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
  63 #include <trace/events/task.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         err = deny_write_access(file);
 788         if (err)
 789                 goto exit;
 790
 791 out:
 792         return file;
 793
 794 exit:
 795         fput(file);
 796         return ERR_PTR(err);
 797 }
 798 EXPORT_SYMBOL(open_exec);
 799
 800 int kernel_read(struct file *file, loff_t offset,
 801                 char *addr, unsigned long count)
 802 {
 803         mm_segment_t old_fs;
 804         loff_t pos = offset;
 805         int result;
 806
 807         old_fs = get_fs();
 808         set_fs(get_ds());
 809         /* The cast to a user pointer is valid due to the set_fs() */
 810         result = vfs_read(file, (void __user *)addr, count, &pos);
 811         set_fs(old_fs);
 812         return result;
 813 }
 814
 815 EXPORT_SYMBOL(kernel_read);
 816
 817 static int exec_mmap(struct mm_struct *mm)
 818 {
 819         struct task_struct *tsk;
 820         struct mm_struct * old_mm, *active_mm;
 821
 822         /* Notify parent that we're no longer interested in the old VM */
 823         tsk = current;
 824         old_mm = current->mm;
 825         sync_mm_rss(tsk, old_mm);
 826         mm_release(tsk, old_mm);
 827
 828         if (old_mm) {
 829                 /*
 830                  * Make sure that if there is a core dump in progress
 831                  * for the old mm, we get out and die instead of going
 832                  * through with the exec.  We must hold mmap_sem around
 833                  * checking core_state and changing tsk->mm.
 834                  */
 835                 down_read(&old_mm->mmap_sem);
 836                 if (unlikely(old_mm->core_state)) {
 837                         up_read(&old_mm->mmap_sem);
 838                         return -EINTR;
 839                 }
 840         }
 841         task_lock(tsk);
 842         active_mm = tsk->active_mm;
 843         tsk->mm = mm;
 844         tsk->active_mm = mm;
 845         activate_mm(active_mm, mm);
 846         task_unlock(tsk);
 847         arch_pick_mmap_layout(mm);
 848         if (old_mm) {
 849                 up_read(&old_mm->mmap_sem);
 850                 BUG_ON(active_mm != old_mm);
 851                 mm_update_next_owner(old_mm);
 852                 mmput(old_mm);
 853                 return 0;
 854         }
 855         mmdrop(active_mm);
 856         return 0;
 857 }
 858
 859 /*
 860  * This function makes sure the current process has its own signal table,
 861  * so that flush_signal_handlers can later reset the handlers without
 862  * disturbing other processes.  (Other processes might share the signal
 863  * table via the CLONE_SIGHAND option to clone().)
 864  */
 865 static int de_thread(struct task_struct *tsk)
 866 {
 867         struct signal_struct *sig = tsk->signal;
 868         struct sighand_struct *oldsighand = tsk->sighand;
 869         spinlock_t *lock = &oldsighand->siglock;
 870
 871         if (thread_group_empty(tsk))
 872                 goto no_thread_group;
 873
 874         /*
 875          * Kill all other threads in the thread group.
 876          */
 877         spin_lock_irq(lock);
 878         if (signal_group_exit(sig)) {
 879                 /*
 880                  * Another group action in progress, just
 881                  * return so that the signal is processed.
 882                  */
 883                 spin_unlock_irq(lock);
 884                 return -EAGAIN;
 885         }
 886
 887         sig->group_exit_task = tsk;
 888         sig->notify_count = zap_other_threads(tsk);
 889         if (!thread_group_leader(tsk))
 890                 sig->notify_count--;
 891
 892         while (sig->notify_count) {
 893                 __set_current_state(TASK_UNINTERRUPTIBLE);
 894                 spin_unlock_irq(lock);
 895                 schedule();
 896                 spin_lock_irq(lock);
 897         }
 898         spin_unlock_irq(lock);
 899
 900         /*
 901          * At this point all other threads have exited, all we have to
 902          * do is to wait for the thread group leader to become inactive,
 903          * and to assume its PID:
 904          */
 905         if (!thread_group_leader(tsk)) {
 906                 struct task_struct *leader = tsk->group_leader;
 907
 908                 sig->notify_count = -1; /* for exit_notify() */
 909                 for (;;) {
 910                         write_lock_irq(&tasklist_lock);
 911                         if (likely(leader->exit_state))
 912                                 break;
 913                         __set_current_state(TASK_UNINTERRUPTIBLE);
 914                         write_unlock_irq(&tasklist_lock);
 915                         schedule();
 916                 }
 917
 918                 /*
 919                  * The only record we have of the real-time age of a
 920                  * process, regardless of execs it's done, is start_time.
 921                  * All the past CPU time is accumulated in signal_struct
 922                  * from sister threads now dead.  But in this non-leader
 923                  * exec, nothing survives from the original leader thread,
 924                  * whose birth marks the true age of this process now.
 925                  * When we take on its identity by switching to its PID, we
 926                  * also take its birthdate (always earlier than our own).
 927                  */
 928                 tsk->start_time = leader->start_time;
 929
 930                 BUG_ON(!same_thread_group(leader, tsk));
 931                 BUG_ON(has_group_leader_pid(tsk));
 932                 /*
 933                  * An exec() starts a new thread group with the
 934                  * TGID of the previous thread group. Rehash the
 935                  * two threads with a switched PID, and release
 936                  * the former thread group leader:
 937                  */
 938
 939                 /* Become a process group leader with the old leader's pid.
 940                  * The old leader becomes a thread of the this thread group.
 941                  * Note: The old leader also uses this pid until release_task
 942                  *       is called.  Odd but simple and correct.
 943                  */
 944                 detach_pid(tsk, PIDTYPE_PID);
 945                 tsk->pid = leader->pid;
 946                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 947                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 948                 transfer_pid(leader, tsk, PIDTYPE_SID);
 949
 950                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 951                 list_replace_init(&leader->sibling, &tsk->sibling);
 952
 953                 tsk->group_leader = tsk;
 954                 leader->group_leader = tsk;
 955
 956                 tsk->exit_signal = SIGCHLD;
 957                 leader->exit_signal = -1;
 958
 959                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 960                 leader->exit_state = EXIT_DEAD;
 961
 962                 /*
 963                  * We are going to release_task()->ptrace_unlink() silently,
 964                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 965                  * the tracer wont't block again waiting for this thread.
 966                  */
 967                 if (unlikely(leader->ptrace))
 968                         __wake_up_parent(leader, leader->parent);
 969                 write_unlock_irq(&tasklist_lock);
 970
 971                 release_task(leader);
 972         }
 973
 974         sig->group_exit_task = NULL;
 975         sig->notify_count = 0;
 976
 977 no_thread_group:
 978         if (current->mm)
 979                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 980
 981         exit_itimers(sig);
 982         flush_itimer_signals();
 983
 984         if (atomic_read(&oldsighand->count) != 1) {
 985                 struct sighand_struct *newsighand;
 986                 /*
 987                  * This ->sighand is shared with the CLONE_SIGHAND
 988                  * but not CLONE_THREAD task, switch to the new one.
 989                  */
 990                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 991                 if (!newsighand)
 992                         return -ENOMEM;
 993
 994                 atomic_set(&newsighand->count, 1);
 995                 memcpy(newsighand->action, oldsighand->action,
 996                        sizeof(newsighand->action));
 997
 998                 write_lock_irq(&tasklist_lock);
 999                 spin_lock(&oldsighand->siglock);
1000                 rcu_assign_pointer(tsk->sighand, newsighand);
1001                 spin_unlock(&oldsighand->siglock);
1002                 write_unlock_irq(&tasklist_lock);
1003
1004                 __cleanup_sighand(oldsighand);
1005         }
1006
1007         BUG_ON(!thread_group_leader(tsk));
1008         return 0;
1009 }
1010
1011 /*
1012  * These functions flushes out all traces of the currently running executable
1013  * so that a new one can be started
1014  */
1015 static void flush_old_files(struct files_struct * files)
1016 {
1017         long j = -1;
1018         struct fdtable *fdt;
1019
1020         spin_lock(&files->file_lock);
1021         for (;;) {
1022                 unsigned long set, i;
1023
1024                 j++;
1025                 i = j * __NFDBITS;
1026                 fdt = files_fdtable(files);
1027                 if (i >= fdt->max_fds)
1028                         break;
1029                 set = fdt->close_on_exec->fds_bits[j];
1030                 if (!set)
1031                         continue;
1032                 fdt->close_on_exec->fds_bits[j] = 0;
1033                 spin_unlock(&files->file_lock);
1034                 for ( ; set ; i++,set >>= 1) {
1035                         if (set & 1) {
1036                                 sys_close(i);
1037                         }
1038                 }
1039                 spin_lock(&files->file_lock);
1040
1041         }
1042         spin_unlock(&files->file_lock);
1043 }
1044
1045 char *get_task_comm(char *buf, struct task_struct *tsk)
1046 {
1047         /* buf must be at least sizeof(tsk->comm) in size */
1048         task_lock(tsk);
1049         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1050         task_unlock(tsk);
1051         return buf;
1052 }
1053 EXPORT_SYMBOL_GPL(get_task_comm);
1054
1055 void set_task_comm(struct task_struct *tsk, char *buf)
1056 {
1057         task_lock(tsk);
1058
1059         trace_task_rename(tsk, buf);
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 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1075 {
1076         int i, ch;
1077
1078         /* Copies the binary name from after last slash */
1079         for (i = 0; (ch = *(fn++)) != '\0';) {
1080                 if (ch == '/')
1081                         i = 0; /* overwrite what we wrote */
1082                 else
1083                         if (i < len - 1)
1084                                 tcomm[i++] = ch;
1085         }
1086         tcomm[i] = '\0';
1087 }
1088
1089 int flush_old_exec(struct linux_binprm * bprm)
1090 {
1091         int retval;
1092
1093         /*
1094          * Make sure we have a private signal table and that
1095          * we are unassociated from the previous thread group.
1096          */
1097         retval = de_thread(current);
1098         if (retval)
1099                 goto out;
1100
1101         set_mm_exe_file(bprm->mm, bprm->file);
1102
1103         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1104         /*
1105          * Release all of the old mmap stuff
1106          */
1107         acct_arg_size(bprm, 0);
1108         retval = exec_mmap(bprm->mm);
1109         if (retval)
1110                 goto out;
1111
1112         bprm->mm = NULL;                /* We're using it now */
1113
1114         set_fs(USER_DS);
1115         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1116         flush_thread();
1117         current->personality &= ~bprm->per_clear;
1118
1119         return 0;
1120
1121 out:
1122         return retval;
1123 }
1124 EXPORT_SYMBOL(flush_old_exec);
1125
1126 void would_dump(struct linux_binprm *bprm, struct file *file)
1127 {
1128         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1129                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1130 }
1131 EXPORT_SYMBOL(would_dump);
1132
1133 void setup_new_exec(struct linux_binprm * bprm)
1134 {
1135         arch_pick_mmap_layout(current->mm);
1136
1137         /* This is the point of no return */
1138         current->sas_ss_sp = current->sas_ss_size = 0;
1139
1140         if (current_euid() == current_uid() && current_egid() == current_gid())
1141                 set_dumpable(current->mm, 1);
1142         else
1143                 set_dumpable(current->mm, suid_dumpable);
1144
1145         set_task_comm(current, bprm->tcomm);
1146
1147         /* Set the new mm task size. We have to do that late because it may
1148          * depend on TIF_32BIT which is only updated in flush_thread() on
1149          * some architectures like powerpc
1150          */
1151         current->mm->task_size = TASK_SIZE;
1152
1153         /* install the new credentials */
1154         if (bprm->cred->uid != current_euid() ||
1155             bprm->cred->gid != current_egid()) {
1156                 current->pdeath_signal = 0;
1157         } else {
1158                 would_dump(bprm, bprm->file);
1159                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1160                         set_dumpable(current->mm, suid_dumpable);
1161         }
1162
1163         /*
1164          * Flush performance counters when crossing a
1165          * security domain:
1166          */
1167         if (!get_dumpable(current->mm))
1168                 perf_event_exit_task(current);
1169
1170         /* An exec changes our domain. We are no longer part of the thread
1171            group */
1172
1173         current->self_exec_id++;
1174
1175         flush_signal_handlers(current, 0);
1176         flush_old_files(current->files);
1177 }
1178 EXPORT_SYMBOL(setup_new_exec);
1179
1180 /*
1181  * Prepare credentials and lock ->cred_guard_mutex.
1182  * install_exec_creds() commits the new creds and drops the lock.
1183  * Or, if exec fails before, free_bprm() should release ->cred and
1184  * and unlock.
1185  */
1186 int prepare_bprm_creds(struct linux_binprm *bprm)
1187 {
1188         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1189                 return -ERESTARTNOINTR;
1190
1191         bprm->cred = prepare_exec_creds();
1192         if (likely(bprm->cred))
1193                 return 0;
1194
1195         mutex_unlock(&current->signal->cred_guard_mutex);
1196         return -ENOMEM;
1197 }
1198
1199 void free_bprm(struct linux_binprm *bprm)
1200 {
1201         free_arg_pages(bprm);
1202         if (bprm->cred) {
1203                 mutex_unlock(&current->signal->cred_guard_mutex);
1204                 abort_creds(bprm->cred);
1205         }
1206         kfree(bprm);
1207 }
1208
1209 /*
1210  * install the new credentials for this executable
1211  */
1212 void install_exec_creds(struct linux_binprm *bprm)
1213 {
1214         security_bprm_committing_creds(bprm);
1215
1216         commit_creds(bprm->cred);
1217         bprm->cred = NULL;
1218         /*
1219          * cred_guard_mutex must be held at least to this point to prevent
1220          * ptrace_attach() from altering our determination of the task's
1221          * credentials; any time after this it may be unlocked.
1222          */
1223         security_bprm_committed_creds(bprm);
1224         mutex_unlock(&current->signal->cred_guard_mutex);
1225 }
1226 EXPORT_SYMBOL(install_exec_creds);
1227
1228 /*
1229  * determine how safe it is to execute the proposed program
1230  * - the caller must hold ->cred_guard_mutex to protect against
1231  *   PTRACE_ATTACH
1232  */
1233 static int check_unsafe_exec(struct linux_binprm *bprm)
1234 {
1235         struct task_struct *p = current, *t;
1236         unsigned n_fs;
1237         int res = 0;
1238
1239         if (p->ptrace) {
1240                 if (p->ptrace & PT_PTRACE_CAP)
1241                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1242                 else
1243                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1244         }
1245
1246         n_fs = 1;
1247         spin_lock(&p->fs->lock);
1248         rcu_read_lock();
1249         for (t = next_thread(p); t != p; t = next_thread(t)) {
1250                 if (t->fs == p->fs)
1251                         n_fs++;
1252         }
1253         rcu_read_unlock();
1254
1255         if (p->fs->users > n_fs) {
1256                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1257         } else {
1258                 res = -EAGAIN;
1259                 if (!p->fs->in_exec) {
1260                         p->fs->in_exec = 1;
1261                         res = 1;
1262                 }
1263         }
1264         spin_unlock(&p->fs->lock);
1265
1266         return res;
1267 }
1268
1269 /*
1270  * Fill the binprm structure from the inode.
1271  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1272  *
1273  * This may be called multiple times for binary chains (scripts for example).
1274  */
1275 int prepare_binprm(struct linux_binprm *bprm)
1276 {
1277         umode_t mode;
1278         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1279         int retval;
1280
1281         mode = inode->i_mode;
1282         if (bprm->file->f_op == NULL)
1283                 return -EACCES;
1284
1285         /* clear any previous set[ug]id data from a previous binary */
1286         bprm->cred->euid = current_euid();
1287         bprm->cred->egid = current_egid();
1288
1289         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1290                 /* Set-uid? */
1291                 if (mode & S_ISUID) {
1292                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1293                         bprm->cred->euid = inode->i_uid;
1294                 }
1295
1296                 /* Set-gid? */
1297                 /*
1298                  * If setgid is set but no group execute bit then this
1299                  * is a candidate for mandatory locking, not a setgid
1300                  * executable.
1301                  */
1302                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1303                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1304                         bprm->cred->egid = inode->i_gid;
1305                 }
1306         }
1307
1308         /* fill in binprm security blob */
1309         retval = security_bprm_set_creds(bprm);
1310         if (retval)
1311                 return retval;
1312         bprm->cred_prepared = 1;
1313
1314         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1315         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1316 }
1317
1318 EXPORT_SYMBOL(prepare_binprm);
1319
1320 /*
1321  * Arguments are '\0' separated strings found at the location bprm->p
1322  * points to; chop off the first by relocating brpm->p to right after
1323  * the first '\0' encountered.
1324  */
1325 int remove_arg_zero(struct linux_binprm *bprm)
1326 {
1327         int ret = 0;
1328         unsigned long offset;
1329         char *kaddr;
1330         struct page *page;
1331
1332         if (!bprm->argc)
1333                 return 0;
1334
1335         do {
1336                 offset = bprm->p & ~PAGE_MASK;
1337                 page = get_arg_page(bprm, bprm->p, 0);
1338                 if (!page) {
1339                         ret = -EFAULT;
1340                         goto out;
1341                 }
1342                 kaddr = kmap_atomic(page, KM_USER0);
1343
1344                 for (; offset < PAGE_SIZE && kaddr[offset];
1345                                 offset++, bprm->p++)
1346                         ;
1347
1348                 kunmap_atomic(kaddr, KM_USER0);
1349                 put_arg_page(page);
1350
1351                 if (offset == PAGE_SIZE)
1352                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1353         } while (offset == PAGE_SIZE);
1354
1355         bprm->p++;
1356         bprm->argc--;
1357         ret = 0;
1358
1359 out:
1360         return ret;
1361 }
1362 EXPORT_SYMBOL(remove_arg_zero);
1363
1364 /*
1365  * cycle the list of binary formats handler, until one recognizes the image
1366  */
1367 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1368 {
1369         unsigned int depth = bprm->recursion_depth;
1370         int try,retval;
1371         struct linux_binfmt *fmt;
1372         pid_t old_pid;
1373
1374         retval = security_bprm_check(bprm);
1375         if (retval)
1376                 return retval;
1377
1378         retval = audit_bprm(bprm);
1379         if (retval)
1380                 return retval;
1381
1382         /* Need to fetch pid before load_binary changes it */
1383         rcu_read_lock();
1384         old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1385         rcu_read_unlock();
1386
1387         retval = -ENOENT;
1388         for (try=0; try<2; try++) {
1389                 read_lock(&binfmt_lock);
1390                 list_for_each_entry(fmt, &formats, lh) {
1391                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1392                         if (!fn)
1393                                 continue;
1394                         if (!try_module_get(fmt->module))
1395                                 continue;
1396                         read_unlock(&binfmt_lock);
1397                         retval = fn(bprm, regs);
1398                         /*
1399                          * Restore the depth counter to its starting value
1400                          * in this call, so we don't have to rely on every
1401                          * load_binary function to restore it on return.
1402                          */
1403                         bprm->recursion_depth = depth;
1404                         if (retval >= 0) {
1405                                 if (depth == 0)
1406                                         ptrace_event(PTRACE_EVENT_EXEC,
1407                                                         old_pid);
1408                                 put_binfmt(fmt);
1409                                 allow_write_access(bprm->file);
1410                                 if (bprm->file)
1411                                         fput(bprm->file);
1412                                 bprm->file = NULL;
1413                                 current->did_exec = 1;
1414                                 proc_exec_connector(current);
1415                                 return retval;
1416                         }
1417                         read_lock(&binfmt_lock);
1418                         put_binfmt(fmt);
1419                         if (retval != -ENOEXEC || bprm->mm == NULL)
1420                                 break;
1421                         if (!bprm->file) {
1422                                 read_unlock(&binfmt_lock);
1423                                 return retval;
1424                         }
1425                 }
1426                 read_unlock(&binfmt_lock);
1427 #ifdef CONFIG_MODULES
1428                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1429                         break;
1430                 } else {
1431 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1432                         if (printable(bprm->buf[0]) &&
1433                             printable(bprm->buf[1]) &&
1434                             printable(bprm->buf[2]) &&
1435                             printable(bprm->buf[3]))
1436                                 break; /* -ENOEXEC */
1437                         if (try)
1438                                 break; /* -ENOEXEC */
1439                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1440                 }
1441 #else
1442                 break;
1443 #endif
1444         }
1445         return retval;
1446 }
1447
1448 EXPORT_SYMBOL(search_binary_handler);
1449
1450 /*
1451  * sys_execve() executes a new program.
1452  */
1453 static int do_execve_common(const char *filename,
1454                                 struct user_arg_ptr argv,
1455                                 struct user_arg_ptr envp,
1456                                 struct pt_regs *regs)
1457 {
1458         struct linux_binprm *bprm;
1459         struct file *file;
1460         struct files_struct *displaced;
1461         bool clear_in_exec;
1462         int retval;
1463         const struct cred *cred = current_cred();
1464
1465         /*
1466          * We move the actual failure in case of RLIMIT_NPROC excess from
1467          * set*uid() to execve() because too many poorly written programs
1468          * don't check setuid() return code.  Here we additionally recheck
1469          * whether NPROC limit is still exceeded.
1470          */
1471         if ((current->flags & PF_NPROC_EXCEEDED) &&
1472             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1473                 retval = -EAGAIN;
1474                 goto out_ret;
1475         }
1476
1477         /* We're below the limit (still or again), so we don't want to make
1478          * further execve() calls fail. */
1479         current->flags &= ~PF_NPROC_EXCEEDED;
1480
1481         retval = unshare_files(&displaced);
1482         if (retval)
1483                 goto out_ret;
1484
1485         retval = -ENOMEM;
1486         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1487         if (!bprm)
1488                 goto out_files;
1489
1490         retval = prepare_bprm_creds(bprm);
1491         if (retval)
1492                 goto out_free;
1493
1494         retval = check_unsafe_exec(bprm);
1495         if (retval < 0)
1496                 goto out_free;
1497         clear_in_exec = retval;
1498         current->in_execve = 1;
1499
1500         file = open_exec(filename);
1501         retval = PTR_ERR(file);
1502         if (IS_ERR(file))
1503                 goto out_unmark;
1504
1505         sched_exec();
1506
1507         bprm->file = file;
1508         bprm->filename = filename;
1509         bprm->interp = filename;
1510
1511         retval = bprm_mm_init(bprm);
1512         if (retval)
1513                 goto out_file;
1514
1515         bprm->argc = count(argv, MAX_ARG_STRINGS);
1516         if ((retval = bprm->argc) < 0)
1517                 goto out;
1518
1519         bprm->envc = count(envp, MAX_ARG_STRINGS);
1520         if ((retval = bprm->envc) < 0)
1521                 goto out;
1522
1523         retval = prepare_binprm(bprm);
1524         if (retval < 0)
1525                 goto out;
1526
1527         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1528         if (retval < 0)
1529                 goto out;
1530
1531         bprm->exec = bprm->p;
1532         retval = copy_strings(bprm->envc, envp, bprm);
1533         if (retval < 0)
1534                 goto out;
1535
1536         retval = copy_strings(bprm->argc, argv, bprm);
1537         if (retval < 0)
1538                 goto out;
1539
1540         retval = search_binary_handler(bprm,regs);
1541         if (retval < 0)
1542                 goto out;
1543
1544         /* execve succeeded */
1545         current->fs->in_exec = 0;
1546         current->in_execve = 0;
1547         acct_update_integrals(current);
1548         free_bprm(bprm);
1549         if (displaced)
1550                 put_files_struct(displaced);
1551         return retval;
1552
1553 out:
1554         if (bprm->mm) {
1555                 acct_arg_size(bprm, 0);
1556                 mmput(bprm->mm);
1557         }
1558
1559 out_file:
1560         if (bprm->file) {
1561                 allow_write_access(bprm->file);
1562                 fput(bprm->file);
1563         }
1564
1565 out_unmark:
1566         if (clear_in_exec)
1567                 current->fs->in_exec = 0;
1568         current->in_execve = 0;
1569
1570 out_free:
1571         free_bprm(bprm);
1572
1573 out_files:
1574         if (displaced)
1575                 reset_files_struct(displaced);
1576 out_ret:
1577         return retval;
1578 }
1579
1580 int do_execve(const char *filename,
1581         const char __user *const __user *__argv,
1582         const char __user *const __user *__envp,
1583         struct pt_regs *regs)
1584 {
1585         struct user_arg_ptr argv = { .ptr.native = __argv };
1586         struct user_arg_ptr envp = { .ptr.native = __envp };
1587         return do_execve_common(filename, argv, envp, regs);
1588 }
1589
1590 #ifdef CONFIG_COMPAT
1591 int compat_do_execve(char *filename,
1592         compat_uptr_t __user *__argv,
1593         compat_uptr_t __user *__envp,
1594         struct pt_regs *regs)
1595 {
1596         struct user_arg_ptr argv = {
1597                 .is_compat = true,
1598                 .ptr.compat = __argv,
1599         };
1600         struct user_arg_ptr envp = {
1601                 .is_compat = true,
1602                 .ptr.compat = __envp,
1603         };
1604         return do_execve_common(filename, argv, envp, regs);
1605 }
1606 #endif
1607
1608 void set_binfmt(struct linux_binfmt *new)
1609 {
1610         struct mm_struct *mm = current->mm;
1611
1612         if (mm->binfmt)
1613                 module_put(mm->binfmt->module);
1614
1615         mm->binfmt = new;
1616         if (new)
1617                 __module_get(new->module);
1618 }
1619
1620 EXPORT_SYMBOL(set_binfmt);
1621
1622 static int expand_corename(struct core_name *cn)
1623 {
1624         char *old_corename = cn->corename;
1625
1626         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1627         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1628
1629         if (!cn->corename) {
1630                 kfree(old_corename);
1631                 return -ENOMEM;
1632         }
1633
1634         return 0;
1635 }
1636
1637 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1638 {
1639         char *cur;
1640         int need;
1641         int ret;
1642         va_list arg;
1643
1644         va_start(arg, fmt);
1645         need = vsnprintf(NULL, 0, fmt, arg);
1646         va_end(arg);
1647
1648         if (likely(need < cn->size - cn->used - 1))
1649                 goto out_printf;
1650
1651         ret = expand_corename(cn);
1652         if (ret)
1653                 goto expand_fail;
1654
1655 out_printf:
1656         cur = cn->corename + cn->used;
1657         va_start(arg, fmt);
1658         vsnprintf(cur, need + 1, fmt, arg);
1659         va_end(arg);
1660         cn->used += need;
1661         return 0;
1662
1663 expand_fail:
1664         return ret;
1665 }
1666
1667 static void cn_escape(char *str)
1668 {
1669         for (; *str; str++)
1670                 if (*str == '/')
1671                         *str = '!';
1672 }
1673
1674 static int cn_print_exe_file(struct core_name *cn)
1675 {
1676         struct file *exe_file;
1677         char *pathbuf, *path;
1678         int ret;
1679
1680         exe_file = get_mm_exe_file(current->mm);
1681         if (!exe_file) {
1682                 char *commstart = cn->corename + cn->used;
1683                 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1684                 cn_escape(commstart);
1685                 return ret;
1686         }
1687
1688         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1689         if (!pathbuf) {
1690                 ret = -ENOMEM;
1691                 goto put_exe_file;
1692         }
1693
1694         path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1695         if (IS_ERR(path)) {
1696                 ret = PTR_ERR(path);
1697                 goto free_buf;
1698         }
1699
1700         cn_escape(path);
1701
1702         ret = cn_printf(cn, "%s", path);
1703
1704 free_buf:
1705         kfree(pathbuf);
1706 put_exe_file:
1707         fput(exe_file);
1708         return ret;
1709 }
1710
1711 /* format_corename will inspect the pattern parameter, and output a
1712  * name into corename, which must have space for at least
1713  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1714  */
1715 static int format_corename(struct core_name *cn, long signr)
1716 {
1717         const struct cred *cred = current_cred();
1718         const char *pat_ptr = core_pattern;
1719         int ispipe = (*pat_ptr == '|');
1720         int pid_in_pattern = 0;
1721         int err = 0;
1722
1723         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1724         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1725         cn->used = 0;
1726
1727         if (!cn->corename)
1728                 return -ENOMEM;
1729
1730         /* Repeat as long as we have more pattern to process and more output
1731            space */
1732         while (*pat_ptr) {
1733                 if (*pat_ptr != '%') {
1734                         if (*pat_ptr == 0)
1735                                 goto out;
1736                         err = cn_printf(cn, "%c", *pat_ptr++);
1737                 } else {
1738                         switch (*++pat_ptr) {
1739                         /* single % at the end, drop that */
1740                         case 0:
1741                                 goto out;
1742                         /* Double percent, output one percent */
1743                         case '%':
1744                                 err = cn_printf(cn, "%c", '%');
1745                                 break;
1746                         /* pid */
1747                         case 'p':
1748                                 pid_in_pattern = 1;
1749                                 err = cn_printf(cn, "%d",
1750                                               task_tgid_vnr(current));
1751                                 break;
1752                         /* uid */
1753                         case 'u':
1754                                 err = cn_printf(cn, "%d", cred->uid);
1755                                 break;
1756                         /* gid */
1757                         case 'g':
1758                                 err = cn_printf(cn, "%d", cred->gid);
1759                                 break;
1760                         /* signal that caused the coredump */
1761                         case 's':
1762                                 err = cn_printf(cn, "%ld", signr);
1763                                 break;
1764                         /* UNIX time of coredump */
1765                         case 't': {
1766                                 struct timeval tv;
1767                                 do_gettimeofday(&tv);
1768                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1769                                 break;
1770                         }
1771                         /* hostname */
1772                         case 'h': {
1773                                 char *namestart = cn->corename + cn->used;
1774                                 down_read(&uts_sem);
1775                                 err = cn_printf(cn, "%s",
1776                                               utsname()->nodename);
1777                                 up_read(&uts_sem);
1778                                 cn_escape(namestart);
1779                                 break;
1780                         }
1781                         /* executable */
1782                         case 'e': {
1783                                 char *commstart = cn->corename + cn->used;
1784                                 err = cn_printf(cn, "%s", current->comm);
1785                                 cn_escape(commstart);
1786                                 break;
1787                         }
1788                         case 'E':
1789                                 err = cn_print_exe_file(cn);
1790                                 break;
1791                         /* core limit size */
1792                         case 'c':
1793                                 err = cn_printf(cn, "%lu",
1794                                               rlimit(RLIMIT_CORE));
1795                                 break;
1796                         default:
1797                                 break;
1798                         }
1799                         ++pat_ptr;
1800                 }
1801
1802                 if (err)
1803                         return err;
1804         }
1805
1806         /* Backward compatibility with core_uses_pid:
1807          *
1808          * If core_pattern does not include a %p (as is the default)
1809          * and core_uses_pid is set, then .%pid will be appended to
1810          * the filename. Do not do this for piped commands. */
1811         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1812                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1813                 if (err)
1814                         return err;
1815         }
1816 out:
1817         return ispipe;
1818 }
1819
1820 static int zap_process(struct task_struct *start, int exit_code)
1821 {
1822         struct task_struct *t;
1823         int nr = 0;
1824
1825         start->signal->flags = SIGNAL_GROUP_EXIT;
1826         start->signal->group_exit_code = exit_code;
1827         start->signal->group_stop_count = 0;
1828
1829         t = start;
1830         do {
1831                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1832                 if (t != current && t->mm) {
1833                         sigaddset(&t->pending.signal, SIGKILL);
1834                         signal_wake_up(t, 1);
1835                         nr++;
1836                 }
1837         } while_each_thread(start, t);
1838
1839         return nr;
1840 }
1841
1842 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1843                                 struct core_state *core_state, int exit_code)
1844 {
1845         struct task_struct *g, *p;
1846         unsigned long flags;
1847         int nr = -EAGAIN;
1848
1849         spin_lock_irq(&tsk->sighand->siglock);
1850         if (!signal_group_exit(tsk->signal)) {
1851                 mm->core_state = core_state;
1852                 nr = zap_process(tsk, exit_code);
1853         }
1854         spin_unlock_irq(&tsk->sighand->siglock);
1855         if (unlikely(nr < 0))
1856                 return nr;
1857
1858         if (atomic_read(&mm->mm_users) == nr + 1)
1859                 goto done;
1860         /*
1861          * We should find and kill all tasks which use this mm, and we should
1862          * count them correctly into ->nr_threads. We don't take tasklist
1863          * lock, but this is safe wrt:
1864          *
1865          * fork:
1866          *      None of sub-threads can fork after zap_process(leader). All
1867          *      processes which were created before this point should be
1868          *      visible to zap_threads() because copy_process() adds the new
1869          *      process to the tail of init_task.tasks list, and lock/unlock
1870          *      of ->siglock provides a memory barrier.
1871          *
1872          * do_exit:
1873          *      The caller holds mm->mmap_sem. This means that the task which
1874          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1875          *      its ->mm.
1876          *
1877          * de_thread:
1878          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1879          *      we must see either old or new leader, this does not matter.
1880          *      However, it can change p->sighand, so lock_task_sighand(p)
1881          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1882          *      it can't fail.
1883          *
1884          *      Note also that "g" can be the old leader with ->mm == NULL
1885          *      and already unhashed and thus removed from ->thread_group.
1886          *      This is OK, __unhash_process()->list_del_rcu() does not
1887          *      clear the ->next pointer, we will find the new leader via
1888          *      next_thread().
1889          */
1890         rcu_read_lock();
1891         for_each_process(g) {
1892                 if (g == tsk->group_leader)
1893                         continue;
1894                 if (g->flags & PF_KTHREAD)
1895                         continue;
1896                 p = g;
1897                 do {
1898                         if (p->mm) {
1899                                 if (unlikely(p->mm == mm)) {
1900                                         lock_task_sighand(p, &flags);
1901                                         nr += zap_process(p, exit_code);
1902                                         unlock_task_sighand(p, &flags);
1903                                 }
1904                                 break;
1905                         }
1906                 } while_each_thread(g, p);
1907         }
1908         rcu_read_unlock();
1909 done:
1910         atomic_set(&core_state->nr_threads, nr);
1911         return nr;
1912 }
1913
1914 static int coredump_wait(int exit_code, struct core_state *core_state)
1915 {
1916         struct task_struct *tsk = current;
1917         struct mm_struct *mm = tsk->mm;
1918         struct completion *vfork_done;
1919         int core_waiters = -EBUSY;
1920
1921         init_completion(&core_state->startup);
1922         core_state->dumper.task = tsk;
1923         core_state->dumper.next = NULL;
1924
1925         down_write(&mm->mmap_sem);
1926         if (!mm->core_state)
1927                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1928         up_write(&mm->mmap_sem);
1929
1930         if (unlikely(core_waiters < 0))
1931                 goto fail;
1932
1933         /*
1934          * Make sure nobody is waiting for us to release the VM,
1935          * otherwise we can deadlock when we wait on each other
1936          */
1937         vfork_done = tsk->vfork_done;
1938         if (vfork_done) {
1939                 tsk->vfork_done = NULL;
1940                 complete(vfork_done);
1941         }
1942
1943         if (core_waiters)
1944                 wait_for_completion(&core_state->startup);
1945 fail:
1946         return core_waiters;
1947 }
1948
1949 static void coredump_finish(struct mm_struct *mm)
1950 {
1951         struct core_thread *curr, *next;
1952         struct task_struct *task;
1953
1954         next = mm->core_state->dumper.next;
1955         while ((curr = next) != NULL) {
1956                 next = curr->next;
1957                 task = curr->task;
1958                 /*
1959                  * see exit_mm(), curr->task must not see
1960                  * ->task == NULL before we read ->next.
1961                  */
1962                 smp_mb();
1963                 curr->task = NULL;
1964                 wake_up_process(task);
1965         }
1966
1967         mm->core_state = NULL;
1968 }
1969
1970 /*
1971  * set_dumpable converts traditional three-value dumpable to two flags and
1972  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1973  * these bits are not changed atomically.  So get_dumpable can observe the
1974  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1975  * return either old dumpable or new one by paying attention to the order of
1976  * modifying the bits.
1977  *
1978  * dumpable |   mm->flags (binary)
1979  * old  new | initial interim  final
1980  * ---------+-----------------------
1981  *  0    1  |   00      01      01
1982  *  0    2  |   00      10(*)   11
1983  *  1    0  |   01      00      00
1984  *  1    2  |   01      11      11
1985  *  2    0  |   11      10(*)   00
1986  *  2    1  |   11      11      01
1987  *
1988  * (*) get_dumpable regards interim value of 10 as 11.
1989  */
1990 void set_dumpable(struct mm_struct *mm, int value)
1991 {
1992         switch (value) {
1993         case 0:
1994                 clear_bit(MMF_DUMPABLE, &mm->flags);
1995                 smp_wmb();
1996                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1997                 break;
1998         case 1:
1999                 set_bit(MMF_DUMPABLE, &mm->flags);
2000                 smp_wmb();
2001                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2002                 break;
2003         case 2:
2004                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2005                 smp_wmb();
2006                 set_bit(MMF_DUMPABLE, &mm->flags);
2007                 break;
2008         }
2009 }
2010
2011 static int __get_dumpable(unsigned long mm_flags)
2012 {
2013         int ret;
2014
2015         ret = mm_flags & MMF_DUMPABLE_MASK;
2016         return (ret >= 2) ? 2 : ret;
2017 }
2018
2019 int get_dumpable(struct mm_struct *mm)
2020 {
2021         return __get_dumpable(mm->flags);
2022 }
2023
2024 static void wait_for_dump_helpers(struct file *file)
2025 {
2026         struct pipe_inode_info *pipe;
2027
2028         pipe = file->f_path.dentry->d_inode->i_pipe;
2029
2030         pipe_lock(pipe);
2031         pipe->readers++;
2032         pipe->writers--;
2033
2034         while ((pipe->readers > 1) && (!signal_pending(current))) {
2035                 wake_up_interruptible_sync(&pipe->wait);
2036                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2037                 pipe_wait(pipe);
2038         }
2039
2040         pipe->readers--;
2041         pipe->writers++;
2042         pipe_unlock(pipe);
2043
2044 }
2045
2046
2047 /*
2048  * umh_pipe_setup
2049  * helper function to customize the process used
2050  * to collect the core in userspace.  Specifically
2051  * it sets up a pipe and installs it as fd 0 (stdin)
2052  * for the process.  Returns 0 on success, or
2053  * PTR_ERR on failure.
2054  * Note that it also sets the core limit to 1.  This
2055  * is a special value that we use to trap recursive
2056  * core dumps
2057  */
2058 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2059 {
2060         struct file *rp, *wp;
2061         struct fdtable *fdt;
2062         struct coredump_params *cp = (struct coredump_params *)info->data;
2063         struct files_struct *cf = current->files;
2064
2065         wp = create_write_pipe(0);
2066         if (IS_ERR(wp))
2067                 return PTR_ERR(wp);
2068
2069         rp = create_read_pipe(wp, 0);
2070         if (IS_ERR(rp)) {
2071                 free_write_pipe(wp);
2072                 return PTR_ERR(rp);
2073         }
2074
2075         cp->file = wp;
2076
2077         sys_close(0);
2078         fd_install(0, rp);
2079         spin_lock(&cf->file_lock);
2080         fdt = files_fdtable(cf);
2081         FD_SET(0, fdt->open_fds);
2082         FD_CLR(0, fdt->close_on_exec);
2083         spin_unlock(&cf->file_lock);
2084
2085         /* and disallow core files too */
2086         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2087
2088         return 0;
2089 }
2090
2091 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2092 {
2093         struct core_state core_state;
2094         struct core_name cn;
2095         struct mm_struct *mm = current->mm;
2096         struct linux_binfmt * binfmt;
2097         const struct cred *old_cred;
2098         struct cred *cred;
2099         int retval = 0;
2100         int flag = 0;
2101         int ispipe;
2102         static atomic_t core_dump_count = ATOMIC_INIT(0);
2103         struct coredump_params cprm = {
2104                 .signr = signr,
2105                 .regs = regs,
2106                 .limit = rlimit(RLIMIT_CORE),
2107                 /*
2108                  * We must use the same mm->flags while dumping core to avoid
2109                  * inconsistency of bit flags, since this flag is not protected
2110                  * by any locks.
2111                  */
2112                 .mm_flags = mm->flags,
2113         };
2114
2115         audit_core_dumps(signr);
2116
2117         binfmt = mm->binfmt;
2118         if (!binfmt || !binfmt->core_dump)
2119                 goto fail;
2120         if (!__get_dumpable(cprm.mm_flags))
2121                 goto fail;
2122
2123         cred = prepare_creds();
2124         if (!cred)
2125                 goto fail;
2126         /*
2127          *      We cannot trust fsuid as being the "true" uid of the
2128          *      process nor do we know its entire history. We only know it
2129          *      was tainted so we dump it as root in mode 2.
2130          */
2131         if (__get_dumpable(cprm.mm_flags) == 2) {
2132                 /* Setuid core dump mode */
2133                 flag = O_EXCL;          /* Stop rewrite attacks */
2134                 cred->fsuid = 0;        /* Dump root private */
2135         }
2136
2137         retval = coredump_wait(exit_code, &core_state);
2138         if (retval < 0)
2139                 goto fail_creds;
2140
2141         old_cred = override_creds(cred);
2142
2143         /*
2144          * Clear any false indication of pending signals that might
2145          * be seen by the filesystem code called to write the core file.
2146          */
2147         clear_thread_flag(TIF_SIGPENDING);
2148
2149         ispipe = format_corename(&cn, signr);
2150
2151         if (ispipe) {
2152                 int dump_count;
2153                 char **helper_argv;
2154
2155                 if (ispipe < 0) {
2156                         printk(KERN_WARNING "format_corename failed\n");
2157                         printk(KERN_WARNING "Aborting core\n");
2158                         goto fail_corename;
2159                 }
2160
2161                 if (cprm.limit == 1) {
2162                         /*
2163                          * Normally core limits are irrelevant to pipes, since
2164                          * we're not writing to the file system, but we use
2165                          * cprm.limit of 1 here as a speacial value. Any
2166                          * non-1 limit gets set to RLIM_INFINITY below, but
2167                          * a limit of 0 skips the dump.  This is a consistent
2168                          * way to catch recursive crashes.  We can still crash
2169                          * if the core_pattern binary sets RLIM_CORE =  !1
2170                          * but it runs as root, and can do lots of stupid things
2171                          * Note that we use task_tgid_vnr here to grab the pid
2172                          * of the process group leader.  That way we get the
2173                          * right pid if a thread in a multi-threaded
2174                          * core_pattern process dies.
2175                          */
2176                         printk(KERN_WARNING
2177                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2178                                 task_tgid_vnr(current), current->comm);
2179                         printk(KERN_WARNING "Aborting core\n");
2180                         goto fail_unlock;
2181                 }
2182                 cprm.limit = RLIM_INFINITY;
2183
2184                 dump_count = atomic_inc_return(&core_dump_count);
2185                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2186                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2187                                task_tgid_vnr(current), current->comm);
2188                         printk(KERN_WARNING "Skipping core dump\n");
2189                         goto fail_dropcount;
2190                 }
2191
2192                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2193                 if (!helper_argv) {
2194                         printk(KERN_WARNING "%s failed to allocate memory\n",
2195                                __func__);
2196                         goto fail_dropcount;
2197                 }
2198
2199                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2200                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2201                                         NULL, &cprm);
2202                 argv_free(helper_argv);
2203                 if (retval) {
2204                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2205                                cn.corename);
2206                         goto close_fail;
2207                 }
2208         } else {
2209                 struct inode *inode;
2210
2211                 if (cprm.limit < binfmt->min_coredump)
2212                         goto fail_unlock;
2213
2214                 cprm.file = filp_open(cn.corename,
2215                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2216                                  0600);
2217                 if (IS_ERR(cprm.file))
2218                         goto fail_unlock;
2219
2220                 inode = cprm.file->f_path.dentry->d_inode;
2221                 if (inode->i_nlink > 1)
2222                         goto close_fail;
2223                 if (d_unhashed(cprm.file->f_path.dentry))
2224                         goto close_fail;
2225                 /*
2226                  * AK: actually i see no reason to not allow this for named
2227                  * pipes etc, but keep the previous behaviour for now.
2228                  */
2229                 if (!S_ISREG(inode->i_mode))
2230                         goto close_fail;
2231                 /*
2232                  * Dont allow local users get cute and trick others to coredump
2233                  * into their pre-created files.
2234                  */
2235                 if (inode->i_uid != current_fsuid())
2236                         goto close_fail;
2237                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2238                         goto close_fail;
2239                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2240                         goto close_fail;
2241         }
2242
2243         retval = binfmt->core_dump(&cprm);
2244         if (retval)
2245                 current->signal->group_exit_code |= 0x80;
2246
2247         if (ispipe && core_pipe_limit)
2248                 wait_for_dump_helpers(cprm.file);
2249 close_fail:
2250         if (cprm.file)
2251                 filp_close(cprm.file, NULL);
2252 fail_dropcount:
2253         if (ispipe)
2254                 atomic_dec(&core_dump_count);
2255 fail_unlock:
2256         kfree(cn.corename);
2257 fail_corename:
2258         coredump_finish(mm);
2259         revert_creds(old_cred);
2260 fail_creds:
2261         put_cred(cred);
2262 fail:
2263         return;
2264 }
2265
2266 /*
2267  * Core dumping helper functions.  These are the only things you should
2268  * do on a core-file: use only these functions to write out all the
2269  * necessary info.
2270  */
2271 int dump_write(struct file *file, const void *addr, int nr)
2272 {
2273         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2274 }
2275 EXPORT_SYMBOL(dump_write);
2276
2277 int dump_seek(struct file *file, loff_t off)
2278 {
2279         int ret = 1;
2280
2281         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2282                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2283                         return 0;
2284         } else {
2285                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2286
2287                 if (!buf)
2288                         return 0;
2289                 while (off > 0) {
2290                         unsigned long n = off;
2291
2292                         if (n > PAGE_SIZE)
2293                                 n = PAGE_SIZE;
2294                         if (!dump_write(file, buf, n)) {
2295                                 ret = 0;
2296                                 break;
2297                         }
2298                         off -= n;
2299                 }
2300                 free_page((unsigned long)buf);
2301         }
2302         return ret;
2303 }
2304 EXPORT_SYMBOL(dump_seek);