projects/linux·blob v3.9-rc1 · refs

1699 lines · 39.0 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/stat.h>
30#include <linux/fcntl.h>
31#include <linux/swap.h>
32#include <linux/string.h>
33#include <linux/init.h>
34#include <linux/pagemap.h>
35#include <linux/perf_event.h>
36#include <linux/highmem.h>
37#include <linux/spinlock.h>
38#include <linux/key.h>
39#include <linux/personality.h>
40#include <linux/binfmts.h>
41#include <linux/utsname.h>
42#include <linux/pid_namespace.h>
43#include <linux/module.h>
44#include <linux/namei.h>
45#include <linux/mount.h>
46#include <linux/security.h>
47#include <linux/syscalls.h>
48#include <linux/tsacct_kern.h>
49#include <linux/cn_proc.h>
50#include <linux/audit.h>
51#include <linux/tracehook.h>
52#include <linux/kmod.h>
53#include <linux/fsnotify.h>
54#include <linux/fs_struct.h>
55#include <linux/pipe_fs_i.h>
56#include <linux/oom.h>
57#include <linux/compat.h>
58
59#include <asm/uaccess.h>
60#include <asm/mmu_context.h>
61#include <asm/tlb.h>
62
63#include <trace/events/task.h>
64#include "internal.h"
65#include "coredump.h"
66
67#include <trace/events/sched.h>
68
69int suid_dumpable = 0;
70
71static LIST_HEAD(formats);
72static DEFINE_RWLOCK(binfmt_lock);
73
74void __register_binfmt(struct linux_binfmt * fmt, int insert)
75{
76 BUG_ON(!fmt);
77 write_lock(&binfmt_lock);
78 insert ? list_add(&fmt->lh, &formats) :
79 list_add_tail(&fmt->lh, &formats);
80 write_unlock(&binfmt_lock);
81}
82
83EXPORT_SYMBOL(__register_binfmt);
84
85void unregister_binfmt(struct linux_binfmt * fmt)
86{
87 write_lock(&binfmt_lock);
88 list_del(&fmt->lh);
89 write_unlock(&binfmt_lock);
90}
91
92EXPORT_SYMBOL(unregister_binfmt);
93
94static inline void put_binfmt(struct linux_binfmt * fmt)
95{
96 module_put(fmt->module);
97}
98
99/*
100 * Note that a shared library must be both readable and executable due to
101 * security reasons.
102 *
103 * Also note that we take the address to load from from the file itself.
104 */
105SYSCALL_DEFINE1(uselib, const char __user *, library)
106{
107 struct file *file;
108 struct filename *tmp = getname(library);
109 int error = PTR_ERR(tmp);
110 static const struct open_flags uselib_flags = {
111 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
112 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
113 .intent = LOOKUP_OPEN
114 };
115
116 if (IS_ERR(tmp))
117 goto out;
118
119 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
120 putname(tmp);
121 error = PTR_ERR(file);
122 if (IS_ERR(file))
123 goto out;
124
125 error = -EINVAL;
126 if (!S_ISREG(file_inode(file)->i_mode))
127 goto exit;
128
129 error = -EACCES;
130 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
131 goto exit;
132
133 fsnotify_open(file);
134
135 error = -ENOEXEC;
136 if(file->f_op) {
137 struct linux_binfmt * fmt;
138
139 read_lock(&binfmt_lock);
140 list_for_each_entry(fmt, &formats, lh) {
141 if (!fmt->load_shlib)
142 continue;
143 if (!try_module_get(fmt->module))
144 continue;
145 read_unlock(&binfmt_lock);
146 error = fmt->load_shlib(file);
147 read_lock(&binfmt_lock);
148 put_binfmt(fmt);
149 if (error != -ENOEXEC)
150 break;
151 }
152 read_unlock(&binfmt_lock);
153 }
154exit:
155 fput(file);
156out:
157 return error;
158}
159
160#ifdef CONFIG_MMU
161/*
162 * The nascent bprm->mm is not visible until exec_mmap() but it can
163 * use a lot of memory, account these pages in current->mm temporary
164 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
165 * change the counter back via acct_arg_size(0).
166 */
167static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
168{
169 struct mm_struct *mm = current->mm;
170 long diff = (long)(pages - bprm->vma_pages);
171
172 if (!mm || !diff)
173 return;
174
175 bprm->vma_pages = pages;
176 add_mm_counter(mm, MM_ANONPAGES, diff);
177}
178
179static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
180 int write)
181{
182 struct page *page;
183 int ret;
184
185#ifdef CONFIG_STACK_GROWSUP
186 if (write) {
187 ret = expand_downwards(bprm->vma, pos);
188 if (ret < 0)
189 return NULL;
190 }
191#endif
192 ret = get_user_pages(current, bprm->mm, pos,
193 1, write, 1, &page, NULL);
194 if (ret <= 0)
195 return NULL;
196
197 if (write) {
198 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
199 struct rlimit *rlim;
200
201 acct_arg_size(bprm, size / PAGE_SIZE);
202
203 /*
204 * We've historically supported up to 32 pages (ARG_MAX)
205 * of argument strings even with small stacks
206 */
207 if (size <= ARG_MAX)
208 return page;
209
210 /*
211 * Limit to 1/4-th the stack size for the argv+env strings.
212 * This ensures that:
213 * - the remaining binfmt code will not run out of stack space,
214 * - the program will have a reasonable amount of stack left
215 * to work from.
216 */
217 rlim = current->signal->rlim;
218 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
219 put_page(page);
220 return NULL;
221 }
222 }
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_page(struct linux_binprm *bprm, int i)
233{
234}
235
236static void free_arg_pages(struct linux_binprm *bprm)
237{
238}
239
240static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
241 struct page *page)
242{
243 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
244}
245
246static int __bprm_mm_init(struct linux_binprm *bprm)
247{
248 int err;
249 struct vm_area_struct *vma = NULL;
250 struct mm_struct *mm = bprm->mm;
251
252 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
253 if (!vma)
254 return -ENOMEM;
255
256 down_write(&mm->mmap_sem);
257 vma->vm_mm = mm;
258
259 /*
260 * Place the stack at the largest stack address the architecture
261 * supports. Later, we'll move this to an appropriate place. We don't
262 * use STACK_TOP because that can depend on attributes which aren't
263 * configured yet.
264 */
265 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
266 vma->vm_end = STACK_TOP_MAX;
267 vma->vm_start = vma->vm_end - PAGE_SIZE;
268 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
269 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
270 INIT_LIST_HEAD(&vma->anon_vma_chain);
271
272 err = insert_vm_struct(mm, vma);
273 if (err)
274 goto err;
275
276 mm->stack_vm = mm->total_vm = 1;
277 up_write(&mm->mmap_sem);
278 bprm->p = vma->vm_end - sizeof(void *);
279 return 0;
280err:
281 up_write(&mm->mmap_sem);
282 bprm->vma = NULL;
283 kmem_cache_free(vm_area_cachep, 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 err = init_new_context(current, mm);
369 if (err)
370 goto err;
371
372 err = __bprm_mm_init(bprm);
373 if (err)
374 goto err;
375
376 return 0;
377
378err:
379 if (mm) {
380 bprm->mm = NULL;
381 mmdrop(mm);
382 }
383
384 return err;
385}
386
387struct user_arg_ptr {
388#ifdef CONFIG_COMPAT
389 bool is_compat;
390#endif
391 union {
392 const char __user *const __user *native;
393#ifdef CONFIG_COMPAT
394 const compat_uptr_t __user *compat;
395#endif
396 } ptr;
397};
398
399static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
400{
401 const char __user *native;
402
403#ifdef CONFIG_COMPAT
404 if (unlikely(argv.is_compat)) {
405 compat_uptr_t compat;
406
407 if (get_user(compat, argv.ptr.compat + nr))
408 return ERR_PTR(-EFAULT);
409
410 return compat_ptr(compat);
411 }
412#endif
413
414 if (get_user(native, argv.ptr.native + nr))
415 return ERR_PTR(-EFAULT);
416
417 return native;
418}
419
420/*
421 * count() counts the number of strings in array ARGV.
422 */
423static int count(struct user_arg_ptr argv, int max)
424{
425 int i = 0;
426
427 if (argv.ptr.native != NULL) {
428 for (;;) {
429 const char __user *p = get_user_arg_ptr(argv, i);
430
431 if (!p)
432 break;
433
434 if (IS_ERR(p))
435 return -EFAULT;
436
437 if (i >= max)
438 return -E2BIG;
439 ++i;
440
441 if (fatal_signal_pending(current))
442 return -ERESTARTNOHAND;
443 cond_resched();
444 }
445 }
446 return i;
447}
448
449/*
450 * 'copy_strings()' copies argument/environment strings from the old
451 * processes's memory to the new process's stack. The call to get_user_pages()
452 * ensures the destination page is created and not swapped out.
453 */
454static int copy_strings(int argc, struct user_arg_ptr argv,
455 struct linux_binprm *bprm)
456{
457 struct page *kmapped_page = NULL;
458 char *kaddr = NULL;
459 unsigned long kpos = 0;
460 int ret;
461
462 while (argc-- > 0) {
463 const char __user *str;
464 int len;
465 unsigned long pos;
466
467 ret = -EFAULT;
468 str = get_user_arg_ptr(argv, argc);
469 if (IS_ERR(str))
470 goto out;
471
472 len = strnlen_user(str, MAX_ARG_STRLEN);
473 if (!len)
474 goto out;
475
476 ret = -E2BIG;
477 if (!valid_arg_len(bprm, len))
478 goto out;
479
480 /* We're going to work our way backwords. */
481 pos = bprm->p;
482 str += len;
483 bprm->p -= len;
484
485 while (len > 0) {
486 int offset, bytes_to_copy;
487
488 if (fatal_signal_pending(current)) {
489 ret = -ERESTARTNOHAND;
490 goto out;
491 }
492 cond_resched();
493
494 offset = pos % PAGE_SIZE;
495 if (offset == 0)
496 offset = PAGE_SIZE;
497
498 bytes_to_copy = offset;
499 if (bytes_to_copy > len)
500 bytes_to_copy = len;
501
502 offset -= bytes_to_copy;
503 pos -= bytes_to_copy;
504 str -= bytes_to_copy;
505 len -= bytes_to_copy;
506
507 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
508 struct page *page;
509
510 page = get_arg_page(bprm, pos, 1);
511 if (!page) {
512 ret = -E2BIG;
513 goto out;
514 }
515
516 if (kmapped_page) {
517 flush_kernel_dcache_page(kmapped_page);
518 kunmap(kmapped_page);
519 put_arg_page(kmapped_page);
520 }
521 kmapped_page = page;
522 kaddr = kmap(kmapped_page);
523 kpos = pos & PAGE_MASK;
524 flush_arg_page(bprm, kpos, kmapped_page);
525 }
526 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
527 ret = -EFAULT;
528 goto out;
529 }
530 }
531 }
532 ret = 0;
533out:
534 if (kmapped_page) {
535 flush_kernel_dcache_page(kmapped_page);
536 kunmap(kmapped_page);
537 put_arg_page(kmapped_page);
538 }
539 return ret;
540}
541
542/*
543 * Like copy_strings, but get argv and its values from kernel memory.
544 */
545int copy_strings_kernel(int argc, const char *const *__argv,
546 struct linux_binprm *bprm)
547{
548 int r;
549 mm_segment_t oldfs = get_fs();
550 struct user_arg_ptr argv = {
551 .ptr.native = (const char __user *const __user *)__argv,
552 };
553
554 set_fs(KERNEL_DS);
555 r = copy_strings(argc, argv, bprm);
556 set_fs(oldfs);
557
558 return r;
559}
560EXPORT_SYMBOL(copy_strings_kernel);
561
562#ifdef CONFIG_MMU
563
564/*
565 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
566 * the binfmt code determines where the new stack should reside, we shift it to
567 * its final location. The process proceeds as follows:
568 *
569 * 1) Use shift to calculate the new vma endpoints.
570 * 2) Extend vma to cover both the old and new ranges. This ensures the
571 * arguments passed to subsequent functions are consistent.
572 * 3) Move vma's page tables to the new range.
573 * 4) Free up any cleared pgd range.
574 * 5) Shrink the vma to cover only the new range.
575 */
576static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
577{
578 struct mm_struct *mm = vma->vm_mm;
579 unsigned long old_start = vma->vm_start;
580 unsigned long old_end = vma->vm_end;
581 unsigned long length = old_end - old_start;
582 unsigned long new_start = old_start - shift;
583 unsigned long new_end = old_end - shift;
584 struct mmu_gather tlb;
585
586 BUG_ON(new_start > new_end);
587
588 /*
589 * ensure there are no vmas between where we want to go
590 * and where we are
591 */
592 if (vma != find_vma(mm, new_start))
593 return -EFAULT;
594
595 /*
596 * cover the whole range: [new_start, old_end)
597 */
598 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
599 return -ENOMEM;
600
601 /*
602 * move the page tables downwards, on failure we rely on
603 * process cleanup to remove whatever mess we made.
604 */
605 if (length != move_page_tables(vma, old_start,
606 vma, new_start, length, false))
607 return -ENOMEM;
608
609 lru_add_drain();
610 tlb_gather_mmu(&tlb, mm, 0);
611 if (new_end > old_start) {
612 /*
613 * when the old and new regions overlap clear from new_end.
614 */
615 free_pgd_range(&tlb, new_end, old_end, new_end,
616 vma->vm_next ? vma->vm_next->vm_start : 0);
617 } else {
618 /*
619 * otherwise, clean from old_start; this is done to not touch
620 * the address space in [new_end, old_start) some architectures
621 * have constraints on va-space that make this illegal (IA64) -
622 * for the others its just a little faster.
623 */
624 free_pgd_range(&tlb, old_start, old_end, new_end,
625 vma->vm_next ? vma->vm_next->vm_start : 0);
626 }
627 tlb_finish_mmu(&tlb, new_end, old_end);
628
629 /*
630 * Shrink the vma to just the new range. Always succeeds.
631 */
632 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
633
634 return 0;
635}
636
637/*
638 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
639 * the stack is optionally relocated, and some extra space is added.
640 */
641int setup_arg_pages(struct linux_binprm *bprm,
642 unsigned long stack_top,
643 int executable_stack)
644{
645 unsigned long ret;
646 unsigned long stack_shift;
647 struct mm_struct *mm = current->mm;
648 struct vm_area_struct *vma = bprm->vma;
649 struct vm_area_struct *prev = NULL;
650 unsigned long vm_flags;
651 unsigned long stack_base;
652 unsigned long stack_size;
653 unsigned long stack_expand;
654 unsigned long rlim_stack;
655
656#ifdef CONFIG_STACK_GROWSUP
657 /* Limit stack size to 1GB */
658 stack_base = rlimit_max(RLIMIT_STACK);
659 if (stack_base > (1 << 30))
660 stack_base = 1 << 30;
661
662 /* Make sure we didn't let the argument array grow too large. */
663 if (vma->vm_end - vma->vm_start > stack_base)
664 return -ENOMEM;
665
666 stack_base = PAGE_ALIGN(stack_top - stack_base);
667
668 stack_shift = vma->vm_start - stack_base;
669 mm->arg_start = bprm->p - stack_shift;
670 bprm->p = vma->vm_end - stack_shift;
671#else
672 stack_top = arch_align_stack(stack_top);
673 stack_top = PAGE_ALIGN(stack_top);
674
675 if (unlikely(stack_top < mmap_min_addr) ||
676 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
677 return -ENOMEM;
678
679 stack_shift = vma->vm_end - stack_top;
680
681 bprm->p -= stack_shift;
682 mm->arg_start = bprm->p;
683#endif
684
685 if (bprm->loader)
686 bprm->loader -= stack_shift;
687 bprm->exec -= stack_shift;
688
689 down_write(&mm->mmap_sem);
690 vm_flags = VM_STACK_FLAGS;
691
692 /*
693 * Adjust stack execute permissions; explicitly enable for
694 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
695 * (arch default) otherwise.
696 */
697 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
698 vm_flags |= VM_EXEC;
699 else if (executable_stack == EXSTACK_DISABLE_X)
700 vm_flags &= ~VM_EXEC;
701 vm_flags |= mm->def_flags;
702 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
703
704 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
705 vm_flags);
706 if (ret)
707 goto out_unlock;
708 BUG_ON(prev != vma);
709
710 /* Move stack pages down in memory. */
711 if (stack_shift) {
712 ret = shift_arg_pages(vma, stack_shift);
713 if (ret)
714 goto out_unlock;
715 }
716
717 /* mprotect_fixup is overkill to remove the temporary stack flags */
718 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
719
720 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
721 stack_size = vma->vm_end - vma->vm_start;
722 /*
723 * Align this down to a page boundary as expand_stack
724 * will align it up.
725 */
726 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
727#ifdef CONFIG_STACK_GROWSUP
728 if (stack_size + stack_expand > rlim_stack)
729 stack_base = vma->vm_start + rlim_stack;
730 else
731 stack_base = vma->vm_end + stack_expand;
732#else
733 if (stack_size + stack_expand > rlim_stack)
734 stack_base = vma->vm_end - rlim_stack;
735 else
736 stack_base = vma->vm_start - stack_expand;
737#endif
738 current->mm->start_stack = bprm->p;
739 ret = expand_stack(vma, stack_base);
740 if (ret)
741 ret = -EFAULT;
742
743out_unlock:
744 up_write(&mm->mmap_sem);
745 return ret;
746}
747EXPORT_SYMBOL(setup_arg_pages);
748
749#endif /* CONFIG_MMU */
750
751struct file *open_exec(const char *name)
752{
753 struct file *file;
754 int err;
755 struct filename tmp = { .name = name };
756 static const struct open_flags open_exec_flags = {
757 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
758 .acc_mode = MAY_EXEC | MAY_OPEN,
759 .intent = LOOKUP_OPEN
760 };
761
762 file = do_filp_open(AT_FDCWD, &tmp, &open_exec_flags, LOOKUP_FOLLOW);
763 if (IS_ERR(file))
764 goto out;
765
766 err = -EACCES;
767 if (!S_ISREG(file_inode(file)->i_mode))
768 goto exit;
769
770 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
771 goto exit;
772
773 fsnotify_open(file);
774
775 err = deny_write_access(file);
776 if (err)
777 goto exit;
778
779out:
780 return file;
781
782exit:
783 fput(file);
784 return ERR_PTR(err);
785}
786EXPORT_SYMBOL(open_exec);
787
788int kernel_read(struct file *file, loff_t offset,
789 char *addr, unsigned long count)
790{
791 mm_segment_t old_fs;
792 loff_t pos = offset;
793 int result;
794
795 old_fs = get_fs();
796 set_fs(get_ds());
797 /* The cast to a user pointer is valid due to the set_fs() */
798 result = vfs_read(file, (void __user *)addr, count, &pos);
799 set_fs(old_fs);
800 return result;
801}
802
803EXPORT_SYMBOL(kernel_read);
804
805static int exec_mmap(struct mm_struct *mm)
806{
807 struct task_struct *tsk;
808 struct mm_struct * old_mm, *active_mm;
809
810 /* Notify parent that we're no longer interested in the old VM */
811 tsk = current;
812 old_mm = current->mm;
813 mm_release(tsk, old_mm);
814
815 if (old_mm) {
816 sync_mm_rss(old_mm);
817 /*
818 * Make sure that if there is a core dump in progress
819 * for the old mm, we get out and die instead of going
820 * through with the exec. We must hold mmap_sem around
821 * checking core_state and changing tsk->mm.
822 */
823 down_read(&old_mm->mmap_sem);
824 if (unlikely(old_mm->core_state)) {
825 up_read(&old_mm->mmap_sem);
826 return -EINTR;
827 }
828 }
829 task_lock(tsk);
830 active_mm = tsk->active_mm;
831 tsk->mm = mm;
832 tsk->active_mm = mm;
833 activate_mm(active_mm, mm);
834 task_unlock(tsk);
835 arch_pick_mmap_layout(mm);
836 if (old_mm) {
837 up_read(&old_mm->mmap_sem);
838 BUG_ON(active_mm != old_mm);
839 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
840 mm_update_next_owner(old_mm);
841 mmput(old_mm);
842 return 0;
843 }
844 mmdrop(active_mm);
845 return 0;
846}
847
848/*
849 * This function makes sure the current process has its own signal table,
850 * so that flush_signal_handlers can later reset the handlers without
851 * disturbing other processes. (Other processes might share the signal
852 * table via the CLONE_SIGHAND option to clone().)
853 */
854static int de_thread(struct task_struct *tsk)
855{
856 struct signal_struct *sig = tsk->signal;
857 struct sighand_struct *oldsighand = tsk->sighand;
858 spinlock_t *lock = &oldsighand->siglock;
859
860 if (thread_group_empty(tsk))
861 goto no_thread_group;
862
863 /*
864 * Kill all other threads in the thread group.
865 */
866 spin_lock_irq(lock);
867 if (signal_group_exit(sig)) {
868 /*
869 * Another group action in progress, just
870 * return so that the signal is processed.
871 */
872 spin_unlock_irq(lock);
873 return -EAGAIN;
874 }
875
876 sig->group_exit_task = tsk;
877 sig->notify_count = zap_other_threads(tsk);
878 if (!thread_group_leader(tsk))
879 sig->notify_count--;
880
881 while (sig->notify_count) {
882 __set_current_state(TASK_KILLABLE);
883 spin_unlock_irq(lock);
884 schedule();
885 if (unlikely(__fatal_signal_pending(tsk)))
886 goto killed;
887 spin_lock_irq(lock);
888 }
889 spin_unlock_irq(lock);
890
891 /*
892 * At this point all other threads have exited, all we have to
893 * do is to wait for the thread group leader to become inactive,
894 * and to assume its PID:
895 */
896 if (!thread_group_leader(tsk)) {
897 struct task_struct *leader = tsk->group_leader;
898
899 sig->notify_count = -1; /* for exit_notify() */
900 for (;;) {
901 write_lock_irq(&tasklist_lock);
902 if (likely(leader->exit_state))
903 break;
904 __set_current_state(TASK_KILLABLE);
905 write_unlock_irq(&tasklist_lock);
906 schedule();
907 if (unlikely(__fatal_signal_pending(tsk)))
908 goto killed;
909 }
910
911 /*
912 * The only record we have of the real-time age of a
913 * process, regardless of execs it's done, is start_time.
914 * All the past CPU time is accumulated in signal_struct
915 * from sister threads now dead. But in this non-leader
916 * exec, nothing survives from the original leader thread,
917 * whose birth marks the true age of this process now.
918 * When we take on its identity by switching to its PID, we
919 * also take its birthdate (always earlier than our own).
920 */
921 tsk->start_time = leader->start_time;
922
923 BUG_ON(!same_thread_group(leader, tsk));
924 BUG_ON(has_group_leader_pid(tsk));
925 /*
926 * An exec() starts a new thread group with the
927 * TGID of the previous thread group. Rehash the
928 * two threads with a switched PID, and release
929 * the former thread group leader:
930 */
931
932 /* Become a process group leader with the old leader's pid.
933 * The old leader becomes a thread of the this thread group.
934 * Note: The old leader also uses this pid until release_task
935 * is called. Odd but simple and correct.
936 */
937 detach_pid(tsk, PIDTYPE_PID);
938 tsk->pid = leader->pid;
939 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
940 transfer_pid(leader, tsk, PIDTYPE_PGID);
941 transfer_pid(leader, tsk, PIDTYPE_SID);
942
943 list_replace_rcu(&leader->tasks, &tsk->tasks);
944 list_replace_init(&leader->sibling, &tsk->sibling);
945
946 tsk->group_leader = tsk;
947 leader->group_leader = tsk;
948
949 tsk->exit_signal = SIGCHLD;
950 leader->exit_signal = -1;
951
952 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
953 leader->exit_state = EXIT_DEAD;
954
955 /*
956 * We are going to release_task()->ptrace_unlink() silently,
957 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
958 * the tracer wont't block again waiting for this thread.
959 */
960 if (unlikely(leader->ptrace))
961 __wake_up_parent(leader, leader->parent);
962 write_unlock_irq(&tasklist_lock);
963
964 release_task(leader);
965 }
966
967 sig->group_exit_task = NULL;
968 sig->notify_count = 0;
969
970no_thread_group:
971 /* we have changed execution domain */
972 tsk->exit_signal = SIGCHLD;
973
974 exit_itimers(sig);
975 flush_itimer_signals();
976
977 if (atomic_read(&oldsighand->count) != 1) {
978 struct sighand_struct *newsighand;
979 /*
980 * This ->sighand is shared with the CLONE_SIGHAND
981 * but not CLONE_THREAD task, switch to the new one.
982 */
983 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
984 if (!newsighand)
985 return -ENOMEM;
986
987 atomic_set(&newsighand->count, 1);
988 memcpy(newsighand->action, oldsighand->action,
989 sizeof(newsighand->action));
990
991 write_lock_irq(&tasklist_lock);
992 spin_lock(&oldsighand->siglock);
993 rcu_assign_pointer(tsk->sighand, newsighand);
994 spin_unlock(&oldsighand->siglock);
995 write_unlock_irq(&tasklist_lock);
996
997 __cleanup_sighand(oldsighand);
998 }
999
1000 BUG_ON(!thread_group_leader(tsk));
1001 return 0;
1002
1003killed:
1004 /* protects against exit_notify() and __exit_signal() */
1005 read_lock(&tasklist_lock);
1006 sig->group_exit_task = NULL;
1007 sig->notify_count = 0;
1008 read_unlock(&tasklist_lock);
1009 return -EAGAIN;
1010}
1011
1012char *get_task_comm(char *buf, struct task_struct *tsk)
1013{
1014 /* buf must be at least sizeof(tsk->comm) in size */
1015 task_lock(tsk);
1016 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1017 task_unlock(tsk);
1018 return buf;
1019}
1020EXPORT_SYMBOL_GPL(get_task_comm);
1021
1022/*
1023 * These functions flushes out all traces of the currently running executable
1024 * so that a new one can be started
1025 */
1026
1027void set_task_comm(struct task_struct *tsk, char *buf)
1028{
1029 task_lock(tsk);
1030
1031 trace_task_rename(tsk, buf);
1032
1033 /*
1034 * Threads may access current->comm without holding
1035 * the task lock, so write the string carefully.
1036 * Readers without a lock may see incomplete new
1037 * names but are safe from non-terminating string reads.
1038 */
1039 memset(tsk->comm, 0, TASK_COMM_LEN);
1040 wmb();
1041 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1042 task_unlock(tsk);
1043 perf_event_comm(tsk);
1044}
1045
1046static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1047{
1048 int i, ch;
1049
1050 /* Copies the binary name from after last slash */
1051 for (i = 0; (ch = *(fn++)) != '\0';) {
1052 if (ch == '/')
1053 i = 0; /* overwrite what we wrote */
1054 else
1055 if (i < len - 1)
1056 tcomm[i++] = ch;
1057 }
1058 tcomm[i] = '\0';
1059}
1060
1061int flush_old_exec(struct linux_binprm * bprm)
1062{
1063 int retval;
1064
1065 /*
1066 * Make sure we have a private signal table and that
1067 * we are unassociated from the previous thread group.
1068 */
1069 retval = de_thread(current);
1070 if (retval)
1071 goto out;
1072
1073 set_mm_exe_file(bprm->mm, bprm->file);
1074
1075 filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1076 /*
1077 * Release all of the old mmap stuff
1078 */
1079 acct_arg_size(bprm, 0);
1080 retval = exec_mmap(bprm->mm);
1081 if (retval)
1082 goto out;
1083
1084 bprm->mm = NULL; /* We're using it now */
1085
1086 set_fs(USER_DS);
1087 current->flags &=
1088 ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | PF_NOFREEZE);
1089 flush_thread();
1090 current->personality &= ~bprm->per_clear;
1091
1092 return 0;
1093
1094out:
1095 return retval;
1096}
1097EXPORT_SYMBOL(flush_old_exec);
1098
1099void would_dump(struct linux_binprm *bprm, struct file *file)
1100{
1101 if (inode_permission(file_inode(file), MAY_READ) < 0)
1102 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1103}
1104EXPORT_SYMBOL(would_dump);
1105
1106void setup_new_exec(struct linux_binprm * bprm)
1107{
1108 arch_pick_mmap_layout(current->mm);
1109
1110 /* This is the point of no return */
1111 current->sas_ss_sp = current->sas_ss_size = 0;
1112
1113 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1114 set_dumpable(current->mm, SUID_DUMP_USER);
1115 else
1116 set_dumpable(current->mm, suid_dumpable);
1117
1118 set_task_comm(current, bprm->tcomm);
1119
1120 /* Set the new mm task size. We have to do that late because it may
1121 * depend on TIF_32BIT which is only updated in flush_thread() on
1122 * some architectures like powerpc
1123 */
1124 current->mm->task_size = TASK_SIZE;
1125
1126 /* install the new credentials */
1127 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1128 !gid_eq(bprm->cred->gid, current_egid())) {
1129 current->pdeath_signal = 0;
1130 } else {
1131 would_dump(bprm, bprm->file);
1132 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1133 set_dumpable(current->mm, suid_dumpable);
1134 }
1135
1136 /*
1137 * Flush performance counters when crossing a
1138 * security domain:
1139 */
1140 if (!get_dumpable(current->mm))
1141 perf_event_exit_task(current);
1142
1143 /* An exec changes our domain. We are no longer part of the thread
1144 group */
1145
1146 current->self_exec_id++;
1147
1148 flush_signal_handlers(current, 0);
1149 do_close_on_exec(current->files);
1150}
1151EXPORT_SYMBOL(setup_new_exec);
1152
1153/*
1154 * Prepare credentials and lock ->cred_guard_mutex.
1155 * install_exec_creds() commits the new creds and drops the lock.
1156 * Or, if exec fails before, free_bprm() should release ->cred and
1157 * and unlock.
1158 */
1159int prepare_bprm_creds(struct linux_binprm *bprm)
1160{
1161 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1162 return -ERESTARTNOINTR;
1163
1164 bprm->cred = prepare_exec_creds();
1165 if (likely(bprm->cred))
1166 return 0;
1167
1168 mutex_unlock(&current->signal->cred_guard_mutex);
1169 return -ENOMEM;
1170}
1171
1172void free_bprm(struct linux_binprm *bprm)
1173{
1174 free_arg_pages(bprm);
1175 if (bprm->cred) {
1176 mutex_unlock(&current->signal->cred_guard_mutex);
1177 abort_creds(bprm->cred);
1178 }
1179 /* If a binfmt changed the interp, free it. */
1180 if (bprm->interp != bprm->filename)
1181 kfree(bprm->interp);
1182 kfree(bprm);
1183}
1184
1185int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1186{
1187 /* If a binfmt changed the interp, free it first. */
1188 if (bprm->interp != bprm->filename)
1189 kfree(bprm->interp);
1190 bprm->interp = kstrdup(interp, GFP_KERNEL);
1191 if (!bprm->interp)
1192 return -ENOMEM;
1193 return 0;
1194}
1195EXPORT_SYMBOL(bprm_change_interp);
1196
1197/*
1198 * install the new credentials for this executable
1199 */
1200void install_exec_creds(struct linux_binprm *bprm)
1201{
1202 security_bprm_committing_creds(bprm);
1203
1204 commit_creds(bprm->cred);
1205 bprm->cred = NULL;
1206 /*
1207 * cred_guard_mutex must be held at least to this point to prevent
1208 * ptrace_attach() from altering our determination of the task's
1209 * credentials; any time after this it may be unlocked.
1210 */
1211 security_bprm_committed_creds(bprm);
1212 mutex_unlock(&current->signal->cred_guard_mutex);
1213}
1214EXPORT_SYMBOL(install_exec_creds);
1215
1216/*
1217 * determine how safe it is to execute the proposed program
1218 * - the caller must hold ->cred_guard_mutex to protect against
1219 * PTRACE_ATTACH
1220 */
1221static int check_unsafe_exec(struct linux_binprm *bprm)
1222{
1223 struct task_struct *p = current, *t;
1224 unsigned n_fs;
1225 int res = 0;
1226
1227 if (p->ptrace) {
1228 if (p->ptrace & PT_PTRACE_CAP)
1229 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1230 else
1231 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1232 }
1233
1234 /*
1235 * This isn't strictly necessary, but it makes it harder for LSMs to
1236 * mess up.
1237 */
1238 if (current->no_new_privs)
1239 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1240
1241 n_fs = 1;
1242 spin_lock(&p->fs->lock);
1243 rcu_read_lock();
1244 for (t = next_thread(p); t != p; t = next_thread(t)) {
1245 if (t->fs == p->fs)
1246 n_fs++;
1247 }
1248 rcu_read_unlock();
1249
1250 if (p->fs->users > n_fs) {
1251 bprm->unsafe |= LSM_UNSAFE_SHARE;
1252 } else {
1253 res = -EAGAIN;
1254 if (!p->fs->in_exec) {
1255 p->fs->in_exec = 1;
1256 res = 1;
1257 }
1258 }
1259 spin_unlock(&p->fs->lock);
1260
1261 return res;
1262}
1263
1264/*
1265 * Fill the binprm structure from the inode.
1266 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1267 *
1268 * This may be called multiple times for binary chains (scripts for example).
1269 */
1270int prepare_binprm(struct linux_binprm *bprm)
1271{
1272 umode_t mode;
1273 struct inode * inode = file_inode(bprm->file);
1274 int retval;
1275
1276 mode = inode->i_mode;
1277 if (bprm->file->f_op == NULL)
1278 return -EACCES;
1279
1280 /* clear any previous set[ug]id data from a previous binary */
1281 bprm->cred->euid = current_euid();
1282 bprm->cred->egid = current_egid();
1283
1284 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1285 !current->no_new_privs &&
1286 kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1287 kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1288 /* Set-uid? */
1289 if (mode & S_ISUID) {
1290 bprm->per_clear |= PER_CLEAR_ON_SETID;
1291 bprm->cred->euid = inode->i_uid;
1292 }
1293
1294 /* Set-gid? */
1295 /*
1296 * If setgid is set but no group execute bit then this
1297 * is a candidate for mandatory locking, not a setgid
1298 * executable.
1299 */
1300 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1301 bprm->per_clear |= PER_CLEAR_ON_SETID;
1302 bprm->cred->egid = inode->i_gid;
1303 }
1304 }
1305
1306 /* fill in binprm security blob */
1307 retval = security_bprm_set_creds(bprm);
1308 if (retval)
1309 return retval;
1310 bprm->cred_prepared = 1;
1311
1312 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1313 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1314}
1315
1316EXPORT_SYMBOL(prepare_binprm);
1317
1318/*
1319 * Arguments are '\0' separated strings found at the location bprm->p
1320 * points to; chop off the first by relocating brpm->p to right after
1321 * the first '\0' encountered.
1322 */
1323int remove_arg_zero(struct linux_binprm *bprm)
1324{
1325 int ret = 0;
1326 unsigned long offset;
1327 char *kaddr;
1328 struct page *page;
1329
1330 if (!bprm->argc)
1331 return 0;
1332
1333 do {
1334 offset = bprm->p & ~PAGE_MASK;
1335 page = get_arg_page(bprm, bprm->p, 0);
1336 if (!page) {
1337 ret = -EFAULT;
1338 goto out;
1339 }
1340 kaddr = kmap_atomic(page);
1341
1342 for (; offset < PAGE_SIZE && kaddr[offset];
1343 offset++, bprm->p++)
1344 ;
1345
1346 kunmap_atomic(kaddr);
1347 put_arg_page(page);
1348
1349 if (offset == PAGE_SIZE)
1350 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1351 } while (offset == PAGE_SIZE);
1352
1353 bprm->p++;
1354 bprm->argc--;
1355 ret = 0;
1356
1357out:
1358 return ret;
1359}
1360EXPORT_SYMBOL(remove_arg_zero);
1361
1362/*
1363 * cycle the list of binary formats handler, until one recognizes the image
1364 */
1365int search_binary_handler(struct linux_binprm *bprm)
1366{
1367 unsigned int depth = bprm->recursion_depth;
1368 int try,retval;
1369 struct linux_binfmt *fmt;
1370 pid_t old_pid, old_vpid;
1371
1372 /* This allows 4 levels of binfmt rewrites before failing hard. */
1373 if (depth > 5)
1374 return -ELOOP;
1375
1376 retval = security_bprm_check(bprm);
1377 if (retval)
1378 return retval;
1379
1380 retval = audit_bprm(bprm);
1381 if (retval)
1382 return retval;
1383
1384 /* Need to fetch pid before load_binary changes it */
1385 old_pid = current->pid;
1386 rcu_read_lock();
1387 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1388 rcu_read_unlock();
1389
1390 retval = -ENOENT;
1391 for (try=0; try<2; try++) {
1392 read_lock(&binfmt_lock);
1393 list_for_each_entry(fmt, &formats, lh) {
1394 int (*fn)(struct linux_binprm *) = fmt->load_binary;
1395 if (!fn)
1396 continue;
1397 if (!try_module_get(fmt->module))
1398 continue;
1399 read_unlock(&binfmt_lock);
1400 bprm->recursion_depth = depth + 1;
1401 retval = fn(bprm);
1402 bprm->recursion_depth = depth;
1403 if (retval >= 0) {
1404 if (depth == 0) {
1405 trace_sched_process_exec(current, old_pid, bprm);
1406 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1407 }
1408 put_binfmt(fmt);
1409 allow_write_access(bprm->file);
1410 if (bprm->file)
1411 fput(bprm->file);
1412 bprm->file = NULL;
1413 current->did_exec = 1;
1414 proc_exec_connector(current);
1415 return retval;
1416 }
1417 read_lock(&binfmt_lock);
1418 put_binfmt(fmt);
1419 if (retval != -ENOEXEC || bprm->mm == NULL)
1420 break;
1421 if (!bprm->file) {
1422 read_unlock(&binfmt_lock);
1423 return retval;
1424 }
1425 }
1426 read_unlock(&binfmt_lock);
1427#ifdef CONFIG_MODULES
1428 if (retval != -ENOEXEC || bprm->mm == NULL) {
1429 break;
1430 } else {
1431#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1432 if (printable(bprm->buf[0]) &&
1433 printable(bprm->buf[1]) &&
1434 printable(bprm->buf[2]) &&
1435 printable(bprm->buf[3]))
1436 break; /* -ENOEXEC */
1437 if (try)
1438 break; /* -ENOEXEC */
1439 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1440 }
1441#else
1442 break;
1443#endif
1444 }
1445 return retval;
1446}
1447
1448EXPORT_SYMBOL(search_binary_handler);
1449
1450/*
1451 * sys_execve() executes a new program.
1452 */
1453static int do_execve_common(const char *filename,
1454 struct user_arg_ptr argv,
1455 struct user_arg_ptr envp)
1456{
1457 struct linux_binprm *bprm;
1458 struct file *file;
1459 struct files_struct *displaced;
1460 bool clear_in_exec;
1461 int retval;
1462 const struct cred *cred = current_cred();
1463
1464 /*
1465 * We move the actual failure in case of RLIMIT_NPROC excess from
1466 * set*uid() to execve() because too many poorly written programs
1467 * don't check setuid() return code. Here we additionally recheck
1468 * whether NPROC limit is still exceeded.
1469 */
1470 if ((current->flags & PF_NPROC_EXCEEDED) &&
1471 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1472 retval = -EAGAIN;
1473 goto out_ret;
1474 }
1475
1476 /* We're below the limit (still or again), so we don't want to make
1477 * further execve() calls fail. */
1478 current->flags &= ~PF_NPROC_EXCEEDED;
1479
1480 retval = unshare_files(&displaced);
1481 if (retval)
1482 goto out_ret;
1483
1484 retval = -ENOMEM;
1485 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1486 if (!bprm)
1487 goto out_files;
1488
1489 retval = prepare_bprm_creds(bprm);
1490 if (retval)
1491 goto out_free;
1492
1493 retval = check_unsafe_exec(bprm);
1494 if (retval < 0)
1495 goto out_free;
1496 clear_in_exec = retval;
1497 current->in_execve = 1;
1498
1499 file = open_exec(filename);
1500 retval = PTR_ERR(file);
1501 if (IS_ERR(file))
1502 goto out_unmark;
1503
1504 sched_exec();
1505
1506 bprm->file = file;
1507 bprm->filename = filename;
1508 bprm->interp = filename;
1509
1510 retval = bprm_mm_init(bprm);
1511 if (retval)
1512 goto out_file;
1513
1514 bprm->argc = count(argv, MAX_ARG_STRINGS);
1515 if ((retval = bprm->argc) < 0)
1516 goto out;
1517
1518 bprm->envc = count(envp, MAX_ARG_STRINGS);
1519 if ((retval = bprm->envc) < 0)
1520 goto out;
1521
1522 retval = prepare_binprm(bprm);
1523 if (retval < 0)
1524 goto out;
1525
1526 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1527 if (retval < 0)
1528 goto out;
1529
1530 bprm->exec = bprm->p;
1531 retval = copy_strings(bprm->envc, envp, bprm);
1532 if (retval < 0)
1533 goto out;
1534
1535 retval = copy_strings(bprm->argc, argv, bprm);
1536 if (retval < 0)
1537 goto out;
1538
1539 retval = search_binary_handler(bprm);
1540 if (retval < 0)
1541 goto out;
1542
1543 /* execve succeeded */
1544 current->fs->in_exec = 0;
1545 current->in_execve = 0;
1546 acct_update_integrals(current);
1547 free_bprm(bprm);
1548 if (displaced)
1549 put_files_struct(displaced);
1550 return retval;
1551
1552out:
1553 if (bprm->mm) {
1554 acct_arg_size(bprm, 0);
1555 mmput(bprm->mm);
1556 }
1557
1558out_file:
1559 if (bprm->file) {
1560 allow_write_access(bprm->file);
1561 fput(bprm->file);
1562 }
1563
1564out_unmark:
1565 if (clear_in_exec)
1566 current->fs->in_exec = 0;
1567 current->in_execve = 0;
1568
1569out_free:
1570 free_bprm(bprm);
1571
1572out_files:
1573 if (displaced)
1574 reset_files_struct(displaced);
1575out_ret:
1576 return retval;
1577}
1578
1579int do_execve(const char *filename,
1580 const char __user *const __user *__argv,
1581 const char __user *const __user *__envp)
1582{
1583 struct user_arg_ptr argv = { .ptr.native = __argv };
1584 struct user_arg_ptr envp = { .ptr.native = __envp };
1585 return do_execve_common(filename, argv, envp);
1586}
1587
1588#ifdef CONFIG_COMPAT
1589static int compat_do_execve(const char *filename,
1590 const compat_uptr_t __user *__argv,
1591 const compat_uptr_t __user *__envp)
1592{
1593 struct user_arg_ptr argv = {
1594 .is_compat = true,
1595 .ptr.compat = __argv,
1596 };
1597 struct user_arg_ptr envp = {
1598 .is_compat = true,
1599 .ptr.compat = __envp,
1600 };
1601 return do_execve_common(filename, argv, envp);
1602}
1603#endif
1604
1605void set_binfmt(struct linux_binfmt *new)
1606{
1607 struct mm_struct *mm = current->mm;
1608
1609 if (mm->binfmt)
1610 module_put(mm->binfmt->module);
1611
1612 mm->binfmt = new;
1613 if (new)
1614 __module_get(new->module);
1615}
1616
1617EXPORT_SYMBOL(set_binfmt);
1618
1619/*
1620 * set_dumpable converts traditional three-value dumpable to two flags and
1621 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1622 * these bits are not changed atomically. So get_dumpable can observe the
1623 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1624 * return either old dumpable or new one by paying attention to the order of
1625 * modifying the bits.
1626 *
1627 * dumpable | mm->flags (binary)
1628 * old new | initial interim final
1629 * ---------+-----------------------
1630 * 0 1 | 00 01 01
1631 * 0 2 | 00 10(*) 11
1632 * 1 0 | 01 00 00
1633 * 1 2 | 01 11 11
1634 * 2 0 | 11 10(*) 00
1635 * 2 1 | 11 11 01
1636 *
1637 * (*) get_dumpable regards interim value of 10 as 11.
1638 */
1639void set_dumpable(struct mm_struct *mm, int value)
1640{
1641 switch (value) {
1642 case SUID_DUMP_DISABLE:
1643 clear_bit(MMF_DUMPABLE, &mm->flags);
1644 smp_wmb();
1645 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1646 break;
1647 case SUID_DUMP_USER:
1648 set_bit(MMF_DUMPABLE, &mm->flags);
1649 smp_wmb();
1650 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1651 break;
1652 case SUID_DUMP_ROOT:
1653 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1654 smp_wmb();
1655 set_bit(MMF_DUMPABLE, &mm->flags);
1656 break;
1657 }
1658}
1659
1660int __get_dumpable(unsigned long mm_flags)
1661{
1662 int ret;
1663
1664 ret = mm_flags & MMF_DUMPABLE_MASK;
1665 return (ret > SUID_DUMP_USER) ? SUID_DUMP_ROOT : ret;
1666}
1667
1668int get_dumpable(struct mm_struct *mm)
1669{
1670 return __get_dumpable(mm->flags);
1671}
1672
1673SYSCALL_DEFINE3(execve,
1674 const char __user *, filename,
1675 const char __user *const __user *, argv,
1676 const char __user *const __user *, envp)
1677{
1678 struct filename *path = getname(filename);
1679 int error = PTR_ERR(path);
1680 if (!IS_ERR(path)) {
1681 error = do_execve(path->name, argv, envp);
1682 putname(path);
1683 }
1684 return error;
1685}
1686#ifdef CONFIG_COMPAT
1687asmlinkage long compat_sys_execve(const char __user * filename,
1688 const compat_uptr_t __user * argv,
1689 const compat_uptr_t __user * envp)
1690{
1691 struct filename *path = getname(filename);
1692 int error = PTR_ERR(path);
1693 if (!IS_ERR(path)) {
1694 error = compat_do_execve(path->name, argv, envp);
1695 putname(path);
1696 }
1697 return error;
1698}
1699#endif