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
  68 #include <linux/uaccess.h>
  69 #include <asm/mmu_context.h>
  70 #include <asm/tlb.h>
  71
  72 #include <trace/events/task.h>
  73 #include "internal.h"
  74
  75 #include <trace/events/sched.h>
  76
  77 static int bprm_creds_from_file(struct linux_binprm *bprm);
  78
  79 int suid_dumpable = 0;
  80
  81 static LIST_HEAD(formats);
  82 static DEFINE_RWLOCK(binfmt_lock);
  83
  84 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  85 {
  86         BUG_ON(!fmt);
  87         if (WARN_ON(!fmt->load_binary))
  88                 return;
  89         write_lock(&binfmt_lock);
  90         insert ? list_add(&fmt->lh, &formats) :
  91                  list_add_tail(&fmt->lh, &formats);
  92         write_unlock(&binfmt_lock);
  93 }
  94
  95 EXPORT_SYMBOL(__register_binfmt);
  96
  97 void unregister_binfmt(struct linux_binfmt * fmt)
  98 {
  99         write_lock(&binfmt_lock);
 100         list_del(&fmt->lh);
 101         write_unlock(&binfmt_lock);
 102 }
 103
 104 EXPORT_SYMBOL(unregister_binfmt);
 105
 106 static inline void put_binfmt(struct linux_binfmt * fmt)
 107 {
 108         module_put(fmt->module);
 109 }
 110
 111 bool path_noexec(const struct path *path)
 112 {
 113         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 114                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 115 }
 116
 117 #ifdef CONFIG_USELIB
 118 /*
 119  * Note that a shared library must be both readable and executable due to
 120  * security reasons.
 121  *
 122  * Also note that we take the address to load from from the file itself.
 123  */
 124 SYSCALL_DEFINE1(uselib, const char __user *, library)
 125 {
 126         struct linux_binfmt *fmt;
 127         struct file *file;
 128         struct filename *tmp = getname(library);
 129         int error = PTR_ERR(tmp);
 130         static const struct open_flags uselib_flags = {
 131                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 132                 .acc_mode = MAY_READ | MAY_EXEC,
 133                 .intent = LOOKUP_OPEN,
 134                 .lookup_flags = LOOKUP_FOLLOW,
 135         };
 136
 137         if (IS_ERR(tmp))
 138                 goto out;
 139
 140         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 141         putname(tmp);
 142         error = PTR_ERR(file);
 143         if (IS_ERR(file))
 144                 goto out;
 145
 146         /*
 147          * may_open() has already checked for this, so it should be
 148          * impossible to trip now. But we need to be extra cautious
 149          * and check again at the very end too.
 150          */
 151         error = -EACCES;
 152         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 153                          path_noexec(&file->f_path)))
 154                 goto exit;
 155
 156         fsnotify_open(file);
 157
 158         error = -ENOEXEC;
 159
 160         read_lock(&binfmt_lock);
 161         list_for_each_entry(fmt, &formats, lh) {
 162                 if (!fmt->load_shlib)
 163                         continue;
 164                 if (!try_module_get(fmt->module))
 165                         continue;
 166                 read_unlock(&binfmt_lock);
 167                 error = fmt->load_shlib(file);
 168                 read_lock(&binfmt_lock);
 169                 put_binfmt(fmt);
 170                 if (error != -ENOEXEC)
 171                         break;
 172         }
 173         read_unlock(&binfmt_lock);
 174 exit:
 175         fput(file);
 176 out:
 177         return error;
 178 }
 179 #endif /* #ifdef CONFIG_USELIB */
 180
 181 #ifdef CONFIG_MMU
 182 /*
 183  * The nascent bprm->mm is not visible until exec_mmap() but it can
 184  * use a lot of memory, account these pages in current->mm temporary
 185  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 186  * change the counter back via acct_arg_size(0).
 187  */
 188 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 189 {
 190         struct mm_struct *mm = current->mm;
 191         long diff = (long)(pages - bprm->vma_pages);
 192
 193         if (!mm || !diff)
 194                 return;
 195
 196         bprm->vma_pages = pages;
 197         add_mm_counter(mm, MM_ANONPAGES, diff);
 198 }
 199
 200 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 201                 int write)
 202 {
 203         struct page *page;
 204         int ret;
 205         unsigned int gup_flags = FOLL_FORCE;
 206
 207 #ifdef CONFIG_STACK_GROWSUP
 208         if (write) {
 209                 ret = expand_downwards(bprm->vma, pos);
 210                 if (ret < 0)
 211                         return NULL;
 212         }
 213 #endif
 214
 215         if (write)
 216                 gup_flags |= FOLL_WRITE;
 217
 218         /*
 219          * We are doing an exec().  'current' is the process
 220          * doing the exec and bprm->mm is the new process's mm.
 221          */
 222         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 223                         &page, NULL, NULL);
 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_kernel_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_kernel_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_kernel_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, old_start, old_end);
 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, old_start, old_end);
 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         if (stack_base > STACK_SIZE_MAX)
 766                 stack_base = STACK_SIZE_MAX;
 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         bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
 892         *sp_location = sp;
 893
 894 out:
 895         return ret;
 896 }
 897 EXPORT_SYMBOL(transfer_args_to_stack);
 898
 899 #endif /* CONFIG_MMU */
 900
 901 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 902 {
 903         struct file *file;
 904         int err;
 905         struct open_flags open_exec_flags = {
 906                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 907                 .acc_mode = MAY_EXEC,
 908                 .intent = LOOKUP_OPEN,
 909                 .lookup_flags = LOOKUP_FOLLOW,
 910         };
 911
 912         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 913                 return ERR_PTR(-EINVAL);
 914         if (flags & AT_SYMLINK_NOFOLLOW)
 915                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 916         if (flags & AT_EMPTY_PATH)
 917                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 918
 919         file = do_filp_open(fd, name, &open_exec_flags);
 920         if (IS_ERR(file))
 921                 goto out;
 922
 923         /*
 924          * may_open() has already checked for this, so it should be
 925          * impossible to trip now. But we need to be extra cautious
 926          * and check again at the very end too.
 927          */
 928         err = -EACCES;
 929         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 930                          path_noexec(&file->f_path)))
 931                 goto exit;
 932
 933         err = deny_write_access(file);
 934         if (err)
 935                 goto exit;
 936
 937         if (name->name[0] != '\0')
 938                 fsnotify_open(file);
 939
 940 out:
 941         return file;
 942
 943 exit:
 944         fput(file);
 945         return ERR_PTR(err);
 946 }
 947
 948 struct file *open_exec(const char *name)
 949 {
 950         struct filename *filename = getname_kernel(name);
 951         struct file *f = ERR_CAST(filename);
 952
 953         if (!IS_ERR(filename)) {
 954                 f = do_open_execat(AT_FDCWD, filename, 0);
 955                 putname(filename);
 956         }
 957         return f;
 958 }
 959 EXPORT_SYMBOL(open_exec);
 960
 961 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
 962     defined(CONFIG_BINFMT_ELF_FDPIC)
 963 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 964 {
 965         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 966         if (res > 0)
 967                 flush_icache_user_range(addr, addr + len);
 968         return res;
 969 }
 970 EXPORT_SYMBOL(read_code);
 971 #endif
 972
 973 /*
 974  * Maps the mm_struct mm into the current task struct.
 975  * On success, this function returns with exec_update_lock
 976  * held for writing.
 977  */
 978 static int exec_mmap(struct mm_struct *mm)
 979 {
 980         struct task_struct *tsk;
 981         struct mm_struct *old_mm, *active_mm;
 982         int ret;
 983
 984         /* Notify parent that we're no longer interested in the old VM */
 985         tsk = current;
 986         old_mm = current->mm;
 987         exec_mm_release(tsk, old_mm);
 988         if (old_mm)
 989                 sync_mm_rss(old_mm);
 990
 991         ret = down_write_killable(&tsk->signal->exec_update_lock);
 992         if (ret)
 993                 return ret;
 994
 995         if (old_mm) {
 996                 /*
 997                  * Make sure that if there is a core dump in progress
 998                  * for the old mm, we get out and die instead of going
 999                  * through with the exec.  We must hold mmap_lock around
1000                  * checking core_state and changing tsk->mm.
1001                  */
1002                 mmap_read_lock(old_mm);
1003                 if (unlikely(old_mm->core_state)) {
1004                         mmap_read_unlock(old_mm);
1005                         up_write(&tsk->signal->exec_update_lock);
1006                         return -EINTR;
1007                 }
1008         }
1009
1010         task_lock(tsk);
1011         membarrier_exec_mmap(mm);
1012
1013         local_irq_disable();
1014         active_mm = tsk->active_mm;
1015         tsk->active_mm = mm;
1016         tsk->mm = mm;
1017         /*
1018          * This prevents preemption while active_mm is being loaded and
1019          * it and mm are being updated, which could cause problems for
1020          * lazy tlb mm refcounting when these are updated by context
1021          * switches. Not all architectures can handle irqs off over
1022          * activate_mm yet.
1023          */
1024         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1025                 local_irq_enable();
1026         activate_mm(active_mm, mm);
1027         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1028                 local_irq_enable();
1029         tsk->mm->vmacache_seqnum = 0;
1030         vmacache_flush(tsk);
1031         task_unlock(tsk);
1032         if (old_mm) {
1033                 mmap_read_unlock(old_mm);
1034                 BUG_ON(active_mm != old_mm);
1035                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1036                 mm_update_next_owner(old_mm);
1037                 mmput(old_mm);
1038                 return 0;
1039         }
1040         mmdrop(active_mm);
1041         return 0;
1042 }
1043
1044 static int de_thread(struct task_struct *tsk)
1045 {
1046         struct signal_struct *sig = tsk->signal;
1047         struct sighand_struct *oldsighand = tsk->sighand;
1048         spinlock_t *lock = &oldsighand->siglock;
1049
1050         if (thread_group_empty(tsk))
1051                 goto no_thread_group;
1052
1053         /*
1054          * Kill all other threads in the thread group.
1055          */
1056         spin_lock_irq(lock);
1057         if (signal_group_exit(sig)) {
1058                 /*
1059                  * Another group action in progress, just
1060                  * return so that the signal is processed.
1061                  */
1062                 spin_unlock_irq(lock);
1063                 return -EAGAIN;
1064         }
1065
1066         sig->group_exit_task = tsk;
1067         sig->notify_count = zap_other_threads(tsk);
1068         if (!thread_group_leader(tsk))
1069                 sig->notify_count--;
1070
1071         while (sig->notify_count) {
1072                 __set_current_state(TASK_KILLABLE);
1073                 spin_unlock_irq(lock);
1074                 schedule();
1075                 if (__fatal_signal_pending(tsk))
1076                         goto killed;
1077                 spin_lock_irq(lock);
1078         }
1079         spin_unlock_irq(lock);
1080
1081         /*
1082          * At this point all other threads have exited, all we have to
1083          * do is to wait for the thread group leader to become inactive,
1084          * and to assume its PID:
1085          */
1086         if (!thread_group_leader(tsk)) {
1087                 struct task_struct *leader = tsk->group_leader;
1088
1089                 for (;;) {
1090                         cgroup_threadgroup_change_begin(tsk);
1091                         write_lock_irq(&tasklist_lock);
1092                         /*
1093                          * Do this under tasklist_lock to ensure that
1094                          * exit_notify() can't miss ->group_exit_task
1095                          */
1096                         sig->notify_count = -1;
1097                         if (likely(leader->exit_state))
1098                                 break;
1099                         __set_current_state(TASK_KILLABLE);
1100                         write_unlock_irq(&tasklist_lock);
1101                         cgroup_threadgroup_change_end(tsk);
1102                         schedule();
1103                         if (__fatal_signal_pending(tsk))
1104                                 goto killed;
1105                 }
1106
1107                 /*
1108                  * The only record we have of the real-time age of a
1109                  * process, regardless of execs it's done, is start_time.
1110                  * All the past CPU time is accumulated in signal_struct
1111                  * from sister threads now dead.  But in this non-leader
1112                  * exec, nothing survives from the original leader thread,
1113                  * whose birth marks the true age of this process now.
1114                  * When we take on its identity by switching to its PID, we
1115                  * also take its birthdate (always earlier than our own).
1116                  */
1117                 tsk->start_time = leader->start_time;
1118                 tsk->start_boottime = leader->start_boottime;
1119
1120                 BUG_ON(!same_thread_group(leader, tsk));
1121                 /*
1122                  * An exec() starts a new thread group with the
1123                  * TGID of the previous thread group. Rehash the
1124                  * two threads with a switched PID, and release
1125                  * the former thread group leader:
1126                  */
1127
1128                 /* Become a process group leader with the old leader's pid.
1129                  * The old leader becomes a thread of the this thread group.
1130                  */
1131                 exchange_tids(tsk, leader);
1132                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1133                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1134                 transfer_pid(leader, tsk, PIDTYPE_SID);
1135
1136                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1137                 list_replace_init(&leader->sibling, &tsk->sibling);
1138
1139                 tsk->group_leader = tsk;
1140                 leader->group_leader = tsk;
1141
1142                 tsk->exit_signal = SIGCHLD;
1143                 leader->exit_signal = -1;
1144
1145                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1146                 leader->exit_state = EXIT_DEAD;
1147
1148                 /*
1149                  * We are going to release_task()->ptrace_unlink() silently,
1150                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1151                  * the tracer wont't block again waiting for this thread.
1152                  */
1153                 if (unlikely(leader->ptrace))
1154                         __wake_up_parent(leader, leader->parent);
1155                 write_unlock_irq(&tasklist_lock);
1156                 cgroup_threadgroup_change_end(tsk);
1157
1158                 release_task(leader);
1159         }
1160
1161         sig->group_exit_task = NULL;
1162         sig->notify_count = 0;
1163
1164 no_thread_group:
1165         /* we have changed execution domain */
1166         tsk->exit_signal = SIGCHLD;
1167
1168         BUG_ON(!thread_group_leader(tsk));
1169         return 0;
1170
1171 killed:
1172         /* protects against exit_notify() and __exit_signal() */
1173         read_lock(&tasklist_lock);
1174         sig->group_exit_task = NULL;
1175         sig->notify_count = 0;
1176         read_unlock(&tasklist_lock);
1177         return -EAGAIN;
1178 }
1179
1180
1181 /*
1182  * This function makes sure the current process has its own signal table,
1183  * so that flush_signal_handlers can later reset the handlers without
1184  * disturbing other processes.  (Other processes might share the signal
1185  * table via the CLONE_SIGHAND option to clone().)
1186  */
1187 static int unshare_sighand(struct task_struct *me)
1188 {
1189         struct sighand_struct *oldsighand = me->sighand;
1190
1191         if (refcount_read(&oldsighand->count) != 1) {
1192                 struct sighand_struct *newsighand;
1193                 /*
1194                  * This ->sighand is shared with the CLONE_SIGHAND
1195                  * but not CLONE_THREAD task, switch to the new one.
1196                  */
1197                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1198                 if (!newsighand)
1199                         return -ENOMEM;
1200
1201                 refcount_set(&newsighand->count, 1);
1202
1203                 write_lock_irq(&tasklist_lock);
1204                 spin_lock(&oldsighand->siglock);
1205                 memcpy(newsighand->action, oldsighand->action,
1206                        sizeof(newsighand->action));
1207                 rcu_assign_pointer(me->sighand, newsighand);
1208                 spin_unlock(&oldsighand->siglock);
1209                 write_unlock_irq(&tasklist_lock);
1210
1211                 __cleanup_sighand(oldsighand);
1212         }
1213         return 0;
1214 }
1215
1216 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1217 {
1218         task_lock(tsk);
1219         strncpy(buf, tsk->comm, buf_size);
1220         task_unlock(tsk);
1221         return buf;
1222 }
1223 EXPORT_SYMBOL_GPL(__get_task_comm);
1224
1225 /*
1226  * These functions flushes out all traces of the currently running executable
1227  * so that a new one can be started
1228  */
1229
1230 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1231 {
1232         task_lock(tsk);
1233         trace_task_rename(tsk, buf);
1234         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1235         task_unlock(tsk);
1236         perf_event_comm(tsk, exec);
1237 }
1238
1239 /*
1240  * Calling this is the point of no return. None of the failures will be
1241  * seen by userspace since either the process is already taking a fatal
1242  * signal (via de_thread() or coredump), or will have SEGV raised
1243  * (after exec_mmap()) by search_binary_handler (see below).
1244  */
1245 int begin_new_exec(struct linux_binprm * bprm)
1246 {
1247         struct task_struct *me = current;
1248         int retval;
1249
1250         /* Once we are committed compute the creds */
1251         retval = bprm_creds_from_file(bprm);
1252         if (retval)
1253                 return retval;
1254
1255         /*
1256          * Ensure all future errors are fatal.
1257          */
1258         bprm->point_of_no_return = true;
1259
1260         /*
1261          * Make this the only thread in the thread group.
1262          */
1263         retval = de_thread(me);
1264         if (retval)
1265                 goto out;
1266
1267         /*
1268          * Must be called _before_ exec_mmap() as bprm->mm is
1269          * not visibile until then. This also enables the update
1270          * to be lockless.
1271          */
1272         set_mm_exe_file(bprm->mm, bprm->file);
1273
1274         /* If the binary is not readable then enforce mm->dumpable=0 */
1275         would_dump(bprm, bprm->file);
1276         if (bprm->have_execfd)
1277                 would_dump(bprm, bprm->executable);
1278
1279         /*
1280          * Release all of the old mmap stuff
1281          */
1282         acct_arg_size(bprm, 0);
1283         retval = exec_mmap(bprm->mm);
1284         if (retval)
1285                 goto out;
1286
1287         bprm->mm = NULL;
1288
1289 #ifdef CONFIG_POSIX_TIMERS
1290         spin_lock_irq(&me->sighand->siglock);
1291         posix_cpu_timers_exit(me);
1292         spin_unlock_irq(&me->sighand->siglock);
1293         exit_itimers(me);
1294         flush_itimer_signals();
1295 #endif
1296
1297         /*
1298          * Make the signal table private.
1299          */
1300         retval = unshare_sighand(me);
1301         if (retval)
1302                 goto out_unlock;
1303
1304         /*
1305          * Ensure that the uaccess routines can actually operate on userspace
1306          * pointers:
1307          */
1308         force_uaccess_begin();
1309
1310         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1311                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1312         flush_thread();
1313         me->personality &= ~bprm->per_clear;
1314
1315         /*
1316          * We have to apply CLOEXEC before we change whether the process is
1317          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1318          * trying to access the should-be-closed file descriptors of a process
1319          * undergoing exec(2).
1320          */
1321         do_close_on_exec(me->files);
1322
1323         if (bprm->secureexec) {
1324                 /* Make sure parent cannot signal privileged process. */
1325                 me->pdeath_signal = 0;
1326
1327                 /*
1328                  * For secureexec, reset the stack limit to sane default to
1329                  * avoid bad behavior from the prior rlimits. This has to
1330                  * happen before arch_pick_mmap_layout(), which examines
1331                  * RLIMIT_STACK, but after the point of no return to avoid
1332                  * needing to clean up the change on failure.
1333                  */
1334                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1335                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1336         }
1337
1338         me->sas_ss_sp = me->sas_ss_size = 0;
1339
1340         /*
1341          * Figure out dumpability. Note that this checking only of current
1342          * is wrong, but userspace depends on it. This should be testing
1343          * bprm->secureexec instead.
1344          */
1345         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1346             !(uid_eq(current_euid(), current_uid()) &&
1347               gid_eq(current_egid(), current_gid())))
1348                 set_dumpable(current->mm, suid_dumpable);
1349         else
1350                 set_dumpable(current->mm, SUID_DUMP_USER);
1351
1352         perf_event_exec();
1353         __set_task_comm(me, kbasename(bprm->filename), true);
1354
1355         /* An exec changes our domain. We are no longer part of the thread
1356            group */
1357         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1358         flush_signal_handlers(me, 0);
1359
1360         /*
1361          * install the new credentials for this executable
1362          */
1363         security_bprm_committing_creds(bprm);
1364
1365         commit_creds(bprm->cred);
1366         bprm->cred = NULL;
1367
1368         /*
1369          * Disable monitoring for regular users
1370          * when executing setuid binaries. Must
1371          * wait until new credentials are committed
1372          * by commit_creds() above
1373          */
1374         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1375                 perf_event_exit_task(me);
1376         /*
1377          * cred_guard_mutex must be held at least to this point to prevent
1378          * ptrace_attach() from altering our determination of the task's
1379          * credentials; any time after this it may be unlocked.
1380          */
1381         security_bprm_committed_creds(bprm);
1382
1383         /* Pass the opened binary to the interpreter. */
1384         if (bprm->have_execfd) {
1385                 retval = get_unused_fd_flags(0);
1386                 if (retval < 0)
1387                         goto out_unlock;
1388                 fd_install(retval, bprm->executable);
1389                 bprm->executable = NULL;
1390                 bprm->execfd = retval;
1391         }
1392         return 0;
1393
1394 out_unlock:
1395         up_write(&me->signal->exec_update_lock);
1396         if (!bprm->cred)
1397                 mutex_unlock(&me->signal->cred_guard_mutex);
1398
1399 out:
1400         return retval;
1401 }
1402 EXPORT_SYMBOL(begin_new_exec);
1403
1404 void would_dump(struct linux_binprm *bprm, struct file *file)
1405 {
1406         struct inode *inode = file_inode(file);
1407         if (inode_permission(inode, MAY_READ) < 0) {
1408                 struct user_namespace *old, *user_ns;
1409                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1410
1411                 /* Ensure mm->user_ns contains the executable */
1412                 user_ns = old = bprm->mm->user_ns;
1413                 while ((user_ns != &init_user_ns) &&
1414                        !privileged_wrt_inode_uidgid(user_ns, inode))
1415                         user_ns = user_ns->parent;
1416
1417                 if (old != user_ns) {
1418                         bprm->mm->user_ns = get_user_ns(user_ns);
1419                         put_user_ns(old);
1420                 }
1421         }
1422 }
1423 EXPORT_SYMBOL(would_dump);
1424
1425 void setup_new_exec(struct linux_binprm * bprm)
1426 {
1427         /* Setup things that can depend upon the personality */
1428         struct task_struct *me = current;
1429
1430         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1431
1432         arch_setup_new_exec();
1433
1434         /* Set the new mm task size. We have to do that late because it may
1435          * depend on TIF_32BIT which is only updated in flush_thread() on
1436          * some architectures like powerpc
1437          */
1438         me->mm->task_size = TASK_SIZE;
1439         up_write(&me->signal->exec_update_lock);
1440         mutex_unlock(&me->signal->cred_guard_mutex);
1441 }
1442 EXPORT_SYMBOL(setup_new_exec);
1443
1444 /* Runs immediately before start_thread() takes over. */
1445 void finalize_exec(struct linux_binprm *bprm)
1446 {
1447         /* Store any stack rlimit changes before starting thread. */
1448         task_lock(current->group_leader);
1449         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1450         task_unlock(current->group_leader);
1451 }
1452 EXPORT_SYMBOL(finalize_exec);
1453
1454 /*
1455  * Prepare credentials and lock ->cred_guard_mutex.
1456  * setup_new_exec() commits the new creds and drops the lock.
1457  * Or, if exec fails before, free_bprm() should release ->cred and
1458  * and unlock.
1459  */
1460 static int prepare_bprm_creds(struct linux_binprm *bprm)
1461 {
1462         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1463                 return -ERESTARTNOINTR;
1464
1465         bprm->cred = prepare_exec_creds();
1466         if (likely(bprm->cred))
1467                 return 0;
1468
1469         mutex_unlock(&current->signal->cred_guard_mutex);
1470         return -ENOMEM;
1471 }
1472
1473 static void free_bprm(struct linux_binprm *bprm)
1474 {
1475         if (bprm->mm) {
1476                 acct_arg_size(bprm, 0);
1477                 mmput(bprm->mm);
1478         }
1479         free_arg_pages(bprm);
1480         if (bprm->cred) {
1481                 mutex_unlock(&current->signal->cred_guard_mutex);
1482                 abort_creds(bprm->cred);
1483         }
1484         if (bprm->file) {
1485                 allow_write_access(bprm->file);
1486                 fput(bprm->file);
1487         }
1488         if (bprm->executable)
1489                 fput(bprm->executable);
1490         /* If a binfmt changed the interp, free it. */
1491         if (bprm->interp != bprm->filename)
1492                 kfree(bprm->interp);
1493         kfree(bprm->fdpath);
1494         kfree(bprm);
1495 }
1496
1497 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1498 {
1499         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1500         int retval = -ENOMEM;
1501         if (!bprm)
1502                 goto out;
1503
1504         if (fd == AT_FDCWD || filename->name[0] == '/') {
1505                 bprm->filename = filename->name;
1506         } else {
1507                 if (filename->name[0] == '\0')
1508                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1509                 else
1510                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1511                                                   fd, filename->name);
1512                 if (!bprm->fdpath)
1513                         goto out_free;
1514
1515                 bprm->filename = bprm->fdpath;
1516         }
1517         bprm->interp = bprm->filename;
1518
1519         retval = bprm_mm_init(bprm);
1520         if (retval)
1521                 goto out_free;
1522         return bprm;
1523
1524 out_free:
1525         free_bprm(bprm);
1526 out:
1527         return ERR_PTR(retval);
1528 }
1529
1530 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1531 {
1532         /* If a binfmt changed the interp, free it first. */
1533         if (bprm->interp != bprm->filename)
1534                 kfree(bprm->interp);
1535         bprm->interp = kstrdup(interp, GFP_KERNEL);
1536         if (!bprm->interp)
1537                 return -ENOMEM;
1538         return 0;
1539 }
1540 EXPORT_SYMBOL(bprm_change_interp);
1541
1542 /*
1543  * determine how safe it is to execute the proposed program
1544  * - the caller must hold ->cred_guard_mutex to protect against
1545  *   PTRACE_ATTACH or seccomp thread-sync
1546  */
1547 static void check_unsafe_exec(struct linux_binprm *bprm)
1548 {
1549         struct task_struct *p = current, *t;
1550         unsigned n_fs;
1551
1552         if (p->ptrace)
1553                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1554
1555         /*
1556          * This isn't strictly necessary, but it makes it harder for LSMs to
1557          * mess up.
1558          */
1559         if (task_no_new_privs(current))
1560                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1561
1562         t = p;
1563         n_fs = 1;
1564         spin_lock(&p->fs->lock);
1565         rcu_read_lock();
1566         while_each_thread(p, t) {
1567                 if (t->fs == p->fs)
1568                         n_fs++;
1569         }
1570         rcu_read_unlock();
1571
1572         if (p->fs->users > n_fs)
1573                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1574         else
1575                 p->fs->in_exec = 1;
1576         spin_unlock(&p->fs->lock);
1577 }
1578
1579 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1580 {
1581         /* Handle suid and sgid on files */
1582         struct inode *inode;
1583         unsigned int mode;
1584         kuid_t uid;
1585         kgid_t gid;
1586
1587         if (!mnt_may_suid(file->f_path.mnt))
1588                 return;
1589
1590         if (task_no_new_privs(current))
1591                 return;
1592
1593         inode = file->f_path.dentry->d_inode;
1594         mode = READ_ONCE(inode->i_mode);
1595         if (!(mode & (S_ISUID|S_ISGID)))
1596                 return;
1597
1598         /* Be careful if suid/sgid is set */
1599         inode_lock(inode);
1600
1601         /* reload atomically mode/uid/gid now that lock held */
1602         mode = inode->i_mode;
1603         uid = inode->i_uid;
1604         gid = inode->i_gid;
1605         inode_unlock(inode);
1606
1607         /* We ignore suid/sgid if there are no mappings for them in the ns */
1608         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1609                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1610                 return;
1611
1612         if (mode & S_ISUID) {
1613                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1614                 bprm->cred->euid = uid;
1615         }
1616
1617         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1618                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1619                 bprm->cred->egid = gid;
1620         }
1621 }
1622
1623 /*
1624  * Compute brpm->cred based upon the final binary.
1625  */
1626 static int bprm_creds_from_file(struct linux_binprm *bprm)
1627 {
1628         /* Compute creds based on which file? */
1629         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1630
1631         bprm_fill_uid(bprm, file);
1632         return security_bprm_creds_from_file(bprm, file);
1633 }
1634
1635 /*
1636  * Fill the binprm structure from the inode.
1637  * Read the first BINPRM_BUF_SIZE bytes
1638  *
1639  * This may be called multiple times for binary chains (scripts for example).
1640  */
1641 static int prepare_binprm(struct linux_binprm *bprm)
1642 {
1643         loff_t pos = 0;
1644
1645         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1646         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1647 }
1648
1649 /*
1650  * Arguments are '\0' separated strings found at the location bprm->p
1651  * points to; chop off the first by relocating brpm->p to right after
1652  * the first '\0' encountered.
1653  */
1654 int remove_arg_zero(struct linux_binprm *bprm)
1655 {
1656         int ret = 0;
1657         unsigned long offset;
1658         char *kaddr;
1659         struct page *page;
1660
1661         if (!bprm->argc)
1662                 return 0;
1663
1664         do {
1665                 offset = bprm->p & ~PAGE_MASK;
1666                 page = get_arg_page(bprm, bprm->p, 0);
1667                 if (!page) {
1668                         ret = -EFAULT;
1669                         goto out;
1670                 }
1671                 kaddr = kmap_atomic(page);
1672
1673                 for (; offset < PAGE_SIZE && kaddr[offset];
1674                                 offset++, bprm->p++)
1675                         ;
1676
1677                 kunmap_atomic(kaddr);
1678                 put_arg_page(page);
1679         } while (offset == PAGE_SIZE);
1680
1681         bprm->p++;
1682         bprm->argc--;
1683         ret = 0;
1684
1685 out:
1686         return ret;
1687 }
1688 EXPORT_SYMBOL(remove_arg_zero);
1689
1690 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1691 /*
1692  * cycle the list of binary formats handler, until one recognizes the image
1693  */
1694 static int search_binary_handler(struct linux_binprm *bprm)
1695 {
1696         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1697         struct linux_binfmt *fmt;
1698         int retval;
1699
1700         retval = prepare_binprm(bprm);
1701         if (retval < 0)
1702                 return retval;
1703
1704         retval = security_bprm_check(bprm);
1705         if (retval)
1706                 return retval;
1707
1708         retval = -ENOENT;
1709  retry:
1710         read_lock(&binfmt_lock);
1711         list_for_each_entry(fmt, &formats, lh) {
1712                 if (!try_module_get(fmt->module))
1713                         continue;
1714                 read_unlock(&binfmt_lock);
1715
1716                 retval = fmt->load_binary(bprm);
1717
1718                 read_lock(&binfmt_lock);
1719                 put_binfmt(fmt);
1720                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1721                         read_unlock(&binfmt_lock);
1722                         return retval;
1723                 }
1724         }
1725         read_unlock(&binfmt_lock);
1726
1727         if (need_retry) {
1728                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1729                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1730                         return retval;
1731                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1732                         return retval;
1733                 need_retry = false;
1734                 goto retry;
1735         }
1736
1737         return retval;
1738 }
1739
1740 static int exec_binprm(struct linux_binprm *bprm)
1741 {
1742         pid_t old_pid, old_vpid;
1743         int ret, depth;
1744
1745         /* Need to fetch pid before load_binary changes it */
1746         old_pid = current->pid;
1747         rcu_read_lock();
1748         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1749         rcu_read_unlock();
1750
1751         /* This allows 4 levels of binfmt rewrites before failing hard. */
1752         for (depth = 0;; depth++) {
1753                 struct file *exec;
1754                 if (depth > 5)
1755                         return -ELOOP;
1756
1757                 ret = search_binary_handler(bprm);
1758                 if (ret < 0)
1759                         return ret;
1760                 if (!bprm->interpreter)
1761                         break;
1762
1763                 exec = bprm->file;
1764                 bprm->file = bprm->interpreter;
1765                 bprm->interpreter = NULL;
1766
1767                 allow_write_access(exec);
1768                 if (unlikely(bprm->have_execfd)) {
1769                         if (bprm->executable) {
1770                                 fput(exec);
1771                                 return -ENOEXEC;
1772                         }
1773                         bprm->executable = exec;
1774                 } else
1775                         fput(exec);
1776         }
1777
1778         audit_bprm(bprm);
1779         trace_sched_process_exec(current, old_pid, bprm);
1780         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1781         proc_exec_connector(current);
1782         return 0;
1783 }
1784
1785 /*
1786  * sys_execve() executes a new program.
1787  */
1788 static int bprm_execve(struct linux_binprm *bprm,
1789                        int fd, struct filename *filename, int flags)
1790 {
1791         struct file *file;
1792         struct files_struct *displaced;
1793         int retval;
1794
1795         /*
1796          * Cancel any io_uring activity across execve
1797          */
1798         io_uring_task_cancel();
1799
1800         retval = unshare_files(&displaced);
1801         if (retval)
1802                 return retval;
1803
1804         retval = prepare_bprm_creds(bprm);
1805         if (retval)
1806                 goto out_files;
1807
1808         check_unsafe_exec(bprm);
1809         current->in_execve = 1;
1810
1811         file = do_open_execat(fd, filename, flags);
1812         retval = PTR_ERR(file);
1813         if (IS_ERR(file))
1814                 goto out_unmark;
1815
1816         sched_exec();
1817
1818         bprm->file = file;
1819         /*
1820          * Record that a name derived from an O_CLOEXEC fd will be
1821          * inaccessible after exec. Relies on having exclusive access to
1822          * current->files (due to unshare_files above).
1823          */
1824         if (bprm->fdpath &&
1825             close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1826                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1827
1828         /* Set the unchanging part of bprm->cred */
1829         retval = security_bprm_creds_for_exec(bprm);
1830         if (retval)
1831                 goto out;
1832
1833         retval = exec_binprm(bprm);
1834         if (retval < 0)
1835                 goto out;
1836
1837         /* execve succeeded */
1838         current->fs->in_exec = 0;
1839         current->in_execve = 0;
1840         rseq_execve(current);
1841         acct_update_integrals(current);
1842         task_numa_free(current, false);
1843         if (displaced)
1844                 put_files_struct(displaced);
1845         return retval;
1846
1847 out:
1848         /*
1849          * If past the point of no return ensure the the code never
1850          * returns to the userspace process.  Use an existing fatal
1851          * signal if present otherwise terminate the process with
1852          * SIGSEGV.
1853          */
1854         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1855                 force_sigsegv(SIGSEGV);
1856
1857 out_unmark:
1858         current->fs->in_exec = 0;
1859         current->in_execve = 0;
1860
1861 out_files:
1862         if (displaced)
1863                 reset_files_struct(displaced);
1864
1865         return retval;
1866 }
1867
1868 static int do_execveat_common(int fd, struct filename *filename,
1869                               struct user_arg_ptr argv,
1870                               struct user_arg_ptr envp,
1871                               int flags)
1872 {
1873         struct linux_binprm *bprm;
1874         int retval;
1875
1876         if (IS_ERR(filename))
1877                 return PTR_ERR(filename);
1878
1879         /*
1880          * We move the actual failure in case of RLIMIT_NPROC excess from
1881          * set*uid() to execve() because too many poorly written programs
1882          * don't check setuid() return code.  Here we additionally recheck
1883          * whether NPROC limit is still exceeded.
1884          */
1885         if ((current->flags & PF_NPROC_EXCEEDED) &&
1886             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1887                 retval = -EAGAIN;
1888                 goto out_ret;
1889         }
1890
1891         /* We're below the limit (still or again), so we don't want to make
1892          * further execve() calls fail. */
1893         current->flags &= ~PF_NPROC_EXCEEDED;
1894
1895         bprm = alloc_bprm(fd, filename);
1896         if (IS_ERR(bprm)) {
1897                 retval = PTR_ERR(bprm);
1898                 goto out_ret;
1899         }
1900
1901         retval = count(argv, MAX_ARG_STRINGS);
1902         if (retval == 0)
1903                 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1904                              current->comm, bprm->filename);
1905         if (retval < 0)
1906                 goto out_free;
1907         bprm->argc = retval;
1908
1909         retval = count(envp, MAX_ARG_STRINGS);
1910         if (retval < 0)
1911                 goto out_free;
1912         bprm->envc = retval;
1913
1914         retval = bprm_stack_limits(bprm);
1915         if (retval < 0)
1916                 goto out_free;
1917
1918         retval = copy_string_kernel(bprm->filename, bprm);
1919         if (retval < 0)
1920                 goto out_free;
1921         bprm->exec = bprm->p;
1922
1923         retval = copy_strings(bprm->envc, envp, bprm);
1924         if (retval < 0)
1925                 goto out_free;
1926
1927         retval = copy_strings(bprm->argc, argv, bprm);
1928         if (retval < 0)
1929                 goto out_free;
1930
1931         /*
1932          * When argv is empty, add an empty string ("") as argv[0] to
1933          * ensure confused userspace programs that start processing
1934          * from argv[1] won't end up walking envp. See also
1935          * bprm_stack_limits().
1936          */
1937         if (bprm->argc == 0) {
1938                 retval = copy_string_kernel("", bprm);
1939                 if (retval < 0)
1940                         goto out_free;
1941                 bprm->argc = 1;
1942         }
1943
1944         retval = bprm_execve(bprm, fd, filename, flags);
1945 out_free:
1946         free_bprm(bprm);
1947
1948 out_ret:
1949         putname(filename);
1950         return retval;
1951 }
1952
1953 int kernel_execve(const char *kernel_filename,
1954                   const char *const *argv, const char *const *envp)
1955 {
1956         struct filename *filename;
1957         struct linux_binprm *bprm;
1958         int fd = AT_FDCWD;
1959         int retval;
1960
1961         filename = getname_kernel(kernel_filename);
1962         if (IS_ERR(filename))
1963                 return PTR_ERR(filename);
1964
1965         bprm = alloc_bprm(fd, filename);
1966         if (IS_ERR(bprm)) {
1967                 retval = PTR_ERR(bprm);
1968                 goto out_ret;
1969         }
1970
1971         retval = count_strings_kernel(argv);
1972         if (WARN_ON_ONCE(retval == 0))
1973                 retval = -EINVAL;
1974         if (retval < 0)
1975                 goto out_free;
1976         bprm->argc = retval;
1977
1978         retval = count_strings_kernel(envp);
1979         if (retval < 0)
1980                 goto out_free;
1981         bprm->envc = retval;
1982
1983         retval = bprm_stack_limits(bprm);
1984         if (retval < 0)
1985                 goto out_free;
1986
1987         retval = copy_string_kernel(bprm->filename, bprm);
1988         if (retval < 0)
1989                 goto out_free;
1990         bprm->exec = bprm->p;
1991
1992         retval = copy_strings_kernel(bprm->envc, envp, bprm);
1993         if (retval < 0)
1994                 goto out_free;
1995
1996         retval = copy_strings_kernel(bprm->argc, argv, bprm);
1997         if (retval < 0)
1998                 goto out_free;
1999
2000         retval = bprm_execve(bprm, fd, filename, 0);
2001 out_free:
2002         free_bprm(bprm);
2003 out_ret:
2004         putname(filename);
2005         return retval;
2006 }
2007
2008 static int do_execve(struct filename *filename,
2009         const char __user *const __user *__argv,
2010         const char __user *const __user *__envp)
2011 {
2012         struct user_arg_ptr argv = { .ptr.native = __argv };
2013         struct user_arg_ptr envp = { .ptr.native = __envp };
2014         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2015 }
2016
2017 static int do_execveat(int fd, struct filename *filename,
2018                 const char __user *const __user *__argv,
2019                 const char __user *const __user *__envp,
2020                 int flags)
2021 {
2022         struct user_arg_ptr argv = { .ptr.native = __argv };
2023         struct user_arg_ptr envp = { .ptr.native = __envp };
2024
2025         return do_execveat_common(fd, filename, argv, envp, flags);
2026 }
2027
2028 #ifdef CONFIG_COMPAT
2029 static int compat_do_execve(struct filename *filename,
2030         const compat_uptr_t __user *__argv,
2031         const compat_uptr_t __user *__envp)
2032 {
2033         struct user_arg_ptr argv = {
2034                 .is_compat = true,
2035                 .ptr.compat = __argv,
2036         };
2037         struct user_arg_ptr envp = {
2038                 .is_compat = true,
2039                 .ptr.compat = __envp,
2040         };
2041         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2042 }
2043
2044 static int compat_do_execveat(int fd, struct filename *filename,
2045                               const compat_uptr_t __user *__argv,
2046                               const compat_uptr_t __user *__envp,
2047                               int flags)
2048 {
2049         struct user_arg_ptr argv = {
2050                 .is_compat = true,
2051                 .ptr.compat = __argv,
2052         };
2053         struct user_arg_ptr envp = {
2054                 .is_compat = true,
2055                 .ptr.compat = __envp,
2056         };
2057         return do_execveat_common(fd, filename, argv, envp, flags);
2058 }
2059 #endif
2060
2061 void set_binfmt(struct linux_binfmt *new)
2062 {
2063         struct mm_struct *mm = current->mm;
2064
2065         if (mm->binfmt)
2066                 module_put(mm->binfmt->module);
2067
2068         mm->binfmt = new;
2069         if (new)
2070                 __module_get(new->module);
2071 }
2072 EXPORT_SYMBOL(set_binfmt);
2073
2074 /*
2075  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2076  */
2077 void set_dumpable(struct mm_struct *mm, int value)
2078 {
2079         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2080                 return;
2081
2082         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2083 }
2084
2085 SYSCALL_DEFINE3(execve,
2086                 const char __user *, filename,
2087                 const char __user *const __user *, argv,
2088                 const char __user *const __user *, envp)
2089 {
2090         return do_execve(getname(filename), argv, envp);
2091 }
2092
2093 SYSCALL_DEFINE5(execveat,
2094                 int, fd, const char __user *, filename,
2095                 const char __user *const __user *, argv,
2096                 const char __user *const __user *, envp,
2097                 int, flags)
2098 {
2099         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2100
2101         return do_execveat(fd,
2102                            getname_flags(filename, lookup_flags, NULL),
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         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2121
2122         return compat_do_execveat(fd,
2123                                   getname_flags(filename, lookup_flags, NULL),
2124                                   argv, envp, flags);
2125 }
2126 #endif