projects/linux·blob v6.0-rc7 · refs

2158 lines · 51.0 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/vmacache.h>
32#include <linux/stat.h>
33#include <linux/fcntl.h>
34#include <linux/swap.h>
35#include <linux/string.h>
36#include <linux/init.h>
37#include <linux/sched/mm.h>
38#include <linux/sched/coredump.h>
39#include <linux/sched/signal.h>
40#include <linux/sched/numa_balancing.h>
41#include <linux/sched/task.h>
42#include <linux/pagemap.h>
43#include <linux/perf_event.h>
44#include <linux/highmem.h>
45#include <linux/spinlock.h>
46#include <linux/key.h>
47#include <linux/personality.h>
48#include <linux/binfmts.h>
49#include <linux/utsname.h>
50#include <linux/pid_namespace.h>
51#include <linux/module.h>
52#include <linux/namei.h>
53#include <linux/mount.h>
54#include <linux/security.h>
55#include <linux/syscalls.h>
56#include <linux/tsacct_kern.h>
57#include <linux/cn_proc.h>
58#include <linux/audit.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#include <linux/syscall_user_dispatch.h>
67#include <linux/coredump.h>
68
69#include <linux/uaccess.h>
70#include <asm/mmu_context.h>
71#include <asm/tlb.h>
72
73#include <trace/events/task.h>
74#include "internal.h"
75
76#include <trace/events/sched.h>
77
78static int bprm_creds_from_file(struct linux_binprm *bprm);
79
80int suid_dumpable = 0;
81
82static LIST_HEAD(formats);
83static DEFINE_RWLOCK(binfmt_lock);
84
85void __register_binfmt(struct linux_binfmt * fmt, int insert)
86{
87 write_lock(&binfmt_lock);
88 insert ? list_add(&fmt->lh, &formats) :
89 list_add_tail(&fmt->lh, &formats);
90 write_unlock(&binfmt_lock);
91}
92
93EXPORT_SYMBOL(__register_binfmt);
94
95void unregister_binfmt(struct linux_binfmt * fmt)
96{
97 write_lock(&binfmt_lock);
98 list_del(&fmt->lh);
99 write_unlock(&binfmt_lock);
100}
101
102EXPORT_SYMBOL(unregister_binfmt);
103
104static inline void put_binfmt(struct linux_binfmt * fmt)
105{
106 module_put(fmt->module);
107}
108
109bool path_noexec(const struct path *path)
110{
111 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
112 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
113}
114
115#ifdef CONFIG_USELIB
116/*
117 * Note that a shared library must be both readable and executable due to
118 * security reasons.
119 *
120 * Also note that we take the address to load from the file itself.
121 */
122SYSCALL_DEFINE1(uselib, const char __user *, library)
123{
124 struct linux_binfmt *fmt;
125 struct file *file;
126 struct filename *tmp = getname(library);
127 int error = PTR_ERR(tmp);
128 static const struct open_flags uselib_flags = {
129 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
130 .acc_mode = MAY_READ | MAY_EXEC,
131 .intent = LOOKUP_OPEN,
132 .lookup_flags = LOOKUP_FOLLOW,
133 };
134
135 if (IS_ERR(tmp))
136 goto out;
137
138 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
139 putname(tmp);
140 error = PTR_ERR(file);
141 if (IS_ERR(file))
142 goto out;
143
144 /*
145 * may_open() has already checked for this, so it should be
146 * impossible to trip now. But we need to be extra cautious
147 * and check again at the very end too.
148 */
149 error = -EACCES;
150 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
151 path_noexec(&file->f_path)))
152 goto exit;
153
154 fsnotify_open(file);
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 int ret;
203 unsigned int gup_flags = FOLL_FORCE;
204
205#ifdef CONFIG_STACK_GROWSUP
206 if (write) {
207 ret = expand_downwards(bprm->vma, pos);
208 if (ret < 0)
209 return NULL;
210 }
211#endif
212
213 if (write)
214 gup_flags |= FOLL_WRITE;
215
216 /*
217 * We are doing an exec(). 'current' is the process
218 * doing the exec and bprm->mm is the new process's mm.
219 */
220 mmap_read_lock(bprm->mm);
221 ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
222 &page, NULL, NULL);
223 mmap_read_unlock(bprm->mm);
224 if (ret <= 0)
225 return NULL;
226
227 if (write)
228 acct_arg_size(bprm, vma_pages(bprm->vma));
229
230 return page;
231}
232
233static void put_arg_page(struct page *page)
234{
235 put_page(page);
236}
237
238static void free_arg_pages(struct linux_binprm *bprm)
239{
240}
241
242static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
243 struct page *page)
244{
245 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
246}
247
248static int __bprm_mm_init(struct linux_binprm *bprm)
249{
250 int err;
251 struct vm_area_struct *vma = NULL;
252 struct mm_struct *mm = bprm->mm;
253
254 bprm->vma = vma = vm_area_alloc(mm);
255 if (!vma)
256 return -ENOMEM;
257 vma_set_anonymous(vma);
258
259 if (mmap_write_lock_killable(mm)) {
260 err = -EINTR;
261 goto err_free;
262 }
263
264 /*
265 * Place the stack at the largest stack address the architecture
266 * supports. Later, we'll move this to an appropriate place. We don't
267 * use STACK_TOP because that can depend on attributes which aren't
268 * configured yet.
269 */
270 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
271 vma->vm_end = STACK_TOP_MAX;
272 vma->vm_start = vma->vm_end - PAGE_SIZE;
273 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
274 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
275
276 err = insert_vm_struct(mm, vma);
277 if (err)
278 goto err;
279
280 mm->stack_vm = mm->total_vm = 1;
281 mmap_write_unlock(mm);
282 bprm->p = vma->vm_end - sizeof(void *);
283 return 0;
284err:
285 mmap_write_unlock(mm);
286err_free:
287 bprm->vma = NULL;
288 vm_area_free(vma);
289 return err;
290}
291
292static bool valid_arg_len(struct linux_binprm *bprm, long len)
293{
294 return len <= MAX_ARG_STRLEN;
295}
296
297#else
298
299static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
300{
301}
302
303static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
304 int write)
305{
306 struct page *page;
307
308 page = bprm->page[pos / PAGE_SIZE];
309 if (!page && write) {
310 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
311 if (!page)
312 return NULL;
313 bprm->page[pos / PAGE_SIZE] = page;
314 }
315
316 return page;
317}
318
319static void put_arg_page(struct page *page)
320{
321}
322
323static void free_arg_page(struct linux_binprm *bprm, int i)
324{
325 if (bprm->page[i]) {
326 __free_page(bprm->page[i]);
327 bprm->page[i] = NULL;
328 }
329}
330
331static void free_arg_pages(struct linux_binprm *bprm)
332{
333 int i;
334
335 for (i = 0; i < MAX_ARG_PAGES; i++)
336 free_arg_page(bprm, i);
337}
338
339static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
340 struct page *page)
341{
342}
343
344static int __bprm_mm_init(struct linux_binprm *bprm)
345{
346 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
347 return 0;
348}
349
350static bool valid_arg_len(struct linux_binprm *bprm, long len)
351{
352 return len <= bprm->p;
353}
354
355#endif /* CONFIG_MMU */
356
357/*
358 * Create a new mm_struct and populate it with a temporary stack
359 * vm_area_struct. We don't have enough context at this point to set the stack
360 * flags, permissions, and offset, so we use temporary values. We'll update
361 * them later in setup_arg_pages().
362 */
363static int bprm_mm_init(struct linux_binprm *bprm)
364{
365 int err;
366 struct mm_struct *mm = NULL;
367
368 bprm->mm = mm = mm_alloc();
369 err = -ENOMEM;
370 if (!mm)
371 goto err;
372
373 /* Save current stack limit for all calculations made during exec. */
374 task_lock(current->group_leader);
375 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
376 task_unlock(current->group_leader);
377
378 err = __bprm_mm_init(bprm);
379 if (err)
380 goto err;
381
382 return 0;
383
384err:
385 if (mm) {
386 bprm->mm = NULL;
387 mmdrop(mm);
388 }
389
390 return err;
391}
392
393struct user_arg_ptr {
394#ifdef CONFIG_COMPAT
395 bool is_compat;
396#endif
397 union {
398 const char __user *const __user *native;
399#ifdef CONFIG_COMPAT
400 const compat_uptr_t __user *compat;
401#endif
402 } ptr;
403};
404
405static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
406{
407 const char __user *native;
408
409#ifdef CONFIG_COMPAT
410 if (unlikely(argv.is_compat)) {
411 compat_uptr_t compat;
412
413 if (get_user(compat, argv.ptr.compat + nr))
414 return ERR_PTR(-EFAULT);
415
416 return compat_ptr(compat);
417 }
418#endif
419
420 if (get_user(native, argv.ptr.native + nr))
421 return ERR_PTR(-EFAULT);
422
423 return native;
424}
425
426/*
427 * count() counts the number of strings in array ARGV.
428 */
429static int count(struct user_arg_ptr argv, int max)
430{
431 int i = 0;
432
433 if (argv.ptr.native != NULL) {
434 for (;;) {
435 const char __user *p = get_user_arg_ptr(argv, i);
436
437 if (!p)
438 break;
439
440 if (IS_ERR(p))
441 return -EFAULT;
442
443 if (i >= max)
444 return -E2BIG;
445 ++i;
446
447 if (fatal_signal_pending(current))
448 return -ERESTARTNOHAND;
449 cond_resched();
450 }
451 }
452 return i;
453}
454
455static int count_strings_kernel(const char *const *argv)
456{
457 int i;
458
459 if (!argv)
460 return 0;
461
462 for (i = 0; argv[i]; ++i) {
463 if (i >= MAX_ARG_STRINGS)
464 return -E2BIG;
465 if (fatal_signal_pending(current))
466 return -ERESTARTNOHAND;
467 cond_resched();
468 }
469 return i;
470}
471
472static int bprm_stack_limits(struct linux_binprm *bprm)
473{
474 unsigned long limit, ptr_size;
475
476 /*
477 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
478 * (whichever is smaller) for the argv+env strings.
479 * This ensures that:
480 * - the remaining binfmt code will not run out of stack space,
481 * - the program will have a reasonable amount of stack left
482 * to work from.
483 */
484 limit = _STK_LIM / 4 * 3;
485 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
486 /*
487 * We've historically supported up to 32 pages (ARG_MAX)
488 * of argument strings even with small stacks
489 */
490 limit = max_t(unsigned long, limit, ARG_MAX);
491 /*
492 * We must account for the size of all the argv and envp pointers to
493 * the argv and envp strings, since they will also take up space in
494 * the stack. They aren't stored until much later when we can't
495 * signal to the parent that the child has run out of stack space.
496 * Instead, calculate it here so it's possible to fail gracefully.
497 *
498 * In the case of argc = 0, make sure there is space for adding a
499 * empty string (which will bump argc to 1), to ensure confused
500 * userspace programs don't start processing from argv[1], thinking
501 * argc can never be 0, to keep them from walking envp by accident.
502 * See do_execveat_common().
503 */
504 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
505 if (limit <= ptr_size)
506 return -E2BIG;
507 limit -= ptr_size;
508
509 bprm->argmin = bprm->p - limit;
510 return 0;
511}
512
513/*
514 * 'copy_strings()' copies argument/environment strings from the old
515 * processes's memory to the new process's stack. The call to get_user_pages()
516 * ensures the destination page is created and not swapped out.
517 */
518static int copy_strings(int argc, struct user_arg_ptr argv,
519 struct linux_binprm *bprm)
520{
521 struct page *kmapped_page = NULL;
522 char *kaddr = NULL;
523 unsigned long kpos = 0;
524 int ret;
525
526 while (argc-- > 0) {
527 const char __user *str;
528 int len;
529 unsigned long pos;
530
531 ret = -EFAULT;
532 str = get_user_arg_ptr(argv, argc);
533 if (IS_ERR(str))
534 goto out;
535
536 len = strnlen_user(str, MAX_ARG_STRLEN);
537 if (!len)
538 goto out;
539
540 ret = -E2BIG;
541 if (!valid_arg_len(bprm, len))
542 goto out;
543
544 /* We're going to work our way backwards. */
545 pos = bprm->p;
546 str += len;
547 bprm->p -= len;
548#ifdef CONFIG_MMU
549 if (bprm->p < bprm->argmin)
550 goto out;
551#endif
552
553 while (len > 0) {
554 int offset, bytes_to_copy;
555
556 if (fatal_signal_pending(current)) {
557 ret = -ERESTARTNOHAND;
558 goto out;
559 }
560 cond_resched();
561
562 offset = pos % PAGE_SIZE;
563 if (offset == 0)
564 offset = PAGE_SIZE;
565
566 bytes_to_copy = offset;
567 if (bytes_to_copy > len)
568 bytes_to_copy = len;
569
570 offset -= bytes_to_copy;
571 pos -= bytes_to_copy;
572 str -= bytes_to_copy;
573 len -= bytes_to_copy;
574
575 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
576 struct page *page;
577
578 page = get_arg_page(bprm, pos, 1);
579 if (!page) {
580 ret = -E2BIG;
581 goto out;
582 }
583
584 if (kmapped_page) {
585 flush_dcache_page(kmapped_page);
586 kunmap_local(kaddr);
587 put_arg_page(kmapped_page);
588 }
589 kmapped_page = page;
590 kaddr = kmap_local_page(kmapped_page);
591 kpos = pos & PAGE_MASK;
592 flush_arg_page(bprm, kpos, kmapped_page);
593 }
594 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
595 ret = -EFAULT;
596 goto out;
597 }
598 }
599 }
600 ret = 0;
601out:
602 if (kmapped_page) {
603 flush_dcache_page(kmapped_page);
604 kunmap_local(kaddr);
605 put_arg_page(kmapped_page);
606 }
607 return ret;
608}
609
610/*
611 * Copy and argument/environment string from the kernel to the processes stack.
612 */
613int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
614{
615 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
616 unsigned long pos = bprm->p;
617
618 if (len == 0)
619 return -EFAULT;
620 if (!valid_arg_len(bprm, len))
621 return -E2BIG;
622
623 /* We're going to work our way backwards. */
624 arg += len;
625 bprm->p -= len;
626 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
627 return -E2BIG;
628
629 while (len > 0) {
630 unsigned int bytes_to_copy = min_t(unsigned int, len,
631 min_not_zero(offset_in_page(pos), PAGE_SIZE));
632 struct page *page;
633
634 pos -= bytes_to_copy;
635 arg -= bytes_to_copy;
636 len -= bytes_to_copy;
637
638 page = get_arg_page(bprm, pos, 1);
639 if (!page)
640 return -E2BIG;
641 flush_arg_page(bprm, pos & PAGE_MASK, page);
642 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
643 put_arg_page(page);
644 }
645
646 return 0;
647}
648EXPORT_SYMBOL(copy_string_kernel);
649
650static int copy_strings_kernel(int argc, const char *const *argv,
651 struct linux_binprm *bprm)
652{
653 while (argc-- > 0) {
654 int ret = copy_string_kernel(argv[argc], bprm);
655 if (ret < 0)
656 return ret;
657 if (fatal_signal_pending(current))
658 return -ERESTARTNOHAND;
659 cond_resched();
660 }
661 return 0;
662}
663
664#ifdef CONFIG_MMU
665
666/*
667 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
668 * the binfmt code determines where the new stack should reside, we shift it to
669 * its final location. The process proceeds as follows:
670 *
671 * 1) Use shift to calculate the new vma endpoints.
672 * 2) Extend vma to cover both the old and new ranges. This ensures the
673 * arguments passed to subsequent functions are consistent.
674 * 3) Move vma's page tables to the new range.
675 * 4) Free up any cleared pgd range.
676 * 5) Shrink the vma to cover only the new range.
677 */
678static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
679{
680 struct mm_struct *mm = vma->vm_mm;
681 unsigned long old_start = vma->vm_start;
682 unsigned long old_end = vma->vm_end;
683 unsigned long length = old_end - old_start;
684 unsigned long new_start = old_start - shift;
685 unsigned long new_end = old_end - shift;
686 struct mmu_gather tlb;
687
688 BUG_ON(new_start > new_end);
689
690 /*
691 * ensure there are no vmas between where we want to go
692 * and where we are
693 */
694 if (vma != find_vma(mm, new_start))
695 return -EFAULT;
696
697 /*
698 * cover the whole range: [new_start, old_end)
699 */
700 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
701 return -ENOMEM;
702
703 /*
704 * move the page tables downwards, on failure we rely on
705 * process cleanup to remove whatever mess we made.
706 */
707 if (length != move_page_tables(vma, old_start,
708 vma, new_start, length, false))
709 return -ENOMEM;
710
711 lru_add_drain();
712 tlb_gather_mmu(&tlb, mm);
713 if (new_end > old_start) {
714 /*
715 * when the old and new regions overlap clear from new_end.
716 */
717 free_pgd_range(&tlb, new_end, old_end, new_end,
718 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
719 } else {
720 /*
721 * otherwise, clean from old_start; this is done to not touch
722 * the address space in [new_end, old_start) some architectures
723 * have constraints on va-space that make this illegal (IA64) -
724 * for the others its just a little faster.
725 */
726 free_pgd_range(&tlb, old_start, old_end, new_end,
727 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
728 }
729 tlb_finish_mmu(&tlb);
730
731 /*
732 * Shrink the vma to just the new range. Always succeeds.
733 */
734 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
735
736 return 0;
737}
738
739/*
740 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
741 * the stack is optionally relocated, and some extra space is added.
742 */
743int setup_arg_pages(struct linux_binprm *bprm,
744 unsigned long stack_top,
745 int executable_stack)
746{
747 unsigned long ret;
748 unsigned long stack_shift;
749 struct mm_struct *mm = current->mm;
750 struct vm_area_struct *vma = bprm->vma;
751 struct vm_area_struct *prev = NULL;
752 unsigned long vm_flags;
753 unsigned long stack_base;
754 unsigned long stack_size;
755 unsigned long stack_expand;
756 unsigned long rlim_stack;
757 struct mmu_gather tlb;
758
759#ifdef CONFIG_STACK_GROWSUP
760 /* Limit stack size */
761 stack_base = bprm->rlim_stack.rlim_max;
762
763 stack_base = calc_max_stack_size(stack_base);
764
765 /* Add space for stack randomization. */
766 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
767
768 /* Make sure we didn't let the argument array grow too large. */
769 if (vma->vm_end - vma->vm_start > stack_base)
770 return -ENOMEM;
771
772 stack_base = PAGE_ALIGN(stack_top - stack_base);
773
774 stack_shift = vma->vm_start - stack_base;
775 mm->arg_start = bprm->p - stack_shift;
776 bprm->p = vma->vm_end - stack_shift;
777#else
778 stack_top = arch_align_stack(stack_top);
779 stack_top = PAGE_ALIGN(stack_top);
780
781 if (unlikely(stack_top < mmap_min_addr) ||
782 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
783 return -ENOMEM;
784
785 stack_shift = vma->vm_end - stack_top;
786
787 bprm->p -= stack_shift;
788 mm->arg_start = bprm->p;
789#endif
790
791 if (bprm->loader)
792 bprm->loader -= stack_shift;
793 bprm->exec -= stack_shift;
794
795 if (mmap_write_lock_killable(mm))
796 return -EINTR;
797
798 vm_flags = VM_STACK_FLAGS;
799
800 /*
801 * Adjust stack execute permissions; explicitly enable for
802 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
803 * (arch default) otherwise.
804 */
805 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
806 vm_flags |= VM_EXEC;
807 else if (executable_stack == EXSTACK_DISABLE_X)
808 vm_flags &= ~VM_EXEC;
809 vm_flags |= mm->def_flags;
810 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
811
812 tlb_gather_mmu(&tlb, mm);
813 ret = mprotect_fixup(&tlb, vma, &prev, vma->vm_start, vma->vm_end,
814 vm_flags);
815 tlb_finish_mmu(&tlb);
816
817 if (ret)
818 goto out_unlock;
819 BUG_ON(prev != vma);
820
821 if (unlikely(vm_flags & VM_EXEC)) {
822 pr_warn_once("process '%pD4' started with executable stack\n",
823 bprm->file);
824 }
825
826 /* Move stack pages down in memory. */
827 if (stack_shift) {
828 ret = shift_arg_pages(vma, stack_shift);
829 if (ret)
830 goto out_unlock;
831 }
832
833 /* mprotect_fixup is overkill to remove the temporary stack flags */
834 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
835
836 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
837 stack_size = vma->vm_end - vma->vm_start;
838 /*
839 * Align this down to a page boundary as expand_stack
840 * will align it up.
841 */
842 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
843#ifdef CONFIG_STACK_GROWSUP
844 if (stack_size + stack_expand > rlim_stack)
845 stack_base = vma->vm_start + rlim_stack;
846 else
847 stack_base = vma->vm_end + stack_expand;
848#else
849 if (stack_size + stack_expand > rlim_stack)
850 stack_base = vma->vm_end - rlim_stack;
851 else
852 stack_base = vma->vm_start - stack_expand;
853#endif
854 current->mm->start_stack = bprm->p;
855 ret = expand_stack(vma, stack_base);
856 if (ret)
857 ret = -EFAULT;
858
859out_unlock:
860 mmap_write_unlock(mm);
861 return ret;
862}
863EXPORT_SYMBOL(setup_arg_pages);
864
865#else
866
867/*
868 * Transfer the program arguments and environment from the holding pages
869 * onto the stack. The provided stack pointer is adjusted accordingly.
870 */
871int transfer_args_to_stack(struct linux_binprm *bprm,
872 unsigned long *sp_location)
873{
874 unsigned long index, stop, sp;
875 int ret = 0;
876
877 stop = bprm->p >> PAGE_SHIFT;
878 sp = *sp_location;
879
880 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
881 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
882 char *src = kmap_local_page(bprm->page[index]) + offset;
883 sp -= PAGE_SIZE - offset;
884 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
885 ret = -EFAULT;
886 kunmap_local(src);
887 if (ret)
888 goto out;
889 }
890
891 *sp_location = sp;
892
893out:
894 return ret;
895}
896EXPORT_SYMBOL(transfer_args_to_stack);
897
898#endif /* CONFIG_MMU */
899
900static struct file *do_open_execat(int fd, struct filename *name, int flags)
901{
902 struct file *file;
903 int err;
904 struct open_flags open_exec_flags = {
905 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
906 .acc_mode = MAY_EXEC,
907 .intent = LOOKUP_OPEN,
908 .lookup_flags = LOOKUP_FOLLOW,
909 };
910
911 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
912 return ERR_PTR(-EINVAL);
913 if (flags & AT_SYMLINK_NOFOLLOW)
914 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
915 if (flags & AT_EMPTY_PATH)
916 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
917
918 file = do_filp_open(fd, name, &open_exec_flags);
919 if (IS_ERR(file))
920 goto out;
921
922 /*
923 * may_open() has already checked for this, so it should be
924 * impossible to trip now. But we need to be extra cautious
925 * and check again at the very end too.
926 */
927 err = -EACCES;
928 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
929 path_noexec(&file->f_path)))
930 goto exit;
931
932 err = deny_write_access(file);
933 if (err)
934 goto exit;
935
936 if (name->name[0] != '\0')
937 fsnotify_open(file);
938
939out:
940 return file;
941
942exit:
943 fput(file);
944 return ERR_PTR(err);
945}
946
947struct file *open_exec(const char *name)
948{
949 struct filename *filename = getname_kernel(name);
950 struct file *f = ERR_CAST(filename);
951
952 if (!IS_ERR(filename)) {
953 f = do_open_execat(AT_FDCWD, filename, 0);
954 putname(filename);
955 }
956 return f;
957}
958EXPORT_SYMBOL(open_exec);
959
960#if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
961 defined(CONFIG_BINFMT_ELF_FDPIC)
962ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
963{
964 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
965 if (res > 0)
966 flush_icache_user_range(addr, addr + len);
967 return res;
968}
969EXPORT_SYMBOL(read_code);
970#endif
971
972/*
973 * Maps the mm_struct mm into the current task struct.
974 * On success, this function returns with exec_update_lock
975 * held for writing.
976 */
977static int exec_mmap(struct mm_struct *mm)
978{
979 struct task_struct *tsk;
980 struct mm_struct *old_mm, *active_mm;
981 int ret;
982
983 /* Notify parent that we're no longer interested in the old VM */
984 tsk = current;
985 old_mm = current->mm;
986 exec_mm_release(tsk, old_mm);
987 if (old_mm)
988 sync_mm_rss(old_mm);
989
990 ret = down_write_killable(&tsk->signal->exec_update_lock);
991 if (ret)
992 return ret;
993
994 if (old_mm) {
995 /*
996 * If there is a pending fatal signal perhaps a signal
997 * whose default action is to create a coredump get
998 * out and die instead of going through with the exec.
999 */
1000 ret = mmap_read_lock_killable(old_mm);
1001 if (ret) {
1002 up_write(&tsk->signal->exec_update_lock);
1003 return ret;
1004 }
1005 }
1006
1007 task_lock(tsk);
1008 membarrier_exec_mmap(mm);
1009
1010 local_irq_disable();
1011 active_mm = tsk->active_mm;
1012 tsk->active_mm = mm;
1013 tsk->mm = mm;
1014 /*
1015 * This prevents preemption while active_mm is being loaded and
1016 * it and mm are being updated, which could cause problems for
1017 * lazy tlb mm refcounting when these are updated by context
1018 * switches. Not all architectures can handle irqs off over
1019 * activate_mm yet.
1020 */
1021 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1022 local_irq_enable();
1023 activate_mm(active_mm, mm);
1024 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1025 local_irq_enable();
1026 tsk->mm->vmacache_seqnum = 0;
1027 vmacache_flush(tsk);
1028 task_unlock(tsk);
1029 if (old_mm) {
1030 mmap_read_unlock(old_mm);
1031 BUG_ON(active_mm != old_mm);
1032 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1033 mm_update_next_owner(old_mm);
1034 mmput(old_mm);
1035 return 0;
1036 }
1037 mmdrop(active_mm);
1038 return 0;
1039}
1040
1041static int de_thread(struct task_struct *tsk)
1042{
1043 struct signal_struct *sig = tsk->signal;
1044 struct sighand_struct *oldsighand = tsk->sighand;
1045 spinlock_t *lock = &oldsighand->siglock;
1046
1047 if (thread_group_empty(tsk))
1048 goto no_thread_group;
1049
1050 /*
1051 * Kill all other threads in the thread group.
1052 */
1053 spin_lock_irq(lock);
1054 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1055 /*
1056 * Another group action in progress, just
1057 * return so that the signal is processed.
1058 */
1059 spin_unlock_irq(lock);
1060 return -EAGAIN;
1061 }
1062
1063 sig->group_exec_task = tsk;
1064 sig->notify_count = zap_other_threads(tsk);
1065 if (!thread_group_leader(tsk))
1066 sig->notify_count--;
1067
1068 while (sig->notify_count) {
1069 __set_current_state(TASK_KILLABLE);
1070 spin_unlock_irq(lock);
1071 schedule();
1072 if (__fatal_signal_pending(tsk))
1073 goto killed;
1074 spin_lock_irq(lock);
1075 }
1076 spin_unlock_irq(lock);
1077
1078 /*
1079 * At this point all other threads have exited, all we have to
1080 * do is to wait for the thread group leader to become inactive,
1081 * and to assume its PID:
1082 */
1083 if (!thread_group_leader(tsk)) {
1084 struct task_struct *leader = tsk->group_leader;
1085
1086 for (;;) {
1087 cgroup_threadgroup_change_begin(tsk);
1088 write_lock_irq(&tasklist_lock);
1089 /*
1090 * Do this under tasklist_lock to ensure that
1091 * exit_notify() can't miss ->group_exec_task
1092 */
1093 sig->notify_count = -1;
1094 if (likely(leader->exit_state))
1095 break;
1096 __set_current_state(TASK_KILLABLE);
1097 write_unlock_irq(&tasklist_lock);
1098 cgroup_threadgroup_change_end(tsk);
1099 schedule();
1100 if (__fatal_signal_pending(tsk))
1101 goto killed;
1102 }
1103
1104 /*
1105 * The only record we have of the real-time age of a
1106 * process, regardless of execs it's done, is start_time.
1107 * All the past CPU time is accumulated in signal_struct
1108 * from sister threads now dead. But in this non-leader
1109 * exec, nothing survives from the original leader thread,
1110 * whose birth marks the true age of this process now.
1111 * When we take on its identity by switching to its PID, we
1112 * also take its birthdate (always earlier than our own).
1113 */
1114 tsk->start_time = leader->start_time;
1115 tsk->start_boottime = leader->start_boottime;
1116
1117 BUG_ON(!same_thread_group(leader, tsk));
1118 /*
1119 * An exec() starts a new thread group with the
1120 * TGID of the previous thread group. Rehash the
1121 * two threads with a switched PID, and release
1122 * the former thread group leader:
1123 */
1124
1125 /* Become a process group leader with the old leader's pid.
1126 * The old leader becomes a thread of the this thread group.
1127 */
1128 exchange_tids(tsk, leader);
1129 transfer_pid(leader, tsk, PIDTYPE_TGID);
1130 transfer_pid(leader, tsk, PIDTYPE_PGID);
1131 transfer_pid(leader, tsk, PIDTYPE_SID);
1132
1133 list_replace_rcu(&leader->tasks, &tsk->tasks);
1134 list_replace_init(&leader->sibling, &tsk->sibling);
1135
1136 tsk->group_leader = tsk;
1137 leader->group_leader = tsk;
1138
1139 tsk->exit_signal = SIGCHLD;
1140 leader->exit_signal = -1;
1141
1142 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1143 leader->exit_state = EXIT_DEAD;
1144
1145 /*
1146 * We are going to release_task()->ptrace_unlink() silently,
1147 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1148 * the tracer won't block again waiting for this thread.
1149 */
1150 if (unlikely(leader->ptrace))
1151 __wake_up_parent(leader, leader->parent);
1152 write_unlock_irq(&tasklist_lock);
1153 cgroup_threadgroup_change_end(tsk);
1154
1155 release_task(leader);
1156 }
1157
1158 sig->group_exec_task = NULL;
1159 sig->notify_count = 0;
1160
1161no_thread_group:
1162 /* we have changed execution domain */
1163 tsk->exit_signal = SIGCHLD;
1164
1165 BUG_ON(!thread_group_leader(tsk));
1166 return 0;
1167
1168killed:
1169 /* protects against exit_notify() and __exit_signal() */
1170 read_lock(&tasklist_lock);
1171 sig->group_exec_task = NULL;
1172 sig->notify_count = 0;
1173 read_unlock(&tasklist_lock);
1174 return -EAGAIN;
1175}
1176
1177
1178/*
1179 * This function makes sure the current process has its own signal table,
1180 * so that flush_signal_handlers can later reset the handlers without
1181 * disturbing other processes. (Other processes might share the signal
1182 * table via the CLONE_SIGHAND option to clone().)
1183 */
1184static int unshare_sighand(struct task_struct *me)
1185{
1186 struct sighand_struct *oldsighand = me->sighand;
1187
1188 if (refcount_read(&oldsighand->count) != 1) {
1189 struct sighand_struct *newsighand;
1190 /*
1191 * This ->sighand is shared with the CLONE_SIGHAND
1192 * but not CLONE_THREAD task, switch to the new one.
1193 */
1194 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1195 if (!newsighand)
1196 return -ENOMEM;
1197
1198 refcount_set(&newsighand->count, 1);
1199 memcpy(newsighand->action, oldsighand->action,
1200 sizeof(newsighand->action));
1201
1202 write_lock_irq(&tasklist_lock);
1203 spin_lock(&oldsighand->siglock);
1204 rcu_assign_pointer(me->sighand, newsighand);
1205 spin_unlock(&oldsighand->siglock);
1206 write_unlock_irq(&tasklist_lock);
1207
1208 __cleanup_sighand(oldsighand);
1209 }
1210 return 0;
1211}
1212
1213char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1214{
1215 task_lock(tsk);
1216 /* Always NUL terminated and zero-padded */
1217 strscpy_pad(buf, tsk->comm, buf_size);
1218 task_unlock(tsk);
1219 return buf;
1220}
1221EXPORT_SYMBOL_GPL(__get_task_comm);
1222
1223/*
1224 * These functions flushes out all traces of the currently running executable
1225 * so that a new one can be started
1226 */
1227
1228void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1229{
1230 task_lock(tsk);
1231 trace_task_rename(tsk, buf);
1232 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1233 task_unlock(tsk);
1234 perf_event_comm(tsk, exec);
1235}
1236
1237/*
1238 * Calling this is the point of no return. None of the failures will be
1239 * seen by userspace since either the process is already taking a fatal
1240 * signal (via de_thread() or coredump), or will have SEGV raised
1241 * (after exec_mmap()) by search_binary_handler (see below).
1242 */
1243int begin_new_exec(struct linux_binprm * bprm)
1244{
1245 struct task_struct *me = current;
1246 int retval;
1247
1248 /* Once we are committed compute the creds */
1249 retval = bprm_creds_from_file(bprm);
1250 if (retval)
1251 return retval;
1252
1253 /*
1254 * Ensure all future errors are fatal.
1255 */
1256 bprm->point_of_no_return = true;
1257
1258 /*
1259 * Make this the only thread in the thread group.
1260 */
1261 retval = de_thread(me);
1262 if (retval)
1263 goto out;
1264
1265 /*
1266 * Cancel any io_uring activity across execve
1267 */
1268 io_uring_task_cancel();
1269
1270 /* Ensure the files table is not shared. */
1271 retval = unshare_files();
1272 if (retval)
1273 goto out;
1274
1275 /*
1276 * Must be called _before_ exec_mmap() as bprm->mm is
1277 * not visible until then. This also enables the update
1278 * to be lockless.
1279 */
1280 retval = set_mm_exe_file(bprm->mm, bprm->file);
1281 if (retval)
1282 goto out;
1283
1284 /* If the binary is not readable then enforce mm->dumpable=0 */
1285 would_dump(bprm, bprm->file);
1286 if (bprm->have_execfd)
1287 would_dump(bprm, bprm->executable);
1288
1289 /*
1290 * Release all of the old mmap stuff
1291 */
1292 acct_arg_size(bprm, 0);
1293 retval = exec_mmap(bprm->mm);
1294 if (retval)
1295 goto out;
1296
1297 bprm->mm = NULL;
1298
1299#ifdef CONFIG_POSIX_TIMERS
1300 spin_lock_irq(&me->sighand->siglock);
1301 posix_cpu_timers_exit(me);
1302 spin_unlock_irq(&me->sighand->siglock);
1303 exit_itimers(me);
1304 flush_itimer_signals();
1305#endif
1306
1307 /*
1308 * Make the signal table private.
1309 */
1310 retval = unshare_sighand(me);
1311 if (retval)
1312 goto out_unlock;
1313
1314 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1315 PF_NOFREEZE | PF_NO_SETAFFINITY);
1316 flush_thread();
1317 me->personality &= ~bprm->per_clear;
1318
1319 clear_syscall_work_syscall_user_dispatch(me);
1320
1321 /*
1322 * We have to apply CLOEXEC before we change whether the process is
1323 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1324 * trying to access the should-be-closed file descriptors of a process
1325 * undergoing exec(2).
1326 */
1327 do_close_on_exec(me->files);
1328
1329 if (bprm->secureexec) {
1330 /* Make sure parent cannot signal privileged process. */
1331 me->pdeath_signal = 0;
1332
1333 /*
1334 * For secureexec, reset the stack limit to sane default to
1335 * avoid bad behavior from the prior rlimits. This has to
1336 * happen before arch_pick_mmap_layout(), which examines
1337 * RLIMIT_STACK, but after the point of no return to avoid
1338 * needing to clean up the change on failure.
1339 */
1340 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1341 bprm->rlim_stack.rlim_cur = _STK_LIM;
1342 }
1343
1344 me->sas_ss_sp = me->sas_ss_size = 0;
1345
1346 /*
1347 * Figure out dumpability. Note that this checking only of current
1348 * is wrong, but userspace depends on it. This should be testing
1349 * bprm->secureexec instead.
1350 */
1351 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1352 !(uid_eq(current_euid(), current_uid()) &&
1353 gid_eq(current_egid(), current_gid())))
1354 set_dumpable(current->mm, suid_dumpable);
1355 else
1356 set_dumpable(current->mm, SUID_DUMP_USER);
1357
1358 perf_event_exec();
1359 __set_task_comm(me, kbasename(bprm->filename), true);
1360
1361 /* An exec changes our domain. We are no longer part of the thread
1362 group */
1363 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1364 flush_signal_handlers(me, 0);
1365
1366 retval = set_cred_ucounts(bprm->cred);
1367 if (retval < 0)
1368 goto out_unlock;
1369
1370 /*
1371 * install the new credentials for this executable
1372 */
1373 security_bprm_committing_creds(bprm);
1374
1375 commit_creds(bprm->cred);
1376 bprm->cred = NULL;
1377
1378 /*
1379 * Disable monitoring for regular users
1380 * when executing setuid binaries. Must
1381 * wait until new credentials are committed
1382 * by commit_creds() above
1383 */
1384 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1385 perf_event_exit_task(me);
1386 /*
1387 * cred_guard_mutex must be held at least to this point to prevent
1388 * ptrace_attach() from altering our determination of the task's
1389 * credentials; any time after this it may be unlocked.
1390 */
1391 security_bprm_committed_creds(bprm);
1392
1393 /* Pass the opened binary to the interpreter. */
1394 if (bprm->have_execfd) {
1395 retval = get_unused_fd_flags(0);
1396 if (retval < 0)
1397 goto out_unlock;
1398 fd_install(retval, bprm->executable);
1399 bprm->executable = NULL;
1400 bprm->execfd = retval;
1401 }
1402 return 0;
1403
1404out_unlock:
1405 up_write(&me->signal->exec_update_lock);
1406out:
1407 return retval;
1408}
1409EXPORT_SYMBOL(begin_new_exec);
1410
1411void would_dump(struct linux_binprm *bprm, struct file *file)
1412{
1413 struct inode *inode = file_inode(file);
1414 struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1415 if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1416 struct user_namespace *old, *user_ns;
1417 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1418
1419 /* Ensure mm->user_ns contains the executable */
1420 user_ns = old = bprm->mm->user_ns;
1421 while ((user_ns != &init_user_ns) &&
1422 !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1423 user_ns = user_ns->parent;
1424
1425 if (old != user_ns) {
1426 bprm->mm->user_ns = get_user_ns(user_ns);
1427 put_user_ns(old);
1428 }
1429 }
1430}
1431EXPORT_SYMBOL(would_dump);
1432
1433void setup_new_exec(struct linux_binprm * bprm)
1434{
1435 /* Setup things that can depend upon the personality */
1436 struct task_struct *me = current;
1437
1438 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1439
1440 arch_setup_new_exec();
1441
1442 /* Set the new mm task size. We have to do that late because it may
1443 * depend on TIF_32BIT which is only updated in flush_thread() on
1444 * some architectures like powerpc
1445 */
1446 me->mm->task_size = TASK_SIZE;
1447 up_write(&me->signal->exec_update_lock);
1448 mutex_unlock(&me->signal->cred_guard_mutex);
1449}
1450EXPORT_SYMBOL(setup_new_exec);
1451
1452/* Runs immediately before start_thread() takes over. */
1453void finalize_exec(struct linux_binprm *bprm)
1454{
1455 /* Store any stack rlimit changes before starting thread. */
1456 task_lock(current->group_leader);
1457 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1458 task_unlock(current->group_leader);
1459}
1460EXPORT_SYMBOL(finalize_exec);
1461
1462/*
1463 * Prepare credentials and lock ->cred_guard_mutex.
1464 * setup_new_exec() commits the new creds and drops the lock.
1465 * Or, if exec fails before, free_bprm() should release ->cred
1466 * and unlock.
1467 */
1468static int prepare_bprm_creds(struct linux_binprm *bprm)
1469{
1470 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1471 return -ERESTARTNOINTR;
1472
1473 bprm->cred = prepare_exec_creds();
1474 if (likely(bprm->cred))
1475 return 0;
1476
1477 mutex_unlock(&current->signal->cred_guard_mutex);
1478 return -ENOMEM;
1479}
1480
1481static void free_bprm(struct linux_binprm *bprm)
1482{
1483 if (bprm->mm) {
1484 acct_arg_size(bprm, 0);
1485 mmput(bprm->mm);
1486 }
1487 free_arg_pages(bprm);
1488 if (bprm->cred) {
1489 mutex_unlock(&current->signal->cred_guard_mutex);
1490 abort_creds(bprm->cred);
1491 }
1492 if (bprm->file) {
1493 allow_write_access(bprm->file);
1494 fput(bprm->file);
1495 }
1496 if (bprm->executable)
1497 fput(bprm->executable);
1498 /* If a binfmt changed the interp, free it. */
1499 if (bprm->interp != bprm->filename)
1500 kfree(bprm->interp);
1501 kfree(bprm->fdpath);
1502 kfree(bprm);
1503}
1504
1505static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1506{
1507 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1508 int retval = -ENOMEM;
1509 if (!bprm)
1510 goto out;
1511
1512 if (fd == AT_FDCWD || filename->name[0] == '/') {
1513 bprm->filename = filename->name;
1514 } else {
1515 if (filename->name[0] == '\0')
1516 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1517 else
1518 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1519 fd, filename->name);
1520 if (!bprm->fdpath)
1521 goto out_free;
1522
1523 bprm->filename = bprm->fdpath;
1524 }
1525 bprm->interp = bprm->filename;
1526
1527 retval = bprm_mm_init(bprm);
1528 if (retval)
1529 goto out_free;
1530 return bprm;
1531
1532out_free:
1533 free_bprm(bprm);
1534out:
1535 return ERR_PTR(retval);
1536}
1537
1538int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1539{
1540 /* If a binfmt changed the interp, free it first. */
1541 if (bprm->interp != bprm->filename)
1542 kfree(bprm->interp);
1543 bprm->interp = kstrdup(interp, GFP_KERNEL);
1544 if (!bprm->interp)
1545 return -ENOMEM;
1546 return 0;
1547}
1548EXPORT_SYMBOL(bprm_change_interp);
1549
1550/*
1551 * determine how safe it is to execute the proposed program
1552 * - the caller must hold ->cred_guard_mutex to protect against
1553 * PTRACE_ATTACH or seccomp thread-sync
1554 */
1555static void check_unsafe_exec(struct linux_binprm *bprm)
1556{
1557 struct task_struct *p = current, *t;
1558 unsigned n_fs;
1559
1560 if (p->ptrace)
1561 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1562
1563 /*
1564 * This isn't strictly necessary, but it makes it harder for LSMs to
1565 * mess up.
1566 */
1567 if (task_no_new_privs(current))
1568 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1569
1570 t = p;
1571 n_fs = 1;
1572 spin_lock(&p->fs->lock);
1573 rcu_read_lock();
1574 while_each_thread(p, t) {
1575 if (t->fs == p->fs)
1576 n_fs++;
1577 }
1578 rcu_read_unlock();
1579
1580 if (p->fs->users > n_fs)
1581 bprm->unsafe |= LSM_UNSAFE_SHARE;
1582 else
1583 p->fs->in_exec = 1;
1584 spin_unlock(&p->fs->lock);
1585}
1586
1587static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1588{
1589 /* Handle suid and sgid on files */
1590 struct user_namespace *mnt_userns;
1591 struct inode *inode;
1592 unsigned int mode;
1593 kuid_t uid;
1594 kgid_t gid;
1595
1596 if (!mnt_may_suid(file->f_path.mnt))
1597 return;
1598
1599 if (task_no_new_privs(current))
1600 return;
1601
1602 inode = file->f_path.dentry->d_inode;
1603 mode = READ_ONCE(inode->i_mode);
1604 if (!(mode & (S_ISUID|S_ISGID)))
1605 return;
1606
1607 mnt_userns = file_mnt_user_ns(file);
1608
1609 /* Be careful if suid/sgid is set */
1610 inode_lock(inode);
1611
1612 /* reload atomically mode/uid/gid now that lock held */
1613 mode = inode->i_mode;
1614 uid = i_uid_into_mnt(mnt_userns, inode);
1615 gid = i_gid_into_mnt(mnt_userns, inode);
1616 inode_unlock(inode);
1617
1618 /* We ignore suid/sgid if there are no mappings for them in the ns */
1619 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1620 !kgid_has_mapping(bprm->cred->user_ns, gid))
1621 return;
1622
1623 if (mode & S_ISUID) {
1624 bprm->per_clear |= PER_CLEAR_ON_SETID;
1625 bprm->cred->euid = uid;
1626 }
1627
1628 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1629 bprm->per_clear |= PER_CLEAR_ON_SETID;
1630 bprm->cred->egid = gid;
1631 }
1632}
1633
1634/*
1635 * Compute brpm->cred based upon the final binary.
1636 */
1637static int bprm_creds_from_file(struct linux_binprm *bprm)
1638{
1639 /* Compute creds based on which file? */
1640 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1641
1642 bprm_fill_uid(bprm, file);
1643 return security_bprm_creds_from_file(bprm, file);
1644}
1645
1646/*
1647 * Fill the binprm structure from the inode.
1648 * Read the first BINPRM_BUF_SIZE bytes
1649 *
1650 * This may be called multiple times for binary chains (scripts for example).
1651 */
1652static int prepare_binprm(struct linux_binprm *bprm)
1653{
1654 loff_t pos = 0;
1655
1656 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1657 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1658}
1659
1660/*
1661 * Arguments are '\0' separated strings found at the location bprm->p
1662 * points to; chop off the first by relocating brpm->p to right after
1663 * the first '\0' encountered.
1664 */
1665int remove_arg_zero(struct linux_binprm *bprm)
1666{
1667 int ret = 0;
1668 unsigned long offset;
1669 char *kaddr;
1670 struct page *page;
1671
1672 if (!bprm->argc)
1673 return 0;
1674
1675 do {
1676 offset = bprm->p & ~PAGE_MASK;
1677 page = get_arg_page(bprm, bprm->p, 0);
1678 if (!page) {
1679 ret = -EFAULT;
1680 goto out;
1681 }
1682 kaddr = kmap_local_page(page);
1683
1684 for (; offset < PAGE_SIZE && kaddr[offset];
1685 offset++, bprm->p++)
1686 ;
1687
1688 kunmap_local(kaddr);
1689 put_arg_page(page);
1690 } while (offset == PAGE_SIZE);
1691
1692 bprm->p++;
1693 bprm->argc--;
1694 ret = 0;
1695
1696out:
1697 return ret;
1698}
1699EXPORT_SYMBOL(remove_arg_zero);
1700
1701#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1702/*
1703 * cycle the list of binary formats handler, until one recognizes the image
1704 */
1705static int search_binary_handler(struct linux_binprm *bprm)
1706{
1707 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1708 struct linux_binfmt *fmt;
1709 int retval;
1710
1711 retval = prepare_binprm(bprm);
1712 if (retval < 0)
1713 return retval;
1714
1715 retval = security_bprm_check(bprm);
1716 if (retval)
1717 return retval;
1718
1719 retval = -ENOENT;
1720 retry:
1721 read_lock(&binfmt_lock);
1722 list_for_each_entry(fmt, &formats, lh) {
1723 if (!try_module_get(fmt->module))
1724 continue;
1725 read_unlock(&binfmt_lock);
1726
1727 retval = fmt->load_binary(bprm);
1728
1729 read_lock(&binfmt_lock);
1730 put_binfmt(fmt);
1731 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1732 read_unlock(&binfmt_lock);
1733 return retval;
1734 }
1735 }
1736 read_unlock(&binfmt_lock);
1737
1738 if (need_retry) {
1739 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1740 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1741 return retval;
1742 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1743 return retval;
1744 need_retry = false;
1745 goto retry;
1746 }
1747
1748 return retval;
1749}
1750
1751static int exec_binprm(struct linux_binprm *bprm)
1752{
1753 pid_t old_pid, old_vpid;
1754 int ret, depth;
1755
1756 /* Need to fetch pid before load_binary changes it */
1757 old_pid = current->pid;
1758 rcu_read_lock();
1759 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1760 rcu_read_unlock();
1761
1762 /* This allows 4 levels of binfmt rewrites before failing hard. */
1763 for (depth = 0;; depth++) {
1764 struct file *exec;
1765 if (depth > 5)
1766 return -ELOOP;
1767
1768 ret = search_binary_handler(bprm);
1769 if (ret < 0)
1770 return ret;
1771 if (!bprm->interpreter)
1772 break;
1773
1774 exec = bprm->file;
1775 bprm->file = bprm->interpreter;
1776 bprm->interpreter = NULL;
1777
1778 allow_write_access(exec);
1779 if (unlikely(bprm->have_execfd)) {
1780 if (bprm->executable) {
1781 fput(exec);
1782 return -ENOEXEC;
1783 }
1784 bprm->executable = exec;
1785 } else
1786 fput(exec);
1787 }
1788
1789 audit_bprm(bprm);
1790 trace_sched_process_exec(current, old_pid, bprm);
1791 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1792 proc_exec_connector(current);
1793 return 0;
1794}
1795
1796/*
1797 * sys_execve() executes a new program.
1798 */
1799static int bprm_execve(struct linux_binprm *bprm,
1800 int fd, struct filename *filename, int flags)
1801{
1802 struct file *file;
1803 int retval;
1804
1805 retval = prepare_bprm_creds(bprm);
1806 if (retval)
1807 return retval;
1808
1809 check_unsafe_exec(bprm);
1810 current->in_execve = 1;
1811
1812 file = do_open_execat(fd, filename, flags);
1813 retval = PTR_ERR(file);
1814 if (IS_ERR(file))
1815 goto out_unmark;
1816
1817 sched_exec();
1818
1819 bprm->file = file;
1820 /*
1821 * Record that a name derived from an O_CLOEXEC fd will be
1822 * inaccessible after exec. This allows the code in exec to
1823 * choose to fail when the executable is not mmaped into the
1824 * interpreter and an open file descriptor is not passed to
1825 * the interpreter. This makes for a better user experience
1826 * than having the interpreter start and then immediately fail
1827 * when it finds the executable is inaccessible.
1828 */
1829 if (bprm->fdpath && get_close_on_exec(fd))
1830 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1831
1832 /* Set the unchanging part of bprm->cred */
1833 retval = security_bprm_creds_for_exec(bprm);
1834 if (retval)
1835 goto out;
1836
1837 retval = exec_binprm(bprm);
1838 if (retval < 0)
1839 goto out;
1840
1841 /* execve succeeded */
1842 current->fs->in_exec = 0;
1843 current->in_execve = 0;
1844 rseq_execve(current);
1845 acct_update_integrals(current);
1846 task_numa_free(current, false);
1847 return retval;
1848
1849out:
1850 /*
1851 * If past the point of no return ensure the code never
1852 * returns to the userspace process. Use an existing fatal
1853 * signal if present otherwise terminate the process with
1854 * SIGSEGV.
1855 */
1856 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1857 force_fatal_sig(SIGSEGV);
1858
1859out_unmark:
1860 current->fs->in_exec = 0;
1861 current->in_execve = 0;
1862
1863 return retval;
1864}
1865
1866static int do_execveat_common(int fd, struct filename *filename,
1867 struct user_arg_ptr argv,
1868 struct user_arg_ptr envp,
1869 int flags)
1870{
1871 struct linux_binprm *bprm;
1872 int retval;
1873
1874 if (IS_ERR(filename))
1875 return PTR_ERR(filename);
1876
1877 /*
1878 * We move the actual failure in case of RLIMIT_NPROC excess from
1879 * set*uid() to execve() because too many poorly written programs
1880 * don't check setuid() return code. Here we additionally recheck
1881 * whether NPROC limit is still exceeded.
1882 */
1883 if ((current->flags & PF_NPROC_EXCEEDED) &&
1884 is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1885 retval = -EAGAIN;
1886 goto out_ret;
1887 }
1888
1889 /* We're below the limit (still or again), so we don't want to make
1890 * further execve() calls fail. */
1891 current->flags &= ~PF_NPROC_EXCEEDED;
1892
1893 bprm = alloc_bprm(fd, filename);
1894 if (IS_ERR(bprm)) {
1895 retval = PTR_ERR(bprm);
1896 goto out_ret;
1897 }
1898
1899 retval = count(argv, MAX_ARG_STRINGS);
1900 if (retval == 0)
1901 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1902 current->comm, bprm->filename);
1903 if (retval < 0)
1904 goto out_free;
1905 bprm->argc = retval;
1906
1907 retval = count(envp, MAX_ARG_STRINGS);
1908 if (retval < 0)
1909 goto out_free;
1910 bprm->envc = retval;
1911
1912 retval = bprm_stack_limits(bprm);
1913 if (retval < 0)
1914 goto out_free;
1915
1916 retval = copy_string_kernel(bprm->filename, bprm);
1917 if (retval < 0)
1918 goto out_free;
1919 bprm->exec = bprm->p;
1920
1921 retval = copy_strings(bprm->envc, envp, bprm);
1922 if (retval < 0)
1923 goto out_free;
1924
1925 retval = copy_strings(bprm->argc, argv, bprm);
1926 if (retval < 0)
1927 goto out_free;
1928
1929 /*
1930 * When argv is empty, add an empty string ("") as argv[0] to
1931 * ensure confused userspace programs that start processing
1932 * from argv[1] won't end up walking envp. See also
1933 * bprm_stack_limits().
1934 */
1935 if (bprm->argc == 0) {
1936 retval = copy_string_kernel("", bprm);
1937 if (retval < 0)
1938 goto out_free;
1939 bprm->argc = 1;
1940 }
1941
1942 retval = bprm_execve(bprm, fd, filename, flags);
1943out_free:
1944 free_bprm(bprm);
1945
1946out_ret:
1947 putname(filename);
1948 return retval;
1949}
1950
1951int kernel_execve(const char *kernel_filename,
1952 const char *const *argv, const char *const *envp)
1953{
1954 struct filename *filename;
1955 struct linux_binprm *bprm;
1956 int fd = AT_FDCWD;
1957 int retval;
1958
1959 /* It is non-sense for kernel threads to call execve */
1960 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1961 return -EINVAL;
1962
1963 filename = getname_kernel(kernel_filename);
1964 if (IS_ERR(filename))
1965 return PTR_ERR(filename);
1966
1967 bprm = alloc_bprm(fd, filename);
1968 if (IS_ERR(bprm)) {
1969 retval = PTR_ERR(bprm);
1970 goto out_ret;
1971 }
1972
1973 retval = count_strings_kernel(argv);
1974 if (WARN_ON_ONCE(retval == 0))
1975 retval = -EINVAL;
1976 if (retval < 0)
1977 goto out_free;
1978 bprm->argc = retval;
1979
1980 retval = count_strings_kernel(envp);
1981 if (retval < 0)
1982 goto out_free;
1983 bprm->envc = retval;
1984
1985 retval = bprm_stack_limits(bprm);
1986 if (retval < 0)
1987 goto out_free;
1988
1989 retval = copy_string_kernel(bprm->filename, bprm);
1990 if (retval < 0)
1991 goto out_free;
1992 bprm->exec = bprm->p;
1993
1994 retval = copy_strings_kernel(bprm->envc, envp, bprm);
1995 if (retval < 0)
1996 goto out_free;
1997
1998 retval = copy_strings_kernel(bprm->argc, argv, bprm);
1999 if (retval < 0)
2000 goto out_free;
2001
2002 retval = bprm_execve(bprm, fd, filename, 0);
2003out_free:
2004 free_bprm(bprm);
2005out_ret:
2006 putname(filename);
2007 return retval;
2008}
2009
2010static int do_execve(struct filename *filename,
2011 const char __user *const __user *__argv,
2012 const char __user *const __user *__envp)
2013{
2014 struct user_arg_ptr argv = { .ptr.native = __argv };
2015 struct user_arg_ptr envp = { .ptr.native = __envp };
2016 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2017}
2018
2019static int do_execveat(int fd, struct filename *filename,
2020 const char __user *const __user *__argv,
2021 const char __user *const __user *__envp,
2022 int flags)
2023{
2024 struct user_arg_ptr argv = { .ptr.native = __argv };
2025 struct user_arg_ptr envp = { .ptr.native = __envp };
2026
2027 return do_execveat_common(fd, filename, argv, envp, flags);
2028}
2029
2030#ifdef CONFIG_COMPAT
2031static int compat_do_execve(struct filename *filename,
2032 const compat_uptr_t __user *__argv,
2033 const compat_uptr_t __user *__envp)
2034{
2035 struct user_arg_ptr argv = {
2036 .is_compat = true,
2037 .ptr.compat = __argv,
2038 };
2039 struct user_arg_ptr envp = {
2040 .is_compat = true,
2041 .ptr.compat = __envp,
2042 };
2043 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2044}
2045
2046static int compat_do_execveat(int fd, struct filename *filename,
2047 const compat_uptr_t __user *__argv,
2048 const compat_uptr_t __user *__envp,
2049 int flags)
2050{
2051 struct user_arg_ptr argv = {
2052 .is_compat = true,
2053 .ptr.compat = __argv,
2054 };
2055 struct user_arg_ptr envp = {
2056 .is_compat = true,
2057 .ptr.compat = __envp,
2058 };
2059 return do_execveat_common(fd, filename, argv, envp, flags);
2060}
2061#endif
2062
2063void set_binfmt(struct linux_binfmt *new)
2064{
2065 struct mm_struct *mm = current->mm;
2066
2067 if (mm->binfmt)
2068 module_put(mm->binfmt->module);
2069
2070 mm->binfmt = new;
2071 if (new)
2072 __module_get(new->module);
2073}
2074EXPORT_SYMBOL(set_binfmt);
2075
2076/*
2077 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2078 */
2079void set_dumpable(struct mm_struct *mm, int value)
2080{
2081 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2082 return;
2083
2084 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2085}
2086
2087SYSCALL_DEFINE3(execve,
2088 const char __user *, filename,
2089 const char __user *const __user *, argv,
2090 const char __user *const __user *, envp)
2091{
2092 return do_execve(getname(filename), argv, envp);
2093}
2094
2095SYSCALL_DEFINE5(execveat,
2096 int, fd, const char __user *, filename,
2097 const char __user *const __user *, argv,
2098 const char __user *const __user *, envp,
2099 int, flags)
2100{
2101 return do_execveat(fd,
2102 getname_uflags(filename, flags),
2103 argv, envp, flags);
2104}
2105
2106#ifdef CONFIG_COMPAT
2107COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2108 const compat_uptr_t __user *, argv,
2109 const compat_uptr_t __user *, envp)
2110{
2111 return compat_do_execve(getname(filename), argv, envp);
2112}
2113
2114COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2115 const char __user *, filename,
2116 const compat_uptr_t __user *, argv,
2117 const compat_uptr_t __user *, envp,
2118 int, flags)
2119{
2120 return compat_do_execveat(fd,
2121 getname_uflags(filename, flags),
2122 argv, envp, flags);
2123}
2124#endif
2125
2126#ifdef CONFIG_SYSCTL
2127
2128static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2129 void *buffer, size_t *lenp, loff_t *ppos)
2130{
2131 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2132
2133 if (!error)
2134 validate_coredump_safety();
2135 return error;
2136}
2137
2138static struct ctl_table fs_exec_sysctls[] = {
2139 {
2140 .procname = "suid_dumpable",
2141 .data = &suid_dumpable,
2142 .maxlen = sizeof(int),
2143 .mode = 0644,
2144 .proc_handler = proc_dointvec_minmax_coredump,
2145 .extra1 = SYSCTL_ZERO,
2146 .extra2 = SYSCTL_TWO,
2147 },
2148 { }
2149};
2150
2151static int __init init_fs_exec_sysctls(void)
2152{
2153 register_sysctl_init("fs", fs_exec_sysctls);
2154 return 0;
2155}
2156
2157fs_initcall(init_fs_exec_sysctls);
2158#endif /* CONFIG_SYSCTL */