fs/exec.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *  linux/fs/exec.c
   4  *
   5  *  Copyright (C) 1991, 1992  Linus Torvalds
   6  */
   7
   8 /*
   9  * #!-checking implemented by tytso.
  10  */
  11 /*
  12  * Demand-loading implemented 01.12.91 - no need to read anything but
  13  * the header into memory. The inode of the executable is put into
  14  * "current->executable", and page faults do the actual loading. Clean.
  15  *
  16  * Once more I can proudly say that linux stood up to being changed: it
  17  * was less than 2 hours work to get demand-loading completely implemented.
  18  *
  19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
  20  * current->executable is only used by the procfs.  This allows a dispatch
  21  * table to check for several different types  of binary formats.  We keep
  22  * trying until we recognize the file or we run out of supported binary
  23  * formats.
  24  */
  25
  26 #include <linux/kernel_read_file.h>
  27 #include <linux/slab.h>
  28 #include <linux/file.h>
  29 #include <linux/fdtable.h>
  30 #include <linux/mm.h>
  31 #include <linux/vmacache.h>
  32 #include <linux/stat.h>
  33 #include <linux/fcntl.h>
  34 #include <linux/swap.h>
  35 #include <linux/string.h>
  36 #include <linux/init.h>
  37 #include <linux/sched/mm.h>
  38 #include <linux/sched/coredump.h>
  39 #include <linux/sched/signal.h>
  40 #include <linux/sched/numa_balancing.h>
  41 #include <linux/sched/task.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/perf_event.h>
  44 #include <linux/highmem.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/key.h>
  47 #include <linux/personality.h>
  48 #include <linux/binfmts.h>
  49 #include <linux/utsname.h>
  50 #include <linux/pid_namespace.h>
  51 #include <linux/module.h>
  52 #include <linux/namei.h>
  53 #include <linux/mount.h>
  54 #include <linux/security.h>
  55 #include <linux/syscalls.h>
  56 #include <linux/tsacct_kern.h>
  57 #include <linux/cn_proc.h>
  58 #include <linux/audit.h>
  59 #include <linux/tracehook.h>
  60 #include <linux/kmod.h>
  61 #include <linux/fsnotify.h>
  62 #include <linux/fs_struct.h>
  63 #include <linux/oom.h>
  64 #include <linux/compat.h>
  65 #include <linux/vmalloc.h>
  66 #include <linux/io_uring.h>
  67 #include <linux/syscall_user_dispatch.h>
  68
  69 #include <linux/uaccess.h>
  70 #include <asm/mmu_context.h>
  71 #include <asm/tlb.h>
  72
  73 #include <trace/events/task.h>
  74 #include "internal.h"
  75
  76 #include <trace/events/sched.h>
  77
  78 static int bprm_creds_from_file(struct linux_binprm *bprm);
  79
  80 int suid_dumpable = 0;
  81
  82 static LIST_HEAD(formats);
  83 static DEFINE_RWLOCK(binfmt_lock);
  84
  85 void __register_binfmt(struct linux_binfmt * fmt, int insert)
  86 {
  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         /*
 145          * may_open() has already checked for this, so it should be
 146          * impossible to trip now. But we need to be extra cautious
 147          * and check again at the very end too.
 148          */
 149         error = -EACCES;
 150         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
 151                          path_noexec(&file->f_path)))
 152                 goto exit;
 153
 154         fsnotify_open(file);
 155
 156         error = -ENOEXEC;
 157
 158         read_lock(&binfmt_lock);
 159         list_for_each_entry(fmt, &formats, lh) {
 160                 if (!fmt->load_shlib)
 161                         continue;
 162                 if (!try_module_get(fmt->module))
 163                         continue;
 164                 read_unlock(&binfmt_lock);
 165                 error = fmt->load_shlib(file);
 166                 read_lock(&binfmt_lock);
 167                 put_binfmt(fmt);
 168                 if (error != -ENOEXEC)
 169                         break;
 170         }
 171         read_unlock(&binfmt_lock);
 172 exit:
 173         fput(file);
 174 out:
 175         return error;
 176 }
 177 #endif /* #ifdef CONFIG_USELIB */
 178
 179 #ifdef CONFIG_MMU
 180 /*
 181  * The nascent bprm->mm is not visible until exec_mmap() but it can
 182  * use a lot of memory, account these pages in current->mm temporary
 183  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
 184  * change the counter back via acct_arg_size(0).
 185  */
 186 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
 187 {
 188         struct mm_struct *mm = current->mm;
 189         long diff = (long)(pages - bprm->vma_pages);
 190
 191         if (!mm || !diff)
 192                 return;
 193
 194         bprm->vma_pages = pages;
 195         add_mm_counter(mm, MM_ANONPAGES, diff);
 196 }
 197
 198 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 199                 int write)
 200 {
 201         struct page *page;
 202         int ret;
 203         unsigned int gup_flags = FOLL_FORCE;
 204
 205 #ifdef CONFIG_STACK_GROWSUP
 206         if (write) {
 207                 ret = expand_downwards(bprm->vma, pos);
 208                 if (ret < 0)
 209                         return NULL;
 210         }
 211 #endif
 212
 213         if (write)
 214                 gup_flags |= FOLL_WRITE;
 215
 216         /*
 217          * We are doing an exec().  'current' is the process
 218          * doing the exec and bprm->mm is the new process's mm.
 219          */
 220         mmap_read_lock(bprm->mm);
 221         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
 222                         &page, NULL, NULL);
 223         mmap_read_unlock(bprm->mm);
 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_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_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_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);
 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);
 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
 760         stack_base = calc_max_stack_size(stack_base);
 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 exec_update_lock
 969  * held for writing.
 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 = down_write_killable(&tsk->signal->exec_update_lock);
 985         if (ret)
 986                 return ret;
 987
 988         if (old_mm) {
 989                 /*
 990                  * If there is a pending fatal signal perhaps a signal
 991                  * whose default action is to create a coredump get
 992                  * out and die instead of going through with the exec.
 993                  */
 994                 ret = mmap_read_lock_killable(old_mm);
 995                 if (ret) {
 996                         up_write(&tsk->signal->exec_update_lock);
 997                         return ret;
 998                 }
 999         }
1000
1001         task_lock(tsk);
1002         membarrier_exec_mmap(mm);
1003
1004         local_irq_disable();
1005         active_mm = tsk->active_mm;
1006         tsk->active_mm = mm;
1007         tsk->mm = mm;
1008         /*
1009          * This prevents preemption while active_mm is being loaded and
1010          * it and mm are being updated, which could cause problems for
1011          * lazy tlb mm refcounting when these are updated by context
1012          * switches. Not all architectures can handle irqs off over
1013          * activate_mm yet.
1014          */
1015         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1016                 local_irq_enable();
1017         activate_mm(active_mm, mm);
1018         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1019                 local_irq_enable();
1020         tsk->mm->vmacache_seqnum = 0;
1021         vmacache_flush(tsk);
1022         task_unlock(tsk);
1023         if (old_mm) {
1024                 mmap_read_unlock(old_mm);
1025                 BUG_ON(active_mm != old_mm);
1026                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1027                 mm_update_next_owner(old_mm);
1028                 mmput(old_mm);
1029                 return 0;
1030         }
1031         mmdrop(active_mm);
1032         return 0;
1033 }
1034
1035 static int de_thread(struct task_struct *tsk)
1036 {
1037         struct signal_struct *sig = tsk->signal;
1038         struct sighand_struct *oldsighand = tsk->sighand;
1039         spinlock_t *lock = &oldsighand->siglock;
1040
1041         if (thread_group_empty(tsk))
1042                 goto no_thread_group;
1043
1044         /*
1045          * Kill all other threads in the thread group.
1046          */
1047         spin_lock_irq(lock);
1048         if (signal_group_exit(sig)) {
1049                 /*
1050                  * Another group action in progress, just
1051                  * return so that the signal is processed.
1052                  */
1053                 spin_unlock_irq(lock);
1054                 return -EAGAIN;
1055         }
1056
1057         sig->group_exit_task = tsk;
1058         sig->notify_count = zap_other_threads(tsk);
1059         if (!thread_group_leader(tsk))
1060                 sig->notify_count--;
1061
1062         while (sig->notify_count) {
1063                 __set_current_state(TASK_KILLABLE);
1064                 spin_unlock_irq(lock);
1065                 schedule();
1066                 if (__fatal_signal_pending(tsk))
1067                         goto killed;
1068                 spin_lock_irq(lock);
1069         }
1070         spin_unlock_irq(lock);
1071
1072         /*
1073          * At this point all other threads have exited, all we have to
1074          * do is to wait for the thread group leader to become inactive,
1075          * and to assume its PID:
1076          */
1077         if (!thread_group_leader(tsk)) {
1078                 struct task_struct *leader = tsk->group_leader;
1079
1080                 for (;;) {
1081                         cgroup_threadgroup_change_begin(tsk);
1082                         write_lock_irq(&tasklist_lock);
1083                         /*
1084                          * Do this under tasklist_lock to ensure that
1085                          * exit_notify() can't miss ->group_exit_task
1086                          */
1087                         sig->notify_count = -1;
1088                         if (likely(leader->exit_state))
1089                                 break;
1090                         __set_current_state(TASK_KILLABLE);
1091                         write_unlock_irq(&tasklist_lock);
1092                         cgroup_threadgroup_change_end(tsk);
1093                         schedule();
1094                         if (__fatal_signal_pending(tsk))
1095                                 goto killed;
1096                 }
1097
1098                 /*
1099                  * The only record we have of the real-time age of a
1100                  * process, regardless of execs it's done, is start_time.
1101                  * All the past CPU time is accumulated in signal_struct
1102                  * from sister threads now dead.  But in this non-leader
1103                  * exec, nothing survives from the original leader thread,
1104                  * whose birth marks the true age of this process now.
1105                  * When we take on its identity by switching to its PID, we
1106                  * also take its birthdate (always earlier than our own).
1107                  */
1108                 tsk->start_time = leader->start_time;
1109                 tsk->start_boottime = leader->start_boottime;
1110
1111                 BUG_ON(!same_thread_group(leader, tsk));
1112                 /*
1113                  * An exec() starts a new thread group with the
1114                  * TGID of the previous thread group. Rehash the
1115                  * two threads with a switched PID, and release
1116                  * the former thread group leader:
1117                  */
1118
1119                 /* Become a process group leader with the old leader's pid.
1120                  * The old leader becomes a thread of the this thread group.
1121                  */
1122                 exchange_tids(tsk, leader);
1123                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1124                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1125                 transfer_pid(leader, tsk, PIDTYPE_SID);
1126
1127                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1128                 list_replace_init(&leader->sibling, &tsk->sibling);
1129
1130                 tsk->group_leader = tsk;
1131                 leader->group_leader = tsk;
1132
1133                 tsk->exit_signal = SIGCHLD;
1134                 leader->exit_signal = -1;
1135
1136                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1137                 leader->exit_state = EXIT_DEAD;
1138
1139                 /*
1140                  * We are going to release_task()->ptrace_unlink() silently,
1141                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1142                  * the tracer wont't block again waiting for this thread.
1143                  */
1144                 if (unlikely(leader->ptrace))
1145                         __wake_up_parent(leader, leader->parent);
1146                 write_unlock_irq(&tasklist_lock);
1147                 cgroup_threadgroup_change_end(tsk);
1148
1149                 release_task(leader);
1150         }
1151
1152         sig->group_exit_task = NULL;
1153         sig->notify_count = 0;
1154
1155 no_thread_group:
1156         /* we have changed execution domain */
1157         tsk->exit_signal = SIGCHLD;
1158
1159         BUG_ON(!thread_group_leader(tsk));
1160         return 0;
1161
1162 killed:
1163         /* protects against exit_notify() and __exit_signal() */
1164         read_lock(&tasklist_lock);
1165         sig->group_exit_task = NULL;
1166         sig->notify_count = 0;
1167         read_unlock(&tasklist_lock);
1168         return -EAGAIN;
1169 }
1170
1171
1172 /*
1173  * This function makes sure the current process has its own signal table,
1174  * so that flush_signal_handlers can later reset the handlers without
1175  * disturbing other processes.  (Other processes might share the signal
1176  * table via the CLONE_SIGHAND option to clone().)
1177  */
1178 static int unshare_sighand(struct task_struct *me)
1179 {
1180         struct sighand_struct *oldsighand = me->sighand;
1181
1182         if (refcount_read(&oldsighand->count) != 1) {
1183                 struct sighand_struct *newsighand;
1184                 /*
1185                  * This ->sighand is shared with the CLONE_SIGHAND
1186                  * but not CLONE_THREAD task, switch to the new one.
1187                  */
1188                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1189                 if (!newsighand)
1190                         return -ENOMEM;
1191
1192                 refcount_set(&newsighand->count, 1);
1193                 memcpy(newsighand->action, oldsighand->action,
1194                        sizeof(newsighand->action));
1195
1196                 write_lock_irq(&tasklist_lock);
1197                 spin_lock(&oldsighand->siglock);
1198                 rcu_assign_pointer(me->sighand, newsighand);
1199                 spin_unlock(&oldsighand->siglock);
1200                 write_unlock_irq(&tasklist_lock);
1201
1202                 __cleanup_sighand(oldsighand);
1203         }
1204         return 0;
1205 }
1206
1207 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1208 {
1209         task_lock(tsk);
1210         strncpy(buf, tsk->comm, buf_size);
1211         task_unlock(tsk);
1212         return buf;
1213 }
1214 EXPORT_SYMBOL_GPL(__get_task_comm);
1215
1216 /*
1217  * These functions flushes out all traces of the currently running executable
1218  * so that a new one can be started
1219  */
1220
1221 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1222 {
1223         task_lock(tsk);
1224         trace_task_rename(tsk, buf);
1225         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1226         task_unlock(tsk);
1227         perf_event_comm(tsk, exec);
1228 }
1229
1230 /*
1231  * Calling this is the point of no return. None of the failures will be
1232  * seen by userspace since either the process is already taking a fatal
1233  * signal (via de_thread() or coredump), or will have SEGV raised
1234  * (after exec_mmap()) by search_binary_handler (see below).
1235  */
1236 int begin_new_exec(struct linux_binprm * bprm)
1237 {
1238         struct task_struct *me = current;
1239         int retval;
1240
1241         /* Once we are committed compute the creds */
1242         retval = bprm_creds_from_file(bprm);
1243         if (retval)
1244                 return retval;
1245
1246         /*
1247          * Ensure all future errors are fatal.
1248          */
1249         bprm->point_of_no_return = true;
1250
1251         /*
1252          * Make this the only thread in the thread group.
1253          */
1254         retval = de_thread(me);
1255         if (retval)
1256                 goto out;
1257
1258         /*
1259          * Cancel any io_uring activity across execve
1260          */
1261         io_uring_task_cancel();
1262
1263         /* Ensure the files table is not shared. */
1264         retval = unshare_files();
1265         if (retval)
1266                 goto out;
1267
1268         /*
1269          * Must be called _before_ exec_mmap() as bprm->mm is
1270          * not visibile until then. This also enables the update
1271          * to be lockless.
1272          */
1273         retval = set_mm_exe_file(bprm->mm, bprm->file);
1274         if (retval)
1275                 goto out;
1276
1277         /* If the binary is not readable then enforce mm->dumpable=0 */
1278         would_dump(bprm, bprm->file);
1279         if (bprm->have_execfd)
1280                 would_dump(bprm, bprm->executable);
1281
1282         /*
1283          * Release all of the old mmap stuff
1284          */
1285         acct_arg_size(bprm, 0);
1286         retval = exec_mmap(bprm->mm);
1287         if (retval)
1288                 goto out;
1289
1290         bprm->mm = NULL;
1291
1292 #ifdef CONFIG_POSIX_TIMERS
1293         exit_itimers(me->signal);
1294         flush_itimer_signals();
1295 #endif
1296
1297         /*
1298          * Make the signal table private.
1299          */
1300         retval = unshare_sighand(me);
1301         if (retval)
1302                 goto out_unlock;
1303
1304         /*
1305          * Ensure that the uaccess routines can actually operate on userspace
1306          * pointers:
1307          */
1308         force_uaccess_begin();
1309
1310         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1311                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1312         flush_thread();
1313         me->personality &= ~bprm->per_clear;
1314
1315         clear_syscall_work_syscall_user_dispatch(me);
1316
1317         /*
1318          * We have to apply CLOEXEC before we change whether the process is
1319          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1320          * trying to access the should-be-closed file descriptors of a process
1321          * undergoing exec(2).
1322          */
1323         do_close_on_exec(me->files);
1324
1325         if (bprm->secureexec) {
1326                 /* Make sure parent cannot signal privileged process. */
1327                 me->pdeath_signal = 0;
1328
1329                 /*
1330                  * For secureexec, reset the stack limit to sane default to
1331                  * avoid bad behavior from the prior rlimits. This has to
1332                  * happen before arch_pick_mmap_layout(), which examines
1333                  * RLIMIT_STACK, but after the point of no return to avoid
1334                  * needing to clean up the change on failure.
1335                  */
1336                 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1337                         bprm->rlim_stack.rlim_cur = _STK_LIM;
1338         }
1339
1340         me->sas_ss_sp = me->sas_ss_size = 0;
1341
1342         /*
1343          * Figure out dumpability. Note that this checking only of current
1344          * is wrong, but userspace depends on it. This should be testing
1345          * bprm->secureexec instead.
1346          */
1347         if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1348             !(uid_eq(current_euid(), current_uid()) &&
1349               gid_eq(current_egid(), current_gid())))
1350                 set_dumpable(current->mm, suid_dumpable);
1351         else
1352                 set_dumpable(current->mm, SUID_DUMP_USER);
1353
1354         perf_event_exec();
1355         __set_task_comm(me, kbasename(bprm->filename), true);
1356
1357         /* An exec changes our domain. We are no longer part of the thread
1358            group */
1359         WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1360         flush_signal_handlers(me, 0);
1361
1362         retval = set_cred_ucounts(bprm->cred);
1363         if (retval < 0)
1364                 goto out_unlock;
1365
1366         /*
1367          * install the new credentials for this executable
1368          */
1369         security_bprm_committing_creds(bprm);
1370
1371         commit_creds(bprm->cred);
1372         bprm->cred = NULL;
1373
1374         /*
1375          * Disable monitoring for regular users
1376          * when executing setuid binaries. Must
1377          * wait until new credentials are committed
1378          * by commit_creds() above
1379          */
1380         if (get_dumpable(me->mm) != SUID_DUMP_USER)
1381                 perf_event_exit_task(me);
1382         /*
1383          * cred_guard_mutex must be held at least to this point to prevent
1384          * ptrace_attach() from altering our determination of the task's
1385          * credentials; any time after this it may be unlocked.
1386          */
1387         security_bprm_committed_creds(bprm);
1388
1389         /* Pass the opened binary to the interpreter. */
1390         if (bprm->have_execfd) {
1391                 retval = get_unused_fd_flags(0);
1392                 if (retval < 0)
1393                         goto out_unlock;
1394                 fd_install(retval, bprm->executable);
1395                 bprm->executable = NULL;
1396                 bprm->execfd = retval;
1397         }
1398         return 0;
1399
1400 out_unlock:
1401         up_write(&me->signal->exec_update_lock);
1402 out:
1403         return retval;
1404 }
1405 EXPORT_SYMBOL(begin_new_exec);
1406
1407 void would_dump(struct linux_binprm *bprm, struct file *file)
1408 {
1409         struct inode *inode = file_inode(file);
1410         struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1411         if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1412                 struct user_namespace *old, *user_ns;
1413                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1414
1415                 /* Ensure mm->user_ns contains the executable */
1416                 user_ns = old = bprm->mm->user_ns;
1417                 while ((user_ns != &init_user_ns) &&
1418                        !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1419                         user_ns = user_ns->parent;
1420
1421                 if (old != user_ns) {
1422                         bprm->mm->user_ns = get_user_ns(user_ns);
1423                         put_user_ns(old);
1424                 }
1425         }
1426 }
1427 EXPORT_SYMBOL(would_dump);
1428
1429 void setup_new_exec(struct linux_binprm * bprm)
1430 {
1431         /* Setup things that can depend upon the personality */
1432         struct task_struct *me = current;
1433
1434         arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1435
1436         arch_setup_new_exec();
1437
1438         /* Set the new mm task size. We have to do that late because it may
1439          * depend on TIF_32BIT which is only updated in flush_thread() on
1440          * some architectures like powerpc
1441          */
1442         me->mm->task_size = TASK_SIZE;
1443         up_write(&me->signal->exec_update_lock);
1444         mutex_unlock(&me->signal->cred_guard_mutex);
1445 }
1446 EXPORT_SYMBOL(setup_new_exec);
1447
1448 /* Runs immediately before start_thread() takes over. */
1449 void finalize_exec(struct linux_binprm *bprm)
1450 {
1451         /* Store any stack rlimit changes before starting thread. */
1452         task_lock(current->group_leader);
1453         current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1454         task_unlock(current->group_leader);
1455 }
1456 EXPORT_SYMBOL(finalize_exec);
1457
1458 /*
1459  * Prepare credentials and lock ->cred_guard_mutex.
1460  * setup_new_exec() commits the new creds and drops the lock.
1461  * Or, if exec fails before, free_bprm() should release ->cred
1462  * and unlock.
1463  */
1464 static int prepare_bprm_creds(struct linux_binprm *bprm)
1465 {
1466         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1467                 return -ERESTARTNOINTR;
1468
1469         bprm->cred = prepare_exec_creds();
1470         if (likely(bprm->cred))
1471                 return 0;
1472
1473         mutex_unlock(&current->signal->cred_guard_mutex);
1474         return -ENOMEM;
1475 }
1476
1477 static void free_bprm(struct linux_binprm *bprm)
1478 {
1479         if (bprm->mm) {
1480                 acct_arg_size(bprm, 0);
1481                 mmput(bprm->mm);
1482         }
1483         free_arg_pages(bprm);
1484         if (bprm->cred) {
1485                 mutex_unlock(&current->signal->cred_guard_mutex);
1486                 abort_creds(bprm->cred);
1487         }
1488         if (bprm->file) {
1489                 allow_write_access(bprm->file);
1490                 fput(bprm->file);
1491         }
1492         if (bprm->executable)
1493                 fput(bprm->executable);
1494         /* If a binfmt changed the interp, free it. */
1495         if (bprm->interp != bprm->filename)
1496                 kfree(bprm->interp);
1497         kfree(bprm->fdpath);
1498         kfree(bprm);
1499 }
1500
1501 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1502 {
1503         struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1504         int retval = -ENOMEM;
1505         if (!bprm)
1506                 goto out;
1507
1508         if (fd == AT_FDCWD || filename->name[0] == '/') {
1509                 bprm->filename = filename->name;
1510         } else {
1511                 if (filename->name[0] == '\0')
1512                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1513                 else
1514                         bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1515                                                   fd, filename->name);
1516                 if (!bprm->fdpath)
1517                         goto out_free;
1518
1519                 bprm->filename = bprm->fdpath;
1520         }
1521         bprm->interp = bprm->filename;
1522
1523         retval = bprm_mm_init(bprm);
1524         if (retval)
1525                 goto out_free;
1526         return bprm;
1527
1528 out_free:
1529         free_bprm(bprm);
1530 out:
1531         return ERR_PTR(retval);
1532 }
1533
1534 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1535 {
1536         /* If a binfmt changed the interp, free it first. */
1537         if (bprm->interp != bprm->filename)
1538                 kfree(bprm->interp);
1539         bprm->interp = kstrdup(interp, GFP_KERNEL);
1540         if (!bprm->interp)
1541                 return -ENOMEM;
1542         return 0;
1543 }
1544 EXPORT_SYMBOL(bprm_change_interp);
1545
1546 /*
1547  * determine how safe it is to execute the proposed program
1548  * - the caller must hold ->cred_guard_mutex to protect against
1549  *   PTRACE_ATTACH or seccomp thread-sync
1550  */
1551 static void check_unsafe_exec(struct linux_binprm *bprm)
1552 {
1553         struct task_struct *p = current, *t;
1554         unsigned n_fs;
1555
1556         if (p->ptrace)
1557                 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1558
1559         /*
1560          * This isn't strictly necessary, but it makes it harder for LSMs to
1561          * mess up.
1562          */
1563         if (task_no_new_privs(current))
1564                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1565
1566         t = p;
1567         n_fs = 1;
1568         spin_lock(&p->fs->lock);
1569         rcu_read_lock();
1570         while_each_thread(p, t) {
1571                 if (t->fs == p->fs)
1572                         n_fs++;
1573         }
1574         rcu_read_unlock();
1575
1576         if (p->fs->users > n_fs)
1577                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1578         else
1579                 p->fs->in_exec = 1;
1580         spin_unlock(&p->fs->lock);
1581 }
1582
1583 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1584 {
1585         /* Handle suid and sgid on files */
1586         struct user_namespace *mnt_userns;
1587         struct inode *inode;
1588         unsigned int mode;
1589         kuid_t uid;
1590         kgid_t gid;
1591
1592         if (!mnt_may_suid(file->f_path.mnt))
1593                 return;
1594
1595         if (task_no_new_privs(current))
1596                 return;
1597
1598         inode = file->f_path.dentry->d_inode;
1599         mode = READ_ONCE(inode->i_mode);
1600         if (!(mode & (S_ISUID|S_ISGID)))
1601                 return;
1602
1603         mnt_userns = file_mnt_user_ns(file);
1604
1605         /* Be careful if suid/sgid is set */
1606         inode_lock(inode);
1607
1608         /* reload atomically mode/uid/gid now that lock held */
1609         mode = inode->i_mode;
1610         uid = i_uid_into_mnt(mnt_userns, inode);
1611         gid = i_gid_into_mnt(mnt_userns, inode);
1612         inode_unlock(inode);
1613
1614         /* We ignore suid/sgid if there are no mappings for them in the ns */
1615         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1616                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1617                 return;
1618
1619         if (mode & S_ISUID) {
1620                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1621                 bprm->cred->euid = uid;
1622         }
1623
1624         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1625                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1626                 bprm->cred->egid = gid;
1627         }
1628 }
1629
1630 /*
1631  * Compute brpm->cred based upon the final binary.
1632  */
1633 static int bprm_creds_from_file(struct linux_binprm *bprm)
1634 {
1635         /* Compute creds based on which file? */
1636         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1637
1638         bprm_fill_uid(bprm, file);
1639         return security_bprm_creds_from_file(bprm, file);
1640 }
1641
1642 /*
1643  * Fill the binprm structure from the inode.
1644  * Read the first BINPRM_BUF_SIZE bytes
1645  *
1646  * This may be called multiple times for binary chains (scripts for example).
1647  */
1648 static int prepare_binprm(struct linux_binprm *bprm)
1649 {
1650         loff_t pos = 0;
1651
1652         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1653         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1654 }
1655
1656 /*
1657  * Arguments are '\0' separated strings found at the location bprm->p
1658  * points to; chop off the first by relocating brpm->p to right after
1659  * the first '\0' encountered.
1660  */
1661 int remove_arg_zero(struct linux_binprm *bprm)
1662 {
1663         int ret = 0;
1664         unsigned long offset;
1665         char *kaddr;
1666         struct page *page;
1667
1668         if (!bprm->argc)
1669                 return 0;
1670
1671         do {
1672                 offset = bprm->p & ~PAGE_MASK;
1673                 page = get_arg_page(bprm, bprm->p, 0);
1674                 if (!page) {
1675                         ret = -EFAULT;
1676                         goto out;
1677                 }
1678                 kaddr = kmap_atomic(page);
1679
1680                 for (; offset < PAGE_SIZE && kaddr[offset];
1681                                 offset++, bprm->p++)
1682                         ;
1683
1684                 kunmap_atomic(kaddr);
1685                 put_arg_page(page);
1686         } while (offset == PAGE_SIZE);
1687
1688         bprm->p++;
1689         bprm->argc--;
1690         ret = 0;
1691
1692 out:
1693         return ret;
1694 }
1695 EXPORT_SYMBOL(remove_arg_zero);
1696
1697 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1698 /*
1699  * cycle the list of binary formats handler, until one recognizes the image
1700  */
1701 static int search_binary_handler(struct linux_binprm *bprm)
1702 {
1703         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1704         struct linux_binfmt *fmt;
1705         int retval;
1706
1707         retval = prepare_binprm(bprm);
1708         if (retval < 0)
1709                 return retval;
1710
1711         retval = security_bprm_check(bprm);
1712         if (retval)
1713                 return retval;
1714
1715         retval = -ENOENT;
1716  retry:
1717         read_lock(&binfmt_lock);
1718         list_for_each_entry(fmt, &formats, lh) {
1719                 if (!try_module_get(fmt->module))
1720                         continue;
1721                 read_unlock(&binfmt_lock);
1722
1723                 retval = fmt->load_binary(bprm);
1724
1725                 read_lock(&binfmt_lock);
1726                 put_binfmt(fmt);
1727                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1728                         read_unlock(&binfmt_lock);
1729                         return retval;
1730                 }
1731         }
1732         read_unlock(&binfmt_lock);
1733
1734         if (need_retry) {
1735                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1736                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1737                         return retval;
1738                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1739                         return retval;
1740                 need_retry = false;
1741                 goto retry;
1742         }
1743
1744         return retval;
1745 }
1746
1747 static int exec_binprm(struct linux_binprm *bprm)
1748 {
1749         pid_t old_pid, old_vpid;
1750         int ret, depth;
1751
1752         /* Need to fetch pid before load_binary changes it */
1753         old_pid = current->pid;
1754         rcu_read_lock();
1755         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1756         rcu_read_unlock();
1757
1758         /* This allows 4 levels of binfmt rewrites before failing hard. */
1759         for (depth = 0;; depth++) {
1760                 struct file *exec;
1761                 if (depth > 5)
1762                         return -ELOOP;
1763
1764                 ret = search_binary_handler(bprm);
1765                 if (ret < 0)
1766                         return ret;
1767                 if (!bprm->interpreter)
1768                         break;
1769
1770                 exec = bprm->file;
1771                 bprm->file = bprm->interpreter;
1772                 bprm->interpreter = NULL;
1773
1774                 allow_write_access(exec);
1775                 if (unlikely(bprm->have_execfd)) {
1776                         if (bprm->executable) {
1777                                 fput(exec);
1778                                 return -ENOEXEC;
1779                         }
1780                         bprm->executable = exec;
1781                 } else
1782                         fput(exec);
1783         }
1784
1785         audit_bprm(bprm);
1786         trace_sched_process_exec(current, old_pid, bprm);
1787         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1788         proc_exec_connector(current);
1789         return 0;
1790 }
1791
1792 /*
1793  * sys_execve() executes a new program.
1794  */
1795 static int bprm_execve(struct linux_binprm *bprm,
1796                        int fd, struct filename *filename, int flags)
1797 {
1798         struct file *file;
1799         int retval;
1800
1801         retval = prepare_bprm_creds(bprm);
1802         if (retval)
1803                 return retval;
1804
1805         check_unsafe_exec(bprm);
1806         current->in_execve = 1;
1807
1808         file = do_open_execat(fd, filename, flags);
1809         retval = PTR_ERR(file);
1810         if (IS_ERR(file))
1811                 goto out_unmark;
1812
1813         sched_exec();
1814
1815         bprm->file = file;
1816         /*
1817          * Record that a name derived from an O_CLOEXEC fd will be
1818          * inaccessible after exec.  This allows the code in exec to
1819          * choose to fail when the executable is not mmaped into the
1820          * interpreter and an open file descriptor is not passed to
1821          * the interpreter.  This makes for a better user experience
1822          * than having the interpreter start and then immediately fail
1823          * when it finds the executable is inaccessible.
1824          */
1825         if (bprm->fdpath && get_close_on_exec(fd))
1826                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1827
1828         /* Set the unchanging part of bprm->cred */
1829         retval = security_bprm_creds_for_exec(bprm);
1830         if (retval)
1831                 goto out;
1832
1833         retval = exec_binprm(bprm);
1834         if (retval < 0)
1835                 goto out;
1836
1837         /* execve succeeded */
1838         current->fs->in_exec = 0;
1839         current->in_execve = 0;
1840         rseq_execve(current);
1841         acct_update_integrals(current);
1842         task_numa_free(current, false);
1843         return retval;
1844
1845 out:
1846         /*
1847          * If past the point of no return ensure the code never
1848          * returns to the userspace process.  Use an existing fatal
1849          * signal if present otherwise terminate the process with
1850          * SIGSEGV.
1851          */
1852         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1853                 force_fatal_sig(SIGSEGV);
1854
1855 out_unmark:
1856         current->fs->in_exec = 0;
1857         current->in_execve = 0;
1858
1859         return retval;
1860 }
1861
1862 static int do_execveat_common(int fd, struct filename *filename,
1863                               struct user_arg_ptr argv,
1864                               struct user_arg_ptr envp,
1865                               int flags)
1866 {
1867         struct linux_binprm *bprm;
1868         int retval;
1869
1870         if (IS_ERR(filename))
1871                 return PTR_ERR(filename);
1872
1873         /*
1874          * We move the actual failure in case of RLIMIT_NPROC excess from
1875          * set*uid() to execve() because too many poorly written programs
1876          * don't check setuid() return code.  Here we additionally recheck
1877          * whether NPROC limit is still exceeded.
1878          */
1879         if ((current->flags & PF_NPROC_EXCEEDED) &&
1880             is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1881                 retval = -EAGAIN;
1882                 goto out_ret;
1883         }
1884
1885         /* We're below the limit (still or again), so we don't want to make
1886          * further execve() calls fail. */
1887         current->flags &= ~PF_NPROC_EXCEEDED;
1888
1889         bprm = alloc_bprm(fd, filename);
1890         if (IS_ERR(bprm)) {
1891                 retval = PTR_ERR(bprm);
1892                 goto out_ret;
1893         }
1894
1895         retval = count(argv, MAX_ARG_STRINGS);
1896         if (retval < 0)
1897                 goto out_free;
1898         bprm->argc = retval;
1899
1900         retval = count(envp, MAX_ARG_STRINGS);
1901         if (retval < 0)
1902                 goto out_free;
1903         bprm->envc = retval;
1904
1905         retval = bprm_stack_limits(bprm);
1906         if (retval < 0)
1907                 goto out_free;
1908
1909         retval = copy_string_kernel(bprm->filename, bprm);
1910         if (retval < 0)
1911                 goto out_free;
1912         bprm->exec = bprm->p;
1913
1914         retval = copy_strings(bprm->envc, envp, bprm);
1915         if (retval < 0)
1916                 goto out_free;
1917
1918         retval = copy_strings(bprm->argc, argv, bprm);
1919         if (retval < 0)
1920                 goto out_free;
1921
1922         retval = bprm_execve(bprm, fd, filename, flags);
1923 out_free:
1924         free_bprm(bprm);
1925
1926 out_ret:
1927         putname(filename);
1928         return retval;
1929 }
1930
1931 int kernel_execve(const char *kernel_filename,
1932                   const char *const *argv, const char *const *envp)
1933 {
1934         struct filename *filename;
1935         struct linux_binprm *bprm;
1936         int fd = AT_FDCWD;
1937         int retval;
1938
1939         filename = getname_kernel(kernel_filename);
1940         if (IS_ERR(filename))
1941                 return PTR_ERR(filename);
1942
1943         bprm = alloc_bprm(fd, filename);
1944         if (IS_ERR(bprm)) {
1945                 retval = PTR_ERR(bprm);
1946                 goto out_ret;
1947         }
1948
1949         retval = count_strings_kernel(argv);
1950         if (retval < 0)
1951                 goto out_free;
1952         bprm->argc = retval;
1953
1954         retval = count_strings_kernel(envp);
1955         if (retval < 0)
1956                 goto out_free;
1957         bprm->envc = retval;
1958
1959         retval = bprm_stack_limits(bprm);
1960         if (retval < 0)
1961                 goto out_free;
1962
1963         retval = copy_string_kernel(bprm->filename, bprm);
1964         if (retval < 0)
1965                 goto out_free;
1966         bprm->exec = bprm->p;
1967
1968         retval = copy_strings_kernel(bprm->envc, envp, bprm);
1969         if (retval < 0)
1970                 goto out_free;
1971
1972         retval = copy_strings_kernel(bprm->argc, argv, bprm);
1973         if (retval < 0)
1974                 goto out_free;
1975
1976         retval = bprm_execve(bprm, fd, filename, 0);
1977 out_free:
1978         free_bprm(bprm);
1979 out_ret:
1980         putname(filename);
1981         return retval;
1982 }
1983
1984 static int do_execve(struct filename *filename,
1985         const char __user *const __user *__argv,
1986         const char __user *const __user *__envp)
1987 {
1988         struct user_arg_ptr argv = { .ptr.native = __argv };
1989         struct user_arg_ptr envp = { .ptr.native = __envp };
1990         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1991 }
1992
1993 static int do_execveat(int fd, struct filename *filename,
1994                 const char __user *const __user *__argv,
1995                 const char __user *const __user *__envp,
1996                 int flags)
1997 {
1998         struct user_arg_ptr argv = { .ptr.native = __argv };
1999         struct user_arg_ptr envp = { .ptr.native = __envp };
2000
2001         return do_execveat_common(fd, filename, argv, envp, flags);
2002 }
2003
2004 #ifdef CONFIG_COMPAT
2005 static int compat_do_execve(struct filename *filename,
2006         const compat_uptr_t __user *__argv,
2007         const compat_uptr_t __user *__envp)
2008 {
2009         struct user_arg_ptr argv = {
2010                 .is_compat = true,
2011                 .ptr.compat = __argv,
2012         };
2013         struct user_arg_ptr envp = {
2014                 .is_compat = true,
2015                 .ptr.compat = __envp,
2016         };
2017         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2018 }
2019
2020 static int compat_do_execveat(int fd, struct filename *filename,
2021                               const compat_uptr_t __user *__argv,
2022                               const compat_uptr_t __user *__envp,
2023                               int flags)
2024 {
2025         struct user_arg_ptr argv = {
2026                 .is_compat = true,
2027                 .ptr.compat = __argv,
2028         };
2029         struct user_arg_ptr envp = {
2030                 .is_compat = true,
2031                 .ptr.compat = __envp,
2032         };
2033         return do_execveat_common(fd, filename, argv, envp, flags);
2034 }
2035 #endif
2036
2037 void set_binfmt(struct linux_binfmt *new)
2038 {
2039         struct mm_struct *mm = current->mm;
2040
2041         if (mm->binfmt)
2042                 module_put(mm->binfmt->module);
2043
2044         mm->binfmt = new;
2045         if (new)
2046                 __module_get(new->module);
2047 }
2048 EXPORT_SYMBOL(set_binfmt);
2049
2050 /*
2051  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2052  */
2053 void set_dumpable(struct mm_struct *mm, int value)
2054 {
2055         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2056                 return;
2057
2058         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2059 }
2060
2061 SYSCALL_DEFINE3(execve,
2062                 const char __user *, filename,
2063                 const char __user *const __user *, argv,
2064                 const char __user *const __user *, envp)
2065 {
2066         return do_execve(getname(filename), argv, envp);
2067 }
2068
2069 SYSCALL_DEFINE5(execveat,
2070                 int, fd, const char __user *, filename,
2071                 const char __user *const __user *, argv,
2072                 const char __user *const __user *, envp,
2073                 int, flags)
2074 {
2075         return do_execveat(fd,
2076                            getname_uflags(filename, flags),
2077                            argv, envp, flags);
2078 }
2079
2080 #ifdef CONFIG_COMPAT
2081 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2082         const compat_uptr_t __user *, argv,
2083         const compat_uptr_t __user *, envp)
2084 {
2085         return compat_do_execve(getname(filename), argv, envp);
2086 }
2087
2088 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2089                        const char __user *, filename,
2090                        const compat_uptr_t __user *, argv,
2091                        const compat_uptr_t __user *, envp,
2092                        int,  flags)
2093 {
2094         return compat_do_execveat(fd,
2095                                   getname_uflags(filename, flags),
2096                                   argv, envp, flags);
2097 }
2098 #endif