1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * #!-checking implemented by tytso.
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.
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.
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
26 #include <linux/kernel_read_file.h>
27 #include <linux/slab.h>
28 #include <linux/file.h>
29 #include <linux/fdtable.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 #include <linux/user_events.h>
70 #include <linux/uaccess.h>
71 #include <asm/mmu_context.h>
74 #include <trace/events/task.h>
77 #include <trace/events/sched.h>
79 static int bprm_creds_from_file(struct linux_binprm
*bprm
);
81 int suid_dumpable
= 0;
83 static LIST_HEAD(formats
);
84 static DEFINE_RWLOCK(binfmt_lock
);
86 void __register_binfmt(struct linux_binfmt
* fmt
, int insert
)
88 write_lock(&binfmt_lock
);
89 insert
? list_add(&fmt
->lh
, &formats
) :
90 list_add_tail(&fmt
->lh
, &formats
);
91 write_unlock(&binfmt_lock
);
94 EXPORT_SYMBOL(__register_binfmt
);
96 void unregister_binfmt(struct linux_binfmt
* fmt
)
98 write_lock(&binfmt_lock
);
100 write_unlock(&binfmt_lock
);
103 EXPORT_SYMBOL(unregister_binfmt
);
105 static inline void put_binfmt(struct linux_binfmt
* fmt
)
107 module_put(fmt
->module
);
110 bool path_noexec(const struct path
*path
)
112 return (path
->mnt
->mnt_flags
& MNT_NOEXEC
) ||
113 (path
->mnt
->mnt_sb
->s_iflags
& SB_I_NOEXEC
);
118 * Note that a shared library must be both readable and executable due to
121 * Also note that we take the address to load from the file itself.
123 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
125 struct linux_binfmt
*fmt
;
127 struct filename
*tmp
= getname(library
);
128 int error
= PTR_ERR(tmp
);
129 static const struct open_flags uselib_flags
= {
130 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
131 .acc_mode
= MAY_READ
| MAY_EXEC
,
132 .intent
= LOOKUP_OPEN
,
133 .lookup_flags
= LOOKUP_FOLLOW
,
139 file
= do_filp_open(AT_FDCWD
, tmp
, &uselib_flags
);
141 error
= PTR_ERR(file
);
146 * may_open() has already checked for this, so it should be
147 * impossible to trip now. But we need to be extra cautious
148 * and check again at the very end too.
151 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
152 path_noexec(&file
->f_path
)))
157 read_lock(&binfmt_lock
);
158 list_for_each_entry(fmt
, &formats
, lh
) {
159 if (!fmt
->load_shlib
)
161 if (!try_module_get(fmt
->module
))
163 read_unlock(&binfmt_lock
);
164 error
= fmt
->load_shlib(file
);
165 read_lock(&binfmt_lock
);
167 if (error
!= -ENOEXEC
)
170 read_unlock(&binfmt_lock
);
176 #endif /* #ifdef CONFIG_USELIB */
180 * The nascent bprm->mm is not visible until exec_mmap() but it can
181 * use a lot of memory, account these pages in current->mm temporary
182 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
183 * change the counter back via acct_arg_size(0).
185 static void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
187 struct mm_struct
*mm
= current
->mm
;
188 long diff
= (long)(pages
- bprm
->vma_pages
);
193 bprm
->vma_pages
= pages
;
194 add_mm_counter(mm
, MM_ANONPAGES
, diff
);
197 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
201 struct vm_area_struct
*vma
= bprm
->vma
;
202 struct mm_struct
*mm
= bprm
->mm
;
206 * Avoid relying on expanding the stack down in GUP (which
207 * does not work for STACK_GROWSUP anyway), and just do it
208 * by hand ahead of time.
210 if (write
&& pos
< vma
->vm_start
) {
212 ret
= expand_downwards(vma
, pos
);
213 if (unlikely(ret
< 0)) {
214 mmap_write_unlock(mm
);
217 mmap_write_downgrade(mm
);
222 * We are doing an exec(). 'current' is the process
223 * doing the exec and 'mm' is the new process's mm.
225 ret
= get_user_pages_remote(mm
, pos
, 1,
226 write
? FOLL_WRITE
: 0,
228 mmap_read_unlock(mm
);
233 acct_arg_size(bprm
, vma_pages(vma
));
238 static void put_arg_page(struct page
*page
)
243 static void free_arg_pages(struct linux_binprm
*bprm
)
247 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
250 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
253 static int __bprm_mm_init(struct linux_binprm
*bprm
)
256 struct vm_area_struct
*vma
= NULL
;
257 struct mm_struct
*mm
= bprm
->mm
;
259 bprm
->vma
= vma
= vm_area_alloc(mm
);
262 vma_set_anonymous(vma
);
264 if (mmap_write_lock_killable(mm
)) {
270 * Place the stack at the largest stack address the architecture
271 * supports. Later, we'll move this to an appropriate place. We don't
272 * use STACK_TOP because that can depend on attributes which aren't
275 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
276 vma
->vm_end
= STACK_TOP_MAX
;
277 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
278 vm_flags_init(vma
, VM_SOFTDIRTY
| VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
);
279 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
281 err
= insert_vm_struct(mm
, vma
);
285 mm
->stack_vm
= mm
->total_vm
= 1;
286 mmap_write_unlock(mm
);
287 bprm
->p
= vma
->vm_end
- sizeof(void *);
290 mmap_write_unlock(mm
);
297 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
299 return len
<= MAX_ARG_STRLEN
;
304 static inline void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
308 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
313 page
= bprm
->page
[pos
/ PAGE_SIZE
];
314 if (!page
&& write
) {
315 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
318 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
324 static void put_arg_page(struct page
*page
)
328 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
331 __free_page(bprm
->page
[i
]);
332 bprm
->page
[i
] = NULL
;
336 static void free_arg_pages(struct linux_binprm
*bprm
)
340 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
341 free_arg_page(bprm
, i
);
344 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
349 static int __bprm_mm_init(struct linux_binprm
*bprm
)
351 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
355 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
357 return len
<= bprm
->p
;
360 #endif /* CONFIG_MMU */
363 * Create a new mm_struct and populate it with a temporary stack
364 * vm_area_struct. We don't have enough context at this point to set the stack
365 * flags, permissions, and offset, so we use temporary values. We'll update
366 * them later in setup_arg_pages().
368 static int bprm_mm_init(struct linux_binprm
*bprm
)
371 struct mm_struct
*mm
= NULL
;
373 bprm
->mm
= mm
= mm_alloc();
378 /* Save current stack limit for all calculations made during exec. */
379 task_lock(current
->group_leader
);
380 bprm
->rlim_stack
= current
->signal
->rlim
[RLIMIT_STACK
];
381 task_unlock(current
->group_leader
);
383 err
= __bprm_mm_init(bprm
);
398 struct user_arg_ptr
{
403 const char __user
*const __user
*native
;
405 const compat_uptr_t __user
*compat
;
410 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
412 const char __user
*native
;
415 if (unlikely(argv
.is_compat
)) {
416 compat_uptr_t compat
;
418 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
419 return ERR_PTR(-EFAULT
);
421 return compat_ptr(compat
);
425 if (get_user(native
, argv
.ptr
.native
+ nr
))
426 return ERR_PTR(-EFAULT
);
432 * count() counts the number of strings in array ARGV.
434 static int count(struct user_arg_ptr argv
, int max
)
438 if (argv
.ptr
.native
!= NULL
) {
440 const char __user
*p
= get_user_arg_ptr(argv
, i
);
452 if (fatal_signal_pending(current
))
453 return -ERESTARTNOHAND
;
460 static int count_strings_kernel(const char *const *argv
)
467 for (i
= 0; argv
[i
]; ++i
) {
468 if (i
>= MAX_ARG_STRINGS
)
470 if (fatal_signal_pending(current
))
471 return -ERESTARTNOHAND
;
477 static int bprm_stack_limits(struct linux_binprm
*bprm
)
479 unsigned long limit
, ptr_size
;
482 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
483 * (whichever is smaller) for the argv+env strings.
485 * - the remaining binfmt code will not run out of stack space,
486 * - the program will have a reasonable amount of stack left
489 limit
= _STK_LIM
/ 4 * 3;
490 limit
= min(limit
, bprm
->rlim_stack
.rlim_cur
/ 4);
492 * We've historically supported up to 32 pages (ARG_MAX)
493 * of argument strings even with small stacks
495 limit
= max_t(unsigned long, limit
, ARG_MAX
);
497 * We must account for the size of all the argv and envp pointers to
498 * the argv and envp strings, since they will also take up space in
499 * the stack. They aren't stored until much later when we can't
500 * signal to the parent that the child has run out of stack space.
501 * Instead, calculate it here so it's possible to fail gracefully.
503 * In the case of argc = 0, make sure there is space for adding a
504 * empty string (which will bump argc to 1), to ensure confused
505 * userspace programs don't start processing from argv[1], thinking
506 * argc can never be 0, to keep them from walking envp by accident.
507 * See do_execveat_common().
509 ptr_size
= (max(bprm
->argc
, 1) + bprm
->envc
) * sizeof(void *);
510 if (limit
<= ptr_size
)
514 bprm
->argmin
= bprm
->p
- limit
;
519 * 'copy_strings()' copies argument/environment strings from the old
520 * processes's memory to the new process's stack. The call to get_user_pages()
521 * ensures the destination page is created and not swapped out.
523 static int copy_strings(int argc
, struct user_arg_ptr argv
,
524 struct linux_binprm
*bprm
)
526 struct page
*kmapped_page
= NULL
;
528 unsigned long kpos
= 0;
532 const char __user
*str
;
537 str
= get_user_arg_ptr(argv
, argc
);
541 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
546 if (!valid_arg_len(bprm
, len
))
549 /* We're going to work our way backwards. */
554 if (bprm
->p
< bprm
->argmin
)
559 int offset
, bytes_to_copy
;
561 if (fatal_signal_pending(current
)) {
562 ret
= -ERESTARTNOHAND
;
567 offset
= pos
% PAGE_SIZE
;
571 bytes_to_copy
= offset
;
572 if (bytes_to_copy
> len
)
575 offset
-= bytes_to_copy
;
576 pos
-= bytes_to_copy
;
577 str
-= bytes_to_copy
;
578 len
-= bytes_to_copy
;
580 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
583 page
= get_arg_page(bprm
, pos
, 1);
590 flush_dcache_page(kmapped_page
);
592 put_arg_page(kmapped_page
);
595 kaddr
= kmap_local_page(kmapped_page
);
596 kpos
= pos
& PAGE_MASK
;
597 flush_arg_page(bprm
, kpos
, kmapped_page
);
599 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
608 flush_dcache_page(kmapped_page
);
610 put_arg_page(kmapped_page
);
616 * Copy and argument/environment string from the kernel to the processes stack.
618 int copy_string_kernel(const char *arg
, struct linux_binprm
*bprm
)
620 int len
= strnlen(arg
, MAX_ARG_STRLEN
) + 1 /* terminating NUL */;
621 unsigned long pos
= bprm
->p
;
625 if (!valid_arg_len(bprm
, len
))
628 /* We're going to work our way backwards. */
631 if (IS_ENABLED(CONFIG_MMU
) && bprm
->p
< bprm
->argmin
)
635 unsigned int bytes_to_copy
= min_t(unsigned int, len
,
636 min_not_zero(offset_in_page(pos
), PAGE_SIZE
));
639 pos
-= bytes_to_copy
;
640 arg
-= bytes_to_copy
;
641 len
-= bytes_to_copy
;
643 page
= get_arg_page(bprm
, pos
, 1);
646 flush_arg_page(bprm
, pos
& PAGE_MASK
, page
);
647 memcpy_to_page(page
, offset_in_page(pos
), arg
, bytes_to_copy
);
653 EXPORT_SYMBOL(copy_string_kernel
);
655 static int copy_strings_kernel(int argc
, const char *const *argv
,
656 struct linux_binprm
*bprm
)
659 int ret
= copy_string_kernel(argv
[argc
], bprm
);
662 if (fatal_signal_pending(current
))
663 return -ERESTARTNOHAND
;
672 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
673 * the binfmt code determines where the new stack should reside, we shift it to
674 * its final location. The process proceeds as follows:
676 * 1) Use shift to calculate the new vma endpoints.
677 * 2) Extend vma to cover both the old and new ranges. This ensures the
678 * arguments passed to subsequent functions are consistent.
679 * 3) Move vma's page tables to the new range.
680 * 4) Free up any cleared pgd range.
681 * 5) Shrink the vma to cover only the new range.
683 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
685 struct mm_struct
*mm
= vma
->vm_mm
;
686 unsigned long old_start
= vma
->vm_start
;
687 unsigned long old_end
= vma
->vm_end
;
688 unsigned long length
= old_end
- old_start
;
689 unsigned long new_start
= old_start
- shift
;
690 unsigned long new_end
= old_end
- shift
;
691 VMA_ITERATOR(vmi
, mm
, new_start
);
692 struct vm_area_struct
*next
;
693 struct mmu_gather tlb
;
695 BUG_ON(new_start
> new_end
);
698 * ensure there are no vmas between where we want to go
701 if (vma
!= vma_next(&vmi
))
704 vma_iter_prev_range(&vmi
);
706 * cover the whole range: [new_start, old_end)
708 if (vma_expand(&vmi
, vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
712 * move the page tables downwards, on failure we rely on
713 * process cleanup to remove whatever mess we made.
715 if (length
!= move_page_tables(vma
, old_start
,
716 vma
, new_start
, length
, false, true))
720 tlb_gather_mmu(&tlb
, mm
);
721 next
= vma_next(&vmi
);
722 if (new_end
> old_start
) {
724 * when the old and new regions overlap clear from new_end.
726 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
727 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
730 * otherwise, clean from old_start; this is done to not touch
731 * the address space in [new_end, old_start) some architectures
732 * have constraints on va-space that make this illegal (IA64) -
733 * for the others its just a little faster.
735 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
736 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
738 tlb_finish_mmu(&tlb
);
741 /* Shrink the vma to just the new range */
742 return vma_shrink(&vmi
, vma
, new_start
, new_end
, vma
->vm_pgoff
);
746 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
747 * the stack is optionally relocated, and some extra space is added.
749 int setup_arg_pages(struct linux_binprm
*bprm
,
750 unsigned long stack_top
,
751 int executable_stack
)
754 unsigned long stack_shift
;
755 struct mm_struct
*mm
= current
->mm
;
756 struct vm_area_struct
*vma
= bprm
->vma
;
757 struct vm_area_struct
*prev
= NULL
;
758 unsigned long vm_flags
;
759 unsigned long stack_base
;
760 unsigned long stack_size
;
761 unsigned long stack_expand
;
762 unsigned long rlim_stack
;
763 struct mmu_gather tlb
;
764 struct vma_iterator vmi
;
766 #ifdef CONFIG_STACK_GROWSUP
767 /* Limit stack size */
768 stack_base
= bprm
->rlim_stack
.rlim_max
;
770 stack_base
= calc_max_stack_size(stack_base
);
772 /* Add space for stack randomization. */
773 stack_base
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
775 /* Make sure we didn't let the argument array grow too large. */
776 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
779 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
781 stack_shift
= vma
->vm_start
- stack_base
;
782 mm
->arg_start
= bprm
->p
- stack_shift
;
783 bprm
->p
= vma
->vm_end
- stack_shift
;
785 stack_top
= arch_align_stack(stack_top
);
786 stack_top
= PAGE_ALIGN(stack_top
);
788 if (unlikely(stack_top
< mmap_min_addr
) ||
789 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
792 stack_shift
= vma
->vm_end
- stack_top
;
794 bprm
->p
-= stack_shift
;
795 mm
->arg_start
= bprm
->p
;
799 bprm
->loader
-= stack_shift
;
800 bprm
->exec
-= stack_shift
;
802 if (mmap_write_lock_killable(mm
))
805 vm_flags
= VM_STACK_FLAGS
;
808 * Adjust stack execute permissions; explicitly enable for
809 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
810 * (arch default) otherwise.
812 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
814 else if (executable_stack
== EXSTACK_DISABLE_X
)
815 vm_flags
&= ~VM_EXEC
;
816 vm_flags
|= mm
->def_flags
;
817 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
819 vma_iter_init(&vmi
, mm
, vma
->vm_start
);
821 tlb_gather_mmu(&tlb
, mm
);
822 ret
= mprotect_fixup(&vmi
, &tlb
, vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
824 tlb_finish_mmu(&tlb
);
830 if (unlikely(vm_flags
& VM_EXEC
)) {
831 pr_warn_once("process '%pD4' started with executable stack\n",
835 /* Move stack pages down in memory. */
837 ret
= shift_arg_pages(vma
, stack_shift
);
842 /* mprotect_fixup is overkill to remove the temporary stack flags */
843 vm_flags_clear(vma
, VM_STACK_INCOMPLETE_SETUP
);
845 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
846 stack_size
= vma
->vm_end
- vma
->vm_start
;
848 * Align this down to a page boundary as expand_stack
851 rlim_stack
= bprm
->rlim_stack
.rlim_cur
& PAGE_MASK
;
853 stack_expand
= min(rlim_stack
, stack_size
+ stack_expand
);
855 #ifdef CONFIG_STACK_GROWSUP
856 stack_base
= vma
->vm_start
+ stack_expand
;
858 stack_base
= vma
->vm_end
- stack_expand
;
860 current
->mm
->start_stack
= bprm
->p
;
861 ret
= expand_stack_locked(vma
, stack_base
);
866 mmap_write_unlock(mm
);
869 EXPORT_SYMBOL(setup_arg_pages
);
874 * Transfer the program arguments and environment from the holding pages
875 * onto the stack. The provided stack pointer is adjusted accordingly.
877 int transfer_args_to_stack(struct linux_binprm
*bprm
,
878 unsigned long *sp_location
)
880 unsigned long index
, stop
, sp
;
883 stop
= bprm
->p
>> PAGE_SHIFT
;
886 for (index
= MAX_ARG_PAGES
- 1; index
>= stop
; index
--) {
887 unsigned int offset
= index
== stop
? bprm
->p
& ~PAGE_MASK
: 0;
888 char *src
= kmap_local_page(bprm
->page
[index
]) + offset
;
889 sp
-= PAGE_SIZE
- offset
;
890 if (copy_to_user((void *) sp
, src
, PAGE_SIZE
- offset
) != 0)
902 EXPORT_SYMBOL(transfer_args_to_stack
);
904 #endif /* CONFIG_MMU */
906 static struct file
*do_open_execat(int fd
, struct filename
*name
, int flags
)
910 struct open_flags open_exec_flags
= {
911 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
912 .acc_mode
= MAY_EXEC
,
913 .intent
= LOOKUP_OPEN
,
914 .lookup_flags
= LOOKUP_FOLLOW
,
917 if ((flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
918 return ERR_PTR(-EINVAL
);
919 if (flags
& AT_SYMLINK_NOFOLLOW
)
920 open_exec_flags
.lookup_flags
&= ~LOOKUP_FOLLOW
;
921 if (flags
& AT_EMPTY_PATH
)
922 open_exec_flags
.lookup_flags
|= LOOKUP_EMPTY
;
924 file
= do_filp_open(fd
, name
, &open_exec_flags
);
929 * may_open() has already checked for this, so it should be
930 * impossible to trip now. But we need to be extra cautious
931 * and check again at the very end too.
934 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
935 path_noexec(&file
->f_path
)))
938 err
= deny_write_access(file
);
950 struct file
*open_exec(const char *name
)
952 struct filename
*filename
= getname_kernel(name
);
953 struct file
*f
= ERR_CAST(filename
);
955 if (!IS_ERR(filename
)) {
956 f
= do_open_execat(AT_FDCWD
, filename
, 0);
961 EXPORT_SYMBOL(open_exec
);
963 #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
964 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
966 ssize_t res
= vfs_read(file
, (void __user
*)addr
, len
, &pos
);
968 flush_icache_user_range(addr
, addr
+ len
);
971 EXPORT_SYMBOL(read_code
);
975 * Maps the mm_struct mm into the current task struct.
976 * On success, this function returns with exec_update_lock
979 static int exec_mmap(struct mm_struct
*mm
)
981 struct task_struct
*tsk
;
982 struct mm_struct
*old_mm
, *active_mm
;
985 /* Notify parent that we're no longer interested in the old VM */
987 old_mm
= current
->mm
;
988 exec_mm_release(tsk
, old_mm
);
990 ret
= down_write_killable(&tsk
->signal
->exec_update_lock
);
996 * If there is a pending fatal signal perhaps a signal
997 * whose default action is to create a coredump get
998 * out and die instead of going through with the exec.
1000 ret
= mmap_read_lock_killable(old_mm
);
1002 up_write(&tsk
->signal
->exec_update_lock
);
1008 membarrier_exec_mmap(mm
);
1010 local_irq_disable();
1011 active_mm
= tsk
->active_mm
;
1012 tsk
->active_mm
= mm
;
1016 * This prevents preemption while active_mm is being loaded and
1017 * it and mm are being updated, which could cause problems for
1018 * lazy tlb mm refcounting when these are updated by context
1019 * switches. Not all architectures can handle irqs off over
1022 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1024 activate_mm(active_mm
, mm
);
1025 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1031 mmap_read_unlock(old_mm
);
1032 BUG_ON(active_mm
!= old_mm
);
1033 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
1034 mm_update_next_owner(old_mm
);
1038 mmdrop_lazy_tlb(active_mm
);
1042 static int de_thread(struct task_struct
*tsk
)
1044 struct signal_struct
*sig
= tsk
->signal
;
1045 struct sighand_struct
*oldsighand
= tsk
->sighand
;
1046 spinlock_t
*lock
= &oldsighand
->siglock
;
1048 if (thread_group_empty(tsk
))
1049 goto no_thread_group
;
1052 * Kill all other threads in the thread group.
1054 spin_lock_irq(lock
);
1055 if ((sig
->flags
& SIGNAL_GROUP_EXIT
) || sig
->group_exec_task
) {
1057 * Another group action in progress, just
1058 * return so that the signal is processed.
1060 spin_unlock_irq(lock
);
1064 sig
->group_exec_task
= tsk
;
1065 sig
->notify_count
= zap_other_threads(tsk
);
1066 if (!thread_group_leader(tsk
))
1067 sig
->notify_count
--;
1069 while (sig
->notify_count
) {
1070 __set_current_state(TASK_KILLABLE
);
1071 spin_unlock_irq(lock
);
1073 if (__fatal_signal_pending(tsk
))
1075 spin_lock_irq(lock
);
1077 spin_unlock_irq(lock
);
1080 * At this point all other threads have exited, all we have to
1081 * do is to wait for the thread group leader to become inactive,
1082 * and to assume its PID:
1084 if (!thread_group_leader(tsk
)) {
1085 struct task_struct
*leader
= tsk
->group_leader
;
1088 cgroup_threadgroup_change_begin(tsk
);
1089 write_lock_irq(&tasklist_lock
);
1091 * Do this under tasklist_lock to ensure that
1092 * exit_notify() can't miss ->group_exec_task
1094 sig
->notify_count
= -1;
1095 if (likely(leader
->exit_state
))
1097 __set_current_state(TASK_KILLABLE
);
1098 write_unlock_irq(&tasklist_lock
);
1099 cgroup_threadgroup_change_end(tsk
);
1101 if (__fatal_signal_pending(tsk
))
1106 * The only record we have of the real-time age of a
1107 * process, regardless of execs it's done, is start_time.
1108 * All the past CPU time is accumulated in signal_struct
1109 * from sister threads now dead. But in this non-leader
1110 * exec, nothing survives from the original leader thread,
1111 * whose birth marks the true age of this process now.
1112 * When we take on its identity by switching to its PID, we
1113 * also take its birthdate (always earlier than our own).
1115 tsk
->start_time
= leader
->start_time
;
1116 tsk
->start_boottime
= leader
->start_boottime
;
1118 BUG_ON(!same_thread_group(leader
, tsk
));
1120 * An exec() starts a new thread group with the
1121 * TGID of the previous thread group. Rehash the
1122 * two threads with a switched PID, and release
1123 * the former thread group leader:
1126 /* Become a process group leader with the old leader's pid.
1127 * The old leader becomes a thread of the this thread group.
1129 exchange_tids(tsk
, leader
);
1130 transfer_pid(leader
, tsk
, PIDTYPE_TGID
);
1131 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
1132 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
1134 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
1135 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
1137 tsk
->group_leader
= tsk
;
1138 leader
->group_leader
= tsk
;
1140 tsk
->exit_signal
= SIGCHLD
;
1141 leader
->exit_signal
= -1;
1143 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
1144 leader
->exit_state
= EXIT_DEAD
;
1147 * We are going to release_task()->ptrace_unlink() silently,
1148 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1149 * the tracer won't block again waiting for this thread.
1151 if (unlikely(leader
->ptrace
))
1152 __wake_up_parent(leader
, leader
->parent
);
1153 write_unlock_irq(&tasklist_lock
);
1154 cgroup_threadgroup_change_end(tsk
);
1156 release_task(leader
);
1159 sig
->group_exec_task
= NULL
;
1160 sig
->notify_count
= 0;
1163 /* we have changed execution domain */
1164 tsk
->exit_signal
= SIGCHLD
;
1166 BUG_ON(!thread_group_leader(tsk
));
1170 /* protects against exit_notify() and __exit_signal() */
1171 read_lock(&tasklist_lock
);
1172 sig
->group_exec_task
= NULL
;
1173 sig
->notify_count
= 0;
1174 read_unlock(&tasklist_lock
);
1180 * This function makes sure the current process has its own signal table,
1181 * so that flush_signal_handlers can later reset the handlers without
1182 * disturbing other processes. (Other processes might share the signal
1183 * table via the CLONE_SIGHAND option to clone().)
1185 static int unshare_sighand(struct task_struct
*me
)
1187 struct sighand_struct
*oldsighand
= me
->sighand
;
1189 if (refcount_read(&oldsighand
->count
) != 1) {
1190 struct sighand_struct
*newsighand
;
1192 * This ->sighand is shared with the CLONE_SIGHAND
1193 * but not CLONE_THREAD task, switch to the new one.
1195 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1199 refcount_set(&newsighand
->count
, 1);
1201 write_lock_irq(&tasklist_lock
);
1202 spin_lock(&oldsighand
->siglock
);
1203 memcpy(newsighand
->action
, oldsighand
->action
,
1204 sizeof(newsighand
->action
));
1205 rcu_assign_pointer(me
->sighand
, newsighand
);
1206 spin_unlock(&oldsighand
->siglock
);
1207 write_unlock_irq(&tasklist_lock
);
1209 __cleanup_sighand(oldsighand
);
1214 char *__get_task_comm(char *buf
, size_t buf_size
, struct task_struct
*tsk
)
1217 /* Always NUL terminated and zero-padded */
1218 strscpy_pad(buf
, tsk
->comm
, buf_size
);
1222 EXPORT_SYMBOL_GPL(__get_task_comm
);
1225 * These functions flushes out all traces of the currently running executable
1226 * so that a new one can be started
1229 void __set_task_comm(struct task_struct
*tsk
, const char *buf
, bool exec
)
1232 trace_task_rename(tsk
, buf
);
1233 strscpy_pad(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1235 perf_event_comm(tsk
, exec
);
1239 * Calling this is the point of no return. None of the failures will be
1240 * seen by userspace since either the process is already taking a fatal
1241 * signal (via de_thread() or coredump), or will have SEGV raised
1242 * (after exec_mmap()) by search_binary_handler (see below).
1244 int begin_new_exec(struct linux_binprm
* bprm
)
1246 struct task_struct
*me
= current
;
1249 /* Once we are committed compute the creds */
1250 retval
= bprm_creds_from_file(bprm
);
1255 * Ensure all future errors are fatal.
1257 bprm
->point_of_no_return
= true;
1260 * Make this the only thread in the thread group.
1262 retval
= de_thread(me
);
1267 * Cancel any io_uring activity across execve
1269 io_uring_task_cancel();
1271 /* Ensure the files table is not shared. */
1272 retval
= unshare_files();
1277 * Must be called _before_ exec_mmap() as bprm->mm is
1278 * not visible until then. Doing it here also ensures
1279 * we don't race against replace_mm_exe_file().
1281 retval
= set_mm_exe_file(bprm
->mm
, bprm
->file
);
1285 /* If the binary is not readable then enforce mm->dumpable=0 */
1286 would_dump(bprm
, bprm
->file
);
1287 if (bprm
->have_execfd
)
1288 would_dump(bprm
, bprm
->executable
);
1291 * Release all of the old mmap stuff
1293 acct_arg_size(bprm
, 0);
1294 retval
= exec_mmap(bprm
->mm
);
1300 retval
= exec_task_namespaces();
1304 #ifdef CONFIG_POSIX_TIMERS
1305 spin_lock_irq(&me
->sighand
->siglock
);
1306 posix_cpu_timers_exit(me
);
1307 spin_unlock_irq(&me
->sighand
->siglock
);
1309 flush_itimer_signals();
1313 * Make the signal table private.
1315 retval
= unshare_sighand(me
);
1319 me
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
|
1320 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1322 me
->personality
&= ~bprm
->per_clear
;
1324 clear_syscall_work_syscall_user_dispatch(me
);
1327 * We have to apply CLOEXEC before we change whether the process is
1328 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1329 * trying to access the should-be-closed file descriptors of a process
1330 * undergoing exec(2).
1332 do_close_on_exec(me
->files
);
1334 if (bprm
->secureexec
) {
1335 /* Make sure parent cannot signal privileged process. */
1336 me
->pdeath_signal
= 0;
1339 * For secureexec, reset the stack limit to sane default to
1340 * avoid bad behavior from the prior rlimits. This has to
1341 * happen before arch_pick_mmap_layout(), which examines
1342 * RLIMIT_STACK, but after the point of no return to avoid
1343 * needing to clean up the change on failure.
1345 if (bprm
->rlim_stack
.rlim_cur
> _STK_LIM
)
1346 bprm
->rlim_stack
.rlim_cur
= _STK_LIM
;
1349 me
->sas_ss_sp
= me
->sas_ss_size
= 0;
1352 * Figure out dumpability. Note that this checking only of current
1353 * is wrong, but userspace depends on it. This should be testing
1354 * bprm->secureexec instead.
1356 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
||
1357 !(uid_eq(current_euid(), current_uid()) &&
1358 gid_eq(current_egid(), current_gid())))
1359 set_dumpable(current
->mm
, suid_dumpable
);
1361 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1364 __set_task_comm(me
, kbasename(bprm
->filename
), true);
1366 /* An exec changes our domain. We are no longer part of the thread
1368 WRITE_ONCE(me
->self_exec_id
, me
->self_exec_id
+ 1);
1369 flush_signal_handlers(me
, 0);
1371 retval
= set_cred_ucounts(bprm
->cred
);
1376 * install the new credentials for this executable
1378 security_bprm_committing_creds(bprm
);
1380 commit_creds(bprm
->cred
);
1384 * Disable monitoring for regular users
1385 * when executing setuid binaries. Must
1386 * wait until new credentials are committed
1387 * by commit_creds() above
1389 if (get_dumpable(me
->mm
) != SUID_DUMP_USER
)
1390 perf_event_exit_task(me
);
1392 * cred_guard_mutex must be held at least to this point to prevent
1393 * ptrace_attach() from altering our determination of the task's
1394 * credentials; any time after this it may be unlocked.
1396 security_bprm_committed_creds(bprm
);
1398 /* Pass the opened binary to the interpreter. */
1399 if (bprm
->have_execfd
) {
1400 retval
= get_unused_fd_flags(0);
1403 fd_install(retval
, bprm
->executable
);
1404 bprm
->executable
= NULL
;
1405 bprm
->execfd
= retval
;
1410 up_write(&me
->signal
->exec_update_lock
);
1414 EXPORT_SYMBOL(begin_new_exec
);
1416 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1418 struct inode
*inode
= file_inode(file
);
1419 struct mnt_idmap
*idmap
= file_mnt_idmap(file
);
1420 if (inode_permission(idmap
, inode
, MAY_READ
) < 0) {
1421 struct user_namespace
*old
, *user_ns
;
1422 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1424 /* Ensure mm->user_ns contains the executable */
1425 user_ns
= old
= bprm
->mm
->user_ns
;
1426 while ((user_ns
!= &init_user_ns
) &&
1427 !privileged_wrt_inode_uidgid(user_ns
, idmap
, inode
))
1428 user_ns
= user_ns
->parent
;
1430 if (old
!= user_ns
) {
1431 bprm
->mm
->user_ns
= get_user_ns(user_ns
);
1436 EXPORT_SYMBOL(would_dump
);
1438 void setup_new_exec(struct linux_binprm
* bprm
)
1440 /* Setup things that can depend upon the personality */
1441 struct task_struct
*me
= current
;
1443 arch_pick_mmap_layout(me
->mm
, &bprm
->rlim_stack
);
1445 arch_setup_new_exec();
1447 /* Set the new mm task size. We have to do that late because it may
1448 * depend on TIF_32BIT which is only updated in flush_thread() on
1449 * some architectures like powerpc
1451 me
->mm
->task_size
= TASK_SIZE
;
1452 up_write(&me
->signal
->exec_update_lock
);
1453 mutex_unlock(&me
->signal
->cred_guard_mutex
);
1455 EXPORT_SYMBOL(setup_new_exec
);
1457 /* Runs immediately before start_thread() takes over. */
1458 void finalize_exec(struct linux_binprm
*bprm
)
1460 /* Store any stack rlimit changes before starting thread. */
1461 task_lock(current
->group_leader
);
1462 current
->signal
->rlim
[RLIMIT_STACK
] = bprm
->rlim_stack
;
1463 task_unlock(current
->group_leader
);
1465 EXPORT_SYMBOL(finalize_exec
);
1468 * Prepare credentials and lock ->cred_guard_mutex.
1469 * setup_new_exec() commits the new creds and drops the lock.
1470 * Or, if exec fails before, free_bprm() should release ->cred
1473 static int prepare_bprm_creds(struct linux_binprm
*bprm
)
1475 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1476 return -ERESTARTNOINTR
;
1478 bprm
->cred
= prepare_exec_creds();
1479 if (likely(bprm
->cred
))
1482 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1486 static void free_bprm(struct linux_binprm
*bprm
)
1489 acct_arg_size(bprm
, 0);
1492 free_arg_pages(bprm
);
1494 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1495 abort_creds(bprm
->cred
);
1498 allow_write_access(bprm
->file
);
1501 if (bprm
->executable
)
1502 fput(bprm
->executable
);
1503 /* If a binfmt changed the interp, free it. */
1504 if (bprm
->interp
!= bprm
->filename
)
1505 kfree(bprm
->interp
);
1506 kfree(bprm
->fdpath
);
1510 static struct linux_binprm
*alloc_bprm(int fd
, struct filename
*filename
)
1512 struct linux_binprm
*bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1513 int retval
= -ENOMEM
;
1517 if (fd
== AT_FDCWD
|| filename
->name
[0] == '/') {
1518 bprm
->filename
= filename
->name
;
1520 if (filename
->name
[0] == '\0')
1521 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d", fd
);
1523 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d/%s",
1524 fd
, filename
->name
);
1528 bprm
->filename
= bprm
->fdpath
;
1530 bprm
->interp
= bprm
->filename
;
1532 retval
= bprm_mm_init(bprm
);
1540 return ERR_PTR(retval
);
1543 int bprm_change_interp(const char *interp
, struct linux_binprm
*bprm
)
1545 /* If a binfmt changed the interp, free it first. */
1546 if (bprm
->interp
!= bprm
->filename
)
1547 kfree(bprm
->interp
);
1548 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1553 EXPORT_SYMBOL(bprm_change_interp
);
1556 * determine how safe it is to execute the proposed program
1557 * - the caller must hold ->cred_guard_mutex to protect against
1558 * PTRACE_ATTACH or seccomp thread-sync
1560 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1562 struct task_struct
*p
= current
, *t
;
1566 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1569 * This isn't strictly necessary, but it makes it harder for LSMs to
1572 if (task_no_new_privs(current
))
1573 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1576 * If another task is sharing our fs, we cannot safely
1577 * suid exec because the differently privileged task
1578 * will be able to manipulate the current directory, etc.
1579 * It would be nice to force an unshare instead...
1583 spin_lock(&p
->fs
->lock
);
1585 while_each_thread(p
, t
) {
1591 if (p
->fs
->users
> n_fs
)
1592 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1595 spin_unlock(&p
->fs
->lock
);
1598 static void bprm_fill_uid(struct linux_binprm
*bprm
, struct file
*file
)
1600 /* Handle suid and sgid on files */
1601 struct mnt_idmap
*idmap
;
1602 struct inode
*inode
= file_inode(file
);
1607 if (!mnt_may_suid(file
->f_path
.mnt
))
1610 if (task_no_new_privs(current
))
1613 mode
= READ_ONCE(inode
->i_mode
);
1614 if (!(mode
& (S_ISUID
|S_ISGID
)))
1617 idmap
= file_mnt_idmap(file
);
1619 /* Be careful if suid/sgid is set */
1622 /* reload atomically mode/uid/gid now that lock held */
1623 mode
= inode
->i_mode
;
1624 vfsuid
= i_uid_into_vfsuid(idmap
, inode
);
1625 vfsgid
= i_gid_into_vfsgid(idmap
, inode
);
1626 inode_unlock(inode
);
1628 /* We ignore suid/sgid if there are no mappings for them in the ns */
1629 if (!vfsuid_has_mapping(bprm
->cred
->user_ns
, vfsuid
) ||
1630 !vfsgid_has_mapping(bprm
->cred
->user_ns
, vfsgid
))
1633 if (mode
& S_ISUID
) {
1634 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1635 bprm
->cred
->euid
= vfsuid_into_kuid(vfsuid
);
1638 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1639 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1640 bprm
->cred
->egid
= vfsgid_into_kgid(vfsgid
);
1645 * Compute brpm->cred based upon the final binary.
1647 static int bprm_creds_from_file(struct linux_binprm
*bprm
)
1649 /* Compute creds based on which file? */
1650 struct file
*file
= bprm
->execfd_creds
? bprm
->executable
: bprm
->file
;
1652 bprm_fill_uid(bprm
, file
);
1653 return security_bprm_creds_from_file(bprm
, file
);
1657 * Fill the binprm structure from the inode.
1658 * Read the first BINPRM_BUF_SIZE bytes
1660 * This may be called multiple times for binary chains (scripts for example).
1662 static int prepare_binprm(struct linux_binprm
*bprm
)
1666 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1667 return kernel_read(bprm
->file
, bprm
->buf
, BINPRM_BUF_SIZE
, &pos
);
1671 * Arguments are '\0' separated strings found at the location bprm->p
1672 * points to; chop off the first by relocating brpm->p to right after
1673 * the first '\0' encountered.
1675 int remove_arg_zero(struct linux_binprm
*bprm
)
1678 unsigned long offset
;
1686 offset
= bprm
->p
& ~PAGE_MASK
;
1687 page
= get_arg_page(bprm
, bprm
->p
, 0);
1692 kaddr
= kmap_local_page(page
);
1694 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1695 offset
++, bprm
->p
++)
1698 kunmap_local(kaddr
);
1700 } while (offset
== PAGE_SIZE
);
1709 EXPORT_SYMBOL(remove_arg_zero
);
1711 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1713 * cycle the list of binary formats handler, until one recognizes the image
1715 static int search_binary_handler(struct linux_binprm
*bprm
)
1717 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1718 struct linux_binfmt
*fmt
;
1721 retval
= prepare_binprm(bprm
);
1725 retval
= security_bprm_check(bprm
);
1731 read_lock(&binfmt_lock
);
1732 list_for_each_entry(fmt
, &formats
, lh
) {
1733 if (!try_module_get(fmt
->module
))
1735 read_unlock(&binfmt_lock
);
1737 retval
= fmt
->load_binary(bprm
);
1739 read_lock(&binfmt_lock
);
1741 if (bprm
->point_of_no_return
|| (retval
!= -ENOEXEC
)) {
1742 read_unlock(&binfmt_lock
);
1746 read_unlock(&binfmt_lock
);
1749 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1750 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1752 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1761 /* binfmt handlers will call back into begin_new_exec() on success. */
1762 static int exec_binprm(struct linux_binprm
*bprm
)
1764 pid_t old_pid
, old_vpid
;
1767 /* Need to fetch pid before load_binary changes it */
1768 old_pid
= current
->pid
;
1770 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1773 /* This allows 4 levels of binfmt rewrites before failing hard. */
1774 for (depth
= 0;; depth
++) {
1779 ret
= search_binary_handler(bprm
);
1782 if (!bprm
->interpreter
)
1786 bprm
->file
= bprm
->interpreter
;
1787 bprm
->interpreter
= NULL
;
1789 allow_write_access(exec
);
1790 if (unlikely(bprm
->have_execfd
)) {
1791 if (bprm
->executable
) {
1795 bprm
->executable
= exec
;
1801 trace_sched_process_exec(current
, old_pid
, bprm
);
1802 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1803 proc_exec_connector(current
);
1808 * sys_execve() executes a new program.
1810 static int bprm_execve(struct linux_binprm
*bprm
,
1811 int fd
, struct filename
*filename
, int flags
)
1816 retval
= prepare_bprm_creds(bprm
);
1821 * Check for unsafe execution states before exec_binprm(), which
1822 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1823 * where setuid-ness is evaluated.
1825 check_unsafe_exec(bprm
);
1826 current
->in_execve
= 1;
1827 sched_mm_cid_before_execve(current
);
1829 file
= do_open_execat(fd
, filename
, flags
);
1830 retval
= PTR_ERR(file
);
1838 * Record that a name derived from an O_CLOEXEC fd will be
1839 * inaccessible after exec. This allows the code in exec to
1840 * choose to fail when the executable is not mmaped into the
1841 * interpreter and an open file descriptor is not passed to
1842 * the interpreter. This makes for a better user experience
1843 * than having the interpreter start and then immediately fail
1844 * when it finds the executable is inaccessible.
1846 if (bprm
->fdpath
&& get_close_on_exec(fd
))
1847 bprm
->interp_flags
|= BINPRM_FLAGS_PATH_INACCESSIBLE
;
1849 /* Set the unchanging part of bprm->cred */
1850 retval
= security_bprm_creds_for_exec(bprm
);
1854 retval
= exec_binprm(bprm
);
1858 sched_mm_cid_after_execve(current
);
1859 /* execve succeeded */
1860 current
->fs
->in_exec
= 0;
1861 current
->in_execve
= 0;
1862 rseq_execve(current
);
1863 user_events_execve(current
);
1864 acct_update_integrals(current
);
1865 task_numa_free(current
, false);
1870 * If past the point of no return ensure the code never
1871 * returns to the userspace process. Use an existing fatal
1872 * signal if present otherwise terminate the process with
1875 if (bprm
->point_of_no_return
&& !fatal_signal_pending(current
))
1876 force_fatal_sig(SIGSEGV
);
1879 sched_mm_cid_after_execve(current
);
1880 current
->fs
->in_exec
= 0;
1881 current
->in_execve
= 0;
1886 static int do_execveat_common(int fd
, struct filename
*filename
,
1887 struct user_arg_ptr argv
,
1888 struct user_arg_ptr envp
,
1891 struct linux_binprm
*bprm
;
1894 if (IS_ERR(filename
))
1895 return PTR_ERR(filename
);
1898 * We move the actual failure in case of RLIMIT_NPROC excess from
1899 * set*uid() to execve() because too many poorly written programs
1900 * don't check setuid() return code. Here we additionally recheck
1901 * whether NPROC limit is still exceeded.
1903 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1904 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC
, rlimit(RLIMIT_NPROC
))) {
1909 /* We're below the limit (still or again), so we don't want to make
1910 * further execve() calls fail. */
1911 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1913 bprm
= alloc_bprm(fd
, filename
);
1915 retval
= PTR_ERR(bprm
);
1919 retval
= count(argv
, MAX_ARG_STRINGS
);
1921 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1922 current
->comm
, bprm
->filename
);
1925 bprm
->argc
= retval
;
1927 retval
= count(envp
, MAX_ARG_STRINGS
);
1930 bprm
->envc
= retval
;
1932 retval
= bprm_stack_limits(bprm
);
1936 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1939 bprm
->exec
= bprm
->p
;
1941 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1945 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1950 * When argv is empty, add an empty string ("") as argv[0] to
1951 * ensure confused userspace programs that start processing
1952 * from argv[1] won't end up walking envp. See also
1953 * bprm_stack_limits().
1955 if (bprm
->argc
== 0) {
1956 retval
= copy_string_kernel("", bprm
);
1962 retval
= bprm_execve(bprm
, fd
, filename
, flags
);
1971 int kernel_execve(const char *kernel_filename
,
1972 const char *const *argv
, const char *const *envp
)
1974 struct filename
*filename
;
1975 struct linux_binprm
*bprm
;
1979 /* It is non-sense for kernel threads to call execve */
1980 if (WARN_ON_ONCE(current
->flags
& PF_KTHREAD
))
1983 filename
= getname_kernel(kernel_filename
);
1984 if (IS_ERR(filename
))
1985 return PTR_ERR(filename
);
1987 bprm
= alloc_bprm(fd
, filename
);
1989 retval
= PTR_ERR(bprm
);
1993 retval
= count_strings_kernel(argv
);
1994 if (WARN_ON_ONCE(retval
== 0))
1998 bprm
->argc
= retval
;
2000 retval
= count_strings_kernel(envp
);
2003 bprm
->envc
= retval
;
2005 retval
= bprm_stack_limits(bprm
);
2009 retval
= copy_string_kernel(bprm
->filename
, bprm
);
2012 bprm
->exec
= bprm
->p
;
2014 retval
= copy_strings_kernel(bprm
->envc
, envp
, bprm
);
2018 retval
= copy_strings_kernel(bprm
->argc
, argv
, bprm
);
2022 retval
= bprm_execve(bprm
, fd
, filename
, 0);
2030 static int do_execve(struct filename
*filename
,
2031 const char __user
*const __user
*__argv
,
2032 const char __user
*const __user
*__envp
)
2034 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2035 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2036 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2039 static int do_execveat(int fd
, struct filename
*filename
,
2040 const char __user
*const __user
*__argv
,
2041 const char __user
*const __user
*__envp
,
2044 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2045 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2047 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2050 #ifdef CONFIG_COMPAT
2051 static int compat_do_execve(struct filename
*filename
,
2052 const compat_uptr_t __user
*__argv
,
2053 const compat_uptr_t __user
*__envp
)
2055 struct user_arg_ptr argv
= {
2057 .ptr
.compat
= __argv
,
2059 struct user_arg_ptr envp
= {
2061 .ptr
.compat
= __envp
,
2063 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2066 static int compat_do_execveat(int fd
, struct filename
*filename
,
2067 const compat_uptr_t __user
*__argv
,
2068 const compat_uptr_t __user
*__envp
,
2071 struct user_arg_ptr argv
= {
2073 .ptr
.compat
= __argv
,
2075 struct user_arg_ptr envp
= {
2077 .ptr
.compat
= __envp
,
2079 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2083 void set_binfmt(struct linux_binfmt
*new)
2085 struct mm_struct
*mm
= current
->mm
;
2088 module_put(mm
->binfmt
->module
);
2092 __module_get(new->module
);
2094 EXPORT_SYMBOL(set_binfmt
);
2097 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2099 void set_dumpable(struct mm_struct
*mm
, int value
)
2101 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
2104 set_mask_bits(&mm
->flags
, MMF_DUMPABLE_MASK
, value
);
2107 SYSCALL_DEFINE3(execve
,
2108 const char __user
*, filename
,
2109 const char __user
*const __user
*, argv
,
2110 const char __user
*const __user
*, envp
)
2112 return do_execve(getname(filename
), argv
, envp
);
2115 SYSCALL_DEFINE5(execveat
,
2116 int, fd
, const char __user
*, filename
,
2117 const char __user
*const __user
*, argv
,
2118 const char __user
*const __user
*, envp
,
2121 return do_execveat(fd
,
2122 getname_uflags(filename
, flags
),
2126 #ifdef CONFIG_COMPAT
2127 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
2128 const compat_uptr_t __user
*, argv
,
2129 const compat_uptr_t __user
*, envp
)
2131 return compat_do_execve(getname(filename
), argv
, envp
);
2134 COMPAT_SYSCALL_DEFINE5(execveat
, int, fd
,
2135 const char __user
*, filename
,
2136 const compat_uptr_t __user
*, argv
,
2137 const compat_uptr_t __user
*, envp
,
2140 return compat_do_execveat(fd
,
2141 getname_uflags(filename
, flags
),
2146 #ifdef CONFIG_SYSCTL
2148 static int proc_dointvec_minmax_coredump(struct ctl_table
*table
, int write
,
2149 void *buffer
, size_t *lenp
, loff_t
*ppos
)
2151 int error
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
2154 validate_coredump_safety();
2158 static struct ctl_table fs_exec_sysctls
[] = {
2160 .procname
= "suid_dumpable",
2161 .data
= &suid_dumpable
,
2162 .maxlen
= sizeof(int),
2164 .proc_handler
= proc_dointvec_minmax_coredump
,
2165 .extra1
= SYSCTL_ZERO
,
2166 .extra2
= SYSCTL_TWO
,
2171 static int __init
init_fs_exec_sysctls(void)
2173 register_sysctl_init("fs", fs_exec_sysctls
);
2177 fs_initcall(init_fs_exec_sysctls
);
2178 #endif /* CONFIG_SYSCTL */