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/swap.h>
  32 #include <linux/string.h>
  33 #include <linux/init.h>
  34 #include <linux/pagemap.h>
  35 #include <linux/perf_event.h>
  36 #include <linux/highmem.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/key.h>
  39 #include <linux/personality.h>
  40 #include <linux/binfmts.h>
  41 #include <linux/utsname.h>
  42 #include <linux/pid_namespace.h>
  43 #include <linux/module.h>
  44 #include <linux/namei.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 #include <linux/kmod.h>
  53 #include <linux/fsnotify.h>
  54 #include <linux/fs_struct.h>
  55 #include <linux/pipe_fs_i.h>
  56 #include <linux/oom.h>
  57 #include <linux/compat.h>
  58
  59 #include <asm/uaccess.h>
  60 #include <asm/mmu_context.h>
  61 #include <asm/tlb.h>
  62 #include "internal.h"
  63
  64 int core_uses_pid;
  65 char core_pattern[CORENAME_MAX_SIZE] = "core";
  66 unsigned int core_pipe_limit;
  67 int suid_dumpable = 0;
  68
  69 struct core_name {
  70         char *corename;
  71         int used, size;
  72 };
  73 static atomic_t call_count = ATOMIC_INIT(1);
  74
  75 /* The maximal length of core_pattern is also specified in sysctl.c */
  76
  77 static LIST_HEAD(formats);
  78 static DEFINE_RWLOCK(binfmt_lock);
  79
  80 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  81 {
  82         if (!fmt)
  83                 return -EINVAL;
  84         write_lock(&binfmt_lock);
  85         insert ? list_add(&fmt->lh, &formats) :
  86                  list_add_tail(&fmt->lh, &formats);
  87         write_unlock(&binfmt_lock);
  88         return 0;
  89 }
  90
  91 EXPORT_SYMBOL(__register_binfmt);
  92
  93 void unregister_binfmt(struct linux_binfmt * fmt)
  94 {
  95         write_lock(&binfmt_lock);
  96         list_del(&fmt->lh);
  97         write_unlock(&binfmt_lock);
  98 }
  99
 100 EXPORT_SYMBOL(unregister_binfmt);
 101
 102 static inline void put_binfmt(struct linux_binfmt * fmt)
 103 {
 104         module_put(fmt->module);
 105 }
 106
 107 /*
 108  * Note that a shared library must be both readable and executable due to
 109  * security reasons.
 110  *
 111  * Also note that we take the address to load from from the file itself.
 112  */
 113 SYSCALL_DEFINE1(uselib, const char __user *, library)
 114 {
 115         struct file *file;
 116         char *tmp = getname(library);
 117         int error = PTR_ERR(tmp);
 118         static const struct open_flags uselib_flags = {
 119                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 120                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 121                 .intent = LOOKUP_OPEN
 122         };
 123
 124         if (IS_ERR(tmp))
 125                 goto out;
 126
 127         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 128         putname(tmp);
 129         error = PTR_ERR(file);
 130         if (IS_ERR(file))
 131                 goto out;
 132
 133         error = -EINVAL;
 134         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 135                 goto exit;
 136
 137         error = -EACCES;
 138         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 139                 goto exit;
 140
 141         fsnotify_open(file);
 142
 143         error = -ENOEXEC;
 144         if(file->f_op) {
 145                 struct linux_binfmt * fmt;
 146
 147                 read_lock(&binfmt_lock);
 148                 list_for_each_entry(fmt, &formats, lh) {
 149                         if (!fmt->load_shlib)
 150                                 continue;
 151                         if (!try_module_get(fmt->module))
 152                                 continue;
 153                         read_unlock(&binfmt_lock);
 154                         error = fmt->load_shlib(file);
 155                         read_lock(&binfmt_lock);
 156                         put_binfmt(fmt);
 157                         if (error != -ENOEXEC)
 158                                 break;
 159                 }
 160                 read_unlock(&binfmt_lock);
 161         }
 162 exit:
 163         fput(file);
 164 out:
 165         return error;
 166 }
 167
 168 #ifdef CONFIG_MMU
 169 /*
 170  * The nascent bprm->mm is not visible until exec_mmap() but it can
 171  * use a lot of memory, account these pages in current->mm temporary
 172  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 173  * change the counter back via acct_arg_size(0).
 174  */
 175 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 176 {
 177         struct mm_struct *mm = current->mm;
 178         long diff = (long)(pages - bprm->vma_pages);
 179
 180         if (!mm || !diff)
 181                 return;
 182
 183         bprm->vma_pages = pages;
 184
 185 #ifdef SPLIT_RSS_COUNTING
 186         add_mm_counter(mm, MM_ANONPAGES, diff);
 187 #else
 188         spin_lock(&mm->page_table_lock);
 189         add_mm_counter(mm, MM_ANONPAGES, diff);
 190         spin_unlock(&mm->page_table_lock);
 191 #endif
 192 }
 193
 194 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 195                 int write)
 196 {
 197         struct page *page;
 198         int ret;
 199
 200 #ifdef CONFIG_STACK_GROWSUP
 201         if (write) {
 202                 ret = expand_downwards(bprm->vma, pos);
 203                 if (ret < 0)
 204                         return NULL;
 205         }
 206 #endif
 207         ret = get_user_pages(current, bprm->mm, pos,
 208                         1, write, 1, &page, NULL);
 209         if (ret <= 0)
 210                 return NULL;
 211
 212         if (write) {
 213                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 214                 struct rlimit *rlim;
 215
 216                 acct_arg_size(bprm, size / PAGE_SIZE);
 217
 218                 /*
 219                  * We've historically supported up to 32 pages (ARG_MAX)
 220                  * of argument strings even with small stacks
 221                  */
 222                 if (size <= ARG_MAX)
 223                         return page;
 224
 225                 /*
 226                  * Limit to 1/4-th the stack size for the argv+env strings.
 227                  * This ensures that:
 228                  *  - the remaining binfmt code will not run out of stack space,
 229                  *  - the program will have a reasonable amount of stack left
 230                  *    to work from.
 231                  */
 232                 rlim = current->signal->rlim;
 233                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 234                         put_page(page);
 235                         return NULL;
 236                 }
 237         }
 238
 239         return page;
 240 }
 241
 242 static void put_arg_page(struct page *page)
 243 {
 244         put_page(page);
 245 }
 246
 247 static void free_arg_page(struct linux_binprm *bprm, int i)
 248 {
 249 }
 250
 251 static void free_arg_pages(struct linux_binprm *bprm)
 252 {
 253 }
 254
 255 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 256                 struct page *page)
 257 {
 258         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 259 }
 260
 261 static int __bprm_mm_init(struct linux_binprm *bprm)
 262 {
 263         int err;
 264         struct vm_area_struct *vma = NULL;
 265         struct mm_struct *mm = bprm->mm;
 266
 267         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 268         if (!vma)
 269                 return -ENOMEM;
 270
 271         down_write(&mm->mmap_sem);
 272         vma->vm_mm = mm;
 273
 274         /*
 275          * Place the stack at the largest stack address the architecture
 276          * supports. Later, we'll move this to an appropriate place. We don't
 277          * use STACK_TOP because that can depend on attributes which aren't
 278          * configured yet.
 279          */
 280         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 281         vma->vm_end = STACK_TOP_MAX;
 282         vma->vm_start = vma->vm_end - PAGE_SIZE;
 283         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 284         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 285         INIT_LIST_HEAD(&vma->anon_vma_chain);
 286
 287         err = insert_vm_struct(mm, vma);
 288         if (err)
 289                 goto err;
 290
 291         mm->stack_vm = mm->total_vm = 1;
 292         up_write(&mm->mmap_sem);
 293         bprm->p = vma->vm_end - sizeof(void *);
 294         return 0;
 295 err:
 296         up_write(&mm->mmap_sem);
 297         bprm->vma = NULL;
 298         kmem_cache_free(vm_area_cachep, vma);
 299         return err;
 300 }
 301
 302 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 303 {
 304         return len <= MAX_ARG_STRLEN;
 305 }
 306
 307 #else
 308
 309 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 310 {
 311 }
 312
 313 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 314                 int write)
 315 {
 316         struct page *page;
 317
 318         page = bprm->page[pos / PAGE_SIZE];
 319         if (!page && write) {
 320                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 321                 if (!page)
 322                         return NULL;
 323                 bprm->page[pos / PAGE_SIZE] = page;
 324         }
 325
 326         return page;
 327 }
 328
 329 static void put_arg_page(struct page *page)
 330 {
 331 }
 332
 333 static void free_arg_page(struct linux_binprm *bprm, int i)
 334 {
 335         if (bprm->page[i]) {
 336                 __free_page(bprm->page[i]);
 337                 bprm->page[i] = NULL;
 338         }
 339 }
 340
 341 static void free_arg_pages(struct linux_binprm *bprm)
 342 {
 343         int i;
 344
 345         for (i = 0; i < MAX_ARG_PAGES; i++)
 346                 free_arg_page(bprm, i);
 347 }
 348
 349 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 350                 struct page *page)
 351 {
 352 }
 353
 354 static int __bprm_mm_init(struct linux_binprm *bprm)
 355 {
 356         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 357         return 0;
 358 }
 359
 360 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 361 {
 362         return len <= bprm->p;
 363 }
 364
 365 #endif /* CONFIG_MMU */
 366
 367 /*
 368  * Create a new mm_struct and populate it with a temporary stack
 369  * vm_area_struct.  We don't have enough context at this point to set the stack
 370  * flags, permissions, and offset, so we use temporary values.  We'll update
 371  * them later in setup_arg_pages().
 372  */
 373 int bprm_mm_init(struct linux_binprm *bprm)
 374 {
 375         int err;
 376         struct mm_struct *mm = NULL;
 377
 378         bprm->mm = mm = mm_alloc();
 379         err = -ENOMEM;
 380         if (!mm)
 381                 goto err;
 382
 383         err = init_new_context(current, mm);
 384         if (err)
 385                 goto err;
 386
 387         err = __bprm_mm_init(bprm);
 388         if (err)
 389                 goto err;
 390
 391         return 0;
 392
 393 err:
 394         if (mm) {
 395                 bprm->mm = NULL;
 396                 mmdrop(mm);
 397         }
 398
 399         return err;
 400 }
 401
 402 struct user_arg_ptr {
 403 #ifdef CONFIG_COMPAT
 404         bool is_compat;
 405 #endif
 406         union {
 407                 const char __user *const __user *native;
 408 #ifdef CONFIG_COMPAT
 409                 compat_uptr_t __user *compat;
 410 #endif
 411         } ptr;
 412 };
 413
 414 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 415 {
 416         const char __user *native;
 417
 418 #ifdef CONFIG_COMPAT
 419         if (unlikely(argv.is_compat)) {
 420                 compat_uptr_t compat;
 421
 422                 if (get_user(compat, argv.ptr.compat + nr))
 423                         return ERR_PTR(-EFAULT);
 424
 425                 return compat_ptr(compat);
 426         }
 427 #endif
 428
 429         if (get_user(native, argv.ptr.native + nr))
 430                 return ERR_PTR(-EFAULT);
 431
 432         return native;
 433 }
 434
 435 /*
 436  * count() counts the number of strings in array ARGV.
 437  */
 438 static int count(struct user_arg_ptr argv, int max)
 439 {
 440         int i = 0;
 441
 442         if (argv.ptr.native != NULL) {
 443                 for (;;) {
 444                         const char __user *p = get_user_arg_ptr(argv, i);
 445
 446                         if (!p)
 447                                 break;
 448
 449                         if (IS_ERR(p))
 450                                 return -EFAULT;
 451
 452                         if (i++ >= max)
 453                                 return -E2BIG;
 454
 455                         if (fatal_signal_pending(current))
 456                                 return -ERESTARTNOHAND;
 457                         cond_resched();
 458                 }
 459         }
 460         return i;
 461 }
 462
 463 /*
 464  * 'copy_strings()' copies argument/environment strings from the old
 465  * processes's memory to the new process's stack.  The call to get_user_pages()
 466  * ensures the destination page is created and not swapped out.
 467  */
 468 static int copy_strings(int argc, struct user_arg_ptr argv,
 469                         struct linux_binprm *bprm)
 470 {
 471         struct page *kmapped_page = NULL;
 472         char *kaddr = NULL;
 473         unsigned long kpos = 0;
 474         int ret;
 475
 476         while (argc-- > 0) {
 477                 const char __user *str;
 478                 int len;
 479                 unsigned long pos;
 480
 481                 ret = -EFAULT;
 482                 str = get_user_arg_ptr(argv, argc);
 483                 if (IS_ERR(str))
 484                         goto out;
 485
 486                 len = strnlen_user(str, MAX_ARG_STRLEN);
 487                 if (!len)
 488                         goto out;
 489
 490                 ret = -E2BIG;
 491                 if (!valid_arg_len(bprm, len))
 492                         goto out;
 493
 494                 /* We're going to work our way backwords. */
 495                 pos = bprm->p;
 496                 str += len;
 497                 bprm->p -= len;
 498
 499                 while (len > 0) {
 500                         int offset, bytes_to_copy;
 501
 502                         if (fatal_signal_pending(current)) {
 503                                 ret = -ERESTARTNOHAND;
 504                                 goto out;
 505                         }
 506                         cond_resched();
 507
 508                         offset = pos % PAGE_SIZE;
 509                         if (offset == 0)
 510                                 offset = PAGE_SIZE;
 511
 512                         bytes_to_copy = offset;
 513                         if (bytes_to_copy > len)
 514                                 bytes_to_copy = len;
 515
 516                         offset -= bytes_to_copy;
 517                         pos -= bytes_to_copy;
 518                         str -= bytes_to_copy;
 519                         len -= bytes_to_copy;
 520
 521                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 522                                 struct page *page;
 523
 524                                 page = get_arg_page(bprm, pos, 1);
 525                                 if (!page) {
 526                                         ret = -E2BIG;
 527                                         goto out;
 528                                 }
 529
 530                                 if (kmapped_page) {
 531                                         flush_kernel_dcache_page(kmapped_page);
 532                                         kunmap(kmapped_page);
 533                                         put_arg_page(kmapped_page);
 534                                 }
 535                                 kmapped_page = page;
 536                                 kaddr = kmap(kmapped_page);
 537                                 kpos = pos & PAGE_MASK;
 538                                 flush_arg_page(bprm, kpos, kmapped_page);
 539                         }
 540                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 541                                 ret = -EFAULT;
 542                                 goto out;
 543                         }
 544                 }
 545         }
 546         ret = 0;
 547 out:
 548         if (kmapped_page) {
 549                 flush_kernel_dcache_page(kmapped_page);
 550                 kunmap(kmapped_page);
 551                 put_arg_page(kmapped_page);
 552         }
 553         return ret;
 554 }
 555
 556 /*
 557  * Like copy_strings, but get argv and its values from kernel memory.
 558  */
 559 int copy_strings_kernel(int argc, const char *const *__argv,
 560                         struct linux_binprm *bprm)
 561 {
 562         int r;
 563         mm_segment_t oldfs = get_fs();
 564         struct user_arg_ptr argv = {
 565                 .ptr.native = (const char __user *const  __user *)__argv,
 566         };
 567
 568         set_fs(KERNEL_DS);
 569         r = copy_strings(argc, argv, bprm);
 570         set_fs(oldfs);
 571
 572         return r;
 573 }
 574 EXPORT_SYMBOL(copy_strings_kernel);
 575
 576 #ifdef CONFIG_MMU
 577
 578 /*
 579  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 580  * the binfmt code determines where the new stack should reside, we shift it to
 581  * its final location.  The process proceeds as follows:
 582  *
 583  * 1) Use shift to calculate the new vma endpoints.
 584  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 585  *    arguments passed to subsequent functions are consistent.
 586  * 3) Move vma's page tables to the new range.
 587  * 4) Free up any cleared pgd range.
 588  * 5) Shrink the vma to cover only the new range.
 589  */
 590 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 591 {
 592         struct mm_struct *mm = vma->vm_mm;
 593         unsigned long old_start = vma->vm_start;
 594         unsigned long old_end = vma->vm_end;
 595         unsigned long length = old_end - old_start;
 596         unsigned long new_start = old_start - shift;
 597         unsigned long new_end = old_end - shift;
 598         struct mmu_gather tlb;
 599
 600         BUG_ON(new_start > new_end);
 601
 602         /*
 603          * ensure there are no vmas between where we want to go
 604          * and where we are
 605          */
 606         if (vma != find_vma(mm, new_start))
 607                 return -EFAULT;
 608
 609         /*
 610          * cover the whole range: [new_start, old_end)
 611          */
 612         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 613                 return -ENOMEM;
 614
 615         /*
 616          * move the page tables downwards, on failure we rely on
 617          * process cleanup to remove whatever mess we made.
 618          */
 619         if (length != move_page_tables(vma, old_start,
 620                                        vma, new_start, length))
 621                 return -ENOMEM;
 622
 623         lru_add_drain();
 624         tlb_gather_mmu(&tlb, mm, 0);
 625         if (new_end > old_start) {
 626                 /*
 627                  * when the old and new regions overlap clear from new_end.
 628                  */
 629                 free_pgd_range(&tlb, new_end, old_end, new_end,
 630                         vma->vm_next ? vma->vm_next->vm_start : 0);
 631         } else {
 632                 /*
 633                  * otherwise, clean from old_start; this is done to not touch
 634                  * the address space in [new_end, old_start) some architectures
 635                  * have constraints on va-space that make this illegal (IA64) -
 636                  * for the others its just a little faster.
 637                  */
 638                 free_pgd_range(&tlb, old_start, old_end, new_end,
 639                         vma->vm_next ? vma->vm_next->vm_start : 0);
 640         }
 641         tlb_finish_mmu(&tlb, new_end, old_end);
 642
 643         /*
 644          * Shrink the vma to just the new range.  Always succeeds.
 645          */
 646         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 647
 648         return 0;
 649 }
 650
 651 /*
 652  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 653  * the stack is optionally relocated, and some extra space is added.
 654  */
 655 int setup_arg_pages(struct linux_binprm *bprm,
 656                     unsigned long stack_top,
 657                     int executable_stack)
 658 {
 659         unsigned long ret;
 660         unsigned long stack_shift;
 661         struct mm_struct *mm = current->mm;
 662         struct vm_area_struct *vma = bprm->vma;
 663         struct vm_area_struct *prev = NULL;
 664         unsigned long vm_flags;
 665         unsigned long stack_base;
 666         unsigned long stack_size;
 667         unsigned long stack_expand;
 668         unsigned long rlim_stack;
 669
 670 #ifdef CONFIG_STACK_GROWSUP
 671         /* Limit stack size to 1GB */
 672         stack_base = rlimit_max(RLIMIT_STACK);
 673         if (stack_base > (1 << 30))
 674                 stack_base = 1 << 30;
 675
 676         /* Make sure we didn't let the argument array grow too large. */
 677         if (vma->vm_end - vma->vm_start > stack_base)
 678                 return -ENOMEM;
 679
 680         stack_base = PAGE_ALIGN(stack_top - stack_base);
 681
 682         stack_shift = vma->vm_start - stack_base;
 683         mm->arg_start = bprm->p - stack_shift;
 684         bprm->p = vma->vm_end - stack_shift;
 685 #else
 686         stack_top = arch_align_stack(stack_top);
 687         stack_top = PAGE_ALIGN(stack_top);
 688
 689         if (unlikely(stack_top < mmap_min_addr) ||
 690             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 691                 return -ENOMEM;
 692
 693         stack_shift = vma->vm_end - stack_top;
 694
 695         bprm->p -= stack_shift;
 696         mm->arg_start = bprm->p;
 697 #endif
 698
 699         if (bprm->loader)
 700                 bprm->loader -= stack_shift;
 701         bprm->exec -= stack_shift;
 702
 703         down_write(&mm->mmap_sem);
 704         vm_flags = VM_STACK_FLAGS;
 705
 706         /*
 707          * Adjust stack execute permissions; explicitly enable for
 708          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 709          * (arch default) otherwise.
 710          */
 711         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 712                 vm_flags |= VM_EXEC;
 713         else if (executable_stack == EXSTACK_DISABLE_X)
 714                 vm_flags &= ~VM_EXEC;
 715         vm_flags |= mm->def_flags;
 716         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 717
 718         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 719                         vm_flags);
 720         if (ret)
 721                 goto out_unlock;
 722         BUG_ON(prev != vma);
 723
 724         /* Move stack pages down in memory. */
 725         if (stack_shift) {
 726                 ret = shift_arg_pages(vma, stack_shift);
 727                 if (ret)
 728                         goto out_unlock;
 729         }
 730
 731         /* mprotect_fixup is overkill to remove the temporary stack flags */
 732         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 733
 734         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 735         stack_size = vma->vm_end - vma->vm_start;
 736         /*
 737          * Align this down to a page boundary as expand_stack
 738          * will align it up.
 739          */
 740         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 741 #ifdef CONFIG_STACK_GROWSUP
 742         if (stack_size + stack_expand > rlim_stack)
 743                 stack_base = vma->vm_start + rlim_stack;
 744         else
 745                 stack_base = vma->vm_end + stack_expand;
 746 #else
 747         if (stack_size + stack_expand > rlim_stack)
 748                 stack_base = vma->vm_end - rlim_stack;
 749         else
 750                 stack_base = vma->vm_start - stack_expand;
 751 #endif
 752         current->mm->start_stack = bprm->p;
 753         ret = expand_stack(vma, stack_base);
 754         if (ret)
 755                 ret = -EFAULT;
 756
 757 out_unlock:
 758         up_write(&mm->mmap_sem);
 759         return ret;
 760 }
 761 EXPORT_SYMBOL(setup_arg_pages);
 762
 763 #endif /* CONFIG_MMU */
 764
 765 struct file *open_exec(const char *name)
 766 {
 767         struct file *file;
 768         int err;
 769         static const struct open_flags open_exec_flags = {
 770                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 771                 .acc_mode = MAY_EXEC | MAY_OPEN,
 772                 .intent = LOOKUP_OPEN
 773         };
 774
 775         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 776         if (IS_ERR(file))
 777                 goto out;
 778
 779         err = -EACCES;
 780         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 781                 goto exit;
 782
 783         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 784                 goto exit;
 785
 786         fsnotify_open(file);
 787
 788         err = deny_write_access(file);
 789         if (err)
 790                 goto exit;
 791
 792 out:
 793         return file;
 794
 795 exit:
 796         fput(file);
 797         return ERR_PTR(err);
 798 }
 799 EXPORT_SYMBOL(open_exec);
 800
 801 int kernel_read(struct file *file, loff_t offset,
 802                 char *addr, unsigned long count)
 803 {
 804         mm_segment_t old_fs;
 805         loff_t pos = offset;
 806         int result;
 807
 808         old_fs = get_fs();
 809         set_fs(get_ds());
 810         /* The cast to a user pointer is valid due to the set_fs() */
 811         result = vfs_read(file, (void __user *)addr, count, &pos);
 812         set_fs(old_fs);
 813         return result;
 814 }
 815
 816 EXPORT_SYMBOL(kernel_read);
 817
 818 static int exec_mmap(struct mm_struct *mm)
 819 {
 820         struct task_struct *tsk;
 821         struct mm_struct * old_mm, *active_mm;
 822
 823         /* Notify parent that we're no longer interested in the old VM */
 824         tsk = current;
 825         old_mm = current->mm;
 826         sync_mm_rss(tsk, old_mm);
 827         mm_release(tsk, old_mm);
 828
 829         if (old_mm) {
 830                 /*
 831                  * Make sure that if there is a core dump in progress
 832                  * for the old mm, we get out and die instead of going
 833                  * through with the exec.  We must hold mmap_sem around
 834                  * checking core_state and changing tsk->mm.
 835                  */
 836                 down_read(&old_mm->mmap_sem);
 837                 if (unlikely(old_mm->core_state)) {
 838                         up_read(&old_mm->mmap_sem);
 839                         return -EINTR;
 840                 }
 841         }
 842         task_lock(tsk);
 843         active_mm = tsk->active_mm;
 844         tsk->mm = mm;
 845         tsk->active_mm = mm;
 846         activate_mm(active_mm, mm);
 847         if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
 848                 atomic_dec(&old_mm->oom_disable_count);
 849                 atomic_inc(&tsk->mm->oom_disable_count);
 850         }
 851         task_unlock(tsk);
 852         arch_pick_mmap_layout(mm);
 853         if (old_mm) {
 854                 up_read(&old_mm->mmap_sem);
 855                 BUG_ON(active_mm != old_mm);
 856                 mm_update_next_owner(old_mm);
 857                 mmput(old_mm);
 858                 return 0;
 859         }
 860         mmdrop(active_mm);
 861         return 0;
 862 }
 863
 864 /*
 865  * This function makes sure the current process has its own signal table,
 866  * so that flush_signal_handlers can later reset the handlers without
 867  * disturbing other processes.  (Other processes might share the signal
 868  * table via the CLONE_SIGHAND option to clone().)
 869  */
 870 static int de_thread(struct task_struct *tsk)
 871 {
 872         struct signal_struct *sig = tsk->signal;
 873         struct sighand_struct *oldsighand = tsk->sighand;
 874         spinlock_t *lock = &oldsighand->siglock;
 875
 876         if (thread_group_empty(tsk))
 877                 goto no_thread_group;
 878
 879         /*
 880          * Kill all other threads in the thread group.
 881          */
 882         spin_lock_irq(lock);
 883         if (signal_group_exit(sig)) {
 884                 /*
 885                  * Another group action in progress, just
 886                  * return so that the signal is processed.
 887                  */
 888                 spin_unlock_irq(lock);
 889                 return -EAGAIN;
 890         }
 891
 892         sig->group_exit_task = tsk;
 893         sig->notify_count = zap_other_threads(tsk);
 894         if (!thread_group_leader(tsk))
 895                 sig->notify_count--;
 896
 897         while (sig->notify_count) {
 898                 __set_current_state(TASK_UNINTERRUPTIBLE);
 899                 spin_unlock_irq(lock);
 900                 schedule();
 901                 spin_lock_irq(lock);
 902         }
 903         spin_unlock_irq(lock);
 904
 905         /*
 906          * At this point all other threads have exited, all we have to
 907          * do is to wait for the thread group leader to become inactive,
 908          * and to assume its PID:
 909          */
 910         if (!thread_group_leader(tsk)) {
 911                 struct task_struct *leader = tsk->group_leader;
 912
 913                 sig->notify_count = -1; /* for exit_notify() */
 914                 for (;;) {
 915                         write_lock_irq(&tasklist_lock);
 916                         if (likely(leader->exit_state))
 917                                 break;
 918                         __set_current_state(TASK_UNINTERRUPTIBLE);
 919                         write_unlock_irq(&tasklist_lock);
 920                         schedule();
 921                 }
 922
 923                 /*
 924                  * The only record we have of the real-time age of a
 925                  * process, regardless of execs it's done, is start_time.
 926                  * All the past CPU time is accumulated in signal_struct
 927                  * from sister threads now dead.  But in this non-leader
 928                  * exec, nothing survives from the original leader thread,
 929                  * whose birth marks the true age of this process now.
 930                  * When we take on its identity by switching to its PID, we
 931                  * also take its birthdate (always earlier than our own).
 932                  */
 933                 tsk->start_time = leader->start_time;
 934
 935                 BUG_ON(!same_thread_group(leader, tsk));
 936                 BUG_ON(has_group_leader_pid(tsk));
 937                 /*
 938                  * An exec() starts a new thread group with the
 939                  * TGID of the previous thread group. Rehash the
 940                  * two threads with a switched PID, and release
 941                  * the former thread group leader:
 942                  */
 943
 944                 /* Become a process group leader with the old leader's pid.
 945                  * The old leader becomes a thread of the this thread group.
 946                  * Note: The old leader also uses this pid until release_task
 947                  *       is called.  Odd but simple and correct.
 948                  */
 949                 detach_pid(tsk, PIDTYPE_PID);
 950                 tsk->pid = leader->pid;
 951                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 952                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 953                 transfer_pid(leader, tsk, PIDTYPE_SID);
 954
 955                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 956                 list_replace_init(&leader->sibling, &tsk->sibling);
 957
 958                 tsk->group_leader = tsk;
 959                 leader->group_leader = tsk;
 960
 961                 tsk->exit_signal = SIGCHLD;
 962
 963                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 964                 leader->exit_state = EXIT_DEAD;
 965                 write_unlock_irq(&tasklist_lock);
 966
 967                 release_task(leader);
 968         }
 969
 970         sig->group_exit_task = NULL;
 971         sig->notify_count = 0;
 972
 973 no_thread_group:
 974         if (current->mm)
 975                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 976
 977         exit_itimers(sig);
 978         flush_itimer_signals();
 979
 980         if (atomic_read(&oldsighand->count) != 1) {
 981                 struct sighand_struct *newsighand;
 982                 /*
 983                  * This ->sighand is shared with the CLONE_SIGHAND
 984                  * but not CLONE_THREAD task, switch to the new one.
 985                  */
 986                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 987                 if (!newsighand)
 988                         return -ENOMEM;
 989
 990                 atomic_set(&newsighand->count, 1);
 991                 memcpy(newsighand->action, oldsighand->action,
 992                        sizeof(newsighand->action));
 993
 994                 write_lock_irq(&tasklist_lock);
 995                 spin_lock(&oldsighand->siglock);
 996                 rcu_assign_pointer(tsk->sighand, newsighand);
 997                 spin_unlock(&oldsighand->siglock);
 998                 write_unlock_irq(&tasklist_lock);
 999
1000                 __cleanup_sighand(oldsighand);
1001         }
1002
1003         BUG_ON(!thread_group_leader(tsk));
1004         return 0;
1005 }
1006
1007 /*
1008  * These functions flushes out all traces of the currently running executable
1009  * so that a new one can be started
1010  */
1011 static void flush_old_files(struct files_struct * files)
1012 {
1013         long j = -1;
1014         struct fdtable *fdt;
1015
1016         spin_lock(&files->file_lock);
1017         for (;;) {
1018                 unsigned long set, i;
1019
1020                 j++;
1021                 i = j * __NFDBITS;
1022                 fdt = files_fdtable(files);
1023                 if (i >= fdt->max_fds)
1024                         break;
1025                 set = fdt->close_on_exec->fds_bits[j];
1026                 if (!set)
1027                         continue;
1028                 fdt->close_on_exec->fds_bits[j] = 0;
1029                 spin_unlock(&files->file_lock);
1030                 for ( ; set ; i++,set >>= 1) {
1031                         if (set & 1) {
1032                                 sys_close(i);
1033                         }
1034                 }
1035                 spin_lock(&files->file_lock);
1036
1037         }
1038         spin_unlock(&files->file_lock);
1039 }
1040
1041 char *get_task_comm(char *buf, struct task_struct *tsk)
1042 {
1043         /* buf must be at least sizeof(tsk->comm) in size */
1044         task_lock(tsk);
1045         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1046         task_unlock(tsk);
1047         return buf;
1048 }
1049 EXPORT_SYMBOL_GPL(get_task_comm);
1050
1051 void set_task_comm(struct task_struct *tsk, char *buf)
1052 {
1053         task_lock(tsk);
1054
1055         /*
1056          * Threads may access current->comm without holding
1057          * the task lock, so write the string carefully.
1058          * Readers without a lock may see incomplete new
1059          * names but are safe from non-terminating string reads.
1060          */
1061         memset(tsk->comm, 0, TASK_COMM_LEN);
1062         wmb();
1063         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1064         task_unlock(tsk);
1065         perf_event_comm(tsk);
1066 }
1067
1068 int flush_old_exec(struct linux_binprm * bprm)
1069 {
1070         int retval;
1071
1072         /*
1073          * Make sure we have a private signal table and that
1074          * we are unassociated from the previous thread group.
1075          */
1076         retval = de_thread(current);
1077         if (retval)
1078                 goto out;
1079
1080         set_mm_exe_file(bprm->mm, bprm->file);
1081
1082         /*
1083          * Release all of the old mmap stuff
1084          */
1085         acct_arg_size(bprm, 0);
1086         retval = exec_mmap(bprm->mm);
1087         if (retval)
1088                 goto out;
1089
1090         bprm->mm = NULL;                /* We're using it now */
1091
1092         set_fs(USER_DS);
1093         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1094         flush_thread();
1095         current->personality &= ~bprm->per_clear;
1096
1097         return 0;
1098
1099 out:
1100         return retval;
1101 }
1102 EXPORT_SYMBOL(flush_old_exec);
1103
1104 void setup_new_exec(struct linux_binprm * bprm)
1105 {
1106         int i, ch;
1107         const char *name;
1108         char tcomm[sizeof(current->comm)];
1109
1110         arch_pick_mmap_layout(current->mm);
1111
1112         /* This is the point of no return */
1113         current->sas_ss_sp = current->sas_ss_size = 0;
1114
1115         if (current_euid() == current_uid() && current_egid() == current_gid())
1116                 set_dumpable(current->mm, 1);
1117         else
1118                 set_dumpable(current->mm, suid_dumpable);
1119
1120         name = bprm->filename;
1121
1122         /* Copies the binary name from after last slash */
1123         for (i=0; (ch = *(name++)) != '\0';) {
1124                 if (ch == '/')
1125                         i = 0; /* overwrite what we wrote */
1126                 else
1127                         if (i < (sizeof(tcomm) - 1))
1128                                 tcomm[i++] = ch;
1129         }
1130         tcomm[i] = '\0';
1131         set_task_comm(current, tcomm);
1132
1133         /* Set the new mm task size. We have to do that late because it may
1134          * depend on TIF_32BIT which is only updated in flush_thread() on
1135          * some architectures like powerpc
1136          */
1137         current->mm->task_size = TASK_SIZE;
1138
1139         /* install the new credentials */
1140         if (bprm->cred->uid != current_euid() ||
1141             bprm->cred->gid != current_egid()) {
1142                 current->pdeath_signal = 0;
1143         } else if (file_permission(bprm->file, MAY_READ) ||
1144                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1145                 set_dumpable(current->mm, suid_dumpable);
1146         }
1147
1148         /*
1149          * Flush performance counters when crossing a
1150          * security domain:
1151          */
1152         if (!get_dumpable(current->mm))
1153                 perf_event_exit_task(current);
1154
1155         /* An exec changes our domain. We are no longer part of the thread
1156            group */
1157
1158         current->self_exec_id++;
1159
1160         flush_signal_handlers(current, 0);
1161         flush_old_files(current->files);
1162 }
1163 EXPORT_SYMBOL(setup_new_exec);
1164
1165 /*
1166  * Prepare credentials and lock ->cred_guard_mutex.
1167  * install_exec_creds() commits the new creds and drops the lock.
1168  * Or, if exec fails before, free_bprm() should release ->cred and
1169  * and unlock.
1170  */
1171 int prepare_bprm_creds(struct linux_binprm *bprm)
1172 {
1173         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1174                 return -ERESTARTNOINTR;
1175
1176         bprm->cred = prepare_exec_creds();
1177         if (likely(bprm->cred))
1178                 return 0;
1179
1180         mutex_unlock(&current->signal->cred_guard_mutex);
1181         return -ENOMEM;
1182 }
1183
1184 void free_bprm(struct linux_binprm *bprm)
1185 {
1186         free_arg_pages(bprm);
1187         if (bprm->cred) {
1188                 mutex_unlock(&current->signal->cred_guard_mutex);
1189                 abort_creds(bprm->cred);
1190         }
1191         kfree(bprm);
1192 }
1193
1194 /*
1195  * install the new credentials for this executable
1196  */
1197 void install_exec_creds(struct linux_binprm *bprm)
1198 {
1199         security_bprm_committing_creds(bprm);
1200
1201         commit_creds(bprm->cred);
1202         bprm->cred = NULL;
1203         /*
1204          * cred_guard_mutex must be held at least to this point to prevent
1205          * ptrace_attach() from altering our determination of the task's
1206          * credentials; any time after this it may be unlocked.
1207          */
1208         security_bprm_committed_creds(bprm);
1209         mutex_unlock(&current->signal->cred_guard_mutex);
1210 }
1211 EXPORT_SYMBOL(install_exec_creds);
1212
1213 /*
1214  * determine how safe it is to execute the proposed program
1215  * - the caller must hold ->cred_guard_mutex to protect against
1216  *   PTRACE_ATTACH
1217  */
1218 int check_unsafe_exec(struct linux_binprm *bprm)
1219 {
1220         struct task_struct *p = current, *t;
1221         unsigned n_fs;
1222         int res = 0;
1223
1224         bprm->unsafe = tracehook_unsafe_exec(p);
1225
1226         n_fs = 1;
1227         spin_lock(&p->fs->lock);
1228         rcu_read_lock();
1229         for (t = next_thread(p); t != p; t = next_thread(t)) {
1230                 if (t->fs == p->fs)
1231                         n_fs++;
1232         }
1233         rcu_read_unlock();
1234
1235         if (p->fs->users > n_fs) {
1236                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1237         } else {
1238                 res = -EAGAIN;
1239                 if (!p->fs->in_exec) {
1240                         p->fs->in_exec = 1;
1241                         res = 1;
1242                 }
1243         }
1244         spin_unlock(&p->fs->lock);
1245
1246         return res;
1247 }
1248
1249 /*
1250  * Fill the binprm structure from the inode.
1251  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1252  *
1253  * This may be called multiple times for binary chains (scripts for example).
1254  */
1255 int prepare_binprm(struct linux_binprm *bprm)
1256 {
1257         umode_t mode;
1258         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1259         int retval;
1260
1261         mode = inode->i_mode;
1262         if (bprm->file->f_op == NULL)
1263                 return -EACCES;
1264
1265         /* clear any previous set[ug]id data from a previous binary */
1266         bprm->cred->euid = current_euid();
1267         bprm->cred->egid = current_egid();
1268
1269         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1270                 /* Set-uid? */
1271                 if (mode & S_ISUID) {
1272                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1273                         bprm->cred->euid = inode->i_uid;
1274                 }
1275
1276                 /* Set-gid? */
1277                 /*
1278                  * If setgid is set but no group execute bit then this
1279                  * is a candidate for mandatory locking, not a setgid
1280                  * executable.
1281                  */
1282                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1283                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1284                         bprm->cred->egid = inode->i_gid;
1285                 }
1286         }
1287
1288         /* fill in binprm security blob */
1289         retval = security_bprm_set_creds(bprm);
1290         if (retval)
1291                 return retval;
1292         bprm->cred_prepared = 1;
1293
1294         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1295         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1296 }
1297
1298 EXPORT_SYMBOL(prepare_binprm);
1299
1300 /*
1301  * Arguments are '\0' separated strings found at the location bprm->p
1302  * points to; chop off the first by relocating brpm->p to right after
1303  * the first '\0' encountered.
1304  */
1305 int remove_arg_zero(struct linux_binprm *bprm)
1306 {
1307         int ret = 0;
1308         unsigned long offset;
1309         char *kaddr;
1310         struct page *page;
1311
1312         if (!bprm->argc)
1313                 return 0;
1314
1315         do {
1316                 offset = bprm->p & ~PAGE_MASK;
1317                 page = get_arg_page(bprm, bprm->p, 0);
1318                 if (!page) {
1319                         ret = -EFAULT;
1320                         goto out;
1321                 }
1322                 kaddr = kmap_atomic(page, KM_USER0);
1323
1324                 for (; offset < PAGE_SIZE && kaddr[offset];
1325                                 offset++, bprm->p++)
1326                         ;
1327
1328                 kunmap_atomic(kaddr, KM_USER0);
1329                 put_arg_page(page);
1330
1331                 if (offset == PAGE_SIZE)
1332                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1333         } while (offset == PAGE_SIZE);
1334
1335         bprm->p++;
1336         bprm->argc--;
1337         ret = 0;
1338
1339 out:
1340         return ret;
1341 }
1342 EXPORT_SYMBOL(remove_arg_zero);
1343
1344 /*
1345  * cycle the list of binary formats handler, until one recognizes the image
1346  */
1347 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1348 {
1349         unsigned int depth = bprm->recursion_depth;
1350         int try,retval;
1351         struct linux_binfmt *fmt;
1352
1353         retval = security_bprm_check(bprm);
1354         if (retval)
1355                 return retval;
1356
1357         retval = audit_bprm(bprm);
1358         if (retval)
1359                 return retval;
1360
1361         retval = -ENOENT;
1362         for (try=0; try<2; try++) {
1363                 read_lock(&binfmt_lock);
1364                 list_for_each_entry(fmt, &formats, lh) {
1365                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1366                         if (!fn)
1367                                 continue;
1368                         if (!try_module_get(fmt->module))
1369                                 continue;
1370                         read_unlock(&binfmt_lock);
1371                         retval = fn(bprm, regs);
1372                         /*
1373                          * Restore the depth counter to its starting value
1374                          * in this call, so we don't have to rely on every
1375                          * load_binary function to restore it on return.
1376                          */
1377                         bprm->recursion_depth = depth;
1378                         if (retval >= 0) {
1379                                 if (depth == 0)
1380                                         tracehook_report_exec(fmt, bprm, regs);
1381                                 put_binfmt(fmt);
1382                                 allow_write_access(bprm->file);
1383                                 if (bprm->file)
1384                                         fput(bprm->file);
1385                                 bprm->file = NULL;
1386                                 current->did_exec = 1;
1387                                 proc_exec_connector(current);
1388                                 return retval;
1389                         }
1390                         read_lock(&binfmt_lock);
1391                         put_binfmt(fmt);
1392                         if (retval != -ENOEXEC || bprm->mm == NULL)
1393                                 break;
1394                         if (!bprm->file) {
1395                                 read_unlock(&binfmt_lock);
1396                                 return retval;
1397                         }
1398                 }
1399                 read_unlock(&binfmt_lock);
1400                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1401                         break;
1402 #ifdef CONFIG_MODULES
1403                 } else {
1404 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1405                         if (printable(bprm->buf[0]) &&
1406                             printable(bprm->buf[1]) &&
1407                             printable(bprm->buf[2]) &&
1408                             printable(bprm->buf[3]))
1409                                 break; /* -ENOEXEC */
1410                         if (try)
1411                                 break; /* -ENOEXEC */
1412                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1413 #endif
1414                 }
1415         }
1416         return retval;
1417 }
1418
1419 EXPORT_SYMBOL(search_binary_handler);
1420
1421 /*
1422  * sys_execve() executes a new program.
1423  */
1424 static int do_execve_common(const char *filename,
1425                                 struct user_arg_ptr argv,
1426                                 struct user_arg_ptr envp,
1427                                 struct pt_regs *regs)
1428 {
1429         struct linux_binprm *bprm;
1430         struct file *file;
1431         struct files_struct *displaced;
1432         bool clear_in_exec;
1433         int retval;
1434
1435         retval = unshare_files(&displaced);
1436         if (retval)
1437                 goto out_ret;
1438
1439         retval = -ENOMEM;
1440         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1441         if (!bprm)
1442                 goto out_files;
1443
1444         retval = prepare_bprm_creds(bprm);
1445         if (retval)
1446                 goto out_free;
1447
1448         retval = check_unsafe_exec(bprm);
1449         if (retval < 0)
1450                 goto out_free;
1451         clear_in_exec = retval;
1452         current->in_execve = 1;
1453
1454         file = open_exec(filename);
1455         retval = PTR_ERR(file);
1456         if (IS_ERR(file))
1457                 goto out_unmark;
1458
1459         sched_exec();
1460
1461         bprm->file = file;
1462         bprm->filename = filename;
1463         bprm->interp = filename;
1464
1465         retval = bprm_mm_init(bprm);
1466         if (retval)
1467                 goto out_file;
1468
1469         bprm->argc = count(argv, MAX_ARG_STRINGS);
1470         if ((retval = bprm->argc) < 0)
1471                 goto out;
1472
1473         bprm->envc = count(envp, MAX_ARG_STRINGS);
1474         if ((retval = bprm->envc) < 0)
1475                 goto out;
1476
1477         retval = prepare_binprm(bprm);
1478         if (retval < 0)
1479                 goto out;
1480
1481         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1482         if (retval < 0)
1483                 goto out;
1484
1485         bprm->exec = bprm->p;
1486         retval = copy_strings(bprm->envc, envp, bprm);
1487         if (retval < 0)
1488                 goto out;
1489
1490         retval = copy_strings(bprm->argc, argv, bprm);
1491         if (retval < 0)
1492                 goto out;
1493
1494         retval = search_binary_handler(bprm,regs);
1495         if (retval < 0)
1496                 goto out;
1497
1498         /* execve succeeded */
1499         current->fs->in_exec = 0;
1500         current->in_execve = 0;
1501         acct_update_integrals(current);
1502         free_bprm(bprm);
1503         if (displaced)
1504                 put_files_struct(displaced);
1505         return retval;
1506
1507 out:
1508         if (bprm->mm) {
1509                 acct_arg_size(bprm, 0);
1510                 mmput(bprm->mm);
1511         }
1512
1513 out_file:
1514         if (bprm->file) {
1515                 allow_write_access(bprm->file);
1516                 fput(bprm->file);
1517         }
1518
1519 out_unmark:
1520         if (clear_in_exec)
1521                 current->fs->in_exec = 0;
1522         current->in_execve = 0;
1523
1524 out_free:
1525         free_bprm(bprm);
1526
1527 out_files:
1528         if (displaced)
1529                 reset_files_struct(displaced);
1530 out_ret:
1531         return retval;
1532 }
1533
1534 int do_execve(const char *filename,
1535         const char __user *const __user *__argv,
1536         const char __user *const __user *__envp,
1537         struct pt_regs *regs)
1538 {
1539         struct user_arg_ptr argv = { .ptr.native = __argv };
1540         struct user_arg_ptr envp = { .ptr.native = __envp };
1541         return do_execve_common(filename, argv, envp, regs);
1542 }
1543
1544 #ifdef CONFIG_COMPAT
1545 int compat_do_execve(char *filename,
1546         compat_uptr_t __user *__argv,
1547         compat_uptr_t __user *__envp,
1548         struct pt_regs *regs)
1549 {
1550         struct user_arg_ptr argv = {
1551                 .is_compat = true,
1552                 .ptr.compat = __argv,
1553         };
1554         struct user_arg_ptr envp = {
1555                 .is_compat = true,
1556                 .ptr.compat = __envp,
1557         };
1558         return do_execve_common(filename, argv, envp, regs);
1559 }
1560 #endif
1561
1562 void set_binfmt(struct linux_binfmt *new)
1563 {
1564         struct mm_struct *mm = current->mm;
1565
1566         if (mm->binfmt)
1567                 module_put(mm->binfmt->module);
1568
1569         mm->binfmt = new;
1570         if (new)
1571                 __module_get(new->module);
1572 }
1573
1574 EXPORT_SYMBOL(set_binfmt);
1575
1576 static int expand_corename(struct core_name *cn)
1577 {
1578         char *old_corename = cn->corename;
1579
1580         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1581         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1582
1583         if (!cn->corename) {
1584                 kfree(old_corename);
1585                 return -ENOMEM;
1586         }
1587
1588         return 0;
1589 }
1590
1591 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1592 {
1593         char *cur;
1594         int need;
1595         int ret;
1596         va_list arg;
1597
1598         va_start(arg, fmt);
1599         need = vsnprintf(NULL, 0, fmt, arg);
1600         va_end(arg);
1601
1602         if (likely(need < cn->size - cn->used - 1))
1603                 goto out_printf;
1604
1605         ret = expand_corename(cn);
1606         if (ret)
1607                 goto expand_fail;
1608
1609 out_printf:
1610         cur = cn->corename + cn->used;
1611         va_start(arg, fmt);
1612         vsnprintf(cur, need + 1, fmt, arg);
1613         va_end(arg);
1614         cn->used += need;
1615         return 0;
1616
1617 expand_fail:
1618         return ret;
1619 }
1620
1621 static int cn_print_exe_file(struct core_name *cn)
1622 {
1623         struct file *exe_file;
1624         char *pathbuf, *path, *p;
1625         int ret;
1626
1627         exe_file = get_mm_exe_file(current->mm);
1628         if (!exe_file)
1629                 return cn_printf(cn, "(unknown)");
1630
1631         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1632         if (!pathbuf) {
1633                 ret = -ENOMEM;
1634                 goto put_exe_file;
1635         }
1636
1637         path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1638         if (IS_ERR(path)) {
1639                 ret = PTR_ERR(path);
1640                 goto free_buf;
1641         }
1642
1643         for (p = path; *p; p++)
1644                 if (*p == '/')
1645                         *p = '!';
1646
1647         ret = cn_printf(cn, "%s", path);
1648
1649 free_buf:
1650         kfree(pathbuf);
1651 put_exe_file:
1652         fput(exe_file);
1653         return ret;
1654 }
1655
1656 /* format_corename will inspect the pattern parameter, and output a
1657  * name into corename, which must have space for at least
1658  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1659  */
1660 static int format_corename(struct core_name *cn, long signr)
1661 {
1662         const struct cred *cred = current_cred();
1663         const char *pat_ptr = core_pattern;
1664         int ispipe = (*pat_ptr == '|');
1665         int pid_in_pattern = 0;
1666         int err = 0;
1667
1668         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1669         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1670         cn->used = 0;
1671
1672         if (!cn->corename)
1673                 return -ENOMEM;
1674
1675         /* Repeat as long as we have more pattern to process and more output
1676            space */
1677         while (*pat_ptr) {
1678                 if (*pat_ptr != '%') {
1679                         if (*pat_ptr == 0)
1680                                 goto out;
1681                         err = cn_printf(cn, "%c", *pat_ptr++);
1682                 } else {
1683                         switch (*++pat_ptr) {
1684                         /* single % at the end, drop that */
1685                         case 0:
1686                                 goto out;
1687                         /* Double percent, output one percent */
1688                         case '%':
1689                                 err = cn_printf(cn, "%c", '%');
1690                                 break;
1691                         /* pid */
1692                         case 'p':
1693                                 pid_in_pattern = 1;
1694                                 err = cn_printf(cn, "%d",
1695                                               task_tgid_vnr(current));
1696                                 break;
1697                         /* uid */
1698                         case 'u':
1699                                 err = cn_printf(cn, "%d", cred->uid);
1700                                 break;
1701                         /* gid */
1702                         case 'g':
1703                                 err = cn_printf(cn, "%d", cred->gid);
1704                                 break;
1705                         /* signal that caused the coredump */
1706                         case 's':
1707                                 err = cn_printf(cn, "%ld", signr);
1708                                 break;
1709                         /* UNIX time of coredump */
1710                         case 't': {
1711                                 struct timeval tv;
1712                                 do_gettimeofday(&tv);
1713                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1714                                 break;
1715                         }
1716                         /* hostname */
1717                         case 'h':
1718                                 down_read(&uts_sem);
1719                                 err = cn_printf(cn, "%s",
1720                                               utsname()->nodename);
1721                                 up_read(&uts_sem);
1722                                 break;
1723                         /* executable */
1724                         case 'e':
1725                                 err = cn_printf(cn, "%s", current->comm);
1726                                 break;
1727                         case 'E':
1728                                 err = cn_print_exe_file(cn);
1729                                 break;
1730                         /* core limit size */
1731                         case 'c':
1732                                 err = cn_printf(cn, "%lu",
1733                                               rlimit(RLIMIT_CORE));
1734                                 break;
1735                         default:
1736                                 break;
1737                         }
1738                         ++pat_ptr;
1739                 }
1740
1741                 if (err)
1742                         return err;
1743         }
1744
1745         /* Backward compatibility with core_uses_pid:
1746          *
1747          * If core_pattern does not include a %p (as is the default)
1748          * and core_uses_pid is set, then .%pid will be appended to
1749          * the filename. Do not do this for piped commands. */
1750         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1751                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1752                 if (err)
1753                         return err;
1754         }
1755 out:
1756         return ispipe;
1757 }
1758
1759 static int zap_process(struct task_struct *start, int exit_code)
1760 {
1761         struct task_struct *t;
1762         int nr = 0;
1763
1764         start->signal->flags = SIGNAL_GROUP_EXIT;
1765         start->signal->group_exit_code = exit_code;
1766         start->signal->group_stop_count = 0;
1767
1768         t = start;
1769         do {
1770                 task_clear_group_stop_pending(t);
1771                 if (t != current && t->mm) {
1772                         sigaddset(&t->pending.signal, SIGKILL);
1773                         signal_wake_up(t, 1);
1774                         nr++;
1775                 }
1776         } while_each_thread(start, t);
1777
1778         return nr;
1779 }
1780
1781 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1782                                 struct core_state *core_state, int exit_code)
1783 {
1784         struct task_struct *g, *p;
1785         unsigned long flags;
1786         int nr = -EAGAIN;
1787
1788         spin_lock_irq(&tsk->sighand->siglock);
1789         if (!signal_group_exit(tsk->signal)) {
1790                 mm->core_state = core_state;
1791                 nr = zap_process(tsk, exit_code);
1792         }
1793         spin_unlock_irq(&tsk->sighand->siglock);
1794         if (unlikely(nr < 0))
1795                 return nr;
1796
1797         if (atomic_read(&mm->mm_users) == nr + 1)
1798                 goto done;
1799         /*
1800          * We should find and kill all tasks which use this mm, and we should
1801          * count them correctly into ->nr_threads. We don't take tasklist
1802          * lock, but this is safe wrt:
1803          *
1804          * fork:
1805          *      None of sub-threads can fork after zap_process(leader). All
1806          *      processes which were created before this point should be
1807          *      visible to zap_threads() because copy_process() adds the new
1808          *      process to the tail of init_task.tasks list, and lock/unlock
1809          *      of ->siglock provides a memory barrier.
1810          *
1811          * do_exit:
1812          *      The caller holds mm->mmap_sem. This means that the task which
1813          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1814          *      its ->mm.
1815          *
1816          * de_thread:
1817          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1818          *      we must see either old or new leader, this does not matter.
1819          *      However, it can change p->sighand, so lock_task_sighand(p)
1820          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1821          *      it can't fail.
1822          *
1823          *      Note also that "g" can be the old leader with ->mm == NULL
1824          *      and already unhashed and thus removed from ->thread_group.
1825          *      This is OK, __unhash_process()->list_del_rcu() does not
1826          *      clear the ->next pointer, we will find the new leader via
1827          *      next_thread().
1828          */
1829         rcu_read_lock();
1830         for_each_process(g) {
1831                 if (g == tsk->group_leader)
1832                         continue;
1833                 if (g->flags & PF_KTHREAD)
1834                         continue;
1835                 p = g;
1836                 do {
1837                         if (p->mm) {
1838                                 if (unlikely(p->mm == mm)) {
1839                                         lock_task_sighand(p, &flags);
1840                                         nr += zap_process(p, exit_code);
1841                                         unlock_task_sighand(p, &flags);
1842                                 }
1843                                 break;
1844                         }
1845                 } while_each_thread(g, p);
1846         }
1847         rcu_read_unlock();
1848 done:
1849         atomic_set(&core_state->nr_threads, nr);
1850         return nr;
1851 }
1852
1853 static int coredump_wait(int exit_code, struct core_state *core_state)
1854 {
1855         struct task_struct *tsk = current;
1856         struct mm_struct *mm = tsk->mm;
1857         struct completion *vfork_done;
1858         int core_waiters = -EBUSY;
1859
1860         init_completion(&core_state->startup);
1861         core_state->dumper.task = tsk;
1862         core_state->dumper.next = NULL;
1863
1864         down_write(&mm->mmap_sem);
1865         if (!mm->core_state)
1866                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1867         up_write(&mm->mmap_sem);
1868
1869         if (unlikely(core_waiters < 0))
1870                 goto fail;
1871
1872         /*
1873          * Make sure nobody is waiting for us to release the VM,
1874          * otherwise we can deadlock when we wait on each other
1875          */
1876         vfork_done = tsk->vfork_done;
1877         if (vfork_done) {
1878                 tsk->vfork_done = NULL;
1879                 complete(vfork_done);
1880         }
1881
1882         if (core_waiters)
1883                 wait_for_completion(&core_state->startup);
1884 fail:
1885         return core_waiters;
1886 }
1887
1888 static void coredump_finish(struct mm_struct *mm)
1889 {
1890         struct core_thread *curr, *next;
1891         struct task_struct *task;
1892
1893         next = mm->core_state->dumper.next;
1894         while ((curr = next) != NULL) {
1895                 next = curr->next;
1896                 task = curr->task;
1897                 /*
1898                  * see exit_mm(), curr->task must not see
1899                  * ->task == NULL before we read ->next.
1900                  */
1901                 smp_mb();
1902                 curr->task = NULL;
1903                 wake_up_process(task);
1904         }
1905
1906         mm->core_state = NULL;
1907 }
1908
1909 /*
1910  * set_dumpable converts traditional three-value dumpable to two flags and
1911  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1912  * these bits are not changed atomically.  So get_dumpable can observe the
1913  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1914  * return either old dumpable or new one by paying attention to the order of
1915  * modifying the bits.
1916  *
1917  * dumpable |   mm->flags (binary)
1918  * old  new | initial interim  final
1919  * ---------+-----------------------
1920  *  0    1  |   00      01      01
1921  *  0    2  |   00      10(*)   11
1922  *  1    0  |   01      00      00
1923  *  1    2  |   01      11      11
1924  *  2    0  |   11      10(*)   00
1925  *  2    1  |   11      11      01
1926  *
1927  * (*) get_dumpable regards interim value of 10 as 11.
1928  */
1929 void set_dumpable(struct mm_struct *mm, int value)
1930 {
1931         switch (value) {
1932         case 0:
1933                 clear_bit(MMF_DUMPABLE, &mm->flags);
1934                 smp_wmb();
1935                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1936                 break;
1937         case 1:
1938                 set_bit(MMF_DUMPABLE, &mm->flags);
1939                 smp_wmb();
1940                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1941                 break;
1942         case 2:
1943                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1944                 smp_wmb();
1945                 set_bit(MMF_DUMPABLE, &mm->flags);
1946                 break;
1947         }
1948 }
1949
1950 static int __get_dumpable(unsigned long mm_flags)
1951 {
1952         int ret;
1953
1954         ret = mm_flags & MMF_DUMPABLE_MASK;
1955         return (ret >= 2) ? 2 : ret;
1956 }
1957
1958 int get_dumpable(struct mm_struct *mm)
1959 {
1960         return __get_dumpable(mm->flags);
1961 }
1962
1963 static void wait_for_dump_helpers(struct file *file)
1964 {
1965         struct pipe_inode_info *pipe;
1966
1967         pipe = file->f_path.dentry->d_inode->i_pipe;
1968
1969         pipe_lock(pipe);
1970         pipe->readers++;
1971         pipe->writers--;
1972
1973         while ((pipe->readers > 1) && (!signal_pending(current))) {
1974                 wake_up_interruptible_sync(&pipe->wait);
1975                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1976                 pipe_wait(pipe);
1977         }
1978
1979         pipe->readers--;
1980         pipe->writers++;
1981         pipe_unlock(pipe);
1982
1983 }
1984
1985
1986 /*
1987  * umh_pipe_setup
1988  * helper function to customize the process used
1989  * to collect the core in userspace.  Specifically
1990  * it sets up a pipe and installs it as fd 0 (stdin)
1991  * for the process.  Returns 0 on success, or
1992  * PTR_ERR on failure.
1993  * Note that it also sets the core limit to 1.  This
1994  * is a special value that we use to trap recursive
1995  * core dumps
1996  */
1997 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
1998 {
1999         struct file *rp, *wp;
2000         struct fdtable *fdt;
2001         struct coredump_params *cp = (struct coredump_params *)info->data;
2002         struct files_struct *cf = current->files;
2003
2004         wp = create_write_pipe(0);
2005         if (IS_ERR(wp))
2006                 return PTR_ERR(wp);
2007
2008         rp = create_read_pipe(wp, 0);
2009         if (IS_ERR(rp)) {
2010                 free_write_pipe(wp);
2011                 return PTR_ERR(rp);
2012         }
2013
2014         cp->file = wp;
2015
2016         sys_close(0);
2017         fd_install(0, rp);
2018         spin_lock(&cf->file_lock);
2019         fdt = files_fdtable(cf);
2020         FD_SET(0, fdt->open_fds);
2021         FD_CLR(0, fdt->close_on_exec);
2022         spin_unlock(&cf->file_lock);
2023
2024         /* and disallow core files too */
2025         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2026
2027         return 0;
2028 }
2029
2030 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2031 {
2032         struct core_state core_state;
2033         struct core_name cn;
2034         struct mm_struct *mm = current->mm;
2035         struct linux_binfmt * binfmt;
2036         const struct cred *old_cred;
2037         struct cred *cred;
2038         int retval = 0;
2039         int flag = 0;
2040         int ispipe;
2041         static atomic_t core_dump_count = ATOMIC_INIT(0);
2042         struct coredump_params cprm = {
2043                 .signr = signr,
2044                 .regs = regs,
2045                 .limit = rlimit(RLIMIT_CORE),
2046                 /*
2047                  * We must use the same mm->flags while dumping core to avoid
2048                  * inconsistency of bit flags, since this flag is not protected
2049                  * by any locks.
2050                  */
2051                 .mm_flags = mm->flags,
2052         };
2053
2054         printk(KERN_ERR "starting coredump : %s\n", current->comm);
2055         audit_core_dumps(signr);
2056
2057         binfmt = mm->binfmt;
2058         if (!binfmt || !binfmt->core_dump) {
2059                 printk(KERN_ERR " binfmt failed\n");
2060                 goto fail;
2061         }
2062         if (!__get_dumpable(cprm.mm_flags)) {
2063                 printk(KERN_ERR " get_dumpable fail\n");
2064                 goto fail;
2065         }
2066         cred = prepare_creds();
2067         if (!cred) {
2068                 printk(KERN_ERR " prepare_creds fail\n");
2069                 goto fail;
2070         }
2071
2072         /*
2073          *      We cannot trust fsuid as being the "true" uid of the
2074          *      process nor do we know its entire history. We only know it
2075          *      was tainted so we dump it as root in mode 2.
2076          */
2077         if (__get_dumpable(cprm.mm_flags) == 2) {
2078                 /* Setuid core dump mode */
2079                 flag = O_EXCL;          /* Stop rewrite attacks */
2080                 cred->fsuid = 0;        /* Dump root private */
2081         }
2082
2083         retval = coredump_wait(exit_code, &core_state);
2084         if (retval < 0) {
2085                 printk(KERN_ERR " coredump_wait fail_creds\n");
2086                 goto fail_creds;
2087         }
2088
2089         old_cred = override_creds(cred);
2090
2091         /*
2092          * Clear any false indication of pending signals that might
2093          * be seen by the filesystem code called to write the core file.
2094          */
2095         clear_thread_flag(TIF_SIGPENDING);
2096
2097         ispipe = format_corename(&cn, signr);
2098
2099         if (ispipe == -ENOMEM) {
2100                 printk(KERN_ERR "format_corename failed\n");
2101                 printk(KERN_ERR "Aborting core\n");
2102                 goto fail_corename;
2103         }
2104
2105         if (ispipe) {
2106                 int dump_count;
2107                 char **helper_argv;
2108
2109                 if (cprm.limit == 1) {
2110                         /*
2111                          * Normally core limits are irrelevant to pipes, since
2112                          * we're not writing to the file system, but we use
2113                          * cprm.limit of 1 here as a speacial value. Any
2114                          * non-1 limit gets set to RLIM_INFINITY below, but
2115                          * a limit of 0 skips the dump.  This is a consistent
2116                          * way to catch recursive crashes.  We can still crash
2117                          * if the core_pattern binary sets RLIM_CORE =  !1
2118                          * but it runs as root, and can do lots of stupid things
2119                          * Note that we use task_tgid_vnr here to grab the pid
2120                          * of the process group leader.  That way we get the
2121                          * right pid if a thread in a multi-threaded
2122                          * core_pattern process dies.
2123                          */
2124                         printk(KERN_ERR
2125                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2126                                 task_tgid_vnr(current), current->comm);
2127                         printk(KERN_ERR "Aborting core\n");
2128                         goto fail_unlock;
2129                 }
2130                 cprm.limit = RLIM_INFINITY;
2131
2132                 dump_count = atomic_inc_return(&core_dump_count);
2133                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2134                         printk(KERN_ERR "Pid %d(%s) over core_pipe_limit\n",
2135                                task_tgid_vnr(current), current->comm);
2136                         printk(KERN_ERR "Skipping core dump\n");
2137                         goto fail_dropcount;
2138                 }
2139
2140                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2141                 if (!helper_argv) {
2142                         printk(KERN_ERR "%s failed to allocate memory\n",
2143                                __func__);
2144                         goto fail_dropcount;
2145                 }
2146
2147                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2148                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2149                                         NULL, &cprm);
2150                 argv_free(helper_argv);
2151                 if (retval) {
2152                         printk(KERN_ERR "Core dump to %s pipe failed\n",
2153                                cn.corename);
2154                         goto close_fail;
2155                 }
2156         } else {
2157                 struct inode *inode;
2158
2159                 if (cprm.limit < binfmt->min_coredump) {
2160                         printk(KERN_ERR " min_coredump: %x less than %x\n",
2161                                         cprm.limit, binfmt->min_coredump);
2162                         goto fail_unlock;
2163                 }
2164
2165                 cprm.file = filp_open(cn.corename,
2166                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2167                                  0600);
2168                 if (IS_ERR(cprm.file)) {
2169                         printk(KERN_ERR " cprm file fail unlock\n");
2170                         goto fail_unlock;
2171                 }
2172
2173                 inode = cprm.file->f_path.dentry->d_inode;
2174                 if (inode->i_nlink > 1)
2175                         goto close_fail;
2176                 if (d_unhashed(cprm.file->f_path.dentry))
2177                         goto close_fail;
2178                 /*
2179                  * AK: actually i see no reason to not allow this for named
2180                  * pipes etc, but keep the previous behaviour for now.
2181                  */
2182                 if (!S_ISREG(inode->i_mode))
2183                         goto close_fail;
2184                 /*
2185                  * Dont allow local users get cute and trick others to coredump
2186                  * into their pre-created files.
2187                  */
2188                 if (inode->i_uid != current_fsuid())
2189                         goto close_fail;
2190                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2191                         goto close_fail;
2192                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2193                         goto close_fail;
2194         }
2195
2196         retval = binfmt->core_dump(&cprm);
2197         if (retval)
2198                 current->signal->group_exit_code |= 0x80;
2199
2200         if (ispipe && core_pipe_limit)
2201                 wait_for_dump_helpers(cprm.file);
2202 close_fail:
2203         if (cprm.file)
2204                 filp_close(cprm.file, NULL);
2205 fail_dropcount:
2206         if (ispipe)
2207                 atomic_dec(&core_dump_count);
2208 fail_unlock:
2209         kfree(cn.corename);
2210 fail_corename:
2211         coredump_finish(mm);
2212         revert_creds(old_cred);
2213 fail_creds:
2214         put_cred(cred);
2215 fail:
2216         return;
2217 }
2218
2219 /*
2220  * Core dumping helper functions.  These are the only things you should
2221  * do on a core-file: use only these functions to write out all the
2222  * necessary info.
2223  */
2224 int dump_write(struct file *file, const void *addr, int nr)
2225 {
2226         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2227 }
2228 EXPORT_SYMBOL(dump_write);
2229
2230 int dump_seek(struct file *file, loff_t off)
2231 {
2232         int ret = 1;
2233
2234         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2235                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2236                         return 0;
2237         } else {
2238                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2239
2240                 if (!buf)
2241                         return 0;
2242                 while (off > 0) {
2243                         unsigned long n = off;
2244
2245                         if (n > PAGE_SIZE)
2246                                 n = PAGE_SIZE;
2247                         if (!dump_write(file, buf, n)) {
2248                                 ret = 0;
2249                                 break;
2250                         }
2251                         off -= n;
2252                 }
2253                 free_page((unsigned long)buf);
2254         }
2255         return ret;
2256 }
2257 EXPORT_SYMBOL(dump_seek);