Loading...
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
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 = 0;
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 VMA_ITERATOR(vmi, mm, new_start);
687 struct vm_area_struct *next;
688 struct mmu_gather tlb;
689
690 BUG_ON(new_start > new_end);
691
692 /*
693 * ensure there are no vmas between where we want to go
694 * and where we are
695 */
696 if (vma != vma_next(&vmi))
697 return -EFAULT;
698
699 /*
700 * cover the whole range: [new_start, old_end)
701 */
702 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
703 return -ENOMEM;
704
705 /*
706 * move the page tables downwards, on failure we rely on
707 * process cleanup to remove whatever mess we made.
708 */
709 if (length != move_page_tables(vma, old_start,
710 vma, new_start, length, false))
711 return -ENOMEM;
712
713 lru_add_drain();
714 tlb_gather_mmu(&tlb, mm);
715 next = vma_next(&vmi);
716 if (new_end > old_start) {
717 /*
718 * when the old and new regions overlap clear from new_end.
719 */
720 free_pgd_range(&tlb, new_end, old_end, new_end,
721 next ? next->vm_start : USER_PGTABLES_CEILING);
722 } else {
723 /*
724 * otherwise, clean from old_start; this is done to not touch
725 * the address space in [new_end, old_start) some architectures
726 * have constraints on va-space that make this illegal (IA64) -
727 * for the others its just a little faster.
728 */
729 free_pgd_range(&tlb, old_start, old_end, new_end,
730 next ? next->vm_start : USER_PGTABLES_CEILING);
731 }
732 tlb_finish_mmu(&tlb);
733
734 /*
735 * Shrink the vma to just the new range. Always succeeds.
736 */
737 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
738
739 return 0;
740}
741
742/*
743 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
744 * the stack is optionally relocated, and some extra space is added.
745 */
746int setup_arg_pages(struct linux_binprm *bprm,
747 unsigned long stack_top,
748 int executable_stack)
749{
750 unsigned long ret;
751 unsigned long stack_shift;
752 struct mm_struct *mm = current->mm;
753 struct vm_area_struct *vma = bprm->vma;
754 struct vm_area_struct *prev = NULL;
755 unsigned long vm_flags;
756 unsigned long stack_base;
757 unsigned long stack_size;
758 unsigned long stack_expand;
759 unsigned long rlim_stack;
760 struct mmu_gather tlb;
761
762#ifdef CONFIG_STACK_GROWSUP
763 /* Limit stack size */
764 stack_base = bprm->rlim_stack.rlim_max;
765
766 stack_base = calc_max_stack_size(stack_base);
767
768 /* Add space for stack randomization. */
769 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
770
771 /* Make sure we didn't let the argument array grow too large. */
772 if (vma->vm_end - vma->vm_start > stack_base)
773 return -ENOMEM;
774
775 stack_base = PAGE_ALIGN(stack_top - stack_base);
776
777 stack_shift = vma->vm_start - stack_base;
778 mm->arg_start = bprm->p - stack_shift;
779 bprm->p = vma->vm_end - stack_shift;
780#else
781 stack_top = arch_align_stack(stack_top);
782 stack_top = PAGE_ALIGN(stack_top);
783
784 if (unlikely(stack_top < mmap_min_addr) ||
785 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
786 return -ENOMEM;
787
788 stack_shift = vma->vm_end - stack_top;
789
790 bprm->p -= stack_shift;
791 mm->arg_start = bprm->p;
792#endif
793
794 if (bprm->loader)
795 bprm->loader -= stack_shift;
796 bprm->exec -= stack_shift;
797
798 if (mmap_write_lock_killable(mm))
799 return -EINTR;
800
801 vm_flags = VM_STACK_FLAGS;
802
803 /*
804 * Adjust stack execute permissions; explicitly enable for
805 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
806 * (arch default) otherwise.
807 */
808 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
809 vm_flags |= VM_EXEC;
810 else if (executable_stack == EXSTACK_DISABLE_X)
811 vm_flags &= ~VM_EXEC;
812 vm_flags |= mm->def_flags;
813 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
814
815 tlb_gather_mmu(&tlb, mm);
816 ret = mprotect_fixup(&tlb, vma, &prev, vma->vm_start, vma->vm_end,
817 vm_flags);
818 tlb_finish_mmu(&tlb);
819
820 if (ret)
821 goto out_unlock;
822 BUG_ON(prev != vma);
823
824 if (unlikely(vm_flags & VM_EXEC)) {
825 pr_warn_once("process '%pD4' started with executable stack\n",
826 bprm->file);
827 }
828
829 /* Move stack pages down in memory. */
830 if (stack_shift) {
831 ret = shift_arg_pages(vma, stack_shift);
832 if (ret)
833 goto out_unlock;
834 }
835
836 /* mprotect_fixup is overkill to remove the temporary stack flags */
837 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
838
839 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
840 stack_size = vma->vm_end - vma->vm_start;
841 /*
842 * Align this down to a page boundary as expand_stack
843 * will align it up.
844 */
845 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
846
847 stack_expand = min(rlim_stack, stack_size + stack_expand);
848
849#ifdef CONFIG_STACK_GROWSUP
850 stack_base = vma->vm_start + stack_expand;
851#else
852 stack_base = vma->vm_end - 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_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
961ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
962{
963 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
964 if (res > 0)
965 flush_icache_user_range(addr, addr + len);
966 return res;
967}
968EXPORT_SYMBOL(read_code);
969#endif
970
971/*
972 * Maps the mm_struct mm into the current task struct.
973 * On success, this function returns with exec_update_lock
974 * held for writing.
975 */
976static int exec_mmap(struct mm_struct *mm)
977{
978 struct task_struct *tsk;
979 struct mm_struct *old_mm, *active_mm;
980 int ret;
981
982 /* Notify parent that we're no longer interested in the old VM */
983 tsk = current;
984 old_mm = current->mm;
985 exec_mm_release(tsk, old_mm);
986 if (old_mm)
987 sync_mm_rss(old_mm);
988
989 ret = down_write_killable(&tsk->signal->exec_update_lock);
990 if (ret)
991 return ret;
992
993 if (old_mm) {
994 /*
995 * If there is a pending fatal signal perhaps a signal
996 * whose default action is to create a coredump get
997 * out and die instead of going through with the exec.
998 */
999 ret = mmap_read_lock_killable(old_mm);
1000 if (ret) {
1001 up_write(&tsk->signal->exec_update_lock);
1002 return ret;
1003 }
1004 }
1005
1006 task_lock(tsk);
1007 membarrier_exec_mmap(mm);
1008
1009 local_irq_disable();
1010 active_mm = tsk->active_mm;
1011 tsk->active_mm = mm;
1012 tsk->mm = mm;
1013 /*
1014 * This prevents preemption while active_mm is being loaded and
1015 * it and mm are being updated, which could cause problems for
1016 * lazy tlb mm refcounting when these are updated by context
1017 * switches. Not all architectures can handle irqs off over
1018 * activate_mm yet.
1019 */
1020 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1021 local_irq_enable();
1022 activate_mm(active_mm, mm);
1023 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1024 local_irq_enable();
1025 lru_gen_add_mm(mm);
1026 task_unlock(tsk);
1027 lru_gen_use_mm(mm);
1028 if (old_mm) {
1029 mmap_read_unlock(old_mm);
1030 BUG_ON(active_mm != old_mm);
1031 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1032 mm_update_next_owner(old_mm);
1033 mmput(old_mm);
1034 return 0;
1035 }
1036 mmdrop(active_mm);
1037 return 0;
1038}
1039
1040static int de_thread(struct task_struct *tsk)
1041{
1042 struct signal_struct *sig = tsk->signal;
1043 struct sighand_struct *oldsighand = tsk->sighand;
1044 spinlock_t *lock = &oldsighand->siglock;
1045
1046 if (thread_group_empty(tsk))
1047 goto no_thread_group;
1048
1049 /*
1050 * Kill all other threads in the thread group.
1051 */
1052 spin_lock_irq(lock);
1053 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1054 /*
1055 * Another group action in progress, just
1056 * return so that the signal is processed.
1057 */
1058 spin_unlock_irq(lock);
1059 return -EAGAIN;
1060 }
1061
1062 sig->group_exec_task = tsk;
1063 sig->notify_count = zap_other_threads(tsk);
1064 if (!thread_group_leader(tsk))
1065 sig->notify_count--;
1066
1067 while (sig->notify_count) {
1068 __set_current_state(TASK_KILLABLE);
1069 spin_unlock_irq(lock);
1070 schedule();
1071 if (__fatal_signal_pending(tsk))
1072 goto killed;
1073 spin_lock_irq(lock);
1074 }
1075 spin_unlock_irq(lock);
1076
1077 /*
1078 * At this point all other threads have exited, all we have to
1079 * do is to wait for the thread group leader to become inactive,
1080 * and to assume its PID:
1081 */
1082 if (!thread_group_leader(tsk)) {
1083 struct task_struct *leader = tsk->group_leader;
1084
1085 for (;;) {
1086 cgroup_threadgroup_change_begin(tsk);
1087 write_lock_irq(&tasklist_lock);
1088 /*
1089 * Do this under tasklist_lock to ensure that
1090 * exit_notify() can't miss ->group_exec_task
1091 */
1092 sig->notify_count = -1;
1093 if (likely(leader->exit_state))
1094 break;
1095 __set_current_state(TASK_KILLABLE);
1096 write_unlock_irq(&tasklist_lock);
1097 cgroup_threadgroup_change_end(tsk);
1098 schedule();
1099 if (__fatal_signal_pending(tsk))
1100 goto killed;
1101 }
1102
1103 /*
1104 * The only record we have of the real-time age of a
1105 * process, regardless of execs it's done, is start_time.
1106 * All the past CPU time is accumulated in signal_struct
1107 * from sister threads now dead. But in this non-leader
1108 * exec, nothing survives from the original leader thread,
1109 * whose birth marks the true age of this process now.
1110 * When we take on its identity by switching to its PID, we
1111 * also take its birthdate (always earlier than our own).
1112 */
1113 tsk->start_time = leader->start_time;
1114 tsk->start_boottime = leader->start_boottime;
1115
1116 BUG_ON(!same_thread_group(leader, tsk));
1117 /*
1118 * An exec() starts a new thread group with the
1119 * TGID of the previous thread group. Rehash the
1120 * two threads with a switched PID, and release
1121 * the former thread group leader:
1122 */
1123
1124 /* Become a process group leader with the old leader's pid.
1125 * The old leader becomes a thread of the this thread group.
1126 */
1127 exchange_tids(tsk, leader);
1128 transfer_pid(leader, tsk, PIDTYPE_TGID);
1129 transfer_pid(leader, tsk, PIDTYPE_PGID);
1130 transfer_pid(leader, tsk, PIDTYPE_SID);
1131
1132 list_replace_rcu(&leader->tasks, &tsk->tasks);
1133 list_replace_init(&leader->sibling, &tsk->sibling);
1134
1135 tsk->group_leader = tsk;
1136 leader->group_leader = tsk;
1137
1138 tsk->exit_signal = SIGCHLD;
1139 leader->exit_signal = -1;
1140
1141 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1142 leader->exit_state = EXIT_DEAD;
1143
1144 /*
1145 * We are going to release_task()->ptrace_unlink() silently,
1146 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1147 * the tracer won't block again waiting for this thread.
1148 */
1149 if (unlikely(leader->ptrace))
1150 __wake_up_parent(leader, leader->parent);
1151 write_unlock_irq(&tasklist_lock);
1152 cgroup_threadgroup_change_end(tsk);
1153
1154 release_task(leader);
1155 }
1156
1157 sig->group_exec_task = NULL;
1158 sig->notify_count = 0;
1159
1160no_thread_group:
1161 /* we have changed execution domain */
1162 tsk->exit_signal = SIGCHLD;
1163
1164 BUG_ON(!thread_group_leader(tsk));
1165 return 0;
1166
1167killed:
1168 /* protects against exit_notify() and __exit_signal() */
1169 read_lock(&tasklist_lock);
1170 sig->group_exec_task = NULL;
1171 sig->notify_count = 0;
1172 read_unlock(&tasklist_lock);
1173 return -EAGAIN;
1174}
1175
1176
1177/*
1178 * This function makes sure the current process has its own signal table,
1179 * so that flush_signal_handlers can later reset the handlers without
1180 * disturbing other processes. (Other processes might share the signal
1181 * table via the CLONE_SIGHAND option to clone().)
1182 */
1183static int unshare_sighand(struct task_struct *me)
1184{
1185 struct sighand_struct *oldsighand = me->sighand;
1186
1187 if (refcount_read(&oldsighand->count) != 1) {
1188 struct sighand_struct *newsighand;
1189 /*
1190 * This ->sighand is shared with the CLONE_SIGHAND
1191 * but not CLONE_THREAD task, switch to the new one.
1192 */
1193 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1194 if (!newsighand)
1195 return -ENOMEM;
1196
1197 refcount_set(&newsighand->count, 1);
1198
1199 write_lock_irq(&tasklist_lock);
1200 spin_lock(&oldsighand->siglock);
1201 memcpy(newsighand->action, oldsighand->action,
1202 sizeof(newsighand->action));
1203 rcu_assign_pointer(me->sighand, newsighand);
1204 spin_unlock(&oldsighand->siglock);
1205 write_unlock_irq(&tasklist_lock);
1206
1207 __cleanup_sighand(oldsighand);
1208 }
1209 return 0;
1210}
1211
1212char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1213{
1214 task_lock(tsk);
1215 /* Always NUL terminated and zero-padded */
1216 strscpy_pad(buf, tsk->comm, buf_size);
1217 task_unlock(tsk);
1218 return buf;
1219}
1220EXPORT_SYMBOL_GPL(__get_task_comm);
1221
1222/*
1223 * These functions flushes out all traces of the currently running executable
1224 * so that a new one can be started
1225 */
1226
1227void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1228{
1229 task_lock(tsk);
1230 trace_task_rename(tsk, buf);
1231 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1232 task_unlock(tsk);
1233 perf_event_comm(tsk, exec);
1234}
1235
1236/*
1237 * Calling this is the point of no return. None of the failures will be
1238 * seen by userspace since either the process is already taking a fatal
1239 * signal (via de_thread() or coredump), or will have SEGV raised
1240 * (after exec_mmap()) by search_binary_handler (see below).
1241 */
1242int begin_new_exec(struct linux_binprm * bprm)
1243{
1244 struct task_struct *me = current;
1245 int retval;
1246
1247 /* Once we are committed compute the creds */
1248 retval = bprm_creds_from_file(bprm);
1249 if (retval)
1250 return retval;
1251
1252 /*
1253 * Ensure all future errors are fatal.
1254 */
1255 bprm->point_of_no_return = true;
1256
1257 /*
1258 * Make this the only thread in the thread group.
1259 */
1260 retval = de_thread(me);
1261 if (retval)
1262 goto out;
1263
1264 /*
1265 * Cancel any io_uring activity across execve
1266 */
1267 io_uring_task_cancel();
1268
1269 /* Ensure the files table is not shared. */
1270 retval = unshare_files();
1271 if (retval)
1272 goto out;
1273
1274 /*
1275 * Must be called _before_ exec_mmap() as bprm->mm is
1276 * not visible until then. This also enables the update
1277 * to be lockless.
1278 */
1279 retval = set_mm_exe_file(bprm->mm, bprm->file);
1280 if (retval)
1281 goto out;
1282
1283 /* If the binary is not readable then enforce mm->dumpable=0 */
1284 would_dump(bprm, bprm->file);
1285 if (bprm->have_execfd)
1286 would_dump(bprm, bprm->executable);
1287
1288 /*
1289 * Release all of the old mmap stuff
1290 */
1291 acct_arg_size(bprm, 0);
1292 retval = exec_mmap(bprm->mm);
1293 if (retval)
1294 goto out;
1295
1296 bprm->mm = NULL;
1297
1298 retval = exec_task_namespaces();
1299 if (retval)
1300 goto out_unlock;
1301
1302#ifdef CONFIG_POSIX_TIMERS
1303 spin_lock_irq(&me->sighand->siglock);
1304 posix_cpu_timers_exit(me);
1305 spin_unlock_irq(&me->sighand->siglock);
1306 exit_itimers(me);
1307 flush_itimer_signals();
1308#endif
1309
1310 /*
1311 * Make the signal table private.
1312 */
1313 retval = unshare_sighand(me);
1314 if (retval)
1315 goto out_unlock;
1316
1317 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1318 PF_NOFREEZE | PF_NO_SETAFFINITY);
1319 flush_thread();
1320 me->personality &= ~bprm->per_clear;
1321
1322 clear_syscall_work_syscall_user_dispatch(me);
1323
1324 /*
1325 * We have to apply CLOEXEC before we change whether the process is
1326 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1327 * trying to access the should-be-closed file descriptors of a process
1328 * undergoing exec(2).
1329 */
1330 do_close_on_exec(me->files);
1331
1332 if (bprm->secureexec) {
1333 /* Make sure parent cannot signal privileged process. */
1334 me->pdeath_signal = 0;
1335
1336 /*
1337 * For secureexec, reset the stack limit to sane default to
1338 * avoid bad behavior from the prior rlimits. This has to
1339 * happen before arch_pick_mmap_layout(), which examines
1340 * RLIMIT_STACK, but after the point of no return to avoid
1341 * needing to clean up the change on failure.
1342 */
1343 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1344 bprm->rlim_stack.rlim_cur = _STK_LIM;
1345 }
1346
1347 me->sas_ss_sp = me->sas_ss_size = 0;
1348
1349 /*
1350 * Figure out dumpability. Note that this checking only of current
1351 * is wrong, but userspace depends on it. This should be testing
1352 * bprm->secureexec instead.
1353 */
1354 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1355 !(uid_eq(current_euid(), current_uid()) &&
1356 gid_eq(current_egid(), current_gid())))
1357 set_dumpable(current->mm, suid_dumpable);
1358 else
1359 set_dumpable(current->mm, SUID_DUMP_USER);
1360
1361 perf_event_exec();
1362 __set_task_comm(me, kbasename(bprm->filename), true);
1363
1364 /* An exec changes our domain. We are no longer part of the thread
1365 group */
1366 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1367 flush_signal_handlers(me, 0);
1368
1369 retval = set_cred_ucounts(bprm->cred);
1370 if (retval < 0)
1371 goto out_unlock;
1372
1373 /*
1374 * install the new credentials for this executable
1375 */
1376 security_bprm_committing_creds(bprm);
1377
1378 commit_creds(bprm->cred);
1379 bprm->cred = NULL;
1380
1381 /*
1382 * Disable monitoring for regular users
1383 * when executing setuid binaries. Must
1384 * wait until new credentials are committed
1385 * by commit_creds() above
1386 */
1387 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1388 perf_event_exit_task(me);
1389 /*
1390 * cred_guard_mutex must be held at least to this point to prevent
1391 * ptrace_attach() from altering our determination of the task's
1392 * credentials; any time after this it may be unlocked.
1393 */
1394 security_bprm_committed_creds(bprm);
1395
1396 /* Pass the opened binary to the interpreter. */
1397 if (bprm->have_execfd) {
1398 retval = get_unused_fd_flags(0);
1399 if (retval < 0)
1400 goto out_unlock;
1401 fd_install(retval, bprm->executable);
1402 bprm->executable = NULL;
1403 bprm->execfd = retval;
1404 }
1405 return 0;
1406
1407out_unlock:
1408 up_write(&me->signal->exec_update_lock);
1409out:
1410 return retval;
1411}
1412EXPORT_SYMBOL(begin_new_exec);
1413
1414void would_dump(struct linux_binprm *bprm, struct file *file)
1415{
1416 struct inode *inode = file_inode(file);
1417 struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1418 if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1419 struct user_namespace *old, *user_ns;
1420 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1421
1422 /* Ensure mm->user_ns contains the executable */
1423 user_ns = old = bprm->mm->user_ns;
1424 while ((user_ns != &init_user_ns) &&
1425 !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1426 user_ns = user_ns->parent;
1427
1428 if (old != user_ns) {
1429 bprm->mm->user_ns = get_user_ns(user_ns);
1430 put_user_ns(old);
1431 }
1432 }
1433}
1434EXPORT_SYMBOL(would_dump);
1435
1436void setup_new_exec(struct linux_binprm * bprm)
1437{
1438 /* Setup things that can depend upon the personality */
1439 struct task_struct *me = current;
1440
1441 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1442
1443 arch_setup_new_exec();
1444
1445 /* Set the new mm task size. We have to do that late because it may
1446 * depend on TIF_32BIT which is only updated in flush_thread() on
1447 * some architectures like powerpc
1448 */
1449 me->mm->task_size = TASK_SIZE;
1450 up_write(&me->signal->exec_update_lock);
1451 mutex_unlock(&me->signal->cred_guard_mutex);
1452}
1453EXPORT_SYMBOL(setup_new_exec);
1454
1455/* Runs immediately before start_thread() takes over. */
1456void finalize_exec(struct linux_binprm *bprm)
1457{
1458 /* Store any stack rlimit changes before starting thread. */
1459 task_lock(current->group_leader);
1460 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1461 task_unlock(current->group_leader);
1462}
1463EXPORT_SYMBOL(finalize_exec);
1464
1465/*
1466 * Prepare credentials and lock ->cred_guard_mutex.
1467 * setup_new_exec() commits the new creds and drops the lock.
1468 * Or, if exec fails before, free_bprm() should release ->cred
1469 * and unlock.
1470 */
1471static int prepare_bprm_creds(struct linux_binprm *bprm)
1472{
1473 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1474 return -ERESTARTNOINTR;
1475
1476 bprm->cred = prepare_exec_creds();
1477 if (likely(bprm->cred))
1478 return 0;
1479
1480 mutex_unlock(¤t->signal->cred_guard_mutex);
1481 return -ENOMEM;
1482}
1483
1484static void free_bprm(struct linux_binprm *bprm)
1485{
1486 if (bprm->mm) {
1487 acct_arg_size(bprm, 0);
1488 mmput(bprm->mm);
1489 }
1490 free_arg_pages(bprm);
1491 if (bprm->cred) {
1492 mutex_unlock(¤t->signal->cred_guard_mutex);
1493 abort_creds(bprm->cred);
1494 }
1495 if (bprm->file) {
1496 allow_write_access(bprm->file);
1497 fput(bprm->file);
1498 }
1499 if (bprm->executable)
1500 fput(bprm->executable);
1501 /* If a binfmt changed the interp, free it. */
1502 if (bprm->interp != bprm->filename)
1503 kfree(bprm->interp);
1504 kfree(bprm->fdpath);
1505 kfree(bprm);
1506}
1507
1508static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1509{
1510 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1511 int retval = -ENOMEM;
1512 if (!bprm)
1513 goto out;
1514
1515 if (fd == AT_FDCWD || filename->name[0] == '/') {
1516 bprm->filename = filename->name;
1517 } else {
1518 if (filename->name[0] == '\0')
1519 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1520 else
1521 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1522 fd, filename->name);
1523 if (!bprm->fdpath)
1524 goto out_free;
1525
1526 bprm->filename = bprm->fdpath;
1527 }
1528 bprm->interp = bprm->filename;
1529
1530 retval = bprm_mm_init(bprm);
1531 if (retval)
1532 goto out_free;
1533 return bprm;
1534
1535out_free:
1536 free_bprm(bprm);
1537out:
1538 return ERR_PTR(retval);
1539}
1540
1541int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1542{
1543 /* If a binfmt changed the interp, free it first. */
1544 if (bprm->interp != bprm->filename)
1545 kfree(bprm->interp);
1546 bprm->interp = kstrdup(interp, GFP_KERNEL);
1547 if (!bprm->interp)
1548 return -ENOMEM;
1549 return 0;
1550}
1551EXPORT_SYMBOL(bprm_change_interp);
1552
1553/*
1554 * determine how safe it is to execute the proposed program
1555 * - the caller must hold ->cred_guard_mutex to protect against
1556 * PTRACE_ATTACH or seccomp thread-sync
1557 */
1558static void check_unsafe_exec(struct linux_binprm *bprm)
1559{
1560 struct task_struct *p = current, *t;
1561 unsigned n_fs;
1562
1563 if (p->ptrace)
1564 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1565
1566 /*
1567 * This isn't strictly necessary, but it makes it harder for LSMs to
1568 * mess up.
1569 */
1570 if (task_no_new_privs(current))
1571 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1572
1573 /*
1574 * If another task is sharing our fs, we cannot safely
1575 * suid exec because the differently privileged task
1576 * will be able to manipulate the current directory, etc.
1577 * It would be nice to force an unshare instead...
1578 */
1579 t = p;
1580 n_fs = 1;
1581 spin_lock(&p->fs->lock);
1582 rcu_read_lock();
1583 while_each_thread(p, t) {
1584 if (t->fs == p->fs)
1585 n_fs++;
1586 }
1587 rcu_read_unlock();
1588
1589 if (p->fs->users > n_fs)
1590 bprm->unsafe |= LSM_UNSAFE_SHARE;
1591 else
1592 p->fs->in_exec = 1;
1593 spin_unlock(&p->fs->lock);
1594}
1595
1596static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1597{
1598 /* Handle suid and sgid on files */
1599 struct user_namespace *mnt_userns;
1600 struct inode *inode = file_inode(file);
1601 unsigned int mode;
1602 vfsuid_t vfsuid;
1603 vfsgid_t vfsgid;
1604
1605 if (!mnt_may_suid(file->f_path.mnt))
1606 return;
1607
1608 if (task_no_new_privs(current))
1609 return;
1610
1611 mode = READ_ONCE(inode->i_mode);
1612 if (!(mode & (S_ISUID|S_ISGID)))
1613 return;
1614
1615 mnt_userns = file_mnt_user_ns(file);
1616
1617 /* Be careful if suid/sgid is set */
1618 inode_lock(inode);
1619
1620 /* reload atomically mode/uid/gid now that lock held */
1621 mode = inode->i_mode;
1622 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
1623 vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
1624 inode_unlock(inode);
1625
1626 /* We ignore suid/sgid if there are no mappings for them in the ns */
1627 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1628 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1629 return;
1630
1631 if (mode & S_ISUID) {
1632 bprm->per_clear |= PER_CLEAR_ON_SETID;
1633 bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1634 }
1635
1636 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1637 bprm->per_clear |= PER_CLEAR_ON_SETID;
1638 bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1639 }
1640}
1641
1642/*
1643 * Compute brpm->cred based upon the final binary.
1644 */
1645static int bprm_creds_from_file(struct linux_binprm *bprm)
1646{
1647 /* Compute creds based on which file? */
1648 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1649
1650 bprm_fill_uid(bprm, file);
1651 return security_bprm_creds_from_file(bprm, file);
1652}
1653
1654/*
1655 * Fill the binprm structure from the inode.
1656 * Read the first BINPRM_BUF_SIZE bytes
1657 *
1658 * This may be called multiple times for binary chains (scripts for example).
1659 */
1660static int prepare_binprm(struct linux_binprm *bprm)
1661{
1662 loff_t pos = 0;
1663
1664 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1665 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1666}
1667
1668/*
1669 * Arguments are '\0' separated strings found at the location bprm->p
1670 * points to; chop off the first by relocating brpm->p to right after
1671 * the first '\0' encountered.
1672 */
1673int remove_arg_zero(struct linux_binprm *bprm)
1674{
1675 int ret = 0;
1676 unsigned long offset;
1677 char *kaddr;
1678 struct page *page;
1679
1680 if (!bprm->argc)
1681 return 0;
1682
1683 do {
1684 offset = bprm->p & ~PAGE_MASK;
1685 page = get_arg_page(bprm, bprm->p, 0);
1686 if (!page) {
1687 ret = -EFAULT;
1688 goto out;
1689 }
1690 kaddr = kmap_local_page(page);
1691
1692 for (; offset < PAGE_SIZE && kaddr[offset];
1693 offset++, bprm->p++)
1694 ;
1695
1696 kunmap_local(kaddr);
1697 put_arg_page(page);
1698 } while (offset == PAGE_SIZE);
1699
1700 bprm->p++;
1701 bprm->argc--;
1702 ret = 0;
1703
1704out:
1705 return ret;
1706}
1707EXPORT_SYMBOL(remove_arg_zero);
1708
1709#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1710/*
1711 * cycle the list of binary formats handler, until one recognizes the image
1712 */
1713static int search_binary_handler(struct linux_binprm *bprm)
1714{
1715 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1716 struct linux_binfmt *fmt;
1717 int retval;
1718
1719 retval = prepare_binprm(bprm);
1720 if (retval < 0)
1721 return retval;
1722
1723 retval = security_bprm_check(bprm);
1724 if (retval)
1725 return retval;
1726
1727 retval = -ENOENT;
1728 retry:
1729 read_lock(&binfmt_lock);
1730 list_for_each_entry(fmt, &formats, lh) {
1731 if (!try_module_get(fmt->module))
1732 continue;
1733 read_unlock(&binfmt_lock);
1734
1735 retval = fmt->load_binary(bprm);
1736
1737 read_lock(&binfmt_lock);
1738 put_binfmt(fmt);
1739 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1740 read_unlock(&binfmt_lock);
1741 return retval;
1742 }
1743 }
1744 read_unlock(&binfmt_lock);
1745
1746 if (need_retry) {
1747 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1748 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1749 return retval;
1750 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1751 return retval;
1752 need_retry = false;
1753 goto retry;
1754 }
1755
1756 return retval;
1757}
1758
1759/* binfmt handlers will call back into begin_new_exec() on success. */
1760static int exec_binprm(struct linux_binprm *bprm)
1761{
1762 pid_t old_pid, old_vpid;
1763 int ret, depth;
1764
1765 /* Need to fetch pid before load_binary changes it */
1766 old_pid = current->pid;
1767 rcu_read_lock();
1768 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1769 rcu_read_unlock();
1770
1771 /* This allows 4 levels of binfmt rewrites before failing hard. */
1772 for (depth = 0;; depth++) {
1773 struct file *exec;
1774 if (depth > 5)
1775 return -ELOOP;
1776
1777 ret = search_binary_handler(bprm);
1778 if (ret < 0)
1779 return ret;
1780 if (!bprm->interpreter)
1781 break;
1782
1783 exec = bprm->file;
1784 bprm->file = bprm->interpreter;
1785 bprm->interpreter = NULL;
1786
1787 allow_write_access(exec);
1788 if (unlikely(bprm->have_execfd)) {
1789 if (bprm->executable) {
1790 fput(exec);
1791 return -ENOEXEC;
1792 }
1793 bprm->executable = exec;
1794 } else
1795 fput(exec);
1796 }
1797
1798 audit_bprm(bprm);
1799 trace_sched_process_exec(current, old_pid, bprm);
1800 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1801 proc_exec_connector(current);
1802 return 0;
1803}
1804
1805/*
1806 * sys_execve() executes a new program.
1807 */
1808static int bprm_execve(struct linux_binprm *bprm,
1809 int fd, struct filename *filename, int flags)
1810{
1811 struct file *file;
1812 int retval;
1813
1814 retval = prepare_bprm_creds(bprm);
1815 if (retval)
1816 return retval;
1817
1818 /*
1819 * Check for unsafe execution states before exec_binprm(), which
1820 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1821 * where setuid-ness is evaluated.
1822 */
1823 check_unsafe_exec(bprm);
1824 current->in_execve = 1;
1825
1826 file = do_open_execat(fd, filename, flags);
1827 retval = PTR_ERR(file);
1828 if (IS_ERR(file))
1829 goto out_unmark;
1830
1831 sched_exec();
1832
1833 bprm->file = file;
1834 /*
1835 * Record that a name derived from an O_CLOEXEC fd will be
1836 * inaccessible after exec. This allows the code in exec to
1837 * choose to fail when the executable is not mmaped into the
1838 * interpreter and an open file descriptor is not passed to
1839 * the interpreter. This makes for a better user experience
1840 * than having the interpreter start and then immediately fail
1841 * when it finds the executable is inaccessible.
1842 */
1843 if (bprm->fdpath && get_close_on_exec(fd))
1844 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1845
1846 /* Set the unchanging part of bprm->cred */
1847 retval = security_bprm_creds_for_exec(bprm);
1848 if (retval)
1849 goto out;
1850
1851 retval = exec_binprm(bprm);
1852 if (retval < 0)
1853 goto out;
1854
1855 /* execve succeeded */
1856 current->fs->in_exec = 0;
1857 current->in_execve = 0;
1858 rseq_execve(current);
1859 acct_update_integrals(current);
1860 task_numa_free(current, false);
1861 return retval;
1862
1863out:
1864 /*
1865 * If past the point of no return ensure the code never
1866 * returns to the userspace process. Use an existing fatal
1867 * signal if present otherwise terminate the process with
1868 * SIGSEGV.
1869 */
1870 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1871 force_fatal_sig(SIGSEGV);
1872
1873out_unmark:
1874 current->fs->in_exec = 0;
1875 current->in_execve = 0;
1876
1877 return retval;
1878}
1879
1880static int do_execveat_common(int fd, struct filename *filename,
1881 struct user_arg_ptr argv,
1882 struct user_arg_ptr envp,
1883 int flags)
1884{
1885 struct linux_binprm *bprm;
1886 int retval;
1887
1888 if (IS_ERR(filename))
1889 return PTR_ERR(filename);
1890
1891 /*
1892 * We move the actual failure in case of RLIMIT_NPROC excess from
1893 * set*uid() to execve() because too many poorly written programs
1894 * don't check setuid() return code. Here we additionally recheck
1895 * whether NPROC limit is still exceeded.
1896 */
1897 if ((current->flags & PF_NPROC_EXCEEDED) &&
1898 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1899 retval = -EAGAIN;
1900 goto out_ret;
1901 }
1902
1903 /* We're below the limit (still or again), so we don't want to make
1904 * further execve() calls fail. */
1905 current->flags &= ~PF_NPROC_EXCEEDED;
1906
1907 bprm = alloc_bprm(fd, filename);
1908 if (IS_ERR(bprm)) {
1909 retval = PTR_ERR(bprm);
1910 goto out_ret;
1911 }
1912
1913 retval = count(argv, MAX_ARG_STRINGS);
1914 if (retval == 0)
1915 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1916 current->comm, bprm->filename);
1917 if (retval < 0)
1918 goto out_free;
1919 bprm->argc = retval;
1920
1921 retval = count(envp, MAX_ARG_STRINGS);
1922 if (retval < 0)
1923 goto out_free;
1924 bprm->envc = retval;
1925
1926 retval = bprm_stack_limits(bprm);
1927 if (retval < 0)
1928 goto out_free;
1929
1930 retval = copy_string_kernel(bprm->filename, bprm);
1931 if (retval < 0)
1932 goto out_free;
1933 bprm->exec = bprm->p;
1934
1935 retval = copy_strings(bprm->envc, envp, bprm);
1936 if (retval < 0)
1937 goto out_free;
1938
1939 retval = copy_strings(bprm->argc, argv, bprm);
1940 if (retval < 0)
1941 goto out_free;
1942
1943 /*
1944 * When argv is empty, add an empty string ("") as argv[0] to
1945 * ensure confused userspace programs that start processing
1946 * from argv[1] won't end up walking envp. See also
1947 * bprm_stack_limits().
1948 */
1949 if (bprm->argc == 0) {
1950 retval = copy_string_kernel("", bprm);
1951 if (retval < 0)
1952 goto out_free;
1953 bprm->argc = 1;
1954 }
1955
1956 retval = bprm_execve(bprm, fd, filename, flags);
1957out_free:
1958 free_bprm(bprm);
1959
1960out_ret:
1961 putname(filename);
1962 return retval;
1963}
1964
1965int kernel_execve(const char *kernel_filename,
1966 const char *const *argv, const char *const *envp)
1967{
1968 struct filename *filename;
1969 struct linux_binprm *bprm;
1970 int fd = AT_FDCWD;
1971 int retval;
1972
1973 /* It is non-sense for kernel threads to call execve */
1974 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1975 return -EINVAL;
1976
1977 filename = getname_kernel(kernel_filename);
1978 if (IS_ERR(filename))
1979 return PTR_ERR(filename);
1980
1981 bprm = alloc_bprm(fd, filename);
1982 if (IS_ERR(bprm)) {
1983 retval = PTR_ERR(bprm);
1984 goto out_ret;
1985 }
1986
1987 retval = count_strings_kernel(argv);
1988 if (WARN_ON_ONCE(retval == 0))
1989 retval = -EINVAL;
1990 if (retval < 0)
1991 goto out_free;
1992 bprm->argc = retval;
1993
1994 retval = count_strings_kernel(envp);
1995 if (retval < 0)
1996 goto out_free;
1997 bprm->envc = retval;
1998
1999 retval = bprm_stack_limits(bprm);
2000 if (retval < 0)
2001 goto out_free;
2002
2003 retval = copy_string_kernel(bprm->filename, bprm);
2004 if (retval < 0)
2005 goto out_free;
2006 bprm->exec = bprm->p;
2007
2008 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2009 if (retval < 0)
2010 goto out_free;
2011
2012 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2013 if (retval < 0)
2014 goto out_free;
2015
2016 retval = bprm_execve(bprm, fd, filename, 0);
2017out_free:
2018 free_bprm(bprm);
2019out_ret:
2020 putname(filename);
2021 return retval;
2022}
2023
2024static int do_execve(struct filename *filename,
2025 const char __user *const __user *__argv,
2026 const char __user *const __user *__envp)
2027{
2028 struct user_arg_ptr argv = { .ptr.native = __argv };
2029 struct user_arg_ptr envp = { .ptr.native = __envp };
2030 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2031}
2032
2033static int do_execveat(int fd, struct filename *filename,
2034 const char __user *const __user *__argv,
2035 const char __user *const __user *__envp,
2036 int flags)
2037{
2038 struct user_arg_ptr argv = { .ptr.native = __argv };
2039 struct user_arg_ptr envp = { .ptr.native = __envp };
2040
2041 return do_execveat_common(fd, filename, argv, envp, flags);
2042}
2043
2044#ifdef CONFIG_COMPAT
2045static int compat_do_execve(struct filename *filename,
2046 const compat_uptr_t __user *__argv,
2047 const compat_uptr_t __user *__envp)
2048{
2049 struct user_arg_ptr argv = {
2050 .is_compat = true,
2051 .ptr.compat = __argv,
2052 };
2053 struct user_arg_ptr envp = {
2054 .is_compat = true,
2055 .ptr.compat = __envp,
2056 };
2057 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2058}
2059
2060static int compat_do_execveat(int fd, struct filename *filename,
2061 const compat_uptr_t __user *__argv,
2062 const compat_uptr_t __user *__envp,
2063 int flags)
2064{
2065 struct user_arg_ptr argv = {
2066 .is_compat = true,
2067 .ptr.compat = __argv,
2068 };
2069 struct user_arg_ptr envp = {
2070 .is_compat = true,
2071 .ptr.compat = __envp,
2072 };
2073 return do_execveat_common(fd, filename, argv, envp, flags);
2074}
2075#endif
2076
2077void set_binfmt(struct linux_binfmt *new)
2078{
2079 struct mm_struct *mm = current->mm;
2080
2081 if (mm->binfmt)
2082 module_put(mm->binfmt->module);
2083
2084 mm->binfmt = new;
2085 if (new)
2086 __module_get(new->module);
2087}
2088EXPORT_SYMBOL(set_binfmt);
2089
2090/*
2091 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2092 */
2093void set_dumpable(struct mm_struct *mm, int value)
2094{
2095 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2096 return;
2097
2098 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2099}
2100
2101SYSCALL_DEFINE3(execve,
2102 const char __user *, filename,
2103 const char __user *const __user *, argv,
2104 const char __user *const __user *, envp)
2105{
2106 return do_execve(getname(filename), argv, envp);
2107}
2108
2109SYSCALL_DEFINE5(execveat,
2110 int, fd, const char __user *, filename,
2111 const char __user *const __user *, argv,
2112 const char __user *const __user *, envp,
2113 int, flags)
2114{
2115 return do_execveat(fd,
2116 getname_uflags(filename, flags),
2117 argv, envp, flags);
2118}
2119
2120#ifdef CONFIG_COMPAT
2121COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2122 const compat_uptr_t __user *, argv,
2123 const compat_uptr_t __user *, envp)
2124{
2125 return compat_do_execve(getname(filename), argv, envp);
2126}
2127
2128COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2129 const char __user *, filename,
2130 const compat_uptr_t __user *, argv,
2131 const compat_uptr_t __user *, envp,
2132 int, flags)
2133{
2134 return compat_do_execveat(fd,
2135 getname_uflags(filename, flags),
2136 argv, envp, flags);
2137}
2138#endif
2139
2140#ifdef CONFIG_SYSCTL
2141
2142static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2143 void *buffer, size_t *lenp, loff_t *ppos)
2144{
2145 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2146
2147 if (!error)
2148 validate_coredump_safety();
2149 return error;
2150}
2151
2152static struct ctl_table fs_exec_sysctls[] = {
2153 {
2154 .procname = "suid_dumpable",
2155 .data = &suid_dumpable,
2156 .maxlen = sizeof(int),
2157 .mode = 0644,
2158 .proc_handler = proc_dointvec_minmax_coredump,
2159 .extra1 = SYSCTL_ZERO,
2160 .extra2 = SYSCTL_TWO,
2161 },
2162 { }
2163};
2164
2165static int __init init_fs_exec_sysctls(void)
2166{
2167 register_sysctl_init("fs", fs_exec_sysctls);
2168 return 0;
2169}
2170
2171fs_initcall(init_fs_exec_sysctls);
2172#endif /* CONFIG_SYSCTL */
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 *sp_location = sp;
899
900out:
901 return ret;
902}
903EXPORT_SYMBOL(transfer_args_to_stack);
904
905#endif /* CONFIG_MMU */
906
907/*
908 * On success, caller must call do_close_execat() on the returned
909 * struct file to close it.
910 */
911static struct file *do_open_execat(int fd, struct filename *name, int flags)
912{
913 struct file *file;
914 int err;
915 struct open_flags open_exec_flags = {
916 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
917 .acc_mode = MAY_EXEC,
918 .intent = LOOKUP_OPEN,
919 .lookup_flags = LOOKUP_FOLLOW,
920 };
921
922 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
923 return ERR_PTR(-EINVAL);
924 if (flags & AT_SYMLINK_NOFOLLOW)
925 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
926 if (flags & AT_EMPTY_PATH)
927 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
928
929 file = do_filp_open(fd, name, &open_exec_flags);
930 if (IS_ERR(file))
931 goto out;
932
933 /*
934 * may_open() has already checked for this, so it should be
935 * impossible to trip now. But we need to be extra cautious
936 * and check again at the very end too.
937 */
938 err = -EACCES;
939 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
940 path_noexec(&file->f_path)))
941 goto exit;
942
943 err = deny_write_access(file);
944 if (err)
945 goto exit;
946
947out:
948 return file;
949
950exit:
951 fput(file);
952 return ERR_PTR(err);
953}
954
955/**
956 * open_exec - Open a path name for execution
957 *
958 * @name: path name to open with the intent of executing it.
959 *
960 * Returns ERR_PTR on failure or allocated struct file on success.
961 *
962 * As this is a wrapper for the internal do_open_execat(), callers
963 * must call allow_write_access() before fput() on release. Also see
964 * do_close_execat().
965 */
966struct file *open_exec(const char *name)
967{
968 struct filename *filename = getname_kernel(name);
969 struct file *f = ERR_CAST(filename);
970
971 if (!IS_ERR(filename)) {
972 f = do_open_execat(AT_FDCWD, filename, 0);
973 putname(filename);
974 }
975 return f;
976}
977EXPORT_SYMBOL(open_exec);
978
979#if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
980ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
981{
982 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
983 if (res > 0)
984 flush_icache_user_range(addr, addr + len);
985 return res;
986}
987EXPORT_SYMBOL(read_code);
988#endif
989
990/*
991 * Maps the mm_struct mm into the current task struct.
992 * On success, this function returns with exec_update_lock
993 * held for writing.
994 */
995static int exec_mmap(struct mm_struct *mm)
996{
997 struct task_struct *tsk;
998 struct mm_struct *old_mm, *active_mm;
999 int ret;
1000
1001 /* Notify parent that we're no longer interested in the old VM */
1002 tsk = current;
1003 old_mm = current->mm;
1004 exec_mm_release(tsk, old_mm);
1005
1006 ret = down_write_killable(&tsk->signal->exec_update_lock);
1007 if (ret)
1008 return ret;
1009
1010 if (old_mm) {
1011 /*
1012 * If there is a pending fatal signal perhaps a signal
1013 * whose default action is to create a coredump get
1014 * out and die instead of going through with the exec.
1015 */
1016 ret = mmap_read_lock_killable(old_mm);
1017 if (ret) {
1018 up_write(&tsk->signal->exec_update_lock);
1019 return ret;
1020 }
1021 }
1022
1023 task_lock(tsk);
1024 membarrier_exec_mmap(mm);
1025
1026 local_irq_disable();
1027 active_mm = tsk->active_mm;
1028 tsk->active_mm = mm;
1029 tsk->mm = mm;
1030 mm_init_cid(mm);
1031 /*
1032 * This prevents preemption while active_mm is being loaded and
1033 * it and mm are being updated, which could cause problems for
1034 * lazy tlb mm refcounting when these are updated by context
1035 * switches. Not all architectures can handle irqs off over
1036 * activate_mm yet.
1037 */
1038 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1039 local_irq_enable();
1040 activate_mm(active_mm, mm);
1041 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1042 local_irq_enable();
1043 lru_gen_add_mm(mm);
1044 task_unlock(tsk);
1045 lru_gen_use_mm(mm);
1046 if (old_mm) {
1047 mmap_read_unlock(old_mm);
1048 BUG_ON(active_mm != old_mm);
1049 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1050 mm_update_next_owner(old_mm);
1051 mmput(old_mm);
1052 return 0;
1053 }
1054 mmdrop_lazy_tlb(active_mm);
1055 return 0;
1056}
1057
1058static int de_thread(struct task_struct *tsk)
1059{
1060 struct signal_struct *sig = tsk->signal;
1061 struct sighand_struct *oldsighand = tsk->sighand;
1062 spinlock_t *lock = &oldsighand->siglock;
1063
1064 if (thread_group_empty(tsk))
1065 goto no_thread_group;
1066
1067 /*
1068 * Kill all other threads in the thread group.
1069 */
1070 spin_lock_irq(lock);
1071 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1072 /*
1073 * Another group action in progress, just
1074 * return so that the signal is processed.
1075 */
1076 spin_unlock_irq(lock);
1077 return -EAGAIN;
1078 }
1079
1080 sig->group_exec_task = tsk;
1081 sig->notify_count = zap_other_threads(tsk);
1082 if (!thread_group_leader(tsk))
1083 sig->notify_count--;
1084
1085 while (sig->notify_count) {
1086 __set_current_state(TASK_KILLABLE);
1087 spin_unlock_irq(lock);
1088 schedule();
1089 if (__fatal_signal_pending(tsk))
1090 goto killed;
1091 spin_lock_irq(lock);
1092 }
1093 spin_unlock_irq(lock);
1094
1095 /*
1096 * At this point all other threads have exited, all we have to
1097 * do is to wait for the thread group leader to become inactive,
1098 * and to assume its PID:
1099 */
1100 if (!thread_group_leader(tsk)) {
1101 struct task_struct *leader = tsk->group_leader;
1102
1103 for (;;) {
1104 cgroup_threadgroup_change_begin(tsk);
1105 write_lock_irq(&tasklist_lock);
1106 /*
1107 * Do this under tasklist_lock to ensure that
1108 * exit_notify() can't miss ->group_exec_task
1109 */
1110 sig->notify_count = -1;
1111 if (likely(leader->exit_state))
1112 break;
1113 __set_current_state(TASK_KILLABLE);
1114 write_unlock_irq(&tasklist_lock);
1115 cgroup_threadgroup_change_end(tsk);
1116 schedule();
1117 if (__fatal_signal_pending(tsk))
1118 goto killed;
1119 }
1120
1121 /*
1122 * The only record we have of the real-time age of a
1123 * process, regardless of execs it's done, is start_time.
1124 * All the past CPU time is accumulated in signal_struct
1125 * from sister threads now dead. But in this non-leader
1126 * exec, nothing survives from the original leader thread,
1127 * whose birth marks the true age of this process now.
1128 * When we take on its identity by switching to its PID, we
1129 * also take its birthdate (always earlier than our own).
1130 */
1131 tsk->start_time = leader->start_time;
1132 tsk->start_boottime = leader->start_boottime;
1133
1134 BUG_ON(!same_thread_group(leader, tsk));
1135 /*
1136 * An exec() starts a new thread group with the
1137 * TGID of the previous thread group. Rehash the
1138 * two threads with a switched PID, and release
1139 * the former thread group leader:
1140 */
1141
1142 /* Become a process group leader with the old leader's pid.
1143 * The old leader becomes a thread of the this thread group.
1144 */
1145 exchange_tids(tsk, leader);
1146 transfer_pid(leader, tsk, PIDTYPE_TGID);
1147 transfer_pid(leader, tsk, PIDTYPE_PGID);
1148 transfer_pid(leader, tsk, PIDTYPE_SID);
1149
1150 list_replace_rcu(&leader->tasks, &tsk->tasks);
1151 list_replace_init(&leader->sibling, &tsk->sibling);
1152
1153 tsk->group_leader = tsk;
1154 leader->group_leader = tsk;
1155
1156 tsk->exit_signal = SIGCHLD;
1157 leader->exit_signal = -1;
1158
1159 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1160 leader->exit_state = EXIT_DEAD;
1161
1162 /*
1163 * We are going to release_task()->ptrace_unlink() silently,
1164 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1165 * the tracer won't block again waiting for this thread.
1166 */
1167 if (unlikely(leader->ptrace))
1168 __wake_up_parent(leader, leader->parent);
1169 write_unlock_irq(&tasklist_lock);
1170 cgroup_threadgroup_change_end(tsk);
1171
1172 release_task(leader);
1173 }
1174
1175 sig->group_exec_task = NULL;
1176 sig->notify_count = 0;
1177
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(¤t->signal->cred_guard_mutex))
1495 return -ERESTARTNOINTR;
1496
1497 bprm->cred = prepare_exec_creds();
1498 if (likely(bprm->cred))
1499 return 0;
1500
1501 mutex_unlock(¤t->signal->cred_guard_mutex);
1502 return -ENOMEM;
1503}
1504
1505/* Matches do_open_execat() */
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(¤t->signal->cred_guard_mutex);
1523 abort_creds(bprm->cred);
1524 }
1525 do_close_execat(bprm->file);
1526 if (bprm->executable)
1527 fput(bprm->executable);
1528 /* If a binfmt changed the interp, free it. */
1529 if (bprm->interp != bprm->filename)
1530 kfree(bprm->interp);
1531 kfree(bprm->fdpath);
1532 kfree(bprm);
1533}
1534
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 int ret = 0;
1724 unsigned long offset;
1725 char *kaddr;
1726 struct page *page;
1727
1728 if (!bprm->argc)
1729 return 0;
1730
1731 do {
1732 offset = bprm->p & ~PAGE_MASK;
1733 page = get_arg_page(bprm, bprm->p, 0);
1734 if (!page) {
1735 ret = -EFAULT;
1736 goto out;
1737 }
1738 kaddr = kmap_local_page(page);
1739
1740 for (; offset < PAGE_SIZE && kaddr[offset];
1741 offset++, bprm->p++)
1742 ;
1743
1744 kunmap_local(kaddr);
1745 put_arg_page(page);
1746 } while (offset == PAGE_SIZE);
1747
1748 bprm->p++;
1749 bprm->argc--;
1750 ret = 0;
1751
1752out:
1753 return ret;
1754}
1755EXPORT_SYMBOL(remove_arg_zero);
1756
1757#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1758/*
1759 * cycle the list of binary formats handler, until one recognizes the image
1760 */
1761static int search_binary_handler(struct linux_binprm *bprm)
1762{
1763 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1764 struct linux_binfmt *fmt;
1765 int retval;
1766
1767 retval = prepare_binprm(bprm);
1768 if (retval < 0)
1769 return retval;
1770
1771 retval = security_bprm_check(bprm);
1772 if (retval)
1773 return retval;
1774
1775 retval = -ENOENT;
1776 retry:
1777 read_lock(&binfmt_lock);
1778 list_for_each_entry(fmt, &formats, lh) {
1779 if (!try_module_get(fmt->module))
1780 continue;
1781 read_unlock(&binfmt_lock);
1782
1783 retval = fmt->load_binary(bprm);
1784
1785 read_lock(&binfmt_lock);
1786 put_binfmt(fmt);
1787 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1788 read_unlock(&binfmt_lock);
1789 return retval;
1790 }
1791 }
1792 read_unlock(&binfmt_lock);
1793
1794 if (need_retry) {
1795 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1796 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1797 return retval;
1798 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1799 return retval;
1800 need_retry = false;
1801 goto retry;
1802 }
1803
1804 return retval;
1805}
1806
1807/* binfmt handlers will call back into begin_new_exec() on success. */
1808static int exec_binprm(struct linux_binprm *bprm)
1809{
1810 pid_t old_pid, old_vpid;
1811 int ret, depth;
1812
1813 /* Need to fetch pid before load_binary changes it */
1814 old_pid = current->pid;
1815 rcu_read_lock();
1816 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1817 rcu_read_unlock();
1818
1819 /* This allows 4 levels of binfmt rewrites before failing hard. */
1820 for (depth = 0;; depth++) {
1821 struct file *exec;
1822 if (depth > 5)
1823 return -ELOOP;
1824
1825 ret = search_binary_handler(bprm);
1826 if (ret < 0)
1827 return ret;
1828 if (!bprm->interpreter)
1829 break;
1830
1831 exec = bprm->file;
1832 bprm->file = bprm->interpreter;
1833 bprm->interpreter = NULL;
1834
1835 allow_write_access(exec);
1836 if (unlikely(bprm->have_execfd)) {
1837 if (bprm->executable) {
1838 fput(exec);
1839 return -ENOEXEC;
1840 }
1841 bprm->executable = exec;
1842 } else
1843 fput(exec);
1844 }
1845
1846 audit_bprm(bprm);
1847 trace_sched_process_exec(current, old_pid, bprm);
1848 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1849 proc_exec_connector(current);
1850 return 0;
1851}
1852
1853static int bprm_execve(struct linux_binprm *bprm)
1854{
1855 int retval;
1856
1857 retval = prepare_bprm_creds(bprm);
1858 if (retval)
1859 return retval;
1860
1861 /*
1862 * Check for unsafe execution states before exec_binprm(), which
1863 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1864 * where setuid-ness is evaluated.
1865 */
1866 check_unsafe_exec(bprm);
1867 current->in_execve = 1;
1868 sched_mm_cid_before_execve(current);
1869
1870 sched_exec();
1871
1872 /* Set the unchanging part of bprm->cred */
1873 retval = security_bprm_creds_for_exec(bprm);
1874 if (retval)
1875 goto out;
1876
1877 retval = exec_binprm(bprm);
1878 if (retval < 0)
1879 goto out;
1880
1881 sched_mm_cid_after_execve(current);
1882 /* execve succeeded */
1883 current->fs->in_exec = 0;
1884 current->in_execve = 0;
1885 rseq_execve(current);
1886 user_events_execve(current);
1887 acct_update_integrals(current);
1888 task_numa_free(current, false);
1889 return retval;
1890
1891out:
1892 /*
1893 * If past the point of no return ensure the code never
1894 * returns to the userspace process. Use an existing fatal
1895 * signal if present otherwise terminate the process with
1896 * SIGSEGV.
1897 */
1898 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1899 force_fatal_sig(SIGSEGV);
1900
1901 sched_mm_cid_after_execve(current);
1902 current->fs->in_exec = 0;
1903 current->in_execve = 0;
1904
1905 return retval;
1906}
1907
1908static int do_execveat_common(int fd, struct filename *filename,
1909 struct user_arg_ptr argv,
1910 struct user_arg_ptr envp,
1911 int flags)
1912{
1913 struct linux_binprm *bprm;
1914 int retval;
1915
1916 if (IS_ERR(filename))
1917 return PTR_ERR(filename);
1918
1919 /*
1920 * We move the actual failure in case of RLIMIT_NPROC excess from
1921 * set*uid() to execve() because too many poorly written programs
1922 * don't check setuid() return code. Here we additionally recheck
1923 * whether NPROC limit is still exceeded.
1924 */
1925 if ((current->flags & PF_NPROC_EXCEEDED) &&
1926 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1927 retval = -EAGAIN;
1928 goto out_ret;
1929 }
1930
1931 /* We're below the limit (still or again), so we don't want to make
1932 * further execve() calls fail. */
1933 current->flags &= ~PF_NPROC_EXCEEDED;
1934
1935 bprm = alloc_bprm(fd, filename, flags);
1936 if (IS_ERR(bprm)) {
1937 retval = PTR_ERR(bprm);
1938 goto out_ret;
1939 }
1940
1941 retval = count(argv, MAX_ARG_STRINGS);
1942 if (retval == 0)
1943 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1944 current->comm, bprm->filename);
1945 if (retval < 0)
1946 goto out_free;
1947 bprm->argc = retval;
1948
1949 retval = count(envp, MAX_ARG_STRINGS);
1950 if (retval < 0)
1951 goto out_free;
1952 bprm->envc = retval;
1953
1954 retval = bprm_stack_limits(bprm);
1955 if (retval < 0)
1956 goto out_free;
1957
1958 retval = copy_string_kernel(bprm->filename, bprm);
1959 if (retval < 0)
1960 goto out_free;
1961 bprm->exec = bprm->p;
1962
1963 retval = copy_strings(bprm->envc, envp, bprm);
1964 if (retval < 0)
1965 goto out_free;
1966
1967 retval = copy_strings(bprm->argc, argv, bprm);
1968 if (retval < 0)
1969 goto out_free;
1970
1971 /*
1972 * When argv is empty, add an empty string ("") as argv[0] to
1973 * ensure confused userspace programs that start processing
1974 * from argv[1] won't end up walking envp. See also
1975 * bprm_stack_limits().
1976 */
1977 if (bprm->argc == 0) {
1978 retval = copy_string_kernel("", bprm);
1979 if (retval < 0)
1980 goto out_free;
1981 bprm->argc = 1;
1982 }
1983
1984 retval = bprm_execve(bprm);
1985out_free:
1986 free_bprm(bprm);
1987
1988out_ret:
1989 putname(filename);
1990 return retval;
1991}
1992
1993int kernel_execve(const char *kernel_filename,
1994 const char *const *argv, const char *const *envp)
1995{
1996 struct filename *filename;
1997 struct linux_binprm *bprm;
1998 int fd = AT_FDCWD;
1999 int retval;
2000
2001 /* It is non-sense for kernel threads to call execve */
2002 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
2003 return -EINVAL;
2004
2005 filename = getname_kernel(kernel_filename);
2006 if (IS_ERR(filename))
2007 return PTR_ERR(filename);
2008
2009 bprm = alloc_bprm(fd, filename, 0);
2010 if (IS_ERR(bprm)) {
2011 retval = PTR_ERR(bprm);
2012 goto out_ret;
2013 }
2014
2015 retval = count_strings_kernel(argv);
2016 if (WARN_ON_ONCE(retval == 0))
2017 retval = -EINVAL;
2018 if (retval < 0)
2019 goto out_free;
2020 bprm->argc = retval;
2021
2022 retval = count_strings_kernel(envp);
2023 if (retval < 0)
2024 goto out_free;
2025 bprm->envc = retval;
2026
2027 retval = bprm_stack_limits(bprm);
2028 if (retval < 0)
2029 goto out_free;
2030
2031 retval = copy_string_kernel(bprm->filename, bprm);
2032 if (retval < 0)
2033 goto out_free;
2034 bprm->exec = bprm->p;
2035
2036 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2037 if (retval < 0)
2038 goto out_free;
2039
2040 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2041 if (retval < 0)
2042 goto out_free;
2043
2044 retval = bprm_execve(bprm);
2045out_free:
2046 free_bprm(bprm);
2047out_ret:
2048 putname(filename);
2049 return retval;
2050}
2051
2052static int do_execve(struct filename *filename,
2053 const char __user *const __user *__argv,
2054 const char __user *const __user *__envp)
2055{
2056 struct user_arg_ptr argv = { .ptr.native = __argv };
2057 struct user_arg_ptr envp = { .ptr.native = __envp };
2058 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2059}
2060
2061static int do_execveat(int fd, struct filename *filename,
2062 const char __user *const __user *__argv,
2063 const char __user *const __user *__envp,
2064 int flags)
2065{
2066 struct user_arg_ptr argv = { .ptr.native = __argv };
2067 struct user_arg_ptr envp = { .ptr.native = __envp };
2068
2069 return do_execveat_common(fd, filename, argv, envp, flags);
2070}
2071
2072#ifdef CONFIG_COMPAT
2073static int compat_do_execve(struct filename *filename,
2074 const compat_uptr_t __user *__argv,
2075 const compat_uptr_t __user *__envp)
2076{
2077 struct user_arg_ptr argv = {
2078 .is_compat = true,
2079 .ptr.compat = __argv,
2080 };
2081 struct user_arg_ptr envp = {
2082 .is_compat = true,
2083 .ptr.compat = __envp,
2084 };
2085 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2086}
2087
2088static int compat_do_execveat(int fd, struct filename *filename,
2089 const compat_uptr_t __user *__argv,
2090 const compat_uptr_t __user *__envp,
2091 int flags)
2092{
2093 struct user_arg_ptr argv = {
2094 .is_compat = true,
2095 .ptr.compat = __argv,
2096 };
2097 struct user_arg_ptr envp = {
2098 .is_compat = true,
2099 .ptr.compat = __envp,
2100 };
2101 return do_execveat_common(fd, filename, argv, envp, flags);
2102}
2103#endif
2104
2105void set_binfmt(struct linux_binfmt *new)
2106{
2107 struct mm_struct *mm = current->mm;
2108
2109 if (mm->binfmt)
2110 module_put(mm->binfmt->module);
2111
2112 mm->binfmt = new;
2113 if (new)
2114 __module_get(new->module);
2115}
2116EXPORT_SYMBOL(set_binfmt);
2117
2118/*
2119 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2120 */
2121void set_dumpable(struct mm_struct *mm, int value)
2122{
2123 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2124 return;
2125
2126 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2127}
2128
2129SYSCALL_DEFINE3(execve,
2130 const char __user *, filename,
2131 const char __user *const __user *, argv,
2132 const char __user *const __user *, envp)
2133{
2134 return do_execve(getname(filename), argv, envp);
2135}
2136
2137SYSCALL_DEFINE5(execveat,
2138 int, fd, const char __user *, filename,
2139 const char __user *const __user *, argv,
2140 const char __user *const __user *, envp,
2141 int, flags)
2142{
2143 return do_execveat(fd,
2144 getname_uflags(filename, flags),
2145 argv, envp, flags);
2146}
2147
2148#ifdef CONFIG_COMPAT
2149COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2150 const compat_uptr_t __user *, argv,
2151 const compat_uptr_t __user *, envp)
2152{
2153 return compat_do_execve(getname(filename), argv, envp);
2154}
2155
2156COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2157 const char __user *, filename,
2158 const compat_uptr_t __user *, argv,
2159 const compat_uptr_t __user *, envp,
2160 int, flags)
2161{
2162 return compat_do_execveat(fd,
2163 getname_uflags(filename, flags),
2164 argv, envp, flags);
2165}
2166#endif
2167
2168#ifdef CONFIG_SYSCTL
2169
2170static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2171 void *buffer, size_t *lenp, loff_t *ppos)
2172{
2173 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2174
2175 if (!error)
2176 validate_coredump_safety();
2177 return error;
2178}
2179
2180static struct ctl_table fs_exec_sysctls[] = {
2181 {
2182 .procname = "suid_dumpable",
2183 .data = &suid_dumpable,
2184 .maxlen = sizeof(int),
2185 .mode = 0644,
2186 .proc_handler = proc_dointvec_minmax_coredump,
2187 .extra1 = SYSCTL_ZERO,
2188 .extra2 = SYSCTL_TWO,
2189 },
2190};
2191
2192static int __init init_fs_exec_sysctls(void)
2193{
2194 register_sysctl_init("fs", fs_exec_sysctls);
2195 return 0;
2196}
2197
2198fs_initcall(init_fs_exec_sysctls);
2199#endif /* CONFIG_SYSCTL */