git.oblomov.eu Git - linux-2.6/blob - fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/mman.h>
  28 #include <linux/a.out.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/init.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/highmem.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/key.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/pid_namespace.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/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/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53 #include <linux/signalfd.h>
  54
  55 #include <asm/uaccess.h>
  56 #include <asm/mmu_context.h>
  57
  58 #ifdef CONFIG_KMOD
  59 #include <linux/kmod.h>
  60 #endif
  61
  62 int core_uses_pid;
  63 char core_pattern[CORENAME_MAX_SIZE] = "core";
  64 int suid_dumpable = 0;
  65
  66 EXPORT_SYMBOL(suid_dumpable);
  67 /* The maximal length of core_pattern is also specified in sysctl.c */
  68
  69 static struct linux_binfmt *formats;
  70 static DEFINE_RWLOCK(binfmt_lock);
  71
  72 int register_binfmt(struct linux_binfmt * fmt)
  73 {
  74         struct linux_binfmt ** tmp = &formats;
  75
  76         if (!fmt)
  77                 return -EINVAL;
  78         if (fmt->next)
  79                 return -EBUSY;
  80         write_lock(&binfmt_lock);
  81         while (*tmp) {
  82                 if (fmt == *tmp) {
  83                         write_unlock(&binfmt_lock);
  84                         return -EBUSY;
  85                 }
  86                 tmp = &(*tmp)->next;
  87         }
  88         fmt->next = formats;
  89         formats = fmt;
  90         write_unlock(&binfmt_lock);
  91         return 0;
  92 }
  93
  94 EXPORT_SYMBOL(register_binfmt);
  95
  96 int unregister_binfmt(struct linux_binfmt * fmt)
  97 {
  98         struct linux_binfmt ** tmp = &formats;
  99
 100         write_lock(&binfmt_lock);
 101         while (*tmp) {
 102                 if (fmt == *tmp) {
 103                         *tmp = fmt->next;
 104                         fmt->next = NULL;
 105                         write_unlock(&binfmt_lock);
 106                         return 0;
 107                 }
 108                 tmp = &(*tmp)->next;
 109         }
 110         write_unlock(&binfmt_lock);
 111         return -EINVAL;
 112 }
 113
 114 EXPORT_SYMBOL(unregister_binfmt);
 115
 116 static inline void put_binfmt(struct linux_binfmt * fmt)
 117 {
 118         module_put(fmt->module);
 119 }
 120
 121 /*
 122  * Note that a shared library must be both readable and executable due to
 123  * security reasons.
 124  *
 125  * Also note that we take the address to load from from the file itself.
 126  */
 127 asmlinkage long sys_uselib(const char __user * library)
 128 {
 129         struct file * file;
 130         struct nameidata nd;
 131         int error;
 132
 133         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 134         if (error)
 135                 goto out;
 136
 137         error = -EINVAL;
 138         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 139                 goto exit;
 140
 141         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 142         if (error)
 143                 goto exit;
 144
 145         file = nameidata_to_filp(&nd, O_RDONLY);
 146         error = PTR_ERR(file);
 147         if (IS_ERR(file))
 148                 goto out;
 149
 150         error = -ENOEXEC;
 151         if(file->f_op) {
 152                 struct linux_binfmt * fmt;
 153
 154                 read_lock(&binfmt_lock);
 155                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 156                         if (!fmt->load_shlib)
 157                                 continue;
 158                         if (!try_module_get(fmt->module))
 159                                 continue;
 160                         read_unlock(&binfmt_lock);
 161                         error = fmt->load_shlib(file);
 162                         read_lock(&binfmt_lock);
 163                         put_binfmt(fmt);
 164                         if (error != -ENOEXEC)
 165                                 break;
 166                 }
 167                 read_unlock(&binfmt_lock);
 168         }
 169         fput(file);
 170 out:
 171         return error;
 172 exit:
 173         release_open_intent(&nd);
 174         path_release(&nd);
 175         goto out;
 176 }
 177
 178 /*
 179  * count() counts the number of strings in array ARGV.
 180  */
 181 static int count(char __user * __user * argv, int max)
 182 {
 183         int i = 0;
 184
 185         if (argv != NULL) {
 186                 for (;;) {
 187                         char __user * p;
 188
 189                         if (get_user(p, argv))
 190                                 return -EFAULT;
 191                         if (!p)
 192                                 break;
 193                         argv++;
 194                         if(++i > max)
 195                                 return -E2BIG;
 196                         cond_resched();
 197                 }
 198         }
 199         return i;
 200 }
 201
 202 /*
 203  * 'copy_strings()' copies argument/environment strings from user
 204  * memory to free pages in kernel mem. These are in a format ready
 205  * to be put directly into the top of new user memory.
 206  */
 207 static int copy_strings(int argc, char __user * __user * argv,
 208                         struct linux_binprm *bprm)
 209 {
 210         struct page *kmapped_page = NULL;
 211         char *kaddr = NULL;
 212         int ret;
 213
 214         while (argc-- > 0) {
 215                 char __user *str;
 216                 int len;
 217                 unsigned long pos;
 218
 219                 if (get_user(str, argv+argc) ||
 220                                 !(len = strnlen_user(str, bprm->p))) {
 221                         ret = -EFAULT;
 222                         goto out;
 223                 }
 224
 225                 if (bprm->p < len)  {
 226                         ret = -E2BIG;
 227                         goto out;
 228                 }
 229
 230                 bprm->p -= len;
 231                 /* XXX: add architecture specific overflow check here. */
 232                 pos = bprm->p;
 233
 234                 while (len > 0) {
 235                         int i, new, err;
 236                         int offset, bytes_to_copy;
 237                         struct page *page;
 238
 239                         offset = pos % PAGE_SIZE;
 240                         i = pos/PAGE_SIZE;
 241                         page = bprm->page[i];
 242                         new = 0;
 243                         if (!page) {
 244                                 page = alloc_page(GFP_HIGHUSER);
 245                                 bprm->page[i] = page;
 246                                 if (!page) {
 247                                         ret = -ENOMEM;
 248                                         goto out;
 249                                 }
 250                                 new = 1;
 251                         }
 252
 253                         if (page != kmapped_page) {
 254                                 if (kmapped_page)
 255                                         kunmap(kmapped_page);
 256                                 kmapped_page = page;
 257                                 kaddr = kmap(kmapped_page);
 258                         }
 259                         if (new && offset)
 260                                 memset(kaddr, 0, offset);
 261                         bytes_to_copy = PAGE_SIZE - offset;
 262                         if (bytes_to_copy > len) {
 263                                 bytes_to_copy = len;
 264                                 if (new)
 265                                         memset(kaddr+offset+len, 0,
 266                                                 PAGE_SIZE-offset-len);
 267                         }
 268                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 269                         if (err) {
 270                                 ret = -EFAULT;
 271                                 goto out;
 272                         }
 273
 274                         pos += bytes_to_copy;
 275                         str += bytes_to_copy;
 276                         len -= bytes_to_copy;
 277                 }
 278         }
 279         ret = 0;
 280 out:
 281         if (kmapped_page)
 282                 kunmap(kmapped_page);
 283         return ret;
 284 }
 285
 286 /*
 287  * Like copy_strings, but get argv and its values from kernel memory.
 288  */
 289 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 290 {
 291         int r;
 292         mm_segment_t oldfs = get_fs();
 293         set_fs(KERNEL_DS);
 294         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 295         set_fs(oldfs);
 296         return r;
 297 }
 298
 299 EXPORT_SYMBOL(copy_strings_kernel);
 300
 301 #ifdef CONFIG_MMU
 302 /*
 303  * This routine is used to map in a page into an address space: needed by
 304  * execve() for the initial stack and environment pages.
 305  *
 306  * vma->vm_mm->mmap_sem is held for writing.
 307  */
 308 void install_arg_page(struct vm_area_struct *vma,
 309                         struct page *page, unsigned long address)
 310 {
 311         struct mm_struct *mm = vma->vm_mm;
 312         pte_t * pte;
 313         spinlock_t *ptl;
 314
 315         if (unlikely(anon_vma_prepare(vma)))
 316                 goto out;
 317
 318         flush_dcache_page(page);
 319         pte = get_locked_pte(mm, address, &ptl);
 320         if (!pte)
 321                 goto out;
 322         if (!pte_none(*pte)) {
 323                 pte_unmap_unlock(pte, ptl);
 324                 goto out;
 325         }
 326         inc_mm_counter(mm, anon_rss);
 327         lru_cache_add_active(page);
 328         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
 329                                         page, vma->vm_page_prot))));
 330         page_add_new_anon_rmap(page, vma, address);
 331         pte_unmap_unlock(pte, ptl);
 332
 333         /* no need for flush_tlb */
 334         return;
 335 out:
 336         __free_page(page);
 337         force_sig(SIGKILL, current);
 338 }
 339
 340 #define EXTRA_STACK_VM_PAGES    20      /* random */
 341
 342 int setup_arg_pages(struct linux_binprm *bprm,
 343                     unsigned long stack_top,
 344                     int executable_stack)
 345 {
 346         unsigned long stack_base;
 347         struct vm_area_struct *mpnt;
 348         struct mm_struct *mm = current->mm;
 349         int i, ret;
 350         long arg_size;
 351
 352 #ifdef CONFIG_STACK_GROWSUP
 353         /* Move the argument and environment strings to the bottom of the
 354          * stack space.
 355          */
 356         int offset, j;
 357         char *to, *from;
 358
 359         /* Start by shifting all the pages down */
 360         i = 0;
 361         for (j = 0; j < MAX_ARG_PAGES; j++) {
 362                 struct page *page = bprm->page[j];
 363                 if (!page)
 364                         continue;
 365                 bprm->page[i++] = page;
 366         }
 367
 368         /* Now move them within their pages */
 369         offset = bprm->p % PAGE_SIZE;
 370         to = kmap(bprm->page[0]);
 371         for (j = 1; j < i; j++) {
 372                 memmove(to, to + offset, PAGE_SIZE - offset);
 373                 from = kmap(bprm->page[j]);
 374                 memcpy(to + PAGE_SIZE - offset, from, offset);
 375                 kunmap(bprm->page[j - 1]);
 376                 to = from;
 377         }
 378         memmove(to, to + offset, PAGE_SIZE - offset);
 379         kunmap(bprm->page[j - 1]);
 380
 381         /* Limit stack size to 1GB */
 382         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 383         if (stack_base > (1 << 30))
 384                 stack_base = 1 << 30;
 385         stack_base = PAGE_ALIGN(stack_top - stack_base);
 386
 387         /* Adjust bprm->p to point to the end of the strings. */
 388         bprm->p = stack_base + PAGE_SIZE * i - offset;
 389
 390         mm->arg_start = stack_base;
 391         arg_size = i << PAGE_SHIFT;
 392
 393         /* zero pages that were copied above */
 394         while (i < MAX_ARG_PAGES)
 395                 bprm->page[i++] = NULL;
 396 #else
 397         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
 398         stack_base = PAGE_ALIGN(stack_base);
 399         bprm->p += stack_base;
 400         mm->arg_start = bprm->p;
 401         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 402 #endif
 403
 404         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 405
 406         if (bprm->loader)
 407                 bprm->loader += stack_base;
 408         bprm->exec += stack_base;
 409
 410         mpnt = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 411         if (!mpnt)
 412                 return -ENOMEM;
 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                         file = ERR_PTR(err);
 491                         if (!err) {
 492                                 file = nameidata_to_filp(&nd, O_RDONLY);
 493                                 if (!IS_ERR(file)) {
 494                                         err = deny_write_access(file);
 495                                         if (err) {
 496                                                 fput(file);
 497                                                 file = ERR_PTR(err);
 498                                         }
 499                                 }
 500 out:
 501                                 return file;
 502                         }
 503                 }
 504                 release_open_intent(&nd);
 505                 path_release(&nd);
 506         }
 507         goto out;
 508 }
 509
 510 EXPORT_SYMBOL(open_exec);
 511
 512 int kernel_read(struct file *file, unsigned long offset,
 513         char *addr, unsigned long count)
 514 {
 515         mm_segment_t old_fs;
 516         loff_t pos = offset;
 517         int result;
 518
 519         old_fs = get_fs();
 520         set_fs(get_ds());
 521         /* The cast to a user pointer is valid due to the set_fs() */
 522         result = vfs_read(file, (void __user *)addr, count, &pos);
 523         set_fs(old_fs);
 524         return result;
 525 }
 526
 527 EXPORT_SYMBOL(kernel_read);
 528
 529 static int exec_mmap(struct mm_struct *mm)
 530 {
 531         struct task_struct *tsk;
 532         struct mm_struct * old_mm, *active_mm;
 533
 534         /* Notify parent that we're no longer interested in the old VM */
 535         tsk = current;
 536         old_mm = current->mm;
 537         mm_release(tsk, old_mm);
 538
 539         if (old_mm) {
 540                 /*
 541                  * Make sure that if there is a core dump in progress
 542                  * for the old mm, we get out and die instead of going
 543                  * through with the exec.  We must hold mmap_sem around
 544                  * checking core_waiters and changing tsk->mm.  The
 545                  * core-inducing thread will increment core_waiters for
 546                  * each thread whose ->mm == old_mm.
 547                  */
 548                 down_read(&old_mm->mmap_sem);
 549                 if (unlikely(old_mm->core_waiters)) {
 550                         up_read(&old_mm->mmap_sem);
 551                         return -EINTR;
 552                 }
 553         }
 554         task_lock(tsk);
 555         active_mm = tsk->active_mm;
 556         tsk->mm = mm;
 557         tsk->active_mm = mm;
 558         activate_mm(active_mm, mm);
 559         task_unlock(tsk);
 560         arch_pick_mmap_layout(mm);
 561         if (old_mm) {
 562                 up_read(&old_mm->mmap_sem);
 563                 BUG_ON(active_mm != old_mm);
 564                 mmput(old_mm);
 565                 return 0;
 566         }
 567         mmdrop(active_mm);
 568         return 0;
 569 }
 570
 571 /*
 572  * This function makes sure the current process has its own signal table,
 573  * so that flush_signal_handlers can later reset the handlers without
 574  * disturbing other processes.  (Other processes might share the signal
 575  * table via the CLONE_SIGHAND option to clone().)
 576  */
 577 static int de_thread(struct task_struct *tsk)
 578 {
 579         struct signal_struct *sig = tsk->signal;
 580         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 581         spinlock_t *lock = &oldsighand->siglock;
 582         struct task_struct *leader = NULL;
 583         int count;
 584
 585         /*
 586          * Tell all the sighand listeners that this sighand has
 587          * been detached. The signalfd_detach() function grabs the
 588          * sighand lock, if signal listeners are present on the sighand.
 589          */
 590         signalfd_detach(tsk);
 591
 592         /*
 593          * If we don't share sighandlers, then we aren't sharing anything
 594          * and we can just re-use it all.
 595          */
 596         if (atomic_read(&oldsighand->count) <= 1) {
 597                 BUG_ON(atomic_read(&sig->count) != 1);
 598                 exit_itimers(sig);
 599                 return 0;
 600         }
 601
 602         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 603         if (!newsighand)
 604                 return -ENOMEM;
 605
 606         if (thread_group_empty(tsk))
 607                 goto no_thread_group;
 608
 609         /*
 610          * Kill all other threads in the thread group.
 611          * We must hold tasklist_lock to call zap_other_threads.
 612          */
 613         read_lock(&tasklist_lock);
 614         spin_lock_irq(lock);
 615         if (sig->flags & SIGNAL_GROUP_EXIT) {
 616                 /*
 617                  * Another group action in progress, just
 618                  * return so that the signal is processed.
 619                  */
 620                 spin_unlock_irq(lock);
 621                 read_unlock(&tasklist_lock);
 622                 kmem_cache_free(sighand_cachep, newsighand);
 623                 return -EAGAIN;
 624         }
 625
 626         /*
 627          * child_reaper ignores SIGKILL, change it now.
 628          * Reparenting needs write_lock on tasklist_lock,
 629          * so it is safe to do it under read_lock.
 630          */
 631         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 632                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 633
 634         zap_other_threads(tsk);
 635         read_unlock(&tasklist_lock);
 636
 637         /*
 638          * Account for the thread group leader hanging around:
 639          */
 640         count = 1;
 641         if (!thread_group_leader(tsk)) {
 642                 count = 2;
 643                 /*
 644                  * The SIGALRM timer survives the exec, but needs to point
 645                  * at us as the new group leader now.  We have a race with
 646                  * a timer firing now getting the old leader, so we need to
 647                  * synchronize with any firing (by calling del_timer_sync)
 648                  * before we can safely let the old group leader die.
 649                  */
 650                 sig->tsk = tsk;
 651                 spin_unlock_irq(lock);
 652                 if (hrtimer_cancel(&sig->real_timer))
 653                         hrtimer_restart(&sig->real_timer);
 654                 spin_lock_irq(lock);
 655         }
 656         while (atomic_read(&sig->count) > count) {
 657                 sig->group_exit_task = tsk;
 658                 sig->notify_count = count;
 659                 __set_current_state(TASK_UNINTERRUPTIBLE);
 660                 spin_unlock_irq(lock);
 661                 schedule();
 662                 spin_lock_irq(lock);
 663         }
 664         sig->group_exit_task = NULL;
 665         sig->notify_count = 0;
 666         spin_unlock_irq(lock);
 667
 668         /*
 669          * At this point all other threads have exited, all we have to
 670          * do is to wait for the thread group leader to become inactive,
 671          * and to assume its PID:
 672          */
 673         if (!thread_group_leader(tsk)) {
 674                 /*
 675                  * Wait for the thread group leader to be a zombie.
 676                  * It should already be zombie at this point, most
 677                  * of the time.
 678                  */
 679                 leader = tsk->group_leader;
 680                 while (leader->exit_state != EXIT_ZOMBIE)
 681                         yield();
 682
 683                 /*
 684                  * The only record we have of the real-time age of a
 685                  * process, regardless of execs it's done, is start_time.
 686                  * All the past CPU time is accumulated in signal_struct
 687                  * from sister threads now dead.  But in this non-leader
 688                  * exec, nothing survives from the original leader thread,
 689                  * whose birth marks the true age of this process now.
 690                  * When we take on its identity by switching to its PID, we
 691                  * also take its birthdate (always earlier than our own).
 692                  */
 693                 tsk->start_time = leader->start_time;
 694
 695                 write_lock_irq(&tasklist_lock);
 696
 697                 BUG_ON(leader->tgid != tsk->tgid);
 698                 BUG_ON(tsk->pid == tsk->tgid);
 699                 /*
 700                  * An exec() starts a new thread group with the
 701                  * TGID of the previous thread group. Rehash the
 702                  * two threads with a switched PID, and release
 703                  * the former thread group leader:
 704                  */
 705
 706                 /* Become a process group leader with the old leader's pid.
 707                  * The old leader becomes a thread of the this thread group.
 708                  * Note: The old leader also uses this pid until release_task
 709                  *       is called.  Odd but simple and correct.
 710                  */
 711                 detach_pid(tsk, PIDTYPE_PID);
 712                 tsk->pid = leader->pid;
 713                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
 714                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 715                 transfer_pid(leader, tsk, PIDTYPE_SID);
 716                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 717
 718                 tsk->group_leader = tsk;
 719                 leader->group_leader = tsk;
 720
 721                 tsk->exit_signal = SIGCHLD;
 722
 723                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 724                 leader->exit_state = EXIT_DEAD;
 725
 726                 write_unlock_irq(&tasklist_lock);
 727         }
 728
 729         /*
 730          * There may be one thread left which is just exiting,
 731          * but it's safe to stop telling the group to kill themselves.
 732          */
 733         sig->flags = 0;
 734
 735 no_thread_group:
 736         exit_itimers(sig);
 737         if (leader)
 738                 release_task(leader);
 739
 740         BUG_ON(atomic_read(&sig->count) != 1);
 741
 742         if (atomic_read(&oldsighand->count) == 1) {
 743                 /*
 744                  * Now that we nuked the rest of the thread group,
 745                  * it turns out we are not sharing sighand any more either.
 746                  * So we can just keep it.
 747                  */
 748                 kmem_cache_free(sighand_cachep, newsighand);
 749         } else {
 750                 /*
 751                  * Move our state over to newsighand and switch it in.
 752                  */
 753                 atomic_set(&newsighand->count, 1);
 754                 memcpy(newsighand->action, oldsighand->action,
 755                        sizeof(newsighand->action));
 756
 757                 write_lock_irq(&tasklist_lock);
 758                 spin_lock(&oldsighand->siglock);
 759                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 760
 761                 rcu_assign_pointer(tsk->sighand, newsighand);
 762                 recalc_sigpending();
 763
 764                 spin_unlock(&newsighand->siglock);
 765                 spin_unlock(&oldsighand->siglock);
 766                 write_unlock_irq(&tasklist_lock);
 767
 768                 __cleanup_sighand(oldsighand);
 769         }
 770
 771         BUG_ON(!thread_group_leader(tsk));
 772         return 0;
 773 }
 774
 775 /*
 776  * These functions flushes out all traces of the currently running executable
 777  * so that a new one can be started
 778  */
 779
 780 static void flush_old_files(struct files_struct * files)
 781 {
 782         long j = -1;
 783         struct fdtable *fdt;
 784
 785         spin_lock(&files->file_lock);
 786         for (;;) {
 787                 unsigned long set, i;
 788
 789                 j++;
 790                 i = j * __NFDBITS;
 791                 fdt = files_fdtable(files);
 792                 if (i >= fdt->max_fds)
 793                         break;
 794                 set = fdt->close_on_exec->fds_bits[j];
 795                 if (!set)
 796                         continue;
 797                 fdt->close_on_exec->fds_bits[j] = 0;
 798                 spin_unlock(&files->file_lock);
 799                 for ( ; set ; i++,set >>= 1) {
 800                         if (set & 1) {
 801                                 sys_close(i);
 802                         }
 803                 }
 804                 spin_lock(&files->file_lock);
 805
 806         }
 807         spin_unlock(&files->file_lock);
 808 }
 809
 810 void get_task_comm(char *buf, struct task_struct *tsk)
 811 {
 812         /* buf must be at least sizeof(tsk->comm) in size */
 813         task_lock(tsk);
 814         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 815         task_unlock(tsk);
 816 }
 817
 818 void set_task_comm(struct task_struct *tsk, char *buf)
 819 {
 820         task_lock(tsk);
 821         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 822         task_unlock(tsk);
 823 }
 824
 825 int flush_old_exec(struct linux_binprm * bprm)
 826 {
 827         char * name;
 828         int i, ch, retval;
 829         struct files_struct *files;
 830         char tcomm[sizeof(current->comm)];
 831
 832         /*
 833          * Make sure we have a private signal table and that
 834          * we are unassociated from the previous thread group.
 835          */
 836         retval = de_thread(current);
 837         if (retval)
 838                 goto out;
 839
 840         /*
 841          * Make sure we have private file handles. Ask the
 842          * fork helper to do the work for us and the exit
 843          * helper to do the cleanup of the old one.
 844          */
 845         files = current->files;         /* refcounted so safe to hold */
 846         retval = unshare_files();
 847         if (retval)
 848                 goto out;
 849         /*
 850          * Release all of the old mmap stuff
 851          */
 852         retval = exec_mmap(bprm->mm);
 853         if (retval)
 854                 goto mmap_failed;
 855
 856         bprm->mm = NULL;                /* We're using it now */
 857
 858         /* This is the point of no return */
 859         put_files_struct(files);
 860
 861         current->sas_ss_sp = current->sas_ss_size = 0;
 862
 863         if (current->euid == current->uid && current->egid == current->gid)
 864                 current->mm->dumpable = 1;
 865         else
 866                 current->mm->dumpable = suid_dumpable;
 867
 868         name = bprm->filename;
 869
 870         /* Copies the binary name from after last slash */
 871         for (i=0; (ch = *(name++)) != '\0';) {
 872                 if (ch == '/')
 873                         i = 0; /* overwrite what we wrote */
 874                 else
 875                         if (i < (sizeof(tcomm) - 1))
 876                                 tcomm[i++] = ch;
 877         }
 878         tcomm[i] = '\0';
 879         set_task_comm(current, tcomm);
 880
 881         current->flags &= ~PF_RANDOMIZE;
 882         flush_thread();
 883
 884         /* Set the new mm task size. We have to do that late because it may
 885          * depend on TIF_32BIT which is only updated in flush_thread() on
 886          * some architectures like powerpc
 887          */
 888         current->mm->task_size = TASK_SIZE;
 889
 890         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 891             file_permission(bprm->file, MAY_READ) ||
 892             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 893                 suid_keys(current);
 894                 current->mm->dumpable = suid_dumpable;
 895         }
 896
 897         /* An exec changes our domain. We are no longer part of the thread
 898            group */
 899
 900         current->self_exec_id++;
 901
 902         flush_signal_handlers(current, 0);
 903         flush_old_files(current->files);
 904
 905         return 0;
 906
 907 mmap_failed:
 908         reset_files_struct(current, files);
 909 out:
 910         return retval;
 911 }
 912
 913 EXPORT_SYMBOL(flush_old_exec);
 914
 915 /*
 916  * Fill the binprm structure from the inode.
 917  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 918  */
 919 int prepare_binprm(struct linux_binprm *bprm)
 920 {
 921         int mode;
 922         struct inode * inode = bprm->file->f_path.dentry->d_inode;
 923         int retval;
 924
 925         mode = inode->i_mode;
 926         if (bprm->file->f_op == NULL)
 927                 return -EACCES;
 928
 929         bprm->e_uid = current->euid;
 930         bprm->e_gid = current->egid;
 931
 932         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
 933                 /* Set-uid? */
 934                 if (mode & S_ISUID) {
 935                         current->personality &= ~PER_CLEAR_ON_SETID;
 936                         bprm->e_uid = inode->i_uid;
 937                 }
 938
 939                 /* Set-gid? */
 940                 /*
 941                  * If setgid is set but no group execute bit then this
 942                  * is a candidate for mandatory locking, not a setgid
 943                  * executable.
 944                  */
 945                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 946                         current->personality &= ~PER_CLEAR_ON_SETID;
 947                         bprm->e_gid = inode->i_gid;
 948                 }
 949         }
 950
 951         /* fill in binprm security blob */
 952         retval = security_bprm_set(bprm);
 953         if (retval)
 954                 return retval;
 955
 956         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 957         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 958 }
 959
 960 EXPORT_SYMBOL(prepare_binprm);
 961
 962 static int unsafe_exec(struct task_struct *p)
 963 {
 964         int unsafe = 0;
 965         if (p->ptrace & PT_PTRACED) {
 966                 if (p->ptrace & PT_PTRACE_CAP)
 967                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 968                 else
 969                         unsafe |= LSM_UNSAFE_PTRACE;
 970         }
 971         if (atomic_read(&p->fs->count) > 1 ||
 972             atomic_read(&p->files->count) > 1 ||
 973             atomic_read(&p->sighand->count) > 1)
 974                 unsafe |= LSM_UNSAFE_SHARE;
 975
 976         return unsafe;
 977 }
 978
 979 void compute_creds(struct linux_binprm *bprm)
 980 {
 981         int unsafe;
 982
 983         if (bprm->e_uid != current->uid)
 984                 suid_keys(current);
 985         exec_keys(current);
 986
 987         task_lock(current);
 988         unsafe = unsafe_exec(current);
 989         security_bprm_apply_creds(bprm, unsafe);
 990         task_unlock(current);
 991         security_bprm_post_apply_creds(bprm);
 992 }
 993 EXPORT_SYMBOL(compute_creds);
 994
 995 /*
 996  * Arguments are '\0' separated strings found at the location bprm->p
 997  * points to; chop off the first by relocating brpm->p to right after
 998  * the first '\0' encountered.
 999  */
1000 void remove_arg_zero(struct linux_binprm *bprm)
1001 {
1002         if (bprm->argc) {
1003                 char ch;
1004
1005                 do {
1006                         unsigned long offset;
1007                         unsigned long index;
1008                         char *kaddr;
1009                         struct page *page;
1010
1011                         offset = bprm->p & ~PAGE_MASK;
1012                         index = bprm->p >> PAGE_SHIFT;
1013
1014                         page = bprm->page[index];
1015                         kaddr = kmap_atomic(page, KM_USER0);
1016
1017                         /* run through page until we reach end or find NUL */
1018                         do {
1019                                 ch = *(kaddr + offset);
1020
1021                                 /* discard that character... */
1022                                 bprm->p++;
1023                                 offset++;
1024                         } while (offset < PAGE_SIZE && ch != '\0');
1025
1026                         kunmap_atomic(kaddr, KM_USER0);
1027
1028                         /* free the old page */
1029                         if (offset == PAGE_SIZE) {
1030                                 __free_page(page);
1031                                 bprm->page[index] = NULL;
1032                         }
1033                 } while (ch != '\0');
1034
1035                 bprm->argc--;
1036         }
1037 }
1038 EXPORT_SYMBOL(remove_arg_zero);
1039
1040 /*
1041  * cycle the list of binary formats handler, until one recognizes the image
1042  */
1043 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1044 {
1045         int try,retval;
1046         struct linux_binfmt *fmt;
1047 #ifdef __alpha__
1048         /* handle /sbin/loader.. */
1049         {
1050             struct exec * eh = (struct exec *) bprm->buf;
1051
1052             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1053                 (eh->fh.f_flags & 0x3000) == 0x3000)
1054             {
1055                 struct file * file;
1056                 unsigned long loader;
1057
1058                 allow_write_access(bprm->file);
1059                 fput(bprm->file);
1060                 bprm->file = NULL;
1061
1062                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1063
1064                 file = open_exec("/sbin/loader");
1065                 retval = PTR_ERR(file);
1066                 if (IS_ERR(file))
1067                         return retval;
1068
1069                 /* Remember if the application is TASO.  */
1070                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1071
1072                 bprm->file = file;
1073                 bprm->loader = loader;
1074                 retval = prepare_binprm(bprm);
1075                 if (retval<0)
1076                         return retval;
1077                 /* should call search_binary_handler recursively here,
1078                    but it does not matter */
1079             }
1080         }
1081 #endif
1082         retval = security_bprm_check(bprm);
1083         if (retval)
1084                 return retval;
1085
1086         /* kernel module loader fixup */
1087         /* so we don't try to load run modprobe in kernel space. */
1088         set_fs(USER_DS);
1089
1090         retval = audit_bprm(bprm);
1091         if (retval)
1092                 return retval;
1093
1094         retval = -ENOENT;
1095         for (try=0; try<2; try++) {
1096                 read_lock(&binfmt_lock);
1097                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1098                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1099                         if (!fn)
1100                                 continue;
1101                         if (!try_module_get(fmt->module))
1102                                 continue;
1103                         read_unlock(&binfmt_lock);
1104                         retval = fn(bprm, regs);
1105                         if (retval >= 0) {
1106                                 put_binfmt(fmt);
1107                                 allow_write_access(bprm->file);
1108                                 if (bprm->file)
1109                                         fput(bprm->file);
1110                                 bprm->file = NULL;
1111                                 current->did_exec = 1;
1112                                 proc_exec_connector(current);
1113                                 return retval;
1114                         }
1115                         read_lock(&binfmt_lock);
1116                         put_binfmt(fmt);
1117                         if (retval != -ENOEXEC || bprm->mm == NULL)
1118                                 break;
1119                         if (!bprm->file) {
1120                                 read_unlock(&binfmt_lock);
1121                                 return retval;
1122                         }
1123                 }
1124                 read_unlock(&binfmt_lock);
1125                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1126                         break;
1127 #ifdef CONFIG_KMOD
1128                 }else{
1129 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1130                         if (printable(bprm->buf[0]) &&
1131                             printable(bprm->buf[1]) &&
1132                             printable(bprm->buf[2]) &&
1133                             printable(bprm->buf[3]))
1134                                 break; /* -ENOEXEC */
1135                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1136 #endif
1137                 }
1138         }
1139         return retval;
1140 }
1141
1142 EXPORT_SYMBOL(search_binary_handler);
1143
1144 /*
1145  * sys_execve() executes a new program.
1146  */
1147 int do_execve(char * filename,
1148         char __user *__user *argv,
1149         char __user *__user *envp,
1150         struct pt_regs * regs)
1151 {
1152         struct linux_binprm *bprm;
1153         struct file *file;
1154         int retval;
1155         int i;
1156
1157         retval = -ENOMEM;
1158         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1159         if (!bprm)
1160                 goto out_ret;
1161
1162         file = open_exec(filename);
1163         retval = PTR_ERR(file);
1164         if (IS_ERR(file))
1165                 goto out_kfree;
1166
1167         sched_exec();
1168
1169         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1170
1171         bprm->file = file;
1172         bprm->filename = filename;
1173         bprm->interp = filename;
1174         bprm->mm = mm_alloc();
1175         retval = -ENOMEM;
1176         if (!bprm->mm)
1177                 goto out_file;
1178
1179         retval = init_new_context(current, bprm->mm);
1180         if (retval < 0)
1181                 goto out_mm;
1182
1183         bprm->argc = count(argv, bprm->p / sizeof(void *));
1184         if ((retval = bprm->argc) < 0)
1185                 goto out_mm;
1186
1187         bprm->envc = count(envp, bprm->p / sizeof(void *));
1188         if ((retval = bprm->envc) < 0)
1189                 goto out_mm;
1190
1191         retval = security_bprm_alloc(bprm);
1192         if (retval)
1193                 goto out;
1194
1195         retval = prepare_binprm(bprm);
1196         if (retval < 0)
1197                 goto out;
1198
1199         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1200         if (retval < 0)
1201                 goto out;
1202
1203         bprm->exec = bprm->p;
1204         retval = copy_strings(bprm->envc, envp, bprm);
1205         if (retval < 0)
1206                 goto out;
1207
1208         retval = copy_strings(bprm->argc, argv, bprm);
1209         if (retval < 0)
1210                 goto out;
1211
1212         retval = search_binary_handler(bprm,regs);
1213         if (retval >= 0) {
1214                 free_arg_pages(bprm);
1215
1216                 /* execve success */
1217                 security_bprm_free(bprm);
1218                 acct_update_integrals(current);
1219                 kfree(bprm);
1220                 return retval;
1221         }
1222
1223 out:
1224         /* Something went wrong, return the inode and free the argument pages*/
1225         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1226                 struct page * page = bprm->page[i];
1227                 if (page)
1228                         __free_page(page);
1229         }
1230
1231         if (bprm->security)
1232                 security_bprm_free(bprm);
1233
1234 out_mm:
1235         if (bprm->mm)
1236                 mmdrop(bprm->mm);
1237
1238 out_file:
1239         if (bprm->file) {
1240                 allow_write_access(bprm->file);
1241                 fput(bprm->file);
1242         }
1243
1244 out_kfree:
1245         kfree(bprm);
1246
1247 out_ret:
1248         return retval;
1249 }
1250
1251 int set_binfmt(struct linux_binfmt *new)
1252 {
1253         struct linux_binfmt *old = current->binfmt;
1254
1255         if (new) {
1256                 if (!try_module_get(new->module))
1257                         return -1;
1258         }
1259         current->binfmt = new;
1260         if (old)
1261                 module_put(old->module);
1262         return 0;
1263 }
1264
1265 EXPORT_SYMBOL(set_binfmt);
1266
1267 /* format_corename will inspect the pattern parameter, and output a
1268  * name into corename, which must have space for at least
1269  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1270  */
1271 static int format_corename(char *corename, const char *pattern, long signr)
1272 {
1273         const char *pat_ptr = pattern;
1274         char *out_ptr = corename;
1275         char *const out_end = corename + CORENAME_MAX_SIZE;
1276         int rc;
1277         int pid_in_pattern = 0;
1278         int ispipe = 0;
1279
1280         if (*pattern == '|')
1281                 ispipe = 1;
1282
1283         /* Repeat as long as we have more pattern to process and more output
1284            space */
1285         while (*pat_ptr) {
1286                 if (*pat_ptr != '%') {
1287                         if (out_ptr == out_end)
1288                                 goto out;
1289                         *out_ptr++ = *pat_ptr++;
1290                 } else {
1291                         switch (*++pat_ptr) {
1292                         case 0:
1293                                 goto out;
1294                         /* Double percent, output one percent */
1295                         case '%':
1296                                 if (out_ptr == out_end)
1297                                         goto out;
1298                                 *out_ptr++ = '%';
1299                                 break;
1300                         /* pid */
1301                         case 'p':
1302                                 pid_in_pattern = 1;
1303                                 rc = snprintf(out_ptr, out_end - out_ptr,
1304                                               "%d", current->tgid);
1305                                 if (rc > out_end - out_ptr)
1306                                         goto out;
1307                                 out_ptr += rc;
1308                                 break;
1309                         /* uid */
1310                         case 'u':
1311                                 rc = snprintf(out_ptr, out_end - out_ptr,
1312                                               "%d", current->uid);
1313                                 if (rc > out_end - out_ptr)
1314                                         goto out;
1315                                 out_ptr += rc;
1316                                 break;
1317                         /* gid */
1318                         case 'g':
1319                                 rc = snprintf(out_ptr, out_end - out_ptr,
1320                                               "%d", current->gid);
1321                                 if (rc > out_end - out_ptr)
1322                                         goto out;
1323                                 out_ptr += rc;
1324                                 break;
1325                         /* signal that caused the coredump */
1326                         case 's':
1327                                 rc = snprintf(out_ptr, out_end - out_ptr,
1328                                               "%ld", signr);
1329                                 if (rc > out_end - out_ptr)
1330                                         goto out;
1331                                 out_ptr += rc;
1332                                 break;
1333                         /* UNIX time of coredump */
1334                         case 't': {
1335                                 struct timeval tv;
1336                                 do_gettimeofday(&tv);
1337                                 rc = snprintf(out_ptr, out_end - out_ptr,
1338                                               "%lu", tv.tv_sec);
1339                                 if (rc > out_end - out_ptr)
1340                                         goto out;
1341                                 out_ptr += rc;
1342                                 break;
1343                         }
1344                         /* hostname */
1345                         case 'h':
1346                                 down_read(&uts_sem);
1347                                 rc = snprintf(out_ptr, out_end - out_ptr,
1348                                               "%s", utsname()->nodename);
1349                                 up_read(&uts_sem);
1350                                 if (rc > out_end - out_ptr)
1351                                         goto out;
1352                                 out_ptr += rc;
1353                                 break;
1354                         /* executable */
1355                         case 'e':
1356                                 rc = snprintf(out_ptr, out_end - out_ptr,
1357                                               "%s", current->comm);
1358                                 if (rc > out_end - out_ptr)
1359                                         goto out;
1360                                 out_ptr += rc;
1361                                 break;
1362                         default:
1363                                 break;
1364                         }
1365                         ++pat_ptr;
1366                 }
1367         }
1368         /* Backward compatibility with core_uses_pid:
1369          *
1370          * If core_pattern does not include a %p (as is the default)
1371          * and core_uses_pid is set, then .%pid will be appended to
1372          * the filename. Do not do this for piped commands. */
1373         if (!ispipe && !pid_in_pattern
1374             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1375                 rc = snprintf(out_ptr, out_end - out_ptr,
1376                               ".%d", current->tgid);
1377                 if (rc > out_end - out_ptr)
1378                         goto out;
1379                 out_ptr += rc;
1380         }
1381 out:
1382         *out_ptr = 0;
1383         return ispipe;
1384 }
1385
1386 static void zap_process(struct task_struct *start)
1387 {
1388         struct task_struct *t;
1389
1390         start->signal->flags = SIGNAL_GROUP_EXIT;
1391         start->signal->group_stop_count = 0;
1392
1393         t = start;
1394         do {
1395                 if (t != current && t->mm) {
1396                         t->mm->core_waiters++;
1397                         sigaddset(&t->pending.signal, SIGKILL);
1398                         signal_wake_up(t, 1);
1399                 }
1400         } while ((t = next_thread(t)) != start);
1401 }
1402
1403 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1404                                 int exit_code)
1405 {
1406         struct task_struct *g, *p;
1407         unsigned long flags;
1408         int err = -EAGAIN;
1409
1410         spin_lock_irq(&tsk->sighand->siglock);
1411         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1412                 tsk->signal->group_exit_code = exit_code;
1413                 zap_process(tsk);
1414                 err = 0;
1415         }
1416         spin_unlock_irq(&tsk->sighand->siglock);
1417         if (err)
1418                 return err;
1419
1420         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1421                 goto done;
1422
1423         rcu_read_lock();
1424         for_each_process(g) {
1425                 if (g == tsk->group_leader)
1426                         continue;
1427
1428                 p = g;
1429                 do {
1430                         if (p->mm) {
1431                                 if (p->mm == mm) {
1432                                         /*
1433                                          * p->sighand can't disappear, but
1434                                          * may be changed by de_thread()
1435                                          */
1436                                         lock_task_sighand(p, &flags);
1437                                         zap_process(p);
1438                                         unlock_task_sighand(p, &flags);
1439                                 }
1440                                 break;
1441                         }
1442                 } while ((p = next_thread(p)) != g);
1443         }
1444         rcu_read_unlock();
1445 done:
1446         return mm->core_waiters;
1447 }
1448
1449 static int coredump_wait(int exit_code)
1450 {
1451         struct task_struct *tsk = current;
1452         struct mm_struct *mm = tsk->mm;
1453         struct completion startup_done;
1454         struct completion *vfork_done;
1455         int core_waiters;
1456
1457         init_completion(&mm->core_done);
1458         init_completion(&startup_done);
1459         mm->core_startup_done = &startup_done;
1460
1461         core_waiters = zap_threads(tsk, mm, exit_code);
1462         up_write(&mm->mmap_sem);
1463
1464         if (unlikely(core_waiters < 0))
1465                 goto fail;
1466
1467         /*
1468          * Make sure nobody is waiting for us to release the VM,
1469          * otherwise we can deadlock when we wait on each other
1470          */
1471         vfork_done = tsk->vfork_done;
1472         if (vfork_done) {
1473                 tsk->vfork_done = NULL;
1474                 complete(vfork_done);
1475         }
1476
1477         if (core_waiters)
1478                 wait_for_completion(&startup_done);
1479 fail:
1480         BUG_ON(mm->core_waiters);
1481         return core_waiters;
1482 }
1483
1484 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1485 {
1486         char corename[CORENAME_MAX_SIZE + 1];
1487         struct mm_struct *mm = current->mm;
1488         struct linux_binfmt * binfmt;
1489         struct inode * inode;
1490         struct file * file;
1491         int retval = 0;
1492         int fsuid = current->fsuid;
1493         int flag = 0;
1494         int ispipe = 0;
1495
1496         audit_core_dumps(signr);
1497
1498         binfmt = current->binfmt;
1499         if (!binfmt || !binfmt->core_dump)
1500                 goto fail;
1501         down_write(&mm->mmap_sem);
1502         if (!mm->dumpable) {
1503                 up_write(&mm->mmap_sem);
1504                 goto fail;
1505         }
1506
1507         /*
1508          *      We cannot trust fsuid as being the "true" uid of the
1509          *      process nor do we know its entire history. We only know it
1510          *      was tainted so we dump it as root in mode 2.
1511          */
1512         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1513                 flag = O_EXCL;          /* Stop rewrite attacks */
1514                 current->fsuid = 0;     /* Dump root private */
1515         }
1516         mm->dumpable = 0;
1517
1518         retval = coredump_wait(exit_code);
1519         if (retval < 0)
1520                 goto fail;
1521
1522         /*
1523          * Clear any false indication of pending signals that might
1524          * be seen by the filesystem code called to write the core file.
1525          */
1526         clear_thread_flag(TIF_SIGPENDING);
1527
1528         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1529                 goto fail_unlock;
1530
1531         /*
1532          * lock_kernel() because format_corename() is controlled by sysctl, which
1533          * uses lock_kernel()
1534          */
1535         lock_kernel();
1536         ispipe = format_corename(corename, core_pattern, signr);
1537         unlock_kernel();
1538         if (ispipe) {
1539                 /* SIGPIPE can happen, but it's just never processed */
1540                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1541                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1542                                corename);
1543                         goto fail_unlock;
1544                 }
1545         } else
1546                 file = filp_open(corename,
1547                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1548                                  0600);
1549         if (IS_ERR(file))
1550                 goto fail_unlock;
1551         inode = file->f_path.dentry->d_inode;
1552         if (inode->i_nlink > 1)
1553                 goto close_fail;        /* multiple links - don't dump */
1554         if (!ispipe && d_unhashed(file->f_path.dentry))
1555                 goto close_fail;
1556
1557         /* AK: actually i see no reason to not allow this for named pipes etc.,
1558            but keep the previous behaviour for now. */
1559         if (!ispipe && !S_ISREG(inode->i_mode))
1560                 goto close_fail;
1561         if (!file->f_op)
1562                 goto close_fail;
1563         if (!file->f_op->write)
1564                 goto close_fail;
1565         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1566                 goto close_fail;
1567
1568         retval = binfmt->core_dump(signr, regs, file);
1569
1570         if (retval)
1571                 current->signal->group_exit_code |= 0x80;
1572 close_fail:
1573         filp_close(file, NULL);
1574 fail_unlock:
1575         current->fsuid = fsuid;
1576         complete_all(&mm->core_done);
1577 fail:
1578         return retval;
1579 }