linux/fs/exec.c
<<
>>
Prefs
   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
  69int suid_dumpable = 0;
  70
  71static LIST_HEAD(formats);
  72static DEFINE_RWLOCK(binfmt_lock);
  73
  74void __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
  83EXPORT_SYMBOL(__register_binfmt);
  84
  85void unregister_binfmt(struct linux_binfmt * fmt)
  86{
  87        write_lock(&binfmt_lock);
  88        list_del(&fmt->lh);
  89        write_unlock(&binfmt_lock);
  90}
  91
  92EXPORT_SYMBOL(unregister_binfmt);
  93
  94static 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 */
 105SYSCALL_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        }
 154exit:
 155        fput(file);
 156out:
 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 */
 167static 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
 179static 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
 227static void put_arg_page(struct page *page)
 228{
 229        put_page(page);
 230}
 231
 232static void free_arg_page(struct linux_binprm *bprm, int i)
 233{
 234}
 235
 236static void free_arg_pages(struct linux_binprm *bprm)
 237{
 238}
 239
 240static 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
 246static 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;
 280err:
 281        up_write(&mm->mmap_sem);
 282        bprm->vma = NULL;
 283        kmem_cache_free(vm_area_cachep, vma);
 284        return err;
 285}
 286
 287static bool valid_arg_len(struct linux_binprm *bprm, long len)
 288{
 289        return len <= MAX_ARG_STRLEN;
 290}
 291
 292#else
 293
 294static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 295{
 296}
 297
 298static 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
 314static void put_arg_page(struct page *page)
 315{
 316}
 317
 318static 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
 326static 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
 334static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 335                struct page *page)
 336{
 337}
 338
 339static int __bprm_mm_init(struct linux_binprm *bprm)
 340{
 341        bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 342        return 0;
 343}
 344
 345static 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 */
 358static 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
 378err:
 379        if (mm) {
 380                bprm->mm = NULL;
 381                mmdrop(mm);
 382        }
 383
 384        return err;
 385}
 386
 387struct 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
 399static 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 */
 423static 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 */
 454static 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;
 533out:
 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 */
 545int 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}
 560EXPORT_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 */
 576static 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, 0);
 611        if (new_end > old_start) {
 612                /*
 613                 * when the old and new regions overlap clear from new_end.
 614                 */
 615                free_pgd_range(&tlb, new_end, old_end, new_end,
 616                        vma->vm_next ? vma->vm_next->vm_start : 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, new_end, old_end);
 628
 629        /*
 630         * Shrink the vma to just the new range.  Always succeeds.
 631         */
 632        vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 633
 634        return 0;
 635}
 636
 637/*
 638 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 639 * the stack is optionally relocated, and some extra space is added.
 640 */
 641int setup_arg_pages(struct linux_binprm *bprm,
 642                    unsigned long stack_top,
 643                    int executable_stack)
 644{
 645        unsigned long ret;
 646        unsigned long stack_shift;
 647        struct mm_struct *mm = current->mm;
 648        struct vm_area_struct *vma = bprm->vma;
 649        struct vm_area_struct *prev = NULL;
 650        unsigned long vm_flags;
 651        unsigned long stack_base;
 652        unsigned long stack_size;
 653        unsigned long stack_expand;
 654        unsigned long rlim_stack;
 655
 656#ifdef CONFIG_STACK_GROWSUP
 657        /* Limit stack size to 1GB */
 658        stack_base = rlimit_max(RLIMIT_STACK);
 659        if (stack_base > (1 << 30))
 660                stack_base = 1 << 30;
 661
 662        /* Make sure we didn't let the argument array grow too large. */
 663        if (vma->vm_end - vma->vm_start > stack_base)
 664                return -ENOMEM;
 665
 666        stack_base = PAGE_ALIGN(stack_top - stack_base);
 667
 668        stack_shift = vma->vm_start - stack_base;
 669        mm->arg_start = bprm->p - stack_shift;
 670        bprm->p = vma->vm_end - stack_shift;
 671#else
 672        stack_top = arch_align_stack(stack_top);
 673        stack_top = PAGE_ALIGN(stack_top);
 674
 675        if (unlikely(stack_top < mmap_min_addr) ||
 676            unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 677                return -ENOMEM;
 678
 679        stack_shift = vma->vm_end - stack_top;
 680
 681        bprm->p -= stack_shift;
 682        mm->arg_start = bprm->p;
 683#endif
 684
 685        if (bprm->loader)
 686                bprm->loader -= stack_shift;
 687        bprm->exec -= stack_shift;
 688
 689        down_write(&mm->mmap_sem);
 690        vm_flags = VM_STACK_FLAGS;
 691
 692        /*
 693         * Adjust stack execute permissions; explicitly enable for
 694         * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 695         * (arch default) otherwise.
 696         */
 697        if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 698                vm_flags |= VM_EXEC;
 699        else if (executable_stack == EXSTACK_DISABLE_X)
 700                vm_flags &= ~VM_EXEC;
 701        vm_flags |= mm->def_flags;
 702        vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 703
 704        ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 705                        vm_flags);
 706        if (ret)
 707                goto out_unlock;
 708        BUG_ON(prev != vma);
 709
 710        /* Move stack pages down in memory. */
 711        if (stack_shift) {
 712                ret = shift_arg_pages(vma, stack_shift);
 713                if (ret)
 714                        goto out_unlock;
 715        }
 716
 717        /* mprotect_fixup is overkill to remove the temporary stack flags */
 718        vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 719
 720        stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 721        stack_size = vma->vm_end - vma->vm_start;
 722        /*
 723         * Align this down to a page boundary as expand_stack
 724         * will align it up.
 725         */
 726        rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 727#ifdef CONFIG_STACK_GROWSUP
 728        if (stack_size + stack_expand > rlim_stack)
 729                stack_base = vma->vm_start + rlim_stack;
 730        else
 731                stack_base = vma->vm_end + stack_expand;
 732#else
 733        if (stack_size + stack_expand > rlim_stack)
 734                stack_base = vma->vm_end - rlim_stack;
 735        else
 736                stack_base = vma->vm_start - stack_expand;
 737#endif
 738        current->mm->start_stack = bprm->p;
 739        ret = expand_stack(vma, stack_base);
 740        if (ret)
 741                ret = -EFAULT;
 742
 743out_unlock:
 744        up_write(&mm->mmap_sem);
 745        return ret;
 746}
 747EXPORT_SYMBOL(setup_arg_pages);
 748
 749#endif /* CONFIG_MMU */
 750
 751struct 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
 779out:
 780        return file;
 781
 782exit:
 783        fput(file);
 784        return ERR_PTR(err);
 785}
 786EXPORT_SYMBOL(open_exec);
 787
 788int 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
 803EXPORT_SYMBOL(kernel_read);
 804
 805static int exec_mmap(struct mm_struct *mm)
 806{
 807        struct task_struct *tsk;
 808        struct mm_struct * old_mm, *active_mm;
 809
 810        /* Notify parent that we're no longer interested in the old VM */
 811        tsk = current;
 812        old_mm = current->mm;
 813        mm_release(tsk, old_mm);
 814
 815        if (old_mm) {
 816                sync_mm_rss(old_mm);
 817                /*
 818                 * Make sure that if there is a core dump in progress
 819                 * for the old mm, we get out and die instead of going
 820                 * through with the exec.  We must hold mmap_sem around
 821                 * checking core_state and changing tsk->mm.
 822                 */
 823                down_read(&old_mm->mmap_sem);
 824                if (unlikely(old_mm->core_state)) {
 825                        up_read(&old_mm->mmap_sem);
 826                        return -EINTR;
 827                }
 828        }
 829        task_lock(tsk);
 830        active_mm = tsk->active_mm;
 831        tsk->mm = mm;
 832        tsk->active_mm = mm;
 833        activate_mm(active_mm, mm);
 834        task_unlock(tsk);
 835        arch_pick_mmap_layout(mm);
 836        if (old_mm) {
 837                up_read(&old_mm->mmap_sem);
 838                BUG_ON(active_mm != old_mm);
 839                setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 840                mm_update_next_owner(old_mm);
 841                mmput(old_mm);
 842                return 0;
 843        }
 844        mmdrop(active_mm);
 845        return 0;
 846}
 847
 848/*
 849 * This function makes sure the current process has its own signal table,
 850 * so that flush_signal_handlers can later reset the handlers without
 851 * disturbing other processes.  (Other processes might share the signal
 852 * table via the CLONE_SIGHAND option to clone().)
 853 */
 854static int de_thread(struct task_struct *tsk)
 855{
 856        struct signal_struct *sig = tsk->signal;
 857        struct sighand_struct *oldsighand = tsk->sighand;
 858        spinlock_t *lock = &oldsighand->siglock;
 859
 860        if (thread_group_empty(tsk))
 861                goto no_thread_group;
 862
 863        /*
 864         * Kill all other threads in the thread group.
 865         */
 866        spin_lock_irq(lock);
 867        if (signal_group_exit(sig)) {
 868                /*
 869                 * Another group action in progress, just
 870                 * return so that the signal is processed.
 871                 */
 872                spin_unlock_irq(lock);
 873                return -EAGAIN;
 874        }
 875
 876        sig->group_exit_task = tsk;
 877        sig->notify_count = zap_other_threads(tsk);
 878        if (!thread_group_leader(tsk))
 879                sig->notify_count--;
 880
 881        while (sig->notify_count) {
 882                __set_current_state(TASK_KILLABLE);
 883                spin_unlock_irq(lock);
 884                schedule();
 885                if (unlikely(__fatal_signal_pending(tsk)))
 886                        goto killed;
 887                spin_lock_irq(lock);
 888        }
 889        spin_unlock_irq(lock);
 890
 891        /*
 892         * At this point all other threads have exited, all we have to
 893         * do is to wait for the thread group leader to become inactive,
 894         * and to assume its PID:
 895         */
 896        if (!thread_group_leader(tsk)) {
 897                struct task_struct *leader = tsk->group_leader;
 898
 899                sig->notify_count = -1; /* for exit_notify() */
 900                for (;;) {
 901                        threadgroup_change_begin(tsk);
 902                        write_lock_irq(&tasklist_lock);
 903                        if (likely(leader->exit_state))
 904                                break;
 905                        __set_current_state(TASK_KILLABLE);
 906                        write_unlock_irq(&tasklist_lock);
 907                        threadgroup_change_end(tsk);
 908                        schedule();
 909                        if (unlikely(__fatal_signal_pending(tsk)))
 910                                goto killed;
 911                }
 912
 913                /*
 914                 * The only record we have of the real-time age of a
 915                 * process, regardless of execs it's done, is start_time.
 916                 * All the past CPU time is accumulated in signal_struct
 917                 * from sister threads now dead.  But in this non-leader
 918                 * exec, nothing survives from the original leader thread,
 919                 * whose birth marks the true age of this process now.
 920                 * When we take on its identity by switching to its PID, we
 921                 * also take its birthdate (always earlier than our own).
 922                 */
 923                tsk->start_time = leader->start_time;
 924
 925                BUG_ON(!same_thread_group(leader, tsk));
 926                BUG_ON(has_group_leader_pid(tsk));
 927                /*
 928                 * An exec() starts a new thread group with the
 929                 * TGID of the previous thread group. Rehash the
 930                 * two threads with a switched PID, and release
 931                 * the former thread group leader:
 932                 */
 933
 934                /* Become a process group leader with the old leader's pid.
 935                 * The old leader becomes a thread of the this thread group.
 936                 * Note: The old leader also uses this pid until release_task
 937                 *       is called.  Odd but simple and correct.
 938                 */
 939                detach_pid(tsk, PIDTYPE_PID);
 940                tsk->pid = leader->pid;
 941                attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
 942                transfer_pid(leader, tsk, PIDTYPE_PGID);
 943                transfer_pid(leader, tsk, PIDTYPE_SID);
 944
 945                list_replace_rcu(&leader->tasks, &tsk->tasks);
 946                list_replace_init(&leader->sibling, &tsk->sibling);
 947
 948                tsk->group_leader = tsk;
 949                leader->group_leader = tsk;
 950
 951                tsk->exit_signal = SIGCHLD;
 952                leader->exit_signal = -1;
 953
 954                BUG_ON(leader->exit_state != EXIT_ZOMBIE);
 955                leader->exit_state = EXIT_DEAD;
 956
 957                /*
 958                 * We are going to release_task()->ptrace_unlink() silently,
 959                 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
 960                 * the tracer wont't block again waiting for this thread.
 961                 */
 962                if (unlikely(leader->ptrace))
 963                        __wake_up_parent(leader, leader->parent);
 964                write_unlock_irq(&tasklist_lock);
 965                threadgroup_change_end(tsk);
 966
 967                release_task(leader);
 968        }
 969
 970        sig->group_exit_task = NULL;
 971        sig->notify_count = 0;
 972
 973no_thread_group:
 974        /* we have changed execution domain */
 975        tsk->exit_signal = SIGCHLD;
 976
 977        exit_itimers(sig);
 978        flush_itimer_signals();
 979
 980        if (atomic_read(&oldsighand->count) != 1) {
 981                struct sighand_struct *newsighand;
 982                /*
 983                 * This ->sighand is shared with the CLONE_SIGHAND
 984                 * but not CLONE_THREAD task, switch to the new one.
 985                 */
 986                newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
 987                if (!newsighand)
 988                        return -ENOMEM;
 989
 990                atomic_set(&newsighand->count, 1);
 991                memcpy(newsighand->action, oldsighand->action,
 992                       sizeof(newsighand->action));
 993
 994                write_lock_irq(&tasklist_lock);
 995                spin_lock(&oldsighand->siglock);
 996                rcu_assign_pointer(tsk->sighand, newsighand);
 997                spin_unlock(&oldsighand->siglock);
 998                write_unlock_irq(&tasklist_lock);
 999
1000                __cleanup_sighand(oldsighand);
1001        }
1002
1003        BUG_ON(!thread_group_leader(tsk));
1004        return 0;
1005
1006killed:
1007        /* protects against exit_notify() and __exit_signal() */
1008        read_lock(&tasklist_lock);
1009        sig->group_exit_task = NULL;
1010        sig->notify_count = 0;
1011        read_unlock(&tasklist_lock);
1012        return -EAGAIN;
1013}
1014
1015char *get_task_comm(char *buf, struct task_struct *tsk)
1016{
1017        /* buf must be at least sizeof(tsk->comm) in size */
1018        task_lock(tsk);
1019        strncpy(buf, tsk->comm, sizeof(tsk->comm));
1020        task_unlock(tsk);
1021        return buf;
1022}
1023EXPORT_SYMBOL_GPL(get_task_comm);
1024
1025/*
1026 * These functions flushes out all traces of the currently running executable
1027 * so that a new one can be started
1028 */
1029
1030void set_task_comm(struct task_struct *tsk, char *buf)
1031{
1032        task_lock(tsk);
1033
1034        trace_task_rename(tsk, buf);
1035
1036        /*
1037         * Threads may access current->comm without holding
1038         * the task lock, so write the string carefully.
1039         * Readers without a lock may see incomplete new
1040         * names but are safe from non-terminating string reads.
1041         */
1042        memset(tsk->comm, 0, TASK_COMM_LEN);
1043        wmb();
1044        strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1045        task_unlock(tsk);
1046        perf_event_comm(tsk);
1047}
1048
1049static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1050{
1051        int i, ch;
1052
1053        /* Copies the binary name from after last slash */
1054        for (i = 0; (ch = *(fn++)) != '\0';) {
1055                if (ch == '/')
1056                        i = 0; /* overwrite what we wrote */
1057                else
1058                        if (i < len - 1)
1059                                tcomm[i++] = ch;
1060        }
1061        tcomm[i] = '\0';
1062}
1063
1064int flush_old_exec(struct linux_binprm * bprm)
1065{
1066        int retval;
1067
1068        /*
1069         * Make sure we have a private signal table and that
1070         * we are unassociated from the previous thread group.
1071         */
1072        retval = de_thread(current);
1073        if (retval)
1074                goto out;
1075
1076        set_mm_exe_file(bprm->mm, bprm->file);
1077
1078        filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1079        /*
1080         * Release all of the old mmap stuff
1081         */
1082        acct_arg_size(bprm, 0);
1083        retval = exec_mmap(bprm->mm);
1084        if (retval)
1085                goto out;
1086
1087        bprm->mm = NULL;                /* We're using it now */
1088
1089        set_fs(USER_DS);
1090        current->flags &=
1091                ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | PF_NOFREEZE);
1092        flush_thread();
1093        current->personality &= ~bprm->per_clear;
1094
1095        return 0;
1096
1097out:
1098        return retval;
1099}
1100EXPORT_SYMBOL(flush_old_exec);
1101
1102void would_dump(struct linux_binprm *bprm, struct file *file)
1103{
1104        if (inode_permission(file_inode(file), MAY_READ) < 0)
1105                bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1106}
1107EXPORT_SYMBOL(would_dump);
1108
1109void setup_new_exec(struct linux_binprm * bprm)
1110{
1111        arch_pick_mmap_layout(current->mm);
1112
1113        /* This is the point of no return */
1114        current->sas_ss_sp = current->sas_ss_size = 0;
1115
1116        if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1117                set_dumpable(current->mm, SUID_DUMP_USER);
1118        else
1119                set_dumpable(current->mm, suid_dumpable);
1120
1121        set_task_comm(current, bprm->tcomm);
1122
1123        /* Set the new mm task size. We have to do that late because it may
1124         * depend on TIF_32BIT which is only updated in flush_thread() on
1125         * some architectures like powerpc
1126         */
1127        current->mm->task_size = TASK_SIZE;
1128
1129        /* install the new credentials */
1130        if (!uid_eq(bprm->cred->uid, current_euid()) ||
1131            !gid_eq(bprm->cred->gid, current_egid())) {
1132                current->pdeath_signal = 0;
1133        } else {
1134                would_dump(bprm, bprm->file);
1135                if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1136                        set_dumpable(current->mm, suid_dumpable);
1137        }
1138
1139        /*
1140         * Flush performance counters when crossing a
1141         * security domain:
1142         */
1143        if (!get_dumpable(current->mm))
1144                perf_event_exit_task(current);
1145
1146        /* An exec changes our domain. We are no longer part of the thread
1147           group */
1148
1149        current->self_exec_id++;
1150                        
1151        flush_signal_handlers(current, 0);
1152        do_close_on_exec(current->files);
1153}
1154EXPORT_SYMBOL(setup_new_exec);
1155
1156/*
1157 * Prepare credentials and lock ->cred_guard_mutex.
1158 * install_exec_creds() commits the new creds and drops the lock.
1159 * Or, if exec fails before, free_bprm() should release ->cred and
1160 * and unlock.
1161 */
1162int prepare_bprm_creds(struct linux_binprm *bprm)
1163{
1164        if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1165                return -ERESTARTNOINTR;
1166
1167        bprm->cred = prepare_exec_creds();
1168        if (likely(bprm->cred))
1169                return 0;
1170
1171        mutex_unlock(&current->signal->cred_guard_mutex);
1172        return -ENOMEM;
1173}
1174
1175void free_bprm(struct linux_binprm *bprm)
1176{
1177        free_arg_pages(bprm);
1178        if (bprm->cred) {
1179                mutex_unlock(&current->signal->cred_guard_mutex);
1180                abort_creds(bprm->cred);
1181        }
1182        /* If a binfmt changed the interp, free it. */
1183        if (bprm->interp != bprm->filename)
1184                kfree(bprm->interp);
1185        kfree(bprm);
1186}
1187
1188int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1189{
1190        /* If a binfmt changed the interp, free it first. */
1191        if (bprm->interp != bprm->filename)
1192                kfree(bprm->interp);
1193        bprm->interp = kstrdup(interp, GFP_KERNEL);
1194        if (!bprm->interp)
1195                return -ENOMEM;
1196        return 0;
1197}
1198EXPORT_SYMBOL(bprm_change_interp);
1199
1200/*
1201 * install the new credentials for this executable
1202 */
1203void install_exec_creds(struct linux_binprm *bprm)
1204{
1205        security_bprm_committing_creds(bprm);
1206
1207        commit_creds(bprm->cred);
1208        bprm->cred = NULL;
1209        /*
1210         * cred_guard_mutex must be held at least to this point to prevent
1211         * ptrace_attach() from altering our determination of the task's
1212         * credentials; any time after this it may be unlocked.
1213         */
1214        security_bprm_committed_creds(bprm);
1215        mutex_unlock(&current->signal->cred_guard_mutex);
1216}
1217EXPORT_SYMBOL(install_exec_creds);
1218
1219/*
1220 * determine how safe it is to execute the proposed program
1221 * - the caller must hold ->cred_guard_mutex to protect against
1222 *   PTRACE_ATTACH
1223 */
1224static int check_unsafe_exec(struct linux_binprm *bprm)
1225{
1226        struct task_struct *p = current, *t;
1227        unsigned n_fs;
1228        int res = 0;
1229
1230        if (p->ptrace) {
1231                if (p->ptrace & PT_PTRACE_CAP)
1232                        bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1233                else
1234                        bprm->unsafe |= LSM_UNSAFE_PTRACE;
1235        }
1236
1237        /*
1238         * This isn't strictly necessary, but it makes it harder for LSMs to
1239         * mess up.
1240         */
1241        if (current->no_new_privs)
1242                bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1243
1244        n_fs = 1;
1245        spin_lock(&p->fs->lock);
1246        rcu_read_lock();
1247        for (t = next_thread(p); t != p; t = next_thread(t)) {
1248                if (t->fs == p->fs)
1249                        n_fs++;
1250        }
1251        rcu_read_unlock();
1252
1253        if (p->fs->users > n_fs) {
1254                bprm->unsafe |= LSM_UNSAFE_SHARE;
1255        } else {
1256                res = -EAGAIN;
1257                if (!p->fs->in_exec) {
1258                        p->fs->in_exec = 1;
1259                        res = 1;
1260                }
1261        }
1262        spin_unlock(&p->fs->lock);
1263
1264        return res;
1265}
1266
1267/* 
1268 * Fill the binprm structure from the inode. 
1269 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1270 *
1271 * This may be called multiple times for binary chains (scripts for example).
1272 */
1273int prepare_binprm(struct linux_binprm *bprm)
1274{
1275        umode_t mode;
1276        struct inode * inode = file_inode(bprm->file);
1277        int retval;
1278
1279        mode = inode->i_mode;
1280        if (bprm->file->f_op == NULL)
1281                return -EACCES;
1282
1283        /* clear any previous set[ug]id data from a previous binary */
1284        bprm->cred->euid = current_euid();
1285        bprm->cred->egid = current_egid();
1286
1287        if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1288            !current->no_new_privs &&
1289            kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1290            kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1291                /* Set-uid? */
1292                if (mode & S_ISUID) {
1293                        bprm->per_clear |= PER_CLEAR_ON_SETID;
1294                        bprm->cred->euid = inode->i_uid;
1295                }
1296
1297                /* Set-gid? */
1298                /*
1299                 * If setgid is set but no group execute bit then this
1300                 * is a candidate for mandatory locking, not a setgid
1301                 * executable.
1302                 */
1303                if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1304                        bprm->per_clear |= PER_CLEAR_ON_SETID;
1305                        bprm->cred->egid = inode->i_gid;
1306                }
1307        }
1308
1309        /* fill in binprm security blob */
1310        retval = security_bprm_set_creds(bprm);
1311        if (retval)
1312                return retval;
1313        bprm->cred_prepared = 1;
1314
1315        memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1316        return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1317}
1318
1319EXPORT_SYMBOL(prepare_binprm);
1320
1321/*
1322 * Arguments are '\0' separated strings found at the location bprm->p
1323 * points to; chop off the first by relocating brpm->p to right after
1324 * the first '\0' encountered.
1325 */
1326int remove_arg_zero(struct linux_binprm *bprm)
1327{
1328        int ret = 0;
1329        unsigned long offset;
1330        char *kaddr;
1331        struct page *page;
1332
1333        if (!bprm->argc)
1334                return 0;
1335
1336        do {
1337                offset = bprm->p & ~PAGE_MASK;
1338                page = get_arg_page(bprm, bprm->p, 0);
1339                if (!page) {
1340                        ret = -EFAULT;
1341                        goto out;
1342                }
1343                kaddr = kmap_atomic(page);
1344
1345                for (; offset < PAGE_SIZE && kaddr[offset];
1346                                offset++, bprm->p++)
1347                        ;
1348
1349                kunmap_atomic(kaddr);
1350                put_arg_page(page);
1351
1352                if (offset == PAGE_SIZE)
1353                        free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1354        } while (offset == PAGE_SIZE);
1355
1356        bprm->p++;
1357        bprm->argc--;
1358        ret = 0;
1359
1360out:
1361        return ret;
1362}
1363EXPORT_SYMBOL(remove_arg_zero);
1364
1365/*
1366 * cycle the list of binary formats handler, until one recognizes the image
1367 */
1368int search_binary_handler(struct linux_binprm *bprm)
1369{
1370        unsigned int depth = bprm->recursion_depth;
1371        int try,retval;
1372        struct linux_binfmt *fmt;
1373        pid_t old_pid, old_vpid;
1374
1375        /* This allows 4 levels of binfmt rewrites before failing hard. */
1376        if (depth > 5)
1377                return -ELOOP;
1378
1379        retval = security_bprm_check(bprm);
1380        if (retval)
1381                return retval;
1382
1383        retval = audit_bprm(bprm);
1384        if (retval)
1385                return retval;
1386
1387        /* Need to fetch pid before load_binary changes it */
1388        old_pid = current->pid;
1389        rcu_read_lock();
1390        old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1391        rcu_read_unlock();
1392
1393        retval = -ENOENT;
1394        for (try=0; try<2; try++) {
1395                read_lock(&binfmt_lock);
1396                list_for_each_entry(fmt, &formats, lh) {
1397                        int (*fn)(struct linux_binprm *) = fmt->load_binary;
1398                        if (!fn)
1399                                continue;
1400                        if (!try_module_get(fmt->module))
1401                                continue;
1402                        read_unlock(&binfmt_lock);
1403                        bprm->recursion_depth = depth + 1;
1404                        retval = fn(bprm);
1405                        bprm->recursion_depth = depth;
1406                        if (retval >= 0) {
1407                                if (depth == 0) {
1408                                        trace_sched_process_exec(current, old_pid, bprm);
1409                                        ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1410                                }
1411                                put_binfmt(fmt);
1412                                allow_write_access(bprm->file);
1413                                if (bprm->file)
1414                                        fput(bprm->file);
1415                                bprm->file = NULL;
1416                                current->did_exec = 1;
1417                                proc_exec_connector(current);
1418                                return retval;
1419                        }
1420                        read_lock(&binfmt_lock);
1421                        put_binfmt(fmt);
1422                        if (retval != -ENOEXEC || bprm->mm == NULL)
1423                                break;
1424                        if (!bprm->file) {
1425                                read_unlock(&binfmt_lock);
1426                                return retval;
1427                        }
1428                }
1429                read_unlock(&binfmt_lock);
1430#ifdef CONFIG_MODULES
1431                if (retval != -ENOEXEC || bprm->mm == NULL) {
1432                        break;
1433                } else {
1434#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1435                        if (printable(bprm->buf[0]) &&
1436                            printable(bprm->buf[1]) &&
1437                            printable(bprm->buf[2]) &&
1438                            printable(bprm->buf[3]))
1439                                break; /* -ENOEXEC */
1440                        if (try)
1441                                break; /* -ENOEXEC */
1442                        request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1443                }
1444#else
1445                break;
1446#endif
1447        }
1448        return retval;
1449}
1450
1451EXPORT_SYMBOL(search_binary_handler);
1452
1453/*
1454 * sys_execve() executes a new program.
1455 */
1456static int do_execve_common(const char *filename,
1457                                struct user_arg_ptr argv,
1458                                struct user_arg_ptr envp)
1459{
1460        struct linux_binprm *bprm;
1461        struct file *file;
1462        struct files_struct *displaced;
1463        bool clear_in_exec;
1464        int retval;
1465        const struct cred *cred = current_cred();
1466
1467        /*
1468         * We move the actual failure in case of RLIMIT_NPROC excess from
1469         * set*uid() to execve() because too many poorly written programs
1470         * don't check setuid() return code.  Here we additionally recheck
1471         * whether NPROC limit is still exceeded.
1472         */
1473        if ((current->flags & PF_NPROC_EXCEEDED) &&
1474            atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1475                retval = -EAGAIN;
1476                goto out_ret;
1477        }
1478
1479        /* We're below the limit (still or again), so we don't want to make
1480         * further execve() calls fail. */
1481        current->flags &= ~PF_NPROC_EXCEEDED;
1482
1483        retval = unshare_files(&displaced);
1484        if (retval)
1485                goto out_ret;
1486
1487        retval = -ENOMEM;
1488        bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1489        if (!bprm)
1490                goto out_files;
1491
1492        retval = prepare_bprm_creds(bprm);
1493        if (retval)
1494                goto out_free;
1495
1496        retval = check_unsafe_exec(bprm);
1497        if (retval < 0)
1498                goto out_free;
1499        clear_in_exec = retval;
1500        current->in_execve = 1;
1501
1502        file = open_exec(filename);
1503        retval = PTR_ERR(file);
1504        if (IS_ERR(file))
1505                goto out_unmark;
1506
1507        sched_exec();
1508
1509        bprm->file = file;
1510        bprm->filename = filename;
1511        bprm->interp = filename;
1512
1513        retval = bprm_mm_init(bprm);
1514        if (retval)
1515                goto out_file;
1516
1517        bprm->argc = count(argv, MAX_ARG_STRINGS);
1518        if ((retval = bprm->argc) < 0)
1519                goto out;
1520
1521        bprm->envc = count(envp, MAX_ARG_STRINGS);
1522        if ((retval = bprm->envc) < 0)
1523                goto out;
1524
1525        retval = prepare_binprm(bprm);
1526        if (retval < 0)
1527                goto out;
1528
1529        retval = copy_strings_kernel(1, &bprm->filename, bprm);
1530        if (retval < 0)
1531                goto out;
1532
1533        bprm->exec = bprm->p;
1534        retval = copy_strings(bprm->envc, envp, bprm);
1535        if (retval < 0)
1536                goto out;
1537
1538        retval = copy_strings(bprm->argc, argv, bprm);
1539        if (retval < 0)
1540                goto out;
1541
1542        retval = search_binary_handler(bprm);
1543        if (retval < 0)
1544                goto out;
1545
1546        /* execve succeeded */
1547        current->fs->in_exec = 0;
1548        current->in_execve = 0;
1549        acct_update_integrals(current);
1550        free_bprm(bprm);
1551        if (displaced)
1552                put_files_struct(displaced);
1553        return retval;
1554
1555out:
1556        if (bprm->mm) {
1557                acct_arg_size(bprm, 0);
1558                mmput(bprm->mm);
1559        }
1560
1561out_file:
1562        if (bprm->file) {
1563                allow_write_access(bprm->file);
1564                fput(bprm->file);
1565        }
1566
1567out_unmark:
1568        if (clear_in_exec)
1569                current->fs->in_exec = 0;
1570        current->in_execve = 0;
1571
1572out_free:
1573        free_bprm(bprm);
1574
1575out_files:
1576        if (displaced)
1577                reset_files_struct(displaced);
1578out_ret:
1579        return retval;
1580}
1581
1582int do_execve(const char *filename,
1583        const char __user *const __user *__argv,
1584        const char __user *const __user *__envp)
1585{
1586        struct user_arg_ptr argv = { .ptr.native = __argv };
1587        struct user_arg_ptr envp = { .ptr.native = __envp };
1588        return do_execve_common(filename, argv, envp);
1589}
1590
1591#ifdef CONFIG_COMPAT
1592static int compat_do_execve(const char *filename,
1593        const compat_uptr_t __user *__argv,
1594        const compat_uptr_t __user *__envp)
1595{
1596        struct user_arg_ptr argv = {
1597                .is_compat = true,
1598                .ptr.compat = __argv,
1599        };
1600        struct user_arg_ptr envp = {
1601                .is_compat = true,
1602                .ptr.compat = __envp,
1603        };
1604        return do_execve_common(filename, argv, envp);
1605}
1606#endif
1607
1608void set_binfmt(struct linux_binfmt *new)
1609{
1610        struct mm_struct *mm = current->mm;
1611
1612        if (mm->binfmt)
1613                module_put(mm->binfmt->module);
1614
1615        mm->binfmt = new;
1616        if (new)
1617                __module_get(new->module);
1618}
1619
1620EXPORT_SYMBOL(set_binfmt);
1621
1622/*
1623 * set_dumpable converts traditional three-value dumpable to two flags and
1624 * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1625 * these bits are not changed atomically.  So get_dumpable can observe the
1626 * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1627 * return either old dumpable or new one by paying attention to the order of
1628 * modifying the bits.
1629 *
1630 * dumpable |   mm->flags (binary)
1631 * old  new | initial interim  final
1632 * ---------+-----------------------
1633 *  0    1  |   00      01      01
1634 *  0    2  |   00      10(*)   11
1635 *  1    0  |   01      00      00
1636 *  1    2  |   01      11      11
1637 *  2    0  |   11      10(*)   00
1638 *  2    1  |   11      11      01
1639 *
1640 * (*) get_dumpable regards interim value of 10 as 11.
1641 */
1642void set_dumpable(struct mm_struct *mm, int value)
1643{
1644        switch (value) {
1645        case SUID_DUMP_DISABLE:
1646                clear_bit(MMF_DUMPABLE, &mm->flags);
1647                smp_wmb();
1648                clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1649                break;
1650        case SUID_DUMP_USER:
1651                set_bit(MMF_DUMPABLE, &mm->flags);
1652                smp_wmb();
1653                clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1654                break;
1655        case SUID_DUMP_ROOT:
1656                set_bit(MMF_DUMP_SECURELY, &mm->flags);
1657                smp_wmb();
1658                set_bit(MMF_DUMPABLE, &mm->flags);
1659                break;
1660        }
1661}
1662
1663int __get_dumpable(unsigned long mm_flags)
1664{
1665        int ret;
1666
1667        ret = mm_flags & MMF_DUMPABLE_MASK;
1668        return (ret > SUID_DUMP_USER) ? SUID_DUMP_ROOT : ret;
1669}
1670
1671int get_dumpable(struct mm_struct *mm)
1672{
1673        return __get_dumpable(mm->flags);
1674}
1675
1676SYSCALL_DEFINE3(execve,
1677                const char __user *, filename,
1678                const char __user *const __user *, argv,
1679                const char __user *const __user *, envp)
1680{
1681        struct filename *path = getname(filename);
1682        int error = PTR_ERR(path);
1683        if (!IS_ERR(path)) {
1684                error = do_execve(path->name, argv, envp);
1685                putname(path);
1686        }
1687        return error;
1688}
1689#ifdef CONFIG_COMPAT
1690asmlinkage long compat_sys_execve(const char __user * filename,
1691        const compat_uptr_t __user * argv,
1692        const compat_uptr_t __user * envp)
1693{
1694        struct filename *path = getname(filename);
1695        int error = PTR_ERR(path);
1696        if (!IS_ERR(path)) {
1697                error = compat_do_execve(path->name, argv, envp);
1698                putname(path);
1699        }
1700        return error;
1701}
1702#endif
1703
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.