4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/vmacache.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/perf_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55 #include <linux/fs_struct.h>
56 #include <linux/pipe_fs_i.h>
57 #include <linux/oom.h>
58 #include <linux/compat.h>
60 #include <trace/events/fs.h>
62 #include <asm/uaccess.h>
63 #include <asm/mmu_context.h>
66 #include <trace/events/task.h>
69 #include <trace/events/sched.h>
71 int suid_dumpable
= 0;
73 static LIST_HEAD(formats
);
74 static DEFINE_RWLOCK(binfmt_lock
);
76 void __register_binfmt(struct linux_binfmt
* fmt
, int insert
)
79 if (WARN_ON(!fmt
->load_binary
))
81 write_lock(&binfmt_lock
);
82 insert
? list_add(&fmt
->lh
, &formats
) :
83 list_add_tail(&fmt
->lh
, &formats
);
84 write_unlock(&binfmt_lock
);
87 EXPORT_SYMBOL(__register_binfmt
);
89 void unregister_binfmt(struct linux_binfmt
* fmt
)
91 write_lock(&binfmt_lock
);
93 write_unlock(&binfmt_lock
);
96 EXPORT_SYMBOL(unregister_binfmt
);
98 static inline void put_binfmt(struct linux_binfmt
* fmt
)
100 module_put(fmt
->module
);
103 bool path_noexec(const struct path
*path
)
105 return (path
->mnt
->mnt_flags
& MNT_NOEXEC
) ||
106 (path
->mnt
->mnt_sb
->s_iflags
& SB_I_NOEXEC
);
108 EXPORT_SYMBOL(path_noexec
);
110 bool path_nosuid(const struct path
*path
)
112 return !mnt_may_suid(path
->mnt
) ||
113 (path
->mnt
->mnt_sb
->s_iflags
& SB_I_NOSUID
);
115 EXPORT_SYMBOL(path_nosuid
);
119 * Note that a shared library must be both readable and executable due to
122 * Also note that we take the address to load from from the file itself.
124 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
126 struct linux_binfmt
*fmt
;
128 struct filename
*tmp
= getname(library
);
129 int error
= PTR_ERR(tmp
);
130 static const struct open_flags uselib_flags
= {
131 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
132 .acc_mode
= MAY_READ
| MAY_EXEC
| MAY_OPEN
,
133 .intent
= LOOKUP_OPEN
,
134 .lookup_flags
= LOOKUP_FOLLOW
,
140 file
= do_filp_open(AT_FDCWD
, tmp
, &uselib_flags
);
142 error
= PTR_ERR(file
);
147 if (!S_ISREG(file_inode(file
)->i_mode
))
151 if (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
,
204 #ifdef CONFIG_STACK_GROWSUP
206 ret
= expand_downwards(bprm
->vma
, pos
);
211 ret
= get_user_pages(current
, bprm
->mm
, pos
,
212 1, write
, 1, &page
, NULL
);
217 unsigned long size
= bprm
->vma
->vm_end
- bprm
->vma
->vm_start
;
220 acct_arg_size(bprm
, size
/ PAGE_SIZE
);
223 * We've historically supported up to 32 pages (ARG_MAX)
224 * of argument strings even with small stacks
230 * Limit to 1/4-th the stack size for the argv+env strings.
232 * - the remaining binfmt code will not run out of stack space,
233 * - the program will have a reasonable amount of stack left
236 rlim
= current
->signal
->rlim
;
237 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
) / 4) {
246 static void put_arg_page(struct page
*page
)
251 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
255 static void free_arg_pages(struct linux_binprm
*bprm
)
259 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
262 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
265 static int __bprm_mm_init(struct linux_binprm
*bprm
)
268 struct vm_area_struct
*vma
= NULL
;
269 struct mm_struct
*mm
= bprm
->mm
;
271 bprm
->vma
= vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
275 down_write(&mm
->mmap_sem
);
279 * Place the stack at the largest stack address the architecture
280 * supports. Later, we'll move this to an appropriate place. We don't
281 * use STACK_TOP because that can depend on attributes which aren't
284 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
285 vma
->vm_end
= STACK_TOP_MAX
;
286 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
287 vma
->vm_flags
= VM_SOFTDIRTY
| VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
;
288 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
289 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
291 err
= insert_vm_struct(mm
, vma
);
295 mm
->stack_vm
= mm
->total_vm
= 1;
296 arch_bprm_mm_init(mm
, vma
);
297 up_write(&mm
->mmap_sem
);
298 bprm
->p
= vma
->vm_end
- sizeof(void *);
301 up_write(&mm
->mmap_sem
);
303 kmem_cache_free(vm_area_cachep
, vma
);
307 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
309 return len
<= MAX_ARG_STRLEN
;
314 static inline void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
318 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
323 page
= bprm
->page
[pos
/ PAGE_SIZE
];
324 if (!page
&& write
) {
325 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
328 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
334 static void put_arg_page(struct page
*page
)
338 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
341 __free_page(bprm
->page
[i
]);
342 bprm
->page
[i
] = NULL
;
346 static void free_arg_pages(struct linux_binprm
*bprm
)
350 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
351 free_arg_page(bprm
, i
);
354 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
359 static int __bprm_mm_init(struct linux_binprm
*bprm
)
361 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
365 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
367 return len
<= bprm
->p
;
370 #endif /* CONFIG_MMU */
373 * Create a new mm_struct and populate it with a temporary stack
374 * vm_area_struct. We don't have enough context at this point to set the stack
375 * flags, permissions, and offset, so we use temporary values. We'll update
376 * them later in setup_arg_pages().
378 static int bprm_mm_init(struct linux_binprm
*bprm
)
381 struct mm_struct
*mm
= NULL
;
383 bprm
->mm
= mm
= mm_alloc();
388 err
= __bprm_mm_init(bprm
);
403 struct user_arg_ptr
{
408 const char __user
*const __user
*native
;
410 const compat_uptr_t __user
*compat
;
415 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
417 const char __user
*native
;
420 if (unlikely(argv
.is_compat
)) {
421 compat_uptr_t compat
;
423 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
424 return ERR_PTR(-EFAULT
);
426 return compat_ptr(compat
);
430 if (get_user(native
, argv
.ptr
.native
+ nr
))
431 return ERR_PTR(-EFAULT
);
437 * count() counts the number of strings in array ARGV.
439 static int count(struct user_arg_ptr argv
, int max
)
443 if (argv
.ptr
.native
!= NULL
) {
445 const char __user
*p
= get_user_arg_ptr(argv
, i
);
457 if (fatal_signal_pending(current
))
458 return -ERESTARTNOHAND
;
466 * 'copy_strings()' copies argument/environment strings from the old
467 * processes's memory to the new process's stack. The call to get_user_pages()
468 * ensures the destination page is created and not swapped out.
470 static int copy_strings(int argc
, struct user_arg_ptr argv
,
471 struct linux_binprm
*bprm
)
473 struct page
*kmapped_page
= NULL
;
475 unsigned long kpos
= 0;
479 const char __user
*str
;
484 str
= get_user_arg_ptr(argv
, argc
);
488 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
493 if (!valid_arg_len(bprm
, len
))
496 /* We're going to work our way backwords. */
502 int offset
, bytes_to_copy
;
504 if (fatal_signal_pending(current
)) {
505 ret
= -ERESTARTNOHAND
;
510 offset
= pos
% PAGE_SIZE
;
514 bytes_to_copy
= offset
;
515 if (bytes_to_copy
> len
)
518 offset
-= bytes_to_copy
;
519 pos
-= bytes_to_copy
;
520 str
-= bytes_to_copy
;
521 len
-= bytes_to_copy
;
523 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
526 page
= get_arg_page(bprm
, pos
, 1);
533 flush_kernel_dcache_page(kmapped_page
);
534 kunmap(kmapped_page
);
535 put_arg_page(kmapped_page
);
538 kaddr
= kmap(kmapped_page
);
539 kpos
= pos
& PAGE_MASK
;
540 flush_arg_page(bprm
, kpos
, kmapped_page
);
542 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
551 flush_kernel_dcache_page(kmapped_page
);
552 kunmap(kmapped_page
);
553 put_arg_page(kmapped_page
);
559 * Like copy_strings, but get argv and its values from kernel memory.
561 int copy_strings_kernel(int argc
, const char *const *__argv
,
562 struct linux_binprm
*bprm
)
565 mm_segment_t oldfs
= get_fs();
566 struct user_arg_ptr argv
= {
567 .ptr
.native
= (const char __user
*const __user
*)__argv
,
571 r
= copy_strings(argc
, argv
, bprm
);
576 EXPORT_SYMBOL(copy_strings_kernel
);
581 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
582 * the binfmt code determines where the new stack should reside, we shift it to
583 * its final location. The process proceeds as follows:
585 * 1) Use shift to calculate the new vma endpoints.
586 * 2) Extend vma to cover both the old and new ranges. This ensures the
587 * arguments passed to subsequent functions are consistent.
588 * 3) Move vma's page tables to the new range.
589 * 4) Free up any cleared pgd range.
590 * 5) Shrink the vma to cover only the new range.
592 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
594 struct mm_struct
*mm
= vma
->vm_mm
;
595 unsigned long old_start
= vma
->vm_start
;
596 unsigned long old_end
= vma
->vm_end
;
597 unsigned long length
= old_end
- old_start
;
598 unsigned long new_start
= old_start
- shift
;
599 unsigned long new_end
= old_end
- shift
;
600 struct mmu_gather tlb
;
602 BUG_ON(new_start
> new_end
);
605 * ensure there are no vmas between where we want to go
608 if (vma
!= find_vma(mm
, new_start
))
612 * cover the whole range: [new_start, old_end)
614 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
618 * move the page tables downwards, on failure we rely on
619 * process cleanup to remove whatever mess we made.
621 if (length
!= move_page_tables(vma
, old_start
,
622 vma
, new_start
, length
, false))
626 tlb_gather_mmu(&tlb
, mm
, old_start
, old_end
);
627 if (new_end
> old_start
) {
629 * when the old and new regions overlap clear from new_end.
631 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
632 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
635 * otherwise, clean from old_start; this is done to not touch
636 * the address space in [new_end, old_start) some architectures
637 * have constraints on va-space that make this illegal (IA64) -
638 * for the others its just a little faster.
640 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
641 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
643 tlb_finish_mmu(&tlb
, old_start
, old_end
);
646 * Shrink the vma to just the new range. Always succeeds.
648 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
654 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
655 * the stack is optionally relocated, and some extra space is added.
657 int setup_arg_pages(struct linux_binprm
*bprm
,
658 unsigned long stack_top
,
659 int executable_stack
)
662 unsigned long stack_shift
;
663 struct mm_struct
*mm
= current
->mm
;
664 struct vm_area_struct
*vma
= bprm
->vma
;
665 struct vm_area_struct
*prev
= NULL
;
666 unsigned long vm_flags
;
667 unsigned long stack_base
;
668 unsigned long stack_size
;
669 unsigned long stack_expand
;
670 unsigned long rlim_stack
;
672 #ifdef CONFIG_STACK_GROWSUP
673 /* Limit stack size */
674 stack_base
= rlimit_max(RLIMIT_STACK
);
675 if (stack_base
> STACK_SIZE_MAX
)
676 stack_base
= STACK_SIZE_MAX
;
678 /* Add space for stack randomization. */
679 stack_base
+= (STACK_RND_MASK
<< PAGE_SHIFT
);
681 /* Make sure we didn't let the argument array grow too large. */
682 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
685 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
687 stack_shift
= vma
->vm_start
- stack_base
;
688 mm
->arg_start
= bprm
->p
- stack_shift
;
689 bprm
->p
= vma
->vm_end
- stack_shift
;
691 stack_top
= arch_align_stack(stack_top
);
692 stack_top
= PAGE_ALIGN(stack_top
);
694 if (unlikely(stack_top
< mmap_min_addr
) ||
695 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
698 stack_shift
= vma
->vm_end
- stack_top
;
700 bprm
->p
-= stack_shift
;
701 mm
->arg_start
= bprm
->p
;
705 bprm
->loader
-= stack_shift
;
706 bprm
->exec
-= stack_shift
;
708 down_write(&mm
->mmap_sem
);
709 vm_flags
= VM_STACK_FLAGS
;
712 * Adjust stack execute permissions; explicitly enable for
713 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
714 * (arch default) otherwise.
716 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
718 else if (executable_stack
== EXSTACK_DISABLE_X
)
719 vm_flags
&= ~VM_EXEC
;
720 vm_flags
|= mm
->def_flags
;
721 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
723 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
729 /* Move stack pages down in memory. */
731 ret
= shift_arg_pages(vma
, stack_shift
);
736 /* mprotect_fixup is overkill to remove the temporary stack flags */
737 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
739 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
740 stack_size
= vma
->vm_end
- vma
->vm_start
;
742 * Align this down to a page boundary as expand_stack
745 rlim_stack
= rlimit(RLIMIT_STACK
) & PAGE_MASK
;
746 #ifdef CONFIG_STACK_GROWSUP
747 if (stack_size
+ stack_expand
> rlim_stack
)
748 stack_base
= vma
->vm_start
+ rlim_stack
;
750 stack_base
= vma
->vm_end
+ stack_expand
;
752 if (stack_size
+ stack_expand
> rlim_stack
)
753 stack_base
= vma
->vm_end
- rlim_stack
;
755 stack_base
= vma
->vm_start
- stack_expand
;
757 current
->mm
->start_stack
= bprm
->p
;
758 ret
= expand_stack(vma
, stack_base
);
763 up_write(&mm
->mmap_sem
);
766 EXPORT_SYMBOL(setup_arg_pages
);
768 #endif /* CONFIG_MMU */
770 static struct file
*do_open_execat(int fd
, struct filename
*name
, int flags
)
774 struct open_flags open_exec_flags
= {
775 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
776 .acc_mode
= MAY_EXEC
| MAY_OPEN
,
777 .intent
= LOOKUP_OPEN
,
778 .lookup_flags
= LOOKUP_FOLLOW
,
781 if ((flags
& ~(AT_SYMLINK_NOFOLLOW
| AT_EMPTY_PATH
)) != 0)
782 return ERR_PTR(-EINVAL
);
783 if (flags
& AT_SYMLINK_NOFOLLOW
)
784 open_exec_flags
.lookup_flags
&= ~LOOKUP_FOLLOW
;
785 if (flags
& AT_EMPTY_PATH
)
786 open_exec_flags
.lookup_flags
|= LOOKUP_EMPTY
;
788 file
= do_filp_open(fd
, name
, &open_exec_flags
);
793 if (!S_ISREG(file_inode(file
)->i_mode
))
796 if (path_noexec(&file
->f_path
))
799 err
= deny_write_access(file
);
803 if (name
->name
[0] != '\0')
806 trace_open_exec(name
->name
);
816 struct file
*open_exec(const char *name
)
818 struct filename
*filename
= getname_kernel(name
);
819 struct file
*f
= ERR_CAST(filename
);
821 if (!IS_ERR(filename
)) {
822 f
= do_open_execat(AT_FDCWD
, filename
, 0);
827 EXPORT_SYMBOL(open_exec
);
829 int kernel_read(struct file
*file
, loff_t offset
,
830 char *addr
, unsigned long count
)
838 /* The cast to a user pointer is valid due to the set_fs() */
839 result
= vfs_read(file
, (void __user
*)addr
, count
, &pos
);
844 EXPORT_SYMBOL(kernel_read
);
846 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
848 ssize_t res
= vfs_read(file
, (void __user
*)addr
, len
, &pos
);
850 flush_icache_range(addr
, addr
+ len
);
853 EXPORT_SYMBOL(read_code
);
855 static int exec_mmap(struct mm_struct
*mm
)
857 struct task_struct
*tsk
;
858 struct mm_struct
*old_mm
, *active_mm
;
860 /* Notify parent that we're no longer interested in the old VM */
862 old_mm
= current
->mm
;
863 mm_release(tsk
, old_mm
);
868 * Make sure that if there is a core dump in progress
869 * for the old mm, we get out and die instead of going
870 * through with the exec. We must hold mmap_sem around
871 * checking core_state and changing tsk->mm.
873 down_read(&old_mm
->mmap_sem
);
874 if (unlikely(old_mm
->core_state
)) {
875 up_read(&old_mm
->mmap_sem
);
880 active_mm
= tsk
->active_mm
;
883 activate_mm(active_mm
, mm
);
884 tsk
->mm
->vmacache_seqnum
= 0;
888 up_read(&old_mm
->mmap_sem
);
889 BUG_ON(active_mm
!= old_mm
);
890 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
891 mm_update_next_owner(old_mm
);
900 * This function makes sure the current process has its own signal table,
901 * so that flush_signal_handlers can later reset the handlers without
902 * disturbing other processes. (Other processes might share the signal
903 * table via the CLONE_SIGHAND option to clone().)
905 static int de_thread(struct task_struct
*tsk
)
907 struct signal_struct
*sig
= tsk
->signal
;
908 struct sighand_struct
*oldsighand
= tsk
->sighand
;
909 spinlock_t
*lock
= &oldsighand
->siglock
;
911 if (thread_group_empty(tsk
))
912 goto no_thread_group
;
915 * Kill all other threads in the thread group.
918 if (signal_group_exit(sig
)) {
920 * Another group action in progress, just
921 * return so that the signal is processed.
923 spin_unlock_irq(lock
);
927 sig
->group_exit_task
= tsk
;
928 sig
->notify_count
= zap_other_threads(tsk
);
929 if (!thread_group_leader(tsk
))
932 while (sig
->notify_count
) {
933 __set_current_state(TASK_KILLABLE
);
934 spin_unlock_irq(lock
);
936 if (unlikely(__fatal_signal_pending(tsk
)))
940 spin_unlock_irq(lock
);
943 * At this point all other threads have exited, all we have to
944 * do is to wait for the thread group leader to become inactive,
945 * and to assume its PID:
947 if (!thread_group_leader(tsk
)) {
948 struct task_struct
*leader
= tsk
->group_leader
;
951 threadgroup_change_begin(tsk
);
952 write_lock_irq(&tasklist_lock
);
954 * Do this under tasklist_lock to ensure that
955 * exit_notify() can't miss ->group_exit_task
957 sig
->notify_count
= -1;
958 if (likely(leader
->exit_state
))
960 __set_current_state(TASK_KILLABLE
);
961 write_unlock_irq(&tasklist_lock
);
962 threadgroup_change_end(tsk
);
964 if (unlikely(__fatal_signal_pending(tsk
)))
969 * The only record we have of the real-time age of a
970 * process, regardless of execs it's done, is start_time.
971 * All the past CPU time is accumulated in signal_struct
972 * from sister threads now dead. But in this non-leader
973 * exec, nothing survives from the original leader thread,
974 * whose birth marks the true age of this process now.
975 * When we take on its identity by switching to its PID, we
976 * also take its birthdate (always earlier than our own).
978 tsk
->start_time
= leader
->start_time
;
979 tsk
->real_start_time
= leader
->real_start_time
;
981 BUG_ON(!same_thread_group(leader
, tsk
));
982 BUG_ON(has_group_leader_pid(tsk
));
984 * An exec() starts a new thread group with the
985 * TGID of the previous thread group. Rehash the
986 * two threads with a switched PID, and release
987 * the former thread group leader:
990 /* Become a process group leader with the old leader's pid.
991 * The old leader becomes a thread of the this thread group.
992 * Note: The old leader also uses this pid until release_task
993 * is called. Odd but simple and correct.
995 tsk
->pid
= leader
->pid
;
996 change_pid(tsk
, PIDTYPE_PID
, task_pid(leader
));
997 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
998 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
1000 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
1001 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
1003 tsk
->group_leader
= tsk
;
1004 leader
->group_leader
= tsk
;
1006 tsk
->exit_signal
= SIGCHLD
;
1007 leader
->exit_signal
= -1;
1009 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
1010 leader
->exit_state
= EXIT_DEAD
;
1013 * We are going to release_task()->ptrace_unlink() silently,
1014 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1015 * the tracer wont't block again waiting for this thread.
1017 if (unlikely(leader
->ptrace
))
1018 __wake_up_parent(leader
, leader
->parent
);
1019 write_unlock_irq(&tasklist_lock
);
1020 threadgroup_change_end(tsk
);
1022 release_task(leader
);
1025 sig
->group_exit_task
= NULL
;
1026 sig
->notify_count
= 0;
1029 /* we have changed execution domain */
1030 tsk
->exit_signal
= SIGCHLD
;
1033 flush_itimer_signals();
1035 if (atomic_read(&oldsighand
->count
) != 1) {
1036 struct sighand_struct
*newsighand
;
1038 * This ->sighand is shared with the CLONE_SIGHAND
1039 * but not CLONE_THREAD task, switch to the new one.
1041 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1045 atomic_set(&newsighand
->count
, 1);
1046 memcpy(newsighand
->action
, oldsighand
->action
,
1047 sizeof(newsighand
->action
));
1049 write_lock_irq(&tasklist_lock
);
1050 spin_lock(&oldsighand
->siglock
);
1051 rcu_assign_pointer(tsk
->sighand
, newsighand
);
1052 spin_unlock(&oldsighand
->siglock
);
1053 write_unlock_irq(&tasklist_lock
);
1055 __cleanup_sighand(oldsighand
);
1058 BUG_ON(!thread_group_leader(tsk
));
1062 /* protects against exit_notify() and __exit_signal() */
1063 read_lock(&tasklist_lock
);
1064 sig
->group_exit_task
= NULL
;
1065 sig
->notify_count
= 0;
1066 read_unlock(&tasklist_lock
);
1070 char *get_task_comm(char *buf
, struct task_struct
*tsk
)
1072 /* buf must be at least sizeof(tsk->comm) in size */
1074 strncpy(buf
, tsk
->comm
, sizeof(tsk
->comm
));
1078 EXPORT_SYMBOL_GPL(get_task_comm
);
1081 * These functions flushes out all traces of the currently running executable
1082 * so that a new one can be started
1085 void __set_task_comm(struct task_struct
*tsk
, const char *buf
, bool exec
)
1088 trace_task_rename(tsk
, buf
);
1089 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1091 perf_event_comm(tsk
, exec
);
1094 int flush_old_exec(struct linux_binprm
* bprm
)
1099 * Make sure we have a private signal table and that
1100 * we are unassociated from the previous thread group.
1102 retval
= de_thread(current
);
1107 * Must be called _before_ exec_mmap() as bprm->mm is
1108 * not visibile until then. This also enables the update
1111 set_mm_exe_file(bprm
->mm
, bprm
->file
);
1114 * Release all of the old mmap stuff
1116 acct_arg_size(bprm
, 0);
1117 retval
= exec_mmap(bprm
->mm
);
1121 bprm
->mm
= NULL
; /* We're using it now */
1124 current
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
| PF_KTHREAD
|
1125 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1127 current
->personality
&= ~bprm
->per_clear
;
1134 EXPORT_SYMBOL(flush_old_exec
);
1136 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1138 if (inode_permission(file_inode(file
), MAY_READ
) < 0)
1139 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1141 EXPORT_SYMBOL(would_dump
);
1143 void setup_new_exec(struct linux_binprm
* bprm
)
1145 arch_pick_mmap_layout(current
->mm
);
1147 /* This is the point of no return */
1148 current
->sas_ss_sp
= current
->sas_ss_size
= 0;
1150 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1151 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1153 set_dumpable(current
->mm
, suid_dumpable
);
1156 __set_task_comm(current
, kbasename(bprm
->filename
), true);
1158 /* Set the new mm task size. We have to do that late because it may
1159 * depend on TIF_32BIT which is only updated in flush_thread() on
1160 * some architectures like powerpc
1162 current
->mm
->task_size
= TASK_SIZE
;
1164 /* install the new credentials */
1165 if (!uid_eq(bprm
->cred
->uid
, current_euid()) ||
1166 !gid_eq(bprm
->cred
->gid
, current_egid())) {
1167 current
->pdeath_signal
= 0;
1169 would_dump(bprm
, bprm
->file
);
1170 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
)
1171 set_dumpable(current
->mm
, suid_dumpable
);
1174 /* An exec changes our domain. We are no longer part of the thread
1176 current
->self_exec_id
++;
1177 flush_signal_handlers(current
, 0);
1178 do_close_on_exec(current
->files
);
1180 EXPORT_SYMBOL(setup_new_exec
);
1183 * Prepare credentials and lock ->cred_guard_mutex.
1184 * install_exec_creds() commits the new creds and drops the lock.
1185 * Or, if exec fails before, free_bprm() should release ->cred and
1188 int prepare_bprm_creds(struct linux_binprm
*bprm
)
1190 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1191 return -ERESTARTNOINTR
;
1193 bprm
->cred
= prepare_exec_creds();
1194 if (likely(bprm
->cred
))
1197 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1201 static void free_bprm(struct linux_binprm
*bprm
)
1203 free_arg_pages(bprm
);
1205 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1206 abort_creds(bprm
->cred
);
1209 allow_write_access(bprm
->file
);
1212 /* If a binfmt changed the interp, free it. */
1213 if (bprm
->interp
!= bprm
->filename
)
1214 kfree(bprm
->interp
);
1218 int bprm_change_interp(char *interp
, struct linux_binprm
*bprm
)
1220 /* If a binfmt changed the interp, free it first. */
1221 if (bprm
->interp
!= bprm
->filename
)
1222 kfree(bprm
->interp
);
1223 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1228 EXPORT_SYMBOL(bprm_change_interp
);
1231 * install the new credentials for this executable
1233 void install_exec_creds(struct linux_binprm
*bprm
)
1235 security_bprm_committing_creds(bprm
);
1237 commit_creds(bprm
->cred
);
1241 * Disable monitoring for regular users
1242 * when executing setuid binaries. Must
1243 * wait until new credentials are committed
1244 * by commit_creds() above
1246 if (get_dumpable(current
->mm
) != SUID_DUMP_USER
)
1247 perf_event_exit_task(current
);
1249 * cred_guard_mutex must be held at least to this point to prevent
1250 * ptrace_attach() from altering our determination of the task's
1251 * credentials; any time after this it may be unlocked.
1253 security_bprm_committed_creds(bprm
);
1254 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1256 EXPORT_SYMBOL(install_exec_creds
);
1259 * determine how safe it is to execute the proposed program
1260 * - the caller must hold ->cred_guard_mutex to protect against
1261 * PTRACE_ATTACH or seccomp thread-sync
1263 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1265 struct task_struct
*p
= current
, *t
;
1269 if (p
->ptrace
& PT_PTRACE_CAP
)
1270 bprm
->unsafe
|= LSM_UNSAFE_PTRACE_CAP
;
1272 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1276 * This isn't strictly necessary, but it makes it harder for LSMs to
1279 if (task_no_new_privs(current
))
1280 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1284 spin_lock(&p
->fs
->lock
);
1286 while_each_thread(p
, t
) {
1292 if (p
->fs
->users
> n_fs
)
1293 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1296 spin_unlock(&p
->fs
->lock
);
1299 static void bprm_fill_uid(struct linux_binprm
*bprm
)
1301 struct inode
*inode
;
1306 /* clear any previous set[ug]id data from a previous binary */
1307 bprm
->cred
->euid
= current_euid();
1308 bprm
->cred
->egid
= current_egid();
1310 if (path_nosuid(&bprm
->file
->f_path
))
1313 if (task_no_new_privs(current
))
1316 inode
= file_inode(bprm
->file
);
1317 mode
= READ_ONCE(inode
->i_mode
);
1318 if (!(mode
& (S_ISUID
|S_ISGID
)))
1321 /* Be careful if suid/sgid is set */
1322 mutex_lock(&inode
->i_mutex
);
1324 /* reload atomically mode/uid/gid now that lock held */
1325 mode
= inode
->i_mode
;
1328 mutex_unlock(&inode
->i_mutex
);
1330 /* We ignore suid/sgid if there are no mappings for them in the ns */
1331 if (!kuid_has_mapping(bprm
->cred
->user_ns
, uid
) ||
1332 !kgid_has_mapping(bprm
->cred
->user_ns
, gid
))
1335 if (mode
& S_ISUID
) {
1336 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1337 bprm
->cred
->euid
= uid
;
1340 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1341 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1342 bprm
->cred
->egid
= gid
;
1347 * Fill the binprm structure from the inode.
1348 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1350 * This may be called multiple times for binary chains (scripts for example).
1352 int prepare_binprm(struct linux_binprm
*bprm
)
1356 bprm_fill_uid(bprm
);
1358 /* fill in binprm security blob */
1359 retval
= security_bprm_set_creds(bprm
);
1362 bprm
->cred_prepared
= 1;
1364 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1365 return kernel_read(bprm
->file
, 0, bprm
->buf
, BINPRM_BUF_SIZE
);
1368 EXPORT_SYMBOL(prepare_binprm
);
1371 * Arguments are '\0' separated strings found at the location bprm->p
1372 * points to; chop off the first by relocating brpm->p to right after
1373 * the first '\0' encountered.
1375 int remove_arg_zero(struct linux_binprm
*bprm
)
1378 unsigned long offset
;
1386 offset
= bprm
->p
& ~PAGE_MASK
;
1387 page
= get_arg_page(bprm
, bprm
->p
, 0);
1392 kaddr
= kmap_atomic(page
);
1394 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1395 offset
++, bprm
->p
++)
1398 kunmap_atomic(kaddr
);
1401 if (offset
== PAGE_SIZE
)
1402 free_arg_page(bprm
, (bprm
->p
>> PAGE_SHIFT
) - 1);
1403 } while (offset
== PAGE_SIZE
);
1412 EXPORT_SYMBOL(remove_arg_zero
);
1414 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1416 * cycle the list of binary formats handler, until one recognizes the image
1418 int search_binary_handler(struct linux_binprm
*bprm
)
1420 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1421 struct linux_binfmt
*fmt
;
1424 /* This allows 4 levels of binfmt rewrites before failing hard. */
1425 if (bprm
->recursion_depth
> 5)
1428 retval
= security_bprm_check(bprm
);
1434 read_lock(&binfmt_lock
);
1435 list_for_each_entry(fmt
, &formats
, lh
) {
1436 if (!try_module_get(fmt
->module
))
1438 read_unlock(&binfmt_lock
);
1439 bprm
->recursion_depth
++;
1440 retval
= fmt
->load_binary(bprm
);
1441 read_lock(&binfmt_lock
);
1443 bprm
->recursion_depth
--;
1444 if (retval
< 0 && !bprm
->mm
) {
1445 /* we got to flush_old_exec() and failed after it */
1446 read_unlock(&binfmt_lock
);
1447 force_sigsegv(SIGSEGV
, current
);
1450 if (retval
!= -ENOEXEC
|| !bprm
->file
) {
1451 read_unlock(&binfmt_lock
);
1455 read_unlock(&binfmt_lock
);
1458 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1459 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1461 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1469 EXPORT_SYMBOL(search_binary_handler
);
1471 static int exec_binprm(struct linux_binprm
*bprm
)
1473 pid_t old_pid
, old_vpid
;
1476 /* Need to fetch pid before load_binary changes it */
1477 old_pid
= current
->pid
;
1479 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1482 ret
= search_binary_handler(bprm
);
1485 trace_sched_process_exec(current
, old_pid
, bprm
);
1486 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1487 proc_exec_connector(current
);
1494 * sys_execve() executes a new program.
1496 static int do_execveat_common(int fd
, struct filename
*filename
,
1497 struct user_arg_ptr argv
,
1498 struct user_arg_ptr envp
,
1501 char *pathbuf
= NULL
;
1502 struct linux_binprm
*bprm
;
1504 struct files_struct
*displaced
;
1507 if (IS_ERR(filename
))
1508 return PTR_ERR(filename
);
1511 * We move the actual failure in case of RLIMIT_NPROC excess from
1512 * set*uid() to execve() because too many poorly written programs
1513 * don't check setuid() return code. Here we additionally recheck
1514 * whether NPROC limit is still exceeded.
1516 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1517 atomic_read(¤t_user()->processes
) > rlimit(RLIMIT_NPROC
)) {
1522 /* We're below the limit (still or again), so we don't want to make
1523 * further execve() calls fail. */
1524 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1526 retval
= unshare_files(&displaced
);
1531 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1535 retval
= prepare_bprm_creds(bprm
);
1539 check_unsafe_exec(bprm
);
1540 current
->in_execve
= 1;
1542 file
= do_open_execat(fd
, filename
, flags
);
1543 retval
= PTR_ERR(file
);
1550 if (fd
== AT_FDCWD
|| filename
->name
[0] == '/') {
1551 bprm
->filename
= filename
->name
;
1553 if (filename
->name
[0] == '\0')
1554 pathbuf
= kasprintf(GFP_TEMPORARY
, "/dev/fd/%d", fd
);
1556 pathbuf
= kasprintf(GFP_TEMPORARY
, "/dev/fd/%d/%s",
1557 fd
, filename
->name
);
1563 * Record that a name derived from an O_CLOEXEC fd will be
1564 * inaccessible after exec. Relies on having exclusive access to
1565 * current->files (due to unshare_files above).
1567 if (close_on_exec(fd
, rcu_dereference_raw(current
->files
->fdt
)))
1568 bprm
->interp_flags
|= BINPRM_FLAGS_PATH_INACCESSIBLE
;
1569 bprm
->filename
= pathbuf
;
1571 bprm
->interp
= bprm
->filename
;
1573 retval
= bprm_mm_init(bprm
);
1577 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
1578 if ((retval
= bprm
->argc
) < 0)
1581 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
1582 if ((retval
= bprm
->envc
) < 0)
1585 retval
= prepare_binprm(bprm
);
1589 retval
= copy_strings_kernel(1, &bprm
->filename
, bprm
);
1593 bprm
->exec
= bprm
->p
;
1594 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1598 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1602 retval
= exec_binprm(bprm
);
1606 /* execve succeeded */
1607 current
->fs
->in_exec
= 0;
1608 current
->in_execve
= 0;
1609 acct_update_integrals(current
);
1610 task_numa_free(current
);
1615 put_files_struct(displaced
);
1620 acct_arg_size(bprm
, 0);
1625 current
->fs
->in_exec
= 0;
1626 current
->in_execve
= 0;
1634 reset_files_struct(displaced
);
1640 int do_execve(struct filename
*filename
,
1641 const char __user
*const __user
*__argv
,
1642 const char __user
*const __user
*__envp
)
1644 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1645 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1646 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
1649 int do_execveat(int fd
, struct filename
*filename
,
1650 const char __user
*const __user
*__argv
,
1651 const char __user
*const __user
*__envp
,
1654 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1655 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1657 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
1660 #ifdef CONFIG_COMPAT
1661 static int compat_do_execve(struct filename
*filename
,
1662 const compat_uptr_t __user
*__argv
,
1663 const compat_uptr_t __user
*__envp
)
1665 struct user_arg_ptr argv
= {
1667 .ptr
.compat
= __argv
,
1669 struct user_arg_ptr envp
= {
1671 .ptr
.compat
= __envp
,
1673 return do_execveat_common(AT_FDCWD
, filename
, argv
, envp
, 0);
1676 static int compat_do_execveat(int fd
, struct filename
*filename
,
1677 const compat_uptr_t __user
*__argv
,
1678 const compat_uptr_t __user
*__envp
,
1681 struct user_arg_ptr argv
= {
1683 .ptr
.compat
= __argv
,
1685 struct user_arg_ptr envp
= {
1687 .ptr
.compat
= __envp
,
1689 return do_execveat_common(fd
, filename
, argv
, envp
, flags
);
1693 void set_binfmt(struct linux_binfmt
*new)
1695 struct mm_struct
*mm
= current
->mm
;
1698 module_put(mm
->binfmt
->module
);
1702 __module_get(new->module
);
1704 EXPORT_SYMBOL(set_binfmt
);
1707 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1709 void set_dumpable(struct mm_struct
*mm
, int value
)
1711 unsigned long old
, new;
1713 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
1717 old
= ACCESS_ONCE(mm
->flags
);
1718 new = (old
& ~MMF_DUMPABLE_MASK
) | value
;
1719 } while (cmpxchg(&mm
->flags
, old
, new) != old
);
1722 SYSCALL_DEFINE3(execve
,
1723 const char __user
*, filename
,
1724 const char __user
*const __user
*, argv
,
1725 const char __user
*const __user
*, envp
)
1727 return do_execve(getname(filename
), argv
, envp
);
1730 SYSCALL_DEFINE5(execveat
,
1731 int, fd
, const char __user
*, filename
,
1732 const char __user
*const __user
*, argv
,
1733 const char __user
*const __user
*, envp
,
1736 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
1738 return do_execveat(fd
,
1739 getname_flags(filename
, lookup_flags
, NULL
),
1743 #ifdef CONFIG_COMPAT
1744 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
1745 const compat_uptr_t __user
*, argv
,
1746 const compat_uptr_t __user
*, envp
)
1748 return compat_do_execve(getname(filename
), argv
, envp
);
1751 COMPAT_SYSCALL_DEFINE5(execveat
, int, fd
,
1752 const char __user
*, filename
,
1753 const compat_uptr_t __user
*, argv
,
1754 const compat_uptr_t __user
*, envp
,
1757 int lookup_flags
= (flags
& AT_EMPTY_PATH
) ? LOOKUP_EMPTY
: 0;
1759 return compat_do_execveat(fd
,
1760 getname_flags(filename
, lookup_flags
, NULL
),