fs/exec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/exec.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  * #!-checking implemented by tytso.
  10  */
  11 /*
  12  * Demand-loading implemented 01.12.91 - no need to read anything but
  13  * the header into memory. The inode of the executable is put into
  14  * "current->executable", and page faults do the actual loading. Clean.
  15  *
  16  * Once more I can proudly say that linux stood up to being changed: it
  17  * was less than 2 hours work to get demand-loading completely implemented.
  18  *
  19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20  * current->executable is only used by the procfs.  This allows a dispatch
  21  * table to check for several different types  of binary formats.  We keep
  22  * trying until we recognize the file or we run out of supported binary
  23  * formats.
  24  */
  25
  26 #include <linux/kernel_read_file.h>
  27 #include <linux/slab.h>
  28 #include <linux/file.h>
  29 #include <linux/fdtable.h>
  30 #include <linux/mm.h>
  31 #include <linux/vmacache.h>
  32 #include <linux/stat.h>
  33 #include <linux/fcntl.h>
  34 #include <linux/swap.h>
  35 #include <linux/string.h>
  36 #include <linux/init.h>
  37 #include <linux/sched/mm.h>
  38 #include <linux/sched/coredump.h>
  39 #include <linux/sched/signal.h>
  40 #include <linux/sched/numa_balancing.h>
  41 #include <linux/sched/task.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/perf_event.h>
  44 #include <linux/highmem.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/key.h>
  47 #include <linux/personality.h>
  48 #include <linux/binfmts.h>
  49 #include <linux/utsname.h>
  50 #include <linux/pid_namespace.h>
  51 #include <linux/module.h>
  52 #include <linux/namei.h>
  53 #include <linux/mount.h>
  54 #include <linux/security.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/audit.h>
  59 #include <linux/tracehook.h>
  60 #include <linux/kmod.h>
  61 #include <linux/fsnotify.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/oom.h>
  64 #include <linux/compat.h>
  65 #include <linux/vmalloc.h>
  66 #include <linux/io_uring.h>
  67 #include <linux/syscall_user_dispatch.h>
  68 #include <linux/coredump.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                 char *kaddr;
 635
 636                 pos -= bytes_to_copy;
 637                 arg -= bytes_to_copy;
 638                 len -= bytes_to_copy;
 639
 640                 page = get_arg_page(bprm, pos, 1);
 641                 if (!page)
 642                         return -E2BIG;
 643                 kaddr = kmap_atomic(page);
 644                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 645                 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
 646                 flush_dcache_page(page);
 647                 kunmap_atomic(kaddr);
 648                 put_arg_page(page);
 649         }
 650
 651         return 0;
 652 }
 653 EXPORT_SYMBOL(copy_string_kernel);
 654
 655 static int copy_strings_kernel(int argc, const char *const *argv,
 656                                struct linux_binprm *bprm)
 657 {
 658         while (argc-- > 0) {
 659                 int ret = copy_string_kernel(argv[argc], bprm);
 660                 if (ret < 0)
 661                         return ret;
 662                 if (fatal_signal_pending(current))
 663                         return -ERESTARTNOHAND;
 664                 cond_resched();
 665         }
 666         return 0;
 667 }
 668
 669 #ifdef CONFIG_MMU
 670
 671 /*
 672  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 673  * the binfmt code determines where the new stack should reside, we shift it to
 674  * its final location.  The process proceeds as follows:
 675  *
 676  * 1) Use shift to calculate the new vma endpoints.
 677  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 678  *    arguments passed to subsequent functions are consistent.
 679  * 3) Move vma's page tables to the new range.
 680  * 4) Free up any cleared pgd range.
 681  * 5) Shrink the vma to cover only the new range.
 682  */
 683 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 684 {
 685         struct mm_struct *mm = vma->vm_mm;
 686         unsigned long old_start = vma->vm_start;
 687         unsigned long old_end = vma->vm_end;
 688         unsigned long length = old_end - old_start;
 689         unsigned long new_start = old_start - shift;
 690         unsigned long new_end = old_end - shift;
 691         struct mmu_gather tlb;
 692
 693         BUG_ON(new_start > new_end);
 694
 695         /*
 696          * ensure there are no vmas between where we want to go
 697          * and where we are
 698          */
 699         if (vma != find_vma(mm, new_start))
 700                 return -EFAULT;
 701
 702         /*
 703          * cover the whole range: [new_start, old_end)
 704          */
 705         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 706                 return -ENOMEM;
 707
 708         /*
 709          * move the page tables downwards, on failure we rely on
 710          * process cleanup to remove whatever mess we made.
 711          */
 712         if (length != move_page_tables(vma, old_start,
 713                                        vma, new_start, length, false))
 714                 return -ENOMEM;
 715
 716         lru_add_drain();
 717         tlb_gather_mmu(&tlb, mm);
 718         if (new_end > old_start) {
 719                 /*
 720                  * when the old and new regions overlap clear from new_end.
 721                  */
 722                 free_pgd_range(&tlb, new_end, old_end, new_end,
 723                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 724         } else {
 725                 /*
 726                  * otherwise, clean from old_start; this is done to not touch
 727                  * the address space in [new_end, old_start) some architectures
 728                  * have constraints on va-space that make this illegal (IA64) -
 729                  * for the others its just a little faster.
 730                  */
 731                 free_pgd_range(&tlb, old_start, old_end, new_end,
 732                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 733         }
 734         tlb_finish_mmu(&tlb);
 735
 736         /*
 737          * Shrink the vma to just the new range.  Always succeeds.
 738          */
 739         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 740
 741         return 0;
 742 }
 743
 744 /*
 745  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 746  * the stack is optionally relocated, and some extra space is added.
 747  */
 748 int setup_arg_pages(struct linux_binprm *bprm,
 749                     unsigned long stack_top,
 750                     int executable_stack)
 751 {
 752         unsigned long ret;
 753         unsigned long stack_shift;
 754         struct mm_struct *mm = current->mm;
 755         struct vm_area_struct *vma = bprm->vma;
 756         struct vm_area_struct *prev = NULL;
 757         unsigned long vm_flags;
 758         unsigned long stack_base;
 759         unsigned long stack_size;
 760         unsigned long stack_expand;
 761         unsigned long rlim_stack;
 762
 763 #ifdef CONFIG_STACK_GROWSUP
 764         /* Limit stack size */
 765         stack_base = bprm->rlim_stack.rlim_max;
 766
 767         stack_base = calc_max_stack_size(stack_base);
 768
 769         /* Add space for stack randomization. */
 770         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 771
 772         /* Make sure we didn't let the argument array grow too large. */
 773         if (vma->vm_end - vma->vm_start > stack_base)
 774                 return -ENOMEM;
 775
 776         stack_base = PAGE_ALIGN(stack_top - stack_base);
 777
 778         stack_shift = vma->vm_start - stack_base;
 779         mm->arg_start = bprm->p - stack_shift;
 780         bprm->p = vma->vm_end - stack_shift;
 781 #else
 782         stack_top = arch_align_stack(stack_top);
 783         stack_top = PAGE_ALIGN(stack_top);
 784
 785         if (unlikely(stack_top < mmap_min_addr) ||
 786             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 787                 return -ENOMEM;
 788
 789         stack_shift = vma->vm_end - stack_top;
 790
 791         bprm->p -= stack_shift;
 792         mm->arg_start = bprm->p;
 793 #endif
 794
 795         if (bprm->loader)
 796                 bprm->loader -= stack_shift;
 797         bprm->exec -= stack_shift;
 798
 799         if (mmap_write_lock_killable(mm))
 800                 return -EINTR;
 801
 802         vm_flags = VM_STACK_FLAGS;
 803
 804         /*
 805          * Adjust stack execute permissions; explicitly enable for
 806          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 807          * (arch default) otherwise.
 808          */
 809         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 810                 vm_flags |= VM_EXEC;
 811         else if (executable_stack == EXSTACK_DISABLE_X)
 812                 vm_flags &= ~VM_EXEC;
 813         vm_flags |= mm->def_flags;
 814         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 815
 816         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 817                         vm_flags);
 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         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                  * If there is a pending fatal signal perhaps a signal
 998                  * whose default action is to create a coredump get
 999                  * out and die instead of going through with the exec.
1000                  */
1001                 ret = mmap_read_lock_killable(old_mm);
1002                 if (ret) {
1003                         up_write(&tsk->signal->exec_update_lock);
1004                         return ret;
1005                 }
1006         }
1007
1008         task_lock(tsk);
1009         membarrier_exec_mmap(mm);
1010
1011         local_irq_disable();
1012         active_mm = tsk->active_mm;
1013         tsk->active_mm = mm;
1014         tsk->mm = mm;
1015         /*
1016          * This prevents preemption while active_mm is being loaded and
1017          * it and mm are being updated, which could cause problems for
1018          * lazy tlb mm refcounting when these are updated by context
1019          * switches. Not all architectures can handle irqs off over
1020          * activate_mm yet.
1021          */
1022         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1023                 local_irq_enable();
1024         activate_mm(active_mm, mm);
1025         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1026                 local_irq_enable();
1027         tsk->mm->vmacache_seqnum = 0;
1028         vmacache_flush(tsk);
1029         task_unlock(tsk);
1030         if (old_mm) {
1031                 mmap_read_unlock(old_mm);
1032                 BUG_ON(active_mm != old_mm);
1033                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1034                 mm_update_next_owner(old_mm);
1035                 mmput(old_mm);
1036                 return 0;
1037         }
1038         mmdrop(active_mm);
1039         return 0;
1040 }
1041
1042 static int de_thread(struct task_struct *tsk)
1043 {
1044         struct signal_struct *sig = tsk->signal;
1045         struct sighand_struct *oldsighand = tsk->sighand;
1046         spinlock_t *lock = &oldsighand->siglock;
1047
1048         if (thread_group_empty(tsk))
1049                 goto no_thread_group;
1050
1051         /*
1052          * Kill all other threads in the thread group.
1053          */
1054         spin_lock_irq(lock);
1055         if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1056                 /*
1057                  * Another group action in progress, just
1058                  * return so that the signal is processed.
1059                  */
1060                 spin_unlock_irq(lock);
1061                 return -EAGAIN;
1062         }
1063
1064         sig->group_exec_task = tsk;
1065         sig->notify_count = zap_other_threads(tsk);
1066         if (!thread_group_leader(tsk))
1067                 sig->notify_count--;
1068
1069         while (sig->notify_count) {
1070                 __set_current_state(TASK_KILLABLE);
1071                 spin_unlock_irq(lock);
1072                 schedule();
1073                 if (__fatal_signal_pending(tsk))
1074                         goto killed;
1075                 spin_lock_irq(lock);
1076         }
1077         spin_unlock_irq(lock);
1078
1079         /*
1080          * At this point all other threads have exited, all we have to
1081          * do is to wait for the thread group leader to become inactive,
1082          * and to assume its PID:
1083          */
1084         if (!thread_group_leader(tsk)) {
1085                 struct task_struct *leader = tsk->group_leader;
1086
1087                 for (;;) {
1088                         cgroup_threadgroup_change_begin(tsk);
1089                         write_lock_irq(&tasklist_lock);
1090                         /*
1091                          * Do this under tasklist_lock to ensure that
1092                          * exit_notify() can't miss ->group_exec_task
1093                          */
1094                         sig->notify_count = -1;
1095                         if (likely(leader->exit_state))
1096                                 break;
1097                         __set_current_state(TASK_KILLABLE);
1098                         write_unlock_irq(&tasklist_lock);
1099                         cgroup_threadgroup_change_end(tsk);
1100                         schedule();
1101                         if (__fatal_signal_pending(tsk))
1102                                 goto killed;
1103                 }
1104
1105                 /*
1106                  * The only record we have of the real-time age of a
1107                  * process, regardless of execs it's done, is start_time.
1108                  * All the past CPU time is accumulated in signal_struct
1109                  * from sister threads now dead.  But in this non-leader
1110                  * exec, nothing survives from the original leader thread,
1111                  * whose birth marks the true age of this process now.
1112                  * When we take on its identity by switching to its PID, we
1113                  * also take its birthdate (always earlier than our own).
1114                  */
1115                 tsk->start_time = leader->start_time;
1116                 tsk->start_boottime = leader->start_boottime;
1117
1118                 BUG_ON(!same_thread_group(leader, tsk));
1119                 /*
1120                  * An exec() starts a new thread group with the
1121                  * TGID of the previous thread group. Rehash the
1122                  * two threads with a switched PID, and release
1123                  * the former thread group leader:
1124                  */
1125
1126                 /* Become a process group leader with the old leader's pid.
1127                  * The old leader becomes a thread of the this thread group.
1128                  */
1129                 exchange_tids(tsk, leader);
1130                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1131                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1132                 transfer_pid(leader, tsk, PIDTYPE_SID);
1133
1134                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1135                 list_replace_init(&leader->sibling, &tsk->sibling);
1136
1137                 tsk->group_leader = tsk;
1138                 leader->group_leader = tsk;
1139
1140                 tsk->exit_signal = SIGCHLD;
1141                 leader->exit_signal = -1;
1142
1143                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1144                 leader->exit_state = EXIT_DEAD;
1145
1146                 /*
1147                  * We are going to release_task()->ptrace_unlink() silently,
1148                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1149                  * the tracer wont't block again waiting for this thread.
1150                  */
1151                 if (unlikely(leader->ptrace))
1152                         __wake_up_parent(leader, leader->parent);
1153                 write_unlock_irq(&tasklist_lock);
1154                 cgroup_threadgroup_change_end(tsk);
1155
1156                 release_task(leader);
1157         }
1158
1159         sig->group_exec_task = NULL;
1160         sig->notify_count = 0;
1161
1162 no_thread_group:
1163         /* we have changed execution domain */
1164         tsk->exit_signal = SIGCHLD;
1165
1166         BUG_ON(!thread_group_leader(tsk));
1167         return 0;
1168
1169 killed:
1170         /* protects against exit_notify() and __exit_signal() */
1171         read_lock(&tasklist_lock);
1172         sig->group_exec_task = NULL;
1173         sig->notify_count = 0;
1174         read_unlock(&tasklist_lock);
1175         return -EAGAIN;
1176 }
1177
1178
1179 /*
1180  * This function makes sure the current process has its own signal table,
1181  * so that flush_signal_handlers can later reset the handlers without
1182  * disturbing other processes.  (Other processes might share the signal
1183  * table via the CLONE_SIGHAND option to clone().)
1184  */
1185 static int unshare_sighand(struct task_struct *me)
1186 {
1187         struct sighand_struct *oldsighand = me->sighand;
1188
1189         if (refcount_read(&oldsighand->count) != 1) {
1190                 struct sighand_struct *newsighand;
1191                 /*
1192                  * This ->sighand is shared with the CLONE_SIGHAND
1193                  * but not CLONE_THREAD task, switch to the new one.
1194                  */
1195                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1196                 if (!newsighand)
1197                         return -ENOMEM;
1198
1199                 refcount_set(&newsighand->count, 1);
1200                 memcpy(newsighand->action, oldsighand->action,
1201                        sizeof(newsighand->action));
1202
1203                 write_lock_irq(&tasklist_lock);
1204                 spin_lock(&oldsighand->siglock);
1205                 rcu_assign_pointer(me->sighand, newsighand);
1206                 spin_unlock(&oldsighand->siglock);
1207                 write_unlock_irq(&tasklist_lock);
1208
1209                 __cleanup_sighand(oldsighand);
1210         }
1211         return 0;
1212 }
1213
1214 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1215 {
1216         task_lock(tsk);
1217         /* Always NUL terminated and zero-padded */
1218         strscpy_pad(buf, tsk->comm, buf_size);
1219         task_unlock(tsk);
1220         return buf;
1221 }
1222 EXPORT_SYMBOL_GPL(__get_task_comm);
1223
1224 /*
1225  * These functions flushes out all traces of the currently running executable
1226  * so that a new one can be started
1227  */
1228
1229 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1230 {
1231         task_lock(tsk);
1232         trace_task_rename(tsk, buf);
1233         strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1234         task_unlock(tsk);
1235         perf_event_comm(tsk, exec);
1236 }
1237
1238 /*
1239  * Calling this is the point of no return. None of the failures will be
1240  * seen by userspace since either the process is already taking a fatal
1241  * signal (via de_thread() or coredump), or will have SEGV raised
1242  * (after exec_mmap()) by search_binary_handler (see below).
1243  */
1244 int begin_new_exec(struct linux_binprm * bprm)
1245 {
1246         struct task_struct *me = current;
1247         int retval;
1248
1249         /* Once we are committed compute the creds */
1250         retval = bprm_creds_from_file(bprm);
1251         if (retval)
1252                 return retval;
1253
1254         /*
1255          * Ensure all future errors are fatal.
1256          */
1257         bprm->point_of_no_return = true;
1258
1259         /*
1260          * Make this the only thread in the thread group.
1261          */
1262         retval = de_thread(me);
1263         if (retval)
1264                 goto out;
1265
1266         /*
1267          * Cancel any io_uring activity across execve
1268          */
1269         io_uring_task_cancel();
1270
1271         /* Ensure the files table is not shared. */
1272         retval = unshare_files();
1273         if (retval)
1274                 goto out;
1275
1276         /*
1277          * Must be called _before_ exec_mmap() as bprm->mm is
1278          * not visible until then. This also enables the update
1279          * to be lockless.
1280          */
1281         retval = set_mm_exe_file(bprm->mm, bprm->file);
1282         if (retval)
1283                 goto out;
1284
1285         /* If the binary is not readable then enforce mm->dumpable=0 */
1286         would_dump(bprm, bprm->file);
1287         if (bprm->have_execfd)
1288                 would_dump(bprm, bprm->executable);
1289
1290         /*
1291          * Release all of the old mmap stuff
1292          */
1293         acct_arg_size(bprm, 0);
1294         retval = exec_mmap(bprm->mm);
1295         if (retval)
1296                 goto out;
1297
1298         bprm->mm = NULL;
1299
1300 #ifdef CONFIG_POSIX_TIMERS
1301         exit_itimers(me->signal);
1302         flush_itimer_signals();
1303 #endif
1304
1305         /*
1306          * Make the signal table private.
1307          */
1308         retval = unshare_sighand(me);
1309         if (retval)
1310                 goto out_unlock;
1311
1312         /*
1313          * Ensure that the uaccess routines can actually operate on userspace
1314          * pointers:
1315          */
1316         force_uaccess_begin();
1317
1318         if (me->flags & PF_KTHREAD)
1319                 free_kthread_struct(me);
1320         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1321                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1322         flush_thread();
1323         me->personality &= ~bprm->per_clear;
1324
1325         clear_syscall_work_syscall_user_dispatch(me);
1326
1327         /*
1328          * We have to apply CLOEXEC before we change whether the process is
1329          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1330          * trying to access the should-be-closed file descriptors of a process
1331          * undergoing exec(2).
1332          */
1333         do_close_on_exec(me->files);
1334
1335         if (bprm->secureexec) {
1336                 /* Make sure parent cannot signal privileged process. */
1337                 me->pdeath_signal = 0;
1338
1339                 /*
1340                  * For secureexec, reset the stack limit to sane default to
1341                  * avoid bad behavior from the prior rlimits. This has to
1342                  * happen before arch_pick_mmap_layout(), which examines
1343                  * RLIMIT_STACK, but after the point of no return to avoid
1344                  * needing to clean up the change on failure.
1345                  */
1346                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1347                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1348         }
1349
1350         me->sas_ss_sp = me->sas_ss_size = 0;
1351
1352         /*
1353          * Figure out dumpability. Note that this checking only of current
1354          * is wrong, but userspace depends on it. This should be testing
1355          * bprm->secureexec instead.
1356          */
1357         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1358             !(uid_eq(current_euid(), current_uid()) &&
1359               gid_eq(current_egid(), current_gid())))
1360                 set_dumpable(current->mm, suid_dumpable);
1361         else
1362                 set_dumpable(current->mm, SUID_DUMP_USER);
1363
1364         perf_event_exec();
1365         __set_task_comm(me, kbasename(bprm->filename), true);
1366
1367         /* An exec changes our domain. We are no longer part of the thread
1368            group */
1369         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1370         flush_signal_handlers(me, 0);
1371
1372         retval = set_cred_ucounts(bprm->cred);
1373         if (retval < 0)
1374                 goto out_unlock;
1375
1376         /*
1377          * install the new credentials for this executable
1378          */
1379         security_bprm_committing_creds(bprm);
1380
1381         commit_creds(bprm->cred);
1382         bprm->cred = NULL;
1383
1384         /*
1385          * Disable monitoring for regular users
1386          * when executing setuid binaries. Must
1387          * wait until new credentials are committed
1388          * by commit_creds() above
1389          */
1390         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1391                 perf_event_exit_task(me);
1392         /*
1393          * cred_guard_mutex must be held at least to this point to prevent
1394          * ptrace_attach() from altering our determination of the task's
1395          * credentials; any time after this it may be unlocked.
1396          */
1397         security_bprm_committed_creds(bprm);
1398
1399         /* Pass the opened binary to the interpreter. */
1400         if (bprm->have_execfd) {
1401                 retval = get_unused_fd_flags(0);
1402                 if (retval < 0)
1403                         goto out_unlock;
1404                 fd_install(retval, bprm->executable);
1405                 bprm->executable = NULL;
1406                 bprm->execfd = retval;
1407         }
1408         return 0;
1409
1410 out_unlock:
1411         up_write(&me->signal->exec_update_lock);
1412 out:
1413         return retval;
1414 }
1415 EXPORT_SYMBOL(begin_new_exec);
1416
1417 void would_dump(struct linux_binprm *bprm, struct file *file)
1418 {
1419         struct inode *inode = file_inode(file);
1420         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1421         if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1422                 struct user_namespace *old, *user_ns;
1423                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1424
1425                 /* Ensure mm->user_ns contains the executable */
1426                 user_ns = old = bprm->mm->user_ns;
1427                 while ((user_ns != &init_user_ns) &&
1428                        !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1429                         user_ns = user_ns->parent;
1430
1431                 if (old != user_ns) {
1432                         bprm->mm->user_ns = get_user_ns(user_ns);
1433                         put_user_ns(old);
1434                 }
1435         }
1436 }
1437 EXPORT_SYMBOL(would_dump);
1438
1439 void setup_new_exec(struct linux_binprm * bprm)
1440 {
1441         /* Setup things that can depend upon the personality */
1442         struct task_struct *me = current;
1443
1444         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1445
1446         arch_setup_new_exec();
1447
1448         /* Set the new mm task size. We have to do that late because it may
1449          * depend on TIF_32BIT which is only updated in flush_thread() on
1450          * some architectures like powerpc
1451          */
1452         me->mm->task_size = TASK_SIZE;
1453         up_write(&me->signal->exec_update_lock);
1454         mutex_unlock(&me->signal->cred_guard_mutex);
1455 }
1456 EXPORT_SYMBOL(setup_new_exec);
1457
1458 /* Runs immediately before start_thread() takes over. */
1459 void finalize_exec(struct linux_binprm *bprm)
1460 {
1461         /* Store any stack rlimit changes before starting thread. */
1462         task_lock(current->group_leader);
1463         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1464         task_unlock(current->group_leader);
1465 }
1466 EXPORT_SYMBOL(finalize_exec);
1467
1468 /*
1469  * Prepare credentials and lock ->cred_guard_mutex.
1470  * setup_new_exec() commits the new creds and drops the lock.
1471  * Or, if exec fails before, free_bprm() should release ->cred
1472  * and unlock.
1473  */
1474 static int prepare_bprm_creds(struct linux_binprm *bprm)
1475 {
1476         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1477                 return -ERESTARTNOINTR;
1478
1479         bprm->cred = prepare_exec_creds();
1480         if (likely(bprm->cred))
1481                 return 0;
1482
1483         mutex_unlock(&current->signal->cred_guard_mutex);
1484         return -ENOMEM;
1485 }
1486
1487 static void free_bprm(struct linux_binprm *bprm)
1488 {
1489         if (bprm->mm) {
1490                 acct_arg_size(bprm, 0);
1491                 mmput(bprm->mm);
1492         }
1493         free_arg_pages(bprm);
1494         if (bprm->cred) {
1495                 mutex_unlock(&current->signal->cred_guard_mutex);
1496                 abort_creds(bprm->cred);
1497         }
1498         if (bprm->file) {
1499                 allow_write_access(bprm->file);
1500                 fput(bprm->file);
1501         }
1502         if (bprm->executable)
1503                 fput(bprm->executable);
1504         /* If a binfmt changed the interp, free it. */
1505         if (bprm->interp != bprm->filename)
1506                 kfree(bprm->interp);
1507         kfree(bprm->fdpath);
1508         kfree(bprm);
1509 }
1510
1511 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1512 {
1513         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1514         int retval = -ENOMEM;
1515         if (!bprm)
1516                 goto out;
1517
1518         if (fd == AT_FDCWD || filename->name[0] == '/') {
1519                 bprm->filename = filename->name;
1520         } else {
1521                 if (filename->name[0] == '\0')
1522                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1523                 else
1524                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1525                                                   fd, filename->name);
1526                 if (!bprm->fdpath)
1527                         goto out_free;
1528
1529                 bprm->filename = bprm->fdpath;
1530         }
1531         bprm->interp = bprm->filename;
1532
1533         retval = bprm_mm_init(bprm);
1534         if (retval)
1535                 goto out_free;
1536         return bprm;
1537
1538 out_free:
1539         free_bprm(bprm);
1540 out:
1541         return ERR_PTR(retval);
1542 }
1543
1544 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1545 {
1546         /* If a binfmt changed the interp, free it first. */
1547         if (bprm->interp != bprm->filename)
1548                 kfree(bprm->interp);
1549         bprm->interp = kstrdup(interp, GFP_KERNEL);
1550         if (!bprm->interp)
1551                 return -ENOMEM;
1552         return 0;
1553 }
1554 EXPORT_SYMBOL(bprm_change_interp);
1555
1556 /*
1557  * determine how safe it is to execute the proposed program
1558  * - the caller must hold ->cred_guard_mutex to protect against
1559  *   PTRACE_ATTACH or seccomp thread-sync
1560  */
1561 static void check_unsafe_exec(struct linux_binprm *bprm)
1562 {
1563         struct task_struct *p = current, *t;
1564         unsigned n_fs;
1565
1566         if (p->ptrace)
1567                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1568
1569         /*
1570          * This isn't strictly necessary, but it makes it harder for LSMs to
1571          * mess up.
1572          */
1573         if (task_no_new_privs(current))
1574                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1575
1576         t = p;
1577         n_fs = 1;
1578         spin_lock(&p->fs->lock);
1579         rcu_read_lock();
1580         while_each_thread(p, t) {
1581                 if (t->fs == p->fs)
1582                         n_fs++;
1583         }
1584         rcu_read_unlock();
1585
1586         if (p->fs->users > n_fs)
1587                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1588         else
1589                 p->fs->in_exec = 1;
1590         spin_unlock(&p->fs->lock);
1591 }
1592
1593 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1594 {
1595         /* Handle suid and sgid on files */
1596         struct user_namespace *mnt_userns;
1597         struct inode *inode;
1598         unsigned int mode;
1599         kuid_t uid;
1600         kgid_t gid;
1601
1602         if (!mnt_may_suid(file->f_path.mnt))
1603                 return;
1604
1605         if (task_no_new_privs(current))
1606                 return;
1607
1608         inode = file->f_path.dentry->d_inode;
1609         mode = READ_ONCE(inode->i_mode);
1610         if (!(mode & (S_ISUID|S_ISGID)))
1611                 return;
1612
1613         mnt_userns = file_mnt_user_ns(file);
1614
1615         /* Be careful if suid/sgid is set */
1616         inode_lock(inode);
1617
1618         /* reload atomically mode/uid/gid now that lock held */
1619         mode = inode->i_mode;
1620         uid = i_uid_into_mnt(mnt_userns, inode);
1621         gid = i_gid_into_mnt(mnt_userns, inode);
1622         inode_unlock(inode);
1623
1624         /* We ignore suid/sgid if there are no mappings for them in the ns */
1625         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1626                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1627                 return;
1628
1629         if (mode & S_ISUID) {
1630                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1631                 bprm->cred->euid = uid;
1632         }
1633
1634         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1635                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1636                 bprm->cred->egid = gid;
1637         }
1638 }
1639
1640 /*
1641  * Compute brpm->cred based upon the final binary.
1642  */
1643 static int bprm_creds_from_file(struct linux_binprm *bprm)
1644 {
1645         /* Compute creds based on which file? */
1646         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1647
1648         bprm_fill_uid(bprm, file);
1649         return security_bprm_creds_from_file(bprm, file);
1650 }
1651
1652 /*
1653  * Fill the binprm structure from the inode.
1654  * Read the first BINPRM_BUF_SIZE bytes
1655  *
1656  * This may be called multiple times for binary chains (scripts for example).
1657  */
1658 static int prepare_binprm(struct linux_binprm *bprm)
1659 {
1660         loff_t pos = 0;
1661
1662         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1663         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1664 }
1665
1666 /*
1667  * Arguments are '\0' separated strings found at the location bprm->p
1668  * points to; chop off the first by relocating brpm->p to right after
1669  * the first '\0' encountered.
1670  */
1671 int remove_arg_zero(struct linux_binprm *bprm)
1672 {
1673         int ret = 0;
1674         unsigned long offset;
1675         char *kaddr;
1676         struct page *page;
1677
1678         if (!bprm->argc)
1679                 return 0;
1680
1681         do {
1682                 offset = bprm->p & ~PAGE_MASK;
1683                 page = get_arg_page(bprm, bprm->p, 0);
1684                 if (!page) {
1685                         ret = -EFAULT;
1686                         goto out;
1687                 }
1688                 kaddr = kmap_atomic(page);
1689
1690                 for (; offset < PAGE_SIZE && kaddr[offset];
1691                                 offset++, bprm->p++)
1692                         ;
1693
1694                 kunmap_atomic(kaddr);
1695                 put_arg_page(page);
1696         } while (offset == PAGE_SIZE);
1697
1698         bprm->p++;
1699         bprm->argc--;
1700         ret = 0;
1701
1702 out:
1703         return ret;
1704 }
1705 EXPORT_SYMBOL(remove_arg_zero);
1706
1707 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1708 /*
1709  * cycle the list of binary formats handler, until one recognizes the image
1710  */
1711 static int search_binary_handler(struct linux_binprm *bprm)
1712 {
1713         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1714         struct linux_binfmt *fmt;
1715         int retval;
1716
1717         retval = prepare_binprm(bprm);
1718         if (retval < 0)
1719                 return retval;
1720
1721         retval = security_bprm_check(bprm);
1722         if (retval)
1723                 return retval;
1724
1725         retval = -ENOENT;
1726  retry:
1727         read_lock(&binfmt_lock);
1728         list_for_each_entry(fmt, &formats, lh) {
1729                 if (!try_module_get(fmt->module))
1730                         continue;
1731                 read_unlock(&binfmt_lock);
1732
1733                 retval = fmt->load_binary(bprm);
1734
1735                 read_lock(&binfmt_lock);
1736                 put_binfmt(fmt);
1737                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1738                         read_unlock(&binfmt_lock);
1739                         return retval;
1740                 }
1741         }
1742         read_unlock(&binfmt_lock);
1743
1744         if (need_retry) {
1745                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1746                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1747                         return retval;
1748                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1749                         return retval;
1750                 need_retry = false;
1751                 goto retry;
1752         }
1753
1754         return retval;
1755 }
1756
1757 static int exec_binprm(struct linux_binprm *bprm)
1758 {
1759         pid_t old_pid, old_vpid;
1760         int ret, depth;
1761
1762         /* Need to fetch pid before load_binary changes it */
1763         old_pid = current->pid;
1764         rcu_read_lock();
1765         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1766         rcu_read_unlock();
1767
1768         /* This allows 4 levels of binfmt rewrites before failing hard. */
1769         for (depth = 0;; depth++) {
1770                 struct file *exec;
1771                 if (depth > 5)
1772                         return -ELOOP;
1773
1774                 ret = search_binary_handler(bprm);
1775                 if (ret < 0)
1776                         return ret;
1777                 if (!bprm->interpreter)
1778                         break;
1779
1780                 exec = bprm->file;
1781                 bprm->file = bprm->interpreter;
1782                 bprm->interpreter = NULL;
1783
1784                 allow_write_access(exec);
1785                 if (unlikely(bprm->have_execfd)) {
1786                         if (bprm->executable) {
1787                                 fput(exec);
1788                                 return -ENOEXEC;
1789                         }
1790                         bprm->executable = exec;
1791                 } else
1792                         fput(exec);
1793         }
1794
1795         audit_bprm(bprm);
1796         trace_sched_process_exec(current, old_pid, bprm);
1797         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1798         proc_exec_connector(current);
1799         return 0;
1800 }
1801
1802 /*
1803  * sys_execve() executes a new program.
1804  */
1805 static int bprm_execve(struct linux_binprm *bprm,
1806                        int fd, struct filename *filename, int flags)
1807 {
1808         struct file *file;
1809         int retval;
1810
1811         retval = prepare_bprm_creds(bprm);
1812         if (retval)
1813                 return retval;
1814
1815         check_unsafe_exec(bprm);
1816         current->in_execve = 1;
1817
1818         file = do_open_execat(fd, filename, flags);
1819         retval = PTR_ERR(file);
1820         if (IS_ERR(file))
1821                 goto out_unmark;
1822
1823         sched_exec();
1824
1825         bprm->file = file;
1826         /*
1827          * Record that a name derived from an O_CLOEXEC fd will be
1828          * inaccessible after exec.  This allows the code in exec to
1829          * choose to fail when the executable is not mmaped into the
1830          * interpreter and an open file descriptor is not passed to
1831          * the interpreter.  This makes for a better user experience
1832          * than having the interpreter start and then immediately fail
1833          * when it finds the executable is inaccessible.
1834          */
1835         if (bprm->fdpath && get_close_on_exec(fd))
1836                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1837
1838         /* Set the unchanging part of bprm->cred */
1839         retval = security_bprm_creds_for_exec(bprm);
1840         if (retval)
1841                 goto out;
1842
1843         retval = exec_binprm(bprm);
1844         if (retval < 0)
1845                 goto out;
1846
1847         /* execve succeeded */
1848         current->fs->in_exec = 0;
1849         current->in_execve = 0;
1850         rseq_execve(current);
1851         acct_update_integrals(current);
1852         task_numa_free(current, false);
1853         return retval;
1854
1855 out:
1856         /*
1857          * If past the point of no return ensure the code never
1858          * returns to the userspace process.  Use an existing fatal
1859          * signal if present otherwise terminate the process with
1860          * SIGSEGV.
1861          */
1862         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1863                 force_fatal_sig(SIGSEGV);
1864
1865 out_unmark:
1866         current->fs->in_exec = 0;
1867         current->in_execve = 0;
1868
1869         return retval;
1870 }
1871
1872 static int do_execveat_common(int fd, struct filename *filename,
1873                               struct user_arg_ptr argv,
1874                               struct user_arg_ptr envp,
1875                               int flags)
1876 {
1877         struct linux_binprm *bprm;
1878         int retval;
1879
1880         if (IS_ERR(filename))
1881                 return PTR_ERR(filename);
1882
1883         /*
1884          * We move the actual failure in case of RLIMIT_NPROC excess from
1885          * set*uid() to execve() because too many poorly written programs
1886          * don't check setuid() return code.  Here we additionally recheck
1887          * whether NPROC limit is still exceeded.
1888          */
1889         if ((current->flags & PF_NPROC_EXCEEDED) &&
1890             is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1891                 retval = -EAGAIN;
1892                 goto out_ret;
1893         }
1894
1895         /* We're below the limit (still or again), so we don't want to make
1896          * further execve() calls fail. */
1897         current->flags &= ~PF_NPROC_EXCEEDED;
1898
1899         bprm = alloc_bprm(fd, filename);
1900         if (IS_ERR(bprm)) {
1901                 retval = PTR_ERR(bprm);
1902                 goto out_ret;
1903         }
1904
1905         retval = count(argv, MAX_ARG_STRINGS);
1906         if (retval == 0)
1907                 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1908                              current->comm, bprm->filename);
1909         if (retval < 0)
1910                 goto out_free;
1911         bprm->argc = retval;
1912
1913         retval = count(envp, MAX_ARG_STRINGS);
1914         if (retval < 0)
1915                 goto out_free;
1916         bprm->envc = retval;
1917
1918         retval = bprm_stack_limits(bprm);
1919         if (retval < 0)
1920                 goto out_free;
1921
1922         retval = copy_string_kernel(bprm->filename, bprm);
1923         if (retval < 0)
1924                 goto out_free;
1925         bprm->exec = bprm->p;
1926
1927         retval = copy_strings(bprm->envc, envp, bprm);
1928         if (retval < 0)
1929                 goto out_free;
1930
1931         retval = copy_strings(bprm->argc, argv, bprm);
1932         if (retval < 0)
1933                 goto out_free;
1934
1935         /*
1936          * When argv is empty, add an empty string ("") as argv[0] to
1937          * ensure confused userspace programs that start processing
1938          * from argv[1] won't end up walking envp. See also
1939          * bprm_stack_limits().
1940          */
1941         if (bprm->argc == 0) {
1942                 retval = copy_string_kernel("", bprm);
1943                 if (retval < 0)
1944                         goto out_free;
1945                 bprm->argc = 1;
1946         }
1947
1948         retval = bprm_execve(bprm, fd, filename, flags);
1949 out_free:
1950         free_bprm(bprm);
1951
1952 out_ret:
1953         putname(filename);
1954         return retval;
1955 }
1956
1957 int kernel_execve(const char *kernel_filename,
1958                   const char *const *argv, const char *const *envp)
1959 {
1960         struct filename *filename;
1961         struct linux_binprm *bprm;
1962         int fd = AT_FDCWD;
1963         int retval;
1964
1965         filename = getname_kernel(kernel_filename);
1966         if (IS_ERR(filename))
1967                 return PTR_ERR(filename);
1968
1969         bprm = alloc_bprm(fd, filename);
1970         if (IS_ERR(bprm)) {
1971                 retval = PTR_ERR(bprm);
1972                 goto out_ret;
1973         }
1974
1975         retval = count_strings_kernel(argv);
1976         if (WARN_ON_ONCE(retval == 0))
1977                 retval = -EINVAL;
1978         if (retval < 0)
1979                 goto out_free;
1980         bprm->argc = retval;
1981
1982         retval = count_strings_kernel(envp);
1983         if (retval < 0)
1984                 goto out_free;
1985         bprm->envc = retval;
1986
1987         retval = bprm_stack_limits(bprm);
1988         if (retval < 0)
1989                 goto out_free;
1990
1991         retval = copy_string_kernel(bprm->filename, bprm);
1992         if (retval < 0)
1993                 goto out_free;
1994         bprm->exec = bprm->p;
1995
1996         retval = copy_strings_kernel(bprm->envc, envp, bprm);
1997         if (retval < 0)
1998                 goto out_free;
1999
2000         retval = copy_strings_kernel(bprm->argc, argv, bprm);
2001         if (retval < 0)
2002                 goto out_free;
2003
2004         retval = bprm_execve(bprm, fd, filename, 0);
2005 out_free:
2006         free_bprm(bprm);
2007 out_ret:
2008         putname(filename);
2009         return retval;
2010 }
2011
2012 static int do_execve(struct filename *filename,
2013         const char __user *const __user *__argv,
2014         const char __user *const __user *__envp)
2015 {
2016         struct user_arg_ptr argv = { .ptr.native = __argv };
2017         struct user_arg_ptr envp = { .ptr.native = __envp };
2018         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2019 }
2020
2021 static int do_execveat(int fd, struct filename *filename,
2022                 const char __user *const __user *__argv,
2023                 const char __user *const __user *__envp,
2024                 int flags)
2025 {
2026         struct user_arg_ptr argv = { .ptr.native = __argv };
2027         struct user_arg_ptr envp = { .ptr.native = __envp };
2028
2029         return do_execveat_common(fd, filename, argv, envp, flags);
2030 }
2031
2032 #ifdef CONFIG_COMPAT
2033 static int compat_do_execve(struct filename *filename,
2034         const compat_uptr_t __user *__argv,
2035         const compat_uptr_t __user *__envp)
2036 {
2037         struct user_arg_ptr argv = {
2038                 .is_compat = true,
2039                 .ptr.compat = __argv,
2040         };
2041         struct user_arg_ptr envp = {
2042                 .is_compat = true,
2043                 .ptr.compat = __envp,
2044         };
2045         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2046 }
2047
2048 static int compat_do_execveat(int fd, struct filename *filename,
2049                               const compat_uptr_t __user *__argv,
2050                               const compat_uptr_t __user *__envp,
2051                               int flags)
2052 {
2053         struct user_arg_ptr argv = {
2054                 .is_compat = true,
2055                 .ptr.compat = __argv,
2056         };
2057         struct user_arg_ptr envp = {
2058                 .is_compat = true,
2059                 .ptr.compat = __envp,
2060         };
2061         return do_execveat_common(fd, filename, argv, envp, flags);
2062 }
2063 #endif
2064
2065 void set_binfmt(struct linux_binfmt *new)
2066 {
2067         struct mm_struct *mm = current->mm;
2068
2069         if (mm->binfmt)
2070                 module_put(mm->binfmt->module);
2071
2072         mm->binfmt = new;
2073         if (new)
2074                 __module_get(new->module);
2075 }
2076 EXPORT_SYMBOL(set_binfmt);
2077
2078 /*
2079  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2080  */
2081 void set_dumpable(struct mm_struct *mm, int value)
2082 {
2083         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2084                 return;
2085
2086         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2087 }
2088
2089 SYSCALL_DEFINE3(execve,
2090                 const char __user *, filename,
2091                 const char __user *const __user *, argv,
2092                 const char __user *const __user *, envp)
2093 {
2094         return do_execve(getname(filename), argv, envp);
2095 }
2096
2097 SYSCALL_DEFINE5(execveat,
2098                 int, fd, const char __user *, filename,
2099                 const char __user *const __user *, argv,
2100                 const char __user *const __user *, envp,
2101                 int, flags)
2102 {
2103         return do_execveat(fd,
2104                            getname_uflags(filename, flags),
2105                            argv, envp, flags);
2106 }
2107
2108 #ifdef CONFIG_COMPAT
2109 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2110         const compat_uptr_t __user *, argv,
2111         const compat_uptr_t __user *, envp)
2112 {
2113         return compat_do_execve(getname(filename), argv, envp);
2114 }
2115
2116 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2117                        const char __user *, filename,
2118                        const compat_uptr_t __user *, argv,
2119                        const compat_uptr_t __user *, envp,
2120                        int,  flags)
2121 {
2122         return compat_do_execveat(fd,
2123                                   getname_uflags(filename, flags),
2124                                   argv, envp, flags);
2125 }
2126 #endif
2127
2128 #ifdef CONFIG_SYSCTL
2129
2130 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2131                 void *buffer, size_t *lenp, loff_t *ppos)
2132 {
2133         int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2134
2135         if (!error)
2136                 validate_coredump_safety();
2137         return error;
2138 }
2139
2140 static struct ctl_table fs_exec_sysctls[] = {
2141         {
2142                 .procname       = "suid_dumpable",
2143                 .data           = &suid_dumpable,
2144                 .maxlen         = sizeof(int),
2145                 .mode           = 0644,
2146                 .proc_handler   = proc_dointvec_minmax_coredump,
2147                 .extra1         = SYSCTL_ZERO,
2148                 .extra2         = SYSCTL_TWO,
2149         },
2150         { }
2151 };
2152
2153 static int __init init_fs_exec_sysctls(void)
2154 {
2155         register_sysctl_init("fs", fs_exec_sysctls);
2156         return 0;
2157 }
2158
2159 fs_initcall(init_fs_exec_sysctls);
2160 #endif /* CONFIG_SYSCTL */