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