xref: /openbmc/linux/fs/exec.c (revision 3e5f1ff7e4f297b928fb579fa1e8ec65a400162e)
1  // SPDX-License-Identifier: GPL-2.0-only
2  /*
3   *  linux/fs/exec.c
4   *
5   *  Copyright (C) 1991, 1992  Linus Torvalds
6   */
7  
8  /*
9   * #!-checking implemented by tytso.
10   */
11  /*
12   * Demand-loading implemented 01.12.91 - no need to read anything but
13   * the header into memory. The inode of the executable is put into
14   * "current->executable", and page faults do the actual loading. Clean.
15   *
16   * Once more I can proudly say that linux stood up to being changed: it
17   * was less than 2 hours work to get demand-loading completely implemented.
18   *
19   * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
20   * current->executable is only used by the procfs.  This allows a dispatch
21   * table to check for several different types  of binary formats.  We keep
22   * trying until we recognize the file or we run out of supported binary
23   * formats.
24   */
25  
26  #include <linux/kernel_read_file.h>
27  #include <linux/slab.h>
28  #include <linux/file.h>
29  #include <linux/fdtable.h>
30  #include <linux/mm.h>
31  #include <linux/vmacache.h>
32  #include <linux/stat.h>
33  #include <linux/fcntl.h>
34  #include <linux/swap.h>
35  #include <linux/string.h>
36  #include <linux/init.h>
37  #include <linux/sched/mm.h>
38  #include <linux/sched/coredump.h>
39  #include <linux/sched/signal.h>
40  #include <linux/sched/numa_balancing.h>
41  #include <linux/sched/task.h>
42  #include <linux/pagemap.h>
43  #include <linux/perf_event.h>
44  #include <linux/highmem.h>
45  #include <linux/spinlock.h>
46  #include <linux/key.h>
47  #include <linux/personality.h>
48  #include <linux/binfmts.h>
49  #include <linux/utsname.h>
50  #include <linux/pid_namespace.h>
51  #include <linux/module.h>
52  #include <linux/namei.h>
53  #include <linux/mount.h>
54  #include <linux/security.h>
55  #include <linux/syscalls.h>
56  #include <linux/tsacct_kern.h>
57  #include <linux/cn_proc.h>
58  #include <linux/audit.h>
59  #include <linux/tracehook.h>
60  #include <linux/kmod.h>
61  #include <linux/fsnotify.h>
62  #include <linux/fs_struct.h>
63  #include <linux/oom.h>
64  #include <linux/compat.h>
65  #include <linux/vmalloc.h>
66  #include <linux/io_uring.h>
67  #include <linux/syscall_user_dispatch.h>
68  #include <linux/coredump.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 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  	fsnotify_open(file);
156  
157  	error = -ENOEXEC;
158  
159  	read_lock(&binfmt_lock);
160  	list_for_each_entry(fmt, &formats, lh) {
161  		if (!fmt->load_shlib)
162  			continue;
163  		if (!try_module_get(fmt->module))
164  			continue;
165  		read_unlock(&binfmt_lock);
166  		error = fmt->load_shlib(file);
167  		read_lock(&binfmt_lock);
168  		put_binfmt(fmt);
169  		if (error != -ENOEXEC)
170  			break;
171  	}
172  	read_unlock(&binfmt_lock);
173  exit:
174  	fput(file);
175  out:
176    	return error;
177  }
178  #endif /* #ifdef CONFIG_USELIB */
179  
180  #ifdef CONFIG_MMU
181  /*
182   * The nascent bprm->mm is not visible until exec_mmap() but it can
183   * use a lot of memory, account these pages in current->mm temporary
184   * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
185   * change the counter back via acct_arg_size(0).
186   */
187  static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
188  {
189  	struct mm_struct *mm = current->mm;
190  	long diff = (long)(pages - bprm->vma_pages);
191  
192  	if (!mm || !diff)
193  		return;
194  
195  	bprm->vma_pages = pages;
196  	add_mm_counter(mm, MM_ANONPAGES, diff);
197  }
198  
199  static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
200  		int write)
201  {
202  	struct page *page;
203  	int ret;
204  	unsigned int gup_flags = FOLL_FORCE;
205  
206  #ifdef CONFIG_STACK_GROWSUP
207  	if (write) {
208  		ret = expand_downwards(bprm->vma, pos);
209  		if (ret < 0)
210  			return NULL;
211  	}
212  #endif
213  
214  	if (write)
215  		gup_flags |= FOLL_WRITE;
216  
217  	/*
218  	 * We are doing an exec().  'current' is the process
219  	 * doing the exec and bprm->mm is the new process's mm.
220  	 */
221  	mmap_read_lock(bprm->mm);
222  	ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
223  			&page, NULL, NULL);
224  	mmap_read_unlock(bprm->mm);
225  	if (ret <= 0)
226  		return NULL;
227  
228  	if (write)
229  		acct_arg_size(bprm, vma_pages(bprm->vma));
230  
231  	return page;
232  }
233  
234  static void put_arg_page(struct page *page)
235  {
236  	put_page(page);
237  }
238  
239  static void free_arg_pages(struct linux_binprm *bprm)
240  {
241  }
242  
243  static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
244  		struct page *page)
245  {
246  	flush_cache_page(bprm->vma, pos, page_to_pfn(page));
247  }
248  
249  static int __bprm_mm_init(struct linux_binprm *bprm)
250  {
251  	int err;
252  	struct vm_area_struct *vma = NULL;
253  	struct mm_struct *mm = bprm->mm;
254  
255  	bprm->vma = vma = vm_area_alloc(mm);
256  	if (!vma)
257  		return -ENOMEM;
258  	vma_set_anonymous(vma);
259  
260  	if (mmap_write_lock_killable(mm)) {
261  		err = -EINTR;
262  		goto err_free;
263  	}
264  
265  	/*
266  	 * Place the stack at the largest stack address the architecture
267  	 * supports. Later, we'll move this to an appropriate place. We don't
268  	 * use STACK_TOP because that can depend on attributes which aren't
269  	 * configured yet.
270  	 */
271  	BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
272  	vma->vm_end = STACK_TOP_MAX;
273  	vma->vm_start = vma->vm_end - PAGE_SIZE;
274  	vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
275  	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
276  
277  	err = insert_vm_struct(mm, vma);
278  	if (err)
279  		goto err;
280  
281  	mm->stack_vm = mm->total_vm = 1;
282  	mmap_write_unlock(mm);
283  	bprm->p = vma->vm_end - sizeof(void *);
284  	return 0;
285  err:
286  	mmap_write_unlock(mm);
287  err_free:
288  	bprm->vma = NULL;
289  	vm_area_free(vma);
290  	return err;
291  }
292  
293  static bool valid_arg_len(struct linux_binprm *bprm, long len)
294  {
295  	return len <= MAX_ARG_STRLEN;
296  }
297  
298  #else
299  
300  static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
301  {
302  }
303  
304  static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
305  		int write)
306  {
307  	struct page *page;
308  
309  	page = bprm->page[pos / PAGE_SIZE];
310  	if (!page && write) {
311  		page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
312  		if (!page)
313  			return NULL;
314  		bprm->page[pos / PAGE_SIZE] = page;
315  	}
316  
317  	return page;
318  }
319  
320  static void put_arg_page(struct page *page)
321  {
322  }
323  
324  static void free_arg_page(struct linux_binprm *bprm, int i)
325  {
326  	if (bprm->page[i]) {
327  		__free_page(bprm->page[i]);
328  		bprm->page[i] = NULL;
329  	}
330  }
331  
332  static void free_arg_pages(struct linux_binprm *bprm)
333  {
334  	int i;
335  
336  	for (i = 0; i < MAX_ARG_PAGES; i++)
337  		free_arg_page(bprm, i);
338  }
339  
340  static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
341  		struct page *page)
342  {
343  }
344  
345  static int __bprm_mm_init(struct linux_binprm *bprm)
346  {
347  	bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
348  	return 0;
349  }
350  
351  static bool valid_arg_len(struct linux_binprm *bprm, long len)
352  {
353  	return len <= bprm->p;
354  }
355  
356  #endif /* CONFIG_MMU */
357  
358  /*
359   * Create a new mm_struct and populate it with a temporary stack
360   * vm_area_struct.  We don't have enough context at this point to set the stack
361   * flags, permissions, and offset, so we use temporary values.  We'll update
362   * them later in setup_arg_pages().
363   */
364  static int bprm_mm_init(struct linux_binprm *bprm)
365  {
366  	int err;
367  	struct mm_struct *mm = NULL;
368  
369  	bprm->mm = mm = mm_alloc();
370  	err = -ENOMEM;
371  	if (!mm)
372  		goto err;
373  
374  	/* Save current stack limit for all calculations made during exec. */
375  	task_lock(current->group_leader);
376  	bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
377  	task_unlock(current->group_leader);
378  
379  	err = __bprm_mm_init(bprm);
380  	if (err)
381  		goto err;
382  
383  	return 0;
384  
385  err:
386  	if (mm) {
387  		bprm->mm = NULL;
388  		mmdrop(mm);
389  	}
390  
391  	return err;
392  }
393  
394  struct user_arg_ptr {
395  #ifdef CONFIG_COMPAT
396  	bool is_compat;
397  #endif
398  	union {
399  		const char __user *const __user *native;
400  #ifdef CONFIG_COMPAT
401  		const compat_uptr_t __user *compat;
402  #endif
403  	} ptr;
404  };
405  
406  static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
407  {
408  	const char __user *native;
409  
410  #ifdef CONFIG_COMPAT
411  	if (unlikely(argv.is_compat)) {
412  		compat_uptr_t compat;
413  
414  		if (get_user(compat, argv.ptr.compat + nr))
415  			return ERR_PTR(-EFAULT);
416  
417  		return compat_ptr(compat);
418  	}
419  #endif
420  
421  	if (get_user(native, argv.ptr.native + nr))
422  		return ERR_PTR(-EFAULT);
423  
424  	return native;
425  }
426  
427  /*
428   * count() counts the number of strings in array ARGV.
429   */
430  static int count(struct user_arg_ptr argv, int max)
431  {
432  	int i = 0;
433  
434  	if (argv.ptr.native != NULL) {
435  		for (;;) {
436  			const char __user *p = get_user_arg_ptr(argv, i);
437  
438  			if (!p)
439  				break;
440  
441  			if (IS_ERR(p))
442  				return -EFAULT;
443  
444  			if (i >= max)
445  				return -E2BIG;
446  			++i;
447  
448  			if (fatal_signal_pending(current))
449  				return -ERESTARTNOHAND;
450  			cond_resched();
451  		}
452  	}
453  	return i;
454  }
455  
456  static int count_strings_kernel(const char *const *argv)
457  {
458  	int i;
459  
460  	if (!argv)
461  		return 0;
462  
463  	for (i = 0; argv[i]; ++i) {
464  		if (i >= MAX_ARG_STRINGS)
465  			return -E2BIG;
466  		if (fatal_signal_pending(current))
467  			return -ERESTARTNOHAND;
468  		cond_resched();
469  	}
470  	return i;
471  }
472  
473  static int bprm_stack_limits(struct linux_binprm *bprm)
474  {
475  	unsigned long limit, ptr_size;
476  
477  	/*
478  	 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
479  	 * (whichever is smaller) for the argv+env strings.
480  	 * This ensures that:
481  	 *  - the remaining binfmt code will not run out of stack space,
482  	 *  - the program will have a reasonable amount of stack left
483  	 *    to work from.
484  	 */
485  	limit = _STK_LIM / 4 * 3;
486  	limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
487  	/*
488  	 * We've historically supported up to 32 pages (ARG_MAX)
489  	 * of argument strings even with small stacks
490  	 */
491  	limit = max_t(unsigned long, limit, ARG_MAX);
492  	/*
493  	 * We must account for the size of all the argv and envp pointers to
494  	 * the argv and envp strings, since they will also take up space in
495  	 * the stack. They aren't stored until much later when we can't
496  	 * signal to the parent that the child has run out of stack space.
497  	 * Instead, calculate it here so it's possible to fail gracefully.
498  	 */
499  	ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
500  	if (limit <= ptr_size)
501  		return -E2BIG;
502  	limit -= ptr_size;
503  
504  	bprm->argmin = bprm->p - limit;
505  	return 0;
506  }
507  
508  /*
509   * 'copy_strings()' copies argument/environment strings from the old
510   * processes's memory to the new process's stack.  The call to get_user_pages()
511   * ensures the destination page is created and not swapped out.
512   */
513  static int copy_strings(int argc, struct user_arg_ptr argv,
514  			struct linux_binprm *bprm)
515  {
516  	struct page *kmapped_page = NULL;
517  	char *kaddr = NULL;
518  	unsigned long kpos = 0;
519  	int ret;
520  
521  	while (argc-- > 0) {
522  		const char __user *str;
523  		int len;
524  		unsigned long pos;
525  
526  		ret = -EFAULT;
527  		str = get_user_arg_ptr(argv, argc);
528  		if (IS_ERR(str))
529  			goto out;
530  
531  		len = strnlen_user(str, MAX_ARG_STRLEN);
532  		if (!len)
533  			goto out;
534  
535  		ret = -E2BIG;
536  		if (!valid_arg_len(bprm, len))
537  			goto out;
538  
539  		/* We're going to work our way backwords. */
540  		pos = bprm->p;
541  		str += len;
542  		bprm->p -= len;
543  #ifdef CONFIG_MMU
544  		if (bprm->p < bprm->argmin)
545  			goto out;
546  #endif
547  
548  		while (len > 0) {
549  			int offset, bytes_to_copy;
550  
551  			if (fatal_signal_pending(current)) {
552  				ret = -ERESTARTNOHAND;
553  				goto out;
554  			}
555  			cond_resched();
556  
557  			offset = pos % PAGE_SIZE;
558  			if (offset == 0)
559  				offset = PAGE_SIZE;
560  
561  			bytes_to_copy = offset;
562  			if (bytes_to_copy > len)
563  				bytes_to_copy = len;
564  
565  			offset -= bytes_to_copy;
566  			pos -= bytes_to_copy;
567  			str -= bytes_to_copy;
568  			len -= bytes_to_copy;
569  
570  			if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
571  				struct page *page;
572  
573  				page = get_arg_page(bprm, pos, 1);
574  				if (!page) {
575  					ret = -E2BIG;
576  					goto out;
577  				}
578  
579  				if (kmapped_page) {
580  					flush_dcache_page(kmapped_page);
581  					kunmap(kmapped_page);
582  					put_arg_page(kmapped_page);
583  				}
584  				kmapped_page = page;
585  				kaddr = kmap(kmapped_page);
586  				kpos = pos & PAGE_MASK;
587  				flush_arg_page(bprm, kpos, kmapped_page);
588  			}
589  			if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
590  				ret = -EFAULT;
591  				goto out;
592  			}
593  		}
594  	}
595  	ret = 0;
596  out:
597  	if (kmapped_page) {
598  		flush_dcache_page(kmapped_page);
599  		kunmap(kmapped_page);
600  		put_arg_page(kmapped_page);
601  	}
602  	return ret;
603  }
604  
605  /*
606   * Copy and argument/environment string from the kernel to the processes stack.
607   */
608  int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
609  {
610  	int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
611  	unsigned long pos = bprm->p;
612  
613  	if (len == 0)
614  		return -EFAULT;
615  	if (!valid_arg_len(bprm, len))
616  		return -E2BIG;
617  
618  	/* We're going to work our way backwards. */
619  	arg += len;
620  	bprm->p -= len;
621  	if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
622  		return -E2BIG;
623  
624  	while (len > 0) {
625  		unsigned int bytes_to_copy = min_t(unsigned int, len,
626  				min_not_zero(offset_in_page(pos), PAGE_SIZE));
627  		struct page *page;
628  		char *kaddr;
629  
630  		pos -= bytes_to_copy;
631  		arg -= bytes_to_copy;
632  		len -= bytes_to_copy;
633  
634  		page = get_arg_page(bprm, pos, 1);
635  		if (!page)
636  			return -E2BIG;
637  		kaddr = kmap_atomic(page);
638  		flush_arg_page(bprm, pos & PAGE_MASK, page);
639  		memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
640  		flush_dcache_page(page);
641  		kunmap_atomic(kaddr);
642  		put_arg_page(page);
643  	}
644  
645  	return 0;
646  }
647  EXPORT_SYMBOL(copy_string_kernel);
648  
649  static int copy_strings_kernel(int argc, const char *const *argv,
650  			       struct linux_binprm *bprm)
651  {
652  	while (argc-- > 0) {
653  		int ret = copy_string_kernel(argv[argc], bprm);
654  		if (ret < 0)
655  			return ret;
656  		if (fatal_signal_pending(current))
657  			return -ERESTARTNOHAND;
658  		cond_resched();
659  	}
660  	return 0;
661  }
662  
663  #ifdef CONFIG_MMU
664  
665  /*
666   * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
667   * the binfmt code determines where the new stack should reside, we shift it to
668   * its final location.  The process proceeds as follows:
669   *
670   * 1) Use shift to calculate the new vma endpoints.
671   * 2) Extend vma to cover both the old and new ranges.  This ensures the
672   *    arguments passed to subsequent functions are consistent.
673   * 3) Move vma's page tables to the new range.
674   * 4) Free up any cleared pgd range.
675   * 5) Shrink the vma to cover only the new range.
676   */
677  static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
678  {
679  	struct mm_struct *mm = vma->vm_mm;
680  	unsigned long old_start = vma->vm_start;
681  	unsigned long old_end = vma->vm_end;
682  	unsigned long length = old_end - old_start;
683  	unsigned long new_start = old_start - shift;
684  	unsigned long new_end = old_end - shift;
685  	struct mmu_gather tlb;
686  
687  	BUG_ON(new_start > new_end);
688  
689  	/*
690  	 * ensure there are no vmas between where we want to go
691  	 * and where we are
692  	 */
693  	if (vma != find_vma(mm, new_start))
694  		return -EFAULT;
695  
696  	/*
697  	 * cover the whole range: [new_start, old_end)
698  	 */
699  	if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
700  		return -ENOMEM;
701  
702  	/*
703  	 * move the page tables downwards, on failure we rely on
704  	 * process cleanup to remove whatever mess we made.
705  	 */
706  	if (length != move_page_tables(vma, old_start,
707  				       vma, new_start, length, false))
708  		return -ENOMEM;
709  
710  	lru_add_drain();
711  	tlb_gather_mmu(&tlb, mm);
712  	if (new_end > old_start) {
713  		/*
714  		 * when the old and new regions overlap clear from new_end.
715  		 */
716  		free_pgd_range(&tlb, new_end, old_end, new_end,
717  			vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
718  	} else {
719  		/*
720  		 * otherwise, clean from old_start; this is done to not touch
721  		 * the address space in [new_end, old_start) some architectures
722  		 * have constraints on va-space that make this illegal (IA64) -
723  		 * for the others its just a little faster.
724  		 */
725  		free_pgd_range(&tlb, old_start, old_end, new_end,
726  			vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
727  	}
728  	tlb_finish_mmu(&tlb);
729  
730  	/*
731  	 * Shrink the vma to just the new range.  Always succeeds.
732  	 */
733  	vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
734  
735  	return 0;
736  }
737  
738  /*
739   * Finalizes the stack vm_area_struct. The flags and permissions are updated,
740   * the stack is optionally relocated, and some extra space is added.
741   */
742  int setup_arg_pages(struct linux_binprm *bprm,
743  		    unsigned long stack_top,
744  		    int executable_stack)
745  {
746  	unsigned long ret;
747  	unsigned long stack_shift;
748  	struct mm_struct *mm = current->mm;
749  	struct vm_area_struct *vma = bprm->vma;
750  	struct vm_area_struct *prev = NULL;
751  	unsigned long vm_flags;
752  	unsigned long stack_base;
753  	unsigned long stack_size;
754  	unsigned long stack_expand;
755  	unsigned long rlim_stack;
756  
757  #ifdef CONFIG_STACK_GROWSUP
758  	/* Limit stack size */
759  	stack_base = bprm->rlim_stack.rlim_max;
760  
761  	stack_base = calc_max_stack_size(stack_base);
762  
763  	/* Add space for stack randomization. */
764  	stack_base += (STACK_RND_MASK << PAGE_SHIFT);
765  
766  	/* Make sure we didn't let the argument array grow too large. */
767  	if (vma->vm_end - vma->vm_start > stack_base)
768  		return -ENOMEM;
769  
770  	stack_base = PAGE_ALIGN(stack_top - stack_base);
771  
772  	stack_shift = vma->vm_start - stack_base;
773  	mm->arg_start = bprm->p - stack_shift;
774  	bprm->p = vma->vm_end - stack_shift;
775  #else
776  	stack_top = arch_align_stack(stack_top);
777  	stack_top = PAGE_ALIGN(stack_top);
778  
779  	if (unlikely(stack_top < mmap_min_addr) ||
780  	    unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
781  		return -ENOMEM;
782  
783  	stack_shift = vma->vm_end - stack_top;
784  
785  	bprm->p -= stack_shift;
786  	mm->arg_start = bprm->p;
787  #endif
788  
789  	if (bprm->loader)
790  		bprm->loader -= stack_shift;
791  	bprm->exec -= stack_shift;
792  
793  	if (mmap_write_lock_killable(mm))
794  		return -EINTR;
795  
796  	vm_flags = VM_STACK_FLAGS;
797  
798  	/*
799  	 * Adjust stack execute permissions; explicitly enable for
800  	 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
801  	 * (arch default) otherwise.
802  	 */
803  	if (unlikely(executable_stack == EXSTACK_ENABLE_X))
804  		vm_flags |= VM_EXEC;
805  	else if (executable_stack == EXSTACK_DISABLE_X)
806  		vm_flags &= ~VM_EXEC;
807  	vm_flags |= mm->def_flags;
808  	vm_flags |= VM_STACK_INCOMPLETE_SETUP;
809  
810  	ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
811  			vm_flags);
812  	if (ret)
813  		goto out_unlock;
814  	BUG_ON(prev != vma);
815  
816  	if (unlikely(vm_flags & VM_EXEC)) {
817  		pr_warn_once("process '%pD4' started with executable stack\n",
818  			     bprm->file);
819  	}
820  
821  	/* Move stack pages down in memory. */
822  	if (stack_shift) {
823  		ret = shift_arg_pages(vma, stack_shift);
824  		if (ret)
825  			goto out_unlock;
826  	}
827  
828  	/* mprotect_fixup is overkill to remove the temporary stack flags */
829  	vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
830  
831  	stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
832  	stack_size = vma->vm_end - vma->vm_start;
833  	/*
834  	 * Align this down to a page boundary as expand_stack
835  	 * will align it up.
836  	 */
837  	rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
838  #ifdef CONFIG_STACK_GROWSUP
839  	if (stack_size + stack_expand > rlim_stack)
840  		stack_base = vma->vm_start + rlim_stack;
841  	else
842  		stack_base = vma->vm_end + stack_expand;
843  #else
844  	if (stack_size + stack_expand > rlim_stack)
845  		stack_base = vma->vm_end - rlim_stack;
846  	else
847  		stack_base = vma->vm_start - stack_expand;
848  #endif
849  	current->mm->start_stack = bprm->p;
850  	ret = expand_stack(vma, stack_base);
851  	if (ret)
852  		ret = -EFAULT;
853  
854  out_unlock:
855  	mmap_write_unlock(mm);
856  	return ret;
857  }
858  EXPORT_SYMBOL(setup_arg_pages);
859  
860  #else
861  
862  /*
863   * Transfer the program arguments and environment from the holding pages
864   * onto the stack. The provided stack pointer is adjusted accordingly.
865   */
866  int transfer_args_to_stack(struct linux_binprm *bprm,
867  			   unsigned long *sp_location)
868  {
869  	unsigned long index, stop, sp;
870  	int ret = 0;
871  
872  	stop = bprm->p >> PAGE_SHIFT;
873  	sp = *sp_location;
874  
875  	for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
876  		unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
877  		char *src = kmap(bprm->page[index]) + offset;
878  		sp -= PAGE_SIZE - offset;
879  		if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
880  			ret = -EFAULT;
881  		kunmap(bprm->page[index]);
882  		if (ret)
883  			goto out;
884  	}
885  
886  	*sp_location = sp;
887  
888  out:
889  	return ret;
890  }
891  EXPORT_SYMBOL(transfer_args_to_stack);
892  
893  #endif /* CONFIG_MMU */
894  
895  static struct file *do_open_execat(int fd, struct filename *name, int flags)
896  {
897  	struct file *file;
898  	int err;
899  	struct open_flags open_exec_flags = {
900  		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
901  		.acc_mode = MAY_EXEC,
902  		.intent = LOOKUP_OPEN,
903  		.lookup_flags = LOOKUP_FOLLOW,
904  	};
905  
906  	if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
907  		return ERR_PTR(-EINVAL);
908  	if (flags & AT_SYMLINK_NOFOLLOW)
909  		open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
910  	if (flags & AT_EMPTY_PATH)
911  		open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
912  
913  	file = do_filp_open(fd, name, &open_exec_flags);
914  	if (IS_ERR(file))
915  		goto out;
916  
917  	/*
918  	 * may_open() has already checked for this, so it should be
919  	 * impossible to trip now. But we need to be extra cautious
920  	 * and check again at the very end too.
921  	 */
922  	err = -EACCES;
923  	if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
924  			 path_noexec(&file->f_path)))
925  		goto exit;
926  
927  	err = deny_write_access(file);
928  	if (err)
929  		goto exit;
930  
931  	if (name->name[0] != '\0')
932  		fsnotify_open(file);
933  
934  out:
935  	return file;
936  
937  exit:
938  	fput(file);
939  	return ERR_PTR(err);
940  }
941  
942  struct file *open_exec(const char *name)
943  {
944  	struct filename *filename = getname_kernel(name);
945  	struct file *f = ERR_CAST(filename);
946  
947  	if (!IS_ERR(filename)) {
948  		f = do_open_execat(AT_FDCWD, filename, 0);
949  		putname(filename);
950  	}
951  	return f;
952  }
953  EXPORT_SYMBOL(open_exec);
954  
955  #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
956      defined(CONFIG_BINFMT_ELF_FDPIC)
957  ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
958  {
959  	ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
960  	if (res > 0)
961  		flush_icache_user_range(addr, addr + len);
962  	return res;
963  }
964  EXPORT_SYMBOL(read_code);
965  #endif
966  
967  /*
968   * Maps the mm_struct mm into the current task struct.
969   * On success, this function returns with exec_update_lock
970   * held for writing.
971   */
972  static int exec_mmap(struct mm_struct *mm)
973  {
974  	struct task_struct *tsk;
975  	struct mm_struct *old_mm, *active_mm;
976  	int ret;
977  
978  	/* Notify parent that we're no longer interested in the old VM */
979  	tsk = current;
980  	old_mm = current->mm;
981  	exec_mm_release(tsk, old_mm);
982  	if (old_mm)
983  		sync_mm_rss(old_mm);
984  
985  	ret = down_write_killable(&tsk->signal->exec_update_lock);
986  	if (ret)
987  		return ret;
988  
989  	if (old_mm) {
990  		/*
991  		 * If there is a pending fatal signal perhaps a signal
992  		 * whose default action is to create a coredump get
993  		 * out and die instead of going through with the exec.
994  		 */
995  		ret = mmap_read_lock_killable(old_mm);
996  		if (ret) {
997  			up_write(&tsk->signal->exec_update_lock);
998  			return ret;
999  		}
1000  	}
1001  
1002  	task_lock(tsk);
1003  	membarrier_exec_mmap(mm);
1004  
1005  	local_irq_disable();
1006  	active_mm = tsk->active_mm;
1007  	tsk->active_mm = mm;
1008  	tsk->mm = mm;
1009  	/*
1010  	 * This prevents preemption while active_mm is being loaded and
1011  	 * it and mm are being updated, which could cause problems for
1012  	 * lazy tlb mm refcounting when these are updated by context
1013  	 * switches. Not all architectures can handle irqs off over
1014  	 * activate_mm yet.
1015  	 */
1016  	if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1017  		local_irq_enable();
1018  	activate_mm(active_mm, mm);
1019  	if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1020  		local_irq_enable();
1021  	tsk->mm->vmacache_seqnum = 0;
1022  	vmacache_flush(tsk);
1023  	task_unlock(tsk);
1024  	if (old_mm) {
1025  		mmap_read_unlock(old_mm);
1026  		BUG_ON(active_mm != old_mm);
1027  		setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1028  		mm_update_next_owner(old_mm);
1029  		mmput(old_mm);
1030  		return 0;
1031  	}
1032  	mmdrop(active_mm);
1033  	return 0;
1034  }
1035  
1036  static int de_thread(struct task_struct *tsk)
1037  {
1038  	struct signal_struct *sig = tsk->signal;
1039  	struct sighand_struct *oldsighand = tsk->sighand;
1040  	spinlock_t *lock = &oldsighand->siglock;
1041  
1042  	if (thread_group_empty(tsk))
1043  		goto no_thread_group;
1044  
1045  	/*
1046  	 * Kill all other threads in the thread group.
1047  	 */
1048  	spin_lock_irq(lock);
1049  	if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1050  		/*
1051  		 * Another group action in progress, just
1052  		 * return so that the signal is processed.
1053  		 */
1054  		spin_unlock_irq(lock);
1055  		return -EAGAIN;
1056  	}
1057  
1058  	sig->group_exec_task = tsk;
1059  	sig->notify_count = zap_other_threads(tsk);
1060  	if (!thread_group_leader(tsk))
1061  		sig->notify_count--;
1062  
1063  	while (sig->notify_count) {
1064  		__set_current_state(TASK_KILLABLE);
1065  		spin_unlock_irq(lock);
1066  		schedule();
1067  		if (__fatal_signal_pending(tsk))
1068  			goto killed;
1069  		spin_lock_irq(lock);
1070  	}
1071  	spin_unlock_irq(lock);
1072  
1073  	/*
1074  	 * At this point all other threads have exited, all we have to
1075  	 * do is to wait for the thread group leader to become inactive,
1076  	 * and to assume its PID:
1077  	 */
1078  	if (!thread_group_leader(tsk)) {
1079  		struct task_struct *leader = tsk->group_leader;
1080  
1081  		for (;;) {
1082  			cgroup_threadgroup_change_begin(tsk);
1083  			write_lock_irq(&tasklist_lock);
1084  			/*
1085  			 * Do this under tasklist_lock to ensure that
1086  			 * exit_notify() can't miss ->group_exec_task
1087  			 */
1088  			sig->notify_count = -1;
1089  			if (likely(leader->exit_state))
1090  				break;
1091  			__set_current_state(TASK_KILLABLE);
1092  			write_unlock_irq(&tasklist_lock);
1093  			cgroup_threadgroup_change_end(tsk);
1094  			schedule();
1095  			if (__fatal_signal_pending(tsk))
1096  				goto killed;
1097  		}
1098  
1099  		/*
1100  		 * The only record we have of the real-time age of a
1101  		 * process, regardless of execs it's done, is start_time.
1102  		 * All the past CPU time is accumulated in signal_struct
1103  		 * from sister threads now dead.  But in this non-leader
1104  		 * exec, nothing survives from the original leader thread,
1105  		 * whose birth marks the true age of this process now.
1106  		 * When we take on its identity by switching to its PID, we
1107  		 * also take its birthdate (always earlier than our own).
1108  		 */
1109  		tsk->start_time = leader->start_time;
1110  		tsk->start_boottime = leader->start_boottime;
1111  
1112  		BUG_ON(!same_thread_group(leader, tsk));
1113  		/*
1114  		 * An exec() starts a new thread group with the
1115  		 * TGID of the previous thread group. Rehash the
1116  		 * two threads with a switched PID, and release
1117  		 * the former thread group leader:
1118  		 */
1119  
1120  		/* Become a process group leader with the old leader's pid.
1121  		 * The old leader becomes a thread of the this thread group.
1122  		 */
1123  		exchange_tids(tsk, leader);
1124  		transfer_pid(leader, tsk, PIDTYPE_TGID);
1125  		transfer_pid(leader, tsk, PIDTYPE_PGID);
1126  		transfer_pid(leader, tsk, PIDTYPE_SID);
1127  
1128  		list_replace_rcu(&leader->tasks, &tsk->tasks);
1129  		list_replace_init(&leader->sibling, &tsk->sibling);
1130  
1131  		tsk->group_leader = tsk;
1132  		leader->group_leader = tsk;
1133  
1134  		tsk->exit_signal = SIGCHLD;
1135  		leader->exit_signal = -1;
1136  
1137  		BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1138  		leader->exit_state = EXIT_DEAD;
1139  
1140  		/*
1141  		 * We are going to release_task()->ptrace_unlink() silently,
1142  		 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1143  		 * the tracer wont't block again waiting for this thread.
1144  		 */
1145  		if (unlikely(leader->ptrace))
1146  			__wake_up_parent(leader, leader->parent);
1147  		write_unlock_irq(&tasklist_lock);
1148  		cgroup_threadgroup_change_end(tsk);
1149  
1150  		release_task(leader);
1151  	}
1152  
1153  	sig->group_exec_task = NULL;
1154  	sig->notify_count = 0;
1155  
1156  no_thread_group:
1157  	/* we have changed execution domain */
1158  	tsk->exit_signal = SIGCHLD;
1159  
1160  	BUG_ON(!thread_group_leader(tsk));
1161  	return 0;
1162  
1163  killed:
1164  	/* protects against exit_notify() and __exit_signal() */
1165  	read_lock(&tasklist_lock);
1166  	sig->group_exec_task = NULL;
1167  	sig->notify_count = 0;
1168  	read_unlock(&tasklist_lock);
1169  	return -EAGAIN;
1170  }
1171  
1172  
1173  /*
1174   * This function makes sure the current process has its own signal table,
1175   * so that flush_signal_handlers can later reset the handlers without
1176   * disturbing other processes.  (Other processes might share the signal
1177   * table via the CLONE_SIGHAND option to clone().)
1178   */
1179  static int unshare_sighand(struct task_struct *me)
1180  {
1181  	struct sighand_struct *oldsighand = me->sighand;
1182  
1183  	if (refcount_read(&oldsighand->count) != 1) {
1184  		struct sighand_struct *newsighand;
1185  		/*
1186  		 * This ->sighand is shared with the CLONE_SIGHAND
1187  		 * but not CLONE_THREAD task, switch to the new one.
1188  		 */
1189  		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1190  		if (!newsighand)
1191  			return -ENOMEM;
1192  
1193  		refcount_set(&newsighand->count, 1);
1194  		memcpy(newsighand->action, oldsighand->action,
1195  		       sizeof(newsighand->action));
1196  
1197  		write_lock_irq(&tasklist_lock);
1198  		spin_lock(&oldsighand->siglock);
1199  		rcu_assign_pointer(me->sighand, newsighand);
1200  		spin_unlock(&oldsighand->siglock);
1201  		write_unlock_irq(&tasklist_lock);
1202  
1203  		__cleanup_sighand(oldsighand);
1204  	}
1205  	return 0;
1206  }
1207  
1208  char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1209  {
1210  	task_lock(tsk);
1211  	/* Always NUL terminated and zero-padded */
1212  	strscpy_pad(buf, tsk->comm, buf_size);
1213  	task_unlock(tsk);
1214  	return buf;
1215  }
1216  EXPORT_SYMBOL_GPL(__get_task_comm);
1217  
1218  /*
1219   * These functions flushes out all traces of the currently running executable
1220   * so that a new one can be started
1221   */
1222  
1223  void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1224  {
1225  	task_lock(tsk);
1226  	trace_task_rename(tsk, buf);
1227  	strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1228  	task_unlock(tsk);
1229  	perf_event_comm(tsk, exec);
1230  }
1231  
1232  /*
1233   * Calling this is the point of no return. None of the failures will be
1234   * seen by userspace since either the process is already taking a fatal
1235   * signal (via de_thread() or coredump), or will have SEGV raised
1236   * (after exec_mmap()) by search_binary_handler (see below).
1237   */
1238  int begin_new_exec(struct linux_binprm * bprm)
1239  {
1240  	struct task_struct *me = current;
1241  	int retval;
1242  
1243  	/* Once we are committed compute the creds */
1244  	retval = bprm_creds_from_file(bprm);
1245  	if (retval)
1246  		return retval;
1247  
1248  	/*
1249  	 * Ensure all future errors are fatal.
1250  	 */
1251  	bprm->point_of_no_return = true;
1252  
1253  	/*
1254  	 * Make this the only thread in the thread group.
1255  	 */
1256  	retval = de_thread(me);
1257  	if (retval)
1258  		goto out;
1259  
1260  	/*
1261  	 * Cancel any io_uring activity across execve
1262  	 */
1263  	io_uring_task_cancel();
1264  
1265  	/* Ensure the files table is not shared. */
1266  	retval = unshare_files();
1267  	if (retval)
1268  		goto out;
1269  
1270  	/*
1271  	 * Must be called _before_ exec_mmap() as bprm->mm is
1272  	 * not visibile until then. This also enables the update
1273  	 * to be lockless.
1274  	 */
1275  	retval = set_mm_exe_file(bprm->mm, bprm->file);
1276  	if (retval)
1277  		goto out;
1278  
1279  	/* If the binary is not readable then enforce mm->dumpable=0 */
1280  	would_dump(bprm, bprm->file);
1281  	if (bprm->have_execfd)
1282  		would_dump(bprm, bprm->executable);
1283  
1284  	/*
1285  	 * Release all of the old mmap stuff
1286  	 */
1287  	acct_arg_size(bprm, 0);
1288  	retval = exec_mmap(bprm->mm);
1289  	if (retval)
1290  		goto out;
1291  
1292  	bprm->mm = NULL;
1293  
1294  #ifdef CONFIG_POSIX_TIMERS
1295  	exit_itimers(me->signal);
1296  	flush_itimer_signals();
1297  #endif
1298  
1299  	/*
1300  	 * Make the signal table private.
1301  	 */
1302  	retval = unshare_sighand(me);
1303  	if (retval)
1304  		goto out_unlock;
1305  
1306  	/*
1307  	 * Ensure that the uaccess routines can actually operate on userspace
1308  	 * pointers:
1309  	 */
1310  	force_uaccess_begin();
1311  
1312  	if (me->flags & PF_KTHREAD)
1313  		free_kthread_struct(me);
1314  	me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1315  					PF_NOFREEZE | PF_NO_SETAFFINITY);
1316  	flush_thread();
1317  	me->personality &= ~bprm->per_clear;
1318  
1319  	clear_syscall_work_syscall_user_dispatch(me);
1320  
1321  	/*
1322  	 * We have to apply CLOEXEC before we change whether the process is
1323  	 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1324  	 * trying to access the should-be-closed file descriptors of a process
1325  	 * undergoing exec(2).
1326  	 */
1327  	do_close_on_exec(me->files);
1328  
1329  	if (bprm->secureexec) {
1330  		/* Make sure parent cannot signal privileged process. */
1331  		me->pdeath_signal = 0;
1332  
1333  		/*
1334  		 * For secureexec, reset the stack limit to sane default to
1335  		 * avoid bad behavior from the prior rlimits. This has to
1336  		 * happen before arch_pick_mmap_layout(), which examines
1337  		 * RLIMIT_STACK, but after the point of no return to avoid
1338  		 * needing to clean up the change on failure.
1339  		 */
1340  		if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1341  			bprm->rlim_stack.rlim_cur = _STK_LIM;
1342  	}
1343  
1344  	me->sas_ss_sp = me->sas_ss_size = 0;
1345  
1346  	/*
1347  	 * Figure out dumpability. Note that this checking only of current
1348  	 * is wrong, but userspace depends on it. This should be testing
1349  	 * bprm->secureexec instead.
1350  	 */
1351  	if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1352  	    !(uid_eq(current_euid(), current_uid()) &&
1353  	      gid_eq(current_egid(), current_gid())))
1354  		set_dumpable(current->mm, suid_dumpable);
1355  	else
1356  		set_dumpable(current->mm, SUID_DUMP_USER);
1357  
1358  	perf_event_exec();
1359  	__set_task_comm(me, kbasename(bprm->filename), true);
1360  
1361  	/* An exec changes our domain. We are no longer part of the thread
1362  	   group */
1363  	WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1364  	flush_signal_handlers(me, 0);
1365  
1366  	retval = set_cred_ucounts(bprm->cred);
1367  	if (retval < 0)
1368  		goto out_unlock;
1369  
1370  	/*
1371  	 * install the new credentials for this executable
1372  	 */
1373  	security_bprm_committing_creds(bprm);
1374  
1375  	commit_creds(bprm->cred);
1376  	bprm->cred = NULL;
1377  
1378  	/*
1379  	 * Disable monitoring for regular users
1380  	 * when executing setuid binaries. Must
1381  	 * wait until new credentials are committed
1382  	 * by commit_creds() above
1383  	 */
1384  	if (get_dumpable(me->mm) != SUID_DUMP_USER)
1385  		perf_event_exit_task(me);
1386  	/*
1387  	 * cred_guard_mutex must be held at least to this point to prevent
1388  	 * ptrace_attach() from altering our determination of the task's
1389  	 * credentials; any time after this it may be unlocked.
1390  	 */
1391  	security_bprm_committed_creds(bprm);
1392  
1393  	/* Pass the opened binary to the interpreter. */
1394  	if (bprm->have_execfd) {
1395  		retval = get_unused_fd_flags(0);
1396  		if (retval < 0)
1397  			goto out_unlock;
1398  		fd_install(retval, bprm->executable);
1399  		bprm->executable = NULL;
1400  		bprm->execfd = retval;
1401  	}
1402  	return 0;
1403  
1404  out_unlock:
1405  	up_write(&me->signal->exec_update_lock);
1406  out:
1407  	return retval;
1408  }
1409  EXPORT_SYMBOL(begin_new_exec);
1410  
1411  void would_dump(struct linux_binprm *bprm, struct file *file)
1412  {
1413  	struct inode *inode = file_inode(file);
1414  	struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1415  	if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1416  		struct user_namespace *old, *user_ns;
1417  		bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1418  
1419  		/* Ensure mm->user_ns contains the executable */
1420  		user_ns = old = bprm->mm->user_ns;
1421  		while ((user_ns != &init_user_ns) &&
1422  		       !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1423  			user_ns = user_ns->parent;
1424  
1425  		if (old != user_ns) {
1426  			bprm->mm->user_ns = get_user_ns(user_ns);
1427  			put_user_ns(old);
1428  		}
1429  	}
1430  }
1431  EXPORT_SYMBOL(would_dump);
1432  
1433  void setup_new_exec(struct linux_binprm * bprm)
1434  {
1435  	/* Setup things that can depend upon the personality */
1436  	struct task_struct *me = current;
1437  
1438  	arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1439  
1440  	arch_setup_new_exec();
1441  
1442  	/* Set the new mm task size. We have to do that late because it may
1443  	 * depend on TIF_32BIT which is only updated in flush_thread() on
1444  	 * some architectures like powerpc
1445  	 */
1446  	me->mm->task_size = TASK_SIZE;
1447  	up_write(&me->signal->exec_update_lock);
1448  	mutex_unlock(&me->signal->cred_guard_mutex);
1449  }
1450  EXPORT_SYMBOL(setup_new_exec);
1451  
1452  /* Runs immediately before start_thread() takes over. */
1453  void finalize_exec(struct linux_binprm *bprm)
1454  {
1455  	/* Store any stack rlimit changes before starting thread. */
1456  	task_lock(current->group_leader);
1457  	current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1458  	task_unlock(current->group_leader);
1459  }
1460  EXPORT_SYMBOL(finalize_exec);
1461  
1462  /*
1463   * Prepare credentials and lock ->cred_guard_mutex.
1464   * setup_new_exec() commits the new creds and drops the lock.
1465   * Or, if exec fails before, free_bprm() should release ->cred
1466   * and unlock.
1467   */
1468  static int prepare_bprm_creds(struct linux_binprm *bprm)
1469  {
1470  	if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1471  		return -ERESTARTNOINTR;
1472  
1473  	bprm->cred = prepare_exec_creds();
1474  	if (likely(bprm->cred))
1475  		return 0;
1476  
1477  	mutex_unlock(&current->signal->cred_guard_mutex);
1478  	return -ENOMEM;
1479  }
1480  
1481  static void free_bprm(struct linux_binprm *bprm)
1482  {
1483  	if (bprm->mm) {
1484  		acct_arg_size(bprm, 0);
1485  		mmput(bprm->mm);
1486  	}
1487  	free_arg_pages(bprm);
1488  	if (bprm->cred) {
1489  		mutex_unlock(&current->signal->cred_guard_mutex);
1490  		abort_creds(bprm->cred);
1491  	}
1492  	if (bprm->file) {
1493  		allow_write_access(bprm->file);
1494  		fput(bprm->file);
1495  	}
1496  	if (bprm->executable)
1497  		fput(bprm->executable);
1498  	/* If a binfmt changed the interp, free it. */
1499  	if (bprm->interp != bprm->filename)
1500  		kfree(bprm->interp);
1501  	kfree(bprm->fdpath);
1502  	kfree(bprm);
1503  }
1504  
1505  static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1506  {
1507  	struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1508  	int retval = -ENOMEM;
1509  	if (!bprm)
1510  		goto out;
1511  
1512  	if (fd == AT_FDCWD || filename->name[0] == '/') {
1513  		bprm->filename = filename->name;
1514  	} else {
1515  		if (filename->name[0] == '\0')
1516  			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1517  		else
1518  			bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1519  						  fd, filename->name);
1520  		if (!bprm->fdpath)
1521  			goto out_free;
1522  
1523  		bprm->filename = bprm->fdpath;
1524  	}
1525  	bprm->interp = bprm->filename;
1526  
1527  	retval = bprm_mm_init(bprm);
1528  	if (retval)
1529  		goto out_free;
1530  	return bprm;
1531  
1532  out_free:
1533  	free_bprm(bprm);
1534  out:
1535  	return ERR_PTR(retval);
1536  }
1537  
1538  int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1539  {
1540  	/* If a binfmt changed the interp, free it first. */
1541  	if (bprm->interp != bprm->filename)
1542  		kfree(bprm->interp);
1543  	bprm->interp = kstrdup(interp, GFP_KERNEL);
1544  	if (!bprm->interp)
1545  		return -ENOMEM;
1546  	return 0;
1547  }
1548  EXPORT_SYMBOL(bprm_change_interp);
1549  
1550  /*
1551   * determine how safe it is to execute the proposed program
1552   * - the caller must hold ->cred_guard_mutex to protect against
1553   *   PTRACE_ATTACH or seccomp thread-sync
1554   */
1555  static void check_unsafe_exec(struct linux_binprm *bprm)
1556  {
1557  	struct task_struct *p = current, *t;
1558  	unsigned n_fs;
1559  
1560  	if (p->ptrace)
1561  		bprm->unsafe |= LSM_UNSAFE_PTRACE;
1562  
1563  	/*
1564  	 * This isn't strictly necessary, but it makes it harder for LSMs to
1565  	 * mess up.
1566  	 */
1567  	if (task_no_new_privs(current))
1568  		bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1569  
1570  	t = p;
1571  	n_fs = 1;
1572  	spin_lock(&p->fs->lock);
1573  	rcu_read_lock();
1574  	while_each_thread(p, t) {
1575  		if (t->fs == p->fs)
1576  			n_fs++;
1577  	}
1578  	rcu_read_unlock();
1579  
1580  	if (p->fs->users > n_fs)
1581  		bprm->unsafe |= LSM_UNSAFE_SHARE;
1582  	else
1583  		p->fs->in_exec = 1;
1584  	spin_unlock(&p->fs->lock);
1585  }
1586  
1587  static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1588  {
1589  	/* Handle suid and sgid on files */
1590  	struct user_namespace *mnt_userns;
1591  	struct inode *inode;
1592  	unsigned int mode;
1593  	kuid_t uid;
1594  	kgid_t gid;
1595  
1596  	if (!mnt_may_suid(file->f_path.mnt))
1597  		return;
1598  
1599  	if (task_no_new_privs(current))
1600  		return;
1601  
1602  	inode = file->f_path.dentry->d_inode;
1603  	mode = READ_ONCE(inode->i_mode);
1604  	if (!(mode & (S_ISUID|S_ISGID)))
1605  		return;
1606  
1607  	mnt_userns = file_mnt_user_ns(file);
1608  
1609  	/* Be careful if suid/sgid is set */
1610  	inode_lock(inode);
1611  
1612  	/* reload atomically mode/uid/gid now that lock held */
1613  	mode = inode->i_mode;
1614  	uid = i_uid_into_mnt(mnt_userns, inode);
1615  	gid = i_gid_into_mnt(mnt_userns, inode);
1616  	inode_unlock(inode);
1617  
1618  	/* We ignore suid/sgid if there are no mappings for them in the ns */
1619  	if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1620  		 !kgid_has_mapping(bprm->cred->user_ns, gid))
1621  		return;
1622  
1623  	if (mode & S_ISUID) {
1624  		bprm->per_clear |= PER_CLEAR_ON_SETID;
1625  		bprm->cred->euid = uid;
1626  	}
1627  
1628  	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1629  		bprm->per_clear |= PER_CLEAR_ON_SETID;
1630  		bprm->cred->egid = gid;
1631  	}
1632  }
1633  
1634  /*
1635   * Compute brpm->cred based upon the final binary.
1636   */
1637  static int bprm_creds_from_file(struct linux_binprm *bprm)
1638  {
1639  	/* Compute creds based on which file? */
1640  	struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1641  
1642  	bprm_fill_uid(bprm, file);
1643  	return security_bprm_creds_from_file(bprm, file);
1644  }
1645  
1646  /*
1647   * Fill the binprm structure from the inode.
1648   * Read the first BINPRM_BUF_SIZE bytes
1649   *
1650   * This may be called multiple times for binary chains (scripts for example).
1651   */
1652  static int prepare_binprm(struct linux_binprm *bprm)
1653  {
1654  	loff_t pos = 0;
1655  
1656  	memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1657  	return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1658  }
1659  
1660  /*
1661   * Arguments are '\0' separated strings found at the location bprm->p
1662   * points to; chop off the first by relocating brpm->p to right after
1663   * the first '\0' encountered.
1664   */
1665  int remove_arg_zero(struct linux_binprm *bprm)
1666  {
1667  	int ret = 0;
1668  	unsigned long offset;
1669  	char *kaddr;
1670  	struct page *page;
1671  
1672  	if (!bprm->argc)
1673  		return 0;
1674  
1675  	do {
1676  		offset = bprm->p & ~PAGE_MASK;
1677  		page = get_arg_page(bprm, bprm->p, 0);
1678  		if (!page) {
1679  			ret = -EFAULT;
1680  			goto out;
1681  		}
1682  		kaddr = kmap_atomic(page);
1683  
1684  		for (; offset < PAGE_SIZE && kaddr[offset];
1685  				offset++, bprm->p++)
1686  			;
1687  
1688  		kunmap_atomic(kaddr);
1689  		put_arg_page(page);
1690  	} while (offset == PAGE_SIZE);
1691  
1692  	bprm->p++;
1693  	bprm->argc--;
1694  	ret = 0;
1695  
1696  out:
1697  	return ret;
1698  }
1699  EXPORT_SYMBOL(remove_arg_zero);
1700  
1701  #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1702  /*
1703   * cycle the list of binary formats handler, until one recognizes the image
1704   */
1705  static int search_binary_handler(struct linux_binprm *bprm)
1706  {
1707  	bool need_retry = IS_ENABLED(CONFIG_MODULES);
1708  	struct linux_binfmt *fmt;
1709  	int retval;
1710  
1711  	retval = prepare_binprm(bprm);
1712  	if (retval < 0)
1713  		return retval;
1714  
1715  	retval = security_bprm_check(bprm);
1716  	if (retval)
1717  		return retval;
1718  
1719  	retval = -ENOENT;
1720   retry:
1721  	read_lock(&binfmt_lock);
1722  	list_for_each_entry(fmt, &formats, lh) {
1723  		if (!try_module_get(fmt->module))
1724  			continue;
1725  		read_unlock(&binfmt_lock);
1726  
1727  		retval = fmt->load_binary(bprm);
1728  
1729  		read_lock(&binfmt_lock);
1730  		put_binfmt(fmt);
1731  		if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1732  			read_unlock(&binfmt_lock);
1733  			return retval;
1734  		}
1735  	}
1736  	read_unlock(&binfmt_lock);
1737  
1738  	if (need_retry) {
1739  		if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1740  		    printable(bprm->buf[2]) && printable(bprm->buf[3]))
1741  			return retval;
1742  		if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1743  			return retval;
1744  		need_retry = false;
1745  		goto retry;
1746  	}
1747  
1748  	return retval;
1749  }
1750  
1751  static int exec_binprm(struct linux_binprm *bprm)
1752  {
1753  	pid_t old_pid, old_vpid;
1754  	int ret, depth;
1755  
1756  	/* Need to fetch pid before load_binary changes it */
1757  	old_pid = current->pid;
1758  	rcu_read_lock();
1759  	old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1760  	rcu_read_unlock();
1761  
1762  	/* This allows 4 levels of binfmt rewrites before failing hard. */
1763  	for (depth = 0;; depth++) {
1764  		struct file *exec;
1765  		if (depth > 5)
1766  			return -ELOOP;
1767  
1768  		ret = search_binary_handler(bprm);
1769  		if (ret < 0)
1770  			return ret;
1771  		if (!bprm->interpreter)
1772  			break;
1773  
1774  		exec = bprm->file;
1775  		bprm->file = bprm->interpreter;
1776  		bprm->interpreter = NULL;
1777  
1778  		allow_write_access(exec);
1779  		if (unlikely(bprm->have_execfd)) {
1780  			if (bprm->executable) {
1781  				fput(exec);
1782  				return -ENOEXEC;
1783  			}
1784  			bprm->executable = exec;
1785  		} else
1786  			fput(exec);
1787  	}
1788  
1789  	audit_bprm(bprm);
1790  	trace_sched_process_exec(current, old_pid, bprm);
1791  	ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1792  	proc_exec_connector(current);
1793  	return 0;
1794  }
1795  
1796  /*
1797   * sys_execve() executes a new program.
1798   */
1799  static int bprm_execve(struct linux_binprm *bprm,
1800  		       int fd, struct filename *filename, int flags)
1801  {
1802  	struct file *file;
1803  	int retval;
1804  
1805  	retval = prepare_bprm_creds(bprm);
1806  	if (retval)
1807  		return retval;
1808  
1809  	check_unsafe_exec(bprm);
1810  	current->in_execve = 1;
1811  
1812  	file = do_open_execat(fd, filename, flags);
1813  	retval = PTR_ERR(file);
1814  	if (IS_ERR(file))
1815  		goto out_unmark;
1816  
1817  	sched_exec();
1818  
1819  	bprm->file = file;
1820  	/*
1821  	 * Record that a name derived from an O_CLOEXEC fd will be
1822  	 * inaccessible after exec.  This allows the code in exec to
1823  	 * choose to fail when the executable is not mmaped into the
1824  	 * interpreter and an open file descriptor is not passed to
1825  	 * the interpreter.  This makes for a better user experience
1826  	 * than having the interpreter start and then immediately fail
1827  	 * when it finds the executable is inaccessible.
1828  	 */
1829  	if (bprm->fdpath && get_close_on_exec(fd))
1830  		bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1831  
1832  	/* Set the unchanging part of bprm->cred */
1833  	retval = security_bprm_creds_for_exec(bprm);
1834  	if (retval)
1835  		goto out;
1836  
1837  	retval = exec_binprm(bprm);
1838  	if (retval < 0)
1839  		goto out;
1840  
1841  	/* execve succeeded */
1842  	current->fs->in_exec = 0;
1843  	current->in_execve = 0;
1844  	rseq_execve(current);
1845  	acct_update_integrals(current);
1846  	task_numa_free(current, false);
1847  	return retval;
1848  
1849  out:
1850  	/*
1851  	 * If past the point of no return ensure the code never
1852  	 * returns to the userspace process.  Use an existing fatal
1853  	 * signal if present otherwise terminate the process with
1854  	 * SIGSEGV.
1855  	 */
1856  	if (bprm->point_of_no_return && !fatal_signal_pending(current))
1857  		force_fatal_sig(SIGSEGV);
1858  
1859  out_unmark:
1860  	current->fs->in_exec = 0;
1861  	current->in_execve = 0;
1862  
1863  	return retval;
1864  }
1865  
1866  static int do_execveat_common(int fd, struct filename *filename,
1867  			      struct user_arg_ptr argv,
1868  			      struct user_arg_ptr envp,
1869  			      int flags)
1870  {
1871  	struct linux_binprm *bprm;
1872  	int retval;
1873  
1874  	if (IS_ERR(filename))
1875  		return PTR_ERR(filename);
1876  
1877  	/*
1878  	 * We move the actual failure in case of RLIMIT_NPROC excess from
1879  	 * set*uid() to execve() because too many poorly written programs
1880  	 * don't check setuid() return code.  Here we additionally recheck
1881  	 * whether NPROC limit is still exceeded.
1882  	 */
1883  	if ((current->flags & PF_NPROC_EXCEEDED) &&
1884  	    is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1885  		retval = -EAGAIN;
1886  		goto out_ret;
1887  	}
1888  
1889  	/* We're below the limit (still or again), so we don't want to make
1890  	 * further execve() calls fail. */
1891  	current->flags &= ~PF_NPROC_EXCEEDED;
1892  
1893  	bprm = alloc_bprm(fd, filename);
1894  	if (IS_ERR(bprm)) {
1895  		retval = PTR_ERR(bprm);
1896  		goto out_ret;
1897  	}
1898  
1899  	retval = count(argv, MAX_ARG_STRINGS);
1900  	if (retval < 0)
1901  		goto out_free;
1902  	bprm->argc = retval;
1903  
1904  	retval = count(envp, MAX_ARG_STRINGS);
1905  	if (retval < 0)
1906  		goto out_free;
1907  	bprm->envc = retval;
1908  
1909  	retval = bprm_stack_limits(bprm);
1910  	if (retval < 0)
1911  		goto out_free;
1912  
1913  	retval = copy_string_kernel(bprm->filename, bprm);
1914  	if (retval < 0)
1915  		goto out_free;
1916  	bprm->exec = bprm->p;
1917  
1918  	retval = copy_strings(bprm->envc, envp, bprm);
1919  	if (retval < 0)
1920  		goto out_free;
1921  
1922  	retval = copy_strings(bprm->argc, argv, bprm);
1923  	if (retval < 0)
1924  		goto out_free;
1925  
1926  	retval = bprm_execve(bprm, fd, filename, flags);
1927  out_free:
1928  	free_bprm(bprm);
1929  
1930  out_ret:
1931  	putname(filename);
1932  	return retval;
1933  }
1934  
1935  int kernel_execve(const char *kernel_filename,
1936  		  const char *const *argv, const char *const *envp)
1937  {
1938  	struct filename *filename;
1939  	struct linux_binprm *bprm;
1940  	int fd = AT_FDCWD;
1941  	int retval;
1942  
1943  	filename = getname_kernel(kernel_filename);
1944  	if (IS_ERR(filename))
1945  		return PTR_ERR(filename);
1946  
1947  	bprm = alloc_bprm(fd, filename);
1948  	if (IS_ERR(bprm)) {
1949  		retval = PTR_ERR(bprm);
1950  		goto out_ret;
1951  	}
1952  
1953  	retval = count_strings_kernel(argv);
1954  	if (retval < 0)
1955  		goto out_free;
1956  	bprm->argc = retval;
1957  
1958  	retval = count_strings_kernel(envp);
1959  	if (retval < 0)
1960  		goto out_free;
1961  	bprm->envc = retval;
1962  
1963  	retval = bprm_stack_limits(bprm);
1964  	if (retval < 0)
1965  		goto out_free;
1966  
1967  	retval = copy_string_kernel(bprm->filename, bprm);
1968  	if (retval < 0)
1969  		goto out_free;
1970  	bprm->exec = bprm->p;
1971  
1972  	retval = copy_strings_kernel(bprm->envc, envp, bprm);
1973  	if (retval < 0)
1974  		goto out_free;
1975  
1976  	retval = copy_strings_kernel(bprm->argc, argv, bprm);
1977  	if (retval < 0)
1978  		goto out_free;
1979  
1980  	retval = bprm_execve(bprm, fd, filename, 0);
1981  out_free:
1982  	free_bprm(bprm);
1983  out_ret:
1984  	putname(filename);
1985  	return retval;
1986  }
1987  
1988  static int do_execve(struct filename *filename,
1989  	const char __user *const __user *__argv,
1990  	const char __user *const __user *__envp)
1991  {
1992  	struct user_arg_ptr argv = { .ptr.native = __argv };
1993  	struct user_arg_ptr envp = { .ptr.native = __envp };
1994  	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1995  }
1996  
1997  static int do_execveat(int fd, struct filename *filename,
1998  		const char __user *const __user *__argv,
1999  		const char __user *const __user *__envp,
2000  		int flags)
2001  {
2002  	struct user_arg_ptr argv = { .ptr.native = __argv };
2003  	struct user_arg_ptr envp = { .ptr.native = __envp };
2004  
2005  	return do_execveat_common(fd, filename, argv, envp, flags);
2006  }
2007  
2008  #ifdef CONFIG_COMPAT
2009  static int compat_do_execve(struct filename *filename,
2010  	const compat_uptr_t __user *__argv,
2011  	const compat_uptr_t __user *__envp)
2012  {
2013  	struct user_arg_ptr argv = {
2014  		.is_compat = true,
2015  		.ptr.compat = __argv,
2016  	};
2017  	struct user_arg_ptr envp = {
2018  		.is_compat = true,
2019  		.ptr.compat = __envp,
2020  	};
2021  	return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2022  }
2023  
2024  static int compat_do_execveat(int fd, struct filename *filename,
2025  			      const compat_uptr_t __user *__argv,
2026  			      const compat_uptr_t __user *__envp,
2027  			      int flags)
2028  {
2029  	struct user_arg_ptr argv = {
2030  		.is_compat = true,
2031  		.ptr.compat = __argv,
2032  	};
2033  	struct user_arg_ptr envp = {
2034  		.is_compat = true,
2035  		.ptr.compat = __envp,
2036  	};
2037  	return do_execveat_common(fd, filename, argv, envp, flags);
2038  }
2039  #endif
2040  
2041  void set_binfmt(struct linux_binfmt *new)
2042  {
2043  	struct mm_struct *mm = current->mm;
2044  
2045  	if (mm->binfmt)
2046  		module_put(mm->binfmt->module);
2047  
2048  	mm->binfmt = new;
2049  	if (new)
2050  		__module_get(new->module);
2051  }
2052  EXPORT_SYMBOL(set_binfmt);
2053  
2054  /*
2055   * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2056   */
2057  void set_dumpable(struct mm_struct *mm, int value)
2058  {
2059  	if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2060  		return;
2061  
2062  	set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2063  }
2064  
2065  SYSCALL_DEFINE3(execve,
2066  		const char __user *, filename,
2067  		const char __user *const __user *, argv,
2068  		const char __user *const __user *, envp)
2069  {
2070  	return do_execve(getname(filename), argv, envp);
2071  }
2072  
2073  SYSCALL_DEFINE5(execveat,
2074  		int, fd, const char __user *, filename,
2075  		const char __user *const __user *, argv,
2076  		const char __user *const __user *, envp,
2077  		int, flags)
2078  {
2079  	return do_execveat(fd,
2080  			   getname_uflags(filename, flags),
2081  			   argv, envp, flags);
2082  }
2083  
2084  #ifdef CONFIG_COMPAT
2085  COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2086  	const compat_uptr_t __user *, argv,
2087  	const compat_uptr_t __user *, envp)
2088  {
2089  	return compat_do_execve(getname(filename), argv, envp);
2090  }
2091  
2092  COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2093  		       const char __user *, filename,
2094  		       const compat_uptr_t __user *, argv,
2095  		       const compat_uptr_t __user *, envp,
2096  		       int,  flags)
2097  {
2098  	return compat_do_execveat(fd,
2099  				  getname_uflags(filename, flags),
2100  				  argv, envp, flags);
2101  }
2102  #endif
2103  
2104  #ifdef CONFIG_SYSCTL
2105  
2106  static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2107  		void *buffer, size_t *lenp, loff_t *ppos)
2108  {
2109  	int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2110  
2111  	if (!error)
2112  		validate_coredump_safety();
2113  	return error;
2114  }
2115  
2116  static struct ctl_table fs_exec_sysctls[] = {
2117  	{
2118  		.procname	= "suid_dumpable",
2119  		.data		= &suid_dumpable,
2120  		.maxlen		= sizeof(int),
2121  		.mode		= 0644,
2122  		.proc_handler	= proc_dointvec_minmax_coredump,
2123  		.extra1		= SYSCTL_ZERO,
2124  		.extra2		= SYSCTL_TWO,
2125  	},
2126  	{ }
2127  };
2128  
2129  static int __init init_fs_exec_sysctls(void)
2130  {
2131  	register_sysctl_init("fs", fs_exec_sysctls);
2132  	return 0;
2133  }
2134  
2135  fs_initcall(init_fs_exec_sysctls);
2136  #endif /* CONFIG_SYSCTL */
2137