projects/linux·blob v4.12-rc4 · refs

1930 lines · 44.5 KiB

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