xref: /linux/fs/exec.c (revision 6b3f7af57881f6d6250c6dcc4d910fe8e855a607)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/exec.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 /*
9  * #!-checking implemented by tytso.
10  */
11 /*
12  * Demand-loading implemented 01.12.91 - no need to read anything but
13  * the header into memory. The inode of the executable is put into
14  * "current->executable", and page faults do the actual loading. Clean.
15  *
16  * Once more I can proudly say that linux stood up to being changed: it
17  * was less than 2 hours work to get demand-loading completely implemented.
18  *
19  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
20  * current->executable is only used by the procfs.  This allows a dispatch
21  * table to check for several different types  of binary formats.  We keep
22  * trying until we recognize the file or we run out of supported binary
23  * formats.
24  */
25 
26 #include <linux/kernel_read_file.h>
27 #include <linux/slab.h>
28 #include <linux/file.h>
29 #include <linux/fdtable.h>
30 #include <linux/mm.h>
31 #include <linux/stat.h>
32 #include <linux/fcntl.h>
33 #include <linux/swap.h>
34 #include <linux/string.h>
35 #include <linux/init.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/coredump.h>
38 #include <linux/sched/exec_state.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/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 #include <linux/io_uring.h>
66 #include <linux/syscall_user_dispatch.h>
67 #include <linux/coredump.h>
68 #include <linux/time_namespace.h>
69 #include <linux/user_events.h>
70 #include <linux/rseq.h>
71 #include <linux/ksm.h>
72 
73 #include <linux/uaccess.h>
74 #include <asm/mmu_context.h>
75 #include <asm/tlb.h>
76 
77 #include <trace/events/task.h>
78 #include "internal.h"
79 
80 #include <trace/events/sched.h>
81 
82 /* For vma exec functions. */
83 #include "../mm/internal.h"
84 
85 static int bprm_creds_from_file(struct linux_binprm *bprm);
86 
87 int suid_dumpable = 0;
88 
89 static LIST_HEAD(formats);
90 static DEFINE_RWLOCK(binfmt_lock);
91 
92 void __register_binfmt(struct linux_binfmt * fmt, int insert)
93 {
94 	write_lock(&binfmt_lock);
95 	insert ? list_add(&fmt->lh, &formats) :
96 		 list_add_tail(&fmt->lh, &formats);
97 	write_unlock(&binfmt_lock);
98 }
99 
100 EXPORT_SYMBOL(__register_binfmt);
101 
102 void unregister_binfmt(struct linux_binfmt * fmt)
103 {
104 	write_lock(&binfmt_lock);
105 	list_del(&fmt->lh);
106 	write_unlock(&binfmt_lock);
107 }
108 
109 EXPORT_SYMBOL(unregister_binfmt);
110 
111 static inline void put_binfmt(struct linux_binfmt * fmt)
112 {
113 	module_put(fmt->module);
114 }
115 
116 bool path_noexec(const struct path *path)
117 {
118 	/* If it's an anonymous inode make sure that we catch any shenanigans. */
119 	VFS_WARN_ON_ONCE(IS_ANON_FILE(d_inode(path->dentry)) &&
120 			 !(path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC));
121 	return (path->mnt->mnt_flags & MNT_NOEXEC) ||
122 	       (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
123 }
124 
125 #ifdef CONFIG_MMU
126 /*
127  * The nascent bprm->mm is not visible until exec_mmap() but it can
128  * use a lot of memory, account these pages in current->mm temporary
129  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
130  * change the counter back via acct_arg_size(0).
131  */
132 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
133 {
134 	struct mm_struct *mm = current->mm;
135 	long diff = (long)(pages - bprm->vma_pages);
136 
137 	if (!mm || !diff)
138 		return;
139 
140 	bprm->vma_pages = pages;
141 	add_mm_counter(mm, MM_ANONPAGES, diff);
142 }
143 
144 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
145 		int write)
146 {
147 	struct page *page;
148 	struct vm_area_struct *vma = bprm->vma;
149 	struct mm_struct *mm = bprm->mm;
150 	int ret;
151 
152 	/*
153 	 * Avoid relying on expanding the stack down in GUP (which
154 	 * does not work for STACK_GROWSUP anyway), and just do it
155 	 * ahead of time.
156 	 */
157 	if (!mmap_read_lock_maybe_expand(mm, vma, pos, write))
158 		return NULL;
159 
160 	/*
161 	 * We are doing an exec().  'current' is the process
162 	 * doing the exec and 'mm' is the new process's mm.
163 	 */
164 	ret = get_user_pages_remote(mm, pos, 1,
165 			write ? FOLL_WRITE : 0,
166 			&page, NULL);
167 	mmap_read_unlock(mm);
168 	if (ret <= 0)
169 		return NULL;
170 
171 	if (write)
172 		acct_arg_size(bprm, vma_pages(vma));
173 
174 	return page;
175 }
176 
177 static void put_arg_page(struct page *page)
178 {
179 	put_page(page);
180 }
181 
182 static void free_arg_pages(struct linux_binprm *bprm)
183 {
184 }
185 
186 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
187 		struct page *page)
188 {
189 	flush_cache_page(bprm->vma, pos, page_to_pfn(page));
190 }
191 
192 static bool valid_arg_len(struct linux_binprm *bprm, long len)
193 {
194 	return len <= MAX_ARG_STRLEN;
195 }
196 
197 #else
198 
199 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
200 {
201 }
202 
203 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
204 		int write)
205 {
206 	struct page *page;
207 
208 	page = bprm->page[pos / PAGE_SIZE];
209 	if (!page && write) {
210 		page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
211 		if (!page)
212 			return NULL;
213 		bprm->page[pos / PAGE_SIZE] = page;
214 	}
215 
216 	return page;
217 }
218 
219 static void put_arg_page(struct page *page)
220 {
221 }
222 
223 static void free_arg_page(struct linux_binprm *bprm, int i)
224 {
225 	if (bprm->page[i]) {
226 		__free_page(bprm->page[i]);
227 		bprm->page[i] = NULL;
228 	}
229 }
230 
231 static void free_arg_pages(struct linux_binprm *bprm)
232 {
233 	int i;
234 
235 	for (i = 0; i < MAX_ARG_PAGES; i++)
236 		free_arg_page(bprm, i);
237 }
238 
239 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
240 		struct page *page)
241 {
242 }
243 
244 static bool valid_arg_len(struct linux_binprm *bprm, long len)
245 {
246 	return len <= bprm->p;
247 }
248 
249 #endif /* CONFIG_MMU */
250 
251 /*
252  * Create a new mm_struct and populate it with a temporary stack
253  * vm_area_struct.  We don't have enough context at this point to set the stack
254  * flags, permissions, and offset, so we use temporary values.  We'll update
255  * them later in setup_arg_pages().
256  */
257 static int bprm_mm_init(struct linux_binprm *bprm)
258 {
259 	int err;
260 	struct mm_struct *mm = NULL;
261 
262 	bprm->mm = mm = mm_alloc();
263 	err = -ENOMEM;
264 	if (!mm)
265 		goto err;
266 
267 	/* Staged for would_dump() narrowing; consumed by begin_new_exec(). */
268 	bprm->user_ns = get_user_ns(current_user_ns());
269 
270 	/* Save current stack limit for all calculations made during exec. */
271 	task_lock(current->group_leader);
272 	bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
273 	task_unlock(current->group_leader);
274 
275 #ifndef CONFIG_MMU
276 	bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
277 #else
278 	err = create_init_stack_vma(bprm->mm, &bprm->vma, &bprm->p);
279 	if (err)
280 		goto err;
281 #endif
282 
283 	return 0;
284 
285 err:
286 	if (mm) {
287 		bprm->mm = NULL;
288 		mmdrop(mm);
289 	}
290 
291 	return err;
292 }
293 
294 struct user_arg_ptr {
295 #ifdef CONFIG_COMPAT
296 	bool is_compat;
297 #endif
298 	union {
299 		const char __user *const __user *native;
300 #ifdef CONFIG_COMPAT
301 		const compat_uptr_t __user *compat;
302 #endif
303 	} ptr;
304 };
305 
306 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
307 {
308 	const char __user *native;
309 
310 #ifdef CONFIG_COMPAT
311 	if (unlikely(argv.is_compat)) {
312 		compat_uptr_t compat;
313 
314 		if (get_user(compat, argv.ptr.compat + nr))
315 			return ERR_PTR(-EFAULT);
316 
317 		return compat_ptr(compat);
318 	}
319 #endif
320 
321 	if (get_user(native, argv.ptr.native + nr))
322 		return ERR_PTR(-EFAULT);
323 
324 	return native;
325 }
326 
327 /*
328  * count() counts the number of strings in array ARGV.
329  */
330 static int count(struct user_arg_ptr argv, int max)
331 {
332 	int i = 0;
333 
334 	if (argv.ptr.native != NULL) {
335 		for (;;) {
336 			const char __user *p = get_user_arg_ptr(argv, i);
337 
338 			if (!p)
339 				break;
340 
341 			if (IS_ERR(p))
342 				return -EFAULT;
343 
344 			if (i >= max)
345 				return -E2BIG;
346 			++i;
347 
348 			if (fatal_signal_pending(current))
349 				return -ERESTARTNOHAND;
350 			cond_resched();
351 		}
352 	}
353 	return i;
354 }
355 
356 static int count_strings_kernel(const char *const *argv)
357 {
358 	int i;
359 
360 	if (!argv)
361 		return 0;
362 
363 	for (i = 0; argv[i]; ++i) {
364 		if (i >= MAX_ARG_STRINGS)
365 			return -E2BIG;
366 		if (fatal_signal_pending(current))
367 			return -ERESTARTNOHAND;
368 		cond_resched();
369 	}
370 	return i;
371 }
372 
373 static inline int bprm_set_stack_limit(struct linux_binprm *bprm,
374 				       unsigned long limit)
375 {
376 #ifdef CONFIG_MMU
377 	/* Avoid a pathological bprm->p. */
378 	if (bprm->p < limit)
379 		return -E2BIG;
380 	bprm->argmin = bprm->p - limit;
381 #endif
382 	return 0;
383 }
384 static inline bool bprm_hit_stack_limit(struct linux_binprm *bprm)
385 {
386 #ifdef CONFIG_MMU
387 	return bprm->p < bprm->argmin;
388 #else
389 	return false;
390 #endif
391 }
392 
393 /*
394  * Calculate bprm->argmin from:
395  * - _STK_LIM
396  * - ARG_MAX
397  * - bprm->rlim_stack.rlim_cur
398  * - bprm->argc
399  * - bprm->envc
400  * - bprm->p
401  */
402 static int bprm_stack_limits(struct linux_binprm *bprm)
403 {
404 	unsigned long limit, ptr_size;
405 
406 	/*
407 	 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
408 	 * (whichever is smaller) for the argv+env strings.
409 	 * This ensures that:
410 	 *  - the remaining binfmt code will not run out of stack space,
411 	 *  - the program will have a reasonable amount of stack left
412 	 *    to work from.
413 	 */
414 	limit = _STK_LIM / 4 * 3;
415 	limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
416 	/*
417 	 * We've historically supported up to 32 pages (ARG_MAX)
418 	 * of argument strings even with small stacks
419 	 */
420 	limit = max_t(unsigned long, limit, ARG_MAX);
421 	/* Reject totally pathological counts. */
422 	if (bprm->argc < 0 || bprm->envc < 0)
423 		return -E2BIG;
424 	/*
425 	 * We must account for the size of all the argv and envp pointers to
426 	 * the argv and envp strings, since they will also take up space in
427 	 * the stack. They aren't stored until much later when we can't
428 	 * signal to the parent that the child has run out of stack space.
429 	 * Instead, calculate it here so it's possible to fail gracefully.
430 	 *
431 	 * In the case of argc = 0, make sure there is space for adding a
432 	 * empty string (which will bump argc to 1), to ensure confused
433 	 * userspace programs don't start processing from argv[1], thinking
434 	 * argc can never be 0, to keep them from walking envp by accident.
435 	 * See do_execveat_common().
436 	 */
437 	if (check_add_overflow(max(bprm->argc, 1), bprm->envc, &ptr_size) ||
438 	    check_mul_overflow(ptr_size, sizeof(void *), &ptr_size))
439 		return -E2BIG;
440 	if (limit <= ptr_size)
441 		return -E2BIG;
442 	limit -= ptr_size;
443 
444 	return bprm_set_stack_limit(bprm, limit);
445 }
446 
447 /*
448  * 'copy_strings()' copies argument/environment strings from the old
449  * processes's memory to the new process's stack.  The call to get_user_pages()
450  * ensures the destination page is created and not swapped out.
451  */
452 static int copy_strings(int argc, struct user_arg_ptr argv,
453 			struct linux_binprm *bprm)
454 {
455 	struct page *kmapped_page = NULL;
456 	char *kaddr = NULL;
457 	unsigned long kpos = 0;
458 	int ret;
459 
460 	while (argc-- > 0) {
461 		const char __user *str;
462 		int len;
463 		unsigned long pos;
464 
465 		ret = -EFAULT;
466 		str = get_user_arg_ptr(argv, argc);
467 		if (IS_ERR(str))
468 			goto out;
469 
470 		len = strnlen_user(str, MAX_ARG_STRLEN);
471 		if (!len)
472 			goto out;
473 
474 		ret = -E2BIG;
475 		if (!valid_arg_len(bprm, len))
476 			goto out;
477 
478 		/* We're going to work our way backwards. */
479 		pos = bprm->p;
480 		str += len;
481 		bprm->p -= len;
482 		if (bprm_hit_stack_limit(bprm))
483 			goto out;
484 
485 		while (len > 0) {
486 			int offset, bytes_to_copy;
487 
488 			if (fatal_signal_pending(current)) {
489 				ret = -ERESTARTNOHAND;
490 				goto out;
491 			}
492 			cond_resched();
493 
494 			offset = pos % PAGE_SIZE;
495 			if (offset == 0)
496 				offset = PAGE_SIZE;
497 
498 			bytes_to_copy = offset;
499 			if (bytes_to_copy > len)
500 				bytes_to_copy = len;
501 
502 			offset -= bytes_to_copy;
503 			pos -= bytes_to_copy;
504 			str -= bytes_to_copy;
505 			len -= bytes_to_copy;
506 
507 			if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
508 				struct page *page;
509 
510 				page = get_arg_page(bprm, pos, 1);
511 				if (!page) {
512 					ret = -E2BIG;
513 					goto out;
514 				}
515 
516 				if (kmapped_page) {
517 					flush_dcache_page(kmapped_page);
518 					kunmap_local(kaddr);
519 					put_arg_page(kmapped_page);
520 				}
521 				kmapped_page = page;
522 				kaddr = kmap_local_page(kmapped_page);
523 				kpos = pos & PAGE_MASK;
524 				flush_arg_page(bprm, kpos, kmapped_page);
525 			}
526 			if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
527 				ret = -EFAULT;
528 				goto out;
529 			}
530 		}
531 	}
532 	ret = 0;
533 out:
534 	if (kmapped_page) {
535 		flush_dcache_page(kmapped_page);
536 		kunmap_local(kaddr);
537 		put_arg_page(kmapped_page);
538 	}
539 	return ret;
540 }
541 
542 /*
543  * Copy and argument/environment string from the kernel to the processes stack.
544  */
545 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
546 {
547 	int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
548 	unsigned long pos = bprm->p;
549 
550 	if (len == 0)
551 		return -EFAULT;
552 	if (!valid_arg_len(bprm, len))
553 		return -E2BIG;
554 
555 	/* We're going to work our way backwards. */
556 	arg += len;
557 	bprm->p -= len;
558 	if (bprm_hit_stack_limit(bprm))
559 		return -E2BIG;
560 
561 	while (len > 0) {
562 		unsigned int bytes_to_copy = min(len,
563 				min_not_zero(offset_in_page(pos), PAGE_SIZE));
564 		struct page *page;
565 
566 		pos -= bytes_to_copy;
567 		arg -= bytes_to_copy;
568 		len -= bytes_to_copy;
569 
570 		page = get_arg_page(bprm, pos, 1);
571 		if (!page)
572 			return -E2BIG;
573 		flush_arg_page(bprm, pos & PAGE_MASK, page);
574 		memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
575 		put_arg_page(page);
576 	}
577 
578 	return 0;
579 }
580 EXPORT_SYMBOL(copy_string_kernel);
581 
582 static int copy_strings_kernel(int argc, const char *const *argv,
583 			       struct linux_binprm *bprm)
584 {
585 	while (argc-- > 0) {
586 		int ret = copy_string_kernel(argv[argc], bprm);
587 		if (ret < 0)
588 			return ret;
589 		if (fatal_signal_pending(current))
590 			return -ERESTARTNOHAND;
591 		cond_resched();
592 	}
593 	return 0;
594 }
595 
596 #ifdef CONFIG_MMU
597 
598 /*
599  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
600  * the stack is optionally relocated, and some extra space is added.
601  */
602 int setup_arg_pages(struct linux_binprm *bprm,
603 		    unsigned long stack_top,
604 		    int executable_stack)
605 {
606 	int ret;
607 	unsigned long stack_shift;
608 	struct mm_struct *mm = current->mm;
609 	struct vm_area_struct *vma = bprm->vma;
610 	struct vm_area_struct *prev = NULL;
611 	vm_flags_t vm_flags;
612 	unsigned long stack_base;
613 	unsigned long stack_size;
614 	unsigned long stack_expand;
615 	unsigned long rlim_stack;
616 	struct mmu_gather tlb;
617 	struct vma_iterator vmi;
618 
619 #ifdef CONFIG_STACK_GROWSUP
620 	/* Limit stack size */
621 	stack_base = bprm->rlim_stack.rlim_max;
622 
623 	stack_base = calc_max_stack_size(stack_base);
624 
625 	/* Add space for stack randomization. */
626 	if (current->flags & PF_RANDOMIZE)
627 		stack_base += (STACK_RND_MASK << PAGE_SHIFT);
628 
629 	/* Make sure we didn't let the argument array grow too large. */
630 	if (vma->vm_end - vma->vm_start > stack_base)
631 		return -ENOMEM;
632 
633 	stack_base = PAGE_ALIGN(stack_top - stack_base);
634 
635 	stack_shift = vma->vm_start - stack_base;
636 	mm->arg_start = bprm->p - stack_shift;
637 	bprm->p = vma->vm_end - stack_shift;
638 #else
639 	stack_top = arch_align_stack(stack_top);
640 	stack_top = PAGE_ALIGN(stack_top);
641 
642 	if (unlikely(stack_top < mmap_min_addr) ||
643 	    unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
644 		return -ENOMEM;
645 
646 	stack_shift = vma->vm_end - stack_top;
647 
648 	bprm->p -= stack_shift;
649 	mm->arg_start = bprm->p;
650 #endif
651 
652 	bprm->exec -= stack_shift;
653 
654 	if (mmap_write_lock_killable(mm))
655 		return -EINTR;
656 
657 	vm_flags = VM_STACK_FLAGS;
658 
659 	/*
660 	 * Adjust stack execute permissions; explicitly enable for
661 	 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
662 	 * (arch default) otherwise.
663 	 */
664 	if (unlikely(executable_stack == EXSTACK_ENABLE_X))
665 		vm_flags |= VM_EXEC;
666 	else if (executable_stack == EXSTACK_DISABLE_X)
667 		vm_flags &= ~VM_EXEC;
668 	vm_flags |= mm->def_flags;
669 	vm_flags |= VM_STACK_INCOMPLETE_SETUP;
670 
671 	vma_iter_init(&vmi, mm, vma->vm_start);
672 
673 	tlb_gather_mmu(&tlb, mm);
674 	ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
675 			vm_flags);
676 	tlb_finish_mmu(&tlb);
677 
678 	if (ret)
679 		goto out_unlock;
680 	BUG_ON(prev != vma);
681 
682 	if (unlikely(vm_flags & VM_EXEC)) {
683 		pr_warn_once("process '%pD4' started with executable stack\n",
684 			     bprm->file);
685 	}
686 
687 	/* Move stack pages down in memory. */
688 	if (stack_shift) {
689 		/*
690 		 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
691 		 * the binfmt code determines where the new stack should reside, we shift it to
692 		 * its final location.
693 		 */
694 		ret = relocate_vma_down(vma, stack_shift);
695 		if (ret)
696 			goto out_unlock;
697 	}
698 
699 	/* mprotect_fixup is overkill to remove the temporary stack flags */
700 	vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
701 
702 	stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
703 	stack_size = vma->vm_end - vma->vm_start;
704 	/*
705 	 * Align this down to a page boundary as expand_stack
706 	 * will align it up.
707 	 */
708 	rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
709 
710 	stack_expand = min(rlim_stack, stack_size + stack_expand);
711 
712 #ifdef CONFIG_STACK_GROWSUP
713 	stack_base = vma->vm_start + stack_expand;
714 #else
715 	stack_base = vma->vm_end - stack_expand;
716 #endif
717 	current->mm->start_stack = bprm->p;
718 	ret = expand_stack_locked(vma, stack_base);
719 	if (ret)
720 		ret = -EFAULT;
721 
722 out_unlock:
723 	mmap_write_unlock(mm);
724 	return ret;
725 }
726 EXPORT_SYMBOL(setup_arg_pages);
727 
728 #else
729 
730 /*
731  * Transfer the program arguments and environment from the holding pages
732  * onto the stack. The provided stack pointer is adjusted accordingly.
733  */
734 int transfer_args_to_stack(struct linux_binprm *bprm,
735 			   unsigned long *sp_location)
736 {
737 	unsigned long index, stop, sp;
738 	int ret = 0;
739 
740 	stop = bprm->p >> PAGE_SHIFT;
741 	sp = *sp_location;
742 
743 	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
744 		unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
745 		char *src = kmap_local_page(bprm->page[index]) + offset;
746 		sp -= PAGE_SIZE - offset;
747 		if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
748 			ret = -EFAULT;
749 		kunmap_local(src);
750 		if (ret)
751 			goto out;
752 	}
753 
754 	bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
755 	*sp_location = sp;
756 
757 out:
758 	return ret;
759 }
760 EXPORT_SYMBOL(transfer_args_to_stack);
761 
762 #endif /* CONFIG_MMU */
763 
764 /*
765  * On success, caller must call do_close_execat() on the returned
766  * struct file to close it.
767  */
768 static struct file *do_open_execat(int fd, struct filename *name, int flags)
769 {
770 	int err;
771 	struct file *file __free(fput) = NULL;
772 	struct open_flags open_exec_flags = {
773 		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
774 		.acc_mode = MAY_EXEC,
775 		.intent = LOOKUP_OPEN,
776 		.lookup_flags = LOOKUP_FOLLOW,
777 	};
778 
779 	if ((flags &
780 	     ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH | AT_EXECVE_CHECK)) != 0)
781 		return ERR_PTR(-EINVAL);
782 	if (flags & AT_SYMLINK_NOFOLLOW)
783 		open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
784 
785 	file = do_file_open(fd, name, &open_exec_flags);
786 	if (IS_ERR(file))
787 		return file;
788 
789 	if (path_noexec(&file->f_path))
790 		return ERR_PTR(-EACCES);
791 
792 	/*
793 	 * In the past the regular type check was here. It moved to may_open() in
794 	 * 633fb6ac3980 ("exec: move S_ISREG() check earlier"). Since then it is
795 	 * an invariant that all non-regular files error out before we get here.
796 	 */
797 	if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode)))
798 		return ERR_PTR(-EACCES);
799 
800 	err = exe_file_deny_write_access(file);
801 	if (err)
802 		return ERR_PTR(err);
803 
804 	return no_free_ptr(file);
805 }
806 
807 /**
808  * open_exec - Open a path name for execution
809  *
810  * @name: path name to open with the intent of executing it.
811  *
812  * Returns ERR_PTR on failure or allocated struct file on success.
813  *
814  * As this is a wrapper for the internal do_open_execat(), callers
815  * must call exe_file_allow_write_access() before fput() on release. Also see
816  * do_close_execat().
817  */
818 struct file *open_exec(const char *name)
819 {
820 	CLASS(filename_kernel, filename)(name);
821 	return do_open_execat(AT_FDCWD, filename, 0);
822 }
823 EXPORT_SYMBOL(open_exec);
824 
825 #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
826 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
827 {
828 	ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
829 	if (res > 0)
830 		flush_icache_user_range(addr, addr + len);
831 	return res;
832 }
833 EXPORT_SYMBOL(read_code);
834 #endif
835 
836 /*
837  * Maps the mm_struct mm into the current task struct.
838  * On success, this function returns with exec_update_lock
839  * held for writing. The replaced address space is stashed in
840  * bprm->old_mm for setup_new_exec() to release outside the lock.
841  */
842 static int exec_mmap(struct linux_binprm *bprm)
843 {
844 	struct task_exec_state *exec_state __free(put_task_exec_state) = NULL;
845 	struct mm_struct *mm = bprm->mm;
846 	struct task_struct *tsk;
847 	struct mm_struct *old_mm, *active_mm;
848 	int ret;
849 
850 	exec_state = alloc_task_exec_state(bprm->user_ns);
851 	if (!exec_state)
852 		return -ENOMEM;
853 
854 	/* Notify parent that we're no longer interested in the old VM */
855 	tsk = current;
856 	old_mm = current->mm;
857 	exec_mm_release(tsk, old_mm);
858 
859 	ret = down_write_killable(&tsk->signal->exec_update_lock);
860 	if (ret)
861 		return ret;
862 
863 	if (old_mm) {
864 		/*
865 		 * If there is a pending fatal signal perhaps a signal
866 		 * whose default action is to create a coredump get
867 		 * out and die instead of going through with the exec.
868 		 */
869 		ret = mmap_read_lock_killable(old_mm);
870 		if (ret) {
871 			up_write(&tsk->signal->exec_update_lock);
872 			return ret;
873 		}
874 	}
875 
876 	task_lock(tsk);
877 	membarrier_exec_mmap(mm);
878 
879 	local_irq_disable();
880 	active_mm = tsk->active_mm;
881 	tsk->active_mm = mm;
882 	tsk->mm = mm;
883 	mm_init_cid(mm, tsk);
884 	exec_state = task_exec_state_replace(tsk, exec_state);
885 	/*
886 	 * This prevents preemption while active_mm is being loaded and
887 	 * it and mm are being updated, which could cause problems for
888 	 * lazy tlb mm refcounting when these are updated by context
889 	 * switches. Not all architectures can handle irqs off over
890 	 * activate_mm yet.
891 	 */
892 	if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
893 		local_irq_enable();
894 	activate_mm(active_mm, mm);
895 	if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
896 		local_irq_enable();
897 	lru_gen_add_mm(mm);
898 	task_unlock(tsk);
899 	lru_gen_use_mm(mm);
900 	if (old_mm) {
901 		mmap_read_unlock(old_mm);
902 		BUG_ON(active_mm != old_mm);
903 		/* Defer teardown to setup_new_exec(), outside the exec locks. */
904 		bprm->old_mm = old_mm;
905 		return 0;
906 	}
907 	mmdrop_lazy_tlb(active_mm);
908 	return 0;
909 }
910 
911 /* Release the address space replaced by exec, outside the exec locks. */
912 static void exec_mm_put_old(struct mm_struct *old_mm)
913 {
914 	setmax_mm_hiwater_rss(&current->signal->maxrss, old_mm);
915 	mm_update_next_owner(old_mm);
916 	mmput(old_mm);
917 }
918 
919 static int de_thread(struct task_struct *tsk)
920 {
921 	struct signal_struct *sig = tsk->signal;
922 	struct sighand_struct *oldsighand = tsk->sighand;
923 	spinlock_t *lock = &oldsighand->siglock;
924 
925 	if (thread_group_empty(tsk))
926 		goto no_thread_group;
927 
928 	/*
929 	 * Kill all other threads in the thread group.
930 	 */
931 	spin_lock_irq(lock);
932 	if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
933 		/*
934 		 * Another group action in progress, just
935 		 * return so that the signal is processed.
936 		 */
937 		spin_unlock_irq(lock);
938 		return -EAGAIN;
939 	}
940 
941 	sig->group_exec_task = tsk;
942 	sig->notify_count = zap_other_threads(tsk);
943 	if (!thread_group_leader(tsk))
944 		sig->notify_count--;
945 
946 	while (sig->notify_count) {
947 		__set_current_state(TASK_KILLABLE);
948 		spin_unlock_irq(lock);
949 		schedule();
950 		if (__fatal_signal_pending(tsk))
951 			goto killed;
952 		spin_lock_irq(lock);
953 	}
954 	spin_unlock_irq(lock);
955 
956 	/*
957 	 * At this point all other threads have exited, all we have to
958 	 * do is to wait for the thread group leader to become inactive,
959 	 * and to assume its PID:
960 	 */
961 	if (!thread_group_leader(tsk)) {
962 		struct task_struct *leader = tsk->group_leader;
963 
964 		for (;;) {
965 			cgroup_threadgroup_change_begin(tsk);
966 			write_lock_irq(&tasklist_lock);
967 			/*
968 			 * Do this under tasklist_lock to ensure that
969 			 * exit_notify() can't miss ->group_exec_task
970 			 */
971 			sig->notify_count = -1;
972 			if (likely(leader->exit_state))
973 				break;
974 			__set_current_state(TASK_KILLABLE);
975 			write_unlock_irq(&tasklist_lock);
976 			cgroup_threadgroup_change_end(tsk);
977 			schedule();
978 			if (__fatal_signal_pending(tsk))
979 				goto killed;
980 		}
981 
982 		/*
983 		 * The only record we have of the real-time age of a
984 		 * process, regardless of execs it's done, is start_time.
985 		 * All the past CPU time is accumulated in signal_struct
986 		 * from sister threads now dead.  But in this non-leader
987 		 * exec, nothing survives from the original leader thread,
988 		 * whose birth marks the true age of this process now.
989 		 * When we take on its identity by switching to its PID, we
990 		 * also take its birthdate (always earlier than our own).
991 		 */
992 		tsk->start_time = leader->start_time;
993 		tsk->start_boottime = leader->start_boottime;
994 
995 		BUG_ON(!same_thread_group(leader, tsk));
996 		/*
997 		 * An exec() starts a new thread group with the
998 		 * TGID of the previous thread group. Rehash the
999 		 * two threads with a switched PID, and release
1000 		 * the former thread group leader:
1001 		 */
1002 
1003 		/* Become a process group leader with the old leader's pid.
1004 		 * The old leader becomes a thread of the this thread group.
1005 		 */
1006 		exchange_tids(tsk, leader);
1007 		transfer_pid(leader, tsk, PIDTYPE_TGID);
1008 		transfer_pid(leader, tsk, PIDTYPE_PGID);
1009 		transfer_pid(leader, tsk, PIDTYPE_SID);
1010 
1011 		list_replace_rcu(&leader->tasks, &tsk->tasks);
1012 		list_replace_init(&leader->sibling, &tsk->sibling);
1013 
1014 		tsk->group_leader = tsk;
1015 		leader->group_leader = tsk;
1016 
1017 		tsk->exit_signal = SIGCHLD;
1018 		leader->exit_signal = -1;
1019 
1020 		BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1021 		leader->exit_state = EXIT_DEAD;
1022 		/*
1023 		 * We are going to release_task()->ptrace_unlink() silently,
1024 		 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1025 		 * the tracer won't block again waiting for this thread.
1026 		 */
1027 		if (unlikely(leader->ptrace))
1028 			__wake_up_parent(leader, leader->parent);
1029 		write_unlock_irq(&tasklist_lock);
1030 		cgroup_threadgroup_change_end(tsk);
1031 
1032 		release_task(leader);
1033 	}
1034 
1035 	sig->group_exec_task = NULL;
1036 	sig->notify_count = 0;
1037 
1038 no_thread_group:
1039 	/* we have changed execution domain */
1040 	tsk->exit_signal = SIGCHLD;
1041 
1042 	BUG_ON(!thread_group_leader(tsk));
1043 	return 0;
1044 
1045 killed:
1046 	/* protects against exit_notify() and __exit_signal() */
1047 	read_lock(&tasklist_lock);
1048 	sig->group_exec_task = NULL;
1049 	sig->notify_count = 0;
1050 	read_unlock(&tasklist_lock);
1051 	return -EAGAIN;
1052 }
1053 
1054 
1055 /*
1056  * This function makes sure the current process has its own signal table,
1057  * so that flush_signal_handlers can later reset the handlers without
1058  * disturbing other processes.  (Other processes might share the signal
1059  * table via the CLONE_SIGHAND option to clone().)
1060  */
1061 static int unshare_sighand(struct task_struct *me)
1062 {
1063 	struct sighand_struct *oldsighand = me->sighand;
1064 
1065 	if (refcount_read(&oldsighand->count) != 1) {
1066 		struct sighand_struct *newsighand;
1067 		/*
1068 		 * This ->sighand is shared with the CLONE_SIGHAND
1069 		 * but not CLONE_THREAD task, switch to the new one.
1070 		 */
1071 		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1072 		if (!newsighand)
1073 			return -ENOMEM;
1074 
1075 		refcount_set(&newsighand->count, 1);
1076 
1077 		write_lock_irq(&tasklist_lock);
1078 		spin_lock(&oldsighand->siglock);
1079 		memcpy(newsighand->action, oldsighand->action,
1080 		       sizeof(newsighand->action));
1081 		rcu_assign_pointer(me->sighand, newsighand);
1082 		spin_unlock(&oldsighand->siglock);
1083 		write_unlock_irq(&tasklist_lock);
1084 
1085 		__cleanup_sighand(oldsighand);
1086 	}
1087 	return 0;
1088 }
1089 
1090 /*
1091  * This is unlocked -- the string will always be NUL-terminated, but
1092  * may show overlapping contents if racing concurrent reads.
1093  */
1094 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1095 {
1096 	size_t len = strnlen(buf, sizeof(tsk->comm) - 1);
1097 
1098 	trace_task_rename(tsk, buf);
1099 	memcpy(tsk->comm, buf, len);
1100 	memset(&tsk->comm[len], 0, sizeof(tsk->comm) - len);
1101 	perf_event_comm(tsk, exec);
1102 }
1103 
1104 /*
1105  * Calling this is the point of no return. None of the failures will be
1106  * seen by userspace since either the process is already taking a fatal
1107  * signal (via de_thread() or coredump), or will have SEGV raised
1108  * (after exec_mmap()) by search_binary_handler (see below).
1109  */
1110 int begin_new_exec(struct linux_binprm * bprm)
1111 {
1112 	struct task_struct *me = current;
1113 	int retval;
1114 
1115 	/* Once we are committed compute the creds */
1116 	retval = bprm_creds_from_file(bprm);
1117 	if (retval)
1118 		return retval;
1119 
1120 	/*
1121 	 * This tracepoint marks the point before flushing the old exec where
1122 	 * the current task is still unchanged, but errors are fatal (point of
1123 	 * no return). The later "sched_process_exec" tracepoint is called after
1124 	 * the current task has successfully switched to the new exec.
1125 	 */
1126 	trace_sched_prepare_exec(current, bprm);
1127 
1128 	/*
1129 	 * Ensure all future errors are fatal.
1130 	 */
1131 	bprm->point_of_no_return = true;
1132 
1133 	/* Make this the only thread in the thread group */
1134 	retval = de_thread(me);
1135 	if (retval)
1136 		goto out;
1137 	/* see the comment in check_unsafe_exec() */
1138 	current->fs->in_exec = 0;
1139 	/*
1140 	 * Cancel any io_uring activity across execve
1141 	 */
1142 	io_uring_task_cancel();
1143 
1144 	/* Ensure the files table is not shared. */
1145 	retval = unshare_files();
1146 	if (retval)
1147 		goto out;
1148 
1149 	/*
1150 	 * Must be called _before_ exec_mmap() as bprm->mm is
1151 	 * not visible until then. Doing it here also ensures
1152 	 * we don't race against replace_mm_exe_file().
1153 	 */
1154 	retval = set_mm_exe_file(bprm->mm, bprm->file);
1155 	if (retval)
1156 		goto out;
1157 
1158 	/* If the binary is not readable then enforce mm->dumpable=0 */
1159 	would_dump(bprm, bprm->file);
1160 	if (bprm->have_execfd)
1161 		would_dump(bprm, bprm->executable);
1162 
1163 	/*
1164 	 * Release all of the old mmap stuff
1165 	 */
1166 	acct_arg_size(bprm, 0);
1167 	retval = exec_mmap(bprm);
1168 	if (retval)
1169 		goto out;
1170 
1171 	bprm->mm = NULL;
1172 
1173 	retval = exec_task_namespaces();
1174 	if (retval)
1175 		goto out_unlock;
1176 
1177 #ifdef CONFIG_POSIX_TIMERS
1178 	spin_lock_irq(&me->sighand->siglock);
1179 	posix_cpu_timers_exit(me);
1180 	spin_unlock_irq(&me->sighand->siglock);
1181 	exit_itimers(me);
1182 	flush_itimer_signals();
1183 #endif
1184 
1185 	/*
1186 	 * Make the signal table private.
1187 	 */
1188 	retval = unshare_sighand(me);
1189 	if (retval)
1190 		goto out_unlock;
1191 
1192 	me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1193 					PF_NOFREEZE | PF_NO_SETAFFINITY);
1194 	flush_thread();
1195 	me->personality &= ~bprm->per_clear;
1196 
1197 	clear_syscall_work_syscall_user_dispatch(me);
1198 
1199 	/*
1200 	 * We have to apply CLOEXEC before we change whether the process is
1201 	 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1202 	 * trying to access the should-be-closed file descriptors of a process
1203 	 * undergoing exec(2).
1204 	 */
1205 	do_close_on_exec(me->files);
1206 
1207 	if (bprm->secureexec) {
1208 		/* Make sure parent cannot signal privileged process. */
1209 		me->pdeath_signal = 0;
1210 
1211 		/*
1212 		 * For secureexec, reset the stack limit to sane default to
1213 		 * avoid bad behavior from the prior rlimits. This has to
1214 		 * happen before arch_pick_mmap_layout(), which examines
1215 		 * RLIMIT_STACK, but after the point of no return to avoid
1216 		 * needing to clean up the change on failure.
1217 		 */
1218 		if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1219 			bprm->rlim_stack.rlim_cur = _STK_LIM;
1220 	}
1221 
1222 	me->sas_ss_sp = me->sas_ss_size = 0;
1223 
1224 	/*
1225 	 * Figure out dumpability. Note that this checking only of current
1226 	 * is wrong, but userspace depends on it. This should be testing
1227 	 * bprm->secureexec instead.
1228 	 */
1229 	if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1230 	    !(uid_eq(current_euid(), current_uid()) &&
1231 	      gid_eq(current_egid(), current_gid())))
1232 		task_exec_state_set_dumpable(suid_dumpable);
1233 	else
1234 		task_exec_state_set_dumpable(TASK_DUMPABLE_OWNER);
1235 
1236 	perf_event_exec();
1237 
1238 	/*
1239 	 * If the original filename was empty, alloc_bprm() made up a path
1240 	 * that will probably not be useful to admins running ps or similar.
1241 	 * Let's fix it up to be something reasonable.
1242 	 */
1243 	if (bprm->comm_from_dentry) {
1244 		/*
1245 		 * Hold RCU lock to keep the name from being freed behind our back.
1246 		 * Use acquire semantics to make sure the terminating NUL from
1247 		 * __d_alloc() is seen.
1248 		 *
1249 		 * Note, we're deliberately sloppy here. We don't need to care about
1250 		 * detecting a concurrent rename and just want a terminated name.
1251 		 */
1252 		rcu_read_lock();
1253 		__set_task_comm(me, smp_load_acquire(&bprm->file->f_path.dentry->d_name.name),
1254 				true);
1255 		rcu_read_unlock();
1256 	} else {
1257 		__set_task_comm(me, kbasename(bprm->filename), true);
1258 	}
1259 
1260 	/* An exec changes our domain. We are no longer part of the thread
1261 	   group */
1262 	WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1263 	flush_signal_handlers(me, 0);
1264 
1265 	retval = set_cred_ucounts(bprm->cred);
1266 	if (retval < 0)
1267 		goto out_unlock;
1268 
1269 	/*
1270 	 * install the new credentials for this executable
1271 	 */
1272 	security_bprm_committing_creds(bprm);
1273 
1274 	commit_creds(bprm->cred);
1275 	bprm->cred = NULL;
1276 
1277 	/*
1278 	 * Disable monitoring for regular users
1279 	 * when executing setuid binaries. Must
1280 	 * wait until new credentials are committed
1281 	 * by commit_creds() above
1282 	 */
1283 	if (task_exec_state_get_dumpable(me) != TASK_DUMPABLE_OWNER)
1284 		perf_event_exit_task(me);
1285 	/*
1286 	 * cred_guard_mutex must be held at least to this point to prevent
1287 	 * ptrace_attach() from altering our determination of the task's
1288 	 * credentials; any time after this it may be unlocked.
1289 	 */
1290 	security_bprm_committed_creds(bprm);
1291 
1292 	/* Pass the opened binary to the interpreter. */
1293 	if (bprm->have_execfd) {
1294 		retval = FD_ADD(0, bprm->executable);
1295 		if (retval < 0)
1296 			goto out_unlock;
1297 		bprm->executable = NULL;
1298 		bprm->execfd = retval;
1299 	}
1300 	return 0;
1301 
1302 out_unlock:
1303 	up_write(&me->signal->exec_update_lock);
1304 	if (!bprm->cred)
1305 		mutex_unlock(&me->signal->cred_guard_mutex);
1306 
1307 out:
1308 	return retval;
1309 }
1310 EXPORT_SYMBOL(begin_new_exec);
1311 
1312 void would_dump(struct linux_binprm *bprm, struct file *file)
1313 {
1314 	struct inode *inode = file_inode(file);
1315 	struct mnt_idmap *idmap = file_mnt_idmap(file);
1316 	if (inode_permission(idmap, inode, MAY_READ) < 0) {
1317 		struct user_namespace *old, *user_ns;
1318 		bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1319 
1320 		/* Ensure bprm->user_ns contains the executable. */
1321 		user_ns = old = bprm->user_ns;
1322 		while ((user_ns != &init_user_ns) &&
1323 		       !privileged_wrt_inode_uidgid(user_ns, idmap, inode))
1324 			user_ns = user_ns->parent;
1325 
1326 		if (old != user_ns) {
1327 			bprm->user_ns = get_user_ns(user_ns);
1328 			put_user_ns(old);
1329 		}
1330 	}
1331 }
1332 EXPORT_SYMBOL(would_dump);
1333 
1334 void setup_new_exec(struct linux_binprm * bprm)
1335 {
1336 	/* Setup things that can depend upon the personality */
1337 	struct task_struct *me = current;
1338 
1339 	arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1340 
1341 	arch_setup_new_exec();
1342 
1343 	/* Set the new mm task size. We have to do that late because it may
1344 	 * depend on TIF_32BIT which is only updated in flush_thread() on
1345 	 * some architectures like powerpc
1346 	 */
1347 	me->mm->task_size = TASK_SIZE;
1348 	up_write(&me->signal->exec_update_lock);
1349 	mutex_unlock(&me->signal->cred_guard_mutex);
1350 
1351 	/* The exec locks are dropped: release the old address space now. */
1352 	if (bprm->old_mm) {
1353 		exec_mm_put_old(bprm->old_mm);
1354 		bprm->old_mm = NULL;
1355 	}
1356 }
1357 EXPORT_SYMBOL(setup_new_exec);
1358 
1359 /* Runs immediately before start_thread() takes over. */
1360 void finalize_exec(struct linux_binprm *bprm)
1361 {
1362 	/* Store any stack rlimit changes before starting thread. */
1363 	task_lock(current->group_leader);
1364 	current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1365 	task_unlock(current->group_leader);
1366 }
1367 EXPORT_SYMBOL(finalize_exec);
1368 
1369 /*
1370  * Prepare credentials and lock ->cred_guard_mutex.
1371  * setup_new_exec() commits the new creds and drops the lock.
1372  * Or, if exec fails before, free_bprm() should release ->cred
1373  * and unlock.
1374  */
1375 static int prepare_bprm_creds(struct linux_binprm *bprm)
1376 {
1377 	if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1378 		return -ERESTARTNOINTR;
1379 
1380 	bprm->cred = prepare_exec_creds();
1381 	if (likely(bprm->cred))
1382 		return 0;
1383 
1384 	mutex_unlock(&current->signal->cred_guard_mutex);
1385 	return -ENOMEM;
1386 }
1387 
1388 /* Matches do_open_execat() */
1389 static void do_close_execat(struct file *file)
1390 {
1391 	if (!file)
1392 		return;
1393 	exe_file_allow_write_access(file);
1394 	fput(file);
1395 }
1396 
1397 static void free_bprm(struct linux_binprm *bprm)
1398 {
1399 	if (bprm->mm) {
1400 		acct_arg_size(bprm, 0);
1401 		mmput(bprm->mm);
1402 	}
1403 	if (bprm->user_ns)
1404 		put_user_ns(bprm->user_ns);
1405 	free_arg_pages(bprm);
1406 	if (bprm->cred) {
1407 		/* in case exec fails before de_thread() succeeds */
1408 		current->fs->in_exec = 0;
1409 		mutex_unlock(&current->signal->cred_guard_mutex);
1410 		abort_creds(bprm->cred);
1411 	}
1412 	/* exec swapped the mm but failed before setup_new_exec() freed it */
1413 	if (bprm->old_mm)
1414 		exec_mm_put_old(bprm->old_mm);
1415 	do_close_execat(bprm->file);
1416 	if (bprm->executable)
1417 		fput(bprm->executable);
1418 	/* If a binfmt changed the interp, free it. */
1419 	if (bprm->interp != bprm->filename)
1420 		kfree(bprm->interp);
1421 	kfree(bprm->fdpath);
1422 	kfree(bprm);
1423 }
1424 
1425 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename, int flags)
1426 {
1427 	struct linux_binprm *bprm;
1428 	struct file *file;
1429 	int retval = -ENOMEM;
1430 
1431 	file = do_open_execat(fd, filename, flags);
1432 	if (IS_ERR(file))
1433 		return ERR_CAST(file);
1434 
1435 	bprm = kzalloc_obj(*bprm);
1436 	if (!bprm) {
1437 		do_close_execat(file);
1438 		return ERR_PTR(-ENOMEM);
1439 	}
1440 
1441 	bprm->file = file;
1442 
1443 	if (fd == AT_FDCWD || filename->name[0] == '/') {
1444 		bprm->filename = filename->name;
1445 	} else {
1446 		if (filename->name[0] == '\0') {
1447 			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1448 			bprm->comm_from_dentry = 1;
1449 		} else {
1450 			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1451 						  fd, filename->name);
1452 		}
1453 		if (!bprm->fdpath)
1454 			goto out_free;
1455 
1456 		/*
1457 		 * Record that a name derived from an O_CLOEXEC fd will be
1458 		 * inaccessible after exec.  This allows the code in exec to
1459 		 * choose to fail when the executable is not mmaped into the
1460 		 * interpreter and an open file descriptor is not passed to
1461 		 * the interpreter.  This makes for a better user experience
1462 		 * than having the interpreter start and then immediately fail
1463 		 * when it finds the executable is inaccessible.
1464 		 */
1465 		if (get_close_on_exec(fd))
1466 			bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1467 
1468 		bprm->filename = bprm->fdpath;
1469 	}
1470 	bprm->interp = bprm->filename;
1471 
1472 	/*
1473 	 * At this point, security_file_open() has already been called (with
1474 	 * __FMODE_EXEC) and access control checks for AT_EXECVE_CHECK will
1475 	 * stop just after the security_bprm_creds_for_exec() call in
1476 	 * bprm_execve().  Indeed, the kernel should not try to parse the
1477 	 * content of the file with exec_binprm() nor change the calling
1478 	 * thread, which means that the following security functions will not
1479 	 * be called:
1480 	 * - security_bprm_check()
1481 	 * - security_bprm_creds_from_file()
1482 	 * - security_bprm_committing_creds()
1483 	 * - security_bprm_committed_creds()
1484 	 */
1485 	bprm->is_check = !!(flags & AT_EXECVE_CHECK);
1486 
1487 	retval = bprm_mm_init(bprm);
1488 	if (!retval)
1489 		return bprm;
1490 
1491 out_free:
1492 	free_bprm(bprm);
1493 	return ERR_PTR(retval);
1494 }
1495 
1496 DEFINE_CLASS(bprm, struct linux_binprm *, if (!IS_ERR(_T)) free_bprm(_T),
1497 	alloc_bprm(fd, name, flags), int fd, struct filename *name, int flags)
1498 
1499 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1500 {
1501 	/* If a binfmt changed the interp, free it first. */
1502 	if (bprm->interp != bprm->filename)
1503 		kfree(bprm->interp);
1504 	bprm->interp = kstrdup(interp, GFP_KERNEL);
1505 	if (!bprm->interp)
1506 		return -ENOMEM;
1507 	return 0;
1508 }
1509 EXPORT_SYMBOL(bprm_change_interp);
1510 
1511 /*
1512  * determine how safe it is to execute the proposed program
1513  * - the caller must hold ->cred_guard_mutex to protect against
1514  *   PTRACE_ATTACH or seccomp thread-sync
1515  */
1516 static void check_unsafe_exec(struct linux_binprm *bprm)
1517 {
1518 	struct task_struct *p = current, *t;
1519 	unsigned n_fs;
1520 
1521 	if (p->ptrace)
1522 		bprm->unsafe |= LSM_UNSAFE_PTRACE;
1523 
1524 	/*
1525 	 * This isn't strictly necessary, but it makes it harder for LSMs to
1526 	 * mess up.
1527 	 */
1528 	if (task_no_new_privs(current))
1529 		bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1530 
1531 	/*
1532 	 * If another task is sharing our fs, we cannot safely
1533 	 * suid exec because the differently privileged task
1534 	 * will be able to manipulate the current directory, etc.
1535 	 * It would be nice to force an unshare instead...
1536 	 *
1537 	 * Otherwise we set fs->in_exec = 1 to deny clone(CLONE_FS)
1538 	 * from another sub-thread until de_thread() succeeds, this
1539 	 * state is protected by cred_guard_mutex we hold.
1540 	 */
1541 	n_fs = 1;
1542 	read_seqlock_excl(&p->fs->seq);
1543 	rcu_read_lock();
1544 	for_other_threads(p, t) {
1545 		if (t->fs == p->fs)
1546 			n_fs++;
1547 	}
1548 	rcu_read_unlock();
1549 
1550 	/* "users" and "in_exec" locked for copy_fs() */
1551 	if (p->fs->users > n_fs)
1552 		bprm->unsafe |= LSM_UNSAFE_SHARE;
1553 	else
1554 		p->fs->in_exec = 1;
1555 	read_sequnlock_excl(&p->fs->seq);
1556 }
1557 
1558 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1559 {
1560 	/* Handle suid and sgid on files */
1561 	struct mnt_idmap *idmap;
1562 	struct inode *inode = file_inode(file);
1563 	unsigned int mode;
1564 	vfsuid_t vfsuid;
1565 	vfsgid_t vfsgid;
1566 	int err;
1567 
1568 	if (!mnt_may_suid(file->f_path.mnt))
1569 		return;
1570 
1571 	if (task_no_new_privs(current))
1572 		return;
1573 
1574 	mode = READ_ONCE(inode->i_mode);
1575 	if (!(mode & (S_ISUID|S_ISGID)))
1576 		return;
1577 
1578 	idmap = file_mnt_idmap(file);
1579 
1580 	/* Be careful if suid/sgid is set */
1581 	inode_lock(inode);
1582 
1583 	/* Atomically reload and check mode/uid/gid now that lock held. */
1584 	mode = inode->i_mode;
1585 	vfsuid = i_uid_into_vfsuid(idmap, inode);
1586 	vfsgid = i_gid_into_vfsgid(idmap, inode);
1587 	err = inode_permission(idmap, inode, MAY_EXEC);
1588 	inode_unlock(inode);
1589 
1590 	/* Did the exec bit vanish out from under us? Give up. */
1591 	if (err)
1592 		return;
1593 
1594 	/* We ignore suid/sgid if there are no mappings for them in the ns */
1595 	if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1596 	    !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1597 		return;
1598 
1599 	if (mode & S_ISUID) {
1600 		bprm->per_clear |= PER_CLEAR_ON_SETID;
1601 		bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1602 	}
1603 
1604 	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1605 		bprm->per_clear |= PER_CLEAR_ON_SETID;
1606 		bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1607 	}
1608 }
1609 
1610 /*
1611  * Compute brpm->cred based upon the final binary.
1612  */
1613 static int bprm_creds_from_file(struct linux_binprm *bprm)
1614 {
1615 	/* Compute creds based on which file? */
1616 	struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1617 
1618 	bprm_fill_uid(bprm, file);
1619 	return security_bprm_creds_from_file(bprm, file);
1620 }
1621 
1622 /*
1623  * Fill the binprm structure from the inode.
1624  * Read the first BINPRM_BUF_SIZE bytes
1625  *
1626  * This may be called multiple times for binary chains (scripts for example).
1627  */
1628 static int prepare_binprm(struct linux_binprm *bprm)
1629 {
1630 	loff_t pos = 0;
1631 
1632 	memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1633 	return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1634 }
1635 
1636 /*
1637  * Arguments are '\0' separated strings found at the location bprm->p
1638  * points to; chop off the first by relocating brpm->p to right after
1639  * the first '\0' encountered.
1640  */
1641 int remove_arg_zero(struct linux_binprm *bprm)
1642 {
1643 	unsigned long offset;
1644 	char *kaddr;
1645 	struct page *page;
1646 
1647 	if (!bprm->argc)
1648 		return 0;
1649 
1650 	do {
1651 		offset = bprm->p & ~PAGE_MASK;
1652 		page = get_arg_page(bprm, bprm->p, 0);
1653 		if (!page)
1654 			return -EFAULT;
1655 		kaddr = kmap_local_page(page);
1656 
1657 		for (; offset < PAGE_SIZE && kaddr[offset];
1658 				offset++, bprm->p++)
1659 			;
1660 
1661 		kunmap_local(kaddr);
1662 		put_arg_page(page);
1663 	} while (offset == PAGE_SIZE);
1664 
1665 	bprm->p++;
1666 	bprm->argc--;
1667 
1668 	return 0;
1669 }
1670 EXPORT_SYMBOL(remove_arg_zero);
1671 
1672 /*
1673  * cycle the list of binary formats handler, until one recognizes the image
1674  */
1675 static int search_binary_handler(struct linux_binprm *bprm)
1676 {
1677 	struct linux_binfmt *fmt;
1678 	int retval;
1679 
1680 	retval = prepare_binprm(bprm);
1681 	if (retval < 0)
1682 		return retval;
1683 
1684 	retval = security_bprm_check(bprm);
1685 	if (retval)
1686 		return retval;
1687 
1688 	read_lock(&binfmt_lock);
1689 	list_for_each_entry(fmt, &formats, lh) {
1690 		if (!try_module_get(fmt->module))
1691 			continue;
1692 		read_unlock(&binfmt_lock);
1693 
1694 		retval = fmt->load_binary(bprm);
1695 
1696 		read_lock(&binfmt_lock);
1697 		put_binfmt(fmt);
1698 		if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1699 			read_unlock(&binfmt_lock);
1700 			return retval;
1701 		}
1702 	}
1703 	read_unlock(&binfmt_lock);
1704 
1705 	return -ENOEXEC;
1706 }
1707 
1708 /* binfmt handlers will call back into begin_new_exec() on success. */
1709 static int exec_binprm(struct linux_binprm *bprm)
1710 {
1711 	pid_t old_pid, old_vpid;
1712 	int ret, depth;
1713 
1714 	/* Need to fetch pid before load_binary changes it */
1715 	old_pid = current->pid;
1716 	rcu_read_lock();
1717 	old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1718 	rcu_read_unlock();
1719 
1720 	/* This allows 4 levels of binfmt rewrites before failing hard. */
1721 	for (depth = 0;; depth++) {
1722 		struct file *exec;
1723 		if (depth > 5)
1724 			return -ELOOP;
1725 
1726 		ret = search_binary_handler(bprm);
1727 		if (ret < 0)
1728 			return ret;
1729 		if (!bprm->interpreter)
1730 			break;
1731 
1732 		exec = bprm->file;
1733 		bprm->file = bprm->interpreter;
1734 		bprm->interpreter = NULL;
1735 
1736 		exe_file_allow_write_access(exec);
1737 		if (unlikely(bprm->have_execfd)) {
1738 			if (bprm->executable) {
1739 				fput(exec);
1740 				return -ENOEXEC;
1741 			}
1742 			bprm->executable = exec;
1743 		} else
1744 			fput(exec);
1745 	}
1746 
1747 	audit_bprm(bprm);
1748 	trace_sched_process_exec(current, old_pid, bprm);
1749 	ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1750 	proc_exec_connector(current);
1751 	return 0;
1752 }
1753 
1754 static int bprm_execve(struct linux_binprm *bprm)
1755 {
1756 	int retval;
1757 
1758 	retval = prepare_bprm_creds(bprm);
1759 	if (retval)
1760 		return retval;
1761 
1762 	/*
1763 	 * Check for unsafe execution states before exec_binprm(), which
1764 	 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1765 	 * where setuid-ness is evaluated.
1766 	 */
1767 	check_unsafe_exec(bprm);
1768 	current->in_execve = 1;
1769 	sched_mm_cid_before_execve(current);
1770 
1771 	sched_exec();
1772 
1773 	/* Set the unchanging part of bprm->cred */
1774 	retval = security_bprm_creds_for_exec(bprm);
1775 	if (retval || bprm->is_check)
1776 		goto out;
1777 
1778 	retval = exec_binprm(bprm);
1779 	if (retval < 0)
1780 		goto out;
1781 
1782 	sched_mm_cid_after_execve(current);
1783 	rseq_execve(current);
1784 	/* execve succeeded */
1785 	current->in_execve = 0;
1786 	user_events_execve(current);
1787 	acct_update_integrals(current);
1788 	task_numa_free(current, false);
1789 	return retval;
1790 
1791 out:
1792 	/*
1793 	 * If past the point of no return ensure the code never
1794 	 * returns to the userspace process.  Use an existing fatal
1795 	 * signal if present otherwise terminate the process with
1796 	 * SIGSEGV.
1797 	 */
1798 	if (bprm->point_of_no_return && !fatal_signal_pending(current))
1799 		force_fatal_sig(SIGSEGV);
1800 
1801 	sched_mm_cid_after_execve(current);
1802 	rseq_force_update();
1803 	current->in_execve = 0;
1804 
1805 	return retval;
1806 }
1807 
1808 static int do_execveat_common(int fd, struct filename *filename,
1809 			      struct user_arg_ptr argv,
1810 			      struct user_arg_ptr envp,
1811 			      int flags)
1812 {
1813 	int retval;
1814 
1815 	/*
1816 	 * We move the actual failure in case of RLIMIT_NPROC excess from
1817 	 * set*uid() to execve() because too many poorly written programs
1818 	 * don't check setuid() return code.  Here we additionally recheck
1819 	 * whether NPROC limit is still exceeded.
1820 	 */
1821 	if ((current->flags & PF_NPROC_EXCEEDED) &&
1822 	    is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)))
1823 		return -EAGAIN;
1824 
1825 	/* We're below the limit (still or again), so we don't want to make
1826 	 * further execve() calls fail. */
1827 	current->flags &= ~PF_NPROC_EXCEEDED;
1828 
1829 	CLASS(bprm, bprm)(fd, filename, flags);
1830 	if (IS_ERR(bprm))
1831 		return PTR_ERR(bprm);
1832 
1833 	retval = count(argv, MAX_ARG_STRINGS);
1834 	if (retval < 0)
1835 		return retval;
1836 	bprm->argc = retval;
1837 
1838 	retval = count(envp, MAX_ARG_STRINGS);
1839 	if (retval < 0)
1840 		return retval;
1841 	bprm->envc = retval;
1842 
1843 	retval = bprm_stack_limits(bprm);
1844 	if (retval < 0)
1845 		return retval;
1846 
1847 	retval = copy_string_kernel(bprm->filename, bprm);
1848 	if (retval < 0)
1849 		return retval;
1850 	bprm->exec = bprm->p;
1851 
1852 	retval = copy_strings(bprm->envc, envp, bprm);
1853 	if (retval < 0)
1854 		return retval;
1855 
1856 	retval = copy_strings(bprm->argc, argv, bprm);
1857 	if (retval < 0)
1858 		return retval;
1859 
1860 	/*
1861 	 * When argv is empty, add an empty string ("") as argv[0] to
1862 	 * ensure confused userspace programs that start processing
1863 	 * from argv[1] won't end up walking envp. See also
1864 	 * bprm_stack_limits().
1865 	 */
1866 	if (bprm->argc == 0) {
1867 		retval = copy_string_kernel("", bprm);
1868 		if (retval < 0)
1869 			return retval;
1870 		bprm->argc = 1;
1871 
1872 		pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1873 			     current->comm, bprm->filename);
1874 	}
1875 
1876 	return bprm_execve(bprm);
1877 }
1878 
1879 int kernel_execve(const char *kernel_filename,
1880 		  const char *const *argv, const char *const *envp)
1881 {
1882 	int retval;
1883 
1884 	/* It is non-sense for kernel threads to call execve */
1885 	if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1886 		return -EINVAL;
1887 
1888 	CLASS(filename_kernel, filename)(kernel_filename);
1889 	CLASS(bprm, bprm)(AT_FDCWD, filename, 0);
1890 	if (IS_ERR(bprm))
1891 		return PTR_ERR(bprm);
1892 
1893 	retval = count_strings_kernel(argv);
1894 	if (WARN_ON_ONCE(retval == 0))
1895 		return -EINVAL;
1896 	if (retval < 0)
1897 		return retval;
1898 	bprm->argc = retval;
1899 
1900 	retval = count_strings_kernel(envp);
1901 	if (retval < 0)
1902 		return retval;
1903 	bprm->envc = retval;
1904 
1905 	retval = bprm_stack_limits(bprm);
1906 	if (retval < 0)
1907 		return retval;
1908 
1909 	retval = copy_string_kernel(bprm->filename, bprm);
1910 	if (retval < 0)
1911 		return retval;
1912 	bprm->exec = bprm->p;
1913 
1914 	retval = copy_strings_kernel(bprm->envc, envp, bprm);
1915 	if (retval < 0)
1916 		return retval;
1917 
1918 	retval = copy_strings_kernel(bprm->argc, argv, bprm);
1919 	if (retval < 0)
1920 		return retval;
1921 
1922 	return bprm_execve(bprm);
1923 }
1924 
1925 void set_binfmt(struct linux_binfmt *new)
1926 {
1927 	struct mm_struct *mm = current->mm;
1928 
1929 	if (mm->binfmt)
1930 		module_put(mm->binfmt->module);
1931 
1932 	mm->binfmt = new;
1933 	if (new)
1934 		__module_get(new->module);
1935 }
1936 EXPORT_SYMBOL(set_binfmt);
1937 
1938 static inline struct user_arg_ptr native_arg(const char __user *const __user *p)
1939 {
1940 	return (struct user_arg_ptr){.ptr.native = p};
1941 }
1942 
1943 SYSCALL_DEFINE3(execve,
1944 		const char __user *, filename,
1945 		const char __user *const __user *, argv,
1946 		const char __user *const __user *, envp)
1947 {
1948 	CLASS(filename, name)(filename);
1949 	return do_execveat_common(AT_FDCWD, name,
1950 				  native_arg(argv), native_arg(envp), 0);
1951 }
1952 
1953 SYSCALL_DEFINE5(execveat,
1954 		int, fd, const char __user *, filename,
1955 		const char __user *const __user *, argv,
1956 		const char __user *const __user *, envp,
1957 		int, flags)
1958 {
1959 	CLASS(filename_uflags, name)(filename, flags);
1960 	return do_execveat_common(fd, name,
1961 				  native_arg(argv), native_arg(envp), flags);
1962 }
1963 
1964 #ifdef CONFIG_COMPAT
1965 
1966 static inline struct user_arg_ptr compat_arg(const compat_uptr_t __user *p)
1967 {
1968 	return (struct user_arg_ptr){.is_compat = true, .ptr.compat = p};
1969 }
1970 
1971 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1972 	const compat_uptr_t __user *, argv,
1973 	const compat_uptr_t __user *, envp)
1974 {
1975 	CLASS(filename, name)(filename);
1976 	return do_execveat_common(AT_FDCWD, name,
1977 				  compat_arg(argv), compat_arg(envp), 0);
1978 }
1979 
1980 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1981 		       const char __user *, filename,
1982 		       const compat_uptr_t __user *, argv,
1983 		       const compat_uptr_t __user *, envp,
1984 		       int,  flags)
1985 {
1986 	CLASS(filename_uflags, name)(filename, flags);
1987 	return do_execveat_common(fd, name,
1988 				  compat_arg(argv), compat_arg(envp), flags);
1989 }
1990 #endif
1991 
1992 #ifdef CONFIG_SYSCTL
1993 
1994 static int proc_dointvec_minmax_coredump(const struct ctl_table *table, int write,
1995 		void *buffer, size_t *lenp, loff_t *ppos)
1996 {
1997 	int error, old = READ_ONCE(suid_dumpable);
1998 
1999 	error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2000 
2001 	if (!error && write && (old != READ_ONCE(suid_dumpable)))
2002 		validate_coredump_safety();
2003 	return error;
2004 }
2005 
2006 static const struct ctl_table fs_exec_sysctls[] = {
2007 	{
2008 		.procname	= "suid_dumpable",
2009 		.data		= &suid_dumpable,
2010 		.maxlen		= sizeof(int),
2011 		.mode		= 0644,
2012 		.proc_handler	= proc_dointvec_minmax_coredump,
2013 		.extra1		= SYSCTL_ZERO,
2014 		.extra2		= SYSCTL_TWO,
2015 	},
2016 };
2017 
2018 static int __init init_fs_exec_sysctls(void)
2019 {
2020 	register_sysctl_init("fs", fs_exec_sysctls);
2021 	return 0;
2022 }
2023 
2024 fs_initcall(init_fs_exec_sysctls);
2025 #endif /* CONFIG_SYSCTL */
2026 
2027 #ifdef CONFIG_EXEC_KUNIT_TEST
2028 #include "tests/exec_kunit.c"
2029 #endif
2030