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
 195#ifdef CONFIG_STACK_GROWSUP
 196        if (write) {
 197                ret = expand_downwards(bprm->vma, pos);
 198                if (ret < 0)
 199                        return NULL;
 200        }
 201#endif
 202        /*
 203         * We are doing an exec().  'current' is the process
 204         * doing the exec and bprm->mm is the new process's mm.
 205         */
 206        ret = get_user_pages_remote(current, bprm->mm, pos, 1, write,
 207                        1, &page, NULL);
 208        if (ret <= 0)
 209                return NULL;
 210
 211        if (write) {
 212                unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 213                struct rlimit *rlim;
 214
 215                acct_arg_size(bprm, size / PAGE_SIZE);
 216
 217                /*
 218                 * We've historically supported up to 32 pages (ARG_MAX)
 219                 * of argument strings even with small stacks
 220                 */
 221                if (size <= ARG_MAX)
 222                        return page;
 223
 224                /*
 225                 * Limit to 1/4-th the stack size for the argv+env strings.
 226                 * This ensures that:
 227                 *  - the remaining binfmt code will not run out of stack space,
 228                 *  - the program will have a reasonable amount of stack left
 229                 *    to work from.
 230                 */
 231                rlim = current->signal->rlim;
 232                if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
 233                        put_page(page);
 234                        return NULL;
 235                }
 236        }
 237
 238        return page;
 239}
 240
 241static void put_arg_page(struct page *page)
 242{
 243        put_page(page);
 244}
 245
 246static void free_arg_pages(struct linux_binprm *bprm)
 247{
 248}
 249
 250static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 251                struct page *page)
 252{
 253        flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 254}
 255
 256static int __bprm_mm_init(struct linux_binprm *bprm)
 257{
 258        int err;
 259        struct vm_area_struct *vma = NULL;
 260        struct mm_struct *mm = bprm->mm;
 261
 262        bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 263        if (!vma)
 264                return -ENOMEM;
 265
 266        if (down_write_killable(&mm->mmap_sem)) {
 267                err = -EINTR;
 268                goto err_free;
 269        }
 270        vma->vm_mm = mm;
 271
 272        /*
 273         * Place the stack at the largest stack address the architecture
 274         * supports. Later, we'll move this to an appropriate place. We don't
 275         * use STACK_TOP because that can depend on attributes which aren't
 276         * configured yet.
 277         */
 278        BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 279        vma->vm_end = STACK_TOP_MAX;
 280        vma->vm_start = vma->vm_end - PAGE_SIZE;
 281        vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 282        vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 283        INIT_LIST_HEAD(&vma->anon_vma_chain);
 284
 285        err = insert_vm_struct(mm, vma);
 286        if (err)
 287                goto err;
 288
 289        mm->stack_vm = mm->total_vm = 1;
 290        arch_bprm_mm_init(mm, vma);
 291        up_write(&mm->mmap_sem);
 292        bprm->p = vma->vm_end - sizeof(void *);
 293        return 0;
 294err:
 295        up_write(&mm->mmap_sem);
 296err_free:
 297        bprm->vma = NULL;
 298        kmem_cache_free(vm_area_cachep, vma);
 299        return err;
 300}
 301
 302static bool valid_arg_len(struct linux_binprm *bprm, long len)
 303{
 304        return len <= MAX_ARG_STRLEN;
 305}
 306
 307#else
 308
 309static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 310{
 311}
 312
 313static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 314                int write)
 315{
 316        struct page *page;
 317
 318        page = bprm->page[pos / PAGE_SIZE];
 319        if (!page && write) {
 320                page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 321                if (!page)
 322                        return NULL;
 323                bprm->page[pos / PAGE_SIZE] = page;
 324        }
 325
 326        return page;
 327}
 328
 329static void put_arg_page(struct page *page)
 330{
 331}
 332
 333static void free_arg_page(struct linux_binprm *bprm, int i)
 334{
 335        if (bprm->page[i]) {
 336                __free_page(bprm->page[i]);
 337                bprm->page[i] = NULL;
 338        }
 339}
 340
 341static void free_arg_pages(struct linux_binprm *bprm)
 342{
 343        int i;
 344
 345        for (i = 0; i < MAX_ARG_PAGES; i++)
 346                free_arg_page(bprm, i);
 347}
 348
 349static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 350                struct page *page)
 351{
 352}
 353
 354static int __bprm_mm_init(struct linux_binprm *bprm)
 355{
 356        bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 357        return 0;
 358}
 359
 360static bool valid_arg_len(struct linux_binprm *bprm, long len)
 361{
 362        return len <= bprm->p;
 363}
 364
 365#endif /* CONFIG_MMU */
 366
 367/*
 368 * Create a new mm_struct and populate it with a temporary stack
 369 * vm_area_struct.  We don't have enough context at this point to set the stack
 370 * flags, permissions, and offset, so we use temporary values.  We'll update
 371 * them later in setup_arg_pages().
 372 */
 373static int bprm_mm_init(struct linux_binprm *bprm)
 374{
 375        int err;
 376        struct mm_struct *mm = NULL;
 377
 378        bprm->mm = mm = mm_alloc();
 379        err = -ENOMEM;
 380        if (!mm)
 381                goto err;
 382
 383        err = __bprm_mm_init(bprm);
 384        if (err)
 385                goto err;
 386
 387        return 0;
 388
 389err:
 390        if (mm) {
 391                bprm->mm = NULL;
 392                mmdrop(mm);
 393        }
 394
 395        return err;
 396}
 397
 398struct user_arg_ptr {
 399#ifdef CONFIG_COMPAT
 400        bool is_compat;
 401#endif
 402        union {
 403                const char __user *const __user *native;
 404#ifdef CONFIG_COMPAT
 405                const compat_uptr_t __user *compat;
 406#endif
 407        } ptr;
 408};
 409
 410static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 411{
 412        const char __user *native;
 413
 414#ifdef CONFIG_COMPAT
 415        if (unlikely(argv.is_compat)) {
 416                compat_uptr_t compat;
 417
 418                if (get_user(compat, argv.ptr.compat + nr))
 419                        return ERR_PTR(-EFAULT);
 420
 421                return compat_ptr(compat);
 422        }
 423#endif
 424
 425        if (get_user(native, argv.ptr.native + nr))
 426                return ERR_PTR(-EFAULT);
 427
 428        return native;
 429}
 430
 431/*
 432 * count() counts the number of strings in array ARGV.
 433 */
 434static int count(struct user_arg_ptr argv, int max)
 435{
 436        int i = 0;
 437
 438        if (argv.ptr.native != NULL) {
 439                for (;;) {
 440                        const char __user *p = get_user_arg_ptr(argv, i);
 441
 442                        if (!p)
 443                                break;
 444
 445                        if (IS_ERR(p))
 446                                return -EFAULT;
 447
 448                        if (i >= max)
 449                                return -E2BIG;
 450                        ++i;
 451
 452                        if (fatal_signal_pending(current))
 453                                return -ERESTARTNOHAND;
 454                        cond_resched();
 455                }
 456        }
 457        return i;
 458}
 459
 460/*
 461 * 'copy_strings()' copies argument/environment strings from the old
 462 * processes's memory to the new process's stack.  The call to get_user_pages()
 463 * ensures the destination page is created and not swapped out.
 464 */
 465static int copy_strings(int argc, struct user_arg_ptr argv,
 466                        struct linux_binprm *bprm)
 467{
 468        struct page *kmapped_page = NULL;
 469        char *kaddr = NULL;
 470        unsigned long kpos = 0;
 471        int ret;
 472
 473        while (argc-- > 0) {
 474                const char __user *str;
 475                int len;
 476                unsigned long pos;
 477
 478                ret = -EFAULT;
 479                str = get_user_arg_ptr(argv, argc);
 480                if (IS_ERR(str))
 481                        goto out;
 482
 483                len = strnlen_user(str, MAX_ARG_STRLEN);
 484                if (!len)
 485                        goto out;
 486
 487                ret = -E2BIG;
 488                if (!valid_arg_len(bprm, len))
 489                        goto out;
 490
 491                /* We're going to work our way backwords. */
 492                pos = bprm->p;
 493                str += len;
 494                bprm->p -= len;
 495
 496                while (len > 0) {
 497                        int offset, bytes_to_copy;
 498
 499                        if (fatal_signal_pending(current)) {
 500                                ret = -ERESTARTNOHAND;
 501                                goto out;
 502                        }
 503                        cond_resched();
 504
 505                        offset = pos % PAGE_SIZE;
 506                        if (offset == 0)
 507                                offset = PAGE_SIZE;
 508
 509                        bytes_to_copy = offset;
 510                        if (bytes_to_copy > len)
 511                                bytes_to_copy = len;
 512
 513                        offset -= bytes_to_copy;
 514                        pos -= bytes_to_copy;
 515                        str -= bytes_to_copy;
 516                        len -= bytes_to_copy;
 517
 518                        if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 519                                struct page *page;
 520
 521                                page = get_arg_page(bprm, pos, 1);
 522                                if (!page) {
 523                                        ret = -E2BIG;
 524                                        goto out;
 525                                }
 526
 527                                if (kmapped_page) {
 528                                        flush_kernel_dcache_page(kmapped_page);
 529                                        kunmap(kmapped_page);
 530                                        put_arg_page(kmapped_page);
 531                                }
 532                                kmapped_page = page;
 533                                kaddr = kmap(kmapped_page);
 534                                kpos = pos & PAGE_MASK;
 535                                flush_arg_page(bprm, kpos, kmapped_page);
 536                        }
 537                        if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 538                                ret = -EFAULT;
 539                                goto out;
 540                        }
 541                }
 542        }
 543        ret = 0;
 544out:
 545        if (kmapped_page) {
 546                flush_kernel_dcache_page(kmapped_page);
 547                kunmap(kmapped_page);
 548                put_arg_page(kmapped_page);
 549        }
 550        return ret;
 551}
 552
 553/*
 554 * Like copy_strings, but get argv and its values from kernel memory.
 555 */
 556int copy_strings_kernel(int argc, const char *const *__argv,
 557                        struct linux_binprm *bprm)
 558{
 559        int r;
 560        mm_segment_t oldfs = get_fs();
 561        struct user_arg_ptr argv = {
 562                .ptr.native = (const char __user *const  __user *)__argv,
 563        };
 564
 565        set_fs(KERNEL_DS);
 566        r = copy_strings(argc, argv, bprm);
 567        set_fs(oldfs);
 568
 569        return r;
 570}
 571EXPORT_SYMBOL(copy_strings_kernel);
 572
 573#ifdef CONFIG_MMU
 574
 575/*
 576 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 577 * the binfmt code determines where the new stack should reside, we shift it to
 578 * its final location.  The process proceeds as follows:
 579 *
 580 * 1) Use shift to calculate the new vma endpoints.
 581 * 2) Extend vma to cover both the old and new ranges.  This ensures the
 582 *    arguments passed to subsequent functions are consistent.
 583 * 3) Move vma's page tables to the new range.
 584 * 4) Free up any cleared pgd range.
 585 * 5) Shrink the vma to cover only the new range.
 586 */
 587static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 588{
 589        struct mm_struct *mm = vma->vm_mm;
 590        unsigned long old_start = vma->vm_start;
 591        unsigned long old_end = vma->vm_end;
 592        unsigned long length = old_end - old_start;
 593        unsigned long new_start = old_start - shift;
 594        unsigned long new_end = old_end - shift;
 595        struct mmu_gather tlb;
 596
 597        BUG_ON(new_start > new_end);
 598
 599        /*
 600         * ensure there are no vmas between where we want to go
 601         * and where we are
 602         */
 603        if (vma != find_vma(mm, new_start))
 604                return -EFAULT;
 605
 606        /*
 607         * cover the whole range: [new_start, old_end)
 608         */
 609        if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 610                return -ENOMEM;
 611
 612        /*
 613         * move the page tables downwards, on failure we rely on
 614         * process cleanup to remove whatever mess we made.
 615         */
 616        if (length != move_page_tables(vma, old_start,
 617                                       vma, new_start, length, false))
 618                return -ENOMEM;
 619
 620        lru_add_drain();
 621        tlb_gather_mmu(&tlb, mm, old_start, old_end);
 622        if (new_end > old_start) {
 623                /*
 624                 * when the old and new regions overlap clear from new_end.
 625                 */
 626                free_pgd_range(&tlb, new_end, old_end, new_end,
 627                        vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 628        } else {
 629                /*
 630                 * otherwise, clean from old_start; this is done to not touch
 631                 * the address space in [new_end, old_start) some architectures
 632                 * have constraints on va-space that make this illegal (IA64) -
 633                 * for the others its just a little faster.
 634                 */
 635                free_pgd_range(&tlb, old_start, old_end, new_end,
 636                        vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 637        }
 638        tlb_finish_mmu(&tlb, old_start, old_end);
 639
 640        /*
 641         * Shrink the vma to just the new range.  Always succeeds.
 642         */
 643        vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 644
 645        return 0;
 646}
 647
 648/*
 649 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 650 * the stack is optionally relocated, and some extra space is added.
 651 */
 652int setup_arg_pages(struct linux_binprm *bprm,
 653                    unsigned long stack_top,
 654                    int executable_stack)
 655{
 656        unsigned long ret;
 657        unsigned long stack_shift;
 658        struct mm_struct *mm = current->mm;
 659        struct vm_area_struct *vma = bprm->vma;
 660        struct vm_area_struct *prev = NULL;
 661        unsigned long vm_flags;
 662        unsigned long stack_base;
 663        unsigned long stack_size;
 664        unsigned long stack_expand;
 665        unsigned long rlim_stack;
 666
 667#ifdef CONFIG_STACK_GROWSUP
 668        /* Limit stack size */
 669        stack_base = rlimit_max(RLIMIT_STACK);
 670        if (stack_base > STACK_SIZE_MAX)
 671                stack_base = STACK_SIZE_MAX;
 672
 673        /* Add space for stack randomization. */
 674        stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 675
 676        /* Make sure we didn't let the argument array grow too large. */
 677        if (vma->vm_end - vma->vm_start > stack_base)
 678                return -ENOMEM;
 679
 680        stack_base = PAGE_ALIGN(stack_top - stack_base);
 681
 682        stack_shift = vma->vm_start - stack_base;
 683        mm->arg_start = bprm->p - stack_shift;
 684        bprm->p = vma->vm_end - stack_shift;
 685#else
 686        stack_top = arch_align_stack(stack_top);
 687        stack_top = PAGE_ALIGN(stack_top);
 688
 689        if (unlikely(stack_top < mmap_min_addr) ||
 690            unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 691                return -ENOMEM;
 692
 693        stack_shift = vma->vm_end - stack_top;
 694
 695        bprm->p -= stack_shift;
 696        mm->arg_start = bprm->p;
 697#endif
 698
 699        if (bprm->loader)
 700                bprm->loader -= stack_shift;
 701        bprm->exec -= stack_shift;
 702
 703        if (down_write_killable(&mm->mmap_sem))
 704                return -EINTR;
 705
 706        vm_flags = VM_STACK_FLAGS;
 707
 708        /*
 709         * Adjust stack execute permissions; explicitly enable for
 710         * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 711         * (arch default) otherwise.
 712         */
 713        if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 714                vm_flags |= VM_EXEC;
 715        else if (executable_stack == EXSTACK_DISABLE_X)
 716                vm_flags &= ~VM_EXEC;
 717        vm_flags |= mm->def_flags;
 718        vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 719
 720        ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 721                        vm_flags);
 722        if (ret)
 723                goto out_unlock;
 724        BUG_ON(prev != vma);
 725
 726        /* Move stack pages down in memory. */
 727        if (stack_shift) {
 728                ret = shift_arg_pages(vma, stack_shift);
 729                if (ret)
 730                        goto out_unlock;
 731        }
 732
 733        /* mprotect_fixup is overkill to remove the temporary stack flags */
 734        vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 735
 736        stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 737        stack_size = vma->vm_end - vma->vm_start;
 738        /*
 739         * Align this down to a page boundary as expand_stack
 740         * will align it up.
 741         */
 742        rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 743#ifdef CONFIG_STACK_GROWSUP
 744        if (stack_size + stack_expand > rlim_stack)
 745                stack_base = vma->vm_start + rlim_stack;
 746        else
 747                stack_base = vma->vm_end + stack_expand;
 748#else
 749        if (stack_size + stack_expand > rlim_stack)
 750                stack_base = vma->vm_end - rlim_stack;
 751        else
 752                stack_base = vma->vm_start - stack_expand;
 753#endif
 754        current->mm->start_stack = bprm->p;
 755        ret = expand_stack(vma, stack_base);
 756        if (ret)
 757                ret = -EFAULT;
 758
 759out_unlock:
 760        up_write(&mm->mmap_sem);
 761        return ret;
 762}
 763EXPORT_SYMBOL(setup_arg_pages);
 764
 765#endif /* CONFIG_MMU */
 766
 767static struct file *do_open_execat(int fd, struct filename *name, int flags)
 768{
 769        struct file *file;
 770        int err;
 771        struct open_flags open_exec_flags = {
 772                .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 773                .acc_mode = MAY_EXEC,
 774                .intent = LOOKUP_OPEN,
 775                .lookup_flags = LOOKUP_FOLLOW,
 776        };
 777
 778        if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 779                return ERR_PTR(-EINVAL);
 780        if (flags & AT_SYMLINK_NOFOLLOW)
 781                open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 782        if (flags & AT_EMPTY_PATH)
 783                open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 784
 785        file = do_filp_open(fd, name, &open_exec_flags);
 786        if (IS_ERR(file))
 787                goto out;
 788
 789        err = -EACCES;
 790        if (!S_ISREG(file_inode(file)->i_mode))
 791                goto exit;
 792
 793        if (path_noexec(&file->f_path))
 794                goto exit;
 795
 796        err = deny_write_access(file);
 797        if (err)
 798                goto exit;
 799
 800        if (name->name[0] != '\0')
 801                fsnotify_open(file);
 802
 803out:
 804        return file;
 805
 806exit:
 807        fput(file);
 808        return ERR_PTR(err);
 809}
 810
 811struct file *open_exec(const char *name)
 812{
 813        struct filename *filename = getname_kernel(name);
 814        struct file *f = ERR_CAST(filename);
 815
 816        if (!IS_ERR(filename)) {
 817                f = do_open_execat(AT_FDCWD, filename, 0);
 818                putname(filename);
 819        }
 820        return f;
 821}
 822EXPORT_SYMBOL(open_exec);
 823
 824int kernel_read(struct file *file, loff_t offset,
 825                char *addr, unsigned long count)
 826{
 827        mm_segment_t old_fs;
 828        loff_t pos = offset;
 829        int result;
 830
 831        old_fs = get_fs();
 832        set_fs(get_ds());
 833        /* The cast to a user pointer is valid due to the set_fs() */
 834        result = vfs_read(file, (void __user *)addr, count, &pos);
 835        set_fs(old_fs);
 836        return result;
 837}
 838
 839EXPORT_SYMBOL(kernel_read);
 840
 841int kernel_read_file(struct file *file, void **buf, loff_t *size,
 842                     loff_t max_size, enum kernel_read_file_id id)
 843{
 844        loff_t i_size, pos;
 845        ssize_t bytes = 0;
 846        int ret;
 847
 848        if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 849                return -EINVAL;
 850
 851        ret = security_kernel_read_file(file, id);
 852        if (ret)
 853                return ret;
 854
 855        ret = deny_write_access(file);
 856        if (ret)
 857                return ret;
 858
 859        i_size = i_size_read(file_inode(file));
 860        if (max_size > 0 && i_size > max_size) {
 861                ret = -EFBIG;
 862                goto out;
 863        }
 864        if (i_size <= 0) {
 865                ret = -EINVAL;
 866                goto out;
 867        }
 868
 869        *buf = vmalloc(i_size);
 870        if (!*buf) {
 871                ret = -ENOMEM;
 872                goto out;
 873        }
 874
 875        pos = 0;
 876        while (pos < i_size) {
 877                bytes = kernel_read(file, pos, (char *)(*buf) + pos,
 878                                    i_size - pos);
 879                if (bytes < 0) {
 880                        ret = bytes;
 881                        goto out;
 882                }
 883
 884                if (bytes == 0)
 885                        break;
 886                pos += bytes;
 887        }
 888
 889        if (pos != i_size) {
 890                ret = -EIO;
 891                goto out_free;
 892        }
 893
 894        ret = security_kernel_post_read_file(file, *buf, i_size, id);
 895        if (!ret)
 896                *size = pos;
 897
 898out_free:
 899        if (ret < 0) {
 900                vfree(*buf);
 901                *buf = NULL;
 902        }
 903
 904out:
 905        allow_write_access(file);
 906        return ret;
 907}
 908EXPORT_SYMBOL_GPL(kernel_read_file);
 909
 910int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
 911                               loff_t max_size, enum kernel_read_file_id id)
 912{
 913        struct file *file;
 914        int ret;
 915
 916        if (!path || !*path)
 917                return -EINVAL;
 918
 919        file = filp_open(path, O_RDONLY, 0);
 920        if (IS_ERR(file))
 921                return PTR_ERR(file);
 922
 923        ret = kernel_read_file(file, buf, size, max_size, id);
 924        fput(file);
 925        return ret;
 926}
 927EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 928
 929int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 930                             enum kernel_read_file_id id)
 931{
 932        struct fd f = fdget(fd);
 933        int ret = -EBADF;
 934
 935        if (!f.file)
 936                goto out;
 937
 938        ret = kernel_read_file(f.file, buf, size, max_size, id);
 939out:
 940        fdput(f);
 941        return ret;
 942}
 943EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 944
 945ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 946{
 947        ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 948        if (res > 0)
 949                flush_icache_range(addr, addr + len);
 950        return res;
 951}
 952EXPORT_SYMBOL(read_code);
 953
 954static int exec_mmap(struct mm_struct *mm)
 955{
 956        struct task_struct *tsk;
 957        struct mm_struct *old_mm, *active_mm;
 958
 959        /* Notify parent that we're no longer interested in the old VM */
 960        tsk = current;
 961        old_mm = current->mm;
 962        mm_release(tsk, old_mm);
 963
 964        if (old_mm) {
 965                sync_mm_rss(old_mm);
 966                /*
 967                 * Make sure that if there is a core dump in progress
 968                 * for the old mm, we get out and die instead of going
 969                 * through with the exec.  We must hold mmap_sem around
 970                 * checking core_state and changing tsk->mm.
 971                 */
 972                down_read(&old_mm->mmap_sem);
 973                if (unlikely(old_mm->core_state)) {
 974                        up_read(&old_mm->mmap_sem);
 975                        return -EINTR;
 976                }
 977        }
 978        task_lock(tsk);
 979        active_mm = tsk->active_mm;
 980        tsk->mm = mm;
 981        tsk->active_mm = mm;
 982        activate_mm(active_mm, mm);
 983        tsk->mm->vmacache_seqnum = 0;
 984        vmacache_flush(tsk);
 985        task_unlock(tsk);
 986        if (old_mm) {
 987                up_read(&old_mm->mmap_sem);
 988                BUG_ON(active_mm != old_mm);
 989                setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
 990                mm_update_next_owner(old_mm);
 991                mmput(old_mm);
 992                return 0;
 993        }
 994        mmdrop(active_mm);
 995        return 0;
 996}
 997
 998/*
 999 * This function makes sure the current process has its own signal table,
1000 * so that flush_signal_handlers can later reset the handlers without
1001 * disturbing other processes.  (Other processes might share the signal
1002 * table via the CLONE_SIGHAND option to clone().)
1003 */
1004static int de_thread(struct task_struct *tsk)
1005{
1006        struct signal_struct *sig = tsk->signal;
1007        struct sighand_struct *oldsighand = tsk->sighand;
1008        spinlock_t *lock = &oldsighand->siglock;
1009
1010        if (thread_group_empty(tsk))
1011                goto no_thread_group;
1012
1013        /*
1014         * Kill all other threads in the thread group.
1015         */
1016        spin_lock_irq(lock);
1017        if (signal_group_exit(sig)) {
1018                /*
1019                 * Another group action in progress, just
1020                 * return so that the signal is processed.
1021                 */
1022                spin_unlock_irq(lock);
1023                return -EAGAIN;
1024        }
1025
1026        sig->group_exit_task = tsk;
1027        sig->notify_count = zap_other_threads(tsk);
1028        if (!thread_group_leader(tsk))
1029                sig->notify_count--;
1030
1031        while (sig->notify_count) {
1032                __set_current_state(TASK_KILLABLE);
1033                spin_unlock_irq(lock);
1034                schedule();
1035                if (unlikely(__fatal_signal_pending(tsk)))
1036                        goto killed;
1037                spin_lock_irq(lock);
1038        }
1039        spin_unlock_irq(lock);
1040
1041        /*
1042         * At this point all other threads have exited, all we have to
1043         * do is to wait for the thread group leader to become inactive,
1044         * and to assume its PID:
1045         */
1046        if (!thread_group_leader(tsk)) {
1047                struct task_struct *leader = tsk->group_leader;
1048
1049                for (;;) {
1050                        threadgroup_change_begin(tsk);
1051                        write_lock_irq(&tasklist_lock);
1052                        /*
1053                         * Do this under tasklist_lock to ensure that
1054                         * exit_notify() can't miss ->group_exit_task
1055                         */
1056                        sig->notify_count = -1;
1057                        if (likely(leader->exit_state))
1058                                break;
1059                        __set_current_state(TASK_KILLABLE);
1060                        write_unlock_irq(&tasklist_lock);
1061                        threadgroup_change_end(tsk);
1062                        schedule();
1063                        if (unlikely(__fatal_signal_pending(tsk)))
1064                                goto killed;
1065                }
1066
1067                /*
1068                 * The only record we have of the real-time age of a
1069                 * process, regardless of execs it's done, is start_time.
1070                 * All the past CPU time is accumulated in signal_struct
1071                 * from sister threads now dead.  But in this non-leader
1072                 * exec, nothing survives from the original leader thread,
1073                 * whose birth marks the true age of this process now.
1074                 * When we take on its identity by switching to its PID, we
1075                 * also take its birthdate (always earlier than our own).
1076                 */
1077                tsk->start_time = leader->start_time;
1078                tsk->real_start_time = leader->real_start_time;
1079
1080                BUG_ON(!same_thread_group(leader, tsk));
1081                BUG_ON(has_group_leader_pid(tsk));
1082                /*
1083                 * An exec() starts a new thread group with the
1084                 * TGID of the previous thread group. Rehash the
1085                 * two threads with a switched PID, and release
1086                 * the former thread group leader:
1087                 */
1088
1089                /* Become a process group leader with the old leader's pid.
1090                 * The old leader becomes a thread of the this thread group.
1091                 * Note: The old leader also uses this pid until release_task
1092                 *       is called.  Odd but simple and correct.
1093                 */
1094                tsk->pid = leader->pid;
1095                change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1096                transfer_pid(leader, tsk, PIDTYPE_PGID);
1097                transfer_pid(leader, tsk, PIDTYPE_SID);
1098
1099                list_replace_rcu(&leader->tasks, &tsk->tasks);
1100                list_replace_init(&leader->sibling, &tsk->sibling);
1101
1102                tsk->group_leader = tsk;
1103                leader->group_leader = tsk;
1104
1105                tsk->exit_signal = SIGCHLD;
1106                leader->exit_signal = -1;
1107
1108                BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1109                leader->exit_state = EXIT_DEAD;
1110
1111                /*
1112                 * We are going to release_task()->ptrace_unlink() silently,
1113                 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1114                 * the tracer wont't block again waiting for this thread.
1115                 */
1116                if (unlikely(leader->ptrace))
1117                        __wake_up_parent(leader, leader->parent);
1118                write_unlock_irq(&tasklist_lock);
1119                threadgroup_change_end(tsk);
1120
1121                release_task(leader);
1122        }
1123
1124        sig->group_exit_task = NULL;
1125        sig->notify_count = 0;
1126
1127no_thread_group:
1128        /* we have changed execution domain */
1129        tsk->exit_signal = SIGCHLD;
1130
1131        exit_itimers(sig);
1132        flush_itimer_signals();
1133
1134        if (atomic_read(&oldsighand->count) != 1) {
1135                struct sighand_struct *newsighand;
1136                /*
1137                 * This ->sighand is shared with the CLONE_SIGHAND
1138                 * but not CLONE_THREAD task, switch to the new one.
1139                 */
1140                newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1141                if (!newsighand)
1142                        return -ENOMEM;
1143
1144                atomic_set(&newsighand->count, 1);
1145                memcpy(newsighand->action, oldsighand->action,
1146                       sizeof(newsighand->action));
1147
1148                write_lock_irq(&tasklist_lock);
1149                spin_lock(&oldsighand->siglock);
1150                rcu_assign_pointer(tsk->sighand, newsighand);
1151                spin_unlock(&oldsighand->siglock);
1152                write_unlock_irq(&tasklist_lock);
1153
1154                __cleanup_sighand(oldsighand);
1155        }
1156
1157        BUG_ON(!thread_group_leader(tsk));
1158        return 0;
1159
1160killed:
1161        /* protects against exit_notify() and __exit_signal() */
1162        read_lock(&tasklist_lock);
1163        sig->group_exit_task = NULL;
1164        sig->notify_count = 0;
1165        read_unlock(&tasklist_lock);
1166        return -EAGAIN;
1167}
1168
1169char *get_task_comm(char *buf, struct task_struct *tsk)
1170{
1171        /* buf must be at least sizeof(tsk->comm) in size */
1172        task_lock(tsk);
1173        strncpy(buf, tsk->comm, sizeof(tsk->comm));
1174        task_unlock(tsk);
1175        return buf;
1176}
1177EXPORT_SYMBOL_GPL(get_task_comm);
1178
1179/*
1180 * These functions flushes out all traces of the currently running executable
1181 * so that a new one can be started
1182 */
1183
1184void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1185{
1186        task_lock(tsk);
1187        trace_task_rename(tsk, buf);
1188        strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1189        task_unlock(tsk);
1190        perf_event_comm(tsk, exec);
1191}
1192
1193int flush_old_exec(struct linux_binprm * bprm)
1194{
1195        int retval;
1196
1197        /*
1198         * Make sure we have a private signal table and that
1199         * we are unassociated from the previous thread group.
1200         */
1201        retval = de_thread(current);
1202        if (retval)
1203                goto out;
1204
1205        /*
1206         * Must be called _before_ exec_mmap() as bprm->mm is
1207         * not visibile until then. This also enables the update
1208         * to be lockless.
1209         */
1210        set_mm_exe_file(bprm->mm, bprm->file);
1211
1212        /*
1213         * Release all of the old mmap stuff
1214         */
1215        acct_arg_size(bprm, 0);
1216        retval = exec_mmap(bprm->mm);
1217        if (retval)
1218                goto out;
1219
1220        bprm->mm = NULL;                /* We're using it now */
1221
1222        set_fs(USER_DS);
1223        current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1224                                        PF_NOFREEZE | PF_NO_SETAFFINITY);
1225        flush_thread();
1226        current->personality &= ~bprm->per_clear;
1227
1228        return 0;
1229
1230out:
1231        return retval;
1232}
1233EXPORT_SYMBOL(flush_old_exec);
1234
1235void would_dump(struct linux_binprm *bprm, struct file *file)
1236{
1237        if (inode_permission(file_inode(file), MAY_READ) < 0)
1238                bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1239}
1240EXPORT_SYMBOL(would_dump);
1241
1242void setup_new_exec(struct linux_binprm * bprm)
1243{
1244        arch_pick_mmap_layout(current->mm);
1245
1246        /* This is the point of no return */
1247        current->sas_ss_sp = current->sas_ss_size = 0;
1248
1249        if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1250                set_dumpable(current->mm, SUID_DUMP_USER);
1251        else
1252                set_dumpable(current->mm, suid_dumpable);
1253
1254        perf_event_exec();
1255        __set_task_comm(current, kbasename(bprm->filename), true);
1256
1257        /* Set the new mm task size. We have to do that late because it may
1258         * depend on TIF_32BIT which is only updated in flush_thread() on
1259         * some architectures like powerpc
1260         */
1261        current->mm->task_size = TASK_SIZE;
1262
1263        /* install the new credentials */
1264        if (!uid_eq(bprm->cred->uid, current_euid()) ||
1265            !gid_eq(bprm->cred->gid, current_egid())) {
1266                current->pdeath_signal = 0;
1267        } else {
1268                would_dump(bprm, bprm->file);
1269                if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1270                        set_dumpable(current->mm, suid_dumpable);
1271        }
1272
1273        /* An exec changes our domain. We are no longer part of the thread
1274           group */
1275        current->self_exec_id++;
1276        flush_signal_handlers(current, 0);
1277        do_close_on_exec(current->files);
1278}
1279EXPORT_SYMBOL(setup_new_exec);
1280
1281/*
1282 * Prepare credentials and lock ->cred_guard_mutex.
1283 * install_exec_creds() commits the new creds and drops the lock.
1284 * Or, if exec fails before, free_bprm() should release ->cred and
1285 * and unlock.
1286 */
1287int prepare_bprm_creds(struct linux_binprm *bprm)
1288{
1289        if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1290                return -ERESTARTNOINTR;
1291
1292        bprm->cred = prepare_exec_creds();
1293        if (likely(bprm->cred))
1294                return 0;
1295
1296        mutex_unlock(&current->signal->cred_guard_mutex);
1297        return -ENOMEM;
1298}
1299
1300static void free_bprm(struct linux_binprm *bprm)
1301{
1302        free_arg_pages(bprm);
1303        if (bprm->cred) {
1304                mutex_unlock(&current->signal->cred_guard_mutex);
1305                abort_creds(bprm->cred);
1306        }
1307        if (bprm->file) {
1308                allow_write_access(bprm->file);
1309                fput(bprm->file);
1310        }
1311        /* If a binfmt changed the interp, free it. */
1312        if (bprm->interp != bprm->filename)
1313                kfree(bprm->interp);
1314        kfree(bprm);
1315}
1316
1317int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1318{
1319        /* If a binfmt changed the interp, free it first. */
1320        if (bprm->interp != bprm->filename)
1321                kfree(bprm->interp);
1322        bprm->interp = kstrdup(interp, GFP_KERNEL);
1323        if (!bprm->interp)
1324                return -ENOMEM;
1325        return 0;
1326}
1327EXPORT_SYMBOL(bprm_change_interp);
1328
1329/*
1330 * install the new credentials for this executable
1331 */
1332void install_exec_creds(struct linux_binprm *bprm)
1333{
1334        security_bprm_committing_creds(bprm);
1335
1336        commit_creds(bprm->cred);
1337        bprm->cred = NULL;
1338
1339        /*
1340         * Disable monitoring for regular users
1341         * when executing setuid binaries. Must
1342         * wait until new credentials are committed
1343         * by commit_creds() above
1344         */
1345        if (get_dumpable(current->mm) != SUID_DUMP_USER)
1346                perf_event_exit_task(current);
1347        /*
1348         * cred_guard_mutex must be held at least to this point to prevent
1349         * ptrace_attach() from altering our determination of the task's
1350         * credentials; any time after this it may be unlocked.
1351         */
1352        security_bprm_committed_creds(bprm);
1353        mutex_unlock(&current->signal->cred_guard_mutex);
1354}
1355EXPORT_SYMBOL(install_exec_creds);
1356
1357/*
1358 * determine how safe it is to execute the proposed program
1359 * - the caller must hold ->cred_guard_mutex to protect against
1360 *   PTRACE_ATTACH or seccomp thread-sync
1361 */
1362static void check_unsafe_exec(struct linux_binprm *bprm)
1363{
1364        struct task_struct *p = current, *t;
1365        unsigned n_fs;
1366
1367        if (p->ptrace) {
1368                if (p->ptrace & PT_PTRACE_CAP)
1369                        bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1370                else
1371                        bprm->unsafe |= LSM_UNSAFE_PTRACE;
1372        }
1373
1374        /*
1375         * This isn't strictly necessary, but it makes it harder for LSMs to
1376         * mess up.
1377         */
1378        if (task_no_new_privs(current))
1379                bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1380
1381        t = p;
1382        n_fs = 1;
1383        spin_lock(&p->fs->lock);
1384        rcu_read_lock();
1385        while_each_thread(p, t) {
1386                if (t->fs == p->fs)
1387                        n_fs++;
1388        }
1389        rcu_read_unlock();
1390
1391        if (p->fs->users > n_fs)
1392                bprm->unsafe |= LSM_UNSAFE_SHARE;
1393        else
1394                p->fs->in_exec = 1;
1395        spin_unlock(&p->fs->lock);
1396}
1397
1398static void bprm_fill_uid(struct linux_binprm *bprm)
1399{
1400        struct inode *inode;
1401        unsigned int mode;
1402        kuid_t uid;
1403        kgid_t gid;
1404
1405        /*
1406         * Since this can be called multiple times (via prepare_binprm),
1407         * we must clear any previous work done when setting set[ug]id
1408         * bits from any earlier bprm->file uses (for example when run
1409         * first for a setuid script then again for its interpreter).
1410         */
1411        bprm->cred->euid = current_euid();
1412        bprm->cred->egid = current_egid();
1413
1414        if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1415                return;
1416
1417        if (task_no_new_privs(current))
1418                return;
1419
1420        inode = file_inode(bprm->file);
1421        mode = READ_ONCE(inode->i_mode);
1422        if (!(mode & (S_ISUID|S_ISGID)))
1423                return;
1424
1425        /* Be careful if suid/sgid is set */
1426        inode_lock(inode);
1427
1428        /* reload atomically mode/uid/gid now that lock held */
1429        mode = inode->i_mode;
1430        uid = inode->i_uid;
1431        gid = inode->i_gid;
1432        inode_unlock(inode);
1433
1434        /* We ignore suid/sgid if there are no mappings for them in the ns */
1435        if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1436                 !kgid_has_mapping(bprm->cred->user_ns, gid))
1437                return;
1438
1439        if (mode & S_ISUID) {
1440                bprm->per_clear |= PER_CLEAR_ON_SETID;
1441                bprm->cred->euid = uid;
1442        }
1443
1444        if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1445                bprm->per_clear |= PER_CLEAR_ON_SETID;
1446                bprm->cred->egid = gid;
1447        }
1448}
1449
1450/*
1451 * Fill the binprm structure from the inode.
1452 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1453 *
1454 * This may be called multiple times for binary chains (scripts for example).
1455 */
1456int prepare_binprm(struct linux_binprm *bprm)
1457{
1458        int retval;
1459
1460        bprm_fill_uid(bprm);
1461
1462        /* fill in binprm security blob */
1463        retval = security_bprm_set_creds(bprm);
1464        if (retval)
1465                return retval;
1466        bprm->cred_prepared = 1;
1467
1468        memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1469        return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1470}
1471
1472EXPORT_SYMBOL(prepare_binprm);
1473
1474/*
1475 * Arguments are '\0' separated strings found at the location bprm->p
1476 * points to; chop off the first by relocating brpm->p to right after
1477 * the first '\0' encountered.
1478 */
1479int remove_arg_zero(struct linux_binprm *bprm)
1480{
1481        int ret = 0;
1482        unsigned long offset;
1483        char *kaddr;
1484        struct page *page;
1485
1486        if (!bprm->argc)
1487                return 0;
1488
1489        do {
1490                offset = bprm->p & ~PAGE_MASK;
1491                page = get_arg_page(bprm, bprm->p, 0);
1492                if (!page) {
1493                        ret = -EFAULT;
1494                        goto out;
1495                }
1496                kaddr = kmap_atomic(page);
1497
1498                for (; offset < PAGE_SIZE && kaddr[offset];
1499                                offset++, bprm->p++)
1500                        ;
1501
1502                kunmap_atomic(kaddr);
1503                put_arg_page(page);
1504        } while (offset == PAGE_SIZE);
1505
1506        bprm->p++;
1507        bprm->argc--;
1508        ret = 0;
1509
1510out:
1511        return ret;
1512}
1513EXPORT_SYMBOL(remove_arg_zero);
1514
1515#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1516/*
1517 * cycle the list of binary formats handler, until one recognizes the image
1518 */
1519int search_binary_handler(struct linux_binprm *bprm)
1520{
1521        bool need_retry = IS_ENABLED(CONFIG_MODULES);
1522        struct linux_binfmt *fmt;
1523        int retval;
1524
1525        /* This allows 4 levels of binfmt rewrites before failing hard. */
1526        if (bprm->recursion_depth > 5)
1527                return -ELOOP;
1528
1529        retval = security_bprm_check(bprm);
1530        if (retval)
1531                return retval;
1532
1533        retval = -ENOENT;
1534 retry:
1535        read_lock(&binfmt_lock);
1536        list_for_each_entry(fmt, &formats, lh) {
1537                if (!try_module_get(fmt->module))
1538                        continue;
1539                read_unlock(&binfmt_lock);
1540                bprm->recursion_depth++;
1541                retval = fmt->load_binary(bprm);
1542                read_lock(&binfmt_lock);
1543                put_binfmt(fmt);
1544                bprm->recursion_depth--;
1545                if (retval < 0 && !bprm->mm) {
1546                        /* we got to flush_old_exec() and failed after it */
1547                        read_unlock(&binfmt_lock);
1548                        force_sigsegv(SIGSEGV, current);
1549                        return retval;
1550                }
1551                if (retval != -ENOEXEC || !bprm->file) {
1552                        read_unlock(&binfmt_lock);
1553                        return retval;
1554                }
1555        }
1556        read_unlock(&binfmt_lock);
1557
1558        if (need_retry) {
1559                if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1560                    printable(bprm->buf[2]) && printable(bprm->buf[3]))
1561                        return retval;
1562                if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1563                        return retval;
1564                need_retry = false;
1565                goto retry;
1566        }
1567
1568        return retval;
1569}
1570EXPORT_SYMBOL(search_binary_handler);
1571
1572static int exec_binprm(struct linux_binprm *bprm)
1573{
1574        pid_t old_pid, old_vpid;
1575        int ret;
1576
1577        /* Need to fetch pid before load_binary changes it */
1578        old_pid = current->pid;
1579        rcu_read_lock();
1580        old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1581        rcu_read_unlock();
1582
1583        ret = search_binary_handler(bprm);
1584        if (ret >= 0) {
1585                audit_bprm(bprm);
1586                trace_sched_process_exec(current, old_pid, bprm);
1587                ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1588                proc_exec_connector(current);
1589        }
1590
1591        return ret;
1592}
1593
1594/*
1595 * sys_execve() executes a new program.
1596 */
1597static int do_execveat_common(int fd, struct filename *filename,
1598                              struct user_arg_ptr argv,
1599                              struct user_arg_ptr envp,
1600                              int flags)
1601{
1602        char *pathbuf = NULL;
1603        struct linux_binprm *bprm;
1604        struct file *file;
1605        struct files_struct *displaced;
1606        int retval;
1607
1608        if (IS_ERR(filename))
1609                return PTR_ERR(filename);
1610
1611        /*
1612         * We move the actual failure in case of RLIMIT_NPROC excess from
1613         * set*uid() to execve() because too many poorly written programs
1614         * don't check setuid() return code.  Here we additionally recheck
1615         * whether NPROC limit is still exceeded.
1616         */
1617        if ((current->flags & PF_NPROC_EXCEEDED) &&
1618            atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1619                retval = -EAGAIN;
1620                goto out_ret;
1621        }
1622
1623        /* We're below the limit (still or again), so we don't want to make
1624         * further execve() calls fail. */
1625        current->flags &= ~PF_NPROC_EXCEEDED;
1626
1627        retval = unshare_files(&displaced);
1628        if (retval)
1629                goto out_ret;
1630
1631        retval = -ENOMEM;
1632        bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1633        if (!bprm)
1634                goto out_files;
1635
1636        retval = prepare_bprm_creds(bprm);
1637        if (retval)
1638                goto out_free;
1639
1640        check_unsafe_exec(bprm);
1641        current->in_execve = 1;
1642
1643        file = do_open_execat(fd, filename, flags);
1644        retval = PTR_ERR(file);
1645        if (IS_ERR(file))
1646                goto out_unmark;
1647
1648        sched_exec();
1649
1650        bprm->file = file;
1651        if (fd == AT_FDCWD || filename->name[0] == '/') {
1652                bprm->filename = filename->name;
1653        } else {
1654                if (filename->name[0] == '\0')
1655                        pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1656                else
1657                        pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1658                                            fd, filename->name);
1659                if (!pathbuf) {
1660                        retval = -ENOMEM;
1661                        goto out_unmark;
1662                }
1663                /*
1664                 * Record that a name derived from an O_CLOEXEC fd will be
1665                 * inaccessible after exec. Relies on having exclusive access to
1666                 * current->files (due to unshare_files above).
1667                 */
1668                if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1669                        bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1670                bprm->filename = pathbuf;
1671        }
1672        bprm->interp = bprm->filename;
1673
1674        retval = bprm_mm_init(bprm);
1675        if (retval)
1676                goto out_unmark;
1677
1678        bprm->argc = count(argv, MAX_ARG_STRINGS);
1679        if ((retval = bprm->argc) < 0)
1680                goto out;
1681
1682        bprm->envc = count(envp, MAX_ARG_STRINGS);
1683        if ((retval = bprm->envc) < 0)
1684                goto out;
1685
1686        retval = prepare_binprm(bprm);
1687        if (retval < 0)
1688                goto out;
1689
1690        retval = copy_strings_kernel(1, &bprm->filename, bprm);
1691        if (retval < 0)
1692                goto out;
1693
1694        bprm->exec = bprm->p;
1695        retval = copy_strings(bprm->envc, envp, bprm);
1696        if (retval < 0)
1697                goto out;
1698
1699        retval = copy_strings(bprm->argc, argv, bprm);
1700        if (retval < 0)
1701                goto out;
1702
1703        retval = exec_binprm(bprm);
1704        if (retval < 0)
1705                goto out;
1706
1707        /* execve succeeded */
1708        current->fs->in_exec = 0;
1709        current->in_execve = 0;
1710        acct_update_integrals(current);
1711        task_numa_free(current);
1712        free_bprm(bprm);
1713        kfree(pathbuf);
1714        putname(filename);
1715        if (displaced)
1716                put_files_struct(displaced);
1717        return retval;
1718
1719out:
1720        if (bprm->mm) {
1721                acct_arg_size(bprm, 0);
1722                mmput(bprm->mm);
1723        }
1724
1725out_unmark:
1726        current->fs->in_exec = 0;
1727        current->in_execve = 0;
1728
1729out_free:
1730        free_bprm(bprm);
1731        kfree(pathbuf);
1732
1733out_files:
1734        if (displaced)
1735                reset_files_struct(displaced);
1736out_ret:
1737        putname(filename);
1738        return retval;
1739}
1740
1741int do_execve(struct filename *filename,
1742        const char __user *const __user *__argv,
1743        const char __user *const __user *__envp)
1744{
1745        struct user_arg_ptr argv = { .ptr.native = __argv };
1746        struct user_arg_ptr envp = { .ptr.native = __envp };
1747        return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1748}
1749
1750int do_execveat(int fd, struct filename *filename,
1751                const char __user *const __user *__argv,
1752                const char __user *const __user *__envp,
1753                int flags)
1754{
1755        struct user_arg_ptr argv = { .ptr.native = __argv };
1756        struct user_arg_ptr envp = { .ptr.native = __envp };
1757
1758        return do_execveat_common(fd, filename, argv, envp, flags);
1759}
1760
1761#ifdef CONFIG_COMPAT
1762static int compat_do_execve(struct filename *filename,
1763        const compat_uptr_t __user *__argv,
1764        const compat_uptr_t __user *__envp)
1765{
1766        struct user_arg_ptr argv = {
1767                .is_compat = true,
1768                .ptr.compat = __argv,
1769        };
1770        struct user_arg_ptr envp = {
1771                .is_compat = true,
1772                .ptr.compat = __envp,
1773        };
1774        return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1775}
1776
1777static int compat_do_execveat(int fd, struct filename *filename,
1778                              const compat_uptr_t __user *__argv,
1779                              const compat_uptr_t __user *__envp,
1780                              int flags)
1781{
1782        struct user_arg_ptr argv = {
1783                .is_compat = true,
1784                .ptr.compat = __argv,
1785        };
1786        struct user_arg_ptr envp = {
1787                .is_compat = true,
1788                .ptr.compat = __envp,
1789        };
1790        return do_execveat_common(fd, filename, argv, envp, flags);
1791}
1792#endif
1793
1794void set_binfmt(struct linux_binfmt *new)
1795{
1796        struct mm_struct *mm = current->mm;
1797
1798        if (mm->binfmt)
1799                module_put(mm->binfmt->module);
1800
1801        mm->binfmt = new;
1802        if (new)
1803                __module_get(new->module);
1804}
1805EXPORT_SYMBOL(set_binfmt);
1806
1807/*
1808 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1809 */
1810void set_dumpable(struct mm_struct *mm, int value)
1811{
1812        unsigned long old, new;
1813
1814        if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1815                return;
1816
1817        do {
1818                old = ACCESS_ONCE(mm->flags);
1819                new = (old & ~MMF_DUMPABLE_MASK) | value;
1820        } while (cmpxchg(&mm->flags, old, new) != old);
1821}
1822
1823SYSCALL_DEFINE3(execve,
1824                const char __user *, filename,
1825                const char __user *const __user *, argv,
1826                const char __user *const __user *, envp)
1827{
1828        return do_execve(getname(filename), argv, envp);
1829}
1830
1831SYSCALL_DEFINE5(execveat,
1832                int, fd, const char __user *, filename,
1833                const char __user *const __user *, argv,
1834                const char __user *const __user *, envp,
1835                int, flags)
1836{
1837        int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1838
1839        return do_execveat(fd,
1840                           getname_flags(filename, lookup_flags, NULL),
1841                           argv, envp, flags);
1842}
1843
1844#ifdef CONFIG_COMPAT
1845COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1846        const compat_uptr_t __user *, argv,
1847        const compat_uptr_t __user *, envp)
1848{
1849        return compat_do_execve(getname(filename), argv, envp);
1850}
1851
1852COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1853                       const char __user *, filename,
1854                       const compat_uptr_t __user *, argv,
1855                       const compat_uptr_t __user *, envp,
1856                       int,  flags)
1857{
1858        int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1859
1860        return compat_do_execveat(fd,
1861                                  getname_flags(filename, lookup_flags, NULL),
1862                                  argv, envp, flags);
1863}
1864#endif
1865
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.