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/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>
65 #include <linux/io_uring.h>
66 #include <linux/syscall_user_dispatch.h>
67 #include <linux/coredump.h>
69 #include <linux/uaccess.h>
70 #include <asm/mmu_context.h>
73 #include <trace/events/task.h>
76 #include <trace/events/sched.h>
78 static int bprm_creds_from_file(struct linux_binprm
*bprm
);
80 int suid_dumpable
= 0;
82 static LIST_HEAD(formats
);
83 static DEFINE_RWLOCK(binfmt_lock
);
85 void __register_binfmt(struct linux_binfmt
* fmt
, int insert
)
87 write_lock(&binfmt_lock
);
88 insert
? list_add(&fmt
->lh
, &formats
) :
89 list_add_tail(&fmt
->lh
, &formats
);
90 write_unlock(&binfmt_lock
);
93 EXPORT_SYMBOL(__register_binfmt
);
95 void unregister_binfmt(struct linux_binfmt
* fmt
)
97 write_lock(&binfmt_lock
);
99 write_unlock(&binfmt_lock
);
102 EXPORT_SYMBOL(unregister_binfmt
);
104 static inline void put_binfmt(struct linux_binfmt
* fmt
)
106 module_put(fmt
->module
);
109 bool path_noexec(const struct path
*path
)
111 return (path
->mnt
->mnt_flags
& MNT_NOEXEC
) ||
112 (path
->mnt
->mnt_sb
->s_iflags
& SB_I_NOEXEC
);
117 * Note that a shared library must be both readable and executable due to
120 * Also note that we take the address to load from the file itself.
122 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
124 struct linux_binfmt
*fmt
;
126 struct filename
*tmp
= getname(library
);
127 int error
= PTR_ERR(tmp
);
128 static const struct open_flags uselib_flags
= {
129 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
130 .acc_mode
= MAY_READ
| MAY_EXEC
,
131 .intent
= LOOKUP_OPEN
,
132 .lookup_flags
= LOOKUP_FOLLOW
,
138 file
= do_filp_open(AT_FDCWD
, tmp
, &uselib_flags
);
140 error
= PTR_ERR(file
);
145 * may_open() has already checked for this, so it should be
146 * impossible to trip now. But we need to be extra cautious
147 * and check again at the very end too.
150 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
151 path_noexec(&file
->f_path
)))
158 read_lock(&binfmt_lock
);
159 list_for_each_entry(fmt
, &formats
, lh
) {
160 if (!fmt
->load_shlib
)
162 if (!try_module_get(fmt
->module
))
164 read_unlock(&binfmt_lock
);
165 error
= fmt
->load_shlib(file
);
166 read_lock(&binfmt_lock
);
168 if (error
!= -ENOEXEC
)
171 read_unlock(&binfmt_lock
);
177 #endif /* #ifdef CONFIG_USELIB */
181 * The nascent bprm->mm is not visible until exec_mmap() but it can
182 * use a lot of memory, account these pages in current->mm temporary
183 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
184 * change the counter back via acct_arg_size(0).
186 static void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
188 struct mm_struct
*mm
= current
->mm
;
189 long diff
= (long)(pages
- bprm
->vma_pages
);
194 bprm
->vma_pages
= pages
;
195 add_mm_counter(mm
, MM_ANONPAGES
, diff
);
198 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
203 unsigned int gup_flags
= FOLL_FORCE
;
205 #ifdef CONFIG_STACK_GROWSUP
207 ret
= expand_downwards(bprm
->vma
, pos
);
214 gup_flags
|= FOLL_WRITE
;
217 * We are doing an exec(). 'current' is the process
218 * doing the exec and bprm->mm is the new process's mm.
220 mmap_read_lock(bprm
->mm
);
221 ret
= get_user_pages_remote(bprm
->mm
, pos
, 1, gup_flags
,
223 mmap_read_unlock(bprm
->mm
);
228 acct_arg_size(bprm
, vma_pages(bprm
->vma
));
233 static void put_arg_page(struct page
*page
)
238 static void free_arg_pages(struct linux_binprm
*bprm
)
242 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
245 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
248 static int __bprm_mm_init(struct linux_binprm
*bprm
)
251 struct vm_area_struct
*vma
= NULL
;
252 struct mm_struct
*mm
= bprm
->mm
;
254 bprm
->vma
= vma
= vm_area_alloc(mm
);
257 vma_set_anonymous(vma
);
259 if (mmap_write_lock_killable(mm
)) {
265 * Place the stack at the largest stack address the architecture
266 * supports. Later, we'll move this to an appropriate place. We don't
267 * use STACK_TOP because that can depend on attributes which aren't
270 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
271 vma
->vm_end
= STACK_TOP_MAX
;
272 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
273 vma
->vm_flags
= VM_SOFTDIRTY
| VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
;
274 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
276 err
= insert_vm_struct(mm
, vma
);
280 mm
->stack_vm
= mm
->total_vm
= 1;
281 mmap_write_unlock(mm
);
282 bprm
->p
= vma
->vm_end
- sizeof(void *);
285 mmap_write_unlock(mm
);
292 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
294 return len
<= MAX_ARG_STRLEN
;
299 static inline void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
303 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
308 page
= bprm
->page
[pos
/ PAGE_SIZE
];
309 if (!page
&& write
) {
310 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
313 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
319 static void put_arg_page(struct page
*page
)
323 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
326 __free_page(bprm
->page
[i
]);
327 bprm
->page
[i
] = NULL
;
331 static void free_arg_pages(struct linux_binprm
*bprm
)
335 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
336 free_arg_page(bprm
, i
);
339 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
344 static int __bprm_mm_init(struct linux_binprm
*bprm
)
346 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
350 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
352 return len
<= bprm
->p
;
355 #endif /* CONFIG_MMU */
358 * Create a new mm_struct and populate it with a temporary stack
359 * vm_area_struct. We don't have enough context at this point to set the stack
360 * flags, permissions, and offset, so we use temporary values. We'll update
361 * them later in setup_arg_pages().
363 static int bprm_mm_init(struct linux_binprm
*bprm
)
366 struct mm_struct
*mm
= NULL
;
368 bprm
->mm
= mm
= mm_alloc();
373 /* Save current stack limit for all calculations made during exec. */
374 task_lock(current
->group_leader
);
375 bprm
->rlim_stack
= current
->signal
->rlim
[RLIMIT_STACK
];
376 task_unlock(current
->group_leader
);
378 err
= __bprm_mm_init(bprm
);
393 struct user_arg_ptr
{
398 const char __user
*const __user
*native
;
400 const compat_uptr_t __user
*compat
;
405 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
407 const char __user
*native
;
410 if (unlikely(argv
.is_compat
)) {
411 compat_uptr_t compat
;
413 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
414 return ERR_PTR(-EFAULT
);
416 return compat_ptr(compat
);
420 if (get_user(native
, argv
.ptr
.native
+ nr
))
421 return ERR_PTR(-EFAULT
);
427 * count() counts the number of strings in array ARGV.
429 static int count(struct user_arg_ptr argv
, int max
)
433 if (argv
.ptr
.native
!= NULL
) {
435 const char __user
*p
= get_user_arg_ptr(argv
, i
);
447 if (fatal_signal_pending(current
))
448 return -ERESTARTNOHAND
;
455 static int count_strings_kernel(const char *const *argv
)
462 for (i
= 0; argv
[i
]; ++i
) {
463 if (i
>= MAX_ARG_STRINGS
)
465 if (fatal_signal_pending(current
))
466 return -ERESTARTNOHAND
;
472 static int bprm_stack_limits(struct linux_binprm
*bprm
)
474 unsigned long limit
, ptr_size
;
477 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
478 * (whichever is smaller) for the argv+env strings.
480 * - the remaining binfmt code will not run out of stack space,
481 * - the program will have a reasonable amount of stack left
484 limit
= _STK_LIM
/ 4 * 3;
485 limit
= min(limit
, bprm
->rlim_stack
.rlim_cur
/ 4);
487 * We've historically supported up to 32 pages (ARG_MAX)
488 * of argument strings even with small stacks
490 limit
= max_t(unsigned long, limit
, ARG_MAX
);
492 * We must account for the size of all the argv and envp pointers to
493 * the argv and envp strings, since they will also take up space in
494 * the stack. They aren't stored until much later when we can't
495 * signal to the parent that the child has run out of stack space.
496 * Instead, calculate it here so it's possible to fail gracefully.
498 * In the case of argc = 0, make sure there is space for adding a
499 * empty string (which will bump argc to 1), to ensure confused
500 * userspace programs don't start processing from argv[1], thinking
501 * argc can never be 0, to keep them from walking envp by accident.
502 * See do_execveat_common().
504 ptr_size
= (max(bprm
->argc
, 1) + bprm
->envc
) * sizeof(void *);
505 if (limit
<= ptr_size
)
509 bprm
->argmin
= bprm
->p
- limit
;
514 * 'copy_strings()' copies argument/environment strings from the old
515 * processes's memory to the new process's stack. The call to get_user_pages()
516 * ensures the destination page is created and not swapped out.
518 static int copy_strings(int argc
, struct user_arg_ptr argv
,
519 struct linux_binprm
*bprm
)
521 struct page
*kmapped_page
= NULL
;
523 unsigned long kpos
= 0;
527 const char __user
*str
;
532 str
= get_user_arg_ptr(argv
, argc
);
536 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
541 if (!valid_arg_len(bprm
, len
))
544 /* We're going to work our way backwards. */
549 if (bprm
->p
< bprm
->argmin
)
554 int offset
, bytes_to_copy
;
556 if (fatal_signal_pending(current
)) {
557 ret
= -ERESTARTNOHAND
;
562 offset
= pos
% PAGE_SIZE
;
566 bytes_to_copy
= offset
;
567 if (bytes_to_copy
> len
)
570 offset
-= bytes_to_copy
;
571 pos
-= bytes_to_copy
;
572 str
-= bytes_to_copy
;
573 len
-= bytes_to_copy
;
575 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
578 page
= get_arg_page(bprm
, pos
, 1);
585 flush_dcache_page(kmapped_page
);
586 kunmap(kmapped_page
);
587 put_arg_page(kmapped_page
);
590 kaddr
= kmap(kmapped_page
);
591 kpos
= pos
& PAGE_MASK
;
592 flush_arg_page(bprm
, kpos
, kmapped_page
);
594 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
603 flush_dcache_page(kmapped_page
);
604 kunmap(kmapped_page
);
605 put_arg_page(kmapped_page
);
611 * Copy and argument/environment string from the kernel to the processes stack.
613 int copy_string_kernel(const char *arg
, struct linux_binprm
*bprm
)
615 int len
= strnlen(arg
, MAX_ARG_STRLEN
) + 1 /* terminating NUL */;
616 unsigned long pos
= bprm
->p
;
620 if (!valid_arg_len(bprm
, len
))
623 /* We're going to work our way backwards. */
626 if (IS_ENABLED(CONFIG_MMU
) && bprm
->p
< bprm
->argmin
)
630 unsigned int bytes_to_copy
= min_t(unsigned int, len
,
631 min_not_zero(offset_in_page(pos
), PAGE_SIZE
));
635 pos
-= bytes_to_copy
;
636 arg
-= bytes_to_copy
;
637 len
-= bytes_to_copy
;
639 page
= get_arg_page(bprm
, pos
, 1);
642 kaddr
= kmap_atomic(page
);
643 flush_arg_page(bprm
, pos
& PAGE_MASK
, page
);
644 memcpy(kaddr
+ offset_in_page(pos
), arg
, bytes_to_copy
);
645 flush_dcache_page(page
);
646 kunmap_atomic(kaddr
);
652 EXPORT_SYMBOL(copy_string_kernel
);
654 static int copy_strings_kernel(int argc
, const char *const *argv
,
655 struct linux_binprm
*bprm
)
658 int ret
= copy_string_kernel(argv
[argc
], bprm
);
661 if (fatal_signal_pending(current
))
662 return -ERESTARTNOHAND
;
671 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
672 * the binfmt code determines where the new stack should reside, we shift it to
673 * its final location. The process proceeds as follows:
675 * 1) Use shift to calculate the new vma endpoints.
676 * 2) Extend vma to cover both the old and new ranges. This ensures the
677 * arguments passed to subsequent functions are consistent.
678 * 3) Move vma's page tables to the new range.
679 * 4) Free up any cleared pgd range.
680 * 5) Shrink the vma to cover only the new range.
682 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
684 struct mm_struct
*mm
= vma
->vm_mm
;
685 unsigned long old_start
= vma
->vm_start
;
686 unsigned long old_end
= vma
->vm_end
;
687 unsigned long length
= old_end
- old_start
;
688 unsigned long new_start
= old_start
- shift
;
689 unsigned long new_end
= old_end
- shift
;
690 struct mmu_gather tlb
;
692 BUG_ON(new_start
> new_end
);
695 * ensure there are no vmas between where we want to go
698 if (vma
!= find_vma(mm
, new_start
))
702 * cover the whole range: [new_start, old_end)
704 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
708 * move the page tables downwards, on failure we rely on
709 * process cleanup to remove whatever mess we made.
711 if (length
!= move_page_tables(vma
, old_start
,
712 vma
, new_start
, length
, false))
716 tlb_gather_mmu(&tlb
, mm
);
717 if (new_end
> old_start
) {
719 * when the old and new regions overlap clear from new_end.
721 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
722 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
725 * otherwise, clean from old_start; this is done to not touch
726 * the address space in [new_end, old_start) some architectures
727 * have constraints on va-space that make this illegal (IA64) -
728 * for the others its just a little faster.
730 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
731 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
733 tlb_finish_mmu(&tlb
);
736 * Shrink the vma to just the new range. Always succeeds.
738 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
744 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
745 * the stack is optionally relocated, and some extra space is added.
747 int setup_arg_pages(struct linux_binprm
*bprm
,
748 unsigned long stack_top
,
749 int executable_stack
)
752 unsigned long stack_shift
;
753 struct mm_struct
*mm
= current
->mm
;
754 struct vm_area_struct
*vma
= bprm
->vma
;
755 struct vm_area_struct
*prev
= NULL
;
756 unsigned long vm_flags
;
757 unsigned long stack_base
;
758 unsigned long stack_size
;
759 unsigned long stack_expand
;
760 unsigned long rlim_stack
;
761 struct mmu_gather tlb
;
763 #ifdef CONFIG_STACK_GROWSUP
764 /* Limit stack size */
765 stack_base
= bprm
->rlim_stack
.rlim_max
;
767 stack_base
= calc_max_stack_size(stack_base
);
769 /* Add space for stack randomization. */
770 stack_base
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
772 /* Make sure we didn't let the argument array grow too large. */
773 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
776 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
778 stack_shift
= vma
->vm_start
- stack_base
;
779 mm
->arg_start
= bprm
->p
- stack_shift
;
780 bprm
->p
= vma
->vm_end
- stack_shift
;
782 stack_top
= arch_align_stack(stack_top
);
783 stack_top
= PAGE_ALIGN(stack_top
);
785 if (unlikely(stack_top
< mmap_min_addr
) ||
786 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
789 stack_shift
= vma
->vm_end
- stack_top
;
791 bprm
->p
-= stack_shift
;
792 mm
->arg_start
= bprm
->p
;
796 bprm
->loader
-= stack_shift
;
797 bprm
->exec
-= stack_shift
;
799 if (mmap_write_lock_killable(mm
))
802 vm_flags
= VM_STACK_FLAGS
;
805 * Adjust stack execute permissions; explicitly enable for
806 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
807 * (arch default) otherwise.
809 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
811 else if (executable_stack
== EXSTACK_DISABLE_X
)
812 vm_flags
&= ~VM_EXEC
;
813 vm_flags
|= mm
->def_flags
;
814 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
816 tlb_gather_mmu(&tlb
, mm
);
817 ret
= mprotect_fixup(&tlb
, vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
819 tlb_finish_mmu(&tlb
);
825 if (unlikely(vm_flags
& VM_EXEC
)) {
826 pr_warn_once("process '%pD4' started with executable stack\n",
830 /* Move stack pages down in memory. */
832 ret
= shift_arg_pages(vma
, stack_shift
);
837 /* mprotect_fixup is overkill to remove the temporary stack flags */
838 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
840 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
841 stack_size
= vma
->vm_end
- vma
->vm_start
;
843 * Align this down to a page boundary as expand_stack
846 rlim_stack
= bprm
->rlim_stack
.rlim_cur
& PAGE_MASK
;
847 #ifdef CONFIG_STACK_GROWSUP
848 if (stack_size
+ stack_expand
> rlim_stack
)
849 stack_base
= vma
->vm_start
+ rlim_stack
;
851 stack_base
= vma
->vm_end
+ stack_expand
;
853 if (stack_size
+ stack_expand
> rlim_stack
)
854 stack_base
= vma
->vm_end
- rlim_stack
;
856 stack_base
= vma
->vm_start
- stack_expand
;
858 current
->mm
->start_stack
= bprm
->p
;
859 ret
= expand_stack(vma
, stack_base
);
864 mmap_write_unlock(mm
);
867 EXPORT_SYMBOL(setup_arg_pages
);
872 * Transfer the program arguments and environment from the holding pages
873 * onto the stack. The provided stack pointer is adjusted accordingly.
875 int transfer_args_to_stack(struct linux_binprm
*bprm
,
876 unsigned long *sp_location
)
878 unsigned long index
, stop
, sp
;
881 stop
= bprm
->p
>> PAGE_SHIFT
;
884 for (index
= MAX_ARG_PAGES
- 1; index
>= stop
; index
--) {
885 unsigned int offset
= index
== stop
? bprm
->p
& ~PAGE_MASK
: 0;
886 char *src
= kmap(bprm
->page
[index
]) + offset
;
887 sp
-= PAGE_SIZE
- offset
;
888 if (copy_to_user((void *) sp
, src
, PAGE_SIZE
- offset
) != 0)
890 kunmap(bprm
->page
[index
]);
900 EXPORT_SYMBOL(transfer_args_to_stack
);
902 #endif /* CONFIG_MMU */
904 static struct file
*do_open_execat(int fd
, struct filename
*name
, int flags
)
908 struct open_flags open_exec_flags
= {
909 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
910 .acc_mode
= MAY_EXEC
,
911 .intent
= LOOKUP_OPEN
,
912 .lookup_flags
= LOOKUP_FOLLOW
,
915 if ((flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
916 return ERR_PTR(-EINVAL
);
917 if (flags
& AT_SYMLINK_NOFOLLOW
)
918 open_exec_flags
.lookup_flags
&= ~LOOKUP_FOLLOW
;
919 if (flags
& AT_EMPTY_PATH
)
920 open_exec_flags
.lookup_flags
|= LOOKUP_EMPTY
;
922 file
= do_filp_open(fd
, name
, &open_exec_flags
);
927 * may_open() has already checked for this, so it should be
928 * impossible to trip now. But we need to be extra cautious
929 * and check again at the very end too.
932 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
933 path_noexec(&file
->f_path
)))
936 err
= deny_write_access(file
);
940 if (name
->name
[0] != '\0')
951 struct file
*open_exec(const char *name
)
953 struct filename
*filename
= getname_kernel(name
);
954 struct file
*f
= ERR_CAST(filename
);
956 if (!IS_ERR(filename
)) {
957 f
= do_open_execat(AT_FDCWD
, filename
, 0);
962 EXPORT_SYMBOL(open_exec
);
964 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
965 defined(CONFIG_BINFMT_ELF_FDPIC)
966 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
968 ssize_t res
= vfs_read(file
, (void __user
*)addr
, len
, &pos
);
970 flush_icache_user_range(addr
, addr
+ len
);
973 EXPORT_SYMBOL(read_code
);
977 * Maps the mm_struct mm into the current task struct.
978 * On success, this function returns with exec_update_lock
981 static int exec_mmap(struct mm_struct
*mm
)
983 struct task_struct
*tsk
;
984 struct mm_struct
*old_mm
, *active_mm
;
987 /* Notify parent that we're no longer interested in the old VM */
989 old_mm
= current
->mm
;
990 exec_mm_release(tsk
, old_mm
);
994 ret
= down_write_killable(&tsk
->signal
->exec_update_lock
);
1000 * If there is a pending fatal signal perhaps a signal
1001 * whose default action is to create a coredump get
1002 * out and die instead of going through with the exec.
1004 ret
= mmap_read_lock_killable(old_mm
);
1006 up_write(&tsk
->signal
->exec_update_lock
);
1012 membarrier_exec_mmap(mm
);
1014 local_irq_disable();
1015 active_mm
= tsk
->active_mm
;
1016 tsk
->active_mm
= mm
;
1019 * This prevents preemption while active_mm is being loaded and
1020 * it and mm are being updated, which could cause problems for
1021 * lazy tlb mm refcounting when these are updated by context
1022 * switches. Not all architectures can handle irqs off over
1025 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1027 activate_mm(active_mm
, mm
);
1028 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1030 tsk
->mm
->vmacache_seqnum
= 0;
1031 vmacache_flush(tsk
);
1034 mmap_read_unlock(old_mm
);
1035 BUG_ON(active_mm
!= old_mm
);
1036 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
1037 mm_update_next_owner(old_mm
);
1045 static int de_thread(struct task_struct
*tsk
)
1047 struct signal_struct
*sig
= tsk
->signal
;
1048 struct sighand_struct
*oldsighand
= tsk
->sighand
;
1049 spinlock_t
*lock
= &oldsighand
->siglock
;
1051 if (thread_group_empty(tsk
))
1052 goto no_thread_group
;
1055 * Kill all other threads in the thread group.
1057 spin_lock_irq(lock
);
1058 if ((sig
->flags
& SIGNAL_GROUP_EXIT
) || sig
->group_exec_task
) {
1060 * Another group action in progress, just
1061 * return so that the signal is processed.
1063 spin_unlock_irq(lock
);
1067 sig
->group_exec_task
= tsk
;
1068 sig
->notify_count
= zap_other_threads(tsk
);
1069 if (!thread_group_leader(tsk
))
1070 sig
->notify_count
--;
1072 while (sig
->notify_count
) {
1073 __set_current_state(TASK_KILLABLE
);
1074 spin_unlock_irq(lock
);
1076 if (__fatal_signal_pending(tsk
))
1078 spin_lock_irq(lock
);
1080 spin_unlock_irq(lock
);
1083 * At this point all other threads have exited, all we have to
1084 * do is to wait for the thread group leader to become inactive,
1085 * and to assume its PID:
1087 if (!thread_group_leader(tsk
)) {
1088 struct task_struct
*leader
= tsk
->group_leader
;
1091 cgroup_threadgroup_change_begin(tsk
);
1092 write_lock_irq(&tasklist_lock
);
1094 * Do this under tasklist_lock to ensure that
1095 * exit_notify() can't miss ->group_exec_task
1097 sig
->notify_count
= -1;
1098 if (likely(leader
->exit_state
))
1100 __set_current_state(TASK_KILLABLE
);
1101 write_unlock_irq(&tasklist_lock
);
1102 cgroup_threadgroup_change_end(tsk
);
1104 if (__fatal_signal_pending(tsk
))
1109 * The only record we have of the real-time age of a
1110 * process, regardless of execs it's done, is start_time.
1111 * All the past CPU time is accumulated in signal_struct
1112 * from sister threads now dead. But in this non-leader
1113 * exec, nothing survives from the original leader thread,
1114 * whose birth marks the true age of this process now.
1115 * When we take on its identity by switching to its PID, we
1116 * also take its birthdate (always earlier than our own).
1118 tsk
->start_time
= leader
->start_time
;
1119 tsk
->start_boottime
= leader
->start_boottime
;
1121 BUG_ON(!same_thread_group(leader
, tsk
));
1123 * An exec() starts a new thread group with the
1124 * TGID of the previous thread group. Rehash the
1125 * two threads with a switched PID, and release
1126 * the former thread group leader:
1129 /* Become a process group leader with the old leader's pid.
1130 * The old leader becomes a thread of the this thread group.
1132 exchange_tids(tsk
, leader
);
1133 transfer_pid(leader
, tsk
, PIDTYPE_TGID
);
1134 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
1135 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
1137 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
1138 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
1140 tsk
->group_leader
= tsk
;
1141 leader
->group_leader
= tsk
;
1143 tsk
->exit_signal
= SIGCHLD
;
1144 leader
->exit_signal
= -1;
1146 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
1147 leader
->exit_state
= EXIT_DEAD
;
1150 * We are going to release_task()->ptrace_unlink() silently,
1151 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1152 * the tracer wont't block again waiting for this thread.
1154 if (unlikely(leader
->ptrace
))
1155 __wake_up_parent(leader
, leader
->parent
);
1156 write_unlock_irq(&tasklist_lock
);
1157 cgroup_threadgroup_change_end(tsk
);
1159 release_task(leader
);
1162 sig
->group_exec_task
= NULL
;
1163 sig
->notify_count
= 0;
1166 /* we have changed execution domain */
1167 tsk
->exit_signal
= SIGCHLD
;
1169 BUG_ON(!thread_group_leader(tsk
));
1173 /* protects against exit_notify() and __exit_signal() */
1174 read_lock(&tasklist_lock
);
1175 sig
->group_exec_task
= NULL
;
1176 sig
->notify_count
= 0;
1177 read_unlock(&tasklist_lock
);
1183 * This function makes sure the current process has its own signal table,
1184 * so that flush_signal_handlers can later reset the handlers without
1185 * disturbing other processes. (Other processes might share the signal
1186 * table via the CLONE_SIGHAND option to clone().)
1188 static int unshare_sighand(struct task_struct
*me
)
1190 struct sighand_struct
*oldsighand
= me
->sighand
;
1192 if (refcount_read(&oldsighand
->count
) != 1) {
1193 struct sighand_struct
*newsighand
;
1195 * This ->sighand is shared with the CLONE_SIGHAND
1196 * but not CLONE_THREAD task, switch to the new one.
1198 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1202 refcount_set(&newsighand
->count
, 1);
1203 memcpy(newsighand
->action
, oldsighand
->action
,
1204 sizeof(newsighand
->action
));
1206 write_lock_irq(&tasklist_lock
);
1207 spin_lock(&oldsighand
->siglock
);
1208 rcu_assign_pointer(me
->sighand
, newsighand
);
1209 spin_unlock(&oldsighand
->siglock
);
1210 write_unlock_irq(&tasklist_lock
);
1212 __cleanup_sighand(oldsighand
);
1217 char *__get_task_comm(char *buf
, size_t buf_size
, struct task_struct
*tsk
)
1220 /* Always NUL terminated and zero-padded */
1221 strscpy_pad(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 strscpy_pad(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 visible until then. This also enables the update
1284 retval
= set_mm_exe_file(bprm
->mm
, bprm
->file
);
1288 /* If the binary is not readable then enforce mm->dumpable=0 */
1289 would_dump(bprm
, bprm
->file
);
1290 if (bprm
->have_execfd
)
1291 would_dump(bprm
, bprm
->executable
);
1294 * Release all of the old mmap stuff
1296 acct_arg_size(bprm
, 0);
1297 retval
= exec_mmap(bprm
->mm
);
1303 #ifdef CONFIG_POSIX_TIMERS
1304 exit_itimers(me
->signal
);
1305 flush_itimer_signals();
1309 * Make the signal table private.
1311 retval
= unshare_sighand(me
);
1315 me
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
|
1316 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1318 me
->personality
&= ~bprm
->per_clear
;
1320 clear_syscall_work_syscall_user_dispatch(me
);
1323 * We have to apply CLOEXEC before we change whether the process is
1324 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1325 * trying to access the should-be-closed file descriptors of a process
1326 * undergoing exec(2).
1328 do_close_on_exec(me
->files
);
1330 if (bprm
->secureexec
) {
1331 /* Make sure parent cannot signal privileged process. */
1332 me
->pdeath_signal
= 0;
1335 * For secureexec, reset the stack limit to sane default to
1336 * avoid bad behavior from the prior rlimits. This has to
1337 * happen before arch_pick_mmap_layout(), which examines
1338 * RLIMIT_STACK, but after the point of no return to avoid
1339 * needing to clean up the change on failure.
1341 if (bprm
->rlim_stack
.rlim_cur
> _STK_LIM
)
1342 bprm
->rlim_stack
.rlim_cur
= _STK_LIM
;
1345 me
->sas_ss_sp
= me
->sas_ss_size
= 0;
1348 * Figure out dumpability. Note that this checking only of current
1349 * is wrong, but userspace depends on it. This should be testing
1350 * bprm->secureexec instead.
1352 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
||
1353 !(uid_eq(current_euid(), current_uid()) &&
1354 gid_eq(current_egid(), current_gid())))
1355 set_dumpable(current
->mm
, suid_dumpable
);
1357 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1360 __set_task_comm(me
, kbasename(bprm
->filename
), true);
1362 /* An exec changes our domain. We are no longer part of the thread
1364 WRITE_ONCE(me
->self_exec_id
, me
->self_exec_id
+ 1);
1365 flush_signal_handlers(me
, 0);
1367 retval
= set_cred_ucounts(bprm
->cred
);
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 struct user_namespace
*mnt_userns
= file_mnt_user_ns(file
);
1416 if (inode_permission(mnt_userns
, inode
, MAY_READ
) < 0) {
1417 struct user_namespace
*old
, *user_ns
;
1418 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1420 /* Ensure mm->user_ns contains the executable */
1421 user_ns
= old
= bprm
->mm
->user_ns
;
1422 while ((user_ns
!= &init_user_ns
) &&
1423 !privileged_wrt_inode_uidgid(user_ns
, mnt_userns
, inode
))
1424 user_ns
= user_ns
->parent
;
1426 if (old
!= user_ns
) {
1427 bprm
->mm
->user_ns
= get_user_ns(user_ns
);
1432 EXPORT_SYMBOL(would_dump
);
1434 void setup_new_exec(struct linux_binprm
* bprm
)
1436 /* Setup things that can depend upon the personality */
1437 struct task_struct
*me
= current
;
1439 arch_pick_mmap_layout(me
->mm
, &bprm
->rlim_stack
);
1441 arch_setup_new_exec();
1443 /* Set the new mm task size. We have to do that late because it may
1444 * depend on TIF_32BIT which is only updated in flush_thread() on
1445 * some architectures like powerpc
1447 me
->mm
->task_size
= TASK_SIZE
;
1448 up_write(&me
->signal
->exec_update_lock
);
1449 mutex_unlock(&me
->signal
->cred_guard_mutex
);
1451 EXPORT_SYMBOL(setup_new_exec
);
1453 /* Runs immediately before start_thread() takes over. */
1454 void finalize_exec(struct linux_binprm
*bprm
)
1456 /* Store any stack rlimit changes before starting thread. */
1457 task_lock(current
->group_leader
);
1458 current
->signal
->rlim
[RLIMIT_STACK
] = bprm
->rlim_stack
;
1459 task_unlock(current
->group_leader
);
1461 EXPORT_SYMBOL(finalize_exec
);
1464 * Prepare credentials and lock ->cred_guard_mutex.
1465 * setup_new_exec() commits the new creds and drops the lock.
1466 * Or, if exec fails before, free_bprm() should release ->cred
1469 static int prepare_bprm_creds(struct linux_binprm
*bprm
)
1471 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1472 return -ERESTARTNOINTR
;
1474 bprm
->cred
= prepare_exec_creds();
1475 if (likely(bprm
->cred
))
1478 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1482 static void free_bprm(struct linux_binprm
*bprm
)
1485 acct_arg_size(bprm
, 0);
1488 free_arg_pages(bprm
);
1490 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1491 abort_creds(bprm
->cred
);
1494 allow_write_access(bprm
->file
);
1497 if (bprm
->executable
)
1498 fput(bprm
->executable
);
1499 /* If a binfmt changed the interp, free it. */
1500 if (bprm
->interp
!= bprm
->filename
)
1501 kfree(bprm
->interp
);
1502 kfree(bprm
->fdpath
);
1506 static struct linux_binprm
*alloc_bprm(int fd
, struct filename
*filename
)
1508 struct linux_binprm
*bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1509 int retval
= -ENOMEM
;
1513 if (fd
== AT_FDCWD
|| filename
->name
[0] == '/') {
1514 bprm
->filename
= filename
->name
;
1516 if (filename
->name
[0] == '\0')
1517 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d", fd
);
1519 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d/%s",
1520 fd
, filename
->name
);
1524 bprm
->filename
= bprm
->fdpath
;
1526 bprm
->interp
= bprm
->filename
;
1528 retval
= bprm_mm_init(bprm
);
1536 return ERR_PTR(retval
);
1539 int bprm_change_interp(const char *interp
, struct linux_binprm
*bprm
)
1541 /* If a binfmt changed the interp, free it first. */
1542 if (bprm
->interp
!= bprm
->filename
)
1543 kfree(bprm
->interp
);
1544 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1549 EXPORT_SYMBOL(bprm_change_interp
);
1552 * determine how safe it is to execute the proposed program
1553 * - the caller must hold ->cred_guard_mutex to protect against
1554 * PTRACE_ATTACH or seccomp thread-sync
1556 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1558 struct task_struct
*p
= current
, *t
;
1562 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1565 * This isn't strictly necessary, but it makes it harder for LSMs to
1568 if (task_no_new_privs(current
))
1569 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1573 spin_lock(&p
->fs
->lock
);
1575 while_each_thread(p
, t
) {
1581 if (p
->fs
->users
> n_fs
)
1582 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1585 spin_unlock(&p
->fs
->lock
);
1588 static void bprm_fill_uid(struct linux_binprm
*bprm
, struct file
*file
)
1590 /* Handle suid and sgid on files */
1591 struct user_namespace
*mnt_userns
;
1592 struct inode
*inode
;
1597 if (!mnt_may_suid(file
->f_path
.mnt
))
1600 if (task_no_new_privs(current
))
1603 inode
= file
->f_path
.dentry
->d_inode
;
1604 mode
= READ_ONCE(inode
->i_mode
);
1605 if (!(mode
& (S_ISUID
|S_ISGID
)))
1608 mnt_userns
= file_mnt_user_ns(file
);
1610 /* Be careful if suid/sgid is set */
1613 /* reload atomically mode/uid/gid now that lock held */
1614 mode
= inode
->i_mode
;
1615 uid
= i_uid_into_mnt(mnt_userns
, inode
);
1616 gid
= i_gid_into_mnt(mnt_userns
, inode
);
1617 inode_unlock(inode
);
1619 /* We ignore suid/sgid if there are no mappings for them in the ns */
1620 if (!kuid_has_mapping(bprm
->cred
->user_ns
, uid
) ||
1621 !kgid_has_mapping(bprm
->cred
->user_ns
, gid
))
1624 if (mode
& S_ISUID
) {
1625 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1626 bprm
->cred
->euid
= uid
;
1629 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1630 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1631 bprm
->cred
->egid
= gid
;
1636 * Compute brpm->cred based upon the final binary.
1638 static int bprm_creds_from_file(struct linux_binprm
*bprm
)
1640 /* Compute creds based on which file? */
1641 struct file
*file
= bprm
->execfd_creds
? bprm
->executable
: bprm
->file
;
1643 bprm_fill_uid(bprm
, file
);
1644 return security_bprm_creds_from_file(bprm
, file
);
1648 * Fill the binprm structure from the inode.
1649 * Read the first BINPRM_BUF_SIZE bytes
1651 * This may be called multiple times for binary chains (scripts for example).
1653 static int prepare_binprm(struct linux_binprm
*bprm
)
1657 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1658 return kernel_read(bprm
->file
, bprm
->buf
, BINPRM_BUF_SIZE
, &pos
);
1662 * Arguments are '\0' separated strings found at the location bprm->p
1663 * points to; chop off the first by relocating brpm->p to right after
1664 * the first '\0' encountered.
1666 int remove_arg_zero(struct linux_binprm
*bprm
)
1669 unsigned long offset
;
1677 offset
= bprm
->p
& ~PAGE_MASK
;
1678 page
= get_arg_page(bprm
, bprm
->p
, 0);
1683 kaddr
= kmap_atomic(page
);
1685 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1686 offset
++, bprm
->p
++)
1689 kunmap_atomic(kaddr
);
1691 } while (offset
== PAGE_SIZE
);
1700 EXPORT_SYMBOL(remove_arg_zero
);
1702 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1704 * cycle the list of binary formats handler, until one recognizes the image
1706 static int search_binary_handler(struct linux_binprm
*bprm
)
1708 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1709 struct linux_binfmt
*fmt
;
1712 retval
= prepare_binprm(bprm
);
1716 retval
= security_bprm_check(bprm
);
1722 read_lock(&binfmt_lock
);
1723 list_for_each_entry(fmt
, &formats
, lh
) {
1724 if (!try_module_get(fmt
->module
))
1726 read_unlock(&binfmt_lock
);
1728 retval
= fmt
->load_binary(bprm
);
1730 read_lock(&binfmt_lock
);
1732 if (bprm
->point_of_no_return
|| (retval
!= -ENOEXEC
)) {
1733 read_unlock(&binfmt_lock
);
1737 read_unlock(&binfmt_lock
);
1740 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1741 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1743 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1752 static int exec_binprm(struct linux_binprm
*bprm
)
1754 pid_t old_pid
, old_vpid
;
1757 /* Need to fetch pid before load_binary changes it */
1758 old_pid
= current
->pid
;
1760 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1763 /* This allows 4 levels of binfmt rewrites before failing hard. */
1764 for (depth
= 0;; depth
++) {
1769 ret
= search_binary_handler(bprm
);
1772 if (!bprm
->interpreter
)
1776 bprm
->file
= bprm
->interpreter
;
1777 bprm
->interpreter
= NULL
;
1779 allow_write_access(exec
);
1780 if (unlikely(bprm
->have_execfd
)) {
1781 if (bprm
->executable
) {
1785 bprm
->executable
= exec
;
1791 trace_sched_process_exec(current
, old_pid
, bprm
);
1792 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1793 proc_exec_connector(current
);
1798 * sys_execve() executes a new program.
1800 static int bprm_execve(struct linux_binprm
*bprm
,
1801 int fd
, struct filename
*filename
, int flags
)
1806 retval
= prepare_bprm_creds(bprm
);
1810 check_unsafe_exec(bprm
);
1811 current
->in_execve
= 1;
1813 file
= do_open_execat(fd
, filename
, flags
);
1814 retval
= PTR_ERR(file
);
1822 * Record that a name derived from an O_CLOEXEC fd will be
1823 * inaccessible after exec. This allows the code in exec to
1824 * choose to fail when the executable is not mmaped into the
1825 * interpreter and an open file descriptor is not passed to
1826 * the interpreter. This makes for a better user experience
1827 * than having the interpreter start and then immediately fail
1828 * when it finds the executable is inaccessible.
1830 if (bprm
->fdpath
&& get_close_on_exec(fd
))
1831 bprm
->interp_flags
|= BINPRM_FLAGS_PATH_INACCESSIBLE
;
1833 /* Set the unchanging part of bprm->cred */
1834 retval
= security_bprm_creds_for_exec(bprm
);
1838 retval
= exec_binprm(bprm
);
1842 /* execve succeeded */
1843 current
->fs
->in_exec
= 0;
1844 current
->in_execve
= 0;
1845 rseq_execve(current
);
1846 acct_update_integrals(current
);
1847 task_numa_free(current
, false);
1852 * If past the point of no return ensure the code never
1853 * returns to the userspace process. Use an existing fatal
1854 * signal if present otherwise terminate the process with
1857 if (bprm
->point_of_no_return
&& !fatal_signal_pending(current
))
1858 force_fatal_sig(SIGSEGV
);
1861 current
->fs
->in_exec
= 0;
1862 current
->in_execve
= 0;
1867 static int do_execveat_common(int fd
, struct filename
*filename
,
1868 struct user_arg_ptr argv
,
1869 struct user_arg_ptr envp
,
1872 struct linux_binprm
*bprm
;
1875 if (IS_ERR(filename
))
1876 return PTR_ERR(filename
);
1879 * We move the actual failure in case of RLIMIT_NPROC excess from
1880 * set*uid() to execve() because too many poorly written programs
1881 * don't check setuid() return code. Here we additionally recheck
1882 * whether NPROC limit is still exceeded.
1884 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1885 is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC
, rlimit(RLIMIT_NPROC
))) {
1890 /* We're below the limit (still or again), so we don't want to make
1891 * further execve() calls fail. */
1892 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1894 bprm
= alloc_bprm(fd
, filename
);
1896 retval
= PTR_ERR(bprm
);
1900 retval
= count(argv
, MAX_ARG_STRINGS
);
1902 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1903 current
->comm
, bprm
->filename
);
1906 bprm
->argc
= retval
;
1908 retval
= count(envp
, MAX_ARG_STRINGS
);
1911 bprm
->envc
= retval
;
1913 retval
= bprm_stack_limits(bprm
);
1917 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1920 bprm
->exec
= bprm
->p
;
1922 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1926 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1931 * When argv is empty, add an empty string ("") as argv[0] to
1932 * ensure confused userspace programs that start processing
1933 * from argv[1] won't end up walking envp. See also
1934 * bprm_stack_limits().
1936 if (bprm
->argc
== 0) {
1937 retval
= copy_string_kernel("", bprm
);
1943 retval
= bprm_execve(bprm
, fd
, filename
, flags
);
1952 int kernel_execve(const char *kernel_filename
,
1953 const char *const *argv
, const char *const *envp
)
1955 struct filename
*filename
;
1956 struct linux_binprm
*bprm
;
1960 /* It is non-sense for kernel threads to call execve */
1961 if (WARN_ON_ONCE(current
->flags
& PF_KTHREAD
))
1964 filename
= getname_kernel(kernel_filename
);
1965 if (IS_ERR(filename
))
1966 return PTR_ERR(filename
);
1968 bprm
= alloc_bprm(fd
, filename
);
1970 retval
= PTR_ERR(bprm
);
1974 retval
= count_strings_kernel(argv
);
1975 if (WARN_ON_ONCE(retval
== 0))
1979 bprm
->argc
= retval
;
1981 retval
= count_strings_kernel(envp
);
1984 bprm
->envc
= retval
;
1986 retval
= bprm_stack_limits(bprm
);
1990 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1993 bprm
->exec
= bprm
->p
;
1995 retval
= copy_strings_kernel(bprm
->envc
, envp
, bprm
);
1999 retval
= copy_strings_kernel(bprm
->argc
, argv
, bprm
);
2003 retval
= bprm_execve(bprm
, fd
, filename
, 0);
2011 static int do_execve(struct filename
*filename
,
2012 const char __user
*const __user
*__argv
,
2013 const char __user
*const __user
*__envp
)
2015 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2016 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2017 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2020 static int do_execveat(int fd
, struct filename
*filename
,
2021 const char __user
*const __user
*__argv
,
2022 const char __user
*const __user
*__envp
,
2025 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2026 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2028 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2031 #ifdef CONFIG_COMPAT
2032 static int compat_do_execve(struct filename
*filename
,
2033 const compat_uptr_t __user
*__argv
,
2034 const compat_uptr_t __user
*__envp
)
2036 struct user_arg_ptr argv
= {
2038 .ptr
.compat
= __argv
,
2040 struct user_arg_ptr envp
= {
2042 .ptr
.compat
= __envp
,
2044 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2047 static int compat_do_execveat(int fd
, struct filename
*filename
,
2048 const compat_uptr_t __user
*__argv
,
2049 const compat_uptr_t __user
*__envp
,
2052 struct user_arg_ptr argv
= {
2054 .ptr
.compat
= __argv
,
2056 struct user_arg_ptr envp
= {
2058 .ptr
.compat
= __envp
,
2060 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2064 void set_binfmt(struct linux_binfmt
*new)
2066 struct mm_struct
*mm
= current
->mm
;
2069 module_put(mm
->binfmt
->module
);
2073 __module_get(new->module
);
2075 EXPORT_SYMBOL(set_binfmt
);
2078 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2080 void set_dumpable(struct mm_struct
*mm
, int value
)
2082 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
2085 set_mask_bits(&mm
->flags
, MMF_DUMPABLE_MASK
, value
);
2088 SYSCALL_DEFINE3(execve
,
2089 const char __user
*, filename
,
2090 const char __user
*const __user
*, argv
,
2091 const char __user
*const __user
*, envp
)
2093 return do_execve(getname(filename
), argv
, envp
);
2096 SYSCALL_DEFINE5(execveat
,
2097 int, fd
, const char __user
*, filename
,
2098 const char __user
*const __user
*, argv
,
2099 const char __user
*const __user
*, envp
,
2102 return do_execveat(fd
,
2103 getname_uflags(filename
, flags
),
2107 #ifdef CONFIG_COMPAT
2108 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
2109 const compat_uptr_t __user
*, argv
,
2110 const compat_uptr_t __user
*, envp
)
2112 return compat_do_execve(getname(filename
), argv
, envp
);
2115 COMPAT_SYSCALL_DEFINE5(execveat
, int, fd
,
2116 const char __user
*, filename
,
2117 const compat_uptr_t __user
*, argv
,
2118 const compat_uptr_t __user
*, envp
,
2121 return compat_do_execveat(fd
,
2122 getname_uflags(filename
, flags
),
2127 #ifdef CONFIG_SYSCTL
2129 static int proc_dointvec_minmax_coredump(struct ctl_table
*table
, int write
,
2130 void *buffer
, size_t *lenp
, loff_t
*ppos
)
2132 int error
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
2135 validate_coredump_safety();
2139 static struct ctl_table fs_exec_sysctls
[] = {
2141 .procname
= "suid_dumpable",
2142 .data
= &suid_dumpable
,
2143 .maxlen
= sizeof(int),
2145 .proc_handler
= proc_dointvec_minmax_coredump
,
2146 .extra1
= SYSCTL_ZERO
,
2147 .extra2
= SYSCTL_TWO
,
2152 static int __init
init_fs_exec_sysctls(void)
2154 register_sysctl_init("fs", fs_exec_sysctls
);
2158 fs_initcall(init_fs_exec_sysctls
);
2159 #endif /* CONFIG_SYSCTL */