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 int suid_dumpable = 0;
  63
  64 EXPORT_SYMBOL(suid_dumpable);
  65
  66 /* The maximal length of core_pattern is also specified in sysctl.c */
  67
  68 static struct linux_binfmt *formats;
  69 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  70
  71 int register_binfmt(struct linux_binfmt * fmt)
  72 {
  73         struct linux_binfmt ** tmp = &formats;
  74
  75         if (!fmt)
  76                 return -EINVAL;
  77         if (fmt->next)
  78                 return -EBUSY;
  79         write_lock(&binfmt_lock);
  80         while (*tmp) {
  81                 if (fmt == *tmp) {
  82                         write_unlock(&binfmt_lock);
  83                         return -EBUSY;
  84                 }
  85                 tmp = &(*tmp)->next;
  86         }
  87         fmt->next = formats;
  88         formats = fmt;
  89         write_unlock(&binfmt_lock);
  90         return 0;
  91 }
  92
  93 EXPORT_SYMBOL(register_binfmt);
  94
  95 int unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         struct linux_binfmt ** tmp = &formats;
  98
  99         write_lock(&binfmt_lock);
 100         while (*tmp) {
 101                 if (fmt == *tmp) {
 102                         *tmp = fmt->next;
 103                         write_unlock(&binfmt_lock);
 104                         return 0;
 105                 }
 106                 tmp = &(*tmp)->next;
 107         }
 108         write_unlock(&binfmt_lock);
 109         return -EINVAL;
 110 }
 111
 112 EXPORT_SYMBOL(unregister_binfmt);
 113
 114 static inline void put_binfmt(struct linux_binfmt * fmt)
 115 {
 116         module_put(fmt->module);
 117 }
 118
 119 /*
 120  * Note that a shared library must be both readable and executable due to
 121  * security reasons.
 122  *
 123  * Also note that we take the address to load from from the file itself.
 124  */
 125 asmlinkage long sys_uselib(const char __user * library)
 126 {
 127         struct file * file;
 128         struct nameidata nd;
 129         int error;
 130
 131         nd.intent.open.flags = FMODE_READ;
 132         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 133         if (error)
 134                 goto out;
 135
 136         error = -EINVAL;
 137         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 138                 goto exit;
 139
 140         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 141         if (error)
 142                 goto exit;
 143
 144         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 145         error = PTR_ERR(file);
 146         if (IS_ERR(file))
 147                 goto out;
 148
 149         error = -ENOEXEC;
 150         if(file->f_op) {
 151                 struct linux_binfmt * fmt;
 152
 153                 read_lock(&binfmt_lock);
 154                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 155                         if (!fmt->load_shlib)
 156                                 continue;
 157                         if (!try_module_get(fmt->module))
 158                                 continue;
 159                         read_unlock(&binfmt_lock);
 160                         error = fmt->load_shlib(file);
 161                         read_lock(&binfmt_lock);
 162                         put_binfmt(fmt);
 163                         if (error != -ENOEXEC)
 164                                 break;
 165                 }
 166                 read_unlock(&binfmt_lock);
 167         }
 168         fput(file);
 169 out:
 170         return error;
 171 exit:
 172         path_release(&nd);
 173         goto out;
 174 }
 175
 176 /*
 177  * count() counts the number of strings in array ARGV.
 178  */
 179 static int count(char __user * __user * argv, int max)
 180 {
 181         int i = 0;
 182
 183         if (argv != NULL) {
 184                 for (;;) {
 185                         char __user * p;
 186
 187                         if (get_user(p, argv))
 188                                 return -EFAULT;
 189                         if (!p)
 190                                 break;
 191                         argv++;
 192                         if(++i > max)
 193                                 return -E2BIG;
 194                 }
 195         }
 196         return i;
 197 }
 198
 199 /*
 200  * 'copy_strings()' copies argument/environment strings from user
 201  * memory to free pages in kernel mem. These are in a format ready
 202  * to be put directly into the top of new user memory.
 203  */
 204 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 205 {
 206         struct page *kmapped_page = NULL;
 207         char *kaddr = NULL;
 208         int ret;
 209
 210         while (argc-- > 0) {
 211                 char __user *str;
 212                 int len;
 213                 unsigned long pos;
 214
 215                 if (get_user(str, argv+argc) ||
 216                                 !(len = strnlen_user(str, bprm->p))) {
 217                         ret = -EFAULT;
 218                         goto out;
 219                 }
 220
 221                 if (bprm->p < len)  {
 222                         ret = -E2BIG;
 223                         goto out;
 224                 }
 225
 226                 bprm->p -= len;
 227                 /* XXX: add architecture specific overflow check here. */
 228                 pos = bprm->p;
 229
 230                 while (len > 0) {
 231                         int i, new, err;
 232                         int offset, bytes_to_copy;
 233                         struct page *page;
 234
 235                         offset = pos % PAGE_SIZE;
 236                         i = pos/PAGE_SIZE;
 237                         page = bprm->page[i];
 238                         new = 0;
 239                         if (!page) {
 240                                 page = alloc_page(GFP_HIGHUSER);
 241                                 bprm->page[i] = page;
 242                                 if (!page) {
 243                                         ret = -ENOMEM;
 244                                         goto out;
 245                                 }
 246                                 new = 1;
 247                         }
 248
 249                         if (page != kmapped_page) {
 250                                 if (kmapped_page)
 251                                         kunmap(kmapped_page);
 252                                 kmapped_page = page;
 253                                 kaddr = kmap(kmapped_page);
 254                         }
 255                         if (new && offset)
 256                                 memset(kaddr, 0, offset);
 257                         bytes_to_copy = PAGE_SIZE - offset;
 258                         if (bytes_to_copy > len) {
 259                                 bytes_to_copy = len;
 260                                 if (new)
 261                                         memset(kaddr+offset+len, 0,
 262                                                 PAGE_SIZE-offset-len);
 263                         }
 264                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 265                         if (err) {
 266                                 ret = -EFAULT;
 267                                 goto out;
 268                         }
 269
 270                         pos += bytes_to_copy;
 271                         str += bytes_to_copy;
 272                         len -= bytes_to_copy;
 273                 }
 274         }
 275         ret = 0;
 276 out:
 277         if (kmapped_page)
 278                 kunmap(kmapped_page);
 279         return ret;
 280 }
 281
 282 /*
 283  * Like copy_strings, but get argv and its values from kernel memory.
 284  */
 285 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 286 {
 287         int r;
 288         mm_segment_t oldfs = get_fs();
 289         set_fs(KERNEL_DS);
 290         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 291         set_fs(oldfs);
 292         return r;
 293 }
 294
 295 EXPORT_SYMBOL(copy_strings_kernel);
 296
 297 #ifdef CONFIG_MMU
 298 /*
 299  * This routine is used to map in a page into an address space: needed by
 300  * execve() for the initial stack and environment pages.
 301  *
 302  * vma->vm_mm->mmap_sem is held for writing.
 303  */
 304 void install_arg_page(struct vm_area_struct *vma,
 305                         struct page *page, unsigned long address)
 306 {
 307         struct mm_struct *mm = vma->vm_mm;
 308         pgd_t * pgd;
 309         pmd_t * pmd;
 310         pte_t * pte;
 311
 312         if (unlikely(anon_vma_prepare(vma)))
 313                 goto out_sig;
 314
 315         flush_dcache_page(page);
 316         pgd = pgd_offset(mm, address);
 317
 318         spin_lock(&mm->page_table_lock);
 319         pmd = pmd_alloc(mm, pgd, address);
 320         if (!pmd)
 321                 goto out;
 322         pte = pte_alloc_map(mm, pmd, address);
 323         if (!pte)
 324                 goto out;
 325         if (!pte_none(*pte)) {
 326                 pte_unmap(pte);
 327                 goto out;
 328         }
 329         // mm->rss++;
 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 int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
 348 {
 349         unsigned long stack_base;
 350         struct vm_area_struct *mpnt;
 351         struct mm_struct *mm = current->mm;
 352         int i, ret;
 353         long arg_size;
 354
 355 #ifdef CONFIG_STACK_GROWSUP
 356         /* Move the argument and environment strings to the bottom of the
 357          * stack space.
 358          */
 359         int offset, j;
 360         char *to, *from;
 361
 362         /* Start by shifting all the pages down */
 363         i = 0;
 364         for (j = 0; j < MAX_ARG_PAGES; j++) {
 365                 struct page *page = bprm->page[j];
 366                 if (!page)
 367                         continue;
 368                 bprm->page[i++] = page;
 369         }
 370
 371         /* Now move them within their pages */
 372         offset = bprm->p % PAGE_SIZE;
 373         to = kmap(bprm->page[0]);
 374         for (j = 1; j < i; j++) {
 375                 memmove(to, to + offset, PAGE_SIZE - offset);
 376                 from = kmap(bprm->page[j]);
 377                 memcpy(to + PAGE_SIZE - offset, from, offset);
 378                 kunmap(bprm->page[j - 1]);
 379                 to = from;
 380         }
 381         memmove(to, to + offset, PAGE_SIZE - offset);
 382         kunmap(bprm->page[j - 1]);
 383
 384         /* Adjust bprm->p to point to the end of the strings. */
 385         bprm->p = PAGE_SIZE * i - offset;
 386
 387         /* Limit stack size to 1GB */
 388         stack_base = current->rlim[RLIMIT_STACK].rlim_max;
 389         if (stack_base > (1 << 30))
 390                 stack_base = 1 << 30;
 391         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
 392
 393         mm->arg_start = stack_base;
 394         arg_size = i << PAGE_SHIFT;
 395
 396         /* zero pages that were copied above */
 397         while (i < MAX_ARG_PAGES)
 398                 bprm->page[i++] = NULL;
 399 #else
 400 #ifdef __HAVE_ARCH_ALIGN_STACK
 401         stack_base = arch_align_stack(STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE);
 402         stack_base = PAGE_ALIGN(stack_base);
 403 #else
 404         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 405 #endif
 406         mm->arg_start = bprm->p + stack_base;
 407         arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
 408 #endif
 409
 410         bprm->p += stack_base;
 411         if (bprm->loader)
 412                 bprm->loader += stack_base;
 413         bprm->exec += stack_base;
 414
 415         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 416         if (!mpnt)
 417                 return -ENOMEM;
 418
 419         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT) ||
 420                 !vx_vmpages_avail(mm, arg_size >> PAGE_SHIFT)) {
 421                 kmem_cache_free(vm_area_cachep, mpnt);
 422                 return -ENOMEM;
 423         }
 424
 425         memset(mpnt, 0, sizeof(*mpnt));
 426
 427         down_write(&mm->mmap_sem);
 428         {
 429                 mpnt->vm_mm = mm;
 430 #ifdef CONFIG_STACK_GROWSUP
 431                 mpnt->vm_start = stack_base;
 432                 mpnt->vm_end = PAGE_MASK &
 433                         (PAGE_SIZE - 1 + (unsigned long) bprm->p);
 434 #else
 435                 mpnt->vm_start = PAGE_MASK & (unsigned long) bprm->p;
 436                 mpnt->vm_end = STACK_TOP;
 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                 // mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 455                 vx_vmpages_sub(mm, mm->total_vm - vma_pages(mpnt));
 456                 mm->stack_vm = mm->total_vm;
 457         }
 458
 459         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 460                 struct page *page = bprm->page[i];
 461                 if (page) {
 462                         bprm->page[i] = NULL;
 463                         install_arg_page(mpnt, page, stack_base);
 464                 }
 465                 stack_base += PAGE_SIZE;
 466         }
 467         up_write(&mm->mmap_sem);
 468
 469         return 0;
 470 }
 471
 472 EXPORT_SYMBOL(setup_arg_pages);
 473
 474 #define free_arg_pages(bprm) do { } while (0)
 475
 476 #else
 477
 478 static inline void free_arg_pages(struct linux_binprm *bprm)
 479 {
 480         int i;
 481
 482         for (i = 0; i < MAX_ARG_PAGES; i++) {
 483                 if (bprm->page[i])
 484                         __free_page(bprm->page[i]);
 485                 bprm->page[i] = NULL;
 486         }
 487 }
 488
 489 #endif /* CONFIG_MMU */
 490
 491 struct file *open_exec(const char *name)
 492 {
 493         struct nameidata nd;
 494         int err;
 495         struct file *file;
 496
 497         nd.intent.open.flags = FMODE_READ;
 498         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 499         file = ERR_PTR(err);
 500
 501         if (!err) {
 502                 struct inode *inode = nd.dentry->d_inode;
 503                 file = ERR_PTR(-EACCES);
 504                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 505                     S_ISREG(inode->i_mode)) {
 506                         int err = permission(inode, MAY_EXEC, &nd);
 507                         if (!err && !(inode->i_mode & 0111))
 508                                 err = -EACCES;
 509                         file = ERR_PTR(err);
 510                         if (!err) {
 511                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 512                                 if (!IS_ERR(file)) {
 513                                         err = deny_write_access(file);
 514                                         if (err) {
 515                                                 fput(file);
 516                                                 file = ERR_PTR(err);
 517                                         }
 518                                 }
 519 out:
 520                                 return file;
 521                         }
 522                 }
 523                 path_release(&nd);
 524         }
 525         goto out;
 526 }
 527
 528 EXPORT_SYMBOL(open_exec);
 529
 530 int kernel_read(struct file *file, unsigned long offset,
 531         char *addr, unsigned long count)
 532 {
 533         mm_segment_t old_fs;
 534         loff_t pos = offset;
 535         int result;
 536
 537         old_fs = get_fs();
 538         set_fs(get_ds());
 539         /* The cast to a user pointer is valid due to the set_fs() */
 540         result = vfs_read(file, (void __user *)addr, count, &pos);
 541         set_fs(old_fs);
 542         return result;
 543 }
 544
 545 EXPORT_SYMBOL(kernel_read);
 546
 547 static int exec_mmap(struct mm_struct *mm)
 548 {
 549         struct task_struct *tsk;
 550         struct mm_struct * old_mm, *active_mm;
 551
 552         /* Add it to the list of mm's */
 553         spin_lock(&mmlist_lock);
 554         list_add(&mm->mmlist, &init_mm.mmlist);
 555         mmlist_nr++;
 556         spin_unlock(&mmlist_lock);
 557
 558         /* Notify parent that we're no longer interested in the old VM */
 559         tsk = current;
 560         old_mm = current->mm;
 561         mm_release(tsk, old_mm);
 562
 563         task_lock(tsk);
 564         active_mm = tsk->active_mm;
 565         tsk->mm = mm;
 566         tsk->active_mm = mm;
 567         activate_mm(active_mm, mm);
 568         task_unlock(tsk);
 569         arch_pick_mmap_layout(mm);
 570 #ifdef CONFIG_CKRM_RES_MEM
 571         if (old_mm) {
 572                 spin_lock(&old_mm->peertask_lock);
 573                 list_del(&tsk->mm_peers);
 574                 ckrm_mem_evaluate_mm(old_mm);
 575                 spin_unlock(&old_mm->peertask_lock);
 576         }
 577         spin_lock(&mm->peertask_lock);
 578         list_add_tail(&tsk->mm_peers, &mm->tasklist);
 579         ckrm_mem_evaluate_mm(mm);
 580         spin_unlock(&mm->peertask_lock);
 581 #endif
 582         if (old_mm) {
 583                 if (active_mm != old_mm) BUG();
 584                 mmput(old_mm);
 585                 return 0;
 586         }
 587         mmdrop(active_mm);
 588         return 0;
 589 }
 590
 591 /*
 592  * This function makes sure the current process has its own signal table,
 593  * so that flush_signal_handlers can later reset the handlers without
 594  * disturbing other processes.  (Other processes might share the signal
 595  * table via the CLONE_SIGHAND option to clone().)
 596  */
 597 static inline int de_thread(struct task_struct *tsk)
 598 {
 599         struct signal_struct *newsig, *oldsig = tsk->signal;
 600         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 601         spinlock_t *lock = &oldsighand->siglock;
 602         int count;
 603
 604         /*
 605          * If we don't share sighandlers, then we aren't sharing anything
 606          * and we can just re-use it all.
 607          */
 608         if (atomic_read(&oldsighand->count) <= 1)
 609                 return 0;
 610
 611         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 612         if (!newsighand)
 613                 return -ENOMEM;
 614
 615         spin_lock_init(&newsighand->siglock);
 616         atomic_set(&newsighand->count, 1);
 617         memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action));
 618
 619         /*
 620          * See if we need to allocate a new signal structure
 621          */
 622         newsig = NULL;
 623         if (atomic_read(&oldsig->count) > 1) {
 624                 newsig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
 625                 if (!newsig) {
 626                         kmem_cache_free(sighand_cachep, newsighand);
 627                         return -ENOMEM;
 628                 }
 629                 atomic_set(&newsig->count, 1);
 630                 newsig->group_exit = 0;
 631                 newsig->group_exit_code = 0;
 632                 newsig->group_exit_task = NULL;
 633                 newsig->group_stop_count = 0;
 634                 newsig->curr_target = NULL;
 635                 init_sigpending(&newsig->shared_pending);
 636                 INIT_LIST_HEAD(&newsig->posix_timers);
 637
 638                 newsig->tty = oldsig->tty;
 639                 newsig->pgrp = oldsig->pgrp;
 640                 newsig->session = oldsig->session;
 641                 newsig->leader = oldsig->leader;
 642                 newsig->tty_old_pgrp = oldsig->tty_old_pgrp;
 643         }
 644
 645         if (thread_group_empty(current))
 646                 goto no_thread_group;
 647
 648         /*
 649          * Kill all other threads in the thread group.
 650          * We must hold tasklist_lock to call zap_other_threads.
 651          */
 652         read_lock(&tasklist_lock);
 653         spin_lock_irq(lock);
 654         if (oldsig->group_exit) {
 655                 /*
 656                  * Another group action in progress, just
 657                  * return so that the signal is processed.
 658                  */
 659                 spin_unlock_irq(lock);
 660                 read_unlock(&tasklist_lock);
 661                 kmem_cache_free(sighand_cachep, newsighand);
 662                 if (newsig)
 663                         kmem_cache_free(signal_cachep, newsig);
 664                 return -EAGAIN;
 665         }
 666         oldsig->group_exit = 1;
 667         zap_other_threads(current);
 668         read_unlock(&tasklist_lock);
 669
 670         /*
 671          * Account for the thread group leader hanging around:
 672          */
 673         count = 2;
 674         if (current->pid == current->tgid)
 675                 count = 1;
 676         while (atomic_read(&oldsig->count) > count) {
 677                 oldsig->group_exit_task = current;
 678                 oldsig->notify_count = count;
 679                 __set_current_state(TASK_UNINTERRUPTIBLE);
 680                 spin_unlock_irq(lock);
 681                 schedule();
 682                 spin_lock_irq(lock);
 683         }
 684         spin_unlock_irq(lock);
 685
 686         /*
 687          * At this point all other threads have exited, all we have to
 688          * do is to wait for the thread group leader to become inactive,
 689          * and to assume its PID:
 690          */
 691         if (current->pid != current->tgid) {
 692                 struct task_struct *leader = current->group_leader, *parent;
 693                 struct dentry *proc_dentry1, *proc_dentry2;
 694                 unsigned long exit_state, ptrace;
 695
 696                 /*
 697                  * Wait for the thread group leader to be a zombie.
 698                  * It should already be zombie at this point, most
 699                  * of the time.
 700                  */
 701                 while (leader->exit_state != EXIT_ZOMBIE)
 702                         yield();
 703
 704                 spin_lock(&leader->proc_lock);
 705                 spin_lock(&current->proc_lock);
 706                 proc_dentry1 = proc_pid_unhash(current);
 707                 proc_dentry2 = proc_pid_unhash(leader);
 708                 write_lock_irq(&tasklist_lock);
 709
 710                 if (leader->tgid != current->tgid)
 711                         BUG();
 712                 if (current->pid == current->tgid)
 713                         BUG();
 714                 /*
 715                  * An exec() starts a new thread group with the
 716                  * TGID of the previous thread group. Rehash the
 717                  * two threads with a switched PID, and release
 718                  * the former thread group leader:
 719                  */
 720                 ptrace = leader->ptrace;
 721                 parent = leader->parent;
 722
 723                 ptrace_unlink(current);
 724                 ptrace_unlink(leader);
 725                 remove_parent(current);
 726                 remove_parent(leader);
 727
 728                 switch_exec_pids(leader, current);
 729
 730                 current->parent = current->real_parent = leader->real_parent;
 731                 leader->parent = leader->real_parent = child_reaper;
 732                 current->group_leader = current;
 733                 leader->group_leader = leader;
 734
 735                 add_parent(current, current->parent);
 736                 add_parent(leader, leader->parent);
 737                 if (ptrace) {
 738                         current->ptrace = ptrace;
 739                         __ptrace_link(current, parent);
 740                 }
 741
 742                 list_del(&current->tasks);
 743                 list_add_tail(&current->tasks, &init_task.tasks);
 744                 current->exit_signal = SIGCHLD;
 745                 exit_state = leader->exit_state;
 746
 747                 write_unlock_irq(&tasklist_lock);
 748                 spin_unlock(&leader->proc_lock);
 749                 spin_unlock(&current->proc_lock);
 750                 proc_pid_flush(proc_dentry1);
 751                 proc_pid_flush(proc_dentry2);
 752
 753                 if (exit_state != EXIT_ZOMBIE)
 754                         BUG();
 755                 release_task(leader);
 756         }
 757
 758 no_thread_group:
 759
 760         write_lock_irq(&tasklist_lock);
 761         spin_lock(&oldsighand->siglock);
 762         spin_lock(&newsighand->siglock);
 763
 764         if (current == oldsig->curr_target)
 765                 oldsig->curr_target = next_thread(current);
 766         if (newsig)
 767                 current->signal = newsig;
 768         current->sighand = newsighand;
 769         init_sigpending(&current->pending);
 770         recalc_sigpending();
 771
 772         spin_unlock(&newsighand->siglock);
 773         spin_unlock(&oldsighand->siglock);
 774         write_unlock_irq(&tasklist_lock);
 775
 776         if (newsig && atomic_dec_and_test(&oldsig->count)) {
 777                 exit_itimers(oldsig);
 778                 kmem_cache_free(signal_cachep, oldsig);
 779         }
 780
 781         if (atomic_dec_and_test(&oldsighand->count))
 782                 kmem_cache_free(sighand_cachep, oldsighand);
 783
 784         if (!thread_group_empty(current))
 785                 BUG();
 786         if (current->tgid != current->pid)
 787                 BUG();
 788         return 0;
 789 }
 790
 791 /*
 792  * These functions flushes out all traces of the currently running executable
 793  * so that a new one can be started
 794  */
 795
 796 static inline void flush_old_files(struct files_struct * files)
 797 {
 798         long j = -1;
 799
 800         spin_lock(&files->file_lock);
 801         for (;;) {
 802                 unsigned long set, i;
 803
 804                 j++;
 805                 i = j * __NFDBITS;
 806                 if (i >= files->max_fds || i >= files->max_fdset)
 807                         break;
 808                 set = files->close_on_exec->fds_bits[j];
 809                 if (!set)
 810                         continue;
 811                 files->close_on_exec->fds_bits[j] = 0;
 812                 spin_unlock(&files->file_lock);
 813                 for ( ; set ; i++,set >>= 1) {
 814                         if (set & 1) {
 815                                 sys_close(i);
 816                         }
 817                 }
 818                 spin_lock(&files->file_lock);
 819
 820         }
 821         spin_unlock(&files->file_lock);
 822 }
 823
 824 void get_task_comm(char *buf, struct task_struct *tsk)
 825 {
 826         /* buf must be at least sizeof(tsk->comm) in size */
 827         task_lock(tsk);
 828         memcpy(buf, tsk->comm, sizeof(tsk->comm));
 829         task_unlock(tsk);
 830 }
 831
 832 void set_task_comm(struct task_struct *tsk, char *buf)
 833 {
 834         task_lock(tsk);
 835         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 836         task_unlock(tsk);
 837 }
 838
 839 int flush_old_exec(struct linux_binprm * bprm)
 840 {
 841         char * name;
 842         int i, ch, retval;
 843         struct files_struct *files;
 844         char tcomm[sizeof(current->comm)];
 845
 846         /*
 847          * Make sure we have a private signal table and that
 848          * we are unassociated from the previous thread group.
 849          */
 850         retval = de_thread(current);
 851         if (retval)
 852                 goto out;
 853
 854         /*
 855          * Make sure we have private file handles. Ask the
 856          * fork helper to do the work for us and the exit
 857          * helper to do the cleanup of the old one.
 858          */
 859         files = current->files;         /* refcounted so safe to hold */
 860         retval = unshare_files();
 861         if (retval)
 862                 goto out;
 863         /*
 864          * Release all of the old mmap stuff
 865          */
 866         retval = exec_mmap(bprm->mm);
 867         if (retval)
 868                 goto mmap_failed;
 869
 870         bprm->mm = NULL;                /* We're using it now */
 871
 872         /* This is the point of no return */
 873         steal_locks(files);
 874         put_files_struct(files);
 875
 876         current->sas_ss_sp = current->sas_ss_size = 0;
 877
 878         if (current->euid == current->uid && current->egid == current->gid)
 879                 current->mm->dumpable = 1;
 880         else
 881                 current->mm->dumpable = suid_dumpable;
 882
 883         name = bprm->filename;
 884         for (i=0; (ch = *(name++)) != '\0';) {
 885                 if (ch == '/')
 886                         i = 0;
 887                 else
 888                         if (i < (sizeof(tcomm) - 1))
 889                                 tcomm[i++] = ch;
 890         }
 891         tcomm[i] = '\0';
 892         set_task_comm(current, tcomm);
 893
 894         current->flags &= ~PF_RELOCEXEC;
 895         flush_thread();
 896
 897         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 898             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
 899             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP))
 900                 current->mm->dumpable = suid_dumpable;
 901
 902         /* An exec changes our domain. We are no longer part of the thread
 903            group */
 904
 905         current->self_exec_id++;
 906
 907         flush_signal_handlers(current, 0);
 908         flush_old_files(current->files);
 909
 910         return 0;
 911
 912 mmap_failed:
 913         put_files_struct(current->files);
 914         current->files = files;
 915 out:
 916         return retval;
 917 }
 918
 919 EXPORT_SYMBOL(flush_old_exec);
 920
 921 /*
 922  * Fill the binprm structure from the inode.
 923  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 924  */
 925 int prepare_binprm(struct linux_binprm *bprm)
 926 {
 927         int mode;
 928         struct inode * inode = bprm->file->f_dentry->d_inode;
 929         int retval;
 930
 931         mode = inode->i_mode;
 932         /*
 933          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 934          * vfs_permission lets a non-executable through
 935          */
 936         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 937                 return -EACCES;
 938         if (bprm->file->f_op == NULL)
 939                 return -EACCES;
 940
 941         bprm->e_uid = current->euid;
 942         bprm->e_gid = current->egid;
 943
 944         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 945                 /* Set-uid? */
 946                 if (mode & S_ISUID) {
 947                         current->personality &= ~PER_CLEAR_ON_SETID;
 948                         bprm->e_uid = inode->i_uid;
 949                 }
 950
 951                 /* Set-gid? */
 952                 /*
 953                  * If setgid is set but no group execute bit then this
 954                  * is a candidate for mandatory locking, not a setgid
 955                  * executable.
 956                  */
 957                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 958                         current->personality &= ~PER_CLEAR_ON_SETID;
 959                         bprm->e_gid = inode->i_gid;
 960                 }
 961         }
 962
 963         /* fill in binprm security blob */
 964         retval = security_bprm_set(bprm);
 965         if (retval)
 966                 return retval;
 967
 968         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 969         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 970 }
 971
 972 EXPORT_SYMBOL(prepare_binprm);
 973
 974 static inline int unsafe_exec(struct task_struct *p)
 975 {
 976         int unsafe = 0;
 977         if (p->ptrace & PT_PTRACED) {
 978                 if (p->ptrace & PT_PTRACE_CAP)
 979                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 980                 else
 981                         unsafe |= LSM_UNSAFE_PTRACE;
 982         }
 983         if (atomic_read(&p->fs->count) > 1 ||
 984             atomic_read(&p->files->count) > 1 ||
 985             atomic_read(&p->sighand->count) > 1)
 986                 unsafe |= LSM_UNSAFE_SHARE;
 987
 988         return unsafe;
 989 }
 990
 991 void compute_creds(struct linux_binprm *bprm)
 992 {
 993         int unsafe;
 994         task_lock(current);
 995         unsafe = unsafe_exec(current);
 996         security_bprm_apply_creds(bprm, unsafe);
 997         task_unlock(current);
 998 }
 999
1000 EXPORT_SYMBOL(compute_creds);
1001
1002 void remove_arg_zero(struct linux_binprm *bprm)
1003 {
1004         if (bprm->argc) {
1005                 unsigned long offset;
1006                 char * kaddr;
1007                 struct page *page;
1008
1009                 offset = bprm->p % PAGE_SIZE;
1010                 goto inside;
1011
1012                 while (bprm->p++, *(kaddr+offset++)) {
1013                         if (offset != PAGE_SIZE)
1014                                 continue;
1015                         offset = 0;
1016                         kunmap_atomic(kaddr, KM_USER0);
1017 inside:
1018                         page = bprm->page[bprm->p/PAGE_SIZE];
1019                         kaddr = kmap_atomic(page, KM_USER0);
1020                 }
1021                 kunmap_atomic(kaddr, KM_USER0);
1022                 bprm->argc--;
1023         }
1024 }
1025
1026 EXPORT_SYMBOL(remove_arg_zero);
1027
1028 /*
1029  * cycle the list of binary formats handler, until one recognizes the image
1030  */
1031 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1032 {
1033         int try,retval;
1034         struct linux_binfmt *fmt;
1035 #ifdef __alpha__
1036         /* handle /sbin/loader.. */
1037         {
1038             struct exec * eh = (struct exec *) bprm->buf;
1039
1040             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1041                 (eh->fh.f_flags & 0x3000) == 0x3000)
1042             {
1043                 struct file * file;
1044                 unsigned long loader;
1045
1046                 allow_write_access(bprm->file);
1047                 fput(bprm->file);
1048                 bprm->file = NULL;
1049
1050                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1051
1052                 file = open_exec("/sbin/loader");
1053                 retval = PTR_ERR(file);
1054                 if (IS_ERR(file))
1055                         return retval;
1056
1057                 /* Remember if the application is TASO.  */
1058                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1059
1060                 bprm->file = file;
1061                 bprm->loader = loader;
1062                 retval = prepare_binprm(bprm);
1063                 if (retval<0)
1064                         return retval;
1065                 /* should call search_binary_handler recursively here,
1066                    but it does not matter */
1067             }
1068         }
1069 #endif
1070         retval = security_bprm_check(bprm);
1071         if (retval)
1072                 return retval;
1073
1074         /* kernel module loader fixup */
1075         /* so we don't try to load run modprobe in kernel space. */
1076         set_fs(USER_DS);
1077         retval = -ENOENT;
1078         for (try=0; try<2; try++) {
1079                 read_lock(&binfmt_lock);
1080                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1081                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1082                         if (!fn)
1083                                 continue;
1084                         if (!try_module_get(fmt->module))
1085                                 continue;
1086                         read_unlock(&binfmt_lock);
1087                         retval = fn(bprm, regs);
1088                         if (retval >= 0) {
1089                                 put_binfmt(fmt);
1090                                 allow_write_access(bprm->file);
1091                                 if (bprm->file)
1092                                         fput(bprm->file);
1093                                 bprm->file = NULL;
1094                                 current->did_exec = 1;
1095                                 ckrm_cb_exec(bprm->filename);
1096                                 return retval;
1097                         }
1098                         read_lock(&binfmt_lock);
1099                         put_binfmt(fmt);
1100                         if (retval != -ENOEXEC || bprm->mm == NULL)
1101                                 break;
1102                         if (!bprm->file) {
1103                                 read_unlock(&binfmt_lock);
1104                                 return retval;
1105                         }
1106                 }
1107                 read_unlock(&binfmt_lock);
1108                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1109                         break;
1110 #ifdef CONFIG_KMOD
1111                 }else{
1112 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1113                         if (printable(bprm->buf[0]) &&
1114                             printable(bprm->buf[1]) &&
1115                             printable(bprm->buf[2]) &&
1116                             printable(bprm->buf[3]))
1117                                 break; /* -ENOEXEC */
1118                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1119 #endif
1120                 }
1121         }
1122         return retval;
1123 }
1124
1125 EXPORT_SYMBOL(search_binary_handler);
1126
1127 /*
1128  * sys_execve() executes a new program.
1129  */
1130 int do_execve(char * filename,
1131         char __user *__user *argv,
1132         char __user *__user *envp,
1133         struct pt_regs * regs)
1134 {
1135         struct linux_binprm *bprm;
1136         struct file *file;
1137         int retval;
1138         int i;
1139
1140         file = open_exec(filename);
1141
1142         retval = PTR_ERR(file);
1143         if (IS_ERR(file))
1144                 return retval;
1145
1146         sched_exec();
1147
1148         retval = -ENOMEM;
1149         bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1150         if (!bprm)
1151                 goto out_ret;
1152         memset(bprm, 0, sizeof(*bprm));
1153
1154         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1155
1156         bprm->file = file;
1157         bprm->filename = filename;
1158         bprm->interp = filename;
1159         bprm->mm = mm_alloc();
1160         if (!bprm->mm)
1161                 goto out_file;
1162
1163         retval = init_new_context(current, bprm->mm);
1164         if (retval < 0)
1165                 goto out_mm;
1166
1167         bprm->argc = count(argv, bprm->p / sizeof(void *));
1168         if ((retval = bprm->argc) < 0)
1169                 goto out_mm;
1170
1171         bprm->envc = count(envp, bprm->p / sizeof(void *));
1172         if ((retval = bprm->envc) < 0)
1173                 goto out_mm;
1174
1175         retval = security_bprm_alloc(bprm);
1176         if (retval)
1177                 goto out;
1178
1179         retval = prepare_binprm(bprm);
1180         if (retval < 0)
1181                 goto out;
1182
1183         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1184         if (retval < 0)
1185                 goto out;
1186
1187         bprm->exec = bprm->p;
1188         retval = copy_strings(bprm->envc, envp, bprm);
1189         if (retval < 0)
1190                 goto out;
1191
1192         retval = copy_strings(bprm->argc, argv, bprm);
1193         if (retval < 0)
1194                 goto out;
1195
1196         retval = search_binary_handler(bprm,regs);
1197         if (retval >= 0) {
1198                 free_arg_pages(bprm);
1199
1200                 /* execve success */
1201                 security_bprm_free(bprm);
1202                 kfree(bprm);
1203                 return retval;
1204         }
1205
1206 out:
1207         /* Something went wrong, return the inode and free the argument pages*/
1208         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1209                 struct page * page = bprm->page[i];
1210                 if (page)
1211                         __free_page(page);
1212         }
1213
1214         if (bprm->security)
1215                 security_bprm_free(bprm);
1216
1217 out_mm:
1218         if (bprm->mm)
1219                 mmdrop(bprm->mm);
1220
1221 out_file:
1222         if (bprm->file) {
1223                 allow_write_access(bprm->file);
1224                 fput(bprm->file);
1225         }
1226         kfree(bprm);
1227
1228 out_ret:
1229         return retval;
1230 }
1231
1232 int set_binfmt(struct linux_binfmt *new)
1233 {
1234         struct linux_binfmt *old = current->binfmt;
1235
1236         if (new) {
1237                 if (!try_module_get(new->module))
1238                         return -1;
1239         }
1240         current->binfmt = new;
1241         if (old)
1242                 module_put(old->module);
1243         return 0;
1244 }
1245
1246 EXPORT_SYMBOL(set_binfmt);
1247
1248 #define CORENAME_MAX_SIZE 64
1249
1250 /* format_corename will inspect the pattern parameter, and output a
1251  * name into corename, which must have space for at least
1252  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1253  */
1254 static void format_corename(char *corename, const char *pattern, long signr)
1255 {
1256         const char *pat_ptr = pattern;
1257         char *out_ptr = corename;
1258         char *const out_end = corename + CORENAME_MAX_SIZE;
1259         int rc;
1260         int pid_in_pattern = 0;
1261
1262         /* Repeat as long as we have more pattern to process and more output
1263            space */
1264         while (*pat_ptr) {
1265                 if (*pat_ptr != '%') {
1266                         if (out_ptr == out_end)
1267                                 goto out;
1268                         *out_ptr++ = *pat_ptr++;
1269                 } else {
1270                         switch (*++pat_ptr) {
1271                         case 0:
1272                                 goto out;
1273                         /* Double percent, output one percent */
1274                         case '%':
1275                                 if (out_ptr == out_end)
1276                                         goto out;
1277                                 *out_ptr++ = '%';
1278                                 break;
1279                         /* pid */
1280                         case 'p':
1281                                 pid_in_pattern = 1;
1282                                 rc = snprintf(out_ptr, out_end - out_ptr,
1283                                               "%d", current->tgid);
1284                                 if (rc > out_end - out_ptr)
1285                                         goto out;
1286                                 out_ptr += rc;
1287                                 break;
1288                         /* uid */
1289                         case 'u':
1290                                 rc = snprintf(out_ptr, out_end - out_ptr,
1291                                               "%d", current->uid);
1292                                 if (rc > out_end - out_ptr)
1293                                         goto out;
1294                                 out_ptr += rc;
1295                                 break;
1296                         /* gid */
1297                         case 'g':
1298                                 rc = snprintf(out_ptr, out_end - out_ptr,
1299                                               "%d", current->gid);
1300                                 if (rc > out_end - out_ptr)
1301                                         goto out;
1302                                 out_ptr += rc;
1303                                 break;
1304                         /* signal that caused the coredump */
1305                         case 's':
1306                                 rc = snprintf(out_ptr, out_end - out_ptr,
1307                                               "%ld", signr);
1308                                 if (rc > out_end - out_ptr)
1309                                         goto out;
1310                                 out_ptr += rc;
1311                                 break;
1312                         /* UNIX time of coredump */
1313                         case 't': {
1314                                 struct timeval tv;
1315                                 do_gettimeofday(&tv);
1316                                 rc = snprintf(out_ptr, out_end - out_ptr,
1317                                               "%lu", tv.tv_sec);
1318                                 if (rc > out_end - out_ptr)
1319                                         goto out;
1320                                 out_ptr += rc;
1321                                 break;
1322                         }
1323                         /* hostname */
1324                         case 'h':
1325                                 down_read(&uts_sem);
1326                                 rc = snprintf(out_ptr, out_end - out_ptr,
1327                                               "%s", system_utsname.nodename);
1328                                 up_read(&uts_sem);
1329                                 if (rc > out_end - out_ptr)
1330                                         goto out;
1331                                 out_ptr += rc;
1332                                 break;
1333                         /* executable */
1334                         case 'e':
1335                                 rc = snprintf(out_ptr, out_end - out_ptr,
1336                                               "%s", current->comm);
1337                                 if (rc > out_end - out_ptr)
1338                                         goto out;
1339                                 out_ptr += rc;
1340                                 break;
1341                         default:
1342                                 break;
1343                         }
1344                         ++pat_ptr;
1345                 }
1346         }
1347         /* Backward compatibility with core_uses_pid:
1348          *
1349          * If core_pattern does not include a %p (as is the default)
1350          * and core_uses_pid is set, then .%pid will be appended to
1351          * the filename */
1352         if (!pid_in_pattern
1353             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1354                 rc = snprintf(out_ptr, out_end - out_ptr,
1355                               ".%d", current->tgid);
1356                 if (rc > out_end - out_ptr)
1357                         goto out;
1358                 out_ptr += rc;
1359         }
1360       out:
1361         *out_ptr = 0;
1362 }
1363
1364 static void zap_threads (struct mm_struct *mm)
1365 {
1366         struct task_struct *g, *p;
1367         struct task_struct *tsk = current;
1368         struct completion *vfork_done = tsk->vfork_done;
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                 }
1385         while_each_thread(g,p);
1386
1387         read_unlock(&tasklist_lock);
1388 }
1389
1390 static void coredump_wait(struct mm_struct *mm)
1391 {
1392         DECLARE_COMPLETION(startup_done);
1393
1394         mm->core_waiters++; /* let other threads block */
1395         mm->core_startup_done = &startup_done;
1396
1397         /* give other threads a chance to run: */
1398         yield();
1399
1400         zap_threads(mm);
1401         if (--mm->core_waiters) {
1402                 up_write(&mm->mmap_sem);
1403                 wait_for_completion(&startup_done);
1404         } else
1405                 up_write(&mm->mmap_sem);
1406         BUG_ON(mm->core_waiters);
1407 }
1408
1409 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1410 {
1411         char corename[CORENAME_MAX_SIZE + 1];
1412         struct mm_struct *mm = current->mm;
1413         struct linux_binfmt * binfmt;
1414         struct inode * inode;
1415         struct file * file;
1416         int retval = 0;
1417         int fsuid = current->fsuid;
1418         int flag = 0;
1419
1420         binfmt = current->binfmt;
1421         if (!binfmt || !binfmt->core_dump)
1422                 goto fail;
1423         if (current->tux_exit)
1424                 current->tux_exit();
1425         down_write(&mm->mmap_sem);
1426         if (!mm->dumpable) {
1427                 up_write(&mm->mmap_sem);
1428                 goto fail;
1429         }
1430
1431         /*
1432          *      We cannot trust fsuid as being the "true" uid of the
1433          *      process nor do we know its entire history. We only know it
1434          *      was tainted so we dump it as root in mode 2.
1435          */
1436         if (mm->dumpable == 2)          /* Setuid core dump mode */
1437         {
1438                 flag = O_EXCL;          /* Stop rewrite attacks */
1439                 current->fsuid = 0;     /* Dump root private */
1440         }
1441         mm->dumpable = 0;
1442         init_completion(&mm->core_done);
1443         current->signal->group_exit = 1;
1444         current->signal->group_exit_code = exit_code;
1445         coredump_wait(mm);
1446
1447         if (current->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1448                 goto fail_unlock;
1449
1450         /*
1451          * lock_kernel() because format_corename() is controlled by sysctl, which
1452          * uses lock_kernel()
1453          */
1454         lock_kernel();
1455         format_corename(corename, core_pattern, signr);
1456         unlock_kernel();
1457         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1458         if (IS_ERR(file))
1459                 goto fail_unlock;
1460         inode = file->f_dentry->d_inode;
1461         if (inode->i_nlink > 1)
1462                 goto close_fail;        /* multiple links - don't dump */
1463         if (d_unhashed(file->f_dentry))
1464                 goto close_fail;
1465
1466         if (!S_ISREG(inode->i_mode))
1467                 goto close_fail;
1468         if (!file->f_op)
1469                 goto close_fail;
1470         if (!file->f_op->write)
1471                 goto close_fail;
1472         if (do_truncate(file->f_dentry, 0) != 0)
1473                 goto close_fail;
1474
1475         retval = binfmt->core_dump(signr, regs, file);
1476
1477         current->signal->group_exit_code |= 0x80;
1478 close_fail:
1479         filp_close(file, NULL);
1480 fail_unlock:
1481         current->fsuid = fsuid;
1482         complete_all(&mm->core_done);
1483 fail:
1484         return retval;
1485 }