개념 설명 전체 · v6.6 / fs/exec.c
1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/fs/exec.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 */ 7 8 /* 9 * #!-checking implemented by tytso. 10 */ 11 /* 12 * Demand-loading implemented 01.12.91 - no need to read anything but 13 * the header into memory. The inode of the executable is put into 14 * "current->executable", and page faults do the actual loading. Clean. 15 * 16 * Once more I can proudly say that linux stood up to being changed: it 17 * was less than 2 hours work to get demand-loading completely implemented. 18 * 19 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead, 20 * current->executable is only used by the procfs. This allows a dispatch 21 * table to check for several different types of binary formats. We keep 22 * trying until we recognize the file or we run out of supported binary 23 * formats. 24 */ 25 26 #include <linux/kernel_read_file.h> 27 #include <linux/slab.h> 28 #include <linux/file.h> 29 #include <linux/fdtable.h> 30 #include <linux/mm.h> 31 #include <linux/stat.h> 32 #include <linux/fcntl.h> 33 #include <linux/swap.h> 34 #include <linux/string.h> 35 #include <linux/init.h> 36 #include <linux/sched/mm.h> 37 #include <linux/sched/coredump.h> 38 #include <linux/sched/signal.h> 39 #include <linux/sched/numa_balancing.h> 40 #include <linux/sched/task.h> 41 #include <linux/pagemap.h> 42 #include <linux/perf_event.h> 43 #include <linux/highmem.h> 44 #include <linux/spinlock.h> 45 #include <linux/key.h> 46 #include <linux/personality.h> 47 #include <linux/binfmts.h> 48 #include <linux/utsname.h> 49 #include <linux/pid_namespace.h> 50 #include <linux/module.h> 51 #include <linux/namei.h> 52 #include <linux/mount.h> 53 #include <linux/security.h> 54 #include <linux/syscalls.h> 55 #include <linux/tsacct_kern.h> 56 #include <linux/cn_proc.h> 57 #include <linux/audit.h> 58 #include <linux/kmod.h> 59 #include <linux/fsnotify.h> 60 #include <linux/fs_struct.h> 61 #include <linux/oom.h> 62 #include <linux/compat.h> 63 #include <linux/vmalloc.h> 64 #include <linux/io_uring.h> 65 #include <linux/syscall_user_dispatch.h> 66 #include <linux/coredump.h> 67 #include <linux/time_namespace.h> 68 #include <linux/user_events.h> 69 70 #include <linux/uaccess.h> 71 #include <asm/mmu_context.h> 72 #include <asm/tlb.h> 73 74 #include <trace/events/task.h> 75 #include "internal.h" 76 77 #include <trace/events/sched.h> 78 79 static int bprm_creds_from_file(struct linux_binprm *bprm); 80 81 int suid_dumpable = 0; 82 83 static LIST_HEAD(formats); 84 static DEFINE_RWLOCK(binfmt_lock); 85 86 void __register_binfmt(struct linux_binfmt * fmt, int insert) 87 { 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 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 error = -ENOEXEC; 156 157 read_lock(&binfmt_lock); 158 list_for_each_entry(fmt, &formats, lh) { 159 if (!fmt->load_shlib) 160 continue; 161 if (!try_module_get(fmt->module)) 162 continue; 163 read_unlock(&binfmt_lock); 164 error = fmt->load_shlib(file); 165 read_lock(&binfmt_lock); 166 put_binfmt(fmt); 167 if (error != -ENOEXEC) 168 break; 169 } 170 read_unlock(&binfmt_lock); 171 exit: 172 fput(file); 173 out: 174 return error; 175 } 176 #endif /* #ifdef CONFIG_USELIB */ 177 178 #ifdef CONFIG_MMU 179 /* 180 * The nascent bprm->mm is not visible until exec_mmap() but it can 181 * use a lot of memory, account these pages in current->mm temporary 182 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we 183 * change the counter back via acct_arg_size(0). 184 */ 185 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages) 186 { 187 struct mm_struct *mm = current->mm; 188 long diff = (long)(pages - bprm->vma_pages); 189 190 if (!mm || !diff) 191 return; 192 193 bprm->vma_pages = pages; 194 add_mm_counter(mm, MM_ANONPAGES, diff); 195 } 196 197 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, 198 int write) 199 { 200 struct page *page; 201 struct vm_area_struct *vma = bprm->vma; 202 struct mm_struct *mm = bprm->mm; 203 int ret; 204 205 /* 206 * Avoid relying on expanding the stack down in GUP (which 207 * does not work for STACK_GROWSUP anyway), and just do it 208 * by hand ahead of time. 209 */ 210 if (write && pos < vma->vm_start) { 211 mmap_write_lock(mm); 212 ret = expand_downwards(vma, pos); 213 if (unlikely(ret < 0)) { 214 mmap_write_unlock(mm); 215 return NULL; 216 } 217 mmap_write_downgrade(mm); 218 } else 219 mmap_read_lock(mm); 220 221 /* 222 * We are doing an exec(). 'current' is the process 223 * doing the exec and 'mm' is the new process's mm. 224 */ 225 ret = get_user_pages_remote(mm, pos, 1, 226 write ? FOLL_WRITE : 0, 227 &page, NULL); 228 mmap_read_unlock(mm); 229 if (ret <= 0) 230 return NULL; 231 232 if (write) 233 acct_arg_size(bprm, vma_pages(vma)); 234 235 return page; 236 } 237 238 static void put_arg_page(struct page *page) 239 { 240 put_page(page); 241 } 242 243 static void free_arg_pages(struct linux_binprm *bprm) 244 { 245 } 246 247 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, 248 struct page *page) 249 { 250 flush_cache_page(bprm->vma, pos, page_to_pfn(page)); 251 } 252 253 static int __bprm_mm_init(struct linux_binprm *bprm) 254 { 255 int err; 256 struct vm_area_struct *vma = NULL; 257 struct mm_struct *mm = bprm->mm; 258 259 bprm->vma = vma = vm_area_alloc(mm); 260 if (!vma) 261 return -ENOMEM; 262 vma_set_anonymous(vma); 263 264 if (mmap_write_lock_killable(mm)) { 265 err = -EINTR; 266 goto err_free; 267 } 268 269 /* 270 * Place the stack at the largest stack address the architecture 271 * supports. Later, we'll move this to an appropriate place. We don't 272 * use STACK_TOP because that can depend on attributes which aren't 273 * configured yet. 274 */ 275 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP); 276 vma->vm_end = STACK_TOP_MAX; 277 vma->vm_start = vma->vm_end - PAGE_SIZE; 278 vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP); 279 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 280 281 err = insert_vm_struct(mm, vma); 282 if (err) 283 goto err; 284 285 mm->stack_vm = mm->total_vm = 1; 286 mmap_write_unlock(mm); 287 bprm->p = vma->vm_end - sizeof(void *); 288 return 0; 289 err: 290 mmap_write_unlock(mm); 291 err_free: 292 bprm->vma = NULL; 293 vm_area_free(vma); 294 return err; 295 } 296 297 static bool valid_arg_len(struct linux_binprm *bprm, long len) 298 { 299 return len <= MAX_ARG_STRLEN; 300 } 301 302 #else 303 304 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages) 305 { 306 } 307 308 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, 309 int write) 310 { 311 struct page *page; 312 313 page = bprm->page[pos / PAGE_SIZE]; 314 if (!page && write) { 315 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO); 316 if (!page) 317 return NULL; 318 bprm->page[pos / PAGE_SIZE] = page; 319 } 320 321 return page; 322 } 323 324 static void put_arg_page(struct page *page) 325 { 326 } 327 328 static void free_arg_page(struct linux_binprm *bprm, int i) 329 { 330 if (bprm->page[i]) { 331 __free_page(bprm->page[i]); 332 bprm->page[i] = NULL; 333 } 334 } 335 336 static void free_arg_pages(struct linux_binprm *bprm) 337 { 338 int i; 339 340 for (i = 0; i < MAX_ARG_PAGES; i++) 341 free_arg_page(bprm, i); 342 } 343 344 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, 345 struct page *page) 346 { 347 } 348 349 static int __bprm_mm_init(struct linux_binprm *bprm) 350 { 351 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *); 352 return 0; 353 } 354 355 static bool valid_arg_len(struct linux_binprm *bprm, long len) 356 { 357 return len <= bprm->p; 358 } 359 360 #endif /* CONFIG_MMU */ 361 362 /* 363 * Create a new mm_struct and populate it with a temporary stack 364 * vm_area_struct. We don't have enough context at this point to set the stack 365 * flags, permissions, and offset, so we use temporary values. We'll update 366 * them later in setup_arg_pages(). 367 */ 368 static int bprm_mm_init(struct linux_binprm *bprm) 369 { 370 int err; 371 struct mm_struct *mm = NULL; 372 373 bprm->mm = mm = mm_alloc(); 374 err = -ENOMEM; 375 if (!mm) 376 goto err; 377 378 /* Save current stack limit for all calculations made during exec. */ 379 task_lock(current->group_leader); 380 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK]; 381 task_unlock(current->group_leader); 382 383 err = __bprm_mm_init(bprm); 384 if (err) 385 goto err; 386 387 return 0; 388 389 err: 390 if (mm) { 391 bprm->mm = NULL; 392 mmdrop(mm); 393 } 394 395 return err; 396 } 397 398 struct user_arg_ptr { 399 #ifdef CONFIG_COMPAT 400 bool is_compat; 401 #endif 402 union { 403 const char __user *const __user *native; 404 #ifdef CONFIG_COMPAT 405 const compat_uptr_t __user *compat; 406 #endif 407 } ptr; 408 }; 409 410 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr) 411 { 412 const char __user *native; 413 414 #ifdef CONFIG_COMPAT 415 if (unlikely(argv.is_compat)) { 416 compat_uptr_t compat; 417 418 if (get_user(compat, argv.ptr.compat + nr)) 419 return ERR_PTR(-EFAULT); 420 421 return compat_ptr(compat); 422 } 423 #endif 424 425 if (get_user(native, argv.ptr.native + nr)) 426 return ERR_PTR(-EFAULT); 427 428 return native; 429 } 430 431 /* 432 * count() counts the number of strings in array ARGV. 433 */ 434 static int count(struct user_arg_ptr argv, int max) 435 { 436 int i = 0; 437 438 if (argv.ptr.native != NULL) { 439 for (;;) { 440 const char __user *p = get_user_arg_ptr(argv, i); 441 442 if (!p) 443 break; 444 445 if (IS_ERR(p)) 446 return -EFAULT; 447 448 if (i >= max) 449 return -E2BIG; 450 ++i; 451 452 if (fatal_signal_pending(current)) 453 return -ERESTARTNOHAND; 454 cond_resched(); 455 } 456 } 457 return i; 458 } 459 460 static int count_strings_kernel(const char *const *argv) 461 { 462 int i; 463 464 if (!argv) 465 return 0; 466 467 for (i = 0; argv[i]; ++i) { 468 if (i >= MAX_ARG_STRINGS) 469 return -E2BIG; 470 if (fatal_signal_pending(current)) 471 return -ERESTARTNOHAND; 472 cond_resched(); 473 } 474 return i; 475 } 476 477 static int bprm_stack_limits(struct linux_binprm *bprm) 478 { 479 unsigned long limit, ptr_size; 480 481 /* 482 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM 483 * (whichever is smaller) for the argv+env strings. 484 * This ensures that: 485 * - the remaining binfmt code will not run out of stack space, 486 * - the program will have a reasonable amount of stack left 487 * to work from. 488 */ 489 limit = _STK_LIM / 4 * 3; 490 limit = min(limit, bprm->rlim_stack.rlim_cur / 4); 491 /* 492 * We've historically supported up to 32 pages (ARG_MAX) 493 * of argument strings even with small stacks 494 */ 495 limit = max_t(unsigned long, limit, ARG_MAX); 496 /* 497 * We must account for the size of all the argv and envp pointers to 498 * the argv and envp strings, since they will also take up space in 499 * the stack. They aren't stored until much later when we can't 500 * signal to the parent that the child has run out of stack space. 501 * Instead, calculate it here so it's possible to fail gracefully. 502 * 503 * In the case of argc = 0, make sure there is space for adding a 504 * empty string (which will bump argc to 1), to ensure confused 505 * userspace programs don't start processing from argv[1], thinking 506 * argc can never be 0, to keep them from walking envp by accident. 507 * See do_execveat_common(). 508 */ 509 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *); 510 if (limit <= ptr_size) 511 return -E2BIG; 512 limit -= ptr_size; 513 514 bprm->argmin = bprm->p - limit; 515 return 0; 516 } 517 518 /* 519 * 'copy_strings()' copies argument/environment strings from the old 520 * processes's memory to the new process's stack. The call to get_user_pages() 521 * ensures the destination page is created and not swapped out. 522 */ 523 static int copy_strings(int argc, struct user_arg_ptr argv, 524 struct linux_binprm *bprm) 525 { 526 struct page *kmapped_page = NULL; 527 char *kaddr = NULL; 528 unsigned long kpos = 0; 529 int ret; 530 531 while (argc-- > 0) { 532 const char __user *str; 533 int len; 534 unsigned long pos; 535 536 ret = -EFAULT; 537 str = get_user_arg_ptr(argv, argc); 538 if (IS_ERR(str)) 539 goto out; 540 541 len = strnlen_user(str, MAX_ARG_STRLEN); 542 if (!len) 543 goto out; 544 545 ret = -E2BIG; 546 if (!valid_arg_len(bprm, len)) 547 goto out; 548 549 /* We're going to work our way backwards. */ 550 pos = bprm->p; 551 str += len; 552 bprm->p -= len; 553 #ifdef CONFIG_MMU 554 if (bprm->p < bprm->argmin) 555 goto out; 556 #endif 557 558 while (len > 0) { 559 int offset, bytes_to_copy; 560 561 if (fatal_signal_pending(current)) { 562 ret = -ERESTARTNOHAND; 563 goto out; 564 } 565 cond_resched(); 566 567 offset = pos % PAGE_SIZE; 568 if (offset == 0) 569 offset = PAGE_SIZE; 570 571 bytes_to_copy = offset; 572 if (bytes_to_copy > len) 573 bytes_to_copy = len; 574 575 offset -= bytes_to_copy; 576 pos -= bytes_to_copy; 577 str -= bytes_to_copy; 578 len -= bytes_to_copy; 579 580 if (!kmapped_page || kpos != (pos & PAGE_MASK)) { 581 struct page *page; 582 583 page = get_arg_page(bprm, pos, 1); 584 if (!page) { 585 ret = -E2BIG; 586 goto out; 587 } 588 589 if (kmapped_page) { 590 flush_dcache_page(kmapped_page); 591 kunmap_local(kaddr); 592 put_arg_page(kmapped_page); 593 } 594 kmapped_page = page; 595 kaddr = kmap_local_page(kmapped_page); 596 kpos = pos & PAGE_MASK; 597 flush_arg_page(bprm, kpos, kmapped_page); 598 } 599 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) { 600 ret = -EFAULT; 601 goto out; 602 } 603 } 604 } 605 ret = 0; 606 out: 607 if (kmapped_page) { 608 flush_dcache_page(kmapped_page); 609 kunmap_local(kaddr); 610 put_arg_page(kmapped_page); 611 } 612 return ret; 613 } 614 615 /* 616 * Copy and argument/environment string from the kernel to the processes stack. 617 */ 618 int copy_string_kernel(const char *arg, struct linux_binprm *bprm) 619 { 620 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */; 621 unsigned long pos = bprm->p; 622 623 if (len == 0) 624 return -EFAULT; 625 if (!valid_arg_len(bprm, len)) 626 return -E2BIG; 627 628 /* We're going to work our way backwards. */ 629 arg += len; 630 bprm->p -= len; 631 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin) 632 return -E2BIG; 633 634 while (len > 0) { 635 unsigned int bytes_to_copy = min_t(unsigned int, len, 636 min_not_zero(offset_in_page(pos), PAGE_SIZE)); 637 struct page *page; 638 639 pos -= bytes_to_copy; 640 arg -= bytes_to_copy; 641 len -= bytes_to_copy; 642 643 page = get_arg_page(bprm, pos, 1); 644 if (!page) 645 return -E2BIG; 646 flush_arg_page(bprm, pos & PAGE_MASK, page); 647 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy); 648 put_arg_page(page); 649 } 650 651 return 0; 652 } 653 EXPORT_SYMBOL(copy_string_kernel); 654 655 static int copy_strings_kernel(int argc, const char *const *argv, 656 struct linux_binprm *bprm) 657 { 658 while (argc-- > 0) { 659 int ret = copy_string_kernel(argv[argc], bprm); 660 if (ret < 0) 661 return ret; 662 if (fatal_signal_pending(current)) 663 return -ERESTARTNOHAND; 664 cond_resched(); 665 } 666 return 0; 667 } 668 669 #ifdef CONFIG_MMU 670 671 /* 672 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once 673 * the binfmt code determines where the new stack should reside, we shift it to 674 * its final location. The process proceeds as follows: 675 * 676 * 1) Use shift to calculate the new vma endpoints. 677 * 2) Extend vma to cover both the old and new ranges. This ensures the 678 * arguments passed to subsequent functions are consistent. 679 * 3) Move vma's page tables to the new range. 680 * 4) Free up any cleared pgd range. 681 * 5) Shrink the vma to cover only the new range. 682 */ 683 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift) 684 { 685 struct mm_struct *mm = vma->vm_mm; 686 unsigned long old_start = vma->vm_start; 687 unsigned long old_end = vma->vm_end; 688 unsigned long length = old_end - old_start; 689 unsigned long new_start = old_start - shift; 690 unsigned long new_end = old_end - shift; 691 VMA_ITERATOR(vmi, mm, new_start); 692 struct vm_area_struct *next; 693 struct mmu_gather tlb; 694 695 BUG_ON(new_start > new_end); 696 697 /* 698 * ensure there are no vmas between where we want to go 699 * and where we are 700 */ 701 if (vma != vma_next(&vmi)) 702 return -EFAULT; 703 704 vma_iter_prev_range(&vmi); 705 /* 706 * cover the whole range: [new_start, old_end) 707 */ 708 if (vma_expand(&vmi, vma, new_start, old_end, vma->vm_pgoff, NULL)) 709 return -ENOMEM; 710 711 /* 712 * move the page tables downwards, on failure we rely on 713 * process cleanup to remove whatever mess we made. 714 */ 715 if (length != move_page_tables(vma, old_start, 716 vma, new_start, length, false)) 717 return -ENOMEM; 718 719 lru_add_drain(); 720 tlb_gather_mmu(&tlb, mm); 721 next = vma_next(&vmi); 722 if (new_end > old_start) { 723 /* 724 * when the old and new regions overlap clear from new_end. 725 */ 726 free_pgd_range(&tlb, new_end, old_end, new_end, 727 next ? next->vm_start : USER_PGTABLES_CEILING); 728 } else { 729 /* 730 * otherwise, clean from old_start; this is done to not touch 731 * the address space in [new_end, old_start) some architectures 732 * have constraints on va-space that make this illegal (IA64) - 733 * for the others its just a little faster. 734 */ 735 free_pgd_range(&tlb, old_start, old_end, new_end, 736 next ? next->vm_start : USER_PGTABLES_CEILING); 737 } 738 tlb_finish_mmu(&tlb); 739 740 vma_prev(&vmi); 741 /* Shrink the vma to just the new range */ 742 return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff); 743 } 744 745 /* 746 * Finalizes the stack vm_area_struct. The flags and permissions are updated, 747 * the stack is optionally relocated, and some extra space is added. 748 */ 749 int setup_arg_pages(struct linux_binprm *bprm, 750 unsigned long stack_top, 751 int executable_stack) 752 { 753 unsigned long ret; 754 unsigned long stack_shift; 755 struct mm_struct *mm = current->mm; 756 struct vm_area_struct *vma = bprm->vma; 757 struct vm_area_struct *prev = NULL; 758 unsigned long vm_flags; 759 unsigned long stack_base; 760 unsigned long stack_size; 761 unsigned long stack_expand; 762 unsigned long rlim_stack; 763 struct mmu_gather tlb; 764 struct vma_iterator vmi; 765 766 #ifdef CONFIG_STACK_GROWSUP 767 /* Limit stack size */ 768 stack_base = bprm->rlim_stack.rlim_max; 769 770 stack_base = calc_max_stack_size(stack_base); 771 772 /* Add space for stack randomization. */ 773 stack_base += (STACK_RND_MASK << PAGE_SHIFT); 774 775 /* Make sure we didn't let the argument array grow too large. */ 776 if (vma->vm_end - vma->vm_start > stack_base) 777 return -ENOMEM; 778 779 stack_base = PAGE_ALIGN(stack_top - stack_base); 780 781 stack_shift = vma->vm_start - stack_base; 782 mm->arg_start = bprm->p - stack_shift; 783 bprm->p = vma->vm_end - stack_shift; 784 #else 785 stack_top = arch_align_stack(stack_top); 786 stack_top = PAGE_ALIGN(stack_top); 787 788 if (unlikely(stack_top < mmap_min_addr) || 789 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr)) 790 return -ENOMEM; 791 792 stack_shift = vma->vm_end - stack_top; 793 794 bprm->p -= stack_shift; 795 mm->arg_start = bprm->p; 796 #endif 797 798 if (bprm->loader) 799 bprm->loader -= stack_shift; 800 bprm->exec -= stack_shift; 801 802 if (mmap_write_lock_killable(mm)) 803 return -EINTR; 804 805 vm_flags = VM_STACK_FLAGS; 806 807 /* 808 * Adjust stack execute permissions; explicitly enable for 809 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone 810 * (arch default) otherwise. 811 */ 812 if (unlikely(executable_stack == EXSTACK_ENABLE_X)) 813 vm_flags |= VM_EXEC; 814 else if (executable_stack == EXSTACK_DISABLE_X) 815 vm_flags &= ~VM_EXEC; 816 vm_flags |= mm->def_flags; 817 vm_flags |= VM_STACK_INCOMPLETE_SETUP; 818 819 vma_iter_init(&vmi, mm, vma->vm_start); 820 821 tlb_gather_mmu(&tlb, mm); 822 ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end, 823 vm_flags); 824 tlb_finish_mmu(&tlb); 825 826 if (ret) 827 goto out_unlock; 828 BUG_ON(prev != vma); 829 830 if (unlikely(vm_flags & VM_EXEC)) { 831 pr_warn_once("process '%pD4' started with executable stack\n", 832 bprm->file); 833 } 834 835 /* Move stack pages down in memory. */ 836 if (stack_shift) { 837 ret = shift_arg_pages(vma, stack_shift); 838 if (ret) 839 goto out_unlock; 840 } 841 842 /* mprotect_fixup is overkill to remove the temporary stack flags */ 843 vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP); 844 845 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */ 846 stack_size = vma->vm_end - vma->vm_start; 847 /* 848 * Align this down to a page boundary as expand_stack 849 * will align it up. 850 */ 851 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK; 852 853 stack_expand = min(rlim_stack, stack_size + stack_expand); 854 855 #ifdef CONFIG_STACK_GROWSUP 856 stack_base = vma->vm_start + stack_expand; 857 #else 858 stack_base = vma->vm_end - stack_expand; 859 #endif 860 current->mm->start_stack = bprm->p; 861 ret = expand_stack_locked(vma, stack_base); 862 if (ret) 863 ret = -EFAULT; 864 865 out_unlock: 866 mmap_write_unlock(mm); 867 return ret; 868 } 869 EXPORT_SYMBOL(setup_arg_pages); 870 871 #else 872 873 /* 874 * Transfer the program arguments and environment from the holding pages 875 * onto the stack. The provided stack pointer is adjusted accordingly. 876 */ 877 int transfer_args_to_stack(struct linux_binprm *bprm, 878 unsigned long *sp_location) 879 { 880 unsigned long index, stop, sp; 881 int ret = 0; 882 883 stop = bprm->p >> PAGE_SHIFT; 884 sp = *sp_location; 885 886 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) { 887 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0; 888 char *src = kmap_local_page(bprm->page[index]) + offset; 889 sp -= PAGE_SIZE - offset; 890 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0) 891 ret = -EFAULT; 892 kunmap_local(src); 893 if (ret) 894 goto out; 895 } 896 897 *sp_location = sp; 898 899 out: 900 return ret; 901 } 902 EXPORT_SYMBOL(transfer_args_to_stack); 903 904 #endif /* CONFIG_MMU */ 905 906 static struct file *do_open_execat(int fd, struct filename *name, int flags) 907 { 908 struct file *file; 909 int err; 910 struct open_flags open_exec_flags = { 911 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC, 912 .acc_mode = MAY_EXEC, 913 .intent = LOOKUP_OPEN, 914 .lookup_flags = LOOKUP_FOLLOW, 915 }; 916 917 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0) 918 return ERR_PTR(-EINVAL); 919 if (flags & AT_SYMLINK_NOFOLLOW) 920 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW; 921 if (flags & AT_EMPTY_PATH) 922 open_exec_flags.lookup_flags |= LOOKUP_EMPTY; 923 924 file = do_filp_open(fd, name, &open_exec_flags); 925 if (IS_ERR(file)) 926 goto out; 927 928 /* 929 * may_open() has already checked for this, so it should be 930 * impossible to trip now. But we need to be extra cautious 931 * and check again at the very end too. 932 */ 933 err = -EACCES; 934 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) || 935 path_noexec(&file->f_path))) 936 goto exit; 937 938 err = deny_write_access(file); 939 if (err) 940 goto exit; 941 942 out: 943 return file; 944 945 exit: 946 fput(file); 947 return ERR_PTR(err); 948 } 949 950 struct file *open_exec(const char *name) 951 { 952 struct filename *filename = getname_kernel(name); 953 struct file *f = ERR_CAST(filename); 954 955 if (!IS_ERR(filename)) { 956 f = do_open_execat(AT_FDCWD, filename, 0); 957 putname(filename); 958 } 959 return f; 960 } 961 EXPORT_SYMBOL(open_exec); 962 963 #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC) 964 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len) 965 { 966 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos); 967 if (res > 0) 968 flush_icache_user_range(addr, addr + len); 969 return res; 970 } 971 EXPORT_SYMBOL(read_code); 972 #endif 973 974 /* 975 * Maps the mm_struct mm into the current task struct. 976 * On success, this function returns with exec_update_lock 977 * held for writing. 978 */ 979 static int exec_mmap(struct mm_struct *mm) 980 { 981 struct task_struct *tsk; 982 struct mm_struct *old_mm, *active_mm; 983 int ret; 984 985 /* Notify parent that we're no longer interested in the old VM */ 986 tsk = current; 987 old_mm = current->mm; 988 exec_mm_release(tsk, old_mm); 989 if (old_mm) 990 sync_mm_rss(old_mm); 991 992 ret = down_write_killable(&tsk->signal->exec_update_lock); 993 if (ret) 994 return ret; 995 996 if (old_mm) { 997 /* 998 * If there is a pending fatal signal perhaps a signal 999 * whose default action is to create a coredump get 1000 * out and die instead of going through with the exec. 1001 */ 1002 ret = mmap_read_lock_killable(old_mm); 1003 if (ret) { 1004 up_write(&tsk->signal->exec_update_lock); 1005 return ret; 1006 } 1007 } 1008 1009 task_lock(tsk); 1010 membarrier_exec_mmap(mm); 1011 1012 local_irq_disable(); 1013 active_mm = tsk->active_mm; 1014 tsk->active_mm = mm; 1015 tsk->mm = mm; 1016 mm_init_cid(mm); 1017 /* 1018 * This prevents preemption while active_mm is being loaded and 1019 * it and mm are being updated, which could cause problems for 1020 * lazy tlb mm refcounting when these are updated by context 1021 * switches. Not all architectures can handle irqs off over 1022 * activate_mm yet. 1023 */ 1024 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM)) 1025 local_irq_enable(); 1026 activate_mm(active_mm, mm); 1027 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM)) 1028 local_irq_enable(); 1029 lru_gen_add_mm(mm); 1030 task_unlock(tsk); 1031 lru_gen_use_mm(mm); 1032 if (old_mm) { 1033 mmap_read_unlock(old_mm); 1034 BUG_ON(active_mm != old_mm); 1035 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm); 1036 mm_update_next_owner(old_mm); 1037 mmput(old_mm); 1038 return 0; 1039 } 1040 mmdrop_lazy_tlb(active_mm); 1041 return 0; 1042 } 1043 1044 static int de_thread(struct task_struct *tsk) 1045 { 1046 struct signal_struct *sig = tsk->signal; 1047 struct sighand_struct *oldsighand = tsk->sighand; 1048 spinlock_t *lock = &oldsighand->siglock; 1049 1050 if (thread_group_empty(tsk)) 1051 goto no_thread_group; 1052 1053 /* 1054 * Kill all other threads in the thread group. 1055 */ 1056 spin_lock_irq(lock); 1057 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) { 1058 /* 1059 * Another group action in progress, just 1060 * return so that the signal is processed. 1061 */ 1062 spin_unlock_irq(lock); 1063 return -EAGAIN; 1064 } 1065 1066 sig->group_exec_task = tsk; 1067 sig->notify_count = zap_other_threads(tsk); 1068 if (!thread_group_leader(tsk)) 1069 sig->notify_count--; 1070 1071 while (sig->notify_count) { 1072 __set_current_state(TASK_KILLABLE); 1073 spin_unlock_irq(lock); 1074 schedule(); 1075 if (__fatal_signal_pending(tsk)) 1076 goto killed; 1077 spin_lock_irq(lock); 1078 } 1079 spin_unlock_irq(lock); 1080 1081 /* 1082 * At this point all other threads have exited, all we have to 1083 * do is to wait for the thread group leader to become inactive, 1084 * and to assume its PID: 1085 */ 1086 if (!thread_group_leader(tsk)) { 1087 struct task_struct *leader = tsk->group_leader; 1088 1089 for (;;) { 1090 cgroup_threadgroup_change_begin(tsk); 1091 write_lock_irq(&tasklist_lock); 1092 /* 1093 * Do this under tasklist_lock to ensure that 1094 * exit_notify() can't miss ->group_exec_task 1095 */ 1096 sig->notify_count = -1; 1097 if (likely(leader->exit_state)) 1098 break; 1099 __set_current_state(TASK_KILLABLE); 1100 write_unlock_irq(&tasklist_lock); 1101 cgroup_threadgroup_change_end(tsk); 1102 schedule(); 1103 if (__fatal_signal_pending(tsk)) 1104 goto killed; 1105 } 1106 1107 /* 1108 * The only record we have of the real-time age of a 1109 * process, regardless of execs it's done, is start_time. 1110 * All the past CPU time is accumulated in signal_struct 1111 * from sister threads now dead. But in this non-leader 1112 * exec, nothing survives from the original leader thread, 1113 * whose birth marks the true age of this process now. 1114 * When we take on its identity by switching to its PID, we 1115 * also take its birthdate (always earlier than our own). 1116 */ 1117 tsk->start_time = leader->start_time; 1118 tsk->start_boottime = leader->start_boottime; 1119 1120 BUG_ON(!same_thread_group(leader, tsk)); 1121 /* 1122 * An exec() starts a new thread group with the 1123 * TGID of the previous thread group. Rehash the 1124 * two threads with a switched PID, and release 1125 * the former thread group leader: 1126 */ 1127 1128 /* Become a process group leader with the old leader's pid. 1129 * The old leader becomes a thread of the this thread group. 1130 */ 1131 exchange_tids(tsk, leader); 1132 transfer_pid(leader, tsk, PIDTYPE_TGID); 1133 transfer_pid(leader, tsk, PIDTYPE_PGID); 1134 transfer_pid(leader, tsk, PIDTYPE_SID); 1135 1136 list_replace_rcu(&leader->tasks, &tsk->tasks); 1137 list_replace_init(&leader->sibling, &tsk->sibling); 1138 1139 tsk->group_leader = tsk; 1140 leader->group_leader = tsk; 1141 1142 tsk->exit_signal = SIGCHLD; 1143 leader->exit_signal = -1; 1144 1145 BUG_ON(leader->exit_state != EXIT_ZOMBIE); 1146 leader->exit_state = EXIT_DEAD; 1147 1148 /* 1149 * We are going to release_task()->ptrace_unlink() silently, 1150 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees 1151 * the tracer won't block again waiting for this thread. 1152 */ 1153 if (unlikely(leader->ptrace)) 1154 __wake_up_parent(leader, leader->parent); 1155 write_unlock_irq(&tasklist_lock); 1156 cgroup_threadgroup_change_end(tsk); 1157 1158 release_task(leader); 1159 } 1160 1161 sig->group_exec_task = NULL; 1162 sig->notify_count = 0; 1163 1164 no_thread_group: 1165 /* we have changed execution domain */ 1166 tsk->exit_signal = SIGCHLD; 1167 1168 BUG_ON(!thread_group_leader(tsk)); 1169 return 0; 1170 1171 killed: 1172 /* protects against exit_notify() and __exit_signal() */ 1173 read_lock(&tasklist_lock); 1174 sig->group_exec_task = NULL; 1175 sig->notify_count = 0; 1176 read_unlock(&tasklist_lock); 1177 return -EAGAIN; 1178 } 1179 1180 1181 /* 1182 * This function makes sure the current process has its own signal table, 1183 * so that flush_signal_handlers can later reset the handlers without 1184 * disturbing other processes. (Other processes might share the signal 1185 * table via the CLONE_SIGHAND option to clone().) 1186 */ 1187 static int unshare_sighand(struct task_struct *me) 1188 { 1189 struct sighand_struct *oldsighand = me->sighand; 1190 1191 if (refcount_read(&oldsighand->count) != 1) { 1192 struct sighand_struct *newsighand; 1193 /* 1194 * This ->sighand is shared with the CLONE_SIGHAND 1195 * but not CLONE_THREAD task, switch to the new one. 1196 */ 1197 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1198 if (!newsighand) 1199 return -ENOMEM; 1200 1201 refcount_set(&newsighand->count, 1); 1202 1203 write_lock_irq(&tasklist_lock); 1204 spin_lock(&oldsighand->siglock); 1205 memcpy(newsighand->action, oldsighand->action, 1206 sizeof(newsighand->action)); 1207 rcu_assign_pointer(me->sighand, newsighand); 1208 spin_unlock(&oldsighand->siglock); 1209 write_unlock_irq(&tasklist_lock); 1210 1211 __cleanup_sighand(oldsighand); 1212 } 1213 return 0; 1214 } 1215 1216 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk) 1217 { 1218 task_lock(tsk); 1219 /* Always NUL terminated and zero-padded */ 1220 strscpy_pad(buf, tsk->comm, buf_size); 1221 task_unlock(tsk); 1222 return buf; 1223 } 1224 EXPORT_SYMBOL_GPL(__get_task_comm); 1225 1226 /* 1227 * These functions flushes out all traces of the currently running executable 1228 * so that a new one can be started 1229 */ 1230 1231 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec) 1232 { 1233 task_lock(tsk); 1234 trace_task_rename(tsk, buf); 1235 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm)); 1236 task_unlock(tsk); 1237 perf_event_comm(tsk, exec); 1238 } 1239 1240 /* 1241 * Calling this is the point of no return. None of the failures will be 1242 * seen by userspace since either the process is already taking a fatal 1243 * signal (via de_thread() or coredump), or will have SEGV raised 1244 * (after exec_mmap()) by search_binary_handler (see below). 1245 */ 1246 int begin_new_exec(struct linux_binprm * bprm) 1247 { 1248 struct task_struct *me = current; 1249 int retval; 1250 1251 /* Once we are committed compute the creds */ 1252 retval = bprm_creds_from_file(bprm); 1253 if (retval) 1254 return retval; 1255 1256 /* 1257 * Ensure all future errors are fatal. 1258 */ 1259 bprm->point_of_no_return = true; 1260 1261 /* 1262 * Make this the only thread in the thread group. 1263 */ 1264 retval = de_thread(me); 1265 if (retval) 1266 goto out; 1267 1268 /* 1269 * Cancel any io_uring activity across execve 1270 */ 1271 io_uring_task_cancel(); 1272 1273 /* Ensure the files table is not shared. */ 1274 retval = unshare_files(); 1275 if (retval) 1276 goto out; 1277 1278 /* 1279 * Must be called _before_ exec_mmap() as bprm->mm is 1280 * not visible until then. Doing it here also ensures 1281 * we don't race against replace_mm_exe_file(). 1282 */ 1283 retval = set_mm_exe_file(bprm->mm, bprm->file); 1284 if (retval) 1285 goto out; 1286 1287 /* If the binary is not readable then enforce mm->dumpable=0 */ 1288 would_dump(bprm, bprm->file); 1289 if (bprm->have_execfd) 1290 would_dump(bprm, bprm->executable); 1291 1292 /* 1293 * Release all of the old mmap stuff 1294 */ 1295 acct_arg_size(bprm, 0); 1296 retval = exec_mmap(bprm->mm); 1297 if (retval) 1298 goto out; 1299 1300 bprm->mm = NULL; 1301 1302 retval = exec_task_namespaces(); 1303 if (retval) 1304 goto out_unlock; 1305 1306 #ifdef CONFIG_POSIX_TIMERS 1307 spin_lock_irq(&me->sighand->siglock); 1308 posix_cpu_timers_exit(me); 1309 spin_unlock_irq(&me->sighand->siglock); 1310 exit_itimers(me); 1311 flush_itimer_signals(); 1312 #endif 1313 1314 /* 1315 * Make the signal table private. 1316 */ 1317 retval = unshare_sighand(me); 1318 if (retval) 1319 goto out_unlock; 1320 1321 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | 1322 PF_NOFREEZE | PF_NO_SETAFFINITY); 1323 flush_thread(); 1324 me->personality &= ~bprm->per_clear; 1325 1326 clear_syscall_work_syscall_user_dispatch(me); 1327 1328 /* 1329 * We have to apply CLOEXEC before we change whether the process is 1330 * dumpable (in setup_new_exec) to avoid a race with a process in userspace 1331 * trying to access the should-be-closed file descriptors of a process 1332 * undergoing exec(2). 1333 */ 1334 do_close_on_exec(me->files); 1335 1336 if (bprm->secureexec) { 1337 /* Make sure parent cannot signal privileged process. */ 1338 me->pdeath_signal = 0; 1339 1340 /* 1341 * For secureexec, reset the stack limit to sane default to 1342 * avoid bad behavior from the prior rlimits. This has to 1343 * happen before arch_pick_mmap_layout(), which examines 1344 * RLIMIT_STACK, but after the point of no return to avoid 1345 * needing to clean up the change on failure. 1346 */ 1347 if (bprm->rlim_stack.rlim_cur > _STK_LIM) 1348 bprm->rlim_stack.rlim_cur = _STK_LIM; 1349 } 1350 1351 me->sas_ss_sp = me->sas_ss_size = 0; 1352 1353 /* 1354 * Figure out dumpability. Note that this checking only of current 1355 * is wrong, but userspace depends on it. This should be testing 1356 * bprm->secureexec instead. 1357 */ 1358 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP || 1359 !(uid_eq(current_euid(), current_uid()) && 1360 gid_eq(current_egid(), current_gid()))) 1361 set_dumpable(current->mm, suid_dumpable); 1362 else 1363 set_dumpable(current->mm, SUID_DUMP_USER); 1364 1365 perf_event_exec(); 1366 __set_task_comm(me, kbasename(bprm->filename), true); 1367 1368 /* An exec changes our domain. We are no longer part of the thread 1369 group */ 1370 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1); 1371 flush_signal_handlers(me, 0); 1372 1373 retval = set_cred_ucounts(bprm->cred); 1374 if (retval < 0) 1375 goto out_unlock; 1376 1377 /* 1378 * install the new credentials for this executable 1379 */ 1380 security_bprm_committing_creds(bprm); 1381 1382 commit_creds(bprm->cred); 1383 bprm->cred = NULL; 1384 1385 /* 1386 * Disable monitoring for regular users 1387 * when executing setuid binaries. Must 1388 * wait until new credentials are committed 1389 * by commit_creds() above 1390 */ 1391 if (get_dumpable(me->mm) != SUID_DUMP_USER) 1392 perf_event_exit_task(me); 1393 /* 1394 * cred_guard_mutex must be held at least to this point to prevent 1395 * ptrace_attach() from altering our determination of the task's 1396 * credentials; any time after this it may be unlocked. 1397 */ 1398 security_bprm_committed_creds(bprm); 1399 1400 /* Pass the opened binary to the interpreter. */ 1401 if (bprm->have_execfd) { 1402 retval = get_unused_fd_flags(0); 1403 if (retval < 0) 1404 goto out_unlock; 1405 fd_install(retval, bprm->executable); 1406 bprm->executable = NULL; 1407 bprm->execfd = retval; 1408 } 1409 return 0; 1410 1411 out_unlock: 1412 up_write(&me->signal->exec_update_lock); 1413 out: 1414 return retval; 1415 } 1416 EXPORT_SYMBOL(begin_new_exec); 1417 1418 void would_dump(struct linux_binprm *bprm, struct file *file) 1419 { 1420 struct inode *inode = file_inode(file); 1421 struct mnt_idmap *idmap = file_mnt_idmap(file); 1422 if (inode_permission(idmap, inode, MAY_READ) < 0) { 1423 struct user_namespace *old, *user_ns; 1424 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP; 1425 1426 /* Ensure mm->user_ns contains the executable */ 1427 user_ns = old = bprm->mm->user_ns; 1428 while ((user_ns != &init_user_ns) && 1429 !privileged_wrt_inode_uidgid(user_ns, idmap, inode)) 1430 user_ns = user_ns->parent; 1431 1432 if (old != user_ns) { 1433 bprm->mm->user_ns = get_user_ns(user_ns); 1434 put_user_ns(old); 1435 } 1436 } 1437 } 1438 EXPORT_SYMBOL(would_dump); 1439 1440 void setup_new_exec(struct linux_binprm * bprm) 1441 { 1442 /* Setup things that can depend upon the personality */ 1443 struct task_struct *me = current; 1444 1445 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack); 1446 1447 arch_setup_new_exec(); 1448 1449 /* Set the new mm task size. We have to do that late because it may 1450 * depend on TIF_32BIT which is only updated in flush_thread() on 1451 * some architectures like powerpc 1452 */ 1453 me->mm->task_size = TASK_SIZE; 1454 up_write(&me->signal->exec_update_lock); 1455 mutex_unlock(&me->signal->cred_guard_mutex); 1456 } 1457 EXPORT_SYMBOL(setup_new_exec); 1458 1459 /* Runs immediately before start_thread() takes over. */ 1460 void finalize_exec(struct linux_binprm *bprm) 1461 { 1462 /* Store any stack rlimit changes before starting thread. */ 1463 task_lock(current->group_leader); 1464 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack; 1465 task_unlock(current->group_leader); 1466 } 1467 EXPORT_SYMBOL(finalize_exec); 1468 1469 /* 1470 * Prepare credentials and lock ->cred_guard_mutex. 1471 * setup_new_exec() commits the new creds and drops the lock. 1472 * Or, if exec fails before, free_bprm() should release ->cred 1473 * and unlock. 1474 */ 1475 static int prepare_bprm_creds(struct linux_binprm *bprm) 1476 { 1477 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex)) 1478 return -ERESTARTNOINTR; 1479 1480 bprm->cred = prepare_exec_creds(); 1481 if (likely(bprm->cred)) 1482 return 0; 1483 1484 mutex_unlock(¤t->signal->cred_guard_mutex); 1485 return -ENOMEM; 1486 } 1487 1488 static void free_bprm(struct linux_binprm *bprm) 1489 { 1490 if (bprm->mm) { 1491 acct_arg_size(bprm, 0); 1492 mmput(bprm->mm); 1493 } 1494 free_arg_pages(bprm); 1495 if (bprm->cred) { 1496 mutex_unlock(¤t->signal->cred_guard_mutex); 1497 abort_creds(bprm->cred); 1498 } 1499 if (bprm->file) { 1500 allow_write_access(bprm->file); 1501 fput(bprm->file); 1502 } 1503 if (bprm->executable) 1504 fput(bprm->executable); 1505 /* If a binfmt changed the interp, free it. */ 1506 if (bprm->interp != bprm->filename) 1507 kfree(bprm->interp); 1508 kfree(bprm->fdpath); 1509 kfree(bprm); 1510 } 1511 1512 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename) 1513 { 1514 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL); 1515 int retval = -ENOMEM; 1516 if (!bprm) 1517 goto out; 1518 1519 if (fd == AT_FDCWD || filename->name[0] == '/') { 1520 bprm->filename = filename->name; 1521 } else { 1522 if (filename->name[0] == '\0') 1523 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd); 1524 else 1525 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s", 1526 fd, filename->name); 1527 if (!bprm->fdpath) 1528 goto out_free; 1529 1530 bprm->filename = bprm->fdpath; 1531 } 1532 bprm->interp = bprm->filename; 1533 1534 retval = bprm_mm_init(bprm); 1535 if (retval) 1536 goto out_free; 1537 return bprm; 1538 1539 out_free: 1540 free_bprm(bprm); 1541 out: 1542 return ERR_PTR(retval); 1543 } 1544 1545 int bprm_change_interp(const char *interp, struct linux_binprm *bprm) 1546 { 1547 /* If a binfmt changed the interp, free it first. */ 1548 if (bprm->interp != bprm->filename) 1549 kfree(bprm->interp); 1550 bprm->interp = kstrdup(interp, GFP_KERNEL); 1551 if (!bprm->interp) 1552 return -ENOMEM; 1553 return 0; 1554 } 1555 EXPORT_SYMBOL(bprm_change_interp); 1556 1557 /* 1558 * determine how safe it is to execute the proposed program 1559 * - the caller must hold ->cred_guard_mutex to protect against 1560 * PTRACE_ATTACH or seccomp thread-sync 1561 */ 1562 static void check_unsafe_exec(struct linux_binprm *bprm) 1563 { 1564 struct task_struct *p = current, *t; 1565 unsigned n_fs; 1566 1567 if (p->ptrace) 1568 bprm->unsafe |= LSM_UNSAFE_PTRACE; 1569 1570 /* 1571 * This isn't strictly necessary, but it makes it harder for LSMs to 1572 * mess up. 1573 */ 1574 if (task_no_new_privs(current)) 1575 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS; 1576 1577 /* 1578 * If another task is sharing our fs, we cannot safely 1579 * suid exec because the differently privileged task 1580 * will be able to manipulate the current directory, etc. 1581 * It would be nice to force an unshare instead... 1582 */ 1583 t = p; 1584 n_fs = 1; 1585 spin_lock(&p->fs->lock); 1586 rcu_read_lock(); 1587 while_each_thread(p, t) { 1588 if (t->fs == p->fs) 1589 n_fs++; 1590 } 1591 rcu_read_unlock(); 1592 1593 if (p->fs->users > n_fs) 1594 bprm->unsafe |= LSM_UNSAFE_SHARE; 1595 else 1596 p->fs->in_exec = 1; 1597 spin_unlock(&p->fs->lock); 1598 } 1599 1600 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file) 1601 { 1602 /* Handle suid and sgid on files */ 1603 struct mnt_idmap *idmap; 1604 struct inode *inode = file_inode(file); 1605 unsigned int mode; 1606 vfsuid_t vfsuid; 1607 vfsgid_t vfsgid; 1608 1609 if (!mnt_may_suid(file->f_path.mnt)) 1610 return; 1611 1612 if (task_no_new_privs(current)) 1613 return; 1614 1615 mode = READ_ONCE(inode->i_mode); 1616 if (!(mode & (S_ISUID|S_ISGID))) 1617 return; 1618 1619 idmap = file_mnt_idmap(file); 1620 1621 /* Be careful if suid/sgid is set */ 1622 inode_lock(inode); 1623 1624 /* reload atomically mode/uid/gid now that lock held */ 1625 mode = inode->i_mode; 1626 vfsuid = i_uid_into_vfsuid(idmap, inode); 1627 vfsgid = i_gid_into_vfsgid(idmap, inode); 1628 inode_unlock(inode); 1629 1630 /* We ignore suid/sgid if there are no mappings for them in the ns */ 1631 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) || 1632 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid)) 1633 return; 1634 1635 if (mode & S_ISUID) { 1636 bprm->per_clear |= PER_CLEAR_ON_SETID; 1637 bprm->cred->euid = vfsuid_into_kuid(vfsuid); 1638 } 1639 1640 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) { 1641 bprm->per_clear |= PER_CLEAR_ON_SETID; 1642 bprm->cred->egid = vfsgid_into_kgid(vfsgid); 1643 } 1644 } 1645 1646 /* 1647 * Compute brpm->cred based upon the final binary. 1648 */ 1649 static int bprm_creds_from_file(struct linux_binprm *bprm) 1650 { 1651 /* Compute creds based on which file? */ 1652 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file; 1653 1654 bprm_fill_uid(bprm, file); 1655 return security_bprm_creds_from_file(bprm, file); 1656 } 1657 1658 /* 1659 * Fill the binprm structure from the inode. 1660 * Read the first BINPRM_BUF_SIZE bytes 1661 * 1662 * This may be called multiple times for binary chains (scripts for example). 1663 */ 1664 static int prepare_binprm(struct linux_binprm *bprm) 1665 { 1666 loff_t pos = 0; 1667 1668 memset(bprm->buf, 0, BINPRM_BUF_SIZE); 1669 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos); 1670 } 1671 1672 /* 1673 * Arguments are '\0' separated strings found at the location bprm->p 1674 * points to; chop off the first by relocating brpm->p to right after 1675 * the first '\0' encountered. 1676 */ 1677 int remove_arg_zero(struct linux_binprm *bprm) 1678 { 1679 int ret = 0; 1680 unsigned long offset; 1681 char *kaddr; 1682 struct page *page; 1683 1684 if (!bprm->argc) 1685 return 0; 1686 1687 do { 1688 offset = bprm->p & ~PAGE_MASK; 1689 page = get_arg_page(bprm, bprm->p, 0); 1690 if (!page) { 1691 ret = -EFAULT; 1692 goto out; 1693 } 1694 kaddr = kmap_local_page(page); 1695 1696 for (; offset < PAGE_SIZE && kaddr[offset]; 1697 offset++, bprm->p++) 1698 ; 1699 1700 kunmap_local(kaddr); 1701 put_arg_page(page); 1702 } while (offset == PAGE_SIZE); 1703 1704 bprm->p++; 1705 bprm->argc--; 1706 ret = 0; 1707 1708 out: 1709 return ret; 1710 } 1711 EXPORT_SYMBOL(remove_arg_zero); 1712 1713 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e)) 1714 /* 1715 * cycle the list of binary formats handler, until one recognizes the image 1716 */ 1717 static int search_binary_handler(struct linux_binprm *bprm) 1718 { 1719 bool need_retry = IS_ENABLED(CONFIG_MODULES); 1720 struct linux_binfmt *fmt; 1721 int retval; 1722 1723 retval = prepare_binprm(bprm); 1724 if (retval < 0) 1725 return retval; 1726 1727 retval = security_bprm_check(bprm); 1728 if (retval) 1729 return retval; 1730 1731 retval = -ENOENT; 1732 retry: 1733 read_lock(&binfmt_lock); 1734 list_for_each_entry(fmt, &formats, lh) { 1735 if (!try_module_get(fmt->module)) 1736 continue; 1737 read_unlock(&binfmt_lock); 1738 1739 retval = fmt->load_binary(bprm); 1740 1741 read_lock(&binfmt_lock); 1742 put_binfmt(fmt); 1743 if (bprm->point_of_no_return || (retval != -ENOEXEC)) { 1744 read_unlock(&binfmt_lock); 1745 return retval; 1746 } 1747 } 1748 read_unlock(&binfmt_lock); 1749 1750 if (need_retry) { 1751 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) && 1752 printable(bprm->buf[2]) && printable(bprm->buf[3])) 1753 return retval; 1754 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0) 1755 return retval; 1756 need_retry = false; 1757 goto retry; 1758 } 1759 1760 return retval; 1761 } 1762 1763 /* binfmt handlers will call back into begin_new_exec() on success. */ 1764 static int exec_binprm(struct linux_binprm *bprm) 1765 { 1766 pid_t old_pid, old_vpid; 1767 int ret, depth; 1768 1769 /* Need to fetch pid before load_binary changes it */ 1770 old_pid = current->pid; 1771 rcu_read_lock(); 1772 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent)); 1773 rcu_read_unlock(); 1774 1775 /* This allows 4 levels of binfmt rewrites before failing hard. */ 1776 for (depth = 0;; depth++) { 1777 struct file *exec; 1778 if (depth > 5) 1779 return -ELOOP; 1780 1781 ret = search_binary_handler(bprm); 1782 if (ret < 0) 1783 return ret; 1784 if (!bprm->interpreter) 1785 break; 1786 1787 exec = bprm->file; 1788 bprm->file = bprm->interpreter; 1789 bprm->interpreter = NULL; 1790 1791 allow_write_access(exec); 1792 if (unlikely(bprm->have_execfd)) { 1793 if (bprm->executable) { 1794 fput(exec); 1795 return -ENOEXEC; 1796 } 1797 bprm->executable = exec; 1798 } else 1799 fput(exec); 1800 } 1801 1802 audit_bprm(bprm); 1803 trace_sched_process_exec(current, old_pid, bprm); 1804 ptrace_event(PTRACE_EVENT_EXEC, old_vpid); 1805 proc_exec_connector(current); 1806 return 0; 1807 } 1808 1809 /* 1810 * sys_execve() executes a new program. 1811 */ 1812 static int bprm_execve(struct linux_binprm *bprm, 1813 int fd, struct filename *filename, int flags) 1814 { 1815 struct file *file; 1816 int retval; 1817 1818 retval = prepare_bprm_creds(bprm); 1819 if (retval) 1820 return retval; 1821 1822 /* 1823 * Check for unsafe execution states before exec_binprm(), which 1824 * will call back into begin_new_exec(), into bprm_creds_from_file(), 1825 * where setuid-ness is evaluated. 1826 */ 1827 check_unsafe_exec(bprm); 1828 current->in_execve = 1; 1829 sched_mm_cid_before_execve(current); 1830 1831 file = do_open_execat(fd, filename, flags); 1832 retval = PTR_ERR(file); 1833 if (IS_ERR(file)) 1834 goto out_unmark; 1835 1836 sched_exec(); 1837 1838 bprm->file = file; 1839 /* 1840 * Record that a name derived from an O_CLOEXEC fd will be 1841 * inaccessible after exec. This allows the code in exec to 1842 * choose to fail when the executable is not mmaped into the 1843 * interpreter and an open file descriptor is not passed to 1844 * the interpreter. This makes for a better user experience 1845 * than having the interpreter start and then immediately fail 1846 * when it finds the executable is inaccessible. 1847 */ 1848 if (bprm->fdpath && get_close_on_exec(fd)) 1849 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE; 1850 1851 /* Set the unchanging part of bprm->cred */ 1852 retval = security_bprm_creds_for_exec(bprm); 1853 if (retval) 1854 goto out; 1855 1856 retval = exec_binprm(bprm); 1857 if (retval < 0) 1858 goto out; 1859 1860 sched_mm_cid_after_execve(current); 1861 /* execve succeeded */ 1862 current->fs->in_exec = 0; 1863 current->in_execve = 0; 1864 rseq_execve(current); 1865 user_events_execve(current); 1866 acct_update_integrals(current); 1867 task_numa_free(current, false); 1868 return retval; 1869 1870 out: 1871 /* 1872 * If past the point of no return ensure the code never 1873 * returns to the userspace process. Use an existing fatal 1874 * signal if present otherwise terminate the process with 1875 * SIGSEGV. 1876 */ 1877 if (bprm->point_of_no_return && !fatal_signal_pending(current)) 1878 force_fatal_sig(SIGSEGV); 1879 1880 out_unmark: 1881 sched_mm_cid_after_execve(current); 1882 current->fs->in_exec = 0; 1883 current->in_execve = 0; 1884 1885 return retval; 1886 } 1887 1888 static int do_execveat_common(int fd, struct filename *filename, 1889 struct user_arg_ptr argv, 1890 struct user_arg_ptr envp, 1891 int flags) 1892 { 1893 struct linux_binprm *bprm; 1894 int retval; 1895 1896 if (IS_ERR(filename)) 1897 return PTR_ERR(filename); 1898 1899 /* 1900 * We move the actual failure in case of RLIMIT_NPROC excess from 1901 * set*uid() to execve() because too many poorly written programs 1902 * don't check setuid() return code. Here we additionally recheck 1903 * whether NPROC limit is still exceeded. 1904 */ 1905 if ((current->flags & PF_NPROC_EXCEEDED) && 1906 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) { 1907 retval = -EAGAIN; 1908 goto out_ret; 1909 } 1910 1911 /* We're below the limit (still or again), so we don't want to make 1912 * further execve() calls fail. */ 1913 current->flags &= ~PF_NPROC_EXCEEDED; 1914 1915 bprm = alloc_bprm(fd, filename); 1916 if (IS_ERR(bprm)) { 1917 retval = PTR_ERR(bprm); 1918 goto out_ret; 1919 } 1920 1921 retval = count(argv, MAX_ARG_STRINGS); 1922 if (retval == 0) 1923 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n", 1924 current->comm, bprm->filename); 1925 if (retval < 0) 1926 goto out_free; 1927 bprm->argc = retval; 1928 1929 retval = count(envp, MAX_ARG_STRINGS); 1930 if (retval < 0) 1931 goto out_free; 1932 bprm->envc = retval; 1933 1934 retval = bprm_stack_limits(bprm); 1935 if (retval < 0) 1936 goto out_free; 1937 1938 retval = copy_string_kernel(bprm->filename, bprm); 1939 if (retval < 0) 1940 goto out_free; 1941 bprm->exec = bprm->p; 1942 1943 retval = copy_strings(bprm->envc, envp, bprm); 1944 if (retval < 0) 1945 goto out_free; 1946 1947 retval = copy_strings(bprm->argc, argv, bprm); 1948 if (retval < 0) 1949 goto out_free; 1950 1951 /* 1952 * When argv is empty, add an empty string ("") as argv[0] to 1953 * ensure confused userspace programs that start processing 1954 * from argv[1] won't end up walking envp. See also 1955 * bprm_stack_limits(). 1956 */ 1957 if (bprm->argc == 0) { 1958 retval = copy_string_kernel("", bprm); 1959 if (retval < 0) 1960 goto out_free; 1961 bprm->argc = 1; 1962 } 1963 1964 retval = bprm_execve(bprm, fd, filename, flags); 1965 out_free: 1966 free_bprm(bprm); 1967 1968 out_ret: 1969 putname(filename); 1970 return retval; 1971 } 1972 1973 int kernel_execve(const char *kernel_filename, 1974 const char *const *argv, const char *const *envp) 1975 { 1976 struct filename *filename; 1977 struct linux_binprm *bprm; 1978 int fd = AT_FDCWD; 1979 int retval; 1980 1981 /* It is non-sense for kernel threads to call execve */ 1982 if (WARN_ON_ONCE(current->flags & PF_KTHREAD)) 1983 return -EINVAL; 1984 1985 filename = getname_kernel(kernel_filename); 1986 if (IS_ERR(filename)) 1987 return PTR_ERR(filename); 1988 1989 bprm = alloc_bprm(fd, filename); 1990 if (IS_ERR(bprm)) { 1991 retval = PTR_ERR(bprm); 1992 goto out_ret; 1993 } 1994 1995 retval = count_strings_kernel(argv); 1996 if (WARN_ON_ONCE(retval == 0)) 1997 retval = -EINVAL; 1998 if (retval < 0) 1999 goto out_free; 2000 bprm->argc = retval; 2001 2002 retval = count_strings_kernel(envp); 2003 if (retval < 0) 2004 goto out_free; 2005 bprm->envc = retval; 2006 2007 retval = bprm_stack_limits(bprm); 2008 if (retval < 0) 2009 goto out_free; 2010 2011 retval = copy_string_kernel(bprm->filename, bprm); 2012 if (retval < 0) 2013 goto out_free; 2014 bprm->exec = bprm->p; 2015 2016 retval = copy_strings_kernel(bprm->envc, envp, bprm); 2017 if (retval < 0) 2018 goto out_free; 2019 2020 retval = copy_strings_kernel(bprm->argc, argv, bprm); 2021 if (retval < 0) 2022 goto out_free; 2023 2024 retval = bprm_execve(bprm, fd, filename, 0); 2025 out_free: 2026 free_bprm(bprm); 2027 out_ret: 2028 putname(filename); 2029 return retval; 2030 } 2031 2032 static int do_execve(struct filename *filename, 2033 const char __user *const __user *__argv, 2034 const char __user *const __user *__envp) 2035 { 2036 struct user_arg_ptr argv = { .ptr.native = __argv }; 2037 struct user_arg_ptr envp = { .ptr.native = __envp }; 2038 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0); 2039 } 2040 2041 static int do_execveat(int fd, struct filename *filename, 2042 const char __user *const __user *__argv, 2043 const char __user *const __user *__envp, 2044 int flags) 2045 { 2046 struct user_arg_ptr argv = { .ptr.native = __argv }; 2047 struct user_arg_ptr envp = { .ptr.native = __envp }; 2048 2049 return do_execveat_common(fd, filename, argv, envp, flags); 2050 } 2051 2052 #ifdef CONFIG_COMPAT 2053 static int compat_do_execve(struct filename *filename, 2054 const compat_uptr_t __user *__argv, 2055 const compat_uptr_t __user *__envp) 2056 { 2057 struct user_arg_ptr argv = { 2058 .is_compat = true, 2059 .ptr.compat = __argv, 2060 }; 2061 struct user_arg_ptr envp = { 2062 .is_compat = true, 2063 .ptr.compat = __envp, 2064 }; 2065 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0); 2066 } 2067 2068 static int compat_do_execveat(int fd, struct filename *filename, 2069 const compat_uptr_t __user *__argv, 2070 const compat_uptr_t __user *__envp, 2071 int flags) 2072 { 2073 struct user_arg_ptr argv = { 2074 .is_compat = true, 2075 .ptr.compat = __argv, 2076 }; 2077 struct user_arg_ptr envp = { 2078 .is_compat = true, 2079 .ptr.compat = __envp, 2080 }; 2081 return do_execveat_common(fd, filename, argv, envp, flags); 2082 } 2083 #endif 2084 2085 void set_binfmt(struct linux_binfmt *new) 2086 { 2087 struct mm_struct *mm = current->mm; 2088 2089 if (mm->binfmt) 2090 module_put(mm->binfmt->module); 2091 2092 mm->binfmt = new; 2093 if (new) 2094 __module_get(new->module); 2095 } 2096 EXPORT_SYMBOL(set_binfmt); 2097 2098 /* 2099 * set_dumpable stores three-value SUID_DUMP_* into mm->flags. 2100 */ 2101 void set_dumpable(struct mm_struct *mm, int value) 2102 { 2103 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT)) 2104 return; 2105 2106 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value); 2107 } 2108 2109 SYSCALL_DEFINE3(execve, 2110 const char __user *, filename, 2111 const char __user *const __user *, argv, 2112 const char __user *const __user *, envp) 2113 { 2114 return do_execve(getname(filename), argv, envp); 2115 } 2116 2117 SYSCALL_DEFINE5(execveat, 2118 int, fd, const char __user *, filename, 2119 const char __user *const __user *, argv, 2120 const char __user *const __user *, envp, 2121 int, flags) 2122 { 2123 return do_execveat(fd, 2124 getname_uflags(filename, flags), 2125 argv, envp, flags); 2126 } 2127 2128 #ifdef CONFIG_COMPAT 2129 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename, 2130 const compat_uptr_t __user *, argv, 2131 const compat_uptr_t __user *, envp) 2132 { 2133 return compat_do_execve(getname(filename), argv, envp); 2134 } 2135 2136 COMPAT_SYSCALL_DEFINE5(execveat, int, fd, 2137 const char __user *, filename, 2138 const compat_uptr_t __user *, argv, 2139 const compat_uptr_t __user *, envp, 2140 int, flags) 2141 { 2142 return compat_do_execveat(fd, 2143 getname_uflags(filename, flags), 2144 argv, envp, flags); 2145 } 2146 #endif 2147 2148 #ifdef CONFIG_SYSCTL 2149 2150 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write, 2151 void *buffer, size_t *lenp, loff_t *ppos) 2152 { 2153 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 2154 2155 if (!error) 2156 validate_coredump_safety(); 2157 return error; 2158 } 2159 2160 static struct ctl_table fs_exec_sysctls[] = { 2161 { 2162 .procname = "suid_dumpable", 2163 .data = &suid_dumpable, 2164 .maxlen = sizeof(int), 2165 .mode = 0644, 2166 .proc_handler = proc_dointvec_minmax_coredump, 2167 .extra1 = SYSCTL_ZERO, 2168 .extra2 = SYSCTL_TWO, 2169 }, 2170 { } 2171 }; 2172 2173 static int __init init_fs_exec_sysctls(void) 2174 { 2175 register_sysctl_init("fs", fs_exec_sysctls); 2176 return 0; 2177 } 2178 2179 fs_initcall(init_fs_exec_sysctls); 2180 #endif /* CONFIG_SYSCTL */