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>
67 #include <trace/events/sched.h>
69 int suid_dumpable
= 0;
71 static LIST_HEAD(formats
);
72 static DEFINE_RWLOCK(binfmt_lock
);
74 void __register_binfmt(struct linux_binfmt
* fmt
, int insert
)
77 write_lock(&binfmt_lock
);
78 insert
? list_add(&fmt
->lh
, &formats
) :
79 list_add_tail(&fmt
->lh
, &formats
);
80 write_unlock(&binfmt_lock
);
83 EXPORT_SYMBOL(__register_binfmt
);
85 void unregister_binfmt(struct linux_binfmt
* fmt
)
87 write_lock(&binfmt_lock
);
89 write_unlock(&binfmt_lock
);
92 EXPORT_SYMBOL(unregister_binfmt
);
94 static inline void put_binfmt(struct linux_binfmt
* fmt
)
96 module_put(fmt
->module
);
100 * Note that a shared library must be both readable and executable due to
103 * Also note that we take the address to load from from the file itself.
105 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
108 struct filename
*tmp
= getname(library
);
109 int error
= PTR_ERR(tmp
);
110 static const struct open_flags uselib_flags
= {
111 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
112 .acc_mode
= MAY_READ
| MAY_EXEC
| MAY_OPEN
,
113 .intent
= LOOKUP_OPEN
119 file
= do_filp_open(AT_FDCWD
, tmp
, &uselib_flags
, LOOKUP_FOLLOW
);
121 error
= PTR_ERR(file
);
126 if (!S_ISREG(file_inode(file
)->i_mode
))
130 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
137 struct linux_binfmt
* fmt
;
139 read_lock(&binfmt_lock
);
140 list_for_each_entry(fmt
, &formats
, lh
) {
141 if (!fmt
->load_shlib
)
143 if (!try_module_get(fmt
->module
))
145 read_unlock(&binfmt_lock
);
146 error
= fmt
->load_shlib(file
);
147 read_lock(&binfmt_lock
);
149 if (error
!= -ENOEXEC
)
152 read_unlock(&binfmt_lock
);
162 * The nascent bprm->mm is not visible until exec_mmap() but it can
163 * use a lot of memory, account these pages in current->mm temporary
164 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
165 * change the counter back via acct_arg_size(0).
167 static void acct_arg_size(struct linux_binprm
*bprm
, unsigned long pages
)
169 struct mm_struct
*mm
= current
->mm
;
170 long diff
= (long)(pages
- bprm
->vma_pages
);
175 bprm
->vma_pages
= pages
;
176 add_mm_counter(mm
, MM_ANONPAGES
, diff
);
179 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
185 #ifdef CONFIG_STACK_GROWSUP
187 ret
= expand_downwards(bprm
->vma
, pos
);
192 ret
= get_user_pages(current
, bprm
->mm
, pos
,
193 1, write
, 1, &page
, NULL
);
198 unsigned long size
= bprm
->vma
->vm_end
- bprm
->vma
->vm_start
;
199 unsigned long ptr_size
;
203 * Since the stack will hold pointers to the strings, we
204 * must account for them as well.
206 * The size calculation is the entire vma while each arg page is
207 * built, so each time we get here it's calculating how far it
208 * is currently (rather than each call being just the newly
209 * added size from the arg page). As a result, we need to
210 * always add the entire size of the pointers, so that on the
211 * last call to get_arg_page() we'll actually have the entire
214 ptr_size
= (bprm
->argc
+ bprm
->envc
) * sizeof(void *);
215 if (ptr_size
> ULONG_MAX
- size
)
219 acct_arg_size(bprm
, size
/ PAGE_SIZE
);
222 * We've historically supported up to 32 pages (ARG_MAX)
223 * of argument strings even with small stacks
229 * Limit to 1/4-th the stack size for the argv+env strings.
231 * - the remaining binfmt code will not run out of stack space,
232 * - the program will have a reasonable amount of stack left
235 rlim
= current
->signal
->rlim
;
236 if (size
> ACCESS_ONCE(rlim
[RLIMIT_STACK
].rlim_cur
) / 4)
247 static void put_arg_page(struct page
*page
)
252 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
256 static void free_arg_pages(struct linux_binprm
*bprm
)
260 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
263 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
266 static int __bprm_mm_init(struct linux_binprm
*bprm
)
269 struct vm_area_struct
*vma
= NULL
;
270 struct mm_struct
*mm
= bprm
->mm
;
272 bprm
->vma
= vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
276 down_write(&mm
->mmap_sem
);
280 * Place the stack at the largest stack address the architecture
281 * supports. Later, we'll move this to an appropriate place. We don't
282 * use STACK_TOP because that can depend on attributes which aren't
285 BUILD_BUG_ON(VM_STACK_FLAGS
& VM_STACK_INCOMPLETE_SETUP
);
286 vma
->vm_end
= STACK_TOP_MAX
;
287 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
288 vma
->vm_flags
= VM_STACK_FLAGS
| VM_STACK_INCOMPLETE_SETUP
;
289 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
290 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
292 err
= insert_vm_struct(mm
, vma
);
296 mm
->stack_vm
= mm
->total_vm
= 1;
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
= init_new_context(current
, mm
);
392 err
= __bprm_mm_init(bprm
);
407 struct user_arg_ptr
{
412 const char __user
*const __user
*native
;
414 const compat_uptr_t __user
*compat
;
419 static const char __user
*get_user_arg_ptr(struct user_arg_ptr argv
, int nr
)
421 const char __user
*native
;
424 if (unlikely(argv
.is_compat
)) {
425 compat_uptr_t compat
;
427 if (get_user(compat
, argv
.ptr
.compat
+ nr
))
428 return ERR_PTR(-EFAULT
);
430 return compat_ptr(compat
);
434 if (get_user(native
, argv
.ptr
.native
+ nr
))
435 return ERR_PTR(-EFAULT
);
441 * count() counts the number of strings in array ARGV.
443 static int count(struct user_arg_ptr argv
, int max
)
447 if (argv
.ptr
.native
!= NULL
) {
449 const char __user
*p
= get_user_arg_ptr(argv
, i
);
461 if (fatal_signal_pending(current
))
462 return -ERESTARTNOHAND
;
470 * 'copy_strings()' copies argument/environment strings from the old
471 * processes's memory to the new process's stack. The call to get_user_pages()
472 * ensures the destination page is created and not swapped out.
474 static int copy_strings(int argc
, struct user_arg_ptr argv
,
475 struct linux_binprm
*bprm
)
477 struct page
*kmapped_page
= NULL
;
479 unsigned long kpos
= 0;
483 const char __user
*str
;
488 str
= get_user_arg_ptr(argv
, argc
);
492 len
= strnlen_user(str
, MAX_ARG_STRLEN
);
497 if (!valid_arg_len(bprm
, len
))
500 /* We're going to work our way backwords. */
506 int offset
, bytes_to_copy
;
508 if (fatal_signal_pending(current
)) {
509 ret
= -ERESTARTNOHAND
;
514 offset
= pos
% PAGE_SIZE
;
518 bytes_to_copy
= offset
;
519 if (bytes_to_copy
> len
)
522 offset
-= bytes_to_copy
;
523 pos
-= bytes_to_copy
;
524 str
-= bytes_to_copy
;
525 len
-= bytes_to_copy
;
527 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
530 page
= get_arg_page(bprm
, pos
, 1);
537 flush_kernel_dcache_page(kmapped_page
);
538 kunmap(kmapped_page
);
539 put_arg_page(kmapped_page
);
542 kaddr
= kmap(kmapped_page
);
543 kpos
= pos
& PAGE_MASK
;
544 flush_arg_page(bprm
, kpos
, kmapped_page
);
546 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
555 flush_kernel_dcache_page(kmapped_page
);
556 kunmap(kmapped_page
);
557 put_arg_page(kmapped_page
);
563 * Like copy_strings, but get argv and its values from kernel memory.
565 int copy_strings_kernel(int argc
, const char *const *__argv
,
566 struct linux_binprm
*bprm
)
569 mm_segment_t oldfs
= get_fs();
570 struct user_arg_ptr argv
= {
571 .ptr
.native
= (const char __user
*const __user
*)__argv
,
575 r
= copy_strings(argc
, argv
, bprm
);
580 EXPORT_SYMBOL(copy_strings_kernel
);
585 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
586 * the binfmt code determines where the new stack should reside, we shift it to
587 * its final location. The process proceeds as follows:
589 * 1) Use shift to calculate the new vma endpoints.
590 * 2) Extend vma to cover both the old and new ranges. This ensures the
591 * arguments passed to subsequent functions are consistent.
592 * 3) Move vma's page tables to the new range.
593 * 4) Free up any cleared pgd range.
594 * 5) Shrink the vma to cover only the new range.
596 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
598 struct mm_struct
*mm
= vma
->vm_mm
;
599 unsigned long old_start
= vma
->vm_start
;
600 unsigned long old_end
= vma
->vm_end
;
601 unsigned long length
= old_end
- old_start
;
602 unsigned long new_start
= old_start
- shift
;
603 unsigned long new_end
= old_end
- shift
;
604 struct mmu_gather tlb
;
606 BUG_ON(new_start
> new_end
);
609 * ensure there are no vmas between where we want to go
612 if (vma
!= find_vma(mm
, new_start
))
616 * cover the whole range: [new_start, old_end)
618 if (vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
))
622 * move the page tables downwards, on failure we rely on
623 * process cleanup to remove whatever mess we made.
625 if (length
!= move_page_tables(vma
, old_start
,
626 vma
, new_start
, length
, false))
630 tlb_gather_mmu(&tlb
, mm
, old_start
, old_end
);
631 if (new_end
> old_start
) {
633 * when the old and new regions overlap clear from new_end.
635 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
636 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
639 * otherwise, clean from old_start; this is done to not touch
640 * the address space in [new_end, old_start) some architectures
641 * have constraints on va-space that make this illegal (IA64) -
642 * for the others its just a little faster.
644 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
645 vma
->vm_next
? vma
->vm_next
->vm_start
: USER_PGTABLES_CEILING
);
647 tlb_finish_mmu(&tlb
, old_start
, old_end
);
650 * Shrink the vma to just the new range. Always succeeds.
652 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
658 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
659 * the stack is optionally relocated, and some extra space is added.
661 int setup_arg_pages(struct linux_binprm
*bprm
,
662 unsigned long stack_top
,
663 int executable_stack
)
666 unsigned long stack_shift
;
667 struct mm_struct
*mm
= current
->mm
;
668 struct vm_area_struct
*vma
= bprm
->vma
;
669 struct vm_area_struct
*prev
= NULL
;
670 unsigned long vm_flags
;
671 unsigned long stack_base
;
672 unsigned long stack_size
;
673 unsigned long stack_expand
;
674 unsigned long rlim_stack
;
676 #ifdef CONFIG_STACK_GROWSUP
677 /* Limit stack size */
678 stack_base
= rlimit_max(RLIMIT_STACK
);
679 if (stack_base
> STACK_SIZE_MAX
)
680 stack_base
= STACK_SIZE_MAX
;
682 /* Make sure we didn't let the argument array grow too large. */
683 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
686 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
688 stack_shift
= vma
->vm_start
- stack_base
;
689 mm
->arg_start
= bprm
->p
- stack_shift
;
690 bprm
->p
= vma
->vm_end
- stack_shift
;
692 stack_top
= arch_align_stack(stack_top
);
693 stack_top
= PAGE_ALIGN(stack_top
);
695 if (unlikely(stack_top
< mmap_min_addr
) ||
696 unlikely(vma
->vm_end
- vma
->vm_start
>= stack_top
- mmap_min_addr
))
699 stack_shift
= vma
->vm_end
- stack_top
;
701 bprm
->p
-= stack_shift
;
702 mm
->arg_start
= bprm
->p
;
706 bprm
->loader
-= stack_shift
;
707 bprm
->exec
-= stack_shift
;
709 down_write(&mm
->mmap_sem
);
710 vm_flags
= VM_STACK_FLAGS
;
713 * Adjust stack execute permissions; explicitly enable for
714 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
715 * (arch default) otherwise.
717 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
719 else if (executable_stack
== EXSTACK_DISABLE_X
)
720 vm_flags
&= ~VM_EXEC
;
721 vm_flags
|= mm
->def_flags
;
722 vm_flags
|= VM_STACK_INCOMPLETE_SETUP
;
724 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
730 /* Move stack pages down in memory. */
732 ret
= shift_arg_pages(vma
, stack_shift
);
737 /* mprotect_fixup is overkill to remove the temporary stack flags */
738 vma
->vm_flags
&= ~VM_STACK_INCOMPLETE_SETUP
;
740 stack_expand
= 131072UL; /* randomly 32*4k (or 2*64k) pages */
741 stack_size
= vma
->vm_end
- vma
->vm_start
;
743 * Align this down to a page boundary as expand_stack
746 rlim_stack
= rlimit(RLIMIT_STACK
) & PAGE_MASK
;
747 #ifdef CONFIG_STACK_GROWSUP
748 if (stack_size
+ stack_expand
> rlim_stack
)
749 stack_base
= vma
->vm_start
+ rlim_stack
;
751 stack_base
= vma
->vm_end
+ stack_expand
;
753 if (stack_size
+ stack_expand
> rlim_stack
)
754 stack_base
= vma
->vm_end
- rlim_stack
;
756 stack_base
= vma
->vm_start
- stack_expand
;
758 current
->mm
->start_stack
= bprm
->p
;
759 ret
= expand_stack(vma
, stack_base
);
764 up_write(&mm
->mmap_sem
);
767 EXPORT_SYMBOL(setup_arg_pages
);
769 #endif /* CONFIG_MMU */
771 struct file
*open_exec(const char *name
)
775 struct filename tmp
= { .name
= name
};
776 static const struct open_flags open_exec_flags
= {
777 .open_flag
= O_LARGEFILE
| O_RDONLY
| __FMODE_EXEC
,
778 .acc_mode
= MAY_EXEC
| MAY_OPEN
,
779 .intent
= LOOKUP_OPEN
782 file
= do_filp_open(AT_FDCWD
, &tmp
, &open_exec_flags
, LOOKUP_FOLLOW
);
787 if (!S_ISREG(file_inode(file
)->i_mode
))
790 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
795 err
= deny_write_access(file
);
806 EXPORT_SYMBOL(open_exec
);
808 int kernel_read(struct file
*file
, loff_t offset
,
809 char *addr
, unsigned long count
)
817 /* The cast to a user pointer is valid due to the set_fs() */
818 result
= vfs_read(file
, (void __user
*)addr
, count
, &pos
);
823 EXPORT_SYMBOL(kernel_read
);
825 ssize_t
read_code(struct file
*file
, unsigned long addr
, loff_t pos
, size_t len
)
827 ssize_t res
= file
->f_op
->read(file
, (void __user
*)addr
, len
, &pos
);
829 flush_icache_range(addr
, addr
+ len
);
832 EXPORT_SYMBOL(read_code
);
834 static int exec_mmap(struct mm_struct
*mm
)
836 struct task_struct
*tsk
;
837 struct mm_struct
* old_mm
, *active_mm
;
839 /* Notify parent that we're no longer interested in the old VM */
841 old_mm
= current
->mm
;
842 mm_release(tsk
, old_mm
);
847 * Make sure that if there is a core dump in progress
848 * for the old mm, we get out and die instead of going
849 * through with the exec. We must hold mmap_sem around
850 * checking core_state and changing tsk->mm.
852 down_read(&old_mm
->mmap_sem
);
853 if (unlikely(old_mm
->core_state
)) {
854 up_read(&old_mm
->mmap_sem
);
859 active_mm
= tsk
->active_mm
;
862 activate_mm(active_mm
, mm
);
864 arch_pick_mmap_layout(mm
);
866 up_read(&old_mm
->mmap_sem
);
867 BUG_ON(active_mm
!= old_mm
);
868 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, old_mm
);
869 mm_update_next_owner(old_mm
);
878 * This function makes sure the current process has its own signal table,
879 * so that flush_signal_handlers can later reset the handlers without
880 * disturbing other processes. (Other processes might share the signal
881 * table via the CLONE_SIGHAND option to clone().)
883 static int de_thread(struct task_struct
*tsk
)
885 struct signal_struct
*sig
= tsk
->signal
;
886 struct sighand_struct
*oldsighand
= tsk
->sighand
;
887 spinlock_t
*lock
= &oldsighand
->siglock
;
889 if (thread_group_empty(tsk
))
890 goto no_thread_group
;
893 * Kill all other threads in the thread group.
896 if (signal_group_exit(sig
)) {
898 * Another group action in progress, just
899 * return so that the signal is processed.
901 spin_unlock_irq(lock
);
905 sig
->group_exit_task
= tsk
;
906 sig
->notify_count
= zap_other_threads(tsk
);
907 if (!thread_group_leader(tsk
))
910 while (sig
->notify_count
) {
911 __set_current_state(TASK_KILLABLE
);
912 spin_unlock_irq(lock
);
914 if (unlikely(__fatal_signal_pending(tsk
)))
918 spin_unlock_irq(lock
);
921 * At this point all other threads have exited, all we have to
922 * do is to wait for the thread group leader to become inactive,
923 * and to assume its PID:
925 if (!thread_group_leader(tsk
)) {
926 struct task_struct
*leader
= tsk
->group_leader
;
928 sig
->notify_count
= -1; /* for exit_notify() */
930 threadgroup_change_begin(tsk
);
931 write_lock_irq(&tasklist_lock
);
932 if (likely(leader
->exit_state
))
934 __set_current_state(TASK_KILLABLE
);
935 write_unlock_irq(&tasklist_lock
);
936 threadgroup_change_end(tsk
);
938 if (unlikely(__fatal_signal_pending(tsk
)))
943 * The only record we have of the real-time age of a
944 * process, regardless of execs it's done, is start_time.
945 * All the past CPU time is accumulated in signal_struct
946 * from sister threads now dead. But in this non-leader
947 * exec, nothing survives from the original leader thread,
948 * whose birth marks the true age of this process now.
949 * When we take on its identity by switching to its PID, we
950 * also take its birthdate (always earlier than our own).
952 tsk
->start_time
= leader
->start_time
;
954 BUG_ON(!same_thread_group(leader
, tsk
));
955 BUG_ON(has_group_leader_pid(tsk
));
957 * An exec() starts a new thread group with the
958 * TGID of the previous thread group. Rehash the
959 * two threads with a switched PID, and release
960 * the former thread group leader:
963 /* Become a process group leader with the old leader's pid.
964 * The old leader becomes a thread of the this thread group.
965 * Note: The old leader also uses this pid until release_task
966 * is called. Odd but simple and correct.
968 detach_pid(tsk
, PIDTYPE_PID
);
969 tsk
->pid
= leader
->pid
;
970 attach_pid(tsk
, PIDTYPE_PID
, task_pid(leader
));
971 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
972 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
974 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
975 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
977 tsk
->group_leader
= tsk
;
978 leader
->group_leader
= tsk
;
980 tsk
->exit_signal
= SIGCHLD
;
981 leader
->exit_signal
= -1;
983 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
984 leader
->exit_state
= EXIT_DEAD
;
987 * We are going to release_task()->ptrace_unlink() silently,
988 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
989 * the tracer wont't block again waiting for this thread.
991 if (unlikely(leader
->ptrace
))
992 __wake_up_parent(leader
, leader
->parent
);
993 write_unlock_irq(&tasklist_lock
);
994 threadgroup_change_end(tsk
);
996 release_task(leader
);
999 sig
->group_exit_task
= NULL
;
1000 sig
->notify_count
= 0;
1003 /* we have changed execution domain */
1004 tsk
->exit_signal
= SIGCHLD
;
1007 flush_itimer_signals();
1009 if (atomic_read(&oldsighand
->count
) != 1) {
1010 struct sighand_struct
*newsighand
;
1012 * This ->sighand is shared with the CLONE_SIGHAND
1013 * but not CLONE_THREAD task, switch to the new one.
1015 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1019 atomic_set(&newsighand
->count
, 1);
1020 memcpy(newsighand
->action
, oldsighand
->action
,
1021 sizeof(newsighand
->action
));
1023 write_lock_irq(&tasklist_lock
);
1024 spin_lock(&oldsighand
->siglock
);
1025 rcu_assign_pointer(tsk
->sighand
, newsighand
);
1026 spin_unlock(&oldsighand
->siglock
);
1027 write_unlock_irq(&tasklist_lock
);
1029 __cleanup_sighand(oldsighand
);
1032 BUG_ON(!thread_group_leader(tsk
));
1036 /* protects against exit_notify() and __exit_signal() */
1037 read_lock(&tasklist_lock
);
1038 sig
->group_exit_task
= NULL
;
1039 sig
->notify_count
= 0;
1040 read_unlock(&tasklist_lock
);
1044 char *get_task_comm(char *buf
, struct task_struct
*tsk
)
1046 /* buf must be at least sizeof(tsk->comm) in size */
1048 strncpy(buf
, tsk
->comm
, sizeof(tsk
->comm
));
1052 EXPORT_SYMBOL_GPL(get_task_comm
);
1055 * These functions flushes out all traces of the currently running executable
1056 * so that a new one can be started
1059 void set_task_comm(struct task_struct
*tsk
, char *buf
)
1062 trace_task_rename(tsk
, buf
);
1063 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
1065 perf_event_comm(tsk
);
1068 static void filename_to_taskname(char *tcomm
, const char *fn
, unsigned int len
)
1072 /* Copies the binary name from after last slash */
1073 for (i
= 0; (ch
= *(fn
++)) != '\0';) {
1075 i
= 0; /* overwrite what we wrote */
1083 int flush_old_exec(struct linux_binprm
* bprm
)
1088 * Make sure we have a private signal table and that
1089 * we are unassociated from the previous thread group.
1091 retval
= de_thread(current
);
1095 set_mm_exe_file(bprm
->mm
, bprm
->file
);
1097 filename_to_taskname(bprm
->tcomm
, bprm
->filename
, sizeof(bprm
->tcomm
));
1099 * Release all of the old mmap stuff
1101 acct_arg_size(bprm
, 0);
1102 retval
= exec_mmap(bprm
->mm
);
1106 bprm
->mm
= NULL
; /* We're using it now */
1110 ~(PF_RANDOMIZE
| PF_FORKNOEXEC
| PF_KTHREAD
| PF_NOFREEZE
);
1112 current
->personality
&= ~bprm
->per_clear
;
1115 * We have to apply CLOEXEC before we change whether the process is
1116 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1117 * trying to access the should-be-closed file descriptors of a process
1118 * undergoing exec(2).
1120 do_close_on_exec(current
->files
);
1126 EXPORT_SYMBOL(flush_old_exec
);
1128 void would_dump(struct linux_binprm
*bprm
, struct file
*file
)
1130 if (inode_permission(file_inode(file
), MAY_READ
) < 0)
1131 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
1133 EXPORT_SYMBOL(would_dump
);
1135 void setup_new_exec(struct linux_binprm
* bprm
)
1137 arch_pick_mmap_layout(current
->mm
);
1139 /* This is the point of no return */
1140 current
->sas_ss_sp
= current
->sas_ss_size
= 0;
1142 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1143 set_dumpable(current
->mm
, SUID_DUMP_USER
);
1145 set_dumpable(current
->mm
, suid_dumpable
);
1147 set_task_comm(current
, bprm
->tcomm
);
1149 /* Set the new mm task size. We have to do that late because it may
1150 * depend on TIF_32BIT which is only updated in flush_thread() on
1151 * some architectures like powerpc
1153 current
->mm
->task_size
= TASK_SIZE
;
1155 /* install the new credentials */
1156 if (!uid_eq(bprm
->cred
->uid
, current_euid()) ||
1157 !gid_eq(bprm
->cred
->gid
, current_egid())) {
1158 current
->pdeath_signal
= 0;
1160 would_dump(bprm
, bprm
->file
);
1161 if (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
)
1162 set_dumpable(current
->mm
, suid_dumpable
);
1165 /* An exec changes our domain. We are no longer part of the thread
1168 current
->self_exec_id
++;
1170 flush_signal_handlers(current
, 0);
1172 EXPORT_SYMBOL(setup_new_exec
);
1175 * Prepare credentials and lock ->cred_guard_mutex.
1176 * install_exec_creds() commits the new creds and drops the lock.
1177 * Or, if exec fails before, free_bprm() should release ->cred and
1180 int prepare_bprm_creds(struct linux_binprm
*bprm
)
1182 if (mutex_lock_interruptible(¤t
->signal
->cred_guard_mutex
))
1183 return -ERESTARTNOINTR
;
1185 bprm
->cred
= prepare_exec_creds();
1186 if (likely(bprm
->cred
))
1189 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1193 void free_bprm(struct linux_binprm
*bprm
)
1195 free_arg_pages(bprm
);
1197 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1198 abort_creds(bprm
->cred
);
1200 /* If a binfmt changed the interp, free it. */
1201 if (bprm
->interp
!= bprm
->filename
)
1202 kfree(bprm
->interp
);
1206 int bprm_change_interp(char *interp
, struct linux_binprm
*bprm
)
1208 /* If a binfmt changed the interp, free it first. */
1209 if (bprm
->interp
!= bprm
->filename
)
1210 kfree(bprm
->interp
);
1211 bprm
->interp
= kstrdup(interp
, GFP_KERNEL
);
1216 EXPORT_SYMBOL(bprm_change_interp
);
1219 * install the new credentials for this executable
1221 void install_exec_creds(struct linux_binprm
*bprm
)
1223 security_bprm_committing_creds(bprm
);
1225 commit_creds(bprm
->cred
);
1229 * Disable monitoring for regular users
1230 * when executing setuid binaries. Must
1231 * wait until new credentials are committed
1232 * by commit_creds() above
1234 if (get_dumpable(current
->mm
) != SUID_DUMP_USER
)
1235 perf_event_exit_task(current
);
1237 * cred_guard_mutex must be held at least to this point to prevent
1238 * ptrace_attach() from altering our determination of the task's
1239 * credentials; any time after this it may be unlocked.
1241 security_bprm_committed_creds(bprm
);
1242 mutex_unlock(¤t
->signal
->cred_guard_mutex
);
1244 EXPORT_SYMBOL(install_exec_creds
);
1247 * determine how safe it is to execute the proposed program
1248 * - the caller must hold ->cred_guard_mutex to protect against
1249 * PTRACE_ATTACH or seccomp thread-sync
1251 static int check_unsafe_exec(struct linux_binprm
*bprm
)
1253 struct task_struct
*p
= current
, *t
;
1258 if (p
->ptrace
& PT_PTRACE_CAP
)
1259 bprm
->unsafe
|= LSM_UNSAFE_PTRACE_CAP
;
1261 bprm
->unsafe
|= LSM_UNSAFE_PTRACE
;
1265 * This isn't strictly necessary, but it makes it harder for LSMs to
1268 if (task_no_new_privs(current
))
1269 bprm
->unsafe
|= LSM_UNSAFE_NO_NEW_PRIVS
;
1272 spin_lock(&p
->fs
->lock
);
1274 for (t
= next_thread(p
); t
!= p
; t
= next_thread(t
)) {
1280 if (p
->fs
->users
> n_fs
) {
1281 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1284 if (!p
->fs
->in_exec
) {
1289 spin_unlock(&p
->fs
->lock
);
1294 static void bprm_fill_uid(struct linux_binprm
*bprm
)
1296 struct inode
*inode
;
1301 /* clear any previous set[ug]id data from a previous binary */
1302 bprm
->cred
->euid
= current_euid();
1303 bprm
->cred
->egid
= current_egid();
1305 if (bprm
->file
->f_path
.mnt
->mnt_flags
& MNT_NOSUID
)
1308 if (current
->no_new_privs
)
1311 inode
= file_inode(bprm
->file
);
1312 mode
= ACCESS_ONCE(inode
->i_mode
);
1313 if (!(mode
& (S_ISUID
|S_ISGID
)))
1316 /* Be careful if suid/sgid is set */
1317 mutex_lock(&inode
->i_mutex
);
1319 /* reload atomically mode/uid/gid now that lock held */
1320 mode
= inode
->i_mode
;
1323 mutex_unlock(&inode
->i_mutex
);
1325 /* We ignore suid/sgid if there are no mappings for them in the ns */
1326 if (!kuid_has_mapping(bprm
->cred
->user_ns
, uid
) ||
1327 !kgid_has_mapping(bprm
->cred
->user_ns
, gid
))
1330 if (mode
& S_ISUID
) {
1331 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1332 bprm
->cred
->euid
= uid
;
1335 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1336 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1337 bprm
->cred
->egid
= gid
;
1342 * Fill the binprm structure from the inode.
1343 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1345 * This may be called multiple times for binary chains (scripts for example).
1347 int prepare_binprm(struct linux_binprm
*bprm
)
1351 if (bprm
->file
->f_op
== NULL
)
1354 bprm_fill_uid(bprm
);
1356 /* fill in binprm security blob */
1357 retval
= security_bprm_set_creds(bprm
);
1360 bprm
->cred_prepared
= 1;
1362 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1363 return kernel_read(bprm
->file
, 0, bprm
->buf
, BINPRM_BUF_SIZE
);
1366 EXPORT_SYMBOL(prepare_binprm
);
1369 * Arguments are '\0' separated strings found at the location bprm->p
1370 * points to; chop off the first by relocating brpm->p to right after
1371 * the first '\0' encountered.
1373 int remove_arg_zero(struct linux_binprm
*bprm
)
1376 unsigned long offset
;
1384 offset
= bprm
->p
& ~PAGE_MASK
;
1385 page
= get_arg_page(bprm
, bprm
->p
, 0);
1390 kaddr
= kmap_atomic(page
);
1392 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1393 offset
++, bprm
->p
++)
1396 kunmap_atomic(kaddr
);
1399 if (offset
== PAGE_SIZE
)
1400 free_arg_page(bprm
, (bprm
->p
>> PAGE_SHIFT
) - 1);
1401 } while (offset
== PAGE_SIZE
);
1410 EXPORT_SYMBOL(remove_arg_zero
);
1413 * cycle the list of binary formats handler, until one recognizes the image
1415 int search_binary_handler(struct linux_binprm
*bprm
)
1417 unsigned int depth
= bprm
->recursion_depth
;
1419 struct linux_binfmt
*fmt
;
1420 pid_t old_pid
, old_vpid
;
1422 /* This allows 4 levels of binfmt rewrites before failing hard. */
1426 retval
= security_bprm_check(bprm
);
1430 retval
= audit_bprm(bprm
);
1434 /* Need to fetch pid before load_binary changes it */
1435 old_pid
= current
->pid
;
1437 old_vpid
= task_pid_nr_ns(current
, task_active_pid_ns(current
->parent
));
1441 for (try=0; try<2; try++) {
1442 read_lock(&binfmt_lock
);
1443 list_for_each_entry(fmt
, &formats
, lh
) {
1444 int (*fn
)(struct linux_binprm
*) = fmt
->load_binary
;
1447 if (!try_module_get(fmt
->module
))
1449 read_unlock(&binfmt_lock
);
1450 bprm
->recursion_depth
= depth
+ 1;
1452 bprm
->recursion_depth
= depth
;
1455 trace_sched_process_exec(current
, old_pid
, bprm
);
1456 ptrace_event(PTRACE_EVENT_EXEC
, old_vpid
);
1459 allow_write_access(bprm
->file
);
1463 current
->did_exec
= 1;
1464 proc_exec_connector(current
);
1467 read_lock(&binfmt_lock
);
1469 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
)
1472 read_unlock(&binfmt_lock
);
1476 read_unlock(&binfmt_lock
);
1477 #ifdef CONFIG_MODULES
1478 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
) {
1481 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1482 if (printable(bprm
->buf
[0]) &&
1483 printable(bprm
->buf
[1]) &&
1484 printable(bprm
->buf
[2]) &&
1485 printable(bprm
->buf
[3]))
1486 break; /* -ENOEXEC */
1488 break; /* -ENOEXEC */
1489 request_module("binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
1498 EXPORT_SYMBOL(search_binary_handler
);
1501 * sys_execve() executes a new program.
1503 static int do_execve_common(const char *filename
,
1504 struct user_arg_ptr argv
,
1505 struct user_arg_ptr envp
)
1507 struct linux_binprm
*bprm
;
1509 struct files_struct
*displaced
;
1512 const struct cred
*cred
= current_cred();
1515 * We move the actual failure in case of RLIMIT_NPROC excess from
1516 * set*uid() to execve() because too many poorly written programs
1517 * don't check setuid() return code. Here we additionally recheck
1518 * whether NPROC limit is still exceeded.
1520 if ((current
->flags
& PF_NPROC_EXCEEDED
) &&
1521 atomic_read(&cred
->user
->processes
) > rlimit(RLIMIT_NPROC
)) {
1526 /* We're below the limit (still or again), so we don't want to make
1527 * further execve() calls fail. */
1528 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1530 retval
= unshare_files(&displaced
);
1535 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1539 retval
= prepare_bprm_creds(bprm
);
1543 retval
= check_unsafe_exec(bprm
);
1546 clear_in_exec
= retval
;
1547 current
->in_execve
= 1;
1549 file
= open_exec(filename
);
1550 retval
= PTR_ERR(file
);
1557 bprm
->filename
= filename
;
1558 bprm
->interp
= filename
;
1560 retval
= bprm_mm_init(bprm
);
1564 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
1565 if ((retval
= bprm
->argc
) < 0)
1568 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
1569 if ((retval
= bprm
->envc
) < 0)
1572 retval
= prepare_binprm(bprm
);
1576 retval
= copy_strings_kernel(1, &bprm
->filename
, bprm
);
1580 bprm
->exec
= bprm
->p
;
1581 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1585 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1589 retval
= search_binary_handler(bprm
);
1593 /* execve succeeded */
1594 current
->fs
->in_exec
= 0;
1595 current
->in_execve
= 0;
1596 acct_update_integrals(current
);
1599 put_files_struct(displaced
);
1604 acct_arg_size(bprm
, 0);
1610 allow_write_access(bprm
->file
);
1616 current
->fs
->in_exec
= 0;
1617 current
->in_execve
= 0;
1624 reset_files_struct(displaced
);
1629 int do_execve(const char *filename
,
1630 const char __user
*const __user
*__argv
,
1631 const char __user
*const __user
*__envp
)
1633 struct user_arg_ptr argv
= { .ptr
.native
= __argv
};
1634 struct user_arg_ptr envp
= { .ptr
.native
= __envp
};
1635 return do_execve_common(filename
, argv
, envp
);
1638 #ifdef CONFIG_COMPAT
1639 static int compat_do_execve(const char *filename
,
1640 const compat_uptr_t __user
*__argv
,
1641 const compat_uptr_t __user
*__envp
)
1643 struct user_arg_ptr argv
= {
1645 .ptr
.compat
= __argv
,
1647 struct user_arg_ptr envp
= {
1649 .ptr
.compat
= __envp
,
1651 return do_execve_common(filename
, argv
, envp
);
1655 void set_binfmt(struct linux_binfmt
*new)
1657 struct mm_struct
*mm
= current
->mm
;
1660 module_put(mm
->binfmt
->module
);
1664 __module_get(new->module
);
1667 EXPORT_SYMBOL(set_binfmt
);
1670 * set_dumpable converts traditional three-value dumpable to two flags and
1671 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1672 * these bits are not changed atomically. So get_dumpable can observe the
1673 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1674 * return either old dumpable or new one by paying attention to the order of
1675 * modifying the bits.
1677 * dumpable | mm->flags (binary)
1678 * old new | initial interim final
1679 * ---------+-----------------------
1687 * (*) get_dumpable regards interim value of 10 as 11.
1689 void set_dumpable(struct mm_struct
*mm
, int value
)
1692 case SUID_DUMP_DISABLE
:
1693 clear_bit(MMF_DUMPABLE
, &mm
->flags
);
1695 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1697 case SUID_DUMP_USER
:
1698 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1700 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1702 case SUID_DUMP_ROOT
:
1703 set_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1705 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1710 int __get_dumpable(unsigned long mm_flags
)
1714 ret
= mm_flags
& MMF_DUMPABLE_MASK
;
1715 return (ret
> SUID_DUMP_USER
) ? SUID_DUMP_ROOT
: ret
;
1719 * This returns the actual value of the suid_dumpable flag. For things
1720 * that are using this for checking for privilege transitions, it must
1721 * test against SUID_DUMP_USER rather than treating it as a boolean
1724 int get_dumpable(struct mm_struct
*mm
)
1726 return __get_dumpable(mm
->flags
);
1729 SYSCALL_DEFINE3(execve
,
1730 const char __user
*, filename
,
1731 const char __user
*const __user
*, argv
,
1732 const char __user
*const __user
*, envp
)
1734 struct filename
*path
= getname(filename
);
1735 int error
= PTR_ERR(path
);
1736 if (!IS_ERR(path
)) {
1737 error
= do_execve(path
->name
, argv
, envp
);
1742 #ifdef CONFIG_COMPAT
1743 asmlinkage
long compat_sys_execve(const char __user
* filename
,
1744 const compat_uptr_t __user
* argv
,
1745 const compat_uptr_t __user
* envp
)
1747 struct filename
*path
= getname(filename
);
1748 int error
= PTR_ERR(path
);
1749 if (!IS_ERR(path
)) {
1750 error
= compat_do_execve(path
->name
, argv
, envp
);