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 #include <linux/audit.h>
  53
  54 #include <asm/uaccess.h>
  55 #include <asm/mmu_context.h>
  56
  57 #ifdef CONFIG_KMOD
  58 #include <linux/kmod.h>
  59 #endif
  60
  61 int core_uses_pid;
  62 char core_pattern[65] = "core";
  63 int suid_dumpable = 0;
  64
  65 EXPORT_SYMBOL(suid_dumpable);
  66 /* The maximal length of core_pattern is also specified in sysctl.c */
  67
  68 static struct linux_binfmt *formats;
  69 static DEFINE_RWLOCK(binfmt_lock);
  70
  71 int register_binfmt(struct linux_binfmt * fmt)
  72 {
  73         struct linux_binfmt ** tmp = &formats;
  74
  75         if (!fmt)
  76                 return -EINVAL;
  77         if (fmt->next)
  78                 return -EBUSY;
  79         write_lock(&binfmt_lock);
  80         while (*tmp) {
  81                 if (fmt == *tmp) {
  82                         write_unlock(&binfmt_lock);
  83                         return -EBUSY;
  84                 }
  85                 tmp = &(*tmp)->next;
  86         }
  87         fmt->next = formats;
  88         formats = fmt;
  89         write_unlock(&binfmt_lock);
  90         return 0;
  91 }
  92
  93 EXPORT_SYMBOL(register_binfmt);
  94
  95 int unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         struct linux_binfmt ** tmp = &formats;
  98
  99         write_lock(&binfmt_lock);
 100         while (*tmp) {
 101                 if (fmt == *tmp) {
 102                         *tmp = fmt->next;
 103                         write_unlock(&binfmt_lock);
 104                         return 0;
 105                 }
 106                 tmp = &(*tmp)->next;
 107         }
 108         write_unlock(&binfmt_lock);
 109         return -EINVAL;
 110 }
 111
 112 EXPORT_SYMBOL(unregister_binfmt);
 113
 114 static inline void put_binfmt(struct linux_binfmt * fmt)
 115 {
 116         module_put(fmt->module);
 117 }
 118
 119 /*
 120  * Note that a shared library must be both readable and executable due to
 121  * security reasons.
 122  *
 123  * Also note that we take the address to load from from the file itself.
 124  */
 125 asmlinkage long sys_uselib(const char __user * library)
 126 {
 127         struct file * file;
 128         struct nameidata nd;
 129         int error;
 130
 131         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 132         if (error)
 133                 goto out;
 134
 135         error = -EINVAL;
 136         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 137                 goto exit;
 138
 139         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 140         if (error)
 141                 goto exit;
 142
 143         file = nameidata_to_filp(&nd, O_RDONLY);
 144         error = PTR_ERR(file);
 145         if (IS_ERR(file))
 146                 goto out;
 147
 148         error = -ENOEXEC;
 149         if(file->f_op) {
 150                 struct linux_binfmt * fmt;
 151
 152                 read_lock(&binfmt_lock);
 153                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 154                         if (!fmt->load_shlib)
 155                                 continue;
 156                         if (!try_module_get(fmt->module))
 157                                 continue;
 158                         read_unlock(&binfmt_lock);
 159                         error = fmt->load_shlib(file);
 160                         read_lock(&binfmt_lock);
 161                         put_binfmt(fmt);
 162                         if (error != -ENOEXEC)
 163                                 break;
 164                 }
 165                 read_unlock(&binfmt_lock);
 166         }
 167         fput(file);
 168 out:
 169         return error;
 170 exit:
 171         release_open_intent(&nd);
 172         path_release(&nd);
 173         goto out;
 174 }
 175
 176 /*
 177  * count() counts the number of strings in array ARGV.
 178  */
 179 static int count(char __user * __user * argv, int max)
 180 {
 181         int i = 0;
 182
 183         if (argv != NULL) {
 184                 for (;;) {
 185                         char __user * p;
 186
 187                         if (get_user(p, argv))
 188                                 return -EFAULT;
 189                         if (!p)
 190                                 break;
 191                         argv++;
 192                         if(++i > max)
 193                                 return -E2BIG;
 194                         cond_resched();
 195                 }
 196         }
 197         return i;
 198 }
 199
 200 /*
 201  * 'copy_strings()' copies argument/environment strings from user
 202  * memory to free pages in kernel mem. These are in a format ready
 203  * to be put directly into the top of new user memory.
 204  */
 205 static int copy_strings(int argc, char __user * __user * argv,
 206                         struct linux_binprm *bprm)
 207 {
 208         struct page *kmapped_page = NULL;
 209         char *kaddr = NULL;
 210         int ret;
 211
 212         while (argc-- > 0) {
 213                 char __user *str;
 214                 int len;
 215                 unsigned long pos;
 216
 217                 if (get_user(str, argv+argc) ||
 218                                 !(len = strnlen_user(str, bprm->p))) {
 219                         ret = -EFAULT;
 220                         goto out;
 221                 }
 222
 223                 if (bprm->p < len)  {
 224                         ret = -E2BIG;
 225                         goto out;
 226                 }
 227
 228                 bprm->p -= len;
 229                 /* XXX: add architecture specific overflow check here. */
 230                 pos = bprm->p;
 231
 232                 while (len > 0) {
 233                         int i, new, err;
 234                         int offset, bytes_to_copy;
 235                         struct page *page;
 236
 237                         offset = pos % PAGE_SIZE;
 238                         i = pos/PAGE_SIZE;
 239                         page = bprm->page[i];
 240                         new = 0;
 241                         if (!page) {
 242                                 page = alloc_page(GFP_HIGHUSER);
 243                                 bprm->page[i] = page;
 244                                 if (!page) {
 245                                         ret = -ENOMEM;
 246                                         goto out;
 247                                 }
 248                                 new = 1;
 249                         }
 250
 251                         if (page != kmapped_page) {
 252                                 if (kmapped_page)
 253                                         kunmap(kmapped_page);
 254                                 kmapped_page = page;
 255                                 kaddr = kmap(kmapped_page);
 256                         }
 257                         if (new && offset)
 258                                 memset(kaddr, 0, offset);
 259                         bytes_to_copy = PAGE_SIZE - offset;
 260                         if (bytes_to_copy > len) {
 261                                 bytes_to_copy = len;
 262                                 if (new)
 263                                         memset(kaddr+offset+len, 0,
 264                                                 PAGE_SIZE-offset-len);
 265                         }
 266                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 267                         if (err) {
 268                                 ret = -EFAULT;
 269                                 goto out;
 270                         }
 271
 272                         pos += bytes_to_copy;
 273                         str += bytes_to_copy;
 274                         len -= bytes_to_copy;
 275                 }
 276         }
 277         ret = 0;
 278 out:
 279         if (kmapped_page)
 280                 kunmap(kmapped_page);
 281         return ret;
 282 }
 283
 284 /*
 285  * Like copy_strings, but get argv and its values from kernel memory.
 286  */
 287 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 288 {
 289         int r;
 290         mm_segment_t oldfs = get_fs();
 291         set_fs(KERNEL_DS);
 292         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 293         set_fs(oldfs);
 294         return r;
 295 }
 296
 297 EXPORT_SYMBOL(copy_strings_kernel);
 298
 299 #ifdef CONFIG_MMU
 300 /*
 301  * This routine is used to map in a page into an address space: needed by
 302  * execve() for the initial stack and environment pages.
 303  *
 304  * vma->vm_mm->mmap_sem is held for writing.
 305  */
 306 void install_arg_page(struct vm_area_struct *vma,
 307                         struct page *page, unsigned long address)
 308 {
 309         struct mm_struct *mm = vma->vm_mm;
 310         pte_t * pte;
 311         spinlock_t *ptl;
 312
 313         if (unlikely(anon_vma_prepare(vma)))
 314                 goto out;
 315
 316         flush_dcache_page(page);
 317         pte = get_locked_pte(mm, address, &ptl);
 318         if (!pte)
 319                 goto out;
 320         if (!pte_none(*pte)) {
 321                 pte_unmap_unlock(pte, ptl);
 322                 goto out;
 323         }
 324         inc_mm_counter(mm, anon_rss);
 325         lru_cache_add_active(page);
 326         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
 327                                         page, vma->vm_page_prot))));
 328         page_add_new_anon_rmap(page, vma, address);
 329         pte_unmap_unlock(pte, ptl);
 330
 331         /* no need for flush_tlb */
 332         return;
 333 out:
 334         __free_page(page);
 335         force_sig(SIGKILL, current);
 336 }
 337
 338 #define EXTRA_STACK_VM_PAGES    20      /* random */
 339
 340 int setup_arg_pages(struct linux_binprm *bprm,
 341                     unsigned long stack_top,
 342                     int executable_stack)
 343 {
 344         unsigned long stack_base;
 345         struct vm_area_struct *mpnt;
 346         struct mm_struct *mm = current->mm;
 347         int i, ret;
 348         long arg_size;
 349
 350 #ifdef CONFIG_STACK_GROWSUP
 351         /* Move the argument and environment strings to the bottom of the
 352          * stack space.
 353          */
 354         int offset, j;
 355         char *to, *from;
 356
 357         /* Start by shifting all the pages down */
 358         i = 0;
 359         for (j = 0; j < MAX_ARG_PAGES; j++) {
 360                 struct page *page = bprm->page[j];
 361                 if (!page)
 362                         continue;
 363                 bprm->page[i++] = page;
 364         }
 365
 366         /* Now move them within their pages */
 367         offset = bprm->p % PAGE_SIZE;
 368         to = kmap(bprm->page[0]);
 369         for (j = 1; j < i; j++) {
 370                 memmove(to, to + offset, PAGE_SIZE - offset);
 371                 from = kmap(bprm->page[j]);
 372                 memcpy(to + PAGE_SIZE - offset, from, offset);
 373                 kunmap(bprm->page[j - 1]);
 374                 to = from;
 375         }
 376         memmove(to, to + offset, PAGE_SIZE - offset);
 377         kunmap(bprm->page[j - 1]);
 378
 379         /* Limit stack size to 1GB */
 380         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 381         if (stack_base > (1 << 30))
 382                 stack_base = 1 << 30;
 383         stack_base = PAGE_ALIGN(stack_top - stack_base);
 384
 385         /* Adjust bprm->p to point to the end of the strings. */
 386         bprm->p = stack_base + PAGE_SIZE * i - offset;
 387
 388         mm->arg_start = stack_base;
 389         arg_size = i << PAGE_SHIFT;
 390
 391         /* zero pages that were copied above */
 392         while (i < MAX_ARG_PAGES)
 393                 bprm->page[i++] = NULL;
 394 #else
 395         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
 396         stack_base = PAGE_ALIGN(stack_base);
 397         bprm->p += stack_base;
 398         mm->arg_start = bprm->p;
 399         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 400 #endif
 401
 402         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 403
 404         if (bprm->loader)
 405                 bprm->loader += stack_base;
 406         bprm->exec += stack_base;
 407
 408         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 409         if (!mpnt)
 410                 return -ENOMEM;
 411
 412         memset(mpnt, 0, sizeof(*mpnt));
 413
 414         down_write(&mm->mmap_sem);
 415         {
 416                 mpnt->vm_mm = mm;
 417 #ifdef CONFIG_STACK_GROWSUP
 418                 mpnt->vm_start = stack_base;
 419                 mpnt->vm_end = stack_base + arg_size;
 420 #else
 421                 mpnt->vm_end = stack_top;
 422                 mpnt->vm_start = mpnt->vm_end - arg_size;
 423 #endif
 424                 /* Adjust stack execute permissions; explicitly enable
 425                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 426                  * and leave alone (arch default) otherwise. */
 427                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 428                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 429                 else if (executable_stack == EXSTACK_DISABLE_X)
 430                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 431                 else
 432                         mpnt->vm_flags = VM_STACK_FLAGS;
 433                 mpnt->vm_flags |= mm->def_flags;
 434                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 435                 if ((ret = insert_vm_struct(mm, mpnt))) {
 436                         up_write(&mm->mmap_sem);
 437                         kmem_cache_free(vm_area_cachep, mpnt);
 438                         return ret;
 439                 }
 440                 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 441         }
 442
 443         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 444                 struct page *page = bprm->page[i];
 445                 if (page) {
 446                         bprm->page[i] = NULL;
 447                         install_arg_page(mpnt, page, stack_base);
 448                 }
 449                 stack_base += PAGE_SIZE;
 450         }
 451         up_write(&mm->mmap_sem);
 452
 453         return 0;
 454 }
 455
 456 EXPORT_SYMBOL(setup_arg_pages);
 457
 458 #define free_arg_pages(bprm) do { } while (0)
 459
 460 #else
 461
 462 static inline void free_arg_pages(struct linux_binprm *bprm)
 463 {
 464         int i;
 465
 466         for (i = 0; i < MAX_ARG_PAGES; i++) {
 467                 if (bprm->page[i])
 468                         __free_page(bprm->page[i]);
 469                 bprm->page[i] = NULL;
 470         }
 471 }
 472
 473 #endif /* CONFIG_MMU */
 474
 475 struct file *open_exec(const char *name)
 476 {
 477         struct nameidata nd;
 478         int err;
 479         struct file *file;
 480
 481         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 482         file = ERR_PTR(err);
 483
 484         if (!err) {
 485                 struct inode *inode = nd.dentry->d_inode;
 486                 file = ERR_PTR(-EACCES);
 487                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 488                     S_ISREG(inode->i_mode)) {
 489                         int err = vfs_permission(&nd, MAY_EXEC);
 490                         if (!err && !(inode->i_mode & 0111))
 491                                 err = -EACCES;
 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(current))
 602                 goto no_thread_group;
 603
 604         /*
 605          * Kill all other threads in the thread group.
 606          * We must hold tasklist_lock to call zap_other_threads.
 607          */
 608         read_lock(&tasklist_lock);
 609         spin_lock_irq(lock);
 610         if (sig->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(current->group_leader == child_reaper))
 627                 child_reaper = current;
 628
 629         zap_other_threads(current);
 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(current)) {
 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 = current;
 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 = current;
 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(current)) {
 669                 struct dentry *proc_dentry1, *proc_dentry2;
 670
 671                 /*
 672                  * Wait for the thread group leader to be a zombie.
 673                  * It should already be zombie at this point, most
 674                  * of the time.
 675                  */
 676                 leader = current->group_leader;
 677                 while (leader->exit_state != EXIT_ZOMBIE)
 678                         yield();
 679
 680                 /*
 681                  * The only record we have of the real-time age of a
 682                  * process, regardless of execs it's done, is start_time.
 683                  * All the past CPU time is accumulated in signal_struct
 684                  * from sister threads now dead.  But in this non-leader
 685                  * exec, nothing survives from the original leader thread,
 686                  * whose birth marks the true age of this process now.
 687                  * When we take on its identity by switching to its PID, we
 688                  * also take its birthdate (always earlier than our own).
 689                  */
 690                 current->start_time = leader->start_time;
 691
 692                 spin_lock(&leader->proc_lock);
 693                 spin_lock(&current->proc_lock);
 694                 proc_dentry1 = proc_pid_unhash(current);
 695                 proc_dentry2 = proc_pid_unhash(leader);
 696                 write_lock_irq(&tasklist_lock);
 697
 698                 BUG_ON(leader->tgid != current->tgid);
 699                 BUG_ON(current->pid == current->tgid);
 700                 /*
 701                  * An exec() starts a new thread group with the
 702                  * TGID of the previous thread group. Rehash the
 703                  * two threads with a switched PID, and release
 704                  * the former thread group leader:
 705                  */
 706
 707                 /* Become a process group leader with the old leader's pid.
 708                  * Note: The old leader also uses thispid until release_task
 709                  *       is called.  Odd but simple and correct.
 710                  */
 711                 detach_pid(current, PIDTYPE_PID);
 712                 current->pid = leader->pid;
 713                 attach_pid(current, PIDTYPE_PID,  current->pid);
 714                 attach_pid(current, PIDTYPE_PGID, current->signal->pgrp);
 715                 attach_pid(current, PIDTYPE_SID,  current->signal->session);
 716                 list_add_tail_rcu(&current->tasks, &init_task.tasks);
 717
 718                 current->group_leader = current;
 719                 leader->group_leader = current;
 720
 721                 /* Reduce leader to a thread */
 722                 detach_pid(leader, PIDTYPE_PGID);
 723                 detach_pid(leader, PIDTYPE_SID);
 724                 list_del_init(&leader->tasks);
 725
 726                 current->exit_signal = SIGCHLD;
 727
 728                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 729                 leader->exit_state = EXIT_DEAD;
 730
 731                 write_unlock_irq(&tasklist_lock);
 732                 spin_unlock(&leader->proc_lock);
 733                 spin_unlock(&current->proc_lock);
 734                 proc_pid_flush(proc_dentry1);
 735                 proc_pid_flush(proc_dentry2);
 736         }
 737
 738         /*
 739          * There may be one thread left which is just exiting,
 740          * but it's safe to stop telling the group to kill themselves.
 741          */
 742         sig->flags = 0;
 743
 744 no_thread_group:
 745         exit_itimers(sig);
 746         if (leader)
 747                 release_task(leader);
 748
 749         BUG_ON(atomic_read(&sig->count) != 1);
 750
 751         if (atomic_read(&oldsighand->count) == 1) {
 752                 /*
 753                  * Now that we nuked the rest of the thread group,
 754                  * it turns out we are not sharing sighand any more either.
 755                  * So we can just keep it.
 756                  */
 757                 kmem_cache_free(sighand_cachep, newsighand);
 758         } else {
 759                 /*
 760                  * Move our state over to newsighand and switch it in.
 761                  */
 762                 atomic_set(&newsighand->count, 1);
 763                 memcpy(newsighand->action, oldsighand->action,
 764                        sizeof(newsighand->action));
 765
 766                 write_lock_irq(&tasklist_lock);
 767                 spin_lock(&oldsighand->siglock);
 768                 spin_lock(&newsighand->siglock);
 769
 770                 rcu_assign_pointer(current->sighand, newsighand);
 771                 recalc_sigpending();
 772
 773                 spin_unlock(&newsighand->siglock);
 774                 spin_unlock(&oldsighand->siglock);
 775                 write_unlock_irq(&tasklist_lock);
 776
 777                 if (atomic_dec_and_test(&oldsighand->count))
 778                         kmem_cache_free(sighand_cachep, oldsighand);
 779         }
 780
 781         BUG_ON(!thread_group_leader(current));
 782         return 0;
 783 }
 784
 785 /*
 786  * These functions flushes out all traces of the currently running executable
 787  * so that a new one can be started
 788  */
 789
 790 static void flush_old_files(struct files_struct * files)
 791 {
 792         long j = -1;
 793         struct fdtable *fdt;
 794
 795         spin_lock(&files->file_lock);
 796         for (;;) {
 797                 unsigned long set, i;
 798
 799                 j++;
 800                 i = j * __NFDBITS;
 801                 fdt = files_fdtable(files);
 802                 if (i >= fdt->max_fds || i >= fdt->max_fdset)
 803                         break;
 804                 set = fdt->close_on_exec->fds_bits[j];
 805                 if (!set)
 806                         continue;
 807                 fdt->close_on_exec->fds_bits[j] = 0;
 808                 spin_unlock(&files->file_lock);
 809                 for ( ; set ; i++,set >>= 1) {
 810                         if (set & 1) {
 811                                 sys_close(i);
 812                         }
 813                 }
 814                 spin_lock(&files->file_lock);
 815
 816         }
 817         spin_unlock(&files->file_lock);
 818 }
 819
 820 void get_task_comm(char *buf, struct task_struct *tsk)
 821 {
 822         /* buf must be at least sizeof(tsk->comm) in size */
 823         task_lock(tsk);
 824         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 825         task_unlock(tsk);
 826 }
 827
 828 void set_task_comm(struct task_struct *tsk, char *buf)
 829 {
 830         task_lock(tsk);
 831         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 832         task_unlock(tsk);
 833 }
 834
 835 int flush_old_exec(struct linux_binprm * bprm)
 836 {
 837         char * name;
 838         int i, ch, retval;
 839         struct files_struct *files;
 840         char tcomm[sizeof(current->comm)];
 841
 842         /*
 843          * Make sure we have a private signal table and that
 844          * we are unassociated from the previous thread group.
 845          */
 846         retval = de_thread(current);
 847         if (retval)
 848                 goto out;
 849
 850         /*
 851          * Make sure we have private file handles. Ask the
 852          * fork helper to do the work for us and the exit
 853          * helper to do the cleanup of the old one.
 854          */
 855         files = current->files;         /* refcounted so safe to hold */
 856         retval = unshare_files();
 857         if (retval)
 858                 goto out;
 859         /*
 860          * Release all of the old mmap stuff
 861          */
 862         retval = exec_mmap(bprm->mm);
 863         if (retval)
 864                 goto mmap_failed;
 865
 866         bprm->mm = NULL;                /* We're using it now */
 867
 868         /* This is the point of no return */
 869         put_files_struct(files);
 870
 871         current->sas_ss_sp = current->sas_ss_size = 0;
 872
 873         if (current->euid == current->uid && current->egid == current->gid)
 874                 current->mm->dumpable = 1;
 875         else
 876                 current->mm->dumpable = suid_dumpable;
 877
 878         name = bprm->filename;
 879
 880         /* Copies the binary name from after last slash */
 881         for (i=0; (ch = *(name++)) != '\0';) {
 882                 if (ch == '/')
 883                         i = 0; /* overwrite what we wrote */
 884                 else
 885                         if (i < (sizeof(tcomm) - 1))
 886                                 tcomm[i++] = ch;
 887         }
 888         tcomm[i] = '\0';
 889         set_task_comm(current, tcomm);
 890
 891         current->flags &= ~PF_RANDOMIZE;
 892         flush_thread();
 893
 894         /* Set the new mm task size. We have to do that late because it may
 895          * depend on TIF_32BIT which is only updated in flush_thread() on
 896          * some architectures like powerpc
 897          */
 898         current->mm->task_size = TASK_SIZE;
 899
 900         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 901             file_permission(bprm->file, MAY_READ) ||
 902             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 903                 suid_keys(current);
 904                 current->mm->dumpable = suid_dumpable;
 905         }
 906
 907         /* An exec changes our domain. We are no longer part of the thread
 908            group */
 909
 910         current->self_exec_id++;
 911
 912         flush_signal_handlers(current, 0);
 913         flush_old_files(current->files);
 914
 915         return 0;
 916
 917 mmap_failed:
 918         put_files_struct(current->files);
 919         current->files = files;
 920 out:
 921         return retval;
 922 }
 923
 924 EXPORT_SYMBOL(flush_old_exec);
 925
 926 /*
 927  * Fill the binprm structure from the inode.
 928  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 929  */
 930 int prepare_binprm(struct linux_binprm *bprm)
 931 {
 932         int mode;
 933         struct inode * inode = bprm->file->f_dentry->d_inode;
 934         int retval;
 935
 936         mode = inode->i_mode;
 937         /*
 938          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 939          * generic_permission lets a non-executable through
 940          */
 941         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 942                 return -EACCES;
 943         if (bprm->file->f_op == NULL)
 944                 return -EACCES;
 945
 946         bprm->e_uid = current->euid;
 947         bprm->e_gid = current->egid;
 948
 949         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 950                 /* Set-uid? */
 951                 if (mode & S_ISUID) {
 952                         current->personality &= ~PER_CLEAR_ON_SETID;
 953                         bprm->e_uid = inode->i_uid;
 954                 }
 955
 956                 /* Set-gid? */
 957                 /*
 958                  * If setgid is set but no group execute bit then this
 959                  * is a candidate for mandatory locking, not a setgid
 960                  * executable.
 961                  */
 962                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 963                         current->personality &= ~PER_CLEAR_ON_SETID;
 964                         bprm->e_gid = inode->i_gid;
 965                 }
 966         }
 967
 968         /* fill in binprm security blob */
 969         retval = security_bprm_set(bprm);
 970         if (retval)
 971                 return retval;
 972
 973         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 974         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 975 }
 976
 977 EXPORT_SYMBOL(prepare_binprm);
 978
 979 static int unsafe_exec(struct task_struct *p)
 980 {
 981         int unsafe = 0;
 982         if (p->ptrace & PT_PTRACED) {
 983                 if (p->ptrace & PT_PTRACE_CAP)
 984                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 985                 else
 986                         unsafe |= LSM_UNSAFE_PTRACE;
 987         }
 988         if (atomic_read(&p->fs->count) > 1 ||
 989             atomic_read(&p->files->count) > 1 ||
 990             atomic_read(&p->sighand->count) > 1)
 991                 unsafe |= LSM_UNSAFE_SHARE;
 992
 993         return unsafe;
 994 }
 995
 996 void compute_creds(struct linux_binprm *bprm)
 997 {
 998         int unsafe;
 999
1000         if (bprm->e_uid != current->uid)
1001                 suid_keys(current);
1002         exec_keys(current);
1003
1004         task_lock(current);
1005         unsafe = unsafe_exec(current);
1006         security_bprm_apply_creds(bprm, unsafe);
1007         task_unlock(current);
1008         security_bprm_post_apply_creds(bprm);
1009 }
1010
1011 EXPORT_SYMBOL(compute_creds);
1012
1013 void remove_arg_zero(struct linux_binprm *bprm)
1014 {
1015         if (bprm->argc) {
1016                 unsigned long offset;
1017                 char * kaddr;
1018                 struct page *page;
1019
1020                 offset = bprm->p % PAGE_SIZE;
1021                 goto inside;
1022
1023                 while (bprm->p++, *(kaddr+offset++)) {
1024                         if (offset != PAGE_SIZE)
1025                                 continue;
1026                         offset = 0;
1027                         kunmap_atomic(kaddr, KM_USER0);
1028 inside:
1029                         page = bprm->page[bprm->p/PAGE_SIZE];
1030                         kaddr = kmap_atomic(page, KM_USER0);
1031                 }
1032                 kunmap_atomic(kaddr, KM_USER0);
1033                 bprm->argc--;
1034         }
1035 }
1036
1037 EXPORT_SYMBOL(remove_arg_zero);
1038
1039 /*
1040  * cycle the list of binary formats handler, until one recognizes the image
1041  */
1042 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1043 {
1044         int try,retval;
1045         struct linux_binfmt *fmt;
1046 #ifdef __alpha__
1047         /* handle /sbin/loader.. */
1048         {
1049             struct exec * eh = (struct exec *) bprm->buf;
1050
1051             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1052                 (eh->fh.f_flags & 0x3000) == 0x3000)
1053             {
1054                 struct file * file;
1055                 unsigned long loader;
1056
1057                 allow_write_access(bprm->file);
1058                 fput(bprm->file);
1059                 bprm->file = NULL;
1060
1061                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1062
1063                 file = open_exec("/sbin/loader");
1064                 retval = PTR_ERR(file);
1065                 if (IS_ERR(file))
1066                         return retval;
1067
1068                 /* Remember if the application is TASO.  */
1069                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1070
1071                 bprm->file = file;
1072                 bprm->loader = loader;
1073                 retval = prepare_binprm(bprm);
1074                 if (retval<0)
1075                         return retval;
1076                 /* should call search_binary_handler recursively here,
1077                    but it does not matter */
1078             }
1079         }
1080 #endif
1081         retval = security_bprm_check(bprm);
1082         if (retval)
1083                 return retval;
1084
1085         /* kernel module loader fixup */
1086         /* so we don't try to load run modprobe in kernel space. */
1087         set_fs(USER_DS);
1088
1089         retval = audit_bprm(bprm);
1090         if (retval)
1091                 return retval;
1092
1093         retval = -ENOENT;
1094         for (try=0; try<2; try++) {
1095                 read_lock(&binfmt_lock);
1096                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1097                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1098                         if (!fn)
1099                                 continue;
1100                         if (!try_module_get(fmt->module))
1101                                 continue;
1102                         read_unlock(&binfmt_lock);
1103                         retval = fn(bprm, regs);
1104                         if (retval >= 0) {
1105                                 put_binfmt(fmt);
1106                                 allow_write_access(bprm->file);
1107                                 if (bprm->file)
1108                                         fput(bprm->file);
1109                                 bprm->file = NULL;
1110                                 current->did_exec = 1;
1111                                 proc_exec_connector(current);
1112                                 return retval;
1113                         }
1114                         read_lock(&binfmt_lock);
1115                         put_binfmt(fmt);
1116                         if (retval != -ENOEXEC || bprm->mm == NULL)
1117                                 break;
1118                         if (!bprm->file) {
1119                                 read_unlock(&binfmt_lock);
1120                                 return retval;
1121                         }
1122                 }
1123                 read_unlock(&binfmt_lock);
1124                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1125                         break;
1126 #ifdef CONFIG_KMOD
1127                 }else{
1128 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1129                         if (printable(bprm->buf[0]) &&
1130                             printable(bprm->buf[1]) &&
1131                             printable(bprm->buf[2]) &&
1132                             printable(bprm->buf[3]))
1133                                 break; /* -ENOEXEC */
1134                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1135 #endif
1136                 }
1137         }
1138         return retval;
1139 }
1140
1141 EXPORT_SYMBOL(search_binary_handler);
1142
1143 /*
1144  * sys_execve() executes a new program.
1145  */
1146 int do_execve(char * filename,
1147         char __user *__user *argv,
1148         char __user *__user *envp,
1149         struct pt_regs * regs)
1150 {
1151         struct linux_binprm *bprm;
1152         struct file *file;
1153         int retval;
1154         int i;
1155
1156         retval = -ENOMEM;
1157         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1158         if (!bprm)
1159                 goto out_ret;
1160
1161         file = open_exec(filename);
1162         retval = PTR_ERR(file);
1163         if (IS_ERR(file))
1164                 goto out_kfree;
1165
1166         sched_exec();
1167
1168         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1169
1170         bprm->file = file;
1171         bprm->filename = filename;
1172         bprm->interp = filename;
1173         bprm->mm = mm_alloc();
1174         retval = -ENOMEM;
1175         if (!bprm->mm)
1176                 goto out_file;
1177
1178         retval = init_new_context(current, bprm->mm);
1179         if (retval < 0)
1180                 goto out_mm;
1181
1182         bprm->argc = count(argv, bprm->p / sizeof(void *));
1183         if ((retval = bprm->argc) < 0)
1184                 goto out_mm;
1185
1186         bprm->envc = count(envp, bprm->p / sizeof(void *));
1187         if ((retval = bprm->envc) < 0)
1188                 goto out_mm;
1189
1190         retval = security_bprm_alloc(bprm);
1191         if (retval)
1192                 goto out;
1193
1194         retval = prepare_binprm(bprm);
1195         if (retval < 0)
1196                 goto out;
1197
1198         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1199         if (retval < 0)
1200                 goto out;
1201
1202         bprm->exec = bprm->p;
1203         retval = copy_strings(bprm->envc, envp, bprm);
1204         if (retval < 0)
1205                 goto out;
1206
1207         retval = copy_strings(bprm->argc, argv, bprm);
1208         if (retval < 0)
1209                 goto out;
1210
1211         retval = search_binary_handler(bprm,regs);
1212         if (retval >= 0) {
1213                 free_arg_pages(bprm);
1214
1215                 /* execve success */
1216                 security_bprm_free(bprm);
1217                 acct_update_integrals(current);
1218                 kfree(bprm);
1219                 return retval;
1220         }
1221
1222 out:
1223         /* Something went wrong, return the inode and free the argument pages*/
1224         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1225                 struct page * page = bprm->page[i];
1226                 if (page)
1227                         __free_page(page);
1228         }
1229
1230         if (bprm->security)
1231                 security_bprm_free(bprm);
1232
1233 out_mm:
1234         if (bprm->mm)
1235                 mmdrop(bprm->mm);
1236
1237 out_file:
1238         if (bprm->file) {
1239                 allow_write_access(bprm->file);
1240                 fput(bprm->file);
1241         }
1242
1243 out_kfree:
1244         kfree(bprm);
1245
1246 out_ret:
1247         return retval;
1248 }
1249
1250 int set_binfmt(struct linux_binfmt *new)
1251 {
1252         struct linux_binfmt *old = current->binfmt;
1253
1254         if (new) {
1255                 if (!try_module_get(new->module))
1256                         return -1;
1257         }
1258         current->binfmt = new;
1259         if (old)
1260                 module_put(old->module);
1261         return 0;
1262 }
1263
1264 EXPORT_SYMBOL(set_binfmt);
1265
1266 #define CORENAME_MAX_SIZE 64
1267
1268 /* format_corename will inspect the pattern parameter, and output a
1269  * name into corename, which must have space for at least
1270  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1271  */
1272 static void format_corename(char *corename, const char *pattern, long signr)
1273 {
1274         const char *pat_ptr = pattern;
1275         char *out_ptr = corename;
1276         char *const out_end = corename + CORENAME_MAX_SIZE;
1277         int rc;
1278         int pid_in_pattern = 0;
1279
1280         /* Repeat as long as we have more pattern to process and more output
1281            space */
1282         while (*pat_ptr) {
1283                 if (*pat_ptr != '%') {
1284                         if (out_ptr == out_end)
1285                                 goto out;
1286                         *out_ptr++ = *pat_ptr++;
1287                 } else {
1288                         switch (*++pat_ptr) {
1289                         case 0:
1290                                 goto out;
1291                         /* Double percent, output one percent */
1292                         case '%':
1293                                 if (out_ptr == out_end)
1294                                         goto out;
1295                                 *out_ptr++ = '%';
1296                                 break;
1297                         /* pid */
1298                         case 'p':
1299                                 pid_in_pattern = 1;
1300                                 rc = snprintf(out_ptr, out_end - out_ptr,
1301                                               "%d", current->tgid);
1302                                 if (rc > out_end - out_ptr)
1303                                         goto out;
1304                                 out_ptr += rc;
1305                                 break;
1306                         /* uid */
1307                         case 'u':
1308                                 rc = snprintf(out_ptr, out_end - out_ptr,
1309                                               "%d", current->uid);
1310                                 if (rc > out_end - out_ptr)
1311                                         goto out;
1312                                 out_ptr += rc;
1313                                 break;
1314                         /* gid */
1315                         case 'g':
1316                                 rc = snprintf(out_ptr, out_end - out_ptr,
1317                                               "%d", current->gid);
1318                                 if (rc > out_end - out_ptr)
1319                                         goto out;
1320                                 out_ptr += rc;
1321                                 break;
1322                         /* signal that caused the coredump */
1323                         case 's':
1324                                 rc = snprintf(out_ptr, out_end - out_ptr,
1325                                               "%ld", signr);
1326                                 if (rc > out_end - out_ptr)
1327                                         goto out;
1328                                 out_ptr += rc;
1329                                 break;
1330                         /* UNIX time of coredump */
1331                         case 't': {
1332                                 struct timeval tv;
1333                                 do_gettimeofday(&tv);
1334                                 rc = snprintf(out_ptr, out_end - out_ptr,
1335                                               "%lu", tv.tv_sec);
1336                                 if (rc > out_end - out_ptr)
1337                                         goto out;
1338                                 out_ptr += rc;
1339                                 break;
1340                         }
1341                         /* hostname */
1342                         case 'h':
1343                                 down_read(&uts_sem);
1344                                 rc = snprintf(out_ptr, out_end - out_ptr,
1345                                               "%s", system_utsname.nodename);
1346                                 up_read(&uts_sem);
1347                                 if (rc > out_end - out_ptr)
1348                                         goto out;
1349                                 out_ptr += rc;
1350                                 break;
1351                         /* executable */
1352                         case 'e':
1353                                 rc = snprintf(out_ptr, out_end - out_ptr,
1354                                               "%s", current->comm);
1355                                 if (rc > out_end - out_ptr)
1356                                         goto out;
1357                                 out_ptr += rc;
1358                                 break;
1359                         default:
1360                                 break;
1361                         }
1362                         ++pat_ptr;
1363                 }
1364         }
1365         /* Backward compatibility with core_uses_pid:
1366          *
1367          * If core_pattern does not include a %p (as is the default)
1368          * and core_uses_pid is set, then .%pid will be appended to
1369          * the filename */
1370         if (!pid_in_pattern
1371             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1372                 rc = snprintf(out_ptr, out_end - out_ptr,
1373                               ".%d", current->tgid);
1374                 if (rc > out_end - out_ptr)
1375                         goto out;
1376                 out_ptr += rc;
1377         }
1378       out:
1379         *out_ptr = 0;
1380 }
1381
1382 static void zap_threads (struct mm_struct *mm)
1383 {
1384         struct task_struct *g, *p;
1385         struct task_struct *tsk = current;
1386         struct completion *vfork_done = tsk->vfork_done;
1387         int traced = 0;
1388
1389         /*
1390          * Make sure nobody is waiting for us to release the VM,
1391          * otherwise we can deadlock when we wait on each other
1392          */
1393         if (vfork_done) {
1394                 tsk->vfork_done = NULL;
1395                 complete(vfork_done);
1396         }
1397
1398         read_lock(&tasklist_lock);
1399         do_each_thread(g,p)
1400                 if (mm == p->mm && p != tsk) {
1401                         force_sig_specific(SIGKILL, p);
1402                         mm->core_waiters++;
1403                         if (unlikely(p->ptrace) &&
1404                             unlikely(p->parent->mm == mm))
1405                                 traced = 1;
1406                 }
1407         while_each_thread(g,p);
1408
1409         read_unlock(&tasklist_lock);
1410
1411         if (unlikely(traced)) {
1412                 /*
1413                  * We are zapping a thread and the thread it ptraces.
1414                  * If the tracee went into a ptrace stop for exit tracing,
1415                  * we could deadlock since the tracer is waiting for this
1416                  * coredump to finish.  Detach them so they can both die.
1417                  */
1418                 write_lock_irq(&tasklist_lock);
1419                 do_each_thread(g,p) {
1420                         if (mm == p->mm && p != tsk &&
1421                             p->ptrace && p->parent->mm == mm) {
1422                                 __ptrace_detach(p, 0);
1423                         }
1424                 } while_each_thread(g,p);
1425                 write_unlock_irq(&tasklist_lock);
1426         }
1427 }
1428
1429 static void coredump_wait(struct mm_struct *mm)
1430 {
1431         DECLARE_COMPLETION(startup_done);
1432         int core_waiters;
1433
1434         mm->core_startup_done = &startup_done;
1435
1436         zap_threads(mm);
1437         core_waiters = mm->core_waiters;
1438         up_write(&mm->mmap_sem);
1439
1440         if (core_waiters)
1441                 wait_for_completion(&startup_done);
1442         BUG_ON(mm->core_waiters);
1443 }
1444
1445 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1446 {
1447         char corename[CORENAME_MAX_SIZE + 1];
1448         struct mm_struct *mm = current->mm;
1449         struct linux_binfmt * binfmt;
1450         struct inode * inode;
1451         struct file * file;
1452         int retval = 0;
1453         int fsuid = current->fsuid;
1454         int flag = 0;
1455
1456         binfmt = current->binfmt;
1457         if (!binfmt || !binfmt->core_dump)
1458                 goto fail;
1459         down_write(&mm->mmap_sem);
1460         if (!mm->dumpable) {
1461                 up_write(&mm->mmap_sem);
1462                 goto fail;
1463         }
1464
1465         /*
1466          *      We cannot trust fsuid as being the "true" uid of the
1467          *      process nor do we know its entire history. We only know it
1468          *      was tainted so we dump it as root in mode 2.
1469          */
1470         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1471                 flag = O_EXCL;          /* Stop rewrite attacks */
1472                 current->fsuid = 0;     /* Dump root private */
1473         }
1474         mm->dumpable = 0;
1475
1476         retval = -EAGAIN;
1477         spin_lock_irq(&current->sighand->siglock);
1478         if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1479                 current->signal->flags = SIGNAL_GROUP_EXIT;
1480                 current->signal->group_exit_code = exit_code;
1481                 current->signal->group_stop_count = 0;
1482                 retval = 0;
1483         }
1484         spin_unlock_irq(&current->sighand->siglock);
1485         if (retval) {
1486                 up_write(&mm->mmap_sem);
1487                 goto fail;
1488         }
1489
1490         init_completion(&mm->core_done);
1491         coredump_wait(mm);
1492
1493         /*
1494          * Clear any false indication of pending signals that might
1495          * be seen by the filesystem code called to write the core file.
1496          */
1497         clear_thread_flag(TIF_SIGPENDING);
1498
1499         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1500                 goto fail_unlock;
1501
1502         /*
1503          * lock_kernel() because format_corename() is controlled by sysctl, which
1504          * uses lock_kernel()
1505          */
1506         lock_kernel();
1507         format_corename(corename, core_pattern, signr);
1508         unlock_kernel();
1509         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1510         if (IS_ERR(file))
1511                 goto fail_unlock;
1512         inode = file->f_dentry->d_inode;
1513         if (inode->i_nlink > 1)
1514                 goto close_fail;        /* multiple links - don't dump */
1515         if (d_unhashed(file->f_dentry))
1516                 goto close_fail;
1517
1518         if (!S_ISREG(inode->i_mode))
1519                 goto close_fail;
1520         if (!file->f_op)
1521                 goto close_fail;
1522         if (!file->f_op->write)
1523                 goto close_fail;
1524         if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1525                 goto close_fail;
1526
1527         retval = binfmt->core_dump(signr, regs, file);
1528
1529         if (retval)
1530                 current->signal->group_exit_code |= 0x80;
1531 close_fail:
1532         filp_close(file, NULL);
1533 fail_unlock:
1534         current->fsuid = fsuid;
1535         complete_all(&mm->core_done);
1536 fail:
1537         return retval;
1538 }