fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/config.h>
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/mman.h>
  29 #include <linux/a.out.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/key.h>
  38 #include <linux/personality.h>
  39 #include <linux/binfmts.h>
  40 #include <linux/swap.h>
  41 #include <linux/utsname.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/acct.h>
  51 #include <linux/cn_proc.h>
  52
  53 #include <asm/uaccess.h>
  54 #include <asm/mmu_context.h>
  55
  56 #ifdef CONFIG_KMOD
  57 #include <linux/kmod.h>
  58 #endif
  59
  60 int core_uses_pid;
  61 char core_pattern[65] = "core";
  62 int suid_dumpable = 0;
  63
  64 EXPORT_SYMBOL(suid_dumpable);
  65 /* The maximal length of core_pattern is also specified in sysctl.c */
  66
  67 static struct linux_binfmt *formats;
  68 static DEFINE_RWLOCK(binfmt_lock);
  69
  70 int register_binfmt(struct linux_binfmt * fmt)
  71 {
  72         struct linux_binfmt ** tmp = &formats;
  73
  74         if (!fmt)
  75                 return -EINVAL;
  76         if (fmt->next)
  77                 return -EBUSY;
  78         write_lock(&binfmt_lock);
  79         while (*tmp) {
  80                 if (fmt == *tmp) {
  81                         write_unlock(&binfmt_lock);
  82                         return -EBUSY;
  83                 }
  84                 tmp = &(*tmp)->next;
  85         }
  86         fmt->next = formats;
  87         formats = fmt;
  88         write_unlock(&binfmt_lock);
  89         return 0;
  90 }
  91
  92 EXPORT_SYMBOL(register_binfmt);
  93
  94 int unregister_binfmt(struct linux_binfmt * fmt)
  95 {
  96         struct linux_binfmt ** tmp = &formats;
  97
  98         write_lock(&binfmt_lock);
  99         while (*tmp) {
 100                 if (fmt == *tmp) {
 101                         *tmp = fmt->next;
 102                         write_unlock(&binfmt_lock);
 103                         return 0;
 104                 }
 105                 tmp = &(*tmp)->next;
 106         }
 107         write_unlock(&binfmt_lock);
 108         return -EINVAL;
 109 }
 110
 111 EXPORT_SYMBOL(unregister_binfmt);
 112
 113 static inline void put_binfmt(struct linux_binfmt * fmt)
 114 {
 115         module_put(fmt->module);
 116 }
 117
 118 /*
 119  * Note that a shared library must be both readable and executable due to
 120  * security reasons.
 121  *
 122  * Also note that we take the address to load from from the file itself.
 123  */
 124 asmlinkage long sys_uselib(const char __user * library)
 125 {
 126         struct file * file;
 127         struct nameidata nd;
 128         int error;
 129
 130         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 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 = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 139         if (error)
 140                 goto exit;
 141
 142         file = nameidata_to_filp(&nd, 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         release_open_intent(&nd);
 171         path_release(&nd);
 172         goto out;
 173 }
 174
 175 /*
 176  * count() counts the number of strings in array ARGV.
 177  */
 178 static int count(char __user * __user * argv, int max)
 179 {
 180         int i = 0;
 181
 182         if (argv != NULL) {
 183                 for (;;) {
 184                         char __user * p;
 185
 186                         if (get_user(p, argv))
 187                                 return -EFAULT;
 188                         if (!p)
 189                                 break;
 190                         argv++;
 191                         if(++i > max)
 192                                 return -E2BIG;
 193                         cond_resched();
 194                 }
 195         }
 196         return i;
 197 }
 198
 199 /*
 200  * 'copy_strings()' copies argument/environment strings from user
 201  * memory to free pages in kernel mem. These are in a format ready
 202  * to be put directly into the top of new user memory.
 203  */
 204 static int copy_strings(int argc, char __user * __user * argv,
 205                         struct linux_binprm *bprm)
 206 {
 207         struct page *kmapped_page = NULL;
 208         char *kaddr = NULL;
 209         int ret;
 210
 211         while (argc-- > 0) {
 212                 char __user *str;
 213                 int len;
 214                 unsigned long pos;
 215
 216                 if (get_user(str, argv+argc) ||
 217                                 !(len = strnlen_user(str, bprm->p))) {
 218                         ret = -EFAULT;
 219                         goto out;
 220                 }
 221
 222                 if (bprm->p < len)  {
 223                         ret = -E2BIG;
 224                         goto out;
 225                 }
 226
 227                 bprm->p -= len;
 228                 /* XXX: add architecture specific overflow check here. */
 229                 pos = bprm->p;
 230
 231                 while (len > 0) {
 232                         int i, new, err;
 233                         int offset, bytes_to_copy;
 234                         struct page *page;
 235
 236                         offset = pos % PAGE_SIZE;
 237                         i = pos/PAGE_SIZE;
 238                         page = bprm->page[i];
 239                         new = 0;
 240                         if (!page) {
 241                                 page = alloc_page(GFP_HIGHUSER);
 242                                 bprm->page[i] = page;
 243                                 if (!page) {
 244                                         ret = -ENOMEM;
 245                                         goto out;
 246                                 }
 247                                 new = 1;
 248                         }
 249
 250                         if (page != kmapped_page) {
 251                                 if (kmapped_page)
 252                                         kunmap(kmapped_page);
 253                                 kmapped_page = page;
 254                                 kaddr = kmap(kmapped_page);
 255                         }
 256                         if (new && offset)
 257                                 memset(kaddr, 0, offset);
 258                         bytes_to_copy = PAGE_SIZE - offset;
 259                         if (bytes_to_copy > len) {
 260                                 bytes_to_copy = len;
 261                                 if (new)
 262                                         memset(kaddr+offset+len, 0,
 263                                                 PAGE_SIZE-offset-len);
 264                         }
 265                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 266                         if (err) {
 267                                 ret = -EFAULT;
 268                                 goto out;
 269                         }
 270
 271                         pos += bytes_to_copy;
 272                         str += bytes_to_copy;
 273                         len -= bytes_to_copy;
 274                 }
 275         }
 276         ret = 0;
 277 out:
 278         if (kmapped_page)
 279                 kunmap(kmapped_page);
 280         return ret;
 281 }
 282
 283 /*
 284  * Like copy_strings, but get argv and its values from kernel memory.
 285  */
 286 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 287 {
 288         int r;
 289         mm_segment_t oldfs = get_fs();
 290         set_fs(KERNEL_DS);
 291         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 292         set_fs(oldfs);
 293         return r;
 294 }
 295
 296 EXPORT_SYMBOL(copy_strings_kernel);
 297
 298 #ifdef CONFIG_MMU
 299 /*
 300  * This routine is used to map in a page into an address space: needed by
 301  * execve() for the initial stack and environment pages.
 302  *
 303  * vma->vm_mm->mmap_sem is held for writing.
 304  */
 305 void install_arg_page(struct vm_area_struct *vma,
 306                         struct page *page, unsigned long address)
 307 {
 308         struct mm_struct *mm = vma->vm_mm;
 309         pte_t * pte;
 310         spinlock_t *ptl;
 311
 312         if (unlikely(anon_vma_prepare(vma)))
 313                 goto out;
 314
 315         flush_dcache_page(page);
 316         pte = get_locked_pte(mm, address, &ptl);
 317         if (!pte)
 318                 goto out;
 319         if (!pte_none(*pte)) {
 320                 pte_unmap_unlock(pte, ptl);
 321                 goto out;
 322         }
 323         inc_mm_counter(mm, anon_rss);
 324         lru_cache_add_active(page);
 325         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
 326                                         page, vma->vm_page_prot))));
 327         page_add_new_anon_rmap(page, vma, address);
 328         pte_unmap_unlock(pte, ptl);
 329
 330         /* no need for flush_tlb */
 331         return;
 332 out:
 333         __free_page(page);
 334         force_sig(SIGKILL, current);
 335 }
 336
 337 #define EXTRA_STACK_VM_PAGES    20      /* random */
 338
 339 int setup_arg_pages(struct linux_binprm *bprm,
 340                     unsigned long stack_top,
 341                     int executable_stack)
 342 {
 343         unsigned long stack_base;
 344         struct vm_area_struct *mpnt;
 345         struct mm_struct *mm = current->mm;
 346         int i, ret;
 347         long arg_size;
 348
 349 #ifdef CONFIG_STACK_GROWSUP
 350         /* Move the argument and environment strings to the bottom of the
 351          * stack space.
 352          */
 353         int offset, j;
 354         char *to, *from;
 355
 356         /* Start by shifting all the pages down */
 357         i = 0;
 358         for (j = 0; j < MAX_ARG_PAGES; j++) {
 359                 struct page *page = bprm->page[j];
 360                 if (!page)
 361                         continue;
 362                 bprm->page[i++] = page;
 363         }
 364
 365         /* Now move them within their pages */
 366         offset = bprm->p % PAGE_SIZE;
 367         to = kmap(bprm->page[0]);
 368         for (j = 1; j < i; j++) {
 369                 memmove(to, to + offset, PAGE_SIZE - offset);
 370                 from = kmap(bprm->page[j]);
 371                 memcpy(to + PAGE_SIZE - offset, from, offset);
 372                 kunmap(bprm->page[j - 1]);
 373                 to = from;
 374         }
 375         memmove(to, to + offset, PAGE_SIZE - offset);
 376         kunmap(bprm->page[j - 1]);
 377
 378         /* Limit stack size to 1GB */
 379         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 380         if (stack_base > (1 << 30))
 381                 stack_base = 1 << 30;
 382         stack_base = PAGE_ALIGN(stack_top - stack_base);
 383
 384         /* Adjust bprm->p to point to the end of the strings. */
 385         bprm->p = stack_base + PAGE_SIZE * i - offset;
 386
 387         mm->arg_start = stack_base;
 388         arg_size = i << PAGE_SHIFT;
 389
 390         /* zero pages that were copied above */
 391         while (i < MAX_ARG_PAGES)
 392                 bprm->page[i++] = NULL;
 393 #else
 394         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
 395         stack_base = PAGE_ALIGN(stack_base);
 396         bprm->p += stack_base;
 397         mm->arg_start = bprm->p;
 398         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 399 #endif
 400
 401         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 402
 403         if (bprm->loader)
 404                 bprm->loader += stack_base;
 405         bprm->exec += stack_base;
 406
 407         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 408         if (!mpnt)
 409                 return -ENOMEM;
 410
 411         memset(mpnt, 0, sizeof(*mpnt));
 412
 413         down_write(&mm->mmap_sem);
 414         {
 415                 mpnt->vm_mm = mm;
 416 #ifdef CONFIG_STACK_GROWSUP
 417                 mpnt->vm_start = stack_base;
 418                 mpnt->vm_end = stack_base + arg_size;
 419 #else
 420                 mpnt->vm_end = stack_top;
 421                 mpnt->vm_start = mpnt->vm_end - arg_size;
 422 #endif
 423                 /* Adjust stack execute permissions; explicitly enable
 424                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 425                  * and leave alone (arch default) otherwise. */
 426                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 427                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 428                 else if (executable_stack == EXSTACK_DISABLE_X)
 429                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 430                 else
 431                         mpnt->vm_flags = VM_STACK_FLAGS;
 432                 mpnt->vm_flags |= mm->def_flags;
 433                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 434                 if ((ret = insert_vm_struct(mm, mpnt))) {
 435                         up_write(&mm->mmap_sem);
 436                         kmem_cache_free(vm_area_cachep, mpnt);
 437                         return ret;
 438                 }
 439                 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 440         }
 441
 442         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 443                 struct page *page = bprm->page[i];
 444                 if (page) {
 445                         bprm->page[i] = NULL;
 446                         install_arg_page(mpnt, page, stack_base);
 447                 }
 448                 stack_base += PAGE_SIZE;
 449         }
 450         up_write(&mm->mmap_sem);
 451
 452         return 0;
 453 }
 454
 455 EXPORT_SYMBOL(setup_arg_pages);
 456
 457 #define free_arg_pages(bprm) do { } while (0)
 458
 459 #else
 460
 461 static inline void free_arg_pages(struct linux_binprm *bprm)
 462 {
 463         int i;
 464
 465         for (i = 0; i < MAX_ARG_PAGES; i++) {
 466                 if (bprm->page[i])
 467                         __free_page(bprm->page[i]);
 468                 bprm->page[i] = NULL;
 469         }
 470 }
 471
 472 #endif /* CONFIG_MMU */
 473
 474 struct file *open_exec(const char *name)
 475 {
 476         struct nameidata nd;
 477         int err;
 478         struct file *file;
 479
 480         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 481         file = ERR_PTR(err);
 482
 483         if (!err) {
 484                 struct inode *inode = nd.dentry->d_inode;
 485                 file = ERR_PTR(-EACCES);
 486                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 487                     S_ISREG(inode->i_mode)) {
 488                         int err = vfs_permission(&nd, MAY_EXEC);
 489                         if (!err && !(inode->i_mode & 0111))
 490                                 err = -EACCES;
 491                         file = ERR_PTR(err);
 492                         if (!err) {
 493                                 file = nameidata_to_filp(&nd, O_RDONLY);
 494                                 if (!IS_ERR(file)) {
 495                                         err = deny_write_access(file);
 496                                         if (err) {
 497                                                 fput(file);
 498                                                 file = ERR_PTR(err);
 499                                         }
 500                                 }
 501 out:
 502                                 return file;
 503                         }
 504                 }
 505                 release_open_intent(&nd);
 506                 path_release(&nd);
 507         }
 508         goto out;
 509 }
 510
 511 EXPORT_SYMBOL(open_exec);
 512
 513 int kernel_read(struct file *file, unsigned long offset,
 514         char *addr, unsigned long count)
 515 {
 516         mm_segment_t old_fs;
 517         loff_t pos = offset;
 518         int result;
 519
 520         old_fs = get_fs();
 521         set_fs(get_ds());
 522         /* The cast to a user pointer is valid due to the set_fs() */
 523         result = vfs_read(file, (void __user *)addr, count, &pos);
 524         set_fs(old_fs);
 525         return result;
 526 }
 527
 528 EXPORT_SYMBOL(kernel_read);
 529
 530 static int exec_mmap(struct mm_struct *mm)
 531 {
 532         struct task_struct *tsk;
 533         struct mm_struct * old_mm, *active_mm;
 534
 535         /* Notify parent that we're no longer interested in the old VM */
 536         tsk = current;
 537         old_mm = current->mm;
 538         mm_release(tsk, old_mm);
 539
 540         if (old_mm) {
 541                 /*
 542                  * Make sure that if there is a core dump in progress
 543                  * for the old mm, we get out and die instead of going
 544                  * through with the exec.  We must hold mmap_sem around
 545                  * checking core_waiters and changing tsk->mm.  The
 546                  * core-inducing thread will increment core_waiters for
 547                  * each thread whose ->mm == old_mm.
 548                  */
 549                 down_read(&old_mm->mmap_sem);
 550                 if (unlikely(old_mm->core_waiters)) {
 551                         up_read(&old_mm->mmap_sem);
 552                         return -EINTR;
 553                 }
 554         }
 555         task_lock(tsk);
 556         active_mm = tsk->active_mm;
 557         tsk->mm = mm;
 558         tsk->active_mm = mm;
 559         activate_mm(active_mm, mm);
 560         task_unlock(tsk);
 561         arch_pick_mmap_layout(mm);
 562         if (old_mm) {
 563                 up_read(&old_mm->mmap_sem);
 564                 BUG_ON(active_mm != old_mm);
 565                 mmput(old_mm);
 566                 return 0;
 567         }
 568         mmdrop(active_mm);
 569         return 0;
 570 }
 571
 572 /*
 573  * This function makes sure the current process has its own signal table,
 574  * so that flush_signal_handlers can later reset the handlers without
 575  * disturbing other processes.  (Other processes might share the signal
 576  * table via the CLONE_SIGHAND option to clone().)
 577  */
 578 static int de_thread(struct task_struct *tsk)
 579 {
 580         struct signal_struct *sig = tsk->signal;
 581         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 582         spinlock_t *lock = &oldsighand->siglock;
 583         struct task_struct *leader = NULL;
 584         int count;
 585
 586         /*
 587          * If we don't share sighandlers, then we aren't sharing anything
 588          * and we can just re-use it all.
 589          */
 590         if (atomic_read(&oldsighand->count) <= 1) {
 591                 BUG_ON(atomic_read(&sig->count) != 1);
 592                 exit_itimers(sig);
 593                 return 0;
 594         }
 595
 596         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 597         if (!newsighand)
 598                 return -ENOMEM;
 599
 600         if (thread_group_empty(current))
 601                 goto no_thread_group;
 602
 603         /*
 604          * Kill all other threads in the thread group.
 605          * We must hold tasklist_lock to call zap_other_threads.
 606          */
 607         read_lock(&tasklist_lock);
 608         spin_lock_irq(lock);
 609         if (sig->flags & SIGNAL_GROUP_EXIT) {
 610                 /*
 611                  * Another group action in progress, just
 612                  * return so that the signal is processed.
 613                  */
 614                 spin_unlock_irq(lock);
 615                 read_unlock(&tasklist_lock);
 616                 kmem_cache_free(sighand_cachep, newsighand);
 617                 return -EAGAIN;
 618         }
 619
 620         /*
 621          * child_reaper ignores SIGKILL, change it now.
 622          * Reparenting needs write_lock on tasklist_lock,
 623          * so it is safe to do it under read_lock.
 624          */
 625         if (unlikely(current->group_leader == child_reaper))
 626                 child_reaper = current;
 627
 628         zap_other_threads(current);
 629         read_unlock(&tasklist_lock);
 630
 631         /*
 632          * Account for the thread group leader hanging around:
 633          */
 634         count = 1;
 635         if (!thread_group_leader(current)) {
 636                 count = 2;
 637                 /*
 638                  * The SIGALRM timer survives the exec, but needs to point
 639                  * at us as the new group leader now.  We have a race with
 640                  * a timer firing now getting the old leader, so we need to
 641                  * synchronize with any firing (by calling del_timer_sync)
 642                  * before we can safely let the old group leader die.
 643                  */
 644                 sig->tsk = current;
 645                 spin_unlock_irq(lock);
 646                 if (hrtimer_cancel(&sig->real_timer))
 647                         hrtimer_restart(&sig->real_timer);
 648                 spin_lock_irq(lock);
 649         }
 650         while (atomic_read(&sig->count) > count) {
 651                 sig->group_exit_task = current;
 652                 sig->notify_count = count;
 653                 __set_current_state(TASK_UNINTERRUPTIBLE);
 654                 spin_unlock_irq(lock);
 655                 schedule();
 656                 spin_lock_irq(lock);
 657         }
 658         sig->group_exit_task = NULL;
 659         sig->notify_count = 0;
 660         spin_unlock_irq(lock);
 661
 662         /*
 663          * At this point all other threads have exited, all we have to
 664          * do is to wait for the thread group leader to become inactive,
 665          * and to assume its PID:
 666          */
 667         if (!thread_group_leader(current)) {
 668                 struct task_struct *parent;
 669                 struct dentry *proc_dentry1, *proc_dentry2;
 670                 unsigned long ptrace;
 671
 672                 /*
 673                  * Wait for the thread group leader to be a zombie.
 674                  * It should already be zombie at this point, most
 675                  * of the time.
 676                  */
 677                 leader = current->group_leader;
 678                 while (leader->exit_state != EXIT_ZOMBIE)
 679                         yield();
 680
 681                 /*
 682                  * The only record we have of the real-time age of a
 683                  * process, regardless of execs it's done, is start_time.
 684                  * All the past CPU time is accumulated in signal_struct
 685                  * from sister threads now dead.  But in this non-leader
 686                  * exec, nothing survives from the original leader thread,
 687                  * whose birth marks the true age of this process now.
 688                  * When we take on its identity by switching to its PID, we
 689                  * also take its birthdate (always earlier than our own).
 690                  */
 691                 current->start_time = leader->start_time;
 692
 693                 spin_lock(&leader->proc_lock);
 694                 spin_lock(&current->proc_lock);
 695                 proc_dentry1 = proc_pid_unhash(current);
 696                 proc_dentry2 = proc_pid_unhash(leader);
 697                 write_lock_irq(&tasklist_lock);
 698
 699                 BUG_ON(leader->tgid != current->tgid);
 700                 BUG_ON(current->pid == current->tgid);
 701                 /*
 702                  * An exec() starts a new thread group with the
 703                  * TGID of the previous thread group. Rehash the
 704                  * two threads with a switched PID, and release
 705                  * the former thread group leader:
 706                  */
 707                 ptrace = leader->ptrace;
 708                 parent = leader->parent;
 709                 if (unlikely(ptrace) && unlikely(parent == current)) {
 710                         /*
 711                          * Joker was ptracing his own group leader,
 712                          * and now he wants to be his own parent!
 713                          * We can't have that.
 714                          */
 715                         ptrace = 0;
 716                 }
 717
 718                 ptrace_unlink(current);
 719                 ptrace_unlink(leader);
 720                 remove_parent(current);
 721                 remove_parent(leader);
 722
 723
 724                 /* Become a process group leader with the old leader's pid.
 725                  * Note: The old leader also uses thispid until release_task
 726                  *       is called.  Odd but simple and correct.
 727                  */
 728                 detach_pid(current, PIDTYPE_PID);
 729                 current->pid = leader->pid;
 730                 attach_pid(current, PIDTYPE_PID,  current->pid);
 731                 attach_pid(current, PIDTYPE_PGID, current->signal->pgrp);
 732                 attach_pid(current, PIDTYPE_SID,  current->signal->session);
 733                 list_add_tail(&current->tasks, &init_task.tasks);
 734
 735                 current->parent = current->real_parent = leader->real_parent;
 736                 leader->parent = leader->real_parent = child_reaper;
 737                 current->group_leader = current;
 738                 leader->group_leader = current;
 739
 740                 /* Reduce leader to a thread */
 741                 detach_pid(leader, PIDTYPE_PGID);
 742                 detach_pid(leader, PIDTYPE_SID);
 743                 list_del_init(&leader->tasks);
 744
 745                 add_parent(current);
 746                 add_parent(leader);
 747                 if (ptrace) {
 748                         current->ptrace = ptrace;
 749                         __ptrace_link(current, parent);
 750                 }
 751
 752                 current->exit_signal = SIGCHLD;
 753
 754                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 755                 leader->exit_state = EXIT_DEAD;
 756
 757                 write_unlock_irq(&tasklist_lock);
 758                 spin_unlock(&leader->proc_lock);
 759                 spin_unlock(&current->proc_lock);
 760                 proc_pid_flush(proc_dentry1);
 761                 proc_pid_flush(proc_dentry2);
 762         }
 763
 764         /*
 765          * There may be one thread left which is just exiting,
 766          * but it's safe to stop telling the group to kill themselves.
 767          */
 768         sig->flags = 0;
 769
 770 no_thread_group:
 771         exit_itimers(sig);
 772         if (leader)
 773                 release_task(leader);
 774
 775         BUG_ON(atomic_read(&sig->count) != 1);
 776
 777         if (atomic_read(&oldsighand->count) == 1) {
 778                 /*
 779                  * Now that we nuked the rest of the thread group,
 780                  * it turns out we are not sharing sighand any more either.
 781                  * So we can just keep it.
 782                  */
 783                 kmem_cache_free(sighand_cachep, newsighand);
 784         } else {
 785                 /*
 786                  * Move our state over to newsighand and switch it in.
 787                  */
 788                 atomic_set(&newsighand->count, 1);
 789                 memcpy(newsighand->action, oldsighand->action,
 790                        sizeof(newsighand->action));
 791
 792                 write_lock_irq(&tasklist_lock);
 793                 spin_lock(&oldsighand->siglock);
 794                 spin_lock(&newsighand->siglock);
 795
 796                 rcu_assign_pointer(current->sighand, newsighand);
 797                 recalc_sigpending();
 798
 799                 spin_unlock(&newsighand->siglock);
 800                 spin_unlock(&oldsighand->siglock);
 801                 write_unlock_irq(&tasklist_lock);
 802
 803                 if (atomic_dec_and_test(&oldsighand->count))
 804                         kmem_cache_free(sighand_cachep, oldsighand);
 805         }
 806
 807         BUG_ON(!thread_group_leader(current));
 808         return 0;
 809 }
 810
 811 /*
 812  * These functions flushes out all traces of the currently running executable
 813  * so that a new one can be started
 814  */
 815
 816 static void flush_old_files(struct files_struct * files)
 817 {
 818         long j = -1;
 819         struct fdtable *fdt;
 820
 821         spin_lock(&files->file_lock);
 822         for (;;) {
 823                 unsigned long set, i;
 824
 825                 j++;
 826                 i = j * __NFDBITS;
 827                 fdt = files_fdtable(files);
 828                 if (i >= fdt->max_fds || i >= fdt->max_fdset)
 829                         break;
 830                 set = fdt->close_on_exec->fds_bits[j];
 831                 if (!set)
 832                         continue;
 833                 fdt->close_on_exec->fds_bits[j] = 0;
 834                 spin_unlock(&files->file_lock);
 835                 for ( ; set ; i++,set >>= 1) {
 836                         if (set & 1) {
 837                                 sys_close(i);
 838                         }
 839                 }
 840                 spin_lock(&files->file_lock);
 841
 842         }
 843         spin_unlock(&files->file_lock);
 844 }
 845
 846 void get_task_comm(char *buf, struct task_struct *tsk)
 847 {
 848         /* buf must be at least sizeof(tsk->comm) in size */
 849         task_lock(tsk);
 850         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 851         task_unlock(tsk);
 852 }
 853
 854 void set_task_comm(struct task_struct *tsk, char *buf)
 855 {
 856         task_lock(tsk);
 857         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 858         task_unlock(tsk);
 859 }
 860
 861 int flush_old_exec(struct linux_binprm * bprm)
 862 {
 863         char * name;
 864         int i, ch, retval;
 865         struct files_struct *files;
 866         char tcomm[sizeof(current->comm)];
 867
 868         /*
 869          * Make sure we have a private signal table and that
 870          * we are unassociated from the previous thread group.
 871          */
 872         retval = de_thread(current);
 873         if (retval)
 874                 goto out;
 875
 876         /*
 877          * Make sure we have private file handles. Ask the
 878          * fork helper to do the work for us and the exit
 879          * helper to do the cleanup of the old one.
 880          */
 881         files = current->files;         /* refcounted so safe to hold */
 882         retval = unshare_files();
 883         if (retval)
 884                 goto out;
 885         /*
 886          * Release all of the old mmap stuff
 887          */
 888         retval = exec_mmap(bprm->mm);
 889         if (retval)
 890                 goto mmap_failed;
 891
 892         bprm->mm = NULL;                /* We're using it now */
 893
 894         /* This is the point of no return */
 895         steal_locks(files);
 896         put_files_struct(files);
 897
 898         current->sas_ss_sp = current->sas_ss_size = 0;
 899
 900         if (current->euid == current->uid && current->egid == current->gid)
 901                 current->mm->dumpable = 1;
 902         else
 903                 current->mm->dumpable = suid_dumpable;
 904
 905         name = bprm->filename;
 906
 907         /* Copies the binary name from after last slash */
 908         for (i=0; (ch = *(name++)) != '\0';) {
 909                 if (ch == '/')
 910                         i = 0; /* overwrite what we wrote */
 911                 else
 912                         if (i < (sizeof(tcomm) - 1))
 913                                 tcomm[i++] = ch;
 914         }
 915         tcomm[i] = '\0';
 916         set_task_comm(current, tcomm);
 917
 918         current->flags &= ~PF_RANDOMIZE;
 919         flush_thread();
 920
 921         /* Set the new mm task size. We have to do that late because it may
 922          * depend on TIF_32BIT which is only updated in flush_thread() on
 923          * some architectures like powerpc
 924          */
 925         current->mm->task_size = TASK_SIZE;
 926
 927         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 928             file_permission(bprm->file, MAY_READ) ||
 929             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 930                 suid_keys(current);
 931                 current->mm->dumpable = suid_dumpable;
 932         }
 933
 934         /* An exec changes our domain. We are no longer part of the thread
 935            group */
 936
 937         current->self_exec_id++;
 938
 939         flush_signal_handlers(current, 0);
 940         flush_old_files(current->files);
 941
 942         return 0;
 943
 944 mmap_failed:
 945         put_files_struct(current->files);
 946         current->files = files;
 947 out:
 948         return retval;
 949 }
 950
 951 EXPORT_SYMBOL(flush_old_exec);
 952
 953 /*
 954  * Fill the binprm structure from the inode.
 955  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 956  */
 957 int prepare_binprm(struct linux_binprm *bprm)
 958 {
 959         int mode;
 960         struct inode * inode = bprm->file->f_dentry->d_inode;
 961         int retval;
 962
 963         mode = inode->i_mode;
 964         /*
 965          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 966          * generic_permission lets a non-executable through
 967          */
 968         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 969                 return -EACCES;
 970         if (bprm->file->f_op == NULL)
 971                 return -EACCES;
 972
 973         bprm->e_uid = current->euid;
 974         bprm->e_gid = current->egid;
 975
 976         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 977                 /* Set-uid? */
 978                 if (mode & S_ISUID) {
 979                         current->personality &= ~PER_CLEAR_ON_SETID;
 980                         bprm->e_uid = inode->i_uid;
 981                 }
 982
 983                 /* Set-gid? */
 984                 /*
 985                  * If setgid is set but no group execute bit then this
 986                  * is a candidate for mandatory locking, not a setgid
 987                  * executable.
 988                  */
 989                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 990                         current->personality &= ~PER_CLEAR_ON_SETID;
 991                         bprm->e_gid = inode->i_gid;
 992                 }
 993         }
 994
 995         /* fill in binprm security blob */
 996         retval = security_bprm_set(bprm);
 997         if (retval)
 998                 return retval;
 999
1000         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1001         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1002 }
1003
1004 EXPORT_SYMBOL(prepare_binprm);
1005
1006 static int unsafe_exec(struct task_struct *p)
1007 {
1008         int unsafe = 0;
1009         if (p->ptrace & PT_PTRACED) {
1010                 if (p->ptrace & PT_PTRACE_CAP)
1011                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1012                 else
1013                         unsafe |= LSM_UNSAFE_PTRACE;
1014         }
1015         if (atomic_read(&p->fs->count) > 1 ||
1016             atomic_read(&p->files->count) > 1 ||
1017             atomic_read(&p->sighand->count) > 1)
1018                 unsafe |= LSM_UNSAFE_SHARE;
1019
1020         return unsafe;
1021 }
1022
1023 void compute_creds(struct linux_binprm *bprm)
1024 {
1025         int unsafe;
1026
1027         if (bprm->e_uid != current->uid)
1028                 suid_keys(current);
1029         exec_keys(current);
1030
1031         task_lock(current);
1032         unsafe = unsafe_exec(current);
1033         security_bprm_apply_creds(bprm, unsafe);
1034         task_unlock(current);
1035         security_bprm_post_apply_creds(bprm);
1036 }
1037
1038 EXPORT_SYMBOL(compute_creds);
1039
1040 void remove_arg_zero(struct linux_binprm *bprm)
1041 {
1042         if (bprm->argc) {
1043                 unsigned long offset;
1044                 char * kaddr;
1045                 struct page *page;
1046
1047                 offset = bprm->p % PAGE_SIZE;
1048                 goto inside;
1049
1050                 while (bprm->p++, *(kaddr+offset++)) {
1051                         if (offset != PAGE_SIZE)
1052                                 continue;
1053                         offset = 0;
1054                         kunmap_atomic(kaddr, KM_USER0);
1055 inside:
1056                         page = bprm->page[bprm->p/PAGE_SIZE];
1057                         kaddr = kmap_atomic(page, KM_USER0);
1058                 }
1059                 kunmap_atomic(kaddr, KM_USER0);
1060                 bprm->argc--;
1061         }
1062 }
1063
1064 EXPORT_SYMBOL(remove_arg_zero);
1065
1066 /*
1067  * cycle the list of binary formats handler, until one recognizes the image
1068  */
1069 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1070 {
1071         int try,retval;
1072         struct linux_binfmt *fmt;
1073 #ifdef __alpha__
1074         /* handle /sbin/loader.. */
1075         {
1076             struct exec * eh = (struct exec *) bprm->buf;
1077
1078             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1079                 (eh->fh.f_flags & 0x3000) == 0x3000)
1080             {
1081                 struct file * file;
1082                 unsigned long loader;
1083
1084                 allow_write_access(bprm->file);
1085                 fput(bprm->file);
1086                 bprm->file = NULL;
1087
1088                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1089
1090                 file = open_exec("/sbin/loader");
1091                 retval = PTR_ERR(file);
1092                 if (IS_ERR(file))
1093                         return retval;
1094
1095                 /* Remember if the application is TASO.  */
1096                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1097
1098                 bprm->file = file;
1099                 bprm->loader = loader;
1100                 retval = prepare_binprm(bprm);
1101                 if (retval<0)
1102                         return retval;
1103                 /* should call search_binary_handler recursively here,
1104                    but it does not matter */
1105             }
1106         }
1107 #endif
1108         retval = security_bprm_check(bprm);
1109         if (retval)
1110                 return retval;
1111
1112         /* kernel module loader fixup */
1113         /* so we don't try to load run modprobe in kernel space. */
1114         set_fs(USER_DS);
1115         retval = -ENOENT;
1116         for (try=0; try<2; try++) {
1117                 read_lock(&binfmt_lock);
1118                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1119                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1120                         if (!fn)
1121                                 continue;
1122                         if (!try_module_get(fmt->module))
1123                                 continue;
1124                         read_unlock(&binfmt_lock);
1125                         retval = fn(bprm, regs);
1126                         if (retval >= 0) {
1127                                 put_binfmt(fmt);
1128                                 allow_write_access(bprm->file);
1129                                 if (bprm->file)
1130                                         fput(bprm->file);
1131                                 bprm->file = NULL;
1132                                 current->did_exec = 1;
1133                                 proc_exec_connector(current);
1134                                 return retval;
1135                         }
1136                         read_lock(&binfmt_lock);
1137                         put_binfmt(fmt);
1138                         if (retval != -ENOEXEC || bprm->mm == NULL)
1139                                 break;
1140                         if (!bprm->file) {
1141                                 read_unlock(&binfmt_lock);
1142                                 return retval;
1143                         }
1144                 }
1145                 read_unlock(&binfmt_lock);
1146                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1147                         break;
1148 #ifdef CONFIG_KMOD
1149                 }else{
1150 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1151                         if (printable(bprm->buf[0]) &&
1152                             printable(bprm->buf[1]) &&
1153                             printable(bprm->buf[2]) &&
1154                             printable(bprm->buf[3]))
1155                                 break; /* -ENOEXEC */
1156                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1157 #endif
1158                 }
1159         }
1160         return retval;
1161 }
1162
1163 EXPORT_SYMBOL(search_binary_handler);
1164
1165 /*
1166  * sys_execve() executes a new program.
1167  */
1168 int do_execve(char * filename,
1169         char __user *__user *argv,
1170         char __user *__user *envp,
1171         struct pt_regs * regs)
1172 {
1173         struct linux_binprm *bprm;
1174         struct file *file;
1175         int retval;
1176         int i;
1177
1178         retval = -ENOMEM;
1179         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1180         if (!bprm)
1181                 goto out_ret;
1182
1183         file = open_exec(filename);
1184         retval = PTR_ERR(file);
1185         if (IS_ERR(file))
1186                 goto out_kfree;
1187
1188         sched_exec();
1189
1190         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1191
1192         bprm->file = file;
1193         bprm->filename = filename;
1194         bprm->interp = filename;
1195         bprm->mm = mm_alloc();
1196         retval = -ENOMEM;
1197         if (!bprm->mm)
1198                 goto out_file;
1199
1200         retval = init_new_context(current, bprm->mm);
1201         if (retval < 0)
1202                 goto out_mm;
1203
1204         bprm->argc = count(argv, bprm->p / sizeof(void *));
1205         if ((retval = bprm->argc) < 0)
1206                 goto out_mm;
1207
1208         bprm->envc = count(envp, bprm->p / sizeof(void *));
1209         if ((retval = bprm->envc) < 0)
1210                 goto out_mm;
1211
1212         retval = security_bprm_alloc(bprm);
1213         if (retval)
1214                 goto out;
1215
1216         retval = prepare_binprm(bprm);
1217         if (retval < 0)
1218                 goto out;
1219
1220         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1221         if (retval < 0)
1222                 goto out;
1223
1224         bprm->exec = bprm->p;
1225         retval = copy_strings(bprm->envc, envp, bprm);
1226         if (retval < 0)
1227                 goto out;
1228
1229         retval = copy_strings(bprm->argc, argv, bprm);
1230         if (retval < 0)
1231                 goto out;
1232
1233         retval = search_binary_handler(bprm,regs);
1234         if (retval >= 0) {
1235                 free_arg_pages(bprm);
1236
1237                 /* execve success */
1238                 security_bprm_free(bprm);
1239                 acct_update_integrals(current);
1240                 kfree(bprm);
1241                 return retval;
1242         }
1243
1244 out:
1245         /* Something went wrong, return the inode and free the argument pages*/
1246         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1247                 struct page * page = bprm->page[i];
1248                 if (page)
1249                         __free_page(page);
1250         }
1251
1252         if (bprm->security)
1253                 security_bprm_free(bprm);
1254
1255 out_mm:
1256         if (bprm->mm)
1257                 mmdrop(bprm->mm);
1258
1259 out_file:
1260         if (bprm->file) {
1261                 allow_write_access(bprm->file);
1262                 fput(bprm->file);
1263         }
1264
1265 out_kfree:
1266         kfree(bprm);
1267
1268 out_ret:
1269         return retval;
1270 }
1271
1272 int set_binfmt(struct linux_binfmt *new)
1273 {
1274         struct linux_binfmt *old = current->binfmt;
1275
1276         if (new) {
1277                 if (!try_module_get(new->module))
1278                         return -1;
1279         }
1280         current->binfmt = new;
1281         if (old)
1282                 module_put(old->module);
1283         return 0;
1284 }
1285
1286 EXPORT_SYMBOL(set_binfmt);
1287
1288 #define CORENAME_MAX_SIZE 64
1289
1290 /* format_corename will inspect the pattern parameter, and output a
1291  * name into corename, which must have space for at least
1292  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1293  */
1294 static void format_corename(char *corename, const char *pattern, long signr)
1295 {
1296         const char *pat_ptr = pattern;
1297         char *out_ptr = corename;
1298         char *const out_end = corename + CORENAME_MAX_SIZE;
1299         int rc;
1300         int pid_in_pattern = 0;
1301
1302         /* Repeat as long as we have more pattern to process and more output
1303            space */
1304         while (*pat_ptr) {
1305                 if (*pat_ptr != '%') {
1306                         if (out_ptr == out_end)
1307                                 goto out;
1308                         *out_ptr++ = *pat_ptr++;
1309                 } else {
1310                         switch (*++pat_ptr) {
1311                         case 0:
1312                                 goto out;
1313                         /* Double percent, output one percent */
1314                         case '%':
1315                                 if (out_ptr == out_end)
1316                                         goto out;
1317                                 *out_ptr++ = '%';
1318                                 break;
1319                         /* pid */
1320                         case 'p':
1321                                 pid_in_pattern = 1;
1322                                 rc = snprintf(out_ptr, out_end - out_ptr,
1323                                               "%d", current->tgid);
1324                                 if (rc > out_end - out_ptr)
1325                                         goto out;
1326                                 out_ptr += rc;
1327                                 break;
1328                         /* uid */
1329                         case 'u':
1330                                 rc = snprintf(out_ptr, out_end - out_ptr,
1331                                               "%d", current->uid);
1332                                 if (rc > out_end - out_ptr)
1333                                         goto out;
1334                                 out_ptr += rc;
1335                                 break;
1336                         /* gid */
1337                         case 'g':
1338                                 rc = snprintf(out_ptr, out_end - out_ptr,
1339                                               "%d", current->gid);
1340                                 if (rc > out_end - out_ptr)
1341                                         goto out;
1342                                 out_ptr += rc;
1343                                 break;
1344                         /* signal that caused the coredump */
1345                         case 's':
1346                                 rc = snprintf(out_ptr, out_end - out_ptr,
1347                                               "%ld", signr);
1348                                 if (rc > out_end - out_ptr)
1349                                         goto out;
1350                                 out_ptr += rc;
1351                                 break;
1352                         /* UNIX time of coredump */
1353                         case 't': {
1354                                 struct timeval tv;
1355                                 do_gettimeofday(&tv);
1356                                 rc = snprintf(out_ptr, out_end - out_ptr,
1357                                               "%lu", tv.tv_sec);
1358                                 if (rc > out_end - out_ptr)
1359                                         goto out;
1360                                 out_ptr += rc;
1361                                 break;
1362                         }
1363                         /* hostname */
1364                         case 'h':
1365                                 down_read(&uts_sem);
1366                                 rc = snprintf(out_ptr, out_end - out_ptr,
1367                                               "%s", system_utsname.nodename);
1368                                 up_read(&uts_sem);
1369                                 if (rc > out_end - out_ptr)
1370                                         goto out;
1371                                 out_ptr += rc;
1372                                 break;
1373                         /* executable */
1374                         case 'e':
1375                                 rc = snprintf(out_ptr, out_end - out_ptr,
1376                                               "%s", current->comm);
1377                                 if (rc > out_end - out_ptr)
1378                                         goto out;
1379                                 out_ptr += rc;
1380                                 break;
1381                         default:
1382                                 break;
1383                         }
1384                         ++pat_ptr;
1385                 }
1386         }
1387         /* Backward compatibility with core_uses_pid:
1388          *
1389          * If core_pattern does not include a %p (as is the default)
1390          * and core_uses_pid is set, then .%pid will be appended to
1391          * the filename */
1392         if (!pid_in_pattern
1393             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1394                 rc = snprintf(out_ptr, out_end - out_ptr,
1395                               ".%d", current->tgid);
1396                 if (rc > out_end - out_ptr)
1397                         goto out;
1398                 out_ptr += rc;
1399         }
1400       out:
1401         *out_ptr = 0;
1402 }
1403
1404 static void zap_threads (struct mm_struct *mm)
1405 {
1406         struct task_struct *g, *p;
1407         struct task_struct *tsk = current;
1408         struct completion *vfork_done = tsk->vfork_done;
1409         int traced = 0;
1410
1411         /*
1412          * Make sure nobody is waiting for us to release the VM,
1413          * otherwise we can deadlock when we wait on each other
1414          */
1415         if (vfork_done) {
1416                 tsk->vfork_done = NULL;
1417                 complete(vfork_done);
1418         }
1419
1420         read_lock(&tasklist_lock);
1421         do_each_thread(g,p)
1422                 if (mm == p->mm && p != tsk) {
1423                         force_sig_specific(SIGKILL, p);
1424                         mm->core_waiters++;
1425                         if (unlikely(p->ptrace) &&
1426                             unlikely(p->parent->mm == mm))
1427                                 traced = 1;
1428                 }
1429         while_each_thread(g,p);
1430
1431         read_unlock(&tasklist_lock);
1432
1433         if (unlikely(traced)) {
1434                 /*
1435                  * We are zapping a thread and the thread it ptraces.
1436                  * If the tracee went into a ptrace stop for exit tracing,
1437                  * we could deadlock since the tracer is waiting for this
1438                  * coredump to finish.  Detach them so they can both die.
1439                  */
1440                 write_lock_irq(&tasklist_lock);
1441                 do_each_thread(g,p) {
1442                         if (mm == p->mm && p != tsk &&
1443                             p->ptrace && p->parent->mm == mm) {
1444                                 __ptrace_detach(p, 0);
1445                         }
1446                 } while_each_thread(g,p);
1447                 write_unlock_irq(&tasklist_lock);
1448         }
1449 }
1450
1451 static void coredump_wait(struct mm_struct *mm)
1452 {
1453         DECLARE_COMPLETION(startup_done);
1454         int core_waiters;
1455
1456         mm->core_startup_done = &startup_done;
1457
1458         zap_threads(mm);
1459         core_waiters = mm->core_waiters;
1460         up_write(&mm->mmap_sem);
1461
1462         if (core_waiters)
1463                 wait_for_completion(&startup_done);
1464         BUG_ON(mm->core_waiters);
1465 }
1466
1467 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1468 {
1469         char corename[CORENAME_MAX_SIZE + 1];
1470         struct mm_struct *mm = current->mm;
1471         struct linux_binfmt * binfmt;
1472         struct inode * inode;
1473         struct file * file;
1474         int retval = 0;
1475         int fsuid = current->fsuid;
1476         int flag = 0;
1477
1478         binfmt = current->binfmt;
1479         if (!binfmt || !binfmt->core_dump)
1480                 goto fail;
1481         down_write(&mm->mmap_sem);
1482         if (!mm->dumpable) {
1483                 up_write(&mm->mmap_sem);
1484                 goto fail;
1485         }
1486
1487         /*
1488          *      We cannot trust fsuid as being the "true" uid of the
1489          *      process nor do we know its entire history. We only know it
1490          *      was tainted so we dump it as root in mode 2.
1491          */
1492         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1493                 flag = O_EXCL;          /* Stop rewrite attacks */
1494                 current->fsuid = 0;     /* Dump root private */
1495         }
1496         mm->dumpable = 0;
1497
1498         retval = -EAGAIN;
1499         spin_lock_irq(&current->sighand->siglock);
1500         if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1501                 current->signal->flags = SIGNAL_GROUP_EXIT;
1502                 current->signal->group_exit_code = exit_code;
1503                 current->signal->group_stop_count = 0;
1504                 retval = 0;
1505         }
1506         spin_unlock_irq(&current->sighand->siglock);
1507         if (retval) {
1508                 up_write(&mm->mmap_sem);
1509                 goto fail;
1510         }
1511
1512         init_completion(&mm->core_done);
1513         coredump_wait(mm);
1514
1515         /*
1516          * Clear any false indication of pending signals that might
1517          * be seen by the filesystem code called to write the core file.
1518          */
1519         clear_thread_flag(TIF_SIGPENDING);
1520
1521         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1522                 goto fail_unlock;
1523
1524         /*
1525          * lock_kernel() because format_corename() is controlled by sysctl, which
1526          * uses lock_kernel()
1527          */
1528         lock_kernel();
1529         format_corename(corename, core_pattern, signr);
1530         unlock_kernel();
1531         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1532         if (IS_ERR(file))
1533                 goto fail_unlock;
1534         inode = file->f_dentry->d_inode;
1535         if (inode->i_nlink > 1)
1536                 goto close_fail;        /* multiple links - don't dump */
1537         if (d_unhashed(file->f_dentry))
1538                 goto close_fail;
1539
1540         if (!S_ISREG(inode->i_mode))
1541                 goto close_fail;
1542         if (!file->f_op)
1543                 goto close_fail;
1544         if (!file->f_op->write)
1545                 goto close_fail;
1546         if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1547                 goto close_fail;
1548
1549         retval = binfmt->core_dump(signr, regs, file);
1550
1551         if (retval)
1552                 current->signal->group_exit_code |= 0x80;
1553 close_fail:
1554         filp_close(file, NULL);
1555 fail_unlock:
1556         current->fsuid = fsuid;
1557         complete_all(&mm->core_done);
1558 fail:
1559         return retval;
1560 }