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/vmacache.h>
  30#include <linux/stat.h>
  31#include <linux/fcntl.h>
  32#include <linux/swap.h>
  33#include <linux/string.h>
  34#include <linux/init.h>
  35#include <linux/pagemap.h>
  36#include <linux/perf_event.h>
  37#include <linux/highmem.h>
  38#include <linux/spinlock.h>
  39#include <linux/key.h>
  40#include <linux/personality.h>
  41#include <linux/binfmts.h>
  42#include <linux/utsname.h>
  43#include <linux/pid_namespace.h>
  44#include <linux/module.h>
  45#include <linux/namei.h>
  46#include <linux/mount.h>
  47#include <linux/security.h>
  48#include <linux/syscalls.h>
  49#include <linux/tsacct_kern.h>
  50#include <linux/cn_proc.h>
  51#include <linux/audit.h>
  52#include <linux/tracehook.h>
  53#include <linux/kmod.h>
  54#include <linux/fsnotify.h>
  55#include <linux/fs_struct.h>
  56#include <linux/pipe_fs_i.h>
  57#include <linux/oom.h>
  58#include <linux/compat.h>
  59#include <linux/vmalloc.h>
  60
  61#include <asm/uaccess.h>
  62#include <asm/mmu_context.h>
  63#include <asm/tlb.h>
  64
  65#include <trace/events/task.h>
  66#include "internal.h"
  67
  68#include <trace/events/sched.h>
  69
  70int suid_dumpable = 0;
  71
  72static LIST_HEAD(formats);
  73static DEFINE_RWLOCK(binfmt_lock);
  74
  75void __register_binfmt(struct linux_binfmt * fmt, int insert)
  76{
  77        BUG_ON(!fmt);
  78        if (WARN_ON(!fmt->load_binary))
  79                return;
  80        write_lock(&binfmt_lock);
  81        insert ? list_add(&fmt->lh, &formats) :
  82                 list_add_tail(&fmt->lh, &formats);
  83        write_unlock(&binfmt_lock);
  84}
  85
  86EXPORT_SYMBOL(__register_binfmt);
  87
  88void unregister_binfmt(struct linux_binfmt * fmt)
  89{
  90        write_lock(&binfmt_lock);
  91        list_del(&fmt->lh);
  92        write_unlock(&binfmt_lock);
  93}
  94
  95EXPORT_SYMBOL(unregister_binfmt);
  96
  97static inline void put_binfmt(struct linux_binfmt * fmt)
  98{
  99        module_put(fmt->module);
 100}
 101
 102bool path_noexec(const struct path *path)
 103{
 104        return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 105               (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 106}
 107
 108#ifdef CONFIG_USELIB
 109/*
 110 * Note that a shared library must be both readable and executable due to
 111 * security reasons.
 112 *
 113 * Also note that we take the address to load from from the file itself.
 114 */
 115SYSCALL_DEFINE1(uselib, const char __user *, library)
 116{
 117        struct linux_binfmt *fmt;
 118        struct file *file;
 119        struct filename *tmp = getname(library);
 120        int error = PTR_ERR(tmp);
 121        static const struct open_flags uselib_flags = {
 122                .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 123                .acc_mode = MAY_READ | MAY_EXEC,
 124                .intent = LOOKUP_OPEN,
 125                .lookup_flags = LOOKUP_FOLLOW,
 126        };
 127
 128        if (IS_ERR(tmp))
 129                goto out;
 130
 131        file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 132        putname(tmp);
 133        error = PTR_ERR(file);
 134        if (IS_ERR(file))
 135                goto out;
 136
 137        error = -EINVAL;
 138        if (!S_ISREG(file_inode(file)->i_mode))
 139                goto exit;
 140
 141        error = -EACCES;
 142        if (path_noexec(&file->f_path))
 143                goto exit;
 144
 145        fsnotify_open(file);
 146
 147        error = -ENOEXEC;
 148
 149        read_lock(&binfmt_lock);
 150        list_for_each_entry(fmt, &formats, lh) {
 151                if (!fmt->load_shlib)
 152                        continue;
 153                if (!try_module_get(fmt->module))
 154                        continue;
 155                read_unlock(&binfmt_lock);
 156                error = fmt->load_shlib(file);
 157                read_lock(&binfmt_lock);
 158                put_binfmt(fmt);
 159                if (error != -ENOEXEC)
 160                        break;
 161        }
 162        read_unlock(&binfmt_lock);
 163exit:
 164        fput(file);
 165out:
 166        return error;
 167}
 168#endif /* #ifdef CONFIG_USELIB */
 169
 170#ifdef CONFIG_MMU
 171/*
 172 * The nascent bprm->mm is not visible until exec_mmap() but it can
 173 * use a lot of memory, account these pages in current->mm temporary
 174 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 175 * change the counter back via acct_arg_size(0).
 176 */
 177static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 178{
 179        struct mm_struct *mm = current->mm;
 180        long diff = (long)(pages - bprm->vma_pages);
 181
 182        if (!mm || !diff)
 183                return;
 184
 185        bprm->vma_pages = pages;
 186        add_mm_counter(mm, MM_ANONPAGES, diff);
 187}
 188
 189static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 190                int write)
 191{
 192        struct page *page;
 193        int ret;
 194        unsigned int gup_flags = FOLL_FORCE;
 195
 196#ifdef CONFIG_STACK_GROWSUP
 197        if (write) {
 198                ret = expand_downwards(bprm->vma, pos);
 199                if (ret < 0)
 200                        return NULL;
 201        }
 202#endif
 203
 204        if (write)
 205                gup_flags |= FOLL_WRITE;
 206
 207        /*
 208         * We are doing an exec().  'current' is the process
 209         * doing the exec and bprm->mm is the new process's mm.
 210         */
 211        ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 212                        &page, NULL);
 213        if (ret <= 0)
 214                return NULL;
 215
 216        if (write) {
 217                unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 218                struct rlimit *rlim;
 219
 220                acct_arg_size(bprm, size / PAGE_SIZE);
 221
 222                /*
 223                 * We've historically supported up to 32 pages (ARG_MAX)
 224                 * of argument strings even with small stacks
 225                 */
 226                if (size <= ARG_MAX)
 227                        return page;
 228
 229                /*
 230                 * Limit to 1/4-th the stack size for the argv+env strings.
 231                 * This ensures that:
 232                 *  - the remaining binfmt code will not run out of stack space,
 233                 *  - the program will have a reasonable amount of stack left
 234                 *    to work from.
 235                 */
 236                rlim = current->signal->rlim;
 237                if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 238                        put_page(page);
 239                        return NULL;
 240                }
 241        }
 242
 243        return page;
 244}
 245
 246static void put_arg_page(struct page *page)
 247{
 248        put_page(page);
 249}
 250
 251static void free_arg_pages(struct linux_binprm *bprm)
 252{
 253}
 254
 255static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 256                struct page *page)
 257{
 258        flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 259}
 260
 261static int __bprm_mm_init(struct linux_binprm *bprm)
 262{
 263        int err;
 264        struct vm_area_struct *vma = NULL;
 265        struct mm_struct *mm = bprm->mm;
 266
 267        bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 268        if (!vma)
 269                return -ENOMEM;
 270
 271        if (down_write_killable(&mm->mmap_sem)) {
 272                err = -EINTR;
 273                goto err_free;
 274        }
 275        vma->vm_mm = mm;
 276
 277        /*
 278         * Place the stack at the largest stack address the architecture
 279         * supports. Later, we'll move this to an appropriate place. We don't
 280         * use STACK_TOP because that can depend on attributes which aren't
 281         * configured yet.
 282         */
 283        BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 284        vma->vm_end = STACK_TOP_MAX;
 285        vma->vm_start = vma->vm_end - PAGE_SIZE;
 286        vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 287        vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 288        INIT_LIST_HEAD(&vma->anon_vma_chain);
 289
 290        err = insert_vm_struct(mm, vma);
 291        if (err)
 292                goto err;
 293
 294        mm->stack_vm = mm->total_vm = 1;
 295        arch_bprm_mm_init(mm, vma);
 296        up_write(&mm->mmap_sem);
 297        bprm->p = vma->vm_end - sizeof(void *);
 298        return 0;
 299err:
 300        up_write(&mm->mmap_sem);
 301err_free:
 302        bprm->vma = NULL;
 303        kmem_cache_free(vm_area_cachep, vma);
 304        return err;
 305}
 306
 307static bool valid_arg_len(struct linux_binprm *bprm, long len)
 308{
 309        return len <= MAX_ARG_STRLEN;
 310}
 311
 312#else
 313
 314static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 315{
 316}
 317
 318static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 319                int write)
 320{
 321        struct page *page;
 322
 323        page = bprm->page[pos / PAGE_SIZE];
 324        if (!page && write) {
 325                page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 326                if (!page)
 327                        return NULL;
 328                bprm->page[pos / PAGE_SIZE] = page;
 329        }
 330
 331        return page;
 332}
 333
 334static void put_arg_page(struct page *page)
 335{
 336}
 337
 338static void free_arg_page(struct linux_binprm *bprm, int i)
 339{
 340        if (bprm->page[i]) {
 341                __free_page(bprm->page[i]);
 342                bprm->page[i] = NULL;
 343        }
 344}
 345
 346static void free_arg_pages(struct linux_binprm *bprm)
 347{
 348        int i;
 349
 350        for (i = 0; i < MAX_ARG_PAGES; i++)
 351                free_arg_page(bprm, i);
 352}
 353
 354static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 355                struct page *page)
 356{
 357}
 358
 359static int __bprm_mm_init(struct linux_binprm *bprm)
 360{
 361        bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 362        return 0;
 363}
 364
 365static bool valid_arg_len(struct linux_binprm *bprm, long len)
 366{
 367        return len <= bprm->p;
 368}
 369
 370#endif /* CONFIG_MMU */
 371
 372/*
 373 * Create a new mm_struct and populate it with a temporary stack
 374 * vm_area_struct.  We don't have enough context at this point to set the stack
 375 * flags, permissions, and offset, so we use temporary values.  We'll update
 376 * them later in setup_arg_pages().
 377 */
 378static int bprm_mm_init(struct linux_binprm *bprm)
 379{
 380        int err;
 381        struct mm_struct *mm = NULL;
 382
 383        bprm->mm = mm = mm_alloc();
 384        err = -ENOMEM;
 385        if (!mm)
 386                goto err;
 387
 388        err = __bprm_mm_init(bprm);
 389        if (err)
 390                goto err;
 391
 392        return 0;
 393
 394err:
 395        if (mm) {
 396                bprm->mm = NULL;
 397                mmdrop(mm);
 398        }
 399
 400        return err;
 401}
 402
 403struct user_arg_ptr {
 404#ifdef CONFIG_COMPAT
 405        bool is_compat;
 406#endif
 407        union {
 408                const char __user *const __user *native;
 409#ifdef CONFIG_COMPAT
 410                const compat_uptr_t __user *compat;
 411#endif
 412        } ptr;
 413};
 414
 415static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 416{
 417        const char __user *native;
 418
 419#ifdef CONFIG_COMPAT
 420        if (unlikely(argv.is_compat)) {
 421                compat_uptr_t compat;
 422
 423                if (get_user(compat, argv.ptr.compat + nr))
 424                        return ERR_PTR(-EFAULT);
 425
 426                return compat_ptr(compat);
 427        }
 428#endif
 429
 430        if (get_user(native, argv.ptr.native + nr))
 431                return ERR_PTR(-EFAULT);
 432
 433        return native;
 434}
 435
 436/*
 437 * count() counts the number of strings in array ARGV.
 438 */
 439static int count(struct user_arg_ptr argv, int max)
 440{
 441        int i = 0;
 442
 443        if (argv.ptr.native != NULL) {
 444                for (;;) {
 445                        const char __user *p = get_user_arg_ptr(argv, i);
 446
 447                        if (!p)
 448                                break;
 449
 450                        if (IS_ERR(p))
 451                                return -EFAULT;
 452
 453                        if (i >= max)
 454                                return -E2BIG;
 455                        ++i;
 456
 457                        if (fatal_signal_pending(current))
 458                                return -ERESTARTNOHAND;
 459                        cond_resched();
 460                }
 461        }
 462        return i;
 463}
 464
 465/*
 466 * 'copy_strings()' copies argument/environment strings from the old
 467 * processes's memory to the new process's stack.  The call to get_user_pages()
 468 * ensures the destination page is created and not swapped out.
 469 */
 470static int copy_strings(int argc, struct user_arg_ptr argv,
 471                        struct linux_binprm *bprm)
 472{
 473        struct page *kmapped_page = NULL;
 474        char *kaddr = NULL;
 475        unsigned long kpos = 0;
 476        int ret;
 477
 478        while (argc-- > 0) {
 479                const char __user *str;
 480                int len;
 481                unsigned long pos;
 482
 483                ret = -EFAULT;
 484                str = get_user_arg_ptr(argv, argc);
 485                if (IS_ERR(str))
 486                        goto out;
 487
 488                len = strnlen_user(str, MAX_ARG_STRLEN);
 489                if (!len)
 490                        goto out;
 491
 492                ret = -E2BIG;
 493                if (!valid_arg_len(bprm, len))
 494                        goto out;
 495
 496                /* We're going to work our way backwords. */
 497                pos = bprm->p;
 498                str += len;
 499                bprm->p -= len;
 500
 501                while (len > 0) {
 502                        int offset, bytes_to_copy;
 503
 504                        if (fatal_signal_pending(current)) {
 505                                ret = -ERESTARTNOHAND;
 506                                goto out;
 507                        }
 508                        cond_resched();
 509
 510                        offset = pos % PAGE_SIZE;
 511                        if (offset == 0)
 512                                offset = PAGE_SIZE;
 513
 514                        bytes_to_copy = offset;
 515                        if (bytes_to_copy > len)
 516                                bytes_to_copy = len;
 517
 518                        offset -= bytes_to_copy;
 519                        pos -= bytes_to_copy;
 520                        str -= bytes_to_copy;
 521                        len -= bytes_to_copy;
 522
 523                        if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 524                                struct page *page;
 525
 526                                page = get_arg_page(bprm, pos, 1);
 527                                if (!page) {
 528                                        ret = -E2BIG;
 529                                        goto out;
 530                                }
 531
 532                                if (kmapped_page) {
 533                                        flush_kernel_dcache_page(kmapped_page);
 534                                        kunmap(kmapped_page);
 535                                        put_arg_page(kmapped_page);
 536                                }
 537                                kmapped_page = page;
 538                                kaddr = kmap(kmapped_page);
 539                                kpos = pos & PAGE_MASK;
 540                                flush_arg_page(bprm, kpos, kmapped_page);
 541                        }
 542                        if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 543                                ret = -EFAULT;
 544                                goto out;
 545                        }
 546                }
 547        }
 548        ret = 0;
 549out:
 550        if (kmapped_page) {
 551                flush_kernel_dcache_page(kmapped_page);
 552                kunmap(kmapped_page);
 553                put_arg_page(kmapped_page);
 554        }
 555        return ret;
 556}
 557
 558/*
 559 * Like copy_strings, but get argv and its values from kernel memory.
 560 */
 561int copy_strings_kernel(int argc, const char *const *__argv,
 562                        struct linux_binprm *bprm)
 563{
 564        int r;
 565        mm_segment_t oldfs = get_fs();
 566        struct user_arg_ptr argv = {
 567                .ptr.native = (const char __user *const  __user *)__argv,
 568        };
 569
 570        set_fs(KERNEL_DS);
 571        r = copy_strings(argc, argv, bprm);
 572        set_fs(oldfs);
 573
 574        return r;
 575}
 576EXPORT_SYMBOL(copy_strings_kernel);
 577
 578#ifdef CONFIG_MMU
 579
 580/*
 581 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 582 * the binfmt code determines where the new stack should reside, we shift it to
 583 * its final location.  The process proceeds as follows:
 584 *
 585 * 1) Use shift to calculate the new vma endpoints.
 586 * 2) Extend vma to cover both the old and new ranges.  This ensures the
 587 *    arguments passed to subsequent functions are consistent.
 588 * 3) Move vma's page tables to the new range.
 589 * 4) Free up any cleared pgd range.
 590 * 5) Shrink the vma to cover only the new range.
 591 */
 592static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 593{
 594        struct mm_struct *mm = vma->vm_mm;
 595        unsigned long old_start = vma->vm_start;
 596        unsigned long old_end = vma->vm_end;
 597        unsigned long length = old_end - old_start;
 598        unsigned long new_start = old_start - shift;
 599        unsigned long new_end = old_end - shift;
 600        struct mmu_gather tlb;
 601
 602        BUG_ON(new_start > new_end);
 603
 604        /*
 605         * ensure there are no vmas between where we want to go
 606         * and where we are
 607         */
 608        if (vma != find_vma(mm, new_start))
 609                return -EFAULT;
 610
 611        /*
 612         * cover the whole range: [new_start, old_end)
 613         */
 614        if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 615                return -ENOMEM;
 616
 617        /*
 618         * move the page tables downwards, on failure we rely on
 619         * process cleanup to remove whatever mess we made.
 620         */
 621        if (length != move_page_tables(vma, old_start,
 622                                       vma, new_start, length, false))
 623                return -ENOMEM;
 624
 625        lru_add_drain();
 626        tlb_gather_mmu(&tlb, mm, old_start, old_end);
 627        if (new_end > old_start) {
 628                /*
 629                 * when the old and new regions overlap clear from new_end.
 630                 */
 631                free_pgd_range(&tlb, new_end, old_end, new_end,
 632                        vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 633        } else {
 634                /*
 635                 * otherwise, clean from old_start; this is done to not touch
 636                 * the address space in [new_end, old_start) some architectures
 637                 * have constraints on va-space that make this illegal (IA64) -
 638                 * for the others its just a little faster.
 639                 */
 640                free_pgd_range(&tlb, old_start, old_end, new_end,
 641                        vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 642        }
 643        tlb_finish_mmu(&tlb, old_start, old_end);
 644
 645        /*
 646         * Shrink the vma to just the new range.  Always succeeds.
 647         */
 648        vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 649
 650        return 0;
 651}
 652
 653/*
 654 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 655 * the stack is optionally relocated, and some extra space is added.
 656 */
 657int setup_arg_pages(struct linux_binprm *bprm,
 658                    unsigned long stack_top,
 659                    int executable_stack)
 660{
 661        unsigned long ret;
 662        unsigned long stack_shift;
 663        struct mm_struct *mm = current->mm;
 664        struct vm_area_struct *vma = bprm->vma;
 665        struct vm_area_struct *prev = NULL;
 666        unsigned long vm_flags;
 667        unsigned long stack_base;
 668        unsigned long stack_size;
 669        unsigned long stack_expand;
 670        unsigned long rlim_stack;
 671
 672#ifdef CONFIG_STACK_GROWSUP
 673        /* Limit stack size */
 674        stack_base = rlimit_max(RLIMIT_STACK);
 675        if (stack_base > STACK_SIZE_MAX)
 676                stack_base = STACK_SIZE_MAX;
 677
 678        /* Add space for stack randomization. */
 679        stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 680
 681        /* Make sure we didn't let the argument array grow too large. */
 682        if (vma->vm_end - vma->vm_start > stack_base)
 683                return -ENOMEM;
 684
 685        stack_base = PAGE_ALIGN(stack_top - stack_base);
 686
 687        stack_shift = vma->vm_start - stack_base;
 688        mm->arg_start = bprm->p - stack_shift;
 689        bprm->p = vma->vm_end - stack_shift;
 690#else
 691        stack_top = arch_align_stack(stack_top);
 692        stack_top = PAGE_ALIGN(stack_top);
 693
 694        if (unlikely(stack_top < mmap_min_addr) ||
 695            unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 696                return -ENOMEM;
 697
 698        stack_shift = vma->vm_end - stack_top;
 699
 700        bprm->p -= stack_shift;
 701        mm->arg_start = bprm->p;
 702#endif
 703
 704        if (bprm->loader)
 705                bprm->loader -= stack_shift;
 706        bprm->exec -= stack_shift;
 707
 708        if (down_write_killable(&mm->mmap_sem))
 709                return -EINTR;
 710
 711        vm_flags = VM_STACK_FLAGS;
 712
 713        /*
 714         * Adjust stack execute permissions; explicitly enable for
 715         * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 716         * (arch default) otherwise.
 717         */
 718        if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 719                vm_flags |= VM_EXEC;
 720        else if (executable_stack == EXSTACK_DISABLE_X)
 721                vm_flags &= ~VM_EXEC;
 722        vm_flags |= mm->def_flags;
 723        vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 724
 725        ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 726                        vm_flags);
 727        if (ret)
 728                goto out_unlock;
 729        BUG_ON(prev != vma);
 730
 731        /* Move stack pages down in memory. */
 732        if (stack_shift) {
 733                ret = shift_arg_pages(vma, stack_shift);
 734                if (ret)
 735                        goto out_unlock;
 736        }
 737
 738        /* mprotect_fixup is overkill to remove the temporary stack flags */
 739        vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 740
 741        stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 742        stack_size = vma->vm_end - vma->vm_start;
 743        /*
 744         * Align this down to a page boundary as expand_stack
 745         * will align it up.
 746         */
 747        rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 748#ifdef CONFIG_STACK_GROWSUP
 749        if (stack_size + stack_expand > rlim_stack)
 750                stack_base = vma->vm_start + rlim_stack;
 751        else
 752                stack_base = vma->vm_end + stack_expand;
 753#else
 754        if (stack_size + stack_expand > rlim_stack)
 755                stack_base = vma->vm_end - rlim_stack;
 756        else
 757                stack_base = vma->vm_start - stack_expand;
 758#endif
 759        current->mm->start_stack = bprm->p;
 760        ret = expand_stack(vma, stack_base);
 761        if (ret)
 762                ret = -EFAULT;
 763
 764out_unlock:
 765        up_write(&mm->mmap_sem);
 766        return ret;
 767}
 768EXPORT_SYMBOL(setup_arg_pages);
 769
 770#else
 771
 772/*
 773 * Transfer the program arguments and environment from the holding pages
 774 * onto the stack. The provided stack pointer is adjusted accordingly.
 775 */
 776int transfer_args_to_stack(struct linux_binprm *bprm,
 777                           unsigned long *sp_location)
 778{
 779        unsigned long index, stop, sp;
 780        int ret = 0;
 781
 782        stop = bprm->p >> PAGE_SHIFT;
 783        sp = *sp_location;
 784
 785        for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 786                unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 787                char *src = kmap(bprm->page[index]) + offset;
 788                sp -= PAGE_SIZE - offset;
 789                if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 790                        ret = -EFAULT;
 791                kunmap(bprm->page[index]);
 792                if (ret)
 793                        goto out;
 794        }
 795
 796        *sp_location = sp;
 797
 798out:
 799        return ret;
 800}
 801EXPORT_SYMBOL(transfer_args_to_stack);
 802
 803#endif /* CONFIG_MMU */
 804
 805static struct file *do_open_execat(int fd, struct filename *name, int flags)
 806{
 807        struct file *file;
 808        int err;
 809        struct open_flags open_exec_flags = {
 810                .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 811                .acc_mode = MAY_EXEC,
 812                .intent = LOOKUP_OPEN,
 813                .lookup_flags = LOOKUP_FOLLOW,
 814        };
 815
 816        if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 817                return ERR_PTR(-EINVAL);
 818        if (flags & AT_SYMLINK_NOFOLLOW)
 819                open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 820        if (flags & AT_EMPTY_PATH)
 821                open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 822
 823        file = do_filp_open(fd, name, &open_exec_flags);
 824        if (IS_ERR(file))
 825                goto out;
 826
 827        err = -EACCES;
 828        if (!S_ISREG(file_inode(file)->i_mode))
 829                goto exit;
 830
 831        if (path_noexec(&file->f_path))
 832                goto exit;
 833
 834        err = deny_write_access(file);
 835        if (err)
 836                goto exit;
 837
 838        if (name->name[0] != '\0')
 839                fsnotify_open(file);
 840
 841out:
 842        return file;
 843
 844exit:
 845        fput(file);
 846        return ERR_PTR(err);
 847}
 848
 849struct file *open_exec(const char *name)
 850{
 851        struct filename *filename = getname_kernel(name);
 852        struct file *f = ERR_CAST(filename);
 853
 854        if (!IS_ERR(filename)) {
 855                f = do_open_execat(AT_FDCWD, filename, 0);
 856                putname(filename);
 857        }
 858        return f;
 859}
 860EXPORT_SYMBOL(open_exec);
 861
 862int kernel_read(struct file *file, loff_t offset,
 863                char *addr, unsigned long count)
 864{
 865        mm_segment_t old_fs;
 866        loff_t pos = offset;
 867        int result;
 868
 869        old_fs = get_fs();
 870        set_fs(get_ds());
 871        /* The cast to a user pointer is valid due to the set_fs() */
 872        result = vfs_read(file, (void __user *)addr, count, &pos);
 873        set_fs(old_fs);
 874        return result;
 875}
 876
 877EXPORT_SYMBOL(kernel_read);
 878
 879int kernel_read_file(struct file *file, void **buf, loff_t *size,
 880                     loff_t max_size, enum kernel_read_file_id id)
 881{
 882        loff_t i_size, pos;
 883        ssize_t bytes = 0;
 884        int ret;
 885
 886        if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 887                return -EINVAL;
 888
 889        ret = security_kernel_read_file(file, id);
 890        if (ret)
 891                return ret;
 892
 893        ret = deny_write_access(file);
 894        if (ret)
 895                return ret;
 896
 897        i_size = i_size_read(file_inode(file));
 898        if (max_size > 0 && i_size > max_size) {
 899                ret = -EFBIG;
 900                goto out;
 901        }
 902        if (i_size <= 0) {
 903                ret = -EINVAL;
 904                goto out;
 905        }
 906
 907        if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 908                *buf = vmalloc(i_size);
 909        if (!*buf) {
 910                ret = -ENOMEM;
 911                goto out;
 912        }
 913
 914        pos = 0;
 915        while (pos < i_size) {
 916                bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 917                                    i_size - pos);
 918                if (bytes < 0) {
 919                        ret = bytes;
 920                        goto out;
 921                }
 922
 923                if (bytes == 0)
 924                        break;
 925                pos += bytes;
 926        }
 927
 928        if (pos != i_size) {
 929                ret = -EIO;
 930                goto out_free;
 931        }
 932
 933        ret = security_kernel_post_read_file(file, *buf, i_size, id);
 934        if (!ret)
 935                *size = pos;
 936
 937out_free:
 938        if (ret < 0) {
 939                if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 940                        vfree(*buf);
 941                        *buf = NULL;
 942                }
 943        }
 944
 945out:
 946        allow_write_access(file);
 947        return ret;
 948}
 949EXPORT_SYMBOL_GPL(kernel_read_file);
 950
 951int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 952                               loff_t max_size, enum kernel_read_file_id id)
 953{
 954        struct file *file;
 955        int ret;
 956
 957        if (!path || !*path)
 958                return -EINVAL;
 959
 960        file = filp_open(path, O_RDONLY, 0);
 961        if (IS_ERR(file))
 962                return PTR_ERR(file);
 963
 964        ret = kernel_read_file(file, buf, size, max_size, id);
 965        fput(file);
 966        return ret;
 967}
 968EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 969
 970int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 971                             enum kernel_read_file_id id)
 972{
 973        struct fd f = fdget(fd);
 974        int ret = -EBADF;
 975
 976        if (!f.file)
 977                goto out;
 978
 979        ret = kernel_read_file(f.file, buf, size, max_size, id);
 980out:
 981        fdput(f);
 982        return ret;
 983}
 984EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 985
 986ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 987{
 988        ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 989        if (res > 0)
 990                flush_icache_range(addr, addr + len);
 991        return res;
 992}
 993EXPORT_SYMBOL(read_code);
 994
 995static int exec_mmap(struct mm_struct *mm)
 996{
 997        struct task_struct *tsk;
 998        struct mm_struct *old_mm, *active_mm;
 999
1000        /* Notify parent that we're no longer interested in the old VM */
1001        tsk = current;
1002        old_mm = current->mm;
1003        mm_release(tsk, old_mm);
1004
1005        if (old_mm) {
1006                sync_mm_rss(old_mm);
1007                /*
1008                 * Make sure that if there is a core dump in progress
1009                 * for the old mm, we get out and die instead of going
1010                 * through with the exec.  We must hold mmap_sem around
1011                 * checking core_state and changing tsk->mm.
1012                 */
1013                down_read(&old_mm->mmap_sem);
1014                if (unlikely(old_mm->core_state)) {
1015                        up_read(&old_mm->mmap_sem);
1016                        return -EINTR;
1017                }
1018        }
1019        task_lock(tsk);
1020        active_mm = tsk->active_mm;
1021        tsk->mm = mm;
1022        tsk->active_mm = mm;
1023        activate_mm(active_mm, mm);
1024        tsk->mm->vmacache_seqnum = 0;
1025        vmacache_flush(tsk);
1026        task_unlock(tsk);
1027        if (old_mm) {
1028                up_read(&old_mm->mmap_sem);
1029                BUG_ON(active_mm != old_mm);
1030                setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1031                mm_update_next_owner(old_mm);
1032                mmput(old_mm);
1033                return 0;
1034        }
1035        mmdrop(active_mm);
1036        return 0;
1037}
1038
1039/*
1040 * This function makes sure the current process has its own signal table,
1041 * so that flush_signal_handlers can later reset the handlers without
1042 * disturbing other processes.  (Other processes might share the signal
1043 * table via the CLONE_SIGHAND option to clone().)
1044 */
1045static int de_thread(struct task_struct *tsk)
1046{
1047        struct signal_struct *sig = tsk->signal;
1048        struct sighand_struct *oldsighand = tsk->sighand;
1049        spinlock_t *lock = &oldsighand->siglock;
1050
1051        if (thread_group_empty(tsk))
1052                goto no_thread_group;
1053
1054        /*
1055         * Kill all other threads in the thread group.
1056         */
1057        spin_lock_irq(lock);
1058        if (signal_group_exit(sig)) {
1059                /*
1060                 * Another group action in progress, just
1061                 * return so that the signal is processed.
1062                 */
1063                spin_unlock_irq(lock);
1064                return -EAGAIN;
1065        }
1066
1067        sig->group_exit_task = tsk;
1068        sig->notify_count = zap_other_threads(tsk);
1069        if (!thread_group_leader(tsk))
1070                sig->notify_count--;
1071
1072        while (sig->notify_count) {
1073                __set_current_state(TASK_KILLABLE);
1074                spin_unlock_irq(lock);
1075                schedule();
1076                if (unlikely(__fatal_signal_pending(tsk)))
1077                        goto killed;
1078                spin_lock_irq(lock);
1079        }
1080        spin_unlock_irq(lock);
1081
1082        /*
1083         * At this point all other threads have exited, all we have to
1084         * do is to wait for the thread group leader to become inactive,
1085         * and to assume its PID:
1086         */
1087        if (!thread_group_leader(tsk)) {
1088                struct task_struct *leader = tsk->group_leader;
1089
1090                for (;;) {
1091                        threadgroup_change_begin(tsk);
1092                        write_lock_irq(&tasklist_lock);
1093                        /*
1094                         * Do this under tasklist_lock to ensure that
1095                         * exit_notify() can't miss ->group_exit_task
1096                         */
1097                        sig->notify_count = -1;
1098                        if (likely(leader->exit_state))
1099                                break;
1100                        __set_current_state(TASK_KILLABLE);
1101                        write_unlock_irq(&tasklist_lock);
1102                        threadgroup_change_end(tsk);
1103                        schedule();
1104                        if (unlikely(__fatal_signal_pending(tsk)))
1105                                goto killed;
1106                }
1107
1108                /*
1109                 * The only record we have of the real-time age of a
1110                 * process, regardless of execs it's done, is start_time.
1111                 * All the past CPU time is accumulated in signal_struct
1112                 * from sister threads now dead.  But in this non-leader
1113                 * exec, nothing survives from the original leader thread,
1114                 * whose birth marks the true age of this process now.
1115                 * When we take on its identity by switching to its PID, we
1116                 * also take its birthdate (always earlier than our own).
1117                 */
1118                tsk->start_time = leader->start_time;
1119                tsk->real_start_time = leader->real_start_time;
1120
1121                BUG_ON(!same_thread_group(leader, tsk));
1122                BUG_ON(has_group_leader_pid(tsk));
1123                /*
1124                 * An exec() starts a new thread group with the
1125                 * TGID of the previous thread group. Rehash the
1126                 * two threads with a switched PID, and release
1127                 * the former thread group leader:
1128                 */
1129
1130                /* Become a process group leader with the old leader's pid.
1131                 * The old leader becomes a thread of the this thread group.
1132                 * Note: The old leader also uses this pid until release_task
1133                 *       is called.  Odd but simple and correct.
1134                 */
1135                tsk->pid = leader->pid;
1136                change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1137                transfer_pid(leader, tsk, PIDTYPE_PGID);
1138                transfer_pid(leader, tsk, PIDTYPE_SID);
1139
1140                list_replace_rcu(&leader->tasks, &tsk->tasks);
1141                list_replace_init(&leader->sibling, &tsk->sibling);
1142
1143                tsk->group_leader = tsk;
1144                leader->group_leader = tsk;
1145
1146                tsk->exit_signal = SIGCHLD;
1147                leader->exit_signal = -1;
1148
1149                BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1150                leader->exit_state = EXIT_DEAD;
1151
1152                /*
1153                 * We are going to release_task()->ptrace_unlink() silently,
1154                 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1155                 * the tracer wont't block again waiting for this thread.
1156                 */
1157                if (unlikely(leader->ptrace))
1158                        __wake_up_parent(leader, leader->parent);
1159                write_unlock_irq(&tasklist_lock);
1160                threadgroup_change_end(tsk);
1161
1162                release_task(leader);
1163        }
1164
1165        sig->group_exit_task = NULL;
1166        sig->notify_count = 0;
1167
1168no_thread_group:
1169        /* we have changed execution domain */
1170        tsk->exit_signal = SIGCHLD;
1171
1172        exit_itimers(sig);
1173        flush_itimer_signals();
1174
1175        if (atomic_read(&oldsighand->count) != 1) {
1176                struct sighand_struct *newsighand;
1177                /*
1178                 * This ->sighand is shared with the CLONE_SIGHAND
1179                 * but not CLONE_THREAD task, switch to the new one.
1180                 */
1181                newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1182                if (!newsighand)
1183                        return -ENOMEM;
1184
1185                atomic_set(&newsighand->count, 1);
1186                memcpy(newsighand->action, oldsighand->action,
1187                       sizeof(newsighand->action));
1188
1189                write_lock_irq(&tasklist_lock);
1190                spin_lock(&oldsighand->siglock);
1191                rcu_assign_pointer(tsk->sighand, newsighand);
1192                spin_unlock(&oldsighand->siglock);
1193                write_unlock_irq(&tasklist_lock);
1194
1195                __cleanup_sighand(oldsighand);
1196        }
1197
1198        BUG_ON(!thread_group_leader(tsk));
1199        return 0;
1200
1201killed:
1202        /* protects against exit_notify() and __exit_signal() */
1203        read_lock(&tasklist_lock);
1204        sig->group_exit_task = NULL;
1205        sig->notify_count = 0;
1206        read_unlock(&tasklist_lock);
1207        return -EAGAIN;
1208}
1209
1210char *get_task_comm(char *buf, struct task_struct *tsk)
1211{
1212        /* buf must be at least sizeof(tsk->comm) in size */
1213        task_lock(tsk);
1214        strncpy(buf, tsk->comm, sizeof(tsk->comm));
1215        task_unlock(tsk);
1216        return buf;
1217}
1218EXPORT_SYMBOL_GPL(get_task_comm);
1219
1220/*
1221 * These functions flushes out all traces of the currently running executable
1222 * so that a new one can be started
1223 */
1224
1225void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1226{
1227        task_lock(tsk);
1228        trace_task_rename(tsk, buf);
1229        strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1230        task_unlock(tsk);
1231        perf_event_comm(tsk, exec);
1232}
1233
1234int flush_old_exec(struct linux_binprm * bprm)
1235{
1236        int retval;
1237
1238        /*
1239         * Make sure we have a private signal table and that
1240         * we are unassociated from the previous thread group.
1241         */
1242        retval = de_thread(current);
1243        if (retval)
1244                goto out;
1245
1246        /*
1247         * Must be called _before_ exec_mmap() as bprm->mm is
1248         * not visibile until then. This also enables the update
1249         * to be lockless.
1250         */
1251        set_mm_exe_file(bprm->mm, bprm->file);
1252
1253        /*
1254         * Release all of the old mmap stuff
1255         */
1256        acct_arg_size(bprm, 0);
1257        retval = exec_mmap(bprm->mm);
1258        if (retval)
1259                goto out;
1260
1261        bprm->mm = NULL;                /* We're using it now */
1262
1263        set_fs(USER_DS);
1264        current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1265                                        PF_NOFREEZE | PF_NO_SETAFFINITY);
1266        flush_thread();
1267        current->personality &= ~bprm->per_clear;
1268
1269        return 0;
1270
1271out:
1272        return retval;
1273}
1274EXPORT_SYMBOL(flush_old_exec);
1275
1276void would_dump(struct linux_binprm *bprm, struct file *file)
1277{
1278        if (inode_permission(file_inode(file), MAY_READ) < 0)
1279                bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1280}
1281EXPORT_SYMBOL(would_dump);
1282
1283void setup_new_exec(struct linux_binprm * bprm)
1284{
1285        arch_pick_mmap_layout(current->mm);
1286
1287        /* This is the point of no return */
1288        current->sas_ss_sp = current->sas_ss_size = 0;
1289
1290        if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1291                set_dumpable(current->mm, SUID_DUMP_USER);
1292        else
1293                set_dumpable(current->mm, suid_dumpable);
1294
1295        perf_event_exec();
1296        __set_task_comm(current, kbasename(bprm->filename), true);
1297
1298        /* Set the new mm task size. We have to do that late because it may
1299         * depend on TIF_32BIT which is only updated in flush_thread() on
1300         * some architectures like powerpc
1301         */
1302        current->mm->task_size = TASK_SIZE;
1303
1304        /* install the new credentials */
1305        if (!uid_eq(bprm->cred->uid, current_euid()) ||
1306            !gid_eq(bprm->cred->gid, current_egid())) {
1307                current->pdeath_signal = 0;
1308        } else {
1309                would_dump(bprm, bprm->file);
1310                if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1311                        set_dumpable(current->mm, suid_dumpable);
1312        }
1313
1314        /* An exec changes our domain. We are no longer part of the thread
1315           group */
1316        current->self_exec_id++;
1317        flush_signal_handlers(current, 0);
1318        do_close_on_exec(current->files);
1319}
1320EXPORT_SYMBOL(setup_new_exec);
1321
1322/*
1323 * Prepare credentials and lock ->cred_guard_mutex.
1324 * install_exec_creds() commits the new creds and drops the lock.
1325 * Or, if exec fails before, free_bprm() should release ->cred and
1326 * and unlock.
1327 */
1328int prepare_bprm_creds(struct linux_binprm *bprm)
1329{
1330        if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1331                return -ERESTARTNOINTR;
1332
1333        bprm->cred = prepare_exec_creds();
1334        if (likely(bprm->cred))
1335                return 0;
1336
1337        mutex_unlock(&current->signal->cred_guard_mutex);
1338        return -ENOMEM;
1339}
1340
1341static void free_bprm(struct linux_binprm *bprm)
1342{
1343        free_arg_pages(bprm);
1344        if (bprm->cred) {
1345                mutex_unlock(&current->signal->cred_guard_mutex);
1346                abort_creds(bprm->cred);
1347        }
1348        if (bprm->file) {
1349                allow_write_access(bprm->file);
1350                fput(bprm->file);
1351        }
1352        /* If a binfmt changed the interp, free it. */
1353        if (bprm->interp != bprm->filename)
1354                kfree(bprm->interp);
1355        kfree(bprm);
1356}
1357
1358int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1359{
1360        /* If a binfmt changed the interp, free it first. */
1361        if (bprm->interp != bprm->filename)
1362                kfree(bprm->interp);
1363        bprm->interp = kstrdup(interp, GFP_KERNEL);
1364        if (!bprm->interp)
1365                return -ENOMEM;
1366        return 0;
1367}
1368EXPORT_SYMBOL(bprm_change_interp);
1369
1370/*
1371 * install the new credentials for this executable
1372 */
1373void install_exec_creds(struct linux_binprm *bprm)
1374{
1375        security_bprm_committing_creds(bprm);
1376
1377        commit_creds(bprm->cred);
1378        bprm->cred = NULL;
1379
1380        /*
1381         * Disable monitoring for regular users
1382         * when executing setuid binaries. Must
1383         * wait until new credentials are committed
1384         * by commit_creds() above
1385         */
1386        if (get_dumpable(current->mm) != SUID_DUMP_USER)
1387                perf_event_exit_task(current);
1388        /*
1389         * cred_guard_mutex must be held at least to this point to prevent
1390         * ptrace_attach() from altering our determination of the task's
1391         * credentials; any time after this it may be unlocked.
1392         */
1393        security_bprm_committed_creds(bprm);
1394        mutex_unlock(&current->signal->cred_guard_mutex);
1395}
1396EXPORT_SYMBOL(install_exec_creds);
1397
1398/*
1399 * determine how safe it is to execute the proposed program
1400 * - the caller must hold ->cred_guard_mutex to protect against
1401 *   PTRACE_ATTACH or seccomp thread-sync
1402 */
1403static void check_unsafe_exec(struct linux_binprm *bprm)
1404{
1405        struct task_struct *p = current, *t;
1406        unsigned n_fs;
1407
1408        if (p->ptrace) {
1409                if (p->ptrace & PT_PTRACE_CAP)
1410                        bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1411                else
1412                        bprm->unsafe |= LSM_UNSAFE_PTRACE;
1413        }
1414
1415        /*
1416         * This isn't strictly necessary, but it makes it harder for LSMs to
1417         * mess up.
1418         */
1419        if (task_no_new_privs(current))
1420                bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1421
1422        t = p;
1423        n_fs = 1;
1424        spin_lock(&p->fs->lock);
1425        rcu_read_lock();
1426        while_each_thread(p, t) {
1427                if (t->fs == p->fs)
1428                        n_fs++;
1429        }
1430        rcu_read_unlock();
1431
1432        if (p->fs->users > n_fs)
1433                bprm->unsafe |= LSM_UNSAFE_SHARE;
1434        else
1435                p->fs->in_exec = 1;
1436        spin_unlock(&p->fs->lock);
1437}
1438
1439static void bprm_fill_uid(struct linux_binprm *bprm)
1440{
1441        struct inode *inode;
1442        unsigned int mode;
1443        kuid_t uid;
1444        kgid_t gid;
1445
1446        /*
1447         * Since this can be called multiple times (via prepare_binprm),
1448         * we must clear any previous work done when setting set[ug]id
1449         * bits from any earlier bprm->file uses (for example when run
1450         * first for a setuid script then again for its interpreter).
1451         */
1452        bprm->cred->euid = current_euid();
1453        bprm->cred->egid = current_egid();
1454
1455        if (!mnt_may_suid(bprm->file->f_path.mnt))
1456                return;
1457
1458        if (task_no_new_privs(current))
1459                return;
1460
1461        inode = file_inode(bprm->file);
1462        mode = READ_ONCE(inode->i_mode);
1463        if (!(mode & (S_ISUID|S_ISGID)))
1464                return;
1465
1466        /* Be careful if suid/sgid is set */
1467        inode_lock(inode);
1468
1469        /* reload atomically mode/uid/gid now that lock held */
1470        mode = inode->i_mode;
1471        uid = inode->i_uid;
1472        gid = inode->i_gid;
1473        inode_unlock(inode);
1474
1475        /* We ignore suid/sgid if there are no mappings for them in the ns */
1476        if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1477                 !kgid_has_mapping(bprm->cred->user_ns, gid))
1478                return;
1479
1480        if (mode & S_ISUID) {
1481                bprm->per_clear |= PER_CLEAR_ON_SETID;
1482                bprm->cred->euid = uid;
1483        }
1484
1485        if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1486                bprm->per_clear |= PER_CLEAR_ON_SETID;
1487                bprm->cred->egid = gid;
1488        }
1489}
1490
1491/*
1492 * Fill the binprm structure from the inode.
1493 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1494 *
1495 * This may be called multiple times for binary chains (scripts for example).
1496 */
1497int prepare_binprm(struct linux_binprm *bprm)
1498{
1499        int retval;
1500
1501        bprm_fill_uid(bprm);
1502
1503        /* fill in binprm security blob */
1504        retval = security_bprm_set_creds(bprm);
1505        if (retval)
1506                return retval;
1507        bprm->cred_prepared = 1;
1508
1509        memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1510        return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1511}
1512
1513EXPORT_SYMBOL(prepare_binprm);
1514
1515/*
1516 * Arguments are '\0' separated strings found at the location bprm->p
1517 * points to; chop off the first by relocating brpm->p to right after
1518 * the first '\0' encountered.
1519 */
1520int remove_arg_zero(struct linux_binprm *bprm)
1521{
1522        int ret = 0;
1523        unsigned long offset;
1524        char *kaddr;
1525        struct page *page;
1526
1527        if (!bprm->argc)
1528                return 0;
1529
1530        do {
1531                offset = bprm->p & ~PAGE_MASK;
1532                page = get_arg_page(bprm, bprm->p, 0);
1533                if (!page) {
1534                        ret = -EFAULT;
1535                        goto out;
1536                }
1537                kaddr = kmap_atomic(page);
1538
1539                for (; offset < PAGE_SIZE && kaddr[offset];
1540                                offset++, bprm->p++)
1541                        ;
1542
1543                kunmap_atomic(kaddr);
1544                put_arg_page(page);
1545        } while (offset == PAGE_SIZE);
1546
1547        bprm->p++;
1548        bprm->argc--;
1549        ret = 0;
1550
1551out:
1552        return ret;
1553}
1554EXPORT_SYMBOL(remove_arg_zero);
1555
1556#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1557/*
1558 * cycle the list of binary formats handler, until one recognizes the image
1559 */
1560int search_binary_handler(struct linux_binprm *bprm)
1561{
1562        bool need_retry = IS_ENABLED(CONFIG_MODULES);
1563        struct linux_binfmt *fmt;
1564        int retval;
1565
1566        /* This allows 4 levels of binfmt rewrites before failing hard. */
1567        if (bprm->recursion_depth > 5)
1568                return -ELOOP;
1569
1570        retval = security_bprm_check(bprm);
1571        if (retval)
1572                return retval;
1573
1574        retval = -ENOENT;
1575 retry:
1576        read_lock(&binfmt_lock);
1577        list_for_each_entry(fmt, &formats, lh) {
1578                if (!try_module_get(fmt->module))
1579                        continue;
1580                read_unlock(&binfmt_lock);
1581                bprm->recursion_depth++;
1582                retval = fmt->load_binary(bprm);
1583                read_lock(&binfmt_lock);
1584                put_binfmt(fmt);
1585                bprm->recursion_depth--;
1586                if (retval < 0 && !bprm->mm) {
1587                        /* we got to flush_old_exec() and failed after it */
1588                        read_unlock(&binfmt_lock);
1589                        force_sigsegv(SIGSEGV, current);
1590                        return retval;
1591                }
1592                if (retval != -ENOEXEC || !bprm->file) {
1593                        read_unlock(&binfmt_lock);
1594                        return retval;
1595                }
1596        }
1597        read_unlock(&binfmt_lock);
1598
1599        if (need_retry) {
1600                if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1601                    printable(bprm->buf[2]) && printable(bprm->buf[3]))
1602                        return retval;
1603                if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1604                        return retval;
1605                need_retry = false;
1606                goto retry;
1607        }
1608
1609        return retval;
1610}
1611EXPORT_SYMBOL(search_binary_handler);
1612
1613static int exec_binprm(struct linux_binprm *bprm)
1614{
1615        pid_t old_pid, old_vpid;
1616        int ret;
1617
1618        /* Need to fetch pid before load_binary changes it */
1619        old_pid = current->pid;
1620        rcu_read_lock();
1621        old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1622        rcu_read_unlock();
1623
1624        ret = search_binary_handler(bprm);
1625        if (ret >= 0) {
1626                audit_bprm(bprm);
1627                trace_sched_process_exec(current, old_pid, bprm);
1628                ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1629                proc_exec_connector(current);
1630        }
1631
1632        return ret;
1633}
1634
1635/*
1636 * sys_execve() executes a new program.
1637 */
1638static int do_execveat_common(int fd, struct filename *filename,
1639                              struct user_arg_ptr argv,
1640                              struct user_arg_ptr envp,
1641                              int flags)
1642{
1643        char *pathbuf = NULL;
1644        struct linux_binprm *bprm;
1645        struct file *file;
1646        struct files_struct *displaced;
1647        int retval;
1648
1649        if (IS_ERR(filename))
1650                return PTR_ERR(filename);
1651
1652        /*
1653         * We move the actual failure in case of RLIMIT_NPROC excess from
1654         * set*uid() to execve() because too many poorly written programs
1655         * don't check setuid() return code.  Here we additionally recheck
1656         * whether NPROC limit is still exceeded.
1657         */
1658        if ((current->flags & PF_NPROC_EXCEEDED) &&
1659            atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1660                retval = -EAGAIN;
1661                goto out_ret;
1662        }
1663
1664        /* We're below the limit (still or again), so we don't want to make
1665         * further execve() calls fail. */
1666        current->flags &= ~PF_NPROC_EXCEEDED;
1667
1668        retval = unshare_files(&displaced);
1669        if (retval)
1670                goto out_ret;
1671
1672        retval = -ENOMEM;
1673        bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1674        if (!bprm)
1675                goto out_files;
1676
1677        retval = prepare_bprm_creds(bprm);
1678        if (retval)
1679                goto out_free;
1680
1681        check_unsafe_exec(bprm);
1682        current->in_execve = 1;
1683
1684        file = do_open_execat(fd, filename, flags);
1685        retval = PTR_ERR(file);
1686        if (IS_ERR(file))
1687                goto out_unmark;
1688
1689        sched_exec();
1690
1691        bprm->file = file;
1692        if (fd == AT_FDCWD || filename->name[0] == '/') {
1693                bprm->filename = filename->name;
1694        } else {
1695                if (filename->name[0] == '\0')
1696                        pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1697                else
1698                        pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1699                                            fd, filename->name);
1700                if (!pathbuf) {
1701                        retval = -ENOMEM;
1702                        goto out_unmark;
1703                }
1704                /*
1705                 * Record that a name derived from an O_CLOEXEC fd will be
1706                 * inaccessible after exec. Relies on having exclusive access to
1707                 * current->files (due to unshare_files above).
1708                 */
1709                if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1710                        bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1711                bprm->filename = pathbuf;
1712        }
1713        bprm->interp = bprm->filename;
1714
1715        retval = bprm_mm_init(bprm);
1716        if (retval)
1717                goto out_unmark;
1718
1719        bprm->argc = count(argv, MAX_ARG_STRINGS);
1720        if ((retval = bprm->argc) < 0)
1721                goto out;
1722
1723        bprm->envc = count(envp, MAX_ARG_STRINGS);
1724        if ((retval = bprm->envc) < 0)
1725                goto out;
1726
1727        retval = prepare_binprm(bprm);
1728        if (retval < 0)
1729                goto out;
1730
1731        retval = copy_strings_kernel(1, &bprm->filename, bprm);
1732        if (retval < 0)
1733                goto out;
1734
1735        bprm->exec = bprm->p;
1736        retval = copy_strings(bprm->envc, envp, bprm);
1737        if (retval < 0)
1738                goto out;
1739
1740        retval = copy_strings(bprm->argc, argv, bprm);
1741        if (retval < 0)
1742                goto out;
1743
1744        retval = exec_binprm(bprm);
1745        if (retval < 0)
1746                goto out;
1747
1748        /* execve succeeded */
1749        current->fs->in_exec = 0;
1750        current->in_execve = 0;
1751        acct_update_integrals(current);
1752        task_numa_free(current);
1753        free_bprm(bprm);
1754        kfree(pathbuf);
1755        putname(filename);
1756        if (displaced)
1757                put_files_struct(displaced);
1758        return retval;
1759
1760out:
1761        if (bprm->mm) {
1762                acct_arg_size(bprm, 0);
1763                mmput(bprm->mm);
1764        }
1765
1766out_unmark:
1767        current->fs->in_exec = 0;
1768        current->in_execve = 0;
1769
1770out_free:
1771        free_bprm(bprm);
1772        kfree(pathbuf);
1773
1774out_files:
1775        if (displaced)
1776                reset_files_struct(displaced);
1777out_ret:
1778        putname(filename);
1779        return retval;
1780}
1781
1782int do_execve(struct filename *filename,
1783        const char __user *const __user *__argv,
1784        const char __user *const __user *__envp)
1785{
1786        struct user_arg_ptr argv = { .ptr.native = __argv };
1787        struct user_arg_ptr envp = { .ptr.native = __envp };
1788        return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1789}
1790
1791int do_execveat(int fd, struct filename *filename,
1792                const char __user *const __user *__argv,
1793                const char __user *const __user *__envp,
1794                int flags)
1795{
1796        struct user_arg_ptr argv = { .ptr.native = __argv };
1797        struct user_arg_ptr envp = { .ptr.native = __envp };
1798
1799        return do_execveat_common(fd, filename, argv, envp, flags);
1800}
1801
1802#ifdef CONFIG_COMPAT
1803static int compat_do_execve(struct filename *filename,
1804        const compat_uptr_t __user *__argv,
1805        const compat_uptr_t __user *__envp)
1806{
1807        struct user_arg_ptr argv = {
1808                .is_compat = true,
1809                .ptr.compat = __argv,
1810        };
1811        struct user_arg_ptr envp = {
1812                .is_compat = true,
1813                .ptr.compat = __envp,
1814        };
1815        return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1816}
1817
1818static int compat_do_execveat(int fd, struct filename *filename,
1819                              const compat_uptr_t __user *__argv,
1820                              const compat_uptr_t __user *__envp,
1821                              int flags)
1822{
1823        struct user_arg_ptr argv = {
1824                .is_compat = true,
1825                .ptr.compat = __argv,
1826        };
1827        struct user_arg_ptr envp = {
1828                .is_compat = true,
1829                .ptr.compat = __envp,
1830        };
1831        return do_execveat_common(fd, filename, argv, envp, flags);
1832}
1833#endif
1834
1835void set_binfmt(struct linux_binfmt *new)
1836{
1837        struct mm_struct *mm = current->mm;
1838
1839        if (mm->binfmt)
1840                module_put(mm->binfmt->module);
1841
1842        mm->binfmt = new;
1843        if (new)
1844                __module_get(new->module);
1845}
1846EXPORT_SYMBOL(set_binfmt);
1847
1848/*
1849 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1850 */
1851void set_dumpable(struct mm_struct *mm, int value)
1852{
1853        unsigned long old, new;
1854
1855        if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1856                return;
1857
1858        do {
1859                old = ACCESS_ONCE(mm->flags);
1860                new = (old & ~MMF_DUMPABLE_MASK) | value;
1861        } while (cmpxchg(&mm->flags, old, new) != old);
1862}
1863
1864SYSCALL_DEFINE3(execve,
1865                const char __user *, filename,
1866                const char __user *const __user *, argv,
1867                const char __user *const __user *, envp)
1868{
1869        return do_execve(getname(filename), argv, envp);
1870}
1871
1872SYSCALL_DEFINE5(execveat,
1873                int, fd, const char __user *, filename,
1874                const char __user *const __user *, argv,
1875                const char __user *const __user *, envp,
1876                int, flags)
1877{
1878        int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1879
1880        return do_execveat(fd,
1881                           getname_flags(filename, lookup_flags, NULL),
1882                           argv, envp, flags);
1883}
1884
1885#ifdef CONFIG_COMPAT
1886COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1887        const compat_uptr_t __user *, argv,
1888        const compat_uptr_t __user *, envp)
1889{
1890        return compat_do_execve(getname(filename), argv, envp);
1891}
1892
1893COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1894                       const char __user *, filename,
1895                       const compat_uptr_t __user *, argv,
1896                       const compat_uptr_t __user *, envp,
1897                       int,  flags)
1898{
1899        int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1900
1901        return compat_do_execveat(fd,
1902                                  getname_flags(filename, lookup_flags, NULL),
1903                                  argv, envp, flags);
1904}
1905#endif
1906