fs/exec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/exec.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  * #!-checking implemented by tytso.
  10  */
  11 /*
  12  * Demand-loading implemented 01.12.91 - no need to read anything but
  13  * the header into memory. The inode of the executable is put into
  14  * "current->executable", and page faults do the actual loading. Clean.
  15  *
  16  * Once more I can proudly say that linux stood up to being changed: it
  17  * was less than 2 hours work to get demand-loading completely implemented.
  18  *
  19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20  * current->executable is only used by the procfs.  This allows a dispatch
  21  * table to check for several different types  of binary formats.  We keep
  22  * trying until we recognize the file or we run out of supported binary
  23  * formats.
  24  */
  25
  26 #include <linux/kernel_read_file.h>
  27 #include <linux/slab.h>
  28 #include <linux/file.h>
  29 #include <linux/fdtable.h>
  30 #include <linux/mm.h>
  31 #include <linux/vmacache.h>
  32 #include <linux/stat.h>
  33 #include <linux/fcntl.h>
  34 #include <linux/swap.h>
  35 #include <linux/string.h>
  36 #include <linux/init.h>
  37 #include <linux/sched/mm.h>
  38 #include <linux/sched/coredump.h>
  39 #include <linux/sched/signal.h>
  40 #include <linux/sched/numa_balancing.h>
  41 #include <linux/sched/task.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/perf_event.h>
  44 #include <linux/highmem.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/key.h>
  47 #include <linux/personality.h>
  48 #include <linux/binfmts.h>
  49 #include <linux/utsname.h>
  50 #include <linux/pid_namespace.h>
  51 #include <linux/module.h>
  52 #include <linux/namei.h>
  53 #include <linux/mount.h>
  54 #include <linux/security.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/audit.h>
  59 #include <linux/tracehook.h>
  60 #include <linux/kmod.h>
  61 #include <linux/fsnotify.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/oom.h>
  64 #include <linux/compat.h>
  65 #include <linux/vmalloc.h>
  66 #include <linux/io_uring.h>
  67
  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         BUG_ON(!fmt);
  87         if (WARN_ON(!fmt->load_binary))
  88                 return;
  89         write_lock(&binfmt_lock);
  90         insert ? list_add(&fmt->lh, &formats) :
  91                  list_add_tail(&fmt->lh, &formats);
  92         write_unlock(&binfmt_lock);
  93 }
  94
  95 EXPORT_SYMBOL(__register_binfmt);
  96
  97 void unregister_binfmt(struct linux_binfmt * fmt)
  98 {
  99         write_lock(&binfmt_lock);
 100         list_del(&fmt->lh);
 101         write_unlock(&binfmt_lock);
 102 }
 103
 104 EXPORT_SYMBOL(unregister_binfmt);
 105
 106 static inline void put_binfmt(struct linux_binfmt * fmt)
 107 {
 108         module_put(fmt->module);
 109 }
 110
 111 bool path_noexec(const struct path *path)
 112 {
 113         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 114                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 115 }
 116
 117 #ifdef CONFIG_USELIB
 118 /*
 119  * Note that a shared library must be both readable and executable due to
 120  * security reasons.
 121  *
 122  * Also note that we take the address to load from from the file itself.
 123  */
 124 SYSCALL_DEFINE1(uselib, const char __user *, library)
 125 {
 126         struct linux_binfmt *fmt;
 127         struct file *file;
 128         struct filename *tmp = getname(library);
 129         int error = PTR_ERR(tmp);
 130         static const struct open_flags uselib_flags = {
 131                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 132                 .acc_mode = MAY_READ | MAY_EXEC,
 133                 .intent = LOOKUP_OPEN,
 134                 .lookup_flags = LOOKUP_FOLLOW,
 135         };
 136
 137         if (IS_ERR(tmp))
 138                 goto out;
 139
 140         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
 141         putname(tmp);
 142         error = PTR_ERR(file);
 143         if (IS_ERR(file))
 144                 goto out;
 145
 146         /*
 147          * may_open() has already checked for this, so it should be
 148          * impossible to trip now. But we need to be extra cautious
 149          * and check again at the very end too.
 150          */
 151         error = -EACCES;
 152         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 153                          path_noexec(&file->f_path)))
 154                 goto exit;
 155
 156         fsnotify_open(file);
 157
 158         error = -ENOEXEC;
 159
 160         read_lock(&binfmt_lock);
 161         list_for_each_entry(fmt, &formats, lh) {
 162                 if (!fmt->load_shlib)
 163                         continue;
 164                 if (!try_module_get(fmt->module))
 165                         continue;
 166                 read_unlock(&binfmt_lock);
 167                 error = fmt->load_shlib(file);
 168                 read_lock(&binfmt_lock);
 169                 put_binfmt(fmt);
 170                 if (error != -ENOEXEC)
 171                         break;
 172         }
 173         read_unlock(&binfmt_lock);
 174 exit:
 175         fput(file);
 176 out:
 177         return error;
 178 }
 179 #endif /* #ifdef CONFIG_USELIB */
 180
 181 #ifdef CONFIG_MMU
 182 /*
 183  * The nascent bprm->mm is not visible until exec_mmap() but it can
 184  * use a lot of memory, account these pages in current->mm temporary
 185  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 186  * change the counter back via acct_arg_size(0).
 187  */
 188 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 189 {
 190         struct mm_struct *mm = current->mm;
 191         long diff = (long)(pages - bprm->vma_pages);
 192
 193         if (!mm || !diff)
 194                 return;
 195
 196         bprm->vma_pages = pages;
 197         add_mm_counter(mm, MM_ANONPAGES, diff);
 198 }
 199
 200 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 201                 int write)
 202 {
 203         struct page *page;
 204         int ret;
 205         unsigned int gup_flags = FOLL_FORCE;
 206
 207 #ifdef CONFIG_STACK_GROWSUP
 208         if (write) {
 209                 ret = expand_downwards(bprm->vma, pos);
 210                 if (ret < 0)
 211                         return NULL;
 212         }
 213 #endif
 214
 215         if (write)
 216                 gup_flags |= FOLL_WRITE;
 217
 218         /*
 219          * We are doing an exec().  'current' is the process
 220          * doing the exec and bprm->mm is the new process's mm.
 221          */
 222         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 223                         &page, NULL, NULL);
 224         if (ret <= 0)
 225                 return NULL;
 226
 227         if (write)
 228                 acct_arg_size(bprm, vma_pages(bprm->vma));
 229
 230         return page;
 231 }
 232
 233 static void put_arg_page(struct page *page)
 234 {
 235         put_page(page);
 236 }
 237
 238 static void free_arg_pages(struct linux_binprm *bprm)
 239 {
 240 }
 241
 242 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 243                 struct page *page)
 244 {
 245         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
 246 }
 247
 248 static int __bprm_mm_init(struct linux_binprm *bprm)
 249 {
 250         int err;
 251         struct vm_area_struct *vma = NULL;
 252         struct mm_struct *mm = bprm->mm;
 253
 254         bprm->vma = vma = vm_area_alloc(mm);
 255         if (!vma)
 256                 return -ENOMEM;
 257         vma_set_anonymous(vma);
 258
 259         if (mmap_write_lock_killable(mm)) {
 260                 err = -EINTR;
 261                 goto err_free;
 262         }
 263
 264         /*
 265          * Place the stack at the largest stack address the architecture
 266          * supports. Later, we'll move this to an appropriate place. We don't
 267          * use STACK_TOP because that can depend on attributes which aren't
 268          * configured yet.
 269          */
 270         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
 271         vma->vm_end = STACK_TOP_MAX;
 272         vma->vm_start = vma->vm_end - PAGE_SIZE;
 273         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 274         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
 275
 276         err = insert_vm_struct(mm, vma);
 277         if (err)
 278                 goto err;
 279
 280         mm->stack_vm = mm->total_vm = 1;
 281         mmap_write_unlock(mm);
 282         bprm->p = vma->vm_end - sizeof(void *);
 283         return 0;
 284 err:
 285         mmap_write_unlock(mm);
 286 err_free:
 287         bprm->vma = NULL;
 288         vm_area_free(vma);
 289         return err;
 290 }
 291
 292 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 293 {
 294         return len <= MAX_ARG_STRLEN;
 295 }
 296
 297 #else
 298
 299 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 300 {
 301 }
 302
 303 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 304                 int write)
 305 {
 306         struct page *page;
 307
 308         page = bprm->page[pos / PAGE_SIZE];
 309         if (!page && write) {
 310                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
 311                 if (!page)
 312                         return NULL;
 313                 bprm->page[pos / PAGE_SIZE] = page;
 314         }
 315
 316         return page;
 317 }
 318
 319 static void put_arg_page(struct page *page)
 320 {
 321 }
 322
 323 static void free_arg_page(struct linux_binprm *bprm, int i)
 324 {
 325         if (bprm->page[i]) {
 326                 __free_page(bprm->page[i]);
 327                 bprm->page[i] = NULL;
 328         }
 329 }
 330
 331 static void free_arg_pages(struct linux_binprm *bprm)
 332 {
 333         int i;
 334
 335         for (i = 0; i < MAX_ARG_PAGES; i++)
 336                 free_arg_page(bprm, i);
 337 }
 338
 339 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
 340                 struct page *page)
 341 {
 342 }
 343
 344 static int __bprm_mm_init(struct linux_binprm *bprm)
 345 {
 346         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
 347         return 0;
 348 }
 349
 350 static bool valid_arg_len(struct linux_binprm *bprm, long len)
 351 {
 352         return len <= bprm->p;
 353 }
 354
 355 #endif /* CONFIG_MMU */
 356
 357 /*
 358  * Create a new mm_struct and populate it with a temporary stack
 359  * vm_area_struct.  We don't have enough context at this point to set the stack
 360  * flags, permissions, and offset, so we use temporary values.  We'll update
 361  * them later in setup_arg_pages().
 362  */
 363 static int bprm_mm_init(struct linux_binprm *bprm)
 364 {
 365         int err;
 366         struct mm_struct *mm = NULL;
 367
 368         bprm->mm = mm = mm_alloc();
 369         err = -ENOMEM;
 370         if (!mm)
 371                 goto err;
 372
 373         /* Save current stack limit for all calculations made during exec. */
 374         task_lock(current->group_leader);
 375         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
 376         task_unlock(current->group_leader);
 377
 378         err = __bprm_mm_init(bprm);
 379         if (err)
 380                 goto err;
 381
 382         return 0;
 383
 384 err:
 385         if (mm) {
 386                 bprm->mm = NULL;
 387                 mmdrop(mm);
 388         }
 389
 390         return err;
 391 }
 392
 393 struct user_arg_ptr {
 394 #ifdef CONFIG_COMPAT
 395         bool is_compat;
 396 #endif
 397         union {
 398                 const char __user *const __user *native;
 399 #ifdef CONFIG_COMPAT
 400                 const compat_uptr_t __user *compat;
 401 #endif
 402         } ptr;
 403 };
 404
 405 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
 406 {
 407         const char __user *native;
 408
 409 #ifdef CONFIG_COMPAT
 410         if (unlikely(argv.is_compat)) {
 411                 compat_uptr_t compat;
 412
 413                 if (get_user(compat, argv.ptr.compat + nr))
 414                         return ERR_PTR(-EFAULT);
 415
 416                 return compat_ptr(compat);
 417         }
 418 #endif
 419
 420         if (get_user(native, argv.ptr.native + nr))
 421                 return ERR_PTR(-EFAULT);
 422
 423         return native;
 424 }
 425
 426 /*
 427  * count() counts the number of strings in array ARGV.
 428  */
 429 static int count(struct user_arg_ptr argv, int max)
 430 {
 431         int i = 0;
 432
 433         if (argv.ptr.native != NULL) {
 434                 for (;;) {
 435                         const char __user *p = get_user_arg_ptr(argv, i);
 436
 437                         if (!p)
 438                                 break;
 439
 440                         if (IS_ERR(p))
 441                                 return -EFAULT;
 442
 443                         if (i >= max)
 444                                 return -E2BIG;
 445                         ++i;
 446
 447                         if (fatal_signal_pending(current))
 448                                 return -ERESTARTNOHAND;
 449                         cond_resched();
 450                 }
 451         }
 452         return i;
 453 }
 454
 455 static int count_strings_kernel(const char *const *argv)
 456 {
 457         int i;
 458
 459         if (!argv)
 460                 return 0;
 461
 462         for (i = 0; argv[i]; ++i) {
 463                 if (i >= MAX_ARG_STRINGS)
 464                         return -E2BIG;
 465                 if (fatal_signal_pending(current))
 466                         return -ERESTARTNOHAND;
 467                 cond_resched();
 468         }
 469         return i;
 470 }
 471
 472 static int bprm_stack_limits(struct linux_binprm *bprm)
 473 {
 474         unsigned long limit, ptr_size;
 475
 476         /*
 477          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
 478          * (whichever is smaller) for the argv+env strings.
 479          * This ensures that:
 480          *  - the remaining binfmt code will not run out of stack space,
 481          *  - the program will have a reasonable amount of stack left
 482          *    to work from.
 483          */
 484         limit = _STK_LIM / 4 * 3;
 485         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
 486         /*
 487          * We've historically supported up to 32 pages (ARG_MAX)
 488          * of argument strings even with small stacks
 489          */
 490         limit = max_t(unsigned long, limit, ARG_MAX);
 491         /*
 492          * We must account for the size of all the argv and envp pointers to
 493          * the argv and envp strings, since they will also take up space in
 494          * the stack. They aren't stored until much later when we can't
 495          * signal to the parent that the child has run out of stack space.
 496          * Instead, calculate it here so it's possible to fail gracefully.
 497          */
 498         ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
 499         if (limit <= ptr_size)
 500                 return -E2BIG;
 501         limit -= ptr_size;
 502
 503         bprm->argmin = bprm->p - limit;
 504         return 0;
 505 }
 506
 507 /*
 508  * 'copy_strings()' copies argument/environment strings from the old
 509  * processes's memory to the new process's stack.  The call to get_user_pages()
 510  * ensures the destination page is created and not swapped out.
 511  */
 512 static int copy_strings(int argc, struct user_arg_ptr argv,
 513                         struct linux_binprm *bprm)
 514 {
 515         struct page *kmapped_page = NULL;
 516         char *kaddr = NULL;
 517         unsigned long kpos = 0;
 518         int ret;
 519
 520         while (argc-- > 0) {
 521                 const char __user *str;
 522                 int len;
 523                 unsigned long pos;
 524
 525                 ret = -EFAULT;
 526                 str = get_user_arg_ptr(argv, argc);
 527                 if (IS_ERR(str))
 528                         goto out;
 529
 530                 len = strnlen_user(str, MAX_ARG_STRLEN);
 531                 if (!len)
 532                         goto out;
 533
 534                 ret = -E2BIG;
 535                 if (!valid_arg_len(bprm, len))
 536                         goto out;
 537
 538                 /* We're going to work our way backwords. */
 539                 pos = bprm->p;
 540                 str += len;
 541                 bprm->p -= len;
 542 #ifdef CONFIG_MMU
 543                 if (bprm->p < bprm->argmin)
 544                         goto out;
 545 #endif
 546
 547                 while (len > 0) {
 548                         int offset, bytes_to_copy;
 549
 550                         if (fatal_signal_pending(current)) {
 551                                 ret = -ERESTARTNOHAND;
 552                                 goto out;
 553                         }
 554                         cond_resched();
 555
 556                         offset = pos % PAGE_SIZE;
 557                         if (offset == 0)
 558                                 offset = PAGE_SIZE;
 559
 560                         bytes_to_copy = offset;
 561                         if (bytes_to_copy > len)
 562                                 bytes_to_copy = len;
 563
 564                         offset -= bytes_to_copy;
 565                         pos -= bytes_to_copy;
 566                         str -= bytes_to_copy;
 567                         len -= bytes_to_copy;
 568
 569                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
 570                                 struct page *page;
 571
 572                                 page = get_arg_page(bprm, pos, 1);
 573                                 if (!page) {
 574                                         ret = -E2BIG;
 575                                         goto out;
 576                                 }
 577
 578                                 if (kmapped_page) {
 579                                         flush_kernel_dcache_page(kmapped_page);
 580                                         kunmap(kmapped_page);
 581                                         put_arg_page(kmapped_page);
 582                                 }
 583                                 kmapped_page = page;
 584                                 kaddr = kmap(kmapped_page);
 585                                 kpos = pos & PAGE_MASK;
 586                                 flush_arg_page(bprm, kpos, kmapped_page);
 587                         }
 588                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
 589                                 ret = -EFAULT;
 590                                 goto out;
 591                         }
 592                 }
 593         }
 594         ret = 0;
 595 out:
 596         if (kmapped_page) {
 597                 flush_kernel_dcache_page(kmapped_page);
 598                 kunmap(kmapped_page);
 599                 put_arg_page(kmapped_page);
 600         }
 601         return ret;
 602 }
 603
 604 /*
 605  * Copy and argument/environment string from the kernel to the processes stack.
 606  */
 607 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
 608 {
 609         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
 610         unsigned long pos = bprm->p;
 611
 612         if (len == 0)
 613                 return -EFAULT;
 614         if (!valid_arg_len(bprm, len))
 615                 return -E2BIG;
 616
 617         /* We're going to work our way backwards. */
 618         arg += len;
 619         bprm->p -= len;
 620         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
 621                 return -E2BIG;
 622
 623         while (len > 0) {
 624                 unsigned int bytes_to_copy = min_t(unsigned int, len,
 625                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
 626                 struct page *page;
 627                 char *kaddr;
 628
 629                 pos -= bytes_to_copy;
 630                 arg -= bytes_to_copy;
 631                 len -= bytes_to_copy;
 632
 633                 page = get_arg_page(bprm, pos, 1);
 634                 if (!page)
 635                         return -E2BIG;
 636                 kaddr = kmap_atomic(page);
 637                 flush_arg_page(bprm, pos & PAGE_MASK, page);
 638                 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
 639                 flush_kernel_dcache_page(page);
 640                 kunmap_atomic(kaddr);
 641                 put_arg_page(page);
 642         }
 643
 644         return 0;
 645 }
 646 EXPORT_SYMBOL(copy_string_kernel);
 647
 648 static int copy_strings_kernel(int argc, const char *const *argv,
 649                                struct linux_binprm *bprm)
 650 {
 651         while (argc-- > 0) {
 652                 int ret = copy_string_kernel(argv[argc], bprm);
 653                 if (ret < 0)
 654                         return ret;
 655                 if (fatal_signal_pending(current))
 656                         return -ERESTARTNOHAND;
 657                 cond_resched();
 658         }
 659         return 0;
 660 }
 661
 662 #ifdef CONFIG_MMU
 663
 664 /*
 665  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
 666  * the binfmt code determines where the new stack should reside, we shift it to
 667  * its final location.  The process proceeds as follows:
 668  *
 669  * 1) Use shift to calculate the new vma endpoints.
 670  * 2) Extend vma to cover both the old and new ranges.  This ensures the
 671  *    arguments passed to subsequent functions are consistent.
 672  * 3) Move vma's page tables to the new range.
 673  * 4) Free up any cleared pgd range.
 674  * 5) Shrink the vma to cover only the new range.
 675  */
 676 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
 677 {
 678         struct mm_struct *mm = vma->vm_mm;
 679         unsigned long old_start = vma->vm_start;
 680         unsigned long old_end = vma->vm_end;
 681         unsigned long length = old_end - old_start;
 682         unsigned long new_start = old_start - shift;
 683         unsigned long new_end = old_end - shift;
 684         struct mmu_gather tlb;
 685
 686         BUG_ON(new_start > new_end);
 687
 688         /*
 689          * ensure there are no vmas between where we want to go
 690          * and where we are
 691          */
 692         if (vma != find_vma(mm, new_start))
 693                 return -EFAULT;
 694
 695         /*
 696          * cover the whole range: [new_start, old_end)
 697          */
 698         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
 699                 return -ENOMEM;
 700
 701         /*
 702          * move the page tables downwards, on failure we rely on
 703          * process cleanup to remove whatever mess we made.
 704          */
 705         if (length != move_page_tables(vma, old_start,
 706                                        vma, new_start, length, false))
 707                 return -ENOMEM;
 708
 709         lru_add_drain();
 710         tlb_gather_mmu(&tlb, mm, old_start, old_end);
 711         if (new_end > old_start) {
 712                 /*
 713                  * when the old and new regions overlap clear from new_end.
 714                  */
 715                 free_pgd_range(&tlb, new_end, old_end, new_end,
 716                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 717         } else {
 718                 /*
 719                  * otherwise, clean from old_start; this is done to not touch
 720                  * the address space in [new_end, old_start) some architectures
 721                  * have constraints on va-space that make this illegal (IA64) -
 722                  * for the others its just a little faster.
 723                  */
 724                 free_pgd_range(&tlb, old_start, old_end, new_end,
 725                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
 726         }
 727         tlb_finish_mmu(&tlb, old_start, old_end);
 728
 729         /*
 730          * Shrink the vma to just the new range.  Always succeeds.
 731          */
 732         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
 733
 734         return 0;
 735 }
 736
 737 /*
 738  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
 739  * the stack is optionally relocated, and some extra space is added.
 740  */
 741 int setup_arg_pages(struct linux_binprm *bprm,
 742                     unsigned long stack_top,
 743                     int executable_stack)
 744 {
 745         unsigned long ret;
 746         unsigned long stack_shift;
 747         struct mm_struct *mm = current->mm;
 748         struct vm_area_struct *vma = bprm->vma;
 749         struct vm_area_struct *prev = NULL;
 750         unsigned long vm_flags;
 751         unsigned long stack_base;
 752         unsigned long stack_size;
 753         unsigned long stack_expand;
 754         unsigned long rlim_stack;
 755
 756 #ifdef CONFIG_STACK_GROWSUP
 757         /* Limit stack size */
 758         stack_base = bprm->rlim_stack.rlim_max;
 759         if (stack_base > STACK_SIZE_MAX)
 760                 stack_base = STACK_SIZE_MAX;
 761
 762         /* Add space for stack randomization. */
 763         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
 764
 765         /* Make sure we didn't let the argument array grow too large. */
 766         if (vma->vm_end - vma->vm_start > stack_base)
 767                 return -ENOMEM;
 768
 769         stack_base = PAGE_ALIGN(stack_top - stack_base);
 770
 771         stack_shift = vma->vm_start - stack_base;
 772         mm->arg_start = bprm->p - stack_shift;
 773         bprm->p = vma->vm_end - stack_shift;
 774 #else
 775         stack_top = arch_align_stack(stack_top);
 776         stack_top = PAGE_ALIGN(stack_top);
 777
 778         if (unlikely(stack_top < mmap_min_addr) ||
 779             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
 780                 return -ENOMEM;
 781
 782         stack_shift = vma->vm_end - stack_top;
 783
 784         bprm->p -= stack_shift;
 785         mm->arg_start = bprm->p;
 786 #endif
 787
 788         if (bprm->loader)
 789                 bprm->loader -= stack_shift;
 790         bprm->exec -= stack_shift;
 791
 792         if (mmap_write_lock_killable(mm))
 793                 return -EINTR;
 794
 795         vm_flags = VM_STACK_FLAGS;
 796
 797         /*
 798          * Adjust stack execute permissions; explicitly enable for
 799          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
 800          * (arch default) otherwise.
 801          */
 802         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
 803                 vm_flags |= VM_EXEC;
 804         else if (executable_stack == EXSTACK_DISABLE_X)
 805                 vm_flags &= ~VM_EXEC;
 806         vm_flags |= mm->def_flags;
 807         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
 808
 809         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
 810                         vm_flags);
 811         if (ret)
 812                 goto out_unlock;
 813         BUG_ON(prev != vma);
 814
 815         if (unlikely(vm_flags & VM_EXEC)) {
 816                 pr_warn_once("process '%pD4' started with executable stack\n",
 817                              bprm->file);
 818         }
 819
 820         /* Move stack pages down in memory. */
 821         if (stack_shift) {
 822                 ret = shift_arg_pages(vma, stack_shift);
 823                 if (ret)
 824                         goto out_unlock;
 825         }
 826
 827         /* mprotect_fixup is overkill to remove the temporary stack flags */
 828         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
 829
 830         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
 831         stack_size = vma->vm_end - vma->vm_start;
 832         /*
 833          * Align this down to a page boundary as expand_stack
 834          * will align it up.
 835          */
 836         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
 837 #ifdef CONFIG_STACK_GROWSUP
 838         if (stack_size + stack_expand > rlim_stack)
 839                 stack_base = vma->vm_start + rlim_stack;
 840         else
 841                 stack_base = vma->vm_end + stack_expand;
 842 #else
 843         if (stack_size + stack_expand > rlim_stack)
 844                 stack_base = vma->vm_end - rlim_stack;
 845         else
 846                 stack_base = vma->vm_start - stack_expand;
 847 #endif
 848         current->mm->start_stack = bprm->p;
 849         ret = expand_stack(vma, stack_base);
 850         if (ret)
 851                 ret = -EFAULT;
 852
 853 out_unlock:
 854         mmap_write_unlock(mm);
 855         return ret;
 856 }
 857 EXPORT_SYMBOL(setup_arg_pages);
 858
 859 #else
 860
 861 /*
 862  * Transfer the program arguments and environment from the holding pages
 863  * onto the stack. The provided stack pointer is adjusted accordingly.
 864  */
 865 int transfer_args_to_stack(struct linux_binprm *bprm,
 866                            unsigned long *sp_location)
 867 {
 868         unsigned long index, stop, sp;
 869         int ret = 0;
 870
 871         stop = bprm->p >> PAGE_SHIFT;
 872         sp = *sp_location;
 873
 874         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
 875                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
 876                 char *src = kmap(bprm->page[index]) + offset;
 877                 sp -= PAGE_SIZE - offset;
 878                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
 879                         ret = -EFAULT;
 880                 kunmap(bprm->page[index]);
 881                 if (ret)
 882                         goto out;
 883         }
 884
 885         *sp_location = sp;
 886
 887 out:
 888         return ret;
 889 }
 890 EXPORT_SYMBOL(transfer_args_to_stack);
 891
 892 #endif /* CONFIG_MMU */
 893
 894 static struct file *do_open_execat(int fd, struct filename *name, int flags)
 895 {
 896         struct file *file;
 897         int err;
 898         struct open_flags open_exec_flags = {
 899                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
 900                 .acc_mode = MAY_EXEC,
 901                 .intent = LOOKUP_OPEN,
 902                 .lookup_flags = LOOKUP_FOLLOW,
 903         };
 904
 905         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
 906                 return ERR_PTR(-EINVAL);
 907         if (flags & AT_SYMLINK_NOFOLLOW)
 908                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
 909         if (flags & AT_EMPTY_PATH)
 910                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
 911
 912         file = do_filp_open(fd, name, &open_exec_flags);
 913         if (IS_ERR(file))
 914                 goto out;
 915
 916         /*
 917          * may_open() has already checked for this, so it should be
 918          * impossible to trip now. But we need to be extra cautious
 919          * and check again at the very end too.
 920          */
 921         err = -EACCES;
 922         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 923                          path_noexec(&file->f_path)))
 924                 goto exit;
 925
 926         err = deny_write_access(file);
 927         if (err)
 928                 goto exit;
 929
 930         if (name->name[0] != '\0')
 931                 fsnotify_open(file);
 932
 933 out:
 934         return file;
 935
 936 exit:
 937         fput(file);
 938         return ERR_PTR(err);
 939 }
 940
 941 struct file *open_exec(const char *name)
 942 {
 943         struct filename *filename = getname_kernel(name);
 944         struct file *f = ERR_CAST(filename);
 945
 946         if (!IS_ERR(filename)) {
 947                 f = do_open_execat(AT_FDCWD, filename, 0);
 948                 putname(filename);
 949         }
 950         return f;
 951 }
 952 EXPORT_SYMBOL(open_exec);
 953
 954 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
 955     defined(CONFIG_BINFMT_ELF_FDPIC)
 956 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
 957 {
 958         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
 959         if (res > 0)
 960                 flush_icache_user_range(addr, addr + len);
 961         return res;
 962 }
 963 EXPORT_SYMBOL(read_code);
 964 #endif
 965
 966 /*
 967  * Maps the mm_struct mm into the current task struct.
 968  * On success, this function returns with the mutex
 969  * exec_update_mutex locked.
 970  */
 971 static int exec_mmap(struct mm_struct *mm)
 972 {
 973         struct task_struct *tsk;
 974         struct mm_struct *old_mm, *active_mm;
 975         int ret;
 976
 977         /* Notify parent that we're no longer interested in the old VM */
 978         tsk = current;
 979         old_mm = current->mm;
 980         exec_mm_release(tsk, old_mm);
 981         if (old_mm)
 982                 sync_mm_rss(old_mm);
 983
 984         ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
 985         if (ret)
 986                 return ret;
 987
 988         if (old_mm) {
 989                 /*
 990                  * Make sure that if there is a core dump in progress
 991                  * for the old mm, we get out and die instead of going
 992                  * through with the exec.  We must hold mmap_lock around
 993                  * checking core_state and changing tsk->mm.
 994                  */
 995                 mmap_read_lock(old_mm);
 996                 if (unlikely(old_mm->core_state)) {
 997                         mmap_read_unlock(old_mm);
 998                         mutex_unlock(&tsk->signal->exec_update_mutex);
 999                         return -EINTR;
1000                 }
1001         }
1002
1003         task_lock(tsk);
1004         membarrier_exec_mmap(mm);
1005
1006         local_irq_disable();
1007         active_mm = tsk->active_mm;
1008         tsk->active_mm = mm;
1009         tsk->mm = mm;
1010         /*
1011          * This prevents preemption while active_mm is being loaded and
1012          * it and mm are being updated, which could cause problems for
1013          * lazy tlb mm refcounting when these are updated by context
1014          * switches. Not all architectures can handle irqs off over
1015          * activate_mm yet.
1016          */
1017         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1018                 local_irq_enable();
1019         activate_mm(active_mm, mm);
1020         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1021                 local_irq_enable();
1022         tsk->mm->vmacache_seqnum = 0;
1023         vmacache_flush(tsk);
1024         task_unlock(tsk);
1025         if (old_mm) {
1026                 mmap_read_unlock(old_mm);
1027                 BUG_ON(active_mm != old_mm);
1028                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1029                 mm_update_next_owner(old_mm);
1030                 mmput(old_mm);
1031                 return 0;
1032         }
1033         mmdrop(active_mm);
1034         return 0;
1035 }
1036
1037 static int de_thread(struct task_struct *tsk)
1038 {
1039         struct signal_struct *sig = tsk->signal;
1040         struct sighand_struct *oldsighand = tsk->sighand;
1041         spinlock_t *lock = &oldsighand->siglock;
1042
1043         if (thread_group_empty(tsk))
1044                 goto no_thread_group;
1045
1046         /*
1047          * Kill all other threads in the thread group.
1048          */
1049         spin_lock_irq(lock);
1050         if (signal_group_exit(sig)) {
1051                 /*
1052                  * Another group action in progress, just
1053                  * return so that the signal is processed.
1054                  */
1055                 spin_unlock_irq(lock);
1056                 return -EAGAIN;
1057         }
1058
1059         sig->group_exit_task = tsk;
1060         sig->notify_count = zap_other_threads(tsk);
1061         if (!thread_group_leader(tsk))
1062                 sig->notify_count--;
1063
1064         while (sig->notify_count) {
1065                 __set_current_state(TASK_KILLABLE);
1066                 spin_unlock_irq(lock);
1067                 schedule();
1068                 if (__fatal_signal_pending(tsk))
1069                         goto killed;
1070                 spin_lock_irq(lock);
1071         }
1072         spin_unlock_irq(lock);
1073
1074         /*
1075          * At this point all other threads have exited, all we have to
1076          * do is to wait for the thread group leader to become inactive,
1077          * and to assume its PID:
1078          */
1079         if (!thread_group_leader(tsk)) {
1080                 struct task_struct *leader = tsk->group_leader;
1081
1082                 for (;;) {
1083                         cgroup_threadgroup_change_begin(tsk);
1084                         write_lock_irq(&tasklist_lock);
1085                         /*
1086                          * Do this under tasklist_lock to ensure that
1087                          * exit_notify() can't miss ->group_exit_task
1088                          */
1089                         sig->notify_count = -1;
1090                         if (likely(leader->exit_state))
1091                                 break;
1092                         __set_current_state(TASK_KILLABLE);
1093                         write_unlock_irq(&tasklist_lock);
1094                         cgroup_threadgroup_change_end(tsk);
1095                         schedule();
1096                         if (__fatal_signal_pending(tsk))
1097                                 goto killed;
1098                 }
1099
1100                 /*
1101                  * The only record we have of the real-time age of a
1102                  * process, regardless of execs it's done, is start_time.
1103                  * All the past CPU time is accumulated in signal_struct
1104                  * from sister threads now dead.  But in this non-leader
1105                  * exec, nothing survives from the original leader thread,
1106                  * whose birth marks the true age of this process now.
1107                  * When we take on its identity by switching to its PID, we
1108                  * also take its birthdate (always earlier than our own).
1109                  */
1110                 tsk->start_time = leader->start_time;
1111                 tsk->start_boottime = leader->start_boottime;
1112
1113                 BUG_ON(!same_thread_group(leader, tsk));
1114                 /*
1115                  * An exec() starts a new thread group with the
1116                  * TGID of the previous thread group. Rehash the
1117                  * two threads with a switched PID, and release
1118                  * the former thread group leader:
1119                  */
1120
1121                 /* Become a process group leader with the old leader's pid.
1122                  * The old leader becomes a thread of the this thread group.
1123                  */
1124                 exchange_tids(tsk, leader);
1125                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1126                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1127                 transfer_pid(leader, tsk, PIDTYPE_SID);
1128
1129                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1130                 list_replace_init(&leader->sibling, &tsk->sibling);
1131
1132                 tsk->group_leader = tsk;
1133                 leader->group_leader = tsk;
1134
1135                 tsk->exit_signal = SIGCHLD;
1136                 leader->exit_signal = -1;
1137
1138                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1139                 leader->exit_state = EXIT_DEAD;
1140
1141                 /*
1142                  * We are going to release_task()->ptrace_unlink() silently,
1143                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1144                  * the tracer wont't block again waiting for this thread.
1145                  */
1146                 if (unlikely(leader->ptrace))
1147                         __wake_up_parent(leader, leader->parent);
1148                 write_unlock_irq(&tasklist_lock);
1149                 cgroup_threadgroup_change_end(tsk);
1150
1151                 release_task(leader);
1152         }
1153
1154         sig->group_exit_task = NULL;
1155         sig->notify_count = 0;
1156
1157 no_thread_group:
1158         /* we have changed execution domain */
1159         tsk->exit_signal = SIGCHLD;
1160
1161         BUG_ON(!thread_group_leader(tsk));
1162         return 0;
1163
1164 killed:
1165         /* protects against exit_notify() and __exit_signal() */
1166         read_lock(&tasklist_lock);
1167         sig->group_exit_task = NULL;
1168         sig->notify_count = 0;
1169         read_unlock(&tasklist_lock);
1170         return -EAGAIN;
1171 }
1172
1173
1174 /*
1175  * This function makes sure the current process has its own signal table,
1176  * so that flush_signal_handlers can later reset the handlers without
1177  * disturbing other processes.  (Other processes might share the signal
1178  * table via the CLONE_SIGHAND option to clone().)
1179  */
1180 static int unshare_sighand(struct task_struct *me)
1181 {
1182         struct sighand_struct *oldsighand = me->sighand;
1183
1184         if (refcount_read(&oldsighand->count) != 1) {
1185                 struct sighand_struct *newsighand;
1186                 /*
1187                  * This ->sighand is shared with the CLONE_SIGHAND
1188                  * but not CLONE_THREAD task, switch to the new one.
1189                  */
1190                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1191                 if (!newsighand)
1192                         return -ENOMEM;
1193
1194                 refcount_set(&newsighand->count, 1);
1195                 memcpy(newsighand->action, oldsighand->action,
1196                        sizeof(newsighand->action));
1197
1198                 write_lock_irq(&tasklist_lock);
1199                 spin_lock(&oldsighand->siglock);
1200                 rcu_assign_pointer(me->sighand, newsighand);
1201                 spin_unlock(&oldsighand->siglock);
1202                 write_unlock_irq(&tasklist_lock);
1203
1204                 __cleanup_sighand(oldsighand);
1205         }
1206         return 0;
1207 }
1208
1209 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1210 {
1211         task_lock(tsk);
1212         strncpy(buf, tsk->comm, buf_size);
1213         task_unlock(tsk);
1214         return buf;
1215 }
1216 EXPORT_SYMBOL_GPL(__get_task_comm);
1217
1218 /*
1219  * These functions flushes out all traces of the currently running executable
1220  * so that a new one can be started
1221  */
1222
1223 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1224 {
1225         task_lock(tsk);
1226         trace_task_rename(tsk, buf);
1227         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1228         task_unlock(tsk);
1229         perf_event_comm(tsk, exec);
1230 }
1231
1232 /*
1233  * Calling this is the point of no return. None of the failures will be
1234  * seen by userspace since either the process is already taking a fatal
1235  * signal (via de_thread() or coredump), or will have SEGV raised
1236  * (after exec_mmap()) by search_binary_handler (see below).
1237  */
1238 int begin_new_exec(struct linux_binprm * bprm)
1239 {
1240         struct task_struct *me = current;
1241         int retval;
1242
1243         /* Once we are committed compute the creds */
1244         retval = bprm_creds_from_file(bprm);
1245         if (retval)
1246                 return retval;
1247
1248         /*
1249          * Ensure all future errors are fatal.
1250          */
1251         bprm->point_of_no_return = true;
1252
1253         /*
1254          * Make this the only thread in the thread group.
1255          */
1256         retval = de_thread(me);
1257         if (retval)
1258                 goto out;
1259
1260         /*
1261          * Must be called _before_ exec_mmap() as bprm->mm is
1262          * not visibile until then. This also enables the update
1263          * to be lockless.
1264          */
1265         set_mm_exe_file(bprm->mm, bprm->file);
1266
1267         /* If the binary is not readable then enforce mm->dumpable=0 */
1268         would_dump(bprm, bprm->file);
1269         if (bprm->have_execfd)
1270                 would_dump(bprm, bprm->executable);
1271
1272         /*
1273          * Release all of the old mmap stuff
1274          */
1275         acct_arg_size(bprm, 0);
1276         retval = exec_mmap(bprm->mm);
1277         if (retval)
1278                 goto out;
1279
1280         bprm->mm = NULL;
1281
1282 #ifdef CONFIG_POSIX_TIMERS
1283         exit_itimers(me->signal);
1284         flush_itimer_signals();
1285 #endif
1286
1287         /*
1288          * Make the signal table private.
1289          */
1290         retval = unshare_sighand(me);
1291         if (retval)
1292                 goto out_unlock;
1293
1294         /*
1295          * Ensure that the uaccess routines can actually operate on userspace
1296          * pointers:
1297          */
1298         force_uaccess_begin();
1299
1300         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1301                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1302         flush_thread();
1303         me->personality &= ~bprm->per_clear;
1304
1305         /*
1306          * We have to apply CLOEXEC before we change whether the process is
1307          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1308          * trying to access the should-be-closed file descriptors of a process
1309          * undergoing exec(2).
1310          */
1311         do_close_on_exec(me->files);
1312
1313         if (bprm->secureexec) {
1314                 /* Make sure parent cannot signal privileged process. */
1315                 me->pdeath_signal = 0;
1316
1317                 /*
1318                  * For secureexec, reset the stack limit to sane default to
1319                  * avoid bad behavior from the prior rlimits. This has to
1320                  * happen before arch_pick_mmap_layout(), which examines
1321                  * RLIMIT_STACK, but after the point of no return to avoid
1322                  * needing to clean up the change on failure.
1323                  */
1324                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1325                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1326         }
1327
1328         me->sas_ss_sp = me->sas_ss_size = 0;
1329
1330         /*
1331          * Figure out dumpability. Note that this checking only of current
1332          * is wrong, but userspace depends on it. This should be testing
1333          * bprm->secureexec instead.
1334          */
1335         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1336             !(uid_eq(current_euid(), current_uid()) &&
1337               gid_eq(current_egid(), current_gid())))
1338                 set_dumpable(current->mm, suid_dumpable);
1339         else
1340                 set_dumpable(current->mm, SUID_DUMP_USER);
1341
1342         perf_event_exec();
1343         __set_task_comm(me, kbasename(bprm->filename), true);
1344
1345         /* An exec changes our domain. We are no longer part of the thread
1346            group */
1347         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1348         flush_signal_handlers(me, 0);
1349
1350         /*
1351          * install the new credentials for this executable
1352          */
1353         security_bprm_committing_creds(bprm);
1354
1355         commit_creds(bprm->cred);
1356         bprm->cred = NULL;
1357
1358         /*
1359          * Disable monitoring for regular users
1360          * when executing setuid binaries. Must
1361          * wait until new credentials are committed
1362          * by commit_creds() above
1363          */
1364         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1365                 perf_event_exit_task(me);
1366         /*
1367          * cred_guard_mutex must be held at least to this point to prevent
1368          * ptrace_attach() from altering our determination of the task's
1369          * credentials; any time after this it may be unlocked.
1370          */
1371         security_bprm_committed_creds(bprm);
1372
1373         /* Pass the opened binary to the interpreter. */
1374         if (bprm->have_execfd) {
1375                 retval = get_unused_fd_flags(0);
1376                 if (retval < 0)
1377                         goto out_unlock;
1378                 fd_install(retval, bprm->executable);
1379                 bprm->executable = NULL;
1380                 bprm->execfd = retval;
1381         }
1382         return 0;
1383
1384 out_unlock:
1385         mutex_unlock(&me->signal->exec_update_mutex);
1386 out:
1387         return retval;
1388 }
1389 EXPORT_SYMBOL(begin_new_exec);
1390
1391 void would_dump(struct linux_binprm *bprm, struct file *file)
1392 {
1393         struct inode *inode = file_inode(file);
1394         if (inode_permission(inode, MAY_READ) < 0) {
1395                 struct user_namespace *old, *user_ns;
1396                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1397
1398                 /* Ensure mm->user_ns contains the executable */
1399                 user_ns = old = bprm->mm->user_ns;
1400                 while ((user_ns != &init_user_ns) &&
1401                        !privileged_wrt_inode_uidgid(user_ns, inode))
1402                         user_ns = user_ns->parent;
1403
1404                 if (old != user_ns) {
1405                         bprm->mm->user_ns = get_user_ns(user_ns);
1406                         put_user_ns(old);
1407                 }
1408         }
1409 }
1410 EXPORT_SYMBOL(would_dump);
1411
1412 void setup_new_exec(struct linux_binprm * bprm)
1413 {
1414         /* Setup things that can depend upon the personality */
1415         struct task_struct *me = current;
1416
1417         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1418
1419         arch_setup_new_exec();
1420
1421         /* Set the new mm task size. We have to do that late because it may
1422          * depend on TIF_32BIT which is only updated in flush_thread() on
1423          * some architectures like powerpc
1424          */
1425         me->mm->task_size = TASK_SIZE;
1426         mutex_unlock(&me->signal->exec_update_mutex);
1427         mutex_unlock(&me->signal->cred_guard_mutex);
1428 }
1429 EXPORT_SYMBOL(setup_new_exec);
1430
1431 /* Runs immediately before start_thread() takes over. */
1432 void finalize_exec(struct linux_binprm *bprm)
1433 {
1434         /* Store any stack rlimit changes before starting thread. */
1435         task_lock(current->group_leader);
1436         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1437         task_unlock(current->group_leader);
1438 }
1439 EXPORT_SYMBOL(finalize_exec);
1440
1441 /*
1442  * Prepare credentials and lock ->cred_guard_mutex.
1443  * setup_new_exec() commits the new creds and drops the lock.
1444  * Or, if exec fails before, free_bprm() should release ->cred and
1445  * and unlock.
1446  */
1447 static int prepare_bprm_creds(struct linux_binprm *bprm)
1448 {
1449         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1450                 return -ERESTARTNOINTR;
1451
1452         bprm->cred = prepare_exec_creds();
1453         if (likely(bprm->cred))
1454                 return 0;
1455
1456         mutex_unlock(&current->signal->cred_guard_mutex);
1457         return -ENOMEM;
1458 }
1459
1460 static void free_bprm(struct linux_binprm *bprm)
1461 {
1462         if (bprm->mm) {
1463                 acct_arg_size(bprm, 0);
1464                 mmput(bprm->mm);
1465         }
1466         free_arg_pages(bprm);
1467         if (bprm->cred) {
1468                 mutex_unlock(&current->signal->cred_guard_mutex);
1469                 abort_creds(bprm->cred);
1470         }
1471         if (bprm->file) {
1472                 allow_write_access(bprm->file);
1473                 fput(bprm->file);
1474         }
1475         if (bprm->executable)
1476                 fput(bprm->executable);
1477         /* If a binfmt changed the interp, free it. */
1478         if (bprm->interp != bprm->filename)
1479                 kfree(bprm->interp);
1480         kfree(bprm->fdpath);
1481         kfree(bprm);
1482 }
1483
1484 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1485 {
1486         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1487         int retval = -ENOMEM;
1488         if (!bprm)
1489                 goto out;
1490
1491         if (fd == AT_FDCWD || filename->name[0] == '/') {
1492                 bprm->filename = filename->name;
1493         } else {
1494                 if (filename->name[0] == '\0')
1495                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1496                 else
1497                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1498                                                   fd, filename->name);
1499                 if (!bprm->fdpath)
1500                         goto out_free;
1501
1502                 bprm->filename = bprm->fdpath;
1503         }
1504         bprm->interp = bprm->filename;
1505
1506         retval = bprm_mm_init(bprm);
1507         if (retval)
1508                 goto out_free;
1509         return bprm;
1510
1511 out_free:
1512         free_bprm(bprm);
1513 out:
1514         return ERR_PTR(retval);
1515 }
1516
1517 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1518 {
1519         /* If a binfmt changed the interp, free it first. */
1520         if (bprm->interp != bprm->filename)
1521                 kfree(bprm->interp);
1522         bprm->interp = kstrdup(interp, GFP_KERNEL);
1523         if (!bprm->interp)
1524                 return -ENOMEM;
1525         return 0;
1526 }
1527 EXPORT_SYMBOL(bprm_change_interp);
1528
1529 /*
1530  * determine how safe it is to execute the proposed program
1531  * - the caller must hold ->cred_guard_mutex to protect against
1532  *   PTRACE_ATTACH or seccomp thread-sync
1533  */
1534 static void check_unsafe_exec(struct linux_binprm *bprm)
1535 {
1536         struct task_struct *p = current, *t;
1537         unsigned n_fs;
1538
1539         if (p->ptrace)
1540                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1541
1542         /*
1543          * This isn't strictly necessary, but it makes it harder for LSMs to
1544          * mess up.
1545          */
1546         if (task_no_new_privs(current))
1547                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1548
1549         t = p;
1550         n_fs = 1;
1551         spin_lock(&p->fs->lock);
1552         rcu_read_lock();
1553         while_each_thread(p, t) {
1554                 if (t->fs == p->fs)
1555                         n_fs++;
1556         }
1557         rcu_read_unlock();
1558
1559         if (p->fs->users > n_fs)
1560                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1561         else
1562                 p->fs->in_exec = 1;
1563         spin_unlock(&p->fs->lock);
1564 }
1565
1566 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1567 {
1568         /* Handle suid and sgid on files */
1569         struct inode *inode;
1570         unsigned int mode;
1571         kuid_t uid;
1572         kgid_t gid;
1573
1574         if (!mnt_may_suid(file->f_path.mnt))
1575                 return;
1576
1577         if (task_no_new_privs(current))
1578                 return;
1579
1580         inode = file->f_path.dentry->d_inode;
1581         mode = READ_ONCE(inode->i_mode);
1582         if (!(mode & (S_ISUID|S_ISGID)))
1583                 return;
1584
1585         /* Be careful if suid/sgid is set */
1586         inode_lock(inode);
1587
1588         /* reload atomically mode/uid/gid now that lock held */
1589         mode = inode->i_mode;
1590         uid = inode->i_uid;
1591         gid = inode->i_gid;
1592         inode_unlock(inode);
1593
1594         /* We ignore suid/sgid if there are no mappings for them in the ns */
1595         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1596                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1597                 return;
1598
1599         if (mode & S_ISUID) {
1600                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1601                 bprm->cred->euid = uid;
1602         }
1603
1604         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1605                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1606                 bprm->cred->egid = gid;
1607         }
1608 }
1609
1610 /*
1611  * Compute brpm->cred based upon the final binary.
1612  */
1613 static int bprm_creds_from_file(struct linux_binprm *bprm)
1614 {
1615         /* Compute creds based on which file? */
1616         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1617
1618         bprm_fill_uid(bprm, file);
1619         return security_bprm_creds_from_file(bprm, file);
1620 }
1621
1622 /*
1623  * Fill the binprm structure from the inode.
1624  * Read the first BINPRM_BUF_SIZE bytes
1625  *
1626  * This may be called multiple times for binary chains (scripts for example).
1627  */
1628 static int prepare_binprm(struct linux_binprm *bprm)
1629 {
1630         loff_t pos = 0;
1631
1632         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1633         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1634 }
1635
1636 /*
1637  * Arguments are '\0' separated strings found at the location bprm->p
1638  * points to; chop off the first by relocating brpm->p to right after
1639  * the first '\0' encountered.
1640  */
1641 int remove_arg_zero(struct linux_binprm *bprm)
1642 {
1643         int ret = 0;
1644         unsigned long offset;
1645         char *kaddr;
1646         struct page *page;
1647
1648         if (!bprm->argc)
1649                 return 0;
1650
1651         do {
1652                 offset = bprm->p & ~PAGE_MASK;
1653                 page = get_arg_page(bprm, bprm->p, 0);
1654                 if (!page) {
1655                         ret = -EFAULT;
1656                         goto out;
1657                 }
1658                 kaddr = kmap_atomic(page);
1659
1660                 for (; offset < PAGE_SIZE && kaddr[offset];
1661                                 offset++, bprm->p++)
1662                         ;
1663
1664                 kunmap_atomic(kaddr);
1665                 put_arg_page(page);
1666         } while (offset == PAGE_SIZE);
1667
1668         bprm->p++;
1669         bprm->argc--;
1670         ret = 0;
1671
1672 out:
1673         return ret;
1674 }
1675 EXPORT_SYMBOL(remove_arg_zero);
1676
1677 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1678 /*
1679  * cycle the list of binary formats handler, until one recognizes the image
1680  */
1681 static int search_binary_handler(struct linux_binprm *bprm)
1682 {
1683         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1684         struct linux_binfmt *fmt;
1685         int retval;
1686
1687         retval = prepare_binprm(bprm);
1688         if (retval < 0)
1689                 return retval;
1690
1691         retval = security_bprm_check(bprm);
1692         if (retval)
1693                 return retval;
1694
1695         retval = -ENOENT;
1696  retry:
1697         read_lock(&binfmt_lock);
1698         list_for_each_entry(fmt, &formats, lh) {
1699                 if (!try_module_get(fmt->module))
1700                         continue;
1701                 read_unlock(&binfmt_lock);
1702
1703                 retval = fmt->load_binary(bprm);
1704
1705                 read_lock(&binfmt_lock);
1706                 put_binfmt(fmt);
1707                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1708                         read_unlock(&binfmt_lock);
1709                         return retval;
1710                 }
1711         }
1712         read_unlock(&binfmt_lock);
1713
1714         if (need_retry) {
1715                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1716                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1717                         return retval;
1718                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1719                         return retval;
1720                 need_retry = false;
1721                 goto retry;
1722         }
1723
1724         return retval;
1725 }
1726
1727 static int exec_binprm(struct linux_binprm *bprm)
1728 {
1729         pid_t old_pid, old_vpid;
1730         int ret, depth;
1731
1732         /* Need to fetch pid before load_binary changes it */
1733         old_pid = current->pid;
1734         rcu_read_lock();
1735         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1736         rcu_read_unlock();
1737
1738         /* This allows 4 levels of binfmt rewrites before failing hard. */
1739         for (depth = 0;; depth++) {
1740                 struct file *exec;
1741                 if (depth > 5)
1742                         return -ELOOP;
1743
1744                 ret = search_binary_handler(bprm);
1745                 if (ret < 0)
1746                         return ret;
1747                 if (!bprm->interpreter)
1748                         break;
1749
1750                 exec = bprm->file;
1751                 bprm->file = bprm->interpreter;
1752                 bprm->interpreter = NULL;
1753
1754                 allow_write_access(exec);
1755                 if (unlikely(bprm->have_execfd)) {
1756                         if (bprm->executable) {
1757                                 fput(exec);
1758                                 return -ENOEXEC;
1759                         }
1760                         bprm->executable = exec;
1761                 } else
1762                         fput(exec);
1763         }
1764
1765         audit_bprm(bprm);
1766         trace_sched_process_exec(current, old_pid, bprm);
1767         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1768         proc_exec_connector(current);
1769         return 0;
1770 }
1771
1772 /*
1773  * sys_execve() executes a new program.
1774  */
1775 static int bprm_execve(struct linux_binprm *bprm,
1776                        int fd, struct filename *filename, int flags)
1777 {
1778         struct file *file;
1779         struct files_struct *displaced;
1780         int retval;
1781
1782         /*
1783          * Cancel any io_uring activity across execve
1784          */
1785         io_uring_task_cancel();
1786
1787         retval = unshare_files(&displaced);
1788         if (retval)
1789                 return retval;
1790
1791         retval = prepare_bprm_creds(bprm);
1792         if (retval)
1793                 goto out_files;
1794
1795         check_unsafe_exec(bprm);
1796         current->in_execve = 1;
1797
1798         file = do_open_execat(fd, filename, flags);
1799         retval = PTR_ERR(file);
1800         if (IS_ERR(file))
1801                 goto out_unmark;
1802
1803         sched_exec();
1804
1805         bprm->file = file;
1806         /*
1807          * Record that a name derived from an O_CLOEXEC fd will be
1808          * inaccessible after exec. Relies on having exclusive access to
1809          * current->files (due to unshare_files above).
1810          */
1811         if (bprm->fdpath &&
1812             close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1813                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1814
1815         /* Set the unchanging part of bprm->cred */
1816         retval = security_bprm_creds_for_exec(bprm);
1817         if (retval)
1818                 goto out;
1819
1820         retval = exec_binprm(bprm);
1821         if (retval < 0)
1822                 goto out;
1823
1824         /* execve succeeded */
1825         current->fs->in_exec = 0;
1826         current->in_execve = 0;
1827         rseq_execve(current);
1828         acct_update_integrals(current);
1829         task_numa_free(current, false);
1830         if (displaced)
1831                 put_files_struct(displaced);
1832         return retval;
1833
1834 out:
1835         /*
1836          * If past the point of no return ensure the the code never
1837          * returns to the userspace process.  Use an existing fatal
1838          * signal if present otherwise terminate the process with
1839          * SIGSEGV.
1840          */
1841         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1842                 force_sigsegv(SIGSEGV);
1843
1844 out_unmark:
1845         current->fs->in_exec = 0;
1846         current->in_execve = 0;
1847
1848 out_files:
1849         if (displaced)
1850                 reset_files_struct(displaced);
1851
1852         return retval;
1853 }
1854
1855 static int do_execveat_common(int fd, struct filename *filename,
1856                               struct user_arg_ptr argv,
1857                               struct user_arg_ptr envp,
1858                               int flags)
1859 {
1860         struct linux_binprm *bprm;
1861         int retval;
1862
1863         if (IS_ERR(filename))
1864                 return PTR_ERR(filename);
1865
1866         /*
1867          * We move the actual failure in case of RLIMIT_NPROC excess from
1868          * set*uid() to execve() because too many poorly written programs
1869          * don't check setuid() return code.  Here we additionally recheck
1870          * whether NPROC limit is still exceeded.
1871          */
1872         if ((current->flags & PF_NPROC_EXCEEDED) &&
1873             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1874                 retval = -EAGAIN;
1875                 goto out_ret;
1876         }
1877
1878         /* We're below the limit (still or again), so we don't want to make
1879          * further execve() calls fail. */
1880         current->flags &= ~PF_NPROC_EXCEEDED;
1881
1882         bprm = alloc_bprm(fd, filename);
1883         if (IS_ERR(bprm)) {
1884                 retval = PTR_ERR(bprm);
1885                 goto out_ret;
1886         }
1887
1888         retval = count(argv, MAX_ARG_STRINGS);
1889         if (retval < 0)
1890                 goto out_free;
1891         bprm->argc = retval;
1892
1893         retval = count(envp, MAX_ARG_STRINGS);
1894         if (retval < 0)
1895                 goto out_free;
1896         bprm->envc = retval;
1897
1898         retval = bprm_stack_limits(bprm);
1899         if (retval < 0)
1900                 goto out_free;
1901
1902         retval = copy_string_kernel(bprm->filename, bprm);
1903         if (retval < 0)
1904                 goto out_free;
1905         bprm->exec = bprm->p;
1906
1907         retval = copy_strings(bprm->envc, envp, bprm);
1908         if (retval < 0)
1909                 goto out_free;
1910
1911         retval = copy_strings(bprm->argc, argv, bprm);
1912         if (retval < 0)
1913                 goto out_free;
1914
1915         retval = bprm_execve(bprm, fd, filename, flags);
1916 out_free:
1917         free_bprm(bprm);
1918
1919 out_ret:
1920         putname(filename);
1921         return retval;
1922 }
1923
1924 int kernel_execve(const char *kernel_filename,
1925                   const char *const *argv, const char *const *envp)
1926 {
1927         struct filename *filename;
1928         struct linux_binprm *bprm;
1929         int fd = AT_FDCWD;
1930         int retval;
1931
1932         filename = getname_kernel(kernel_filename);
1933         if (IS_ERR(filename))
1934                 return PTR_ERR(filename);
1935
1936         bprm = alloc_bprm(fd, filename);
1937         if (IS_ERR(bprm)) {
1938                 retval = PTR_ERR(bprm);
1939                 goto out_ret;
1940         }
1941
1942         retval = count_strings_kernel(argv);
1943         if (retval < 0)
1944                 goto out_free;
1945         bprm->argc = retval;
1946
1947         retval = count_strings_kernel(envp);
1948         if (retval < 0)
1949                 goto out_free;
1950         bprm->envc = retval;
1951
1952         retval = bprm_stack_limits(bprm);
1953         if (retval < 0)
1954                 goto out_free;
1955
1956         retval = copy_string_kernel(bprm->filename, bprm);
1957         if (retval < 0)
1958                 goto out_free;
1959         bprm->exec = bprm->p;
1960
1961         retval = copy_strings_kernel(bprm->envc, envp, bprm);
1962         if (retval < 0)
1963                 goto out_free;
1964
1965         retval = copy_strings_kernel(bprm->argc, argv, bprm);
1966         if (retval < 0)
1967                 goto out_free;
1968
1969         retval = bprm_execve(bprm, fd, filename, 0);
1970 out_free:
1971         free_bprm(bprm);
1972 out_ret:
1973         putname(filename);
1974         return retval;
1975 }
1976
1977 static int do_execve(struct filename *filename,
1978         const char __user *const __user *__argv,
1979         const char __user *const __user *__envp)
1980 {
1981         struct user_arg_ptr argv = { .ptr.native = __argv };
1982         struct user_arg_ptr envp = { .ptr.native = __envp };
1983         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1984 }
1985
1986 static int do_execveat(int fd, struct filename *filename,
1987                 const char __user *const __user *__argv,
1988                 const char __user *const __user *__envp,
1989                 int flags)
1990 {
1991         struct user_arg_ptr argv = { .ptr.native = __argv };
1992         struct user_arg_ptr envp = { .ptr.native = __envp };
1993
1994         return do_execveat_common(fd, filename, argv, envp, flags);
1995 }
1996
1997 #ifdef CONFIG_COMPAT
1998 static int compat_do_execve(struct filename *filename,
1999         const compat_uptr_t __user *__argv,
2000         const compat_uptr_t __user *__envp)
2001 {
2002         struct user_arg_ptr argv = {
2003                 .is_compat = true,
2004                 .ptr.compat = __argv,
2005         };
2006         struct user_arg_ptr envp = {
2007                 .is_compat = true,
2008                 .ptr.compat = __envp,
2009         };
2010         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2011 }
2012
2013 static int compat_do_execveat(int fd, struct filename *filename,
2014                               const compat_uptr_t __user *__argv,
2015                               const compat_uptr_t __user *__envp,
2016                               int flags)
2017 {
2018         struct user_arg_ptr argv = {
2019                 .is_compat = true,
2020                 .ptr.compat = __argv,
2021         };
2022         struct user_arg_ptr envp = {
2023                 .is_compat = true,
2024                 .ptr.compat = __envp,
2025         };
2026         return do_execveat_common(fd, filename, argv, envp, flags);
2027 }
2028 #endif
2029
2030 void set_binfmt(struct linux_binfmt *new)
2031 {
2032         struct mm_struct *mm = current->mm;
2033
2034         if (mm->binfmt)
2035                 module_put(mm->binfmt->module);
2036
2037         mm->binfmt = new;
2038         if (new)
2039                 __module_get(new->module);
2040 }
2041 EXPORT_SYMBOL(set_binfmt);
2042
2043 /*
2044  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2045  */
2046 void set_dumpable(struct mm_struct *mm, int value)
2047 {
2048         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2049                 return;
2050
2051         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2052 }
2053
2054 SYSCALL_DEFINE3(execve,
2055                 const char __user *, filename,
2056                 const char __user *const __user *, argv,
2057                 const char __user *const __user *, envp)
2058 {
2059         return do_execve(getname(filename), argv, envp);
2060 }
2061
2062 SYSCALL_DEFINE5(execveat,
2063                 int, fd, const char __user *, filename,
2064                 const char __user *const __user *, argv,
2065                 const char __user *const __user *, envp,
2066                 int, flags)
2067 {
2068         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2069
2070         return do_execveat(fd,
2071                            getname_flags(filename, lookup_flags, NULL),
2072                            argv, envp, flags);
2073 }
2074
2075 #ifdef CONFIG_COMPAT
2076 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2077         const compat_uptr_t __user *, argv,
2078         const compat_uptr_t __user *, envp)
2079 {
2080         return compat_do_execve(getname(filename), argv, envp);
2081 }
2082
2083 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2084                        const char __user *, filename,
2085                        const compat_uptr_t __user *, argv,
2086                        const compat_uptr_t __user *, envp,
2087                        int,  flags)
2088 {
2089         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2090
2091         return compat_do_execveat(fd,
2092                                   getname_flags(filename, lookup_flags, NULL),
2093                                   argv, envp, flags);
2094 }
2095 #endif