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