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