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/sched/mm.h>
  36#include <linux/sched/coredump.h>
  37#include <linux/sched/signal.h>
  38#include <linux/sched/numa_balancing.h>
  39#include <linux/sched/task.h>
  40#include <linux/pagemap.h>
  41#include <linux/perf_event.h>
  42#include <linux/highmem.h>
  43#include <linux/spinlock.h>
  44#include <linux/key.h>
  45#include <linux/personality.h>
  46#include <linux/binfmts.h>
  47#include <linux/utsname.h>
  48#include <linux/pid_namespace.h>
  49#include <linux/module.h>
  50#include <linux/namei.h>
  51#include <linux/mount.h>
  52#include <linux/security.h>
  53#include <linux/syscalls.h>
  54#include <linux/tsacct_kern.h>
  55#include <linux/cn_proc.h>
  56#include <linux/audit.h>
  57#include <linux/tracehook.h>
  58#include <linux/kmod.h>
  59#include <linux/fsnotify.h>
  60#include <linux/fs_struct.h>
  61#include <linux/pipe_fs_i.h>
  62#include <linux/oom.h>
  63#include <linux/compat.h>
  64#include <linux/vmalloc.h>
  65
  66#include <linux/uaccess.h>
  67#include <asm/mmu_context.h>
  68#include <asm/tlb.h>
  69
  70#include <trace/events/task.h>
  71#include "internal.h"
  72
  73#include <trace/events/sched.h>
  74
  75int suid_dumpable = 0;
  76
  77static LIST_HEAD(formats);
  78static DEFINE_RWLOCK(binfmt_lock);
  79
  80void __register_binfmt(struct linux_binfmt * fmt, int insert)
  81{
  82        BUG_ON(!fmt);
  83        if (WARN_ON(!fmt->load_binary))
  84                return;
  85        write_lock(&binfmt_lock);
  86        insert ? list_add(&fmt->lh, &formats) :
  87                 list_add_tail(&fmt->lh, &formats);
  88        write_unlock(&binfmt_lock);
  89}
  90
  91EXPORT_SYMBOL(__register_binfmt);
  92
  93void unregister_binfmt(struct linux_binfmt * fmt)
  94{
  95        write_lock(&binfmt_lock);
  96        list_del(&fmt->lh);
  97        write_unlock(&binfmt_lock);
  98}
  99
 100EXPORT_SYMBOL(unregister_binfmt);
 101
 102static inline void put_binfmt(struct linux_binfmt * fmt)
 103{
 104        module_put(fmt->module);
 105}
 106
 107bool path_noexec(const struct path *path)
 108{
 109        return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 110               (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 111}
 112
 113#ifdef CONFIG_USELIB
 114/*
 115 * Note that a shared library must be both readable and executable due to
 116 * security reasons.
 117 *
 118 * Also note that we take the address to load from from the file itself.
 119 */
 120SYSCALL_DEFINE1(uselib, const char __user *, library)
 121{
 122        struct linux_binfmt *fmt;
 123        struct file *file;
 124        struct filename *tmp = getname(library);
 125        int error = PTR_ERR(tmp);
 126        static const struct open_flags uselib_flags = {
 127                .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 128                .acc_mode = MAY_READ | MAY_EXEC,
 129                .intent = LOOKUP_OPEN,
 130                .lookup_flags = LOOKUP_FOLLOW,
 131        };
 132
 133        if (IS_ERR(tmp))
 134                goto out;
 135
 136        file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 137        putname(tmp);
 138        error = PTR_ERR(file);
 139        if (IS_ERR(file))
 140                goto out;
 141
 142        error = -EINVAL;
 143        if (!S_ISREG(file_inode(file)->i_mode))
 144                goto exit;
 145
 146        error = -EACCES;
 147        if (path_noexec(&file->f_path))
 148                goto exit;
 149
 150        fsnotify_open(file);
 151
 152        error = -ENOEXEC;
 153
 154        read_lock(&binfmt_lock);
 155        list_for_each_entry(fmt, &formats, lh) {
 156                if (!fmt->load_shlib)
 157                        continue;
 158                if (!try_module_get(fmt->module))
 159                        continue;
 160                read_unlock(&binfmt_lock);
 161                error = fmt->load_shlib(file);
 162                read_lock(&binfmt_lock);
 163                put_binfmt(fmt);
 164                if (error != -ENOEXEC)
 165                        break;
 166        }
 167        read_unlock(&binfmt_lock);
 168exit:
 169        fput(file);
 170out:
 171        return error;
 172}
 173#endif /* #ifdef CONFIG_USELIB */
 174
 175#ifdef CONFIG_MMU
 176/*
 177 * The nascent bprm->mm is not visible until exec_mmap() but it can
 178 * use a lot of memory, account these pages in current->mm temporary
 179 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 180 * change the counter back via acct_arg_size(0).
 181 */
 182static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 183{
 184        struct mm_struct *mm = current->mm;
 185        long diff = (long)(pages - bprm->vma_pages);
 186
 187        if (!mm || !diff)
 188                return;
 189
 190        bprm->vma_pages = pages;
 191        add_mm_counter(mm, MM_ANONPAGES, diff);
 192}
 193
 194static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 195                int write)
 196{
 197        struct page *page;
 198        int ret;
 199        unsigned int gup_flags = FOLL_FORCE;
 200
 201#ifdef CONFIG_STACK_GROWSUP
 202        if (write) {
 203                ret = expand_downwards(bprm->vma, pos);
 204                if (ret < 0)
 205                        return NULL;
 206        }
 207#endif
 208
 209        if (write)
 210                gup_flags |= FOLL_WRITE;
 211
 212        /*
 213         * We are doing an exec().  'current' is the process
 214         * doing the exec and bprm->mm is the new process's mm.
 215         */
 216        ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 217                        &page, NULL, NULL);
 218        if (ret <= 0)
 219                return NULL;
 220
 221        if (write) {
 222                unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
 223                unsigned long ptr_size, limit;
 224
 225                /*
 226                 * Since the stack will hold pointers to the strings, we
 227                 * must account for them as well.
 228                 *
 229                 * The size calculation is the entire vma while each arg page is
 230                 * built, so each time we get here it's calculating how far it
 231                 * is currently (rather than each call being just the newly
 232                 * added size from the arg page).  As a result, we need to
 233                 * always add the entire size of the pointers, so that on the
 234                 * last call to get_arg_page() we'll actually have the entire
 235                 * correct size.
 236                 */
 237                ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 238                if (ptr_size > ULONG_MAX - size)
 239                        goto fail;
 240                size += ptr_size;
 241
 242                acct_arg_size(bprm, size / PAGE_SIZE);
 243
 244                /*
 245                 * We've historically supported up to 32 pages (ARG_MAX)
 246                 * of argument strings even with small stacks
 247                 */
 248                if (size <= ARG_MAX)
 249                        return page;
 250
 251                /*
 252                 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 253                 * (whichever is smaller) for the argv+env strings.
 254                 * This ensures that:
 255                 *  - the remaining binfmt code will not run out of stack space,
 256                 *  - the program will have a reasonable amount of stack left
 257                 *    to work from.
 258                 */
 259                limit = _STK_LIM / 4 * 3;
 260                limit = min(limit, rlimit(RLIMIT_STACK) / 4);
 261                if (size > limit)
 262                        goto fail;
 263        }
 264
 265        return page;
 266
 267fail:
 268        put_page(page);
 269        return NULL;
 270}
 271
 272static void put_arg_page(struct page *page)
 273{
 274        put_page(page);
 275}
 276
 277static void free_arg_pages(struct linux_binprm *bprm)
 278{
 279}
 280
 281static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 282                struct page *page)
 283{
 284        flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 285}
 286
 287static int __bprm_mm_init(struct linux_binprm *bprm)
 288{
 289        int err;
 290        struct vm_area_struct *vma = NULL;
 291        struct mm_struct *mm = bprm->mm;
 292
 293        bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 294        if (!vma)
 295                return -ENOMEM;
 296
 297        if (down_write_killable(&mm->mmap_sem)) {
 298                err = -EINTR;
 299                goto err_free;
 300        }
 301        vma->vm_mm = mm;
 302
 303        /*
 304         * Place the stack at the largest stack address the architecture
 305         * supports. Later, we'll move this to an appropriate place. We don't
 306         * use STACK_TOP because that can depend on attributes which aren't
 307         * configured yet.
 308         */
 309        BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 310        vma->vm_end = STACK_TOP_MAX;
 311        vma->vm_start = vma->vm_end - PAGE_SIZE;
 312        vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 313        vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 314        INIT_LIST_HEAD(&vma->anon_vma_chain);
 315
 316        err = insert_vm_struct(mm, vma);
 317        if (err)
 318                goto err;
 319
 320        mm->stack_vm = mm->total_vm = 1;
 321        arch_bprm_mm_init(mm, vma);
 322        up_write(&mm->mmap_sem);
 323        bprm->p = vma->vm_end - sizeof(void *);
 324        return 0;
 325err:
 326        up_write(&mm->mmap_sem);
 327err_free:
 328        bprm->vma = NULL;
 329        kmem_cache_free(vm_area_cachep, vma);
 330        return err;
 331}
 332
 333static bool valid_arg_len(struct linux_binprm *bprm, long len)
 334{
 335        return len <= MAX_ARG_STRLEN;
 336}
 337
 338#else
 339
 340static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 341{
 342}
 343
 344static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 345                int write)
 346{
 347        struct page *page;
 348
 349        page = bprm->page[pos / PAGE_SIZE];
 350        if (!page && write) {
 351                page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 352                if (!page)
 353                        return NULL;
 354                bprm->page[pos / PAGE_SIZE] = page;
 355        }
 356
 357        return page;
 358}
 359
 360static void put_arg_page(struct page *page)
 361{
 362}
 363
 364static void free_arg_page(struct linux_binprm *bprm, int i)
 365{
 366        if (bprm->page[i]) {
 367                __free_page(bprm->page[i]);
 368                bprm->page[i] = NULL;
 369        }
 370}
 371
 372static void free_arg_pages(struct linux_binprm *bprm)
 373{
 374        int i;
 375
 376        for (i = 0; i < MAX_ARG_PAGES; i++)
 377                free_arg_page(bprm, i);
 378}
 379
 380static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 381                struct page *page)
 382{
 383}
 384
 385static int __bprm_mm_init(struct linux_binprm *bprm)
 386{
 387        bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 388        return 0;
 389}
 390
 391static bool valid_arg_len(struct linux_binprm *bprm, long len)
 392{
 393        return len <= bprm->p;
 394}
 395
 396#endif /* CONFIG_MMU */
 397
 398/*
 399 * Create a new mm_struct and populate it with a temporary stack
 400 * vm_area_struct.  We don't have enough context at this point to set the stack
 401 * flags, permissions, and offset, so we use temporary values.  We'll update
 402 * them later in setup_arg_pages().
 403 */
 404static int bprm_mm_init(struct linux_binprm *bprm)
 405{
 406        int err;
 407        struct mm_struct *mm = NULL;
 408
 409        bprm->mm = mm = mm_alloc();
 410        err = -ENOMEM;
 411        if (!mm)
 412                goto err;
 413
 414        err = __bprm_mm_init(bprm);
 415        if (err)
 416                goto err;
 417
 418        return 0;
 419
 420err:
 421        if (mm) {
 422                bprm->mm = NULL;
 423                mmdrop(mm);
 424        }
 425
 426        return err;
 427}
 428
 429struct user_arg_ptr {
 430#ifdef CONFIG_COMPAT
 431        bool is_compat;
 432#endif
 433        union {
 434                const char __user *const __user *native;
 435#ifdef CONFIG_COMPAT
 436                const compat_uptr_t __user *compat;
 437#endif
 438        } ptr;
 439};
 440
 441static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 442{
 443        const char __user *native;
 444
 445#ifdef CONFIG_COMPAT
 446        if (unlikely(argv.is_compat)) {
 447                compat_uptr_t compat;
 448
 449                if (get_user(compat, argv.ptr.compat + nr))
 450                        return ERR_PTR(-EFAULT);
 451
 452                return compat_ptr(compat);
 453        }
 454#endif
 455
 456        if (get_user(native, argv.ptr.native + nr))
 457                return ERR_PTR(-EFAULT);
 458
 459        return native;
 460}
 461
 462/*
 463 * count() counts the number of strings in array ARGV.
 464 */
 465static int count(struct user_arg_ptr argv, int max)
 466{
 467        int i = 0;
 468
 469        if (argv.ptr.native != NULL) {
 470                for (;;) {
 471                        const char __user *p = get_user_arg_ptr(argv, i);
 472
 473                        if (!p)
 474                                break;
 475
 476                        if (IS_ERR(p))
 477                                return -EFAULT;
 478
 479                        if (i >= max)
 480                                return -E2BIG;
 481                        ++i;
 482
 483                        if (fatal_signal_pending(current))
 484                                return -ERESTARTNOHAND;
 485                        cond_resched();
 486                }
 487        }
 488        return i;
 489}
 490
 491/*
 492 * 'copy_strings()' copies argument/environment strings from the old
 493 * processes's memory to the new process's stack.  The call to get_user_pages()
 494 * ensures the destination page is created and not swapped out.
 495 */
 496static int copy_strings(int argc, struct user_arg_ptr argv,
 497                        struct linux_binprm *bprm)
 498{
 499        struct page *kmapped_page = NULL;
 500        char *kaddr = NULL;
 501        unsigned long kpos = 0;
 502        int ret;
 503
 504        while (argc-- > 0) {
 505                const char __user *str;
 506                int len;
 507                unsigned long pos;
 508
 509                ret = -EFAULT;
 510                str = get_user_arg_ptr(argv, argc);
 511                if (IS_ERR(str))
 512                        goto out;
 513
 514                len = strnlen_user(str, MAX_ARG_STRLEN);
 515                if (!len)
 516                        goto out;
 517
 518                ret = -E2BIG;
 519                if (!valid_arg_len(bprm, len))
 520                        goto out;
 521
 522                /* We're going to work our way backwords. */
 523                pos = bprm->p;
 524                str += len;
 525                bprm->p -= len;
 526
 527                while (len > 0) {
 528                        int offset, bytes_to_copy;
 529
 530                        if (fatal_signal_pending(current)) {
 531                                ret = -ERESTARTNOHAND;
 532                                goto out;
 533                        }
 534                        cond_resched();
 535
 536                        offset = pos % PAGE_SIZE;
 537                        if (offset == 0)
 538                                offset = PAGE_SIZE;
 539
 540                        bytes_to_copy = offset;
 541                        if (bytes_to_copy > len)
 542                                bytes_to_copy = len;
 543
 544                        offset -= bytes_to_copy;
 545                        pos -= bytes_to_copy;
 546                        str -= bytes_to_copy;
 547                        len -= bytes_to_copy;
 548
 549                        if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 550                                struct page *page;
 551
 552                                page = get_arg_page(bprm, pos, 1);
 553                                if (!page) {
 554                                        ret = -E2BIG;
 555                                        goto out;
 556                                }
 557
 558                                if (kmapped_page) {
 559                                        flush_kernel_dcache_page(kmapped_page);
 560                                        kunmap(kmapped_page);
 561                                        put_arg_page(kmapped_page);
 562                                }
 563                                kmapped_page = page;
 564                                kaddr = kmap(kmapped_page);
 565                                kpos = pos & PAGE_MASK;
 566                                flush_arg_page(bprm, kpos, kmapped_page);
 567                        }
 568                        if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 569                                ret = -EFAULT;
 570                                goto out;
 571                        }
 572                }
 573        }
 574        ret = 0;
 575out:
 576        if (kmapped_page) {
 577                flush_kernel_dcache_page(kmapped_page);
 578                kunmap(kmapped_page);
 579                put_arg_page(kmapped_page);
 580        }
 581        return ret;
 582}
 583
 584/*
 585 * Like copy_strings, but get argv and its values from kernel memory.
 586 */
 587int copy_strings_kernel(int argc, const char *const *__argv,
 588                        struct linux_binprm *bprm)
 589{
 590        int r;
 591        mm_segment_t oldfs = get_fs();
 592        struct user_arg_ptr argv = {
 593                .ptr.native = (const char __user *const  __user *)__argv,
 594        };
 595
 596        set_fs(KERNEL_DS);
 597        r = copy_strings(argc, argv, bprm);
 598        set_fs(oldfs);
 599
 600        return r;
 601}
 602EXPORT_SYMBOL(copy_strings_kernel);
 603
 604#ifdef CONFIG_MMU
 605
 606/*
 607 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 608 * the binfmt code determines where the new stack should reside, we shift it to
 609 * its final location.  The process proceeds as follows:
 610 *
 611 * 1) Use shift to calculate the new vma endpoints.
 612 * 2) Extend vma to cover both the old and new ranges.  This ensures the
 613 *    arguments passed to subsequent functions are consistent.
 614 * 3) Move vma's page tables to the new range.
 615 * 4) Free up any cleared pgd range.
 616 * 5) Shrink the vma to cover only the new range.
 617 */
 618static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 619{
 620        struct mm_struct *mm = vma->vm_mm;
 621        unsigned long old_start = vma->vm_start;
 622        unsigned long old_end = vma->vm_end;
 623        unsigned long length = old_end - old_start;
 624        unsigned long new_start = old_start - shift;
 625        unsigned long new_end = old_end - shift;
 626        struct mmu_gather tlb;
 627
 628        BUG_ON(new_start > new_end);
 629
 630        /*
 631         * ensure there are no vmas between where we want to go
 632         * and where we are
 633         */
 634        if (vma != find_vma(mm, new_start))
 635                return -EFAULT;
 636
 637        /*
 638         * cover the whole range: [new_start, old_end)
 639         */
 640        if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 641                return -ENOMEM;
 642
 643        /*
 644         * move the page tables downwards, on failure we rely on
 645         * process cleanup to remove whatever mess we made.
 646         */
 647        if (length != move_page_tables(vma, old_start,
 648                                       vma, new_start, length, false))
 649                return -ENOMEM;
 650
 651        lru_add_drain();
 652        tlb_gather_mmu(&tlb, mm, old_start, old_end);
 653        if (new_end > old_start) {
 654                /*
 655                 * when the old and new regions overlap clear from new_end.
 656                 */
 657                free_pgd_range(&tlb, new_end, old_end, new_end,
 658                        vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 659        } else {
 660                /*
 661                 * otherwise, clean from old_start; this is done to not touch
 662                 * the address space in [new_end, old_start) some architectures
 663                 * have constraints on va-space that make this illegal (IA64) -
 664                 * for the others its just a little faster.
 665                 */
 666                free_pgd_range(&tlb, old_start, old_end, new_end,
 667                        vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 668        }
 669        tlb_finish_mmu(&tlb, old_start, old_end);
 670
 671        /*
 672         * Shrink the vma to just the new range.  Always succeeds.
 673         */
 674        vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 675
 676        return 0;
 677}
 678
 679/*
 680 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 681 * the stack is optionally relocated, and some extra space is added.
 682 */
 683int setup_arg_pages(struct linux_binprm *bprm,
 684                    unsigned long stack_top,
 685                    int executable_stack)
 686{
 687        unsigned long ret;
 688        unsigned long stack_shift;
 689        struct mm_struct *mm = current->mm;
 690        struct vm_area_struct *vma = bprm->vma;
 691        struct vm_area_struct *prev = NULL;
 692        unsigned long vm_flags;
 693        unsigned long stack_base;
 694        unsigned long stack_size;
 695        unsigned long stack_expand;
 696        unsigned long rlim_stack;
 697
 698#ifdef CONFIG_STACK_GROWSUP
 699        /* Limit stack size */
 700        stack_base = rlimit_max(RLIMIT_STACK);
 701        if (stack_base > STACK_SIZE_MAX)
 702                stack_base = STACK_SIZE_MAX;
 703
 704        /* Add space for stack randomization. */
 705        stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 706
 707        /* Make sure we didn't let the argument array grow too large. */
 708        if (vma->vm_end - vma->vm_start > stack_base)
 709                return -ENOMEM;
 710
 711        stack_base = PAGE_ALIGN(stack_top - stack_base);
 712
 713        stack_shift = vma->vm_start - stack_base;
 714        mm->arg_start = bprm->p - stack_shift;
 715        bprm->p = vma->vm_end - stack_shift;
 716#else
 717        stack_top = arch_align_stack(stack_top);
 718        stack_top = PAGE_ALIGN(stack_top);
 719
 720        if (unlikely(stack_top < mmap_min_addr) ||
 721            unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 722                return -ENOMEM;
 723
 724        stack_shift = vma->vm_end - stack_top;
 725
 726        bprm->p -= stack_shift;
 727        mm->arg_start = bprm->p;
 728#endif
 729
 730        if (bprm->loader)
 731                bprm->loader -= stack_shift;
 732        bprm->exec -= stack_shift;
 733
 734        if (down_write_killable(&mm->mmap_sem))
 735                return -EINTR;
 736
 737        vm_flags = VM_STACK_FLAGS;
 738
 739        /*
 740         * Adjust stack execute permissions; explicitly enable for
 741         * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 742         * (arch default) otherwise.
 743         */
 744        if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 745                vm_flags |= VM_EXEC;
 746        else if (executable_stack == EXSTACK_DISABLE_X)
 747                vm_flags &= ~VM_EXEC;
 748        vm_flags |= mm->def_flags;
 749        vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 750
 751        ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 752                        vm_flags);
 753        if (ret)
 754                goto out_unlock;
 755        BUG_ON(prev != vma);
 756
 757        /* Move stack pages down in memory. */
 758        if (stack_shift) {
 759                ret = shift_arg_pages(vma, stack_shift);
 760                if (ret)
 761                        goto out_unlock;
 762        }
 763
 764        /* mprotect_fixup is overkill to remove the temporary stack flags */
 765        vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 766
 767        stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 768        stack_size = vma->vm_end - vma->vm_start;
 769        /*
 770         * Align this down to a page boundary as expand_stack
 771         * will align it up.
 772         */
 773        rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
 774#ifdef CONFIG_STACK_GROWSUP
 775        if (stack_size + stack_expand > rlim_stack)
 776                stack_base = vma->vm_start + rlim_stack;
 777        else
 778                stack_base = vma->vm_end + stack_expand;
 779#else
 780        if (stack_size + stack_expand > rlim_stack)
 781                stack_base = vma->vm_end - rlim_stack;
 782        else
 783                stack_base = vma->vm_start - stack_expand;
 784#endif
 785        current->mm->start_stack = bprm->p;
 786        ret = expand_stack(vma, stack_base);
 787        if (ret)
 788                ret = -EFAULT;
 789
 790out_unlock:
 791        up_write(&mm->mmap_sem);
 792        return ret;
 793}
 794EXPORT_SYMBOL(setup_arg_pages);
 795
 796#else
 797
 798/*
 799 * Transfer the program arguments and environment from the holding pages
 800 * onto the stack. The provided stack pointer is adjusted accordingly.
 801 */
 802int transfer_args_to_stack(struct linux_binprm *bprm,
 803                           unsigned long *sp_location)
 804{
 805        unsigned long index, stop, sp;
 806        int ret = 0;
 807
 808        stop = bprm->p >> PAGE_SHIFT;
 809        sp = *sp_location;
 810
 811        for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 812                unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 813                char *src = kmap(bprm->page[index]) + offset;
 814                sp -= PAGE_SIZE - offset;
 815                if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 816                        ret = -EFAULT;
 817                kunmap(bprm->page[index]);
 818                if (ret)
 819                        goto out;
 820        }
 821
 822        *sp_location = sp;
 823
 824out:
 825        return ret;
 826}
 827EXPORT_SYMBOL(transfer_args_to_stack);
 828
 829#endif /* CONFIG_MMU */
 830
 831static struct file *do_open_execat(int fd, struct filename *name, int flags)
 832{
 833        struct file *file;
 834        int err;
 835        struct open_flags open_exec_flags = {
 836                .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 837                .acc_mode = MAY_EXEC,
 838                .intent = LOOKUP_OPEN,
 839                .lookup_flags = LOOKUP_FOLLOW,
 840        };
 841
 842        if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 843                return ERR_PTR(-EINVAL);
 844        if (flags & AT_SYMLINK_NOFOLLOW)
 845                open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 846        if (flags & AT_EMPTY_PATH)
 847                open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 848
 849        file = do_filp_open(fd, name, &open_exec_flags);
 850        if (IS_ERR(file))
 851                goto out;
 852
 853        err = -EACCES;
 854        if (!S_ISREG(file_inode(file)->i_mode))
 855                goto exit;
 856
 857        if (path_noexec(&file->f_path))
 858                goto exit;
 859
 860        err = deny_write_access(file);
 861        if (err)
 862                goto exit;
 863
 864        if (name->name[0] != '\0')
 865                fsnotify_open(file);
 866
 867out:
 868        return file;
 869
 870exit:
 871        fput(file);
 872        return ERR_PTR(err);
 873}
 874
 875struct file *open_exec(const char *name)
 876{
 877        struct filename *filename = getname_kernel(name);
 878        struct file *f = ERR_CAST(filename);
 879
 880        if (!IS_ERR(filename)) {
 881                f = do_open_execat(AT_FDCWD, filename, 0);
 882                putname(filename);
 883        }
 884        return f;
 885}
 886EXPORT_SYMBOL(open_exec);
 887
 888int kernel_read_file(struct file *file, void **buf, loff_t *size,
 889                     loff_t max_size, enum kernel_read_file_id id)
 890{
 891        loff_t i_size, pos;
 892        ssize_t bytes = 0;
 893        int ret;
 894
 895        if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 896                return -EINVAL;
 897
 898        ret = security_kernel_read_file(file, id);
 899        if (ret)
 900                return ret;
 901
 902        ret = deny_write_access(file);
 903        if (ret)
 904                return ret;
 905
 906        i_size = i_size_read(file_inode(file));
 907        if (max_size > 0 && i_size > max_size) {
 908                ret = -EFBIG;
 909                goto out;
 910        }
 911        if (i_size <= 0) {
 912                ret = -EINVAL;
 913                goto out;
 914        }
 915
 916        if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 917                *buf = vmalloc(i_size);
 918        if (!*buf) {
 919                ret = -ENOMEM;
 920                goto out;
 921        }
 922
 923        pos = 0;
 924        while (pos < i_size) {
 925                bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 926                if (bytes < 0) {
 927                        ret = bytes;
 928                        goto out;
 929                }
 930
 931                if (bytes == 0)
 932                        break;
 933        }
 934
 935        if (pos != i_size) {
 936                ret = -EIO;
 937                goto out_free;
 938        }
 939
 940        ret = security_kernel_post_read_file(file, *buf, i_size, id);
 941        if (!ret)
 942                *size = pos;
 943
 944out_free:
 945        if (ret < 0) {
 946                if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
 947                        vfree(*buf);
 948                        *buf = NULL;
 949                }
 950        }
 951
 952out:
 953        allow_write_access(file);
 954        return ret;
 955}
 956EXPORT_SYMBOL_GPL(kernel_read_file);
 957
 958int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
 959                               loff_t max_size, enum kernel_read_file_id id)
 960{
 961        struct file *file;
 962        int ret;
 963
 964        if (!path || !*path)
 965                return -EINVAL;
 966
 967        file = filp_open(path, O_RDONLY, 0);
 968        if (IS_ERR(file))
 969                return PTR_ERR(file);
 970
 971        ret = kernel_read_file(file, buf, size, max_size, id);
 972        fput(file);
 973        return ret;
 974}
 975EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
 976
 977int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
 978                             enum kernel_read_file_id id)
 979{
 980        struct fd f = fdget(fd);
 981        int ret = -EBADF;
 982
 983        if (!f.file)
 984                goto out;
 985
 986        ret = kernel_read_file(f.file, buf, size, max_size, id);
 987out:
 988        fdput(f);
 989        return ret;
 990}
 991EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
 992
 993ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 994{
 995        ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 996        if (res > 0)
 997                flush_icache_range(addr, addr + len);
 998        return res;
 999}
1000EXPORT_SYMBOL(read_code);
1001
1002static int exec_mmap(struct mm_struct *mm)
1003{
1004        struct task_struct *tsk;
1005        struct mm_struct *old_mm, *active_mm;
1006
1007        /* Notify parent that we're no longer interested in the old VM */
1008        tsk = current;
1009        old_mm = current->mm;
1010        mm_release(tsk, old_mm);
1011
1012        if (old_mm) {
1013                sync_mm_rss(old_mm);
1014                /*
1015                 * Make sure that if there is a core dump in progress
1016                 * for the old mm, we get out and die instead of going
1017                 * through with the exec.  We must hold mmap_sem around
1018                 * checking core_state and changing tsk->mm.
1019                 */
1020                down_read(&old_mm->mmap_sem);
1021                if (unlikely(old_mm->core_state)) {
1022                        up_read(&old_mm->mmap_sem);
1023                        return -EINTR;
1024                }
1025        }
1026        task_lock(tsk);
1027        active_mm = tsk->active_mm;
1028        tsk->mm = mm;
1029        tsk->active_mm = mm;
1030        activate_mm(active_mm, mm);
1031        tsk->mm->vmacache_seqnum = 0;
1032        vmacache_flush(tsk);
1033        task_unlock(tsk);
1034        if (old_mm) {
1035                up_read(&old_mm->mmap_sem);
1036                BUG_ON(active_mm != old_mm);
1037                setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1038                mm_update_next_owner(old_mm);
1039                mmput(old_mm);
1040                return 0;
1041        }
1042        mmdrop(active_mm);
1043        return 0;
1044}
1045
1046/*
1047 * This function makes sure the current process has its own signal table,
1048 * so that flush_signal_handlers can later reset the handlers without
1049 * disturbing other processes.  (Other processes might share the signal
1050 * table via the CLONE_SIGHAND option to clone().)
1051 */
1052static int de_thread(struct task_struct *tsk)
1053{
1054        struct signal_struct *sig = tsk->signal;
1055        struct sighand_struct *oldsighand = tsk->sighand;
1056        spinlock_t *lock = &oldsighand->siglock;
1057
1058        if (thread_group_empty(tsk))
1059                goto no_thread_group;
1060
1061        /*
1062         * Kill all other threads in the thread group.
1063         */
1064        spin_lock_irq(lock);
1065        if (signal_group_exit(sig)) {
1066                /*
1067                 * Another group action in progress, just
1068                 * return so that the signal is processed.
1069                 */
1070                spin_unlock_irq(lock);
1071                return -EAGAIN;
1072        }
1073
1074        sig->group_exit_task = tsk;
1075        sig->notify_count = zap_other_threads(tsk);
1076        if (!thread_group_leader(tsk))
1077                sig->notify_count--;
1078
1079        while (sig->notify_count) {
1080                __set_current_state(TASK_KILLABLE);
1081                spin_unlock_irq(lock);
1082                schedule();
1083                if (unlikely(__fatal_signal_pending(tsk)))
1084                        goto killed;
1085                spin_lock_irq(lock);
1086        }
1087        spin_unlock_irq(lock);
1088
1089        /*
1090         * At this point all other threads have exited, all we have to
1091         * do is to wait for the thread group leader to become inactive,
1092         * and to assume its PID:
1093         */
1094        if (!thread_group_leader(tsk)) {
1095                struct task_struct *leader = tsk->group_leader;
1096
1097                for (;;) {
1098                        cgroup_threadgroup_change_begin(tsk);
1099                        write_lock_irq(&tasklist_lock);
1100                        /*
1101                         * Do this under tasklist_lock to ensure that
1102                         * exit_notify() can't miss ->group_exit_task
1103                         */
1104                        sig->notify_count = -1;
1105                        if (likely(leader->exit_state))
1106                                break;
1107                        __set_current_state(TASK_KILLABLE);
1108                        write_unlock_irq(&tasklist_lock);
1109                        cgroup_threadgroup_change_end(tsk);
1110                        schedule();
1111                        if (unlikely(__fatal_signal_pending(tsk)))
1112                                goto killed;
1113                }
1114
1115                /*
1116                 * The only record we have of the real-time age of a
1117                 * process, regardless of execs it's done, is start_time.
1118                 * All the past CPU time is accumulated in signal_struct
1119                 * from sister threads now dead.  But in this non-leader
1120                 * exec, nothing survives from the original leader thread,
1121                 * whose birth marks the true age of this process now.
1122                 * When we take on its identity by switching to its PID, we
1123                 * also take its birthdate (always earlier than our own).
1124                 */
1125                tsk->start_time = leader->start_time;
1126                tsk->real_start_time = leader->real_start_time;
1127
1128                BUG_ON(!same_thread_group(leader, tsk));
1129                BUG_ON(has_group_leader_pid(tsk));
1130                /*
1131                 * An exec() starts a new thread group with the
1132                 * TGID of the previous thread group. Rehash the
1133                 * two threads with a switched PID, and release
1134                 * the former thread group leader:
1135                 */
1136
1137                /* Become a process group leader with the old leader's pid.
1138                 * The old leader becomes a thread of the this thread group.
1139                 * Note: The old leader also uses this pid until release_task
1140                 *       is called.  Odd but simple and correct.
1141                 */
1142                tsk->pid = leader->pid;
1143                change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1144                transfer_pid(leader, tsk, PIDTYPE_PGID);
1145                transfer_pid(leader, tsk, PIDTYPE_SID);
1146
1147                list_replace_rcu(&leader->tasks, &tsk->tasks);
1148                list_replace_init(&leader->sibling, &tsk->sibling);
1149
1150                tsk->group_leader = tsk;
1151                leader->group_leader = tsk;
1152
1153                tsk->exit_signal = SIGCHLD;
1154                leader->exit_signal = -1;
1155
1156                BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1157                leader->exit_state = EXIT_DEAD;
1158
1159                /*
1160                 * We are going to release_task()->ptrace_unlink() silently,
1161                 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1162                 * the tracer wont't block again waiting for this thread.
1163                 */
1164                if (unlikely(leader->ptrace))
1165                        __wake_up_parent(leader, leader->parent);
1166                write_unlock_irq(&tasklist_lock);
1167                cgroup_threadgroup_change_end(tsk);
1168
1169                release_task(leader);
1170        }
1171
1172        sig->group_exit_task = NULL;
1173        sig->notify_count = 0;
1174
1175no_thread_group:
1176        /* we have changed execution domain */
1177        tsk->exit_signal = SIGCHLD;
1178
1179#ifdef CONFIG_POSIX_TIMERS
1180        exit_itimers(sig);
1181        flush_itimer_signals();
1182#endif
1183
1184        if (atomic_read(&oldsighand->count) != 1) {
1185                struct sighand_struct *newsighand;
1186                /*
1187                 * This ->sighand is shared with the CLONE_SIGHAND
1188                 * but not CLONE_THREAD task, switch to the new one.
1189                 */
1190                newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1191                if (!newsighand)
1192                        return -ENOMEM;
1193
1194                atomic_set(&newsighand->count, 1);
1195                memcpy(newsighand->action, oldsighand->action,
1196                       sizeof(newsighand->action));
1197
1198                write_lock_irq(&tasklist_lock);
1199                spin_lock(&oldsighand->siglock);
1200                rcu_assign_pointer(tsk->sighand, newsighand);
1201                spin_unlock(&oldsighand->siglock);
1202                write_unlock_irq(&tasklist_lock);
1203
1204                __cleanup_sighand(oldsighand);
1205        }
1206
1207        BUG_ON(!thread_group_leader(tsk));
1208        return 0;
1209
1210killed:
1211        /* protects against exit_notify() and __exit_signal() */
1212        read_lock(&tasklist_lock);
1213        sig->group_exit_task = NULL;
1214        sig->notify_count = 0;
1215        read_unlock(&tasklist_lock);
1216        return -EAGAIN;
1217}
1218
1219char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1220{
1221        task_lock(tsk);
1222        strncpy(buf, tsk->comm, buf_size);
1223        task_unlock(tsk);
1224        return buf;
1225}
1226EXPORT_SYMBOL_GPL(__get_task_comm);
1227
1228/*
1229 * These functions flushes out all traces of the currently running executable
1230 * so that a new one can be started
1231 */
1232
1233void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1234{
1235        task_lock(tsk);
1236        trace_task_rename(tsk, buf);
1237        strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1238        task_unlock(tsk);
1239        perf_event_comm(tsk, exec);
1240}
1241
1242/*
1243 * Calling this is the point of no return. None of the failures will be
1244 * seen by userspace since either the process is already taking a fatal
1245 * signal (via de_thread() or coredump), or will have SEGV raised
1246 * (after exec_mmap()) by search_binary_handlers (see below).
1247 */
1248int flush_old_exec(struct linux_binprm * bprm)
1249{
1250        int retval;
1251
1252        /*
1253         * Make sure we have a private signal table and that
1254         * we are unassociated from the previous thread group.
1255         */
1256        retval = de_thread(current);
1257        if (retval)
1258                goto out;
1259
1260        /*
1261         * Must be called _before_ exec_mmap() as bprm->mm is
1262         * not visibile until then. This also enables the update
1263         * to be lockless.
1264         */
1265        set_mm_exe_file(bprm->mm, bprm->file);
1266
1267        /*
1268         * Release all of the old mmap stuff
1269         */
1270        acct_arg_size(bprm, 0);
1271        retval = exec_mmap(bprm->mm);
1272        if (retval)
1273                goto out;
1274
1275        /*
1276         * After clearing bprm->mm (to mark that current is using the
1277         * prepared mm now), we have nothing left of the original
1278         * process. If anything from here on returns an error, the check
1279         * in search_binary_handler() will SEGV current.
1280         */
1281        bprm->mm = NULL;
1282
1283        set_fs(USER_DS);
1284        current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1285                                        PF_NOFREEZE | PF_NO_SETAFFINITY);
1286        flush_thread();
1287        current->personality &= ~bprm->per_clear;
1288
1289        /*
1290         * We have to apply CLOEXEC before we change whether the process is
1291         * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1292         * trying to access the should-be-closed file descriptors of a process
1293         * undergoing exec(2).
1294         */
1295        do_close_on_exec(current->files);
1296        return 0;
1297
1298out:
1299        return retval;
1300}
1301EXPORT_SYMBOL(flush_old_exec);
1302
1303void would_dump(struct linux_binprm *bprm, struct file *file)
1304{
1305        struct inode *inode = file_inode(file);
1306        if (inode_permission(inode, MAY_READ) < 0) {
1307                struct user_namespace *old, *user_ns;
1308                bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1309
1310                /* Ensure mm->user_ns contains the executable */
1311                user_ns = old = bprm->mm->user_ns;
1312                while ((user_ns != &init_user_ns) &&
1313                       !privileged_wrt_inode_uidgid(user_ns, inode))
1314                        user_ns = user_ns->parent;
1315
1316                if (old != user_ns) {
1317                        bprm->mm->user_ns = get_user_ns(user_ns);
1318                        put_user_ns(old);
1319                }
1320        }
1321}
1322EXPORT_SYMBOL(would_dump);
1323
1324void setup_new_exec(struct linux_binprm * bprm)
1325{
1326        /*
1327         * Once here, prepare_binrpm() will not be called any more, so
1328         * the final state of setuid/setgid/fscaps can be merged into the
1329         * secureexec flag.
1330         */
1331        bprm->secureexec |= bprm->cap_elevated;
1332
1333        if (bprm->secureexec) {
1334                /* Make sure parent cannot signal privileged process. */
1335                current->pdeath_signal = 0;
1336
1337                /*
1338                 * For secureexec, reset the stack limit to sane default to
1339                 * avoid bad behavior from the prior rlimits. This has to
1340                 * happen before arch_pick_mmap_layout(), which examines
1341                 * RLIMIT_STACK, but after the point of no return to avoid
1342                 * needing to clean up the change on failure.
1343                 */
1344                if (current->signal->rlim[RLIMIT_STACK].rlim_cur > _STK_LIM)
1345                        current->signal->rlim[RLIMIT_STACK].rlim_cur = _STK_LIM;
1346        }
1347
1348        arch_pick_mmap_layout(current->mm);
1349
1350        current->sas_ss_sp = current->sas_ss_size = 0;
1351
1352        /*
1353         * Figure out dumpability. Note that this checking only of current
1354         * is wrong, but userspace depends on it. This should be testing
1355         * bprm->secureexec instead.
1356         */
1357        if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1358            !(uid_eq(current_euid(), current_uid()) &&
1359              gid_eq(current_egid(), current_gid())))
1360                set_dumpable(current->mm, suid_dumpable);
1361        else
1362                set_dumpable(current->mm, SUID_DUMP_USER);
1363
1364        arch_setup_new_exec();
1365        perf_event_exec();
1366        __set_task_comm(current, kbasename(bprm->filename), true);
1367
1368        /* Set the new mm task size. We have to do that late because it may
1369         * depend on TIF_32BIT which is only updated in flush_thread() on
1370         * some architectures like powerpc
1371         */
1372        current->mm->task_size = TASK_SIZE;
1373
1374        /* An exec changes our domain. We are no longer part of the thread
1375           group */
1376        current->self_exec_id++;
1377        flush_signal_handlers(current, 0);
1378}
1379EXPORT_SYMBOL(setup_new_exec);
1380
1381/*
1382 * Prepare credentials and lock ->cred_guard_mutex.
1383 * install_exec_creds() commits the new creds and drops the lock.
1384 * Or, if exec fails before, free_bprm() should release ->cred and
1385 * and unlock.
1386 */
1387int prepare_bprm_creds(struct linux_binprm *bprm)
1388{
1389        if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1390                return -ERESTARTNOINTR;
1391
1392        bprm->cred = prepare_exec_creds();
1393        if (likely(bprm->cred))
1394                return 0;
1395
1396        mutex_unlock(&current->signal->cred_guard_mutex);
1397        return -ENOMEM;
1398}
1399
1400static void free_bprm(struct linux_binprm *bprm)
1401{
1402        free_arg_pages(bprm);
1403        if (bprm->cred) {
1404                mutex_unlock(&current->signal->cred_guard_mutex);
1405                abort_creds(bprm->cred);
1406        }
1407        if (bprm->file) {
1408                allow_write_access(bprm->file);
1409                fput(bprm->file);
1410        }
1411        /* If a binfmt changed the interp, free it. */
1412        if (bprm->interp != bprm->filename)
1413                kfree(bprm->interp);
1414        kfree(bprm);
1415}
1416
1417int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1418{
1419        /* If a binfmt changed the interp, free it first. */
1420        if (bprm->interp != bprm->filename)
1421                kfree(bprm->interp);
1422        bprm->interp = kstrdup(interp, GFP_KERNEL);
1423        if (!bprm->interp)
1424                return -ENOMEM;
1425        return 0;
1426}
1427EXPORT_SYMBOL(bprm_change_interp);
1428
1429/*
1430 * install the new credentials for this executable
1431 */
1432void install_exec_creds(struct linux_binprm *bprm)
1433{
1434        security_bprm_committing_creds(bprm);
1435
1436        commit_creds(bprm->cred);
1437        bprm->cred = NULL;
1438
1439        /*
1440         * Disable monitoring for regular users
1441         * when executing setuid binaries. Must
1442         * wait until new credentials are committed
1443         * by commit_creds() above
1444         */
1445        if (get_dumpable(current->mm) != SUID_DUMP_USER)
1446                perf_event_exit_task(current);
1447        /*
1448         * cred_guard_mutex must be held at least to this point to prevent
1449         * ptrace_attach() from altering our determination of the task's
1450         * credentials; any time after this it may be unlocked.
1451         */
1452        security_bprm_committed_creds(bprm);
1453        mutex_unlock(&current->signal->cred_guard_mutex);
1454}
1455EXPORT_SYMBOL(install_exec_creds);
1456
1457/*
1458 * determine how safe it is to execute the proposed program
1459 * - the caller must hold ->cred_guard_mutex to protect against
1460 *   PTRACE_ATTACH or seccomp thread-sync
1461 */
1462static void check_unsafe_exec(struct linux_binprm *bprm)
1463{
1464        struct task_struct *p = current, *t;
1465        unsigned n_fs;
1466
1467        if (p->ptrace)
1468                bprm->unsafe |= LSM_UNSAFE_PTRACE;
1469
1470        /*
1471         * This isn't strictly necessary, but it makes it harder for LSMs to
1472         * mess up.
1473         */
1474        if (task_no_new_privs(current))
1475                bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1476
1477        t = p;
1478        n_fs = 1;
1479        spin_lock(&p->fs->lock);
1480        rcu_read_lock();
1481        while_each_thread(p, t) {
1482                if (t->fs == p->fs)
1483                        n_fs++;
1484        }
1485        rcu_read_unlock();
1486
1487        if (p->fs->users > n_fs)
1488                bprm->unsafe |= LSM_UNSAFE_SHARE;
1489        else
1490                p->fs->in_exec = 1;
1491        spin_unlock(&p->fs->lock);
1492}
1493
1494static void bprm_fill_uid(struct linux_binprm *bprm)
1495{
1496        struct inode *inode;
1497        unsigned int mode;
1498        kuid_t uid;
1499        kgid_t gid;
1500
1501        /*
1502         * Since this can be called multiple times (via prepare_binprm),
1503         * we must clear any previous work done when setting set[ug]id
1504         * bits from any earlier bprm->file uses (for example when run
1505         * first for a setuid script then again for its interpreter).
1506         */
1507        bprm->cred->euid = current_euid();
1508        bprm->cred->egid = current_egid();
1509
1510        if (!mnt_may_suid(bprm->file->f_path.mnt))
1511                return;
1512
1513        if (task_no_new_privs(current))
1514                return;
1515
1516        inode = bprm->file->f_path.dentry->d_inode;
1517        mode = READ_ONCE(inode->i_mode);
1518        if (!(mode & (S_ISUID|S_ISGID)))
1519                return;
1520
1521        /* Be careful if suid/sgid is set */
1522        inode_lock(inode);
1523
1524        /* reload atomically mode/uid/gid now that lock held */
1525        mode = inode->i_mode;
1526        uid = inode->i_uid;
1527        gid = inode->i_gid;
1528        inode_unlock(inode);
1529
1530        /* We ignore suid/sgid if there are no mappings for them in the ns */
1531        if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1532                 !kgid_has_mapping(bprm->cred->user_ns, gid))
1533                return;
1534
1535        if (mode & S_ISUID) {
1536                bprm->per_clear |= PER_CLEAR_ON_SETID;
1537                bprm->cred->euid = uid;
1538        }
1539
1540        if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1541                bprm->per_clear |= PER_CLEAR_ON_SETID;
1542                bprm->cred->egid = gid;
1543        }
1544}
1545
1546/*
1547 * Fill the binprm structure from the inode.
1548 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1549 *
1550 * This may be called multiple times for binary chains (scripts for example).
1551 */
1552int prepare_binprm(struct linux_binprm *bprm)
1553{
1554        int retval;
1555        loff_t pos = 0;
1556
1557        bprm_fill_uid(bprm);
1558
1559        /* fill in binprm security blob */
1560        retval = security_bprm_set_creds(bprm);
1561        if (retval)
1562                return retval;
1563        bprm->called_set_creds = 1;
1564
1565        memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1566        return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1567}
1568
1569EXPORT_SYMBOL(prepare_binprm);
1570
1571/*
1572 * Arguments are '\0' separated strings found at the location bprm->p
1573 * points to; chop off the first by relocating brpm->p to right after
1574 * the first '\0' encountered.
1575 */
1576int remove_arg_zero(struct linux_binprm *bprm)
1577{
1578        int ret = 0;
1579        unsigned long offset;
1580        char *kaddr;
1581        struct page *page;
1582
1583        if (!bprm->argc)
1584                return 0;
1585
1586        do {
1587                offset = bprm->p & ~PAGE_MASK;
1588                page = get_arg_page(bprm, bprm->p, 0);
1589                if (!page) {
1590                        ret = -EFAULT;
1591                        goto out;
1592                }
1593                kaddr = kmap_atomic(page);
1594
1595                for (; offset < PAGE_SIZE && kaddr[offset];
1596                                offset++, bprm->p++)
1597                        ;
1598
1599                kunmap_atomic(kaddr);
1600                put_arg_page(page);
1601        } while (offset == PAGE_SIZE);
1602
1603        bprm->p++;
1604        bprm->argc--;
1605        ret = 0;
1606
1607out:
1608        return ret;
1609}
1610EXPORT_SYMBOL(remove_arg_zero);
1611
1612#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1613/*
1614 * cycle the list of binary formats handler, until one recognizes the image
1615 */
1616int search_binary_handler(struct linux_binprm *bprm)
1617{
1618        bool need_retry = IS_ENABLED(CONFIG_MODULES);
1619        struct linux_binfmt *fmt;
1620        int retval;
1621
1622        /* This allows 4 levels of binfmt rewrites before failing hard. */
1623        if (bprm->recursion_depth > 5)
1624                return -ELOOP;
1625
1626        retval = security_bprm_check(bprm);
1627        if (retval)
1628                return retval;
1629
1630        retval = -ENOENT;
1631 retry:
1632        read_lock(&binfmt_lock);
1633        list_for_each_entry(fmt, &formats, lh) {
1634                if (!try_module_get(fmt->module))
1635                        continue;
1636                read_unlock(&binfmt_lock);
1637                bprm->recursion_depth++;
1638                retval = fmt->load_binary(bprm);
1639                read_lock(&binfmt_lock);
1640                put_binfmt(fmt);
1641                bprm->recursion_depth--;
1642                if (retval < 0 && !bprm->mm) {
1643                        /* we got to flush_old_exec() and failed after it */
1644                        read_unlock(&binfmt_lock);
1645                        force_sigsegv(SIGSEGV, current);
1646                        return retval;
1647                }
1648                if (retval != -ENOEXEC || !bprm->file) {
1649                        read_unlock(&binfmt_lock);
1650                        return retval;
1651                }
1652        }
1653        read_unlock(&binfmt_lock);
1654
1655        if (need_retry) {
1656                if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1657                    printable(bprm->buf[2]) && printable(bprm->buf[3]))
1658                        return retval;
1659                if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1660                        return retval;
1661                need_retry = false;
1662                goto retry;
1663        }
1664
1665        return retval;
1666}
1667EXPORT_SYMBOL(search_binary_handler);
1668
1669static int exec_binprm(struct linux_binprm *bprm)
1670{
1671        pid_t old_pid, old_vpid;
1672        int ret;
1673
1674        /* Need to fetch pid before load_binary changes it */
1675        old_pid = current->pid;
1676        rcu_read_lock();
1677        old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1678        rcu_read_unlock();
1679
1680        ret = search_binary_handler(bprm);
1681        if (ret >= 0) {
1682                audit_bprm(bprm);
1683                trace_sched_process_exec(current, old_pid, bprm);
1684                ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1685                proc_exec_connector(current);
1686        }
1687
1688        return ret;
1689}
1690
1691/*
1692 * sys_execve() executes a new program.
1693 */
1694static int do_execveat_common(int fd, struct filename *filename,
1695                              struct user_arg_ptr argv,
1696                              struct user_arg_ptr envp,
1697                              int flags)
1698{
1699        char *pathbuf = NULL;
1700        struct linux_binprm *bprm;
1701        struct file *file;
1702        struct files_struct *displaced;
1703        int retval;
1704
1705        if (IS_ERR(filename))
1706                return PTR_ERR(filename);
1707
1708        /*
1709         * We move the actual failure in case of RLIMIT_NPROC excess from
1710         * set*uid() to execve() because too many poorly written programs
1711         * don't check setuid() return code.  Here we additionally recheck
1712         * whether NPROC limit is still exceeded.
1713         */
1714        if ((current->flags & PF_NPROC_EXCEEDED) &&
1715            atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1716                retval = -EAGAIN;
1717                goto out_ret;
1718        }
1719
1720        /* We're below the limit (still or again), so we don't want to make
1721         * further execve() calls fail. */
1722        current->flags &= ~PF_NPROC_EXCEEDED;
1723
1724        retval = unshare_files(&displaced);
1725        if (retval)
1726                goto out_ret;
1727
1728        retval = -ENOMEM;
1729        bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1730        if (!bprm)
1731                goto out_files;
1732
1733        retval = prepare_bprm_creds(bprm);
1734        if (retval)
1735                goto out_free;
1736
1737        check_unsafe_exec(bprm);
1738        current->in_execve = 1;
1739
1740        file = do_open_execat(fd, filename, flags);
1741        retval = PTR_ERR(file);
1742        if (IS_ERR(file))
1743                goto out_unmark;
1744
1745        sched_exec();
1746
1747        bprm->file = file;
1748        if (fd == AT_FDCWD || filename->name[0] == '/') {
1749                bprm->filename = filename->name;
1750        } else {
1751                if (filename->name[0] == '\0')
1752                        pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1753                else
1754                        pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1755                                            fd, filename->name);
1756                if (!pathbuf) {
1757                        retval = -ENOMEM;
1758                        goto out_unmark;
1759                }
1760                /*
1761                 * Record that a name derived from an O_CLOEXEC fd will be
1762                 * inaccessible after exec. Relies on having exclusive access to
1763                 * current->files (due to unshare_files above).
1764                 */
1765                if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1766                        bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1767                bprm->filename = pathbuf;
1768        }
1769        bprm->interp = bprm->filename;
1770
1771        retval = bprm_mm_init(bprm);
1772        if (retval)
1773                goto out_unmark;
1774
1775        bprm->argc = count(argv, MAX_ARG_STRINGS);
1776        if ((retval = bprm->argc) < 0)
1777                goto out;
1778
1779        bprm->envc = count(envp, MAX_ARG_STRINGS);
1780        if ((retval = bprm->envc) < 0)
1781                goto out;
1782
1783        retval = prepare_binprm(bprm);
1784        if (retval < 0)
1785                goto out;
1786
1787        retval = copy_strings_kernel(1, &bprm->filename, bprm);
1788        if (retval < 0)
1789                goto out;
1790
1791        bprm->exec = bprm->p;
1792        retval = copy_strings(bprm->envc, envp, bprm);
1793        if (retval < 0)
1794                goto out;
1795
1796        retval = copy_strings(bprm->argc, argv, bprm);
1797        if (retval < 0)
1798                goto out;
1799
1800        would_dump(bprm, bprm->file);
1801
1802        retval = exec_binprm(bprm);
1803        if (retval < 0)
1804                goto out;
1805
1806        /* execve succeeded */
1807        current->fs->in_exec = 0;
1808        current->in_execve = 0;
1809        membarrier_execve(current);
1810        acct_update_integrals(current);
1811        task_numa_free(current);
1812        free_bprm(bprm);
1813        kfree(pathbuf);
1814        putname(filename);
1815        if (displaced)
1816                put_files_struct(displaced);
1817        return retval;
1818
1819out:
1820        if (bprm->mm) {
1821                acct_arg_size(bprm, 0);
1822                mmput(bprm->mm);
1823        }
1824
1825out_unmark:
1826        current->fs->in_exec = 0;
1827        current->in_execve = 0;
1828
1829out_free:
1830        free_bprm(bprm);
1831        kfree(pathbuf);
1832
1833out_files:
1834        if (displaced)
1835                reset_files_struct(displaced);
1836out_ret:
1837        putname(filename);
1838        return retval;
1839}
1840
1841int do_execve(struct filename *filename,
1842        const char __user *const __user *__argv,
1843        const char __user *const __user *__envp)
1844{
1845        struct user_arg_ptr argv = { .ptr.native = __argv };
1846        struct user_arg_ptr envp = { .ptr.native = __envp };
1847        return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1848}
1849
1850int do_execveat(int fd, struct filename *filename,
1851                const char __user *const __user *__argv,
1852                const char __user *const __user *__envp,
1853                int flags)
1854{
1855        struct user_arg_ptr argv = { .ptr.native = __argv };
1856        struct user_arg_ptr envp = { .ptr.native = __envp };
1857
1858        return do_execveat_common(fd, filename, argv, envp, flags);
1859}
1860
1861#ifdef CONFIG_COMPAT
1862static int compat_do_execve(struct filename *filename,
1863        const compat_uptr_t __user *__argv,
1864        const compat_uptr_t __user *__envp)
1865{
1866        struct user_arg_ptr argv = {
1867                .is_compat = true,
1868                .ptr.compat = __argv,
1869        };
1870        struct user_arg_ptr envp = {
1871                .is_compat = true,
1872                .ptr.compat = __envp,
1873        };
1874        return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1875}
1876
1877static int compat_do_execveat(int fd, struct filename *filename,
1878                              const compat_uptr_t __user *__argv,
1879                              const compat_uptr_t __user *__envp,
1880                              int flags)
1881{
1882        struct user_arg_ptr argv = {
1883                .is_compat = true,
1884                .ptr.compat = __argv,
1885        };
1886        struct user_arg_ptr envp = {
1887                .is_compat = true,
1888                .ptr.compat = __envp,
1889        };
1890        return do_execveat_common(fd, filename, argv, envp, flags);
1891}
1892#endif
1893
1894void set_binfmt(struct linux_binfmt *new)
1895{
1896        struct mm_struct *mm = current->mm;
1897
1898        if (mm->binfmt)
1899                module_put(mm->binfmt->module);
1900
1901        mm->binfmt = new;
1902        if (new)
1903                __module_get(new->module);
1904}
1905EXPORT_SYMBOL(set_binfmt);
1906
1907/*
1908 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1909 */
1910void set_dumpable(struct mm_struct *mm, int value)
1911{
1912        unsigned long old, new;
1913
1914        if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1915                return;
1916
1917        do {
1918                old = READ_ONCE(mm->flags);
1919                new = (old & ~MMF_DUMPABLE_MASK) | value;
1920        } while (cmpxchg(&mm->flags, old, new) != old);
1921}
1922
1923SYSCALL_DEFINE3(execve,
1924                const char __user *, filename,
1925                const char __user *const __user *, argv,
1926                const char __user *const __user *, envp)
1927{
1928        return do_execve(getname(filename), argv, envp);
1929}
1930
1931SYSCALL_DEFINE5(execveat,
1932                int, fd, const char __user *, filename,
1933                const char __user *const __user *, argv,
1934                const char __user *const __user *, envp,
1935                int, flags)
1936{
1937        int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1938
1939        return do_execveat(fd,
1940                           getname_flags(filename, lookup_flags, NULL),
1941                           argv, envp, flags);
1942}
1943
1944#ifdef CONFIG_COMPAT
1945COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1946        const compat_uptr_t __user *, argv,
1947        const compat_uptr_t __user *, envp)
1948{
1949        return compat_do_execve(getname(filename), argv, envp);
1950}
1951
1952COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1953                       const char __user *, filename,
1954                       const compat_uptr_t __user *, argv,
1955                       const compat_uptr_t __user *, envp,
1956                       int,  flags)
1957{
1958        int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1959
1960        return compat_do_execveat(fd,
1961                                  getname_flags(filename, lookup_flags, NULL),
1962                                  argv, envp, flags);
1963}
1964#endif
1965
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.