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/vmacache.h>
32 #include <linux/stat.h>
33 #include <linux/fcntl.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/signal.h>
40 #include <linux/sched/numa_balancing.h>
41 #include <linux/sched/task.h>
42 #include <linux/pagemap.h>
43 #include <linux/perf_event.h>
44 #include <linux/highmem.h>
45 #include <linux/spinlock.h>
46 #include <linux/key.h>
47 #include <linux/personality.h>
48 #include <linux/binfmts.h>
49 #include <linux/utsname.h>
50 #include <linux/pid_namespace.h>
51 #include <linux/module.h>
52 #include <linux/namei.h>
53 #include <linux/mount.h>
54 #include <linux/security.h>
55 #include <linux/syscalls.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/audit.h>
59 #include <linux/tracehook.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>
69 #include <linux/uaccess.h>
70 #include <asm/mmu_context.h>
73 #include <trace/events/task.h>
76 #include <trace/events/sched.h>
78 static int bprm_creds_from_file(struct linux_binprm
*bprm
);
80 int suid_dumpable
= 0;
82 static LIST_HEAD(formats
);
83 static DEFINE_RWLOCK(binfmt_lock
);
85 void __register_binfmt(struct linux_binfmt
* fmt
, int insert
)
87 write_lock(&binfmt_lock
);
88 insert
? list_add(&fmt
->lh
, &formats
) :
89 list_add_tail(&fmt
->lh
, &formats
);
90 write_unlock(&binfmt_lock
);
93 EXPORT_SYMBOL(__register_binfmt
);
95 void unregister_binfmt(struct linux_binfmt
* fmt
)
97 write_lock(&binfmt_lock
);
99 write_unlock(&binfmt_lock
);
102 EXPORT_SYMBOL(unregister_binfmt
);
104 static inline void put_binfmt(struct linux_binfmt
* fmt
)
106 module_put(fmt
->module
);
109 bool path_noexec(const struct path
*path
)
111 return (path
->mnt
->mnt_flags
& MNT_NOEXEC
) ||
112 (path
->mnt
->mnt_sb
->s_iflags
& SB_I_NOEXEC
);
117 * Note that a shared library must be both readable and executable due to
120 * Also note that we take the address to load from from the file itself.
122 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
124 struct linux_binfmt
*fmt
;
126 struct filename
*tmp
= getname(library
);
127 int error
= PTR_ERR(tmp
);
128 static const struct open_flags uselib_flags
= {
129 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
130 .acc_mode
= MAY_READ
| MAY_EXEC
,
131 .intent
= LOOKUP_OPEN
,
132 .lookup_flags
= LOOKUP_FOLLOW
,
138 file
= do_filp_open(AT_FDCWD
, tmp
, &uselib_flags
);
140 error
= PTR_ERR(file
);
145 * may_open() has already checked for this, so it should be
146 * impossible to trip now. But we need to be extra cautious
147 * and check again at the very end too.
150 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
151 path_noexec(&file
->f_path
)))
158 read_lock(&binfmt_lock
);
159 list_for_each_entry(fmt
, &formats
, lh
) {
160 if (!fmt
->load_shlib
)
162 if (!try_module_get(fmt
->module
))
164 read_unlock(&binfmt_lock
);
165 error
= fmt
->load_shlib(file
);
166 read_lock(&binfmt_lock
);
168 if (error
!= -ENOEXEC
)
171 read_unlock(&binfmt_lock
);
177 #endif /* #ifdef CONFIG_USELIB */
181 * The nascent bprm->mm is not visible until exec_mmap() but it can
182 * use a lot of memory, account these pages in current->mm temporary
183 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
184 * change the counter back via acct_arg_size(0).
186 static void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
188 struct mm_struct
*mm
= current
->mm
;
189 long diff
= (long)(pages
- bprm
->vma_pages
);
194 bprm
->vma_pages
= pages
;
195 add_mm_counter(mm
, MM_ANONPAGES
, diff
);
198 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
203 unsigned int gup_flags
= FOLL_FORCE
;
205 #ifdef CONFIG_STACK_GROWSUP
207 ret
= expand_downwards(bprm
->vma
, pos
);
214 gup_flags
|= FOLL_WRITE
;
217 * We are doing an exec(). 'current' is the process
218 * doing the exec and bprm->mm is the new process's mm.
220 ret
= get_user_pages_remote(bprm
->mm
, pos
, 1, gup_flags
,
226 acct_arg_size(bprm
, vma_pages(bprm
->vma
));
231 static void put_arg_page(struct page
*page
)
236 static void free_arg_pages(struct linux_binprm
*bprm
)
240 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
243 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
246 static int __bprm_mm_init(struct linux_binprm
*bprm
)
249 struct vm_area_struct
*vma
= NULL
;
250 struct mm_struct
*mm
= bprm
->mm
;
252 bprm
->vma
= vma
= vm_area_alloc(mm
);
255 vma_set_anonymous(vma
);
257 if (mmap_write_lock_killable(mm
)) {
263 * Place the stack at the largest stack address the architecture
264 * supports. Later, we'll move this to an appropriate place. We don't
265 * use STACK_TOP because that can depend on attributes which aren't
268 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
269 vma
->vm_end
= STACK_TOP_MAX
;
270 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
271 vma
->vm_flags
= VM_SOFTDIRTY
| VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
;
272 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
274 err
= insert_vm_struct(mm
, vma
);
278 mm
->stack_vm
= mm
->total_vm
= 1;
279 mmap_write_unlock(mm
);
280 bprm
->p
= vma
->vm_end
- sizeof(void *);
283 mmap_write_unlock(mm
);
290 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
292 return len
<= MAX_ARG_STRLEN
;
297 static inline void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
301 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
306 page
= bprm
->page
[pos
/ PAGE_SIZE
];
307 if (!page
&& write
) {
308 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
311 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
317 static void put_arg_page(struct page
*page
)
321 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
324 __free_page(bprm
->page
[i
]);
325 bprm
->page
[i
] = NULL
;
329 static void free_arg_pages(struct linux_binprm
*bprm
)
333 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
334 free_arg_page(bprm
, i
);
337 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
342 static int __bprm_mm_init(struct linux_binprm
*bprm
)
344 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
348 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
350 return len
<= bprm
->p
;
353 #endif /* CONFIG_MMU */
356 * Create a new mm_struct and populate it with a temporary stack
357 * vm_area_struct. We don't have enough context at this point to set the stack
358 * flags, permissions, and offset, so we use temporary values. We'll update
359 * them later in setup_arg_pages().
361 static int bprm_mm_init(struct linux_binprm
*bprm
)
364 struct mm_struct
*mm
= NULL
;
366 bprm
->mm
= mm
= mm_alloc();
371 /* Save current stack limit for all calculations made during exec. */
372 task_lock(current
->group_leader
);
373 bprm
->rlim_stack
= current
->signal
->rlim
[RLIMIT_STACK
];
374 task_unlock(current
->group_leader
);
376 err
= __bprm_mm_init(bprm
);
391 struct user_arg_ptr
{
396 const char __user
*const __user
*native
;
398 const compat_uptr_t __user
*compat
;
403 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
405 const char __user
*native
;
408 if (unlikely(argv
.is_compat
)) {
409 compat_uptr_t compat
;
411 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
412 return ERR_PTR(-EFAULT
);
414 return compat_ptr(compat
);
418 if (get_user(native
, argv
.ptr
.native
+ nr
))
419 return ERR_PTR(-EFAULT
);
425 * count() counts the number of strings in array ARGV.
427 static int count(struct user_arg_ptr argv
, int max
)
431 if (argv
.ptr
.native
!= NULL
) {
433 const char __user
*p
= get_user_arg_ptr(argv
, i
);
445 if (fatal_signal_pending(current
))
446 return -ERESTARTNOHAND
;
453 static int count_strings_kernel(const char *const *argv
)
460 for (i
= 0; argv
[i
]; ++i
) {
461 if (i
>= MAX_ARG_STRINGS
)
463 if (fatal_signal_pending(current
))
464 return -ERESTARTNOHAND
;
470 static int bprm_stack_limits(struct linux_binprm
*bprm
)
472 unsigned long limit
, ptr_size
;
475 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
476 * (whichever is smaller) for the argv+env strings.
478 * - the remaining binfmt code will not run out of stack space,
479 * - the program will have a reasonable amount of stack left
482 limit
= _STK_LIM
/ 4 * 3;
483 limit
= min(limit
, bprm
->rlim_stack
.rlim_cur
/ 4);
485 * We've historically supported up to 32 pages (ARG_MAX)
486 * of argument strings even with small stacks
488 limit
= max_t(unsigned long, limit
, ARG_MAX
);
490 * We must account for the size of all the argv and envp pointers to
491 * the argv and envp strings, since they will also take up space in
492 * the stack. They aren't stored until much later when we can't
493 * signal to the parent that the child has run out of stack space.
494 * Instead, calculate it here so it's possible to fail gracefully.
496 ptr_size
= (bprm
->argc
+ bprm
->envc
) * sizeof(void *);
497 if (limit
<= ptr_size
)
501 bprm
->argmin
= bprm
->p
- limit
;
506 * 'copy_strings()' copies argument/environment strings from the old
507 * processes's memory to the new process's stack. The call to get_user_pages()
508 * ensures the destination page is created and not swapped out.
510 static int copy_strings(int argc
, struct user_arg_ptr argv
,
511 struct linux_binprm
*bprm
)
513 struct page
*kmapped_page
= NULL
;
515 unsigned long kpos
= 0;
519 const char __user
*str
;
524 str
= get_user_arg_ptr(argv
, argc
);
528 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
533 if (!valid_arg_len(bprm
, len
))
536 /* We're going to work our way backwords. */
541 if (bprm
->p
< bprm
->argmin
)
546 int offset
, bytes_to_copy
;
548 if (fatal_signal_pending(current
)) {
549 ret
= -ERESTARTNOHAND
;
554 offset
= pos
% PAGE_SIZE
;
558 bytes_to_copy
= offset
;
559 if (bytes_to_copy
> len
)
562 offset
-= bytes_to_copy
;
563 pos
-= bytes_to_copy
;
564 str
-= bytes_to_copy
;
565 len
-= bytes_to_copy
;
567 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
570 page
= get_arg_page(bprm
, pos
, 1);
577 flush_kernel_dcache_page(kmapped_page
);
578 kunmap(kmapped_page
);
579 put_arg_page(kmapped_page
);
582 kaddr
= kmap(kmapped_page
);
583 kpos
= pos
& PAGE_MASK
;
584 flush_arg_page(bprm
, kpos
, kmapped_page
);
586 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
595 flush_kernel_dcache_page(kmapped_page
);
596 kunmap(kmapped_page
);
597 put_arg_page(kmapped_page
);
603 * Copy and argument/environment string from the kernel to the processes stack.
605 int copy_string_kernel(const char *arg
, struct linux_binprm
*bprm
)
607 int len
= strnlen(arg
, MAX_ARG_STRLEN
) + 1 /* terminating NUL */;
608 unsigned long pos
= bprm
->p
;
612 if (!valid_arg_len(bprm
, len
))
615 /* We're going to work our way backwards. */
618 if (IS_ENABLED(CONFIG_MMU
) && bprm
->p
< bprm
->argmin
)
622 unsigned int bytes_to_copy
= min_t(unsigned int, len
,
623 min_not_zero(offset_in_page(pos
), PAGE_SIZE
));
627 pos
-= bytes_to_copy
;
628 arg
-= bytes_to_copy
;
629 len
-= bytes_to_copy
;
631 page
= get_arg_page(bprm
, pos
, 1);
634 kaddr
= kmap_atomic(page
);
635 flush_arg_page(bprm
, pos
& PAGE_MASK
, page
);
636 memcpy(kaddr
+ offset_in_page(pos
), arg
, bytes_to_copy
);
637 flush_kernel_dcache_page(page
);
638 kunmap_atomic(kaddr
);
644 EXPORT_SYMBOL(copy_string_kernel
);
646 static int copy_strings_kernel(int argc
, const char *const *argv
,
647 struct linux_binprm
*bprm
)
650 int ret
= copy_string_kernel(argv
[argc
], bprm
);
653 if (fatal_signal_pending(current
))
654 return -ERESTARTNOHAND
;
663 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
664 * the binfmt code determines where the new stack should reside, we shift it to
665 * its final location. The process proceeds as follows:
667 * 1) Use shift to calculate the new vma endpoints.
668 * 2) Extend vma to cover both the old and new ranges. This ensures the
669 * arguments passed to subsequent functions are consistent.
670 * 3) Move vma's page tables to the new range.
671 * 4) Free up any cleared pgd range.
672 * 5) Shrink the vma to cover only the new range.
674 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
676 struct mm_struct
*mm
= vma
->vm_mm
;
677 unsigned long old_start
= vma
->vm_start
;
678 unsigned long old_end
= vma
->vm_end
;
679 unsigned long length
= old_end
- old_start
;
680 unsigned long new_start
= old_start
- shift
;
681 unsigned long new_end
= old_end
- shift
;
682 struct mmu_gather tlb
;
684 BUG_ON(new_start
> new_end
);
687 * ensure there are no vmas between where we want to go
690 if (vma
!= find_vma(mm
, new_start
))
694 * cover the whole range: [new_start, old_end)
696 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
700 * move the page tables downwards, on failure we rely on
701 * process cleanup to remove whatever mess we made.
703 if (length
!= move_page_tables(vma
, old_start
,
704 vma
, new_start
, length
, false))
708 tlb_gather_mmu(&tlb
, mm
);
709 if (new_end
> old_start
) {
711 * when the old and new regions overlap clear from new_end.
713 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
714 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
717 * otherwise, clean from old_start; this is done to not touch
718 * the address space in [new_end, old_start) some architectures
719 * have constraints on va-space that make this illegal (IA64) -
720 * for the others its just a little faster.
722 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
723 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
725 tlb_finish_mmu(&tlb
);
728 * Shrink the vma to just the new range. Always succeeds.
730 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
736 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
737 * the stack is optionally relocated, and some extra space is added.
739 int setup_arg_pages(struct linux_binprm
*bprm
,
740 unsigned long stack_top
,
741 int executable_stack
)
744 unsigned long stack_shift
;
745 struct mm_struct
*mm
= current
->mm
;
746 struct vm_area_struct
*vma
= bprm
->vma
;
747 struct vm_area_struct
*prev
= NULL
;
748 unsigned long vm_flags
;
749 unsigned long stack_base
;
750 unsigned long stack_size
;
751 unsigned long stack_expand
;
752 unsigned long rlim_stack
;
754 #ifdef CONFIG_STACK_GROWSUP
755 /* Limit stack size */
756 stack_base
= bprm
->rlim_stack
.rlim_max
;
758 stack_base
= calc_max_stack_size(stack_base
);
760 /* Add space for stack randomization. */
761 stack_base
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
763 /* Make sure we didn't let the argument array grow too large. */
764 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
767 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
769 stack_shift
= vma
->vm_start
- stack_base
;
770 mm
->arg_start
= bprm
->p
- stack_shift
;
771 bprm
->p
= vma
->vm_end
- stack_shift
;
773 stack_top
= arch_align_stack(stack_top
);
774 stack_top
= PAGE_ALIGN(stack_top
);
776 if (unlikely(stack_top
< mmap_min_addr
) ||
777 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
780 stack_shift
= vma
->vm_end
- stack_top
;
782 bprm
->p
-= stack_shift
;
783 mm
->arg_start
= bprm
->p
;
787 bprm
->loader
-= stack_shift
;
788 bprm
->exec
-= stack_shift
;
790 if (mmap_write_lock_killable(mm
))
793 vm_flags
= VM_STACK_FLAGS
;
796 * Adjust stack execute permissions; explicitly enable for
797 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
798 * (arch default) otherwise.
800 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
802 else if (executable_stack
== EXSTACK_DISABLE_X
)
803 vm_flags
&= ~VM_EXEC
;
804 vm_flags
|= mm
->def_flags
;
805 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
807 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
813 if (unlikely(vm_flags
& VM_EXEC
)) {
814 pr_warn_once("process '%pD4' started with executable stack\n",
818 /* Move stack pages down in memory. */
820 ret
= shift_arg_pages(vma
, stack_shift
);
825 /* mprotect_fixup is overkill to remove the temporary stack flags */
826 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
828 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
829 stack_size
= vma
->vm_end
- vma
->vm_start
;
831 * Align this down to a page boundary as expand_stack
834 rlim_stack
= bprm
->rlim_stack
.rlim_cur
& PAGE_MASK
;
835 #ifdef CONFIG_STACK_GROWSUP
836 if (stack_size
+ stack_expand
> rlim_stack
)
837 stack_base
= vma
->vm_start
+ rlim_stack
;
839 stack_base
= vma
->vm_end
+ stack_expand
;
841 if (stack_size
+ stack_expand
> rlim_stack
)
842 stack_base
= vma
->vm_end
- rlim_stack
;
844 stack_base
= vma
->vm_start
- stack_expand
;
846 current
->mm
->start_stack
= bprm
->p
;
847 ret
= expand_stack(vma
, stack_base
);
852 mmap_write_unlock(mm
);
855 EXPORT_SYMBOL(setup_arg_pages
);
860 * Transfer the program arguments and environment from the holding pages
861 * onto the stack. The provided stack pointer is adjusted accordingly.
863 int transfer_args_to_stack(struct linux_binprm
*bprm
,
864 unsigned long *sp_location
)
866 unsigned long index
, stop
, sp
;
869 stop
= bprm
->p
>> PAGE_SHIFT
;
872 for (index
= MAX_ARG_PAGES
- 1; index
>= stop
; index
--) {
873 unsigned int offset
= index
== stop
? bprm
->p
& ~PAGE_MASK
: 0;
874 char *src
= kmap(bprm
->page
[index
]) + offset
;
875 sp
-= PAGE_SIZE
- offset
;
876 if (copy_to_user((void *) sp
, src
, PAGE_SIZE
- offset
) != 0)
878 kunmap(bprm
->page
[index
]);
888 EXPORT_SYMBOL(transfer_args_to_stack
);
890 #endif /* CONFIG_MMU */
892 static struct file
*do_open_execat(int fd
, struct filename
*name
, int flags
)
896 struct open_flags open_exec_flags
= {
897 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
898 .acc_mode
= MAY_EXEC
,
899 .intent
= LOOKUP_OPEN
,
900 .lookup_flags
= LOOKUP_FOLLOW
,
903 if ((flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
904 return ERR_PTR(-EINVAL
);
905 if (flags
& AT_SYMLINK_NOFOLLOW
)
906 open_exec_flags
.lookup_flags
&= ~LOOKUP_FOLLOW
;
907 if (flags
& AT_EMPTY_PATH
)
908 open_exec_flags
.lookup_flags
|= LOOKUP_EMPTY
;
910 file
= do_filp_open(fd
, name
, &open_exec_flags
);
915 * may_open() has already checked for this, so it should be
916 * impossible to trip now. But we need to be extra cautious
917 * and check again at the very end too.
920 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
921 path_noexec(&file
->f_path
)))
924 err
= deny_write_access(file
);
928 if (name
->name
[0] != '\0')
939 struct file
*open_exec(const char *name
)
941 struct filename
*filename
= getname_kernel(name
);
942 struct file
*f
= ERR_CAST(filename
);
944 if (!IS_ERR(filename
)) {
945 f
= do_open_execat(AT_FDCWD
, filename
, 0);
950 EXPORT_SYMBOL(open_exec
);
952 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
953 defined(CONFIG_BINFMT_ELF_FDPIC)
954 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
956 ssize_t res
= vfs_read(file
, (void __user
*)addr
, len
, &pos
);
958 flush_icache_user_range(addr
, addr
+ len
);
961 EXPORT_SYMBOL(read_code
);
965 * Maps the mm_struct mm into the current task struct.
966 * On success, this function returns with exec_update_lock
969 static int exec_mmap(struct mm_struct
*mm
)
971 struct task_struct
*tsk
;
972 struct mm_struct
*old_mm
, *active_mm
;
975 /* Notify parent that we're no longer interested in the old VM */
977 old_mm
= current
->mm
;
978 exec_mm_release(tsk
, old_mm
);
982 ret
= down_write_killable(&tsk
->signal
->exec_update_lock
);
988 * Make sure that if there is a core dump in progress
989 * for the old mm, we get out and die instead of going
990 * through with the exec. We must hold mmap_lock around
991 * checking core_state and changing tsk->mm.
993 mmap_read_lock(old_mm
);
994 if (unlikely(old_mm
->core_state
)) {
995 mmap_read_unlock(old_mm
);
996 up_write(&tsk
->signal
->exec_update_lock
);
1002 membarrier_exec_mmap(mm
);
1004 local_irq_disable();
1005 active_mm
= tsk
->active_mm
;
1006 tsk
->active_mm
= mm
;
1009 * This prevents preemption while active_mm is being loaded and
1010 * it and mm are being updated, which could cause problems for
1011 * lazy tlb mm refcounting when these are updated by context
1012 * switches. Not all architectures can handle irqs off over
1015 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1017 activate_mm(active_mm
, mm
);
1018 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1020 tsk
->mm
->vmacache_seqnum
= 0;
1021 vmacache_flush(tsk
);
1024 mmap_read_unlock(old_mm
);
1025 BUG_ON(active_mm
!= old_mm
);
1026 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
1027 mm_update_next_owner(old_mm
);
1035 static int de_thread(struct task_struct
*tsk
)
1037 struct signal_struct
*sig
= tsk
->signal
;
1038 struct sighand_struct
*oldsighand
= tsk
->sighand
;
1039 spinlock_t
*lock
= &oldsighand
->siglock
;
1041 if (thread_group_empty(tsk
))
1042 goto no_thread_group
;
1045 * Kill all other threads in the thread group.
1047 spin_lock_irq(lock
);
1048 if (signal_group_exit(sig
)) {
1050 * Another group action in progress, just
1051 * return so that the signal is processed.
1053 spin_unlock_irq(lock
);
1057 sig
->group_exit_task
= tsk
;
1058 sig
->notify_count
= zap_other_threads(tsk
);
1059 if (!thread_group_leader(tsk
))
1060 sig
->notify_count
--;
1062 while (sig
->notify_count
) {
1063 __set_current_state(TASK_KILLABLE
);
1064 spin_unlock_irq(lock
);
1066 if (__fatal_signal_pending(tsk
))
1068 spin_lock_irq(lock
);
1070 spin_unlock_irq(lock
);
1073 * At this point all other threads have exited, all we have to
1074 * do is to wait for the thread group leader to become inactive,
1075 * and to assume its PID:
1077 if (!thread_group_leader(tsk
)) {
1078 struct task_struct
*leader
= tsk
->group_leader
;
1081 cgroup_threadgroup_change_begin(tsk
);
1082 write_lock_irq(&tasklist_lock
);
1084 * Do this under tasklist_lock to ensure that
1085 * exit_notify() can't miss ->group_exit_task
1087 sig
->notify_count
= -1;
1088 if (likely(leader
->exit_state
))
1090 __set_current_state(TASK_KILLABLE
);
1091 write_unlock_irq(&tasklist_lock
);
1092 cgroup_threadgroup_change_end(tsk
);
1094 if (__fatal_signal_pending(tsk
))
1099 * The only record we have of the real-time age of a
1100 * process, regardless of execs it's done, is start_time.
1101 * All the past CPU time is accumulated in signal_struct
1102 * from sister threads now dead. But in this non-leader
1103 * exec, nothing survives from the original leader thread,
1104 * whose birth marks the true age of this process now.
1105 * When we take on its identity by switching to its PID, we
1106 * also take its birthdate (always earlier than our own).
1108 tsk
->start_time
= leader
->start_time
;
1109 tsk
->start_boottime
= leader
->start_boottime
;
1111 BUG_ON(!same_thread_group(leader
, tsk
));
1113 * An exec() starts a new thread group with the
1114 * TGID of the previous thread group. Rehash the
1115 * two threads with a switched PID, and release
1116 * the former thread group leader:
1119 /* Become a process group leader with the old leader's pid.
1120 * The old leader becomes a thread of the this thread group.
1122 exchange_tids(tsk
, leader
);
1123 transfer_pid(leader
, tsk
, PIDTYPE_TGID
);
1124 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
1125 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
1127 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
1128 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
1130 tsk
->group_leader
= tsk
;
1131 leader
->group_leader
= tsk
;
1133 tsk
->exit_signal
= SIGCHLD
;
1134 leader
->exit_signal
= -1;
1136 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
1137 leader
->exit_state
= EXIT_DEAD
;
1140 * We are going to release_task()->ptrace_unlink() silently,
1141 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1142 * the tracer wont't block again waiting for this thread.
1144 if (unlikely(leader
->ptrace
))
1145 __wake_up_parent(leader
, leader
->parent
);
1146 write_unlock_irq(&tasklist_lock
);
1147 cgroup_threadgroup_change_end(tsk
);
1149 release_task(leader
);
1152 sig
->group_exit_task
= NULL
;
1153 sig
->notify_count
= 0;
1156 /* we have changed execution domain */
1157 tsk
->exit_signal
= SIGCHLD
;
1159 BUG_ON(!thread_group_leader(tsk
));
1163 /* protects against exit_notify() and __exit_signal() */
1164 read_lock(&tasklist_lock
);
1165 sig
->group_exit_task
= NULL
;
1166 sig
->notify_count
= 0;
1167 read_unlock(&tasklist_lock
);
1173 * This function makes sure the current process has its own signal table,
1174 * so that flush_signal_handlers can later reset the handlers without
1175 * disturbing other processes. (Other processes might share the signal
1176 * table via the CLONE_SIGHAND option to clone().)
1178 static int unshare_sighand(struct task_struct
*me
)
1180 struct sighand_struct
*oldsighand
= me
->sighand
;
1182 if (refcount_read(&oldsighand
->count
) != 1) {
1183 struct sighand_struct
*newsighand
;
1185 * This ->sighand is shared with the CLONE_SIGHAND
1186 * but not CLONE_THREAD task, switch to the new one.
1188 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1192 refcount_set(&newsighand
->count
, 1);
1193 memcpy(newsighand
->action
, oldsighand
->action
,
1194 sizeof(newsighand
->action
));
1196 write_lock_irq(&tasklist_lock
);
1197 spin_lock(&oldsighand
->siglock
);
1198 rcu_assign_pointer(me
->sighand
, newsighand
);
1199 spin_unlock(&oldsighand
->siglock
);
1200 write_unlock_irq(&tasklist_lock
);
1202 __cleanup_sighand(oldsighand
);
1207 char *__get_task_comm(char *buf
, size_t buf_size
, struct task_struct
*tsk
)
1210 strncpy(buf
, tsk
->comm
, buf_size
);
1214 EXPORT_SYMBOL_GPL(__get_task_comm
);
1217 * These functions flushes out all traces of the currently running executable
1218 * so that a new one can be started
1221 void __set_task_comm(struct task_struct
*tsk
, const char *buf
, bool exec
)
1224 trace_task_rename(tsk
, buf
);
1225 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1227 perf_event_comm(tsk
, exec
);
1231 * Calling this is the point of no return. None of the failures will be
1232 * seen by userspace since either the process is already taking a fatal
1233 * signal (via de_thread() or coredump), or will have SEGV raised
1234 * (after exec_mmap()) by search_binary_handler (see below).
1236 int begin_new_exec(struct linux_binprm
* bprm
)
1238 struct task_struct
*me
= current
;
1241 /* Once we are committed compute the creds */
1242 retval
= bprm_creds_from_file(bprm
);
1247 * Ensure all future errors are fatal.
1249 bprm
->point_of_no_return
= true;
1252 * Make this the only thread in the thread group.
1254 retval
= de_thread(me
);
1259 * Cancel any io_uring activity across execve
1261 io_uring_task_cancel();
1263 /* Ensure the files table is not shared. */
1264 retval
= unshare_files();
1269 * Must be called _before_ exec_mmap() as bprm->mm is
1270 * not visibile until then. This also enables the update
1273 set_mm_exe_file(bprm
->mm
, bprm
->file
);
1275 /* If the binary is not readable then enforce mm->dumpable=0 */
1276 would_dump(bprm
, bprm
->file
);
1277 if (bprm
->have_execfd
)
1278 would_dump(bprm
, bprm
->executable
);
1281 * Release all of the old mmap stuff
1283 acct_arg_size(bprm
, 0);
1284 retval
= exec_mmap(bprm
->mm
);
1290 #ifdef CONFIG_POSIX_TIMERS
1291 exit_itimers(me
->signal
);
1292 flush_itimer_signals();
1296 * Make the signal table private.
1298 retval
= unshare_sighand(me
);
1303 * Ensure that the uaccess routines can actually operate on userspace
1306 force_uaccess_begin();
1308 me
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
| PF_KTHREAD
|
1309 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1311 me
->personality
&= ~bprm
->per_clear
;
1313 clear_syscall_work_syscall_user_dispatch(me
);
1316 * We have to apply CLOEXEC before we change whether the process is
1317 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1318 * trying to access the should-be-closed file descriptors of a process
1319 * undergoing exec(2).
1321 do_close_on_exec(me
->files
);
1323 if (bprm
->secureexec
) {
1324 /* Make sure parent cannot signal privileged process. */
1325 me
->pdeath_signal
= 0;
1328 * For secureexec, reset the stack limit to sane default to
1329 * avoid bad behavior from the prior rlimits. This has to
1330 * happen before arch_pick_mmap_layout(), which examines
1331 * RLIMIT_STACK, but after the point of no return to avoid
1332 * needing to clean up the change on failure.
1334 if (bprm
->rlim_stack
.rlim_cur
> _STK_LIM
)
1335 bprm
->rlim_stack
.rlim_cur
= _STK_LIM
;
1338 me
->sas_ss_sp
= me
->sas_ss_size
= 0;
1341 * Figure out dumpability. Note that this checking only of current
1342 * is wrong, but userspace depends on it. This should be testing
1343 * bprm->secureexec instead.
1345 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
||
1346 !(uid_eq(current_euid(), current_uid()) &&
1347 gid_eq(current_egid(), current_gid())))
1348 set_dumpable(current
->mm
, suid_dumpable
);
1350 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1353 __set_task_comm(me
, kbasename(bprm
->filename
), true);
1355 /* An exec changes our domain. We are no longer part of the thread
1357 WRITE_ONCE(me
->self_exec_id
, me
->self_exec_id
+ 1);
1358 flush_signal_handlers(me
, 0);
1360 retval
= set_cred_ucounts(bprm
->cred
);
1365 * install the new credentials for this executable
1367 security_bprm_committing_creds(bprm
);
1369 commit_creds(bprm
->cred
);
1373 * Disable monitoring for regular users
1374 * when executing setuid binaries. Must
1375 * wait until new credentials are committed
1376 * by commit_creds() above
1378 if (get_dumpable(me
->mm
) != SUID_DUMP_USER
)
1379 perf_event_exit_task(me
);
1381 * cred_guard_mutex must be held at least to this point to prevent
1382 * ptrace_attach() from altering our determination of the task's
1383 * credentials; any time after this it may be unlocked.
1385 security_bprm_committed_creds(bprm
);
1387 /* Pass the opened binary to the interpreter. */
1388 if (bprm
->have_execfd
) {
1389 retval
= get_unused_fd_flags(0);
1392 fd_install(retval
, bprm
->executable
);
1393 bprm
->executable
= NULL
;
1394 bprm
->execfd
= retval
;
1399 up_write(&me
->signal
->exec_update_lock
);
1403 EXPORT_SYMBOL(begin_new_exec
);
1405 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1407 struct inode
*inode
= file_inode(file
);
1408 struct user_namespace
*mnt_userns
= file_mnt_user_ns(file
);
1409 if (inode_permission(mnt_userns
, inode
, MAY_READ
) < 0) {
1410 struct user_namespace
*old
, *user_ns
;
1411 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1413 /* Ensure mm->user_ns contains the executable */
1414 user_ns
= old
= bprm
->mm
->user_ns
;
1415 while ((user_ns
!= &init_user_ns
) &&
1416 !privileged_wrt_inode_uidgid(user_ns
, mnt_userns
, inode
))
1417 user_ns
= user_ns
->parent
;
1419 if (old
!= user_ns
) {
1420 bprm
->mm
->user_ns
= get_user_ns(user_ns
);
1425 EXPORT_SYMBOL(would_dump
);
1427 void setup_new_exec(struct linux_binprm
* bprm
)
1429 /* Setup things that can depend upon the personality */
1430 struct task_struct
*me
= current
;
1432 arch_pick_mmap_layout(me
->mm
, &bprm
->rlim_stack
);
1434 arch_setup_new_exec();
1436 /* Set the new mm task size. We have to do that late because it may
1437 * depend on TIF_32BIT which is only updated in flush_thread() on
1438 * some architectures like powerpc
1440 me
->mm
->task_size
= TASK_SIZE
;
1441 up_write(&me
->signal
->exec_update_lock
);
1442 mutex_unlock(&me
->signal
->cred_guard_mutex
);
1444 EXPORT_SYMBOL(setup_new_exec
);
1446 /* Runs immediately before start_thread() takes over. */
1447 void finalize_exec(struct linux_binprm
*bprm
)
1449 /* Store any stack rlimit changes before starting thread. */
1450 task_lock(current
->group_leader
);
1451 current
->signal
->rlim
[RLIMIT_STACK
] = bprm
->rlim_stack
;
1452 task_unlock(current
->group_leader
);
1454 EXPORT_SYMBOL(finalize_exec
);
1457 * Prepare credentials and lock ->cred_guard_mutex.
1458 * setup_new_exec() commits the new creds and drops the lock.
1459 * Or, if exec fails before, free_bprm() should release ->cred
1462 static int prepare_bprm_creds(struct linux_binprm
*bprm
)
1464 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1465 return -ERESTARTNOINTR
;
1467 bprm
->cred
= prepare_exec_creds();
1468 if (likely(bprm
->cred
))
1471 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1475 static void free_bprm(struct linux_binprm
*bprm
)
1478 acct_arg_size(bprm
, 0);
1481 free_arg_pages(bprm
);
1483 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1484 abort_creds(bprm
->cred
);
1487 allow_write_access(bprm
->file
);
1490 if (bprm
->executable
)
1491 fput(bprm
->executable
);
1492 /* If a binfmt changed the interp, free it. */
1493 if (bprm
->interp
!= bprm
->filename
)
1494 kfree(bprm
->interp
);
1495 kfree(bprm
->fdpath
);
1499 static struct linux_binprm
*alloc_bprm(int fd
, struct filename
*filename
)
1501 struct linux_binprm
*bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1502 int retval
= -ENOMEM
;
1506 if (fd
== AT_FDCWD
|| filename
->name
[0] == '/') {
1507 bprm
->filename
= filename
->name
;
1509 if (filename
->name
[0] == '\0')
1510 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d", fd
);
1512 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d/%s",
1513 fd
, filename
->name
);
1517 bprm
->filename
= bprm
->fdpath
;
1519 bprm
->interp
= bprm
->filename
;
1521 retval
= bprm_mm_init(bprm
);
1529 return ERR_PTR(retval
);
1532 int bprm_change_interp(const char *interp
, struct linux_binprm
*bprm
)
1534 /* If a binfmt changed the interp, free it first. */
1535 if (bprm
->interp
!= bprm
->filename
)
1536 kfree(bprm
->interp
);
1537 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1542 EXPORT_SYMBOL(bprm_change_interp
);
1545 * determine how safe it is to execute the proposed program
1546 * - the caller must hold ->cred_guard_mutex to protect against
1547 * PTRACE_ATTACH or seccomp thread-sync
1549 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1551 struct task_struct
*p
= current
, *t
;
1555 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1558 * This isn't strictly necessary, but it makes it harder for LSMs to
1561 if (task_no_new_privs(current
))
1562 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1566 spin_lock(&p
->fs
->lock
);
1568 while_each_thread(p
, t
) {
1574 if (p
->fs
->users
> n_fs
)
1575 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1578 spin_unlock(&p
->fs
->lock
);
1581 static void bprm_fill_uid(struct linux_binprm
*bprm
, struct file
*file
)
1583 /* Handle suid and sgid on files */
1584 struct user_namespace
*mnt_userns
;
1585 struct inode
*inode
;
1590 if (!mnt_may_suid(file
->f_path
.mnt
))
1593 if (task_no_new_privs(current
))
1596 inode
= file
->f_path
.dentry
->d_inode
;
1597 mode
= READ_ONCE(inode
->i_mode
);
1598 if (!(mode
& (S_ISUID
|S_ISGID
)))
1601 mnt_userns
= file_mnt_user_ns(file
);
1603 /* Be careful if suid/sgid is set */
1606 /* reload atomically mode/uid/gid now that lock held */
1607 mode
= inode
->i_mode
;
1608 uid
= i_uid_into_mnt(mnt_userns
, inode
);
1609 gid
= i_gid_into_mnt(mnt_userns
, inode
);
1610 inode_unlock(inode
);
1612 /* We ignore suid/sgid if there are no mappings for them in the ns */
1613 if (!kuid_has_mapping(bprm
->cred
->user_ns
, uid
) ||
1614 !kgid_has_mapping(bprm
->cred
->user_ns
, gid
))
1617 if (mode
& S_ISUID
) {
1618 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1619 bprm
->cred
->euid
= uid
;
1622 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1623 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1624 bprm
->cred
->egid
= gid
;
1629 * Compute brpm->cred based upon the final binary.
1631 static int bprm_creds_from_file(struct linux_binprm
*bprm
)
1633 /* Compute creds based on which file? */
1634 struct file
*file
= bprm
->execfd_creds
? bprm
->executable
: bprm
->file
;
1636 bprm_fill_uid(bprm
, file
);
1637 return security_bprm_creds_from_file(bprm
, file
);
1641 * Fill the binprm structure from the inode.
1642 * Read the first BINPRM_BUF_SIZE bytes
1644 * This may be called multiple times for binary chains (scripts for example).
1646 static int prepare_binprm(struct linux_binprm
*bprm
)
1650 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1651 return kernel_read(bprm
->file
, bprm
->buf
, BINPRM_BUF_SIZE
, &pos
);
1655 * Arguments are '\0' separated strings found at the location bprm->p
1656 * points to; chop off the first by relocating brpm->p to right after
1657 * the first '\0' encountered.
1659 int remove_arg_zero(struct linux_binprm
*bprm
)
1662 unsigned long offset
;
1670 offset
= bprm
->p
& ~PAGE_MASK
;
1671 page
= get_arg_page(bprm
, bprm
->p
, 0);
1676 kaddr
= kmap_atomic(page
);
1678 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1679 offset
++, bprm
->p
++)
1682 kunmap_atomic(kaddr
);
1684 } while (offset
== PAGE_SIZE
);
1693 EXPORT_SYMBOL(remove_arg_zero
);
1695 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1697 * cycle the list of binary formats handler, until one recognizes the image
1699 static int search_binary_handler(struct linux_binprm
*bprm
)
1701 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1702 struct linux_binfmt
*fmt
;
1705 retval
= prepare_binprm(bprm
);
1709 retval
= security_bprm_check(bprm
);
1715 read_lock(&binfmt_lock
);
1716 list_for_each_entry(fmt
, &formats
, lh
) {
1717 if (!try_module_get(fmt
->module
))
1719 read_unlock(&binfmt_lock
);
1721 retval
= fmt
->load_binary(bprm
);
1723 read_lock(&binfmt_lock
);
1725 if (bprm
->point_of_no_return
|| (retval
!= -ENOEXEC
)) {
1726 read_unlock(&binfmt_lock
);
1730 read_unlock(&binfmt_lock
);
1733 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1734 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1736 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1745 static int exec_binprm(struct linux_binprm
*bprm
)
1747 pid_t old_pid
, old_vpid
;
1750 /* Need to fetch pid before load_binary changes it */
1751 old_pid
= current
->pid
;
1753 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1756 /* This allows 4 levels of binfmt rewrites before failing hard. */
1757 for (depth
= 0;; depth
++) {
1762 ret
= search_binary_handler(bprm
);
1765 if (!bprm
->interpreter
)
1769 bprm
->file
= bprm
->interpreter
;
1770 bprm
->interpreter
= NULL
;
1772 allow_write_access(exec
);
1773 if (unlikely(bprm
->have_execfd
)) {
1774 if (bprm
->executable
) {
1778 bprm
->executable
= exec
;
1784 trace_sched_process_exec(current
, old_pid
, bprm
);
1785 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1786 proc_exec_connector(current
);
1791 * sys_execve() executes a new program.
1793 static int bprm_execve(struct linux_binprm
*bprm
,
1794 int fd
, struct filename
*filename
, int flags
)
1799 retval
= prepare_bprm_creds(bprm
);
1803 check_unsafe_exec(bprm
);
1804 current
->in_execve
= 1;
1806 file
= do_open_execat(fd
, filename
, flags
);
1807 retval
= PTR_ERR(file
);
1815 * Record that a name derived from an O_CLOEXEC fd will be
1816 * inaccessible after exec. This allows the code in exec to
1817 * choose to fail when the executable is not mmaped into the
1818 * interpreter and an open file descriptor is not passed to
1819 * the interpreter. This makes for a better user experience
1820 * than having the interpreter start and then immediately fail
1821 * when it finds the executable is inaccessible.
1823 if (bprm
->fdpath
&& get_close_on_exec(fd
))
1824 bprm
->interp_flags
|= BINPRM_FLAGS_PATH_INACCESSIBLE
;
1826 /* Set the unchanging part of bprm->cred */
1827 retval
= security_bprm_creds_for_exec(bprm
);
1831 retval
= exec_binprm(bprm
);
1835 /* execve succeeded */
1836 current
->fs
->in_exec
= 0;
1837 current
->in_execve
= 0;
1838 rseq_execve(current
);
1839 acct_update_integrals(current
);
1840 task_numa_free(current
, false);
1845 * If past the point of no return ensure the code never
1846 * returns to the userspace process. Use an existing fatal
1847 * signal if present otherwise terminate the process with
1850 if (bprm
->point_of_no_return
&& !fatal_signal_pending(current
))
1851 force_sigsegv(SIGSEGV
);
1854 current
->fs
->in_exec
= 0;
1855 current
->in_execve
= 0;
1860 static int do_execveat_common(int fd
, struct filename
*filename
,
1861 struct user_arg_ptr argv
,
1862 struct user_arg_ptr envp
,
1865 struct linux_binprm
*bprm
;
1868 if (IS_ERR(filename
))
1869 return PTR_ERR(filename
);
1872 * We move the actual failure in case of RLIMIT_NPROC excess from
1873 * set*uid() to execve() because too many poorly written programs
1874 * don't check setuid() return code. Here we additionally recheck
1875 * whether NPROC limit is still exceeded.
1877 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1878 is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC
, rlimit(RLIMIT_NPROC
))) {
1883 /* We're below the limit (still or again), so we don't want to make
1884 * further execve() calls fail. */
1885 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1887 bprm
= alloc_bprm(fd
, filename
);
1889 retval
= PTR_ERR(bprm
);
1893 retval
= count(argv
, MAX_ARG_STRINGS
);
1896 bprm
->argc
= retval
;
1898 retval
= count(envp
, MAX_ARG_STRINGS
);
1901 bprm
->envc
= retval
;
1903 retval
= bprm_stack_limits(bprm
);
1907 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1910 bprm
->exec
= bprm
->p
;
1912 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1916 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1920 retval
= bprm_execve(bprm
, fd
, filename
, flags
);
1929 int kernel_execve(const char *kernel_filename
,
1930 const char *const *argv
, const char *const *envp
)
1932 struct filename
*filename
;
1933 struct linux_binprm
*bprm
;
1937 filename
= getname_kernel(kernel_filename
);
1938 if (IS_ERR(filename
))
1939 return PTR_ERR(filename
);
1941 bprm
= alloc_bprm(fd
, filename
);
1943 retval
= PTR_ERR(bprm
);
1947 retval
= count_strings_kernel(argv
);
1950 bprm
->argc
= retval
;
1952 retval
= count_strings_kernel(envp
);
1955 bprm
->envc
= retval
;
1957 retval
= bprm_stack_limits(bprm
);
1961 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1964 bprm
->exec
= bprm
->p
;
1966 retval
= copy_strings_kernel(bprm
->envc
, envp
, bprm
);
1970 retval
= copy_strings_kernel(bprm
->argc
, argv
, bprm
);
1974 retval
= bprm_execve(bprm
, fd
, filename
, 0);
1982 static int do_execve(struct filename
*filename
,
1983 const char __user
*const __user
*__argv
,
1984 const char __user
*const __user
*__envp
)
1986 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1987 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1988 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
1991 static int do_execveat(int fd
, struct filename
*filename
,
1992 const char __user
*const __user
*__argv
,
1993 const char __user
*const __user
*__envp
,
1996 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1997 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1999 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2002 #ifdef CONFIG_COMPAT
2003 static int compat_do_execve(struct filename
*filename
,
2004 const compat_uptr_t __user
*__argv
,
2005 const compat_uptr_t __user
*__envp
)
2007 struct user_arg_ptr argv
= {
2009 .ptr
.compat
= __argv
,
2011 struct user_arg_ptr envp
= {
2013 .ptr
.compat
= __envp
,
2015 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2018 static int compat_do_execveat(int fd
, struct filename
*filename
,
2019 const compat_uptr_t __user
*__argv
,
2020 const compat_uptr_t __user
*__envp
,
2023 struct user_arg_ptr argv
= {
2025 .ptr
.compat
= __argv
,
2027 struct user_arg_ptr envp
= {
2029 .ptr
.compat
= __envp
,
2031 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2035 void set_binfmt(struct linux_binfmt
*new)
2037 struct mm_struct
*mm
= current
->mm
;
2040 module_put(mm
->binfmt
->module
);
2044 __module_get(new->module
);
2046 EXPORT_SYMBOL(set_binfmt
);
2049 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2051 void set_dumpable(struct mm_struct
*mm
, int value
)
2053 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
2056 set_mask_bits(&mm
->flags
, MMF_DUMPABLE_MASK
, value
);
2059 SYSCALL_DEFINE3(execve
,
2060 const char __user
*, filename
,
2061 const char __user
*const __user
*, argv
,
2062 const char __user
*const __user
*, envp
)
2064 return do_execve(getname(filename
), argv
, envp
);
2067 SYSCALL_DEFINE5(execveat
,
2068 int, fd
, const char __user
*, filename
,
2069 const char __user
*const __user
*, argv
,
2070 const char __user
*const __user
*, envp
,
2073 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
2075 return do_execveat(fd
,
2076 getname_flags(filename
, lookup_flags
, NULL
),
2080 #ifdef CONFIG_COMPAT
2081 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
2082 const compat_uptr_t __user
*, argv
,
2083 const compat_uptr_t __user
*, envp
)
2085 return compat_do_execve(getname(filename
), argv
, envp
);
2088 COMPAT_SYSCALL_DEFINE5(execveat
, int, fd
,
2089 const char __user
*, filename
,
2090 const compat_uptr_t __user
*, argv
,
2091 const compat_uptr_t __user
*, envp
,
2094 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
2096 return compat_do_execveat(fd
,
2097 getname_flags(filename
, lookup_flags
, NULL
),