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         };
 115
 116         if (IS_ERR(tmp))
 117                 goto out;
 118
 119         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
 120         putname(tmp);
 121         error = PTR_ERR(file);
 122         if (IS_ERR(file))
 123                 goto out;
 124
 125         error = -EINVAL;
 126         if (!S_ISREG(file_inode(file)->i_mode))
 127                 goto exit;
 128
 129         error = -EACCES;
 130         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 131                 goto exit;
 132
 133         fsnotify_open(file);
 134
 135         error = -ENOEXEC;
 136         if(file->f_op) {
 137                 struct linux_binfmt * fmt;
 138
 139                 read_lock(&binfmt_lock);
 140                 list_for_each_entry(fmt, &formats, lh) {
 141                         if (!fmt->load_shlib)
 142                                 continue;
 143                         if (!try_module_get(fmt->module))
 144                                 continue;
 145                         read_unlock(&binfmt_lock);
 146                         error = fmt->load_shlib(file);
 147                         read_lock(&binfmt_lock);
 148                         put_binfmt(fmt);
 149                         if (error != -ENOEXEC)
 150                                 break;
 151                 }
 152                 read_unlock(&binfmt_lock);
 153         }
 154 exit:
 155         fput(file);
 156 out:
 157         return error;
 158 }
 159
 160 #ifdef CONFIG_MMU
 161 /*
 162  * The nascent bprm->mm is not visible until exec_mmap() but it can
 163  * use a lot of memory, account these pages in current->mm temporary
 164  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 165  * change the counter back via acct_arg_size(0).
 166  */
 167 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 168 {
 169         struct mm_struct *mm = current->mm;
 170         long diff = (long)(pages - bprm->vma_pages);
 171
 172         if (!mm || !diff)
 173                 return;
 174
 175         bprm->vma_pages = pages;
 176         add_mm_counter(mm, MM_ANONPAGES, diff);
 177 }
 178
 179 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 180                 int write)
 181 {
 182         struct page *page;
 183         int ret;
 184
 185 #ifdef CONFIG_STACK_GROWSUP
 186         if (write) {
 187                 ret = expand_downwards(bprm->vma, pos);
 188                 if (ret < 0)
 189                         return NULL;
 190         }
 191 #endif
 192         ret = get_user_pages(current, bprm->mm, pos,
 193                         1, write, 1, &page, NULL);
 194         if (ret <= 0)
 195                 return NULL;
 196
 197         if (write) {
 198                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 199                 struct rlimit *rlim;
 200
 201                 acct_arg_size(bprm, size / PAGE_SIZE);
 202
 203                 /*
 204                  * We've historically supported up to 32 pages (ARG_MAX)
 205                  * of argument strings even with small stacks
 206                  */
 207                 if (size <= ARG_MAX)
 208                         return page;
 209
 210                 /*
 211                  * Limit to 1/4-th the stack size for the argv+env strings.
 212                  * This ensures that:
 213                  *  - the remaining binfmt code will not run out of stack space,
 214                  *  - the program will have a reasonable amount of stack left
 215                  *    to work from.
 216                  */
 217                 rlim = current->signal->rlim;
 218                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 219                         put_page(page);
 220                         return NULL;
 221                 }
 222         }
 223
 224         return page;
 225 }
 226
 227 static void put_arg_page(struct page *page)
 228 {
 229         put_page(page);
 230 }
 231
 232 static void free_arg_page(struct linux_binprm *bprm, int i)
 233 {
 234 }
 235
 236 static void free_arg_pages(struct linux_binprm *bprm)
 237 {
 238 }
 239
 240 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 241                 struct page *page)
 242 {
 243         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 244 }
 245
 246 static int __bprm_mm_init(struct linux_binprm *bprm)
 247 {
 248         int err;
 249         struct vm_area_struct *vma = NULL;
 250         struct mm_struct *mm = bprm->mm;
 251
 252         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 253         if (!vma)
 254                 return -ENOMEM;
 255
 256         down_write(&mm->mmap_sem);
 257         vma->vm_mm = mm;
 258
 259         /*
 260          * Place the stack at the largest stack address the architecture
 261          * supports. Later, we'll move this to an appropriate place. We don't
 262          * use STACK_TOP because that can depend on attributes which aren't
 263          * configured yet.
 264          */
 265         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 266         vma->vm_end = STACK_TOP_MAX;
 267         vma->vm_start = vma->vm_end - PAGE_SIZE;
 268         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 269         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 270         INIT_LIST_HEAD(&vma->anon_vma_chain);
 271
 272         err = insert_vm_struct(mm, vma);
 273         if (err)
 274                 goto err;
 275
 276         mm->stack_vm = mm->total_vm = 1;
 277         up_write(&mm->mmap_sem);
 278         bprm->p = vma->vm_end - sizeof(void *);
 279         return 0;
 280 err:
 281         up_write(&mm->mmap_sem);
 282         bprm->vma = NULL;
 283         kmem_cache_free(vm_area_cachep, vma);
 284         return err;
 285 }
 286
 287 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 288 {
 289         return len <= MAX_ARG_STRLEN;
 290 }
 291
 292 #else
 293
 294 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 295 {
 296 }
 297
 298 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 299                 int write)
 300 {
 301         struct page *page;
 302
 303         page = bprm->page[pos / PAGE_SIZE];
 304         if (!page && write) {
 305                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 306                 if (!page)
 307                         return NULL;
 308                 bprm->page[pos / PAGE_SIZE] = page;
 309         }
 310
 311         return page;
 312 }
 313
 314 static void put_arg_page(struct page *page)
 315 {
 316 }
 317
 318 static void free_arg_page(struct linux_binprm *bprm, int i)
 319 {
 320         if (bprm->page[i]) {
 321                 __free_page(bprm->page[i]);
 322                 bprm->page[i] = NULL;
 323         }
 324 }
 325
 326 static void free_arg_pages(struct linux_binprm *bprm)
 327 {
 328         int i;
 329
 330         for (i = 0; i < MAX_ARG_PAGES; i++)
 331                 free_arg_page(bprm, i);
 332 }
 333
 334 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 335                 struct page *page)
 336 {
 337 }
 338
 339 static int __bprm_mm_init(struct linux_binprm *bprm)
 340 {
 341         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 342         return 0;
 343 }
 344
 345 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 346 {
 347         return len <= bprm->p;
 348 }
 349
 350 #endif /* CONFIG_MMU */
 351
 352 /*
 353  * Create a new mm_struct and populate it with a temporary stack
 354  * vm_area_struct.  We don't have enough context at this point to set the stack
 355  * flags, permissions, and offset, so we use temporary values.  We'll update
 356  * them later in setup_arg_pages().
 357  */
 358 static int bprm_mm_init(struct linux_binprm *bprm)
 359 {
 360         int err;
 361         struct mm_struct *mm = NULL;
 362
 363         bprm->mm = mm = mm_alloc();
 364         err = -ENOMEM;
 365         if (!mm)
 366                 goto err;
 367
 368         err = init_new_context(current, mm);
 369         if (err)
 370                 goto err;
 371
 372         err = __bprm_mm_init(bprm);
 373         if (err)
 374                 goto err;
 375
 376         return 0;
 377
 378 err:
 379         if (mm) {
 380                 bprm->mm = NULL;
 381                 mmdrop(mm);
 382         }
 383
 384         return err;
 385 }
 386
 387 struct user_arg_ptr {
 388 #ifdef CONFIG_COMPAT
 389         bool is_compat;
 390 #endif
 391         union {
 392                 const char __user *const __user *native;
 393 #ifdef CONFIG_COMPAT
 394                 const compat_uptr_t __user *compat;
 395 #endif
 396         } ptr;
 397 };
 398
 399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 400 {
 401         const char __user *native;
 402
 403 #ifdef CONFIG_COMPAT
 404         if (unlikely(argv.is_compat)) {
 405                 compat_uptr_t compat;
 406
 407                 if (get_user(compat, argv.ptr.compat + nr))
 408                         return ERR_PTR(-EFAULT);
 409
 410                 return compat_ptr(compat);
 411         }
 412 #endif
 413
 414         if (get_user(native, argv.ptr.native + nr))
 415                 return ERR_PTR(-EFAULT);
 416
 417         return native;
 418 }
 419
 420 /*
 421  * count() counts the number of strings in array ARGV.
 422  */
 423 static int count(struct user_arg_ptr argv, int max)
 424 {
 425         int i = 0;
 426
 427         if (argv.ptr.native != NULL) {
 428                 for (;;) {
 429                         const char __user *p = get_user_arg_ptr(argv, i);
 430
 431                         if (!p)
 432                                 break;
 433
 434                         if (IS_ERR(p))
 435                                 return -EFAULT;
 436
 437                         if (i >= max)
 438                                 return -E2BIG;
 439                         ++i;
 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, false))
 607                 return -ENOMEM;
 608
 609         lru_add_drain();
 610         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 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 : USER_PGTABLES_CEILING);
 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 : USER_PGTABLES_CEILING);
 626         }
 627         tlb_finish_mmu(&tlb, old_start, 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 */
 658         stack_base = rlimit_max(RLIMIT_STACK);
 659         if (stack_base > STACK_SIZE_MAX)
 660                 stack_base = STACK_SIZE_MAX;
 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         struct filename tmp = { .name = name };
 756         static const struct open_flags open_exec_flags = {
 757                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 758                 .acc_mode = MAY_EXEC | MAY_OPEN,
 759                 .intent = LOOKUP_OPEN
 760         };
 761
 762         file = do_filp_open(AT_FDCWD, &tmp, &open_exec_flags, LOOKUP_FOLLOW);
 763         if (IS_ERR(file))
 764                 goto out;
 765
 766         err = -EACCES;
 767         if (!S_ISREG(file_inode(file)->i_mode))
 768                 goto exit;
 769
 770         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 771                 goto exit;
 772
 773         fsnotify_open(file);
 774
 775         err = deny_write_access(file);
 776         if (err)
 777                 goto exit;
 778
 779 out:
 780         return file;
 781
 782 exit:
 783         fput(file);
 784         return ERR_PTR(err);
 785 }
 786 EXPORT_SYMBOL(open_exec);
 787
 788 int kernel_read(struct file *file, loff_t offset,
 789                 char *addr, unsigned long count)
 790 {
 791         mm_segment_t old_fs;
 792         loff_t pos = offset;
 793         int result;
 794
 795         old_fs = get_fs();
 796         set_fs(get_ds());
 797         /* The cast to a user pointer is valid due to the set_fs() */
 798         result = vfs_read(file, (void __user *)addr, count, &pos);
 799         set_fs(old_fs);
 800         return result;
 801 }
 802
 803 EXPORT_SYMBOL(kernel_read);
 804
 805 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 806 {
 807         ssize_t res = file->f_op->read(file, (void __user *)addr, len, &pos);
 808         if (res > 0)
 809                 flush_icache_range(addr, addr + len);
 810         return res;
 811 }
 812 EXPORT_SYMBOL(read_code);
 813
 814 static int exec_mmap(struct mm_struct *mm)
 815 {
 816         struct task_struct *tsk;
 817         struct mm_struct * old_mm, *active_mm;
 818
 819         /* Notify parent that we're no longer interested in the old VM */
 820         tsk = current;
 821         old_mm = current->mm;
 822         mm_release(tsk, old_mm);
 823
 824         if (old_mm) {
 825                 sync_mm_rss(old_mm);
 826                 /*
 827                  * Make sure that if there is a core dump in progress
 828                  * for the old mm, we get out and die instead of going
 829                  * through with the exec.  We must hold mmap_sem around
 830                  * checking core_state and changing tsk->mm.
 831                  */
 832                 down_read(&old_mm->mmap_sem);
 833                 if (unlikely(old_mm->core_state)) {
 834                         up_read(&old_mm->mmap_sem);
 835                         return -EINTR;
 836                 }
 837         }
 838         task_lock(tsk);
 839         active_mm = tsk->active_mm;
 840         tsk->mm = mm;
 841         tsk->active_mm = mm;
 842         activate_mm(active_mm, mm);
 843         task_unlock(tsk);
 844         arch_pick_mmap_layout(mm);
 845         if (old_mm) {
 846                 up_read(&old_mm->mmap_sem);
 847                 BUG_ON(active_mm != old_mm);
 848                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 849                 mm_update_next_owner(old_mm);
 850                 mmput(old_mm);
 851                 return 0;
 852         }
 853         mmdrop(active_mm);
 854         return 0;
 855 }
 856
 857 /*
 858  * This function makes sure the current process has its own signal table,
 859  * so that flush_signal_handlers can later reset the handlers without
 860  * disturbing other processes.  (Other processes might share the signal
 861  * table via the CLONE_SIGHAND option to clone().)
 862  */
 863 static int de_thread(struct task_struct *tsk)
 864 {
 865         struct signal_struct *sig = tsk->signal;
 866         struct sighand_struct *oldsighand = tsk->sighand;
 867         spinlock_t *lock = &oldsighand->siglock;
 868
 869         if (thread_group_empty(tsk))
 870                 goto no_thread_group;
 871
 872         /*
 873          * Kill all other threads in the thread group.
 874          */
 875         spin_lock_irq(lock);
 876         if (signal_group_exit(sig)) {
 877                 /*
 878                  * Another group action in progress, just
 879                  * return so that the signal is processed.
 880                  */
 881                 spin_unlock_irq(lock);
 882                 return -EAGAIN;
 883         }
 884
 885         sig->group_exit_task = tsk;
 886         sig->notify_count = zap_other_threads(tsk);
 887         if (!thread_group_leader(tsk))
 888                 sig->notify_count--;
 889
 890         while (sig->notify_count) {
 891                 __set_current_state(TASK_KILLABLE);
 892                 spin_unlock_irq(lock);
 893                 schedule();
 894                 if (unlikely(__fatal_signal_pending(tsk)))
 895                         goto killed;
 896                 spin_lock_irq(lock);
 897         }
 898         spin_unlock_irq(lock);
 899
 900         /*
 901          * At this point all other threads have exited, all we have to
 902          * do is to wait for the thread group leader to become inactive,
 903          * and to assume its PID:
 904          */
 905         if (!thread_group_leader(tsk)) {
 906                 struct task_struct *leader = tsk->group_leader;
 907
 908                 sig->notify_count = -1; /* for exit_notify() */
 909                 for (;;) {
 910                         threadgroup_change_begin(tsk);
 911                         write_lock_irq(&tasklist_lock);
 912                         if (likely(leader->exit_state))
 913                                 break;
 914                         __set_current_state(TASK_KILLABLE);
 915                         write_unlock_irq(&tasklist_lock);
 916                         threadgroup_change_end(tsk);
 917                         schedule();
 918                         if (unlikely(__fatal_signal_pending(tsk)))
 919                                 goto killed;
 920                 }
 921
 922                 /*
 923                  * The only record we have of the real-time age of a
 924                  * process, regardless of execs it's done, is start_time.
 925                  * All the past CPU time is accumulated in signal_struct
 926                  * from sister threads now dead.  But in this non-leader
 927                  * exec, nothing survives from the original leader thread,
 928                  * whose birth marks the true age of this process now.
 929                  * When we take on its identity by switching to its PID, we
 930                  * also take its birthdate (always earlier than our own).
 931                  */
 932                 tsk->start_time = leader->start_time;
 933
 934                 BUG_ON(!same_thread_group(leader, tsk));
 935                 BUG_ON(has_group_leader_pid(tsk));
 936                 /*
 937                  * An exec() starts a new thread group with the
 938                  * TGID of the previous thread group. Rehash the
 939                  * two threads with a switched PID, and release
 940                  * the former thread group leader:
 941                  */
 942
 943                 /* Become a process group leader with the old leader's pid.
 944                  * The old leader becomes a thread of the this thread group.
 945                  * Note: The old leader also uses this pid until release_task
 946                  *       is called.  Odd but simple and correct.
 947                  */
 948                 detach_pid(tsk, PIDTYPE_PID);
 949                 tsk->pid = leader->pid;
 950                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 951                 transfer_pid(leader, tsk, PIDTYPE_PGID);
 952                 transfer_pid(leader, tsk, PIDTYPE_SID);
 953
 954                 list_replace_rcu(&leader->tasks, &tsk->tasks);
 955                 list_replace_init(&leader->sibling, &tsk->sibling);
 956
 957                 tsk->group_leader = tsk;
 958                 leader->group_leader = tsk;
 959
 960                 tsk->exit_signal = SIGCHLD;
 961                 leader->exit_signal = -1;
 962
 963                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 964                 leader->exit_state = EXIT_DEAD;
 965
 966                 /*
 967                  * We are going to release_task()->ptrace_unlink() silently,
 968                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 969                  * the tracer wont't block again waiting for this thread.
 970                  */
 971                 if (unlikely(leader->ptrace))
 972                         __wake_up_parent(leader, leader->parent);
 973                 write_unlock_irq(&tasklist_lock);
 974                 threadgroup_change_end(tsk);
 975
 976                 release_task(leader);
 977         }
 978
 979         sig->group_exit_task = NULL;
 980         sig->notify_count = 0;
 981
 982 no_thread_group:
 983         /* we have changed execution domain */
 984         tsk->exit_signal = SIGCHLD;
 985
 986         exit_itimers(sig);
 987         flush_itimer_signals();
 988
 989         if (atomic_read(&oldsighand->count) != 1) {
 990                 struct sighand_struct *newsighand;
 991                 /*
 992                  * This ->sighand is shared with the CLONE_SIGHAND
 993                  * but not CLONE_THREAD task, switch to the new one.
 994                  */
 995                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 996                 if (!newsighand)
 997                         return -ENOMEM;
 998
 999                 atomic_set(&newsighand->count, 1);
1000                 memcpy(newsighand->action, oldsighand->action,
1001                        sizeof(newsighand->action));
1002
1003                 write_lock_irq(&tasklist_lock);
1004                 spin_lock(&oldsighand->siglock);
1005                 rcu_assign_pointer(tsk->sighand, newsighand);
1006                 spin_unlock(&oldsighand->siglock);
1007                 write_unlock_irq(&tasklist_lock);
1008
1009                 __cleanup_sighand(oldsighand);
1010         }
1011
1012         BUG_ON(!thread_group_leader(tsk));
1013         return 0;
1014
1015 killed:
1016         /* protects against exit_notify() and __exit_signal() */
1017         read_lock(&tasklist_lock);
1018         sig->group_exit_task = NULL;
1019         sig->notify_count = 0;
1020         read_unlock(&tasklist_lock);
1021         return -EAGAIN;
1022 }
1023
1024 char *get_task_comm(char *buf, struct task_struct *tsk)
1025 {
1026         /* buf must be at least sizeof(tsk->comm) in size */
1027         task_lock(tsk);
1028         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1029         task_unlock(tsk);
1030         return buf;
1031 }
1032 EXPORT_SYMBOL_GPL(get_task_comm);
1033
1034 /*
1035  * These functions flushes out all traces of the currently running executable
1036  * so that a new one can be started
1037  */
1038
1039 void set_task_comm(struct task_struct *tsk, char *buf)
1040 {
1041         task_lock(tsk);
1042         trace_task_rename(tsk, buf);
1043         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1044         task_unlock(tsk);
1045         perf_event_comm(tsk);
1046 }
1047
1048 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1049 {
1050         int i, ch;
1051
1052         /* Copies the binary name from after last slash */
1053         for (i = 0; (ch = *(fn++)) != '\0';) {
1054                 if (ch == '/')
1055                         i = 0; /* overwrite what we wrote */
1056                 else
1057                         if (i < len - 1)
1058                                 tcomm[i++] = ch;
1059         }
1060         tcomm[i] = '\0';
1061 }
1062
1063 int flush_old_exec(struct linux_binprm * bprm)
1064 {
1065         int retval;
1066
1067         /*
1068          * Make sure we have a private signal table and that
1069          * we are unassociated from the previous thread group.
1070          */
1071         retval = de_thread(current);
1072         if (retval)
1073                 goto out;
1074
1075         set_mm_exe_file(bprm->mm, bprm->file);
1076
1077         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1078         /*
1079          * Release all of the old mmap stuff
1080          */
1081         acct_arg_size(bprm, 0);
1082         retval = exec_mmap(bprm->mm);
1083         if (retval)
1084                 goto out;
1085
1086         bprm->mm = NULL;                /* We're using it now */
1087
1088         set_fs(USER_DS);
1089         current->flags &=
1090                 ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | PF_NOFREEZE);
1091         flush_thread();
1092         current->personality &= ~bprm->per_clear;
1093
1094         return 0;
1095
1096 out:
1097         return retval;
1098 }
1099 EXPORT_SYMBOL(flush_old_exec);
1100
1101 void would_dump(struct linux_binprm *bprm, struct file *file)
1102 {
1103         if (inode_permission(file_inode(file), MAY_READ) < 0)
1104                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1105 }
1106 EXPORT_SYMBOL(would_dump);
1107
1108 void setup_new_exec(struct linux_binprm * bprm)
1109 {
1110         arch_pick_mmap_layout(current->mm);
1111
1112         /* This is the point of no return */
1113         current->sas_ss_sp = current->sas_ss_size = 0;
1114
1115         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1116                 set_dumpable(current->mm, SUID_DUMP_USER);
1117         else
1118                 set_dumpable(current->mm, suid_dumpable);
1119
1120         set_task_comm(current, bprm->tcomm);
1121
1122         /* Set the new mm task size. We have to do that late because it may
1123          * depend on TIF_32BIT which is only updated in flush_thread() on
1124          * some architectures like powerpc
1125          */
1126         current->mm->task_size = TASK_SIZE;
1127
1128         /* install the new credentials */
1129         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1130             !gid_eq(bprm->cred->gid, current_egid())) {
1131                 current->pdeath_signal = 0;
1132         } else {
1133                 would_dump(bprm, bprm->file);
1134                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1135                         set_dumpable(current->mm, suid_dumpable);
1136         }
1137
1138         /* An exec changes our domain. We are no longer part of the thread
1139            group */
1140
1141         current->self_exec_id++;
1142
1143         flush_signal_handlers(current, 0);
1144         do_close_on_exec(current->files);
1145 }
1146 EXPORT_SYMBOL(setup_new_exec);
1147
1148 /*
1149  * Prepare credentials and lock ->cred_guard_mutex.
1150  * install_exec_creds() commits the new creds and drops the lock.
1151  * Or, if exec fails before, free_bprm() should release ->cred and
1152  * and unlock.
1153  */
1154 int prepare_bprm_creds(struct linux_binprm *bprm)
1155 {
1156         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1157                 return -ERESTARTNOINTR;
1158
1159         bprm->cred = prepare_exec_creds();
1160         if (likely(bprm->cred))
1161                 return 0;
1162
1163         mutex_unlock(&current->signal->cred_guard_mutex);
1164         return -ENOMEM;
1165 }
1166
1167 void free_bprm(struct linux_binprm *bprm)
1168 {
1169         free_arg_pages(bprm);
1170         if (bprm->cred) {
1171                 mutex_unlock(&current->signal->cred_guard_mutex);
1172                 abort_creds(bprm->cred);
1173         }
1174         /* If a binfmt changed the interp, free it. */
1175         if (bprm->interp != bprm->filename)
1176                 kfree(bprm->interp);
1177         kfree(bprm);
1178 }
1179
1180 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1181 {
1182         /* If a binfmt changed the interp, free it first. */
1183         if (bprm->interp != bprm->filename)
1184                 kfree(bprm->interp);
1185         bprm->interp = kstrdup(interp, GFP_KERNEL);
1186         if (!bprm->interp)
1187                 return -ENOMEM;
1188         return 0;
1189 }
1190 EXPORT_SYMBOL(bprm_change_interp);
1191
1192 /*
1193  * install the new credentials for this executable
1194  */
1195 void install_exec_creds(struct linux_binprm *bprm)
1196 {
1197         security_bprm_committing_creds(bprm);
1198
1199         commit_creds(bprm->cred);
1200         bprm->cred = NULL;
1201
1202         /*
1203          * Disable monitoring for regular users
1204          * when executing setuid binaries. Must
1205          * wait until new credentials are committed
1206          * by commit_creds() above
1207          */
1208         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1209                 perf_event_exit_task(current);
1210         /*
1211          * cred_guard_mutex must be held at least to this point to prevent
1212          * ptrace_attach() from altering our determination of the task's
1213          * credentials; any time after this it may be unlocked.
1214          */
1215         security_bprm_committed_creds(bprm);
1216         mutex_unlock(&current->signal->cred_guard_mutex);
1217 }
1218 EXPORT_SYMBOL(install_exec_creds);
1219
1220 /*
1221  * determine how safe it is to execute the proposed program
1222  * - the caller must hold ->cred_guard_mutex to protect against
1223  *   PTRACE_ATTACH or seccomp thread-sync
1224  */
1225 static int check_unsafe_exec(struct linux_binprm *bprm)
1226 {
1227         struct task_struct *p = current, *t;
1228         unsigned n_fs;
1229         int res = 0;
1230
1231         if (p->ptrace) {
1232                 if (p->ptrace & PT_PTRACE_CAP)
1233                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1234                 else
1235                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1236         }
1237
1238         /*
1239          * This isn't strictly necessary, but it makes it harder for LSMs to
1240          * mess up.
1241          */
1242         if (task_no_new_privs(current))
1243                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1244
1245         n_fs = 1;
1246         spin_lock(&p->fs->lock);
1247         rcu_read_lock();
1248         for (t = next_thread(p); t != p; t = next_thread(t)) {
1249                 if (t->fs == p->fs)
1250                         n_fs++;
1251         }
1252         rcu_read_unlock();
1253
1254         if (p->fs->users > n_fs) {
1255                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1256         } else {
1257                 res = -EAGAIN;
1258                 if (!p->fs->in_exec) {
1259                         p->fs->in_exec = 1;
1260                         res = 1;
1261                 }
1262         }
1263         spin_unlock(&p->fs->lock);
1264
1265         return res;
1266 }
1267
1268 static void bprm_fill_uid(struct linux_binprm *bprm)
1269 {
1270         struct inode *inode;
1271         unsigned int mode;
1272         kuid_t uid;
1273         kgid_t gid;
1274
1275         /* clear any previous set[ug]id data from a previous binary */
1276         bprm->cred->euid = current_euid();
1277         bprm->cred->egid = current_egid();
1278
1279         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1280                 return;
1281
1282         if (task_no_new_privs(current))
1283                 return;
1284
1285         inode = file_inode(bprm->file);
1286         mode = ACCESS_ONCE(inode->i_mode);
1287         if (!(mode & (S_ISUID|S_ISGID)))
1288                 return;
1289
1290         /* Be careful if suid/sgid is set */
1291         mutex_lock(&inode->i_mutex);
1292
1293         /* reload atomically mode/uid/gid now that lock held */
1294         mode = inode->i_mode;
1295         uid = inode->i_uid;
1296         gid = inode->i_gid;
1297         mutex_unlock(&inode->i_mutex);
1298
1299         /* We ignore suid/sgid if there are no mappings for them in the ns */
1300         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1301                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1302                 return;
1303
1304         if (mode & S_ISUID) {
1305                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1306                 bprm->cred->euid = uid;
1307         }
1308
1309         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1310                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1311                 bprm->cred->egid = gid;
1312         }
1313 }
1314
1315 /*
1316  * Fill the binprm structure from the inode.
1317  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1318  *
1319  * This may be called multiple times for binary chains (scripts for example).
1320  */
1321 int prepare_binprm(struct linux_binprm *bprm)
1322 {
1323         int retval;
1324
1325         if (bprm->file->f_op == NULL)
1326                 return -EACCES;
1327
1328         bprm_fill_uid(bprm);
1329
1330         /* fill in binprm security blob */
1331         retval = security_bprm_set_creds(bprm);
1332         if (retval)
1333                 return retval;
1334         bprm->cred_prepared = 1;
1335
1336         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1337         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1338 }
1339
1340 EXPORT_SYMBOL(prepare_binprm);
1341
1342 /*
1343  * Arguments are '\0' separated strings found at the location bprm->p
1344  * points to; chop off the first by relocating brpm->p to right after
1345  * the first '\0' encountered.
1346  */
1347 int remove_arg_zero(struct linux_binprm *bprm)
1348 {
1349         int ret = 0;
1350         unsigned long offset;
1351         char *kaddr;
1352         struct page *page;
1353
1354         if (!bprm->argc)
1355                 return 0;
1356
1357         do {
1358                 offset = bprm->p & ~PAGE_MASK;
1359                 page = get_arg_page(bprm, bprm->p, 0);
1360                 if (!page) {
1361                         ret = -EFAULT;
1362                         goto out;
1363                 }
1364                 kaddr = kmap_atomic(page);
1365
1366                 for (; offset < PAGE_SIZE && kaddr[offset];
1367                                 offset++, bprm->p++)
1368                         ;
1369
1370                 kunmap_atomic(kaddr);
1371                 put_arg_page(page);
1372
1373                 if (offset == PAGE_SIZE)
1374                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1375         } while (offset == PAGE_SIZE);
1376
1377         bprm->p++;
1378         bprm->argc--;
1379         ret = 0;
1380
1381 out:
1382         return ret;
1383 }
1384 EXPORT_SYMBOL(remove_arg_zero);
1385
1386 /*
1387  * cycle the list of binary formats handler, until one recognizes the image
1388  */
1389 int search_binary_handler(struct linux_binprm *bprm)
1390 {
1391         unsigned int depth = bprm->recursion_depth;
1392         int try,retval;
1393         struct linux_binfmt *fmt;
1394         pid_t old_pid, old_vpid;
1395
1396         /* This allows 4 levels of binfmt rewrites before failing hard. */
1397         if (depth > 5)
1398                 return -ELOOP;
1399
1400         retval = security_bprm_check(bprm);
1401         if (retval)
1402                 return retval;
1403
1404         retval = audit_bprm(bprm);
1405         if (retval)
1406                 return retval;
1407
1408         /* Need to fetch pid before load_binary changes it */
1409         old_pid = current->pid;
1410         rcu_read_lock();
1411         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1412         rcu_read_unlock();
1413
1414         retval = -ENOENT;
1415         for (try=0; try<2; try++) {
1416                 read_lock(&binfmt_lock);
1417                 list_for_each_entry(fmt, &formats, lh) {
1418                         int (*fn)(struct linux_binprm *) = fmt->load_binary;
1419                         if (!fn)
1420                                 continue;
1421                         if (!try_module_get(fmt->module))
1422                                 continue;
1423                         read_unlock(&binfmt_lock);
1424                         bprm->recursion_depth = depth + 1;
1425                         retval = fn(bprm);
1426                         bprm->recursion_depth = depth;
1427                         if (retval >= 0) {
1428                                 if (depth == 0) {
1429                                         trace_sched_process_exec(current, old_pid, bprm);
1430                                         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1431                                 }
1432                                 put_binfmt(fmt);
1433                                 allow_write_access(bprm->file);
1434                                 if (bprm->file)
1435                                         fput(bprm->file);
1436                                 bprm->file = NULL;
1437                                 current->did_exec = 1;
1438                                 proc_exec_connector(current);
1439                                 return retval;
1440                         }
1441                         read_lock(&binfmt_lock);
1442                         put_binfmt(fmt);
1443                         if (retval != -ENOEXEC || bprm->mm == NULL)
1444                                 break;
1445                         if (!bprm->file) {
1446                                 read_unlock(&binfmt_lock);
1447                                 return retval;
1448                         }
1449                 }
1450                 read_unlock(&binfmt_lock);
1451 #ifdef CONFIG_MODULES
1452                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1453                         break;
1454                 } else {
1455 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1456                         if (printable(bprm->buf[0]) &&
1457                             printable(bprm->buf[1]) &&
1458                             printable(bprm->buf[2]) &&
1459                             printable(bprm->buf[3]))
1460                                 break; /* -ENOEXEC */
1461                         if (try)
1462                                 break; /* -ENOEXEC */
1463                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1464                 }
1465 #else
1466                 break;
1467 #endif
1468         }
1469         return retval;
1470 }
1471
1472 EXPORT_SYMBOL(search_binary_handler);
1473
1474 /*
1475  * sys_execve() executes a new program.
1476  */
1477 static int do_execve_common(const char *filename,
1478                                 struct user_arg_ptr argv,
1479                                 struct user_arg_ptr envp)
1480 {
1481         struct linux_binprm *bprm;
1482         struct file *file;
1483         struct files_struct *displaced;
1484         bool clear_in_exec;
1485         int retval;
1486         const struct cred *cred = current_cred();
1487
1488         /*
1489          * We move the actual failure in case of RLIMIT_NPROC excess from
1490          * set*uid() to execve() because too many poorly written programs
1491          * don't check setuid() return code.  Here we additionally recheck
1492          * whether NPROC limit is still exceeded.
1493          */
1494         if ((current->flags & PF_NPROC_EXCEEDED) &&
1495             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1496                 retval = -EAGAIN;
1497                 goto out_ret;
1498         }
1499
1500         /* We're below the limit (still or again), so we don't want to make
1501          * further execve() calls fail. */
1502         current->flags &= ~PF_NPROC_EXCEEDED;
1503
1504         retval = unshare_files(&displaced);
1505         if (retval)
1506                 goto out_ret;
1507
1508         retval = -ENOMEM;
1509         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1510         if (!bprm)
1511                 goto out_files;
1512
1513         retval = prepare_bprm_creds(bprm);
1514         if (retval)
1515                 goto out_free;
1516
1517         retval = check_unsafe_exec(bprm);
1518         if (retval < 0)
1519                 goto out_free;
1520         clear_in_exec = retval;
1521         current->in_execve = 1;
1522
1523         file = open_exec(filename);
1524         retval = PTR_ERR(file);
1525         if (IS_ERR(file))
1526                 goto out_unmark;
1527
1528         sched_exec();
1529
1530         bprm->file = file;
1531         bprm->filename = filename;
1532         bprm->interp = filename;
1533
1534         retval = bprm_mm_init(bprm);
1535         if (retval)
1536                 goto out_file;
1537
1538         bprm->argc = count(argv, MAX_ARG_STRINGS);
1539         if ((retval = bprm->argc) < 0)
1540                 goto out;
1541
1542         bprm->envc = count(envp, MAX_ARG_STRINGS);
1543         if ((retval = bprm->envc) < 0)
1544                 goto out;
1545
1546         retval = prepare_binprm(bprm);
1547         if (retval < 0)
1548                 goto out;
1549
1550         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1551         if (retval < 0)
1552                 goto out;
1553
1554         bprm->exec = bprm->p;
1555         retval = copy_strings(bprm->envc, envp, bprm);
1556         if (retval < 0)
1557                 goto out;
1558
1559         retval = copy_strings(bprm->argc, argv, bprm);
1560         if (retval < 0)
1561                 goto out;
1562
1563         retval = search_binary_handler(bprm);
1564         if (retval < 0)
1565                 goto out;
1566
1567         /* execve succeeded */
1568         current->fs->in_exec = 0;
1569         current->in_execve = 0;
1570         acct_update_integrals(current);
1571         free_bprm(bprm);
1572         if (displaced)
1573                 put_files_struct(displaced);
1574         return retval;
1575
1576 out:
1577         if (bprm->mm) {
1578                 acct_arg_size(bprm, 0);
1579                 mmput(bprm->mm);
1580         }
1581
1582 out_file:
1583         if (bprm->file) {
1584                 allow_write_access(bprm->file);
1585                 fput(bprm->file);
1586         }
1587
1588 out_unmark:
1589         if (clear_in_exec)
1590                 current->fs->in_exec = 0;
1591         current->in_execve = 0;
1592
1593 out_free:
1594         free_bprm(bprm);
1595
1596 out_files:
1597         if (displaced)
1598                 reset_files_struct(displaced);
1599 out_ret:
1600         return retval;
1601 }
1602
1603 int do_execve(const char *filename,
1604         const char __user *const __user *__argv,
1605         const char __user *const __user *__envp)
1606 {
1607         struct user_arg_ptr argv = { .ptr.native = __argv };
1608         struct user_arg_ptr envp = { .ptr.native = __envp };
1609         return do_execve_common(filename, argv, envp);
1610 }
1611
1612 #ifdef CONFIG_COMPAT
1613 static int compat_do_execve(const char *filename,
1614         const compat_uptr_t __user *__argv,
1615         const compat_uptr_t __user *__envp)
1616 {
1617         struct user_arg_ptr argv = {
1618                 .is_compat = true,
1619                 .ptr.compat = __argv,
1620         };
1621         struct user_arg_ptr envp = {
1622                 .is_compat = true,
1623                 .ptr.compat = __envp,
1624         };
1625         return do_execve_common(filename, argv, envp);
1626 }
1627 #endif
1628
1629 void set_binfmt(struct linux_binfmt *new)
1630 {
1631         struct mm_struct *mm = current->mm;
1632
1633         if (mm->binfmt)
1634                 module_put(mm->binfmt->module);
1635
1636         mm->binfmt = new;
1637         if (new)
1638                 __module_get(new->module);
1639 }
1640
1641 EXPORT_SYMBOL(set_binfmt);
1642
1643 /*
1644  * set_dumpable converts traditional three-value dumpable to two flags and
1645  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1646  * these bits are not changed atomically.  So get_dumpable can observe the
1647  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1648  * return either old dumpable or new one by paying attention to the order of
1649  * modifying the bits.
1650  *
1651  * dumpable |   mm->flags (binary)
1652  * old  new | initial interim  final
1653  * ---------+-----------------------
1654  *  0    1  |   00      01      01
1655  *  0    2  |   00      10(*)   11
1656  *  1    0  |   01      00      00
1657  *  1    2  |   01      11      11
1658  *  2    0  |   11      10(*)   00
1659  *  2    1  |   11      11      01
1660  *
1661  * (*) get_dumpable regards interim value of 10 as 11.
1662  */
1663 void set_dumpable(struct mm_struct *mm, int value)
1664 {
1665         switch (value) {
1666         case SUID_DUMP_DISABLE:
1667                 clear_bit(MMF_DUMPABLE, &mm->flags);
1668                 smp_wmb();
1669                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1670                 break;
1671         case SUID_DUMP_USER:
1672                 set_bit(MMF_DUMPABLE, &mm->flags);
1673                 smp_wmb();
1674                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1675                 break;
1676         case SUID_DUMP_ROOT:
1677                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1678                 smp_wmb();
1679                 set_bit(MMF_DUMPABLE, &mm->flags);
1680                 break;
1681         }
1682 }
1683
1684 int __get_dumpable(unsigned long mm_flags)
1685 {
1686         int ret;
1687
1688         ret = mm_flags & MMF_DUMPABLE_MASK;
1689         return (ret > SUID_DUMP_USER) ? SUID_DUMP_ROOT : ret;
1690 }
1691
1692 /*
1693  * This returns the actual value of the suid_dumpable flag. For things
1694  * that are using this for checking for privilege transitions, it must
1695  * test against SUID_DUMP_USER rather than treating it as a boolean
1696  * value.
1697  */
1698 int get_dumpable(struct mm_struct *mm)
1699 {
1700         return __get_dumpable(mm->flags);
1701 }
1702
1703 SYSCALL_DEFINE3(execve,
1704                 const char __user *, filename,
1705                 const char __user *const __user *, argv,
1706                 const char __user *const __user *, envp)
1707 {
1708         struct filename *path = getname(filename);
1709         int error = PTR_ERR(path);
1710         if (!IS_ERR(path)) {
1711                 error = do_execve(path->name, argv, envp);
1712                 putname(path);
1713         }
1714         return error;
1715 }
1716 #ifdef CONFIG_COMPAT
1717 asmlinkage long compat_sys_execve(const char __user * filename,
1718         const compat_uptr_t __user * argv,
1719         const compat_uptr_t __user * envp)
1720 {
1721         struct filename *path = getname(filename);
1722         int error = PTR_ERR(path);
1723         if (!IS_ERR(path)) {
1724                 error = compat_do_execve(path->name, argv, envp);
1725                 putname(path);
1726         }
1727         return error;
1728 }
1729 #endif