fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/config.h>
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/mman.h>
  29 #include <linux/a.out.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/key.h>
  38 #include <linux/personality.h>
  39 #include <linux/binfmts.h>
  40 #include <linux/swap.h>
  41 #include <linux/utsname.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/vs_memory.h>
  51
  52 #include <asm/uaccess.h>
  53 #include <asm/mmu_context.h>
  54
  55 #ifdef CONFIG_KMOD
  56 #include <linux/kmod.h>
  57 #endif
  58
  59 int core_uses_pid;
  60 char core_pattern[65] = "core";
  61 int suid_dumpable = 0;
  62
  63 EXPORT_SYMBOL(suid_dumpable);
  64 /* The maximal length of core_pattern is also specified in sysctl.c */
  65
  66 static struct linux_binfmt *formats;
  67 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  68
  69 int register_binfmt(struct linux_binfmt * fmt)
  70 {
  71         struct linux_binfmt ** tmp = &formats;
  72
  73         if (!fmt)
  74                 return -EINVAL;
  75         if (fmt->next)
  76                 return -EBUSY;
  77         write_lock(&binfmt_lock);
  78         while (*tmp) {
  79                 if (fmt == *tmp) {
  80                         write_unlock(&binfmt_lock);
  81                         return -EBUSY;
  82                 }
  83                 tmp = &(*tmp)->next;
  84         }
  85         fmt->next = formats;
  86         formats = fmt;
  87         write_unlock(&binfmt_lock);
  88         return 0;
  89 }
  90
  91 EXPORT_SYMBOL(register_binfmt);
  92
  93 int unregister_binfmt(struct linux_binfmt * fmt)
  94 {
  95         struct linux_binfmt ** tmp = &formats;
  96
  97         write_lock(&binfmt_lock);
  98         while (*tmp) {
  99                 if (fmt == *tmp) {
 100                         *tmp = fmt->next;
 101                         write_unlock(&binfmt_lock);
 102                         return 0;
 103                 }
 104                 tmp = &(*tmp)->next;
 105         }
 106         write_unlock(&binfmt_lock);
 107         return -EINVAL;
 108 }
 109
 110 EXPORT_SYMBOL(unregister_binfmt);
 111
 112 static inline void put_binfmt(struct linux_binfmt * fmt)
 113 {
 114         module_put(fmt->module);
 115 }
 116
 117 /*
 118  * Note that a shared library must be both readable and executable due to
 119  * security reasons.
 120  *
 121  * Also note that we take the address to load from from the file itself.
 122  */
 123 asmlinkage long sys_uselib(const char __user * library)
 124 {
 125         struct file * file;
 126         struct nameidata nd;
 127         int error;
 128
 129         nd.intent.open.flags = FMODE_READ;
 130         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 131         if (error)
 132                 goto out;
 133
 134         error = -EINVAL;
 135         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 136                 goto exit;
 137
 138         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 139         if (error)
 140                 goto exit;
 141
 142         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 143         error = PTR_ERR(file);
 144         if (IS_ERR(file))
 145                 goto out;
 146
 147         error = -ENOEXEC;
 148         if(file->f_op) {
 149                 struct linux_binfmt * fmt;
 150
 151                 read_lock(&binfmt_lock);
 152                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 153                         if (!fmt->load_shlib)
 154                                 continue;
 155                         if (!try_module_get(fmt->module))
 156                                 continue;
 157                         read_unlock(&binfmt_lock);
 158                         error = fmt->load_shlib(file);
 159                         read_lock(&binfmt_lock);
 160                         put_binfmt(fmt);
 161                         if (error != -ENOEXEC)
 162                                 break;
 163                 }
 164                 read_unlock(&binfmt_lock);
 165         }
 166         fput(file);
 167 out:
 168         return error;
 169 exit:
 170         path_release(&nd);
 171         goto out;
 172 }
 173
 174 /*
 175  * count() counts the number of strings in array ARGV.
 176  */
 177 static int count(char __user * __user * argv, int max)
 178 {
 179         int i = 0;
 180
 181         if (argv != NULL) {
 182                 for (;;) {
 183                         char __user * p;
 184
 185                         if (get_user(p, argv))
 186                                 return -EFAULT;
 187                         if (!p)
 188                                 break;
 189                         argv++;
 190                         if(++i > max)
 191                                 return -E2BIG;
 192                 }
 193         }
 194         return i;
 195 }
 196
 197 /*
 198  * 'copy_strings()' copies argument/environment strings from user
 199  * memory to free pages in kernel mem. These are in a format ready
 200  * to be put directly into the top of new user memory.
 201  */
 202 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 203 {
 204         struct page *kmapped_page = NULL;
 205         char *kaddr = NULL;
 206         int ret;
 207
 208         while (argc-- > 0) {
 209                 char __user *str;
 210                 int len;
 211                 unsigned long pos;
 212
 213                 if (get_user(str, argv+argc) ||
 214                                 !(len = strnlen_user(str, bprm->p))) {
 215                         ret = -EFAULT;
 216                         goto out;
 217                 }
 218
 219                 if (bprm->p < len)  {
 220                         ret = -E2BIG;
 221                         goto out;
 222                 }
 223
 224                 bprm->p -= len;
 225                 /* XXX: add architecture specific overflow check here. */
 226                 pos = bprm->p;
 227
 228                 while (len > 0) {
 229                         int i, new, err;
 230                         int offset, bytes_to_copy;
 231                         struct page *page;
 232
 233                         offset = pos % PAGE_SIZE;
 234                         i = pos/PAGE_SIZE;
 235                         page = bprm->page[i];
 236                         new = 0;
 237                         if (!page) {
 238                                 page = alloc_page(GFP_HIGHUSER);
 239                                 bprm->page[i] = page;
 240                                 if (!page) {
 241                                         ret = -ENOMEM;
 242                                         goto out;
 243                                 }
 244                                 new = 1;
 245                         }
 246
 247                         if (page != kmapped_page) {
 248                                 if (kmapped_page)
 249                                         kunmap(kmapped_page);
 250                                 kmapped_page = page;
 251                                 kaddr = kmap(kmapped_page);
 252                         }
 253                         if (new && offset)
 254                                 memset(kaddr, 0, offset);
 255                         bytes_to_copy = PAGE_SIZE - offset;
 256                         if (bytes_to_copy > len) {
 257                                 bytes_to_copy = len;
 258                                 if (new)
 259                                         memset(kaddr+offset+len, 0,
 260                                                 PAGE_SIZE-offset-len);
 261                         }
 262                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 263                         if (err) {
 264                                 ret = -EFAULT;
 265                                 goto out;
 266                         }
 267
 268                         pos += bytes_to_copy;
 269                         str += bytes_to_copy;
 270                         len -= bytes_to_copy;
 271                 }
 272         }
 273         ret = 0;
 274 out:
 275         if (kmapped_page)
 276                 kunmap(kmapped_page);
 277         return ret;
 278 }
 279
 280 /*
 281  * Like copy_strings, but get argv and its values from kernel memory.
 282  */
 283 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 284 {
 285         int r;
 286         mm_segment_t oldfs = get_fs();
 287         set_fs(KERNEL_DS);
 288         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 289         set_fs(oldfs);
 290         return r;
 291 }
 292
 293 EXPORT_SYMBOL(copy_strings_kernel);
 294
 295 #ifdef CONFIG_MMU
 296 /*
 297  * This routine is used to map in a page into an address space: needed by
 298  * execve() for the initial stack and environment pages.
 299  *
 300  * vma->vm_mm->mmap_sem is held for writing.
 301  */
 302 void install_arg_page(struct vm_area_struct *vma,
 303                         struct page *page, unsigned long address)
 304 {
 305         struct mm_struct *mm = vma->vm_mm;
 306         pgd_t * pgd;
 307         pmd_t * pmd;
 308         pte_t * pte;
 309
 310         if (unlikely(anon_vma_prepare(vma)))
 311                 goto out_sig;
 312
 313         flush_dcache_page(page);
 314         pgd = pgd_offset(mm, address);
 315
 316         spin_lock(&mm->page_table_lock);
 317         pmd = pmd_alloc(mm, pgd, address);
 318         if (!pmd)
 319                 goto out;
 320         pte = pte_alloc_map(mm, pmd, address);
 321         if (!pte)
 322                 goto out;
 323         if (!pte_none(*pte)) {
 324                 pte_unmap(pte);
 325                 goto out;
 326         }
 327         // mm->rss++;
 328         vx_rsspages_inc(mm);
 329         lru_cache_add_active(page);
 330         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
 331                                         page, vma->vm_page_prot))));
 332         page_add_anon_rmap(page, vma, address);
 333         pte_unmap(pte);
 334         spin_unlock(&mm->page_table_lock);
 335
 336         /* no need for flush_tlb */
 337         return;
 338 out:
 339         spin_unlock(&mm->page_table_lock);
 340 out_sig:
 341         __free_page(page);
 342         force_sig(SIGKILL, current);
 343 }
 344
 345 #define EXTRA_STACK_VM_PAGES    20      /* random */
 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         /* Limit stack size to 1GB */
 385         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 386         if (stack_base > (1 << 30))
 387                 stack_base = 1 << 30;
 388         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
 389
 390         /* Adjust bprm->p to point to the end of the strings. */
 391         bprm->p = stack_base + PAGE_SIZE * i - offset;
 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         bprm->p += stack_base;
 407         mm->arg_start = bprm->p;
 408         arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
 409 #endif
 410
 411         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 412
 413         if (bprm->loader)
 414                 bprm->loader += stack_base;
 415         bprm->exec += stack_base;
 416
 417         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 418         if (!mpnt)
 419                 return -ENOMEM;
 420
 421         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT) ||
 422                 !vx_vmpages_avail(mm, arg_size >> PAGE_SHIFT)) {
 423                 kmem_cache_free(vm_area_cachep, mpnt);
 424                 return -ENOMEM;
 425         }
 426
 427         memset(mpnt, 0, sizeof(*mpnt));
 428
 429         down_write(&mm->mmap_sem);
 430         {
 431                 mpnt->vm_mm = mm;
 432 #ifdef CONFIG_STACK_GROWSUP
 433                 mpnt->vm_start = stack_base;
 434                 mpnt->vm_end = stack_base + arg_size;
 435 #else
 436                 mpnt->vm_end = STACK_TOP;
 437                 mpnt->vm_start = mpnt->vm_end - arg_size;
 438 #endif
 439                 /* Adjust stack execute permissions; explicitly enable
 440                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 441                  * and leave alone (arch default) otherwise. */
 442                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 443                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 444                 else if (executable_stack == EXSTACK_DISABLE_X)
 445                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 446                 else
 447                         mpnt->vm_flags = VM_STACK_FLAGS;
 448                 mpnt->vm_flags |= mm->def_flags;
 449                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 450                 if ((ret = insert_vm_struct(mm, mpnt))) {
 451                         up_write(&mm->mmap_sem);
 452                         kmem_cache_free(vm_area_cachep, mpnt);
 453                         return ret;
 454                 }
 455                 // mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 456                 vx_vmpages_sub(mm, mm->total_vm - vma_pages(mpnt));
 457                 mm->stack_vm = mm->total_vm;
 458         }
 459
 460         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 461                 struct page *page = bprm->page[i];
 462                 if (page) {
 463                         bprm->page[i] = NULL;
 464                         install_arg_page(mpnt, page, stack_base);
 465                 }
 466                 stack_base += PAGE_SIZE;
 467         }
 468         up_write(&mm->mmap_sem);
 469
 470         return 0;
 471 }
 472
 473 EXPORT_SYMBOL(setup_arg_pages);
 474
 475 #define free_arg_pages(bprm) do { } while (0)
 476
 477 #else
 478
 479 static inline void free_arg_pages(struct linux_binprm *bprm)
 480 {
 481         int i;
 482
 483         for (i = 0; i < MAX_ARG_PAGES; i++) {
 484                 if (bprm->page[i])
 485                         __free_page(bprm->page[i]);
 486                 bprm->page[i] = NULL;
 487         }
 488 }
 489
 490 #endif /* CONFIG_MMU */
 491
 492 struct file *open_exec(const char *name)
 493 {
 494         struct nameidata nd;
 495         int err;
 496         struct file *file;
 497
 498         nd.intent.open.flags = FMODE_READ;
 499         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 500         file = ERR_PTR(err);
 501
 502         if (!err) {
 503                 struct inode *inode = nd.dentry->d_inode;
 504                 file = ERR_PTR(-EACCES);
 505                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 506                     S_ISREG(inode->i_mode)) {
 507                         int err = permission(inode, MAY_EXEC, &nd);
 508                         if (!err && !(inode->i_mode & 0111))
 509                                 err = -EACCES;
 510                         file = ERR_PTR(err);
 511                         if (!err) {
 512                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 513                                 if (!IS_ERR(file)) {
 514                                         err = deny_write_access(file);
 515                                         if (err) {
 516                                                 fput(file);
 517                                                 file = ERR_PTR(err);
 518                                         }
 519                                 }
 520 out:
 521                                 return file;
 522                         }
 523                 }
 524                 path_release(&nd);
 525         }
 526         goto out;
 527 }
 528
 529 EXPORT_SYMBOL(open_exec);
 530
 531 int kernel_read(struct file *file, unsigned long offset,
 532         char *addr, unsigned long count)
 533 {
 534         mm_segment_t old_fs;
 535         loff_t pos = offset;
 536         int result;
 537
 538         old_fs = get_fs();
 539         set_fs(get_ds());
 540         /* The cast to a user pointer is valid due to the set_fs() */
 541         result = vfs_read(file, (void __user *)addr, count, &pos);
 542         set_fs(old_fs);
 543         return result;
 544 }
 545
 546 EXPORT_SYMBOL(kernel_read);
 547
 548 static int exec_mmap(struct mm_struct *mm)
 549 {
 550         struct task_struct *tsk;
 551         struct mm_struct * old_mm, *active_mm;
 552
 553         /* Notify parent that we're no longer interested in the old VM */
 554         tsk = current;
 555         old_mm = current->mm;
 556         mm_release(tsk, old_mm);
 557
 558         task_lock(tsk);
 559         active_mm = tsk->active_mm;
 560         tsk->mm = mm;
 561         tsk->active_mm = mm;
 562         activate_mm(active_mm, mm);
 563         task_unlock(tsk);
 564         arch_pick_mmap_layout(mm);
 565         if (old_mm) {
 566                 if (active_mm != old_mm) BUG();
 567                 mmput(old_mm);
 568                 return 0;
 569         }
 570         mmdrop(active_mm);
 571         return 0;
 572 }
 573
 574 /*
 575  * This function makes sure the current process has its own signal table,
 576  * so that flush_signal_handlers can later reset the handlers without
 577  * disturbing other processes.  (Other processes might share the signal
 578  * table via the CLONE_SIGHAND option to clone().)
 579  */
 580 static inline int de_thread(struct task_struct *tsk)
 581 {
 582         struct signal_struct *sig = tsk->signal;
 583         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 584         spinlock_t *lock = &oldsighand->siglock;
 585         int count;
 586
 587         /*
 588          * If we don't share sighandlers, then we aren't sharing anything
 589          * and we can just re-use it all.
 590          */
 591         if (atomic_read(&oldsighand->count) <= 1) {
 592                 BUG_ON(atomic_read(&sig->count) != 1);
 593                 exit_itimers(sig);
 594                 return 0;
 595         }
 596
 597         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 598         if (!newsighand)
 599                 return -ENOMEM;
 600
 601         if (thread_group_empty(current))
 602                 goto no_thread_group;
 603
 604         /*
 605          * Kill all other threads in the thread group.
 606          * We must hold tasklist_lock to call zap_other_threads.
 607          */
 608         read_lock(&tasklist_lock);
 609         spin_lock_irq(lock);
 610         if (sig->group_exit) {
 611                 /*
 612                  * Another group action in progress, just
 613                  * return so that the signal is processed.
 614                  */
 615                 spin_unlock_irq(lock);
 616                 read_unlock(&tasklist_lock);
 617                 kmem_cache_free(sighand_cachep, newsighand);
 618                 return -EAGAIN;
 619         }
 620         sig->group_exit = 1;
 621         zap_other_threads(current);
 622         read_unlock(&tasklist_lock);
 623
 624         /*
 625          * Account for the thread group leader hanging around:
 626          */
 627         count = 2;
 628         if (current->pid == current->tgid)
 629                 count = 1;
 630         while (atomic_read(&sig->count) > count) {
 631                 sig->group_exit_task = current;
 632                 sig->notify_count = count;
 633                 __set_current_state(TASK_UNINTERRUPTIBLE);
 634                 spin_unlock_irq(lock);
 635                 schedule();
 636                 spin_lock_irq(lock);
 637         }
 638         sig->group_exit_task = NULL;
 639         sig->notify_count = 0;
 640         spin_unlock_irq(lock);
 641
 642         /*
 643          * At this point all other threads have exited, all we have to
 644          * do is to wait for the thread group leader to become inactive,
 645          * and to assume its PID:
 646          */
 647         if (current->pid != current->tgid) {
 648                 struct task_struct *leader = current->group_leader, *parent;
 649                 struct dentry *proc_dentry1, *proc_dentry2;
 650                 unsigned long exit_state, ptrace;
 651
 652                 /*
 653                  * Wait for the thread group leader to be a zombie.
 654                  * It should already be zombie at this point, most
 655                  * of the time.
 656                  */
 657                 while (leader->exit_state != EXIT_ZOMBIE)
 658                         yield();
 659
 660                 spin_lock(&leader->proc_lock);
 661                 spin_lock(&current->proc_lock);
 662                 proc_dentry1 = proc_pid_unhash(current);
 663                 proc_dentry2 = proc_pid_unhash(leader);
 664                 write_lock_irq(&tasklist_lock);
 665
 666                 if (leader->tgid != current->tgid)
 667                         BUG();
 668                 if (current->pid == current->tgid)
 669                         BUG();
 670                 /*
 671                  * An exec() starts a new thread group with the
 672                  * TGID of the previous thread group. Rehash the
 673                  * two threads with a switched PID, and release
 674                  * the former thread group leader:
 675                  */
 676                 ptrace = leader->ptrace;
 677                 parent = leader->parent;
 678
 679                 ptrace_unlink(current);
 680                 ptrace_unlink(leader);
 681                 remove_parent(current);
 682                 remove_parent(leader);
 683
 684                 switch_exec_pids(leader, current);
 685
 686                 current->parent = current->real_parent = leader->real_parent;
 687                 leader->parent = leader->real_parent = child_reaper;
 688                 current->group_leader = current;
 689                 leader->group_leader = leader;
 690
 691                 add_parent(current, current->parent);
 692                 add_parent(leader, leader->parent);
 693                 if (ptrace) {
 694                         current->ptrace = ptrace;
 695                         __ptrace_link(current, parent);
 696                 }
 697
 698                 list_del(&current->tasks);
 699                 list_add_tail(&current->tasks, &init_task.tasks);
 700                 current->exit_signal = SIGCHLD;
 701                 exit_state = leader->exit_state;
 702
 703                 write_unlock_irq(&tasklist_lock);
 704                 spin_unlock(&leader->proc_lock);
 705                 spin_unlock(&current->proc_lock);
 706                 proc_pid_flush(proc_dentry1);
 707                 proc_pid_flush(proc_dentry2);
 708
 709                 if (exit_state != EXIT_ZOMBIE)
 710                         BUG();
 711                 release_task(leader);
 712         }
 713
 714         /*
 715          * Now there are really no other threads at all,
 716          * so it's safe to stop telling them to kill themselves.
 717          */
 718         sig->group_exit = 0;
 719
 720 no_thread_group:
 721         BUG_ON(atomic_read(&sig->count) != 1);
 722         exit_itimers(sig);
 723
 724         if (atomic_read(&oldsighand->count) == 1) {
 725                 /*
 726                  * Now that we nuked the rest of the thread group,
 727                  * it turns out we are not sharing sighand any more either.
 728                  * So we can just keep it.
 729                  */
 730                 kmem_cache_free(sighand_cachep, newsighand);
 731         } else {
 732                 /*
 733                  * Move our state over to newsighand and switch it in.
 734                  */
 735                 spin_lock_init(&newsighand->siglock);
 736                 atomic_set(&newsighand->count, 1);
 737                 memcpy(newsighand->action, oldsighand->action,
 738                        sizeof(newsighand->action));
 739                 write_lock_irq(&tasklist_lock);
 740                 spin_lock(&oldsighand->siglock);
 741                 spin_lock(&newsighand->siglock);
 742                 current->sighand = newsighand;
 743                 recalc_sigpending();
 744                 spin_unlock(&newsighand->siglock);
 745                 spin_unlock(&oldsighand->siglock);
 746                 write_unlock_irq(&tasklist_lock);
 747
 748                 if (atomic_dec_and_test(&oldsighand->count))
 749                         kmem_cache_free(sighand_cachep, oldsighand);
 750         }
 751
 752         if (!thread_group_empty(current))
 753                 BUG();
 754         if (current->tgid != current->pid)
 755                 BUG();
 756         return 0;
 757 }
 758
 759 /*
 760  * These functions flushes out all traces of the currently running executable
 761  * so that a new one can be started
 762  */
 763
 764 static inline void flush_old_files(struct files_struct * files)
 765 {
 766         long j = -1;
 767
 768         spin_lock(&files->file_lock);
 769         for (;;) {
 770                 unsigned long set, i;
 771
 772                 j++;
 773                 i = j * __NFDBITS;
 774                 if (i >= files->max_fds || i >= files->max_fdset)
 775                         break;
 776                 set = files->close_on_exec->fds_bits[j];
 777                 if (!set)
 778                         continue;
 779                 files->close_on_exec->fds_bits[j] = 0;
 780                 spin_unlock(&files->file_lock);
 781                 for ( ; set ; i++,set >>= 1) {
 782                         if (set & 1) {
 783                                 sys_close(i);
 784                         }
 785                 }
 786                 spin_lock(&files->file_lock);
 787
 788         }
 789         spin_unlock(&files->file_lock);
 790 }
 791
 792 void get_task_comm(char *buf, struct task_struct *tsk)
 793 {
 794         /* buf must be at least sizeof(tsk->comm) in size */
 795         task_lock(tsk);
 796         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 797         task_unlock(tsk);
 798 }
 799
 800 void set_task_comm(struct task_struct *tsk, char *buf)
 801 {
 802         task_lock(tsk);
 803         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 804         task_unlock(tsk);
 805 }
 806
 807 int flush_old_exec(struct linux_binprm * bprm)
 808 {
 809         char * name;
 810         int i, ch, retval;
 811         struct files_struct *files;
 812         char tcomm[sizeof(current->comm)];
 813
 814         /*
 815          * Make sure we have a private signal table and that
 816          * we are unassociated from the previous thread group.
 817          */
 818         retval = de_thread(current);
 819         if (retval)
 820                 goto out;
 821
 822         /*
 823          * Make sure we have private file handles. Ask the
 824          * fork helper to do the work for us and the exit
 825          * helper to do the cleanup of the old one.
 826          */
 827         files = current->files;         /* refcounted so safe to hold */
 828         retval = unshare_files();
 829         if (retval)
 830                 goto out;
 831         /*
 832          * Release all of the old mmap stuff
 833          */
 834         retval = exec_mmap(bprm->mm);
 835         if (retval)
 836                 goto mmap_failed;
 837
 838         bprm->mm = NULL;                /* We're using it now */
 839
 840         /* This is the point of no return */
 841         steal_locks(files);
 842         put_files_struct(files);
 843
 844         current->sas_ss_sp = current->sas_ss_size = 0;
 845
 846         if (current->euid == current->uid && current->egid == current->gid)
 847                 current->mm->dumpable = 1;
 848         else
 849                 current->mm->dumpable = suid_dumpable;
 850
 851         name = bprm->filename;
 852         for (i=0; (ch = *(name++)) != '\0';) {
 853                 if (ch == '/')
 854                         i = 0;
 855                 else
 856                         if (i < (sizeof(tcomm) - 1))
 857                                 tcomm[i++] = ch;
 858         }
 859         tcomm[i] = '\0';
 860         set_task_comm(current, tcomm);
 861
 862         current->flags &= ~PF_RELOCEXEC;
 863         flush_thread();
 864
 865         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 866             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
 867             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 868                 suid_keys(current);
 869                 current->mm->dumpable = suid_dumpable;
 870         }
 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          * generic_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
 965         if (bprm->e_uid != current->uid)
 966                 suid_keys(current);
 967         exec_keys(current);
 968
 969         task_lock(current);
 970         unsafe = unsafe_exec(current);
 971         security_bprm_apply_creds(bprm, unsafe);
 972         task_unlock(current);
 973 }
 974
 975 EXPORT_SYMBOL(compute_creds);
 976
 977 void remove_arg_zero(struct linux_binprm *bprm)
 978 {
 979         if (bprm->argc) {
 980                 unsigned long offset;
 981                 char * kaddr;
 982                 struct page *page;
 983
 984                 offset = bprm->p % PAGE_SIZE;
 985                 goto inside;
 986
 987                 while (bprm->p++, *(kaddr+offset++)) {
 988                         if (offset != PAGE_SIZE)
 989                                 continue;
 990                         offset = 0;
 991                         kunmap_atomic(kaddr, KM_USER0);
 992 inside:
 993                         page = bprm->page[bprm->p/PAGE_SIZE];
 994                         kaddr = kmap_atomic(page, KM_USER0);
 995                 }
 996                 kunmap_atomic(kaddr, KM_USER0);
 997                 bprm->argc--;
 998         }
 999 }
1000
1001 EXPORT_SYMBOL(remove_arg_zero);
1002
1003 /*
1004  * cycle the list of binary formats handler, until one recognizes the image
1005  */
1006 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1007 {
1008         int try,retval;
1009         struct linux_binfmt *fmt;
1010 #ifdef __alpha__
1011         /* handle /sbin/loader.. */
1012         {
1013             struct exec * eh = (struct exec *) bprm->buf;
1014
1015             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1016                 (eh->fh.f_flags & 0x3000) == 0x3000)
1017             {
1018                 struct file * file;
1019                 unsigned long loader;
1020
1021                 allow_write_access(bprm->file);
1022                 fput(bprm->file);
1023                 bprm->file = NULL;
1024
1025                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1026
1027                 file = open_exec("/sbin/loader");
1028                 retval = PTR_ERR(file);
1029                 if (IS_ERR(file))
1030                         return retval;
1031
1032                 /* Remember if the application is TASO.  */
1033                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1034
1035                 bprm->file = file;
1036                 bprm->loader = loader;
1037                 retval = prepare_binprm(bprm);
1038                 if (retval<0)
1039                         return retval;
1040                 /* should call search_binary_handler recursively here,
1041                    but it does not matter */
1042             }
1043         }
1044 #endif
1045         retval = security_bprm_check(bprm);
1046         if (retval)
1047                 return retval;
1048
1049         /* kernel module loader fixup */
1050         /* so we don't try to load run modprobe in kernel space. */
1051         set_fs(USER_DS);
1052         retval = -ENOENT;
1053         for (try=0; try<2; try++) {
1054                 read_lock(&binfmt_lock);
1055                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1056                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1057                         if (!fn)
1058                                 continue;
1059                         if (!try_module_get(fmt->module))
1060                                 continue;
1061                         read_unlock(&binfmt_lock);
1062                         retval = fn(bprm, regs);
1063                         if (retval >= 0) {
1064                                 put_binfmt(fmt);
1065                                 allow_write_access(bprm->file);
1066                                 if (bprm->file)
1067                                         fput(bprm->file);
1068                                 bprm->file = NULL;
1069                                 current->did_exec = 1;
1070                                 return retval;
1071                         }
1072                         read_lock(&binfmt_lock);
1073                         put_binfmt(fmt);
1074                         if (retval != -ENOEXEC || bprm->mm == NULL)
1075                                 break;
1076                         if (!bprm->file) {
1077                                 read_unlock(&binfmt_lock);
1078                                 return retval;
1079                         }
1080                 }
1081                 read_unlock(&binfmt_lock);
1082                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1083                         break;
1084 #ifdef CONFIG_KMOD
1085                 }else{
1086 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1087                         if (printable(bprm->buf[0]) &&
1088                             printable(bprm->buf[1]) &&
1089                             printable(bprm->buf[2]) &&
1090                             printable(bprm->buf[3]))
1091                                 break; /* -ENOEXEC */
1092                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1093 #endif
1094                 }
1095         }
1096         return retval;
1097 }
1098
1099 EXPORT_SYMBOL(search_binary_handler);
1100
1101 /*
1102  * sys_execve() executes a new program.
1103  */
1104 int do_execve(char * filename,
1105         char __user *__user *argv,
1106         char __user *__user *envp,
1107         struct pt_regs * regs)
1108 {
1109         struct linux_binprm *bprm;
1110         struct file *file;
1111         int retval;
1112         int i;
1113
1114         retval = -ENOMEM;
1115         bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1116         if (!bprm)
1117                 goto out_ret;
1118         memset(bprm, 0, sizeof(*bprm));
1119
1120         file = open_exec(filename);
1121         retval = PTR_ERR(file);
1122         if (IS_ERR(file))
1123                 goto out_kfree;
1124
1125         sched_exec();
1126
1127         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1128
1129         bprm->file = file;
1130         bprm->filename = filename;
1131         bprm->interp = filename;
1132         bprm->mm = mm_alloc();
1133         retval = -ENOMEM;
1134         if (!bprm->mm)
1135                 goto out_file;
1136
1137         retval = init_new_context(current, bprm->mm);
1138         if (retval < 0)
1139                 goto out_mm;
1140
1141         bprm->argc = count(argv, bprm->p / sizeof(void *));
1142         if ((retval = bprm->argc) < 0)
1143                 goto out_mm;
1144
1145         bprm->envc = count(envp, bprm->p / sizeof(void *));
1146         if ((retval = bprm->envc) < 0)
1147                 goto out_mm;
1148
1149         retval = security_bprm_alloc(bprm);
1150         if (retval)
1151                 goto out;
1152
1153         retval = prepare_binprm(bprm);
1154         if (retval < 0)
1155                 goto out;
1156
1157         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1158         if (retval < 0)
1159                 goto out;
1160
1161         bprm->exec = bprm->p;
1162         retval = copy_strings(bprm->envc, envp, bprm);
1163         if (retval < 0)
1164                 goto out;
1165
1166         retval = copy_strings(bprm->argc, argv, bprm);
1167         if (retval < 0)
1168                 goto out;
1169
1170         retval = search_binary_handler(bprm,regs);
1171         if (retval >= 0) {
1172                 free_arg_pages(bprm);
1173
1174                 /* execve success */
1175                 security_bprm_free(bprm);
1176                 kfree(bprm);
1177                 return retval;
1178         }
1179
1180 out:
1181         /* Something went wrong, return the inode and free the argument pages*/
1182         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1183                 struct page * page = bprm->page[i];
1184                 if (page)
1185                         __free_page(page);
1186         }
1187
1188         if (bprm->security)
1189                 security_bprm_free(bprm);
1190
1191 out_mm:
1192         if (bprm->mm)
1193                 mmdrop(bprm->mm);
1194
1195 out_file:
1196         if (bprm->file) {
1197                 allow_write_access(bprm->file);
1198                 fput(bprm->file);
1199         }
1200
1201 out_kfree:
1202         kfree(bprm);
1203
1204 out_ret:
1205         return retval;
1206 }
1207
1208 int set_binfmt(struct linux_binfmt *new)
1209 {
1210         struct linux_binfmt *old = current->binfmt;
1211
1212         if (new) {
1213                 if (!try_module_get(new->module))
1214                         return -1;
1215         }
1216         current->binfmt = new;
1217         if (old)
1218                 module_put(old->module);
1219         return 0;
1220 }
1221
1222 EXPORT_SYMBOL(set_binfmt);
1223
1224 #define CORENAME_MAX_SIZE 64
1225
1226 /* format_corename will inspect the pattern parameter, and output a
1227  * name into corename, which must have space for at least
1228  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1229  */
1230 static void format_corename(char *corename, const char *pattern, long signr)
1231 {
1232         const char *pat_ptr = pattern;
1233         char *out_ptr = corename;
1234         char *const out_end = corename + CORENAME_MAX_SIZE;
1235         int rc;
1236         int pid_in_pattern = 0;
1237
1238         /* Repeat as long as we have more pattern to process and more output
1239            space */
1240         while (*pat_ptr) {
1241                 if (*pat_ptr != '%') {
1242                         if (out_ptr == out_end)
1243                                 goto out;
1244                         *out_ptr++ = *pat_ptr++;
1245                 } else {
1246                         switch (*++pat_ptr) {
1247                         case 0:
1248                                 goto out;
1249                         /* Double percent, output one percent */
1250                         case '%':
1251                                 if (out_ptr == out_end)
1252                                         goto out;
1253                                 *out_ptr++ = '%';
1254                                 break;
1255                         /* pid */
1256                         case 'p':
1257                                 pid_in_pattern = 1;
1258                                 rc = snprintf(out_ptr, out_end - out_ptr,
1259                                               "%d", current->tgid);
1260                                 if (rc > out_end - out_ptr)
1261                                         goto out;
1262                                 out_ptr += rc;
1263                                 break;
1264                         /* uid */
1265                         case 'u':
1266                                 rc = snprintf(out_ptr, out_end - out_ptr,
1267                                               "%d", current->uid);
1268                                 if (rc > out_end - out_ptr)
1269                                         goto out;
1270                                 out_ptr += rc;
1271                                 break;
1272                         /* gid */
1273                         case 'g':
1274                                 rc = snprintf(out_ptr, out_end - out_ptr,
1275                                               "%d", current->gid);
1276                                 if (rc > out_end - out_ptr)
1277                                         goto out;
1278                                 out_ptr += rc;
1279                                 break;
1280                         /* signal that caused the coredump */
1281                         case 's':
1282                                 rc = snprintf(out_ptr, out_end - out_ptr,
1283                                               "%ld", signr);
1284                                 if (rc > out_end - out_ptr)
1285                                         goto out;
1286                                 out_ptr += rc;
1287                                 break;
1288                         /* UNIX time of coredump */
1289                         case 't': {
1290                                 struct timeval tv;
1291                                 do_gettimeofday(&tv);
1292                                 rc = snprintf(out_ptr, out_end - out_ptr,
1293                                               "%lu", tv.tv_sec);
1294                                 if (rc > out_end - out_ptr)
1295                                         goto out;
1296                                 out_ptr += rc;
1297                                 break;
1298                         }
1299                         /* hostname */
1300                         case 'h':
1301                                 down_read(&uts_sem);
1302                                 rc = snprintf(out_ptr, out_end - out_ptr,
1303                                               "%s", system_utsname.nodename);
1304                                 up_read(&uts_sem);
1305                                 if (rc > out_end - out_ptr)
1306                                         goto out;
1307                                 out_ptr += rc;
1308                                 break;
1309                         /* executable */
1310                         case 'e':
1311                                 rc = snprintf(out_ptr, out_end - out_ptr,
1312                                               "%s", current->comm);
1313                                 if (rc > out_end - out_ptr)
1314                                         goto out;
1315                                 out_ptr += rc;
1316                                 break;
1317                         default:
1318                                 break;
1319                         }
1320                         ++pat_ptr;
1321                 }
1322         }
1323         /* Backward compatibility with core_uses_pid:
1324          *
1325          * If core_pattern does not include a %p (as is the default)
1326          * and core_uses_pid is set, then .%pid will be appended to
1327          * the filename */
1328         if (!pid_in_pattern
1329             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1330                 rc = snprintf(out_ptr, out_end - out_ptr,
1331                               ".%d", current->tgid);
1332                 if (rc > out_end - out_ptr)
1333                         goto out;
1334                 out_ptr += rc;
1335         }
1336       out:
1337         *out_ptr = 0;
1338 }
1339
1340 static void zap_threads (struct mm_struct *mm)
1341 {
1342         struct task_struct *g, *p;
1343         struct task_struct *tsk = current;
1344         struct completion *vfork_done = tsk->vfork_done;
1345         int traced = 0;
1346
1347         /*
1348          * Make sure nobody is waiting for us to release the VM,
1349          * otherwise we can deadlock when we wait on each other
1350          */
1351         if (vfork_done) {
1352                 tsk->vfork_done = NULL;
1353                 complete(vfork_done);
1354         }
1355
1356         read_lock(&tasklist_lock);
1357         do_each_thread(g,p)
1358                 if (mm == p->mm && p != tsk) {
1359                         force_sig_specific(SIGKILL, p);
1360                         mm->core_waiters++;
1361                         if (unlikely(p->ptrace) &&
1362                             unlikely(p->parent->mm == mm))
1363                                 traced = 1;
1364                 }
1365         while_each_thread(g,p);
1366
1367         read_unlock(&tasklist_lock);
1368
1369         while (unlikely(traced)) {
1370                 /*
1371                  * We are zapping a thread and the thread it ptraces.
1372                  * The tracee won't come out of TASK_TRACED state until
1373                  * its ptracer detaches.  That happens when the ptracer
1374                  * dies, but it synchronizes with us and so won't get
1375                  * that far until we finish the core dump.  If we're
1376                  * waiting for the tracee to synchronize but it stays
1377                  * blocked in TASK_TRACED, then we deadlock.  So, for
1378                  * this weirdo case we have to do another round with
1379                  * tasklist_lock write-locked to __ptrace_unlink the
1380                  * children that might cause this deadlock.  That will
1381                  * wake them up to process their pending SIGKILL.
1382                  *
1383                  * First, give everyone we just killed a chance to run
1384                  * so they can all get into the coredump synchronization.
1385                  * That should leave only the TASK_TRACED stragglers for
1386                  * us to wake up.  If a ptracer is still running, we'll
1387                  * have to come around again after letting it finish.
1388                  */
1389                 yield();
1390                 traced = 0;
1391                 write_lock_irq(&tasklist_lock);
1392                 do_each_thread(g,p) {
1393                         if (mm != p->mm || p == tsk ||
1394                             !p->ptrace || p->parent->mm != mm)
1395                                 continue;
1396                         if ((p->parent->flags & (PF_SIGNALED|PF_EXITING)) ||
1397                             (p->parent->state & (TASK_TRACED|TASK_STOPPED))) {
1398                                 /*
1399                                  * The parent is in the process of exiting
1400                                  * itself, or else it's stopped right now.
1401                                  * It cannot be in a ptrace call, and would
1402                                  * have to read_lock tasklist_lock before
1403                                  * it could start one, so we are safe here.
1404                                  */
1405                                 __ptrace_unlink(p);
1406                         } else {
1407                                 /*
1408                                  * Blargh!  The ptracer is not dying
1409                                  * yet, so we cannot be sure that it
1410                                  * isn't in the middle of a ptrace call.
1411                                  * We'll have to let it run to get into
1412                                  * coredump_wait and come around another
1413                                  * time to detach its tracee.
1414                                  */
1415                                 traced = 1;
1416                         }
1417                 } while_each_thread(g,p);
1418                 write_unlock_irq(&tasklist_lock);
1419         }
1420 }
1421
1422 static void coredump_wait(struct mm_struct *mm)
1423 {
1424         DECLARE_COMPLETION(startup_done);
1425
1426         mm->core_waiters++; /* let other threads block */
1427         mm->core_startup_done = &startup_done;
1428
1429         /* give other threads a chance to run: */
1430         yield();
1431
1432         zap_threads(mm);
1433         if (--mm->core_waiters) {
1434                 up_write(&mm->mmap_sem);
1435                 wait_for_completion(&startup_done);
1436         } else
1437                 up_write(&mm->mmap_sem);
1438         BUG_ON(mm->core_waiters);
1439 }
1440
1441 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1442 {
1443         char corename[CORENAME_MAX_SIZE + 1];
1444         struct mm_struct *mm = current->mm;
1445         struct linux_binfmt * binfmt;
1446         struct inode * inode;
1447         struct file * file;
1448         int retval = 0;
1449         int fsuid = current->fsuid;
1450         int flag = 0;
1451
1452         binfmt = current->binfmt;
1453         if (!binfmt || !binfmt->core_dump)
1454                 goto fail;
1455         if (current->tux_exit)
1456                 current->tux_exit();
1457         down_write(&mm->mmap_sem);
1458         if (!mm->dumpable) {
1459                 up_write(&mm->mmap_sem);
1460                 goto fail;
1461         }
1462
1463         /*
1464          *      We cannot trust fsuid as being the "true" uid of the
1465          *      process nor do we know its entire history. We only know it
1466          *      was tainted so we dump it as root in mode 2.
1467          */
1468         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1469                 flag = O_EXCL;          /* Stop rewrite attacks */
1470                 current->fsuid = 0;     /* Dump root private */
1471         }
1472         mm->dumpable = 0;
1473         init_completion(&mm->core_done);
1474         current->signal->group_exit = 1;
1475         current->signal->group_exit_code = exit_code;
1476         coredump_wait(mm);
1477
1478         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1479                 goto fail_unlock;
1480
1481         /*
1482          * lock_kernel() because format_corename() is controlled by sysctl, which
1483          * uses lock_kernel()
1484          */
1485         lock_kernel();
1486         format_corename(corename, core_pattern, signr);
1487         unlock_kernel();
1488         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1489         if (IS_ERR(file))
1490                 goto fail_unlock;
1491         inode = file->f_dentry->d_inode;
1492         if (inode->i_nlink > 1)
1493                 goto close_fail;        /* multiple links - don't dump */
1494         if (d_unhashed(file->f_dentry))
1495                 goto close_fail;
1496
1497         if (!S_ISREG(inode->i_mode))
1498                 goto close_fail;
1499         if (!file->f_op)
1500                 goto close_fail;
1501         if (!file->f_op->write)
1502                 goto close_fail;
1503         if (do_truncate(file->f_dentry, 0) != 0)
1504                 goto close_fail;
1505
1506         retval = binfmt->core_dump(signr, regs, file);
1507
1508         if (retval)
1509                 current->signal->group_exit_code |= 0x80;
1510 close_fail:
1511         filp_close(file, NULL);
1512 fail_unlock:
1513         current->fsuid = fsuid;
1514         complete_all(&mm->core_done);
1515 fail:
1516         return retval;
1517 }