projects/linux·blob master · refs

2124 lines · 52.9 KiB

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/futex.h>
34#include <linux/swap.h>
35#include <linux/string.h>
36#include <linux/init.h>
37#include <linux/sched/mm.h>
38#include <linux/sched/coredump.h>
39#include <linux/sched/exec_state.h>
40#include <linux/sched/signal.h>
41#include <linux/sched/numa_balancing.h>
42#include <linux/sched/task.h>
43#include <linux/pagemap.h>
44#include <linux/perf_event.h>
45#include <linux/highmem.h>
46#include <linux/spinlock.h>
47#include <linux/key.h>
48#include <linux/personality.h>
49#include <linux/binfmts.h>
50#include <linux/utsname.h>
51#include <linux/pid_namespace.h>
52#include <linux/module.h>
53#include <linux/namei.h>
54#include <linux/mount.h>
55#include <linux/security.h>
56#include <linux/syscalls.h>
57#include <linux/tsacct_kern.h>
58#include <linux/cn_proc.h>
59#include <linux/audit.h>
60#include <linux/kmod.h>
61#include <linux/fsnotify.h>
62#include <linux/fs_struct.h>
63#include <linux/oom.h>
64#include <linux/compat.h>
65#include <linux/vmalloc.h>
66#include <linux/io_uring.h>
67#include <linux/syscall_user_dispatch.h>
68#include <linux/coredump.h>
69#include <linux/time_namespace.h>
70#include <linux/user_events.h>
71#include <linux/rseq.h>
72#include <linux/ksm.h>
73
74#include <linux/uaccess.h>
75#include <asm/mmu_context.h>
76#include <asm/tlb.h>
77
78#include <trace/events/task.h>
79#include "internal.h"
80
81#include <trace/events/sched.h>
82
83/* For vma exec functions. */
84#include "../mm/internal.h"
85
86static int bprm_creds_from_file(struct linux_binprm *bprm);
87
88int suid_dumpable = 0;
89
90static LIST_HEAD(formats);
91static DEFINE_RWLOCK(binfmt_lock);
92
93void __register_binfmt(struct linux_binfmt * fmt, int insert)
94{
95 write_lock(&binfmt_lock);
96 insert ? list_add(&fmt->lh, &formats) :
97 list_add_tail(&fmt->lh, &formats);
98 write_unlock(&binfmt_lock);
99}
100
101EXPORT_SYMBOL(__register_binfmt);
102
103void unregister_binfmt(struct linux_binfmt * fmt)
104{
105 write_lock(&binfmt_lock);
106 list_del(&fmt->lh);
107 write_unlock(&binfmt_lock);
108}
109
110EXPORT_SYMBOL(unregister_binfmt);
111
112static inline void put_binfmt(struct linux_binfmt * fmt)
113{
114 module_put(fmt->module);
115}
116
117bool path_noexec(const struct path *path)
118{
119 /* If it's an anonymous inode make sure that we catch any shenanigans. */
120 VFS_WARN_ON_ONCE(IS_ANON_FILE(d_inode(path->dentry)) &&
121 !(path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC));
122 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
123 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
124}
125
126#ifdef CONFIG_MMU
127/*
128 * The nascent bprm->mm is not visible until exec_mmap() but it can
129 * use a lot of memory, account these pages in current->mm temporary
130 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
131 * change the counter back via acct_arg_size(0).
132 */
133static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
134{
135 struct mm_struct *mm = current->mm;
136 long diff = (long)(pages - bprm->vma_pages);
137
138 if (!mm || !diff)
139 return;
140
141 bprm->vma_pages = pages;
142 add_mm_counter(mm, MM_ANONPAGES, diff);
143}
144
145static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
146 int write)
147{
148 struct page *page;
149 struct vm_area_struct *vma = bprm->vma;
150 struct mm_struct *mm = bprm->mm;
151 int ret;
152
153 /*
154 * Avoid relying on expanding the stack down in GUP (which
155 * does not work for STACK_GROWSUP anyway), and just do it
156 * ahead of time.
157 */
158 if (!mmap_read_lock_maybe_expand(mm, vma, pos, write))
159 return NULL;
160
161 /*
162 * We are doing an exec(). 'current' is the process
163 * doing the exec and 'mm' is the new process's mm.
164 */
165 ret = get_user_pages_remote(mm, pos, 1,
166 write ? FOLL_WRITE : 0,
167 &page, NULL);
168 mmap_read_unlock(mm);
169 if (ret <= 0)
170 return NULL;
171
172 if (write)
173 acct_arg_size(bprm, vma_pages(vma));
174
175 return page;
176}
177
178static void put_arg_page(struct page *page)
179{
180 put_page(page);
181}
182
183static void free_arg_pages(struct linux_binprm *bprm)
184{
185}
186
187static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
188 struct page *page)
189{
190 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
191}
192
193static bool valid_arg_len(struct linux_binprm *bprm, long len)
194{
195 return len <= MAX_ARG_STRLEN;
196}
197
198#else
199
200static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
201{
202}
203
204static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
205 int write)
206{
207 struct page *page;
208
209 page = bprm->page[pos / PAGE_SIZE];
210 if (!page && write) {
211 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
212 if (!page)
213 return NULL;
214 bprm->page[pos / PAGE_SIZE] = page;
215 }
216
217 return page;
218}
219
220static void put_arg_page(struct page *page)
221{
222}
223
224static void free_arg_page(struct linux_binprm *bprm, int i)
225{
226 if (bprm->page[i]) {
227 __free_page(bprm->page[i]);
228 bprm->page[i] = NULL;
229 }
230}
231
232static void free_arg_pages(struct linux_binprm *bprm)
233{
234 int i;
235
236 for (i = 0; i < MAX_ARG_PAGES; i++)
237 free_arg_page(bprm, i);
238}
239
240static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
241 struct page *page)
242{
243}
244
245static bool valid_arg_len(struct linux_binprm *bprm, long len)
246{
247 return len <= bprm->p;
248}
249
250#endif /* CONFIG_MMU */
251
252/*
253 * Create a new mm_struct and populate it with a temporary stack
254 * vm_area_struct. We don't have enough context at this point to set the stack
255 * flags, permissions, and offset, so we use temporary values. We'll update
256 * them later in setup_arg_pages().
257 */
258static int bprm_mm_init(struct linux_binprm *bprm)
259{
260 int err;
261 struct mm_struct *mm = NULL;
262
263 bprm->mm = mm = mm_alloc();
264 err = -ENOMEM;
265 if (!mm)
266 goto err;
267
268 /* Staged for would_dump() narrowing; consumed by begin_new_exec(). */
269 bprm->user_ns = get_user_ns(current_user_ns());
270
271 /* Save current stack limit for all calculations made during exec. */
272 task_lock(current->group_leader);
273 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
274 task_unlock(current->group_leader);
275
276#ifndef CONFIG_MMU
277 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
278#else
279 err = create_init_stack_vma(bprm->mm, &bprm->vma, &bprm->p);
280 if (err)
281 goto err;
282#endif
283
284 return 0;
285
286err:
287 if (mm) {
288 bprm->mm = NULL;
289 mmdrop(mm);
290 }
291
292 return err;
293}
294
295struct user_arg_ptr {
296#ifdef CONFIG_COMPAT
297 bool is_compat;
298#endif
299 union {
300 const char __user *const __user *native;
301#ifdef CONFIG_COMPAT
302 const compat_uptr_t __user *compat;
303#endif
304 } ptr;
305};
306
307static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
308{
309 const char __user *native;
310
311#ifdef CONFIG_COMPAT
312 if (unlikely(argv.is_compat)) {
313 compat_uptr_t compat;
314
315 if (get_user(compat, argv.ptr.compat + nr))
316 return ERR_PTR(-EFAULT);
317
318 return compat_ptr(compat);
319 }
320#endif
321
322 if (get_user(native, argv.ptr.native + nr))
323 return ERR_PTR(-EFAULT);
324
325 return native;
326}
327
328/*
329 * count() counts the number of strings in array ARGV.
330 */
331static int count(struct user_arg_ptr argv, int max)
332{
333 int i = 0;
334
335 if (argv.ptr.native != NULL) {
336 for (;;) {
337 const char __user *p = get_user_arg_ptr(argv, i);
338
339 if (!p)
340 break;
341
342 if (IS_ERR(p))
343 return -EFAULT;
344
345 if (i >= max)
346 return -E2BIG;
347 ++i;
348
349 if (fatal_signal_pending(current))
350 return -ERESTARTNOHAND;
351 cond_resched();
352 }
353 }
354 return i;
355}
356
357static int count_strings_kernel(const char *const *argv)
358{
359 int i;
360
361 if (!argv)
362 return 0;
363
364 for (i = 0; argv[i]; ++i) {
365 if (i >= MAX_ARG_STRINGS)
366 return -E2BIG;
367 if (fatal_signal_pending(current))
368 return -ERESTARTNOHAND;
369 cond_resched();
370 }
371 return i;
372}
373
374static inline int bprm_set_stack_limit(struct linux_binprm *bprm,
375 unsigned long limit)
376{
377#ifdef CONFIG_MMU
378 /* Avoid a pathological bprm->p. */
379 if (bprm->p < limit)
380 return -E2BIG;
381 bprm->argmin = bprm->p - limit;
382#endif
383 return 0;
384}
385static inline bool bprm_hit_stack_limit(struct linux_binprm *bprm)
386{
387#ifdef CONFIG_MMU
388 return bprm->p < bprm->argmin;
389#else
390 return false;
391#endif
392}
393
394/*
395 * Calculate bprm->argmin from:
396 * - _STK_LIM
397 * - ARG_MAX
398 * - bprm->rlim_stack.rlim_cur
399 * - bprm->argc
400 * - bprm->envc
401 * - bprm->p
402 */
403static int bprm_stack_limits(struct linux_binprm *bprm)
404{
405 unsigned long limit, ptr_size;
406
407 /*
408 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
409 * (whichever is smaller) for the argv+env strings.
410 * This ensures that:
411 * - the remaining binfmt code will not run out of stack space,
412 * - the program will have a reasonable amount of stack left
413 * to work from.
414 */
415 limit = _STK_LIM / 4 * 3;
416 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
417 /*
418 * We've historically supported up to 32 pages (ARG_MAX)
419 * of argument strings even with small stacks
420 */
421 limit = max_t(unsigned long, limit, ARG_MAX);
422 /* Reject totally pathological counts. */
423 if (bprm->argc < 0 || bprm->envc < 0)
424 return -E2BIG;
425 /*
426 * We must account for the size of all the argv and envp pointers to
427 * the argv and envp strings, since they will also take up space in
428 * the stack. They aren't stored until much later when we can't
429 * signal to the parent that the child has run out of stack space.
430 * Instead, calculate it here so it's possible to fail gracefully.
431 *
432 * In the case of argc = 0, make sure there is space for adding a
433 * empty string (which will bump argc to 1), to ensure confused
434 * userspace programs don't start processing from argv[1], thinking
435 * argc can never be 0, to keep them from walking envp by accident.
436 * See do_execveat_common().
437 */
438 if (check_add_overflow(max(bprm->argc, 1), bprm->envc, &ptr_size) ||
439 check_mul_overflow(ptr_size, sizeof(void *), &ptr_size))
440 return -E2BIG;
441 if (limit <= ptr_size)
442 return -E2BIG;
443 limit -= ptr_size;
444
445 return bprm_set_stack_limit(bprm, limit);
446}
447
448/*
449 * 'copy_strings()' copies argument/environment strings from the old
450 * processes's memory to the new process's stack. The call to get_user_pages()
451 * ensures the destination page is created and not swapped out.
452 */
453static int copy_strings(int argc, struct user_arg_ptr argv,
454 struct linux_binprm *bprm)
455{
456 struct page *kmapped_page = NULL;
457 char *kaddr = NULL;
458 unsigned long kpos = 0;
459 int ret;
460
461 while (argc-- > 0) {
462 const char __user *str;
463 int len;
464 unsigned long pos;
465
466 ret = -EFAULT;
467 str = get_user_arg_ptr(argv, argc);
468 if (IS_ERR(str))
469 goto out;
470
471 len = strnlen_user(str, MAX_ARG_STRLEN);
472 if (!len)
473 goto out;
474
475 ret = -E2BIG;
476 if (!valid_arg_len(bprm, len))
477 goto out;
478
479 /* We're going to work our way backwards. */
480 pos = bprm->p;
481 str += len;
482 bprm->p -= len;
483 if (bprm_hit_stack_limit(bprm))
484 goto out;
485
486 while (len > 0) {
487 int offset, bytes_to_copy;
488
489 if (fatal_signal_pending(current)) {
490 ret = -ERESTARTNOHAND;
491 goto out;
492 }
493 cond_resched();
494
495 offset = pos % PAGE_SIZE;
496 if (offset == 0)
497 offset = PAGE_SIZE;
498
499 bytes_to_copy = offset;
500 if (bytes_to_copy > len)
501 bytes_to_copy = len;
502
503 offset -= bytes_to_copy;
504 pos -= bytes_to_copy;
505 str -= bytes_to_copy;
506 len -= bytes_to_copy;
507
508 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
509 struct page *page;
510
511 page = get_arg_page(bprm, pos, 1);
512 if (!page) {
513 ret = -E2BIG;
514 goto out;
515 }
516
517 if (kmapped_page) {
518 flush_dcache_page(kmapped_page);
519 kunmap_local(kaddr);
520 put_arg_page(kmapped_page);
521 }
522 kmapped_page = page;
523 kaddr = kmap_local_page(kmapped_page);
524 kpos = pos & PAGE_MASK;
525 flush_arg_page(bprm, kpos, kmapped_page);
526 }
527 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
528 ret = -EFAULT;
529 goto out;
530 }
531 }
532 }
533 ret = 0;
534out:
535 if (kmapped_page) {
536 flush_dcache_page(kmapped_page);
537 kunmap_local(kaddr);
538 put_arg_page(kmapped_page);
539 }
540 return ret;
541}
542
543/*
544 * Copy and argument/environment string from the kernel to the processes stack.
545 */
546int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
547{
548 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
549 unsigned long pos = bprm->p;
550
551 if (len == 0)
552 return -EFAULT;
553 if (!valid_arg_len(bprm, len))
554 return -E2BIG;
555
556 /* We're going to work our way backwards. */
557 arg += len;
558 bprm->p -= len;
559 if (bprm_hit_stack_limit(bprm))
560 return -E2BIG;
561
562 while (len > 0) {
563 unsigned int bytes_to_copy = min(len,
564 min_not_zero(offset_in_page(pos), PAGE_SIZE));
565 struct page *page;
566
567 pos -= bytes_to_copy;
568 arg -= bytes_to_copy;
569 len -= bytes_to_copy;
570
571 page = get_arg_page(bprm, pos, 1);
572 if (!page)
573 return -E2BIG;
574 flush_arg_page(bprm, pos & PAGE_MASK, page);
575 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
576 put_arg_page(page);
577 }
578
579 return 0;
580}
581EXPORT_SYMBOL(copy_string_kernel);
582
583static int copy_strings_kernel(int argc, const char *const *argv,
584 struct linux_binprm *bprm)
585{
586 while (argc-- > 0) {
587 int ret = copy_string_kernel(argv[argc], bprm);
588 if (ret < 0)
589 return ret;
590 if (fatal_signal_pending(current))
591 return -ERESTARTNOHAND;
592 cond_resched();
593 }
594 return 0;
595}
596
597#ifdef CONFIG_MMU
598
599/*
600 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
601 * the stack is optionally relocated, and some extra space is added.
602 */
603int setup_arg_pages(struct linux_binprm *bprm,
604 unsigned long stack_top,
605 int executable_stack)
606{
607 int ret;
608 unsigned long stack_shift;
609 struct mm_struct *mm = current->mm;
610 struct vm_area_struct *vma = bprm->vma;
611 struct vm_area_struct *prev = NULL;
612 vm_flags_t vm_flags;
613 unsigned long stack_base;
614 unsigned long stack_size;
615 unsigned long stack_expand;
616 unsigned long rlim_stack;
617 struct mmu_gather tlb;
618 struct vma_iterator vmi;
619
620#ifdef CONFIG_STACK_GROWSUP
621 /* Limit stack size */
622 stack_base = bprm->rlim_stack.rlim_max;
623
624 stack_base = calc_max_stack_size(stack_base);
625
626 /* Add space for stack randomization. */
627 if (current->flags & PF_RANDOMIZE)
628 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
629
630 /* Make sure we didn't let the argument array grow too large. */
631 if (vma->vm_end - vma->vm_start > stack_base)
632 return -ENOMEM;
633
634 stack_base = PAGE_ALIGN(stack_top - stack_base);
635
636 stack_shift = vma->vm_start - stack_base;
637 mm->arg_start = bprm->p - stack_shift;
638 bprm->p = vma->vm_end - stack_shift;
639#else
640 stack_top = arch_align_stack(stack_top);
641 stack_top = PAGE_ALIGN(stack_top);
642
643 if (unlikely(stack_top < mmap_min_addr) ||
644 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
645 return -ENOMEM;
646
647 stack_shift = vma->vm_end - stack_top;
648
649 bprm->p -= stack_shift;
650 mm->arg_start = bprm->p;
651#endif
652
653 bprm->exec -= stack_shift;
654
655 if (mmap_write_lock_killable(mm))
656 return -EINTR;
657
658 vm_flags = VM_STACK_FLAGS;
659
660 /*
661 * Adjust stack execute permissions; explicitly enable for
662 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
663 * (arch default) otherwise.
664 */
665 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
666 vm_flags |= VM_EXEC;
667 else if (executable_stack == EXSTACK_DISABLE_X)
668 vm_flags &= ~VM_EXEC;
669 vm_flags |= mm->def_flags;
670 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
671
672 vma_iter_init(&vmi, mm, vma->vm_start);
673
674 tlb_gather_mmu(&tlb, mm);
675 ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
676 vm_flags);
677 tlb_finish_mmu(&tlb);
678
679 if (ret)
680 goto out_unlock;
681 BUG_ON(prev != vma);
682
683 if (unlikely(vm_flags & VM_EXEC)) {
684 pr_warn_once("process '%pD4' started with executable stack\n",
685 bprm->file);
686 }
687
688 /* Move stack pages down in memory. */
689 if (stack_shift) {
690 /*
691 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
692 * the binfmt code determines where the new stack should reside, we shift it to
693 * its final location.
694 */
695 ret = relocate_vma_down(vma, stack_shift);
696 if (ret)
697 goto out_unlock;
698 }
699
700 /* mprotect_fixup is overkill to remove the temporary stack flags */
701 vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
702
703 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
704 stack_size = vma->vm_end - vma->vm_start;
705 /*
706 * Align this down to a page boundary as expand_stack
707 * will align it up.
708 */
709 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
710
711 stack_expand = min(rlim_stack, stack_size + stack_expand);
712
713#ifdef CONFIG_STACK_GROWSUP
714 stack_base = vma->vm_start + stack_expand;
715#else
716 stack_base = vma->vm_end - stack_expand;
717#endif
718 current->mm->start_stack = bprm->p;
719 ret = expand_stack_locked(vma, stack_base);
720 if (ret)
721 ret = -EFAULT;
722
723out_unlock:
724 mmap_write_unlock(mm);
725 return ret;
726}
727EXPORT_SYMBOL(setup_arg_pages);
728
729#else
730
731/*
732 * Transfer the program arguments and environment from the holding pages
733 * onto the stack. The provided stack pointer is adjusted accordingly.
734 */
735int transfer_args_to_stack(struct linux_binprm *bprm,
736 unsigned long *sp_location)
737{
738 unsigned long index, stop, sp;
739 int ret = 0;
740
741 stop = bprm->p >> PAGE_SHIFT;
742 sp = *sp_location;
743
744 for (index = MAX_ARG_PAGES; index-- > stop; ) {
745 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
746 char *src = kmap_local_page(bprm->page[index]) + offset;
747 sp -= PAGE_SIZE - offset;
748 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
749 ret = -EFAULT;
750 kunmap_local(src);
751 if (ret)
752 goto out;
753 }
754
755 bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
756 *sp_location = sp;
757
758out:
759 return ret;
760}
761EXPORT_SYMBOL(transfer_args_to_stack);
762
763#endif /* CONFIG_MMU */
764
765/*
766 * On success, caller must call do_close_execat() on the returned
767 * struct file to close it.
768 */
769static struct file *do_open_execat(int fd, struct filename *name, int flags)
770{
771 int err;
772 struct file *file __free(fput) = NULL;
773 struct open_flags open_exec_flags = {
774 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
775 .acc_mode = MAY_EXEC,
776 .intent = LOOKUP_OPEN,
777 .lookup_flags = LOOKUP_FOLLOW,
778 };
779
780 if ((flags &
781 ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH | AT_EXECVE_CHECK)) != 0)
782 return ERR_PTR(-EINVAL);
783 if (flags & AT_SYMLINK_NOFOLLOW)
784 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
785
786 file = do_file_open(fd, name, &open_exec_flags);
787 if (IS_ERR(file))
788 return file;
789
790 if (path_noexec(&file->f_path))
791 return ERR_PTR(-EACCES);
792
793 /*
794 * In the past the regular type check was here. It moved to may_open() in
795 * 633fb6ac3980 ("exec: move S_ISREG() check earlier"). Since then it is
796 * an invariant that all non-regular files error out before we get here.
797 */
798 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode)))
799 return ERR_PTR(-EACCES);
800
801 err = exe_file_deny_write_access(file);
802 if (err)
803 return ERR_PTR(err);
804
805 return no_free_ptr(file);
806}
807
808/**
809 * open_exec - Open a path name for execution
810 *
811 * @name: path name to open with the intent of executing it.
812 *
813 * Returns ERR_PTR on failure or allocated struct file on success.
814 *
815 * As this is a wrapper for the internal do_open_execat(), callers
816 * must call exe_file_allow_write_access() before fput() on release. Also see
817 * do_close_execat().
818 */
819struct file *open_exec(const char *name)
820{
821 CLASS(filename_kernel, filename)(name);
822 return do_open_execat(AT_FDCWD, filename, 0);
823}
824EXPORT_SYMBOL(open_exec);
825
826#if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
827ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
828{
829 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
830 if (res > 0)
831 flush_icache_user_range(addr, addr + len);
832 return res;
833}
834EXPORT_SYMBOL(read_code);
835#endif
836
837/*
838 * Maps the mm_struct mm into the current task struct.
839 * On success, this function returns with exec_update_lock
840 * held for writing. The replaced address space is stashed in
841 * bprm->old_mm for setup_new_exec() to release outside the lock.
842 */
843static int exec_mmap(struct linux_binprm *bprm)
844{
845 struct task_exec_state *exec_state __free(put_task_exec_state) = NULL;
846 struct mm_struct *mm = bprm->mm;
847 struct task_struct *tsk;
848 struct mm_struct *old_mm, *active_mm;
849 int ret;
850
851 exec_state = alloc_task_exec_state(bprm->user_ns);
852 if (!exec_state)
853 return -ENOMEM;
854
855 /* Notify parent that we're no longer interested in the old VM */
856 tsk = current;
857 old_mm = current->mm;
858 /* Clean up futexes and release the mm */
859 mm_exit_exec_release(tsk, old_mm);
860
861 ret = down_write_killable(&tsk->signal->exec_update_lock);
862 if (ret)
863 return ret;
864
865 if (old_mm) {
866 /*
867 * If there is a pending fatal signal perhaps a signal
868 * whose default action is to create a coredump get
869 * out and die instead of going through with the exec.
870 */
871 ret = mmap_read_lock_killable(old_mm);
872 if (ret) {
873 up_write(&tsk->signal->exec_update_lock);
874 return ret;
875 }
876 }
877
878 task_lock(tsk);
879 membarrier_exec_mmap(mm);
880
881 local_irq_disable();
882 active_mm = tsk->active_mm;
883 tsk->active_mm = mm;
884 tsk->mm = mm;
885 sched_cache_exec_mmap(tsk, mm);
886 mm_init_cid(mm, tsk);
887 exec_state = task_exec_state_replace(tsk, exec_state);
888 /*
889 * This prevents preemption while active_mm is being loaded and
890 * it and mm are being updated, which could cause problems for
891 * lazy tlb mm refcounting when these are updated by context
892 * switches. Not all architectures can handle irqs off over
893 * activate_mm yet.
894 */
895 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
896 local_irq_enable();
897 activate_mm(active_mm, mm);
898 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
899 local_irq_enable();
900 lru_gen_add_mm(mm);
901 task_unlock(tsk);
902 lru_gen_use_mm(mm);
903 if (old_mm) {
904 mmap_read_unlock(old_mm);
905 BUG_ON(active_mm != old_mm);
906 /* Defer teardown to setup_new_exec(), outside the exec locks. */
907 bprm->old_mm = old_mm;
908 } else {
909 mmdrop_lazy_tlb(active_mm);
910 }
911 futex_exec_done(tsk);
912 return 0;
913}
914
915/* Release the address space replaced by exec, outside the exec locks. */
916static void exec_mm_put_old(struct mm_struct *old_mm)
917{
918 setmax_mm_hiwater_rss(&current->signal->maxrss, old_mm);
919 mm_update_next_owner(old_mm);
920 mmput(old_mm);
921}
922
923static int de_thread(struct task_struct *tsk)
924{
925 struct signal_struct *sig = tsk->signal;
926 struct sighand_struct *oldsighand = tsk->sighand;
927 spinlock_t *lock = &oldsighand->siglock;
928
929 if (thread_group_empty(tsk))
930 goto no_thread_group;
931
932 /*
933 * Kill all other threads in the thread group.
934 */
935 spin_lock_irq(lock);
936 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
937 /*
938 * Another group action in progress, just
939 * return so that the signal is processed.
940 */
941 spin_unlock_irq(lock);
942 return -EAGAIN;
943 }
944
945 sig->group_exec_task = tsk;
946 sig->notify_count = zap_other_threads(tsk);
947 if (!thread_group_leader(tsk))
948 sig->notify_count--;
949
950 while (sig->notify_count) {
951 __set_current_state(TASK_KILLABLE);
952 spin_unlock_irq(lock);
953 schedule();
954 if (__fatal_signal_pending(tsk))
955 goto killed;
956 spin_lock_irq(lock);
957 }
958 spin_unlock_irq(lock);
959
960 /*
961 * At this point all other threads have exited, all we have to
962 * do is to wait for the thread group leader to become inactive,
963 * and to assume its PID:
964 */
965 if (!thread_group_leader(tsk)) {
966 struct task_struct *leader = tsk->group_leader;
967
968 for (;;) {
969 cgroup_threadgroup_change_begin(tsk);
970 write_lock_irq(&tasklist_lock);
971 /*
972 * Do this under tasklist_lock to ensure that
973 * exit_notify() can't miss ->group_exec_task
974 */
975 sig->notify_count = -1;
976 if (likely(leader->exit_state))
977 break;
978 __set_current_state(TASK_KILLABLE);
979 write_unlock_irq(&tasklist_lock);
980 cgroup_threadgroup_change_end(tsk);
981 schedule();
982 if (__fatal_signal_pending(tsk))
983 goto killed;
984 }
985
986 /*
987 * The only record we have of the real-time age of a
988 * process, regardless of execs it's done, is start_time.
989 * All the past CPU time is accumulated in signal_struct
990 * from sister threads now dead. But in this non-leader
991 * exec, nothing survives from the original leader thread,
992 * whose birth marks the true age of this process now.
993 * When we take on its identity by switching to its PID, we
994 * also take its birthdate (always earlier than our own).
995 */
996 tsk->start_time = leader->start_time;
997 tsk->start_boottime = leader->start_boottime;
998
999 BUG_ON(!same_thread_group(leader, tsk));
1000 /*
1001 * An exec() starts a new thread group with the
1002 * TGID of the previous thread group. Rehash the
1003 * two threads with a switched PID, and release
1004 * the former thread group leader:
1005 */
1006
1007 /* Become a process group leader with the old leader's pid.
1008 * The old leader becomes a thread of the this thread group.
1009 */
1010 exchange_tids(tsk, leader);
1011 transfer_pid(leader, tsk, PIDTYPE_TGID);
1012 transfer_pid(leader, tsk, PIDTYPE_PGID);
1013 transfer_pid(leader, tsk, PIDTYPE_SID);
1014
1015 list_replace_rcu(&leader->tasks, &tsk->tasks);
1016 list_replace_init(&leader->sibling, &tsk->sibling);
1017
1018 tsk->group_leader = tsk;
1019 leader->group_leader = tsk;
1020
1021 tsk->exit_signal = SIGCHLD;
1022 leader->exit_signal = -1;
1023
1024 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1025 leader->exit_state = EXIT_DEAD;
1026 /*
1027 * We are going to release_task()->ptrace_unlink() silently,
1028 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1029 * the tracer won't block again waiting for this thread.
1030 */
1031 if (unlikely(leader->ptrace))
1032 __wake_up_parent(leader, leader->parent);
1033 write_unlock_irq(&tasklist_lock);
1034 cgroup_threadgroup_change_end(tsk);
1035
1036 release_task(leader);
1037 }
1038
1039 sig->group_exec_task = NULL;
1040 sig->notify_count = 0;
1041
1042no_thread_group:
1043 /* we have changed execution domain */
1044 tsk->exit_signal = SIGCHLD;
1045
1046 BUG_ON(!thread_group_leader(tsk));
1047 return 0;
1048
1049killed:
1050 /* protects against exit_notify() and __exit_signal() */
1051 read_lock(&tasklist_lock);
1052 sig->group_exec_task = NULL;
1053 sig->notify_count = 0;
1054 read_unlock(&tasklist_lock);
1055 return -EAGAIN;
1056}
1057
1058
1059/*
1060 * This function makes sure the current process has its own signal table,
1061 * so that flush_signal_handlers can later reset the handlers without
1062 * disturbing other processes. (Other processes might share the signal
1063 * table via the CLONE_SIGHAND option to clone().)
1064 */
1065static int unshare_sighand(struct task_struct *me)
1066{
1067 struct sighand_struct *oldsighand = me->sighand;
1068
1069 if (refcount_read(&oldsighand->count) != 1) {
1070 struct sighand_struct *newsighand;
1071 /*
1072 * This ->sighand is shared with the CLONE_SIGHAND
1073 * but not CLONE_THREAD task, switch to the new one.
1074 */
1075 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1076 if (!newsighand)
1077 return -ENOMEM;
1078
1079 refcount_set(&newsighand->count, 1);
1080
1081 write_lock_irq(&tasklist_lock);
1082 spin_lock(&oldsighand->siglock);
1083 memcpy(newsighand->action, oldsighand->action,
1084 sizeof(newsighand->action));
1085 rcu_assign_pointer(me->sighand, newsighand);
1086 spin_unlock(&oldsighand->siglock);
1087 write_unlock_irq(&tasklist_lock);
1088
1089 __cleanup_sighand(oldsighand);
1090 }
1091 return 0;
1092}
1093
1094/*
1095 * This is unlocked -- the string will always be NUL-terminated, but
1096 * may show overlapping contents if racing concurrent reads.
1097 */
1098void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1099{
1100 size_t len = strnlen(buf, sizeof(tsk->comm) - 1);
1101
1102 trace_task_rename(tsk, buf);
1103 memcpy(tsk->comm, buf, len);
1104 memset(&tsk->comm[len], 0, sizeof(tsk->comm) - len);
1105 perf_event_comm(tsk, exec);
1106}
1107
1108/*
1109 * The file the process presents as: its exe link and comm. A transparent
1110 * dispatch presents as the binary, which is bprm->executable.
1111 */
1112static struct file *bprm_identity_file(const struct linux_binprm *bprm)
1113{
1114 if (bprm->interp_flags & BINPRM_FLAGS_TRANSPARENT_INTERP)
1115 return bprm->executable;
1116 return bprm->file;
1117}
1118
1119static void posixtimer_exec(struct task_struct *me)
1120{
1121#ifdef CONFIG_POSIX_TIMERS
1122 spin_lock_irq(&me->sighand->siglock);
1123 posix_cpu_timers_exit(me);
1124 spin_unlock_irq(&me->sighand->siglock);
1125 exit_itimers(me);
1126 flush_itimer_signals();
1127#endif
1128}
1129
1130/*
1131 * Calling this is the point of no return. None of the failures will be
1132 * seen by userspace since either the process is already taking a fatal
1133 * signal (via de_thread() or coredump), or will have SEGV raised
1134 * (after exec_mmap()) by search_binary_handler (see below).
1135 */
1136int begin_new_exec(struct linux_binprm * bprm)
1137{
1138 struct task_struct *me = current;
1139 struct files_struct *files = NULL;
1140 int retval;
1141
1142 /* A pending PT_INTERP substitution this format cannot consume. */
1143 if (bprm->loader)
1144 return -ENOEXEC;
1145
1146 /* Once we are committed compute the creds */
1147 retval = bprm_creds_from_file(bprm);
1148 if (retval)
1149 return retval;
1150
1151 /*
1152 * This tracepoint marks the point before flushing the old exec where
1153 * the current task is still unchanged, but errors are fatal (point of
1154 * no return). The later "sched_process_exec" tracepoint is called after
1155 * the current task has successfully switched to the new exec.
1156 */
1157 trace_sched_prepare_exec(current, bprm);
1158
1159 /*
1160 * Ensure all future errors are fatal.
1161 */
1162 bprm->point_of_no_return = true;
1163
1164 /*
1165 * Cancel any io_uring activity across execve. This runs task work
1166 * that may still create an io-wq worker, so do it while de_thread()
1167 * can still zap it.
1168 */
1169 io_uring_task_cancel();
1170
1171 /* Make this the only thread in the thread group */
1172 retval = de_thread(me);
1173 if (retval)
1174 goto out;
1175
1176 /*
1177 * This must be done here to ensure that POSIX CPU timers which were
1178 * armed on the current task are dequeued from me::posix_cputimers.
1179 * Otherwise in case of a TID switch the deletion of the related POSIX
1180 * timer would not remove an enqueued timer because the TID lookup
1181 * of the old TID fails.
1182 */
1183 posixtimer_exec(me);
1184
1185 /* see the comment in check_unsafe_exec() */
1186 current->fs->in_exec = 0;
1187
1188 /* Ensure the files table is not shared. */
1189 retval = unshare_fd(CLONE_FILES, &files);
1190 if (retval)
1191 goto out;
1192 if (files)
1193 switch_files_struct(me, files);
1194
1195 /*
1196 * We have to apply CLOEXEC before we change whether the process is
1197 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1198 * trying to access the should-be-closed file descriptors of a process
1199 * undergoing exec(2).
1200 *
1201 * This can block on filesystem ->flush() and ->release() handlers,
1202 * including waiting for FUSE daemons, so do it before exec_mmap
1203 * takes the exec_update_lock.
1204 * This must happen after the point of no return, and after unsharing
1205 * the FD table.
1206 */
1207 close_cloexec_files(me->files);
1208
1209 /*
1210 * Must be called _before_ exec_mmap() as bprm->mm is
1211 * not visible until then. Doing it here also ensures
1212 * we don't race against replace_mm_exe_file().
1213 */
1214 retval = set_mm_exe_file(bprm->mm, bprm_identity_file(bprm));
1215 if (retval)
1216 goto out;
1217
1218 /* If the binary is not readable then enforce mm->dumpable=0 */
1219 would_dump(bprm, bprm->file);
1220 if (bprm->have_execfd)
1221 would_dump(bprm, bprm->executable);
1222
1223 /*
1224 * Release all of the old mmap stuff
1225 */
1226 acct_arg_size(bprm, 0);
1227 retval = exec_mmap(bprm);
1228 if (retval)
1229 goto out;
1230
1231 bprm->mm = NULL;
1232
1233 retval = exec_task_namespaces();
1234 if (retval)
1235 goto out_unlock;
1236
1237 /*
1238 * Make the signal table private.
1239 */
1240 retval = unshare_sighand(me);
1241 if (retval)
1242 goto out_unlock;
1243
1244 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1245 PF_NOFREEZE | PF_NO_SETAFFINITY);
1246 flush_thread();
1247 me->personality &= ~bprm->per_clear;
1248
1249 clear_syscall_work_syscall_user_dispatch(me);
1250
1251 if (bprm->secureexec) {
1252 /* Make sure parent cannot signal privileged process. */
1253 me->pdeath_signal = 0;
1254
1255 /*
1256 * For secureexec, reset the stack limit to sane default to
1257 * avoid bad behavior from the prior rlimits. This has to
1258 * happen before arch_pick_mmap_layout(), which examines
1259 * RLIMIT_STACK, but after the point of no return to avoid
1260 * needing to clean up the change on failure.
1261 */
1262 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1263 bprm->rlim_stack.rlim_cur = _STK_LIM;
1264 }
1265
1266 me->sas_ss_sp = me->sas_ss_size = 0;
1267
1268 /*
1269 * Figure out dumpability. Note that this checking only of current
1270 * is wrong, but userspace depends on it. This should be testing
1271 * bprm->secureexec instead.
1272 */
1273 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1274 !(uid_eq(current_euid(), current_uid()) &&
1275 gid_eq(current_egid(), current_gid())))
1276 task_exec_state_set_dumpable(suid_dumpable);
1277 else
1278 task_exec_state_set_dumpable(TASK_DUMPABLE_OWNER);
1279
1280 perf_event_exec();
1281
1282 /*
1283 * If the original filename was empty, alloc_bprm() made up a path
1284 * that will probably not be useful to admins running ps or similar.
1285 * Let's fix it up to be something reasonable.
1286 */
1287 if (bprm->comm_from_dentry) {
1288 struct file *comm_file = bprm_identity_file(bprm);
1289
1290 /*
1291 * Hold RCU lock to keep the name from being freed behind our back.
1292 * Use acquire semantics to make sure the terminating NUL from
1293 * __d_alloc() is seen.
1294 *
1295 * Note, we're deliberately sloppy here. We don't need to care about
1296 * detecting a concurrent rename and just want a terminated name.
1297 */
1298 rcu_read_lock();
1299 __set_task_comm(me, smp_load_acquire(&comm_file->f_path.dentry->d_name.name),
1300 true);
1301 rcu_read_unlock();
1302 } else {
1303 __set_task_comm(me, kbasename(bprm->filename), true);
1304 }
1305
1306 /* An exec changes our domain. We are no longer part of the thread
1307 group */
1308 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1309 flush_signal_handlers(me, 0);
1310
1311 retval = set_cred_ucounts(bprm->cred);
1312 if (retval < 0)
1313 goto out_unlock;
1314
1315 /*
1316 * install the new credentials for this executable
1317 */
1318 security_bprm_committing_creds(bprm);
1319
1320 commit_creds(bprm->cred);
1321 bprm->cred = NULL;
1322
1323 /*
1324 * Disable monitoring for regular users
1325 * when executing setuid binaries. Must
1326 * wait until new credentials are committed
1327 * by commit_creds() above
1328 */
1329 if (task_exec_state_get_dumpable(me) != TASK_DUMPABLE_OWNER)
1330 perf_event_exit_task(me);
1331 /*
1332 * cred_guard_mutex must be held at least to this point to prevent
1333 * ptrace_attach() from altering our determination of the task's
1334 * credentials; any time after this it may be unlocked.
1335 */
1336 security_bprm_committed_creds(bprm);
1337
1338 /* Pass the opened binary to the interpreter. */
1339 if (bprm->have_execfd) {
1340 struct file *executable = bprm->executable;
1341
1342 /* mm->exe_file carries its own write denial now so drop it. */
1343 exe_file_allow_write_access(executable);
1344 bprm->executable = NULL;
1345 retval = FD_ADD(0, executable);
1346 if (retval < 0) {
1347 /* The reference was not consumed. */
1348 fput(executable);
1349 goto out_unlock;
1350 }
1351 bprm->execfd = retval;
1352 }
1353 return 0;
1354
1355out_unlock:
1356 up_write(&me->signal->exec_update_lock);
1357 if (!bprm->cred)
1358 mutex_unlock(&me->signal->cred_guard_mutex);
1359
1360out:
1361 return retval;
1362}
1363EXPORT_SYMBOL(begin_new_exec);
1364
1365void would_dump(struct linux_binprm *bprm, struct file *file)
1366{
1367 struct inode *inode = file_inode(file);
1368 const struct mnt_idmap *idmap = file_mnt_idmap(file);
1369 if (inode_permission(idmap, inode, MAY_READ) < 0) {
1370 struct user_namespace *old, *user_ns;
1371 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1372
1373 /* Ensure bprm->user_ns contains the executable. */
1374 user_ns = old = bprm->user_ns;
1375 while ((user_ns != &init_user_ns) &&
1376 !privileged_wrt_inode_uidgid(user_ns, idmap, inode))
1377 user_ns = user_ns->parent;
1378
1379 if (old != user_ns) {
1380 bprm->user_ns = get_user_ns(user_ns);
1381 put_user_ns(old);
1382 }
1383 }
1384}
1385EXPORT_SYMBOL(would_dump);
1386
1387void setup_new_exec(struct linux_binprm * bprm)
1388{
1389 /* Setup things that can depend upon the personality */
1390 struct task_struct *me = current;
1391
1392 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1393
1394 arch_setup_new_exec();
1395
1396 /* Set the new mm task size. We have to do that late because it may
1397 * depend on TIF_32BIT which is only updated in flush_thread() on
1398 * some architectures like powerpc
1399 */
1400 me->mm->task_size = TASK_SIZE;
1401 up_write(&me->signal->exec_update_lock);
1402 mutex_unlock(&me->signal->cred_guard_mutex);
1403
1404 /* The exec locks are dropped: release the old address space now. */
1405 if (bprm->old_mm) {
1406 exec_mm_put_old(bprm->old_mm);
1407 bprm->old_mm = NULL;
1408 }
1409}
1410EXPORT_SYMBOL(setup_new_exec);
1411
1412/* Runs immediately before start_thread() takes over. */
1413void finalize_exec(struct linux_binprm *bprm)
1414{
1415 /* Store any stack rlimit changes before starting thread. */
1416 task_lock(current->group_leader);
1417 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1418 task_unlock(current->group_leader);
1419}
1420EXPORT_SYMBOL(finalize_exec);
1421
1422/*
1423 * Prepare credentials and lock ->cred_guard_mutex.
1424 * setup_new_exec() commits the new creds and drops the lock.
1425 * Or, if exec fails before, free_bprm() should release ->cred
1426 * and unlock.
1427 */
1428static int prepare_bprm_creds(struct linux_binprm *bprm)
1429{
1430 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1431 return -ERESTARTNOINTR;
1432
1433 bprm->cred = prepare_exec_creds();
1434 if (likely(bprm->cred))
1435 return 0;
1436
1437 mutex_unlock(&current->signal->cred_guard_mutex);
1438 return -ENOMEM;
1439}
1440
1441/* Matches do_open_execat() */
1442static void do_close_execat(struct file *file)
1443{
1444 if (!file)
1445 return;
1446 exe_file_allow_write_access(file);
1447 fput(file);
1448}
1449
1450/**
1451 * bprm_open_interpreter - open the interpreter the binary asks for
1452 * @bprm: binary that is being executed
1453 * @path: the interpreter path named in the binary's PT_INTERP
1454 *
1455 * A binfmt_misc loader entry substitutes for the interpreter the binary
1456 * names. Hand out the stashed substitute if there is one and open @path
1457 * if there is not. The caller owns the reference either way and releases
1458 * it like any other open_exec() one.
1459 *
1460 * Return: the interpreter on success, an ERR_PTR on failure
1461 */
1462struct file *bprm_open_interpreter(struct linux_binprm *bprm, const char *path)
1463{
1464 if (bprm->loader)
1465 return no_free_ptr(bprm->loader);
1466 return open_exec(path);
1467}
1468
1469/**
1470 * bprm_drop_loader - discard a PT_INTERP substitute that does not apply
1471 * @bprm: binary that is being executed
1472 *
1473 * A binary without PT_INTERP has nothing to substitute for, so drop the
1474 * override and let the binary load natively rather than have
1475 * begin_new_exec() refuse it. A no-op once bprm_open_interpreter() took
1476 * the substitute.
1477 */
1478void bprm_drop_loader(struct linux_binprm *bprm)
1479{
1480 do_close_execat(no_free_ptr(bprm->loader));
1481}
1482
1483static void free_bprm(struct linux_binprm *bprm)
1484{
1485 if (bprm->mm) {
1486 acct_arg_size(bprm, 0);
1487 mmput(bprm->mm);
1488 }
1489 if (bprm->user_ns)
1490 put_user_ns(bprm->user_ns);
1491 free_arg_pages(bprm);
1492 if (bprm->cred) {
1493 /* in case exec fails before de_thread() succeeds */
1494 current->fs->in_exec = 0;
1495 mutex_unlock(&current->signal->cred_guard_mutex);
1496 abort_creds(bprm->cred);
1497 }
1498 /* exec swapped the mm but failed before setup_new_exec() freed it */
1499 if (bprm->old_mm)
1500 exec_mm_put_old(bprm->old_mm);
1501 /* An unconsumed PT_INTERP substitute from a binfmt_misc loader entry. */
1502 bprm_drop_loader(bprm);
1503 do_close_execat(bprm->file);
1504 do_close_execat(bprm->executable);
1505 /* If a binfmt changed the interp, free it. */
1506 if (bprm->interp != bprm->filename)
1507 kfree(bprm->interp);
1508 kfree(bprm->bpf_interp);
1509 if (bprm->bpf_interp_file)
1510 fput(bprm->bpf_interp_file);
1511 kfree(bprm->bpf_interp_arg);
1512 kfree(bprm->fdpath);
1513 kfree(bprm);
1514}
1515
1516static struct linux_binprm *alloc_bprm(int fd, struct filename *filename, int flags)
1517{
1518 struct linux_binprm *bprm;
1519 struct file *file;
1520 int retval = -ENOMEM;
1521
1522 file = do_open_execat(fd, filename, flags);
1523 if (IS_ERR(file))
1524 return ERR_CAST(file);
1525
1526 bprm = kzalloc_obj(*bprm);
1527 if (!bprm) {
1528 do_close_execat(file);
1529 return ERR_PTR(-ENOMEM);
1530 }
1531
1532 bprm->file = file;
1533
1534 if (fd == AT_FDCWD || filename->name[0] == '/') {
1535 bprm->filename = filename->name;
1536 } else {
1537 if (filename->name[0] == '\0') {
1538 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1539 bprm->comm_from_dentry = 1;
1540 } else {
1541 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1542 fd, filename->name);
1543 }
1544 if (!bprm->fdpath)
1545 goto out_free;
1546
1547 /*
1548 * Record that a name derived from an O_CLOEXEC fd will be
1549 * inaccessible after exec. This allows the code in exec to
1550 * choose to fail when the executable is not mmaped into the
1551 * interpreter and an open file descriptor is not passed to
1552 * the interpreter. This makes for a better user experience
1553 * than having the interpreter start and then immediately fail
1554 * when it finds the executable is inaccessible.
1555 */
1556 if (get_close_on_exec(fd))
1557 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1558
1559 bprm->filename = bprm->fdpath;
1560 }
1561 bprm->interp = bprm->filename;
1562
1563 /*
1564 * At this point, security_file_open() has already been called (with
1565 * __FMODE_EXEC) and access control checks for AT_EXECVE_CHECK will
1566 * stop just after the security_bprm_creds_for_exec() call in
1567 * bprm_execve(). Indeed, the kernel should not try to parse the
1568 * content of the file with exec_binprm() nor change the calling
1569 * thread, which means that the following security functions will not
1570 * be called:
1571 * - security_bprm_check()
1572 * - security_bprm_creds_from_file()
1573 * - security_bprm_committing_creds()
1574 * - security_bprm_committed_creds()
1575 */
1576 bprm->is_check = !!(flags & AT_EXECVE_CHECK);
1577
1578 retval = bprm_mm_init(bprm);
1579 if (!retval)
1580 return bprm;
1581
1582out_free:
1583 free_bprm(bprm);
1584 return ERR_PTR(retval);
1585}
1586
1587DEFINE_CLASS(bprm, struct linux_binprm *, if (!IS_ERR(_T)) free_bprm(_T),
1588 alloc_bprm(fd, name, flags), int fd, struct filename *name, int flags)
1589
1590int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1591{
1592 /* If a binfmt changed the interp, free it first. */
1593 if (bprm->interp != bprm->filename)
1594 kfree(bprm->interp);
1595 bprm->interp = kstrdup(interp, GFP_KERNEL);
1596 if (!bprm->interp)
1597 return -ENOMEM;
1598 return 0;
1599}
1600EXPORT_SYMBOL(bprm_change_interp);
1601
1602/*
1603 * determine how safe it is to execute the proposed program
1604 * - the caller must hold ->cred_guard_mutex to protect against
1605 * PTRACE_ATTACH or seccomp thread-sync
1606 */
1607static void check_unsafe_exec(struct linux_binprm *bprm)
1608{
1609 struct task_struct *p = current, *t;
1610 unsigned n_fs;
1611
1612 if (p->ptrace)
1613 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1614
1615 /*
1616 * This isn't strictly necessary, but it makes it harder for LSMs to
1617 * mess up.
1618 */
1619 if (task_no_new_privs(current))
1620 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1621
1622 /*
1623 * If another task is sharing our fs, we cannot safely
1624 * suid exec because the differently privileged task
1625 * will be able to manipulate the current directory, etc.
1626 * It would be nice to force an unshare instead...
1627 *
1628 * Otherwise we set fs->in_exec = 1 to deny clone(CLONE_FS)
1629 * from another sub-thread until de_thread() succeeds, this
1630 * state is protected by cred_guard_mutex we hold.
1631 */
1632 n_fs = 1;
1633 read_seqlock_excl(&p->fs->seq);
1634 rcu_read_lock();
1635 for_other_threads(p, t) {
1636 if (t->fs == p->fs)
1637 n_fs++;
1638 }
1639 rcu_read_unlock();
1640
1641 /* "users" and "in_exec" locked for copy_fs() */
1642 if (p->fs->users > n_fs)
1643 bprm->unsafe |= LSM_UNSAFE_SHARE;
1644 else
1645 p->fs->in_exec = 1;
1646 read_sequnlock_excl(&p->fs->seq);
1647}
1648
1649static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1650{
1651 /* Handle suid and sgid on files */
1652 const struct mnt_idmap *idmap;
1653 struct inode *inode = file_inode(file);
1654 unsigned int mode;
1655 vfsuid_t vfsuid;
1656 vfsgid_t vfsgid;
1657 int err;
1658
1659 if (!mnt_may_suid(file->f_path.mnt))
1660 return;
1661
1662 if (task_no_new_privs(current))
1663 return;
1664
1665 mode = READ_ONCE(inode->i_mode);
1666 if (!(mode & (S_ISUID|S_ISGID)))
1667 return;
1668
1669 idmap = file_mnt_idmap(file);
1670
1671 /* Be careful if suid/sgid is set */
1672 inode_lock(inode);
1673
1674 /* Atomically reload and check mode/uid/gid now that lock held. */
1675 mode = inode->i_mode;
1676 vfsuid = i_uid_into_vfsuid(idmap, inode);
1677 vfsgid = i_gid_into_vfsgid(idmap, inode);
1678 err = inode_permission(idmap, inode, MAY_EXEC);
1679 inode_unlock(inode);
1680
1681 /* Did the exec bit vanish out from under us? Give up. */
1682 if (err)
1683 return;
1684
1685 /* We ignore suid/sgid if there are no mappings for them in the ns */
1686 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1687 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1688 return;
1689
1690 if (mode & S_ISUID) {
1691 bprm->per_clear |= PER_CLEAR_ON_SETID;
1692 bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1693 }
1694
1695 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1696 bprm->per_clear |= PER_CLEAR_ON_SETID;
1697 bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1698 }
1699}
1700
1701/*
1702 * Compute brpm->cred based upon the final binary.
1703 */
1704static int bprm_creds_from_file(struct linux_binprm *bprm)
1705{
1706 /* Compute creds based on which file? */
1707 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1708
1709 bprm_fill_uid(bprm, file);
1710 return security_bprm_creds_from_file(bprm, file);
1711}
1712
1713/*
1714 * Fill the binprm structure from the inode.
1715 * Read the first BINPRM_BUF_SIZE bytes
1716 *
1717 * This may be called multiple times for binary chains (scripts for example).
1718 */
1719static int prepare_binprm(struct linux_binprm *bprm)
1720{
1721 loff_t pos = 0;
1722
1723 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1724 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1725}
1726
1727/*
1728 * Arguments are '\0' separated strings found at the location bprm->p
1729 * points to; chop off the first by relocating brpm->p to right after
1730 * the first '\0' encountered.
1731 */
1732int remove_arg_zero(struct linux_binprm *bprm)
1733{
1734 unsigned long offset;
1735 char *kaddr;
1736 struct page *page;
1737
1738 if (!bprm->argc)
1739 return 0;
1740
1741 do {
1742 offset = bprm->p & ~PAGE_MASK;
1743 page = get_arg_page(bprm, bprm->p, 0);
1744 if (!page)
1745 return -EFAULT;
1746 kaddr = kmap_local_page(page);
1747
1748 for (; offset < PAGE_SIZE && kaddr[offset];
1749 offset++, bprm->p++)
1750 ;
1751
1752 kunmap_local(kaddr);
1753 put_arg_page(page);
1754 } while (offset == PAGE_SIZE);
1755
1756 bprm->p++;
1757 bprm->argc--;
1758
1759 return 0;
1760}
1761EXPORT_SYMBOL(remove_arg_zero);
1762
1763/*
1764 * cycle the list of binary formats handler, until one recognizes the image
1765 */
1766static int search_binary_handler(struct linux_binprm *bprm)
1767{
1768 struct linux_binfmt *fmt;
1769 int retval;
1770
1771 retval = prepare_binprm(bprm);
1772 if (retval < 0)
1773 return retval;
1774
1775 retval = security_bprm_check(bprm);
1776 if (retval)
1777 return retval;
1778
1779 read_lock(&binfmt_lock);
1780 list_for_each_entry(fmt, &formats, lh) {
1781 if (!try_module_get(fmt->module))
1782 continue;
1783 read_unlock(&binfmt_lock);
1784
1785 retval = fmt->load_binary(bprm);
1786
1787 read_lock(&binfmt_lock);
1788 put_binfmt(fmt);
1789 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1790 read_unlock(&binfmt_lock);
1791 return retval;
1792 }
1793 }
1794 read_unlock(&binfmt_lock);
1795
1796 return -ENOEXEC;
1797}
1798
1799/* binfmt handlers will call back into begin_new_exec() on success. */
1800static int exec_binprm(struct linux_binprm *bprm)
1801{
1802 pid_t old_pid, old_vpid;
1803 int ret, depth;
1804
1805 /* Need to fetch pid before load_binary changes it */
1806 old_pid = current->pid;
1807 rcu_read_lock();
1808 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1809 rcu_read_unlock();
1810
1811 /* This allows 5 levels of binfmt rewrites before failing hard. */
1812 for (depth = 0;; depth++) {
1813 struct file *exec;
1814 if (depth > 5)
1815 return -ELOOP;
1816
1817 ret = search_binary_handler(bprm);
1818 if (ret < 0)
1819 return ret;
1820 if (!bprm->interpreter)
1821 break;
1822
1823 /* A stashed PT_INTERP substitute belonged to the replaced file. */
1824 bprm_drop_loader(bprm);
1825
1826 exec = bprm->file;
1827 bprm->file = bprm->interpreter;
1828 bprm->interpreter = NULL;
1829
1830 if (unlikely(bprm->have_execfd)) {
1831 if (bprm->executable) {
1832 do_close_execat(exec);
1833 return -ENOEXEC;
1834 }
1835 /* Kept for AT_EXECFD; the write denial rides along until hand-over. */
1836 bprm->executable = exec;
1837 } else {
1838 do_close_execat(exec);
1839 }
1840 }
1841
1842 audit_bprm(bprm);
1843 trace_sched_process_exec(current, old_pid, bprm);
1844 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1845 proc_exec_connector(current);
1846 return 0;
1847}
1848
1849static int bprm_execve(struct linux_binprm *bprm)
1850{
1851 int retval;
1852
1853 retval = prepare_bprm_creds(bprm);
1854 if (retval)
1855 return retval;
1856
1857 /*
1858 * Check for unsafe execution states before exec_binprm(), which
1859 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1860 * where setuid-ness is evaluated.
1861 */
1862 check_unsafe_exec(bprm);
1863 current->in_execve = 1;
1864 sched_mm_cid_before_execve(current);
1865
1866 sched_exec();
1867
1868 /* Set the unchanging part of bprm->cred */
1869 retval = security_bprm_creds_for_exec(bprm);
1870 if (retval || bprm->is_check)
1871 goto out;
1872
1873 retval = exec_binprm(bprm);
1874 if (retval < 0)
1875 goto out;
1876
1877 sched_mm_cid_after_execve(current);
1878 rseq_execve(current);
1879 /* execve succeeded */
1880 current->in_execve = 0;
1881 user_events_execve(current);
1882 acct_update_integrals(current);
1883 task_numa_free(current, false);
1884 return retval;
1885
1886out:
1887 /*
1888 * If past the point of no return ensure the code never
1889 * returns to the userspace process. Use an existing fatal
1890 * signal if present otherwise terminate the process with
1891 * SIGSEGV.
1892 */
1893 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1894 force_fatal_sig(SIGSEGV);
1895
1896 sched_mm_cid_after_execve(current);
1897 rseq_force_update();
1898 current->in_execve = 0;
1899
1900 return retval;
1901}
1902
1903static int do_execveat_common(int fd, struct filename *filename,
1904 struct user_arg_ptr argv,
1905 struct user_arg_ptr envp,
1906 int flags)
1907{
1908 int retval;
1909
1910 /*
1911 * We move the actual failure in case of RLIMIT_NPROC excess from
1912 * set*uid() to execve() because too many poorly written programs
1913 * don't check setuid() return code. Here we additionally recheck
1914 * whether NPROC limit is still exceeded.
1915 */
1916 if ((current->flags & PF_NPROC_EXCEEDED) &&
1917 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)))
1918 return -EAGAIN;
1919
1920 /* We're below the limit (still or again), so we don't want to make
1921 * further execve() calls fail. */
1922 current->flags &= ~PF_NPROC_EXCEEDED;
1923
1924 CLASS(bprm, bprm)(fd, filename, flags);
1925 if (IS_ERR(bprm))
1926 return PTR_ERR(bprm);
1927
1928 retval = count(argv, MAX_ARG_STRINGS);
1929 if (retval < 0)
1930 return retval;
1931 bprm->argc = retval;
1932
1933 retval = count(envp, MAX_ARG_STRINGS);
1934 if (retval < 0)
1935 return retval;
1936 bprm->envc = retval;
1937
1938 retval = bprm_stack_limits(bprm);
1939 if (retval < 0)
1940 return retval;
1941
1942 retval = copy_string_kernel(bprm->filename, bprm);
1943 if (retval < 0)
1944 return retval;
1945 bprm->exec = bprm->p;
1946
1947 retval = copy_strings(bprm->envc, envp, bprm);
1948 if (retval < 0)
1949 return retval;
1950
1951 retval = copy_strings(bprm->argc, argv, bprm);
1952 if (retval < 0)
1953 return retval;
1954
1955 /*
1956 * When argv is empty, add an empty string ("") as argv[0] to
1957 * ensure confused userspace programs that start processing
1958 * from argv[1] won't end up walking envp. See also
1959 * bprm_stack_limits().
1960 */
1961 if (bprm->argc == 0) {
1962 retval = copy_string_kernel("", bprm);
1963 if (retval < 0)
1964 return retval;
1965 bprm->argc = 1;
1966
1967 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1968 current->comm, bprm->filename);
1969 }
1970
1971 return bprm_execve(bprm);
1972}
1973
1974int kernel_execve(const char *kernel_filename,
1975 const char *const *argv, const char *const *envp)
1976{
1977 int retval;
1978
1979 /* It is non-sense for kernel threads to call execve */
1980 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1981 return -EINVAL;
1982
1983 CLASS(filename_kernel, filename)(kernel_filename);
1984 CLASS(bprm, bprm)(AT_FDCWD, filename, 0);
1985 if (IS_ERR(bprm))
1986 return PTR_ERR(bprm);
1987
1988 retval = count_strings_kernel(argv);
1989 if (WARN_ON_ONCE(retval == 0))
1990 return -EINVAL;
1991 if (retval < 0)
1992 return retval;
1993 bprm->argc = retval;
1994
1995 retval = count_strings_kernel(envp);
1996 if (retval < 0)
1997 return retval;
1998 bprm->envc = retval;
1999
2000 retval = bprm_stack_limits(bprm);
2001 if (retval < 0)
2002 return retval;
2003
2004 retval = copy_string_kernel(bprm->filename, bprm);
2005 if (retval < 0)
2006 return retval;
2007 bprm->exec = bprm->p;
2008
2009 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2010 if (retval < 0)
2011 return retval;
2012
2013 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2014 if (retval < 0)
2015 return retval;
2016
2017 return bprm_execve(bprm);
2018}
2019
2020void set_binfmt(struct linux_binfmt *new)
2021{
2022 struct mm_struct *mm = current->mm;
2023
2024 if (mm->binfmt)
2025 module_put(mm->binfmt->module);
2026
2027 mm->binfmt = new;
2028 if (new)
2029 __module_get(new->module);
2030}
2031EXPORT_SYMBOL(set_binfmt);
2032
2033static inline struct user_arg_ptr native_arg(const char __user *const __user *p)
2034{
2035 return (struct user_arg_ptr){.ptr.native = p};
2036}
2037
2038SYSCALL_DEFINE3(execve,
2039 const char __user *, filename,
2040 const char __user *const __user *, argv,
2041 const char __user *const __user *, envp)
2042{
2043 CLASS(filename, name)(filename);
2044 return do_execveat_common(AT_FDCWD, name,
2045 native_arg(argv), native_arg(envp), 0);
2046}
2047
2048SYSCALL_DEFINE5(execveat,
2049 int, fd, const char __user *, filename,
2050 const char __user *const __user *, argv,
2051 const char __user *const __user *, envp,
2052 int, flags)
2053{
2054 CLASS(filename_uflags, name)(filename, flags);
2055 return do_execveat_common(fd, name,
2056 native_arg(argv), native_arg(envp), flags);
2057}
2058
2059#ifdef CONFIG_COMPAT
2060
2061static inline struct user_arg_ptr compat_arg(const compat_uptr_t __user *p)
2062{
2063 return (struct user_arg_ptr){.is_compat = true, .ptr.compat = p};
2064}
2065
2066COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2067 const compat_uptr_t __user *, argv,
2068 const compat_uptr_t __user *, envp)
2069{
2070 CLASS(filename, name)(filename);
2071 return do_execveat_common(AT_FDCWD, name,
2072 compat_arg(argv), compat_arg(envp), 0);
2073}
2074
2075COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2076 const char __user *, filename,
2077 const compat_uptr_t __user *, argv,
2078 const compat_uptr_t __user *, envp,
2079 int, flags)
2080{
2081 CLASS(filename_uflags, name)(filename, flags);
2082 return do_execveat_common(fd, name,
2083 compat_arg(argv), compat_arg(envp), flags);
2084}
2085#endif
2086
2087#ifdef CONFIG_SYSCTL
2088
2089static int proc_dointvec_minmax_coredump(const struct ctl_table *table, int write,
2090 void *buffer, size_t *lenp, loff_t *ppos)
2091{
2092 int error, old = READ_ONCE(suid_dumpable);
2093
2094 error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2095
2096 if (!error && write && (old != READ_ONCE(suid_dumpable)))
2097 validate_coredump_safety();
2098 return error;
2099}
2100
2101static const struct ctl_table fs_exec_sysctls[] = {
2102 {
2103 .procname = "suid_dumpable",
2104 .data = &suid_dumpable,
2105 .maxlen = sizeof(int),
2106 .mode = 0644,
2107 .proc_handler = proc_dointvec_minmax_coredump,
2108 .extra1 = SYSCTL_ZERO,
2109 .extra2 = SYSCTL_TWO,
2110 },
2111};
2112
2113static int __init init_fs_exec_sysctls(void)
2114{
2115 register_sysctl_init("fs", fs_exec_sysctls);
2116 return 0;
2117}
2118
2119fs_initcall(init_fs_exec_sysctls);
2120#endif /* CONFIG_SYSCTL */
2121
2122#ifdef CONFIG_EXEC_KUNIT_TEST
2123#include "tests/exec_kunit.c"
2124#endif