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
  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[128] = "core";
  63 int suid_dumpable = 0;
  64
  65 EXPORT_SYMBOL(suid_dumpable);
  66 /* The maximal length of core_pattern is also specified in sysctl.c */
  67
  68 static struct linux_binfmt *formats;
  69 static DEFINE_RWLOCK(binfmt_lock);
  70
  71 int register_binfmt(struct linux_binfmt * fmt)
  72 {
  73         struct linux_binfmt ** tmp = &formats;
  74
  75         if (!fmt)
  76                 return -EINVAL;
  77         if (fmt->next)
  78                 return -EBUSY;
  79         write_lock(&binfmt_lock);
  80         while (*tmp) {
  81                 if (fmt == *tmp) {
  82                         write_unlock(&binfmt_lock);
  83                         return -EBUSY;
  84                 }
  85                 tmp = &(*tmp)->next;
  86         }
  87         fmt->next = formats;
  88         formats = fmt;
  89         write_unlock(&binfmt_lock);
  90         return 0;
  91 }
  92
  93 EXPORT_SYMBOL(register_binfmt);
  94
  95 int unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         struct linux_binfmt ** tmp = &formats;
  98
  99         write_lock(&binfmt_lock);
 100         while (*tmp) {
 101                 if (fmt == *tmp) {
 102                         *tmp = fmt->next;
 103                         fmt->next = NULL;
 104                         write_unlock(&binfmt_lock);
 105                         return 0;
 106                 }
 107                 tmp = &(*tmp)->next;
 108         }
 109         write_unlock(&binfmt_lock);
 110         return -EINVAL;
 111 }
 112
 113 EXPORT_SYMBOL(unregister_binfmt);
 114
 115 static inline void put_binfmt(struct linux_binfmt * fmt)
 116 {
 117         module_put(fmt->module);
 118 }
 119
 120 /*
 121  * Note that a shared library must be both readable and executable due to
 122  * security reasons.
 123  *
 124  * Also note that we take the address to load from from the file itself.
 125  */
 126 asmlinkage long sys_uselib(const char __user * library)
 127 {
 128         struct file * file;
 129         struct nameidata nd;
 130         int error;
 131
 132         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 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 = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 141         if (error)
 142                 goto exit;
 143
 144         file = nameidata_to_filp(&nd, 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         release_open_intent(&nd);
 173         path_release(&nd);
 174         goto out;
 175 }
 176
 177 /*
 178  * count() counts the number of strings in array ARGV.
 179  */
 180 static int count(char __user * __user * argv, int max)
 181 {
 182         int i = 0;
 183
 184         if (argv != NULL) {
 185                 for (;;) {
 186                         char __user * p;
 187
 188                         if (get_user(p, argv))
 189                                 return -EFAULT;
 190                         if (!p)
 191                                 break;
 192                         argv++;
 193                         if(++i > max)
 194                                 return -E2BIG;
 195                         cond_resched();
 196                 }
 197         }
 198         return i;
 199 }
 200
 201 /*
 202  * 'copy_strings()' copies argument/environment strings from user
 203  * memory to free pages in kernel mem. These are in a format ready
 204  * to be put directly into the top of new user memory.
 205  */
 206 static int copy_strings(int argc, char __user * __user * argv,
 207                         struct linux_binprm *bprm)
 208 {
 209         struct page *kmapped_page = NULL;
 210         char *kaddr = NULL;
 211         int ret;
 212
 213         while (argc-- > 0) {
 214                 char __user *str;
 215                 int len;
 216                 unsigned long pos;
 217
 218                 if (get_user(str, argv+argc) ||
 219                                 !(len = strnlen_user(str, bprm->p))) {
 220                         ret = -EFAULT;
 221                         goto out;
 222                 }
 223
 224                 if (bprm->p < len)  {
 225                         ret = -E2BIG;
 226                         goto out;
 227                 }
 228
 229                 bprm->p -= len;
 230                 /* XXX: add architecture specific overflow check here. */
 231                 pos = bprm->p;
 232
 233                 while (len > 0) {
 234                         int i, new, err;
 235                         int offset, bytes_to_copy;
 236                         struct page *page;
 237
 238                         offset = pos % PAGE_SIZE;
 239                         i = pos/PAGE_SIZE;
 240                         page = bprm->page[i];
 241                         new = 0;
 242                         if (!page) {
 243                                 page = alloc_page(GFP_HIGHUSER);
 244                                 bprm->page[i] = page;
 245                                 if (!page) {
 246                                         ret = -ENOMEM;
 247                                         goto out;
 248                                 }
 249                                 new = 1;
 250                         }
 251
 252                         if (page != kmapped_page) {
 253                                 if (kmapped_page)
 254                                         kunmap(kmapped_page);
 255                                 kmapped_page = page;
 256                                 kaddr = kmap(kmapped_page);
 257                         }
 258                         if (new && offset)
 259                                 memset(kaddr, 0, offset);
 260                         bytes_to_copy = PAGE_SIZE - offset;
 261                         if (bytes_to_copy > len) {
 262                                 bytes_to_copy = len;
 263                                 if (new)
 264                                         memset(kaddr+offset+len, 0,
 265                                                 PAGE_SIZE-offset-len);
 266                         }
 267                         err = copy_from_user(kaddr+offset, str, bytes_to_copy);
 268                         if (err) {
 269                                 ret = -EFAULT;
 270                                 goto out;
 271                         }
 272
 273                         pos += bytes_to_copy;
 274                         str += bytes_to_copy;
 275                         len -= bytes_to_copy;
 276                 }
 277         }
 278         ret = 0;
 279 out:
 280         if (kmapped_page)
 281                 kunmap(kmapped_page);
 282         return ret;
 283 }
 284
 285 /*
 286  * Like copy_strings, but get argv and its values from kernel memory.
 287  */
 288 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 289 {
 290         int r;
 291         mm_segment_t oldfs = get_fs();
 292         set_fs(KERNEL_DS);
 293         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 294         set_fs(oldfs);
 295         return r;
 296 }
 297
 298 EXPORT_SYMBOL(copy_strings_kernel);
 299
 300 #ifdef CONFIG_MMU
 301 /*
 302  * This routine is used to map in a page into an address space: needed by
 303  * execve() for the initial stack and environment pages.
 304  *
 305  * vma->vm_mm->mmap_sem is held for writing.
 306  */
 307 void install_arg_page(struct vm_area_struct *vma,
 308                         struct page *page, unsigned long address)
 309 {
 310         struct mm_struct *mm = vma->vm_mm;
 311         pte_t * pte;
 312         spinlock_t *ptl;
 313
 314         if (unlikely(anon_vma_prepare(vma)))
 315                 goto out;
 316
 317         flush_dcache_page(page);
 318         pte = get_locked_pte(mm, address, &ptl);
 319         if (!pte)
 320                 goto out;
 321         if (!pte_none(*pte)) {
 322                 pte_unmap_unlock(pte, ptl);
 323                 goto out;
 324         }
 325         inc_mm_counter(mm, anon_rss);
 326         lru_cache_add_active(page);
 327         set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
 328                                         page, vma->vm_page_prot))));
 329         page_add_new_anon_rmap(page, vma, address);
 330         pte_unmap_unlock(pte, ptl);
 331
 332         /* no need for flush_tlb */
 333         return;
 334 out:
 335         __free_page(page);
 336         force_sig(SIGKILL, current);
 337 }
 338
 339 #define EXTRA_STACK_VM_PAGES    20      /* random */
 340
 341 int setup_arg_pages(struct linux_binprm *bprm,
 342                     unsigned long stack_top,
 343                     int executable_stack)
 344 {
 345         unsigned long stack_base;
 346         struct vm_area_struct *mpnt;
 347         struct mm_struct *mm = current->mm;
 348         int i, ret;
 349         long arg_size;
 350
 351 #ifdef CONFIG_STACK_GROWSUP
 352         /* Move the argument and environment strings to the bottom of the
 353          * stack space.
 354          */
 355         int offset, j;
 356         char *to, *from;
 357
 358         /* Start by shifting all the pages down */
 359         i = 0;
 360         for (j = 0; j < MAX_ARG_PAGES; j++) {
 361                 struct page *page = bprm->page[j];
 362                 if (!page)
 363                         continue;
 364                 bprm->page[i++] = page;
 365         }
 366
 367         /* Now move them within their pages */
 368         offset = bprm->p % PAGE_SIZE;
 369         to = kmap(bprm->page[0]);
 370         for (j = 1; j < i; j++) {
 371                 memmove(to, to + offset, PAGE_SIZE - offset);
 372                 from = kmap(bprm->page[j]);
 373                 memcpy(to + PAGE_SIZE - offset, from, offset);
 374                 kunmap(bprm->page[j - 1]);
 375                 to = from;
 376         }
 377         memmove(to, to + offset, PAGE_SIZE - offset);
 378         kunmap(bprm->page[j - 1]);
 379
 380         /* Limit stack size to 1GB */
 381         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 382         if (stack_base > (1 << 30))
 383                 stack_base = 1 << 30;
 384         stack_base = PAGE_ALIGN(stack_top - stack_base);
 385
 386         /* Adjust bprm->p to point to the end of the strings. */
 387         bprm->p = stack_base + PAGE_SIZE * i - offset;
 388
 389         mm->arg_start = stack_base;
 390         arg_size = i << PAGE_SHIFT;
 391
 392         /* zero pages that were copied above */
 393         while (i < MAX_ARG_PAGES)
 394                 bprm->page[i++] = NULL;
 395 #else
 396         stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
 397         stack_base = PAGE_ALIGN(stack_base);
 398         bprm->p += stack_base;
 399         mm->arg_start = bprm->p;
 400         arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
 401 #endif
 402
 403         arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 404
 405         if (bprm->loader)
 406                 bprm->loader += stack_base;
 407         bprm->exec += stack_base;
 408
 409         mpnt = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 410         if (!mpnt)
 411                 return -ENOMEM;
 412
 413         down_write(&mm->mmap_sem);
 414         {
 415                 mpnt->vm_mm = mm;
 416 #ifdef CONFIG_STACK_GROWSUP
 417                 mpnt->vm_start = stack_base;
 418                 mpnt->vm_end = stack_base + arg_size;
 419 #else
 420                 mpnt->vm_end = stack_top;
 421                 mpnt->vm_start = mpnt->vm_end - arg_size;
 422 #endif
 423                 /* Adjust stack execute permissions; explicitly enable
 424                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 425                  * and leave alone (arch default) otherwise. */
 426                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 427                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 428                 else if (executable_stack == EXSTACK_DISABLE_X)
 429                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 430                 else
 431                         mpnt->vm_flags = VM_STACK_FLAGS;
 432                 mpnt->vm_flags |= mm->def_flags;
 433                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 434                 if ((ret = insert_vm_struct(mm, mpnt))) {
 435                         up_write(&mm->mmap_sem);
 436                         kmem_cache_free(vm_area_cachep, mpnt);
 437                         return ret;
 438                 }
 439                 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
 440         }
 441
 442         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 443                 struct page *page = bprm->page[i];
 444                 if (page) {
 445                         bprm->page[i] = NULL;
 446                         install_arg_page(mpnt, page, stack_base);
 447                 }
 448                 stack_base += PAGE_SIZE;
 449         }
 450         up_write(&mm->mmap_sem);
 451
 452         return 0;
 453 }
 454
 455 EXPORT_SYMBOL(setup_arg_pages);
 456
 457 #define free_arg_pages(bprm) do { } while (0)
 458
 459 #else
 460
 461 static inline void free_arg_pages(struct linux_binprm *bprm)
 462 {
 463         int i;
 464
 465         for (i = 0; i < MAX_ARG_PAGES; i++) {
 466                 if (bprm->page[i])
 467                         __free_page(bprm->page[i]);
 468                 bprm->page[i] = NULL;
 469         }
 470 }
 471
 472 #endif /* CONFIG_MMU */
 473
 474 struct file *open_exec(const char *name)
 475 {
 476         struct nameidata nd;
 477         int err;
 478         struct file *file;
 479
 480         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 481         file = ERR_PTR(err);
 482
 483         if (!err) {
 484                 struct inode *inode = nd.dentry->d_inode;
 485                 file = ERR_PTR(-EACCES);
 486                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 487                     S_ISREG(inode->i_mode)) {
 488                         int err = vfs_permission(&nd, MAY_EXEC);
 489                         file = ERR_PTR(err);
 490                         if (!err) {
 491                                 file = nameidata_to_filp(&nd, O_RDONLY);
 492                                 if (!IS_ERR(file)) {
 493                                         err = deny_write_access(file);
 494                                         if (err) {
 495                                                 fput(file);
 496                                                 file = ERR_PTR(err);
 497                                         }
 498                                 }
 499 out:
 500                                 return file;
 501                         }
 502                 }
 503                 release_open_intent(&nd);
 504                 path_release(&nd);
 505         }
 506         goto out;
 507 }
 508
 509 EXPORT_SYMBOL(open_exec);
 510
 511 int kernel_read(struct file *file, unsigned long offset,
 512         char *addr, unsigned long count)
 513 {
 514         mm_segment_t old_fs;
 515         loff_t pos = offset;
 516         int result;
 517
 518         old_fs = get_fs();
 519         set_fs(get_ds());
 520         /* The cast to a user pointer is valid due to the set_fs() */
 521         result = vfs_read(file, (void __user *)addr, count, &pos);
 522         set_fs(old_fs);
 523         return result;
 524 }
 525
 526 EXPORT_SYMBOL(kernel_read);
 527
 528 static int exec_mmap(struct mm_struct *mm)
 529 {
 530         struct task_struct *tsk;
 531         struct mm_struct * old_mm, *active_mm;
 532
 533         /* Notify parent that we're no longer interested in the old VM */
 534         tsk = current;
 535         old_mm = current->mm;
 536         mm_release(tsk, old_mm);
 537
 538         if (old_mm) {
 539                 /*
 540                  * Make sure that if there is a core dump in progress
 541                  * for the old mm, we get out and die instead of going
 542                  * through with the exec.  We must hold mmap_sem around
 543                  * checking core_waiters and changing tsk->mm.  The
 544                  * core-inducing thread will increment core_waiters for
 545                  * each thread whose ->mm == old_mm.
 546                  */
 547                 down_read(&old_mm->mmap_sem);
 548                 if (unlikely(old_mm->core_waiters)) {
 549                         up_read(&old_mm->mmap_sem);
 550                         return -EINTR;
 551                 }
 552         }
 553         task_lock(tsk);
 554         active_mm = tsk->active_mm;
 555         tsk->mm = mm;
 556         tsk->active_mm = mm;
 557         activate_mm(active_mm, mm);
 558         task_unlock(tsk);
 559         arch_pick_mmap_layout(mm);
 560         if (old_mm) {
 561                 up_read(&old_mm->mmap_sem);
 562                 BUG_ON(active_mm != old_mm);
 563                 mmput(old_mm);
 564                 return 0;
 565         }
 566         mmdrop(active_mm);
 567         return 0;
 568 }
 569
 570 /*
 571  * This function makes sure the current process has its own signal table,
 572  * so that flush_signal_handlers can later reset the handlers without
 573  * disturbing other processes.  (Other processes might share the signal
 574  * table via the CLONE_SIGHAND option to clone().)
 575  */
 576 static int de_thread(struct task_struct *tsk)
 577 {
 578         struct signal_struct *sig = tsk->signal;
 579         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 580         spinlock_t *lock = &oldsighand->siglock;
 581         struct task_struct *leader = NULL;
 582         int count;
 583
 584         /*
 585          * If we don't share sighandlers, then we aren't sharing anything
 586          * and we can just re-use it all.
 587          */
 588         if (atomic_read(&oldsighand->count) <= 1) {
 589                 BUG_ON(atomic_read(&sig->count) != 1);
 590                 exit_itimers(sig);
 591                 return 0;
 592         }
 593
 594         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 595         if (!newsighand)
 596                 return -ENOMEM;
 597
 598         if (thread_group_empty(tsk))
 599                 goto no_thread_group;
 600
 601         /*
 602          * Kill all other threads in the thread group.
 603          * We must hold tasklist_lock to call zap_other_threads.
 604          */
 605         read_lock(&tasklist_lock);
 606         spin_lock_irq(lock);
 607         if (sig->flags & SIGNAL_GROUP_EXIT) {
 608                 /*
 609                  * Another group action in progress, just
 610                  * return so that the signal is processed.
 611                  */
 612                 spin_unlock_irq(lock);
 613                 read_unlock(&tasklist_lock);
 614                 kmem_cache_free(sighand_cachep, newsighand);
 615                 return -EAGAIN;
 616         }
 617
 618         /*
 619          * child_reaper ignores SIGKILL, change it now.
 620          * Reparenting needs write_lock on tasklist_lock,
 621          * so it is safe to do it under read_lock.
 622          */
 623         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 624                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 625
 626         zap_other_threads(tsk);
 627         read_unlock(&tasklist_lock);
 628
 629         /*
 630          * Account for the thread group leader hanging around:
 631          */
 632         count = 1;
 633         if (!thread_group_leader(tsk)) {
 634                 count = 2;
 635                 /*
 636                  * The SIGALRM timer survives the exec, but needs to point
 637                  * at us as the new group leader now.  We have a race with
 638                  * a timer firing now getting the old leader, so we need to
 639                  * synchronize with any firing (by calling del_timer_sync)
 640                  * before we can safely let the old group leader die.
 641                  */
 642                 sig->tsk = tsk;
 643                 spin_unlock_irq(lock);
 644                 if (hrtimer_cancel(&sig->real_timer))
 645                         hrtimer_restart(&sig->real_timer);
 646                 spin_lock_irq(lock);
 647         }
 648         while (atomic_read(&sig->count) > count) {
 649                 sig->group_exit_task = tsk;
 650                 sig->notify_count = count;
 651                 __set_current_state(TASK_UNINTERRUPTIBLE);
 652                 spin_unlock_irq(lock);
 653                 schedule();
 654                 spin_lock_irq(lock);
 655         }
 656         sig->group_exit_task = NULL;
 657         sig->notify_count = 0;
 658         spin_unlock_irq(lock);
 659
 660         /*
 661          * At this point all other threads have exited, all we have to
 662          * do is to wait for the thread group leader to become inactive,
 663          * and to assume its PID:
 664          */
 665         if (!thread_group_leader(tsk)) {
 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                 leader = tsk->group_leader;
 672                 while (leader->exit_state != EXIT_ZOMBIE)
 673                         yield();
 674
 675                 /*
 676                  * The only record we have of the real-time age of a
 677                  * process, regardless of execs it's done, is start_time.
 678                  * All the past CPU time is accumulated in signal_struct
 679                  * from sister threads now dead.  But in this non-leader
 680                  * exec, nothing survives from the original leader thread,
 681                  * whose birth marks the true age of this process now.
 682                  * When we take on its identity by switching to its PID, we
 683                  * also take its birthdate (always earlier than our own).
 684                  */
 685                 tsk->start_time = leader->start_time;
 686
 687                 write_lock_irq(&tasklist_lock);
 688
 689                 BUG_ON(leader->tgid != tsk->tgid);
 690                 BUG_ON(tsk->pid == tsk->tgid);
 691                 /*
 692                  * An exec() starts a new thread group with the
 693                  * TGID of the previous thread group. Rehash the
 694                  * two threads with a switched PID, and release
 695                  * the former thread group leader:
 696                  */
 697
 698                 /* Become a process group leader with the old leader's pid.
 699                  * The old leader becomes a thread of the this thread group.
 700                  * Note: The old leader also uses this pid until release_task
 701                  *       is called.  Odd but simple and correct.
 702                  */
 703                 detach_pid(tsk, PIDTYPE_PID);
 704                 tsk->pid = leader->pid;
 705                 attach_pid(tsk, PIDTYPE_PID,  tsk->pid);
 706                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 707                 transfer_pid(leader, tsk, PIDTYPE_SID);
 708                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 709
 710                 tsk->group_leader = tsk;
 711                 leader->group_leader = tsk;
 712
 713                 tsk->exit_signal = SIGCHLD;
 714
 715                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 716                 leader->exit_state = EXIT_DEAD;
 717
 718                 write_unlock_irq(&tasklist_lock);
 719         }
 720
 721         /*
 722          * There may be one thread left which is just exiting,
 723          * but it's safe to stop telling the group to kill themselves.
 724          */
 725         sig->flags = 0;
 726
 727 no_thread_group:
 728         exit_itimers(sig);
 729         if (leader)
 730                 release_task(leader);
 731
 732         BUG_ON(atomic_read(&sig->count) != 1);
 733
 734         if (atomic_read(&oldsighand->count) == 1) {
 735                 /*
 736                  * Now that we nuked the rest of the thread group,
 737                  * it turns out we are not sharing sighand any more either.
 738                  * So we can just keep it.
 739                  */
 740                 kmem_cache_free(sighand_cachep, newsighand);
 741         } else {
 742                 /*
 743                  * Move our state over to newsighand and switch it in.
 744                  */
 745                 atomic_set(&newsighand->count, 1);
 746                 memcpy(newsighand->action, oldsighand->action,
 747                        sizeof(newsighand->action));
 748
 749                 write_lock_irq(&tasklist_lock);
 750                 spin_lock(&oldsighand->siglock);
 751                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 752
 753                 rcu_assign_pointer(tsk->sighand, newsighand);
 754                 recalc_sigpending();
 755
 756                 spin_unlock(&newsighand->siglock);
 757                 spin_unlock(&oldsighand->siglock);
 758                 write_unlock_irq(&tasklist_lock);
 759
 760                 if (atomic_dec_and_test(&oldsighand->count))
 761                         kmem_cache_free(sighand_cachep, oldsighand);
 762         }
 763
 764         BUG_ON(!thread_group_leader(tsk));
 765         return 0;
 766 }
 767
 768 /*
 769  * These functions flushes out all traces of the currently running executable
 770  * so that a new one can be started
 771  */
 772
 773 static void flush_old_files(struct files_struct * files)
 774 {
 775         long j = -1;
 776         struct fdtable *fdt;
 777
 778         spin_lock(&files->file_lock);
 779         for (;;) {
 780                 unsigned long set, i;
 781
 782                 j++;
 783                 i = j * __NFDBITS;
 784                 fdt = files_fdtable(files);
 785                 if (i >= fdt->max_fds)
 786                         break;
 787                 set = fdt->close_on_exec->fds_bits[j];
 788                 if (!set)
 789                         continue;
 790                 fdt->close_on_exec->fds_bits[j] = 0;
 791                 spin_unlock(&files->file_lock);
 792                 for ( ; set ; i++,set >>= 1) {
 793                         if (set & 1) {
 794                                 sys_close(i);
 795                         }
 796                 }
 797                 spin_lock(&files->file_lock);
 798
 799         }
 800         spin_unlock(&files->file_lock);
 801 }
 802
 803 void get_task_comm(char *buf, struct task_struct *tsk)
 804 {
 805         /* buf must be at least sizeof(tsk->comm) in size */
 806         task_lock(tsk);
 807         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 808         task_unlock(tsk);
 809 }
 810
 811 void set_task_comm(struct task_struct *tsk, char *buf)
 812 {
 813         task_lock(tsk);
 814         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 815         task_unlock(tsk);
 816 }
 817
 818 int flush_old_exec(struct linux_binprm * bprm)
 819 {
 820         char * name;
 821         int i, ch, retval;
 822         struct files_struct *files;
 823         char tcomm[sizeof(current->comm)];
 824
 825         /*
 826          * Make sure we have a private signal table and that
 827          * we are unassociated from the previous thread group.
 828          */
 829         retval = de_thread(current);
 830         if (retval)
 831                 goto out;
 832
 833         /*
 834          * Make sure we have private file handles. Ask the
 835          * fork helper to do the work for us and the exit
 836          * helper to do the cleanup of the old one.
 837          */
 838         files = current->files;         /* refcounted so safe to hold */
 839         retval = unshare_files();
 840         if (retval)
 841                 goto out;
 842         /*
 843          * Release all of the old mmap stuff
 844          */
 845         retval = exec_mmap(bprm->mm);
 846         if (retval)
 847                 goto mmap_failed;
 848
 849         bprm->mm = NULL;                /* We're using it now */
 850
 851         /* This is the point of no return */
 852         put_files_struct(files);
 853
 854         current->sas_ss_sp = current->sas_ss_size = 0;
 855
 856         if (current->euid == current->uid && current->egid == current->gid)
 857                 current->mm->dumpable = 1;
 858         else
 859                 current->mm->dumpable = suid_dumpable;
 860
 861         name = bprm->filename;
 862
 863         /* Copies the binary name from after last slash */
 864         for (i=0; (ch = *(name++)) != '\0';) {
 865                 if (ch == '/')
 866                         i = 0; /* overwrite what we wrote */
 867                 else
 868                         if (i < (sizeof(tcomm) - 1))
 869                                 tcomm[i++] = ch;
 870         }
 871         tcomm[i] = '\0';
 872         set_task_comm(current, tcomm);
 873
 874         current->flags &= ~PF_RANDOMIZE;
 875         flush_thread();
 876
 877         /* Set the new mm task size. We have to do that late because it may
 878          * depend on TIF_32BIT which is only updated in flush_thread() on
 879          * some architectures like powerpc
 880          */
 881         current->mm->task_size = TASK_SIZE;
 882
 883         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 884             file_permission(bprm->file, MAY_READ) ||
 885             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 886                 suid_keys(current);
 887                 current->mm->dumpable = suid_dumpable;
 888         }
 889
 890         /* An exec changes our domain. We are no longer part of the thread
 891            group */
 892
 893         current->self_exec_id++;
 894
 895         flush_signal_handlers(current, 0);
 896         flush_old_files(current->files);
 897
 898         return 0;
 899
 900 mmap_failed:
 901         reset_files_struct(current, 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_path.dentry->d_inode;
 916         int retval;
 917
 918         mode = inode->i_mode;
 919         if (bprm->file->f_op == NULL)
 920                 return -EACCES;
 921
 922         bprm->e_uid = current->euid;
 923         bprm->e_gid = current->egid;
 924
 925         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
 926                 /* Set-uid? */
 927                 if (mode & S_ISUID) {
 928                         current->personality &= ~PER_CLEAR_ON_SETID;
 929                         bprm->e_uid = inode->i_uid;
 930                 }
 931
 932                 /* Set-gid? */
 933                 /*
 934                  * If setgid is set but no group execute bit then this
 935                  * is a candidate for mandatory locking, not a setgid
 936                  * executable.
 937                  */
 938                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 939                         current->personality &= ~PER_CLEAR_ON_SETID;
 940                         bprm->e_gid = inode->i_gid;
 941                 }
 942         }
 943
 944         /* fill in binprm security blob */
 945         retval = security_bprm_set(bprm);
 946         if (retval)
 947                 return retval;
 948
 949         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 950         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 951 }
 952
 953 EXPORT_SYMBOL(prepare_binprm);
 954
 955 static int unsafe_exec(struct task_struct *p)
 956 {
 957         int unsafe = 0;
 958         if (p->ptrace & PT_PTRACED) {
 959                 if (p->ptrace & PT_PTRACE_CAP)
 960                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 961                 else
 962                         unsafe |= LSM_UNSAFE_PTRACE;
 963         }
 964         if (atomic_read(&p->fs->count) > 1 ||
 965             atomic_read(&p->files->count) > 1 ||
 966             atomic_read(&p->sighand->count) > 1)
 967                 unsafe |= LSM_UNSAFE_SHARE;
 968
 969         return unsafe;
 970 }
 971
 972 void compute_creds(struct linux_binprm *bprm)
 973 {
 974         int unsafe;
 975
 976         if (bprm->e_uid != current->uid)
 977                 suid_keys(current);
 978         exec_keys(current);
 979
 980         task_lock(current);
 981         unsafe = unsafe_exec(current);
 982         security_bprm_apply_creds(bprm, unsafe);
 983         task_unlock(current);
 984         security_bprm_post_apply_creds(bprm);
 985 }
 986 EXPORT_SYMBOL(compute_creds);
 987
 988 /*
 989  * Arguments are '\0' separated strings found at the location bprm->p
 990  * points to; chop off the first by relocating brpm->p to right after
 991  * the first '\0' encountered.
 992  */
 993 void remove_arg_zero(struct linux_binprm *bprm)
 994 {
 995         if (bprm->argc) {
 996                 char ch;
 997
 998                 do {
 999                         unsigned long offset;
1000                         unsigned long index;
1001                         char *kaddr;
1002                         struct page *page;
1003
1004                         offset = bprm->p & ~PAGE_MASK;
1005                         index = bprm->p >> PAGE_SHIFT;
1006
1007                         page = bprm->page[index];
1008                         kaddr = kmap_atomic(page, KM_USER0);
1009
1010                         /* run through page until we reach end or find NUL */
1011                         do {
1012                                 ch = *(kaddr + offset);
1013
1014                                 /* discard that character... */
1015                                 bprm->p++;
1016                                 offset++;
1017                         } while (offset < PAGE_SIZE && ch != '\0');
1018
1019                         kunmap_atomic(kaddr, KM_USER0);
1020
1021                         /* free the old page */
1022                         if (offset == PAGE_SIZE) {
1023                                 __free_page(page);
1024                                 bprm->page[index] = NULL;
1025                         }
1026                 } while (ch != '\0');
1027
1028                 bprm->argc--;
1029         }
1030 }
1031 EXPORT_SYMBOL(remove_arg_zero);
1032
1033 /*
1034  * cycle the list of binary formats handler, until one recognizes the image
1035  */
1036 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1037 {
1038         int try,retval;
1039         struct linux_binfmt *fmt;
1040 #ifdef __alpha__
1041         /* handle /sbin/loader.. */
1042         {
1043             struct exec * eh = (struct exec *) bprm->buf;
1044
1045             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1046                 (eh->fh.f_flags & 0x3000) == 0x3000)
1047             {
1048                 struct file * file;
1049                 unsigned long loader;
1050
1051                 allow_write_access(bprm->file);
1052                 fput(bprm->file);
1053                 bprm->file = NULL;
1054
1055                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1056
1057                 file = open_exec("/sbin/loader");
1058                 retval = PTR_ERR(file);
1059                 if (IS_ERR(file))
1060                         return retval;
1061
1062                 /* Remember if the application is TASO.  */
1063                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1064
1065                 bprm->file = file;
1066                 bprm->loader = loader;
1067                 retval = prepare_binprm(bprm);
1068                 if (retval<0)
1069                         return retval;
1070                 /* should call search_binary_handler recursively here,
1071                    but it does not matter */
1072             }
1073         }
1074 #endif
1075         retval = security_bprm_check(bprm);
1076         if (retval)
1077                 return retval;
1078
1079         /* kernel module loader fixup */
1080         /* so we don't try to load run modprobe in kernel space. */
1081         set_fs(USER_DS);
1082
1083         retval = audit_bprm(bprm);
1084         if (retval)
1085                 return retval;
1086
1087         retval = -ENOENT;
1088         for (try=0; try<2; try++) {
1089                 read_lock(&binfmt_lock);
1090                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1091                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1092                         if (!fn)
1093                                 continue;
1094                         if (!try_module_get(fmt->module))
1095                                 continue;
1096                         read_unlock(&binfmt_lock);
1097                         retval = fn(bprm, regs);
1098                         if (retval >= 0) {
1099                                 put_binfmt(fmt);
1100                                 allow_write_access(bprm->file);
1101                                 if (bprm->file)
1102                                         fput(bprm->file);
1103                                 bprm->file = NULL;
1104                                 current->did_exec = 1;
1105                                 proc_exec_connector(current);
1106                                 return retval;
1107                         }
1108                         read_lock(&binfmt_lock);
1109                         put_binfmt(fmt);
1110                         if (retval != -ENOEXEC || bprm->mm == NULL)
1111                                 break;
1112                         if (!bprm->file) {
1113                                 read_unlock(&binfmt_lock);
1114                                 return retval;
1115                         }
1116                 }
1117                 read_unlock(&binfmt_lock);
1118                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1119                         break;
1120 #ifdef CONFIG_KMOD
1121                 }else{
1122 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1123                         if (printable(bprm->buf[0]) &&
1124                             printable(bprm->buf[1]) &&
1125                             printable(bprm->buf[2]) &&
1126                             printable(bprm->buf[3]))
1127                                 break; /* -ENOEXEC */
1128                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1129 #endif
1130                 }
1131         }
1132         return retval;
1133 }
1134
1135 EXPORT_SYMBOL(search_binary_handler);
1136
1137 /*
1138  * sys_execve() executes a new program.
1139  */
1140 int do_execve(char * filename,
1141         char __user *__user *argv,
1142         char __user *__user *envp,
1143         struct pt_regs * regs)
1144 {
1145         struct linux_binprm *bprm;
1146         struct file *file;
1147         int retval;
1148         int i;
1149
1150         retval = -ENOMEM;
1151         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1152         if (!bprm)
1153                 goto out_ret;
1154
1155         file = open_exec(filename);
1156         retval = PTR_ERR(file);
1157         if (IS_ERR(file))
1158                 goto out_kfree;
1159
1160         sched_exec();
1161
1162         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1163
1164         bprm->file = file;
1165         bprm->filename = filename;
1166         bprm->interp = filename;
1167         bprm->mm = mm_alloc();
1168         retval = -ENOMEM;
1169         if (!bprm->mm)
1170                 goto out_file;
1171
1172         retval = init_new_context(current, bprm->mm);
1173         if (retval < 0)
1174                 goto out_mm;
1175
1176         bprm->argc = count(argv, bprm->p / sizeof(void *));
1177         if ((retval = bprm->argc) < 0)
1178                 goto out_mm;
1179
1180         bprm->envc = count(envp, bprm->p / sizeof(void *));
1181         if ((retval = bprm->envc) < 0)
1182                 goto out_mm;
1183
1184         retval = security_bprm_alloc(bprm);
1185         if (retval)
1186                 goto out;
1187
1188         retval = prepare_binprm(bprm);
1189         if (retval < 0)
1190                 goto out;
1191
1192         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1193         if (retval < 0)
1194                 goto out;
1195
1196         bprm->exec = bprm->p;
1197         retval = copy_strings(bprm->envc, envp, bprm);
1198         if (retval < 0)
1199                 goto out;
1200
1201         retval = copy_strings(bprm->argc, argv, bprm);
1202         if (retval < 0)
1203                 goto out;
1204
1205         retval = search_binary_handler(bprm,regs);
1206         if (retval >= 0) {
1207                 free_arg_pages(bprm);
1208
1209                 /* execve success */
1210                 security_bprm_free(bprm);
1211                 acct_update_integrals(current);
1212                 kfree(bprm);
1213                 return retval;
1214         }
1215
1216 out:
1217         /* Something went wrong, return the inode and free the argument pages*/
1218         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1219                 struct page * page = bprm->page[i];
1220                 if (page)
1221                         __free_page(page);
1222         }
1223
1224         if (bprm->security)
1225                 security_bprm_free(bprm);
1226
1227 out_mm:
1228         if (bprm->mm)
1229                 mmdrop(bprm->mm);
1230
1231 out_file:
1232         if (bprm->file) {
1233                 allow_write_access(bprm->file);
1234                 fput(bprm->file);
1235         }
1236
1237 out_kfree:
1238         kfree(bprm);
1239
1240 out_ret:
1241         return retval;
1242 }
1243
1244 int set_binfmt(struct linux_binfmt *new)
1245 {
1246         struct linux_binfmt *old = current->binfmt;
1247
1248         if (new) {
1249                 if (!try_module_get(new->module))
1250                         return -1;
1251         }
1252         current->binfmt = new;
1253         if (old)
1254                 module_put(old->module);
1255         return 0;
1256 }
1257
1258 EXPORT_SYMBOL(set_binfmt);
1259
1260 #define CORENAME_MAX_SIZE 64
1261
1262 /* format_corename will inspect the pattern parameter, and output a
1263  * name into corename, which must have space for at least
1264  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1265  */
1266 static int format_corename(char *corename, const char *pattern, long signr)
1267 {
1268         const char *pat_ptr = pattern;
1269         char *out_ptr = corename;
1270         char *const out_end = corename + CORENAME_MAX_SIZE;
1271         int rc;
1272         int pid_in_pattern = 0;
1273         int ispipe = 0;
1274
1275         if (*pattern == '|')
1276                 ispipe = 1;
1277
1278         /* Repeat as long as we have more pattern to process and more output
1279            space */
1280         while (*pat_ptr) {
1281                 if (*pat_ptr != '%') {
1282                         if (out_ptr == out_end)
1283                                 goto out;
1284                         *out_ptr++ = *pat_ptr++;
1285                 } else {
1286                         switch (*++pat_ptr) {
1287                         case 0:
1288                                 goto out;
1289                         /* Double percent, output one percent */
1290                         case '%':
1291                                 if (out_ptr == out_end)
1292                                         goto out;
1293                                 *out_ptr++ = '%';
1294                                 break;
1295                         /* pid */
1296                         case 'p':
1297                                 pid_in_pattern = 1;
1298                                 rc = snprintf(out_ptr, out_end - out_ptr,
1299                                               "%d", current->tgid);
1300                                 if (rc > out_end - out_ptr)
1301                                         goto out;
1302                                 out_ptr += rc;
1303                                 break;
1304                         /* uid */
1305                         case 'u':
1306                                 rc = snprintf(out_ptr, out_end - out_ptr,
1307                                               "%d", current->uid);
1308                                 if (rc > out_end - out_ptr)
1309                                         goto out;
1310                                 out_ptr += rc;
1311                                 break;
1312                         /* gid */
1313                         case 'g':
1314                                 rc = snprintf(out_ptr, out_end - out_ptr,
1315                                               "%d", current->gid);
1316                                 if (rc > out_end - out_ptr)
1317                                         goto out;
1318                                 out_ptr += rc;
1319                                 break;
1320                         /* signal that caused the coredump */
1321                         case 's':
1322                                 rc = snprintf(out_ptr, out_end - out_ptr,
1323                                               "%ld", signr);
1324                                 if (rc > out_end - out_ptr)
1325                                         goto out;
1326                                 out_ptr += rc;
1327                                 break;
1328                         /* UNIX time of coredump */
1329                         case 't': {
1330                                 struct timeval tv;
1331                                 do_gettimeofday(&tv);
1332                                 rc = snprintf(out_ptr, out_end - out_ptr,
1333                                               "%lu", tv.tv_sec);
1334                                 if (rc > out_end - out_ptr)
1335                                         goto out;
1336                                 out_ptr += rc;
1337                                 break;
1338                         }
1339                         /* hostname */
1340                         case 'h':
1341                                 down_read(&uts_sem);
1342                                 rc = snprintf(out_ptr, out_end - out_ptr,
1343                                               "%s", utsname()->nodename);
1344                                 up_read(&uts_sem);
1345                                 if (rc > out_end - out_ptr)
1346                                         goto out;
1347                                 out_ptr += rc;
1348                                 break;
1349                         /* executable */
1350                         case 'e':
1351                                 rc = snprintf(out_ptr, out_end - out_ptr,
1352                                               "%s", current->comm);
1353                                 if (rc > out_end - out_ptr)
1354                                         goto out;
1355                                 out_ptr += rc;
1356                                 break;
1357                         default:
1358                                 break;
1359                         }
1360                         ++pat_ptr;
1361                 }
1362         }
1363         /* Backward compatibility with core_uses_pid:
1364          *
1365          * If core_pattern does not include a %p (as is the default)
1366          * and core_uses_pid is set, then .%pid will be appended to
1367          * the filename. Do not do this for piped commands. */
1368         if (!ispipe && !pid_in_pattern
1369             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1370                 rc = snprintf(out_ptr, out_end - out_ptr,
1371                               ".%d", current->tgid);
1372                 if (rc > out_end - out_ptr)
1373                         goto out;
1374                 out_ptr += rc;
1375         }
1376 out:
1377         *out_ptr = 0;
1378         return ispipe;
1379 }
1380
1381 static void zap_process(struct task_struct *start)
1382 {
1383         struct task_struct *t;
1384
1385         start->signal->flags = SIGNAL_GROUP_EXIT;
1386         start->signal->group_stop_count = 0;
1387
1388         t = start;
1389         do {
1390                 if (t != current && t->mm) {
1391                         t->mm->core_waiters++;
1392                         sigaddset(&t->pending.signal, SIGKILL);
1393                         signal_wake_up(t, 1);
1394                 }
1395         } while ((t = next_thread(t)) != start);
1396 }
1397
1398 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1399                                 int exit_code)
1400 {
1401         struct task_struct *g, *p;
1402         unsigned long flags;
1403         int err = -EAGAIN;
1404
1405         spin_lock_irq(&tsk->sighand->siglock);
1406         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1407                 tsk->signal->group_exit_code = exit_code;
1408                 zap_process(tsk);
1409                 err = 0;
1410         }
1411         spin_unlock_irq(&tsk->sighand->siglock);
1412         if (err)
1413                 return err;
1414
1415         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1416                 goto done;
1417
1418         rcu_read_lock();
1419         for_each_process(g) {
1420                 if (g == tsk->group_leader)
1421                         continue;
1422
1423                 p = g;
1424                 do {
1425                         if (p->mm) {
1426                                 if (p->mm == mm) {
1427                                         /*
1428                                          * p->sighand can't disappear, but
1429                                          * may be changed by de_thread()
1430                                          */
1431                                         lock_task_sighand(p, &flags);
1432                                         zap_process(p);
1433                                         unlock_task_sighand(p, &flags);
1434                                 }
1435                                 break;
1436                         }
1437                 } while ((p = next_thread(p)) != g);
1438         }
1439         rcu_read_unlock();
1440 done:
1441         return mm->core_waiters;
1442 }
1443
1444 static int coredump_wait(int exit_code)
1445 {
1446         struct task_struct *tsk = current;
1447         struct mm_struct *mm = tsk->mm;
1448         struct completion startup_done;
1449         struct completion *vfork_done;
1450         int core_waiters;
1451
1452         init_completion(&mm->core_done);
1453         init_completion(&startup_done);
1454         mm->core_startup_done = &startup_done;
1455
1456         core_waiters = zap_threads(tsk, mm, exit_code);
1457         up_write(&mm->mmap_sem);
1458
1459         if (unlikely(core_waiters < 0))
1460                 goto fail;
1461
1462         /*
1463          * Make sure nobody is waiting for us to release the VM,
1464          * otherwise we can deadlock when we wait on each other
1465          */
1466         vfork_done = tsk->vfork_done;
1467         if (vfork_done) {
1468                 tsk->vfork_done = NULL;
1469                 complete(vfork_done);
1470         }
1471
1472         if (core_waiters)
1473                 wait_for_completion(&startup_done);
1474 fail:
1475         BUG_ON(mm->core_waiters);
1476         return core_waiters;
1477 }
1478
1479 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1480 {
1481         char corename[CORENAME_MAX_SIZE + 1];
1482         struct mm_struct *mm = current->mm;
1483         struct linux_binfmt * binfmt;
1484         struct inode * inode;
1485         struct file * file;
1486         int retval = 0;
1487         int fsuid = current->fsuid;
1488         int flag = 0;
1489         int ispipe = 0;
1490
1491         audit_core_dumps(signr);
1492
1493         binfmt = current->binfmt;
1494         if (!binfmt || !binfmt->core_dump)
1495                 goto fail;
1496         down_write(&mm->mmap_sem);
1497         if (!mm->dumpable) {
1498                 up_write(&mm->mmap_sem);
1499                 goto fail;
1500         }
1501
1502         /*
1503          *      We cannot trust fsuid as being the "true" uid of the
1504          *      process nor do we know its entire history. We only know it
1505          *      was tainted so we dump it as root in mode 2.
1506          */
1507         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1508                 flag = O_EXCL;          /* Stop rewrite attacks */
1509                 current->fsuid = 0;     /* Dump root private */
1510         }
1511         mm->dumpable = 0;
1512
1513         retval = coredump_wait(exit_code);
1514         if (retval < 0)
1515                 goto fail;
1516
1517         /*
1518          * Clear any false indication of pending signals that might
1519          * be seen by the filesystem code called to write the core file.
1520          */
1521         clear_thread_flag(TIF_SIGPENDING);
1522
1523         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1524                 goto fail_unlock;
1525
1526         /*
1527          * lock_kernel() because format_corename() is controlled by sysctl, which
1528          * uses lock_kernel()
1529          */
1530         lock_kernel();
1531         ispipe = format_corename(corename, core_pattern, signr);
1532         unlock_kernel();
1533         if (ispipe) {
1534                 /* SIGPIPE can happen, but it's just never processed */
1535                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1536                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1537                                corename);
1538                         goto fail_unlock;
1539                 }
1540         } else
1541                 file = filp_open(corename,
1542                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1543                                  0600);
1544         if (IS_ERR(file))
1545                 goto fail_unlock;
1546         inode = file->f_path.dentry->d_inode;
1547         if (inode->i_nlink > 1)
1548                 goto close_fail;        /* multiple links - don't dump */
1549         if (!ispipe && d_unhashed(file->f_path.dentry))
1550                 goto close_fail;
1551
1552         /* AK: actually i see no reason to not allow this for named pipes etc.,
1553            but keep the previous behaviour for now. */
1554         if (!ispipe && !S_ISREG(inode->i_mode))
1555                 goto close_fail;
1556         if (!file->f_op)
1557                 goto close_fail;
1558         if (!file->f_op->write)
1559                 goto close_fail;
1560         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1561                 goto close_fail;
1562
1563         retval = binfmt->core_dump(signr, regs, file);
1564
1565         if (retval)
1566                 current->signal->group_exit_code |= 0x80;
1567 close_fail:
1568         filp_close(file, NULL);
1569 fail_unlock:
1570         current->fsuid = fsuid;
1571         complete_all(&mm->core_done);
1572 fail:
1573         return retval;
1574 }