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/slab.h> 27 #include <linux/file.h> 28 #include <linux/fdtable.h> 29 #include <linux/mm.h> 30 #include <linux/vmacache.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/tracehook.h> 59 #include <linux/kmod.h> 60 #include <linux/fsnotify.h> 61 #include <linux/fs_struct.h> 62 #include <linux/oom.h> 63 #include <linux/compat.h> 64 #include <linux/vmalloc.h> 65 66 #include <linux/uaccess.h> 67 #include <asm/mmu_context.h> 68 #include <asm/tlb.h> 69 70 #include <trace/events/task.h> 71 #include "internal.h" 72 73 #include <trace/events/sched.h> 74 75 int suid_dumpable = 0; 76 77 static LIST_HEAD(formats); 78 static DEFINE_RWLOCK(binfmt_lock); 79 80 void __register_binfmt(struct linux_binfmt * fmt, int insert) 81 { 82 BUG_ON(!fmt); 83 if (WARN_ON(!fmt->load_binary)) 84 return; 85 write_lock(&binfmt_lock); 86 insert ? list_add(&fmt->lh, &formats) : 87 list_add_tail(&fmt->lh, &formats); 88 write_unlock(&binfmt_lock); 89 } 90 91 EXPORT_SYMBOL(__register_binfmt); 92 93 void unregister_binfmt(struct linux_binfmt * fmt) 94 { 95 write_lock(&binfmt_lock); 96 list_del(&fmt->lh); 97 write_unlock(&binfmt_lock); 98 } 99 100 EXPORT_SYMBOL(unregister_binfmt); 101 102 static inline void put_binfmt(struct linux_binfmt * fmt) 103 { 104 module_put(fmt->module); 105 } 106 107 bool path_noexec(const struct path *path) 108 { 109 return (path->mnt->mnt_flags & MNT_NOEXEC) || 110 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC); 111 } 112 113 #ifdef CONFIG_USELIB 114 /* 115 * Note that a shared library must be both readable and executable due to 116 * security reasons. 117 * 118 * Also note that we take the address to load from from the file itself. 119 */ 120 SYSCALL_DEFINE1(uselib, const char __user *, library) 121 { 122 struct linux_binfmt *fmt; 123 struct file *file; 124 struct filename *tmp = getname(library); 125 int error = PTR_ERR(tmp); 126 static const struct open_flags uselib_flags = { 127 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC, 128 .acc_mode = MAY_READ | MAY_EXEC, 129 .intent = LOOKUP_OPEN, 130 .lookup_flags = LOOKUP_FOLLOW, 131 }; 132 133 if (IS_ERR(tmp)) 134 goto out; 135 136 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags); 137 putname(tmp); 138 error = PTR_ERR(file); 139 if (IS_ERR(file)) 140 goto out; 141 142 error = -EINVAL; 143 if (!S_ISREG(file_inode(file)->i_mode)) 144 goto exit; 145 146 error = -EACCES; 147 if (path_noexec(&file->f_path)) 148 goto exit; 149 150 fsnotify_open(file); 151 152 error = -ENOEXEC; 153 154 read_lock(&binfmt_lock); 155 list_for_each_entry(fmt, &formats, lh) { 156 if (!fmt->load_shlib) 157 continue; 158 if (!try_module_get(fmt->module)) 159 continue; 160 read_unlock(&binfmt_lock); 161 error = fmt->load_shlib(file); 162 read_lock(&binfmt_lock); 163 put_binfmt(fmt); 164 if (error != -ENOEXEC) 165 break; 166 } 167 read_unlock(&binfmt_lock); 168 exit: 169 fput(file); 170 out: 171 return error; 172 } 173 #endif /* #ifdef CONFIG_USELIB */ 174 175 #ifdef CONFIG_MMU 176 /* 177 * The nascent bprm->mm is not visible until exec_mmap() but it can 178 * use a lot of memory, account these pages in current->mm temporary 179 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we 180 * change the counter back via acct_arg_size(0). 181 */ 182 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages) 183 { 184 struct mm_struct *mm = current->mm; 185 long diff = (long)(pages - bprm->vma_pages); 186 187 if (!mm || !diff) 188 return; 189 190 bprm->vma_pages = pages; 191 add_mm_counter(mm, MM_ANONPAGES, diff); 192 } 193 194 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, 195 int write) 196 { 197 struct page *page; 198 int ret; 199 unsigned int gup_flags = FOLL_FORCE; 200 201 #ifdef CONFIG_STACK_GROWSUP 202 if (write) { 203 ret = expand_downwards(bprm->vma, pos); 204 if (ret < 0) 205 return NULL; 206 } 207 #endif 208 209 if (write) 210 gup_flags |= FOLL_WRITE; 211 212 /* 213 * We are doing an exec(). 'current' is the process 214 * doing the exec and bprm->mm is the new process's mm. 215 */ 216 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags, 217 &page, NULL, NULL); 218 if (ret <= 0) 219 return NULL; 220 221 if (write) 222 acct_arg_size(bprm, vma_pages(bprm->vma)); 223 224 return page; 225 } 226 227 static void put_arg_page(struct page *page) 228 { 229 put_page(page); 230 } 231 232 static void free_arg_pages(struct linux_binprm *bprm) 233 { 234 } 235 236 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, 237 struct page *page) 238 { 239 flush_cache_page(bprm->vma, pos, page_to_pfn(page)); 240 } 241 242 static int __bprm_mm_init(struct linux_binprm *bprm) 243 { 244 int err; 245 struct vm_area_struct *vma = NULL; 246 struct mm_struct *mm = bprm->mm; 247 248 bprm->vma = vma = vm_area_alloc(mm); 249 if (!vma) 250 return -ENOMEM; 251 vma_set_anonymous(vma); 252 253 if (down_write_killable(&mm->mmap_sem)) { 254 err = -EINTR; 255 goto err_free; 256 } 257 258 /* 259 * Place the stack at the largest stack address the architecture 260 * supports. Later, we'll move this to an appropriate place. We don't 261 * use STACK_TOP because that can depend on attributes which aren't 262 * configured yet. 263 */ 264 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP); 265 vma->vm_end = STACK_TOP_MAX; 266 vma->vm_start = vma->vm_end - PAGE_SIZE; 267 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP; 268 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 269 270 err = insert_vm_struct(mm, vma); 271 if (err) 272 goto err; 273 274 mm->stack_vm = mm->total_vm = 1; 275 up_write(&mm->mmap_sem); 276 bprm->p = vma->vm_end - sizeof(void *); 277 return 0; 278 err: 279 up_write(&mm->mmap_sem); 280 err_free: 281 bprm->vma = NULL; 282 vm_area_free(vma); 283 return err; 284 } 285 286 static bool valid_arg_len(struct linux_binprm *bprm, long len) 287 { 288 return len <= MAX_ARG_STRLEN; 289 } 290 291 #else 292 293 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages) 294 { 295 } 296 297 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos, 298 int write) 299 { 300 struct page *page; 301 302 page = bprm->page[pos / PAGE_SIZE]; 303 if (!page && write) { 304 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO); 305 if (!page) 306 return NULL; 307 bprm->page[pos / PAGE_SIZE] = page; 308 } 309 310 return page; 311 } 312 313 static void put_arg_page(struct page *page) 314 { 315 } 316 317 static void free_arg_page(struct linux_binprm *bprm, int i) 318 { 319 if (bprm->page[i]) { 320 __free_page(bprm->page[i]); 321 bprm->page[i] = NULL; 322 } 323 } 324 325 static void free_arg_pages(struct linux_binprm *bprm) 326 { 327 int i; 328 329 for (i = 0; i < MAX_ARG_PAGES; i++) 330 free_arg_page(bprm, i); 331 } 332 333 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos, 334 struct page *page) 335 { 336 } 337 338 static int __bprm_mm_init(struct linux_binprm *bprm) 339 { 340 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *); 341 return 0; 342 } 343 344 static bool valid_arg_len(struct linux_binprm *bprm, long len) 345 { 346 return len <= bprm->p; 347 } 348 349 #endif /* CONFIG_MMU */ 350 351 /* 352 * Create a new mm_struct and populate it with a temporary stack 353 * vm_area_struct. We don't have enough context at this point to set the stack 354 * flags, permissions, and offset, so we use temporary values. We'll update 355 * them later in setup_arg_pages(). 356 */ 357 static int bprm_mm_init(struct linux_binprm *bprm) 358 { 359 int err; 360 struct mm_struct *mm = NULL; 361 362 bprm->mm = mm = mm_alloc(); 363 err = -ENOMEM; 364 if (!mm) 365 goto err; 366 367 /* Save current stack limit for all calculations made during exec. */ 368 task_lock(current->group_leader); 369 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK]; 370 task_unlock(current->group_leader); 371 372 err = __bprm_mm_init(bprm); 373 if (err) 374 goto err; 375 376 return 0; 377 378 err: 379 if (mm) { 380 bprm->mm = NULL; 381 mmdrop(mm); 382 } 383 384 return err; 385 } 386 387 struct user_arg_ptr { 388 #ifdef CONFIG_COMPAT 389 bool is_compat; 390 #endif 391 union { 392 const char __user *const __user *native; 393 #ifdef CONFIG_COMPAT 394 const compat_uptr_t __user *compat; 395 #endif 396 } ptr; 397 }; 398 399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr) 400 { 401 const char __user *native; 402 403 #ifdef CONFIG_COMPAT 404 if (unlikely(argv.is_compat)) { 405 compat_uptr_t compat; 406 407 if (get_user(compat, argv.ptr.compat + nr)) 408 return ERR_PTR(-EFAULT); 409 410 return compat_ptr(compat); 411 } 412 #endif 413 414 if (get_user(native, argv.ptr.native + nr)) 415 return ERR_PTR(-EFAULT); 416 417 return native; 418 } 419 420 /* 421 * count() counts the number of strings in array ARGV. 422 */ 423 static int count(struct user_arg_ptr argv, int max) 424 { 425 int i = 0; 426 427 if (argv.ptr.native != NULL) { 428 for (;;) { 429 const char __user *p = get_user_arg_ptr(argv, i); 430 431 if (!p) 432 break; 433 434 if (IS_ERR(p)) 435 return -EFAULT; 436 437 if (i >= max) 438 return -E2BIG; 439 ++i; 440 441 if (fatal_signal_pending(current)) 442 return -ERESTARTNOHAND; 443 cond_resched(); 444 } 445 } 446 return i; 447 } 448 449 static int prepare_arg_pages(struct linux_binprm *bprm, 450 struct user_arg_ptr argv, struct user_arg_ptr envp) 451 { 452 unsigned long limit, ptr_size; 453 454 bprm->argc = count(argv, MAX_ARG_STRINGS); 455 if (bprm->argc < 0) 456 return bprm->argc; 457 458 bprm->envc = count(envp, MAX_ARG_STRINGS); 459 if (bprm->envc < 0) 460 return bprm->envc; 461 462 /* 463 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM 464 * (whichever is smaller) for the argv+env strings. 465 * This ensures that: 466 * - the remaining binfmt code will not run out of stack space, 467 * - the program will have a reasonable amount of stack left 468 * to work from. 469 */ 470 limit = _STK_LIM / 4 * 3; 471 limit = min(limit, bprm->rlim_stack.rlim_cur / 4); 472 /* 473 * We've historically supported up to 32 pages (ARG_MAX) 474 * of argument strings even with small stacks 475 */ 476 limit = max_t(unsigned long, limit, ARG_MAX); 477 /* 478 * We must account for the size of all the argv and envp pointers to 479 * the argv and envp strings, since they will also take up space in 480 * the stack. They aren't stored until much later when we can't 481 * signal to the parent that the child has run out of stack space. 482 * Instead, calculate it here so it's possible to fail gracefully. 483 */ 484 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *); 485 if (limit <= ptr_size) 486 return -E2BIG; 487 limit -= ptr_size; 488 489 bprm->argmin = bprm->p - limit; 490 return 0; 491 } 492 493 /* 494 * 'copy_strings()' copies argument/environment strings from the old 495 * processes's memory to the new process's stack. The call to get_user_pages() 496 * ensures the destination page is created and not swapped out. 497 */ 498 static int copy_strings(int argc, struct user_arg_ptr argv, 499 struct linux_binprm *bprm) 500 { 501 struct page *kmapped_page = NULL; 502 char *kaddr = NULL; 503 unsigned long kpos = 0; 504 int ret; 505 506 while (argc-- > 0) { 507 const char __user *str; 508 int len; 509 unsigned long pos; 510 511 ret = -EFAULT; 512 str = get_user_arg_ptr(argv, argc); 513 if (IS_ERR(str)) 514 goto out; 515 516 len = strnlen_user(str, MAX_ARG_STRLEN); 517 if (!len) 518 goto out; 519 520 ret = -E2BIG; 521 if (!valid_arg_len(bprm, len)) 522 goto out; 523 524 /* We're going to work our way backwords. */ 525 pos = bprm->p; 526 str += len; 527 bprm->p -= len; 528 #ifdef CONFIG_MMU 529 if (bprm->p < bprm->argmin) 530 goto out; 531 #endif 532 533 while (len > 0) { 534 int offset, bytes_to_copy; 535 536 if (fatal_signal_pending(current)) { 537 ret = -ERESTARTNOHAND; 538 goto out; 539 } 540 cond_resched(); 541 542 offset = pos % PAGE_SIZE; 543 if (offset == 0) 544 offset = PAGE_SIZE; 545 546 bytes_to_copy = offset; 547 if (bytes_to_copy > len) 548 bytes_to_copy = len; 549 550 offset -= bytes_to_copy; 551 pos -= bytes_to_copy; 552 str -= bytes_to_copy; 553 len -= bytes_to_copy; 554 555 if (!kmapped_page || kpos != (pos & PAGE_MASK)) { 556 struct page *page; 557 558 page = get_arg_page(bprm, pos, 1); 559 if (!page) { 560 ret = -E2BIG; 561 goto out; 562 } 563 564 if (kmapped_page) { 565 flush_kernel_dcache_page(kmapped_page); 566 kunmap(kmapped_page); 567 put_arg_page(kmapped_page); 568 } 569 kmapped_page = page; 570 kaddr = kmap(kmapped_page); 571 kpos = pos & PAGE_MASK; 572 flush_arg_page(bprm, kpos, kmapped_page); 573 } 574 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) { 575 ret = -EFAULT; 576 goto out; 577 } 578 } 579 } 580 ret = 0; 581 out: 582 if (kmapped_page) { 583 flush_kernel_dcache_page(kmapped_page); 584 kunmap(kmapped_page); 585 put_arg_page(kmapped_page); 586 } 587 return ret; 588 } 589 590 /* 591 * Like copy_strings, but get argv and its values from kernel memory. 592 */ 593 int copy_strings_kernel(int argc, const char *const *__argv, 594 struct linux_binprm *bprm) 595 { 596 int r; 597 mm_segment_t oldfs = get_fs(); 598 struct user_arg_ptr argv = { 599 .ptr.native = (const char __user *const __user *)__argv, 600 }; 601 602 set_fs(KERNEL_DS); 603 r = copy_strings(argc, argv, bprm); 604 set_fs(oldfs); 605 606 return r; 607 } 608 EXPORT_SYMBOL(copy_strings_kernel); 609 610 #ifdef CONFIG_MMU 611 612 /* 613 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once 614 * the binfmt code determines where the new stack should reside, we shift it to 615 * its final location. The process proceeds as follows: 616 * 617 * 1) Use shift to calculate the new vma endpoints. 618 * 2) Extend vma to cover both the old and new ranges. This ensures the 619 * arguments passed to subsequent functions are consistent. 620 * 3) Move vma's page tables to the new range. 621 * 4) Free up any cleared pgd range. 622 * 5) Shrink the vma to cover only the new range. 623 */ 624 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift) 625 { 626 struct mm_struct *mm = vma->vm_mm; 627 unsigned long old_start = vma->vm_start; 628 unsigned long old_end = vma->vm_end; 629 unsigned long length = old_end - old_start; 630 unsigned long new_start = old_start - shift; 631 unsigned long new_end = old_end - shift; 632 struct mmu_gather tlb; 633 634 BUG_ON(new_start > new_end); 635 636 /* 637 * ensure there are no vmas between where we want to go 638 * and where we are 639 */ 640 if (vma != find_vma(mm, new_start)) 641 return -EFAULT; 642 643 /* 644 * cover the whole range: [new_start, old_end) 645 */ 646 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL)) 647 return -ENOMEM; 648 649 /* 650 * move the page tables downwards, on failure we rely on 651 * process cleanup to remove whatever mess we made. 652 */ 653 if (length != move_page_tables(vma, old_start, 654 vma, new_start, length, false)) 655 return -ENOMEM; 656 657 lru_add_drain(); 658 tlb_gather_mmu(&tlb, mm, old_start, old_end); 659 if (new_end > old_start) { 660 /* 661 * when the old and new regions overlap clear from new_end. 662 */ 663 free_pgd_range(&tlb, new_end, old_end, new_end, 664 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING); 665 } else { 666 /* 667 * otherwise, clean from old_start; this is done to not touch 668 * the address space in [new_end, old_start) some architectures 669 * have constraints on va-space that make this illegal (IA64) - 670 * for the others its just a little faster. 671 */ 672 free_pgd_range(&tlb, old_start, old_end, new_end, 673 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING); 674 } 675 tlb_finish_mmu(&tlb, old_start, old_end); 676 677 /* 678 * Shrink the vma to just the new range. Always succeeds. 679 */ 680 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL); 681 682 return 0; 683 } 684 685 /* 686 * Finalizes the stack vm_area_struct. The flags and permissions are updated, 687 * the stack is optionally relocated, and some extra space is added. 688 */ 689 int setup_arg_pages(struct linux_binprm *bprm, 690 unsigned long stack_top, 691 int executable_stack) 692 { 693 unsigned long ret; 694 unsigned long stack_shift; 695 struct mm_struct *mm = current->mm; 696 struct vm_area_struct *vma = bprm->vma; 697 struct vm_area_struct *prev = NULL; 698 unsigned long vm_flags; 699 unsigned long stack_base; 700 unsigned long stack_size; 701 unsigned long stack_expand; 702 unsigned long rlim_stack; 703 704 #ifdef CONFIG_STACK_GROWSUP 705 /* Limit stack size */ 706 stack_base = bprm->rlim_stack.rlim_max; 707 if (stack_base > STACK_SIZE_MAX) 708 stack_base = STACK_SIZE_MAX; 709 710 /* Add space for stack randomization. */ 711 stack_base += (STACK_RND_MASK << PAGE_SHIFT); 712 713 /* Make sure we didn't let the argument array grow too large. */ 714 if (vma->vm_end - vma->vm_start > stack_base) 715 return -ENOMEM; 716 717 stack_base = PAGE_ALIGN(stack_top - stack_base); 718 719 stack_shift = vma->vm_start - stack_base; 720 mm->arg_start = bprm->p - stack_shift; 721 bprm->p = vma->vm_end - stack_shift; 722 #else 723 stack_top = arch_align_stack(stack_top); 724 stack_top = PAGE_ALIGN(stack_top); 725 726 if (unlikely(stack_top < mmap_min_addr) || 727 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr)) 728 return -ENOMEM; 729 730 stack_shift = vma->vm_end - stack_top; 731 732 bprm->p -= stack_shift; 733 mm->arg_start = bprm->p; 734 #endif 735 736 if (bprm->loader) 737 bprm->loader -= stack_shift; 738 bprm->exec -= stack_shift; 739 740 if (down_write_killable(&mm->mmap_sem)) 741 return -EINTR; 742 743 vm_flags = VM_STACK_FLAGS; 744 745 /* 746 * Adjust stack execute permissions; explicitly enable for 747 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone 748 * (arch default) otherwise. 749 */ 750 if (unlikely(executable_stack == EXSTACK_ENABLE_X)) 751 vm_flags |= VM_EXEC; 752 else if (executable_stack == EXSTACK_DISABLE_X) 753 vm_flags &= ~VM_EXEC; 754 vm_flags |= mm->def_flags; 755 vm_flags |= VM_STACK_INCOMPLETE_SETUP; 756 757 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end, 758 vm_flags); 759 if (ret) 760 goto out_unlock; 761 BUG_ON(prev != vma); 762 763 if (unlikely(vm_flags & VM_EXEC)) { 764 pr_warn_once("process '%pD4' started with executable stack\n", 765 bprm->file); 766 } 767 768 /* Move stack pages down in memory. */ 769 if (stack_shift) { 770 ret = shift_arg_pages(vma, stack_shift); 771 if (ret) 772 goto out_unlock; 773 } 774 775 /* mprotect_fixup is overkill to remove the temporary stack flags */ 776 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP; 777 778 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */ 779 stack_size = vma->vm_end - vma->vm_start; 780 /* 781 * Align this down to a page boundary as expand_stack 782 * will align it up. 783 */ 784 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK; 785 #ifdef CONFIG_STACK_GROWSUP 786 if (stack_size + stack_expand > rlim_stack) 787 stack_base = vma->vm_start + rlim_stack; 788 else 789 stack_base = vma->vm_end + stack_expand; 790 #else 791 if (stack_size + stack_expand > rlim_stack) 792 stack_base = vma->vm_end - rlim_stack; 793 else 794 stack_base = vma->vm_start - stack_expand; 795 #endif 796 current->mm->start_stack = bprm->p; 797 ret = expand_stack(vma, stack_base); 798 if (ret) 799 ret = -EFAULT; 800 801 out_unlock: 802 up_write(&mm->mmap_sem); 803 return ret; 804 } 805 EXPORT_SYMBOL(setup_arg_pages); 806 807 #else 808 809 /* 810 * Transfer the program arguments and environment from the holding pages 811 * onto the stack. The provided stack pointer is adjusted accordingly. 812 */ 813 int transfer_args_to_stack(struct linux_binprm *bprm, 814 unsigned long *sp_location) 815 { 816 unsigned long index, stop, sp; 817 int ret = 0; 818 819 stop = bprm->p >> PAGE_SHIFT; 820 sp = *sp_location; 821 822 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) { 823 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0; 824 char *src = kmap(bprm->page[index]) + offset; 825 sp -= PAGE_SIZE - offset; 826 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0) 827 ret = -EFAULT; 828 kunmap(bprm->page[index]); 829 if (ret) 830 goto out; 831 } 832 833 *sp_location = sp; 834 835 out: 836 return ret; 837 } 838 EXPORT_SYMBOL(transfer_args_to_stack); 839 840 #endif /* CONFIG_MMU */ 841 842 static struct file *do_open_execat(int fd, struct filename *name, int flags) 843 { 844 struct file *file; 845 int err; 846 struct open_flags open_exec_flags = { 847 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC, 848 .acc_mode = MAY_EXEC, 849 .intent = LOOKUP_OPEN, 850 .lookup_flags = LOOKUP_FOLLOW, 851 }; 852 853 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0) 854 return ERR_PTR(-EINVAL); 855 if (flags & AT_SYMLINK_NOFOLLOW) 856 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW; 857 if (flags & AT_EMPTY_PATH) 858 open_exec_flags.lookup_flags |= LOOKUP_EMPTY; 859 860 file = do_filp_open(fd, name, &open_exec_flags); 861 if (IS_ERR(file)) 862 goto out; 863 864 err = -EACCES; 865 if (!S_ISREG(file_inode(file)->i_mode)) 866 goto exit; 867 868 if (path_noexec(&file->f_path)) 869 goto exit; 870 871 err = deny_write_access(file); 872 if (err) 873 goto exit; 874 875 if (name->name[0] != '\0') 876 fsnotify_open(file); 877 878 out: 879 return file; 880 881 exit: 882 fput(file); 883 return ERR_PTR(err); 884 } 885 886 struct file *open_exec(const char *name) 887 { 888 struct filename *filename = getname_kernel(name); 889 struct file *f = ERR_CAST(filename); 890 891 if (!IS_ERR(filename)) { 892 f = do_open_execat(AT_FDCWD, filename, 0); 893 putname(filename); 894 } 895 return f; 896 } 897 EXPORT_SYMBOL(open_exec); 898 899 int kernel_read_file(struct file *file, void **buf, loff_t *size, 900 loff_t max_size, enum kernel_read_file_id id) 901 { 902 loff_t i_size, pos; 903 ssize_t bytes = 0; 904 int ret; 905 906 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0) 907 return -EINVAL; 908 909 ret = deny_write_access(file); 910 if (ret) 911 return ret; 912 913 ret = security_kernel_read_file(file, id); 914 if (ret) 915 goto out; 916 917 i_size = i_size_read(file_inode(file)); 918 if (i_size <= 0) { 919 ret = -EINVAL; 920 goto out; 921 } 922 if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) { 923 ret = -EFBIG; 924 goto out; 925 } 926 927 if (id != READING_FIRMWARE_PREALLOC_BUFFER) 928 *buf = vmalloc(i_size); 929 if (!*buf) { 930 ret = -ENOMEM; 931 goto out; 932 } 933 934 pos = 0; 935 while (pos < i_size) { 936 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos); 937 if (bytes < 0) { 938 ret = bytes; 939 goto out_free; 940 } 941 942 if (bytes == 0) 943 break; 944 } 945 946 if (pos != i_size) { 947 ret = -EIO; 948 goto out_free; 949 } 950 951 ret = security_kernel_post_read_file(file, *buf, i_size, id); 952 if (!ret) 953 *size = pos; 954 955 out_free: 956 if (ret < 0) { 957 if (id != READING_FIRMWARE_PREALLOC_BUFFER) { 958 vfree(*buf); 959 *buf = NULL; 960 } 961 } 962 963 out: 964 allow_write_access(file); 965 return ret; 966 } 967 EXPORT_SYMBOL_GPL(kernel_read_file); 968 969 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size, 970 loff_t max_size, enum kernel_read_file_id id) 971 { 972 struct file *file; 973 int ret; 974 975 if (!path || !*path) 976 return -EINVAL; 977 978 file = filp_open(path, O_RDONLY, 0); 979 if (IS_ERR(file)) 980 return PTR_ERR(file); 981 982 ret = kernel_read_file(file, buf, size, max_size, id); 983 fput(file); 984 return ret; 985 } 986 EXPORT_SYMBOL_GPL(kernel_read_file_from_path); 987 988 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size, 989 enum kernel_read_file_id id) 990 { 991 struct fd f = fdget(fd); 992 int ret = -EBADF; 993 994 if (!f.file) 995 goto out; 996 997 ret = kernel_read_file(f.file, buf, size, max_size, id); 998 out: 999 fdput(f); 1000 return ret; 1001 } 1002 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd); 1003 1004 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len) 1005 { 1006 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos); 1007 if (res > 0) 1008 flush_icache_range(addr, addr + len); 1009 return res; 1010 } 1011 EXPORT_SYMBOL(read_code); 1012 1013 static int exec_mmap(struct mm_struct *mm) 1014 { 1015 struct task_struct *tsk; 1016 struct mm_struct *old_mm, *active_mm; 1017 1018 /* Notify parent that we're no longer interested in the old VM */ 1019 tsk = current; 1020 old_mm = current->mm; 1021 exec_mm_release(tsk, old_mm); 1022 1023 if (old_mm) { 1024 sync_mm_rss(old_mm); 1025 /* 1026 * Make sure that if there is a core dump in progress 1027 * for the old mm, we get out and die instead of going 1028 * through with the exec. We must hold mmap_sem around 1029 * checking core_state and changing tsk->mm. 1030 */ 1031 down_read(&old_mm->mmap_sem); 1032 if (unlikely(old_mm->core_state)) { 1033 up_read(&old_mm->mmap_sem); 1034 return -EINTR; 1035 } 1036 } 1037 task_lock(tsk); 1038 active_mm = tsk->active_mm; 1039 membarrier_exec_mmap(mm); 1040 tsk->mm = mm; 1041 tsk->active_mm = mm; 1042 activate_mm(active_mm, mm); 1043 tsk->mm->vmacache_seqnum = 0; 1044 vmacache_flush(tsk); 1045 task_unlock(tsk); 1046 if (old_mm) { 1047 up_read(&old_mm->mmap_sem); 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(active_mm); 1055 return 0; 1056 } 1057 1058 /* 1059 * This function makes sure the current process has its own signal table, 1060 * so that flush_signal_handlers can later reset the handlers without 1061 * disturbing other processes. (Other processes might share the signal 1062 * table via the CLONE_SIGHAND option to clone().) 1063 */ 1064 static int de_thread(struct task_struct *tsk) 1065 { 1066 struct signal_struct *sig = tsk->signal; 1067 struct sighand_struct *oldsighand = tsk->sighand; 1068 spinlock_t *lock = &oldsighand->siglock; 1069 1070 if (thread_group_empty(tsk)) 1071 goto no_thread_group; 1072 1073 /* 1074 * Kill all other threads in the thread group. 1075 */ 1076 spin_lock_irq(lock); 1077 if (signal_group_exit(sig)) { 1078 /* 1079 * Another group action in progress, just 1080 * return so that the signal is processed. 1081 */ 1082 spin_unlock_irq(lock); 1083 return -EAGAIN; 1084 } 1085 1086 sig->group_exit_task = tsk; 1087 sig->notify_count = zap_other_threads(tsk); 1088 if (!thread_group_leader(tsk)) 1089 sig->notify_count--; 1090 1091 while (sig->notify_count) { 1092 __set_current_state(TASK_KILLABLE); 1093 spin_unlock_irq(lock); 1094 schedule(); 1095 if (__fatal_signal_pending(tsk)) 1096 goto killed; 1097 spin_lock_irq(lock); 1098 } 1099 spin_unlock_irq(lock); 1100 1101 /* 1102 * At this point all other threads have exited, all we have to 1103 * do is to wait for the thread group leader to become inactive, 1104 * and to assume its PID: 1105 */ 1106 if (!thread_group_leader(tsk)) { 1107 struct task_struct *leader = tsk->group_leader; 1108 1109 for (;;) { 1110 cgroup_threadgroup_change_begin(tsk); 1111 write_lock_irq(&tasklist_lock); 1112 /* 1113 * Do this under tasklist_lock to ensure that 1114 * exit_notify() can't miss ->group_exit_task 1115 */ 1116 sig->notify_count = -1; 1117 if (likely(leader->exit_state)) 1118 break; 1119 __set_current_state(TASK_KILLABLE); 1120 write_unlock_irq(&tasklist_lock); 1121 cgroup_threadgroup_change_end(tsk); 1122 schedule(); 1123 if (__fatal_signal_pending(tsk)) 1124 goto killed; 1125 } 1126 1127 /* 1128 * The only record we have of the real-time age of a 1129 * process, regardless of execs it's done, is start_time. 1130 * All the past CPU time is accumulated in signal_struct 1131 * from sister threads now dead. But in this non-leader 1132 * exec, nothing survives from the original leader thread, 1133 * whose birth marks the true age of this process now. 1134 * When we take on its identity by switching to its PID, we 1135 * also take its birthdate (always earlier than our own). 1136 */ 1137 tsk->start_time = leader->start_time; 1138 tsk->start_boottime = leader->start_boottime; 1139 1140 BUG_ON(!same_thread_group(leader, tsk)); 1141 BUG_ON(has_group_leader_pid(tsk)); 1142 /* 1143 * An exec() starts a new thread group with the 1144 * TGID of the previous thread group. Rehash the 1145 * two threads with a switched PID, and release 1146 * the former thread group leader: 1147 */ 1148 1149 /* Become a process group leader with the old leader's pid. 1150 * The old leader becomes a thread of the this thread group. 1151 * Note: The old leader also uses this pid until release_task 1152 * is called. Odd but simple and correct. 1153 */ 1154 tsk->pid = leader->pid; 1155 change_pid(tsk, PIDTYPE_PID, task_pid(leader)); 1156 transfer_pid(leader, tsk, PIDTYPE_TGID); 1157 transfer_pid(leader, tsk, PIDTYPE_PGID); 1158 transfer_pid(leader, tsk, PIDTYPE_SID); 1159 1160 list_replace_rcu(&leader->tasks, &tsk->tasks); 1161 list_replace_init(&leader->sibling, &tsk->sibling); 1162 1163 tsk->group_leader = tsk; 1164 leader->group_leader = tsk; 1165 1166 tsk->exit_signal = SIGCHLD; 1167 leader->exit_signal = -1; 1168 1169 BUG_ON(leader->exit_state != EXIT_ZOMBIE); 1170 leader->exit_state = EXIT_DEAD; 1171 1172 /* 1173 * We are going to release_task()->ptrace_unlink() silently, 1174 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees 1175 * the tracer wont't block again waiting for this thread. 1176 */ 1177 if (unlikely(leader->ptrace)) 1178 __wake_up_parent(leader, leader->parent); 1179 write_unlock_irq(&tasklist_lock); 1180 cgroup_threadgroup_change_end(tsk); 1181 1182 release_task(leader); 1183 } 1184 1185 sig->group_exit_task = NULL; 1186 sig->notify_count = 0; 1187 1188 no_thread_group: 1189 /* we have changed execution domain */ 1190 tsk->exit_signal = SIGCHLD; 1191 1192 #ifdef CONFIG_POSIX_TIMERS 1193 exit_itimers(sig); 1194 flush_itimer_signals(); 1195 #endif 1196 1197 if (refcount_read(&oldsighand->count) != 1) { 1198 struct sighand_struct *newsighand; 1199 /* 1200 * This ->sighand is shared with the CLONE_SIGHAND 1201 * but not CLONE_THREAD task, switch to the new one. 1202 */ 1203 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); 1204 if (!newsighand) 1205 return -ENOMEM; 1206 1207 refcount_set(&newsighand->count, 1); 1208 memcpy(newsighand->action, oldsighand->action, 1209 sizeof(newsighand->action)); 1210 1211 write_lock_irq(&tasklist_lock); 1212 spin_lock(&oldsighand->siglock); 1213 rcu_assign_pointer(tsk->sighand, newsighand); 1214 spin_unlock(&oldsighand->siglock); 1215 write_unlock_irq(&tasklist_lock); 1216 1217 __cleanup_sighand(oldsighand); 1218 } 1219 1220 BUG_ON(!thread_group_leader(tsk)); 1221 return 0; 1222 1223 killed: 1224 /* protects against exit_notify() and __exit_signal() */ 1225 read_lock(&tasklist_lock); 1226 sig->group_exit_task = NULL; 1227 sig->notify_count = 0; 1228 read_unlock(&tasklist_lock); 1229 return -EAGAIN; 1230 } 1231 1232 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk) 1233 { 1234 task_lock(tsk); 1235 strncpy(buf, tsk->comm, buf_size); 1236 task_unlock(tsk); 1237 return buf; 1238 } 1239 EXPORT_SYMBOL_GPL(__get_task_comm); 1240 1241 /* 1242 * These functions flushes out all traces of the currently running executable 1243 * so that a new one can be started 1244 */ 1245 1246 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec) 1247 { 1248 task_lock(tsk); 1249 trace_task_rename(tsk, buf); 1250 strlcpy(tsk->comm, buf, sizeof(tsk->comm)); 1251 task_unlock(tsk); 1252 perf_event_comm(tsk, exec); 1253 } 1254 1255 /* 1256 * Calling this is the point of no return. None of the failures will be 1257 * seen by userspace since either the process is already taking a fatal 1258 * signal (via de_thread() or coredump), or will have SEGV raised 1259 * (after exec_mmap()) by search_binary_handlers (see below). 1260 */ 1261 int flush_old_exec(struct linux_binprm * bprm) 1262 { 1263 int retval; 1264 1265 /* 1266 * Make sure we have a private signal table and that 1267 * we are unassociated from the previous thread group. 1268 */ 1269 retval = de_thread(current); 1270 if (retval) 1271 goto out; 1272 1273 /* 1274 * Must be called _before_ exec_mmap() as bprm->mm is 1275 * not visibile until then. This also enables the update 1276 * to be lockless. 1277 */ 1278 set_mm_exe_file(bprm->mm, bprm->file); 1279 1280 /* 1281 * Release all of the old mmap stuff 1282 */ 1283 acct_arg_size(bprm, 0); 1284 retval = exec_mmap(bprm->mm); 1285 if (retval) 1286 goto out; 1287 1288 /* 1289 * After clearing bprm->mm (to mark that current is using the 1290 * prepared mm now), we have nothing left of the original 1291 * process. If anything from here on returns an error, the check 1292 * in search_binary_handler() will SEGV current. 1293 */ 1294 bprm->mm = NULL; 1295 1296 set_fs(USER_DS); 1297 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | 1298 PF_NOFREEZE | PF_NO_SETAFFINITY); 1299 flush_thread(); 1300 current->personality &= ~bprm->per_clear; 1301 1302 /* 1303 * We have to apply CLOEXEC before we change whether the process is 1304 * dumpable (in setup_new_exec) to avoid a race with a process in userspace 1305 * trying to access the should-be-closed file descriptors of a process 1306 * undergoing exec(2). 1307 */ 1308 do_close_on_exec(current->files); 1309 return 0; 1310 1311 out: 1312 return retval; 1313 } 1314 EXPORT_SYMBOL(flush_old_exec); 1315 1316 void would_dump(struct linux_binprm *bprm, struct file *file) 1317 { 1318 struct inode *inode = file_inode(file); 1319 if (inode_permission(inode, MAY_READ) < 0) { 1320 struct user_namespace *old, *user_ns; 1321 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP; 1322 1323 /* Ensure mm->user_ns contains the executable */ 1324 user_ns = old = bprm->mm->user_ns; 1325 while ((user_ns != &init_user_ns) && 1326 !privileged_wrt_inode_uidgid(user_ns, inode)) 1327 user_ns = user_ns->parent; 1328 1329 if (old != user_ns) { 1330 bprm->mm->user_ns = get_user_ns(user_ns); 1331 put_user_ns(old); 1332 } 1333 } 1334 } 1335 EXPORT_SYMBOL(would_dump); 1336 1337 void setup_new_exec(struct linux_binprm * bprm) 1338 { 1339 /* 1340 * Once here, prepare_binrpm() will not be called any more, so 1341 * the final state of setuid/setgid/fscaps can be merged into the 1342 * secureexec flag. 1343 */ 1344 bprm->secureexec |= bprm->cap_elevated; 1345 1346 if (bprm->secureexec) { 1347 /* Make sure parent cannot signal privileged process. */ 1348 current->pdeath_signal = 0; 1349 1350 /* 1351 * For secureexec, reset the stack limit to sane default to 1352 * avoid bad behavior from the prior rlimits. This has to 1353 * happen before arch_pick_mmap_layout(), which examines 1354 * RLIMIT_STACK, but after the point of no return to avoid 1355 * needing to clean up the change on failure. 1356 */ 1357 if (bprm->rlim_stack.rlim_cur > _STK_LIM) 1358 bprm->rlim_stack.rlim_cur = _STK_LIM; 1359 } 1360 1361 arch_pick_mmap_layout(current->mm, &bprm->rlim_stack); 1362 1363 current->sas_ss_sp = current->sas_ss_size = 0; 1364 1365 /* 1366 * Figure out dumpability. Note that this checking only of current 1367 * is wrong, but userspace depends on it. This should be testing 1368 * bprm->secureexec instead. 1369 */ 1370 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP || 1371 !(uid_eq(current_euid(), current_uid()) && 1372 gid_eq(current_egid(), current_gid()))) 1373 set_dumpable(current->mm, suid_dumpable); 1374 else 1375 set_dumpable(current->mm, SUID_DUMP_USER); 1376 1377 arch_setup_new_exec(); 1378 perf_event_exec(); 1379 __set_task_comm(current, kbasename(bprm->filename), true); 1380 1381 /* Set the new mm task size. We have to do that late because it may 1382 * depend on TIF_32BIT which is only updated in flush_thread() on 1383 * some architectures like powerpc 1384 */ 1385 current->mm->task_size = TASK_SIZE; 1386 1387 /* An exec changes our domain. We are no longer part of the thread 1388 group */ 1389 current->self_exec_id++; 1390 flush_signal_handlers(current, 0); 1391 } 1392 EXPORT_SYMBOL(setup_new_exec); 1393 1394 /* Runs immediately before start_thread() takes over. */ 1395 void finalize_exec(struct linux_binprm *bprm) 1396 { 1397 /* Store any stack rlimit changes before starting thread. */ 1398 task_lock(current->group_leader); 1399 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack; 1400 task_unlock(current->group_leader); 1401 } 1402 EXPORT_SYMBOL(finalize_exec); 1403 1404 /* 1405 * Prepare credentials and lock ->cred_guard_mutex. 1406 * install_exec_creds() commits the new creds and drops the lock. 1407 * Or, if exec fails before, free_bprm() should release ->cred and 1408 * and unlock. 1409 */ 1410 static int prepare_bprm_creds(struct linux_binprm *bprm) 1411 { 1412 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex)) 1413 return -ERESTARTNOINTR; 1414 1415 bprm->cred = prepare_exec_creds(); 1416 if (likely(bprm->cred)) 1417 return 0; 1418 1419 mutex_unlock(¤t->signal->cred_guard_mutex); 1420 return -ENOMEM; 1421 } 1422 1423 static void free_bprm(struct linux_binprm *bprm) 1424 { 1425 free_arg_pages(bprm); 1426 if (bprm->cred) { 1427 mutex_unlock(¤t->signal->cred_guard_mutex); 1428 abort_creds(bprm->cred); 1429 } 1430 if (bprm->file) { 1431 allow_write_access(bprm->file); 1432 fput(bprm->file); 1433 } 1434 /* If a binfmt changed the interp, free it. */ 1435 if (bprm->interp != bprm->filename) 1436 kfree(bprm->interp); 1437 kfree(bprm); 1438 } 1439 1440 int bprm_change_interp(const char *interp, struct linux_binprm *bprm) 1441 { 1442 /* If a binfmt changed the interp, free it first. */ 1443 if (bprm->interp != bprm->filename) 1444 kfree(bprm->interp); 1445 bprm->interp = kstrdup(interp, GFP_KERNEL); 1446 if (!bprm->interp) 1447 return -ENOMEM; 1448 return 0; 1449 } 1450 EXPORT_SYMBOL(bprm_change_interp); 1451 1452 /* 1453 * install the new credentials for this executable 1454 */ 1455 void install_exec_creds(struct linux_binprm *bprm) 1456 { 1457 security_bprm_committing_creds(bprm); 1458 1459 commit_creds(bprm->cred); 1460 bprm->cred = NULL; 1461 1462 /* 1463 * Disable monitoring for regular users 1464 * when executing setuid binaries. Must 1465 * wait until new credentials are committed 1466 * by commit_creds() above 1467 */ 1468 if (get_dumpable(current->mm) != SUID_DUMP_USER) 1469 perf_event_exit_task(current); 1470 /* 1471 * cred_guard_mutex must be held at least to this point to prevent 1472 * ptrace_attach() from altering our determination of the task's 1473 * credentials; any time after this it may be unlocked. 1474 */ 1475 security_bprm_committed_creds(bprm); 1476 mutex_unlock(¤t->signal->cred_guard_mutex); 1477 } 1478 EXPORT_SYMBOL(install_exec_creds); 1479 1480 /* 1481 * determine how safe it is to execute the proposed program 1482 * - the caller must hold ->cred_guard_mutex to protect against 1483 * PTRACE_ATTACH or seccomp thread-sync 1484 */ 1485 static void check_unsafe_exec(struct linux_binprm *bprm) 1486 { 1487 struct task_struct *p = current, *t; 1488 unsigned n_fs; 1489 1490 if (p->ptrace) 1491 bprm->unsafe |= LSM_UNSAFE_PTRACE; 1492 1493 /* 1494 * This isn't strictly necessary, but it makes it harder for LSMs to 1495 * mess up. 1496 */ 1497 if (task_no_new_privs(current)) 1498 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS; 1499 1500 t = p; 1501 n_fs = 1; 1502 spin_lock(&p->fs->lock); 1503 rcu_read_lock(); 1504 while_each_thread(p, t) { 1505 if (t->fs == p->fs) 1506 n_fs++; 1507 } 1508 rcu_read_unlock(); 1509 1510 if (p->fs->users > n_fs) 1511 bprm->unsafe |= LSM_UNSAFE_SHARE; 1512 else 1513 p->fs->in_exec = 1; 1514 spin_unlock(&p->fs->lock); 1515 } 1516 1517 static void bprm_fill_uid(struct linux_binprm *bprm) 1518 { 1519 struct inode *inode; 1520 unsigned int mode; 1521 kuid_t uid; 1522 kgid_t gid; 1523 1524 /* 1525 * Since this can be called multiple times (via prepare_binprm), 1526 * we must clear any previous work done when setting set[ug]id 1527 * bits from any earlier bprm->file uses (for example when run 1528 * first for a setuid script then again for its interpreter). 1529 */ 1530 bprm->cred->euid = current_euid(); 1531 bprm->cred->egid = current_egid(); 1532 1533 if (!mnt_may_suid(bprm->file->f_path.mnt)) 1534 return; 1535 1536 if (task_no_new_privs(current)) 1537 return; 1538 1539 inode = bprm->file->f_path.dentry->d_inode; 1540 mode = READ_ONCE(inode->i_mode); 1541 if (!(mode & (S_ISUID|S_ISGID))) 1542 return; 1543 1544 /* Be careful if suid/sgid is set */ 1545 inode_lock(inode); 1546 1547 /* reload atomically mode/uid/gid now that lock held */ 1548 mode = inode->i_mode; 1549 uid = inode->i_uid; 1550 gid = inode->i_gid; 1551 inode_unlock(inode); 1552 1553 /* We ignore suid/sgid if there are no mappings for them in the ns */ 1554 if (!kuid_has_mapping(bprm->cred->user_ns, uid) || 1555 !kgid_has_mapping(bprm->cred->user_ns, gid)) 1556 return; 1557 1558 if (mode & S_ISUID) { 1559 bprm->per_clear |= PER_CLEAR_ON_SETID; 1560 bprm->cred->euid = uid; 1561 } 1562 1563 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) { 1564 bprm->per_clear |= PER_CLEAR_ON_SETID; 1565 bprm->cred->egid = gid; 1566 } 1567 } 1568 1569 /* 1570 * Fill the binprm structure from the inode. 1571 * Check permissions, then read the first BINPRM_BUF_SIZE bytes 1572 * 1573 * This may be called multiple times for binary chains (scripts for example). 1574 */ 1575 int prepare_binprm(struct linux_binprm *bprm) 1576 { 1577 int retval; 1578 loff_t pos = 0; 1579 1580 bprm_fill_uid(bprm); 1581 1582 /* fill in binprm security blob */ 1583 retval = security_bprm_set_creds(bprm); 1584 if (retval) 1585 return retval; 1586 bprm->called_set_creds = 1; 1587 1588 memset(bprm->buf, 0, BINPRM_BUF_SIZE); 1589 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos); 1590 } 1591 1592 EXPORT_SYMBOL(prepare_binprm); 1593 1594 /* 1595 * Arguments are '\0' separated strings found at the location bprm->p 1596 * points to; chop off the first by relocating brpm->p to right after 1597 * the first '\0' encountered. 1598 */ 1599 int remove_arg_zero(struct linux_binprm *bprm) 1600 { 1601 int ret = 0; 1602 unsigned long offset; 1603 char *kaddr; 1604 struct page *page; 1605 1606 if (!bprm->argc) 1607 return 0; 1608 1609 do { 1610 offset = bprm->p & ~PAGE_MASK; 1611 page = get_arg_page(bprm, bprm->p, 0); 1612 if (!page) { 1613 ret = -EFAULT; 1614 goto out; 1615 } 1616 kaddr = kmap_atomic(page); 1617 1618 for (; offset < PAGE_SIZE && kaddr[offset]; 1619 offset++, bprm->p++) 1620 ; 1621 1622 kunmap_atomic(kaddr); 1623 put_arg_page(page); 1624 } while (offset == PAGE_SIZE); 1625 1626 bprm->p++; 1627 bprm->argc--; 1628 ret = 0; 1629 1630 out: 1631 return ret; 1632 } 1633 EXPORT_SYMBOL(remove_arg_zero); 1634 1635 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e)) 1636 /* 1637 * cycle the list of binary formats handler, until one recognizes the image 1638 */ 1639 int search_binary_handler(struct linux_binprm *bprm) 1640 { 1641 bool need_retry = IS_ENABLED(CONFIG_MODULES); 1642 struct linux_binfmt *fmt; 1643 int retval; 1644 1645 /* This allows 4 levels of binfmt rewrites before failing hard. */ 1646 if (bprm->recursion_depth > 5) 1647 return -ELOOP; 1648 1649 retval = security_bprm_check(bprm); 1650 if (retval) 1651 return retval; 1652 1653 retval = -ENOENT; 1654 retry: 1655 read_lock(&binfmt_lock); 1656 list_for_each_entry(fmt, &formats, lh) { 1657 if (!try_module_get(fmt->module)) 1658 continue; 1659 read_unlock(&binfmt_lock); 1660 1661 bprm->recursion_depth++; 1662 retval = fmt->load_binary(bprm); 1663 bprm->recursion_depth--; 1664 1665 read_lock(&binfmt_lock); 1666 put_binfmt(fmt); 1667 if (retval < 0 && !bprm->mm) { 1668 /* we got to flush_old_exec() and failed after it */ 1669 read_unlock(&binfmt_lock); 1670 force_sigsegv(SIGSEGV); 1671 return retval; 1672 } 1673 if (retval != -ENOEXEC || !bprm->file) { 1674 read_unlock(&binfmt_lock); 1675 return retval; 1676 } 1677 } 1678 read_unlock(&binfmt_lock); 1679 1680 if (need_retry) { 1681 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) && 1682 printable(bprm->buf[2]) && printable(bprm->buf[3])) 1683 return retval; 1684 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0) 1685 return retval; 1686 need_retry = false; 1687 goto retry; 1688 } 1689 1690 return retval; 1691 } 1692 EXPORT_SYMBOL(search_binary_handler); 1693 1694 static int exec_binprm(struct linux_binprm *bprm) 1695 { 1696 pid_t old_pid, old_vpid; 1697 int ret; 1698 1699 /* Need to fetch pid before load_binary changes it */ 1700 old_pid = current->pid; 1701 rcu_read_lock(); 1702 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent)); 1703 rcu_read_unlock(); 1704 1705 ret = search_binary_handler(bprm); 1706 if (ret >= 0) { 1707 audit_bprm(bprm); 1708 trace_sched_process_exec(current, old_pid, bprm); 1709 ptrace_event(PTRACE_EVENT_EXEC, old_vpid); 1710 proc_exec_connector(current); 1711 } 1712 1713 return ret; 1714 } 1715 1716 /* 1717 * sys_execve() executes a new program. 1718 */ 1719 static int __do_execve_file(int fd, struct filename *filename, 1720 struct user_arg_ptr argv, 1721 struct user_arg_ptr envp, 1722 int flags, struct file *file) 1723 { 1724 char *pathbuf = NULL; 1725 struct linux_binprm *bprm; 1726 struct files_struct *displaced; 1727 int retval; 1728 1729 if (IS_ERR(filename)) 1730 return PTR_ERR(filename); 1731 1732 /* 1733 * We move the actual failure in case of RLIMIT_NPROC excess from 1734 * set*uid() to execve() because too many poorly written programs 1735 * don't check setuid() return code. Here we additionally recheck 1736 * whether NPROC limit is still exceeded. 1737 */ 1738 if ((current->flags & PF_NPROC_EXCEEDED) && 1739 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) { 1740 retval = -EAGAIN; 1741 goto out_ret; 1742 } 1743 1744 /* We're below the limit (still or again), so we don't want to make 1745 * further execve() calls fail. */ 1746 current->flags &= ~PF_NPROC_EXCEEDED; 1747 1748 retval = unshare_files(&displaced); 1749 if (retval) 1750 goto out_ret; 1751 1752 retval = -ENOMEM; 1753 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL); 1754 if (!bprm) 1755 goto out_files; 1756 1757 retval = prepare_bprm_creds(bprm); 1758 if (retval) 1759 goto out_free; 1760 1761 check_unsafe_exec(bprm); 1762 current->in_execve = 1; 1763 1764 if (!file) 1765 file = do_open_execat(fd, filename, flags); 1766 retval = PTR_ERR(file); 1767 if (IS_ERR(file)) 1768 goto out_unmark; 1769 1770 sched_exec(); 1771 1772 bprm->file = file; 1773 if (!filename) { 1774 bprm->filename = "none"; 1775 } else if (fd == AT_FDCWD || filename->name[0] == '/') { 1776 bprm->filename = filename->name; 1777 } else { 1778 if (filename->name[0] == '\0') 1779 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd); 1780 else 1781 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s", 1782 fd, filename->name); 1783 if (!pathbuf) { 1784 retval = -ENOMEM; 1785 goto out_unmark; 1786 } 1787 /* 1788 * Record that a name derived from an O_CLOEXEC fd will be 1789 * inaccessible after exec. Relies on having exclusive access to 1790 * current->files (due to unshare_files above). 1791 */ 1792 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt))) 1793 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE; 1794 bprm->filename = pathbuf; 1795 } 1796 bprm->interp = bprm->filename; 1797 1798 retval = bprm_mm_init(bprm); 1799 if (retval) 1800 goto out_unmark; 1801 1802 retval = prepare_arg_pages(bprm, argv, envp); 1803 if (retval < 0) 1804 goto out; 1805 1806 retval = prepare_binprm(bprm); 1807 if (retval < 0) 1808 goto out; 1809 1810 retval = copy_strings_kernel(1, &bprm->filename, bprm); 1811 if (retval < 0) 1812 goto out; 1813 1814 bprm->exec = bprm->p; 1815 retval = copy_strings(bprm->envc, envp, bprm); 1816 if (retval < 0) 1817 goto out; 1818 1819 retval = copy_strings(bprm->argc, argv, bprm); 1820 if (retval < 0) 1821 goto out; 1822 1823 would_dump(bprm, bprm->file); 1824 1825 retval = exec_binprm(bprm); 1826 if (retval < 0) 1827 goto out; 1828 1829 /* execve succeeded */ 1830 current->fs->in_exec = 0; 1831 current->in_execve = 0; 1832 rseq_execve(current); 1833 acct_update_integrals(current); 1834 task_numa_free(current, false); 1835 free_bprm(bprm); 1836 kfree(pathbuf); 1837 if (filename) 1838 putname(filename); 1839 if (displaced) 1840 put_files_struct(displaced); 1841 return retval; 1842 1843 out: 1844 if (bprm->mm) { 1845 acct_arg_size(bprm, 0); 1846 mmput(bprm->mm); 1847 } 1848 1849 out_unmark: 1850 current->fs->in_exec = 0; 1851 current->in_execve = 0; 1852 1853 out_free: 1854 free_bprm(bprm); 1855 kfree(pathbuf); 1856 1857 out_files: 1858 if (displaced) 1859 reset_files_struct(displaced); 1860 out_ret: 1861 if (filename) 1862 putname(filename); 1863 return retval; 1864 } 1865 1866 static int do_execveat_common(int fd, struct filename *filename, 1867 struct user_arg_ptr argv, 1868 struct user_arg_ptr envp, 1869 int flags) 1870 { 1871 return __do_execve_file(fd, filename, argv, envp, flags, NULL); 1872 } 1873 1874 int do_execve_file(struct file *file, void *__argv, void *__envp) 1875 { 1876 struct user_arg_ptr argv = { .ptr.native = __argv }; 1877 struct user_arg_ptr envp = { .ptr.native = __envp }; 1878 1879 return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file); 1880 } 1881 1882 int do_execve(struct filename *filename, 1883 const char __user *const __user *__argv, 1884 const char __user *const __user *__envp) 1885 { 1886 struct user_arg_ptr argv = { .ptr.native = __argv }; 1887 struct user_arg_ptr envp = { .ptr.native = __envp }; 1888 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0); 1889 } 1890 1891 int do_execveat(int fd, struct filename *filename, 1892 const char __user *const __user *__argv, 1893 const char __user *const __user *__envp, 1894 int flags) 1895 { 1896 struct user_arg_ptr argv = { .ptr.native = __argv }; 1897 struct user_arg_ptr envp = { .ptr.native = __envp }; 1898 1899 return do_execveat_common(fd, filename, argv, envp, flags); 1900 } 1901 1902 #ifdef CONFIG_COMPAT 1903 static int compat_do_execve(struct filename *filename, 1904 const compat_uptr_t __user *__argv, 1905 const compat_uptr_t __user *__envp) 1906 { 1907 struct user_arg_ptr argv = { 1908 .is_compat = true, 1909 .ptr.compat = __argv, 1910 }; 1911 struct user_arg_ptr envp = { 1912 .is_compat = true, 1913 .ptr.compat = __envp, 1914 }; 1915 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0); 1916 } 1917 1918 static int compat_do_execveat(int fd, struct filename *filename, 1919 const compat_uptr_t __user *__argv, 1920 const compat_uptr_t __user *__envp, 1921 int flags) 1922 { 1923 struct user_arg_ptr argv = { 1924 .is_compat = true, 1925 .ptr.compat = __argv, 1926 }; 1927 struct user_arg_ptr envp = { 1928 .is_compat = true, 1929 .ptr.compat = __envp, 1930 }; 1931 return do_execveat_common(fd, filename, argv, envp, flags); 1932 } 1933 #endif 1934 1935 void set_binfmt(struct linux_binfmt *new) 1936 { 1937 struct mm_struct *mm = current->mm; 1938 1939 if (mm->binfmt) 1940 module_put(mm->binfmt->module); 1941 1942 mm->binfmt = new; 1943 if (new) 1944 __module_get(new->module); 1945 } 1946 EXPORT_SYMBOL(set_binfmt); 1947 1948 /* 1949 * set_dumpable stores three-value SUID_DUMP_* into mm->flags. 1950 */ 1951 void set_dumpable(struct mm_struct *mm, int value) 1952 { 1953 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT)) 1954 return; 1955 1956 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value); 1957 } 1958 1959 SYSCALL_DEFINE3(execve, 1960 const char __user *, filename, 1961 const char __user *const __user *, argv, 1962 const char __user *const __user *, envp) 1963 { 1964 return do_execve(getname(filename), argv, envp); 1965 } 1966 1967 SYSCALL_DEFINE5(execveat, 1968 int, fd, const char __user *, filename, 1969 const char __user *const __user *, argv, 1970 const char __user *const __user *, envp, 1971 int, flags) 1972 { 1973 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 1974 1975 return do_execveat(fd, 1976 getname_flags(filename, lookup_flags, NULL), 1977 argv, envp, flags); 1978 } 1979 1980 #ifdef CONFIG_COMPAT 1981 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename, 1982 const compat_uptr_t __user *, argv, 1983 const compat_uptr_t __user *, envp) 1984 { 1985 return compat_do_execve(getname(filename), argv, envp); 1986 } 1987 1988 COMPAT_SYSCALL_DEFINE5(execveat, int, fd, 1989 const char __user *, filename, 1990 const compat_uptr_t __user *, argv, 1991 const compat_uptr_t __user *, envp, 1992 int, flags) 1993 { 1994 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 1995 1996 return compat_do_execveat(fd, 1997 getname_flags(filename, lookup_flags, NULL), 1998 argv, envp, flags); 1999 } 2000 #endif 2001