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/ima.h>
50 #include <linux/syscalls.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>
57 #include <linux/fs_struct.h>
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
65 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
66 int suid_dumpable
= 0;
68 /* The maximal length of core_pattern is also specified in sysctl.c */
70 static LIST_HEAD(formats
);
71 static DEFINE_RWLOCK(binfmt_lock
);
73 int __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
);
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
);
101 * Note that a shared library must be both readable and executable due to
104 * Also note that we take the address to load from from the file itself.
106 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
109 char *tmp
= getname(library
);
110 int error
= PTR_ERR(tmp
);
115 file
= do_filp_open(AT_FDCWD
, tmp
,
116 O_LARGEFILE
| O_RDONLY
| FMODE_EXEC
, 0,
117 MAY_READ
| MAY_EXEC
| MAY_OPEN
);
119 error
= PTR_ERR(file
);
124 if (!S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
128 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
131 fsnotify_open(file
->f_path
.dentry
);
135 struct linux_binfmt
* fmt
;
137 read_lock(&binfmt_lock
);
138 list_for_each_entry(fmt
, &formats
, lh
) {
139 if (!fmt
->load_shlib
)
141 if (!try_module_get(fmt
->module
))
143 read_unlock(&binfmt_lock
);
144 error
= fmt
->load_shlib(file
);
145 read_lock(&binfmt_lock
);
147 if (error
!= -ENOEXEC
)
150 read_unlock(&binfmt_lock
);
160 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
166 #ifdef CONFIG_STACK_GROWSUP
168 ret
= expand_stack_downwards(bprm
->vma
, pos
);
173 ret
= get_user_pages(current
, bprm
->mm
, pos
,
174 1, write
, 1, &page
, NULL
);
179 unsigned long size
= bprm
->vma
->vm_end
- bprm
->vma
->vm_start
;
183 * We've historically supported up to 32 pages (ARG_MAX)
184 * of argument strings even with small stacks
190 * Limit to 1/4-th the stack size for the argv+env strings.
192 * - the remaining binfmt code will not run out of stack space,
193 * - the program will have a reasonable amount of stack left
196 rlim
= current
->signal
->rlim
;
197 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
/ 4) {
206 static void put_arg_page(struct page
*page
)
211 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
215 static void free_arg_pages(struct linux_binprm
*bprm
)
219 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
222 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
225 static int __bprm_mm_init(struct linux_binprm
*bprm
)
228 struct vm_area_struct
*vma
= NULL
;
229 struct mm_struct
*mm
= bprm
->mm
;
231 bprm
->vma
= vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
235 down_write(&mm
->mmap_sem
);
239 * Place the stack at the largest stack address the architecture
240 * supports. Later, we'll move this to an appropriate place. We don't
241 * use STACK_TOP because that can depend on attributes which aren't
244 vma
->vm_end
= STACK_TOP_MAX
;
245 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
246 vma
->vm_flags
= VM_STACK_FLAGS
;
247 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
248 err
= insert_vm_struct(mm
, vma
);
252 mm
->stack_vm
= mm
->total_vm
= 1;
253 up_write(&mm
->mmap_sem
);
254 bprm
->p
= vma
->vm_end
- sizeof(void *);
257 up_write(&mm
->mmap_sem
);
259 kmem_cache_free(vm_area_cachep
, vma
);
263 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
265 return len
<= MAX_ARG_STRLEN
;
270 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
275 page
= bprm
->page
[pos
/ PAGE_SIZE
];
276 if (!page
&& write
) {
277 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
280 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
286 static void put_arg_page(struct page
*page
)
290 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
293 __free_page(bprm
->page
[i
]);
294 bprm
->page
[i
] = NULL
;
298 static void free_arg_pages(struct linux_binprm
*bprm
)
302 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
303 free_arg_page(bprm
, i
);
306 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
311 static int __bprm_mm_init(struct linux_binprm
*bprm
)
313 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
317 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
319 return len
<= bprm
->p
;
322 #endif /* CONFIG_MMU */
325 * Create a new mm_struct and populate it with a temporary stack
326 * vm_area_struct. We don't have enough context at this point to set the stack
327 * flags, permissions, and offset, so we use temporary values. We'll update
328 * them later in setup_arg_pages().
330 int bprm_mm_init(struct linux_binprm
*bprm
)
333 struct mm_struct
*mm
= NULL
;
335 bprm
->mm
= mm
= mm_alloc();
340 err
= init_new_context(current
, mm
);
344 err
= __bprm_mm_init(bprm
);
360 * count() counts the number of strings in array ARGV.
362 static int count(char __user
* __user
* argv
, int max
)
370 if (get_user(p
, argv
))
384 * 'copy_strings()' copies argument/environment strings from the old
385 * processes's memory to the new process's stack. The call to get_user_pages()
386 * ensures the destination page is created and not swapped out.
388 static int copy_strings(int argc
, char __user
* __user
* argv
,
389 struct linux_binprm
*bprm
)
391 struct page
*kmapped_page
= NULL
;
393 unsigned long kpos
= 0;
401 if (get_user(str
, argv
+argc
) ||
402 !(len
= strnlen_user(str
, MAX_ARG_STRLEN
))) {
407 if (!valid_arg_len(bprm
, len
)) {
412 /* We're going to work our way backwords. */
418 int offset
, bytes_to_copy
;
420 offset
= pos
% PAGE_SIZE
;
424 bytes_to_copy
= offset
;
425 if (bytes_to_copy
> len
)
428 offset
-= bytes_to_copy
;
429 pos
-= bytes_to_copy
;
430 str
-= bytes_to_copy
;
431 len
-= bytes_to_copy
;
433 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
436 page
= get_arg_page(bprm
, pos
, 1);
443 flush_kernel_dcache_page(kmapped_page
);
444 kunmap(kmapped_page
);
445 put_arg_page(kmapped_page
);
448 kaddr
= kmap(kmapped_page
);
449 kpos
= pos
& PAGE_MASK
;
450 flush_arg_page(bprm
, kpos
, kmapped_page
);
452 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
461 flush_kernel_dcache_page(kmapped_page
);
462 kunmap(kmapped_page
);
463 put_arg_page(kmapped_page
);
469 * Like copy_strings, but get argv and its values from kernel memory.
471 int copy_strings_kernel(int argc
,char ** argv
, struct linux_binprm
*bprm
)
474 mm_segment_t oldfs
= get_fs();
476 r
= copy_strings(argc
, (char __user
* __user
*)argv
, bprm
);
480 EXPORT_SYMBOL(copy_strings_kernel
);
485 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
486 * the binfmt code determines where the new stack should reside, we shift it to
487 * its final location. The process proceeds as follows:
489 * 1) Use shift to calculate the new vma endpoints.
490 * 2) Extend vma to cover both the old and new ranges. This ensures the
491 * arguments passed to subsequent functions are consistent.
492 * 3) Move vma's page tables to the new range.
493 * 4) Free up any cleared pgd range.
494 * 5) Shrink the vma to cover only the new range.
496 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
498 struct mm_struct
*mm
= vma
->vm_mm
;
499 unsigned long old_start
= vma
->vm_start
;
500 unsigned long old_end
= vma
->vm_end
;
501 unsigned long length
= old_end
- old_start
;
502 unsigned long new_start
= old_start
- shift
;
503 unsigned long new_end
= old_end
- shift
;
504 struct mmu_gather
*tlb
;
506 BUG_ON(new_start
> new_end
);
509 * ensure there are no vmas between where we want to go
512 if (vma
!= find_vma(mm
, new_start
))
516 * cover the whole range: [new_start, old_end)
518 vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
);
521 * move the page tables downwards, on failure we rely on
522 * process cleanup to remove whatever mess we made.
524 if (length
!= move_page_tables(vma
, old_start
,
525 vma
, new_start
, length
))
529 tlb
= tlb_gather_mmu(mm
, 0);
530 if (new_end
> old_start
) {
532 * when the old and new regions overlap clear from new_end.
534 free_pgd_range(tlb
, new_end
, old_end
, new_end
,
535 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
538 * otherwise, clean from old_start; this is done to not touch
539 * the address space in [new_end, old_start) some architectures
540 * have constraints on va-space that make this illegal (IA64) -
541 * for the others its just a little faster.
543 free_pgd_range(tlb
, old_start
, old_end
, new_end
,
544 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
546 tlb_finish_mmu(tlb
, new_end
, old_end
);
549 * shrink the vma to just the new range.
551 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
556 #define EXTRA_STACK_VM_PAGES 20 /* random */
559 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
560 * the stack is optionally relocated, and some extra space is added.
562 int setup_arg_pages(struct linux_binprm
*bprm
,
563 unsigned long stack_top
,
564 int executable_stack
)
567 unsigned long stack_shift
;
568 struct mm_struct
*mm
= current
->mm
;
569 struct vm_area_struct
*vma
= bprm
->vma
;
570 struct vm_area_struct
*prev
= NULL
;
571 unsigned long vm_flags
;
572 unsigned long stack_base
;
574 #ifdef CONFIG_STACK_GROWSUP
575 /* Limit stack size to 1GB */
576 stack_base
= current
->signal
->rlim
[RLIMIT_STACK
].rlim_max
;
577 if (stack_base
> (1 << 30))
578 stack_base
= 1 << 30;
580 /* Make sure we didn't let the argument array grow too large. */
581 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
584 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
586 stack_shift
= vma
->vm_start
- stack_base
;
587 mm
->arg_start
= bprm
->p
- stack_shift
;
588 bprm
->p
= vma
->vm_end
- stack_shift
;
590 stack_top
= arch_align_stack(stack_top
);
591 stack_top
= PAGE_ALIGN(stack_top
);
592 stack_shift
= vma
->vm_end
- stack_top
;
594 bprm
->p
-= stack_shift
;
595 mm
->arg_start
= bprm
->p
;
599 bprm
->loader
-= stack_shift
;
600 bprm
->exec
-= stack_shift
;
602 down_write(&mm
->mmap_sem
);
603 vm_flags
= VM_STACK_FLAGS
;
606 * Adjust stack execute permissions; explicitly enable for
607 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
608 * (arch default) otherwise.
610 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
612 else if (executable_stack
== EXSTACK_DISABLE_X
)
613 vm_flags
&= ~VM_EXEC
;
614 vm_flags
|= mm
->def_flags
;
616 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
622 /* Move stack pages down in memory. */
624 ret
= shift_arg_pages(vma
, stack_shift
);
626 up_write(&mm
->mmap_sem
);
631 #ifdef CONFIG_STACK_GROWSUP
632 stack_base
= vma
->vm_end
+ EXTRA_STACK_VM_PAGES
* PAGE_SIZE
;
634 stack_base
= vma
->vm_start
- EXTRA_STACK_VM_PAGES
* PAGE_SIZE
;
636 ret
= expand_stack(vma
, stack_base
);
641 up_write(&mm
->mmap_sem
);
644 EXPORT_SYMBOL(setup_arg_pages
);
646 #endif /* CONFIG_MMU */
648 struct file
*open_exec(const char *name
)
653 file
= do_filp_open(AT_FDCWD
, name
,
654 O_LARGEFILE
| O_RDONLY
| FMODE_EXEC
, 0,
655 MAY_EXEC
| MAY_OPEN
);
660 if (!S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
663 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
666 fsnotify_open(file
->f_path
.dentry
);
668 err
= deny_write_access(file
);
679 EXPORT_SYMBOL(open_exec
);
681 int kernel_read(struct file
*file
, loff_t offset
,
682 char *addr
, unsigned long count
)
690 /* The cast to a user pointer is valid due to the set_fs() */
691 result
= vfs_read(file
, (void __user
*)addr
, count
, &pos
);
696 EXPORT_SYMBOL(kernel_read
);
698 static int exec_mmap(struct mm_struct
*mm
)
700 struct task_struct
*tsk
;
701 struct mm_struct
* old_mm
, *active_mm
;
703 /* Notify parent that we're no longer interested in the old VM */
705 old_mm
= current
->mm
;
706 mm_release(tsk
, old_mm
);
710 * Make sure that if there is a core dump in progress
711 * for the old mm, we get out and die instead of going
712 * through with the exec. We must hold mmap_sem around
713 * checking core_state and changing tsk->mm.
715 down_read(&old_mm
->mmap_sem
);
716 if (unlikely(old_mm
->core_state
)) {
717 up_read(&old_mm
->mmap_sem
);
722 active_mm
= tsk
->active_mm
;
725 activate_mm(active_mm
, mm
);
727 arch_pick_mmap_layout(mm
);
729 up_read(&old_mm
->mmap_sem
);
730 BUG_ON(active_mm
!= old_mm
);
731 mm_update_next_owner(old_mm
);
740 * This function makes sure the current process has its own signal table,
741 * so that flush_signal_handlers can later reset the handlers without
742 * disturbing other processes. (Other processes might share the signal
743 * table via the CLONE_SIGHAND option to clone().)
745 static int de_thread(struct task_struct
*tsk
)
747 struct signal_struct
*sig
= tsk
->signal
;
748 struct sighand_struct
*oldsighand
= tsk
->sighand
;
749 spinlock_t
*lock
= &oldsighand
->siglock
;
752 if (thread_group_empty(tsk
))
753 goto no_thread_group
;
756 * Kill all other threads in the thread group.
759 if (signal_group_exit(sig
)) {
761 * Another group action in progress, just
762 * return so that the signal is processed.
764 spin_unlock_irq(lock
);
767 sig
->group_exit_task
= tsk
;
768 zap_other_threads(tsk
);
770 /* Account for the thread group leader hanging around: */
771 count
= thread_group_leader(tsk
) ? 1 : 2;
772 sig
->notify_count
= count
;
773 while (atomic_read(&sig
->count
) > count
) {
774 __set_current_state(TASK_UNINTERRUPTIBLE
);
775 spin_unlock_irq(lock
);
779 spin_unlock_irq(lock
);
782 * At this point all other threads have exited, all we have to
783 * do is to wait for the thread group leader to become inactive,
784 * and to assume its PID:
786 if (!thread_group_leader(tsk
)) {
787 struct task_struct
*leader
= tsk
->group_leader
;
789 sig
->notify_count
= -1; /* for exit_notify() */
791 write_lock_irq(&tasklist_lock
);
792 if (likely(leader
->exit_state
))
794 __set_current_state(TASK_UNINTERRUPTIBLE
);
795 write_unlock_irq(&tasklist_lock
);
800 * The only record we have of the real-time age of a
801 * process, regardless of execs it's done, is start_time.
802 * All the past CPU time is accumulated in signal_struct
803 * from sister threads now dead. But in this non-leader
804 * exec, nothing survives from the original leader thread,
805 * whose birth marks the true age of this process now.
806 * When we take on its identity by switching to its PID, we
807 * also take its birthdate (always earlier than our own).
809 tsk
->start_time
= leader
->start_time
;
811 BUG_ON(!same_thread_group(leader
, tsk
));
812 BUG_ON(has_group_leader_pid(tsk
));
814 * An exec() starts a new thread group with the
815 * TGID of the previous thread group. Rehash the
816 * two threads with a switched PID, and release
817 * the former thread group leader:
820 /* Become a process group leader with the old leader's pid.
821 * The old leader becomes a thread of the this thread group.
822 * Note: The old leader also uses this pid until release_task
823 * is called. Odd but simple and correct.
825 detach_pid(tsk
, PIDTYPE_PID
);
826 tsk
->pid
= leader
->pid
;
827 attach_pid(tsk
, PIDTYPE_PID
, task_pid(leader
));
828 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
829 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
830 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
832 tsk
->group_leader
= tsk
;
833 leader
->group_leader
= tsk
;
835 tsk
->exit_signal
= SIGCHLD
;
837 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
838 leader
->exit_state
= EXIT_DEAD
;
839 write_unlock_irq(&tasklist_lock
);
841 release_task(leader
);
844 sig
->group_exit_task
= NULL
;
845 sig
->notify_count
= 0;
849 flush_itimer_signals();
851 if (atomic_read(&oldsighand
->count
) != 1) {
852 struct sighand_struct
*newsighand
;
854 * This ->sighand is shared with the CLONE_SIGHAND
855 * but not CLONE_THREAD task, switch to the new one.
857 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
861 atomic_set(&newsighand
->count
, 1);
862 memcpy(newsighand
->action
, oldsighand
->action
,
863 sizeof(newsighand
->action
));
865 write_lock_irq(&tasklist_lock
);
866 spin_lock(&oldsighand
->siglock
);
867 rcu_assign_pointer(tsk
->sighand
, newsighand
);
868 spin_unlock(&oldsighand
->siglock
);
869 write_unlock_irq(&tasklist_lock
);
871 __cleanup_sighand(oldsighand
);
874 BUG_ON(!thread_group_leader(tsk
));
879 * These functions flushes out all traces of the currently running executable
880 * so that a new one can be started
882 static void flush_old_files(struct files_struct
* files
)
887 spin_lock(&files
->file_lock
);
889 unsigned long set
, i
;
893 fdt
= files_fdtable(files
);
894 if (i
>= fdt
->max_fds
)
896 set
= fdt
->close_on_exec
->fds_bits
[j
];
899 fdt
->close_on_exec
->fds_bits
[j
] = 0;
900 spin_unlock(&files
->file_lock
);
901 for ( ; set
; i
++,set
>>= 1) {
906 spin_lock(&files
->file_lock
);
909 spin_unlock(&files
->file_lock
);
912 char *get_task_comm(char *buf
, struct task_struct
*tsk
)
914 /* buf must be at least sizeof(tsk->comm) in size */
916 strncpy(buf
, tsk
->comm
, sizeof(tsk
->comm
));
921 void set_task_comm(struct task_struct
*tsk
, char *buf
)
924 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
926 perf_counter_comm(tsk
);
929 int flush_old_exec(struct linux_binprm
* bprm
)
933 char tcomm
[sizeof(current
->comm
)];
936 * Make sure we have a private signal table and that
937 * we are unassociated from the previous thread group.
939 retval
= de_thread(current
);
943 set_mm_exe_file(bprm
->mm
, bprm
->file
);
946 * Release all of the old mmap stuff
948 retval
= exec_mmap(bprm
->mm
);
952 bprm
->mm
= NULL
; /* We're using it now */
954 /* This is the point of no return */
955 current
->sas_ss_sp
= current
->sas_ss_size
= 0;
957 if (current_euid() == current_uid() && current_egid() == current_gid())
958 set_dumpable(current
->mm
, 1);
960 set_dumpable(current
->mm
, suid_dumpable
);
962 name
= bprm
->filename
;
964 /* Copies the binary name from after last slash */
965 for (i
=0; (ch
= *(name
++)) != '\0';) {
967 i
= 0; /* overwrite what we wrote */
969 if (i
< (sizeof(tcomm
) - 1))
973 set_task_comm(current
, tcomm
);
975 current
->flags
&= ~PF_RANDOMIZE
;
978 /* Set the new mm task size. We have to do that late because it may
979 * depend on TIF_32BIT which is only updated in flush_thread() on
980 * some architectures like powerpc
982 current
->mm
->task_size
= TASK_SIZE
;
984 /* install the new credentials */
985 if (bprm
->cred
->uid
!= current_euid() ||
986 bprm
->cred
->gid
!= current_egid()) {
987 current
->pdeath_signal
= 0;
988 } else if (file_permission(bprm
->file
, MAY_READ
) ||
989 bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
) {
990 set_dumpable(current
->mm
, suid_dumpable
);
993 current
->personality
&= ~bprm
->per_clear
;
996 * Flush performance counters when crossing a
999 if (!get_dumpable(current
->mm
))
1000 perf_counter_exit_task(current
);
1002 /* An exec changes our domain. We are no longer part of the thread
1005 current
->self_exec_id
++;
1007 flush_signal_handlers(current
, 0);
1008 flush_old_files(current
->files
);
1016 EXPORT_SYMBOL(flush_old_exec
);
1019 * install the new credentials for this executable
1021 void install_exec_creds(struct linux_binprm
*bprm
)
1023 security_bprm_committing_creds(bprm
);
1025 commit_creds(bprm
->cred
);
1028 /* cred_guard_mutex must be held at least to this point to prevent
1029 * ptrace_attach() from altering our determination of the task's
1030 * credentials; any time after this it may be unlocked */
1032 security_bprm_committed_creds(bprm
);
1034 EXPORT_SYMBOL(install_exec_creds
);
1037 * determine how safe it is to execute the proposed program
1038 * - the caller must hold current->cred_guard_mutex to protect against
1041 int check_unsafe_exec(struct linux_binprm
*bprm
)
1043 struct task_struct
*p
= current
, *t
;
1047 bprm
->unsafe
= tracehook_unsafe_exec(p
);
1050 write_lock(&p
->fs
->lock
);
1052 for (t
= next_thread(p
); t
!= p
; t
= next_thread(t
)) {
1058 if (p
->fs
->users
> n_fs
) {
1059 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1062 if (!p
->fs
->in_exec
) {
1067 write_unlock(&p
->fs
->lock
);
1073 * Fill the binprm structure from the inode.
1074 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1076 * This may be called multiple times for binary chains (scripts for example).
1078 int prepare_binprm(struct linux_binprm
*bprm
)
1081 struct inode
* inode
= bprm
->file
->f_path
.dentry
->d_inode
;
1084 mode
= inode
->i_mode
;
1085 if (bprm
->file
->f_op
== NULL
)
1088 /* clear any previous set[ug]id data from a previous binary */
1089 bprm
->cred
->euid
= current_euid();
1090 bprm
->cred
->egid
= current_egid();
1092 if (!(bprm
->file
->f_path
.mnt
->mnt_flags
& MNT_NOSUID
)) {
1094 if (mode
& S_ISUID
) {
1095 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1096 bprm
->cred
->euid
= inode
->i_uid
;
1101 * If setgid is set but no group execute bit then this
1102 * is a candidate for mandatory locking, not a setgid
1105 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1106 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1107 bprm
->cred
->egid
= inode
->i_gid
;
1111 /* fill in binprm security blob */
1112 retval
= security_bprm_set_creds(bprm
);
1115 bprm
->cred_prepared
= 1;
1117 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1118 return kernel_read(bprm
->file
, 0, bprm
->buf
, BINPRM_BUF_SIZE
);
1121 EXPORT_SYMBOL(prepare_binprm
);
1124 * Arguments are '\0' separated strings found at the location bprm->p
1125 * points to; chop off the first by relocating brpm->p to right after
1126 * the first '\0' encountered.
1128 int remove_arg_zero(struct linux_binprm
*bprm
)
1131 unsigned long offset
;
1139 offset
= bprm
->p
& ~PAGE_MASK
;
1140 page
= get_arg_page(bprm
, bprm
->p
, 0);
1145 kaddr
= kmap_atomic(page
, KM_USER0
);
1147 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1148 offset
++, bprm
->p
++)
1151 kunmap_atomic(kaddr
, KM_USER0
);
1154 if (offset
== PAGE_SIZE
)
1155 free_arg_page(bprm
, (bprm
->p
>> PAGE_SHIFT
) - 1);
1156 } while (offset
== PAGE_SIZE
);
1165 EXPORT_SYMBOL(remove_arg_zero
);
1168 * cycle the list of binary formats handler, until one recognizes the image
1170 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
1172 unsigned int depth
= bprm
->recursion_depth
;
1174 struct linux_binfmt
*fmt
;
1176 retval
= security_bprm_check(bprm
);
1179 retval
= ima_bprm_check(bprm
);
1183 /* kernel module loader fixup */
1184 /* so we don't try to load run modprobe in kernel space. */
1187 retval
= audit_bprm(bprm
);
1192 for (try=0; try<2; try++) {
1193 read_lock(&binfmt_lock
);
1194 list_for_each_entry(fmt
, &formats
, lh
) {
1195 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
1198 if (!try_module_get(fmt
->module
))
1200 read_unlock(&binfmt_lock
);
1201 retval
= fn(bprm
, regs
);
1203 * Restore the depth counter to its starting value
1204 * in this call, so we don't have to rely on every
1205 * load_binary function to restore it on return.
1207 bprm
->recursion_depth
= depth
;
1210 tracehook_report_exec(fmt
, bprm
, regs
);
1212 allow_write_access(bprm
->file
);
1216 current
->did_exec
= 1;
1217 proc_exec_connector(current
);
1220 read_lock(&binfmt_lock
);
1222 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
)
1225 read_unlock(&binfmt_lock
);
1229 read_unlock(&binfmt_lock
);
1230 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
) {
1232 #ifdef CONFIG_MODULES
1234 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1235 if (printable(bprm
->buf
[0]) &&
1236 printable(bprm
->buf
[1]) &&
1237 printable(bprm
->buf
[2]) &&
1238 printable(bprm
->buf
[3]))
1239 break; /* -ENOEXEC */
1240 request_module("binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
1247 EXPORT_SYMBOL(search_binary_handler
);
1249 void free_bprm(struct linux_binprm
*bprm
)
1251 free_arg_pages(bprm
);
1253 abort_creds(bprm
->cred
);
1258 * sys_execve() executes a new program.
1260 int do_execve(char * filename
,
1261 char __user
*__user
*argv
,
1262 char __user
*__user
*envp
,
1263 struct pt_regs
* regs
)
1265 struct linux_binprm
*bprm
;
1267 struct files_struct
*displaced
;
1271 retval
= unshare_files(&displaced
);
1276 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1280 retval
= -ERESTARTNOINTR
;
1281 if (mutex_lock_interruptible(¤t
->cred_guard_mutex
))
1283 current
->in_execve
= 1;
1286 bprm
->cred
= prepare_exec_creds();
1290 retval
= check_unsafe_exec(bprm
);
1293 clear_in_exec
= retval
;
1295 file
= open_exec(filename
);
1296 retval
= PTR_ERR(file
);
1303 bprm
->filename
= filename
;
1304 bprm
->interp
= filename
;
1306 retval
= bprm_mm_init(bprm
);
1310 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
1311 if ((retval
= bprm
->argc
) < 0)
1314 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
1315 if ((retval
= bprm
->envc
) < 0)
1318 retval
= prepare_binprm(bprm
);
1322 retval
= copy_strings_kernel(1, &bprm
->filename
, bprm
);
1326 bprm
->exec
= bprm
->p
;
1327 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1331 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1335 current
->flags
&= ~PF_KTHREAD
;
1336 retval
= search_binary_handler(bprm
,regs
);
1340 /* execve succeeded */
1341 current
->fs
->in_exec
= 0;
1342 current
->in_execve
= 0;
1343 mutex_unlock(¤t
->cred_guard_mutex
);
1344 acct_update_integrals(current
);
1347 put_files_struct(displaced
);
1356 allow_write_access(bprm
->file
);
1362 current
->fs
->in_exec
= 0;
1365 current
->in_execve
= 0;
1366 mutex_unlock(¤t
->cred_guard_mutex
);
1373 reset_files_struct(displaced
);
1378 int set_binfmt(struct linux_binfmt
*new)
1380 struct linux_binfmt
*old
= current
->binfmt
;
1383 if (!try_module_get(new->module
))
1386 current
->binfmt
= new;
1388 module_put(old
->module
);
1392 EXPORT_SYMBOL(set_binfmt
);
1394 /* format_corename will inspect the pattern parameter, and output a
1395 * name into corename, which must have space for at least
1396 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1398 static int format_corename(char *corename
, long signr
)
1400 const struct cred
*cred
= current_cred();
1401 const char *pat_ptr
= core_pattern
;
1402 int ispipe
= (*pat_ptr
== '|');
1403 char *out_ptr
= corename
;
1404 char *const out_end
= corename
+ CORENAME_MAX_SIZE
;
1406 int pid_in_pattern
= 0;
1408 /* Repeat as long as we have more pattern to process and more output
1411 if (*pat_ptr
!= '%') {
1412 if (out_ptr
== out_end
)
1414 *out_ptr
++ = *pat_ptr
++;
1416 switch (*++pat_ptr
) {
1419 /* Double percent, output one percent */
1421 if (out_ptr
== out_end
)
1428 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1429 "%d", task_tgid_vnr(current
));
1430 if (rc
> out_end
- out_ptr
)
1436 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1438 if (rc
> out_end
- out_ptr
)
1444 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1446 if (rc
> out_end
- out_ptr
)
1450 /* signal that caused the coredump */
1452 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1454 if (rc
> out_end
- out_ptr
)
1458 /* UNIX time of coredump */
1461 do_gettimeofday(&tv
);
1462 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1464 if (rc
> out_end
- out_ptr
)
1471 down_read(&uts_sem
);
1472 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1473 "%s", utsname()->nodename
);
1475 if (rc
> out_end
- out_ptr
)
1481 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1482 "%s", current
->comm
);
1483 if (rc
> out_end
- out_ptr
)
1487 /* core limit size */
1489 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1490 "%lu", current
->signal
->rlim
[RLIMIT_CORE
].rlim_cur
);
1491 if (rc
> out_end
- out_ptr
)
1501 /* Backward compatibility with core_uses_pid:
1503 * If core_pattern does not include a %p (as is the default)
1504 * and core_uses_pid is set, then .%pid will be appended to
1505 * the filename. Do not do this for piped commands. */
1506 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
1507 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1508 ".%d", task_tgid_vnr(current
));
1509 if (rc
> out_end
- out_ptr
)
1518 static int zap_process(struct task_struct
*start
)
1520 struct task_struct
*t
;
1523 start
->signal
->flags
= SIGNAL_GROUP_EXIT
;
1524 start
->signal
->group_stop_count
= 0;
1528 if (t
!= current
&& t
->mm
) {
1529 sigaddset(&t
->pending
.signal
, SIGKILL
);
1530 signal_wake_up(t
, 1);
1533 } while_each_thread(start
, t
);
1538 static inline int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
1539 struct core_state
*core_state
, int exit_code
)
1541 struct task_struct
*g
, *p
;
1542 unsigned long flags
;
1545 spin_lock_irq(&tsk
->sighand
->siglock
);
1546 if (!signal_group_exit(tsk
->signal
)) {
1547 mm
->core_state
= core_state
;
1548 tsk
->signal
->group_exit_code
= exit_code
;
1549 nr
= zap_process(tsk
);
1551 spin_unlock_irq(&tsk
->sighand
->siglock
);
1552 if (unlikely(nr
< 0))
1555 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
1558 * We should find and kill all tasks which use this mm, and we should
1559 * count them correctly into ->nr_threads. We don't take tasklist
1560 * lock, but this is safe wrt:
1563 * None of sub-threads can fork after zap_process(leader). All
1564 * processes which were created before this point should be
1565 * visible to zap_threads() because copy_process() adds the new
1566 * process to the tail of init_task.tasks list, and lock/unlock
1567 * of ->siglock provides a memory barrier.
1570 * The caller holds mm->mmap_sem. This means that the task which
1571 * uses this mm can't pass exit_mm(), so it can't exit or clear
1575 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1576 * we must see either old or new leader, this does not matter.
1577 * However, it can change p->sighand, so lock_task_sighand(p)
1578 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1581 * Note also that "g" can be the old leader with ->mm == NULL
1582 * and already unhashed and thus removed from ->thread_group.
1583 * This is OK, __unhash_process()->list_del_rcu() does not
1584 * clear the ->next pointer, we will find the new leader via
1588 for_each_process(g
) {
1589 if (g
== tsk
->group_leader
)
1591 if (g
->flags
& PF_KTHREAD
)
1596 if (unlikely(p
->mm
== mm
)) {
1597 lock_task_sighand(p
, &flags
);
1598 nr
+= zap_process(p
);
1599 unlock_task_sighand(p
, &flags
);
1603 } while_each_thread(g
, p
);
1607 atomic_set(&core_state
->nr_threads
, nr
);
1611 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
1613 struct task_struct
*tsk
= current
;
1614 struct mm_struct
*mm
= tsk
->mm
;
1615 struct completion
*vfork_done
;
1618 init_completion(&core_state
->startup
);
1619 core_state
->dumper
.task
= tsk
;
1620 core_state
->dumper
.next
= NULL
;
1621 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
1622 up_write(&mm
->mmap_sem
);
1624 if (unlikely(core_waiters
< 0))
1628 * Make sure nobody is waiting for us to release the VM,
1629 * otherwise we can deadlock when we wait on each other
1631 vfork_done
= tsk
->vfork_done
;
1633 tsk
->vfork_done
= NULL
;
1634 complete(vfork_done
);
1638 wait_for_completion(&core_state
->startup
);
1640 return core_waiters
;
1643 static void coredump_finish(struct mm_struct
*mm
)
1645 struct core_thread
*curr
, *next
;
1646 struct task_struct
*task
;
1648 next
= mm
->core_state
->dumper
.next
;
1649 while ((curr
= next
) != NULL
) {
1653 * see exit_mm(), curr->task must not see
1654 * ->task == NULL before we read ->next.
1658 wake_up_process(task
);
1661 mm
->core_state
= NULL
;
1665 * set_dumpable converts traditional three-value dumpable to two flags and
1666 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1667 * these bits are not changed atomically. So get_dumpable can observe the
1668 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1669 * return either old dumpable or new one by paying attention to the order of
1670 * modifying the bits.
1672 * dumpable | mm->flags (binary)
1673 * old new | initial interim final
1674 * ---------+-----------------------
1682 * (*) get_dumpable regards interim value of 10 as 11.
1684 void set_dumpable(struct mm_struct
*mm
, int value
)
1688 clear_bit(MMF_DUMPABLE
, &mm
->flags
);
1690 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1693 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1695 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1698 set_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1700 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1705 int get_dumpable(struct mm_struct
*mm
)
1709 ret
= mm
->flags
& 0x3;
1710 return (ret
>= 2) ? 2 : ret
;
1713 void do_coredump(long signr
, int exit_code
, struct pt_regs
*regs
)
1715 struct core_state core_state
;
1716 char corename
[CORENAME_MAX_SIZE
+ 1];
1717 struct mm_struct
*mm
= current
->mm
;
1718 struct linux_binfmt
* binfmt
;
1719 struct inode
* inode
;
1721 const struct cred
*old_cred
;
1726 unsigned long core_limit
= current
->signal
->rlim
[RLIMIT_CORE
].rlim_cur
;
1727 char **helper_argv
= NULL
;
1728 int helper_argc
= 0;
1731 audit_core_dumps(signr
);
1733 binfmt
= current
->binfmt
;
1734 if (!binfmt
|| !binfmt
->core_dump
)
1737 cred
= prepare_creds();
1743 down_write(&mm
->mmap_sem
);
1745 * If another thread got here first, or we are not dumpable, bail out.
1747 if (mm
->core_state
|| !get_dumpable(mm
)) {
1748 up_write(&mm
->mmap_sem
);
1754 * We cannot trust fsuid as being the "true" uid of the
1755 * process nor do we know its entire history. We only know it
1756 * was tainted so we dump it as root in mode 2.
1758 if (get_dumpable(mm
) == 2) { /* Setuid core dump mode */
1759 flag
= O_EXCL
; /* Stop rewrite attacks */
1760 cred
->fsuid
= 0; /* Dump root private */
1763 retval
= coredump_wait(exit_code
, &core_state
);
1769 old_cred
= override_creds(cred
);
1772 * Clear any false indication of pending signals that might
1773 * be seen by the filesystem code called to write the core file.
1775 clear_thread_flag(TIF_SIGPENDING
);
1778 * lock_kernel() because format_corename() is controlled by sysctl, which
1779 * uses lock_kernel()
1782 ispipe
= format_corename(corename
, signr
);
1785 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1786 * to a pipe. Since we're not writing directly to the filesystem
1787 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1788 * created unless the pipe reader choses to write out the core file
1789 * at which point file size limits and permissions will be imposed
1790 * as it does with any other process
1792 if ((!ispipe
) && (core_limit
< binfmt
->min_coredump
))
1796 helper_argv
= argv_split(GFP_KERNEL
, corename
+1, &helper_argc
);
1798 printk(KERN_WARNING
"%s failed to allocate memory\n",
1802 /* Terminate the string before the first option */
1803 delimit
= strchr(corename
, ' ');
1806 delimit
= strrchr(helper_argv
[0], '/');
1810 delimit
= helper_argv
[0];
1811 if (!strcmp(delimit
, current
->comm
)) {
1812 printk(KERN_NOTICE
"Recursive core dump detected, "
1817 core_limit
= RLIM_INFINITY
;
1819 /* SIGPIPE can happen, but it's just never processed */
1820 if (call_usermodehelper_pipe(corename
+1, helper_argv
, NULL
,
1822 printk(KERN_INFO
"Core dump to %s pipe failed\n",
1827 file
= filp_open(corename
,
1828 O_CREAT
| 2 | O_NOFOLLOW
| O_LARGEFILE
| flag
,
1832 inode
= file
->f_path
.dentry
->d_inode
;
1833 if (inode
->i_nlink
> 1)
1834 goto close_fail
; /* multiple links - don't dump */
1835 if (!ispipe
&& d_unhashed(file
->f_path
.dentry
))
1838 /* AK: actually i see no reason to not allow this for named pipes etc.,
1839 but keep the previous behaviour for now. */
1840 if (!ispipe
&& !S_ISREG(inode
->i_mode
))
1843 * Dont allow local users get cute and trick others to coredump
1844 * into their pre-created files:
1846 if (inode
->i_uid
!= current_fsuid())
1850 if (!file
->f_op
->write
)
1852 if (!ispipe
&& do_truncate(file
->f_path
.dentry
, 0, 0, file
) != 0)
1855 retval
= binfmt
->core_dump(signr
, regs
, file
, core_limit
);
1858 current
->signal
->group_exit_code
|= 0x80;
1860 filp_close(file
, NULL
);
1863 argv_free(helper_argv
);
1865 revert_creds(old_cred
);
1867 coredump_finish(mm
);