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/smp_lock.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_counter.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/proc_fs.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/delay.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54 #include <linux/tracehook.h>
55 #include <linux/kmod.h>
56 #include <linux/fsnotify.h>
58 #include <asm/uaccess.h>
59 #include <asm/mmu_context.h>
64 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
65 int suid_dumpable
= 0;
67 /* The maximal length of core_pattern is also specified in sysctl.c */
69 static LIST_HEAD(formats
);
70 static DEFINE_RWLOCK(binfmt_lock
);
72 int register_binfmt(struct linux_binfmt
* fmt
)
76 write_lock(&binfmt_lock
);
77 list_add(&fmt
->lh
, &formats
);
78 write_unlock(&binfmt_lock
);
82 EXPORT_SYMBOL(register_binfmt
);
84 void unregister_binfmt(struct linux_binfmt
* fmt
)
86 write_lock(&binfmt_lock
);
88 write_unlock(&binfmt_lock
);
91 EXPORT_SYMBOL(unregister_binfmt
);
93 static inline void put_binfmt(struct linux_binfmt
* fmt
)
95 module_put(fmt
->module
);
99 * Note that a shared library must be both readable and executable due to
102 * Also note that we take the address to load from from the file itself.
104 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
108 char *tmp
= getname(library
);
109 int error
= PTR_ERR(tmp
);
112 error
= path_lookup_open(AT_FDCWD
, tmp
,
114 FMODE_READ
|FMODE_EXEC
);
121 if (!S_ISREG(nd
.path
.dentry
->d_inode
->i_mode
))
125 if (nd
.path
.mnt
->mnt_flags
& MNT_NOEXEC
)
128 error
= inode_permission(nd
.path
.dentry
->d_inode
,
129 MAY_READ
| MAY_EXEC
| MAY_OPEN
);
133 file
= nameidata_to_filp(&nd
, O_RDONLY
|O_LARGEFILE
);
134 error
= PTR_ERR(file
);
138 fsnotify_open(file
->f_path
.dentry
);
142 struct linux_binfmt
* fmt
;
144 read_lock(&binfmt_lock
);
145 list_for_each_entry(fmt
, &formats
, lh
) {
146 if (!fmt
->load_shlib
)
148 if (!try_module_get(fmt
->module
))
150 read_unlock(&binfmt_lock
);
151 error
= fmt
->load_shlib(file
);
152 read_lock(&binfmt_lock
);
154 if (error
!= -ENOEXEC
)
157 read_unlock(&binfmt_lock
);
163 release_open_intent(&nd
);
170 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
176 #ifdef CONFIG_STACK_GROWSUP
178 ret
= expand_stack_downwards(bprm
->vma
, pos
);
183 ret
= get_user_pages(current
, bprm
->mm
, pos
,
184 1, write
, 1, &page
, NULL
);
189 unsigned long size
= bprm
->vma
->vm_end
- bprm
->vma
->vm_start
;
193 * We've historically supported up to 32 pages (ARG_MAX)
194 * of argument strings even with small stacks
200 * Limit to 1/4-th the stack size for the argv+env strings.
202 * - the remaining binfmt code will not run out of stack space,
203 * - the program will have a reasonable amount of stack left
206 rlim
= current
->signal
->rlim
;
207 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
/ 4) {
216 static void put_arg_page(struct page
*page
)
221 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
225 static void free_arg_pages(struct linux_binprm
*bprm
)
229 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
232 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
235 static int __bprm_mm_init(struct linux_binprm
*bprm
)
238 struct vm_area_struct
*vma
= NULL
;
239 struct mm_struct
*mm
= bprm
->mm
;
241 bprm
->vma
= vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
245 down_write(&mm
->mmap_sem
);
249 * Place the stack at the largest stack address the architecture
250 * supports. Later, we'll move this to an appropriate place. We don't
251 * use STACK_TOP because that can depend on attributes which aren't
254 vma
->vm_end
= STACK_TOP_MAX
;
255 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
256 vma
->vm_flags
= VM_STACK_FLAGS
;
257 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
258 err
= insert_vm_struct(mm
, vma
);
262 mm
->stack_vm
= mm
->total_vm
= 1;
263 up_write(&mm
->mmap_sem
);
264 bprm
->p
= vma
->vm_end
- sizeof(void *);
267 up_write(&mm
->mmap_sem
);
269 kmem_cache_free(vm_area_cachep
, vma
);
273 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
275 return len
<= MAX_ARG_STRLEN
;
280 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
285 page
= bprm
->page
[pos
/ PAGE_SIZE
];
286 if (!page
&& write
) {
287 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
290 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
296 static void put_arg_page(struct page
*page
)
300 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
303 __free_page(bprm
->page
[i
]);
304 bprm
->page
[i
] = NULL
;
308 static void free_arg_pages(struct linux_binprm
*bprm
)
312 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
313 free_arg_page(bprm
, i
);
316 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
321 static int __bprm_mm_init(struct linux_binprm
*bprm
)
323 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
327 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
329 return len
<= bprm
->p
;
332 #endif /* CONFIG_MMU */
335 * Create a new mm_struct and populate it with a temporary stack
336 * vm_area_struct. We don't have enough context at this point to set the stack
337 * flags, permissions, and offset, so we use temporary values. We'll update
338 * them later in setup_arg_pages().
340 int bprm_mm_init(struct linux_binprm
*bprm
)
343 struct mm_struct
*mm
= NULL
;
345 bprm
->mm
= mm
= mm_alloc();
350 err
= init_new_context(current
, mm
);
354 err
= __bprm_mm_init(bprm
);
370 * count() counts the number of strings in array ARGV.
372 static int count(char __user
* __user
* argv
, int max
)
380 if (get_user(p
, argv
))
394 * 'copy_strings()' copies argument/environment strings from the old
395 * processes's memory to the new process's stack. The call to get_user_pages()
396 * ensures the destination page is created and not swapped out.
398 static int copy_strings(int argc
, char __user
* __user
* argv
,
399 struct linux_binprm
*bprm
)
401 struct page
*kmapped_page
= NULL
;
403 unsigned long kpos
= 0;
411 if (get_user(str
, argv
+argc
) ||
412 !(len
= strnlen_user(str
, MAX_ARG_STRLEN
))) {
417 if (!valid_arg_len(bprm
, len
)) {
422 /* We're going to work our way backwords. */
428 int offset
, bytes_to_copy
;
430 offset
= pos
% PAGE_SIZE
;
434 bytes_to_copy
= offset
;
435 if (bytes_to_copy
> len
)
438 offset
-= bytes_to_copy
;
439 pos
-= bytes_to_copy
;
440 str
-= bytes_to_copy
;
441 len
-= bytes_to_copy
;
443 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
446 page
= get_arg_page(bprm
, pos
, 1);
453 flush_kernel_dcache_page(kmapped_page
);
454 kunmap(kmapped_page
);
455 put_arg_page(kmapped_page
);
458 kaddr
= kmap(kmapped_page
);
459 kpos
= pos
& PAGE_MASK
;
460 flush_arg_page(bprm
, kpos
, kmapped_page
);
462 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
471 flush_kernel_dcache_page(kmapped_page
);
472 kunmap(kmapped_page
);
473 put_arg_page(kmapped_page
);
479 * Like copy_strings, but get argv and its values from kernel memory.
481 int copy_strings_kernel(int argc
,char ** argv
, struct linux_binprm
*bprm
)
484 mm_segment_t oldfs
= get_fs();
486 r
= copy_strings(argc
, (char __user
* __user
*)argv
, bprm
);
490 EXPORT_SYMBOL(copy_strings_kernel
);
495 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
496 * the binfmt code determines where the new stack should reside, we shift it to
497 * its final location. The process proceeds as follows:
499 * 1) Use shift to calculate the new vma endpoints.
500 * 2) Extend vma to cover both the old and new ranges. This ensures the
501 * arguments passed to subsequent functions are consistent.
502 * 3) Move vma's page tables to the new range.
503 * 4) Free up any cleared pgd range.
504 * 5) Shrink the vma to cover only the new range.
506 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
508 struct mm_struct
*mm
= vma
->vm_mm
;
509 unsigned long old_start
= vma
->vm_start
;
510 unsigned long old_end
= vma
->vm_end
;
511 unsigned long length
= old_end
- old_start
;
512 unsigned long new_start
= old_start
- shift
;
513 unsigned long new_end
= old_end
- shift
;
514 struct mmu_gather tlb
;
516 BUG_ON(new_start
> new_end
);
519 * ensure there are no vmas between where we want to go
522 if (vma
!= find_vma(mm
, new_start
))
526 * cover the whole range: [new_start, old_end)
528 vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
);
531 * move the page tables downwards, on failure we rely on
532 * process cleanup to remove whatever mess we made.
534 if (length
!= move_page_tables(vma
, old_start
,
535 vma
, new_start
, length
))
539 tlb_gather_mmu(&tlb
, mm
, 0);
540 if (new_end
> old_start
) {
542 * when the old and new regions overlap clear from new_end.
544 free_pgd_range(&tlb
, new_end
, old_end
, new_end
,
545 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
548 * otherwise, clean from old_start; this is done to not touch
549 * the address space in [new_end, old_start) some architectures
550 * have constraints on va-space that make this illegal (IA64) -
551 * for the others its just a little faster.
553 free_pgd_range(&tlb
, old_start
, old_end
, new_end
,
554 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
556 tlb_finish_mmu(&tlb
, new_end
, old_end
);
559 * shrink the vma to just the new range.
561 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
566 #define EXTRA_STACK_VM_PAGES 20 /* random */
569 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
570 * the stack is optionally relocated, and some extra space is added.
572 int setup_arg_pages(struct linux_binprm
*bprm
,
573 unsigned long stack_top
,
574 int executable_stack
)
577 unsigned long stack_shift
;
578 struct mm_struct
*mm
= current
->mm
;
579 struct vm_area_struct
*vma
= bprm
->vma
;
580 struct vm_area_struct
*prev
= NULL
;
581 unsigned long vm_flags
;
582 unsigned long stack_base
;
584 #ifdef CONFIG_STACK_GROWSUP
585 /* Limit stack size to 1GB */
586 stack_base
= current
->signal
->rlim
[RLIMIT_STACK
].rlim_max
;
587 if (stack_base
> (1 << 30))
588 stack_base
= 1 << 30;
590 /* Make sure we didn't let the argument array grow too large. */
591 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
594 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
596 stack_shift
= vma
->vm_start
- stack_base
;
597 mm
->arg_start
= bprm
->p
- stack_shift
;
598 bprm
->p
= vma
->vm_end
- stack_shift
;
600 stack_top
= arch_align_stack(stack_top
);
601 stack_top
= PAGE_ALIGN(stack_top
);
602 stack_shift
= vma
->vm_end
- stack_top
;
604 bprm
->p
-= stack_shift
;
605 mm
->arg_start
= bprm
->p
;
609 bprm
->loader
-= stack_shift
;
610 bprm
->exec
-= stack_shift
;
612 down_write(&mm
->mmap_sem
);
613 vm_flags
= VM_STACK_FLAGS
;
616 * Adjust stack execute permissions; explicitly enable for
617 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
618 * (arch default) otherwise.
620 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
622 else if (executable_stack
== EXSTACK_DISABLE_X
)
623 vm_flags
&= ~VM_EXEC
;
624 vm_flags
|= mm
->def_flags
;
626 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
632 /* Move stack pages down in memory. */
634 ret
= shift_arg_pages(vma
, stack_shift
);
636 up_write(&mm
->mmap_sem
);
641 #ifdef CONFIG_STACK_GROWSUP
642 stack_base
= vma
->vm_end
+ EXTRA_STACK_VM_PAGES
* PAGE_SIZE
;
644 stack_base
= vma
->vm_start
- EXTRA_STACK_VM_PAGES
* PAGE_SIZE
;
646 ret
= expand_stack(vma
, stack_base
);
651 up_write(&mm
->mmap_sem
);
654 EXPORT_SYMBOL(setup_arg_pages
);
656 #endif /* CONFIG_MMU */
658 struct file
*open_exec(const char *name
)
664 err
= path_lookup_open(AT_FDCWD
, name
, LOOKUP_FOLLOW
, &nd
,
665 FMODE_READ
|FMODE_EXEC
);
670 if (!S_ISREG(nd
.path
.dentry
->d_inode
->i_mode
))
673 if (nd
.path
.mnt
->mnt_flags
& MNT_NOEXEC
)
676 err
= inode_permission(nd
.path
.dentry
->d_inode
, MAY_EXEC
| MAY_OPEN
);
680 file
= nameidata_to_filp(&nd
, O_RDONLY
|O_LARGEFILE
);
684 fsnotify_open(file
->f_path
.dentry
);
686 err
= deny_write_access(file
);
695 release_open_intent(&nd
);
700 EXPORT_SYMBOL(open_exec
);
702 int kernel_read(struct file
*file
, unsigned long offset
,
703 char *addr
, unsigned long count
)
711 /* The cast to a user pointer is valid due to the set_fs() */
712 result
= vfs_read(file
, (void __user
*)addr
, count
, &pos
);
717 EXPORT_SYMBOL(kernel_read
);
719 static int exec_mmap(struct mm_struct
*mm
)
721 struct task_struct
*tsk
;
722 struct mm_struct
* old_mm
, *active_mm
;
724 /* Notify parent that we're no longer interested in the old VM */
726 old_mm
= current
->mm
;
727 mm_release(tsk
, old_mm
);
731 * Make sure that if there is a core dump in progress
732 * for the old mm, we get out and die instead of going
733 * through with the exec. We must hold mmap_sem around
734 * checking core_state and changing tsk->mm.
736 down_read(&old_mm
->mmap_sem
);
737 if (unlikely(old_mm
->core_state
)) {
738 up_read(&old_mm
->mmap_sem
);
744 active_mm
= tsk
->active_mm
;
745 activate_mm(active_mm
, mm
);
750 arch_pick_mmap_layout(mm
);
752 up_read(&old_mm
->mmap_sem
);
753 BUG_ON(active_mm
!= old_mm
);
754 mm_update_next_owner(old_mm
);
763 * This function makes sure the current process has its own signal table,
764 * so that flush_signal_handlers can later reset the handlers without
765 * disturbing other processes. (Other processes might share the signal
766 * table via the CLONE_SIGHAND option to clone().)
768 static int de_thread(struct task_struct
*tsk
)
770 struct signal_struct
*sig
= tsk
->signal
;
771 struct sighand_struct
*oldsighand
= tsk
->sighand
;
772 spinlock_t
*lock
= &oldsighand
->siglock
;
775 if (thread_group_empty(tsk
))
776 goto no_thread_group
;
779 * Kill all other threads in the thread group.
782 if (signal_group_exit(sig
)) {
784 * Another group action in progress, just
785 * return so that the signal is processed.
787 spin_unlock_irq(lock
);
790 sig
->group_exit_task
= tsk
;
791 zap_other_threads(tsk
);
793 /* Account for the thread group leader hanging around: */
794 count
= thread_group_leader(tsk
) ? 1 : 2;
795 sig
->notify_count
= count
;
796 while (atomic_read(&sig
->count
) > count
) {
797 __set_current_state(TASK_UNINTERRUPTIBLE
);
798 spin_unlock_irq(lock
);
802 spin_unlock_irq(lock
);
805 * At this point all other threads have exited, all we have to
806 * do is to wait for the thread group leader to become inactive,
807 * and to assume its PID:
809 if (!thread_group_leader(tsk
)) {
810 struct task_struct
*leader
= tsk
->group_leader
;
812 sig
->notify_count
= -1; /* for exit_notify() */
814 write_lock_irq(&tasklist_lock
);
815 if (likely(leader
->exit_state
))
817 __set_current_state(TASK_UNINTERRUPTIBLE
);
818 write_unlock_irq(&tasklist_lock
);
823 * The only record we have of the real-time age of a
824 * process, regardless of execs it's done, is start_time.
825 * All the past CPU time is accumulated in signal_struct
826 * from sister threads now dead. But in this non-leader
827 * exec, nothing survives from the original leader thread,
828 * whose birth marks the true age of this process now.
829 * When we take on its identity by switching to its PID, we
830 * also take its birthdate (always earlier than our own).
832 tsk
->start_time
= leader
->start_time
;
834 BUG_ON(!same_thread_group(leader
, tsk
));
835 BUG_ON(has_group_leader_pid(tsk
));
837 * An exec() starts a new thread group with the
838 * TGID of the previous thread group. Rehash the
839 * two threads with a switched PID, and release
840 * the former thread group leader:
843 /* Become a process group leader with the old leader's pid.
844 * The old leader becomes a thread of the this thread group.
845 * Note: The old leader also uses this pid until release_task
846 * is called. Odd but simple and correct.
848 detach_pid(tsk
, PIDTYPE_PID
);
849 tsk
->pid
= leader
->pid
;
850 attach_pid(tsk
, PIDTYPE_PID
, task_pid(leader
));
851 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
852 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
853 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
855 tsk
->group_leader
= tsk
;
856 leader
->group_leader
= tsk
;
858 tsk
->exit_signal
= SIGCHLD
;
860 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
861 leader
->exit_state
= EXIT_DEAD
;
862 write_unlock_irq(&tasklist_lock
);
864 release_task(leader
);
867 sig
->group_exit_task
= NULL
;
868 sig
->notify_count
= 0;
872 flush_itimer_signals();
874 if (atomic_read(&oldsighand
->count
) != 1) {
875 struct sighand_struct
*newsighand
;
877 * This ->sighand is shared with the CLONE_SIGHAND
878 * but not CLONE_THREAD task, switch to the new one.
880 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
884 atomic_set(&newsighand
->count
, 1);
885 memcpy(newsighand
->action
, oldsighand
->action
,
886 sizeof(newsighand
->action
));
888 write_lock_irq(&tasklist_lock
);
889 spin_lock(&oldsighand
->siglock
);
890 rcu_assign_pointer(tsk
->sighand
, newsighand
);
891 spin_unlock(&oldsighand
->siglock
);
892 write_unlock_irq(&tasklist_lock
);
894 __cleanup_sighand(oldsighand
);
897 BUG_ON(!thread_group_leader(tsk
));
902 * These functions flushes out all traces of the currently running executable
903 * so that a new one can be started
905 static void flush_old_files(struct files_struct
* files
)
910 spin_lock(&files
->file_lock
);
912 unsigned long set
, i
;
916 fdt
= files_fdtable(files
);
917 if (i
>= fdt
->max_fds
)
919 set
= fdt
->close_on_exec
->fds_bits
[j
];
922 fdt
->close_on_exec
->fds_bits
[j
] = 0;
923 spin_unlock(&files
->file_lock
);
924 for ( ; set
; i
++,set
>>= 1) {
929 spin_lock(&files
->file_lock
);
932 spin_unlock(&files
->file_lock
);
935 char *get_task_comm(char *buf
, struct task_struct
*tsk
)
937 /* buf must be at least sizeof(tsk->comm) in size */
939 strncpy(buf
, tsk
->comm
, sizeof(tsk
->comm
));
944 void set_task_comm(struct task_struct
*tsk
, char *buf
)
947 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
951 int flush_old_exec(struct linux_binprm
* bprm
)
955 char tcomm
[sizeof(current
->comm
)];
958 * Make sure we have a private signal table and that
959 * we are unassociated from the previous thread group.
961 retval
= de_thread(current
);
965 set_mm_exe_file(bprm
->mm
, bprm
->file
);
968 * Release all of the old mmap stuff
970 retval
= exec_mmap(bprm
->mm
);
974 bprm
->mm
= NULL
; /* We're using it now */
976 /* This is the point of no return */
977 current
->sas_ss_sp
= current
->sas_ss_size
= 0;
979 if (current_euid() == current_uid() && current_egid() == current_gid())
980 set_dumpable(current
->mm
, 1);
982 set_dumpable(current
->mm
, suid_dumpable
);
984 name
= bprm
->filename
;
986 /* Copies the binary name from after last slash */
987 for (i
=0; (ch
= *(name
++)) != '\0';) {
989 i
= 0; /* overwrite what we wrote */
991 if (i
< (sizeof(tcomm
) - 1))
995 set_task_comm(current
, tcomm
);
997 current
->flags
&= ~PF_RANDOMIZE
;
1000 /* Set the new mm task size. We have to do that late because it may
1001 * depend on TIF_32BIT which is only updated in flush_thread() on
1002 * some architectures like powerpc
1004 current
->mm
->task_size
= TASK_SIZE
;
1006 /* install the new credentials */
1007 if (bprm
->cred
->uid
!= current_euid() ||
1008 bprm
->cred
->gid
!= current_egid()) {
1009 current
->pdeath_signal
= 0;
1010 } else if (file_permission(bprm
->file
, MAY_READ
) ||
1011 bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
) {
1012 set_dumpable(current
->mm
, suid_dumpable
);
1015 current
->personality
&= ~bprm
->per_clear
;
1018 * Flush performance counters when crossing a
1021 if (!get_dumpable(current
->mm
))
1022 perf_counter_exit_task(current
);
1024 /* An exec changes our domain. We are no longer part of the thread
1027 current
->self_exec_id
++;
1029 flush_signal_handlers(current
, 0);
1030 flush_old_files(current
->files
);
1038 EXPORT_SYMBOL(flush_old_exec
);
1041 * install the new credentials for this executable
1043 void install_exec_creds(struct linux_binprm
*bprm
)
1045 security_bprm_committing_creds(bprm
);
1047 commit_creds(bprm
->cred
);
1050 /* cred_exec_mutex must be held at least to this point to prevent
1051 * ptrace_attach() from altering our determination of the task's
1052 * credentials; any time after this it may be unlocked */
1054 security_bprm_committed_creds(bprm
);
1056 EXPORT_SYMBOL(install_exec_creds
);
1059 * determine how safe it is to execute the proposed program
1060 * - the caller must hold current->cred_exec_mutex to protect against
1063 int check_unsafe_exec(struct linux_binprm
*bprm
)
1065 struct task_struct
*p
= current
, *t
;
1069 bprm
->unsafe
= tracehook_unsafe_exec(p
);
1072 write_lock(&p
->fs
->lock
);
1074 for (t
= next_thread(p
); t
!= p
; t
= next_thread(t
)) {
1080 if (p
->fs
->users
> n_fs
) {
1081 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1084 if (!p
->fs
->in_exec
) {
1089 write_unlock(&p
->fs
->lock
);
1095 * Fill the binprm structure from the inode.
1096 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1098 * This may be called multiple times for binary chains (scripts for example).
1100 int prepare_binprm(struct linux_binprm
*bprm
)
1103 struct inode
* inode
= bprm
->file
->f_path
.dentry
->d_inode
;
1106 mode
= inode
->i_mode
;
1107 if (bprm
->file
->f_op
== NULL
)
1110 /* clear any previous set[ug]id data from a previous binary */
1111 bprm
->cred
->euid
= current_euid();
1112 bprm
->cred
->egid
= current_egid();
1114 if (!(bprm
->file
->f_path
.mnt
->mnt_flags
& MNT_NOSUID
)) {
1116 if (mode
& S_ISUID
) {
1117 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1118 bprm
->cred
->euid
= inode
->i_uid
;
1123 * If setgid is set but no group execute bit then this
1124 * is a candidate for mandatory locking, not a setgid
1127 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1128 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1129 bprm
->cred
->egid
= inode
->i_gid
;
1133 /* fill in binprm security blob */
1134 retval
= security_bprm_set_creds(bprm
);
1137 bprm
->cred_prepared
= 1;
1139 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1140 return kernel_read(bprm
->file
, 0, bprm
->buf
, BINPRM_BUF_SIZE
);
1143 EXPORT_SYMBOL(prepare_binprm
);
1146 * Arguments are '\0' separated strings found at the location bprm->p
1147 * points to; chop off the first by relocating brpm->p to right after
1148 * the first '\0' encountered.
1150 int remove_arg_zero(struct linux_binprm
*bprm
)
1153 unsigned long offset
;
1161 offset
= bprm
->p
& ~PAGE_MASK
;
1162 page
= get_arg_page(bprm
, bprm
->p
, 0);
1167 kaddr
= kmap_atomic(page
, KM_USER0
);
1169 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1170 offset
++, bprm
->p
++)
1173 kunmap_atomic(kaddr
, KM_USER0
);
1176 if (offset
== PAGE_SIZE
)
1177 free_arg_page(bprm
, (bprm
->p
>> PAGE_SHIFT
) - 1);
1178 } while (offset
== PAGE_SIZE
);
1187 EXPORT_SYMBOL(remove_arg_zero
);
1190 * cycle the list of binary formats handler, until one recognizes the image
1192 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
1194 unsigned int depth
= bprm
->recursion_depth
;
1196 struct linux_binfmt
*fmt
;
1198 retval
= security_bprm_check(bprm
);
1202 /* kernel module loader fixup */
1203 /* so we don't try to load run modprobe in kernel space. */
1206 retval
= audit_bprm(bprm
);
1211 for (try=0; try<2; try++) {
1212 read_lock(&binfmt_lock
);
1213 list_for_each_entry(fmt
, &formats
, lh
) {
1214 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
1217 if (!try_module_get(fmt
->module
))
1219 read_unlock(&binfmt_lock
);
1220 retval
= fn(bprm
, regs
);
1222 * Restore the depth counter to its starting value
1223 * in this call, so we don't have to rely on every
1224 * load_binary function to restore it on return.
1226 bprm
->recursion_depth
= depth
;
1229 tracehook_report_exec(fmt
, bprm
, regs
);
1231 allow_write_access(bprm
->file
);
1235 current
->did_exec
= 1;
1236 proc_exec_connector(current
);
1239 read_lock(&binfmt_lock
);
1241 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
)
1244 read_unlock(&binfmt_lock
);
1248 read_unlock(&binfmt_lock
);
1249 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
) {
1251 #ifdef CONFIG_MODULES
1253 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1254 if (printable(bprm
->buf
[0]) &&
1255 printable(bprm
->buf
[1]) &&
1256 printable(bprm
->buf
[2]) &&
1257 printable(bprm
->buf
[3]))
1258 break; /* -ENOEXEC */
1259 request_module("binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
1266 EXPORT_SYMBOL(search_binary_handler
);
1268 void free_bprm(struct linux_binprm
*bprm
)
1270 free_arg_pages(bprm
);
1272 abort_creds(bprm
->cred
);
1277 * sys_execve() executes a new program.
1279 int do_execve(char * filename
,
1280 char __user
*__user
*argv
,
1281 char __user
*__user
*envp
,
1282 struct pt_regs
* regs
)
1284 struct linux_binprm
*bprm
;
1286 struct files_struct
*displaced
;
1290 retval
= unshare_files(&displaced
);
1295 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1299 retval
= mutex_lock_interruptible(¤t
->cred_exec_mutex
);
1304 bprm
->cred
= prepare_exec_creds();
1308 retval
= check_unsafe_exec(bprm
);
1311 clear_in_exec
= retval
;
1313 file
= open_exec(filename
);
1314 retval
= PTR_ERR(file
);
1321 bprm
->filename
= filename
;
1322 bprm
->interp
= filename
;
1324 retval
= bprm_mm_init(bprm
);
1328 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
1329 if ((retval
= bprm
->argc
) < 0)
1332 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
1333 if ((retval
= bprm
->envc
) < 0)
1336 retval
= prepare_binprm(bprm
);
1340 retval
= copy_strings_kernel(1, &bprm
->filename
, bprm
);
1344 bprm
->exec
= bprm
->p
;
1345 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1349 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1353 current
->flags
&= ~PF_KTHREAD
;
1354 retval
= search_binary_handler(bprm
,regs
);
1358 /* execve succeeded */
1359 current
->fs
->in_exec
= 0;
1360 mutex_unlock(¤t
->cred_exec_mutex
);
1361 acct_update_integrals(current
);
1364 put_files_struct(displaced
);
1373 allow_write_access(bprm
->file
);
1379 current
->fs
->in_exec
= 0;
1382 mutex_unlock(¤t
->cred_exec_mutex
);
1389 reset_files_struct(displaced
);
1394 int set_binfmt(struct linux_binfmt
*new)
1396 struct linux_binfmt
*old
= current
->binfmt
;
1399 if (!try_module_get(new->module
))
1402 current
->binfmt
= new;
1404 module_put(old
->module
);
1408 EXPORT_SYMBOL(set_binfmt
);
1410 /* format_corename will inspect the pattern parameter, and output a
1411 * name into corename, which must have space for at least
1412 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1414 static int format_corename(char *corename
, long signr
)
1416 const struct cred
*cred
= current_cred();
1417 const char *pat_ptr
= core_pattern
;
1418 int ispipe
= (*pat_ptr
== '|');
1419 char *out_ptr
= corename
;
1420 char *const out_end
= corename
+ CORENAME_MAX_SIZE
;
1422 int pid_in_pattern
= 0;
1424 /* Repeat as long as we have more pattern to process and more output
1427 if (*pat_ptr
!= '%') {
1428 if (out_ptr
== out_end
)
1430 *out_ptr
++ = *pat_ptr
++;
1432 switch (*++pat_ptr
) {
1435 /* Double percent, output one percent */
1437 if (out_ptr
== out_end
)
1444 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1445 "%d", task_tgid_vnr(current
));
1446 if (rc
> out_end
- out_ptr
)
1452 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1454 if (rc
> out_end
- out_ptr
)
1460 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1462 if (rc
> out_end
- out_ptr
)
1466 /* signal that caused the coredump */
1468 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1470 if (rc
> out_end
- out_ptr
)
1474 /* UNIX time of coredump */
1477 do_gettimeofday(&tv
);
1478 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1480 if (rc
> out_end
- out_ptr
)
1487 down_read(&uts_sem
);
1488 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1489 "%s", utsname()->nodename
);
1491 if (rc
> out_end
- out_ptr
)
1497 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1498 "%s", current
->comm
);
1499 if (rc
> out_end
- out_ptr
)
1503 /* core limit size */
1505 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1506 "%lu", current
->signal
->rlim
[RLIMIT_CORE
].rlim_cur
);
1507 if (rc
> out_end
- out_ptr
)
1517 /* Backward compatibility with core_uses_pid:
1519 * If core_pattern does not include a %p (as is the default)
1520 * and core_uses_pid is set, then .%pid will be appended to
1521 * the filename. Do not do this for piped commands. */
1522 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
1523 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1524 ".%d", task_tgid_vnr(current
));
1525 if (rc
> out_end
- out_ptr
)
1534 static int zap_process(struct task_struct
*start
)
1536 struct task_struct
*t
;
1539 start
->signal
->flags
= SIGNAL_GROUP_EXIT
;
1540 start
->signal
->group_stop_count
= 0;
1544 if (t
!= current
&& t
->mm
) {
1545 sigaddset(&t
->pending
.signal
, SIGKILL
);
1546 signal_wake_up(t
, 1);
1549 } while_each_thread(start
, t
);
1554 static inline int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
1555 struct core_state
*core_state
, int exit_code
)
1557 struct task_struct
*g
, *p
;
1558 unsigned long flags
;
1561 spin_lock_irq(&tsk
->sighand
->siglock
);
1562 if (!signal_group_exit(tsk
->signal
)) {
1563 mm
->core_state
= core_state
;
1564 tsk
->signal
->group_exit_code
= exit_code
;
1565 nr
= zap_process(tsk
);
1567 spin_unlock_irq(&tsk
->sighand
->siglock
);
1568 if (unlikely(nr
< 0))
1571 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
1574 * We should find and kill all tasks which use this mm, and we should
1575 * count them correctly into ->nr_threads. We don't take tasklist
1576 * lock, but this is safe wrt:
1579 * None of sub-threads can fork after zap_process(leader). All
1580 * processes which were created before this point should be
1581 * visible to zap_threads() because copy_process() adds the new
1582 * process to the tail of init_task.tasks list, and lock/unlock
1583 * of ->siglock provides a memory barrier.
1586 * The caller holds mm->mmap_sem. This means that the task which
1587 * uses this mm can't pass exit_mm(), so it can't exit or clear
1591 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1592 * we must see either old or new leader, this does not matter.
1593 * However, it can change p->sighand, so lock_task_sighand(p)
1594 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1597 * Note also that "g" can be the old leader with ->mm == NULL
1598 * and already unhashed and thus removed from ->thread_group.
1599 * This is OK, __unhash_process()->list_del_rcu() does not
1600 * clear the ->next pointer, we will find the new leader via
1604 for_each_process(g
) {
1605 if (g
== tsk
->group_leader
)
1607 if (g
->flags
& PF_KTHREAD
)
1612 if (unlikely(p
->mm
== mm
)) {
1613 lock_task_sighand(p
, &flags
);
1614 nr
+= zap_process(p
);
1615 unlock_task_sighand(p
, &flags
);
1619 } while_each_thread(g
, p
);
1623 atomic_set(&core_state
->nr_threads
, nr
);
1627 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
1629 struct task_struct
*tsk
= current
;
1630 struct mm_struct
*mm
= tsk
->mm
;
1631 struct completion
*vfork_done
;
1634 init_completion(&core_state
->startup
);
1635 core_state
->dumper
.task
= tsk
;
1636 core_state
->dumper
.next
= NULL
;
1637 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
1638 up_write(&mm
->mmap_sem
);
1640 if (unlikely(core_waiters
< 0))
1644 * Make sure nobody is waiting for us to release the VM,
1645 * otherwise we can deadlock when we wait on each other
1647 vfork_done
= tsk
->vfork_done
;
1649 tsk
->vfork_done
= NULL
;
1650 complete(vfork_done
);
1654 wait_for_completion(&core_state
->startup
);
1656 return core_waiters
;
1659 static void coredump_finish(struct mm_struct
*mm
)
1661 struct core_thread
*curr
, *next
;
1662 struct task_struct
*task
;
1664 next
= mm
->core_state
->dumper
.next
;
1665 while ((curr
= next
) != NULL
) {
1669 * see exit_mm(), curr->task must not see
1670 * ->task == NULL before we read ->next.
1674 wake_up_process(task
);
1677 mm
->core_state
= NULL
;
1681 * set_dumpable converts traditional three-value dumpable to two flags and
1682 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1683 * these bits are not changed atomically. So get_dumpable can observe the
1684 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1685 * return either old dumpable or new one by paying attention to the order of
1686 * modifying the bits.
1688 * dumpable | mm->flags (binary)
1689 * old new | initial interim final
1690 * ---------+-----------------------
1698 * (*) get_dumpable regards interim value of 10 as 11.
1700 void set_dumpable(struct mm_struct
*mm
, int value
)
1704 clear_bit(MMF_DUMPABLE
, &mm
->flags
);
1706 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1709 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1711 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1714 set_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1716 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1721 int get_dumpable(struct mm_struct
*mm
)
1725 ret
= mm
->flags
& 0x3;
1726 return (ret
>= 2) ? 2 : ret
;
1729 void do_coredump(long signr
, int exit_code
, struct pt_regs
*regs
)
1731 struct core_state core_state
;
1732 char corename
[CORENAME_MAX_SIZE
+ 1];
1733 struct mm_struct
*mm
= current
->mm
;
1734 struct linux_binfmt
* binfmt
;
1735 struct inode
* inode
;
1737 const struct cred
*old_cred
;
1742 unsigned long core_limit
= current
->signal
->rlim
[RLIMIT_CORE
].rlim_cur
;
1743 char **helper_argv
= NULL
;
1744 int helper_argc
= 0;
1747 audit_core_dumps(signr
);
1749 binfmt
= current
->binfmt
;
1750 if (!binfmt
|| !binfmt
->core_dump
)
1753 cred
= prepare_creds();
1759 down_write(&mm
->mmap_sem
);
1761 * If another thread got here first, or we are not dumpable, bail out.
1763 if (mm
->core_state
|| !get_dumpable(mm
)) {
1764 up_write(&mm
->mmap_sem
);
1770 * We cannot trust fsuid as being the "true" uid of the
1771 * process nor do we know its entire history. We only know it
1772 * was tainted so we dump it as root in mode 2.
1774 if (get_dumpable(mm
) == 2) { /* Setuid core dump mode */
1775 flag
= O_EXCL
; /* Stop rewrite attacks */
1776 cred
->fsuid
= 0; /* Dump root private */
1779 retval
= coredump_wait(exit_code
, &core_state
);
1785 old_cred
= override_creds(cred
);
1788 * Clear any false indication of pending signals that might
1789 * be seen by the filesystem code called to write the core file.
1791 clear_thread_flag(TIF_SIGPENDING
);
1794 * lock_kernel() because format_corename() is controlled by sysctl, which
1795 * uses lock_kernel()
1798 ispipe
= format_corename(corename
, signr
);
1801 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1802 * to a pipe. Since we're not writing directly to the filesystem
1803 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1804 * created unless the pipe reader choses to write out the core file
1805 * at which point file size limits and permissions will be imposed
1806 * as it does with any other process
1808 if ((!ispipe
) && (core_limit
< binfmt
->min_coredump
))
1812 helper_argv
= argv_split(GFP_KERNEL
, corename
+1, &helper_argc
);
1814 printk(KERN_WARNING
"%s failed to allocate memory\n",
1818 /* Terminate the string before the first option */
1819 delimit
= strchr(corename
, ' ');
1822 delimit
= strrchr(helper_argv
[0], '/');
1826 delimit
= helper_argv
[0];
1827 if (!strcmp(delimit
, current
->comm
)) {
1828 printk(KERN_NOTICE
"Recursive core dump detected, "
1833 core_limit
= RLIM_INFINITY
;
1835 /* SIGPIPE can happen, but it's just never processed */
1836 if (call_usermodehelper_pipe(corename
+1, helper_argv
, NULL
,
1838 printk(KERN_INFO
"Core dump to %s pipe failed\n",
1843 file
= filp_open(corename
,
1844 O_CREAT
| 2 | O_NOFOLLOW
| O_LARGEFILE
| flag
,
1848 inode
= file
->f_path
.dentry
->d_inode
;
1849 if (inode
->i_nlink
> 1)
1850 goto close_fail
; /* multiple links - don't dump */
1851 if (!ispipe
&& d_unhashed(file
->f_path
.dentry
))
1854 /* AK: actually i see no reason to not allow this for named pipes etc.,
1855 but keep the previous behaviour for now. */
1856 if (!ispipe
&& !S_ISREG(inode
->i_mode
))
1859 * Dont allow local users get cute and trick others to coredump
1860 * into their pre-created files:
1862 if (inode
->i_uid
!= current_fsuid())
1866 if (!file
->f_op
->write
)
1868 if (!ispipe
&& do_truncate(file
->f_path
.dentry
, 0, 0, file
) != 0)
1871 retval
= binfmt
->core_dump(signr
, regs
, file
, core_limit
);
1874 current
->signal
->group_exit_code
|= 0x80;
1876 filp_close(file
, NULL
);
1879 argv_free(helper_argv
);
1881 revert_creds(old_cred
);
1883 coredump_finish(mm
);