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
)
88 if (WARN_ON(!fmt
->load_binary
))
90 write_lock(&binfmt_lock
);
91 insert
? list_add(&fmt
->lh
, &formats
) :
92 list_add_tail(&fmt
->lh
, &formats
);
93 write_unlock(&binfmt_lock
);
96 EXPORT_SYMBOL(__register_binfmt
);
98 void unregister_binfmt(struct linux_binfmt
* fmt
)
100 write_lock(&binfmt_lock
);
102 write_unlock(&binfmt_lock
);
105 EXPORT_SYMBOL(unregister_binfmt
);
107 static inline void put_binfmt(struct linux_binfmt
* fmt
)
109 module_put(fmt
->module
);
112 bool path_noexec(const struct path
*path
)
114 return (path
->mnt
->mnt_flags
& MNT_NOEXEC
) ||
115 (path
->mnt
->mnt_sb
->s_iflags
& SB_I_NOEXEC
);
117 EXPORT_SYMBOL_GPL(path_noexec
);
119 bool path_nosuid(const struct path
*path
)
121 return !mnt_may_suid(path
->mnt
) ||
122 (path
->mnt
->mnt_sb
->s_iflags
& SB_I_NOSUID
);
124 EXPORT_SYMBOL(path_nosuid
);
128 * Note that a shared library must be both readable and executable due to
131 * Also note that we take the address to load from from the file itself.
133 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
135 struct linux_binfmt
*fmt
;
137 struct filename
*tmp
= getname(library
);
138 int error
= PTR_ERR(tmp
);
139 static const struct open_flags uselib_flags
= {
140 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
141 .acc_mode
= MAY_READ
| MAY_EXEC
,
142 .intent
= LOOKUP_OPEN
,
143 .lookup_flags
= LOOKUP_FOLLOW
,
149 file
= do_filp_open(AT_FDCWD
, tmp
, &uselib_flags
);
151 error
= PTR_ERR(file
);
156 * may_open() has already checked for this, so it should be
157 * impossible to trip now. But we need to be extra cautious
158 * and check again at the very end too.
161 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
162 path_noexec(&file
->f_path
)))
169 read_lock(&binfmt_lock
);
170 list_for_each_entry(fmt
, &formats
, lh
) {
171 if (!fmt
->load_shlib
)
173 if (!try_module_get(fmt
->module
))
175 read_unlock(&binfmt_lock
);
176 error
= fmt
->load_shlib(file
);
177 read_lock(&binfmt_lock
);
179 if (error
!= -ENOEXEC
)
182 read_unlock(&binfmt_lock
);
188 #endif /* #ifdef CONFIG_USELIB */
192 * The nascent bprm->mm is not visible until exec_mmap() but it can
193 * use a lot of memory, account these pages in current->mm temporary
194 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
195 * change the counter back via acct_arg_size(0).
197 static void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
199 struct mm_struct
*mm
= current
->mm
;
200 long diff
= (long)(pages
- bprm
->vma_pages
);
205 bprm
->vma_pages
= pages
;
206 add_mm_counter(mm
, MM_ANONPAGES
, diff
);
209 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
214 unsigned int gup_flags
= FOLL_FORCE
;
216 #ifdef CONFIG_STACK_GROWSUP
218 ret
= expand_downwards(bprm
->vma
, pos
);
225 gup_flags
|= FOLL_WRITE
;
228 * We are doing an exec(). 'current' is the process
229 * doing the exec and bprm->mm is the new process's mm.
231 ret
= get_user_pages_remote(bprm
->mm
, pos
, 1, gup_flags
,
237 acct_arg_size(bprm
, vma_pages(bprm
->vma
));
242 static void put_arg_page(struct page
*page
)
247 static void free_arg_pages(struct linux_binprm
*bprm
)
251 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
254 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
257 static int __bprm_mm_init(struct linux_binprm
*bprm
)
260 struct vm_area_struct
*vma
= NULL
;
261 struct mm_struct
*mm
= bprm
->mm
;
263 bprm
->vma
= vma
= vm_area_alloc(mm
);
266 vma_set_anonymous(vma
);
268 if (mmap_write_lock_killable(mm
)) {
274 * Place the stack at the largest stack address the architecture
275 * supports. Later, we'll move this to an appropriate place. We don't
276 * use STACK_TOP because that can depend on attributes which aren't
279 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
280 vma
->vm_end
= STACK_TOP_MAX
;
281 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
282 vma
->vm_flags
= VM_SOFTDIRTY
| VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
;
283 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
285 err
= insert_vm_struct(mm
, vma
);
289 mm
->stack_vm
= mm
->total_vm
= 1;
290 mmap_write_unlock(mm
);
291 bprm
->p
= vma
->vm_end
- sizeof(void *);
294 mmap_write_unlock(mm
);
301 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
303 return len
<= MAX_ARG_STRLEN
;
308 static inline void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
312 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
317 page
= bprm
->page
[pos
/ PAGE_SIZE
];
318 if (!page
&& write
) {
319 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
322 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
328 static void put_arg_page(struct page
*page
)
332 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
335 __free_page(bprm
->page
[i
]);
336 bprm
->page
[i
] = NULL
;
340 static void free_arg_pages(struct linux_binprm
*bprm
)
344 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
345 free_arg_page(bprm
, i
);
348 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
353 static int __bprm_mm_init(struct linux_binprm
*bprm
)
355 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
359 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
361 return len
<= bprm
->p
;
364 #endif /* CONFIG_MMU */
367 * Create a new mm_struct and populate it with a temporary stack
368 * vm_area_struct. We don't have enough context at this point to set the stack
369 * flags, permissions, and offset, so we use temporary values. We'll update
370 * them later in setup_arg_pages().
372 static int bprm_mm_init(struct linux_binprm
*bprm
)
375 struct mm_struct
*mm
= NULL
;
377 bprm
->mm
= mm
= mm_alloc();
382 /* Save current stack limit for all calculations made during exec. */
383 task_lock(current
->group_leader
);
384 bprm
->rlim_stack
= current
->signal
->rlim
[RLIMIT_STACK
];
385 task_unlock(current
->group_leader
);
387 err
= __bprm_mm_init(bprm
);
402 struct user_arg_ptr
{
407 const char __user
*const __user
*native
;
409 const compat_uptr_t __user
*compat
;
414 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
416 const char __user
*native
;
419 if (unlikely(argv
.is_compat
)) {
420 compat_uptr_t compat
;
422 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
423 return ERR_PTR(-EFAULT
);
425 return compat_ptr(compat
);
429 if (get_user(native
, argv
.ptr
.native
+ nr
))
430 return ERR_PTR(-EFAULT
);
436 * count() counts the number of strings in array ARGV.
438 static int count(struct user_arg_ptr argv
, int max
)
442 if (argv
.ptr
.native
!= NULL
) {
444 const char __user
*p
= get_user_arg_ptr(argv
, i
);
456 if (fatal_signal_pending(current
))
457 return -ERESTARTNOHAND
;
464 static int count_strings_kernel(const char *const *argv
)
471 for (i
= 0; argv
[i
]; ++i
) {
472 if (i
>= MAX_ARG_STRINGS
)
474 if (fatal_signal_pending(current
))
475 return -ERESTARTNOHAND
;
481 static int bprm_stack_limits(struct linux_binprm
*bprm
)
483 unsigned long limit
, ptr_size
;
486 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
487 * (whichever is smaller) for the argv+env strings.
489 * - the remaining binfmt code will not run out of stack space,
490 * - the program will have a reasonable amount of stack left
493 limit
= _STK_LIM
/ 4 * 3;
494 limit
= min(limit
, bprm
->rlim_stack
.rlim_cur
/ 4);
496 * We've historically supported up to 32 pages (ARG_MAX)
497 * of argument strings even with small stacks
499 limit
= max_t(unsigned long, limit
, ARG_MAX
);
501 * We must account for the size of all the argv and envp pointers to
502 * the argv and envp strings, since they will also take up space in
503 * the stack. They aren't stored until much later when we can't
504 * signal to the parent that the child has run out of stack space.
505 * Instead, calculate it here so it's possible to fail gracefully.
507 ptr_size
= (bprm
->argc
+ bprm
->envc
) * sizeof(void *);
508 if (limit
<= ptr_size
)
512 bprm
->argmin
= bprm
->p
- limit
;
517 * 'copy_strings()' copies argument/environment strings from the old
518 * processes's memory to the new process's stack. The call to get_user_pages()
519 * ensures the destination page is created and not swapped out.
521 static int copy_strings(int argc
, struct user_arg_ptr argv
,
522 struct linux_binprm
*bprm
)
524 struct page
*kmapped_page
= NULL
;
526 unsigned long kpos
= 0;
530 const char __user
*str
;
535 str
= get_user_arg_ptr(argv
, argc
);
539 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
544 if (!valid_arg_len(bprm
, len
))
547 /* We're going to work our way backwords. */
552 if (bprm
->p
< bprm
->argmin
)
557 int offset
, bytes_to_copy
;
559 if (fatal_signal_pending(current
)) {
560 ret
= -ERESTARTNOHAND
;
565 offset
= pos
% PAGE_SIZE
;
569 bytes_to_copy
= offset
;
570 if (bytes_to_copy
> len
)
573 offset
-= bytes_to_copy
;
574 pos
-= bytes_to_copy
;
575 str
-= bytes_to_copy
;
576 len
-= bytes_to_copy
;
578 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
581 page
= get_arg_page(bprm
, pos
, 1);
588 flush_kernel_dcache_page(kmapped_page
);
589 kunmap(kmapped_page
);
590 put_arg_page(kmapped_page
);
593 kaddr
= kmap(kmapped_page
);
594 kpos
= pos
& PAGE_MASK
;
595 flush_arg_page(bprm
, kpos
, kmapped_page
);
597 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
606 flush_kernel_dcache_page(kmapped_page
);
607 kunmap(kmapped_page
);
608 put_arg_page(kmapped_page
);
614 * Copy and argument/environment string from the kernel to the processes stack.
616 int copy_string_kernel(const char *arg
, struct linux_binprm
*bprm
)
618 int len
= strnlen(arg
, MAX_ARG_STRLEN
) + 1 /* terminating NUL */;
619 unsigned long pos
= bprm
->p
;
623 if (!valid_arg_len(bprm
, len
))
626 /* We're going to work our way backwards. */
629 if (IS_ENABLED(CONFIG_MMU
) && bprm
->p
< bprm
->argmin
)
633 unsigned int bytes_to_copy
= min_t(unsigned int, len
,
634 min_not_zero(offset_in_page(pos
), PAGE_SIZE
));
638 pos
-= bytes_to_copy
;
639 arg
-= bytes_to_copy
;
640 len
-= bytes_to_copy
;
642 page
= get_arg_page(bprm
, pos
, 1);
645 kaddr
= kmap_atomic(page
);
646 flush_arg_page(bprm
, pos
& PAGE_MASK
, page
);
647 memcpy(kaddr
+ offset_in_page(pos
), arg
, bytes_to_copy
);
648 flush_kernel_dcache_page(page
);
649 kunmap_atomic(kaddr
);
655 EXPORT_SYMBOL(copy_string_kernel
);
657 static int copy_strings_kernel(int argc
, const char *const *argv
,
658 struct linux_binprm
*bprm
)
661 int ret
= copy_string_kernel(argv
[argc
], bprm
);
664 if (fatal_signal_pending(current
))
665 return -ERESTARTNOHAND
;
674 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
675 * the binfmt code determines where the new stack should reside, we shift it to
676 * its final location. The process proceeds as follows:
678 * 1) Use shift to calculate the new vma endpoints.
679 * 2) Extend vma to cover both the old and new ranges. This ensures the
680 * arguments passed to subsequent functions are consistent.
681 * 3) Move vma's page tables to the new range.
682 * 4) Free up any cleared pgd range.
683 * 5) Shrink the vma to cover only the new range.
685 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
687 struct mm_struct
*mm
= vma
->vm_mm
;
688 unsigned long old_start
= vma
->vm_start
;
689 unsigned long old_end
= vma
->vm_end
;
690 unsigned long length
= old_end
- old_start
;
691 unsigned long new_start
= old_start
- shift
;
692 unsigned long new_end
= old_end
- shift
;
693 struct mmu_gather tlb
;
695 BUG_ON(new_start
> new_end
);
698 * ensure there are no vmas between where we want to go
701 if (vma
!= find_vma(mm
, new_start
))
705 * cover the whole range: [new_start, old_end)
707 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
711 * move the page tables downwards, on failure we rely on
712 * process cleanup to remove whatever mess we made.
714 if (length
!= move_page_tables(vma
, old_start
,
715 vma
, new_start
, length
, false))
719 tlb_gather_mmu(&tlb
, mm
, old_start
, old_end
);
720 if (new_end
> old_start
) {
722 * when the old and new regions overlap clear from new_end.
724 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
725 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
728 * otherwise, clean from old_start; this is done to not touch
729 * the address space in [new_end, old_start) some architectures
730 * have constraints on va-space that make this illegal (IA64) -
731 * for the others its just a little faster.
733 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
734 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
736 tlb_finish_mmu(&tlb
, old_start
, old_end
);
739 * Shrink the vma to just the new range. Always succeeds.
741 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
747 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
748 * the stack is optionally relocated, and some extra space is added.
750 int setup_arg_pages(struct linux_binprm
*bprm
,
751 unsigned long stack_top
,
752 int executable_stack
)
755 unsigned long stack_shift
;
756 struct mm_struct
*mm
= current
->mm
;
757 struct vm_area_struct
*vma
= bprm
->vma
;
758 struct vm_area_struct
*prev
= NULL
;
759 unsigned long vm_flags
;
760 unsigned long stack_base
;
761 unsigned long stack_size
;
762 unsigned long stack_expand
;
763 unsigned long rlim_stack
;
765 #ifdef CONFIG_STACK_GROWSUP
766 /* Limit stack size */
767 stack_base
= bprm
->rlim_stack
.rlim_max
;
769 stack_base
= calc_max_stack_size(stack_base
);
771 /* Add space for stack randomization. */
772 stack_base
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
774 /* Make sure we didn't let the argument array grow too large. */
775 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
778 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
780 stack_shift
= vma
->vm_start
- stack_base
;
781 mm
->arg_start
= bprm
->p
- stack_shift
;
782 bprm
->p
= vma
->vm_end
- stack_shift
;
784 stack_top
= arch_align_stack(stack_top
);
785 stack_top
= PAGE_ALIGN(stack_top
);
787 if (unlikely(stack_top
< mmap_min_addr
) ||
788 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
791 stack_shift
= vma
->vm_end
- stack_top
;
793 bprm
->p
-= stack_shift
;
794 mm
->arg_start
= bprm
->p
;
798 bprm
->loader
-= stack_shift
;
799 bprm
->exec
-= stack_shift
;
801 if (mmap_write_lock_killable(mm
))
804 vm_flags
= VM_STACK_FLAGS
;
807 * Adjust stack execute permissions; explicitly enable for
808 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
809 * (arch default) otherwise.
811 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
813 else if (executable_stack
== EXSTACK_DISABLE_X
)
814 vm_flags
&= ~VM_EXEC
;
815 vm_flags
|= mm
->def_flags
;
816 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
818 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
824 if (unlikely(vm_flags
& VM_EXEC
)) {
825 pr_warn_once("process '%pD4' started with executable stack\n",
829 /* Move stack pages down in memory. */
831 ret
= shift_arg_pages(vma
, stack_shift
);
836 /* mprotect_fixup is overkill to remove the temporary stack flags */
837 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
839 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
840 stack_size
= vma
->vm_end
- vma
->vm_start
;
842 * Align this down to a page boundary as expand_stack
845 rlim_stack
= bprm
->rlim_stack
.rlim_cur
& PAGE_MASK
;
846 #ifdef CONFIG_STACK_GROWSUP
847 if (stack_size
+ stack_expand
> rlim_stack
)
848 stack_base
= vma
->vm_start
+ rlim_stack
;
850 stack_base
= vma
->vm_end
+ stack_expand
;
852 if (stack_size
+ stack_expand
> rlim_stack
)
853 stack_base
= vma
->vm_end
- rlim_stack
;
855 stack_base
= vma
->vm_start
- stack_expand
;
857 current
->mm
->start_stack
= bprm
->p
;
858 ret
= expand_stack(vma
, stack_base
);
863 mmap_write_unlock(mm
);
866 EXPORT_SYMBOL(setup_arg_pages
);
871 * Transfer the program arguments and environment from the holding pages
872 * onto the stack. The provided stack pointer is adjusted accordingly.
874 int transfer_args_to_stack(struct linux_binprm
*bprm
,
875 unsigned long *sp_location
)
877 unsigned long index
, stop
, sp
;
880 stop
= bprm
->p
>> PAGE_SHIFT
;
883 for (index
= MAX_ARG_PAGES
- 1; index
>= stop
; index
--) {
884 unsigned int offset
= index
== stop
? bprm
->p
& ~PAGE_MASK
: 0;
885 char *src
= kmap(bprm
->page
[index
]) + offset
;
886 sp
-= PAGE_SIZE
- offset
;
887 if (copy_to_user((void *) sp
, src
, PAGE_SIZE
- offset
) != 0)
889 kunmap(bprm
->page
[index
]);
899 EXPORT_SYMBOL(transfer_args_to_stack
);
901 #endif /* CONFIG_MMU */
903 static struct file
*do_open_execat(int fd
, struct filename
*name
, int flags
)
907 struct open_flags open_exec_flags
= {
908 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
909 .acc_mode
= MAY_EXEC
,
910 .intent
= LOOKUP_OPEN
,
911 .lookup_flags
= LOOKUP_FOLLOW
,
914 if ((flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
915 return ERR_PTR(-EINVAL
);
916 if (flags
& AT_SYMLINK_NOFOLLOW
)
917 open_exec_flags
.lookup_flags
&= ~LOOKUP_FOLLOW
;
918 if (flags
& AT_EMPTY_PATH
)
919 open_exec_flags
.lookup_flags
|= LOOKUP_EMPTY
;
921 file
= do_filp_open(fd
, name
, &open_exec_flags
);
926 * may_open() has already checked for this, so it should be
927 * impossible to trip now. But we need to be extra cautious
928 * and check again at the very end too.
931 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
932 path_noexec(&file
->f_path
)))
935 err
= deny_write_access(file
);
939 if (name
->name
[0] != '\0')
950 struct file
*open_exec(const char *name
)
952 struct filename
*filename
= getname_kernel(name
);
953 struct file
*f
= ERR_CAST(filename
);
955 if (!IS_ERR(filename
)) {
956 f
= do_open_execat(AT_FDCWD
, filename
, 0);
961 EXPORT_SYMBOL(open_exec
);
963 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
964 defined(CONFIG_BINFMT_ELF_FDPIC)
965 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
967 ssize_t res
= vfs_read(file
, (void __user
*)addr
, len
, &pos
);
969 flush_icache_user_range(addr
, addr
+ len
);
972 EXPORT_SYMBOL(read_code
);
976 * Maps the mm_struct mm into the current task struct.
977 * On success, this function returns with exec_update_lock
980 static int exec_mmap(struct mm_struct
*mm
)
982 struct task_struct
*tsk
;
983 struct mm_struct
*old_mm
, *active_mm
;
986 /* Notify parent that we're no longer interested in the old VM */
988 old_mm
= current
->mm
;
989 exec_mm_release(tsk
, old_mm
);
993 ret
= down_write_killable(&tsk
->signal
->exec_update_lock
);
999 * Make sure that if there is a core dump in progress
1000 * for the old mm, we get out and die instead of going
1001 * through with the exec. We must hold mmap_lock around
1002 * checking core_state and changing tsk->mm.
1004 mmap_read_lock(old_mm
);
1005 if (unlikely(old_mm
->core_state
)) {
1006 mmap_read_unlock(old_mm
);
1007 up_write(&tsk
->signal
->exec_update_lock
);
1013 membarrier_exec_mmap(mm
);
1015 local_irq_disable();
1016 active_mm
= tsk
->active_mm
;
1017 tsk
->active_mm
= mm
;
1020 * This prevents preemption while active_mm is being loaded and
1021 * it and mm are being updated, which could cause problems for
1022 * lazy tlb mm refcounting when these are updated by context
1023 * switches. Not all architectures can handle irqs off over
1026 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1028 activate_mm(active_mm
, mm
);
1029 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1031 tsk
->mm
->vmacache_seqnum
= 0;
1032 vmacache_flush(tsk
);
1035 mmap_read_unlock(old_mm
);
1036 BUG_ON(active_mm
!= old_mm
);
1037 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
1038 mm_update_next_owner(old_mm
);
1046 static int de_thread(struct task_struct
*tsk
)
1048 struct signal_struct
*sig
= tsk
->signal
;
1049 struct sighand_struct
*oldsighand
= tsk
->sighand
;
1050 spinlock_t
*lock
= &oldsighand
->siglock
;
1052 if (thread_group_empty(tsk
))
1053 goto no_thread_group
;
1056 * Kill all other threads in the thread group.
1058 spin_lock_irq(lock
);
1059 if (signal_group_exit(sig
)) {
1061 * Another group action in progress, just
1062 * return so that the signal is processed.
1064 spin_unlock_irq(lock
);
1068 sig
->group_exit_task
= tsk
;
1069 sig
->notify_count
= zap_other_threads(tsk
);
1070 if (!thread_group_leader(tsk
))
1071 sig
->notify_count
--;
1073 while (sig
->notify_count
) {
1074 __set_current_state(TASK_KILLABLE
);
1075 spin_unlock_irq(lock
);
1077 if (__fatal_signal_pending(tsk
))
1079 spin_lock_irq(lock
);
1081 spin_unlock_irq(lock
);
1084 * At this point all other threads have exited, all we have to
1085 * do is to wait for the thread group leader to become inactive,
1086 * and to assume its PID:
1088 if (!thread_group_leader(tsk
)) {
1089 struct task_struct
*leader
= tsk
->group_leader
;
1092 cgroup_threadgroup_change_begin(tsk
);
1093 write_lock_irq(&tasklist_lock
);
1095 * Do this under tasklist_lock to ensure that
1096 * exit_notify() can't miss ->group_exit_task
1098 sig
->notify_count
= -1;
1099 if (likely(leader
->exit_state
))
1101 __set_current_state(TASK_KILLABLE
);
1102 write_unlock_irq(&tasklist_lock
);
1103 cgroup_threadgroup_change_end(tsk
);
1105 if (__fatal_signal_pending(tsk
))
1110 * The only record we have of the real-time age of a
1111 * process, regardless of execs it's done, is start_time.
1112 * All the past CPU time is accumulated in signal_struct
1113 * from sister threads now dead. But in this non-leader
1114 * exec, nothing survives from the original leader thread,
1115 * whose birth marks the true age of this process now.
1116 * When we take on its identity by switching to its PID, we
1117 * also take its birthdate (always earlier than our own).
1119 tsk
->start_time
= leader
->start_time
;
1120 tsk
->start_boottime
= leader
->start_boottime
;
1122 BUG_ON(!same_thread_group(leader
, tsk
));
1124 * An exec() starts a new thread group with the
1125 * TGID of the previous thread group. Rehash the
1126 * two threads with a switched PID, and release
1127 * the former thread group leader:
1130 /* Become a process group leader with the old leader's pid.
1131 * The old leader becomes a thread of the this thread group.
1133 exchange_tids(tsk
, leader
);
1134 transfer_pid(leader
, tsk
, PIDTYPE_TGID
);
1135 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
1136 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
1138 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
1139 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
1141 tsk
->group_leader
= tsk
;
1142 leader
->group_leader
= tsk
;
1144 tsk
->exit_signal
= SIGCHLD
;
1145 leader
->exit_signal
= -1;
1147 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
1148 leader
->exit_state
= EXIT_DEAD
;
1151 * We are going to release_task()->ptrace_unlink() silently,
1152 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1153 * the tracer wont't block again waiting for this thread.
1155 if (unlikely(leader
->ptrace
))
1156 __wake_up_parent(leader
, leader
->parent
);
1157 write_unlock_irq(&tasklist_lock
);
1158 cgroup_threadgroup_change_end(tsk
);
1160 release_task(leader
);
1163 sig
->group_exit_task
= NULL
;
1164 sig
->notify_count
= 0;
1167 /* we have changed execution domain */
1168 tsk
->exit_signal
= SIGCHLD
;
1170 BUG_ON(!thread_group_leader(tsk
));
1174 /* protects against exit_notify() and __exit_signal() */
1175 read_lock(&tasklist_lock
);
1176 sig
->group_exit_task
= NULL
;
1177 sig
->notify_count
= 0;
1178 read_unlock(&tasklist_lock
);
1184 * This function makes sure the current process has its own signal table,
1185 * so that flush_signal_handlers can later reset the handlers without
1186 * disturbing other processes. (Other processes might share the signal
1187 * table via the CLONE_SIGHAND option to clone().)
1189 static int unshare_sighand(struct task_struct
*me
)
1191 struct sighand_struct
*oldsighand
= me
->sighand
;
1193 if (refcount_read(&oldsighand
->count
) != 1) {
1194 struct sighand_struct
*newsighand
;
1196 * This ->sighand is shared with the CLONE_SIGHAND
1197 * but not CLONE_THREAD task, switch to the new one.
1199 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1203 refcount_set(&newsighand
->count
, 1);
1204 memcpy(newsighand
->action
, oldsighand
->action
,
1205 sizeof(newsighand
->action
));
1207 write_lock_irq(&tasklist_lock
);
1208 spin_lock(&oldsighand
->siglock
);
1209 rcu_assign_pointer(me
->sighand
, newsighand
);
1210 spin_unlock(&oldsighand
->siglock
);
1211 write_unlock_irq(&tasklist_lock
);
1213 __cleanup_sighand(oldsighand
);
1218 char *__get_task_comm(char *buf
, size_t buf_size
, struct task_struct
*tsk
)
1221 strncpy(buf
, tsk
->comm
, buf_size
);
1225 EXPORT_SYMBOL_GPL(__get_task_comm
);
1228 * These functions flushes out all traces of the currently running executable
1229 * so that a new one can be started
1232 void __set_task_comm(struct task_struct
*tsk
, const char *buf
, bool exec
)
1235 trace_task_rename(tsk
, buf
);
1236 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1238 perf_event_comm(tsk
, exec
);
1242 * Calling this is the point of no return. None of the failures will be
1243 * seen by userspace since either the process is already taking a fatal
1244 * signal (via de_thread() or coredump), or will have SEGV raised
1245 * (after exec_mmap()) by search_binary_handler (see below).
1247 int begin_new_exec(struct linux_binprm
* bprm
)
1249 struct task_struct
*me
= current
;
1252 /* Once we are committed compute the creds */
1253 retval
= bprm_creds_from_file(bprm
);
1258 * Ensure all future errors are fatal.
1260 bprm
->point_of_no_return
= true;
1263 * Make this the only thread in the thread group.
1265 retval
= de_thread(me
);
1270 * Cancel any io_uring activity across execve
1272 io_uring_task_cancel();
1274 /* Ensure the files table is not shared. */
1275 retval
= unshare_files();
1280 * Must be called _before_ exec_mmap() as bprm->mm is
1281 * not visibile until then. This also enables the update
1284 set_mm_exe_file(bprm
->mm
, bprm
->file
);
1286 /* If the binary is not readable then enforce mm->dumpable=0 */
1287 would_dump(bprm
, bprm
->file
);
1288 if (bprm
->have_execfd
)
1289 would_dump(bprm
, bprm
->executable
);
1292 * Release all of the old mmap stuff
1294 acct_arg_size(bprm
, 0);
1295 retval
= exec_mmap(bprm
->mm
);
1301 #ifdef CONFIG_POSIX_TIMERS
1302 exit_itimers(me
->signal
);
1303 flush_itimer_signals();
1307 * Make the signal table private.
1309 retval
= unshare_sighand(me
);
1314 * Ensure that the uaccess routines can actually operate on userspace
1317 force_uaccess_begin();
1319 me
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
| PF_KTHREAD
|
1320 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1322 me
->personality
&= ~bprm
->per_clear
;
1324 clear_syscall_work_syscall_user_dispatch(me
);
1327 * We have to apply CLOEXEC before we change whether the process is
1328 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1329 * trying to access the should-be-closed file descriptors of a process
1330 * undergoing exec(2).
1332 do_close_on_exec(me
->files
);
1334 if (bprm
->secureexec
) {
1335 /* Make sure parent cannot signal privileged process. */
1336 me
->pdeath_signal
= 0;
1339 * For secureexec, reset the stack limit to sane default to
1340 * avoid bad behavior from the prior rlimits. This has to
1341 * happen before arch_pick_mmap_layout(), which examines
1342 * RLIMIT_STACK, but after the point of no return to avoid
1343 * needing to clean up the change on failure.
1345 if (bprm
->rlim_stack
.rlim_cur
> _STK_LIM
)
1346 bprm
->rlim_stack
.rlim_cur
= _STK_LIM
;
1349 me
->sas_ss_sp
= me
->sas_ss_size
= 0;
1352 * Figure out dumpability. Note that this checking only of current
1353 * is wrong, but userspace depends on it. This should be testing
1354 * bprm->secureexec instead.
1356 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
||
1357 !(uid_eq(current_euid(), current_uid()) &&
1358 gid_eq(current_egid(), current_gid())))
1359 set_dumpable(current
->mm
, suid_dumpable
);
1361 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1364 __set_task_comm(me
, kbasename(bprm
->filename
), true);
1366 /* An exec changes our domain. We are no longer part of the thread
1368 WRITE_ONCE(me
->self_exec_id
, me
->self_exec_id
+ 1);
1369 flush_signal_handlers(me
, 0);
1372 * install the new credentials for this executable
1374 security_bprm_committing_creds(bprm
);
1376 commit_creds(bprm
->cred
);
1380 * Disable monitoring for regular users
1381 * when executing setuid binaries. Must
1382 * wait until new credentials are committed
1383 * by commit_creds() above
1385 if (get_dumpable(me
->mm
) != SUID_DUMP_USER
)
1386 perf_event_exit_task(me
);
1388 * cred_guard_mutex must be held at least to this point to prevent
1389 * ptrace_attach() from altering our determination of the task's
1390 * credentials; any time after this it may be unlocked.
1392 security_bprm_committed_creds(bprm
);
1394 /* Pass the opened binary to the interpreter. */
1395 if (bprm
->have_execfd
) {
1396 retval
= get_unused_fd_flags(0);
1399 fd_install(retval
, bprm
->executable
);
1400 bprm
->executable
= NULL
;
1401 bprm
->execfd
= retval
;
1406 up_write(&me
->signal
->exec_update_lock
);
1410 EXPORT_SYMBOL(begin_new_exec
);
1412 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1414 struct inode
*inode
= file_inode(file
);
1415 if (inode_permission(inode
, MAY_READ
) < 0) {
1416 struct user_namespace
*old
, *user_ns
;
1417 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1419 /* Ensure mm->user_ns contains the executable */
1420 user_ns
= old
= bprm
->mm
->user_ns
;
1421 while ((user_ns
!= &init_user_ns
) &&
1422 !privileged_wrt_inode_uidgid(user_ns
, inode
))
1423 user_ns
= user_ns
->parent
;
1425 if (old
!= user_ns
) {
1426 bprm
->mm
->user_ns
= get_user_ns(user_ns
);
1431 EXPORT_SYMBOL(would_dump
);
1433 void setup_new_exec(struct linux_binprm
* bprm
)
1435 /* Setup things that can depend upon the personality */
1436 struct task_struct
*me
= current
;
1438 arch_pick_mmap_layout(me
->mm
, &bprm
->rlim_stack
);
1440 arch_setup_new_exec();
1442 /* Set the new mm task size. We have to do that late because it may
1443 * depend on TIF_32BIT which is only updated in flush_thread() on
1444 * some architectures like powerpc
1446 me
->mm
->task_size
= TASK_SIZE
;
1447 up_write(&me
->signal
->exec_update_lock
);
1448 mutex_unlock(&me
->signal
->cred_guard_mutex
);
1450 EXPORT_SYMBOL(setup_new_exec
);
1452 /* Runs immediately before start_thread() takes over. */
1453 void finalize_exec(struct linux_binprm
*bprm
)
1455 /* Store any stack rlimit changes before starting thread. */
1456 task_lock(current
->group_leader
);
1457 current
->signal
->rlim
[RLIMIT_STACK
] = bprm
->rlim_stack
;
1458 task_unlock(current
->group_leader
);
1460 EXPORT_SYMBOL(finalize_exec
);
1463 * Prepare credentials and lock ->cred_guard_mutex.
1464 * setup_new_exec() commits the new creds and drops the lock.
1465 * Or, if exec fails before, free_bprm() should release ->cred and
1468 static int prepare_bprm_creds(struct linux_binprm
*bprm
)
1470 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1471 return -ERESTARTNOINTR
;
1473 bprm
->cred
= prepare_exec_creds();
1474 if (likely(bprm
->cred
))
1477 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1481 static void free_bprm(struct linux_binprm
*bprm
)
1484 acct_arg_size(bprm
, 0);
1487 free_arg_pages(bprm
);
1489 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1490 abort_creds(bprm
->cred
);
1493 allow_write_access(bprm
->file
);
1496 if (bprm
->executable
)
1497 fput(bprm
->executable
);
1498 /* If a binfmt changed the interp, free it. */
1499 if (bprm
->interp
!= bprm
->filename
)
1500 kfree(bprm
->interp
);
1501 kfree(bprm
->fdpath
);
1505 static struct linux_binprm
*alloc_bprm(int fd
, struct filename
*filename
)
1507 struct linux_binprm
*bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1508 int retval
= -ENOMEM
;
1512 if (fd
== AT_FDCWD
|| filename
->name
[0] == '/') {
1513 bprm
->filename
= filename
->name
;
1515 if (filename
->name
[0] == '\0')
1516 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d", fd
);
1518 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d/%s",
1519 fd
, filename
->name
);
1523 bprm
->filename
= bprm
->fdpath
;
1525 bprm
->interp
= bprm
->filename
;
1527 retval
= bprm_mm_init(bprm
);
1535 return ERR_PTR(retval
);
1538 int bprm_change_interp(const char *interp
, struct linux_binprm
*bprm
)
1540 /* If a binfmt changed the interp, free it first. */
1541 if (bprm
->interp
!= bprm
->filename
)
1542 kfree(bprm
->interp
);
1543 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1548 EXPORT_SYMBOL(bprm_change_interp
);
1551 * determine how safe it is to execute the proposed program
1552 * - the caller must hold ->cred_guard_mutex to protect against
1553 * PTRACE_ATTACH or seccomp thread-sync
1555 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1557 struct task_struct
*p
= current
, *t
;
1561 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1564 * This isn't strictly necessary, but it makes it harder for LSMs to
1567 if (task_no_new_privs(current
))
1568 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1572 spin_lock(&p
->fs
->lock
);
1574 while_each_thread(p
, t
) {
1580 if (p
->fs
->users
> n_fs
)
1581 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1584 spin_unlock(&p
->fs
->lock
);
1587 static void bprm_fill_uid(struct linux_binprm
*bprm
, struct file
*file
)
1589 /* Handle suid and sgid on files */
1590 struct inode
*inode
;
1595 if (path_nosuid(&file
->f_path
))
1598 if (task_no_new_privs(current
))
1601 inode
= file
->f_path
.dentry
->d_inode
;
1602 mode
= READ_ONCE(inode
->i_mode
);
1603 if (!(mode
& (S_ISUID
|S_ISGID
)))
1606 /* Be careful if suid/sgid is set */
1609 /* reload atomically mode/uid/gid now that lock held */
1610 mode
= inode
->i_mode
;
1613 inode_unlock(inode
);
1615 /* We ignore suid/sgid if there are no mappings for them in the ns */
1616 if (!kuid_has_mapping(bprm
->cred
->user_ns
, uid
) ||
1617 !kgid_has_mapping(bprm
->cred
->user_ns
, gid
))
1620 if (mode
& S_ISUID
) {
1621 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1622 bprm
->cred
->euid
= uid
;
1625 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1626 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1627 bprm
->cred
->egid
= gid
;
1632 * Compute brpm->cred based upon the final binary.
1634 static int bprm_creds_from_file(struct linux_binprm
*bprm
)
1636 /* Compute creds based on which file? */
1637 struct file
*file
= bprm
->execfd_creds
? bprm
->executable
: bprm
->file
;
1639 bprm_fill_uid(bprm
, file
);
1640 return security_bprm_creds_from_file(bprm
, file
);
1644 * Fill the binprm structure from the inode.
1645 * Read the first BINPRM_BUF_SIZE bytes
1647 * This may be called multiple times for binary chains (scripts for example).
1649 static int prepare_binprm(struct linux_binprm
*bprm
)
1653 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1654 return kernel_read(bprm
->file
, bprm
->buf
, BINPRM_BUF_SIZE
, &pos
);
1658 * Arguments are '\0' separated strings found at the location bprm->p
1659 * points to; chop off the first by relocating brpm->p to right after
1660 * the first '\0' encountered.
1662 int remove_arg_zero(struct linux_binprm
*bprm
)
1665 unsigned long offset
;
1673 offset
= bprm
->p
& ~PAGE_MASK
;
1674 page
= get_arg_page(bprm
, bprm
->p
, 0);
1679 kaddr
= kmap_atomic(page
);
1681 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1682 offset
++, bprm
->p
++)
1685 kunmap_atomic(kaddr
);
1687 } while (offset
== PAGE_SIZE
);
1696 EXPORT_SYMBOL(remove_arg_zero
);
1698 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1700 * cycle the list of binary formats handler, until one recognizes the image
1702 static int search_binary_handler(struct linux_binprm
*bprm
)
1704 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1705 struct linux_binfmt
*fmt
;
1708 retval
= prepare_binprm(bprm
);
1712 retval
= security_bprm_check(bprm
);
1718 read_lock(&binfmt_lock
);
1719 list_for_each_entry(fmt
, &formats
, lh
) {
1720 if (!try_module_get(fmt
->module
))
1722 read_unlock(&binfmt_lock
);
1724 retval
= fmt
->load_binary(bprm
);
1726 read_lock(&binfmt_lock
);
1728 if (bprm
->point_of_no_return
|| (retval
!= -ENOEXEC
)) {
1729 read_unlock(&binfmt_lock
);
1733 read_unlock(&binfmt_lock
);
1736 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1737 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1739 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1748 static int exec_binprm(struct linux_binprm
*bprm
)
1750 pid_t old_pid
, old_vpid
;
1753 /* Need to fetch pid before load_binary changes it */
1754 old_pid
= current
->pid
;
1756 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1759 /* This allows 4 levels of binfmt rewrites before failing hard. */
1760 for (depth
= 0;; depth
++) {
1765 ret
= search_binary_handler(bprm
);
1768 if (!bprm
->interpreter
)
1772 bprm
->file
= bprm
->interpreter
;
1773 bprm
->interpreter
= NULL
;
1775 allow_write_access(exec
);
1776 if (unlikely(bprm
->have_execfd
)) {
1777 if (bprm
->executable
) {
1781 bprm
->executable
= exec
;
1787 trace_sched_process_exec(current
, old_pid
, bprm
);
1788 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1789 proc_exec_connector(current
);
1794 * sys_execve() executes a new program.
1796 static int bprm_execve(struct linux_binprm
*bprm
,
1797 int fd
, struct filename
*filename
, int flags
)
1802 retval
= prepare_bprm_creds(bprm
);
1806 check_unsafe_exec(bprm
);
1807 current
->in_execve
= 1;
1809 file
= do_open_execat(fd
, filename
, flags
);
1810 retval
= PTR_ERR(file
);
1818 * Record that a name derived from an O_CLOEXEC fd will be
1819 * inaccessible after exec. This allows the code in exec to
1820 * choose to fail when the executable is not mmaped into the
1821 * interpreter and an open file descriptor is not passed to
1822 * the interpreter. This makes for a better user experience
1823 * than having the interpreter start and then immediately fail
1824 * when it finds the executable is inaccessible.
1826 if (bprm
->fdpath
&& get_close_on_exec(fd
))
1827 bprm
->interp_flags
|= BINPRM_FLAGS_PATH_INACCESSIBLE
;
1829 /* Set the unchanging part of bprm->cred */
1830 retval
= security_bprm_creds_for_exec(bprm
);
1834 retval
= exec_binprm(bprm
);
1838 /* execve succeeded */
1839 current
->fs
->in_exec
= 0;
1840 current
->in_execve
= 0;
1841 rseq_execve(current
);
1842 acct_update_integrals(current
);
1843 task_numa_free(current
, false);
1848 * If past the point of no return ensure the the code never
1849 * returns to the userspace process. Use an existing fatal
1850 * signal if present otherwise terminate the process with
1853 if (bprm
->point_of_no_return
&& !fatal_signal_pending(current
))
1854 force_sigsegv(SIGSEGV
);
1857 current
->fs
->in_exec
= 0;
1858 current
->in_execve
= 0;
1863 static int do_execveat_common(int fd
, struct filename
*filename
,
1864 struct user_arg_ptr argv
,
1865 struct user_arg_ptr envp
,
1868 struct linux_binprm
*bprm
;
1871 if (IS_ERR(filename
))
1872 return PTR_ERR(filename
);
1875 * We move the actual failure in case of RLIMIT_NPROC excess from
1876 * set*uid() to execve() because too many poorly written programs
1877 * don't check setuid() return code. Here we additionally recheck
1878 * whether NPROC limit is still exceeded.
1880 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1881 atomic_read(¤t_user()->processes
) > rlimit(RLIMIT_NPROC
)) {
1886 /* We're below the limit (still or again), so we don't want to make
1887 * further execve() calls fail. */
1888 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1890 bprm
= alloc_bprm(fd
, filename
);
1892 retval
= PTR_ERR(bprm
);
1896 retval
= count(argv
, MAX_ARG_STRINGS
);
1899 bprm
->argc
= retval
;
1901 retval
= count(envp
, MAX_ARG_STRINGS
);
1904 bprm
->envc
= retval
;
1906 retval
= bprm_stack_limits(bprm
);
1910 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1913 bprm
->exec
= bprm
->p
;
1915 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1919 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1923 retval
= bprm_execve(bprm
, fd
, filename
, flags
);
1932 int kernel_execve(const char *kernel_filename
,
1933 const char *const *argv
, const char *const *envp
)
1935 struct filename
*filename
;
1936 struct linux_binprm
*bprm
;
1940 filename
= getname_kernel(kernel_filename
);
1941 if (IS_ERR(filename
))
1942 return PTR_ERR(filename
);
1944 bprm
= alloc_bprm(fd
, filename
);
1946 retval
= PTR_ERR(bprm
);
1950 retval
= count_strings_kernel(argv
);
1953 bprm
->argc
= retval
;
1955 retval
= count_strings_kernel(envp
);
1958 bprm
->envc
= retval
;
1960 retval
= bprm_stack_limits(bprm
);
1964 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1967 bprm
->exec
= bprm
->p
;
1969 retval
= copy_strings_kernel(bprm
->envc
, envp
, bprm
);
1973 retval
= copy_strings_kernel(bprm
->argc
, argv
, bprm
);
1977 retval
= bprm_execve(bprm
, fd
, filename
, 0);
1985 static int do_execve(struct filename
*filename
,
1986 const char __user
*const __user
*__argv
,
1987 const char __user
*const __user
*__envp
)
1989 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1990 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1991 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
1994 static int do_execveat(int fd
, struct filename
*filename
,
1995 const char __user
*const __user
*__argv
,
1996 const char __user
*const __user
*__envp
,
1999 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2000 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2002 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2005 #ifdef CONFIG_COMPAT
2006 static int compat_do_execve(struct filename
*filename
,
2007 const compat_uptr_t __user
*__argv
,
2008 const compat_uptr_t __user
*__envp
)
2010 struct user_arg_ptr argv
= {
2012 .ptr
.compat
= __argv
,
2014 struct user_arg_ptr envp
= {
2016 .ptr
.compat
= __envp
,
2018 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2021 static int compat_do_execveat(int fd
, struct filename
*filename
,
2022 const compat_uptr_t __user
*__argv
,
2023 const compat_uptr_t __user
*__envp
,
2026 struct user_arg_ptr argv
= {
2028 .ptr
.compat
= __argv
,
2030 struct user_arg_ptr envp
= {
2032 .ptr
.compat
= __envp
,
2034 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2038 void set_binfmt(struct linux_binfmt
*new)
2040 struct mm_struct
*mm
= current
->mm
;
2043 module_put(mm
->binfmt
->module
);
2047 __module_get(new->module
);
2049 EXPORT_SYMBOL(set_binfmt
);
2052 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2054 void set_dumpable(struct mm_struct
*mm
, int value
)
2056 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
2059 set_mask_bits(&mm
->flags
, MMF_DUMPABLE_MASK
, value
);
2062 SYSCALL_DEFINE3(execve
,
2063 const char __user
*, filename
,
2064 const char __user
*const __user
*, argv
,
2065 const char __user
*const __user
*, envp
)
2067 return do_execve(getname(filename
), argv
, envp
);
2070 SYSCALL_DEFINE5(execveat
,
2071 int, fd
, const char __user
*, filename
,
2072 const char __user
*const __user
*, argv
,
2073 const char __user
*const __user
*, envp
,
2076 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
2078 return do_execveat(fd
,
2079 getname_flags(filename
, lookup_flags
, NULL
),
2083 #ifdef CONFIG_COMPAT
2084 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
2085 const compat_uptr_t __user
*, argv
,
2086 const compat_uptr_t __user
*, envp
)
2088 return compat_do_execve(getname(filename
), argv
, envp
);
2091 COMPAT_SYSCALL_DEFINE5(execveat
, int, fd
,
2092 const char __user
*, filename
,
2093 const compat_uptr_t __user
*, argv
,
2094 const compat_uptr_t __user
*, envp
,
2097 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
2099 return compat_do_execveat(fd
,
2100 getname_flags(filename
, lookup_flags
, NULL
),