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 <trace/events/fs.h>
  60
  61 #include <asm/uaccess.h>
  62 #include <asm/mmu_context.h>
  63 #include <asm/tlb.h>
  64 #include "internal.h"
  65
  66 int core_uses_pid;
  67 char core_pattern[CORENAME_MAX_SIZE] = "core";
  68 unsigned int core_pipe_limit;
  69 int suid_dumpable = 0;
  70
  71 struct core_name {
  72         char *corename;
  73         int used, size;
  74 };
  75 static atomic_t call_count = ATOMIC_INIT(1);
  76
  77 /* The maximal length of core_pattern is also specified in sysctl.c */
  78
  79 static LIST_HEAD(formats);
  80 static DEFINE_RWLOCK(binfmt_lock);
  81
  82 int __register_binfmt(struct linux_binfmt * fmt, int insert)
  83 {
  84         if (!fmt)
  85                 return -EINVAL;
  86         write_lock(&binfmt_lock);
  87         insert ? list_add(&fmt->lh, &formats) :
  88                  list_add_tail(&fmt->lh, &formats);
  89         write_unlock(&binfmt_lock);
  90         return 0;
  91 }
  92
  93 EXPORT_SYMBOL(__register_binfmt);
  94
  95 void unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         write_lock(&binfmt_lock);
  98         list_del(&fmt->lh);
  99         write_unlock(&binfmt_lock);
 100 }
 101
 102 EXPORT_SYMBOL(unregister_binfmt);
 103
 104 static inline void put_binfmt(struct linux_binfmt * fmt)
 105 {
 106         module_put(fmt->module);
 107 }
 108
 109 /*
 110  * Note that a shared library must be both readable and executable due to
 111  * security reasons.
 112  *
 113  * Also note that we take the address to load from from the file itself.
 114  */
 115 SYSCALL_DEFINE1(uselib, const char __user *, library)
 116 {
 117         struct file *file;
 118         char *tmp = getname(library);
 119         int error = PTR_ERR(tmp);
 120         static const struct open_flags uselib_flags = {
 121                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 122                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 123                 .intent = LOOKUP_OPEN
 124         };
 125
 126         if (IS_ERR(tmp))
 127                 goto out;
 128
 129         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 130         putname(tmp);
 131         error = PTR_ERR(file);
 132         if (IS_ERR(file))
 133                 goto out;
 134
 135         error = -EINVAL;
 136         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 137                 goto exit;
 138
 139         error = -EACCES;
 140         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 141                 goto exit;
 142
 143         fsnotify_open(file);
 144
 145         error = -ENOEXEC;
 146         if(file->f_op) {
 147                 struct linux_binfmt * fmt;
 148
 149                 read_lock(&binfmt_lock);
 150                 list_for_each_entry(fmt, &formats, lh) {
 151                         if (!fmt->load_shlib)
 152                                 continue;
 153                         if (!try_module_get(fmt->module))
 154                                 continue;
 155                         read_unlock(&binfmt_lock);
 156                         error = fmt->load_shlib(file);
 157                         read_lock(&binfmt_lock);
 158                         put_binfmt(fmt);
 159                         if (error != -ENOEXEC)
 160                                 break;
 161                 }
 162                 read_unlock(&binfmt_lock);
 163         }
 164 exit:
 165         fput(file);
 166 out:
 167         return error;
 168 }
 169
 170 #ifdef CONFIG_MMU
 171 /*
 172  * The nascent bprm->mm is not visible until exec_mmap() but it can
 173  * use a lot of memory, account these pages in current->mm temporary
 174  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 175  * change the counter back via acct_arg_size(0).
 176  */
 177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 178 {
 179         struct mm_struct *mm = current->mm;
 180         long diff = (long)(pages - bprm->vma_pages);
 181
 182         if (!mm || !diff)
 183                 return;
 184
 185         bprm->vma_pages = pages;
 186         add_mm_counter(mm, MM_ANONPAGES, diff);
 187 }
 188
 189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 190                 int write)
 191 {
 192         struct page *page;
 193         int ret;
 194
 195 #ifdef CONFIG_STACK_GROWSUP
 196         if (write) {
 197                 ret = expand_downwards(bprm->vma, pos);
 198                 if (ret < 0)
 199                         return NULL;
 200         }
 201 #endif
 202         ret = get_user_pages(current, bprm->mm, pos,
 203                         1, write, 1, &page, NULL);
 204         if (ret <= 0)
 205                 return NULL;
 206
 207         if (write) {
 208                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 209                 struct rlimit *rlim;
 210
 211                 acct_arg_size(bprm, size / PAGE_SIZE);
 212
 213                 /*
 214                  * We've historically supported up to 32 pages (ARG_MAX)
 215                  * of argument strings even with small stacks
 216                  */
 217                 if (size <= ARG_MAX)
 218                         return page;
 219
 220                 /*
 221                  * Limit to 1/4-th the stack size for the argv+env strings.
 222                  * This ensures that:
 223                  *  - the remaining binfmt code will not run out of stack space,
 224                  *  - the program will have a reasonable amount of stack left
 225                  *    to work from.
 226                  */
 227                 rlim = current->signal->rlim;
 228                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 229                         put_page(page);
 230                         return NULL;
 231                 }
 232         }
 233
 234         return page;
 235 }
 236
 237 static void put_arg_page(struct page *page)
 238 {
 239         put_page(page);
 240 }
 241
 242 static void free_arg_page(struct linux_binprm *bprm, int i)
 243 {
 244 }
 245
 246 static void free_arg_pages(struct linux_binprm *bprm)
 247 {
 248 }
 249
 250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 251                 struct page *page)
 252 {
 253         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 254 }
 255
 256 static int __bprm_mm_init(struct linux_binprm *bprm)
 257 {
 258         int err;
 259         struct vm_area_struct *vma = NULL;
 260         struct mm_struct *mm = bprm->mm;
 261
 262         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 263         if (!vma)
 264                 return -ENOMEM;
 265
 266         down_write(&mm->mmap_sem);
 267         vma->vm_mm = mm;
 268
 269         /*
 270          * Place the stack at the largest stack address the architecture
 271          * supports. Later, we'll move this to an appropriate place. We don't
 272          * use STACK_TOP because that can depend on attributes which aren't
 273          * configured yet.
 274          */
 275         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 276         vma->vm_end = STACK_TOP_MAX;
 277         vma->vm_start = vma->vm_end - PAGE_SIZE;
 278         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 279         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 280         INIT_LIST_HEAD(&vma->anon_vma_chain);
 281
 282         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
 283         if (err)
 284                 goto err;
 285
 286         err = insert_vm_struct(mm, vma);
 287         if (err)
 288                 goto err;
 289
 290         mm->stack_vm = mm->total_vm = 1;
 291         up_write(&mm->mmap_sem);
 292         bprm->p = vma->vm_end - sizeof(void *);
 293         return 0;
 294 err:
 295         up_write(&mm->mmap_sem);
 296         bprm->vma = NULL;
 297         kmem_cache_free(vm_area_cachep, vma);
 298         return err;
 299 }
 300
 301 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 302 {
 303         return len <= MAX_ARG_STRLEN;
 304 }
 305
 306 #else
 307
 308 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 309 {
 310 }
 311
 312 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 313                 int write)
 314 {
 315         struct page *page;
 316
 317         page = bprm->page[pos / PAGE_SIZE];
 318         if (!page && write) {
 319                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 320                 if (!page)
 321                         return NULL;
 322                 bprm->page[pos / PAGE_SIZE] = page;
 323         }
 324
 325         return page;
 326 }
 327
 328 static void put_arg_page(struct page *page)
 329 {
 330 }
 331
 332 static void free_arg_page(struct linux_binprm *bprm, int i)
 333 {
 334         if (bprm->page[i]) {
 335                 __free_page(bprm->page[i]);
 336                 bprm->page[i] = NULL;
 337         }
 338 }
 339
 340 static void free_arg_pages(struct linux_binprm *bprm)
 341 {
 342         int i;
 343
 344         for (i = 0; i < MAX_ARG_PAGES; i++)
 345                 free_arg_page(bprm, i);
 346 }
 347
 348 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 349                 struct page *page)
 350 {
 351 }
 352
 353 static int __bprm_mm_init(struct linux_binprm *bprm)
 354 {
 355         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 356         return 0;
 357 }
 358
 359 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 360 {
 361         return len <= bprm->p;
 362 }
 363
 364 #endif /* CONFIG_MMU */
 365
 366 /*
 367  * Create a new mm_struct and populate it with a temporary stack
 368  * vm_area_struct.  We don't have enough context at this point to set the stack
 369  * flags, permissions, and offset, so we use temporary values.  We'll update
 370  * them later in setup_arg_pages().
 371  */
 372 int bprm_mm_init(struct linux_binprm *bprm)
 373 {
 374         int err;
 375         struct mm_struct *mm = NULL;
 376
 377         bprm->mm = mm = mm_alloc();
 378         err = -ENOMEM;
 379         if (!mm)
 380                 goto err;
 381
 382         err = init_new_context(current, mm);
 383         if (err)
 384                 goto err;
 385
 386         err = __bprm_mm_init(bprm);
 387         if (err)
 388                 goto err;
 389
 390         return 0;
 391
 392 err:
 393         if (mm) {
 394                 bprm->mm = NULL;
 395                 mmdrop(mm);
 396         }
 397
 398         return err;
 399 }
 400
 401 struct user_arg_ptr {
 402 #ifdef CONFIG_COMPAT
 403         bool is_compat;
 404 #endif
 405         union {
 406                 const char __user *const __user *native;
 407 #ifdef CONFIG_COMPAT
 408                 compat_uptr_t __user *compat;
 409 #endif
 410         } ptr;
 411 };
 412
 413 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 414 {
 415         const char __user *native;
 416
 417 #ifdef CONFIG_COMPAT
 418         if (unlikely(argv.is_compat)) {
 419                 compat_uptr_t compat;
 420
 421                 if (get_user(compat, argv.ptr.compat + nr))
 422                         return ERR_PTR(-EFAULT);
 423
 424                 return compat_ptr(compat);
 425         }
 426 #endif
 427
 428         if (get_user(native, argv.ptr.native + nr))
 429                 return ERR_PTR(-EFAULT);
 430
 431         return native;
 432 }
 433
 434 /*
 435  * count() counts the number of strings in array ARGV.
 436  */
 437 static int count(struct user_arg_ptr argv, int max)
 438 {
 439         int i = 0;
 440
 441         if (argv.ptr.native != NULL) {
 442                 for (;;) {
 443                         const char __user *p = get_user_arg_ptr(argv, i);
 444
 445                         if (!p)
 446                                 break;
 447
 448                         if (IS_ERR(p))
 449                                 return -EFAULT;
 450
 451                         if (i++ >= max)
 452                                 return -E2BIG;
 453
 454                         if (fatal_signal_pending(current))
 455                                 return -ERESTARTNOHAND;
 456                         cond_resched();
 457                 }
 458         }
 459         return i;
 460 }
 461
 462 /*
 463  * 'copy_strings()' copies argument/environment strings from the old
 464  * processes's memory to the new process's stack.  The call to get_user_pages()
 465  * ensures the destination page is created and not swapped out.
 466  */
 467 static int copy_strings(int argc, struct user_arg_ptr argv,
 468                         struct linux_binprm *bprm)
 469 {
 470         struct page *kmapped_page = NULL;
 471         char *kaddr = NULL;
 472         unsigned long kpos = 0;
 473         int ret;
 474
 475         while (argc-- > 0) {
 476                 const char __user *str;
 477                 int len;
 478                 unsigned long pos;
 479
 480                 ret = -EFAULT;
 481                 str = get_user_arg_ptr(argv, argc);
 482                 if (IS_ERR(str))
 483                         goto out;
 484
 485                 len = strnlen_user(str, MAX_ARG_STRLEN);
 486                 if (!len)
 487                         goto out;
 488
 489                 ret = -E2BIG;
 490                 if (!valid_arg_len(bprm, len))
 491                         goto out;
 492
 493                 /* We're going to work our way backwords. */
 494                 pos = bprm->p;
 495                 str += len;
 496                 bprm->p -= len;
 497
 498                 while (len > 0) {
 499                         int offset, bytes_to_copy;
 500
 501                         if (fatal_signal_pending(current)) {
 502                                 ret = -ERESTARTNOHAND;
 503                                 goto out;
 504                         }
 505                         cond_resched();
 506
 507                         offset = pos % PAGE_SIZE;
 508                         if (offset == 0)
 509                                 offset = PAGE_SIZE;
 510
 511                         bytes_to_copy = offset;
 512                         if (bytes_to_copy > len)
 513                                 bytes_to_copy = len;
 514
 515                         offset -= bytes_to_copy;
 516                         pos -= bytes_to_copy;
 517                         str -= bytes_to_copy;
 518                         len -= bytes_to_copy;
 519
 520                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 521                                 struct page *page;
 522
 523                                 page = get_arg_page(bprm, pos, 1);
 524                                 if (!page) {
 525                                         ret = -E2BIG;
 526                                         goto out;
 527                                 }
 528
 529                                 if (kmapped_page) {
 530                                         flush_kernel_dcache_page(kmapped_page);
 531                                         kunmap(kmapped_page);
 532                                         put_arg_page(kmapped_page);
 533                                 }
 534                                 kmapped_page = page;
 535                                 kaddr = kmap(kmapped_page);
 536                                 kpos = pos & PAGE_MASK;
 537                                 flush_arg_page(bprm, kpos, kmapped_page);
 538                         }
 539                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 540                                 ret = -EFAULT;
 541                                 goto out;
 542                         }
 543                 }
 544         }
 545         ret = 0;
 546 out:
 547         if (kmapped_page) {
 548                 flush_kernel_dcache_page(kmapped_page);
 549                 kunmap(kmapped_page);
 550                 put_arg_page(kmapped_page);
 551         }
 552         return ret;
 553 }
 554
 555 /*
 556  * Like copy_strings, but get argv and its values from kernel memory.
 557  */
 558 int copy_strings_kernel(int argc, const char *const *__argv,
 559                         struct linux_binprm *bprm)
 560 {
 561         int r;
 562         mm_segment_t oldfs = get_fs();
 563         struct user_arg_ptr argv = {
 564                 .ptr.native = (const char __user *const  __user *)__argv,
 565         };
 566
 567         set_fs(KERNEL_DS);
 568         r = copy_strings(argc, argv, bprm);
 569         set_fs(oldfs);
 570
 571         return r;
 572 }
 573 EXPORT_SYMBOL(copy_strings_kernel);
 574
 575 #ifdef CONFIG_MMU
 576
 577 /*
 578  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 579  * the binfmt code determines where the new stack should reside, we shift it to
 580  * its final location.  The process proceeds as follows:
 581  *
 582  * 1) Use shift to calculate the new vma endpoints.
 583  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 584  *    arguments passed to subsequent functions are consistent.
 585  * 3) Move vma's page tables to the new range.
 586  * 4) Free up any cleared pgd range.
 587  * 5) Shrink the vma to cover only the new range.
 588  */
 589 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 590 {
 591         struct mm_struct *mm = vma->vm_mm;
 592         unsigned long old_start = vma->vm_start;
 593         unsigned long old_end = vma->vm_end;
 594         unsigned long length = old_end - old_start;
 595         unsigned long new_start = old_start - shift;
 596         unsigned long new_end = old_end - shift;
 597         struct mmu_gather tlb;
 598
 599         BUG_ON(new_start > new_end);
 600
 601         /*
 602          * ensure there are no vmas between where we want to go
 603          * and where we are
 604          */
 605         if (vma != find_vma(mm, new_start))
 606                 return -EFAULT;
 607
 608         /*
 609          * cover the whole range: [new_start, old_end)
 610          */
 611         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 612                 return -ENOMEM;
 613
 614         /*
 615          * move the page tables downwards, on failure we rely on
 616          * process cleanup to remove whatever mess we made.
 617          */
 618         if (length != move_page_tables(vma, old_start,
 619                                        vma, new_start, length))
 620                 return -ENOMEM;
 621
 622         lru_add_drain();
 623         tlb_gather_mmu(&tlb, mm, 0);
 624         if (new_end > old_start) {
 625                 /*
 626                  * when the old and new regions overlap clear from new_end.
 627                  */
 628                 free_pgd_range(&tlb, new_end, old_end, new_end,
 629                         vma->vm_next ? vma->vm_next->vm_start : 0);
 630         } else {
 631                 /*
 632                  * otherwise, clean from old_start; this is done to not touch
 633                  * the address space in [new_end, old_start) some architectures
 634                  * have constraints on va-space that make this illegal (IA64) -
 635                  * for the others its just a little faster.
 636                  */
 637                 free_pgd_range(&tlb, old_start, old_end, new_end,
 638                         vma->vm_next ? vma->vm_next->vm_start : 0);
 639         }
 640         tlb_finish_mmu(&tlb, new_end, old_end);
 641
 642         /*
 643          * Shrink the vma to just the new range.  Always succeeds.
 644          */
 645         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 646
 647         return 0;
 648 }
 649
 650 /*
 651  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 652  * the stack is optionally relocated, and some extra space is added.
 653  */
 654 int setup_arg_pages(struct linux_binprm *bprm,
 655                     unsigned long stack_top,
 656                     int executable_stack)
 657 {
 658         unsigned long ret;
 659         unsigned long stack_shift;
 660         struct mm_struct *mm = current->mm;
 661         struct vm_area_struct *vma = bprm->vma;
 662         struct vm_area_struct *prev = NULL;
 663         unsigned long vm_flags;
 664         unsigned long stack_base;
 665         unsigned long stack_size;
 666         unsigned long stack_expand;
 667         unsigned long rlim_stack;
 668
 669 #ifdef CONFIG_STACK_GROWSUP
 670         /* Limit stack size to 1GB */
 671         stack_base = rlimit_max(RLIMIT_STACK);
 672         if (stack_base > (1 << 30))
 673                 stack_base = 1 << 30;
 674
 675         /* Make sure we didn't let the argument array grow too large. */
 676         if (vma->vm_end - vma->vm_start > stack_base)
 677                 return -ENOMEM;
 678
 679         stack_base = PAGE_ALIGN(stack_top - stack_base);
 680
 681         stack_shift = vma->vm_start - stack_base;
 682         mm->arg_start = bprm->p - stack_shift;
 683         bprm->p = vma->vm_end - stack_shift;
 684 #else
 685         stack_top = arch_align_stack(stack_top);
 686         stack_top = PAGE_ALIGN(stack_top);
 687
 688         if (unlikely(stack_top < mmap_min_addr) ||
 689             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 690                 return -ENOMEM;
 691
 692         stack_shift = vma->vm_end - stack_top;
 693
 694         bprm->p -= stack_shift;
 695         mm->arg_start = bprm->p;
 696 #endif
 697
 698         if (bprm->loader)
 699                 bprm->loader -= stack_shift;
 700         bprm->exec -= stack_shift;
 701
 702         down_write(&mm->mmap_sem);
 703         vm_flags = VM_STACK_FLAGS;
 704
 705         /*
 706          * Adjust stack execute permissions; explicitly enable for
 707          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 708          * (arch default) otherwise.
 709          */
 710         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 711                 vm_flags |= VM_EXEC;
 712         else if (executable_stack == EXSTACK_DISABLE_X)
 713                 vm_flags &= ~VM_EXEC;
 714         vm_flags |= mm->def_flags;
 715         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 716
 717         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 718                         vm_flags);
 719         if (ret)
 720                 goto out_unlock;
 721         BUG_ON(prev != vma);
 722
 723         /* Move stack pages down in memory. */
 724         if (stack_shift) {
 725                 ret = shift_arg_pages(vma, stack_shift);
 726                 if (ret)
 727                         goto out_unlock;
 728         }
 729
 730         /* mprotect_fixup is overkill to remove the temporary stack flags */
 731         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 732
 733         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 734         stack_size = vma->vm_end - vma->vm_start;
 735         /*
 736          * Align this down to a page boundary as expand_stack
 737          * will align it up.
 738          */
 739         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 740 #ifdef CONFIG_STACK_GROWSUP
 741         if (stack_size + stack_expand > rlim_stack)
 742                 stack_base = vma->vm_start + rlim_stack;
 743         else
 744                 stack_base = vma->vm_end + stack_expand;
 745 #else
 746         if (stack_size + stack_expand > rlim_stack)
 747                 stack_base = vma->vm_end - rlim_stack;
 748         else
 749                 stack_base = vma->vm_start - stack_expand;
 750 #endif
 751         current->mm->start_stack = bprm->p;
 752         ret = expand_stack(vma, stack_base);
 753         if (ret)
 754                 ret = -EFAULT;
 755
 756 out_unlock:
 757         up_write(&mm->mmap_sem);
 758         return ret;
 759 }
 760 EXPORT_SYMBOL(setup_arg_pages);
 761
 762 #endif /* CONFIG_MMU */
 763
 764 struct file *open_exec(const char *name)
 765 {
 766         struct file *file;
 767         int err;
 768         static const struct open_flags open_exec_flags = {
 769                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 770                 .acc_mode = MAY_EXEC | MAY_OPEN,
 771                 .intent = LOOKUP_OPEN
 772         };
 773
 774         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 775         if (IS_ERR(file))
 776                 goto out;
 777
 778         err = -EACCES;
 779         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 780                 goto exit;
 781
 782         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 783                 goto exit;
 784
 785         fsnotify_open(file);
 786
 787         trace_open_exec(name);
 788
 789         err = deny_write_access(file);
 790         if (err)
 791                 goto exit;
 792
 793 out:
 794         return file;
 795
 796 exit:
 797         fput(file);
 798         return ERR_PTR(err);
 799 }
 800 EXPORT_SYMBOL(open_exec);
 801
 802 int kernel_read(struct file *file, loff_t offset,
 803                 char *addr, unsigned long count)
 804 {
 805         mm_segment_t old_fs;
 806         loff_t pos = offset;
 807         int result;
 808
 809         old_fs = get_fs();
 810         set_fs(get_ds());
 811         /* The cast to a user pointer is valid due to the set_fs() */
 812         result = vfs_read(file, (void __user *)addr, count, &pos);
 813         set_fs(old_fs);
 814         return result;
 815 }
 816
 817 EXPORT_SYMBOL(kernel_read);
 818
 819 static int exec_mmap(struct mm_struct *mm)
 820 {
 821         struct task_struct *tsk;
 822         struct mm_struct * old_mm, *active_mm;
 823
 824         /* Notify parent that we're no longer interested in the old VM */
 825         tsk = current;
 826         old_mm = current->mm;
 827         sync_mm_rss(tsk, old_mm);
 828         mm_release(tsk, old_mm);
 829
 830         if (old_mm) {
 831                 /*
 832                  * Make sure that if there is a core dump in progress
 833                  * for the old mm, we get out and die instead of going
 834                  * through with the exec.  We must hold mmap_sem around
 835                  * checking core_state and changing tsk->mm.
 836                  */
 837                 down_read(&old_mm->mmap_sem);
 838                 if (unlikely(old_mm->core_state)) {
 839                         up_read(&old_mm->mmap_sem);
 840                         return -EINTR;
 841                 }
 842         }
 843         task_lock(tsk);
 844         active_mm = tsk->active_mm;
 845         tsk->mm = mm;
 846         tsk->active_mm = mm;
 847         activate_mm(active_mm, mm);
 848         task_unlock(tsk);
 849         arch_pick_mmap_layout(mm);
 850         if (old_mm) {
 851                 up_read(&old_mm->mmap_sem);
 852                 BUG_ON(active_mm != old_mm);
 853                 mm_update_next_owner(old_mm);
 854                 mmput(old_mm);
 855                 return 0;
 856         }
 857         mmdrop(active_mm);
 858         return 0;
 859 }
 860
 861 /*
 862  * This function makes sure the current process has its own signal table,
 863  * so that flush_signal_handlers can later reset the handlers without
 864  * disturbing other processes.  (Other processes might share the signal
 865  * table via the CLONE_SIGHAND option to clone().)
 866  */
 867 static int de_thread(struct task_struct *tsk)
 868 {
 869         struct signal_struct *sig = tsk->signal;
 870         struct sighand_struct *oldsighand = tsk->sighand;
 871         spinlock_t *lock = &oldsighand->siglock;
 872
 873         if (thread_group_empty(tsk))
 874                 goto no_thread_group;
 875
 876         /*
 877          * Kill all other threads in the thread group.
 878          */
 879         spin_lock_irq(lock);
 880         if (signal_group_exit(sig)) {
 881                 /*
 882                  * Another group action in progress, just
 883                  * return so that the signal is processed.
 884                  */
 885                 spin_unlock_irq(lock);
 886                 return -EAGAIN;
 887         }
 888
 889         sig->group_exit_task = tsk;
 890         sig->notify_count = zap_other_threads(tsk);
 891         if (!thread_group_leader(tsk))
 892                 sig->notify_count--;
 893
 894         while (sig->notify_count) {
 895                 __set_current_state(TASK_UNINTERRUPTIBLE);
 896                 spin_unlock_irq(lock);
 897                 schedule();
 898                 spin_lock_irq(lock);
 899         }
 900         spin_unlock_irq(lock);
 901
 902         /*
 903          * At this point all other threads have exited, all we have to
 904          * do is to wait for the thread group leader to become inactive,
 905          * and to assume its PID:
 906          */
 907         if (!thread_group_leader(tsk)) {
 908                 struct task_struct *leader = tsk->group_leader;
 909
 910                 sig->notify_count = -1; /* for exit_notify() */
 911                 for (;;) {
 912                         write_lock_irq(&tasklist_lock);
 913                         if (likely(leader->exit_state))
 914                                 break;
 915                         __set_current_state(TASK_UNINTERRUPTIBLE);
 916                         write_unlock_irq(&tasklist_lock);
 917                         schedule();
 918                 }
 919
 920                 /*
 921                  * The only record we have of the real-time age of a
 922                  * process, regardless of execs it's done, is start_time.
 923                  * All the past CPU time is accumulated in signal_struct
 924                  * from sister threads now dead.  But in this non-leader
 925                  * exec, nothing survives from the original leader thread,
 926                  * whose birth marks the true age of this process now.
 927                  * When we take on its identity by switching to its PID, we
 928                  * also take its birthdate (always earlier than our own).
 929                  */
 930                 tsk->start_time = leader->start_time;
 931
 932                 BUG_ON(!same_thread_group(leader, tsk));
 933                 BUG_ON(has_group_leader_pid(tsk));
 934                 /*
 935                  * An exec() starts a new thread group with the
 936                  * TGID of the previous thread group. Rehash the
 937                  * two threads with a switched PID, and release
 938                  * the former thread group leader:
 939                  */
 940
 941                 /* Become a process group leader with the old leader's pid.
 942                  * The old leader becomes a thread of the this thread group.
 943                  * Note: The old leader also uses this pid until release_task
 944                  *       is called.  Odd but simple and correct.
 945                  */
 946                 detach_pid(tsk, PIDTYPE_PID);
 947                 tsk->pid = leader->pid;
 948                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 949                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 950                 transfer_pid(leader, tsk, PIDTYPE_SID);
 951
 952                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 953                 list_replace_init(&leader->sibling, &tsk->sibling);
 954
 955                 tsk->group_leader = tsk;
 956                 leader->group_leader = tsk;
 957
 958                 tsk->exit_signal = SIGCHLD;
 959                 leader->exit_signal = -1;
 960
 961                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 962                 leader->exit_state = EXIT_DEAD;
 963
 964                 /*
 965                  * We are going to release_task()->ptrace_unlink() silently,
 966                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 967                  * the tracer wont't block again waiting for this thread.
 968                  */
 969                 if (unlikely(leader->ptrace))
 970                         __wake_up_parent(leader, leader->parent);
 971                 write_unlock_irq(&tasklist_lock);
 972
 973                 release_task(leader);
 974         }
 975
 976         sig->group_exit_task = NULL;
 977         sig->notify_count = 0;
 978
 979 no_thread_group:
 980         if (current->mm)
 981                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 982
 983         exit_itimers(sig);
 984         flush_itimer_signals();
 985
 986         if (atomic_read(&oldsighand->count) != 1) {
 987                 struct sighand_struct *newsighand;
 988                 /*
 989                  * This ->sighand is shared with the CLONE_SIGHAND
 990                  * but not CLONE_THREAD task, switch to the new one.
 991                  */
 992                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 993                 if (!newsighand)
 994                         return -ENOMEM;
 995
 996                 atomic_set(&newsighand->count, 1);
 997                 memcpy(newsighand->action, oldsighand->action,
 998                        sizeof(newsighand->action));
 999
1000                 write_lock_irq(&tasklist_lock);
1001                 spin_lock(&oldsighand->siglock);
1002                 rcu_assign_pointer(tsk->sighand, newsighand);
1003                 spin_unlock(&oldsighand->siglock);
1004                 write_unlock_irq(&tasklist_lock);
1005
1006                 __cleanup_sighand(oldsighand);
1007         }
1008
1009         BUG_ON(!thread_group_leader(tsk));
1010         return 0;
1011 }
1012
1013 /*
1014  * These functions flushes out all traces of the currently running executable
1015  * so that a new one can be started
1016  */
1017 static void flush_old_files(struct files_struct * files)
1018 {
1019         long j = -1;
1020         struct fdtable *fdt;
1021
1022         spin_lock(&files->file_lock);
1023         for (;;) {
1024                 unsigned long set, i;
1025
1026                 j++;
1027                 i = j * __NFDBITS;
1028                 fdt = files_fdtable(files);
1029                 if (i >= fdt->max_fds)
1030                         break;
1031                 set = fdt->close_on_exec->fds_bits[j];
1032                 if (!set)
1033                         continue;
1034                 fdt->close_on_exec->fds_bits[j] = 0;
1035                 spin_unlock(&files->file_lock);
1036                 for ( ; set ; i++,set >>= 1) {
1037                         if (set & 1) {
1038                                 sys_close(i);
1039                         }
1040                 }
1041                 spin_lock(&files->file_lock);
1042
1043         }
1044         spin_unlock(&files->file_lock);
1045 }
1046
1047 char *get_task_comm(char *buf, struct task_struct *tsk)
1048 {
1049         /* buf must be at least sizeof(tsk->comm) in size */
1050         task_lock(tsk);
1051         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1052         task_unlock(tsk);
1053         return buf;
1054 }
1055 EXPORT_SYMBOL_GPL(get_task_comm);
1056
1057 void set_task_comm(struct task_struct *tsk, char *buf)
1058 {
1059         task_lock(tsk);
1060
1061         /*
1062          * Threads may access current->comm without holding
1063          * the task lock, so write the string carefully.
1064          * Readers without a lock may see incomplete new
1065          * names but are safe from non-terminating string reads.
1066          */
1067         memset(tsk->comm, 0, TASK_COMM_LEN);
1068         wmb();
1069         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1070         task_unlock(tsk);
1071         perf_event_comm(tsk);
1072 }
1073
1074 int flush_old_exec(struct linux_binprm * bprm)
1075 {
1076         int retval;
1077
1078         /*
1079          * Make sure we have a private signal table and that
1080          * we are unassociated from the previous thread group.
1081          */
1082         retval = de_thread(current);
1083         if (retval)
1084                 goto out;
1085
1086         set_mm_exe_file(bprm->mm, bprm->file);
1087
1088         /*
1089          * Release all of the old mmap stuff
1090          */
1091         acct_arg_size(bprm, 0);
1092         retval = exec_mmap(bprm->mm);
1093         if (retval)
1094                 goto out;
1095
1096         bprm->mm = NULL;                /* We're using it now */
1097
1098         set_fs(USER_DS);
1099         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1100         flush_thread();
1101         current->personality &= ~bprm->per_clear;
1102
1103         return 0;
1104
1105 out:
1106         return retval;
1107 }
1108 EXPORT_SYMBOL(flush_old_exec);
1109
1110 void would_dump(struct linux_binprm *bprm, struct file *file)
1111 {
1112         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1113                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1114 }
1115 EXPORT_SYMBOL(would_dump);
1116
1117 void setup_new_exec(struct linux_binprm * bprm)
1118 {
1119         int i, ch;
1120         const char *name;
1121         char tcomm[sizeof(current->comm)];
1122
1123         arch_pick_mmap_layout(current->mm);
1124
1125         /* This is the point of no return */
1126         current->sas_ss_sp = current->sas_ss_size = 0;
1127
1128         if (current_euid() == current_uid() && current_egid() == current_gid())
1129                 set_dumpable(current->mm, 1);
1130         else
1131                 set_dumpable(current->mm, suid_dumpable);
1132
1133         name = bprm->filename;
1134
1135         /* Copies the binary name from after last slash */
1136         for (i=0; (ch = *(name++)) != '\0';) {
1137                 if (ch == '/')
1138                         i = 0; /* overwrite what we wrote */
1139                 else
1140                         if (i < (sizeof(tcomm) - 1))
1141                                 tcomm[i++] = ch;
1142         }
1143         tcomm[i] = '\0';
1144         set_task_comm(current, tcomm);
1145
1146         /* Set the new mm task size. We have to do that late because it may
1147          * depend on TIF_32BIT which is only updated in flush_thread() on
1148          * some architectures like powerpc
1149          */
1150         current->mm->task_size = TASK_SIZE;
1151
1152         /* install the new credentials */
1153         if (bprm->cred->uid != current_euid() ||
1154             bprm->cred->gid != current_egid()) {
1155                 current->pdeath_signal = 0;
1156         } else {
1157                 would_dump(bprm, bprm->file);
1158                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1159                         set_dumpable(current->mm, suid_dumpable);
1160         }
1161
1162         /*
1163          * Flush performance counters when crossing a
1164          * security domain:
1165          */
1166         if (!get_dumpable(current->mm))
1167                 perf_event_exit_task(current);
1168
1169         /* An exec changes our domain. We are no longer part of the thread
1170            group */
1171
1172         current->self_exec_id++;
1173
1174         flush_signal_handlers(current, 0);
1175         flush_old_files(current->files);
1176 }
1177 EXPORT_SYMBOL(setup_new_exec);
1178
1179 /*
1180  * Prepare credentials and lock ->cred_guard_mutex.
1181  * install_exec_creds() commits the new creds and drops the lock.
1182  * Or, if exec fails before, free_bprm() should release ->cred and
1183  * and unlock.
1184  */
1185 int prepare_bprm_creds(struct linux_binprm *bprm)
1186 {
1187         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1188                 return -ERESTARTNOINTR;
1189
1190         bprm->cred = prepare_exec_creds();
1191         if (likely(bprm->cred))
1192                 return 0;
1193
1194         mutex_unlock(&current->signal->cred_guard_mutex);
1195         return -ENOMEM;
1196 }
1197
1198 void free_bprm(struct linux_binprm *bprm)
1199 {
1200         free_arg_pages(bprm);
1201         if (bprm->cred) {
1202                 mutex_unlock(&current->signal->cred_guard_mutex);
1203                 abort_creds(bprm->cred);
1204         }
1205         kfree(bprm);
1206 }
1207
1208 /*
1209  * install the new credentials for this executable
1210  */
1211 void install_exec_creds(struct linux_binprm *bprm)
1212 {
1213         security_bprm_committing_creds(bprm);
1214
1215         commit_creds(bprm->cred);
1216         bprm->cred = NULL;
1217         /*
1218          * cred_guard_mutex must be held at least to this point to prevent
1219          * ptrace_attach() from altering our determination of the task's
1220          * credentials; any time after this it may be unlocked.
1221          */
1222         security_bprm_committed_creds(bprm);
1223         mutex_unlock(&current->signal->cred_guard_mutex);
1224 }
1225 EXPORT_SYMBOL(install_exec_creds);
1226
1227 /*
1228  * determine how safe it is to execute the proposed program
1229  * - the caller must hold ->cred_guard_mutex to protect against
1230  *   PTRACE_ATTACH
1231  */
1232 int check_unsafe_exec(struct linux_binprm *bprm)
1233 {
1234         struct task_struct *p = current, *t;
1235         unsigned n_fs;
1236         int res = 0;
1237
1238         if (p->ptrace) {
1239                 if (p->ptrace & PT_PTRACE_CAP)
1240                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1241                 else
1242                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1243         }
1244
1245         /*
1246          * This isn't strictly necessary, but it makes it harder for LSMs to
1247          * mess up.
1248          */
1249         if (current->no_new_privs)
1250                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1251
1252         n_fs = 1;
1253         spin_lock(&p->fs->lock);
1254         rcu_read_lock();
1255         for (t = next_thread(p); t != p; t = next_thread(t)) {
1256                 if (t->fs == p->fs)
1257                         n_fs++;
1258         }
1259         rcu_read_unlock();
1260
1261         if (p->fs->users > n_fs) {
1262                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1263         } else {
1264                 res = -EAGAIN;
1265                 if (!p->fs->in_exec) {
1266                         p->fs->in_exec = 1;
1267                         res = 1;
1268                 }
1269         }
1270         spin_unlock(&p->fs->lock);
1271
1272         return res;
1273 }
1274
1275 /*
1276  * Fill the binprm structure from the inode.
1277  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1278  *
1279  * This may be called multiple times for binary chains (scripts for example).
1280  */
1281 int prepare_binprm(struct linux_binprm *bprm)
1282 {
1283         umode_t mode;
1284         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1285         int retval;
1286
1287         mode = inode->i_mode;
1288         if (bprm->file->f_op == NULL)
1289                 return -EACCES;
1290
1291         /* clear any previous set[ug]id data from a previous binary */
1292         bprm->cred->euid = current_euid();
1293         bprm->cred->egid = current_egid();
1294
1295         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1296             !current->no_new_privs) {
1297                 /* Set-uid? */
1298                 if (mode & S_ISUID) {
1299                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1300                         bprm->cred->euid = inode->i_uid;
1301                 }
1302
1303                 /* Set-gid? */
1304                 /*
1305                  * If setgid is set but no group execute bit then this
1306                  * is a candidate for mandatory locking, not a setgid
1307                  * executable.
1308                  */
1309                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1310                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1311                         bprm->cred->egid = inode->i_gid;
1312                 }
1313         }
1314
1315         /* fill in binprm security blob */
1316         retval = security_bprm_set_creds(bprm);
1317         if (retval)
1318                 return retval;
1319         bprm->cred_prepared = 1;
1320
1321         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1322         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1323 }
1324
1325 EXPORT_SYMBOL(prepare_binprm);
1326
1327 /*
1328  * Arguments are '\0' separated strings found at the location bprm->p
1329  * points to; chop off the first by relocating brpm->p to right after
1330  * the first '\0' encountered.
1331  */
1332 int remove_arg_zero(struct linux_binprm *bprm)
1333 {
1334         int ret = 0;
1335         unsigned long offset;
1336         char *kaddr;
1337         struct page *page;
1338
1339         if (!bprm->argc)
1340                 return 0;
1341
1342         do {
1343                 offset = bprm->p & ~PAGE_MASK;
1344                 page = get_arg_page(bprm, bprm->p, 0);
1345                 if (!page) {
1346                         ret = -EFAULT;
1347                         goto out;
1348                 }
1349                 kaddr = kmap_atomic(page, KM_USER0);
1350
1351                 for (; offset < PAGE_SIZE && kaddr[offset];
1352                                 offset++, bprm->p++)
1353                         ;
1354
1355                 kunmap_atomic(kaddr, KM_USER0);
1356                 put_arg_page(page);
1357
1358                 if (offset == PAGE_SIZE)
1359                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1360         } while (offset == PAGE_SIZE);
1361
1362         bprm->p++;
1363         bprm->argc--;
1364         ret = 0;
1365
1366 out:
1367         return ret;
1368 }
1369 EXPORT_SYMBOL(remove_arg_zero);
1370
1371 /*
1372  * cycle the list of binary formats handler, until one recognizes the image
1373  */
1374 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1375 {
1376         unsigned int depth = bprm->recursion_depth;
1377         int try,retval;
1378         struct linux_binfmt *fmt;
1379         pid_t old_pid;
1380
1381         retval = security_bprm_check(bprm);
1382         if (retval)
1383                 return retval;
1384
1385         retval = audit_bprm(bprm);
1386         if (retval)
1387                 return retval;
1388
1389         /* Need to fetch pid before load_binary changes it */
1390         rcu_read_lock();
1391         old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1392         rcu_read_unlock();
1393
1394         retval = -ENOENT;
1395         for (try=0; try<2; try++) {
1396                 read_lock(&binfmt_lock);
1397                 list_for_each_entry(fmt, &formats, lh) {
1398                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1399                         if (!fn)
1400                                 continue;
1401                         if (!try_module_get(fmt->module))
1402                                 continue;
1403                         read_unlock(&binfmt_lock);
1404                         retval = fn(bprm, regs);
1405                         /*
1406                          * Restore the depth counter to its starting value
1407                          * in this call, so we don't have to rely on every
1408                          * load_binary function to restore it on return.
1409                          */
1410                         bprm->recursion_depth = depth;
1411                         if (retval >= 0) {
1412                                 if (depth == 0)
1413                                         ptrace_event(PTRACE_EVENT_EXEC,
1414                                                         old_pid);
1415                                 put_binfmt(fmt);
1416                                 allow_write_access(bprm->file);
1417                                 if (bprm->file)
1418                                         fput(bprm->file);
1419                                 bprm->file = NULL;
1420                                 current->did_exec = 1;
1421                                 proc_exec_connector(current);
1422                                 return retval;
1423                         }
1424                         read_lock(&binfmt_lock);
1425                         put_binfmt(fmt);
1426                         if (retval != -ENOEXEC || bprm->mm == NULL)
1427                                 break;
1428                         if (!bprm->file) {
1429                                 read_unlock(&binfmt_lock);
1430                                 return retval;
1431                         }
1432                 }
1433                 read_unlock(&binfmt_lock);
1434 #ifdef CONFIG_MODULES
1435                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1436                         break;
1437                 } else {
1438 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1439                         if (printable(bprm->buf[0]) &&
1440                             printable(bprm->buf[1]) &&
1441                             printable(bprm->buf[2]) &&
1442                             printable(bprm->buf[3]))
1443                                 break; /* -ENOEXEC */
1444                         if (try)
1445                                 break; /* -ENOEXEC */
1446                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1447                 }
1448 #else
1449                 break;
1450 #endif
1451         }
1452         return retval;
1453 }
1454
1455 EXPORT_SYMBOL(search_binary_handler);
1456
1457 /*
1458  * sys_execve() executes a new program.
1459  */
1460 static int do_execve_common(const char *filename,
1461                                 struct user_arg_ptr argv,
1462                                 struct user_arg_ptr envp,
1463                                 struct pt_regs *regs)
1464 {
1465         struct linux_binprm *bprm;
1466         struct file *file;
1467         struct files_struct *displaced;
1468         bool clear_in_exec;
1469         int retval;
1470         const struct cred *cred = current_cred();
1471
1472         /*
1473          * We move the actual failure in case of RLIMIT_NPROC excess from
1474          * set*uid() to execve() because too many poorly written programs
1475          * don't check setuid() return code.  Here we additionally recheck
1476          * whether NPROC limit is still exceeded.
1477          */
1478         if ((current->flags & PF_NPROC_EXCEEDED) &&
1479             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1480                 retval = -EAGAIN;
1481                 goto out_ret;
1482         }
1483
1484         /* We're below the limit (still or again), so we don't want to make
1485          * further execve() calls fail. */
1486         current->flags &= ~PF_NPROC_EXCEEDED;
1487
1488         retval = unshare_files(&displaced);
1489         if (retval)
1490                 goto out_ret;
1491
1492         retval = -ENOMEM;
1493         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1494         if (!bprm)
1495                 goto out_files;
1496
1497         retval = prepare_bprm_creds(bprm);
1498         if (retval)
1499                 goto out_free;
1500
1501         retval = check_unsafe_exec(bprm);
1502         if (retval < 0)
1503                 goto out_free;
1504         clear_in_exec = retval;
1505         current->in_execve = 1;
1506
1507         file = open_exec(filename);
1508         retval = PTR_ERR(file);
1509         if (IS_ERR(file))
1510                 goto out_unmark;
1511
1512         sched_exec();
1513
1514         bprm->file = file;
1515         bprm->filename = filename;
1516         bprm->interp = filename;
1517
1518         retval = bprm_mm_init(bprm);
1519         if (retval)
1520                 goto out_file;
1521
1522         bprm->argc = count(argv, MAX_ARG_STRINGS);
1523         if ((retval = bprm->argc) < 0)
1524                 goto out;
1525
1526         bprm->envc = count(envp, MAX_ARG_STRINGS);
1527         if ((retval = bprm->envc) < 0)
1528                 goto out;
1529
1530         retval = prepare_binprm(bprm);
1531         if (retval < 0)
1532                 goto out;
1533
1534         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1535         if (retval < 0)
1536                 goto out;
1537
1538         bprm->exec = bprm->p;
1539         retval = copy_strings(bprm->envc, envp, bprm);
1540         if (retval < 0)
1541                 goto out;
1542
1543         retval = copy_strings(bprm->argc, argv, bprm);
1544         if (retval < 0)
1545                 goto out;
1546
1547         retval = search_binary_handler(bprm,regs);
1548         if (retval < 0)
1549                 goto out;
1550
1551         /* execve succeeded */
1552         current->fs->in_exec = 0;
1553         current->in_execve = 0;
1554         acct_update_integrals(current);
1555         free_bprm(bprm);
1556         if (displaced)
1557                 put_files_struct(displaced);
1558         return retval;
1559
1560 out:
1561         if (bprm->mm) {
1562                 acct_arg_size(bprm, 0);
1563                 mmput(bprm->mm);
1564         }
1565
1566 out_file:
1567         if (bprm->file) {
1568                 allow_write_access(bprm->file);
1569                 fput(bprm->file);
1570         }
1571
1572 out_unmark:
1573         if (clear_in_exec)
1574                 current->fs->in_exec = 0;
1575         current->in_execve = 0;
1576
1577 out_free:
1578         free_bprm(bprm);
1579
1580 out_files:
1581         if (displaced)
1582                 reset_files_struct(displaced);
1583 out_ret:
1584         return retval;
1585 }
1586
1587 int do_execve(const char *filename,
1588         const char __user *const __user *__argv,
1589         const char __user *const __user *__envp,
1590         struct pt_regs *regs)
1591 {
1592         struct user_arg_ptr argv = { .ptr.native = __argv };
1593         struct user_arg_ptr envp = { .ptr.native = __envp };
1594         return do_execve_common(filename, argv, envp, regs);
1595 }
1596
1597 #ifdef CONFIG_COMPAT
1598 int compat_do_execve(char *filename,
1599         compat_uptr_t __user *__argv,
1600         compat_uptr_t __user *__envp,
1601         struct pt_regs *regs)
1602 {
1603         struct user_arg_ptr argv = {
1604                 .is_compat = true,
1605                 .ptr.compat = __argv,
1606         };
1607         struct user_arg_ptr envp = {
1608                 .is_compat = true,
1609                 .ptr.compat = __envp,
1610         };
1611         return do_execve_common(filename, argv, envp, regs);
1612 }
1613 #endif
1614
1615 void set_binfmt(struct linux_binfmt *new)
1616 {
1617         struct mm_struct *mm = current->mm;
1618
1619         if (mm->binfmt)
1620                 module_put(mm->binfmt->module);
1621
1622         mm->binfmt = new;
1623         if (new)
1624                 __module_get(new->module);
1625 }
1626
1627 EXPORT_SYMBOL(set_binfmt);
1628
1629 static int expand_corename(struct core_name *cn)
1630 {
1631         char *old_corename = cn->corename;
1632
1633         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1634         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1635
1636         if (!cn->corename) {
1637                 kfree(old_corename);
1638                 return -ENOMEM;
1639         }
1640
1641         return 0;
1642 }
1643
1644 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1645 {
1646         char *cur;
1647         int need;
1648         int ret;
1649         va_list arg;
1650
1651         va_start(arg, fmt);
1652         need = vsnprintf(NULL, 0, fmt, arg);
1653         va_end(arg);
1654
1655         if (likely(need < cn->size - cn->used - 1))
1656                 goto out_printf;
1657
1658         ret = expand_corename(cn);
1659         if (ret)
1660                 goto expand_fail;
1661
1662 out_printf:
1663         cur = cn->corename + cn->used;
1664         va_start(arg, fmt);
1665         vsnprintf(cur, need + 1, fmt, arg);
1666         va_end(arg);
1667         cn->used += need;
1668         return 0;
1669
1670 expand_fail:
1671         return ret;
1672 }
1673
1674 static void cn_escape(char *str)
1675 {
1676         for (; *str; str++)
1677                 if (*str == '/')
1678                         *str = '!';
1679 }
1680
1681 static int cn_print_exe_file(struct core_name *cn)
1682 {
1683         struct file *exe_file;
1684         char *pathbuf, *path;
1685         int ret;
1686
1687         exe_file = get_mm_exe_file(current->mm);
1688         if (!exe_file) {
1689                 char *commstart = cn->corename + cn->used;
1690                 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1691                 cn_escape(commstart);
1692                 return ret;
1693         }
1694
1695         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1696         if (!pathbuf) {
1697                 ret = -ENOMEM;
1698                 goto put_exe_file;
1699         }
1700
1701         path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1702         if (IS_ERR(path)) {
1703                 ret = PTR_ERR(path);
1704                 goto free_buf;
1705         }
1706
1707         cn_escape(path);
1708
1709         ret = cn_printf(cn, "%s", path);
1710
1711 free_buf:
1712         kfree(pathbuf);
1713 put_exe_file:
1714         fput(exe_file);
1715         return ret;
1716 }
1717
1718 /* format_corename will inspect the pattern parameter, and output a
1719  * name into corename, which must have space for at least
1720  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1721  */
1722 static int format_corename(struct core_name *cn, long signr)
1723 {
1724         const struct cred *cred = current_cred();
1725         const char *pat_ptr = core_pattern;
1726         int ispipe = (*pat_ptr == '|');
1727         int pid_in_pattern = 0;
1728         int err = 0;
1729
1730         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1731         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1732         cn->used = 0;
1733
1734         if (!cn->corename)
1735                 return -ENOMEM;
1736
1737         /* Repeat as long as we have more pattern to process and more output
1738            space */
1739         while (*pat_ptr) {
1740                 if (*pat_ptr != '%') {
1741                         if (*pat_ptr == 0)
1742                                 goto out;
1743                         err = cn_printf(cn, "%c", *pat_ptr++);
1744                 } else {
1745                         switch (*++pat_ptr) {
1746                         /* single % at the end, drop that */
1747                         case 0:
1748                                 goto out;
1749                         /* Double percent, output one percent */
1750                         case '%':
1751                                 err = cn_printf(cn, "%c", '%');
1752                                 break;
1753                         /* pid */
1754                         case 'p':
1755                                 pid_in_pattern = 1;
1756                                 err = cn_printf(cn, "%d",
1757                                               task_tgid_vnr(current));
1758                                 break;
1759                         /* uid */
1760                         case 'u':
1761                                 err = cn_printf(cn, "%d", cred->uid);
1762                                 break;
1763                         /* gid */
1764                         case 'g':
1765                                 err = cn_printf(cn, "%d", cred->gid);
1766                                 break;
1767                         /* signal that caused the coredump */
1768                         case 's':
1769                                 err = cn_printf(cn, "%ld", signr);
1770                                 break;
1771                         /* UNIX time of coredump */
1772                         case 't': {
1773                                 struct timeval tv;
1774                                 do_gettimeofday(&tv);
1775                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1776                                 break;
1777                         }
1778                         /* hostname */
1779                         case 'h': {
1780                                 char *namestart = cn->corename + cn->used;
1781                                 down_read(&uts_sem);
1782                                 err = cn_printf(cn, "%s",
1783                                               utsname()->nodename);
1784                                 up_read(&uts_sem);
1785                                 cn_escape(namestart);
1786                                 break;
1787                         }
1788                         /* executable */
1789                         case 'e': {
1790                                 char *commstart = cn->corename + cn->used;
1791                                 err = cn_printf(cn, "%s", current->comm);
1792                                 cn_escape(commstart);
1793                                 break;
1794                         }
1795                         case 'E':
1796                                 err = cn_print_exe_file(cn);
1797                                 break;
1798                         /* core limit size */
1799                         case 'c':
1800                                 err = cn_printf(cn, "%lu",
1801                                               rlimit(RLIMIT_CORE));
1802                                 break;
1803                         default:
1804                                 break;
1805                         }
1806                         ++pat_ptr;
1807                 }
1808
1809                 if (err)
1810                         return err;
1811         }
1812
1813         /* Backward compatibility with core_uses_pid:
1814          *
1815          * If core_pattern does not include a %p (as is the default)
1816          * and core_uses_pid is set, then .%pid will be appended to
1817          * the filename. Do not do this for piped commands. */
1818         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1819                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1820                 if (err)
1821                         return err;
1822         }
1823 out:
1824         return ispipe;
1825 }
1826
1827 static int zap_process(struct task_struct *start, int exit_code)
1828 {
1829         struct task_struct *t;
1830         int nr = 0;
1831
1832         start->signal->flags = SIGNAL_GROUP_EXIT;
1833         start->signal->group_exit_code = exit_code;
1834         start->signal->group_stop_count = 0;
1835
1836         t = start;
1837         do {
1838                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1839                 if (t != current && t->mm) {
1840                         sigaddset(&t->pending.signal, SIGKILL);
1841                         signal_wake_up(t, 1);
1842                         nr++;
1843                 }
1844         } while_each_thread(start, t);
1845
1846         return nr;
1847 }
1848
1849 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1850                                 struct core_state *core_state, int exit_code)
1851 {
1852         struct task_struct *g, *p;
1853         unsigned long flags;
1854         int nr = -EAGAIN;
1855
1856         spin_lock_irq(&tsk->sighand->siglock);
1857         if (!signal_group_exit(tsk->signal)) {
1858                 mm->core_state = core_state;
1859                 nr = zap_process(tsk, exit_code);
1860         }
1861         spin_unlock_irq(&tsk->sighand->siglock);
1862         if (unlikely(nr < 0))
1863                 return nr;
1864
1865         if (atomic_read(&mm->mm_users) == nr + 1)
1866                 goto done;
1867         /*
1868          * We should find and kill all tasks which use this mm, and we should
1869          * count them correctly into ->nr_threads. We don't take tasklist
1870          * lock, but this is safe wrt:
1871          *
1872          * fork:
1873          *      None of sub-threads can fork after zap_process(leader). All
1874          *      processes which were created before this point should be
1875          *      visible to zap_threads() because copy_process() adds the new
1876          *      process to the tail of init_task.tasks list, and lock/unlock
1877          *      of ->siglock provides a memory barrier.
1878          *
1879          * do_exit:
1880          *      The caller holds mm->mmap_sem. This means that the task which
1881          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1882          *      its ->mm.
1883          *
1884          * de_thread:
1885          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1886          *      we must see either old or new leader, this does not matter.
1887          *      However, it can change p->sighand, so lock_task_sighand(p)
1888          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1889          *      it can't fail.
1890          *
1891          *      Note also that "g" can be the old leader with ->mm == NULL
1892          *      and already unhashed and thus removed from ->thread_group.
1893          *      This is OK, __unhash_process()->list_del_rcu() does not
1894          *      clear the ->next pointer, we will find the new leader via
1895          *      next_thread().
1896          */
1897         rcu_read_lock();
1898         for_each_process(g) {
1899                 if (g == tsk->group_leader)
1900                         continue;
1901                 if (g->flags & PF_KTHREAD)
1902                         continue;
1903                 p = g;
1904                 do {
1905                         if (p->mm) {
1906                                 if (unlikely(p->mm == mm)) {
1907                                         lock_task_sighand(p, &flags);
1908                                         nr += zap_process(p, exit_code);
1909                                         unlock_task_sighand(p, &flags);
1910                                 }
1911                                 break;
1912                         }
1913                 } while_each_thread(g, p);
1914         }
1915         rcu_read_unlock();
1916 done:
1917         atomic_set(&core_state->nr_threads, nr);
1918         return nr;
1919 }
1920
1921 static int coredump_wait(int exit_code, struct core_state *core_state)
1922 {
1923         struct task_struct *tsk = current;
1924         struct mm_struct *mm = tsk->mm;
1925         struct completion *vfork_done;
1926         int core_waiters = -EBUSY;
1927
1928         init_completion(&core_state->startup);
1929         core_state->dumper.task = tsk;
1930         core_state->dumper.next = NULL;
1931
1932         down_write(&mm->mmap_sem);
1933         if (!mm->core_state)
1934                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1935         up_write(&mm->mmap_sem);
1936
1937         if (unlikely(core_waiters < 0))
1938                 goto fail;
1939
1940         /*
1941          * Make sure nobody is waiting for us to release the VM,
1942          * otherwise we can deadlock when we wait on each other
1943          */
1944         vfork_done = tsk->vfork_done;
1945         if (vfork_done) {
1946                 tsk->vfork_done = NULL;
1947                 complete(vfork_done);
1948         }
1949
1950         if (core_waiters)
1951                 wait_for_completion(&core_state->startup);
1952 fail:
1953         return core_waiters;
1954 }
1955
1956 static void coredump_finish(struct mm_struct *mm)
1957 {
1958         struct core_thread *curr, *next;
1959         struct task_struct *task;
1960
1961         next = mm->core_state->dumper.next;
1962         while ((curr = next) != NULL) {
1963                 next = curr->next;
1964                 task = curr->task;
1965                 /*
1966                  * see exit_mm(), curr->task must not see
1967                  * ->task == NULL before we read ->next.
1968                  */
1969                 smp_mb();
1970                 curr->task = NULL;
1971                 wake_up_process(task);
1972         }
1973
1974         mm->core_state = NULL;
1975 }
1976
1977 /*
1978  * set_dumpable converts traditional three-value dumpable to two flags and
1979  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1980  * these bits are not changed atomically.  So get_dumpable can observe the
1981  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1982  * return either old dumpable or new one by paying attention to the order of
1983  * modifying the bits.
1984  *
1985  * dumpable |   mm->flags (binary)
1986  * old  new | initial interim  final
1987  * ---------+-----------------------
1988  *  0    1  |   00      01      01
1989  *  0    2  |   00      10(*)   11
1990  *  1    0  |   01      00      00
1991  *  1    2  |   01      11      11
1992  *  2    0  |   11      10(*)   00
1993  *  2    1  |   11      11      01
1994  *
1995  * (*) get_dumpable regards interim value of 10 as 11.
1996  */
1997 void set_dumpable(struct mm_struct *mm, int value)
1998 {
1999         switch (value) {
2000         case 0:
2001                 clear_bit(MMF_DUMPABLE, &mm->flags);
2002                 smp_wmb();
2003                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2004                 break;
2005         case 1:
2006                 set_bit(MMF_DUMPABLE, &mm->flags);
2007                 smp_wmb();
2008                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2009                 break;
2010         case 2:
2011                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2012                 smp_wmb();
2013                 set_bit(MMF_DUMPABLE, &mm->flags);
2014                 break;
2015         }
2016 }
2017
2018 static int __get_dumpable(unsigned long mm_flags)
2019 {
2020         int ret;
2021
2022         ret = mm_flags & MMF_DUMPABLE_MASK;
2023         return (ret >= 2) ? 2 : ret;
2024 }
2025
2026 int get_dumpable(struct mm_struct *mm)
2027 {
2028         return __get_dumpable(mm->flags);
2029 }
2030
2031 static void wait_for_dump_helpers(struct file *file)
2032 {
2033         struct pipe_inode_info *pipe;
2034
2035         pipe = file->f_path.dentry->d_inode->i_pipe;
2036
2037         pipe_lock(pipe);
2038         pipe->readers++;
2039         pipe->writers--;
2040
2041         while ((pipe->readers > 1) && (!signal_pending(current))) {
2042                 wake_up_interruptible_sync(&pipe->wait);
2043                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2044                 pipe_wait(pipe);
2045         }
2046
2047         pipe->readers--;
2048         pipe->writers++;
2049         pipe_unlock(pipe);
2050
2051 }
2052
2053
2054 /*
2055  * umh_pipe_setup
2056  * helper function to customize the process used
2057  * to collect the core in userspace.  Specifically
2058  * it sets up a pipe and installs it as fd 0 (stdin)
2059  * for the process.  Returns 0 on success, or
2060  * PTR_ERR on failure.
2061  * Note that it also sets the core limit to 1.  This
2062  * is a special value that we use to trap recursive
2063  * core dumps
2064  */
2065 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2066 {
2067         struct file *rp, *wp;
2068         struct fdtable *fdt;
2069         struct coredump_params *cp = (struct coredump_params *)info->data;
2070         struct files_struct *cf = current->files;
2071
2072         wp = create_write_pipe(0);
2073         if (IS_ERR(wp))
2074                 return PTR_ERR(wp);
2075
2076         rp = create_read_pipe(wp, 0);
2077         if (IS_ERR(rp)) {
2078                 free_write_pipe(wp);
2079                 return PTR_ERR(rp);
2080         }
2081
2082         cp->file = wp;
2083
2084         sys_close(0);
2085         fd_install(0, rp);
2086         spin_lock(&cf->file_lock);
2087         fdt = files_fdtable(cf);
2088         FD_SET(0, fdt->open_fds);
2089         FD_CLR(0, fdt->close_on_exec);
2090         spin_unlock(&cf->file_lock);
2091
2092         /* and disallow core files too */
2093         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2094
2095         return 0;
2096 }
2097
2098 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2099 {
2100         struct core_state core_state;
2101         struct core_name cn;
2102         struct mm_struct *mm = current->mm;
2103         struct linux_binfmt * binfmt;
2104         const struct cred *old_cred;
2105         struct cred *cred;
2106         int retval = 0;
2107         int flag = 0;
2108         int ispipe;
2109         static atomic_t core_dump_count = ATOMIC_INIT(0);
2110         struct coredump_params cprm = {
2111                 .signr = signr,
2112                 .regs = regs,
2113                 .limit = rlimit(RLIMIT_CORE),
2114                 /*
2115                  * We must use the same mm->flags while dumping core to avoid
2116                  * inconsistency of bit flags, since this flag is not protected
2117                  * by any locks.
2118                  */
2119                 .mm_flags = mm->flags,
2120         };
2121
2122         audit_core_dumps(signr);
2123
2124         binfmt = mm->binfmt;
2125         if (!binfmt || !binfmt->core_dump)
2126                 goto fail;
2127         if (!__get_dumpable(cprm.mm_flags))
2128                 goto fail;
2129
2130         cred = prepare_creds();
2131         if (!cred)
2132                 goto fail;
2133         /*
2134          *      We cannot trust fsuid as being the "true" uid of the
2135          *      process nor do we know its entire history. We only know it
2136          *      was tainted so we dump it as root in mode 2.
2137          */
2138         if (__get_dumpable(cprm.mm_flags) == 2) {
2139                 /* Setuid core dump mode */
2140                 flag = O_EXCL;          /* Stop rewrite attacks */
2141                 cred->fsuid = 0;        /* Dump root private */
2142         }
2143
2144         retval = coredump_wait(exit_code, &core_state);
2145         if (retval < 0)
2146                 goto fail_creds;
2147
2148         old_cred = override_creds(cred);
2149
2150         /*
2151          * Clear any false indication of pending signals that might
2152          * be seen by the filesystem code called to write the core file.
2153          */
2154         clear_thread_flag(TIF_SIGPENDING);
2155
2156         ispipe = format_corename(&cn, signr);
2157
2158         if (ispipe) {
2159                 int dump_count;
2160                 char **helper_argv;
2161
2162                 if (ispipe < 0) {
2163                         printk(KERN_WARNING "format_corename failed\n");
2164                         printk(KERN_WARNING "Aborting core\n");
2165                         goto fail_corename;
2166                 }
2167
2168                 if (cprm.limit == 1) {
2169                         /*
2170                          * Normally core limits are irrelevant to pipes, since
2171                          * we're not writing to the file system, but we use
2172                          * cprm.limit of 1 here as a speacial value. Any
2173                          * non-1 limit gets set to RLIM_INFINITY below, but
2174                          * a limit of 0 skips the dump.  This is a consistent
2175                          * way to catch recursive crashes.  We can still crash
2176                          * if the core_pattern binary sets RLIM_CORE =  !1
2177                          * but it runs as root, and can do lots of stupid things
2178                          * Note that we use task_tgid_vnr here to grab the pid
2179                          * of the process group leader.  That way we get the
2180                          * right pid if a thread in a multi-threaded
2181                          * core_pattern process dies.
2182                          */
2183                         printk(KERN_WARNING
2184                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2185                                 task_tgid_vnr(current), current->comm);
2186                         printk(KERN_WARNING "Aborting core\n");
2187                         goto fail_unlock;
2188                 }
2189                 cprm.limit = RLIM_INFINITY;
2190
2191                 dump_count = atomic_inc_return(&core_dump_count);
2192                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2193                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2194                                task_tgid_vnr(current), current->comm);
2195                         printk(KERN_WARNING "Skipping core dump\n");
2196                         goto fail_dropcount;
2197                 }
2198
2199                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2200                 if (!helper_argv) {
2201                         printk(KERN_WARNING "%s failed to allocate memory\n",
2202                                __func__);
2203                         goto fail_dropcount;
2204                 }
2205
2206                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2207                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2208                                         NULL, &cprm);
2209                 argv_free(helper_argv);
2210                 if (retval) {
2211                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2212                                cn.corename);
2213                         goto close_fail;
2214                 }
2215         } else {
2216                 struct inode *inode;
2217
2218                 if (cprm.limit < binfmt->min_coredump)
2219                         goto fail_unlock;
2220
2221                 cprm.file = filp_open(cn.corename,
2222                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2223                                  0600);
2224                 if (IS_ERR(cprm.file))
2225                         goto fail_unlock;
2226
2227                 inode = cprm.file->f_path.dentry->d_inode;
2228                 if (inode->i_nlink > 1)
2229                         goto close_fail;
2230                 if (d_unhashed(cprm.file->f_path.dentry))
2231                         goto close_fail;
2232                 /*
2233                  * AK: actually i see no reason to not allow this for named
2234                  * pipes etc, but keep the previous behaviour for now.
2235                  */
2236                 if (!S_ISREG(inode->i_mode))
2237                         goto close_fail;
2238                 /*
2239                  * Dont allow local users get cute and trick others to coredump
2240                  * into their pre-created files.
2241                  */
2242                 if (inode->i_uid != current_fsuid())
2243                         goto close_fail;
2244                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2245                         goto close_fail;
2246                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2247                         goto close_fail;
2248         }
2249
2250         retval = binfmt->core_dump(&cprm);
2251         if (retval)
2252                 current->signal->group_exit_code |= 0x80;
2253
2254         if (ispipe && core_pipe_limit)
2255                 wait_for_dump_helpers(cprm.file);
2256 close_fail:
2257         if (cprm.file)
2258                 filp_close(cprm.file, NULL);
2259 fail_dropcount:
2260         if (ispipe)
2261                 atomic_dec(&core_dump_count);
2262 fail_unlock:
2263         kfree(cn.corename);
2264 fail_corename:
2265         coredump_finish(mm);
2266         revert_creds(old_cred);
2267 fail_creds:
2268         put_cred(cred);
2269 fail:
2270         return;
2271 }
2272
2273 /*
2274  * Core dumping helper functions.  These are the only things you should
2275  * do on a core-file: use only these functions to write out all the
2276  * necessary info.
2277  */
2278 int dump_write(struct file *file, const void *addr, int nr)
2279 {
2280         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2281 }
2282 EXPORT_SYMBOL(dump_write);
2283
2284 int dump_seek(struct file *file, loff_t off)
2285 {
2286         int ret = 1;
2287
2288         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2289                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2290                         return 0;
2291         } else {
2292                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2293
2294                 if (!buf)
2295                         return 0;
2296                 while (off > 0) {
2297                         unsigned long n = off;
2298
2299                         if (n > PAGE_SIZE)
2300                                 n = PAGE_SIZE;
2301                         if (!dump_write(file, buf, n)) {
2302                                 ret = 0;
2303                                 break;
2304                         }
2305                         off -= n;
2306                 }
2307                 free_page((unsigned long)buf);
2308         }
2309         return ret;
2310 }
2311 EXPORT_SYMBOL(dump_seek);