projects/linux·blob v5.8-rc3 · refs

2109 lines · 49.2 KiB

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