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         write_lock(&binfmt_lock);
  78         insert ? list_add(&fmt->lh, &formats) :
  79                  list_add_tail(&fmt->lh, &formats);
  80         write_unlock(&binfmt_lock);
  81 }
  82
  83 EXPORT_SYMBOL(__register_binfmt);
  84
  85 void unregister_binfmt(struct linux_binfmt * fmt)
  86 {
  87         write_lock(&binfmt_lock);
  88         list_del(&fmt->lh);
  89         write_unlock(&binfmt_lock);
  90 }
  91
  92 EXPORT_SYMBOL(unregister_binfmt);
  93
  94 static inline void put_binfmt(struct linux_binfmt * fmt)
  95 {
  96         module_put(fmt->module);
  97 }
  98
  99 /*
 100  * Note that a shared library must be both readable and executable due to
 101  * security reasons.
 102  *
 103  * Also note that we take the address to load from from the file itself.
 104  */
 105 SYSCALL_DEFINE1(uselib, const char __user *, library)
 106 {
 107         struct file *file;
 108         struct filename *tmp = getname(library);
 109         int error = PTR_ERR(tmp);
 110         static const struct open_flags uselib_flags = {
 111                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 112                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
 113                 .intent = LOOKUP_OPEN,
 114                 .lookup_flags = LOOKUP_FOLLOW,
 115         };
 116
 117         if (IS_ERR(tmp))
 118                 goto out;
 119
 120         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 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_inode(file)->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 static 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                 const 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                         ++i;
 441
 442                         if (fatal_signal_pending(current))
 443                                 return -ERESTARTNOHAND;
 444                         cond_resched();
 445                 }
 446         }
 447         return i;
 448 }
 449
 450 /*
 451  * 'copy_strings()' copies argument/environment strings from the old
 452  * processes's memory to the new process's stack.  The call to get_user_pages()
 453  * ensures the destination page is created and not swapped out.
 454  */
 455 static int copy_strings(int argc, struct user_arg_ptr argv,
 456                         struct linux_binprm *bprm)
 457 {
 458         struct page *kmapped_page = NULL;
 459         char *kaddr = NULL;
 460         unsigned long kpos = 0;
 461         int ret;
 462
 463         while (argc-- > 0) {
 464                 const char __user *str;
 465                 int len;
 466                 unsigned long pos;
 467
 468                 ret = -EFAULT;
 469                 str = get_user_arg_ptr(argv, argc);
 470                 if (IS_ERR(str))
 471                         goto out;
 472
 473                 len = strnlen_user(str, MAX_ARG_STRLEN);
 474                 if (!len)
 475                         goto out;
 476
 477                 ret = -E2BIG;
 478                 if (!valid_arg_len(bprm, len))
 479                         goto out;
 480
 481                 /* We're going to work our way backwords. */
 482                 pos = bprm->p;
 483                 str += len;
 484                 bprm->p -= len;
 485
 486                 while (len > 0) {
 487                         int offset, bytes_to_copy;
 488
 489                         if (fatal_signal_pending(current)) {
 490                                 ret = -ERESTARTNOHAND;
 491                                 goto out;
 492                         }
 493                         cond_resched();
 494
 495                         offset = pos % PAGE_SIZE;
 496                         if (offset == 0)
 497                                 offset = PAGE_SIZE;
 498
 499                         bytes_to_copy = offset;
 500                         if (bytes_to_copy > len)
 501                                 bytes_to_copy = len;
 502
 503                         offset -= bytes_to_copy;
 504                         pos -= bytes_to_copy;
 505                         str -= bytes_to_copy;
 506                         len -= bytes_to_copy;
 507
 508                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 509                                 struct page *page;
 510
 511                                 page = get_arg_page(bprm, pos, 1);
 512                                 if (!page) {
 513                                         ret = -E2BIG;
 514                                         goto out;
 515                                 }
 516
 517                                 if (kmapped_page) {
 518                                         flush_kernel_dcache_page(kmapped_page);
 519                                         kunmap(kmapped_page);
 520                                         put_arg_page(kmapped_page);
 521                                 }
 522                                 kmapped_page = page;
 523                                 kaddr = kmap(kmapped_page);
 524                                 kpos = pos & PAGE_MASK;
 525                                 flush_arg_page(bprm, kpos, kmapped_page);
 526                         }
 527                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 528                                 ret = -EFAULT;
 529                                 goto out;
 530                         }
 531                 }
 532         }
 533         ret = 0;
 534 out:
 535         if (kmapped_page) {
 536                 flush_kernel_dcache_page(kmapped_page);
 537                 kunmap(kmapped_page);
 538                 put_arg_page(kmapped_page);
 539         }
 540         return ret;
 541 }
 542
 543 /*
 544  * Like copy_strings, but get argv and its values from kernel memory.
 545  */
 546 int copy_strings_kernel(int argc, const char *const *__argv,
 547                         struct linux_binprm *bprm)
 548 {
 549         int r;
 550         mm_segment_t oldfs = get_fs();
 551         struct user_arg_ptr argv = {
 552                 .ptr.native = (const char __user *const  __user *)__argv,
 553         };
 554
 555         set_fs(KERNEL_DS);
 556         r = copy_strings(argc, argv, bprm);
 557         set_fs(oldfs);
 558
 559         return r;
 560 }
 561 EXPORT_SYMBOL(copy_strings_kernel);
 562
 563 #ifdef CONFIG_MMU
 564
 565 /*
 566  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 567  * the binfmt code determines where the new stack should reside, we shift it to
 568  * its final location.  The process proceeds as follows:
 569  *
 570  * 1) Use shift to calculate the new vma endpoints.
 571  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 572  *    arguments passed to subsequent functions are consistent.
 573  * 3) Move vma's page tables to the new range.
 574  * 4) Free up any cleared pgd range.
 575  * 5) Shrink the vma to cover only the new range.
 576  */
 577 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 578 {
 579         struct mm_struct *mm = vma->vm_mm;
 580         unsigned long old_start = vma->vm_start;
 581         unsigned long old_end = vma->vm_end;
 582         unsigned long length = old_end - old_start;
 583         unsigned long new_start = old_start - shift;
 584         unsigned long new_end = old_end - shift;
 585         struct mmu_gather tlb;
 586
 587         BUG_ON(new_start > new_end);
 588
 589         /*
 590          * ensure there are no vmas between where we want to go
 591          * and where we are
 592          */
 593         if (vma != find_vma(mm, new_start))
 594                 return -EFAULT;
 595
 596         /*
 597          * cover the whole range: [new_start, old_end)
 598          */
 599         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 600                 return -ENOMEM;
 601
 602         /*
 603          * move the page tables downwards, on failure we rely on
 604          * process cleanup to remove whatever mess we made.
 605          */
 606         if (length != move_page_tables(vma, old_start,
 607                                        vma, new_start, length, false))
 608                 return -ENOMEM;
 609
 610         lru_add_drain();
 611         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 612         if (new_end > old_start) {
 613                 /*
 614                  * when the old and new regions overlap clear from new_end.
 615                  */
 616                 free_pgd_range(&tlb, new_end, old_end, new_end,
 617                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 618         } else {
 619                 /*
 620                  * otherwise, clean from old_start; this is done to not touch
 621                  * the address space in [new_end, old_start) some architectures
 622                  * have constraints on va-space that make this illegal (IA64) -
 623                  * for the others its just a little faster.
 624                  */
 625                 free_pgd_range(&tlb, old_start, old_end, new_end,
 626                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 627         }
 628         tlb_finish_mmu(&tlb, old_start, old_end);
 629
 630         /*
 631          * Shrink the vma to just the new range.  Always succeeds.
 632          */
 633         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 634
 635         return 0;
 636 }
 637
 638 /*
 639  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 640  * the stack is optionally relocated, and some extra space is added.
 641  */
 642 int setup_arg_pages(struct linux_binprm *bprm,
 643                     unsigned long stack_top,
 644                     int executable_stack)
 645 {
 646         unsigned long ret;
 647         unsigned long stack_shift;
 648         struct mm_struct *mm = current->mm;
 649         struct vm_area_struct *vma = bprm->vma;
 650         struct vm_area_struct *prev = NULL;
 651         unsigned long vm_flags;
 652         unsigned long stack_base;
 653         unsigned long stack_size;
 654         unsigned long stack_expand;
 655         unsigned long rlim_stack;
 656
 657 #ifdef CONFIG_STACK_GROWSUP
 658         /* Limit stack size to 1GB */
 659         stack_base = rlimit_max(RLIMIT_STACK);
 660         if (stack_base > (1 << 30))
 661                 stack_base = 1 << 30;
 662
 663         /* Make sure we didn't let the argument array grow too large. */
 664         if (vma->vm_end - vma->vm_start > stack_base)
 665                 return -ENOMEM;
 666
 667         stack_base = PAGE_ALIGN(stack_top - stack_base);
 668
 669         stack_shift = vma->vm_start - stack_base;
 670         mm->arg_start = bprm->p - stack_shift;
 671         bprm->p = vma->vm_end - stack_shift;
 672 #else
 673         stack_top = arch_align_stack(stack_top);
 674         stack_top = PAGE_ALIGN(stack_top);
 675
 676         if (unlikely(stack_top < mmap_min_addr) ||
 677             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 678                 return -ENOMEM;
 679
 680         stack_shift = vma->vm_end - stack_top;
 681
 682         bprm->p -= stack_shift;
 683         mm->arg_start = bprm->p;
 684 #endif
 685
 686         if (bprm->loader)
 687                 bprm->loader -= stack_shift;
 688         bprm->exec -= stack_shift;
 689
 690         down_write(&mm->mmap_sem);
 691         vm_flags = VM_STACK_FLAGS;
 692
 693         /*
 694          * Adjust stack execute permissions; explicitly enable for
 695          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 696          * (arch default) otherwise.
 697          */
 698         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 699                 vm_flags |= VM_EXEC;
 700         else if (executable_stack == EXSTACK_DISABLE_X)
 701                 vm_flags &= ~VM_EXEC;
 702         vm_flags |= mm->def_flags;
 703         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 704
 705         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 706                         vm_flags);
 707         if (ret)
 708                 goto out_unlock;
 709         BUG_ON(prev != vma);
 710
 711         /* Move stack pages down in memory. */
 712         if (stack_shift) {
 713                 ret = shift_arg_pages(vma, stack_shift);
 714                 if (ret)
 715                         goto out_unlock;
 716         }
 717
 718         /* mprotect_fixup is overkill to remove the temporary stack flags */
 719         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 720
 721         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 722         stack_size = vma->vm_end - vma->vm_start;
 723         /*
 724          * Align this down to a page boundary as expand_stack
 725          * will align it up.
 726          */
 727         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 728 #ifdef CONFIG_STACK_GROWSUP
 729         if (stack_size + stack_expand > rlim_stack)
 730                 stack_base = vma->vm_start + rlim_stack;
 731         else
 732                 stack_base = vma->vm_end + stack_expand;
 733 #else
 734         if (stack_size + stack_expand > rlim_stack)
 735                 stack_base = vma->vm_end - rlim_stack;
 736         else
 737                 stack_base = vma->vm_start - stack_expand;
 738 #endif
 739         current->mm->start_stack = bprm->p;
 740         ret = expand_stack(vma, stack_base);
 741         if (ret)
 742                 ret = -EFAULT;
 743
 744 out_unlock:
 745         up_write(&mm->mmap_sem);
 746         return ret;
 747 }
 748 EXPORT_SYMBOL(setup_arg_pages);
 749
 750 #endif /* CONFIG_MMU */
 751
 752 struct file *open_exec(const char *name)
 753 {
 754         struct file *file;
 755         int err;
 756         struct filename tmp = { .name = name };
 757         static const struct open_flags open_exec_flags = {
 758                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 759                 .acc_mode = MAY_EXEC | MAY_OPEN,
 760                 .intent = LOOKUP_OPEN,
 761                 .lookup_flags = LOOKUP_FOLLOW,
 762         };
 763
 764         file = do_filp_open(AT_FDCWD, &tmp, &open_exec_flags);
 765         if (IS_ERR(file))
 766                 goto out;
 767
 768         err = -EACCES;
 769         if (!S_ISREG(file_inode(file)->i_mode))
 770                 goto exit;
 771
 772         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 773                 goto exit;
 774
 775         fsnotify_open(file);
 776
 777         err = deny_write_access(file);
 778         if (err)
 779                 goto exit;
 780
 781 out:
 782         return file;
 783
 784 exit:
 785         fput(file);
 786         return ERR_PTR(err);
 787 }
 788 EXPORT_SYMBOL(open_exec);
 789
 790 int kernel_read(struct file *file, loff_t offset,
 791                 char *addr, unsigned long count)
 792 {
 793         mm_segment_t old_fs;
 794         loff_t pos = offset;
 795         int result;
 796
 797         old_fs = get_fs();
 798         set_fs(get_ds());
 799         /* The cast to a user pointer is valid due to the set_fs() */
 800         result = vfs_read(file, (void __user *)addr, count, &pos);
 801         set_fs(old_fs);
 802         return result;
 803 }
 804
 805 EXPORT_SYMBOL(kernel_read);
 806
 807 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 808 {
 809         ssize_t res = file->f_op->read(file, (void __user *)addr, len, &pos);
 810         if (res > 0)
 811                 flush_icache_range(addr, addr + len);
 812         return res;
 813 }
 814 EXPORT_SYMBOL(read_code);
 815
 816 static int exec_mmap(struct mm_struct *mm)
 817 {
 818         struct task_struct *tsk;
 819         struct mm_struct * old_mm, *active_mm;
 820
 821         /* Notify parent that we're no longer interested in the old VM */
 822         tsk = current;
 823         old_mm = current->mm;
 824         mm_release(tsk, old_mm);
 825
 826         if (old_mm) {
 827                 sync_mm_rss(old_mm);
 828                 /*
 829                  * Make sure that if there is a core dump in progress
 830                  * for the old mm, we get out and die instead of going
 831                  * through with the exec.  We must hold mmap_sem around
 832                  * checking core_state and changing tsk->mm.
 833                  */
 834                 down_read(&old_mm->mmap_sem);
 835                 if (unlikely(old_mm->core_state)) {
 836                         up_read(&old_mm->mmap_sem);
 837                         return -EINTR;
 838                 }
 839         }
 840         task_lock(tsk);
 841         active_mm = tsk->active_mm;
 842         tsk->mm = mm;
 843         tsk->active_mm = mm;
 844         activate_mm(active_mm, mm);
 845         task_unlock(tsk);
 846         arch_pick_mmap_layout(mm);
 847         if (old_mm) {
 848                 up_read(&old_mm->mmap_sem);
 849                 BUG_ON(active_mm != old_mm);
 850                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 851                 mm_update_next_owner(old_mm);
 852                 mmput(old_mm);
 853                 return 0;
 854         }
 855         mmdrop(active_mm);
 856         return 0;
 857 }
 858
 859 /*
 860  * This function makes sure the current process has its own signal table,
 861  * so that flush_signal_handlers can later reset the handlers without
 862  * disturbing other processes.  (Other processes might share the signal
 863  * table via the CLONE_SIGHAND option to clone().)
 864  */
 865 static int de_thread(struct task_struct *tsk)
 866 {
 867         struct signal_struct *sig = tsk->signal;
 868         struct sighand_struct *oldsighand = tsk->sighand;
 869         spinlock_t *lock = &oldsighand->siglock;
 870
 871         if (thread_group_empty(tsk))
 872                 goto no_thread_group;
 873
 874         /*
 875          * Kill all other threads in the thread group.
 876          */
 877         spin_lock_irq(lock);
 878         if (signal_group_exit(sig)) {
 879                 /*
 880                  * Another group action in progress, just
 881                  * return so that the signal is processed.
 882                  */
 883                 spin_unlock_irq(lock);
 884                 return -EAGAIN;
 885         }
 886
 887         sig->group_exit_task = tsk;
 888         sig->notify_count = zap_other_threads(tsk);
 889         if (!thread_group_leader(tsk))
 890                 sig->notify_count--;
 891
 892         while (sig->notify_count) {
 893                 __set_current_state(TASK_KILLABLE);
 894                 spin_unlock_irq(lock);
 895                 schedule();
 896                 if (unlikely(__fatal_signal_pending(tsk)))
 897                         goto killed;
 898                 spin_lock_irq(lock);
 899         }
 900         spin_unlock_irq(lock);
 901
 902         /*
 903          * At this point all other threads have exited, all we have to
 904          * do is to wait for the thread group leader to become inactive,
 905          * and to assume its PID:
 906          */
 907         if (!thread_group_leader(tsk)) {
 908                 struct task_struct *leader = tsk->group_leader;
 909
 910                 sig->notify_count = -1; /* for exit_notify() */
 911                 for (;;) {
 912                         threadgroup_change_begin(tsk);
 913                         write_lock_irq(&tasklist_lock);
 914                         if (likely(leader->exit_state))
 915                                 break;
 916                         __set_current_state(TASK_KILLABLE);
 917                         write_unlock_irq(&tasklist_lock);
 918                         threadgroup_change_end(tsk);
 919                         schedule();
 920                         if (unlikely(__fatal_signal_pending(tsk)))
 921                                 goto killed;
 922                 }
 923
 924                 /*
 925                  * The only record we have of the real-time age of a
 926                  * process, regardless of execs it's done, is start_time.
 927                  * All the past CPU time is accumulated in signal_struct
 928                  * from sister threads now dead.  But in this non-leader
 929                  * exec, nothing survives from the original leader thread,
 930                  * whose birth marks the true age of this process now.
 931                  * When we take on its identity by switching to its PID, we
 932                  * also take its birthdate (always earlier than our own).
 933                  */
 934                 tsk->start_time = leader->start_time;
 935                 tsk->real_start_time = leader->real_start_time;
 936
 937                 BUG_ON(!same_thread_group(leader, tsk));
 938                 BUG_ON(has_group_leader_pid(tsk));
 939                 /*
 940                  * An exec() starts a new thread group with the
 941                  * TGID of the previous thread group. Rehash the
 942                  * two threads with a switched PID, and release
 943                  * the former thread group leader:
 944                  */
 945
 946                 /* Become a process group leader with the old leader's pid.
 947                  * The old leader becomes a thread of the this thread group.
 948                  * Note: The old leader also uses this pid until release_task
 949                  *       is called.  Odd but simple and correct.
 950                  */
 951                 tsk->pid = leader->pid;
 952                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
 953                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 954                 transfer_pid(leader, tsk, PIDTYPE_SID);
 955
 956                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 957                 list_replace_init(&leader->sibling, &tsk->sibling);
 958
 959                 tsk->group_leader = tsk;
 960                 leader->group_leader = tsk;
 961
 962                 tsk->exit_signal = SIGCHLD;
 963                 leader->exit_signal = -1;
 964
 965                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 966                 leader->exit_state = EXIT_DEAD;
 967
 968                 /*
 969                  * We are going to release_task()->ptrace_unlink() silently,
 970                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 971                  * the tracer wont't block again waiting for this thread.
 972                  */
 973                 if (unlikely(leader->ptrace))
 974                         __wake_up_parent(leader, leader->parent);
 975                 write_unlock_irq(&tasklist_lock);
 976                 threadgroup_change_end(tsk);
 977
 978                 release_task(leader);
 979         }
 980
 981         sig->group_exit_task = NULL;
 982         sig->notify_count = 0;
 983
 984 no_thread_group:
 985         /* we have changed execution domain */
 986         tsk->exit_signal = SIGCHLD;
 987
 988         exit_itimers(sig);
 989         flush_itimer_signals();
 990
 991         if (atomic_read(&oldsighand->count) != 1) {
 992                 struct sighand_struct *newsighand;
 993                 /*
 994                  * This ->sighand is shared with the CLONE_SIGHAND
 995                  * but not CLONE_THREAD task, switch to the new one.
 996                  */
 997                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 998                 if (!newsighand)
 999                         return -ENOMEM;
1000
1001                 atomic_set(&newsighand->count, 1);
1002                 memcpy(newsighand->action, oldsighand->action,
1003                        sizeof(newsighand->action));
1004
1005                 write_lock_irq(&tasklist_lock);
1006                 spin_lock(&oldsighand->siglock);
1007                 rcu_assign_pointer(tsk->sighand, newsighand);
1008                 spin_unlock(&oldsighand->siglock);
1009                 write_unlock_irq(&tasklist_lock);
1010
1011                 __cleanup_sighand(oldsighand);
1012         }
1013
1014         BUG_ON(!thread_group_leader(tsk));
1015         return 0;
1016
1017 killed:
1018         /* protects against exit_notify() and __exit_signal() */
1019         read_lock(&tasklist_lock);
1020         sig->group_exit_task = NULL;
1021         sig->notify_count = 0;
1022         read_unlock(&tasklist_lock);
1023         return -EAGAIN;
1024 }
1025
1026 char *get_task_comm(char *buf, struct task_struct *tsk)
1027 {
1028         /* buf must be at least sizeof(tsk->comm) in size */
1029         task_lock(tsk);
1030         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1031         task_unlock(tsk);
1032         return buf;
1033 }
1034 EXPORT_SYMBOL_GPL(get_task_comm);
1035
1036 /*
1037  * These functions flushes out all traces of the currently running executable
1038  * so that a new one can be started
1039  */
1040
1041 void set_task_comm(struct task_struct *tsk, char *buf)
1042 {
1043         task_lock(tsk);
1044         trace_task_rename(tsk, buf);
1045         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1046         task_unlock(tsk);
1047         perf_event_comm(tsk);
1048 }
1049
1050 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1051 {
1052         int i, ch;
1053
1054         /* Copies the binary name from after last slash */
1055         for (i = 0; (ch = *(fn++)) != '\0';) {
1056                 if (ch == '/')
1057                         i = 0; /* overwrite what we wrote */
1058                 else
1059                         if (i < len - 1)
1060                                 tcomm[i++] = ch;
1061         }
1062         tcomm[i] = '\0';
1063 }
1064
1065 int flush_old_exec(struct linux_binprm * bprm)
1066 {
1067         int retval;
1068
1069         /*
1070          * Make sure we have a private signal table and that
1071          * we are unassociated from the previous thread group.
1072          */
1073         retval = de_thread(current);
1074         if (retval)
1075                 goto out;
1076
1077         set_mm_exe_file(bprm->mm, bprm->file);
1078
1079         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1080         /*
1081          * Release all of the old mmap stuff
1082          */
1083         acct_arg_size(bprm, 0);
1084         retval = exec_mmap(bprm->mm);
1085         if (retval)
1086                 goto out;
1087
1088         bprm->mm = NULL;                /* We're using it now */
1089
1090         set_fs(USER_DS);
1091         current->flags &=
1092                 ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | PF_NOFREEZE);
1093         flush_thread();
1094         current->personality &= ~bprm->per_clear;
1095
1096         return 0;
1097
1098 out:
1099         return retval;
1100 }
1101 EXPORT_SYMBOL(flush_old_exec);
1102
1103 void would_dump(struct linux_binprm *bprm, struct file *file)
1104 {
1105         if (inode_permission(file_inode(file), MAY_READ) < 0)
1106                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1107 }
1108 EXPORT_SYMBOL(would_dump);
1109
1110 void setup_new_exec(struct linux_binprm * bprm)
1111 {
1112         arch_pick_mmap_layout(current->mm);
1113
1114         /* This is the point of no return */
1115         current->sas_ss_sp = current->sas_ss_size = 0;
1116
1117         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1118                 set_dumpable(current->mm, SUID_DUMP_USER);
1119         else
1120                 set_dumpable(current->mm, suid_dumpable);
1121
1122         set_task_comm(current, bprm->tcomm);
1123
1124         /* Set the new mm task size. We have to do that late because it may
1125          * depend on TIF_32BIT which is only updated in flush_thread() on
1126          * some architectures like powerpc
1127          */
1128         current->mm->task_size = TASK_SIZE;
1129
1130         /* install the new credentials */
1131         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1132             !gid_eq(bprm->cred->gid, current_egid())) {
1133                 current->pdeath_signal = 0;
1134         } else {
1135                 would_dump(bprm, bprm->file);
1136                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1137                         set_dumpable(current->mm, suid_dumpable);
1138         }
1139
1140         /* An exec changes our domain. We are no longer part of the thread
1141            group */
1142
1143         current->self_exec_id++;
1144
1145         flush_signal_handlers(current, 0);
1146         do_close_on_exec(current->files);
1147 }
1148 EXPORT_SYMBOL(setup_new_exec);
1149
1150 /*
1151  * Prepare credentials and lock ->cred_guard_mutex.
1152  * install_exec_creds() commits the new creds and drops the lock.
1153  * Or, if exec fails before, free_bprm() should release ->cred and
1154  * and unlock.
1155  */
1156 int prepare_bprm_creds(struct linux_binprm *bprm)
1157 {
1158         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1159                 return -ERESTARTNOINTR;
1160
1161         bprm->cred = prepare_exec_creds();
1162         if (likely(bprm->cred))
1163                 return 0;
1164
1165         mutex_unlock(&current->signal->cred_guard_mutex);
1166         return -ENOMEM;
1167 }
1168
1169 void free_bprm(struct linux_binprm *bprm)
1170 {
1171         free_arg_pages(bprm);
1172         if (bprm->cred) {
1173                 mutex_unlock(&current->signal->cred_guard_mutex);
1174                 abort_creds(bprm->cred);
1175         }
1176         /* If a binfmt changed the interp, free it. */
1177         if (bprm->interp != bprm->filename)
1178                 kfree(bprm->interp);
1179         kfree(bprm);
1180 }
1181
1182 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1183 {
1184         /* If a binfmt changed the interp, free it first. */
1185         if (bprm->interp != bprm->filename)
1186                 kfree(bprm->interp);
1187         bprm->interp = kstrdup(interp, GFP_KERNEL);
1188         if (!bprm->interp)
1189                 return -ENOMEM;
1190         return 0;
1191 }
1192 EXPORT_SYMBOL(bprm_change_interp);
1193
1194 /*
1195  * install the new credentials for this executable
1196  */
1197 void install_exec_creds(struct linux_binprm *bprm)
1198 {
1199         security_bprm_committing_creds(bprm);
1200
1201         commit_creds(bprm->cred);
1202         bprm->cred = NULL;
1203
1204         /*
1205          * Disable monitoring for regular users
1206          * when executing setuid binaries. Must
1207          * wait until new credentials are committed
1208          * by commit_creds() above
1209          */
1210         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1211                 perf_event_exit_task(current);
1212         /*
1213          * cred_guard_mutex must be held at least to this point to prevent
1214          * ptrace_attach() from altering our determination of the task's
1215          * credentials; any time after this it may be unlocked.
1216          */
1217         security_bprm_committed_creds(bprm);
1218         mutex_unlock(&current->signal->cred_guard_mutex);
1219 }
1220 EXPORT_SYMBOL(install_exec_creds);
1221
1222 /*
1223  * determine how safe it is to execute the proposed program
1224  * - the caller must hold ->cred_guard_mutex to protect against
1225  *   PTRACE_ATTACH
1226  */
1227 static int check_unsafe_exec(struct linux_binprm *bprm)
1228 {
1229         struct task_struct *p = current, *t;
1230         unsigned n_fs;
1231         int res = 0;
1232
1233         if (p->ptrace) {
1234                 if (p->ptrace & PT_PTRACE_CAP)
1235                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1236                 else
1237                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1238         }
1239
1240         /*
1241          * This isn't strictly necessary, but it makes it harder for LSMs to
1242          * mess up.
1243          */
1244         if (current->no_new_privs)
1245                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1246
1247         n_fs = 1;
1248         spin_lock(&p->fs->lock);
1249         rcu_read_lock();
1250         for (t = next_thread(p); t != p; t = next_thread(t)) {
1251                 if (t->fs == p->fs)
1252                         n_fs++;
1253         }
1254         rcu_read_unlock();
1255
1256         if (p->fs->users > n_fs) {
1257                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1258         } else {
1259                 res = -EAGAIN;
1260                 if (!p->fs->in_exec) {
1261                         p->fs->in_exec = 1;
1262                         res = 1;
1263                 }
1264         }
1265         spin_unlock(&p->fs->lock);
1266
1267         return res;
1268 }
1269
1270 /*
1271  * Fill the binprm structure from the inode.
1272  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1273  *
1274  * This may be called multiple times for binary chains (scripts for example).
1275  */
1276 int prepare_binprm(struct linux_binprm *bprm)
1277 {
1278         umode_t mode;
1279         struct inode * inode = file_inode(bprm->file);
1280         int retval;
1281
1282         mode = inode->i_mode;
1283         if (bprm->file->f_op == NULL)
1284                 return -EACCES;
1285
1286         /* clear any previous set[ug]id data from a previous binary */
1287         bprm->cred->euid = current_euid();
1288         bprm->cred->egid = current_egid();
1289
1290         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1291             !current->no_new_privs &&
1292             kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1293             kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1294                 /* Set-uid? */
1295                 if (mode & S_ISUID) {
1296                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1297                         bprm->cred->euid = inode->i_uid;
1298                 }
1299
1300                 /* Set-gid? */
1301                 /*
1302                  * If setgid is set but no group execute bit then this
1303                  * is a candidate for mandatory locking, not a setgid
1304                  * executable.
1305                  */
1306                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1307                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1308                         bprm->cred->egid = inode->i_gid;
1309                 }
1310         }
1311
1312         /* fill in binprm security blob */
1313         retval = security_bprm_set_creds(bprm);
1314         if (retval)
1315                 return retval;
1316         bprm->cred_prepared = 1;
1317
1318         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1319         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1320 }
1321
1322 EXPORT_SYMBOL(prepare_binprm);
1323
1324 /*
1325  * Arguments are '\0' separated strings found at the location bprm->p
1326  * points to; chop off the first by relocating brpm->p to right after
1327  * the first '\0' encountered.
1328  */
1329 int remove_arg_zero(struct linux_binprm *bprm)
1330 {
1331         int ret = 0;
1332         unsigned long offset;
1333         char *kaddr;
1334         struct page *page;
1335
1336         if (!bprm->argc)
1337                 return 0;
1338
1339         do {
1340                 offset = bprm->p & ~PAGE_MASK;
1341                 page = get_arg_page(bprm, bprm->p, 0);
1342                 if (!page) {
1343                         ret = -EFAULT;
1344                         goto out;
1345                 }
1346                 kaddr = kmap_atomic(page);
1347
1348                 for (; offset < PAGE_SIZE && kaddr[offset];
1349                                 offset++, bprm->p++)
1350                         ;
1351
1352                 kunmap_atomic(kaddr);
1353                 put_arg_page(page);
1354
1355                 if (offset == PAGE_SIZE)
1356                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1357         } while (offset == PAGE_SIZE);
1358
1359         bprm->p++;
1360         bprm->argc--;
1361         ret = 0;
1362
1363 out:
1364         return ret;
1365 }
1366 EXPORT_SYMBOL(remove_arg_zero);
1367
1368 /*
1369  * cycle the list of binary formats handler, until one recognizes the image
1370  */
1371 int search_binary_handler(struct linux_binprm *bprm)
1372 {
1373         unsigned int depth = bprm->recursion_depth;
1374         int try,retval;
1375         struct linux_binfmt *fmt;
1376         pid_t old_pid, old_vpid;
1377
1378         /* This allows 4 levels of binfmt rewrites before failing hard. */
1379         if (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         /* Need to fetch pid before load_binary changes it */
1391         old_pid = current->pid;
1392         rcu_read_lock();
1393         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1394         rcu_read_unlock();
1395
1396         retval = -ENOENT;
1397         for (try=0; try<2; try++) {
1398                 read_lock(&binfmt_lock);
1399                 list_for_each_entry(fmt, &formats, lh) {
1400                         int (*fn)(struct linux_binprm *) = fmt->load_binary;
1401                         if (!fn)
1402                                 continue;
1403                         if (!try_module_get(fmt->module))
1404                                 continue;
1405                         read_unlock(&binfmt_lock);
1406                         bprm->recursion_depth = depth + 1;
1407                         retval = fn(bprm);
1408                         bprm->recursion_depth = depth;
1409                         if (retval >= 0) {
1410                                 if (depth == 0) {
1411                                         trace_sched_process_exec(current, old_pid, bprm);
1412                                         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1413                                 }
1414                                 put_binfmt(fmt);
1415                                 allow_write_access(bprm->file);
1416                                 if (bprm->file)
1417                                         fput(bprm->file);
1418                                 bprm->file = NULL;
1419                                 current->did_exec = 1;
1420                                 proc_exec_connector(current);
1421                                 return retval;
1422                         }
1423                         read_lock(&binfmt_lock);
1424                         put_binfmt(fmt);
1425                         if (retval != -ENOEXEC || bprm->mm == NULL)
1426                                 break;
1427                         if (!bprm->file) {
1428                                 read_unlock(&binfmt_lock);
1429                                 return retval;
1430                         }
1431                 }
1432                 read_unlock(&binfmt_lock);
1433 #ifdef CONFIG_MODULES
1434                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1435                         break;
1436                 } else {
1437 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1438                         if (printable(bprm->buf[0]) &&
1439                             printable(bprm->buf[1]) &&
1440                             printable(bprm->buf[2]) &&
1441                             printable(bprm->buf[3]))
1442                                 break; /* -ENOEXEC */
1443                         if (try)
1444                                 break; /* -ENOEXEC */
1445                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1446                 }
1447 #else
1448                 break;
1449 #endif
1450         }
1451         return retval;
1452 }
1453
1454 EXPORT_SYMBOL(search_binary_handler);
1455
1456 /*
1457  * sys_execve() executes a new program.
1458  */
1459 static int do_execve_common(const char *filename,
1460                                 struct user_arg_ptr argv,
1461                                 struct user_arg_ptr envp)
1462 {
1463         struct linux_binprm *bprm;
1464         struct file *file;
1465         struct files_struct *displaced;
1466         bool clear_in_exec;
1467         int retval;
1468
1469         /*
1470          * We move the actual failure in case of RLIMIT_NPROC excess from
1471          * set*uid() to execve() because too many poorly written programs
1472          * don't check setuid() return code.  Here we additionally recheck
1473          * whether NPROC limit is still exceeded.
1474          */
1475         if ((current->flags & PF_NPROC_EXCEEDED) &&
1476             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1477                 retval = -EAGAIN;
1478                 goto out_ret;
1479         }
1480
1481         /* We're below the limit (still or again), so we don't want to make
1482          * further execve() calls fail. */
1483         current->flags &= ~PF_NPROC_EXCEEDED;
1484
1485         retval = unshare_files(&displaced);
1486         if (retval)
1487                 goto out_ret;
1488
1489         retval = -ENOMEM;
1490         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1491         if (!bprm)
1492                 goto out_files;
1493
1494         retval = prepare_bprm_creds(bprm);
1495         if (retval)
1496                 goto out_free;
1497
1498         retval = check_unsafe_exec(bprm);
1499         if (retval < 0)
1500                 goto out_free;
1501         clear_in_exec = retval;
1502         current->in_execve = 1;
1503
1504         file = open_exec(filename);
1505         retval = PTR_ERR(file);
1506         if (IS_ERR(file))
1507                 goto out_unmark;
1508
1509         sched_exec();
1510
1511         bprm->file = file;
1512         bprm->filename = filename;
1513         bprm->interp = filename;
1514
1515         retval = bprm_mm_init(bprm);
1516         if (retval)
1517                 goto out_file;
1518
1519         bprm->argc = count(argv, MAX_ARG_STRINGS);
1520         if ((retval = bprm->argc) < 0)
1521                 goto out;
1522
1523         bprm->envc = count(envp, MAX_ARG_STRINGS);
1524         if ((retval = bprm->envc) < 0)
1525                 goto out;
1526
1527         retval = prepare_binprm(bprm);
1528         if (retval < 0)
1529                 goto out;
1530
1531         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1532         if (retval < 0)
1533                 goto out;
1534
1535         bprm->exec = bprm->p;
1536         retval = copy_strings(bprm->envc, envp, bprm);
1537         if (retval < 0)
1538                 goto out;
1539
1540         retval = copy_strings(bprm->argc, argv, bprm);
1541         if (retval < 0)
1542                 goto out;
1543
1544         retval = search_binary_handler(bprm);
1545         if (retval < 0)
1546                 goto out;
1547
1548         /* execve succeeded */
1549         current->fs->in_exec = 0;
1550         current->in_execve = 0;
1551         acct_update_integrals(current);
1552         free_bprm(bprm);
1553         if (displaced)
1554                 put_files_struct(displaced);
1555         return retval;
1556
1557 out:
1558         if (bprm->mm) {
1559                 acct_arg_size(bprm, 0);
1560                 mmput(bprm->mm);
1561         }
1562
1563 out_file:
1564         if (bprm->file) {
1565                 allow_write_access(bprm->file);
1566                 fput(bprm->file);
1567         }
1568
1569 out_unmark:
1570         if (clear_in_exec)
1571                 current->fs->in_exec = 0;
1572         current->in_execve = 0;
1573
1574 out_free:
1575         free_bprm(bprm);
1576
1577 out_files:
1578         if (displaced)
1579                 reset_files_struct(displaced);
1580 out_ret:
1581         return retval;
1582 }
1583
1584 int do_execve(const char *filename,
1585         const char __user *const __user *__argv,
1586         const char __user *const __user *__envp)
1587 {
1588         struct user_arg_ptr argv = { .ptr.native = __argv };
1589         struct user_arg_ptr envp = { .ptr.native = __envp };
1590         return do_execve_common(filename, argv, envp);
1591 }
1592
1593 #ifdef CONFIG_COMPAT
1594 static int compat_do_execve(const char *filename,
1595         const compat_uptr_t __user *__argv,
1596         const compat_uptr_t __user *__envp)
1597 {
1598         struct user_arg_ptr argv = {
1599                 .is_compat = true,
1600                 .ptr.compat = __argv,
1601         };
1602         struct user_arg_ptr envp = {
1603                 .is_compat = true,
1604                 .ptr.compat = __envp,
1605         };
1606         return do_execve_common(filename, argv, envp);
1607 }
1608 #endif
1609
1610 void set_binfmt(struct linux_binfmt *new)
1611 {
1612         struct mm_struct *mm = current->mm;
1613
1614         if (mm->binfmt)
1615                 module_put(mm->binfmt->module);
1616
1617         mm->binfmt = new;
1618         if (new)
1619                 __module_get(new->module);
1620 }
1621
1622 EXPORT_SYMBOL(set_binfmt);
1623
1624 /*
1625  * set_dumpable converts traditional three-value dumpable to two flags and
1626  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1627  * these bits are not changed atomically.  So get_dumpable can observe the
1628  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1629  * return either old dumpable or new one by paying attention to the order of
1630  * modifying the bits.
1631  *
1632  * dumpable |   mm->flags (binary)
1633  * old  new | initial interim  final
1634  * ---------+-----------------------
1635  *  0    1  |   00      01      01
1636  *  0    2  |   00      10(*)   11
1637  *  1    0  |   01      00      00
1638  *  1    2  |   01      11      11
1639  *  2    0  |   11      10(*)   00
1640  *  2    1  |   11      11      01
1641  *
1642  * (*) get_dumpable regards interim value of 10 as 11.
1643  */
1644 void set_dumpable(struct mm_struct *mm, int value)
1645 {
1646         switch (value) {
1647         case SUID_DUMP_DISABLE:
1648                 clear_bit(MMF_DUMPABLE, &mm->flags);
1649                 smp_wmb();
1650                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1651                 break;
1652         case SUID_DUMP_USER:
1653                 set_bit(MMF_DUMPABLE, &mm->flags);
1654                 smp_wmb();
1655                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1656                 break;
1657         case SUID_DUMP_ROOT:
1658                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1659                 smp_wmb();
1660                 set_bit(MMF_DUMPABLE, &mm->flags);
1661                 break;
1662         }
1663 }
1664
1665 int __get_dumpable(unsigned long mm_flags)
1666 {
1667         int ret;
1668
1669         ret = mm_flags & MMF_DUMPABLE_MASK;
1670         return (ret > SUID_DUMP_USER) ? SUID_DUMP_ROOT : ret;
1671 }
1672
1673 int get_dumpable(struct mm_struct *mm)
1674 {
1675         return __get_dumpable(mm->flags);
1676 }
1677
1678 SYSCALL_DEFINE3(execve,
1679                 const char __user *, filename,
1680                 const char __user *const __user *, argv,
1681                 const char __user *const __user *, envp)
1682 {
1683         struct filename *path = getname(filename);
1684         int error = PTR_ERR(path);
1685         if (!IS_ERR(path)) {
1686                 error = do_execve(path->name, argv, envp);
1687                 putname(path);
1688         }
1689         return error;
1690 }
1691 #ifdef CONFIG_COMPAT
1692 asmlinkage long compat_sys_execve(const char __user * filename,
1693         const compat_uptr_t __user * argv,
1694         const compat_uptr_t __user * envp)
1695 {
1696         struct filename *path = getname(filename);
1697         int error = PTR_ERR(path);
1698         if (!IS_ERR(path)) {
1699                 error = compat_do_execve(path->name, argv, envp);
1700                 putname(path);
1701         }
1702         return error;
1703 }
1704 #endif