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/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/kmod.h>
59 #include <linux/fsnotify.h>
60 #include <linux/fs_struct.h>
61 #include <linux/oom.h>
62 #include <linux/compat.h>
63 #include <linux/vmalloc.h>
64 #include <linux/io_uring.h>
65 #include <linux/syscall_user_dispatch.h>
66 #include <linux/coredump.h>
67 #include <linux/time_namespace.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
= 0;
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 vm_flags_init(vma
, 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
);
587 put_arg_page(kmapped_page
);
590 kaddr
= kmap_local_page(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
);
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
));
634 pos
-= bytes_to_copy
;
635 arg
-= bytes_to_copy
;
636 len
-= bytes_to_copy
;
638 page
= get_arg_page(bprm
, pos
, 1);
641 flush_arg_page(bprm
, pos
& PAGE_MASK
, page
);
642 memcpy_to_page(page
, offset_in_page(pos
), arg
, bytes_to_copy
);
648 EXPORT_SYMBOL(copy_string_kernel
);
650 static int copy_strings_kernel(int argc
, const char *const *argv
,
651 struct linux_binprm
*bprm
)
654 int ret
= copy_string_kernel(argv
[argc
], bprm
);
657 if (fatal_signal_pending(current
))
658 return -ERESTARTNOHAND
;
667 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
668 * the binfmt code determines where the new stack should reside, we shift it to
669 * its final location. The process proceeds as follows:
671 * 1) Use shift to calculate the new vma endpoints.
672 * 2) Extend vma to cover both the old and new ranges. This ensures the
673 * arguments passed to subsequent functions are consistent.
674 * 3) Move vma's page tables to the new range.
675 * 4) Free up any cleared pgd range.
676 * 5) Shrink the vma to cover only the new range.
678 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
680 struct mm_struct
*mm
= vma
->vm_mm
;
681 unsigned long old_start
= vma
->vm_start
;
682 unsigned long old_end
= vma
->vm_end
;
683 unsigned long length
= old_end
- old_start
;
684 unsigned long new_start
= old_start
- shift
;
685 unsigned long new_end
= old_end
- shift
;
686 VMA_ITERATOR(vmi
, mm
, new_start
);
687 struct vm_area_struct
*next
;
688 struct mmu_gather tlb
;
690 BUG_ON(new_start
> new_end
);
693 * ensure there are no vmas between where we want to go
696 if (vma
!= vma_next(&vmi
))
700 * cover the whole range: [new_start, old_end)
702 if (vma_expand(&vmi
, vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
706 * move the page tables downwards, on failure we rely on
707 * process cleanup to remove whatever mess we made.
709 if (length
!= move_page_tables(vma
, old_start
,
710 vma
, new_start
, length
, false))
714 tlb_gather_mmu(&tlb
, mm
);
715 next
= vma_next(&vmi
);
716 if (new_end
> old_start
) {
718 * when the old and new regions overlap clear from new_end.
720 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
721 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
724 * otherwise, clean from old_start; this is done to not touch
725 * the address space in [new_end, old_start) some architectures
726 * have constraints on va-space that make this illegal (IA64) -
727 * for the others its just a little faster.
729 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
730 next
? next
->vm_start
: USER_PGTABLES_CEILING
);
732 tlb_finish_mmu(&tlb
);
735 /* Shrink the vma to just the new range */
736 return vma_shrink(&vmi
, vma
, new_start
, new_end
, vma
->vm_pgoff
);
740 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
741 * the stack is optionally relocated, and some extra space is added.
743 int setup_arg_pages(struct linux_binprm
*bprm
,
744 unsigned long stack_top
,
745 int executable_stack
)
748 unsigned long stack_shift
;
749 struct mm_struct
*mm
= current
->mm
;
750 struct vm_area_struct
*vma
= bprm
->vma
;
751 struct vm_area_struct
*prev
= NULL
;
752 unsigned long vm_flags
;
753 unsigned long stack_base
;
754 unsigned long stack_size
;
755 unsigned long stack_expand
;
756 unsigned long rlim_stack
;
757 struct mmu_gather tlb
;
758 struct vma_iterator vmi
;
760 #ifdef CONFIG_STACK_GROWSUP
761 /* Limit stack size */
762 stack_base
= bprm
->rlim_stack
.rlim_max
;
764 stack_base
= calc_max_stack_size(stack_base
);
766 /* Add space for stack randomization. */
767 stack_base
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
769 /* Make sure we didn't let the argument array grow too large. */
770 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
773 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
775 stack_shift
= vma
->vm_start
- stack_base
;
776 mm
->arg_start
= bprm
->p
- stack_shift
;
777 bprm
->p
= vma
->vm_end
- stack_shift
;
779 stack_top
= arch_align_stack(stack_top
);
780 stack_top
= PAGE_ALIGN(stack_top
);
782 if (unlikely(stack_top
< mmap_min_addr
) ||
783 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
786 stack_shift
= vma
->vm_end
- stack_top
;
788 bprm
->p
-= stack_shift
;
789 mm
->arg_start
= bprm
->p
;
793 bprm
->loader
-= stack_shift
;
794 bprm
->exec
-= stack_shift
;
796 if (mmap_write_lock_killable(mm
))
799 vm_flags
= VM_STACK_FLAGS
;
802 * Adjust stack execute permissions; explicitly enable for
803 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
804 * (arch default) otherwise.
806 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
808 else if (executable_stack
== EXSTACK_DISABLE_X
)
809 vm_flags
&= ~VM_EXEC
;
810 vm_flags
|= mm
->def_flags
;
811 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
813 vma_iter_init(&vmi
, mm
, vma
->vm_start
);
815 tlb_gather_mmu(&tlb
, mm
);
816 ret
= mprotect_fixup(&vmi
, &tlb
, vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
818 tlb_finish_mmu(&tlb
);
824 if (unlikely(vm_flags
& VM_EXEC
)) {
825 pr_warn_once("process '%pD4' started with executable stack\n",
829 /* Move stack pages down in memory. */
831 ret
= shift_arg_pages(vma
, stack_shift
);
836 /* mprotect_fixup is overkill to remove the temporary stack flags */
837 vm_flags_clear(vma
, VM_STACK_INCOMPLETE_SETUP
);
839 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
840 stack_size
= vma
->vm_end
- vma
->vm_start
;
842 * Align this down to a page boundary as expand_stack
845 rlim_stack
= bprm
->rlim_stack
.rlim_cur
& PAGE_MASK
;
847 stack_expand
= min(rlim_stack
, stack_size
+ stack_expand
);
849 #ifdef CONFIG_STACK_GROWSUP
850 stack_base
= vma
->vm_start
+ stack_expand
;
852 stack_base
= vma
->vm_end
- stack_expand
;
854 current
->mm
->start_stack
= bprm
->p
;
855 ret
= expand_stack(vma
, stack_base
);
860 mmap_write_unlock(mm
);
863 EXPORT_SYMBOL(setup_arg_pages
);
868 * Transfer the program arguments and environment from the holding pages
869 * onto the stack. The provided stack pointer is adjusted accordingly.
871 int transfer_args_to_stack(struct linux_binprm
*bprm
,
872 unsigned long *sp_location
)
874 unsigned long index
, stop
, sp
;
877 stop
= bprm
->p
>> PAGE_SHIFT
;
880 for (index
= MAX_ARG_PAGES
- 1; index
>= stop
; index
--) {
881 unsigned int offset
= index
== stop
? bprm
->p
& ~PAGE_MASK
: 0;
882 char *src
= kmap_local_page(bprm
->page
[index
]) + offset
;
883 sp
-= PAGE_SIZE
- offset
;
884 if (copy_to_user((void *) sp
, src
, PAGE_SIZE
- offset
) != 0)
896 EXPORT_SYMBOL(transfer_args_to_stack
);
898 #endif /* CONFIG_MMU */
900 static struct file
*do_open_execat(int fd
, struct filename
*name
, int flags
)
904 struct open_flags open_exec_flags
= {
905 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
906 .acc_mode
= MAY_EXEC
,
907 .intent
= LOOKUP_OPEN
,
908 .lookup_flags
= LOOKUP_FOLLOW
,
911 if ((flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
912 return ERR_PTR(-EINVAL
);
913 if (flags
& AT_SYMLINK_NOFOLLOW
)
914 open_exec_flags
.lookup_flags
&= ~LOOKUP_FOLLOW
;
915 if (flags
& AT_EMPTY_PATH
)
916 open_exec_flags
.lookup_flags
|= LOOKUP_EMPTY
;
918 file
= do_filp_open(fd
, name
, &open_exec_flags
);
923 * may_open() has already checked for this, so it should be
924 * impossible to trip now. But we need to be extra cautious
925 * and check again at the very end too.
928 if (WARN_ON_ONCE(!S_ISREG(file_inode(file
)->i_mode
) ||
929 path_noexec(&file
->f_path
)))
932 err
= deny_write_access(file
);
936 if (name
->name
[0] != '\0')
947 struct file
*open_exec(const char *name
)
949 struct filename
*filename
= getname_kernel(name
);
950 struct file
*f
= ERR_CAST(filename
);
952 if (!IS_ERR(filename
)) {
953 f
= do_open_execat(AT_FDCWD
, filename
, 0);
958 EXPORT_SYMBOL(open_exec
);
960 #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
961 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
963 ssize_t res
= vfs_read(file
, (void __user
*)addr
, len
, &pos
);
965 flush_icache_user_range(addr
, addr
+ len
);
968 EXPORT_SYMBOL(read_code
);
972 * Maps the mm_struct mm into the current task struct.
973 * On success, this function returns with exec_update_lock
976 static int exec_mmap(struct mm_struct
*mm
)
978 struct task_struct
*tsk
;
979 struct mm_struct
*old_mm
, *active_mm
;
982 /* Notify parent that we're no longer interested in the old VM */
984 old_mm
= current
->mm
;
985 exec_mm_release(tsk
, old_mm
);
989 ret
= down_write_killable(&tsk
->signal
->exec_update_lock
);
995 * If there is a pending fatal signal perhaps a signal
996 * whose default action is to create a coredump get
997 * out and die instead of going through with the exec.
999 ret
= mmap_read_lock_killable(old_mm
);
1001 up_write(&tsk
->signal
->exec_update_lock
);
1007 membarrier_exec_mmap(mm
);
1009 local_irq_disable();
1010 active_mm
= tsk
->active_mm
;
1011 tsk
->active_mm
= mm
;
1015 * This prevents preemption while active_mm is being loaded and
1016 * it and mm are being updated, which could cause problems for
1017 * lazy tlb mm refcounting when these are updated by context
1018 * switches. Not all architectures can handle irqs off over
1021 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1023 activate_mm(active_mm
, mm
);
1024 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM
))
1030 mmap_read_unlock(old_mm
);
1031 BUG_ON(active_mm
!= old_mm
);
1032 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
1033 mm_update_next_owner(old_mm
);
1041 static int de_thread(struct task_struct
*tsk
)
1043 struct signal_struct
*sig
= tsk
->signal
;
1044 struct sighand_struct
*oldsighand
= tsk
->sighand
;
1045 spinlock_t
*lock
= &oldsighand
->siglock
;
1047 if (thread_group_empty(tsk
))
1048 goto no_thread_group
;
1051 * Kill all other threads in the thread group.
1053 spin_lock_irq(lock
);
1054 if ((sig
->flags
& SIGNAL_GROUP_EXIT
) || sig
->group_exec_task
) {
1056 * Another group action in progress, just
1057 * return so that the signal is processed.
1059 spin_unlock_irq(lock
);
1063 sig
->group_exec_task
= tsk
;
1064 sig
->notify_count
= zap_other_threads(tsk
);
1065 if (!thread_group_leader(tsk
))
1066 sig
->notify_count
--;
1068 while (sig
->notify_count
) {
1069 __set_current_state(TASK_KILLABLE
);
1070 spin_unlock_irq(lock
);
1072 if (__fatal_signal_pending(tsk
))
1074 spin_lock_irq(lock
);
1076 spin_unlock_irq(lock
);
1079 * At this point all other threads have exited, all we have to
1080 * do is to wait for the thread group leader to become inactive,
1081 * and to assume its PID:
1083 if (!thread_group_leader(tsk
)) {
1084 struct task_struct
*leader
= tsk
->group_leader
;
1087 cgroup_threadgroup_change_begin(tsk
);
1088 write_lock_irq(&tasklist_lock
);
1090 * Do this under tasklist_lock to ensure that
1091 * exit_notify() can't miss ->group_exec_task
1093 sig
->notify_count
= -1;
1094 if (likely(leader
->exit_state
))
1096 __set_current_state(TASK_KILLABLE
);
1097 write_unlock_irq(&tasklist_lock
);
1098 cgroup_threadgroup_change_end(tsk
);
1100 if (__fatal_signal_pending(tsk
))
1105 * The only record we have of the real-time age of a
1106 * process, regardless of execs it's done, is start_time.
1107 * All the past CPU time is accumulated in signal_struct
1108 * from sister threads now dead. But in this non-leader
1109 * exec, nothing survives from the original leader thread,
1110 * whose birth marks the true age of this process now.
1111 * When we take on its identity by switching to its PID, we
1112 * also take its birthdate (always earlier than our own).
1114 tsk
->start_time
= leader
->start_time
;
1115 tsk
->start_boottime
= leader
->start_boottime
;
1117 BUG_ON(!same_thread_group(leader
, tsk
));
1119 * An exec() starts a new thread group with the
1120 * TGID of the previous thread group. Rehash the
1121 * two threads with a switched PID, and release
1122 * the former thread group leader:
1125 /* Become a process group leader with the old leader's pid.
1126 * The old leader becomes a thread of the this thread group.
1128 exchange_tids(tsk
, leader
);
1129 transfer_pid(leader
, tsk
, PIDTYPE_TGID
);
1130 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
1131 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
1133 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
1134 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
1136 tsk
->group_leader
= tsk
;
1137 leader
->group_leader
= tsk
;
1139 tsk
->exit_signal
= SIGCHLD
;
1140 leader
->exit_signal
= -1;
1142 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
1143 leader
->exit_state
= EXIT_DEAD
;
1146 * We are going to release_task()->ptrace_unlink() silently,
1147 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1148 * the tracer won't block again waiting for this thread.
1150 if (unlikely(leader
->ptrace
))
1151 __wake_up_parent(leader
, leader
->parent
);
1152 write_unlock_irq(&tasklist_lock
);
1153 cgroup_threadgroup_change_end(tsk
);
1155 release_task(leader
);
1158 sig
->group_exec_task
= NULL
;
1159 sig
->notify_count
= 0;
1162 /* we have changed execution domain */
1163 tsk
->exit_signal
= SIGCHLD
;
1165 BUG_ON(!thread_group_leader(tsk
));
1169 /* protects against exit_notify() and __exit_signal() */
1170 read_lock(&tasklist_lock
);
1171 sig
->group_exec_task
= NULL
;
1172 sig
->notify_count
= 0;
1173 read_unlock(&tasklist_lock
);
1179 * This function makes sure the current process has its own signal table,
1180 * so that flush_signal_handlers can later reset the handlers without
1181 * disturbing other processes. (Other processes might share the signal
1182 * table via the CLONE_SIGHAND option to clone().)
1184 static int unshare_sighand(struct task_struct
*me
)
1186 struct sighand_struct
*oldsighand
= me
->sighand
;
1188 if (refcount_read(&oldsighand
->count
) != 1) {
1189 struct sighand_struct
*newsighand
;
1191 * This ->sighand is shared with the CLONE_SIGHAND
1192 * but not CLONE_THREAD task, switch to the new one.
1194 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1198 refcount_set(&newsighand
->count
, 1);
1200 write_lock_irq(&tasklist_lock
);
1201 spin_lock(&oldsighand
->siglock
);
1202 memcpy(newsighand
->action
, oldsighand
->action
,
1203 sizeof(newsighand
->action
));
1204 rcu_assign_pointer(me
->sighand
, newsighand
);
1205 spin_unlock(&oldsighand
->siglock
);
1206 write_unlock_irq(&tasklist_lock
);
1208 __cleanup_sighand(oldsighand
);
1213 char *__get_task_comm(char *buf
, size_t buf_size
, struct task_struct
*tsk
)
1216 /* Always NUL terminated and zero-padded */
1217 strscpy_pad(buf
, tsk
->comm
, buf_size
);
1221 EXPORT_SYMBOL_GPL(__get_task_comm
);
1224 * These functions flushes out all traces of the currently running executable
1225 * so that a new one can be started
1228 void __set_task_comm(struct task_struct
*tsk
, const char *buf
, bool exec
)
1231 trace_task_rename(tsk
, buf
);
1232 strscpy_pad(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1234 perf_event_comm(tsk
, exec
);
1238 * Calling this is the point of no return. None of the failures will be
1239 * seen by userspace since either the process is already taking a fatal
1240 * signal (via de_thread() or coredump), or will have SEGV raised
1241 * (after exec_mmap()) by search_binary_handler (see below).
1243 int begin_new_exec(struct linux_binprm
* bprm
)
1245 struct task_struct
*me
= current
;
1248 /* Once we are committed compute the creds */
1249 retval
= bprm_creds_from_file(bprm
);
1254 * Ensure all future errors are fatal.
1256 bprm
->point_of_no_return
= true;
1259 * Make this the only thread in the thread group.
1261 retval
= de_thread(me
);
1266 * Cancel any io_uring activity across execve
1268 io_uring_task_cancel();
1270 /* Ensure the files table is not shared. */
1271 retval
= unshare_files();
1276 * Must be called _before_ exec_mmap() as bprm->mm is
1277 * not visible until then. This also enables the update
1280 retval
= set_mm_exe_file(bprm
->mm
, bprm
->file
);
1284 /* If the binary is not readable then enforce mm->dumpable=0 */
1285 would_dump(bprm
, bprm
->file
);
1286 if (bprm
->have_execfd
)
1287 would_dump(bprm
, bprm
->executable
);
1290 * Release all of the old mmap stuff
1292 acct_arg_size(bprm
, 0);
1293 retval
= exec_mmap(bprm
->mm
);
1299 retval
= exec_task_namespaces();
1303 #ifdef CONFIG_POSIX_TIMERS
1304 spin_lock_irq(&me
->sighand
->siglock
);
1305 posix_cpu_timers_exit(me
);
1306 spin_unlock_irq(&me
->sighand
->siglock
);
1308 flush_itimer_signals();
1312 * Make the signal table private.
1314 retval
= unshare_sighand(me
);
1318 me
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
|
1319 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1321 me
->personality
&= ~bprm
->per_clear
;
1323 clear_syscall_work_syscall_user_dispatch(me
);
1326 * We have to apply CLOEXEC before we change whether the process is
1327 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1328 * trying to access the should-be-closed file descriptors of a process
1329 * undergoing exec(2).
1331 do_close_on_exec(me
->files
);
1333 if (bprm
->secureexec
) {
1334 /* Make sure parent cannot signal privileged process. */
1335 me
->pdeath_signal
= 0;
1338 * For secureexec, reset the stack limit to sane default to
1339 * avoid bad behavior from the prior rlimits. This has to
1340 * happen before arch_pick_mmap_layout(), which examines
1341 * RLIMIT_STACK, but after the point of no return to avoid
1342 * needing to clean up the change on failure.
1344 if (bprm
->rlim_stack
.rlim_cur
> _STK_LIM
)
1345 bprm
->rlim_stack
.rlim_cur
= _STK_LIM
;
1348 me
->sas_ss_sp
= me
->sas_ss_size
= 0;
1351 * Figure out dumpability. Note that this checking only of current
1352 * is wrong, but userspace depends on it. This should be testing
1353 * bprm->secureexec instead.
1355 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
||
1356 !(uid_eq(current_euid(), current_uid()) &&
1357 gid_eq(current_egid(), current_gid())))
1358 set_dumpable(current
->mm
, suid_dumpable
);
1360 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1363 __set_task_comm(me
, kbasename(bprm
->filename
), true);
1365 /* An exec changes our domain. We are no longer part of the thread
1367 WRITE_ONCE(me
->self_exec_id
, me
->self_exec_id
+ 1);
1368 flush_signal_handlers(me
, 0);
1370 retval
= set_cred_ucounts(bprm
->cred
);
1375 * install the new credentials for this executable
1377 security_bprm_committing_creds(bprm
);
1379 commit_creds(bprm
->cred
);
1383 * Disable monitoring for regular users
1384 * when executing setuid binaries. Must
1385 * wait until new credentials are committed
1386 * by commit_creds() above
1388 if (get_dumpable(me
->mm
) != SUID_DUMP_USER
)
1389 perf_event_exit_task(me
);
1391 * cred_guard_mutex must be held at least to this point to prevent
1392 * ptrace_attach() from altering our determination of the task's
1393 * credentials; any time after this it may be unlocked.
1395 security_bprm_committed_creds(bprm
);
1397 /* Pass the opened binary to the interpreter. */
1398 if (bprm
->have_execfd
) {
1399 retval
= get_unused_fd_flags(0);
1402 fd_install(retval
, bprm
->executable
);
1403 bprm
->executable
= NULL
;
1404 bprm
->execfd
= retval
;
1409 up_write(&me
->signal
->exec_update_lock
);
1413 EXPORT_SYMBOL(begin_new_exec
);
1415 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1417 struct inode
*inode
= file_inode(file
);
1418 struct mnt_idmap
*idmap
= file_mnt_idmap(file
);
1419 if (inode_permission(idmap
, inode
, MAY_READ
) < 0) {
1420 struct user_namespace
*old
, *user_ns
;
1421 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1423 /* Ensure mm->user_ns contains the executable */
1424 user_ns
= old
= bprm
->mm
->user_ns
;
1425 while ((user_ns
!= &init_user_ns
) &&
1426 !privileged_wrt_inode_uidgid(user_ns
, idmap
, inode
))
1427 user_ns
= user_ns
->parent
;
1429 if (old
!= user_ns
) {
1430 bprm
->mm
->user_ns
= get_user_ns(user_ns
);
1435 EXPORT_SYMBOL(would_dump
);
1437 void setup_new_exec(struct linux_binprm
* bprm
)
1439 /* Setup things that can depend upon the personality */
1440 struct task_struct
*me
= current
;
1442 arch_pick_mmap_layout(me
->mm
, &bprm
->rlim_stack
);
1444 arch_setup_new_exec();
1446 /* Set the new mm task size. We have to do that late because it may
1447 * depend on TIF_32BIT which is only updated in flush_thread() on
1448 * some architectures like powerpc
1450 me
->mm
->task_size
= TASK_SIZE
;
1451 up_write(&me
->signal
->exec_update_lock
);
1452 mutex_unlock(&me
->signal
->cred_guard_mutex
);
1454 EXPORT_SYMBOL(setup_new_exec
);
1456 /* Runs immediately before start_thread() takes over. */
1457 void finalize_exec(struct linux_binprm
*bprm
)
1459 /* Store any stack rlimit changes before starting thread. */
1460 task_lock(current
->group_leader
);
1461 current
->signal
->rlim
[RLIMIT_STACK
] = bprm
->rlim_stack
;
1462 task_unlock(current
->group_leader
);
1464 EXPORT_SYMBOL(finalize_exec
);
1467 * Prepare credentials and lock ->cred_guard_mutex.
1468 * setup_new_exec() commits the new creds and drops the lock.
1469 * Or, if exec fails before, free_bprm() should release ->cred
1472 static int prepare_bprm_creds(struct linux_binprm
*bprm
)
1474 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1475 return -ERESTARTNOINTR
;
1477 bprm
->cred
= prepare_exec_creds();
1478 if (likely(bprm
->cred
))
1481 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1485 static void free_bprm(struct linux_binprm
*bprm
)
1488 acct_arg_size(bprm
, 0);
1491 free_arg_pages(bprm
);
1493 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1494 abort_creds(bprm
->cred
);
1497 allow_write_access(bprm
->file
);
1500 if (bprm
->executable
)
1501 fput(bprm
->executable
);
1502 /* If a binfmt changed the interp, free it. */
1503 if (bprm
->interp
!= bprm
->filename
)
1504 kfree(bprm
->interp
);
1505 kfree(bprm
->fdpath
);
1509 static struct linux_binprm
*alloc_bprm(int fd
, struct filename
*filename
)
1511 struct linux_binprm
*bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1512 int retval
= -ENOMEM
;
1516 if (fd
== AT_FDCWD
|| filename
->name
[0] == '/') {
1517 bprm
->filename
= filename
->name
;
1519 if (filename
->name
[0] == '\0')
1520 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d", fd
);
1522 bprm
->fdpath
= kasprintf(GFP_KERNEL
, "/dev/fd/%d/%s",
1523 fd
, filename
->name
);
1527 bprm
->filename
= bprm
->fdpath
;
1529 bprm
->interp
= bprm
->filename
;
1531 retval
= bprm_mm_init(bprm
);
1539 return ERR_PTR(retval
);
1542 int bprm_change_interp(const char *interp
, struct linux_binprm
*bprm
)
1544 /* If a binfmt changed the interp, free it first. */
1545 if (bprm
->interp
!= bprm
->filename
)
1546 kfree(bprm
->interp
);
1547 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1552 EXPORT_SYMBOL(bprm_change_interp
);
1555 * determine how safe it is to execute the proposed program
1556 * - the caller must hold ->cred_guard_mutex to protect against
1557 * PTRACE_ATTACH or seccomp thread-sync
1559 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1561 struct task_struct
*p
= current
, *t
;
1565 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1568 * This isn't strictly necessary, but it makes it harder for LSMs to
1571 if (task_no_new_privs(current
))
1572 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1575 * If another task is sharing our fs, we cannot safely
1576 * suid exec because the differently privileged task
1577 * will be able to manipulate the current directory, etc.
1578 * It would be nice to force an unshare instead...
1582 spin_lock(&p
->fs
->lock
);
1584 while_each_thread(p
, t
) {
1590 if (p
->fs
->users
> n_fs
)
1591 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1594 spin_unlock(&p
->fs
->lock
);
1597 static void bprm_fill_uid(struct linux_binprm
*bprm
, struct file
*file
)
1599 /* Handle suid and sgid on files */
1600 struct mnt_idmap
*idmap
;
1601 struct inode
*inode
= file_inode(file
);
1606 if (!mnt_may_suid(file
->f_path
.mnt
))
1609 if (task_no_new_privs(current
))
1612 mode
= READ_ONCE(inode
->i_mode
);
1613 if (!(mode
& (S_ISUID
|S_ISGID
)))
1616 idmap
= file_mnt_idmap(file
);
1618 /* Be careful if suid/sgid is set */
1621 /* reload atomically mode/uid/gid now that lock held */
1622 mode
= inode
->i_mode
;
1623 vfsuid
= i_uid_into_vfsuid(idmap
, inode
);
1624 vfsgid
= i_gid_into_vfsgid(idmap
, inode
);
1625 inode_unlock(inode
);
1627 /* We ignore suid/sgid if there are no mappings for them in the ns */
1628 if (!vfsuid_has_mapping(bprm
->cred
->user_ns
, vfsuid
) ||
1629 !vfsgid_has_mapping(bprm
->cred
->user_ns
, vfsgid
))
1632 if (mode
& S_ISUID
) {
1633 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1634 bprm
->cred
->euid
= vfsuid_into_kuid(vfsuid
);
1637 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1638 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1639 bprm
->cred
->egid
= vfsgid_into_kgid(vfsgid
);
1644 * Compute brpm->cred based upon the final binary.
1646 static int bprm_creds_from_file(struct linux_binprm
*bprm
)
1648 /* Compute creds based on which file? */
1649 struct file
*file
= bprm
->execfd_creds
? bprm
->executable
: bprm
->file
;
1651 bprm_fill_uid(bprm
, file
);
1652 return security_bprm_creds_from_file(bprm
, file
);
1656 * Fill the binprm structure from the inode.
1657 * Read the first BINPRM_BUF_SIZE bytes
1659 * This may be called multiple times for binary chains (scripts for example).
1661 static int prepare_binprm(struct linux_binprm
*bprm
)
1665 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1666 return kernel_read(bprm
->file
, bprm
->buf
, BINPRM_BUF_SIZE
, &pos
);
1670 * Arguments are '\0' separated strings found at the location bprm->p
1671 * points to; chop off the first by relocating brpm->p to right after
1672 * the first '\0' encountered.
1674 int remove_arg_zero(struct linux_binprm
*bprm
)
1677 unsigned long offset
;
1685 offset
= bprm
->p
& ~PAGE_MASK
;
1686 page
= get_arg_page(bprm
, bprm
->p
, 0);
1691 kaddr
= kmap_local_page(page
);
1693 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1694 offset
++, bprm
->p
++)
1697 kunmap_local(kaddr
);
1699 } while (offset
== PAGE_SIZE
);
1708 EXPORT_SYMBOL(remove_arg_zero
);
1710 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1712 * cycle the list of binary formats handler, until one recognizes the image
1714 static int search_binary_handler(struct linux_binprm
*bprm
)
1716 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1717 struct linux_binfmt
*fmt
;
1720 retval
= prepare_binprm(bprm
);
1724 retval
= security_bprm_check(bprm
);
1730 read_lock(&binfmt_lock
);
1731 list_for_each_entry(fmt
, &formats
, lh
) {
1732 if (!try_module_get(fmt
->module
))
1734 read_unlock(&binfmt_lock
);
1736 retval
= fmt
->load_binary(bprm
);
1738 read_lock(&binfmt_lock
);
1740 if (bprm
->point_of_no_return
|| (retval
!= -ENOEXEC
)) {
1741 read_unlock(&binfmt_lock
);
1745 read_unlock(&binfmt_lock
);
1748 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1749 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1751 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1760 /* binfmt handlers will call back into begin_new_exec() on success. */
1761 static int exec_binprm(struct linux_binprm
*bprm
)
1763 pid_t old_pid
, old_vpid
;
1766 /* Need to fetch pid before load_binary changes it */
1767 old_pid
= current
->pid
;
1769 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1772 /* This allows 4 levels of binfmt rewrites before failing hard. */
1773 for (depth
= 0;; depth
++) {
1778 ret
= search_binary_handler(bprm
);
1781 if (!bprm
->interpreter
)
1785 bprm
->file
= bprm
->interpreter
;
1786 bprm
->interpreter
= NULL
;
1788 allow_write_access(exec
);
1789 if (unlikely(bprm
->have_execfd
)) {
1790 if (bprm
->executable
) {
1794 bprm
->executable
= exec
;
1800 trace_sched_process_exec(current
, old_pid
, bprm
);
1801 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1802 proc_exec_connector(current
);
1807 * sys_execve() executes a new program.
1809 static int bprm_execve(struct linux_binprm
*bprm
,
1810 int fd
, struct filename
*filename
, int flags
)
1815 retval
= prepare_bprm_creds(bprm
);
1820 * Check for unsafe execution states before exec_binprm(), which
1821 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1822 * where setuid-ness is evaluated.
1824 check_unsafe_exec(bprm
);
1825 current
->in_execve
= 1;
1826 sched_mm_cid_before_execve(current
);
1828 file
= do_open_execat(fd
, filename
, flags
);
1829 retval
= PTR_ERR(file
);
1837 * Record that a name derived from an O_CLOEXEC fd will be
1838 * inaccessible after exec. This allows the code in exec to
1839 * choose to fail when the executable is not mmaped into the
1840 * interpreter and an open file descriptor is not passed to
1841 * the interpreter. This makes for a better user experience
1842 * than having the interpreter start and then immediately fail
1843 * when it finds the executable is inaccessible.
1845 if (bprm
->fdpath
&& get_close_on_exec(fd
))
1846 bprm
->interp_flags
|= BINPRM_FLAGS_PATH_INACCESSIBLE
;
1848 /* Set the unchanging part of bprm->cred */
1849 retval
= security_bprm_creds_for_exec(bprm
);
1853 retval
= exec_binprm(bprm
);
1857 sched_mm_cid_after_execve(current
);
1858 /* execve succeeded */
1859 current
->fs
->in_exec
= 0;
1860 current
->in_execve
= 0;
1861 rseq_execve(current
);
1862 acct_update_integrals(current
);
1863 task_numa_free(current
, false);
1868 * If past the point of no return ensure the code never
1869 * returns to the userspace process. Use an existing fatal
1870 * signal if present otherwise terminate the process with
1873 if (bprm
->point_of_no_return
&& !fatal_signal_pending(current
))
1874 force_fatal_sig(SIGSEGV
);
1877 sched_mm_cid_after_execve(current
);
1878 current
->fs
->in_exec
= 0;
1879 current
->in_execve
= 0;
1884 static int do_execveat_common(int fd
, struct filename
*filename
,
1885 struct user_arg_ptr argv
,
1886 struct user_arg_ptr envp
,
1889 struct linux_binprm
*bprm
;
1892 if (IS_ERR(filename
))
1893 return PTR_ERR(filename
);
1896 * We move the actual failure in case of RLIMIT_NPROC excess from
1897 * set*uid() to execve() because too many poorly written programs
1898 * don't check setuid() return code. Here we additionally recheck
1899 * whether NPROC limit is still exceeded.
1901 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1902 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC
, rlimit(RLIMIT_NPROC
))) {
1907 /* We're below the limit (still or again), so we don't want to make
1908 * further execve() calls fail. */
1909 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1911 bprm
= alloc_bprm(fd
, filename
);
1913 retval
= PTR_ERR(bprm
);
1917 retval
= count(argv
, MAX_ARG_STRINGS
);
1919 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1920 current
->comm
, bprm
->filename
);
1923 bprm
->argc
= retval
;
1925 retval
= count(envp
, MAX_ARG_STRINGS
);
1928 bprm
->envc
= retval
;
1930 retval
= bprm_stack_limits(bprm
);
1934 retval
= copy_string_kernel(bprm
->filename
, bprm
);
1937 bprm
->exec
= bprm
->p
;
1939 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1943 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1948 * When argv is empty, add an empty string ("") as argv[0] to
1949 * ensure confused userspace programs that start processing
1950 * from argv[1] won't end up walking envp. See also
1951 * bprm_stack_limits().
1953 if (bprm
->argc
== 0) {
1954 retval
= copy_string_kernel("", bprm
);
1960 retval
= bprm_execve(bprm
, fd
, filename
, flags
);
1969 int kernel_execve(const char *kernel_filename
,
1970 const char *const *argv
, const char *const *envp
)
1972 struct filename
*filename
;
1973 struct linux_binprm
*bprm
;
1977 /* It is non-sense for kernel threads to call execve */
1978 if (WARN_ON_ONCE(current
->flags
& PF_KTHREAD
))
1981 filename
= getname_kernel(kernel_filename
);
1982 if (IS_ERR(filename
))
1983 return PTR_ERR(filename
);
1985 bprm
= alloc_bprm(fd
, filename
);
1987 retval
= PTR_ERR(bprm
);
1991 retval
= count_strings_kernel(argv
);
1992 if (WARN_ON_ONCE(retval
== 0))
1996 bprm
->argc
= retval
;
1998 retval
= count_strings_kernel(envp
);
2001 bprm
->envc
= retval
;
2003 retval
= bprm_stack_limits(bprm
);
2007 retval
= copy_string_kernel(bprm
->filename
, bprm
);
2010 bprm
->exec
= bprm
->p
;
2012 retval
= copy_strings_kernel(bprm
->envc
, envp
, bprm
);
2016 retval
= copy_strings_kernel(bprm
->argc
, argv
, bprm
);
2020 retval
= bprm_execve(bprm
, fd
, filename
, 0);
2028 static int do_execve(struct filename
*filename
,
2029 const char __user
*const __user
*__argv
,
2030 const char __user
*const __user
*__envp
)
2032 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2033 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2034 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2037 static int do_execveat(int fd
, struct filename
*filename
,
2038 const char __user
*const __user
*__argv
,
2039 const char __user
*const __user
*__envp
,
2042 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
2043 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
2045 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2048 #ifdef CONFIG_COMPAT
2049 static int compat_do_execve(struct filename
*filename
,
2050 const compat_uptr_t __user
*__argv
,
2051 const compat_uptr_t __user
*__envp
)
2053 struct user_arg_ptr argv
= {
2055 .ptr
.compat
= __argv
,
2057 struct user_arg_ptr envp
= {
2059 .ptr
.compat
= __envp
,
2061 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
2064 static int compat_do_execveat(int fd
, struct filename
*filename
,
2065 const compat_uptr_t __user
*__argv
,
2066 const compat_uptr_t __user
*__envp
,
2069 struct user_arg_ptr argv
= {
2071 .ptr
.compat
= __argv
,
2073 struct user_arg_ptr envp
= {
2075 .ptr
.compat
= __envp
,
2077 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
2081 void set_binfmt(struct linux_binfmt
*new)
2083 struct mm_struct
*mm
= current
->mm
;
2086 module_put(mm
->binfmt
->module
);
2090 __module_get(new->module
);
2092 EXPORT_SYMBOL(set_binfmt
);
2095 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2097 void set_dumpable(struct mm_struct
*mm
, int value
)
2099 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
2102 set_mask_bits(&mm
->flags
, MMF_DUMPABLE_MASK
, value
);
2105 SYSCALL_DEFINE3(execve
,
2106 const char __user
*, filename
,
2107 const char __user
*const __user
*, argv
,
2108 const char __user
*const __user
*, envp
)
2110 return do_execve(getname(filename
), argv
, envp
);
2113 SYSCALL_DEFINE5(execveat
,
2114 int, fd
, const char __user
*, filename
,
2115 const char __user
*const __user
*, argv
,
2116 const char __user
*const __user
*, envp
,
2119 return do_execveat(fd
,
2120 getname_uflags(filename
, flags
),
2124 #ifdef CONFIG_COMPAT
2125 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
2126 const compat_uptr_t __user
*, argv
,
2127 const compat_uptr_t __user
*, envp
)
2129 return compat_do_execve(getname(filename
), argv
, envp
);
2132 COMPAT_SYSCALL_DEFINE5(execveat
, int, fd
,
2133 const char __user
*, filename
,
2134 const compat_uptr_t __user
*, argv
,
2135 const compat_uptr_t __user
*, envp
,
2138 return compat_do_execveat(fd
,
2139 getname_uflags(filename
, flags
),
2144 #ifdef CONFIG_SYSCTL
2146 static int proc_dointvec_minmax_coredump(struct ctl_table
*table
, int write
,
2147 void *buffer
, size_t *lenp
, loff_t
*ppos
)
2149 int error
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
2152 validate_coredump_safety();
2156 static struct ctl_table fs_exec_sysctls
[] = {
2158 .procname
= "suid_dumpable",
2159 .data
= &suid_dumpable
,
2160 .maxlen
= sizeof(int),
2162 .proc_handler
= proc_dointvec_minmax_coredump
,
2163 .extra1
= SYSCTL_ZERO
,
2164 .extra2
= SYSCTL_TWO
,
2169 static int __init
init_fs_exec_sysctls(void)
2171 register_sysctl_init("fs", fs_exec_sysctls
);
2175 fs_initcall(init_fs_exec_sysctls
);
2176 #endif /* CONFIG_SYSCTL */