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