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 #include <linux/rseq.h> 70 71 #include <linux/uaccess.h> 72 #include <asm/mmu_context.h> 73 #include <asm/tlb.h> 74 75 #include <trace/events/task.h> 76 #include "internal.h" 77 78 #include <trace/events/sched.h> 79 80 static int bprm_creds_from_file(struct linux_binprm *bprm); 81 82 int suid_dumpable = 0; 83 84 static LIST_HEAD(formats); 85 static DEFINE_RWLOCK(binfmt_lock); 86 87 void __register_binfmt(struct linux_binfmt * fmt, int insert) 88 { 89 write_lock(&binfmt_lock); 90 insert ? list_add(&fmt->lh, &formats) : 91 list_add_tail(&fmt->lh, &formats); 92 write_unlock(&binfmt_lock); 93 } 94 95 EXPORT_SYMBOL(__register_binfmt); 96 97 void unregister_binfmt(struct linux_binfmt * fmt) 98 { 99 write_lock(&binfmt_lock); 100 list_del(&fmt->lh); 101 write_unlock(&binfmt_lock); 102 } 103 104 EXPORT_SYMBOL(unregister_binfmt); 105 106 static inline void put_binfmt(struct linux_binfmt * fmt) 107 { 108 module_put(fmt->module); 109 } 110 111 bool path_noexec(const struct path *path) 112 { 113 return (path->mnt->mnt_flags & MNT_NOEXEC) || 114 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC); 115 } 116 117 #ifdef CONFIG_USELIB 118 /* 119 * Note that a shared library must be both readable and executable due to 120 * security reasons. 121 * 122 * Also note that we take the address to load from the file itself. 123 */ 124 SYSCALL_DEFINE1(uselib, const char __user *, library) 125 { 126 struct linux_binfmt *fmt; 127 struct file *file; 128 struct filename *tmp = getname(library); 129 int error = PTR_ERR(tmp); 130 static const struct open_flags uselib_flags = { 131 .open_flag = O_LARGEFILE | O_RDONLY, 132 .acc_mode = MAY_READ | MAY_EXEC, 133 .intent = LOOKUP_OPEN, 134 .lookup_flags = LOOKUP_FOLLOW, 135 }; 136 137 if (IS_ERR(tmp)) 138 goto out; 139 140 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags); 141 putname(tmp); 142 error = PTR_ERR(file); 143 if (IS_ERR(file)) 144 goto out; 145 146 /* 147 * may_open() has already checked for this, so it should be 148 * impossible to trip now. But we need to be extra cautious 149 * and check again at the very end too. 150 */ 151 error = -EACCES; 152 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) || 153 path_noexec(&file->f_path))) 154 goto exit; 155 156 error = -ENOEXEC; 157 158 read_lock(&binfmt_lock); 159 list_for_each_entry(fmt, &formats, lh) { 160 if (!fmt->load_shlib) 161 continue; 162 if (!try_module_get(fmt->module)) 163 continue; 164 read_unlock(&binfmt_lock); 165 error = fmt->load_shlib(file); 166 read_lock(&binfmt_lock); 167 put_binfmt(fmt); 168 if (error != -ENOEXEC) 169 break; 170 } 171 read_unlock(&binfmt_lock); 172 exit: 173 fput(file); 174 out: 175 return error; 176 } 177 #endif /* #ifdef CONFIG_USELIB */ 178 179 #ifdef CONFIG_MMU 180 /* 181 * The nascent bprm->mm is not visible until exec_mmap() but it can 182 * use a lot of memory, account these pages in current->mm temporary 183 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we 184 * change the counter back via acct_arg_size(0). 185 */ 186 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages) 187 { 188 struct mm_struct *mm = current->mm; 189 long diff = (long)(pages - bprm->vma_pages); 190 191 if (!mm || !diff) 192 return; 193 194 bprm->vma_pages = pages; 195 add_mm_counter(mm, MM_ANONPAGES, diff); 196 } 197 198 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, 199 int write) 200 { 201 struct page *page; 202 struct vm_area_struct *vma = bprm->vma; 203 struct mm_struct *mm = bprm->mm; 204 int ret; 205 206 /* 207 * Avoid relying on expanding the stack down in GUP (which 208 * does not work for STACK_GROWSUP anyway), and just do it 209 * by hand ahead of time. 210 */ 211 if (write && pos < vma->vm_start) { 212 mmap_write_lock(mm); 213 ret = expand_downwards(vma, pos); 214 if (unlikely(ret < 0)) { 215 mmap_write_unlock(mm); 216 return NULL; 217 } 218 mmap_write_downgrade(mm); 219 } else 220 mmap_read_lock(mm); 221 222 /* 223 * We are doing an exec(). 'current' is the process 224 * doing the exec and 'mm' is the new process's mm. 225 */ 226 ret = get_user_pages_remote(mm, pos, 1, 227 write ? FOLL_WRITE : 0, 228 &page, NULL); 229 mmap_read_unlock(mm); 230 if (ret <= 0) 231 return NULL; 232 233 if (write) 234 acct_arg_size(bprm, vma_pages(vma)); 235 236 return page; 237 } 238 239 static void put_arg_page(struct page *page) 240 { 241 put_page(page); 242 } 243 244 static void free_arg_pages(struct linux_binprm *bprm) 245 { 246 } 247 248 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, 249 struct page *page) 250 { 251 flush_cache_page(bprm->vma, pos, page_to_pfn(page)); 252 } 253 254 static int __bprm_mm_init(struct linux_binprm *bprm) 255 { 256 int err; 257 struct vm_area_struct *vma = NULL; 258 struct mm_struct *mm = bprm->mm; 259 260 bprm->vma = vma = vm_area_alloc(mm); 261 if (!vma) 262 return -ENOMEM; 263 vma_set_anonymous(vma); 264 265 if (mmap_write_lock_killable(mm)) { 266 err = -EINTR; 267 goto err_free; 268 } 269 270 /* 271 * Place the stack at the largest stack address the architecture 272 * supports. Later, we'll move this to an appropriate place. We don't 273 * use STACK_TOP because that can depend on attributes which aren't 274 * configured yet. 275 */ 276 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP); 277 vma->vm_end = STACK_TOP_MAX; 278 vma->vm_start = vma->vm_end - PAGE_SIZE; 279 vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP); 280 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 281 282 err = insert_vm_struct(mm, vma); 283 if (err) 284 goto err; 285 286 mm->stack_vm = mm->total_vm = 1; 287 mmap_write_unlock(mm); 288 bprm->p = vma->vm_end - sizeof(void *); 289 return 0; 290 err: 291 mmap_write_unlock(mm); 292 err_free: 293 bprm->vma = NULL; 294 vm_area_free(vma); 295 return err; 296 } 297 298 static bool valid_arg_len(struct linux_binprm *bprm, long len) 299 { 300 return len <= MAX_ARG_STRLEN; 301 } 302 303 #else 304 305 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages) 306 { 307 } 308 309 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, 310 int write) 311 { 312 struct page *page; 313 314 page = bprm->page[pos / PAGE_SIZE]; 315 if (!page && write) { 316 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO); 317 if (!page) 318 return NULL; 319 bprm->page[pos / PAGE_SIZE] = page; 320 } 321 322 return page; 323 } 324 325 static void put_arg_page(struct page *page) 326 { 327 } 328 329 static void free_arg_page(struct linux_binprm *bprm, int i) 330 { 331 if (bprm->page[i]) { 332 __free_page(bprm->page[i]); 333 bprm->page[i] = NULL; 334 } 335 } 336 337 static void free_arg_pages(struct linux_binprm *bprm) 338 { 339 int i; 340 341 for (i = 0; i < MAX_ARG_PAGES; i++) 342 free_arg_page(bprm, i); 343 } 344 345 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, 346 struct page *page) 347 { 348 } 349 350 static int __bprm_mm_init(struct linux_binprm *bprm) 351 { 352 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *); 353 return 0; 354 } 355 356 static bool valid_arg_len(struct linux_binprm *bprm, long len) 357 { 358 return len <= bprm->p; 359 } 360 361 #endif /* CONFIG_MMU */ 362 363 /* 364 * Create a new mm_struct and populate it with a temporary stack 365 * vm_area_struct. We don't have enough context at this point to set the stack 366 * flags, permissions, and offset, so we use temporary values. We'll update 367 * them later in setup_arg_pages(). 368 */ 369 static int bprm_mm_init(struct linux_binprm *bprm) 370 { 371 int err; 372 struct mm_struct *mm = NULL; 373 374 bprm->mm = mm = mm_alloc(); 375 err = -ENOMEM; 376 if (!mm) 377 goto err; 378 379 /* Save current stack limit for all calculations made during exec. */ 380 task_lock(current->group_leader); 381 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK]; 382 task_unlock(current->group_leader); 383 384 err = __bprm_mm_init(bprm); 385 if (err) 386 goto err; 387 388 return 0; 389 390 err: 391 if (mm) { 392 bprm->mm = NULL; 393 mmdrop(mm); 394 } 395 396 return err; 397 } 398 399 struct user_arg_ptr { 400 #ifdef CONFIG_COMPAT 401 bool is_compat; 402 #endif 403 union { 404 const char __user *const __user *native; 405 #ifdef CONFIG_COMPAT 406 const compat_uptr_t __user *compat; 407 #endif 408 } ptr; 409 }; 410 411 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr) 412 { 413 const char __user *native; 414 415 #ifdef CONFIG_COMPAT 416 if (unlikely(argv.is_compat)) { 417 compat_uptr_t compat; 418 419 if (get_user(compat, argv.ptr.compat + nr)) 420 return ERR_PTR(-EFAULT); 421 422 return compat_ptr(compat); 423 } 424 #endif 425 426 if (get_user(native, argv.ptr.native + nr)) 427 return ERR_PTR(-EFAULT); 428 429 return native; 430 } 431 432 /* 433 * count() counts the number of strings in array ARGV. 434 */ 435 static int count(struct user_arg_ptr argv, int max) 436 { 437 int i = 0; 438 439 if (argv.ptr.native != NULL) { 440 for (;;) { 441 const char __user *p = get_user_arg_ptr(argv, i); 442 443 if (!p) 444 break; 445 446 if (IS_ERR(p)) 447 return -EFAULT; 448 449 if (i >= max) 450 return -E2BIG; 451 ++i; 452 453 if (fatal_signal_pending(current)) 454 return -ERESTARTNOHAND; 455 cond_resched(); 456 } 457 } 458 return i; 459 } 460 461 static int count_strings_kernel(const char *const *argv) 462 { 463 int i; 464 465 if (!argv) 466 return 0; 467 468 for (i = 0; argv[i]; ++i) { 469 if (i >= MAX_ARG_STRINGS) 470 return -E2BIG; 471 if (fatal_signal_pending(current)) 472 return -ERESTARTNOHAND; 473 cond_resched(); 474 } 475 return i; 476 } 477 478 static int bprm_stack_limits(struct linux_binprm *bprm) 479 { 480 unsigned long limit, ptr_size; 481 482 /* 483 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM 484 * (whichever is smaller) for the argv+env strings. 485 * This ensures that: 486 * - the remaining binfmt code will not run out of stack space, 487 * - the program will have a reasonable amount of stack left 488 * to work from. 489 */ 490 limit = _STK_LIM / 4 * 3; 491 limit = min(limit, bprm->rlim_stack.rlim_cur / 4); 492 /* 493 * We've historically supported up to 32 pages (ARG_MAX) 494 * of argument strings even with small stacks 495 */ 496 limit = max_t(unsigned long, limit, ARG_MAX); 497 /* 498 * We must account for the size of all the argv and envp pointers to 499 * the argv and envp strings, since they will also take up space in 500 * the stack. They aren't stored until much later when we can't 501 * signal to the parent that the child has run out of stack space. 502 * Instead, calculate it here so it's possible to fail gracefully. 503 * 504 * In the case of argc = 0, make sure there is space for adding a 505 * empty string (which will bump argc to 1), to ensure confused 506 * userspace programs don't start processing from argv[1], thinking 507 * argc can never be 0, to keep them from walking envp by accident. 508 * See do_execveat_common(). 509 */ 510 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *); 511 if (limit <= ptr_size) 512 return -E2BIG; 513 limit -= ptr_size; 514 515 bprm->argmin = bprm->p - limit; 516 return 0; 517 } 518 519 /* 520 * 'copy_strings()' copies argument/environment strings from the old 521 * processes's memory to the new process's stack. The call to get_user_pages() 522 * ensures the destination page is created and not swapped out. 523 */ 524 static int copy_strings(int argc, struct user_arg_ptr argv, 525 struct linux_binprm *bprm) 526 { 527 struct page *kmapped_page = NULL; 528 char *kaddr = NULL; 529 unsigned long kpos = 0; 530 int ret; 531 532 while (argc-- > 0) { 533 const char __user *str; 534 int len; 535 unsigned long pos; 536 537 ret = -EFAULT; 538 str = get_user_arg_ptr(argv, argc); 539 if (IS_ERR(str)) 540 goto out; 541 542 len = strnlen_user(str, MAX_ARG_STRLEN); 543 if (!len) 544 goto out; 545 546 ret = -E2BIG; 547 if (!valid_arg_len(bprm, len)) 548 goto out; 549 550 /* We're going to work our way backwards. */ 551 pos = bprm->p; 552 str += len; 553 bprm->p -= len; 554 #ifdef CONFIG_MMU 555 if (bprm->p < bprm->argmin) 556 goto out; 557 #endif 558 559 while (len > 0) { 560 int offset, bytes_to_copy; 561 562 if (fatal_signal_pending(current)) { 563 ret = -ERESTARTNOHAND; 564 goto out; 565 } 566 cond_resched(); 567 568 offset = pos % PAGE_SIZE; 569 if (offset == 0) 570 offset = PAGE_SIZE; 571 572 bytes_to_copy = offset; 573 if (bytes_to_copy > len) 574 bytes_to_copy = len; 575 576 offset -= bytes_to_copy; 577 pos -= bytes_to_copy; 578 str -= bytes_to_copy; 579 len -= bytes_to_copy; 580 581 if (!kmapped_page || kpos != (pos & PAGE_MASK)) { 582 struct page *page; 583 584 page = get_arg_page(bprm, pos, 1); 585 if (!page) { 586 ret = -E2BIG; 587 goto out; 588 } 589 590 if (kmapped_page) { 591 flush_dcache_page(kmapped_page); 592 kunmap_local(kaddr); 593 put_arg_page(kmapped_page); 594 } 595 kmapped_page = page; 596 kaddr = kmap_local_page(kmapped_page); 597 kpos = pos & PAGE_MASK; 598 flush_arg_page(bprm, kpos, kmapped_page); 599 } 600 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) { 601 ret = -EFAULT; 602 goto out; 603 } 604 } 605 } 606 ret = 0; 607 out: 608 if (kmapped_page) { 609 flush_dcache_page(kmapped_page); 610 kunmap_local(kaddr); 611 put_arg_page(kmapped_page); 612 } 613 return ret; 614 } 615 616 /* 617 * Copy and argument/environment string from the kernel to the processes stack. 618 */ 619 int copy_string_kernel(const char *arg, struct linux_binprm *bprm) 620 { 621 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */; 622 unsigned long pos = bprm->p; 623 624 if (len == 0) 625 return -EFAULT; 626 if (!valid_arg_len(bprm, len)) 627 return -E2BIG; 628 629 /* We're going to work our way backwards. */ 630 arg += len; 631 bprm->p -= len; 632 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin) 633 return -E2BIG; 634 635 while (len > 0) { 636 unsigned int bytes_to_copy = min_t(unsigned int, len, 637 min_not_zero(offset_in_page(pos), PAGE_SIZE)); 638 struct page *page; 639 640 pos -= bytes_to_copy; 641 arg -= bytes_to_copy; 642 len -= bytes_to_copy; 643 644 page = get_arg_page(bprm, pos, 1); 645 if (!page) 646 return -E2BIG; 647 flush_arg_page(bprm, pos & PAGE_MASK, page); 648 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy); 649 put_arg_page(page); 650 } 651 652 return 0; 653 } 654 EXPORT_SYMBOL(copy_string_kernel); 655 656 static int copy_strings_kernel(int argc, const char *const *argv, 657 struct linux_binprm *bprm) 658 { 659 while (argc-- > 0) { 660 int ret = copy_string_kernel(argv[argc], bprm); 661 if (ret < 0) 662 return ret; 663 if (fatal_signal_pending(current)) 664 return -ERESTARTNOHAND; 665 cond_resched(); 666 } 667 return 0; 668 } 669 670 #ifdef CONFIG_MMU 671 672 /* 673 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once 674 * the binfmt code determines where the new stack should reside, we shift it to 675 * its final location. The process proceeds as follows: 676 * 677 * 1) Use shift to calculate the new vma endpoints. 678 * 2) Extend vma to cover both the old and new ranges. This ensures the 679 * arguments passed to subsequent functions are consistent. 680 * 3) Move vma's page tables to the new range. 681 * 4) Free up any cleared pgd range. 682 * 5) Shrink the vma to cover only the new range. 683 */ 684 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift) 685 { 686 struct mm_struct *mm = vma->vm_mm; 687 unsigned long old_start = vma->vm_start; 688 unsigned long old_end = vma->vm_end; 689 unsigned long length = old_end - old_start; 690 unsigned long new_start = old_start - shift; 691 unsigned long new_end = old_end - shift; 692 VMA_ITERATOR(vmi, mm, new_start); 693 struct vm_area_struct *next; 694 struct mmu_gather tlb; 695 696 BUG_ON(new_start > new_end); 697 698 /* 699 * ensure there are no vmas between where we want to go 700 * and where we are 701 */ 702 if (vma != vma_next(&vmi)) 703 return -EFAULT; 704 705 vma_iter_prev_range(&vmi); 706 /* 707 * cover the whole range: [new_start, old_end) 708 */ 709 if (vma_expand(&vmi, vma, new_start, old_end, vma->vm_pgoff, NULL)) 710 return -ENOMEM; 711 712 /* 713 * move the page tables downwards, on failure we rely on 714 * process cleanup to remove whatever mess we made. 715 */ 716 if (length != move_page_tables(vma, old_start, 717 vma, new_start, length, false, true)) 718 return -ENOMEM; 719 720 lru_add_drain(); 721 tlb_gather_mmu(&tlb, mm); 722 next = vma_next(&vmi); 723 if (new_end > old_start) { 724 /* 725 * when the old and new regions overlap clear from new_end. 726 */ 727 free_pgd_range(&tlb, new_end, old_end, new_end, 728 next ? next->vm_start : USER_PGTABLES_CEILING); 729 } else { 730 /* 731 * otherwise, clean from old_start; this is done to not touch 732 * the address space in [new_end, old_start) some architectures 733 * have constraints on va-space that make this illegal (IA64) - 734 * for the others its just a little faster. 735 */ 736 free_pgd_range(&tlb, old_start, old_end, new_end, 737 next ? next->vm_start : USER_PGTABLES_CEILING); 738 } 739 tlb_finish_mmu(&tlb); 740 741 vma_prev(&vmi); 742 /* Shrink the vma to just the new range */ 743 return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff); 744 } 745 746 /* 747 * Finalizes the stack vm_area_struct. The flags and permissions are updated, 748 * the stack is optionally relocated, and some extra space is added. 749 */ 750 int setup_arg_pages(struct linux_binprm *bprm, 751 unsigned long stack_top, 752 int executable_stack) 753 { 754 unsigned long ret; 755 unsigned long stack_shift; 756 struct mm_struct *mm = current->mm; 757 struct vm_area_struct *vma = bprm->vma; 758 struct vm_area_struct *prev = NULL; 759 unsigned long vm_flags; 760 unsigned long stack_base; 761 unsigned long stack_size; 762 unsigned long stack_expand; 763 unsigned long rlim_stack; 764 struct mmu_gather tlb; 765 struct vma_iterator vmi; 766 767 #ifdef CONFIG_STACK_GROWSUP 768 /* Limit stack size */ 769 stack_base = bprm->rlim_stack.rlim_max; 770 771 stack_base = calc_max_stack_size(stack_base); 772 773 /* Add space for stack randomization. */ 774 stack_base += (STACK_RND_MASK << PAGE_SHIFT); 775 776 /* Make sure we didn't let the argument array grow too large. */ 777 if (vma->vm_end - vma->vm_start > stack_base) 778 return -ENOMEM; 779 780 stack_base = PAGE_ALIGN(stack_top - stack_base); 781 782 stack_shift = vma->vm_start - stack_base; 783 mm->arg_start = bprm->p - stack_shift; 784 bprm->p = vma->vm_end - stack_shift; 785 #else 786 stack_top = arch_align_stack(stack_top); 787 stack_top = PAGE_ALIGN(stack_top); 788 789 if (unlikely(stack_top < mmap_min_addr) || 790 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr)) 791 return -ENOMEM; 792 793 stack_shift = vma->vm_end - stack_top; 794 795 bprm->p -= stack_shift; 796 mm->arg_start = bprm->p; 797 #endif 798 799 if (bprm->loader) 800 bprm->loader -= stack_shift; 801 bprm->exec -= stack_shift; 802 803 if (mmap_write_lock_killable(mm)) 804 return -EINTR; 805 806 vm_flags = VM_STACK_FLAGS; 807 808 /* 809 * Adjust stack execute permissions; explicitly enable for 810 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone 811 * (arch default) otherwise. 812 */ 813 if (unlikely(executable_stack == EXSTACK_ENABLE_X)) 814 vm_flags |= VM_EXEC; 815 else if (executable_stack == EXSTACK_DISABLE_X) 816 vm_flags &= ~VM_EXEC; 817 vm_flags |= mm->def_flags; 818 vm_flags |= VM_STACK_INCOMPLETE_SETUP; 819 820 vma_iter_init(&vmi, mm, vma->vm_start); 821 822 tlb_gather_mmu(&tlb, mm); 823 ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end, 824 vm_flags); 825 tlb_finish_mmu(&tlb); 826 827 if (ret) 828 goto out_unlock; 829 BUG_ON(prev != vma); 830 831 if (unlikely(vm_flags & VM_EXEC)) { 832 pr_warn_once("process '%pD4' started with executable stack\n", 833 bprm->file); 834 } 835 836 /* Move stack pages down in memory. */ 837 if (stack_shift) { 838 ret = shift_arg_pages(vma, stack_shift); 839 if (ret) 840 goto out_unlock; 841 } 842 843 /* mprotect_fixup is overkill to remove the temporary stack flags */ 844 vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP); 845 846 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */ 847 stack_size = vma->vm_end - vma->vm_start; 848 /* 849 * Align this down to a page boundary as expand_stack 850 * will align it up. 851 */ 852 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK; 853 854 stack_expand = min(rlim_stack, stack_size + stack_expand); 855 856 #ifdef CONFIG_STACK_GROWSUP 857 stack_base = vma->vm_start + stack_expand; 858 #else 859 stack_base = vma->vm_end - stack_expand; 860 #endif 861 current->mm->start_stack = bprm->p; 862 ret = expand_stack_locked(vma, stack_base); 863 if (ret) 864 ret = -EFAULT; 865 866 out_unlock: 867 mmap_write_unlock(mm); 868 return ret; 869 } 870 EXPORT_SYMBOL(setup_arg_pages); 871 872 #else 873 874 /* 875 * Transfer the program arguments and environment from the holding pages 876 * onto the stack. The provided stack pointer is adjusted accordingly. 877 */ 878 int transfer_args_to_stack(struct linux_binprm *bprm, 879 unsigned long *sp_location) 880 { 881 unsigned long index, stop, sp; 882 int ret = 0; 883 884 stop = bprm->p >> PAGE_SHIFT; 885 sp = *sp_location; 886 887 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) { 888 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0; 889 char *src = kmap_local_page(bprm->page[index]) + offset; 890 sp -= PAGE_SIZE - offset; 891 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0) 892 ret = -EFAULT; 893 kunmap_local(src); 894 if (ret) 895 goto out; 896 } 897 898 *sp_location = sp; 899 900 out: 901 return ret; 902 } 903 EXPORT_SYMBOL(transfer_args_to_stack); 904 905 #endif /* CONFIG_MMU */ 906 907 /* 908 * On success, caller must call do_close_execat() on the returned 909 * struct file to close it. 910 */ 911 static struct file *do_open_execat(int fd, struct filename *name, int flags) 912 { 913 struct file *file; 914 int err; 915 struct open_flags open_exec_flags = { 916 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC, 917 .acc_mode = MAY_EXEC, 918 .intent = LOOKUP_OPEN, 919 .lookup_flags = LOOKUP_FOLLOW, 920 }; 921 922 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0) 923 return ERR_PTR(-EINVAL); 924 if (flags & AT_SYMLINK_NOFOLLOW) 925 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW; 926 if (flags & AT_EMPTY_PATH) 927 open_exec_flags.lookup_flags |= LOOKUP_EMPTY; 928 929 file = do_filp_open(fd, name, &open_exec_flags); 930 if (IS_ERR(file)) 931 goto out; 932 933 /* 934 * may_open() has already checked for this, so it should be 935 * impossible to trip now. But we need to be extra cautious 936 * and check again at the very end too. 937 */ 938 err = -EACCES; 939 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) || 940 path_noexec(&file->f_path))) 941 goto exit; 942 943 err = deny_write_access(file); 944 if (err) 945 goto exit; 946 947 out: 948 return file; 949 950 exit: 951 fput(file); 952 return ERR_PTR(err); 953 } 954 955 /** 956 * open_exec - Open a path name for execution 957 * 958 * @name: path name to open with the intent of executing it. 959 * 960 * Returns ERR_PTR on failure or allocated struct file on success. 961 * 962 * As this is a wrapper for the internal do_open_execat(), callers 963 * must call allow_write_access() before fput() on release. Also see 964 * do_close_execat(). 965 */ 966 struct file *open_exec(const char *name) 967 { 968 struct filename *filename = getname_kernel(name); 969 struct file *f = ERR_CAST(filename); 970 971 if (!IS_ERR(filename)) { 972 f = do_open_execat(AT_FDCWD, filename, 0); 973 putname(filename); 974 } 975 return f; 976 } 977 EXPORT_SYMBOL(open_exec); 978 979 #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC) 980 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len) 981 { 982 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos); 983 if (res > 0) 984 flush_icache_user_range(addr, addr + len); 985 return res; 986 } 987 EXPORT_SYMBOL(read_code); 988 #endif 989 990 /* 991 * Maps the mm_struct mm into the current task struct. 992 * On success, this function returns with exec_update_lock 993 * held for writing. 994 */ 995 static int exec_mmap(struct mm_struct *mm) 996 { 997 struct task_struct *tsk; 998 struct mm_struct *old_mm, *active_mm; 999 int ret; 1000 1001 /* Notify parent that we're no longer interested in the old VM */ 1002 tsk = current; 1003 old_mm = current->mm; 1004 exec_mm_release(tsk, old_mm); 1005 1006 ret = down_write_killable(&tsk->signal->exec_update_lock); 1007 if (ret) 1008 return ret; 1009 1010 if (old_mm) { 1011 /* 1012 * If there is a pending fatal signal perhaps a signal 1013 * whose default action is to create a coredump get 1014 * out and die instead of going through with the exec. 1015 */ 1016 ret = mmap_read_lock_killable(old_mm); 1017 if (ret) { 1018 up_write(&tsk->signal->exec_update_lock); 1019 return ret; 1020 } 1021 } 1022 1023 task_lock(tsk); 1024 membarrier_exec_mmap(mm); 1025 1026 local_irq_disable(); 1027 active_mm = tsk->active_mm; 1028 tsk->active_mm = mm; 1029 tsk->mm = mm; 1030 mm_init_cid(mm); 1031 /* 1032 * This prevents preemption while active_mm is being loaded and 1033 * it and mm are being updated, which could cause problems for 1034 * lazy tlb mm refcounting when these are updated by context 1035 * switches. Not all architectures can handle irqs off over 1036 * activate_mm yet. 1037 */ 1038 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM)) 1039 local_irq_enable(); 1040 activate_mm(active_mm, mm); 1041 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM)) 1042 local_irq_enable(); 1043 lru_gen_add_mm(mm); 1044 task_unlock(tsk); 1045 lru_gen_use_mm(mm); 1046 if (old_mm) { 1047 mmap_read_unlock(old_mm); 1048 BUG_ON(active_mm != old_mm); 1049 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm); 1050 mm_update_next_owner(old_mm); 1051 mmput(old_mm); 1052 return 0; 1053 } 1054 mmdrop_lazy_tlb(active_mm); 1055 return 0; 1056 } 1057 1058 static int de_thread(struct task_struct *tsk) 1059 { 1060 struct signal_struct *sig = tsk->signal; 1061 struct sighand_struct *oldsighand = tsk->sighand; 1062 spinlock_t *lock = &oldsighand->siglock; 1063 1064 if (thread_group_empty(tsk)) 1065 goto no_thread_group; 1066 1067 /* 1068 * Kill all other threads in the thread group. 1069 */ 1070 spin_lock_irq(lock); 1071 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) { 1072 /* 1073 * Another group action in progress, just 1074 * return so that the signal is processed. 1075 */ 1076 spin_unlock_irq(lock); 1077 return -EAGAIN; 1078 } 1079 1080 sig->group_exec_task = tsk; 1081 sig->notify_count = zap_other_threads(tsk); 1082 if (!thread_group_leader(tsk)) 1083 sig->notify_count--; 1084 1085 while (sig->notify_count) { 1086 __set_current_state(TASK_KILLABLE); 1087 spin_unlock_irq(lock); 1088 schedule(); 1089 if (__fatal_signal_pending(tsk)) 1090 goto killed; 1091 spin_lock_irq(lock); 1092 } 1093 spin_unlock_irq(lock); 1094 1095 /* 1096 * At this point all other threads have exited, all we have to 1097 * do is to wait for the thread group leader to become inactive, 1098 * and to assume its PID: 1099 */ 1100 if (!thread_group_leader(tsk)) { 1101 struct task_struct *leader = tsk->group_leader; 1102 1103 for (;;) { 1104 cgroup_threadgroup_change_begin(tsk); 1105 write_lock_irq(&tasklist_lock); 1106 /* 1107 * Do this under tasklist_lock to ensure that 1108 * exit_notify() can't miss ->group_exec_task 1109 */ 1110 sig->notify_count = -1; 1111 if (likely(leader->exit_state)) 1112 break; 1113 __set_current_state(TASK_KILLABLE); 1114 write_unlock_irq(&tasklist_lock); 1115 cgroup_threadgroup_change_end(tsk); 1116 schedule(); 1117 if (__fatal_signal_pending(tsk)) 1118 goto killed; 1119 } 1120 1121 /* 1122 * The only record we have of the real-time age of a 1123 * process, regardless of execs it's done, is start_time. 1124 * All the past CPU time is accumulated in signal_struct 1125 * from sister threads now dead. But in this non-leader 1126 * exec, nothing survives from the original leader thread, 1127 * whose birth marks the true age of this process now. 1128 * When we take on its identity by switching to its PID, we 1129 * also take its birthdate (always earlier than our own). 1130 */ 1131 tsk->start_time = leader->start_time; 1132 tsk->start_boottime = leader->start_boottime; 1133 1134 BUG_ON(!same_thread_group(leader, tsk)); 1135 /* 1136 * An exec() starts a new thread group with the 1137 * TGID of the previous thread group. Rehash the 1138 * two threads with a switched PID, and release 1139 * the former thread group leader: 1140 */ 1141 1142 /* Become a process group leader with the old leader's pid. 1143 * The old leader becomes a thread of the this thread group. 1144 */ 1145 exchange_tids(tsk, leader); 1146 transfer_pid(leader, tsk, PIDTYPE_TGID); 1147 transfer_pid(leader, tsk, PIDTYPE_PGID); 1148 transfer_pid(leader, tsk, PIDTYPE_SID); 1149 1150 list_replace_rcu(&leader->tasks, &tsk->tasks); 1151 list_replace_init(&leader->sibling, &tsk->sibling); 1152 1153 tsk->group_leader = tsk; 1154 leader->group_leader = tsk; 1155 1156 tsk->exit_signal = SIGCHLD; 1157 leader->exit_signal = -1; 1158 1159 BUG_ON(leader->exit_state != EXIT_ZOMBIE); 1160 leader->exit_state = EXIT_DEAD; 1161 1162 /* 1163 * We are going to release_task()->ptrace_unlink() silently, 1164 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees 1165 * the tracer won't block again waiting for this thread. 1166 */ 1167 if (unlikely(leader->ptrace)) 1168 __wake_up_parent(leader, leader->parent); 1169 write_unlock_irq(&tasklist_lock); 1170 cgroup_threadgroup_change_end(tsk); 1171 1172 release_task(leader); 1173 } 1174 1175 sig->group_exec_task = NULL; 1176 sig->notify_count = 0; 1177 1178 no_thread_group: 1179 /* we have changed execution domain */ 1180 tsk->exit_signal = SIGCHLD; 1181 1182 BUG_ON(!thread_group_leader(tsk)); 1183 return 0; 1184 1185 killed: 1186 /* protects against exit_notify() and __exit_signal() */ 1187 read_lock(&tasklist_lock); 1188 sig->group_exec_task = NULL; 1189 sig->notify_count = 0; 1190 read_unlock(&tasklist_lock); 1191 return -EAGAIN; 1192 } 1193 1194 1195 /* 1196 * This function makes sure the current process has its own signal table, 1197 * so that flush_signal_handlers can later reset the handlers without 1198 * disturbing other processes. (Other processes might share the signal 1199 * table via the CLONE_SIGHAND option to clone().) 1200 */ 1201 static int unshare_sighand(struct task_struct *me) 1202 { 1203 struct sighand_struct *oldsighand = me->sighand; 1204 1205 if (refcount_read(&oldsighand->count) != 1) { 1206 struct sighand_struct *newsighand; 1207 /* 1208 * This ->sighand is shared with the CLONE_SIGHAND 1209 * but not CLONE_THREAD task, switch to the new one. 1210 */ 1211 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1212 if (!newsighand) 1213 return -ENOMEM; 1214 1215 refcount_set(&newsighand->count, 1); 1216 1217 write_lock_irq(&tasklist_lock); 1218 spin_lock(&oldsighand->siglock); 1219 memcpy(newsighand->action, oldsighand->action, 1220 sizeof(newsighand->action)); 1221 rcu_assign_pointer(me->sighand, newsighand); 1222 spin_unlock(&oldsighand->siglock); 1223 write_unlock_irq(&tasklist_lock); 1224 1225 __cleanup_sighand(oldsighand); 1226 } 1227 return 0; 1228 } 1229 1230 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk) 1231 { 1232 task_lock(tsk); 1233 /* Always NUL terminated and zero-padded */ 1234 strscpy_pad(buf, tsk->comm, buf_size); 1235 task_unlock(tsk); 1236 return buf; 1237 } 1238 EXPORT_SYMBOL_GPL(__get_task_comm); 1239 1240 /* 1241 * These functions flushes out all traces of the currently running executable 1242 * so that a new one can be started 1243 */ 1244 1245 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec) 1246 { 1247 task_lock(tsk); 1248 trace_task_rename(tsk, buf); 1249 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm)); 1250 task_unlock(tsk); 1251 perf_event_comm(tsk, exec); 1252 } 1253 1254 /* 1255 * Calling this is the point of no return. None of the failures will be 1256 * seen by userspace since either the process is already taking a fatal 1257 * signal (via de_thread() or coredump), or will have SEGV raised 1258 * (after exec_mmap()) by search_binary_handler (see below). 1259 */ 1260 int begin_new_exec(struct linux_binprm * bprm) 1261 { 1262 struct task_struct *me = current; 1263 int retval; 1264 1265 /* Once we are committed compute the creds */ 1266 retval = bprm_creds_from_file(bprm); 1267 if (retval) 1268 return retval; 1269 1270 /* 1271 * Ensure all future errors are fatal. 1272 */ 1273 bprm->point_of_no_return = true; 1274 1275 /* 1276 * Make this the only thread in the thread group. 1277 */ 1278 retval = de_thread(me); 1279 if (retval) 1280 goto out; 1281 1282 /* 1283 * Cancel any io_uring activity across execve 1284 */ 1285 io_uring_task_cancel(); 1286 1287 /* Ensure the files table is not shared. */ 1288 retval = unshare_files(); 1289 if (retval) 1290 goto out; 1291 1292 /* 1293 * Must be called _before_ exec_mmap() as bprm->mm is 1294 * not visible until then. Doing it here also ensures 1295 * we don't race against replace_mm_exe_file(). 1296 */ 1297 retval = set_mm_exe_file(bprm->mm, bprm->file); 1298 if (retval) 1299 goto out; 1300 1301 /* If the binary is not readable then enforce mm->dumpable=0 */ 1302 would_dump(bprm, bprm->file); 1303 if (bprm->have_execfd) 1304 would_dump(bprm, bprm->executable); 1305 1306 /* 1307 * Release all of the old mmap stuff 1308 */ 1309 acct_arg_size(bprm, 0); 1310 retval = exec_mmap(bprm->mm); 1311 if (retval) 1312 goto out; 1313 1314 bprm->mm = NULL; 1315 1316 retval = exec_task_namespaces(); 1317 if (retval) 1318 goto out_unlock; 1319 1320 #ifdef CONFIG_POSIX_TIMERS 1321 spin_lock_irq(&me->sighand->siglock); 1322 posix_cpu_timers_exit(me); 1323 spin_unlock_irq(&me->sighand->siglock); 1324 exit_itimers(me); 1325 flush_itimer_signals(); 1326 #endif 1327 1328 /* 1329 * Make the signal table private. 1330 */ 1331 retval = unshare_sighand(me); 1332 if (retval) 1333 goto out_unlock; 1334 1335 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | 1336 PF_NOFREEZE | PF_NO_SETAFFINITY); 1337 flush_thread(); 1338 me->personality &= ~bprm->per_clear; 1339 1340 clear_syscall_work_syscall_user_dispatch(me); 1341 1342 /* 1343 * We have to apply CLOEXEC before we change whether the process is 1344 * dumpable (in setup_new_exec) to avoid a race with a process in userspace 1345 * trying to access the should-be-closed file descriptors of a process 1346 * undergoing exec(2). 1347 */ 1348 do_close_on_exec(me->files); 1349 1350 if (bprm->secureexec) { 1351 /* Make sure parent cannot signal privileged process. */ 1352 me->pdeath_signal = 0; 1353 1354 /* 1355 * For secureexec, reset the stack limit to sane default to 1356 * avoid bad behavior from the prior rlimits. This has to 1357 * happen before arch_pick_mmap_layout(), which examines 1358 * RLIMIT_STACK, but after the point of no return to avoid 1359 * needing to clean up the change on failure. 1360 */ 1361 if (bprm->rlim_stack.rlim_cur > _STK_LIM) 1362 bprm->rlim_stack.rlim_cur = _STK_LIM; 1363 } 1364 1365 me->sas_ss_sp = me->sas_ss_size = 0; 1366 1367 /* 1368 * Figure out dumpability. Note that this checking only of current 1369 * is wrong, but userspace depends on it. This should be testing 1370 * bprm->secureexec instead. 1371 */ 1372 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP || 1373 !(uid_eq(current_euid(), current_uid()) && 1374 gid_eq(current_egid(), current_gid()))) 1375 set_dumpable(current->mm, suid_dumpable); 1376 else 1377 set_dumpable(current->mm, SUID_DUMP_USER); 1378 1379 perf_event_exec(); 1380 __set_task_comm(me, kbasename(bprm->filename), true); 1381 1382 /* An exec changes our domain. We are no longer part of the thread 1383 group */ 1384 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1); 1385 flush_signal_handlers(me, 0); 1386 1387 retval = set_cred_ucounts(bprm->cred); 1388 if (retval < 0) 1389 goto out_unlock; 1390 1391 /* 1392 * install the new credentials for this executable 1393 */ 1394 security_bprm_committing_creds(bprm); 1395 1396 commit_creds(bprm->cred); 1397 bprm->cred = NULL; 1398 1399 /* 1400 * Disable monitoring for regular users 1401 * when executing setuid binaries. Must 1402 * wait until new credentials are committed 1403 * by commit_creds() above 1404 */ 1405 if (get_dumpable(me->mm) != SUID_DUMP_USER) 1406 perf_event_exit_task(me); 1407 /* 1408 * cred_guard_mutex must be held at least to this point to prevent 1409 * ptrace_attach() from altering our determination of the task's 1410 * credentials; any time after this it may be unlocked. 1411 */ 1412 security_bprm_committed_creds(bprm); 1413 1414 /* Pass the opened binary to the interpreter. */ 1415 if (bprm->have_execfd) { 1416 retval = get_unused_fd_flags(0); 1417 if (retval < 0) 1418 goto out_unlock; 1419 fd_install(retval, bprm->executable); 1420 bprm->executable = NULL; 1421 bprm->execfd = retval; 1422 } 1423 return 0; 1424 1425 out_unlock: 1426 up_write(&me->signal->exec_update_lock); 1427 if (!bprm->cred) 1428 mutex_unlock(&me->signal->cred_guard_mutex); 1429 1430 out: 1431 return retval; 1432 } 1433 EXPORT_SYMBOL(begin_new_exec); 1434 1435 void would_dump(struct linux_binprm *bprm, struct file *file) 1436 { 1437 struct inode *inode = file_inode(file); 1438 struct mnt_idmap *idmap = file_mnt_idmap(file); 1439 if (inode_permission(idmap, inode, MAY_READ) < 0) { 1440 struct user_namespace *old, *user_ns; 1441 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP; 1442 1443 /* Ensure mm->user_ns contains the executable */ 1444 user_ns = old = bprm->mm->user_ns; 1445 while ((user_ns != &init_user_ns) && 1446 !privileged_wrt_inode_uidgid(user_ns, idmap, inode)) 1447 user_ns = user_ns->parent; 1448 1449 if (old != user_ns) { 1450 bprm->mm->user_ns = get_user_ns(user_ns); 1451 put_user_ns(old); 1452 } 1453 } 1454 } 1455 EXPORT_SYMBOL(would_dump); 1456 1457 void setup_new_exec(struct linux_binprm * bprm) 1458 { 1459 /* Setup things that can depend upon the personality */ 1460 struct task_struct *me = current; 1461 1462 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack); 1463 1464 arch_setup_new_exec(); 1465 1466 /* Set the new mm task size. We have to do that late because it may 1467 * depend on TIF_32BIT which is only updated in flush_thread() on 1468 * some architectures like powerpc 1469 */ 1470 me->mm->task_size = TASK_SIZE; 1471 up_write(&me->signal->exec_update_lock); 1472 mutex_unlock(&me->signal->cred_guard_mutex); 1473 } 1474 EXPORT_SYMBOL(setup_new_exec); 1475 1476 /* Runs immediately before start_thread() takes over. */ 1477 void finalize_exec(struct linux_binprm *bprm) 1478 { 1479 /* Store any stack rlimit changes before starting thread. */ 1480 task_lock(current->group_leader); 1481 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack; 1482 task_unlock(current->group_leader); 1483 } 1484 EXPORT_SYMBOL(finalize_exec); 1485 1486 /* 1487 * Prepare credentials and lock ->cred_guard_mutex. 1488 * setup_new_exec() commits the new creds and drops the lock. 1489 * Or, if exec fails before, free_bprm() should release ->cred 1490 * and unlock. 1491 */ 1492 static int prepare_bprm_creds(struct linux_binprm *bprm) 1493 { 1494 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex)) 1495 return -ERESTARTNOINTR; 1496 1497 bprm->cred = prepare_exec_creds(); 1498 if (likely(bprm->cred)) 1499 return 0; 1500 1501 mutex_unlock(¤t->signal->cred_guard_mutex); 1502 return -ENOMEM; 1503 } 1504 1505 /* Matches do_open_execat() */ 1506 static void do_close_execat(struct file *file) 1507 { 1508 if (!file) 1509 return; 1510 allow_write_access(file); 1511 fput(file); 1512 } 1513 1514 static void free_bprm(struct linux_binprm *bprm) 1515 { 1516 if (bprm->mm) { 1517 acct_arg_size(bprm, 0); 1518 mmput(bprm->mm); 1519 } 1520 free_arg_pages(bprm); 1521 if (bprm->cred) { 1522 mutex_unlock(¤t->signal->cred_guard_mutex); 1523 abort_creds(bprm->cred); 1524 } 1525 do_close_execat(bprm->file); 1526 if (bprm->executable) 1527 fput(bprm->executable); 1528 /* If a binfmt changed the interp, free it. */ 1529 if (bprm->interp != bprm->filename) 1530 kfree(bprm->interp); 1531 kfree(bprm->fdpath); 1532 kfree(bprm); 1533 } 1534 1535 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename, int flags) 1536 { 1537 struct linux_binprm *bprm; 1538 struct file *file; 1539 int retval = -ENOMEM; 1540 1541 file = do_open_execat(fd, filename, flags); 1542 if (IS_ERR(file)) 1543 return ERR_CAST(file); 1544 1545 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL); 1546 if (!bprm) { 1547 do_close_execat(file); 1548 return ERR_PTR(-ENOMEM); 1549 } 1550 1551 bprm->file = file; 1552 1553 if (fd == AT_FDCWD || filename->name[0] == '/') { 1554 bprm->filename = filename->name; 1555 } else { 1556 if (filename->name[0] == '\0') 1557 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd); 1558 else 1559 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s", 1560 fd, filename->name); 1561 if (!bprm->fdpath) 1562 goto out_free; 1563 1564 /* 1565 * Record that a name derived from an O_CLOEXEC fd will be 1566 * inaccessible after exec. This allows the code in exec to 1567 * choose to fail when the executable is not mmaped into the 1568 * interpreter and an open file descriptor is not passed to 1569 * the interpreter. This makes for a better user experience 1570 * than having the interpreter start and then immediately fail 1571 * when it finds the executable is inaccessible. 1572 */ 1573 if (get_close_on_exec(fd)) 1574 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE; 1575 1576 bprm->filename = bprm->fdpath; 1577 } 1578 bprm->interp = bprm->filename; 1579 1580 retval = bprm_mm_init(bprm); 1581 if (!retval) 1582 return bprm; 1583 1584 out_free: 1585 free_bprm(bprm); 1586 return ERR_PTR(retval); 1587 } 1588 1589 int bprm_change_interp(const char *interp, struct linux_binprm *bprm) 1590 { 1591 /* If a binfmt changed the interp, free it first. */ 1592 if (bprm->interp != bprm->filename) 1593 kfree(bprm->interp); 1594 bprm->interp = kstrdup(interp, GFP_KERNEL); 1595 if (!bprm->interp) 1596 return -ENOMEM; 1597 return 0; 1598 } 1599 EXPORT_SYMBOL(bprm_change_interp); 1600 1601 /* 1602 * determine how safe it is to execute the proposed program 1603 * - the caller must hold ->cred_guard_mutex to protect against 1604 * PTRACE_ATTACH or seccomp thread-sync 1605 */ 1606 static void check_unsafe_exec(struct linux_binprm *bprm) 1607 { 1608 struct task_struct *p = current, *t; 1609 unsigned n_fs; 1610 1611 if (p->ptrace) 1612 bprm->unsafe |= LSM_UNSAFE_PTRACE; 1613 1614 /* 1615 * This isn't strictly necessary, but it makes it harder for LSMs to 1616 * mess up. 1617 */ 1618 if (task_no_new_privs(current)) 1619 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS; 1620 1621 /* 1622 * If another task is sharing our fs, we cannot safely 1623 * suid exec because the differently privileged task 1624 * will be able to manipulate the current directory, etc. 1625 * It would be nice to force an unshare instead... 1626 */ 1627 n_fs = 1; 1628 spin_lock(&p->fs->lock); 1629 rcu_read_lock(); 1630 for_other_threads(p, t) { 1631 if (t->fs == p->fs) 1632 n_fs++; 1633 } 1634 rcu_read_unlock(); 1635 1636 /* "users" and "in_exec" locked for copy_fs() */ 1637 if (p->fs->users > n_fs) 1638 bprm->unsafe |= LSM_UNSAFE_SHARE; 1639 else 1640 p->fs->in_exec = 1; 1641 spin_unlock(&p->fs->lock); 1642 } 1643 1644 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file) 1645 { 1646 /* Handle suid and sgid on files */ 1647 struct mnt_idmap *idmap; 1648 struct inode *inode = file_inode(file); 1649 unsigned int mode; 1650 vfsuid_t vfsuid; 1651 vfsgid_t vfsgid; 1652 1653 if (!mnt_may_suid(file->f_path.mnt)) 1654 return; 1655 1656 if (task_no_new_privs(current)) 1657 return; 1658 1659 mode = READ_ONCE(inode->i_mode); 1660 if (!(mode & (S_ISUID|S_ISGID))) 1661 return; 1662 1663 idmap = file_mnt_idmap(file); 1664 1665 /* Be careful if suid/sgid is set */ 1666 inode_lock(inode); 1667 1668 /* reload atomically mode/uid/gid now that lock held */ 1669 mode = inode->i_mode; 1670 vfsuid = i_uid_into_vfsuid(idmap, inode); 1671 vfsgid = i_gid_into_vfsgid(idmap, inode); 1672 inode_unlock(inode); 1673 1674 /* We ignore suid/sgid if there are no mappings for them in the ns */ 1675 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) || 1676 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid)) 1677 return; 1678 1679 if (mode & S_ISUID) { 1680 bprm->per_clear |= PER_CLEAR_ON_SETID; 1681 bprm->cred->euid = vfsuid_into_kuid(vfsuid); 1682 } 1683 1684 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) { 1685 bprm->per_clear |= PER_CLEAR_ON_SETID; 1686 bprm->cred->egid = vfsgid_into_kgid(vfsgid); 1687 } 1688 } 1689 1690 /* 1691 * Compute brpm->cred based upon the final binary. 1692 */ 1693 static int bprm_creds_from_file(struct linux_binprm *bprm) 1694 { 1695 /* Compute creds based on which file? */ 1696 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file; 1697 1698 bprm_fill_uid(bprm, file); 1699 return security_bprm_creds_from_file(bprm, file); 1700 } 1701 1702 /* 1703 * Fill the binprm structure from the inode. 1704 * Read the first BINPRM_BUF_SIZE bytes 1705 * 1706 * This may be called multiple times for binary chains (scripts for example). 1707 */ 1708 static int prepare_binprm(struct linux_binprm *bprm) 1709 { 1710 loff_t pos = 0; 1711 1712 memset(bprm->buf, 0, BINPRM_BUF_SIZE); 1713 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos); 1714 } 1715 1716 /* 1717 * Arguments are '\0' separated strings found at the location bprm->p 1718 * points to; chop off the first by relocating brpm->p to right after 1719 * the first '\0' encountered. 1720 */ 1721 int remove_arg_zero(struct linux_binprm *bprm) 1722 { 1723 int ret = 0; 1724 unsigned long offset; 1725 char *kaddr; 1726 struct page *page; 1727 1728 if (!bprm->argc) 1729 return 0; 1730 1731 do { 1732 offset = bprm->p & ~PAGE_MASK; 1733 page = get_arg_page(bprm, bprm->p, 0); 1734 if (!page) { 1735 ret = -EFAULT; 1736 goto out; 1737 } 1738 kaddr = kmap_local_page(page); 1739 1740 for (; offset < PAGE_SIZE && kaddr[offset]; 1741 offset++, bprm->p++) 1742 ; 1743 1744 kunmap_local(kaddr); 1745 put_arg_page(page); 1746 } while (offset == PAGE_SIZE); 1747 1748 bprm->p++; 1749 bprm->argc--; 1750 ret = 0; 1751 1752 out: 1753 return ret; 1754 } 1755 EXPORT_SYMBOL(remove_arg_zero); 1756 1757 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e)) 1758 /* 1759 * cycle the list of binary formats handler, until one recognizes the image 1760 */ 1761 static int search_binary_handler(struct linux_binprm *bprm) 1762 { 1763 bool need_retry = IS_ENABLED(CONFIG_MODULES); 1764 struct linux_binfmt *fmt; 1765 int retval; 1766 1767 retval = prepare_binprm(bprm); 1768 if (retval < 0) 1769 return retval; 1770 1771 retval = security_bprm_check(bprm); 1772 if (retval) 1773 return retval; 1774 1775 retval = -ENOENT; 1776 retry: 1777 read_lock(&binfmt_lock); 1778 list_for_each_entry(fmt, &formats, lh) { 1779 if (!try_module_get(fmt->module)) 1780 continue; 1781 read_unlock(&binfmt_lock); 1782 1783 retval = fmt->load_binary(bprm); 1784 1785 read_lock(&binfmt_lock); 1786 put_binfmt(fmt); 1787 if (bprm->point_of_no_return || (retval != -ENOEXEC)) { 1788 read_unlock(&binfmt_lock); 1789 return retval; 1790 } 1791 } 1792 read_unlock(&binfmt_lock); 1793 1794 if (need_retry) { 1795 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) && 1796 printable(bprm->buf[2]) && printable(bprm->buf[3])) 1797 return retval; 1798 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0) 1799 return retval; 1800 need_retry = false; 1801 goto retry; 1802 } 1803 1804 return retval; 1805 } 1806 1807 /* binfmt handlers will call back into begin_new_exec() on success. */ 1808 static int exec_binprm(struct linux_binprm *bprm) 1809 { 1810 pid_t old_pid, old_vpid; 1811 int ret, depth; 1812 1813 /* Need to fetch pid before load_binary changes it */ 1814 old_pid = current->pid; 1815 rcu_read_lock(); 1816 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent)); 1817 rcu_read_unlock(); 1818 1819 /* This allows 4 levels of binfmt rewrites before failing hard. */ 1820 for (depth = 0;; depth++) { 1821 struct file *exec; 1822 if (depth > 5) 1823 return -ELOOP; 1824 1825 ret = search_binary_handler(bprm); 1826 if (ret < 0) 1827 return ret; 1828 if (!bprm->interpreter) 1829 break; 1830 1831 exec = bprm->file; 1832 bprm->file = bprm->interpreter; 1833 bprm->interpreter = NULL; 1834 1835 allow_write_access(exec); 1836 if (unlikely(bprm->have_execfd)) { 1837 if (bprm->executable) { 1838 fput(exec); 1839 return -ENOEXEC; 1840 } 1841 bprm->executable = exec; 1842 } else 1843 fput(exec); 1844 } 1845 1846 audit_bprm(bprm); 1847 trace_sched_process_exec(current, old_pid, bprm); 1848 ptrace_event(PTRACE_EVENT_EXEC, old_vpid); 1849 proc_exec_connector(current); 1850 return 0; 1851 } 1852 1853 static int bprm_execve(struct linux_binprm *bprm) 1854 { 1855 int retval; 1856 1857 retval = prepare_bprm_creds(bprm); 1858 if (retval) 1859 return retval; 1860 1861 /* 1862 * Check for unsafe execution states before exec_binprm(), which 1863 * will call back into begin_new_exec(), into bprm_creds_from_file(), 1864 * where setuid-ness is evaluated. 1865 */ 1866 check_unsafe_exec(bprm); 1867 current->in_execve = 1; 1868 sched_mm_cid_before_execve(current); 1869 1870 sched_exec(); 1871 1872 /* Set the unchanging part of bprm->cred */ 1873 retval = security_bprm_creds_for_exec(bprm); 1874 if (retval) 1875 goto out; 1876 1877 retval = exec_binprm(bprm); 1878 if (retval < 0) 1879 goto out; 1880 1881 sched_mm_cid_after_execve(current); 1882 /* execve succeeded */ 1883 current->fs->in_exec = 0; 1884 current->in_execve = 0; 1885 rseq_execve(current); 1886 user_events_execve(current); 1887 acct_update_integrals(current); 1888 task_numa_free(current, false); 1889 return retval; 1890 1891 out: 1892 /* 1893 * If past the point of no return ensure the code never 1894 * returns to the userspace process. Use an existing fatal 1895 * signal if present otherwise terminate the process with 1896 * SIGSEGV. 1897 */ 1898 if (bprm->point_of_no_return && !fatal_signal_pending(current)) 1899 force_fatal_sig(SIGSEGV); 1900 1901 sched_mm_cid_after_execve(current); 1902 current->fs->in_exec = 0; 1903 current->in_execve = 0; 1904 1905 return retval; 1906 } 1907 1908 static int do_execveat_common(int fd, struct filename *filename, 1909 struct user_arg_ptr argv, 1910 struct user_arg_ptr envp, 1911 int flags) 1912 { 1913 struct linux_binprm *bprm; 1914 int retval; 1915 1916 if (IS_ERR(filename)) 1917 return PTR_ERR(filename); 1918 1919 /* 1920 * We move the actual failure in case of RLIMIT_NPROC excess from 1921 * set*uid() to execve() because too many poorly written programs 1922 * don't check setuid() return code. Here we additionally recheck 1923 * whether NPROC limit is still exceeded. 1924 */ 1925 if ((current->flags & PF_NPROC_EXCEEDED) && 1926 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) { 1927 retval = -EAGAIN; 1928 goto out_ret; 1929 } 1930 1931 /* We're below the limit (still or again), so we don't want to make 1932 * further execve() calls fail. */ 1933 current->flags &= ~PF_NPROC_EXCEEDED; 1934 1935 bprm = alloc_bprm(fd, filename, flags); 1936 if (IS_ERR(bprm)) { 1937 retval = PTR_ERR(bprm); 1938 goto out_ret; 1939 } 1940 1941 retval = count(argv, MAX_ARG_STRINGS); 1942 if (retval == 0) 1943 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n", 1944 current->comm, bprm->filename); 1945 if (retval < 0) 1946 goto out_free; 1947 bprm->argc = retval; 1948 1949 retval = count(envp, MAX_ARG_STRINGS); 1950 if (retval < 0) 1951 goto out_free; 1952 bprm->envc = retval; 1953 1954 retval = bprm_stack_limits(bprm); 1955 if (retval < 0) 1956 goto out_free; 1957 1958 retval = copy_string_kernel(bprm->filename, bprm); 1959 if (retval < 0) 1960 goto out_free; 1961 bprm->exec = bprm->p; 1962 1963 retval = copy_strings(bprm->envc, envp, bprm); 1964 if (retval < 0) 1965 goto out_free; 1966 1967 retval = copy_strings(bprm->argc, argv, bprm); 1968 if (retval < 0) 1969 goto out_free; 1970 1971 /* 1972 * When argv is empty, add an empty string ("") as argv[0] to 1973 * ensure confused userspace programs that start processing 1974 * from argv[1] won't end up walking envp. See also 1975 * bprm_stack_limits(). 1976 */ 1977 if (bprm->argc == 0) { 1978 retval = copy_string_kernel("", bprm); 1979 if (retval < 0) 1980 goto out_free; 1981 bprm->argc = 1; 1982 } 1983 1984 retval = bprm_execve(bprm); 1985 out_free: 1986 free_bprm(bprm); 1987 1988 out_ret: 1989 putname(filename); 1990 return retval; 1991 } 1992 1993 int kernel_execve(const char *kernel_filename, 1994 const char *const *argv, const char *const *envp) 1995 { 1996 struct filename *filename; 1997 struct linux_binprm *bprm; 1998 int fd = AT_FDCWD; 1999 int retval; 2000 2001 /* It is non-sense for kernel threads to call execve */ 2002 if (WARN_ON_ONCE(current->flags & PF_KTHREAD)) 2003 return -EINVAL; 2004 2005 filename = getname_kernel(kernel_filename); 2006 if (IS_ERR(filename)) 2007 return PTR_ERR(filename); 2008 2009 bprm = alloc_bprm(fd, filename, 0); 2010 if (IS_ERR(bprm)) { 2011 retval = PTR_ERR(bprm); 2012 goto out_ret; 2013 } 2014 2015 retval = count_strings_kernel(argv); 2016 if (WARN_ON_ONCE(retval == 0)) 2017 retval = -EINVAL; 2018 if (retval < 0) 2019 goto out_free; 2020 bprm->argc = retval; 2021 2022 retval = count_strings_kernel(envp); 2023 if (retval < 0) 2024 goto out_free; 2025 bprm->envc = retval; 2026 2027 retval = bprm_stack_limits(bprm); 2028 if (retval < 0) 2029 goto out_free; 2030 2031 retval = copy_string_kernel(bprm->filename, bprm); 2032 if (retval < 0) 2033 goto out_free; 2034 bprm->exec = bprm->p; 2035 2036 retval = copy_strings_kernel(bprm->envc, envp, bprm); 2037 if (retval < 0) 2038 goto out_free; 2039 2040 retval = copy_strings_kernel(bprm->argc, argv, bprm); 2041 if (retval < 0) 2042 goto out_free; 2043 2044 retval = bprm_execve(bprm); 2045 out_free: 2046 free_bprm(bprm); 2047 out_ret: 2048 putname(filename); 2049 return retval; 2050 } 2051 2052 static int do_execve(struct filename *filename, 2053 const char __user *const __user *__argv, 2054 const char __user *const __user *__envp) 2055 { 2056 struct user_arg_ptr argv = { .ptr.native = __argv }; 2057 struct user_arg_ptr envp = { .ptr.native = __envp }; 2058 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0); 2059 } 2060 2061 static int do_execveat(int fd, struct filename *filename, 2062 const char __user *const __user *__argv, 2063 const char __user *const __user *__envp, 2064 int flags) 2065 { 2066 struct user_arg_ptr argv = { .ptr.native = __argv }; 2067 struct user_arg_ptr envp = { .ptr.native = __envp }; 2068 2069 return do_execveat_common(fd, filename, argv, envp, flags); 2070 } 2071 2072 #ifdef CONFIG_COMPAT 2073 static int compat_do_execve(struct filename *filename, 2074 const compat_uptr_t __user *__argv, 2075 const compat_uptr_t __user *__envp) 2076 { 2077 struct user_arg_ptr argv = { 2078 .is_compat = true, 2079 .ptr.compat = __argv, 2080 }; 2081 struct user_arg_ptr envp = { 2082 .is_compat = true, 2083 .ptr.compat = __envp, 2084 }; 2085 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0); 2086 } 2087 2088 static int compat_do_execveat(int fd, struct filename *filename, 2089 const compat_uptr_t __user *__argv, 2090 const compat_uptr_t __user *__envp, 2091 int flags) 2092 { 2093 struct user_arg_ptr argv = { 2094 .is_compat = true, 2095 .ptr.compat = __argv, 2096 }; 2097 struct user_arg_ptr envp = { 2098 .is_compat = true, 2099 .ptr.compat = __envp, 2100 }; 2101 return do_execveat_common(fd, filename, argv, envp, flags); 2102 } 2103 #endif 2104 2105 void set_binfmt(struct linux_binfmt *new) 2106 { 2107 struct mm_struct *mm = current->mm; 2108 2109 if (mm->binfmt) 2110 module_put(mm->binfmt->module); 2111 2112 mm->binfmt = new; 2113 if (new) 2114 __module_get(new->module); 2115 } 2116 EXPORT_SYMBOL(set_binfmt); 2117 2118 /* 2119 * set_dumpable stores three-value SUID_DUMP_* into mm->flags. 2120 */ 2121 void set_dumpable(struct mm_struct *mm, int value) 2122 { 2123 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT)) 2124 return; 2125 2126 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value); 2127 } 2128 2129 SYSCALL_DEFINE3(execve, 2130 const char __user *, filename, 2131 const char __user *const __user *, argv, 2132 const char __user *const __user *, envp) 2133 { 2134 return do_execve(getname(filename), argv, envp); 2135 } 2136 2137 SYSCALL_DEFINE5(execveat, 2138 int, fd, const char __user *, filename, 2139 const char __user *const __user *, argv, 2140 const char __user *const __user *, envp, 2141 int, flags) 2142 { 2143 return do_execveat(fd, 2144 getname_uflags(filename, flags), 2145 argv, envp, flags); 2146 } 2147 2148 #ifdef CONFIG_COMPAT 2149 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename, 2150 const compat_uptr_t __user *, argv, 2151 const compat_uptr_t __user *, envp) 2152 { 2153 return compat_do_execve(getname(filename), argv, envp); 2154 } 2155 2156 COMPAT_SYSCALL_DEFINE5(execveat, int, fd, 2157 const char __user *, filename, 2158 const compat_uptr_t __user *, argv, 2159 const compat_uptr_t __user *, envp, 2160 int, flags) 2161 { 2162 return compat_do_execveat(fd, 2163 getname_uflags(filename, flags), 2164 argv, envp, flags); 2165 } 2166 #endif 2167 2168 #ifdef CONFIG_SYSCTL 2169 2170 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write, 2171 void *buffer, size_t *lenp, loff_t *ppos) 2172 { 2173 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 2174 2175 if (!error) 2176 validate_coredump_safety(); 2177 return error; 2178 } 2179 2180 static struct ctl_table fs_exec_sysctls[] = { 2181 { 2182 .procname = "suid_dumpable", 2183 .data = &suid_dumpable, 2184 .maxlen = sizeof(int), 2185 .mode = 0644, 2186 .proc_handler = proc_dointvec_minmax_coredump, 2187 .extra1 = SYSCTL_ZERO, 2188 .extra2 = SYSCTL_TWO, 2189 }, 2190 }; 2191 2192 static int __init init_fs_exec_sysctls(void) 2193 { 2194 register_sysctl_init("fs", fs_exec_sysctls); 2195 return 0; 2196 } 2197 2198 fs_initcall(init_fs_exec_sysctls); 2199 #endif /* CONFIG_SYSCTL */ 2200