projects/linux·blob v6.9-rc4 · refs

2194 lines · 52.4 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/exec.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * #!-checking implemented by tytso.
10 */
11/*
12 * Demand-loading implemented 01.12.91 - no need to read anything but
13 * the header into memory. The inode of the executable is put into
14 * "current->executable", and page faults do the actual loading. Clean.
15 *
16 * Once more I can proudly say that linux stood up to being changed: it
17 * was less than 2 hours work to get demand-loading completely implemented.
18 *
19 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
20 * current->executable is only used by the procfs. This allows a dispatch
21 * table to check for several different types of binary formats. We keep
22 * trying until we recognize the file or we run out of supported binary
23 * formats.
24 */
25
26#include <linux/kernel_read_file.h>
27#include <linux/slab.h>
28#include <linux/file.h>
29#include <linux/fdtable.h>
30#include <linux/mm.h>
31#include <linux/stat.h>
32#include <linux/fcntl.h>
33#include <linux/swap.h>
34#include <linux/string.h>
35#include <linux/init.h>
36#include <linux/sched/mm.h>
37#include <linux/sched/coredump.h>
38#include <linux/sched/signal.h>
39#include <linux/sched/numa_balancing.h>
40#include <linux/sched/task.h>
41#include <linux/pagemap.h>
42#include <linux/perf_event.h>
43#include <linux/highmem.h>
44#include <linux/spinlock.h>
45#include <linux/key.h>
46#include <linux/personality.h>
47#include <linux/binfmts.h>
48#include <linux/utsname.h>
49#include <linux/pid_namespace.h>
50#include <linux/module.h>
51#include <linux/namei.h>
52#include <linux/mount.h>
53#include <linux/security.h>
54#include <linux/syscalls.h>
55#include <linux/tsacct_kern.h>
56#include <linux/cn_proc.h>
57#include <linux/audit.h>
58#include <linux/kmod.h>
59#include <linux/fsnotify.h>
60#include <linux/fs_struct.h>
61#include <linux/oom.h>
62#include <linux/compat.h>
63#include <linux/vmalloc.h>
64#include <linux/io_uring.h>
65#include <linux/syscall_user_dispatch.h>
66#include <linux/coredump.h>
67#include <linux/time_namespace.h>
68#include <linux/user_events.h>
69#include <linux/rseq.h>
70
71#include <linux/uaccess.h>
72#include <asm/mmu_context.h>
73#include <asm/tlb.h>
74
75#include <trace/events/task.h>
76#include "internal.h"
77
78#include <trace/events/sched.h>
79
80static int bprm_creds_from_file(struct linux_binprm *bprm);
81
82int suid_dumpable = 0;
83
84static LIST_HEAD(formats);
85static DEFINE_RWLOCK(binfmt_lock);
86
87void __register_binfmt(struct linux_binfmt * fmt, int insert)
88{
89 write_lock(&binfmt_lock);
90 insert ? list_add(&fmt->lh, &formats) :
91 list_add_tail(&fmt->lh, &formats);
92 write_unlock(&binfmt_lock);
93}
94
95EXPORT_SYMBOL(__register_binfmt);
96
97void unregister_binfmt(struct linux_binfmt * fmt)
98{
99 write_lock(&binfmt_lock);
100 list_del(&fmt->lh);
101 write_unlock(&binfmt_lock);
102}
103
104EXPORT_SYMBOL(unregister_binfmt);
105
106static inline void put_binfmt(struct linux_binfmt * fmt)
107{
108 module_put(fmt->module);
109}
110
111bool path_noexec(const struct path *path)
112{
113 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
114 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
115}
116
117#ifdef CONFIG_USELIB
118/*
119 * Note that a shared library must be both readable and executable due to
120 * security reasons.
121 *
122 * Also note that we take the address to load from the file itself.
123 */
124SYSCALL_DEFINE1(uselib, const char __user *, library)
125{
126 struct linux_binfmt *fmt;
127 struct file *file;
128 struct filename *tmp = getname(library);
129 int error = PTR_ERR(tmp);
130 static const struct open_flags uselib_flags = {
131 .open_flag = O_LARGEFILE | O_RDONLY,
132 .acc_mode = MAY_READ | MAY_EXEC,
133 .intent = LOOKUP_OPEN,
134 .lookup_flags = LOOKUP_FOLLOW,
135 };
136
137 if (IS_ERR(tmp))
138 goto out;
139
140 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
141 putname(tmp);
142 error = PTR_ERR(file);
143 if (IS_ERR(file))
144 goto out;
145
146 /*
147 * may_open() has already checked for this, so it should be
148 * impossible to trip now. But we need to be extra cautious
149 * and check again at the very end too.
150 */
151 error = -EACCES;
152 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
153 path_noexec(&file->f_path)))
154 goto exit;
155
156 error = -ENOEXEC;
157
158 read_lock(&binfmt_lock);
159 list_for_each_entry(fmt, &formats, lh) {
160 if (!fmt->load_shlib)
161 continue;
162 if (!try_module_get(fmt->module))
163 continue;
164 read_unlock(&binfmt_lock);
165 error = fmt->load_shlib(file);
166 read_lock(&binfmt_lock);
167 put_binfmt(fmt);
168 if (error != -ENOEXEC)
169 break;
170 }
171 read_unlock(&binfmt_lock);
172exit:
173 fput(file);
174out:
175 return error;
176}
177#endif /* #ifdef CONFIG_USELIB */
178
179#ifdef CONFIG_MMU
180/*
181 * The nascent bprm->mm is not visible until exec_mmap() but it can
182 * use a lot of memory, account these pages in current->mm temporary
183 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
184 * change the counter back via acct_arg_size(0).
185 */
186static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
187{
188 struct mm_struct *mm = current->mm;
189 long diff = (long)(pages - bprm->vma_pages);
190
191 if (!mm || !diff)
192 return;
193
194 bprm->vma_pages = pages;
195 add_mm_counter(mm, MM_ANONPAGES, diff);
196}
197
198static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
199 int write)
200{
201 struct page *page;
202 struct vm_area_struct *vma = bprm->vma;
203 struct mm_struct *mm = bprm->mm;
204 int ret;
205
206 /*
207 * Avoid relying on expanding the stack down in GUP (which
208 * does not work for STACK_GROWSUP anyway), and just do it
209 * by hand ahead of time.
210 */
211 if (write && pos < vma->vm_start) {
212 mmap_write_lock(mm);
213 ret = expand_downwards(vma, pos);
214 if (unlikely(ret < 0)) {
215 mmap_write_unlock(mm);
216 return NULL;
217 }
218 mmap_write_downgrade(mm);
219 } else
220 mmap_read_lock(mm);
221
222 /*
223 * We are doing an exec(). 'current' is the process
224 * doing the exec and 'mm' is the new process's mm.
225 */
226 ret = get_user_pages_remote(mm, pos, 1,
227 write ? FOLL_WRITE : 0,
228 &page, NULL);
229 mmap_read_unlock(mm);
230 if (ret <= 0)
231 return NULL;
232
233 if (write)
234 acct_arg_size(bprm, vma_pages(vma));
235
236 return page;
237}
238
239static void put_arg_page(struct page *page)
240{
241 put_page(page);
242}
243
244static void free_arg_pages(struct linux_binprm *bprm)
245{
246}
247
248static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
249 struct page *page)
250{
251 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
252}
253
254static int __bprm_mm_init(struct linux_binprm *bprm)
255{
256 int err;
257 struct vm_area_struct *vma = NULL;
258 struct mm_struct *mm = bprm->mm;
259
260 bprm->vma = vma = vm_area_alloc(mm);
261 if (!vma)
262 return -ENOMEM;
263 vma_set_anonymous(vma);
264
265 if (mmap_write_lock_killable(mm)) {
266 err = -EINTR;
267 goto err_free;
268 }
269
270 /*
271 * Place the stack at the largest stack address the architecture
272 * supports. Later, we'll move this to an appropriate place. We don't
273 * use STACK_TOP because that can depend on attributes which aren't
274 * configured yet.
275 */
276 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
277 vma->vm_end = STACK_TOP_MAX;
278 vma->vm_start = vma->vm_end - PAGE_SIZE;
279 vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);
280 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
281
282 err = insert_vm_struct(mm, vma);
283 if (err)
284 goto err;
285
286 mm->stack_vm = mm->total_vm = 1;
287 mmap_write_unlock(mm);
288 bprm->p = vma->vm_end - sizeof(void *);
289 return 0;
290err:
291 mmap_write_unlock(mm);
292err_free:
293 bprm->vma = NULL;
294 vm_area_free(vma);
295 return err;
296}
297
298static bool valid_arg_len(struct linux_binprm *bprm, long len)
299{
300 return len <= MAX_ARG_STRLEN;
301}
302
303#else
304
305static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
306{
307}
308
309static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
310 int write)
311{
312 struct page *page;
313
314 page = bprm->page[pos / PAGE_SIZE];
315 if (!page && write) {
316 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
317 if (!page)
318 return NULL;
319 bprm->page[pos / PAGE_SIZE] = page;
320 }
321
322 return page;
323}
324
325static void put_arg_page(struct page *page)
326{
327}
328
329static void free_arg_page(struct linux_binprm *bprm, int i)
330{
331 if (bprm->page[i]) {
332 __free_page(bprm->page[i]);
333 bprm->page[i] = NULL;
334 }
335}
336
337static void free_arg_pages(struct linux_binprm *bprm)
338{
339 int i;
340
341 for (i = 0; i < MAX_ARG_PAGES; i++)
342 free_arg_page(bprm, i);
343}
344
345static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
346 struct page *page)
347{
348}
349
350static int __bprm_mm_init(struct linux_binprm *bprm)
351{
352 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
353 return 0;
354}
355
356static bool valid_arg_len(struct linux_binprm *bprm, long len)
357{
358 return len <= bprm->p;
359}
360
361#endif /* CONFIG_MMU */
362
363/*
364 * Create a new mm_struct and populate it with a temporary stack
365 * vm_area_struct. We don't have enough context at this point to set the stack
366 * flags, permissions, and offset, so we use temporary values. We'll update
367 * them later in setup_arg_pages().
368 */
369static int bprm_mm_init(struct linux_binprm *bprm)
370{
371 int err;
372 struct mm_struct *mm = NULL;
373
374 bprm->mm = mm = mm_alloc();
375 err = -ENOMEM;
376 if (!mm)
377 goto err;
378
379 /* Save current stack limit for all calculations made during exec. */
380 task_lock(current->group_leader);
381 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
382 task_unlock(current->group_leader);
383
384 err = __bprm_mm_init(bprm);
385 if (err)
386 goto err;
387
388 return 0;
389
390err:
391 if (mm) {
392 bprm->mm = NULL;
393 mmdrop(mm);
394 }
395
396 return err;
397}
398
399struct user_arg_ptr {
400#ifdef CONFIG_COMPAT
401 bool is_compat;
402#endif
403 union {
404 const char __user *const __user *native;
405#ifdef CONFIG_COMPAT
406 const compat_uptr_t __user *compat;
407#endif
408 } ptr;
409};
410
411static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
412{
413 const char __user *native;
414
415#ifdef CONFIG_COMPAT
416 if (unlikely(argv.is_compat)) {
417 compat_uptr_t compat;
418
419 if (get_user(compat, argv.ptr.compat + nr))
420 return ERR_PTR(-EFAULT);
421
422 return compat_ptr(compat);
423 }
424#endif
425
426 if (get_user(native, argv.ptr.native + nr))
427 return ERR_PTR(-EFAULT);
428
429 return native;
430}
431
432/*
433 * count() counts the number of strings in array ARGV.
434 */
435static int count(struct user_arg_ptr argv, int max)
436{
437 int i = 0;
438
439 if (argv.ptr.native != NULL) {
440 for (;;) {
441 const char __user *p = get_user_arg_ptr(argv, i);
442
443 if (!p)
444 break;
445
446 if (IS_ERR(p))
447 return -EFAULT;
448
449 if (i >= max)
450 return -E2BIG;
451 ++i;
452
453 if (fatal_signal_pending(current))
454 return -ERESTARTNOHAND;
455 cond_resched();
456 }
457 }
458 return i;
459}
460
461static int count_strings_kernel(const char *const *argv)
462{
463 int i;
464
465 if (!argv)
466 return 0;
467
468 for (i = 0; argv[i]; ++i) {
469 if (i >= MAX_ARG_STRINGS)
470 return -E2BIG;
471 if (fatal_signal_pending(current))
472 return -ERESTARTNOHAND;
473 cond_resched();
474 }
475 return i;
476}
477
478static int bprm_stack_limits(struct linux_binprm *bprm)
479{
480 unsigned long limit, ptr_size;
481
482 /*
483 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
484 * (whichever is smaller) for the argv+env strings.
485 * This ensures that:
486 * - the remaining binfmt code will not run out of stack space,
487 * - the program will have a reasonable amount of stack left
488 * to work from.
489 */
490 limit = _STK_LIM / 4 * 3;
491 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
492 /*
493 * We've historically supported up to 32 pages (ARG_MAX)
494 * of argument strings even with small stacks
495 */
496 limit = max_t(unsigned long, limit, ARG_MAX);
497 /*
498 * We must account for the size of all the argv and envp pointers to
499 * the argv and envp strings, since they will also take up space in
500 * the stack. They aren't stored until much later when we can't
501 * signal to the parent that the child has run out of stack space.
502 * Instead, calculate it here so it's possible to fail gracefully.
503 *
504 * In the case of argc = 0, make sure there is space for adding a
505 * empty string (which will bump argc to 1), to ensure confused
506 * userspace programs don't start processing from argv[1], thinking
507 * argc can never be 0, to keep them from walking envp by accident.
508 * See do_execveat_common().
509 */
510 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
511 if (limit <= ptr_size)
512 return -E2BIG;
513 limit -= ptr_size;
514
515 bprm->argmin = bprm->p - limit;
516 return 0;
517}
518
519/*
520 * 'copy_strings()' copies argument/environment strings from the old
521 * processes's memory to the new process's stack. The call to get_user_pages()
522 * ensures the destination page is created and not swapped out.
523 */
524static int copy_strings(int argc, struct user_arg_ptr argv,
525 struct linux_binprm *bprm)
526{
527 struct page *kmapped_page = NULL;
528 char *kaddr = NULL;
529 unsigned long kpos = 0;
530 int ret;
531
532 while (argc-- > 0) {
533 const char __user *str;
534 int len;
535 unsigned long pos;
536
537 ret = -EFAULT;
538 str = get_user_arg_ptr(argv, argc);
539 if (IS_ERR(str))
540 goto out;
541
542 len = strnlen_user(str, MAX_ARG_STRLEN);
543 if (!len)
544 goto out;
545
546 ret = -E2BIG;
547 if (!valid_arg_len(bprm, len))
548 goto out;
549
550 /* We're going to work our way backwards. */
551 pos = bprm->p;
552 str += len;
553 bprm->p -= len;
554#ifdef CONFIG_MMU
555 if (bprm->p < bprm->argmin)
556 goto out;
557#endif
558
559 while (len > 0) {
560 int offset, bytes_to_copy;
561
562 if (fatal_signal_pending(current)) {
563 ret = -ERESTARTNOHAND;
564 goto out;
565 }
566 cond_resched();
567
568 offset = pos % PAGE_SIZE;
569 if (offset == 0)
570 offset = PAGE_SIZE;
571
572 bytes_to_copy = offset;
573 if (bytes_to_copy > len)
574 bytes_to_copy = len;
575
576 offset -= bytes_to_copy;
577 pos -= bytes_to_copy;
578 str -= bytes_to_copy;
579 len -= bytes_to_copy;
580
581 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
582 struct page *page;
583
584 page = get_arg_page(bprm, pos, 1);
585 if (!page) {
586 ret = -E2BIG;
587 goto out;
588 }
589
590 if (kmapped_page) {
591 flush_dcache_page(kmapped_page);
592 kunmap_local(kaddr);
593 put_arg_page(kmapped_page);
594 }
595 kmapped_page = page;
596 kaddr = kmap_local_page(kmapped_page);
597 kpos = pos & PAGE_MASK;
598 flush_arg_page(bprm, kpos, kmapped_page);
599 }
600 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
601 ret = -EFAULT;
602 goto out;
603 }
604 }
605 }
606 ret = 0;
607out:
608 if (kmapped_page) {
609 flush_dcache_page(kmapped_page);
610 kunmap_local(kaddr);
611 put_arg_page(kmapped_page);
612 }
613 return ret;
614}
615
616/*
617 * Copy and argument/environment string from the kernel to the processes stack.
618 */
619int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
620{
621 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
622 unsigned long pos = bprm->p;
623
624 if (len == 0)
625 return -EFAULT;
626 if (!valid_arg_len(bprm, len))
627 return -E2BIG;
628
629 /* We're going to work our way backwards. */
630 arg += len;
631 bprm->p -= len;
632 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
633 return -E2BIG;
634
635 while (len > 0) {
636 unsigned int bytes_to_copy = min_t(unsigned int, len,
637 min_not_zero(offset_in_page(pos), PAGE_SIZE));
638 struct page *page;
639
640 pos -= bytes_to_copy;
641 arg -= bytes_to_copy;
642 len -= bytes_to_copy;
643
644 page = get_arg_page(bprm, pos, 1);
645 if (!page)
646 return -E2BIG;
647 flush_arg_page(bprm, pos & PAGE_MASK, page);
648 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
649 put_arg_page(page);
650 }
651
652 return 0;
653}
654EXPORT_SYMBOL(copy_string_kernel);
655
656static int copy_strings_kernel(int argc, const char *const *argv,
657 struct linux_binprm *bprm)
658{
659 while (argc-- > 0) {
660 int ret = copy_string_kernel(argv[argc], bprm);
661 if (ret < 0)
662 return ret;
663 if (fatal_signal_pending(current))
664 return -ERESTARTNOHAND;
665 cond_resched();
666 }
667 return 0;
668}
669
670#ifdef CONFIG_MMU
671
672/*
673 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
674 * the binfmt code determines where the new stack should reside, we shift it to
675 * its final location. The process proceeds as follows:
676 *
677 * 1) Use shift to calculate the new vma endpoints.
678 * 2) Extend vma to cover both the old and new ranges. This ensures the
679 * arguments passed to subsequent functions are consistent.
680 * 3) Move vma's page tables to the new range.
681 * 4) Free up any cleared pgd range.
682 * 5) Shrink the vma to cover only the new range.
683 */
684static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
685{
686 struct mm_struct *mm = vma->vm_mm;
687 unsigned long old_start = vma->vm_start;
688 unsigned long old_end = vma->vm_end;
689 unsigned long length = old_end - old_start;
690 unsigned long new_start = old_start - shift;
691 unsigned long new_end = old_end - shift;
692 VMA_ITERATOR(vmi, mm, new_start);
693 struct vm_area_struct *next;
694 struct mmu_gather tlb;
695
696 BUG_ON(new_start > new_end);
697
698 /*
699 * ensure there are no vmas between where we want to go
700 * and where we are
701 */
702 if (vma != vma_next(&vmi))
703 return -EFAULT;
704
705 vma_iter_prev_range(&vmi);
706 /*
707 * cover the whole range: [new_start, old_end)
708 */
709 if (vma_expand(&vmi, vma, new_start, old_end, vma->vm_pgoff, NULL))
710 return -ENOMEM;
711
712 /*
713 * move the page tables downwards, on failure we rely on
714 * process cleanup to remove whatever mess we made.
715 */
716 if (length != move_page_tables(vma, old_start,
717 vma, new_start, length, false, true))
718 return -ENOMEM;
719
720 lru_add_drain();
721 tlb_gather_mmu(&tlb, mm);
722 next = vma_next(&vmi);
723 if (new_end > old_start) {
724 /*
725 * when the old and new regions overlap clear from new_end.
726 */
727 free_pgd_range(&tlb, new_end, old_end, new_end,
728 next ? next->vm_start : USER_PGTABLES_CEILING);
729 } else {
730 /*
731 * otherwise, clean from old_start; this is done to not touch
732 * the address space in [new_end, old_start) some architectures
733 * have constraints on va-space that make this illegal (IA64) -
734 * for the others its just a little faster.
735 */
736 free_pgd_range(&tlb, old_start, old_end, new_end,
737 next ? next->vm_start : USER_PGTABLES_CEILING);
738 }
739 tlb_finish_mmu(&tlb);
740
741 vma_prev(&vmi);
742 /* Shrink the vma to just the new range */
743 return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff);
744}
745
746/*
747 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
748 * the stack is optionally relocated, and some extra space is added.
749 */
750int setup_arg_pages(struct linux_binprm *bprm,
751 unsigned long stack_top,
752 int executable_stack)
753{
754 unsigned long ret;
755 unsigned long stack_shift;
756 struct mm_struct *mm = current->mm;
757 struct vm_area_struct *vma = bprm->vma;
758 struct vm_area_struct *prev = NULL;
759 unsigned long vm_flags;
760 unsigned long stack_base;
761 unsigned long stack_size;
762 unsigned long stack_expand;
763 unsigned long rlim_stack;
764 struct mmu_gather tlb;
765 struct vma_iterator vmi;
766
767#ifdef CONFIG_STACK_GROWSUP
768 /* Limit stack size */
769 stack_base = bprm->rlim_stack.rlim_max;
770
771 stack_base = calc_max_stack_size(stack_base);
772
773 /* Add space for stack randomization. */
774 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
775
776 /* Make sure we didn't let the argument array grow too large. */
777 if (vma->vm_end - vma->vm_start > stack_base)
778 return -ENOMEM;
779
780 stack_base = PAGE_ALIGN(stack_top - stack_base);
781
782 stack_shift = vma->vm_start - stack_base;
783 mm->arg_start = bprm->p - stack_shift;
784 bprm->p = vma->vm_end - stack_shift;
785#else
786 stack_top = arch_align_stack(stack_top);
787 stack_top = PAGE_ALIGN(stack_top);
788
789 if (unlikely(stack_top < mmap_min_addr) ||
790 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
791 return -ENOMEM;
792
793 stack_shift = vma->vm_end - stack_top;
794
795 bprm->p -= stack_shift;
796 mm->arg_start = bprm->p;
797#endif
798
799 if (bprm->loader)
800 bprm->loader -= stack_shift;
801 bprm->exec -= stack_shift;
802
803 if (mmap_write_lock_killable(mm))
804 return -EINTR;
805
806 vm_flags = VM_STACK_FLAGS;
807
808 /*
809 * Adjust stack execute permissions; explicitly enable for
810 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
811 * (arch default) otherwise.
812 */
813 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
814 vm_flags |= VM_EXEC;
815 else if (executable_stack == EXSTACK_DISABLE_X)
816 vm_flags &= ~VM_EXEC;
817 vm_flags |= mm->def_flags;
818 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
819
820 vma_iter_init(&vmi, mm, vma->vm_start);
821
822 tlb_gather_mmu(&tlb, mm);
823 ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
824 vm_flags);
825 tlb_finish_mmu(&tlb);
826
827 if (ret)
828 goto out_unlock;
829 BUG_ON(prev != vma);
830
831 if (unlikely(vm_flags & VM_EXEC)) {
832 pr_warn_once("process '%pD4' started with executable stack\n",
833 bprm->file);
834 }
835
836 /* Move stack pages down in memory. */
837 if (stack_shift) {
838 ret = shift_arg_pages(vma, stack_shift);
839 if (ret)
840 goto out_unlock;
841 }
842
843 /* mprotect_fixup is overkill to remove the temporary stack flags */
844 vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
845
846 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
847 stack_size = vma->vm_end - vma->vm_start;
848 /*
849 * Align this down to a page boundary as expand_stack
850 * will align it up.
851 */
852 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
853
854 stack_expand = min(rlim_stack, stack_size + stack_expand);
855
856#ifdef CONFIG_STACK_GROWSUP
857 stack_base = vma->vm_start + stack_expand;
858#else
859 stack_base = vma->vm_end - stack_expand;
860#endif
861 current->mm->start_stack = bprm->p;
862 ret = expand_stack_locked(vma, stack_base);
863 if (ret)
864 ret = -EFAULT;
865
866out_unlock:
867 mmap_write_unlock(mm);
868 return ret;
869}
870EXPORT_SYMBOL(setup_arg_pages);
871
872#else
873
874/*
875 * Transfer the program arguments and environment from the holding pages
876 * onto the stack. The provided stack pointer is adjusted accordingly.
877 */
878int transfer_args_to_stack(struct linux_binprm *bprm,
879 unsigned long *sp_location)
880{
881 unsigned long index, stop, sp;
882 int ret = 0;
883
884 stop = bprm->p >> PAGE_SHIFT;
885 sp = *sp_location;
886
887 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
888 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
889 char *src = kmap_local_page(bprm->page[index]) + offset;
890 sp -= PAGE_SIZE - offset;
891 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
892 ret = -EFAULT;
893 kunmap_local(src);
894 if (ret)
895 goto out;
896 }
897
898 bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
899 *sp_location = sp;
900
901out:
902 return ret;
903}
904EXPORT_SYMBOL(transfer_args_to_stack);
905
906#endif /* CONFIG_MMU */
907
908/*
909 * On success, caller must call do_close_execat() on the returned
910 * struct file to close it.
911 */
912static struct file *do_open_execat(int fd, struct filename *name, int flags)
913{
914 struct file *file;
915 int err;
916 struct open_flags open_exec_flags = {
917 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
918 .acc_mode = MAY_EXEC,
919 .intent = LOOKUP_OPEN,
920 .lookup_flags = LOOKUP_FOLLOW,
921 };
922
923 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
924 return ERR_PTR(-EINVAL);
925 if (flags & AT_SYMLINK_NOFOLLOW)
926 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
927 if (flags & AT_EMPTY_PATH)
928 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
929
930 file = do_filp_open(fd, name, &open_exec_flags);
931 if (IS_ERR(file))
932 goto out;
933
934 /*
935 * may_open() has already checked for this, so it should be
936 * impossible to trip now. But we need to be extra cautious
937 * and check again at the very end too.
938 */
939 err = -EACCES;
940 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
941 path_noexec(&file->f_path)))
942 goto exit;
943
944 err = deny_write_access(file);
945 if (err)
946 goto exit;
947
948out:
949 return file;
950
951exit:
952 fput(file);
953 return ERR_PTR(err);
954}
955
956/**
957 * open_exec - Open a path name for execution
958 *
959 * @name: path name to open with the intent of executing it.
960 *
961 * Returns ERR_PTR on failure or allocated struct file on success.
962 *
963 * As this is a wrapper for the internal do_open_execat(), callers
964 * must call allow_write_access() before fput() on release. Also see
965 * do_close_execat().
966 */
967struct file *open_exec(const char *name)
968{
969 struct filename *filename = getname_kernel(name);
970 struct file *f = ERR_CAST(filename);
971
972 if (!IS_ERR(filename)) {
973 f = do_open_execat(AT_FDCWD, filename, 0);
974 putname(filename);
975 }
976 return f;
977}
978EXPORT_SYMBOL(open_exec);
979
980#if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
981ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
982{
983 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
984 if (res > 0)
985 flush_icache_user_range(addr, addr + len);
986 return res;
987}
988EXPORT_SYMBOL(read_code);
989#endif
990
991/*
992 * Maps the mm_struct mm into the current task struct.
993 * On success, this function returns with exec_update_lock
994 * held for writing.
995 */
996static int exec_mmap(struct mm_struct *mm)
997{
998 struct task_struct *tsk;
999 struct mm_struct *old_mm, *active_mm;
1000 int ret;
1001
1002 /* Notify parent that we're no longer interested in the old VM */
1003 tsk = current;
1004 old_mm = current->mm;
1005 exec_mm_release(tsk, old_mm);
1006
1007 ret = down_write_killable(&tsk->signal->exec_update_lock);
1008 if (ret)
1009 return ret;
1010
1011 if (old_mm) {
1012 /*
1013 * If there is a pending fatal signal perhaps a signal
1014 * whose default action is to create a coredump get
1015 * out and die instead of going through with the exec.
1016 */
1017 ret = mmap_read_lock_killable(old_mm);
1018 if (ret) {
1019 up_write(&tsk->signal->exec_update_lock);
1020 return ret;
1021 }
1022 }
1023
1024 task_lock(tsk);
1025 membarrier_exec_mmap(mm);
1026
1027 local_irq_disable();
1028 active_mm = tsk->active_mm;
1029 tsk->active_mm = mm;
1030 tsk->mm = mm;
1031 mm_init_cid(mm);
1032 /*
1033 * This prevents preemption while active_mm is being loaded and
1034 * it and mm are being updated, which could cause problems for
1035 * lazy tlb mm refcounting when these are updated by context
1036 * switches. Not all architectures can handle irqs off over
1037 * activate_mm yet.
1038 */
1039 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1040 local_irq_enable();
1041 activate_mm(active_mm, mm);
1042 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1043 local_irq_enable();
1044 lru_gen_add_mm(mm);
1045 task_unlock(tsk);
1046 lru_gen_use_mm(mm);
1047 if (old_mm) {
1048 mmap_read_unlock(old_mm);
1049 BUG_ON(active_mm != old_mm);
1050 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1051 mm_update_next_owner(old_mm);
1052 mmput(old_mm);
1053 return 0;
1054 }
1055 mmdrop_lazy_tlb(active_mm);
1056 return 0;
1057}
1058
1059static int de_thread(struct task_struct *tsk)
1060{
1061 struct signal_struct *sig = tsk->signal;
1062 struct sighand_struct *oldsighand = tsk->sighand;
1063 spinlock_t *lock = &oldsighand->siglock;
1064
1065 if (thread_group_empty(tsk))
1066 goto no_thread_group;
1067
1068 /*
1069 * Kill all other threads in the thread group.
1070 */
1071 spin_lock_irq(lock);
1072 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1073 /*
1074 * Another group action in progress, just
1075 * return so that the signal is processed.
1076 */
1077 spin_unlock_irq(lock);
1078 return -EAGAIN;
1079 }
1080
1081 sig->group_exec_task = tsk;
1082 sig->notify_count = zap_other_threads(tsk);
1083 if (!thread_group_leader(tsk))
1084 sig->notify_count--;
1085
1086 while (sig->notify_count) {
1087 __set_current_state(TASK_KILLABLE);
1088 spin_unlock_irq(lock);
1089 schedule();
1090 if (__fatal_signal_pending(tsk))
1091 goto killed;
1092 spin_lock_irq(lock);
1093 }
1094 spin_unlock_irq(lock);
1095
1096 /*
1097 * At this point all other threads have exited, all we have to
1098 * do is to wait for the thread group leader to become inactive,
1099 * and to assume its PID:
1100 */
1101 if (!thread_group_leader(tsk)) {
1102 struct task_struct *leader = tsk->group_leader;
1103
1104 for (;;) {
1105 cgroup_threadgroup_change_begin(tsk);
1106 write_lock_irq(&tasklist_lock);
1107 /*
1108 * Do this under tasklist_lock to ensure that
1109 * exit_notify() can't miss ->group_exec_task
1110 */
1111 sig->notify_count = -1;
1112 if (likely(leader->exit_state))
1113 break;
1114 __set_current_state(TASK_KILLABLE);
1115 write_unlock_irq(&tasklist_lock);
1116 cgroup_threadgroup_change_end(tsk);
1117 schedule();
1118 if (__fatal_signal_pending(tsk))
1119 goto killed;
1120 }
1121
1122 /*
1123 * The only record we have of the real-time age of a
1124 * process, regardless of execs it's done, is start_time.
1125 * All the past CPU time is accumulated in signal_struct
1126 * from sister threads now dead. But in this non-leader
1127 * exec, nothing survives from the original leader thread,
1128 * whose birth marks the true age of this process now.
1129 * When we take on its identity by switching to its PID, we
1130 * also take its birthdate (always earlier than our own).
1131 */
1132 tsk->start_time = leader->start_time;
1133 tsk->start_boottime = leader->start_boottime;
1134
1135 BUG_ON(!same_thread_group(leader, tsk));
1136 /*
1137 * An exec() starts a new thread group with the
1138 * TGID of the previous thread group. Rehash the
1139 * two threads with a switched PID, and release
1140 * the former thread group leader:
1141 */
1142
1143 /* Become a process group leader with the old leader's pid.
1144 * The old leader becomes a thread of the this thread group.
1145 */
1146 exchange_tids(tsk, leader);
1147 transfer_pid(leader, tsk, PIDTYPE_TGID);
1148 transfer_pid(leader, tsk, PIDTYPE_PGID);
1149 transfer_pid(leader, tsk, PIDTYPE_SID);
1150
1151 list_replace_rcu(&leader->tasks, &tsk->tasks);
1152 list_replace_init(&leader->sibling, &tsk->sibling);
1153
1154 tsk->group_leader = tsk;
1155 leader->group_leader = tsk;
1156
1157 tsk->exit_signal = SIGCHLD;
1158 leader->exit_signal = -1;
1159
1160 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1161 leader->exit_state = EXIT_DEAD;
1162 /*
1163 * We are going to release_task()->ptrace_unlink() silently,
1164 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1165 * the tracer won't block again waiting for this thread.
1166 */
1167 if (unlikely(leader->ptrace))
1168 __wake_up_parent(leader, leader->parent);
1169 write_unlock_irq(&tasklist_lock);
1170 cgroup_threadgroup_change_end(tsk);
1171
1172 release_task(leader);
1173 }
1174
1175 sig->group_exec_task = NULL;
1176 sig->notify_count = 0;
1177
1178no_thread_group:
1179 /* we have changed execution domain */
1180 tsk->exit_signal = SIGCHLD;
1181
1182 BUG_ON(!thread_group_leader(tsk));
1183 return 0;
1184
1185killed:
1186 /* protects against exit_notify() and __exit_signal() */
1187 read_lock(&tasklist_lock);
1188 sig->group_exec_task = NULL;
1189 sig->notify_count = 0;
1190 read_unlock(&tasklist_lock);
1191 return -EAGAIN;
1192}
1193
1194
1195/*
1196 * This function makes sure the current process has its own signal table,
1197 * so that flush_signal_handlers can later reset the handlers without
1198 * disturbing other processes. (Other processes might share the signal
1199 * table via the CLONE_SIGHAND option to clone().)
1200 */
1201static int unshare_sighand(struct task_struct *me)
1202{
1203 struct sighand_struct *oldsighand = me->sighand;
1204
1205 if (refcount_read(&oldsighand->count) != 1) {
1206 struct sighand_struct *newsighand;
1207 /*
1208 * This ->sighand is shared with the CLONE_SIGHAND
1209 * but not CLONE_THREAD task, switch to the new one.
1210 */
1211 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1212 if (!newsighand)
1213 return -ENOMEM;
1214
1215 refcount_set(&newsighand->count, 1);
1216
1217 write_lock_irq(&tasklist_lock);
1218 spin_lock(&oldsighand->siglock);
1219 memcpy(newsighand->action, oldsighand->action,
1220 sizeof(newsighand->action));
1221 rcu_assign_pointer(me->sighand, newsighand);
1222 spin_unlock(&oldsighand->siglock);
1223 write_unlock_irq(&tasklist_lock);
1224
1225 __cleanup_sighand(oldsighand);
1226 }
1227 return 0;
1228}
1229
1230char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1231{
1232 task_lock(tsk);
1233 /* Always NUL terminated and zero-padded */
1234 strscpy_pad(buf, tsk->comm, buf_size);
1235 task_unlock(tsk);
1236 return buf;
1237}
1238EXPORT_SYMBOL_GPL(__get_task_comm);
1239
1240/*
1241 * These functions flushes out all traces of the currently running executable
1242 * so that a new one can be started
1243 */
1244
1245void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1246{
1247 task_lock(tsk);
1248 trace_task_rename(tsk, buf);
1249 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1250 task_unlock(tsk);
1251 perf_event_comm(tsk, exec);
1252}
1253
1254/*
1255 * Calling this is the point of no return. None of the failures will be
1256 * seen by userspace since either the process is already taking a fatal
1257 * signal (via de_thread() or coredump), or will have SEGV raised
1258 * (after exec_mmap()) by search_binary_handler (see below).
1259 */
1260int begin_new_exec(struct linux_binprm * bprm)
1261{
1262 struct task_struct *me = current;
1263 int retval;
1264
1265 /* Once we are committed compute the creds */
1266 retval = bprm_creds_from_file(bprm);
1267 if (retval)
1268 return retval;
1269
1270 /*
1271 * Ensure all future errors are fatal.
1272 */
1273 bprm->point_of_no_return = true;
1274
1275 /*
1276 * Make this the only thread in the thread group.
1277 */
1278 retval = de_thread(me);
1279 if (retval)
1280 goto out;
1281
1282 /*
1283 * Cancel any io_uring activity across execve
1284 */
1285 io_uring_task_cancel();
1286
1287 /* Ensure the files table is not shared. */
1288 retval = unshare_files();
1289 if (retval)
1290 goto out;
1291
1292 /*
1293 * Must be called _before_ exec_mmap() as bprm->mm is
1294 * not visible until then. Doing it here also ensures
1295 * we don't race against replace_mm_exe_file().
1296 */
1297 retval = set_mm_exe_file(bprm->mm, bprm->file);
1298 if (retval)
1299 goto out;
1300
1301 /* If the binary is not readable then enforce mm->dumpable=0 */
1302 would_dump(bprm, bprm->file);
1303 if (bprm->have_execfd)
1304 would_dump(bprm, bprm->executable);
1305
1306 /*
1307 * Release all of the old mmap stuff
1308 */
1309 acct_arg_size(bprm, 0);
1310 retval = exec_mmap(bprm->mm);
1311 if (retval)
1312 goto out;
1313
1314 bprm->mm = NULL;
1315
1316 retval = exec_task_namespaces();
1317 if (retval)
1318 goto out_unlock;
1319
1320#ifdef CONFIG_POSIX_TIMERS
1321 spin_lock_irq(&me->sighand->siglock);
1322 posix_cpu_timers_exit(me);
1323 spin_unlock_irq(&me->sighand->siglock);
1324 exit_itimers(me);
1325 flush_itimer_signals();
1326#endif
1327
1328 /*
1329 * Make the signal table private.
1330 */
1331 retval = unshare_sighand(me);
1332 if (retval)
1333 goto out_unlock;
1334
1335 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1336 PF_NOFREEZE | PF_NO_SETAFFINITY);
1337 flush_thread();
1338 me->personality &= ~bprm->per_clear;
1339
1340 clear_syscall_work_syscall_user_dispatch(me);
1341
1342 /*
1343 * We have to apply CLOEXEC before we change whether the process is
1344 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1345 * trying to access the should-be-closed file descriptors of a process
1346 * undergoing exec(2).
1347 */
1348 do_close_on_exec(me->files);
1349
1350 if (bprm->secureexec) {
1351 /* Make sure parent cannot signal privileged process. */
1352 me->pdeath_signal = 0;
1353
1354 /*
1355 * For secureexec, reset the stack limit to sane default to
1356 * avoid bad behavior from the prior rlimits. This has to
1357 * happen before arch_pick_mmap_layout(), which examines
1358 * RLIMIT_STACK, but after the point of no return to avoid
1359 * needing to clean up the change on failure.
1360 */
1361 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1362 bprm->rlim_stack.rlim_cur = _STK_LIM;
1363 }
1364
1365 me->sas_ss_sp = me->sas_ss_size = 0;
1366
1367 /*
1368 * Figure out dumpability. Note that this checking only of current
1369 * is wrong, but userspace depends on it. This should be testing
1370 * bprm->secureexec instead.
1371 */
1372 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1373 !(uid_eq(current_euid(), current_uid()) &&
1374 gid_eq(current_egid(), current_gid())))
1375 set_dumpable(current->mm, suid_dumpable);
1376 else
1377 set_dumpable(current->mm, SUID_DUMP_USER);
1378
1379 perf_event_exec();
1380 __set_task_comm(me, kbasename(bprm->filename), true);
1381
1382 /* An exec changes our domain. We are no longer part of the thread
1383 group */
1384 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1385 flush_signal_handlers(me, 0);
1386
1387 retval = set_cred_ucounts(bprm->cred);
1388 if (retval < 0)
1389 goto out_unlock;
1390
1391 /*
1392 * install the new credentials for this executable
1393 */
1394 security_bprm_committing_creds(bprm);
1395
1396 commit_creds(bprm->cred);
1397 bprm->cred = NULL;
1398
1399 /*
1400 * Disable monitoring for regular users
1401 * when executing setuid binaries. Must
1402 * wait until new credentials are committed
1403 * by commit_creds() above
1404 */
1405 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1406 perf_event_exit_task(me);
1407 /*
1408 * cred_guard_mutex must be held at least to this point to prevent
1409 * ptrace_attach() from altering our determination of the task's
1410 * credentials; any time after this it may be unlocked.
1411 */
1412 security_bprm_committed_creds(bprm);
1413
1414 /* Pass the opened binary to the interpreter. */
1415 if (bprm->have_execfd) {
1416 retval = get_unused_fd_flags(0);
1417 if (retval < 0)
1418 goto out_unlock;
1419 fd_install(retval, bprm->executable);
1420 bprm->executable = NULL;
1421 bprm->execfd = retval;
1422 }
1423 return 0;
1424
1425out_unlock:
1426 up_write(&me->signal->exec_update_lock);
1427 if (!bprm->cred)
1428 mutex_unlock(&me->signal->cred_guard_mutex);
1429
1430out:
1431 return retval;
1432}
1433EXPORT_SYMBOL(begin_new_exec);
1434
1435void would_dump(struct linux_binprm *bprm, struct file *file)
1436{
1437 struct inode *inode = file_inode(file);
1438 struct mnt_idmap *idmap = file_mnt_idmap(file);
1439 if (inode_permission(idmap, inode, MAY_READ) < 0) {
1440 struct user_namespace *old, *user_ns;
1441 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1442
1443 /* Ensure mm->user_ns contains the executable */
1444 user_ns = old = bprm->mm->user_ns;
1445 while ((user_ns != &init_user_ns) &&
1446 !privileged_wrt_inode_uidgid(user_ns, idmap, inode))
1447 user_ns = user_ns->parent;
1448
1449 if (old != user_ns) {
1450 bprm->mm->user_ns = get_user_ns(user_ns);
1451 put_user_ns(old);
1452 }
1453 }
1454}
1455EXPORT_SYMBOL(would_dump);
1456
1457void setup_new_exec(struct linux_binprm * bprm)
1458{
1459 /* Setup things that can depend upon the personality */
1460 struct task_struct *me = current;
1461
1462 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1463
1464 arch_setup_new_exec();
1465
1466 /* Set the new mm task size. We have to do that late because it may
1467 * depend on TIF_32BIT which is only updated in flush_thread() on
1468 * some architectures like powerpc
1469 */
1470 me->mm->task_size = TASK_SIZE;
1471 up_write(&me->signal->exec_update_lock);
1472 mutex_unlock(&me->signal->cred_guard_mutex);
1473}
1474EXPORT_SYMBOL(setup_new_exec);
1475
1476/* Runs immediately before start_thread() takes over. */
1477void finalize_exec(struct linux_binprm *bprm)
1478{
1479 /* Store any stack rlimit changes before starting thread. */
1480 task_lock(current->group_leader);
1481 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1482 task_unlock(current->group_leader);
1483}
1484EXPORT_SYMBOL(finalize_exec);
1485
1486/*
1487 * Prepare credentials and lock ->cred_guard_mutex.
1488 * setup_new_exec() commits the new creds and drops the lock.
1489 * Or, if exec fails before, free_bprm() should release ->cred
1490 * and unlock.
1491 */
1492static int prepare_bprm_creds(struct linux_binprm *bprm)
1493{
1494 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1495 return -ERESTARTNOINTR;
1496
1497 bprm->cred = prepare_exec_creds();
1498 if (likely(bprm->cred))
1499 return 0;
1500
1501 mutex_unlock(&current->signal->cred_guard_mutex);
1502 return -ENOMEM;
1503}
1504
1505/* Matches do_open_execat() */
1506static void do_close_execat(struct file *file)
1507{
1508 if (!file)
1509 return;
1510 allow_write_access(file);
1511 fput(file);
1512}
1513
1514static void free_bprm(struct linux_binprm *bprm)
1515{
1516 if (bprm->mm) {
1517 acct_arg_size(bprm, 0);
1518 mmput(bprm->mm);
1519 }
1520 free_arg_pages(bprm);
1521 if (bprm->cred) {
1522 mutex_unlock(&current->signal->cred_guard_mutex);
1523 abort_creds(bprm->cred);
1524 }
1525 do_close_execat(bprm->file);
1526 if (bprm->executable)
1527 fput(bprm->executable);
1528 /* If a binfmt changed the interp, free it. */
1529 if (bprm->interp != bprm->filename)
1530 kfree(bprm->interp);
1531 kfree(bprm->fdpath);
1532 kfree(bprm);
1533}
1534
1535static struct linux_binprm *alloc_bprm(int fd, struct filename *filename, int flags)
1536{
1537 struct linux_binprm *bprm;
1538 struct file *file;
1539 int retval = -ENOMEM;
1540
1541 file = do_open_execat(fd, filename, flags);
1542 if (IS_ERR(file))
1543 return ERR_CAST(file);
1544
1545 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1546 if (!bprm) {
1547 do_close_execat(file);
1548 return ERR_PTR(-ENOMEM);
1549 }
1550
1551 bprm->file = file;
1552
1553 if (fd == AT_FDCWD || filename->name[0] == '/') {
1554 bprm->filename = filename->name;
1555 } else {
1556 if (filename->name[0] == '\0')
1557 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1558 else
1559 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1560 fd, filename->name);
1561 if (!bprm->fdpath)
1562 goto out_free;
1563
1564 /*
1565 * Record that a name derived from an O_CLOEXEC fd will be
1566 * inaccessible after exec. This allows the code in exec to
1567 * choose to fail when the executable is not mmaped into the
1568 * interpreter and an open file descriptor is not passed to
1569 * the interpreter. This makes for a better user experience
1570 * than having the interpreter start and then immediately fail
1571 * when it finds the executable is inaccessible.
1572 */
1573 if (get_close_on_exec(fd))
1574 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1575
1576 bprm->filename = bprm->fdpath;
1577 }
1578 bprm->interp = bprm->filename;
1579
1580 retval = bprm_mm_init(bprm);
1581 if (!retval)
1582 return bprm;
1583
1584out_free:
1585 free_bprm(bprm);
1586 return ERR_PTR(retval);
1587}
1588
1589int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1590{
1591 /* If a binfmt changed the interp, free it first. */
1592 if (bprm->interp != bprm->filename)
1593 kfree(bprm->interp);
1594 bprm->interp = kstrdup(interp, GFP_KERNEL);
1595 if (!bprm->interp)
1596 return -ENOMEM;
1597 return 0;
1598}
1599EXPORT_SYMBOL(bprm_change_interp);
1600
1601/*
1602 * determine how safe it is to execute the proposed program
1603 * - the caller must hold ->cred_guard_mutex to protect against
1604 * PTRACE_ATTACH or seccomp thread-sync
1605 */
1606static void check_unsafe_exec(struct linux_binprm *bprm)
1607{
1608 struct task_struct *p = current, *t;
1609 unsigned n_fs;
1610
1611 if (p->ptrace)
1612 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1613
1614 /*
1615 * This isn't strictly necessary, but it makes it harder for LSMs to
1616 * mess up.
1617 */
1618 if (task_no_new_privs(current))
1619 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1620
1621 /*
1622 * If another task is sharing our fs, we cannot safely
1623 * suid exec because the differently privileged task
1624 * will be able to manipulate the current directory, etc.
1625 * It would be nice to force an unshare instead...
1626 */
1627 n_fs = 1;
1628 spin_lock(&p->fs->lock);
1629 rcu_read_lock();
1630 for_other_threads(p, t) {
1631 if (t->fs == p->fs)
1632 n_fs++;
1633 }
1634 rcu_read_unlock();
1635
1636 /* "users" and "in_exec" locked for copy_fs() */
1637 if (p->fs->users > n_fs)
1638 bprm->unsafe |= LSM_UNSAFE_SHARE;
1639 else
1640 p->fs->in_exec = 1;
1641 spin_unlock(&p->fs->lock);
1642}
1643
1644static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1645{
1646 /* Handle suid and sgid on files */
1647 struct mnt_idmap *idmap;
1648 struct inode *inode = file_inode(file);
1649 unsigned int mode;
1650 vfsuid_t vfsuid;
1651 vfsgid_t vfsgid;
1652
1653 if (!mnt_may_suid(file->f_path.mnt))
1654 return;
1655
1656 if (task_no_new_privs(current))
1657 return;
1658
1659 mode = READ_ONCE(inode->i_mode);
1660 if (!(mode & (S_ISUID|S_ISGID)))
1661 return;
1662
1663 idmap = file_mnt_idmap(file);
1664
1665 /* Be careful if suid/sgid is set */
1666 inode_lock(inode);
1667
1668 /* reload atomically mode/uid/gid now that lock held */
1669 mode = inode->i_mode;
1670 vfsuid = i_uid_into_vfsuid(idmap, inode);
1671 vfsgid = i_gid_into_vfsgid(idmap, inode);
1672 inode_unlock(inode);
1673
1674 /* We ignore suid/sgid if there are no mappings for them in the ns */
1675 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1676 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1677 return;
1678
1679 if (mode & S_ISUID) {
1680 bprm->per_clear |= PER_CLEAR_ON_SETID;
1681 bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1682 }
1683
1684 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1685 bprm->per_clear |= PER_CLEAR_ON_SETID;
1686 bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1687 }
1688}
1689
1690/*
1691 * Compute brpm->cred based upon the final binary.
1692 */
1693static int bprm_creds_from_file(struct linux_binprm *bprm)
1694{
1695 /* Compute creds based on which file? */
1696 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1697
1698 bprm_fill_uid(bprm, file);
1699 return security_bprm_creds_from_file(bprm, file);
1700}
1701
1702/*
1703 * Fill the binprm structure from the inode.
1704 * Read the first BINPRM_BUF_SIZE bytes
1705 *
1706 * This may be called multiple times for binary chains (scripts for example).
1707 */
1708static int prepare_binprm(struct linux_binprm *bprm)
1709{
1710 loff_t pos = 0;
1711
1712 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1713 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1714}
1715
1716/*
1717 * Arguments are '\0' separated strings found at the location bprm->p
1718 * points to; chop off the first by relocating brpm->p to right after
1719 * the first '\0' encountered.
1720 */
1721int remove_arg_zero(struct linux_binprm *bprm)
1722{
1723 unsigned long offset;
1724 char *kaddr;
1725 struct page *page;
1726
1727 if (!bprm->argc)
1728 return 0;
1729
1730 do {
1731 offset = bprm->p & ~PAGE_MASK;
1732 page = get_arg_page(bprm, bprm->p, 0);
1733 if (!page)
1734 return -EFAULT;
1735 kaddr = kmap_local_page(page);
1736
1737 for (; offset < PAGE_SIZE && kaddr[offset];
1738 offset++, bprm->p++)
1739 ;
1740
1741 kunmap_local(kaddr);
1742 put_arg_page(page);
1743 } while (offset == PAGE_SIZE);
1744
1745 bprm->p++;
1746 bprm->argc--;
1747
1748 return 0;
1749}
1750EXPORT_SYMBOL(remove_arg_zero);
1751
1752#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1753/*
1754 * cycle the list of binary formats handler, until one recognizes the image
1755 */
1756static int search_binary_handler(struct linux_binprm *bprm)
1757{
1758 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1759 struct linux_binfmt *fmt;
1760 int retval;
1761
1762 retval = prepare_binprm(bprm);
1763 if (retval < 0)
1764 return retval;
1765
1766 retval = security_bprm_check(bprm);
1767 if (retval)
1768 return retval;
1769
1770 retval = -ENOENT;
1771 retry:
1772 read_lock(&binfmt_lock);
1773 list_for_each_entry(fmt, &formats, lh) {
1774 if (!try_module_get(fmt->module))
1775 continue;
1776 read_unlock(&binfmt_lock);
1777
1778 retval = fmt->load_binary(bprm);
1779
1780 read_lock(&binfmt_lock);
1781 put_binfmt(fmt);
1782 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1783 read_unlock(&binfmt_lock);
1784 return retval;
1785 }
1786 }
1787 read_unlock(&binfmt_lock);
1788
1789 if (need_retry) {
1790 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1791 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1792 return retval;
1793 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1794 return retval;
1795 need_retry = false;
1796 goto retry;
1797 }
1798
1799 return retval;
1800}
1801
1802/* binfmt handlers will call back into begin_new_exec() on success. */
1803static int exec_binprm(struct linux_binprm *bprm)
1804{
1805 pid_t old_pid, old_vpid;
1806 int ret, depth;
1807
1808 /* Need to fetch pid before load_binary changes it */
1809 old_pid = current->pid;
1810 rcu_read_lock();
1811 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1812 rcu_read_unlock();
1813
1814 /* This allows 4 levels of binfmt rewrites before failing hard. */
1815 for (depth = 0;; depth++) {
1816 struct file *exec;
1817 if (depth > 5)
1818 return -ELOOP;
1819
1820 ret = search_binary_handler(bprm);
1821 if (ret < 0)
1822 return ret;
1823 if (!bprm->interpreter)
1824 break;
1825
1826 exec = bprm->file;
1827 bprm->file = bprm->interpreter;
1828 bprm->interpreter = NULL;
1829
1830 allow_write_access(exec);
1831 if (unlikely(bprm->have_execfd)) {
1832 if (bprm->executable) {
1833 fput(exec);
1834 return -ENOEXEC;
1835 }
1836 bprm->executable = exec;
1837 } else
1838 fput(exec);
1839 }
1840
1841 audit_bprm(bprm);
1842 trace_sched_process_exec(current, old_pid, bprm);
1843 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1844 proc_exec_connector(current);
1845 return 0;
1846}
1847
1848static int bprm_execve(struct linux_binprm *bprm)
1849{
1850 int retval;
1851
1852 retval = prepare_bprm_creds(bprm);
1853 if (retval)
1854 return retval;
1855
1856 /*
1857 * Check for unsafe execution states before exec_binprm(), which
1858 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1859 * where setuid-ness is evaluated.
1860 */
1861 check_unsafe_exec(bprm);
1862 current->in_execve = 1;
1863 sched_mm_cid_before_execve(current);
1864
1865 sched_exec();
1866
1867 /* Set the unchanging part of bprm->cred */
1868 retval = security_bprm_creds_for_exec(bprm);
1869 if (retval)
1870 goto out;
1871
1872 retval = exec_binprm(bprm);
1873 if (retval < 0)
1874 goto out;
1875
1876 sched_mm_cid_after_execve(current);
1877 /* execve succeeded */
1878 current->fs->in_exec = 0;
1879 current->in_execve = 0;
1880 rseq_execve(current);
1881 user_events_execve(current);
1882 acct_update_integrals(current);
1883 task_numa_free(current, false);
1884 return retval;
1885
1886out:
1887 /*
1888 * If past the point of no return ensure the code never
1889 * returns to the userspace process. Use an existing fatal
1890 * signal if present otherwise terminate the process with
1891 * SIGSEGV.
1892 */
1893 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1894 force_fatal_sig(SIGSEGV);
1895
1896 sched_mm_cid_after_execve(current);
1897 current->fs->in_exec = 0;
1898 current->in_execve = 0;
1899
1900 return retval;
1901}
1902
1903static int do_execveat_common(int fd, struct filename *filename,
1904 struct user_arg_ptr argv,
1905 struct user_arg_ptr envp,
1906 int flags)
1907{
1908 struct linux_binprm *bprm;
1909 int retval;
1910
1911 if (IS_ERR(filename))
1912 return PTR_ERR(filename);
1913
1914 /*
1915 * We move the actual failure in case of RLIMIT_NPROC excess from
1916 * set*uid() to execve() because too many poorly written programs
1917 * don't check setuid() return code. Here we additionally recheck
1918 * whether NPROC limit is still exceeded.
1919 */
1920 if ((current->flags & PF_NPROC_EXCEEDED) &&
1921 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1922 retval = -EAGAIN;
1923 goto out_ret;
1924 }
1925
1926 /* We're below the limit (still or again), so we don't want to make
1927 * further execve() calls fail. */
1928 current->flags &= ~PF_NPROC_EXCEEDED;
1929
1930 bprm = alloc_bprm(fd, filename, flags);
1931 if (IS_ERR(bprm)) {
1932 retval = PTR_ERR(bprm);
1933 goto out_ret;
1934 }
1935
1936 retval = count(argv, MAX_ARG_STRINGS);
1937 if (retval == 0)
1938 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1939 current->comm, bprm->filename);
1940 if (retval < 0)
1941 goto out_free;
1942 bprm->argc = retval;
1943
1944 retval = count(envp, MAX_ARG_STRINGS);
1945 if (retval < 0)
1946 goto out_free;
1947 bprm->envc = retval;
1948
1949 retval = bprm_stack_limits(bprm);
1950 if (retval < 0)
1951 goto out_free;
1952
1953 retval = copy_string_kernel(bprm->filename, bprm);
1954 if (retval < 0)
1955 goto out_free;
1956 bprm->exec = bprm->p;
1957
1958 retval = copy_strings(bprm->envc, envp, bprm);
1959 if (retval < 0)
1960 goto out_free;
1961
1962 retval = copy_strings(bprm->argc, argv, bprm);
1963 if (retval < 0)
1964 goto out_free;
1965
1966 /*
1967 * When argv is empty, add an empty string ("") as argv[0] to
1968 * ensure confused userspace programs that start processing
1969 * from argv[1] won't end up walking envp. See also
1970 * bprm_stack_limits().
1971 */
1972 if (bprm->argc == 0) {
1973 retval = copy_string_kernel("", bprm);
1974 if (retval < 0)
1975 goto out_free;
1976 bprm->argc = 1;
1977 }
1978
1979 retval = bprm_execve(bprm);
1980out_free:
1981 free_bprm(bprm);
1982
1983out_ret:
1984 putname(filename);
1985 return retval;
1986}
1987
1988int kernel_execve(const char *kernel_filename,
1989 const char *const *argv, const char *const *envp)
1990{
1991 struct filename *filename;
1992 struct linux_binprm *bprm;
1993 int fd = AT_FDCWD;
1994 int retval;
1995
1996 /* It is non-sense for kernel threads to call execve */
1997 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1998 return -EINVAL;
1999
2000 filename = getname_kernel(kernel_filename);
2001 if (IS_ERR(filename))
2002 return PTR_ERR(filename);
2003
2004 bprm = alloc_bprm(fd, filename, 0);
2005 if (IS_ERR(bprm)) {
2006 retval = PTR_ERR(bprm);
2007 goto out_ret;
2008 }
2009
2010 retval = count_strings_kernel(argv);
2011 if (WARN_ON_ONCE(retval == 0))
2012 retval = -EINVAL;
2013 if (retval < 0)
2014 goto out_free;
2015 bprm->argc = retval;
2016
2017 retval = count_strings_kernel(envp);
2018 if (retval < 0)
2019 goto out_free;
2020 bprm->envc = retval;
2021
2022 retval = bprm_stack_limits(bprm);
2023 if (retval < 0)
2024 goto out_free;
2025
2026 retval = copy_string_kernel(bprm->filename, bprm);
2027 if (retval < 0)
2028 goto out_free;
2029 bprm->exec = bprm->p;
2030
2031 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2032 if (retval < 0)
2033 goto out_free;
2034
2035 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2036 if (retval < 0)
2037 goto out_free;
2038
2039 retval = bprm_execve(bprm);
2040out_free:
2041 free_bprm(bprm);
2042out_ret:
2043 putname(filename);
2044 return retval;
2045}
2046
2047static int do_execve(struct filename *filename,
2048 const char __user *const __user *__argv,
2049 const char __user *const __user *__envp)
2050{
2051 struct user_arg_ptr argv = { .ptr.native = __argv };
2052 struct user_arg_ptr envp = { .ptr.native = __envp };
2053 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2054}
2055
2056static int do_execveat(int fd, struct filename *filename,
2057 const char __user *const __user *__argv,
2058 const char __user *const __user *__envp,
2059 int flags)
2060{
2061 struct user_arg_ptr argv = { .ptr.native = __argv };
2062 struct user_arg_ptr envp = { .ptr.native = __envp };
2063
2064 return do_execveat_common(fd, filename, argv, envp, flags);
2065}
2066
2067#ifdef CONFIG_COMPAT
2068static int compat_do_execve(struct filename *filename,
2069 const compat_uptr_t __user *__argv,
2070 const compat_uptr_t __user *__envp)
2071{
2072 struct user_arg_ptr argv = {
2073 .is_compat = true,
2074 .ptr.compat = __argv,
2075 };
2076 struct user_arg_ptr envp = {
2077 .is_compat = true,
2078 .ptr.compat = __envp,
2079 };
2080 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2081}
2082
2083static int compat_do_execveat(int fd, struct filename *filename,
2084 const compat_uptr_t __user *__argv,
2085 const compat_uptr_t __user *__envp,
2086 int flags)
2087{
2088 struct user_arg_ptr argv = {
2089 .is_compat = true,
2090 .ptr.compat = __argv,
2091 };
2092 struct user_arg_ptr envp = {
2093 .is_compat = true,
2094 .ptr.compat = __envp,
2095 };
2096 return do_execveat_common(fd, filename, argv, envp, flags);
2097}
2098#endif
2099
2100void set_binfmt(struct linux_binfmt *new)
2101{
2102 struct mm_struct *mm = current->mm;
2103
2104 if (mm->binfmt)
2105 module_put(mm->binfmt->module);
2106
2107 mm->binfmt = new;
2108 if (new)
2109 __module_get(new->module);
2110}
2111EXPORT_SYMBOL(set_binfmt);
2112
2113/*
2114 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2115 */
2116void set_dumpable(struct mm_struct *mm, int value)
2117{
2118 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2119 return;
2120
2121 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2122}
2123
2124SYSCALL_DEFINE3(execve,
2125 const char __user *, filename,
2126 const char __user *const __user *, argv,
2127 const char __user *const __user *, envp)
2128{
2129 return do_execve(getname(filename), argv, envp);
2130}
2131
2132SYSCALL_DEFINE5(execveat,
2133 int, fd, const char __user *, filename,
2134 const char __user *const __user *, argv,
2135 const char __user *const __user *, envp,
2136 int, flags)
2137{
2138 return do_execveat(fd,
2139 getname_uflags(filename, flags),
2140 argv, envp, flags);
2141}
2142
2143#ifdef CONFIG_COMPAT
2144COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2145 const compat_uptr_t __user *, argv,
2146 const compat_uptr_t __user *, envp)
2147{
2148 return compat_do_execve(getname(filename), argv, envp);
2149}
2150
2151COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2152 const char __user *, filename,
2153 const compat_uptr_t __user *, argv,
2154 const compat_uptr_t __user *, envp,
2155 int, flags)
2156{
2157 return compat_do_execveat(fd,
2158 getname_uflags(filename, flags),
2159 argv, envp, flags);
2160}
2161#endif
2162
2163#ifdef CONFIG_SYSCTL
2164
2165static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2166 void *buffer, size_t *lenp, loff_t *ppos)
2167{
2168 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2169
2170 if (!error)
2171 validate_coredump_safety();
2172 return error;
2173}
2174
2175static struct ctl_table fs_exec_sysctls[] = {
2176 {
2177 .procname = "suid_dumpable",
2178 .data = &suid_dumpable,
2179 .maxlen = sizeof(int),
2180 .mode = 0644,
2181 .proc_handler = proc_dointvec_minmax_coredump,
2182 .extra1 = SYSCTL_ZERO,
2183 .extra2 = SYSCTL_TWO,
2184 },
2185};
2186
2187static int __init init_fs_exec_sysctls(void)
2188{
2189 register_sysctl_init("fs", fs_exec_sysctls);
2190 return 0;
2191}
2192
2193fs_initcall(init_fs_exec_sysctls);
2194#endif /* CONFIG_SYSCTL */