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/kmod.h>
  60 #include <linux/fsnotify.h>
  61 #include <linux/fs_struct.h>
  62 #include <linux/oom.h>
  63 #include <linux/compat.h>
  64 #include <linux/vmalloc.h>
  65 #include <linux/io_uring.h>
  66 #include <linux/syscall_user_dispatch.h>
  67 #include <linux/coredump.h>
  68 #include <linux/time_namespace.h>
  69
  70 #include <linux/uaccess.h>
  71 #include <asm/mmu_context.h>
  72 #include <asm/tlb.h>
  73
  74 #include <trace/events/task.h>
  75 #include "internal.h"
  76
  77 #include <trace/events/sched.h>
  78
  79 static int bprm_creds_from_file(struct linux_binprm *bprm);
  80
  81 int suid_dumpable = 0;
  82
  83 static LIST_HEAD(formats);
  84 static DEFINE_RWLOCK(binfmt_lock);
  85
  86 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  87 {
  88         write_lock(&binfmt_lock);
  89         insert ? list_add(&fmt->lh, &formats) :
  90                  list_add_tail(&fmt->lh, &formats);
  91         write_unlock(&binfmt_lock);
  92 }
  93
  94 EXPORT_SYMBOL(__register_binfmt);
  95
  96 void unregister_binfmt(struct linux_binfmt * fmt)
  97 {
  98         write_lock(&binfmt_lock);
  99         list_del(&fmt->lh);
 100         write_unlock(&binfmt_lock);
 101 }
 102
 103 EXPORT_SYMBOL(unregister_binfmt);
 104
 105 static inline void put_binfmt(struct linux_binfmt * fmt)
 106 {
 107         module_put(fmt->module);
 108 }
 109
 110 bool path_noexec(const struct path *path)
 111 {
 112         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 113                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 114 }
 115
 116 #ifdef CONFIG_USELIB
 117 /*
 118  * Note that a shared library must be both readable and executable due to
 119  * security reasons.
 120  *
 121  * Also note that we take the address to load from the file itself.
 122  */
 123 SYSCALL_DEFINE1(uselib, const char __user *, library)
 124 {
 125         struct linux_binfmt *fmt;
 126         struct file *file;
 127         struct filename *tmp = getname(library);
 128         int error = PTR_ERR(tmp);
 129         static const struct open_flags uselib_flags = {
 130                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 131                 .acc_mode = MAY_READ | MAY_EXEC,
 132                 .intent = LOOKUP_OPEN,
 133                 .lookup_flags = LOOKUP_FOLLOW,
 134         };
 135
 136         if (IS_ERR(tmp))
 137                 goto out;
 138
 139         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 140         putname(tmp);
 141         error = PTR_ERR(file);
 142         if (IS_ERR(file))
 143                 goto out;
 144
 145         /*
 146          * may_open() has already checked for this, so it should be
 147          * impossible to trip now. But we need to be extra cautious
 148          * and check again at the very end too.
 149          */
 150         error = -EACCES;
 151         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 152                          path_noexec(&file->f_path)))
 153                 goto exit;
 154
 155         fsnotify_open(file);
 156
 157         error = -ENOEXEC;
 158
 159         read_lock(&binfmt_lock);
 160         list_for_each_entry(fmt, &formats, lh) {
 161                 if (!fmt->load_shlib)
 162                         continue;
 163                 if (!try_module_get(fmt->module))
 164                         continue;
 165                 read_unlock(&binfmt_lock);
 166                 error = fmt->load_shlib(file);
 167                 read_lock(&binfmt_lock);
 168                 put_binfmt(fmt);
 169                 if (error != -ENOEXEC)
 170                         break;
 171         }
 172         read_unlock(&binfmt_lock);
 173 exit:
 174         fput(file);
 175 out:
 176         return error;
 177 }
 178 #endif /* #ifdef CONFIG_USELIB */
 179
 180 #ifdef CONFIG_MMU
 181 /*
 182  * The nascent bprm->mm is not visible until exec_mmap() but it can
 183  * use a lot of memory, account these pages in current->mm temporary
 184  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 185  * change the counter back via acct_arg_size(0).
 186  */
 187 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 188 {
 189         struct mm_struct *mm = current->mm;
 190         long diff = (long)(pages - bprm->vma_pages);
 191
 192         if (!mm || !diff)
 193                 return;
 194
 195         bprm->vma_pages = pages;
 196         add_mm_counter(mm, MM_ANONPAGES, diff);
 197 }
 198
 199 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 200                 int write)
 201 {
 202         struct page *page;
 203         int ret;
 204         unsigned int gup_flags = FOLL_FORCE;
 205
 206 #ifdef CONFIG_STACK_GROWSUP
 207         if (write) {
 208                 ret = expand_downwards(bprm->vma, pos);
 209                 if (ret < 0)
 210                         return NULL;
 211         }
 212 #endif
 213
 214         if (write)
 215                 gup_flags |= FOLL_WRITE;
 216
 217         /*
 218          * We are doing an exec().  'current' is the process
 219          * doing the exec and bprm->mm is the new process's mm.
 220          */
 221         mmap_read_lock(bprm->mm);
 222         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 223                         &page, NULL, NULL);
 224         mmap_read_unlock(bprm->mm);
 225         if (ret <= 0)
 226                 return NULL;
 227
 228         if (write)
 229                 acct_arg_size(bprm, vma_pages(bprm->vma));
 230
 231         return page;
 232 }
 233
 234 static void put_arg_page(struct page *page)
 235 {
 236         put_page(page);
 237 }
 238
 239 static void free_arg_pages(struct linux_binprm *bprm)
 240 {
 241 }
 242
 243 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 244                 struct page *page)
 245 {
 246         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 247 }
 248
 249 static int __bprm_mm_init(struct linux_binprm *bprm)
 250 {
 251         int err;
 252         struct vm_area_struct *vma = NULL;
 253         struct mm_struct *mm = bprm->mm;
 254
 255         bprm->vma = vma = vm_area_alloc(mm);
 256         if (!vma)
 257                 return -ENOMEM;
 258         vma_set_anonymous(vma);
 259
 260         if (mmap_write_lock_killable(mm)) {
 261                 err = -EINTR;
 262                 goto err_free;
 263         }
 264
 265         /*
 266          * Place the stack at the largest stack address the architecture
 267          * supports. Later, we'll move this to an appropriate place. We don't
 268          * use STACK_TOP because that can depend on attributes which aren't
 269          * configured yet.
 270          */
 271         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 272         vma->vm_end = STACK_TOP_MAX;
 273         vma->vm_start = vma->vm_end - PAGE_SIZE;
 274         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 275         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 276
 277         err = insert_vm_struct(mm, vma);
 278         if (err)
 279                 goto err;
 280
 281         mm->stack_vm = mm->total_vm = 1;
 282         mmap_write_unlock(mm);
 283         bprm->p = vma->vm_end - sizeof(void *);
 284         return 0;
 285 err:
 286         mmap_write_unlock(mm);
 287 err_free:
 288         bprm->vma = NULL;
 289         vm_area_free(vma);
 290         return err;
 291 }
 292
 293 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 294 {
 295         return len <= MAX_ARG_STRLEN;
 296 }
 297
 298 #else
 299
 300 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 301 {
 302 }
 303
 304 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 305                 int write)
 306 {
 307         struct page *page;
 308
 309         page = bprm->page[pos / PAGE_SIZE];
 310         if (!page && write) {
 311                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 312                 if (!page)
 313                         return NULL;
 314                 bprm->page[pos / PAGE_SIZE] = page;
 315         }
 316
 317         return page;
 318 }
 319
 320 static void put_arg_page(struct page *page)
 321 {
 322 }
 323
 324 static void free_arg_page(struct linux_binprm *bprm, int i)
 325 {
 326         if (bprm->page[i]) {
 327                 __free_page(bprm->page[i]);
 328                 bprm->page[i] = NULL;
 329         }
 330 }
 331
 332 static void free_arg_pages(struct linux_binprm *bprm)
 333 {
 334         int i;
 335
 336         for (i = 0; i < MAX_ARG_PAGES; i++)
 337                 free_arg_page(bprm, i);
 338 }
 339
 340 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 341                 struct page *page)
 342 {
 343 }
 344
 345 static int __bprm_mm_init(struct linux_binprm *bprm)
 346 {
 347         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 348         return 0;
 349 }
 350
 351 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 352 {
 353         return len <= bprm->p;
 354 }
 355
 356 #endif /* CONFIG_MMU */
 357
 358 /*
 359  * Create a new mm_struct and populate it with a temporary stack
 360  * vm_area_struct.  We don't have enough context at this point to set the stack
 361  * flags, permissions, and offset, so we use temporary values.  We'll update
 362  * them later in setup_arg_pages().
 363  */
 364 static int bprm_mm_init(struct linux_binprm *bprm)
 365 {
 366         int err;
 367         struct mm_struct *mm = NULL;
 368
 369         bprm->mm = mm = mm_alloc();
 370         err = -ENOMEM;
 371         if (!mm)
 372                 goto err;
 373
 374         /* Save current stack limit for all calculations made during exec. */
 375         task_lock(current->group_leader);
 376         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 377         task_unlock(current->group_leader);
 378
 379         err = __bprm_mm_init(bprm);
 380         if (err)
 381                 goto err;
 382
 383         return 0;
 384
 385 err:
 386         if (mm) {
 387                 bprm->mm = NULL;
 388                 mmdrop(mm);
 389         }
 390
 391         return err;
 392 }
 393
 394 struct user_arg_ptr {
 395 #ifdef CONFIG_COMPAT
 396         bool is_compat;
 397 #endif
 398         union {
 399                 const char __user *const __user *native;
 400 #ifdef CONFIG_COMPAT
 401                 const compat_uptr_t __user *compat;
 402 #endif
 403         } ptr;
 404 };
 405
 406 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 407 {
 408         const char __user *native;
 409
 410 #ifdef CONFIG_COMPAT
 411         if (unlikely(argv.is_compat)) {
 412                 compat_uptr_t compat;
 413
 414                 if (get_user(compat, argv.ptr.compat + nr))
 415                         return ERR_PTR(-EFAULT);
 416
 417                 return compat_ptr(compat);
 418         }
 419 #endif
 420
 421         if (get_user(native, argv.ptr.native + nr))
 422                 return ERR_PTR(-EFAULT);
 423
 424         return native;
 425 }
 426
 427 /*
 428  * count() counts the number of strings in array ARGV.
 429  */
 430 static int count(struct user_arg_ptr argv, int max)
 431 {
 432         int i = 0;
 433
 434         if (argv.ptr.native != NULL) {
 435                 for (;;) {
 436                         const char __user *p = get_user_arg_ptr(argv, i);
 437
 438                         if (!p)
 439                                 break;
 440
 441                         if (IS_ERR(p))
 442                                 return -EFAULT;
 443
 444                         if (i >= max)
 445                                 return -E2BIG;
 446                         ++i;
 447
 448                         if (fatal_signal_pending(current))
 449                                 return -ERESTARTNOHAND;
 450                         cond_resched();
 451                 }
 452         }
 453         return i;
 454 }
 455
 456 static int count_strings_kernel(const char *const *argv)
 457 {
 458         int i;
 459
 460         if (!argv)
 461                 return 0;
 462
 463         for (i = 0; argv[i]; ++i) {
 464                 if (i >= MAX_ARG_STRINGS)
 465                         return -E2BIG;
 466                 if (fatal_signal_pending(current))
 467                         return -ERESTARTNOHAND;
 468                 cond_resched();
 469         }
 470         return i;
 471 }
 472
 473 static int bprm_stack_limits(struct linux_binprm *bprm)
 474 {
 475         unsigned long limit, ptr_size;
 476
 477         /*
 478          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 479          * (whichever is smaller) for the argv+env strings.
 480          * This ensures that:
 481          *  - the remaining binfmt code will not run out of stack space,
 482          *  - the program will have a reasonable amount of stack left
 483          *    to work from.
 484          */
 485         limit = _STK_LIM / 4 * 3;
 486         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 487         /*
 488          * We've historically supported up to 32 pages (ARG_MAX)
 489          * of argument strings even with small stacks
 490          */
 491         limit = max_t(unsigned long, limit, ARG_MAX);
 492         /*
 493          * We must account for the size of all the argv and envp pointers to
 494          * the argv and envp strings, since they will also take up space in
 495          * the stack. They aren't stored until much later when we can't
 496          * signal to the parent that the child has run out of stack space.
 497          * Instead, calculate it here so it's possible to fail gracefully.
 498          *
 499          * In the case of argc = 0, make sure there is space for adding a
 500          * empty string (which will bump argc to 1), to ensure confused
 501          * userspace programs don't start processing from argv[1], thinking
 502          * argc can never be 0, to keep them from walking envp by accident.
 503          * See do_execveat_common().
 504          */
 505         ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
 506         if (limit <= ptr_size)
 507                 return -E2BIG;
 508         limit -= ptr_size;
 509
 510         bprm->argmin = bprm->p - limit;
 511         return 0;
 512 }
 513
 514 /*
 515  * 'copy_strings()' copies argument/environment strings from the old
 516  * processes's memory to the new process's stack.  The call to get_user_pages()
 517  * ensures the destination page is created and not swapped out.
 518  */
 519 static int copy_strings(int argc, struct user_arg_ptr argv,
 520                         struct linux_binprm *bprm)
 521 {
 522         struct page *kmapped_page = NULL;
 523         char *kaddr = NULL;
 524         unsigned long kpos = 0;
 525         int ret;
 526
 527         while (argc-- > 0) {
 528                 const char __user *str;
 529                 int len;
 530                 unsigned long pos;
 531
 532                 ret = -EFAULT;
 533                 str = get_user_arg_ptr(argv, argc);
 534                 if (IS_ERR(str))
 535                         goto out;
 536
 537                 len = strnlen_user(str, MAX_ARG_STRLEN);
 538                 if (!len)
 539                         goto out;
 540
 541                 ret = -E2BIG;
 542                 if (!valid_arg_len(bprm, len))
 543                         goto out;
 544
 545                 /* We're going to work our way backwards. */
 546                 pos = bprm->p;
 547                 str += len;
 548                 bprm->p -= len;
 549 #ifdef CONFIG_MMU
 550                 if (bprm->p < bprm->argmin)
 551                         goto out;
 552 #endif
 553
 554                 while (len > 0) {
 555                         int offset, bytes_to_copy;
 556
 557                         if (fatal_signal_pending(current)) {
 558                                 ret = -ERESTARTNOHAND;
 559                                 goto out;
 560                         }
 561                         cond_resched();
 562
 563                         offset = pos % PAGE_SIZE;
 564                         if (offset == 0)
 565                                 offset = PAGE_SIZE;
 566
 567                         bytes_to_copy = offset;
 568                         if (bytes_to_copy > len)
 569                                 bytes_to_copy = len;
 570
 571                         offset -= bytes_to_copy;
 572                         pos -= bytes_to_copy;
 573                         str -= bytes_to_copy;
 574                         len -= bytes_to_copy;
 575
 576                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 577                                 struct page *page;
 578
 579                                 page = get_arg_page(bprm, pos, 1);
 580                                 if (!page) {
 581                                         ret = -E2BIG;
 582                                         goto out;
 583                                 }
 584
 585                                 if (kmapped_page) {
 586                                         flush_dcache_page(kmapped_page);
 587                                         kunmap(kmapped_page);
 588                                         put_arg_page(kmapped_page);
 589                                 }
 590                                 kmapped_page = page;
 591                                 kaddr = kmap(kmapped_page);
 592                                 kpos = pos & PAGE_MASK;
 593                                 flush_arg_page(bprm, kpos, kmapped_page);
 594                         }
 595                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 596                                 ret = -EFAULT;
 597                                 goto out;
 598                         }
 599                 }
 600         }
 601         ret = 0;
 602 out:
 603         if (kmapped_page) {
 604                 flush_dcache_page(kmapped_page);
 605                 kunmap(kmapped_page);
 606                 put_arg_page(kmapped_page);
 607         }
 608         return ret;
 609 }
 610
 611 /*
 612  * Copy and argument/environment string from the kernel to the processes stack.
 613  */
 614 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 615 {
 616         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 617         unsigned long pos = bprm->p;
 618
 619         if (len == 0)
 620                 return -EFAULT;
 621         if (!valid_arg_len(bprm, len))
 622                 return -E2BIG;
 623
 624         /* We're going to work our way backwards. */
 625         arg += len;
 626         bprm->p -= len;
 627         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 628                 return -E2BIG;
 629
 630         while (len > 0) {
 631                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 632                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 633                 struct page *page;
 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                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 643                 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
 644                 put_arg_page(page);
 645         }
 646
 647         return 0;
 648 }
 649 EXPORT_SYMBOL(copy_string_kernel);
 650
 651 static int copy_strings_kernel(int argc, const char *const *argv,
 652                                struct linux_binprm *bprm)
 653 {
 654         while (argc-- > 0) {
 655                 int ret = copy_string_kernel(argv[argc], bprm);
 656                 if (ret < 0)
 657                         return ret;
 658                 if (fatal_signal_pending(current))
 659                         return -ERESTARTNOHAND;
 660                 cond_resched();
 661         }
 662         return 0;
 663 }
 664
 665 #ifdef CONFIG_MMU
 666
 667 /*
 668  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 669  * the binfmt code determines where the new stack should reside, we shift it to
 670  * its final location.  The process proceeds as follows:
 671  *
 672  * 1) Use shift to calculate the new vma endpoints.
 673  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 674  *    arguments passed to subsequent functions are consistent.
 675  * 3) Move vma's page tables to the new range.
 676  * 4) Free up any cleared pgd range.
 677  * 5) Shrink the vma to cover only the new range.
 678  */
 679 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 680 {
 681         struct mm_struct *mm = vma->vm_mm;
 682         unsigned long old_start = vma->vm_start;
 683         unsigned long old_end = vma->vm_end;
 684         unsigned long length = old_end - old_start;
 685         unsigned long new_start = old_start - shift;
 686         unsigned long new_end = old_end - shift;
 687         struct mmu_gather tlb;
 688
 689         BUG_ON(new_start > new_end);
 690
 691         /*
 692          * ensure there are no vmas between where we want to go
 693          * and where we are
 694          */
 695         if (vma != find_vma(mm, new_start))
 696                 return -EFAULT;
 697
 698         /*
 699          * cover the whole range: [new_start, old_end)
 700          */
 701         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 702                 return -ENOMEM;
 703
 704         /*
 705          * move the page tables downwards, on failure we rely on
 706          * process cleanup to remove whatever mess we made.
 707          */
 708         if (length != move_page_tables(vma, old_start,
 709                                        vma, new_start, length, false))
 710                 return -ENOMEM;
 711
 712         lru_add_drain();
 713         tlb_gather_mmu(&tlb, mm);
 714         if (new_end > old_start) {
 715                 /*
 716                  * when the old and new regions overlap clear from new_end.
 717                  */
 718                 free_pgd_range(&tlb, new_end, old_end, new_end,
 719                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 720         } else {
 721                 /*
 722                  * otherwise, clean from old_start; this is done to not touch
 723                  * the address space in [new_end, old_start) some architectures
 724                  * have constraints on va-space that make this illegal (IA64) -
 725                  * for the others its just a little faster.
 726                  */
 727                 free_pgd_range(&tlb, old_start, old_end, new_end,
 728                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 729         }
 730         tlb_finish_mmu(&tlb);
 731
 732         /*
 733          * Shrink the vma to just the new range.  Always succeeds.
 734          */
 735         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 736
 737         return 0;
 738 }
 739
 740 /*
 741  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 742  * the stack is optionally relocated, and some extra space is added.
 743  */
 744 int setup_arg_pages(struct linux_binprm *bprm,
 745                     unsigned long stack_top,
 746                     int executable_stack)
 747 {
 748         unsigned long ret;
 749         unsigned long stack_shift;
 750         struct mm_struct *mm = current->mm;
 751         struct vm_area_struct *vma = bprm->vma;
 752         struct vm_area_struct *prev = NULL;
 753         unsigned long vm_flags;
 754         unsigned long stack_base;
 755         unsigned long stack_size;
 756         unsigned long stack_expand;
 757         unsigned long rlim_stack;
 758         struct mmu_gather tlb;
 759
 760 #ifdef CONFIG_STACK_GROWSUP
 761         /* Limit stack size */
 762         stack_base = bprm->rlim_stack.rlim_max;
 763
 764         stack_base = calc_max_stack_size(stack_base);
 765
 766         /* Add space for stack randomization. */
 767         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 768
 769         /* Make sure we didn't let the argument array grow too large. */
 770         if (vma->vm_end - vma->vm_start > stack_base)
 771                 return -ENOMEM;
 772
 773         stack_base = PAGE_ALIGN(stack_top - stack_base);
 774
 775         stack_shift = vma->vm_start - stack_base;
 776         mm->arg_start = bprm->p - stack_shift;
 777         bprm->p = vma->vm_end - stack_shift;
 778 #else
 779         stack_top = arch_align_stack(stack_top);
 780         stack_top = PAGE_ALIGN(stack_top);
 781
 782         if (unlikely(stack_top < mmap_min_addr) ||
 783             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 784                 return -ENOMEM;
 785
 786         stack_shift = vma->vm_end - stack_top;
 787
 788         bprm->p -= stack_shift;
 789         mm->arg_start = bprm->p;
 790 #endif
 791
 792         if (bprm->loader)
 793                 bprm->loader -= stack_shift;
 794         bprm->exec -= stack_shift;
 795
 796         if (mmap_write_lock_killable(mm))
 797                 return -EINTR;
 798
 799         vm_flags = VM_STACK_FLAGS;
 800
 801         /*
 802          * Adjust stack execute permissions; explicitly enable for
 803          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 804          * (arch default) otherwise.
 805          */
 806         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 807                 vm_flags |= VM_EXEC;
 808         else if (executable_stack == EXSTACK_DISABLE_X)
 809                 vm_flags &= ~VM_EXEC;
 810         vm_flags |= mm->def_flags;
 811         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 812
 813         tlb_gather_mmu(&tlb, mm);
 814         ret = mprotect_fixup(&tlb, vma, &prev, vma->vm_start, vma->vm_end,
 815                         vm_flags);
 816         tlb_finish_mmu(&tlb);
 817
 818         if (ret)
 819                 goto out_unlock;
 820         BUG_ON(prev != vma);
 821
 822         if (unlikely(vm_flags & VM_EXEC)) {
 823                 pr_warn_once("process '%pD4' started with executable stack\n",
 824                              bprm->file);
 825         }
 826
 827         /* Move stack pages down in memory. */
 828         if (stack_shift) {
 829                 ret = shift_arg_pages(vma, stack_shift);
 830                 if (ret)
 831                         goto out_unlock;
 832         }
 833
 834         /* mprotect_fixup is overkill to remove the temporary stack flags */
 835         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 836
 837         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 838         stack_size = vma->vm_end - vma->vm_start;
 839         /*
 840          * Align this down to a page boundary as expand_stack
 841          * will align it up.
 842          */
 843         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 844 #ifdef CONFIG_STACK_GROWSUP
 845         if (stack_size + stack_expand > rlim_stack)
 846                 stack_base = vma->vm_start + rlim_stack;
 847         else
 848                 stack_base = vma->vm_end + stack_expand;
 849 #else
 850         if (stack_size + stack_expand > rlim_stack)
 851                 stack_base = vma->vm_end - rlim_stack;
 852         else
 853                 stack_base = vma->vm_start - stack_expand;
 854 #endif
 855         current->mm->start_stack = bprm->p;
 856         ret = expand_stack(vma, stack_base);
 857         if (ret)
 858                 ret = -EFAULT;
 859
 860 out_unlock:
 861         mmap_write_unlock(mm);
 862         return ret;
 863 }
 864 EXPORT_SYMBOL(setup_arg_pages);
 865
 866 #else
 867
 868 /*
 869  * Transfer the program arguments and environment from the holding pages
 870  * onto the stack. The provided stack pointer is adjusted accordingly.
 871  */
 872 int transfer_args_to_stack(struct linux_binprm *bprm,
 873                            unsigned long *sp_location)
 874 {
 875         unsigned long index, stop, sp;
 876         int ret = 0;
 877
 878         stop = bprm->p >> PAGE_SHIFT;
 879         sp = *sp_location;
 880
 881         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 882                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 883                 char *src = kmap(bprm->page[index]) + offset;
 884                 sp -= PAGE_SIZE - offset;
 885                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 886                         ret = -EFAULT;
 887                 kunmap(bprm->page[index]);
 888                 if (ret)
 889                         goto out;
 890         }
 891
 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         bool vfork;
 983         int ret;
 984
 985         /* Notify parent that we're no longer interested in the old VM */
 986         tsk = current;
 987         vfork = !!tsk->vfork_done;
 988         old_mm = current->mm;
 989         exec_mm_release(tsk, old_mm);
 990         if (old_mm)
 991                 sync_mm_rss(old_mm);
 992
 993         ret = down_write_killable(&tsk->signal->exec_update_lock);
 994         if (ret)
 995                 return ret;
 996
 997         if (old_mm) {
 998                 /*
 999                  * If there is a pending fatal signal perhaps a signal
1000                  * whose default action is to create a coredump get
1001                  * out and die instead of going through with the exec.
1002                  */
1003                 ret = mmap_read_lock_killable(old_mm);
1004                 if (ret) {
1005                         up_write(&tsk->signal->exec_update_lock);
1006                         return ret;
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
1033         if (vfork)
1034                 timens_on_fork(tsk->nsproxy, tsk);
1035
1036         if (old_mm) {
1037                 mmap_read_unlock(old_mm);
1038                 BUG_ON(active_mm != old_mm);
1039                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1040                 mm_update_next_owner(old_mm);
1041                 mmput(old_mm);
1042                 return 0;
1043         }
1044         mmdrop(active_mm);
1045         return 0;
1046 }
1047
1048 static int de_thread(struct task_struct *tsk)
1049 {
1050         struct signal_struct *sig = tsk->signal;
1051         struct sighand_struct *oldsighand = tsk->sighand;
1052         spinlock_t *lock = &oldsighand->siglock;
1053
1054         if (thread_group_empty(tsk))
1055                 goto no_thread_group;
1056
1057         /*
1058          * Kill all other threads in the thread group.
1059          */
1060         spin_lock_irq(lock);
1061         if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1062                 /*
1063                  * Another group action in progress, just
1064                  * return so that the signal is processed.
1065                  */
1066                 spin_unlock_irq(lock);
1067                 return -EAGAIN;
1068         }
1069
1070         sig->group_exec_task = tsk;
1071         sig->notify_count = zap_other_threads(tsk);
1072         if (!thread_group_leader(tsk))
1073                 sig->notify_count--;
1074
1075         while (sig->notify_count) {
1076                 __set_current_state(TASK_KILLABLE);
1077                 spin_unlock_irq(lock);
1078                 schedule();
1079                 if (__fatal_signal_pending(tsk))
1080                         goto killed;
1081                 spin_lock_irq(lock);
1082         }
1083         spin_unlock_irq(lock);
1084
1085         /*
1086          * At this point all other threads have exited, all we have to
1087          * do is to wait for the thread group leader to become inactive,
1088          * and to assume its PID:
1089          */
1090         if (!thread_group_leader(tsk)) {
1091                 struct task_struct *leader = tsk->group_leader;
1092
1093                 for (;;) {
1094                         cgroup_threadgroup_change_begin(tsk);
1095                         write_lock_irq(&tasklist_lock);
1096                         /*
1097                          * Do this under tasklist_lock to ensure that
1098                          * exit_notify() can't miss ->group_exec_task
1099                          */
1100                         sig->notify_count = -1;
1101                         if (likely(leader->exit_state))
1102                                 break;
1103                         __set_current_state(TASK_KILLABLE);
1104                         write_unlock_irq(&tasklist_lock);
1105                         cgroup_threadgroup_change_end(tsk);
1106                         schedule();
1107                         if (__fatal_signal_pending(tsk))
1108                                 goto killed;
1109                 }
1110
1111                 /*
1112                  * The only record we have of the real-time age of a
1113                  * process, regardless of execs it's done, is start_time.
1114                  * All the past CPU time is accumulated in signal_struct
1115                  * from sister threads now dead.  But in this non-leader
1116                  * exec, nothing survives from the original leader thread,
1117                  * whose birth marks the true age of this process now.
1118                  * When we take on its identity by switching to its PID, we
1119                  * also take its birthdate (always earlier than our own).
1120                  */
1121                 tsk->start_time = leader->start_time;
1122                 tsk->start_boottime = leader->start_boottime;
1123
1124                 BUG_ON(!same_thread_group(leader, tsk));
1125                 /*
1126                  * An exec() starts a new thread group with the
1127                  * TGID of the previous thread group. Rehash the
1128                  * two threads with a switched PID, and release
1129                  * the former thread group leader:
1130                  */
1131
1132                 /* Become a process group leader with the old leader's pid.
1133                  * The old leader becomes a thread of the this thread group.
1134                  */
1135                 exchange_tids(tsk, leader);
1136                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1137                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1138                 transfer_pid(leader, tsk, PIDTYPE_SID);
1139
1140                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1141                 list_replace_init(&leader->sibling, &tsk->sibling);
1142
1143                 tsk->group_leader = tsk;
1144                 leader->group_leader = tsk;
1145
1146                 tsk->exit_signal = SIGCHLD;
1147                 leader->exit_signal = -1;
1148
1149                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1150                 leader->exit_state = EXIT_DEAD;
1151
1152                 /*
1153                  * We are going to release_task()->ptrace_unlink() silently,
1154                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1155                  * the tracer won't block again waiting for this thread.
1156                  */
1157                 if (unlikely(leader->ptrace))
1158                         __wake_up_parent(leader, leader->parent);
1159                 write_unlock_irq(&tasklist_lock);
1160                 cgroup_threadgroup_change_end(tsk);
1161
1162                 release_task(leader);
1163         }
1164
1165         sig->group_exec_task = NULL;
1166         sig->notify_count = 0;
1167
1168 no_thread_group:
1169         /* we have changed execution domain */
1170         tsk->exit_signal = SIGCHLD;
1171
1172         BUG_ON(!thread_group_leader(tsk));
1173         return 0;
1174
1175 killed:
1176         /* protects against exit_notify() and __exit_signal() */
1177         read_lock(&tasklist_lock);
1178         sig->group_exec_task = NULL;
1179         sig->notify_count = 0;
1180         read_unlock(&tasklist_lock);
1181         return -EAGAIN;
1182 }
1183
1184
1185 /*
1186  * This function makes sure the current process has its own signal table,
1187  * so that flush_signal_handlers can later reset the handlers without
1188  * disturbing other processes.  (Other processes might share the signal
1189  * table via the CLONE_SIGHAND option to clone().)
1190  */
1191 static int unshare_sighand(struct task_struct *me)
1192 {
1193         struct sighand_struct *oldsighand = me->sighand;
1194
1195         if (refcount_read(&oldsighand->count) != 1) {
1196                 struct sighand_struct *newsighand;
1197                 /*
1198                  * This ->sighand is shared with the CLONE_SIGHAND
1199                  * but not CLONE_THREAD task, switch to the new one.
1200                  */
1201                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1202                 if (!newsighand)
1203                         return -ENOMEM;
1204
1205                 refcount_set(&newsighand->count, 1);
1206                 memcpy(newsighand->action, oldsighand->action,
1207                        sizeof(newsighand->action));
1208
1209                 write_lock_irq(&tasklist_lock);
1210                 spin_lock(&oldsighand->siglock);
1211                 rcu_assign_pointer(me->sighand, newsighand);
1212                 spin_unlock(&oldsighand->siglock);
1213                 write_unlock_irq(&tasklist_lock);
1214
1215                 __cleanup_sighand(oldsighand);
1216         }
1217         return 0;
1218 }
1219
1220 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1221 {
1222         task_lock(tsk);
1223         /* Always NUL terminated and zero-padded */
1224         strscpy_pad(buf, tsk->comm, buf_size);
1225         task_unlock(tsk);
1226         return buf;
1227 }
1228 EXPORT_SYMBOL_GPL(__get_task_comm);
1229
1230 /*
1231  * These functions flushes out all traces of the currently running executable
1232  * so that a new one can be started
1233  */
1234
1235 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1236 {
1237         task_lock(tsk);
1238         trace_task_rename(tsk, buf);
1239         strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1240         task_unlock(tsk);
1241         perf_event_comm(tsk, exec);
1242 }
1243
1244 /*
1245  * Calling this is the point of no return. None of the failures will be
1246  * seen by userspace since either the process is already taking a fatal
1247  * signal (via de_thread() or coredump), or will have SEGV raised
1248  * (after exec_mmap()) by search_binary_handler (see below).
1249  */
1250 int begin_new_exec(struct linux_binprm * bprm)
1251 {
1252         struct task_struct *me = current;
1253         int retval;
1254
1255         /* Once we are committed compute the creds */
1256         retval = bprm_creds_from_file(bprm);
1257         if (retval)
1258                 return retval;
1259
1260         /*
1261          * Ensure all future errors are fatal.
1262          */
1263         bprm->point_of_no_return = true;
1264
1265         /*
1266          * Make this the only thread in the thread group.
1267          */
1268         retval = de_thread(me);
1269         if (retval)
1270                 goto out;
1271
1272         /*
1273          * Cancel any io_uring activity across execve
1274          */
1275         io_uring_task_cancel();
1276
1277         /* Ensure the files table is not shared. */
1278         retval = unshare_files();
1279         if (retval)
1280                 goto out;
1281
1282         /*
1283          * Must be called _before_ exec_mmap() as bprm->mm is
1284          * not visible until then. This also enables the update
1285          * to be lockless.
1286          */
1287         retval = set_mm_exe_file(bprm->mm, bprm->file);
1288         if (retval)
1289                 goto out;
1290
1291         /* If the binary is not readable then enforce mm->dumpable=0 */
1292         would_dump(bprm, bprm->file);
1293         if (bprm->have_execfd)
1294                 would_dump(bprm, bprm->executable);
1295
1296         /*
1297          * Release all of the old mmap stuff
1298          */
1299         acct_arg_size(bprm, 0);
1300         retval = exec_mmap(bprm->mm);
1301         if (retval)
1302                 goto out;
1303
1304         bprm->mm = NULL;
1305
1306 #ifdef CONFIG_POSIX_TIMERS
1307         exit_itimers(me);
1308         flush_itimer_signals();
1309 #endif
1310
1311         /*
1312          * Make the signal table private.
1313          */
1314         retval = unshare_sighand(me);
1315         if (retval)
1316                 goto out_unlock;
1317
1318         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
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         /* It is non-sense for kernel threads to call execve */
1964         if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1965                 return -EINVAL;
1966
1967         filename = getname_kernel(kernel_filename);
1968         if (IS_ERR(filename))
1969                 return PTR_ERR(filename);
1970
1971         bprm = alloc_bprm(fd, filename);
1972         if (IS_ERR(bprm)) {
1973                 retval = PTR_ERR(bprm);
1974                 goto out_ret;
1975         }
1976
1977         retval = count_strings_kernel(argv);
1978         if (WARN_ON_ONCE(retval == 0))
1979                 retval = -EINVAL;
1980         if (retval < 0)
1981                 goto out_free;
1982         bprm->argc = retval;
1983
1984         retval = count_strings_kernel(envp);
1985         if (retval < 0)
1986                 goto out_free;
1987         bprm->envc = retval;
1988
1989         retval = bprm_stack_limits(bprm);
1990         if (retval < 0)
1991                 goto out_free;
1992
1993         retval = copy_string_kernel(bprm->filename, bprm);
1994         if (retval < 0)
1995                 goto out_free;
1996         bprm->exec = bprm->p;
1997
1998         retval = copy_strings_kernel(bprm->envc, envp, bprm);
1999         if (retval < 0)
2000                 goto out_free;
2001
2002         retval = copy_strings_kernel(bprm->argc, argv, bprm);
2003         if (retval < 0)
2004                 goto out_free;
2005
2006         retval = bprm_execve(bprm, fd, filename, 0);
2007 out_free:
2008         free_bprm(bprm);
2009 out_ret:
2010         putname(filename);
2011         return retval;
2012 }
2013
2014 static int do_execve(struct filename *filename,
2015         const char __user *const __user *__argv,
2016         const char __user *const __user *__envp)
2017 {
2018         struct user_arg_ptr argv = { .ptr.native = __argv };
2019         struct user_arg_ptr envp = { .ptr.native = __envp };
2020         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2021 }
2022
2023 static int do_execveat(int fd, struct filename *filename,
2024                 const char __user *const __user *__argv,
2025                 const char __user *const __user *__envp,
2026                 int flags)
2027 {
2028         struct user_arg_ptr argv = { .ptr.native = __argv };
2029         struct user_arg_ptr envp = { .ptr.native = __envp };
2030
2031         return do_execveat_common(fd, filename, argv, envp, flags);
2032 }
2033
2034 #ifdef CONFIG_COMPAT
2035 static int compat_do_execve(struct filename *filename,
2036         const compat_uptr_t __user *__argv,
2037         const compat_uptr_t __user *__envp)
2038 {
2039         struct user_arg_ptr argv = {
2040                 .is_compat = true,
2041                 .ptr.compat = __argv,
2042         };
2043         struct user_arg_ptr envp = {
2044                 .is_compat = true,
2045                 .ptr.compat = __envp,
2046         };
2047         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2048 }
2049
2050 static int compat_do_execveat(int fd, struct filename *filename,
2051                               const compat_uptr_t __user *__argv,
2052                               const compat_uptr_t __user *__envp,
2053                               int flags)
2054 {
2055         struct user_arg_ptr argv = {
2056                 .is_compat = true,
2057                 .ptr.compat = __argv,
2058         };
2059         struct user_arg_ptr envp = {
2060                 .is_compat = true,
2061                 .ptr.compat = __envp,
2062         };
2063         return do_execveat_common(fd, filename, argv, envp, flags);
2064 }
2065 #endif
2066
2067 void set_binfmt(struct linux_binfmt *new)
2068 {
2069         struct mm_struct *mm = current->mm;
2070
2071         if (mm->binfmt)
2072                 module_put(mm->binfmt->module);
2073
2074         mm->binfmt = new;
2075         if (new)
2076                 __module_get(new->module);
2077 }
2078 EXPORT_SYMBOL(set_binfmt);
2079
2080 /*
2081  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2082  */
2083 void set_dumpable(struct mm_struct *mm, int value)
2084 {
2085         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2086                 return;
2087
2088         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2089 }
2090
2091 SYSCALL_DEFINE3(execve,
2092                 const char __user *, filename,
2093                 const char __user *const __user *, argv,
2094                 const char __user *const __user *, envp)
2095 {
2096         return do_execve(getname(filename), argv, envp);
2097 }
2098
2099 SYSCALL_DEFINE5(execveat,
2100                 int, fd, const char __user *, filename,
2101                 const char __user *const __user *, argv,
2102                 const char __user *const __user *, envp,
2103                 int, flags)
2104 {
2105         return do_execveat(fd,
2106                            getname_uflags(filename, flags),
2107                            argv, envp, flags);
2108 }
2109
2110 #ifdef CONFIG_COMPAT
2111 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2112         const compat_uptr_t __user *, argv,
2113         const compat_uptr_t __user *, envp)
2114 {
2115         return compat_do_execve(getname(filename), argv, envp);
2116 }
2117
2118 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2119                        const char __user *, filename,
2120                        const compat_uptr_t __user *, argv,
2121                        const compat_uptr_t __user *, envp,
2122                        int,  flags)
2123 {
2124         return compat_do_execveat(fd,
2125                                   getname_uflags(filename, flags),
2126                                   argv, envp, flags);
2127 }
2128 #endif
2129
2130 #ifdef CONFIG_SYSCTL
2131
2132 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2133                 void *buffer, size_t *lenp, loff_t *ppos)
2134 {
2135         int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2136
2137         if (!error)
2138                 validate_coredump_safety();
2139         return error;
2140 }
2141
2142 static struct ctl_table fs_exec_sysctls[] = {
2143         {
2144                 .procname       = "suid_dumpable",
2145                 .data           = &suid_dumpable,
2146                 .maxlen         = sizeof(int),
2147                 .mode           = 0644,
2148                 .proc_handler   = proc_dointvec_minmax_coredump,
2149                 .extra1         = SYSCTL_ZERO,
2150                 .extra2         = SYSCTL_TWO,
2151         },
2152         { }
2153 };
2154
2155 static int __init init_fs_exec_sysctls(void)
2156 {
2157         register_sysctl_init("fs", fs_exec_sysctls);
2158         return 0;
2159 }
2160
2161 fs_initcall(init_fs_exec_sysctls);
2162 #endif /* CONFIG_SYSCTL */