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