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/stat.h>
  32 #include <linux/fcntl.h>
  33 #include <linux/swap.h>
  34 #include <linux/string.h>
  35 #include <linux/init.h>
  36 #include <linux/sched/mm.h>
  37 #include <linux/sched/coredump.h>
  38 #include <linux/sched/signal.h>
  39 #include <linux/sched/numa_balancing.h>
  40 #include <linux/sched/task.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/perf_event.h>
  43 #include <linux/highmem.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/key.h>
  46 #include <linux/personality.h>
  47 #include <linux/binfmts.h>
  48 #include <linux/utsname.h>
  49 #include <linux/pid_namespace.h>
  50 #include <linux/module.h>
  51 #include <linux/namei.h>
  52 #include <linux/mount.h>
  53 #include <linux/security.h>
  54 #include <linux/syscalls.h>
  55 #include <linux/tsacct_kern.h>
  56 #include <linux/cn_proc.h>
  57 #include <linux/audit.h>
  58 #include <linux/kmod.h>
  59 #include <linux/fsnotify.h>
  60 #include <linux/fs_struct.h>
  61 #include <linux/oom.h>
  62 #include <linux/compat.h>
  63 #include <linux/vmalloc.h>
  64 #include <linux/io_uring.h>
  65 #include <linux/syscall_user_dispatch.h>
  66 #include <linux/coredump.h>
  67
  68 #include <linux/uaccess.h>
  69 #include <asm/mmu_context.h>
  70 #include <asm/tlb.h>
  71
  72 #include <trace/events/task.h>
  73 #include "internal.h"
  74
  75 #include <trace/events/sched.h>
  76
  77 static int bprm_creds_from_file(struct linux_binprm *bprm);
  78
  79 int suid_dumpable = 0;
  80
  81 static LIST_HEAD(formats);
  82 static DEFINE_RWLOCK(binfmt_lock);
  83
  84 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  85 {
  86         write_lock(&binfmt_lock);
  87         insert ? list_add(&fmt->lh, &formats) :
  88                  list_add_tail(&fmt->lh, &formats);
  89         write_unlock(&binfmt_lock);
  90 }
  91
  92 EXPORT_SYMBOL(__register_binfmt);
  93
  94 void unregister_binfmt(struct linux_binfmt * fmt)
  95 {
  96         write_lock(&binfmt_lock);
  97         list_del(&fmt->lh);
  98         write_unlock(&binfmt_lock);
  99 }
 100
 101 EXPORT_SYMBOL(unregister_binfmt);
 102
 103 static inline void put_binfmt(struct linux_binfmt * fmt)
 104 {
 105         module_put(fmt->module);
 106 }
 107
 108 bool path_noexec(const struct path *path)
 109 {
 110         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 111                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 112 }
 113
 114 #ifdef CONFIG_USELIB
 115 /*
 116  * Note that a shared library must be both readable and executable due to
 117  * security reasons.
 118  *
 119  * Also note that we take the address to load from the file itself.
 120  */
 121 SYSCALL_DEFINE1(uselib, const char __user *, library)
 122 {
 123         struct linux_binfmt *fmt;
 124         struct file *file;
 125         struct filename *tmp = getname(library);
 126         int error = PTR_ERR(tmp);
 127         static const struct open_flags uselib_flags = {
 128                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 129                 .acc_mode = MAY_READ | MAY_EXEC,
 130                 .intent = LOOKUP_OPEN,
 131                 .lookup_flags = LOOKUP_FOLLOW,
 132         };
 133
 134         if (IS_ERR(tmp))
 135                 goto out;
 136
 137         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 138         putname(tmp);
 139         error = PTR_ERR(file);
 140         if (IS_ERR(file))
 141                 goto out;
 142
 143         /*
 144          * may_open() has already checked for this, so it should be
 145          * impossible to trip now. But we need to be extra cautious
 146          * and check again at the very end too.
 147          */
 148         error = -EACCES;
 149         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 150                          path_noexec(&file->f_path)))
 151                 goto exit;
 152
 153         fsnotify_open(file);
 154
 155         error = -ENOEXEC;
 156
 157         read_lock(&binfmt_lock);
 158         list_for_each_entry(fmt, &formats, lh) {
 159                 if (!fmt->load_shlib)
 160                         continue;
 161                 if (!try_module_get(fmt->module))
 162                         continue;
 163                 read_unlock(&binfmt_lock);
 164                 error = fmt->load_shlib(file);
 165                 read_lock(&binfmt_lock);
 166                 put_binfmt(fmt);
 167                 if (error != -ENOEXEC)
 168                         break;
 169         }
 170         read_unlock(&binfmt_lock);
 171 exit:
 172         fput(file);
 173 out:
 174         return error;
 175 }
 176 #endif /* #ifdef CONFIG_USELIB */
 177
 178 #ifdef CONFIG_MMU
 179 /*
 180  * The nascent bprm->mm is not visible until exec_mmap() but it can
 181  * use a lot of memory, account these pages in current->mm temporary
 182  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 183  * change the counter back via acct_arg_size(0).
 184  */
 185 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 186 {
 187         struct mm_struct *mm = current->mm;
 188         long diff = (long)(pages - bprm->vma_pages);
 189
 190         if (!mm || !diff)
 191                 return;
 192
 193         bprm->vma_pages = pages;
 194         add_mm_counter(mm, MM_ANONPAGES, diff);
 195 }
 196
 197 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 198                 int write)
 199 {
 200         struct page *page;
 201         struct vm_area_struct *vma = bprm->vma;
 202         struct mm_struct *mm = bprm->mm;
 203         int ret;
 204
 205         /*
 206          * Avoid relying on expanding the stack down in GUP (which
 207          * does not work for STACK_GROWSUP anyway), and just do it
 208          * by hand ahead of time.
 209          */
 210         if (write && pos < vma->vm_start) {
 211                 mmap_write_lock(mm);
 212                 ret = expand_downwards(vma, pos);
 213                 if (unlikely(ret < 0)) {
 214                         mmap_write_unlock(mm);
 215                         return NULL;
 216                 }
 217                 mmap_write_downgrade(mm);
 218         } else
 219                 mmap_read_lock(mm);
 220
 221         /*
 222          * We are doing an exec().  'current' is the process
 223          * doing the exec and 'mm' is the new process's mm.
 224          */
 225         ret = get_user_pages_remote(mm, pos, 1,
 226                         write ? FOLL_WRITE : 0,
 227                         &page, NULL, NULL);
 228         mmap_read_unlock(mm);
 229         if (ret <= 0)
 230                 return NULL;
 231
 232         if (write)
 233                 acct_arg_size(bprm, vma_pages(vma));
 234
 235         return page;
 236 }
 237
 238 static void put_arg_page(struct page *page)
 239 {
 240         put_page(page);
 241 }
 242
 243 static void free_arg_pages(struct linux_binprm *bprm)
 244 {
 245 }
 246
 247 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 248                 struct page *page)
 249 {
 250         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 251 }
 252
 253 static int __bprm_mm_init(struct linux_binprm *bprm)
 254 {
 255         int err;
 256         struct vm_area_struct *vma = NULL;
 257         struct mm_struct *mm = bprm->mm;
 258
 259         bprm->vma = vma = vm_area_alloc(mm);
 260         if (!vma)
 261                 return -ENOMEM;
 262         vma_set_anonymous(vma);
 263
 264         if (mmap_write_lock_killable(mm)) {
 265                 err = -EINTR;
 266                 goto err_free;
 267         }
 268
 269         /*
 270          * Place the stack at the largest stack address the architecture
 271          * supports. Later, we'll move this to an appropriate place. We don't
 272          * use STACK_TOP because that can depend on attributes which aren't
 273          * configured yet.
 274          */
 275         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 276         vma->vm_end = STACK_TOP_MAX;
 277         vma->vm_start = vma->vm_end - PAGE_SIZE;
 278         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 279         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 280
 281         err = insert_vm_struct(mm, vma);
 282         if (err)
 283                 goto err;
 284
 285         mm->stack_vm = mm->total_vm = 1;
 286         mmap_write_unlock(mm);
 287         bprm->p = vma->vm_end - sizeof(void *);
 288         return 0;
 289 err:
 290         mmap_write_unlock(mm);
 291 err_free:
 292         bprm->vma = NULL;
 293         vm_area_free(vma);
 294         return err;
 295 }
 296
 297 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 298 {
 299         return len <= MAX_ARG_STRLEN;
 300 }
 301
 302 #else
 303
 304 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 305 {
 306 }
 307
 308 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 309                 int write)
 310 {
 311         struct page *page;
 312
 313         page = bprm->page[pos / PAGE_SIZE];
 314         if (!page && write) {
 315                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 316                 if (!page)
 317                         return NULL;
 318                 bprm->page[pos / PAGE_SIZE] = page;
 319         }
 320
 321         return page;
 322 }
 323
 324 static void put_arg_page(struct page *page)
 325 {
 326 }
 327
 328 static void free_arg_page(struct linux_binprm *bprm, int i)
 329 {
 330         if (bprm->page[i]) {
 331                 __free_page(bprm->page[i]);
 332                 bprm->page[i] = NULL;
 333         }
 334 }
 335
 336 static void free_arg_pages(struct linux_binprm *bprm)
 337 {
 338         int i;
 339
 340         for (i = 0; i < MAX_ARG_PAGES; i++)
 341                 free_arg_page(bprm, i);
 342 }
 343
 344 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 345                 struct page *page)
 346 {
 347 }
 348
 349 static int __bprm_mm_init(struct linux_binprm *bprm)
 350 {
 351         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 352         return 0;
 353 }
 354
 355 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 356 {
 357         return len <= bprm->p;
 358 }
 359
 360 #endif /* CONFIG_MMU */
 361
 362 /*
 363  * Create a new mm_struct and populate it with a temporary stack
 364  * vm_area_struct.  We don't have enough context at this point to set the stack
 365  * flags, permissions, and offset, so we use temporary values.  We'll update
 366  * them later in setup_arg_pages().
 367  */
 368 static int bprm_mm_init(struct linux_binprm *bprm)
 369 {
 370         int err;
 371         struct mm_struct *mm = NULL;
 372
 373         bprm->mm = mm = mm_alloc();
 374         err = -ENOMEM;
 375         if (!mm)
 376                 goto err;
 377
 378         /* Save current stack limit for all calculations made during exec. */
 379         task_lock(current->group_leader);
 380         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 381         task_unlock(current->group_leader);
 382
 383         err = __bprm_mm_init(bprm);
 384         if (err)
 385                 goto err;
 386
 387         return 0;
 388
 389 err:
 390         if (mm) {
 391                 bprm->mm = NULL;
 392                 mmdrop(mm);
 393         }
 394
 395         return err;
 396 }
 397
 398 struct user_arg_ptr {
 399 #ifdef CONFIG_COMPAT
 400         bool is_compat;
 401 #endif
 402         union {
 403                 const char __user *const __user *native;
 404 #ifdef CONFIG_COMPAT
 405                 const compat_uptr_t __user *compat;
 406 #endif
 407         } ptr;
 408 };
 409
 410 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 411 {
 412         const char __user *native;
 413
 414 #ifdef CONFIG_COMPAT
 415         if (unlikely(argv.is_compat)) {
 416                 compat_uptr_t compat;
 417
 418                 if (get_user(compat, argv.ptr.compat + nr))
 419                         return ERR_PTR(-EFAULT);
 420
 421                 return compat_ptr(compat);
 422         }
 423 #endif
 424
 425         if (get_user(native, argv.ptr.native + nr))
 426                 return ERR_PTR(-EFAULT);
 427
 428         return native;
 429 }
 430
 431 /*
 432  * count() counts the number of strings in array ARGV.
 433  */
 434 static int count(struct user_arg_ptr argv, int max)
 435 {
 436         int i = 0;
 437
 438         if (argv.ptr.native != NULL) {
 439                 for (;;) {
 440                         const char __user *p = get_user_arg_ptr(argv, i);
 441
 442                         if (!p)
 443                                 break;
 444
 445                         if (IS_ERR(p))
 446                                 return -EFAULT;
 447
 448                         if (i >= max)
 449                                 return -E2BIG;
 450                         ++i;
 451
 452                         if (fatal_signal_pending(current))
 453                                 return -ERESTARTNOHAND;
 454                         cond_resched();
 455                 }
 456         }
 457         return i;
 458 }
 459
 460 static int count_strings_kernel(const char *const *argv)
 461 {
 462         int i;
 463
 464         if (!argv)
 465                 return 0;
 466
 467         for (i = 0; argv[i]; ++i) {
 468                 if (i >= MAX_ARG_STRINGS)
 469                         return -E2BIG;
 470                 if (fatal_signal_pending(current))
 471                         return -ERESTARTNOHAND;
 472                 cond_resched();
 473         }
 474         return i;
 475 }
 476
 477 static int bprm_stack_limits(struct linux_binprm *bprm)
 478 {
 479         unsigned long limit, ptr_size;
 480
 481         /*
 482          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 483          * (whichever is smaller) for the argv+env strings.
 484          * This ensures that:
 485          *  - the remaining binfmt code will not run out of stack space,
 486          *  - the program will have a reasonable amount of stack left
 487          *    to work from.
 488          */
 489         limit = _STK_LIM / 4 * 3;
 490         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 491         /*
 492          * We've historically supported up to 32 pages (ARG_MAX)
 493          * of argument strings even with small stacks
 494          */
 495         limit = max_t(unsigned long, limit, ARG_MAX);
 496         /*
 497          * We must account for the size of all the argv and envp pointers to
 498          * the argv and envp strings, since they will also take up space in
 499          * the stack. They aren't stored until much later when we can't
 500          * signal to the parent that the child has run out of stack space.
 501          * Instead, calculate it here so it's possible to fail gracefully.
 502          *
 503          * In the case of argc = 0, make sure there is space for adding a
 504          * empty string (which will bump argc to 1), to ensure confused
 505          * userspace programs don't start processing from argv[1], thinking
 506          * argc can never be 0, to keep them from walking envp by accident.
 507          * See do_execveat_common().
 508          */
 509         ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
 510         if (limit <= ptr_size)
 511                 return -E2BIG;
 512         limit -= ptr_size;
 513
 514         bprm->argmin = bprm->p - limit;
 515         return 0;
 516 }
 517
 518 /*
 519  * 'copy_strings()' copies argument/environment strings from the old
 520  * processes's memory to the new process's stack.  The call to get_user_pages()
 521  * ensures the destination page is created and not swapped out.
 522  */
 523 static int copy_strings(int argc, struct user_arg_ptr argv,
 524                         struct linux_binprm *bprm)
 525 {
 526         struct page *kmapped_page = NULL;
 527         char *kaddr = NULL;
 528         unsigned long kpos = 0;
 529         int ret;
 530
 531         while (argc-- > 0) {
 532                 const char __user *str;
 533                 int len;
 534                 unsigned long pos;
 535
 536                 ret = -EFAULT;
 537                 str = get_user_arg_ptr(argv, argc);
 538                 if (IS_ERR(str))
 539                         goto out;
 540
 541                 len = strnlen_user(str, MAX_ARG_STRLEN);
 542                 if (!len)
 543                         goto out;
 544
 545                 ret = -E2BIG;
 546                 if (!valid_arg_len(bprm, len))
 547                         goto out;
 548
 549                 /* We're going to work our way backwards. */
 550                 pos = bprm->p;
 551                 str += len;
 552                 bprm->p -= len;
 553 #ifdef CONFIG_MMU
 554                 if (bprm->p < bprm->argmin)
 555                         goto out;
 556 #endif
 557
 558                 while (len > 0) {
 559                         int offset, bytes_to_copy;
 560
 561                         if (fatal_signal_pending(current)) {
 562                                 ret = -ERESTARTNOHAND;
 563                                 goto out;
 564                         }
 565                         cond_resched();
 566
 567                         offset = pos % PAGE_SIZE;
 568                         if (offset == 0)
 569                                 offset = PAGE_SIZE;
 570
 571                         bytes_to_copy = offset;
 572                         if (bytes_to_copy > len)
 573                                 bytes_to_copy = len;
 574
 575                         offset -= bytes_to_copy;
 576                         pos -= bytes_to_copy;
 577                         str -= bytes_to_copy;
 578                         len -= bytes_to_copy;
 579
 580                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 581                                 struct page *page;
 582
 583                                 page = get_arg_page(bprm, pos, 1);
 584                                 if (!page) {
 585                                         ret = -E2BIG;
 586                                         goto out;
 587                                 }
 588
 589                                 if (kmapped_page) {
 590                                         flush_dcache_page(kmapped_page);
 591                                         kunmap_local(kaddr);
 592                                         put_arg_page(kmapped_page);
 593                                 }
 594                                 kmapped_page = page;
 595                                 kaddr = kmap_local_page(kmapped_page);
 596                                 kpos = pos & PAGE_MASK;
 597                                 flush_arg_page(bprm, kpos, kmapped_page);
 598                         }
 599                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 600                                 ret = -EFAULT;
 601                                 goto out;
 602                         }
 603                 }
 604         }
 605         ret = 0;
 606 out:
 607         if (kmapped_page) {
 608                 flush_dcache_page(kmapped_page);
 609                 kunmap_local(kaddr);
 610                 put_arg_page(kmapped_page);
 611         }
 612         return ret;
 613 }
 614
 615 /*
 616  * Copy and argument/environment string from the kernel to the processes stack.
 617  */
 618 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 619 {
 620         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 621         unsigned long pos = bprm->p;
 622
 623         if (len == 0)
 624                 return -EFAULT;
 625         if (!valid_arg_len(bprm, len))
 626                 return -E2BIG;
 627
 628         /* We're going to work our way backwards. */
 629         arg += len;
 630         bprm->p -= len;
 631         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 632                 return -E2BIG;
 633
 634         while (len > 0) {
 635                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 636                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 637                 struct page *page;
 638
 639                 pos -= bytes_to_copy;
 640                 arg -= bytes_to_copy;
 641                 len -= bytes_to_copy;
 642
 643                 page = get_arg_page(bprm, pos, 1);
 644                 if (!page)
 645                         return -E2BIG;
 646                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 647                 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
 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         VMA_ITERATOR(vmi, mm, new_start);
 692         struct vm_area_struct *next;
 693         struct mmu_gather tlb;
 694
 695         BUG_ON(new_start > new_end);
 696
 697         /*
 698          * ensure there are no vmas between where we want to go
 699          * and where we are
 700          */
 701         if (vma != vma_next(&vmi))
 702                 return -EFAULT;
 703
 704         /*
 705          * cover the whole range: [new_start, old_end)
 706          */
 707         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 708                 return -ENOMEM;
 709
 710         /*
 711          * move the page tables downwards, on failure we rely on
 712          * process cleanup to remove whatever mess we made.
 713          */
 714         if (length != move_page_tables(vma, old_start,
 715                                        vma, new_start, length, false))
 716                 return -ENOMEM;
 717
 718         lru_add_drain();
 719         tlb_gather_mmu(&tlb, mm);
 720         next = vma_next(&vmi);
 721         if (new_end > old_start) {
 722                 /*
 723                  * when the old and new regions overlap clear from new_end.
 724                  */
 725                 free_pgd_range(&tlb, new_end, old_end, new_end,
 726                         next ? next->vm_start : USER_PGTABLES_CEILING);
 727         } else {
 728                 /*
 729                  * otherwise, clean from old_start; this is done to not touch
 730                  * the address space in [new_end, old_start) some architectures
 731                  * have constraints on va-space that make this illegal (IA64) -
 732                  * for the others its just a little faster.
 733                  */
 734                 free_pgd_range(&tlb, old_start, old_end, new_end,
 735                         next ? next->vm_start : USER_PGTABLES_CEILING);
 736         }
 737         tlb_finish_mmu(&tlb);
 738
 739         /*
 740          * Shrink the vma to just the new range.  Always succeeds.
 741          */
 742         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 743
 744         return 0;
 745 }
 746
 747 /*
 748  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 749  * the stack is optionally relocated, and some extra space is added.
 750  */
 751 int setup_arg_pages(struct linux_binprm *bprm,
 752                     unsigned long stack_top,
 753                     int executable_stack)
 754 {
 755         unsigned long ret;
 756         unsigned long stack_shift;
 757         struct mm_struct *mm = current->mm;
 758         struct vm_area_struct *vma = bprm->vma;
 759         struct vm_area_struct *prev = NULL;
 760         unsigned long vm_flags;
 761         unsigned long stack_base;
 762         unsigned long stack_size;
 763         unsigned long stack_expand;
 764         unsigned long rlim_stack;
 765         struct mmu_gather tlb;
 766
 767 #ifdef CONFIG_STACK_GROWSUP
 768         /* Limit stack size */
 769         stack_base = bprm->rlim_stack.rlim_max;
 770
 771         stack_base = calc_max_stack_size(stack_base);
 772
 773         /* Add space for stack randomization. */
 774         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 775
 776         /* Make sure we didn't let the argument array grow too large. */
 777         if (vma->vm_end - vma->vm_start > stack_base)
 778                 return -ENOMEM;
 779
 780         stack_base = PAGE_ALIGN(stack_top - stack_base);
 781
 782         stack_shift = vma->vm_start - stack_base;
 783         mm->arg_start = bprm->p - stack_shift;
 784         bprm->p = vma->vm_end - stack_shift;
 785 #else
 786         stack_top = arch_align_stack(stack_top);
 787         stack_top = PAGE_ALIGN(stack_top);
 788
 789         if (unlikely(stack_top < mmap_min_addr) ||
 790             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 791                 return -ENOMEM;
 792
 793         stack_shift = vma->vm_end - stack_top;
 794
 795         bprm->p -= stack_shift;
 796         mm->arg_start = bprm->p;
 797 #endif
 798
 799         if (bprm->loader)
 800                 bprm->loader -= stack_shift;
 801         bprm->exec -= stack_shift;
 802
 803         if (mmap_write_lock_killable(mm))
 804                 return -EINTR;
 805
 806         vm_flags = VM_STACK_FLAGS;
 807
 808         /*
 809          * Adjust stack execute permissions; explicitly enable for
 810          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 811          * (arch default) otherwise.
 812          */
 813         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 814                 vm_flags |= VM_EXEC;
 815         else if (executable_stack == EXSTACK_DISABLE_X)
 816                 vm_flags &= ~VM_EXEC;
 817         vm_flags |= mm->def_flags;
 818         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 819
 820         tlb_gather_mmu(&tlb, mm);
 821         ret = mprotect_fixup(&tlb, vma, &prev, vma->vm_start, vma->vm_end,
 822                         vm_flags);
 823         tlb_finish_mmu(&tlb);
 824
 825         if (ret)
 826                 goto out_unlock;
 827         BUG_ON(prev != vma);
 828
 829         if (unlikely(vm_flags & VM_EXEC)) {
 830                 pr_warn_once("process '%pD4' started with executable stack\n",
 831                              bprm->file);
 832         }
 833
 834         /* Move stack pages down in memory. */
 835         if (stack_shift) {
 836                 ret = shift_arg_pages(vma, stack_shift);
 837                 if (ret)
 838                         goto out_unlock;
 839         }
 840
 841         /* mprotect_fixup is overkill to remove the temporary stack flags */
 842         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 843
 844         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 845         stack_size = vma->vm_end - vma->vm_start;
 846         /*
 847          * Align this down to a page boundary as expand_stack
 848          * will align it up.
 849          */
 850         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 851 #ifdef CONFIG_STACK_GROWSUP
 852         if (stack_size + stack_expand > rlim_stack)
 853                 stack_base = vma->vm_start + rlim_stack;
 854         else
 855                 stack_base = vma->vm_end + stack_expand;
 856 #else
 857         if (stack_size + stack_expand > rlim_stack)
 858                 stack_base = vma->vm_end - rlim_stack;
 859         else
 860                 stack_base = vma->vm_start - stack_expand;
 861 #endif
 862         current->mm->start_stack = bprm->p;
 863         ret = expand_stack_locked(vma, stack_base);
 864         if (ret)
 865                 ret = -EFAULT;
 866
 867 out_unlock:
 868         mmap_write_unlock(mm);
 869         return ret;
 870 }
 871 EXPORT_SYMBOL(setup_arg_pages);
 872
 873 #else
 874
 875 /*
 876  * Transfer the program arguments and environment from the holding pages
 877  * onto the stack. The provided stack pointer is adjusted accordingly.
 878  */
 879 int transfer_args_to_stack(struct linux_binprm *bprm,
 880                            unsigned long *sp_location)
 881 {
 882         unsigned long index, stop, sp;
 883         int ret = 0;
 884
 885         stop = bprm->p >> PAGE_SHIFT;
 886         sp = *sp_location;
 887
 888         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 889                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 890                 char *src = kmap_local_page(bprm->page[index]) + offset;
 891                 sp -= PAGE_SIZE - offset;
 892                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 893                         ret = -EFAULT;
 894                 kunmap_local(src);
 895                 if (ret)
 896                         goto out;
 897         }
 898
 899         bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
 900         *sp_location = sp;
 901
 902 out:
 903         return ret;
 904 }
 905 EXPORT_SYMBOL(transfer_args_to_stack);
 906
 907 #endif /* CONFIG_MMU */
 908
 909 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 910 {
 911         struct file *file;
 912         int err;
 913         struct open_flags open_exec_flags = {
 914                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 915                 .acc_mode = MAY_EXEC,
 916                 .intent = LOOKUP_OPEN,
 917                 .lookup_flags = LOOKUP_FOLLOW,
 918         };
 919
 920         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 921                 return ERR_PTR(-EINVAL);
 922         if (flags & AT_SYMLINK_NOFOLLOW)
 923                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 924         if (flags & AT_EMPTY_PATH)
 925                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 926
 927         file = do_filp_open(fd, name, &open_exec_flags);
 928         if (IS_ERR(file))
 929                 goto out;
 930
 931         /*
 932          * may_open() has already checked for this, so it should be
 933          * impossible to trip now. But we need to be extra cautious
 934          * and check again at the very end too.
 935          */
 936         err = -EACCES;
 937         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 938                          path_noexec(&file->f_path)))
 939                 goto exit;
 940
 941         err = deny_write_access(file);
 942         if (err)
 943                 goto exit;
 944
 945         if (name->name[0] != '\0')
 946                 fsnotify_open(file);
 947
 948 out:
 949         return file;
 950
 951 exit:
 952         fput(file);
 953         return ERR_PTR(err);
 954 }
 955
 956 struct file *open_exec(const char *name)
 957 {
 958         struct filename *filename = getname_kernel(name);
 959         struct file *f = ERR_CAST(filename);
 960
 961         if (!IS_ERR(filename)) {
 962                 f = do_open_execat(AT_FDCWD, filename, 0);
 963                 putname(filename);
 964         }
 965         return f;
 966 }
 967 EXPORT_SYMBOL(open_exec);
 968
 969 #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
 970 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 971 {
 972         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 973         if (res > 0)
 974                 flush_icache_user_range(addr, addr + len);
 975         return res;
 976 }
 977 EXPORT_SYMBOL(read_code);
 978 #endif
 979
 980 /*
 981  * Maps the mm_struct mm into the current task struct.
 982  * On success, this function returns with exec_update_lock
 983  * held for writing.
 984  */
 985 static int exec_mmap(struct mm_struct *mm)
 986 {
 987         struct task_struct *tsk;
 988         struct mm_struct *old_mm, *active_mm;
 989         int ret;
 990
 991         /* Notify parent that we're no longer interested in the old VM */
 992         tsk = current;
 993         old_mm = current->mm;
 994         exec_mm_release(tsk, old_mm);
 995         if (old_mm)
 996                 sync_mm_rss(old_mm);
 997
 998         ret = down_write_killable(&tsk->signal->exec_update_lock);
 999         if (ret)
1000                 return ret;
1001
1002         if (old_mm) {
1003                 /*
1004                  * If there is a pending fatal signal perhaps a signal
1005                  * whose default action is to create a coredump get
1006                  * out and die instead of going through with the exec.
1007                  */
1008                 ret = mmap_read_lock_killable(old_mm);
1009                 if (ret) {
1010                         up_write(&tsk->signal->exec_update_lock);
1011                         return ret;
1012                 }
1013         }
1014
1015         task_lock(tsk);
1016         membarrier_exec_mmap(mm);
1017
1018         local_irq_disable();
1019         active_mm = tsk->active_mm;
1020         tsk->active_mm = mm;
1021         tsk->mm = mm;
1022         /*
1023          * This prevents preemption while active_mm is being loaded and
1024          * it and mm are being updated, which could cause problems for
1025          * lazy tlb mm refcounting when these are updated by context
1026          * switches. Not all architectures can handle irqs off over
1027          * activate_mm yet.
1028          */
1029         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1030                 local_irq_enable();
1031         activate_mm(active_mm, mm);
1032         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1033                 local_irq_enable();
1034         lru_gen_add_mm(mm);
1035         task_unlock(tsk);
1036         lru_gen_use_mm(mm);
1037         if (old_mm) {
1038                 mmap_read_unlock(old_mm);
1039                 BUG_ON(active_mm != old_mm);
1040                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1041                 mm_update_next_owner(old_mm);
1042                 mmput(old_mm);
1043                 return 0;
1044         }
1045         mmdrop(active_mm);
1046         return 0;
1047 }
1048
1049 static int de_thread(struct task_struct *tsk)
1050 {
1051         struct signal_struct *sig = tsk->signal;
1052         struct sighand_struct *oldsighand = tsk->sighand;
1053         spinlock_t *lock = &oldsighand->siglock;
1054
1055         if (thread_group_empty(tsk))
1056                 goto no_thread_group;
1057
1058         /*
1059          * Kill all other threads in the thread group.
1060          */
1061         spin_lock_irq(lock);
1062         if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1063                 /*
1064                  * Another group action in progress, just
1065                  * return so that the signal is processed.
1066                  */
1067                 spin_unlock_irq(lock);
1068                 return -EAGAIN;
1069         }
1070
1071         sig->group_exec_task = tsk;
1072         sig->notify_count = zap_other_threads(tsk);
1073         if (!thread_group_leader(tsk))
1074                 sig->notify_count--;
1075
1076         while (sig->notify_count) {
1077                 __set_current_state(TASK_KILLABLE);
1078                 spin_unlock_irq(lock);
1079                 schedule();
1080                 if (__fatal_signal_pending(tsk))
1081                         goto killed;
1082                 spin_lock_irq(lock);
1083         }
1084         spin_unlock_irq(lock);
1085
1086         /*
1087          * At this point all other threads have exited, all we have to
1088          * do is to wait for the thread group leader to become inactive,
1089          * and to assume its PID:
1090          */
1091         if (!thread_group_leader(tsk)) {
1092                 struct task_struct *leader = tsk->group_leader;
1093
1094                 for (;;) {
1095                         cgroup_threadgroup_change_begin(tsk);
1096                         write_lock_irq(&tasklist_lock);
1097                         /*
1098                          * Do this under tasklist_lock to ensure that
1099                          * exit_notify() can't miss ->group_exec_task
1100                          */
1101                         sig->notify_count = -1;
1102                         if (likely(leader->exit_state))
1103                                 break;
1104                         __set_current_state(TASK_KILLABLE);
1105                         write_unlock_irq(&tasklist_lock);
1106                         cgroup_threadgroup_change_end(tsk);
1107                         schedule();
1108                         if (__fatal_signal_pending(tsk))
1109                                 goto killed;
1110                 }
1111
1112                 /*
1113                  * The only record we have of the real-time age of a
1114                  * process, regardless of execs it's done, is start_time.
1115                  * All the past CPU time is accumulated in signal_struct
1116                  * from sister threads now dead.  But in this non-leader
1117                  * exec, nothing survives from the original leader thread,
1118                  * whose birth marks the true age of this process now.
1119                  * When we take on its identity by switching to its PID, we
1120                  * also take its birthdate (always earlier than our own).
1121                  */
1122                 tsk->start_time = leader->start_time;
1123                 tsk->start_boottime = leader->start_boottime;
1124
1125                 BUG_ON(!same_thread_group(leader, tsk));
1126                 /*
1127                  * An exec() starts a new thread group with the
1128                  * TGID of the previous thread group. Rehash the
1129                  * two threads with a switched PID, and release
1130                  * the former thread group leader:
1131                  */
1132
1133                 /* Become a process group leader with the old leader's pid.
1134                  * The old leader becomes a thread of the this thread group.
1135                  */
1136                 exchange_tids(tsk, leader);
1137                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1138                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1139                 transfer_pid(leader, tsk, PIDTYPE_SID);
1140
1141                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1142                 list_replace_init(&leader->sibling, &tsk->sibling);
1143
1144                 tsk->group_leader = tsk;
1145                 leader->group_leader = tsk;
1146
1147                 tsk->exit_signal = SIGCHLD;
1148                 leader->exit_signal = -1;
1149
1150                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1151                 leader->exit_state = EXIT_DEAD;
1152
1153                 /*
1154                  * We are going to release_task()->ptrace_unlink() silently,
1155                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1156                  * the tracer won't block again waiting for this thread.
1157                  */
1158                 if (unlikely(leader->ptrace))
1159                         __wake_up_parent(leader, leader->parent);
1160                 write_unlock_irq(&tasklist_lock);
1161                 cgroup_threadgroup_change_end(tsk);
1162
1163                 release_task(leader);
1164         }
1165
1166         sig->group_exec_task = NULL;
1167         sig->notify_count = 0;
1168
1169 no_thread_group:
1170         /* we have changed execution domain */
1171         tsk->exit_signal = SIGCHLD;
1172
1173         BUG_ON(!thread_group_leader(tsk));
1174         return 0;
1175
1176 killed:
1177         /* protects against exit_notify() and __exit_signal() */
1178         read_lock(&tasklist_lock);
1179         sig->group_exec_task = NULL;
1180         sig->notify_count = 0;
1181         read_unlock(&tasklist_lock);
1182         return -EAGAIN;
1183 }
1184
1185
1186 /*
1187  * This function makes sure the current process has its own signal table,
1188  * so that flush_signal_handlers can later reset the handlers without
1189  * disturbing other processes.  (Other processes might share the signal
1190  * table via the CLONE_SIGHAND option to clone().)
1191  */
1192 static int unshare_sighand(struct task_struct *me)
1193 {
1194         struct sighand_struct *oldsighand = me->sighand;
1195
1196         if (refcount_read(&oldsighand->count) != 1) {
1197                 struct sighand_struct *newsighand;
1198                 /*
1199                  * This ->sighand is shared with the CLONE_SIGHAND
1200                  * but not CLONE_THREAD task, switch to the new one.
1201                  */
1202                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1203                 if (!newsighand)
1204                         return -ENOMEM;
1205
1206                 refcount_set(&newsighand->count, 1);
1207
1208                 write_lock_irq(&tasklist_lock);
1209                 spin_lock(&oldsighand->siglock);
1210                 memcpy(newsighand->action, oldsighand->action,
1211                        sizeof(newsighand->action));
1212                 rcu_assign_pointer(me->sighand, newsighand);
1213                 spin_unlock(&oldsighand->siglock);
1214                 write_unlock_irq(&tasklist_lock);
1215
1216                 __cleanup_sighand(oldsighand);
1217         }
1218         return 0;
1219 }
1220
1221 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1222 {
1223         task_lock(tsk);
1224         /* Always NUL terminated and zero-padded */
1225         strscpy_pad(buf, tsk->comm, buf_size);
1226         task_unlock(tsk);
1227         return buf;
1228 }
1229 EXPORT_SYMBOL_GPL(__get_task_comm);
1230
1231 /*
1232  * These functions flushes out all traces of the currently running executable
1233  * so that a new one can be started
1234  */
1235
1236 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1237 {
1238         task_lock(tsk);
1239         trace_task_rename(tsk, buf);
1240         strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1241         task_unlock(tsk);
1242         perf_event_comm(tsk, exec);
1243 }
1244
1245 /*
1246  * Calling this is the point of no return. None of the failures will be
1247  * seen by userspace since either the process is already taking a fatal
1248  * signal (via de_thread() or coredump), or will have SEGV raised
1249  * (after exec_mmap()) by search_binary_handler (see below).
1250  */
1251 int begin_new_exec(struct linux_binprm * bprm)
1252 {
1253         struct task_struct *me = current;
1254         int retval;
1255
1256         /* Once we are committed compute the creds */
1257         retval = bprm_creds_from_file(bprm);
1258         if (retval)
1259                 return retval;
1260
1261         /*
1262          * Ensure all future errors are fatal.
1263          */
1264         bprm->point_of_no_return = true;
1265
1266         /*
1267          * Make this the only thread in the thread group.
1268          */
1269         retval = de_thread(me);
1270         if (retval)
1271                 goto out;
1272
1273         /*
1274          * Cancel any io_uring activity across execve
1275          */
1276         io_uring_task_cancel();
1277
1278         /* Ensure the files table is not shared. */
1279         retval = unshare_files();
1280         if (retval)
1281                 goto out;
1282
1283         /*
1284          * Must be called _before_ exec_mmap() as bprm->mm is
1285          * not visible until then. This also enables the update
1286          * to be lockless.
1287          */
1288         retval = set_mm_exe_file(bprm->mm, bprm->file);
1289         if (retval)
1290                 goto out;
1291
1292         /* If the binary is not readable then enforce mm->dumpable=0 */
1293         would_dump(bprm, bprm->file);
1294         if (bprm->have_execfd)
1295                 would_dump(bprm, bprm->executable);
1296
1297         /*
1298          * Release all of the old mmap stuff
1299          */
1300         acct_arg_size(bprm, 0);
1301         retval = exec_mmap(bprm->mm);
1302         if (retval)
1303                 goto out;
1304
1305         bprm->mm = NULL;
1306
1307 #ifdef CONFIG_POSIX_TIMERS
1308         spin_lock_irq(&me->sighand->siglock);
1309         posix_cpu_timers_exit(me);
1310         spin_unlock_irq(&me->sighand->siglock);
1311         exit_itimers(me);
1312         flush_itimer_signals();
1313 #endif
1314
1315         /*
1316          * Make the signal table private.
1317          */
1318         retval = unshare_sighand(me);
1319         if (retval)
1320                 goto out_unlock;
1321
1322         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1323                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1324         flush_thread();
1325         me->personality &= ~bprm->per_clear;
1326
1327         clear_syscall_work_syscall_user_dispatch(me);
1328
1329         /*
1330          * We have to apply CLOEXEC before we change whether the process is
1331          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1332          * trying to access the should-be-closed file descriptors of a process
1333          * undergoing exec(2).
1334          */
1335         do_close_on_exec(me->files);
1336
1337         if (bprm->secureexec) {
1338                 /* Make sure parent cannot signal privileged process. */
1339                 me->pdeath_signal = 0;
1340
1341                 /*
1342                  * For secureexec, reset the stack limit to sane default to
1343                  * avoid bad behavior from the prior rlimits. This has to
1344                  * happen before arch_pick_mmap_layout(), which examines
1345                  * RLIMIT_STACK, but after the point of no return to avoid
1346                  * needing to clean up the change on failure.
1347                  */
1348                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1349                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1350         }
1351
1352         me->sas_ss_sp = me->sas_ss_size = 0;
1353
1354         /*
1355          * Figure out dumpability. Note that this checking only of current
1356          * is wrong, but userspace depends on it. This should be testing
1357          * bprm->secureexec instead.
1358          */
1359         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1360             !(uid_eq(current_euid(), current_uid()) &&
1361               gid_eq(current_egid(), current_gid())))
1362                 set_dumpable(current->mm, suid_dumpable);
1363         else
1364                 set_dumpable(current->mm, SUID_DUMP_USER);
1365
1366         perf_event_exec();
1367         __set_task_comm(me, kbasename(bprm->filename), true);
1368
1369         /* An exec changes our domain. We are no longer part of the thread
1370            group */
1371         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1372         flush_signal_handlers(me, 0);
1373
1374         retval = set_cred_ucounts(bprm->cred);
1375         if (retval < 0)
1376                 goto out_unlock;
1377
1378         /*
1379          * install the new credentials for this executable
1380          */
1381         security_bprm_committing_creds(bprm);
1382
1383         commit_creds(bprm->cred);
1384         bprm->cred = NULL;
1385
1386         /*
1387          * Disable monitoring for regular users
1388          * when executing setuid binaries. Must
1389          * wait until new credentials are committed
1390          * by commit_creds() above
1391          */
1392         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1393                 perf_event_exit_task(me);
1394         /*
1395          * cred_guard_mutex must be held at least to this point to prevent
1396          * ptrace_attach() from altering our determination of the task's
1397          * credentials; any time after this it may be unlocked.
1398          */
1399         security_bprm_committed_creds(bprm);
1400
1401         /* Pass the opened binary to the interpreter. */
1402         if (bprm->have_execfd) {
1403                 retval = get_unused_fd_flags(0);
1404                 if (retval < 0)
1405                         goto out_unlock;
1406                 fd_install(retval, bprm->executable);
1407                 bprm->executable = NULL;
1408                 bprm->execfd = retval;
1409         }
1410         return 0;
1411
1412 out_unlock:
1413         up_write(&me->signal->exec_update_lock);
1414         if (!bprm->cred)
1415                 mutex_unlock(&me->signal->cred_guard_mutex);
1416
1417 out:
1418         return retval;
1419 }
1420 EXPORT_SYMBOL(begin_new_exec);
1421
1422 void would_dump(struct linux_binprm *bprm, struct file *file)
1423 {
1424         struct inode *inode = file_inode(file);
1425         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1426         if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1427                 struct user_namespace *old, *user_ns;
1428                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1429
1430                 /* Ensure mm->user_ns contains the executable */
1431                 user_ns = old = bprm->mm->user_ns;
1432                 while ((user_ns != &init_user_ns) &&
1433                        !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1434                         user_ns = user_ns->parent;
1435
1436                 if (old != user_ns) {
1437                         bprm->mm->user_ns = get_user_ns(user_ns);
1438                         put_user_ns(old);
1439                 }
1440         }
1441 }
1442 EXPORT_SYMBOL(would_dump);
1443
1444 void setup_new_exec(struct linux_binprm * bprm)
1445 {
1446         /* Setup things that can depend upon the personality */
1447         struct task_struct *me = current;
1448
1449         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1450
1451         arch_setup_new_exec();
1452
1453         /* Set the new mm task size. We have to do that late because it may
1454          * depend on TIF_32BIT which is only updated in flush_thread() on
1455          * some architectures like powerpc
1456          */
1457         me->mm->task_size = TASK_SIZE;
1458         up_write(&me->signal->exec_update_lock);
1459         mutex_unlock(&me->signal->cred_guard_mutex);
1460 }
1461 EXPORT_SYMBOL(setup_new_exec);
1462
1463 /* Runs immediately before start_thread() takes over. */
1464 void finalize_exec(struct linux_binprm *bprm)
1465 {
1466         /* Store any stack rlimit changes before starting thread. */
1467         task_lock(current->group_leader);
1468         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1469         task_unlock(current->group_leader);
1470 }
1471 EXPORT_SYMBOL(finalize_exec);
1472
1473 /*
1474  * Prepare credentials and lock ->cred_guard_mutex.
1475  * setup_new_exec() commits the new creds and drops the lock.
1476  * Or, if exec fails before, free_bprm() should release ->cred
1477  * and unlock.
1478  */
1479 static int prepare_bprm_creds(struct linux_binprm *bprm)
1480 {
1481         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1482                 return -ERESTARTNOINTR;
1483
1484         bprm->cred = prepare_exec_creds();
1485         if (likely(bprm->cred))
1486                 return 0;
1487
1488         mutex_unlock(&current->signal->cred_guard_mutex);
1489         return -ENOMEM;
1490 }
1491
1492 static void free_bprm(struct linux_binprm *bprm)
1493 {
1494         if (bprm->mm) {
1495                 acct_arg_size(bprm, 0);
1496                 mmput(bprm->mm);
1497         }
1498         free_arg_pages(bprm);
1499         if (bprm->cred) {
1500                 mutex_unlock(&current->signal->cred_guard_mutex);
1501                 abort_creds(bprm->cred);
1502         }
1503         if (bprm->file) {
1504                 allow_write_access(bprm->file);
1505                 fput(bprm->file);
1506         }
1507         if (bprm->executable)
1508                 fput(bprm->executable);
1509         /* If a binfmt changed the interp, free it. */
1510         if (bprm->interp != bprm->filename)
1511                 kfree(bprm->interp);
1512         kfree(bprm->fdpath);
1513         kfree(bprm);
1514 }
1515
1516 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1517 {
1518         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1519         int retval = -ENOMEM;
1520         if (!bprm)
1521                 goto out;
1522
1523         if (fd == AT_FDCWD || filename->name[0] == '/') {
1524                 bprm->filename = filename->name;
1525         } else {
1526                 if (filename->name[0] == '\0')
1527                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1528                 else
1529                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1530                                                   fd, filename->name);
1531                 if (!bprm->fdpath)
1532                         goto out_free;
1533
1534                 bprm->filename = bprm->fdpath;
1535         }
1536         bprm->interp = bprm->filename;
1537
1538         retval = bprm_mm_init(bprm);
1539         if (retval)
1540                 goto out_free;
1541         return bprm;
1542
1543 out_free:
1544         free_bprm(bprm);
1545 out:
1546         return ERR_PTR(retval);
1547 }
1548
1549 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1550 {
1551         /* If a binfmt changed the interp, free it first. */
1552         if (bprm->interp != bprm->filename)
1553                 kfree(bprm->interp);
1554         bprm->interp = kstrdup(interp, GFP_KERNEL);
1555         if (!bprm->interp)
1556                 return -ENOMEM;
1557         return 0;
1558 }
1559 EXPORT_SYMBOL(bprm_change_interp);
1560
1561 /*
1562  * determine how safe it is to execute the proposed program
1563  * - the caller must hold ->cred_guard_mutex to protect against
1564  *   PTRACE_ATTACH or seccomp thread-sync
1565  */
1566 static void check_unsafe_exec(struct linux_binprm *bprm)
1567 {
1568         struct task_struct *p = current, *t;
1569         unsigned n_fs;
1570
1571         if (p->ptrace)
1572                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1573
1574         /*
1575          * This isn't strictly necessary, but it makes it harder for LSMs to
1576          * mess up.
1577          */
1578         if (task_no_new_privs(current))
1579                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1580
1581         t = p;
1582         n_fs = 1;
1583         spin_lock(&p->fs->lock);
1584         rcu_read_lock();
1585         while_each_thread(p, t) {
1586                 if (t->fs == p->fs)
1587                         n_fs++;
1588         }
1589         rcu_read_unlock();
1590
1591         if (p->fs->users > n_fs)
1592                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1593         else
1594                 p->fs->in_exec = 1;
1595         spin_unlock(&p->fs->lock);
1596 }
1597
1598 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1599 {
1600         /* Handle suid and sgid on files */
1601         struct user_namespace *mnt_userns;
1602         struct inode *inode = file_inode(file);
1603         unsigned int mode;
1604         kuid_t uid;
1605         kgid_t gid;
1606
1607         if (!mnt_may_suid(file->f_path.mnt))
1608                 return;
1609
1610         if (task_no_new_privs(current))
1611                 return;
1612
1613         mode = READ_ONCE(inode->i_mode);
1614         if (!(mode & (S_ISUID|S_ISGID)))
1615                 return;
1616
1617         mnt_userns = file_mnt_user_ns(file);
1618
1619         /* Be careful if suid/sgid is set */
1620         inode_lock(inode);
1621
1622         /* reload atomically mode/uid/gid now that lock held */
1623         mode = inode->i_mode;
1624         uid = i_uid_into_mnt(mnt_userns, inode);
1625         gid = i_gid_into_mnt(mnt_userns, inode);
1626         inode_unlock(inode);
1627
1628         /* We ignore suid/sgid if there are no mappings for them in the ns */
1629         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1630                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1631                 return;
1632
1633         if (mode & S_ISUID) {
1634                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1635                 bprm->cred->euid = uid;
1636         }
1637
1638         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1639                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1640                 bprm->cred->egid = gid;
1641         }
1642 }
1643
1644 /*
1645  * Compute brpm->cred based upon the final binary.
1646  */
1647 static int bprm_creds_from_file(struct linux_binprm *bprm)
1648 {
1649         /* Compute creds based on which file? */
1650         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1651
1652         bprm_fill_uid(bprm, file);
1653         return security_bprm_creds_from_file(bprm, file);
1654 }
1655
1656 /*
1657  * Fill the binprm structure from the inode.
1658  * Read the first BINPRM_BUF_SIZE bytes
1659  *
1660  * This may be called multiple times for binary chains (scripts for example).
1661  */
1662 static int prepare_binprm(struct linux_binprm *bprm)
1663 {
1664         loff_t pos = 0;
1665
1666         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1667         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1668 }
1669
1670 /*
1671  * Arguments are '\0' separated strings found at the location bprm->p
1672  * points to; chop off the first by relocating brpm->p to right after
1673  * the first '\0' encountered.
1674  */
1675 int remove_arg_zero(struct linux_binprm *bprm)
1676 {
1677         int ret = 0;
1678         unsigned long offset;
1679         char *kaddr;
1680         struct page *page;
1681
1682         if (!bprm->argc)
1683                 return 0;
1684
1685         do {
1686                 offset = bprm->p & ~PAGE_MASK;
1687                 page = get_arg_page(bprm, bprm->p, 0);
1688                 if (!page) {
1689                         ret = -EFAULT;
1690                         goto out;
1691                 }
1692                 kaddr = kmap_local_page(page);
1693
1694                 for (; offset < PAGE_SIZE && kaddr[offset];
1695                                 offset++, bprm->p++)
1696                         ;
1697
1698                 kunmap_local(kaddr);
1699                 put_arg_page(page);
1700         } while (offset == PAGE_SIZE);
1701
1702         bprm->p++;
1703         bprm->argc--;
1704         ret = 0;
1705
1706 out:
1707         return ret;
1708 }
1709 EXPORT_SYMBOL(remove_arg_zero);
1710
1711 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1712 /*
1713  * cycle the list of binary formats handler, until one recognizes the image
1714  */
1715 static int search_binary_handler(struct linux_binprm *bprm)
1716 {
1717         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1718         struct linux_binfmt *fmt;
1719         int retval;
1720
1721         retval = prepare_binprm(bprm);
1722         if (retval < 0)
1723                 return retval;
1724
1725         retval = security_bprm_check(bprm);
1726         if (retval)
1727                 return retval;
1728
1729         retval = -ENOENT;
1730  retry:
1731         read_lock(&binfmt_lock);
1732         list_for_each_entry(fmt, &formats, lh) {
1733                 if (!try_module_get(fmt->module))
1734                         continue;
1735                 read_unlock(&binfmt_lock);
1736
1737                 retval = fmt->load_binary(bprm);
1738
1739                 read_lock(&binfmt_lock);
1740                 put_binfmt(fmt);
1741                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1742                         read_unlock(&binfmt_lock);
1743                         return retval;
1744                 }
1745         }
1746         read_unlock(&binfmt_lock);
1747
1748         if (need_retry) {
1749                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1750                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1751                         return retval;
1752                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1753                         return retval;
1754                 need_retry = false;
1755                 goto retry;
1756         }
1757
1758         return retval;
1759 }
1760
1761 static int exec_binprm(struct linux_binprm *bprm)
1762 {
1763         pid_t old_pid, old_vpid;
1764         int ret, depth;
1765
1766         /* Need to fetch pid before load_binary changes it */
1767         old_pid = current->pid;
1768         rcu_read_lock();
1769         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1770         rcu_read_unlock();
1771
1772         /* This allows 4 levels of binfmt rewrites before failing hard. */
1773         for (depth = 0;; depth++) {
1774                 struct file *exec;
1775                 if (depth > 5)
1776                         return -ELOOP;
1777
1778                 ret = search_binary_handler(bprm);
1779                 if (ret < 0)
1780                         return ret;
1781                 if (!bprm->interpreter)
1782                         break;
1783
1784                 exec = bprm->file;
1785                 bprm->file = bprm->interpreter;
1786                 bprm->interpreter = NULL;
1787
1788                 allow_write_access(exec);
1789                 if (unlikely(bprm->have_execfd)) {
1790                         if (bprm->executable) {
1791                                 fput(exec);
1792                                 return -ENOEXEC;
1793                         }
1794                         bprm->executable = exec;
1795                 } else
1796                         fput(exec);
1797         }
1798
1799         audit_bprm(bprm);
1800         trace_sched_process_exec(current, old_pid, bprm);
1801         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1802         proc_exec_connector(current);
1803         return 0;
1804 }
1805
1806 /*
1807  * sys_execve() executes a new program.
1808  */
1809 static int bprm_execve(struct linux_binprm *bprm,
1810                        int fd, struct filename *filename, int flags)
1811 {
1812         struct file *file;
1813         int retval;
1814
1815         retval = prepare_bprm_creds(bprm);
1816         if (retval)
1817                 return retval;
1818
1819         check_unsafe_exec(bprm);
1820         current->in_execve = 1;
1821
1822         file = do_open_execat(fd, filename, flags);
1823         retval = PTR_ERR(file);
1824         if (IS_ERR(file))
1825                 goto out_unmark;
1826
1827         sched_exec();
1828
1829         bprm->file = file;
1830         /*
1831          * Record that a name derived from an O_CLOEXEC fd will be
1832          * inaccessible after exec.  This allows the code in exec to
1833          * choose to fail when the executable is not mmaped into the
1834          * interpreter and an open file descriptor is not passed to
1835          * the interpreter.  This makes for a better user experience
1836          * than having the interpreter start and then immediately fail
1837          * when it finds the executable is inaccessible.
1838          */
1839         if (bprm->fdpath && get_close_on_exec(fd))
1840                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1841
1842         /* Set the unchanging part of bprm->cred */
1843         retval = security_bprm_creds_for_exec(bprm);
1844         if (retval)
1845                 goto out;
1846
1847         retval = exec_binprm(bprm);
1848         if (retval < 0)
1849                 goto out;
1850
1851         /* execve succeeded */
1852         current->fs->in_exec = 0;
1853         current->in_execve = 0;
1854         rseq_execve(current);
1855         acct_update_integrals(current);
1856         task_numa_free(current, false);
1857         return retval;
1858
1859 out:
1860         /*
1861          * If past the point of no return ensure the code never
1862          * returns to the userspace process.  Use an existing fatal
1863          * signal if present otherwise terminate the process with
1864          * SIGSEGV.
1865          */
1866         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1867                 force_fatal_sig(SIGSEGV);
1868
1869 out_unmark:
1870         current->fs->in_exec = 0;
1871         current->in_execve = 0;
1872
1873         return retval;
1874 }
1875
1876 static int do_execveat_common(int fd, struct filename *filename,
1877                               struct user_arg_ptr argv,
1878                               struct user_arg_ptr envp,
1879                               int flags)
1880 {
1881         struct linux_binprm *bprm;
1882         int retval;
1883
1884         if (IS_ERR(filename))
1885                 return PTR_ERR(filename);
1886
1887         /*
1888          * We move the actual failure in case of RLIMIT_NPROC excess from
1889          * set*uid() to execve() because too many poorly written programs
1890          * don't check setuid() return code.  Here we additionally recheck
1891          * whether NPROC limit is still exceeded.
1892          */
1893         if ((current->flags & PF_NPROC_EXCEEDED) &&
1894             is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1895                 retval = -EAGAIN;
1896                 goto out_ret;
1897         }
1898
1899         /* We're below the limit (still or again), so we don't want to make
1900          * further execve() calls fail. */
1901         current->flags &= ~PF_NPROC_EXCEEDED;
1902
1903         bprm = alloc_bprm(fd, filename);
1904         if (IS_ERR(bprm)) {
1905                 retval = PTR_ERR(bprm);
1906                 goto out_ret;
1907         }
1908
1909         retval = count(argv, MAX_ARG_STRINGS);
1910         if (retval == 0)
1911                 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1912                              current->comm, bprm->filename);
1913         if (retval < 0)
1914                 goto out_free;
1915         bprm->argc = retval;
1916
1917         retval = count(envp, MAX_ARG_STRINGS);
1918         if (retval < 0)
1919                 goto out_free;
1920         bprm->envc = retval;
1921
1922         retval = bprm_stack_limits(bprm);
1923         if (retval < 0)
1924                 goto out_free;
1925
1926         retval = copy_string_kernel(bprm->filename, bprm);
1927         if (retval < 0)
1928                 goto out_free;
1929         bprm->exec = bprm->p;
1930
1931         retval = copy_strings(bprm->envc, envp, bprm);
1932         if (retval < 0)
1933                 goto out_free;
1934
1935         retval = copy_strings(bprm->argc, argv, bprm);
1936         if (retval < 0)
1937                 goto out_free;
1938
1939         /*
1940          * When argv is empty, add an empty string ("") as argv[0] to
1941          * ensure confused userspace programs that start processing
1942          * from argv[1] won't end up walking envp. See also
1943          * bprm_stack_limits().
1944          */
1945         if (bprm->argc == 0) {
1946                 retval = copy_string_kernel("", bprm);
1947                 if (retval < 0)
1948                         goto out_free;
1949                 bprm->argc = 1;
1950         }
1951
1952         retval = bprm_execve(bprm, fd, filename, flags);
1953 out_free:
1954         free_bprm(bprm);
1955
1956 out_ret:
1957         putname(filename);
1958         return retval;
1959 }
1960
1961 int kernel_execve(const char *kernel_filename,
1962                   const char *const *argv, const char *const *envp)
1963 {
1964         struct filename *filename;
1965         struct linux_binprm *bprm;
1966         int fd = AT_FDCWD;
1967         int retval;
1968
1969         /* It is non-sense for kernel threads to call execve */
1970         if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1971                 return -EINVAL;
1972
1973         filename = getname_kernel(kernel_filename);
1974         if (IS_ERR(filename))
1975                 return PTR_ERR(filename);
1976
1977         bprm = alloc_bprm(fd, filename);
1978         if (IS_ERR(bprm)) {
1979                 retval = PTR_ERR(bprm);
1980                 goto out_ret;
1981         }
1982
1983         retval = count_strings_kernel(argv);
1984         if (WARN_ON_ONCE(retval == 0))
1985                 retval = -EINVAL;
1986         if (retval < 0)
1987                 goto out_free;
1988         bprm->argc = retval;
1989
1990         retval = count_strings_kernel(envp);
1991         if (retval < 0)
1992                 goto out_free;
1993         bprm->envc = retval;
1994
1995         retval = bprm_stack_limits(bprm);
1996         if (retval < 0)
1997                 goto out_free;
1998
1999         retval = copy_string_kernel(bprm->filename, bprm);
2000         if (retval < 0)
2001                 goto out_free;
2002         bprm->exec = bprm->p;
2003
2004         retval = copy_strings_kernel(bprm->envc, envp, bprm);
2005         if (retval < 0)
2006                 goto out_free;
2007
2008         retval = copy_strings_kernel(bprm->argc, argv, bprm);
2009         if (retval < 0)
2010                 goto out_free;
2011
2012         retval = bprm_execve(bprm, fd, filename, 0);
2013 out_free:
2014         free_bprm(bprm);
2015 out_ret:
2016         putname(filename);
2017         return retval;
2018 }
2019
2020 static int do_execve(struct filename *filename,
2021         const char __user *const __user *__argv,
2022         const char __user *const __user *__envp)
2023 {
2024         struct user_arg_ptr argv = { .ptr.native = __argv };
2025         struct user_arg_ptr envp = { .ptr.native = __envp };
2026         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2027 }
2028
2029 static int do_execveat(int fd, struct filename *filename,
2030                 const char __user *const __user *__argv,
2031                 const char __user *const __user *__envp,
2032                 int flags)
2033 {
2034         struct user_arg_ptr argv = { .ptr.native = __argv };
2035         struct user_arg_ptr envp = { .ptr.native = __envp };
2036
2037         return do_execveat_common(fd, filename, argv, envp, flags);
2038 }
2039
2040 #ifdef CONFIG_COMPAT
2041 static int compat_do_execve(struct filename *filename,
2042         const compat_uptr_t __user *__argv,
2043         const compat_uptr_t __user *__envp)
2044 {
2045         struct user_arg_ptr argv = {
2046                 .is_compat = true,
2047                 .ptr.compat = __argv,
2048         };
2049         struct user_arg_ptr envp = {
2050                 .is_compat = true,
2051                 .ptr.compat = __envp,
2052         };
2053         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2054 }
2055
2056 static int compat_do_execveat(int fd, struct filename *filename,
2057                               const compat_uptr_t __user *__argv,
2058                               const compat_uptr_t __user *__envp,
2059                               int flags)
2060 {
2061         struct user_arg_ptr argv = {
2062                 .is_compat = true,
2063                 .ptr.compat = __argv,
2064         };
2065         struct user_arg_ptr envp = {
2066                 .is_compat = true,
2067                 .ptr.compat = __envp,
2068         };
2069         return do_execveat_common(fd, filename, argv, envp, flags);
2070 }
2071 #endif
2072
2073 void set_binfmt(struct linux_binfmt *new)
2074 {
2075         struct mm_struct *mm = current->mm;
2076
2077         if (mm->binfmt)
2078                 module_put(mm->binfmt->module);
2079
2080         mm->binfmt = new;
2081         if (new)
2082                 __module_get(new->module);
2083 }
2084 EXPORT_SYMBOL(set_binfmt);
2085
2086 /*
2087  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2088  */
2089 void set_dumpable(struct mm_struct *mm, int value)
2090 {
2091         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2092                 return;
2093
2094         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2095 }
2096
2097 SYSCALL_DEFINE3(execve,
2098                 const char __user *, filename,
2099                 const char __user *const __user *, argv,
2100                 const char __user *const __user *, envp)
2101 {
2102         return do_execve(getname(filename), argv, envp);
2103 }
2104
2105 SYSCALL_DEFINE5(execveat,
2106                 int, fd, const char __user *, filename,
2107                 const char __user *const __user *, argv,
2108                 const char __user *const __user *, envp,
2109                 int, flags)
2110 {
2111         return do_execveat(fd,
2112                            getname_uflags(filename, flags),
2113                            argv, envp, flags);
2114 }
2115
2116 #ifdef CONFIG_COMPAT
2117 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2118         const compat_uptr_t __user *, argv,
2119         const compat_uptr_t __user *, envp)
2120 {
2121         return compat_do_execve(getname(filename), argv, envp);
2122 }
2123
2124 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2125                        const char __user *, filename,
2126                        const compat_uptr_t __user *, argv,
2127                        const compat_uptr_t __user *, envp,
2128                        int,  flags)
2129 {
2130         return compat_do_execveat(fd,
2131                                   getname_uflags(filename, flags),
2132                                   argv, envp, flags);
2133 }
2134 #endif
2135
2136 #ifdef CONFIG_SYSCTL
2137
2138 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2139                 void *buffer, size_t *lenp, loff_t *ppos)
2140 {
2141         int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2142
2143         if (!error)
2144                 validate_coredump_safety();
2145         return error;
2146 }
2147
2148 static struct ctl_table fs_exec_sysctls[] = {
2149         {
2150                 .procname       = "suid_dumpable",
2151                 .data           = &suid_dumpable,
2152                 .maxlen         = sizeof(int),
2153                 .mode           = 0644,
2154                 .proc_handler   = proc_dointvec_minmax_coredump,
2155                 .extra1         = SYSCTL_ZERO,
2156                 .extra2         = SYSCTL_TWO,
2157         },
2158         { }
2159 };
2160
2161 static int __init init_fs_exec_sysctls(void)
2162 {
2163         register_sysctl_init("fs", fs_exec_sysctls);
2164         return 0;
2165 }
2166
2167 fs_initcall(init_fs_exec_sysctls);
2168 #endif /* CONFIG_SYSCTL */