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/slab.h>
  27 #include <linux/file.h>
  28 #include <linux/fdtable.h>
  29 #include <linux/mm.h>
  30 #include <linux/vmacache.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/tracehook.h>
  59 #include <linux/kmod.h>
  60 #include <linux/fsnotify.h>
  61 #include <linux/fs_struct.h>
  62 #include <linux/oom.h>
  63 #include <linux/compat.h>
  64 #include <linux/vmalloc.h>
  65
  66 #include <linux/uaccess.h>
  67 #include <asm/mmu_context.h>
  68 #include <asm/tlb.h>
  69
  70 #include <trace/events/task.h>
  71 #include "internal.h"
  72
  73 #include <trace/events/sched.h>
  74
  75 static int bprm_creds_from_file(struct linux_binprm *bprm);
  76
  77 int suid_dumpable = 0;
  78
  79 static LIST_HEAD(formats);
  80 static DEFINE_RWLOCK(binfmt_lock);
  81
  82 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  83 {
  84         BUG_ON(!fmt);
  85         if (WARN_ON(!fmt->load_binary))
  86                 return;
  87         write_lock(&binfmt_lock);
  88         insert ? list_add(&fmt->lh, &formats) :
  89                  list_add_tail(&fmt->lh, &formats);
  90         write_unlock(&binfmt_lock);
  91 }
  92
  93 EXPORT_SYMBOL(__register_binfmt);
  94
  95 void unregister_binfmt(struct linux_binfmt * fmt)
  96 {
  97         write_lock(&binfmt_lock);
  98         list_del(&fmt->lh);
  99         write_unlock(&binfmt_lock);
 100 }
 101
 102 EXPORT_SYMBOL(unregister_binfmt);
 103
 104 static inline void put_binfmt(struct linux_binfmt * fmt)
 105 {
 106         module_put(fmt->module);
 107 }
 108
 109 bool path_noexec(const struct path *path)
 110 {
 111         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 112                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 113 }
 114
 115 #ifdef CONFIG_USELIB
 116 /*
 117  * Note that a shared library must be both readable and executable due to
 118  * security reasons.
 119  *
 120  * Also note that we take the address to load from from the file itself.
 121  */
 122 SYSCALL_DEFINE1(uselib, const char __user *, library)
 123 {
 124         struct linux_binfmt *fmt;
 125         struct file *file;
 126         struct filename *tmp = getname(library);
 127         int error = PTR_ERR(tmp);
 128         static const struct open_flags uselib_flags = {
 129                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 130                 .acc_mode = MAY_READ | MAY_EXEC,
 131                 .intent = LOOKUP_OPEN,
 132                 .lookup_flags = LOOKUP_FOLLOW,
 133         };
 134
 135         if (IS_ERR(tmp))
 136                 goto out;
 137
 138         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 139         putname(tmp);
 140         error = PTR_ERR(file);
 141         if (IS_ERR(file))
 142                 goto out;
 143
 144         error = -EINVAL;
 145         if (!S_ISREG(file_inode(file)->i_mode))
 146                 goto exit;
 147
 148         error = -EACCES;
 149         if (path_noexec(&file->f_path))
 150                 goto exit;
 151
 152         fsnotify_open(file);
 153
 154         error = -ENOEXEC;
 155
 156         read_lock(&binfmt_lock);
 157         list_for_each_entry(fmt, &formats, lh) {
 158                 if (!fmt->load_shlib)
 159                         continue;
 160                 if (!try_module_get(fmt->module))
 161                         continue;
 162                 read_unlock(&binfmt_lock);
 163                 error = fmt->load_shlib(file);
 164                 read_lock(&binfmt_lock);
 165                 put_binfmt(fmt);
 166                 if (error != -ENOEXEC)
 167                         break;
 168         }
 169         read_unlock(&binfmt_lock);
 170 exit:
 171         fput(file);
 172 out:
 173         return error;
 174 }
 175 #endif /* #ifdef CONFIG_USELIB */
 176
 177 #ifdef CONFIG_MMU
 178 /*
 179  * The nascent bprm->mm is not visible until exec_mmap() but it can
 180  * use a lot of memory, account these pages in current->mm temporary
 181  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 182  * change the counter back via acct_arg_size(0).
 183  */
 184 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 185 {
 186         struct mm_struct *mm = current->mm;
 187         long diff = (long)(pages - bprm->vma_pages);
 188
 189         if (!mm || !diff)
 190                 return;
 191
 192         bprm->vma_pages = pages;
 193         add_mm_counter(mm, MM_ANONPAGES, diff);
 194 }
 195
 196 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 197                 int write)
 198 {
 199         struct page *page;
 200         int ret;
 201         unsigned int gup_flags = FOLL_FORCE;
 202
 203 #ifdef CONFIG_STACK_GROWSUP
 204         if (write) {
 205                 ret = expand_downwards(bprm->vma, pos);
 206                 if (ret < 0)
 207                         return NULL;
 208         }
 209 #endif
 210
 211         if (write)
 212                 gup_flags |= FOLL_WRITE;
 213
 214         /*
 215          * We are doing an exec().  'current' is the process
 216          * doing the exec and bprm->mm is the new process's mm.
 217          */
 218         ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 219                         &page, NULL, NULL);
 220         if (ret <= 0)
 221                 return NULL;
 222
 223         if (write)
 224                 acct_arg_size(bprm, vma_pages(bprm->vma));
 225
 226         return page;
 227 }
 228
 229 static void put_arg_page(struct page *page)
 230 {
 231         put_page(page);
 232 }
 233
 234 static void free_arg_pages(struct linux_binprm *bprm)
 235 {
 236 }
 237
 238 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 239                 struct page *page)
 240 {
 241         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 242 }
 243
 244 static int __bprm_mm_init(struct linux_binprm *bprm)
 245 {
 246         int err;
 247         struct vm_area_struct *vma = NULL;
 248         struct mm_struct *mm = bprm->mm;
 249
 250         bprm->vma = vma = vm_area_alloc(mm);
 251         if (!vma)
 252                 return -ENOMEM;
 253         vma_set_anonymous(vma);
 254
 255         if (mmap_write_lock_killable(mm)) {
 256                 err = -EINTR;
 257                 goto err_free;
 258         }
 259
 260         /*
 261          * Place the stack at the largest stack address the architecture
 262          * supports. Later, we'll move this to an appropriate place. We don't
 263          * use STACK_TOP because that can depend on attributes which aren't
 264          * configured yet.
 265          */
 266         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 267         vma->vm_end = STACK_TOP_MAX;
 268         vma->vm_start = vma->vm_end - PAGE_SIZE;
 269         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 270         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 271
 272         err = insert_vm_struct(mm, vma);
 273         if (err)
 274                 goto err;
 275
 276         mm->stack_vm = mm->total_vm = 1;
 277         mmap_write_unlock(mm);
 278         bprm->p = vma->vm_end - sizeof(void *);
 279         return 0;
 280 err:
 281         mmap_write_unlock(mm);
 282 err_free:
 283         bprm->vma = NULL;
 284         vm_area_free(vma);
 285         return err;
 286 }
 287
 288 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 289 {
 290         return len <= MAX_ARG_STRLEN;
 291 }
 292
 293 #else
 294
 295 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 296 {
 297 }
 298
 299 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 300                 int write)
 301 {
 302         struct page *page;
 303
 304         page = bprm->page[pos / PAGE_SIZE];
 305         if (!page && write) {
 306                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 307                 if (!page)
 308                         return NULL;
 309                 bprm->page[pos / PAGE_SIZE] = page;
 310         }
 311
 312         return page;
 313 }
 314
 315 static void put_arg_page(struct page *page)
 316 {
 317 }
 318
 319 static void free_arg_page(struct linux_binprm *bprm, int i)
 320 {
 321         if (bprm->page[i]) {
 322                 __free_page(bprm->page[i]);
 323                 bprm->page[i] = NULL;
 324         }
 325 }
 326
 327 static void free_arg_pages(struct linux_binprm *bprm)
 328 {
 329         int i;
 330
 331         for (i = 0; i < MAX_ARG_PAGES; i++)
 332                 free_arg_page(bprm, i);
 333 }
 334
 335 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 336                 struct page *page)
 337 {
 338 }
 339
 340 static int __bprm_mm_init(struct linux_binprm *bprm)
 341 {
 342         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 343         return 0;
 344 }
 345
 346 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 347 {
 348         return len <= bprm->p;
 349 }
 350
 351 #endif /* CONFIG_MMU */
 352
 353 /*
 354  * Create a new mm_struct and populate it with a temporary stack
 355  * vm_area_struct.  We don't have enough context at this point to set the stack
 356  * flags, permissions, and offset, so we use temporary values.  We'll update
 357  * them later in setup_arg_pages().
 358  */
 359 static int bprm_mm_init(struct linux_binprm *bprm)
 360 {
 361         int err;
 362         struct mm_struct *mm = NULL;
 363
 364         bprm->mm = mm = mm_alloc();
 365         err = -ENOMEM;
 366         if (!mm)
 367                 goto err;
 368
 369         /* Save current stack limit for all calculations made during exec. */
 370         task_lock(current->group_leader);
 371         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 372         task_unlock(current->group_leader);
 373
 374         err = __bprm_mm_init(bprm);
 375         if (err)
 376                 goto err;
 377
 378         return 0;
 379
 380 err:
 381         if (mm) {
 382                 bprm->mm = NULL;
 383                 mmdrop(mm);
 384         }
 385
 386         return err;
 387 }
 388
 389 struct user_arg_ptr {
 390 #ifdef CONFIG_COMPAT
 391         bool is_compat;
 392 #endif
 393         union {
 394                 const char __user *const __user *native;
 395 #ifdef CONFIG_COMPAT
 396                 const compat_uptr_t __user *compat;
 397 #endif
 398         } ptr;
 399 };
 400
 401 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 402 {
 403         const char __user *native;
 404
 405 #ifdef CONFIG_COMPAT
 406         if (unlikely(argv.is_compat)) {
 407                 compat_uptr_t compat;
 408
 409                 if (get_user(compat, argv.ptr.compat + nr))
 410                         return ERR_PTR(-EFAULT);
 411
 412                 return compat_ptr(compat);
 413         }
 414 #endif
 415
 416         if (get_user(native, argv.ptr.native + nr))
 417                 return ERR_PTR(-EFAULT);
 418
 419         return native;
 420 }
 421
 422 /*
 423  * count() counts the number of strings in array ARGV.
 424  */
 425 static int count(struct user_arg_ptr argv, int max)
 426 {
 427         int i = 0;
 428
 429         if (argv.ptr.native != NULL) {
 430                 for (;;) {
 431                         const char __user *p = get_user_arg_ptr(argv, i);
 432
 433                         if (!p)
 434                                 break;
 435
 436                         if (IS_ERR(p))
 437                                 return -EFAULT;
 438
 439                         if (i >= max)
 440                                 return -E2BIG;
 441                         ++i;
 442
 443                         if (fatal_signal_pending(current))
 444                                 return -ERESTARTNOHAND;
 445                         cond_resched();
 446                 }
 447         }
 448         return i;
 449 }
 450
 451 static int count_strings_kernel(const char *const *argv)
 452 {
 453         int i;
 454
 455         if (!argv)
 456                 return 0;
 457
 458         for (i = 0; argv[i]; ++i) {
 459                 if (i >= MAX_ARG_STRINGS)
 460                         return -E2BIG;
 461                 if (fatal_signal_pending(current))
 462                         return -ERESTARTNOHAND;
 463                 cond_resched();
 464         }
 465         return i;
 466 }
 467
 468 static int bprm_stack_limits(struct linux_binprm *bprm)
 469 {
 470         unsigned long limit, ptr_size;
 471
 472         /*
 473          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 474          * (whichever is smaller) for the argv+env strings.
 475          * This ensures that:
 476          *  - the remaining binfmt code will not run out of stack space,
 477          *  - the program will have a reasonable amount of stack left
 478          *    to work from.
 479          */
 480         limit = _STK_LIM / 4 * 3;
 481         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 482         /*
 483          * We've historically supported up to 32 pages (ARG_MAX)
 484          * of argument strings even with small stacks
 485          */
 486         limit = max_t(unsigned long, limit, ARG_MAX);
 487         /*
 488          * We must account for the size of all the argv and envp pointers to
 489          * the argv and envp strings, since they will also take up space in
 490          * the stack. They aren't stored until much later when we can't
 491          * signal to the parent that the child has run out of stack space.
 492          * Instead, calculate it here so it's possible to fail gracefully.
 493          */
 494         ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 495         if (limit <= ptr_size)
 496                 return -E2BIG;
 497         limit -= ptr_size;
 498
 499         bprm->argmin = bprm->p - limit;
 500         return 0;
 501 }
 502
 503 /*
 504  * 'copy_strings()' copies argument/environment strings from the old
 505  * processes's memory to the new process's stack.  The call to get_user_pages()
 506  * ensures the destination page is created and not swapped out.
 507  */
 508 static int copy_strings(int argc, struct user_arg_ptr argv,
 509                         struct linux_binprm *bprm)
 510 {
 511         struct page *kmapped_page = NULL;
 512         char *kaddr = NULL;
 513         unsigned long kpos = 0;
 514         int ret;
 515
 516         while (argc-- > 0) {
 517                 const char __user *str;
 518                 int len;
 519                 unsigned long pos;
 520
 521                 ret = -EFAULT;
 522                 str = get_user_arg_ptr(argv, argc);
 523                 if (IS_ERR(str))
 524                         goto out;
 525
 526                 len = strnlen_user(str, MAX_ARG_STRLEN);
 527                 if (!len)
 528                         goto out;
 529
 530                 ret = -E2BIG;
 531                 if (!valid_arg_len(bprm, len))
 532                         goto out;
 533
 534                 /* We're going to work our way backwords. */
 535                 pos = bprm->p;
 536                 str += len;
 537                 bprm->p -= len;
 538 #ifdef CONFIG_MMU
 539                 if (bprm->p < bprm->argmin)
 540                         goto out;
 541 #endif
 542
 543                 while (len > 0) {
 544                         int offset, bytes_to_copy;
 545
 546                         if (fatal_signal_pending(current)) {
 547                                 ret = -ERESTARTNOHAND;
 548                                 goto out;
 549                         }
 550                         cond_resched();
 551
 552                         offset = pos % PAGE_SIZE;
 553                         if (offset == 0)
 554                                 offset = PAGE_SIZE;
 555
 556                         bytes_to_copy = offset;
 557                         if (bytes_to_copy > len)
 558                                 bytes_to_copy = len;
 559
 560                         offset -= bytes_to_copy;
 561                         pos -= bytes_to_copy;
 562                         str -= bytes_to_copy;
 563                         len -= bytes_to_copy;
 564
 565                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 566                                 struct page *page;
 567
 568                                 page = get_arg_page(bprm, pos, 1);
 569                                 if (!page) {
 570                                         ret = -E2BIG;
 571                                         goto out;
 572                                 }
 573
 574                                 if (kmapped_page) {
 575                                         flush_kernel_dcache_page(kmapped_page);
 576                                         kunmap(kmapped_page);
 577                                         put_arg_page(kmapped_page);
 578                                 }
 579                                 kmapped_page = page;
 580                                 kaddr = kmap(kmapped_page);
 581                                 kpos = pos & PAGE_MASK;
 582                                 flush_arg_page(bprm, kpos, kmapped_page);
 583                         }
 584                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 585                                 ret = -EFAULT;
 586                                 goto out;
 587                         }
 588                 }
 589         }
 590         ret = 0;
 591 out:
 592         if (kmapped_page) {
 593                 flush_kernel_dcache_page(kmapped_page);
 594                 kunmap(kmapped_page);
 595                 put_arg_page(kmapped_page);
 596         }
 597         return ret;
 598 }
 599
 600 /*
 601  * Copy and argument/environment string from the kernel to the processes stack.
 602  */
 603 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 604 {
 605         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 606         unsigned long pos = bprm->p;
 607
 608         if (len == 0)
 609                 return -EFAULT;
 610         if (!valid_arg_len(bprm, len))
 611                 return -E2BIG;
 612
 613         /* We're going to work our way backwards. */
 614         arg += len;
 615         bprm->p -= len;
 616         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 617                 return -E2BIG;
 618
 619         while (len > 0) {
 620                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 621                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 622                 struct page *page;
 623                 char *kaddr;
 624
 625                 pos -= bytes_to_copy;
 626                 arg -= bytes_to_copy;
 627                 len -= bytes_to_copy;
 628
 629                 page = get_arg_page(bprm, pos, 1);
 630                 if (!page)
 631                         return -E2BIG;
 632                 kaddr = kmap_atomic(page);
 633                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 634                 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
 635                 flush_kernel_dcache_page(page);
 636                 kunmap_atomic(kaddr);
 637                 put_arg_page(page);
 638         }
 639
 640         return 0;
 641 }
 642 EXPORT_SYMBOL(copy_string_kernel);
 643
 644 static int copy_strings_kernel(int argc, const char *const *argv,
 645                                struct linux_binprm *bprm)
 646 {
 647         while (argc-- > 0) {
 648                 int ret = copy_string_kernel(argv[argc], bprm);
 649                 if (ret < 0)
 650                         return ret;
 651                 if (fatal_signal_pending(current))
 652                         return -ERESTARTNOHAND;
 653                 cond_resched();
 654         }
 655         return 0;
 656 }
 657
 658 #ifdef CONFIG_MMU
 659
 660 /*
 661  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 662  * the binfmt code determines where the new stack should reside, we shift it to
 663  * its final location.  The process proceeds as follows:
 664  *
 665  * 1) Use shift to calculate the new vma endpoints.
 666  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 667  *    arguments passed to subsequent functions are consistent.
 668  * 3) Move vma's page tables to the new range.
 669  * 4) Free up any cleared pgd range.
 670  * 5) Shrink the vma to cover only the new range.
 671  */
 672 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 673 {
 674         struct mm_struct *mm = vma->vm_mm;
 675         unsigned long old_start = vma->vm_start;
 676         unsigned long old_end = vma->vm_end;
 677         unsigned long length = old_end - old_start;
 678         unsigned long new_start = old_start - shift;
 679         unsigned long new_end = old_end - shift;
 680         struct mmu_gather tlb;
 681
 682         BUG_ON(new_start > new_end);
 683
 684         /*
 685          * ensure there are no vmas between where we want to go
 686          * and where we are
 687          */
 688         if (vma != find_vma(mm, new_start))
 689                 return -EFAULT;
 690
 691         /*
 692          * cover the whole range: [new_start, old_end)
 693          */
 694         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 695                 return -ENOMEM;
 696
 697         /*
 698          * move the page tables downwards, on failure we rely on
 699          * process cleanup to remove whatever mess we made.
 700          */
 701         if (length != move_page_tables(vma, old_start,
 702                                        vma, new_start, length, false))
 703                 return -ENOMEM;
 704
 705         lru_add_drain();
 706         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 707         if (new_end > old_start) {
 708                 /*
 709                  * when the old and new regions overlap clear from new_end.
 710                  */
 711                 free_pgd_range(&tlb, new_end, old_end, new_end,
 712                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 713         } else {
 714                 /*
 715                  * otherwise, clean from old_start; this is done to not touch
 716                  * the address space in [new_end, old_start) some architectures
 717                  * have constraints on va-space that make this illegal (IA64) -
 718                  * for the others its just a little faster.
 719                  */
 720                 free_pgd_range(&tlb, old_start, old_end, new_end,
 721                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 722         }
 723         tlb_finish_mmu(&tlb, old_start, old_end);
 724
 725         /*
 726          * Shrink the vma to just the new range.  Always succeeds.
 727          */
 728         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 729
 730         return 0;
 731 }
 732
 733 /*
 734  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 735  * the stack is optionally relocated, and some extra space is added.
 736  */
 737 int setup_arg_pages(struct linux_binprm *bprm,
 738                     unsigned long stack_top,
 739                     int executable_stack)
 740 {
 741         unsigned long ret;
 742         unsigned long stack_shift;
 743         struct mm_struct *mm = current->mm;
 744         struct vm_area_struct *vma = bprm->vma;
 745         struct vm_area_struct *prev = NULL;
 746         unsigned long vm_flags;
 747         unsigned long stack_base;
 748         unsigned long stack_size;
 749         unsigned long stack_expand;
 750         unsigned long rlim_stack;
 751
 752 #ifdef CONFIG_STACK_GROWSUP
 753         /* Limit stack size */
 754         stack_base = bprm->rlim_stack.rlim_max;
 755         if (stack_base > STACK_SIZE_MAX)
 756                 stack_base = STACK_SIZE_MAX;
 757
 758         /* Add space for stack randomization. */
 759         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 760
 761         /* Make sure we didn't let the argument array grow too large. */
 762         if (vma->vm_end - vma->vm_start > stack_base)
 763                 return -ENOMEM;
 764
 765         stack_base = PAGE_ALIGN(stack_top - stack_base);
 766
 767         stack_shift = vma->vm_start - stack_base;
 768         mm->arg_start = bprm->p - stack_shift;
 769         bprm->p = vma->vm_end - stack_shift;
 770 #else
 771         stack_top = arch_align_stack(stack_top);
 772         stack_top = PAGE_ALIGN(stack_top);
 773
 774         if (unlikely(stack_top < mmap_min_addr) ||
 775             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 776                 return -ENOMEM;
 777
 778         stack_shift = vma->vm_end - stack_top;
 779
 780         bprm->p -= stack_shift;
 781         mm->arg_start = bprm->p;
 782 #endif
 783
 784         if (bprm->loader)
 785                 bprm->loader -= stack_shift;
 786         bprm->exec -= stack_shift;
 787
 788         if (mmap_write_lock_killable(mm))
 789                 return -EINTR;
 790
 791         vm_flags = VM_STACK_FLAGS;
 792
 793         /*
 794          * Adjust stack execute permissions; explicitly enable for
 795          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 796          * (arch default) otherwise.
 797          */
 798         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 799                 vm_flags |= VM_EXEC;
 800         else if (executable_stack == EXSTACK_DISABLE_X)
 801                 vm_flags &= ~VM_EXEC;
 802         vm_flags |= mm->def_flags;
 803         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 804
 805         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 806                         vm_flags);
 807         if (ret)
 808                 goto out_unlock;
 809         BUG_ON(prev != vma);
 810
 811         if (unlikely(vm_flags & VM_EXEC)) {
 812                 pr_warn_once("process '%pD4' started with executable stack\n",
 813                              bprm->file);
 814         }
 815
 816         /* Move stack pages down in memory. */
 817         if (stack_shift) {
 818                 ret = shift_arg_pages(vma, stack_shift);
 819                 if (ret)
 820                         goto out_unlock;
 821         }
 822
 823         /* mprotect_fixup is overkill to remove the temporary stack flags */
 824         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 825
 826         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 827         stack_size = vma->vm_end - vma->vm_start;
 828         /*
 829          * Align this down to a page boundary as expand_stack
 830          * will align it up.
 831          */
 832         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 833 #ifdef CONFIG_STACK_GROWSUP
 834         if (stack_size + stack_expand > rlim_stack)
 835                 stack_base = vma->vm_start + rlim_stack;
 836         else
 837                 stack_base = vma->vm_end + stack_expand;
 838 #else
 839         if (stack_size + stack_expand > rlim_stack)
 840                 stack_base = vma->vm_end - rlim_stack;
 841         else
 842                 stack_base = vma->vm_start - stack_expand;
 843 #endif
 844         current->mm->start_stack = bprm->p;
 845         ret = expand_stack(vma, stack_base);
 846         if (ret)
 847                 ret = -EFAULT;
 848
 849 out_unlock:
 850         mmap_write_unlock(mm);
 851         return ret;
 852 }
 853 EXPORT_SYMBOL(setup_arg_pages);
 854
 855 #else
 856
 857 /*
 858  * Transfer the program arguments and environment from the holding pages
 859  * onto the stack. The provided stack pointer is adjusted accordingly.
 860  */
 861 int transfer_args_to_stack(struct linux_binprm *bprm,
 862                            unsigned long *sp_location)
 863 {
 864         unsigned long index, stop, sp;
 865         int ret = 0;
 866
 867         stop = bprm->p >> PAGE_SHIFT;
 868         sp = *sp_location;
 869
 870         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 871                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 872                 char *src = kmap(bprm->page[index]) + offset;
 873                 sp -= PAGE_SIZE - offset;
 874                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 875                         ret = -EFAULT;
 876                 kunmap(bprm->page[index]);
 877                 if (ret)
 878                         goto out;
 879         }
 880
 881         *sp_location = sp;
 882
 883 out:
 884         return ret;
 885 }
 886 EXPORT_SYMBOL(transfer_args_to_stack);
 887
 888 #endif /* CONFIG_MMU */
 889
 890 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 891 {
 892         struct file *file;
 893         int err;
 894         struct open_flags open_exec_flags = {
 895                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 896                 .acc_mode = MAY_EXEC,
 897                 .intent = LOOKUP_OPEN,
 898                 .lookup_flags = LOOKUP_FOLLOW,
 899         };
 900
 901         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 902                 return ERR_PTR(-EINVAL);
 903         if (flags & AT_SYMLINK_NOFOLLOW)
 904                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 905         if (flags & AT_EMPTY_PATH)
 906                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 907
 908         file = do_filp_open(fd, name, &open_exec_flags);
 909         if (IS_ERR(file))
 910                 goto out;
 911
 912         err = -EACCES;
 913         if (!S_ISREG(file_inode(file)->i_mode))
 914                 goto exit;
 915
 916         if (path_noexec(&file->f_path))
 917                 goto exit;
 918
 919         err = deny_write_access(file);
 920         if (err)
 921                 goto exit;
 922
 923         if (name->name[0] != '\0')
 924                 fsnotify_open(file);
 925
 926 out:
 927         return file;
 928
 929 exit:
 930         fput(file);
 931         return ERR_PTR(err);
 932 }
 933
 934 struct file *open_exec(const char *name)
 935 {
 936         struct filename *filename = getname_kernel(name);
 937         struct file *f = ERR_CAST(filename);
 938
 939         if (!IS_ERR(filename)) {
 940                 f = do_open_execat(AT_FDCWD, filename, 0);
 941                 putname(filename);
 942         }
 943         return f;
 944 }
 945 EXPORT_SYMBOL(open_exec);
 946
 947 int kernel_read_file(struct file *file, void **buf, loff_t *size,
 948                      loff_t max_size, enum kernel_read_file_id id)
 949 {
 950         loff_t i_size, pos;
 951         ssize_t bytes = 0;
 952         int ret;
 953
 954         if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
 955                 return -EINVAL;
 956
 957         ret = deny_write_access(file);
 958         if (ret)
 959                 return ret;
 960
 961         ret = security_kernel_read_file(file, id);
 962         if (ret)
 963                 goto out;
 964
 965         i_size = i_size_read(file_inode(file));
 966         if (i_size <= 0) {
 967                 ret = -EINVAL;
 968                 goto out;
 969         }
 970         if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
 971                 ret = -EFBIG;
 972                 goto out;
 973         }
 974
 975         if (id != READING_FIRMWARE_PREALLOC_BUFFER)
 976                 *buf = vmalloc(i_size);
 977         if (!*buf) {
 978                 ret = -ENOMEM;
 979                 goto out;
 980         }
 981
 982         pos = 0;
 983         while (pos < i_size) {
 984                 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
 985                 if (bytes < 0) {
 986                         ret = bytes;
 987                         goto out_free;
 988                 }
 989
 990                 if (bytes == 0)
 991                         break;
 992         }
 993
 994         if (pos != i_size) {
 995                 ret = -EIO;
 996                 goto out_free;
 997         }
 998
 999         ret = security_kernel_post_read_file(file, *buf, i_size, id);
1000         if (!ret)
1001                 *size = pos;
1002
1003 out_free:
1004         if (ret < 0) {
1005                 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
1006                         vfree(*buf);
1007                         *buf = NULL;
1008                 }
1009         }
1010
1011 out:
1012         allow_write_access(file);
1013         return ret;
1014 }
1015 EXPORT_SYMBOL_GPL(kernel_read_file);
1016
1017 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
1018                                loff_t max_size, enum kernel_read_file_id id)
1019 {
1020         struct file *file;
1021         int ret;
1022
1023         if (!path || !*path)
1024                 return -EINVAL;
1025
1026         file = filp_open(path, O_RDONLY, 0);
1027         if (IS_ERR(file))
1028                 return PTR_ERR(file);
1029
1030         ret = kernel_read_file(file, buf, size, max_size, id);
1031         fput(file);
1032         return ret;
1033 }
1034 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
1035
1036 int kernel_read_file_from_path_initns(const char *path, void **buf,
1037                                       loff_t *size, loff_t max_size,
1038                                       enum kernel_read_file_id id)
1039 {
1040         struct file *file;
1041         struct path root;
1042         int ret;
1043
1044         if (!path || !*path)
1045                 return -EINVAL;
1046
1047         task_lock(&init_task);
1048         get_fs_root(init_task.fs, &root);
1049         task_unlock(&init_task);
1050
1051         file = file_open_root(root.dentry, root.mnt, path, O_RDONLY, 0);
1052         path_put(&root);
1053         if (IS_ERR(file))
1054                 return PTR_ERR(file);
1055
1056         ret = kernel_read_file(file, buf, size, max_size, id);
1057         fput(file);
1058         return ret;
1059 }
1060 EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns);
1061
1062 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
1063                              enum kernel_read_file_id id)
1064 {
1065         struct fd f = fdget(fd);
1066         int ret = -EBADF;
1067
1068         if (!f.file)
1069                 goto out;
1070
1071         ret = kernel_read_file(f.file, buf, size, max_size, id);
1072 out:
1073         fdput(f);
1074         return ret;
1075 }
1076 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1077
1078 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
1079     defined(CONFIG_BINFMT_ELF_FDPIC)
1080 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1081 {
1082         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1083         if (res > 0)
1084                 flush_icache_user_range(addr, addr + len);
1085         return res;
1086 }
1087 EXPORT_SYMBOL(read_code);
1088 #endif
1089
1090 /*
1091  * Maps the mm_struct mm into the current task struct.
1092  * On success, this function returns with the mutex
1093  * exec_update_mutex locked.
1094  */
1095 static int exec_mmap(struct mm_struct *mm)
1096 {
1097         struct task_struct *tsk;
1098         struct mm_struct *old_mm, *active_mm;
1099         int ret;
1100
1101         /* Notify parent that we're no longer interested in the old VM */
1102         tsk = current;
1103         old_mm = current->mm;
1104         exec_mm_release(tsk, old_mm);
1105         if (old_mm)
1106                 sync_mm_rss(old_mm);
1107
1108         ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
1109         if (ret)
1110                 return ret;
1111
1112         if (old_mm) {
1113                 /*
1114                  * Make sure that if there is a core dump in progress
1115                  * for the old mm, we get out and die instead of going
1116                  * through with the exec.  We must hold mmap_lock around
1117                  * checking core_state and changing tsk->mm.
1118                  */
1119                 mmap_read_lock(old_mm);
1120                 if (unlikely(old_mm->core_state)) {
1121                         mmap_read_unlock(old_mm);
1122                         mutex_unlock(&tsk->signal->exec_update_mutex);
1123                         return -EINTR;
1124                 }
1125         }
1126
1127         task_lock(tsk);
1128         active_mm = tsk->active_mm;
1129         membarrier_exec_mmap(mm);
1130         tsk->mm = mm;
1131         tsk->active_mm = mm;
1132         activate_mm(active_mm, mm);
1133         tsk->mm->vmacache_seqnum = 0;
1134         vmacache_flush(tsk);
1135         task_unlock(tsk);
1136         if (old_mm) {
1137                 mmap_read_unlock(old_mm);
1138                 BUG_ON(active_mm != old_mm);
1139                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1140                 mm_update_next_owner(old_mm);
1141                 mmput(old_mm);
1142                 return 0;
1143         }
1144         mmdrop(active_mm);
1145         return 0;
1146 }
1147
1148 static int de_thread(struct task_struct *tsk)
1149 {
1150         struct signal_struct *sig = tsk->signal;
1151         struct sighand_struct *oldsighand = tsk->sighand;
1152         spinlock_t *lock = &oldsighand->siglock;
1153
1154         if (thread_group_empty(tsk))
1155                 goto no_thread_group;
1156
1157         /*
1158          * Kill all other threads in the thread group.
1159          */
1160         spin_lock_irq(lock);
1161         if (signal_group_exit(sig)) {
1162                 /*
1163                  * Another group action in progress, just
1164                  * return so that the signal is processed.
1165                  */
1166                 spin_unlock_irq(lock);
1167                 return -EAGAIN;
1168         }
1169
1170         sig->group_exit_task = tsk;
1171         sig->notify_count = zap_other_threads(tsk);
1172         if (!thread_group_leader(tsk))
1173                 sig->notify_count--;
1174
1175         while (sig->notify_count) {
1176                 __set_current_state(TASK_KILLABLE);
1177                 spin_unlock_irq(lock);
1178                 schedule();
1179                 if (__fatal_signal_pending(tsk))
1180                         goto killed;
1181                 spin_lock_irq(lock);
1182         }
1183         spin_unlock_irq(lock);
1184
1185         /*
1186          * At this point all other threads have exited, all we have to
1187          * do is to wait for the thread group leader to become inactive,
1188          * and to assume its PID:
1189          */
1190         if (!thread_group_leader(tsk)) {
1191                 struct task_struct *leader = tsk->group_leader;
1192
1193                 for (;;) {
1194                         cgroup_threadgroup_change_begin(tsk);
1195                         write_lock_irq(&tasklist_lock);
1196                         /*
1197                          * Do this under tasklist_lock to ensure that
1198                          * exit_notify() can't miss ->group_exit_task
1199                          */
1200                         sig->notify_count = -1;
1201                         if (likely(leader->exit_state))
1202                                 break;
1203                         __set_current_state(TASK_KILLABLE);
1204                         write_unlock_irq(&tasklist_lock);
1205                         cgroup_threadgroup_change_end(tsk);
1206                         schedule();
1207                         if (__fatal_signal_pending(tsk))
1208                                 goto killed;
1209                 }
1210
1211                 /*
1212                  * The only record we have of the real-time age of a
1213                  * process, regardless of execs it's done, is start_time.
1214                  * All the past CPU time is accumulated in signal_struct
1215                  * from sister threads now dead.  But in this non-leader
1216                  * exec, nothing survives from the original leader thread,
1217                  * whose birth marks the true age of this process now.
1218                  * When we take on its identity by switching to its PID, we
1219                  * also take its birthdate (always earlier than our own).
1220                  */
1221                 tsk->start_time = leader->start_time;
1222                 tsk->start_boottime = leader->start_boottime;
1223
1224                 BUG_ON(!same_thread_group(leader, tsk));
1225                 /*
1226                  * An exec() starts a new thread group with the
1227                  * TGID of the previous thread group. Rehash the
1228                  * two threads with a switched PID, and release
1229                  * the former thread group leader:
1230                  */
1231
1232                 /* Become a process group leader with the old leader's pid.
1233                  * The old leader becomes a thread of the this thread group.
1234                  */
1235                 exchange_tids(tsk, leader);
1236                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1237                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1238                 transfer_pid(leader, tsk, PIDTYPE_SID);
1239
1240                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1241                 list_replace_init(&leader->sibling, &tsk->sibling);
1242
1243                 tsk->group_leader = tsk;
1244                 leader->group_leader = tsk;
1245
1246                 tsk->exit_signal = SIGCHLD;
1247                 leader->exit_signal = -1;
1248
1249                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1250                 leader->exit_state = EXIT_DEAD;
1251
1252                 /*
1253                  * We are going to release_task()->ptrace_unlink() silently,
1254                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1255                  * the tracer wont't block again waiting for this thread.
1256                  */
1257                 if (unlikely(leader->ptrace))
1258                         __wake_up_parent(leader, leader->parent);
1259                 write_unlock_irq(&tasklist_lock);
1260                 cgroup_threadgroup_change_end(tsk);
1261
1262                 release_task(leader);
1263         }
1264
1265         sig->group_exit_task = NULL;
1266         sig->notify_count = 0;
1267
1268 no_thread_group:
1269         /* we have changed execution domain */
1270         tsk->exit_signal = SIGCHLD;
1271
1272         BUG_ON(!thread_group_leader(tsk));
1273         return 0;
1274
1275 killed:
1276         /* protects against exit_notify() and __exit_signal() */
1277         read_lock(&tasklist_lock);
1278         sig->group_exit_task = NULL;
1279         sig->notify_count = 0;
1280         read_unlock(&tasklist_lock);
1281         return -EAGAIN;
1282 }
1283
1284
1285 /*
1286  * This function makes sure the current process has its own signal table,
1287  * so that flush_signal_handlers can later reset the handlers without
1288  * disturbing other processes.  (Other processes might share the signal
1289  * table via the CLONE_SIGHAND option to clone().)
1290  */
1291 static int unshare_sighand(struct task_struct *me)
1292 {
1293         struct sighand_struct *oldsighand = me->sighand;
1294
1295         if (refcount_read(&oldsighand->count) != 1) {
1296                 struct sighand_struct *newsighand;
1297                 /*
1298                  * This ->sighand is shared with the CLONE_SIGHAND
1299                  * but not CLONE_THREAD task, switch to the new one.
1300                  */
1301                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1302                 if (!newsighand)
1303                         return -ENOMEM;
1304
1305                 refcount_set(&newsighand->count, 1);
1306                 memcpy(newsighand->action, oldsighand->action,
1307                        sizeof(newsighand->action));
1308
1309                 write_lock_irq(&tasklist_lock);
1310                 spin_lock(&oldsighand->siglock);
1311                 rcu_assign_pointer(me->sighand, newsighand);
1312                 spin_unlock(&oldsighand->siglock);
1313                 write_unlock_irq(&tasklist_lock);
1314
1315                 __cleanup_sighand(oldsighand);
1316         }
1317         return 0;
1318 }
1319
1320 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1321 {
1322         task_lock(tsk);
1323         strncpy(buf, tsk->comm, buf_size);
1324         task_unlock(tsk);
1325         return buf;
1326 }
1327 EXPORT_SYMBOL_GPL(__get_task_comm);
1328
1329 /*
1330  * These functions flushes out all traces of the currently running executable
1331  * so that a new one can be started
1332  */
1333
1334 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1335 {
1336         task_lock(tsk);
1337         trace_task_rename(tsk, buf);
1338         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1339         task_unlock(tsk);
1340         perf_event_comm(tsk, exec);
1341 }
1342
1343 /*
1344  * Calling this is the point of no return. None of the failures will be
1345  * seen by userspace since either the process is already taking a fatal
1346  * signal (via de_thread() or coredump), or will have SEGV raised
1347  * (after exec_mmap()) by search_binary_handler (see below).
1348  */
1349 int begin_new_exec(struct linux_binprm * bprm)
1350 {
1351         struct task_struct *me = current;
1352         int retval;
1353
1354         /* Once we are committed compute the creds */
1355         retval = bprm_creds_from_file(bprm);
1356         if (retval)
1357                 return retval;
1358
1359         /*
1360          * Ensure all future errors are fatal.
1361          */
1362         bprm->point_of_no_return = true;
1363
1364         /*
1365          * Make this the only thread in the thread group.
1366          */
1367         retval = de_thread(me);
1368         if (retval)
1369                 goto out;
1370
1371         /*
1372          * Must be called _before_ exec_mmap() as bprm->mm is
1373          * not visibile until then. This also enables the update
1374          * to be lockless.
1375          */
1376         set_mm_exe_file(bprm->mm, bprm->file);
1377
1378         /* If the binary is not readable then enforce mm->dumpable=0 */
1379         would_dump(bprm, bprm->file);
1380         if (bprm->have_execfd)
1381                 would_dump(bprm, bprm->executable);
1382
1383         /*
1384          * Release all of the old mmap stuff
1385          */
1386         acct_arg_size(bprm, 0);
1387         retval = exec_mmap(bprm->mm);
1388         if (retval)
1389                 goto out;
1390
1391         bprm->mm = NULL;
1392
1393 #ifdef CONFIG_POSIX_TIMERS
1394         exit_itimers(me->signal);
1395         flush_itimer_signals();
1396 #endif
1397
1398         /*
1399          * Make the signal table private.
1400          */
1401         retval = unshare_sighand(me);
1402         if (retval)
1403                 goto out_unlock;
1404
1405         set_fs(USER_DS);
1406         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1407                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1408         flush_thread();
1409         me->personality &= ~bprm->per_clear;
1410
1411         /*
1412          * We have to apply CLOEXEC before we change whether the process is
1413          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1414          * trying to access the should-be-closed file descriptors of a process
1415          * undergoing exec(2).
1416          */
1417         do_close_on_exec(me->files);
1418
1419         if (bprm->secureexec) {
1420                 /* Make sure parent cannot signal privileged process. */
1421                 me->pdeath_signal = 0;
1422
1423                 /*
1424                  * For secureexec, reset the stack limit to sane default to
1425                  * avoid bad behavior from the prior rlimits. This has to
1426                  * happen before arch_pick_mmap_layout(), which examines
1427                  * RLIMIT_STACK, but after the point of no return to avoid
1428                  * needing to clean up the change on failure.
1429                  */
1430                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1431                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1432         }
1433
1434         me->sas_ss_sp = me->sas_ss_size = 0;
1435
1436         /*
1437          * Figure out dumpability. Note that this checking only of current
1438          * is wrong, but userspace depends on it. This should be testing
1439          * bprm->secureexec instead.
1440          */
1441         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1442             !(uid_eq(current_euid(), current_uid()) &&
1443               gid_eq(current_egid(), current_gid())))
1444                 set_dumpable(current->mm, suid_dumpable);
1445         else
1446                 set_dumpable(current->mm, SUID_DUMP_USER);
1447
1448         perf_event_exec();
1449         __set_task_comm(me, kbasename(bprm->filename), true);
1450
1451         /* An exec changes our domain. We are no longer part of the thread
1452            group */
1453         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1454         flush_signal_handlers(me, 0);
1455
1456         /*
1457          * install the new credentials for this executable
1458          */
1459         security_bprm_committing_creds(bprm);
1460
1461         commit_creds(bprm->cred);
1462         bprm->cred = NULL;
1463
1464         /*
1465          * Disable monitoring for regular users
1466          * when executing setuid binaries. Must
1467          * wait until new credentials are committed
1468          * by commit_creds() above
1469          */
1470         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1471                 perf_event_exit_task(me);
1472         /*
1473          * cred_guard_mutex must be held at least to this point to prevent
1474          * ptrace_attach() from altering our determination of the task's
1475          * credentials; any time after this it may be unlocked.
1476          */
1477         security_bprm_committed_creds(bprm);
1478
1479         /* Pass the opened binary to the interpreter. */
1480         if (bprm->have_execfd) {
1481                 retval = get_unused_fd_flags(0);
1482                 if (retval < 0)
1483                         goto out_unlock;
1484                 fd_install(retval, bprm->executable);
1485                 bprm->executable = NULL;
1486                 bprm->execfd = retval;
1487         }
1488         return 0;
1489
1490 out_unlock:
1491         mutex_unlock(&me->signal->exec_update_mutex);
1492 out:
1493         return retval;
1494 }
1495 EXPORT_SYMBOL(begin_new_exec);
1496
1497 void would_dump(struct linux_binprm *bprm, struct file *file)
1498 {
1499         struct inode *inode = file_inode(file);
1500         if (inode_permission(inode, MAY_READ) < 0) {
1501                 struct user_namespace *old, *user_ns;
1502                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1503
1504                 /* Ensure mm->user_ns contains the executable */
1505                 user_ns = old = bprm->mm->user_ns;
1506                 while ((user_ns != &init_user_ns) &&
1507                        !privileged_wrt_inode_uidgid(user_ns, inode))
1508                         user_ns = user_ns->parent;
1509
1510                 if (old != user_ns) {
1511                         bprm->mm->user_ns = get_user_ns(user_ns);
1512                         put_user_ns(old);
1513                 }
1514         }
1515 }
1516 EXPORT_SYMBOL(would_dump);
1517
1518 void setup_new_exec(struct linux_binprm * bprm)
1519 {
1520         /* Setup things that can depend upon the personality */
1521         struct task_struct *me = current;
1522
1523         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1524
1525         arch_setup_new_exec();
1526
1527         /* Set the new mm task size. We have to do that late because it may
1528          * depend on TIF_32BIT which is only updated in flush_thread() on
1529          * some architectures like powerpc
1530          */
1531         me->mm->task_size = TASK_SIZE;
1532         mutex_unlock(&me->signal->exec_update_mutex);
1533         mutex_unlock(&me->signal->cred_guard_mutex);
1534 }
1535 EXPORT_SYMBOL(setup_new_exec);
1536
1537 /* Runs immediately before start_thread() takes over. */
1538 void finalize_exec(struct linux_binprm *bprm)
1539 {
1540         /* Store any stack rlimit changes before starting thread. */
1541         task_lock(current->group_leader);
1542         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1543         task_unlock(current->group_leader);
1544 }
1545 EXPORT_SYMBOL(finalize_exec);
1546
1547 /*
1548  * Prepare credentials and lock ->cred_guard_mutex.
1549  * setup_new_exec() commits the new creds and drops the lock.
1550  * Or, if exec fails before, free_bprm() should release ->cred and
1551  * and unlock.
1552  */
1553 static int prepare_bprm_creds(struct linux_binprm *bprm)
1554 {
1555         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1556                 return -ERESTARTNOINTR;
1557
1558         bprm->cred = prepare_exec_creds();
1559         if (likely(bprm->cred))
1560                 return 0;
1561
1562         mutex_unlock(&current->signal->cred_guard_mutex);
1563         return -ENOMEM;
1564 }
1565
1566 static void free_bprm(struct linux_binprm *bprm)
1567 {
1568         if (bprm->mm) {
1569                 acct_arg_size(bprm, 0);
1570                 mmput(bprm->mm);
1571         }
1572         free_arg_pages(bprm);
1573         if (bprm->cred) {
1574                 mutex_unlock(&current->signal->cred_guard_mutex);
1575                 abort_creds(bprm->cred);
1576         }
1577         if (bprm->file) {
1578                 allow_write_access(bprm->file);
1579                 fput(bprm->file);
1580         }
1581         if (bprm->executable)
1582                 fput(bprm->executable);
1583         /* If a binfmt changed the interp, free it. */
1584         if (bprm->interp != bprm->filename)
1585                 kfree(bprm->interp);
1586         kfree(bprm->fdpath);
1587         kfree(bprm);
1588 }
1589
1590 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1591 {
1592         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1593         int retval = -ENOMEM;
1594         if (!bprm)
1595                 goto out;
1596
1597         if (fd == AT_FDCWD || filename->name[0] == '/') {
1598                 bprm->filename = filename->name;
1599         } else {
1600                 if (filename->name[0] == '\0')
1601                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1602                 else
1603                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1604                                                   fd, filename->name);
1605                 if (!bprm->fdpath)
1606                         goto out_free;
1607
1608                 bprm->filename = bprm->fdpath;
1609         }
1610         bprm->interp = bprm->filename;
1611
1612         retval = bprm_mm_init(bprm);
1613         if (retval)
1614                 goto out_free;
1615         return bprm;
1616
1617 out_free:
1618         free_bprm(bprm);
1619 out:
1620         return ERR_PTR(retval);
1621 }
1622
1623 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1624 {
1625         /* If a binfmt changed the interp, free it first. */
1626         if (bprm->interp != bprm->filename)
1627                 kfree(bprm->interp);
1628         bprm->interp = kstrdup(interp, GFP_KERNEL);
1629         if (!bprm->interp)
1630                 return -ENOMEM;
1631         return 0;
1632 }
1633 EXPORT_SYMBOL(bprm_change_interp);
1634
1635 /*
1636  * determine how safe it is to execute the proposed program
1637  * - the caller must hold ->cred_guard_mutex to protect against
1638  *   PTRACE_ATTACH or seccomp thread-sync
1639  */
1640 static void check_unsafe_exec(struct linux_binprm *bprm)
1641 {
1642         struct task_struct *p = current, *t;
1643         unsigned n_fs;
1644
1645         if (p->ptrace)
1646                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1647
1648         /*
1649          * This isn't strictly necessary, but it makes it harder for LSMs to
1650          * mess up.
1651          */
1652         if (task_no_new_privs(current))
1653                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1654
1655         t = p;
1656         n_fs = 1;
1657         spin_lock(&p->fs->lock);
1658         rcu_read_lock();
1659         while_each_thread(p, t) {
1660                 if (t->fs == p->fs)
1661                         n_fs++;
1662         }
1663         rcu_read_unlock();
1664
1665         if (p->fs->users > n_fs)
1666                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1667         else
1668                 p->fs->in_exec = 1;
1669         spin_unlock(&p->fs->lock);
1670 }
1671
1672 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1673 {
1674         /* Handle suid and sgid on files */
1675         struct inode *inode;
1676         unsigned int mode;
1677         kuid_t uid;
1678         kgid_t gid;
1679
1680         if (!mnt_may_suid(file->f_path.mnt))
1681                 return;
1682
1683         if (task_no_new_privs(current))
1684                 return;
1685
1686         inode = file->f_path.dentry->d_inode;
1687         mode = READ_ONCE(inode->i_mode);
1688         if (!(mode & (S_ISUID|S_ISGID)))
1689                 return;
1690
1691         /* Be careful if suid/sgid is set */
1692         inode_lock(inode);
1693
1694         /* reload atomically mode/uid/gid now that lock held */
1695         mode = inode->i_mode;
1696         uid = inode->i_uid;
1697         gid = inode->i_gid;
1698         inode_unlock(inode);
1699
1700         /* We ignore suid/sgid if there are no mappings for them in the ns */
1701         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1702                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1703                 return;
1704
1705         if (mode & S_ISUID) {
1706                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1707                 bprm->cred->euid = uid;
1708         }
1709
1710         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1711                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1712                 bprm->cred->egid = gid;
1713         }
1714 }
1715
1716 /*
1717  * Compute brpm->cred based upon the final binary.
1718  */
1719 static int bprm_creds_from_file(struct linux_binprm *bprm)
1720 {
1721         /* Compute creds based on which file? */
1722         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1723
1724         bprm_fill_uid(bprm, file);
1725         return security_bprm_creds_from_file(bprm, file);
1726 }
1727
1728 /*
1729  * Fill the binprm structure from the inode.
1730  * Read the first BINPRM_BUF_SIZE bytes
1731  *
1732  * This may be called multiple times for binary chains (scripts for example).
1733  */
1734 static int prepare_binprm(struct linux_binprm *bprm)
1735 {
1736         loff_t pos = 0;
1737
1738         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1739         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1740 }
1741
1742 /*
1743  * Arguments are '\0' separated strings found at the location bprm->p
1744  * points to; chop off the first by relocating brpm->p to right after
1745  * the first '\0' encountered.
1746  */
1747 int remove_arg_zero(struct linux_binprm *bprm)
1748 {
1749         int ret = 0;
1750         unsigned long offset;
1751         char *kaddr;
1752         struct page *page;
1753
1754         if (!bprm->argc)
1755                 return 0;
1756
1757         do {
1758                 offset = bprm->p & ~PAGE_MASK;
1759                 page = get_arg_page(bprm, bprm->p, 0);
1760                 if (!page) {
1761                         ret = -EFAULT;
1762                         goto out;
1763                 }
1764                 kaddr = kmap_atomic(page);
1765
1766                 for (; offset < PAGE_SIZE && kaddr[offset];
1767                                 offset++, bprm->p++)
1768                         ;
1769
1770                 kunmap_atomic(kaddr);
1771                 put_arg_page(page);
1772         } while (offset == PAGE_SIZE);
1773
1774         bprm->p++;
1775         bprm->argc--;
1776         ret = 0;
1777
1778 out:
1779         return ret;
1780 }
1781 EXPORT_SYMBOL(remove_arg_zero);
1782
1783 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1784 /*
1785  * cycle the list of binary formats handler, until one recognizes the image
1786  */
1787 static int search_binary_handler(struct linux_binprm *bprm)
1788 {
1789         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1790         struct linux_binfmt *fmt;
1791         int retval;
1792
1793         retval = prepare_binprm(bprm);
1794         if (retval < 0)
1795                 return retval;
1796
1797         retval = security_bprm_check(bprm);
1798         if (retval)
1799                 return retval;
1800
1801         retval = -ENOENT;
1802  retry:
1803         read_lock(&binfmt_lock);
1804         list_for_each_entry(fmt, &formats, lh) {
1805                 if (!try_module_get(fmt->module))
1806                         continue;
1807                 read_unlock(&binfmt_lock);
1808
1809                 retval = fmt->load_binary(bprm);
1810
1811                 read_lock(&binfmt_lock);
1812                 put_binfmt(fmt);
1813                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1814                         read_unlock(&binfmt_lock);
1815                         return retval;
1816                 }
1817         }
1818         read_unlock(&binfmt_lock);
1819
1820         if (need_retry) {
1821                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1822                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1823                         return retval;
1824                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1825                         return retval;
1826                 need_retry = false;
1827                 goto retry;
1828         }
1829
1830         return retval;
1831 }
1832
1833 static int exec_binprm(struct linux_binprm *bprm)
1834 {
1835         pid_t old_pid, old_vpid;
1836         int ret, depth;
1837
1838         /* Need to fetch pid before load_binary changes it */
1839         old_pid = current->pid;
1840         rcu_read_lock();
1841         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1842         rcu_read_unlock();
1843
1844         /* This allows 4 levels of binfmt rewrites before failing hard. */
1845         for (depth = 0;; depth++) {
1846                 struct file *exec;
1847                 if (depth > 5)
1848                         return -ELOOP;
1849
1850                 ret = search_binary_handler(bprm);
1851                 if (ret < 0)
1852                         return ret;
1853                 if (!bprm->interpreter)
1854                         break;
1855
1856                 exec = bprm->file;
1857                 bprm->file = bprm->interpreter;
1858                 bprm->interpreter = NULL;
1859
1860                 allow_write_access(exec);
1861                 if (unlikely(bprm->have_execfd)) {
1862                         if (bprm->executable) {
1863                                 fput(exec);
1864                                 return -ENOEXEC;
1865                         }
1866                         bprm->executable = exec;
1867                 } else
1868                         fput(exec);
1869         }
1870
1871         audit_bprm(bprm);
1872         trace_sched_process_exec(current, old_pid, bprm);
1873         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1874         proc_exec_connector(current);
1875         return 0;
1876 }
1877
1878 /*
1879  * sys_execve() executes a new program.
1880  */
1881 static int bprm_execve(struct linux_binprm *bprm,
1882                        int fd, struct filename *filename, int flags)
1883 {
1884         struct file *file;
1885         struct files_struct *displaced;
1886         int retval;
1887
1888         retval = unshare_files(&displaced);
1889         if (retval)
1890                 return retval;
1891
1892         retval = prepare_bprm_creds(bprm);
1893         if (retval)
1894                 goto out_files;
1895
1896         check_unsafe_exec(bprm);
1897         current->in_execve = 1;
1898
1899         file = do_open_execat(fd, filename, flags);
1900         retval = PTR_ERR(file);
1901         if (IS_ERR(file))
1902                 goto out_unmark;
1903
1904         sched_exec();
1905
1906         bprm->file = file;
1907         /*
1908          * Record that a name derived from an O_CLOEXEC fd will be
1909          * inaccessible after exec. Relies on having exclusive access to
1910          * current->files (due to unshare_files above).
1911          */
1912         if (bprm->fdpath &&
1913             close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1914                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1915
1916         /* Set the unchanging part of bprm->cred */
1917         retval = security_bprm_creds_for_exec(bprm);
1918         if (retval)
1919                 goto out;
1920
1921         retval = exec_binprm(bprm);
1922         if (retval < 0)
1923                 goto out;
1924
1925         /* execve succeeded */
1926         current->fs->in_exec = 0;
1927         current->in_execve = 0;
1928         rseq_execve(current);
1929         acct_update_integrals(current);
1930         task_numa_free(current, false);
1931         if (displaced)
1932                 put_files_struct(displaced);
1933         return retval;
1934
1935 out:
1936         /*
1937          * If past the point of no return ensure the the code never
1938          * returns to the userspace process.  Use an existing fatal
1939          * signal if present otherwise terminate the process with
1940          * SIGSEGV.
1941          */
1942         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1943                 force_sigsegv(SIGSEGV);
1944
1945 out_unmark:
1946         current->fs->in_exec = 0;
1947         current->in_execve = 0;
1948
1949 out_files:
1950         if (displaced)
1951                 reset_files_struct(displaced);
1952
1953         return retval;
1954 }
1955
1956 static int do_execveat_common(int fd, struct filename *filename,
1957                               struct user_arg_ptr argv,
1958                               struct user_arg_ptr envp,
1959                               int flags)
1960 {
1961         struct linux_binprm *bprm;
1962         int retval;
1963
1964         if (IS_ERR(filename))
1965                 return PTR_ERR(filename);
1966
1967         /*
1968          * We move the actual failure in case of RLIMIT_NPROC excess from
1969          * set*uid() to execve() because too many poorly written programs
1970          * don't check setuid() return code.  Here we additionally recheck
1971          * whether NPROC limit is still exceeded.
1972          */
1973         if ((current->flags & PF_NPROC_EXCEEDED) &&
1974             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1975                 retval = -EAGAIN;
1976                 goto out_ret;
1977         }
1978
1979         /* We're below the limit (still or again), so we don't want to make
1980          * further execve() calls fail. */
1981         current->flags &= ~PF_NPROC_EXCEEDED;
1982
1983         bprm = alloc_bprm(fd, filename);
1984         if (IS_ERR(bprm)) {
1985                 retval = PTR_ERR(bprm);
1986                 goto out_ret;
1987         }
1988
1989         retval = count(argv, MAX_ARG_STRINGS);
1990         if (retval < 0)
1991                 goto out_free;
1992         bprm->argc = retval;
1993
1994         retval = count(envp, MAX_ARG_STRINGS);
1995         if (retval < 0)
1996                 goto out_free;
1997         bprm->envc = retval;
1998
1999         retval = bprm_stack_limits(bprm);
2000         if (retval < 0)
2001                 goto out_free;
2002
2003         retval = copy_string_kernel(bprm->filename, bprm);
2004         if (retval < 0)
2005                 goto out_free;
2006         bprm->exec = bprm->p;
2007
2008         retval = copy_strings(bprm->envc, envp, bprm);
2009         if (retval < 0)
2010                 goto out_free;
2011
2012         retval = copy_strings(bprm->argc, argv, bprm);
2013         if (retval < 0)
2014                 goto out_free;
2015
2016         retval = bprm_execve(bprm, fd, filename, flags);
2017 out_free:
2018         free_bprm(bprm);
2019
2020 out_ret:
2021         putname(filename);
2022         return retval;
2023 }
2024
2025 int kernel_execve(const char *kernel_filename,
2026                   const char *const *argv, const char *const *envp)
2027 {
2028         struct filename *filename;
2029         struct linux_binprm *bprm;
2030         int fd = AT_FDCWD;
2031         int retval;
2032
2033         filename = getname_kernel(kernel_filename);
2034         if (IS_ERR(filename))
2035                 return PTR_ERR(filename);
2036
2037         bprm = alloc_bprm(fd, filename);
2038         if (IS_ERR(bprm)) {
2039                 retval = PTR_ERR(bprm);
2040                 goto out_ret;
2041         }
2042
2043         retval = count_strings_kernel(argv);
2044         if (retval < 0)
2045                 goto out_free;
2046         bprm->argc = retval;
2047
2048         retval = count_strings_kernel(envp);
2049         if (retval < 0)
2050                 goto out_free;
2051         bprm->envc = retval;
2052
2053         retval = bprm_stack_limits(bprm);
2054         if (retval < 0)
2055                 goto out_free;
2056
2057         retval = copy_string_kernel(bprm->filename, bprm);
2058         if (retval < 0)
2059                 goto out_free;
2060         bprm->exec = bprm->p;
2061
2062         retval = copy_strings_kernel(bprm->envc, envp, bprm);
2063         if (retval < 0)
2064                 goto out_free;
2065
2066         retval = copy_strings_kernel(bprm->argc, argv, bprm);
2067         if (retval < 0)
2068                 goto out_free;
2069
2070         retval = bprm_execve(bprm, fd, filename, 0);
2071 out_free:
2072         free_bprm(bprm);
2073 out_ret:
2074         putname(filename);
2075         return retval;
2076 }
2077
2078 static int do_execve(struct filename *filename,
2079         const char __user *const __user *__argv,
2080         const char __user *const __user *__envp)
2081 {
2082         struct user_arg_ptr argv = { .ptr.native = __argv };
2083         struct user_arg_ptr envp = { .ptr.native = __envp };
2084         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2085 }
2086
2087 static int do_execveat(int fd, struct filename *filename,
2088                 const char __user *const __user *__argv,
2089                 const char __user *const __user *__envp,
2090                 int flags)
2091 {
2092         struct user_arg_ptr argv = { .ptr.native = __argv };
2093         struct user_arg_ptr envp = { .ptr.native = __envp };
2094
2095         return do_execveat_common(fd, filename, argv, envp, flags);
2096 }
2097
2098 #ifdef CONFIG_COMPAT
2099 static int compat_do_execve(struct filename *filename,
2100         const compat_uptr_t __user *__argv,
2101         const compat_uptr_t __user *__envp)
2102 {
2103         struct user_arg_ptr argv = {
2104                 .is_compat = true,
2105                 .ptr.compat = __argv,
2106         };
2107         struct user_arg_ptr envp = {
2108                 .is_compat = true,
2109                 .ptr.compat = __envp,
2110         };
2111         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2112 }
2113
2114 static int compat_do_execveat(int fd, struct filename *filename,
2115                               const compat_uptr_t __user *__argv,
2116                               const compat_uptr_t __user *__envp,
2117                               int flags)
2118 {
2119         struct user_arg_ptr argv = {
2120                 .is_compat = true,
2121                 .ptr.compat = __argv,
2122         };
2123         struct user_arg_ptr envp = {
2124                 .is_compat = true,
2125                 .ptr.compat = __envp,
2126         };
2127         return do_execveat_common(fd, filename, argv, envp, flags);
2128 }
2129 #endif
2130
2131 void set_binfmt(struct linux_binfmt *new)
2132 {
2133         struct mm_struct *mm = current->mm;
2134
2135         if (mm->binfmt)
2136                 module_put(mm->binfmt->module);
2137
2138         mm->binfmt = new;
2139         if (new)
2140                 __module_get(new->module);
2141 }
2142 EXPORT_SYMBOL(set_binfmt);
2143
2144 /*
2145  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2146  */
2147 void set_dumpable(struct mm_struct *mm, int value)
2148 {
2149         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2150                 return;
2151
2152         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2153 }
2154
2155 SYSCALL_DEFINE3(execve,
2156                 const char __user *, filename,
2157                 const char __user *const __user *, argv,
2158                 const char __user *const __user *, envp)
2159 {
2160         return do_execve(getname(filename), argv, envp);
2161 }
2162
2163 SYSCALL_DEFINE5(execveat,
2164                 int, fd, const char __user *, filename,
2165                 const char __user *const __user *, argv,
2166                 const char __user *const __user *, envp,
2167                 int, flags)
2168 {
2169         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2170
2171         return do_execveat(fd,
2172                            getname_flags(filename, lookup_flags, NULL),
2173                            argv, envp, flags);
2174 }
2175
2176 #ifdef CONFIG_COMPAT
2177 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2178         const compat_uptr_t __user *, argv,
2179         const compat_uptr_t __user *, envp)
2180 {
2181         return compat_do_execve(getname(filename), argv, envp);
2182 }
2183
2184 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2185                        const char __user *, filename,
2186                        const compat_uptr_t __user *, argv,
2187                        const compat_uptr_t __user *, envp,
2188                        int,  flags)
2189 {
2190         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2191
2192         return compat_do_execveat(fd,
2193                                   getname_flags(filename, lookup_flags, NULL),
2194                                   argv, envp, flags);
2195 }
2196 #endif