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/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/swap.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/perf_event.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52 #include <linux/kmod.h>
53 #include <linux/fsnotify.h>
54 #include <linux/fs_struct.h>
55 #include <linux/pipe_fs_i.h>
56 #include <linux/oom.h>
57 #include <linux/compat.h>
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
63 #include <trace/events/task.h>
66 #include <trace/events/sched.h>
68 int suid_dumpable
= 0;
70 static LIST_HEAD(formats
);
71 static DEFINE_RWLOCK(binfmt_lock
);
73 void __register_binfmt(struct linux_binfmt
* fmt
, int insert
)
76 if (WARN_ON(!fmt
->load_binary
))
78 write_lock(&binfmt_lock
);
79 insert
? list_add(&fmt
->lh
, &formats
) :
80 list_add_tail(&fmt
->lh
, &formats
);
81 write_unlock(&binfmt_lock
);
84 EXPORT_SYMBOL(__register_binfmt
);
86 void unregister_binfmt(struct linux_binfmt
* fmt
)
88 write_lock(&binfmt_lock
);
90 write_unlock(&binfmt_lock
);
93 EXPORT_SYMBOL(unregister_binfmt
);
95 static inline void put_binfmt(struct linux_binfmt
* fmt
)
97 module_put(fmt
->module
);
102 * Note that a shared library must be both readable and executable due to
105 * Also note that we take the address to load from from the file itself.
107 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
109 struct linux_binfmt
*fmt
;
111 struct filename
*tmp
= getname(library
);
112 int error
= PTR_ERR(tmp
);
113 static const struct open_flags uselib_flags
= {
114 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
115 .acc_mode
= MAY_READ
| MAY_EXEC
| MAY_OPEN
,
116 .intent
= LOOKUP_OPEN
,
117 .lookup_flags
= LOOKUP_FOLLOW
,
123 file
= do_filp_open(AT_FDCWD
, tmp
, &uselib_flags
);
125 error
= PTR_ERR(file
);
130 if (!S_ISREG(file_inode(file
)->i_mode
))
134 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
141 read_lock(&binfmt_lock
);
142 list_for_each_entry(fmt
, &formats
, lh
) {
143 if (!fmt
->load_shlib
)
145 if (!try_module_get(fmt
->module
))
147 read_unlock(&binfmt_lock
);
148 error
= fmt
->load_shlib(file
);
149 read_lock(&binfmt_lock
);
151 if (error
!= -ENOEXEC
)
154 read_unlock(&binfmt_lock
);
160 #endif /* #ifdef CONFIG_USELIB */
164 * The nascent bprm->mm is not visible until exec_mmap() but it can
165 * use a lot of memory, account these pages in current->mm temporary
166 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
167 * change the counter back via acct_arg_size(0).
169 static void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
171 struct mm_struct
*mm
= current
->mm
;
172 long diff
= (long)(pages
- bprm
->vma_pages
);
177 bprm
->vma_pages
= pages
;
178 add_mm_counter(mm
, MM_ANONPAGES
, diff
);
181 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
187 #ifdef CONFIG_STACK_GROWSUP
189 ret
= expand_downwards(bprm
->vma
, pos
);
194 ret
= get_user_pages(current
, bprm
->mm
, pos
,
195 1, write
, 1, &page
, NULL
);
200 unsigned long size
= bprm
->vma
->vm_end
- bprm
->vma
->vm_start
;
203 acct_arg_size(bprm
, size
/ PAGE_SIZE
);
206 * We've historically supported up to 32 pages (ARG_MAX)
207 * of argument strings even with small stacks
213 * Limit to 1/4-th the stack size for the argv+env strings.
215 * - the remaining binfmt code will not run out of stack space,
216 * - the program will have a reasonable amount of stack left
219 rlim
= current
->signal
->rlim
;
220 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
) / 4) {
229 static void put_arg_page(struct page
*page
)
234 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
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
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
258 down_write(&mm
->mmap_sem
);
262 * Place the stack at the largest stack address the architecture
263 * supports. Later, we'll move this to an appropriate place. We don't
264 * use STACK_TOP because that can depend on attributes which aren't
267 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
268 vma
->vm_end
= STACK_TOP_MAX
;
269 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
270 vma
->vm_flags
= VM_SOFTDIRTY
| VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
;
271 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
272 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
274 err
= insert_vm_struct(mm
, vma
);
278 mm
->stack_vm
= mm
->total_vm
= 1;
279 up_write(&mm
->mmap_sem
);
280 bprm
->p
= vma
->vm_end
- sizeof(void *);
283 up_write(&mm
->mmap_sem
);
285 kmem_cache_free(vm_area_cachep
, vma
);
289 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
291 return len
<= MAX_ARG_STRLEN
;
296 static inline void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
300 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
305 page
= bprm
->page
[pos
/ PAGE_SIZE
];
306 if (!page
&& write
) {
307 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
310 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
316 static void put_arg_page(struct page
*page
)
320 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
323 __free_page(bprm
->page
[i
]);
324 bprm
->page
[i
] = NULL
;
328 static void free_arg_pages(struct linux_binprm
*bprm
)
332 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
333 free_arg_page(bprm
, i
);
336 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
341 static int __bprm_mm_init(struct linux_binprm
*bprm
)
343 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
347 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
349 return len
<= bprm
->p
;
352 #endif /* CONFIG_MMU */
355 * Create a new mm_struct and populate it with a temporary stack
356 * vm_area_struct. We don't have enough context at this point to set the stack
357 * flags, permissions, and offset, so we use temporary values. We'll update
358 * them later in setup_arg_pages().
360 static int bprm_mm_init(struct linux_binprm
*bprm
)
363 struct mm_struct
*mm
= NULL
;
365 bprm
->mm
= mm
= mm_alloc();
370 err
= init_new_context(current
, mm
);
374 err
= __bprm_mm_init(bprm
);
389 struct user_arg_ptr
{
394 const char __user
*const __user
*native
;
396 const compat_uptr_t __user
*compat
;
401 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
403 const char __user
*native
;
406 if (unlikely(argv
.is_compat
)) {
407 compat_uptr_t compat
;
409 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
410 return ERR_PTR(-EFAULT
);
412 return compat_ptr(compat
);
416 if (get_user(native
, argv
.ptr
.native
+ nr
))
417 return ERR_PTR(-EFAULT
);
423 * count() counts the number of strings in array ARGV.
425 static int count(struct user_arg_ptr argv
, int max
)
429 if (argv
.ptr
.native
!= NULL
) {
431 const char __user
*p
= get_user_arg_ptr(argv
, i
);
443 if (fatal_signal_pending(current
))
444 return -ERESTARTNOHAND
;
452 * 'copy_strings()' copies argument/environment strings from the old
453 * processes's memory to the new process's stack. The call to get_user_pages()
454 * ensures the destination page is created and not swapped out.
456 static int copy_strings(int argc
, struct user_arg_ptr argv
,
457 struct linux_binprm
*bprm
)
459 struct page
*kmapped_page
= NULL
;
461 unsigned long kpos
= 0;
465 const char __user
*str
;
470 str
= get_user_arg_ptr(argv
, argc
);
474 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
479 if (!valid_arg_len(bprm
, len
))
482 /* We're going to work our way backwords. */
488 int offset
, bytes_to_copy
;
490 if (fatal_signal_pending(current
)) {
491 ret
= -ERESTARTNOHAND
;
496 offset
= pos
% PAGE_SIZE
;
500 bytes_to_copy
= offset
;
501 if (bytes_to_copy
> len
)
504 offset
-= bytes_to_copy
;
505 pos
-= bytes_to_copy
;
506 str
-= bytes_to_copy
;
507 len
-= bytes_to_copy
;
509 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
512 page
= get_arg_page(bprm
, pos
, 1);
519 flush_kernel_dcache_page(kmapped_page
);
520 kunmap(kmapped_page
);
521 put_arg_page(kmapped_page
);
524 kaddr
= kmap(kmapped_page
);
525 kpos
= pos
& PAGE_MASK
;
526 flush_arg_page(bprm
, kpos
, kmapped_page
);
528 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
537 flush_kernel_dcache_page(kmapped_page
);
538 kunmap(kmapped_page
);
539 put_arg_page(kmapped_page
);
545 * Like copy_strings, but get argv and its values from kernel memory.
547 int copy_strings_kernel(int argc
, const char *const *__argv
,
548 struct linux_binprm
*bprm
)
551 mm_segment_t oldfs
= get_fs();
552 struct user_arg_ptr argv
= {
553 .ptr
.native
= (const char __user
*const __user
*)__argv
,
557 r
= copy_strings(argc
, argv
, bprm
);
562 EXPORT_SYMBOL(copy_strings_kernel
);
567 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
568 * the binfmt code determines where the new stack should reside, we shift it to
569 * its final location. The process proceeds as follows:
571 * 1) Use shift to calculate the new vma endpoints.
572 * 2) Extend vma to cover both the old and new ranges. This ensures the
573 * arguments passed to subsequent functions are consistent.
574 * 3) Move vma's page tables to the new range.
575 * 4) Free up any cleared pgd range.
576 * 5) Shrink the vma to cover only the new range.
578 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
580 struct mm_struct
*mm
= vma
->vm_mm
;
581 unsigned long old_start
= vma
->vm_start
;
582 unsigned long old_end
= vma
->vm_end
;
583 unsigned long length
= old_end
- old_start
;
584 unsigned long new_start
= old_start
- shift
;
585 unsigned long new_end
= old_end
- shift
;
586 struct mmu_gather tlb
;
588 BUG_ON(new_start
> new_end
);
591 * ensure there are no vmas between where we want to go
594 if (vma
!= find_vma(mm
, new_start
))
598 * cover the whole range: [new_start, old_end)
600 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
604 * move the page tables downwards, on failure we rely on
605 * process cleanup to remove whatever mess we made.
607 if (length
!= move_page_tables(vma
, old_start
,
608 vma
, new_start
, length
, false))
612 tlb_gather_mmu(&tlb
, mm
, old_start
, old_end
);
613 if (new_end
> old_start
) {
615 * when the old and new regions overlap clear from new_end.
617 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
618 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
621 * otherwise, clean from old_start; this is done to not touch
622 * the address space in [new_end, old_start) some architectures
623 * have constraints on va-space that make this illegal (IA64) -
624 * for the others its just a little faster.
626 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
627 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
629 tlb_finish_mmu(&tlb
, old_start
, old_end
);
632 * Shrink the vma to just the new range. Always succeeds.
634 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
640 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
641 * the stack is optionally relocated, and some extra space is added.
643 int setup_arg_pages(struct linux_binprm
*bprm
,
644 unsigned long stack_top
,
645 int executable_stack
)
648 unsigned long stack_shift
;
649 struct mm_struct
*mm
= current
->mm
;
650 struct vm_area_struct
*vma
= bprm
->vma
;
651 struct vm_area_struct
*prev
= NULL
;
652 unsigned long vm_flags
;
653 unsigned long stack_base
;
654 unsigned long stack_size
;
655 unsigned long stack_expand
;
656 unsigned long rlim_stack
;
658 #ifdef CONFIG_STACK_GROWSUP
659 /* Limit stack size to 1GB */
660 stack_base
= rlimit_max(RLIMIT_STACK
);
661 if (stack_base
> (1 << 30))
662 stack_base
= 1 << 30;
664 /* Make sure we didn't let the argument array grow too large. */
665 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
668 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
670 stack_shift
= vma
->vm_start
- stack_base
;
671 mm
->arg_start
= bprm
->p
- stack_shift
;
672 bprm
->p
= vma
->vm_end
- stack_shift
;
674 stack_top
= arch_align_stack(stack_top
);
675 stack_top
= PAGE_ALIGN(stack_top
);
677 if (unlikely(stack_top
< mmap_min_addr
) ||
678 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
681 stack_shift
= vma
->vm_end
- stack_top
;
683 bprm
->p
-= stack_shift
;
684 mm
->arg_start
= bprm
->p
;
688 bprm
->loader
-= stack_shift
;
689 bprm
->exec
-= stack_shift
;
691 down_write(&mm
->mmap_sem
);
692 vm_flags
= VM_STACK_FLAGS
;
695 * Adjust stack execute permissions; explicitly enable for
696 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
697 * (arch default) otherwise.
699 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
701 else if (executable_stack
== EXSTACK_DISABLE_X
)
702 vm_flags
&= ~VM_EXEC
;
703 vm_flags
|= mm
->def_flags
;
704 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
706 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
712 /* Move stack pages down in memory. */
714 ret
= shift_arg_pages(vma
, stack_shift
);
719 /* mprotect_fixup is overkill to remove the temporary stack flags */
720 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
722 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
723 stack_size
= vma
->vm_end
- vma
->vm_start
;
725 * Align this down to a page boundary as expand_stack
728 rlim_stack
= rlimit(RLIMIT_STACK
) & PAGE_MASK
;
729 #ifdef CONFIG_STACK_GROWSUP
730 if (stack_size
+ stack_expand
> rlim_stack
)
731 stack_base
= vma
->vm_start
+ rlim_stack
;
733 stack_base
= vma
->vm_end
+ stack_expand
;
735 if (stack_size
+ stack_expand
> rlim_stack
)
736 stack_base
= vma
->vm_end
- rlim_stack
;
738 stack_base
= vma
->vm_start
- stack_expand
;
740 current
->mm
->start_stack
= bprm
->p
;
741 ret
= expand_stack(vma
, stack_base
);
746 up_write(&mm
->mmap_sem
);
749 EXPORT_SYMBOL(setup_arg_pages
);
751 #endif /* CONFIG_MMU */
753 static struct file
*do_open_exec(struct filename
*name
)
757 static const struct open_flags open_exec_flags
= {
758 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
759 .acc_mode
= MAY_EXEC
| MAY_OPEN
,
760 .intent
= LOOKUP_OPEN
,
761 .lookup_flags
= LOOKUP_FOLLOW
,
764 file
= do_filp_open(AT_FDCWD
, name
, &open_exec_flags
);
769 if (!S_ISREG(file_inode(file
)->i_mode
))
772 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
777 err
= deny_write_access(file
);
789 struct file
*open_exec(const char *name
)
791 struct filename tmp
= { .name
= name
};
792 return do_open_exec(&tmp
);
794 EXPORT_SYMBOL(open_exec
);
796 int kernel_read(struct file
*file
, loff_t offset
,
797 char *addr
, unsigned long count
)
805 /* The cast to a user pointer is valid due to the set_fs() */
806 result
= vfs_read(file
, (void __user
*)addr
, count
, &pos
);
811 EXPORT_SYMBOL(kernel_read
);
813 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
815 ssize_t res
= file
->f_op
->read(file
, (void __user
*)addr
, len
, &pos
);
817 flush_icache_range(addr
, addr
+ len
);
820 EXPORT_SYMBOL(read_code
);
822 static int exec_mmap(struct mm_struct
*mm
)
824 struct task_struct
*tsk
;
825 struct mm_struct
* old_mm
, *active_mm
;
827 /* Notify parent that we're no longer interested in the old VM */
829 old_mm
= current
->mm
;
830 mm_release(tsk
, old_mm
);
835 * Make sure that if there is a core dump in progress
836 * for the old mm, we get out and die instead of going
837 * through with the exec. We must hold mmap_sem around
838 * checking core_state and changing tsk->mm.
840 down_read(&old_mm
->mmap_sem
);
841 if (unlikely(old_mm
->core_state
)) {
842 up_read(&old_mm
->mmap_sem
);
847 active_mm
= tsk
->active_mm
;
850 activate_mm(active_mm
, mm
);
853 up_read(&old_mm
->mmap_sem
);
854 BUG_ON(active_mm
!= old_mm
);
855 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
856 mm_update_next_owner(old_mm
);
865 * This function makes sure the current process has its own signal table,
866 * so that flush_signal_handlers can later reset the handlers without
867 * disturbing other processes. (Other processes might share the signal
868 * table via the CLONE_SIGHAND option to clone().)
870 static int de_thread(struct task_struct
*tsk
)
872 struct signal_struct
*sig
= tsk
->signal
;
873 struct sighand_struct
*oldsighand
= tsk
->sighand
;
874 spinlock_t
*lock
= &oldsighand
->siglock
;
876 if (thread_group_empty(tsk
))
877 goto no_thread_group
;
880 * Kill all other threads in the thread group.
883 if (signal_group_exit(sig
)) {
885 * Another group action in progress, just
886 * return so that the signal is processed.
888 spin_unlock_irq(lock
);
892 sig
->group_exit_task
= tsk
;
893 sig
->notify_count
= zap_other_threads(tsk
);
894 if (!thread_group_leader(tsk
))
897 while (sig
->notify_count
) {
898 __set_current_state(TASK_KILLABLE
);
899 spin_unlock_irq(lock
);
901 if (unlikely(__fatal_signal_pending(tsk
)))
905 spin_unlock_irq(lock
);
908 * At this point all other threads have exited, all we have to
909 * do is to wait for the thread group leader to become inactive,
910 * and to assume its PID:
912 if (!thread_group_leader(tsk
)) {
913 struct task_struct
*leader
= tsk
->group_leader
;
915 sig
->notify_count
= -1; /* for exit_notify() */
917 threadgroup_change_begin(tsk
);
918 write_lock_irq(&tasklist_lock
);
919 if (likely(leader
->exit_state
))
921 __set_current_state(TASK_KILLABLE
);
922 write_unlock_irq(&tasklist_lock
);
923 threadgroup_change_end(tsk
);
925 if (unlikely(__fatal_signal_pending(tsk
)))
930 * The only record we have of the real-time age of a
931 * process, regardless of execs it's done, is start_time.
932 * All the past CPU time is accumulated in signal_struct
933 * from sister threads now dead. But in this non-leader
934 * exec, nothing survives from the original leader thread,
935 * whose birth marks the true age of this process now.
936 * When we take on its identity by switching to its PID, we
937 * also take its birthdate (always earlier than our own).
939 tsk
->start_time
= leader
->start_time
;
940 tsk
->real_start_time
= leader
->real_start_time
;
942 BUG_ON(!same_thread_group(leader
, tsk
));
943 BUG_ON(has_group_leader_pid(tsk
));
945 * An exec() starts a new thread group with the
946 * TGID of the previous thread group. Rehash the
947 * two threads with a switched PID, and release
948 * the former thread group leader:
951 /* Become a process group leader with the old leader's pid.
952 * The old leader becomes a thread of the this thread group.
953 * Note: The old leader also uses this pid until release_task
954 * is called. Odd but simple and correct.
956 tsk
->pid
= leader
->pid
;
957 change_pid(tsk
, PIDTYPE_PID
, task_pid(leader
));
958 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
959 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
961 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
962 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
964 tsk
->group_leader
= tsk
;
965 leader
->group_leader
= tsk
;
967 tsk
->exit_signal
= SIGCHLD
;
968 leader
->exit_signal
= -1;
970 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
971 leader
->exit_state
= EXIT_DEAD
;
974 * We are going to release_task()->ptrace_unlink() silently,
975 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
976 * the tracer wont't block again waiting for this thread.
978 if (unlikely(leader
->ptrace
))
979 __wake_up_parent(leader
, leader
->parent
);
980 write_unlock_irq(&tasklist_lock
);
981 threadgroup_change_end(tsk
);
983 release_task(leader
);
986 sig
->group_exit_task
= NULL
;
987 sig
->notify_count
= 0;
990 /* we have changed execution domain */
991 tsk
->exit_signal
= SIGCHLD
;
994 flush_itimer_signals();
996 if (atomic_read(&oldsighand
->count
) != 1) {
997 struct sighand_struct
*newsighand
;
999 * This ->sighand is shared with the CLONE_SIGHAND
1000 * but not CLONE_THREAD task, switch to the new one.
1002 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1006 atomic_set(&newsighand
->count
, 1);
1007 memcpy(newsighand
->action
, oldsighand
->action
,
1008 sizeof(newsighand
->action
));
1010 write_lock_irq(&tasklist_lock
);
1011 spin_lock(&oldsighand
->siglock
);
1012 rcu_assign_pointer(tsk
->sighand
, newsighand
);
1013 spin_unlock(&oldsighand
->siglock
);
1014 write_unlock_irq(&tasklist_lock
);
1016 __cleanup_sighand(oldsighand
);
1019 BUG_ON(!thread_group_leader(tsk
));
1023 /* protects against exit_notify() and __exit_signal() */
1024 read_lock(&tasklist_lock
);
1025 sig
->group_exit_task
= NULL
;
1026 sig
->notify_count
= 0;
1027 read_unlock(&tasklist_lock
);
1031 char *get_task_comm(char *buf
, struct task_struct
*tsk
)
1033 /* buf must be at least sizeof(tsk->comm) in size */
1035 strncpy(buf
, tsk
->comm
, sizeof(tsk
->comm
));
1039 EXPORT_SYMBOL_GPL(get_task_comm
);
1042 * These functions flushes out all traces of the currently running executable
1043 * so that a new one can be started
1046 void set_task_comm(struct task_struct
*tsk
, char *buf
)
1049 trace_task_rename(tsk
, buf
);
1050 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1052 perf_event_comm(tsk
);
1055 static void filename_to_taskname(char *tcomm
, const char *fn
, unsigned int len
)
1059 /* Copies the binary name from after last slash */
1060 for (i
= 0; (ch
= *(fn
++)) != '\0';) {
1062 i
= 0; /* overwrite what we wrote */
1070 int flush_old_exec(struct linux_binprm
* bprm
)
1075 * Make sure we have a private signal table and that
1076 * we are unassociated from the previous thread group.
1078 retval
= de_thread(current
);
1082 set_mm_exe_file(bprm
->mm
, bprm
->file
);
1084 filename_to_taskname(bprm
->tcomm
, bprm
->filename
, sizeof(bprm
->tcomm
));
1086 * Release all of the old mmap stuff
1088 acct_arg_size(bprm
, 0);
1089 retval
= exec_mmap(bprm
->mm
);
1093 bprm
->mm
= NULL
; /* We're using it now */
1096 current
->flags
&= ~(PF_RANDOMIZE
| PF_FORKNOEXEC
| PF_KTHREAD
|
1097 PF_NOFREEZE
| PF_NO_SETAFFINITY
);
1099 current
->personality
&= ~bprm
->per_clear
;
1106 EXPORT_SYMBOL(flush_old_exec
);
1108 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1110 if (inode_permission(file_inode(file
), MAY_READ
) < 0)
1111 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1113 EXPORT_SYMBOL(would_dump
);
1115 void setup_new_exec(struct linux_binprm
* bprm
)
1117 arch_pick_mmap_layout(current
->mm
);
1119 /* This is the point of no return */
1120 current
->sas_ss_sp
= current
->sas_ss_size
= 0;
1122 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1123 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1125 set_dumpable(current
->mm
, suid_dumpable
);
1127 set_task_comm(current
, bprm
->tcomm
);
1129 /* Set the new mm task size. We have to do that late because it may
1130 * depend on TIF_32BIT which is only updated in flush_thread() on
1131 * some architectures like powerpc
1133 current
->mm
->task_size
= TASK_SIZE
;
1135 /* install the new credentials */
1136 if (!uid_eq(bprm
->cred
->uid
, current_euid()) ||
1137 !gid_eq(bprm
->cred
->gid
, current_egid())) {
1138 current
->pdeath_signal
= 0;
1140 would_dump(bprm
, bprm
->file
);
1141 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
)
1142 set_dumpable(current
->mm
, suid_dumpable
);
1145 /* An exec changes our domain. We are no longer part of the thread
1147 current
->self_exec_id
++;
1148 flush_signal_handlers(current
, 0);
1149 do_close_on_exec(current
->files
);
1151 EXPORT_SYMBOL(setup_new_exec
);
1154 * Prepare credentials and lock ->cred_guard_mutex.
1155 * install_exec_creds() commits the new creds and drops the lock.
1156 * Or, if exec fails before, free_bprm() should release ->cred and
1159 int prepare_bprm_creds(struct linux_binprm
*bprm
)
1161 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1162 return -ERESTARTNOINTR
;
1164 bprm
->cred
= prepare_exec_creds();
1165 if (likely(bprm
->cred
))
1168 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1172 static void free_bprm(struct linux_binprm
*bprm
)
1174 free_arg_pages(bprm
);
1176 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1177 abort_creds(bprm
->cred
);
1180 allow_write_access(bprm
->file
);
1183 /* If a binfmt changed the interp, free it. */
1184 if (bprm
->interp
!= bprm
->filename
)
1185 kfree(bprm
->interp
);
1189 int bprm_change_interp(char *interp
, struct linux_binprm
*bprm
)
1191 /* If a binfmt changed the interp, free it first. */
1192 if (bprm
->interp
!= bprm
->filename
)
1193 kfree(bprm
->interp
);
1194 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1199 EXPORT_SYMBOL(bprm_change_interp
);
1202 * install the new credentials for this executable
1204 void install_exec_creds(struct linux_binprm
*bprm
)
1206 security_bprm_committing_creds(bprm
);
1208 commit_creds(bprm
->cred
);
1212 * Disable monitoring for regular users
1213 * when executing setuid binaries. Must
1214 * wait until new credentials are committed
1215 * by commit_creds() above
1217 if (get_dumpable(current
->mm
) != SUID_DUMP_USER
)
1218 perf_event_exit_task(current
);
1220 * cred_guard_mutex must be held at least to this point to prevent
1221 * ptrace_attach() from altering our determination of the task's
1222 * credentials; any time after this it may be unlocked.
1224 security_bprm_committed_creds(bprm
);
1225 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1227 EXPORT_SYMBOL(install_exec_creds
);
1230 * determine how safe it is to execute the proposed program
1231 * - the caller must hold ->cred_guard_mutex to protect against
1234 static void check_unsafe_exec(struct linux_binprm
*bprm
)
1236 struct task_struct
*p
= current
, *t
;
1240 if (p
->ptrace
& PT_PTRACE_CAP
)
1241 bprm
->unsafe
|= LSM_UNSAFE_PTRACE_CAP
;
1243 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1247 * This isn't strictly necessary, but it makes it harder for LSMs to
1250 if (current
->no_new_privs
)
1251 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1255 spin_lock(&p
->fs
->lock
);
1257 while_each_thread(p
, t
) {
1263 if (p
->fs
->users
> n_fs
)
1264 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1267 spin_unlock(&p
->fs
->lock
);
1271 * Fill the binprm structure from the inode.
1272 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1274 * This may be called multiple times for binary chains (scripts for example).
1276 int prepare_binprm(struct linux_binprm
*bprm
)
1278 struct inode
*inode
= file_inode(bprm
->file
);
1279 umode_t mode
= inode
->i_mode
;
1283 /* clear any previous set[ug]id data from a previous binary */
1284 bprm
->cred
->euid
= current_euid();
1285 bprm
->cred
->egid
= current_egid();
1287 if (!(bprm
->file
->f_path
.mnt
->mnt_flags
& MNT_NOSUID
) &&
1288 !current
->no_new_privs
&&
1289 kuid_has_mapping(bprm
->cred
->user_ns
, inode
->i_uid
) &&
1290 kgid_has_mapping(bprm
->cred
->user_ns
, inode
->i_gid
)) {
1292 if (mode
& S_ISUID
) {
1293 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1294 bprm
->cred
->euid
= inode
->i_uid
;
1299 * If setgid is set but no group execute bit then this
1300 * is a candidate for mandatory locking, not a setgid
1303 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1304 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1305 bprm
->cred
->egid
= inode
->i_gid
;
1309 /* fill in binprm security blob */
1310 retval
= security_bprm_set_creds(bprm
);
1313 bprm
->cred_prepared
= 1;
1315 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1316 return kernel_read(bprm
->file
, 0, bprm
->buf
, BINPRM_BUF_SIZE
);
1319 EXPORT_SYMBOL(prepare_binprm
);
1322 * Arguments are '\0' separated strings found at the location bprm->p
1323 * points to; chop off the first by relocating brpm->p to right after
1324 * the first '\0' encountered.
1326 int remove_arg_zero(struct linux_binprm
*bprm
)
1329 unsigned long offset
;
1337 offset
= bprm
->p
& ~PAGE_MASK
;
1338 page
= get_arg_page(bprm
, bprm
->p
, 0);
1343 kaddr
= kmap_atomic(page
);
1345 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1346 offset
++, bprm
->p
++)
1349 kunmap_atomic(kaddr
);
1352 if (offset
== PAGE_SIZE
)
1353 free_arg_page(bprm
, (bprm
->p
>> PAGE_SHIFT
) - 1);
1354 } while (offset
== PAGE_SIZE
);
1363 EXPORT_SYMBOL(remove_arg_zero
);
1365 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1367 * cycle the list of binary formats handler, until one recognizes the image
1369 int search_binary_handler(struct linux_binprm
*bprm
)
1371 bool need_retry
= IS_ENABLED(CONFIG_MODULES
);
1372 struct linux_binfmt
*fmt
;
1375 /* This allows 4 levels of binfmt rewrites before failing hard. */
1376 if (bprm
->recursion_depth
> 5)
1379 retval
= security_bprm_check(bprm
);
1385 read_lock(&binfmt_lock
);
1386 list_for_each_entry(fmt
, &formats
, lh
) {
1387 if (!try_module_get(fmt
->module
))
1389 read_unlock(&binfmt_lock
);
1390 bprm
->recursion_depth
++;
1391 retval
= fmt
->load_binary(bprm
);
1392 bprm
->recursion_depth
--;
1393 if (retval
>= 0 || retval
!= -ENOEXEC
||
1394 bprm
->mm
== NULL
|| bprm
->file
== NULL
) {
1398 read_lock(&binfmt_lock
);
1401 read_unlock(&binfmt_lock
);
1403 if (need_retry
&& retval
== -ENOEXEC
) {
1404 if (printable(bprm
->buf
[0]) && printable(bprm
->buf
[1]) &&
1405 printable(bprm
->buf
[2]) && printable(bprm
->buf
[3]))
1407 if (request_module("binfmt-%04x", *(ushort
*)(bprm
->buf
+ 2)) < 0)
1415 EXPORT_SYMBOL(search_binary_handler
);
1417 static int exec_binprm(struct linux_binprm
*bprm
)
1419 pid_t old_pid
, old_vpid
;
1422 /* Need to fetch pid before load_binary changes it */
1423 old_pid
= current
->pid
;
1425 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1428 ret
= search_binary_handler(bprm
);
1431 trace_sched_process_exec(current
, old_pid
, bprm
);
1432 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1433 proc_exec_connector(current
);
1440 * sys_execve() executes a new program.
1442 static int do_execve_common(struct filename
*filename
,
1443 struct user_arg_ptr argv
,
1444 struct user_arg_ptr envp
)
1446 struct linux_binprm
*bprm
;
1448 struct files_struct
*displaced
;
1451 if (IS_ERR(filename
))
1452 return PTR_ERR(filename
);
1455 * We move the actual failure in case of RLIMIT_NPROC excess from
1456 * set*uid() to execve() because too many poorly written programs
1457 * don't check setuid() return code. Here we additionally recheck
1458 * whether NPROC limit is still exceeded.
1460 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1461 atomic_read(¤t_user()->processes
) > rlimit(RLIMIT_NPROC
)) {
1466 /* We're below the limit (still or again), so we don't want to make
1467 * further execve() calls fail. */
1468 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1470 retval
= unshare_files(&displaced
);
1475 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1479 retval
= prepare_bprm_creds(bprm
);
1483 check_unsafe_exec(bprm
);
1484 current
->in_execve
= 1;
1486 file
= do_open_exec(filename
);
1487 retval
= PTR_ERR(file
);
1494 bprm
->filename
= bprm
->interp
= filename
->name
;
1496 retval
= bprm_mm_init(bprm
);
1500 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
1501 if ((retval
= bprm
->argc
) < 0)
1504 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
1505 if ((retval
= bprm
->envc
) < 0)
1508 retval
= prepare_binprm(bprm
);
1512 retval
= copy_strings_kernel(1, &bprm
->filename
, bprm
);
1516 bprm
->exec
= bprm
->p
;
1517 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1521 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1525 retval
= exec_binprm(bprm
);
1529 /* execve succeeded */
1530 current
->fs
->in_exec
= 0;
1531 current
->in_execve
= 0;
1532 acct_update_integrals(current
);
1533 task_numa_free(current
);
1537 put_files_struct(displaced
);
1542 acct_arg_size(bprm
, 0);
1547 current
->fs
->in_exec
= 0;
1548 current
->in_execve
= 0;
1555 reset_files_struct(displaced
);
1561 int do_execve(struct filename
*filename
,
1562 const char __user
*const __user
*__argv
,
1563 const char __user
*const __user
*__envp
)
1565 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1566 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1567 return do_execve_common(filename
, argv
, envp
);
1570 #ifdef CONFIG_COMPAT
1571 static int compat_do_execve(struct filename
*filename
,
1572 const compat_uptr_t __user
*__argv
,
1573 const compat_uptr_t __user
*__envp
)
1575 struct user_arg_ptr argv
= {
1577 .ptr
.compat
= __argv
,
1579 struct user_arg_ptr envp
= {
1581 .ptr
.compat
= __envp
,
1583 return do_execve_common(filename
, argv
, envp
);
1587 void set_binfmt(struct linux_binfmt
*new)
1589 struct mm_struct
*mm
= current
->mm
;
1592 module_put(mm
->binfmt
->module
);
1596 __module_get(new->module
);
1598 EXPORT_SYMBOL(set_binfmt
);
1601 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1603 void set_dumpable(struct mm_struct
*mm
, int value
)
1605 unsigned long old
, new;
1607 if (WARN_ON((unsigned)value
> SUID_DUMP_ROOT
))
1611 old
= ACCESS_ONCE(mm
->flags
);
1612 new = (old
& ~MMF_DUMPABLE_MASK
) | value
;
1613 } while (cmpxchg(&mm
->flags
, old
, new) != old
);
1616 SYSCALL_DEFINE3(execve
,
1617 const char __user
*, filename
,
1618 const char __user
*const __user
*, argv
,
1619 const char __user
*const __user
*, envp
)
1621 return do_execve(getname(filename
), argv
, envp
);
1623 #ifdef CONFIG_COMPAT
1624 COMPAT_SYSCALL_DEFINE3(execve
, const char __user
*, filename
,
1625 const compat_uptr_t __user
*, argv
,
1626 const compat_uptr_t __user
*, envp
)
1628 return compat_do_execve(getname(filename
), argv
, envp
);