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_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/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>
58 #include <linux/pipe_fs_i.h>
60 #include <asm/uaccess.h>
61 #include <asm/mmu_context.h>
66 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
67 unsigned int core_pipe_limit
;
68 int suid_dumpable
= 0;
70 /* The maximal length of core_pattern is also specified in sysctl.c */
72 static LIST_HEAD(formats
);
73 static DEFINE_RWLOCK(binfmt_lock
);
75 int __register_binfmt(struct linux_binfmt
* fmt
, int insert
)
79 write_lock(&binfmt_lock
);
80 insert
? list_add(&fmt
->lh
, &formats
) :
81 list_add_tail(&fmt
->lh
, &formats
);
82 write_unlock(&binfmt_lock
);
86 EXPORT_SYMBOL(__register_binfmt
);
88 void unregister_binfmt(struct linux_binfmt
* fmt
)
90 write_lock(&binfmt_lock
);
92 write_unlock(&binfmt_lock
);
95 EXPORT_SYMBOL(unregister_binfmt
);
97 static inline void put_binfmt(struct linux_binfmt
* fmt
)
99 module_put(fmt
->module
);
103 * Note that a shared library must be both readable and executable due to
106 * Also note that we take the address to load from from the file itself.
108 SYSCALL_DEFINE1(uselib
, const char __user
*, library
)
111 char *tmp
= getname(library
);
112 int error
= PTR_ERR(tmp
);
117 file
= do_filp_open(AT_FDCWD
, tmp
,
118 O_LARGEFILE
| O_RDONLY
| FMODE_EXEC
, 0,
119 MAY_READ
| MAY_EXEC
| MAY_OPEN
);
121 error
= PTR_ERR(file
);
126 if (!S_ISREG(file
->f_path
.dentry
->d_inode
->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 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
168 #ifdef CONFIG_STACK_GROWSUP
170 ret
= expand_stack_downwards(bprm
->vma
, pos
);
175 ret
= get_user_pages(current
, bprm
->mm
, pos
,
176 1, write
, 1, &page
, NULL
);
181 unsigned long size
= bprm
->vma
->vm_end
- bprm
->vma
->vm_start
;
185 * We've historically supported up to 32 pages (ARG_MAX)
186 * of argument strings even with small stacks
192 * Limit to 1/4-th the stack size for the argv+env strings.
194 * - the remaining binfmt code will not run out of stack space,
195 * - the program will have a reasonable amount of stack left
198 rlim
= current
->signal
->rlim
;
199 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
/ 4) {
208 static void put_arg_page(struct page
*page
)
213 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
217 static void free_arg_pages(struct linux_binprm
*bprm
)
221 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
224 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
227 static int __bprm_mm_init(struct linux_binprm
*bprm
)
230 struct vm_area_struct
*vma
= NULL
;
231 struct mm_struct
*mm
= bprm
->mm
;
233 bprm
->vma
= vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
237 down_write(&mm
->mmap_sem
);
241 * Place the stack at the largest stack address the architecture
242 * supports. Later, we'll move this to an appropriate place. We don't
243 * use STACK_TOP because that can depend on attributes which aren't
246 vma
->vm_end
= STACK_TOP_MAX
;
247 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
248 vma
->vm_flags
= VM_STACK_FLAGS
;
249 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
250 err
= insert_vm_struct(mm
, vma
);
254 mm
->stack_vm
= mm
->total_vm
= 1;
255 up_write(&mm
->mmap_sem
);
256 bprm
->p
= vma
->vm_end
- sizeof(void *);
259 up_write(&mm
->mmap_sem
);
261 kmem_cache_free(vm_area_cachep
, vma
);
265 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
267 return len
<= MAX_ARG_STRLEN
;
272 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
277 page
= bprm
->page
[pos
/ PAGE_SIZE
];
278 if (!page
&& write
) {
279 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
282 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
288 static void put_arg_page(struct page
*page
)
292 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
295 __free_page(bprm
->page
[i
]);
296 bprm
->page
[i
] = NULL
;
300 static void free_arg_pages(struct linux_binprm
*bprm
)
304 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
305 free_arg_page(bprm
, i
);
308 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
313 static int __bprm_mm_init(struct linux_binprm
*bprm
)
315 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
319 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
321 return len
<= bprm
->p
;
324 #endif /* CONFIG_MMU */
327 * Create a new mm_struct and populate it with a temporary stack
328 * vm_area_struct. We don't have enough context at this point to set the stack
329 * flags, permissions, and offset, so we use temporary values. We'll update
330 * them later in setup_arg_pages().
332 int bprm_mm_init(struct linux_binprm
*bprm
)
335 struct mm_struct
*mm
= NULL
;
337 bprm
->mm
= mm
= mm_alloc();
342 err
= init_new_context(current
, mm
);
346 err
= __bprm_mm_init(bprm
);
362 * count() counts the number of strings in array ARGV.
364 static int count(char __user
* __user
* argv
, int max
)
372 if (get_user(p
, argv
))
386 * 'copy_strings()' copies argument/environment strings from the old
387 * processes's memory to the new process's stack. The call to get_user_pages()
388 * ensures the destination page is created and not swapped out.
390 static int copy_strings(int argc
, char __user
* __user
* argv
,
391 struct linux_binprm
*bprm
)
393 struct page
*kmapped_page
= NULL
;
395 unsigned long kpos
= 0;
403 if (get_user(str
, argv
+argc
) ||
404 !(len
= strnlen_user(str
, MAX_ARG_STRLEN
))) {
409 if (!valid_arg_len(bprm
, len
)) {
414 /* We're going to work our way backwords. */
420 int offset
, bytes_to_copy
;
422 offset
= pos
% PAGE_SIZE
;
426 bytes_to_copy
= offset
;
427 if (bytes_to_copy
> len
)
430 offset
-= bytes_to_copy
;
431 pos
-= bytes_to_copy
;
432 str
-= bytes_to_copy
;
433 len
-= bytes_to_copy
;
435 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
438 page
= get_arg_page(bprm
, pos
, 1);
445 flush_kernel_dcache_page(kmapped_page
);
446 kunmap(kmapped_page
);
447 put_arg_page(kmapped_page
);
450 kaddr
= kmap(kmapped_page
);
451 kpos
= pos
& PAGE_MASK
;
452 flush_arg_page(bprm
, kpos
, kmapped_page
);
454 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
463 flush_kernel_dcache_page(kmapped_page
);
464 kunmap(kmapped_page
);
465 put_arg_page(kmapped_page
);
471 * Like copy_strings, but get argv and its values from kernel memory.
473 int copy_strings_kernel(int argc
,char ** argv
, struct linux_binprm
*bprm
)
476 mm_segment_t oldfs
= get_fs();
478 r
= copy_strings(argc
, (char __user
* __user
*)argv
, bprm
);
482 EXPORT_SYMBOL(copy_strings_kernel
);
487 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
488 * the binfmt code determines where the new stack should reside, we shift it to
489 * its final location. The process proceeds as follows:
491 * 1) Use shift to calculate the new vma endpoints.
492 * 2) Extend vma to cover both the old and new ranges. This ensures the
493 * arguments passed to subsequent functions are consistent.
494 * 3) Move vma's page tables to the new range.
495 * 4) Free up any cleared pgd range.
496 * 5) Shrink the vma to cover only the new range.
498 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
500 struct mm_struct
*mm
= vma
->vm_mm
;
501 unsigned long old_start
= vma
->vm_start
;
502 unsigned long old_end
= vma
->vm_end
;
503 unsigned long length
= old_end
- old_start
;
504 unsigned long new_start
= old_start
- shift
;
505 unsigned long new_end
= old_end
- shift
;
506 struct mmu_gather
*tlb
;
508 BUG_ON(new_start
> new_end
);
511 * ensure there are no vmas between where we want to go
514 if (vma
!= find_vma(mm
, new_start
))
518 * cover the whole range: [new_start, old_end)
520 vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
);
523 * move the page tables downwards, on failure we rely on
524 * process cleanup to remove whatever mess we made.
526 if (length
!= move_page_tables(vma
, old_start
,
527 vma
, new_start
, length
))
531 tlb
= tlb_gather_mmu(mm
, 0);
532 if (new_end
> old_start
) {
534 * when the old and new regions overlap clear from new_end.
536 free_pgd_range(tlb
, new_end
, old_end
, new_end
,
537 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
540 * otherwise, clean from old_start; this is done to not touch
541 * the address space in [new_end, old_start) some architectures
542 * have constraints on va-space that make this illegal (IA64) -
543 * for the others its just a little faster.
545 free_pgd_range(tlb
, old_start
, old_end
, new_end
,
546 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
548 tlb_finish_mmu(tlb
, new_end
, old_end
);
551 * shrink the vma to just the new range.
553 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
558 #define EXTRA_STACK_VM_PAGES 20 /* random */
561 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
562 * the stack is optionally relocated, and some extra space is added.
564 int setup_arg_pages(struct linux_binprm
*bprm
,
565 unsigned long stack_top
,
566 int executable_stack
)
569 unsigned long stack_shift
;
570 struct mm_struct
*mm
= current
->mm
;
571 struct vm_area_struct
*vma
= bprm
->vma
;
572 struct vm_area_struct
*prev
= NULL
;
573 unsigned long vm_flags
;
574 unsigned long stack_base
;
576 #ifdef CONFIG_STACK_GROWSUP
577 /* Limit stack size to 1GB */
578 stack_base
= current
->signal
->rlim
[RLIMIT_STACK
].rlim_max
;
579 if (stack_base
> (1 << 30))
580 stack_base
= 1 << 30;
582 /* Make sure we didn't let the argument array grow too large. */
583 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
586 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
588 stack_shift
= vma
->vm_start
- stack_base
;
589 mm
->arg_start
= bprm
->p
- stack_shift
;
590 bprm
->p
= vma
->vm_end
- stack_shift
;
592 stack_top
= arch_align_stack(stack_top
);
593 stack_top
= PAGE_ALIGN(stack_top
);
594 stack_shift
= vma
->vm_end
- stack_top
;
596 bprm
->p
-= stack_shift
;
597 mm
->arg_start
= bprm
->p
;
601 bprm
->loader
-= stack_shift
;
602 bprm
->exec
-= stack_shift
;
604 down_write(&mm
->mmap_sem
);
605 vm_flags
= VM_STACK_FLAGS
;
608 * Adjust stack execute permissions; explicitly enable for
609 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
610 * (arch default) otherwise.
612 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
614 else if (executable_stack
== EXSTACK_DISABLE_X
)
615 vm_flags
&= ~VM_EXEC
;
616 vm_flags
|= mm
->def_flags
;
618 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
624 /* Move stack pages down in memory. */
626 ret
= shift_arg_pages(vma
, stack_shift
);
628 up_write(&mm
->mmap_sem
);
633 #ifdef CONFIG_STACK_GROWSUP
634 stack_base
= vma
->vm_end
+ EXTRA_STACK_VM_PAGES
* PAGE_SIZE
;
636 stack_base
= vma
->vm_start
- EXTRA_STACK_VM_PAGES
* PAGE_SIZE
;
638 ret
= expand_stack(vma
, stack_base
);
643 up_write(&mm
->mmap_sem
);
646 EXPORT_SYMBOL(setup_arg_pages
);
648 #endif /* CONFIG_MMU */
650 struct file
*open_exec(const char *name
)
655 file
= do_filp_open(AT_FDCWD
, name
,
656 O_LARGEFILE
| O_RDONLY
| FMODE_EXEC
, 0,
657 MAY_EXEC
| MAY_OPEN
);
662 if (!S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
))
665 if (file
->f_path
.mnt
->mnt_flags
& MNT_NOEXEC
)
670 err
= deny_write_access(file
);
681 EXPORT_SYMBOL(open_exec
);
683 int kernel_read(struct file
*file
, loff_t offset
,
684 char *addr
, unsigned long count
)
692 /* The cast to a user pointer is valid due to the set_fs() */
693 result
= vfs_read(file
, (void __user
*)addr
, count
, &pos
);
698 EXPORT_SYMBOL(kernel_read
);
700 static int exec_mmap(struct mm_struct
*mm
)
702 struct task_struct
*tsk
;
703 struct mm_struct
* old_mm
, *active_mm
;
705 /* Notify parent that we're no longer interested in the old VM */
707 old_mm
= current
->mm
;
708 mm_release(tsk
, old_mm
);
712 * Make sure that if there is a core dump in progress
713 * for the old mm, we get out and die instead of going
714 * through with the exec. We must hold mmap_sem around
715 * checking core_state and changing tsk->mm.
717 down_read(&old_mm
->mmap_sem
);
718 if (unlikely(old_mm
->core_state
)) {
719 up_read(&old_mm
->mmap_sem
);
724 active_mm
= tsk
->active_mm
;
727 activate_mm(active_mm
, mm
);
729 arch_pick_mmap_layout(mm
);
731 up_read(&old_mm
->mmap_sem
);
732 BUG_ON(active_mm
!= old_mm
);
733 mm_update_next_owner(old_mm
);
742 * This function makes sure the current process has its own signal table,
743 * so that flush_signal_handlers can later reset the handlers without
744 * disturbing other processes. (Other processes might share the signal
745 * table via the CLONE_SIGHAND option to clone().)
747 static int de_thread(struct task_struct
*tsk
)
749 struct signal_struct
*sig
= tsk
->signal
;
750 struct sighand_struct
*oldsighand
= tsk
->sighand
;
751 spinlock_t
*lock
= &oldsighand
->siglock
;
754 if (thread_group_empty(tsk
))
755 goto no_thread_group
;
758 * Kill all other threads in the thread group.
761 if (signal_group_exit(sig
)) {
763 * Another group action in progress, just
764 * return so that the signal is processed.
766 spin_unlock_irq(lock
);
769 sig
->group_exit_task
= tsk
;
770 zap_other_threads(tsk
);
772 /* Account for the thread group leader hanging around: */
773 count
= thread_group_leader(tsk
) ? 1 : 2;
774 sig
->notify_count
= count
;
775 while (atomic_read(&sig
->count
) > count
) {
776 __set_current_state(TASK_UNINTERRUPTIBLE
);
777 spin_unlock_irq(lock
);
781 spin_unlock_irq(lock
);
784 * At this point all other threads have exited, all we have to
785 * do is to wait for the thread group leader to become inactive,
786 * and to assume its PID:
788 if (!thread_group_leader(tsk
)) {
789 struct task_struct
*leader
= tsk
->group_leader
;
791 sig
->notify_count
= -1; /* for exit_notify() */
793 write_lock_irq(&tasklist_lock
);
794 if (likely(leader
->exit_state
))
796 __set_current_state(TASK_UNINTERRUPTIBLE
);
797 write_unlock_irq(&tasklist_lock
);
802 * The only record we have of the real-time age of a
803 * process, regardless of execs it's done, is start_time.
804 * All the past CPU time is accumulated in signal_struct
805 * from sister threads now dead. But in this non-leader
806 * exec, nothing survives from the original leader thread,
807 * whose birth marks the true age of this process now.
808 * When we take on its identity by switching to its PID, we
809 * also take its birthdate (always earlier than our own).
811 tsk
->start_time
= leader
->start_time
;
813 BUG_ON(!same_thread_group(leader
, tsk
));
814 BUG_ON(has_group_leader_pid(tsk
));
816 * An exec() starts a new thread group with the
817 * TGID of the previous thread group. Rehash the
818 * two threads with a switched PID, and release
819 * the former thread group leader:
822 /* Become a process group leader with the old leader's pid.
823 * The old leader becomes a thread of the this thread group.
824 * Note: The old leader also uses this pid until release_task
825 * is called. Odd but simple and correct.
827 detach_pid(tsk
, PIDTYPE_PID
);
828 tsk
->pid
= leader
->pid
;
829 attach_pid(tsk
, PIDTYPE_PID
, task_pid(leader
));
830 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
831 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
833 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
834 list_replace_init(&leader
->sibling
, &tsk
->sibling
);
836 tsk
->group_leader
= tsk
;
837 leader
->group_leader
= tsk
;
839 tsk
->exit_signal
= SIGCHLD
;
841 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
842 leader
->exit_state
= EXIT_DEAD
;
843 write_unlock_irq(&tasklist_lock
);
845 release_task(leader
);
848 sig
->group_exit_task
= NULL
;
849 sig
->notify_count
= 0;
853 setmax_mm_hiwater_rss(&sig
->maxrss
, current
->mm
);
856 flush_itimer_signals();
858 if (atomic_read(&oldsighand
->count
) != 1) {
859 struct sighand_struct
*newsighand
;
861 * This ->sighand is shared with the CLONE_SIGHAND
862 * but not CLONE_THREAD task, switch to the new one.
864 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
868 atomic_set(&newsighand
->count
, 1);
869 memcpy(newsighand
->action
, oldsighand
->action
,
870 sizeof(newsighand
->action
));
872 write_lock_irq(&tasklist_lock
);
873 spin_lock(&oldsighand
->siglock
);
874 rcu_assign_pointer(tsk
->sighand
, newsighand
);
875 spin_unlock(&oldsighand
->siglock
);
876 write_unlock_irq(&tasklist_lock
);
878 __cleanup_sighand(oldsighand
);
881 BUG_ON(!thread_group_leader(tsk
));
886 * These functions flushes out all traces of the currently running executable
887 * so that a new one can be started
889 static void flush_old_files(struct files_struct
* files
)
894 spin_lock(&files
->file_lock
);
896 unsigned long set
, i
;
900 fdt
= files_fdtable(files
);
901 if (i
>= fdt
->max_fds
)
903 set
= fdt
->close_on_exec
->fds_bits
[j
];
906 fdt
->close_on_exec
->fds_bits
[j
] = 0;
907 spin_unlock(&files
->file_lock
);
908 for ( ; set
; i
++,set
>>= 1) {
913 spin_lock(&files
->file_lock
);
916 spin_unlock(&files
->file_lock
);
919 char *get_task_comm(char *buf
, struct task_struct
*tsk
)
921 /* buf must be at least sizeof(tsk->comm) in size */
923 strncpy(buf
, tsk
->comm
, sizeof(tsk
->comm
));
928 void set_task_comm(struct task_struct
*tsk
, char *buf
)
931 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
933 perf_event_comm(tsk
);
936 int flush_old_exec(struct linux_binprm
* bprm
)
940 char tcomm
[sizeof(current
->comm
)];
943 * Make sure we have a private signal table and that
944 * we are unassociated from the previous thread group.
946 retval
= de_thread(current
);
950 set_mm_exe_file(bprm
->mm
, bprm
->file
);
953 * Release all of the old mmap stuff
955 retval
= exec_mmap(bprm
->mm
);
959 bprm
->mm
= NULL
; /* We're using it now */
961 /* This is the point of no return */
962 current
->sas_ss_sp
= current
->sas_ss_size
= 0;
964 if (current_euid() == current_uid() && current_egid() == current_gid())
965 set_dumpable(current
->mm
, 1);
967 set_dumpable(current
->mm
, suid_dumpable
);
969 name
= bprm
->filename
;
971 /* Copies the binary name from after last slash */
972 for (i
=0; (ch
= *(name
++)) != '\0';) {
974 i
= 0; /* overwrite what we wrote */
976 if (i
< (sizeof(tcomm
) - 1))
980 set_task_comm(current
, tcomm
);
982 current
->flags
&= ~PF_RANDOMIZE
;
985 /* Set the new mm task size. We have to do that late because it may
986 * depend on TIF_32BIT which is only updated in flush_thread() on
987 * some architectures like powerpc
989 current
->mm
->task_size
= TASK_SIZE
;
991 /* install the new credentials */
992 if (bprm
->cred
->uid
!= current_euid() ||
993 bprm
->cred
->gid
!= current_egid()) {
994 current
->pdeath_signal
= 0;
995 } else if (file_permission(bprm
->file
, MAY_READ
) ||
996 bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
) {
997 set_dumpable(current
->mm
, suid_dumpable
);
1000 current
->personality
&= ~bprm
->per_clear
;
1003 * Flush performance counters when crossing a
1006 if (!get_dumpable(current
->mm
))
1007 perf_event_exit_task(current
);
1009 /* An exec changes our domain. We are no longer part of the thread
1012 current
->self_exec_id
++;
1014 flush_signal_handlers(current
, 0);
1015 flush_old_files(current
->files
);
1023 EXPORT_SYMBOL(flush_old_exec
);
1026 * Prepare credentials and lock ->cred_guard_mutex.
1027 * install_exec_creds() commits the new creds and drops the lock.
1028 * Or, if exec fails before, free_bprm() should release ->cred and
1031 int prepare_bprm_creds(struct linux_binprm
*bprm
)
1033 if (mutex_lock_interruptible(¤t
->cred_guard_mutex
))
1034 return -ERESTARTNOINTR
;
1036 bprm
->cred
= prepare_exec_creds();
1037 if (likely(bprm
->cred
))
1040 mutex_unlock(¤t
->cred_guard_mutex
);
1044 void free_bprm(struct linux_binprm
*bprm
)
1046 free_arg_pages(bprm
);
1048 mutex_unlock(¤t
->cred_guard_mutex
);
1049 abort_creds(bprm
->cred
);
1055 * install the new credentials for this executable
1057 void install_exec_creds(struct linux_binprm
*bprm
)
1059 security_bprm_committing_creds(bprm
);
1061 commit_creds(bprm
->cred
);
1064 * cred_guard_mutex must be held at least to this point to prevent
1065 * ptrace_attach() from altering our determination of the task's
1066 * credentials; any time after this it may be unlocked.
1068 security_bprm_committed_creds(bprm
);
1069 mutex_unlock(¤t
->cred_guard_mutex
);
1071 EXPORT_SYMBOL(install_exec_creds
);
1074 * determine how safe it is to execute the proposed program
1075 * - the caller must hold current->cred_guard_mutex to protect against
1078 int check_unsafe_exec(struct linux_binprm
*bprm
)
1080 struct task_struct
*p
= current
, *t
;
1084 bprm
->unsafe
= tracehook_unsafe_exec(p
);
1087 write_lock(&p
->fs
->lock
);
1089 for (t
= next_thread(p
); t
!= p
; t
= next_thread(t
)) {
1095 if (p
->fs
->users
> n_fs
) {
1096 bprm
->unsafe
|= LSM_UNSAFE_SHARE
;
1099 if (!p
->fs
->in_exec
) {
1104 write_unlock(&p
->fs
->lock
);
1110 * Fill the binprm structure from the inode.
1111 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1113 * This may be called multiple times for binary chains (scripts for example).
1115 int prepare_binprm(struct linux_binprm
*bprm
)
1118 struct inode
* inode
= bprm
->file
->f_path
.dentry
->d_inode
;
1121 mode
= inode
->i_mode
;
1122 if (bprm
->file
->f_op
== NULL
)
1125 /* clear any previous set[ug]id data from a previous binary */
1126 bprm
->cred
->euid
= current_euid();
1127 bprm
->cred
->egid
= current_egid();
1129 if (!(bprm
->file
->f_path
.mnt
->mnt_flags
& MNT_NOSUID
)) {
1131 if (mode
& S_ISUID
) {
1132 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1133 bprm
->cred
->euid
= inode
->i_uid
;
1138 * If setgid is set but no group execute bit then this
1139 * is a candidate for mandatory locking, not a setgid
1142 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1143 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
1144 bprm
->cred
->egid
= inode
->i_gid
;
1148 /* fill in binprm security blob */
1149 retval
= security_bprm_set_creds(bprm
);
1152 bprm
->cred_prepared
= 1;
1154 memset(bprm
->buf
, 0, BINPRM_BUF_SIZE
);
1155 return kernel_read(bprm
->file
, 0, bprm
->buf
, BINPRM_BUF_SIZE
);
1158 EXPORT_SYMBOL(prepare_binprm
);
1161 * Arguments are '\0' separated strings found at the location bprm->p
1162 * points to; chop off the first by relocating brpm->p to right after
1163 * the first '\0' encountered.
1165 int remove_arg_zero(struct linux_binprm
*bprm
)
1168 unsigned long offset
;
1176 offset
= bprm
->p
& ~PAGE_MASK
;
1177 page
= get_arg_page(bprm
, bprm
->p
, 0);
1182 kaddr
= kmap_atomic(page
, KM_USER0
);
1184 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1185 offset
++, bprm
->p
++)
1188 kunmap_atomic(kaddr
, KM_USER0
);
1191 if (offset
== PAGE_SIZE
)
1192 free_arg_page(bprm
, (bprm
->p
>> PAGE_SHIFT
) - 1);
1193 } while (offset
== PAGE_SIZE
);
1202 EXPORT_SYMBOL(remove_arg_zero
);
1205 * cycle the list of binary formats handler, until one recognizes the image
1207 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
1209 unsigned int depth
= bprm
->recursion_depth
;
1211 struct linux_binfmt
*fmt
;
1213 retval
= security_bprm_check(bprm
);
1216 retval
= ima_bprm_check(bprm
);
1220 /* kernel module loader fixup */
1221 /* so we don't try to load run modprobe in kernel space. */
1224 retval
= audit_bprm(bprm
);
1229 for (try=0; try<2; try++) {
1230 read_lock(&binfmt_lock
);
1231 list_for_each_entry(fmt
, &formats
, lh
) {
1232 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
1235 if (!try_module_get(fmt
->module
))
1237 read_unlock(&binfmt_lock
);
1238 retval
= fn(bprm
, regs
);
1240 * Restore the depth counter to its starting value
1241 * in this call, so we don't have to rely on every
1242 * load_binary function to restore it on return.
1244 bprm
->recursion_depth
= depth
;
1247 tracehook_report_exec(fmt
, bprm
, regs
);
1249 allow_write_access(bprm
->file
);
1253 current
->did_exec
= 1;
1254 proc_exec_connector(current
);
1257 read_lock(&binfmt_lock
);
1259 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
)
1262 read_unlock(&binfmt_lock
);
1266 read_unlock(&binfmt_lock
);
1267 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
) {
1269 #ifdef CONFIG_MODULES
1271 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1272 if (printable(bprm
->buf
[0]) &&
1273 printable(bprm
->buf
[1]) &&
1274 printable(bprm
->buf
[2]) &&
1275 printable(bprm
->buf
[3]))
1276 break; /* -ENOEXEC */
1277 request_module("binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
1284 EXPORT_SYMBOL(search_binary_handler
);
1287 * sys_execve() executes a new program.
1289 int do_execve(char * filename
,
1290 char __user
*__user
*argv
,
1291 char __user
*__user
*envp
,
1292 struct pt_regs
* regs
)
1294 struct linux_binprm
*bprm
;
1296 struct files_struct
*displaced
;
1300 retval
= unshare_files(&displaced
);
1305 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1309 retval
= prepare_bprm_creds(bprm
);
1313 retval
= check_unsafe_exec(bprm
);
1316 clear_in_exec
= retval
;
1317 current
->in_execve
= 1;
1319 file
= open_exec(filename
);
1320 retval
= PTR_ERR(file
);
1327 bprm
->filename
= filename
;
1328 bprm
->interp
= filename
;
1330 retval
= bprm_mm_init(bprm
);
1334 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
1335 if ((retval
= bprm
->argc
) < 0)
1338 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
1339 if ((retval
= bprm
->envc
) < 0)
1342 retval
= prepare_binprm(bprm
);
1346 retval
= copy_strings_kernel(1, &bprm
->filename
, bprm
);
1350 bprm
->exec
= bprm
->p
;
1351 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1355 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1359 current
->flags
&= ~PF_KTHREAD
;
1360 retval
= search_binary_handler(bprm
,regs
);
1364 current
->stack_start
= current
->mm
->start_stack
;
1366 /* execve succeeded */
1367 current
->fs
->in_exec
= 0;
1368 current
->in_execve
= 0;
1369 acct_update_integrals(current
);
1372 put_files_struct(displaced
);
1381 allow_write_access(bprm
->file
);
1387 current
->fs
->in_exec
= 0;
1388 current
->in_execve
= 0;
1395 reset_files_struct(displaced
);
1400 void set_binfmt(struct linux_binfmt
*new)
1402 struct mm_struct
*mm
= current
->mm
;
1405 module_put(mm
->binfmt
->module
);
1409 __module_get(new->module
);
1412 EXPORT_SYMBOL(set_binfmt
);
1414 /* format_corename will inspect the pattern parameter, and output a
1415 * name into corename, which must have space for at least
1416 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1418 static int format_corename(char *corename
, long signr
)
1420 const struct cred
*cred
= current_cred();
1421 const char *pat_ptr
= core_pattern
;
1422 int ispipe
= (*pat_ptr
== '|');
1423 char *out_ptr
= corename
;
1424 char *const out_end
= corename
+ CORENAME_MAX_SIZE
;
1426 int pid_in_pattern
= 0;
1428 /* Repeat as long as we have more pattern to process and more output
1431 if (*pat_ptr
!= '%') {
1432 if (out_ptr
== out_end
)
1434 *out_ptr
++ = *pat_ptr
++;
1436 switch (*++pat_ptr
) {
1439 /* Double percent, output one percent */
1441 if (out_ptr
== out_end
)
1448 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1449 "%d", task_tgid_vnr(current
));
1450 if (rc
> out_end
- out_ptr
)
1456 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1458 if (rc
> out_end
- out_ptr
)
1464 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1466 if (rc
> out_end
- out_ptr
)
1470 /* signal that caused the coredump */
1472 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1474 if (rc
> out_end
- out_ptr
)
1478 /* UNIX time of coredump */
1481 do_gettimeofday(&tv
);
1482 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1484 if (rc
> out_end
- out_ptr
)
1491 down_read(&uts_sem
);
1492 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1493 "%s", utsname()->nodename
);
1495 if (rc
> out_end
- out_ptr
)
1501 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1502 "%s", current
->comm
);
1503 if (rc
> out_end
- out_ptr
)
1507 /* core limit size */
1509 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1510 "%lu", current
->signal
->rlim
[RLIMIT_CORE
].rlim_cur
);
1511 if (rc
> out_end
- out_ptr
)
1521 /* Backward compatibility with core_uses_pid:
1523 * If core_pattern does not include a %p (as is the default)
1524 * and core_uses_pid is set, then .%pid will be appended to
1525 * the filename. Do not do this for piped commands. */
1526 if (!ispipe
&& !pid_in_pattern
&& core_uses_pid
) {
1527 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1528 ".%d", task_tgid_vnr(current
));
1529 if (rc
> out_end
- out_ptr
)
1538 static int zap_process(struct task_struct
*start
)
1540 struct task_struct
*t
;
1543 start
->signal
->flags
= SIGNAL_GROUP_EXIT
;
1544 start
->signal
->group_stop_count
= 0;
1548 if (t
!= current
&& t
->mm
) {
1549 sigaddset(&t
->pending
.signal
, SIGKILL
);
1550 signal_wake_up(t
, 1);
1553 } while_each_thread(start
, t
);
1558 static inline int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
1559 struct core_state
*core_state
, int exit_code
)
1561 struct task_struct
*g
, *p
;
1562 unsigned long flags
;
1565 spin_lock_irq(&tsk
->sighand
->siglock
);
1566 if (!signal_group_exit(tsk
->signal
)) {
1567 mm
->core_state
= core_state
;
1568 tsk
->signal
->group_exit_code
= exit_code
;
1569 nr
= zap_process(tsk
);
1571 spin_unlock_irq(&tsk
->sighand
->siglock
);
1572 if (unlikely(nr
< 0))
1575 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
1578 * We should find and kill all tasks which use this mm, and we should
1579 * count them correctly into ->nr_threads. We don't take tasklist
1580 * lock, but this is safe wrt:
1583 * None of sub-threads can fork after zap_process(leader). All
1584 * processes which were created before this point should be
1585 * visible to zap_threads() because copy_process() adds the new
1586 * process to the tail of init_task.tasks list, and lock/unlock
1587 * of ->siglock provides a memory barrier.
1590 * The caller holds mm->mmap_sem. This means that the task which
1591 * uses this mm can't pass exit_mm(), so it can't exit or clear
1595 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1596 * we must see either old or new leader, this does not matter.
1597 * However, it can change p->sighand, so lock_task_sighand(p)
1598 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1601 * Note also that "g" can be the old leader with ->mm == NULL
1602 * and already unhashed and thus removed from ->thread_group.
1603 * This is OK, __unhash_process()->list_del_rcu() does not
1604 * clear the ->next pointer, we will find the new leader via
1608 for_each_process(g
) {
1609 if (g
== tsk
->group_leader
)
1611 if (g
->flags
& PF_KTHREAD
)
1616 if (unlikely(p
->mm
== mm
)) {
1617 lock_task_sighand(p
, &flags
);
1618 nr
+= zap_process(p
);
1619 unlock_task_sighand(p
, &flags
);
1623 } while_each_thread(g
, p
);
1627 atomic_set(&core_state
->nr_threads
, nr
);
1631 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
1633 struct task_struct
*tsk
= current
;
1634 struct mm_struct
*mm
= tsk
->mm
;
1635 struct completion
*vfork_done
;
1638 init_completion(&core_state
->startup
);
1639 core_state
->dumper
.task
= tsk
;
1640 core_state
->dumper
.next
= NULL
;
1641 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
1642 up_write(&mm
->mmap_sem
);
1644 if (unlikely(core_waiters
< 0))
1648 * Make sure nobody is waiting for us to release the VM,
1649 * otherwise we can deadlock when we wait on each other
1651 vfork_done
= tsk
->vfork_done
;
1653 tsk
->vfork_done
= NULL
;
1654 complete(vfork_done
);
1658 wait_for_completion(&core_state
->startup
);
1660 return core_waiters
;
1663 static void coredump_finish(struct mm_struct
*mm
)
1665 struct core_thread
*curr
, *next
;
1666 struct task_struct
*task
;
1668 next
= mm
->core_state
->dumper
.next
;
1669 while ((curr
= next
) != NULL
) {
1673 * see exit_mm(), curr->task must not see
1674 * ->task == NULL before we read ->next.
1678 wake_up_process(task
);
1681 mm
->core_state
= NULL
;
1685 * set_dumpable converts traditional three-value dumpable to two flags and
1686 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1687 * these bits are not changed atomically. So get_dumpable can observe the
1688 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1689 * return either old dumpable or new one by paying attention to the order of
1690 * modifying the bits.
1692 * dumpable | mm->flags (binary)
1693 * old new | initial interim final
1694 * ---------+-----------------------
1702 * (*) get_dumpable regards interim value of 10 as 11.
1704 void set_dumpable(struct mm_struct
*mm
, int value
)
1708 clear_bit(MMF_DUMPABLE
, &mm
->flags
);
1710 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1713 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1715 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1718 set_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1720 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1725 int get_dumpable(struct mm_struct
*mm
)
1729 ret
= mm
->flags
& 0x3;
1730 return (ret
>= 2) ? 2 : ret
;
1733 static void wait_for_dump_helpers(struct file
*file
)
1735 struct pipe_inode_info
*pipe
;
1737 pipe
= file
->f_path
.dentry
->d_inode
->i_pipe
;
1743 while ((pipe
->readers
> 1) && (!signal_pending(current
))) {
1744 wake_up_interruptible_sync(&pipe
->wait
);
1745 kill_fasync(&pipe
->fasync_readers
, SIGIO
, POLL_IN
);
1756 void do_coredump(long signr
, int exit_code
, struct pt_regs
*regs
)
1758 struct core_state core_state
;
1759 char corename
[CORENAME_MAX_SIZE
+ 1];
1760 struct mm_struct
*mm
= current
->mm
;
1761 struct linux_binfmt
* binfmt
;
1762 struct inode
* inode
;
1764 const struct cred
*old_cred
;
1769 unsigned long core_limit
= current
->signal
->rlim
[RLIMIT_CORE
].rlim_cur
;
1770 char **helper_argv
= NULL
;
1771 int helper_argc
= 0;
1773 static atomic_t core_dump_count
= ATOMIC_INIT(0);
1775 audit_core_dumps(signr
);
1777 binfmt
= mm
->binfmt
;
1778 if (!binfmt
|| !binfmt
->core_dump
)
1781 cred
= prepare_creds();
1787 down_write(&mm
->mmap_sem
);
1789 * If another thread got here first, or we are not dumpable, bail out.
1791 if (mm
->core_state
|| !get_dumpable(mm
)) {
1792 up_write(&mm
->mmap_sem
);
1798 * We cannot trust fsuid as being the "true" uid of the
1799 * process nor do we know its entire history. We only know it
1800 * was tainted so we dump it as root in mode 2.
1802 if (get_dumpable(mm
) == 2) { /* Setuid core dump mode */
1803 flag
= O_EXCL
; /* Stop rewrite attacks */
1804 cred
->fsuid
= 0; /* Dump root private */
1807 retval
= coredump_wait(exit_code
, &core_state
);
1813 old_cred
= override_creds(cred
);
1816 * Clear any false indication of pending signals that might
1817 * be seen by the filesystem code called to write the core file.
1819 clear_thread_flag(TIF_SIGPENDING
);
1822 * lock_kernel() because format_corename() is controlled by sysctl, which
1823 * uses lock_kernel()
1826 ispipe
= format_corename(corename
, signr
);
1829 if ((!ispipe
) && (core_limit
< binfmt
->min_coredump
))
1833 if (core_limit
== 0) {
1835 * Normally core limits are irrelevant to pipes, since
1836 * we're not writing to the file system, but we use
1837 * core_limit of 0 here as a speacial value. Any
1838 * non-zero limit gets set to RLIM_INFINITY below, but
1839 * a limit of 0 skips the dump. This is a consistent
1840 * way to catch recursive crashes. We can still crash
1841 * if the core_pattern binary sets RLIM_CORE = !0
1842 * but it runs as root, and can do lots of stupid things
1843 * Note that we use task_tgid_vnr here to grab the pid
1844 * of the process group leader. That way we get the
1845 * right pid if a thread in a multi-threaded
1846 * core_pattern process dies.
1849 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1850 task_tgid_vnr(current
), current
->comm
);
1851 printk(KERN_WARNING
"Aborting core\n");
1855 dump_count
= atomic_inc_return(&core_dump_count
);
1856 if (core_pipe_limit
&& (core_pipe_limit
< dump_count
)) {
1857 printk(KERN_WARNING
"Pid %d(%s) over core_pipe_limit\n",
1858 task_tgid_vnr(current
), current
->comm
);
1859 printk(KERN_WARNING
"Skipping core dump\n");
1860 goto fail_dropcount
;
1863 helper_argv
= argv_split(GFP_KERNEL
, corename
+1, &helper_argc
);
1865 printk(KERN_WARNING
"%s failed to allocate memory\n",
1867 goto fail_dropcount
;
1870 core_limit
= RLIM_INFINITY
;
1872 /* SIGPIPE can happen, but it's just never processed */
1873 if (call_usermodehelper_pipe(helper_argv
[0], helper_argv
, NULL
,
1875 printk(KERN_INFO
"Core dump to %s pipe failed\n",
1877 goto fail_dropcount
;
1880 file
= filp_open(corename
,
1881 O_CREAT
| 2 | O_NOFOLLOW
| O_LARGEFILE
| flag
,
1884 goto fail_dropcount
;
1885 inode
= file
->f_path
.dentry
->d_inode
;
1886 if (inode
->i_nlink
> 1)
1887 goto close_fail
; /* multiple links - don't dump */
1888 if (!ispipe
&& d_unhashed(file
->f_path
.dentry
))
1891 /* AK: actually i see no reason to not allow this for named pipes etc.,
1892 but keep the previous behaviour for now. */
1893 if (!ispipe
&& !S_ISREG(inode
->i_mode
))
1896 * Dont allow local users get cute and trick others to coredump
1897 * into their pre-created files:
1899 if (inode
->i_uid
!= current_fsuid())
1903 if (!file
->f_op
->write
)
1905 if (!ispipe
&& do_truncate(file
->f_path
.dentry
, 0, 0, file
) != 0)
1908 retval
= binfmt
->core_dump(signr
, regs
, file
, core_limit
);
1911 current
->signal
->group_exit_code
|= 0x80;
1913 if (ispipe
&& core_pipe_limit
)
1914 wait_for_dump_helpers(file
);
1915 filp_close(file
, NULL
);
1918 atomic_dec(&core_dump_count
);
1921 argv_free(helper_argv
);
1923 revert_creds(old_cred
);
1925 coredump_finish(mm
);