fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/config.h>
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/mman.h>
  29 #include <linux/a.out.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/key.h>
  38 #include <linux/personality.h>
  39 #include <linux/binfmts.h>
  40 #include <linux/swap.h>
  41 #include <linux/utsname.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/ckrm.h>
  51 #include <linux/vs_memory.h>
  52 #include <linux/ckrm_mem.h>
  53
  54 #include <asm/uaccess.h>
  55 #include <asm/mmu_context.h>
  56
  57 #ifdef CONFIG_KMOD
  58 #include <linux/kmod.h>
  59 #endif
  60
  61 int core_uses_pid;
  62 char core_pattern[65] = "core";
  63 int suid_dumpable = 0;
  64
  65 EXPORT_SYMBOL(suid_dumpable);
  66 /* The maximal length of core_pattern is also specified in sysctl.c */
  67
  68 static struct linux_binfmt *formats;
  69 static rwlock_t binfmt_lock = RW_LOCK_UNLOCKED;
  70
  71 int register_binfmt(struct linux_binfmt * fmt)
  72 {
  73         struct linux_binfmt ** tmp = &formats;
  74
  75         if (!fmt)
  76                 return -EINVAL;
  77         if (fmt->next)
  78                 return -EBUSY;
  79         write_lock(&binfmt_lock);
  80         while (*tmp) {
  81                 if (fmt == *tmp) {
  82                         write_unlock(&binfmt_lock);
  83                         return -EBUSY;
  84                 }
  85                 tmp = &(*tmp)->next;
  86         }
  87         fmt->next = formats;
  88         formats = fmt;
  89         write_unlock(&binfmt_lock);
  90         return 0;
  91 }
  92
  93 EXPORT_SYMBOL(register_binfmt);
  94
  95 int unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         struct linux_binfmt ** tmp = &formats;
  98
  99         write_lock(&binfmt_lock);
 100         while (*tmp) {
 101                 if (fmt == *tmp) {
 102                         *tmp = fmt->next;
 103                         write_unlock(&binfmt_lock);
 104                         return 0;
 105                 }
 106                 tmp = &(*tmp)->next;
 107         }
 108         write_unlock(&binfmt_lock);
 109         return -EINVAL;
 110 }
 111
 112 EXPORT_SYMBOL(unregister_binfmt);
 113
 114 static inline void put_binfmt(struct linux_binfmt * fmt)
 115 {
 116         module_put(fmt->module);
 117 }
 118
 119 /*
 120  * Note that a shared library must be both readable and executable due to
 121  * security reasons.
 122  *
 123  * Also note that we take the address to load from from the file itself.
 124  */
 125 asmlinkage long sys_uselib(const char __user * library)
 126 {
 127         struct file * file;
 128         struct nameidata nd;
 129         int error;
 130
 131         nd.intent.open.flags = FMODE_READ;
 132         error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 133         if (error)
 134                 goto out;
 135
 136         error = -EINVAL;
 137         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 138                 goto exit;
 139
 140         error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
 141         if (error)
 142                 goto exit;
 143
 144         file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 145         error = PTR_ERR(file);
 146         if (IS_ERR(file))
 147                 goto out;
 148
 149         error = -ENOEXEC;
 150         if(file->f_op) {
 151                 struct linux_binfmt * fmt;
 152
 153                 read_lock(&binfmt_lock);
 154                 for (fmt = formats ; fmt ; fmt = fmt->next) {
 155                         if (!fmt->load_shlib)
 156                                 continue;
 157                         if (!try_module_get(fmt->module))
 158                                 continue;
 159                         read_unlock(&binfmt_lock);
 160                         error = fmt->load_shlib(file);
 161                         read_lock(&binfmt_lock);
 162                         put_binfmt(fmt);
 163                         if (error != -ENOEXEC)
 164                                 break;
 165                 }
 166                 read_unlock(&binfmt_lock);
 167         }
 168         fput(file);
 169 out:
 170         return error;
 171 exit:
 172         path_release(&nd);
 173         goto out;
 174 }
 175
 176 /*
 177  * count() counts the number of strings in array ARGV.
 178  */
 179 static int count(char __user * __user * argv, int max)
 180 {
 181         int i = 0;
 182
 183         if (argv != NULL) {
 184                 for (;;) {
 185                         char __user * p;
 186
 187                         if (get_user(p, argv))
 188                                 return -EFAULT;
 189                         if (!p)
 190                                 break;
 191                         argv++;
 192                         if(++i > max)
 193                                 return -E2BIG;
 194                 }
 195         }
 196         return i;
 197 }
 198
 199 /*
 200  * 'copy_strings()' copies argument/environment strings from user
 201  * memory to free pages in kernel mem. These are in a format ready
 202  * to be put directly into the top of new user memory.
 203  */
 204 int copy_strings(int argc,char __user * __user * argv, struct linux_binprm *bprm)
 205 {
 206         struct page *kmapped_page = NULL;
 207         char *kaddr = NULL;
 208         int ret;
 209
 210         while (argc-- > 0) {
 211                 char __user *str;
 212                 int len;
 213                 unsigned long pos;
 214
 215                 if (get_user(str, argv+argc) ||
 216                                 !(len = strnlen_user(str, bprm->p))) {
 217                         ret = -EFAULT;
 218                         goto out;
 219                 }
 220
 221                 if (bprm->p < len)  {
 222                         ret = -E2BIG;
 223                         goto out;
 224                 }
 225
 226                 bprm->p -= len;
 227                 /* XXX: add architecture specific overflow check here. */
 228                 pos = bprm->p;
 229
 230                 while (len > 0) {
 231                         int i, new, err;
 232                         int offset, bytes_to_copy;
 233                         struct page *page;
 234
 235                         offset = pos % PAGE_SIZE;
 236                         i = pos/PAGE_SIZE;
 237                         page = bprm->page[i];
 238                         new = 0;
 239                         if (!page) {
 240                                 page = alloc_page(GFP_HIGHUSER);
 241                                 bprm->page[i] = page;
 242                                 if (!page) {
 243                                         ret = -ENOMEM;
 244                                         goto out;
 245                                 }
 246                                 new = 1;
 247                         }
 248
 249                         if (page != kmapped_page) {
 250                                 if (kmapped_page)
 251                                         kunmap(kmapped_page);
 252                                 kmapped_page = page;
 253                                 kaddr = kmap(kmapped_page);
 254                         }
 255                         if (new && offset)
 256                                 memset(kaddr, 0, offset);
 257                         bytes_to_copy = PAGE_SIZE - offset;
 258                         if (bytes_to_copy > len) {
 259                                 bytes_to_copy = len;
 260                                 if (new)
 261                                         memset(kaddr+offset+len, 0,
 262                                                 PAGE_SIZE-offset-len);
 263                         }
 264                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 265                         if (err) {
 266                                 ret = -EFAULT;
 267                                 goto out;
 268                         }
 269
 270                         pos += bytes_to_copy;
 271                         str += bytes_to_copy;
 272                         len -= bytes_to_copy;
 273                 }
 274         }
 275         ret = 0;
 276 out:
 277         if (kmapped_page)
 278                 kunmap(kmapped_page);
 279         return ret;
 280 }
 281
 282 /*
 283  * Like copy_strings, but get argv and its values from kernel memory.
 284  */
 285 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 286 {
 287         int r;
 288         mm_segment_t oldfs = get_fs();
 289         set_fs(KERNEL_DS);
 290         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 291         set_fs(oldfs);
 292         return r;
 293 }
 294
 295 EXPORT_SYMBOL(copy_strings_kernel);
 296
 297 #ifdef CONFIG_MMU
 298 /*
 299  * This routine is used to map in a page into an address space: needed by
 300  * execve() for the initial stack and environment pages.
 301  *
 302  * vma->vm_mm->mmap_sem is held for writing.
 303  */
 304 void install_arg_page(struct vm_area_struct *vma,
 305                         struct page *page, unsigned long address)
 306 {
 307         struct mm_struct *mm = vma->vm_mm;
 308         pgd_t * pgd;
 309         pmd_t * pmd;
 310         pte_t * pte;
 311
 312         if (unlikely(anon_vma_prepare(vma)))
 313                 goto out_sig;
 314
 315         flush_dcache_page(page);
 316         pgd = pgd_offset(mm, address);
 317
 318         spin_lock(&mm->page_table_lock);
 319         pmd = pmd_alloc(mm, pgd, address);
 320         if (!pmd)
 321                 goto out;
 322         pte = pte_alloc_map(mm, pmd, address);
 323         if (!pte)
 324                 goto out;
 325         if (!pte_none(*pte)) {
 326                 pte_unmap(pte);
 327                 goto out;
 328         }
 329         // mm->rss++;
 330         vx_rsspages_inc(mm);
 331         lru_cache_add_active(page);
 332         set_pte(pte, pte_mkdirty(pte_mkwrite(mk_pte(
 333                                         page, vma->vm_page_prot))));
 334         page_add_anon_rmap(page, vma, address);
 335         pte_unmap(pte);
 336         spin_unlock(&mm->page_table_lock);
 337
 338         /* no need for flush_tlb */
 339         return;
 340 out:
 341         spin_unlock(&mm->page_table_lock);
 342 out_sig:
 343         __free_page(page);
 344         force_sig(SIGKILL, current);
 345 }
 346
 347 #define EXTRA_STACK_VM_PAGES    20      /* random */
 348
 349 int setup_arg_pages(struct linux_binprm *bprm, int executable_stack)
 350 {
 351         unsigned long stack_base;
 352         struct vm_area_struct *mpnt;
 353         struct mm_struct *mm = current->mm;
 354         int i, ret;
 355         long arg_size;
 356
 357 #ifdef CONFIG_STACK_GROWSUP
 358         /* Move the argument and environment strings to the bottom of the
 359          * stack space.
 360          */
 361         int offset, j;
 362         char *to, *from;
 363
 364         /* Start by shifting all the pages down */
 365         i = 0;
 366         for (j = 0; j < MAX_ARG_PAGES; j++) {
 367                 struct page *page = bprm->page[j];
 368                 if (!page)
 369                         continue;
 370                 bprm->page[i++] = page;
 371         }
 372
 373         /* Now move them within their pages */
 374         offset = bprm->p % PAGE_SIZE;
 375         to = kmap(bprm->page[0]);
 376         for (j = 1; j < i; j++) {
 377                 memmove(to, to + offset, PAGE_SIZE - offset);
 378                 from = kmap(bprm->page[j]);
 379                 memcpy(to + PAGE_SIZE - offset, from, offset);
 380                 kunmap(bprm->page[j - 1]);
 381                 to = from;
 382         }
 383         memmove(to, to + offset, PAGE_SIZE - offset);
 384         kunmap(bprm->page[j - 1]);
 385
 386         /* Limit stack size to 1GB */
 387         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 388         if (stack_base > (1 << 30))
 389                 stack_base = 1 << 30;
 390         stack_base = PAGE_ALIGN(STACK_TOP - stack_base);
 391
 392         /* Adjust bprm->p to point to the end of the strings. */
 393         bprm->p = stack_base + PAGE_SIZE * i - offset;
 394
 395         mm->arg_start = stack_base;
 396         arg_size = i << PAGE_SHIFT;
 397
 398         /* zero pages that were copied above */
 399         while (i < MAX_ARG_PAGES)
 400                 bprm->page[i++] = NULL;
 401 #else
 402 #ifdef __HAVE_ARCH_ALIGN_STACK
 403         stack_base = arch_align_stack(STACK_TOP - MAX_ARG_PAGES*PAGE_SIZE);
 404         stack_base = PAGE_ALIGN(stack_base);
 405 #else
 406         stack_base = STACK_TOP - MAX_ARG_PAGES * PAGE_SIZE;
 407 #endif
 408         bprm->p += stack_base;
 409         mm->arg_start = bprm->p;
 410         arg_size = STACK_TOP - (PAGE_MASK & (unsigned long) mm->arg_start);
 411 #endif
 412
 413         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 414
 415         if (bprm->loader)
 416                 bprm->loader += stack_base;
 417         bprm->exec += stack_base;
 418
 419         mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
 420         if (!mpnt)
 421                 return -ENOMEM;
 422
 423         if (security_vm_enough_memory(arg_size >> PAGE_SHIFT) ||
 424                 !vx_vmpages_avail(mm, arg_size >> PAGE_SHIFT)) {
 425                 kmem_cache_free(vm_area_cachep, mpnt);
 426                 return -ENOMEM;
 427         }
 428
 429         memset(mpnt, 0, sizeof(*mpnt));
 430
 431         down_write(&mm->mmap_sem);
 432         {
 433                 mpnt->vm_mm = mm;
 434 #ifdef CONFIG_STACK_GROWSUP
 435                 mpnt->vm_start = stack_base;
 436                 mpnt->vm_end = stack_base + arg_size;
 437 #else
 438                 mpnt->vm_end = STACK_TOP;
 439                 mpnt->vm_start = mpnt->vm_end - arg_size;
 440 #endif
 441                 /* Adjust stack execute permissions; explicitly enable
 442                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 443                  * and leave alone (arch default) otherwise. */
 444                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 445                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 446                 else if (executable_stack == EXSTACK_DISABLE_X)
 447                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 448                 else
 449                         mpnt->vm_flags = VM_STACK_FLAGS;
 450                 mpnt->vm_flags |= mm->def_flags;
 451                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 452                 if ((ret = insert_vm_struct(mm, mpnt))) {
 453                         up_write(&mm->mmap_sem);
 454                         kmem_cache_free(vm_area_cachep, mpnt);
 455                         return ret;
 456                 }
 457                 // mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 458                 vx_vmpages_sub(mm, mm->total_vm - vma_pages(mpnt));
 459                 mm->stack_vm = mm->total_vm;
 460         }
 461
 462         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 463                 struct page *page = bprm->page[i];
 464                 if (page) {
 465                         bprm->page[i] = NULL;
 466                         install_arg_page(mpnt, page, stack_base);
 467                 }
 468                 stack_base += PAGE_SIZE;
 469         }
 470         up_write(&mm->mmap_sem);
 471
 472         return 0;
 473 }
 474
 475 EXPORT_SYMBOL(setup_arg_pages);
 476
 477 #define free_arg_pages(bprm) do { } while (0)
 478
 479 #else
 480
 481 static inline void free_arg_pages(struct linux_binprm *bprm)
 482 {
 483         int i;
 484
 485         for (i = 0; i < MAX_ARG_PAGES; i++) {
 486                 if (bprm->page[i])
 487                         __free_page(bprm->page[i]);
 488                 bprm->page[i] = NULL;
 489         }
 490 }
 491
 492 #endif /* CONFIG_MMU */
 493
 494 struct file *open_exec(const char *name)
 495 {
 496         struct nameidata nd;
 497         int err;
 498         struct file *file;
 499
 500         nd.intent.open.flags = FMODE_READ;
 501         err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
 502         file = ERR_PTR(err);
 503
 504         if (!err) {
 505                 struct inode *inode = nd.dentry->d_inode;
 506                 file = ERR_PTR(-EACCES);
 507                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 508                     S_ISREG(inode->i_mode)) {
 509                         int err = permission(inode, MAY_EXEC, &nd);
 510                         if (!err && !(inode->i_mode & 0111))
 511                                 err = -EACCES;
 512                         file = ERR_PTR(err);
 513                         if (!err) {
 514                                 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
 515                                 if (!IS_ERR(file)) {
 516                                         err = deny_write_access(file);
 517                                         if (err) {
 518                                                 fput(file);
 519                                                 file = ERR_PTR(err);
 520                                         }
 521                                 }
 522 out:
 523                                 return file;
 524                         }
 525                 }
 526                 path_release(&nd);
 527         }
 528         goto out;
 529 }
 530
 531 EXPORT_SYMBOL(open_exec);
 532
 533 int kernel_read(struct file *file, unsigned long offset,
 534         char *addr, unsigned long count)
 535 {
 536         mm_segment_t old_fs;
 537         loff_t pos = offset;
 538         int result;
 539
 540         old_fs = get_fs();
 541         set_fs(get_ds());
 542         /* The cast to a user pointer is valid due to the set_fs() */
 543         result = vfs_read(file, (void __user *)addr, count, &pos);
 544         set_fs(old_fs);
 545         return result;
 546 }
 547
 548 EXPORT_SYMBOL(kernel_read);
 549
 550 static int exec_mmap(struct mm_struct *mm)
 551 {
 552         struct task_struct *tsk;
 553         struct mm_struct * old_mm, *active_mm;
 554
 555         /* Notify parent that we're no longer interested in the old VM */
 556         tsk = current;
 557         old_mm = current->mm;
 558         mm_release(tsk, old_mm);
 559
 560         task_lock(tsk);
 561         active_mm = tsk->active_mm;
 562         tsk->mm = mm;
 563         tsk->active_mm = mm;
 564         activate_mm(active_mm, mm);
 565         task_unlock(tsk);
 566         arch_pick_mmap_layout(mm);
 567 #ifdef CONFIG_CKRM_RES_MEM
 568         if (old_mm) {
 569                 spin_lock(&old_mm->peertask_lock);
 570                 list_del(&tsk->mm_peers);
 571                 ckrm_mem_evaluate_mm(old_mm);
 572                 spin_unlock(&old_mm->peertask_lock);
 573         }
 574         spin_lock(&mm->peertask_lock);
 575         list_add_tail(&tsk->mm_peers, &mm->tasklist);
 576         ckrm_mem_evaluate_mm(mm);
 577         spin_unlock(&mm->peertask_lock);
 578 #endif
 579         if (old_mm) {
 580                 if (active_mm != old_mm) BUG();
 581                 mmput(old_mm);
 582                 return 0;
 583         }
 584         mmdrop(active_mm);
 585         return 0;
 586 }
 587
 588 /*
 589  * This function makes sure the current process has its own signal table,
 590  * so that flush_signal_handlers can later reset the handlers without
 591  * disturbing other processes.  (Other processes might share the signal
 592  * table via the CLONE_SIGHAND option to clone().)
 593  */
 594 static inline int de_thread(struct task_struct *tsk)
 595 {
 596         struct signal_struct *sig = tsk->signal;
 597         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 598         spinlock_t *lock = &oldsighand->siglock;
 599         int count;
 600
 601         /*
 602          * If we don't share sighandlers, then we aren't sharing anything
 603          * and we can just re-use it all.
 604          */
 605         if (atomic_read(&oldsighand->count) <= 1) {
 606                 BUG_ON(atomic_read(&sig->count) != 1);
 607                 exit_itimers(sig);
 608                 return 0;
 609         }
 610
 611         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 612         if (!newsighand)
 613                 return -ENOMEM;
 614
 615         if (thread_group_empty(current))
 616                 goto no_thread_group;
 617
 618         /*
 619          * Kill all other threads in the thread group.
 620          * We must hold tasklist_lock to call zap_other_threads.
 621          */
 622         read_lock(&tasklist_lock);
 623         spin_lock_irq(lock);
 624         if (sig->group_exit) {
 625                 /*
 626                  * Another group action in progress, just
 627                  * return so that the signal is processed.
 628                  */
 629                 spin_unlock_irq(lock);
 630                 read_unlock(&tasklist_lock);
 631                 kmem_cache_free(sighand_cachep, newsighand);
 632                 return -EAGAIN;
 633         }
 634         sig->group_exit = 1;
 635         zap_other_threads(current);
 636         read_unlock(&tasklist_lock);
 637
 638         /*
 639          * Account for the thread group leader hanging around:
 640          */
 641         count = 2;
 642         if (current->pid == current->tgid)
 643                 count = 1;
 644         while (atomic_read(&sig->count) > count) {
 645                 sig->group_exit_task = current;
 646                 sig->notify_count = count;
 647                 __set_current_state(TASK_UNINTERRUPTIBLE);
 648                 spin_unlock_irq(lock);
 649                 schedule();
 650                 spin_lock_irq(lock);
 651         }
 652         sig->group_exit_task = NULL;
 653         sig->notify_count = 0;
 654         spin_unlock_irq(lock);
 655
 656         /*
 657          * At this point all other threads have exited, all we have to
 658          * do is to wait for the thread group leader to become inactive,
 659          * and to assume its PID:
 660          */
 661         if (current->pid != current->tgid) {
 662                 struct task_struct *leader = current->group_leader, *parent;
 663                 struct dentry *proc_dentry1, *proc_dentry2;
 664                 unsigned long exit_state, ptrace;
 665
 666                 /*
 667                  * Wait for the thread group leader to be a zombie.
 668                  * It should already be zombie at this point, most
 669                  * of the time.
 670                  */
 671                 while (leader->exit_state != EXIT_ZOMBIE)
 672                         yield();
 673
 674                 spin_lock(&leader->proc_lock);
 675                 spin_lock(&current->proc_lock);
 676                 proc_dentry1 = proc_pid_unhash(current);
 677                 proc_dentry2 = proc_pid_unhash(leader);
 678                 write_lock_irq(&tasklist_lock);
 679
 680                 if (leader->tgid != current->tgid)
 681                         BUG();
 682                 if (current->pid == current->tgid)
 683                         BUG();
 684                 /*
 685                  * An exec() starts a new thread group with the
 686                  * TGID of the previous thread group. Rehash the
 687                  * two threads with a switched PID, and release
 688                  * the former thread group leader:
 689                  */
 690                 ptrace = leader->ptrace;
 691                 parent = leader->parent;
 692
 693                 ptrace_unlink(current);
 694                 ptrace_unlink(leader);
 695                 remove_parent(current);
 696                 remove_parent(leader);
 697
 698                 switch_exec_pids(leader, current);
 699
 700                 current->parent = current->real_parent = leader->real_parent;
 701                 leader->parent = leader->real_parent = child_reaper;
 702                 current->group_leader = current;
 703                 leader->group_leader = leader;
 704
 705                 add_parent(current, current->parent);
 706                 add_parent(leader, leader->parent);
 707                 if (ptrace) {
 708                         current->ptrace = ptrace;
 709                         __ptrace_link(current, parent);
 710                 }
 711
 712                 list_del(&current->tasks);
 713                 list_add_tail(&current->tasks, &init_task.tasks);
 714                 current->exit_signal = SIGCHLD;
 715                 exit_state = leader->exit_state;
 716
 717                 write_unlock_irq(&tasklist_lock);
 718                 spin_unlock(&leader->proc_lock);
 719                 spin_unlock(&current->proc_lock);
 720                 proc_pid_flush(proc_dentry1);
 721                 proc_pid_flush(proc_dentry2);
 722
 723                 if (exit_state != EXIT_ZOMBIE)
 724                         BUG();
 725                 release_task(leader);
 726         }
 727
 728         /*
 729          * Now there are really no other threads at all,
 730          * so it's safe to stop telling them to kill themselves.
 731          */
 732         sig->group_exit = 0;
 733
 734 no_thread_group:
 735         BUG_ON(atomic_read(&sig->count) != 1);
 736         exit_itimers(sig);
 737
 738         if (atomic_read(&oldsighand->count) == 1) {
 739                 /*
 740                  * Now that we nuked the rest of the thread group,
 741                  * it turns out we are not sharing sighand any more either.
 742                  * So we can just keep it.
 743                  */
 744                 kmem_cache_free(sighand_cachep, newsighand);
 745         } else {
 746                 /*
 747                  * Move our state over to newsighand and switch it in.
 748                  */
 749                 spin_lock_init(&newsighand->siglock);
 750                 atomic_set(&newsighand->count, 1);
 751                 memcpy(newsighand->action, oldsighand->action,
 752                        sizeof(newsighand->action));
 753
 754                 write_lock_irq(&tasklist_lock);
 755                 spin_lock(&oldsighand->siglock);
 756                 spin_lock(&newsighand->siglock);
 757
 758                 current->sighand = newsighand;
 759                 recalc_sigpending();
 760
 761                 spin_unlock(&newsighand->siglock);
 762                 spin_unlock(&oldsighand->siglock);
 763                 write_unlock_irq(&tasklist_lock);
 764
 765                 if (atomic_dec_and_test(&oldsighand->count))
 766                         kmem_cache_free(sighand_cachep, oldsighand);
 767         }
 768
 769         if (!thread_group_empty(current))
 770                 BUG();
 771         if (current->tgid != current->pid)
 772                 BUG();
 773         return 0;
 774 }
 775
 776 /*
 777  * These functions flushes out all traces of the currently running executable
 778  * so that a new one can be started
 779  */
 780
 781 static inline void flush_old_files(struct files_struct * files)
 782 {
 783         long j = -1;
 784
 785         spin_lock(&files->file_lock);
 786         for (;;) {
 787                 unsigned long set, i;
 788
 789                 j++;
 790                 i = j * __NFDBITS;
 791                 if (i >= files->max_fds || i >= files->max_fdset)
 792                         break;
 793                 set = files->close_on_exec->fds_bits[j];
 794                 if (!set)
 795                         continue;
 796                 files->close_on_exec->fds_bits[j] = 0;
 797                 spin_unlock(&files->file_lock);
 798                 for ( ; set ; i++,set >>= 1) {
 799                         if (set & 1) {
 800                                 sys_close(i);
 801                         }
 802                 }
 803                 spin_lock(&files->file_lock);
 804
 805         }
 806         spin_unlock(&files->file_lock);
 807 }
 808
 809 void get_task_comm(char *buf, struct task_struct *tsk)
 810 {
 811         /* buf must be at least sizeof(tsk->comm) in size */
 812         task_lock(tsk);
 813         memcpy(buf, tsk->comm, sizeof(tsk->comm));
 814         task_unlock(tsk);
 815 }
 816
 817 void set_task_comm(struct task_struct *tsk, char *buf)
 818 {
 819         task_lock(tsk);
 820         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 821         task_unlock(tsk);
 822 }
 823
 824 int flush_old_exec(struct linux_binprm * bprm)
 825 {
 826         char * name;
 827         int i, ch, retval;
 828         struct files_struct *files;
 829         char tcomm[sizeof(current->comm)];
 830
 831         /*
 832          * Make sure we have a private signal table and that
 833          * we are unassociated from the previous thread group.
 834          */
 835         retval = de_thread(current);
 836         if (retval)
 837                 goto out;
 838
 839         /*
 840          * Make sure we have private file handles. Ask the
 841          * fork helper to do the work for us and the exit
 842          * helper to do the cleanup of the old one.
 843          */
 844         files = current->files;         /* refcounted so safe to hold */
 845         retval = unshare_files();
 846         if (retval)
 847                 goto out;
 848         /*
 849          * Release all of the old mmap stuff
 850          */
 851         retval = exec_mmap(bprm->mm);
 852         if (retval)
 853                 goto mmap_failed;
 854
 855         bprm->mm = NULL;                /* We're using it now */
 856
 857         /* This is the point of no return */
 858         steal_locks(files);
 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         for (i=0; (ch = *(name++)) != '\0';) {
 870                 if (ch == '/')
 871                         i = 0;
 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_RELOCEXEC;
 880         flush_thread();
 881
 882         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 883             permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
 884             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 885                 suid_keys(current);
 886                 current->mm->dumpable = suid_dumpable;
 887         }
 888
 889         /* An exec changes our domain. We are no longer part of the thread
 890            group */
 891
 892         current->self_exec_id++;
 893
 894         flush_signal_handlers(current, 0);
 895         flush_old_files(current->files);
 896
 897         return 0;
 898
 899 mmap_failed:
 900         put_files_struct(current->files);
 901         current->files = files;
 902 out:
 903         return retval;
 904 }
 905
 906 EXPORT_SYMBOL(flush_old_exec);
 907
 908 /*
 909  * Fill the binprm structure from the inode.
 910  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 911  */
 912 int prepare_binprm(struct linux_binprm *bprm)
 913 {
 914         int mode;
 915         struct inode * inode = bprm->file->f_dentry->d_inode;
 916         int retval;
 917
 918         mode = inode->i_mode;
 919         /*
 920          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 921          * generic_permission lets a non-executable through
 922          */
 923         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 924                 return -EACCES;
 925         if (bprm->file->f_op == NULL)
 926                 return -EACCES;
 927
 928         bprm->e_uid = current->euid;
 929         bprm->e_gid = current->egid;
 930
 931         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 932                 /* Set-uid? */
 933                 if (mode & S_ISUID) {
 934                         current->personality &= ~PER_CLEAR_ON_SETID;
 935                         bprm->e_uid = inode->i_uid;
 936                 }
 937
 938                 /* Set-gid? */
 939                 /*
 940                  * If setgid is set but no group execute bit then this
 941                  * is a candidate for mandatory locking, not a setgid
 942                  * executable.
 943                  */
 944                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 945                         current->personality &= ~PER_CLEAR_ON_SETID;
 946                         bprm->e_gid = inode->i_gid;
 947                 }
 948         }
 949
 950         /* fill in binprm security blob */
 951         retval = security_bprm_set(bprm);
 952         if (retval)
 953                 return retval;
 954
 955         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 956         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 957 }
 958
 959 EXPORT_SYMBOL(prepare_binprm);
 960
 961 static inline int unsafe_exec(struct task_struct *p)
 962 {
 963         int unsafe = 0;
 964         if (p->ptrace & PT_PTRACED) {
 965                 if (p->ptrace & PT_PTRACE_CAP)
 966                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 967                 else
 968                         unsafe |= LSM_UNSAFE_PTRACE;
 969         }
 970         if (atomic_read(&p->fs->count) > 1 ||
 971             atomic_read(&p->files->count) > 1 ||
 972             atomic_read(&p->sighand->count) > 1)
 973                 unsafe |= LSM_UNSAFE_SHARE;
 974
 975         return unsafe;
 976 }
 977
 978 void compute_creds(struct linux_binprm *bprm)
 979 {
 980         int unsafe;
 981
 982         if (bprm->e_uid != current->uid)
 983                 suid_keys(current);
 984         exec_keys(current);
 985
 986         task_lock(current);
 987         unsafe = unsafe_exec(current);
 988         security_bprm_apply_creds(bprm, unsafe);
 989         task_unlock(current);
 990 }
 991
 992 EXPORT_SYMBOL(compute_creds);
 993
 994 void remove_arg_zero(struct linux_binprm *bprm)
 995 {
 996         if (bprm->argc) {
 997                 unsigned long offset;
 998                 char * kaddr;
 999                 struct page *page;
1000
1001                 offset = bprm->p % PAGE_SIZE;
1002                 goto inside;
1003
1004                 while (bprm->p++, *(kaddr+offset++)) {
1005                         if (offset != PAGE_SIZE)
1006                                 continue;
1007                         offset = 0;
1008                         kunmap_atomic(kaddr, KM_USER0);
1009 inside:
1010                         page = bprm->page[bprm->p/PAGE_SIZE];
1011                         kaddr = kmap_atomic(page, KM_USER0);
1012                 }
1013                 kunmap_atomic(kaddr, KM_USER0);
1014                 bprm->argc--;
1015         }
1016 }
1017
1018 EXPORT_SYMBOL(remove_arg_zero);
1019
1020 /*
1021  * cycle the list of binary formats handler, until one recognizes the image
1022  */
1023 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1024 {
1025         int try,retval;
1026         struct linux_binfmt *fmt;
1027 #ifdef __alpha__
1028         /* handle /sbin/loader.. */
1029         {
1030             struct exec * eh = (struct exec *) bprm->buf;
1031
1032             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1033                 (eh->fh.f_flags & 0x3000) == 0x3000)
1034             {
1035                 struct file * file;
1036                 unsigned long loader;
1037
1038                 allow_write_access(bprm->file);
1039                 fput(bprm->file);
1040                 bprm->file = NULL;
1041
1042                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1043
1044                 file = open_exec("/sbin/loader");
1045                 retval = PTR_ERR(file);
1046                 if (IS_ERR(file))
1047                         return retval;
1048
1049                 /* Remember if the application is TASO.  */
1050                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1051
1052                 bprm->file = file;
1053                 bprm->loader = loader;
1054                 retval = prepare_binprm(bprm);
1055                 if (retval<0)
1056                         return retval;
1057                 /* should call search_binary_handler recursively here,
1058                    but it does not matter */
1059             }
1060         }
1061 #endif
1062         retval = security_bprm_check(bprm);
1063         if (retval)
1064                 return retval;
1065
1066         /* kernel module loader fixup */
1067         /* so we don't try to load run modprobe in kernel space. */
1068         set_fs(USER_DS);
1069         retval = -ENOENT;
1070         for (try=0; try<2; try++) {
1071                 read_lock(&binfmt_lock);
1072                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1073                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1074                         if (!fn)
1075                                 continue;
1076                         if (!try_module_get(fmt->module))
1077                                 continue;
1078                         read_unlock(&binfmt_lock);
1079                         retval = fn(bprm, regs);
1080                         if (retval >= 0) {
1081                                 put_binfmt(fmt);
1082                                 allow_write_access(bprm->file);
1083                                 if (bprm->file)
1084                                         fput(bprm->file);
1085                                 bprm->file = NULL;
1086                                 current->did_exec = 1;
1087                                 ckrm_cb_exec(bprm->filename);
1088                                 return retval;
1089                         }
1090                         read_lock(&binfmt_lock);
1091                         put_binfmt(fmt);
1092                         if (retval != -ENOEXEC || bprm->mm == NULL)
1093                                 break;
1094                         if (!bprm->file) {
1095                                 read_unlock(&binfmt_lock);
1096                                 return retval;
1097                         }
1098                 }
1099                 read_unlock(&binfmt_lock);
1100                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1101                         break;
1102 #ifdef CONFIG_KMOD
1103                 }else{
1104 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1105                         if (printable(bprm->buf[0]) &&
1106                             printable(bprm->buf[1]) &&
1107                             printable(bprm->buf[2]) &&
1108                             printable(bprm->buf[3]))
1109                                 break; /* -ENOEXEC */
1110                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1111 #endif
1112                 }
1113         }
1114         return retval;
1115 }
1116
1117 EXPORT_SYMBOL(search_binary_handler);
1118
1119 /*
1120  * sys_execve() executes a new program.
1121  */
1122 int do_execve(char * filename,
1123         char __user *__user *argv,
1124         char __user *__user *envp,
1125         struct pt_regs * regs)
1126 {
1127         struct linux_binprm *bprm;
1128         struct file *file;
1129         int retval;
1130         int i;
1131
1132         retval = -ENOMEM;
1133         bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1134         if (!bprm)
1135                 goto out_ret;
1136         memset(bprm, 0, sizeof(*bprm));
1137
1138         file = open_exec(filename);
1139         retval = PTR_ERR(file);
1140         if (IS_ERR(file))
1141                 goto out_kfree;
1142
1143         sched_exec();
1144
1145         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1146
1147         bprm->file = file;
1148         bprm->filename = filename;
1149         bprm->interp = filename;
1150         bprm->mm = mm_alloc();
1151         retval = -ENOMEM;
1152         if (!bprm->mm)
1153                 goto out_file;
1154
1155         retval = init_new_context(current, bprm->mm);
1156         if (retval < 0)
1157                 goto out_mm;
1158
1159         bprm->argc = count(argv, bprm->p / sizeof(void *));
1160         if ((retval = bprm->argc) < 0)
1161                 goto out_mm;
1162
1163         bprm->envc = count(envp, bprm->p / sizeof(void *));
1164         if ((retval = bprm->envc) < 0)
1165                 goto out_mm;
1166
1167         retval = security_bprm_alloc(bprm);
1168         if (retval)
1169                 goto out;
1170
1171         retval = prepare_binprm(bprm);
1172         if (retval < 0)
1173                 goto out;
1174
1175         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1176         if (retval < 0)
1177                 goto out;
1178
1179         bprm->exec = bprm->p;
1180         retval = copy_strings(bprm->envc, envp, bprm);
1181         if (retval < 0)
1182                 goto out;
1183
1184         retval = copy_strings(bprm->argc, argv, bprm);
1185         if (retval < 0)
1186                 goto out;
1187
1188         retval = search_binary_handler(bprm,regs);
1189         if (retval >= 0) {
1190                 free_arg_pages(bprm);
1191
1192                 /* execve success */
1193                 security_bprm_free(bprm);
1194                 kfree(bprm);
1195                 return retval;
1196         }
1197
1198 out:
1199         /* Something went wrong, return the inode and free the argument pages*/
1200         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1201                 struct page * page = bprm->page[i];
1202                 if (page)
1203                         __free_page(page);
1204         }
1205
1206         if (bprm->security)
1207                 security_bprm_free(bprm);
1208
1209 out_mm:
1210         if (bprm->mm)
1211                 mmdrop(bprm->mm);
1212
1213 out_file:
1214         if (bprm->file) {
1215                 allow_write_access(bprm->file);
1216                 fput(bprm->file);
1217         }
1218
1219 out_kfree:
1220         kfree(bprm);
1221
1222 out_ret:
1223         return retval;
1224 }
1225
1226 int set_binfmt(struct linux_binfmt *new)
1227 {
1228         struct linux_binfmt *old = current->binfmt;
1229
1230         if (new) {
1231                 if (!try_module_get(new->module))
1232                         return -1;
1233         }
1234         current->binfmt = new;
1235         if (old)
1236                 module_put(old->module);
1237         return 0;
1238 }
1239
1240 EXPORT_SYMBOL(set_binfmt);
1241
1242 #define CORENAME_MAX_SIZE 64
1243
1244 /* format_corename will inspect the pattern parameter, and output a
1245  * name into corename, which must have space for at least
1246  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1247  */
1248 static void format_corename(char *corename, const char *pattern, long signr)
1249 {
1250         const char *pat_ptr = pattern;
1251         char *out_ptr = corename;
1252         char *const out_end = corename + CORENAME_MAX_SIZE;
1253         int rc;
1254         int pid_in_pattern = 0;
1255
1256         /* Repeat as long as we have more pattern to process and more output
1257            space */
1258         while (*pat_ptr) {
1259                 if (*pat_ptr != '%') {
1260                         if (out_ptr == out_end)
1261                                 goto out;
1262                         *out_ptr++ = *pat_ptr++;
1263                 } else {
1264                         switch (*++pat_ptr) {
1265                         case 0:
1266                                 goto out;
1267                         /* Double percent, output one percent */
1268                         case '%':
1269                                 if (out_ptr == out_end)
1270                                         goto out;
1271                                 *out_ptr++ = '%';
1272                                 break;
1273                         /* pid */
1274                         case 'p':
1275                                 pid_in_pattern = 1;
1276                                 rc = snprintf(out_ptr, out_end - out_ptr,
1277                                               "%d", current->tgid);
1278                                 if (rc > out_end - out_ptr)
1279                                         goto out;
1280                                 out_ptr += rc;
1281                                 break;
1282                         /* uid */
1283                         case 'u':
1284                                 rc = snprintf(out_ptr, out_end - out_ptr,
1285                                               "%d", current->uid);
1286                                 if (rc > out_end - out_ptr)
1287                                         goto out;
1288                                 out_ptr += rc;
1289                                 break;
1290                         /* gid */
1291                         case 'g':
1292                                 rc = snprintf(out_ptr, out_end - out_ptr,
1293                                               "%d", current->gid);
1294                                 if (rc > out_end - out_ptr)
1295                                         goto out;
1296                                 out_ptr += rc;
1297                                 break;
1298                         /* signal that caused the coredump */
1299                         case 's':
1300                                 rc = snprintf(out_ptr, out_end - out_ptr,
1301                                               "%ld", signr);
1302                                 if (rc > out_end - out_ptr)
1303                                         goto out;
1304                                 out_ptr += rc;
1305                                 break;
1306                         /* UNIX time of coredump */
1307                         case 't': {
1308                                 struct timeval tv;
1309                                 do_gettimeofday(&tv);
1310                                 rc = snprintf(out_ptr, out_end - out_ptr,
1311                                               "%lu", tv.tv_sec);
1312                                 if (rc > out_end - out_ptr)
1313                                         goto out;
1314                                 out_ptr += rc;
1315                                 break;
1316                         }
1317                         /* hostname */
1318                         case 'h':
1319                                 down_read(&uts_sem);
1320                                 rc = snprintf(out_ptr, out_end - out_ptr,
1321                                               "%s", system_utsname.nodename);
1322                                 up_read(&uts_sem);
1323                                 if (rc > out_end - out_ptr)
1324                                         goto out;
1325                                 out_ptr += rc;
1326                                 break;
1327                         /* executable */
1328                         case 'e':
1329                                 rc = snprintf(out_ptr, out_end - out_ptr,
1330                                               "%s", current->comm);
1331                                 if (rc > out_end - out_ptr)
1332                                         goto out;
1333                                 out_ptr += rc;
1334                                 break;
1335                         default:
1336                                 break;
1337                         }
1338                         ++pat_ptr;
1339                 }
1340         }
1341         /* Backward compatibility with core_uses_pid:
1342          *
1343          * If core_pattern does not include a %p (as is the default)
1344          * and core_uses_pid is set, then .%pid will be appended to
1345          * the filename */
1346         if (!pid_in_pattern
1347             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1348                 rc = snprintf(out_ptr, out_end - out_ptr,
1349                               ".%d", current->tgid);
1350                 if (rc > out_end - out_ptr)
1351                         goto out;
1352                 out_ptr += rc;
1353         }
1354       out:
1355         *out_ptr = 0;
1356 }
1357
1358 static void zap_threads (struct mm_struct *mm)
1359 {
1360         struct task_struct *g, *p;
1361         struct task_struct *tsk = current;
1362         struct completion *vfork_done = tsk->vfork_done;
1363         int traced = 0;
1364
1365         /*
1366          * Make sure nobody is waiting for us to release the VM,
1367          * otherwise we can deadlock when we wait on each other
1368          */
1369         if (vfork_done) {
1370                 tsk->vfork_done = NULL;
1371                 complete(vfork_done);
1372         }
1373
1374         read_lock(&tasklist_lock);
1375         do_each_thread(g,p)
1376                 if (mm == p->mm && p != tsk) {
1377                         force_sig_specific(SIGKILL, p);
1378                         mm->core_waiters++;
1379                         if (unlikely(p->ptrace) &&
1380                             unlikely(p->parent->mm == mm))
1381                                 traced = 1;
1382                 }
1383         while_each_thread(g,p);
1384
1385         read_unlock(&tasklist_lock);
1386
1387         while (unlikely(traced)) {
1388                 /*
1389                  * We are zapping a thread and the thread it ptraces.
1390                  * The tracee won't come out of TASK_TRACED state until
1391                  * its ptracer detaches.  That happens when the ptracer
1392                  * dies, but it synchronizes with us and so won't get
1393                  * that far until we finish the core dump.  If we're
1394                  * waiting for the tracee to synchronize but it stays
1395                  * blocked in TASK_TRACED, then we deadlock.  So, for
1396                  * this weirdo case we have to do another round with
1397                  * tasklist_lock write-locked to __ptrace_unlink the
1398                  * children that might cause this deadlock.  That will
1399                  * wake them up to process their pending SIGKILL.
1400                  *
1401                  * First, give everyone we just killed a chance to run
1402                  * so they can all get into the coredump synchronization.
1403                  * That should leave only the TASK_TRACED stragglers for
1404                  * us to wake up.  If a ptracer is still running, we'll
1405                  * have to come around again after letting it finish.
1406                  */
1407                 yield();
1408                 traced = 0;
1409                 write_lock_irq(&tasklist_lock);
1410                 do_each_thread(g,p) {
1411                         if (mm != p->mm || p == tsk ||
1412                             !p->ptrace || p->parent->mm != mm)
1413                                 continue;
1414                         if ((p->parent->flags & (PF_SIGNALED|PF_EXITING)) ||
1415                             (p->parent->state & (TASK_TRACED|TASK_STOPPED))) {
1416                                 /*
1417                                  * The parent is in the process of exiting
1418                                  * itself, or else it's stopped right now.
1419                                  * It cannot be in a ptrace call, and would
1420                                  * have to read_lock tasklist_lock before
1421                                  * it could start one, so we are safe here.
1422                                  */
1423                                 __ptrace_unlink(p);
1424                         } else {
1425                                 /*
1426                                  * Blargh!  The ptracer is not dying
1427                                  * yet, so we cannot be sure that it
1428                                  * isn't in the middle of a ptrace call.
1429                                  * We'll have to let it run to get into
1430                                  * coredump_wait and come around another
1431                                  * time to detach its tracee.
1432                                  */
1433                                 traced = 1;
1434                         }
1435                 } while_each_thread(g,p);
1436                 write_unlock_irq(&tasklist_lock);
1437         }
1438 }
1439
1440 static void coredump_wait(struct mm_struct *mm)
1441 {
1442         DECLARE_COMPLETION(startup_done);
1443
1444         mm->core_waiters++; /* let other threads block */
1445         mm->core_startup_done = &startup_done;
1446
1447         /* give other threads a chance to run: */
1448         yield();
1449
1450         zap_threads(mm);
1451         if (--mm->core_waiters) {
1452                 up_write(&mm->mmap_sem);
1453                 wait_for_completion(&startup_done);
1454         } else
1455                 up_write(&mm->mmap_sem);
1456         BUG_ON(mm->core_waiters);
1457 }
1458
1459 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1460 {
1461         char corename[CORENAME_MAX_SIZE + 1];
1462         struct mm_struct *mm = current->mm;
1463         struct linux_binfmt * binfmt;
1464         struct inode * inode;
1465         struct file * file;
1466         int retval = 0;
1467         int fsuid = current->fsuid;
1468         int flag = 0;
1469
1470         binfmt = current->binfmt;
1471         if (!binfmt || !binfmt->core_dump)
1472                 goto fail;
1473         if (current->tux_exit)
1474                 current->tux_exit();
1475         down_write(&mm->mmap_sem);
1476         if (!mm->dumpable) {
1477                 up_write(&mm->mmap_sem);
1478                 goto fail;
1479         }
1480
1481         /*
1482          *      We cannot trust fsuid as being the "true" uid of the
1483          *      process nor do we know its entire history. We only know it
1484          *      was tainted so we dump it as root in mode 2.
1485          */
1486         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1487                 flag = O_EXCL;          /* Stop rewrite attacks */
1488                 current->fsuid = 0;     /* Dump root private */
1489         }
1490         mm->dumpable = 0;
1491         init_completion(&mm->core_done);
1492         current->signal->group_exit = 1;
1493         current->signal->group_exit_code = exit_code;
1494         coredump_wait(mm);
1495
1496         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1497                 goto fail_unlock;
1498
1499         /*
1500          * lock_kernel() because format_corename() is controlled by sysctl, which
1501          * uses lock_kernel()
1502          */
1503         lock_kernel();
1504         format_corename(corename, core_pattern, signr);
1505         unlock_kernel();
1506         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1507         if (IS_ERR(file))
1508                 goto fail_unlock;
1509         inode = file->f_dentry->d_inode;
1510         if (inode->i_nlink > 1)
1511                 goto close_fail;        /* multiple links - don't dump */
1512         if (d_unhashed(file->f_dentry))
1513                 goto close_fail;
1514
1515         if (!S_ISREG(inode->i_mode))
1516                 goto close_fail;
1517         if (!file->f_op)
1518                 goto close_fail;
1519         if (!file->f_op->write)
1520                 goto close_fail;
1521         if (do_truncate(file->f_dentry, 0) != 0)
1522                 goto close_fail;
1523
1524         retval = binfmt->core_dump(signr, regs, file);
1525
1526         if (retval)
1527                 current->signal->group_exit_code |= 0x80;
1528 close_fail:
1529         filp_close(file, NULL);
1530 fail_unlock:
1531         current->fsuid = fsuid;
1532         complete_all(&mm->core_done);
1533 fail:
1534         return retval;
1535 }