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/slab.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
30 #include <linux/vmacache.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/tracehook.h>
59 #include <linux/kmod.h>
60 #include <linux/fsnotify.h>
61 #include <linux/fs_struct.h>
62 #include <linux/oom.h>
63 #include <linux/compat.h>
64 #include <linux/vmalloc.h>
66 #include <linux/uaccess.h>
67 #include <asm/mmu_context.h>
70 #include <trace/events/task.h>
73 #include <trace/events/sched.h>
75 static int bprm_creds_from_file(struct linux_binprm
*bprm
);
77 int suid_dumpable
= 0;
79 static LIST_HEAD(formats
);
80 static DEFINE_RWLOCK(binfmt_lock
);
82 void __register_binfmt(struct linux_binfmt
* fmt
, int insert
)
85 if (WARN_ON(!fmt
->load_binary
))
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 if (!S_ISREG(file_inode(file
)->i_mode
))
149 if (path_noexec(&file
->f_path
))
156 read_lock(&binfmt_lock
);
157 list_for_each_entry(fmt
, &formats
, lh
) {
158 if (!fmt
->load_shlib
)
160 if (!try_module_get(fmt
->module
))
162 read_unlock(&binfmt_lock
);
163 error
= fmt
->load_shlib(file
);
164 read_lock(&binfmt_lock
);
166 if (error
!= -ENOEXEC
)
169 read_unlock(&binfmt_lock
);
175 #endif /* #ifdef CONFIG_USELIB */
179 * The nascent bprm->mm is not visible until exec_mmap() but it can
180 * use a lot of memory, account these pages in current->mm temporary
181 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
182 * change the counter back via acct_arg_size(0).
184 static void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
186 struct mm_struct
*mm
= current
->mm
;
187 long diff
= (long)(pages
- bprm
->vma_pages
);
192 bprm
->vma_pages
= pages
;
193 add_mm_counter(mm
, MM_ANONPAGES
, diff
);
196 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
201 unsigned int gup_flags
= FOLL_FORCE
;
203 #ifdef CONFIG_STACK_GROWSUP
205 ret
= expand_downwards(bprm
->vma
, pos
);
212 gup_flags
|= FOLL_WRITE
;
215 * We are doing an exec(). 'current' is the process
216 * doing the exec and bprm->mm is the new process's mm.
218 ret
= get_user_pages_remote(current
, bprm
->mm
, pos
, 1, gup_flags
,
224 acct_arg_size(bprm
, vma_pages(bprm
->vma
));
229 static void put_arg_page(struct page
*page
)
234 static void free_arg_pages(struct linux_binprm
*bprm
)
238 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
241 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
244 static int __bprm_mm_init(struct linux_binprm
*bprm
)
247 struct vm_area_struct
*vma
= NULL
;
248 struct mm_struct
*mm
= bprm
->mm
;
250 bprm
->vma
= vma
= vm_area_alloc(mm
);
253 vma_set_anonymous(vma
);
255 if (mmap_write_lock_killable(mm
)) {
261 * Place the stack at the largest stack address the architecture
262 * supports. Later, we'll move this to an appropriate place. We don't
263 * use STACK_TOP because that can depend on attributes which aren't
266 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
267 vma
->vm_end
= STACK_TOP_MAX
;
268 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
269 vma
->vm_flags
= VM_SOFTDIRTY
| VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
;
270 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
272 err
= insert_vm_struct(mm
, vma
);
276 mm
->stack_vm
= mm
->total_vm
= 1;
277 mmap_write_unlock(mm
);
278 bprm
->p
= vma
->vm_end
- sizeof(void *);
281 mmap_write_unlock(mm
);
288 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
290 return len
<= MAX_ARG_STRLEN
;
295 static inline void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
299 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
304 page
= bprm
->page
[pos
/ PAGE_SIZE
];
305 if (!page
&& write
) {
306 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
309 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
315 static void put_arg_page(struct page
*page
)
319 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
322 __free_page(bprm
->page
[i
]);
323 bprm
->page
[i
] = NULL
;
327 static void free_arg_pages(struct linux_binprm
*bprm
)
331 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
332 free_arg_page(bprm
, i
);
335 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
340 static int __bprm_mm_init(struct linux_binprm
*bprm
)
342 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
346 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
348 return len
<= bprm
->p
;
351 #endif /* CONFIG_MMU */
354 * Create a new mm_struct and populate it with a temporary stack
355 * vm_area_struct. We don't have enough context at this point to set the stack
356 * flags, permissions, and offset, so we use temporary values. We'll update
357 * them later in setup_arg_pages().
359 static int bprm_mm_init(struct linux_binprm
*bprm
)
362 struct mm_struct
*mm
= NULL
;
364 bprm
->mm
= mm
= mm_alloc();
369 /* Save current stack limit for all calculations made during exec. */
370 task_lock(current
->group_leader
);
371 bprm
->rlim_stack
= current
->signal
->rlim
[RLIMIT_STACK
];
372 task_unlock(current
->group_leader
);
374 err
= __bprm_mm_init(bprm
);
389 struct user_arg_ptr
{
394 const char __user
*const __user
*native
;
396 const compat_uptr_t __user
*compat
;
401 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
403 const char __user
*native
;
406 if (unlikely(argv
.is_compat
)) {
407 compat_uptr_t compat
;
409 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
410 return ERR_PTR(-EFAULT
);
412 return compat_ptr(compat
);
416 if (get_user(native
, argv
.ptr
.native
+ nr
))
417 return ERR_PTR(-EFAULT
);
423 * count() counts the number of strings in array ARGV.
425 static int count(struct user_arg_ptr argv
, int max
)
429 if (argv
.ptr
.native
!= NULL
) {
431 const char __user
*p
= get_user_arg_ptr(argv
, i
);
443 if (fatal_signal_pending(current
))
444 return -ERESTARTNOHAND
;
451 static int count_strings_kernel(const char *const *argv
)
458 for (i
= 0; argv
[i
]; ++i
) {
459 if (i
>= MAX_ARG_STRINGS
)
461 if (fatal_signal_pending(current
))
462 return -ERESTARTNOHAND
;
468 static int bprm_stack_limits(struct linux_binprm
*bprm
)
470 unsigned long limit
, ptr_size
;
473 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
474 * (whichever is smaller) for the argv+env strings.
476 * - the remaining binfmt code will not run out of stack space,
477 * - the program will have a reasonable amount of stack left
480 limit
= _STK_LIM
/ 4 * 3;
481 limit
= min(limit
, bprm
->rlim_stack
.rlim_cur
/ 4);
483 * We've historically supported up to 32 pages (ARG_MAX)
484 * of argument strings even with small stacks
486 limit
= max_t(unsigned long, limit
, ARG_MAX
);
488 * We must account for the size of all the argv and envp pointers to
489 * the argv and envp strings, since they will also take up space in
490 * the stack. They aren't stored until much later when we can't
491 * signal to the parent that the child has run out of stack space.
492 * Instead, calculate it here so it's possible to fail gracefully.
494 ptr_size
= (bprm
->argc
+ bprm
->envc
) * sizeof(void *);
495 if (limit
<= ptr_size
)
499 bprm
->argmin
= bprm
->p
- limit
;
504 * 'copy_strings()' copies argument/environment strings from the old
505 * processes's memory to the new process's stack. The call to get_user_pages()
506 * ensures the destination page is created and not swapped out.
508 static int copy_strings(int argc
, struct user_arg_ptr argv
,
509 struct linux_binprm
*bprm
)
511 struct page
*kmapped_page
= NULL
;
513 unsigned long kpos
= 0;
517 const char __user
*str
;
522 str
= get_user_arg_ptr(argv
, argc
);
526 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
531 if (!valid_arg_len(bprm
, len
))
534 /* We're going to work our way backwords. */
539 if (bprm
->p
< bprm
->argmin
)
544 int offset
, bytes_to_copy
;
546 if (fatal_signal_pending(current
)) {
547 ret
= -ERESTARTNOHAND
;
552 offset
= pos
% PAGE_SIZE
;
556 bytes_to_copy
= offset
;
557 if (bytes_to_copy
> len
)
560 offset
-= bytes_to_copy
;
561 pos
-= bytes_to_copy
;
562 str
-= bytes_to_copy
;
563 len
-= bytes_to_copy
;
565 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
568 page
= get_arg_page(bprm
, pos
, 1);
575 flush_kernel_dcache_page(kmapped_page
);
576 kunmap(kmapped_page
);
577 put_arg_page(kmapped_page
);
580 kaddr
= kmap(kmapped_page
);
581 kpos
= pos
& PAGE_MASK
;
582 flush_arg_page(bprm
, kpos
, kmapped_page
);
584 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
593 flush_kernel_dcache_page(kmapped_page
);
594 kunmap(kmapped_page
);
595 put_arg_page(kmapped_page
);
601 * Copy and argument/environment string from the kernel to the processes stack.
603 int copy_string_kernel(const char *arg
, struct linux_binprm
*bprm
)
605 int len
= strnlen(arg
, MAX_ARG_STRLEN
) + 1 /* terminating NUL */;
606 unsigned long pos
= bprm
->p
;
610 if (!valid_arg_len(bprm
, len
))
613 /* We're going to work our way backwards. */
616 if (IS_ENABLED(CONFIG_MMU
) && bprm
->p
< bprm
->argmin
)
620 unsigned int bytes_to_copy
= min_t(unsigned int, len
,
621 min_not_zero(offset_in_page(pos
), PAGE_SIZE
));
625 pos
-= bytes_to_copy
;
626 arg
-= bytes_to_copy
;
627 len
-= bytes_to_copy
;
629 page
= get_arg_page(bprm
, pos
, 1);
632 kaddr
= kmap_atomic(page
);
633 flush_arg_page(bprm
, pos
& PAGE_MASK
, page
);
634 memcpy(kaddr
+ offset_in_page(pos
), arg
, bytes_to_copy
);
635 flush_kernel_dcache_page(page
);
636 kunmap_atomic(kaddr
);
642 EXPORT_SYMBOL(copy_string_kernel
);
644 static int copy_strings_kernel(int argc
, const char *const *argv
,
645 struct linux_binprm
*bprm
)
648 int ret
= copy_string_kernel(argv
[argc
], bprm
);
651 if (fatal_signal_pending(current
))
652 return -ERESTARTNOHAND
;
661 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
662 * the binfmt code determines where the new stack should reside, we shift it to
663 * its final location. The process proceeds as follows:
665 * 1) Use shift to calculate the new vma endpoints.
666 * 2) Extend vma to cover both the old and new ranges. This ensures the
667 * arguments passed to subsequent functions are consistent.
668 * 3) Move vma's page tables to the new range.
669 * 4) Free up any cleared pgd range.
670 * 5) Shrink the vma to cover only the new range.
672 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
674 struct mm_struct
*mm
= vma
->vm_mm
;
675 unsigned long old_start
= vma
->vm_start
;
676 unsigned long old_end
= vma
->vm_end
;
677 unsigned long length
= old_end
- old_start
;
678 unsigned long new_start
= old_start
- shift
;
679 unsigned long new_end
= old_end
- shift
;
680 struct mmu_gather tlb
;
682 BUG_ON(new_start
> new_end
);
685 * ensure there are no vmas between where we want to go
688 if (vma
!= find_vma(mm
, new_start
))
692 * cover the whole range: [new_start, old_end)
694 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
698 * move the page tables downwards, on failure we rely on
699 * process cleanup to remove whatever mess we made.
701 if (length
!= move_page_tables(vma
, old_start
,
702 vma
, new_start
, length
, false))
706 tlb_gather_mmu(&tlb
, mm
, old_start
, old_end
);
707 if (new_end
> old_start
) {
709 * when the old and new regions overlap clear from new_end.
711 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
712 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
715 * otherwise, clean from old_start; this is done to not touch
716 * the address space in [new_end, old_start) some architectures
717 * have constraints on va-space that make this illegal (IA64) -
718 * for the others its just a little faster.
720 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
721 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
723 tlb_finish_mmu(&tlb
, old_start
, old_end
);
726 * Shrink the vma to just the new range. Always succeeds.
728 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
734 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
735 * the stack is optionally relocated, and some extra space is added.
737 int setup_arg_pages(struct linux_binprm
*bprm
,
738 unsigned long stack_top
,
739 int executable_stack
)
742 unsigned long stack_shift
;
743 struct mm_struct
*mm
= current
->mm
;
744 struct vm_area_struct
*vma
= bprm
->vma
;
745 struct vm_area_struct
*prev
= NULL
;
746 unsigned long vm_flags
;
747 unsigned long stack_base
;
748 unsigned long stack_size
;
749 unsigned long stack_expand
;
750 unsigned long rlim_stack
;
752 #ifdef CONFIG_STACK_GROWSUP
753 /* Limit stack size */
754 stack_base
= bprm
->rlim_stack
.rlim_max
;
755 if (stack_base
> STACK_SIZE_MAX
)
756 stack_base
= STACK_SIZE_MAX
;
758 /* Add space for stack randomization. */
759 stack_base
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
761 /* Make sure we didn't let the argument array grow too large. */
762 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
765 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
767 stack_shift
= vma
->vm_start
- stack_base
;
768 mm
->arg_start
= bprm
->p
- stack_shift
;
769 bprm
->p
= vma
->vm_end
- stack_shift
;
771 stack_top
= arch_align_stack(stack_top
);
772 stack_top
= PAGE_ALIGN(stack_top
);
774 if (unlikely(stack_top
< mmap_min_addr
) ||
775 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
778 stack_shift
= vma
->vm_end
- stack_top
;
780 bprm
->p
-= stack_shift
;
781 mm
->arg_start
= bprm
->p
;
785 bprm
->loader
-= stack_shift
;
786 bprm
->exec
-= stack_shift
;
788 if (mmap_write_lock_killable(mm
))
791 vm_flags
= VM_STACK_FLAGS
;
794 * Adjust stack execute permissions; explicitly enable for
795 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
796 * (arch default) otherwise.
798 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
800 else if (executable_stack
== EXSTACK_DISABLE_X
)
801 vm_flags
&= ~VM_EXEC
;
802 vm_flags
|= mm
->def_flags
;
803 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
805 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
811 if (unlikely(vm_flags
& VM_EXEC
)) {
812 pr_warn_once("process '%pD4' started with executable stack\n",
816 /* Move stack pages down in memory. */
818 ret
= shift_arg_pages(vma
, stack_shift
);
823 /* mprotect_fixup is overkill to remove the temporary stack flags */
824 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
826 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
827 stack_size
= vma
->vm_end
- vma
->vm_start
;
829 * Align this down to a page boundary as expand_stack
832 rlim_stack
= bprm
->rlim_stack
.rlim_cur
& PAGE_MASK
;
833 #ifdef CONFIG_STACK_GROWSUP
834 if (stack_size
+ stack_expand
> rlim_stack
)
835 stack_base
= vma
->vm_start
+ rlim_stack
;
837 stack_base
= vma
->vm_end
+ stack_expand
;
839 if (stack_size
+ stack_expand
> rlim_stack
)
840 stack_base
= vma
->vm_end
- rlim_stack
;
842 stack_base
= vma
->vm_start
- stack_expand
;
844 current
->mm
->start_stack
= bprm
->p
;
845 ret
= expand_stack(vma
, stack_base
);
850 mmap_write_unlock(mm
);
853 EXPORT_SYMBOL(setup_arg_pages
);
858 * Transfer the program arguments and environment from the holding pages
859 * onto the stack. The provided stack pointer is adjusted accordingly.
861 int transfer_args_to_stack(struct linux_binprm
*bprm
,
862 unsigned long *sp_location
)
864 unsigned long index
, stop
, sp
;
867 stop
= bprm
->p
>> PAGE_SHIFT
;
870 for (index
= MAX_ARG_PAGES
- 1; index
>= stop
; index
--) {
871 unsigned int offset
= index
== stop
? bprm
->p
& ~PAGE_MASK
: 0;
872 char *src
= kmap(bprm
->page
[index
]) + offset
;
873 sp
-= PAGE_SIZE
- offset
;
874 if (copy_to_user((void *) sp
, src
, PAGE_SIZE
- offset
) != 0)
876 kunmap(bprm
->page
[index
]);
886 EXPORT_SYMBOL(transfer_args_to_stack
);
888 #endif /* CONFIG_MMU */
890 static struct file
*do_open_execat(int fd
, struct filename
*name
, int flags
)
894 struct open_flags open_exec_flags
= {
895 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
896 .acc_mode
= MAY_EXEC
,
897 .intent
= LOOKUP_OPEN
,
898 .lookup_flags
= LOOKUP_FOLLOW
,
901 if ((flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
902 return ERR_PTR(-EINVAL
);
903 if (flags
& AT_SYMLINK_NOFOLLOW
)
904 open_exec_flags
.lookup_flags
&= ~LOOKUP_FOLLOW
;
905 if (flags
& AT_EMPTY_PATH
)
906 open_exec_flags
.lookup_flags
|= LOOKUP_EMPTY
;
908 file
= do_filp_open(fd
, name
, &open_exec_flags
);
913 if (!S_ISREG(file_inode(file
)->i_mode
))
916 if (path_noexec(&file
->f_path
))
919 err
= deny_write_access(file
);
923 if (name
->name
[0] != '\0')
934 struct file
*open_exec(const char *name
)
936 struct filename
*filename
= getname_kernel(name
);
937 struct file
*f
= ERR_CAST(filename
);
939 if (!IS_ERR(filename
)) {
940 f
= do_open_execat(AT_FDCWD
, filename
, 0);
945 EXPORT_SYMBOL(open_exec
);
947 int kernel_read_file(struct file
*file
, void **buf
, loff_t
*size
,
948 loff_t max_size
, enum kernel_read_file_id id
)
954 if (!S_ISREG(file_inode(file
)->i_mode
) || max_size
< 0)
957 ret
= deny_write_access(file
);
961 ret
= security_kernel_read_file(file
, id
);
965 i_size
= i_size_read(file_inode(file
));
970 if (i_size
> SIZE_MAX
|| (max_size
> 0 && i_size
> max_size
)) {
975 if (id
!= READING_FIRMWARE_PREALLOC_BUFFER
)
976 *buf
= vmalloc(i_size
);
983 while (pos
< i_size
) {
984 bytes
= kernel_read(file
, *buf
+ pos
, i_size
- pos
, &pos
);
999 ret
= security_kernel_post_read_file(file
, *buf
, i_size
, id
);
1005 if (id
!= READING_FIRMWARE_PREALLOC_BUFFER
) {
1012 allow_write_access(file
);
1015 EXPORT_SYMBOL_GPL(kernel_read_file
);
1017 int kernel_read_file_from_path(const char *path
, void **buf
, loff_t
*size
,
1018 loff_t max_size
, enum kernel_read_file_id id
)
1023 if (!path
|| !*path
)
1026 file
= filp_open(path
, O_RDONLY
, 0);
1028 return PTR_ERR(file
);
1030 ret
= kernel_read_file(file
, buf
, size
, max_size
, id
);
1034 EXPORT_SYMBOL_GPL(kernel_read_file_from_path
);
1036 int kernel_read_file_from_path_initns(const char *path
, void **buf
,
1037 loff_t
*size
, loff_t max_size
,
1038 enum kernel_read_file_id id
)
1044 if (!path
|| !*path
)
1047 task_lock(&init_task
);
1048 get_fs_root(init_task
.fs
, &root
);
1049 task_unlock(&init_task
);
1051 file
= file_open_root(root
.dentry
, root
.mnt
, path
, O_RDONLY
, 0);
1054 return PTR_ERR(file
);
1056 ret
= kernel_read_file(file
, buf
, size
, max_size
, id
);
1060 EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns
);
1062 int kernel_read_file_from_fd(int fd
, void **buf
, loff_t
*size
, loff_t max_size
,
1063 enum kernel_read_file_id id
)
1065 struct fd f
= fdget(fd
);
1071 ret
= kernel_read_file(f
.file
, buf
, size
, max_size
, id
);
1076 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd
);
1078 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
1079 defined(CONFIG_BINFMT_ELF_FDPIC)
1080 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
1082 ssize_t res
= vfs_read(file
, (void __user
*)addr
, len
, &pos
);
1084 flush_icache_user_range(addr
, addr
+ len
);
1087 EXPORT_SYMBOL(read_code
);
1091 * Maps the mm_struct mm into the current task struct.
1092 * On success, this function returns with the mutex
1093 * exec_update_mutex locked.
1095 static int exec_mmap(struct mm_struct
*mm
)
1097 struct task_struct
*tsk
;
1098 struct mm_struct
*old_mm
, *active_mm
;
1101 /* Notify parent that we're no longer interested in the old VM */
1103 old_mm
= current
->mm
;
1104 exec_mm_release(tsk
, old_mm
);
1106 sync_mm_rss(old_mm
);
1108 ret
= mutex_lock_killable(&tsk
->signal
->exec_update_mutex
);
1114 * Make sure that if there is a core dump in progress
1115 * for the old mm, we get out and die instead of going
1116 * through with the exec. We must hold mmap_lock around
1117 * checking core_state and changing tsk->mm.
1119 mmap_read_lock(old_mm
);
1120 if (unlikely(old_mm
->core_state
)) {
1121 mmap_read_unlock(old_mm
);
1122 mutex_unlock(&tsk
->signal
->exec_update_mutex
);
1128 active_mm
= tsk
->active_mm
;
1129 membarrier_exec_mmap(mm
);
1131 tsk
->active_mm
= mm
;
1132 activate_mm(active_mm
, mm
);
1133 tsk
->mm
->vmacache_seqnum
= 0;
1134 vmacache_flush(tsk
);
1137 mmap_read_unlock(old_mm
);
1138 BUG_ON(active_mm
!= old_mm
);
1139 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
1140 mm_update_next_owner(old_mm
);
1148 static int de_thread(struct task_struct
*tsk
)
1150 struct signal_struct
*sig
= tsk
->signal
;
1151 struct sighand_struct
*oldsighand
= tsk
->sighand
;
1152 spinlock_t
*lock
= &oldsighand
->siglock
;
1154 if (thread_group_empty(tsk
))
1155 goto no_thread_group
;
1158 * Kill all other threads in the thread group.
1160 spin_lock_irq(lock
);
1161 if (signal_group_exit(sig
)) {
1163 * Another group action in progress, just
1164 * return so that the signal is processed.
1166 spin_unlock_irq(lock
);
1170 sig
->group_exit_task
= tsk
;
1171 sig
->notify_count
= zap_other_threads(tsk
);
1172 if (!thread_group_leader(tsk
))
1173 sig
->notify_count
--;
1175 while (sig
->notify_count
) {
1176 __set_current_state(TASK_KILLABLE
);
1177 spin_unlock_irq(lock
);
1179 if (__fatal_signal_pending(tsk
))
1181 spin_lock_irq(lock
);
1183 spin_unlock_irq(lock
);
1186 * At this point all other threads have exited, all we have to
1187 * do is to wait for the thread group leader to become inactive,
1188 * and to assume its PID:
1190 if (!thread_group_leader(tsk
)) {
1191 struct task_struct
*leader
= tsk
->group_leader
;
1194 cgroup_threadgroup_change_begin(tsk
);
1195 write_lock_irq(&tasklist_lock
);
1197 * Do this under tasklist_lock to ensure that
1198 * exit_notify() can't miss ->group_exit_task
1200 sig
->notify_count
= -1;
1201 if (likely(leader
->exit_state
))
1203 __set_current_state(TASK_KILLABLE
);
1204 write_unlock_irq(&tasklist_lock
);
1205 cgroup_threadgroup_change_end(tsk
);
1207 if (__fatal_signal_pending(tsk
))
1212 * The only record we have of the real-time age of a
1213 * process, regardless of execs it's done, is start_time.
1214 * All the past CPU time is accumulated in signal_struct
1215 * from sister threads now dead. But in this non-leader
1216 * exec, nothing survives from the original leader thread,
1217 * whose birth marks the true age of this process now.
1218 * When we take on its identity by switching to its PID, we
1219 * also take its birthdate (always earlier than our own).
1221 tsk
->start_time
= leader
->start_time
;
1222 tsk
->start_boottime
= leader
->start_boottime
;
1224 BUG_ON(!same_thread_group(leader
, tsk
));
1226 * An exec() starts a new thread group with the
1227 * TGID of the previous thread group. Rehash the
1228 * two threads with a switched PID, and release
1229 * the former thread group leader:
1232 /* Become a process group leader with the old leader's pid.
1233 * The old leader becomes a thread of the this thread group.
1235 exchange_tids(tsk
, leader
);
1236 transfer_pid(leader
, tsk
, PIDTYPE_TGID
);
1237 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
1238 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
1240 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
1241 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
1243 tsk
->group_leader
= tsk
;
1244 leader
->group_leader
= tsk
;
1246 tsk
->exit_signal
= SIGCHLD
;
1247 leader
->exit_signal
= -1;
1249 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
1250 leader
->exit_state
= EXIT_DEAD
;
1253 * We are going to release_task()->ptrace_unlink() silently,
1254 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1255 * the tracer wont't block again waiting for this thread.
1257 if (unlikely(leader
->ptrace
))
1258 __wake_up_parent(leader
, leader
->parent
);
1259 write_unlock_irq(&tasklist_lock
);
1260 cgroup_threadgroup_change_end(tsk
);
1262 release_task(leader
);
1265 sig
->group_exit_task
= NULL
;
1266 sig
->notify_count
= 0;
1269 /* we have changed execution domain */
1270 tsk
->exit_signal
= SIGCHLD
;
1272 BUG_ON(!thread_group_leader(tsk
));
1276 /* protects against exit_notify() and __exit_signal() */
1277 read_lock(&tasklist_lock
);
1278 sig
->group_exit_task
= NULL
;
1279 sig
->notify_count
= 0;
1280 read_unlock(&tasklist_lock
);
1286 * This function makes sure the current process has its own signal table,
1287 * so that flush_signal_handlers can later reset the handlers without
1288 * disturbing other processes. (Other processes might share the signal
1289 * table via the CLONE_SIGHAND option to clone().)
1291 static int unshare_sighand(struct task_struct
*me
)
1293 struct sighand_struct
*oldsighand
= me
->sighand
;
1295 if (refcount_read(&oldsighand
->count
) != 1) {
1296 struct sighand_struct
*newsighand
;
1298 * This ->sighand is shared with the CLONE_SIGHAND
1299 * but not CLONE_THREAD task, switch to the new one.
1301 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1305 refcount_set(&newsighand
->count
, 1);
1306 memcpy(newsighand
->action
, oldsighand
->action
,
1307 sizeof(newsighand
->action
));
1309 write_lock_irq(&tasklist_lock
);
1310 spin_lock(&oldsighand
->siglock
);
1311 rcu_assign_pointer(me
->sighand
, newsighand
);
1312 spin_unlock(&oldsighand
->siglock
);
1313 write_unlock_irq(&tasklist_lock
);
1315 __cleanup_sighand(oldsighand
);
1320 char *__get_task_comm(char *buf
, size_t buf_size
, struct task_struct
*tsk
)
1323 strncpy(buf
, tsk
->comm
, buf_size
);
1327 EXPORT_SYMBOL_GPL(__get_task_comm
);
1330 * These functions flushes out all traces of the currently running executable
1331 * so that a new one can be started
1334 void __set_task_comm(struct task_struct
*tsk
, const char *buf
, bool exec
)
1337 trace_task_rename(tsk
, buf
);
1338 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1340 perf_event_comm(tsk
, exec
);
1344 * Calling this is the point of no return. None of the failures will be
1345 * seen by userspace since either the process is already taking a fatal
1346 * signal (via de_thread() or coredump), or will have SEGV raised
1347 * (after exec_mmap()) by search_binary_handler (see below).
1349 int begin_new_exec(struct linux_binprm
* bprm
)
1351 struct task_struct
*me
= current
;
1354 /* Once we are committed compute the creds */
1355 retval
= bprm_creds_from_file(bprm
);
1360 * Ensure all future errors are fatal.
1362 bprm
->point_of_no_return
= true;
1365 * Make this the only thread in the thread group.
1367 retval
= de_thread(me
);
1372 * Must be called _before_ exec_mmap() as bprm->mm is
1373 * not visibile until then. This also enables the update
1376 set_mm_exe_file(bprm
->mm
, bprm
->file
);
1378 /* If the binary is not readable then enforce mm->dumpable=0 */
1379 would_dump(bprm
, bprm
->file
);
1380 if (bprm
->have_execfd
)
1381 would_dump(bprm
, bprm
->executable
);
1384 * Release all of the old mmap stuff
1386 acct_arg_size(bprm
, 0);
1387 retval
= exec_mmap(bprm
->mm
);
1393 #ifdef CONFIG_POSIX_TIMERS
1394 exit_itimers(me
->signal
);
1395 flush_itimer_signals();
1399 * Make the signal table private.
1401 retval
= unshare_sighand(me
);
1406 * Ensure that the uaccess routines can actually operate on userspace
1409 force_uaccess_begin();
1411 me
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
| PF_KTHREAD
|
1412 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1414 me
->personality
&= ~bprm
->per_clear
;
1417 * We have to apply CLOEXEC before we change whether the process is
1418 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1419 * trying to access the should-be-closed file descriptors of a process
1420 * undergoing exec(2).
1422 do_close_on_exec(me
->files
);
1424 if (bprm
->secureexec
) {
1425 /* Make sure parent cannot signal privileged process. */
1426 me
->pdeath_signal
= 0;
1429 * For secureexec, reset the stack limit to sane default to
1430 * avoid bad behavior from the prior rlimits. This has to
1431 * happen before arch_pick_mmap_layout(), which examines
1432 * RLIMIT_STACK, but after the point of no return to avoid
1433 * needing to clean up the change on failure.
1435 if (bprm
->rlim_stack
.rlim_cur
> _STK_LIM
)
1436 bprm
->rlim_stack
.rlim_cur
= _STK_LIM
;
1439 me
->sas_ss_sp
= me
->sas_ss_size
= 0;
1442 * Figure out dumpability. Note that this checking only of current
1443 * is wrong, but userspace depends on it. This should be testing
1444 * bprm->secureexec instead.
1446 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
||
1447 !(uid_eq(current_euid(), current_uid()) &&
1448 gid_eq(current_egid(), current_gid())))
1449 set_dumpable(current
->mm
, suid_dumpable
);
1451 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1454 __set_task_comm(me
, kbasename(bprm
->filename
), true);
1456 /* An exec changes our domain. We are no longer part of the thread
1458 WRITE_ONCE(me
->self_exec_id
, me
->self_exec_id
+ 1);
1459 flush_signal_handlers(me
, 0);
1462 * install the new credentials for this executable
1464 security_bprm_committing_creds(bprm
);
1466 commit_creds(bprm
->cred
);
1470 * Disable monitoring for regular users
1471 * when executing setuid binaries. Must
1472 * wait until new credentials are committed
1473 * by commit_creds() above
1475 if (get_dumpable(me
->mm
) != SUID_DUMP_USER
)
1476 perf_event_exit_task(me
);
1478 * cred_guard_mutex must be held at least to this point to prevent
1479 * ptrace_attach() from altering our determination of the task's
1480 * credentials; any time after this it may be unlocked.
1482 security_bprm_committed_creds(bprm
);
1484 /* Pass the opened binary to the interpreter. */
1485 if (bprm
->have_execfd
) {
1486 retval
= get_unused_fd_flags(0);
1489 fd_install(retval
, bprm
->executable
);
1490 bprm
->executable
= NULL
;
1491 bprm
->execfd
= retval
;
1496 mutex_unlock(&me
->signal
->exec_update_mutex
);
1500 EXPORT_SYMBOL(begin_new_exec
);
1502 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1504 struct inode
*inode
= file_inode(file
);
1505 if (inode_permission(inode
, MAY_READ
) < 0) {
1506 struct user_namespace
*old
, *user_ns
;
1507 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1509 /* Ensure mm->user_ns contains the executable */
1510 user_ns
= old
= bprm
->mm
->user_ns
;
1511 while ((user_ns
!= &init_user_ns
) &&
1512 !privileged_wrt_inode_uidgid(user_ns
, inode
))
1513 user_ns
= user_ns
->parent
;
1515 if (old
!= user_ns
) {
1516 bprm
->mm
->user_ns
= get_user_ns(user_ns
);
1521 EXPORT_SYMBOL(would_dump
);
1523 void setup_new_exec(struct linux_binprm
* bprm
)
1525 /* Setup things that can depend upon the personality */
1526 struct task_struct
*me
= current
;
1528 arch_pick_mmap_layout(me
->mm
, &bprm
->rlim_stack
);
1530 arch_setup_new_exec();
1532 /* Set the new mm task size. We have to do that late because it may
1533 * depend on TIF_32BIT which is only updated in flush_thread() on
1534 * some architectures like powerpc
1536 me
->mm
->task_size
= TASK_SIZE
;
1537 mutex_unlock(&me
->signal
->exec_update_mutex
);
1538 mutex_unlock(&me
->signal
->cred_guard_mutex
);
1540 EXPORT_SYMBOL(setup_new_exec
);
1542 /* Runs immediately before start_thread() takes over. */
1543 void finalize_exec(struct linux_binprm
*bprm
)
1545 /* Store any stack rlimit changes before starting thread. */
1546 task_lock(current
->group_leader
);
1547 current
->signal
->rlim
[RLIMIT_STACK
] = bprm
->rlim_stack
;
1548 task_unlock(current
->group_leader
);
1550 EXPORT_SYMBOL(finalize_exec
);
1553 * Prepare credentials and lock ->cred_guard_mutex.
1554 * setup_new_exec() commits the new creds and drops the lock.
1555 * Or, if exec fails before, free_bprm() should release ->cred and
1558 static int prepare_bprm_creds(struct linux_binprm
*bprm
)
1560 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1561 return -ERESTARTNOINTR
;
1563 bprm
->cred
= prepare_exec_creds();
1564 if (likely(bprm
->cred
))
1567 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1571 static void free_bprm(struct linux_binprm
*bprm
)
1574 acct_arg_size(bprm
, 0);
1577 free_arg_pages(bprm
);
1579 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1580 abort_creds(bprm
->cred
);
1583 allow_write_access(bprm
->file
);
1586 if (bprm
->executable
)
1587 fput(bprm
->executable
);
1588 /* If a binfmt changed the interp, free it. */
1589 if (bprm
->interp
!= bprm
->filename
)
1590 kfree(bprm
->interp
);
1591 kfree(bprm
->fdpath
);
1595 static struct linux_binprm
*alloc_bprm(int fd
, struct filename
*filename
)
1597 struct linux_binprm
*bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1598 int retval
= -ENOMEM
;
1602 if (fd
== AT_FDCWD
|| filename
->name
[0] == '/') {
1603 bprm
->filename
= filename
->name
;
1605 if (filename
->name
[0] == '\0')
1606 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d", fd
);
1608 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d/%s",
1609 fd
, filename
->name
);
1613 bprm
->filename
= bprm
->fdpath
;
1615 bprm
->interp
= bprm
->filename
;
1617 retval
= bprm_mm_init(bprm
);
1625 return ERR_PTR(retval
);
1628 int bprm_change_interp(const char *interp
, struct linux_binprm
*bprm
)
1630 /* If a binfmt changed the interp, free it first. */
1631 if (bprm
->interp
!= bprm
->filename
)
1632 kfree(bprm
->interp
);
1633 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1638 EXPORT_SYMBOL(bprm_change_interp
);
1641 * determine how safe it is to execute the proposed program
1642 * - the caller must hold ->cred_guard_mutex to protect against
1643 * PTRACE_ATTACH or seccomp thread-sync
1645 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1647 struct task_struct
*p
= current
, *t
;
1651 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1654 * This isn't strictly necessary, but it makes it harder for LSMs to
1657 if (task_no_new_privs(current
))
1658 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1662 spin_lock(&p
->fs
->lock
);
1664 while_each_thread(p
, t
) {
1670 if (p
->fs
->users
> n_fs
)
1671 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1674 spin_unlock(&p
->fs
->lock
);
1677 static void bprm_fill_uid(struct linux_binprm
*bprm
, struct file
*file
)
1679 /* Handle suid and sgid on files */
1680 struct inode
*inode
;
1685 if (!mnt_may_suid(file
->f_path
.mnt
))
1688 if (task_no_new_privs(current
))
1691 inode
= file
->f_path
.dentry
->d_inode
;
1692 mode
= READ_ONCE(inode
->i_mode
);
1693 if (!(mode
& (S_ISUID
|S_ISGID
)))
1696 /* Be careful if suid/sgid is set */
1699 /* reload atomically mode/uid/gid now that lock held */
1700 mode
= inode
->i_mode
;
1703 inode_unlock(inode
);
1705 /* We ignore suid/sgid if there are no mappings for them in the ns */
1706 if (!kuid_has_mapping(bprm
->cred
->user_ns
, uid
) ||
1707 !kgid_has_mapping(bprm
->cred
->user_ns
, gid
))
1710 if (mode
& S_ISUID
) {
1711 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1712 bprm
->cred
->euid
= uid
;
1715 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1716 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1717 bprm
->cred
->egid
= gid
;
1722 * Compute brpm->cred based upon the final binary.
1724 static int bprm_creds_from_file(struct linux_binprm
*bprm
)
1726 /* Compute creds based on which file? */
1727 struct file
*file
= bprm
->execfd_creds
? bprm
->executable
: bprm
->file
;
1729 bprm_fill_uid(bprm
, file
);
1730 return security_bprm_creds_from_file(bprm
, file
);
1734 * Fill the binprm structure from the inode.
1735 * Read the first BINPRM_BUF_SIZE bytes
1737 * This may be called multiple times for binary chains (scripts for example).
1739 static int prepare_binprm(struct linux_binprm
*bprm
)
1743 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1744 return kernel_read(bprm
->file
, bprm
->buf
, BINPRM_BUF_SIZE
, &pos
);
1748 * Arguments are '\0' separated strings found at the location bprm->p
1749 * points to; chop off the first by relocating brpm->p to right after
1750 * the first '\0' encountered.
1752 int remove_arg_zero(struct linux_binprm
*bprm
)
1755 unsigned long offset
;
1763 offset
= bprm
->p
& ~PAGE_MASK
;
1764 page
= get_arg_page(bprm
, bprm
->p
, 0);
1769 kaddr
= kmap_atomic(page
);
1771 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1772 offset
++, bprm
->p
++)
1775 kunmap_atomic(kaddr
);
1777 } while (offset
== PAGE_SIZE
);
1786 EXPORT_SYMBOL(remove_arg_zero
);
1788 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1790 * cycle the list of binary formats handler, until one recognizes the image
1792 static int search_binary_handler(struct linux_binprm
*bprm
)
1794 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1795 struct linux_binfmt
*fmt
;
1798 retval
= prepare_binprm(bprm
);
1802 retval
= security_bprm_check(bprm
);
1808 read_lock(&binfmt_lock
);
1809 list_for_each_entry(fmt
, &formats
, lh
) {
1810 if (!try_module_get(fmt
->module
))
1812 read_unlock(&binfmt_lock
);
1814 retval
= fmt
->load_binary(bprm
);
1816 read_lock(&binfmt_lock
);
1818 if (bprm
->point_of_no_return
|| (retval
!= -ENOEXEC
)) {
1819 read_unlock(&binfmt_lock
);
1823 read_unlock(&binfmt_lock
);
1826 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1827 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1829 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1838 static int exec_binprm(struct linux_binprm
*bprm
)
1840 pid_t old_pid
, old_vpid
;
1843 /* Need to fetch pid before load_binary changes it */
1844 old_pid
= current
->pid
;
1846 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1849 /* This allows 4 levels of binfmt rewrites before failing hard. */
1850 for (depth
= 0;; depth
++) {
1855 ret
= search_binary_handler(bprm
);
1858 if (!bprm
->interpreter
)
1862 bprm
->file
= bprm
->interpreter
;
1863 bprm
->interpreter
= NULL
;
1865 allow_write_access(exec
);
1866 if (unlikely(bprm
->have_execfd
)) {
1867 if (bprm
->executable
) {
1871 bprm
->executable
= exec
;
1877 trace_sched_process_exec(current
, old_pid
, bprm
);
1878 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1879 proc_exec_connector(current
);
1884 * sys_execve() executes a new program.
1886 static int bprm_execve(struct linux_binprm
*bprm
,
1887 int fd
, struct filename
*filename
, int flags
)
1890 struct files_struct
*displaced
;
1893 retval
= unshare_files(&displaced
);
1897 retval
= prepare_bprm_creds(bprm
);
1901 check_unsafe_exec(bprm
);
1902 current
->in_execve
= 1;
1904 file
= do_open_execat(fd
, filename
, flags
);
1905 retval
= PTR_ERR(file
);
1913 * Record that a name derived from an O_CLOEXEC fd will be
1914 * inaccessible after exec. Relies on having exclusive access to
1915 * current->files (due to unshare_files above).
1918 close_on_exec(fd
, rcu_dereference_raw(current
->files
->fdt
)))
1919 bprm
->interp_flags
|= BINPRM_FLAGS_PATH_INACCESSIBLE
;
1921 /* Set the unchanging part of bprm->cred */
1922 retval
= security_bprm_creds_for_exec(bprm
);
1926 retval
= exec_binprm(bprm
);
1930 /* execve succeeded */
1931 current
->fs
->in_exec
= 0;
1932 current
->in_execve
= 0;
1933 rseq_execve(current
);
1934 acct_update_integrals(current
);
1935 task_numa_free(current
, false);
1937 put_files_struct(displaced
);
1942 * If past the point of no return ensure the the code never
1943 * returns to the userspace process. Use an existing fatal
1944 * signal if present otherwise terminate the process with
1947 if (bprm
->point_of_no_return
&& !fatal_signal_pending(current
))
1948 force_sigsegv(SIGSEGV
);
1951 current
->fs
->in_exec
= 0;
1952 current
->in_execve
= 0;
1956 reset_files_struct(displaced
);
1961 static int do_execveat_common(int fd
, struct filename
*filename
,
1962 struct user_arg_ptr argv
,
1963 struct user_arg_ptr envp
,
1966 struct linux_binprm
*bprm
;
1969 if (IS_ERR(filename
))
1970 return PTR_ERR(filename
);
1973 * We move the actual failure in case of RLIMIT_NPROC excess from
1974 * set*uid() to execve() because too many poorly written programs
1975 * don't check setuid() return code. Here we additionally recheck
1976 * whether NPROC limit is still exceeded.
1978 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1979 atomic_read(¤t_user()->processes
) > rlimit(RLIMIT_NPROC
)) {
1984 /* We're below the limit (still or again), so we don't want to make
1985 * further execve() calls fail. */
1986 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1988 bprm
= alloc_bprm(fd
, filename
);
1990 retval
= PTR_ERR(bprm
);
1994 retval
= count(argv
, MAX_ARG_STRINGS
);
1997 bprm
->argc
= retval
;
1999 retval
= count(envp
, MAX_ARG_STRINGS
);
2002 bprm
->envc
= retval
;
2004 retval
= bprm_stack_limits(bprm
);
2008 retval
= copy_string_kernel(bprm
->filename
, bprm
);
2011 bprm
->exec
= bprm
->p
;
2013 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
2017 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
2021 retval
= bprm_execve(bprm
, fd
, filename
, flags
);
2030 int kernel_execve(const char *kernel_filename
,
2031 const char *const *argv
, const char *const *envp
)
2033 struct filename
*filename
;
2034 struct linux_binprm
*bprm
;
2038 filename
= getname_kernel(kernel_filename
);
2039 if (IS_ERR(filename
))
2040 return PTR_ERR(filename
);
2042 bprm
= alloc_bprm(fd
, filename
);
2044 retval
= PTR_ERR(bprm
);
2048 retval
= count_strings_kernel(argv
);
2051 bprm
->argc
= retval
;
2053 retval
= count_strings_kernel(envp
);
2056 bprm
->envc
= retval
;
2058 retval
= bprm_stack_limits(bprm
);
2062 retval
= copy_string_kernel(bprm
->filename
, bprm
);
2065 bprm
->exec
= bprm
->p
;
2067 retval
= copy_strings_kernel(bprm
->envc
, envp
, bprm
);
2071 retval
= copy_strings_kernel(bprm
->argc
, argv
, bprm
);
2075 retval
= bprm_execve(bprm
, fd
, filename
, 0);
2083 static int do_execve(struct filename
*filename
,
2084 const char __user
*const __user
*__argv
,
2085 const char __user
*const __user
*__envp
)
2087 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2088 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2089 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2092 static int do_execveat(int fd
, struct filename
*filename
,
2093 const char __user
*const __user
*__argv
,
2094 const char __user
*const __user
*__envp
,
2097 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2098 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2100 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2103 #ifdef CONFIG_COMPAT
2104 static int compat_do_execve(struct filename
*filename
,
2105 const compat_uptr_t __user
*__argv
,
2106 const compat_uptr_t __user
*__envp
)
2108 struct user_arg_ptr argv
= {
2110 .ptr
.compat
= __argv
,
2112 struct user_arg_ptr envp
= {
2114 .ptr
.compat
= __envp
,
2116 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2119 static int compat_do_execveat(int fd
, struct filename
*filename
,
2120 const compat_uptr_t __user
*__argv
,
2121 const compat_uptr_t __user
*__envp
,
2124 struct user_arg_ptr argv
= {
2126 .ptr
.compat
= __argv
,
2128 struct user_arg_ptr envp
= {
2130 .ptr
.compat
= __envp
,
2132 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2136 void set_binfmt(struct linux_binfmt
*new)
2138 struct mm_struct
*mm
= current
->mm
;
2141 module_put(mm
->binfmt
->module
);
2145 __module_get(new->module
);
2147 EXPORT_SYMBOL(set_binfmt
);
2150 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2152 void set_dumpable(struct mm_struct
*mm
, int value
)
2154 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
2157 set_mask_bits(&mm
->flags
, MMF_DUMPABLE_MASK
, value
);
2160 SYSCALL_DEFINE3(execve
,
2161 const char __user
*, filename
,
2162 const char __user
*const __user
*, argv
,
2163 const char __user
*const __user
*, envp
)
2165 return do_execve(getname(filename
), argv
, envp
);
2168 SYSCALL_DEFINE5(execveat
,
2169 int, fd
, const char __user
*, filename
,
2170 const char __user
*const __user
*, argv
,
2171 const char __user
*const __user
*, envp
,
2174 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
2176 return do_execveat(fd
,
2177 getname_flags(filename
, lookup_flags
, NULL
),
2181 #ifdef CONFIG_COMPAT
2182 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
2183 const compat_uptr_t __user
*, argv
,
2184 const compat_uptr_t __user
*, envp
)
2186 return compat_do_execve(getname(filename
), argv
, envp
);
2189 COMPAT_SYSCALL_DEFINE5(execveat
, int, fd
,
2190 const char __user
*, filename
,
2191 const compat_uptr_t __user
*, argv
,
2192 const compat_uptr_t __user
*, envp
,
2195 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
2197 return compat_do_execveat(fd
,
2198 getname_flags(filename
, lookup_flags
, NULL
),