projects/linux·blob v6.10 · refs

2213 lines · 52.9 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/exec.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * #!-checking implemented by tytso.
10 */
11/*
12 * Demand-loading implemented 01.12.91 - no need to read anything but
13 * the header into memory. The inode of the executable is put into
14 * "current->executable", and page faults do the actual loading. Clean.
15 *
16 * Once more I can proudly say that linux stood up to being changed: it
17 * was less than 2 hours work to get demand-loading completely implemented.
18 *
19 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
20 * current->executable is only used by the procfs. This allows a dispatch
21 * table to check for several different types of binary formats. We keep
22 * trying until we recognize the file or we run out of supported binary
23 * formats.
24 */
25
26#include <linux/kernel_read_file.h>
27#include <linux/slab.h>
28#include <linux/file.h>
29#include <linux/fdtable.h>
30#include <linux/mm.h>
31#include <linux/stat.h>
32#include <linux/fcntl.h>
33#include <linux/swap.h>
34#include <linux/string.h>
35#include <linux/init.h>
36#include <linux/sched/mm.h>
37#include <linux/sched/coredump.h>
38#include <linux/sched/signal.h>
39#include <linux/sched/numa_balancing.h>
40#include <linux/sched/task.h>
41#include <linux/pagemap.h>
42#include <linux/perf_event.h>
43#include <linux/highmem.h>
44#include <linux/spinlock.h>
45#include <linux/key.h>
46#include <linux/personality.h>
47#include <linux/binfmts.h>
48#include <linux/utsname.h>
49#include <linux/pid_namespace.h>
50#include <linux/module.h>
51#include <linux/namei.h>
52#include <linux/mount.h>
53#include <linux/security.h>
54#include <linux/syscalls.h>
55#include <linux/tsacct_kern.h>
56#include <linux/cn_proc.h>
57#include <linux/audit.h>
58#include <linux/kmod.h>
59#include <linux/fsnotify.h>
60#include <linux/fs_struct.h>
61#include <linux/oom.h>
62#include <linux/compat.h>
63#include <linux/vmalloc.h>
64#include <linux/io_uring.h>
65#include <linux/syscall_user_dispatch.h>
66#include <linux/coredump.h>
67#include <linux/time_namespace.h>
68#include <linux/user_events.h>
69#include <linux/rseq.h>
70#include <linux/ksm.h>
71
72#include <linux/uaccess.h>
73#include <asm/mmu_context.h>
74#include <asm/tlb.h>
75
76#include <trace/events/task.h>
77#include "internal.h"
78
79#include <trace/events/sched.h>
80
81static int bprm_creds_from_file(struct linux_binprm *bprm);
82
83int suid_dumpable = 0;
84
85static LIST_HEAD(formats);
86static DEFINE_RWLOCK(binfmt_lock);
87
88void __register_binfmt(struct linux_binfmt * fmt, int insert)
89{
90 write_lock(&binfmt_lock);
91 insert ? list_add(&fmt->lh, &formats) :
92 list_add_tail(&fmt->lh, &formats);
93 write_unlock(&binfmt_lock);
94}
95
96EXPORT_SYMBOL(__register_binfmt);
97
98void unregister_binfmt(struct linux_binfmt * fmt)
99{
100 write_lock(&binfmt_lock);
101 list_del(&fmt->lh);
102 write_unlock(&binfmt_lock);
103}
104
105EXPORT_SYMBOL(unregister_binfmt);
106
107static inline void put_binfmt(struct linux_binfmt * fmt)
108{
109 module_put(fmt->module);
110}
111
112bool path_noexec(const struct path *path)
113{
114 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
115 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
116}
117
118#ifdef CONFIG_USELIB
119/*
120 * Note that a shared library must be both readable and executable due to
121 * security reasons.
122 *
123 * Also note that we take the address to load from the file itself.
124 */
125SYSCALL_DEFINE1(uselib, const char __user *, library)
126{
127 struct linux_binfmt *fmt;
128 struct file *file;
129 struct filename *tmp = getname(library);
130 int error = PTR_ERR(tmp);
131 static const struct open_flags uselib_flags = {
132 .open_flag = O_LARGEFILE | O_RDONLY,
133 .acc_mode = MAY_READ | MAY_EXEC,
134 .intent = LOOKUP_OPEN,
135 .lookup_flags = LOOKUP_FOLLOW,
136 };
137
138 if (IS_ERR(tmp))
139 goto out;
140
141 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
142 putname(tmp);
143 error = PTR_ERR(file);
144 if (IS_ERR(file))
145 goto out;
146
147 /*
148 * may_open() has already checked for this, so it should be
149 * impossible to trip now. But we need to be extra cautious
150 * and check again at the very end too.
151 */
152 error = -EACCES;
153 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
154 path_noexec(&file->f_path)))
155 goto exit;
156
157 error = -ENOEXEC;
158
159 read_lock(&binfmt_lock);
160 list_for_each_entry(fmt, &formats, lh) {
161 if (!fmt->load_shlib)
162 continue;
163 if (!try_module_get(fmt->module))
164 continue;
165 read_unlock(&binfmt_lock);
166 error = fmt->load_shlib(file);
167 read_lock(&binfmt_lock);
168 put_binfmt(fmt);
169 if (error != -ENOEXEC)
170 break;
171 }
172 read_unlock(&binfmt_lock);
173exit:
174 fput(file);
175out:
176 return error;
177}
178#endif /* #ifdef CONFIG_USELIB */
179
180#ifdef CONFIG_MMU
181/*
182 * The nascent bprm->mm is not visible until exec_mmap() but it can
183 * use a lot of memory, account these pages in current->mm temporary
184 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
185 * change the counter back via acct_arg_size(0).
186 */
187static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
188{
189 struct mm_struct *mm = current->mm;
190 long diff = (long)(pages - bprm->vma_pages);
191
192 if (!mm || !diff)
193 return;
194
195 bprm->vma_pages = pages;
196 add_mm_counter(mm, MM_ANONPAGES, diff);
197}
198
199static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
200 int write)
201{
202 struct page *page;
203 struct vm_area_struct *vma = bprm->vma;
204 struct mm_struct *mm = bprm->mm;
205 int ret;
206
207 /*
208 * Avoid relying on expanding the stack down in GUP (which
209 * does not work for STACK_GROWSUP anyway), and just do it
210 * by hand ahead of time.
211 */
212 if (write && pos < vma->vm_start) {
213 mmap_write_lock(mm);
214 ret = expand_downwards(vma, pos);
215 if (unlikely(ret < 0)) {
216 mmap_write_unlock(mm);
217 return NULL;
218 }
219 mmap_write_downgrade(mm);
220 } else
221 mmap_read_lock(mm);
222
223 /*
224 * We are doing an exec(). 'current' is the process
225 * doing the exec and 'mm' is the new process's mm.
226 */
227 ret = get_user_pages_remote(mm, pos, 1,
228 write ? FOLL_WRITE : 0,
229 &page, NULL);
230 mmap_read_unlock(mm);
231 if (ret <= 0)
232 return NULL;
233
234 if (write)
235 acct_arg_size(bprm, vma_pages(vma));
236
237 return page;
238}
239
240static void put_arg_page(struct page *page)
241{
242 put_page(page);
243}
244
245static void free_arg_pages(struct linux_binprm *bprm)
246{
247}
248
249static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
250 struct page *page)
251{
252 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
253}
254
255static int __bprm_mm_init(struct linux_binprm *bprm)
256{
257 int err;
258 struct vm_area_struct *vma = NULL;
259 struct mm_struct *mm = bprm->mm;
260
261 bprm->vma = vma = vm_area_alloc(mm);
262 if (!vma)
263 return -ENOMEM;
264 vma_set_anonymous(vma);
265
266 if (mmap_write_lock_killable(mm)) {
267 err = -EINTR;
268 goto err_free;
269 }
270
271 /*
272 * Need to be called with mmap write lock
273 * held, to avoid race with ksmd.
274 */
275 err = ksm_execve(mm);
276 if (err)
277 goto err_ksm;
278
279 /*
280 * Place the stack at the largest stack address the architecture
281 * supports. Later, we'll move this to an appropriate place. We don't
282 * use STACK_TOP because that can depend on attributes which aren't
283 * configured yet.
284 */
285 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
286 vma->vm_end = STACK_TOP_MAX;
287 vma->vm_start = vma->vm_end - PAGE_SIZE;
288 vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);
289 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
290
291 err = insert_vm_struct(mm, vma);
292 if (err)
293 goto err;
294
295 mm->stack_vm = mm->total_vm = 1;
296 mmap_write_unlock(mm);
297 bprm->p = vma->vm_end - sizeof(void *);
298 return 0;
299err:
300 ksm_exit(mm);
301err_ksm:
302 mmap_write_unlock(mm);
303err_free:
304 bprm->vma = NULL;
305 vm_area_free(vma);
306 return err;
307}
308
309static bool valid_arg_len(struct linux_binprm *bprm, long len)
310{
311 return len <= MAX_ARG_STRLEN;
312}
313
314#else
315
316static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
317{
318}
319
320static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
321 int write)
322{
323 struct page *page;
324
325 page = bprm->page[pos / PAGE_SIZE];
326 if (!page && write) {
327 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
328 if (!page)
329 return NULL;
330 bprm->page[pos / PAGE_SIZE] = page;
331 }
332
333 return page;
334}
335
336static void put_arg_page(struct page *page)
337{
338}
339
340static void free_arg_page(struct linux_binprm *bprm, int i)
341{
342 if (bprm->page[i]) {
343 __free_page(bprm->page[i]);
344 bprm->page[i] = NULL;
345 }
346}
347
348static void free_arg_pages(struct linux_binprm *bprm)
349{
350 int i;
351
352 for (i = 0; i < MAX_ARG_PAGES; i++)
353 free_arg_page(bprm, i);
354}
355
356static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
357 struct page *page)
358{
359}
360
361static int __bprm_mm_init(struct linux_binprm *bprm)
362{
363 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
364 return 0;
365}
366
367static bool valid_arg_len(struct linux_binprm *bprm, long len)
368{
369 return len <= bprm->p;
370}
371
372#endif /* CONFIG_MMU */
373
374/*
375 * Create a new mm_struct and populate it with a temporary stack
376 * vm_area_struct. We don't have enough context at this point to set the stack
377 * flags, permissions, and offset, so we use temporary values. We'll update
378 * them later in setup_arg_pages().
379 */
380static int bprm_mm_init(struct linux_binprm *bprm)
381{
382 int err;
383 struct mm_struct *mm = NULL;
384
385 bprm->mm = mm = mm_alloc();
386 err = -ENOMEM;
387 if (!mm)
388 goto err;
389
390 /* Save current stack limit for all calculations made during exec. */
391 task_lock(current->group_leader);
392 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
393 task_unlock(current->group_leader);
394
395 err = __bprm_mm_init(bprm);
396 if (err)
397 goto err;
398
399 return 0;
400
401err:
402 if (mm) {
403 bprm->mm = NULL;
404 mmdrop(mm);
405 }
406
407 return err;
408}
409
410struct user_arg_ptr {
411#ifdef CONFIG_COMPAT
412 bool is_compat;
413#endif
414 union {
415 const char __user *const __user *native;
416#ifdef CONFIG_COMPAT
417 const compat_uptr_t __user *compat;
418#endif
419 } ptr;
420};
421
422static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
423{
424 const char __user *native;
425
426#ifdef CONFIG_COMPAT
427 if (unlikely(argv.is_compat)) {
428 compat_uptr_t compat;
429
430 if (get_user(compat, argv.ptr.compat + nr))
431 return ERR_PTR(-EFAULT);
432
433 return compat_ptr(compat);
434 }
435#endif
436
437 if (get_user(native, argv.ptr.native + nr))
438 return ERR_PTR(-EFAULT);
439
440 return native;
441}
442
443/*
444 * count() counts the number of strings in array ARGV.
445 */
446static int count(struct user_arg_ptr argv, int max)
447{
448 int i = 0;
449
450 if (argv.ptr.native != NULL) {
451 for (;;) {
452 const char __user *p = get_user_arg_ptr(argv, i);
453
454 if (!p)
455 break;
456
457 if (IS_ERR(p))
458 return -EFAULT;
459
460 if (i >= max)
461 return -E2BIG;
462 ++i;
463
464 if (fatal_signal_pending(current))
465 return -ERESTARTNOHAND;
466 cond_resched();
467 }
468 }
469 return i;
470}
471
472static int count_strings_kernel(const char *const *argv)
473{
474 int i;
475
476 if (!argv)
477 return 0;
478
479 for (i = 0; argv[i]; ++i) {
480 if (i >= MAX_ARG_STRINGS)
481 return -E2BIG;
482 if (fatal_signal_pending(current))
483 return -ERESTARTNOHAND;
484 cond_resched();
485 }
486 return i;
487}
488
489static int bprm_stack_limits(struct linux_binprm *bprm)
490{
491 unsigned long limit, ptr_size;
492
493 /*
494 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
495 * (whichever is smaller) for the argv+env strings.
496 * This ensures that:
497 * - the remaining binfmt code will not run out of stack space,
498 * - the program will have a reasonable amount of stack left
499 * to work from.
500 */
501 limit = _STK_LIM / 4 * 3;
502 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
503 /*
504 * We've historically supported up to 32 pages (ARG_MAX)
505 * of argument strings even with small stacks
506 */
507 limit = max_t(unsigned long, limit, ARG_MAX);
508 /*
509 * We must account for the size of all the argv and envp pointers to
510 * the argv and envp strings, since they will also take up space in
511 * the stack. They aren't stored until much later when we can't
512 * signal to the parent that the child has run out of stack space.
513 * Instead, calculate it here so it's possible to fail gracefully.
514 *
515 * In the case of argc = 0, make sure there is space for adding a
516 * empty string (which will bump argc to 1), to ensure confused
517 * userspace programs don't start processing from argv[1], thinking
518 * argc can never be 0, to keep them from walking envp by accident.
519 * See do_execveat_common().
520 */
521 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
522 if (limit <= ptr_size)
523 return -E2BIG;
524 limit -= ptr_size;
525
526 bprm->argmin = bprm->p - limit;
527 return 0;
528}
529
530/*
531 * 'copy_strings()' copies argument/environment strings from the old
532 * processes's memory to the new process's stack. The call to get_user_pages()
533 * ensures the destination page is created and not swapped out.
534 */
535static int copy_strings(int argc, struct user_arg_ptr argv,
536 struct linux_binprm *bprm)
537{
538 struct page *kmapped_page = NULL;
539 char *kaddr = NULL;
540 unsigned long kpos = 0;
541 int ret;
542
543 while (argc-- > 0) {
544 const char __user *str;
545 int len;
546 unsigned long pos;
547
548 ret = -EFAULT;
549 str = get_user_arg_ptr(argv, argc);
550 if (IS_ERR(str))
551 goto out;
552
553 len = strnlen_user(str, MAX_ARG_STRLEN);
554 if (!len)
555 goto out;
556
557 ret = -E2BIG;
558 if (!valid_arg_len(bprm, len))
559 goto out;
560
561 /* We're going to work our way backwards. */
562 pos = bprm->p;
563 str += len;
564 bprm->p -= len;
565#ifdef CONFIG_MMU
566 if (bprm->p < bprm->argmin)
567 goto out;
568#endif
569
570 while (len > 0) {
571 int offset, bytes_to_copy;
572
573 if (fatal_signal_pending(current)) {
574 ret = -ERESTARTNOHAND;
575 goto out;
576 }
577 cond_resched();
578
579 offset = pos % PAGE_SIZE;
580 if (offset == 0)
581 offset = PAGE_SIZE;
582
583 bytes_to_copy = offset;
584 if (bytes_to_copy > len)
585 bytes_to_copy = len;
586
587 offset -= bytes_to_copy;
588 pos -= bytes_to_copy;
589 str -= bytes_to_copy;
590 len -= bytes_to_copy;
591
592 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
593 struct page *page;
594
595 page = get_arg_page(bprm, pos, 1);
596 if (!page) {
597 ret = -E2BIG;
598 goto out;
599 }
600
601 if (kmapped_page) {
602 flush_dcache_page(kmapped_page);
603 kunmap_local(kaddr);
604 put_arg_page(kmapped_page);
605 }
606 kmapped_page = page;
607 kaddr = kmap_local_page(kmapped_page);
608 kpos = pos & PAGE_MASK;
609 flush_arg_page(bprm, kpos, kmapped_page);
610 }
611 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
612 ret = -EFAULT;
613 goto out;
614 }
615 }
616 }
617 ret = 0;
618out:
619 if (kmapped_page) {
620 flush_dcache_page(kmapped_page);
621 kunmap_local(kaddr);
622 put_arg_page(kmapped_page);
623 }
624 return ret;
625}
626
627/*
628 * Copy and argument/environment string from the kernel to the processes stack.
629 */
630int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
631{
632 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
633 unsigned long pos = bprm->p;
634
635 if (len == 0)
636 return -EFAULT;
637 if (!valid_arg_len(bprm, len))
638 return -E2BIG;
639
640 /* We're going to work our way backwards. */
641 arg += len;
642 bprm->p -= len;
643 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
644 return -E2BIG;
645
646 while (len > 0) {
647 unsigned int bytes_to_copy = min_t(unsigned int, len,
648 min_not_zero(offset_in_page(pos), PAGE_SIZE));
649 struct page *page;
650
651 pos -= bytes_to_copy;
652 arg -= bytes_to_copy;
653 len -= bytes_to_copy;
654
655 page = get_arg_page(bprm, pos, 1);
656 if (!page)
657 return -E2BIG;
658 flush_arg_page(bprm, pos & PAGE_MASK, page);
659 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
660 put_arg_page(page);
661 }
662
663 return 0;
664}
665EXPORT_SYMBOL(copy_string_kernel);
666
667static int copy_strings_kernel(int argc, const char *const *argv,
668 struct linux_binprm *bprm)
669{
670 while (argc-- > 0) {
671 int ret = copy_string_kernel(argv[argc], bprm);
672 if (ret < 0)
673 return ret;
674 if (fatal_signal_pending(current))
675 return -ERESTARTNOHAND;
676 cond_resched();
677 }
678 return 0;
679}
680
681#ifdef CONFIG_MMU
682
683/*
684 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
685 * the binfmt code determines where the new stack should reside, we shift it to
686 * its final location. The process proceeds as follows:
687 *
688 * 1) Use shift to calculate the new vma endpoints.
689 * 2) Extend vma to cover both the old and new ranges. This ensures the
690 * arguments passed to subsequent functions are consistent.
691 * 3) Move vma's page tables to the new range.
692 * 4) Free up any cleared pgd range.
693 * 5) Shrink the vma to cover only the new range.
694 */
695static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
696{
697 struct mm_struct *mm = vma->vm_mm;
698 unsigned long old_start = vma->vm_start;
699 unsigned long old_end = vma->vm_end;
700 unsigned long length = old_end - old_start;
701 unsigned long new_start = old_start - shift;
702 unsigned long new_end = old_end - shift;
703 VMA_ITERATOR(vmi, mm, new_start);
704 struct vm_area_struct *next;
705 struct mmu_gather tlb;
706
707 BUG_ON(new_start > new_end);
708
709 /*
710 * ensure there are no vmas between where we want to go
711 * and where we are
712 */
713 if (vma != vma_next(&vmi))
714 return -EFAULT;
715
716 vma_iter_prev_range(&vmi);
717 /*
718 * cover the whole range: [new_start, old_end)
719 */
720 if (vma_expand(&vmi, vma, new_start, old_end, vma->vm_pgoff, NULL))
721 return -ENOMEM;
722
723 /*
724 * move the page tables downwards, on failure we rely on
725 * process cleanup to remove whatever mess we made.
726 */
727 if (length != move_page_tables(vma, old_start,
728 vma, new_start, length, false, true))
729 return -ENOMEM;
730
731 lru_add_drain();
732 tlb_gather_mmu(&tlb, mm);
733 next = vma_next(&vmi);
734 if (new_end > old_start) {
735 /*
736 * when the old and new regions overlap clear from new_end.
737 */
738 free_pgd_range(&tlb, new_end, old_end, new_end,
739 next ? next->vm_start : USER_PGTABLES_CEILING);
740 } else {
741 /*
742 * otherwise, clean from old_start; this is done to not touch
743 * the address space in [new_end, old_start) some architectures
744 * have constraints on va-space that make this illegal (IA64) -
745 * for the others its just a little faster.
746 */
747 free_pgd_range(&tlb, old_start, old_end, new_end,
748 next ? next->vm_start : USER_PGTABLES_CEILING);
749 }
750 tlb_finish_mmu(&tlb);
751
752 vma_prev(&vmi);
753 /* Shrink the vma to just the new range */
754 return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff);
755}
756
757/*
758 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
759 * the stack is optionally relocated, and some extra space is added.
760 */
761int setup_arg_pages(struct linux_binprm *bprm,
762 unsigned long stack_top,
763 int executable_stack)
764{
765 unsigned long ret;
766 unsigned long stack_shift;
767 struct mm_struct *mm = current->mm;
768 struct vm_area_struct *vma = bprm->vma;
769 struct vm_area_struct *prev = NULL;
770 unsigned long vm_flags;
771 unsigned long stack_base;
772 unsigned long stack_size;
773 unsigned long stack_expand;
774 unsigned long rlim_stack;
775 struct mmu_gather tlb;
776 struct vma_iterator vmi;
777
778#ifdef CONFIG_STACK_GROWSUP
779 /* Limit stack size */
780 stack_base = bprm->rlim_stack.rlim_max;
781
782 stack_base = calc_max_stack_size(stack_base);
783
784 /* Add space for stack randomization. */
785 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
786
787 /* Make sure we didn't let the argument array grow too large. */
788 if (vma->vm_end - vma->vm_start > stack_base)
789 return -ENOMEM;
790
791 stack_base = PAGE_ALIGN(stack_top - stack_base);
792
793 stack_shift = vma->vm_start - stack_base;
794 mm->arg_start = bprm->p - stack_shift;
795 bprm->p = vma->vm_end - stack_shift;
796#else
797 stack_top = arch_align_stack(stack_top);
798 stack_top = PAGE_ALIGN(stack_top);
799
800 if (unlikely(stack_top < mmap_min_addr) ||
801 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
802 return -ENOMEM;
803
804 stack_shift = vma->vm_end - stack_top;
805
806 bprm->p -= stack_shift;
807 mm->arg_start = bprm->p;
808#endif
809
810 if (bprm->loader)
811 bprm->loader -= stack_shift;
812 bprm->exec -= stack_shift;
813
814 if (mmap_write_lock_killable(mm))
815 return -EINTR;
816
817 vm_flags = VM_STACK_FLAGS;
818
819 /*
820 * Adjust stack execute permissions; explicitly enable for
821 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
822 * (arch default) otherwise.
823 */
824 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
825 vm_flags |= VM_EXEC;
826 else if (executable_stack == EXSTACK_DISABLE_X)
827 vm_flags &= ~VM_EXEC;
828 vm_flags |= mm->def_flags;
829 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
830
831 vma_iter_init(&vmi, mm, vma->vm_start);
832
833 tlb_gather_mmu(&tlb, mm);
834 ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
835 vm_flags);
836 tlb_finish_mmu(&tlb);
837
838 if (ret)
839 goto out_unlock;
840 BUG_ON(prev != vma);
841
842 if (unlikely(vm_flags & VM_EXEC)) {
843 pr_warn_once("process '%pD4' started with executable stack\n",
844 bprm->file);
845 }
846
847 /* Move stack pages down in memory. */
848 if (stack_shift) {
849 ret = shift_arg_pages(vma, stack_shift);
850 if (ret)
851 goto out_unlock;
852 }
853
854 /* mprotect_fixup is overkill to remove the temporary stack flags */
855 vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
856
857 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
858 stack_size = vma->vm_end - vma->vm_start;
859 /*
860 * Align this down to a page boundary as expand_stack
861 * will align it up.
862 */
863 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
864
865 stack_expand = min(rlim_stack, stack_size + stack_expand);
866
867#ifdef CONFIG_STACK_GROWSUP
868 stack_base = vma->vm_start + stack_expand;
869#else
870 stack_base = vma->vm_end - stack_expand;
871#endif
872 current->mm->start_stack = bprm->p;
873 ret = expand_stack_locked(vma, stack_base);
874 if (ret)
875 ret = -EFAULT;
876
877out_unlock:
878 mmap_write_unlock(mm);
879 return ret;
880}
881EXPORT_SYMBOL(setup_arg_pages);
882
883#else
884
885/*
886 * Transfer the program arguments and environment from the holding pages
887 * onto the stack. The provided stack pointer is adjusted accordingly.
888 */
889int transfer_args_to_stack(struct linux_binprm *bprm,
890 unsigned long *sp_location)
891{
892 unsigned long index, stop, sp;
893 int ret = 0;
894
895 stop = bprm->p >> PAGE_SHIFT;
896 sp = *sp_location;
897
898 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
899 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
900 char *src = kmap_local_page(bprm->page[index]) + offset;
901 sp -= PAGE_SIZE - offset;
902 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
903 ret = -EFAULT;
904 kunmap_local(src);
905 if (ret)
906 goto out;
907 }
908
909 bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
910 *sp_location = sp;
911
912out:
913 return ret;
914}
915EXPORT_SYMBOL(transfer_args_to_stack);
916
917#endif /* CONFIG_MMU */
918
919/*
920 * On success, caller must call do_close_execat() on the returned
921 * struct file to close it.
922 */
923static struct file *do_open_execat(int fd, struct filename *name, int flags)
924{
925 struct file *file;
926 int err;
927 struct open_flags open_exec_flags = {
928 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
929 .acc_mode = MAY_EXEC,
930 .intent = LOOKUP_OPEN,
931 .lookup_flags = LOOKUP_FOLLOW,
932 };
933
934 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
935 return ERR_PTR(-EINVAL);
936 if (flags & AT_SYMLINK_NOFOLLOW)
937 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
938 if (flags & AT_EMPTY_PATH)
939 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
940
941 file = do_filp_open(fd, name, &open_exec_flags);
942 if (IS_ERR(file))
943 goto out;
944
945 /*
946 * may_open() has already checked for this, so it should be
947 * impossible to trip now. But we need to be extra cautious
948 * and check again at the very end too.
949 */
950 err = -EACCES;
951 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
952 path_noexec(&file->f_path)))
953 goto exit;
954
955 err = deny_write_access(file);
956 if (err)
957 goto exit;
958
959out:
960 return file;
961
962exit:
963 fput(file);
964 return ERR_PTR(err);
965}
966
967/**
968 * open_exec - Open a path name for execution
969 *
970 * @name: path name to open with the intent of executing it.
971 *
972 * Returns ERR_PTR on failure or allocated struct file on success.
973 *
974 * As this is a wrapper for the internal do_open_execat(), callers
975 * must call allow_write_access() before fput() on release. Also see
976 * do_close_execat().
977 */
978struct file *open_exec(const char *name)
979{
980 struct filename *filename = getname_kernel(name);
981 struct file *f = ERR_CAST(filename);
982
983 if (!IS_ERR(filename)) {
984 f = do_open_execat(AT_FDCWD, filename, 0);
985 putname(filename);
986 }
987 return f;
988}
989EXPORT_SYMBOL(open_exec);
990
991#if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
992ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
993{
994 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
995 if (res > 0)
996 flush_icache_user_range(addr, addr + len);
997 return res;
998}
999EXPORT_SYMBOL(read_code);
1000#endif
1001
1002/*
1003 * Maps the mm_struct mm into the current task struct.
1004 * On success, this function returns with exec_update_lock
1005 * held for writing.
1006 */
1007static int exec_mmap(struct mm_struct *mm)
1008{
1009 struct task_struct *tsk;
1010 struct mm_struct *old_mm, *active_mm;
1011 int ret;
1012
1013 /* Notify parent that we're no longer interested in the old VM */
1014 tsk = current;
1015 old_mm = current->mm;
1016 exec_mm_release(tsk, old_mm);
1017
1018 ret = down_write_killable(&tsk->signal->exec_update_lock);
1019 if (ret)
1020 return ret;
1021
1022 if (old_mm) {
1023 /*
1024 * If there is a pending fatal signal perhaps a signal
1025 * whose default action is to create a coredump get
1026 * out and die instead of going through with the exec.
1027 */
1028 ret = mmap_read_lock_killable(old_mm);
1029 if (ret) {
1030 up_write(&tsk->signal->exec_update_lock);
1031 return ret;
1032 }
1033 }
1034
1035 task_lock(tsk);
1036 membarrier_exec_mmap(mm);
1037
1038 local_irq_disable();
1039 active_mm = tsk->active_mm;
1040 tsk->active_mm = mm;
1041 tsk->mm = mm;
1042 mm_init_cid(mm);
1043 /*
1044 * This prevents preemption while active_mm is being loaded and
1045 * it and mm are being updated, which could cause problems for
1046 * lazy tlb mm refcounting when these are updated by context
1047 * switches. Not all architectures can handle irqs off over
1048 * activate_mm yet.
1049 */
1050 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1051 local_irq_enable();
1052 activate_mm(active_mm, mm);
1053 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1054 local_irq_enable();
1055 lru_gen_add_mm(mm);
1056 task_unlock(tsk);
1057 lru_gen_use_mm(mm);
1058 if (old_mm) {
1059 mmap_read_unlock(old_mm);
1060 BUG_ON(active_mm != old_mm);
1061 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1062 mm_update_next_owner(old_mm);
1063 mmput(old_mm);
1064 return 0;
1065 }
1066 mmdrop_lazy_tlb(active_mm);
1067 return 0;
1068}
1069
1070static int de_thread(struct task_struct *tsk)
1071{
1072 struct signal_struct *sig = tsk->signal;
1073 struct sighand_struct *oldsighand = tsk->sighand;
1074 spinlock_t *lock = &oldsighand->siglock;
1075
1076 if (thread_group_empty(tsk))
1077 goto no_thread_group;
1078
1079 /*
1080 * Kill all other threads in the thread group.
1081 */
1082 spin_lock_irq(lock);
1083 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1084 /*
1085 * Another group action in progress, just
1086 * return so that the signal is processed.
1087 */
1088 spin_unlock_irq(lock);
1089 return -EAGAIN;
1090 }
1091
1092 sig->group_exec_task = tsk;
1093 sig->notify_count = zap_other_threads(tsk);
1094 if (!thread_group_leader(tsk))
1095 sig->notify_count--;
1096
1097 while (sig->notify_count) {
1098 __set_current_state(TASK_KILLABLE);
1099 spin_unlock_irq(lock);
1100 schedule();
1101 if (__fatal_signal_pending(tsk))
1102 goto killed;
1103 spin_lock_irq(lock);
1104 }
1105 spin_unlock_irq(lock);
1106
1107 /*
1108 * At this point all other threads have exited, all we have to
1109 * do is to wait for the thread group leader to become inactive,
1110 * and to assume its PID:
1111 */
1112 if (!thread_group_leader(tsk)) {
1113 struct task_struct *leader = tsk->group_leader;
1114
1115 for (;;) {
1116 cgroup_threadgroup_change_begin(tsk);
1117 write_lock_irq(&tasklist_lock);
1118 /*
1119 * Do this under tasklist_lock to ensure that
1120 * exit_notify() can't miss ->group_exec_task
1121 */
1122 sig->notify_count = -1;
1123 if (likely(leader->exit_state))
1124 break;
1125 __set_current_state(TASK_KILLABLE);
1126 write_unlock_irq(&tasklist_lock);
1127 cgroup_threadgroup_change_end(tsk);
1128 schedule();
1129 if (__fatal_signal_pending(tsk))
1130 goto killed;
1131 }
1132
1133 /*
1134 * The only record we have of the real-time age of a
1135 * process, regardless of execs it's done, is start_time.
1136 * All the past CPU time is accumulated in signal_struct
1137 * from sister threads now dead. But in this non-leader
1138 * exec, nothing survives from the original leader thread,
1139 * whose birth marks the true age of this process now.
1140 * When we take on its identity by switching to its PID, we
1141 * also take its birthdate (always earlier than our own).
1142 */
1143 tsk->start_time = leader->start_time;
1144 tsk->start_boottime = leader->start_boottime;
1145
1146 BUG_ON(!same_thread_group(leader, tsk));
1147 /*
1148 * An exec() starts a new thread group with the
1149 * TGID of the previous thread group. Rehash the
1150 * two threads with a switched PID, and release
1151 * the former thread group leader:
1152 */
1153
1154 /* Become a process group leader with the old leader's pid.
1155 * The old leader becomes a thread of the this thread group.
1156 */
1157 exchange_tids(tsk, leader);
1158 transfer_pid(leader, tsk, PIDTYPE_TGID);
1159 transfer_pid(leader, tsk, PIDTYPE_PGID);
1160 transfer_pid(leader, tsk, PIDTYPE_SID);
1161
1162 list_replace_rcu(&leader->tasks, &tsk->tasks);
1163 list_replace_init(&leader->sibling, &tsk->sibling);
1164
1165 tsk->group_leader = tsk;
1166 leader->group_leader = tsk;
1167
1168 tsk->exit_signal = SIGCHLD;
1169 leader->exit_signal = -1;
1170
1171 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1172 leader->exit_state = EXIT_DEAD;
1173 /*
1174 * We are going to release_task()->ptrace_unlink() silently,
1175 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1176 * the tracer won't block again waiting for this thread.
1177 */
1178 if (unlikely(leader->ptrace))
1179 __wake_up_parent(leader, leader->parent);
1180 write_unlock_irq(&tasklist_lock);
1181 cgroup_threadgroup_change_end(tsk);
1182
1183 release_task(leader);
1184 }
1185
1186 sig->group_exec_task = NULL;
1187 sig->notify_count = 0;
1188
1189no_thread_group:
1190 /* we have changed execution domain */
1191 tsk->exit_signal = SIGCHLD;
1192
1193 BUG_ON(!thread_group_leader(tsk));
1194 return 0;
1195
1196killed:
1197 /* protects against exit_notify() and __exit_signal() */
1198 read_lock(&tasklist_lock);
1199 sig->group_exec_task = NULL;
1200 sig->notify_count = 0;
1201 read_unlock(&tasklist_lock);
1202 return -EAGAIN;
1203}
1204
1205
1206/*
1207 * This function makes sure the current process has its own signal table,
1208 * so that flush_signal_handlers can later reset the handlers without
1209 * disturbing other processes. (Other processes might share the signal
1210 * table via the CLONE_SIGHAND option to clone().)
1211 */
1212static int unshare_sighand(struct task_struct *me)
1213{
1214 struct sighand_struct *oldsighand = me->sighand;
1215
1216 if (refcount_read(&oldsighand->count) != 1) {
1217 struct sighand_struct *newsighand;
1218 /*
1219 * This ->sighand is shared with the CLONE_SIGHAND
1220 * but not CLONE_THREAD task, switch to the new one.
1221 */
1222 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1223 if (!newsighand)
1224 return -ENOMEM;
1225
1226 refcount_set(&newsighand->count, 1);
1227
1228 write_lock_irq(&tasklist_lock);
1229 spin_lock(&oldsighand->siglock);
1230 memcpy(newsighand->action, oldsighand->action,
1231 sizeof(newsighand->action));
1232 rcu_assign_pointer(me->sighand, newsighand);
1233 spin_unlock(&oldsighand->siglock);
1234 write_unlock_irq(&tasklist_lock);
1235
1236 __cleanup_sighand(oldsighand);
1237 }
1238 return 0;
1239}
1240
1241char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1242{
1243 task_lock(tsk);
1244 /* Always NUL terminated and zero-padded */
1245 strscpy_pad(buf, tsk->comm, buf_size);
1246 task_unlock(tsk);
1247 return buf;
1248}
1249EXPORT_SYMBOL_GPL(__get_task_comm);
1250
1251/*
1252 * These functions flushes out all traces of the currently running executable
1253 * so that a new one can be started
1254 */
1255
1256void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1257{
1258 task_lock(tsk);
1259 trace_task_rename(tsk, buf);
1260 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1261 task_unlock(tsk);
1262 perf_event_comm(tsk, exec);
1263}
1264
1265/*
1266 * Calling this is the point of no return. None of the failures will be
1267 * seen by userspace since either the process is already taking a fatal
1268 * signal (via de_thread() or coredump), or will have SEGV raised
1269 * (after exec_mmap()) by search_binary_handler (see below).
1270 */
1271int begin_new_exec(struct linux_binprm * bprm)
1272{
1273 struct task_struct *me = current;
1274 int retval;
1275
1276 /* Once we are committed compute the creds */
1277 retval = bprm_creds_from_file(bprm);
1278 if (retval)
1279 return retval;
1280
1281 /*
1282 * This tracepoint marks the point before flushing the old exec where
1283 * the current task is still unchanged, but errors are fatal (point of
1284 * no return). The later "sched_process_exec" tracepoint is called after
1285 * the current task has successfully switched to the new exec.
1286 */
1287 trace_sched_prepare_exec(current, bprm);
1288
1289 /*
1290 * Ensure all future errors are fatal.
1291 */
1292 bprm->point_of_no_return = true;
1293
1294 /*
1295 * Make this the only thread in the thread group.
1296 */
1297 retval = de_thread(me);
1298 if (retval)
1299 goto out;
1300
1301 /*
1302 * Cancel any io_uring activity across execve
1303 */
1304 io_uring_task_cancel();
1305
1306 /* Ensure the files table is not shared. */
1307 retval = unshare_files();
1308 if (retval)
1309 goto out;
1310
1311 /*
1312 * Must be called _before_ exec_mmap() as bprm->mm is
1313 * not visible until then. Doing it here also ensures
1314 * we don't race against replace_mm_exe_file().
1315 */
1316 retval = set_mm_exe_file(bprm->mm, bprm->file);
1317 if (retval)
1318 goto out;
1319
1320 /* If the binary is not readable then enforce mm->dumpable=0 */
1321 would_dump(bprm, bprm->file);
1322 if (bprm->have_execfd)
1323 would_dump(bprm, bprm->executable);
1324
1325 /*
1326 * Release all of the old mmap stuff
1327 */
1328 acct_arg_size(bprm, 0);
1329 retval = exec_mmap(bprm->mm);
1330 if (retval)
1331 goto out;
1332
1333 bprm->mm = NULL;
1334
1335 retval = exec_task_namespaces();
1336 if (retval)
1337 goto out_unlock;
1338
1339#ifdef CONFIG_POSIX_TIMERS
1340 spin_lock_irq(&me->sighand->siglock);
1341 posix_cpu_timers_exit(me);
1342 spin_unlock_irq(&me->sighand->siglock);
1343 exit_itimers(me);
1344 flush_itimer_signals();
1345#endif
1346
1347 /*
1348 * Make the signal table private.
1349 */
1350 retval = unshare_sighand(me);
1351 if (retval)
1352 goto out_unlock;
1353
1354 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1355 PF_NOFREEZE | PF_NO_SETAFFINITY);
1356 flush_thread();
1357 me->personality &= ~bprm->per_clear;
1358
1359 clear_syscall_work_syscall_user_dispatch(me);
1360
1361 /*
1362 * We have to apply CLOEXEC before we change whether the process is
1363 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1364 * trying to access the should-be-closed file descriptors of a process
1365 * undergoing exec(2).
1366 */
1367 do_close_on_exec(me->files);
1368
1369 if (bprm->secureexec) {
1370 /* Make sure parent cannot signal privileged process. */
1371 me->pdeath_signal = 0;
1372
1373 /*
1374 * For secureexec, reset the stack limit to sane default to
1375 * avoid bad behavior from the prior rlimits. This has to
1376 * happen before arch_pick_mmap_layout(), which examines
1377 * RLIMIT_STACK, but after the point of no return to avoid
1378 * needing to clean up the change on failure.
1379 */
1380 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1381 bprm->rlim_stack.rlim_cur = _STK_LIM;
1382 }
1383
1384 me->sas_ss_sp = me->sas_ss_size = 0;
1385
1386 /*
1387 * Figure out dumpability. Note that this checking only of current
1388 * is wrong, but userspace depends on it. This should be testing
1389 * bprm->secureexec instead.
1390 */
1391 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1392 !(uid_eq(current_euid(), current_uid()) &&
1393 gid_eq(current_egid(), current_gid())))
1394 set_dumpable(current->mm, suid_dumpable);
1395 else
1396 set_dumpable(current->mm, SUID_DUMP_USER);
1397
1398 perf_event_exec();
1399 __set_task_comm(me, kbasename(bprm->filename), true);
1400
1401 /* An exec changes our domain. We are no longer part of the thread
1402 group */
1403 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1404 flush_signal_handlers(me, 0);
1405
1406 retval = set_cred_ucounts(bprm->cred);
1407 if (retval < 0)
1408 goto out_unlock;
1409
1410 /*
1411 * install the new credentials for this executable
1412 */
1413 security_bprm_committing_creds(bprm);
1414
1415 commit_creds(bprm->cred);
1416 bprm->cred = NULL;
1417
1418 /*
1419 * Disable monitoring for regular users
1420 * when executing setuid binaries. Must
1421 * wait until new credentials are committed
1422 * by commit_creds() above
1423 */
1424 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1425 perf_event_exit_task(me);
1426 /*
1427 * cred_guard_mutex must be held at least to this point to prevent
1428 * ptrace_attach() from altering our determination of the task's
1429 * credentials; any time after this it may be unlocked.
1430 */
1431 security_bprm_committed_creds(bprm);
1432
1433 /* Pass the opened binary to the interpreter. */
1434 if (bprm->have_execfd) {
1435 retval = get_unused_fd_flags(0);
1436 if (retval < 0)
1437 goto out_unlock;
1438 fd_install(retval, bprm->executable);
1439 bprm->executable = NULL;
1440 bprm->execfd = retval;
1441 }
1442 return 0;
1443
1444out_unlock:
1445 up_write(&me->signal->exec_update_lock);
1446 if (!bprm->cred)
1447 mutex_unlock(&me->signal->cred_guard_mutex);
1448
1449out:
1450 return retval;
1451}
1452EXPORT_SYMBOL(begin_new_exec);
1453
1454void would_dump(struct linux_binprm *bprm, struct file *file)
1455{
1456 struct inode *inode = file_inode(file);
1457 struct mnt_idmap *idmap = file_mnt_idmap(file);
1458 if (inode_permission(idmap, inode, MAY_READ) < 0) {
1459 struct user_namespace *old, *user_ns;
1460 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1461
1462 /* Ensure mm->user_ns contains the executable */
1463 user_ns = old = bprm->mm->user_ns;
1464 while ((user_ns != &init_user_ns) &&
1465 !privileged_wrt_inode_uidgid(user_ns, idmap, inode))
1466 user_ns = user_ns->parent;
1467
1468 if (old != user_ns) {
1469 bprm->mm->user_ns = get_user_ns(user_ns);
1470 put_user_ns(old);
1471 }
1472 }
1473}
1474EXPORT_SYMBOL(would_dump);
1475
1476void setup_new_exec(struct linux_binprm * bprm)
1477{
1478 /* Setup things that can depend upon the personality */
1479 struct task_struct *me = current;
1480
1481 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1482
1483 arch_setup_new_exec();
1484
1485 /* Set the new mm task size. We have to do that late because it may
1486 * depend on TIF_32BIT which is only updated in flush_thread() on
1487 * some architectures like powerpc
1488 */
1489 me->mm->task_size = TASK_SIZE;
1490 up_write(&me->signal->exec_update_lock);
1491 mutex_unlock(&me->signal->cred_guard_mutex);
1492}
1493EXPORT_SYMBOL(setup_new_exec);
1494
1495/* Runs immediately before start_thread() takes over. */
1496void finalize_exec(struct linux_binprm *bprm)
1497{
1498 /* Store any stack rlimit changes before starting thread. */
1499 task_lock(current->group_leader);
1500 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1501 task_unlock(current->group_leader);
1502}
1503EXPORT_SYMBOL(finalize_exec);
1504
1505/*
1506 * Prepare credentials and lock ->cred_guard_mutex.
1507 * setup_new_exec() commits the new creds and drops the lock.
1508 * Or, if exec fails before, free_bprm() should release ->cred
1509 * and unlock.
1510 */
1511static int prepare_bprm_creds(struct linux_binprm *bprm)
1512{
1513 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1514 return -ERESTARTNOINTR;
1515
1516 bprm->cred = prepare_exec_creds();
1517 if (likely(bprm->cred))
1518 return 0;
1519
1520 mutex_unlock(&current->signal->cred_guard_mutex);
1521 return -ENOMEM;
1522}
1523
1524/* Matches do_open_execat() */
1525static void do_close_execat(struct file *file)
1526{
1527 if (!file)
1528 return;
1529 allow_write_access(file);
1530 fput(file);
1531}
1532
1533static void free_bprm(struct linux_binprm *bprm)
1534{
1535 if (bprm->mm) {
1536 acct_arg_size(bprm, 0);
1537 mmput(bprm->mm);
1538 }
1539 free_arg_pages(bprm);
1540 if (bprm->cred) {
1541 mutex_unlock(&current->signal->cred_guard_mutex);
1542 abort_creds(bprm->cred);
1543 }
1544 do_close_execat(bprm->file);
1545 if (bprm->executable)
1546 fput(bprm->executable);
1547 /* If a binfmt changed the interp, free it. */
1548 if (bprm->interp != bprm->filename)
1549 kfree(bprm->interp);
1550 kfree(bprm->fdpath);
1551 kfree(bprm);
1552}
1553
1554static struct linux_binprm *alloc_bprm(int fd, struct filename *filename, int flags)
1555{
1556 struct linux_binprm *bprm;
1557 struct file *file;
1558 int retval = -ENOMEM;
1559
1560 file = do_open_execat(fd, filename, flags);
1561 if (IS_ERR(file))
1562 return ERR_CAST(file);
1563
1564 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1565 if (!bprm) {
1566 do_close_execat(file);
1567 return ERR_PTR(-ENOMEM);
1568 }
1569
1570 bprm->file = file;
1571
1572 if (fd == AT_FDCWD || filename->name[0] == '/') {
1573 bprm->filename = filename->name;
1574 } else {
1575 if (filename->name[0] == '\0')
1576 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1577 else
1578 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1579 fd, filename->name);
1580 if (!bprm->fdpath)
1581 goto out_free;
1582
1583 /*
1584 * Record that a name derived from an O_CLOEXEC fd will be
1585 * inaccessible after exec. This allows the code in exec to
1586 * choose to fail when the executable is not mmaped into the
1587 * interpreter and an open file descriptor is not passed to
1588 * the interpreter. This makes for a better user experience
1589 * than having the interpreter start and then immediately fail
1590 * when it finds the executable is inaccessible.
1591 */
1592 if (get_close_on_exec(fd))
1593 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1594
1595 bprm->filename = bprm->fdpath;
1596 }
1597 bprm->interp = bprm->filename;
1598
1599 retval = bprm_mm_init(bprm);
1600 if (!retval)
1601 return bprm;
1602
1603out_free:
1604 free_bprm(bprm);
1605 return ERR_PTR(retval);
1606}
1607
1608int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1609{
1610 /* If a binfmt changed the interp, free it first. */
1611 if (bprm->interp != bprm->filename)
1612 kfree(bprm->interp);
1613 bprm->interp = kstrdup(interp, GFP_KERNEL);
1614 if (!bprm->interp)
1615 return -ENOMEM;
1616 return 0;
1617}
1618EXPORT_SYMBOL(bprm_change_interp);
1619
1620/*
1621 * determine how safe it is to execute the proposed program
1622 * - the caller must hold ->cred_guard_mutex to protect against
1623 * PTRACE_ATTACH or seccomp thread-sync
1624 */
1625static void check_unsafe_exec(struct linux_binprm *bprm)
1626{
1627 struct task_struct *p = current, *t;
1628 unsigned n_fs;
1629
1630 if (p->ptrace)
1631 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1632
1633 /*
1634 * This isn't strictly necessary, but it makes it harder for LSMs to
1635 * mess up.
1636 */
1637 if (task_no_new_privs(current))
1638 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1639
1640 /*
1641 * If another task is sharing our fs, we cannot safely
1642 * suid exec because the differently privileged task
1643 * will be able to manipulate the current directory, etc.
1644 * It would be nice to force an unshare instead...
1645 */
1646 n_fs = 1;
1647 spin_lock(&p->fs->lock);
1648 rcu_read_lock();
1649 for_other_threads(p, t) {
1650 if (t->fs == p->fs)
1651 n_fs++;
1652 }
1653 rcu_read_unlock();
1654
1655 /* "users" and "in_exec" locked for copy_fs() */
1656 if (p->fs->users > n_fs)
1657 bprm->unsafe |= LSM_UNSAFE_SHARE;
1658 else
1659 p->fs->in_exec = 1;
1660 spin_unlock(&p->fs->lock);
1661}
1662
1663static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1664{
1665 /* Handle suid and sgid on files */
1666 struct mnt_idmap *idmap;
1667 struct inode *inode = file_inode(file);
1668 unsigned int mode;
1669 vfsuid_t vfsuid;
1670 vfsgid_t vfsgid;
1671
1672 if (!mnt_may_suid(file->f_path.mnt))
1673 return;
1674
1675 if (task_no_new_privs(current))
1676 return;
1677
1678 mode = READ_ONCE(inode->i_mode);
1679 if (!(mode & (S_ISUID|S_ISGID)))
1680 return;
1681
1682 idmap = file_mnt_idmap(file);
1683
1684 /* Be careful if suid/sgid is set */
1685 inode_lock(inode);
1686
1687 /* reload atomically mode/uid/gid now that lock held */
1688 mode = inode->i_mode;
1689 vfsuid = i_uid_into_vfsuid(idmap, inode);
1690 vfsgid = i_gid_into_vfsgid(idmap, inode);
1691 inode_unlock(inode);
1692
1693 /* We ignore suid/sgid if there are no mappings for them in the ns */
1694 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1695 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1696 return;
1697
1698 if (mode & S_ISUID) {
1699 bprm->per_clear |= PER_CLEAR_ON_SETID;
1700 bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1701 }
1702
1703 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1704 bprm->per_clear |= PER_CLEAR_ON_SETID;
1705 bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1706 }
1707}
1708
1709/*
1710 * Compute brpm->cred based upon the final binary.
1711 */
1712static int bprm_creds_from_file(struct linux_binprm *bprm)
1713{
1714 /* Compute creds based on which file? */
1715 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1716
1717 bprm_fill_uid(bprm, file);
1718 return security_bprm_creds_from_file(bprm, file);
1719}
1720
1721/*
1722 * Fill the binprm structure from the inode.
1723 * Read the first BINPRM_BUF_SIZE bytes
1724 *
1725 * This may be called multiple times for binary chains (scripts for example).
1726 */
1727static int prepare_binprm(struct linux_binprm *bprm)
1728{
1729 loff_t pos = 0;
1730
1731 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1732 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1733}
1734
1735/*
1736 * Arguments are '\0' separated strings found at the location bprm->p
1737 * points to; chop off the first by relocating brpm->p to right after
1738 * the first '\0' encountered.
1739 */
1740int remove_arg_zero(struct linux_binprm *bprm)
1741{
1742 unsigned long offset;
1743 char *kaddr;
1744 struct page *page;
1745
1746 if (!bprm->argc)
1747 return 0;
1748
1749 do {
1750 offset = bprm->p & ~PAGE_MASK;
1751 page = get_arg_page(bprm, bprm->p, 0);
1752 if (!page)
1753 return -EFAULT;
1754 kaddr = kmap_local_page(page);
1755
1756 for (; offset < PAGE_SIZE && kaddr[offset];
1757 offset++, bprm->p++)
1758 ;
1759
1760 kunmap_local(kaddr);
1761 put_arg_page(page);
1762 } while (offset == PAGE_SIZE);
1763
1764 bprm->p++;
1765 bprm->argc--;
1766
1767 return 0;
1768}
1769EXPORT_SYMBOL(remove_arg_zero);
1770
1771#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1772/*
1773 * cycle the list of binary formats handler, until one recognizes the image
1774 */
1775static int search_binary_handler(struct linux_binprm *bprm)
1776{
1777 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1778 struct linux_binfmt *fmt;
1779 int retval;
1780
1781 retval = prepare_binprm(bprm);
1782 if (retval < 0)
1783 return retval;
1784
1785 retval = security_bprm_check(bprm);
1786 if (retval)
1787 return retval;
1788
1789 retval = -ENOENT;
1790 retry:
1791 read_lock(&binfmt_lock);
1792 list_for_each_entry(fmt, &formats, lh) {
1793 if (!try_module_get(fmt->module))
1794 continue;
1795 read_unlock(&binfmt_lock);
1796
1797 retval = fmt->load_binary(bprm);
1798
1799 read_lock(&binfmt_lock);
1800 put_binfmt(fmt);
1801 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1802 read_unlock(&binfmt_lock);
1803 return retval;
1804 }
1805 }
1806 read_unlock(&binfmt_lock);
1807
1808 if (need_retry) {
1809 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1810 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1811 return retval;
1812 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1813 return retval;
1814 need_retry = false;
1815 goto retry;
1816 }
1817
1818 return retval;
1819}
1820
1821/* binfmt handlers will call back into begin_new_exec() on success. */
1822static int exec_binprm(struct linux_binprm *bprm)
1823{
1824 pid_t old_pid, old_vpid;
1825 int ret, depth;
1826
1827 /* Need to fetch pid before load_binary changes it */
1828 old_pid = current->pid;
1829 rcu_read_lock();
1830 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1831 rcu_read_unlock();
1832
1833 /* This allows 4 levels of binfmt rewrites before failing hard. */
1834 for (depth = 0;; depth++) {
1835 struct file *exec;
1836 if (depth > 5)
1837 return -ELOOP;
1838
1839 ret = search_binary_handler(bprm);
1840 if (ret < 0)
1841 return ret;
1842 if (!bprm->interpreter)
1843 break;
1844
1845 exec = bprm->file;
1846 bprm->file = bprm->interpreter;
1847 bprm->interpreter = NULL;
1848
1849 allow_write_access(exec);
1850 if (unlikely(bprm->have_execfd)) {
1851 if (bprm->executable) {
1852 fput(exec);
1853 return -ENOEXEC;
1854 }
1855 bprm->executable = exec;
1856 } else
1857 fput(exec);
1858 }
1859
1860 audit_bprm(bprm);
1861 trace_sched_process_exec(current, old_pid, bprm);
1862 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1863 proc_exec_connector(current);
1864 return 0;
1865}
1866
1867static int bprm_execve(struct linux_binprm *bprm)
1868{
1869 int retval;
1870
1871 retval = prepare_bprm_creds(bprm);
1872 if (retval)
1873 return retval;
1874
1875 /*
1876 * Check for unsafe execution states before exec_binprm(), which
1877 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1878 * where setuid-ness is evaluated.
1879 */
1880 check_unsafe_exec(bprm);
1881 current->in_execve = 1;
1882 sched_mm_cid_before_execve(current);
1883
1884 sched_exec();
1885
1886 /* Set the unchanging part of bprm->cred */
1887 retval = security_bprm_creds_for_exec(bprm);
1888 if (retval)
1889 goto out;
1890
1891 retval = exec_binprm(bprm);
1892 if (retval < 0)
1893 goto out;
1894
1895 sched_mm_cid_after_execve(current);
1896 /* execve succeeded */
1897 current->fs->in_exec = 0;
1898 current->in_execve = 0;
1899 rseq_execve(current);
1900 user_events_execve(current);
1901 acct_update_integrals(current);
1902 task_numa_free(current, false);
1903 return retval;
1904
1905out:
1906 /*
1907 * If past the point of no return ensure the code never
1908 * returns to the userspace process. Use an existing fatal
1909 * signal if present otherwise terminate the process with
1910 * SIGSEGV.
1911 */
1912 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1913 force_fatal_sig(SIGSEGV);
1914
1915 sched_mm_cid_after_execve(current);
1916 current->fs->in_exec = 0;
1917 current->in_execve = 0;
1918
1919 return retval;
1920}
1921
1922static int do_execveat_common(int fd, struct filename *filename,
1923 struct user_arg_ptr argv,
1924 struct user_arg_ptr envp,
1925 int flags)
1926{
1927 struct linux_binprm *bprm;
1928 int retval;
1929
1930 if (IS_ERR(filename))
1931 return PTR_ERR(filename);
1932
1933 /*
1934 * We move the actual failure in case of RLIMIT_NPROC excess from
1935 * set*uid() to execve() because too many poorly written programs
1936 * don't check setuid() return code. Here we additionally recheck
1937 * whether NPROC limit is still exceeded.
1938 */
1939 if ((current->flags & PF_NPROC_EXCEEDED) &&
1940 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1941 retval = -EAGAIN;
1942 goto out_ret;
1943 }
1944
1945 /* We're below the limit (still or again), so we don't want to make
1946 * further execve() calls fail. */
1947 current->flags &= ~PF_NPROC_EXCEEDED;
1948
1949 bprm = alloc_bprm(fd, filename, flags);
1950 if (IS_ERR(bprm)) {
1951 retval = PTR_ERR(bprm);
1952 goto out_ret;
1953 }
1954
1955 retval = count(argv, MAX_ARG_STRINGS);
1956 if (retval == 0)
1957 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1958 current->comm, bprm->filename);
1959 if (retval < 0)
1960 goto out_free;
1961 bprm->argc = retval;
1962
1963 retval = count(envp, MAX_ARG_STRINGS);
1964 if (retval < 0)
1965 goto out_free;
1966 bprm->envc = retval;
1967
1968 retval = bprm_stack_limits(bprm);
1969 if (retval < 0)
1970 goto out_free;
1971
1972 retval = copy_string_kernel(bprm->filename, bprm);
1973 if (retval < 0)
1974 goto out_free;
1975 bprm->exec = bprm->p;
1976
1977 retval = copy_strings(bprm->envc, envp, bprm);
1978 if (retval < 0)
1979 goto out_free;
1980
1981 retval = copy_strings(bprm->argc, argv, bprm);
1982 if (retval < 0)
1983 goto out_free;
1984
1985 /*
1986 * When argv is empty, add an empty string ("") as argv[0] to
1987 * ensure confused userspace programs that start processing
1988 * from argv[1] won't end up walking envp. See also
1989 * bprm_stack_limits().
1990 */
1991 if (bprm->argc == 0) {
1992 retval = copy_string_kernel("", bprm);
1993 if (retval < 0)
1994 goto out_free;
1995 bprm->argc = 1;
1996 }
1997
1998 retval = bprm_execve(bprm);
1999out_free:
2000 free_bprm(bprm);
2001
2002out_ret:
2003 putname(filename);
2004 return retval;
2005}
2006
2007int kernel_execve(const char *kernel_filename,
2008 const char *const *argv, const char *const *envp)
2009{
2010 struct filename *filename;
2011 struct linux_binprm *bprm;
2012 int fd = AT_FDCWD;
2013 int retval;
2014
2015 /* It is non-sense for kernel threads to call execve */
2016 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
2017 return -EINVAL;
2018
2019 filename = getname_kernel(kernel_filename);
2020 if (IS_ERR(filename))
2021 return PTR_ERR(filename);
2022
2023 bprm = alloc_bprm(fd, filename, 0);
2024 if (IS_ERR(bprm)) {
2025 retval = PTR_ERR(bprm);
2026 goto out_ret;
2027 }
2028
2029 retval = count_strings_kernel(argv);
2030 if (WARN_ON_ONCE(retval == 0))
2031 retval = -EINVAL;
2032 if (retval < 0)
2033 goto out_free;
2034 bprm->argc = retval;
2035
2036 retval = count_strings_kernel(envp);
2037 if (retval < 0)
2038 goto out_free;
2039 bprm->envc = retval;
2040
2041 retval = bprm_stack_limits(bprm);
2042 if (retval < 0)
2043 goto out_free;
2044
2045 retval = copy_string_kernel(bprm->filename, bprm);
2046 if (retval < 0)
2047 goto out_free;
2048 bprm->exec = bprm->p;
2049
2050 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2051 if (retval < 0)
2052 goto out_free;
2053
2054 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2055 if (retval < 0)
2056 goto out_free;
2057
2058 retval = bprm_execve(bprm);
2059out_free:
2060 free_bprm(bprm);
2061out_ret:
2062 putname(filename);
2063 return retval;
2064}
2065
2066static int do_execve(struct filename *filename,
2067 const char __user *const __user *__argv,
2068 const char __user *const __user *__envp)
2069{
2070 struct user_arg_ptr argv = { .ptr.native = __argv };
2071 struct user_arg_ptr envp = { .ptr.native = __envp };
2072 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2073}
2074
2075static int do_execveat(int fd, struct filename *filename,
2076 const char __user *const __user *__argv,
2077 const char __user *const __user *__envp,
2078 int flags)
2079{
2080 struct user_arg_ptr argv = { .ptr.native = __argv };
2081 struct user_arg_ptr envp = { .ptr.native = __envp };
2082
2083 return do_execveat_common(fd, filename, argv, envp, flags);
2084}
2085
2086#ifdef CONFIG_COMPAT
2087static int compat_do_execve(struct filename *filename,
2088 const compat_uptr_t __user *__argv,
2089 const compat_uptr_t __user *__envp)
2090{
2091 struct user_arg_ptr argv = {
2092 .is_compat = true,
2093 .ptr.compat = __argv,
2094 };
2095 struct user_arg_ptr envp = {
2096 .is_compat = true,
2097 .ptr.compat = __envp,
2098 };
2099 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2100}
2101
2102static int compat_do_execveat(int fd, struct filename *filename,
2103 const compat_uptr_t __user *__argv,
2104 const compat_uptr_t __user *__envp,
2105 int flags)
2106{
2107 struct user_arg_ptr argv = {
2108 .is_compat = true,
2109 .ptr.compat = __argv,
2110 };
2111 struct user_arg_ptr envp = {
2112 .is_compat = true,
2113 .ptr.compat = __envp,
2114 };
2115 return do_execveat_common(fd, filename, argv, envp, flags);
2116}
2117#endif
2118
2119void set_binfmt(struct linux_binfmt *new)
2120{
2121 struct mm_struct *mm = current->mm;
2122
2123 if (mm->binfmt)
2124 module_put(mm->binfmt->module);
2125
2126 mm->binfmt = new;
2127 if (new)
2128 __module_get(new->module);
2129}
2130EXPORT_SYMBOL(set_binfmt);
2131
2132/*
2133 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2134 */
2135void set_dumpable(struct mm_struct *mm, int value)
2136{
2137 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2138 return;
2139
2140 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2141}
2142
2143SYSCALL_DEFINE3(execve,
2144 const char __user *, filename,
2145 const char __user *const __user *, argv,
2146 const char __user *const __user *, envp)
2147{
2148 return do_execve(getname(filename), argv, envp);
2149}
2150
2151SYSCALL_DEFINE5(execveat,
2152 int, fd, const char __user *, filename,
2153 const char __user *const __user *, argv,
2154 const char __user *const __user *, envp,
2155 int, flags)
2156{
2157 return do_execveat(fd,
2158 getname_uflags(filename, flags),
2159 argv, envp, flags);
2160}
2161
2162#ifdef CONFIG_COMPAT
2163COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2164 const compat_uptr_t __user *, argv,
2165 const compat_uptr_t __user *, envp)
2166{
2167 return compat_do_execve(getname(filename), argv, envp);
2168}
2169
2170COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2171 const char __user *, filename,
2172 const compat_uptr_t __user *, argv,
2173 const compat_uptr_t __user *, envp,
2174 int, flags)
2175{
2176 return compat_do_execveat(fd,
2177 getname_uflags(filename, flags),
2178 argv, envp, flags);
2179}
2180#endif
2181
2182#ifdef CONFIG_SYSCTL
2183
2184static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2185 void *buffer, size_t *lenp, loff_t *ppos)
2186{
2187 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2188
2189 if (!error)
2190 validate_coredump_safety();
2191 return error;
2192}
2193
2194static struct ctl_table fs_exec_sysctls[] = {
2195 {
2196 .procname = "suid_dumpable",
2197 .data = &suid_dumpable,
2198 .maxlen = sizeof(int),
2199 .mode = 0644,
2200 .proc_handler = proc_dointvec_minmax_coredump,
2201 .extra1 = SYSCTL_ZERO,
2202 .extra2 = SYSCTL_TWO,
2203 },
2204};
2205
2206static int __init init_fs_exec_sysctls(void)
2207{
2208 register_sysctl_init("fs", fs_exec_sysctls);
2209 return 0;
2210}
2211
2212fs_initcall(init_fs_exec_sysctls);
2213#endif /* CONFIG_SYSCTL */