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/fdtable.h>
  28 #include <linux/mm.h>
  29 #include <linux/stat.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/swap.h>
  33 #include <linux/string.h>
  34 #include <linux/init.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/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/mount.h>
  46 #include <linux/security.h>
  47 #include <linux/syscalls.h>
  48 #include <linux/tsacct_kern.h>
  49 #include <linux/cn_proc.h>
  50 #include <linux/audit.h>
  51 #include <linux/tracehook.h>
  52
  53 #include <asm/uaccess.h>
  54 #include <asm/mmu_context.h>
  55 #include <asm/tlb.h>
  56
  57 #ifdef CONFIG_KMOD
  58 #include <linux/kmod.h>
  59 #endif
  60
  61 #ifdef __alpha__
  62 /* for /sbin/loader handling in search_binary_handler() */
  63 #include <linux/a.out.h>
  64 #endif
  65
  66 int core_uses_pid;
  67 char core_pattern[CORENAME_MAX_SIZE] = "core";
  68 int suid_dumpable = 0;
  69
  70 /* The maximal length of core_pattern is also specified in sysctl.c */
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 int register_binfmt(struct linux_binfmt * fmt)
  76 {
  77         if (!fmt)
  78                 return -EINVAL;
  79         write_lock(&binfmt_lock);
  80         list_add(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82         return 0;
  83 }
  84
  85 EXPORT_SYMBOL(register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 asmlinkage long sys_uselib(const char __user * library)
 108 {
 109         struct file * file;
 110         struct nameidata nd;
 111         int error;
 112
 113         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
 114         if (error)
 115                 goto out;
 116
 117         error = -EINVAL;
 118         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 119                 goto exit;
 120
 121         error = -EACCES;
 122         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
 123                 goto exit;
 124
 125         error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
 126         if (error)
 127                 goto exit;
 128
 129         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 130         error = PTR_ERR(file);
 131         if (IS_ERR(file))
 132                 goto out;
 133
 134         error = -ENOEXEC;
 135         if(file->f_op) {
 136                 struct linux_binfmt * fmt;
 137
 138                 read_lock(&binfmt_lock);
 139                 list_for_each_entry(fmt, &formats, lh) {
 140                         if (!fmt->load_shlib)
 141                                 continue;
 142                         if (!try_module_get(fmt->module))
 143                                 continue;
 144                         read_unlock(&binfmt_lock);
 145                         error = fmt->load_shlib(file);
 146                         read_lock(&binfmt_lock);
 147                         put_binfmt(fmt);
 148                         if (error != -ENOEXEC)
 149                                 break;
 150                 }
 151                 read_unlock(&binfmt_lock);
 152         }
 153         fput(file);
 154 out:
 155         return error;
 156 exit:
 157         release_open_intent(&nd);
 158         path_put(&nd.path);
 159         goto out;
 160 }
 161
 162 #ifdef CONFIG_MMU
 163
 164 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 165                 int write)
 166 {
 167         struct page *page;
 168         int ret;
 169
 170 #ifdef CONFIG_STACK_GROWSUP
 171         if (write) {
 172                 ret = expand_stack_downwards(bprm->vma, pos);
 173                 if (ret < 0)
 174                         return NULL;
 175         }
 176 #endif
 177         ret = get_user_pages(current, bprm->mm, pos,
 178                         1, write, 1, &page, NULL);
 179         if (ret <= 0)
 180                 return NULL;
 181
 182         if (write) {
 183                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 184                 struct rlimit *rlim;
 185
 186                 /*
 187                  * We've historically supported up to 32 pages (ARG_MAX)
 188                  * of argument strings even with small stacks
 189                  */
 190                 if (size <= ARG_MAX)
 191                         return page;
 192
 193                 /*
 194                  * Limit to 1/4-th the stack size for the argv+env strings.
 195                  * This ensures that:
 196                  *  - the remaining binfmt code will not run out of stack space,
 197                  *  - the program will have a reasonable amount of stack left
 198                  *    to work from.
 199                  */
 200                 rlim = current->signal->rlim;
 201                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
 202                         put_page(page);
 203                         return NULL;
 204                 }
 205         }
 206
 207         return page;
 208 }
 209
 210 static void put_arg_page(struct page *page)
 211 {
 212         put_page(page);
 213 }
 214
 215 static void free_arg_page(struct linux_binprm *bprm, int i)
 216 {
 217 }
 218
 219 static void free_arg_pages(struct linux_binprm *bprm)
 220 {
 221 }
 222
 223 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 224                 struct page *page)
 225 {
 226         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 227 }
 228
 229 static int __bprm_mm_init(struct linux_binprm *bprm)
 230 {
 231         int err = -ENOMEM;
 232         struct vm_area_struct *vma = NULL;
 233         struct mm_struct *mm = bprm->mm;
 234
 235         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 236         if (!vma)
 237                 goto err;
 238
 239         down_write(&mm->mmap_sem);
 240         vma->vm_mm = mm;
 241
 242         /*
 243          * Place the stack at the largest stack address the architecture
 244          * supports. Later, we'll move this to an appropriate place. We don't
 245          * use STACK_TOP because that can depend on attributes which aren't
 246          * configured yet.
 247          */
 248         vma->vm_end = STACK_TOP_MAX;
 249         vma->vm_start = vma->vm_end - PAGE_SIZE;
 250
 251         vma->vm_flags = VM_STACK_FLAGS;
 252         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 253         err = insert_vm_struct(mm, vma);
 254         if (err) {
 255                 up_write(&mm->mmap_sem);
 256                 goto err;
 257         }
 258
 259         mm->stack_vm = mm->total_vm = 1;
 260         up_write(&mm->mmap_sem);
 261
 262         bprm->p = vma->vm_end - sizeof(void *);
 263
 264         return 0;
 265
 266 err:
 267         if (vma) {
 268                 bprm->vma = NULL;
 269                 kmem_cache_free(vm_area_cachep, vma);
 270         }
 271
 272         return err;
 273 }
 274
 275 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 276 {
 277         return len <= MAX_ARG_STRLEN;
 278 }
 279
 280 #else
 281
 282 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 283                 int write)
 284 {
 285         struct page *page;
 286
 287         page = bprm->page[pos / PAGE_SIZE];
 288         if (!page && write) {
 289                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 290                 if (!page)
 291                         return NULL;
 292                 bprm->page[pos / PAGE_SIZE] = page;
 293         }
 294
 295         return page;
 296 }
 297
 298 static void put_arg_page(struct page *page)
 299 {
 300 }
 301
 302 static void free_arg_page(struct linux_binprm *bprm, int i)
 303 {
 304         if (bprm->page[i]) {
 305                 __free_page(bprm->page[i]);
 306                 bprm->page[i] = NULL;
 307         }
 308 }
 309
 310 static void free_arg_pages(struct linux_binprm *bprm)
 311 {
 312         int i;
 313
 314         for (i = 0; i < MAX_ARG_PAGES; i++)
 315                 free_arg_page(bprm, i);
 316 }
 317
 318 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 319                 struct page *page)
 320 {
 321 }
 322
 323 static int __bprm_mm_init(struct linux_binprm *bprm)
 324 {
 325         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 326         return 0;
 327 }
 328
 329 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 330 {
 331         return len <= bprm->p;
 332 }
 333
 334 #endif /* CONFIG_MMU */
 335
 336 /*
 337  * Create a new mm_struct and populate it with a temporary stack
 338  * vm_area_struct.  We don't have enough context at this point to set the stack
 339  * flags, permissions, and offset, so we use temporary values.  We'll update
 340  * them later in setup_arg_pages().
 341  */
 342 int bprm_mm_init(struct linux_binprm *bprm)
 343 {
 344         int err;
 345         struct mm_struct *mm = NULL;
 346
 347         bprm->mm = mm = mm_alloc();
 348         err = -ENOMEM;
 349         if (!mm)
 350                 goto err;
 351
 352         err = init_new_context(current, mm);
 353         if (err)
 354                 goto err;
 355
 356         err = __bprm_mm_init(bprm);
 357         if (err)
 358                 goto err;
 359
 360         return 0;
 361
 362 err:
 363         if (mm) {
 364                 bprm->mm = NULL;
 365                 mmdrop(mm);
 366         }
 367
 368         return err;
 369 }
 370
 371 /*
 372  * count() counts the number of strings in array ARGV.
 373  */
 374 static int count(char __user * __user * argv, int max)
 375 {
 376         int i = 0;
 377
 378         if (argv != NULL) {
 379                 for (;;) {
 380                         char __user * p;
 381
 382                         if (get_user(p, argv))
 383                                 return -EFAULT;
 384                         if (!p)
 385                                 break;
 386                         argv++;
 387                         if(++i > max)
 388                                 return -E2BIG;
 389                         cond_resched();
 390                 }
 391         }
 392         return i;
 393 }
 394
 395 /*
 396  * 'copy_strings()' copies argument/environment strings from the old
 397  * processes's memory to the new process's stack.  The call to get_user_pages()
 398  * ensures the destination page is created and not swapped out.
 399  */
 400 static int copy_strings(int argc, char __user * __user * argv,
 401                         struct linux_binprm *bprm)
 402 {
 403         struct page *kmapped_page = NULL;
 404         char *kaddr = NULL;
 405         unsigned long kpos = 0;
 406         int ret;
 407
 408         while (argc-- > 0) {
 409                 char __user *str;
 410                 int len;
 411                 unsigned long pos;
 412
 413                 if (get_user(str, argv+argc) ||
 414                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
 415                         ret = -EFAULT;
 416                         goto out;
 417                 }
 418
 419                 if (!valid_arg_len(bprm, len)) {
 420                         ret = -E2BIG;
 421                         goto out;
 422                 }
 423
 424                 /* We're going to work our way backwords. */
 425                 pos = bprm->p;
 426                 str += len;
 427                 bprm->p -= len;
 428
 429                 while (len > 0) {
 430                         int offset, bytes_to_copy;
 431
 432                         offset = pos % PAGE_SIZE;
 433                         if (offset == 0)
 434                                 offset = PAGE_SIZE;
 435
 436                         bytes_to_copy = offset;
 437                         if (bytes_to_copy > len)
 438                                 bytes_to_copy = len;
 439
 440                         offset -= bytes_to_copy;
 441                         pos -= bytes_to_copy;
 442                         str -= bytes_to_copy;
 443                         len -= bytes_to_copy;
 444
 445                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 446                                 struct page *page;
 447
 448                                 page = get_arg_page(bprm, pos, 1);
 449                                 if (!page) {
 450                                         ret = -E2BIG;
 451                                         goto out;
 452                                 }
 453
 454                                 if (kmapped_page) {
 455                                         flush_kernel_dcache_page(kmapped_page);
 456                                         kunmap(kmapped_page);
 457                                         put_arg_page(kmapped_page);
 458                                 }
 459                                 kmapped_page = page;
 460                                 kaddr = kmap(kmapped_page);
 461                                 kpos = pos & PAGE_MASK;
 462                                 flush_arg_page(bprm, kpos, kmapped_page);
 463                         }
 464                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 465                                 ret = -EFAULT;
 466                                 goto out;
 467                         }
 468                 }
 469         }
 470         ret = 0;
 471 out:
 472         if (kmapped_page) {
 473                 flush_kernel_dcache_page(kmapped_page);
 474                 kunmap(kmapped_page);
 475                 put_arg_page(kmapped_page);
 476         }
 477         return ret;
 478 }
 479
 480 /*
 481  * Like copy_strings, but get argv and its values from kernel memory.
 482  */
 483 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
 484 {
 485         int r;
 486         mm_segment_t oldfs = get_fs();
 487         set_fs(KERNEL_DS);
 488         r = copy_strings(argc, (char __user * __user *)argv, bprm);
 489         set_fs(oldfs);
 490         return r;
 491 }
 492 EXPORT_SYMBOL(copy_strings_kernel);
 493
 494 #ifdef CONFIG_MMU
 495
 496 /*
 497  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 498  * the binfmt code determines where the new stack should reside, we shift it to
 499  * its final location.  The process proceeds as follows:
 500  *
 501  * 1) Use shift to calculate the new vma endpoints.
 502  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 503  *    arguments passed to subsequent functions are consistent.
 504  * 3) Move vma's page tables to the new range.
 505  * 4) Free up any cleared pgd range.
 506  * 5) Shrink the vma to cover only the new range.
 507  */
 508 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 509 {
 510         struct mm_struct *mm = vma->vm_mm;
 511         unsigned long old_start = vma->vm_start;
 512         unsigned long old_end = vma->vm_end;
 513         unsigned long length = old_end - old_start;
 514         unsigned long new_start = old_start - shift;
 515         unsigned long new_end = old_end - shift;
 516         struct mmu_gather *tlb;
 517
 518         BUG_ON(new_start > new_end);
 519
 520         /*
 521          * ensure there are no vmas between where we want to go
 522          * and where we are
 523          */
 524         if (vma != find_vma(mm, new_start))
 525                 return -EFAULT;
 526
 527         /*
 528          * cover the whole range: [new_start, old_end)
 529          */
 530         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
 531
 532         /*
 533          * move the page tables downwards, on failure we rely on
 534          * process cleanup to remove whatever mess we made.
 535          */
 536         if (length != move_page_tables(vma, old_start,
 537                                        vma, new_start, length))
 538                 return -ENOMEM;
 539
 540         lru_add_drain();
 541         tlb = tlb_gather_mmu(mm, 0);
 542         if (new_end > old_start) {
 543                 /*
 544                  * when the old and new regions overlap clear from new_end.
 545                  */
 546                 free_pgd_range(tlb, new_end, old_end, new_end,
 547                         vma->vm_next ? vma->vm_next->vm_start : 0);
 548         } else {
 549                 /*
 550                  * otherwise, clean from old_start; this is done to not touch
 551                  * the address space in [new_end, old_start) some architectures
 552                  * have constraints on va-space that make this illegal (IA64) -
 553                  * for the others its just a little faster.
 554                  */
 555                 free_pgd_range(tlb, old_start, old_end, new_end,
 556                         vma->vm_next ? vma->vm_next->vm_start : 0);
 557         }
 558         tlb_finish_mmu(tlb, new_end, old_end);
 559
 560         /*
 561          * shrink the vma to just the new range.
 562          */
 563         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 564
 565         return 0;
 566 }
 567
 568 #define EXTRA_STACK_VM_PAGES    20      /* random */
 569
 570 /*
 571  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 572  * the stack is optionally relocated, and some extra space is added.
 573  */
 574 int setup_arg_pages(struct linux_binprm *bprm,
 575                     unsigned long stack_top,
 576                     int executable_stack)
 577 {
 578         unsigned long ret;
 579         unsigned long stack_shift;
 580         struct mm_struct *mm = current->mm;
 581         struct vm_area_struct *vma = bprm->vma;
 582         struct vm_area_struct *prev = NULL;
 583         unsigned long vm_flags;
 584         unsigned long stack_base;
 585
 586 #ifdef CONFIG_STACK_GROWSUP
 587         /* Limit stack size to 1GB */
 588         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
 589         if (stack_base > (1 << 30))
 590                 stack_base = 1 << 30;
 591
 592         /* Make sure we didn't let the argument array grow too large. */
 593         if (vma->vm_end - vma->vm_start > stack_base)
 594                 return -ENOMEM;
 595
 596         stack_base = PAGE_ALIGN(stack_top - stack_base);
 597
 598         stack_shift = vma->vm_start - stack_base;
 599         mm->arg_start = bprm->p - stack_shift;
 600         bprm->p = vma->vm_end - stack_shift;
 601 #else
 602         stack_top = arch_align_stack(stack_top);
 603         stack_top = PAGE_ALIGN(stack_top);
 604         stack_shift = vma->vm_end - stack_top;
 605
 606         bprm->p -= stack_shift;
 607         mm->arg_start = bprm->p;
 608 #endif
 609
 610         if (bprm->loader)
 611                 bprm->loader -= stack_shift;
 612         bprm->exec -= stack_shift;
 613
 614         down_write(&mm->mmap_sem);
 615         vm_flags = VM_STACK_FLAGS;
 616
 617         /*
 618          * Adjust stack execute permissions; explicitly enable for
 619          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 620          * (arch default) otherwise.
 621          */
 622         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 623                 vm_flags |= VM_EXEC;
 624         else if (executable_stack == EXSTACK_DISABLE_X)
 625                 vm_flags &= ~VM_EXEC;
 626         vm_flags |= mm->def_flags;
 627
 628         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 629                         vm_flags);
 630         if (ret)
 631                 goto out_unlock;
 632         BUG_ON(prev != vma);
 633
 634         /* Move stack pages down in memory. */
 635         if (stack_shift) {
 636                 ret = shift_arg_pages(vma, stack_shift);
 637                 if (ret) {
 638                         up_write(&mm->mmap_sem);
 639                         return ret;
 640                 }
 641         }
 642
 643 #ifdef CONFIG_STACK_GROWSUP
 644         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 645 #else
 646         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
 647 #endif
 648         ret = expand_stack(vma, stack_base);
 649         if (ret)
 650                 ret = -EFAULT;
 651
 652 out_unlock:
 653         up_write(&mm->mmap_sem);
 654         return 0;
 655 }
 656 EXPORT_SYMBOL(setup_arg_pages);
 657
 658 #endif /* CONFIG_MMU */
 659
 660 struct file *open_exec(const char *name)
 661 {
 662         struct nameidata nd;
 663         struct file *file;
 664         int err;
 665
 666         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
 667                                 FMODE_READ|FMODE_EXEC);
 668         if (err)
 669                 goto out;
 670
 671         err = -EACCES;
 672         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 673                 goto out_path_put;
 674
 675         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
 676                 goto out_path_put;
 677
 678         err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
 679         if (err)
 680                 goto out_path_put;
 681
 682         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
 683         if (IS_ERR(file))
 684                 return file;
 685
 686         err = deny_write_access(file);
 687         if (err) {
 688                 fput(file);
 689                 goto out;
 690         }
 691
 692         return file;
 693
 694  out_path_put:
 695         release_open_intent(&nd);
 696         path_put(&nd.path);
 697  out:
 698         return ERR_PTR(err);
 699 }
 700 EXPORT_SYMBOL(open_exec);
 701
 702 int kernel_read(struct file *file, unsigned long offset,
 703         char *addr, unsigned long count)
 704 {
 705         mm_segment_t old_fs;
 706         loff_t pos = offset;
 707         int result;
 708
 709         old_fs = get_fs();
 710         set_fs(get_ds());
 711         /* The cast to a user pointer is valid due to the set_fs() */
 712         result = vfs_read(file, (void __user *)addr, count, &pos);
 713         set_fs(old_fs);
 714         return result;
 715 }
 716
 717 EXPORT_SYMBOL(kernel_read);
 718
 719 static int exec_mmap(struct mm_struct *mm)
 720 {
 721         struct task_struct *tsk;
 722         struct mm_struct * old_mm, *active_mm;
 723
 724         /* Notify parent that we're no longer interested in the old VM */
 725         tsk = current;
 726         old_mm = current->mm;
 727         mm_release(tsk, old_mm);
 728
 729         if (old_mm) {
 730                 /*
 731                  * Make sure that if there is a core dump in progress
 732                  * for the old mm, we get out and die instead of going
 733                  * through with the exec.  We must hold mmap_sem around
 734                  * checking core_state and changing tsk->mm.
 735                  */
 736                 down_read(&old_mm->mmap_sem);
 737                 if (unlikely(old_mm->core_state)) {
 738                         up_read(&old_mm->mmap_sem);
 739                         return -EINTR;
 740                 }
 741         }
 742         task_lock(tsk);
 743         active_mm = tsk->active_mm;
 744         tsk->mm = mm;
 745         tsk->active_mm = mm;
 746         activate_mm(active_mm, mm);
 747         task_unlock(tsk);
 748         mm_update_next_owner(old_mm);
 749         arch_pick_mmap_layout(mm);
 750         if (old_mm) {
 751                 up_read(&old_mm->mmap_sem);
 752                 BUG_ON(active_mm != old_mm);
 753                 mmput(old_mm);
 754                 return 0;
 755         }
 756         mmdrop(active_mm);
 757         return 0;
 758 }
 759
 760 /*
 761  * This function makes sure the current process has its own signal table,
 762  * so that flush_signal_handlers can later reset the handlers without
 763  * disturbing other processes.  (Other processes might share the signal
 764  * table via the CLONE_SIGHAND option to clone().)
 765  */
 766 static int de_thread(struct task_struct *tsk)
 767 {
 768         struct signal_struct *sig = tsk->signal;
 769         struct sighand_struct *oldsighand = tsk->sighand;
 770         spinlock_t *lock = &oldsighand->siglock;
 771         struct task_struct *leader = NULL;
 772         int count;
 773
 774         if (thread_group_empty(tsk))
 775                 goto no_thread_group;
 776
 777         /*
 778          * Kill all other threads in the thread group.
 779          */
 780         spin_lock_irq(lock);
 781         if (signal_group_exit(sig)) {
 782                 /*
 783                  * Another group action in progress, just
 784                  * return so that the signal is processed.
 785                  */
 786                 spin_unlock_irq(lock);
 787                 return -EAGAIN;
 788         }
 789         sig->group_exit_task = tsk;
 790         zap_other_threads(tsk);
 791
 792         /* Account for the thread group leader hanging around: */
 793         count = thread_group_leader(tsk) ? 1 : 2;
 794         sig->notify_count = count;
 795         while (atomic_read(&sig->count) > count) {
 796                 __set_current_state(TASK_UNINTERRUPTIBLE);
 797                 spin_unlock_irq(lock);
 798                 schedule();
 799                 spin_lock_irq(lock);
 800         }
 801         spin_unlock_irq(lock);
 802
 803         /*
 804          * At this point all other threads have exited, all we have to
 805          * do is to wait for the thread group leader to become inactive,
 806          * and to assume its PID:
 807          */
 808         if (!thread_group_leader(tsk)) {
 809                 leader = tsk->group_leader;
 810
 811                 sig->notify_count = -1; /* for exit_notify() */
 812                 for (;;) {
 813                         write_lock_irq(&tasklist_lock);
 814                         if (likely(leader->exit_state))
 815                                 break;
 816                         __set_current_state(TASK_UNINTERRUPTIBLE);
 817                         write_unlock_irq(&tasklist_lock);
 818                         schedule();
 819                 }
 820
 821                 if (unlikely(task_child_reaper(tsk) == leader))
 822                         task_active_pid_ns(tsk)->child_reaper = tsk;
 823                 /*
 824                  * The only record we have of the real-time age of a
 825                  * process, regardless of execs it's done, is start_time.
 826                  * All the past CPU time is accumulated in signal_struct
 827                  * from sister threads now dead.  But in this non-leader
 828                  * exec, nothing survives from the original leader thread,
 829                  * whose birth marks the true age of this process now.
 830                  * When we take on its identity by switching to its PID, we
 831                  * also take its birthdate (always earlier than our own).
 832                  */
 833                 tsk->start_time = leader->start_time;
 834
 835                 BUG_ON(!same_thread_group(leader, tsk));
 836                 BUG_ON(has_group_leader_pid(tsk));
 837                 /*
 838                  * An exec() starts a new thread group with the
 839                  * TGID of the previous thread group. Rehash the
 840                  * two threads with a switched PID, and release
 841                  * the former thread group leader:
 842                  */
 843
 844                 /* Become a process group leader with the old leader's pid.
 845                  * The old leader becomes a thread of the this thread group.
 846                  * Note: The old leader also uses this pid until release_task
 847                  *       is called.  Odd but simple and correct.
 848                  */
 849                 detach_pid(tsk, PIDTYPE_PID);
 850                 tsk->pid = leader->pid;
 851                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 852                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 853                 transfer_pid(leader, tsk, PIDTYPE_SID);
 854                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 855
 856                 tsk->group_leader = tsk;
 857                 leader->group_leader = tsk;
 858
 859                 tsk->exit_signal = SIGCHLD;
 860
 861                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 862                 leader->exit_state = EXIT_DEAD;
 863
 864                 write_unlock_irq(&tasklist_lock);
 865         }
 866
 867         sig->group_exit_task = NULL;
 868         sig->notify_count = 0;
 869
 870 no_thread_group:
 871         exit_itimers(sig);
 872         flush_itimer_signals();
 873         if (leader)
 874                 release_task(leader);
 875
 876         if (atomic_read(&oldsighand->count) != 1) {
 877                 struct sighand_struct *newsighand;
 878                 /*
 879                  * This ->sighand is shared with the CLONE_SIGHAND
 880                  * but not CLONE_THREAD task, switch to the new one.
 881                  */
 882                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 883                 if (!newsighand)
 884                         return -ENOMEM;
 885
 886                 atomic_set(&newsighand->count, 1);
 887                 memcpy(newsighand->action, oldsighand->action,
 888                        sizeof(newsighand->action));
 889
 890                 write_lock_irq(&tasklist_lock);
 891                 spin_lock(&oldsighand->siglock);
 892                 rcu_assign_pointer(tsk->sighand, newsighand);
 893                 spin_unlock(&oldsighand->siglock);
 894                 write_unlock_irq(&tasklist_lock);
 895
 896                 __cleanup_sighand(oldsighand);
 897         }
 898
 899         BUG_ON(!thread_group_leader(tsk));
 900         return 0;
 901 }
 902
 903 /*
 904  * These functions flushes out all traces of the currently running executable
 905  * so that a new one can be started
 906  */
 907 static void flush_old_files(struct files_struct * files)
 908 {
 909         long j = -1;
 910         struct fdtable *fdt;
 911
 912         spin_lock(&files->file_lock);
 913         for (;;) {
 914                 unsigned long set, i;
 915
 916                 j++;
 917                 i = j * __NFDBITS;
 918                 fdt = files_fdtable(files);
 919                 if (i >= fdt->max_fds)
 920                         break;
 921                 set = fdt->close_on_exec->fds_bits[j];
 922                 if (!set)
 923                         continue;
 924                 fdt->close_on_exec->fds_bits[j] = 0;
 925                 spin_unlock(&files->file_lock);
 926                 for ( ; set ; i++,set >>= 1) {
 927                         if (set & 1) {
 928                                 sys_close(i);
 929                         }
 930                 }
 931                 spin_lock(&files->file_lock);
 932
 933         }
 934         spin_unlock(&files->file_lock);
 935 }
 936
 937 char *get_task_comm(char *buf, struct task_struct *tsk)
 938 {
 939         /* buf must be at least sizeof(tsk->comm) in size */
 940         task_lock(tsk);
 941         strncpy(buf, tsk->comm, sizeof(tsk->comm));
 942         task_unlock(tsk);
 943         return buf;
 944 }
 945
 946 void set_task_comm(struct task_struct *tsk, char *buf)
 947 {
 948         task_lock(tsk);
 949         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
 950         task_unlock(tsk);
 951 }
 952
 953 int flush_old_exec(struct linux_binprm * bprm)
 954 {
 955         char * name;
 956         int i, ch, retval;
 957         char tcomm[sizeof(current->comm)];
 958
 959         /*
 960          * Make sure we have a private signal table and that
 961          * we are unassociated from the previous thread group.
 962          */
 963         retval = de_thread(current);
 964         if (retval)
 965                 goto out;
 966
 967         set_mm_exe_file(bprm->mm, bprm->file);
 968
 969         /*
 970          * Release all of the old mmap stuff
 971          */
 972         retval = exec_mmap(bprm->mm);
 973         if (retval)
 974                 goto out;
 975
 976         bprm->mm = NULL;                /* We're using it now */
 977
 978         /* This is the point of no return */
 979         current->sas_ss_sp = current->sas_ss_size = 0;
 980
 981         if (current->euid == current->uid && current->egid == current->gid)
 982                 set_dumpable(current->mm, 1);
 983         else
 984                 set_dumpable(current->mm, suid_dumpable);
 985
 986         name = bprm->filename;
 987
 988         /* Copies the binary name from after last slash */
 989         for (i=0; (ch = *(name++)) != '\0';) {
 990                 if (ch == '/')
 991                         i = 0; /* overwrite what we wrote */
 992                 else
 993                         if (i < (sizeof(tcomm) - 1))
 994                                 tcomm[i++] = ch;
 995         }
 996         tcomm[i] = '\0';
 997         set_task_comm(current, tcomm);
 998
 999         current->flags &= ~PF_RANDOMIZE;
1000         flush_thread();
1001
1002         /* Set the new mm task size. We have to do that late because it may
1003          * depend on TIF_32BIT which is only updated in flush_thread() on
1004          * some architectures like powerpc
1005          */
1006         current->mm->task_size = TASK_SIZE;
1007
1008         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1009                 suid_keys(current);
1010                 set_dumpable(current->mm, suid_dumpable);
1011                 current->pdeath_signal = 0;
1012         } else if (file_permission(bprm->file, MAY_READ) ||
1013                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1014                 suid_keys(current);
1015                 set_dumpable(current->mm, suid_dumpable);
1016         }
1017
1018         /* An exec changes our domain. We are no longer part of the thread
1019            group */
1020
1021         current->self_exec_id++;
1022
1023         flush_signal_handlers(current, 0);
1024         flush_old_files(current->files);
1025
1026         return 0;
1027
1028 out:
1029         return retval;
1030 }
1031
1032 EXPORT_SYMBOL(flush_old_exec);
1033
1034 /*
1035  * Fill the binprm structure from the inode.
1036  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1037  */
1038 int prepare_binprm(struct linux_binprm *bprm)
1039 {
1040         int mode;
1041         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1042         int retval;
1043
1044         mode = inode->i_mode;
1045         if (bprm->file->f_op == NULL)
1046                 return -EACCES;
1047
1048         bprm->e_uid = current->euid;
1049         bprm->e_gid = current->egid;
1050
1051         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1052                 /* Set-uid? */
1053                 if (mode & S_ISUID) {
1054                         current->personality &= ~PER_CLEAR_ON_SETID;
1055                         bprm->e_uid = inode->i_uid;
1056                 }
1057
1058                 /* Set-gid? */
1059                 /*
1060                  * If setgid is set but no group execute bit then this
1061                  * is a candidate for mandatory locking, not a setgid
1062                  * executable.
1063                  */
1064                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1065                         current->personality &= ~PER_CLEAR_ON_SETID;
1066                         bprm->e_gid = inode->i_gid;
1067                 }
1068         }
1069
1070         /* fill in binprm security blob */
1071         retval = security_bprm_set(bprm);
1072         if (retval)
1073                 return retval;
1074
1075         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1076         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1077 }
1078
1079 EXPORT_SYMBOL(prepare_binprm);
1080
1081 static int unsafe_exec(struct task_struct *p)
1082 {
1083         int unsafe = tracehook_unsafe_exec(p);
1084
1085         if (atomic_read(&p->fs->count) > 1 ||
1086             atomic_read(&p->files->count) > 1 ||
1087             atomic_read(&p->sighand->count) > 1)
1088                 unsafe |= LSM_UNSAFE_SHARE;
1089
1090         return unsafe;
1091 }
1092
1093 void compute_creds(struct linux_binprm *bprm)
1094 {
1095         int unsafe;
1096
1097         if (bprm->e_uid != current->uid) {
1098                 suid_keys(current);
1099                 current->pdeath_signal = 0;
1100         }
1101         exec_keys(current);
1102
1103         task_lock(current);
1104         unsafe = unsafe_exec(current);
1105         security_bprm_apply_creds(bprm, unsafe);
1106         task_unlock(current);
1107         security_bprm_post_apply_creds(bprm);
1108 }
1109 EXPORT_SYMBOL(compute_creds);
1110
1111 /*
1112  * Arguments are '\0' separated strings found at the location bprm->p
1113  * points to; chop off the first by relocating brpm->p to right after
1114  * the first '\0' encountered.
1115  */
1116 int remove_arg_zero(struct linux_binprm *bprm)
1117 {
1118         int ret = 0;
1119         unsigned long offset;
1120         char *kaddr;
1121         struct page *page;
1122
1123         if (!bprm->argc)
1124                 return 0;
1125
1126         do {
1127                 offset = bprm->p & ~PAGE_MASK;
1128                 page = get_arg_page(bprm, bprm->p, 0);
1129                 if (!page) {
1130                         ret = -EFAULT;
1131                         goto out;
1132                 }
1133                 kaddr = kmap_atomic(page, KM_USER0);
1134
1135                 for (; offset < PAGE_SIZE && kaddr[offset];
1136                                 offset++, bprm->p++)
1137                         ;
1138
1139                 kunmap_atomic(kaddr, KM_USER0);
1140                 put_arg_page(page);
1141
1142                 if (offset == PAGE_SIZE)
1143                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1144         } while (offset == PAGE_SIZE);
1145
1146         bprm->p++;
1147         bprm->argc--;
1148         ret = 0;
1149
1150 out:
1151         return ret;
1152 }
1153 EXPORT_SYMBOL(remove_arg_zero);
1154
1155 /*
1156  * cycle the list of binary formats handler, until one recognizes the image
1157  */
1158 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1159 {
1160         int try,retval;
1161         struct linux_binfmt *fmt;
1162 #ifdef __alpha__
1163         /* handle /sbin/loader.. */
1164         {
1165             struct exec * eh = (struct exec *) bprm->buf;
1166
1167             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1168                 (eh->fh.f_flags & 0x3000) == 0x3000)
1169             {
1170                 struct file * file;
1171                 unsigned long loader;
1172
1173                 allow_write_access(bprm->file);
1174                 fput(bprm->file);
1175                 bprm->file = NULL;
1176
1177                 loader = bprm->vma->vm_end - sizeof(void *);
1178
1179                 file = open_exec("/sbin/loader");
1180                 retval = PTR_ERR(file);
1181                 if (IS_ERR(file))
1182                         return retval;
1183
1184                 /* Remember if the application is TASO.  */
1185                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1186
1187                 bprm->file = file;
1188                 bprm->loader = loader;
1189                 retval = prepare_binprm(bprm);
1190                 if (retval<0)
1191                         return retval;
1192                 /* should call search_binary_handler recursively here,
1193                    but it does not matter */
1194             }
1195         }
1196 #endif
1197         retval = security_bprm_check(bprm);
1198         if (retval)
1199                 return retval;
1200
1201         /* kernel module loader fixup */
1202         /* so we don't try to load run modprobe in kernel space. */
1203         set_fs(USER_DS);
1204
1205         retval = audit_bprm(bprm);
1206         if (retval)
1207                 return retval;
1208
1209         retval = -ENOENT;
1210         for (try=0; try<2; try++) {
1211                 read_lock(&binfmt_lock);
1212                 list_for_each_entry(fmt, &formats, lh) {
1213                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1214                         if (!fn)
1215                                 continue;
1216                         if (!try_module_get(fmt->module))
1217                                 continue;
1218                         read_unlock(&binfmt_lock);
1219                         retval = fn(bprm, regs);
1220                         if (retval >= 0) {
1221                                 tracehook_report_exec(fmt, bprm, regs);
1222                                 put_binfmt(fmt);
1223                                 allow_write_access(bprm->file);
1224                                 if (bprm->file)
1225                                         fput(bprm->file);
1226                                 bprm->file = NULL;
1227                                 current->did_exec = 1;
1228                                 proc_exec_connector(current);
1229                                 return retval;
1230                         }
1231                         read_lock(&binfmt_lock);
1232                         put_binfmt(fmt);
1233                         if (retval != -ENOEXEC || bprm->mm == NULL)
1234                                 break;
1235                         if (!bprm->file) {
1236                                 read_unlock(&binfmt_lock);
1237                                 return retval;
1238                         }
1239                 }
1240                 read_unlock(&binfmt_lock);
1241                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1242                         break;
1243 #ifdef CONFIG_KMOD
1244                 }else{
1245 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1246                         if (printable(bprm->buf[0]) &&
1247                             printable(bprm->buf[1]) &&
1248                             printable(bprm->buf[2]) &&
1249                             printable(bprm->buf[3]))
1250                                 break; /* -ENOEXEC */
1251                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1252 #endif
1253                 }
1254         }
1255         return retval;
1256 }
1257
1258 EXPORT_SYMBOL(search_binary_handler);
1259
1260 void free_bprm(struct linux_binprm *bprm)
1261 {
1262         free_arg_pages(bprm);
1263         kfree(bprm);
1264 }
1265
1266 /*
1267  * sys_execve() executes a new program.
1268  */
1269 int do_execve(char * filename,
1270         char __user *__user *argv,
1271         char __user *__user *envp,
1272         struct pt_regs * regs)
1273 {
1274         struct linux_binprm *bprm;
1275         struct file *file;
1276         struct files_struct *displaced;
1277         int retval;
1278
1279         retval = unshare_files(&displaced);
1280         if (retval)
1281                 goto out_ret;
1282
1283         retval = -ENOMEM;
1284         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1285         if (!bprm)
1286                 goto out_files;
1287
1288         file = open_exec(filename);
1289         retval = PTR_ERR(file);
1290         if (IS_ERR(file))
1291                 goto out_kfree;
1292
1293         sched_exec();
1294
1295         bprm->file = file;
1296         bprm->filename = filename;
1297         bprm->interp = filename;
1298
1299         retval = bprm_mm_init(bprm);
1300         if (retval)
1301                 goto out_file;
1302
1303         bprm->argc = count(argv, MAX_ARG_STRINGS);
1304         if ((retval = bprm->argc) < 0)
1305                 goto out_mm;
1306
1307         bprm->envc = count(envp, MAX_ARG_STRINGS);
1308         if ((retval = bprm->envc) < 0)
1309                 goto out_mm;
1310
1311         retval = security_bprm_alloc(bprm);
1312         if (retval)
1313                 goto out;
1314
1315         retval = prepare_binprm(bprm);
1316         if (retval < 0)
1317                 goto out;
1318
1319         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1320         if (retval < 0)
1321                 goto out;
1322
1323         bprm->exec = bprm->p;
1324         retval = copy_strings(bprm->envc, envp, bprm);
1325         if (retval < 0)
1326                 goto out;
1327
1328         retval = copy_strings(bprm->argc, argv, bprm);
1329         if (retval < 0)
1330                 goto out;
1331
1332         current->flags &= ~PF_KTHREAD;
1333         retval = search_binary_handler(bprm,regs);
1334         if (retval >= 0) {
1335                 /* execve success */
1336                 security_bprm_free(bprm);
1337                 acct_update_integrals(current);
1338                 free_bprm(bprm);
1339                 if (displaced)
1340                         put_files_struct(displaced);
1341                 return retval;
1342         }
1343
1344 out:
1345         if (bprm->security)
1346                 security_bprm_free(bprm);
1347
1348 out_mm:
1349         if (bprm->mm)
1350                 mmput (bprm->mm);
1351
1352 out_file:
1353         if (bprm->file) {
1354                 allow_write_access(bprm->file);
1355                 fput(bprm->file);
1356         }
1357 out_kfree:
1358         free_bprm(bprm);
1359
1360 out_files:
1361         if (displaced)
1362                 reset_files_struct(displaced);
1363 out_ret:
1364         return retval;
1365 }
1366
1367 int set_binfmt(struct linux_binfmt *new)
1368 {
1369         struct linux_binfmt *old = current->binfmt;
1370
1371         if (new) {
1372                 if (!try_module_get(new->module))
1373                         return -1;
1374         }
1375         current->binfmt = new;
1376         if (old)
1377                 module_put(old->module);
1378         return 0;
1379 }
1380
1381 EXPORT_SYMBOL(set_binfmt);
1382
1383 /* format_corename will inspect the pattern parameter, and output a
1384  * name into corename, which must have space for at least
1385  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1386  */
1387 static int format_corename(char *corename, int nr_threads, long signr)
1388 {
1389         const char *pat_ptr = core_pattern;
1390         int ispipe = (*pat_ptr == '|');
1391         char *out_ptr = corename;
1392         char *const out_end = corename + CORENAME_MAX_SIZE;
1393         int rc;
1394         int pid_in_pattern = 0;
1395
1396         /* Repeat as long as we have more pattern to process and more output
1397            space */
1398         while (*pat_ptr) {
1399                 if (*pat_ptr != '%') {
1400                         if (out_ptr == out_end)
1401                                 goto out;
1402                         *out_ptr++ = *pat_ptr++;
1403                 } else {
1404                         switch (*++pat_ptr) {
1405                         case 0:
1406                                 goto out;
1407                         /* Double percent, output one percent */
1408                         case '%':
1409                                 if (out_ptr == out_end)
1410                                         goto out;
1411                                 *out_ptr++ = '%';
1412                                 break;
1413                         /* pid */
1414                         case 'p':
1415                                 pid_in_pattern = 1;
1416                                 rc = snprintf(out_ptr, out_end - out_ptr,
1417                                               "%d", task_tgid_vnr(current));
1418                                 if (rc > out_end - out_ptr)
1419                                         goto out;
1420                                 out_ptr += rc;
1421                                 break;
1422                         /* uid */
1423                         case 'u':
1424                                 rc = snprintf(out_ptr, out_end - out_ptr,
1425                                               "%d", current->uid);
1426                                 if (rc > out_end - out_ptr)
1427                                         goto out;
1428                                 out_ptr += rc;
1429                                 break;
1430                         /* gid */
1431                         case 'g':
1432                                 rc = snprintf(out_ptr, out_end - out_ptr,
1433                                               "%d", current->gid);
1434                                 if (rc > out_end - out_ptr)
1435                                         goto out;
1436                                 out_ptr += rc;
1437                                 break;
1438                         /* signal that caused the coredump */
1439                         case 's':
1440                                 rc = snprintf(out_ptr, out_end - out_ptr,
1441                                               "%ld", signr);
1442                                 if (rc > out_end - out_ptr)
1443                                         goto out;
1444                                 out_ptr += rc;
1445                                 break;
1446                         /* UNIX time of coredump */
1447                         case 't': {
1448                                 struct timeval tv;
1449                                 do_gettimeofday(&tv);
1450                                 rc = snprintf(out_ptr, out_end - out_ptr,
1451                                               "%lu", tv.tv_sec);
1452                                 if (rc > out_end - out_ptr)
1453                                         goto out;
1454                                 out_ptr += rc;
1455                                 break;
1456                         }
1457                         /* hostname */
1458                         case 'h':
1459                                 down_read(&uts_sem);
1460                                 rc = snprintf(out_ptr, out_end - out_ptr,
1461                                               "%s", utsname()->nodename);
1462                                 up_read(&uts_sem);
1463                                 if (rc > out_end - out_ptr)
1464                                         goto out;
1465                                 out_ptr += rc;
1466                                 break;
1467                         /* executable */
1468                         case 'e':
1469                                 rc = snprintf(out_ptr, out_end - out_ptr,
1470                                               "%s", current->comm);
1471                                 if (rc > out_end - out_ptr)
1472                                         goto out;
1473                                 out_ptr += rc;
1474                                 break;
1475                         /* core limit size */
1476                         case 'c':
1477                                 rc = snprintf(out_ptr, out_end - out_ptr,
1478                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1479                                 if (rc > out_end - out_ptr)
1480                                         goto out;
1481                                 out_ptr += rc;
1482                                 break;
1483                         default:
1484                                 break;
1485                         }
1486                         ++pat_ptr;
1487                 }
1488         }
1489         /* Backward compatibility with core_uses_pid:
1490          *
1491          * If core_pattern does not include a %p (as is the default)
1492          * and core_uses_pid is set, then .%pid will be appended to
1493          * the filename. Do not do this for piped commands. */
1494         if (!ispipe && !pid_in_pattern
1495             && (core_uses_pid || nr_threads)) {
1496                 rc = snprintf(out_ptr, out_end - out_ptr,
1497                               ".%d", task_tgid_vnr(current));
1498                 if (rc > out_end - out_ptr)
1499                         goto out;
1500                 out_ptr += rc;
1501         }
1502 out:
1503         *out_ptr = 0;
1504         return ispipe;
1505 }
1506
1507 static int zap_process(struct task_struct *start)
1508 {
1509         struct task_struct *t;
1510         int nr = 0;
1511
1512         start->signal->flags = SIGNAL_GROUP_EXIT;
1513         start->signal->group_stop_count = 0;
1514
1515         t = start;
1516         do {
1517                 if (t != current && t->mm) {
1518                         sigaddset(&t->pending.signal, SIGKILL);
1519                         signal_wake_up(t, 1);
1520                         nr++;
1521                 }
1522         } while_each_thread(start, t);
1523
1524         return nr;
1525 }
1526
1527 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1528                                 struct core_state *core_state, int exit_code)
1529 {
1530         struct task_struct *g, *p;
1531         unsigned long flags;
1532         int nr = -EAGAIN;
1533
1534         spin_lock_irq(&tsk->sighand->siglock);
1535         if (!signal_group_exit(tsk->signal)) {
1536                 mm->core_state = core_state;
1537                 tsk->signal->group_exit_code = exit_code;
1538                 nr = zap_process(tsk);
1539         }
1540         spin_unlock_irq(&tsk->sighand->siglock);
1541         if (unlikely(nr < 0))
1542                 return nr;
1543
1544         if (atomic_read(&mm->mm_users) == nr + 1)
1545                 goto done;
1546         /*
1547          * We should find and kill all tasks which use this mm, and we should
1548          * count them correctly into ->nr_threads. We don't take tasklist
1549          * lock, but this is safe wrt:
1550          *
1551          * fork:
1552          *      None of sub-threads can fork after zap_process(leader). All
1553          *      processes which were created before this point should be
1554          *      visible to zap_threads() because copy_process() adds the new
1555          *      process to the tail of init_task.tasks list, and lock/unlock
1556          *      of ->siglock provides a memory barrier.
1557          *
1558          * do_exit:
1559          *      The caller holds mm->mmap_sem. This means that the task which
1560          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1561          *      its ->mm.
1562          *
1563          * de_thread:
1564          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1565          *      we must see either old or new leader, this does not matter.
1566          *      However, it can change p->sighand, so lock_task_sighand(p)
1567          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1568          *      it can't fail.
1569          *
1570          *      Note also that "g" can be the old leader with ->mm == NULL
1571          *      and already unhashed and thus removed from ->thread_group.
1572          *      This is OK, __unhash_process()->list_del_rcu() does not
1573          *      clear the ->next pointer, we will find the new leader via
1574          *      next_thread().
1575          */
1576         rcu_read_lock();
1577         for_each_process(g) {
1578                 if (g == tsk->group_leader)
1579                         continue;
1580                 if (g->flags & PF_KTHREAD)
1581                         continue;
1582                 p = g;
1583                 do {
1584                         if (p->mm) {
1585                                 if (unlikely(p->mm == mm)) {
1586                                         lock_task_sighand(p, &flags);
1587                                         nr += zap_process(p);
1588                                         unlock_task_sighand(p, &flags);
1589                                 }
1590                                 break;
1591                         }
1592                 } while_each_thread(g, p);
1593         }
1594         rcu_read_unlock();
1595 done:
1596         atomic_set(&core_state->nr_threads, nr);
1597         return nr;
1598 }
1599
1600 static int coredump_wait(int exit_code, struct core_state *core_state)
1601 {
1602         struct task_struct *tsk = current;
1603         struct mm_struct *mm = tsk->mm;
1604         struct completion *vfork_done;
1605         int core_waiters;
1606
1607         init_completion(&core_state->startup);
1608         core_state->dumper.task = tsk;
1609         core_state->dumper.next = NULL;
1610         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1611         up_write(&mm->mmap_sem);
1612
1613         if (unlikely(core_waiters < 0))
1614                 goto fail;
1615
1616         /*
1617          * Make sure nobody is waiting for us to release the VM,
1618          * otherwise we can deadlock when we wait on each other
1619          */
1620         vfork_done = tsk->vfork_done;
1621         if (vfork_done) {
1622                 tsk->vfork_done = NULL;
1623                 complete(vfork_done);
1624         }
1625
1626         if (core_waiters)
1627                 wait_for_completion(&core_state->startup);
1628 fail:
1629         return core_waiters;
1630 }
1631
1632 static void coredump_finish(struct mm_struct *mm)
1633 {
1634         struct core_thread *curr, *next;
1635         struct task_struct *task;
1636
1637         next = mm->core_state->dumper.next;
1638         while ((curr = next) != NULL) {
1639                 next = curr->next;
1640                 task = curr->task;
1641                 /*
1642                  * see exit_mm(), curr->task must not see
1643                  * ->task == NULL before we read ->next.
1644                  */
1645                 smp_mb();
1646                 curr->task = NULL;
1647                 wake_up_process(task);
1648         }
1649
1650         mm->core_state = NULL;
1651 }
1652
1653 /*
1654  * set_dumpable converts traditional three-value dumpable to two flags and
1655  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1656  * these bits are not changed atomically.  So get_dumpable can observe the
1657  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1658  * return either old dumpable or new one by paying attention to the order of
1659  * modifying the bits.
1660  *
1661  * dumpable |   mm->flags (binary)
1662  * old  new | initial interim  final
1663  * ---------+-----------------------
1664  *  0    1  |   00      01      01
1665  *  0    2  |   00      10(*)   11
1666  *  1    0  |   01      00      00
1667  *  1    2  |   01      11      11
1668  *  2    0  |   11      10(*)   00
1669  *  2    1  |   11      11      01
1670  *
1671  * (*) get_dumpable regards interim value of 10 as 11.
1672  */
1673 void set_dumpable(struct mm_struct *mm, int value)
1674 {
1675         switch (value) {
1676         case 0:
1677                 clear_bit(MMF_DUMPABLE, &mm->flags);
1678                 smp_wmb();
1679                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1680                 break;
1681         case 1:
1682                 set_bit(MMF_DUMPABLE, &mm->flags);
1683                 smp_wmb();
1684                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1685                 break;
1686         case 2:
1687                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1688                 smp_wmb();
1689                 set_bit(MMF_DUMPABLE, &mm->flags);
1690                 break;
1691         }
1692 }
1693
1694 int get_dumpable(struct mm_struct *mm)
1695 {
1696         int ret;
1697
1698         ret = mm->flags & 0x3;
1699         return (ret >= 2) ? 2 : ret;
1700 }
1701
1702 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1703 {
1704         struct core_state core_state;
1705         char corename[CORENAME_MAX_SIZE + 1];
1706         struct mm_struct *mm = current->mm;
1707         struct linux_binfmt * binfmt;
1708         struct inode * inode;
1709         struct file * file;
1710         int retval = 0;
1711         int fsuid = current->fsuid;
1712         int flag = 0;
1713         int ispipe = 0;
1714         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1715         char **helper_argv = NULL;
1716         int helper_argc = 0;
1717         char *delimit;
1718
1719         audit_core_dumps(signr);
1720
1721         binfmt = current->binfmt;
1722         if (!binfmt || !binfmt->core_dump)
1723                 goto fail;
1724         down_write(&mm->mmap_sem);
1725         /*
1726          * If another thread got here first, or we are not dumpable, bail out.
1727          */
1728         if (mm->core_state || !get_dumpable(mm)) {
1729                 up_write(&mm->mmap_sem);
1730                 goto fail;
1731         }
1732
1733         /*
1734          *      We cannot trust fsuid as being the "true" uid of the
1735          *      process nor do we know its entire history. We only know it
1736          *      was tainted so we dump it as root in mode 2.
1737          */
1738         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1739                 flag = O_EXCL;          /* Stop rewrite attacks */
1740                 current->fsuid = 0;     /* Dump root private */
1741         }
1742
1743         retval = coredump_wait(exit_code, &core_state);
1744         if (retval < 0)
1745                 goto fail;
1746
1747         /*
1748          * Clear any false indication of pending signals that might
1749          * be seen by the filesystem code called to write the core file.
1750          */
1751         clear_thread_flag(TIF_SIGPENDING);
1752
1753         /*
1754          * lock_kernel() because format_corename() is controlled by sysctl, which
1755          * uses lock_kernel()
1756          */
1757         lock_kernel();
1758         ispipe = format_corename(corename, retval, signr);
1759         unlock_kernel();
1760         /*
1761          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1762          * to a pipe.  Since we're not writing directly to the filesystem
1763          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1764          * created unless the pipe reader choses to write out the core file
1765          * at which point file size limits and permissions will be imposed
1766          * as it does with any other process
1767          */
1768         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1769                 goto fail_unlock;
1770
1771         if (ispipe) {
1772                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1773                 /* Terminate the string before the first option */
1774                 delimit = strchr(corename, ' ');
1775                 if (delimit)
1776                         *delimit = '\0';
1777                 delimit = strrchr(helper_argv[0], '/');
1778                 if (delimit)
1779                         delimit++;
1780                 else
1781                         delimit = helper_argv[0];
1782                 if (!strcmp(delimit, current->comm)) {
1783                         printk(KERN_NOTICE "Recursive core dump detected, "
1784                                         "aborting\n");
1785                         goto fail_unlock;
1786                 }
1787
1788                 core_limit = RLIM_INFINITY;
1789
1790                 /* SIGPIPE can happen, but it's just never processed */
1791                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1792                                 &file)) {
1793                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1794                                corename);
1795                         goto fail_unlock;
1796                 }
1797         } else
1798                 file = filp_open(corename,
1799                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1800                                  0600);
1801         if (IS_ERR(file))
1802                 goto fail_unlock;
1803         inode = file->f_path.dentry->d_inode;
1804         if (inode->i_nlink > 1)
1805                 goto close_fail;        /* multiple links - don't dump */
1806         if (!ispipe && d_unhashed(file->f_path.dentry))
1807                 goto close_fail;
1808
1809         /* AK: actually i see no reason to not allow this for named pipes etc.,
1810            but keep the previous behaviour for now. */
1811         if (!ispipe && !S_ISREG(inode->i_mode))
1812                 goto close_fail;
1813         /*
1814          * Dont allow local users get cute and trick others to coredump
1815          * into their pre-created files:
1816          */
1817         if (inode->i_uid != current->fsuid)
1818                 goto close_fail;
1819         if (!file->f_op)
1820                 goto close_fail;
1821         if (!file->f_op->write)
1822                 goto close_fail;
1823         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1824                 goto close_fail;
1825
1826         retval = binfmt->core_dump(signr, regs, file, core_limit);
1827
1828         if (retval)
1829                 current->signal->group_exit_code |= 0x80;
1830 close_fail:
1831         filp_close(file, NULL);
1832 fail_unlock:
1833         if (helper_argv)
1834                 argv_free(helper_argv);
1835
1836         current->fsuid = fsuid;
1837         coredump_finish(mm);
1838 fail:
1839         return retval;
1840 }