xref: /openbmc/linux/fs/exec.c (revision a6978d1b7bb8f3a25305e8ff7d367f7289614c5d)
1  // SPDX-License-Identifier: GPL-2.0-only
2  /*
3   *  linux/fs/exec.c
4   *
5   *  Copyright (C) 1991, 1992  Linus Torvalds
6   */
7  
8  /*
9   * #!-checking implemented by tytso.
10   */
11  /*
12   * Demand-loading implemented 01.12.91 - no need to read anything but
13   * the header into memory. The inode of the executable is put into
14   * "current->executable", and page faults do the actual loading. Clean.
15   *
16   * Once more I can proudly say that linux stood up to being changed: it
17   * was less than 2 hours work to get demand-loading completely implemented.
18   *
19   * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
20   * current->executable is only used by the procfs.  This allows a dispatch
21   * table to check for several different types  of binary formats.  We keep
22   * trying until we recognize the file or we run out of supported binary
23   * formats.
24   */
25  
26  #include <linux/kernel_read_file.h>
27  #include <linux/slab.h>
28  #include <linux/file.h>
29  #include <linux/fdtable.h>
30  #include <linux/mm.h>
31  #include <linux/stat.h>
32  #include <linux/fcntl.h>
33  #include <linux/swap.h>
34  #include <linux/string.h>
35  #include <linux/init.h>
36  #include <linux/sched/mm.h>
37  #include <linux/sched/coredump.h>
38  #include <linux/sched/signal.h>
39  #include <linux/sched/numa_balancing.h>
40  #include <linux/sched/task.h>
41  #include <linux/pagemap.h>
42  #include <linux/perf_event.h>
43  #include <linux/highmem.h>
44  #include <linux/spinlock.h>
45  #include <linux/key.h>
46  #include <linux/personality.h>
47  #include <linux/binfmts.h>
48  #include <linux/utsname.h>
49  #include <linux/pid_namespace.h>
50  #include <linux/module.h>
51  #include <linux/namei.h>
52  #include <linux/mount.h>
53  #include <linux/security.h>
54  #include <linux/syscalls.h>
55  #include <linux/tsacct_kern.h>
56  #include <linux/cn_proc.h>
57  #include <linux/audit.h>
58  #include <linux/kmod.h>
59  #include <linux/fsnotify.h>
60  #include <linux/fs_struct.h>
61  #include <linux/oom.h>
62  #include <linux/compat.h>
63  #include <linux/vmalloc.h>
64  #include <linux/io_uring.h>
65  #include <linux/syscall_user_dispatch.h>
66  #include <linux/coredump.h>
67  #include <linux/time_namespace.h>
68  #include <linux/user_events.h>
69  
70  #include <linux/uaccess.h>
71  #include <asm/mmu_context.h>
72  #include <asm/tlb.h>
73  
74  #include <trace/events/task.h>
75  #include "internal.h"
76  
77  #include <trace/events/sched.h>
78  
79  static int bprm_creds_from_file(struct linux_binprm *bprm);
80  
81  int suid_dumpable = 0;
82  
83  static LIST_HEAD(formats);
84  static DEFINE_RWLOCK(binfmt_lock);
85  
86  void __register_binfmt(struct linux_binfmt * fmt, int insert)
87  {
88  	write_lock(&binfmt_lock);
89  	insert ? list_add(&fmt->lh, &formats) :
90  		 list_add_tail(&fmt->lh, &formats);
91  	write_unlock(&binfmt_lock);
92  }
93  
94  EXPORT_SYMBOL(__register_binfmt);
95  
96  void unregister_binfmt(struct linux_binfmt * fmt)
97  {
98  	write_lock(&binfmt_lock);
99  	list_del(&fmt->lh);
100  	write_unlock(&binfmt_lock);
101  }
102  
103  EXPORT_SYMBOL(unregister_binfmt);
104  
105  static inline void put_binfmt(struct linux_binfmt * fmt)
106  {
107  	module_put(fmt->module);
108  }
109  
110  bool path_noexec(const struct path *path)
111  {
112  	return (path->mnt->mnt_flags & MNT_NOEXEC) ||
113  	       (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
114  }
115  
116  #ifdef CONFIG_USELIB
117  /*
118   * Note that a shared library must be both readable and executable due to
119   * security reasons.
120   *
121   * Also note that we take the address to load from the file itself.
122   */
123  SYSCALL_DEFINE1(uselib, const char __user *, library)
124  {
125  	struct linux_binfmt *fmt;
126  	struct file *file;
127  	struct filename *tmp = getname(library);
128  	int error = PTR_ERR(tmp);
129  	static const struct open_flags uselib_flags = {
130  		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
131  		.acc_mode = MAY_READ | MAY_EXEC,
132  		.intent = LOOKUP_OPEN,
133  		.lookup_flags = LOOKUP_FOLLOW,
134  	};
135  
136  	if (IS_ERR(tmp))
137  		goto out;
138  
139  	file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
140  	putname(tmp);
141  	error = PTR_ERR(file);
142  	if (IS_ERR(file))
143  		goto out;
144  
145  	/*
146  	 * may_open() has already checked for this, so it should be
147  	 * impossible to trip now. But we need to be extra cautious
148  	 * and check again at the very end too.
149  	 */
150  	error = -EACCES;
151  	if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
152  			 path_noexec(&file->f_path)))
153  		goto exit;
154  
155  	error = -ENOEXEC;
156  
157  	read_lock(&binfmt_lock);
158  	list_for_each_entry(fmt, &formats, lh) {
159  		if (!fmt->load_shlib)
160  			continue;
161  		if (!try_module_get(fmt->module))
162  			continue;
163  		read_unlock(&binfmt_lock);
164  		error = fmt->load_shlib(file);
165  		read_lock(&binfmt_lock);
166  		put_binfmt(fmt);
167  		if (error != -ENOEXEC)
168  			break;
169  	}
170  	read_unlock(&binfmt_lock);
171  exit:
172  	fput(file);
173  out:
174  	return error;
175  }
176  #endif /* #ifdef CONFIG_USELIB */
177  
178  #ifdef CONFIG_MMU
179  /*
180   * The nascent bprm->mm is not visible until exec_mmap() but it can
181   * use a lot of memory, account these pages in current->mm temporary
182   * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
183   * change the counter back via acct_arg_size(0).
184   */
185  static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
186  {
187  	struct mm_struct *mm = current->mm;
188  	long diff = (long)(pages - bprm->vma_pages);
189  
190  	if (!mm || !diff)
191  		return;
192  
193  	bprm->vma_pages = pages;
194  	add_mm_counter(mm, MM_ANONPAGES, diff);
195  }
196  
197  static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
198  		int write)
199  {
200  	struct page *page;
201  	struct vm_area_struct *vma = bprm->vma;
202  	struct mm_struct *mm = bprm->mm;
203  	int ret;
204  
205  	/*
206  	 * Avoid relying on expanding the stack down in GUP (which
207  	 * does not work for STACK_GROWSUP anyway), and just do it
208  	 * by hand ahead of time.
209  	 */
210  	if (write && pos < vma->vm_start) {
211  		mmap_write_lock(mm);
212  		ret = expand_downwards(vma, pos);
213  		if (unlikely(ret < 0)) {
214  			mmap_write_unlock(mm);
215  			return NULL;
216  		}
217  		mmap_write_downgrade(mm);
218  	} else
219  		mmap_read_lock(mm);
220  
221  	/*
222  	 * We are doing an exec().  'current' is the process
223  	 * doing the exec and 'mm' is the new process's mm.
224  	 */
225  	ret = get_user_pages_remote(mm, pos, 1,
226  			write ? FOLL_WRITE : 0,
227  			&page, NULL);
228  	mmap_read_unlock(mm);
229  	if (ret <= 0)
230  		return NULL;
231  
232  	if (write)
233  		acct_arg_size(bprm, vma_pages(vma));
234  
235  	return page;
236  }
237  
238  static void put_arg_page(struct page *page)
239  {
240  	put_page(page);
241  }
242  
243  static void free_arg_pages(struct linux_binprm *bprm)
244  {
245  }
246  
247  static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
248  		struct page *page)
249  {
250  	flush_cache_page(bprm->vma, pos, page_to_pfn(page));
251  }
252  
253  static int __bprm_mm_init(struct linux_binprm *bprm)
254  {
255  	int err;
256  	struct vm_area_struct *vma = NULL;
257  	struct mm_struct *mm = bprm->mm;
258  
259  	bprm->vma = vma = vm_area_alloc(mm);
260  	if (!vma)
261  		return -ENOMEM;
262  	vma_set_anonymous(vma);
263  
264  	if (mmap_write_lock_killable(mm)) {
265  		err = -EINTR;
266  		goto err_free;
267  	}
268  
269  	/*
270  	 * Place the stack at the largest stack address the architecture
271  	 * supports. Later, we'll move this to an appropriate place. We don't
272  	 * use STACK_TOP because that can depend on attributes which aren't
273  	 * configured yet.
274  	 */
275  	BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
276  	vma->vm_end = STACK_TOP_MAX;
277  	vma->vm_start = vma->vm_end - PAGE_SIZE;
278  	vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);
279  	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
280  
281  	err = insert_vm_struct(mm, vma);
282  	if (err)
283  		goto err;
284  
285  	mm->stack_vm = mm->total_vm = 1;
286  	mmap_write_unlock(mm);
287  	bprm->p = vma->vm_end - sizeof(void *);
288  	return 0;
289  err:
290  	mmap_write_unlock(mm);
291  err_free:
292  	bprm->vma = NULL;
293  	vm_area_free(vma);
294  	return err;
295  }
296  
297  static bool valid_arg_len(struct linux_binprm *bprm, long len)
298  {
299  	return len <= MAX_ARG_STRLEN;
300  }
301  
302  #else
303  
304  static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
305  {
306  }
307  
308  static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
309  		int write)
310  {
311  	struct page *page;
312  
313  	page = bprm->page[pos / PAGE_SIZE];
314  	if (!page && write) {
315  		page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
316  		if (!page)
317  			return NULL;
318  		bprm->page[pos / PAGE_SIZE] = page;
319  	}
320  
321  	return page;
322  }
323  
324  static void put_arg_page(struct page *page)
325  {
326  }
327  
328  static void free_arg_page(struct linux_binprm *bprm, int i)
329  {
330  	if (bprm->page[i]) {
331  		__free_page(bprm->page[i]);
332  		bprm->page[i] = NULL;
333  	}
334  }
335  
336  static void free_arg_pages(struct linux_binprm *bprm)
337  {
338  	int i;
339  
340  	for (i = 0; i < MAX_ARG_PAGES; i++)
341  		free_arg_page(bprm, i);
342  }
343  
344  static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
345  		struct page *page)
346  {
347  }
348  
349  static int __bprm_mm_init(struct linux_binprm *bprm)
350  {
351  	bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
352  	return 0;
353  }
354  
355  static bool valid_arg_len(struct linux_binprm *bprm, long len)
356  {
357  	return len <= bprm->p;
358  }
359  
360  #endif /* CONFIG_MMU */
361  
362  /*
363   * Create a new mm_struct and populate it with a temporary stack
364   * vm_area_struct.  We don't have enough context at this point to set the stack
365   * flags, permissions, and offset, so we use temporary values.  We'll update
366   * them later in setup_arg_pages().
367   */
368  static int bprm_mm_init(struct linux_binprm *bprm)
369  {
370  	int err;
371  	struct mm_struct *mm = NULL;
372  
373  	bprm->mm = mm = mm_alloc();
374  	err = -ENOMEM;
375  	if (!mm)
376  		goto err;
377  
378  	/* Save current stack limit for all calculations made during exec. */
379  	task_lock(current->group_leader);
380  	bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
381  	task_unlock(current->group_leader);
382  
383  	err = __bprm_mm_init(bprm);
384  	if (err)
385  		goto err;
386  
387  	return 0;
388  
389  err:
390  	if (mm) {
391  		bprm->mm = NULL;
392  		mmdrop(mm);
393  	}
394  
395  	return err;
396  }
397  
398  struct user_arg_ptr {
399  #ifdef CONFIG_COMPAT
400  	bool is_compat;
401  #endif
402  	union {
403  		const char __user *const __user *native;
404  #ifdef CONFIG_COMPAT
405  		const compat_uptr_t __user *compat;
406  #endif
407  	} ptr;
408  };
409  
410  static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
411  {
412  	const char __user *native;
413  
414  #ifdef CONFIG_COMPAT
415  	if (unlikely(argv.is_compat)) {
416  		compat_uptr_t compat;
417  
418  		if (get_user(compat, argv.ptr.compat + nr))
419  			return ERR_PTR(-EFAULT);
420  
421  		return compat_ptr(compat);
422  	}
423  #endif
424  
425  	if (get_user(native, argv.ptr.native + nr))
426  		return ERR_PTR(-EFAULT);
427  
428  	return native;
429  }
430  
431  /*
432   * count() counts the number of strings in array ARGV.
433   */
434  static int count(struct user_arg_ptr argv, int max)
435  {
436  	int i = 0;
437  
438  	if (argv.ptr.native != NULL) {
439  		for (;;) {
440  			const char __user *p = get_user_arg_ptr(argv, i);
441  
442  			if (!p)
443  				break;
444  
445  			if (IS_ERR(p))
446  				return -EFAULT;
447  
448  			if (i >= max)
449  				return -E2BIG;
450  			++i;
451  
452  			if (fatal_signal_pending(current))
453  				return -ERESTARTNOHAND;
454  			cond_resched();
455  		}
456  	}
457  	return i;
458  }
459  
460  static int count_strings_kernel(const char *const *argv)
461  {
462  	int i;
463  
464  	if (!argv)
465  		return 0;
466  
467  	for (i = 0; argv[i]; ++i) {
468  		if (i >= MAX_ARG_STRINGS)
469  			return -E2BIG;
470  		if (fatal_signal_pending(current))
471  			return -ERESTARTNOHAND;
472  		cond_resched();
473  	}
474  	return i;
475  }
476  
477  static int bprm_stack_limits(struct linux_binprm *bprm)
478  {
479  	unsigned long limit, ptr_size;
480  
481  	/*
482  	 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
483  	 * (whichever is smaller) for the argv+env strings.
484  	 * This ensures that:
485  	 *  - the remaining binfmt code will not run out of stack space,
486  	 *  - the program will have a reasonable amount of stack left
487  	 *    to work from.
488  	 */
489  	limit = _STK_LIM / 4 * 3;
490  	limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
491  	/*
492  	 * We've historically supported up to 32 pages (ARG_MAX)
493  	 * of argument strings even with small stacks
494  	 */
495  	limit = max_t(unsigned long, limit, ARG_MAX);
496  	/*
497  	 * We must account for the size of all the argv and envp pointers to
498  	 * the argv and envp strings, since they will also take up space in
499  	 * the stack. They aren't stored until much later when we can't
500  	 * signal to the parent that the child has run out of stack space.
501  	 * Instead, calculate it here so it's possible to fail gracefully.
502  	 *
503  	 * In the case of argc = 0, make sure there is space for adding a
504  	 * empty string (which will bump argc to 1), to ensure confused
505  	 * userspace programs don't start processing from argv[1], thinking
506  	 * argc can never be 0, to keep them from walking envp by accident.
507  	 * See do_execveat_common().
508  	 */
509  	ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
510  	if (limit <= ptr_size)
511  		return -E2BIG;
512  	limit -= ptr_size;
513  
514  	bprm->argmin = bprm->p - limit;
515  	return 0;
516  }
517  
518  /*
519   * 'copy_strings()' copies argument/environment strings from the old
520   * processes's memory to the new process's stack.  The call to get_user_pages()
521   * ensures the destination page is created and not swapped out.
522   */
523  static int copy_strings(int argc, struct user_arg_ptr argv,
524  			struct linux_binprm *bprm)
525  {
526  	struct page *kmapped_page = NULL;
527  	char *kaddr = NULL;
528  	unsigned long kpos = 0;
529  	int ret;
530  
531  	while (argc-- > 0) {
532  		const char __user *str;
533  		int len;
534  		unsigned long pos;
535  
536  		ret = -EFAULT;
537  		str = get_user_arg_ptr(argv, argc);
538  		if (IS_ERR(str))
539  			goto out;
540  
541  		len = strnlen_user(str, MAX_ARG_STRLEN);
542  		if (!len)
543  			goto out;
544  
545  		ret = -E2BIG;
546  		if (!valid_arg_len(bprm, len))
547  			goto out;
548  
549  		/* We're going to work our way backwards. */
550  		pos = bprm->p;
551  		str += len;
552  		bprm->p -= len;
553  #ifdef CONFIG_MMU
554  		if (bprm->p < bprm->argmin)
555  			goto out;
556  #endif
557  
558  		while (len > 0) {
559  			int offset, bytes_to_copy;
560  
561  			if (fatal_signal_pending(current)) {
562  				ret = -ERESTARTNOHAND;
563  				goto out;
564  			}
565  			cond_resched();
566  
567  			offset = pos % PAGE_SIZE;
568  			if (offset == 0)
569  				offset = PAGE_SIZE;
570  
571  			bytes_to_copy = offset;
572  			if (bytes_to_copy > len)
573  				bytes_to_copy = len;
574  
575  			offset -= bytes_to_copy;
576  			pos -= bytes_to_copy;
577  			str -= bytes_to_copy;
578  			len -= bytes_to_copy;
579  
580  			if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
581  				struct page *page;
582  
583  				page = get_arg_page(bprm, pos, 1);
584  				if (!page) {
585  					ret = -E2BIG;
586  					goto out;
587  				}
588  
589  				if (kmapped_page) {
590  					flush_dcache_page(kmapped_page);
591  					kunmap_local(kaddr);
592  					put_arg_page(kmapped_page);
593  				}
594  				kmapped_page = page;
595  				kaddr = kmap_local_page(kmapped_page);
596  				kpos = pos & PAGE_MASK;
597  				flush_arg_page(bprm, kpos, kmapped_page);
598  			}
599  			if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
600  				ret = -EFAULT;
601  				goto out;
602  			}
603  		}
604  	}
605  	ret = 0;
606  out:
607  	if (kmapped_page) {
608  		flush_dcache_page(kmapped_page);
609  		kunmap_local(kaddr);
610  		put_arg_page(kmapped_page);
611  	}
612  	return ret;
613  }
614  
615  /*
616   * Copy and argument/environment string from the kernel to the processes stack.
617   */
618  int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
619  {
620  	int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
621  	unsigned long pos = bprm->p;
622  
623  	if (len == 0)
624  		return -EFAULT;
625  	if (!valid_arg_len(bprm, len))
626  		return -E2BIG;
627  
628  	/* We're going to work our way backwards. */
629  	arg += len;
630  	bprm->p -= len;
631  	if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
632  		return -E2BIG;
633  
634  	while (len > 0) {
635  		unsigned int bytes_to_copy = min_t(unsigned int, len,
636  				min_not_zero(offset_in_page(pos), PAGE_SIZE));
637  		struct page *page;
638  
639  		pos -= bytes_to_copy;
640  		arg -= bytes_to_copy;
641  		len -= bytes_to_copy;
642  
643  		page = get_arg_page(bprm, pos, 1);
644  		if (!page)
645  			return -E2BIG;
646  		flush_arg_page(bprm, pos & PAGE_MASK, page);
647  		memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
648  		put_arg_page(page);
649  	}
650  
651  	return 0;
652  }
653  EXPORT_SYMBOL(copy_string_kernel);
654  
655  static int copy_strings_kernel(int argc, const char *const *argv,
656  			       struct linux_binprm *bprm)
657  {
658  	while (argc-- > 0) {
659  		int ret = copy_string_kernel(argv[argc], bprm);
660  		if (ret < 0)
661  			return ret;
662  		if (fatal_signal_pending(current))
663  			return -ERESTARTNOHAND;
664  		cond_resched();
665  	}
666  	return 0;
667  }
668  
669  #ifdef CONFIG_MMU
670  
671  /*
672   * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
673   * the binfmt code determines where the new stack should reside, we shift it to
674   * its final location.  The process proceeds as follows:
675   *
676   * 1) Use shift to calculate the new vma endpoints.
677   * 2) Extend vma to cover both the old and new ranges.  This ensures the
678   *    arguments passed to subsequent functions are consistent.
679   * 3) Move vma's page tables to the new range.
680   * 4) Free up any cleared pgd range.
681   * 5) Shrink the vma to cover only the new range.
682   */
683  static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
684  {
685  	struct mm_struct *mm = vma->vm_mm;
686  	unsigned long old_start = vma->vm_start;
687  	unsigned long old_end = vma->vm_end;
688  	unsigned long length = old_end - old_start;
689  	unsigned long new_start = old_start - shift;
690  	unsigned long new_end = old_end - shift;
691  	VMA_ITERATOR(vmi, mm, new_start);
692  	struct vm_area_struct *next;
693  	struct mmu_gather tlb;
694  
695  	BUG_ON(new_start > new_end);
696  
697  	/*
698  	 * ensure there are no vmas between where we want to go
699  	 * and where we are
700  	 */
701  	if (vma != vma_next(&vmi))
702  		return -EFAULT;
703  
704  	vma_iter_prev_range(&vmi);
705  	/*
706  	 * cover the whole range: [new_start, old_end)
707  	 */
708  	if (vma_expand(&vmi, vma, new_start, old_end, vma->vm_pgoff, NULL))
709  		return -ENOMEM;
710  
711  	/*
712  	 * move the page tables downwards, on failure we rely on
713  	 * process cleanup to remove whatever mess we made.
714  	 */
715  	if (length != move_page_tables(vma, old_start,
716  				       vma, new_start, length, false))
717  		return -ENOMEM;
718  
719  	lru_add_drain();
720  	tlb_gather_mmu(&tlb, mm);
721  	next = vma_next(&vmi);
722  	if (new_end > old_start) {
723  		/*
724  		 * when the old and new regions overlap clear from new_end.
725  		 */
726  		free_pgd_range(&tlb, new_end, old_end, new_end,
727  			next ? next->vm_start : USER_PGTABLES_CEILING);
728  	} else {
729  		/*
730  		 * otherwise, clean from old_start; this is done to not touch
731  		 * the address space in [new_end, old_start) some architectures
732  		 * have constraints on va-space that make this illegal (IA64) -
733  		 * for the others its just a little faster.
734  		 */
735  		free_pgd_range(&tlb, old_start, old_end, new_end,
736  			next ? next->vm_start : USER_PGTABLES_CEILING);
737  	}
738  	tlb_finish_mmu(&tlb);
739  
740  	vma_prev(&vmi);
741  	/* Shrink the vma to just the new range */
742  	return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff);
743  }
744  
745  /*
746   * Finalizes the stack vm_area_struct. The flags and permissions are updated,
747   * the stack is optionally relocated, and some extra space is added.
748   */
749  int setup_arg_pages(struct linux_binprm *bprm,
750  		    unsigned long stack_top,
751  		    int executable_stack)
752  {
753  	unsigned long ret;
754  	unsigned long stack_shift;
755  	struct mm_struct *mm = current->mm;
756  	struct vm_area_struct *vma = bprm->vma;
757  	struct vm_area_struct *prev = NULL;
758  	unsigned long vm_flags;
759  	unsigned long stack_base;
760  	unsigned long stack_size;
761  	unsigned long stack_expand;
762  	unsigned long rlim_stack;
763  	struct mmu_gather tlb;
764  	struct vma_iterator vmi;
765  
766  #ifdef CONFIG_STACK_GROWSUP
767  	/* Limit stack size */
768  	stack_base = bprm->rlim_stack.rlim_max;
769  
770  	stack_base = calc_max_stack_size(stack_base);
771  
772  	/* Add space for stack randomization. */
773  	stack_base += (STACK_RND_MASK << PAGE_SHIFT);
774  
775  	/* Make sure we didn't let the argument array grow too large. */
776  	if (vma->vm_end - vma->vm_start > stack_base)
777  		return -ENOMEM;
778  
779  	stack_base = PAGE_ALIGN(stack_top - stack_base);
780  
781  	stack_shift = vma->vm_start - stack_base;
782  	mm->arg_start = bprm->p - stack_shift;
783  	bprm->p = vma->vm_end - stack_shift;
784  #else
785  	stack_top = arch_align_stack(stack_top);
786  	stack_top = PAGE_ALIGN(stack_top);
787  
788  	if (unlikely(stack_top < mmap_min_addr) ||
789  	    unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
790  		return -ENOMEM;
791  
792  	stack_shift = vma->vm_end - stack_top;
793  
794  	bprm->p -= stack_shift;
795  	mm->arg_start = bprm->p;
796  #endif
797  
798  	if (bprm->loader)
799  		bprm->loader -= stack_shift;
800  	bprm->exec -= stack_shift;
801  
802  	if (mmap_write_lock_killable(mm))
803  		return -EINTR;
804  
805  	vm_flags = VM_STACK_FLAGS;
806  
807  	/*
808  	 * Adjust stack execute permissions; explicitly enable for
809  	 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
810  	 * (arch default) otherwise.
811  	 */
812  	if (unlikely(executable_stack == EXSTACK_ENABLE_X))
813  		vm_flags |= VM_EXEC;
814  	else if (executable_stack == EXSTACK_DISABLE_X)
815  		vm_flags &= ~VM_EXEC;
816  	vm_flags |= mm->def_flags;
817  	vm_flags |= VM_STACK_INCOMPLETE_SETUP;
818  
819  	vma_iter_init(&vmi, mm, vma->vm_start);
820  
821  	tlb_gather_mmu(&tlb, mm);
822  	ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
823  			vm_flags);
824  	tlb_finish_mmu(&tlb);
825  
826  	if (ret)
827  		goto out_unlock;
828  	BUG_ON(prev != vma);
829  
830  	if (unlikely(vm_flags & VM_EXEC)) {
831  		pr_warn_once("process '%pD4' started with executable stack\n",
832  			     bprm->file);
833  	}
834  
835  	/* Move stack pages down in memory. */
836  	if (stack_shift) {
837  		ret = shift_arg_pages(vma, stack_shift);
838  		if (ret)
839  			goto out_unlock;
840  	}
841  
842  	/* mprotect_fixup is overkill to remove the temporary stack flags */
843  	vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
844  
845  	stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
846  	stack_size = vma->vm_end - vma->vm_start;
847  	/*
848  	 * Align this down to a page boundary as expand_stack
849  	 * will align it up.
850  	 */
851  	rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
852  
853  	stack_expand = min(rlim_stack, stack_size + stack_expand);
854  
855  #ifdef CONFIG_STACK_GROWSUP
856  	stack_base = vma->vm_start + stack_expand;
857  #else
858  	stack_base = vma->vm_end - stack_expand;
859  #endif
860  	current->mm->start_stack = bprm->p;
861  	ret = expand_stack_locked(vma, stack_base);
862  	if (ret)
863  		ret = -EFAULT;
864  
865  out_unlock:
866  	mmap_write_unlock(mm);
867  	return ret;
868  }
869  EXPORT_SYMBOL(setup_arg_pages);
870  
871  #else
872  
873  /*
874   * Transfer the program arguments and environment from the holding pages
875   * onto the stack. The provided stack pointer is adjusted accordingly.
876   */
877  int transfer_args_to_stack(struct linux_binprm *bprm,
878  			   unsigned long *sp_location)
879  {
880  	unsigned long index, stop, sp;
881  	int ret = 0;
882  
883  	stop = bprm->p >> PAGE_SHIFT;
884  	sp = *sp_location;
885  
886  	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
887  		unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
888  		char *src = kmap_local_page(bprm->page[index]) + offset;
889  		sp -= PAGE_SIZE - offset;
890  		if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
891  			ret = -EFAULT;
892  		kunmap_local(src);
893  		if (ret)
894  			goto out;
895  	}
896  
897  	bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
898  	*sp_location = sp;
899  
900  out:
901  	return ret;
902  }
903  EXPORT_SYMBOL(transfer_args_to_stack);
904  
905  #endif /* CONFIG_MMU */
906  
907  static struct file *do_open_execat(int fd, struct filename *name, int flags)
908  {
909  	struct file *file;
910  	int err;
911  	struct open_flags open_exec_flags = {
912  		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
913  		.acc_mode = MAY_EXEC,
914  		.intent = LOOKUP_OPEN,
915  		.lookup_flags = LOOKUP_FOLLOW,
916  	};
917  
918  	if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
919  		return ERR_PTR(-EINVAL);
920  	if (flags & AT_SYMLINK_NOFOLLOW)
921  		open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
922  	if (flags & AT_EMPTY_PATH)
923  		open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
924  
925  	file = do_filp_open(fd, name, &open_exec_flags);
926  	if (IS_ERR(file))
927  		goto out;
928  
929  	/*
930  	 * may_open() has already checked for this, so it should be
931  	 * impossible to trip now. But we need to be extra cautious
932  	 * and check again at the very end too.
933  	 */
934  	err = -EACCES;
935  	if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
936  			 path_noexec(&file->f_path)))
937  		goto exit;
938  
939  	err = deny_write_access(file);
940  	if (err)
941  		goto exit;
942  
943  out:
944  	return file;
945  
946  exit:
947  	fput(file);
948  	return ERR_PTR(err);
949  }
950  
951  struct file *open_exec(const char *name)
952  {
953  	struct filename *filename = getname_kernel(name);
954  	struct file *f = ERR_CAST(filename);
955  
956  	if (!IS_ERR(filename)) {
957  		f = do_open_execat(AT_FDCWD, filename, 0);
958  		putname(filename);
959  	}
960  	return f;
961  }
962  EXPORT_SYMBOL(open_exec);
963  
964  #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
965  ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
966  {
967  	ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
968  	if (res > 0)
969  		flush_icache_user_range(addr, addr + len);
970  	return res;
971  }
972  EXPORT_SYMBOL(read_code);
973  #endif
974  
975  /*
976   * Maps the mm_struct mm into the current task struct.
977   * On success, this function returns with exec_update_lock
978   * held for writing.
979   */
980  static int exec_mmap(struct mm_struct *mm)
981  {
982  	struct task_struct *tsk;
983  	struct mm_struct *old_mm, *active_mm;
984  	int ret;
985  
986  	/* Notify parent that we're no longer interested in the old VM */
987  	tsk = current;
988  	old_mm = current->mm;
989  	exec_mm_release(tsk, old_mm);
990  	if (old_mm)
991  		sync_mm_rss(old_mm);
992  
993  	ret = down_write_killable(&tsk->signal->exec_update_lock);
994  	if (ret)
995  		return ret;
996  
997  	if (old_mm) {
998  		/*
999  		 * If there is a pending fatal signal perhaps a signal
1000  		 * whose default action is to create a coredump get
1001  		 * out and die instead of going through with the exec.
1002  		 */
1003  		ret = mmap_read_lock_killable(old_mm);
1004  		if (ret) {
1005  			up_write(&tsk->signal->exec_update_lock);
1006  			return ret;
1007  		}
1008  	}
1009  
1010  	task_lock(tsk);
1011  	membarrier_exec_mmap(mm);
1012  
1013  	local_irq_disable();
1014  	active_mm = tsk->active_mm;
1015  	tsk->active_mm = mm;
1016  	tsk->mm = mm;
1017  	mm_init_cid(mm);
1018  	/*
1019  	 * This prevents preemption while active_mm is being loaded and
1020  	 * it and mm are being updated, which could cause problems for
1021  	 * lazy tlb mm refcounting when these are updated by context
1022  	 * switches. Not all architectures can handle irqs off over
1023  	 * activate_mm yet.
1024  	 */
1025  	if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1026  		local_irq_enable();
1027  	activate_mm(active_mm, mm);
1028  	if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1029  		local_irq_enable();
1030  	lru_gen_add_mm(mm);
1031  	task_unlock(tsk);
1032  	lru_gen_use_mm(mm);
1033  	if (old_mm) {
1034  		mmap_read_unlock(old_mm);
1035  		BUG_ON(active_mm != old_mm);
1036  		setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1037  		mm_update_next_owner(old_mm);
1038  		mmput(old_mm);
1039  		return 0;
1040  	}
1041  	mmdrop_lazy_tlb(active_mm);
1042  	return 0;
1043  }
1044  
1045  static int de_thread(struct task_struct *tsk)
1046  {
1047  	struct signal_struct *sig = tsk->signal;
1048  	struct sighand_struct *oldsighand = tsk->sighand;
1049  	spinlock_t *lock = &oldsighand->siglock;
1050  
1051  	if (thread_group_empty(tsk))
1052  		goto no_thread_group;
1053  
1054  	/*
1055  	 * Kill all other threads in the thread group.
1056  	 */
1057  	spin_lock_irq(lock);
1058  	if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1059  		/*
1060  		 * Another group action in progress, just
1061  		 * return so that the signal is processed.
1062  		 */
1063  		spin_unlock_irq(lock);
1064  		return -EAGAIN;
1065  	}
1066  
1067  	sig->group_exec_task = tsk;
1068  	sig->notify_count = zap_other_threads(tsk);
1069  	if (!thread_group_leader(tsk))
1070  		sig->notify_count--;
1071  
1072  	while (sig->notify_count) {
1073  		__set_current_state(TASK_KILLABLE);
1074  		spin_unlock_irq(lock);
1075  		schedule();
1076  		if (__fatal_signal_pending(tsk))
1077  			goto killed;
1078  		spin_lock_irq(lock);
1079  	}
1080  	spin_unlock_irq(lock);
1081  
1082  	/*
1083  	 * At this point all other threads have exited, all we have to
1084  	 * do is to wait for the thread group leader to become inactive,
1085  	 * and to assume its PID:
1086  	 */
1087  	if (!thread_group_leader(tsk)) {
1088  		struct task_struct *leader = tsk->group_leader;
1089  
1090  		for (;;) {
1091  			cgroup_threadgroup_change_begin(tsk);
1092  			write_lock_irq(&tasklist_lock);
1093  			/*
1094  			 * Do this under tasklist_lock to ensure that
1095  			 * exit_notify() can't miss ->group_exec_task
1096  			 */
1097  			sig->notify_count = -1;
1098  			if (likely(leader->exit_state))
1099  				break;
1100  			__set_current_state(TASK_KILLABLE);
1101  			write_unlock_irq(&tasklist_lock);
1102  			cgroup_threadgroup_change_end(tsk);
1103  			schedule();
1104  			if (__fatal_signal_pending(tsk))
1105  				goto killed;
1106  		}
1107  
1108  		/*
1109  		 * The only record we have of the real-time age of a
1110  		 * process, regardless of execs it's done, is start_time.
1111  		 * All the past CPU time is accumulated in signal_struct
1112  		 * from sister threads now dead.  But in this non-leader
1113  		 * exec, nothing survives from the original leader thread,
1114  		 * whose birth marks the true age of this process now.
1115  		 * When we take on its identity by switching to its PID, we
1116  		 * also take its birthdate (always earlier than our own).
1117  		 */
1118  		tsk->start_time = leader->start_time;
1119  		tsk->start_boottime = leader->start_boottime;
1120  
1121  		BUG_ON(!same_thread_group(leader, tsk));
1122  		/*
1123  		 * An exec() starts a new thread group with the
1124  		 * TGID of the previous thread group. Rehash the
1125  		 * two threads with a switched PID, and release
1126  		 * the former thread group leader:
1127  		 */
1128  
1129  		/* Become a process group leader with the old leader's pid.
1130  		 * The old leader becomes a thread of the this thread group.
1131  		 */
1132  		exchange_tids(tsk, leader);
1133  		transfer_pid(leader, tsk, PIDTYPE_TGID);
1134  		transfer_pid(leader, tsk, PIDTYPE_PGID);
1135  		transfer_pid(leader, tsk, PIDTYPE_SID);
1136  
1137  		list_replace_rcu(&leader->tasks, &tsk->tasks);
1138  		list_replace_init(&leader->sibling, &tsk->sibling);
1139  
1140  		tsk->group_leader = tsk;
1141  		leader->group_leader = tsk;
1142  
1143  		tsk->exit_signal = SIGCHLD;
1144  		leader->exit_signal = -1;
1145  
1146  		BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1147  		leader->exit_state = EXIT_DEAD;
1148  
1149  		/*
1150  		 * We are going to release_task()->ptrace_unlink() silently,
1151  		 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1152  		 * the tracer won't block again waiting for this thread.
1153  		 */
1154  		if (unlikely(leader->ptrace))
1155  			__wake_up_parent(leader, leader->parent);
1156  		write_unlock_irq(&tasklist_lock);
1157  		cgroup_threadgroup_change_end(tsk);
1158  
1159  		release_task(leader);
1160  	}
1161  
1162  	sig->group_exec_task = NULL;
1163  	sig->notify_count = 0;
1164  
1165  no_thread_group:
1166  	/* we have changed execution domain */
1167  	tsk->exit_signal = SIGCHLD;
1168  
1169  	BUG_ON(!thread_group_leader(tsk));
1170  	return 0;
1171  
1172  killed:
1173  	/* protects against exit_notify() and __exit_signal() */
1174  	read_lock(&tasklist_lock);
1175  	sig->group_exec_task = NULL;
1176  	sig->notify_count = 0;
1177  	read_unlock(&tasklist_lock);
1178  	return -EAGAIN;
1179  }
1180  
1181  
1182  /*
1183   * This function makes sure the current process has its own signal table,
1184   * so that flush_signal_handlers can later reset the handlers without
1185   * disturbing other processes.  (Other processes might share the signal
1186   * table via the CLONE_SIGHAND option to clone().)
1187   */
1188  static int unshare_sighand(struct task_struct *me)
1189  {
1190  	struct sighand_struct *oldsighand = me->sighand;
1191  
1192  	if (refcount_read(&oldsighand->count) != 1) {
1193  		struct sighand_struct *newsighand;
1194  		/*
1195  		 * This ->sighand is shared with the CLONE_SIGHAND
1196  		 * but not CLONE_THREAD task, switch to the new one.
1197  		 */
1198  		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1199  		if (!newsighand)
1200  			return -ENOMEM;
1201  
1202  		refcount_set(&newsighand->count, 1);
1203  
1204  		write_lock_irq(&tasklist_lock);
1205  		spin_lock(&oldsighand->siglock);
1206  		memcpy(newsighand->action, oldsighand->action,
1207  		       sizeof(newsighand->action));
1208  		rcu_assign_pointer(me->sighand, newsighand);
1209  		spin_unlock(&oldsighand->siglock);
1210  		write_unlock_irq(&tasklist_lock);
1211  
1212  		__cleanup_sighand(oldsighand);
1213  	}
1214  	return 0;
1215  }
1216  
1217  char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1218  {
1219  	task_lock(tsk);
1220  	/* Always NUL terminated and zero-padded */
1221  	strscpy_pad(buf, tsk->comm, buf_size);
1222  	task_unlock(tsk);
1223  	return buf;
1224  }
1225  EXPORT_SYMBOL_GPL(__get_task_comm);
1226  
1227  /*
1228   * These functions flushes out all traces of the currently running executable
1229   * so that a new one can be started
1230   */
1231  
1232  void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1233  {
1234  	task_lock(tsk);
1235  	trace_task_rename(tsk, buf);
1236  	strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1237  	task_unlock(tsk);
1238  	perf_event_comm(tsk, exec);
1239  }
1240  
1241  /*
1242   * Calling this is the point of no return. None of the failures will be
1243   * seen by userspace since either the process is already taking a fatal
1244   * signal (via de_thread() or coredump), or will have SEGV raised
1245   * (after exec_mmap()) by search_binary_handler (see below).
1246   */
1247  int begin_new_exec(struct linux_binprm * bprm)
1248  {
1249  	struct task_struct *me = current;
1250  	int retval;
1251  
1252  	/* Once we are committed compute the creds */
1253  	retval = bprm_creds_from_file(bprm);
1254  	if (retval)
1255  		return retval;
1256  
1257  	/*
1258  	 * Ensure all future errors are fatal.
1259  	 */
1260  	bprm->point_of_no_return = true;
1261  
1262  	/*
1263  	 * Make this the only thread in the thread group.
1264  	 */
1265  	retval = de_thread(me);
1266  	if (retval)
1267  		goto out;
1268  
1269  	/*
1270  	 * Cancel any io_uring activity across execve
1271  	 */
1272  	io_uring_task_cancel();
1273  
1274  	/* Ensure the files table is not shared. */
1275  	retval = unshare_files();
1276  	if (retval)
1277  		goto out;
1278  
1279  	/*
1280  	 * Must be called _before_ exec_mmap() as bprm->mm is
1281  	 * not visible until then. Doing it here also ensures
1282  	 * we don't race against replace_mm_exe_file().
1283  	 */
1284  	retval = set_mm_exe_file(bprm->mm, bprm->file);
1285  	if (retval)
1286  		goto out;
1287  
1288  	/* If the binary is not readable then enforce mm->dumpable=0 */
1289  	would_dump(bprm, bprm->file);
1290  	if (bprm->have_execfd)
1291  		would_dump(bprm, bprm->executable);
1292  
1293  	/*
1294  	 * Release all of the old mmap stuff
1295  	 */
1296  	acct_arg_size(bprm, 0);
1297  	retval = exec_mmap(bprm->mm);
1298  	if (retval)
1299  		goto out;
1300  
1301  	bprm->mm = NULL;
1302  
1303  	retval = exec_task_namespaces();
1304  	if (retval)
1305  		goto out_unlock;
1306  
1307  #ifdef CONFIG_POSIX_TIMERS
1308  	spin_lock_irq(&me->sighand->siglock);
1309  	posix_cpu_timers_exit(me);
1310  	spin_unlock_irq(&me->sighand->siglock);
1311  	exit_itimers(me);
1312  	flush_itimer_signals();
1313  #endif
1314  
1315  	/*
1316  	 * Make the signal table private.
1317  	 */
1318  	retval = unshare_sighand(me);
1319  	if (retval)
1320  		goto out_unlock;
1321  
1322  	me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1323  					PF_NOFREEZE | PF_NO_SETAFFINITY);
1324  	flush_thread();
1325  	me->personality &= ~bprm->per_clear;
1326  
1327  	clear_syscall_work_syscall_user_dispatch(me);
1328  
1329  	/*
1330  	 * We have to apply CLOEXEC before we change whether the process is
1331  	 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1332  	 * trying to access the should-be-closed file descriptors of a process
1333  	 * undergoing exec(2).
1334  	 */
1335  	do_close_on_exec(me->files);
1336  
1337  	if (bprm->secureexec) {
1338  		/* Make sure parent cannot signal privileged process. */
1339  		me->pdeath_signal = 0;
1340  
1341  		/*
1342  		 * For secureexec, reset the stack limit to sane default to
1343  		 * avoid bad behavior from the prior rlimits. This has to
1344  		 * happen before arch_pick_mmap_layout(), which examines
1345  		 * RLIMIT_STACK, but after the point of no return to avoid
1346  		 * needing to clean up the change on failure.
1347  		 */
1348  		if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1349  			bprm->rlim_stack.rlim_cur = _STK_LIM;
1350  	}
1351  
1352  	me->sas_ss_sp = me->sas_ss_size = 0;
1353  
1354  	/*
1355  	 * Figure out dumpability. Note that this checking only of current
1356  	 * is wrong, but userspace depends on it. This should be testing
1357  	 * bprm->secureexec instead.
1358  	 */
1359  	if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1360  	    !(uid_eq(current_euid(), current_uid()) &&
1361  	      gid_eq(current_egid(), current_gid())))
1362  		set_dumpable(current->mm, suid_dumpable);
1363  	else
1364  		set_dumpable(current->mm, SUID_DUMP_USER);
1365  
1366  	perf_event_exec();
1367  	__set_task_comm(me, kbasename(bprm->filename), true);
1368  
1369  	/* An exec changes our domain. We are no longer part of the thread
1370  	   group */
1371  	WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1372  	flush_signal_handlers(me, 0);
1373  
1374  	retval = set_cred_ucounts(bprm->cred);
1375  	if (retval < 0)
1376  		goto out_unlock;
1377  
1378  	/*
1379  	 * install the new credentials for this executable
1380  	 */
1381  	security_bprm_committing_creds(bprm);
1382  
1383  	commit_creds(bprm->cred);
1384  	bprm->cred = NULL;
1385  
1386  	/*
1387  	 * Disable monitoring for regular users
1388  	 * when executing setuid binaries. Must
1389  	 * wait until new credentials are committed
1390  	 * by commit_creds() above
1391  	 */
1392  	if (get_dumpable(me->mm) != SUID_DUMP_USER)
1393  		perf_event_exit_task(me);
1394  	/*
1395  	 * cred_guard_mutex must be held at least to this point to prevent
1396  	 * ptrace_attach() from altering our determination of the task's
1397  	 * credentials; any time after this it may be unlocked.
1398  	 */
1399  	security_bprm_committed_creds(bprm);
1400  
1401  	/* Pass the opened binary to the interpreter. */
1402  	if (bprm->have_execfd) {
1403  		retval = get_unused_fd_flags(0);
1404  		if (retval < 0)
1405  			goto out_unlock;
1406  		fd_install(retval, bprm->executable);
1407  		bprm->executable = NULL;
1408  		bprm->execfd = retval;
1409  	}
1410  	return 0;
1411  
1412  out_unlock:
1413  	up_write(&me->signal->exec_update_lock);
1414  	if (!bprm->cred)
1415  		mutex_unlock(&me->signal->cred_guard_mutex);
1416  
1417  out:
1418  	return retval;
1419  }
1420  EXPORT_SYMBOL(begin_new_exec);
1421  
1422  void would_dump(struct linux_binprm *bprm, struct file *file)
1423  {
1424  	struct inode *inode = file_inode(file);
1425  	struct mnt_idmap *idmap = file_mnt_idmap(file);
1426  	if (inode_permission(idmap, inode, MAY_READ) < 0) {
1427  		struct user_namespace *old, *user_ns;
1428  		bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1429  
1430  		/* Ensure mm->user_ns contains the executable */
1431  		user_ns = old = bprm->mm->user_ns;
1432  		while ((user_ns != &init_user_ns) &&
1433  		       !privileged_wrt_inode_uidgid(user_ns, idmap, inode))
1434  			user_ns = user_ns->parent;
1435  
1436  		if (old != user_ns) {
1437  			bprm->mm->user_ns = get_user_ns(user_ns);
1438  			put_user_ns(old);
1439  		}
1440  	}
1441  }
1442  EXPORT_SYMBOL(would_dump);
1443  
1444  void setup_new_exec(struct linux_binprm * bprm)
1445  {
1446  	/* Setup things that can depend upon the personality */
1447  	struct task_struct *me = current;
1448  
1449  	arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1450  
1451  	arch_setup_new_exec();
1452  
1453  	/* Set the new mm task size. We have to do that late because it may
1454  	 * depend on TIF_32BIT which is only updated in flush_thread() on
1455  	 * some architectures like powerpc
1456  	 */
1457  	me->mm->task_size = TASK_SIZE;
1458  	up_write(&me->signal->exec_update_lock);
1459  	mutex_unlock(&me->signal->cred_guard_mutex);
1460  }
1461  EXPORT_SYMBOL(setup_new_exec);
1462  
1463  /* Runs immediately before start_thread() takes over. */
1464  void finalize_exec(struct linux_binprm *bprm)
1465  {
1466  	/* Store any stack rlimit changes before starting thread. */
1467  	task_lock(current->group_leader);
1468  	current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1469  	task_unlock(current->group_leader);
1470  }
1471  EXPORT_SYMBOL(finalize_exec);
1472  
1473  /*
1474   * Prepare credentials and lock ->cred_guard_mutex.
1475   * setup_new_exec() commits the new creds and drops the lock.
1476   * Or, if exec fails before, free_bprm() should release ->cred
1477   * and unlock.
1478   */
1479  static int prepare_bprm_creds(struct linux_binprm *bprm)
1480  {
1481  	if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1482  		return -ERESTARTNOINTR;
1483  
1484  	bprm->cred = prepare_exec_creds();
1485  	if (likely(bprm->cred))
1486  		return 0;
1487  
1488  	mutex_unlock(&current->signal->cred_guard_mutex);
1489  	return -ENOMEM;
1490  }
1491  
1492  static void free_bprm(struct linux_binprm *bprm)
1493  {
1494  	if (bprm->mm) {
1495  		acct_arg_size(bprm, 0);
1496  		mmput(bprm->mm);
1497  	}
1498  	free_arg_pages(bprm);
1499  	if (bprm->cred) {
1500  		mutex_unlock(&current->signal->cred_guard_mutex);
1501  		abort_creds(bprm->cred);
1502  	}
1503  	if (bprm->file) {
1504  		allow_write_access(bprm->file);
1505  		fput(bprm->file);
1506  	}
1507  	if (bprm->executable)
1508  		fput(bprm->executable);
1509  	/* If a binfmt changed the interp, free it. */
1510  	if (bprm->interp != bprm->filename)
1511  		kfree(bprm->interp);
1512  	kfree(bprm->fdpath);
1513  	kfree(bprm);
1514  }
1515  
1516  static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1517  {
1518  	struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1519  	int retval = -ENOMEM;
1520  	if (!bprm)
1521  		goto out;
1522  
1523  	if (fd == AT_FDCWD || filename->name[0] == '/') {
1524  		bprm->filename = filename->name;
1525  	} else {
1526  		if (filename->name[0] == '\0')
1527  			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1528  		else
1529  			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1530  						  fd, filename->name);
1531  		if (!bprm->fdpath)
1532  			goto out_free;
1533  
1534  		bprm->filename = bprm->fdpath;
1535  	}
1536  	bprm->interp = bprm->filename;
1537  
1538  	retval = bprm_mm_init(bprm);
1539  	if (retval)
1540  		goto out_free;
1541  	return bprm;
1542  
1543  out_free:
1544  	free_bprm(bprm);
1545  out:
1546  	return ERR_PTR(retval);
1547  }
1548  
1549  int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1550  {
1551  	/* If a binfmt changed the interp, free it first. */
1552  	if (bprm->interp != bprm->filename)
1553  		kfree(bprm->interp);
1554  	bprm->interp = kstrdup(interp, GFP_KERNEL);
1555  	if (!bprm->interp)
1556  		return -ENOMEM;
1557  	return 0;
1558  }
1559  EXPORT_SYMBOL(bprm_change_interp);
1560  
1561  /*
1562   * determine how safe it is to execute the proposed program
1563   * - the caller must hold ->cred_guard_mutex to protect against
1564   *   PTRACE_ATTACH or seccomp thread-sync
1565   */
1566  static void check_unsafe_exec(struct linux_binprm *bprm)
1567  {
1568  	struct task_struct *p = current, *t;
1569  	unsigned n_fs;
1570  
1571  	if (p->ptrace)
1572  		bprm->unsafe |= LSM_UNSAFE_PTRACE;
1573  
1574  	/*
1575  	 * This isn't strictly necessary, but it makes it harder for LSMs to
1576  	 * mess up.
1577  	 */
1578  	if (task_no_new_privs(current))
1579  		bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1580  
1581  	/*
1582  	 * If another task is sharing our fs, we cannot safely
1583  	 * suid exec because the differently privileged task
1584  	 * will be able to manipulate the current directory, etc.
1585  	 * It would be nice to force an unshare instead...
1586  	 */
1587  	t = p;
1588  	n_fs = 1;
1589  	spin_lock(&p->fs->lock);
1590  	rcu_read_lock();
1591  	while_each_thread(p, t) {
1592  		if (t->fs == p->fs)
1593  			n_fs++;
1594  	}
1595  	rcu_read_unlock();
1596  
1597  	if (p->fs->users > n_fs)
1598  		bprm->unsafe |= LSM_UNSAFE_SHARE;
1599  	else
1600  		p->fs->in_exec = 1;
1601  	spin_unlock(&p->fs->lock);
1602  }
1603  
1604  static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1605  {
1606  	/* Handle suid and sgid on files */
1607  	struct mnt_idmap *idmap;
1608  	struct inode *inode = file_inode(file);
1609  	unsigned int mode;
1610  	vfsuid_t vfsuid;
1611  	vfsgid_t vfsgid;
1612  	int err;
1613  
1614  	if (!mnt_may_suid(file->f_path.mnt))
1615  		return;
1616  
1617  	if (task_no_new_privs(current))
1618  		return;
1619  
1620  	mode = READ_ONCE(inode->i_mode);
1621  	if (!(mode & (S_ISUID|S_ISGID)))
1622  		return;
1623  
1624  	idmap = file_mnt_idmap(file);
1625  
1626  	/* Be careful if suid/sgid is set */
1627  	inode_lock(inode);
1628  
1629  	/* Atomically reload and check mode/uid/gid now that lock held. */
1630  	mode = inode->i_mode;
1631  	vfsuid = i_uid_into_vfsuid(idmap, inode);
1632  	vfsgid = i_gid_into_vfsgid(idmap, inode);
1633  	err = inode_permission(idmap, inode, MAY_EXEC);
1634  	inode_unlock(inode);
1635  
1636  	/* Did the exec bit vanish out from under us? Give up. */
1637  	if (err)
1638  		return;
1639  
1640  	/* We ignore suid/sgid if there are no mappings for them in the ns */
1641  	if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1642  	    !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1643  		return;
1644  
1645  	if (mode & S_ISUID) {
1646  		bprm->per_clear |= PER_CLEAR_ON_SETID;
1647  		bprm->cred->euid = vfsuid_into_kuid(vfsuid);
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 = vfsgid_into_kgid(vfsgid);
1653  	}
1654  }
1655  
1656  /*
1657   * Compute brpm->cred based upon the final binary.
1658   */
1659  static 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   */
1674  static 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   */
1687  int 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_local_page(page);
1705  
1706  		for (; offset < PAGE_SIZE && kaddr[offset];
1707  				offset++, bprm->p++)
1708  			;
1709  
1710  		kunmap_local(kaddr);
1711  		put_arg_page(page);
1712  	} while (offset == PAGE_SIZE);
1713  
1714  	bprm->p++;
1715  	bprm->argc--;
1716  	ret = 0;
1717  
1718  out:
1719  	return ret;
1720  }
1721  EXPORT_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   */
1727  static 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  
1773  /* binfmt handlers will call back into begin_new_exec() on success. */
1774  static int exec_binprm(struct linux_binprm *bprm)
1775  {
1776  	pid_t old_pid, old_vpid;
1777  	int ret, depth;
1778  
1779  	/* Need to fetch pid before load_binary changes it */
1780  	old_pid = current->pid;
1781  	rcu_read_lock();
1782  	old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1783  	rcu_read_unlock();
1784  
1785  	/* This allows 4 levels of binfmt rewrites before failing hard. */
1786  	for (depth = 0;; depth++) {
1787  		struct file *exec;
1788  		if (depth > 5)
1789  			return -ELOOP;
1790  
1791  		ret = search_binary_handler(bprm);
1792  		if (ret < 0)
1793  			return ret;
1794  		if (!bprm->interpreter)
1795  			break;
1796  
1797  		exec = bprm->file;
1798  		bprm->file = bprm->interpreter;
1799  		bprm->interpreter = NULL;
1800  
1801  		allow_write_access(exec);
1802  		if (unlikely(bprm->have_execfd)) {
1803  			if (bprm->executable) {
1804  				fput(exec);
1805  				return -ENOEXEC;
1806  			}
1807  			bprm->executable = exec;
1808  		} else
1809  			fput(exec);
1810  	}
1811  
1812  	audit_bprm(bprm);
1813  	trace_sched_process_exec(current, old_pid, bprm);
1814  	ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1815  	proc_exec_connector(current);
1816  	return 0;
1817  }
1818  
1819  /*
1820   * sys_execve() executes a new program.
1821   */
1822  static int bprm_execve(struct linux_binprm *bprm,
1823  		       int fd, struct filename *filename, int flags)
1824  {
1825  	struct file *file;
1826  	int retval;
1827  
1828  	retval = prepare_bprm_creds(bprm);
1829  	if (retval)
1830  		return retval;
1831  
1832  	/*
1833  	 * Check for unsafe execution states before exec_binprm(), which
1834  	 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1835  	 * where setuid-ness is evaluated.
1836  	 */
1837  	check_unsafe_exec(bprm);
1838  	current->in_execve = 1;
1839  	sched_mm_cid_before_execve(current);
1840  
1841  	file = do_open_execat(fd, filename, flags);
1842  	retval = PTR_ERR(file);
1843  	if (IS_ERR(file))
1844  		goto out_unmark;
1845  
1846  	sched_exec();
1847  
1848  	bprm->file = file;
1849  	/*
1850  	 * Record that a name derived from an O_CLOEXEC fd will be
1851  	 * inaccessible after exec.  This allows the code in exec to
1852  	 * choose to fail when the executable is not mmaped into the
1853  	 * interpreter and an open file descriptor is not passed to
1854  	 * the interpreter.  This makes for a better user experience
1855  	 * than having the interpreter start and then immediately fail
1856  	 * when it finds the executable is inaccessible.
1857  	 */
1858  	if (bprm->fdpath && get_close_on_exec(fd))
1859  		bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1860  
1861  	/* Set the unchanging part of bprm->cred */
1862  	retval = security_bprm_creds_for_exec(bprm);
1863  	if (retval)
1864  		goto out;
1865  
1866  	retval = exec_binprm(bprm);
1867  	if (retval < 0)
1868  		goto out;
1869  
1870  	sched_mm_cid_after_execve(current);
1871  	/* execve succeeded */
1872  	current->fs->in_exec = 0;
1873  	current->in_execve = 0;
1874  	rseq_execve(current);
1875  	user_events_execve(current);
1876  	acct_update_integrals(current);
1877  	task_numa_free(current, false);
1878  	return retval;
1879  
1880  out:
1881  	/*
1882  	 * If past the point of no return ensure the code never
1883  	 * returns to the userspace process.  Use an existing fatal
1884  	 * signal if present otherwise terminate the process with
1885  	 * SIGSEGV.
1886  	 */
1887  	if (bprm->point_of_no_return && !fatal_signal_pending(current))
1888  		force_fatal_sig(SIGSEGV);
1889  
1890  out_unmark:
1891  	sched_mm_cid_after_execve(current);
1892  	current->fs->in_exec = 0;
1893  	current->in_execve = 0;
1894  
1895  	return retval;
1896  }
1897  
1898  static int do_execveat_common(int fd, struct filename *filename,
1899  			      struct user_arg_ptr argv,
1900  			      struct user_arg_ptr envp,
1901  			      int flags)
1902  {
1903  	struct linux_binprm *bprm;
1904  	int retval;
1905  
1906  	if (IS_ERR(filename))
1907  		return PTR_ERR(filename);
1908  
1909  	/*
1910  	 * We move the actual failure in case of RLIMIT_NPROC excess from
1911  	 * set*uid() to execve() because too many poorly written programs
1912  	 * don't check setuid() return code.  Here we additionally recheck
1913  	 * whether NPROC limit is still exceeded.
1914  	 */
1915  	if ((current->flags & PF_NPROC_EXCEEDED) &&
1916  	    is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1917  		retval = -EAGAIN;
1918  		goto out_ret;
1919  	}
1920  
1921  	/* We're below the limit (still or again), so we don't want to make
1922  	 * further execve() calls fail. */
1923  	current->flags &= ~PF_NPROC_EXCEEDED;
1924  
1925  	bprm = alloc_bprm(fd, filename);
1926  	if (IS_ERR(bprm)) {
1927  		retval = PTR_ERR(bprm);
1928  		goto out_ret;
1929  	}
1930  
1931  	retval = count(argv, MAX_ARG_STRINGS);
1932  	if (retval == 0)
1933  		pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1934  			     current->comm, bprm->filename);
1935  	if (retval < 0)
1936  		goto out_free;
1937  	bprm->argc = retval;
1938  
1939  	retval = count(envp, MAX_ARG_STRINGS);
1940  	if (retval < 0)
1941  		goto out_free;
1942  	bprm->envc = retval;
1943  
1944  	retval = bprm_stack_limits(bprm);
1945  	if (retval < 0)
1946  		goto out_free;
1947  
1948  	retval = copy_string_kernel(bprm->filename, bprm);
1949  	if (retval < 0)
1950  		goto out_free;
1951  	bprm->exec = bprm->p;
1952  
1953  	retval = copy_strings(bprm->envc, envp, bprm);
1954  	if (retval < 0)
1955  		goto out_free;
1956  
1957  	retval = copy_strings(bprm->argc, argv, bprm);
1958  	if (retval < 0)
1959  		goto out_free;
1960  
1961  	/*
1962  	 * When argv is empty, add an empty string ("") as argv[0] to
1963  	 * ensure confused userspace programs that start processing
1964  	 * from argv[1] won't end up walking envp. See also
1965  	 * bprm_stack_limits().
1966  	 */
1967  	if (bprm->argc == 0) {
1968  		retval = copy_string_kernel("", bprm);
1969  		if (retval < 0)
1970  			goto out_free;
1971  		bprm->argc = 1;
1972  	}
1973  
1974  	retval = bprm_execve(bprm, fd, filename, flags);
1975  out_free:
1976  	free_bprm(bprm);
1977  
1978  out_ret:
1979  	putname(filename);
1980  	return retval;
1981  }
1982  
1983  int kernel_execve(const char *kernel_filename,
1984  		  const char *const *argv, const char *const *envp)
1985  {
1986  	struct filename *filename;
1987  	struct linux_binprm *bprm;
1988  	int fd = AT_FDCWD;
1989  	int retval;
1990  
1991  	/* It is non-sense for kernel threads to call execve */
1992  	if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1993  		return -EINVAL;
1994  
1995  	filename = getname_kernel(kernel_filename);
1996  	if (IS_ERR(filename))
1997  		return PTR_ERR(filename);
1998  
1999  	bprm = alloc_bprm(fd, filename);
2000  	if (IS_ERR(bprm)) {
2001  		retval = PTR_ERR(bprm);
2002  		goto out_ret;
2003  	}
2004  
2005  	retval = count_strings_kernel(argv);
2006  	if (WARN_ON_ONCE(retval == 0))
2007  		retval = -EINVAL;
2008  	if (retval < 0)
2009  		goto out_free;
2010  	bprm->argc = retval;
2011  
2012  	retval = count_strings_kernel(envp);
2013  	if (retval < 0)
2014  		goto out_free;
2015  	bprm->envc = retval;
2016  
2017  	retval = bprm_stack_limits(bprm);
2018  	if (retval < 0)
2019  		goto out_free;
2020  
2021  	retval = copy_string_kernel(bprm->filename, bprm);
2022  	if (retval < 0)
2023  		goto out_free;
2024  	bprm->exec = bprm->p;
2025  
2026  	retval = copy_strings_kernel(bprm->envc, envp, bprm);
2027  	if (retval < 0)
2028  		goto out_free;
2029  
2030  	retval = copy_strings_kernel(bprm->argc, argv, bprm);
2031  	if (retval < 0)
2032  		goto out_free;
2033  
2034  	retval = bprm_execve(bprm, fd, filename, 0);
2035  out_free:
2036  	free_bprm(bprm);
2037  out_ret:
2038  	putname(filename);
2039  	return retval;
2040  }
2041  
2042  static int do_execve(struct filename *filename,
2043  	const char __user *const __user *__argv,
2044  	const char __user *const __user *__envp)
2045  {
2046  	struct user_arg_ptr argv = { .ptr.native = __argv };
2047  	struct user_arg_ptr envp = { .ptr.native = __envp };
2048  	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2049  }
2050  
2051  static int do_execveat(int fd, struct filename *filename,
2052  		const char __user *const __user *__argv,
2053  		const char __user *const __user *__envp,
2054  		int flags)
2055  {
2056  	struct user_arg_ptr argv = { .ptr.native = __argv };
2057  	struct user_arg_ptr envp = { .ptr.native = __envp };
2058  
2059  	return do_execveat_common(fd, filename, argv, envp, flags);
2060  }
2061  
2062  #ifdef CONFIG_COMPAT
2063  static int compat_do_execve(struct filename *filename,
2064  	const compat_uptr_t __user *__argv,
2065  	const compat_uptr_t __user *__envp)
2066  {
2067  	struct user_arg_ptr argv = {
2068  		.is_compat = true,
2069  		.ptr.compat = __argv,
2070  	};
2071  	struct user_arg_ptr envp = {
2072  		.is_compat = true,
2073  		.ptr.compat = __envp,
2074  	};
2075  	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2076  }
2077  
2078  static int compat_do_execveat(int fd, struct filename *filename,
2079  			      const compat_uptr_t __user *__argv,
2080  			      const compat_uptr_t __user *__envp,
2081  			      int flags)
2082  {
2083  	struct user_arg_ptr argv = {
2084  		.is_compat = true,
2085  		.ptr.compat = __argv,
2086  	};
2087  	struct user_arg_ptr envp = {
2088  		.is_compat = true,
2089  		.ptr.compat = __envp,
2090  	};
2091  	return do_execveat_common(fd, filename, argv, envp, flags);
2092  }
2093  #endif
2094  
2095  void set_binfmt(struct linux_binfmt *new)
2096  {
2097  	struct mm_struct *mm = current->mm;
2098  
2099  	if (mm->binfmt)
2100  		module_put(mm->binfmt->module);
2101  
2102  	mm->binfmt = new;
2103  	if (new)
2104  		__module_get(new->module);
2105  }
2106  EXPORT_SYMBOL(set_binfmt);
2107  
2108  /*
2109   * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2110   */
2111  void set_dumpable(struct mm_struct *mm, int value)
2112  {
2113  	if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2114  		return;
2115  
2116  	set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2117  }
2118  
2119  SYSCALL_DEFINE3(execve,
2120  		const char __user *, filename,
2121  		const char __user *const __user *, argv,
2122  		const char __user *const __user *, envp)
2123  {
2124  	return do_execve(getname(filename), argv, envp);
2125  }
2126  
2127  SYSCALL_DEFINE5(execveat,
2128  		int, fd, const char __user *, filename,
2129  		const char __user *const __user *, argv,
2130  		const char __user *const __user *, envp,
2131  		int, flags)
2132  {
2133  	return do_execveat(fd,
2134  			   getname_uflags(filename, flags),
2135  			   argv, envp, flags);
2136  }
2137  
2138  #ifdef CONFIG_COMPAT
2139  COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2140  	const compat_uptr_t __user *, argv,
2141  	const compat_uptr_t __user *, envp)
2142  {
2143  	return compat_do_execve(getname(filename), argv, envp);
2144  }
2145  
2146  COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2147  		       const char __user *, filename,
2148  		       const compat_uptr_t __user *, argv,
2149  		       const compat_uptr_t __user *, envp,
2150  		       int,  flags)
2151  {
2152  	return compat_do_execveat(fd,
2153  				  getname_uflags(filename, flags),
2154  				  argv, envp, flags);
2155  }
2156  #endif
2157  
2158  #ifdef CONFIG_SYSCTL
2159  
2160  static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2161  		void *buffer, size_t *lenp, loff_t *ppos)
2162  {
2163  	int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2164  
2165  	if (!error)
2166  		validate_coredump_safety();
2167  	return error;
2168  }
2169  
2170  static struct ctl_table fs_exec_sysctls[] = {
2171  	{
2172  		.procname	= "suid_dumpable",
2173  		.data		= &suid_dumpable,
2174  		.maxlen		= sizeof(int),
2175  		.mode		= 0644,
2176  		.proc_handler	= proc_dointvec_minmax_coredump,
2177  		.extra1		= SYSCTL_ZERO,
2178  		.extra2		= SYSCTL_TWO,
2179  	},
2180  	{ }
2181  };
2182  
2183  static int __init init_fs_exec_sysctls(void)
2184  {
2185  	register_sysctl_init("fs", fs_exec_sysctls);
2186  	return 0;
2187  }
2188  
2189  fs_initcall(init_fs_exec_sysctls);
2190  #endif /* CONFIG_SYSCTL */
2191