fs/exec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/exec.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  * #!-checking implemented by tytso.
  10  */
  11 /*
  12  * Demand-loading implemented 01.12.91 - no need to read anything but
  13  * the header into memory. The inode of the executable is put into
  14  * "current->executable", and page faults do the actual loading. Clean.
  15  *
  16  * Once more I can proudly say that linux stood up to being changed: it
  17  * was less than 2 hours work to get demand-loading completely implemented.
  18  *
  19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20  * current->executable is only used by the procfs.  This allows a dispatch
  21  * table to check for several different types  of binary formats.  We keep
  22  * trying until we recognize the file or we run out of supported binary
  23  * formats.
  24  */
  25
  26 #include <linux/kernel_read_file.h>
  27 #include <linux/slab.h>
  28 #include <linux/file.h>
  29 #include <linux/fdtable.h>
  30 #include <linux/mm.h>
  31 #include <linux/vmacache.h>
  32 #include <linux/stat.h>
  33 #include <linux/fcntl.h>
  34 #include <linux/swap.h>
  35 #include <linux/string.h>
  36 #include <linux/init.h>
  37 #include <linux/sched/mm.h>
  38 #include <linux/sched/coredump.h>
  39 #include <linux/sched/signal.h>
  40 #include <linux/sched/numa_balancing.h>
  41 #include <linux/sched/task.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/perf_event.h>
  44 #include <linux/highmem.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/key.h>
  47 #include <linux/personality.h>
  48 #include <linux/binfmts.h>
  49 #include <linux/utsname.h>
  50 #include <linux/pid_namespace.h>
  51 #include <linux/module.h>
  52 #include <linux/namei.h>
  53 #include <linux/mount.h>
  54 #include <linux/security.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/audit.h>
  59 #include <linux/tracehook.h>
  60 #include <linux/kmod.h>
  61 #include <linux/fsnotify.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/oom.h>
  64 #include <linux/compat.h>
  65 #include <linux/vmalloc.h>
  66 #include <linux/io_uring.h>
  67 #include <linux/syscall_user_dispatch.h>
  68
  69 #include <linux/uaccess.h>
  70 #include <asm/mmu_context.h>
  71 #include <asm/tlb.h>
  72
  73 #include <trace/events/task.h>
  74 #include "internal.h"
  75
  76 #include <trace/events/sched.h>
  77
  78 static int bprm_creds_from_file(struct linux_binprm *bprm);
  79
  80 int suid_dumpable = 0;
  81
  82 static LIST_HEAD(formats);
  83 static DEFINE_RWLOCK(binfmt_lock);
  84
  85 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  86 {
  87         write_lock(&binfmt_lock);
  88         insert ? list_add(&fmt->lh, &formats) :
  89                  list_add_tail(&fmt->lh, &formats);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(__register_binfmt);
  94
  95 void unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         write_lock(&binfmt_lock);
  98         list_del(&fmt->lh);
  99         write_unlock(&binfmt_lock);
 100 }
 101
 102 EXPORT_SYMBOL(unregister_binfmt);
 103
 104 static inline void put_binfmt(struct linux_binfmt * fmt)
 105 {
 106         module_put(fmt->module);
 107 }
 108
 109 bool path_noexec(const struct path *path)
 110 {
 111         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 112                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 113 }
 114
 115 #ifdef CONFIG_USELIB
 116 /*
 117  * Note that a shared library must be both readable and executable due to
 118  * security reasons.
 119  *
 120  * Also note that we take the address to load from from the file itself.
 121  */
 122 SYSCALL_DEFINE1(uselib, const char __user *, library)
 123 {
 124         struct linux_binfmt *fmt;
 125         struct file *file;
 126         struct filename *tmp = getname(library);
 127         int error = PTR_ERR(tmp);
 128         static const struct open_flags uselib_flags = {
 129                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 130                 .acc_mode = MAY_READ | MAY_EXEC,
 131                 .intent = LOOKUP_OPEN,
 132                 .lookup_flags = LOOKUP_FOLLOW,
 133         };
 134
 135         if (IS_ERR(tmp))
 136                 goto out;
 137
 138         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 139         putname(tmp);
 140         error = PTR_ERR(file);
 141         if (IS_ERR(file))
 142                 goto out;
 143
 144         /*
 145          * may_open() has already checked for this, so it should be
 146          * impossible to trip now. But we need to be extra cautious
 147          * and check again at the very end too.
 148          */
 149         error = -EACCES;
 150         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 151                          path_noexec(&file->f_path)))
 152                 goto exit;
 153
 154         fsnotify_open(file);
 155
 156         error = -ENOEXEC;
 157
 158         read_lock(&binfmt_lock);
 159         list_for_each_entry(fmt, &formats, lh) {
 160                 if (!fmt->load_shlib)
 161                         continue;
 162                 if (!try_module_get(fmt->module))
 163                         continue;
 164                 read_unlock(&binfmt_lock);
 165                 error = fmt->load_shlib(file);
 166                 read_lock(&binfmt_lock);
 167                 put_binfmt(fmt);
 168                 if (error != -ENOEXEC)
 169                         break;
 170         }
 171         read_unlock(&binfmt_lock);
 172 exit:
 173         fput(file);
 174 out:
 175         return error;
 176 }
 177 #endif /* #ifdef CONFIG_USELIB */
 178
 179 #ifdef CONFIG_MMU
 180 /*
 181  * The nascent bprm->mm is not visible until exec_mmap() but it can
 182  * use a lot of memory, account these pages in current->mm temporary
 183  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 184  * change the counter back via acct_arg_size(0).
 185  */
 186 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 187 {
 188         struct mm_struct *mm = current->mm;
 189         long diff = (long)(pages - bprm->vma_pages);
 190
 191         if (!mm || !diff)
 192                 return;
 193
 194         bprm->vma_pages = pages;
 195         add_mm_counter(mm, MM_ANONPAGES, diff);
 196 }
 197
 198 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 199                 int write)
 200 {
 201         struct page *page;
 202         int ret;
 203         unsigned int gup_flags = FOLL_FORCE;
 204
 205 #ifdef CONFIG_STACK_GROWSUP
 206         if (write) {
 207                 ret = expand_downwards(bprm->vma, pos);
 208                 if (ret < 0)
 209                         return NULL;
 210         }
 211 #endif
 212
 213         if (write)
 214                 gup_flags |= FOLL_WRITE;
 215
 216         /*
 217          * We are doing an exec().  'current' is the process
 218          * doing the exec and bprm->mm is the new process's mm.
 219          */
 220         mmap_read_lock(bprm->mm);
 221         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 222                         &page, NULL, NULL);
 223         mmap_read_unlock(bprm->mm);
 224         if (ret <= 0)
 225                 return NULL;
 226
 227         if (write)
 228                 acct_arg_size(bprm, vma_pages(bprm->vma));
 229
 230         return page;
 231 }
 232
 233 static void put_arg_page(struct page *page)
 234 {
 235         put_page(page);
 236 }
 237
 238 static void free_arg_pages(struct linux_binprm *bprm)
 239 {
 240 }
 241
 242 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 243                 struct page *page)
 244 {
 245         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 246 }
 247
 248 static int __bprm_mm_init(struct linux_binprm *bprm)
 249 {
 250         int err;
 251         struct vm_area_struct *vma = NULL;
 252         struct mm_struct *mm = bprm->mm;
 253
 254         bprm->vma = vma = vm_area_alloc(mm);
 255         if (!vma)
 256                 return -ENOMEM;
 257         vma_set_anonymous(vma);
 258
 259         if (mmap_write_lock_killable(mm)) {
 260                 err = -EINTR;
 261                 goto err_free;
 262         }
 263
 264         /*
 265          * Place the stack at the largest stack address the architecture
 266          * supports. Later, we'll move this to an appropriate place. We don't
 267          * use STACK_TOP because that can depend on attributes which aren't
 268          * configured yet.
 269          */
 270         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 271         vma->vm_end = STACK_TOP_MAX;
 272         vma->vm_start = vma->vm_end - PAGE_SIZE;
 273         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 274         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 275
 276         err = insert_vm_struct(mm, vma);
 277         if (err)
 278                 goto err;
 279
 280         mm->stack_vm = mm->total_vm = 1;
 281         mmap_write_unlock(mm);
 282         bprm->p = vma->vm_end - sizeof(void *);
 283         return 0;
 284 err:
 285         mmap_write_unlock(mm);
 286 err_free:
 287         bprm->vma = NULL;
 288         vm_area_free(vma);
 289         return err;
 290 }
 291
 292 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 293 {
 294         return len <= MAX_ARG_STRLEN;
 295 }
 296
 297 #else
 298
 299 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 300 {
 301 }
 302
 303 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 304                 int write)
 305 {
 306         struct page *page;
 307
 308         page = bprm->page[pos / PAGE_SIZE];
 309         if (!page && write) {
 310                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 311                 if (!page)
 312                         return NULL;
 313                 bprm->page[pos / PAGE_SIZE] = page;
 314         }
 315
 316         return page;
 317 }
 318
 319 static void put_arg_page(struct page *page)
 320 {
 321 }
 322
 323 static void free_arg_page(struct linux_binprm *bprm, int i)
 324 {
 325         if (bprm->page[i]) {
 326                 __free_page(bprm->page[i]);
 327                 bprm->page[i] = NULL;
 328         }
 329 }
 330
 331 static void free_arg_pages(struct linux_binprm *bprm)
 332 {
 333         int i;
 334
 335         for (i = 0; i < MAX_ARG_PAGES; i++)
 336                 free_arg_page(bprm, i);
 337 }
 338
 339 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 340                 struct page *page)
 341 {
 342 }
 343
 344 static int __bprm_mm_init(struct linux_binprm *bprm)
 345 {
 346         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 347         return 0;
 348 }
 349
 350 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 351 {
 352         return len <= bprm->p;
 353 }
 354
 355 #endif /* CONFIG_MMU */
 356
 357 /*
 358  * Create a new mm_struct and populate it with a temporary stack
 359  * vm_area_struct.  We don't have enough context at this point to set the stack
 360  * flags, permissions, and offset, so we use temporary values.  We'll update
 361  * them later in setup_arg_pages().
 362  */
 363 static int bprm_mm_init(struct linux_binprm *bprm)
 364 {
 365         int err;
 366         struct mm_struct *mm = NULL;
 367
 368         bprm->mm = mm = mm_alloc();
 369         err = -ENOMEM;
 370         if (!mm)
 371                 goto err;
 372
 373         /* Save current stack limit for all calculations made during exec. */
 374         task_lock(current->group_leader);
 375         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 376         task_unlock(current->group_leader);
 377
 378         err = __bprm_mm_init(bprm);
 379         if (err)
 380                 goto err;
 381
 382         return 0;
 383
 384 err:
 385         if (mm) {
 386                 bprm->mm = NULL;
 387                 mmdrop(mm);
 388         }
 389
 390         return err;
 391 }
 392
 393 struct user_arg_ptr {
 394 #ifdef CONFIG_COMPAT
 395         bool is_compat;
 396 #endif
 397         union {
 398                 const char __user *const __user *native;
 399 #ifdef CONFIG_COMPAT
 400                 const compat_uptr_t __user *compat;
 401 #endif
 402         } ptr;
 403 };
 404
 405 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 406 {
 407         const char __user *native;
 408
 409 #ifdef CONFIG_COMPAT
 410         if (unlikely(argv.is_compat)) {
 411                 compat_uptr_t compat;
 412
 413                 if (get_user(compat, argv.ptr.compat + nr))
 414                         return ERR_PTR(-EFAULT);
 415
 416                 return compat_ptr(compat);
 417         }
 418 #endif
 419
 420         if (get_user(native, argv.ptr.native + nr))
 421                 return ERR_PTR(-EFAULT);
 422
 423         return native;
 424 }
 425
 426 /*
 427  * count() counts the number of strings in array ARGV.
 428  */
 429 static int count(struct user_arg_ptr argv, int max)
 430 {
 431         int i = 0;
 432
 433         if (argv.ptr.native != NULL) {
 434                 for (;;) {
 435                         const char __user *p = get_user_arg_ptr(argv, i);
 436
 437                         if (!p)
 438                                 break;
 439
 440                         if (IS_ERR(p))
 441                                 return -EFAULT;
 442
 443                         if (i >= max)
 444                                 return -E2BIG;
 445                         ++i;
 446
 447                         if (fatal_signal_pending(current))
 448                                 return -ERESTARTNOHAND;
 449                         cond_resched();
 450                 }
 451         }
 452         return i;
 453 }
 454
 455 static int count_strings_kernel(const char *const *argv)
 456 {
 457         int i;
 458
 459         if (!argv)
 460                 return 0;
 461
 462         for (i = 0; argv[i]; ++i) {
 463                 if (i >= MAX_ARG_STRINGS)
 464                         return -E2BIG;
 465                 if (fatal_signal_pending(current))
 466                         return -ERESTARTNOHAND;
 467                 cond_resched();
 468         }
 469         return i;
 470 }
 471
 472 static int bprm_stack_limits(struct linux_binprm *bprm)
 473 {
 474         unsigned long limit, ptr_size;
 475
 476         /*
 477          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 478          * (whichever is smaller) for the argv+env strings.
 479          * This ensures that:
 480          *  - the remaining binfmt code will not run out of stack space,
 481          *  - the program will have a reasonable amount of stack left
 482          *    to work from.
 483          */
 484         limit = _STK_LIM / 4 * 3;
 485         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 486         /*
 487          * We've historically supported up to 32 pages (ARG_MAX)
 488          * of argument strings even with small stacks
 489          */
 490         limit = max_t(unsigned long, limit, ARG_MAX);
 491         /*
 492          * We must account for the size of all the argv and envp pointers to
 493          * the argv and envp strings, since they will also take up space in
 494          * the stack. They aren't stored until much later when we can't
 495          * signal to the parent that the child has run out of stack space.
 496          * Instead, calculate it here so it's possible to fail gracefully.
 497          *
 498          * In the case of argc = 0, make sure there is space for adding a
 499          * empty string (which will bump argc to 1), to ensure confused
 500          * userspace programs don't start processing from argv[1], thinking
 501          * argc can never be 0, to keep them from walking envp by accident.
 502          * See do_execveat_common().
 503          */
 504         ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
 505         if (limit <= ptr_size)
 506                 return -E2BIG;
 507         limit -= ptr_size;
 508
 509         bprm->argmin = bprm->p - limit;
 510         return 0;
 511 }
 512
 513 /*
 514  * 'copy_strings()' copies argument/environment strings from the old
 515  * processes's memory to the new process's stack.  The call to get_user_pages()
 516  * ensures the destination page is created and not swapped out.
 517  */
 518 static int copy_strings(int argc, struct user_arg_ptr argv,
 519                         struct linux_binprm *bprm)
 520 {
 521         struct page *kmapped_page = NULL;
 522         char *kaddr = NULL;
 523         unsigned long kpos = 0;
 524         int ret;
 525
 526         while (argc-- > 0) {
 527                 const char __user *str;
 528                 int len;
 529                 unsigned long pos;
 530
 531                 ret = -EFAULT;
 532                 str = get_user_arg_ptr(argv, argc);
 533                 if (IS_ERR(str))
 534                         goto out;
 535
 536                 len = strnlen_user(str, MAX_ARG_STRLEN);
 537                 if (!len)
 538                         goto out;
 539
 540                 ret = -E2BIG;
 541                 if (!valid_arg_len(bprm, len))
 542                         goto out;
 543
 544                 /* We're going to work our way backwords. */
 545                 pos = bprm->p;
 546                 str += len;
 547                 bprm->p -= len;
 548 #ifdef CONFIG_MMU
 549                 if (bprm->p < bprm->argmin)
 550                         goto out;
 551 #endif
 552
 553                 while (len > 0) {
 554                         int offset, bytes_to_copy;
 555
 556                         if (fatal_signal_pending(current)) {
 557                                 ret = -ERESTARTNOHAND;
 558                                 goto out;
 559                         }
 560                         cond_resched();
 561
 562                         offset = pos % PAGE_SIZE;
 563                         if (offset == 0)
 564                                 offset = PAGE_SIZE;
 565
 566                         bytes_to_copy = offset;
 567                         if (bytes_to_copy > len)
 568                                 bytes_to_copy = len;
 569
 570                         offset -= bytes_to_copy;
 571                         pos -= bytes_to_copy;
 572                         str -= bytes_to_copy;
 573                         len -= bytes_to_copy;
 574
 575                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 576                                 struct page *page;
 577
 578                                 page = get_arg_page(bprm, pos, 1);
 579                                 if (!page) {
 580                                         ret = -E2BIG;
 581                                         goto out;
 582                                 }
 583
 584                                 if (kmapped_page) {
 585                                         flush_dcache_page(kmapped_page);
 586                                         kunmap(kmapped_page);
 587                                         put_arg_page(kmapped_page);
 588                                 }
 589                                 kmapped_page = page;
 590                                 kaddr = kmap(kmapped_page);
 591                                 kpos = pos & PAGE_MASK;
 592                                 flush_arg_page(bprm, kpos, kmapped_page);
 593                         }
 594                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 595                                 ret = -EFAULT;
 596                                 goto out;
 597                         }
 598                 }
 599         }
 600         ret = 0;
 601 out:
 602         if (kmapped_page) {
 603                 flush_dcache_page(kmapped_page);
 604                 kunmap(kmapped_page);
 605                 put_arg_page(kmapped_page);
 606         }
 607         return ret;
 608 }
 609
 610 /*
 611  * Copy and argument/environment string from the kernel to the processes stack.
 612  */
 613 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 614 {
 615         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 616         unsigned long pos = bprm->p;
 617
 618         if (len == 0)
 619                 return -EFAULT;
 620         if (!valid_arg_len(bprm, len))
 621                 return -E2BIG;
 622
 623         /* We're going to work our way backwards. */
 624         arg += len;
 625         bprm->p -= len;
 626         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 627                 return -E2BIG;
 628
 629         while (len > 0) {
 630                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 631                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 632                 struct page *page;
 633                 char *kaddr;
 634
 635                 pos -= bytes_to_copy;
 636                 arg -= bytes_to_copy;
 637                 len -= bytes_to_copy;
 638
 639                 page = get_arg_page(bprm, pos, 1);
 640                 if (!page)
 641                         return -E2BIG;
 642                 kaddr = kmap_atomic(page);
 643                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 644                 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
 645                 flush_dcache_page(page);
 646                 kunmap_atomic(kaddr);
 647                 put_arg_page(page);
 648         }
 649
 650         return 0;
 651 }
 652 EXPORT_SYMBOL(copy_string_kernel);
 653
 654 static int copy_strings_kernel(int argc, const char *const *argv,
 655                                struct linux_binprm *bprm)
 656 {
 657         while (argc-- > 0) {
 658                 int ret = copy_string_kernel(argv[argc], bprm);
 659                 if (ret < 0)
 660                         return ret;
 661                 if (fatal_signal_pending(current))
 662                         return -ERESTARTNOHAND;
 663                 cond_resched();
 664         }
 665         return 0;
 666 }
 667
 668 #ifdef CONFIG_MMU
 669
 670 /*
 671  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 672  * the binfmt code determines where the new stack should reside, we shift it to
 673  * its final location.  The process proceeds as follows:
 674  *
 675  * 1) Use shift to calculate the new vma endpoints.
 676  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 677  *    arguments passed to subsequent functions are consistent.
 678  * 3) Move vma's page tables to the new range.
 679  * 4) Free up any cleared pgd range.
 680  * 5) Shrink the vma to cover only the new range.
 681  */
 682 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 683 {
 684         struct mm_struct *mm = vma->vm_mm;
 685         unsigned long old_start = vma->vm_start;
 686         unsigned long old_end = vma->vm_end;
 687         unsigned long length = old_end - old_start;
 688         unsigned long new_start = old_start - shift;
 689         unsigned long new_end = old_end - shift;
 690         struct mmu_gather tlb;
 691
 692         BUG_ON(new_start > new_end);
 693
 694         /*
 695          * ensure there are no vmas between where we want to go
 696          * and where we are
 697          */
 698         if (vma != find_vma(mm, new_start))
 699                 return -EFAULT;
 700
 701         /*
 702          * cover the whole range: [new_start, old_end)
 703          */
 704         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 705                 return -ENOMEM;
 706
 707         /*
 708          * move the page tables downwards, on failure we rely on
 709          * process cleanup to remove whatever mess we made.
 710          */
 711         if (length != move_page_tables(vma, old_start,
 712                                        vma, new_start, length, false))
 713                 return -ENOMEM;
 714
 715         lru_add_drain();
 716         tlb_gather_mmu(&tlb, mm);
 717         if (new_end > old_start) {
 718                 /*
 719                  * when the old and new regions overlap clear from new_end.
 720                  */
 721                 free_pgd_range(&tlb, new_end, old_end, new_end,
 722                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 723         } else {
 724                 /*
 725                  * otherwise, clean from old_start; this is done to not touch
 726                  * the address space in [new_end, old_start) some architectures
 727                  * have constraints on va-space that make this illegal (IA64) -
 728                  * for the others its just a little faster.
 729                  */
 730                 free_pgd_range(&tlb, old_start, old_end, new_end,
 731                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 732         }
 733         tlb_finish_mmu(&tlb);
 734
 735         /*
 736          * Shrink the vma to just the new range.  Always succeeds.
 737          */
 738         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 739
 740         return 0;
 741 }
 742
 743 /*
 744  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 745  * the stack is optionally relocated, and some extra space is added.
 746  */
 747 int setup_arg_pages(struct linux_binprm *bprm,
 748                     unsigned long stack_top,
 749                     int executable_stack)
 750 {
 751         unsigned long ret;
 752         unsigned long stack_shift;
 753         struct mm_struct *mm = current->mm;
 754         struct vm_area_struct *vma = bprm->vma;
 755         struct vm_area_struct *prev = NULL;
 756         unsigned long vm_flags;
 757         unsigned long stack_base;
 758         unsigned long stack_size;
 759         unsigned long stack_expand;
 760         unsigned long rlim_stack;
 761
 762 #ifdef CONFIG_STACK_GROWSUP
 763         /* Limit stack size */
 764         stack_base = bprm->rlim_stack.rlim_max;
 765
 766         stack_base = calc_max_stack_size(stack_base);
 767
 768         /* Add space for stack randomization. */
 769         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 770
 771         /* Make sure we didn't let the argument array grow too large. */
 772         if (vma->vm_end - vma->vm_start > stack_base)
 773                 return -ENOMEM;
 774
 775         stack_base = PAGE_ALIGN(stack_top - stack_base);
 776
 777         stack_shift = vma->vm_start - stack_base;
 778         mm->arg_start = bprm->p - stack_shift;
 779         bprm->p = vma->vm_end - stack_shift;
 780 #else
 781         stack_top = arch_align_stack(stack_top);
 782         stack_top = PAGE_ALIGN(stack_top);
 783
 784         if (unlikely(stack_top < mmap_min_addr) ||
 785             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 786                 return -ENOMEM;
 787
 788         stack_shift = vma->vm_end - stack_top;
 789
 790         bprm->p -= stack_shift;
 791         mm->arg_start = bprm->p;
 792 #endif
 793
 794         if (bprm->loader)
 795                 bprm->loader -= stack_shift;
 796         bprm->exec -= stack_shift;
 797
 798         if (mmap_write_lock_killable(mm))
 799                 return -EINTR;
 800
 801         vm_flags = VM_STACK_FLAGS;
 802
 803         /*
 804          * Adjust stack execute permissions; explicitly enable for
 805          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 806          * (arch default) otherwise.
 807          */
 808         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 809                 vm_flags |= VM_EXEC;
 810         else if (executable_stack == EXSTACK_DISABLE_X)
 811                 vm_flags &= ~VM_EXEC;
 812         vm_flags |= mm->def_flags;
 813         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 814
 815         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 816                         vm_flags);
 817         if (ret)
 818                 goto out_unlock;
 819         BUG_ON(prev != vma);
 820
 821         if (unlikely(vm_flags & VM_EXEC)) {
 822                 pr_warn_once("process '%pD4' started with executable stack\n",
 823                              bprm->file);
 824         }
 825
 826         /* Move stack pages down in memory. */
 827         if (stack_shift) {
 828                 ret = shift_arg_pages(vma, stack_shift);
 829                 if (ret)
 830                         goto out_unlock;
 831         }
 832
 833         /* mprotect_fixup is overkill to remove the temporary stack flags */
 834         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 835
 836         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 837         stack_size = vma->vm_end - vma->vm_start;
 838         /*
 839          * Align this down to a page boundary as expand_stack
 840          * will align it up.
 841          */
 842         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 843 #ifdef CONFIG_STACK_GROWSUP
 844         if (stack_size + stack_expand > rlim_stack)
 845                 stack_base = vma->vm_start + rlim_stack;
 846         else
 847                 stack_base = vma->vm_end + stack_expand;
 848 #else
 849         if (stack_size + stack_expand > rlim_stack)
 850                 stack_base = vma->vm_end - rlim_stack;
 851         else
 852                 stack_base = vma->vm_start - stack_expand;
 853 #endif
 854         current->mm->start_stack = bprm->p;
 855         ret = expand_stack(vma, stack_base);
 856         if (ret)
 857                 ret = -EFAULT;
 858
 859 out_unlock:
 860         mmap_write_unlock(mm);
 861         return ret;
 862 }
 863 EXPORT_SYMBOL(setup_arg_pages);
 864
 865 #else
 866
 867 /*
 868  * Transfer the program arguments and environment from the holding pages
 869  * onto the stack. The provided stack pointer is adjusted accordingly.
 870  */
 871 int transfer_args_to_stack(struct linux_binprm *bprm,
 872                            unsigned long *sp_location)
 873 {
 874         unsigned long index, stop, sp;
 875         int ret = 0;
 876
 877         stop = bprm->p >> PAGE_SHIFT;
 878         sp = *sp_location;
 879
 880         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 881                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 882                 char *src = kmap(bprm->page[index]) + offset;
 883                 sp -= PAGE_SIZE - offset;
 884                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 885                         ret = -EFAULT;
 886                 kunmap(bprm->page[index]);
 887                 if (ret)
 888                         goto out;
 889         }
 890
 891         *sp_location = sp;
 892
 893 out:
 894         return ret;
 895 }
 896 EXPORT_SYMBOL(transfer_args_to_stack);
 897
 898 #endif /* CONFIG_MMU */
 899
 900 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 901 {
 902         struct file *file;
 903         int err;
 904         struct open_flags open_exec_flags = {
 905                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 906                 .acc_mode = MAY_EXEC,
 907                 .intent = LOOKUP_OPEN,
 908                 .lookup_flags = LOOKUP_FOLLOW,
 909         };
 910
 911         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 912                 return ERR_PTR(-EINVAL);
 913         if (flags & AT_SYMLINK_NOFOLLOW)
 914                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 915         if (flags & AT_EMPTY_PATH)
 916                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 917
 918         file = do_filp_open(fd, name, &open_exec_flags);
 919         if (IS_ERR(file))
 920                 goto out;
 921
 922         /*
 923          * may_open() has already checked for this, so it should be
 924          * impossible to trip now. But we need to be extra cautious
 925          * and check again at the very end too.
 926          */
 927         err = -EACCES;
 928         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 929                          path_noexec(&file->f_path)))
 930                 goto exit;
 931
 932         err = deny_write_access(file);
 933         if (err)
 934                 goto exit;
 935
 936         if (name->name[0] != '\0')
 937                 fsnotify_open(file);
 938
 939 out:
 940         return file;
 941
 942 exit:
 943         fput(file);
 944         return ERR_PTR(err);
 945 }
 946
 947 struct file *open_exec(const char *name)
 948 {
 949         struct filename *filename = getname_kernel(name);
 950         struct file *f = ERR_CAST(filename);
 951
 952         if (!IS_ERR(filename)) {
 953                 f = do_open_execat(AT_FDCWD, filename, 0);
 954                 putname(filename);
 955         }
 956         return f;
 957 }
 958 EXPORT_SYMBOL(open_exec);
 959
 960 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
 961     defined(CONFIG_BINFMT_ELF_FDPIC)
 962 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 963 {
 964         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 965         if (res > 0)
 966                 flush_icache_user_range(addr, addr + len);
 967         return res;
 968 }
 969 EXPORT_SYMBOL(read_code);
 970 #endif
 971
 972 /*
 973  * Maps the mm_struct mm into the current task struct.
 974  * On success, this function returns with exec_update_lock
 975  * held for writing.
 976  */
 977 static int exec_mmap(struct mm_struct *mm)
 978 {
 979         struct task_struct *tsk;
 980         struct mm_struct *old_mm, *active_mm;
 981         int ret;
 982
 983         /* Notify parent that we're no longer interested in the old VM */
 984         tsk = current;
 985         old_mm = current->mm;
 986         exec_mm_release(tsk, old_mm);
 987         if (old_mm)
 988                 sync_mm_rss(old_mm);
 989
 990         ret = down_write_killable(&tsk->signal->exec_update_lock);
 991         if (ret)
 992                 return ret;
 993
 994         if (old_mm) {
 995                 /*
 996                  * Make sure that if there is a core dump in progress
 997                  * for the old mm, we get out and die instead of going
 998                  * through with the exec.  We must hold mmap_lock around
 999                  * checking core_state and changing tsk->mm.
1000                  */
1001                 mmap_read_lock(old_mm);
1002                 if (unlikely(old_mm->core_state)) {
1003                         mmap_read_unlock(old_mm);
1004                         up_write(&tsk->signal->exec_update_lock);
1005                         return -EINTR;
1006                 }
1007         }
1008
1009         task_lock(tsk);
1010         membarrier_exec_mmap(mm);
1011
1012         local_irq_disable();
1013         active_mm = tsk->active_mm;
1014         tsk->active_mm = mm;
1015         tsk->mm = mm;
1016         /*
1017          * This prevents preemption while active_mm is being loaded and
1018          * it and mm are being updated, which could cause problems for
1019          * lazy tlb mm refcounting when these are updated by context
1020          * switches. Not all architectures can handle irqs off over
1021          * activate_mm yet.
1022          */
1023         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1024                 local_irq_enable();
1025         activate_mm(active_mm, mm);
1026         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1027                 local_irq_enable();
1028         tsk->mm->vmacache_seqnum = 0;
1029         vmacache_flush(tsk);
1030         task_unlock(tsk);
1031         if (old_mm) {
1032                 mmap_read_unlock(old_mm);
1033                 BUG_ON(active_mm != old_mm);
1034                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1035                 mm_update_next_owner(old_mm);
1036                 mmput(old_mm);
1037                 return 0;
1038         }
1039         mmdrop(active_mm);
1040         return 0;
1041 }
1042
1043 static int de_thread(struct task_struct *tsk)
1044 {
1045         struct signal_struct *sig = tsk->signal;
1046         struct sighand_struct *oldsighand = tsk->sighand;
1047         spinlock_t *lock = &oldsighand->siglock;
1048
1049         if (thread_group_empty(tsk))
1050                 goto no_thread_group;
1051
1052         /*
1053          * Kill all other threads in the thread group.
1054          */
1055         spin_lock_irq(lock);
1056         if (signal_group_exit(sig)) {
1057                 /*
1058                  * Another group action in progress, just
1059                  * return so that the signal is processed.
1060                  */
1061                 spin_unlock_irq(lock);
1062                 return -EAGAIN;
1063         }
1064
1065         sig->group_exit_task = tsk;
1066         sig->notify_count = zap_other_threads(tsk);
1067         if (!thread_group_leader(tsk))
1068                 sig->notify_count--;
1069
1070         while (sig->notify_count) {
1071                 __set_current_state(TASK_KILLABLE);
1072                 spin_unlock_irq(lock);
1073                 schedule();
1074                 if (__fatal_signal_pending(tsk))
1075                         goto killed;
1076                 spin_lock_irq(lock);
1077         }
1078         spin_unlock_irq(lock);
1079
1080         /*
1081          * At this point all other threads have exited, all we have to
1082          * do is to wait for the thread group leader to become inactive,
1083          * and to assume its PID:
1084          */
1085         if (!thread_group_leader(tsk)) {
1086                 struct task_struct *leader = tsk->group_leader;
1087
1088                 for (;;) {
1089                         cgroup_threadgroup_change_begin(tsk);
1090                         write_lock_irq(&tasklist_lock);
1091                         /*
1092                          * Do this under tasklist_lock to ensure that
1093                          * exit_notify() can't miss ->group_exit_task
1094                          */
1095                         sig->notify_count = -1;
1096                         if (likely(leader->exit_state))
1097                                 break;
1098                         __set_current_state(TASK_KILLABLE);
1099                         write_unlock_irq(&tasklist_lock);
1100                         cgroup_threadgroup_change_end(tsk);
1101                         schedule();
1102                         if (__fatal_signal_pending(tsk))
1103                                 goto killed;
1104                 }
1105
1106                 /*
1107                  * The only record we have of the real-time age of a
1108                  * process, regardless of execs it's done, is start_time.
1109                  * All the past CPU time is accumulated in signal_struct
1110                  * from sister threads now dead.  But in this non-leader
1111                  * exec, nothing survives from the original leader thread,
1112                  * whose birth marks the true age of this process now.
1113                  * When we take on its identity by switching to its PID, we
1114                  * also take its birthdate (always earlier than our own).
1115                  */
1116                 tsk->start_time = leader->start_time;
1117                 tsk->start_boottime = leader->start_boottime;
1118
1119                 BUG_ON(!same_thread_group(leader, tsk));
1120                 /*
1121                  * An exec() starts a new thread group with the
1122                  * TGID of the previous thread group. Rehash the
1123                  * two threads with a switched PID, and release
1124                  * the former thread group leader:
1125                  */
1126
1127                 /* Become a process group leader with the old leader's pid.
1128                  * The old leader becomes a thread of the this thread group.
1129                  */
1130                 exchange_tids(tsk, leader);
1131                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1132                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1133                 transfer_pid(leader, tsk, PIDTYPE_SID);
1134
1135                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1136                 list_replace_init(&leader->sibling, &tsk->sibling);
1137
1138                 tsk->group_leader = tsk;
1139                 leader->group_leader = tsk;
1140
1141                 tsk->exit_signal = SIGCHLD;
1142                 leader->exit_signal = -1;
1143
1144                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1145                 leader->exit_state = EXIT_DEAD;
1146
1147                 /*
1148                  * We are going to release_task()->ptrace_unlink() silently,
1149                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1150                  * the tracer wont't block again waiting for this thread.
1151                  */
1152                 if (unlikely(leader->ptrace))
1153                         __wake_up_parent(leader, leader->parent);
1154                 write_unlock_irq(&tasklist_lock);
1155                 cgroup_threadgroup_change_end(tsk);
1156
1157                 release_task(leader);
1158         }
1159
1160         sig->group_exit_task = NULL;
1161         sig->notify_count = 0;
1162
1163 no_thread_group:
1164         /* we have changed execution domain */
1165         tsk->exit_signal = SIGCHLD;
1166
1167         BUG_ON(!thread_group_leader(tsk));
1168         return 0;
1169
1170 killed:
1171         /* protects against exit_notify() and __exit_signal() */
1172         read_lock(&tasklist_lock);
1173         sig->group_exit_task = NULL;
1174         sig->notify_count = 0;
1175         read_unlock(&tasklist_lock);
1176         return -EAGAIN;
1177 }
1178
1179
1180 /*
1181  * This function makes sure the current process has its own signal table,
1182  * so that flush_signal_handlers can later reset the handlers without
1183  * disturbing other processes.  (Other processes might share the signal
1184  * table via the CLONE_SIGHAND option to clone().)
1185  */
1186 static int unshare_sighand(struct task_struct *me)
1187 {
1188         struct sighand_struct *oldsighand = me->sighand;
1189
1190         if (refcount_read(&oldsighand->count) != 1) {
1191                 struct sighand_struct *newsighand;
1192                 /*
1193                  * This ->sighand is shared with the CLONE_SIGHAND
1194                  * but not CLONE_THREAD task, switch to the new one.
1195                  */
1196                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1197                 if (!newsighand)
1198                         return -ENOMEM;
1199
1200                 refcount_set(&newsighand->count, 1);
1201                 memcpy(newsighand->action, oldsighand->action,
1202                        sizeof(newsighand->action));
1203
1204                 write_lock_irq(&tasklist_lock);
1205                 spin_lock(&oldsighand->siglock);
1206                 rcu_assign_pointer(me->sighand, newsighand);
1207                 spin_unlock(&oldsighand->siglock);
1208                 write_unlock_irq(&tasklist_lock);
1209
1210                 __cleanup_sighand(oldsighand);
1211         }
1212         return 0;
1213 }
1214
1215 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1216 {
1217         task_lock(tsk);
1218         strncpy(buf, tsk->comm, buf_size);
1219         task_unlock(tsk);
1220         return buf;
1221 }
1222 EXPORT_SYMBOL_GPL(__get_task_comm);
1223
1224 /*
1225  * These functions flushes out all traces of the currently running executable
1226  * so that a new one can be started
1227  */
1228
1229 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1230 {
1231         task_lock(tsk);
1232         trace_task_rename(tsk, buf);
1233         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1234         task_unlock(tsk);
1235         perf_event_comm(tsk, exec);
1236 }
1237
1238 /*
1239  * Calling this is the point of no return. None of the failures will be
1240  * seen by userspace since either the process is already taking a fatal
1241  * signal (via de_thread() or coredump), or will have SEGV raised
1242  * (after exec_mmap()) by search_binary_handler (see below).
1243  */
1244 int begin_new_exec(struct linux_binprm * bprm)
1245 {
1246         struct task_struct *me = current;
1247         int retval;
1248
1249         /* Once we are committed compute the creds */
1250         retval = bprm_creds_from_file(bprm);
1251         if (retval)
1252                 return retval;
1253
1254         /*
1255          * Ensure all future errors are fatal.
1256          */
1257         bprm->point_of_no_return = true;
1258
1259         /*
1260          * Make this the only thread in the thread group.
1261          */
1262         retval = de_thread(me);
1263         if (retval)
1264                 goto out;
1265
1266         /*
1267          * Cancel any io_uring activity across execve
1268          */
1269         io_uring_task_cancel();
1270
1271         /* Ensure the files table is not shared. */
1272         retval = unshare_files();
1273         if (retval)
1274                 goto out;
1275
1276         /*
1277          * Must be called _before_ exec_mmap() as bprm->mm is
1278          * not visibile until then. This also enables the update
1279          * to be lockless.
1280          */
1281         retval = set_mm_exe_file(bprm->mm, bprm->file);
1282         if (retval)
1283                 goto out;
1284
1285         /* If the binary is not readable then enforce mm->dumpable=0 */
1286         would_dump(bprm, bprm->file);
1287         if (bprm->have_execfd)
1288                 would_dump(bprm, bprm->executable);
1289
1290         /*
1291          * Release all of the old mmap stuff
1292          */
1293         acct_arg_size(bprm, 0);
1294         retval = exec_mmap(bprm->mm);
1295         if (retval)
1296                 goto out;
1297
1298         bprm->mm = NULL;
1299
1300 #ifdef CONFIG_POSIX_TIMERS
1301         exit_itimers(me->signal);
1302         flush_itimer_signals();
1303 #endif
1304
1305         /*
1306          * Make the signal table private.
1307          */
1308         retval = unshare_sighand(me);
1309         if (retval)
1310                 goto out_unlock;
1311
1312         /*
1313          * Ensure that the uaccess routines can actually operate on userspace
1314          * pointers:
1315          */
1316         force_uaccess_begin();
1317
1318         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1319                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1320         flush_thread();
1321         me->personality &= ~bprm->per_clear;
1322
1323         clear_syscall_work_syscall_user_dispatch(me);
1324
1325         /*
1326          * We have to apply CLOEXEC before we change whether the process is
1327          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1328          * trying to access the should-be-closed file descriptors of a process
1329          * undergoing exec(2).
1330          */
1331         do_close_on_exec(me->files);
1332
1333         if (bprm->secureexec) {
1334                 /* Make sure parent cannot signal privileged process. */
1335                 me->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 (bprm->rlim_stack.rlim_cur > _STK_LIM)
1345                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1346         }
1347
1348         me->sas_ss_sp = me->sas_ss_size = 0;
1349
1350         /*
1351          * Figure out dumpability. Note that this checking only of current
1352          * is wrong, but userspace depends on it. This should be testing
1353          * bprm->secureexec instead.
1354          */
1355         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1356             !(uid_eq(current_euid(), current_uid()) &&
1357               gid_eq(current_egid(), current_gid())))
1358                 set_dumpable(current->mm, suid_dumpable);
1359         else
1360                 set_dumpable(current->mm, SUID_DUMP_USER);
1361
1362         perf_event_exec();
1363         __set_task_comm(me, kbasename(bprm->filename), true);
1364
1365         /* An exec changes our domain. We are no longer part of the thread
1366            group */
1367         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1368         flush_signal_handlers(me, 0);
1369
1370         retval = set_cred_ucounts(bprm->cred);
1371         if (retval < 0)
1372                 goto out_unlock;
1373
1374         /*
1375          * install the new credentials for this executable
1376          */
1377         security_bprm_committing_creds(bprm);
1378
1379         commit_creds(bprm->cred);
1380         bprm->cred = NULL;
1381
1382         /*
1383          * Disable monitoring for regular users
1384          * when executing setuid binaries. Must
1385          * wait until new credentials are committed
1386          * by commit_creds() above
1387          */
1388         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1389                 perf_event_exit_task(me);
1390         /*
1391          * cred_guard_mutex must be held at least to this point to prevent
1392          * ptrace_attach() from altering our determination of the task's
1393          * credentials; any time after this it may be unlocked.
1394          */
1395         security_bprm_committed_creds(bprm);
1396
1397         /* Pass the opened binary to the interpreter. */
1398         if (bprm->have_execfd) {
1399                 retval = get_unused_fd_flags(0);
1400                 if (retval < 0)
1401                         goto out_unlock;
1402                 fd_install(retval, bprm->executable);
1403                 bprm->executable = NULL;
1404                 bprm->execfd = retval;
1405         }
1406         return 0;
1407
1408 out_unlock:
1409         up_write(&me->signal->exec_update_lock);
1410 out:
1411         return retval;
1412 }
1413 EXPORT_SYMBOL(begin_new_exec);
1414
1415 void would_dump(struct linux_binprm *bprm, struct file *file)
1416 {
1417         struct inode *inode = file_inode(file);
1418         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1419         if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1420                 struct user_namespace *old, *user_ns;
1421                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1422
1423                 /* Ensure mm->user_ns contains the executable */
1424                 user_ns = old = bprm->mm->user_ns;
1425                 while ((user_ns != &init_user_ns) &&
1426                        !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1427                         user_ns = user_ns->parent;
1428
1429                 if (old != user_ns) {
1430                         bprm->mm->user_ns = get_user_ns(user_ns);
1431                         put_user_ns(old);
1432                 }
1433         }
1434 }
1435 EXPORT_SYMBOL(would_dump);
1436
1437 void setup_new_exec(struct linux_binprm * bprm)
1438 {
1439         /* Setup things that can depend upon the personality */
1440         struct task_struct *me = current;
1441
1442         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1443
1444         arch_setup_new_exec();
1445
1446         /* Set the new mm task size. We have to do that late because it may
1447          * depend on TIF_32BIT which is only updated in flush_thread() on
1448          * some architectures like powerpc
1449          */
1450         me->mm->task_size = TASK_SIZE;
1451         up_write(&me->signal->exec_update_lock);
1452         mutex_unlock(&me->signal->cred_guard_mutex);
1453 }
1454 EXPORT_SYMBOL(setup_new_exec);
1455
1456 /* Runs immediately before start_thread() takes over. */
1457 void finalize_exec(struct linux_binprm *bprm)
1458 {
1459         /* Store any stack rlimit changes before starting thread. */
1460         task_lock(current->group_leader);
1461         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1462         task_unlock(current->group_leader);
1463 }
1464 EXPORT_SYMBOL(finalize_exec);
1465
1466 /*
1467  * Prepare credentials and lock ->cred_guard_mutex.
1468  * setup_new_exec() commits the new creds and drops the lock.
1469  * Or, if exec fails before, free_bprm() should release ->cred
1470  * and unlock.
1471  */
1472 static int prepare_bprm_creds(struct linux_binprm *bprm)
1473 {
1474         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1475                 return -ERESTARTNOINTR;
1476
1477         bprm->cred = prepare_exec_creds();
1478         if (likely(bprm->cred))
1479                 return 0;
1480
1481         mutex_unlock(&current->signal->cred_guard_mutex);
1482         return -ENOMEM;
1483 }
1484
1485 static void free_bprm(struct linux_binprm *bprm)
1486 {
1487         if (bprm->mm) {
1488                 acct_arg_size(bprm, 0);
1489                 mmput(bprm->mm);
1490         }
1491         free_arg_pages(bprm);
1492         if (bprm->cred) {
1493                 mutex_unlock(&current->signal->cred_guard_mutex);
1494                 abort_creds(bprm->cred);
1495         }
1496         if (bprm->file) {
1497                 allow_write_access(bprm->file);
1498                 fput(bprm->file);
1499         }
1500         if (bprm->executable)
1501                 fput(bprm->executable);
1502         /* If a binfmt changed the interp, free it. */
1503         if (bprm->interp != bprm->filename)
1504                 kfree(bprm->interp);
1505         kfree(bprm->fdpath);
1506         kfree(bprm);
1507 }
1508
1509 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1510 {
1511         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1512         int retval = -ENOMEM;
1513         if (!bprm)
1514                 goto out;
1515
1516         if (fd == AT_FDCWD || filename->name[0] == '/') {
1517                 bprm->filename = filename->name;
1518         } else {
1519                 if (filename->name[0] == '\0')
1520                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1521                 else
1522                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1523                                                   fd, filename->name);
1524                 if (!bprm->fdpath)
1525                         goto out_free;
1526
1527                 bprm->filename = bprm->fdpath;
1528         }
1529         bprm->interp = bprm->filename;
1530
1531         retval = bprm_mm_init(bprm);
1532         if (retval)
1533                 goto out_free;
1534         return bprm;
1535
1536 out_free:
1537         free_bprm(bprm);
1538 out:
1539         return ERR_PTR(retval);
1540 }
1541
1542 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1543 {
1544         /* If a binfmt changed the interp, free it first. */
1545         if (bprm->interp != bprm->filename)
1546                 kfree(bprm->interp);
1547         bprm->interp = kstrdup(interp, GFP_KERNEL);
1548         if (!bprm->interp)
1549                 return -ENOMEM;
1550         return 0;
1551 }
1552 EXPORT_SYMBOL(bprm_change_interp);
1553
1554 /*
1555  * determine how safe it is to execute the proposed program
1556  * - the caller must hold ->cred_guard_mutex to protect against
1557  *   PTRACE_ATTACH or seccomp thread-sync
1558  */
1559 static void check_unsafe_exec(struct linux_binprm *bprm)
1560 {
1561         struct task_struct *p = current, *t;
1562         unsigned n_fs;
1563
1564         if (p->ptrace)
1565                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1566
1567         /*
1568          * This isn't strictly necessary, but it makes it harder for LSMs to
1569          * mess up.
1570          */
1571         if (task_no_new_privs(current))
1572                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1573
1574         t = p;
1575         n_fs = 1;
1576         spin_lock(&p->fs->lock);
1577         rcu_read_lock();
1578         while_each_thread(p, t) {
1579                 if (t->fs == p->fs)
1580                         n_fs++;
1581         }
1582         rcu_read_unlock();
1583
1584         if (p->fs->users > n_fs)
1585                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1586         else
1587                 p->fs->in_exec = 1;
1588         spin_unlock(&p->fs->lock);
1589 }
1590
1591 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1592 {
1593         /* Handle suid and sgid on files */
1594         struct user_namespace *mnt_userns;
1595         struct inode *inode;
1596         unsigned int mode;
1597         kuid_t uid;
1598         kgid_t gid;
1599
1600         if (!mnt_may_suid(file->f_path.mnt))
1601                 return;
1602
1603         if (task_no_new_privs(current))
1604                 return;
1605
1606         inode = file->f_path.dentry->d_inode;
1607         mode = READ_ONCE(inode->i_mode);
1608         if (!(mode & (S_ISUID|S_ISGID)))
1609                 return;
1610
1611         mnt_userns = file_mnt_user_ns(file);
1612
1613         /* Be careful if suid/sgid is set */
1614         inode_lock(inode);
1615
1616         /* reload atomically mode/uid/gid now that lock held */
1617         mode = inode->i_mode;
1618         uid = i_uid_into_mnt(mnt_userns, inode);
1619         gid = i_gid_into_mnt(mnt_userns, inode);
1620         inode_unlock(inode);
1621
1622         /* We ignore suid/sgid if there are no mappings for them in the ns */
1623         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1624                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1625                 return;
1626
1627         if (mode & S_ISUID) {
1628                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1629                 bprm->cred->euid = uid;
1630         }
1631
1632         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1633                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1634                 bprm->cred->egid = gid;
1635         }
1636 }
1637
1638 /*
1639  * Compute brpm->cred based upon the final binary.
1640  */
1641 static int bprm_creds_from_file(struct linux_binprm *bprm)
1642 {
1643         /* Compute creds based on which file? */
1644         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1645
1646         bprm_fill_uid(bprm, file);
1647         return security_bprm_creds_from_file(bprm, file);
1648 }
1649
1650 /*
1651  * Fill the binprm structure from the inode.
1652  * Read the first BINPRM_BUF_SIZE bytes
1653  *
1654  * This may be called multiple times for binary chains (scripts for example).
1655  */
1656 static int prepare_binprm(struct linux_binprm *bprm)
1657 {
1658         loff_t pos = 0;
1659
1660         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1661         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1662 }
1663
1664 /*
1665  * Arguments are '\0' separated strings found at the location bprm->p
1666  * points to; chop off the first by relocating brpm->p to right after
1667  * the first '\0' encountered.
1668  */
1669 int remove_arg_zero(struct linux_binprm *bprm)
1670 {
1671         int ret = 0;
1672         unsigned long offset;
1673         char *kaddr;
1674         struct page *page;
1675
1676         if (!bprm->argc)
1677                 return 0;
1678
1679         do {
1680                 offset = bprm->p & ~PAGE_MASK;
1681                 page = get_arg_page(bprm, bprm->p, 0);
1682                 if (!page) {
1683                         ret = -EFAULT;
1684                         goto out;
1685                 }
1686                 kaddr = kmap_atomic(page);
1687
1688                 for (; offset < PAGE_SIZE && kaddr[offset];
1689                                 offset++, bprm->p++)
1690                         ;
1691
1692                 kunmap_atomic(kaddr);
1693                 put_arg_page(page);
1694         } while (offset == PAGE_SIZE);
1695
1696         bprm->p++;
1697         bprm->argc--;
1698         ret = 0;
1699
1700 out:
1701         return ret;
1702 }
1703 EXPORT_SYMBOL(remove_arg_zero);
1704
1705 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1706 /*
1707  * cycle the list of binary formats handler, until one recognizes the image
1708  */
1709 static int search_binary_handler(struct linux_binprm *bprm)
1710 {
1711         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1712         struct linux_binfmt *fmt;
1713         int retval;
1714
1715         retval = prepare_binprm(bprm);
1716         if (retval < 0)
1717                 return retval;
1718
1719         retval = security_bprm_check(bprm);
1720         if (retval)
1721                 return retval;
1722
1723         retval = -ENOENT;
1724  retry:
1725         read_lock(&binfmt_lock);
1726         list_for_each_entry(fmt, &formats, lh) {
1727                 if (!try_module_get(fmt->module))
1728                         continue;
1729                 read_unlock(&binfmt_lock);
1730
1731                 retval = fmt->load_binary(bprm);
1732
1733                 read_lock(&binfmt_lock);
1734                 put_binfmt(fmt);
1735                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1736                         read_unlock(&binfmt_lock);
1737                         return retval;
1738                 }
1739         }
1740         read_unlock(&binfmt_lock);
1741
1742         if (need_retry) {
1743                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1744                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1745                         return retval;
1746                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1747                         return retval;
1748                 need_retry = false;
1749                 goto retry;
1750         }
1751
1752         return retval;
1753 }
1754
1755 static int exec_binprm(struct linux_binprm *bprm)
1756 {
1757         pid_t old_pid, old_vpid;
1758         int ret, depth;
1759
1760         /* Need to fetch pid before load_binary changes it */
1761         old_pid = current->pid;
1762         rcu_read_lock();
1763         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1764         rcu_read_unlock();
1765
1766         /* This allows 4 levels of binfmt rewrites before failing hard. */
1767         for (depth = 0;; depth++) {
1768                 struct file *exec;
1769                 if (depth > 5)
1770                         return -ELOOP;
1771
1772                 ret = search_binary_handler(bprm);
1773                 if (ret < 0)
1774                         return ret;
1775                 if (!bprm->interpreter)
1776                         break;
1777
1778                 exec = bprm->file;
1779                 bprm->file = bprm->interpreter;
1780                 bprm->interpreter = NULL;
1781
1782                 allow_write_access(exec);
1783                 if (unlikely(bprm->have_execfd)) {
1784                         if (bprm->executable) {
1785                                 fput(exec);
1786                                 return -ENOEXEC;
1787                         }
1788                         bprm->executable = exec;
1789                 } else
1790                         fput(exec);
1791         }
1792
1793         audit_bprm(bprm);
1794         trace_sched_process_exec(current, old_pid, bprm);
1795         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1796         proc_exec_connector(current);
1797         return 0;
1798 }
1799
1800 /*
1801  * sys_execve() executes a new program.
1802  */
1803 static int bprm_execve(struct linux_binprm *bprm,
1804                        int fd, struct filename *filename, int flags)
1805 {
1806         struct file *file;
1807         int retval;
1808
1809         retval = prepare_bprm_creds(bprm);
1810         if (retval)
1811                 return retval;
1812
1813         check_unsafe_exec(bprm);
1814         current->in_execve = 1;
1815
1816         file = do_open_execat(fd, filename, flags);
1817         retval = PTR_ERR(file);
1818         if (IS_ERR(file))
1819                 goto out_unmark;
1820
1821         sched_exec();
1822
1823         bprm->file = file;
1824         /*
1825          * Record that a name derived from an O_CLOEXEC fd will be
1826          * inaccessible after exec.  This allows the code in exec to
1827          * choose to fail when the executable is not mmaped into the
1828          * interpreter and an open file descriptor is not passed to
1829          * the interpreter.  This makes for a better user experience
1830          * than having the interpreter start and then immediately fail
1831          * when it finds the executable is inaccessible.
1832          */
1833         if (bprm->fdpath && get_close_on_exec(fd))
1834                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1835
1836         /* Set the unchanging part of bprm->cred */
1837         retval = security_bprm_creds_for_exec(bprm);
1838         if (retval)
1839                 goto out;
1840
1841         retval = exec_binprm(bprm);
1842         if (retval < 0)
1843                 goto out;
1844
1845         /* execve succeeded */
1846         current->fs->in_exec = 0;
1847         current->in_execve = 0;
1848         rseq_execve(current);
1849         acct_update_integrals(current);
1850         task_numa_free(current, false);
1851         return retval;
1852
1853 out:
1854         /*
1855          * If past the point of no return ensure the code never
1856          * returns to the userspace process.  Use an existing fatal
1857          * signal if present otherwise terminate the process with
1858          * SIGSEGV.
1859          */
1860         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1861                 force_fatal_sig(SIGSEGV);
1862
1863 out_unmark:
1864         current->fs->in_exec = 0;
1865         current->in_execve = 0;
1866
1867         return retval;
1868 }
1869
1870 static int do_execveat_common(int fd, struct filename *filename,
1871                               struct user_arg_ptr argv,
1872                               struct user_arg_ptr envp,
1873                               int flags)
1874 {
1875         struct linux_binprm *bprm;
1876         int retval;
1877
1878         if (IS_ERR(filename))
1879                 return PTR_ERR(filename);
1880
1881         /*
1882          * We move the actual failure in case of RLIMIT_NPROC excess from
1883          * set*uid() to execve() because too many poorly written programs
1884          * don't check setuid() return code.  Here we additionally recheck
1885          * whether NPROC limit is still exceeded.
1886          */
1887         if ((current->flags & PF_NPROC_EXCEEDED) &&
1888             is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1889                 retval = -EAGAIN;
1890                 goto out_ret;
1891         }
1892
1893         /* We're below the limit (still or again), so we don't want to make
1894          * further execve() calls fail. */
1895         current->flags &= ~PF_NPROC_EXCEEDED;
1896
1897         bprm = alloc_bprm(fd, filename);
1898         if (IS_ERR(bprm)) {
1899                 retval = PTR_ERR(bprm);
1900                 goto out_ret;
1901         }
1902
1903         retval = count(argv, MAX_ARG_STRINGS);
1904         if (retval == 0)
1905                 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1906                              current->comm, bprm->filename);
1907         if (retval < 0)
1908                 goto out_free;
1909         bprm->argc = retval;
1910
1911         retval = count(envp, MAX_ARG_STRINGS);
1912         if (retval < 0)
1913                 goto out_free;
1914         bprm->envc = retval;
1915
1916         retval = bprm_stack_limits(bprm);
1917         if (retval < 0)
1918                 goto out_free;
1919
1920         retval = copy_string_kernel(bprm->filename, bprm);
1921         if (retval < 0)
1922                 goto out_free;
1923         bprm->exec = bprm->p;
1924
1925         retval = copy_strings(bprm->envc, envp, bprm);
1926         if (retval < 0)
1927                 goto out_free;
1928
1929         retval = copy_strings(bprm->argc, argv, bprm);
1930         if (retval < 0)
1931                 goto out_free;
1932
1933         /*
1934          * When argv is empty, add an empty string ("") as argv[0] to
1935          * ensure confused userspace programs that start processing
1936          * from argv[1] won't end up walking envp. See also
1937          * bprm_stack_limits().
1938          */
1939         if (bprm->argc == 0) {
1940                 retval = copy_string_kernel("", bprm);
1941                 if (retval < 0)
1942                         goto out_free;
1943                 bprm->argc = 1;
1944         }
1945
1946         retval = bprm_execve(bprm, fd, filename, flags);
1947 out_free:
1948         free_bprm(bprm);
1949
1950 out_ret:
1951         putname(filename);
1952         return retval;
1953 }
1954
1955 int kernel_execve(const char *kernel_filename,
1956                   const char *const *argv, const char *const *envp)
1957 {
1958         struct filename *filename;
1959         struct linux_binprm *bprm;
1960         int fd = AT_FDCWD;
1961         int retval;
1962
1963         filename = getname_kernel(kernel_filename);
1964         if (IS_ERR(filename))
1965                 return PTR_ERR(filename);
1966
1967         bprm = alloc_bprm(fd, filename);
1968         if (IS_ERR(bprm)) {
1969                 retval = PTR_ERR(bprm);
1970                 goto out_ret;
1971         }
1972
1973         retval = count_strings_kernel(argv);
1974         if (WARN_ON_ONCE(retval == 0))
1975                 retval = -EINVAL;
1976         if (retval < 0)
1977                 goto out_free;
1978         bprm->argc = retval;
1979
1980         retval = count_strings_kernel(envp);
1981         if (retval < 0)
1982                 goto out_free;
1983         bprm->envc = retval;
1984
1985         retval = bprm_stack_limits(bprm);
1986         if (retval < 0)
1987                 goto out_free;
1988
1989         retval = copy_string_kernel(bprm->filename, bprm);
1990         if (retval < 0)
1991                 goto out_free;
1992         bprm->exec = bprm->p;
1993
1994         retval = copy_strings_kernel(bprm->envc, envp, bprm);
1995         if (retval < 0)
1996                 goto out_free;
1997
1998         retval = copy_strings_kernel(bprm->argc, argv, bprm);
1999         if (retval < 0)
2000                 goto out_free;
2001
2002         retval = bprm_execve(bprm, fd, filename, 0);
2003 out_free:
2004         free_bprm(bprm);
2005 out_ret:
2006         putname(filename);
2007         return retval;
2008 }
2009
2010 static int do_execve(struct filename *filename,
2011         const char __user *const __user *__argv,
2012         const char __user *const __user *__envp)
2013 {
2014         struct user_arg_ptr argv = { .ptr.native = __argv };
2015         struct user_arg_ptr envp = { .ptr.native = __envp };
2016         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2017 }
2018
2019 static int do_execveat(int fd, struct filename *filename,
2020                 const char __user *const __user *__argv,
2021                 const char __user *const __user *__envp,
2022                 int flags)
2023 {
2024         struct user_arg_ptr argv = { .ptr.native = __argv };
2025         struct user_arg_ptr envp = { .ptr.native = __envp };
2026
2027         return do_execveat_common(fd, filename, argv, envp, flags);
2028 }
2029
2030 #ifdef CONFIG_COMPAT
2031 static int compat_do_execve(struct filename *filename,
2032         const compat_uptr_t __user *__argv,
2033         const compat_uptr_t __user *__envp)
2034 {
2035         struct user_arg_ptr argv = {
2036                 .is_compat = true,
2037                 .ptr.compat = __argv,
2038         };
2039         struct user_arg_ptr envp = {
2040                 .is_compat = true,
2041                 .ptr.compat = __envp,
2042         };
2043         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2044 }
2045
2046 static int compat_do_execveat(int fd, struct filename *filename,
2047                               const compat_uptr_t __user *__argv,
2048                               const compat_uptr_t __user *__envp,
2049                               int flags)
2050 {
2051         struct user_arg_ptr argv = {
2052                 .is_compat = true,
2053                 .ptr.compat = __argv,
2054         };
2055         struct user_arg_ptr envp = {
2056                 .is_compat = true,
2057                 .ptr.compat = __envp,
2058         };
2059         return do_execveat_common(fd, filename, argv, envp, flags);
2060 }
2061 #endif
2062
2063 void set_binfmt(struct linux_binfmt *new)
2064 {
2065         struct mm_struct *mm = current->mm;
2066
2067         if (mm->binfmt)
2068                 module_put(mm->binfmt->module);
2069
2070         mm->binfmt = new;
2071         if (new)
2072                 __module_get(new->module);
2073 }
2074 EXPORT_SYMBOL(set_binfmt);
2075
2076 /*
2077  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2078  */
2079 void set_dumpable(struct mm_struct *mm, int value)
2080 {
2081         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2082                 return;
2083
2084         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2085 }
2086
2087 SYSCALL_DEFINE3(execve,
2088                 const char __user *, filename,
2089                 const char __user *const __user *, argv,
2090                 const char __user *const __user *, envp)
2091 {
2092         return do_execve(getname(filename), argv, envp);
2093 }
2094
2095 SYSCALL_DEFINE5(execveat,
2096                 int, fd, const char __user *, filename,
2097                 const char __user *const __user *, argv,
2098                 const char __user *const __user *, envp,
2099                 int, flags)
2100 {
2101         return do_execveat(fd,
2102                            getname_uflags(filename, flags),
2103                            argv, envp, flags);
2104 }
2105
2106 #ifdef CONFIG_COMPAT
2107 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2108         const compat_uptr_t __user *, argv,
2109         const compat_uptr_t __user *, envp)
2110 {
2111         return compat_do_execve(getname(filename), argv, envp);
2112 }
2113
2114 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2115                        const char __user *, filename,
2116                        const compat_uptr_t __user *, argv,
2117                        const compat_uptr_t __user *, envp,
2118                        int,  flags)
2119 {
2120         return compat_do_execveat(fd,
2121                                   getname_uflags(filename, flags),
2122                                   argv, envp, flags);
2123 }
2124 #endif