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 #include "coredump.h"
  66
  67 #include <trace/events/sched.h>
  68
  69 int suid_dumpable = 0;
  70
  71 static LIST_HEAD(formats);
  72 static DEFINE_RWLOCK(binfmt_lock);
  73
  74 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  75 {
  76         BUG_ON(!fmt);
  77         if (WARN_ON(!fmt->load_binary))
  78                 return;
  79         write_lock(&binfmt_lock);
  80         insert ? list_add(&fmt->lh, &formats) :
  81                  list_add_tail(&fmt->lh, &formats);
  82         write_unlock(&binfmt_lock);
  83 }
  84
  85 EXPORT_SYMBOL(__register_binfmt);
  86
  87 void unregister_binfmt(struct linux_binfmt * fmt)
  88 {
  89         write_lock(&binfmt_lock);
  90         list_del(&fmt->lh);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(unregister_binfmt);
  95
  96 static inline void put_binfmt(struct linux_binfmt * fmt)
  97 {
  98         module_put(fmt->module);
  99 }
 100
 101 /*
 102  * Note that a shared library must be both readable and executable due to
 103  * security reasons.
 104  *
 105  * Also note that we take the address to load from from the file itself.
 106  */
 107 SYSCALL_DEFINE1(uselib, const char __user *, library)
 108 {
 109         struct file *file;
 110         struct filename *tmp = getname(library);
 111         int error = PTR_ERR(tmp);
 112         static const struct open_flags uselib_flags = {
 113                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 114                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 115                 .intent = LOOKUP_OPEN,
 116                 .lookup_flags = LOOKUP_FOLLOW,
 117         };
 118
 119         if (IS_ERR(tmp))
 120                 goto out;
 121
 122         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 123         putname(tmp);
 124         error = PTR_ERR(file);
 125         if (IS_ERR(file))
 126                 goto out;
 127
 128         error = -EINVAL;
 129         if (!S_ISREG(file_inode(file)->i_mode))
 130                 goto exit;
 131
 132         error = -EACCES;
 133         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 134                 goto exit;
 135
 136         fsnotify_open(file);
 137
 138         error = -ENOEXEC;
 139         if(file->f_op) {
 140                 struct linux_binfmt * fmt;
 141
 142                 read_lock(&binfmt_lock);
 143                 list_for_each_entry(fmt, &formats, lh) {
 144                         if (!fmt->load_shlib)
 145                                 continue;
 146                         if (!try_module_get(fmt->module))
 147                                 continue;
 148                         read_unlock(&binfmt_lock);
 149                         error = fmt->load_shlib(file);
 150                         read_lock(&binfmt_lock);
 151                         put_binfmt(fmt);
 152                         if (error != -ENOEXEC)
 153                                 break;
 154                 }
 155                 read_unlock(&binfmt_lock);
 156         }
 157 exit:
 158         fput(file);
 159 out:
 160         return error;
 161 }
 162
 163 #ifdef CONFIG_MMU
 164 /*
 165  * The nascent bprm->mm is not visible until exec_mmap() but it can
 166  * use a lot of memory, account these pages in current->mm temporary
 167  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 168  * change the counter back via acct_arg_size(0).
 169  */
 170 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 171 {
 172         struct mm_struct *mm = current->mm;
 173         long diff = (long)(pages - bprm->vma_pages);
 174
 175         if (!mm || !diff)
 176                 return;
 177
 178         bprm->vma_pages = pages;
 179         add_mm_counter(mm, MM_ANONPAGES, diff);
 180 }
 181
 182 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 183                 int write)
 184 {
 185         struct page *page;
 186         int ret;
 187
 188 #ifdef CONFIG_STACK_GROWSUP
 189         if (write) {
 190                 ret = expand_downwards(bprm->vma, pos);
 191                 if (ret < 0)
 192                         return NULL;
 193         }
 194 #endif
 195         ret = get_user_pages(current, bprm->mm, pos,
 196                         1, write, 1, &page, NULL);
 197         if (ret <= 0)
 198                 return NULL;
 199
 200         if (write) {
 201                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 202                 struct rlimit *rlim;
 203
 204                 acct_arg_size(bprm, size / PAGE_SIZE);
 205
 206                 /*
 207                  * We've historically supported up to 32 pages (ARG_MAX)
 208                  * of argument strings even with small stacks
 209                  */
 210                 if (size <= ARG_MAX)
 211                         return page;
 212
 213                 /*
 214                  * Limit to 1/4-th the stack size for the argv+env strings.
 215                  * This ensures that:
 216                  *  - the remaining binfmt code will not run out of stack space,
 217                  *  - the program will have a reasonable amount of stack left
 218                  *    to work from.
 219                  */
 220                 rlim = current->signal->rlim;
 221                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 222                         put_page(page);
 223                         return NULL;
 224                 }
 225         }
 226
 227         return page;
 228 }
 229
 230 static void put_arg_page(struct page *page)
 231 {
 232         put_page(page);
 233 }
 234
 235 static void free_arg_page(struct linux_binprm *bprm, int i)
 236 {
 237 }
 238
 239 static void free_arg_pages(struct linux_binprm *bprm)
 240 {
 241 }
 242
 243 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 244                 struct page *page)
 245 {
 246         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 247 }
 248
 249 static int __bprm_mm_init(struct linux_binprm *bprm)
 250 {
 251         int err;
 252         struct vm_area_struct *vma = NULL;
 253         struct mm_struct *mm = bprm->mm;
 254
 255         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 256         if (!vma)
 257                 return -ENOMEM;
 258
 259         down_write(&mm->mmap_sem);
 260         vma->vm_mm = mm;
 261
 262         /*
 263          * Place the stack at the largest stack address the architecture
 264          * supports. Later, we'll move this to an appropriate place. We don't
 265          * use STACK_TOP because that can depend on attributes which aren't
 266          * configured yet.
 267          */
 268         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 269         vma->vm_end = STACK_TOP_MAX;
 270         vma->vm_start = vma->vm_end - PAGE_SIZE;
 271         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 272         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 273         INIT_LIST_HEAD(&vma->anon_vma_chain);
 274
 275         err = insert_vm_struct(mm, vma);
 276         if (err)
 277                 goto err;
 278
 279         mm->stack_vm = mm->total_vm = 1;
 280         up_write(&mm->mmap_sem);
 281         bprm->p = vma->vm_end - sizeof(void *);
 282         return 0;
 283 err:
 284         up_write(&mm->mmap_sem);
 285         bprm->vma = NULL;
 286         kmem_cache_free(vm_area_cachep, vma);
 287         return err;
 288 }
 289
 290 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 291 {
 292         return len <= MAX_ARG_STRLEN;
 293 }
 294
 295 #else
 296
 297 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 298 {
 299 }
 300
 301 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 302                 int write)
 303 {
 304         struct page *page;
 305
 306         page = bprm->page[pos / PAGE_SIZE];
 307         if (!page && write) {
 308                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 309                 if (!page)
 310                         return NULL;
 311                 bprm->page[pos / PAGE_SIZE] = page;
 312         }
 313
 314         return page;
 315 }
 316
 317 static void put_arg_page(struct page *page)
 318 {
 319 }
 320
 321 static void free_arg_page(struct linux_binprm *bprm, int i)
 322 {
 323         if (bprm->page[i]) {
 324                 __free_page(bprm->page[i]);
 325                 bprm->page[i] = NULL;
 326         }
 327 }
 328
 329 static void free_arg_pages(struct linux_binprm *bprm)
 330 {
 331         int i;
 332
 333         for (i = 0; i < MAX_ARG_PAGES; i++)
 334                 free_arg_page(bprm, i);
 335 }
 336
 337 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 338                 struct page *page)
 339 {
 340 }
 341
 342 static int __bprm_mm_init(struct linux_binprm *bprm)
 343 {
 344         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 345         return 0;
 346 }
 347
 348 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 349 {
 350         return len <= bprm->p;
 351 }
 352
 353 #endif /* CONFIG_MMU */
 354
 355 /*
 356  * Create a new mm_struct and populate it with a temporary stack
 357  * vm_area_struct.  We don't have enough context at this point to set the stack
 358  * flags, permissions, and offset, so we use temporary values.  We'll update
 359  * them later in setup_arg_pages().
 360  */
 361 static int bprm_mm_init(struct linux_binprm *bprm)
 362 {
 363         int err;
 364         struct mm_struct *mm = NULL;
 365
 366         bprm->mm = mm = mm_alloc();
 367         err = -ENOMEM;
 368         if (!mm)
 369                 goto err;
 370
 371         err = init_new_context(current, mm);
 372         if (err)
 373                 goto err;
 374
 375         err = __bprm_mm_init(bprm);
 376         if (err)
 377                 goto err;
 378
 379         return 0;
 380
 381 err:
 382         if (mm) {
 383                 bprm->mm = NULL;
 384                 mmdrop(mm);
 385         }
 386
 387         return err;
 388 }
 389
 390 struct user_arg_ptr {
 391 #ifdef CONFIG_COMPAT
 392         bool is_compat;
 393 #endif
 394         union {
 395                 const char __user *const __user *native;
 396 #ifdef CONFIG_COMPAT
 397                 const compat_uptr_t __user *compat;
 398 #endif
 399         } ptr;
 400 };
 401
 402 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 403 {
 404         const char __user *native;
 405
 406 #ifdef CONFIG_COMPAT
 407         if (unlikely(argv.is_compat)) {
 408                 compat_uptr_t compat;
 409
 410                 if (get_user(compat, argv.ptr.compat + nr))
 411                         return ERR_PTR(-EFAULT);
 412
 413                 return compat_ptr(compat);
 414         }
 415 #endif
 416
 417         if (get_user(native, argv.ptr.native + nr))
 418                 return ERR_PTR(-EFAULT);
 419
 420         return native;
 421 }
 422
 423 /*
 424  * count() counts the number of strings in array ARGV.
 425  */
 426 static int count(struct user_arg_ptr argv, int max)
 427 {
 428         int i = 0;
 429
 430         if (argv.ptr.native != NULL) {
 431                 for (;;) {
 432                         const char __user *p = get_user_arg_ptr(argv, i);
 433
 434                         if (!p)
 435                                 break;
 436
 437                         if (IS_ERR(p))
 438                                 return -EFAULT;
 439
 440                         if (i >= max)
 441                                 return -E2BIG;
 442                         ++i;
 443
 444                         if (fatal_signal_pending(current))
 445                                 return -ERESTARTNOHAND;
 446                         cond_resched();
 447                 }
 448         }
 449         return i;
 450 }
 451
 452 /*
 453  * 'copy_strings()' copies argument/environment strings from the old
 454  * processes's memory to the new process's stack.  The call to get_user_pages()
 455  * ensures the destination page is created and not swapped out.
 456  */
 457 static int copy_strings(int argc, struct user_arg_ptr argv,
 458                         struct linux_binprm *bprm)
 459 {
 460         struct page *kmapped_page = NULL;
 461         char *kaddr = NULL;
 462         unsigned long kpos = 0;
 463         int ret;
 464
 465         while (argc-- > 0) {
 466                 const char __user *str;
 467                 int len;
 468                 unsigned long pos;
 469
 470                 ret = -EFAULT;
 471                 str = get_user_arg_ptr(argv, argc);
 472                 if (IS_ERR(str))
 473                         goto out;
 474
 475                 len = strnlen_user(str, MAX_ARG_STRLEN);
 476                 if (!len)
 477                         goto out;
 478
 479                 ret = -E2BIG;
 480                 if (!valid_arg_len(bprm, len))
 481                         goto out;
 482
 483                 /* We're going to work our way backwords. */
 484                 pos = bprm->p;
 485                 str += len;
 486                 bprm->p -= len;
 487
 488                 while (len > 0) {
 489                         int offset, bytes_to_copy;
 490
 491                         if (fatal_signal_pending(current)) {
 492                                 ret = -ERESTARTNOHAND;
 493                                 goto out;
 494                         }
 495                         cond_resched();
 496
 497                         offset = pos % PAGE_SIZE;
 498                         if (offset == 0)
 499                                 offset = PAGE_SIZE;
 500
 501                         bytes_to_copy = offset;
 502                         if (bytes_to_copy > len)
 503                                 bytes_to_copy = len;
 504
 505                         offset -= bytes_to_copy;
 506                         pos -= bytes_to_copy;
 507                         str -= bytes_to_copy;
 508                         len -= bytes_to_copy;
 509
 510                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 511                                 struct page *page;
 512
 513                                 page = get_arg_page(bprm, pos, 1);
 514                                 if (!page) {
 515                                         ret = -E2BIG;
 516                                         goto out;
 517                                 }
 518
 519                                 if (kmapped_page) {
 520                                         flush_kernel_dcache_page(kmapped_page);
 521                                         kunmap(kmapped_page);
 522                                         put_arg_page(kmapped_page);
 523                                 }
 524                                 kmapped_page = page;
 525                                 kaddr = kmap(kmapped_page);
 526                                 kpos = pos & PAGE_MASK;
 527                                 flush_arg_page(bprm, kpos, kmapped_page);
 528                         }
 529                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 530                                 ret = -EFAULT;
 531                                 goto out;
 532                         }
 533                 }
 534         }
 535         ret = 0;
 536 out:
 537         if (kmapped_page) {
 538                 flush_kernel_dcache_page(kmapped_page);
 539                 kunmap(kmapped_page);
 540                 put_arg_page(kmapped_page);
 541         }
 542         return ret;
 543 }
 544
 545 /*
 546  * Like copy_strings, but get argv and its values from kernel memory.
 547  */
 548 int copy_strings_kernel(int argc, const char *const *__argv,
 549                         struct linux_binprm *bprm)
 550 {
 551         int r;
 552         mm_segment_t oldfs = get_fs();
 553         struct user_arg_ptr argv = {
 554                 .ptr.native = (const char __user *const  __user *)__argv,
 555         };
 556
 557         set_fs(KERNEL_DS);
 558         r = copy_strings(argc, argv, bprm);
 559         set_fs(oldfs);
 560
 561         return r;
 562 }
 563 EXPORT_SYMBOL(copy_strings_kernel);
 564
 565 #ifdef CONFIG_MMU
 566
 567 /*
 568  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 569  * the binfmt code determines where the new stack should reside, we shift it to
 570  * its final location.  The process proceeds as follows:
 571  *
 572  * 1) Use shift to calculate the new vma endpoints.
 573  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 574  *    arguments passed to subsequent functions are consistent.
 575  * 3) Move vma's page tables to the new range.
 576  * 4) Free up any cleared pgd range.
 577  * 5) Shrink the vma to cover only the new range.
 578  */
 579 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 580 {
 581         struct mm_struct *mm = vma->vm_mm;
 582         unsigned long old_start = vma->vm_start;
 583         unsigned long old_end = vma->vm_end;
 584         unsigned long length = old_end - old_start;
 585         unsigned long new_start = old_start - shift;
 586         unsigned long new_end = old_end - shift;
 587         struct mmu_gather tlb;
 588
 589         BUG_ON(new_start > new_end);
 590
 591         /*
 592          * ensure there are no vmas between where we want to go
 593          * and where we are
 594          */
 595         if (vma != find_vma(mm, new_start))
 596                 return -EFAULT;
 597
 598         /*
 599          * cover the whole range: [new_start, old_end)
 600          */
 601         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 602                 return -ENOMEM;
 603
 604         /*
 605          * move the page tables downwards, on failure we rely on
 606          * process cleanup to remove whatever mess we made.
 607          */
 608         if (length != move_page_tables(vma, old_start,
 609                                        vma, new_start, length, false))
 610                 return -ENOMEM;
 611
 612         lru_add_drain();
 613         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 614         if (new_end > old_start) {
 615                 /*
 616                  * when the old and new regions overlap clear from new_end.
 617                  */
 618                 free_pgd_range(&tlb, new_end, old_end, new_end,
 619                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 620         } else {
 621                 /*
 622                  * otherwise, clean from old_start; this is done to not touch
 623                  * the address space in [new_end, old_start) some architectures
 624                  * have constraints on va-space that make this illegal (IA64) -
 625                  * for the others its just a little faster.
 626                  */
 627                 free_pgd_range(&tlb, old_start, old_end, new_end,
 628                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 629         }
 630         tlb_finish_mmu(&tlb, old_start, old_end);
 631
 632         /*
 633          * Shrink the vma to just the new range.  Always succeeds.
 634          */
 635         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 636
 637         return 0;
 638 }
 639
 640 /*
 641  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 642  * the stack is optionally relocated, and some extra space is added.
 643  */
 644 int setup_arg_pages(struct linux_binprm *bprm,
 645                     unsigned long stack_top,
 646                     int executable_stack)
 647 {
 648         unsigned long ret;
 649         unsigned long stack_shift;
 650         struct mm_struct *mm = current->mm;
 651         struct vm_area_struct *vma = bprm->vma;
 652         struct vm_area_struct *prev = NULL;
 653         unsigned long vm_flags;
 654         unsigned long stack_base;
 655         unsigned long stack_size;
 656         unsigned long stack_expand;
 657         unsigned long rlim_stack;
 658
 659 #ifdef CONFIG_STACK_GROWSUP
 660         /* Limit stack size to 1GB */
 661         stack_base = rlimit_max(RLIMIT_STACK);
 662         if (stack_base > (1 << 30))
 663                 stack_base = 1 << 30;
 664
 665         /* Make sure we didn't let the argument array grow too large. */
 666         if (vma->vm_end - vma->vm_start > stack_base)
 667                 return -ENOMEM;
 668
 669         stack_base = PAGE_ALIGN(stack_top - stack_base);
 670
 671         stack_shift = vma->vm_start - stack_base;
 672         mm->arg_start = bprm->p - stack_shift;
 673         bprm->p = vma->vm_end - stack_shift;
 674 #else
 675         stack_top = arch_align_stack(stack_top);
 676         stack_top = PAGE_ALIGN(stack_top);
 677
 678         if (unlikely(stack_top < mmap_min_addr) ||
 679             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 680                 return -ENOMEM;
 681
 682         stack_shift = vma->vm_end - stack_top;
 683
 684         bprm->p -= stack_shift;
 685         mm->arg_start = bprm->p;
 686 #endif
 687
 688         if (bprm->loader)
 689                 bprm->loader -= stack_shift;
 690         bprm->exec -= stack_shift;
 691
 692         down_write(&mm->mmap_sem);
 693         vm_flags = VM_STACK_FLAGS;
 694
 695         /*
 696          * Adjust stack execute permissions; explicitly enable for
 697          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 698          * (arch default) otherwise.
 699          */
 700         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 701                 vm_flags |= VM_EXEC;
 702         else if (executable_stack == EXSTACK_DISABLE_X)
 703                 vm_flags &= ~VM_EXEC;
 704         vm_flags |= mm->def_flags;
 705         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 706
 707         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 708                         vm_flags);
 709         if (ret)
 710                 goto out_unlock;
 711         BUG_ON(prev != vma);
 712
 713         /* Move stack pages down in memory. */
 714         if (stack_shift) {
 715                 ret = shift_arg_pages(vma, stack_shift);
 716                 if (ret)
 717                         goto out_unlock;
 718         }
 719
 720         /* mprotect_fixup is overkill to remove the temporary stack flags */
 721         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 722
 723         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 724         stack_size = vma->vm_end - vma->vm_start;
 725         /*
 726          * Align this down to a page boundary as expand_stack
 727          * will align it up.
 728          */
 729         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 730 #ifdef CONFIG_STACK_GROWSUP
 731         if (stack_size + stack_expand > rlim_stack)
 732                 stack_base = vma->vm_start + rlim_stack;
 733         else
 734                 stack_base = vma->vm_end + stack_expand;
 735 #else
 736         if (stack_size + stack_expand > rlim_stack)
 737                 stack_base = vma->vm_end - rlim_stack;
 738         else
 739                 stack_base = vma->vm_start - stack_expand;
 740 #endif
 741         current->mm->start_stack = bprm->p;
 742         ret = expand_stack(vma, stack_base);
 743         if (ret)
 744                 ret = -EFAULT;
 745
 746 out_unlock:
 747         up_write(&mm->mmap_sem);
 748         return ret;
 749 }
 750 EXPORT_SYMBOL(setup_arg_pages);
 751
 752 #endif /* CONFIG_MMU */
 753
 754 struct file *open_exec(const char *name)
 755 {
 756         struct file *file;
 757         int err;
 758         struct filename tmp = { .name = name };
 759         static const struct open_flags open_exec_flags = {
 760                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 761                 .acc_mode = MAY_EXEC | MAY_OPEN,
 762                 .intent = LOOKUP_OPEN,
 763                 .lookup_flags = LOOKUP_FOLLOW,
 764         };
 765
 766         file = do_filp_open(AT_FDCWD, &tmp, &open_exec_flags);
 767         if (IS_ERR(file))
 768                 goto out;
 769
 770         err = -EACCES;
 771         if (!S_ISREG(file_inode(file)->i_mode))
 772                 goto exit;
 773
 774         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 775                 goto exit;
 776
 777         fsnotify_open(file);
 778
 779         err = deny_write_access(file);
 780         if (err)
 781                 goto exit;
 782
 783 out:
 784         return file;
 785
 786 exit:
 787         fput(file);
 788         return ERR_PTR(err);
 789 }
 790 EXPORT_SYMBOL(open_exec);
 791
 792 int kernel_read(struct file *file, loff_t offset,
 793                 char *addr, unsigned long count)
 794 {
 795         mm_segment_t old_fs;
 796         loff_t pos = offset;
 797         int result;
 798
 799         old_fs = get_fs();
 800         set_fs(get_ds());
 801         /* The cast to a user pointer is valid due to the set_fs() */
 802         result = vfs_read(file, (void __user *)addr, count, &pos);
 803         set_fs(old_fs);
 804         return result;
 805 }
 806
 807 EXPORT_SYMBOL(kernel_read);
 808
 809 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 810 {
 811         ssize_t res = file->f_op->read(file, (void __user *)addr, len, &pos);
 812         if (res > 0)
 813                 flush_icache_range(addr, addr + len);
 814         return res;
 815 }
 816 EXPORT_SYMBOL(read_code);
 817
 818 static int exec_mmap(struct mm_struct *mm)
 819 {
 820         struct task_struct *tsk;
 821         struct mm_struct * old_mm, *active_mm;
 822
 823         /* Notify parent that we're no longer interested in the old VM */
 824         tsk = current;
 825         old_mm = current->mm;
 826         mm_release(tsk, old_mm);
 827
 828         if (old_mm) {
 829                 sync_mm_rss(old_mm);
 830                 /*
 831                  * Make sure that if there is a core dump in progress
 832                  * for the old mm, we get out and die instead of going
 833                  * through with the exec.  We must hold mmap_sem around
 834                  * checking core_state and changing tsk->mm.
 835                  */
 836                 down_read(&old_mm->mmap_sem);
 837                 if (unlikely(old_mm->core_state)) {
 838                         up_read(&old_mm->mmap_sem);
 839                         return -EINTR;
 840                 }
 841         }
 842         task_lock(tsk);
 843         active_mm = tsk->active_mm;
 844         tsk->mm = mm;
 845         tsk->active_mm = mm;
 846         activate_mm(active_mm, mm);
 847         task_unlock(tsk);
 848         arch_pick_mmap_layout(mm);
 849         if (old_mm) {
 850                 up_read(&old_mm->mmap_sem);
 851                 BUG_ON(active_mm != old_mm);
 852                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, 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_KILLABLE);
 896                 spin_unlock_irq(lock);
 897                 schedule();
 898                 if (unlikely(__fatal_signal_pending(tsk)))
 899                         goto killed;
 900                 spin_lock_irq(lock);
 901         }
 902         spin_unlock_irq(lock);
 903
 904         /*
 905          * At this point all other threads have exited, all we have to
 906          * do is to wait for the thread group leader to become inactive,
 907          * and to assume its PID:
 908          */
 909         if (!thread_group_leader(tsk)) {
 910                 struct task_struct *leader = tsk->group_leader;
 911
 912                 sig->notify_count = -1; /* for exit_notify() */
 913                 for (;;) {
 914                         threadgroup_change_begin(tsk);
 915                         write_lock_irq(&tasklist_lock);
 916                         if (likely(leader->exit_state))
 917                                 break;
 918                         __set_current_state(TASK_KILLABLE);
 919                         write_unlock_irq(&tasklist_lock);
 920                         threadgroup_change_end(tsk);
 921                         schedule();
 922                         if (unlikely(__fatal_signal_pending(tsk)))
 923                                 goto killed;
 924                 }
 925
 926                 /*
 927                  * The only record we have of the real-time age of a
 928                  * process, regardless of execs it's done, is start_time.
 929                  * All the past CPU time is accumulated in signal_struct
 930                  * from sister threads now dead.  But in this non-leader
 931                  * exec, nothing survives from the original leader thread,
 932                  * whose birth marks the true age of this process now.
 933                  * When we take on its identity by switching to its PID, we
 934                  * also take its birthdate (always earlier than our own).
 935                  */
 936                 tsk->start_time = leader->start_time;
 937                 tsk->real_start_time = leader->real_start_time;
 938
 939                 BUG_ON(!same_thread_group(leader, tsk));
 940                 BUG_ON(has_group_leader_pid(tsk));
 941                 /*
 942                  * An exec() starts a new thread group with the
 943                  * TGID of the previous thread group. Rehash the
 944                  * two threads with a switched PID, and release
 945                  * the former thread group leader:
 946                  */
 947
 948                 /* Become a process group leader with the old leader's pid.
 949                  * The old leader becomes a thread of the this thread group.
 950                  * Note: The old leader also uses this pid until release_task
 951                  *       is called.  Odd but simple and correct.
 952                  */
 953                 tsk->pid = leader->pid;
 954                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
 955                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 956                 transfer_pid(leader, tsk, PIDTYPE_SID);
 957
 958                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 959                 list_replace_init(&leader->sibling, &tsk->sibling);
 960
 961                 tsk->group_leader = tsk;
 962                 leader->group_leader = tsk;
 963
 964                 tsk->exit_signal = SIGCHLD;
 965                 leader->exit_signal = -1;
 966
 967                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 968                 leader->exit_state = EXIT_DEAD;
 969
 970                 /*
 971                  * We are going to release_task()->ptrace_unlink() silently,
 972                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 973                  * the tracer wont't block again waiting for this thread.
 974                  */
 975                 if (unlikely(leader->ptrace))
 976                         __wake_up_parent(leader, leader->parent);
 977                 write_unlock_irq(&tasklist_lock);
 978                 threadgroup_change_end(tsk);
 979
 980                 release_task(leader);
 981         }
 982
 983         sig->group_exit_task = NULL;
 984         sig->notify_count = 0;
 985
 986 no_thread_group:
 987         /* we have changed execution domain */
 988         tsk->exit_signal = SIGCHLD;
 989
 990         exit_itimers(sig);
 991         flush_itimer_signals();
 992
 993         if (atomic_read(&oldsighand->count) != 1) {
 994                 struct sighand_struct *newsighand;
 995                 /*
 996                  * This ->sighand is shared with the CLONE_SIGHAND
 997                  * but not CLONE_THREAD task, switch to the new one.
 998                  */
 999                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1000                 if (!newsighand)
1001                         return -ENOMEM;
1002
1003                 atomic_set(&newsighand->count, 1);
1004                 memcpy(newsighand->action, oldsighand->action,
1005                        sizeof(newsighand->action));
1006
1007                 write_lock_irq(&tasklist_lock);
1008                 spin_lock(&oldsighand->siglock);
1009                 rcu_assign_pointer(tsk->sighand, newsighand);
1010                 spin_unlock(&oldsighand->siglock);
1011                 write_unlock_irq(&tasklist_lock);
1012
1013                 __cleanup_sighand(oldsighand);
1014         }
1015
1016         BUG_ON(!thread_group_leader(tsk));
1017         return 0;
1018
1019 killed:
1020         /* protects against exit_notify() and __exit_signal() */
1021         read_lock(&tasklist_lock);
1022         sig->group_exit_task = NULL;
1023         sig->notify_count = 0;
1024         read_unlock(&tasklist_lock);
1025         return -EAGAIN;
1026 }
1027
1028 char *get_task_comm(char *buf, struct task_struct *tsk)
1029 {
1030         /* buf must be at least sizeof(tsk->comm) in size */
1031         task_lock(tsk);
1032         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1033         task_unlock(tsk);
1034         return buf;
1035 }
1036 EXPORT_SYMBOL_GPL(get_task_comm);
1037
1038 /*
1039  * These functions flushes out all traces of the currently running executable
1040  * so that a new one can be started
1041  */
1042
1043 void set_task_comm(struct task_struct *tsk, char *buf)
1044 {
1045         task_lock(tsk);
1046         trace_task_rename(tsk, buf);
1047         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1048         task_unlock(tsk);
1049         perf_event_comm(tsk);
1050 }
1051
1052 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1053 {
1054         int i, ch;
1055
1056         /* Copies the binary name from after last slash */
1057         for (i = 0; (ch = *(fn++)) != '\0';) {
1058                 if (ch == '/')
1059                         i = 0; /* overwrite what we wrote */
1060                 else
1061                         if (i < len - 1)
1062                                 tcomm[i++] = ch;
1063         }
1064         tcomm[i] = '\0';
1065 }
1066
1067 int flush_old_exec(struct linux_binprm * bprm)
1068 {
1069         int retval;
1070
1071         /*
1072          * Make sure we have a private signal table and that
1073          * we are unassociated from the previous thread group.
1074          */
1075         retval = de_thread(current);
1076         if (retval)
1077                 goto out;
1078
1079         set_mm_exe_file(bprm->mm, bprm->file);
1080
1081         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1082         /*
1083          * Release all of the old mmap stuff
1084          */
1085         acct_arg_size(bprm, 0);
1086         retval = exec_mmap(bprm->mm);
1087         if (retval)
1088                 goto out;
1089
1090         bprm->mm = NULL;                /* We're using it now */
1091
1092         set_fs(USER_DS);
1093         current->flags &=
1094                 ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | PF_NOFREEZE);
1095         flush_thread();
1096         current->personality &= ~bprm->per_clear;
1097
1098         return 0;
1099
1100 out:
1101         return retval;
1102 }
1103 EXPORT_SYMBOL(flush_old_exec);
1104
1105 void would_dump(struct linux_binprm *bprm, struct file *file)
1106 {
1107         if (inode_permission(file_inode(file), MAY_READ) < 0)
1108                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1109 }
1110 EXPORT_SYMBOL(would_dump);
1111
1112 void setup_new_exec(struct linux_binprm * bprm)
1113 {
1114         arch_pick_mmap_layout(current->mm);
1115
1116         /* This is the point of no return */
1117         current->sas_ss_sp = current->sas_ss_size = 0;
1118
1119         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1120                 set_dumpable(current->mm, SUID_DUMP_USER);
1121         else
1122                 set_dumpable(current->mm, suid_dumpable);
1123
1124         set_task_comm(current, bprm->tcomm);
1125
1126         /* Set the new mm task size. We have to do that late because it may
1127          * depend on TIF_32BIT which is only updated in flush_thread() on
1128          * some architectures like powerpc
1129          */
1130         current->mm->task_size = TASK_SIZE;
1131
1132         /* install the new credentials */
1133         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1134             !gid_eq(bprm->cred->gid, current_egid())) {
1135                 current->pdeath_signal = 0;
1136         } else {
1137                 would_dump(bprm, bprm->file);
1138                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1139                         set_dumpable(current->mm, suid_dumpable);
1140         }
1141
1142         /* An exec changes our domain. We are no longer part of the thread
1143            group */
1144
1145         current->self_exec_id++;
1146
1147         flush_signal_handlers(current, 0);
1148         do_close_on_exec(current->files);
1149 }
1150 EXPORT_SYMBOL(setup_new_exec);
1151
1152 /*
1153  * Prepare credentials and lock ->cred_guard_mutex.
1154  * install_exec_creds() commits the new creds and drops the lock.
1155  * Or, if exec fails before, free_bprm() should release ->cred and
1156  * and unlock.
1157  */
1158 int prepare_bprm_creds(struct linux_binprm *bprm)
1159 {
1160         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1161                 return -ERESTARTNOINTR;
1162
1163         bprm->cred = prepare_exec_creds();
1164         if (likely(bprm->cred))
1165                 return 0;
1166
1167         mutex_unlock(&current->signal->cred_guard_mutex);
1168         return -ENOMEM;
1169 }
1170
1171 void free_bprm(struct linux_binprm *bprm)
1172 {
1173         free_arg_pages(bprm);
1174         if (bprm->cred) {
1175                 mutex_unlock(&current->signal->cred_guard_mutex);
1176                 abort_creds(bprm->cred);
1177         }
1178         /* If a binfmt changed the interp, free it. */
1179         if (bprm->interp != bprm->filename)
1180                 kfree(bprm->interp);
1181         kfree(bprm);
1182 }
1183
1184 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1185 {
1186         /* If a binfmt changed the interp, free it first. */
1187         if (bprm->interp != bprm->filename)
1188                 kfree(bprm->interp);
1189         bprm->interp = kstrdup(interp, GFP_KERNEL);
1190         if (!bprm->interp)
1191                 return -ENOMEM;
1192         return 0;
1193 }
1194 EXPORT_SYMBOL(bprm_change_interp);
1195
1196 /*
1197  * install the new credentials for this executable
1198  */
1199 void install_exec_creds(struct linux_binprm *bprm)
1200 {
1201         security_bprm_committing_creds(bprm);
1202
1203         commit_creds(bprm->cred);
1204         bprm->cred = NULL;
1205
1206         /*
1207          * Disable monitoring for regular users
1208          * when executing setuid binaries. Must
1209          * wait until new credentials are committed
1210          * by commit_creds() above
1211          */
1212         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1213                 perf_event_exit_task(current);
1214         /*
1215          * cred_guard_mutex must be held at least to this point to prevent
1216          * ptrace_attach() from altering our determination of the task's
1217          * credentials; any time after this it may be unlocked.
1218          */
1219         security_bprm_committed_creds(bprm);
1220         mutex_unlock(&current->signal->cred_guard_mutex);
1221 }
1222 EXPORT_SYMBOL(install_exec_creds);
1223
1224 /*
1225  * determine how safe it is to execute the proposed program
1226  * - the caller must hold ->cred_guard_mutex to protect against
1227  *   PTRACE_ATTACH
1228  */
1229 static int check_unsafe_exec(struct linux_binprm *bprm)
1230 {
1231         struct task_struct *p = current, *t;
1232         unsigned n_fs;
1233         int res = 0;
1234
1235         if (p->ptrace) {
1236                 if (p->ptrace & PT_PTRACE_CAP)
1237                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1238                 else
1239                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1240         }
1241
1242         /*
1243          * This isn't strictly necessary, but it makes it harder for LSMs to
1244          * mess up.
1245          */
1246         if (current->no_new_privs)
1247                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1248
1249         n_fs = 1;
1250         spin_lock(&p->fs->lock);
1251         rcu_read_lock();
1252         for (t = next_thread(p); t != p; t = next_thread(t)) {
1253                 if (t->fs == p->fs)
1254                         n_fs++;
1255         }
1256         rcu_read_unlock();
1257
1258         if (p->fs->users > n_fs) {
1259                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1260         } else {
1261                 res = -EAGAIN;
1262                 if (!p->fs->in_exec) {
1263                         p->fs->in_exec = 1;
1264                         res = 1;
1265                 }
1266         }
1267         spin_unlock(&p->fs->lock);
1268
1269         return res;
1270 }
1271
1272 /*
1273  * Fill the binprm structure from the inode.
1274  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1275  *
1276  * This may be called multiple times for binary chains (scripts for example).
1277  */
1278 int prepare_binprm(struct linux_binprm *bprm)
1279 {
1280         umode_t mode;
1281         struct inode * inode = file_inode(bprm->file);
1282         int retval;
1283
1284         mode = inode->i_mode;
1285         if (bprm->file->f_op == NULL)
1286                 return -EACCES;
1287
1288         /* clear any previous set[ug]id data from a previous binary */
1289         bprm->cred->euid = current_euid();
1290         bprm->cred->egid = current_egid();
1291
1292         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1293             !current->no_new_privs &&
1294             kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1295             kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1296                 /* Set-uid? */
1297                 if (mode & S_ISUID) {
1298                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1299                         bprm->cred->euid = inode->i_uid;
1300                 }
1301
1302                 /* Set-gid? */
1303                 /*
1304                  * If setgid is set but no group execute bit then this
1305                  * is a candidate for mandatory locking, not a setgid
1306                  * executable.
1307                  */
1308                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1309                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1310                         bprm->cred->egid = inode->i_gid;
1311                 }
1312         }
1313
1314         /* fill in binprm security blob */
1315         retval = security_bprm_set_creds(bprm);
1316         if (retval)
1317                 return retval;
1318         bprm->cred_prepared = 1;
1319
1320         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1321         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1322 }
1323
1324 EXPORT_SYMBOL(prepare_binprm);
1325
1326 /*
1327  * Arguments are '\0' separated strings found at the location bprm->p
1328  * points to; chop off the first by relocating brpm->p to right after
1329  * the first '\0' encountered.
1330  */
1331 int remove_arg_zero(struct linux_binprm *bprm)
1332 {
1333         int ret = 0;
1334         unsigned long offset;
1335         char *kaddr;
1336         struct page *page;
1337
1338         if (!bprm->argc)
1339                 return 0;
1340
1341         do {
1342                 offset = bprm->p & ~PAGE_MASK;
1343                 page = get_arg_page(bprm, bprm->p, 0);
1344                 if (!page) {
1345                         ret = -EFAULT;
1346                         goto out;
1347                 }
1348                 kaddr = kmap_atomic(page);
1349
1350                 for (; offset < PAGE_SIZE && kaddr[offset];
1351                                 offset++, bprm->p++)
1352                         ;
1353
1354                 kunmap_atomic(kaddr);
1355                 put_arg_page(page);
1356
1357                 if (offset == PAGE_SIZE)
1358                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1359         } while (offset == PAGE_SIZE);
1360
1361         bprm->p++;
1362         bprm->argc--;
1363         ret = 0;
1364
1365 out:
1366         return ret;
1367 }
1368 EXPORT_SYMBOL(remove_arg_zero);
1369
1370 /*
1371  * cycle the list of binary formats handler, until one recognizes the image
1372  */
1373 int search_binary_handler(struct linux_binprm *bprm)
1374 {
1375         int try, retval;
1376         struct linux_binfmt *fmt;
1377
1378         /* This allows 4 levels of binfmt rewrites before failing hard. */
1379         if (bprm->recursion_depth > 5)
1380                 return -ELOOP;
1381
1382         retval = security_bprm_check(bprm);
1383         if (retval)
1384                 return retval;
1385
1386         retval = audit_bprm(bprm);
1387         if (retval)
1388                 return retval;
1389
1390         retval = -ENOENT;
1391         for (try=0; try<2; try++) {
1392                 read_lock(&binfmt_lock);
1393                 list_for_each_entry(fmt, &formats, lh) {
1394                         if (!try_module_get(fmt->module))
1395                                 continue;
1396                         read_unlock(&binfmt_lock);
1397                         bprm->recursion_depth++;
1398                         retval = fmt->load_binary(bprm);
1399                         bprm->recursion_depth--;
1400                         if (retval >= 0) {
1401                                 put_binfmt(fmt);
1402                                 return retval;
1403                         }
1404                         read_lock(&binfmt_lock);
1405                         put_binfmt(fmt);
1406                         if (retval != -ENOEXEC || bprm->mm == NULL)
1407                                 break;
1408                         if (!bprm->file) {
1409                                 read_unlock(&binfmt_lock);
1410                                 return retval;
1411                         }
1412                 }
1413                 read_unlock(&binfmt_lock);
1414 #ifdef CONFIG_MODULES
1415                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1416                         break;
1417                 } else {
1418 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1419                         if (printable(bprm->buf[0]) &&
1420                             printable(bprm->buf[1]) &&
1421                             printable(bprm->buf[2]) &&
1422                             printable(bprm->buf[3]))
1423                                 break; /* -ENOEXEC */
1424                         if (try)
1425                                 break; /* -ENOEXEC */
1426                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1427                 }
1428 #else
1429                 break;
1430 #endif
1431         }
1432         return retval;
1433 }
1434 EXPORT_SYMBOL(search_binary_handler);
1435
1436 static int exec_binprm(struct linux_binprm *bprm)
1437 {
1438         pid_t old_pid, old_vpid;
1439         int ret;
1440
1441         /* Need to fetch pid before load_binary changes it */
1442         old_pid = current->pid;
1443         rcu_read_lock();
1444         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1445         rcu_read_unlock();
1446
1447         ret = search_binary_handler(bprm);
1448         if (ret >= 0) {
1449                 trace_sched_process_exec(current, old_pid, bprm);
1450                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1451                 current->did_exec = 1;
1452                 proc_exec_connector(current);
1453
1454                 if (bprm->file) {
1455                         allow_write_access(bprm->file);
1456                         fput(bprm->file);
1457                         bprm->file = NULL; /* to catch use-after-free */
1458                 }
1459         }
1460
1461         return ret;
1462 }
1463
1464 /*
1465  * sys_execve() executes a new program.
1466  */
1467 static int do_execve_common(const char *filename,
1468                                 struct user_arg_ptr argv,
1469                                 struct user_arg_ptr envp)
1470 {
1471         struct linux_binprm *bprm;
1472         struct file *file;
1473         struct files_struct *displaced;
1474         bool clear_in_exec;
1475         int retval;
1476
1477         /*
1478          * We move the actual failure in case of RLIMIT_NPROC excess from
1479          * set*uid() to execve() because too many poorly written programs
1480          * don't check setuid() return code.  Here we additionally recheck
1481          * whether NPROC limit is still exceeded.
1482          */
1483         if ((current->flags & PF_NPROC_EXCEEDED) &&
1484             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1485                 retval = -EAGAIN;
1486                 goto out_ret;
1487         }
1488
1489         /* We're below the limit (still or again), so we don't want to make
1490          * further execve() calls fail. */
1491         current->flags &= ~PF_NPROC_EXCEEDED;
1492
1493         retval = unshare_files(&displaced);
1494         if (retval)
1495                 goto out_ret;
1496
1497         retval = -ENOMEM;
1498         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1499         if (!bprm)
1500                 goto out_files;
1501
1502         retval = prepare_bprm_creds(bprm);
1503         if (retval)
1504                 goto out_free;
1505
1506         retval = check_unsafe_exec(bprm);
1507         if (retval < 0)
1508                 goto out_free;
1509         clear_in_exec = retval;
1510         current->in_execve = 1;
1511
1512         file = open_exec(filename);
1513         retval = PTR_ERR(file);
1514         if (IS_ERR(file))
1515                 goto out_unmark;
1516
1517         sched_exec();
1518
1519         bprm->file = file;
1520         bprm->filename = filename;
1521         bprm->interp = filename;
1522
1523         retval = bprm_mm_init(bprm);
1524         if (retval)
1525                 goto out_file;
1526
1527         bprm->argc = count(argv, MAX_ARG_STRINGS);
1528         if ((retval = bprm->argc) < 0)
1529                 goto out;
1530
1531         bprm->envc = count(envp, MAX_ARG_STRINGS);
1532         if ((retval = bprm->envc) < 0)
1533                 goto out;
1534
1535         retval = prepare_binprm(bprm);
1536         if (retval < 0)
1537                 goto out;
1538
1539         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1540         if (retval < 0)
1541                 goto out;
1542
1543         bprm->exec = bprm->p;
1544         retval = copy_strings(bprm->envc, envp, bprm);
1545         if (retval < 0)
1546                 goto out;
1547
1548         retval = copy_strings(bprm->argc, argv, bprm);
1549         if (retval < 0)
1550                 goto out;
1551
1552         retval = exec_binprm(bprm);
1553         if (retval < 0)
1554                 goto out;
1555
1556         /* execve succeeded */
1557         current->fs->in_exec = 0;
1558         current->in_execve = 0;
1559         acct_update_integrals(current);
1560         free_bprm(bprm);
1561         if (displaced)
1562                 put_files_struct(displaced);
1563         return retval;
1564
1565 out:
1566         if (bprm->mm) {
1567                 acct_arg_size(bprm, 0);
1568                 mmput(bprm->mm);
1569         }
1570
1571 out_file:
1572         if (bprm->file) {
1573                 allow_write_access(bprm->file);
1574                 fput(bprm->file);
1575         }
1576
1577 out_unmark:
1578         if (clear_in_exec)
1579                 current->fs->in_exec = 0;
1580         current->in_execve = 0;
1581
1582 out_free:
1583         free_bprm(bprm);
1584
1585 out_files:
1586         if (displaced)
1587                 reset_files_struct(displaced);
1588 out_ret:
1589         return retval;
1590 }
1591
1592 int do_execve(const char *filename,
1593         const char __user *const __user *__argv,
1594         const char __user *const __user *__envp)
1595 {
1596         struct user_arg_ptr argv = { .ptr.native = __argv };
1597         struct user_arg_ptr envp = { .ptr.native = __envp };
1598         return do_execve_common(filename, argv, envp);
1599 }
1600
1601 #ifdef CONFIG_COMPAT
1602 static int compat_do_execve(const char *filename,
1603         const compat_uptr_t __user *__argv,
1604         const compat_uptr_t __user *__envp)
1605 {
1606         struct user_arg_ptr argv = {
1607                 .is_compat = true,
1608                 .ptr.compat = __argv,
1609         };
1610         struct user_arg_ptr envp = {
1611                 .is_compat = true,
1612                 .ptr.compat = __envp,
1613         };
1614         return do_execve_common(filename, argv, envp);
1615 }
1616 #endif
1617
1618 void set_binfmt(struct linux_binfmt *new)
1619 {
1620         struct mm_struct *mm = current->mm;
1621
1622         if (mm->binfmt)
1623                 module_put(mm->binfmt->module);
1624
1625         mm->binfmt = new;
1626         if (new)
1627                 __module_get(new->module);
1628 }
1629
1630 EXPORT_SYMBOL(set_binfmt);
1631
1632 /*
1633  * set_dumpable converts traditional three-value dumpable to two flags and
1634  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1635  * these bits are not changed atomically.  So get_dumpable can observe the
1636  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1637  * return either old dumpable or new one by paying attention to the order of
1638  * modifying the bits.
1639  *
1640  * dumpable |   mm->flags (binary)
1641  * old  new | initial interim  final
1642  * ---------+-----------------------
1643  *  0    1  |   00      01      01
1644  *  0    2  |   00      10(*)   11
1645  *  1    0  |   01      00      00
1646  *  1    2  |   01      11      11
1647  *  2    0  |   11      10(*)   00
1648  *  2    1  |   11      11      01
1649  *
1650  * (*) get_dumpable regards interim value of 10 as 11.
1651  */
1652 void set_dumpable(struct mm_struct *mm, int value)
1653 {
1654         switch (value) {
1655         case SUID_DUMP_DISABLE:
1656                 clear_bit(MMF_DUMPABLE, &mm->flags);
1657                 smp_wmb();
1658                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1659                 break;
1660         case SUID_DUMP_USER:
1661                 set_bit(MMF_DUMPABLE, &mm->flags);
1662                 smp_wmb();
1663                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1664                 break;
1665         case SUID_DUMP_ROOT:
1666                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1667                 smp_wmb();
1668                 set_bit(MMF_DUMPABLE, &mm->flags);
1669                 break;
1670         }
1671 }
1672
1673 int __get_dumpable(unsigned long mm_flags)
1674 {
1675         int ret;
1676
1677         ret = mm_flags & MMF_DUMPABLE_MASK;
1678         return (ret > SUID_DUMP_USER) ? SUID_DUMP_ROOT : ret;
1679 }
1680
1681 int get_dumpable(struct mm_struct *mm)
1682 {
1683         return __get_dumpable(mm->flags);
1684 }
1685
1686 SYSCALL_DEFINE3(execve,
1687                 const char __user *, filename,
1688                 const char __user *const __user *, argv,
1689                 const char __user *const __user *, envp)
1690 {
1691         struct filename *path = getname(filename);
1692         int error = PTR_ERR(path);
1693         if (!IS_ERR(path)) {
1694                 error = do_execve(path->name, argv, envp);
1695                 putname(path);
1696         }
1697         return error;
1698 }
1699 #ifdef CONFIG_COMPAT
1700 asmlinkage long compat_sys_execve(const char __user * filename,
1701         const compat_uptr_t __user * argv,
1702         const compat_uptr_t __user * envp)
1703 {
1704         struct filename *path = getname(filename);
1705         int error = PTR_ERR(path);
1706         if (!IS_ERR(path)) {
1707                 error = compat_do_execve(path->name, argv, envp);
1708                 putname(path);
1709         }
1710         return error;
1711 }
1712 #endif