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 #include <asm/exec.h>
  63
  64 #include <trace/events/task.h>
  65 #include "internal.h"
  66 #include "coredump.h"
  67
  68 #include <trace/events/sched.h>
  69
  70 int suid_dumpable = 0;
  71
  72 static LIST_HEAD(formats);
  73 static DEFINE_RWLOCK(binfmt_lock);
  74
  75 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  76 {
  77         BUG_ON(!fmt);
  78         write_lock(&binfmt_lock);
  79         insert ? list_add(&fmt->lh, &formats) :
  80                  list_add_tail(&fmt->lh, &formats);
  81         write_unlock(&binfmt_lock);
  82 }
  83
  84 EXPORT_SYMBOL(__register_binfmt);
  85
  86 void unregister_binfmt(struct linux_binfmt * fmt)
  87 {
  88         write_lock(&binfmt_lock);
  89         list_del(&fmt->lh);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(unregister_binfmt);
  94
  95 static inline void put_binfmt(struct linux_binfmt * fmt)
  96 {
  97         module_put(fmt->module);
  98 }
  99
 100 /*
 101  * Note that a shared library must be both readable and executable due to
 102  * security reasons.
 103  *
 104  * Also note that we take the address to load from from the file itself.
 105  */
 106 SYSCALL_DEFINE1(uselib, const char __user *, library)
 107 {
 108         struct file *file;
 109         char *tmp = getname(library);
 110         int error = PTR_ERR(tmp);
 111         static const struct open_flags uselib_flags = {
 112                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 113                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 114                 .intent = LOOKUP_OPEN
 115         };
 116
 117         if (IS_ERR(tmp))
 118                 goto out;
 119
 120         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 121         putname(tmp);
 122         error = PTR_ERR(file);
 123         if (IS_ERR(file))
 124                 goto out;
 125
 126         error = -EINVAL;
 127         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 128                 goto exit;
 129
 130         error = -EACCES;
 131         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 132                 goto exit;
 133
 134         fsnotify_open(file);
 135
 136         error = -ENOEXEC;
 137         if(file->f_op) {
 138                 struct linux_binfmt * fmt;
 139
 140                 read_lock(&binfmt_lock);
 141                 list_for_each_entry(fmt, &formats, lh) {
 142                         if (!fmt->load_shlib)
 143                                 continue;
 144                         if (!try_module_get(fmt->module))
 145                                 continue;
 146                         read_unlock(&binfmt_lock);
 147                         error = fmt->load_shlib(file);
 148                         read_lock(&binfmt_lock);
 149                         put_binfmt(fmt);
 150                         if (error != -ENOEXEC)
 151                                 break;
 152                 }
 153                 read_unlock(&binfmt_lock);
 154         }
 155 exit:
 156         fput(file);
 157 out:
 158         return error;
 159 }
 160
 161 #ifdef CONFIG_MMU
 162 /*
 163  * The nascent bprm->mm is not visible until exec_mmap() but it can
 164  * use a lot of memory, account these pages in current->mm temporary
 165  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 166  * change the counter back via acct_arg_size(0).
 167  */
 168 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 169 {
 170         struct mm_struct *mm = current->mm;
 171         long diff = (long)(pages - bprm->vma_pages);
 172
 173         if (!mm || !diff)
 174                 return;
 175
 176         bprm->vma_pages = pages;
 177         add_mm_counter(mm, MM_ANONPAGES, diff);
 178 }
 179
 180 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 181                 int write)
 182 {
 183         struct page *page;
 184         int ret;
 185
 186 #ifdef CONFIG_STACK_GROWSUP
 187         if (write) {
 188                 ret = expand_downwards(bprm->vma, pos);
 189                 if (ret < 0)
 190                         return NULL;
 191         }
 192 #endif
 193         ret = get_user_pages(current, bprm->mm, pos,
 194                         1, write, 1, &page, NULL);
 195         if (ret <= 0)
 196                 return NULL;
 197
 198         if (write) {
 199                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 200                 struct rlimit *rlim;
 201
 202                 acct_arg_size(bprm, size / PAGE_SIZE);
 203
 204                 /*
 205                  * We've historically supported up to 32 pages (ARG_MAX)
 206                  * of argument strings even with small stacks
 207                  */
 208                 if (size <= ARG_MAX)
 209                         return page;
 210
 211                 /*
 212                  * Limit to 1/4-th the stack size for the argv+env strings.
 213                  * This ensures that:
 214                  *  - the remaining binfmt code will not run out of stack space,
 215                  *  - the program will have a reasonable amount of stack left
 216                  *    to work from.
 217                  */
 218                 rlim = current->signal->rlim;
 219                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 220                         put_page(page);
 221                         return NULL;
 222                 }
 223         }
 224
 225         return page;
 226 }
 227
 228 static void put_arg_page(struct page *page)
 229 {
 230         put_page(page);
 231 }
 232
 233 static void free_arg_page(struct linux_binprm *bprm, int i)
 234 {
 235 }
 236
 237 static void free_arg_pages(struct linux_binprm *bprm)
 238 {
 239 }
 240
 241 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 242                 struct page *page)
 243 {
 244         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 245 }
 246
 247 static int __bprm_mm_init(struct linux_binprm *bprm)
 248 {
 249         int err;
 250         struct vm_area_struct *vma = NULL;
 251         struct mm_struct *mm = bprm->mm;
 252
 253         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 254         if (!vma)
 255                 return -ENOMEM;
 256
 257         down_write(&mm->mmap_sem);
 258         vma->vm_mm = mm;
 259
 260         /*
 261          * Place the stack at the largest stack address the architecture
 262          * supports. Later, we'll move this to an appropriate place. We don't
 263          * use STACK_TOP because that can depend on attributes which aren't
 264          * configured yet.
 265          */
 266         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 267         vma->vm_end = STACK_TOP_MAX;
 268         vma->vm_start = vma->vm_end - PAGE_SIZE;
 269         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 270         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 271         INIT_LIST_HEAD(&vma->anon_vma_chain);
 272
 273         err = insert_vm_struct(mm, vma);
 274         if (err)
 275                 goto err;
 276
 277         mm->stack_vm = mm->total_vm = 1;
 278         up_write(&mm->mmap_sem);
 279         bprm->p = vma->vm_end - sizeof(void *);
 280         return 0;
 281 err:
 282         up_write(&mm->mmap_sem);
 283         bprm->vma = NULL;
 284         kmem_cache_free(vm_area_cachep, vma);
 285         return err;
 286 }
 287
 288 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 289 {
 290         return len <= MAX_ARG_STRLEN;
 291 }
 292
 293 #else
 294
 295 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 296 {
 297 }
 298
 299 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 300                 int write)
 301 {
 302         struct page *page;
 303
 304         page = bprm->page[pos / PAGE_SIZE];
 305         if (!page && write) {
 306                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 307                 if (!page)
 308                         return NULL;
 309                 bprm->page[pos / PAGE_SIZE] = page;
 310         }
 311
 312         return page;
 313 }
 314
 315 static void put_arg_page(struct page *page)
 316 {
 317 }
 318
 319 static void free_arg_page(struct linux_binprm *bprm, int i)
 320 {
 321         if (bprm->page[i]) {
 322                 __free_page(bprm->page[i]);
 323                 bprm->page[i] = NULL;
 324         }
 325 }
 326
 327 static void free_arg_pages(struct linux_binprm *bprm)
 328 {
 329         int i;
 330
 331         for (i = 0; i < MAX_ARG_PAGES; i++)
 332                 free_arg_page(bprm, i);
 333 }
 334
 335 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 336                 struct page *page)
 337 {
 338 }
 339
 340 static int __bprm_mm_init(struct linux_binprm *bprm)
 341 {
 342         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 343         return 0;
 344 }
 345
 346 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 347 {
 348         return len <= bprm->p;
 349 }
 350
 351 #endif /* CONFIG_MMU */
 352
 353 /*
 354  * Create a new mm_struct and populate it with a temporary stack
 355  * vm_area_struct.  We don't have enough context at this point to set the stack
 356  * flags, permissions, and offset, so we use temporary values.  We'll update
 357  * them later in setup_arg_pages().
 358  */
 359 int bprm_mm_init(struct linux_binprm *bprm)
 360 {
 361         int err;
 362         struct mm_struct *mm = NULL;
 363
 364         bprm->mm = mm = mm_alloc();
 365         err = -ENOMEM;
 366         if (!mm)
 367                 goto err;
 368
 369         err = init_new_context(current, mm);
 370         if (err)
 371                 goto err;
 372
 373         err = __bprm_mm_init(bprm);
 374         if (err)
 375                 goto err;
 376
 377         return 0;
 378
 379 err:
 380         if (mm) {
 381                 bprm->mm = NULL;
 382                 mmdrop(mm);
 383         }
 384
 385         return err;
 386 }
 387
 388 struct user_arg_ptr {
 389 #ifdef CONFIG_COMPAT
 390         bool is_compat;
 391 #endif
 392         union {
 393                 const char __user *const __user *native;
 394 #ifdef CONFIG_COMPAT
 395                 compat_uptr_t __user *compat;
 396 #endif
 397         } ptr;
 398 };
 399
 400 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 401 {
 402         const char __user *native;
 403
 404 #ifdef CONFIG_COMPAT
 405         if (unlikely(argv.is_compat)) {
 406                 compat_uptr_t compat;
 407
 408                 if (get_user(compat, argv.ptr.compat + nr))
 409                         return ERR_PTR(-EFAULT);
 410
 411                 return compat_ptr(compat);
 412         }
 413 #endif
 414
 415         if (get_user(native, argv.ptr.native + nr))
 416                 return ERR_PTR(-EFAULT);
 417
 418         return native;
 419 }
 420
 421 /*
 422  * count() counts the number of strings in array ARGV.
 423  */
 424 static int count(struct user_arg_ptr argv, int max)
 425 {
 426         int i = 0;
 427
 428         if (argv.ptr.native != NULL) {
 429                 for (;;) {
 430                         const char __user *p = get_user_arg_ptr(argv, i);
 431
 432                         if (!p)
 433                                 break;
 434
 435                         if (IS_ERR(p))
 436                                 return -EFAULT;
 437
 438                         if (i++ >= max)
 439                                 return -E2BIG;
 440
 441                         if (fatal_signal_pending(current))
 442                                 return -ERESTARTNOHAND;
 443                         cond_resched();
 444                 }
 445         }
 446         return i;
 447 }
 448
 449 /*
 450  * 'copy_strings()' copies argument/environment strings from the old
 451  * processes's memory to the new process's stack.  The call to get_user_pages()
 452  * ensures the destination page is created and not swapped out.
 453  */
 454 static int copy_strings(int argc, struct user_arg_ptr argv,
 455                         struct linux_binprm *bprm)
 456 {
 457         struct page *kmapped_page = NULL;
 458         char *kaddr = NULL;
 459         unsigned long kpos = 0;
 460         int ret;
 461
 462         while (argc-- > 0) {
 463                 const char __user *str;
 464                 int len;
 465                 unsigned long pos;
 466
 467                 ret = -EFAULT;
 468                 str = get_user_arg_ptr(argv, argc);
 469                 if (IS_ERR(str))
 470                         goto out;
 471
 472                 len = strnlen_user(str, MAX_ARG_STRLEN);
 473                 if (!len)
 474                         goto out;
 475
 476                 ret = -E2BIG;
 477                 if (!valid_arg_len(bprm, len))
 478                         goto out;
 479
 480                 /* We're going to work our way backwords. */
 481                 pos = bprm->p;
 482                 str += len;
 483                 bprm->p -= len;
 484
 485                 while (len > 0) {
 486                         int offset, bytes_to_copy;
 487
 488                         if (fatal_signal_pending(current)) {
 489                                 ret = -ERESTARTNOHAND;
 490                                 goto out;
 491                         }
 492                         cond_resched();
 493
 494                         offset = pos % PAGE_SIZE;
 495                         if (offset == 0)
 496                                 offset = PAGE_SIZE;
 497
 498                         bytes_to_copy = offset;
 499                         if (bytes_to_copy > len)
 500                                 bytes_to_copy = len;
 501
 502                         offset -= bytes_to_copy;
 503                         pos -= bytes_to_copy;
 504                         str -= bytes_to_copy;
 505                         len -= bytes_to_copy;
 506
 507                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 508                                 struct page *page;
 509
 510                                 page = get_arg_page(bprm, pos, 1);
 511                                 if (!page) {
 512                                         ret = -E2BIG;
 513                                         goto out;
 514                                 }
 515
 516                                 if (kmapped_page) {
 517                                         flush_kernel_dcache_page(kmapped_page);
 518                                         kunmap(kmapped_page);
 519                                         put_arg_page(kmapped_page);
 520                                 }
 521                                 kmapped_page = page;
 522                                 kaddr = kmap(kmapped_page);
 523                                 kpos = pos & PAGE_MASK;
 524                                 flush_arg_page(bprm, kpos, kmapped_page);
 525                         }
 526                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 527                                 ret = -EFAULT;
 528                                 goto out;
 529                         }
 530                 }
 531         }
 532         ret = 0;
 533 out:
 534         if (kmapped_page) {
 535                 flush_kernel_dcache_page(kmapped_page);
 536                 kunmap(kmapped_page);
 537                 put_arg_page(kmapped_page);
 538         }
 539         return ret;
 540 }
 541
 542 /*
 543  * Like copy_strings, but get argv and its values from kernel memory.
 544  */
 545 int copy_strings_kernel(int argc, const char *const *__argv,
 546                         struct linux_binprm *bprm)
 547 {
 548         int r;
 549         mm_segment_t oldfs = get_fs();
 550         struct user_arg_ptr argv = {
 551                 .ptr.native = (const char __user *const  __user *)__argv,
 552         };
 553
 554         set_fs(KERNEL_DS);
 555         r = copy_strings(argc, argv, bprm);
 556         set_fs(oldfs);
 557
 558         return r;
 559 }
 560 EXPORT_SYMBOL(copy_strings_kernel);
 561
 562 #ifdef CONFIG_MMU
 563
 564 /*
 565  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 566  * the binfmt code determines where the new stack should reside, we shift it to
 567  * its final location.  The process proceeds as follows:
 568  *
 569  * 1) Use shift to calculate the new vma endpoints.
 570  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 571  *    arguments passed to subsequent functions are consistent.
 572  * 3) Move vma's page tables to the new range.
 573  * 4) Free up any cleared pgd range.
 574  * 5) Shrink the vma to cover only the new range.
 575  */
 576 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 577 {
 578         struct mm_struct *mm = vma->vm_mm;
 579         unsigned long old_start = vma->vm_start;
 580         unsigned long old_end = vma->vm_end;
 581         unsigned long length = old_end - old_start;
 582         unsigned long new_start = old_start - shift;
 583         unsigned long new_end = old_end - shift;
 584         struct mmu_gather tlb;
 585
 586         BUG_ON(new_start > new_end);
 587
 588         /*
 589          * ensure there are no vmas between where we want to go
 590          * and where we are
 591          */
 592         if (vma != find_vma(mm, new_start))
 593                 return -EFAULT;
 594
 595         /*
 596          * cover the whole range: [new_start, old_end)
 597          */
 598         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 599                 return -ENOMEM;
 600
 601         /*
 602          * move the page tables downwards, on failure we rely on
 603          * process cleanup to remove whatever mess we made.
 604          */
 605         if (length != move_page_tables(vma, old_start,
 606                                        vma, new_start, length))
 607                 return -ENOMEM;
 608
 609         lru_add_drain();
 610         tlb_gather_mmu(&tlb, mm, 0);
 611         if (new_end > old_start) {
 612                 /*
 613                  * when the old and new regions overlap clear from new_end.
 614                  */
 615                 free_pgd_range(&tlb, new_end, old_end, new_end,
 616                         vma->vm_next ? vma->vm_next->vm_start : 0);
 617         } else {
 618                 /*
 619                  * otherwise, clean from old_start; this is done to not touch
 620                  * the address space in [new_end, old_start) some architectures
 621                  * have constraints on va-space that make this illegal (IA64) -
 622                  * for the others its just a little faster.
 623                  */
 624                 free_pgd_range(&tlb, old_start, old_end, new_end,
 625                         vma->vm_next ? vma->vm_next->vm_start : 0);
 626         }
 627         tlb_finish_mmu(&tlb, new_end, old_end);
 628
 629         /*
 630          * Shrink the vma to just the new range.  Always succeeds.
 631          */
 632         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 633
 634         return 0;
 635 }
 636
 637 /*
 638  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 639  * the stack is optionally relocated, and some extra space is added.
 640  */
 641 int setup_arg_pages(struct linux_binprm *bprm,
 642                     unsigned long stack_top,
 643                     int executable_stack)
 644 {
 645         unsigned long ret;
 646         unsigned long stack_shift;
 647         struct mm_struct *mm = current->mm;
 648         struct vm_area_struct *vma = bprm->vma;
 649         struct vm_area_struct *prev = NULL;
 650         unsigned long vm_flags;
 651         unsigned long stack_base;
 652         unsigned long stack_size;
 653         unsigned long stack_expand;
 654         unsigned long rlim_stack;
 655
 656 #ifdef CONFIG_STACK_GROWSUP
 657         /* Limit stack size to 1GB */
 658         stack_base = rlimit_max(RLIMIT_STACK);
 659         if (stack_base > (1 << 30))
 660                 stack_base = 1 << 30;
 661
 662         /* Make sure we didn't let the argument array grow too large. */
 663         if (vma->vm_end - vma->vm_start > stack_base)
 664                 return -ENOMEM;
 665
 666         stack_base = PAGE_ALIGN(stack_top - stack_base);
 667
 668         stack_shift = vma->vm_start - stack_base;
 669         mm->arg_start = bprm->p - stack_shift;
 670         bprm->p = vma->vm_end - stack_shift;
 671 #else
 672         stack_top = arch_align_stack(stack_top);
 673         stack_top = PAGE_ALIGN(stack_top);
 674
 675         if (unlikely(stack_top < mmap_min_addr) ||
 676             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 677                 return -ENOMEM;
 678
 679         stack_shift = vma->vm_end - stack_top;
 680
 681         bprm->p -= stack_shift;
 682         mm->arg_start = bprm->p;
 683 #endif
 684
 685         if (bprm->loader)
 686                 bprm->loader -= stack_shift;
 687         bprm->exec -= stack_shift;
 688
 689         down_write(&mm->mmap_sem);
 690         vm_flags = VM_STACK_FLAGS;
 691
 692         /*
 693          * Adjust stack execute permissions; explicitly enable for
 694          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 695          * (arch default) otherwise.
 696          */
 697         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 698                 vm_flags |= VM_EXEC;
 699         else if (executable_stack == EXSTACK_DISABLE_X)
 700                 vm_flags &= ~VM_EXEC;
 701         vm_flags |= mm->def_flags;
 702         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 703
 704         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 705                         vm_flags);
 706         if (ret)
 707                 goto out_unlock;
 708         BUG_ON(prev != vma);
 709
 710         /* Move stack pages down in memory. */
 711         if (stack_shift) {
 712                 ret = shift_arg_pages(vma, stack_shift);
 713                 if (ret)
 714                         goto out_unlock;
 715         }
 716
 717         /* mprotect_fixup is overkill to remove the temporary stack flags */
 718         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 719
 720         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 721         stack_size = vma->vm_end - vma->vm_start;
 722         /*
 723          * Align this down to a page boundary as expand_stack
 724          * will align it up.
 725          */
 726         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 727 #ifdef CONFIG_STACK_GROWSUP
 728         if (stack_size + stack_expand > rlim_stack)
 729                 stack_base = vma->vm_start + rlim_stack;
 730         else
 731                 stack_base = vma->vm_end + stack_expand;
 732 #else
 733         if (stack_size + stack_expand > rlim_stack)
 734                 stack_base = vma->vm_end - rlim_stack;
 735         else
 736                 stack_base = vma->vm_start - stack_expand;
 737 #endif
 738         current->mm->start_stack = bprm->p;
 739         ret = expand_stack(vma, stack_base);
 740         if (ret)
 741                 ret = -EFAULT;
 742
 743 out_unlock:
 744         up_write(&mm->mmap_sem);
 745         return ret;
 746 }
 747 EXPORT_SYMBOL(setup_arg_pages);
 748
 749 #endif /* CONFIG_MMU */
 750
 751 struct file *open_exec(const char *name)
 752 {
 753         struct file *file;
 754         int err;
 755         static const struct open_flags open_exec_flags = {
 756                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 757                 .acc_mode = MAY_EXEC | MAY_OPEN,
 758                 .intent = LOOKUP_OPEN
 759         };
 760
 761         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
 762         if (IS_ERR(file))
 763                 goto out;
 764
 765         err = -EACCES;
 766         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
 767                 goto exit;
 768
 769         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 770                 goto exit;
 771
 772         fsnotify_open(file);
 773
 774         err = deny_write_access(file);
 775         if (err)
 776                 goto exit;
 777
 778 out:
 779         return file;
 780
 781 exit:
 782         fput(file);
 783         return ERR_PTR(err);
 784 }
 785 EXPORT_SYMBOL(open_exec);
 786
 787 int kernel_read(struct file *file, loff_t offset,
 788                 char *addr, unsigned long count)
 789 {
 790         mm_segment_t old_fs;
 791         loff_t pos = offset;
 792         int result;
 793
 794         old_fs = get_fs();
 795         set_fs(get_ds());
 796         /* The cast to a user pointer is valid due to the set_fs() */
 797         result = vfs_read(file, (void __user *)addr, count, &pos);
 798         set_fs(old_fs);
 799         return result;
 800 }
 801
 802 EXPORT_SYMBOL(kernel_read);
 803
 804 static int exec_mmap(struct mm_struct *mm)
 805 {
 806         struct task_struct *tsk;
 807         struct mm_struct * old_mm, *active_mm;
 808
 809         /* Notify parent that we're no longer interested in the old VM */
 810         tsk = current;
 811         old_mm = current->mm;
 812         mm_release(tsk, old_mm);
 813
 814         if (old_mm) {
 815                 sync_mm_rss(old_mm);
 816                 /*
 817                  * Make sure that if there is a core dump in progress
 818                  * for the old mm, we get out and die instead of going
 819                  * through with the exec.  We must hold mmap_sem around
 820                  * checking core_state and changing tsk->mm.
 821                  */
 822                 down_read(&old_mm->mmap_sem);
 823                 if (unlikely(old_mm->core_state)) {
 824                         up_read(&old_mm->mmap_sem);
 825                         return -EINTR;
 826                 }
 827         }
 828         task_lock(tsk);
 829         active_mm = tsk->active_mm;
 830         tsk->mm = mm;
 831         tsk->active_mm = mm;
 832         activate_mm(active_mm, mm);
 833         task_unlock(tsk);
 834         arch_pick_mmap_layout(mm);
 835         if (old_mm) {
 836                 up_read(&old_mm->mmap_sem);
 837                 BUG_ON(active_mm != old_mm);
 838                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 839                 mm_update_next_owner(old_mm);
 840                 mmput(old_mm);
 841                 return 0;
 842         }
 843         mmdrop(active_mm);
 844         return 0;
 845 }
 846
 847 /*
 848  * This function makes sure the current process has its own signal table,
 849  * so that flush_signal_handlers can later reset the handlers without
 850  * disturbing other processes.  (Other processes might share the signal
 851  * table via the CLONE_SIGHAND option to clone().)
 852  */
 853 static int de_thread(struct task_struct *tsk)
 854 {
 855         struct signal_struct *sig = tsk->signal;
 856         struct sighand_struct *oldsighand = tsk->sighand;
 857         spinlock_t *lock = &oldsighand->siglock;
 858
 859         if (thread_group_empty(tsk))
 860                 goto no_thread_group;
 861
 862         /*
 863          * Kill all other threads in the thread group.
 864          */
 865         spin_lock_irq(lock);
 866         if (signal_group_exit(sig)) {
 867                 /*
 868                  * Another group action in progress, just
 869                  * return so that the signal is processed.
 870                  */
 871                 spin_unlock_irq(lock);
 872                 return -EAGAIN;
 873         }
 874
 875         sig->group_exit_task = tsk;
 876         sig->notify_count = zap_other_threads(tsk);
 877         if (!thread_group_leader(tsk))
 878                 sig->notify_count--;
 879
 880         while (sig->notify_count) {
 881                 __set_current_state(TASK_UNINTERRUPTIBLE);
 882                 spin_unlock_irq(lock);
 883                 schedule();
 884                 spin_lock_irq(lock);
 885         }
 886         spin_unlock_irq(lock);
 887
 888         /*
 889          * At this point all other threads have exited, all we have to
 890          * do is to wait for the thread group leader to become inactive,
 891          * and to assume its PID:
 892          */
 893         if (!thread_group_leader(tsk)) {
 894                 struct task_struct *leader = tsk->group_leader;
 895
 896                 sig->notify_count = -1; /* for exit_notify() */
 897                 for (;;) {
 898                         write_lock_irq(&tasklist_lock);
 899                         if (likely(leader->exit_state))
 900                                 break;
 901                         __set_current_state(TASK_UNINTERRUPTIBLE);
 902                         write_unlock_irq(&tasklist_lock);
 903                         schedule();
 904                 }
 905
 906                 /*
 907                  * The only record we have of the real-time age of a
 908                  * process, regardless of execs it's done, is start_time.
 909                  * All the past CPU time is accumulated in signal_struct
 910                  * from sister threads now dead.  But in this non-leader
 911                  * exec, nothing survives from the original leader thread,
 912                  * whose birth marks the true age of this process now.
 913                  * When we take on its identity by switching to its PID, we
 914                  * also take its birthdate (always earlier than our own).
 915                  */
 916                 tsk->start_time = leader->start_time;
 917
 918                 BUG_ON(!same_thread_group(leader, tsk));
 919                 BUG_ON(has_group_leader_pid(tsk));
 920                 /*
 921                  * An exec() starts a new thread group with the
 922                  * TGID of the previous thread group. Rehash the
 923                  * two threads with a switched PID, and release
 924                  * the former thread group leader:
 925                  */
 926
 927                 /* Become a process group leader with the old leader's pid.
 928                  * The old leader becomes a thread of the this thread group.
 929                  * Note: The old leader also uses this pid until release_task
 930                  *       is called.  Odd but simple and correct.
 931                  */
 932                 detach_pid(tsk, PIDTYPE_PID);
 933                 tsk->pid = leader->pid;
 934                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 935                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 936                 transfer_pid(leader, tsk, PIDTYPE_SID);
 937
 938                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 939                 list_replace_init(&leader->sibling, &tsk->sibling);
 940
 941                 tsk->group_leader = tsk;
 942                 leader->group_leader = tsk;
 943
 944                 tsk->exit_signal = SIGCHLD;
 945                 leader->exit_signal = -1;
 946
 947                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 948                 leader->exit_state = EXIT_DEAD;
 949
 950                 /*
 951                  * We are going to release_task()->ptrace_unlink() silently,
 952                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 953                  * the tracer wont't block again waiting for this thread.
 954                  */
 955                 if (unlikely(leader->ptrace))
 956                         __wake_up_parent(leader, leader->parent);
 957                 write_unlock_irq(&tasklist_lock);
 958
 959                 release_task(leader);
 960         }
 961
 962         sig->group_exit_task = NULL;
 963         sig->notify_count = 0;
 964
 965 no_thread_group:
 966         /* we have changed execution domain */
 967         tsk->exit_signal = SIGCHLD;
 968
 969         exit_itimers(sig);
 970         flush_itimer_signals();
 971
 972         if (atomic_read(&oldsighand->count) != 1) {
 973                 struct sighand_struct *newsighand;
 974                 /*
 975                  * This ->sighand is shared with the CLONE_SIGHAND
 976                  * but not CLONE_THREAD task, switch to the new one.
 977                  */
 978                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 979                 if (!newsighand)
 980                         return -ENOMEM;
 981
 982                 atomic_set(&newsighand->count, 1);
 983                 memcpy(newsighand->action, oldsighand->action,
 984                        sizeof(newsighand->action));
 985
 986                 write_lock_irq(&tasklist_lock);
 987                 spin_lock(&oldsighand->siglock);
 988                 rcu_assign_pointer(tsk->sighand, newsighand);
 989                 spin_unlock(&oldsighand->siglock);
 990                 write_unlock_irq(&tasklist_lock);
 991
 992                 __cleanup_sighand(oldsighand);
 993         }
 994
 995         BUG_ON(!thread_group_leader(tsk));
 996         return 0;
 997 }
 998
 999 char *get_task_comm(char *buf, struct task_struct *tsk)
1000 {
1001         /* buf must be at least sizeof(tsk->comm) in size */
1002         task_lock(tsk);
1003         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1004         task_unlock(tsk);
1005         return buf;
1006 }
1007 EXPORT_SYMBOL_GPL(get_task_comm);
1008
1009 /*
1010  * These functions flushes out all traces of the currently running executable
1011  * so that a new one can be started
1012  */
1013
1014 void set_task_comm(struct task_struct *tsk, char *buf)
1015 {
1016         task_lock(tsk);
1017
1018         trace_task_rename(tsk, buf);
1019
1020         /*
1021          * Threads may access current->comm without holding
1022          * the task lock, so write the string carefully.
1023          * Readers without a lock may see incomplete new
1024          * names but are safe from non-terminating string reads.
1025          */
1026         memset(tsk->comm, 0, TASK_COMM_LEN);
1027         wmb();
1028         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1029         task_unlock(tsk);
1030         perf_event_comm(tsk);
1031 }
1032
1033 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1034 {
1035         int i, ch;
1036
1037         /* Copies the binary name from after last slash */
1038         for (i = 0; (ch = *(fn++)) != '\0';) {
1039                 if (ch == '/')
1040                         i = 0; /* overwrite what we wrote */
1041                 else
1042                         if (i < len - 1)
1043                                 tcomm[i++] = ch;
1044         }
1045         tcomm[i] = '\0';
1046 }
1047
1048 int flush_old_exec(struct linux_binprm * bprm)
1049 {
1050         int retval;
1051
1052         /*
1053          * Make sure we have a private signal table and that
1054          * we are unassociated from the previous thread group.
1055          */
1056         retval = de_thread(current);
1057         if (retval)
1058                 goto out;
1059
1060         set_mm_exe_file(bprm->mm, bprm->file);
1061
1062         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1063         /*
1064          * Release all of the old mmap stuff
1065          */
1066         acct_arg_size(bprm, 0);
1067         retval = exec_mmap(bprm->mm);
1068         if (retval)
1069                 goto out;
1070
1071         bprm->mm = NULL;                /* We're using it now */
1072
1073         set_fs(USER_DS);
1074         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD);
1075         flush_thread();
1076         current->personality &= ~bprm->per_clear;
1077
1078         return 0;
1079
1080 out:
1081         return retval;
1082 }
1083 EXPORT_SYMBOL(flush_old_exec);
1084
1085 void would_dump(struct linux_binprm *bprm, struct file *file)
1086 {
1087         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1088                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1089 }
1090 EXPORT_SYMBOL(would_dump);
1091
1092 void setup_new_exec(struct linux_binprm * bprm)
1093 {
1094         arch_pick_mmap_layout(current->mm);
1095
1096         /* This is the point of no return */
1097         current->sas_ss_sp = current->sas_ss_size = 0;
1098
1099         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1100                 set_dumpable(current->mm, SUID_DUMPABLE_ENABLED);
1101         else
1102                 set_dumpable(current->mm, suid_dumpable);
1103
1104         set_task_comm(current, bprm->tcomm);
1105
1106         /* Set the new mm task size. We have to do that late because it may
1107          * depend on TIF_32BIT which is only updated in flush_thread() on
1108          * some architectures like powerpc
1109          */
1110         current->mm->task_size = TASK_SIZE;
1111
1112         /* install the new credentials */
1113         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1114             !gid_eq(bprm->cred->gid, current_egid())) {
1115                 current->pdeath_signal = 0;
1116         } else {
1117                 would_dump(bprm, bprm->file);
1118                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1119                         set_dumpable(current->mm, suid_dumpable);
1120         }
1121
1122         /*
1123          * Flush performance counters when crossing a
1124          * security domain:
1125          */
1126         if (!get_dumpable(current->mm))
1127                 perf_event_exit_task(current);
1128
1129         /* An exec changes our domain. We are no longer part of the thread
1130            group */
1131
1132         current->self_exec_id++;
1133
1134         flush_signal_handlers(current, 0);
1135         do_close_on_exec(current->files);
1136 }
1137 EXPORT_SYMBOL(setup_new_exec);
1138
1139 /*
1140  * Prepare credentials and lock ->cred_guard_mutex.
1141  * install_exec_creds() commits the new creds and drops the lock.
1142  * Or, if exec fails before, free_bprm() should release ->cred and
1143  * and unlock.
1144  */
1145 int prepare_bprm_creds(struct linux_binprm *bprm)
1146 {
1147         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1148                 return -ERESTARTNOINTR;
1149
1150         bprm->cred = prepare_exec_creds();
1151         if (likely(bprm->cred))
1152                 return 0;
1153
1154         mutex_unlock(&current->signal->cred_guard_mutex);
1155         return -ENOMEM;
1156 }
1157
1158 void free_bprm(struct linux_binprm *bprm)
1159 {
1160         free_arg_pages(bprm);
1161         if (bprm->cred) {
1162                 mutex_unlock(&current->signal->cred_guard_mutex);
1163                 abort_creds(bprm->cred);
1164         }
1165         kfree(bprm);
1166 }
1167
1168 /*
1169  * install the new credentials for this executable
1170  */
1171 void install_exec_creds(struct linux_binprm *bprm)
1172 {
1173         security_bprm_committing_creds(bprm);
1174
1175         commit_creds(bprm->cred);
1176         bprm->cred = NULL;
1177         /*
1178          * cred_guard_mutex must be held at least to this point to prevent
1179          * ptrace_attach() from altering our determination of the task's
1180          * credentials; any time after this it may be unlocked.
1181          */
1182         security_bprm_committed_creds(bprm);
1183         mutex_unlock(&current->signal->cred_guard_mutex);
1184 }
1185 EXPORT_SYMBOL(install_exec_creds);
1186
1187 /*
1188  * determine how safe it is to execute the proposed program
1189  * - the caller must hold ->cred_guard_mutex to protect against
1190  *   PTRACE_ATTACH
1191  */
1192 static int check_unsafe_exec(struct linux_binprm *bprm)
1193 {
1194         struct task_struct *p = current, *t;
1195         unsigned n_fs;
1196         int res = 0;
1197
1198         if (p->ptrace) {
1199                 if (p->ptrace & PT_PTRACE_CAP)
1200                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1201                 else
1202                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1203         }
1204
1205         /*
1206          * This isn't strictly necessary, but it makes it harder for LSMs to
1207          * mess up.
1208          */
1209         if (current->no_new_privs)
1210                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1211
1212         n_fs = 1;
1213         spin_lock(&p->fs->lock);
1214         rcu_read_lock();
1215         for (t = next_thread(p); t != p; t = next_thread(t)) {
1216                 if (t->fs == p->fs)
1217                         n_fs++;
1218         }
1219         rcu_read_unlock();
1220
1221         if (p->fs->users > n_fs) {
1222                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1223         } else {
1224                 res = -EAGAIN;
1225                 if (!p->fs->in_exec) {
1226                         p->fs->in_exec = 1;
1227                         res = 1;
1228                 }
1229         }
1230         spin_unlock(&p->fs->lock);
1231
1232         return res;
1233 }
1234
1235 /*
1236  * Fill the binprm structure from the inode.
1237  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1238  *
1239  * This may be called multiple times for binary chains (scripts for example).
1240  */
1241 int prepare_binprm(struct linux_binprm *bprm)
1242 {
1243         umode_t mode;
1244         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1245         int retval;
1246
1247         mode = inode->i_mode;
1248         if (bprm->file->f_op == NULL)
1249                 return -EACCES;
1250
1251         /* clear any previous set[ug]id data from a previous binary */
1252         bprm->cred->euid = current_euid();
1253         bprm->cred->egid = current_egid();
1254
1255         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1256             !current->no_new_privs) {
1257                 /* Set-uid? */
1258                 if (mode & S_ISUID) {
1259                         if (!kuid_has_mapping(bprm->cred->user_ns, inode->i_uid))
1260                                 return -EPERM;
1261                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1262                         bprm->cred->euid = inode->i_uid;
1263
1264                 }
1265
1266                 /* Set-gid? */
1267                 /*
1268                  * If setgid is set but no group execute bit then this
1269                  * is a candidate for mandatory locking, not a setgid
1270                  * executable.
1271                  */
1272                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1273                         if (!kgid_has_mapping(bprm->cred->user_ns, inode->i_gid))
1274                                 return -EPERM;
1275                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1276                         bprm->cred->egid = inode->i_gid;
1277                 }
1278         }
1279
1280         /* fill in binprm security blob */
1281         retval = security_bprm_set_creds(bprm);
1282         if (retval)
1283                 return retval;
1284         bprm->cred_prepared = 1;
1285
1286         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1287         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1288 }
1289
1290 EXPORT_SYMBOL(prepare_binprm);
1291
1292 /*
1293  * Arguments are '\0' separated strings found at the location bprm->p
1294  * points to; chop off the first by relocating brpm->p to right after
1295  * the first '\0' encountered.
1296  */
1297 int remove_arg_zero(struct linux_binprm *bprm)
1298 {
1299         int ret = 0;
1300         unsigned long offset;
1301         char *kaddr;
1302         struct page *page;
1303
1304         if (!bprm->argc)
1305                 return 0;
1306
1307         do {
1308                 offset = bprm->p & ~PAGE_MASK;
1309                 page = get_arg_page(bprm, bprm->p, 0);
1310                 if (!page) {
1311                         ret = -EFAULT;
1312                         goto out;
1313                 }
1314                 kaddr = kmap_atomic(page);
1315
1316                 for (; offset < PAGE_SIZE && kaddr[offset];
1317                                 offset++, bprm->p++)
1318                         ;
1319
1320                 kunmap_atomic(kaddr);
1321                 put_arg_page(page);
1322
1323                 if (offset == PAGE_SIZE)
1324                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1325         } while (offset == PAGE_SIZE);
1326
1327         bprm->p++;
1328         bprm->argc--;
1329         ret = 0;
1330
1331 out:
1332         return ret;
1333 }
1334 EXPORT_SYMBOL(remove_arg_zero);
1335
1336 /*
1337  * cycle the list of binary formats handler, until one recognizes the image
1338  */
1339 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1340 {
1341         unsigned int depth = bprm->recursion_depth;
1342         int try,retval;
1343         struct linux_binfmt *fmt;
1344         pid_t old_pid, old_vpid;
1345
1346         retval = security_bprm_check(bprm);
1347         if (retval)
1348                 return retval;
1349
1350         retval = audit_bprm(bprm);
1351         if (retval)
1352                 return retval;
1353
1354         /* Need to fetch pid before load_binary changes it */
1355         old_pid = current->pid;
1356         rcu_read_lock();
1357         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1358         rcu_read_unlock();
1359
1360         retval = -ENOENT;
1361         for (try=0; try<2; try++) {
1362                 read_lock(&binfmt_lock);
1363                 list_for_each_entry(fmt, &formats, lh) {
1364                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1365                         if (!fn)
1366                                 continue;
1367                         if (!try_module_get(fmt->module))
1368                                 continue;
1369                         read_unlock(&binfmt_lock);
1370                         retval = fn(bprm, regs);
1371                         /*
1372                          * Restore the depth counter to its starting value
1373                          * in this call, so we don't have to rely on every
1374                          * load_binary function to restore it on return.
1375                          */
1376                         bprm->recursion_depth = depth;
1377                         if (retval >= 0) {
1378                                 if (depth == 0) {
1379                                         trace_sched_process_exec(current, old_pid, bprm);
1380                                         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1381                                 }
1382                                 put_binfmt(fmt);
1383                                 allow_write_access(bprm->file);
1384                                 if (bprm->file)
1385                                         fput(bprm->file);
1386                                 bprm->file = NULL;
1387                                 current->did_exec = 1;
1388                                 proc_exec_connector(current);
1389                                 return retval;
1390                         }
1391                         read_lock(&binfmt_lock);
1392                         put_binfmt(fmt);
1393                         if (retval != -ENOEXEC || bprm->mm == NULL)
1394                                 break;
1395                         if (!bprm->file) {
1396                                 read_unlock(&binfmt_lock);
1397                                 return retval;
1398                         }
1399                 }
1400                 read_unlock(&binfmt_lock);
1401 #ifdef CONFIG_MODULES
1402                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1403                         break;
1404                 } else {
1405 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1406                         if (printable(bprm->buf[0]) &&
1407                             printable(bprm->buf[1]) &&
1408                             printable(bprm->buf[2]) &&
1409                             printable(bprm->buf[3]))
1410                                 break; /* -ENOEXEC */
1411                         if (try)
1412                                 break; /* -ENOEXEC */
1413                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1414                 }
1415 #else
1416                 break;
1417 #endif
1418         }
1419         return retval;
1420 }
1421
1422 EXPORT_SYMBOL(search_binary_handler);
1423
1424 /*
1425  * sys_execve() executes a new program.
1426  */
1427 static int do_execve_common(const char *filename,
1428                                 struct user_arg_ptr argv,
1429                                 struct user_arg_ptr envp,
1430                                 struct pt_regs *regs)
1431 {
1432         struct linux_binprm *bprm;
1433         struct file *file;
1434         struct files_struct *displaced;
1435         bool clear_in_exec;
1436         int retval;
1437         const struct cred *cred = current_cred();
1438
1439         /*
1440          * We move the actual failure in case of RLIMIT_NPROC excess from
1441          * set*uid() to execve() because too many poorly written programs
1442          * don't check setuid() return code.  Here we additionally recheck
1443          * whether NPROC limit is still exceeded.
1444          */
1445         if ((current->flags & PF_NPROC_EXCEEDED) &&
1446             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1447                 retval = -EAGAIN;
1448                 goto out_ret;
1449         }
1450
1451         /* We're below the limit (still or again), so we don't want to make
1452          * further execve() calls fail. */
1453         current->flags &= ~PF_NPROC_EXCEEDED;
1454
1455         retval = unshare_files(&displaced);
1456         if (retval)
1457                 goto out_ret;
1458
1459         retval = -ENOMEM;
1460         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1461         if (!bprm)
1462                 goto out_files;
1463
1464         retval = prepare_bprm_creds(bprm);
1465         if (retval)
1466                 goto out_free;
1467
1468         retval = check_unsafe_exec(bprm);
1469         if (retval < 0)
1470                 goto out_free;
1471         clear_in_exec = retval;
1472         current->in_execve = 1;
1473
1474         file = open_exec(filename);
1475         retval = PTR_ERR(file);
1476         if (IS_ERR(file))
1477                 goto out_unmark;
1478
1479         sched_exec();
1480
1481         bprm->file = file;
1482         bprm->filename = filename;
1483         bprm->interp = filename;
1484
1485         retval = bprm_mm_init(bprm);
1486         if (retval)
1487                 goto out_file;
1488
1489         bprm->argc = count(argv, MAX_ARG_STRINGS);
1490         if ((retval = bprm->argc) < 0)
1491                 goto out;
1492
1493         bprm->envc = count(envp, MAX_ARG_STRINGS);
1494         if ((retval = bprm->envc) < 0)
1495                 goto out;
1496
1497         retval = prepare_binprm(bprm);
1498         if (retval < 0)
1499                 goto out;
1500
1501         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1502         if (retval < 0)
1503                 goto out;
1504
1505         bprm->exec = bprm->p;
1506         retval = copy_strings(bprm->envc, envp, bprm);
1507         if (retval < 0)
1508                 goto out;
1509
1510         retval = copy_strings(bprm->argc, argv, bprm);
1511         if (retval < 0)
1512                 goto out;
1513
1514         retval = search_binary_handler(bprm,regs);
1515         if (retval < 0)
1516                 goto out;
1517
1518         /* execve succeeded */
1519         current->fs->in_exec = 0;
1520         current->in_execve = 0;
1521         acct_update_integrals(current);
1522         free_bprm(bprm);
1523         if (displaced)
1524                 put_files_struct(displaced);
1525         return retval;
1526
1527 out:
1528         if (bprm->mm) {
1529                 acct_arg_size(bprm, 0);
1530                 mmput(bprm->mm);
1531         }
1532
1533 out_file:
1534         if (bprm->file) {
1535                 allow_write_access(bprm->file);
1536                 fput(bprm->file);
1537         }
1538
1539 out_unmark:
1540         if (clear_in_exec)
1541                 current->fs->in_exec = 0;
1542         current->in_execve = 0;
1543
1544 out_free:
1545         free_bprm(bprm);
1546
1547 out_files:
1548         if (displaced)
1549                 reset_files_struct(displaced);
1550 out_ret:
1551         return retval;
1552 }
1553
1554 int do_execve(const char *filename,
1555         const char __user *const __user *__argv,
1556         const char __user *const __user *__envp,
1557         struct pt_regs *regs)
1558 {
1559         struct user_arg_ptr argv = { .ptr.native = __argv };
1560         struct user_arg_ptr envp = { .ptr.native = __envp };
1561         return do_execve_common(filename, argv, envp, regs);
1562 }
1563
1564 #ifdef CONFIG_COMPAT
1565 int compat_do_execve(char *filename,
1566         compat_uptr_t __user *__argv,
1567         compat_uptr_t __user *__envp,
1568         struct pt_regs *regs)
1569 {
1570         struct user_arg_ptr argv = {
1571                 .is_compat = true,
1572                 .ptr.compat = __argv,
1573         };
1574         struct user_arg_ptr envp = {
1575                 .is_compat = true,
1576                 .ptr.compat = __envp,
1577         };
1578         return do_execve_common(filename, argv, envp, regs);
1579 }
1580 #endif
1581
1582 void set_binfmt(struct linux_binfmt *new)
1583 {
1584         struct mm_struct *mm = current->mm;
1585
1586         if (mm->binfmt)
1587                 module_put(mm->binfmt->module);
1588
1589         mm->binfmt = new;
1590         if (new)
1591                 __module_get(new->module);
1592 }
1593
1594 EXPORT_SYMBOL(set_binfmt);
1595
1596 /*
1597  * set_dumpable converts traditional three-value dumpable to two flags and
1598  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1599  * these bits are not changed atomically.  So get_dumpable can observe the
1600  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1601  * return either old dumpable or new one by paying attention to the order of
1602  * modifying the bits.
1603  *
1604  * dumpable |   mm->flags (binary)
1605  * old  new | initial interim  final
1606  * ---------+-----------------------
1607  *  0    1  |   00      01      01
1608  *  0    2  |   00      10(*)   11
1609  *  1    0  |   01      00      00
1610  *  1    2  |   01      11      11
1611  *  2    0  |   11      10(*)   00
1612  *  2    1  |   11      11      01
1613  *
1614  * (*) get_dumpable regards interim value of 10 as 11.
1615  */
1616 void set_dumpable(struct mm_struct *mm, int value)
1617 {
1618         switch (value) {
1619         case SUID_DUMPABLE_DISABLED:
1620                 clear_bit(MMF_DUMPABLE, &mm->flags);
1621                 smp_wmb();
1622                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1623                 break;
1624         case SUID_DUMPABLE_ENABLED:
1625                 set_bit(MMF_DUMPABLE, &mm->flags);
1626                 smp_wmb();
1627                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1628                 break;
1629         case SUID_DUMPABLE_SAFE:
1630                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1631                 smp_wmb();
1632                 set_bit(MMF_DUMPABLE, &mm->flags);
1633                 break;
1634         }
1635 }
1636
1637 int __get_dumpable(unsigned long mm_flags)
1638 {
1639         int ret;
1640
1641         ret = mm_flags & MMF_DUMPABLE_MASK;
1642         return (ret > SUID_DUMPABLE_ENABLED) ? SUID_DUMPABLE_SAFE : ret;
1643 }
1644
1645 int get_dumpable(struct mm_struct *mm)
1646 {
1647         return __get_dumpable(mm->flags);
1648 }