projects/linux·blob v6.13 · refs

2168 lines · 51.4 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/exec.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * #!-checking implemented by tytso.
10 */
11/*
12 * Demand-loading implemented 01.12.91 - no need to read anything but
13 * the header into memory. The inode of the executable is put into
14 * "current->executable", and page faults do the actual loading. Clean.
15 *
16 * Once more I can proudly say that linux stood up to being changed: it
17 * was less than 2 hours work to get demand-loading completely implemented.
18 *
19 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
20 * current->executable is only used by the procfs. This allows a dispatch
21 * table to check for several different types of binary formats. We keep
22 * trying until we recognize the file or we run out of supported binary
23 * formats.
24 */
25
26#include <linux/kernel_read_file.h>
27#include <linux/slab.h>
28#include <linux/file.h>
29#include <linux/fdtable.h>
30#include <linux/mm.h>
31#include <linux/stat.h>
32#include <linux/fcntl.h>
33#include <linux/swap.h>
34#include <linux/string.h>
35#include <linux/init.h>
36#include <linux/sched/mm.h>
37#include <linux/sched/coredump.h>
38#include <linux/sched/signal.h>
39#include <linux/sched/numa_balancing.h>
40#include <linux/sched/task.h>
41#include <linux/pagemap.h>
42#include <linux/perf_event.h>
43#include <linux/highmem.h>
44#include <linux/spinlock.h>
45#include <linux/key.h>
46#include <linux/personality.h>
47#include <linux/binfmts.h>
48#include <linux/utsname.h>
49#include <linux/pid_namespace.h>
50#include <linux/module.h>
51#include <linux/namei.h>
52#include <linux/mount.h>
53#include <linux/security.h>
54#include <linux/syscalls.h>
55#include <linux/tsacct_kern.h>
56#include <linux/cn_proc.h>
57#include <linux/audit.h>
58#include <linux/kmod.h>
59#include <linux/fsnotify.h>
60#include <linux/fs_struct.h>
61#include <linux/oom.h>
62#include <linux/compat.h>
63#include <linux/vmalloc.h>
64#include <linux/io_uring.h>
65#include <linux/syscall_user_dispatch.h>
66#include <linux/coredump.h>
67#include <linux/time_namespace.h>
68#include <linux/user_events.h>
69#include <linux/rseq.h>
70#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 * Check do_open_execat() for an explanation.
149 */
150 error = -EACCES;
151 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode)) ||
152 path_noexec(&file->f_path))
153 goto exit;
154
155 error = -ENOEXEC;
156
157 read_lock(&binfmt_lock);
158 list_for_each_entry(fmt, &formats, lh) {
159 if (!fmt->load_shlib)
160 continue;
161 if (!try_module_get(fmt->module))
162 continue;
163 read_unlock(&binfmt_lock);
164 error = fmt->load_shlib(file);
165 read_lock(&binfmt_lock);
166 put_binfmt(fmt);
167 if (error != -ENOEXEC)
168 break;
169 }
170 read_unlock(&binfmt_lock);
171exit:
172 fput(file);
173out:
174 return error;
175}
176#endif /* #ifdef CONFIG_USELIB */
177
178#ifdef CONFIG_MMU
179/*
180 * The nascent bprm->mm is not visible until exec_mmap() but it can
181 * use a lot of memory, account these pages in current->mm temporary
182 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
183 * change the counter back via acct_arg_size(0).
184 */
185static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
186{
187 struct mm_struct *mm = current->mm;
188 long diff = (long)(pages - bprm->vma_pages);
189
190 if (!mm || !diff)
191 return;
192
193 bprm->vma_pages = pages;
194 add_mm_counter(mm, MM_ANONPAGES, diff);
195}
196
197static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
198 int write)
199{
200 struct page *page;
201 struct vm_area_struct *vma = bprm->vma;
202 struct mm_struct *mm = bprm->mm;
203 int ret;
204
205 /*
206 * Avoid relying on expanding the stack down in GUP (which
207 * does not work for STACK_GROWSUP anyway), and just do it
208 * by hand ahead of time.
209 */
210 if (write && pos < vma->vm_start) {
211 mmap_write_lock(mm);
212 ret = expand_downwards(vma, pos);
213 if (unlikely(ret < 0)) {
214 mmap_write_unlock(mm);
215 return NULL;
216 }
217 mmap_write_downgrade(mm);
218 } else
219 mmap_read_lock(mm);
220
221 /*
222 * We are doing an exec(). 'current' is the process
223 * doing the exec and 'mm' is the new process's mm.
224 */
225 ret = get_user_pages_remote(mm, pos, 1,
226 write ? FOLL_WRITE : 0,
227 &page, NULL);
228 mmap_read_unlock(mm);
229 if (ret <= 0)
230 return NULL;
231
232 if (write)
233 acct_arg_size(bprm, vma_pages(vma));
234
235 return page;
236}
237
238static void put_arg_page(struct page *page)
239{
240 put_page(page);
241}
242
243static void free_arg_pages(struct linux_binprm *bprm)
244{
245}
246
247static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
248 struct page *page)
249{
250 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
251}
252
253static int __bprm_mm_init(struct linux_binprm *bprm)
254{
255 int err;
256 struct vm_area_struct *vma = NULL;
257 struct mm_struct *mm = bprm->mm;
258
259 bprm->vma = vma = vm_area_alloc(mm);
260 if (!vma)
261 return -ENOMEM;
262 vma_set_anonymous(vma);
263
264 if (mmap_write_lock_killable(mm)) {
265 err = -EINTR;
266 goto err_free;
267 }
268
269 /*
270 * Need to be called with mmap write lock
271 * held, to avoid race with ksmd.
272 */
273 err = ksm_execve(mm);
274 if (err)
275 goto err_ksm;
276
277 /*
278 * Place the stack at the largest stack address the architecture
279 * supports. Later, we'll move this to an appropriate place. We don't
280 * use STACK_TOP because that can depend on attributes which aren't
281 * configured yet.
282 */
283 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
284 vma->vm_end = STACK_TOP_MAX;
285 vma->vm_start = vma->vm_end - PAGE_SIZE;
286 vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);
287 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
288
289 err = insert_vm_struct(mm, vma);
290 if (err)
291 goto err;
292
293 mm->stack_vm = mm->total_vm = 1;
294 mmap_write_unlock(mm);
295 bprm->p = vma->vm_end - sizeof(void *);
296 return 0;
297err:
298 ksm_exit(mm);
299err_ksm:
300 mmap_write_unlock(mm);
301err_free:
302 bprm->vma = NULL;
303 vm_area_free(vma);
304 return err;
305}
306
307static bool valid_arg_len(struct linux_binprm *bprm, long len)
308{
309 return len <= MAX_ARG_STRLEN;
310}
311
312#else
313
314static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
315{
316}
317
318static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
319 int write)
320{
321 struct page *page;
322
323 page = bprm->page[pos / PAGE_SIZE];
324 if (!page && write) {
325 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
326 if (!page)
327 return NULL;
328 bprm->page[pos / PAGE_SIZE] = page;
329 }
330
331 return page;
332}
333
334static void put_arg_page(struct page *page)
335{
336}
337
338static void free_arg_page(struct linux_binprm *bprm, int i)
339{
340 if (bprm->page[i]) {
341 __free_page(bprm->page[i]);
342 bprm->page[i] = NULL;
343 }
344}
345
346static void free_arg_pages(struct linux_binprm *bprm)
347{
348 int i;
349
350 for (i = 0; i < MAX_ARG_PAGES; i++)
351 free_arg_page(bprm, i);
352}
353
354static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
355 struct page *page)
356{
357}
358
359static int __bprm_mm_init(struct linux_binprm *bprm)
360{
361 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
362 return 0;
363}
364
365static bool valid_arg_len(struct linux_binprm *bprm, long len)
366{
367 return len <= bprm->p;
368}
369
370#endif /* CONFIG_MMU */
371
372/*
373 * Create a new mm_struct and populate it with a temporary stack
374 * vm_area_struct. We don't have enough context at this point to set the stack
375 * flags, permissions, and offset, so we use temporary values. We'll update
376 * them later in setup_arg_pages().
377 */
378static int bprm_mm_init(struct linux_binprm *bprm)
379{
380 int err;
381 struct mm_struct *mm = NULL;
382
383 bprm->mm = mm = mm_alloc();
384 err = -ENOMEM;
385 if (!mm)
386 goto err;
387
388 /* Save current stack limit for all calculations made during exec. */
389 task_lock(current->group_leader);
390 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
391 task_unlock(current->group_leader);
392
393 err = __bprm_mm_init(bprm);
394 if (err)
395 goto err;
396
397 return 0;
398
399err:
400 if (mm) {
401 bprm->mm = NULL;
402 mmdrop(mm);
403 }
404
405 return err;
406}
407
408struct user_arg_ptr {
409#ifdef CONFIG_COMPAT
410 bool is_compat;
411#endif
412 union {
413 const char __user *const __user *native;
414#ifdef CONFIG_COMPAT
415 const compat_uptr_t __user *compat;
416#endif
417 } ptr;
418};
419
420static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
421{
422 const char __user *native;
423
424#ifdef CONFIG_COMPAT
425 if (unlikely(argv.is_compat)) {
426 compat_uptr_t compat;
427
428 if (get_user(compat, argv.ptr.compat + nr))
429 return ERR_PTR(-EFAULT);
430
431 return compat_ptr(compat);
432 }
433#endif
434
435 if (get_user(native, argv.ptr.native + nr))
436 return ERR_PTR(-EFAULT);
437
438 return native;
439}
440
441/*
442 * count() counts the number of strings in array ARGV.
443 */
444static int count(struct user_arg_ptr argv, int max)
445{
446 int i = 0;
447
448 if (argv.ptr.native != NULL) {
449 for (;;) {
450 const char __user *p = get_user_arg_ptr(argv, i);
451
452 if (!p)
453 break;
454
455 if (IS_ERR(p))
456 return -EFAULT;
457
458 if (i >= max)
459 return -E2BIG;
460 ++i;
461
462 if (fatal_signal_pending(current))
463 return -ERESTARTNOHAND;
464 cond_resched();
465 }
466 }
467 return i;
468}
469
470static int count_strings_kernel(const char *const *argv)
471{
472 int i;
473
474 if (!argv)
475 return 0;
476
477 for (i = 0; argv[i]; ++i) {
478 if (i >= MAX_ARG_STRINGS)
479 return -E2BIG;
480 if (fatal_signal_pending(current))
481 return -ERESTARTNOHAND;
482 cond_resched();
483 }
484 return i;
485}
486
487static inline int bprm_set_stack_limit(struct linux_binprm *bprm,
488 unsigned long limit)
489{
490#ifdef CONFIG_MMU
491 /* Avoid a pathological bprm->p. */
492 if (bprm->p < limit)
493 return -E2BIG;
494 bprm->argmin = bprm->p - limit;
495#endif
496 return 0;
497}
498static inline bool bprm_hit_stack_limit(struct linux_binprm *bprm)
499{
500#ifdef CONFIG_MMU
501 return bprm->p < bprm->argmin;
502#else
503 return false;
504#endif
505}
506
507/*
508 * Calculate bprm->argmin from:
509 * - _STK_LIM
510 * - ARG_MAX
511 * - bprm->rlim_stack.rlim_cur
512 * - bprm->argc
513 * - bprm->envc
514 * - bprm->p
515 */
516static int bprm_stack_limits(struct linux_binprm *bprm)
517{
518 unsigned long limit, ptr_size;
519
520 /*
521 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
522 * (whichever is smaller) for the argv+env strings.
523 * This ensures that:
524 * - the remaining binfmt code will not run out of stack space,
525 * - the program will have a reasonable amount of stack left
526 * to work from.
527 */
528 limit = _STK_LIM / 4 * 3;
529 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
530 /*
531 * We've historically supported up to 32 pages (ARG_MAX)
532 * of argument strings even with small stacks
533 */
534 limit = max_t(unsigned long, limit, ARG_MAX);
535 /* Reject totally pathological counts. */
536 if (bprm->argc < 0 || bprm->envc < 0)
537 return -E2BIG;
538 /*
539 * We must account for the size of all the argv and envp pointers to
540 * the argv and envp strings, since they will also take up space in
541 * the stack. They aren't stored until much later when we can't
542 * signal to the parent that the child has run out of stack space.
543 * Instead, calculate it here so it's possible to fail gracefully.
544 *
545 * In the case of argc = 0, make sure there is space for adding a
546 * empty string (which will bump argc to 1), to ensure confused
547 * userspace programs don't start processing from argv[1], thinking
548 * argc can never be 0, to keep them from walking envp by accident.
549 * See do_execveat_common().
550 */
551 if (check_add_overflow(max(bprm->argc, 1), bprm->envc, &ptr_size) ||
552 check_mul_overflow(ptr_size, sizeof(void *), &ptr_size))
553 return -E2BIG;
554 if (limit <= ptr_size)
555 return -E2BIG;
556 limit -= ptr_size;
557
558 return bprm_set_stack_limit(bprm, limit);
559}
560
561/*
562 * 'copy_strings()' copies argument/environment strings from the old
563 * processes's memory to the new process's stack. The call to get_user_pages()
564 * ensures the destination page is created and not swapped out.
565 */
566static int copy_strings(int argc, struct user_arg_ptr argv,
567 struct linux_binprm *bprm)
568{
569 struct page *kmapped_page = NULL;
570 char *kaddr = NULL;
571 unsigned long kpos = 0;
572 int ret;
573
574 while (argc-- > 0) {
575 const char __user *str;
576 int len;
577 unsigned long pos;
578
579 ret = -EFAULT;
580 str = get_user_arg_ptr(argv, argc);
581 if (IS_ERR(str))
582 goto out;
583
584 len = strnlen_user(str, MAX_ARG_STRLEN);
585 if (!len)
586 goto out;
587
588 ret = -E2BIG;
589 if (!valid_arg_len(bprm, len))
590 goto out;
591
592 /* We're going to work our way backwards. */
593 pos = bprm->p;
594 str += len;
595 bprm->p -= len;
596 if (bprm_hit_stack_limit(bprm))
597 goto out;
598
599 while (len > 0) {
600 int offset, bytes_to_copy;
601
602 if (fatal_signal_pending(current)) {
603 ret = -ERESTARTNOHAND;
604 goto out;
605 }
606 cond_resched();
607
608 offset = pos % PAGE_SIZE;
609 if (offset == 0)
610 offset = PAGE_SIZE;
611
612 bytes_to_copy = offset;
613 if (bytes_to_copy > len)
614 bytes_to_copy = len;
615
616 offset -= bytes_to_copy;
617 pos -= bytes_to_copy;
618 str -= bytes_to_copy;
619 len -= bytes_to_copy;
620
621 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
622 struct page *page;
623
624 page = get_arg_page(bprm, pos, 1);
625 if (!page) {
626 ret = -E2BIG;
627 goto out;
628 }
629
630 if (kmapped_page) {
631 flush_dcache_page(kmapped_page);
632 kunmap_local(kaddr);
633 put_arg_page(kmapped_page);
634 }
635 kmapped_page = page;
636 kaddr = kmap_local_page(kmapped_page);
637 kpos = pos & PAGE_MASK;
638 flush_arg_page(bprm, kpos, kmapped_page);
639 }
640 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
641 ret = -EFAULT;
642 goto out;
643 }
644 }
645 }
646 ret = 0;
647out:
648 if (kmapped_page) {
649 flush_dcache_page(kmapped_page);
650 kunmap_local(kaddr);
651 put_arg_page(kmapped_page);
652 }
653 return ret;
654}
655
656/*
657 * Copy and argument/environment string from the kernel to the processes stack.
658 */
659int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
660{
661 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
662 unsigned long pos = bprm->p;
663
664 if (len == 0)
665 return -EFAULT;
666 if (!valid_arg_len(bprm, len))
667 return -E2BIG;
668
669 /* We're going to work our way backwards. */
670 arg += len;
671 bprm->p -= len;
672 if (bprm_hit_stack_limit(bprm))
673 return -E2BIG;
674
675 while (len > 0) {
676 unsigned int bytes_to_copy = min_t(unsigned int, len,
677 min_not_zero(offset_in_page(pos), PAGE_SIZE));
678 struct page *page;
679
680 pos -= bytes_to_copy;
681 arg -= bytes_to_copy;
682 len -= bytes_to_copy;
683
684 page = get_arg_page(bprm, pos, 1);
685 if (!page)
686 return -E2BIG;
687 flush_arg_page(bprm, pos & PAGE_MASK, page);
688 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
689 put_arg_page(page);
690 }
691
692 return 0;
693}
694EXPORT_SYMBOL(copy_string_kernel);
695
696static int copy_strings_kernel(int argc, const char *const *argv,
697 struct linux_binprm *bprm)
698{
699 while (argc-- > 0) {
700 int ret = copy_string_kernel(argv[argc], bprm);
701 if (ret < 0)
702 return ret;
703 if (fatal_signal_pending(current))
704 return -ERESTARTNOHAND;
705 cond_resched();
706 }
707 return 0;
708}
709
710#ifdef CONFIG_MMU
711
712/*
713 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
714 * the stack is optionally relocated, and some extra space is added.
715 */
716int setup_arg_pages(struct linux_binprm *bprm,
717 unsigned long stack_top,
718 int executable_stack)
719{
720 unsigned long ret;
721 unsigned long stack_shift;
722 struct mm_struct *mm = current->mm;
723 struct vm_area_struct *vma = bprm->vma;
724 struct vm_area_struct *prev = NULL;
725 unsigned long vm_flags;
726 unsigned long stack_base;
727 unsigned long stack_size;
728 unsigned long stack_expand;
729 unsigned long rlim_stack;
730 struct mmu_gather tlb;
731 struct vma_iterator vmi;
732
733#ifdef CONFIG_STACK_GROWSUP
734 /* Limit stack size */
735 stack_base = bprm->rlim_stack.rlim_max;
736
737 stack_base = calc_max_stack_size(stack_base);
738
739 /* Add space for stack randomization. */
740 if (current->flags & PF_RANDOMIZE)
741 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
742
743 /* Make sure we didn't let the argument array grow too large. */
744 if (vma->vm_end - vma->vm_start > stack_base)
745 return -ENOMEM;
746
747 stack_base = PAGE_ALIGN(stack_top - stack_base);
748
749 stack_shift = vma->vm_start - stack_base;
750 mm->arg_start = bprm->p - stack_shift;
751 bprm->p = vma->vm_end - stack_shift;
752#else
753 stack_top = arch_align_stack(stack_top);
754 stack_top = PAGE_ALIGN(stack_top);
755
756 if (unlikely(stack_top < mmap_min_addr) ||
757 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
758 return -ENOMEM;
759
760 stack_shift = vma->vm_end - stack_top;
761
762 bprm->p -= stack_shift;
763 mm->arg_start = bprm->p;
764#endif
765
766 if (bprm->loader)
767 bprm->loader -= stack_shift;
768 bprm->exec -= stack_shift;
769
770 if (mmap_write_lock_killable(mm))
771 return -EINTR;
772
773 vm_flags = VM_STACK_FLAGS;
774
775 /*
776 * Adjust stack execute permissions; explicitly enable for
777 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
778 * (arch default) otherwise.
779 */
780 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
781 vm_flags |= VM_EXEC;
782 else if (executable_stack == EXSTACK_DISABLE_X)
783 vm_flags &= ~VM_EXEC;
784 vm_flags |= mm->def_flags;
785 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
786
787 vma_iter_init(&vmi, mm, vma->vm_start);
788
789 tlb_gather_mmu(&tlb, mm);
790 ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
791 vm_flags);
792 tlb_finish_mmu(&tlb);
793
794 if (ret)
795 goto out_unlock;
796 BUG_ON(prev != vma);
797
798 if (unlikely(vm_flags & VM_EXEC)) {
799 pr_warn_once("process '%pD4' started with executable stack\n",
800 bprm->file);
801 }
802
803 /* Move stack pages down in memory. */
804 if (stack_shift) {
805 /*
806 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
807 * the binfmt code determines where the new stack should reside, we shift it to
808 * its final location.
809 */
810 ret = relocate_vma_down(vma, stack_shift);
811 if (ret)
812 goto out_unlock;
813 }
814
815 /* mprotect_fixup is overkill to remove the temporary stack flags */
816 vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
817
818 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
819 stack_size = vma->vm_end - vma->vm_start;
820 /*
821 * Align this down to a page boundary as expand_stack
822 * will align it up.
823 */
824 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
825
826 stack_expand = min(rlim_stack, stack_size + stack_expand);
827
828#ifdef CONFIG_STACK_GROWSUP
829 stack_base = vma->vm_start + stack_expand;
830#else
831 stack_base = vma->vm_end - stack_expand;
832#endif
833 current->mm->start_stack = bprm->p;
834 ret = expand_stack_locked(vma, stack_base);
835 if (ret)
836 ret = -EFAULT;
837
838out_unlock:
839 mmap_write_unlock(mm);
840 return ret;
841}
842EXPORT_SYMBOL(setup_arg_pages);
843
844#else
845
846/*
847 * Transfer the program arguments and environment from the holding pages
848 * onto the stack. The provided stack pointer is adjusted accordingly.
849 */
850int transfer_args_to_stack(struct linux_binprm *bprm,
851 unsigned long *sp_location)
852{
853 unsigned long index, stop, sp;
854 int ret = 0;
855
856 stop = bprm->p >> PAGE_SHIFT;
857 sp = *sp_location;
858
859 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
860 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
861 char *src = kmap_local_page(bprm->page[index]) + offset;
862 sp -= PAGE_SIZE - offset;
863 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
864 ret = -EFAULT;
865 kunmap_local(src);
866 if (ret)
867 goto out;
868 }
869
870 bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
871 *sp_location = sp;
872
873out:
874 return ret;
875}
876EXPORT_SYMBOL(transfer_args_to_stack);
877
878#endif /* CONFIG_MMU */
879
880/*
881 * On success, caller must call do_close_execat() on the returned
882 * struct file to close it.
883 */
884static struct file *do_open_execat(int fd, struct filename *name, int flags)
885{
886 int err;
887 struct file *file __free(fput) = NULL;
888 struct open_flags open_exec_flags = {
889 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
890 .acc_mode = MAY_EXEC,
891 .intent = LOOKUP_OPEN,
892 .lookup_flags = LOOKUP_FOLLOW,
893 };
894
895 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
896 return ERR_PTR(-EINVAL);
897 if (flags & AT_SYMLINK_NOFOLLOW)
898 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
899 if (flags & AT_EMPTY_PATH)
900 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
901
902 file = do_filp_open(fd, name, &open_exec_flags);
903 if (IS_ERR(file))
904 return file;
905
906 /*
907 * In the past the regular type check was here. It moved to may_open() in
908 * 633fb6ac3980 ("exec: move S_ISREG() check earlier"). Since then it is
909 * an invariant that all non-regular files error out before we get here.
910 */
911 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode)) ||
912 path_noexec(&file->f_path))
913 return ERR_PTR(-EACCES);
914
915 err = deny_write_access(file);
916 if (err)
917 return ERR_PTR(err);
918
919 return no_free_ptr(file);
920}
921
922/**
923 * open_exec - Open a path name for execution
924 *
925 * @name: path name to open with the intent of executing it.
926 *
927 * Returns ERR_PTR on failure or allocated struct file on success.
928 *
929 * As this is a wrapper for the internal do_open_execat(), callers
930 * must call allow_write_access() before fput() on release. Also see
931 * do_close_execat().
932 */
933struct file *open_exec(const char *name)
934{
935 struct filename *filename = getname_kernel(name);
936 struct file *f = ERR_CAST(filename);
937
938 if (!IS_ERR(filename)) {
939 f = do_open_execat(AT_FDCWD, filename, 0);
940 putname(filename);
941 }
942 return f;
943}
944EXPORT_SYMBOL(open_exec);
945
946#if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
947ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
948{
949 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
950 if (res > 0)
951 flush_icache_user_range(addr, addr + len);
952 return res;
953}
954EXPORT_SYMBOL(read_code);
955#endif
956
957/*
958 * Maps the mm_struct mm into the current task struct.
959 * On success, this function returns with exec_update_lock
960 * held for writing.
961 */
962static int exec_mmap(struct mm_struct *mm)
963{
964 struct task_struct *tsk;
965 struct mm_struct *old_mm, *active_mm;
966 int ret;
967
968 /* Notify parent that we're no longer interested in the old VM */
969 tsk = current;
970 old_mm = current->mm;
971 exec_mm_release(tsk, old_mm);
972
973 ret = down_write_killable(&tsk->signal->exec_update_lock);
974 if (ret)
975 return ret;
976
977 if (old_mm) {
978 /*
979 * If there is a pending fatal signal perhaps a signal
980 * whose default action is to create a coredump get
981 * out and die instead of going through with the exec.
982 */
983 ret = mmap_read_lock_killable(old_mm);
984 if (ret) {
985 up_write(&tsk->signal->exec_update_lock);
986 return ret;
987 }
988 }
989
990 task_lock(tsk);
991 membarrier_exec_mmap(mm);
992
993 local_irq_disable();
994 active_mm = tsk->active_mm;
995 tsk->active_mm = mm;
996 tsk->mm = mm;
997 mm_init_cid(mm, tsk);
998 /*
999 * This prevents preemption while active_mm is being loaded and
1000 * it and mm are being updated, which could cause problems for
1001 * lazy tlb mm refcounting when these are updated by context
1002 * switches. Not all architectures can handle irqs off over
1003 * activate_mm yet.
1004 */
1005 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1006 local_irq_enable();
1007 activate_mm(active_mm, mm);
1008 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1009 local_irq_enable();
1010 lru_gen_add_mm(mm);
1011 task_unlock(tsk);
1012 lru_gen_use_mm(mm);
1013 if (old_mm) {
1014 mmap_read_unlock(old_mm);
1015 BUG_ON(active_mm != old_mm);
1016 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1017 mm_update_next_owner(old_mm);
1018 mmput(old_mm);
1019 return 0;
1020 }
1021 mmdrop_lazy_tlb(active_mm);
1022 return 0;
1023}
1024
1025static int de_thread(struct task_struct *tsk)
1026{
1027 struct signal_struct *sig = tsk->signal;
1028 struct sighand_struct *oldsighand = tsk->sighand;
1029 spinlock_t *lock = &oldsighand->siglock;
1030
1031 if (thread_group_empty(tsk))
1032 goto no_thread_group;
1033
1034 /*
1035 * Kill all other threads in the thread group.
1036 */
1037 spin_lock_irq(lock);
1038 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1039 /*
1040 * Another group action in progress, just
1041 * return so that the signal is processed.
1042 */
1043 spin_unlock_irq(lock);
1044 return -EAGAIN;
1045 }
1046
1047 sig->group_exec_task = tsk;
1048 sig->notify_count = zap_other_threads(tsk);
1049 if (!thread_group_leader(tsk))
1050 sig->notify_count--;
1051
1052 while (sig->notify_count) {
1053 __set_current_state(TASK_KILLABLE);
1054 spin_unlock_irq(lock);
1055 schedule();
1056 if (__fatal_signal_pending(tsk))
1057 goto killed;
1058 spin_lock_irq(lock);
1059 }
1060 spin_unlock_irq(lock);
1061
1062 /*
1063 * At this point all other threads have exited, all we have to
1064 * do is to wait for the thread group leader to become inactive,
1065 * and to assume its PID:
1066 */
1067 if (!thread_group_leader(tsk)) {
1068 struct task_struct *leader = tsk->group_leader;
1069
1070 for (;;) {
1071 cgroup_threadgroup_change_begin(tsk);
1072 write_lock_irq(&tasklist_lock);
1073 /*
1074 * Do this under tasklist_lock to ensure that
1075 * exit_notify() can't miss ->group_exec_task
1076 */
1077 sig->notify_count = -1;
1078 if (likely(leader->exit_state))
1079 break;
1080 __set_current_state(TASK_KILLABLE);
1081 write_unlock_irq(&tasklist_lock);
1082 cgroup_threadgroup_change_end(tsk);
1083 schedule();
1084 if (__fatal_signal_pending(tsk))
1085 goto killed;
1086 }
1087
1088 /*
1089 * The only record we have of the real-time age of a
1090 * process, regardless of execs it's done, is start_time.
1091 * All the past CPU time is accumulated in signal_struct
1092 * from sister threads now dead. But in this non-leader
1093 * exec, nothing survives from the original leader thread,
1094 * whose birth marks the true age of this process now.
1095 * When we take on its identity by switching to its PID, we
1096 * also take its birthdate (always earlier than our own).
1097 */
1098 tsk->start_time = leader->start_time;
1099 tsk->start_boottime = leader->start_boottime;
1100
1101 BUG_ON(!same_thread_group(leader, tsk));
1102 /*
1103 * An exec() starts a new thread group with the
1104 * TGID of the previous thread group. Rehash the
1105 * two threads with a switched PID, and release
1106 * the former thread group leader:
1107 */
1108
1109 /* Become a process group leader with the old leader's pid.
1110 * The old leader becomes a thread of the this thread group.
1111 */
1112 exchange_tids(tsk, leader);
1113 transfer_pid(leader, tsk, PIDTYPE_TGID);
1114 transfer_pid(leader, tsk, PIDTYPE_PGID);
1115 transfer_pid(leader, tsk, PIDTYPE_SID);
1116
1117 list_replace_rcu(&leader->tasks, &tsk->tasks);
1118 list_replace_init(&leader->sibling, &tsk->sibling);
1119
1120 tsk->group_leader = tsk;
1121 leader->group_leader = tsk;
1122
1123 tsk->exit_signal = SIGCHLD;
1124 leader->exit_signal = -1;
1125
1126 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1127 leader->exit_state = EXIT_DEAD;
1128 /*
1129 * We are going to release_task()->ptrace_unlink() silently,
1130 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1131 * the tracer won't block again waiting for this thread.
1132 */
1133 if (unlikely(leader->ptrace))
1134 __wake_up_parent(leader, leader->parent);
1135 write_unlock_irq(&tasklist_lock);
1136 cgroup_threadgroup_change_end(tsk);
1137
1138 release_task(leader);
1139 }
1140
1141 sig->group_exec_task = NULL;
1142 sig->notify_count = 0;
1143
1144no_thread_group:
1145 /* we have changed execution domain */
1146 tsk->exit_signal = SIGCHLD;
1147
1148 BUG_ON(!thread_group_leader(tsk));
1149 return 0;
1150
1151killed:
1152 /* protects against exit_notify() and __exit_signal() */
1153 read_lock(&tasklist_lock);
1154 sig->group_exec_task = NULL;
1155 sig->notify_count = 0;
1156 read_unlock(&tasklist_lock);
1157 return -EAGAIN;
1158}
1159
1160
1161/*
1162 * This function makes sure the current process has its own signal table,
1163 * so that flush_signal_handlers can later reset the handlers without
1164 * disturbing other processes. (Other processes might share the signal
1165 * table via the CLONE_SIGHAND option to clone().)
1166 */
1167static int unshare_sighand(struct task_struct *me)
1168{
1169 struct sighand_struct *oldsighand = me->sighand;
1170
1171 if (refcount_read(&oldsighand->count) != 1) {
1172 struct sighand_struct *newsighand;
1173 /*
1174 * This ->sighand is shared with the CLONE_SIGHAND
1175 * but not CLONE_THREAD task, switch to the new one.
1176 */
1177 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1178 if (!newsighand)
1179 return -ENOMEM;
1180
1181 refcount_set(&newsighand->count, 1);
1182
1183 write_lock_irq(&tasklist_lock);
1184 spin_lock(&oldsighand->siglock);
1185 memcpy(newsighand->action, oldsighand->action,
1186 sizeof(newsighand->action));
1187 rcu_assign_pointer(me->sighand, newsighand);
1188 spin_unlock(&oldsighand->siglock);
1189 write_unlock_irq(&tasklist_lock);
1190
1191 __cleanup_sighand(oldsighand);
1192 }
1193 return 0;
1194}
1195
1196/*
1197 * These functions flushes out all traces of the currently running executable
1198 * so that a new one can be started
1199 */
1200
1201void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1202{
1203 task_lock(tsk);
1204 trace_task_rename(tsk, buf);
1205 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1206 task_unlock(tsk);
1207 perf_event_comm(tsk, exec);
1208}
1209
1210/*
1211 * Calling this is the point of no return. None of the failures will be
1212 * seen by userspace since either the process is already taking a fatal
1213 * signal (via de_thread() or coredump), or will have SEGV raised
1214 * (after exec_mmap()) by search_binary_handler (see below).
1215 */
1216int begin_new_exec(struct linux_binprm * bprm)
1217{
1218 struct task_struct *me = current;
1219 int retval;
1220
1221 /* Once we are committed compute the creds */
1222 retval = bprm_creds_from_file(bprm);
1223 if (retval)
1224 return retval;
1225
1226 /*
1227 * This tracepoint marks the point before flushing the old exec where
1228 * the current task is still unchanged, but errors are fatal (point of
1229 * no return). The later "sched_process_exec" tracepoint is called after
1230 * the current task has successfully switched to the new exec.
1231 */
1232 trace_sched_prepare_exec(current, bprm);
1233
1234 /*
1235 * Ensure all future errors are fatal.
1236 */
1237 bprm->point_of_no_return = true;
1238
1239 /*
1240 * Make this the only thread in the thread group.
1241 */
1242 retval = de_thread(me);
1243 if (retval)
1244 goto out;
1245
1246 /*
1247 * Cancel any io_uring activity across execve
1248 */
1249 io_uring_task_cancel();
1250
1251 /* Ensure the files table is not shared. */
1252 retval = unshare_files();
1253 if (retval)
1254 goto out;
1255
1256 /*
1257 * Must be called _before_ exec_mmap() as bprm->mm is
1258 * not visible until then. Doing it here also ensures
1259 * we don't race against replace_mm_exe_file().
1260 */
1261 retval = set_mm_exe_file(bprm->mm, bprm->file);
1262 if (retval)
1263 goto out;
1264
1265 /* If the binary is not readable then enforce mm->dumpable=0 */
1266 would_dump(bprm, bprm->file);
1267 if (bprm->have_execfd)
1268 would_dump(bprm, bprm->executable);
1269
1270 /*
1271 * Release all of the old mmap stuff
1272 */
1273 acct_arg_size(bprm, 0);
1274 retval = exec_mmap(bprm->mm);
1275 if (retval)
1276 goto out;
1277
1278 bprm->mm = NULL;
1279
1280 retval = exec_task_namespaces();
1281 if (retval)
1282 goto out_unlock;
1283
1284#ifdef CONFIG_POSIX_TIMERS
1285 spin_lock_irq(&me->sighand->siglock);
1286 posix_cpu_timers_exit(me);
1287 spin_unlock_irq(&me->sighand->siglock);
1288 exit_itimers(me);
1289 flush_itimer_signals();
1290#endif
1291
1292 /*
1293 * Make the signal table private.
1294 */
1295 retval = unshare_sighand(me);
1296 if (retval)
1297 goto out_unlock;
1298
1299 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1300 PF_NOFREEZE | PF_NO_SETAFFINITY);
1301 flush_thread();
1302 me->personality &= ~bprm->per_clear;
1303
1304 clear_syscall_work_syscall_user_dispatch(me);
1305
1306 /*
1307 * We have to apply CLOEXEC before we change whether the process is
1308 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1309 * trying to access the should-be-closed file descriptors of a process
1310 * undergoing exec(2).
1311 */
1312 do_close_on_exec(me->files);
1313
1314 if (bprm->secureexec) {
1315 /* Make sure parent cannot signal privileged process. */
1316 me->pdeath_signal = 0;
1317
1318 /*
1319 * For secureexec, reset the stack limit to sane default to
1320 * avoid bad behavior from the prior rlimits. This has to
1321 * happen before arch_pick_mmap_layout(), which examines
1322 * RLIMIT_STACK, but after the point of no return to avoid
1323 * needing to clean up the change on failure.
1324 */
1325 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1326 bprm->rlim_stack.rlim_cur = _STK_LIM;
1327 }
1328
1329 me->sas_ss_sp = me->sas_ss_size = 0;
1330
1331 /*
1332 * Figure out dumpability. Note that this checking only of current
1333 * is wrong, but userspace depends on it. This should be testing
1334 * bprm->secureexec instead.
1335 */
1336 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1337 !(uid_eq(current_euid(), current_uid()) &&
1338 gid_eq(current_egid(), current_gid())))
1339 set_dumpable(current->mm, suid_dumpable);
1340 else
1341 set_dumpable(current->mm, SUID_DUMP_USER);
1342
1343 perf_event_exec();
1344 __set_task_comm(me, kbasename(bprm->filename), true);
1345
1346 /* An exec changes our domain. We are no longer part of the thread
1347 group */
1348 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1349 flush_signal_handlers(me, 0);
1350
1351 retval = set_cred_ucounts(bprm->cred);
1352 if (retval < 0)
1353 goto out_unlock;
1354
1355 /*
1356 * install the new credentials for this executable
1357 */
1358 security_bprm_committing_creds(bprm);
1359
1360 commit_creds(bprm->cred);
1361 bprm->cred = NULL;
1362
1363 /*
1364 * Disable monitoring for regular users
1365 * when executing setuid binaries. Must
1366 * wait until new credentials are committed
1367 * by commit_creds() above
1368 */
1369 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1370 perf_event_exit_task(me);
1371 /*
1372 * cred_guard_mutex must be held at least to this point to prevent
1373 * ptrace_attach() from altering our determination of the task's
1374 * credentials; any time after this it may be unlocked.
1375 */
1376 security_bprm_committed_creds(bprm);
1377
1378 /* Pass the opened binary to the interpreter. */
1379 if (bprm->have_execfd) {
1380 retval = get_unused_fd_flags(0);
1381 if (retval < 0)
1382 goto out_unlock;
1383 fd_install(retval, bprm->executable);
1384 bprm->executable = NULL;
1385 bprm->execfd = retval;
1386 }
1387 return 0;
1388
1389out_unlock:
1390 up_write(&me->signal->exec_update_lock);
1391 if (!bprm->cred)
1392 mutex_unlock(&me->signal->cred_guard_mutex);
1393
1394out:
1395 return retval;
1396}
1397EXPORT_SYMBOL(begin_new_exec);
1398
1399void would_dump(struct linux_binprm *bprm, struct file *file)
1400{
1401 struct inode *inode = file_inode(file);
1402 struct mnt_idmap *idmap = file_mnt_idmap(file);
1403 if (inode_permission(idmap, inode, MAY_READ) < 0) {
1404 struct user_namespace *old, *user_ns;
1405 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1406
1407 /* Ensure mm->user_ns contains the executable */
1408 user_ns = old = bprm->mm->user_ns;
1409 while ((user_ns != &init_user_ns) &&
1410 !privileged_wrt_inode_uidgid(user_ns, idmap, inode))
1411 user_ns = user_ns->parent;
1412
1413 if (old != user_ns) {
1414 bprm->mm->user_ns = get_user_ns(user_ns);
1415 put_user_ns(old);
1416 }
1417 }
1418}
1419EXPORT_SYMBOL(would_dump);
1420
1421void setup_new_exec(struct linux_binprm * bprm)
1422{
1423 /* Setup things that can depend upon the personality */
1424 struct task_struct *me = current;
1425
1426 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1427
1428 arch_setup_new_exec();
1429
1430 /* Set the new mm task size. We have to do that late because it may
1431 * depend on TIF_32BIT which is only updated in flush_thread() on
1432 * some architectures like powerpc
1433 */
1434 me->mm->task_size = TASK_SIZE;
1435 up_write(&me->signal->exec_update_lock);
1436 mutex_unlock(&me->signal->cred_guard_mutex);
1437}
1438EXPORT_SYMBOL(setup_new_exec);
1439
1440/* Runs immediately before start_thread() takes over. */
1441void finalize_exec(struct linux_binprm *bprm)
1442{
1443 /* Store any stack rlimit changes before starting thread. */
1444 task_lock(current->group_leader);
1445 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1446 task_unlock(current->group_leader);
1447}
1448EXPORT_SYMBOL(finalize_exec);
1449
1450/*
1451 * Prepare credentials and lock ->cred_guard_mutex.
1452 * setup_new_exec() commits the new creds and drops the lock.
1453 * Or, if exec fails before, free_bprm() should release ->cred
1454 * and unlock.
1455 */
1456static int prepare_bprm_creds(struct linux_binprm *bprm)
1457{
1458 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1459 return -ERESTARTNOINTR;
1460
1461 bprm->cred = prepare_exec_creds();
1462 if (likely(bprm->cred))
1463 return 0;
1464
1465 mutex_unlock(&current->signal->cred_guard_mutex);
1466 return -ENOMEM;
1467}
1468
1469/* Matches do_open_execat() */
1470static void do_close_execat(struct file *file)
1471{
1472 if (!file)
1473 return;
1474 allow_write_access(file);
1475 fput(file);
1476}
1477
1478static void free_bprm(struct linux_binprm *bprm)
1479{
1480 if (bprm->mm) {
1481 acct_arg_size(bprm, 0);
1482 mmput(bprm->mm);
1483 }
1484 free_arg_pages(bprm);
1485 if (bprm->cred) {
1486 mutex_unlock(&current->signal->cred_guard_mutex);
1487 abort_creds(bprm->cred);
1488 }
1489 do_close_execat(bprm->file);
1490 if (bprm->executable)
1491 fput(bprm->executable);
1492 /* If a binfmt changed the interp, free it. */
1493 if (bprm->interp != bprm->filename)
1494 kfree(bprm->interp);
1495 kfree(bprm->fdpath);
1496 kfree(bprm);
1497}
1498
1499static struct linux_binprm *alloc_bprm(int fd, struct filename *filename, int flags)
1500{
1501 struct linux_binprm *bprm;
1502 struct file *file;
1503 int retval = -ENOMEM;
1504
1505 file = do_open_execat(fd, filename, flags);
1506 if (IS_ERR(file))
1507 return ERR_CAST(file);
1508
1509 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1510 if (!bprm) {
1511 do_close_execat(file);
1512 return ERR_PTR(-ENOMEM);
1513 }
1514
1515 bprm->file = file;
1516
1517 if (fd == AT_FDCWD || filename->name[0] == '/') {
1518 bprm->filename = filename->name;
1519 } else {
1520 if (filename->name[0] == '\0')
1521 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1522 else
1523 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1524 fd, filename->name);
1525 if (!bprm->fdpath)
1526 goto out_free;
1527
1528 /*
1529 * Record that a name derived from an O_CLOEXEC fd will be
1530 * inaccessible after exec. This allows the code in exec to
1531 * choose to fail when the executable is not mmaped into the
1532 * interpreter and an open file descriptor is not passed to
1533 * the interpreter. This makes for a better user experience
1534 * than having the interpreter start and then immediately fail
1535 * when it finds the executable is inaccessible.
1536 */
1537 if (get_close_on_exec(fd))
1538 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1539
1540 bprm->filename = bprm->fdpath;
1541 }
1542 bprm->interp = bprm->filename;
1543
1544 retval = bprm_mm_init(bprm);
1545 if (!retval)
1546 return bprm;
1547
1548out_free:
1549 free_bprm(bprm);
1550 return ERR_PTR(retval);
1551}
1552
1553int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1554{
1555 /* If a binfmt changed the interp, free it first. */
1556 if (bprm->interp != bprm->filename)
1557 kfree(bprm->interp);
1558 bprm->interp = kstrdup(interp, GFP_KERNEL);
1559 if (!bprm->interp)
1560 return -ENOMEM;
1561 return 0;
1562}
1563EXPORT_SYMBOL(bprm_change_interp);
1564
1565/*
1566 * determine how safe it is to execute the proposed program
1567 * - the caller must hold ->cred_guard_mutex to protect against
1568 * PTRACE_ATTACH or seccomp thread-sync
1569 */
1570static void check_unsafe_exec(struct linux_binprm *bprm)
1571{
1572 struct task_struct *p = current, *t;
1573 unsigned n_fs;
1574
1575 if (p->ptrace)
1576 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1577
1578 /*
1579 * This isn't strictly necessary, but it makes it harder for LSMs to
1580 * mess up.
1581 */
1582 if (task_no_new_privs(current))
1583 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1584
1585 /*
1586 * If another task is sharing our fs, we cannot safely
1587 * suid exec because the differently privileged task
1588 * will be able to manipulate the current directory, etc.
1589 * It would be nice to force an unshare instead...
1590 */
1591 n_fs = 1;
1592 spin_lock(&p->fs->lock);
1593 rcu_read_lock();
1594 for_other_threads(p, t) {
1595 if (t->fs == p->fs)
1596 n_fs++;
1597 }
1598 rcu_read_unlock();
1599
1600 /* "users" and "in_exec" locked for copy_fs() */
1601 if (p->fs->users > n_fs)
1602 bprm->unsafe |= LSM_UNSAFE_SHARE;
1603 else
1604 p->fs->in_exec = 1;
1605 spin_unlock(&p->fs->lock);
1606}
1607
1608static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1609{
1610 /* Handle suid and sgid on files */
1611 struct mnt_idmap *idmap;
1612 struct inode *inode = file_inode(file);
1613 unsigned int mode;
1614 vfsuid_t vfsuid;
1615 vfsgid_t vfsgid;
1616 int err;
1617
1618 if (!mnt_may_suid(file->f_path.mnt))
1619 return;
1620
1621 if (task_no_new_privs(current))
1622 return;
1623
1624 mode = READ_ONCE(inode->i_mode);
1625 if (!(mode & (S_ISUID|S_ISGID)))
1626 return;
1627
1628 idmap = file_mnt_idmap(file);
1629
1630 /* Be careful if suid/sgid is set */
1631 inode_lock(inode);
1632
1633 /* Atomically reload and check mode/uid/gid now that lock held. */
1634 mode = inode->i_mode;
1635 vfsuid = i_uid_into_vfsuid(idmap, inode);
1636 vfsgid = i_gid_into_vfsgid(idmap, inode);
1637 err = inode_permission(idmap, inode, MAY_EXEC);
1638 inode_unlock(inode);
1639
1640 /* Did the exec bit vanish out from under us? Give up. */
1641 if (err)
1642 return;
1643
1644 /* We ignore suid/sgid if there are no mappings for them in the ns */
1645 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1646 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1647 return;
1648
1649 if (mode & S_ISUID) {
1650 bprm->per_clear |= PER_CLEAR_ON_SETID;
1651 bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1652 }
1653
1654 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1655 bprm->per_clear |= PER_CLEAR_ON_SETID;
1656 bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1657 }
1658}
1659
1660/*
1661 * Compute brpm->cred based upon the final binary.
1662 */
1663static int bprm_creds_from_file(struct linux_binprm *bprm)
1664{
1665 /* Compute creds based on which file? */
1666 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1667
1668 bprm_fill_uid(bprm, file);
1669 return security_bprm_creds_from_file(bprm, file);
1670}
1671
1672/*
1673 * Fill the binprm structure from the inode.
1674 * Read the first BINPRM_BUF_SIZE bytes
1675 *
1676 * This may be called multiple times for binary chains (scripts for example).
1677 */
1678static int prepare_binprm(struct linux_binprm *bprm)
1679{
1680 loff_t pos = 0;
1681
1682 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1683 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1684}
1685
1686/*
1687 * Arguments are '\0' separated strings found at the location bprm->p
1688 * points to; chop off the first by relocating brpm->p to right after
1689 * the first '\0' encountered.
1690 */
1691int remove_arg_zero(struct linux_binprm *bprm)
1692{
1693 unsigned long offset;
1694 char *kaddr;
1695 struct page *page;
1696
1697 if (!bprm->argc)
1698 return 0;
1699
1700 do {
1701 offset = bprm->p & ~PAGE_MASK;
1702 page = get_arg_page(bprm, bprm->p, 0);
1703 if (!page)
1704 return -EFAULT;
1705 kaddr = kmap_local_page(page);
1706
1707 for (; offset < PAGE_SIZE && kaddr[offset];
1708 offset++, bprm->p++)
1709 ;
1710
1711 kunmap_local(kaddr);
1712 put_arg_page(page);
1713 } while (offset == PAGE_SIZE);
1714
1715 bprm->p++;
1716 bprm->argc--;
1717
1718 return 0;
1719}
1720EXPORT_SYMBOL(remove_arg_zero);
1721
1722#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1723/*
1724 * cycle the list of binary formats handler, until one recognizes the image
1725 */
1726static int search_binary_handler(struct linux_binprm *bprm)
1727{
1728 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1729 struct linux_binfmt *fmt;
1730 int retval;
1731
1732 retval = prepare_binprm(bprm);
1733 if (retval < 0)
1734 return retval;
1735
1736 retval = security_bprm_check(bprm);
1737 if (retval)
1738 return retval;
1739
1740 retval = -ENOENT;
1741 retry:
1742 read_lock(&binfmt_lock);
1743 list_for_each_entry(fmt, &formats, lh) {
1744 if (!try_module_get(fmt->module))
1745 continue;
1746 read_unlock(&binfmt_lock);
1747
1748 retval = fmt->load_binary(bprm);
1749
1750 read_lock(&binfmt_lock);
1751 put_binfmt(fmt);
1752 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1753 read_unlock(&binfmt_lock);
1754 return retval;
1755 }
1756 }
1757 read_unlock(&binfmt_lock);
1758
1759 if (need_retry) {
1760 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1761 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1762 return retval;
1763 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1764 return retval;
1765 need_retry = false;
1766 goto retry;
1767 }
1768
1769 return retval;
1770}
1771
1772/* binfmt handlers will call back into begin_new_exec() on success. */
1773static int exec_binprm(struct linux_binprm *bprm)
1774{
1775 pid_t old_pid, old_vpid;
1776 int ret, depth;
1777
1778 /* Need to fetch pid before load_binary changes it */
1779 old_pid = current->pid;
1780 rcu_read_lock();
1781 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1782 rcu_read_unlock();
1783
1784 /* This allows 4 levels of binfmt rewrites before failing hard. */
1785 for (depth = 0;; depth++) {
1786 struct file *exec;
1787 if (depth > 5)
1788 return -ELOOP;
1789
1790 ret = search_binary_handler(bprm);
1791 if (ret < 0)
1792 return ret;
1793 if (!bprm->interpreter)
1794 break;
1795
1796 exec = bprm->file;
1797 bprm->file = bprm->interpreter;
1798 bprm->interpreter = NULL;
1799
1800 allow_write_access(exec);
1801 if (unlikely(bprm->have_execfd)) {
1802 if (bprm->executable) {
1803 fput(exec);
1804 return -ENOEXEC;
1805 }
1806 bprm->executable = exec;
1807 } else
1808 fput(exec);
1809 }
1810
1811 audit_bprm(bprm);
1812 trace_sched_process_exec(current, old_pid, bprm);
1813 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1814 proc_exec_connector(current);
1815 return 0;
1816}
1817
1818static int bprm_execve(struct linux_binprm *bprm)
1819{
1820 int retval;
1821
1822 retval = prepare_bprm_creds(bprm);
1823 if (retval)
1824 return retval;
1825
1826 /*
1827 * Check for unsafe execution states before exec_binprm(), which
1828 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1829 * where setuid-ness is evaluated.
1830 */
1831 check_unsafe_exec(bprm);
1832 current->in_execve = 1;
1833 sched_mm_cid_before_execve(current);
1834
1835 sched_exec();
1836
1837 /* Set the unchanging part of bprm->cred */
1838 retval = security_bprm_creds_for_exec(bprm);
1839 if (retval)
1840 goto out;
1841
1842 retval = exec_binprm(bprm);
1843 if (retval < 0)
1844 goto out;
1845
1846 sched_mm_cid_after_execve(current);
1847 /* execve succeeded */
1848 current->fs->in_exec = 0;
1849 current->in_execve = 0;
1850 rseq_execve(current);
1851 user_events_execve(current);
1852 acct_update_integrals(current);
1853 task_numa_free(current, false);
1854 return retval;
1855
1856out:
1857 /*
1858 * If past the point of no return ensure the code never
1859 * returns to the userspace process. Use an existing fatal
1860 * signal if present otherwise terminate the process with
1861 * SIGSEGV.
1862 */
1863 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1864 force_fatal_sig(SIGSEGV);
1865
1866 sched_mm_cid_after_execve(current);
1867 current->fs->in_exec = 0;
1868 current->in_execve = 0;
1869
1870 return retval;
1871}
1872
1873static int do_execveat_common(int fd, struct filename *filename,
1874 struct user_arg_ptr argv,
1875 struct user_arg_ptr envp,
1876 int flags)
1877{
1878 struct linux_binprm *bprm;
1879 int retval;
1880
1881 if (IS_ERR(filename))
1882 return PTR_ERR(filename);
1883
1884 /*
1885 * We move the actual failure in case of RLIMIT_NPROC excess from
1886 * set*uid() to execve() because too many poorly written programs
1887 * don't check setuid() return code. Here we additionally recheck
1888 * whether NPROC limit is still exceeded.
1889 */
1890 if ((current->flags & PF_NPROC_EXCEEDED) &&
1891 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1892 retval = -EAGAIN;
1893 goto out_ret;
1894 }
1895
1896 /* We're below the limit (still or again), so we don't want to make
1897 * further execve() calls fail. */
1898 current->flags &= ~PF_NPROC_EXCEEDED;
1899
1900 bprm = alloc_bprm(fd, filename, flags);
1901 if (IS_ERR(bprm)) {
1902 retval = PTR_ERR(bprm);
1903 goto out_ret;
1904 }
1905
1906 retval = count(argv, MAX_ARG_STRINGS);
1907 if (retval == 0)
1908 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1909 current->comm, bprm->filename);
1910 if (retval < 0)
1911 goto out_free;
1912 bprm->argc = retval;
1913
1914 retval = count(envp, MAX_ARG_STRINGS);
1915 if (retval < 0)
1916 goto out_free;
1917 bprm->envc = retval;
1918
1919 retval = bprm_stack_limits(bprm);
1920 if (retval < 0)
1921 goto out_free;
1922
1923 retval = copy_string_kernel(bprm->filename, bprm);
1924 if (retval < 0)
1925 goto out_free;
1926 bprm->exec = bprm->p;
1927
1928 retval = copy_strings(bprm->envc, envp, bprm);
1929 if (retval < 0)
1930 goto out_free;
1931
1932 retval = copy_strings(bprm->argc, argv, bprm);
1933 if (retval < 0)
1934 goto out_free;
1935
1936 /*
1937 * When argv is empty, add an empty string ("") as argv[0] to
1938 * ensure confused userspace programs that start processing
1939 * from argv[1] won't end up walking envp. See also
1940 * bprm_stack_limits().
1941 */
1942 if (bprm->argc == 0) {
1943 retval = copy_string_kernel("", bprm);
1944 if (retval < 0)
1945 goto out_free;
1946 bprm->argc = 1;
1947 }
1948
1949 retval = bprm_execve(bprm);
1950out_free:
1951 free_bprm(bprm);
1952
1953out_ret:
1954 putname(filename);
1955 return retval;
1956}
1957
1958int kernel_execve(const char *kernel_filename,
1959 const char *const *argv, const char *const *envp)
1960{
1961 struct filename *filename;
1962 struct linux_binprm *bprm;
1963 int fd = AT_FDCWD;
1964 int retval;
1965
1966 /* It is non-sense for kernel threads to call execve */
1967 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1968 return -EINVAL;
1969
1970 filename = getname_kernel(kernel_filename);
1971 if (IS_ERR(filename))
1972 return PTR_ERR(filename);
1973
1974 bprm = alloc_bprm(fd, filename, 0);
1975 if (IS_ERR(bprm)) {
1976 retval = PTR_ERR(bprm);
1977 goto out_ret;
1978 }
1979
1980 retval = count_strings_kernel(argv);
1981 if (WARN_ON_ONCE(retval == 0))
1982 retval = -EINVAL;
1983 if (retval < 0)
1984 goto out_free;
1985 bprm->argc = retval;
1986
1987 retval = count_strings_kernel(envp);
1988 if (retval < 0)
1989 goto out_free;
1990 bprm->envc = retval;
1991
1992 retval = bprm_stack_limits(bprm);
1993 if (retval < 0)
1994 goto out_free;
1995
1996 retval = copy_string_kernel(bprm->filename, bprm);
1997 if (retval < 0)
1998 goto out_free;
1999 bprm->exec = bprm->p;
2000
2001 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2002 if (retval < 0)
2003 goto out_free;
2004
2005 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2006 if (retval < 0)
2007 goto out_free;
2008
2009 retval = bprm_execve(bprm);
2010out_free:
2011 free_bprm(bprm);
2012out_ret:
2013 putname(filename);
2014 return retval;
2015}
2016
2017static int do_execve(struct filename *filename,
2018 const char __user *const __user *__argv,
2019 const char __user *const __user *__envp)
2020{
2021 struct user_arg_ptr argv = { .ptr.native = __argv };
2022 struct user_arg_ptr envp = { .ptr.native = __envp };
2023 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2024}
2025
2026static int do_execveat(int fd, struct filename *filename,
2027 const char __user *const __user *__argv,
2028 const char __user *const __user *__envp,
2029 int flags)
2030{
2031 struct user_arg_ptr argv = { .ptr.native = __argv };
2032 struct user_arg_ptr envp = { .ptr.native = __envp };
2033
2034 return do_execveat_common(fd, filename, argv, envp, flags);
2035}
2036
2037#ifdef CONFIG_COMPAT
2038static int compat_do_execve(struct filename *filename,
2039 const compat_uptr_t __user *__argv,
2040 const compat_uptr_t __user *__envp)
2041{
2042 struct user_arg_ptr argv = {
2043 .is_compat = true,
2044 .ptr.compat = __argv,
2045 };
2046 struct user_arg_ptr envp = {
2047 .is_compat = true,
2048 .ptr.compat = __envp,
2049 };
2050 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2051}
2052
2053static int compat_do_execveat(int fd, struct filename *filename,
2054 const compat_uptr_t __user *__argv,
2055 const compat_uptr_t __user *__envp,
2056 int flags)
2057{
2058 struct user_arg_ptr argv = {
2059 .is_compat = true,
2060 .ptr.compat = __argv,
2061 };
2062 struct user_arg_ptr envp = {
2063 .is_compat = true,
2064 .ptr.compat = __envp,
2065 };
2066 return do_execveat_common(fd, filename, argv, envp, flags);
2067}
2068#endif
2069
2070void set_binfmt(struct linux_binfmt *new)
2071{
2072 struct mm_struct *mm = current->mm;
2073
2074 if (mm->binfmt)
2075 module_put(mm->binfmt->module);
2076
2077 mm->binfmt = new;
2078 if (new)
2079 __module_get(new->module);
2080}
2081EXPORT_SYMBOL(set_binfmt);
2082
2083/*
2084 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2085 */
2086void set_dumpable(struct mm_struct *mm, int value)
2087{
2088 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2089 return;
2090
2091 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2092}
2093
2094SYSCALL_DEFINE3(execve,
2095 const char __user *, filename,
2096 const char __user *const __user *, argv,
2097 const char __user *const __user *, envp)
2098{
2099 return do_execve(getname(filename), argv, envp);
2100}
2101
2102SYSCALL_DEFINE5(execveat,
2103 int, fd, const char __user *, filename,
2104 const char __user *const __user *, argv,
2105 const char __user *const __user *, envp,
2106 int, flags)
2107{
2108 return do_execveat(fd,
2109 getname_uflags(filename, flags),
2110 argv, envp, flags);
2111}
2112
2113#ifdef CONFIG_COMPAT
2114COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2115 const compat_uptr_t __user *, argv,
2116 const compat_uptr_t __user *, envp)
2117{
2118 return compat_do_execve(getname(filename), argv, envp);
2119}
2120
2121COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2122 const char __user *, filename,
2123 const compat_uptr_t __user *, argv,
2124 const compat_uptr_t __user *, envp,
2125 int, flags)
2126{
2127 return compat_do_execveat(fd,
2128 getname_uflags(filename, flags),
2129 argv, envp, flags);
2130}
2131#endif
2132
2133#ifdef CONFIG_SYSCTL
2134
2135static int proc_dointvec_minmax_coredump(const struct ctl_table *table, int write,
2136 void *buffer, size_t *lenp, loff_t *ppos)
2137{
2138 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2139
2140 if (!error)
2141 validate_coredump_safety();
2142 return error;
2143}
2144
2145static struct ctl_table fs_exec_sysctls[] = {
2146 {
2147 .procname = "suid_dumpable",
2148 .data = &suid_dumpable,
2149 .maxlen = sizeof(int),
2150 .mode = 0644,
2151 .proc_handler = proc_dointvec_minmax_coredump,
2152 .extra1 = SYSCTL_ZERO,
2153 .extra2 = SYSCTL_TWO,
2154 },
2155};
2156
2157static int __init init_fs_exec_sysctls(void)
2158{
2159 register_sysctl_init("fs", fs_exec_sysctls);
2160 return 0;
2161}
2162
2163fs_initcall(init_fs_exec_sysctls);
2164#endif /* CONFIG_SYSCTL */
2165
2166#ifdef CONFIG_EXEC_KUNIT_TEST
2167#include "tests/exec_kunit.c"
2168#endif