fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/config.h>
  26 #include <linux/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/mman.h>
  29 #include <linux/a.out.h>
  30 #include <linux/stat.h>
  31 #include <linux/fcntl.h>
  32 #include <linux/smp_lock.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/highmem.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/key.h>
  38 #include <linux/personality.h>
  39 #include <linux/binfmts.h>
  40 #include <linux/swap.h>
  41 #include <linux/utsname.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/acct.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/vs_memory.h>
  53 #include <linux/vs_cvirt.h>
  54
  55 #include <asm/uaccess.h>
  56 #include <asm/mmu_context.h>
  57
  58 #ifdef CONFIG_KMOD
  59 #include <linux/kmod.h>
  60 #endif
  61
  62 int core_uses_pid;
  63 char core_pattern[65] = "core";
  64 int suid_dumpable = 0;
  65
  66 EXPORT_SYMBOL(suid_dumpable);
  67 /* The maximal length of core_pattern is also specified in sysctl.c */
  68
  69 static struct linux_binfmt *formats;
  70 static DEFINE_RWLOCK(binfmt_lock);
  71
  72 int register_binfmt(struct linux_binfmt * fmt)
  73 {
  74         struct linux_binfmt ** tmp = &formats;
  75
  76         if (!fmt)
  77                 return -EINVAL;
  78         if (fmt->next)
  79                 return -EBUSY;
  80         write_lock(&binfmt_lock);
  81         while (*tmp) {
  82                 if (fmt == *tmp) {
  83                         write_unlock(&binfmt_lock);
  84                         return -EBUSY;
  85                 }
  86                 tmp = &(*tmp)->next;
  87         }
  88         fmt->next = formats;
  89         formats = fmt;
  90         write_unlock(&binfmt_lock);
  91         return 0;
  92 }
  93
  94 EXPORT_SYMBOL(register_binfmt);
  95
  96 int unregister_binfmt(struct linux_binfmt * fmt)
  97 {
  98         struct linux_binfmt ** tmp = &formats;
  99
 100         write_lock(&binfmt_lock);
 101         while (*tmp) {
 102                 if (fmt == *tmp) {
 103                         *tmp = fmt->next;
 104                         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);
 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_alloc(vm_area_cachep, SLAB_KERNEL);
 410         if (!mpnt)
 411                 return -ENOMEM;
 412
 413         memset(mpnt, 0, sizeof(*mpnt));
 414
 415         down_write(&mm->mmap_sem);
 416         {
 417                 mpnt->vm_mm = mm;
 418 #ifdef CONFIG_STACK_GROWSUP
 419                 mpnt->vm_start = stack_base;
 420                 mpnt->vm_end = stack_base + arg_size;
 421 #else
 422                 mpnt->vm_end = stack_top;
 423                 mpnt->vm_start = mpnt->vm_end - arg_size;
 424 #endif
 425                 /* Adjust stack execute permissions; explicitly enable
 426                  * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
 427                  * and leave alone (arch default) otherwise. */
 428                 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 429                         mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
 430                 else if (executable_stack == EXSTACK_DISABLE_X)
 431                         mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
 432                 else
 433                         mpnt->vm_flags = VM_STACK_FLAGS;
 434                 mpnt->vm_flags |= mm->def_flags;
 435                 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
 436                 if ((ret = insert_vm_struct(mm, mpnt))) {
 437                         up_write(&mm->mmap_sem);
 438                         kmem_cache_free(vm_area_cachep, mpnt);
 439                         return ret;
 440                 }
 441                 vx_vmpages_sub(mm, mm->total_vm - vma_pages(mpnt));
 442                 mm->stack_vm = mm->total_vm;
 443         }
 444
 445         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
 446                 struct page *page = bprm->page[i];
 447                 if (page) {
 448                         bprm->page[i] = NULL;
 449                         install_arg_page(mpnt, page, stack_base);
 450                 }
 451                 stack_base += PAGE_SIZE;
 452         }
 453         up_write(&mm->mmap_sem);
 454
 455         return 0;
 456 }
 457
 458 EXPORT_SYMBOL(setup_arg_pages);
 459
 460 #define free_arg_pages(bprm) do { } while (0)
 461
 462 #else
 463
 464 static inline void free_arg_pages(struct linux_binprm *bprm)
 465 {
 466         int i;
 467
 468         for (i = 0; i < MAX_ARG_PAGES; i++) {
 469                 if (bprm->page[i])
 470                         __free_page(bprm->page[i]);
 471                 bprm->page[i] = NULL;
 472         }
 473 }
 474
 475 #endif /* CONFIG_MMU */
 476
 477 struct file *open_exec(const char *name)
 478 {
 479         struct nameidata nd;
 480         int err;
 481         struct file *file;
 482
 483         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ);
 484         file = ERR_PTR(err);
 485
 486         if (!err) {
 487                 struct inode *inode = nd.dentry->d_inode;
 488                 file = ERR_PTR(-EACCES);
 489                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 490                     S_ISREG(inode->i_mode)) {
 491                         int err = vfs_permission(&nd, MAY_EXEC);
 492                         if (!err && !(inode->i_mode & 0111))
 493                                 err = -EACCES;
 494                         file = ERR_PTR(err);
 495                         if (!err) {
 496                                 file = nameidata_to_filp(&nd, O_RDONLY);
 497                                 if (!IS_ERR(file)) {
 498                                         err = deny_write_access(file);
 499                                         if (err) {
 500                                                 fput(file);
 501                                                 file = ERR_PTR(err);
 502                                         }
 503                                 }
 504 out:
 505                                 return file;
 506                         }
 507                 }
 508                 release_open_intent(&nd);
 509                 path_release(&nd);
 510         }
 511         goto out;
 512 }
 513
 514 EXPORT_SYMBOL(open_exec);
 515
 516 int kernel_read(struct file *file, unsigned long offset,
 517         char *addr, unsigned long count)
 518 {
 519         mm_segment_t old_fs;
 520         loff_t pos = offset;
 521         int result;
 522
 523         old_fs = get_fs();
 524         set_fs(get_ds());
 525         /* The cast to a user pointer is valid due to the set_fs() */
 526         result = vfs_read(file, (void __user *)addr, count, &pos);
 527         set_fs(old_fs);
 528         return result;
 529 }
 530
 531 EXPORT_SYMBOL(kernel_read);
 532
 533 static int exec_mmap(struct mm_struct *mm)
 534 {
 535         struct task_struct *tsk;
 536         struct mm_struct * old_mm, *active_mm;
 537
 538         /* Notify parent that we're no longer interested in the old VM */
 539         tsk = current;
 540         old_mm = current->mm;
 541         mm_release(tsk, old_mm);
 542
 543         if (old_mm) {
 544                 /*
 545                  * Make sure that if there is a core dump in progress
 546                  * for the old mm, we get out and die instead of going
 547                  * through with the exec.  We must hold mmap_sem around
 548                  * checking core_waiters and changing tsk->mm.  The
 549                  * core-inducing thread will increment core_waiters for
 550                  * each thread whose ->mm == old_mm.
 551                  */
 552                 down_read(&old_mm->mmap_sem);
 553                 if (unlikely(old_mm->core_waiters)) {
 554                         up_read(&old_mm->mmap_sem);
 555                         return -EINTR;
 556                 }
 557         }
 558         task_lock(tsk);
 559         active_mm = tsk->active_mm;
 560         tsk->mm = mm;
 561         tsk->active_mm = mm;
 562         activate_mm(active_mm, mm);
 563         task_unlock(tsk);
 564         arch_pick_mmap_layout(mm);
 565         if (old_mm) {
 566                 up_read(&old_mm->mmap_sem);
 567                 if (active_mm != old_mm) BUG();
 568                 mmput(old_mm);
 569                 return 0;
 570         }
 571         mmdrop(active_mm);
 572         return 0;
 573 }
 574
 575 /*
 576  * This function makes sure the current process has its own signal table,
 577  * so that flush_signal_handlers can later reset the handlers without
 578  * disturbing other processes.  (Other processes might share the signal
 579  * table via the CLONE_SIGHAND option to clone().)
 580  */
 581 static int de_thread(struct task_struct *tsk)
 582 {
 583         struct signal_struct *sig = tsk->signal;
 584         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 585         spinlock_t *lock = &oldsighand->siglock;
 586         struct task_struct *leader = NULL;
 587         int count;
 588
 589         /*
 590          * If we don't share sighandlers, then we aren't sharing anything
 591          * and we can just re-use it all.
 592          */
 593         if (atomic_read(&oldsighand->count) <= 1) {
 594                 BUG_ON(atomic_read(&sig->count) != 1);
 595                 exit_itimers(sig);
 596                 return 0;
 597         }
 598
 599         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 600         if (!newsighand)
 601                 return -ENOMEM;
 602
 603         if (thread_group_empty(current))
 604                 goto no_thread_group;
 605
 606         /*
 607          * Kill all other threads in the thread group.
 608          * We must hold tasklist_lock to call zap_other_threads.
 609          */
 610         read_lock(&tasklist_lock);
 611         spin_lock_irq(lock);
 612         if (sig->flags & SIGNAL_GROUP_EXIT) {
 613                 /*
 614                  * Another group action in progress, just
 615                  * return so that the signal is processed.
 616                  */
 617                 spin_unlock_irq(lock);
 618                 read_unlock(&tasklist_lock);
 619                 kmem_cache_free(sighand_cachep, newsighand);
 620                 return -EAGAIN;
 621         }
 622         zap_other_threads(current);
 623         read_unlock(&tasklist_lock);
 624
 625         /*
 626          * Account for the thread group leader hanging around:
 627          */
 628         count = 1;
 629         if (!thread_group_leader(current)) {
 630                 count = 2;
 631                 /*
 632                  * The SIGALRM timer survives the exec, but needs to point
 633                  * at us as the new group leader now.  We have a race with
 634                  * a timer firing now getting the old leader, so we need to
 635                  * synchronize with any firing (by calling del_timer_sync)
 636                  * before we can safely let the old group leader die.
 637                  */
 638                 sig->real_timer.data = current;
 639                 spin_unlock_irq(lock);
 640                 if (hrtimer_cancel(&sig->real_timer))
 641                         hrtimer_restart(&sig->real_timer);
 642                 spin_lock_irq(lock);
 643         }
 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 (!thread_group_leader(current)) {
 662                 struct task_struct *parent;
 663                 struct dentry *proc_dentry1, *proc_dentry2;
 664                 unsigned long 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                 leader = current->group_leader;
 672                 while (leader->exit_state != EXIT_ZOMBIE)
 673                         yield();
 674
 675                 spin_lock(&leader->proc_lock);
 676                 spin_lock(&current->proc_lock);
 677                 proc_dentry1 = proc_pid_unhash(current);
 678                 proc_dentry2 = proc_pid_unhash(leader);
 679                 write_lock_irq(&tasklist_lock);
 680
 681                 BUG_ON(leader->tgid != current->tgid);
 682                 BUG_ON(current->pid == current->tgid);
 683                 /*
 684                  * An exec() starts a new thread group with the
 685                  * TGID of the previous thread group. Rehash the
 686                  * two threads with a switched PID, and release
 687                  * the former thread group leader:
 688                  */
 689                 ptrace = leader->ptrace;
 690                 parent = leader->parent;
 691                 if (unlikely(ptrace) && unlikely(parent == current)) {
 692                         /*
 693                          * Joker was ptracing his own group leader,
 694                          * and now he wants to be his own parent!
 695                          * We can't have that.
 696                          */
 697                         ptrace = 0;
 698                 }
 699
 700                 ptrace_unlink(current);
 701                 ptrace_unlink(leader);
 702                 remove_parent(current);
 703                 remove_parent(leader);
 704
 705                 switch_exec_pids(leader, current);
 706
 707                 current->parent = current->real_parent = leader->real_parent;
 708                 leader->parent = leader->real_parent = child_reaper;
 709                 current->group_leader = current;
 710                 leader->group_leader = leader;
 711
 712                 add_parent(current, current->parent);
 713                 add_parent(leader, leader->parent);
 714                 if (ptrace) {
 715                         current->ptrace = ptrace;
 716                         __ptrace_link(current, parent);
 717                 }
 718
 719                 list_del(&current->tasks);
 720                 list_add_tail(&current->tasks, &init_task.tasks);
 721                 current->exit_signal = SIGCHLD;
 722
 723                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 724                 leader->exit_state = EXIT_DEAD;
 725
 726                 write_unlock_irq(&tasklist_lock);
 727                 spin_unlock(&leader->proc_lock);
 728                 spin_unlock(&current->proc_lock);
 729                 proc_pid_flush(proc_dentry1);
 730                 proc_pid_flush(proc_dentry2);
 731         }
 732
 733         /*
 734          * There may be one thread left which is just exiting,
 735          * but it's safe to stop telling the group to kill themselves.
 736          */
 737         sig->flags = 0;
 738
 739 no_thread_group:
 740         exit_itimers(sig);
 741         if (leader)
 742                 release_task(leader);
 743
 744         BUG_ON(atomic_read(&sig->count) != 1);
 745
 746         if (atomic_read(&oldsighand->count) == 1) {
 747                 /*
 748                  * Now that we nuked the rest of the thread group,
 749                  * it turns out we are not sharing sighand any more either.
 750                  * So we can just keep it.
 751                  */
 752                 kmem_cache_free(sighand_cachep, newsighand);
 753         } else {
 754                 /*
 755                  * Move our state over to newsighand and switch it in.
 756                  */
 757                 spin_lock_init(&newsighand->siglock);
 758                 atomic_set(&newsighand->count, 1);
 759                 memcpy(newsighand->action, oldsighand->action,
 760                        sizeof(newsighand->action));
 761
 762                 write_lock_irq(&tasklist_lock);
 763                 spin_lock(&oldsighand->siglock);
 764                 spin_lock(&newsighand->siglock);
 765
 766                 rcu_assign_pointer(current->sighand, newsighand);
 767                 recalc_sigpending();
 768
 769                 spin_unlock(&newsighand->siglock);
 770                 spin_unlock(&oldsighand->siglock);
 771                 write_unlock_irq(&tasklist_lock);
 772
 773                 if (atomic_dec_and_test(&oldsighand->count))
 774                         sighand_free(oldsighand);
 775         }
 776
 777         BUG_ON(!thread_group_leader(current));
 778         return 0;
 779 }
 780
 781 /*
 782  * These functions flushes out all traces of the currently running executable
 783  * so that a new one can be started
 784  */
 785
 786 static void flush_old_files(struct files_struct * files)
 787 {
 788         long j = -1;
 789         struct fdtable *fdt;
 790
 791         spin_lock(&files->file_lock);
 792         for (;;) {
 793                 unsigned long set, i;
 794
 795                 j++;
 796                 i = j * __NFDBITS;
 797                 fdt = files_fdtable(files);
 798                 if (i >= fdt->max_fds || i >= fdt->max_fdset)
 799                         break;
 800                 set = fdt->close_on_exec->fds_bits[j];
 801                 if (!set)
 802                         continue;
 803                 fdt->close_on_exec->fds_bits[j] = 0;
 804                 spin_unlock(&files->file_lock);
 805                 for ( ; set ; i++,set >>= 1) {
 806                         if (set & 1) {
 807                                 sys_close(i);
 808                         }
 809                 }
 810                 spin_lock(&files->file_lock);
 811
 812         }
 813         spin_unlock(&files->file_lock);
 814 }
 815
 816 void get_task_comm(char *buf, struct task_struct *tsk)
 817 {
 818         /* buf must be at least sizeof(tsk->comm) in size */
 819         task_lock(tsk);
 820         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 821         task_unlock(tsk);
 822 }
 823
 824 void set_task_comm(struct task_struct *tsk, char *buf)
 825 {
 826         task_lock(tsk);
 827         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 828         task_unlock(tsk);
 829 }
 830
 831 int flush_old_exec(struct linux_binprm * bprm)
 832 {
 833         char * name;
 834         int i, ch, retval;
 835         struct files_struct *files;
 836         char tcomm[sizeof(current->comm)];
 837
 838         /*
 839          * Make sure we have a private signal table and that
 840          * we are unassociated from the previous thread group.
 841          */
 842         retval = de_thread(current);
 843         if (retval)
 844                 goto out;
 845
 846         /*
 847          * Make sure we have private file handles. Ask the
 848          * fork helper to do the work for us and the exit
 849          * helper to do the cleanup of the old one.
 850          */
 851         files = current->files;         /* refcounted so safe to hold */
 852         retval = unshare_files();
 853         if (retval)
 854                 goto out;
 855         /*
 856          * Release all of the old mmap stuff
 857          */
 858         retval = exec_mmap(bprm->mm);
 859         if (retval)
 860                 goto mmap_failed;
 861
 862         bprm->mm = NULL;                /* We're using it now */
 863
 864         /* This is the point of no return */
 865         steal_locks(files);
 866         put_files_struct(files);
 867
 868         current->sas_ss_sp = current->sas_ss_size = 0;
 869
 870         if (current->euid == current->uid && current->egid == current->gid)
 871                 current->mm->dumpable = 1;
 872         else
 873                 current->mm->dumpable = suid_dumpable;
 874
 875         name = bprm->filename;
 876
 877         /* Copies the binary name from after last slash */
 878         for (i=0; (ch = *(name++)) != '\0';) {
 879                 if (ch == '/')
 880                         i = 0; /* overwrite what we wrote */
 881                 else
 882                         if (i < (sizeof(tcomm) - 1))
 883                                 tcomm[i++] = ch;
 884         }
 885         tcomm[i] = '\0';
 886         set_task_comm(current, tcomm);
 887
 888         current->flags &= ~PF_RANDOMIZE;
 889         flush_thread();
 890
 891         /* Set the new mm task size. We have to do that late because it may
 892          * depend on TIF_32BIT which is only updated in flush_thread() on
 893          * some architectures like powerpc
 894          */
 895         current->mm->task_size = TASK_SIZE;
 896
 897         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
 898             file_permission(bprm->file, MAY_READ) ||
 899             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
 900                 suid_keys(current);
 901                 current->mm->dumpable = suid_dumpable;
 902         }
 903
 904         /* An exec changes our domain. We are no longer part of the thread
 905            group */
 906
 907         current->self_exec_id++;
 908
 909         flush_signal_handlers(current, 0);
 910         flush_old_files(current->files);
 911
 912         return 0;
 913
 914 mmap_failed:
 915         put_files_struct(current->files);
 916         current->files = files;
 917 out:
 918         return retval;
 919 }
 920
 921 EXPORT_SYMBOL(flush_old_exec);
 922
 923 /*
 924  * Fill the binprm structure from the inode.
 925  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
 926  */
 927 int prepare_binprm(struct linux_binprm *bprm)
 928 {
 929         int mode;
 930         struct inode * inode = bprm->file->f_dentry->d_inode;
 931         int retval;
 932
 933         mode = inode->i_mode;
 934         /*
 935          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
 936          * generic_permission lets a non-executable through
 937          */
 938         if (!(mode & 0111))     /* with at least _one_ execute bit set */
 939                 return -EACCES;
 940         if (bprm->file->f_op == NULL)
 941                 return -EACCES;
 942
 943         bprm->e_uid = current->euid;
 944         bprm->e_gid = current->egid;
 945
 946         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
 947                 /* Set-uid? */
 948                 if (mode & S_ISUID) {
 949                         current->personality &= ~PER_CLEAR_ON_SETID;
 950                         bprm->e_uid = inode->i_uid;
 951                 }
 952
 953                 /* Set-gid? */
 954                 /*
 955                  * If setgid is set but no group execute bit then this
 956                  * is a candidate for mandatory locking, not a setgid
 957                  * executable.
 958                  */
 959                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
 960                         current->personality &= ~PER_CLEAR_ON_SETID;
 961                         bprm->e_gid = inode->i_gid;
 962                 }
 963         }
 964
 965         /* fill in binprm security blob */
 966         retval = security_bprm_set(bprm);
 967         if (retval)
 968                 return retval;
 969
 970         memset(bprm->buf,0,BINPRM_BUF_SIZE);
 971         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
 972 }
 973
 974 EXPORT_SYMBOL(prepare_binprm);
 975
 976 static int unsafe_exec(struct task_struct *p)
 977 {
 978         int unsafe = 0;
 979         if (p->ptrace & PT_PTRACED) {
 980                 if (p->ptrace & PT_PTRACE_CAP)
 981                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
 982                 else
 983                         unsafe |= LSM_UNSAFE_PTRACE;
 984         }
 985         if (atomic_read(&p->fs->count) > 1 ||
 986             atomic_read(&p->files->count) > 1 ||
 987             atomic_read(&p->sighand->count) > 1)
 988                 unsafe |= LSM_UNSAFE_SHARE;
 989
 990         return unsafe;
 991 }
 992
 993 void compute_creds(struct linux_binprm *bprm)
 994 {
 995         int unsafe;
 996
 997         if (bprm->e_uid != current->uid)
 998                 suid_keys(current);
 999         exec_keys(current);
1000
1001         task_lock(current);
1002         unsafe = unsafe_exec(current);
1003         security_bprm_apply_creds(bprm, unsafe);
1004         task_unlock(current);
1005         security_bprm_post_apply_creds(bprm);
1006 }
1007
1008 EXPORT_SYMBOL(compute_creds);
1009
1010 void remove_arg_zero(struct linux_binprm *bprm)
1011 {
1012         if (bprm->argc) {
1013                 unsigned long offset;
1014                 char * kaddr;
1015                 struct page *page;
1016
1017                 offset = bprm->p % PAGE_SIZE;
1018                 goto inside;
1019
1020                 while (bprm->p++, *(kaddr+offset++)) {
1021                         if (offset != PAGE_SIZE)
1022                                 continue;
1023                         offset = 0;
1024                         kunmap_atomic(kaddr, KM_USER0);
1025 inside:
1026                         page = bprm->page[bprm->p/PAGE_SIZE];
1027                         kaddr = kmap_atomic(page, KM_USER0);
1028                 }
1029                 kunmap_atomic(kaddr, KM_USER0);
1030                 bprm->argc--;
1031         }
1032 }
1033
1034 EXPORT_SYMBOL(remove_arg_zero);
1035
1036 /*
1037  * cycle the list of binary formats handler, until one recognizes the image
1038  */
1039 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1040 {
1041         int try,retval;
1042         struct linux_binfmt *fmt;
1043 #ifdef __alpha__
1044         /* handle /sbin/loader.. */
1045         {
1046             struct exec * eh = (struct exec *) bprm->buf;
1047
1048             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1049                 (eh->fh.f_flags & 0x3000) == 0x3000)
1050             {
1051                 struct file * file;
1052                 unsigned long loader;
1053
1054                 allow_write_access(bprm->file);
1055                 fput(bprm->file);
1056                 bprm->file = NULL;
1057
1058                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1059
1060                 file = open_exec("/sbin/loader");
1061                 retval = PTR_ERR(file);
1062                 if (IS_ERR(file))
1063                         return retval;
1064
1065                 /* Remember if the application is TASO.  */
1066                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1067
1068                 bprm->file = file;
1069                 bprm->loader = loader;
1070                 retval = prepare_binprm(bprm);
1071                 if (retval<0)
1072                         return retval;
1073                 /* should call search_binary_handler recursively here,
1074                    but it does not matter */
1075             }
1076         }
1077 #endif
1078         retval = security_bprm_check(bprm);
1079         if (retval)
1080                 return retval;
1081
1082         /* kernel module loader fixup */
1083         /* so we don't try to load run modprobe in kernel space. */
1084         set_fs(USER_DS);
1085         retval = -ENOENT;
1086         for (try=0; try<2; try++) {
1087                 read_lock(&binfmt_lock);
1088                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1089                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1090                         if (!fn)
1091                                 continue;
1092                         if (!try_module_get(fmt->module))
1093                                 continue;
1094                         read_unlock(&binfmt_lock);
1095                         retval = fn(bprm, regs);
1096                         if (retval >= 0) {
1097                                 put_binfmt(fmt);
1098                                 allow_write_access(bprm->file);
1099                                 if (bprm->file)
1100                                         fput(bprm->file);
1101                                 bprm->file = NULL;
1102                                 current->did_exec = 1;
1103                                 proc_exec_connector(current);
1104                                 return retval;
1105                         }
1106                         read_lock(&binfmt_lock);
1107                         put_binfmt(fmt);
1108                         if (retval != -ENOEXEC || bprm->mm == NULL)
1109                                 break;
1110                         if (!bprm->file) {
1111                                 read_unlock(&binfmt_lock);
1112                                 return retval;
1113                         }
1114                 }
1115                 read_unlock(&binfmt_lock);
1116                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1117                         break;
1118 #ifdef CONFIG_KMOD
1119                 }else{
1120 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1121                         if (printable(bprm->buf[0]) &&
1122                             printable(bprm->buf[1]) &&
1123                             printable(bprm->buf[2]) &&
1124                             printable(bprm->buf[3]))
1125                                 break; /* -ENOEXEC */
1126                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1127 #endif
1128                 }
1129         }
1130         return retval;
1131 }
1132
1133 EXPORT_SYMBOL(search_binary_handler);
1134
1135 /*
1136  * sys_execve() executes a new program.
1137  */
1138 int do_execve(char * filename,
1139         char __user *__user *argv,
1140         char __user *__user *envp,
1141         struct pt_regs * regs)
1142 {
1143         struct linux_binprm *bprm;
1144         struct file *file;
1145         int retval;
1146         int i;
1147
1148         retval = -ENOMEM;
1149         bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1150         if (!bprm)
1151                 goto out_ret;
1152         memset(bprm, 0, sizeof(*bprm));
1153
1154         file = open_exec(filename);
1155         retval = PTR_ERR(file);
1156         if (IS_ERR(file))
1157                 goto out_kfree;
1158
1159         sched_exec();
1160
1161         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1162
1163         bprm->file = file;
1164         bprm->filename = filename;
1165         bprm->interp = filename;
1166         bprm->mm = mm_alloc();
1167         retval = -ENOMEM;
1168         if (!bprm->mm)
1169                 goto out_file;
1170
1171         retval = init_new_context(current, bprm->mm);
1172         if (retval < 0)
1173                 goto out_mm;
1174
1175         bprm->argc = count(argv, bprm->p / sizeof(void *));
1176         if ((retval = bprm->argc) < 0)
1177                 goto out_mm;
1178
1179         bprm->envc = count(envp, bprm->p / sizeof(void *));
1180         if ((retval = bprm->envc) < 0)
1181                 goto out_mm;
1182
1183         retval = security_bprm_alloc(bprm);
1184         if (retval)
1185                 goto out;
1186
1187         retval = prepare_binprm(bprm);
1188         if (retval < 0)
1189                 goto out;
1190
1191         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1192         if (retval < 0)
1193                 goto out;
1194
1195         bprm->exec = bprm->p;
1196         retval = copy_strings(bprm->envc, envp, bprm);
1197         if (retval < 0)
1198                 goto out;
1199
1200         retval = copy_strings(bprm->argc, argv, bprm);
1201         if (retval < 0)
1202                 goto out;
1203
1204         retval = search_binary_handler(bprm,regs);
1205         if (retval >= 0) {
1206                 free_arg_pages(bprm);
1207
1208                 /* execve success */
1209                 security_bprm_free(bprm);
1210                 acct_update_integrals(current);
1211                 kfree(bprm);
1212                 return retval;
1213         }
1214
1215 out:
1216         /* Something went wrong, return the inode and free the argument pages*/
1217         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1218                 struct page * page = bprm->page[i];
1219                 if (page)
1220                         __free_page(page);
1221         }
1222
1223         if (bprm->security)
1224                 security_bprm_free(bprm);
1225
1226 out_mm:
1227         if (bprm->mm)
1228                 mmdrop(bprm->mm);
1229
1230 out_file:
1231         if (bprm->file) {
1232                 allow_write_access(bprm->file);
1233                 fput(bprm->file);
1234         }
1235
1236 out_kfree:
1237         kfree(bprm);
1238
1239 out_ret:
1240         return retval;
1241 }
1242
1243 int set_binfmt(struct linux_binfmt *new)
1244 {
1245         struct linux_binfmt *old = current->binfmt;
1246
1247         if (new) {
1248                 if (!try_module_get(new->module))
1249                         return -1;
1250         }
1251         current->binfmt = new;
1252         if (old)
1253                 module_put(old->module);
1254         return 0;
1255 }
1256
1257 EXPORT_SYMBOL(set_binfmt);
1258
1259 #define CORENAME_MAX_SIZE 64
1260
1261 /* format_corename will inspect the pattern parameter, and output a
1262  * name into corename, which must have space for at least
1263  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1264  */
1265 static void format_corename(char *corename, const char *pattern, long signr)
1266 {
1267         const char *pat_ptr = pattern;
1268         char *out_ptr = corename;
1269         char *const out_end = corename + CORENAME_MAX_SIZE;
1270         int rc;
1271         int pid_in_pattern = 0;
1272
1273         /* Repeat as long as we have more pattern to process and more output
1274            space */
1275         while (*pat_ptr) {
1276                 if (*pat_ptr != '%') {
1277                         if (out_ptr == out_end)
1278                                 goto out;
1279                         *out_ptr++ = *pat_ptr++;
1280                 } else {
1281                         switch (*++pat_ptr) {
1282                         case 0:
1283                                 goto out;
1284                         /* Double percent, output one percent */
1285                         case '%':
1286                                 if (out_ptr == out_end)
1287                                         goto out;
1288                                 *out_ptr++ = '%';
1289                                 break;
1290                         /* pid */
1291                         case 'p':
1292                                 pid_in_pattern = 1;
1293                                 rc = snprintf(out_ptr, out_end - out_ptr,
1294                                               "%d", current->tgid);
1295                                 if (rc > out_end - out_ptr)
1296                                         goto out;
1297                                 out_ptr += rc;
1298                                 break;
1299                         /* uid */
1300                         case 'u':
1301                                 rc = snprintf(out_ptr, out_end - out_ptr,
1302                                               "%d", current->uid);
1303                                 if (rc > out_end - out_ptr)
1304                                         goto out;
1305                                 out_ptr += rc;
1306                                 break;
1307                         /* gid */
1308                         case 'g':
1309                                 rc = snprintf(out_ptr, out_end - out_ptr,
1310                                               "%d", current->gid);
1311                                 if (rc > out_end - out_ptr)
1312                                         goto out;
1313                                 out_ptr += rc;
1314                                 break;
1315                         /* signal that caused the coredump */
1316                         case 's':
1317                                 rc = snprintf(out_ptr, out_end - out_ptr,
1318                                               "%ld", signr);
1319                                 if (rc > out_end - out_ptr)
1320                                         goto out;
1321                                 out_ptr += rc;
1322                                 break;
1323                         /* UNIX time of coredump */
1324                         case 't': {
1325                                 struct timeval tv;
1326                                 do_gettimeofday(&tv);
1327                                 rc = snprintf(out_ptr, out_end - out_ptr,
1328                                               "%lu", tv.tv_sec);
1329                                 if (rc > out_end - out_ptr)
1330                                         goto out;
1331                                 out_ptr += rc;
1332                                 break;
1333                         }
1334                         /* hostname */
1335                         case 'h':
1336                                 down_read(&uts_sem);
1337                                 rc = snprintf(out_ptr, out_end - out_ptr,
1338                                               "%s", vx_new_uts(nodename));
1339                                 up_read(&uts_sem);
1340                                 if (rc > out_end - out_ptr)
1341                                         goto out;
1342                                 out_ptr += rc;
1343                                 break;
1344                         /* executable */
1345                         case 'e':
1346                                 rc = snprintf(out_ptr, out_end - out_ptr,
1347                                               "%s", current->comm);
1348                                 if (rc > out_end - out_ptr)
1349                                         goto out;
1350                                 out_ptr += rc;
1351                                 break;
1352                         default:
1353                                 break;
1354                         }
1355                         ++pat_ptr;
1356                 }
1357         }
1358         /* Backward compatibility with core_uses_pid:
1359          *
1360          * If core_pattern does not include a %p (as is the default)
1361          * and core_uses_pid is set, then .%pid will be appended to
1362          * the filename */
1363         if (!pid_in_pattern
1364             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1365                 rc = snprintf(out_ptr, out_end - out_ptr,
1366                               ".%d", current->tgid);
1367                 if (rc > out_end - out_ptr)
1368                         goto out;
1369                 out_ptr += rc;
1370         }
1371       out:
1372         *out_ptr = 0;
1373 }
1374
1375 static void zap_threads (struct mm_struct *mm)
1376 {
1377         struct task_struct *g, *p;
1378         struct task_struct *tsk = current;
1379         struct completion *vfork_done = tsk->vfork_done;
1380         int traced = 0;
1381
1382         /*
1383          * Make sure nobody is waiting for us to release the VM,
1384          * otherwise we can deadlock when we wait on each other
1385          */
1386         if (vfork_done) {
1387                 tsk->vfork_done = NULL;
1388                 complete(vfork_done);
1389         }
1390
1391         read_lock(&tasklist_lock);
1392         do_each_thread(g,p)
1393                 if (mm == p->mm && p != tsk) {
1394                         force_sig_specific(SIGKILL, p);
1395                         mm->core_waiters++;
1396                         if (unlikely(p->ptrace) &&
1397                             unlikely(p->parent->mm == mm))
1398                                 traced = 1;
1399                 }
1400         while_each_thread(g,p);
1401
1402         read_unlock(&tasklist_lock);
1403
1404         if (unlikely(traced)) {
1405                 /*
1406                  * We are zapping a thread and the thread it ptraces.
1407                  * If the tracee went into a ptrace stop for exit tracing,
1408                  * we could deadlock since the tracer is waiting for this
1409                  * coredump to finish.  Detach them so they can both die.
1410                  */
1411                 write_lock_irq(&tasklist_lock);
1412                 do_each_thread(g,p) {
1413                         if (mm == p->mm && p != tsk &&
1414                             p->ptrace && p->parent->mm == mm) {
1415                                 __ptrace_detach(p, 0);
1416                         }
1417                 } while_each_thread(g,p);
1418                 write_unlock_irq(&tasklist_lock);
1419         }
1420 }
1421
1422 static void coredump_wait(struct mm_struct *mm)
1423 {
1424         DECLARE_COMPLETION(startup_done);
1425         int core_waiters;
1426
1427         mm->core_startup_done = &startup_done;
1428
1429         zap_threads(mm);
1430         core_waiters = mm->core_waiters;
1431         up_write(&mm->mmap_sem);
1432
1433         if (core_waiters)
1434                 wait_for_completion(&startup_done);
1435         BUG_ON(mm->core_waiters);
1436 }
1437
1438 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1439 {
1440         char corename[CORENAME_MAX_SIZE + 1];
1441         struct mm_struct *mm = current->mm;
1442         struct linux_binfmt * binfmt;
1443         struct inode * inode;
1444         struct file * file;
1445         int retval = 0;
1446         int fsuid = current->fsuid;
1447         int flag = 0;
1448
1449         binfmt = current->binfmt;
1450         if (!binfmt || !binfmt->core_dump)
1451                 goto fail;
1452         down_write(&mm->mmap_sem);
1453         if (!mm->dumpable) {
1454                 up_write(&mm->mmap_sem);
1455                 goto fail;
1456         }
1457
1458         /*
1459          *      We cannot trust fsuid as being the "true" uid of the
1460          *      process nor do we know its entire history. We only know it
1461          *      was tainted so we dump it as root in mode 2.
1462          */
1463         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1464                 flag = O_EXCL;          /* Stop rewrite attacks */
1465                 current->fsuid = 0;     /* Dump root private */
1466         }
1467         mm->dumpable = 0;
1468
1469         retval = -EAGAIN;
1470         spin_lock_irq(&current->sighand->siglock);
1471         if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1472                 current->signal->flags = SIGNAL_GROUP_EXIT;
1473                 current->signal->group_exit_code = exit_code;
1474                 current->signal->group_stop_count = 0;
1475                 retval = 0;
1476         }
1477         spin_unlock_irq(&current->sighand->siglock);
1478         if (retval) {
1479                 up_write(&mm->mmap_sem);
1480                 goto fail;
1481         }
1482
1483         init_completion(&mm->core_done);
1484         coredump_wait(mm);
1485
1486         /*
1487          * Clear any false indication of pending signals that might
1488          * be seen by the filesystem code called to write the core file.
1489          */
1490         clear_thread_flag(TIF_SIGPENDING);
1491
1492         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1493                 goto fail_unlock;
1494
1495         /*
1496          * lock_kernel() because format_corename() is controlled by sysctl, which
1497          * uses lock_kernel()
1498          */
1499         lock_kernel();
1500         format_corename(corename, core_pattern, signr);
1501         unlock_kernel();
1502         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1503         if (IS_ERR(file))
1504                 goto fail_unlock;
1505         inode = file->f_dentry->d_inode;
1506         if (inode->i_nlink > 1)
1507                 goto close_fail;        /* multiple links - don't dump */
1508         if (d_unhashed(file->f_dentry))
1509                 goto close_fail;
1510
1511         if (!S_ISREG(inode->i_mode))
1512                 goto close_fail;
1513         if (!file->f_op)
1514                 goto close_fail;
1515         if (!file->f_op->write)
1516                 goto close_fail;
1517         if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1518                 goto close_fail;
1519
1520         retval = binfmt->core_dump(signr, regs, file);
1521
1522         if (retval)
1523                 current->signal->group_exit_code |= 0x80;
1524 close_fail:
1525         filp_close(file, NULL);
1526 fail_unlock:
1527         current->fsuid = fsuid;
1528         complete_all(&mm->core_done);
1529 fail:
1530         return retval;
1531 }