]> Git Repo - linux.git/blob - fs/exec.c
fsnotify: move fsnotify_open() hook into do_dentry_open()
[linux.git] / fs / exec.c
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         int ret;
202         unsigned int gup_flags = 0;
203
204 #ifdef CONFIG_STACK_GROWSUP
205         if (write) {
206                 ret = expand_downwards(bprm->vma, pos);
207                 if (ret < 0)
208                         return NULL;
209         }
210 #endif
211
212         if (write)
213                 gup_flags |= FOLL_WRITE;
214
215         /*
216          * We are doing an exec().  'current' is the process
217          * doing the exec and bprm->mm is the new process's mm.
218          */
219         mmap_read_lock(bprm->mm);
220         ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
221                         &page, NULL, NULL);
222         mmap_read_unlock(bprm->mm);
223         if (ret <= 0)
224                 return NULL;
225
226         if (write)
227                 acct_arg_size(bprm, vma_pages(bprm->vma));
228
229         return page;
230 }
231
232 static void put_arg_page(struct page *page)
233 {
234         put_page(page);
235 }
236
237 static void free_arg_pages(struct linux_binprm *bprm)
238 {
239 }
240
241 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
242                 struct page *page)
243 {
244         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
245 }
246
247 static int __bprm_mm_init(struct linux_binprm *bprm)
248 {
249         int err;
250         struct vm_area_struct *vma = NULL;
251         struct mm_struct *mm = bprm->mm;
252
253         bprm->vma = vma = vm_area_alloc(mm);
254         if (!vma)
255                 return -ENOMEM;
256         vma_set_anonymous(vma);
257
258         if (mmap_write_lock_killable(mm)) {
259                 err = -EINTR;
260                 goto err_free;
261         }
262
263         /*
264          * Place the stack at the largest stack address the architecture
265          * supports. Later, we'll move this to an appropriate place. We don't
266          * use STACK_TOP because that can depend on attributes which aren't
267          * configured yet.
268          */
269         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
270         vma->vm_end = STACK_TOP_MAX;
271         vma->vm_start = vma->vm_end - PAGE_SIZE;
272         vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);
273         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
274
275         err = insert_vm_struct(mm, vma);
276         if (err)
277                 goto err;
278
279         mm->stack_vm = mm->total_vm = 1;
280         mmap_write_unlock(mm);
281         bprm->p = vma->vm_end - sizeof(void *);
282         return 0;
283 err:
284         mmap_write_unlock(mm);
285 err_free:
286         bprm->vma = NULL;
287         vm_area_free(vma);
288         return err;
289 }
290
291 static bool valid_arg_len(struct linux_binprm *bprm, long len)
292 {
293         return len <= MAX_ARG_STRLEN;
294 }
295
296 #else
297
298 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
299 {
300 }
301
302 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
303                 int write)
304 {
305         struct page *page;
306
307         page = bprm->page[pos / PAGE_SIZE];
308         if (!page && write) {
309                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
310                 if (!page)
311                         return NULL;
312                 bprm->page[pos / PAGE_SIZE] = page;
313         }
314
315         return page;
316 }
317
318 static void put_arg_page(struct page *page)
319 {
320 }
321
322 static void free_arg_page(struct linux_binprm *bprm, int i)
323 {
324         if (bprm->page[i]) {
325                 __free_page(bprm->page[i]);
326                 bprm->page[i] = NULL;
327         }
328 }
329
330 static void free_arg_pages(struct linux_binprm *bprm)
331 {
332         int i;
333
334         for (i = 0; i < MAX_ARG_PAGES; i++)
335                 free_arg_page(bprm, i);
336 }
337
338 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
339                 struct page *page)
340 {
341 }
342
343 static int __bprm_mm_init(struct linux_binprm *bprm)
344 {
345         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
346         return 0;
347 }
348
349 static bool valid_arg_len(struct linux_binprm *bprm, long len)
350 {
351         return len <= bprm->p;
352 }
353
354 #endif /* CONFIG_MMU */
355
356 /*
357  * Create a new mm_struct and populate it with a temporary stack
358  * vm_area_struct.  We don't have enough context at this point to set the stack
359  * flags, permissions, and offset, so we use temporary values.  We'll update
360  * them later in setup_arg_pages().
361  */
362 static int bprm_mm_init(struct linux_binprm *bprm)
363 {
364         int err;
365         struct mm_struct *mm = NULL;
366
367         bprm->mm = mm = mm_alloc();
368         err = -ENOMEM;
369         if (!mm)
370                 goto err;
371
372         /* Save current stack limit for all calculations made during exec. */
373         task_lock(current->group_leader);
374         bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
375         task_unlock(current->group_leader);
376
377         err = __bprm_mm_init(bprm);
378         if (err)
379                 goto err;
380
381         return 0;
382
383 err:
384         if (mm) {
385                 bprm->mm = NULL;
386                 mmdrop(mm);
387         }
388
389         return err;
390 }
391
392 struct user_arg_ptr {
393 #ifdef CONFIG_COMPAT
394         bool is_compat;
395 #endif
396         union {
397                 const char __user *const __user *native;
398 #ifdef CONFIG_COMPAT
399                 const compat_uptr_t __user *compat;
400 #endif
401         } ptr;
402 };
403
404 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
405 {
406         const char __user *native;
407
408 #ifdef CONFIG_COMPAT
409         if (unlikely(argv.is_compat)) {
410                 compat_uptr_t compat;
411
412                 if (get_user(compat, argv.ptr.compat + nr))
413                         return ERR_PTR(-EFAULT);
414
415                 return compat_ptr(compat);
416         }
417 #endif
418
419         if (get_user(native, argv.ptr.native + nr))
420                 return ERR_PTR(-EFAULT);
421
422         return native;
423 }
424
425 /*
426  * count() counts the number of strings in array ARGV.
427  */
428 static int count(struct user_arg_ptr argv, int max)
429 {
430         int i = 0;
431
432         if (argv.ptr.native != NULL) {
433                 for (;;) {
434                         const char __user *p = get_user_arg_ptr(argv, i);
435
436                         if (!p)
437                                 break;
438
439                         if (IS_ERR(p))
440                                 return -EFAULT;
441
442                         if (i >= max)
443                                 return -E2BIG;
444                         ++i;
445
446                         if (fatal_signal_pending(current))
447                                 return -ERESTARTNOHAND;
448                         cond_resched();
449                 }
450         }
451         return i;
452 }
453
454 static int count_strings_kernel(const char *const *argv)
455 {
456         int i;
457
458         if (!argv)
459                 return 0;
460
461         for (i = 0; argv[i]; ++i) {
462                 if (i >= MAX_ARG_STRINGS)
463                         return -E2BIG;
464                 if (fatal_signal_pending(current))
465                         return -ERESTARTNOHAND;
466                 cond_resched();
467         }
468         return i;
469 }
470
471 static int bprm_stack_limits(struct linux_binprm *bprm)
472 {
473         unsigned long limit, ptr_size;
474
475         /*
476          * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
477          * (whichever is smaller) for the argv+env strings.
478          * This ensures that:
479          *  - the remaining binfmt code will not run out of stack space,
480          *  - the program will have a reasonable amount of stack left
481          *    to work from.
482          */
483         limit = _STK_LIM / 4 * 3;
484         limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
485         /*
486          * We've historically supported up to 32 pages (ARG_MAX)
487          * of argument strings even with small stacks
488          */
489         limit = max_t(unsigned long, limit, ARG_MAX);
490         /*
491          * We must account for the size of all the argv and envp pointers to
492          * the argv and envp strings, since they will also take up space in
493          * the stack. They aren't stored until much later when we can't
494          * signal to the parent that the child has run out of stack space.
495          * Instead, calculate it here so it's possible to fail gracefully.
496          *
497          * In the case of argc = 0, make sure there is space for adding a
498          * empty string (which will bump argc to 1), to ensure confused
499          * userspace programs don't start processing from argv[1], thinking
500          * argc can never be 0, to keep them from walking envp by accident.
501          * See do_execveat_common().
502          */
503         ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
504         if (limit <= ptr_size)
505                 return -E2BIG;
506         limit -= ptr_size;
507
508         bprm->argmin = bprm->p - limit;
509         return 0;
510 }
511
512 /*
513  * 'copy_strings()' copies argument/environment strings from the old
514  * processes's memory to the new process's stack.  The call to get_user_pages()
515  * ensures the destination page is created and not swapped out.
516  */
517 static int copy_strings(int argc, struct user_arg_ptr argv,
518                         struct linux_binprm *bprm)
519 {
520         struct page *kmapped_page = NULL;
521         char *kaddr = NULL;
522         unsigned long kpos = 0;
523         int ret;
524
525         while (argc-- > 0) {
526                 const char __user *str;
527                 int len;
528                 unsigned long pos;
529
530                 ret = -EFAULT;
531                 str = get_user_arg_ptr(argv, argc);
532                 if (IS_ERR(str))
533                         goto out;
534
535                 len = strnlen_user(str, MAX_ARG_STRLEN);
536                 if (!len)
537                         goto out;
538
539                 ret = -E2BIG;
540                 if (!valid_arg_len(bprm, len))
541                         goto out;
542
543                 /* We're going to work our way backwards. */
544                 pos = bprm->p;
545                 str += len;
546                 bprm->p -= len;
547 #ifdef CONFIG_MMU
548                 if (bprm->p < bprm->argmin)
549                         goto out;
550 #endif
551
552                 while (len > 0) {
553                         int offset, bytes_to_copy;
554
555                         if (fatal_signal_pending(current)) {
556                                 ret = -ERESTARTNOHAND;
557                                 goto out;
558                         }
559                         cond_resched();
560
561                         offset = pos % PAGE_SIZE;
562                         if (offset == 0)
563                                 offset = PAGE_SIZE;
564
565                         bytes_to_copy = offset;
566                         if (bytes_to_copy > len)
567                                 bytes_to_copy = len;
568
569                         offset -= bytes_to_copy;
570                         pos -= bytes_to_copy;
571                         str -= bytes_to_copy;
572                         len -= bytes_to_copy;
573
574                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
575                                 struct page *page;
576
577                                 page = get_arg_page(bprm, pos, 1);
578                                 if (!page) {
579                                         ret = -E2BIG;
580                                         goto out;
581                                 }
582
583                                 if (kmapped_page) {
584                                         flush_dcache_page(kmapped_page);
585                                         kunmap_local(kaddr);
586                                         put_arg_page(kmapped_page);
587                                 }
588                                 kmapped_page = page;
589                                 kaddr = kmap_local_page(kmapped_page);
590                                 kpos = pos & PAGE_MASK;
591                                 flush_arg_page(bprm, kpos, kmapped_page);
592                         }
593                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
594                                 ret = -EFAULT;
595                                 goto out;
596                         }
597                 }
598         }
599         ret = 0;
600 out:
601         if (kmapped_page) {
602                 flush_dcache_page(kmapped_page);
603                 kunmap_local(kaddr);
604                 put_arg_page(kmapped_page);
605         }
606         return ret;
607 }
608
609 /*
610  * Copy and argument/environment string from the kernel to the processes stack.
611  */
612 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
613 {
614         int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
615         unsigned long pos = bprm->p;
616
617         if (len == 0)
618                 return -EFAULT;
619         if (!valid_arg_len(bprm, len))
620                 return -E2BIG;
621
622         /* We're going to work our way backwards. */
623         arg += len;
624         bprm->p -= len;
625         if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
626                 return -E2BIG;
627
628         while (len > 0) {
629                 unsigned int bytes_to_copy = min_t(unsigned int, len,
630                                 min_not_zero(offset_in_page(pos), PAGE_SIZE));
631                 struct page *page;
632
633                 pos -= bytes_to_copy;
634                 arg -= bytes_to_copy;
635                 len -= bytes_to_copy;
636
637                 page = get_arg_page(bprm, pos, 1);
638                 if (!page)
639                         return -E2BIG;
640                 flush_arg_page(bprm, pos & PAGE_MASK, page);
641                 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
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         VMA_ITERATOR(vmi, mm, new_start);
686         struct vm_area_struct *next;
687         struct mmu_gather tlb;
688
689         BUG_ON(new_start > new_end);
690
691         /*
692          * ensure there are no vmas between where we want to go
693          * and where we are
694          */
695         if (vma != vma_next(&vmi))
696                 return -EFAULT;
697
698         /*
699          * cover the whole range: [new_start, old_end)
700          */
701         if (vma_expand(&vmi, vma, new_start, old_end, vma->vm_pgoff, NULL))
702                 return -ENOMEM;
703
704         /*
705          * move the page tables downwards, on failure we rely on
706          * process cleanup to remove whatever mess we made.
707          */
708         if (length != move_page_tables(vma, old_start,
709                                        vma, new_start, length, false))
710                 return -ENOMEM;
711
712         lru_add_drain();
713         tlb_gather_mmu(&tlb, mm);
714         next = vma_next(&vmi);
715         if (new_end > old_start) {
716                 /*
717                  * when the old and new regions overlap clear from new_end.
718                  */
719                 free_pgd_range(&tlb, new_end, old_end, new_end,
720                         next ? next->vm_start : USER_PGTABLES_CEILING);
721         } else {
722                 /*
723                  * otherwise, clean from old_start; this is done to not touch
724                  * the address space in [new_end, old_start) some architectures
725                  * have constraints on va-space that make this illegal (IA64) -
726                  * for the others its just a little faster.
727                  */
728                 free_pgd_range(&tlb, old_start, old_end, new_end,
729                         next ? next->vm_start : USER_PGTABLES_CEILING);
730         }
731         tlb_finish_mmu(&tlb);
732
733         vma_prev(&vmi);
734         /* Shrink the vma to just the new range */
735         return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff);
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         struct mmu_gather tlb;
757         struct vma_iterator vmi;
758
759 #ifdef CONFIG_STACK_GROWSUP
760         /* Limit stack size */
761         stack_base = bprm->rlim_stack.rlim_max;
762
763         stack_base = calc_max_stack_size(stack_base);
764
765         /* Add space for stack randomization. */
766         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
767
768         /* Make sure we didn't let the argument array grow too large. */
769         if (vma->vm_end - vma->vm_start > stack_base)
770                 return -ENOMEM;
771
772         stack_base = PAGE_ALIGN(stack_top - stack_base);
773
774         stack_shift = vma->vm_start - stack_base;
775         mm->arg_start = bprm->p - stack_shift;
776         bprm->p = vma->vm_end - stack_shift;
777 #else
778         stack_top = arch_align_stack(stack_top);
779         stack_top = PAGE_ALIGN(stack_top);
780
781         if (unlikely(stack_top < mmap_min_addr) ||
782             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
783                 return -ENOMEM;
784
785         stack_shift = vma->vm_end - stack_top;
786
787         bprm->p -= stack_shift;
788         mm->arg_start = bprm->p;
789 #endif
790
791         if (bprm->loader)
792                 bprm->loader -= stack_shift;
793         bprm->exec -= stack_shift;
794
795         if (mmap_write_lock_killable(mm))
796                 return -EINTR;
797
798         vm_flags = VM_STACK_FLAGS;
799
800         /*
801          * Adjust stack execute permissions; explicitly enable for
802          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
803          * (arch default) otherwise.
804          */
805         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
806                 vm_flags |= VM_EXEC;
807         else if (executable_stack == EXSTACK_DISABLE_X)
808                 vm_flags &= ~VM_EXEC;
809         vm_flags |= mm->def_flags;
810         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
811
812         vma_iter_init(&vmi, mm, vma->vm_start);
813
814         tlb_gather_mmu(&tlb, mm);
815         ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
816                         vm_flags);
817         tlb_finish_mmu(&tlb);
818
819         if (ret)
820                 goto out_unlock;
821         BUG_ON(prev != vma);
822
823         if (unlikely(vm_flags & VM_EXEC)) {
824                 pr_warn_once("process '%pD4' started with executable stack\n",
825                              bprm->file);
826         }
827
828         /* Move stack pages down in memory. */
829         if (stack_shift) {
830                 ret = shift_arg_pages(vma, stack_shift);
831                 if (ret)
832                         goto out_unlock;
833         }
834
835         /* mprotect_fixup is overkill to remove the temporary stack flags */
836         vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
837
838         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
839         stack_size = vma->vm_end - vma->vm_start;
840         /*
841          * Align this down to a page boundary as expand_stack
842          * will align it up.
843          */
844         rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
845
846         stack_expand = min(rlim_stack, stack_size + stack_expand);
847
848 #ifdef CONFIG_STACK_GROWSUP
849         stack_base = vma->vm_start + stack_expand;
850 #else
851         stack_base = vma->vm_end - stack_expand;
852 #endif
853         current->mm->start_stack = bprm->p;
854         ret = expand_stack(vma, stack_base);
855         if (ret)
856                 ret = -EFAULT;
857
858 out_unlock:
859         mmap_write_unlock(mm);
860         return ret;
861 }
862 EXPORT_SYMBOL(setup_arg_pages);
863
864 #else
865
866 /*
867  * Transfer the program arguments and environment from the holding pages
868  * onto the stack. The provided stack pointer is adjusted accordingly.
869  */
870 int transfer_args_to_stack(struct linux_binprm *bprm,
871                            unsigned long *sp_location)
872 {
873         unsigned long index, stop, sp;
874         int ret = 0;
875
876         stop = bprm->p >> PAGE_SHIFT;
877         sp = *sp_location;
878
879         for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
880                 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
881                 char *src = kmap_local_page(bprm->page[index]) + offset;
882                 sp -= PAGE_SIZE - offset;
883                 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
884                         ret = -EFAULT;
885                 kunmap_local(src);
886                 if (ret)
887                         goto out;
888         }
889
890         *sp_location = sp;
891
892 out:
893         return ret;
894 }
895 EXPORT_SYMBOL(transfer_args_to_stack);
896
897 #endif /* CONFIG_MMU */
898
899 static struct file *do_open_execat(int fd, struct filename *name, int flags)
900 {
901         struct file *file;
902         int err;
903         struct open_flags open_exec_flags = {
904                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
905                 .acc_mode = MAY_EXEC,
906                 .intent = LOOKUP_OPEN,
907                 .lookup_flags = LOOKUP_FOLLOW,
908         };
909
910         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
911                 return ERR_PTR(-EINVAL);
912         if (flags & AT_SYMLINK_NOFOLLOW)
913                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
914         if (flags & AT_EMPTY_PATH)
915                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
916
917         file = do_filp_open(fd, name, &open_exec_flags);
918         if (IS_ERR(file))
919                 goto out;
920
921         /*
922          * may_open() has already checked for this, so it should be
923          * impossible to trip now. But we need to be extra cautious
924          * and check again at the very end too.
925          */
926         err = -EACCES;
927         if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
928                          path_noexec(&file->f_path)))
929                 goto exit;
930
931         err = deny_write_access(file);
932         if (err)
933                 goto exit;
934
935 out:
936         return file;
937
938 exit:
939         fput(file);
940         return ERR_PTR(err);
941 }
942
943 struct file *open_exec(const char *name)
944 {
945         struct filename *filename = getname_kernel(name);
946         struct file *f = ERR_CAST(filename);
947
948         if (!IS_ERR(filename)) {
949                 f = do_open_execat(AT_FDCWD, filename, 0);
950                 putname(filename);
951         }
952         return f;
953 }
954 EXPORT_SYMBOL(open_exec);
955
956 #if defined(CONFIG_BINFMT_FLAT) || 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         mm_init_cid(mm);
1010         /*
1011          * This prevents preemption while active_mm is being loaded and
1012          * it and mm are being updated, which could cause problems for
1013          * lazy tlb mm refcounting when these are updated by context
1014          * switches. Not all architectures can handle irqs off over
1015          * activate_mm yet.
1016          */
1017         if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1018                 local_irq_enable();
1019         activate_mm(active_mm, mm);
1020         if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1021                 local_irq_enable();
1022         lru_gen_add_mm(mm);
1023         task_unlock(tsk);
1024         lru_gen_use_mm(mm);
1025         if (old_mm) {
1026                 mmap_read_unlock(old_mm);
1027                 BUG_ON(active_mm != old_mm);
1028                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1029                 mm_update_next_owner(old_mm);
1030                 mmput(old_mm);
1031                 return 0;
1032         }
1033         mmdrop_lazy_tlb(active_mm);
1034         return 0;
1035 }
1036
1037 static int de_thread(struct task_struct *tsk)
1038 {
1039         struct signal_struct *sig = tsk->signal;
1040         struct sighand_struct *oldsighand = tsk->sighand;
1041         spinlock_t *lock = &oldsighand->siglock;
1042
1043         if (thread_group_empty(tsk))
1044                 goto no_thread_group;
1045
1046         /*
1047          * Kill all other threads in the thread group.
1048          */
1049         spin_lock_irq(lock);
1050         if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1051                 /*
1052                  * Another group action in progress, just
1053                  * return so that the signal is processed.
1054                  */
1055                 spin_unlock_irq(lock);
1056                 return -EAGAIN;
1057         }
1058
1059         sig->group_exec_task = tsk;
1060         sig->notify_count = zap_other_threads(tsk);
1061         if (!thread_group_leader(tsk))
1062                 sig->notify_count--;
1063
1064         while (sig->notify_count) {
1065                 __set_current_state(TASK_KILLABLE);
1066                 spin_unlock_irq(lock);
1067                 schedule();
1068                 if (__fatal_signal_pending(tsk))
1069                         goto killed;
1070                 spin_lock_irq(lock);
1071         }
1072         spin_unlock_irq(lock);
1073
1074         /*
1075          * At this point all other threads have exited, all we have to
1076          * do is to wait for the thread group leader to become inactive,
1077          * and to assume its PID:
1078          */
1079         if (!thread_group_leader(tsk)) {
1080                 struct task_struct *leader = tsk->group_leader;
1081
1082                 for (;;) {
1083                         cgroup_threadgroup_change_begin(tsk);
1084                         write_lock_irq(&tasklist_lock);
1085                         /*
1086                          * Do this under tasklist_lock to ensure that
1087                          * exit_notify() can't miss ->group_exec_task
1088                          */
1089                         sig->notify_count = -1;
1090                         if (likely(leader->exit_state))
1091                                 break;
1092                         __set_current_state(TASK_KILLABLE);
1093                         write_unlock_irq(&tasklist_lock);
1094                         cgroup_threadgroup_change_end(tsk);
1095                         schedule();
1096                         if (__fatal_signal_pending(tsk))
1097                                 goto killed;
1098                 }
1099
1100                 /*
1101                  * The only record we have of the real-time age of a
1102                  * process, regardless of execs it's done, is start_time.
1103                  * All the past CPU time is accumulated in signal_struct
1104                  * from sister threads now dead.  But in this non-leader
1105                  * exec, nothing survives from the original leader thread,
1106                  * whose birth marks the true age of this process now.
1107                  * When we take on its identity by switching to its PID, we
1108                  * also take its birthdate (always earlier than our own).
1109                  */
1110                 tsk->start_time = leader->start_time;
1111                 tsk->start_boottime = leader->start_boottime;
1112
1113                 BUG_ON(!same_thread_group(leader, tsk));
1114                 /*
1115                  * An exec() starts a new thread group with the
1116                  * TGID of the previous thread group. Rehash the
1117                  * two threads with a switched PID, and release
1118                  * the former thread group leader:
1119                  */
1120
1121                 /* Become a process group leader with the old leader's pid.
1122                  * The old leader becomes a thread of the this thread group.
1123                  */
1124                 exchange_tids(tsk, leader);
1125                 transfer_pid(leader, tsk, PIDTYPE_TGID);
1126                 transfer_pid(leader, tsk, PIDTYPE_PGID);
1127                 transfer_pid(leader, tsk, PIDTYPE_SID);
1128
1129                 list_replace_rcu(&leader->tasks, &tsk->tasks);
1130                 list_replace_init(&leader->sibling, &tsk->sibling);
1131
1132                 tsk->group_leader = tsk;
1133                 leader->group_leader = tsk;
1134
1135                 tsk->exit_signal = SIGCHLD;
1136                 leader->exit_signal = -1;
1137
1138                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1139                 leader->exit_state = EXIT_DEAD;
1140
1141                 /*
1142                  * We are going to release_task()->ptrace_unlink() silently,
1143                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1144                  * the tracer won't block again waiting for this thread.
1145                  */
1146                 if (unlikely(leader->ptrace))
1147                         __wake_up_parent(leader, leader->parent);
1148                 write_unlock_irq(&tasklist_lock);
1149                 cgroup_threadgroup_change_end(tsk);
1150
1151                 release_task(leader);
1152         }
1153
1154         sig->group_exec_task = NULL;
1155         sig->notify_count = 0;
1156
1157 no_thread_group:
1158         /* we have changed execution domain */
1159         tsk->exit_signal = SIGCHLD;
1160
1161         BUG_ON(!thread_group_leader(tsk));
1162         return 0;
1163
1164 killed:
1165         /* protects against exit_notify() and __exit_signal() */
1166         read_lock(&tasklist_lock);
1167         sig->group_exec_task = NULL;
1168         sig->notify_count = 0;
1169         read_unlock(&tasklist_lock);
1170         return -EAGAIN;
1171 }
1172
1173
1174 /*
1175  * This function makes sure the current process has its own signal table,
1176  * so that flush_signal_handlers can later reset the handlers without
1177  * disturbing other processes.  (Other processes might share the signal
1178  * table via the CLONE_SIGHAND option to clone().)
1179  */
1180 static int unshare_sighand(struct task_struct *me)
1181 {
1182         struct sighand_struct *oldsighand = me->sighand;
1183
1184         if (refcount_read(&oldsighand->count) != 1) {
1185                 struct sighand_struct *newsighand;
1186                 /*
1187                  * This ->sighand is shared with the CLONE_SIGHAND
1188                  * but not CLONE_THREAD task, switch to the new one.
1189                  */
1190                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1191                 if (!newsighand)
1192                         return -ENOMEM;
1193
1194                 refcount_set(&newsighand->count, 1);
1195
1196                 write_lock_irq(&tasklist_lock);
1197                 spin_lock(&oldsighand->siglock);
1198                 memcpy(newsighand->action, oldsighand->action,
1199                        sizeof(newsighand->action));
1200                 rcu_assign_pointer(me->sighand, newsighand);
1201                 spin_unlock(&oldsighand->siglock);
1202                 write_unlock_irq(&tasklist_lock);
1203
1204                 __cleanup_sighand(oldsighand);
1205         }
1206         return 0;
1207 }
1208
1209 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1210 {
1211         task_lock(tsk);
1212         /* Always NUL terminated and zero-padded */
1213         strscpy_pad(buf, tsk->comm, buf_size);
1214         task_unlock(tsk);
1215         return buf;
1216 }
1217 EXPORT_SYMBOL_GPL(__get_task_comm);
1218
1219 /*
1220  * These functions flushes out all traces of the currently running executable
1221  * so that a new one can be started
1222  */
1223
1224 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1225 {
1226         task_lock(tsk);
1227         trace_task_rename(tsk, buf);
1228         strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1229         task_unlock(tsk);
1230         perf_event_comm(tsk, exec);
1231 }
1232
1233 /*
1234  * Calling this is the point of no return. None of the failures will be
1235  * seen by userspace since either the process is already taking a fatal
1236  * signal (via de_thread() or coredump), or will have SEGV raised
1237  * (after exec_mmap()) by search_binary_handler (see below).
1238  */
1239 int begin_new_exec(struct linux_binprm * bprm)
1240 {
1241         struct task_struct *me = current;
1242         int retval;
1243
1244         /* Once we are committed compute the creds */
1245         retval = bprm_creds_from_file(bprm);
1246         if (retval)
1247                 return retval;
1248
1249         /*
1250          * Ensure all future errors are fatal.
1251          */
1252         bprm->point_of_no_return = true;
1253
1254         /*
1255          * Make this the only thread in the thread group.
1256          */
1257         retval = de_thread(me);
1258         if (retval)
1259                 goto out;
1260
1261         /*
1262          * Cancel any io_uring activity across execve
1263          */
1264         io_uring_task_cancel();
1265
1266         /* Ensure the files table is not shared. */
1267         retval = unshare_files();
1268         if (retval)
1269                 goto out;
1270
1271         /*
1272          * Must be called _before_ exec_mmap() as bprm->mm is
1273          * not visible until then. This also enables the update
1274          * to be lockless.
1275          */
1276         retval = set_mm_exe_file(bprm->mm, bprm->file);
1277         if (retval)
1278                 goto out;
1279
1280         /* If the binary is not readable then enforce mm->dumpable=0 */
1281         would_dump(bprm, bprm->file);
1282         if (bprm->have_execfd)
1283                 would_dump(bprm, bprm->executable);
1284
1285         /*
1286          * Release all of the old mmap stuff
1287          */
1288         acct_arg_size(bprm, 0);
1289         retval = exec_mmap(bprm->mm);
1290         if (retval)
1291                 goto out;
1292
1293         bprm->mm = NULL;
1294
1295         retval = exec_task_namespaces();
1296         if (retval)
1297                 goto out_unlock;
1298
1299 #ifdef CONFIG_POSIX_TIMERS
1300         spin_lock_irq(&me->sighand->siglock);
1301         posix_cpu_timers_exit(me);
1302         spin_unlock_irq(&me->sighand->siglock);
1303         exit_itimers(me);
1304         flush_itimer_signals();
1305 #endif
1306
1307         /*
1308          * Make the signal table private.
1309          */
1310         retval = unshare_sighand(me);
1311         if (retval)
1312                 goto out_unlock;
1313
1314         me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
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 mnt_idmap *idmap = file_mnt_idmap(file);
1415         if (inode_permission(idmap, 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, idmap, 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         /*
1571          * If another task is sharing our fs, we cannot safely
1572          * suid exec because the differently privileged task
1573          * will be able to manipulate the current directory, etc.
1574          * It would be nice to force an unshare instead...
1575          */
1576         t = p;
1577         n_fs = 1;
1578         spin_lock(&p->fs->lock);
1579         rcu_read_lock();
1580         while_each_thread(p, t) {
1581                 if (t->fs == p->fs)
1582                         n_fs++;
1583         }
1584         rcu_read_unlock();
1585
1586         if (p->fs->users > n_fs)
1587                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1588         else
1589                 p->fs->in_exec = 1;
1590         spin_unlock(&p->fs->lock);
1591 }
1592
1593 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1594 {
1595         /* Handle suid and sgid on files */
1596         struct mnt_idmap *idmap;
1597         struct inode *inode = file_inode(file);
1598         unsigned int mode;
1599         vfsuid_t vfsuid;
1600         vfsgid_t vfsgid;
1601
1602         if (!mnt_may_suid(file->f_path.mnt))
1603                 return;
1604
1605         if (task_no_new_privs(current))
1606                 return;
1607
1608         mode = READ_ONCE(inode->i_mode);
1609         if (!(mode & (S_ISUID|S_ISGID)))
1610                 return;
1611
1612         idmap = file_mnt_idmap(file);
1613
1614         /* Be careful if suid/sgid is set */
1615         inode_lock(inode);
1616
1617         /* reload atomically mode/uid/gid now that lock held */
1618         mode = inode->i_mode;
1619         vfsuid = i_uid_into_vfsuid(idmap, inode);
1620         vfsgid = i_gid_into_vfsgid(idmap, inode);
1621         inode_unlock(inode);
1622
1623         /* We ignore suid/sgid if there are no mappings for them in the ns */
1624         if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1625             !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1626                 return;
1627
1628         if (mode & S_ISUID) {
1629                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1630                 bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1631         }
1632
1633         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1634                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1635                 bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1636         }
1637 }
1638
1639 /*
1640  * Compute brpm->cred based upon the final binary.
1641  */
1642 static int bprm_creds_from_file(struct linux_binprm *bprm)
1643 {
1644         /* Compute creds based on which file? */
1645         struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1646
1647         bprm_fill_uid(bprm, file);
1648         return security_bprm_creds_from_file(bprm, file);
1649 }
1650
1651 /*
1652  * Fill the binprm structure from the inode.
1653  * Read the first BINPRM_BUF_SIZE bytes
1654  *
1655  * This may be called multiple times for binary chains (scripts for example).
1656  */
1657 static int prepare_binprm(struct linux_binprm *bprm)
1658 {
1659         loff_t pos = 0;
1660
1661         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1662         return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1663 }
1664
1665 /*
1666  * Arguments are '\0' separated strings found at the location bprm->p
1667  * points to; chop off the first by relocating brpm->p to right after
1668  * the first '\0' encountered.
1669  */
1670 int remove_arg_zero(struct linux_binprm *bprm)
1671 {
1672         int ret = 0;
1673         unsigned long offset;
1674         char *kaddr;
1675         struct page *page;
1676
1677         if (!bprm->argc)
1678                 return 0;
1679
1680         do {
1681                 offset = bprm->p & ~PAGE_MASK;
1682                 page = get_arg_page(bprm, bprm->p, 0);
1683                 if (!page) {
1684                         ret = -EFAULT;
1685                         goto out;
1686                 }
1687                 kaddr = kmap_local_page(page);
1688
1689                 for (; offset < PAGE_SIZE && kaddr[offset];
1690                                 offset++, bprm->p++)
1691                         ;
1692
1693                 kunmap_local(kaddr);
1694                 put_arg_page(page);
1695         } while (offset == PAGE_SIZE);
1696
1697         bprm->p++;
1698         bprm->argc--;
1699         ret = 0;
1700
1701 out:
1702         return ret;
1703 }
1704 EXPORT_SYMBOL(remove_arg_zero);
1705
1706 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1707 /*
1708  * cycle the list of binary formats handler, until one recognizes the image
1709  */
1710 static int search_binary_handler(struct linux_binprm *bprm)
1711 {
1712         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1713         struct linux_binfmt *fmt;
1714         int retval;
1715
1716         retval = prepare_binprm(bprm);
1717         if (retval < 0)
1718                 return retval;
1719
1720         retval = security_bprm_check(bprm);
1721         if (retval)
1722                 return retval;
1723
1724         retval = -ENOENT;
1725  retry:
1726         read_lock(&binfmt_lock);
1727         list_for_each_entry(fmt, &formats, lh) {
1728                 if (!try_module_get(fmt->module))
1729                         continue;
1730                 read_unlock(&binfmt_lock);
1731
1732                 retval = fmt->load_binary(bprm);
1733
1734                 read_lock(&binfmt_lock);
1735                 put_binfmt(fmt);
1736                 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1737                         read_unlock(&binfmt_lock);
1738                         return retval;
1739                 }
1740         }
1741         read_unlock(&binfmt_lock);
1742
1743         if (need_retry) {
1744                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1745                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1746                         return retval;
1747                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1748                         return retval;
1749                 need_retry = false;
1750                 goto retry;
1751         }
1752
1753         return retval;
1754 }
1755
1756 /* binfmt handlers will call back into begin_new_exec() on success. */
1757 static int exec_binprm(struct linux_binprm *bprm)
1758 {
1759         pid_t old_pid, old_vpid;
1760         int ret, depth;
1761
1762         /* Need to fetch pid before load_binary changes it */
1763         old_pid = current->pid;
1764         rcu_read_lock();
1765         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1766         rcu_read_unlock();
1767
1768         /* This allows 4 levels of binfmt rewrites before failing hard. */
1769         for (depth = 0;; depth++) {
1770                 struct file *exec;
1771                 if (depth > 5)
1772                         return -ELOOP;
1773
1774                 ret = search_binary_handler(bprm);
1775                 if (ret < 0)
1776                         return ret;
1777                 if (!bprm->interpreter)
1778                         break;
1779
1780                 exec = bprm->file;
1781                 bprm->file = bprm->interpreter;
1782                 bprm->interpreter = NULL;
1783
1784                 allow_write_access(exec);
1785                 if (unlikely(bprm->have_execfd)) {
1786                         if (bprm->executable) {
1787                                 fput(exec);
1788                                 return -ENOEXEC;
1789                         }
1790                         bprm->executable = exec;
1791                 } else
1792                         fput(exec);
1793         }
1794
1795         audit_bprm(bprm);
1796         trace_sched_process_exec(current, old_pid, bprm);
1797         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1798         proc_exec_connector(current);
1799         return 0;
1800 }
1801
1802 /*
1803  * sys_execve() executes a new program.
1804  */
1805 static int bprm_execve(struct linux_binprm *bprm,
1806                        int fd, struct filename *filename, int flags)
1807 {
1808         struct file *file;
1809         int retval;
1810
1811         retval = prepare_bprm_creds(bprm);
1812         if (retval)
1813                 return retval;
1814
1815         /*
1816          * Check for unsafe execution states before exec_binprm(), which
1817          * will call back into begin_new_exec(), into bprm_creds_from_file(),
1818          * where setuid-ness is evaluated.
1819          */
1820         check_unsafe_exec(bprm);
1821         current->in_execve = 1;
1822         sched_mm_cid_before_execve(current);
1823
1824         file = do_open_execat(fd, filename, flags);
1825         retval = PTR_ERR(file);
1826         if (IS_ERR(file))
1827                 goto out_unmark;
1828
1829         sched_exec();
1830
1831         bprm->file = file;
1832         /*
1833          * Record that a name derived from an O_CLOEXEC fd will be
1834          * inaccessible after exec.  This allows the code in exec to
1835          * choose to fail when the executable is not mmaped into the
1836          * interpreter and an open file descriptor is not passed to
1837          * the interpreter.  This makes for a better user experience
1838          * than having the interpreter start and then immediately fail
1839          * when it finds the executable is inaccessible.
1840          */
1841         if (bprm->fdpath && get_close_on_exec(fd))
1842                 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1843
1844         /* Set the unchanging part of bprm->cred */
1845         retval = security_bprm_creds_for_exec(bprm);
1846         if (retval)
1847                 goto out;
1848
1849         retval = exec_binprm(bprm);
1850         if (retval < 0)
1851                 goto out;
1852
1853         sched_mm_cid_after_execve(current);
1854         /* execve succeeded */
1855         current->fs->in_exec = 0;
1856         current->in_execve = 0;
1857         rseq_execve(current);
1858         user_events_execve(current);
1859         acct_update_integrals(current);
1860         task_numa_free(current, false);
1861         return retval;
1862
1863 out:
1864         /*
1865          * If past the point of no return ensure the code never
1866          * returns to the userspace process.  Use an existing fatal
1867          * signal if present otherwise terminate the process with
1868          * SIGSEGV.
1869          */
1870         if (bprm->point_of_no_return && !fatal_signal_pending(current))
1871                 force_fatal_sig(SIGSEGV);
1872
1873 out_unmark:
1874         sched_mm_cid_after_execve(current);
1875         current->fs->in_exec = 0;
1876         current->in_execve = 0;
1877
1878         return retval;
1879 }
1880
1881 static int do_execveat_common(int fd, struct filename *filename,
1882                               struct user_arg_ptr argv,
1883                               struct user_arg_ptr envp,
1884                               int flags)
1885 {
1886         struct linux_binprm *bprm;
1887         int retval;
1888
1889         if (IS_ERR(filename))
1890                 return PTR_ERR(filename);
1891
1892         /*
1893          * We move the actual failure in case of RLIMIT_NPROC excess from
1894          * set*uid() to execve() because too many poorly written programs
1895          * don't check setuid() return code.  Here we additionally recheck
1896          * whether NPROC limit is still exceeded.
1897          */
1898         if ((current->flags & PF_NPROC_EXCEEDED) &&
1899             is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1900                 retval = -EAGAIN;
1901                 goto out_ret;
1902         }
1903
1904         /* We're below the limit (still or again), so we don't want to make
1905          * further execve() calls fail. */
1906         current->flags &= ~PF_NPROC_EXCEEDED;
1907
1908         bprm = alloc_bprm(fd, filename);
1909         if (IS_ERR(bprm)) {
1910                 retval = PTR_ERR(bprm);
1911                 goto out_ret;
1912         }
1913
1914         retval = count(argv, MAX_ARG_STRINGS);
1915         if (retval == 0)
1916                 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1917                              current->comm, bprm->filename);
1918         if (retval < 0)
1919                 goto out_free;
1920         bprm->argc = retval;
1921
1922         retval = count(envp, MAX_ARG_STRINGS);
1923         if (retval < 0)
1924                 goto out_free;
1925         bprm->envc = retval;
1926
1927         retval = bprm_stack_limits(bprm);
1928         if (retval < 0)
1929                 goto out_free;
1930
1931         retval = copy_string_kernel(bprm->filename, bprm);
1932         if (retval < 0)
1933                 goto out_free;
1934         bprm->exec = bprm->p;
1935
1936         retval = copy_strings(bprm->envc, envp, bprm);
1937         if (retval < 0)
1938                 goto out_free;
1939
1940         retval = copy_strings(bprm->argc, argv, bprm);
1941         if (retval < 0)
1942                 goto out_free;
1943
1944         /*
1945          * When argv is empty, add an empty string ("") as argv[0] to
1946          * ensure confused userspace programs that start processing
1947          * from argv[1] won't end up walking envp. See also
1948          * bprm_stack_limits().
1949          */
1950         if (bprm->argc == 0) {
1951                 retval = copy_string_kernel("", bprm);
1952                 if (retval < 0)
1953                         goto out_free;
1954                 bprm->argc = 1;
1955         }
1956
1957         retval = bprm_execve(bprm, fd, filename, flags);
1958 out_free:
1959         free_bprm(bprm);
1960
1961 out_ret:
1962         putname(filename);
1963         return retval;
1964 }
1965
1966 int kernel_execve(const char *kernel_filename,
1967                   const char *const *argv, const char *const *envp)
1968 {
1969         struct filename *filename;
1970         struct linux_binprm *bprm;
1971         int fd = AT_FDCWD;
1972         int retval;
1973
1974         /* It is non-sense for kernel threads to call execve */
1975         if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1976                 return -EINVAL;
1977
1978         filename = getname_kernel(kernel_filename);
1979         if (IS_ERR(filename))
1980                 return PTR_ERR(filename);
1981
1982         bprm = alloc_bprm(fd, filename);
1983         if (IS_ERR(bprm)) {
1984                 retval = PTR_ERR(bprm);
1985                 goto out_ret;
1986         }
1987
1988         retval = count_strings_kernel(argv);
1989         if (WARN_ON_ONCE(retval == 0))
1990                 retval = -EINVAL;
1991         if (retval < 0)
1992                 goto out_free;
1993         bprm->argc = retval;
1994
1995         retval = count_strings_kernel(envp);
1996         if (retval < 0)
1997                 goto out_free;
1998         bprm->envc = retval;
1999
2000         retval = bprm_stack_limits(bprm);
2001         if (retval < 0)
2002                 goto out_free;
2003
2004         retval = copy_string_kernel(bprm->filename, bprm);
2005         if (retval < 0)
2006                 goto out_free;
2007         bprm->exec = bprm->p;
2008
2009         retval = copy_strings_kernel(bprm->envc, envp, bprm);
2010         if (retval < 0)
2011                 goto out_free;
2012
2013         retval = copy_strings_kernel(bprm->argc, argv, bprm);
2014         if (retval < 0)
2015                 goto out_free;
2016
2017         retval = bprm_execve(bprm, fd, filename, 0);
2018 out_free:
2019         free_bprm(bprm);
2020 out_ret:
2021         putname(filename);
2022         return retval;
2023 }
2024
2025 static int do_execve(struct filename *filename,
2026         const char __user *const __user *__argv,
2027         const char __user *const __user *__envp)
2028 {
2029         struct user_arg_ptr argv = { .ptr.native = __argv };
2030         struct user_arg_ptr envp = { .ptr.native = __envp };
2031         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2032 }
2033
2034 static int do_execveat(int fd, struct filename *filename,
2035                 const char __user *const __user *__argv,
2036                 const char __user *const __user *__envp,
2037                 int flags)
2038 {
2039         struct user_arg_ptr argv = { .ptr.native = __argv };
2040         struct user_arg_ptr envp = { .ptr.native = __envp };
2041
2042         return do_execveat_common(fd, filename, argv, envp, flags);
2043 }
2044
2045 #ifdef CONFIG_COMPAT
2046 static int compat_do_execve(struct filename *filename,
2047         const compat_uptr_t __user *__argv,
2048         const compat_uptr_t __user *__envp)
2049 {
2050         struct user_arg_ptr argv = {
2051                 .is_compat = true,
2052                 .ptr.compat = __argv,
2053         };
2054         struct user_arg_ptr envp = {
2055                 .is_compat = true,
2056                 .ptr.compat = __envp,
2057         };
2058         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2059 }
2060
2061 static int compat_do_execveat(int fd, struct filename *filename,
2062                               const compat_uptr_t __user *__argv,
2063                               const compat_uptr_t __user *__envp,
2064                               int flags)
2065 {
2066         struct user_arg_ptr argv = {
2067                 .is_compat = true,
2068                 .ptr.compat = __argv,
2069         };
2070         struct user_arg_ptr envp = {
2071                 .is_compat = true,
2072                 .ptr.compat = __envp,
2073         };
2074         return do_execveat_common(fd, filename, argv, envp, flags);
2075 }
2076 #endif
2077
2078 void set_binfmt(struct linux_binfmt *new)
2079 {
2080         struct mm_struct *mm = current->mm;
2081
2082         if (mm->binfmt)
2083                 module_put(mm->binfmt->module);
2084
2085         mm->binfmt = new;
2086         if (new)
2087                 __module_get(new->module);
2088 }
2089 EXPORT_SYMBOL(set_binfmt);
2090
2091 /*
2092  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2093  */
2094 void set_dumpable(struct mm_struct *mm, int value)
2095 {
2096         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2097                 return;
2098
2099         set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2100 }
2101
2102 SYSCALL_DEFINE3(execve,
2103                 const char __user *, filename,
2104                 const char __user *const __user *, argv,
2105                 const char __user *const __user *, envp)
2106 {
2107         return do_execve(getname(filename), argv, envp);
2108 }
2109
2110 SYSCALL_DEFINE5(execveat,
2111                 int, fd, const char __user *, filename,
2112                 const char __user *const __user *, argv,
2113                 const char __user *const __user *, envp,
2114                 int, flags)
2115 {
2116         return do_execveat(fd,
2117                            getname_uflags(filename, flags),
2118                            argv, envp, flags);
2119 }
2120
2121 #ifdef CONFIG_COMPAT
2122 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2123         const compat_uptr_t __user *, argv,
2124         const compat_uptr_t __user *, envp)
2125 {
2126         return compat_do_execve(getname(filename), argv, envp);
2127 }
2128
2129 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2130                        const char __user *, filename,
2131                        const compat_uptr_t __user *, argv,
2132                        const compat_uptr_t __user *, envp,
2133                        int,  flags)
2134 {
2135         return compat_do_execveat(fd,
2136                                   getname_uflags(filename, flags),
2137                                   argv, envp, flags);
2138 }
2139 #endif
2140
2141 #ifdef CONFIG_SYSCTL
2142
2143 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2144                 void *buffer, size_t *lenp, loff_t *ppos)
2145 {
2146         int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2147
2148         if (!error)
2149                 validate_coredump_safety();
2150         return error;
2151 }
2152
2153 static struct ctl_table fs_exec_sysctls[] = {
2154         {
2155                 .procname       = "suid_dumpable",
2156                 .data           = &suid_dumpable,
2157                 .maxlen         = sizeof(int),
2158                 .mode           = 0644,
2159                 .proc_handler   = proc_dointvec_minmax_coredump,
2160                 .extra1         = SYSCTL_ZERO,
2161                 .extra2         = SYSCTL_TWO,
2162         },
2163         { }
2164 };
2165
2166 static int __init init_fs_exec_sysctls(void)
2167 {
2168         register_sysctl_init("fs", fs_exec_sysctls);
2169         return 0;
2170 }
2171
2172 fs_initcall(init_fs_exec_sysctls);
2173 #endif /* CONFIG_SYSCTL */
This page took 0.15772 seconds and 4 git commands to generate.