fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/mman.h>
  28 #include <linux/a.out.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/init.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/highmem.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/key.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/pid_namespace.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/tracehook.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/vs_memory.h>
  54
  55 #include <asm/uaccess.h>
  56 #include <asm/mmu_context.h>
  57
  58 #ifdef CONFIG_KMOD
  59 #include <linux/kmod.h>
  60 #endif
  61
  62 int core_uses_pid;
  63 char core_pattern[128] = "core";
  64 int suid_dumpable = 0;
  65
  66 EXPORT_SYMBOL(suid_dumpable);
  67 /* The maximal length of core_pattern is also specified in sysctl.c */
  68
  69 static struct linux_binfmt *formats;
  70 static DEFINE_RWLOCK(binfmt_lock);
  71
  72 int register_binfmt(struct linux_binfmt * fmt)
  73 {
  74         struct linux_binfmt ** tmp = &formats;
  75
  76         if (!fmt)
  77                 return -EINVAL;
  78         if (fmt->next)
  79                 return -EBUSY;
  80         write_lock(&binfmt_lock);
  81         while (*tmp) {
  82                 if (fmt == *tmp) {
  83                         write_unlock(&binfmt_lock);
  84                         return -EBUSY;
  85                 }
  86                 tmp = &(*tmp)->next;
  87         }
  88         fmt->next = formats;
  89         formats = fmt;
  90         write_unlock(&binfmt_lock);
  91         return 0;
  92 }
  93
  94 EXPORT_SYMBOL(register_binfmt);
  95
  96 int unregister_binfmt(struct linux_binfmt * fmt)
  97 {
  98         struct linux_binfmt ** tmp = &formats;
  99
 100         write_lock(&binfmt_lock);
 101         while (*tmp) {
 102                 if (fmt == *tmp) {
 103                         *tmp = fmt->next;
 104                         write_unlock(&binfmt_lock);
 105                         return 0;
 106                 }
 107                 tmp = &(*tmp)->next;
 108         }
 109         write_unlock(&binfmt_lock);
 110         return -EINVAL;
 111 }
 112
 113 EXPORT_SYMBOL(unregister_binfmt);
 114
 115 static inline void put_binfmt(struct linux_binfmt * fmt)
 116 {
 117         module_put(fmt->module);
 118 }
 119
 120 /*
 121  * Note that a shared library must be both readable and executable due to
 122  * security reasons.
 123  *
 124  * Also note that we take the address to load from from the file itself.
 125  */
 126 asmlinkage long sys_uselib(const char __user * library)
 127 {
 128         struct file * file;
 129         struct nameidata nd;
 130         int error;
 131
 132         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 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 = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 141         if (error)
 142                 goto exit;
 143
 144         file = nameidata_to_filp(&nd, 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         release_open_intent(&nd);
 173         path_release(&nd);
 174         goto out;
 175 }
 176
 177 /*
 178  * count() counts the number of strings in array ARGV.
 179  */
 180 static int count(char __user * __user * argv, int max)
 181 {
 182         int i = 0;
 183
 184         if (argv != NULL) {
 185                 for (;;) {
 186                         char __user * p;
 187
 188                         if (get_user(p, argv))
 189                                 return -EFAULT;
 190                         if (!p)
 191                                 break;
 192                         argv++;
 193                         if(++i > max)
 194                                 return -E2BIG;
 195                         cond_resched();
 196                 }
 197         }
 198         return i;
 199 }
 200
 201 /*
 202  * 'copy_strings()' copies argument/environment strings from user
 203  * memory to free pages in kernel mem. These are in a format ready
 204  * to be put directly into the top of new user memory.
 205  */
 206 static int copy_strings(int argc, char __user * __user * argv,
 207                         struct linux_binprm *bprm)
 208 {
 209         struct page *kmapped_page = NULL;
 210         char *kaddr = NULL;
 211         int ret;
 212
 213         while (argc-- > 0) {
 214                 char __user *str;
 215                 int len;
 216                 unsigned long pos;
 217
 218                 if (get_user(str, argv+argc) ||
 219                                 !(len = strnlen_user(str, bprm->p))) {
 220                         ret = -EFAULT;
 221                         goto out;
 222                 }
 223
 224                 if (bprm->p < len)  {
 225                         ret = -E2BIG;
 226                         goto out;
 227                 }
 228
 229                 bprm->p -= len;
 230                 /* XXX: add architecture specific overflow check here. */
 231                 pos = bprm->p;
 232
 233                 while (len > 0) {
 234                         int i, new, err;
 235                         int offset, bytes_to_copy;
 236                         struct page *page;
 237
 238                         offset = pos % PAGE_SIZE;
 239                         i = pos/PAGE_SIZE;
 240                         page = bprm->page[i];
 241                         new = 0;
 242                         if (!page) {
 243                                 page = alloc_page(GFP_HIGHUSER);
 244                                 bprm->page[i] = page;
 245                                 if (!page) {
 246                                         ret = -ENOMEM;
 247                                         goto out;
 248                                 }
 249                                 new = 1;
 250                         }
 251
 252                         if (page != kmapped_page) {
 253                                 if (kmapped_page)
 254                                         kunmap(kmapped_page);
 255                                 kmapped_page = page;
 256                                 kaddr = kmap(kmapped_page);
 257                         }
 258                         if (new && offset)
 259                                 memset(kaddr, 0, offset);
 260                         bytes_to_copy = PAGE_SIZE - offset;
 261                         if (bytes_to_copy > len) {
 262                                 bytes_to_copy = len;
 263                                 if (new)
 264                                         memset(kaddr+offset+len, 0,
 265                                                 PAGE_SIZE-offset-len);
 266                         }
 267                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 268                         if (err) {
 269                                 ret = -EFAULT;
 270                                 goto out;
 271                         }
 272
 273                         pos += bytes_to_copy;
 274                         str += bytes_to_copy;
 275                         len -= bytes_to_copy;
 276                 }
 277         }
 278         ret = 0;
 279 out:
 280         if (kmapped_page)
 281                 kunmap(kmapped_page);
 282         return ret;
 283 }
 284
 285 /*
 286  * Like copy_strings, but get argv and its values from kernel memory.
 287  */
 288 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 289 {
 290         int r;
 291         mm_segment_t oldfs = get_fs();
 292         set_fs(KERNEL_DS);
 293         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 294         set_fs(oldfs);
 295         return r;
 296 }
 297
 298 EXPORT_SYMBOL(copy_strings_kernel);
 299
 300 #ifdef CONFIG_MMU
 301 /*
 302  * This routine is used to map in a page into an address space: needed by
 303  * execve() for the initial stack and environment pages.
 304  *
 305  * vma->vm_mm->mmap_sem is held for writing.
 306  */
 307 void install_arg_page(struct vm_area_struct *vma,
 308                         struct page *page, unsigned long address)
 309 {
 310         struct mm_struct *mm = vma->vm_mm;
 311         pte_t * pte;
 312         spinlock_t *ptl;
 313
 314         if (unlikely(anon_vma_prepare(vma)))
 315                 goto out;
 316
 317         flush_dcache_page(page);
 318         pte = get_locked_pte(mm, address, &ptl);
 319         if (!pte)
 320                 goto out;
 321         if (!pte_none(*pte)) {
 322                 pte_unmap_unlock(pte, ptl);
 323                 goto out;
 324         }
 325         inc_mm_counter(mm, anon_rss);
 326         lru_cache_add_active(page);
 327         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
 328                                         page, vma->vm_page_prot))));
 329         page_add_new_anon_rmap(page, vma, address);
 330         pte_unmap_unlock(pte, ptl);
 331
 332         /* no need for flush_tlb */
 333         return;
 334 out:
 335         __free_page(page);
 336         force_sig(SIGKILL, current);
 337 }
 338
 339 #define EXTRA_STACK_VM_PAGES    20      /* random */
 340
 341 int setup_arg_pages(struct linux_binprm *bprm,
 342                     unsigned long stack_top,
 343                     int executable_stack)
 344 {
 345         unsigned long stack_base;
 346         struct vm_area_struct *mpnt;
 347         struct mm_struct *mm = current->mm;
 348         int i, ret;
 349         long arg_size;
 350
 351 #ifdef CONFIG_STACK_GROWSUP
 352         /* Move the argument and environment strings to the bottom of the
 353          * stack space.
 354          */
 355         int offset, j;
 356         char *to, *from;
 357
 358         /* Start by shifting all the pages down */
 359         i = 0;
 360         for (j = 0; j < MAX_ARG_PAGES; j++) {
 361                 struct page *page = bprm->page[j];
 362                 if (!page)
 363                         continue;
 364                 bprm->page[i++] = page;
 365         }
 366
 367         /* Now move them within their pages */
 368         offset = bprm->p % PAGE_SIZE;
 369         to = kmap(bprm->page[0]);
 370         for (j = 1; j < i; j++) {
 371                 memmove(to, to + offset, PAGE_SIZE - offset);
 372                 from = kmap(bprm->page[j]);
 373                 memcpy(to + PAGE_SIZE - offset, from, offset);
 374                 kunmap(bprm->page[j - 1]);
 375                 to = from;
 376         }
 377         memmove(to, to + offset, PAGE_SIZE - offset);
 378         kunmap(bprm->page[j - 1]);
 379
 380         /* Limit stack size to 1GB */
 381         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 382         if (stack_base > (1 << 30))
 383                 stack_base = 1 << 30;
 384         stack_base = PAGE_ALIGN(stack_top - stack_base);
 385
 386         /* Adjust bprm->p to point to the end of the strings. */
 387         bprm->p = stack_base + PAGE_SIZE * i - offset;
 388
 389         mm->arg_start = stack_base;
 390         arg_size = i << PAGE_SHIFT;
 391
 392         /* zero pages that were copied above */
 393         while (i < MAX_ARG_PAGES)
 394                 bprm->page[i++] = NULL;
 395 #else
 396         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
 397         stack_base = PAGE_ALIGN(stack_base);
 398         bprm->p += stack_base;
 399         mm->arg_start = bprm->p;
 400         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 401 #endif
 402
 403         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 404
 405         if (bprm->loader)
 406                 bprm->loader += stack_base;
 407         bprm->exec += stack_base;
 408
 409         mpnt = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
 410         if (!mpnt)
 411                 return -ENOMEM;
 412
 413         memset(mpnt, 0, sizeof(*mpnt));
 414
 415         down_write(&mm->mmap_sem);
 416         {
 417                 mpnt->vm_mm = mm;
 418 #ifdef CONFIG_STACK_GROWSUP
 419                 mpnt->vm_start = stack_base;
 420                 mpnt->vm_end = stack_base + arg_size;
 421 #else
 422                 mpnt->vm_end = stack_top;
 423                 mpnt->vm_start = mpnt->vm_end - arg_size;
 424 #endif
 425                 /* Adjust stack execute permissions; explicitly enable
 426                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 427                  * and leave alone (arch default) otherwise. */
 428                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 429                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 430                 else if (executable_stack == EXSTACK_DISABLE_X)
 431                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 432                 else
 433                         mpnt->vm_flags = VM_STACK_FLAGS;
 434                 mpnt->vm_flags |= mm->def_flags;
 435                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 436                 if ((ret = insert_vm_struct(mm, mpnt))) {
 437                         up_write(&mm->mmap_sem);
 438                         kmem_cache_free(vm_area_cachep, mpnt);
 439                         return ret;
 440                 }
 441                 vx_vmpages_sub(mm, mm->total_vm - vma_pages(mpnt));
 442                 mm->stack_vm = mm->total_vm;
 443         }
 444
 445         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 446                 struct page *page = bprm->page[i];
 447                 if (page) {
 448                         bprm->page[i] = NULL;
 449                         install_arg_page(mpnt, page, stack_base);
 450                 }
 451                 stack_base += PAGE_SIZE;
 452         }
 453         up_write(&mm->mmap_sem);
 454
 455         return 0;
 456 }
 457
 458 EXPORT_SYMBOL(setup_arg_pages);
 459
 460 #define free_arg_pages(bprm) do { } while (0)
 461
 462 #else
 463
 464 static inline void free_arg_pages(struct linux_binprm *bprm)
 465 {
 466         int i;
 467
 468         for (i = 0; i < MAX_ARG_PAGES; i++) {
 469                 if (bprm->page[i])
 470                         __free_page(bprm->page[i]);
 471                 bprm->page[i] = NULL;
 472         }
 473 }
 474
 475 #endif /* CONFIG_MMU */
 476
 477 struct file *open_exec(const char *name)
 478 {
 479         struct nameidata nd;
 480         int err;
 481         struct file *file;
 482
 483         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 484         file = ERR_PTR(err);
 485
 486         if (!err) {
 487                 struct inode *inode = nd.dentry->d_inode;
 488                 file = ERR_PTR(-EACCES);
 489                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 490                     S_ISREG(inode->i_mode)) {
 491                         int err = vfs_permission(&nd, MAY_EXEC);
 492                         file = ERR_PTR(err);
 493                         if (!err) {
 494                                 file = nameidata_to_filp(&nd, O_RDONLY);
 495                                 if (!IS_ERR(file)) {
 496                                         err = deny_write_access(file);
 497                                         if (err) {
 498                                                 fput(file);
 499                                                 file = ERR_PTR(err);
 500                                         }
 501                                 }
 502 out:
 503                                 return file;
 504                         }
 505                 }
 506                 release_open_intent(&nd);
 507                 path_release(&nd);
 508         }
 509         goto out;
 510 }
 511
 512 EXPORT_SYMBOL(open_exec);
 513
 514 int kernel_read(struct file *file, unsigned long offset,
 515         char *addr, unsigned long count)
 516 {
 517         mm_segment_t old_fs;
 518         loff_t pos = offset;
 519         int result;
 520
 521         old_fs = get_fs();
 522         set_fs(get_ds());
 523         /* The cast to a user pointer is valid due to the set_fs() */
 524         result = vfs_read(file, (void __user *)addr, count, &pos);
 525         set_fs(old_fs);
 526         return result;
 527 }
 528
 529 EXPORT_SYMBOL(kernel_read);
 530
 531 static int exec_mmap(struct mm_struct *mm)
 532 {
 533         struct task_struct *tsk;
 534         struct mm_struct * old_mm, *active_mm;
 535
 536         /* Notify parent that we're no longer interested in the old VM */
 537         tsk = current;
 538         old_mm = current->mm;
 539         mm_release(tsk, old_mm);
 540
 541         if (old_mm) {
 542                 /*
 543                  * Make sure that if there is a core dump in progress
 544                  * for the old mm, we get out and die instead of going
 545                  * through with the exec.  We must hold mmap_sem around
 546                  * checking core_waiters and changing tsk->mm.  The
 547                  * core-inducing thread will increment core_waiters for
 548                  * each thread whose ->mm == old_mm.
 549                  */
 550                 down_read(&old_mm->mmap_sem);
 551                 if (unlikely(old_mm->core_waiters)) {
 552                         up_read(&old_mm->mmap_sem);
 553                         return -EINTR;
 554                 }
 555         }
 556         task_lock(tsk);
 557         active_mm = tsk->active_mm;
 558         tsk->mm = mm;
 559         tsk->active_mm = mm;
 560         activate_mm(active_mm, mm);
 561         task_unlock(tsk);
 562         arch_pick_mmap_layout(mm);
 563         if (old_mm) {
 564                 up_read(&old_mm->mmap_sem);
 565                 BUG_ON(active_mm != old_mm);
 566                 mmput(old_mm);
 567                 return 0;
 568         }
 569         mmdrop(active_mm);
 570         return 0;
 571 }
 572
 573 /*
 574  * This function makes sure the current process has its own signal table,
 575  * so that flush_signal_handlers can later reset the handlers without
 576  * disturbing other processes.  (Other processes might share the signal
 577  * table via the CLONE_SIGHAND option to clone().)
 578  */
 579 static int de_thread(struct task_struct *tsk)
 580 {
 581         struct signal_struct *sig = tsk->signal;
 582         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 583         spinlock_t *lock = &oldsighand->siglock;
 584         struct task_struct *leader = NULL;
 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(tsk))
 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->flags & SIGNAL_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
 621         /*
 622          * child_reaper ignores SIGKILL, change it now.
 623          * Reparenting needs write_lock on tasklist_lock,
 624          * so it is safe to do it under read_lock.
 625          */
 626         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 627                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 628
 629         zap_other_threads(tsk);
 630         read_unlock(&tasklist_lock);
 631
 632         /*
 633          * Account for the thread group leader hanging around:
 634          */
 635         count = 1;
 636         if (!thread_group_leader(tsk)) {
 637                 count = 2;
 638                 /*
 639                  * The SIGALRM timer survives the exec, but needs to point
 640                  * at us as the new group leader now.  We have a race with
 641                  * a timer firing now getting the old leader, so we need to
 642                  * synchronize with any firing (by calling del_timer_sync)
 643                  * before we can safely let the old group leader die.
 644                  */
 645                 sig->tsk = tsk;
 646                 spin_unlock_irq(lock);
 647                 if (hrtimer_cancel(&sig->real_timer))
 648                         hrtimer_restart(&sig->real_timer);
 649                 spin_lock_irq(lock);
 650         }
 651         while (atomic_read(&sig->count) > count) {
 652                 sig->group_exit_task = tsk;
 653                 sig->notify_count = count;
 654                 __set_current_state(TASK_UNINTERRUPTIBLE);
 655                 spin_unlock_irq(lock);
 656                 schedule();
 657                 spin_lock_irq(lock);
 658         }
 659         sig->group_exit_task = NULL;
 660         sig->notify_count = 0;
 661         spin_unlock_irq(lock);
 662
 663         /*
 664          * At this point all other threads have exited, all we have to
 665          * do is to wait for the thread group leader to become inactive,
 666          * and to assume its PID:
 667          */
 668         if (!thread_group_leader(tsk)) {
 669                 /*
 670                  * Wait for the thread group leader to be a zombie.
 671                  * It should already be zombie at this point, most
 672                  * of the time.
 673                  */
 674                 leader = tsk->group_leader;
 675                 while (leader->exit_state != EXIT_ZOMBIE)
 676                         yield();
 677
 678                 /*
 679                  * The only record we have of the real-time age of a
 680                  * process, regardless of execs it's done, is start_time.
 681                  * All the past CPU time is accumulated in signal_struct
 682                  * from sister threads now dead.  But in this non-leader
 683                  * exec, nothing survives from the original leader thread,
 684                  * whose birth marks the true age of this process now.
 685                  * When we take on its identity by switching to its PID, we
 686                  * also take its birthdate (always earlier than our own).
 687                  */
 688                 tsk->start_time = leader->start_time;
 689
 690                 write_lock_irq(&tasklist_lock);
 691
 692                 BUG_ON(leader->tgid != tsk->tgid);
 693                 BUG_ON(tsk->pid == tsk->tgid);
 694                 /*
 695                  * An exec() starts a new thread group with the
 696                  * TGID of the previous thread group. Rehash the
 697                  * two threads with a switched PID, and release
 698                  * the former thread group leader:
 699                  */
 700
 701                 /* Become a process group leader with the old leader's pid.
 702                  * The old leader becomes a thread of the this thread group.
 703                  * Note: The old leader also uses this pid until release_task
 704                  *       is called.  Odd but simple and correct.
 705                  */
 706                 detach_pid(tsk, PIDTYPE_PID);
 707                 tsk->pid = leader->pid;
 708                 attach_pid(tsk, PIDTYPE_PID,  tsk->pid);
 709                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 710                 transfer_pid(leader, tsk, PIDTYPE_SID);
 711                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 712
 713                 tsk->group_leader = tsk;
 714                 leader->group_leader = tsk;
 715
 716                 tsk->exit_signal = SIGCHLD;
 717
 718                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 719                 leader->exit_state = EXIT_DEAD;
 720
 721                 write_unlock_irq(&tasklist_lock);
 722         }
 723
 724         /*
 725          * There may be one thread left which is just exiting,
 726          * but it's safe to stop telling the group to kill themselves.
 727          */
 728         sig->flags = 0;
 729
 730 no_thread_group:
 731         exit_itimers(sig);
 732         if (leader)
 733                 release_task(leader);
 734
 735         BUG_ON(atomic_read(&sig->count) != 1);
 736
 737         if (atomic_read(&oldsighand->count) == 1) {
 738                 /*
 739                  * Now that we nuked the rest of the thread group,
 740                  * it turns out we are not sharing sighand any more either.
 741                  * So we can just keep it.
 742                  */
 743                 kmem_cache_free(sighand_cachep, newsighand);
 744         } else {
 745                 /*
 746                  * Move our state over to newsighand and switch it in.
 747                  */
 748                 atomic_set(&newsighand->count, 1);
 749                 memcpy(newsighand->action, oldsighand->action,
 750                        sizeof(newsighand->action));
 751
 752                 write_lock_irq(&tasklist_lock);
 753                 spin_lock(&oldsighand->siglock);
 754                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 755
 756                 rcu_assign_pointer(tsk->sighand, newsighand);
 757                 recalc_sigpending();
 758
 759                 spin_unlock(&newsighand->siglock);
 760                 spin_unlock(&oldsighand->siglock);
 761                 write_unlock_irq(&tasklist_lock);
 762
 763                 if (atomic_dec_and_test(&oldsighand->count))
 764                         kmem_cache_free(sighand_cachep, oldsighand);
 765         }
 766
 767         BUG_ON(!thread_group_leader(tsk));
 768         return 0;
 769 }
 770
 771 /*
 772  * These functions flushes out all traces of the currently running executable
 773  * so that a new one can be started
 774  */
 775
 776 static void flush_old_files(struct files_struct * files)
 777 {
 778         long j = -1;
 779         struct fdtable *fdt;
 780
 781         spin_lock(&files->file_lock);
 782         for (;;) {
 783                 unsigned long set, i;
 784
 785                 j++;
 786                 i = j * __NFDBITS;
 787                 fdt = files_fdtable(files);
 788                 if (i >= fdt->max_fds)
 789                         break;
 790                 set = fdt->close_on_exec->fds_bits[j];
 791                 if (!set)
 792                         continue;
 793                 fdt->close_on_exec->fds_bits[j] = 0;
 794                 spin_unlock(&files->file_lock);
 795                 for ( ; set ; i++,set >>= 1) {
 796                         if (set & 1) {
 797                                 sys_close(i);
 798                         }
 799                 }
 800                 spin_lock(&files->file_lock);
 801
 802         }
 803         spin_unlock(&files->file_lock);
 804 }
 805
 806 void get_task_comm(char *buf, struct task_struct *tsk)
 807 {
 808         /* buf must be at least sizeof(tsk->comm) in size */
 809         task_lock(tsk);
 810         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 811         task_unlock(tsk);
 812 }
 813
 814 void set_task_comm(struct task_struct *tsk, char *buf)
 815 {
 816         task_lock(tsk);
 817         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 818         task_unlock(tsk);
 819 }
 820
 821 int flush_old_exec(struct linux_binprm * bprm)
 822 {
 823         char * name;
 824         int i, ch, retval;
 825         struct files_struct *files;
 826         char tcomm[sizeof(current->comm)];
 827
 828         /*
 829          * Make sure we have a private signal table and that
 830          * we are unassociated from the previous thread group.
 831          */
 832         retval = de_thread(current);
 833         if (retval)
 834                 goto out;
 835
 836         /*
 837          * Make sure we have private file handles. Ask the
 838          * fork helper to do the work for us and the exit
 839          * helper to do the cleanup of the old one.
 840          */
 841         files = current->files;         /* refcounted so safe to hold */
 842         retval = unshare_files();
 843         if (retval)
 844                 goto out;
 845         /*
 846          * Release all of the old mmap stuff
 847          */
 848         retval = exec_mmap(bprm->mm);
 849         if (retval)
 850                 goto mmap_failed;
 851
 852         bprm->mm = NULL;                /* We're using it now */
 853
 854         /* This is the point of no return */
 855         put_files_struct(files);
 856
 857         current->sas_ss_sp = current->sas_ss_size = 0;
 858
 859         if (current->euid == current->uid && current->egid == current->gid)
 860                 current->mm->dumpable = 1;
 861         else
 862                 current->mm->dumpable = suid_dumpable;
 863
 864         name = bprm->filename;
 865
 866         /* Copies the binary name from after last slash */
 867         for (i=0; (ch = *(name++)) != '\0';) {
 868                 if (ch == '/')
 869                         i = 0; /* overwrite what we wrote */
 870                 else
 871                         if (i < (sizeof(tcomm) - 1))
 872                                 tcomm[i++] = ch;
 873         }
 874         tcomm[i] = '\0';
 875         set_task_comm(current, tcomm);
 876
 877         current->flags &= ~PF_RANDOMIZE;
 878         flush_thread();
 879
 880         /* Set the new mm task size. We have to do that late because it may
 881          * depend on TIF_32BIT which is only updated in flush_thread() on
 882          * some architectures like powerpc
 883          */
 884         current->mm->task_size = TASK_SIZE;
 885
 886         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 887             file_permission(bprm->file, MAY_READ) ||
 888             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 889                 suid_keys(current);
 890                 current->mm->dumpable = suid_dumpable;
 891         }
 892
 893         /* An exec changes our domain. We are no longer part of the thread
 894            group */
 895
 896         current->self_exec_id++;
 897
 898         flush_signal_handlers(current, 0);
 899         flush_old_files(current->files);
 900
 901         return 0;
 902
 903 mmap_failed:
 904         reset_files_struct(current, files);
 905 out:
 906         return retval;
 907 }
 908
 909 EXPORT_SYMBOL(flush_old_exec);
 910
 911 /*
 912  * Fill the binprm structure from the inode.
 913  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 914  */
 915 int prepare_binprm(struct linux_binprm *bprm)
 916 {
 917         int mode;
 918         struct inode * inode = bprm->file->f_path.dentry->d_inode;
 919         int retval;
 920
 921         mode = inode->i_mode;
 922         if (bprm->file->f_op == NULL)
 923                 return -EACCES;
 924
 925         bprm->e_uid = current->euid;
 926         bprm->e_gid = current->egid;
 927
 928         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
 929                 /* Set-uid? */
 930                 if (mode & S_ISUID) {
 931                         current->personality &= ~PER_CLEAR_ON_SETID;
 932                         bprm->e_uid = inode->i_uid;
 933                 }
 934
 935                 /* Set-gid? */
 936                 /*
 937                  * If setgid is set but no group execute bit then this
 938                  * is a candidate for mandatory locking, not a setgid
 939                  * executable.
 940                  */
 941                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 942                         current->personality &= ~PER_CLEAR_ON_SETID;
 943                         bprm->e_gid = inode->i_gid;
 944                 }
 945         }
 946
 947         /* fill in binprm security blob */
 948         retval = security_bprm_set(bprm);
 949         if (retval)
 950                 return retval;
 951
 952         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 953         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 954 }
 955
 956 EXPORT_SYMBOL(prepare_binprm);
 957
 958 static int unsafe_exec(struct task_struct *p)
 959 {
 960         int unsafe = tracehook_unsafe_exec(p);
 961         if (atomic_read(&p->fs->count) > 1 ||
 962             atomic_read(&p->files->count) > 1 ||
 963             atomic_read(&p->sighand->count) > 1)
 964                 unsafe |= LSM_UNSAFE_SHARE;
 965
 966         return unsafe;
 967 }
 968
 969 void compute_creds(struct linux_binprm *bprm)
 970 {
 971         int unsafe;
 972
 973         if (bprm->e_uid != current->uid)
 974                 suid_keys(current);
 975         exec_keys(current);
 976
 977         task_lock(current);
 978         unsafe = unsafe_exec(current);
 979         security_bprm_apply_creds(bprm, unsafe);
 980         task_unlock(current);
 981         security_bprm_post_apply_creds(bprm);
 982 }
 983
 984 EXPORT_SYMBOL(compute_creds);
 985
 986 void remove_arg_zero(struct linux_binprm *bprm)
 987 {
 988         if (bprm->argc) {
 989                 unsigned long offset;
 990                 char * kaddr;
 991                 struct page *page;
 992
 993                 offset = bprm->p % PAGE_SIZE;
 994                 goto inside;
 995
 996                 while (bprm->p++, *(kaddr+offset++)) {
 997                         if (offset != PAGE_SIZE)
 998                                 continue;
 999                         offset = 0;
1000                         kunmap_atomic(kaddr, KM_USER0);
1001 inside:
1002                         page = bprm->page[bprm->p/PAGE_SIZE];
1003                         kaddr = kmap_atomic(page, KM_USER0);
1004                 }
1005                 kunmap_atomic(kaddr, KM_USER0);
1006                 bprm->argc--;
1007         }
1008 }
1009
1010 EXPORT_SYMBOL(remove_arg_zero);
1011
1012 /*
1013  * cycle the list of binary formats handler, until one recognizes the image
1014  */
1015 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1016 {
1017         int try,retval;
1018         struct linux_binfmt *fmt;
1019 #ifdef __alpha__
1020         /* handle /sbin/loader.. */
1021         {
1022             struct exec * eh = (struct exec *) bprm->buf;
1023
1024             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1025                 (eh->fh.f_flags & 0x3000) == 0x3000)
1026             {
1027                 struct file * file;
1028                 unsigned long loader;
1029
1030                 allow_write_access(bprm->file);
1031                 fput(bprm->file);
1032                 bprm->file = NULL;
1033
1034                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1035
1036                 file = open_exec("/sbin/loader");
1037                 retval = PTR_ERR(file);
1038                 if (IS_ERR(file))
1039                         return retval;
1040
1041                 /* Remember if the application is TASO.  */
1042                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1043
1044                 bprm->file = file;
1045                 bprm->loader = loader;
1046                 retval = prepare_binprm(bprm);
1047                 if (retval<0)
1048                         return retval;
1049                 /* should call search_binary_handler recursively here,
1050                    but it does not matter */
1051             }
1052         }
1053 #endif
1054         retval = security_bprm_check(bprm);
1055         if (retval)
1056                 return retval;
1057
1058         /* kernel module loader fixup */
1059         /* so we don't try to load run modprobe in kernel space. */
1060         set_fs(USER_DS);
1061
1062         retval = audit_bprm(bprm);
1063         if (retval)
1064                 return retval;
1065
1066         retval = -ENOENT;
1067         for (try=0; try<2; try++) {
1068                 read_lock(&binfmt_lock);
1069                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1070                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1071                         if (!fn)
1072                                 continue;
1073                         if (!try_module_get(fmt->module))
1074                                 continue;
1075                         read_unlock(&binfmt_lock);
1076                         retval = fn(bprm, regs);
1077                         if (retval >= 0) {
1078                                 put_binfmt(fmt);
1079                                 allow_write_access(bprm->file);
1080                                 if (bprm->file)
1081                                         fput(bprm->file);
1082                                 bprm->file = NULL;
1083                                 current->did_exec = 1;
1084                                 proc_exec_connector(current);
1085                                 tracehook_report_exec(bprm, regs);
1086                                 return retval;
1087                         }
1088                         read_lock(&binfmt_lock);
1089                         put_binfmt(fmt);
1090                         if (retval != -ENOEXEC || bprm->mm == NULL)
1091                                 break;
1092                         if (!bprm->file) {
1093                                 read_unlock(&binfmt_lock);
1094                                 return retval;
1095                         }
1096                 }
1097                 read_unlock(&binfmt_lock);
1098                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1099                         break;
1100 #ifdef CONFIG_KMOD
1101                 }else{
1102 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1103                         if (printable(bprm->buf[0]) &&
1104                             printable(bprm->buf[1]) &&
1105                             printable(bprm->buf[2]) &&
1106                             printable(bprm->buf[3]))
1107                                 break; /* -ENOEXEC */
1108                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1109 #endif
1110                 }
1111         }
1112         return retval;
1113 }
1114
1115 EXPORT_SYMBOL(search_binary_handler);
1116
1117 /*
1118  * sys_execve() executes a new program.
1119  */
1120 int do_execve(char * filename,
1121         char __user *__user *argv,
1122         char __user *__user *envp,
1123         struct pt_regs * regs)
1124 {
1125         struct linux_binprm *bprm;
1126         struct file *file;
1127         int retval;
1128         int i;
1129
1130         retval = -ENOMEM;
1131         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1132         if (!bprm)
1133                 goto out_ret;
1134
1135         file = open_exec(filename);
1136         retval = PTR_ERR(file);
1137         if (IS_ERR(file))
1138                 goto out_kfree;
1139
1140         sched_exec();
1141
1142         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1143
1144         bprm->file = file;
1145         bprm->filename = filename;
1146         bprm->interp = filename;
1147         bprm->mm = mm_alloc();
1148         retval = -ENOMEM;
1149         if (!bprm->mm)
1150                 goto out_file;
1151
1152         retval = init_new_context(current, bprm->mm);
1153         if (retval < 0)
1154                 goto out_mm;
1155
1156         bprm->argc = count(argv, bprm->p / sizeof(void *));
1157         if ((retval = bprm->argc) < 0)
1158                 goto out_mm;
1159
1160         bprm->envc = count(envp, bprm->p / sizeof(void *));
1161         if ((retval = bprm->envc) < 0)
1162                 goto out_mm;
1163
1164         retval = security_bprm_alloc(bprm);
1165         if (retval)
1166                 goto out;
1167
1168         retval = prepare_binprm(bprm);
1169         if (retval < 0)
1170                 goto out;
1171
1172         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1173         if (retval < 0)
1174                 goto out;
1175
1176         bprm->exec = bprm->p;
1177         retval = copy_strings(bprm->envc, envp, bprm);
1178         if (retval < 0)
1179                 goto out;
1180
1181         retval = copy_strings(bprm->argc, argv, bprm);
1182         if (retval < 0)
1183                 goto out;
1184
1185         retval = search_binary_handler(bprm,regs);
1186         if (retval >= 0) {
1187                 free_arg_pages(bprm);
1188
1189                 /* execve success */
1190                 security_bprm_free(bprm);
1191                 acct_update_integrals(current);
1192                 kfree(bprm);
1193                 return retval;
1194         }
1195
1196 out:
1197         /* Something went wrong, return the inode and free the argument pages*/
1198         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1199                 struct page * page = bprm->page[i];
1200                 if (page)
1201                         __free_page(page);
1202         }
1203
1204         if (bprm->security)
1205                 security_bprm_free(bprm);
1206
1207 out_mm:
1208         if (bprm->mm)
1209                 mmdrop(bprm->mm);
1210
1211 out_file:
1212         if (bprm->file) {
1213                 allow_write_access(bprm->file);
1214                 fput(bprm->file);
1215         }
1216
1217 out_kfree:
1218         kfree(bprm);
1219
1220 out_ret:
1221         return retval;
1222 }
1223
1224 int set_binfmt(struct linux_binfmt *new)
1225 {
1226         struct linux_binfmt *old = current->binfmt;
1227
1228         if (new) {
1229                 if (!try_module_get(new->module))
1230                         return -1;
1231         }
1232         current->binfmt = new;
1233         if (old)
1234                 module_put(old->module);
1235         return 0;
1236 }
1237
1238 EXPORT_SYMBOL(set_binfmt);
1239
1240 #define CORENAME_MAX_SIZE 64
1241
1242 /* format_corename will inspect the pattern parameter, and output a
1243  * name into corename, which must have space for at least
1244  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1245  */
1246 static int format_corename(char *corename, const char *pattern, long signr)
1247 {
1248         const char *pat_ptr = pattern;
1249         char *out_ptr = corename;
1250         char *const out_end = corename + CORENAME_MAX_SIZE;
1251         int rc;
1252         int pid_in_pattern = 0;
1253         int ispipe = 0;
1254
1255         if (*pattern == '|')
1256                 ispipe = 1;
1257
1258         /* Repeat as long as we have more pattern to process and more output
1259            space */
1260         while (*pat_ptr) {
1261                 if (*pat_ptr != '%') {
1262                         if (out_ptr == out_end)
1263                                 goto out;
1264                         *out_ptr++ = *pat_ptr++;
1265                 } else {
1266                         switch (*++pat_ptr) {
1267                         case 0:
1268                                 goto out;
1269                         /* Double percent, output one percent */
1270                         case '%':
1271                                 if (out_ptr == out_end)
1272                                         goto out;
1273                                 *out_ptr++ = '%';
1274                                 break;
1275                         /* pid */
1276                         case 'p':
1277                                 pid_in_pattern = 1;
1278                                 rc = snprintf(out_ptr, out_end - out_ptr,
1279                                               "%d", current->tgid);
1280                                 if (rc > out_end - out_ptr)
1281                                         goto out;
1282                                 out_ptr += rc;
1283                                 break;
1284                         /* uid */
1285                         case 'u':
1286                                 rc = snprintf(out_ptr, out_end - out_ptr,
1287                                               "%d", current->uid);
1288                                 if (rc > out_end - out_ptr)
1289                                         goto out;
1290                                 out_ptr += rc;
1291                                 break;
1292                         /* gid */
1293                         case 'g':
1294                                 rc = snprintf(out_ptr, out_end - out_ptr,
1295                                               "%d", current->gid);
1296                                 if (rc > out_end - out_ptr)
1297                                         goto out;
1298                                 out_ptr += rc;
1299                                 break;
1300                         /* signal that caused the coredump */
1301                         case 's':
1302                                 rc = snprintf(out_ptr, out_end - out_ptr,
1303                                               "%ld", signr);
1304                                 if (rc > out_end - out_ptr)
1305                                         goto out;
1306                                 out_ptr += rc;
1307                                 break;
1308                         /* UNIX time of coredump */
1309                         case 't': {
1310                                 struct timeval tv;
1311                                 vx_gettimeofday(&tv);
1312                                 rc = snprintf(out_ptr, out_end - out_ptr,
1313                                               "%lu", tv.tv_sec);
1314                                 if (rc > out_end - out_ptr)
1315                                         goto out;
1316                                 out_ptr += rc;
1317                                 break;
1318                         }
1319                         /* hostname */
1320                         case 'h':
1321                                 down_read(&uts_sem);
1322                                 rc = snprintf(out_ptr, out_end - out_ptr,
1323                                               "%s", utsname()->nodename);
1324                                 up_read(&uts_sem);
1325                                 if (rc > out_end - out_ptr)
1326                                         goto out;
1327                                 out_ptr += rc;
1328                                 break;
1329                         /* executable */
1330                         case 'e':
1331                                 rc = snprintf(out_ptr, out_end - out_ptr,
1332                                               "%s", current->comm);
1333                                 if (rc > out_end - out_ptr)
1334                                         goto out;
1335                                 out_ptr += rc;
1336                                 break;
1337                         default:
1338                                 break;
1339                         }
1340                         ++pat_ptr;
1341                 }
1342         }
1343         /* Backward compatibility with core_uses_pid:
1344          *
1345          * If core_pattern does not include a %p (as is the default)
1346          * and core_uses_pid is set, then .%pid will be appended to
1347          * the filename. Do not do this for piped commands. */
1348         if (!ispipe && !pid_in_pattern
1349             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1350                 rc = snprintf(out_ptr, out_end - out_ptr,
1351                               ".%d", current->tgid);
1352                 if (rc > out_end - out_ptr)
1353                         goto out;
1354                 out_ptr += rc;
1355         }
1356 out:
1357         *out_ptr = 0;
1358         return ispipe;
1359 }
1360
1361 static void zap_process(struct task_struct *start)
1362 {
1363         struct task_struct *t;
1364
1365         start->signal->flags = SIGNAL_GROUP_EXIT;
1366         start->signal->group_stop_count = 0;
1367
1368         t = start;
1369         do {
1370                 if (t != current && t->mm) {
1371                         t->mm->core_waiters++;
1372                         sigaddset(&t->pending.signal, SIGKILL);
1373                         signal_wake_up(t, 1);
1374                 }
1375         } while ((t = next_thread(t)) != start);
1376 }
1377
1378 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1379                                 int exit_code)
1380 {
1381         struct task_struct *g, *p;
1382         unsigned long flags;
1383         int err = -EAGAIN;
1384
1385         spin_lock_irq(&tsk->sighand->siglock);
1386         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1387                 tsk->signal->group_exit_code = exit_code;
1388                 zap_process(tsk);
1389                 err = 0;
1390         }
1391         spin_unlock_irq(&tsk->sighand->siglock);
1392         if (err)
1393                 return err;
1394
1395         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1396                 goto done;
1397
1398         rcu_read_lock();
1399         for_each_process(g) {
1400                 if (g == tsk->group_leader)
1401                         continue;
1402
1403                 p = g;
1404                 do {
1405                         if (p->mm) {
1406                                 if (p->mm == mm) {
1407                                         /*
1408                                          * p->sighand can't disappear, but
1409                                          * may be changed by de_thread()
1410                                          */
1411                                         lock_task_sighand(p, &flags);
1412                                         zap_process(p);
1413                                         unlock_task_sighand(p, &flags);
1414                                 }
1415                                 break;
1416                         }
1417                 } while ((p = next_thread(p)) != g);
1418         }
1419         rcu_read_unlock();
1420 done:
1421         return mm->core_waiters;
1422 }
1423
1424 static int coredump_wait(int exit_code)
1425 {
1426         struct task_struct *tsk = current;
1427         struct mm_struct *mm = tsk->mm;
1428         struct completion startup_done;
1429         struct completion *vfork_done;
1430         int core_waiters;
1431
1432         init_completion(&mm->core_done);
1433         init_completion(&startup_done);
1434         mm->core_startup_done = &startup_done;
1435
1436         core_waiters = zap_threads(tsk, mm, exit_code);
1437         up_write(&mm->mmap_sem);
1438
1439         if (unlikely(core_waiters < 0))
1440                 goto fail;
1441
1442         /*
1443          * Make sure nobody is waiting for us to release the VM,
1444          * otherwise we can deadlock when we wait on each other
1445          */
1446         vfork_done = tsk->vfork_done;
1447         if (vfork_done) {
1448                 tsk->vfork_done = NULL;
1449                 complete(vfork_done);
1450         }
1451
1452         if (core_waiters)
1453                 wait_for_completion(&startup_done);
1454 fail:
1455         BUG_ON(mm->core_waiters);
1456         return core_waiters;
1457 }
1458
1459 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1460 {
1461         char corename[CORENAME_MAX_SIZE + 1];
1462         struct mm_struct *mm = current->mm;
1463         struct linux_binfmt * binfmt;
1464         struct inode * inode;
1465         struct file * file;
1466         int retval = 0;
1467         int fsuid = current->fsuid;
1468         int flag = 0;
1469         int ispipe = 0;
1470
1471         binfmt = current->binfmt;
1472         if (!binfmt || !binfmt->core_dump)
1473                 goto fail;
1474         down_write(&mm->mmap_sem);
1475         if (!mm->dumpable) {
1476                 up_write(&mm->mmap_sem);
1477                 goto fail;
1478         }
1479
1480         /*
1481          *      We cannot trust fsuid as being the "true" uid of the
1482          *      process nor do we know its entire history. We only know it
1483          *      was tainted so we dump it as root in mode 2.
1484          */
1485         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1486                 flag = O_EXCL;          /* Stop rewrite attacks */
1487                 current->fsuid = 0;     /* Dump root private */
1488         }
1489         mm->dumpable = 0;
1490
1491         retval = coredump_wait(exit_code);
1492         if (retval < 0)
1493                 goto fail;
1494
1495         /*
1496          * Clear any false indication of pending signals that might
1497          * be seen by the filesystem code called to write the core file.
1498          */
1499         clear_thread_flag(TIF_SIGPENDING);
1500
1501         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1502                 goto fail_unlock;
1503
1504         /*
1505          * lock_kernel() because format_corename() is controlled by sysctl, which
1506          * uses lock_kernel()
1507          */
1508         lock_kernel();
1509         ispipe = format_corename(corename, core_pattern, signr);
1510         unlock_kernel();
1511         if (ispipe) {
1512                 /* SIGPIPE can happen, but it's just never processed */
1513                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1514                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1515                                corename);
1516                         goto fail_unlock;
1517                 }
1518         } else
1519                 file = filp_open(corename,
1520                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1521                                  0600);
1522         if (IS_ERR(file))
1523                 goto fail_unlock;
1524         inode = file->f_path.dentry->d_inode;
1525         if (inode->i_nlink > 1)
1526                 goto close_fail;        /* multiple links - don't dump */
1527         if (!ispipe && d_unhashed(file->f_path.dentry))
1528                 goto close_fail;
1529
1530         /* AK: actually i see no reason to not allow this for named pipes etc.,
1531            but keep the previous behaviour for now. */
1532         if (!ispipe && !S_ISREG(inode->i_mode))
1533                 goto close_fail;
1534         if (!file->f_op)
1535                 goto close_fail;
1536         if (!file->f_op->write)
1537                 goto close_fail;
1538         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1539                 goto close_fail;
1540
1541         retval = binfmt->core_dump(signr, regs, file);
1542
1543         if (retval)
1544                 current->signal->group_exit_code |= 0x80;
1545 close_fail:
1546         filp_close(file, NULL);
1547 fail_unlock:
1548         current->fsuid = fsuid;
1549         complete_all(&mm->core_done);
1550 fail:
1551         return retval;
1552 }