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         /* we have changed execution domain */
 981         tsk->exit_signal = SIGCHLD;
 982
 983         if (current->mm)
 984                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
 985
 986         exit_itimers(sig);
 987         flush_itimer_signals();
 988
 989         if (atomic_read(&oldsighand->count) != 1) {
 990                 struct sighand_struct *newsighand;
 991                 /*
 992                  * This ->sighand is shared with the CLONE_SIGHAND
 993                  * but not CLONE_THREAD task, switch to the new one.
 994                  */
 995                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 996                 if (!newsighand)
 997                         return -ENOMEM;
 998
 999                 atomic_set(&newsighand->count, 1);
1000                 memcpy(newsighand->action, oldsighand->action,
1001                        sizeof(newsighand->action));
1002
1003                 write_lock_irq(&tasklist_lock);
1004                 spin_lock(&oldsighand->siglock);
1005                 rcu_assign_pointer(tsk->sighand, newsighand);
1006                 spin_unlock(&oldsighand->siglock);
1007                 write_unlock_irq(&tasklist_lock);
1008
1009                 __cleanup_sighand(oldsighand);
1010         }
1011
1012         BUG_ON(!thread_group_leader(tsk));
1013         return 0;
1014 }
1015
1016 /*
1017  * These functions flushes out all traces of the currently running executable
1018  * so that a new one can be started
1019  */
1020 static void flush_old_files(struct files_struct * files)
1021 {
1022         long j = -1;
1023         struct fdtable *fdt;
1024
1025         spin_lock(&files->file_lock);
1026         for (;;) {
1027                 unsigned long set, i;
1028
1029                 j++;
1030                 i = j * __NFDBITS;
1031                 fdt = files_fdtable(files);
1032                 if (i >= fdt->max_fds)
1033                         break;
1034                 set = fdt->close_on_exec->fds_bits[j];
1035                 if (!set)
1036                         continue;
1037                 fdt->close_on_exec->fds_bits[j] = 0;
1038                 spin_unlock(&files->file_lock);
1039                 for ( ; set ; i++,set >>= 1) {
1040                         if (set & 1) {
1041                                 sys_close(i);
1042                         }
1043                 }
1044                 spin_lock(&files->file_lock);
1045
1046         }
1047         spin_unlock(&files->file_lock);
1048 }
1049
1050 char *get_task_comm(char *buf, struct task_struct *tsk)
1051 {
1052         /* buf must be at least sizeof(tsk->comm) in size */
1053         task_lock(tsk);
1054         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1055         task_unlock(tsk);
1056         return buf;
1057 }
1058 EXPORT_SYMBOL_GPL(get_task_comm);
1059
1060 void set_task_comm(struct task_struct *tsk, char *buf)
1061 {
1062         task_lock(tsk);
1063
1064         trace_task_rename(tsk, buf);
1065
1066         /*
1067          * Threads may access current->comm without holding
1068          * the task lock, so write the string carefully.
1069          * Readers without a lock may see incomplete new
1070          * names but are safe from non-terminating string reads.
1071          */
1072         memset(tsk->comm, 0, TASK_COMM_LEN);
1073         wmb();
1074         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1075         task_unlock(tsk);
1076         perf_event_comm(tsk);
1077 }
1078
1079 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1080 {
1081         int i, ch;
1082
1083         /* Copies the binary name from after last slash */
1084         for (i = 0; (ch = *(fn++)) != '\0';) {
1085                 if (ch == '/')
1086                         i = 0; /* overwrite what we wrote */
1087                 else
1088                         if (i < len - 1)
1089                                 tcomm[i++] = ch;
1090         }
1091         tcomm[i] = '\0';
1092 }
1093
1094 int flush_old_exec(struct linux_binprm * bprm)
1095 {
1096         int retval;
1097
1098         /*
1099          * Make sure we have a private signal table and that
1100          * we are unassociated from the previous thread group.
1101          */
1102         retval = de_thread(current);
1103         if (retval)
1104                 goto out;
1105
1106         set_mm_exe_file(bprm->mm, bprm->file);
1107
1108         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1109         /*
1110          * Release all of the old mmap stuff
1111          */
1112         acct_arg_size(bprm, 0);
1113         retval = exec_mmap(bprm->mm);
1114         if (retval)
1115                 goto out;
1116
1117         bprm->mm = NULL;                /* We're using it now */
1118
1119         set_fs(USER_DS);
1120         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1121         flush_thread();
1122         current->personality &= ~bprm->per_clear;
1123
1124         return 0;
1125
1126 out:
1127         return retval;
1128 }
1129 EXPORT_SYMBOL(flush_old_exec);
1130
1131 void would_dump(struct linux_binprm *bprm, struct file *file)
1132 {
1133         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1134                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1135 }
1136 EXPORT_SYMBOL(would_dump);
1137
1138 void setup_new_exec(struct linux_binprm * bprm)
1139 {
1140         arch_pick_mmap_layout(current->mm);
1141
1142         /* This is the point of no return */
1143         current->sas_ss_sp = current->sas_ss_size = 0;
1144
1145         if (current_euid() == current_uid() && current_egid() == current_gid())
1146                 set_dumpable(current->mm, 1);
1147         else
1148                 set_dumpable(current->mm, suid_dumpable);
1149
1150         set_task_comm(current, bprm->tcomm);
1151
1152         /* Set the new mm task size. We have to do that late because it may
1153          * depend on TIF_32BIT which is only updated in flush_thread() on
1154          * some architectures like powerpc
1155          */
1156         current->mm->task_size = TASK_SIZE;
1157
1158         /* install the new credentials */
1159         if (bprm->cred->uid != current_euid() ||
1160             bprm->cred->gid != current_egid()) {
1161                 current->pdeath_signal = 0;
1162         } else {
1163                 would_dump(bprm, bprm->file);
1164                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1165                         set_dumpable(current->mm, suid_dumpable);
1166         }
1167
1168         /*
1169          * Flush performance counters when crossing a
1170          * security domain:
1171          */
1172         if (!get_dumpable(current->mm))
1173                 perf_event_exit_task(current);
1174
1175         /* An exec changes our domain. We are no longer part of the thread
1176            group */
1177
1178         current->self_exec_id++;
1179
1180         flush_signal_handlers(current, 0);
1181         flush_old_files(current->files);
1182 }
1183 EXPORT_SYMBOL(setup_new_exec);
1184
1185 /*
1186  * Prepare credentials and lock ->cred_guard_mutex.
1187  * install_exec_creds() commits the new creds and drops the lock.
1188  * Or, if exec fails before, free_bprm() should release ->cred and
1189  * and unlock.
1190  */
1191 int prepare_bprm_creds(struct linux_binprm *bprm)
1192 {
1193         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1194                 return -ERESTARTNOINTR;
1195
1196         bprm->cred = prepare_exec_creds();
1197         if (likely(bprm->cred))
1198                 return 0;
1199
1200         mutex_unlock(&current->signal->cred_guard_mutex);
1201         return -ENOMEM;
1202 }
1203
1204 void free_bprm(struct linux_binprm *bprm)
1205 {
1206         free_arg_pages(bprm);
1207         if (bprm->cred) {
1208                 mutex_unlock(&current->signal->cred_guard_mutex);
1209                 abort_creds(bprm->cred);
1210         }
1211         kfree(bprm);
1212 }
1213
1214 /*
1215  * install the new credentials for this executable
1216  */
1217 void install_exec_creds(struct linux_binprm *bprm)
1218 {
1219         security_bprm_committing_creds(bprm);
1220
1221         commit_creds(bprm->cred);
1222         bprm->cred = NULL;
1223         /*
1224          * cred_guard_mutex must be held at least to this point to prevent
1225          * ptrace_attach() from altering our determination of the task's
1226          * credentials; any time after this it may be unlocked.
1227          */
1228         security_bprm_committed_creds(bprm);
1229         mutex_unlock(&current->signal->cred_guard_mutex);
1230 }
1231 EXPORT_SYMBOL(install_exec_creds);
1232
1233 /*
1234  * determine how safe it is to execute the proposed program
1235  * - the caller must hold ->cred_guard_mutex to protect against
1236  *   PTRACE_ATTACH
1237  */
1238 static int check_unsafe_exec(struct linux_binprm *bprm)
1239 {
1240         struct task_struct *p = current, *t;
1241         unsigned n_fs;
1242         int res = 0;
1243
1244         if (p->ptrace) {
1245                 if (p->ptrace & PT_PTRACE_CAP)
1246                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1247                 else
1248                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1249         }
1250
1251         n_fs = 1;
1252         spin_lock(&p->fs->lock);
1253         rcu_read_lock();
1254         for (t = next_thread(p); t != p; t = next_thread(t)) {
1255                 if (t->fs == p->fs)
1256                         n_fs++;
1257         }
1258         rcu_read_unlock();
1259
1260         if (p->fs->users > n_fs) {
1261                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1262         } else {
1263                 res = -EAGAIN;
1264                 if (!p->fs->in_exec) {
1265                         p->fs->in_exec = 1;
1266                         res = 1;
1267                 }
1268         }
1269         spin_unlock(&p->fs->lock);
1270
1271         return res;
1272 }
1273
1274 /*
1275  * Fill the binprm structure from the inode.
1276  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1277  *
1278  * This may be called multiple times for binary chains (scripts for example).
1279  */
1280 int prepare_binprm(struct linux_binprm *bprm)
1281 {
1282         umode_t mode;
1283         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1284         int retval;
1285
1286         mode = inode->i_mode;
1287         if (bprm->file->f_op == NULL)
1288                 return -EACCES;
1289
1290         /* clear any previous set[ug]id data from a previous binary */
1291         bprm->cred->euid = current_euid();
1292         bprm->cred->egid = current_egid();
1293
1294         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1295                 /* Set-uid? */
1296                 if (mode & S_ISUID) {
1297                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1298                         bprm->cred->euid = inode->i_uid;
1299                 }
1300
1301                 /* Set-gid? */
1302                 /*
1303                  * If setgid is set but no group execute bit then this
1304                  * is a candidate for mandatory locking, not a setgid
1305                  * executable.
1306                  */
1307                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1308                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1309                         bprm->cred->egid = inode->i_gid;
1310                 }
1311         }
1312
1313         /* fill in binprm security blob */
1314         retval = security_bprm_set_creds(bprm);
1315         if (retval)
1316                 return retval;
1317         bprm->cred_prepared = 1;
1318
1319         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1320         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1321 }
1322
1323 EXPORT_SYMBOL(prepare_binprm);
1324
1325 /*
1326  * Arguments are '\0' separated strings found at the location bprm->p
1327  * points to; chop off the first by relocating brpm->p to right after
1328  * the first '\0' encountered.
1329  */
1330 int remove_arg_zero(struct linux_binprm *bprm)
1331 {
1332         int ret = 0;
1333         unsigned long offset;
1334         char *kaddr;
1335         struct page *page;
1336
1337         if (!bprm->argc)
1338                 return 0;
1339
1340         do {
1341                 offset = bprm->p & ~PAGE_MASK;
1342                 page = get_arg_page(bprm, bprm->p, 0);
1343                 if (!page) {
1344                         ret = -EFAULT;
1345                         goto out;
1346                 }
1347                 kaddr = kmap_atomic(page, KM_USER0);
1348
1349                 for (; offset < PAGE_SIZE && kaddr[offset];
1350                                 offset++, bprm->p++)
1351                         ;
1352
1353                 kunmap_atomic(kaddr, KM_USER0);
1354                 put_arg_page(page);
1355
1356                 if (offset == PAGE_SIZE)
1357                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1358         } while (offset == PAGE_SIZE);
1359
1360         bprm->p++;
1361         bprm->argc--;
1362         ret = 0;
1363
1364 out:
1365         return ret;
1366 }
1367 EXPORT_SYMBOL(remove_arg_zero);
1368
1369 /*
1370  * cycle the list of binary formats handler, until one recognizes the image
1371  */
1372 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1373 {
1374         unsigned int depth = bprm->recursion_depth;
1375         int try,retval;
1376         struct linux_binfmt *fmt;
1377         pid_t old_pid;
1378
1379         retval = security_bprm_check(bprm);
1380         if (retval)
1381                 return retval;
1382
1383         retval = audit_bprm(bprm);
1384         if (retval)
1385                 return retval;
1386
1387         /* Need to fetch pid before load_binary changes it */
1388         rcu_read_lock();
1389         old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1390         rcu_read_unlock();
1391
1392         retval = -ENOENT;
1393         for (try=0; try<2; try++) {
1394                 read_lock(&binfmt_lock);
1395                 list_for_each_entry(fmt, &formats, lh) {
1396                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1397                         if (!fn)
1398                                 continue;
1399                         if (!try_module_get(fmt->module))
1400                                 continue;
1401                         read_unlock(&binfmt_lock);
1402                         retval = fn(bprm, regs);
1403                         /*
1404                          * Restore the depth counter to its starting value
1405                          * in this call, so we don't have to rely on every
1406                          * load_binary function to restore it on return.
1407                          */
1408                         bprm->recursion_depth = depth;
1409                         if (retval >= 0) {
1410                                 if (depth == 0) {
1411                                         trace_sched_process_exec(current, old_pid, bprm);
1412                                         ptrace_event(PTRACE_EVENT_EXEC, old_pid);
1413                                 }
1414                                 put_binfmt(fmt);
1415                                 allow_write_access(bprm->file);
1416                                 if (bprm->file)
1417                                         fput(bprm->file);
1418                                 bprm->file = NULL;
1419                                 current->did_exec = 1;
1420                                 proc_exec_connector(current);
1421                                 return retval;
1422                         }
1423                         read_lock(&binfmt_lock);
1424                         put_binfmt(fmt);
1425                         if (retval != -ENOEXEC || bprm->mm == NULL)
1426                                 break;
1427                         if (!bprm->file) {
1428                                 read_unlock(&binfmt_lock);
1429                                 return retval;
1430                         }
1431                 }
1432                 read_unlock(&binfmt_lock);
1433 #ifdef CONFIG_MODULES
1434                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1435                         break;
1436                 } else {
1437 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1438                         if (printable(bprm->buf[0]) &&
1439                             printable(bprm->buf[1]) &&
1440                             printable(bprm->buf[2]) &&
1441                             printable(bprm->buf[3]))
1442                                 break; /* -ENOEXEC */
1443                         if (try)
1444                                 break; /* -ENOEXEC */
1445                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1446                 }
1447 #else
1448                 break;
1449 #endif
1450         }
1451         return retval;
1452 }
1453
1454 EXPORT_SYMBOL(search_binary_handler);
1455
1456 /*
1457  * sys_execve() executes a new program.
1458  */
1459 static int do_execve_common(const char *filename,
1460                                 struct user_arg_ptr argv,
1461                                 struct user_arg_ptr envp,
1462                                 struct pt_regs *regs)
1463 {
1464         struct linux_binprm *bprm;
1465         struct file *file;
1466         struct files_struct *displaced;
1467         bool clear_in_exec;
1468         int retval;
1469         const struct cred *cred = current_cred();
1470
1471         /*
1472          * We move the actual failure in case of RLIMIT_NPROC excess from
1473          * set*uid() to execve() because too many poorly written programs
1474          * don't check setuid() return code.  Here we additionally recheck
1475          * whether NPROC limit is still exceeded.
1476          */
1477         if ((current->flags & PF_NPROC_EXCEEDED) &&
1478             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1479                 retval = -EAGAIN;
1480                 goto out_ret;
1481         }
1482
1483         /* We're below the limit (still or again), so we don't want to make
1484          * further execve() calls fail. */
1485         current->flags &= ~PF_NPROC_EXCEEDED;
1486
1487         retval = unshare_files(&displaced);
1488         if (retval)
1489                 goto out_ret;
1490
1491         retval = -ENOMEM;
1492         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1493         if (!bprm)
1494                 goto out_files;
1495
1496         retval = prepare_bprm_creds(bprm);
1497         if (retval)
1498                 goto out_free;
1499
1500         retval = check_unsafe_exec(bprm);
1501         if (retval < 0)
1502                 goto out_free;
1503         clear_in_exec = retval;
1504         current->in_execve = 1;
1505
1506         file = open_exec(filename);
1507         retval = PTR_ERR(file);
1508         if (IS_ERR(file))
1509                 goto out_unmark;
1510
1511         sched_exec();
1512
1513         bprm->file = file;
1514         bprm->filename = filename;
1515         bprm->interp = filename;
1516
1517         retval = bprm_mm_init(bprm);
1518         if (retval)
1519                 goto out_file;
1520
1521         bprm->argc = count(argv, MAX_ARG_STRINGS);
1522         if ((retval = bprm->argc) < 0)
1523                 goto out;
1524
1525         bprm->envc = count(envp, MAX_ARG_STRINGS);
1526         if ((retval = bprm->envc) < 0)
1527                 goto out;
1528
1529         retval = prepare_binprm(bprm);
1530         if (retval < 0)
1531                 goto out;
1532
1533         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1534         if (retval < 0)
1535                 goto out;
1536
1537         bprm->exec = bprm->p;
1538         retval = copy_strings(bprm->envc, envp, bprm);
1539         if (retval < 0)
1540                 goto out;
1541
1542         retval = copy_strings(bprm->argc, argv, bprm);
1543         if (retval < 0)
1544                 goto out;
1545
1546         retval = search_binary_handler(bprm,regs);
1547         if (retval < 0)
1548                 goto out;
1549
1550         /* execve succeeded */
1551         current->fs->in_exec = 0;
1552         current->in_execve = 0;
1553         acct_update_integrals(current);
1554         free_bprm(bprm);
1555         if (displaced)
1556                 put_files_struct(displaced);
1557         return retval;
1558
1559 out:
1560         if (bprm->mm) {
1561                 acct_arg_size(bprm, 0);
1562                 mmput(bprm->mm);
1563         }
1564
1565 out_file:
1566         if (bprm->file) {
1567                 allow_write_access(bprm->file);
1568                 fput(bprm->file);
1569         }
1570
1571 out_unmark:
1572         if (clear_in_exec)
1573                 current->fs->in_exec = 0;
1574         current->in_execve = 0;
1575
1576 out_free:
1577         free_bprm(bprm);
1578
1579 out_files:
1580         if (displaced)
1581                 reset_files_struct(displaced);
1582 out_ret:
1583         return retval;
1584 }
1585
1586 int do_execve(const char *filename,
1587         const char __user *const __user *__argv,
1588         const char __user *const __user *__envp,
1589         struct pt_regs *regs)
1590 {
1591         struct user_arg_ptr argv = { .ptr.native = __argv };
1592         struct user_arg_ptr envp = { .ptr.native = __envp };
1593         return do_execve_common(filename, argv, envp, regs);
1594 }
1595
1596 #ifdef CONFIG_COMPAT
1597 int compat_do_execve(char *filename,
1598         compat_uptr_t __user *__argv,
1599         compat_uptr_t __user *__envp,
1600         struct pt_regs *regs)
1601 {
1602         struct user_arg_ptr argv = {
1603                 .is_compat = true,
1604                 .ptr.compat = __argv,
1605         };
1606         struct user_arg_ptr envp = {
1607                 .is_compat = true,
1608                 .ptr.compat = __envp,
1609         };
1610         return do_execve_common(filename, argv, envp, regs);
1611 }
1612 #endif
1613
1614 void set_binfmt(struct linux_binfmt *new)
1615 {
1616         struct mm_struct *mm = current->mm;
1617
1618         if (mm->binfmt)
1619                 module_put(mm->binfmt->module);
1620
1621         mm->binfmt = new;
1622         if (new)
1623                 __module_get(new->module);
1624 }
1625
1626 EXPORT_SYMBOL(set_binfmt);
1627
1628 static int expand_corename(struct core_name *cn)
1629 {
1630         char *old_corename = cn->corename;
1631
1632         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1633         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1634
1635         if (!cn->corename) {
1636                 kfree(old_corename);
1637                 return -ENOMEM;
1638         }
1639
1640         return 0;
1641 }
1642
1643 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1644 {
1645         char *cur;
1646         int need;
1647         int ret;
1648         va_list arg;
1649
1650         va_start(arg, fmt);
1651         need = vsnprintf(NULL, 0, fmt, arg);
1652         va_end(arg);
1653
1654         if (likely(need < cn->size - cn->used - 1))
1655                 goto out_printf;
1656
1657         ret = expand_corename(cn);
1658         if (ret)
1659                 goto expand_fail;
1660
1661 out_printf:
1662         cur = cn->corename + cn->used;
1663         va_start(arg, fmt);
1664         vsnprintf(cur, need + 1, fmt, arg);
1665         va_end(arg);
1666         cn->used += need;
1667         return 0;
1668
1669 expand_fail:
1670         return ret;
1671 }
1672
1673 static void cn_escape(char *str)
1674 {
1675         for (; *str; str++)
1676                 if (*str == '/')
1677                         *str = '!';
1678 }
1679
1680 static int cn_print_exe_file(struct core_name *cn)
1681 {
1682         struct file *exe_file;
1683         char *pathbuf, *path;
1684         int ret;
1685
1686         exe_file = get_mm_exe_file(current->mm);
1687         if (!exe_file) {
1688                 char *commstart = cn->corename + cn->used;
1689                 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1690                 cn_escape(commstart);
1691                 return ret;
1692         }
1693
1694         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1695         if (!pathbuf) {
1696                 ret = -ENOMEM;
1697                 goto put_exe_file;
1698         }
1699
1700         path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1701         if (IS_ERR(path)) {
1702                 ret = PTR_ERR(path);
1703                 goto free_buf;
1704         }
1705
1706         cn_escape(path);
1707
1708         ret = cn_printf(cn, "%s", path);
1709
1710 free_buf:
1711         kfree(pathbuf);
1712 put_exe_file:
1713         fput(exe_file);
1714         return ret;
1715 }
1716
1717 /* format_corename will inspect the pattern parameter, and output a
1718  * name into corename, which must have space for at least
1719  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1720  */
1721 static int format_corename(struct core_name *cn, long signr)
1722 {
1723         const struct cred *cred = current_cred();
1724         const char *pat_ptr = core_pattern;
1725         int ispipe = (*pat_ptr == '|');
1726         int pid_in_pattern = 0;
1727         int err = 0;
1728
1729         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1730         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1731         cn->used = 0;
1732
1733         if (!cn->corename)
1734                 return -ENOMEM;
1735
1736         /* Repeat as long as we have more pattern to process and more output
1737            space */
1738         while (*pat_ptr) {
1739                 if (*pat_ptr != '%') {
1740                         if (*pat_ptr == 0)
1741                                 goto out;
1742                         err = cn_printf(cn, "%c", *pat_ptr++);
1743                 } else {
1744                         switch (*++pat_ptr) {
1745                         /* single % at the end, drop that */
1746                         case 0:
1747                                 goto out;
1748                         /* Double percent, output one percent */
1749                         case '%':
1750                                 err = cn_printf(cn, "%c", '%');
1751                                 break;
1752                         /* pid */
1753                         case 'p':
1754                                 pid_in_pattern = 1;
1755                                 err = cn_printf(cn, "%d",
1756                                               task_tgid_vnr(current));
1757                                 break;
1758                         /* uid */
1759                         case 'u':
1760                                 err = cn_printf(cn, "%d", cred->uid);
1761                                 break;
1762                         /* gid */
1763                         case 'g':
1764                                 err = cn_printf(cn, "%d", cred->gid);
1765                                 break;
1766                         /* signal that caused the coredump */
1767                         case 's':
1768                                 err = cn_printf(cn, "%ld", signr);
1769                                 break;
1770                         /* UNIX time of coredump */
1771                         case 't': {
1772                                 struct timeval tv;
1773                                 do_gettimeofday(&tv);
1774                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1775                                 break;
1776                         }
1777                         /* hostname */
1778                         case 'h': {
1779                                 char *namestart = cn->corename + cn->used;
1780                                 down_read(&uts_sem);
1781                                 err = cn_printf(cn, "%s",
1782                                               utsname()->nodename);
1783                                 up_read(&uts_sem);
1784                                 cn_escape(namestart);
1785                                 break;
1786                         }
1787                         /* executable */
1788                         case 'e': {
1789                                 char *commstart = cn->corename + cn->used;
1790                                 err = cn_printf(cn, "%s", current->comm);
1791                                 cn_escape(commstart);
1792                                 break;
1793                         }
1794                         case 'E':
1795                                 err = cn_print_exe_file(cn);
1796                                 break;
1797                         /* core limit size */
1798                         case 'c':
1799                                 err = cn_printf(cn, "%lu",
1800                                               rlimit(RLIMIT_CORE));
1801                                 break;
1802                         default:
1803                                 break;
1804                         }
1805                         ++pat_ptr;
1806                 }
1807
1808                 if (err)
1809                         return err;
1810         }
1811
1812         /* Backward compatibility with core_uses_pid:
1813          *
1814          * If core_pattern does not include a %p (as is the default)
1815          * and core_uses_pid is set, then .%pid will be appended to
1816          * the filename. Do not do this for piped commands. */
1817         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1818                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1819                 if (err)
1820                         return err;
1821         }
1822 out:
1823         return ispipe;
1824 }
1825
1826 static int zap_process(struct task_struct *start, int exit_code)
1827 {
1828         struct task_struct *t;
1829         int nr = 0;
1830
1831         start->signal->flags = SIGNAL_GROUP_EXIT;
1832         start->signal->group_exit_code = exit_code;
1833         start->signal->group_stop_count = 0;
1834
1835         t = start;
1836         do {
1837                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1838                 if (t != current && t->mm) {
1839                         sigaddset(&t->pending.signal, SIGKILL);
1840                         signal_wake_up(t, 1);
1841                         nr++;
1842                 }
1843         } while_each_thread(start, t);
1844
1845         return nr;
1846 }
1847
1848 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1849                                 struct core_state *core_state, int exit_code)
1850 {
1851         struct task_struct *g, *p;
1852         unsigned long flags;
1853         int nr = -EAGAIN;
1854
1855         spin_lock_irq(&tsk->sighand->siglock);
1856         if (!signal_group_exit(tsk->signal)) {
1857                 mm->core_state = core_state;
1858                 nr = zap_process(tsk, exit_code);
1859         }
1860         spin_unlock_irq(&tsk->sighand->siglock);
1861         if (unlikely(nr < 0))
1862                 return nr;
1863
1864         if (atomic_read(&mm->mm_users) == nr + 1)
1865                 goto done;
1866         /*
1867          * We should find and kill all tasks which use this mm, and we should
1868          * count them correctly into ->nr_threads. We don't take tasklist
1869          * lock, but this is safe wrt:
1870          *
1871          * fork:
1872          *      None of sub-threads can fork after zap_process(leader). All
1873          *      processes which were created before this point should be
1874          *      visible to zap_threads() because copy_process() adds the new
1875          *      process to the tail of init_task.tasks list, and lock/unlock
1876          *      of ->siglock provides a memory barrier.
1877          *
1878          * do_exit:
1879          *      The caller holds mm->mmap_sem. This means that the task which
1880          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1881          *      its ->mm.
1882          *
1883          * de_thread:
1884          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1885          *      we must see either old or new leader, this does not matter.
1886          *      However, it can change p->sighand, so lock_task_sighand(p)
1887          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1888          *      it can't fail.
1889          *
1890          *      Note also that "g" can be the old leader with ->mm == NULL
1891          *      and already unhashed and thus removed from ->thread_group.
1892          *      This is OK, __unhash_process()->list_del_rcu() does not
1893          *      clear the ->next pointer, we will find the new leader via
1894          *      next_thread().
1895          */
1896         rcu_read_lock();
1897         for_each_process(g) {
1898                 if (g == tsk->group_leader)
1899                         continue;
1900                 if (g->flags & PF_KTHREAD)
1901                         continue;
1902                 p = g;
1903                 do {
1904                         if (p->mm) {
1905                                 if (unlikely(p->mm == mm)) {
1906                                         lock_task_sighand(p, &flags);
1907                                         nr += zap_process(p, exit_code);
1908                                         unlock_task_sighand(p, &flags);
1909                                 }
1910                                 break;
1911                         }
1912                 } while_each_thread(g, p);
1913         }
1914         rcu_read_unlock();
1915 done:
1916         atomic_set(&core_state->nr_threads, nr);
1917         return nr;
1918 }
1919
1920 static int coredump_wait(int exit_code, struct core_state *core_state)
1921 {
1922         struct task_struct *tsk = current;
1923         struct mm_struct *mm = tsk->mm;
1924         int core_waiters = -EBUSY;
1925
1926         init_completion(&core_state->startup);
1927         core_state->dumper.task = tsk;
1928         core_state->dumper.next = NULL;
1929
1930         down_write(&mm->mmap_sem);
1931         if (!mm->core_state)
1932                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1933         up_write(&mm->mmap_sem);
1934
1935         if (core_waiters > 0)
1936                 wait_for_completion(&core_state->startup);
1937
1938         return core_waiters;
1939 }
1940
1941 static void coredump_finish(struct mm_struct *mm)
1942 {
1943         struct core_thread *curr, *next;
1944         struct task_struct *task;
1945
1946         next = mm->core_state->dumper.next;
1947         while ((curr = next) != NULL) {
1948                 next = curr->next;
1949                 task = curr->task;
1950                 /*
1951                  * see exit_mm(), curr->task must not see
1952                  * ->task == NULL before we read ->next.
1953                  */
1954                 smp_mb();
1955                 curr->task = NULL;
1956                 wake_up_process(task);
1957         }
1958
1959         mm->core_state = NULL;
1960 }
1961
1962 /*
1963  * set_dumpable converts traditional three-value dumpable to two flags and
1964  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1965  * these bits are not changed atomically.  So get_dumpable can observe the
1966  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1967  * return either old dumpable or new one by paying attention to the order of
1968  * modifying the bits.
1969  *
1970  * dumpable |   mm->flags (binary)
1971  * old  new | initial interim  final
1972  * ---------+-----------------------
1973  *  0    1  |   00      01      01
1974  *  0    2  |   00      10(*)   11
1975  *  1    0  |   01      00      00
1976  *  1    2  |   01      11      11
1977  *  2    0  |   11      10(*)   00
1978  *  2    1  |   11      11      01
1979  *
1980  * (*) get_dumpable regards interim value of 10 as 11.
1981  */
1982 void set_dumpable(struct mm_struct *mm, int value)
1983 {
1984         switch (value) {
1985         case 0:
1986                 clear_bit(MMF_DUMPABLE, &mm->flags);
1987                 smp_wmb();
1988                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1989                 break;
1990         case 1:
1991                 set_bit(MMF_DUMPABLE, &mm->flags);
1992                 smp_wmb();
1993                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1994                 break;
1995         case 2:
1996                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1997                 smp_wmb();
1998                 set_bit(MMF_DUMPABLE, &mm->flags);
1999                 break;
2000         }
2001 }
2002
2003 static int __get_dumpable(unsigned long mm_flags)
2004 {
2005         int ret;
2006
2007         ret = mm_flags & MMF_DUMPABLE_MASK;
2008         return (ret >= 2) ? 2 : ret;
2009 }
2010
2011 int get_dumpable(struct mm_struct *mm)
2012 {
2013         return __get_dumpable(mm->flags);
2014 }
2015
2016 static void wait_for_dump_helpers(struct file *file)
2017 {
2018         struct pipe_inode_info *pipe;
2019
2020         pipe = file->f_path.dentry->d_inode->i_pipe;
2021
2022         pipe_lock(pipe);
2023         pipe->readers++;
2024         pipe->writers--;
2025
2026         while ((pipe->readers > 1) && (!signal_pending(current))) {
2027                 wake_up_interruptible_sync(&pipe->wait);
2028                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2029                 pipe_wait(pipe);
2030         }
2031
2032         pipe->readers--;
2033         pipe->writers++;
2034         pipe_unlock(pipe);
2035
2036 }
2037
2038
2039 /*
2040  * umh_pipe_setup
2041  * helper function to customize the process used
2042  * to collect the core in userspace.  Specifically
2043  * it sets up a pipe and installs it as fd 0 (stdin)
2044  * for the process.  Returns 0 on success, or
2045  * PTR_ERR on failure.
2046  * Note that it also sets the core limit to 1.  This
2047  * is a special value that we use to trap recursive
2048  * core dumps
2049  */
2050 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2051 {
2052         struct file *rp, *wp;
2053         struct fdtable *fdt;
2054         struct coredump_params *cp = (struct coredump_params *)info->data;
2055         struct files_struct *cf = current->files;
2056
2057         wp = create_write_pipe(0);
2058         if (IS_ERR(wp))
2059                 return PTR_ERR(wp);
2060
2061         rp = create_read_pipe(wp, 0);
2062         if (IS_ERR(rp)) {
2063                 free_write_pipe(wp);
2064                 return PTR_ERR(rp);
2065         }
2066
2067         cp->file = wp;
2068
2069         sys_close(0);
2070         fd_install(0, rp);
2071         spin_lock(&cf->file_lock);
2072         fdt = files_fdtable(cf);
2073         FD_SET(0, fdt->open_fds);
2074         FD_CLR(0, fdt->close_on_exec);
2075         spin_unlock(&cf->file_lock);
2076
2077         /* and disallow core files too */
2078         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2079
2080         return 0;
2081 }
2082
2083 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2084 {
2085         struct core_state core_state;
2086         struct core_name cn;
2087         struct mm_struct *mm = current->mm;
2088         struct linux_binfmt * binfmt;
2089         const struct cred *old_cred;
2090         struct cred *cred;
2091         int retval = 0;
2092         int flag = 0;
2093         int ispipe;
2094         static atomic_t core_dump_count = ATOMIC_INIT(0);
2095         struct coredump_params cprm = {
2096                 .signr = signr,
2097                 .regs = regs,
2098                 .limit = rlimit(RLIMIT_CORE),
2099                 /*
2100                  * We must use the same mm->flags while dumping core to avoid
2101                  * inconsistency of bit flags, since this flag is not protected
2102                  * by any locks.
2103                  */
2104                 .mm_flags = mm->flags,
2105         };
2106
2107         audit_core_dumps(signr);
2108
2109         binfmt = mm->binfmt;
2110         if (!binfmt || !binfmt->core_dump)
2111                 goto fail;
2112         if (!__get_dumpable(cprm.mm_flags))
2113                 goto fail;
2114
2115         cred = prepare_creds();
2116         if (!cred)
2117                 goto fail;
2118         /*
2119          *      We cannot trust fsuid as being the "true" uid of the
2120          *      process nor do we know its entire history. We only know it
2121          *      was tainted so we dump it as root in mode 2.
2122          */
2123         if (__get_dumpable(cprm.mm_flags) == 2) {
2124                 /* Setuid core dump mode */
2125                 flag = O_EXCL;          /* Stop rewrite attacks */
2126                 cred->fsuid = 0;        /* Dump root private */
2127         }
2128
2129         retval = coredump_wait(exit_code, &core_state);
2130         if (retval < 0)
2131                 goto fail_creds;
2132
2133         old_cred = override_creds(cred);
2134
2135         /*
2136          * Clear any false indication of pending signals that might
2137          * be seen by the filesystem code called to write the core file.
2138          */
2139         clear_thread_flag(TIF_SIGPENDING);
2140
2141         ispipe = format_corename(&cn, signr);
2142
2143         if (ispipe) {
2144                 int dump_count;
2145                 char **helper_argv;
2146
2147                 if (ispipe < 0) {
2148                         printk(KERN_WARNING "format_corename failed\n");
2149                         printk(KERN_WARNING "Aborting core\n");
2150                         goto fail_corename;
2151                 }
2152
2153                 if (cprm.limit == 1) {
2154                         /*
2155                          * Normally core limits are irrelevant to pipes, since
2156                          * we're not writing to the file system, but we use
2157                          * cprm.limit of 1 here as a speacial value. Any
2158                          * non-1 limit gets set to RLIM_INFINITY below, but
2159                          * a limit of 0 skips the dump.  This is a consistent
2160                          * way to catch recursive crashes.  We can still crash
2161                          * if the core_pattern binary sets RLIM_CORE =  !1
2162                          * but it runs as root, and can do lots of stupid things
2163                          * Note that we use task_tgid_vnr here to grab the pid
2164                          * of the process group leader.  That way we get the
2165                          * right pid if a thread in a multi-threaded
2166                          * core_pattern process dies.
2167                          */
2168                         printk(KERN_WARNING
2169                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2170                                 task_tgid_vnr(current), current->comm);
2171                         printk(KERN_WARNING "Aborting core\n");
2172                         goto fail_unlock;
2173                 }
2174                 cprm.limit = RLIM_INFINITY;
2175
2176                 dump_count = atomic_inc_return(&core_dump_count);
2177                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2178                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2179                                task_tgid_vnr(current), current->comm);
2180                         printk(KERN_WARNING "Skipping core dump\n");
2181                         goto fail_dropcount;
2182                 }
2183
2184                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2185                 if (!helper_argv) {
2186                         printk(KERN_WARNING "%s failed to allocate memory\n",
2187                                __func__);
2188                         goto fail_dropcount;
2189                 }
2190
2191                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2192                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2193                                         NULL, &cprm);
2194                 argv_free(helper_argv);
2195                 if (retval) {
2196                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2197                                cn.corename);
2198                         goto close_fail;
2199                 }
2200         } else {
2201                 struct inode *inode;
2202
2203                 if (cprm.limit < binfmt->min_coredump)
2204                         goto fail_unlock;
2205
2206                 cprm.file = filp_open(cn.corename,
2207                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2208                                  0600);
2209                 if (IS_ERR(cprm.file))
2210                         goto fail_unlock;
2211
2212                 inode = cprm.file->f_path.dentry->d_inode;
2213                 if (inode->i_nlink > 1)
2214                         goto close_fail;
2215                 if (d_unhashed(cprm.file->f_path.dentry))
2216                         goto close_fail;
2217                 /*
2218                  * AK: actually i see no reason to not allow this for named
2219                  * pipes etc, but keep the previous behaviour for now.
2220                  */
2221                 if (!S_ISREG(inode->i_mode))
2222                         goto close_fail;
2223                 /*
2224                  * Dont allow local users get cute and trick others to coredump
2225                  * into their pre-created files.
2226                  */
2227                 if (inode->i_uid != current_fsuid())
2228                         goto close_fail;
2229                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2230                         goto close_fail;
2231                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2232                         goto close_fail;
2233         }
2234
2235         retval = binfmt->core_dump(&cprm);
2236         if (retval)
2237                 current->signal->group_exit_code |= 0x80;
2238
2239         if (ispipe && core_pipe_limit)
2240                 wait_for_dump_helpers(cprm.file);
2241 close_fail:
2242         if (cprm.file)
2243                 filp_close(cprm.file, NULL);
2244 fail_dropcount:
2245         if (ispipe)
2246                 atomic_dec(&core_dump_count);
2247 fail_unlock:
2248         kfree(cn.corename);
2249 fail_corename:
2250         coredump_finish(mm);
2251         revert_creds(old_cred);
2252 fail_creds:
2253         put_cred(cred);
2254 fail:
2255         return;
2256 }
2257
2258 /*
2259  * Core dumping helper functions.  These are the only things you should
2260  * do on a core-file: use only these functions to write out all the
2261  * necessary info.
2262  */
2263 int dump_write(struct file *file, const void *addr, int nr)
2264 {
2265         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2266 }
2267 EXPORT_SYMBOL(dump_write);
2268
2269 int dump_seek(struct file *file, loff_t off)
2270 {
2271         int ret = 1;
2272
2273         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2274                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2275                         return 0;
2276         } else {
2277                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2278
2279                 if (!buf)
2280                         return 0;
2281                 while (off > 0) {
2282                         unsigned long n = off;
2283
2284                         if (n > PAGE_SIZE)
2285                                 n = PAGE_SIZE;
2286                         if (!dump_write(file, buf, n)) {
2287                                 ret = 0;
2288                                 break;
2289                         }
2290                         off -= n;
2291                 }
2292                 free_page((unsigned long)buf);
2293         }
2294         return ret;
2295 }
2296 EXPORT_SYMBOL(dump_seek);