1/*
2 * linux/fs/exec.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * #!-checking implemented by tytso.
9 */
10/*
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
14 *
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
17 *
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
22 * formats.
23 */
24
25#include <linux/slab.h>
26#include <linux/file.h>
27#include <linux/fdtable.h>
28#include <linux/mm.h>
29#include <linux/vmacache.h>
30#include <linux/stat.h>
31#include <linux/fcntl.h>
32#include <linux/swap.h>
33#include <linux/string.h>
34#include <linux/init.h>
35#include <linux/sched/mm.h>
36#include <linux/sched/coredump.h>
37#include <linux/sched/signal.h>
38#include <linux/sched/numa_balancing.h>
39#include <linux/sched/task.h>
40#include <linux/pagemap.h>
41#include <linux/perf_event.h>
42#include <linux/highmem.h>
43#include <linux/spinlock.h>
44#include <linux/key.h>
45#include <linux/personality.h>
46#include <linux/binfmts.h>
47#include <linux/utsname.h>
48#include <linux/pid_namespace.h>
49#include <linux/module.h>
50#include <linux/namei.h>
51#include <linux/mount.h>
52#include <linux/security.h>
53#include <linux/syscalls.h>
54#include <linux/tsacct_kern.h>
55#include <linux/cn_proc.h>
56#include <linux/audit.h>
57#include <linux/tracehook.h>
58#include <linux/kmod.h>
59#include <linux/fsnotify.h>
60#include <linux/fs_struct.h>
61#include <linux/pipe_fs_i.h>
62#include <linux/oom.h>
63#include <linux/compat.h>
64#include <linux/vmalloc.h>
65
66#include <linux/uaccess.h>
67#include <asm/mmu_context.h>
68#include <asm/tlb.h>
69
70#include <trace/events/task.h>
71#include "internal.h"
72
73#include <trace/events/sched.h>
74
75int suid_dumpable = 0;
76
77static LIST_HEAD(formats);
78static DEFINE_RWLOCK(binfmt_lock);
79
80void __register_binfmt(struct linux_binfmt * fmt, int insert)
81{
82 BUG_ON(!fmt);
83 if (WARN_ON(!fmt->load_binary))
84 return;
85 write_lock(&binfmt_lock);
86 insert ? list_add(&fmt->lh, &formats) :
87 list_add_tail(&fmt->lh, &formats);
88 write_unlock(&binfmt_lock);
89}
90
91EXPORT_SYMBOL(__register_binfmt);
92
93void unregister_binfmt(struct linux_binfmt * fmt)
94{
95 write_lock(&binfmt_lock);
96 list_del(&fmt->lh);
97 write_unlock(&binfmt_lock);
98}
99
100EXPORT_SYMBOL(unregister_binfmt);
101
102static inline void put_binfmt(struct linux_binfmt * fmt)
103{
104 module_put(fmt->module);
105}
106
107bool path_noexec(const struct path *path)
108{
109 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
110 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
111}
112
113#ifdef CONFIG_USELIB
114/*
115 * Note that a shared library must be both readable and executable due to
116 * security reasons.
117 *
118 * Also note that we take the address to load from from the file itself.
119 */
120SYSCALL_DEFINE1(uselib, const char __user *, library)
121{
122 struct linux_binfmt *fmt;
123 struct file *file;
124 struct filename *tmp = getname(library);
125 int error = PTR_ERR(tmp);
126 static const struct open_flags uselib_flags = {
127 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
128 .acc_mode = MAY_READ | MAY_EXEC,
129 .intent = LOOKUP_OPEN,
130 .lookup_flags = LOOKUP_FOLLOW,
131 };
132
133 if (IS_ERR(tmp))
134 goto out;
135
136 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
137 putname(tmp);
138 error = PTR_ERR(file);
139 if (IS_ERR(file))
140 goto out;
141
142 error = -EINVAL;
143 if (!S_ISREG(file_inode(file)->i_mode))
144 goto exit;
145
146 error = -EACCES;
147 if (path_noexec(&file->f_path))
148 goto exit;
149
150 fsnotify_open(file);
151
152 error = -ENOEXEC;
153
154 read_lock(&binfmt_lock);
155 list_for_each_entry(fmt, &formats, lh) {
156 if (!fmt->load_shlib)
157 continue;
158 if (!try_module_get(fmt->module))
159 continue;
160 read_unlock(&binfmt_lock);
161 error = fmt->load_shlib(file);
162 read_lock(&binfmt_lock);
163 put_binfmt(fmt);
164 if (error != -ENOEXEC)
165 break;
166 }
167 read_unlock(&binfmt_lock);
168exit:
169 fput(file);
170out:
171 return error;
172}
173#endif /* #ifdef CONFIG_USELIB */
174
175#ifdef CONFIG_MMU
176/*
177 * The nascent bprm->mm is not visible until exec_mmap() but it can
178 * use a lot of memory, account these pages in current->mm temporary
179 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
180 * change the counter back via acct_arg_size(0).
181 */
182static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
183{
184 struct mm_struct *mm = current->mm;
185 long diff = (long)(pages - bprm->vma_pages);
186
187 if (!mm || !diff)
188 return;
189
190 bprm->vma_pages = pages;
191 add_mm_counter(mm, MM_ANONPAGES, diff);
192}
193
194static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
195 int write)
196{
197 struct page *page;
198 int ret;
199 unsigned int gup_flags = FOLL_FORCE;
200
201#ifdef CONFIG_STACK_GROWSUP
202 if (write) {
203 ret = expand_downwards(bprm->vma, pos);
204 if (ret < 0)
205 return NULL;
206 }
207#endif
208
209 if (write)
210 gup_flags |= FOLL_WRITE;
211
212 /*
213 * We are doing an exec(). 'current' is the process
214 * doing the exec and bprm->mm is the new process's mm.
215 */
216 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
217 &page, NULL, NULL);
218 if (ret <= 0)
219 return NULL;
220
221 if (write)
222 acct_arg_size(bprm, vma_pages(bprm->vma));
223
224 return page;
225}
226
227static void put_arg_page(struct page *page)
228{
229 put_page(page);
230}
231
232static void free_arg_pages(struct linux_binprm *bprm)
233{
234}
235
236static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
237 struct page *page)
238{
239 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
240}
241
242static int __bprm_mm_init(struct linux_binprm *bprm)
243{
244 int err;
245 struct vm_area_struct *vma = NULL;
246 struct mm_struct *mm = bprm->mm;
247
248 bprm->vma = vma = vm_area_alloc(mm);
249 if (!vma)
250 return -ENOMEM;
251 vma_set_anonymous(vma);
252
253 if (down_write_killable(&mm->mmap_sem)) {
254 err = -EINTR;
255 goto err_free;
256 }
257
258 /*
259 * Place the stack at the largest stack address the architecture
260 * supports. Later, we'll move this to an appropriate place. We don't
261 * use STACK_TOP because that can depend on attributes which aren't
262 * configured yet.
263 */
264 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
265 vma->vm_end = STACK_TOP_MAX;
266 vma->vm_start = vma->vm_end - PAGE_SIZE;
267 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
268 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
269
270 err = insert_vm_struct(mm, vma);
271 if (err)
272 goto err;
273
274 mm->stack_vm = mm->total_vm = 1;
275 arch_bprm_mm_init(mm, vma);
276 up_write(&mm->mmap_sem);
277 bprm->p = vma->vm_end - sizeof(void *);
278 return 0;
279err:
280 up_write(&mm->mmap_sem);
281err_free:
282 bprm->vma = NULL;
283 vm_area_free(vma);
284 return err;
285}
286
287static bool valid_arg_len(struct linux_binprm *bprm, long len)
288{
289 return len <= MAX_ARG_STRLEN;
290}
291
292#else
293
294static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
295{
296}
297
298static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
299 int write)
300{
301 struct page *page;
302
303 page = bprm->page[pos / PAGE_SIZE];
304 if (!page && write) {
305 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
306 if (!page)
307 return NULL;
308 bprm->page[pos / PAGE_SIZE] = page;
309 }
310
311 return page;
312}
313
314static void put_arg_page(struct page *page)
315{
316}
317
318static void free_arg_page(struct linux_binprm *bprm, int i)
319{
320 if (bprm->page[i]) {
321 __free_page(bprm->page[i]);
322 bprm->page[i] = NULL;
323 }
324}
325
326static void free_arg_pages(struct linux_binprm *bprm)
327{
328 int i;
329
330 for (i = 0; i < MAX_ARG_PAGES; i++)
331 free_arg_page(bprm, i);
332}
333
334static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
335 struct page *page)
336{
337}
338
339static int __bprm_mm_init(struct linux_binprm *bprm)
340{
341 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
342 return 0;
343}
344
345static bool valid_arg_len(struct linux_binprm *bprm, long len)
346{
347 return len <= bprm->p;
348}
349
350#endif /* CONFIG_MMU */
351
352/*
353 * Create a new mm_struct and populate it with a temporary stack
354 * vm_area_struct. We don't have enough context at this point to set the stack
355 * flags, permissions, and offset, so we use temporary values. We'll update
356 * them later in setup_arg_pages().
357 */
358static int bprm_mm_init(struct linux_binprm *bprm)
359{
360 int err;
361 struct mm_struct *mm = NULL;
362
363 bprm->mm = mm = mm_alloc();
364 err = -ENOMEM;
365 if (!mm)
366 goto err;
367
368 /* Save current stack limit for all calculations made during exec. */
369 task_lock(current->group_leader);
370 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
371 task_unlock(current->group_leader);
372
373 err = __bprm_mm_init(bprm);
374 if (err)
375 goto err;
376
377 return 0;
378
379err:
380 if (mm) {
381 bprm->mm = NULL;
382 mmdrop(mm);
383 }
384
385 return err;
386}
387
388struct user_arg_ptr {
389#ifdef CONFIG_COMPAT
390 bool is_compat;
391#endif
392 union {
393 const char __user *const __user *native;
394#ifdef CONFIG_COMPAT
395 const compat_uptr_t __user *compat;
396#endif
397 } ptr;
398};
399
400static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
401{
402 const char __user *native;
403
404#ifdef CONFIG_COMPAT
405 if (unlikely(argv.is_compat)) {
406 compat_uptr_t compat;
407
408 if (get_user(compat, argv.ptr.compat + nr))
409 return ERR_PTR(-EFAULT);
410
411 return compat_ptr(compat);
412 }
413#endif
414
415 if (get_user(native, argv.ptr.native + nr))
416 return ERR_PTR(-EFAULT);
417
418 return native;
419}
420
421/*
422 * count() counts the number of strings in array ARGV.
423 */
424static int count(struct user_arg_ptr argv, int max)
425{
426 int i = 0;
427
428 if (argv.ptr.native != NULL) {
429 for (;;) {
430 const char __user *p = get_user_arg_ptr(argv, i);
431
432 if (!p)
433 break;
434
435 if (IS_ERR(p))
436 return -EFAULT;
437
438 if (i >= max)
439 return -E2BIG;
440 ++i;
441
442 if (fatal_signal_pending(current))
443 return -ERESTARTNOHAND;
444 cond_resched();
445 }
446 }
447 return i;
448}
449
450static int prepare_arg_pages(struct linux_binprm *bprm,
451 struct user_arg_ptr argv, struct user_arg_ptr envp)
452{
453 unsigned long limit, ptr_size;
454
455 bprm->argc = count(argv, MAX_ARG_STRINGS);
456 if (bprm->argc < 0)
457 return bprm->argc;
458
459 bprm->envc = count(envp, MAX_ARG_STRINGS);
460 if (bprm->envc < 0)
461 return bprm->envc;
462
463 /*
464 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
465 * (whichever is smaller) for the argv+env strings.
466 * This ensures that:
467 * - the remaining binfmt code will not run out of stack space,
468 * - the program will have a reasonable amount of stack left
469 * to work from.
470 */
471 limit = _STK_LIM / 4 * 3;
472 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
473 /*
474 * We've historically supported up to 32 pages (ARG_MAX)
475 * of argument strings even with small stacks
476 */
477 limit = max_t(unsigned long, limit, ARG_MAX);
478 /*
479 * We must account for the size of all the argv and envp pointers to
480 * the argv and envp strings, since they will also take up space in
481 * the stack. They aren't stored until much later when we can't
482 * signal to the parent that the child has run out of stack space.
483 * Instead, calculate it here so it's possible to fail gracefully.
484 */
485 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
486 if (limit <= ptr_size)
487 return -E2BIG;
488 limit -= ptr_size;
489
490 bprm->argmin = bprm->p - limit;
491 return 0;
492}
493
494/*
495 * 'copy_strings()' copies argument/environment strings from the old
496 * processes's memory to the new process's stack. The call to get_user_pages()
497 * ensures the destination page is created and not swapped out.
498 */
499static int copy_strings(int argc, struct user_arg_ptr argv,
500 struct linux_binprm *bprm)
501{
502 struct page *kmapped_page = NULL;
503 char *kaddr = NULL;
504 unsigned long kpos = 0;
505 int ret;
506
507 while (argc-- > 0) {
508 const char __user *str;
509 int len;
510 unsigned long pos;
511
512 ret = -EFAULT;
513 str = get_user_arg_ptr(argv, argc);
514 if (IS_ERR(str))
515 goto out;
516
517 len = strnlen_user(str, MAX_ARG_STRLEN);
518 if (!len)
519 goto out;
520
521 ret = -E2BIG;
522 if (!valid_arg_len(bprm, len))
523 goto out;
524
525 /* We're going to work our way backwords. */
526 pos = bprm->p;
527 str += len;
528 bprm->p -= len;
529#ifdef CONFIG_MMU
530 if (bprm->p < bprm->argmin)
531 goto out;
532#endif
533
534 while (len > 0) {
535 int offset, bytes_to_copy;
536
537 if (fatal_signal_pending(current)) {
538 ret = -ERESTARTNOHAND;
539 goto out;
540 }
541 cond_resched();
542
543 offset = pos % PAGE_SIZE;
544 if (offset == 0)
545 offset = PAGE_SIZE;
546
547 bytes_to_copy = offset;
548 if (bytes_to_copy > len)
549 bytes_to_copy = len;
550
551 offset -= bytes_to_copy;
552 pos -= bytes_to_copy;
553 str -= bytes_to_copy;
554 len -= bytes_to_copy;
555
556 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
557 struct page *page;
558
559 page = get_arg_page(bprm, pos, 1);
560 if (!page) {
561 ret = -E2BIG;
562 goto out;
563 }
564
565 if (kmapped_page) {
566 flush_kernel_dcache_page(kmapped_page);
567 kunmap(kmapped_page);
568 put_arg_page(kmapped_page);
569 }
570 kmapped_page = page;
571 kaddr = kmap(kmapped_page);
572 kpos = pos & PAGE_MASK;
573 flush_arg_page(bprm, kpos, kmapped_page);
574 }
575 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
576 ret = -EFAULT;
577 goto out;
578 }
579 }
580 }
581 ret = 0;
582out:
583 if (kmapped_page) {
584 flush_kernel_dcache_page(kmapped_page);
585 kunmap(kmapped_page);
586 put_arg_page(kmapped_page);
587 }
588 return ret;
589}
590
591/*
592 * Like copy_strings, but get argv and its values from kernel memory.
593 */
594int copy_strings_kernel(int argc, const char *const *__argv,
595 struct linux_binprm *bprm)
596{
597 int r;
598 mm_segment_t oldfs = get_fs();
599 struct user_arg_ptr argv = {
600 .ptr.native = (const char __user *const __user *)__argv,
601 };
602
603 set_fs(KERNEL_DS);
604 r = copy_strings(argc, argv, bprm);
605 set_fs(oldfs);
606
607 return r;
608}
609EXPORT_SYMBOL(copy_strings_kernel);
610
611#ifdef CONFIG_MMU
612
613/*
614 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
615 * the binfmt code determines where the new stack should reside, we shift it to
616 * its final location. The process proceeds as follows:
617 *
618 * 1) Use shift to calculate the new vma endpoints.
619 * 2) Extend vma to cover both the old and new ranges. This ensures the
620 * arguments passed to subsequent functions are consistent.
621 * 3) Move vma's page tables to the new range.
622 * 4) Free up any cleared pgd range.
623 * 5) Shrink the vma to cover only the new range.
624 */
625static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
626{
627 struct mm_struct *mm = vma->vm_mm;
628 unsigned long old_start = vma->vm_start;
629 unsigned long old_end = vma->vm_end;
630 unsigned long length = old_end - old_start;
631 unsigned long new_start = old_start - shift;
632 unsigned long new_end = old_end - shift;
633 struct mmu_gather tlb;
634
635 BUG_ON(new_start > new_end);
636
637 /*
638 * ensure there are no vmas between where we want to go
639 * and where we are
640 */
641 if (vma != find_vma(mm, new_start))
642 return -EFAULT;
643
644 /*
645 * cover the whole range: [new_start, old_end)
646 */
647 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
648 return -ENOMEM;
649
650 /*
651 * move the page tables downwards, on failure we rely on
652 * process cleanup to remove whatever mess we made.
653 */
654 if (length != move_page_tables(vma, old_start,
655 vma, new_start, length, false))
656 return -ENOMEM;
657
658 lru_add_drain();
659 tlb_gather_mmu(&tlb, mm, old_start, old_end);
660 if (new_end > old_start) {
661 /*
662 * when the old and new regions overlap clear from new_end.
663 */
664 free_pgd_range(&tlb, new_end, old_end, new_end,
665 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
666 } else {
667 /*
668 * otherwise, clean from old_start; this is done to not touch
669 * the address space in [new_end, old_start) some architectures
670 * have constraints on va-space that make this illegal (IA64) -
671 * for the others its just a little faster.
672 */
673 free_pgd_range(&tlb, old_start, old_end, new_end,
674 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
675 }
676 tlb_finish_mmu(&tlb, old_start, old_end);
677
678 /*
679 * Shrink the vma to just the new range. Always succeeds.
680 */
681 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
682
683 return 0;
684}
685
686/*
687 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
688 * the stack is optionally relocated, and some extra space is added.
689 */
690int setup_arg_pages(struct linux_binprm *bprm,
691 unsigned long stack_top,
692 int executable_stack)
693{
694 unsigned long ret;
695 unsigned long stack_shift;
696 struct mm_struct *mm = current->mm;
697 struct vm_area_struct *vma = bprm->vma;
698 struct vm_area_struct *prev = NULL;
699 unsigned long vm_flags;
700 unsigned long stack_base;
701 unsigned long stack_size;
702 unsigned long stack_expand;
703 unsigned long rlim_stack;
704
705#ifdef CONFIG_STACK_GROWSUP
706 /* Limit stack size */
707 stack_base = bprm->rlim_stack.rlim_max;
708 if (stack_base > STACK_SIZE_MAX)
709 stack_base = STACK_SIZE_MAX;
710
711 /* Add space for stack randomization. */
712 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
713
714 /* Make sure we didn't let the argument array grow too large. */
715 if (vma->vm_end - vma->vm_start > stack_base)
716 return -ENOMEM;
717
718 stack_base = PAGE_ALIGN(stack_top - stack_base);
719
720 stack_shift = vma->vm_start - stack_base;
721 mm->arg_start = bprm->p - stack_shift;
722 bprm->p = vma->vm_end - stack_shift;
723#else
724 stack_top = arch_align_stack(stack_top);
725 stack_top = PAGE_ALIGN(stack_top);
726
727 if (unlikely(stack_top < mmap_min_addr) ||
728 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
729 return -ENOMEM;
730
731 stack_shift = vma->vm_end - stack_top;
732
733 bprm->p -= stack_shift;
734 mm->arg_start = bprm->p;
735#endif
736
737 if (bprm->loader)
738 bprm->loader -= stack_shift;
739 bprm->exec -= stack_shift;
740
741 if (down_write_killable(&mm->mmap_sem))
742 return -EINTR;
743
744 vm_flags = VM_STACK_FLAGS;
745
746 /*
747 * Adjust stack execute permissions; explicitly enable for
748 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
749 * (arch default) otherwise.
750 */
751 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
752 vm_flags |= VM_EXEC;
753 else if (executable_stack == EXSTACK_DISABLE_X)
754 vm_flags &= ~VM_EXEC;
755 vm_flags |= mm->def_flags;
756 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
757
758 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
759 vm_flags);
760 if (ret)
761 goto out_unlock;
762 BUG_ON(prev != vma);
763
764 /* Move stack pages down in memory. */
765 if (stack_shift) {
766 ret = shift_arg_pages(vma, stack_shift);
767 if (ret)
768 goto out_unlock;
769 }
770
771 /* mprotect_fixup is overkill to remove the temporary stack flags */
772 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
773
774 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
775 stack_size = vma->vm_end - vma->vm_start;
776 /*
777 * Align this down to a page boundary as expand_stack
778 * will align it up.
779 */
780 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
781#ifdef CONFIG_STACK_GROWSUP
782 if (stack_size + stack_expand > rlim_stack)
783 stack_base = vma->vm_start + rlim_stack;
784 else
785 stack_base = vma->vm_end + stack_expand;
786#else
787 if (stack_size + stack_expand > rlim_stack)
788 stack_base = vma->vm_end - rlim_stack;
789 else
790 stack_base = vma->vm_start - stack_expand;
791#endif
792 current->mm->start_stack = bprm->p;
793 ret = expand_stack(vma, stack_base);
794 if (ret)
795 ret = -EFAULT;
796
797out_unlock:
798 up_write(&mm->mmap_sem);
799 return ret;
800}
801EXPORT_SYMBOL(setup_arg_pages);
802
803#else
804
805/*
806 * Transfer the program arguments and environment from the holding pages
807 * onto the stack. The provided stack pointer is adjusted accordingly.
808 */
809int transfer_args_to_stack(struct linux_binprm *bprm,
810 unsigned long *sp_location)
811{
812 unsigned long index, stop, sp;
813 int ret = 0;
814
815 stop = bprm->p >> PAGE_SHIFT;
816 sp = *sp_location;
817
818 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
819 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
820 char *src = kmap(bprm->page[index]) + offset;
821 sp -= PAGE_SIZE - offset;
822 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
823 ret = -EFAULT;
824 kunmap(bprm->page[index]);
825 if (ret)
826 goto out;
827 }
828
829 *sp_location = sp;
830
831out:
832 return ret;
833}
834EXPORT_SYMBOL(transfer_args_to_stack);
835
836#endif /* CONFIG_MMU */
837
838static struct file *do_open_execat(int fd, struct filename *name, int flags)
839{
840 struct file *file;
841 int err;
842 struct open_flags open_exec_flags = {
843 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
844 .acc_mode = MAY_EXEC,
845 .intent = LOOKUP_OPEN,
846 .lookup_flags = LOOKUP_FOLLOW,
847 };
848
849 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
850 return ERR_PTR(-EINVAL);
851 if (flags & AT_SYMLINK_NOFOLLOW)
852 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
853 if (flags & AT_EMPTY_PATH)
854 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
855
856 file = do_filp_open(fd, name, &open_exec_flags);
857 if (IS_ERR(file))
858 goto out;
859
860 err = -EACCES;
861 if (!S_ISREG(file_inode(file)->i_mode))
862 goto exit;
863
864 if (path_noexec(&file->f_path))
865 goto exit;
866
867 err = deny_write_access(file);
868 if (err)
869 goto exit;
870
871 if (name->name[0] != '\0')
872 fsnotify_open(file);
873
874out:
875 return file;
876
877exit:
878 fput(file);
879 return ERR_PTR(err);
880}
881
882struct file *open_exec(const char *name)
883{
884 struct filename *filename = getname_kernel(name);
885 struct file *f = ERR_CAST(filename);
886
887 if (!IS_ERR(filename)) {
888 f = do_open_execat(AT_FDCWD, filename, 0);
889 putname(filename);
890 }
891 return f;
892}
893EXPORT_SYMBOL(open_exec);
894
895int kernel_read_file(struct file *file, void **buf, loff_t *size,
896 loff_t max_size, enum kernel_read_file_id id)
897{
898 loff_t i_size, pos;
899 ssize_t bytes = 0;
900 int ret;
901
902 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
903 return -EINVAL;
904
905 ret = deny_write_access(file);
906 if (ret)
907 return ret;
908
909 ret = security_kernel_read_file(file, id);
910 if (ret)
911 goto out;
912
913 i_size = i_size_read(file_inode(file));
914 if (i_size <= 0) {
915 ret = -EINVAL;
916 goto out;
917 }
918 if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
919 ret = -EFBIG;
920 goto out;
921 }
922
923 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
924 *buf = vmalloc(i_size);
925 if (!*buf) {
926 ret = -ENOMEM;
927 goto out;
928 }
929
930 pos = 0;
931 while (pos < i_size) {
932 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
933 if (bytes < 0) {
934 ret = bytes;
935 goto out_free;
936 }
937
938 if (bytes == 0)
939 break;
940 }
941
942 if (pos != i_size) {
943 ret = -EIO;
944 goto out_free;
945 }
946
947 ret = security_kernel_post_read_file(file, *buf, i_size, id);
948 if (!ret)
949 *size = pos;
950
951out_free:
952 if (ret < 0) {
953 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
954 vfree(*buf);
955 *buf = NULL;
956 }
957 }
958
959out:
960 allow_write_access(file);
961 return ret;
962}
963EXPORT_SYMBOL_GPL(kernel_read_file);
964
965int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
966 loff_t max_size, enum kernel_read_file_id id)
967{
968 struct file *file;
969 int ret;
970
971 if (!path || !*path)
972 return -EINVAL;
973
974 file = filp_open(path, O_RDONLY, 0);
975 if (IS_ERR(file))
976 return PTR_ERR(file);
977
978 ret = kernel_read_file(file, buf, size, max_size, id);
979 fput(file);
980 return ret;
981}
982EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
983
984int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
985 enum kernel_read_file_id id)
986{
987 struct fd f = fdget(fd);
988 int ret = -EBADF;
989
990 if (!f.file)
991 goto out;
992
993 ret = kernel_read_file(f.file, buf, size, max_size, id);
994out:
995 fdput(f);
996 return ret;
997}
998EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
999
1000ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1001{
1002 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1003 if (res > 0)
1004 flush_icache_range(addr, addr + len);
1005 return res;
1006}
1007EXPORT_SYMBOL(read_code);
1008
1009static int exec_mmap(struct mm_struct *mm)
1010{
1011 struct task_struct *tsk;
1012 struct mm_struct *old_mm, *active_mm;
1013
1014 /* Notify parent that we're no longer interested in the old VM */
1015 tsk = current;
1016 old_mm = current->mm;
1017 mm_release(tsk, old_mm);
1018
1019 if (old_mm) {
1020 sync_mm_rss(old_mm);
1021 /*
1022 * Make sure that if there is a core dump in progress
1023 * for the old mm, we get out and die instead of going
1024 * through with the exec. We must hold mmap_sem around
1025 * checking core_state and changing tsk->mm.
1026 */
1027 down_read(&old_mm->mmap_sem);
1028 if (unlikely(old_mm->core_state)) {
1029 up_read(&old_mm->mmap_sem);
1030 return -EINTR;
1031 }
1032 }
1033 task_lock(tsk);
1034 active_mm = tsk->active_mm;
1035 tsk->mm = mm;
1036 tsk->active_mm = mm;
1037 activate_mm(active_mm, mm);
1038 tsk->mm->vmacache_seqnum = 0;
1039 vmacache_flush(tsk);
1040 task_unlock(tsk);
1041 if (old_mm) {
1042 up_read(&old_mm->mmap_sem);
1043 BUG_ON(active_mm != old_mm);
1044 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1045 mm_update_next_owner(old_mm);
1046 mmput(old_mm);
1047 return 0;
1048 }
1049 mmdrop(active_mm);
1050 return 0;
1051}
1052
1053/*
1054 * This function makes sure the current process has its own signal table,
1055 * so that flush_signal_handlers can later reset the handlers without
1056 * disturbing other processes. (Other processes might share the signal
1057 * table via the CLONE_SIGHAND option to clone().)
1058 */
1059static int de_thread(struct task_struct *tsk)
1060{
1061 struct signal_struct *sig = tsk->signal;
1062 struct sighand_struct *oldsighand = tsk->sighand;
1063 spinlock_t *lock = &oldsighand->siglock;
1064
1065 if (thread_group_empty(tsk))
1066 goto no_thread_group;
1067
1068 /*
1069 * Kill all other threads in the thread group.
1070 */
1071 spin_lock_irq(lock);
1072 if (signal_group_exit(sig)) {
1073 /*
1074 * Another group action in progress, just
1075 * return so that the signal is processed.
1076 */
1077 spin_unlock_irq(lock);
1078 return -EAGAIN;
1079 }
1080
1081 sig->group_exit_task = tsk;
1082 sig->notify_count = zap_other_threads(tsk);
1083 if (!thread_group_leader(tsk))
1084 sig->notify_count--;
1085
1086 while (sig->notify_count) {
1087 __set_current_state(TASK_KILLABLE);
1088 spin_unlock_irq(lock);
1089 schedule();
1090 if (__fatal_signal_pending(tsk))
1091 goto killed;
1092 spin_lock_irq(lock);
1093 }
1094 spin_unlock_irq(lock);
1095
1096 /*
1097 * At this point all other threads have exited, all we have to
1098 * do is to wait for the thread group leader to become inactive,
1099 * and to assume its PID:
1100 */
1101 if (!thread_group_leader(tsk)) {
1102 struct task_struct *leader = tsk->group_leader;
1103
1104 for (;;) {
1105 cgroup_threadgroup_change_begin(tsk);
1106 write_lock_irq(&tasklist_lock);
1107 /*
1108 * Do this under tasklist_lock to ensure that
1109 * exit_notify() can't miss ->group_exit_task
1110 */
1111 sig->notify_count = -1;
1112 if (likely(leader->exit_state))
1113 break;
1114 __set_current_state(TASK_KILLABLE);
1115 write_unlock_irq(&tasklist_lock);
1116 cgroup_threadgroup_change_end(tsk);
1117 schedule();
1118 if (__fatal_signal_pending(tsk))
1119 goto killed;
1120 }
1121
1122 /*
1123 * The only record we have of the real-time age of a
1124 * process, regardless of execs it's done, is start_time.
1125 * All the past CPU time is accumulated in signal_struct
1126 * from sister threads now dead. But in this non-leader
1127 * exec, nothing survives from the original leader thread,
1128 * whose birth marks the true age of this process now.
1129 * When we take on its identity by switching to its PID, we
1130 * also take its birthdate (always earlier than our own).
1131 */
1132 tsk->start_time = leader->start_time;
1133 tsk->real_start_time = leader->real_start_time;
1134
1135 BUG_ON(!same_thread_group(leader, tsk));
1136 BUG_ON(has_group_leader_pid(tsk));
1137 /*
1138 * An exec() starts a new thread group with the
1139 * TGID of the previous thread group. Rehash the
1140 * two threads with a switched PID, and release
1141 * the former thread group leader:
1142 */
1143
1144 /* Become a process group leader with the old leader's pid.
1145 * The old leader becomes a thread of the this thread group.
1146 * Note: The old leader also uses this pid until release_task
1147 * is called. Odd but simple and correct.
1148 */
1149 tsk->pid = leader->pid;
1150 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1151 transfer_pid(leader, tsk, PIDTYPE_TGID);
1152 transfer_pid(leader, tsk, PIDTYPE_PGID);
1153 transfer_pid(leader, tsk, PIDTYPE_SID);
1154
1155 list_replace_rcu(&leader->tasks, &tsk->tasks);
1156 list_replace_init(&leader->sibling, &tsk->sibling);
1157
1158 tsk->group_leader = tsk;
1159 leader->group_leader = tsk;
1160
1161 tsk->exit_signal = SIGCHLD;
1162 leader->exit_signal = -1;
1163
1164 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1165 leader->exit_state = EXIT_DEAD;
1166
1167 /*
1168 * We are going to release_task()->ptrace_unlink() silently,
1169 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1170 * the tracer wont't block again waiting for this thread.
1171 */
1172 if (unlikely(leader->ptrace))
1173 __wake_up_parent(leader, leader->parent);
1174 write_unlock_irq(&tasklist_lock);
1175 cgroup_threadgroup_change_end(tsk);
1176
1177 release_task(leader);
1178 }
1179
1180 sig->group_exit_task = NULL;
1181 sig->notify_count = 0;
1182
1183no_thread_group:
1184 /* we have changed execution domain */
1185 tsk->exit_signal = SIGCHLD;
1186
1187#ifdef CONFIG_POSIX_TIMERS
1188 exit_itimers(sig);
1189 flush_itimer_signals();
1190#endif
1191
1192 if (refcount_read(&oldsighand->count) != 1) {
1193 struct sighand_struct *newsighand;
1194 /*
1195 * This ->sighand is shared with the CLONE_SIGHAND
1196 * but not CLONE_THREAD task, switch to the new one.
1197 */
1198 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1199 if (!newsighand)
1200 return -ENOMEM;
1201
1202 refcount_set(&newsighand->count, 1);
1203 memcpy(newsighand->action, oldsighand->action,
1204 sizeof(newsighand->action));
1205
1206 write_lock_irq(&tasklist_lock);
1207 spin_lock(&oldsighand->siglock);
1208 rcu_assign_pointer(tsk->sighand, newsighand);
1209 spin_unlock(&oldsighand->siglock);
1210 write_unlock_irq(&tasklist_lock);
1211
1212 __cleanup_sighand(oldsighand);
1213 }
1214
1215 BUG_ON(!thread_group_leader(tsk));
1216 return 0;
1217
1218killed:
1219 /* protects against exit_notify() and __exit_signal() */
1220 read_lock(&tasklist_lock);
1221 sig->group_exit_task = NULL;
1222 sig->notify_count = 0;
1223 read_unlock(&tasklist_lock);
1224 return -EAGAIN;
1225}
1226
1227char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1228{
1229 task_lock(tsk);
1230 strncpy(buf, tsk->comm, buf_size);
1231 task_unlock(tsk);
1232 return buf;
1233}
1234EXPORT_SYMBOL_GPL(__get_task_comm);
1235
1236/*
1237 * These functions flushes out all traces of the currently running executable
1238 * so that a new one can be started
1239 */
1240
1241void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1242{
1243 task_lock(tsk);
1244 trace_task_rename(tsk, buf);
1245 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1246 task_unlock(tsk);
1247 perf_event_comm(tsk, exec);
1248}
1249
1250/*
1251 * Calling this is the point of no return. None of the failures will be
1252 * seen by userspace since either the process is already taking a fatal
1253 * signal (via de_thread() or coredump), or will have SEGV raised
1254 * (after exec_mmap()) by search_binary_handlers (see below).
1255 */
1256int flush_old_exec(struct linux_binprm * bprm)
1257{
1258 int retval;
1259
1260 /*
1261 * Make sure we have a private signal table and that
1262 * we are unassociated from the previous thread group.
1263 */
1264 retval = de_thread(current);
1265 if (retval)
1266 goto out;
1267
1268 /*
1269 * Must be called _before_ exec_mmap() as bprm->mm is
1270 * not visibile until then. This also enables the update
1271 * to be lockless.
1272 */
1273 set_mm_exe_file(bprm->mm, bprm->file);
1274
1275 /*
1276 * Release all of the old mmap stuff
1277 */
1278 acct_arg_size(bprm, 0);
1279 retval = exec_mmap(bprm->mm);
1280 if (retval)
1281 goto out;
1282
1283 /*
1284 * After clearing bprm->mm (to mark that current is using the
1285 * prepared mm now), we have nothing left of the original
1286 * process. If anything from here on returns an error, the check
1287 * in search_binary_handler() will SEGV current.
1288 */
1289 bprm->mm = NULL;
1290
1291 set_fs(USER_DS);
1292 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1293 PF_NOFREEZE | PF_NO_SETAFFINITY);
1294 flush_thread();
1295 current->personality &= ~bprm->per_clear;
1296
1297 /*
1298 * We have to apply CLOEXEC before we change whether the process is
1299 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1300 * trying to access the should-be-closed file descriptors of a process
1301 * undergoing exec(2).
1302 */
1303 do_close_on_exec(current->files);
1304 return 0;
1305
1306out:
1307 return retval;
1308}
1309EXPORT_SYMBOL(flush_old_exec);
1310
1311void would_dump(struct linux_binprm *bprm, struct file *file)
1312{
1313 struct inode *inode = file_inode(file);
1314 if (inode_permission(inode, MAY_READ) < 0) {
1315 struct user_namespace *old, *user_ns;
1316 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1317
1318 /* Ensure mm->user_ns contains the executable */
1319 user_ns = old = bprm->mm->user_ns;
1320 while ((user_ns != &init_user_ns) &&
1321 !privileged_wrt_inode_uidgid(user_ns, inode))
1322 user_ns = user_ns->parent;
1323
1324 if (old != user_ns) {
1325 bprm->mm->user_ns = get_user_ns(user_ns);
1326 put_user_ns(old);
1327 }
1328 }
1329}
1330EXPORT_SYMBOL(would_dump);
1331
1332void setup_new_exec(struct linux_binprm * bprm)
1333{
1334 /*
1335 * Once here, prepare_binrpm() will not be called any more, so
1336 * the final state of setuid/setgid/fscaps can be merged into the
1337 * secureexec flag.
1338 */
1339 bprm->secureexec |= bprm->cap_elevated;
1340
1341 if (bprm->secureexec) {
1342 /* Make sure parent cannot signal privileged process. */
1343 current->pdeath_signal = 0;
1344
1345 /*
1346 * For secureexec, reset the stack limit to sane default to
1347 * avoid bad behavior from the prior rlimits. This has to
1348 * happen before arch_pick_mmap_layout(), which examines
1349 * RLIMIT_STACK, but after the point of no return to avoid
1350 * needing to clean up the change on failure.
1351 */
1352 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1353 bprm->rlim_stack.rlim_cur = _STK_LIM;
1354 }
1355
1356 arch_pick_mmap_layout(current->mm, &bprm->rlim_stack);
1357
1358 current->sas_ss_sp = current->sas_ss_size = 0;
1359
1360 /*
1361 * Figure out dumpability. Note that this checking only of current
1362 * is wrong, but userspace depends on it. This should be testing
1363 * bprm->secureexec instead.
1364 */
1365 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1366 !(uid_eq(current_euid(), current_uid()) &&
1367 gid_eq(current_egid(), current_gid())))
1368 set_dumpable(current->mm, suid_dumpable);
1369 else
1370 set_dumpable(current->mm, SUID_DUMP_USER);
1371
1372 arch_setup_new_exec();
1373 perf_event_exec();
1374 __set_task_comm(current, kbasename(bprm->filename), true);
1375
1376 /* Set the new mm task size. We have to do that late because it may
1377 * depend on TIF_32BIT which is only updated in flush_thread() on
1378 * some architectures like powerpc
1379 */
1380 current->mm->task_size = TASK_SIZE;
1381
1382 /* An exec changes our domain. We are no longer part of the thread
1383 group */
1384 current->self_exec_id++;
1385 flush_signal_handlers(current, 0);
1386}
1387EXPORT_SYMBOL(setup_new_exec);
1388
1389/* Runs immediately before start_thread() takes over. */
1390void finalize_exec(struct linux_binprm *bprm)
1391{
1392 /* Store any stack rlimit changes before starting thread. */
1393 task_lock(current->group_leader);
1394 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1395 task_unlock(current->group_leader);
1396}
1397EXPORT_SYMBOL(finalize_exec);
1398
1399/*
1400 * Prepare credentials and lock ->cred_guard_mutex.
1401 * install_exec_creds() commits the new creds and drops the lock.
1402 * Or, if exec fails before, free_bprm() should release ->cred and
1403 * and unlock.
1404 */
1405static int prepare_bprm_creds(struct linux_binprm *bprm)
1406{
1407 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1408 return -ERESTARTNOINTR;
1409
1410 bprm->cred = prepare_exec_creds();
1411 if (likely(bprm->cred))
1412 return 0;
1413
1414 mutex_unlock(&current->signal->cred_guard_mutex);
1415 return -ENOMEM;
1416}
1417
1418static void free_bprm(struct linux_binprm *bprm)
1419{
1420 free_arg_pages(bprm);
1421 if (bprm->cred) {
1422 mutex_unlock(&current->signal->cred_guard_mutex);
1423 abort_creds(bprm->cred);
1424 }
1425 if (bprm->file) {
1426 allow_write_access(bprm->file);
1427 fput(bprm->file);
1428 }
1429 /* If a binfmt changed the interp, free it. */
1430 if (bprm->interp != bprm->filename)
1431 kfree(bprm->interp);
1432 kfree(bprm);
1433}
1434
1435int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1436{
1437 /* If a binfmt changed the interp, free it first. */
1438 if (bprm->interp != bprm->filename)
1439 kfree(bprm->interp);
1440 bprm->interp = kstrdup(interp, GFP_KERNEL);
1441 if (!bprm->interp)
1442 return -ENOMEM;
1443 return 0;
1444}
1445EXPORT_SYMBOL(bprm_change_interp);
1446
1447/*
1448 * install the new credentials for this executable
1449 */
1450void install_exec_creds(struct linux_binprm *bprm)
1451{
1452 security_bprm_committing_creds(bprm);
1453
1454 commit_creds(bprm->cred);
1455 bprm->cred = NULL;
1456
1457 /*
1458 * Disable monitoring for regular users
1459 * when executing setuid binaries. Must
1460 * wait until new credentials are committed
1461 * by commit_creds() above
1462 */
1463 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1464 perf_event_exit_task(current);
1465 /*
1466 * cred_guard_mutex must be held at least to this point to prevent
1467 * ptrace_attach() from altering our determination of the task's
1468 * credentials; any time after this it may be unlocked.
1469 */
1470 security_bprm_committed_creds(bprm);
1471 mutex_unlock(&current->signal->cred_guard_mutex);
1472}
1473EXPORT_SYMBOL(install_exec_creds);
1474
1475/*
1476 * determine how safe it is to execute the proposed program
1477 * - the caller must hold ->cred_guard_mutex to protect against
1478 * PTRACE_ATTACH or seccomp thread-sync
1479 */
1480static void check_unsafe_exec(struct linux_binprm *bprm)
1481{
1482 struct task_struct *p = current, *t;
1483 unsigned n_fs;
1484
1485 if (p->ptrace)
1486 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1487
1488 /*
1489 * This isn't strictly necessary, but it makes it harder for LSMs to
1490 * mess up.
1491 */
1492 if (task_no_new_privs(current))
1493 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1494
1495 t = p;
1496 n_fs = 1;
1497 spin_lock(&p->fs->lock);
1498 rcu_read_lock();
1499 while_each_thread(p, t) {
1500 if (t->fs == p->fs)
1501 n_fs++;
1502 }
1503 rcu_read_unlock();
1504
1505 if (p->fs->users > n_fs)
1506 bprm->unsafe |= LSM_UNSAFE_SHARE;
1507 else
1508 p->fs->in_exec = 1;
1509 spin_unlock(&p->fs->lock);
1510}
1511
1512static void bprm_fill_uid(struct linux_binprm *bprm)
1513{
1514 struct inode *inode;
1515 unsigned int mode;
1516 kuid_t uid;
1517 kgid_t gid;
1518
1519 /*
1520 * Since this can be called multiple times (via prepare_binprm),
1521 * we must clear any previous work done when setting set[ug]id
1522 * bits from any earlier bprm->file uses (for example when run
1523 * first for a setuid script then again for its interpreter).
1524 */
1525 bprm->cred->euid = current_euid();
1526 bprm->cred->egid = current_egid();
1527
1528 if (!mnt_may_suid(bprm->file->f_path.mnt))
1529 return;
1530
1531 if (task_no_new_privs(current))
1532 return;
1533
1534 inode = bprm->file->f_path.dentry->d_inode;
1535 mode = READ_ONCE(inode->i_mode);
1536 if (!(mode & (S_ISUID|S_ISGID)))
1537 return;
1538
1539 /* Be careful if suid/sgid is set */
1540 inode_lock(inode);
1541
1542 /* reload atomically mode/uid/gid now that lock held */
1543 mode = inode->i_mode;
1544 uid = inode->i_uid;
1545 gid = inode->i_gid;
1546 inode_unlock(inode);
1547
1548 /* We ignore suid/sgid if there are no mappings for them in the ns */
1549 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1550 !kgid_has_mapping(bprm->cred->user_ns, gid))
1551 return;
1552
1553 if (mode & S_ISUID) {
1554 bprm->per_clear |= PER_CLEAR_ON_SETID;
1555 bprm->cred->euid = uid;
1556 }
1557
1558 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1559 bprm->per_clear |= PER_CLEAR_ON_SETID;
1560 bprm->cred->egid = gid;
1561 }
1562}
1563
1564/*
1565 * Fill the binprm structure from the inode.
1566 * Check permissions, then read the first BINPRM_BUF_SIZE bytes
1567 *
1568 * This may be called multiple times for binary chains (scripts for example).
1569 */
1570int prepare_binprm(struct linux_binprm *bprm)
1571{
1572 int retval;
1573 loff_t pos = 0;
1574
1575 bprm_fill_uid(bprm);
1576
1577 /* fill in binprm security blob */
1578 retval = security_bprm_set_creds(bprm);
1579 if (retval)
1580 return retval;
1581 bprm->called_set_creds = 1;
1582
1583 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1584 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1585}
1586
1587EXPORT_SYMBOL(prepare_binprm);
1588
1589/*
1590 * Arguments are '\0' separated strings found at the location bprm->p
1591 * points to; chop off the first by relocating brpm->p to right after
1592 * the first '\0' encountered.
1593 */
1594int remove_arg_zero(struct linux_binprm *bprm)
1595{
1596 int ret = 0;
1597 unsigned long offset;
1598 char *kaddr;
1599 struct page *page;
1600
1601 if (!bprm->argc)
1602 return 0;
1603
1604 do {
1605 offset = bprm->p & ~PAGE_MASK;
1606 page = get_arg_page(bprm, bprm->p, 0);
1607 if (!page) {
1608 ret = -EFAULT;
1609 goto out;
1610 }
1611 kaddr = kmap_atomic(page);
1612
1613 for (; offset < PAGE_SIZE && kaddr[offset];
1614 offset++, bprm->p++)
1615 ;
1616
1617 kunmap_atomic(kaddr);
1618 put_arg_page(page);
1619 } while (offset == PAGE_SIZE);
1620
1621 bprm->p++;
1622 bprm->argc--;
1623 ret = 0;
1624
1625out:
1626 return ret;
1627}
1628EXPORT_SYMBOL(remove_arg_zero);
1629
1630#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1631/*
1632 * cycle the list of binary formats handler, until one recognizes the image
1633 */
1634int search_binary_handler(struct linux_binprm *bprm)
1635{
1636 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1637 struct linux_binfmt *fmt;
1638 int retval;
1639
1640 /* This allows 4 levels of binfmt rewrites before failing hard. */
1641 if (bprm->recursion_depth > 5)
1642 return -ELOOP;
1643
1644 retval = security_bprm_check(bprm);
1645 if (retval)
1646 return retval;
1647
1648 retval = -ENOENT;
1649 retry:
1650 read_lock(&binfmt_lock);
1651 list_for_each_entry(fmt, &formats, lh) {
1652 if (!try_module_get(fmt->module))
1653 continue;
1654 read_unlock(&binfmt_lock);
1655 bprm->recursion_depth++;
1656 retval = fmt->load_binary(bprm);
1657 read_lock(&binfmt_lock);
1658 put_binfmt(fmt);
1659 bprm->recursion_depth--;
1660 if (retval < 0 && !bprm->mm) {
1661 /* we got to flush_old_exec() and failed after it */
1662 read_unlock(&binfmt_lock);
1663 force_sigsegv(SIGSEGV, current);
1664 return retval;
1665 }
1666 if (retval != -ENOEXEC || !bprm->file) {
1667 read_unlock(&binfmt_lock);
1668 return retval;
1669 }
1670 }
1671 read_unlock(&binfmt_lock);
1672
1673 if (need_retry) {
1674 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1675 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1676 return retval;
1677 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1678 return retval;
1679 need_retry = false;
1680 goto retry;
1681 }
1682
1683 return retval;
1684}
1685EXPORT_SYMBOL(search_binary_handler);
1686
1687static int exec_binprm(struct linux_binprm *bprm)
1688{
1689 pid_t old_pid, old_vpid;
1690 int ret;
1691
1692 /* Need to fetch pid before load_binary changes it */
1693 old_pid = current->pid;
1694 rcu_read_lock();
1695 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1696 rcu_read_unlock();
1697
1698 ret = search_binary_handler(bprm);
1699 if (ret >= 0) {
1700 audit_bprm(bprm);
1701 trace_sched_process_exec(current, old_pid, bprm);
1702 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1703 proc_exec_connector(current);
1704 }
1705
1706 return ret;
1707}
1708
1709/*
1710 * sys_execve() executes a new program.
1711 */
1712static int __do_execve_file(int fd, struct filename *filename,
1713 struct user_arg_ptr argv,
1714 struct user_arg_ptr envp,
1715 int flags, struct file *file)
1716{
1717 char *pathbuf = NULL;
1718 struct linux_binprm *bprm;
1719 struct files_struct *displaced;
1720 int retval;
1721
1722 if (IS_ERR(filename))
1723 return PTR_ERR(filename);
1724
1725 /*
1726 * We move the actual failure in case of RLIMIT_NPROC excess from
1727 * set*uid() to execve() because too many poorly written programs
1728 * don't check setuid() return code. Here we additionally recheck
1729 * whether NPROC limit is still exceeded.
1730 */
1731 if ((current->flags & PF_NPROC_EXCEEDED) &&
1732 atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1733 retval = -EAGAIN;
1734 goto out_ret;
1735 }
1736
1737 /* We're below the limit (still or again), so we don't want to make
1738 * further execve() calls fail. */
1739 current->flags &= ~PF_NPROC_EXCEEDED;
1740
1741 retval = unshare_files(&displaced);
1742 if (retval)
1743 goto out_ret;
1744
1745 retval = -ENOMEM;
1746 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1747 if (!bprm)
1748 goto out_files;
1749
1750 retval = prepare_bprm_creds(bprm);
1751 if (retval)
1752 goto out_free;
1753
1754 check_unsafe_exec(bprm);
1755 current->in_execve = 1;
1756
1757 if (!file)
1758 file = do_open_execat(fd, filename, flags);
1759 retval = PTR_ERR(file);
1760 if (IS_ERR(file))
1761 goto out_unmark;
1762
1763 sched_exec();
1764
1765 bprm->file = file;
1766 if (!filename) {
1767 bprm->filename = "none";
1768 } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1769 bprm->filename = filename->name;
1770 } else {
1771 if (filename->name[0] == '\0')
1772 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1773 else
1774 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1775 fd, filename->name);
1776 if (!pathbuf) {
1777 retval = -ENOMEM;
1778 goto out_unmark;
1779 }
1780 /*
1781 * Record that a name derived from an O_CLOEXEC fd will be
1782 * inaccessible after exec. Relies on having exclusive access to
1783 * current->files (due to unshare_files above).
1784 */
1785 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1786 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1787 bprm->filename = pathbuf;
1788 }
1789 bprm->interp = bprm->filename;
1790
1791 retval = bprm_mm_init(bprm);
1792 if (retval)
1793 goto out_unmark;
1794
1795 retval = prepare_arg_pages(bprm, argv, envp);
1796 if (retval < 0)
1797 goto out;
1798
1799 retval = prepare_binprm(bprm);
1800 if (retval < 0)
1801 goto out;
1802
1803 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1804 if (retval < 0)
1805 goto out;
1806
1807 bprm->exec = bprm->p;
1808 retval = copy_strings(bprm->envc, envp, bprm);
1809 if (retval < 0)
1810 goto out;
1811
1812 retval = copy_strings(bprm->argc, argv, bprm);
1813 if (retval < 0)
1814 goto out;
1815
1816 would_dump(bprm, bprm->file);
1817
1818 retval = exec_binprm(bprm);
1819 if (retval < 0)
1820 goto out;
1821
1822 /* execve succeeded */
1823 current->fs->in_exec = 0;
1824 current->in_execve = 0;
1825 membarrier_execve(current);
1826 rseq_execve(current);
1827 acct_update_integrals(current);
1828 task_numa_free(current);
1829 free_bprm(bprm);
1830 kfree(pathbuf);
1831 if (filename)
1832 putname(filename);
1833 if (displaced)
1834 put_files_struct(displaced);
1835 return retval;
1836
1837out:
1838 if (bprm->mm) {
1839 acct_arg_size(bprm, 0);
1840 mmput(bprm->mm);
1841 }
1842
1843out_unmark:
1844 current->fs->in_exec = 0;
1845 current->in_execve = 0;
1846
1847out_free:
1848 free_bprm(bprm);
1849 kfree(pathbuf);
1850
1851out_files:
1852 if (displaced)
1853 reset_files_struct(displaced);
1854out_ret:
1855 if (filename)
1856 putname(filename);
1857 return retval;
1858}
1859
1860static int do_execveat_common(int fd, struct filename *filename,
1861 struct user_arg_ptr argv,
1862 struct user_arg_ptr envp,
1863 int flags)
1864{
1865 return __do_execve_file(fd, filename, argv, envp, flags, NULL);
1866}
1867
1868int do_execve_file(struct file *file, void *__argv, void *__envp)
1869{
1870 struct user_arg_ptr argv = { .ptr.native = __argv };
1871 struct user_arg_ptr envp = { .ptr.native = __envp };
1872
1873 return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file);
1874}
1875
1876int do_execve(struct filename *filename,
1877 const char __user *const __user *__argv,
1878 const char __user *const __user *__envp)
1879{
1880 struct user_arg_ptr argv = { .ptr.native = __argv };
1881 struct user_arg_ptr envp = { .ptr.native = __envp };
1882 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1883}
1884
1885int do_execveat(int fd, struct filename *filename,
1886 const char __user *const __user *__argv,
1887 const char __user *const __user *__envp,
1888 int flags)
1889{
1890 struct user_arg_ptr argv = { .ptr.native = __argv };
1891 struct user_arg_ptr envp = { .ptr.native = __envp };
1892
1893 return do_execveat_common(fd, filename, argv, envp, flags);
1894}
1895
1896#ifdef CONFIG_COMPAT
1897static int compat_do_execve(struct filename *filename,
1898 const compat_uptr_t __user *__argv,
1899 const compat_uptr_t __user *__envp)
1900{
1901 struct user_arg_ptr argv = {
1902 .is_compat = true,
1903 .ptr.compat = __argv,
1904 };
1905 struct user_arg_ptr envp = {
1906 .is_compat = true,
1907 .ptr.compat = __envp,
1908 };
1909 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1910}
1911
1912static int compat_do_execveat(int fd, struct filename *filename,
1913 const compat_uptr_t __user *__argv,
1914 const compat_uptr_t __user *__envp,
1915 int flags)
1916{
1917 struct user_arg_ptr argv = {
1918 .is_compat = true,
1919 .ptr.compat = __argv,
1920 };
1921 struct user_arg_ptr envp = {
1922 .is_compat = true,
1923 .ptr.compat = __envp,
1924 };
1925 return do_execveat_common(fd, filename, argv, envp, flags);
1926}
1927#endif
1928
1929void set_binfmt(struct linux_binfmt *new)
1930{
1931 struct mm_struct *mm = current->mm;
1932
1933 if (mm->binfmt)
1934 module_put(mm->binfmt->module);
1935
1936 mm->binfmt = new;
1937 if (new)
1938 __module_get(new->module);
1939}
1940EXPORT_SYMBOL(set_binfmt);
1941
1942/*
1943 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1944 */
1945void set_dumpable(struct mm_struct *mm, int value)
1946{
1947 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1948 return;
1949
1950 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
1951}
1952
1953SYSCALL_DEFINE3(execve,
1954 const char __user *, filename,
1955 const char __user *const __user *, argv,
1956 const char __user *const __user *, envp)
1957{
1958 return do_execve(getname(filename), argv, envp);
1959}
1960
1961SYSCALL_DEFINE5(execveat,
1962 int, fd, const char __user *, filename,
1963 const char __user *const __user *, argv,
1964 const char __user *const __user *, envp,
1965 int, flags)
1966{
1967 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1968
1969 return do_execveat(fd,
1970 getname_flags(filename, lookup_flags, NULL),
1971 argv, envp, flags);
1972}
1973
1974#ifdef CONFIG_COMPAT
1975COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1976 const compat_uptr_t __user *, argv,
1977 const compat_uptr_t __user *, envp)
1978{
1979 return compat_do_execve(getname(filename), argv, envp);
1980}
1981
1982COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1983 const char __user *, filename,
1984 const compat_uptr_t __user *, argv,
1985 const compat_uptr_t __user *, envp,
1986 int, flags)
1987{
1988 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1989
1990 return compat_do_execveat(fd,
1991 getname_flags(filename, lookup_flags, NULL),
1992 argv, envp, flags);
1993}
1994#endif
1995