 834         retval = unshare_files();
 835         if (retval)
 836                 goto out;
 837         /*
 838          * Release all of the old mmap stuff
 839          */
 840         retval = exec_mmap(bprm->mm);
 841         if (retval)
 842                 goto mmap_failed;
 843
 844         bprm->mm = NULL;                /* We're using it now */
 845
 846         /* This is the point of no return */
 847         steal_locks(files);
 848         put_files_struct(files);
 849
 850         current->sas_ss_sp = current->sas_ss_size = 0;
 851
 852         if (current->euid == current->uid && current->egid == current->gid)
 853                 current->mm->dumpable = 1;
 854         name = bprm->filename;
 855         for (i=0; (ch = *(name++)) != '\0';) {
 856                 if (ch == '/')
 857                         i = 0;
 858                 else
 859                         if (i < 15)
 860                                 current->comm[i++] = ch;
 861         }
 862         current->comm[i] = '\0';
 863
 864         current->flags &= ~PF_RELOCEXEC;
 865         flush_thread();
 866
 867         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 868             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
 869             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP))
 870                 current->mm->dumpable = 0;
 871
 872         /* An exec changes our domain. We are no longer part of the thread
 873            group */
 874
 875         current->self_exec_id++;
 876
 877         flush_signal_handlers(current, 0);
 878         flush_old_files(current->files);
 879
 880         return 0;
 881
 882 mmap_failed:
 883         put_files_struct(current->files);
 884         current->files = files;
 885 out:
 886         return retval;
 887 }
 888
 889 EXPORT_SYMBOL(flush_old_exec);
 890
 891 /*
 892  * Fill the binprm structure from the inode.
 893  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 894  */
 895 int prepare_binprm(struct linux_binprm *bprm)
 896 {
 897         int mode;
 898         struct inode * inode = bprm->file->f_dentry->d_inode;
 899         int retval;
 900
 901         mode = inode->i_mode;
 902         /*
 903          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 904          * vfs_permission lets a non-executable through
 905          */
 906         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 907                 return -EACCES;
 908         if (bprm->file->f_op == NULL)
 909                 return -EACCES;
 910
 911         bprm->e_uid = current->euid;
 912         bprm->e_gid = current->egid;
 913
 914         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 915                 /* Set-uid? */
 916                 if (mode & S_ISUID) {
 917                         current->personality &= ~PER_CLEAR_ON_SETID;
 918                         bprm->e_uid = inode->i_uid;
 919                 }
 920
 921                 /* Set-gid? */
 922                 /*
 923                  * If setgid is set but no group execute bit then this
 924                  * is a candidate for mandatory locking, not a setgid
 925                  * executable.
 926                  */
 927                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 928                         current->personality &= ~PER_CLEAR_ON_SETID;
 929                         bprm->e_gid = inode->i_gid;
 930                 }
 931         }
 932
 933         /* fill in binprm security blob */
 934         retval = security_bprm_set(bprm);
 935         if (retval)
 936                 return retval;
 937
 938         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 939         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 940 }
 941
 942 EXPORT_SYMBOL(prepare_binprm);
 943
 944 static inline int unsafe_exec(struct task_struct *p)
 945 {
 946         int unsafe = 0;
 947         if (p->ptrace & PT_PTRACED) {
 948                 if (p->ptrace & PT_PTRACE_CAP)
 949                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 950                 else
 951                         unsafe |= LSM_UNSAFE_PTRACE;
 952         }
 953         if (atomic_read(&p->fs->count) > 1 ||
 954             atomic_read(&p->files->count) > 1 ||
 955             atomic_read(&p->sighand->count) > 1)
 956                 unsafe |= LSM_UNSAFE_SHARE;
 957
 958         return unsafe;
 959 }
 960
 961 void compute_creds(struct linux_binprm *bprm)
 962 {
 963         int unsafe;
 964         task_lock(current);
 965         unsafe = unsafe_exec(current);
 966         security_bprm_apply_creds(bprm, unsafe);
 967         task_unlock(current);
 968 }
 969
 970 EXPORT_SYMBOL(compute_creds);
 971
 972 void remove_arg_zero(struct linux_binprm *bprm)
 973 {
 974         if (bprm->argc) {
 975                 unsigned long offset;
 976                 char * kaddr;
 977                 struct page *page;
 978
 979                 offset = bprm->p % PAGE_SIZE;
 980                 goto inside;
 981
 982                 while (bprm->p++, *(kaddr+offset++)) {
 983                         if (offset != PAGE_SIZE)
 984                                 continue;
 985                         offset = 0;
 986                         kunmap_atomic(kaddr, KM_USER0);
 987 inside:
 988                         page = bprm->page[bprm->p/PAGE_SIZE];
 989                         kaddr = kmap_atomic(page, KM_USER0);
 990                 }
 991                 kunmap_atomic(kaddr, KM_USER0);
 992                 bprm->argc--;
 993         }
 994 }
 995
 996 EXPORT_SYMBOL(remove_arg_zero);
 997
 998 /*
 999  * cycle the list of binary formats handler, until one recognizes the image
1000  */
1001 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1002 {
1003         int try,retval=0;
1004         struct linux_binfmt *fmt;
1005 #ifdef __alpha__
1006         /* handle /sbin/loader.. */
1007         {
1008             struct exec * eh = (struct exec *) bprm->buf;
1009
1010             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1011                 (eh->fh.f_flags & 0x3000) == 0x3000)
1012             {
1013                 struct file * file;
1014                 unsigned long loader;
1015
1016                 allow_write_access(bprm->file);
1017                 fput(bprm->file);
1018                 bprm->file = NULL;
1019
1020                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1021
1022                 file = open_exec("/sbin/loader");
1023                 retval = PTR_ERR(file);
1024                 if (IS_ERR(file))
1025                         return retval;
1026
1027                 /* Remember if the application is TASO.  */
1028                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1029
1030                 bprm->file = file;
1031                 bprm->loader = loader;
1032                 retval = prepare_binprm(bprm);
1033                 if (retval<0)
1034                         return retval;
1035                 /* should call search_binary_handler recursively here,
1036                    but it does not matter */
1037             }
1038         }
1039 #endif
1040         retval = security_bprm_check(bprm);
1041         if (retval)
1042                 return retval;
1043
1044         /* kernel module loader fixup */
1045         /* so we don't try to load run modprobe in kernel space. */
1046         set_fs(USER_DS);
1047         for (try=0; try<2; try++) {
1048                 read_lock(&binfmt_lock);
1049                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1050                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1051                         if (!fn)
1052                                 continue;
1053                         if (!try_module_get(fmt->module))
1054                                 continue;
1055                         read_unlock(&binfmt_lock);
1056                         retval = fn(bprm, regs);
1057                         if (retval >= 0) {
1058                                 put_binfmt(fmt);
1059                                 allow_write_access(bprm->file);
1060                                 if (bprm->file)
1061                                         fput(bprm->file);
1062                                 bprm->file = NULL;
1063                                 current->did_exec = 1;
1064                                 ckrm_cb_exec(bprm->filename);
1065                                 return retval;
1066                         }
1067                         read_lock(&binfmt_lock);
1068                         put_binfmt(fmt);
1069                         if (retval != -ENOEXEC || bprm->mm == NULL)
1070                                 break;
1071                         if (!bprm->file) {
1072                                 read_unlock(&binfmt_lock);
1073                                 return retval;
1074                         }
1075                 }
1076                 read_unlock(&binfmt_lock);
1077                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1078                         break;
1079 #ifdef CONFIG_KMOD
1080                 }else{
1081 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1082                         if (printable(bprm->buf[0]) &&
1083                             printable(bprm->buf[1]) &&
1084                             printable(bprm->buf[2]) &&
1085                             printable(bprm->buf[3]))
1086                                 break; /* -ENOEXEC */
1087                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1088 #endif
1089                 }
1090         }
1091         return retval;
1092 }
1093
1094 EXPORT_SYMBOL(search_binary_handler);
1095
1096 /*
1097  * sys_execve() executes a new program.
1098  */
1099 int do_execve(char * filename,
1100         char __user *__user *argv,
1101         char __user *__user *envp,
1102         struct pt_regs * regs)
1103 {
1104         struct linux_binprm bprm;
1105         struct file *file;
1106         int retval;
1107         int i;
1108
1109         file = open_exec(filename);
1110
1111         retval = PTR_ERR(file);
1112         if (IS_ERR(file))
1113                 return retval;
1114
1115         sched_balance_exec();
1116
1117         bprm.p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1118         memset(bprm.page, 0, MAX_ARG_PAGES*sizeof(bprm.page[0]));
1119
1120         bprm.file = file;
1121         bprm.filename = filename;
1122         bprm.interp = filename;
1123         bprm.interp_flags = 0;
1124         bprm.interp_data = 0;
1125         bprm.sh_bang = 0;
1126         bprm.loader = 0;
1127         bprm.exec = 0;
1128         bprm.security = NULL;
1129         bprm.mm = mm_alloc();
1130         retval = -ENOMEM;
1131         if (!bprm.mm)
1132                 goto out_file;
1133
1134         retval = init_new_context(current, bprm.mm);
1135         if (retval < 0)
1136                 goto out_mm;
1137
1138         bprm.argc = count(argv, bprm.p / sizeof(void *));
1139         if ((retval = bprm.argc) < 0)
1140                 goto out_mm;
1141
1142         bprm.envc = count(envp, bprm.p / sizeof(void *));
1143         if ((retval = bprm.envc) < 0)
1144                 goto out_mm;
1145
1146         retval = security_bprm_alloc(&bprm);
1147         if (retval)
1148                 goto out;
1149
1150         retval = prepare_binprm(&bprm);
1151         if (retval < 0)
1152                 goto out;
1153
1154         retval = copy_strings_kernel(1, &bprm.filename, &bprm);
1155         if (retval < 0)
1156                 goto out;
1157
1158         bprm.exec = bprm.p;
1159         retval = copy_strings(bprm.envc, envp, &bprm);
1160         if (retval < 0)
1161                 goto out;
1162
1163         retval = copy_strings(bprm.argc, argv, &bprm);
1164         if (retval < 0)
1165                 goto out;
1166
1167         retval = search_binary_handler(&bprm,regs);
1168         if (retval >= 0) {
1169                 free_arg_pages(&bprm);
1170
1171                 /* execve success */
1172                 security_bprm_free(&bprm);
1173                 return retval;
1174         }
1175
1176 out:
1177         /* Something went wrong, return the inode and free the argument pages*/
1178         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1179                 struct page * page = bprm.page[i];
1180                 if (page)
1181                         __free_page(page);
1182         }
1183
1184         if (bprm.security)
1185                 security_bprm_free(&bprm);
1186
1187 out_mm:
1188         if (bprm.mm)
1189                 mmdrop(bprm.mm);
1190
1191 out_file:
1192         if (bprm.file) {
1193                 allow_write_access(bprm.file);
1194                 fput(bprm.file);
1195         }
1196         return retval;
1197 }
1198
1199 EXPORT_SYMBOL(do_execve);
1200
1201 int set_binfmt(struct linux_binfmt *new)
1202 {
1203         struct linux_binfmt *old = current->binfmt;
1204
1205         if (new) {
1206                 if (!try_module_get(new->module))
1207                         return -1;
1208         }
1209         current->binfmt = new;
1210         if (old)
1211                 module_put(old->module);
1212         return 0;
1213 }
1214
1215 EXPORT_SYMBOL(set_binfmt);
1216
1217 #define CORENAME_MAX_SIZE 64
1218
1219 /* format_corename will inspect the pattern parameter, and output a
1220  * name into corename, which must have space for at least
1221  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1222  */
1223 void format_corename(char *corename, const char *pattern, long signr)
1224 {
1225         const char *pat_ptr = pattern;
1226         char *out_ptr = corename;
1227         char *const out_end = corename + CORENAME_MAX_SIZE;
1228         int rc;
1229         int pid_in_pattern = 0;
1230
1231         /* Repeat as long as we have more pattern to process and more output
1232            space */
1233         while (*pat_ptr) {
1234                 if (*pat_ptr != '%') {
1235                         if (out_ptr == out_end)
1236                                 goto out;
1237                         *out_ptr++ = *pat_ptr++;
1238                 } else {
1239                         switch (*++pat_ptr) {
1240                         case 0:
1241                                 goto out;
1242                         /* Double percent, output one percent */
1243                         case '%':
1244                                 if (out_ptr == out_end)
1245                                         goto out;
1246                                 *out_ptr++ = '%';
1247                                 break;
1248                         /* pid */
1249                         case 'p':
1250                                 pid_in_pattern = 1;
1251                                 rc = snprintf(out_ptr, out_end - out_ptr,
1252                                               "%d", current->tgid);
1253                                 if (rc > out_end - out_ptr)
1254                                         goto out;
1255                                 out_ptr += rc;
1256                                 break;
1257                         /* uid */
1258                         case 'u':
1259                                 rc = snprintf(out_ptr, out_end - out_ptr,
1260                                               "%d", current->uid);
1261                                 if (rc > out_end - out_ptr)
1262                                         goto out;
1263                                 out_ptr += rc;
1264                                 break;
1265                         /* gid */
1266                         case 'g':
1267                                 rc = snprintf(out_ptr, out_end - out_ptr,
1268                                               "%d", current->gid);
1269                                 if (rc > out_end - out_ptr)
1270                                         goto out;
1271                                 out_ptr += rc;
1272                                 break;
1273                         /* signal that caused the coredump */
1274                         case 's':
1275                                 rc = snprintf(out_ptr, out_end - out_ptr,
1276                                               "%ld", signr);
1277                                 if (rc > out_end - out_ptr)
1278                                         goto out;
1279                                 out_ptr += rc;
1280                                 break;
1281                         /* UNIX time of coredump */
1282                         case 't': {
1283                                 struct timeval tv;
1284                                 do_gettimeofday(&tv);
1285                                 rc = snprintf(out_ptr, out_end - out_ptr,
1286                                               "%lu", tv.tv_sec);
1287                                 if (rc > out_end - out_ptr)
1288                                         goto out;
1289                                 out_ptr += rc;
1290                                 break;
1291                         }
1292                         /* hostname */
1293                         case 'h':
1294                                 down_read(&uts_sem);
1295                                 rc = snprintf(out_ptr, out_end - out_ptr,
1296                                               "%s", system_utsname.nodename);
1297                                 up_read(&uts_sem);
1298                                 if (rc > out_end - out_ptr)
1299                                         goto out;
1300                                 out_ptr += rc;
1301                                 break;
1302                         /* executable */
1303                         case 'e':
1304                                 rc = snprintf(out_ptr, out_end - out_ptr,
1305                                               "%s", current->comm);
1306                                 if (rc > out_end - out_ptr)
1307                                         goto out;
1308                                 out_ptr += rc;
1309                                 break;
1310                         default:
1311                                 break;
1312                         }
1313                         ++pat_ptr;
1314                 }
1315         }
1316         /* Backward compatibility with core_uses_pid:
1317          *
1318          * If core_pattern does not include a %p (as is the default)
1319          * and core_uses_pid is set, then .%pid will be appended to
1320          * the filename */
1321         if (!pid_in_pattern
1322             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1323                 rc = snprintf(out_ptr, out_end - out_ptr,
1324                               ".%d", current->tgid);
1325                 if (rc > out_end - out_ptr)
1326                         goto out;
1327                 out_ptr += rc;
1328         }
1329       out:
1330         *out_ptr = 0;
1331 }
1332
1333 static void zap_threads (struct mm_struct *mm)
1334 {
1335         struct task_struct *g, *p;
1336         struct task_struct *tsk = current;
1337         struct completion *vfork_done = tsk->vfork_done;
1338
1339         /*
1340          * Make sure nobody is waiting for us to release the VM,
1341          * otherwise we can deadlock when we wait on each other
1342          */
1343         if (vfork_done) {
1344                 tsk->vfork_done = NULL;
1345                 complete(vfork_done);
1346         }
1347
1348         read_lock(&tasklist_lock);
1349         do_each_thread(g,p)
1350                 if (mm == p->mm && p != tsk) {
1351                         force_sig_specific(SIGKILL, p);
1352                         mm->core_waiters++;
1353                 }
1354         while_each_thread(g,p);
1355
1356         read_unlock(&tasklist_lock);
1357 }
1358
1359 static void coredump_wait(struct mm_struct *mm)
1360 {
1361         DECLARE_COMPLETION(startup_done);
1362
1363         mm->core_waiters++; /* let other threads block */
1364         mm->core_startup_done = &startup_done;
1365
1366         /* give other threads a chance to run: */
1367         yield();
1368
1369         zap_threads(mm);
1370         if (--mm->core_waiters) {
1371                 up_write(&mm->mmap_sem);
1372                 wait_for_completion(&startup_done);
1373         } else
1374                 up_write(&mm->mmap_sem);
1375         BUG_ON(mm->core_waiters);
1376 }
1377
1378 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1379 {
1380         char corename[CORENAME_MAX_SIZE + 1];
1381         struct mm_struct *mm = current->mm;
1382         struct linux_binfmt * binfmt;
1383         struct inode * inode;
1384         struct file * file;
1385         int retval = 0;
1386
1387         lock_kernel();
1388         binfmt = current->binfmt;
1389         if (!binfmt || !binfmt->core_dump)
1390                 goto fail;
1391         down_write(&mm->mmap_sem);
1392         if (!mm->dumpable) {
1393                 up_write(&mm->mmap_sem);
1394                 goto fail;
1395         }
1396         mm->dumpable = 0;
1397         init_completion(&mm->core_done);
1398         current->signal->group_exit = 1;
1399         current->signal->group_exit_code = exit_code;
1400         coredump_wait(mm);
1401
1402         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1403                 goto fail_unlock;
1404
1405         format_corename(corename, core_pattern, signr);
1406         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1407         if (IS_ERR(file))
1408                 goto fail_unlock;
1409         inode = file->f_dentry->d_inode;
1410         if (inode->i_nlink > 1)
1411                 goto close_fail;        /* multiple links - don't dump */
1412         if (d_unhashed(file->f_dentry))
1413                 goto close_fail;
1414
1415         if (!S_ISREG(inode->i_mode))
1416                 goto close_fail;
1417         if (!file->f_op)
1418                 goto close_fail;
1419         if (!file->f_op->write)
1420                 goto close_fail;
1421         if (do_truncate(file->f_dentry, 0) != 0)
1422                 goto close_fail;
1423
1424         retval = binfmt->core_dump(signr, regs, file);
1425
1426         current->signal->group_exit_code |= 0x80;
1427 close_fail:
1428         filp_close(file, NULL);
1429 fail_unlock:
1430         complete_all(&mm->core_done);
1431 fail:
1432         unlock_kernel();
1433         return retval;
1434 }