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         int ret;
 202         unsigned int gup_flags = FOLL_FORCE;
 203
 204 #ifdef CONFIG_STACK_GROWSUP
 205         if (write) {
 206                 ret = expand_downwards(bprm->vma, pos);
 207                 if (ret < 0)
 208                         return NULL;
 209         }
 210 #endif
 211
 212         if (write)
 213                 gup_flags |= FOLL_WRITE;
 214
 215         /*
 216          * We are doing an exec().  'current' is the process
 217          * doing the exec and bprm->mm is the new process's mm.
 218          */
 219         mmap_read_lock(bprm->mm);
 220         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 221                         &page, NULL, NULL);
 222         mmap_read_unlock(bprm->mm);
 223         if (ret <= 0)
 224                 return NULL;
 225
 226         if (write)
 227                 acct_arg_size(bprm, vma_pages(bprm->vma));
 228
 229         return page;
 230 }
 231
 232 static void put_arg_page(struct page *page)
 233 {
 234         put_page(page);
 235 }
 236
 237 static void free_arg_pages(struct linux_binprm *bprm)
 238 {
 239 }
 240
 241 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 242                 struct page *page)
 243 {
 244         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 245 }
 246
 247 static int __bprm_mm_init(struct linux_binprm *bprm)
 248 {
 249         int err;
 250         struct vm_area_struct *vma = NULL;
 251         struct mm_struct *mm = bprm->mm;
 252
 253         bprm->vma = vma = vm_area_alloc(mm);
 254         if (!vma)
 255                 return -ENOMEM;
 256         vma_set_anonymous(vma);
 257
 258         if (mmap_write_lock_killable(mm)) {
 259                 err = -EINTR;
 260                 goto err_free;
 261         }
 262
 263         /*
 264          * Place the stack at the largest stack address the architecture
 265          * supports. Later, we'll move this to an appropriate place. We don't
 266          * use STACK_TOP because that can depend on attributes which aren't
 267          * configured yet.
 268          */
 269         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 270         vma->vm_end = STACK_TOP_MAX;
 271         vma->vm_start = vma->vm_end - PAGE_SIZE;
 272         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 273         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 274
 275         err = insert_vm_struct(mm, vma);
 276         if (err)
 277                 goto err;
 278
 279         mm->stack_vm = mm->total_vm = 1;
 280         mmap_write_unlock(mm);
 281         bprm->p = vma->vm_end - sizeof(void *);
 282         return 0;
 283 err:
 284         mmap_write_unlock(mm);
 285 err_free:
 286         bprm->vma = NULL;
 287         vm_area_free(vma);
 288         return err;
 289 }
 290
 291 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 292 {
 293         return len <= MAX_ARG_STRLEN;
 294 }
 295
 296 #else
 297
 298 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 299 {
 300 }
 301
 302 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 303                 int write)
 304 {
 305         struct page *page;
 306
 307         page = bprm->page[pos / PAGE_SIZE];
 308         if (!page && write) {
 309                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 310                 if (!page)
 311                         return NULL;
 312                 bprm->page[pos / PAGE_SIZE] = page;
 313         }
 314
 315         return page;
 316 }
 317
 318 static void put_arg_page(struct page *page)
 319 {
 320 }
 321
 322 static void free_arg_page(struct linux_binprm *bprm, int i)
 323 {
 324         if (bprm->page[i]) {
 325                 __free_page(bprm->page[i]);
 326                 bprm->page[i] = NULL;
 327         }
 328 }
 329
 330 static void free_arg_pages(struct linux_binprm *bprm)
 331 {
 332         int i;
 333
 334         for (i = 0; i < MAX_ARG_PAGES; i++)
 335                 free_arg_page(bprm, i);
 336 }
 337
 338 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 339                 struct page *page)
 340 {
 341 }
 342
 343 static int __bprm_mm_init(struct linux_binprm *bprm)
 344 {
 345         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 346         return 0;
 347 }
 348
 349 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 350 {
 351         return len <= bprm->p;
 352 }
 353
 354 #endif /* CONFIG_MMU */
 355
 356 /*
 357  * Create a new mm_struct and populate it with a temporary stack
 358  * vm_area_struct.  We don't have enough context at this point to set the stack
 359  * flags, permissions, and offset, so we use temporary values.  We'll update
 360  * them later in setup_arg_pages().
 361  */
 362 static int bprm_mm_init(struct linux_binprm *bprm)
 363 {
 364         int err;
 365         struct mm_struct *mm = NULL;
 366
 367         bprm->mm = mm = mm_alloc();
 368         err = -ENOMEM;
 369         if (!mm)
 370                 goto err;
 371
 372         /* Save current stack limit for all calculations made during exec. */
 373         task_lock(current->group_leader);
 374         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 375         task_unlock(current->group_leader);
 376
 377         err = __bprm_mm_init(bprm);
 378         if (err)
 379                 goto err;
 380
 381         return 0;
 382
 383 err:
 384         if (mm) {
 385                 bprm->mm = NULL;
 386                 mmdrop(mm);
 387         }
 388
 389         return err;
 390 }
 391
 392 struct user_arg_ptr {
 393 #ifdef CONFIG_COMPAT
 394         bool is_compat;
 395 #endif
 396         union {
 397                 const char __user *const __user *native;
 398 #ifdef CONFIG_COMPAT
 399                 const compat_uptr_t __user *compat;
 400 #endif
 401         } ptr;
 402 };
 403
 404 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 405 {
 406         const char __user *native;
 407
 408 #ifdef CONFIG_COMPAT
 409         if (unlikely(argv.is_compat)) {
 410                 compat_uptr_t compat;
 411
 412                 if (get_user(compat, argv.ptr.compat + nr))
 413                         return ERR_PTR(-EFAULT);
 414
 415                 return compat_ptr(compat);
 416         }
 417 #endif
 418
 419         if (get_user(native, argv.ptr.native + nr))
 420                 return ERR_PTR(-EFAULT);
 421
 422         return native;
 423 }
 424
 425 /*
 426  * count() counts the number of strings in array ARGV.
 427  */
 428 static int count(struct user_arg_ptr argv, int max)
 429 {
 430         int i = 0;
 431
 432         if (argv.ptr.native != NULL) {
 433                 for (;;) {
 434                         const char __user *p = get_user_arg_ptr(argv, i);
 435
 436                         if (!p)
 437                                 break;
 438
 439                         if (IS_ERR(p))
 440                                 return -EFAULT;
 441
 442                         if (i >= max)
 443                                 return -E2BIG;
 444                         ++i;
 445
 446                         if (fatal_signal_pending(current))
 447                                 return -ERESTARTNOHAND;
 448                         cond_resched();
 449                 }
 450         }
 451         return i;
 452 }
 453
 454 static int count_strings_kernel(const char *const *argv)
 455 {
 456         int i;
 457
 458         if (!argv)
 459                 return 0;
 460
 461         for (i = 0; argv[i]; ++i) {
 462                 if (i >= MAX_ARG_STRINGS)
 463                         return -E2BIG;
 464                 if (fatal_signal_pending(current))
 465                         return -ERESTARTNOHAND;
 466                 cond_resched();
 467         }
 468         return i;
 469 }
 470
 471 static int bprm_stack_limits(struct linux_binprm *bprm)
 472 {
 473         unsigned long limit, ptr_size;
 474
 475         /*
 476          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 477          * (whichever is smaller) for the argv+env strings.
 478          * This ensures that:
 479          *  - the remaining binfmt code will not run out of stack space,
 480          *  - the program will have a reasonable amount of stack left
 481          *    to work from.
 482          */
 483         limit = _STK_LIM / 4 * 3;
 484         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 485         /*
 486          * We've historically supported up to 32 pages (ARG_MAX)
 487          * of argument strings even with small stacks
 488          */
 489         limit = max_t(unsigned long, limit, ARG_MAX);
 490         /*
 491          * We must account for the size of all the argv and envp pointers to
 492          * the argv and envp strings, since they will also take up space in
 493          * the stack. They aren't stored until much later when we can't
 494          * signal to the parent that the child has run out of stack space.
 495          * Instead, calculate it here so it's possible to fail gracefully.
 496          *
 497          * In the case of argc = 0, make sure there is space for adding a
 498          * empty string (which will bump argc to 1), to ensure confused
 499          * userspace programs don't start processing from argv[1], thinking
 500          * argc can never be 0, to keep them from walking envp by accident.
 501          * See do_execveat_common().
 502          */
 503         ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
 504         if (limit <= ptr_size)
 505                 return -E2BIG;
 506         limit -= ptr_size;
 507
 508         bprm->argmin = bprm->p - limit;
 509         return 0;
 510 }
 511
 512 /*
 513  * 'copy_strings()' copies argument/environment strings from the old
 514  * processes's memory to the new process's stack.  The call to get_user_pages()
 515  * ensures the destination page is created and not swapped out.
 516  */
 517 static int copy_strings(int argc, struct user_arg_ptr argv,
 518                         struct linux_binprm *bprm)
 519 {
 520         struct page *kmapped_page = NULL;
 521         char *kaddr = NULL;
 522         unsigned long kpos = 0;
 523         int ret;
 524
 525         while (argc-- > 0) {
 526                 const char __user *str;
 527                 int len;
 528                 unsigned long pos;
 529
 530                 ret = -EFAULT;
 531                 str = get_user_arg_ptr(argv, argc);
 532                 if (IS_ERR(str))
 533                         goto out;
 534
 535                 len = strnlen_user(str, MAX_ARG_STRLEN);
 536                 if (!len)
 537                         goto out;
 538
 539                 ret = -E2BIG;
 540                 if (!valid_arg_len(bprm, len))
 541                         goto out;
 542
 543                 /* We're going to work our way backwards. */
 544                 pos = bprm->p;
 545                 str += len;
 546                 bprm->p -= len;
 547 #ifdef CONFIG_MMU
 548                 if (bprm->p < bprm->argmin)
 549                         goto out;
 550 #endif
 551
 552                 while (len > 0) {
 553                         int offset, bytes_to_copy;
 554
 555                         if (fatal_signal_pending(current)) {
 556                                 ret = -ERESTARTNOHAND;
 557                                 goto out;
 558                         }
 559                         cond_resched();
 560
 561                         offset = pos % PAGE_SIZE;
 562                         if (offset == 0)
 563                                 offset = PAGE_SIZE;
 564
 565                         bytes_to_copy = offset;
 566                         if (bytes_to_copy > len)
 567                                 bytes_to_copy = len;
 568
 569                         offset -= bytes_to_copy;
 570                         pos -= bytes_to_copy;
 571                         str -= bytes_to_copy;
 572                         len -= bytes_to_copy;
 573
 574                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 575                                 struct page *page;
 576
 577                                 page = get_arg_page(bprm, pos, 1);
 578                                 if (!page) {
 579                                         ret = -E2BIG;
 580                                         goto out;
 581                                 }
 582
 583                                 if (kmapped_page) {
 584                                         flush_dcache_page(kmapped_page);
 585                                         kunmap_local(kaddr);
 586                                         put_arg_page(kmapped_page);
 587                                 }
 588                                 kmapped_page = page;
 589                                 kaddr = kmap_local_page(kmapped_page);
 590                                 kpos = pos & PAGE_MASK;
 591                                 flush_arg_page(bprm, kpos, kmapped_page);
 592                         }
 593                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 594                                 ret = -EFAULT;
 595                                 goto out;
 596                         }
 597                 }
 598         }
 599         ret = 0;
 600 out:
 601         if (kmapped_page) {
 602                 flush_dcache_page(kmapped_page);
 603                 kunmap_local(kaddr);
 604                 put_arg_page(kmapped_page);
 605         }
 606         return ret;
 607 }
 608
 609 /*
 610  * Copy and argument/environment string from the kernel to the processes stack.
 611  */
 612 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 613 {
 614         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 615         unsigned long pos = bprm->p;
 616
 617         if (len == 0)
 618                 return -EFAULT;
 619         if (!valid_arg_len(bprm, len))
 620                 return -E2BIG;
 621
 622         /* We're going to work our way backwards. */
 623         arg += len;
 624         bprm->p -= len;
 625         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 626                 return -E2BIG;
 627
 628         while (len > 0) {
 629                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 630                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 631                 struct page *page;
 632
 633                 pos -= bytes_to_copy;
 634                 arg -= bytes_to_copy;
 635                 len -= bytes_to_copy;
 636
 637                 page = get_arg_page(bprm, pos, 1);
 638                 if (!page)
 639                         return -E2BIG;
 640                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 641                 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
 642                 put_arg_page(page);
 643         }
 644
 645         return 0;
 646 }
 647 EXPORT_SYMBOL(copy_string_kernel);
 648
 649 static int copy_strings_kernel(int argc, const char *const *argv,
 650                                struct linux_binprm *bprm)
 651 {
 652         while (argc-- > 0) {
 653                 int ret = copy_string_kernel(argv[argc], bprm);
 654                 if (ret < 0)
 655                         return ret;
 656                 if (fatal_signal_pending(current))
 657                         return -ERESTARTNOHAND;
 658                 cond_resched();
 659         }
 660         return 0;
 661 }
 662
 663 #ifdef CONFIG_MMU
 664
 665 /*
 666  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 667  * the binfmt code determines where the new stack should reside, we shift it to
 668  * its final location.  The process proceeds as follows:
 669  *
 670  * 1) Use shift to calculate the new vma endpoints.
 671  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 672  *    arguments passed to subsequent functions are consistent.
 673  * 3) Move vma's page tables to the new range.
 674  * 4) Free up any cleared pgd range.
 675  * 5) Shrink the vma to cover only the new range.
 676  */
 677 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 678 {
 679         struct mm_struct *mm = vma->vm_mm;
 680         unsigned long old_start = vma->vm_start;
 681         unsigned long old_end = vma->vm_end;
 682         unsigned long length = old_end - old_start;
 683         unsigned long new_start = old_start - shift;
 684         unsigned long new_end = old_end - shift;
 685         VMA_ITERATOR(vmi, mm, new_start);
 686         struct vm_area_struct *next;
 687         struct mmu_gather tlb;
 688
 689         BUG_ON(new_start > new_end);
 690
 691         /*
 692          * ensure there are no vmas between where we want to go
 693          * and where we are
 694          */
 695         if (vma != vma_next(&vmi))
 696                 return -EFAULT;
 697
 698         /*
 699          * cover the whole range: [new_start, old_end)
 700          */
 701         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 702                 return -ENOMEM;
 703
 704         /*
 705          * move the page tables downwards, on failure we rely on
 706          * process cleanup to remove whatever mess we made.
 707          */
 708         if (length != move_page_tables(vma, old_start,
 709                                        vma, new_start, length, false))
 710                 return -ENOMEM;
 711
 712         lru_add_drain();
 713         tlb_gather_mmu(&tlb, mm);
 714         next = vma_next(&vmi);
 715         if (new_end > old_start) {
 716                 /*
 717                  * when the old and new regions overlap clear from new_end.
 718                  */
 719                 free_pgd_range(&tlb, new_end, old_end, new_end,
 720                         next ? next->vm_start : USER_PGTABLES_CEILING);
 721         } else {
 722                 /*
 723                  * otherwise, clean from old_start; this is done to not touch
 724                  * the address space in [new_end, old_start) some architectures
 725                  * have constraints on va-space that make this illegal (IA64) -
 726                  * for the others its just a little faster.
 727                  */
 728                 free_pgd_range(&tlb, old_start, old_end, new_end,
 729                         next ? next->vm_start : USER_PGTABLES_CEILING);
 730         }
 731         tlb_finish_mmu(&tlb);
 732
 733         /*
 734          * Shrink the vma to just the new range.  Always succeeds.
 735          */
 736         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 737
 738         return 0;
 739 }
 740
 741 /*
 742  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 743  * the stack is optionally relocated, and some extra space is added.
 744  */
 745 int setup_arg_pages(struct linux_binprm *bprm,
 746                     unsigned long stack_top,
 747                     int executable_stack)
 748 {
 749         unsigned long ret;
 750         unsigned long stack_shift;
 751         struct mm_struct *mm = current->mm;
 752         struct vm_area_struct *vma = bprm->vma;
 753         struct vm_area_struct *prev = NULL;
 754         unsigned long vm_flags;
 755         unsigned long stack_base;
 756         unsigned long stack_size;
 757         unsigned long stack_expand;
 758         unsigned long rlim_stack;
 759         struct mmu_gather tlb;
 760
 761 #ifdef CONFIG_STACK_GROWSUP
 762         /* Limit stack size */
 763         stack_base = bprm->rlim_stack.rlim_max;
 764
 765         stack_base = calc_max_stack_size(stack_base);
 766
 767         /* Add space for stack randomization. */
 768         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 769
 770         /* Make sure we didn't let the argument array grow too large. */
 771         if (vma->vm_end - vma->vm_start > stack_base)
 772                 return -ENOMEM;
 773
 774         stack_base = PAGE_ALIGN(stack_top - stack_base);
 775
 776         stack_shift = vma->vm_start - stack_base;
 777         mm->arg_start = bprm->p - stack_shift;
 778         bprm->p = vma->vm_end - stack_shift;
 779 #else
 780         stack_top = arch_align_stack(stack_top);
 781         stack_top = PAGE_ALIGN(stack_top);
 782
 783         if (unlikely(stack_top < mmap_min_addr) ||
 784             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 785                 return -ENOMEM;
 786
 787         stack_shift = vma->vm_end - stack_top;
 788
 789         bprm->p -= stack_shift;
 790         mm->arg_start = bprm->p;
 791 #endif
 792
 793         if (bprm->loader)
 794                 bprm->loader -= stack_shift;
 795         bprm->exec -= stack_shift;
 796
 797         if (mmap_write_lock_killable(mm))
 798                 return -EINTR;
 799
 800         vm_flags = VM_STACK_FLAGS;
 801
 802         /*
 803          * Adjust stack execute permissions; explicitly enable for
 804          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 805          * (arch default) otherwise.
 806          */
 807         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 808                 vm_flags |= VM_EXEC;
 809         else if (executable_stack == EXSTACK_DISABLE_X)
 810                 vm_flags &= ~VM_EXEC;
 811         vm_flags |= mm->def_flags;
 812         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 813
 814         tlb_gather_mmu(&tlb, mm);
 815         ret = mprotect_fixup(&tlb, vma, &prev, vma->vm_start, vma->vm_end,
 816                         vm_flags);
 817         tlb_finish_mmu(&tlb);
 818
 819         if (ret)
 820                 goto out_unlock;
 821         BUG_ON(prev != vma);
 822
 823         if (unlikely(vm_flags & VM_EXEC)) {
 824                 pr_warn_once("process '%pD4' started with executable stack\n",
 825                              bprm->file);
 826         }
 827
 828         /* Move stack pages down in memory. */
 829         if (stack_shift) {
 830                 ret = shift_arg_pages(vma, stack_shift);
 831                 if (ret)
 832                         goto out_unlock;
 833         }
 834
 835         /* mprotect_fixup is overkill to remove the temporary stack flags */
 836         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 837
 838         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 839         stack_size = vma->vm_end - vma->vm_start;
 840         /*
 841          * Align this down to a page boundary as expand_stack
 842          * will align it up.
 843          */
 844         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 845 #ifdef CONFIG_STACK_GROWSUP
 846         if (stack_size + stack_expand > rlim_stack)
 847                 stack_base = vma->vm_start + rlim_stack;
 848         else
 849                 stack_base = vma->vm_end + stack_expand;
 850 #else
 851         if (stack_size + stack_expand > rlim_stack)
 852                 stack_base = vma->vm_end - rlim_stack;
 853         else
 854                 stack_base = vma->vm_start - stack_expand;
 855 #endif
 856         current->mm->start_stack = bprm->p;
 857         ret = expand_stack(vma, stack_base);
 858         if (ret)
 859                 ret = -EFAULT;
 860
 861 out_unlock:
 862         mmap_write_unlock(mm);
 863         return ret;
 864 }
 865 EXPORT_SYMBOL(setup_arg_pages);
 866
 867 #else
 868
 869 /*
 870  * Transfer the program arguments and environment from the holding pages
 871  * onto the stack. The provided stack pointer is adjusted accordingly.
 872  */
 873 int transfer_args_to_stack(struct linux_binprm *bprm,
 874                            unsigned long *sp_location)
 875 {
 876         unsigned long index, stop, sp;
 877         int ret = 0;
 878
 879         stop = bprm->p >> PAGE_SHIFT;
 880         sp = *sp_location;
 881
 882         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 883                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 884                 char *src = kmap_local_page(bprm->page[index]) + offset;
 885                 sp -= PAGE_SIZE - offset;
 886                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 887                         ret = -EFAULT;
 888                 kunmap_local(src);
 889                 if (ret)
 890                         goto out;
 891         }
 892
 893         *sp_location = sp;
 894
 895 out:
 896         return ret;
 897 }
 898 EXPORT_SYMBOL(transfer_args_to_stack);
 899
 900 #endif /* CONFIG_MMU */
 901
 902 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 903 {
 904         struct file *file;
 905         int err;
 906         struct open_flags open_exec_flags = {
 907                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 908                 .acc_mode = MAY_EXEC,
 909                 .intent = LOOKUP_OPEN,
 910                 .lookup_flags = LOOKUP_FOLLOW,
 911         };
 912
 913         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 914                 return ERR_PTR(-EINVAL);
 915         if (flags & AT_SYMLINK_NOFOLLOW)
 916                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 917         if (flags & AT_EMPTY_PATH)
 918                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 919
 920         file = do_filp_open(fd, name, &open_exec_flags);
 921         if (IS_ERR(file))
 922                 goto out;
 923
 924         /*
 925          * may_open() has already checked for this, so it should be
 926          * impossible to trip now. But we need to be extra cautious
 927          * and check again at the very end too.
 928          */
 929         err = -EACCES;
 930         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 931                          path_noexec(&file->f_path)))
 932                 goto exit;
 933
 934         err = deny_write_access(file);
 935         if (err)
 936                 goto exit;
 937
 938         if (name->name[0] != '\0')
 939                 fsnotify_open(file);
 940
 941 out:
 942         return file;
 943
 944 exit:
 945         fput(file);
 946         return ERR_PTR(err);
 947 }
 948
 949 struct file *open_exec(const char *name)
 950 {
 951         struct filename *filename = getname_kernel(name);
 952         struct file *f = ERR_CAST(filename);
 953
 954         if (!IS_ERR(filename)) {
 955                 f = do_open_execat(AT_FDCWD, filename, 0);
 956                 putname(filename);
 957         }
 958         return f;
 959 }
 960 EXPORT_SYMBOL(open_exec);
 961
 962 #if defined(CONFIG_BINFMT_FLAT) || 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         lru_gen_add_mm(mm);
1016         /*
1017          * This prevents preemption while active_mm is being loaded and
1018          * it and mm are being updated, which could cause problems for
1019          * lazy tlb mm refcounting when these are updated by context
1020          * switches. Not all architectures can handle irqs off over
1021          * activate_mm yet.
1022          */
1023         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1024                 local_irq_enable();
1025         activate_mm(active_mm, mm);
1026         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1027                 local_irq_enable();
1028         task_unlock(tsk);
1029         lru_gen_use_mm(mm);
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 won'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         spin_lock_irq(&me->sighand->siglock);
1302         posix_cpu_timers_exit(me);
1303         spin_unlock_irq(&me->sighand->siglock);
1304         exit_itimers(me);
1305         flush_itimer_signals();
1306 #endif
1307
1308         /*
1309          * Make the signal table private.
1310          */
1311         retval = unshare_sighand(me);
1312         if (retval)
1313                 goto out_unlock;
1314
1315         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1316                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1317         flush_thread();
1318         me->personality &= ~bprm->per_clear;
1319
1320         clear_syscall_work_syscall_user_dispatch(me);
1321
1322         /*
1323          * We have to apply CLOEXEC before we change whether the process is
1324          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1325          * trying to access the should-be-closed file descriptors of a process
1326          * undergoing exec(2).
1327          */
1328         do_close_on_exec(me->files);
1329
1330         if (bprm->secureexec) {
1331                 /* Make sure parent cannot signal privileged process. */
1332                 me->pdeath_signal = 0;
1333
1334                 /*
1335                  * For secureexec, reset the stack limit to sane default to
1336                  * avoid bad behavior from the prior rlimits. This has to
1337                  * happen before arch_pick_mmap_layout(), which examines
1338                  * RLIMIT_STACK, but after the point of no return to avoid
1339                  * needing to clean up the change on failure.
1340                  */
1341                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1342                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1343         }
1344
1345         me->sas_ss_sp = me->sas_ss_size = 0;
1346
1347         /*
1348          * Figure out dumpability. Note that this checking only of current
1349          * is wrong, but userspace depends on it. This should be testing
1350          * bprm->secureexec instead.
1351          */
1352         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1353             !(uid_eq(current_euid(), current_uid()) &&
1354               gid_eq(current_egid(), current_gid())))
1355                 set_dumpable(current->mm, suid_dumpable);
1356         else
1357                 set_dumpable(current->mm, SUID_DUMP_USER);
1358
1359         perf_event_exec();
1360         __set_task_comm(me, kbasename(bprm->filename), true);
1361
1362         /* An exec changes our domain. We are no longer part of the thread
1363            group */
1364         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1365         flush_signal_handlers(me, 0);
1366
1367         retval = set_cred_ucounts(bprm->cred);
1368         if (retval < 0)
1369                 goto out_unlock;
1370
1371         /*
1372          * install the new credentials for this executable
1373          */
1374         security_bprm_committing_creds(bprm);
1375
1376         commit_creds(bprm->cred);
1377         bprm->cred = NULL;
1378
1379         /*
1380          * Disable monitoring for regular users
1381          * when executing setuid binaries. Must
1382          * wait until new credentials are committed
1383          * by commit_creds() above
1384          */
1385         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1386                 perf_event_exit_task(me);
1387         /*
1388          * cred_guard_mutex must be held at least to this point to prevent
1389          * ptrace_attach() from altering our determination of the task's
1390          * credentials; any time after this it may be unlocked.
1391          */
1392         security_bprm_committed_creds(bprm);
1393
1394         /* Pass the opened binary to the interpreter. */
1395         if (bprm->have_execfd) {
1396                 retval = get_unused_fd_flags(0);
1397                 if (retval < 0)
1398                         goto out_unlock;
1399                 fd_install(retval, bprm->executable);
1400                 bprm->executable = NULL;
1401                 bprm->execfd = retval;
1402         }
1403         return 0;
1404
1405 out_unlock:
1406         up_write(&me->signal->exec_update_lock);
1407 out:
1408         return retval;
1409 }
1410 EXPORT_SYMBOL(begin_new_exec);
1411
1412 void would_dump(struct linux_binprm *bprm, struct file *file)
1413 {
1414         struct inode *inode = file_inode(file);
1415         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1416         if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1417                 struct user_namespace *old, *user_ns;
1418                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1419
1420                 /* Ensure mm->user_ns contains the executable */
1421                 user_ns = old = bprm->mm->user_ns;
1422                 while ((user_ns != &init_user_ns) &&
1423                        !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1424                         user_ns = user_ns->parent;
1425
1426                 if (old != user_ns) {
1427                         bprm->mm->user_ns = get_user_ns(user_ns);
1428                         put_user_ns(old);
1429                 }
1430         }
1431 }
1432 EXPORT_SYMBOL(would_dump);
1433
1434 void setup_new_exec(struct linux_binprm * bprm)
1435 {
1436         /* Setup things that can depend upon the personality */
1437         struct task_struct *me = current;
1438
1439         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1440
1441         arch_setup_new_exec();
1442
1443         /* Set the new mm task size. We have to do that late because it may
1444          * depend on TIF_32BIT which is only updated in flush_thread() on
1445          * some architectures like powerpc
1446          */
1447         me->mm->task_size = TASK_SIZE;
1448         up_write(&me->signal->exec_update_lock);
1449         mutex_unlock(&me->signal->cred_guard_mutex);
1450 }
1451 EXPORT_SYMBOL(setup_new_exec);
1452
1453 /* Runs immediately before start_thread() takes over. */
1454 void finalize_exec(struct linux_binprm *bprm)
1455 {
1456         /* Store any stack rlimit changes before starting thread. */
1457         task_lock(current->group_leader);
1458         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1459         task_unlock(current->group_leader);
1460 }
1461 EXPORT_SYMBOL(finalize_exec);
1462
1463 /*
1464  * Prepare credentials and lock ->cred_guard_mutex.
1465  * setup_new_exec() commits the new creds and drops the lock.
1466  * Or, if exec fails before, free_bprm() should release ->cred
1467  * and unlock.
1468  */
1469 static int prepare_bprm_creds(struct linux_binprm *bprm)
1470 {
1471         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1472                 return -ERESTARTNOINTR;
1473
1474         bprm->cred = prepare_exec_creds();
1475         if (likely(bprm->cred))
1476                 return 0;
1477
1478         mutex_unlock(&current->signal->cred_guard_mutex);
1479         return -ENOMEM;
1480 }
1481
1482 static void free_bprm(struct linux_binprm *bprm)
1483 {
1484         if (bprm->mm) {
1485                 acct_arg_size(bprm, 0);
1486                 mmput(bprm->mm);
1487         }
1488         free_arg_pages(bprm);
1489         if (bprm->cred) {
1490                 mutex_unlock(&current->signal->cred_guard_mutex);
1491                 abort_creds(bprm->cred);
1492         }
1493         if (bprm->file) {
1494                 allow_write_access(bprm->file);
1495                 fput(bprm->file);
1496         }
1497         if (bprm->executable)
1498                 fput(bprm->executable);
1499         /* If a binfmt changed the interp, free it. */
1500         if (bprm->interp != bprm->filename)
1501                 kfree(bprm->interp);
1502         kfree(bprm->fdpath);
1503         kfree(bprm);
1504 }
1505
1506 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1507 {
1508         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1509         int retval = -ENOMEM;
1510         if (!bprm)
1511                 goto out;
1512
1513         if (fd == AT_FDCWD || filename->name[0] == '/') {
1514                 bprm->filename = filename->name;
1515         } else {
1516                 if (filename->name[0] == '\0')
1517                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1518                 else
1519                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1520                                                   fd, filename->name);
1521                 if (!bprm->fdpath)
1522                         goto out_free;
1523
1524                 bprm->filename = bprm->fdpath;
1525         }
1526         bprm->interp = bprm->filename;
1527
1528         retval = bprm_mm_init(bprm);
1529         if (retval)
1530                 goto out_free;
1531         return bprm;
1532
1533 out_free:
1534         free_bprm(bprm);
1535 out:
1536         return ERR_PTR(retval);
1537 }
1538
1539 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1540 {
1541         /* If a binfmt changed the interp, free it first. */
1542         if (bprm->interp != bprm->filename)
1543                 kfree(bprm->interp);
1544         bprm->interp = kstrdup(interp, GFP_KERNEL);
1545         if (!bprm->interp)
1546                 return -ENOMEM;
1547         return 0;
1548 }
1549 EXPORT_SYMBOL(bprm_change_interp);
1550
1551 /*
1552  * determine how safe it is to execute the proposed program
1553  * - the caller must hold ->cred_guard_mutex to protect against
1554  *   PTRACE_ATTACH or seccomp thread-sync
1555  */
1556 static void check_unsafe_exec(struct linux_binprm *bprm)
1557 {
1558         struct task_struct *p = current, *t;
1559         unsigned n_fs;
1560
1561         if (p->ptrace)
1562                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1563
1564         /*
1565          * This isn't strictly necessary, but it makes it harder for LSMs to
1566          * mess up.
1567          */
1568         if (task_no_new_privs(current))
1569                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1570
1571         t = p;
1572         n_fs = 1;
1573         spin_lock(&p->fs->lock);
1574         rcu_read_lock();
1575         while_each_thread(p, t) {
1576                 if (t->fs == p->fs)
1577                         n_fs++;
1578         }
1579         rcu_read_unlock();
1580
1581         if (p->fs->users > n_fs)
1582                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1583         else
1584                 p->fs->in_exec = 1;
1585         spin_unlock(&p->fs->lock);
1586 }
1587
1588 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1589 {
1590         /* Handle suid and sgid on files */
1591         struct user_namespace *mnt_userns;
1592         struct inode *inode = file_inode(file);
1593         unsigned int mode;
1594         kuid_t uid;
1595         kgid_t gid;
1596
1597         if (!mnt_may_suid(file->f_path.mnt))
1598                 return;
1599
1600         if (task_no_new_privs(current))
1601                 return;
1602
1603         mode = READ_ONCE(inode->i_mode);
1604         if (!(mode & (S_ISUID|S_ISGID)))
1605                 return;
1606
1607         mnt_userns = file_mnt_user_ns(file);
1608
1609         /* Be careful if suid/sgid is set */
1610         inode_lock(inode);
1611
1612         /* reload atomically mode/uid/gid now that lock held */
1613         mode = inode->i_mode;
1614         uid = i_uid_into_mnt(mnt_userns, inode);
1615         gid = i_gid_into_mnt(mnt_userns, inode);
1616         inode_unlock(inode);
1617
1618         /* We ignore suid/sgid if there are no mappings for them in the ns */
1619         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1620                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1621                 return;
1622
1623         if (mode & S_ISUID) {
1624                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1625                 bprm->cred->euid = uid;
1626         }
1627
1628         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1629                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1630                 bprm->cred->egid = gid;
1631         }
1632 }
1633
1634 /*
1635  * Compute brpm->cred based upon the final binary.
1636  */
1637 static int bprm_creds_from_file(struct linux_binprm *bprm)
1638 {
1639         /* Compute creds based on which file? */
1640         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1641
1642         bprm_fill_uid(bprm, file);
1643         return security_bprm_creds_from_file(bprm, file);
1644 }
1645
1646 /*
1647  * Fill the binprm structure from the inode.
1648  * Read the first BINPRM_BUF_SIZE bytes
1649  *
1650  * This may be called multiple times for binary chains (scripts for example).
1651  */
1652 static int prepare_binprm(struct linux_binprm *bprm)
1653 {
1654         loff_t pos = 0;
1655
1656         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1657         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1658 }
1659
1660 /*
1661  * Arguments are '\0' separated strings found at the location bprm->p
1662  * points to; chop off the first by relocating brpm->p to right after
1663  * the first '\0' encountered.
1664  */
1665 int remove_arg_zero(struct linux_binprm *bprm)
1666 {
1667         int ret = 0;
1668         unsigned long offset;
1669         char *kaddr;
1670         struct page *page;
1671
1672         if (!bprm->argc)
1673                 return 0;
1674
1675         do {
1676                 offset = bprm->p & ~PAGE_MASK;
1677                 page = get_arg_page(bprm, bprm->p, 0);
1678                 if (!page) {
1679                         ret = -EFAULT;
1680                         goto out;
1681                 }
1682                 kaddr = kmap_local_page(page);
1683
1684                 for (; offset < PAGE_SIZE && kaddr[offset];
1685                                 offset++, bprm->p++)
1686                         ;
1687
1688                 kunmap_local(kaddr);
1689                 put_arg_page(page);
1690         } while (offset == PAGE_SIZE);
1691
1692         bprm->p++;
1693         bprm->argc--;
1694         ret = 0;
1695
1696 out:
1697         return ret;
1698 }
1699 EXPORT_SYMBOL(remove_arg_zero);
1700
1701 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1702 /*
1703  * cycle the list of binary formats handler, until one recognizes the image
1704  */
1705 static int search_binary_handler(struct linux_binprm *bprm)
1706 {
1707         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1708         struct linux_binfmt *fmt;
1709         int retval;
1710
1711         retval = prepare_binprm(bprm);
1712         if (retval < 0)
1713                 return retval;
1714
1715         retval = security_bprm_check(bprm);
1716         if (retval)
1717                 return retval;
1718
1719         retval = -ENOENT;
1720  retry:
1721         read_lock(&binfmt_lock);
1722         list_for_each_entry(fmt, &formats, lh) {
1723                 if (!try_module_get(fmt->module))
1724                         continue;
1725                 read_unlock(&binfmt_lock);
1726
1727                 retval = fmt->load_binary(bprm);
1728
1729                 read_lock(&binfmt_lock);
1730                 put_binfmt(fmt);
1731                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1732                         read_unlock(&binfmt_lock);
1733                         return retval;
1734                 }
1735         }
1736         read_unlock(&binfmt_lock);
1737
1738         if (need_retry) {
1739                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1740                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1741                         return retval;
1742                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1743                         return retval;
1744                 need_retry = false;
1745                 goto retry;
1746         }
1747
1748         return retval;
1749 }
1750
1751 static int exec_binprm(struct linux_binprm *bprm)
1752 {
1753         pid_t old_pid, old_vpid;
1754         int ret, depth;
1755
1756         /* Need to fetch pid before load_binary changes it */
1757         old_pid = current->pid;
1758         rcu_read_lock();
1759         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1760         rcu_read_unlock();
1761
1762         /* This allows 4 levels of binfmt rewrites before failing hard. */
1763         for (depth = 0;; depth++) {
1764                 struct file *exec;
1765                 if (depth > 5)
1766                         return -ELOOP;
1767
1768                 ret = search_binary_handler(bprm);
1769                 if (ret < 0)
1770                         return ret;
1771                 if (!bprm->interpreter)
1772                         break;
1773
1774                 exec = bprm->file;
1775                 bprm->file = bprm->interpreter;
1776                 bprm->interpreter = NULL;
1777
1778                 allow_write_access(exec);
1779                 if (unlikely(bprm->have_execfd)) {
1780                         if (bprm->executable) {
1781                                 fput(exec);
1782                                 return -ENOEXEC;
1783                         }
1784                         bprm->executable = exec;
1785                 } else
1786                         fput(exec);
1787         }
1788
1789         audit_bprm(bprm);
1790         trace_sched_process_exec(current, old_pid, bprm);
1791         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1792         proc_exec_connector(current);
1793         return 0;
1794 }
1795
1796 /*
1797  * sys_execve() executes a new program.
1798  */
1799 static int bprm_execve(struct linux_binprm *bprm,
1800                        int fd, struct filename *filename, int flags)
1801 {
1802         struct file *file;
1803         int retval;
1804
1805         retval = prepare_bprm_creds(bprm);
1806         if (retval)
1807                 return retval;
1808
1809         check_unsafe_exec(bprm);
1810         current->in_execve = 1;
1811
1812         file = do_open_execat(fd, filename, flags);
1813         retval = PTR_ERR(file);
1814         if (IS_ERR(file))
1815                 goto out_unmark;
1816
1817         sched_exec();
1818
1819         bprm->file = file;
1820         /*
1821          * Record that a name derived from an O_CLOEXEC fd will be
1822          * inaccessible after exec.  This allows the code in exec to
1823          * choose to fail when the executable is not mmaped into the
1824          * interpreter and an open file descriptor is not passed to
1825          * the interpreter.  This makes for a better user experience
1826          * than having the interpreter start and then immediately fail
1827          * when it finds the executable is inaccessible.
1828          */
1829         if (bprm->fdpath && get_close_on_exec(fd))
1830                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1831
1832         /* Set the unchanging part of bprm->cred */
1833         retval = security_bprm_creds_for_exec(bprm);
1834         if (retval)
1835                 goto out;
1836
1837         retval = exec_binprm(bprm);
1838         if (retval < 0)
1839                 goto out;
1840
1841         /* execve succeeded */
1842         current->fs->in_exec = 0;
1843         current->in_execve = 0;
1844         rseq_execve(current);
1845         acct_update_integrals(current);
1846         task_numa_free(current, false);
1847         return retval;
1848
1849 out:
1850         /*
1851          * If past the point of no return ensure the code never
1852          * returns to the userspace process.  Use an existing fatal
1853          * signal if present otherwise terminate the process with
1854          * SIGSEGV.
1855          */
1856         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1857                 force_fatal_sig(SIGSEGV);
1858
1859 out_unmark:
1860         current->fs->in_exec = 0;
1861         current->in_execve = 0;
1862
1863         return retval;
1864 }
1865
1866 static int do_execveat_common(int fd, struct filename *filename,
1867                               struct user_arg_ptr argv,
1868                               struct user_arg_ptr envp,
1869                               int flags)
1870 {
1871         struct linux_binprm *bprm;
1872         int retval;
1873
1874         if (IS_ERR(filename))
1875                 return PTR_ERR(filename);
1876
1877         /*
1878          * We move the actual failure in case of RLIMIT_NPROC excess from
1879          * set*uid() to execve() because too many poorly written programs
1880          * don't check setuid() return code.  Here we additionally recheck
1881          * whether NPROC limit is still exceeded.
1882          */
1883         if ((current->flags & PF_NPROC_EXCEEDED) &&
1884             is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1885                 retval = -EAGAIN;
1886                 goto out_ret;
1887         }
1888
1889         /* We're below the limit (still or again), so we don't want to make
1890          * further execve() calls fail. */
1891         current->flags &= ~PF_NPROC_EXCEEDED;
1892
1893         bprm = alloc_bprm(fd, filename);
1894         if (IS_ERR(bprm)) {
1895                 retval = PTR_ERR(bprm);
1896                 goto out_ret;
1897         }
1898
1899         retval = count(argv, MAX_ARG_STRINGS);
1900         if (retval == 0)
1901                 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1902                              current->comm, bprm->filename);
1903         if (retval < 0)
1904                 goto out_free;
1905         bprm->argc = retval;
1906
1907         retval = count(envp, MAX_ARG_STRINGS);
1908         if (retval < 0)
1909                 goto out_free;
1910         bprm->envc = retval;
1911
1912         retval = bprm_stack_limits(bprm);
1913         if (retval < 0)
1914                 goto out_free;
1915
1916         retval = copy_string_kernel(bprm->filename, bprm);
1917         if (retval < 0)
1918                 goto out_free;
1919         bprm->exec = bprm->p;
1920
1921         retval = copy_strings(bprm->envc, envp, bprm);
1922         if (retval < 0)
1923                 goto out_free;
1924
1925         retval = copy_strings(bprm->argc, argv, bprm);
1926         if (retval < 0)
1927                 goto out_free;
1928
1929         /*
1930          * When argv is empty, add an empty string ("") as argv[0] to
1931          * ensure confused userspace programs that start processing
1932          * from argv[1] won't end up walking envp. See also
1933          * bprm_stack_limits().
1934          */
1935         if (bprm->argc == 0) {
1936                 retval = copy_string_kernel("", bprm);
1937                 if (retval < 0)
1938                         goto out_free;
1939                 bprm->argc = 1;
1940         }
1941
1942         retval = bprm_execve(bprm, fd, filename, flags);
1943 out_free:
1944         free_bprm(bprm);
1945
1946 out_ret:
1947         putname(filename);
1948         return retval;
1949 }
1950
1951 int kernel_execve(const char *kernel_filename,
1952                   const char *const *argv, const char *const *envp)
1953 {
1954         struct filename *filename;
1955         struct linux_binprm *bprm;
1956         int fd = AT_FDCWD;
1957         int retval;
1958
1959         /* It is non-sense for kernel threads to call execve */
1960         if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1961                 return -EINVAL;
1962
1963         filename = getname_kernel(kernel_filename);
1964         if (IS_ERR(filename))
1965                 return PTR_ERR(filename);
1966
1967         bprm = alloc_bprm(fd, filename);
1968         if (IS_ERR(bprm)) {
1969                 retval = PTR_ERR(bprm);
1970                 goto out_ret;
1971         }
1972
1973         retval = count_strings_kernel(argv);
1974         if (WARN_ON_ONCE(retval == 0))
1975                 retval = -EINVAL;
1976         if (retval < 0)
1977                 goto out_free;
1978         bprm->argc = retval;
1979
1980         retval = count_strings_kernel(envp);
1981         if (retval < 0)
1982                 goto out_free;
1983         bprm->envc = retval;
1984
1985         retval = bprm_stack_limits(bprm);
1986         if (retval < 0)
1987                 goto out_free;
1988
1989         retval = copy_string_kernel(bprm->filename, bprm);
1990         if (retval < 0)
1991                 goto out_free;
1992         bprm->exec = bprm->p;
1993
1994         retval = copy_strings_kernel(bprm->envc, envp, bprm);
1995         if (retval < 0)
1996                 goto out_free;
1997
1998         retval = copy_strings_kernel(bprm->argc, argv, bprm);
1999         if (retval < 0)
2000                 goto out_free;
2001
2002         retval = bprm_execve(bprm, fd, filename, 0);
2003 out_free:
2004         free_bprm(bprm);
2005 out_ret:
2006         putname(filename);
2007         return retval;
2008 }
2009
2010 static int do_execve(struct filename *filename,
2011         const char __user *const __user *__argv,
2012         const char __user *const __user *__envp)
2013 {
2014         struct user_arg_ptr argv = { .ptr.native = __argv };
2015         struct user_arg_ptr envp = { .ptr.native = __envp };
2016         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2017 }
2018
2019 static int do_execveat(int fd, struct filename *filename,
2020                 const char __user *const __user *__argv,
2021                 const char __user *const __user *__envp,
2022                 int flags)
2023 {
2024         struct user_arg_ptr argv = { .ptr.native = __argv };
2025         struct user_arg_ptr envp = { .ptr.native = __envp };
2026
2027         return do_execveat_common(fd, filename, argv, envp, flags);
2028 }
2029
2030 #ifdef CONFIG_COMPAT
2031 static int compat_do_execve(struct filename *filename,
2032         const compat_uptr_t __user *__argv,
2033         const compat_uptr_t __user *__envp)
2034 {
2035         struct user_arg_ptr argv = {
2036                 .is_compat = true,
2037                 .ptr.compat = __argv,
2038         };
2039         struct user_arg_ptr envp = {
2040                 .is_compat = true,
2041                 .ptr.compat = __envp,
2042         };
2043         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2044 }
2045
2046 static int compat_do_execveat(int fd, struct filename *filename,
2047                               const compat_uptr_t __user *__argv,
2048                               const compat_uptr_t __user *__envp,
2049                               int flags)
2050 {
2051         struct user_arg_ptr argv = {
2052                 .is_compat = true,
2053                 .ptr.compat = __argv,
2054         };
2055         struct user_arg_ptr envp = {
2056                 .is_compat = true,
2057                 .ptr.compat = __envp,
2058         };
2059         return do_execveat_common(fd, filename, argv, envp, flags);
2060 }
2061 #endif
2062
2063 void set_binfmt(struct linux_binfmt *new)
2064 {
2065         struct mm_struct *mm = current->mm;
2066
2067         if (mm->binfmt)
2068                 module_put(mm->binfmt->module);
2069
2070         mm->binfmt = new;
2071         if (new)
2072                 __module_get(new->module);
2073 }
2074 EXPORT_SYMBOL(set_binfmt);
2075
2076 /*
2077  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2078  */
2079 void set_dumpable(struct mm_struct *mm, int value)
2080 {
2081         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2082                 return;
2083
2084         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2085 }
2086
2087 SYSCALL_DEFINE3(execve,
2088                 const char __user *, filename,
2089                 const char __user *const __user *, argv,
2090                 const char __user *const __user *, envp)
2091 {
2092         return do_execve(getname(filename), argv, envp);
2093 }
2094
2095 SYSCALL_DEFINE5(execveat,
2096                 int, fd, const char __user *, filename,
2097                 const char __user *const __user *, argv,
2098                 const char __user *const __user *, envp,
2099                 int, flags)
2100 {
2101         return do_execveat(fd,
2102                            getname_uflags(filename, flags),
2103                            argv, envp, flags);
2104 }
2105
2106 #ifdef CONFIG_COMPAT
2107 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2108         const compat_uptr_t __user *, argv,
2109         const compat_uptr_t __user *, envp)
2110 {
2111         return compat_do_execve(getname(filename), argv, envp);
2112 }
2113
2114 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2115                        const char __user *, filename,
2116                        const compat_uptr_t __user *, argv,
2117                        const compat_uptr_t __user *, envp,
2118                        int,  flags)
2119 {
2120         return compat_do_execveat(fd,
2121                                   getname_uflags(filename, flags),
2122                                   argv, envp, flags);
2123 }
2124 #endif
2125
2126 #ifdef CONFIG_SYSCTL
2127
2128 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2129                 void *buffer, size_t *lenp, loff_t *ppos)
2130 {
2131         int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2132
2133         if (!error)
2134                 validate_coredump_safety();
2135         return error;
2136 }
2137
2138 static struct ctl_table fs_exec_sysctls[] = {
2139         {
2140                 .procname       = "suid_dumpable",
2141                 .data           = &suid_dumpable,
2142                 .maxlen         = sizeof(int),
2143                 .mode           = 0644,
2144                 .proc_handler   = proc_dointvec_minmax_coredump,
2145                 .extra1         = SYSCTL_ZERO,
2146                 .extra2         = SYSCTL_TWO,
2147         },
2148         { }
2149 };
2150
2151 static int __init init_fs_exec_sysctls(void)
2152 {
2153         register_sysctl_init("fs", fs_exec_sysctls);
2154         return 0;
2155 }
2156
2157 fs_initcall(init_fs_exec_sysctls);
2158 #endif /* CONFIG_SYSCTL */