SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/exec.c

   1 /*
   2  *  linux/fs/exec.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * #!-checking implemented by tytso.
   9  */
  10 /*
  11  * Demand-loading implemented 01.12.91 - no need to read anything but
  12  * the header into memory. The inode of the executable is put into
  13  * "current->executable", and page faults do the actual loading. Clean.
  14  *
  15  * Once more I can proudly say that linux stood up to being changed: it
  16  * was less than 2 hours work to get demand-loading completely implemented.
  17  *
  18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  19  * current->executable is only used by the procfs.  This allows a dispatch
  20  * table to check for several different types  of binary formats.  We keep
  21  * trying until we recognize the file or we run out of supported binary
  22  * formats.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/file.h>
  27 #include <linux/mman.h>
  28 #include <linux/a.out.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/init.h>
  33 #include <linux/pagemap.h>
  34 #include <linux/highmem.h>
  35 #include <linux/spinlock.h>
  36 #include <linux/key.h>
  37 #include <linux/personality.h>
  38 #include <linux/binfmts.h>
  39 #include <linux/swap.h>
  40 #include <linux/utsname.h>
  41 #include <linux/pid_namespace.h>
  42 #include <linux/module.h>
  43 #include <linux/namei.h>
  44 #include <linux/proc_fs.h>
  45 #include <linux/ptrace.h>
  46 #include <linux/mount.h>
  47 #include <linux/security.h>
  48 #include <linux/syscalls.h>
  49 #include <linux/rmap.h>
  50 #include <linux/tsacct_kern.h>
  51 #include <linux/cn_proc.h>
  52 #include <linux/audit.h>
  53
  54 #include <asm/uaccess.h>
  55 #include <asm/mmu_context.h>
  56 #include <asm/tlb.h>
  57
  58 #ifdef CONFIG_KMOD
  59 #include <linux/kmod.h>
  60 #endif
  61
  62 int core_uses_pid;
  63 char core_pattern[CORENAME_MAX_SIZE] = "core";
  64 int suid_dumpable = 0;
  65
  66 EXPORT_SYMBOL(suid_dumpable);
  67 /* The maximal length of core_pattern is also specified in sysctl.c */
  68
  69 static LIST_HEAD(formats);
  70 static DEFINE_RWLOCK(binfmt_lock);
  71
  72 int register_binfmt(struct linux_binfmt * fmt)
  73 {
  74         if (!fmt)
  75                 return -EINVAL;
  76         write_lock(&binfmt_lock);
  77         list_add(&fmt->lh, &formats);
  78         write_unlock(&binfmt_lock);
  79         return 0;
  80 }
  81
  82 EXPORT_SYMBOL(register_binfmt);
  83
  84 void unregister_binfmt(struct linux_binfmt * fmt)
  85 {
  86         write_lock(&binfmt_lock);
  87         list_del(&fmt->lh);
  88         write_unlock(&binfmt_lock);
  89 }
  90
  91 EXPORT_SYMBOL(unregister_binfmt);
  92
  93 static inline void put_binfmt(struct linux_binfmt * fmt)
  94 {
  95         module_put(fmt->module);
  96 }
  97
  98 /*
  99  * Note that a shared library must be both readable and executable due to
 100  * security reasons.
 101  *
 102  * Also note that we take the address to load from from the file itself.
 103  */
 104 asmlinkage long sys_uselib(const char __user * library)
 105 {
 106         struct file * file;
 107         struct nameidata nd;
 108         int error;
 109
 110         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 111         if (error)
 112                 goto out;
 113
 114         error = -EACCES;
 115         if (nd.mnt->mnt_flags & MNT_NOEXEC)
 116                 goto exit;
 117         error = -EINVAL;
 118         if (!S_ISREG(nd.dentry->d_inode->i_mode))
 119                 goto exit;
 120
 121         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
 122         if (error)
 123                 goto exit;
 124
 125         file = nameidata_to_filp(&nd, O_RDONLY);
 126         error = PTR_ERR(file);
 127         if (IS_ERR(file))
 128                 goto out;
 129
 130         error = -ENOEXEC;
 131         if(file->f_op) {
 132                 struct linux_binfmt * fmt;
 133
 134                 read_lock(&binfmt_lock);
 135                 list_for_each_entry(fmt, &formats, lh) {
 136                         if (!fmt->load_shlib)
 137                                 continue;
 138                         if (!try_module_get(fmt->module))
 139                                 continue;
 140                         read_unlock(&binfmt_lock);
 141                         error = fmt->load_shlib(file);
 142                         read_lock(&binfmt_lock);
 143                         put_binfmt(fmt);
 144                         if (error != -ENOEXEC)
 145                                 break;
 146                 }
 147                 read_unlock(&binfmt_lock);
 148         }
 149         fput(file);
 150 out:
 151         return error;
 152 exit:
 153         release_open_intent(&nd);
 154         path_release(&nd);
 155         goto out;
 156 }
 157
 158 #ifdef CONFIG_MMU
 159
 160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 161                 int write)
 162 {
 163         struct page *page;
 164         int ret;
 165
 166 #ifdef CONFIG_STACK_GROWSUP
 167         if (write) {
 168                 ret = expand_stack_downwards(bprm->vma, pos);
 169                 if (ret < 0)
 170                         return NULL;
 171         }
 172 #endif
 173         ret = get_user_pages(current, bprm->mm, pos,
 174                         1, write, 1, &page, NULL);
 175         if (ret <= 0)
 176                 return NULL;
 177
 178         if (write) {
 179                 struct rlimit *rlim = current->signal->rlim;
 180                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 181
 182                 /*
 183                  * Limit to 1/4-th the stack size for the argv+env strings.
 184                  * This ensures that:
 185                  *  - the remaining binfmt code will not run out of stack space,
 186                  *  - the program will have a reasonable amount of stack left
 187                  *    to work from.
 188                  */
 189                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 190                         put_page(page);
 191                         return NULL;
 192                 }
 193         }
 194
 195         return page;
 196 }
 197
 198 static void put_arg_page(struct page *page)
 199 {
 200         put_page(page);
 201 }
 202
 203 static void free_arg_page(struct linux_binprm *bprm, int i)
 204 {
 205 }
 206
 207 static void free_arg_pages(struct linux_binprm *bprm)
 208 {
 209 }
 210
 211 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 212                 struct page *page)
 213 {
 214         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 215 }
 216
 217 static int __bprm_mm_init(struct linux_binprm *bprm)
 218 {
 219         int err = -ENOMEM;
 220         struct vm_area_struct *vma = NULL;
 221         struct mm_struct *mm = bprm->mm;
 222
 223         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 224         if (!vma)
 225                 goto err;
 226
 227         down_write(&mm->mmap_sem);
 228         vma->vm_mm = mm;
 229
 230         /*
 231          * Place the stack at the largest stack address the architecture
 232          * supports. Later, we'll move this to an appropriate place. We don't
 233          * use STACK_TOP because that can depend on attributes which aren't
 234          * configured yet.
 235          */
 236         vma->vm_end = STACK_TOP_MAX;
 237         vma->vm_start = vma->vm_end - PAGE_SIZE;
 238
 239         vma->vm_flags = VM_STACK_FLAGS;
 240         vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
 241         err = insert_vm_struct(mm, vma);
 242         if (err) {
 243                 up_write(&mm->mmap_sem);
 244                 goto err;
 245         }
 246
 247         mm->stack_vm = mm->total_vm = 1;
 248         up_write(&mm->mmap_sem);
 249
 250         bprm->p = vma->vm_end - sizeof(void *);
 251
 252         return 0;
 253
 254 err:
 255         if (vma) {
 256                 bprm->vma = NULL;
 257                 kmem_cache_free(vm_area_cachep, vma);
 258         }
 259
 260         return err;
 261 }
 262
 263 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 264 {
 265         return len <= MAX_ARG_STRLEN;
 266 }
 267
 268 #else
 269
 270 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 271                 int write)
 272 {
 273         struct page *page;
 274
 275         page = bprm->page[pos / PAGE_SIZE];
 276         if (!page && write) {
 277                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 278                 if (!page)
 279                         return NULL;
 280                 bprm->page[pos / PAGE_SIZE] = page;
 281         }
 282
 283         return page;
 284 }
 285
 286 static void put_arg_page(struct page *page)
 287 {
 288 }
 289
 290 static void free_arg_page(struct linux_binprm *bprm, int i)
 291 {
 292         if (bprm->page[i]) {
 293                 __free_page(bprm->page[i]);
 294                 bprm->page[i] = NULL;
 295         }
 296 }
 297
 298 static void free_arg_pages(struct linux_binprm *bprm)
 299 {
 300         int i;
 301
 302         for (i = 0; i < MAX_ARG_PAGES; i++)
 303                 free_arg_page(bprm, i);
 304 }
 305
 306 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 307                 struct page *page)
 308 {
 309 }
 310
 311 static int __bprm_mm_init(struct linux_binprm *bprm)
 312 {
 313         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 314         return 0;
 315 }
 316
 317 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 318 {
 319         return len <= bprm->p;
 320 }
 321
 322 #endif /* CONFIG_MMU */
 323
 324 /*
 325  * Create a new mm_struct and populate it with a temporary stack
 326  * vm_area_struct.  We don't have enough context at this point to set the stack
 327  * flags, permissions, and offset, so we use temporary values.  We'll update
 328  * them later in setup_arg_pages().
 329  */
 330 int bprm_mm_init(struct linux_binprm *bprm)
 331 {
 332         int err;
 333         struct mm_struct *mm = NULL;
 334
 335         bprm->mm = mm = mm_alloc();
 336         err = -ENOMEM;
 337         if (!mm)
 338                 goto err;
 339
 340         err = init_new_context(current, mm);
 341         if (err)
 342                 goto err;
 343
 344         err = __bprm_mm_init(bprm);
 345         if (err)
 346                 goto err;
 347
 348         return 0;
 349
 350 err:
 351         if (mm) {
 352                 bprm->mm = NULL;
 353                 mmdrop(mm);
 354         }
 355
 356         return err;
 357 }
 358
 359 /*
 360  * count() counts the number of strings in array ARGV.
 361  */
 362 static int count(char __user * __user * argv, int max)
 363 {
 364         int i = 0;
 365
 366         if (argv != NULL) {
 367                 for (;;) {
 368                         char __user * p;
 369
 370                         if (get_user(p, argv))
 371                                 return -EFAULT;
 372                         if (!p)
 373                                 break;
 374                         argv++;
 375                         if(++i > max)
 376                                 return -E2BIG;
 377                         cond_resched();
 378                 }
 379         }
 380         return i;
 381 }
 382
 383 /*
 384  * 'copy_strings()' copies argument/environment strings from the old
 385  * processes's memory to the new process's stack.  The call to get_user_pages()
 386  * ensures the destination page is created and not swapped out.
 387  */
 388 static int copy_strings(int argc, char __user * __user * argv,
 389                         struct linux_binprm *bprm)
 390 {
 391         struct page *kmapped_page = NULL;
 392         char *kaddr = NULL;
 393         unsigned long kpos = 0;
 394         int ret;
 395
 396         while (argc-- > 0) {
 397                 char __user *str;
 398                 int len;
 399                 unsigned long pos;
 400
 401                 if (get_user(str, argv+argc) ||
 402                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 403                         ret = -EFAULT;
 404                         goto out;
 405                 }
 406
 407                 if (!valid_arg_len(bprm, len)) {
 408                         ret = -E2BIG;
 409                         goto out;
 410                 }
 411
 412                 /* We're going to work our way backwords. */
 413                 pos = bprm->p;
 414                 str += len;
 415                 bprm->p -= len;
 416
 417                 while (len > 0) {
 418                         int offset, bytes_to_copy;
 419
 420                         offset = pos % PAGE_SIZE;
 421                         if (offset == 0)
 422                                 offset = PAGE_SIZE;
 423
 424                         bytes_to_copy = offset;
 425                         if (bytes_to_copy > len)
 426                                 bytes_to_copy = len;
 427
 428                         offset -= bytes_to_copy;
 429                         pos -= bytes_to_copy;
 430                         str -= bytes_to_copy;
 431                         len -= bytes_to_copy;
 432
 433                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 434                                 struct page *page;
 435
 436                                 page = get_arg_page(bprm, pos, 1);
 437                                 if (!page) {
 438                                         ret = -E2BIG;
 439                                         goto out;
 440                                 }
 441
 442                                 if (kmapped_page) {
 443                                         flush_kernel_dcache_page(kmapped_page);
 444                                         kunmap(kmapped_page);
 445                                         put_arg_page(kmapped_page);
 446                                 }
 447                                 kmapped_page = page;
 448                                 kaddr = kmap(kmapped_page);
 449                                 kpos = pos & PAGE_MASK;
 450                                 flush_arg_page(bprm, kpos, kmapped_page);
 451                         }
 452                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 453                                 ret = -EFAULT;
 454                                 goto out;
 455                         }
 456                 }
 457         }
 458         ret = 0;
 459 out:
 460         if (kmapped_page) {
 461                 flush_kernel_dcache_page(kmapped_page);
 462                 kunmap(kmapped_page);
 463                 put_arg_page(kmapped_page);
 464         }
 465         return ret;
 466 }
 467
 468 /*
 469  * Like copy_strings, but get argv and its values from kernel memory.
 470  */
 471 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 472 {
 473         int r;
 474         mm_segment_t oldfs = get_fs();
 475         set_fs(KERNEL_DS);
 476         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 477         set_fs(oldfs);
 478         return r;
 479 }
 480 EXPORT_SYMBOL(copy_strings_kernel);
 481
 482 #ifdef CONFIG_MMU
 483
 484 /*
 485  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 486  * the binfmt code determines where the new stack should reside, we shift it to
 487  * its final location.  The process proceeds as follows:
 488  *
 489  * 1) Use shift to calculate the new vma endpoints.
 490  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 491  *    arguments passed to subsequent functions are consistent.
 492  * 3) Move vma's page tables to the new range.
 493  * 4) Free up any cleared pgd range.
 494  * 5) Shrink the vma to cover only the new range.
 495  */
 496 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 497 {
 498         struct mm_struct *mm = vma->vm_mm;
 499         unsigned long old_start = vma->vm_start;
 500         unsigned long old_end = vma->vm_end;
 501         unsigned long length = old_end - old_start;
 502         unsigned long new_start = old_start - shift;
 503         unsigned long new_end = old_end - shift;
 504         struct mmu_gather *tlb;
 505
 506         BUG_ON(new_start > new_end);
 507
 508         /*
 509          * ensure there are no vmas between where we want to go
 510          * and where we are
 511          */
 512         if (vma != find_vma(mm, new_start))
 513                 return -EFAULT;
 514
 515         /*
 516          * cover the whole range: [new_start, old_end)
 517          */
 518         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 519
 520         /*
 521          * move the page tables downwards, on failure we rely on
 522          * process cleanup to remove whatever mess we made.
 523          */
 524         if (length != move_page_tables(vma, old_start,
 525                                        vma, new_start, length))
 526                 return -ENOMEM;
 527
 528         lru_add_drain();
 529         tlb = tlb_gather_mmu(mm, 0);
 530         if (new_end > old_start) {
 531                 /*
 532                  * when the old and new regions overlap clear from new_end.
 533                  */
 534                 free_pgd_range(&tlb, new_end, old_end, new_end,
 535                         vma->vm_next ? vma->vm_next->vm_start : 0);
 536         } else {
 537                 /*
 538                  * otherwise, clean from old_start; this is done to not touch
 539                  * the address space in [new_end, old_start) some architectures
 540                  * have constraints on va-space that make this illegal (IA64) -
 541                  * for the others its just a little faster.
 542                  */
 543                 free_pgd_range(&tlb, old_start, old_end, new_end,
 544                         vma->vm_next ? vma->vm_next->vm_start : 0);
 545         }
 546         tlb_finish_mmu(tlb, new_end, old_end);
 547
 548         /*
 549          * shrink the vma to just the new range.
 550          */
 551         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 552
 553         return 0;
 554 }
 555
 556 #define EXTRA_STACK_VM_PAGES    20      /* random */
 557
 558 /*
 559  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 560  * the stack is optionally relocated, and some extra space is added.
 561  */
 562 int setup_arg_pages(struct linux_binprm *bprm,
 563                     unsigned long stack_top,
 564                     int executable_stack)
 565 {
 566         unsigned long ret;
 567         unsigned long stack_shift;
 568         struct mm_struct *mm = current->mm;
 569         struct vm_area_struct *vma = bprm->vma;
 570         struct vm_area_struct *prev = NULL;
 571         unsigned long vm_flags;
 572         unsigned long stack_base;
 573
 574 #ifdef CONFIG_STACK_GROWSUP
 575         /* Limit stack size to 1GB */
 576         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 577         if (stack_base > (1 << 30))
 578                 stack_base = 1 << 30;
 579
 580         /* Make sure we didn't let the argument array grow too large. */
 581         if (vma->vm_end - vma->vm_start > stack_base)
 582                 return -ENOMEM;
 583
 584         stack_base = PAGE_ALIGN(stack_top - stack_base);
 585
 586         stack_shift = vma->vm_start - stack_base;
 587         mm->arg_start = bprm->p - stack_shift;
 588         bprm->p = vma->vm_end - stack_shift;
 589 #else
 590         stack_top = arch_align_stack(stack_top);
 591         stack_top = PAGE_ALIGN(stack_top);
 592         stack_shift = vma->vm_end - stack_top;
 593
 594         bprm->p -= stack_shift;
 595         mm->arg_start = bprm->p;
 596 #endif
 597
 598         if (bprm->loader)
 599                 bprm->loader -= stack_shift;
 600         bprm->exec -= stack_shift;
 601
 602         down_write(&mm->mmap_sem);
 603         vm_flags = vma->vm_flags;
 604
 605         /*
 606          * Adjust stack execute permissions; explicitly enable for
 607          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 608          * (arch default) otherwise.
 609          */
 610         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 611                 vm_flags |= VM_EXEC;
 612         else if (executable_stack == EXSTACK_DISABLE_X)
 613                 vm_flags &= ~VM_EXEC;
 614         vm_flags |= mm->def_flags;
 615
 616         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 617                         vm_flags);
 618         if (ret)
 619                 goto out_unlock;
 620         BUG_ON(prev != vma);
 621
 622         /* Move stack pages down in memory. */
 623         if (stack_shift) {
 624                 ret = shift_arg_pages(vma, stack_shift);
 625                 if (ret) {
 626                         up_write(&mm->mmap_sem);
 627                         return ret;
 628                 }
 629         }
 630
 631 #ifdef CONFIG_STACK_GROWSUP
 632         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 633 #else
 634         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 635 #endif
 636         ret = expand_stack(vma, stack_base);
 637         if (ret)
 638                 ret = -EFAULT;
 639
 640 out_unlock:
 641         up_write(&mm->mmap_sem);
 642         return 0;
 643 }
 644 EXPORT_SYMBOL(setup_arg_pages);
 645
 646 #endif /* CONFIG_MMU */
 647
 648 struct file *open_exec(const char *name)
 649 {
 650         struct nameidata nd;
 651         int err;
 652         struct file *file;
 653
 654         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 655         file = ERR_PTR(err);
 656
 657         if (!err) {
 658                 struct inode *inode = nd.dentry->d_inode;
 659                 file = ERR_PTR(-EACCES);
 660                 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
 661                     S_ISREG(inode->i_mode)) {
 662                         int err = vfs_permission(&nd, MAY_EXEC);
 663                         file = ERR_PTR(err);
 664                         if (!err) {
 665                                 file = nameidata_to_filp(&nd, O_RDONLY);
 666                                 if (!IS_ERR(file)) {
 667                                         err = deny_write_access(file);
 668                                         if (err) {
 669                                                 fput(file);
 670                                                 file = ERR_PTR(err);
 671                                         }
 672                                 }
 673 out:
 674                                 return file;
 675                         }
 676                 }
 677                 release_open_intent(&nd);
 678                 path_release(&nd);
 679         }
 680         goto out;
 681 }
 682
 683 EXPORT_SYMBOL(open_exec);
 684
 685 int kernel_read(struct file *file, unsigned long offset,
 686         char *addr, unsigned long count)
 687 {
 688         mm_segment_t old_fs;
 689         loff_t pos = offset;
 690         int result;
 691
 692         old_fs = get_fs();
 693         set_fs(get_ds());
 694         /* The cast to a user pointer is valid due to the set_fs() */
 695         result = vfs_read(file, (void __user *)addr, count, &pos);
 696         set_fs(old_fs);
 697         return result;
 698 }
 699
 700 EXPORT_SYMBOL(kernel_read);
 701
 702 static int exec_mmap(struct mm_struct *mm)
 703 {
 704         struct task_struct *tsk;
 705         struct mm_struct * old_mm, *active_mm;
 706
 707         /* Notify parent that we're no longer interested in the old VM */
 708         tsk = current;
 709         old_mm = current->mm;
 710         mm_release(tsk, old_mm);
 711
 712         if (old_mm) {
 713                 /*
 714                  * Make sure that if there is a core dump in progress
 715                  * for the old mm, we get out and die instead of going
 716                  * through with the exec.  We must hold mmap_sem around
 717                  * checking core_waiters and changing tsk->mm.  The
 718                  * core-inducing thread will increment core_waiters for
 719                  * each thread whose ->mm == old_mm.
 720                  */
 721                 down_read(&old_mm->mmap_sem);
 722                 if (unlikely(old_mm->core_waiters)) {
 723                         up_read(&old_mm->mmap_sem);
 724                         return -EINTR;
 725                 }
 726         }
 727         task_lock(tsk);
 728         active_mm = tsk->active_mm;
 729         tsk->mm = mm;
 730         tsk->active_mm = mm;
 731         activate_mm(active_mm, mm);
 732         task_unlock(tsk);
 733         arch_pick_mmap_layout(mm);
 734         if (old_mm) {
 735                 up_read(&old_mm->mmap_sem);
 736                 BUG_ON(active_mm != old_mm);
 737                 mmput(old_mm);
 738                 return 0;
 739         }
 740         mmdrop(active_mm);
 741         return 0;
 742 }
 743
 744 /*
 745  * This function makes sure the current process has its own signal table,
 746  * so that flush_signal_handlers can later reset the handlers without
 747  * disturbing other processes.  (Other processes might share the signal
 748  * table via the CLONE_SIGHAND option to clone().)
 749  */
 750 static int de_thread(struct task_struct *tsk)
 751 {
 752         struct signal_struct *sig = tsk->signal;
 753         struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
 754         spinlock_t *lock = &oldsighand->siglock;
 755         struct task_struct *leader = NULL;
 756         int count;
 757
 758         /*
 759          * If we don't share sighandlers, then we aren't sharing anything
 760          * and we can just re-use it all.
 761          */
 762         if (atomic_read(&oldsighand->count) <= 1) {
 763                 exit_itimers(sig);
 764                 return 0;
 765         }
 766
 767         newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 768         if (!newsighand)
 769                 return -ENOMEM;
 770
 771         if (thread_group_empty(tsk))
 772                 goto no_thread_group;
 773
 774         /*
 775          * Kill all other threads in the thread group.
 776          * We must hold tasklist_lock to call zap_other_threads.
 777          */
 778         read_lock(&tasklist_lock);
 779         spin_lock_irq(lock);
 780         if (sig->flags & SIGNAL_GROUP_EXIT) {
 781                 /*
 782                  * Another group action in progress, just
 783                  * return so that the signal is processed.
 784                  */
 785                 spin_unlock_irq(lock);
 786                 read_unlock(&tasklist_lock);
 787                 kmem_cache_free(sighand_cachep, newsighand);
 788                 return -EAGAIN;
 789         }
 790
 791         /*
 792          * child_reaper ignores SIGKILL, change it now.
 793          * Reparenting needs write_lock on tasklist_lock,
 794          * so it is safe to do it under read_lock.
 795          */
 796         if (unlikely(tsk->group_leader == child_reaper(tsk)))
 797                 tsk->nsproxy->pid_ns->child_reaper = tsk;
 798
 799         zap_other_threads(tsk);
 800         read_unlock(&tasklist_lock);
 801
 802         /*
 803          * Account for the thread group leader hanging around:
 804          */
 805         count = 1;
 806         if (!thread_group_leader(tsk)) {
 807                 count = 2;
 808                 /*
 809                  * The SIGALRM timer survives the exec, but needs to point
 810                  * at us as the new group leader now.  We have a race with
 811                  * a timer firing now getting the old leader, so we need to
 812                  * synchronize with any firing (by calling del_timer_sync)
 813                  * before we can safely let the old group leader die.
 814                  */
 815                 sig->tsk = tsk;
 816                 spin_unlock_irq(lock);
 817                 if (hrtimer_cancel(&sig->real_timer))
 818                         hrtimer_restart(&sig->real_timer);
 819                 spin_lock_irq(lock);
 820         }
 821         while (atomic_read(&sig->count) > count) {
 822                 sig->group_exit_task = tsk;
 823                 sig->notify_count = count;
 824                 __set_current_state(TASK_UNINTERRUPTIBLE);
 825                 spin_unlock_irq(lock);
 826                 schedule();
 827                 spin_lock_irq(lock);
 828         }
 829         sig->group_exit_task = NULL;
 830         sig->notify_count = 0;
 831         spin_unlock_irq(lock);
 832
 833         /*
 834          * At this point all other threads have exited, all we have to
 835          * do is to wait for the thread group leader to become inactive,
 836          * and to assume its PID:
 837          */
 838         if (!thread_group_leader(tsk)) {
 839                 /*
 840                  * Wait for the thread group leader to be a zombie.
 841                  * It should already be zombie at this point, most
 842                  * of the time.
 843                  */
 844                 leader = tsk->group_leader;
 845                 while (leader->exit_state != EXIT_ZOMBIE)
 846                         yield();
 847
 848                 /*
 849                  * The only record we have of the real-time age of a
 850                  * process, regardless of execs it's done, is start_time.
 851                  * All the past CPU time is accumulated in signal_struct
 852                  * from sister threads now dead.  But in this non-leader
 853                  * exec, nothing survives from the original leader thread,
 854                  * whose birth marks the true age of this process now.
 855                  * When we take on its identity by switching to its PID, we
 856                  * also take its birthdate (always earlier than our own).
 857                  */
 858                 tsk->start_time = leader->start_time;
 859
 860                 write_lock_irq(&tasklist_lock);
 861
 862                 BUG_ON(leader->tgid != tsk->tgid);
 863                 BUG_ON(tsk->pid == tsk->tgid);
 864                 /*
 865                  * An exec() starts a new thread group with the
 866                  * TGID of the previous thread group. Rehash the
 867                  * two threads with a switched PID, and release
 868                  * the former thread group leader:
 869                  */
 870
 871                 /* Become a process group leader with the old leader's pid.
 872                  * The old leader becomes a thread of the this thread group.
 873                  * Note: The old leader also uses this pid until release_task
 874                  *       is called.  Odd but simple and correct.
 875                  */
 876                 detach_pid(tsk, PIDTYPE_PID);
 877                 tsk->pid = leader->pid;
 878                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
 879                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 880                 transfer_pid(leader, tsk, PIDTYPE_SID);
 881                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 882
 883                 tsk->group_leader = tsk;
 884                 leader->group_leader = tsk;
 885
 886                 tsk->exit_signal = SIGCHLD;
 887
 888                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 889                 leader->exit_state = EXIT_DEAD;
 890
 891                 write_unlock_irq(&tasklist_lock);
 892         }
 893
 894         /*
 895          * There may be one thread left which is just exiting,
 896          * but it's safe to stop telling the group to kill themselves.
 897          */
 898         sig->flags = 0;
 899
 900 no_thread_group:
 901         exit_itimers(sig);
 902         if (leader)
 903                 release_task(leader);
 904
 905         if (atomic_read(&oldsighand->count) == 1) {
 906                 /*
 907                  * Now that we nuked the rest of the thread group,
 908                  * it turns out we are not sharing sighand any more either.
 909                  * So we can just keep it.
 910                  */
 911                 kmem_cache_free(sighand_cachep, newsighand);
 912         } else {
 913                 /*
 914                  * Move our state over to newsighand and switch it in.
 915                  */
 916                 atomic_set(&newsighand->count, 1);
 917                 memcpy(newsighand->action, oldsighand->action,
 918                        sizeof(newsighand->action));
 919
 920                 write_lock_irq(&tasklist_lock);
 921                 spin_lock(&oldsighand->siglock);
 922                 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
 923
 924                 rcu_assign_pointer(tsk->sighand, newsighand);
 925                 recalc_sigpending();
 926
 927                 spin_unlock(&newsighand->siglock);
 928                 spin_unlock(&oldsighand->siglock);
 929                 write_unlock_irq(&tasklist_lock);
 930
 931                 __cleanup_sighand(oldsighand);
 932         }
 933
 934         BUG_ON(!thread_group_leader(tsk));
 935         return 0;
 936 }
 937
 938 /*
 939  * These functions flushes out all traces of the currently running executable
 940  * so that a new one can be started
 941  */
 942
 943 static void flush_old_files(struct files_struct * files)
 944 {
 945         long j = -1;
 946         struct fdtable *fdt;
 947
 948         spin_lock(&files->file_lock);
 949         for (;;) {
 950                 unsigned long set, i;
 951
 952                 j++;
 953                 i = j * __NFDBITS;
 954                 fdt = files_fdtable(files);
 955                 if (i >= fdt->max_fds)
 956                         break;
 957                 set = fdt->close_on_exec->fds_bits[j];
 958                 if (!set)
 959                         continue;
 960                 fdt->close_on_exec->fds_bits[j] = 0;
 961                 spin_unlock(&files->file_lock);
 962                 for ( ; set ; i++,set >>= 1) {
 963                         if (set & 1) {
 964                                 sys_close(i);
 965                         }
 966                 }
 967                 spin_lock(&files->file_lock);
 968
 969         }
 970         spin_unlock(&files->file_lock);
 971 }
 972
 973 void get_task_comm(char *buf, struct task_struct *tsk)
 974 {
 975         /* buf must be at least sizeof(tsk->comm) in size */
 976         task_lock(tsk);
 977         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 978         task_unlock(tsk);
 979 }
 980
 981 void set_task_comm(struct task_struct *tsk, char *buf)
 982 {
 983         task_lock(tsk);
 984         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 985         task_unlock(tsk);
 986 }
 987
 988 int flush_old_exec(struct linux_binprm * bprm)
 989 {
 990         char * name;
 991         int i, ch, retval;
 992         struct files_struct *files;
 993         char tcomm[sizeof(current->comm)];
 994
 995         /*
 996          * Make sure we have a private signal table and that
 997          * we are unassociated from the previous thread group.
 998          */
 999         retval = de_thread(current);
1000         if (retval)
1001                 goto out;
1002
1003         /*
1004          * Make sure we have private file handles. Ask the
1005          * fork helper to do the work for us and the exit
1006          * helper to do the cleanup of the old one.
1007          */
1008         files = current->files;         /* refcounted so safe to hold */
1009         retval = unshare_files();
1010         if (retval)
1011                 goto out;
1012         /*
1013          * Release all of the old mmap stuff
1014          */
1015         retval = exec_mmap(bprm->mm);
1016         if (retval)
1017                 goto mmap_failed;
1018
1019         bprm->mm = NULL;                /* We're using it now */
1020
1021         /* This is the point of no return */
1022         put_files_struct(files);
1023
1024         current->sas_ss_sp = current->sas_ss_size = 0;
1025
1026         if (current->euid == current->uid && current->egid == current->gid)
1027                 set_dumpable(current->mm, 1);
1028         else
1029                 set_dumpable(current->mm, suid_dumpable);
1030
1031         name = bprm->filename;
1032
1033         /* Copies the binary name from after last slash */
1034         for (i=0; (ch = *(name++)) != '\0';) {
1035                 if (ch == '/')
1036                         i = 0; /* overwrite what we wrote */
1037                 else
1038                         if (i < (sizeof(tcomm) - 1))
1039                                 tcomm[i++] = ch;
1040         }
1041         tcomm[i] = '\0';
1042         set_task_comm(current, tcomm);
1043
1044         current->flags &= ~PF_RANDOMIZE;
1045         flush_thread();
1046
1047         /* Set the new mm task size. We have to do that late because it may
1048          * depend on TIF_32BIT which is only updated in flush_thread() on
1049          * some architectures like powerpc
1050          */
1051         current->mm->task_size = TASK_SIZE;
1052
1053         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1054                 suid_keys(current);
1055                 set_dumpable(current->mm, suid_dumpable);
1056                 current->pdeath_signal = 0;
1057         } else if (file_permission(bprm->file, MAY_READ) ||
1058                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1059                 suid_keys(current);
1060                 set_dumpable(current->mm, suid_dumpable);
1061         }
1062
1063         /* An exec changes our domain. We are no longer part of the thread
1064            group */
1065
1066         current->self_exec_id++;
1067
1068         flush_signal_handlers(current, 0);
1069         flush_old_files(current->files);
1070
1071         return 0;
1072
1073 mmap_failed:
1074         reset_files_struct(current, files);
1075 out:
1076         return retval;
1077 }
1078
1079 EXPORT_SYMBOL(flush_old_exec);
1080
1081 /*
1082  * Fill the binprm structure from the inode.
1083  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1084  */
1085 int prepare_binprm(struct linux_binprm *bprm)
1086 {
1087         int mode;
1088         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1089         int retval;
1090
1091         mode = inode->i_mode;
1092         if (bprm->file->f_op == NULL)
1093                 return -EACCES;
1094
1095         bprm->e_uid = current->euid;
1096         bprm->e_gid = current->egid;
1097
1098         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1099                 /* Set-uid? */
1100                 if (mode & S_ISUID) {
1101                         current->personality &= ~PER_CLEAR_ON_SETID;
1102                         bprm->e_uid = inode->i_uid;
1103                 }
1104
1105                 /* Set-gid? */
1106                 /*
1107                  * If setgid is set but no group execute bit then this
1108                  * is a candidate for mandatory locking, not a setgid
1109                  * executable.
1110                  */
1111                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1112                         current->personality &= ~PER_CLEAR_ON_SETID;
1113                         bprm->e_gid = inode->i_gid;
1114                 }
1115         }
1116
1117         /* fill in binprm security blob */
1118         retval = security_bprm_set(bprm);
1119         if (retval)
1120                 return retval;
1121
1122         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1123         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1124 }
1125
1126 EXPORT_SYMBOL(prepare_binprm);
1127
1128 static int unsafe_exec(struct task_struct *p)
1129 {
1130         int unsafe = 0;
1131         if (p->ptrace & PT_PTRACED) {
1132                 if (p->ptrace & PT_PTRACE_CAP)
1133                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1134                 else
1135                         unsafe |= LSM_UNSAFE_PTRACE;
1136         }
1137         if (atomic_read(&p->fs->count) > 1 ||
1138             atomic_read(&p->files->count) > 1 ||
1139             atomic_read(&p->sighand->count) > 1)
1140                 unsafe |= LSM_UNSAFE_SHARE;
1141
1142         return unsafe;
1143 }
1144
1145 void compute_creds(struct linux_binprm *bprm)
1146 {
1147         int unsafe;
1148
1149         if (bprm->e_uid != current->uid) {
1150                 suid_keys(current);
1151                 current->pdeath_signal = 0;
1152         }
1153         exec_keys(current);
1154
1155         task_lock(current);
1156         unsafe = unsafe_exec(current);
1157         security_bprm_apply_creds(bprm, unsafe);
1158         task_unlock(current);
1159         security_bprm_post_apply_creds(bprm);
1160 }
1161 EXPORT_SYMBOL(compute_creds);
1162
1163 /*
1164  * Arguments are '\0' separated strings found at the location bprm->p
1165  * points to; chop off the first by relocating brpm->p to right after
1166  * the first '\0' encountered.
1167  */
1168 int remove_arg_zero(struct linux_binprm *bprm)
1169 {
1170         int ret = 0;
1171         unsigned long offset;
1172         char *kaddr;
1173         struct page *page;
1174
1175         if (!bprm->argc)
1176                 return 0;
1177
1178         do {
1179                 offset = bprm->p & ~PAGE_MASK;
1180                 page = get_arg_page(bprm, bprm->p, 0);
1181                 if (!page) {
1182                         ret = -EFAULT;
1183                         goto out;
1184                 }
1185                 kaddr = kmap_atomic(page, KM_USER0);
1186
1187                 for (; offset < PAGE_SIZE && kaddr[offset];
1188                                 offset++, bprm->p++)
1189                         ;
1190
1191                 kunmap_atomic(kaddr, KM_USER0);
1192                 put_arg_page(page);
1193
1194                 if (offset == PAGE_SIZE)
1195                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1196         } while (offset == PAGE_SIZE);
1197
1198         bprm->p++;
1199         bprm->argc--;
1200         ret = 0;
1201
1202 out:
1203         return ret;
1204 }
1205 EXPORT_SYMBOL(remove_arg_zero);
1206
1207 /*
1208  * cycle the list of binary formats handler, until one recognizes the image
1209  */
1210 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1211 {
1212         int try,retval;
1213         struct linux_binfmt *fmt;
1214 #ifdef __alpha__
1215         /* handle /sbin/loader.. */
1216         {
1217             struct exec * eh = (struct exec *) bprm->buf;
1218
1219             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1220                 (eh->fh.f_flags & 0x3000) == 0x3000)
1221             {
1222                 struct file * file;
1223                 unsigned long loader;
1224
1225                 allow_write_access(bprm->file);
1226                 fput(bprm->file);
1227                 bprm->file = NULL;
1228
1229                 loader = bprm->vma->vm_end - sizeof(void *);
1230
1231                 file = open_exec("/sbin/loader");
1232                 retval = PTR_ERR(file);
1233                 if (IS_ERR(file))
1234                         return retval;
1235
1236                 /* Remember if the application is TASO.  */
1237                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1238
1239                 bprm->file = file;
1240                 bprm->loader = loader;
1241                 retval = prepare_binprm(bprm);
1242                 if (retval<0)
1243                         return retval;
1244                 /* should call search_binary_handler recursively here,
1245                    but it does not matter */
1246             }
1247         }
1248 #endif
1249         retval = security_bprm_check(bprm);
1250         if (retval)
1251                 return retval;
1252
1253         /* kernel module loader fixup */
1254         /* so we don't try to load run modprobe in kernel space. */
1255         set_fs(USER_DS);
1256
1257         retval = audit_bprm(bprm);
1258         if (retval)
1259                 return retval;
1260
1261         retval = -ENOENT;
1262         for (try=0; try<2; try++) {
1263                 read_lock(&binfmt_lock);
1264                 list_for_each_entry(fmt, &formats, lh) {
1265                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1266                         if (!fn)
1267                                 continue;
1268                         if (!try_module_get(fmt->module))
1269                                 continue;
1270                         read_unlock(&binfmt_lock);
1271                         retval = fn(bprm, regs);
1272                         if (retval >= 0) {
1273                                 put_binfmt(fmt);
1274                                 allow_write_access(bprm->file);
1275                                 if (bprm->file)
1276                                         fput(bprm->file);
1277                                 bprm->file = NULL;
1278                                 current->did_exec = 1;
1279                                 proc_exec_connector(current);
1280                                 return retval;
1281                         }
1282                         read_lock(&binfmt_lock);
1283                         put_binfmt(fmt);
1284                         if (retval != -ENOEXEC || bprm->mm == NULL)
1285                                 break;
1286                         if (!bprm->file) {
1287                                 read_unlock(&binfmt_lock);
1288                                 return retval;
1289                         }
1290                 }
1291                 read_unlock(&binfmt_lock);
1292                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1293                         break;
1294 #ifdef CONFIG_KMOD
1295                 }else{
1296 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1297                         if (printable(bprm->buf[0]) &&
1298                             printable(bprm->buf[1]) &&
1299                             printable(bprm->buf[2]) &&
1300                             printable(bprm->buf[3]))
1301                                 break; /* -ENOEXEC */
1302                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1303 #endif
1304                 }
1305         }
1306         return retval;
1307 }
1308
1309 EXPORT_SYMBOL(search_binary_handler);
1310
1311 /*
1312  * sys_execve() executes a new program.
1313  */
1314 int do_execve(char * filename,
1315         char __user *__user *argv,
1316         char __user *__user *envp,
1317         struct pt_regs * regs)
1318 {
1319         struct linux_binprm *bprm;
1320         struct file *file;
1321         unsigned long env_p;
1322         int retval;
1323
1324         retval = -ENOMEM;
1325         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1326         if (!bprm)
1327                 goto out_ret;
1328
1329         file = open_exec(filename);
1330         retval = PTR_ERR(file);
1331         if (IS_ERR(file))
1332                 goto out_kfree;
1333
1334         sched_exec();
1335
1336         bprm->file = file;
1337         bprm->filename = filename;
1338         bprm->interp = filename;
1339
1340         retval = bprm_mm_init(bprm);
1341         if (retval)
1342                 goto out_file;
1343
1344         bprm->argc = count(argv, MAX_ARG_STRINGS);
1345         if ((retval = bprm->argc) < 0)
1346                 goto out_mm;
1347
1348         bprm->envc = count(envp, MAX_ARG_STRINGS);
1349         if ((retval = bprm->envc) < 0)
1350                 goto out_mm;
1351
1352         retval = security_bprm_alloc(bprm);
1353         if (retval)
1354                 goto out;
1355
1356         retval = prepare_binprm(bprm);
1357         if (retval < 0)
1358                 goto out;
1359
1360         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1361         if (retval < 0)
1362                 goto out;
1363
1364         bprm->exec = bprm->p;
1365         retval = copy_strings(bprm->envc, envp, bprm);
1366         if (retval < 0)
1367                 goto out;
1368
1369         env_p = bprm->p;
1370         retval = copy_strings(bprm->argc, argv, bprm);
1371         if (retval < 0)
1372                 goto out;
1373         bprm->argv_len = env_p - bprm->p;
1374
1375         retval = search_binary_handler(bprm,regs);
1376         if (retval >= 0) {
1377                 /* execve success */
1378                 free_arg_pages(bprm);
1379                 security_bprm_free(bprm);
1380                 acct_update_integrals(current);
1381                 kfree(bprm);
1382                 return retval;
1383         }
1384
1385 out:
1386         free_arg_pages(bprm);
1387         if (bprm->security)
1388                 security_bprm_free(bprm);
1389
1390 out_mm:
1391         if (bprm->mm)
1392                 mmput (bprm->mm);
1393
1394 out_file:
1395         if (bprm->file) {
1396                 allow_write_access(bprm->file);
1397                 fput(bprm->file);
1398         }
1399 out_kfree:
1400         kfree(bprm);
1401
1402 out_ret:
1403         return retval;
1404 }
1405
1406 int set_binfmt(struct linux_binfmt *new)
1407 {
1408         struct linux_binfmt *old = current->binfmt;
1409
1410         if (new) {
1411                 if (!try_module_get(new->module))
1412                         return -1;
1413         }
1414         current->binfmt = new;
1415         if (old)
1416                 module_put(old->module);
1417         return 0;
1418 }
1419
1420 EXPORT_SYMBOL(set_binfmt);
1421
1422 /* format_corename will inspect the pattern parameter, and output a
1423  * name into corename, which must have space for at least
1424  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1425  */
1426 static int format_corename(char *corename, const char *pattern, long signr)
1427 {
1428         const char *pat_ptr = pattern;
1429         char *out_ptr = corename;
1430         char *const out_end = corename + CORENAME_MAX_SIZE;
1431         int rc;
1432         int pid_in_pattern = 0;
1433         int ispipe = 0;
1434
1435         if (*pattern == '|')
1436                 ispipe = 1;
1437
1438         /* Repeat as long as we have more pattern to process and more output
1439            space */
1440         while (*pat_ptr) {
1441                 if (*pat_ptr != '%') {
1442                         if (out_ptr == out_end)
1443                                 goto out;
1444                         *out_ptr++ = *pat_ptr++;
1445                 } else {
1446                         switch (*++pat_ptr) {
1447                         case 0:
1448                                 goto out;
1449                         /* Double percent, output one percent */
1450                         case '%':
1451                                 if (out_ptr == out_end)
1452                                         goto out;
1453                                 *out_ptr++ = '%';
1454                                 break;
1455                         /* pid */
1456                         case 'p':
1457                                 pid_in_pattern = 1;
1458                                 rc = snprintf(out_ptr, out_end - out_ptr,
1459                                               "%d", current->tgid);
1460                                 if (rc > out_end - out_ptr)
1461                                         goto out;
1462                                 out_ptr += rc;
1463                                 break;
1464                         /* uid */
1465                         case 'u':
1466                                 rc = snprintf(out_ptr, out_end - out_ptr,
1467                                               "%d", current->uid);
1468                                 if (rc > out_end - out_ptr)
1469                                         goto out;
1470                                 out_ptr += rc;
1471                                 break;
1472                         /* gid */
1473                         case 'g':
1474                                 rc = snprintf(out_ptr, out_end - out_ptr,
1475                                               "%d", current->gid);
1476                                 if (rc > out_end - out_ptr)
1477                                         goto out;
1478                                 out_ptr += rc;
1479                                 break;
1480                         /* signal that caused the coredump */
1481                         case 's':
1482                                 rc = snprintf(out_ptr, out_end - out_ptr,
1483                                               "%ld", signr);
1484                                 if (rc > out_end - out_ptr)
1485                                         goto out;
1486                                 out_ptr += rc;
1487                                 break;
1488                         /* UNIX time of coredump */
1489                         case 't': {
1490                                 struct timeval tv;
1491                                 do_gettimeofday(&tv);
1492                                 rc = snprintf(out_ptr, out_end - out_ptr,
1493                                               "%lu", tv.tv_sec);
1494                                 if (rc > out_end - out_ptr)
1495                                         goto out;
1496                                 out_ptr += rc;
1497                                 break;
1498                         }
1499                         /* hostname */
1500                         case 'h':
1501                                 down_read(&uts_sem);
1502                                 rc = snprintf(out_ptr, out_end - out_ptr,
1503                                               "%s", utsname()->nodename);
1504                                 up_read(&uts_sem);
1505                                 if (rc > out_end - out_ptr)
1506                                         goto out;
1507                                 out_ptr += rc;
1508                                 break;
1509                         /* executable */
1510                         case 'e':
1511                                 rc = snprintf(out_ptr, out_end - out_ptr,
1512                                               "%s", current->comm);
1513                                 if (rc > out_end - out_ptr)
1514                                         goto out;
1515                                 out_ptr += rc;
1516                                 break;
1517                         default:
1518                                 break;
1519                         }
1520                         ++pat_ptr;
1521                 }
1522         }
1523         /* Backward compatibility with core_uses_pid:
1524          *
1525          * If core_pattern does not include a %p (as is the default)
1526          * and core_uses_pid is set, then .%pid will be appended to
1527          * the filename. Do not do this for piped commands. */
1528         if (!ispipe && !pid_in_pattern
1529             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1530                 rc = snprintf(out_ptr, out_end - out_ptr,
1531                               ".%d", current->tgid);
1532                 if (rc > out_end - out_ptr)
1533                         goto out;
1534                 out_ptr += rc;
1535         }
1536 out:
1537         *out_ptr = 0;
1538         return ispipe;
1539 }
1540
1541 static void zap_process(struct task_struct *start)
1542 {
1543         struct task_struct *t;
1544
1545         start->signal->flags = SIGNAL_GROUP_EXIT;
1546         start->signal->group_stop_count = 0;
1547
1548         t = start;
1549         do {
1550                 if (t != current && t->mm) {
1551                         t->mm->core_waiters++;
1552                         sigaddset(&t->pending.signal, SIGKILL);
1553                         signal_wake_up(t, 1);
1554                 }
1555         } while ((t = next_thread(t)) != start);
1556 }
1557
1558 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1559                                 int exit_code)
1560 {
1561         struct task_struct *g, *p;
1562         unsigned long flags;
1563         int err = -EAGAIN;
1564
1565         spin_lock_irq(&tsk->sighand->siglock);
1566         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1567                 tsk->signal->group_exit_code = exit_code;
1568                 zap_process(tsk);
1569                 err = 0;
1570         }
1571         spin_unlock_irq(&tsk->sighand->siglock);
1572         if (err)
1573                 return err;
1574
1575         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1576                 goto done;
1577
1578         rcu_read_lock();
1579         for_each_process(g) {
1580                 if (g == tsk->group_leader)
1581                         continue;
1582
1583                 p = g;
1584                 do {
1585                         if (p->mm) {
1586                                 if (p->mm == mm) {
1587                                         /*
1588                                          * p->sighand can't disappear, but
1589                                          * may be changed by de_thread()
1590                                          */
1591                                         lock_task_sighand(p, &flags);
1592                                         zap_process(p);
1593                                         unlock_task_sighand(p, &flags);
1594                                 }
1595                                 break;
1596                         }
1597                 } while ((p = next_thread(p)) != g);
1598         }
1599         rcu_read_unlock();
1600 done:
1601         return mm->core_waiters;
1602 }
1603
1604 static int coredump_wait(int exit_code)
1605 {
1606         struct task_struct *tsk = current;
1607         struct mm_struct *mm = tsk->mm;
1608         struct completion startup_done;
1609         struct completion *vfork_done;
1610         int core_waiters;
1611
1612         init_completion(&mm->core_done);
1613         init_completion(&startup_done);
1614         mm->core_startup_done = &startup_done;
1615
1616         core_waiters = zap_threads(tsk, mm, exit_code);
1617         up_write(&mm->mmap_sem);
1618
1619         if (unlikely(core_waiters < 0))
1620                 goto fail;
1621
1622         /*
1623          * Make sure nobody is waiting for us to release the VM,
1624          * otherwise we can deadlock when we wait on each other
1625          */
1626         vfork_done = tsk->vfork_done;
1627         if (vfork_done) {
1628                 tsk->vfork_done = NULL;
1629                 complete(vfork_done);
1630         }
1631
1632         if (core_waiters)
1633                 wait_for_completion(&startup_done);
1634 fail:
1635         BUG_ON(mm->core_waiters);
1636         return core_waiters;
1637 }
1638
1639 /*
1640  * set_dumpable converts traditional three-value dumpable to two flags and
1641  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1642  * these bits are not changed atomically.  So get_dumpable can observe the
1643  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1644  * return either old dumpable or new one by paying attention to the order of
1645  * modifying the bits.
1646  *
1647  * dumpable |   mm->flags (binary)
1648  * old  new | initial interim  final
1649  * ---------+-----------------------
1650  *  0    1  |   00      01      01
1651  *  0    2  |   00      10(*)   11
1652  *  1    0  |   01      00      00
1653  *  1    2  |   01      11      11
1654  *  2    0  |   11      10(*)   00
1655  *  2    1  |   11      11      01
1656  *
1657  * (*) get_dumpable regards interim value of 10 as 11.
1658  */
1659 void set_dumpable(struct mm_struct *mm, int value)
1660 {
1661         switch (value) {
1662         case 0:
1663                 clear_bit(MMF_DUMPABLE, &mm->flags);
1664                 smp_wmb();
1665                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1666                 break;
1667         case 1:
1668                 set_bit(MMF_DUMPABLE, &mm->flags);
1669                 smp_wmb();
1670                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1671                 break;
1672         case 2:
1673                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1674                 smp_wmb();
1675                 set_bit(MMF_DUMPABLE, &mm->flags);
1676                 break;
1677         }
1678 }
1679 EXPORT_SYMBOL_GPL(set_dumpable);
1680
1681 int get_dumpable(struct mm_struct *mm)
1682 {
1683         int ret;
1684
1685         ret = mm->flags & 0x3;
1686         return (ret >= 2) ? 2 : ret;
1687 }
1688
1689 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1690 {
1691         char corename[CORENAME_MAX_SIZE + 1];
1692         struct mm_struct *mm = current->mm;
1693         struct linux_binfmt * binfmt;
1694         struct inode * inode;
1695         struct file * file;
1696         int retval = 0;
1697         int fsuid = current->fsuid;
1698         int flag = 0;
1699         int ispipe = 0;
1700
1701         audit_core_dumps(signr);
1702
1703         binfmt = current->binfmt;
1704         if (!binfmt || !binfmt->core_dump)
1705                 goto fail;
1706         down_write(&mm->mmap_sem);
1707         if (!get_dumpable(mm)) {
1708                 up_write(&mm->mmap_sem);
1709                 goto fail;
1710         }
1711
1712         /*
1713          *      We cannot trust fsuid as being the "true" uid of the
1714          *      process nor do we know its entire history. We only know it
1715          *      was tainted so we dump it as root in mode 2.
1716          */
1717         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1718                 flag = O_EXCL;          /* Stop rewrite attacks */
1719                 current->fsuid = 0;     /* Dump root private */
1720         }
1721         set_dumpable(mm, 0);
1722
1723         retval = coredump_wait(exit_code);
1724         if (retval < 0)
1725                 goto fail;
1726
1727         /*
1728          * Clear any false indication of pending signals that might
1729          * be seen by the filesystem code called to write the core file.
1730          */
1731         clear_thread_flag(TIF_SIGPENDING);
1732
1733         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1734                 goto fail_unlock;
1735
1736         /*
1737          * lock_kernel() because format_corename() is controlled by sysctl, which
1738          * uses lock_kernel()
1739          */
1740         lock_kernel();
1741         ispipe = format_corename(corename, core_pattern, signr);
1742         unlock_kernel();
1743         if (ispipe) {
1744                 /* SIGPIPE can happen, but it's just never processed */
1745                 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1746                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1747                                corename);
1748                         goto fail_unlock;
1749                 }
1750         } else
1751                 file = filp_open(corename,
1752                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1753                                  0600);
1754         if (IS_ERR(file))
1755                 goto fail_unlock;
1756         inode = file->f_path.dentry->d_inode;
1757         if (inode->i_nlink > 1)
1758                 goto close_fail;        /* multiple links - don't dump */
1759         if (!ispipe && d_unhashed(file->f_path.dentry))
1760                 goto close_fail;
1761
1762         /* AK: actually i see no reason to not allow this for named pipes etc.,
1763            but keep the previous behaviour for now. */
1764         if (!ispipe && !S_ISREG(inode->i_mode))
1765                 goto close_fail;
1766         if (!file->f_op)
1767                 goto close_fail;
1768         if (!file->f_op->write)
1769                 goto close_fail;
1770         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1771                 goto close_fail;
1772
1773         retval = binfmt->core_dump(signr, regs, file);
1774
1775         if (retval)
1776                 current->signal->group_exit_code |= 0x80;
1777 close_fail:
1778         filp_close(file, NULL);
1779 fail_unlock:
1780         current->fsuid = fsuid;
1781         complete_all(&mm->core_done);
1782 fail:
1783         return retval;
1784 }