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 = -EACCES;
 138         if (nd.mnt->mnt_flags & MNT_NOEXEC)
 139                 goto exit;
 140         error = -EINVAL;
 141         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 142                 goto exit;
 143
 144         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 145         if (error)
 146                 goto exit;
 147
 148         file = nameidata_to_filp(&nd, O_RDONLY);
 149         error = PTR_ERR(file);
 150         if (IS_ERR(file))
 151                 goto out;
 152
 153         error = -ENOEXEC;
 154         if(file->f_op) {
 155                 struct linux_binfmt * fmt;
 156
 157                 read_lock(&binfmt_lock);
 158                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 159                         if (!fmt->load_shlib)
 160                                 continue;
 161                         if (!try_module_get(fmt->module))
 162                                 continue;
 163                         read_unlock(&binfmt_lock);
 164                         error = fmt->load_shlib(file);
 165                         read_lock(&binfmt_lock);
 166                         put_binfmt(fmt);
 167                         if (error != -ENOEXEC)
 168                                 break;
 169                 }
 170                 read_unlock(&binfmt_lock);
 171         }
 172         fput(file);
 173 out:
 174         return error;
 175 exit:
 176         release_open_intent(&nd);
 177         path_release(&nd);
 178         goto out;
 179 }
 180
 181 /*
 182  * count() counts the number of strings in array ARGV.
 183  */
 184 static int count(char __user * __user * argv, int max)
 185 {
 186         int i = 0;
 187
 188         if (argv != NULL) {
 189                 for (;;) {
 190                         char __user * p;
 191
 192                         if (get_user(p, argv))
 193                                 return -EFAULT;
 194                         if (!p)
 195                                 break;
 196                         argv++;
 197                         if(++i > max)
 198                                 return -E2BIG;
 199                         cond_resched();
 200                 }
 201         }
 202         return i;
 203 }
 204
 205 /*
 206  * 'copy_strings()' copies argument/environment strings from user
 207  * memory to free pages in kernel mem. These are in a format ready
 208  * to be put directly into the top of new user memory.
 209  */
 210 static int copy_strings(int argc, char __user * __user * argv,
 211                         struct linux_binprm *bprm)
 212 {
 213         struct page *kmapped_page = NULL;
 214         char *kaddr = NULL;
 215         int ret;
 216
 217         while (argc-- > 0) {
 218                 char __user *str;
 219                 int len;
 220                 unsigned long pos;
 221
 222                 if (get_user(str, argv+argc) ||
 223                                 !(len = strnlen_user(str, bprm->p))) {
 224                         ret = -EFAULT;
 225                         goto out;
 226                 }
 227
 228                 if (bprm->p < len)  {
 229                         ret = -E2BIG;
 230                         goto out;
 231                 }
 232
 233                 bprm->p -= len;
 234                 /* XXX: add architecture specific overflow check here. */
 235                 pos = bprm->p;
 236
 237                 while (len > 0) {
 238                         int i, new, err;
 239                         int offset, bytes_to_copy;
 240                         struct page *page;
 241
 242                         offset = pos % PAGE_SIZE;
 243                         i = pos/PAGE_SIZE;
 244                         page = bprm->page[i];
 245                         new = 0;
 246                         if (!page) {
 247                                 page = alloc_page(GFP_HIGHUSER);
 248                                 bprm->page[i] = page;
 249                                 if (!page) {
 250                                         ret = -ENOMEM;
 251                                         goto out;
 252                                 }
 253                                 new = 1;
 254                         }
 255
 256                         if (page != kmapped_page) {
 257                                 if (kmapped_page)
 258                                         kunmap(kmapped_page);
 259                                 kmapped_page = page;
 260                                 kaddr = kmap(kmapped_page);
 261                         }
 262                         if (new && offset)
 263                                 memset(kaddr, 0, offset);
 264                         bytes_to_copy = PAGE_SIZE - offset;
 265                         if (bytes_to_copy > len) {
 266                                 bytes_to_copy = len;
 267                                 if (new)
 268                                         memset(kaddr+offset+len, 0,
 269                                                 PAGE_SIZE-offset-len);
 270                         }
 271                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 272                         if (err) {
 273                                 ret = -EFAULT;
 274                                 goto out;
 275                         }
 276
 277                         pos += bytes_to_copy;
 278                         str += bytes_to_copy;
 279                         len -= bytes_to_copy;
 280                 }
 281         }
 282         ret = 0;
 283 out:
 284         if (kmapped_page)
 285                 kunmap(kmapped_page);
 286         return ret;
 287 }
 288
 289 /*
 290  * Like copy_strings, but get argv and its values from kernel memory.
 291  */
 292 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 293 {
 294         int r;
 295         mm_segment_t oldfs = get_fs();
 296         set_fs(KERNEL_DS);
 297         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 298         set_fs(oldfs);
 299         return r;
 300 }
 301
 302 EXPORT_SYMBOL(copy_strings_kernel);
 303
 304 #ifdef CONFIG_MMU
 305 /*
 306  * This routine is used to map in a page into an address space: needed by
 307  * execve() for the initial stack and environment pages.
 308  *
 309  * vma->vm_mm->mmap_sem is held for writing.
 310  */
 311 void install_arg_page(struct vm_area_struct *vma,
 312                         struct page *page, unsigned long address)
 313 {
 314         struct mm_struct *mm = vma->vm_mm;
 315         pte_t * pte;
 316         spinlock_t *ptl;
 317
 318         if (unlikely(anon_vma_prepare(vma)))
 319                 goto out;
 320
 321         flush_dcache_page(page);
 322         pte = get_locked_pte(mm, address, &ptl);
 323         if (!pte)
 324                 goto out;
 325         if (!pte_none(*pte)) {
 326                 pte_unmap_unlock(pte, ptl);
 327                 goto out;
 328         }
 329         inc_mm_counter(mm, anon_rss);
 330         lru_cache_add_active(page);
 331         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
 332                                         page, vma->vm_page_prot))));
 333         page_add_new_anon_rmap(page, vma, address);
 334         pte_unmap_unlock(pte, ptl);
 335
 336         /* no need for flush_tlb */
 337         return;
 338 out:
 339         __free_page(page);
 340         force_sig(SIGKILL, current);
 341 }
 342
 343 #define EXTRA_STACK_VM_PAGES    20      /* random */
 344
 345 int setup_arg_pages(struct linux_binprm *bprm,
 346                     unsigned long stack_top,
 347                     int executable_stack)
 348 {
 349         unsigned long stack_base;
 350         struct vm_area_struct *mpnt;
 351         struct mm_struct *mm = current->mm;
 352         int i, ret;
 353         long arg_size;
 354
 355 #ifdef CONFIG_STACK_GROWSUP
 356         /* Move the argument and environment strings to the bottom of the
 357          * stack space.
 358          */
 359         int offset, j;
 360         char *to, *from;
 361
 362         /* Start by shifting all the pages down */
 363         i = 0;
 364         for (j = 0; j < MAX_ARG_PAGES; j++) {
 365                 struct page *page = bprm->page[j];
 366                 if (!page)
 367                         continue;
 368                 bprm->page[i++] = page;
 369         }
 370
 371         /* Now move them within their pages */
 372         offset = bprm->p % PAGE_SIZE;
 373         to = kmap(bprm->page[0]);
 374         for (j = 1; j < i; j++) {
 375                 memmove(to, to + offset, PAGE_SIZE - offset);
 376                 from = kmap(bprm->page[j]);
 377                 memcpy(to + PAGE_SIZE - offset, from, offset);
 378                 kunmap(bprm->page[j - 1]);
 379                 to = from;
 380         }
 381         memmove(to, to + offset, PAGE_SIZE - offset);
 382         kunmap(bprm->page[j - 1]);
 383
 384         /* Limit stack size to 1GB */
 385         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 386         if (stack_base > (1 << 30))
 387                 stack_base = 1 << 30;
 388         stack_base = PAGE_ALIGN(stack_top - stack_base);
 389
 390         /* Adjust bprm->p to point to the end of the strings. */
 391         bprm->p = stack_base + PAGE_SIZE * i - offset;
 392
 393         mm->arg_start = stack_base;
 394         arg_size = i << PAGE_SHIFT;
 395
 396         /* zero pages that were copied above */
 397         while (i < MAX_ARG_PAGES)
 398                 bprm->page[i++] = NULL;
 399 #else
 400         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
 401         stack_base = PAGE_ALIGN(stack_base);
 402         bprm->p += stack_base;
 403         mm->arg_start = bprm->p;
 404         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 405 #endif
 406
 407         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 408
 409         if (bprm->loader)
 410                 bprm->loader += stack_base;
 411         bprm->exec += stack_base;
 412
 413         mpnt = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 414         if (!mpnt)
 415                 return -ENOMEM;
 416
 417         down_write(&mm->mmap_sem);
 418         {
 419                 mpnt->vm_mm = mm;
 420 #ifdef CONFIG_STACK_GROWSUP
 421                 mpnt->vm_start = stack_base;
 422                 mpnt->vm_end = stack_base + arg_size;
 423 #else
 424                 mpnt->vm_end = stack_top;
 425                 mpnt->vm_start = mpnt->vm_end - arg_size;
 426 #endif
 427                 /* Adjust stack execute permissions; explicitly enable
 428                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 429                  * and leave alone (arch default) otherwise. */
 430                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 431                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 432                 else if (executable_stack == EXSTACK_DISABLE_X)
 433                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 434                 else
 435                         mpnt->vm_flags = VM_STACK_FLAGS;
 436                 mpnt->vm_flags |= mm->def_flags;
 437                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 438                 if ((ret = insert_vm_struct(mm, mpnt))) {
 439                         up_write(&mm->mmap_sem);
 440                         kmem_cache_free(vm_area_cachep, mpnt);
 441                         return ret;
 442                 }
 443                 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 444         }
 445
 446         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 447                 struct page *page = bprm->page[i];
 448                 if (page) {
 449                         bprm->page[i] = NULL;
 450                         install_arg_page(mpnt, page, stack_base);
 451                 }
 452                 stack_base += PAGE_SIZE;
 453         }
 454         up_write(&mm->mmap_sem);
 455
 456         return 0;
 457 }
 458
 459 EXPORT_SYMBOL(setup_arg_pages);
 460
 461 #define free_arg_pages(bprm) do { } while (0)
 462
 463 #else
 464
 465 static inline void free_arg_pages(struct linux_binprm *bprm)
 466 {
 467         int i;
 468
 469         for (i = 0; i < MAX_ARG_PAGES; i++) {
 470                 if (bprm->page[i])
 471                         __free_page(bprm->page[i]);
 472                 bprm->page[i] = NULL;
 473         }
 474 }
 475
 476 #endif /* CONFIG_MMU */
 477
 478 struct file *open_exec(const char *name)
 479 {
 480         struct nameidata nd;
 481         int err;
 482         struct file *file;
 483
 484         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 485         file = ERR_PTR(err);
 486
 487         if (!err) {
 488                 struct inode *inode = nd.dentry->d_inode;
 489                 file = ERR_PTR(-EACCES);
 490                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 491                     S_ISREG(inode->i_mode)) {
 492                         int err = vfs_permission(&nd, MAY_EXEC);
 493                         file = ERR_PTR(err);
 494                         if (!err) {
 495                                 file = nameidata_to_filp(&nd, O_RDONLY);
 496                                 if (!IS_ERR(file)) {
 497                                         err = deny_write_access(file);
 498                                         if (err) {
 499                                                 fput(file);
 500                                                 file = ERR_PTR(err);
 501                                         }
 502                                 }
 503 out:
 504                                 return file;
 505                         }
 506                 }
 507                 release_open_intent(&nd);
 508                 path_release(&nd);
 509         }
 510         goto out;
 511 }
 512
 513 EXPORT_SYMBOL(open_exec);
 514
 515 int kernel_read(struct file *file, unsigned long offset,
 516         char *addr, unsigned long count)
 517 {
 518         mm_segment_t old_fs;
 519         loff_t pos = offset;
 520         int result;
 521
 522         old_fs = get_fs();
 523         set_fs(get_ds());
 524         /* The cast to a user pointer is valid due to the set_fs() */
 525         result = vfs_read(file, (void __user *)addr, count, &pos);
 526         set_fs(old_fs);
 527         return result;
 528 }
 529
 530 EXPORT_SYMBOL(kernel_read);
 531
 532 static int exec_mmap(struct mm_struct *mm)
 533 {
 534         struct task_struct *tsk;
 535         struct mm_struct * old_mm, *active_mm;
 536
 537         /* Notify parent that we're no longer interested in the old VM */
 538         tsk = current;
 539         old_mm = current->mm;
 540         mm_release(tsk, old_mm);
 541
 542         if (old_mm) {
 543                 /*
 544                  * Make sure that if there is a core dump in progress
 545                  * for the old mm, we get out and die instead of going
 546                  * through with the exec.  We must hold mmap_sem around
 547                  * checking core_waiters and changing tsk->mm.  The
 548                  * core-inducing thread will increment core_waiters for
 549                  * each thread whose ->mm == old_mm.
 550                  */
 551                 down_read(&old_mm->mmap_sem);
 552                 if (unlikely(old_mm->core_waiters)) {
 553                         up_read(&old_mm->mmap_sem);
 554                         return -EINTR;
 555                 }
 556         }
 557         task_lock(tsk);
 558         active_mm = tsk->active_mm;
 559         tsk->mm = mm;
 560         tsk->active_mm = mm;
 561         activate_mm(active_mm, mm);
 562         task_unlock(tsk);
 563         arch_pick_mmap_layout(mm);
 564         if (old_mm) {
 565                 up_read(&old_mm->mmap_sem);
 566                 BUG_ON(active_mm != old_mm);
 567                 mmput(old_mm);
 568                 return 0;
 569         }
 570         mmdrop(active_mm);
 571         return 0;
 572 }
 573
 574 /*
 575  * This function makes sure the current process has its own signal table,
 576  * so that flush_signal_handlers can later reset the handlers without
 577  * disturbing other processes.  (Other processes might share the signal
 578  * table via the CLONE_SIGHAND option to clone().)
 579  */
 580 static int de_thread(struct task_struct *tsk)
 581 {
 582         struct signal_struct *sig = tsk->signal;
 583         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 584         spinlock_t *lock = &oldsighand->siglock;
 585         struct task_struct *leader = NULL;
 586         int count;
 587
 588         /*
 589          * If we don't share sighandlers, then we aren't sharing anything
 590          * and we can just re-use it all.
 591          */
 592         if (atomic_read(&oldsighand->count) <= 1) {
 593                 BUG_ON(atomic_read(&sig->count) != 1);
 594                 signalfd_detach(tsk);
 595                 exit_itimers(sig);
 596                 return 0;
 597         }
 598
 599         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 600         if (!newsighand)
 601                 return -ENOMEM;
 602
 603         if (thread_group_empty(tsk))
 604                 goto no_thread_group;
 605
 606         /*
 607          * Kill all other threads in the thread group.
 608          * We must hold tasklist_lock to call zap_other_threads.
 609          */
 610         read_lock(&tasklist_lock);
 611         spin_lock_irq(lock);
 612         if (sig->flags & SIGNAL_GROUP_EXIT) {
 613                 /*
 614                  * Another group action in progress, just
 615                  * return so that the signal is processed.
 616                  */
 617                 spin_unlock_irq(lock);
 618                 read_unlock(&tasklist_lock);
 619                 kmem_cache_free(sighand_cachep, newsighand);
 620                 return -EAGAIN;
 621         }
 622
 623         /*
 624          * child_reaper ignores SIGKILL, change it now.
 625          * Reparenting needs write_lock on tasklist_lock,
 626          * so it is safe to do it under read_lock.
 627          */
 628         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 629                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 630
 631         zap_other_threads(tsk);
 632         read_unlock(&tasklist_lock);
 633
 634         /*
 635          * Account for the thread group leader hanging around:
 636          */
 637         count = 1;
 638         if (!thread_group_leader(tsk)) {
 639                 count = 2;
 640                 /*
 641                  * The SIGALRM timer survives the exec, but needs to point
 642                  * at us as the new group leader now.  We have a race with
 643                  * a timer firing now getting the old leader, so we need to
 644                  * synchronize with any firing (by calling del_timer_sync)
 645                  * before we can safely let the old group leader die.
 646                  */
 647                 sig->tsk = tsk;
 648                 spin_unlock_irq(lock);
 649                 if (hrtimer_cancel(&sig->real_timer))
 650                         hrtimer_restart(&sig->real_timer);
 651                 spin_lock_irq(lock);
 652         }
 653         while (atomic_read(&sig->count) > count) {
 654                 sig->group_exit_task = tsk;
 655                 sig->notify_count = count;
 656                 __set_current_state(TASK_UNINTERRUPTIBLE);
 657                 spin_unlock_irq(lock);
 658                 schedule();
 659                 spin_lock_irq(lock);
 660         }
 661         sig->group_exit_task = NULL;
 662         sig->notify_count = 0;
 663         spin_unlock_irq(lock);
 664
 665         /*
 666          * At this point all other threads have exited, all we have to
 667          * do is to wait for the thread group leader to become inactive,
 668          * and to assume its PID:
 669          */
 670         if (!thread_group_leader(tsk)) {
 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 = tsk->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                 tsk->start_time = leader->start_time;
 691
 692                 write_lock_irq(&tasklist_lock);
 693
 694                 BUG_ON(leader->tgid != tsk->tgid);
 695                 BUG_ON(tsk->pid == tsk->tgid);
 696                 /*
 697                  * An exec() starts a new thread group with the
 698                  * TGID of the previous thread group. Rehash the
 699                  * two threads with a switched PID, and release
 700                  * the former thread group leader:
 701                  */
 702
 703                 /* Become a process group leader with the old leader's pid.
 704                  * The old leader becomes a thread of the this thread group.
 705                  * Note: The old leader also uses this pid until release_task
 706                  *       is called.  Odd but simple and correct.
 707                  */
 708                 detach_pid(tsk, PIDTYPE_PID);
 709                 tsk->pid = leader->pid;
 710                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
 711                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 712                 transfer_pid(leader, tsk, PIDTYPE_SID);
 713                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 714
 715                 tsk->group_leader = tsk;
 716                 leader->group_leader = tsk;
 717
 718                 tsk->exit_signal = SIGCHLD;
 719
 720                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 721                 leader->exit_state = EXIT_DEAD;
 722
 723                 write_unlock_irq(&tasklist_lock);
 724         }
 725
 726         /*
 727          * There may be one thread left which is just exiting,
 728          * but it's safe to stop telling the group to kill themselves.
 729          */
 730         sig->flags = 0;
 731
 732 no_thread_group:
 733         signalfd_detach(tsk);
 734         exit_itimers(sig);
 735         if (leader)
 736                 release_task(leader);
 737
 738         BUG_ON(atomic_read(&sig->count) != 1);
 739
 740         if (atomic_read(&oldsighand->count) == 1) {
 741                 /*
 742                  * Now that we nuked the rest of the thread group,
 743                  * it turns out we are not sharing sighand any more either.
 744                  * So we can just keep it.
 745                  */
 746                 kmem_cache_free(sighand_cachep, newsighand);
 747         } else {
 748                 /*
 749                  * Move our state over to newsighand and switch it in.
 750                  */
 751                 atomic_set(&newsighand->count, 1);
 752                 memcpy(newsighand->action, oldsighand->action,
 753                        sizeof(newsighand->action));
 754
 755                 write_lock_irq(&tasklist_lock);
 756                 spin_lock(&oldsighand->siglock);
 757                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 758
 759                 rcu_assign_pointer(tsk->sighand, newsighand);
 760                 recalc_sigpending();
 761
 762                 spin_unlock(&newsighand->siglock);
 763                 spin_unlock(&oldsighand->siglock);
 764                 write_unlock_irq(&tasklist_lock);
 765
 766                 __cleanup_sighand(oldsighand);
 767         }
 768
 769         BUG_ON(!thread_group_leader(tsk));
 770         return 0;
 771 }
 772
 773 /*
 774  * These functions flushes out all traces of the currently running executable
 775  * so that a new one can be started
 776  */
 777
 778 static void flush_old_files(struct files_struct * files)
 779 {
 780         long j = -1;
 781         struct fdtable *fdt;
 782
 783         spin_lock(&files->file_lock);
 784         for (;;) {
 785                 unsigned long set, i;
 786
 787                 j++;
 788                 i = j * __NFDBITS;
 789                 fdt = files_fdtable(files);
 790                 if (i >= fdt->max_fds)
 791                         break;
 792                 set = fdt->close_on_exec->fds_bits[j];
 793                 if (!set)
 794                         continue;
 795                 fdt->close_on_exec->fds_bits[j] = 0;
 796                 spin_unlock(&files->file_lock);
 797                 for ( ; set ; i++,set >>= 1) {
 798                         if (set & 1) {
 799                                 sys_close(i);
 800                         }
 801                 }
 802                 spin_lock(&files->file_lock);
 803
 804         }
 805         spin_unlock(&files->file_lock);
 806 }
 807
 808 void get_task_comm(char *buf, struct task_struct *tsk)
 809 {
 810         /* buf must be at least sizeof(tsk->comm) in size */
 811         task_lock(tsk);
 812         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 813         task_unlock(tsk);
 814 }
 815
 816 void set_task_comm(struct task_struct *tsk, char *buf)
 817 {
 818         task_lock(tsk);
 819         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 820         task_unlock(tsk);
 821 }
 822
 823 int flush_old_exec(struct linux_binprm * bprm)
 824 {
 825         char * name;
 826         int i, ch, retval;
 827         struct files_struct *files;
 828         char tcomm[sizeof(current->comm)];
 829
 830         /*
 831          * Make sure we have a private signal table and that
 832          * we are unassociated from the previous thread group.
 833          */
 834         retval = de_thread(current);
 835         if (retval)
 836                 goto out;
 837
 838         /*
 839          * Make sure we have private file handles. Ask the
 840          * fork helper to do the work for us and the exit
 841          * helper to do the cleanup of the old one.
 842          */
 843         files = current->files;         /* refcounted so safe to hold */
 844         retval = unshare_files();
 845         if (retval)
 846                 goto out;
 847         /*
 848          * Release all of the old mmap stuff
 849          */
 850         retval = exec_mmap(bprm->mm);
 851         if (retval)
 852                 goto mmap_failed;
 853
 854         bprm->mm = NULL;                /* We're using it now */
 855
 856         /* This is the point of no return */
 857         put_files_struct(files);
 858
 859         current->sas_ss_sp = current->sas_ss_size = 0;
 860
 861         if (current->euid == current->uid && current->egid == current->gid)
 862                 current->mm->dumpable = 1;
 863         else
 864                 current->mm->dumpable = suid_dumpable;
 865
 866         name = bprm->filename;
 867
 868         /* Copies the binary name from after last slash */
 869         for (i=0; (ch = *(name++)) != '\0';) {
 870                 if (ch == '/')
 871                         i = 0; /* overwrite what we wrote */
 872                 else
 873                         if (i < (sizeof(tcomm) - 1))
 874                                 tcomm[i++] = ch;
 875         }
 876         tcomm[i] = '\0';
 877         set_task_comm(current, tcomm);
 878
 879         current->flags &= ~PF_RANDOMIZE;
 880         flush_thread();
 881
 882         /* Set the new mm task size. We have to do that late because it may
 883          * depend on TIF_32BIT which is only updated in flush_thread() on
 884          * some architectures like powerpc
 885          */
 886         current->mm->task_size = TASK_SIZE;
 887
 888         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
 889                 suid_keys(current);
 890                 current->mm->dumpable = suid_dumpable;
 891                 current->pdeath_signal = 0;
 892         } else if (file_permission(bprm->file, MAY_READ) ||
 893                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 894                 suid_keys(current);
 895                 current->mm->dumpable = suid_dumpable;
 896         }
 897
 898         /* An exec changes our domain. We are no longer part of the thread
 899            group */
 900
 901         current->self_exec_id++;
 902
 903         flush_signal_handlers(current, 0);
 904         flush_old_files(current->files);
 905
 906         return 0;
 907
 908 mmap_failed:
 909         reset_files_struct(current, files);
 910 out:
 911         return retval;
 912 }
 913
 914 EXPORT_SYMBOL(flush_old_exec);
 915
 916 /*
 917  * Fill the binprm structure from the inode.
 918  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 919  */
 920 int prepare_binprm(struct linux_binprm *bprm)
 921 {
 922         int mode;
 923         struct inode * inode = bprm->file->f_path.dentry->d_inode;
 924         int retval;
 925
 926         mode = inode->i_mode;
 927         if (bprm->file->f_op == NULL)
 928                 return -EACCES;
 929
 930         bprm->e_uid = current->euid;
 931         bprm->e_gid = current->egid;
 932
 933         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
 934                 /* Set-uid? */
 935                 if (mode & S_ISUID) {
 936                         current->personality &= ~PER_CLEAR_ON_SETID;
 937                         bprm->e_uid = inode->i_uid;
 938                 }
 939
 940                 /* Set-gid? */
 941                 /*
 942                  * If setgid is set but no group execute bit then this
 943                  * is a candidate for mandatory locking, not a setgid
 944                  * executable.
 945                  */
 946                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 947                         current->personality &= ~PER_CLEAR_ON_SETID;
 948                         bprm->e_gid = inode->i_gid;
 949                 }
 950         }
 951
 952         /* fill in binprm security blob */
 953         retval = security_bprm_set(bprm);
 954         if (retval)
 955                 return retval;
 956
 957         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 958         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 959 }
 960
 961 EXPORT_SYMBOL(prepare_binprm);
 962
 963 static int unsafe_exec(struct task_struct *p)
 964 {
 965         int unsafe = 0;
 966         if (p->ptrace & PT_PTRACED) {
 967                 if (p->ptrace & PT_PTRACE_CAP)
 968                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 969                 else
 970                         unsafe |= LSM_UNSAFE_PTRACE;
 971         }
 972         if (atomic_read(&p->fs->count) > 1 ||
 973             atomic_read(&p->files->count) > 1 ||
 974             atomic_read(&p->sighand->count) > 1)
 975                 unsafe |= LSM_UNSAFE_SHARE;
 976
 977         return unsafe;
 978 }
 979
 980 void compute_creds(struct linux_binprm *bprm)
 981 {
 982         int unsafe;
 983
 984         if (bprm->e_uid != current->uid) {
 985                 suid_keys(current);
 986                 current->pdeath_signal = 0;
 987         }
 988         exec_keys(current);
 989
 990         task_lock(current);
 991         unsafe = unsafe_exec(current);
 992         security_bprm_apply_creds(bprm, unsafe);
 993         task_unlock(current);
 994         security_bprm_post_apply_creds(bprm);
 995 }
 996 EXPORT_SYMBOL(compute_creds);
 997
 998 /*
 999  * Arguments are '\0' separated strings found at the location bprm->p
1000  * points to; chop off the first by relocating brpm->p to right after
1001  * the first '\0' encountered.
1002  */
1003 void remove_arg_zero(struct linux_binprm *bprm)
1004 {
1005         if (bprm->argc) {
1006                 char ch;
1007
1008                 do {
1009                         unsigned long offset;
1010                         unsigned long index;
1011                         char *kaddr;
1012                         struct page *page;
1013
1014                         offset = bprm->p & ~PAGE_MASK;
1015                         index = bprm->p >> PAGE_SHIFT;
1016
1017                         page = bprm->page[index];
1018                         kaddr = kmap_atomic(page, KM_USER0);
1019
1020                         /* run through page until we reach end or find NUL */
1021                         do {
1022                                 ch = *(kaddr + offset);
1023
1024                                 /* discard that character... */
1025                                 bprm->p++;
1026                                 offset++;
1027                         } while (offset < PAGE_SIZE && ch != '\0');
1028
1029                         kunmap_atomic(kaddr, KM_USER0);
1030
1031                         /* free the old page */
1032                         if (offset == PAGE_SIZE) {
1033                                 __free_page(page);
1034                                 bprm->page[index] = NULL;
1035                         }
1036                 } while (ch != '\0');
1037
1038                 bprm->argc--;
1039         }
1040 }
1041 EXPORT_SYMBOL(remove_arg_zero);
1042
1043 /*
1044  * cycle the list of binary formats handler, until one recognizes the image
1045  */
1046 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1047 {
1048         int try,retval;
1049         struct linux_binfmt *fmt;
1050 #ifdef __alpha__
1051         /* handle /sbin/loader.. */
1052         {
1053             struct exec * eh = (struct exec *) bprm->buf;
1054
1055             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1056                 (eh->fh.f_flags & 0x3000) == 0x3000)
1057             {
1058                 struct file * file;
1059                 unsigned long loader;
1060
1061                 allow_write_access(bprm->file);
1062                 fput(bprm->file);
1063                 bprm->file = NULL;
1064
1065                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1066
1067                 file = open_exec("/sbin/loader");
1068                 retval = PTR_ERR(file);
1069                 if (IS_ERR(file))
1070                         return retval;
1071
1072                 /* Remember if the application is TASO.  */
1073                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1074
1075                 bprm->file = file;
1076                 bprm->loader = loader;
1077                 retval = prepare_binprm(bprm);
1078                 if (retval<0)
1079                         return retval;
1080                 /* should call search_binary_handler recursively here,
1081                    but it does not matter */
1082             }
1083         }
1084 #endif
1085         retval = security_bprm_check(bprm);
1086         if (retval)
1087                 return retval;
1088
1089         /* kernel module loader fixup */
1090         /* so we don't try to load run modprobe in kernel space. */
1091         set_fs(USER_DS);
1092
1093         retval = audit_bprm(bprm);
1094         if (retval)
1095                 return retval;
1096
1097         retval = -ENOENT;
1098         for (try=0; try<2; try++) {
1099                 read_lock(&binfmt_lock);
1100                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1101                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1102                         if (!fn)
1103                                 continue;
1104                         if (!try_module_get(fmt->module))
1105                                 continue;
1106                         read_unlock(&binfmt_lock);
1107                         retval = fn(bprm, regs);
1108                         if (retval >= 0) {
1109                                 put_binfmt(fmt);
1110                                 allow_write_access(bprm->file);
1111                                 if (bprm->file)
1112                                         fput(bprm->file);
1113                                 bprm->file = NULL;
1114                                 current->did_exec = 1;
1115                                 proc_exec_connector(current);
1116                                 return retval;
1117                         }
1118                         read_lock(&binfmt_lock);
1119                         put_binfmt(fmt);
1120                         if (retval != -ENOEXEC || bprm->mm == NULL)
1121                                 break;
1122                         if (!bprm->file) {
1123                                 read_unlock(&binfmt_lock);
1124                                 return retval;
1125                         }
1126                 }
1127                 read_unlock(&binfmt_lock);
1128                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1129                         break;
1130 #ifdef CONFIG_KMOD
1131                 }else{
1132 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1133                         if (printable(bprm->buf[0]) &&
1134                             printable(bprm->buf[1]) &&
1135                             printable(bprm->buf[2]) &&
1136                             printable(bprm->buf[3]))
1137                                 break; /* -ENOEXEC */
1138                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1139 #endif
1140                 }
1141         }
1142         return retval;
1143 }
1144
1145 EXPORT_SYMBOL(search_binary_handler);
1146
1147 /*
1148  * sys_execve() executes a new program.
1149  */
1150 int do_execve(char * filename,
1151         char __user *__user *argv,
1152         char __user *__user *envp,
1153         struct pt_regs * regs)
1154 {
1155         struct linux_binprm *bprm;
1156         struct file *file;
1157         int retval;
1158         int i;
1159
1160         retval = -ENOMEM;
1161         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1162         if (!bprm)
1163                 goto out_ret;
1164
1165         file = open_exec(filename);
1166         retval = PTR_ERR(file);
1167         if (IS_ERR(file))
1168                 goto out_kfree;
1169
1170         sched_exec();
1171
1172         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1173
1174         bprm->file = file;
1175         bprm->filename = filename;
1176         bprm->interp = filename;
1177         bprm->mm = mm_alloc();
1178         retval = -ENOMEM;
1179         if (!bprm->mm)
1180                 goto out_file;
1181
1182         retval = init_new_context(current, bprm->mm);
1183         if (retval < 0)
1184                 goto out_mm;
1185
1186         bprm->argc = count(argv, bprm->p / sizeof(void *));
1187         if ((retval = bprm->argc) < 0)
1188                 goto out_mm;
1189
1190         bprm->envc = count(envp, bprm->p / sizeof(void *));
1191         if ((retval = bprm->envc) < 0)
1192                 goto out_mm;
1193
1194         retval = security_bprm_alloc(bprm);
1195         if (retval)
1196                 goto out;
1197
1198         retval = prepare_binprm(bprm);
1199         if (retval < 0)
1200                 goto out;
1201
1202         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1203         if (retval < 0)
1204                 goto out;
1205
1206         bprm->exec = bprm->p;
1207         retval = copy_strings(bprm->envc, envp, bprm);
1208         if (retval < 0)
1209                 goto out;
1210
1211         retval = copy_strings(bprm->argc, argv, bprm);
1212         if (retval < 0)
1213                 goto out;
1214
1215         retval = search_binary_handler(bprm,regs);
1216         if (retval >= 0) {
1217                 free_arg_pages(bprm);
1218
1219                 /* execve success */
1220                 security_bprm_free(bprm);
1221                 acct_update_integrals(current);
1222                 kfree(bprm);
1223                 return retval;
1224         }
1225
1226 out:
1227         /* Something went wrong, return the inode and free the argument pages*/
1228         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1229                 struct page * page = bprm->page[i];
1230                 if (page)
1231                         __free_page(page);
1232         }
1233
1234         if (bprm->security)
1235                 security_bprm_free(bprm);
1236
1237 out_mm:
1238         if (bprm->mm)
1239                 mmdrop(bprm->mm);
1240
1241 out_file:
1242         if (bprm->file) {
1243                 allow_write_access(bprm->file);
1244                 fput(bprm->file);
1245         }
1246
1247 out_kfree:
1248         kfree(bprm);
1249
1250 out_ret:
1251         return retval;
1252 }
1253
1254 int set_binfmt(struct linux_binfmt *new)
1255 {
1256         struct linux_binfmt *old = current->binfmt;
1257
1258         if (new) {
1259                 if (!try_module_get(new->module))
1260                         return -1;
1261         }
1262         current->binfmt = new;
1263         if (old)
1264                 module_put(old->module);
1265         return 0;
1266 }
1267
1268 EXPORT_SYMBOL(set_binfmt);
1269
1270 /* format_corename will inspect the pattern parameter, and output a
1271  * name into corename, which must have space for at least
1272  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1273  */
1274 static int format_corename(char *corename, const char *pattern, long signr)
1275 {
1276         const char *pat_ptr = pattern;
1277         char *out_ptr = corename;
1278         char *const out_end = corename + CORENAME_MAX_SIZE;
1279         int rc;
1280         int pid_in_pattern = 0;
1281         int ispipe = 0;
1282
1283         if (*pattern == '|')
1284                 ispipe = 1;
1285
1286         /* Repeat as long as we have more pattern to process and more output
1287            space */
1288         while (*pat_ptr) {
1289                 if (*pat_ptr != '%') {
1290                         if (out_ptr == out_end)
1291                                 goto out;
1292                         *out_ptr++ = *pat_ptr++;
1293                 } else {
1294                         switch (*++pat_ptr) {
1295                         case 0:
1296                                 goto out;
1297                         /* Double percent, output one percent */
1298                         case '%':
1299                                 if (out_ptr == out_end)
1300                                         goto out;
1301                                 *out_ptr++ = '%';
1302                                 break;
1303                         /* pid */
1304                         case 'p':
1305                                 pid_in_pattern = 1;
1306                                 rc = snprintf(out_ptr, out_end - out_ptr,
1307                                               "%d", current->tgid);
1308                                 if (rc > out_end - out_ptr)
1309                                         goto out;
1310                                 out_ptr += rc;
1311                                 break;
1312                         /* uid */
1313                         case 'u':
1314                                 rc = snprintf(out_ptr, out_end - out_ptr,
1315                                               "%d", current->uid);
1316                                 if (rc > out_end - out_ptr)
1317                                         goto out;
1318                                 out_ptr += rc;
1319                                 break;
1320                         /* gid */
1321                         case 'g':
1322                                 rc = snprintf(out_ptr, out_end - out_ptr,
1323                                               "%d", current->gid);
1324                                 if (rc > out_end - out_ptr)
1325                                         goto out;
1326                                 out_ptr += rc;
1327                                 break;
1328                         /* signal that caused the coredump */
1329                         case 's':
1330                                 rc = snprintf(out_ptr, out_end - out_ptr,
1331                                               "%ld", signr);
1332                                 if (rc > out_end - out_ptr)
1333                                         goto out;
1334                                 out_ptr += rc;
1335                                 break;
1336                         /* UNIX time of coredump */
1337                         case 't': {
1338                                 struct timeval tv;
1339                                 do_gettimeofday(&tv);
1340                                 rc = snprintf(out_ptr, out_end - out_ptr,
1341                                               "%lu", tv.tv_sec);
1342                                 if (rc > out_end - out_ptr)
1343                                         goto out;
1344                                 out_ptr += rc;
1345                                 break;
1346                         }
1347                         /* hostname */
1348                         case 'h':
1349                                 down_read(&uts_sem);
1350                                 rc = snprintf(out_ptr, out_end - out_ptr,
1351                                               "%s", utsname()->nodename);
1352                                 up_read(&uts_sem);
1353                                 if (rc > out_end - out_ptr)
1354                                         goto out;
1355                                 out_ptr += rc;
1356                                 break;
1357                         /* executable */
1358                         case 'e':
1359                                 rc = snprintf(out_ptr, out_end - out_ptr,
1360                                               "%s", current->comm);
1361                                 if (rc > out_end - out_ptr)
1362                                         goto out;
1363                                 out_ptr += rc;
1364                                 break;
1365                         default:
1366                                 break;
1367                         }
1368                         ++pat_ptr;
1369                 }
1370         }
1371         /* Backward compatibility with core_uses_pid:
1372          *
1373          * If core_pattern does not include a %p (as is the default)
1374          * and core_uses_pid is set, then .%pid will be appended to
1375          * the filename. Do not do this for piped commands. */
1376         if (!ispipe && !pid_in_pattern
1377             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1378                 rc = snprintf(out_ptr, out_end - out_ptr,
1379                               ".%d", current->tgid);
1380                 if (rc > out_end - out_ptr)
1381                         goto out;
1382                 out_ptr += rc;
1383         }
1384 out:
1385         *out_ptr = 0;
1386         return ispipe;
1387 }
1388
1389 static void zap_process(struct task_struct *start)
1390 {
1391         struct task_struct *t;
1392
1393         start->signal->flags = SIGNAL_GROUP_EXIT;
1394         start->signal->group_stop_count = 0;
1395
1396         t = start;
1397         do {
1398                 if (t != current && t->mm) {
1399                         t->mm->core_waiters++;
1400                         sigaddset(&t->pending.signal, SIGKILL);
1401                         signal_wake_up(t, 1);
1402                 }
1403         } while ((t = next_thread(t)) != start);
1404 }
1405
1406 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1407                                 int exit_code)
1408 {
1409         struct task_struct *g, *p;
1410         unsigned long flags;
1411         int err = -EAGAIN;
1412
1413         spin_lock_irq(&tsk->sighand->siglock);
1414         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1415                 tsk->signal->group_exit_code = exit_code;
1416                 zap_process(tsk);
1417                 err = 0;
1418         }
1419         spin_unlock_irq(&tsk->sighand->siglock);
1420         if (err)
1421                 return err;
1422
1423         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1424                 goto done;
1425
1426         rcu_read_lock();
1427         for_each_process(g) {
1428                 if (g == tsk->group_leader)
1429                         continue;
1430
1431                 p = g;
1432                 do {
1433                         if (p->mm) {
1434                                 if (p->mm == mm) {
1435                                         /*
1436                                          * p->sighand can't disappear, but
1437                                          * may be changed by de_thread()
1438                                          */
1439                                         lock_task_sighand(p, &flags);
1440                                         zap_process(p);
1441                                         unlock_task_sighand(p, &flags);
1442                                 }
1443                                 break;
1444                         }
1445                 } while ((p = next_thread(p)) != g);
1446         }
1447         rcu_read_unlock();
1448 done:
1449         return mm->core_waiters;
1450 }
1451
1452 static int coredump_wait(int exit_code)
1453 {
1454         struct task_struct *tsk = current;
1455         struct mm_struct *mm = tsk->mm;
1456         struct completion startup_done;
1457         struct completion *vfork_done;
1458         int core_waiters;
1459
1460         init_completion(&mm->core_done);
1461         init_completion(&startup_done);
1462         mm->core_startup_done = &startup_done;
1463
1464         core_waiters = zap_threads(tsk, mm, exit_code);
1465         up_write(&mm->mmap_sem);
1466
1467         if (unlikely(core_waiters < 0))
1468                 goto fail;
1469
1470         /*
1471          * Make sure nobody is waiting for us to release the VM,
1472          * otherwise we can deadlock when we wait on each other
1473          */
1474         vfork_done = tsk->vfork_done;
1475         if (vfork_done) {
1476                 tsk->vfork_done = NULL;
1477                 complete(vfork_done);
1478         }
1479
1480         if (core_waiters)
1481                 wait_for_completion(&startup_done);
1482 fail:
1483         BUG_ON(mm->core_waiters);
1484         return core_waiters;
1485 }
1486
1487 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1488 {
1489         char corename[CORENAME_MAX_SIZE + 1];
1490         struct mm_struct *mm = current->mm;
1491         struct linux_binfmt * binfmt;
1492         struct inode * inode;
1493         struct file * file;
1494         int retval = 0;
1495         int fsuid = current->fsuid;
1496         int flag = 0;
1497         int ispipe = 0;
1498
1499         audit_core_dumps(signr);
1500
1501         binfmt = current->binfmt;
1502         if (!binfmt || !binfmt->core_dump)
1503                 goto fail;
1504         down_write(&mm->mmap_sem);
1505         if (!mm->dumpable) {
1506                 up_write(&mm->mmap_sem);
1507                 goto fail;
1508         }
1509
1510         /*
1511          *      We cannot trust fsuid as being the "true" uid of the
1512          *      process nor do we know its entire history. We only know it
1513          *      was tainted so we dump it as root in mode 2.
1514          */
1515         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1516                 flag = O_EXCL;          /* Stop rewrite attacks */
1517                 current->fsuid = 0;     /* Dump root private */
1518         }
1519         mm->dumpable = 0;
1520
1521         retval = coredump_wait(exit_code);
1522         if (retval < 0)
1523                 goto fail;
1524
1525         /*
1526          * Clear any false indication of pending signals that might
1527          * be seen by the filesystem code called to write the core file.
1528          */
1529         clear_thread_flag(TIF_SIGPENDING);
1530
1531         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1532                 goto fail_unlock;
1533
1534         /*
1535          * lock_kernel() because format_corename() is controlled by sysctl, which
1536          * uses lock_kernel()
1537          */
1538         lock_kernel();
1539         ispipe = format_corename(corename, core_pattern, signr);
1540         unlock_kernel();
1541         if (ispipe) {
1542                 /* SIGPIPE can happen, but it's just never processed */
1543                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1544                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1545                                corename);
1546                         goto fail_unlock;
1547                 }
1548         } else
1549                 file = filp_open(corename,
1550                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1551                                  0600);
1552         if (IS_ERR(file))
1553                 goto fail_unlock;
1554         inode = file->f_path.dentry->d_inode;
1555         if (inode->i_nlink > 1)
1556                 goto close_fail;        /* multiple links - don't dump */
1557         if (!ispipe && d_unhashed(file->f_path.dentry))
1558                 goto close_fail;
1559
1560         /* AK: actually i see no reason to not allow this for named pipes etc.,
1561            but keep the previous behaviour for now. */
1562         if (!ispipe && !S_ISREG(inode->i_mode))
1563                 goto close_fail;
1564         /*
1565          * Dont allow local users get cute and trick others to coredump
1566          * into their pre-created files:
1567          */
1568         if (inode->i_uid != current->fsuid)
1569                 goto close_fail;
1570         if (!file->f_op)
1571                 goto close_fail;
1572         if (!file->f_op->write)
1573                 goto close_fail;
1574         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1575                 goto close_fail;
1576
1577         retval = binfmt->core_dump(signr, regs, file);
1578
1579         if (retval)
1580                 current->signal->group_exit_code |= 0x80;
1581 close_fail:
1582         filp_close(file, NULL);
1583 fail_unlock:
1584         current->fsuid = fsuid;
1585         complete_all(&mm->core_done);
1586 fail:
1587         return retval;
1588 }