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 (down_write_killable(&mm
->mmap_sem
)) {
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 up_write(&mm
->mmap_sem
);
278 bprm
->p
= vma
->vm_end
- sizeof(void *);
281 up_write(&mm
->mmap_sem
);
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 prepare_arg_pages(struct linux_binprm
*bprm
,
452 struct user_arg_ptr argv
, struct user_arg_ptr envp
)
454 unsigned long limit
, ptr_size
;
456 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
460 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
465 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
466 * (whichever is smaller) for the argv+env strings.
468 * - the remaining binfmt code will not run out of stack space,
469 * - the program will have a reasonable amount of stack left
472 limit
= _STK_LIM
/ 4 * 3;
473 limit
= min(limit
, bprm
->rlim_stack
.rlim_cur
/ 4);
475 * We've historically supported up to 32 pages (ARG_MAX)
476 * of argument strings even with small stacks
478 limit
= max_t(unsigned long, limit
, ARG_MAX
);
480 * We must account for the size of all the argv and envp pointers to
481 * the argv and envp strings, since they will also take up space in
482 * the stack. They aren't stored until much later when we can't
483 * signal to the parent that the child has run out of stack space.
484 * Instead, calculate it here so it's possible to fail gracefully.
486 ptr_size
= (bprm
->argc
+ bprm
->envc
) * sizeof(void *);
487 if (limit
<= ptr_size
)
491 bprm
->argmin
= bprm
->p
- limit
;
496 * 'copy_strings()' copies argument/environment strings from the old
497 * processes's memory to the new process's stack. The call to get_user_pages()
498 * ensures the destination page is created and not swapped out.
500 static int copy_strings(int argc
, struct user_arg_ptr argv
,
501 struct linux_binprm
*bprm
)
503 struct page
*kmapped_page
= NULL
;
505 unsigned long kpos
= 0;
509 const char __user
*str
;
514 str
= get_user_arg_ptr(argv
, argc
);
518 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
523 if (!valid_arg_len(bprm
, len
))
526 /* We're going to work our way backwords. */
531 if (bprm
->p
< bprm
->argmin
)
536 int offset
, bytes_to_copy
;
538 if (fatal_signal_pending(current
)) {
539 ret
= -ERESTARTNOHAND
;
544 offset
= pos
% PAGE_SIZE
;
548 bytes_to_copy
= offset
;
549 if (bytes_to_copy
> len
)
552 offset
-= bytes_to_copy
;
553 pos
-= bytes_to_copy
;
554 str
-= bytes_to_copy
;
555 len
-= bytes_to_copy
;
557 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
560 page
= get_arg_page(bprm
, pos
, 1);
567 flush_kernel_dcache_page(kmapped_page
);
568 kunmap(kmapped_page
);
569 put_arg_page(kmapped_page
);
572 kaddr
= kmap(kmapped_page
);
573 kpos
= pos
& PAGE_MASK
;
574 flush_arg_page(bprm
, kpos
, kmapped_page
);
576 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
585 flush_kernel_dcache_page(kmapped_page
);
586 kunmap(kmapped_page
);
587 put_arg_page(kmapped_page
);
593 * Copy and argument/environment string from the kernel to the processes stack.
595 int copy_string_kernel(const char *arg
, struct linux_binprm
*bprm
)
597 int len
= strnlen(arg
, MAX_ARG_STRLEN
) + 1 /* terminating NUL */;
598 unsigned long pos
= bprm
->p
;
602 if (!valid_arg_len(bprm
, len
))
605 /* We're going to work our way backwards. */
608 if (IS_ENABLED(CONFIG_MMU
) && bprm
->p
< bprm
->argmin
)
612 unsigned int bytes_to_copy
= min_t(unsigned int, len
,
613 min_not_zero(offset_in_page(pos
), PAGE_SIZE
));
617 pos
-= bytes_to_copy
;
618 arg
-= bytes_to_copy
;
619 len
-= bytes_to_copy
;
621 page
= get_arg_page(bprm
, pos
, 1);
624 kaddr
= kmap_atomic(page
);
625 flush_arg_page(bprm
, pos
& PAGE_MASK
, page
);
626 memcpy(kaddr
+ offset_in_page(pos
), arg
, bytes_to_copy
);
627 flush_kernel_dcache_page(page
);
628 kunmap_atomic(kaddr
);
634 EXPORT_SYMBOL(copy_string_kernel
);
639 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
640 * the binfmt code determines where the new stack should reside, we shift it to
641 * its final location. The process proceeds as follows:
643 * 1) Use shift to calculate the new vma endpoints.
644 * 2) Extend vma to cover both the old and new ranges. This ensures the
645 * arguments passed to subsequent functions are consistent.
646 * 3) Move vma's page tables to the new range.
647 * 4) Free up any cleared pgd range.
648 * 5) Shrink the vma to cover only the new range.
650 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
652 struct mm_struct
*mm
= vma
->vm_mm
;
653 unsigned long old_start
= vma
->vm_start
;
654 unsigned long old_end
= vma
->vm_end
;
655 unsigned long length
= old_end
- old_start
;
656 unsigned long new_start
= old_start
- shift
;
657 unsigned long new_end
= old_end
- shift
;
658 struct mmu_gather tlb
;
660 BUG_ON(new_start
> new_end
);
663 * ensure there are no vmas between where we want to go
666 if (vma
!= find_vma(mm
, new_start
))
670 * cover the whole range: [new_start, old_end)
672 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
676 * move the page tables downwards, on failure we rely on
677 * process cleanup to remove whatever mess we made.
679 if (length
!= move_page_tables(vma
, old_start
,
680 vma
, new_start
, length
, false))
684 tlb_gather_mmu(&tlb
, mm
, old_start
, old_end
);
685 if (new_end
> old_start
) {
687 * when the old and new regions overlap clear from new_end.
689 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
690 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
693 * otherwise, clean from old_start; this is done to not touch
694 * the address space in [new_end, old_start) some architectures
695 * have constraints on va-space that make this illegal (IA64) -
696 * for the others its just a little faster.
698 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
699 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
701 tlb_finish_mmu(&tlb
, old_start
, old_end
);
704 * Shrink the vma to just the new range. Always succeeds.
706 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
712 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
713 * the stack is optionally relocated, and some extra space is added.
715 int setup_arg_pages(struct linux_binprm
*bprm
,
716 unsigned long stack_top
,
717 int executable_stack
)
720 unsigned long stack_shift
;
721 struct mm_struct
*mm
= current
->mm
;
722 struct vm_area_struct
*vma
= bprm
->vma
;
723 struct vm_area_struct
*prev
= NULL
;
724 unsigned long vm_flags
;
725 unsigned long stack_base
;
726 unsigned long stack_size
;
727 unsigned long stack_expand
;
728 unsigned long rlim_stack
;
730 #ifdef CONFIG_STACK_GROWSUP
731 /* Limit stack size */
732 stack_base
= bprm
->rlim_stack
.rlim_max
;
733 if (stack_base
> STACK_SIZE_MAX
)
734 stack_base
= STACK_SIZE_MAX
;
736 /* Add space for stack randomization. */
737 stack_base
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
739 /* Make sure we didn't let the argument array grow too large. */
740 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
743 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
745 stack_shift
= vma
->vm_start
- stack_base
;
746 mm
->arg_start
= bprm
->p
- stack_shift
;
747 bprm
->p
= vma
->vm_end
- stack_shift
;
749 stack_top
= arch_align_stack(stack_top
);
750 stack_top
= PAGE_ALIGN(stack_top
);
752 if (unlikely(stack_top
< mmap_min_addr
) ||
753 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
756 stack_shift
= vma
->vm_end
- stack_top
;
758 bprm
->p
-= stack_shift
;
759 mm
->arg_start
= bprm
->p
;
763 bprm
->loader
-= stack_shift
;
764 bprm
->exec
-= stack_shift
;
766 if (down_write_killable(&mm
->mmap_sem
))
769 vm_flags
= VM_STACK_FLAGS
;
772 * Adjust stack execute permissions; explicitly enable for
773 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
774 * (arch default) otherwise.
776 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
778 else if (executable_stack
== EXSTACK_DISABLE_X
)
779 vm_flags
&= ~VM_EXEC
;
780 vm_flags
|= mm
->def_flags
;
781 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
783 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
789 if (unlikely(vm_flags
& VM_EXEC
)) {
790 pr_warn_once("process '%pD4' started with executable stack\n",
794 /* Move stack pages down in memory. */
796 ret
= shift_arg_pages(vma
, stack_shift
);
801 /* mprotect_fixup is overkill to remove the temporary stack flags */
802 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
804 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
805 stack_size
= vma
->vm_end
- vma
->vm_start
;
807 * Align this down to a page boundary as expand_stack
810 rlim_stack
= bprm
->rlim_stack
.rlim_cur
& PAGE_MASK
;
811 #ifdef CONFIG_STACK_GROWSUP
812 if (stack_size
+ stack_expand
> rlim_stack
)
813 stack_base
= vma
->vm_start
+ rlim_stack
;
815 stack_base
= vma
->vm_end
+ stack_expand
;
817 if (stack_size
+ stack_expand
> rlim_stack
)
818 stack_base
= vma
->vm_end
- rlim_stack
;
820 stack_base
= vma
->vm_start
- stack_expand
;
822 current
->mm
->start_stack
= bprm
->p
;
823 ret
= expand_stack(vma
, stack_base
);
828 up_write(&mm
->mmap_sem
);
831 EXPORT_SYMBOL(setup_arg_pages
);
836 * Transfer the program arguments and environment from the holding pages
837 * onto the stack. The provided stack pointer is adjusted accordingly.
839 int transfer_args_to_stack(struct linux_binprm
*bprm
,
840 unsigned long *sp_location
)
842 unsigned long index
, stop
, sp
;
845 stop
= bprm
->p
>> PAGE_SHIFT
;
848 for (index
= MAX_ARG_PAGES
- 1; index
>= stop
; index
--) {
849 unsigned int offset
= index
== stop
? bprm
->p
& ~PAGE_MASK
: 0;
850 char *src
= kmap(bprm
->page
[index
]) + offset
;
851 sp
-= PAGE_SIZE
- offset
;
852 if (copy_to_user((void *) sp
, src
, PAGE_SIZE
- offset
) != 0)
854 kunmap(bprm
->page
[index
]);
864 EXPORT_SYMBOL(transfer_args_to_stack
);
866 #endif /* CONFIG_MMU */
868 static struct file
*do_open_execat(int fd
, struct filename
*name
, int flags
)
872 struct open_flags open_exec_flags
= {
873 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
874 .acc_mode
= MAY_EXEC
,
875 .intent
= LOOKUP_OPEN
,
876 .lookup_flags
= LOOKUP_FOLLOW
,
879 if ((flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
880 return ERR_PTR(-EINVAL
);
881 if (flags
& AT_SYMLINK_NOFOLLOW
)
882 open_exec_flags
.lookup_flags
&= ~LOOKUP_FOLLOW
;
883 if (flags
& AT_EMPTY_PATH
)
884 open_exec_flags
.lookup_flags
|= LOOKUP_EMPTY
;
886 file
= do_filp_open(fd
, name
, &open_exec_flags
);
891 if (!S_ISREG(file_inode(file
)->i_mode
))
894 if (path_noexec(&file
->f_path
))
897 err
= deny_write_access(file
);
901 if (name
->name
[0] != '\0')
912 struct file
*open_exec(const char *name
)
914 struct filename
*filename
= getname_kernel(name
);
915 struct file
*f
= ERR_CAST(filename
);
917 if (!IS_ERR(filename
)) {
918 f
= do_open_execat(AT_FDCWD
, filename
, 0);
923 EXPORT_SYMBOL(open_exec
);
925 int kernel_read_file(struct file
*file
, void **buf
, loff_t
*size
,
926 loff_t max_size
, enum kernel_read_file_id id
)
932 if (!S_ISREG(file_inode(file
)->i_mode
) || max_size
< 0)
935 ret
= deny_write_access(file
);
939 ret
= security_kernel_read_file(file
, id
);
943 i_size
= i_size_read(file_inode(file
));
948 if (i_size
> SIZE_MAX
|| (max_size
> 0 && i_size
> max_size
)) {
953 if (id
!= READING_FIRMWARE_PREALLOC_BUFFER
)
954 *buf
= vmalloc(i_size
);
961 while (pos
< i_size
) {
962 bytes
= kernel_read(file
, *buf
+ pos
, i_size
- pos
, &pos
);
977 ret
= security_kernel_post_read_file(file
, *buf
, i_size
, id
);
983 if (id
!= READING_FIRMWARE_PREALLOC_BUFFER
) {
990 allow_write_access(file
);
993 EXPORT_SYMBOL_GPL(kernel_read_file
);
995 int kernel_read_file_from_path(const char *path
, void **buf
, loff_t
*size
,
996 loff_t max_size
, enum kernel_read_file_id id
)
1001 if (!path
|| !*path
)
1004 file
= filp_open(path
, O_RDONLY
, 0);
1006 return PTR_ERR(file
);
1008 ret
= kernel_read_file(file
, buf
, size
, max_size
, id
);
1012 EXPORT_SYMBOL_GPL(kernel_read_file_from_path
);
1014 int kernel_read_file_from_path_initns(const char *path
, void **buf
,
1015 loff_t
*size
, loff_t max_size
,
1016 enum kernel_read_file_id id
)
1022 if (!path
|| !*path
)
1025 task_lock(&init_task
);
1026 get_fs_root(init_task
.fs
, &root
);
1027 task_unlock(&init_task
);
1029 file
= file_open_root(root
.dentry
, root
.mnt
, path
, O_RDONLY
, 0);
1032 return PTR_ERR(file
);
1034 ret
= kernel_read_file(file
, buf
, size
, max_size
, id
);
1038 EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns
);
1040 int kernel_read_file_from_fd(int fd
, void **buf
, loff_t
*size
, loff_t max_size
,
1041 enum kernel_read_file_id id
)
1043 struct fd f
= fdget(fd
);
1049 ret
= kernel_read_file(f
.file
, buf
, size
, max_size
, id
);
1054 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd
);
1056 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
1058 ssize_t res
= vfs_read(file
, (void __user
*)addr
, len
, &pos
);
1060 flush_icache_range(addr
, addr
+ len
);
1063 EXPORT_SYMBOL(read_code
);
1066 * Maps the mm_struct mm into the current task struct.
1067 * On success, this function returns with the mutex
1068 * exec_update_mutex locked.
1070 static int exec_mmap(struct mm_struct
*mm
)
1072 struct task_struct
*tsk
;
1073 struct mm_struct
*old_mm
, *active_mm
;
1076 /* Notify parent that we're no longer interested in the old VM */
1078 old_mm
= current
->mm
;
1079 exec_mm_release(tsk
, old_mm
);
1081 sync_mm_rss(old_mm
);
1083 ret
= mutex_lock_killable(&tsk
->signal
->exec_update_mutex
);
1089 * Make sure that if there is a core dump in progress
1090 * for the old mm, we get out and die instead of going
1091 * through with the exec. We must hold mmap_sem around
1092 * checking core_state and changing tsk->mm.
1094 down_read(&old_mm
->mmap_sem
);
1095 if (unlikely(old_mm
->core_state
)) {
1096 up_read(&old_mm
->mmap_sem
);
1097 mutex_unlock(&tsk
->signal
->exec_update_mutex
);
1103 active_mm
= tsk
->active_mm
;
1104 membarrier_exec_mmap(mm
);
1106 tsk
->active_mm
= mm
;
1107 activate_mm(active_mm
, mm
);
1108 tsk
->mm
->vmacache_seqnum
= 0;
1109 vmacache_flush(tsk
);
1112 up_read(&old_mm
->mmap_sem
);
1113 BUG_ON(active_mm
!= old_mm
);
1114 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
1115 mm_update_next_owner(old_mm
);
1123 static int de_thread(struct task_struct
*tsk
)
1125 struct signal_struct
*sig
= tsk
->signal
;
1126 struct sighand_struct
*oldsighand
= tsk
->sighand
;
1127 spinlock_t
*lock
= &oldsighand
->siglock
;
1129 if (thread_group_empty(tsk
))
1130 goto no_thread_group
;
1133 * Kill all other threads in the thread group.
1135 spin_lock_irq(lock
);
1136 if (signal_group_exit(sig
)) {
1138 * Another group action in progress, just
1139 * return so that the signal is processed.
1141 spin_unlock_irq(lock
);
1145 sig
->group_exit_task
= tsk
;
1146 sig
->notify_count
= zap_other_threads(tsk
);
1147 if (!thread_group_leader(tsk
))
1148 sig
->notify_count
--;
1150 while (sig
->notify_count
) {
1151 __set_current_state(TASK_KILLABLE
);
1152 spin_unlock_irq(lock
);
1154 if (__fatal_signal_pending(tsk
))
1156 spin_lock_irq(lock
);
1158 spin_unlock_irq(lock
);
1161 * At this point all other threads have exited, all we have to
1162 * do is to wait for the thread group leader to become inactive,
1163 * and to assume its PID:
1165 if (!thread_group_leader(tsk
)) {
1166 struct task_struct
*leader
= tsk
->group_leader
;
1169 cgroup_threadgroup_change_begin(tsk
);
1170 write_lock_irq(&tasklist_lock
);
1172 * Do this under tasklist_lock to ensure that
1173 * exit_notify() can't miss ->group_exit_task
1175 sig
->notify_count
= -1;
1176 if (likely(leader
->exit_state
))
1178 __set_current_state(TASK_KILLABLE
);
1179 write_unlock_irq(&tasklist_lock
);
1180 cgroup_threadgroup_change_end(tsk
);
1182 if (__fatal_signal_pending(tsk
))
1187 * The only record we have of the real-time age of a
1188 * process, regardless of execs it's done, is start_time.
1189 * All the past CPU time is accumulated in signal_struct
1190 * from sister threads now dead. But in this non-leader
1191 * exec, nothing survives from the original leader thread,
1192 * whose birth marks the true age of this process now.
1193 * When we take on its identity by switching to its PID, we
1194 * also take its birthdate (always earlier than our own).
1196 tsk
->start_time
= leader
->start_time
;
1197 tsk
->start_boottime
= leader
->start_boottime
;
1199 BUG_ON(!same_thread_group(leader
, tsk
));
1201 * An exec() starts a new thread group with the
1202 * TGID of the previous thread group. Rehash the
1203 * two threads with a switched PID, and release
1204 * the former thread group leader:
1207 /* Become a process group leader with the old leader's pid.
1208 * The old leader becomes a thread of the this thread group.
1210 exchange_tids(tsk
, leader
);
1211 transfer_pid(leader
, tsk
, PIDTYPE_TGID
);
1212 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
1213 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
1215 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
1216 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
1218 tsk
->group_leader
= tsk
;
1219 leader
->group_leader
= tsk
;
1221 tsk
->exit_signal
= SIGCHLD
;
1222 leader
->exit_signal
= -1;
1224 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
1225 leader
->exit_state
= EXIT_DEAD
;
1228 * We are going to release_task()->ptrace_unlink() silently,
1229 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1230 * the tracer wont't block again waiting for this thread.
1232 if (unlikely(leader
->ptrace
))
1233 __wake_up_parent(leader
, leader
->parent
);
1234 write_unlock_irq(&tasklist_lock
);
1235 cgroup_threadgroup_change_end(tsk
);
1237 release_task(leader
);
1240 sig
->group_exit_task
= NULL
;
1241 sig
->notify_count
= 0;
1244 /* we have changed execution domain */
1245 tsk
->exit_signal
= SIGCHLD
;
1247 BUG_ON(!thread_group_leader(tsk
));
1251 /* protects against exit_notify() and __exit_signal() */
1252 read_lock(&tasklist_lock
);
1253 sig
->group_exit_task
= NULL
;
1254 sig
->notify_count
= 0;
1255 read_unlock(&tasklist_lock
);
1261 * This function makes sure the current process has its own signal table,
1262 * so that flush_signal_handlers can later reset the handlers without
1263 * disturbing other processes. (Other processes might share the signal
1264 * table via the CLONE_SIGHAND option to clone().)
1266 static int unshare_sighand(struct task_struct
*me
)
1268 struct sighand_struct
*oldsighand
= me
->sighand
;
1270 if (refcount_read(&oldsighand
->count
) != 1) {
1271 struct sighand_struct
*newsighand
;
1273 * This ->sighand is shared with the CLONE_SIGHAND
1274 * but not CLONE_THREAD task, switch to the new one.
1276 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1280 refcount_set(&newsighand
->count
, 1);
1281 memcpy(newsighand
->action
, oldsighand
->action
,
1282 sizeof(newsighand
->action
));
1284 write_lock_irq(&tasklist_lock
);
1285 spin_lock(&oldsighand
->siglock
);
1286 rcu_assign_pointer(me
->sighand
, newsighand
);
1287 spin_unlock(&oldsighand
->siglock
);
1288 write_unlock_irq(&tasklist_lock
);
1290 __cleanup_sighand(oldsighand
);
1295 char *__get_task_comm(char *buf
, size_t buf_size
, struct task_struct
*tsk
)
1298 strncpy(buf
, tsk
->comm
, buf_size
);
1302 EXPORT_SYMBOL_GPL(__get_task_comm
);
1305 * These functions flushes out all traces of the currently running executable
1306 * so that a new one can be started
1309 void __set_task_comm(struct task_struct
*tsk
, const char *buf
, bool exec
)
1312 trace_task_rename(tsk
, buf
);
1313 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1315 perf_event_comm(tsk
, exec
);
1319 * Calling this is the point of no return. None of the failures will be
1320 * seen by userspace since either the process is already taking a fatal
1321 * signal (via de_thread() or coredump), or will have SEGV raised
1322 * (after exec_mmap()) by search_binary_handler (see below).
1324 int begin_new_exec(struct linux_binprm
* bprm
)
1326 struct task_struct
*me
= current
;
1329 /* Once we are committed compute the creds */
1330 retval
= bprm_creds_from_file(bprm
);
1335 * Ensure all future errors are fatal.
1337 bprm
->point_of_no_return
= true;
1340 * Make this the only thread in the thread group.
1342 retval
= de_thread(me
);
1347 * Must be called _before_ exec_mmap() as bprm->mm is
1348 * not visibile until then. This also enables the update
1351 set_mm_exe_file(bprm
->mm
, bprm
->file
);
1353 /* If the binary is not readable then enforce mm->dumpable=0 */
1354 would_dump(bprm
, bprm
->file
);
1355 if (bprm
->have_execfd
)
1356 would_dump(bprm
, bprm
->executable
);
1359 * Release all of the old mmap stuff
1361 acct_arg_size(bprm
, 0);
1362 retval
= exec_mmap(bprm
->mm
);
1368 #ifdef CONFIG_POSIX_TIMERS
1369 exit_itimers(me
->signal
);
1370 flush_itimer_signals();
1374 * Make the signal table private.
1376 retval
= unshare_sighand(me
);
1381 me
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
| PF_KTHREAD
|
1382 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1384 me
->personality
&= ~bprm
->per_clear
;
1387 * We have to apply CLOEXEC before we change whether the process is
1388 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1389 * trying to access the should-be-closed file descriptors of a process
1390 * undergoing exec(2).
1392 do_close_on_exec(me
->files
);
1394 if (bprm
->secureexec
) {
1395 /* Make sure parent cannot signal privileged process. */
1396 me
->pdeath_signal
= 0;
1399 * For secureexec, reset the stack limit to sane default to
1400 * avoid bad behavior from the prior rlimits. This has to
1401 * happen before arch_pick_mmap_layout(), which examines
1402 * RLIMIT_STACK, but after the point of no return to avoid
1403 * needing to clean up the change on failure.
1405 if (bprm
->rlim_stack
.rlim_cur
> _STK_LIM
)
1406 bprm
->rlim_stack
.rlim_cur
= _STK_LIM
;
1409 me
->sas_ss_sp
= me
->sas_ss_size
= 0;
1412 * Figure out dumpability. Note that this checking only of current
1413 * is wrong, but userspace depends on it. This should be testing
1414 * bprm->secureexec instead.
1416 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
||
1417 !(uid_eq(current_euid(), current_uid()) &&
1418 gid_eq(current_egid(), current_gid())))
1419 set_dumpable(current
->mm
, suid_dumpable
);
1421 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1424 __set_task_comm(me
, kbasename(bprm
->filename
), true);
1426 /* An exec changes our domain. We are no longer part of the thread
1428 WRITE_ONCE(me
->self_exec_id
, me
->self_exec_id
+ 1);
1429 flush_signal_handlers(me
, 0);
1432 * install the new credentials for this executable
1434 security_bprm_committing_creds(bprm
);
1436 commit_creds(bprm
->cred
);
1440 * Disable monitoring for regular users
1441 * when executing setuid binaries. Must
1442 * wait until new credentials are committed
1443 * by commit_creds() above
1445 if (get_dumpable(me
->mm
) != SUID_DUMP_USER
)
1446 perf_event_exit_task(me
);
1448 * cred_guard_mutex must be held at least to this point to prevent
1449 * ptrace_attach() from altering our determination of the task's
1450 * credentials; any time after this it may be unlocked.
1452 security_bprm_committed_creds(bprm
);
1454 /* Pass the opened binary to the interpreter. */
1455 if (bprm
->have_execfd
) {
1456 retval
= get_unused_fd_flags(0);
1459 fd_install(retval
, bprm
->executable
);
1460 bprm
->executable
= NULL
;
1461 bprm
->execfd
= retval
;
1466 mutex_unlock(&me
->signal
->exec_update_mutex
);
1470 EXPORT_SYMBOL(begin_new_exec
);
1472 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1474 struct inode
*inode
= file_inode(file
);
1475 if (inode_permission(inode
, MAY_READ
) < 0) {
1476 struct user_namespace
*old
, *user_ns
;
1477 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1479 /* Ensure mm->user_ns contains the executable */
1480 user_ns
= old
= bprm
->mm
->user_ns
;
1481 while ((user_ns
!= &init_user_ns
) &&
1482 !privileged_wrt_inode_uidgid(user_ns
, inode
))
1483 user_ns
= user_ns
->parent
;
1485 if (old
!= user_ns
) {
1486 bprm
->mm
->user_ns
= get_user_ns(user_ns
);
1491 EXPORT_SYMBOL(would_dump
);
1493 void setup_new_exec(struct linux_binprm
* bprm
)
1495 /* Setup things that can depend upon the personality */
1496 struct task_struct
*me
= current
;
1498 arch_pick_mmap_layout(me
->mm
, &bprm
->rlim_stack
);
1500 arch_setup_new_exec();
1502 /* Set the new mm task size. We have to do that late because it may
1503 * depend on TIF_32BIT which is only updated in flush_thread() on
1504 * some architectures like powerpc
1506 me
->mm
->task_size
= TASK_SIZE
;
1507 mutex_unlock(&me
->signal
->exec_update_mutex
);
1508 mutex_unlock(&me
->signal
->cred_guard_mutex
);
1510 EXPORT_SYMBOL(setup_new_exec
);
1512 /* Runs immediately before start_thread() takes over. */
1513 void finalize_exec(struct linux_binprm
*bprm
)
1515 /* Store any stack rlimit changes before starting thread. */
1516 task_lock(current
->group_leader
);
1517 current
->signal
->rlim
[RLIMIT_STACK
] = bprm
->rlim_stack
;
1518 task_unlock(current
->group_leader
);
1520 EXPORT_SYMBOL(finalize_exec
);
1523 * Prepare credentials and lock ->cred_guard_mutex.
1524 * setup_new_exec() commits the new creds and drops the lock.
1525 * Or, if exec fails before, free_bprm() should release ->cred and
1528 static int prepare_bprm_creds(struct linux_binprm
*bprm
)
1530 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1531 return -ERESTARTNOINTR
;
1533 bprm
->cred
= prepare_exec_creds();
1534 if (likely(bprm
->cred
))
1537 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1541 static void free_bprm(struct linux_binprm
*bprm
)
1543 free_arg_pages(bprm
);
1545 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1546 abort_creds(bprm
->cred
);
1549 allow_write_access(bprm
->file
);
1552 if (bprm
->executable
)
1553 fput(bprm
->executable
);
1554 /* If a binfmt changed the interp, free it. */
1555 if (bprm
->interp
!= bprm
->filename
)
1556 kfree(bprm
->interp
);
1560 int bprm_change_interp(const char *interp
, struct linux_binprm
*bprm
)
1562 /* If a binfmt changed the interp, free it first. */
1563 if (bprm
->interp
!= bprm
->filename
)
1564 kfree(bprm
->interp
);
1565 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1570 EXPORT_SYMBOL(bprm_change_interp
);
1573 * determine how safe it is to execute the proposed program
1574 * - the caller must hold ->cred_guard_mutex to protect against
1575 * PTRACE_ATTACH or seccomp thread-sync
1577 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1579 struct task_struct
*p
= current
, *t
;
1583 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1586 * This isn't strictly necessary, but it makes it harder for LSMs to
1589 if (task_no_new_privs(current
))
1590 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1594 spin_lock(&p
->fs
->lock
);
1596 while_each_thread(p
, t
) {
1602 if (p
->fs
->users
> n_fs
)
1603 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1606 spin_unlock(&p
->fs
->lock
);
1609 static void bprm_fill_uid(struct linux_binprm
*bprm
, struct file
*file
)
1611 /* Handle suid and sgid on files */
1612 struct inode
*inode
;
1617 if (!mnt_may_suid(file
->f_path
.mnt
))
1620 if (task_no_new_privs(current
))
1623 inode
= file
->f_path
.dentry
->d_inode
;
1624 mode
= READ_ONCE(inode
->i_mode
);
1625 if (!(mode
& (S_ISUID
|S_ISGID
)))
1628 /* Be careful if suid/sgid is set */
1631 /* reload atomically mode/uid/gid now that lock held */
1632 mode
= inode
->i_mode
;
1635 inode_unlock(inode
);
1637 /* We ignore suid/sgid if there are no mappings for them in the ns */
1638 if (!kuid_has_mapping(bprm
->cred
->user_ns
, uid
) ||
1639 !kgid_has_mapping(bprm
->cred
->user_ns
, gid
))
1642 if (mode
& S_ISUID
) {
1643 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1644 bprm
->cred
->euid
= uid
;
1647 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1648 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1649 bprm
->cred
->egid
= gid
;
1654 * Compute brpm->cred based upon the final binary.
1656 static int bprm_creds_from_file(struct linux_binprm
*bprm
)
1658 /* Compute creds based on which file? */
1659 struct file
*file
= bprm
->execfd_creds
? bprm
->executable
: bprm
->file
;
1661 bprm_fill_uid(bprm
, file
);
1662 return security_bprm_creds_from_file(bprm
, file
);
1666 * Fill the binprm structure from the inode.
1667 * Read the first BINPRM_BUF_SIZE bytes
1669 * This may be called multiple times for binary chains (scripts for example).
1671 static int prepare_binprm(struct linux_binprm
*bprm
)
1675 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1676 return kernel_read(bprm
->file
, bprm
->buf
, BINPRM_BUF_SIZE
, &pos
);
1680 * Arguments are '\0' separated strings found at the location bprm->p
1681 * points to; chop off the first by relocating brpm->p to right after
1682 * the first '\0' encountered.
1684 int remove_arg_zero(struct linux_binprm
*bprm
)
1687 unsigned long offset
;
1695 offset
= bprm
->p
& ~PAGE_MASK
;
1696 page
= get_arg_page(bprm
, bprm
->p
, 0);
1701 kaddr
= kmap_atomic(page
);
1703 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1704 offset
++, bprm
->p
++)
1707 kunmap_atomic(kaddr
);
1709 } while (offset
== PAGE_SIZE
);
1718 EXPORT_SYMBOL(remove_arg_zero
);
1720 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1722 * cycle the list of binary formats handler, until one recognizes the image
1724 static int search_binary_handler(struct linux_binprm
*bprm
)
1726 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1727 struct linux_binfmt
*fmt
;
1730 retval
= prepare_binprm(bprm
);
1734 retval
= security_bprm_check(bprm
);
1740 read_lock(&binfmt_lock
);
1741 list_for_each_entry(fmt
, &formats
, lh
) {
1742 if (!try_module_get(fmt
->module
))
1744 read_unlock(&binfmt_lock
);
1746 retval
= fmt
->load_binary(bprm
);
1748 read_lock(&binfmt_lock
);
1750 if (bprm
->point_of_no_return
|| (retval
!= -ENOEXEC
)) {
1751 read_unlock(&binfmt_lock
);
1755 read_unlock(&binfmt_lock
);
1758 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1759 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1761 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1770 static int exec_binprm(struct linux_binprm
*bprm
)
1772 pid_t old_pid
, old_vpid
;
1775 /* Need to fetch pid before load_binary changes it */
1776 old_pid
= current
->pid
;
1778 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1781 /* This allows 4 levels of binfmt rewrites before failing hard. */
1782 for (depth
= 0;; depth
++) {
1787 ret
= search_binary_handler(bprm
);
1790 if (!bprm
->interpreter
)
1794 bprm
->file
= bprm
->interpreter
;
1795 bprm
->interpreter
= NULL
;
1797 allow_write_access(exec
);
1798 if (unlikely(bprm
->have_execfd
)) {
1799 if (bprm
->executable
) {
1803 bprm
->executable
= exec
;
1809 trace_sched_process_exec(current
, old_pid
, bprm
);
1810 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1811 proc_exec_connector(current
);
1816 * sys_execve() executes a new program.
1818 static int __do_execve_file(int fd
, struct filename
*filename
,
1819 struct user_arg_ptr argv
,
1820 struct user_arg_ptr envp
,
1821 int flags
, struct file
*file
)
1823 char *pathbuf
= NULL
;
1824 struct linux_binprm
*bprm
;
1825 struct files_struct
*displaced
;
1828 if (IS_ERR(filename
))
1829 return PTR_ERR(filename
);
1832 * We move the actual failure in case of RLIMIT_NPROC excess from
1833 * set*uid() to execve() because too many poorly written programs
1834 * don't check setuid() return code. Here we additionally recheck
1835 * whether NPROC limit is still exceeded.
1837 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1838 atomic_read(¤t_user()->processes
) > rlimit(RLIMIT_NPROC
)) {
1843 /* We're below the limit (still or again), so we don't want to make
1844 * further execve() calls fail. */
1845 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1847 retval
= unshare_files(&displaced
);
1852 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1856 retval
= prepare_bprm_creds(bprm
);
1860 check_unsafe_exec(bprm
);
1861 current
->in_execve
= 1;
1864 file
= do_open_execat(fd
, filename
, flags
);
1865 retval
= PTR_ERR(file
);
1873 bprm
->filename
= "none";
1874 } else if (fd
== AT_FDCWD
|| filename
->name
[0] == '/') {
1875 bprm
->filename
= filename
->name
;
1877 if (filename
->name
[0] == '\0')
1878 pathbuf
= kasprintf(GFP_KERNEL
, "/dev/fd/%d", fd
);
1880 pathbuf
= kasprintf(GFP_KERNEL
, "/dev/fd/%d/%s",
1881 fd
, filename
->name
);
1887 * Record that a name derived from an O_CLOEXEC fd will be
1888 * inaccessible after exec. Relies on having exclusive access to
1889 * current->files (due to unshare_files above).
1891 if (close_on_exec(fd
, rcu_dereference_raw(current
->files
->fdt
)))
1892 bprm
->interp_flags
|= BINPRM_FLAGS_PATH_INACCESSIBLE
;
1893 bprm
->filename
= pathbuf
;
1895 bprm
->interp
= bprm
->filename
;
1897 retval
= bprm_mm_init(bprm
);
1901 retval
= prepare_arg_pages(bprm
, argv
, envp
);
1905 /* Set the unchanging part of bprm->cred */
1906 retval
= security_bprm_creds_for_exec(bprm
);
1910 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1914 bprm
->exec
= bprm
->p
;
1915 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1919 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1923 retval
= exec_binprm(bprm
);
1927 /* execve succeeded */
1928 current
->fs
->in_exec
= 0;
1929 current
->in_execve
= 0;
1930 rseq_execve(current
);
1931 acct_update_integrals(current
);
1932 task_numa_free(current
, false);
1938 put_files_struct(displaced
);
1943 * If past the point of no return ensure the the code never
1944 * returns to the userspace process. Use an existing fatal
1945 * signal if present otherwise terminate the process with
1948 if (bprm
->point_of_no_return
&& !fatal_signal_pending(current
))
1949 force_sigsegv(SIGSEGV
);
1951 acct_arg_size(bprm
, 0);
1956 current
->fs
->in_exec
= 0;
1957 current
->in_execve
= 0;
1965 reset_files_struct(displaced
);
1972 static int do_execveat_common(int fd
, struct filename
*filename
,
1973 struct user_arg_ptr argv
,
1974 struct user_arg_ptr envp
,
1977 return __do_execve_file(fd
, filename
, argv
, envp
, flags
, NULL
);
1980 int do_execve_file(struct file
*file
, void *__argv
, void *__envp
)
1982 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1983 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1985 return __do_execve_file(AT_FDCWD
, NULL
, argv
, envp
, 0, file
);
1988 int do_execve(struct filename
*filename
,
1989 const char __user
*const __user
*__argv
,
1990 const char __user
*const __user
*__envp
)
1992 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1993 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1994 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
1997 int do_execveat(int fd
, struct filename
*filename
,
1998 const char __user
*const __user
*__argv
,
1999 const char __user
*const __user
*__envp
,
2002 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2003 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2005 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2008 #ifdef CONFIG_COMPAT
2009 static int compat_do_execve(struct filename
*filename
,
2010 const compat_uptr_t __user
*__argv
,
2011 const compat_uptr_t __user
*__envp
)
2013 struct user_arg_ptr argv
= {
2015 .ptr
.compat
= __argv
,
2017 struct user_arg_ptr envp
= {
2019 .ptr
.compat
= __envp
,
2021 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2024 static int compat_do_execveat(int fd
, struct filename
*filename
,
2025 const compat_uptr_t __user
*__argv
,
2026 const compat_uptr_t __user
*__envp
,
2029 struct user_arg_ptr argv
= {
2031 .ptr
.compat
= __argv
,
2033 struct user_arg_ptr envp
= {
2035 .ptr
.compat
= __envp
,
2037 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2041 void set_binfmt(struct linux_binfmt
*new)
2043 struct mm_struct
*mm
= current
->mm
;
2046 module_put(mm
->binfmt
->module
);
2050 __module_get(new->module
);
2052 EXPORT_SYMBOL(set_binfmt
);
2055 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2057 void set_dumpable(struct mm_struct
*mm
, int value
)
2059 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
2062 set_mask_bits(&mm
->flags
, MMF_DUMPABLE_MASK
, value
);
2065 SYSCALL_DEFINE3(execve
,
2066 const char __user
*, filename
,
2067 const char __user
*const __user
*, argv
,
2068 const char __user
*const __user
*, envp
)
2070 return do_execve(getname(filename
), argv
, envp
);
2073 SYSCALL_DEFINE5(execveat
,
2074 int, fd
, const char __user
*, filename
,
2075 const char __user
*const __user
*, argv
,
2076 const char __user
*const __user
*, envp
,
2079 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
2081 return do_execveat(fd
,
2082 getname_flags(filename
, lookup_flags
, NULL
),
2086 #ifdef CONFIG_COMPAT
2087 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
2088 const compat_uptr_t __user
*, argv
,
2089 const compat_uptr_t __user
*, envp
)
2091 return compat_do_execve(getname(filename
), argv
, envp
);
2094 COMPAT_SYSCALL_DEFINE5(execveat
, int, fd
,
2095 const char __user
*, filename
,
2096 const compat_uptr_t __user
*, argv
,
2097 const compat_uptr_t __user
*, envp
,
2100 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
2102 return compat_do_execveat(fd
,
2103 getname_flags(filename
, lookup_flags
, NULL
),