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 #include <linux/io_uring.h>
66
67 #include <linux/uaccess.h>
68 #include <asm/mmu_context.h>
69 #include <asm/tlb.h>
70
71 #include <trace/events/task.h>
72 #include "internal.h"
73
74 #include <trace/events/sched.h>
75
76 static int bprm_creds_from_file(struct linux_binprm *bprm);
77
78 int suid_dumpable = 0;
79
80 static LIST_HEAD(formats);
81 static DEFINE_RWLOCK(binfmt_lock);
82
83 void __register_binfmt(struct linux_binfmt * fmt, int insert)
84 {
85 BUG_ON(!fmt);
86 if (WARN_ON(!fmt->load_binary))
87 return;
88 write_lock(&binfmt_lock);
89 insert ? list_add(&fmt->lh, &formats) :
90 list_add_tail(&fmt->lh, &formats);
91 write_unlock(&binfmt_lock);
92 }
93
94 EXPORT_SYMBOL(__register_binfmt);
95
96 void unregister_binfmt(struct linux_binfmt * fmt)
97 {
98 write_lock(&binfmt_lock);
99 list_del(&fmt->lh);
100 write_unlock(&binfmt_lock);
101 }
102
103 EXPORT_SYMBOL(unregister_binfmt);
104
105 static inline void put_binfmt(struct linux_binfmt * fmt)
106 {
107 module_put(fmt->module);
108 }
109
110 bool path_noexec(const struct path *path)
111 {
112 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
113 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
114 }
115
116 #ifdef CONFIG_USELIB
117 /*
118 * Note that a shared library must be both readable and executable due to
119 * security reasons.
120 *
121 * Also note that we take the address to load from from the file itself.
122 */
123 SYSCALL_DEFINE1(uselib, const char __user *, library)
124 {
125 struct linux_binfmt *fmt;
126 struct file *file;
127 struct filename *tmp = getname(library);
128 int error = PTR_ERR(tmp);
129 static const struct open_flags uselib_flags = {
130 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
131 .acc_mode = MAY_READ | MAY_EXEC,
132 .intent = LOOKUP_OPEN,
133 .lookup_flags = LOOKUP_FOLLOW,
134 };
135
136 if (IS_ERR(tmp))
137 goto out;
138
139 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
140 putname(tmp);
141 error = PTR_ERR(file);
142 if (IS_ERR(file))
143 goto out;
144
145 /*
146 * may_open() has already checked for this, so it should be
147 * impossible to trip now. But we need to be extra cautious
148 * and check again at the very end too.
149 */
150 error = -EACCES;
151 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
152 path_noexec(&file->f_path)))
153 goto exit;
154
155 fsnotify_open(file);
156
157 error = -ENOEXEC;
158
159 read_lock(&binfmt_lock);
160 list_for_each_entry(fmt, &formats, lh) {
161 if (!fmt->load_shlib)
162 continue;
163 if (!try_module_get(fmt->module))
164 continue;
165 read_unlock(&binfmt_lock);
166 error = fmt->load_shlib(file);
167 read_lock(&binfmt_lock);
168 put_binfmt(fmt);
169 if (error != -ENOEXEC)
170 break;
171 }
172 read_unlock(&binfmt_lock);
173 exit:
174 fput(file);
175 out:
176 return error;
177 }
178 #endif /* #ifdef CONFIG_USELIB */
179
180 #ifdef CONFIG_MMU
181 /*
182 * The nascent bprm->mm is not visible until exec_mmap() but it can
183 * use a lot of memory, account these pages in current->mm temporary
184 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
185 * change the counter back via acct_arg_size(0).
186 */
187 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
188 {
189 struct mm_struct *mm = current->mm;
190 long diff = (long)(pages - bprm->vma_pages);
191
192 if (!mm || !diff)
193 return;
194
195 bprm->vma_pages = pages;
196 add_mm_counter(mm, MM_ANONPAGES, diff);
197 }
198
199 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
200 int write)
201 {
202 struct page *page;
203 int ret;
204 unsigned int gup_flags = FOLL_FORCE;
205
206 #ifdef CONFIG_STACK_GROWSUP
207 if (write) {
208 ret = expand_downwards(bprm->vma, pos);
209 if (ret < 0)
210 return NULL;
211 }
212 #endif
213
214 if (write)
215 gup_flags |= FOLL_WRITE;
216
217 /*
218 * We are doing an exec(). 'current' is the process
219 * doing the exec and bprm->mm is the new process's mm.
220 */
221 ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
222 &page, NULL, NULL);
223 if (ret <= 0)
224 return NULL;
225
226 if (write)
227 acct_arg_size(bprm, vma_pages(bprm->vma));
228
229 return page;
230 }
231
232 static void put_arg_page(struct page *page)
233 {
234 put_page(page);
235 }
236
237 static void free_arg_pages(struct linux_binprm *bprm)
238 {
239 }
240
241 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
242 struct page *page)
243 {
244 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
245 }
246
247 static int __bprm_mm_init(struct linux_binprm *bprm)
248 {
249 int err;
250 struct vm_area_struct *vma = NULL;
251 struct mm_struct *mm = bprm->mm;
252
253 bprm->vma = vma = vm_area_alloc(mm);
254 if (!vma)
255 return -ENOMEM;
256 vma_set_anonymous(vma);
257
258 if (mmap_write_lock_killable(mm)) {
259 err = -EINTR;
260 goto err_free;
261 }
262
263 /*
264 * Place the stack at the largest stack address the architecture
265 * supports. Later, we'll move this to an appropriate place. We don't
266 * use STACK_TOP because that can depend on attributes which aren't
267 * configured yet.
268 */
269 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
270 vma->vm_end = STACK_TOP_MAX;
271 vma->vm_start = vma->vm_end - PAGE_SIZE;
272 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
273 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
274
275 err = insert_vm_struct(mm, vma);
276 if (err)
277 goto err;
278
279 mm->stack_vm = mm->total_vm = 1;
280 mmap_write_unlock(mm);
281 bprm->p = vma->vm_end - sizeof(void *);
282 return 0;
283 err:
284 mmap_write_unlock(mm);
285 err_free:
286 bprm->vma = NULL;
287 vm_area_free(vma);
288 return err;
289 }
290
291 static bool valid_arg_len(struct linux_binprm *bprm, long len)
292 {
293 return len <= MAX_ARG_STRLEN;
294 }
295
296 #else
297
298 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
299 {
300 }
301
302 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
303 int write)
304 {
305 struct page *page;
306
307 page = bprm->page[pos / PAGE_SIZE];
308 if (!page && write) {
309 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
310 if (!page)
311 return NULL;
312 bprm->page[pos / PAGE_SIZE] = page;
313 }
314
315 return page;
316 }
317
318 static void put_arg_page(struct page *page)
319 {
320 }
321
322 static void free_arg_page(struct linux_binprm *bprm, int i)
323 {
324 if (bprm->page[i]) {
325 __free_page(bprm->page[i]);
326 bprm->page[i] = NULL;
327 }
328 }
329
330 static void free_arg_pages(struct linux_binprm *bprm)
331 {
332 int i;
333
334 for (i = 0; i < MAX_ARG_PAGES; i++)
335 free_arg_page(bprm, i);
336 }
337
338 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
339 struct page *page)
340 {
341 }
342
343 static int __bprm_mm_init(struct linux_binprm *bprm)
344 {
345 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
346 return 0;
347 }
348
349 static bool valid_arg_len(struct linux_binprm *bprm, long len)
350 {
351 return len <= bprm->p;
352 }
353
354 #endif /* CONFIG_MMU */
355
356 /*
357 * Create a new mm_struct and populate it with a temporary stack
358 * vm_area_struct. We don't have enough context at this point to set the stack
359 * flags, permissions, and offset, so we use temporary values. We'll update
360 * them later in setup_arg_pages().
361 */
362 static int bprm_mm_init(struct linux_binprm *bprm)
363 {
364 int err;
365 struct mm_struct *mm = NULL;
366
367 bprm->mm = mm = mm_alloc();
368 err = -ENOMEM;
369 if (!mm)
370 goto err;
371
372 /* Save current stack limit for all calculations made during exec. */
373 task_lock(current->group_leader);
374 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
375 task_unlock(current->group_leader);
376
377 err = __bprm_mm_init(bprm);
378 if (err)
379 goto err;
380
381 return 0;
382
383 err:
384 if (mm) {
385 bprm->mm = NULL;
386 mmdrop(mm);
387 }
388
389 return err;
390 }
391
392 struct user_arg_ptr {
393 #ifdef CONFIG_COMPAT
394 bool is_compat;
395 #endif
396 union {
397 const char __user *const __user *native;
398 #ifdef CONFIG_COMPAT
399 const compat_uptr_t __user *compat;
400 #endif
401 } ptr;
402 };
403
404 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
405 {
406 const char __user *native;
407
408 #ifdef CONFIG_COMPAT
409 if (unlikely(argv.is_compat)) {
410 compat_uptr_t compat;
411
412 if (get_user(compat, argv.ptr.compat + nr))
413 return ERR_PTR(-EFAULT);
414
415 return compat_ptr(compat);
416 }
417 #endif
418
419 if (get_user(native, argv.ptr.native + nr))
420 return ERR_PTR(-EFAULT);
421
422 return native;
423 }
424
425 /*
426 * count() counts the number of strings in array ARGV.
427 */
428 static int count(struct user_arg_ptr argv, int max)
429 {
430 int i = 0;
431
432 if (argv.ptr.native != NULL) {
433 for (;;) {
434 const char __user *p = get_user_arg_ptr(argv, i);
435
436 if (!p)
437 break;
438
439 if (IS_ERR(p))
440 return -EFAULT;
441
442 if (i >= max)
443 return -E2BIG;
444 ++i;
445
446 if (fatal_signal_pending(current))
447 return -ERESTARTNOHAND;
448 cond_resched();
449 }
450 }
451 return i;
452 }
453
454 static int count_strings_kernel(const char *const *argv)
455 {
456 int i;
457
458 if (!argv)
459 return 0;
460
461 for (i = 0; argv[i]; ++i) {
462 if (i >= MAX_ARG_STRINGS)
463 return -E2BIG;
464 if (fatal_signal_pending(current))
465 return -ERESTARTNOHAND;
466 cond_resched();
467 }
468 return i;
469 }
470
471 static int bprm_stack_limits(struct linux_binprm *bprm)
472 {
473 unsigned long limit, ptr_size;
474
475 /*
476 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
477 * (whichever is smaller) for the argv+env strings.
478 * This ensures that:
479 * - the remaining binfmt code will not run out of stack space,
480 * - the program will have a reasonable amount of stack left
481 * to work from.
482 */
483 limit = _STK_LIM / 4 * 3;
484 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
485 /*
486 * We've historically supported up to 32 pages (ARG_MAX)
487 * of argument strings even with small stacks
488 */
489 limit = max_t(unsigned long, limit, ARG_MAX);
490 /*
491 * We must account for the size of all the argv and envp pointers to
492 * the argv and envp strings, since they will also take up space in
493 * the stack. They aren't stored until much later when we can't
494 * signal to the parent that the child has run out of stack space.
495 * Instead, calculate it here so it's possible to fail gracefully.
496 */
497 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
498 if (limit <= ptr_size)
499 return -E2BIG;
500 limit -= ptr_size;
501
502 bprm->argmin = bprm->p - limit;
503 return 0;
504 }
505
506 /*
507 * 'copy_strings()' copies argument/environment strings from the old
508 * processes's memory to the new process's stack. The call to get_user_pages()
509 * ensures the destination page is created and not swapped out.
510 */
511 static int copy_strings(int argc, struct user_arg_ptr argv,
512 struct linux_binprm *bprm)
513 {
514 struct page *kmapped_page = NULL;
515 char *kaddr = NULL;
516 unsigned long kpos = 0;
517 int ret;
518
519 while (argc-- > 0) {
520 const char __user *str;
521 int len;
522 unsigned long pos;
523
524 ret = -EFAULT;
525 str = get_user_arg_ptr(argv, argc);
526 if (IS_ERR(str))
527 goto out;
528
529 len = strnlen_user(str, MAX_ARG_STRLEN);
530 if (!len)
531 goto out;
532
533 ret = -E2BIG;
534 if (!valid_arg_len(bprm, len))
535 goto out;
536
537 /* We're going to work our way backwords. */
538 pos = bprm->p;
539 str += len;
540 bprm->p -= len;
541 #ifdef CONFIG_MMU
542 if (bprm->p < bprm->argmin)
543 goto out;
544 #endif
545
546 while (len > 0) {
547 int offset, bytes_to_copy;
548
549 if (fatal_signal_pending(current)) {
550 ret = -ERESTARTNOHAND;
551 goto out;
552 }
553 cond_resched();
554
555 offset = pos % PAGE_SIZE;
556 if (offset == 0)
557 offset = PAGE_SIZE;
558
559 bytes_to_copy = offset;
560 if (bytes_to_copy > len)
561 bytes_to_copy = len;
562
563 offset -= bytes_to_copy;
564 pos -= bytes_to_copy;
565 str -= bytes_to_copy;
566 len -= bytes_to_copy;
567
568 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
569 struct page *page;
570
571 page = get_arg_page(bprm, pos, 1);
572 if (!page) {
573 ret = -E2BIG;
574 goto out;
575 }
576
577 if (kmapped_page) {
578 flush_kernel_dcache_page(kmapped_page);
579 kunmap(kmapped_page);
580 put_arg_page(kmapped_page);
581 }
582 kmapped_page = page;
583 kaddr = kmap(kmapped_page);
584 kpos = pos & PAGE_MASK;
585 flush_arg_page(bprm, kpos, kmapped_page);
586 }
587 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
588 ret = -EFAULT;
589 goto out;
590 }
591 }
592 }
593 ret = 0;
594 out:
595 if (kmapped_page) {
596 flush_kernel_dcache_page(kmapped_page);
597 kunmap(kmapped_page);
598 put_arg_page(kmapped_page);
599 }
600 return ret;
601 }
602
603 /*
604 * Copy and argument/environment string from the kernel to the processes stack.
605 */
606 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
607 {
608 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
609 unsigned long pos = bprm->p;
610
611 if (len == 0)
612 return -EFAULT;
613 if (!valid_arg_len(bprm, len))
614 return -E2BIG;
615
616 /* We're going to work our way backwards. */
617 arg += len;
618 bprm->p -= len;
619 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
620 return -E2BIG;
621
622 while (len > 0) {
623 unsigned int bytes_to_copy = min_t(unsigned int, len,
624 min_not_zero(offset_in_page(pos), PAGE_SIZE));
625 struct page *page;
626 char *kaddr;
627
628 pos -= bytes_to_copy;
629 arg -= bytes_to_copy;
630 len -= bytes_to_copy;
631
632 page = get_arg_page(bprm, pos, 1);
633 if (!page)
634 return -E2BIG;
635 kaddr = kmap_atomic(page);
636 flush_arg_page(bprm, pos & PAGE_MASK, page);
637 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
638 flush_kernel_dcache_page(page);
639 kunmap_atomic(kaddr);
640 put_arg_page(page);
641 }
642
643 return 0;
644 }
645 EXPORT_SYMBOL(copy_string_kernel);
646
647 static int copy_strings_kernel(int argc, const char *const *argv,
648 struct linux_binprm *bprm)
649 {
650 while (argc-- > 0) {
651 int ret = copy_string_kernel(argv[argc], bprm);
652 if (ret < 0)
653 return ret;
654 if (fatal_signal_pending(current))
655 return -ERESTARTNOHAND;
656 cond_resched();
657 }
658 return 0;
659 }
660
661 #ifdef CONFIG_MMU
662
663 /*
664 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
665 * the binfmt code determines where the new stack should reside, we shift it to
666 * its final location. The process proceeds as follows:
667 *
668 * 1) Use shift to calculate the new vma endpoints.
669 * 2) Extend vma to cover both the old and new ranges. This ensures the
670 * arguments passed to subsequent functions are consistent.
671 * 3) Move vma's page tables to the new range.
672 * 4) Free up any cleared pgd range.
673 * 5) Shrink the vma to cover only the new range.
674 */
675 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
676 {
677 struct mm_struct *mm = vma->vm_mm;
678 unsigned long old_start = vma->vm_start;
679 unsigned long old_end = vma->vm_end;
680 unsigned long length = old_end - old_start;
681 unsigned long new_start = old_start - shift;
682 unsigned long new_end = old_end - shift;
683 struct mmu_gather tlb;
684
685 BUG_ON(new_start > new_end);
686
687 /*
688 * ensure there are no vmas between where we want to go
689 * and where we are
690 */
691 if (vma != find_vma(mm, new_start))
692 return -EFAULT;
693
694 /*
695 * cover the whole range: [new_start, old_end)
696 */
697 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
698 return -ENOMEM;
699
700 /*
701 * move the page tables downwards, on failure we rely on
702 * process cleanup to remove whatever mess we made.
703 */
704 if (length != move_page_tables(vma, old_start,
705 vma, new_start, length, false))
706 return -ENOMEM;
707
708 lru_add_drain();
709 tlb_gather_mmu(&tlb, mm, old_start, old_end);
710 if (new_end > old_start) {
711 /*
712 * when the old and new regions overlap clear from new_end.
713 */
714 free_pgd_range(&tlb, new_end, old_end, new_end,
715 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
716 } else {
717 /*
718 * otherwise, clean from old_start; this is done to not touch
719 * the address space in [new_end, old_start) some architectures
720 * have constraints on va-space that make this illegal (IA64) -
721 * for the others its just a little faster.
722 */
723 free_pgd_range(&tlb, old_start, old_end, new_end,
724 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
725 }
726 tlb_finish_mmu(&tlb, old_start, old_end);
727
728 /*
729 * Shrink the vma to just the new range. Always succeeds.
730 */
731 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
732
733 return 0;
734 }
735
736 /*
737 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
738 * the stack is optionally relocated, and some extra space is added.
739 */
740 int setup_arg_pages(struct linux_binprm *bprm,
741 unsigned long stack_top,
742 int executable_stack)
743 {
744 unsigned long ret;
745 unsigned long stack_shift;
746 struct mm_struct *mm = current->mm;
747 struct vm_area_struct *vma = bprm->vma;
748 struct vm_area_struct *prev = NULL;
749 unsigned long vm_flags;
750 unsigned long stack_base;
751 unsigned long stack_size;
752 unsigned long stack_expand;
753 unsigned long rlim_stack;
754
755 #ifdef CONFIG_STACK_GROWSUP
756 /* Limit stack size */
757 stack_base = bprm->rlim_stack.rlim_max;
758 if (stack_base > STACK_SIZE_MAX)
759 stack_base = STACK_SIZE_MAX;
760
761 /* Add space for stack randomization. */
762 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
763
764 /* Make sure we didn't let the argument array grow too large. */
765 if (vma->vm_end - vma->vm_start > stack_base)
766 return -ENOMEM;
767
768 stack_base = PAGE_ALIGN(stack_top - stack_base);
769
770 stack_shift = vma->vm_start - stack_base;
771 mm->arg_start = bprm->p - stack_shift;
772 bprm->p = vma->vm_end - stack_shift;
773 #else
774 stack_top = arch_align_stack(stack_top);
775 stack_top = PAGE_ALIGN(stack_top);
776
777 if (unlikely(stack_top < mmap_min_addr) ||
778 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
779 return -ENOMEM;
780
781 stack_shift = vma->vm_end - stack_top;
782
783 bprm->p -= stack_shift;
784 mm->arg_start = bprm->p;
785 #endif
786
787 if (bprm->loader)
788 bprm->loader -= stack_shift;
789 bprm->exec -= stack_shift;
790
791 if (mmap_write_lock_killable(mm))
792 return -EINTR;
793
794 vm_flags = VM_STACK_FLAGS;
795
796 /*
797 * Adjust stack execute permissions; explicitly enable for
798 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
799 * (arch default) otherwise.
800 */
801 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
802 vm_flags |= VM_EXEC;
803 else if (executable_stack == EXSTACK_DISABLE_X)
804 vm_flags &= ~VM_EXEC;
805 vm_flags |= mm->def_flags;
806 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
807
808 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
809 vm_flags);
810 if (ret)
811 goto out_unlock;
812 BUG_ON(prev != vma);
813
814 if (unlikely(vm_flags & VM_EXEC)) {
815 pr_warn_once("process '%pD4' started with executable stack\n",
816 bprm->file);
817 }
818
819 /* Move stack pages down in memory. */
820 if (stack_shift) {
821 ret = shift_arg_pages(vma, stack_shift);
822 if (ret)
823 goto out_unlock;
824 }
825
826 /* mprotect_fixup is overkill to remove the temporary stack flags */
827 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
828
829 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
830 stack_size = vma->vm_end - vma->vm_start;
831 /*
832 * Align this down to a page boundary as expand_stack
833 * will align it up.
834 */
835 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
836 #ifdef CONFIG_STACK_GROWSUP
837 if (stack_size + stack_expand > rlim_stack)
838 stack_base = vma->vm_start + rlim_stack;
839 else
840 stack_base = vma->vm_end + stack_expand;
841 #else
842 if (stack_size + stack_expand > rlim_stack)
843 stack_base = vma->vm_end - rlim_stack;
844 else
845 stack_base = vma->vm_start - stack_expand;
846 #endif
847 current->mm->start_stack = bprm->p;
848 ret = expand_stack(vma, stack_base);
849 if (ret)
850 ret = -EFAULT;
851
852 out_unlock:
853 mmap_write_unlock(mm);
854 return ret;
855 }
856 EXPORT_SYMBOL(setup_arg_pages);
857
858 #else
859
860 /*
861 * Transfer the program arguments and environment from the holding pages
862 * onto the stack. The provided stack pointer is adjusted accordingly.
863 */
864 int transfer_args_to_stack(struct linux_binprm *bprm,
865 unsigned long *sp_location)
866 {
867 unsigned long index, stop, sp;
868 int ret = 0;
869
870 stop = bprm->p >> PAGE_SHIFT;
871 sp = *sp_location;
872
873 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
874 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
875 char *src = kmap(bprm->page[index]) + offset;
876 sp -= PAGE_SIZE - offset;
877 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
878 ret = -EFAULT;
879 kunmap(bprm->page[index]);
880 if (ret)
881 goto out;
882 }
883
884 *sp_location = sp;
885
886 out:
887 return ret;
888 }
889 EXPORT_SYMBOL(transfer_args_to_stack);
890
891 #endif /* CONFIG_MMU */
892
893 static struct file *do_open_execat(int fd, struct filename *name, int flags)
894 {
895 struct file *file;
896 int err;
897 struct open_flags open_exec_flags = {
898 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
899 .acc_mode = MAY_EXEC,
900 .intent = LOOKUP_OPEN,
901 .lookup_flags = LOOKUP_FOLLOW,
902 };
903
904 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
905 return ERR_PTR(-EINVAL);
906 if (flags & AT_SYMLINK_NOFOLLOW)
907 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
908 if (flags & AT_EMPTY_PATH)
909 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
910
911 file = do_filp_open(fd, name, &open_exec_flags);
912 if (IS_ERR(file))
913 goto out;
914
915 /*
916 * may_open() has already checked for this, so it should be
917 * impossible to trip now. But we need to be extra cautious
918 * and check again at the very end too.
919 */
920 err = -EACCES;
921 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
922 path_noexec(&file->f_path)))
923 goto exit;
924
925 err = deny_write_access(file);
926 if (err)
927 goto exit;
928
929 if (name->name[0] != '\0')
930 fsnotify_open(file);
931
932 out:
933 return file;
934
935 exit:
936 fput(file);
937 return ERR_PTR(err);
938 }
939
940 struct file *open_exec(const char *name)
941 {
942 struct filename *filename = getname_kernel(name);
943 struct file *f = ERR_CAST(filename);
944
945 if (!IS_ERR(filename)) {
946 f = do_open_execat(AT_FDCWD, filename, 0);
947 putname(filename);
948 }
949 return f;
950 }
951 EXPORT_SYMBOL(open_exec);
952
953 int kernel_read_file(struct file *file, void **buf, loff_t *size,
954 loff_t max_size, enum kernel_read_file_id id)
955 {
956 loff_t i_size, pos;
957 ssize_t bytes = 0;
958 void *allocated = NULL;
959 int ret;
960
961 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
962 return -EINVAL;
963
964 ret = deny_write_access(file);
965 if (ret)
966 return ret;
967
968 ret = security_kernel_read_file(file, id);
969 if (ret)
970 goto out;
971
972 i_size = i_size_read(file_inode(file));
973 if (i_size <= 0) {
974 ret = -EINVAL;
975 goto out;
976 }
977 if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
978 ret = -EFBIG;
979 goto out;
980 }
981
982 if (!*buf)
983 *buf = allocated = vmalloc(i_size);
984 if (!*buf) {
985 ret = -ENOMEM;
986 goto out;
987 }
988
989 pos = 0;
990 while (pos < i_size) {
991 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
992 if (bytes < 0) {
993 ret = bytes;
994 goto out_free;
995 }
996
997 if (bytes == 0)
998 break;
999 }
1000
1001 if (pos != i_size) {
1002 ret = -EIO;
1003 goto out_free;
1004 }
1005
1006 ret = security_kernel_post_read_file(file, *buf, i_size, id);
1007 if (!ret)
1008 *size = pos;
1009
1010 out_free:
1011 if (ret < 0) {
1012 if (allocated) {
1013 vfree(*buf);
1014 *buf = NULL;
1015 }
1016 }
1017
1018 out:
1019 allow_write_access(file);
1020 return ret;
1021 }
1022 EXPORT_SYMBOL_GPL(kernel_read_file);
1023
1024 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
1025 loff_t max_size, enum kernel_read_file_id id)
1026 {
1027 struct file *file;
1028 int ret;
1029
1030 if (!path || !*path)
1031 return -EINVAL;
1032
1033 file = filp_open(path, O_RDONLY, 0);
1034 if (IS_ERR(file))
1035 return PTR_ERR(file);
1036
1037 ret = kernel_read_file(file, buf, size, max_size, id);
1038 fput(file);
1039 return ret;
1040 }
1041 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
1042
1043 int kernel_read_file_from_path_initns(const char *path, void **buf,
1044 loff_t *size, loff_t max_size,
1045 enum kernel_read_file_id id)
1046 {
1047 struct file *file;
1048 struct path root;
1049 int ret;
1050
1051 if (!path || !*path)
1052 return -EINVAL;
1053
1054 task_lock(&init_task);
1055 get_fs_root(init_task.fs, &root);
1056 task_unlock(&init_task);
1057
1058 file = file_open_root(root.dentry, root.mnt, path, O_RDONLY, 0);
1059 path_put(&root);
1060 if (IS_ERR(file))
1061 return PTR_ERR(file);
1062
1063 ret = kernel_read_file(file, buf, size, max_size, id);
1064 fput(file);
1065 return ret;
1066 }
1067 EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns);
1068
1069 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
1070 enum kernel_read_file_id id)
1071 {
1072 struct fd f = fdget(fd);
1073 int ret = -EBADF;
1074
1075 if (!f.file)
1076 goto out;
1077
1078 ret = kernel_read_file(f.file, buf, size, max_size, id);
1079 out:
1080 fdput(f);
1081 return ret;
1082 }
1083 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1084
1085 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
1086 defined(CONFIG_BINFMT_ELF_FDPIC)
1087 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1088 {
1089 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1090 if (res > 0)
1091 flush_icache_user_range(addr, addr + len);
1092 return res;
1093 }
1094 EXPORT_SYMBOL(read_code);
1095 #endif
1096
1097 /*
1098 * Maps the mm_struct mm into the current task struct.
1099 * On success, this function returns with the mutex
1100 * exec_update_mutex locked.
1101 */
1102 static int exec_mmap(struct mm_struct *mm)
1103 {
1104 struct task_struct *tsk;
1105 struct mm_struct *old_mm, *active_mm;
1106 int ret;
1107
1108 /* Notify parent that we're no longer interested in the old VM */
1109 tsk = current;
1110 old_mm = current->mm;
1111 exec_mm_release(tsk, old_mm);
1112 if (old_mm)
1113 sync_mm_rss(old_mm);
1114
1115 ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
1116 if (ret)
1117 return ret;
1118
1119 if (old_mm) {
1120 /*
1121 * Make sure that if there is a core dump in progress
1122 * for the old mm, we get out and die instead of going
1123 * through with the exec. We must hold mmap_lock around
1124 * checking core_state and changing tsk->mm.
1125 */
1126 mmap_read_lock(old_mm);
1127 if (unlikely(old_mm->core_state)) {
1128 mmap_read_unlock(old_mm);
1129 mutex_unlock(&tsk->signal->exec_update_mutex);
1130 return -EINTR;
1131 }
1132 }
1133
1134 task_lock(tsk);
1135 membarrier_exec_mmap(mm);
1136
1137 local_irq_disable();
1138 active_mm = tsk->active_mm;
1139 tsk->active_mm = mm;
1140 tsk->mm = mm;
1141 /*
1142 * This prevents preemption while active_mm is being loaded and
1143 * it and mm are being updated, which could cause problems for
1144 * lazy tlb mm refcounting when these are updated by context
1145 * switches. Not all architectures can handle irqs off over
1146 * activate_mm yet.
1147 */
1148 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1149 local_irq_enable();
1150 activate_mm(active_mm, mm);
1151 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1152 local_irq_enable();
1153 tsk->mm->vmacache_seqnum = 0;
1154 vmacache_flush(tsk);
1155 task_unlock(tsk);
1156 if (old_mm) {
1157 mmap_read_unlock(old_mm);
1158 BUG_ON(active_mm != old_mm);
1159 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1160 mm_update_next_owner(old_mm);
1161 mmput(old_mm);
1162 return 0;
1163 }
1164 mmdrop(active_mm);
1165 return 0;
1166 }
1167
1168 static int de_thread(struct task_struct *tsk)
1169 {
1170 struct signal_struct *sig = tsk->signal;
1171 struct sighand_struct *oldsighand = tsk->sighand;
1172 spinlock_t *lock = &oldsighand->siglock;
1173
1174 if (thread_group_empty(tsk))
1175 goto no_thread_group;
1176
1177 /*
1178 * Kill all other threads in the thread group.
1179 */
1180 spin_lock_irq(lock);
1181 if (signal_group_exit(sig)) {
1182 /*
1183 * Another group action in progress, just
1184 * return so that the signal is processed.
1185 */
1186 spin_unlock_irq(lock);
1187 return -EAGAIN;
1188 }
1189
1190 sig->group_exit_task = tsk;
1191 sig->notify_count = zap_other_threads(tsk);
1192 if (!thread_group_leader(tsk))
1193 sig->notify_count--;
1194
1195 while (sig->notify_count) {
1196 __set_current_state(TASK_KILLABLE);
1197 spin_unlock_irq(lock);
1198 schedule();
1199 if (__fatal_signal_pending(tsk))
1200 goto killed;
1201 spin_lock_irq(lock);
1202 }
1203 spin_unlock_irq(lock);
1204
1205 /*
1206 * At this point all other threads have exited, all we have to
1207 * do is to wait for the thread group leader to become inactive,
1208 * and to assume its PID:
1209 */
1210 if (!thread_group_leader(tsk)) {
1211 struct task_struct *leader = tsk->group_leader;
1212
1213 for (;;) {
1214 cgroup_threadgroup_change_begin(tsk);
1215 write_lock_irq(&tasklist_lock);
1216 /*
1217 * Do this under tasklist_lock to ensure that
1218 * exit_notify() can't miss ->group_exit_task
1219 */
1220 sig->notify_count = -1;
1221 if (likely(leader->exit_state))
1222 break;
1223 __set_current_state(TASK_KILLABLE);
1224 write_unlock_irq(&tasklist_lock);
1225 cgroup_threadgroup_change_end(tsk);
1226 schedule();
1227 if (__fatal_signal_pending(tsk))
1228 goto killed;
1229 }
1230
1231 /*
1232 * The only record we have of the real-time age of a
1233 * process, regardless of execs it's done, is start_time.
1234 * All the past CPU time is accumulated in signal_struct
1235 * from sister threads now dead. But in this non-leader
1236 * exec, nothing survives from the original leader thread,
1237 * whose birth marks the true age of this process now.
1238 * When we take on its identity by switching to its PID, we
1239 * also take its birthdate (always earlier than our own).
1240 */
1241 tsk->start_time = leader->start_time;
1242 tsk->start_boottime = leader->start_boottime;
1243
1244 BUG_ON(!same_thread_group(leader, tsk));
1245 /*
1246 * An exec() starts a new thread group with the
1247 * TGID of the previous thread group. Rehash the
1248 * two threads with a switched PID, and release
1249 * the former thread group leader:
1250 */
1251
1252 /* Become a process group leader with the old leader's pid.
1253 * The old leader becomes a thread of the this thread group.
1254 */
1255 exchange_tids(tsk, leader);
1256 transfer_pid(leader, tsk, PIDTYPE_TGID);
1257 transfer_pid(leader, tsk, PIDTYPE_PGID);
1258 transfer_pid(leader, tsk, PIDTYPE_SID);
1259
1260 list_replace_rcu(&leader->tasks, &tsk->tasks);
1261 list_replace_init(&leader->sibling, &tsk->sibling);
1262
1263 tsk->group_leader = tsk;
1264 leader->group_leader = tsk;
1265
1266 tsk->exit_signal = SIGCHLD;
1267 leader->exit_signal = -1;
1268
1269 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1270 leader->exit_state = EXIT_DEAD;
1271
1272 /*
1273 * We are going to release_task()->ptrace_unlink() silently,
1274 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1275 * the tracer wont't block again waiting for this thread.
1276 */
1277 if (unlikely(leader->ptrace))
1278 __wake_up_parent(leader, leader->parent);
1279 write_unlock_irq(&tasklist_lock);
1280 cgroup_threadgroup_change_end(tsk);
1281
1282 release_task(leader);
1283 }
1284
1285 sig->group_exit_task = NULL;
1286 sig->notify_count = 0;
1287
1288 no_thread_group:
1289 /* we have changed execution domain */
1290 tsk->exit_signal = SIGCHLD;
1291
1292 BUG_ON(!thread_group_leader(tsk));
1293 return 0;
1294
1295 killed:
1296 /* protects against exit_notify() and __exit_signal() */
1297 read_lock(&tasklist_lock);
1298 sig->group_exit_task = NULL;
1299 sig->notify_count = 0;
1300 read_unlock(&tasklist_lock);
1301 return -EAGAIN;
1302 }
1303
1304
1305 /*
1306 * This function makes sure the current process has its own signal table,
1307 * so that flush_signal_handlers can later reset the handlers without
1308 * disturbing other processes. (Other processes might share the signal
1309 * table via the CLONE_SIGHAND option to clone().)
1310 */
1311 static int unshare_sighand(struct task_struct *me)
1312 {
1313 struct sighand_struct *oldsighand = me->sighand;
1314
1315 if (refcount_read(&oldsighand->count) != 1) {
1316 struct sighand_struct *newsighand;
1317 /*
1318 * This ->sighand is shared with the CLONE_SIGHAND
1319 * but not CLONE_THREAD task, switch to the new one.
1320 */
1321 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1322 if (!newsighand)
1323 return -ENOMEM;
1324
1325 refcount_set(&newsighand->count, 1);
1326 memcpy(newsighand->action, oldsighand->action,
1327 sizeof(newsighand->action));
1328
1329 write_lock_irq(&tasklist_lock);
1330 spin_lock(&oldsighand->siglock);
1331 rcu_assign_pointer(me->sighand, newsighand);
1332 spin_unlock(&oldsighand->siglock);
1333 write_unlock_irq(&tasklist_lock);
1334
1335 __cleanup_sighand(oldsighand);
1336 }
1337 return 0;
1338 }
1339
1340 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1341 {
1342 task_lock(tsk);
1343 strncpy(buf, tsk->comm, buf_size);
1344 task_unlock(tsk);
1345 return buf;
1346 }
1347 EXPORT_SYMBOL_GPL(__get_task_comm);
1348
1349 /*
1350 * These functions flushes out all traces of the currently running executable
1351 * so that a new one can be started
1352 */
1353
1354 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1355 {
1356 task_lock(tsk);
1357 trace_task_rename(tsk, buf);
1358 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1359 task_unlock(tsk);
1360 perf_event_comm(tsk, exec);
1361 }
1362
1363 /*
1364 * Calling this is the point of no return. None of the failures will be
1365 * seen by userspace since either the process is already taking a fatal
1366 * signal (via de_thread() or coredump), or will have SEGV raised
1367 * (after exec_mmap()) by search_binary_handler (see below).
1368 */
1369 int begin_new_exec(struct linux_binprm * bprm)
1370 {
1371 struct task_struct *me = current;
1372 int retval;
1373
1374 /* Once we are committed compute the creds */
1375 retval = bprm_creds_from_file(bprm);
1376 if (retval)
1377 return retval;
1378
1379 /*
1380 * Ensure all future errors are fatal.
1381 */
1382 bprm->point_of_no_return = true;
1383
1384 /*
1385 * Make this the only thread in the thread group.
1386 */
1387 retval = de_thread(me);
1388 if (retval)
1389 goto out;
1390
1391 /*
1392 * Must be called _before_ exec_mmap() as bprm->mm is
1393 * not visibile until then. This also enables the update
1394 * to be lockless.
1395 */
1396 set_mm_exe_file(bprm->mm, bprm->file);
1397
1398 /* If the binary is not readable then enforce mm->dumpable=0 */
1399 would_dump(bprm, bprm->file);
1400 if (bprm->have_execfd)
1401 would_dump(bprm, bprm->executable);
1402
1403 /*
1404 * Release all of the old mmap stuff
1405 */
1406 acct_arg_size(bprm, 0);
1407 retval = exec_mmap(bprm->mm);
1408 if (retval)
1409 goto out;
1410
1411 bprm->mm = NULL;
1412
1413 #ifdef CONFIG_POSIX_TIMERS
1414 exit_itimers(me->signal);
1415 flush_itimer_signals();
1416 #endif
1417
1418 /*
1419 * Make the signal table private.
1420 */
1421 retval = unshare_sighand(me);
1422 if (retval)
1423 goto out_unlock;
1424
1425 /*
1426 * Ensure that the uaccess routines can actually operate on userspace
1427 * pointers:
1428 */
1429 force_uaccess_begin();
1430
1431 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1432 PF_NOFREEZE | PF_NO_SETAFFINITY);
1433 flush_thread();
1434 me->personality &= ~bprm->per_clear;
1435
1436 /*
1437 * We have to apply CLOEXEC before we change whether the process is
1438 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1439 * trying to access the should-be-closed file descriptors of a process
1440 * undergoing exec(2).
1441 */
1442 do_close_on_exec(me->files);
1443
1444 if (bprm->secureexec) {
1445 /* Make sure parent cannot signal privileged process. */
1446 me->pdeath_signal = 0;
1447
1448 /*
1449 * For secureexec, reset the stack limit to sane default to
1450 * avoid bad behavior from the prior rlimits. This has to
1451 * happen before arch_pick_mmap_layout(), which examines
1452 * RLIMIT_STACK, but after the point of no return to avoid
1453 * needing to clean up the change on failure.
1454 */
1455 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1456 bprm->rlim_stack.rlim_cur = _STK_LIM;
1457 }
1458
1459 me->sas_ss_sp = me->sas_ss_size = 0;
1460
1461 /*
1462 * Figure out dumpability. Note that this checking only of current
1463 * is wrong, but userspace depends on it. This should be testing
1464 * bprm->secureexec instead.
1465 */
1466 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1467 !(uid_eq(current_euid(), current_uid()) &&
1468 gid_eq(current_egid(), current_gid())))
1469 set_dumpable(current->mm, suid_dumpable);
1470 else
1471 set_dumpable(current->mm, SUID_DUMP_USER);
1472
1473 perf_event_exec();
1474 __set_task_comm(me, kbasename(bprm->filename), true);
1475
1476 /* An exec changes our domain. We are no longer part of the thread
1477 group */
1478 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1479 flush_signal_handlers(me, 0);
1480
1481 /*
1482 * install the new credentials for this executable
1483 */
1484 security_bprm_committing_creds(bprm);
1485
1486 commit_creds(bprm->cred);
1487 bprm->cred = NULL;
1488
1489 /*
1490 * Disable monitoring for regular users
1491 * when executing setuid binaries. Must
1492 * wait until new credentials are committed
1493 * by commit_creds() above
1494 */
1495 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1496 perf_event_exit_task(me);
1497 /*
1498 * cred_guard_mutex must be held at least to this point to prevent
1499 * ptrace_attach() from altering our determination of the task's
1500 * credentials; any time after this it may be unlocked.
1501 */
1502 security_bprm_committed_creds(bprm);
1503
1504 /* Pass the opened binary to the interpreter. */
1505 if (bprm->have_execfd) {
1506 retval = get_unused_fd_flags(0);
1507 if (retval < 0)
1508 goto out_unlock;
1509 fd_install(retval, bprm->executable);
1510 bprm->executable = NULL;
1511 bprm->execfd = retval;
1512 }
1513 return 0;
1514
1515 out_unlock:
1516 mutex_unlock(&me->signal->exec_update_mutex);
1517 out:
1518 return retval;
1519 }
1520 EXPORT_SYMBOL(begin_new_exec);
1521
1522 void would_dump(struct linux_binprm *bprm, struct file *file)
1523 {
1524 struct inode *inode = file_inode(file);
1525 if (inode_permission(inode, MAY_READ) < 0) {
1526 struct user_namespace *old, *user_ns;
1527 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1528
1529 /* Ensure mm->user_ns contains the executable */
1530 user_ns = old = bprm->mm->user_ns;
1531 while ((user_ns != &init_user_ns) &&
1532 !privileged_wrt_inode_uidgid(user_ns, inode))
1533 user_ns = user_ns->parent;
1534
1535 if (old != user_ns) {
1536 bprm->mm->user_ns = get_user_ns(user_ns);
1537 put_user_ns(old);
1538 }
1539 }
1540 }
1541 EXPORT_SYMBOL(would_dump);
1542
1543 void setup_new_exec(struct linux_binprm * bprm)
1544 {
1545 /* Setup things that can depend upon the personality */
1546 struct task_struct *me = current;
1547
1548 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1549
1550 arch_setup_new_exec();
1551
1552 /* Set the new mm task size. We have to do that late because it may
1553 * depend on TIF_32BIT which is only updated in flush_thread() on
1554 * some architectures like powerpc
1555 */
1556 me->mm->task_size = TASK_SIZE;
1557 mutex_unlock(&me->signal->exec_update_mutex);
1558 mutex_unlock(&me->signal->cred_guard_mutex);
1559 }
1560 EXPORT_SYMBOL(setup_new_exec);
1561
1562 /* Runs immediately before start_thread() takes over. */
1563 void finalize_exec(struct linux_binprm *bprm)
1564 {
1565 /* Store any stack rlimit changes before starting thread. */
1566 task_lock(current->group_leader);
1567 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1568 task_unlock(current->group_leader);
1569 }
1570 EXPORT_SYMBOL(finalize_exec);
1571
1572 /*
1573 * Prepare credentials and lock ->cred_guard_mutex.
1574 * setup_new_exec() commits the new creds and drops the lock.
1575 * Or, if exec fails before, free_bprm() should release ->cred and
1576 * and unlock.
1577 */
1578 static int prepare_bprm_creds(struct linux_binprm *bprm)
1579 {
1580 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1581 return -ERESTARTNOINTR;
1582
1583 bprm->cred = prepare_exec_creds();
1584 if (likely(bprm->cred))
1585 return 0;
1586
1587 mutex_unlock(¤t->signal->cred_guard_mutex);
1588 return -ENOMEM;
1589 }
1590
1591 static void free_bprm(struct linux_binprm *bprm)
1592 {
1593 if (bprm->mm) {
1594 acct_arg_size(bprm, 0);
1595 mmput(bprm->mm);
1596 }
1597 free_arg_pages(bprm);
1598 if (bprm->cred) {
1599 mutex_unlock(¤t->signal->cred_guard_mutex);
1600 abort_creds(bprm->cred);
1601 }
1602 if (bprm->file) {
1603 allow_write_access(bprm->file);
1604 fput(bprm->file);
1605 }
1606 if (bprm->executable)
1607 fput(bprm->executable);
1608 /* If a binfmt changed the interp, free it. */
1609 if (bprm->interp != bprm->filename)
1610 kfree(bprm->interp);
1611 kfree(bprm->fdpath);
1612 kfree(bprm);
1613 }
1614
1615 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1616 {
1617 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1618 int retval = -ENOMEM;
1619 if (!bprm)
1620 goto out;
1621
1622 if (fd == AT_FDCWD || filename->name[0] == '/') {
1623 bprm->filename = filename->name;
1624 } else {
1625 if (filename->name[0] == '\0')
1626 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1627 else
1628 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1629 fd, filename->name);
1630 if (!bprm->fdpath)
1631 goto out_free;
1632
1633 bprm->filename = bprm->fdpath;
1634 }
1635 bprm->interp = bprm->filename;
1636
1637 retval = bprm_mm_init(bprm);
1638 if (retval)
1639 goto out_free;
1640 return bprm;
1641
1642 out_free:
1643 free_bprm(bprm);
1644 out:
1645 return ERR_PTR(retval);
1646 }
1647
1648 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1649 {
1650 /* If a binfmt changed the interp, free it first. */
1651 if (bprm->interp != bprm->filename)
1652 kfree(bprm->interp);
1653 bprm->interp = kstrdup(interp, GFP_KERNEL);
1654 if (!bprm->interp)
1655 return -ENOMEM;
1656 return 0;
1657 }
1658 EXPORT_SYMBOL(bprm_change_interp);
1659
1660 /*
1661 * determine how safe it is to execute the proposed program
1662 * - the caller must hold ->cred_guard_mutex to protect against
1663 * PTRACE_ATTACH or seccomp thread-sync
1664 */
1665 static void check_unsafe_exec(struct linux_binprm *bprm)
1666 {
1667 struct task_struct *p = current, *t;
1668 unsigned n_fs;
1669
1670 if (p->ptrace)
1671 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1672
1673 /*
1674 * This isn't strictly necessary, but it makes it harder for LSMs to
1675 * mess up.
1676 */
1677 if (task_no_new_privs(current))
1678 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1679
1680 t = p;
1681 n_fs = 1;
1682 spin_lock(&p->fs->lock);
1683 rcu_read_lock();
1684 while_each_thread(p, t) {
1685 if (t->fs == p->fs)
1686 n_fs++;
1687 }
1688 rcu_read_unlock();
1689
1690 if (p->fs->users > n_fs)
1691 bprm->unsafe |= LSM_UNSAFE_SHARE;
1692 else
1693 p->fs->in_exec = 1;
1694 spin_unlock(&p->fs->lock);
1695 }
1696
1697 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1698 {
1699 /* Handle suid and sgid on files */
1700 struct inode *inode;
1701 unsigned int mode;
1702 kuid_t uid;
1703 kgid_t gid;
1704
1705 if (!mnt_may_suid(file->f_path.mnt))
1706 return;
1707
1708 if (task_no_new_privs(current))
1709 return;
1710
1711 inode = file->f_path.dentry->d_inode;
1712 mode = READ_ONCE(inode->i_mode);
1713 if (!(mode & (S_ISUID|S_ISGID)))
1714 return;
1715
1716 /* Be careful if suid/sgid is set */
1717 inode_lock(inode);
1718
1719 /* reload atomically mode/uid/gid now that lock held */
1720 mode = inode->i_mode;
1721 uid = inode->i_uid;
1722 gid = inode->i_gid;
1723 inode_unlock(inode);
1724
1725 /* We ignore suid/sgid if there are no mappings for them in the ns */
1726 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1727 !kgid_has_mapping(bprm->cred->user_ns, gid))
1728 return;
1729
1730 if (mode & S_ISUID) {
1731 bprm->per_clear |= PER_CLEAR_ON_SETID;
1732 bprm->cred->euid = uid;
1733 }
1734
1735 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1736 bprm->per_clear |= PER_CLEAR_ON_SETID;
1737 bprm->cred->egid = gid;
1738 }
1739 }
1740
1741 /*
1742 * Compute brpm->cred based upon the final binary.
1743 */
1744 static int bprm_creds_from_file(struct linux_binprm *bprm)
1745 {
1746 /* Compute creds based on which file? */
1747 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1748
1749 bprm_fill_uid(bprm, file);
1750 return security_bprm_creds_from_file(bprm, file);
1751 }
1752
1753 /*
1754 * Fill the binprm structure from the inode.
1755 * Read the first BINPRM_BUF_SIZE bytes
1756 *
1757 * This may be called multiple times for binary chains (scripts for example).
1758 */
1759 static int prepare_binprm(struct linux_binprm *bprm)
1760 {
1761 loff_t pos = 0;
1762
1763 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1764 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1765 }
1766
1767 /*
1768 * Arguments are '\0' separated strings found at the location bprm->p
1769 * points to; chop off the first by relocating brpm->p to right after
1770 * the first '\0' encountered.
1771 */
1772 int remove_arg_zero(struct linux_binprm *bprm)
1773 {
1774 int ret = 0;
1775 unsigned long offset;
1776 char *kaddr;
1777 struct page *page;
1778
1779 if (!bprm->argc)
1780 return 0;
1781
1782 do {
1783 offset = bprm->p & ~PAGE_MASK;
1784 page = get_arg_page(bprm, bprm->p, 0);
1785 if (!page) {
1786 ret = -EFAULT;
1787 goto out;
1788 }
1789 kaddr = kmap_atomic(page);
1790
1791 for (; offset < PAGE_SIZE && kaddr[offset];
1792 offset++, bprm->p++)
1793 ;
1794
1795 kunmap_atomic(kaddr);
1796 put_arg_page(page);
1797 } while (offset == PAGE_SIZE);
1798
1799 bprm->p++;
1800 bprm->argc--;
1801 ret = 0;
1802
1803 out:
1804 return ret;
1805 }
1806 EXPORT_SYMBOL(remove_arg_zero);
1807
1808 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1809 /*
1810 * cycle the list of binary formats handler, until one recognizes the image
1811 */
1812 static int search_binary_handler(struct linux_binprm *bprm)
1813 {
1814 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1815 struct linux_binfmt *fmt;
1816 int retval;
1817
1818 retval = prepare_binprm(bprm);
1819 if (retval < 0)
1820 return retval;
1821
1822 retval = security_bprm_check(bprm);
1823 if (retval)
1824 return retval;
1825
1826 retval = -ENOENT;
1827 retry:
1828 read_lock(&binfmt_lock);
1829 list_for_each_entry(fmt, &formats, lh) {
1830 if (!try_module_get(fmt->module))
1831 continue;
1832 read_unlock(&binfmt_lock);
1833
1834 retval = fmt->load_binary(bprm);
1835
1836 read_lock(&binfmt_lock);
1837 put_binfmt(fmt);
1838 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1839 read_unlock(&binfmt_lock);
1840 return retval;
1841 }
1842 }
1843 read_unlock(&binfmt_lock);
1844
1845 if (need_retry) {
1846 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1847 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1848 return retval;
1849 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1850 return retval;
1851 need_retry = false;
1852 goto retry;
1853 }
1854
1855 return retval;
1856 }
1857
1858 static int exec_binprm(struct linux_binprm *bprm)
1859 {
1860 pid_t old_pid, old_vpid;
1861 int ret, depth;
1862
1863 /* Need to fetch pid before load_binary changes it */
1864 old_pid = current->pid;
1865 rcu_read_lock();
1866 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1867 rcu_read_unlock();
1868
1869 /* This allows 4 levels of binfmt rewrites before failing hard. */
1870 for (depth = 0;; depth++) {
1871 struct file *exec;
1872 if (depth > 5)
1873 return -ELOOP;
1874
1875 ret = search_binary_handler(bprm);
1876 if (ret < 0)
1877 return ret;
1878 if (!bprm->interpreter)
1879 break;
1880
1881 exec = bprm->file;
1882 bprm->file = bprm->interpreter;
1883 bprm->interpreter = NULL;
1884
1885 allow_write_access(exec);
1886 if (unlikely(bprm->have_execfd)) {
1887 if (bprm->executable) {
1888 fput(exec);
1889 return -ENOEXEC;
1890 }
1891 bprm->executable = exec;
1892 } else
1893 fput(exec);
1894 }
1895
1896 audit_bprm(bprm);
1897 trace_sched_process_exec(current, old_pid, bprm);
1898 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1899 proc_exec_connector(current);
1900 return 0;
1901 }
1902
1903 /*
1904 * sys_execve() executes a new program.
1905 */
1906 static int bprm_execve(struct linux_binprm *bprm,
1907 int fd, struct filename *filename, int flags)
1908 {
1909 struct file *file;
1910 struct files_struct *displaced;
1911 int retval;
1912
1913 /*
1914 * Cancel any io_uring activity across execve
1915 */
1916 io_uring_task_cancel();
1917
1918 retval = unshare_files(&displaced);
1919 if (retval)
1920 return retval;
1921
1922 retval = prepare_bprm_creds(bprm);
1923 if (retval)
1924 goto out_files;
1925
1926 check_unsafe_exec(bprm);
1927 current->in_execve = 1;
1928
1929 file = do_open_execat(fd, filename, flags);
1930 retval = PTR_ERR(file);
1931 if (IS_ERR(file))
1932 goto out_unmark;
1933
1934 sched_exec();
1935
1936 bprm->file = file;
1937 /*
1938 * Record that a name derived from an O_CLOEXEC fd will be
1939 * inaccessible after exec. Relies on having exclusive access to
1940 * current->files (due to unshare_files above).
1941 */
1942 if (bprm->fdpath &&
1943 close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1944 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1945
1946 /* Set the unchanging part of bprm->cred */
1947 retval = security_bprm_creds_for_exec(bprm);
1948 if (retval)
1949 goto out;
1950
1951 retval = ccs_exec_binprm(bprm);
1952 if (retval < 0)
1953 goto out;
1954
1955 /* execve succeeded */
1956 current->fs->in_exec = 0;
1957 current->in_execve = 0;
1958 rseq_execve(current);
1959 acct_update_integrals(current);
1960 task_numa_free(current, false);
1961 if (displaced)
1962 put_files_struct(displaced);
1963 return retval;
1964
1965 out:
1966 /*
1967 * If past the point of no return ensure the the code never
1968 * returns to the userspace process. Use an existing fatal
1969 * signal if present otherwise terminate the process with
1970 * SIGSEGV.
1971 */
1972 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1973 force_sigsegv(SIGSEGV);
1974
1975 out_unmark:
1976 current->fs->in_exec = 0;
1977 current->in_execve = 0;
1978
1979 out_files:
1980 if (displaced)
1981 reset_files_struct(displaced);
1982
1983 return retval;
1984 }
1985
1986 static int do_execveat_common(int fd, struct filename *filename,
1987 struct user_arg_ptr argv,
1988 struct user_arg_ptr envp,
1989 int flags)
1990 {
1991 struct linux_binprm *bprm;
1992 int retval;
1993
1994 if (IS_ERR(filename))
1995 return PTR_ERR(filename);
1996
1997 /*
1998 * We move the actual failure in case of RLIMIT_NPROC excess from
1999 * set*uid() to execve() because too many poorly written programs
2000 * don't check setuid() return code. Here we additionally recheck
2001 * whether NPROC limit is still exceeded.
2002 */
2003 if ((current->flags & PF_NPROC_EXCEEDED) &&
2004 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
2005 retval = -EAGAIN;
2006 goto out_ret;
2007 }
2008
2009 /* We're below the limit (still or again), so we don't want to make
2010 * further execve() calls fail. */
2011 current->flags &= ~PF_NPROC_EXCEEDED;
2012
2013 bprm = alloc_bprm(fd, filename);
2014 if (IS_ERR(bprm)) {
2015 retval = PTR_ERR(bprm);
2016 goto out_ret;
2017 }
2018
2019 retval = count(argv, MAX_ARG_STRINGS);
2020 if (retval < 0)
2021 goto out_free;
2022 bprm->argc = retval;
2023
2024 retval = count(envp, MAX_ARG_STRINGS);
2025 if (retval < 0)
2026 goto out_free;
2027 bprm->envc = retval;
2028
2029 retval = bprm_stack_limits(bprm);
2030 if (retval < 0)
2031 goto out_free;
2032
2033 retval = copy_string_kernel(bprm->filename, bprm);
2034 if (retval < 0)
2035 goto out_free;
2036 bprm->exec = bprm->p;
2037
2038 retval = copy_strings(bprm->envc, envp, bprm);
2039 if (retval < 0)
2040 goto out_free;
2041
2042 retval = copy_strings(bprm->argc, argv, bprm);
2043 if (retval < 0)
2044 goto out_free;
2045
2046 retval = bprm_execve(bprm, fd, filename, flags);
2047 out_free:
2048 free_bprm(bprm);
2049
2050 out_ret:
2051 putname(filename);
2052 return retval;
2053 }
2054
2055 int kernel_execve(const char *kernel_filename,
2056 const char *const *argv, const char *const *envp)
2057 {
2058 struct filename *filename;
2059 struct linux_binprm *bprm;
2060 int fd = AT_FDCWD;
2061 int retval;
2062
2063 filename = getname_kernel(kernel_filename);
2064 if (IS_ERR(filename))
2065 return PTR_ERR(filename);
2066
2067 bprm = alloc_bprm(fd, filename);
2068 if (IS_ERR(bprm)) {
2069 retval = PTR_ERR(bprm);
2070 goto out_ret;
2071 }
2072
2073 retval = count_strings_kernel(argv);
2074 if (retval < 0)
2075 goto out_free;
2076 bprm->argc = retval;
2077
2078 retval = count_strings_kernel(envp);
2079 if (retval < 0)
2080 goto out_free;
2081 bprm->envc = retval;
2082
2083 retval = bprm_stack_limits(bprm);
2084 if (retval < 0)
2085 goto out_free;
2086
2087 retval = copy_string_kernel(bprm->filename, bprm);
2088 if (retval < 0)
2089 goto out_free;
2090 bprm->exec = bprm->p;
2091
2092 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2093 if (retval < 0)
2094 goto out_free;
2095
2096 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2097 if (retval < 0)
2098 goto out_free;
2099
2100 retval = bprm_execve(bprm, fd, filename, 0);
2101 out_free:
2102 free_bprm(bprm);
2103 out_ret:
2104 putname(filename);
2105 return retval;
2106 }
2107
2108 static int do_execve(struct filename *filename,
2109 const char __user *const __user *__argv,
2110 const char __user *const __user *__envp)
2111 {
2112 struct user_arg_ptr argv = { .ptr.native = __argv };
2113 struct user_arg_ptr envp = { .ptr.native = __envp };
2114 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2115 }
2116
2117 static int do_execveat(int fd, struct filename *filename,
2118 const char __user *const __user *__argv,
2119 const char __user *const __user *__envp,
2120 int flags)
2121 {
2122 struct user_arg_ptr argv = { .ptr.native = __argv };
2123 struct user_arg_ptr envp = { .ptr.native = __envp };
2124
2125 return do_execveat_common(fd, filename, argv, envp, flags);
2126 }
2127
2128 #ifdef CONFIG_COMPAT
2129 static int compat_do_execve(struct filename *filename,
2130 const compat_uptr_t __user *__argv,
2131 const compat_uptr_t __user *__envp)
2132 {
2133 struct user_arg_ptr argv = {
2134 .is_compat = true,
2135 .ptr.compat = __argv,
2136 };
2137 struct user_arg_ptr envp = {
2138 .is_compat = true,
2139 .ptr.compat = __envp,
2140 };
2141 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2142 }
2143
2144 static int compat_do_execveat(int fd, struct filename *filename,
2145 const compat_uptr_t __user *__argv,
2146 const compat_uptr_t __user *__envp,
2147 int flags)
2148 {
2149 struct user_arg_ptr argv = {
2150 .is_compat = true,
2151 .ptr.compat = __argv,
2152 };
2153 struct user_arg_ptr envp = {
2154 .is_compat = true,
2155 .ptr.compat = __envp,
2156 };
2157 return do_execveat_common(fd, filename, argv, envp, flags);
2158 }
2159 #endif
2160
2161 void set_binfmt(struct linux_binfmt *new)
2162 {
2163 struct mm_struct *mm = current->mm;
2164
2165 if (mm->binfmt)
2166 module_put(mm->binfmt->module);
2167
2168 mm->binfmt = new;
2169 if (new)
2170 __module_get(new->module);
2171 }
2172 EXPORT_SYMBOL(set_binfmt);
2173
2174 /*
2175 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2176 */
2177 void set_dumpable(struct mm_struct *mm, int value)
2178 {
2179 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2180 return;
2181
2182 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2183 }
2184
2185 SYSCALL_DEFINE3(execve,
2186 const char __user *, filename,
2187 const char __user *const __user *, argv,
2188 const char __user *const __user *, envp)
2189 {
2190 return do_execve(getname(filename), argv, envp);
2191 }
2192
2193 SYSCALL_DEFINE5(execveat,
2194 int, fd, const char __user *, filename,
2195 const char __user *const __user *, argv,
2196 const char __user *const __user *, envp,
2197 int, flags)
2198 {
2199 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2200
2201 return do_execveat(fd,
2202 getname_flags(filename, lookup_flags, NULL),
2203 argv, envp, flags);
2204 }
2205
2206 #ifdef CONFIG_COMPAT
2207 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2208 const compat_uptr_t __user *, argv,
2209 const compat_uptr_t __user *, envp)
2210 {
2211 return compat_do_execve(getname(filename), argv, envp);
2212 }
2213
2214 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2215 const char __user *, filename,
2216 const compat_uptr_t __user *, argv,
2217 const compat_uptr_t __user *, envp,
2218 int, flags)
2219 {
2220 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2221
2222 return compat_do_execveat(fd,
2223 getname_flags(filename, lookup_flags, NULL),
2224 argv, envp, flags);
2225 }
2226 #endif
2227
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.