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