Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/exec.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * #!-checking implemented by tytso.
10 */
11/*
12 * Demand-loading implemented 01.12.91 - no need to read anything but
13 * the header into memory. The inode of the executable is put into
14 * "current->executable", and page faults do the actual loading. Clean.
15 *
16 * Once more I can proudly say that linux stood up to being changed: it
17 * was less than 2 hours work to get demand-loading completely implemented.
18 *
19 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
20 * current->executable is only used by the procfs. This allows a dispatch
21 * table to check for several different types of binary formats. We keep
22 * trying until we recognize the file or we run out of supported binary
23 * formats.
24 */
25
26#include <linux/kernel_read_file.h>
27#include <linux/slab.h>
28#include <linux/file.h>
29#include <linux/fdtable.h>
30#include <linux/mm.h>
31#include <linux/stat.h>
32#include <linux/fcntl.h>
33#include <linux/swap.h>
34#include <linux/string.h>
35#include <linux/init.h>
36#include <linux/sched/mm.h>
37#include <linux/sched/coredump.h>
38#include <linux/sched/signal.h>
39#include <linux/sched/numa_balancing.h>
40#include <linux/sched/task.h>
41#include <linux/pagemap.h>
42#include <linux/perf_event.h>
43#include <linux/highmem.h>
44#include <linux/spinlock.h>
45#include <linux/key.h>
46#include <linux/personality.h>
47#include <linux/binfmts.h>
48#include <linux/utsname.h>
49#include <linux/pid_namespace.h>
50#include <linux/module.h>
51#include <linux/namei.h>
52#include <linux/mount.h>
53#include <linux/security.h>
54#include <linux/syscalls.h>
55#include <linux/tsacct_kern.h>
56#include <linux/cn_proc.h>
57#include <linux/audit.h>
58#include <linux/kmod.h>
59#include <linux/fsnotify.h>
60#include <linux/fs_struct.h>
61#include <linux/oom.h>
62#include <linux/compat.h>
63#include <linux/vmalloc.h>
64#include <linux/io_uring.h>
65#include <linux/syscall_user_dispatch.h>
66#include <linux/coredump.h>
67#include <linux/time_namespace.h>
68
69#include <linux/uaccess.h>
70#include <asm/mmu_context.h>
71#include <asm/tlb.h>
72
73#include <trace/events/task.h>
74#include "internal.h"
75
76#include <trace/events/sched.h>
77
78static int bprm_creds_from_file(struct linux_binprm *bprm);
79
80int suid_dumpable = 0;
81
82static LIST_HEAD(formats);
83static DEFINE_RWLOCK(binfmt_lock);
84
85void __register_binfmt(struct linux_binfmt * fmt, int insert)
86{
87 write_lock(&binfmt_lock);
88 insert ? list_add(&fmt->lh, &formats) :
89 list_add_tail(&fmt->lh, &formats);
90 write_unlock(&binfmt_lock);
91}
92
93EXPORT_SYMBOL(__register_binfmt);
94
95void unregister_binfmt(struct linux_binfmt * fmt)
96{
97 write_lock(&binfmt_lock);
98 list_del(&fmt->lh);
99 write_unlock(&binfmt_lock);
100}
101
102EXPORT_SYMBOL(unregister_binfmt);
103
104static inline void put_binfmt(struct linux_binfmt * fmt)
105{
106 module_put(fmt->module);
107}
108
109bool path_noexec(const struct path *path)
110{
111 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
112 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
113}
114
115#ifdef CONFIG_USELIB
116/*
117 * Note that a shared library must be both readable and executable due to
118 * security reasons.
119 *
120 * Also note that we take the address to load from the file itself.
121 */
122SYSCALL_DEFINE1(uselib, const char __user *, library)
123{
124 struct linux_binfmt *fmt;
125 struct file *file;
126 struct filename *tmp = getname(library);
127 int error = PTR_ERR(tmp);
128 static const struct open_flags uselib_flags = {
129 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
130 .acc_mode = MAY_READ | MAY_EXEC,
131 .intent = LOOKUP_OPEN,
132 .lookup_flags = LOOKUP_FOLLOW,
133 };
134
135 if (IS_ERR(tmp))
136 goto out;
137
138 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
139 putname(tmp);
140 error = PTR_ERR(file);
141 if (IS_ERR(file))
142 goto out;
143
144 /*
145 * may_open() has already checked for this, so it should be
146 * impossible to trip now. But we need to be extra cautious
147 * and check again at the very end too.
148 */
149 error = -EACCES;
150 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
151 path_noexec(&file->f_path)))
152 goto exit;
153
154 fsnotify_open(file);
155
156 error = -ENOEXEC;
157
158 read_lock(&binfmt_lock);
159 list_for_each_entry(fmt, &formats, lh) {
160 if (!fmt->load_shlib)
161 continue;
162 if (!try_module_get(fmt->module))
163 continue;
164 read_unlock(&binfmt_lock);
165 error = fmt->load_shlib(file);
166 read_lock(&binfmt_lock);
167 put_binfmt(fmt);
168 if (error != -ENOEXEC)
169 break;
170 }
171 read_unlock(&binfmt_lock);
172exit:
173 fput(file);
174out:
175 return error;
176}
177#endif /* #ifdef CONFIG_USELIB */
178
179#ifdef CONFIG_MMU
180/*
181 * The nascent bprm->mm is not visible until exec_mmap() but it can
182 * use a lot of memory, account these pages in current->mm temporary
183 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
184 * change the counter back via acct_arg_size(0).
185 */
186static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
187{
188 struct mm_struct *mm = current->mm;
189 long diff = (long)(pages - bprm->vma_pages);
190
191 if (!mm || !diff)
192 return;
193
194 bprm->vma_pages = pages;
195 add_mm_counter(mm, MM_ANONPAGES, diff);
196}
197
198static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
199 int write)
200{
201 struct page *page;
202 int ret;
203 unsigned int gup_flags = 0;
204
205#ifdef CONFIG_STACK_GROWSUP
206 if (write) {
207 ret = expand_downwards(bprm->vma, pos);
208 if (ret < 0)
209 return NULL;
210 }
211#endif
212
213 if (write)
214 gup_flags |= FOLL_WRITE;
215
216 /*
217 * We are doing an exec(). 'current' is the process
218 * doing the exec and bprm->mm is the new process's mm.
219 */
220 mmap_read_lock(bprm->mm);
221 ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
222 &page, NULL, NULL);
223 mmap_read_unlock(bprm->mm);
224 if (ret <= 0)
225 return NULL;
226
227 if (write)
228 acct_arg_size(bprm, vma_pages(bprm->vma));
229
230 return page;
231}
232
233static void put_arg_page(struct page *page)
234{
235 put_page(page);
236}
237
238static void free_arg_pages(struct linux_binprm *bprm)
239{
240}
241
242static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
243 struct page *page)
244{
245 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
246}
247
248static int __bprm_mm_init(struct linux_binprm *bprm)
249{
250 int err;
251 struct vm_area_struct *vma = NULL;
252 struct mm_struct *mm = bprm->mm;
253
254 bprm->vma = vma = vm_area_alloc(mm);
255 if (!vma)
256 return -ENOMEM;
257 vma_set_anonymous(vma);
258
259 if (mmap_write_lock_killable(mm)) {
260 err = -EINTR;
261 goto err_free;
262 }
263
264 /*
265 * Place the stack at the largest stack address the architecture
266 * supports. Later, we'll move this to an appropriate place. We don't
267 * use STACK_TOP because that can depend on attributes which aren't
268 * configured yet.
269 */
270 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
271 vma->vm_end = STACK_TOP_MAX;
272 vma->vm_start = vma->vm_end - PAGE_SIZE;
273 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
274 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
275
276 err = insert_vm_struct(mm, vma);
277 if (err)
278 goto err;
279
280 mm->stack_vm = mm->total_vm = 1;
281 mmap_write_unlock(mm);
282 bprm->p = vma->vm_end - sizeof(void *);
283 return 0;
284err:
285 mmap_write_unlock(mm);
286err_free:
287 bprm->vma = NULL;
288 vm_area_free(vma);
289 return err;
290}
291
292static bool valid_arg_len(struct linux_binprm *bprm, long len)
293{
294 return len <= MAX_ARG_STRLEN;
295}
296
297#else
298
299static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
300{
301}
302
303static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
304 int write)
305{
306 struct page *page;
307
308 page = bprm->page[pos / PAGE_SIZE];
309 if (!page && write) {
310 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
311 if (!page)
312 return NULL;
313 bprm->page[pos / PAGE_SIZE] = page;
314 }
315
316 return page;
317}
318
319static void put_arg_page(struct page *page)
320{
321}
322
323static void free_arg_page(struct linux_binprm *bprm, int i)
324{
325 if (bprm->page[i]) {
326 __free_page(bprm->page[i]);
327 bprm->page[i] = NULL;
328 }
329}
330
331static void free_arg_pages(struct linux_binprm *bprm)
332{
333 int i;
334
335 for (i = 0; i < MAX_ARG_PAGES; i++)
336 free_arg_page(bprm, i);
337}
338
339static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
340 struct page *page)
341{
342}
343
344static int __bprm_mm_init(struct linux_binprm *bprm)
345{
346 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
347 return 0;
348}
349
350static bool valid_arg_len(struct linux_binprm *bprm, long len)
351{
352 return len <= bprm->p;
353}
354
355#endif /* CONFIG_MMU */
356
357/*
358 * Create a new mm_struct and populate it with a temporary stack
359 * vm_area_struct. We don't have enough context at this point to set the stack
360 * flags, permissions, and offset, so we use temporary values. We'll update
361 * them later in setup_arg_pages().
362 */
363static int bprm_mm_init(struct linux_binprm *bprm)
364{
365 int err;
366 struct mm_struct *mm = NULL;
367
368 bprm->mm = mm = mm_alloc();
369 err = -ENOMEM;
370 if (!mm)
371 goto err;
372
373 /* Save current stack limit for all calculations made during exec. */
374 task_lock(current->group_leader);
375 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
376 task_unlock(current->group_leader);
377
378 err = __bprm_mm_init(bprm);
379 if (err)
380 goto err;
381
382 return 0;
383
384err:
385 if (mm) {
386 bprm->mm = NULL;
387 mmdrop(mm);
388 }
389
390 return err;
391}
392
393struct user_arg_ptr {
394#ifdef CONFIG_COMPAT
395 bool is_compat;
396#endif
397 union {
398 const char __user *const __user *native;
399#ifdef CONFIG_COMPAT
400 const compat_uptr_t __user *compat;
401#endif
402 } ptr;
403};
404
405static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
406{
407 const char __user *native;
408
409#ifdef CONFIG_COMPAT
410 if (unlikely(argv.is_compat)) {
411 compat_uptr_t compat;
412
413 if (get_user(compat, argv.ptr.compat + nr))
414 return ERR_PTR(-EFAULT);
415
416 return compat_ptr(compat);
417 }
418#endif
419
420 if (get_user(native, argv.ptr.native + nr))
421 return ERR_PTR(-EFAULT);
422
423 return native;
424}
425
426/*
427 * count() counts the number of strings in array ARGV.
428 */
429static int count(struct user_arg_ptr argv, int max)
430{
431 int i = 0;
432
433 if (argv.ptr.native != NULL) {
434 for (;;) {
435 const char __user *p = get_user_arg_ptr(argv, i);
436
437 if (!p)
438 break;
439
440 if (IS_ERR(p))
441 return -EFAULT;
442
443 if (i >= max)
444 return -E2BIG;
445 ++i;
446
447 if (fatal_signal_pending(current))
448 return -ERESTARTNOHAND;
449 cond_resched();
450 }
451 }
452 return i;
453}
454
455static int count_strings_kernel(const char *const *argv)
456{
457 int i;
458
459 if (!argv)
460 return 0;
461
462 for (i = 0; argv[i]; ++i) {
463 if (i >= MAX_ARG_STRINGS)
464 return -E2BIG;
465 if (fatal_signal_pending(current))
466 return -ERESTARTNOHAND;
467 cond_resched();
468 }
469 return i;
470}
471
472static int bprm_stack_limits(struct linux_binprm *bprm)
473{
474 unsigned long limit, ptr_size;
475
476 /*
477 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
478 * (whichever is smaller) for the argv+env strings.
479 * This ensures that:
480 * - the remaining binfmt code will not run out of stack space,
481 * - the program will have a reasonable amount of stack left
482 * to work from.
483 */
484 limit = _STK_LIM / 4 * 3;
485 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
486 /*
487 * We've historically supported up to 32 pages (ARG_MAX)
488 * of argument strings even with small stacks
489 */
490 limit = max_t(unsigned long, limit, ARG_MAX);
491 /*
492 * We must account for the size of all the argv and envp pointers to
493 * the argv and envp strings, since they will also take up space in
494 * the stack. They aren't stored until much later when we can't
495 * signal to the parent that the child has run out of stack space.
496 * Instead, calculate it here so it's possible to fail gracefully.
497 *
498 * In the case of argc = 0, make sure there is space for adding a
499 * empty string (which will bump argc to 1), to ensure confused
500 * userspace programs don't start processing from argv[1], thinking
501 * argc can never be 0, to keep them from walking envp by accident.
502 * See do_execveat_common().
503 */
504 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
505 if (limit <= ptr_size)
506 return -E2BIG;
507 limit -= ptr_size;
508
509 bprm->argmin = bprm->p - limit;
510 return 0;
511}
512
513/*
514 * 'copy_strings()' copies argument/environment strings from the old
515 * processes's memory to the new process's stack. The call to get_user_pages()
516 * ensures the destination page is created and not swapped out.
517 */
518static int copy_strings(int argc, struct user_arg_ptr argv,
519 struct linux_binprm *bprm)
520{
521 struct page *kmapped_page = NULL;
522 char *kaddr = NULL;
523 unsigned long kpos = 0;
524 int ret;
525
526 while (argc-- > 0) {
527 const char __user *str;
528 int len;
529 unsigned long pos;
530
531 ret = -EFAULT;
532 str = get_user_arg_ptr(argv, argc);
533 if (IS_ERR(str))
534 goto out;
535
536 len = strnlen_user(str, MAX_ARG_STRLEN);
537 if (!len)
538 goto out;
539
540 ret = -E2BIG;
541 if (!valid_arg_len(bprm, len))
542 goto out;
543
544 /* We're going to work our way backwards. */
545 pos = bprm->p;
546 str += len;
547 bprm->p -= len;
548#ifdef CONFIG_MMU
549 if (bprm->p < bprm->argmin)
550 goto out;
551#endif
552
553 while (len > 0) {
554 int offset, bytes_to_copy;
555
556 if (fatal_signal_pending(current)) {
557 ret = -ERESTARTNOHAND;
558 goto out;
559 }
560 cond_resched();
561
562 offset = pos % PAGE_SIZE;
563 if (offset == 0)
564 offset = PAGE_SIZE;
565
566 bytes_to_copy = offset;
567 if (bytes_to_copy > len)
568 bytes_to_copy = len;
569
570 offset -= bytes_to_copy;
571 pos -= bytes_to_copy;
572 str -= bytes_to_copy;
573 len -= bytes_to_copy;
574
575 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
576 struct page *page;
577
578 page = get_arg_page(bprm, pos, 1);
579 if (!page) {
580 ret = -E2BIG;
581 goto out;
582 }
583
584 if (kmapped_page) {
585 flush_dcache_page(kmapped_page);
586 kunmap_local(kaddr);
587 put_arg_page(kmapped_page);
588 }
589 kmapped_page = page;
590 kaddr = kmap_local_page(kmapped_page);
591 kpos = pos & PAGE_MASK;
592 flush_arg_page(bprm, kpos, kmapped_page);
593 }
594 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
595 ret = -EFAULT;
596 goto out;
597 }
598 }
599 }
600 ret = 0;
601out:
602 if (kmapped_page) {
603 flush_dcache_page(kmapped_page);
604 kunmap_local(kaddr);
605 put_arg_page(kmapped_page);
606 }
607 return ret;
608}
609
610/*
611 * Copy and argument/environment string from the kernel to the processes stack.
612 */
613int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
614{
615 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
616 unsigned long pos = bprm->p;
617
618 if (len == 0)
619 return -EFAULT;
620 if (!valid_arg_len(bprm, len))
621 return -E2BIG;
622
623 /* We're going to work our way backwards. */
624 arg += len;
625 bprm->p -= len;
626 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
627 return -E2BIG;
628
629 while (len > 0) {
630 unsigned int bytes_to_copy = min_t(unsigned int, len,
631 min_not_zero(offset_in_page(pos), PAGE_SIZE));
632 struct page *page;
633
634 pos -= bytes_to_copy;
635 arg -= bytes_to_copy;
636 len -= bytes_to_copy;
637
638 page = get_arg_page(bprm, pos, 1);
639 if (!page)
640 return -E2BIG;
641 flush_arg_page(bprm, pos & PAGE_MASK, page);
642 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
643 put_arg_page(page);
644 }
645
646 return 0;
647}
648EXPORT_SYMBOL(copy_string_kernel);
649
650static int copy_strings_kernel(int argc, const char *const *argv,
651 struct linux_binprm *bprm)
652{
653 while (argc-- > 0) {
654 int ret = copy_string_kernel(argv[argc], bprm);
655 if (ret < 0)
656 return ret;
657 if (fatal_signal_pending(current))
658 return -ERESTARTNOHAND;
659 cond_resched();
660 }
661 return 0;
662}
663
664#ifdef CONFIG_MMU
665
666/*
667 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
668 * the binfmt code determines where the new stack should reside, we shift it to
669 * its final location. The process proceeds as follows:
670 *
671 * 1) Use shift to calculate the new vma endpoints.
672 * 2) Extend vma to cover both the old and new ranges. This ensures the
673 * arguments passed to subsequent functions are consistent.
674 * 3) Move vma's page tables to the new range.
675 * 4) Free up any cleared pgd range.
676 * 5) Shrink the vma to cover only the new range.
677 */
678static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
679{
680 struct mm_struct *mm = vma->vm_mm;
681 unsigned long old_start = vma->vm_start;
682 unsigned long old_end = vma->vm_end;
683 unsigned long length = old_end - old_start;
684 unsigned long new_start = old_start - shift;
685 unsigned long new_end = old_end - shift;
686 VMA_ITERATOR(vmi, mm, new_start);
687 struct vm_area_struct *next;
688 struct mmu_gather tlb;
689
690 BUG_ON(new_start > new_end);
691
692 /*
693 * ensure there are no vmas between where we want to go
694 * and where we are
695 */
696 if (vma != vma_next(&vmi))
697 return -EFAULT;
698
699 /*
700 * cover the whole range: [new_start, old_end)
701 */
702 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
703 return -ENOMEM;
704
705 /*
706 * move the page tables downwards, on failure we rely on
707 * process cleanup to remove whatever mess we made.
708 */
709 if (length != move_page_tables(vma, old_start,
710 vma, new_start, length, false))
711 return -ENOMEM;
712
713 lru_add_drain();
714 tlb_gather_mmu(&tlb, mm);
715 next = vma_next(&vmi);
716 if (new_end > old_start) {
717 /*
718 * when the old and new regions overlap clear from new_end.
719 */
720 free_pgd_range(&tlb, new_end, old_end, new_end,
721 next ? next->vm_start : USER_PGTABLES_CEILING);
722 } else {
723 /*
724 * otherwise, clean from old_start; this is done to not touch
725 * the address space in [new_end, old_start) some architectures
726 * have constraints on va-space that make this illegal (IA64) -
727 * for the others its just a little faster.
728 */
729 free_pgd_range(&tlb, old_start, old_end, new_end,
730 next ? next->vm_start : USER_PGTABLES_CEILING);
731 }
732 tlb_finish_mmu(&tlb);
733
734 /*
735 * Shrink the vma to just the new range. Always succeeds.
736 */
737 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
738
739 return 0;
740}
741
742/*
743 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
744 * the stack is optionally relocated, and some extra space is added.
745 */
746int setup_arg_pages(struct linux_binprm *bprm,
747 unsigned long stack_top,
748 int executable_stack)
749{
750 unsigned long ret;
751 unsigned long stack_shift;
752 struct mm_struct *mm = current->mm;
753 struct vm_area_struct *vma = bprm->vma;
754 struct vm_area_struct *prev = NULL;
755 unsigned long vm_flags;
756 unsigned long stack_base;
757 unsigned long stack_size;
758 unsigned long stack_expand;
759 unsigned long rlim_stack;
760 struct mmu_gather tlb;
761
762#ifdef CONFIG_STACK_GROWSUP
763 /* Limit stack size */
764 stack_base = bprm->rlim_stack.rlim_max;
765
766 stack_base = calc_max_stack_size(stack_base);
767
768 /* Add space for stack randomization. */
769 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
770
771 /* Make sure we didn't let the argument array grow too large. */
772 if (vma->vm_end - vma->vm_start > stack_base)
773 return -ENOMEM;
774
775 stack_base = PAGE_ALIGN(stack_top - stack_base);
776
777 stack_shift = vma->vm_start - stack_base;
778 mm->arg_start = bprm->p - stack_shift;
779 bprm->p = vma->vm_end - stack_shift;
780#else
781 stack_top = arch_align_stack(stack_top);
782 stack_top = PAGE_ALIGN(stack_top);
783
784 if (unlikely(stack_top < mmap_min_addr) ||
785 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
786 return -ENOMEM;
787
788 stack_shift = vma->vm_end - stack_top;
789
790 bprm->p -= stack_shift;
791 mm->arg_start = bprm->p;
792#endif
793
794 if (bprm->loader)
795 bprm->loader -= stack_shift;
796 bprm->exec -= stack_shift;
797
798 if (mmap_write_lock_killable(mm))
799 return -EINTR;
800
801 vm_flags = VM_STACK_FLAGS;
802
803 /*
804 * Adjust stack execute permissions; explicitly enable for
805 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
806 * (arch default) otherwise.
807 */
808 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
809 vm_flags |= VM_EXEC;
810 else if (executable_stack == EXSTACK_DISABLE_X)
811 vm_flags &= ~VM_EXEC;
812 vm_flags |= mm->def_flags;
813 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
814
815 tlb_gather_mmu(&tlb, mm);
816 ret = mprotect_fixup(&tlb, vma, &prev, vma->vm_start, vma->vm_end,
817 vm_flags);
818 tlb_finish_mmu(&tlb);
819
820 if (ret)
821 goto out_unlock;
822 BUG_ON(prev != vma);
823
824 if (unlikely(vm_flags & VM_EXEC)) {
825 pr_warn_once("process '%pD4' started with executable stack\n",
826 bprm->file);
827 }
828
829 /* Move stack pages down in memory. */
830 if (stack_shift) {
831 ret = shift_arg_pages(vma, stack_shift);
832 if (ret)
833 goto out_unlock;
834 }
835
836 /* mprotect_fixup is overkill to remove the temporary stack flags */
837 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
838
839 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
840 stack_size = vma->vm_end - vma->vm_start;
841 /*
842 * Align this down to a page boundary as expand_stack
843 * will align it up.
844 */
845 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
846
847 stack_expand = min(rlim_stack, stack_size + stack_expand);
848
849#ifdef CONFIG_STACK_GROWSUP
850 stack_base = vma->vm_start + stack_expand;
851#else
852 stack_base = vma->vm_end - stack_expand;
853#endif
854 current->mm->start_stack = bprm->p;
855 ret = expand_stack(vma, stack_base);
856 if (ret)
857 ret = -EFAULT;
858
859out_unlock:
860 mmap_write_unlock(mm);
861 return ret;
862}
863EXPORT_SYMBOL(setup_arg_pages);
864
865#else
866
867/*
868 * Transfer the program arguments and environment from the holding pages
869 * onto the stack. The provided stack pointer is adjusted accordingly.
870 */
871int transfer_args_to_stack(struct linux_binprm *bprm,
872 unsigned long *sp_location)
873{
874 unsigned long index, stop, sp;
875 int ret = 0;
876
877 stop = bprm->p >> PAGE_SHIFT;
878 sp = *sp_location;
879
880 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
881 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
882 char *src = kmap_local_page(bprm->page[index]) + offset;
883 sp -= PAGE_SIZE - offset;
884 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
885 ret = -EFAULT;
886 kunmap_local(src);
887 if (ret)
888 goto out;
889 }
890
891 *sp_location = sp;
892
893out:
894 return ret;
895}
896EXPORT_SYMBOL(transfer_args_to_stack);
897
898#endif /* CONFIG_MMU */
899
900static struct file *do_open_execat(int fd, struct filename *name, int flags)
901{
902 struct file *file;
903 int err;
904 struct open_flags open_exec_flags = {
905 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
906 .acc_mode = MAY_EXEC,
907 .intent = LOOKUP_OPEN,
908 .lookup_flags = LOOKUP_FOLLOW,
909 };
910
911 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
912 return ERR_PTR(-EINVAL);
913 if (flags & AT_SYMLINK_NOFOLLOW)
914 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
915 if (flags & AT_EMPTY_PATH)
916 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
917
918 file = do_filp_open(fd, name, &open_exec_flags);
919 if (IS_ERR(file))
920 goto out;
921
922 /*
923 * may_open() has already checked for this, so it should be
924 * impossible to trip now. But we need to be extra cautious
925 * and check again at the very end too.
926 */
927 err = -EACCES;
928 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
929 path_noexec(&file->f_path)))
930 goto exit;
931
932 err = deny_write_access(file);
933 if (err)
934 goto exit;
935
936 if (name->name[0] != '\0')
937 fsnotify_open(file);
938
939out:
940 return file;
941
942exit:
943 fput(file);
944 return ERR_PTR(err);
945}
946
947struct file *open_exec(const char *name)
948{
949 struct filename *filename = getname_kernel(name);
950 struct file *f = ERR_CAST(filename);
951
952 if (!IS_ERR(filename)) {
953 f = do_open_execat(AT_FDCWD, filename, 0);
954 putname(filename);
955 }
956 return f;
957}
958EXPORT_SYMBOL(open_exec);
959
960#if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
961ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
962{
963 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
964 if (res > 0)
965 flush_icache_user_range(addr, addr + len);
966 return res;
967}
968EXPORT_SYMBOL(read_code);
969#endif
970
971/*
972 * Maps the mm_struct mm into the current task struct.
973 * On success, this function returns with exec_update_lock
974 * held for writing.
975 */
976static int exec_mmap(struct mm_struct *mm)
977{
978 struct task_struct *tsk;
979 struct mm_struct *old_mm, *active_mm;
980 int ret;
981
982 /* Notify parent that we're no longer interested in the old VM */
983 tsk = current;
984 old_mm = current->mm;
985 exec_mm_release(tsk, old_mm);
986 if (old_mm)
987 sync_mm_rss(old_mm);
988
989 ret = down_write_killable(&tsk->signal->exec_update_lock);
990 if (ret)
991 return ret;
992
993 if (old_mm) {
994 /*
995 * If there is a pending fatal signal perhaps a signal
996 * whose default action is to create a coredump get
997 * out and die instead of going through with the exec.
998 */
999 ret = mmap_read_lock_killable(old_mm);
1000 if (ret) {
1001 up_write(&tsk->signal->exec_update_lock);
1002 return ret;
1003 }
1004 }
1005
1006 task_lock(tsk);
1007 membarrier_exec_mmap(mm);
1008
1009 local_irq_disable();
1010 active_mm = tsk->active_mm;
1011 tsk->active_mm = mm;
1012 tsk->mm = mm;
1013 /*
1014 * This prevents preemption while active_mm is being loaded and
1015 * it and mm are being updated, which could cause problems for
1016 * lazy tlb mm refcounting when these are updated by context
1017 * switches. Not all architectures can handle irqs off over
1018 * activate_mm yet.
1019 */
1020 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1021 local_irq_enable();
1022 activate_mm(active_mm, mm);
1023 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1024 local_irq_enable();
1025 lru_gen_add_mm(mm);
1026 task_unlock(tsk);
1027 lru_gen_use_mm(mm);
1028 if (old_mm) {
1029 mmap_read_unlock(old_mm);
1030 BUG_ON(active_mm != old_mm);
1031 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1032 mm_update_next_owner(old_mm);
1033 mmput(old_mm);
1034 return 0;
1035 }
1036 mmdrop(active_mm);
1037 return 0;
1038}
1039
1040static int de_thread(struct task_struct *tsk)
1041{
1042 struct signal_struct *sig = tsk->signal;
1043 struct sighand_struct *oldsighand = tsk->sighand;
1044 spinlock_t *lock = &oldsighand->siglock;
1045
1046 if (thread_group_empty(tsk))
1047 goto no_thread_group;
1048
1049 /*
1050 * Kill all other threads in the thread group.
1051 */
1052 spin_lock_irq(lock);
1053 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1054 /*
1055 * Another group action in progress, just
1056 * return so that the signal is processed.
1057 */
1058 spin_unlock_irq(lock);
1059 return -EAGAIN;
1060 }
1061
1062 sig->group_exec_task = tsk;
1063 sig->notify_count = zap_other_threads(tsk);
1064 if (!thread_group_leader(tsk))
1065 sig->notify_count--;
1066
1067 while (sig->notify_count) {
1068 __set_current_state(TASK_KILLABLE);
1069 spin_unlock_irq(lock);
1070 schedule();
1071 if (__fatal_signal_pending(tsk))
1072 goto killed;
1073 spin_lock_irq(lock);
1074 }
1075 spin_unlock_irq(lock);
1076
1077 /*
1078 * At this point all other threads have exited, all we have to
1079 * do is to wait for the thread group leader to become inactive,
1080 * and to assume its PID:
1081 */
1082 if (!thread_group_leader(tsk)) {
1083 struct task_struct *leader = tsk->group_leader;
1084
1085 for (;;) {
1086 cgroup_threadgroup_change_begin(tsk);
1087 write_lock_irq(&tasklist_lock);
1088 /*
1089 * Do this under tasklist_lock to ensure that
1090 * exit_notify() can't miss ->group_exec_task
1091 */
1092 sig->notify_count = -1;
1093 if (likely(leader->exit_state))
1094 break;
1095 __set_current_state(TASK_KILLABLE);
1096 write_unlock_irq(&tasklist_lock);
1097 cgroup_threadgroup_change_end(tsk);
1098 schedule();
1099 if (__fatal_signal_pending(tsk))
1100 goto killed;
1101 }
1102
1103 /*
1104 * The only record we have of the real-time age of a
1105 * process, regardless of execs it's done, is start_time.
1106 * All the past CPU time is accumulated in signal_struct
1107 * from sister threads now dead. But in this non-leader
1108 * exec, nothing survives from the original leader thread,
1109 * whose birth marks the true age of this process now.
1110 * When we take on its identity by switching to its PID, we
1111 * also take its birthdate (always earlier than our own).
1112 */
1113 tsk->start_time = leader->start_time;
1114 tsk->start_boottime = leader->start_boottime;
1115
1116 BUG_ON(!same_thread_group(leader, tsk));
1117 /*
1118 * An exec() starts a new thread group with the
1119 * TGID of the previous thread group. Rehash the
1120 * two threads with a switched PID, and release
1121 * the former thread group leader:
1122 */
1123
1124 /* Become a process group leader with the old leader's pid.
1125 * The old leader becomes a thread of the this thread group.
1126 */
1127 exchange_tids(tsk, leader);
1128 transfer_pid(leader, tsk, PIDTYPE_TGID);
1129 transfer_pid(leader, tsk, PIDTYPE_PGID);
1130 transfer_pid(leader, tsk, PIDTYPE_SID);
1131
1132 list_replace_rcu(&leader->tasks, &tsk->tasks);
1133 list_replace_init(&leader->sibling, &tsk->sibling);
1134
1135 tsk->group_leader = tsk;
1136 leader->group_leader = tsk;
1137
1138 tsk->exit_signal = SIGCHLD;
1139 leader->exit_signal = -1;
1140
1141 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1142 leader->exit_state = EXIT_DEAD;
1143
1144 /*
1145 * We are going to release_task()->ptrace_unlink() silently,
1146 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1147 * the tracer won't block again waiting for this thread.
1148 */
1149 if (unlikely(leader->ptrace))
1150 __wake_up_parent(leader, leader->parent);
1151 write_unlock_irq(&tasklist_lock);
1152 cgroup_threadgroup_change_end(tsk);
1153
1154 release_task(leader);
1155 }
1156
1157 sig->group_exec_task = NULL;
1158 sig->notify_count = 0;
1159
1160no_thread_group:
1161 /* we have changed execution domain */
1162 tsk->exit_signal = SIGCHLD;
1163
1164 BUG_ON(!thread_group_leader(tsk));
1165 return 0;
1166
1167killed:
1168 /* protects against exit_notify() and __exit_signal() */
1169 read_lock(&tasklist_lock);
1170 sig->group_exec_task = NULL;
1171 sig->notify_count = 0;
1172 read_unlock(&tasklist_lock);
1173 return -EAGAIN;
1174}
1175
1176
1177/*
1178 * This function makes sure the current process has its own signal table,
1179 * so that flush_signal_handlers can later reset the handlers without
1180 * disturbing other processes. (Other processes might share the signal
1181 * table via the CLONE_SIGHAND option to clone().)
1182 */
1183static int unshare_sighand(struct task_struct *me)
1184{
1185 struct sighand_struct *oldsighand = me->sighand;
1186
1187 if (refcount_read(&oldsighand->count) != 1) {
1188 struct sighand_struct *newsighand;
1189 /*
1190 * This ->sighand is shared with the CLONE_SIGHAND
1191 * but not CLONE_THREAD task, switch to the new one.
1192 */
1193 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1194 if (!newsighand)
1195 return -ENOMEM;
1196
1197 refcount_set(&newsighand->count, 1);
1198
1199 write_lock_irq(&tasklist_lock);
1200 spin_lock(&oldsighand->siglock);
1201 memcpy(newsighand->action, oldsighand->action,
1202 sizeof(newsighand->action));
1203 rcu_assign_pointer(me->sighand, newsighand);
1204 spin_unlock(&oldsighand->siglock);
1205 write_unlock_irq(&tasklist_lock);
1206
1207 __cleanup_sighand(oldsighand);
1208 }
1209 return 0;
1210}
1211
1212char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1213{
1214 task_lock(tsk);
1215 /* Always NUL terminated and zero-padded */
1216 strscpy_pad(buf, tsk->comm, buf_size);
1217 task_unlock(tsk);
1218 return buf;
1219}
1220EXPORT_SYMBOL_GPL(__get_task_comm);
1221
1222/*
1223 * These functions flushes out all traces of the currently running executable
1224 * so that a new one can be started
1225 */
1226
1227void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1228{
1229 task_lock(tsk);
1230 trace_task_rename(tsk, buf);
1231 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1232 task_unlock(tsk);
1233 perf_event_comm(tsk, exec);
1234}
1235
1236/*
1237 * Calling this is the point of no return. None of the failures will be
1238 * seen by userspace since either the process is already taking a fatal
1239 * signal (via de_thread() or coredump), or will have SEGV raised
1240 * (after exec_mmap()) by search_binary_handler (see below).
1241 */
1242int begin_new_exec(struct linux_binprm * bprm)
1243{
1244 struct task_struct *me = current;
1245 int retval;
1246
1247 /* Once we are committed compute the creds */
1248 retval = bprm_creds_from_file(bprm);
1249 if (retval)
1250 return retval;
1251
1252 /*
1253 * Ensure all future errors are fatal.
1254 */
1255 bprm->point_of_no_return = true;
1256
1257 /*
1258 * Make this the only thread in the thread group.
1259 */
1260 retval = de_thread(me);
1261 if (retval)
1262 goto out;
1263
1264 /*
1265 * Cancel any io_uring activity across execve
1266 */
1267 io_uring_task_cancel();
1268
1269 /* Ensure the files table is not shared. */
1270 retval = unshare_files();
1271 if (retval)
1272 goto out;
1273
1274 /*
1275 * Must be called _before_ exec_mmap() as bprm->mm is
1276 * not visible until then. This also enables the update
1277 * to be lockless.
1278 */
1279 retval = set_mm_exe_file(bprm->mm, bprm->file);
1280 if (retval)
1281 goto out;
1282
1283 /* If the binary is not readable then enforce mm->dumpable=0 */
1284 would_dump(bprm, bprm->file);
1285 if (bprm->have_execfd)
1286 would_dump(bprm, bprm->executable);
1287
1288 /*
1289 * Release all of the old mmap stuff
1290 */
1291 acct_arg_size(bprm, 0);
1292 retval = exec_mmap(bprm->mm);
1293 if (retval)
1294 goto out;
1295
1296 bprm->mm = NULL;
1297
1298 retval = exec_task_namespaces();
1299 if (retval)
1300 goto out_unlock;
1301
1302#ifdef CONFIG_POSIX_TIMERS
1303 spin_lock_irq(&me->sighand->siglock);
1304 posix_cpu_timers_exit(me);
1305 spin_unlock_irq(&me->sighand->siglock);
1306 exit_itimers(me);
1307 flush_itimer_signals();
1308#endif
1309
1310 /*
1311 * Make the signal table private.
1312 */
1313 retval = unshare_sighand(me);
1314 if (retval)
1315 goto out_unlock;
1316
1317 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1318 PF_NOFREEZE | PF_NO_SETAFFINITY);
1319 flush_thread();
1320 me->personality &= ~bprm->per_clear;
1321
1322 clear_syscall_work_syscall_user_dispatch(me);
1323
1324 /*
1325 * We have to apply CLOEXEC before we change whether the process is
1326 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1327 * trying to access the should-be-closed file descriptors of a process
1328 * undergoing exec(2).
1329 */
1330 do_close_on_exec(me->files);
1331
1332 if (bprm->secureexec) {
1333 /* Make sure parent cannot signal privileged process. */
1334 me->pdeath_signal = 0;
1335
1336 /*
1337 * For secureexec, reset the stack limit to sane default to
1338 * avoid bad behavior from the prior rlimits. This has to
1339 * happen before arch_pick_mmap_layout(), which examines
1340 * RLIMIT_STACK, but after the point of no return to avoid
1341 * needing to clean up the change on failure.
1342 */
1343 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1344 bprm->rlim_stack.rlim_cur = _STK_LIM;
1345 }
1346
1347 me->sas_ss_sp = me->sas_ss_size = 0;
1348
1349 /*
1350 * Figure out dumpability. Note that this checking only of current
1351 * is wrong, but userspace depends on it. This should be testing
1352 * bprm->secureexec instead.
1353 */
1354 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1355 !(uid_eq(current_euid(), current_uid()) &&
1356 gid_eq(current_egid(), current_gid())))
1357 set_dumpable(current->mm, suid_dumpable);
1358 else
1359 set_dumpable(current->mm, SUID_DUMP_USER);
1360
1361 perf_event_exec();
1362 __set_task_comm(me, kbasename(bprm->filename), true);
1363
1364 /* An exec changes our domain. We are no longer part of the thread
1365 group */
1366 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1367 flush_signal_handlers(me, 0);
1368
1369 retval = set_cred_ucounts(bprm->cred);
1370 if (retval < 0)
1371 goto out_unlock;
1372
1373 /*
1374 * install the new credentials for this executable
1375 */
1376 security_bprm_committing_creds(bprm);
1377
1378 commit_creds(bprm->cred);
1379 bprm->cred = NULL;
1380
1381 /*
1382 * Disable monitoring for regular users
1383 * when executing setuid binaries. Must
1384 * wait until new credentials are committed
1385 * by commit_creds() above
1386 */
1387 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1388 perf_event_exit_task(me);
1389 /*
1390 * cred_guard_mutex must be held at least to this point to prevent
1391 * ptrace_attach() from altering our determination of the task's
1392 * credentials; any time after this it may be unlocked.
1393 */
1394 security_bprm_committed_creds(bprm);
1395
1396 /* Pass the opened binary to the interpreter. */
1397 if (bprm->have_execfd) {
1398 retval = get_unused_fd_flags(0);
1399 if (retval < 0)
1400 goto out_unlock;
1401 fd_install(retval, bprm->executable);
1402 bprm->executable = NULL;
1403 bprm->execfd = retval;
1404 }
1405 return 0;
1406
1407out_unlock:
1408 up_write(&me->signal->exec_update_lock);
1409out:
1410 return retval;
1411}
1412EXPORT_SYMBOL(begin_new_exec);
1413
1414void would_dump(struct linux_binprm *bprm, struct file *file)
1415{
1416 struct inode *inode = file_inode(file);
1417 struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1418 if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1419 struct user_namespace *old, *user_ns;
1420 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1421
1422 /* Ensure mm->user_ns contains the executable */
1423 user_ns = old = bprm->mm->user_ns;
1424 while ((user_ns != &init_user_ns) &&
1425 !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1426 user_ns = user_ns->parent;
1427
1428 if (old != user_ns) {
1429 bprm->mm->user_ns = get_user_ns(user_ns);
1430 put_user_ns(old);
1431 }
1432 }
1433}
1434EXPORT_SYMBOL(would_dump);
1435
1436void setup_new_exec(struct linux_binprm * bprm)
1437{
1438 /* Setup things that can depend upon the personality */
1439 struct task_struct *me = current;
1440
1441 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1442
1443 arch_setup_new_exec();
1444
1445 /* Set the new mm task size. We have to do that late because it may
1446 * depend on TIF_32BIT which is only updated in flush_thread() on
1447 * some architectures like powerpc
1448 */
1449 me->mm->task_size = TASK_SIZE;
1450 up_write(&me->signal->exec_update_lock);
1451 mutex_unlock(&me->signal->cred_guard_mutex);
1452}
1453EXPORT_SYMBOL(setup_new_exec);
1454
1455/* Runs immediately before start_thread() takes over. */
1456void finalize_exec(struct linux_binprm *bprm)
1457{
1458 /* Store any stack rlimit changes before starting thread. */
1459 task_lock(current->group_leader);
1460 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1461 task_unlock(current->group_leader);
1462}
1463EXPORT_SYMBOL(finalize_exec);
1464
1465/*
1466 * Prepare credentials and lock ->cred_guard_mutex.
1467 * setup_new_exec() commits the new creds and drops the lock.
1468 * Or, if exec fails before, free_bprm() should release ->cred
1469 * and unlock.
1470 */
1471static int prepare_bprm_creds(struct linux_binprm *bprm)
1472{
1473 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1474 return -ERESTARTNOINTR;
1475
1476 bprm->cred = prepare_exec_creds();
1477 if (likely(bprm->cred))
1478 return 0;
1479
1480 mutex_unlock(¤t->signal->cred_guard_mutex);
1481 return -ENOMEM;
1482}
1483
1484static void free_bprm(struct linux_binprm *bprm)
1485{
1486 if (bprm->mm) {
1487 acct_arg_size(bprm, 0);
1488 mmput(bprm->mm);
1489 }
1490 free_arg_pages(bprm);
1491 if (bprm->cred) {
1492 mutex_unlock(¤t->signal->cred_guard_mutex);
1493 abort_creds(bprm->cred);
1494 }
1495 if (bprm->file) {
1496 allow_write_access(bprm->file);
1497 fput(bprm->file);
1498 }
1499 if (bprm->executable)
1500 fput(bprm->executable);
1501 /* If a binfmt changed the interp, free it. */
1502 if (bprm->interp != bprm->filename)
1503 kfree(bprm->interp);
1504 kfree(bprm->fdpath);
1505 kfree(bprm);
1506}
1507
1508static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1509{
1510 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1511 int retval = -ENOMEM;
1512 if (!bprm)
1513 goto out;
1514
1515 if (fd == AT_FDCWD || filename->name[0] == '/') {
1516 bprm->filename = filename->name;
1517 } else {
1518 if (filename->name[0] == '\0')
1519 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1520 else
1521 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1522 fd, filename->name);
1523 if (!bprm->fdpath)
1524 goto out_free;
1525
1526 bprm->filename = bprm->fdpath;
1527 }
1528 bprm->interp = bprm->filename;
1529
1530 retval = bprm_mm_init(bprm);
1531 if (retval)
1532 goto out_free;
1533 return bprm;
1534
1535out_free:
1536 free_bprm(bprm);
1537out:
1538 return ERR_PTR(retval);
1539}
1540
1541int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1542{
1543 /* If a binfmt changed the interp, free it first. */
1544 if (bprm->interp != bprm->filename)
1545 kfree(bprm->interp);
1546 bprm->interp = kstrdup(interp, GFP_KERNEL);
1547 if (!bprm->interp)
1548 return -ENOMEM;
1549 return 0;
1550}
1551EXPORT_SYMBOL(bprm_change_interp);
1552
1553/*
1554 * determine how safe it is to execute the proposed program
1555 * - the caller must hold ->cred_guard_mutex to protect against
1556 * PTRACE_ATTACH or seccomp thread-sync
1557 */
1558static void check_unsafe_exec(struct linux_binprm *bprm)
1559{
1560 struct task_struct *p = current, *t;
1561 unsigned n_fs;
1562
1563 if (p->ptrace)
1564 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1565
1566 /*
1567 * This isn't strictly necessary, but it makes it harder for LSMs to
1568 * mess up.
1569 */
1570 if (task_no_new_privs(current))
1571 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1572
1573 /*
1574 * If another task is sharing our fs, we cannot safely
1575 * suid exec because the differently privileged task
1576 * will be able to manipulate the current directory, etc.
1577 * It would be nice to force an unshare instead...
1578 */
1579 t = p;
1580 n_fs = 1;
1581 spin_lock(&p->fs->lock);
1582 rcu_read_lock();
1583 while_each_thread(p, t) {
1584 if (t->fs == p->fs)
1585 n_fs++;
1586 }
1587 rcu_read_unlock();
1588
1589 if (p->fs->users > n_fs)
1590 bprm->unsafe |= LSM_UNSAFE_SHARE;
1591 else
1592 p->fs->in_exec = 1;
1593 spin_unlock(&p->fs->lock);
1594}
1595
1596static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1597{
1598 /* Handle suid and sgid on files */
1599 struct user_namespace *mnt_userns;
1600 struct inode *inode = file_inode(file);
1601 unsigned int mode;
1602 vfsuid_t vfsuid;
1603 vfsgid_t vfsgid;
1604
1605 if (!mnt_may_suid(file->f_path.mnt))
1606 return;
1607
1608 if (task_no_new_privs(current))
1609 return;
1610
1611 mode = READ_ONCE(inode->i_mode);
1612 if (!(mode & (S_ISUID|S_ISGID)))
1613 return;
1614
1615 mnt_userns = file_mnt_user_ns(file);
1616
1617 /* Be careful if suid/sgid is set */
1618 inode_lock(inode);
1619
1620 /* reload atomically mode/uid/gid now that lock held */
1621 mode = inode->i_mode;
1622 vfsuid = i_uid_into_vfsuid(mnt_userns, inode);
1623 vfsgid = i_gid_into_vfsgid(mnt_userns, inode);
1624 inode_unlock(inode);
1625
1626 /* We ignore suid/sgid if there are no mappings for them in the ns */
1627 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1628 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1629 return;
1630
1631 if (mode & S_ISUID) {
1632 bprm->per_clear |= PER_CLEAR_ON_SETID;
1633 bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1634 }
1635
1636 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1637 bprm->per_clear |= PER_CLEAR_ON_SETID;
1638 bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1639 }
1640}
1641
1642/*
1643 * Compute brpm->cred based upon the final binary.
1644 */
1645static int bprm_creds_from_file(struct linux_binprm *bprm)
1646{
1647 /* Compute creds based on which file? */
1648 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1649
1650 bprm_fill_uid(bprm, file);
1651 return security_bprm_creds_from_file(bprm, file);
1652}
1653
1654/*
1655 * Fill the binprm structure from the inode.
1656 * Read the first BINPRM_BUF_SIZE bytes
1657 *
1658 * This may be called multiple times for binary chains (scripts for example).
1659 */
1660static int prepare_binprm(struct linux_binprm *bprm)
1661{
1662 loff_t pos = 0;
1663
1664 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1665 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1666}
1667
1668/*
1669 * Arguments are '\0' separated strings found at the location bprm->p
1670 * points to; chop off the first by relocating brpm->p to right after
1671 * the first '\0' encountered.
1672 */
1673int remove_arg_zero(struct linux_binprm *bprm)
1674{
1675 int ret = 0;
1676 unsigned long offset;
1677 char *kaddr;
1678 struct page *page;
1679
1680 if (!bprm->argc)
1681 return 0;
1682
1683 do {
1684 offset = bprm->p & ~PAGE_MASK;
1685 page = get_arg_page(bprm, bprm->p, 0);
1686 if (!page) {
1687 ret = -EFAULT;
1688 goto out;
1689 }
1690 kaddr = kmap_local_page(page);
1691
1692 for (; offset < PAGE_SIZE && kaddr[offset];
1693 offset++, bprm->p++)
1694 ;
1695
1696 kunmap_local(kaddr);
1697 put_arg_page(page);
1698 } while (offset == PAGE_SIZE);
1699
1700 bprm->p++;
1701 bprm->argc--;
1702 ret = 0;
1703
1704out:
1705 return ret;
1706}
1707EXPORT_SYMBOL(remove_arg_zero);
1708
1709#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1710/*
1711 * cycle the list of binary formats handler, until one recognizes the image
1712 */
1713static int search_binary_handler(struct linux_binprm *bprm)
1714{
1715 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1716 struct linux_binfmt *fmt;
1717 int retval;
1718
1719 retval = prepare_binprm(bprm);
1720 if (retval < 0)
1721 return retval;
1722
1723 retval = security_bprm_check(bprm);
1724 if (retval)
1725 return retval;
1726
1727 retval = -ENOENT;
1728 retry:
1729 read_lock(&binfmt_lock);
1730 list_for_each_entry(fmt, &formats, lh) {
1731 if (!try_module_get(fmt->module))
1732 continue;
1733 read_unlock(&binfmt_lock);
1734
1735 retval = fmt->load_binary(bprm);
1736
1737 read_lock(&binfmt_lock);
1738 put_binfmt(fmt);
1739 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1740 read_unlock(&binfmt_lock);
1741 return retval;
1742 }
1743 }
1744 read_unlock(&binfmt_lock);
1745
1746 if (need_retry) {
1747 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1748 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1749 return retval;
1750 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1751 return retval;
1752 need_retry = false;
1753 goto retry;
1754 }
1755
1756 return retval;
1757}
1758
1759/* binfmt handlers will call back into begin_new_exec() on success. */
1760static int exec_binprm(struct linux_binprm *bprm)
1761{
1762 pid_t old_pid, old_vpid;
1763 int ret, depth;
1764
1765 /* Need to fetch pid before load_binary changes it */
1766 old_pid = current->pid;
1767 rcu_read_lock();
1768 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1769 rcu_read_unlock();
1770
1771 /* This allows 4 levels of binfmt rewrites before failing hard. */
1772 for (depth = 0;; depth++) {
1773 struct file *exec;
1774 if (depth > 5)
1775 return -ELOOP;
1776
1777 ret = search_binary_handler(bprm);
1778 if (ret < 0)
1779 return ret;
1780 if (!bprm->interpreter)
1781 break;
1782
1783 exec = bprm->file;
1784 bprm->file = bprm->interpreter;
1785 bprm->interpreter = NULL;
1786
1787 allow_write_access(exec);
1788 if (unlikely(bprm->have_execfd)) {
1789 if (bprm->executable) {
1790 fput(exec);
1791 return -ENOEXEC;
1792 }
1793 bprm->executable = exec;
1794 } else
1795 fput(exec);
1796 }
1797
1798 audit_bprm(bprm);
1799 trace_sched_process_exec(current, old_pid, bprm);
1800 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1801 proc_exec_connector(current);
1802 return 0;
1803}
1804
1805/*
1806 * sys_execve() executes a new program.
1807 */
1808static int bprm_execve(struct linux_binprm *bprm,
1809 int fd, struct filename *filename, int flags)
1810{
1811 struct file *file;
1812 int retval;
1813
1814 retval = prepare_bprm_creds(bprm);
1815 if (retval)
1816 return retval;
1817
1818 /*
1819 * Check for unsafe execution states before exec_binprm(), which
1820 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1821 * where setuid-ness is evaluated.
1822 */
1823 check_unsafe_exec(bprm);
1824 current->in_execve = 1;
1825
1826 file = do_open_execat(fd, filename, flags);
1827 retval = PTR_ERR(file);
1828 if (IS_ERR(file))
1829 goto out_unmark;
1830
1831 sched_exec();
1832
1833 bprm->file = file;
1834 /*
1835 * Record that a name derived from an O_CLOEXEC fd will be
1836 * inaccessible after exec. This allows the code in exec to
1837 * choose to fail when the executable is not mmaped into the
1838 * interpreter and an open file descriptor is not passed to
1839 * the interpreter. This makes for a better user experience
1840 * than having the interpreter start and then immediately fail
1841 * when it finds the executable is inaccessible.
1842 */
1843 if (bprm->fdpath && get_close_on_exec(fd))
1844 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1845
1846 /* Set the unchanging part of bprm->cred */
1847 retval = security_bprm_creds_for_exec(bprm);
1848 if (retval)
1849 goto out;
1850
1851 retval = exec_binprm(bprm);
1852 if (retval < 0)
1853 goto out;
1854
1855 /* execve succeeded */
1856 current->fs->in_exec = 0;
1857 current->in_execve = 0;
1858 rseq_execve(current);
1859 acct_update_integrals(current);
1860 task_numa_free(current, false);
1861 return retval;
1862
1863out:
1864 /*
1865 * If past the point of no return ensure the code never
1866 * returns to the userspace process. Use an existing fatal
1867 * signal if present otherwise terminate the process with
1868 * SIGSEGV.
1869 */
1870 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1871 force_fatal_sig(SIGSEGV);
1872
1873out_unmark:
1874 current->fs->in_exec = 0;
1875 current->in_execve = 0;
1876
1877 return retval;
1878}
1879
1880static int do_execveat_common(int fd, struct filename *filename,
1881 struct user_arg_ptr argv,
1882 struct user_arg_ptr envp,
1883 int flags)
1884{
1885 struct linux_binprm *bprm;
1886 int retval;
1887
1888 if (IS_ERR(filename))
1889 return PTR_ERR(filename);
1890
1891 /*
1892 * We move the actual failure in case of RLIMIT_NPROC excess from
1893 * set*uid() to execve() because too many poorly written programs
1894 * don't check setuid() return code. Here we additionally recheck
1895 * whether NPROC limit is still exceeded.
1896 */
1897 if ((current->flags & PF_NPROC_EXCEEDED) &&
1898 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1899 retval = -EAGAIN;
1900 goto out_ret;
1901 }
1902
1903 /* We're below the limit (still or again), so we don't want to make
1904 * further execve() calls fail. */
1905 current->flags &= ~PF_NPROC_EXCEEDED;
1906
1907 bprm = alloc_bprm(fd, filename);
1908 if (IS_ERR(bprm)) {
1909 retval = PTR_ERR(bprm);
1910 goto out_ret;
1911 }
1912
1913 retval = count(argv, MAX_ARG_STRINGS);
1914 if (retval == 0)
1915 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1916 current->comm, bprm->filename);
1917 if (retval < 0)
1918 goto out_free;
1919 bprm->argc = retval;
1920
1921 retval = count(envp, MAX_ARG_STRINGS);
1922 if (retval < 0)
1923 goto out_free;
1924 bprm->envc = retval;
1925
1926 retval = bprm_stack_limits(bprm);
1927 if (retval < 0)
1928 goto out_free;
1929
1930 retval = copy_string_kernel(bprm->filename, bprm);
1931 if (retval < 0)
1932 goto out_free;
1933 bprm->exec = bprm->p;
1934
1935 retval = copy_strings(bprm->envc, envp, bprm);
1936 if (retval < 0)
1937 goto out_free;
1938
1939 retval = copy_strings(bprm->argc, argv, bprm);
1940 if (retval < 0)
1941 goto out_free;
1942
1943 /*
1944 * When argv is empty, add an empty string ("") as argv[0] to
1945 * ensure confused userspace programs that start processing
1946 * from argv[1] won't end up walking envp. See also
1947 * bprm_stack_limits().
1948 */
1949 if (bprm->argc == 0) {
1950 retval = copy_string_kernel("", bprm);
1951 if (retval < 0)
1952 goto out_free;
1953 bprm->argc = 1;
1954 }
1955
1956 retval = bprm_execve(bprm, fd, filename, flags);
1957out_free:
1958 free_bprm(bprm);
1959
1960out_ret:
1961 putname(filename);
1962 return retval;
1963}
1964
1965int kernel_execve(const char *kernel_filename,
1966 const char *const *argv, const char *const *envp)
1967{
1968 struct filename *filename;
1969 struct linux_binprm *bprm;
1970 int fd = AT_FDCWD;
1971 int retval;
1972
1973 /* It is non-sense for kernel threads to call execve */
1974 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1975 return -EINVAL;
1976
1977 filename = getname_kernel(kernel_filename);
1978 if (IS_ERR(filename))
1979 return PTR_ERR(filename);
1980
1981 bprm = alloc_bprm(fd, filename);
1982 if (IS_ERR(bprm)) {
1983 retval = PTR_ERR(bprm);
1984 goto out_ret;
1985 }
1986
1987 retval = count_strings_kernel(argv);
1988 if (WARN_ON_ONCE(retval == 0))
1989 retval = -EINVAL;
1990 if (retval < 0)
1991 goto out_free;
1992 bprm->argc = retval;
1993
1994 retval = count_strings_kernel(envp);
1995 if (retval < 0)
1996 goto out_free;
1997 bprm->envc = retval;
1998
1999 retval = bprm_stack_limits(bprm);
2000 if (retval < 0)
2001 goto out_free;
2002
2003 retval = copy_string_kernel(bprm->filename, bprm);
2004 if (retval < 0)
2005 goto out_free;
2006 bprm->exec = bprm->p;
2007
2008 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2009 if (retval < 0)
2010 goto out_free;
2011
2012 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2013 if (retval < 0)
2014 goto out_free;
2015
2016 retval = bprm_execve(bprm, fd, filename, 0);
2017out_free:
2018 free_bprm(bprm);
2019out_ret:
2020 putname(filename);
2021 return retval;
2022}
2023
2024static int do_execve(struct filename *filename,
2025 const char __user *const __user *__argv,
2026 const char __user *const __user *__envp)
2027{
2028 struct user_arg_ptr argv = { .ptr.native = __argv };
2029 struct user_arg_ptr envp = { .ptr.native = __envp };
2030 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2031}
2032
2033static int do_execveat(int fd, struct filename *filename,
2034 const char __user *const __user *__argv,
2035 const char __user *const __user *__envp,
2036 int flags)
2037{
2038 struct user_arg_ptr argv = { .ptr.native = __argv };
2039 struct user_arg_ptr envp = { .ptr.native = __envp };
2040
2041 return do_execveat_common(fd, filename, argv, envp, flags);
2042}
2043
2044#ifdef CONFIG_COMPAT
2045static int compat_do_execve(struct filename *filename,
2046 const compat_uptr_t __user *__argv,
2047 const compat_uptr_t __user *__envp)
2048{
2049 struct user_arg_ptr argv = {
2050 .is_compat = true,
2051 .ptr.compat = __argv,
2052 };
2053 struct user_arg_ptr envp = {
2054 .is_compat = true,
2055 .ptr.compat = __envp,
2056 };
2057 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2058}
2059
2060static int compat_do_execveat(int fd, struct filename *filename,
2061 const compat_uptr_t __user *__argv,
2062 const compat_uptr_t __user *__envp,
2063 int flags)
2064{
2065 struct user_arg_ptr argv = {
2066 .is_compat = true,
2067 .ptr.compat = __argv,
2068 };
2069 struct user_arg_ptr envp = {
2070 .is_compat = true,
2071 .ptr.compat = __envp,
2072 };
2073 return do_execveat_common(fd, filename, argv, envp, flags);
2074}
2075#endif
2076
2077void set_binfmt(struct linux_binfmt *new)
2078{
2079 struct mm_struct *mm = current->mm;
2080
2081 if (mm->binfmt)
2082 module_put(mm->binfmt->module);
2083
2084 mm->binfmt = new;
2085 if (new)
2086 __module_get(new->module);
2087}
2088EXPORT_SYMBOL(set_binfmt);
2089
2090/*
2091 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2092 */
2093void set_dumpable(struct mm_struct *mm, int value)
2094{
2095 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2096 return;
2097
2098 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2099}
2100
2101SYSCALL_DEFINE3(execve,
2102 const char __user *, filename,
2103 const char __user *const __user *, argv,
2104 const char __user *const __user *, envp)
2105{
2106 return do_execve(getname(filename), argv, envp);
2107}
2108
2109SYSCALL_DEFINE5(execveat,
2110 int, fd, const char __user *, filename,
2111 const char __user *const __user *, argv,
2112 const char __user *const __user *, envp,
2113 int, flags)
2114{
2115 return do_execveat(fd,
2116 getname_uflags(filename, flags),
2117 argv, envp, flags);
2118}
2119
2120#ifdef CONFIG_COMPAT
2121COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2122 const compat_uptr_t __user *, argv,
2123 const compat_uptr_t __user *, envp)
2124{
2125 return compat_do_execve(getname(filename), argv, envp);
2126}
2127
2128COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2129 const char __user *, filename,
2130 const compat_uptr_t __user *, argv,
2131 const compat_uptr_t __user *, envp,
2132 int, flags)
2133{
2134 return compat_do_execveat(fd,
2135 getname_uflags(filename, flags),
2136 argv, envp, flags);
2137}
2138#endif
2139
2140#ifdef CONFIG_SYSCTL
2141
2142static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2143 void *buffer, size_t *lenp, loff_t *ppos)
2144{
2145 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2146
2147 if (!error)
2148 validate_coredump_safety();
2149 return error;
2150}
2151
2152static struct ctl_table fs_exec_sysctls[] = {
2153 {
2154 .procname = "suid_dumpable",
2155 .data = &suid_dumpable,
2156 .maxlen = sizeof(int),
2157 .mode = 0644,
2158 .proc_handler = proc_dointvec_minmax_coredump,
2159 .extra1 = SYSCTL_ZERO,
2160 .extra2 = SYSCTL_TWO,
2161 },
2162 { }
2163};
2164
2165static int __init init_fs_exec_sysctls(void)
2166{
2167 register_sysctl_init("fs", fs_exec_sysctls);
2168 return 0;
2169}
2170
2171fs_initcall(init_fs_exec_sysctls);
2172#endif /* CONFIG_SYSCTL */
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#include "internal.h"
63
64int core_uses_pid;
65char core_pattern[CORENAME_MAX_SIZE] = "core";
66unsigned int core_pipe_limit;
67int suid_dumpable = 0;
68
69struct core_name {
70 char *corename;
71 int used, size;
72};
73static atomic_t call_count = ATOMIC_INIT(1);
74
75/* The maximal length of core_pattern is also specified in sysctl.c */
76
77static LIST_HEAD(formats);
78static DEFINE_RWLOCK(binfmt_lock);
79
80int __register_binfmt(struct linux_binfmt * fmt, int insert)
81{
82 if (!fmt)
83 return -EINVAL;
84 write_lock(&binfmt_lock);
85 insert ? list_add(&fmt->lh, &formats) :
86 list_add_tail(&fmt->lh, &formats);
87 write_unlock(&binfmt_lock);
88 return 0;
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
107/*
108 * Note that a shared library must be both readable and executable due to
109 * security reasons.
110 *
111 * Also note that we take the address to load from from the file itself.
112 */
113SYSCALL_DEFINE1(uselib, const char __user *, library)
114{
115 struct file *file;
116 char *tmp = getname(library);
117 int error = PTR_ERR(tmp);
118 static const struct open_flags uselib_flags = {
119 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
120 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
121 .intent = LOOKUP_OPEN
122 };
123
124 if (IS_ERR(tmp))
125 goto out;
126
127 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
128 putname(tmp);
129 error = PTR_ERR(file);
130 if (IS_ERR(file))
131 goto out;
132
133 error = -EINVAL;
134 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
135 goto exit;
136
137 error = -EACCES;
138 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
139 goto exit;
140
141 fsnotify_open(file);
142
143 error = -ENOEXEC;
144 if(file->f_op) {
145 struct linux_binfmt * fmt;
146
147 read_lock(&binfmt_lock);
148 list_for_each_entry(fmt, &formats, lh) {
149 if (!fmt->load_shlib)
150 continue;
151 if (!try_module_get(fmt->module))
152 continue;
153 read_unlock(&binfmt_lock);
154 error = fmt->load_shlib(file);
155 read_lock(&binfmt_lock);
156 put_binfmt(fmt);
157 if (error != -ENOEXEC)
158 break;
159 }
160 read_unlock(&binfmt_lock);
161 }
162exit:
163 fput(file);
164out:
165 return error;
166}
167
168#ifdef CONFIG_MMU
169/*
170 * The nascent bprm->mm is not visible until exec_mmap() but it can
171 * use a lot of memory, account these pages in current->mm temporary
172 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
173 * change the counter back via acct_arg_size(0).
174 */
175static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
176{
177 struct mm_struct *mm = current->mm;
178 long diff = (long)(pages - bprm->vma_pages);
179
180 if (!mm || !diff)
181 return;
182
183 bprm->vma_pages = pages;
184 add_mm_counter(mm, MM_ANONPAGES, diff);
185}
186
187static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
188 int write)
189{
190 struct page *page;
191 int ret;
192
193#ifdef CONFIG_STACK_GROWSUP
194 if (write) {
195 ret = expand_downwards(bprm->vma, pos);
196 if (ret < 0)
197 return NULL;
198 }
199#endif
200 ret = get_user_pages(current, bprm->mm, pos,
201 1, write, 1, &page, NULL);
202 if (ret <= 0)
203 return NULL;
204
205 if (write) {
206 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
207 struct rlimit *rlim;
208
209 acct_arg_size(bprm, size / PAGE_SIZE);
210
211 /*
212 * We've historically supported up to 32 pages (ARG_MAX)
213 * of argument strings even with small stacks
214 */
215 if (size <= ARG_MAX)
216 return page;
217
218 /*
219 * Limit to 1/4-th the stack size for the argv+env strings.
220 * This ensures that:
221 * - the remaining binfmt code will not run out of stack space,
222 * - the program will have a reasonable amount of stack left
223 * to work from.
224 */
225 rlim = current->signal->rlim;
226 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
227 put_page(page);
228 return NULL;
229 }
230 }
231
232 return page;
233}
234
235static void put_arg_page(struct page *page)
236{
237 put_page(page);
238}
239
240static void free_arg_page(struct linux_binprm *bprm, int i)
241{
242}
243
244static void free_arg_pages(struct linux_binprm *bprm)
245{
246}
247
248static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
249 struct page *page)
250{
251 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
252}
253
254static int __bprm_mm_init(struct linux_binprm *bprm)
255{
256 int err;
257 struct vm_area_struct *vma = NULL;
258 struct mm_struct *mm = bprm->mm;
259
260 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
261 if (!vma)
262 return -ENOMEM;
263
264 down_write(&mm->mmap_sem);
265 vma->vm_mm = mm;
266
267 /*
268 * Place the stack at the largest stack address the architecture
269 * supports. Later, we'll move this to an appropriate place. We don't
270 * use STACK_TOP because that can depend on attributes which aren't
271 * configured yet.
272 */
273 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
274 vma->vm_end = STACK_TOP_MAX;
275 vma->vm_start = vma->vm_end - PAGE_SIZE;
276 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
277 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
278 INIT_LIST_HEAD(&vma->anon_vma_chain);
279
280 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
281 if (err)
282 goto err;
283
284 err = insert_vm_struct(mm, vma);
285 if (err)
286 goto err;
287
288 mm->stack_vm = mm->total_vm = 1;
289 up_write(&mm->mmap_sem);
290 bprm->p = vma->vm_end - sizeof(void *);
291 return 0;
292err:
293 up_write(&mm->mmap_sem);
294 bprm->vma = NULL;
295 kmem_cache_free(vm_area_cachep, vma);
296 return err;
297}
298
299static bool valid_arg_len(struct linux_binprm *bprm, long len)
300{
301 return len <= MAX_ARG_STRLEN;
302}
303
304#else
305
306static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
307{
308}
309
310static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
311 int write)
312{
313 struct page *page;
314
315 page = bprm->page[pos / PAGE_SIZE];
316 if (!page && write) {
317 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
318 if (!page)
319 return NULL;
320 bprm->page[pos / PAGE_SIZE] = page;
321 }
322
323 return page;
324}
325
326static void put_arg_page(struct page *page)
327{
328}
329
330static void free_arg_page(struct linux_binprm *bprm, int i)
331{
332 if (bprm->page[i]) {
333 __free_page(bprm->page[i]);
334 bprm->page[i] = NULL;
335 }
336}
337
338static void free_arg_pages(struct linux_binprm *bprm)
339{
340 int i;
341
342 for (i = 0; i < MAX_ARG_PAGES; i++)
343 free_arg_page(bprm, i);
344}
345
346static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
347 struct page *page)
348{
349}
350
351static int __bprm_mm_init(struct linux_binprm *bprm)
352{
353 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
354 return 0;
355}
356
357static bool valid_arg_len(struct linux_binprm *bprm, long len)
358{
359 return len <= bprm->p;
360}
361
362#endif /* CONFIG_MMU */
363
364/*
365 * Create a new mm_struct and populate it with a temporary stack
366 * vm_area_struct. We don't have enough context at this point to set the stack
367 * flags, permissions, and offset, so we use temporary values. We'll update
368 * them later in setup_arg_pages().
369 */
370int bprm_mm_init(struct linux_binprm *bprm)
371{
372 int err;
373 struct mm_struct *mm = NULL;
374
375 bprm->mm = mm = mm_alloc();
376 err = -ENOMEM;
377 if (!mm)
378 goto err;
379
380 err = init_new_context(current, mm);
381 if (err)
382 goto err;
383
384 err = __bprm_mm_init(bprm);
385 if (err)
386 goto err;
387
388 return 0;
389
390err:
391 if (mm) {
392 bprm->mm = NULL;
393 mmdrop(mm);
394 }
395
396 return err;
397}
398
399struct user_arg_ptr {
400#ifdef CONFIG_COMPAT
401 bool is_compat;
402#endif
403 union {
404 const char __user *const __user *native;
405#ifdef CONFIG_COMPAT
406 compat_uptr_t __user *compat;
407#endif
408 } ptr;
409};
410
411static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
412{
413 const char __user *native;
414
415#ifdef CONFIG_COMPAT
416 if (unlikely(argv.is_compat)) {
417 compat_uptr_t compat;
418
419 if (get_user(compat, argv.ptr.compat + nr))
420 return ERR_PTR(-EFAULT);
421
422 return compat_ptr(compat);
423 }
424#endif
425
426 if (get_user(native, argv.ptr.native + nr))
427 return ERR_PTR(-EFAULT);
428
429 return native;
430}
431
432/*
433 * count() counts the number of strings in array ARGV.
434 */
435static int count(struct user_arg_ptr argv, int max)
436{
437 int i = 0;
438
439 if (argv.ptr.native != NULL) {
440 for (;;) {
441 const char __user *p = get_user_arg_ptr(argv, i);
442
443 if (!p)
444 break;
445
446 if (IS_ERR(p))
447 return -EFAULT;
448
449 if (i++ >= max)
450 return -E2BIG;
451
452 if (fatal_signal_pending(current))
453 return -ERESTARTNOHAND;
454 cond_resched();
455 }
456 }
457 return i;
458}
459
460/*
461 * 'copy_strings()' copies argument/environment strings from the old
462 * processes's memory to the new process's stack. The call to get_user_pages()
463 * ensures the destination page is created and not swapped out.
464 */
465static int copy_strings(int argc, struct user_arg_ptr argv,
466 struct linux_binprm *bprm)
467{
468 struct page *kmapped_page = NULL;
469 char *kaddr = NULL;
470 unsigned long kpos = 0;
471 int ret;
472
473 while (argc-- > 0) {
474 const char __user *str;
475 int len;
476 unsigned long pos;
477
478 ret = -EFAULT;
479 str = get_user_arg_ptr(argv, argc);
480 if (IS_ERR(str))
481 goto out;
482
483 len = strnlen_user(str, MAX_ARG_STRLEN);
484 if (!len)
485 goto out;
486
487 ret = -E2BIG;
488 if (!valid_arg_len(bprm, len))
489 goto out;
490
491 /* We're going to work our way backwords. */
492 pos = bprm->p;
493 str += len;
494 bprm->p -= len;
495
496 while (len > 0) {
497 int offset, bytes_to_copy;
498
499 if (fatal_signal_pending(current)) {
500 ret = -ERESTARTNOHAND;
501 goto out;
502 }
503 cond_resched();
504
505 offset = pos % PAGE_SIZE;
506 if (offset == 0)
507 offset = PAGE_SIZE;
508
509 bytes_to_copy = offset;
510 if (bytes_to_copy > len)
511 bytes_to_copy = len;
512
513 offset -= bytes_to_copy;
514 pos -= bytes_to_copy;
515 str -= bytes_to_copy;
516 len -= bytes_to_copy;
517
518 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
519 struct page *page;
520
521 page = get_arg_page(bprm, pos, 1);
522 if (!page) {
523 ret = -E2BIG;
524 goto out;
525 }
526
527 if (kmapped_page) {
528 flush_kernel_dcache_page(kmapped_page);
529 kunmap(kmapped_page);
530 put_arg_page(kmapped_page);
531 }
532 kmapped_page = page;
533 kaddr = kmap(kmapped_page);
534 kpos = pos & PAGE_MASK;
535 flush_arg_page(bprm, kpos, kmapped_page);
536 }
537 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
538 ret = -EFAULT;
539 goto out;
540 }
541 }
542 }
543 ret = 0;
544out:
545 if (kmapped_page) {
546 flush_kernel_dcache_page(kmapped_page);
547 kunmap(kmapped_page);
548 put_arg_page(kmapped_page);
549 }
550 return ret;
551}
552
553/*
554 * Like copy_strings, but get argv and its values from kernel memory.
555 */
556int copy_strings_kernel(int argc, const char *const *__argv,
557 struct linux_binprm *bprm)
558{
559 int r;
560 mm_segment_t oldfs = get_fs();
561 struct user_arg_ptr argv = {
562 .ptr.native = (const char __user *const __user *)__argv,
563 };
564
565 set_fs(KERNEL_DS);
566 r = copy_strings(argc, argv, bprm);
567 set_fs(oldfs);
568
569 return r;
570}
571EXPORT_SYMBOL(copy_strings_kernel);
572
573#ifdef CONFIG_MMU
574
575/*
576 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
577 * the binfmt code determines where the new stack should reside, we shift it to
578 * its final location. The process proceeds as follows:
579 *
580 * 1) Use shift to calculate the new vma endpoints.
581 * 2) Extend vma to cover both the old and new ranges. This ensures the
582 * arguments passed to subsequent functions are consistent.
583 * 3) Move vma's page tables to the new range.
584 * 4) Free up any cleared pgd range.
585 * 5) Shrink the vma to cover only the new range.
586 */
587static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
588{
589 struct mm_struct *mm = vma->vm_mm;
590 unsigned long old_start = vma->vm_start;
591 unsigned long old_end = vma->vm_end;
592 unsigned long length = old_end - old_start;
593 unsigned long new_start = old_start - shift;
594 unsigned long new_end = old_end - shift;
595 struct mmu_gather tlb;
596
597 BUG_ON(new_start > new_end);
598
599 /*
600 * ensure there are no vmas between where we want to go
601 * and where we are
602 */
603 if (vma != find_vma(mm, new_start))
604 return -EFAULT;
605
606 /*
607 * cover the whole range: [new_start, old_end)
608 */
609 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
610 return -ENOMEM;
611
612 /*
613 * move the page tables downwards, on failure we rely on
614 * process cleanup to remove whatever mess we made.
615 */
616 if (length != move_page_tables(vma, old_start,
617 vma, new_start, length))
618 return -ENOMEM;
619
620 lru_add_drain();
621 tlb_gather_mmu(&tlb, mm, 0);
622 if (new_end > old_start) {
623 /*
624 * when the old and new regions overlap clear from new_end.
625 */
626 free_pgd_range(&tlb, new_end, old_end, new_end,
627 vma->vm_next ? vma->vm_next->vm_start : 0);
628 } else {
629 /*
630 * otherwise, clean from old_start; this is done to not touch
631 * the address space in [new_end, old_start) some architectures
632 * have constraints on va-space that make this illegal (IA64) -
633 * for the others its just a little faster.
634 */
635 free_pgd_range(&tlb, old_start, old_end, new_end,
636 vma->vm_next ? vma->vm_next->vm_start : 0);
637 }
638 tlb_finish_mmu(&tlb, new_end, old_end);
639
640 /*
641 * Shrink the vma to just the new range. Always succeeds.
642 */
643 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
644
645 return 0;
646}
647
648/*
649 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
650 * the stack is optionally relocated, and some extra space is added.
651 */
652int setup_arg_pages(struct linux_binprm *bprm,
653 unsigned long stack_top,
654 int executable_stack)
655{
656 unsigned long ret;
657 unsigned long stack_shift;
658 struct mm_struct *mm = current->mm;
659 struct vm_area_struct *vma = bprm->vma;
660 struct vm_area_struct *prev = NULL;
661 unsigned long vm_flags;
662 unsigned long stack_base;
663 unsigned long stack_size;
664 unsigned long stack_expand;
665 unsigned long rlim_stack;
666
667#ifdef CONFIG_STACK_GROWSUP
668 /* Limit stack size to 1GB */
669 stack_base = rlimit_max(RLIMIT_STACK);
670 if (stack_base > (1 << 30))
671 stack_base = 1 << 30;
672
673 /* Make sure we didn't let the argument array grow too large. */
674 if (vma->vm_end - vma->vm_start > stack_base)
675 return -ENOMEM;
676
677 stack_base = PAGE_ALIGN(stack_top - stack_base);
678
679 stack_shift = vma->vm_start - stack_base;
680 mm->arg_start = bprm->p - stack_shift;
681 bprm->p = vma->vm_end - stack_shift;
682#else
683 stack_top = arch_align_stack(stack_top);
684 stack_top = PAGE_ALIGN(stack_top);
685
686 if (unlikely(stack_top < mmap_min_addr) ||
687 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
688 return -ENOMEM;
689
690 stack_shift = vma->vm_end - stack_top;
691
692 bprm->p -= stack_shift;
693 mm->arg_start = bprm->p;
694#endif
695
696 if (bprm->loader)
697 bprm->loader -= stack_shift;
698 bprm->exec -= stack_shift;
699
700 down_write(&mm->mmap_sem);
701 vm_flags = VM_STACK_FLAGS;
702
703 /*
704 * Adjust stack execute permissions; explicitly enable for
705 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
706 * (arch default) otherwise.
707 */
708 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
709 vm_flags |= VM_EXEC;
710 else if (executable_stack == EXSTACK_DISABLE_X)
711 vm_flags &= ~VM_EXEC;
712 vm_flags |= mm->def_flags;
713 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
714
715 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
716 vm_flags);
717 if (ret)
718 goto out_unlock;
719 BUG_ON(prev != vma);
720
721 /* Move stack pages down in memory. */
722 if (stack_shift) {
723 ret = shift_arg_pages(vma, stack_shift);
724 if (ret)
725 goto out_unlock;
726 }
727
728 /* mprotect_fixup is overkill to remove the temporary stack flags */
729 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
730
731 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
732 stack_size = vma->vm_end - vma->vm_start;
733 /*
734 * Align this down to a page boundary as expand_stack
735 * will align it up.
736 */
737 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
738#ifdef CONFIG_STACK_GROWSUP
739 if (stack_size + stack_expand > rlim_stack)
740 stack_base = vma->vm_start + rlim_stack;
741 else
742 stack_base = vma->vm_end + stack_expand;
743#else
744 if (stack_size + stack_expand > rlim_stack)
745 stack_base = vma->vm_end - rlim_stack;
746 else
747 stack_base = vma->vm_start - stack_expand;
748#endif
749 current->mm->start_stack = bprm->p;
750 ret = expand_stack(vma, stack_base);
751 if (ret)
752 ret = -EFAULT;
753
754out_unlock:
755 up_write(&mm->mmap_sem);
756 return ret;
757}
758EXPORT_SYMBOL(setup_arg_pages);
759
760#endif /* CONFIG_MMU */
761
762struct file *open_exec(const char *name)
763{
764 struct file *file;
765 int err;
766 static const struct open_flags open_exec_flags = {
767 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
768 .acc_mode = MAY_EXEC | MAY_OPEN,
769 .intent = LOOKUP_OPEN
770 };
771
772 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
773 if (IS_ERR(file))
774 goto out;
775
776 err = -EACCES;
777 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
778 goto exit;
779
780 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
781 goto exit;
782
783 fsnotify_open(file);
784
785 err = deny_write_access(file);
786 if (err)
787 goto exit;
788
789out:
790 return file;
791
792exit:
793 fput(file);
794 return ERR_PTR(err);
795}
796EXPORT_SYMBOL(open_exec);
797
798int kernel_read(struct file *file, loff_t offset,
799 char *addr, unsigned long count)
800{
801 mm_segment_t old_fs;
802 loff_t pos = offset;
803 int result;
804
805 old_fs = get_fs();
806 set_fs(get_ds());
807 /* The cast to a user pointer is valid due to the set_fs() */
808 result = vfs_read(file, (void __user *)addr, count, &pos);
809 set_fs(old_fs);
810 return result;
811}
812
813EXPORT_SYMBOL(kernel_read);
814
815static int exec_mmap(struct mm_struct *mm)
816{
817 struct task_struct *tsk;
818 struct mm_struct * old_mm, *active_mm;
819
820 /* Notify parent that we're no longer interested in the old VM */
821 tsk = current;
822 old_mm = current->mm;
823 sync_mm_rss(tsk, old_mm);
824 mm_release(tsk, old_mm);
825
826 if (old_mm) {
827 /*
828 * Make sure that if there is a core dump in progress
829 * for the old mm, we get out and die instead of going
830 * through with the exec. We must hold mmap_sem around
831 * checking core_state and changing tsk->mm.
832 */
833 down_read(&old_mm->mmap_sem);
834 if (unlikely(old_mm->core_state)) {
835 up_read(&old_mm->mmap_sem);
836 return -EINTR;
837 }
838 }
839 task_lock(tsk);
840 active_mm = tsk->active_mm;
841 tsk->mm = mm;
842 tsk->active_mm = mm;
843 activate_mm(active_mm, mm);
844 if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
845 atomic_dec(&old_mm->oom_disable_count);
846 atomic_inc(&tsk->mm->oom_disable_count);
847 }
848 task_unlock(tsk);
849 arch_pick_mmap_layout(mm);
850 if (old_mm) {
851 up_read(&old_mm->mmap_sem);
852 BUG_ON(active_mm != old_mm);
853 mm_update_next_owner(old_mm);
854 mmput(old_mm);
855 return 0;
856 }
857 mmdrop(active_mm);
858 return 0;
859}
860
861/*
862 * This function makes sure the current process has its own signal table,
863 * so that flush_signal_handlers can later reset the handlers without
864 * disturbing other processes. (Other processes might share the signal
865 * table via the CLONE_SIGHAND option to clone().)
866 */
867static int de_thread(struct task_struct *tsk)
868{
869 struct signal_struct *sig = tsk->signal;
870 struct sighand_struct *oldsighand = tsk->sighand;
871 spinlock_t *lock = &oldsighand->siglock;
872
873 if (thread_group_empty(tsk))
874 goto no_thread_group;
875
876 /*
877 * Kill all other threads in the thread group.
878 */
879 spin_lock_irq(lock);
880 if (signal_group_exit(sig)) {
881 /*
882 * Another group action in progress, just
883 * return so that the signal is processed.
884 */
885 spin_unlock_irq(lock);
886 return -EAGAIN;
887 }
888
889 sig->group_exit_task = tsk;
890 sig->notify_count = zap_other_threads(tsk);
891 if (!thread_group_leader(tsk))
892 sig->notify_count--;
893
894 while (sig->notify_count) {
895 __set_current_state(TASK_UNINTERRUPTIBLE);
896 spin_unlock_irq(lock);
897 schedule();
898 spin_lock_irq(lock);
899 }
900 spin_unlock_irq(lock);
901
902 /*
903 * At this point all other threads have exited, all we have to
904 * do is to wait for the thread group leader to become inactive,
905 * and to assume its PID:
906 */
907 if (!thread_group_leader(tsk)) {
908 struct task_struct *leader = tsk->group_leader;
909
910 sig->notify_count = -1; /* for exit_notify() */
911 for (;;) {
912 write_lock_irq(&tasklist_lock);
913 if (likely(leader->exit_state))
914 break;
915 __set_current_state(TASK_UNINTERRUPTIBLE);
916 write_unlock_irq(&tasklist_lock);
917 schedule();
918 }
919
920 /*
921 * The only record we have of the real-time age of a
922 * process, regardless of execs it's done, is start_time.
923 * All the past CPU time is accumulated in signal_struct
924 * from sister threads now dead. But in this non-leader
925 * exec, nothing survives from the original leader thread,
926 * whose birth marks the true age of this process now.
927 * When we take on its identity by switching to its PID, we
928 * also take its birthdate (always earlier than our own).
929 */
930 tsk->start_time = leader->start_time;
931
932 BUG_ON(!same_thread_group(leader, tsk));
933 BUG_ON(has_group_leader_pid(tsk));
934 /*
935 * An exec() starts a new thread group with the
936 * TGID of the previous thread group. Rehash the
937 * two threads with a switched PID, and release
938 * the former thread group leader:
939 */
940
941 /* Become a process group leader with the old leader's pid.
942 * The old leader becomes a thread of the this thread group.
943 * Note: The old leader also uses this pid until release_task
944 * is called. Odd but simple and correct.
945 */
946 detach_pid(tsk, PIDTYPE_PID);
947 tsk->pid = leader->pid;
948 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
949 transfer_pid(leader, tsk, PIDTYPE_PGID);
950 transfer_pid(leader, tsk, PIDTYPE_SID);
951
952 list_replace_rcu(&leader->tasks, &tsk->tasks);
953 list_replace_init(&leader->sibling, &tsk->sibling);
954
955 tsk->group_leader = tsk;
956 leader->group_leader = tsk;
957
958 tsk->exit_signal = SIGCHLD;
959 leader->exit_signal = -1;
960
961 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
962 leader->exit_state = EXIT_DEAD;
963
964 /*
965 * We are going to release_task()->ptrace_unlink() silently,
966 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
967 * the tracer wont't block again waiting for this thread.
968 */
969 if (unlikely(leader->ptrace))
970 __wake_up_parent(leader, leader->parent);
971 write_unlock_irq(&tasklist_lock);
972
973 release_task(leader);
974 }
975
976 sig->group_exit_task = NULL;
977 sig->notify_count = 0;
978
979no_thread_group:
980 if (current->mm)
981 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
982
983 exit_itimers(sig);
984 flush_itimer_signals();
985
986 if (atomic_read(&oldsighand->count) != 1) {
987 struct sighand_struct *newsighand;
988 /*
989 * This ->sighand is shared with the CLONE_SIGHAND
990 * but not CLONE_THREAD task, switch to the new one.
991 */
992 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
993 if (!newsighand)
994 return -ENOMEM;
995
996 atomic_set(&newsighand->count, 1);
997 memcpy(newsighand->action, oldsighand->action,
998 sizeof(newsighand->action));
999
1000 write_lock_irq(&tasklist_lock);
1001 spin_lock(&oldsighand->siglock);
1002 rcu_assign_pointer(tsk->sighand, newsighand);
1003 spin_unlock(&oldsighand->siglock);
1004 write_unlock_irq(&tasklist_lock);
1005
1006 __cleanup_sighand(oldsighand);
1007 }
1008
1009 BUG_ON(!thread_group_leader(tsk));
1010 return 0;
1011}
1012
1013/*
1014 * These functions flushes out all traces of the currently running executable
1015 * so that a new one can be started
1016 */
1017static void flush_old_files(struct files_struct * files)
1018{
1019 long j = -1;
1020 struct fdtable *fdt;
1021
1022 spin_lock(&files->file_lock);
1023 for (;;) {
1024 unsigned long set, i;
1025
1026 j++;
1027 i = j * __NFDBITS;
1028 fdt = files_fdtable(files);
1029 if (i >= fdt->max_fds)
1030 break;
1031 set = fdt->close_on_exec->fds_bits[j];
1032 if (!set)
1033 continue;
1034 fdt->close_on_exec->fds_bits[j] = 0;
1035 spin_unlock(&files->file_lock);
1036 for ( ; set ; i++,set >>= 1) {
1037 if (set & 1) {
1038 sys_close(i);
1039 }
1040 }
1041 spin_lock(&files->file_lock);
1042
1043 }
1044 spin_unlock(&files->file_lock);
1045}
1046
1047char *get_task_comm(char *buf, struct task_struct *tsk)
1048{
1049 /* buf must be at least sizeof(tsk->comm) in size */
1050 task_lock(tsk);
1051 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1052 task_unlock(tsk);
1053 return buf;
1054}
1055EXPORT_SYMBOL_GPL(get_task_comm);
1056
1057void set_task_comm(struct task_struct *tsk, char *buf)
1058{
1059 task_lock(tsk);
1060
1061 /*
1062 * Threads may access current->comm without holding
1063 * the task lock, so write the string carefully.
1064 * Readers without a lock may see incomplete new
1065 * names but are safe from non-terminating string reads.
1066 */
1067 memset(tsk->comm, 0, TASK_COMM_LEN);
1068 wmb();
1069 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1070 task_unlock(tsk);
1071 perf_event_comm(tsk);
1072}
1073
1074int flush_old_exec(struct linux_binprm * bprm)
1075{
1076 int retval;
1077
1078 /*
1079 * Make sure we have a private signal table and that
1080 * we are unassociated from the previous thread group.
1081 */
1082 retval = de_thread(current);
1083 if (retval)
1084 goto out;
1085
1086 set_mm_exe_file(bprm->mm, bprm->file);
1087
1088 /*
1089 * Release all of the old mmap stuff
1090 */
1091 acct_arg_size(bprm, 0);
1092 retval = exec_mmap(bprm->mm);
1093 if (retval)
1094 goto out;
1095
1096 bprm->mm = NULL; /* We're using it now */
1097
1098 set_fs(USER_DS);
1099 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1100 flush_thread();
1101 current->personality &= ~bprm->per_clear;
1102
1103 return 0;
1104
1105out:
1106 return retval;
1107}
1108EXPORT_SYMBOL(flush_old_exec);
1109
1110void would_dump(struct linux_binprm *bprm, struct file *file)
1111{
1112 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1113 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1114}
1115EXPORT_SYMBOL(would_dump);
1116
1117void setup_new_exec(struct linux_binprm * bprm)
1118{
1119 int i, ch;
1120 const char *name;
1121 char tcomm[sizeof(current->comm)];
1122
1123 arch_pick_mmap_layout(current->mm);
1124
1125 /* This is the point of no return */
1126 current->sas_ss_sp = current->sas_ss_size = 0;
1127
1128 if (current_euid() == current_uid() && current_egid() == current_gid())
1129 set_dumpable(current->mm, 1);
1130 else
1131 set_dumpable(current->mm, suid_dumpable);
1132
1133 name = bprm->filename;
1134
1135 /* Copies the binary name from after last slash */
1136 for (i=0; (ch = *(name++)) != '\0';) {
1137 if (ch == '/')
1138 i = 0; /* overwrite what we wrote */
1139 else
1140 if (i < (sizeof(tcomm) - 1))
1141 tcomm[i++] = ch;
1142 }
1143 tcomm[i] = '\0';
1144 set_task_comm(current, tcomm);
1145
1146 /* Set the new mm task size. We have to do that late because it may
1147 * depend on TIF_32BIT which is only updated in flush_thread() on
1148 * some architectures like powerpc
1149 */
1150 current->mm->task_size = TASK_SIZE;
1151
1152 /* install the new credentials */
1153 if (bprm->cred->uid != current_euid() ||
1154 bprm->cred->gid != current_egid()) {
1155 current->pdeath_signal = 0;
1156 } else {
1157 would_dump(bprm, bprm->file);
1158 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1159 set_dumpable(current->mm, suid_dumpable);
1160 }
1161
1162 /*
1163 * Flush performance counters when crossing a
1164 * security domain:
1165 */
1166 if (!get_dumpable(current->mm))
1167 perf_event_exit_task(current);
1168
1169 /* An exec changes our domain. We are no longer part of the thread
1170 group */
1171
1172 current->self_exec_id++;
1173
1174 flush_signal_handlers(current, 0);
1175 flush_old_files(current->files);
1176}
1177EXPORT_SYMBOL(setup_new_exec);
1178
1179/*
1180 * Prepare credentials and lock ->cred_guard_mutex.
1181 * install_exec_creds() commits the new creds and drops the lock.
1182 * Or, if exec fails before, free_bprm() should release ->cred and
1183 * and unlock.
1184 */
1185int prepare_bprm_creds(struct linux_binprm *bprm)
1186{
1187 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1188 return -ERESTARTNOINTR;
1189
1190 bprm->cred = prepare_exec_creds();
1191 if (likely(bprm->cred))
1192 return 0;
1193
1194 mutex_unlock(¤t->signal->cred_guard_mutex);
1195 return -ENOMEM;
1196}
1197
1198void free_bprm(struct linux_binprm *bprm)
1199{
1200 free_arg_pages(bprm);
1201 if (bprm->cred) {
1202 mutex_unlock(¤t->signal->cred_guard_mutex);
1203 abort_creds(bprm->cred);
1204 }
1205 kfree(bprm);
1206}
1207
1208/*
1209 * install the new credentials for this executable
1210 */
1211void install_exec_creds(struct linux_binprm *bprm)
1212{
1213 security_bprm_committing_creds(bprm);
1214
1215 commit_creds(bprm->cred);
1216 bprm->cred = NULL;
1217 /*
1218 * cred_guard_mutex must be held at least to this point to prevent
1219 * ptrace_attach() from altering our determination of the task's
1220 * credentials; any time after this it may be unlocked.
1221 */
1222 security_bprm_committed_creds(bprm);
1223 mutex_unlock(¤t->signal->cred_guard_mutex);
1224}
1225EXPORT_SYMBOL(install_exec_creds);
1226
1227/*
1228 * determine how safe it is to execute the proposed program
1229 * - the caller must hold ->cred_guard_mutex to protect against
1230 * PTRACE_ATTACH
1231 */
1232int check_unsafe_exec(struct linux_binprm *bprm)
1233{
1234 struct task_struct *p = current, *t;
1235 unsigned n_fs;
1236 int res = 0;
1237
1238 if (p->ptrace) {
1239 if (p->ptrace & PT_PTRACE_CAP)
1240 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1241 else
1242 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1243 }
1244
1245 n_fs = 1;
1246 spin_lock(&p->fs->lock);
1247 rcu_read_lock();
1248 for (t = next_thread(p); t != p; t = next_thread(t)) {
1249 if (t->fs == p->fs)
1250 n_fs++;
1251 }
1252 rcu_read_unlock();
1253
1254 if (p->fs->users > n_fs) {
1255 bprm->unsafe |= LSM_UNSAFE_SHARE;
1256 } else {
1257 res = -EAGAIN;
1258 if (!p->fs->in_exec) {
1259 p->fs->in_exec = 1;
1260 res = 1;
1261 }
1262 }
1263 spin_unlock(&p->fs->lock);
1264
1265 return res;
1266}
1267
1268/*
1269 * Fill the binprm structure from the inode.
1270 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1271 *
1272 * This may be called multiple times for binary chains (scripts for example).
1273 */
1274int prepare_binprm(struct linux_binprm *bprm)
1275{
1276 umode_t mode;
1277 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1278 int retval;
1279
1280 mode = inode->i_mode;
1281 if (bprm->file->f_op == NULL)
1282 return -EACCES;
1283
1284 /* clear any previous set[ug]id data from a previous binary */
1285 bprm->cred->euid = current_euid();
1286 bprm->cred->egid = current_egid();
1287
1288 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1289 /* Set-uid? */
1290 if (mode & S_ISUID) {
1291 bprm->per_clear |= PER_CLEAR_ON_SETID;
1292 bprm->cred->euid = inode->i_uid;
1293 }
1294
1295 /* Set-gid? */
1296 /*
1297 * If setgid is set but no group execute bit then this
1298 * is a candidate for mandatory locking, not a setgid
1299 * executable.
1300 */
1301 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1302 bprm->per_clear |= PER_CLEAR_ON_SETID;
1303 bprm->cred->egid = inode->i_gid;
1304 }
1305 }
1306
1307 /* fill in binprm security blob */
1308 retval = security_bprm_set_creds(bprm);
1309 if (retval)
1310 return retval;
1311 bprm->cred_prepared = 1;
1312
1313 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1314 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1315}
1316
1317EXPORT_SYMBOL(prepare_binprm);
1318
1319/*
1320 * Arguments are '\0' separated strings found at the location bprm->p
1321 * points to; chop off the first by relocating brpm->p to right after
1322 * the first '\0' encountered.
1323 */
1324int remove_arg_zero(struct linux_binprm *bprm)
1325{
1326 int ret = 0;
1327 unsigned long offset;
1328 char *kaddr;
1329 struct page *page;
1330
1331 if (!bprm->argc)
1332 return 0;
1333
1334 do {
1335 offset = bprm->p & ~PAGE_MASK;
1336 page = get_arg_page(bprm, bprm->p, 0);
1337 if (!page) {
1338 ret = -EFAULT;
1339 goto out;
1340 }
1341 kaddr = kmap_atomic(page, KM_USER0);
1342
1343 for (; offset < PAGE_SIZE && kaddr[offset];
1344 offset++, bprm->p++)
1345 ;
1346
1347 kunmap_atomic(kaddr, KM_USER0);
1348 put_arg_page(page);
1349
1350 if (offset == PAGE_SIZE)
1351 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1352 } while (offset == PAGE_SIZE);
1353
1354 bprm->p++;
1355 bprm->argc--;
1356 ret = 0;
1357
1358out:
1359 return ret;
1360}
1361EXPORT_SYMBOL(remove_arg_zero);
1362
1363/*
1364 * cycle the list of binary formats handler, until one recognizes the image
1365 */
1366int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1367{
1368 unsigned int depth = bprm->recursion_depth;
1369 int try,retval;
1370 struct linux_binfmt *fmt;
1371 pid_t old_pid;
1372
1373 retval = security_bprm_check(bprm);
1374 if (retval)
1375 return retval;
1376
1377 retval = audit_bprm(bprm);
1378 if (retval)
1379 return retval;
1380
1381 /* Need to fetch pid before load_binary changes it */
1382 rcu_read_lock();
1383 old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1384 rcu_read_unlock();
1385
1386 retval = -ENOENT;
1387 for (try=0; try<2; try++) {
1388 read_lock(&binfmt_lock);
1389 list_for_each_entry(fmt, &formats, lh) {
1390 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1391 if (!fn)
1392 continue;
1393 if (!try_module_get(fmt->module))
1394 continue;
1395 read_unlock(&binfmt_lock);
1396 retval = fn(bprm, regs);
1397 /*
1398 * Restore the depth counter to its starting value
1399 * in this call, so we don't have to rely on every
1400 * load_binary function to restore it on return.
1401 */
1402 bprm->recursion_depth = depth;
1403 if (retval >= 0) {
1404 if (depth == 0)
1405 ptrace_event(PTRACE_EVENT_EXEC,
1406 old_pid);
1407 put_binfmt(fmt);
1408 allow_write_access(bprm->file);
1409 if (bprm->file)
1410 fput(bprm->file);
1411 bprm->file = NULL;
1412 current->did_exec = 1;
1413 proc_exec_connector(current);
1414 return retval;
1415 }
1416 read_lock(&binfmt_lock);
1417 put_binfmt(fmt);
1418 if (retval != -ENOEXEC || bprm->mm == NULL)
1419 break;
1420 if (!bprm->file) {
1421 read_unlock(&binfmt_lock);
1422 return retval;
1423 }
1424 }
1425 read_unlock(&binfmt_lock);
1426#ifdef CONFIG_MODULES
1427 if (retval != -ENOEXEC || bprm->mm == NULL) {
1428 break;
1429 } else {
1430#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1431 if (printable(bprm->buf[0]) &&
1432 printable(bprm->buf[1]) &&
1433 printable(bprm->buf[2]) &&
1434 printable(bprm->buf[3]))
1435 break; /* -ENOEXEC */
1436 if (try)
1437 break; /* -ENOEXEC */
1438 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1439 }
1440#else
1441 break;
1442#endif
1443 }
1444 return retval;
1445}
1446
1447EXPORT_SYMBOL(search_binary_handler);
1448
1449/*
1450 * sys_execve() executes a new program.
1451 */
1452static int do_execve_common(const char *filename,
1453 struct user_arg_ptr argv,
1454 struct user_arg_ptr envp,
1455 struct pt_regs *regs)
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,regs);
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 struct pt_regs *regs)
1583{
1584 struct user_arg_ptr argv = { .ptr.native = __argv };
1585 struct user_arg_ptr envp = { .ptr.native = __envp };
1586 return do_execve_common(filename, argv, envp, regs);
1587}
1588
1589#ifdef CONFIG_COMPAT
1590int compat_do_execve(char *filename,
1591 compat_uptr_t __user *__argv,
1592 compat_uptr_t __user *__envp,
1593 struct pt_regs *regs)
1594{
1595 struct user_arg_ptr argv = {
1596 .is_compat = true,
1597 .ptr.compat = __argv,
1598 };
1599 struct user_arg_ptr envp = {
1600 .is_compat = true,
1601 .ptr.compat = __envp,
1602 };
1603 return do_execve_common(filename, argv, envp, regs);
1604}
1605#endif
1606
1607void set_binfmt(struct linux_binfmt *new)
1608{
1609 struct mm_struct *mm = current->mm;
1610
1611 if (mm->binfmt)
1612 module_put(mm->binfmt->module);
1613
1614 mm->binfmt = new;
1615 if (new)
1616 __module_get(new->module);
1617}
1618
1619EXPORT_SYMBOL(set_binfmt);
1620
1621static int expand_corename(struct core_name *cn)
1622{
1623 char *old_corename = cn->corename;
1624
1625 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1626 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1627
1628 if (!cn->corename) {
1629 kfree(old_corename);
1630 return -ENOMEM;
1631 }
1632
1633 return 0;
1634}
1635
1636static int cn_printf(struct core_name *cn, const char *fmt, ...)
1637{
1638 char *cur;
1639 int need;
1640 int ret;
1641 va_list arg;
1642
1643 va_start(arg, fmt);
1644 need = vsnprintf(NULL, 0, fmt, arg);
1645 va_end(arg);
1646
1647 if (likely(need < cn->size - cn->used - 1))
1648 goto out_printf;
1649
1650 ret = expand_corename(cn);
1651 if (ret)
1652 goto expand_fail;
1653
1654out_printf:
1655 cur = cn->corename + cn->used;
1656 va_start(arg, fmt);
1657 vsnprintf(cur, need + 1, fmt, arg);
1658 va_end(arg);
1659 cn->used += need;
1660 return 0;
1661
1662expand_fail:
1663 return ret;
1664}
1665
1666static void cn_escape(char *str)
1667{
1668 for (; *str; str++)
1669 if (*str == '/')
1670 *str = '!';
1671}
1672
1673static int cn_print_exe_file(struct core_name *cn)
1674{
1675 struct file *exe_file;
1676 char *pathbuf, *path;
1677 int ret;
1678
1679 exe_file = get_mm_exe_file(current->mm);
1680 if (!exe_file) {
1681 char *commstart = cn->corename + cn->used;
1682 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1683 cn_escape(commstart);
1684 return ret;
1685 }
1686
1687 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1688 if (!pathbuf) {
1689 ret = -ENOMEM;
1690 goto put_exe_file;
1691 }
1692
1693 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1694 if (IS_ERR(path)) {
1695 ret = PTR_ERR(path);
1696 goto free_buf;
1697 }
1698
1699 cn_escape(path);
1700
1701 ret = cn_printf(cn, "%s", path);
1702
1703free_buf:
1704 kfree(pathbuf);
1705put_exe_file:
1706 fput(exe_file);
1707 return ret;
1708}
1709
1710/* format_corename will inspect the pattern parameter, and output a
1711 * name into corename, which must have space for at least
1712 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1713 */
1714static int format_corename(struct core_name *cn, long signr)
1715{
1716 const struct cred *cred = current_cred();
1717 const char *pat_ptr = core_pattern;
1718 int ispipe = (*pat_ptr == '|');
1719 int pid_in_pattern = 0;
1720 int err = 0;
1721
1722 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1723 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1724 cn->used = 0;
1725
1726 if (!cn->corename)
1727 return -ENOMEM;
1728
1729 /* Repeat as long as we have more pattern to process and more output
1730 space */
1731 while (*pat_ptr) {
1732 if (*pat_ptr != '%') {
1733 if (*pat_ptr == 0)
1734 goto out;
1735 err = cn_printf(cn, "%c", *pat_ptr++);
1736 } else {
1737 switch (*++pat_ptr) {
1738 /* single % at the end, drop that */
1739 case 0:
1740 goto out;
1741 /* Double percent, output one percent */
1742 case '%':
1743 err = cn_printf(cn, "%c", '%');
1744 break;
1745 /* pid */
1746 case 'p':
1747 pid_in_pattern = 1;
1748 err = cn_printf(cn, "%d",
1749 task_tgid_vnr(current));
1750 break;
1751 /* uid */
1752 case 'u':
1753 err = cn_printf(cn, "%d", cred->uid);
1754 break;
1755 /* gid */
1756 case 'g':
1757 err = cn_printf(cn, "%d", cred->gid);
1758 break;
1759 /* signal that caused the coredump */
1760 case 's':
1761 err = cn_printf(cn, "%ld", signr);
1762 break;
1763 /* UNIX time of coredump */
1764 case 't': {
1765 struct timeval tv;
1766 do_gettimeofday(&tv);
1767 err = cn_printf(cn, "%lu", tv.tv_sec);
1768 break;
1769 }
1770 /* hostname */
1771 case 'h': {
1772 char *namestart = cn->corename + cn->used;
1773 down_read(&uts_sem);
1774 err = cn_printf(cn, "%s",
1775 utsname()->nodename);
1776 up_read(&uts_sem);
1777 cn_escape(namestart);
1778 break;
1779 }
1780 /* executable */
1781 case 'e': {
1782 char *commstart = cn->corename + cn->used;
1783 err = cn_printf(cn, "%s", current->comm);
1784 cn_escape(commstart);
1785 break;
1786 }
1787 case 'E':
1788 err = cn_print_exe_file(cn);
1789 break;
1790 /* core limit size */
1791 case 'c':
1792 err = cn_printf(cn, "%lu",
1793 rlimit(RLIMIT_CORE));
1794 break;
1795 default:
1796 break;
1797 }
1798 ++pat_ptr;
1799 }
1800
1801 if (err)
1802 return err;
1803 }
1804
1805 /* Backward compatibility with core_uses_pid:
1806 *
1807 * If core_pattern does not include a %p (as is the default)
1808 * and core_uses_pid is set, then .%pid will be appended to
1809 * the filename. Do not do this for piped commands. */
1810 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1811 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1812 if (err)
1813 return err;
1814 }
1815out:
1816 return ispipe;
1817}
1818
1819static int zap_process(struct task_struct *start, int exit_code)
1820{
1821 struct task_struct *t;
1822 int nr = 0;
1823
1824 start->signal->flags = SIGNAL_GROUP_EXIT;
1825 start->signal->group_exit_code = exit_code;
1826 start->signal->group_stop_count = 0;
1827
1828 t = start;
1829 do {
1830 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1831 if (t != current && t->mm) {
1832 sigaddset(&t->pending.signal, SIGKILL);
1833 signal_wake_up(t, 1);
1834 nr++;
1835 }
1836 } while_each_thread(start, t);
1837
1838 return nr;
1839}
1840
1841static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1842 struct core_state *core_state, int exit_code)
1843{
1844 struct task_struct *g, *p;
1845 unsigned long flags;
1846 int nr = -EAGAIN;
1847
1848 spin_lock_irq(&tsk->sighand->siglock);
1849 if (!signal_group_exit(tsk->signal)) {
1850 mm->core_state = core_state;
1851 nr = zap_process(tsk, exit_code);
1852 }
1853 spin_unlock_irq(&tsk->sighand->siglock);
1854 if (unlikely(nr < 0))
1855 return nr;
1856
1857 if (atomic_read(&mm->mm_users) == nr + 1)
1858 goto done;
1859 /*
1860 * We should find and kill all tasks which use this mm, and we should
1861 * count them correctly into ->nr_threads. We don't take tasklist
1862 * lock, but this is safe wrt:
1863 *
1864 * fork:
1865 * None of sub-threads can fork after zap_process(leader). All
1866 * processes which were created before this point should be
1867 * visible to zap_threads() because copy_process() adds the new
1868 * process to the tail of init_task.tasks list, and lock/unlock
1869 * of ->siglock provides a memory barrier.
1870 *
1871 * do_exit:
1872 * The caller holds mm->mmap_sem. This means that the task which
1873 * uses this mm can't pass exit_mm(), so it can't exit or clear
1874 * its ->mm.
1875 *
1876 * de_thread:
1877 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1878 * we must see either old or new leader, this does not matter.
1879 * However, it can change p->sighand, so lock_task_sighand(p)
1880 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1881 * it can't fail.
1882 *
1883 * Note also that "g" can be the old leader with ->mm == NULL
1884 * and already unhashed and thus removed from ->thread_group.
1885 * This is OK, __unhash_process()->list_del_rcu() does not
1886 * clear the ->next pointer, we will find the new leader via
1887 * next_thread().
1888 */
1889 rcu_read_lock();
1890 for_each_process(g) {
1891 if (g == tsk->group_leader)
1892 continue;
1893 if (g->flags & PF_KTHREAD)
1894 continue;
1895 p = g;
1896 do {
1897 if (p->mm) {
1898 if (unlikely(p->mm == mm)) {
1899 lock_task_sighand(p, &flags);
1900 nr += zap_process(p, exit_code);
1901 unlock_task_sighand(p, &flags);
1902 }
1903 break;
1904 }
1905 } while_each_thread(g, p);
1906 }
1907 rcu_read_unlock();
1908done:
1909 atomic_set(&core_state->nr_threads, nr);
1910 return nr;
1911}
1912
1913static int coredump_wait(int exit_code, struct core_state *core_state)
1914{
1915 struct task_struct *tsk = current;
1916 struct mm_struct *mm = tsk->mm;
1917 struct completion *vfork_done;
1918 int core_waiters = -EBUSY;
1919
1920 init_completion(&core_state->startup);
1921 core_state->dumper.task = tsk;
1922 core_state->dumper.next = NULL;
1923
1924 down_write(&mm->mmap_sem);
1925 if (!mm->core_state)
1926 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1927 up_write(&mm->mmap_sem);
1928
1929 if (unlikely(core_waiters < 0))
1930 goto fail;
1931
1932 /*
1933 * Make sure nobody is waiting for us to release the VM,
1934 * otherwise we can deadlock when we wait on each other
1935 */
1936 vfork_done = tsk->vfork_done;
1937 if (vfork_done) {
1938 tsk->vfork_done = NULL;
1939 complete(vfork_done);
1940 }
1941
1942 if (core_waiters)
1943 wait_for_completion(&core_state->startup);
1944fail:
1945 return core_waiters;
1946}
1947
1948static void coredump_finish(struct mm_struct *mm)
1949{
1950 struct core_thread *curr, *next;
1951 struct task_struct *task;
1952
1953 next = mm->core_state->dumper.next;
1954 while ((curr = next) != NULL) {
1955 next = curr->next;
1956 task = curr->task;
1957 /*
1958 * see exit_mm(), curr->task must not see
1959 * ->task == NULL before we read ->next.
1960 */
1961 smp_mb();
1962 curr->task = NULL;
1963 wake_up_process(task);
1964 }
1965
1966 mm->core_state = NULL;
1967}
1968
1969/*
1970 * set_dumpable converts traditional three-value dumpable to two flags and
1971 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1972 * these bits are not changed atomically. So get_dumpable can observe the
1973 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1974 * return either old dumpable or new one by paying attention to the order of
1975 * modifying the bits.
1976 *
1977 * dumpable | mm->flags (binary)
1978 * old new | initial interim final
1979 * ---------+-----------------------
1980 * 0 1 | 00 01 01
1981 * 0 2 | 00 10(*) 11
1982 * 1 0 | 01 00 00
1983 * 1 2 | 01 11 11
1984 * 2 0 | 11 10(*) 00
1985 * 2 1 | 11 11 01
1986 *
1987 * (*) get_dumpable regards interim value of 10 as 11.
1988 */
1989void set_dumpable(struct mm_struct *mm, int value)
1990{
1991 switch (value) {
1992 case 0:
1993 clear_bit(MMF_DUMPABLE, &mm->flags);
1994 smp_wmb();
1995 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1996 break;
1997 case 1:
1998 set_bit(MMF_DUMPABLE, &mm->flags);
1999 smp_wmb();
2000 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2001 break;
2002 case 2:
2003 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2004 smp_wmb();
2005 set_bit(MMF_DUMPABLE, &mm->flags);
2006 break;
2007 }
2008}
2009
2010static int __get_dumpable(unsigned long mm_flags)
2011{
2012 int ret;
2013
2014 ret = mm_flags & MMF_DUMPABLE_MASK;
2015 return (ret >= 2) ? 2 : ret;
2016}
2017
2018int get_dumpable(struct mm_struct *mm)
2019{
2020 return __get_dumpable(mm->flags);
2021}
2022
2023static void wait_for_dump_helpers(struct file *file)
2024{
2025 struct pipe_inode_info *pipe;
2026
2027 pipe = file->f_path.dentry->d_inode->i_pipe;
2028
2029 pipe_lock(pipe);
2030 pipe->readers++;
2031 pipe->writers--;
2032
2033 while ((pipe->readers > 1) && (!signal_pending(current))) {
2034 wake_up_interruptible_sync(&pipe->wait);
2035 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2036 pipe_wait(pipe);
2037 }
2038
2039 pipe->readers--;
2040 pipe->writers++;
2041 pipe_unlock(pipe);
2042
2043}
2044
2045
2046/*
2047 * umh_pipe_setup
2048 * helper function to customize the process used
2049 * to collect the core in userspace. Specifically
2050 * it sets up a pipe and installs it as fd 0 (stdin)
2051 * for the process. Returns 0 on success, or
2052 * PTR_ERR on failure.
2053 * Note that it also sets the core limit to 1. This
2054 * is a special value that we use to trap recursive
2055 * core dumps
2056 */
2057static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2058{
2059 struct file *rp, *wp;
2060 struct fdtable *fdt;
2061 struct coredump_params *cp = (struct coredump_params *)info->data;
2062 struct files_struct *cf = current->files;
2063
2064 wp = create_write_pipe(0);
2065 if (IS_ERR(wp))
2066 return PTR_ERR(wp);
2067
2068 rp = create_read_pipe(wp, 0);
2069 if (IS_ERR(rp)) {
2070 free_write_pipe(wp);
2071 return PTR_ERR(rp);
2072 }
2073
2074 cp->file = wp;
2075
2076 sys_close(0);
2077 fd_install(0, rp);
2078 spin_lock(&cf->file_lock);
2079 fdt = files_fdtable(cf);
2080 FD_SET(0, fdt->open_fds);
2081 FD_CLR(0, fdt->close_on_exec);
2082 spin_unlock(&cf->file_lock);
2083
2084 /* and disallow core files too */
2085 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2086
2087 return 0;
2088}
2089
2090void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2091{
2092 struct core_state core_state;
2093 struct core_name cn;
2094 struct mm_struct *mm = current->mm;
2095 struct linux_binfmt * binfmt;
2096 const struct cred *old_cred;
2097 struct cred *cred;
2098 int retval = 0;
2099 int flag = 0;
2100 int ispipe;
2101 static atomic_t core_dump_count = ATOMIC_INIT(0);
2102 struct coredump_params cprm = {
2103 .signr = signr,
2104 .regs = regs,
2105 .limit = rlimit(RLIMIT_CORE),
2106 /*
2107 * We must use the same mm->flags while dumping core to avoid
2108 * inconsistency of bit flags, since this flag is not protected
2109 * by any locks.
2110 */
2111 .mm_flags = mm->flags,
2112 };
2113
2114 audit_core_dumps(signr);
2115
2116 binfmt = mm->binfmt;
2117 if (!binfmt || !binfmt->core_dump)
2118 goto fail;
2119 if (!__get_dumpable(cprm.mm_flags))
2120 goto fail;
2121
2122 cred = prepare_creds();
2123 if (!cred)
2124 goto fail;
2125 /*
2126 * We cannot trust fsuid as being the "true" uid of the
2127 * process nor do we know its entire history. We only know it
2128 * was tainted so we dump it as root in mode 2.
2129 */
2130 if (__get_dumpable(cprm.mm_flags) == 2) {
2131 /* Setuid core dump mode */
2132 flag = O_EXCL; /* Stop rewrite attacks */
2133 cred->fsuid = 0; /* Dump root private */
2134 }
2135
2136 retval = coredump_wait(exit_code, &core_state);
2137 if (retval < 0)
2138 goto fail_creds;
2139
2140 old_cred = override_creds(cred);
2141
2142 /*
2143 * Clear any false indication of pending signals that might
2144 * be seen by the filesystem code called to write the core file.
2145 */
2146 clear_thread_flag(TIF_SIGPENDING);
2147
2148 ispipe = format_corename(&cn, signr);
2149
2150 if (ispipe) {
2151 int dump_count;
2152 char **helper_argv;
2153
2154 if (ispipe < 0) {
2155 printk(KERN_WARNING "format_corename failed\n");
2156 printk(KERN_WARNING "Aborting core\n");
2157 goto fail_corename;
2158 }
2159
2160 if (cprm.limit == 1) {
2161 /*
2162 * Normally core limits are irrelevant to pipes, since
2163 * we're not writing to the file system, but we use
2164 * cprm.limit of 1 here as a speacial value. Any
2165 * non-1 limit gets set to RLIM_INFINITY below, but
2166 * a limit of 0 skips the dump. This is a consistent
2167 * way to catch recursive crashes. We can still crash
2168 * if the core_pattern binary sets RLIM_CORE = !1
2169 * but it runs as root, and can do lots of stupid things
2170 * Note that we use task_tgid_vnr here to grab the pid
2171 * of the process group leader. That way we get the
2172 * right pid if a thread in a multi-threaded
2173 * core_pattern process dies.
2174 */
2175 printk(KERN_WARNING
2176 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2177 task_tgid_vnr(current), current->comm);
2178 printk(KERN_WARNING "Aborting core\n");
2179 goto fail_unlock;
2180 }
2181 cprm.limit = RLIM_INFINITY;
2182
2183 dump_count = atomic_inc_return(&core_dump_count);
2184 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2185 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2186 task_tgid_vnr(current), current->comm);
2187 printk(KERN_WARNING "Skipping core dump\n");
2188 goto fail_dropcount;
2189 }
2190
2191 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2192 if (!helper_argv) {
2193 printk(KERN_WARNING "%s failed to allocate memory\n",
2194 __func__);
2195 goto fail_dropcount;
2196 }
2197
2198 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2199 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2200 NULL, &cprm);
2201 argv_free(helper_argv);
2202 if (retval) {
2203 printk(KERN_INFO "Core dump to %s pipe failed\n",
2204 cn.corename);
2205 goto close_fail;
2206 }
2207 } else {
2208 struct inode *inode;
2209
2210 if (cprm.limit < binfmt->min_coredump)
2211 goto fail_unlock;
2212
2213 cprm.file = filp_open(cn.corename,
2214 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2215 0600);
2216 if (IS_ERR(cprm.file))
2217 goto fail_unlock;
2218
2219 inode = cprm.file->f_path.dentry->d_inode;
2220 if (inode->i_nlink > 1)
2221 goto close_fail;
2222 if (d_unhashed(cprm.file->f_path.dentry))
2223 goto close_fail;
2224 /*
2225 * AK: actually i see no reason to not allow this for named
2226 * pipes etc, but keep the previous behaviour for now.
2227 */
2228 if (!S_ISREG(inode->i_mode))
2229 goto close_fail;
2230 /*
2231 * Dont allow local users get cute and trick others to coredump
2232 * into their pre-created files.
2233 */
2234 if (inode->i_uid != current_fsuid())
2235 goto close_fail;
2236 if (!cprm.file->f_op || !cprm.file->f_op->write)
2237 goto close_fail;
2238 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2239 goto close_fail;
2240 }
2241
2242 retval = binfmt->core_dump(&cprm);
2243 if (retval)
2244 current->signal->group_exit_code |= 0x80;
2245
2246 if (ispipe && core_pipe_limit)
2247 wait_for_dump_helpers(cprm.file);
2248close_fail:
2249 if (cprm.file)
2250 filp_close(cprm.file, NULL);
2251fail_dropcount:
2252 if (ispipe)
2253 atomic_dec(&core_dump_count);
2254fail_unlock:
2255 kfree(cn.corename);
2256fail_corename:
2257 coredump_finish(mm);
2258 revert_creds(old_cred);
2259fail_creds:
2260 put_cred(cred);
2261fail:
2262 return;
2263}
2264
2265/*
2266 * Core dumping helper functions. These are the only things you should
2267 * do on a core-file: use only these functions to write out all the
2268 * necessary info.
2269 */
2270int dump_write(struct file *file, const void *addr, int nr)
2271{
2272 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2273}
2274EXPORT_SYMBOL(dump_write);
2275
2276int dump_seek(struct file *file, loff_t off)
2277{
2278 int ret = 1;
2279
2280 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2281 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2282 return 0;
2283 } else {
2284 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2285
2286 if (!buf)
2287 return 0;
2288 while (off > 0) {
2289 unsigned long n = off;
2290
2291 if (n > PAGE_SIZE)
2292 n = PAGE_SIZE;
2293 if (!dump_write(file, buf, n)) {
2294 ret = 0;
2295 break;
2296 }
2297 off -= n;
2298 }
2299 free_page((unsigned long)buf);
2300 }
2301 return ret;
2302}
2303EXPORT_SYMBOL(dump_seek);