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/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
54 #include <asm/uaccess.h>
55 #include <asm/mmu_context.h>
59 #include <linux/kmod.h>
63 /* for /sbin/loader handling in search_binary_handler() */
64 #include <linux/a.out.h>
68 char core_pattern
[CORENAME_MAX_SIZE
] = "core";
69 int suid_dumpable
= 0;
71 /* The maximal length of core_pattern is also specified in sysctl.c */
73 static LIST_HEAD(formats
);
74 static DEFINE_RWLOCK(binfmt_lock
);
76 int register_binfmt(struct linux_binfmt
* fmt
)
80 write_lock(&binfmt_lock
);
81 list_add(&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 asmlinkage
long sys_uselib(const char __user
* library
)
112 char *tmp
= getname(library
);
113 int error
= PTR_ERR(tmp
);
116 error
= path_lookup_open(AT_FDCWD
, tmp
,
118 FMODE_READ
|FMODE_EXEC
);
125 if (!S_ISREG(nd
.path
.dentry
->d_inode
->i_mode
))
129 if (nd
.path
.mnt
->mnt_flags
& MNT_NOEXEC
)
132 error
= vfs_permission(&nd
, MAY_READ
| MAY_EXEC
| MAY_OPEN
);
136 file
= nameidata_to_filp(&nd
, O_RDONLY
|O_LARGEFILE
);
137 error
= PTR_ERR(file
);
143 struct linux_binfmt
* fmt
;
145 read_lock(&binfmt_lock
);
146 list_for_each_entry(fmt
, &formats
, lh
) {
147 if (!fmt
->load_shlib
)
149 if (!try_module_get(fmt
->module
))
151 read_unlock(&binfmt_lock
);
152 error
= fmt
->load_shlib(file
);
153 read_lock(&binfmt_lock
);
155 if (error
!= -ENOEXEC
)
158 read_unlock(&binfmt_lock
);
164 release_open_intent(&nd
);
171 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
177 #ifdef CONFIG_STACK_GROWSUP
179 ret
= expand_stack_downwards(bprm
->vma
, pos
);
184 ret
= get_user_pages(current
, bprm
->mm
, pos
,
185 1, write
, 1, &page
, NULL
);
190 unsigned long size
= bprm
->vma
->vm_end
- bprm
->vma
->vm_start
;
194 * We've historically supported up to 32 pages (ARG_MAX)
195 * of argument strings even with small stacks
201 * Limit to 1/4-th the stack size for the argv+env strings.
203 * - the remaining binfmt code will not run out of stack space,
204 * - the program will have a reasonable amount of stack left
207 rlim
= current
->signal
->rlim
;
208 if (size
> rlim
[RLIMIT_STACK
].rlim_cur
/ 4) {
217 static void put_arg_page(struct page
*page
)
222 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
226 static void free_arg_pages(struct linux_binprm
*bprm
)
230 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
233 flush_cache_page(bprm
->vma
, pos
, page_to_pfn(page
));
236 static int __bprm_mm_init(struct linux_binprm
*bprm
)
239 struct vm_area_struct
*vma
= NULL
;
240 struct mm_struct
*mm
= bprm
->mm
;
242 bprm
->vma
= vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
246 down_write(&mm
->mmap_sem
);
250 * Place the stack at the largest stack address the architecture
251 * supports. Later, we'll move this to an appropriate place. We don't
252 * use STACK_TOP because that can depend on attributes which aren't
255 vma
->vm_end
= STACK_TOP_MAX
;
256 vma
->vm_start
= vma
->vm_end
- PAGE_SIZE
;
258 vma
->vm_flags
= VM_STACK_FLAGS
;
259 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
260 err
= insert_vm_struct(mm
, vma
);
262 up_write(&mm
->mmap_sem
);
266 mm
->stack_vm
= mm
->total_vm
= 1;
267 up_write(&mm
->mmap_sem
);
269 bprm
->p
= vma
->vm_end
- sizeof(void *);
276 kmem_cache_free(vm_area_cachep
, vma
);
282 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
284 return len
<= MAX_ARG_STRLEN
;
289 static struct page
*get_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
294 page
= bprm
->page
[pos
/ PAGE_SIZE
];
295 if (!page
&& write
) {
296 page
= alloc_page(GFP_HIGHUSER
|__GFP_ZERO
);
299 bprm
->page
[pos
/ PAGE_SIZE
] = page
;
305 static void put_arg_page(struct page
*page
)
309 static void free_arg_page(struct linux_binprm
*bprm
, int i
)
312 __free_page(bprm
->page
[i
]);
313 bprm
->page
[i
] = NULL
;
317 static void free_arg_pages(struct linux_binprm
*bprm
)
321 for (i
= 0; i
< MAX_ARG_PAGES
; i
++)
322 free_arg_page(bprm
, i
);
325 static void flush_arg_page(struct linux_binprm
*bprm
, unsigned long pos
,
330 static int __bprm_mm_init(struct linux_binprm
*bprm
)
332 bprm
->p
= PAGE_SIZE
* MAX_ARG_PAGES
- sizeof(void *);
336 static bool valid_arg_len(struct linux_binprm
*bprm
, long len
)
338 return len
<= bprm
->p
;
341 #endif /* CONFIG_MMU */
344 * Create a new mm_struct and populate it with a temporary stack
345 * vm_area_struct. We don't have enough context at this point to set the stack
346 * flags, permissions, and offset, so we use temporary values. We'll update
347 * them later in setup_arg_pages().
349 int bprm_mm_init(struct linux_binprm
*bprm
)
352 struct mm_struct
*mm
= NULL
;
354 bprm
->mm
= mm
= mm_alloc();
359 err
= init_new_context(current
, mm
);
363 err
= __bprm_mm_init(bprm
);
379 * count() counts the number of strings in array ARGV.
381 static int count(char __user
* __user
* argv
, int max
)
389 if (get_user(p
, argv
))
403 * 'copy_strings()' copies argument/environment strings from the old
404 * processes's memory to the new process's stack. The call to get_user_pages()
405 * ensures the destination page is created and not swapped out.
407 static int copy_strings(int argc
, char __user
* __user
* argv
,
408 struct linux_binprm
*bprm
)
410 struct page
*kmapped_page
= NULL
;
412 unsigned long kpos
= 0;
420 if (get_user(str
, argv
+argc
) ||
421 !(len
= strnlen_user(str
, MAX_ARG_STRLEN
))) {
426 if (!valid_arg_len(bprm
, len
)) {
431 /* We're going to work our way backwords. */
437 int offset
, bytes_to_copy
;
439 offset
= pos
% PAGE_SIZE
;
443 bytes_to_copy
= offset
;
444 if (bytes_to_copy
> len
)
447 offset
-= bytes_to_copy
;
448 pos
-= bytes_to_copy
;
449 str
-= bytes_to_copy
;
450 len
-= bytes_to_copy
;
452 if (!kmapped_page
|| kpos
!= (pos
& PAGE_MASK
)) {
455 page
= get_arg_page(bprm
, pos
, 1);
462 flush_kernel_dcache_page(kmapped_page
);
463 kunmap(kmapped_page
);
464 put_arg_page(kmapped_page
);
467 kaddr
= kmap(kmapped_page
);
468 kpos
= pos
& PAGE_MASK
;
469 flush_arg_page(bprm
, kpos
, kmapped_page
);
471 if (copy_from_user(kaddr
+offset
, str
, bytes_to_copy
)) {
480 flush_kernel_dcache_page(kmapped_page
);
481 kunmap(kmapped_page
);
482 put_arg_page(kmapped_page
);
488 * Like copy_strings, but get argv and its values from kernel memory.
490 int copy_strings_kernel(int argc
,char ** argv
, struct linux_binprm
*bprm
)
493 mm_segment_t oldfs
= get_fs();
495 r
= copy_strings(argc
, (char __user
* __user
*)argv
, bprm
);
499 EXPORT_SYMBOL(copy_strings_kernel
);
504 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
505 * the binfmt code determines where the new stack should reside, we shift it to
506 * its final location. The process proceeds as follows:
508 * 1) Use shift to calculate the new vma endpoints.
509 * 2) Extend vma to cover both the old and new ranges. This ensures the
510 * arguments passed to subsequent functions are consistent.
511 * 3) Move vma's page tables to the new range.
512 * 4) Free up any cleared pgd range.
513 * 5) Shrink the vma to cover only the new range.
515 static int shift_arg_pages(struct vm_area_struct
*vma
, unsigned long shift
)
517 struct mm_struct
*mm
= vma
->vm_mm
;
518 unsigned long old_start
= vma
->vm_start
;
519 unsigned long old_end
= vma
->vm_end
;
520 unsigned long length
= old_end
- old_start
;
521 unsigned long new_start
= old_start
- shift
;
522 unsigned long new_end
= old_end
- shift
;
523 struct mmu_gather
*tlb
;
525 BUG_ON(new_start
> new_end
);
528 * ensure there are no vmas between where we want to go
531 if (vma
!= find_vma(mm
, new_start
))
535 * cover the whole range: [new_start, old_end)
537 vma_adjust(vma
, new_start
, old_end
, vma
->vm_pgoff
, NULL
);
540 * move the page tables downwards, on failure we rely on
541 * process cleanup to remove whatever mess we made.
543 if (length
!= move_page_tables(vma
, old_start
,
544 vma
, new_start
, length
))
548 tlb
= tlb_gather_mmu(mm
, 0);
549 if (new_end
> old_start
) {
551 * when the old and new regions overlap clear from new_end.
553 free_pgd_range(tlb
, new_end
, old_end
, new_end
,
554 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
557 * otherwise, clean from old_start; this is done to not touch
558 * the address space in [new_end, old_start) some architectures
559 * have constraints on va-space that make this illegal (IA64) -
560 * for the others its just a little faster.
562 free_pgd_range(tlb
, old_start
, old_end
, new_end
,
563 vma
->vm_next
? vma
->vm_next
->vm_start
: 0);
565 tlb_finish_mmu(tlb
, new_end
, old_end
);
568 * shrink the vma to just the new range.
570 vma_adjust(vma
, new_start
, new_end
, vma
->vm_pgoff
, NULL
);
575 #define EXTRA_STACK_VM_PAGES 20 /* random */
578 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
579 * the stack is optionally relocated, and some extra space is added.
581 int setup_arg_pages(struct linux_binprm
*bprm
,
582 unsigned long stack_top
,
583 int executable_stack
)
586 unsigned long stack_shift
;
587 struct mm_struct
*mm
= current
->mm
;
588 struct vm_area_struct
*vma
= bprm
->vma
;
589 struct vm_area_struct
*prev
= NULL
;
590 unsigned long vm_flags
;
591 unsigned long stack_base
;
593 #ifdef CONFIG_STACK_GROWSUP
594 /* Limit stack size to 1GB */
595 stack_base
= current
->signal
->rlim
[RLIMIT_STACK
].rlim_max
;
596 if (stack_base
> (1 << 30))
597 stack_base
= 1 << 30;
599 /* Make sure we didn't let the argument array grow too large. */
600 if (vma
->vm_end
- vma
->vm_start
> stack_base
)
603 stack_base
= PAGE_ALIGN(stack_top
- stack_base
);
605 stack_shift
= vma
->vm_start
- stack_base
;
606 mm
->arg_start
= bprm
->p
- stack_shift
;
607 bprm
->p
= vma
->vm_end
- stack_shift
;
609 stack_top
= arch_align_stack(stack_top
);
610 stack_top
= PAGE_ALIGN(stack_top
);
611 stack_shift
= vma
->vm_end
- stack_top
;
613 bprm
->p
-= stack_shift
;
614 mm
->arg_start
= bprm
->p
;
618 bprm
->loader
-= stack_shift
;
619 bprm
->exec
-= stack_shift
;
621 down_write(&mm
->mmap_sem
);
622 vm_flags
= VM_STACK_FLAGS
;
625 * Adjust stack execute permissions; explicitly enable for
626 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
627 * (arch default) otherwise.
629 if (unlikely(executable_stack
== EXSTACK_ENABLE_X
))
631 else if (executable_stack
== EXSTACK_DISABLE_X
)
632 vm_flags
&= ~VM_EXEC
;
633 vm_flags
|= mm
->def_flags
;
635 ret
= mprotect_fixup(vma
, &prev
, vma
->vm_start
, vma
->vm_end
,
641 /* Move stack pages down in memory. */
643 ret
= shift_arg_pages(vma
, stack_shift
);
645 up_write(&mm
->mmap_sem
);
650 #ifdef CONFIG_STACK_GROWSUP
651 stack_base
= vma
->vm_end
+ EXTRA_STACK_VM_PAGES
* PAGE_SIZE
;
653 stack_base
= vma
->vm_start
- EXTRA_STACK_VM_PAGES
* PAGE_SIZE
;
655 ret
= expand_stack(vma
, stack_base
);
660 up_write(&mm
->mmap_sem
);
663 EXPORT_SYMBOL(setup_arg_pages
);
665 #endif /* CONFIG_MMU */
667 struct file
*open_exec(const char *name
)
673 err
= path_lookup_open(AT_FDCWD
, name
, LOOKUP_FOLLOW
, &nd
,
674 FMODE_READ
|FMODE_EXEC
);
679 if (!S_ISREG(nd
.path
.dentry
->d_inode
->i_mode
))
682 if (nd
.path
.mnt
->mnt_flags
& MNT_NOEXEC
)
685 err
= vfs_permission(&nd
, MAY_EXEC
| MAY_OPEN
);
689 file
= nameidata_to_filp(&nd
, O_RDONLY
|O_LARGEFILE
);
693 err
= deny_write_access(file
);
702 release_open_intent(&nd
);
707 EXPORT_SYMBOL(open_exec
);
709 int kernel_read(struct file
*file
, unsigned long offset
,
710 char *addr
, unsigned long count
)
718 /* The cast to a user pointer is valid due to the set_fs() */
719 result
= vfs_read(file
, (void __user
*)addr
, count
, &pos
);
724 EXPORT_SYMBOL(kernel_read
);
726 static int exec_mmap(struct mm_struct
*mm
)
728 struct task_struct
*tsk
;
729 struct mm_struct
* old_mm
, *active_mm
;
731 /* Notify parent that we're no longer interested in the old VM */
733 old_mm
= current
->mm
;
734 mm_release(tsk
, old_mm
);
738 * Make sure that if there is a core dump in progress
739 * for the old mm, we get out and die instead of going
740 * through with the exec. We must hold mmap_sem around
741 * checking core_state and changing tsk->mm.
743 down_read(&old_mm
->mmap_sem
);
744 if (unlikely(old_mm
->core_state
)) {
745 up_read(&old_mm
->mmap_sem
);
750 active_mm
= tsk
->active_mm
;
753 activate_mm(active_mm
, mm
);
755 arch_pick_mmap_layout(mm
);
757 up_read(&old_mm
->mmap_sem
);
758 BUG_ON(active_mm
!= old_mm
);
759 mm_update_next_owner(old_mm
);
768 * This function makes sure the current process has its own signal table,
769 * so that flush_signal_handlers can later reset the handlers without
770 * disturbing other processes. (Other processes might share the signal
771 * table via the CLONE_SIGHAND option to clone().)
773 static int de_thread(struct task_struct
*tsk
)
775 struct signal_struct
*sig
= tsk
->signal
;
776 struct sighand_struct
*oldsighand
= tsk
->sighand
;
777 spinlock_t
*lock
= &oldsighand
->siglock
;
778 struct task_struct
*leader
= NULL
;
781 if (thread_group_empty(tsk
))
782 goto no_thread_group
;
785 * Kill all other threads in the thread group.
788 if (signal_group_exit(sig
)) {
790 * Another group action in progress, just
791 * return so that the signal is processed.
793 spin_unlock_irq(lock
);
796 sig
->group_exit_task
= tsk
;
797 zap_other_threads(tsk
);
799 /* Account for the thread group leader hanging around: */
800 count
= thread_group_leader(tsk
) ? 1 : 2;
801 sig
->notify_count
= count
;
802 while (atomic_read(&sig
->count
) > count
) {
803 __set_current_state(TASK_UNINTERRUPTIBLE
);
804 spin_unlock_irq(lock
);
808 spin_unlock_irq(lock
);
811 * At this point all other threads have exited, all we have to
812 * do is to wait for the thread group leader to become inactive,
813 * and to assume its PID:
815 if (!thread_group_leader(tsk
)) {
816 leader
= tsk
->group_leader
;
818 sig
->notify_count
= -1; /* for exit_notify() */
820 write_lock_irq(&tasklist_lock
);
821 if (likely(leader
->exit_state
))
823 __set_current_state(TASK_UNINTERRUPTIBLE
);
824 write_unlock_irq(&tasklist_lock
);
829 * The only record we have of the real-time age of a
830 * process, regardless of execs it's done, is start_time.
831 * All the past CPU time is accumulated in signal_struct
832 * from sister threads now dead. But in this non-leader
833 * exec, nothing survives from the original leader thread,
834 * whose birth marks the true age of this process now.
835 * When we take on its identity by switching to its PID, we
836 * also take its birthdate (always earlier than our own).
838 tsk
->start_time
= leader
->start_time
;
840 BUG_ON(!same_thread_group(leader
, tsk
));
841 BUG_ON(has_group_leader_pid(tsk
));
843 * An exec() starts a new thread group with the
844 * TGID of the previous thread group. Rehash the
845 * two threads with a switched PID, and release
846 * the former thread group leader:
849 /* Become a process group leader with the old leader's pid.
850 * The old leader becomes a thread of the this thread group.
851 * Note: The old leader also uses this pid until release_task
852 * is called. Odd but simple and correct.
854 detach_pid(tsk
, PIDTYPE_PID
);
855 tsk
->pid
= leader
->pid
;
856 attach_pid(tsk
, PIDTYPE_PID
, task_pid(leader
));
857 transfer_pid(leader
, tsk
, PIDTYPE_PGID
);
858 transfer_pid(leader
, tsk
, PIDTYPE_SID
);
859 list_replace_rcu(&leader
->tasks
, &tsk
->tasks
);
861 tsk
->group_leader
= tsk
;
862 leader
->group_leader
= tsk
;
864 tsk
->exit_signal
= SIGCHLD
;
866 BUG_ON(leader
->exit_state
!= EXIT_ZOMBIE
);
867 leader
->exit_state
= EXIT_DEAD
;
869 write_unlock_irq(&tasklist_lock
);
872 sig
->group_exit_task
= NULL
;
873 sig
->notify_count
= 0;
877 flush_itimer_signals();
879 release_task(leader
);
881 if (atomic_read(&oldsighand
->count
) != 1) {
882 struct sighand_struct
*newsighand
;
884 * This ->sighand is shared with the CLONE_SIGHAND
885 * but not CLONE_THREAD task, switch to the new one.
887 newsighand
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
891 atomic_set(&newsighand
->count
, 1);
892 memcpy(newsighand
->action
, oldsighand
->action
,
893 sizeof(newsighand
->action
));
895 write_lock_irq(&tasklist_lock
);
896 spin_lock(&oldsighand
->siglock
);
897 rcu_assign_pointer(tsk
->sighand
, newsighand
);
898 spin_unlock(&oldsighand
->siglock
);
899 write_unlock_irq(&tasklist_lock
);
901 __cleanup_sighand(oldsighand
);
904 BUG_ON(!thread_group_leader(tsk
));
909 * These functions flushes out all traces of the currently running executable
910 * so that a new one can be started
912 static void flush_old_files(struct files_struct
* files
)
917 spin_lock(&files
->file_lock
);
919 unsigned long set
, i
;
923 fdt
= files_fdtable(files
);
924 if (i
>= fdt
->max_fds
)
926 set
= fdt
->close_on_exec
->fds_bits
[j
];
929 fdt
->close_on_exec
->fds_bits
[j
] = 0;
930 spin_unlock(&files
->file_lock
);
931 for ( ; set
; i
++,set
>>= 1) {
936 spin_lock(&files
->file_lock
);
939 spin_unlock(&files
->file_lock
);
942 char *get_task_comm(char *buf
, struct task_struct
*tsk
)
944 /* buf must be at least sizeof(tsk->comm) in size */
946 strncpy(buf
, tsk
->comm
, sizeof(tsk
->comm
));
951 void set_task_comm(struct task_struct
*tsk
, char *buf
)
954 strlcpy(tsk
->comm
, buf
, sizeof(tsk
->comm
));
958 int flush_old_exec(struct linux_binprm
* bprm
)
962 char tcomm
[sizeof(current
->comm
)];
965 * Make sure we have a private signal table and that
966 * we are unassociated from the previous thread group.
968 retval
= de_thread(current
);
972 set_mm_exe_file(bprm
->mm
, bprm
->file
);
975 * Release all of the old mmap stuff
977 retval
= exec_mmap(bprm
->mm
);
981 bprm
->mm
= NULL
; /* We're using it now */
983 /* This is the point of no return */
984 current
->sas_ss_sp
= current
->sas_ss_size
= 0;
986 if (current
->euid
== current
->uid
&& current
->egid
== current
->gid
)
987 set_dumpable(current
->mm
, 1);
989 set_dumpable(current
->mm
, suid_dumpable
);
991 name
= bprm
->filename
;
993 /* Copies the binary name from after last slash */
994 for (i
=0; (ch
= *(name
++)) != '\0';) {
996 i
= 0; /* overwrite what we wrote */
998 if (i
< (sizeof(tcomm
) - 1))
1002 set_task_comm(current
, tcomm
);
1004 current
->flags
&= ~PF_RANDOMIZE
;
1007 /* Set the new mm task size. We have to do that late because it may
1008 * depend on TIF_32BIT which is only updated in flush_thread() on
1009 * some architectures like powerpc
1011 current
->mm
->task_size
= TASK_SIZE
;
1013 if (bprm
->e_uid
!= current
->euid
|| bprm
->e_gid
!= current
->egid
) {
1015 set_dumpable(current
->mm
, suid_dumpable
);
1016 current
->pdeath_signal
= 0;
1017 } else if (file_permission(bprm
->file
, MAY_READ
) ||
1018 (bprm
->interp_flags
& BINPRM_FLAGS_ENFORCE_NONDUMP
)) {
1020 set_dumpable(current
->mm
, suid_dumpable
);
1023 /* An exec changes our domain. We are no longer part of the thread
1026 current
->self_exec_id
++;
1028 flush_signal_handlers(current
, 0);
1029 flush_old_files(current
->files
);
1037 EXPORT_SYMBOL(flush_old_exec
);
1040 * Fill the binprm structure from the inode.
1041 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1043 int prepare_binprm(struct linux_binprm
*bprm
)
1046 struct inode
* inode
= bprm
->file
->f_path
.dentry
->d_inode
;
1049 mode
= inode
->i_mode
;
1050 if (bprm
->file
->f_op
== NULL
)
1053 bprm
->e_uid
= current
->euid
;
1054 bprm
->e_gid
= current
->egid
;
1056 if(!(bprm
->file
->f_path
.mnt
->mnt_flags
& MNT_NOSUID
)) {
1058 if (mode
& S_ISUID
) {
1059 current
->personality
&= ~PER_CLEAR_ON_SETID
;
1060 bprm
->e_uid
= inode
->i_uid
;
1065 * If setgid is set but no group execute bit then this
1066 * is a candidate for mandatory locking, not a setgid
1069 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
1070 current
->personality
&= ~PER_CLEAR_ON_SETID
;
1071 bprm
->e_gid
= inode
->i_gid
;
1075 /* fill in binprm security blob */
1076 retval
= security_bprm_set(bprm
);
1080 memset(bprm
->buf
,0,BINPRM_BUF_SIZE
);
1081 return kernel_read(bprm
->file
,0,bprm
->buf
,BINPRM_BUF_SIZE
);
1084 EXPORT_SYMBOL(prepare_binprm
);
1086 static int unsafe_exec(struct task_struct
*p
)
1088 int unsafe
= tracehook_unsafe_exec(p
);
1090 if (atomic_read(&p
->fs
->count
) > 1 ||
1091 atomic_read(&p
->files
->count
) > 1 ||
1092 atomic_read(&p
->sighand
->count
) > 1)
1093 unsafe
|= LSM_UNSAFE_SHARE
;
1098 void compute_creds(struct linux_binprm
*bprm
)
1102 if (bprm
->e_uid
!= current
->uid
) {
1104 current
->pdeath_signal
= 0;
1109 unsafe
= unsafe_exec(current
);
1110 security_bprm_apply_creds(bprm
, unsafe
);
1111 task_unlock(current
);
1112 security_bprm_post_apply_creds(bprm
);
1114 EXPORT_SYMBOL(compute_creds
);
1117 * Arguments are '\0' separated strings found at the location bprm->p
1118 * points to; chop off the first by relocating brpm->p to right after
1119 * the first '\0' encountered.
1121 int remove_arg_zero(struct linux_binprm
*bprm
)
1124 unsigned long offset
;
1132 offset
= bprm
->p
& ~PAGE_MASK
;
1133 page
= get_arg_page(bprm
, bprm
->p
, 0);
1138 kaddr
= kmap_atomic(page
, KM_USER0
);
1140 for (; offset
< PAGE_SIZE
&& kaddr
[offset
];
1141 offset
++, bprm
->p
++)
1144 kunmap_atomic(kaddr
, KM_USER0
);
1147 if (offset
== PAGE_SIZE
)
1148 free_arg_page(bprm
, (bprm
->p
>> PAGE_SHIFT
) - 1);
1149 } while (offset
== PAGE_SIZE
);
1158 EXPORT_SYMBOL(remove_arg_zero
);
1161 * cycle the list of binary formats handler, until one recognizes the image
1163 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
1166 struct linux_binfmt
*fmt
;
1168 /* handle /sbin/loader.. */
1170 struct exec
* eh
= (struct exec
*) bprm
->buf
;
1172 if (!bprm
->loader
&& eh
->fh
.f_magic
== 0x183 &&
1173 (eh
->fh
.f_flags
& 0x3000) == 0x3000)
1176 unsigned long loader
;
1178 allow_write_access(bprm
->file
);
1182 loader
= bprm
->vma
->vm_end
- sizeof(void *);
1184 file
= open_exec("/sbin/loader");
1185 retval
= PTR_ERR(file
);
1189 /* Remember if the application is TASO. */
1190 bprm
->taso
= eh
->ah
.entry
< 0x100000000UL
;
1193 bprm
->loader
= loader
;
1194 retval
= prepare_binprm(bprm
);
1197 /* should call search_binary_handler recursively here,
1198 but it does not matter */
1202 retval
= security_bprm_check(bprm
);
1206 /* kernel module loader fixup */
1207 /* so we don't try to load run modprobe in kernel space. */
1210 retval
= audit_bprm(bprm
);
1215 for (try=0; try<2; try++) {
1216 read_lock(&binfmt_lock
);
1217 list_for_each_entry(fmt
, &formats
, lh
) {
1218 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
1221 if (!try_module_get(fmt
->module
))
1223 read_unlock(&binfmt_lock
);
1224 retval
= fn(bprm
, regs
);
1226 tracehook_report_exec(fmt
, bprm
, regs
);
1228 allow_write_access(bprm
->file
);
1232 current
->did_exec
= 1;
1233 proc_exec_connector(current
);
1236 read_lock(&binfmt_lock
);
1238 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
)
1241 read_unlock(&binfmt_lock
);
1245 read_unlock(&binfmt_lock
);
1246 if (retval
!= -ENOEXEC
|| bprm
->mm
== NULL
) {
1250 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1251 if (printable(bprm
->buf
[0]) &&
1252 printable(bprm
->buf
[1]) &&
1253 printable(bprm
->buf
[2]) &&
1254 printable(bprm
->buf
[3]))
1255 break; /* -ENOEXEC */
1256 request_module("binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
1263 EXPORT_SYMBOL(search_binary_handler
);
1265 void free_bprm(struct linux_binprm
*bprm
)
1267 free_arg_pages(bprm
);
1272 * sys_execve() executes a new program.
1274 int do_execve(char * filename
,
1275 char __user
*__user
*argv
,
1276 char __user
*__user
*envp
,
1277 struct pt_regs
* regs
)
1279 struct linux_binprm
*bprm
;
1281 struct files_struct
*displaced
;
1284 retval
= unshare_files(&displaced
);
1289 bprm
= kzalloc(sizeof(*bprm
), GFP_KERNEL
);
1293 file
= open_exec(filename
);
1294 retval
= PTR_ERR(file
);
1301 bprm
->filename
= filename
;
1302 bprm
->interp
= filename
;
1304 retval
= bprm_mm_init(bprm
);
1308 bprm
->argc
= count(argv
, MAX_ARG_STRINGS
);
1309 if ((retval
= bprm
->argc
) < 0)
1312 bprm
->envc
= count(envp
, MAX_ARG_STRINGS
);
1313 if ((retval
= bprm
->envc
) < 0)
1316 retval
= security_bprm_alloc(bprm
);
1320 retval
= prepare_binprm(bprm
);
1324 retval
= copy_strings_kernel(1, &bprm
->filename
, bprm
);
1328 bprm
->exec
= bprm
->p
;
1329 retval
= copy_strings(bprm
->envc
, envp
, bprm
);
1333 retval
= copy_strings(bprm
->argc
, argv
, bprm
);
1337 current
->flags
&= ~PF_KTHREAD
;
1338 retval
= search_binary_handler(bprm
,regs
);
1340 /* execve success */
1341 security_bprm_free(bprm
);
1342 acct_update_integrals(current
);
1345 put_files_struct(displaced
);
1351 security_bprm_free(bprm
);
1359 allow_write_access(bprm
->file
);
1367 reset_files_struct(displaced
);
1372 int set_binfmt(struct linux_binfmt
*new)
1374 struct linux_binfmt
*old
= current
->binfmt
;
1377 if (!try_module_get(new->module
))
1380 current
->binfmt
= new;
1382 module_put(old
->module
);
1386 EXPORT_SYMBOL(set_binfmt
);
1388 /* format_corename will inspect the pattern parameter, and output a
1389 * name into corename, which must have space for at least
1390 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1392 static int format_corename(char *corename
, int nr_threads
, long signr
)
1394 const char *pat_ptr
= core_pattern
;
1395 int ispipe
= (*pat_ptr
== '|');
1396 char *out_ptr
= corename
;
1397 char *const out_end
= corename
+ CORENAME_MAX_SIZE
;
1399 int pid_in_pattern
= 0;
1401 /* Repeat as long as we have more pattern to process and more output
1404 if (*pat_ptr
!= '%') {
1405 if (out_ptr
== out_end
)
1407 *out_ptr
++ = *pat_ptr
++;
1409 switch (*++pat_ptr
) {
1412 /* Double percent, output one percent */
1414 if (out_ptr
== out_end
)
1421 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1422 "%d", task_tgid_vnr(current
));
1423 if (rc
> out_end
- out_ptr
)
1429 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1430 "%d", current
->uid
);
1431 if (rc
> out_end
- out_ptr
)
1437 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1438 "%d", current
->gid
);
1439 if (rc
> out_end
- out_ptr
)
1443 /* signal that caused the coredump */
1445 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1447 if (rc
> out_end
- out_ptr
)
1451 /* UNIX time of coredump */
1454 do_gettimeofday(&tv
);
1455 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1457 if (rc
> out_end
- out_ptr
)
1464 down_read(&uts_sem
);
1465 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1466 "%s", utsname()->nodename
);
1468 if (rc
> out_end
- out_ptr
)
1474 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1475 "%s", current
->comm
);
1476 if (rc
> out_end
- out_ptr
)
1480 /* core limit size */
1482 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1483 "%lu", current
->signal
->rlim
[RLIMIT_CORE
].rlim_cur
);
1484 if (rc
> out_end
- out_ptr
)
1494 /* Backward compatibility with core_uses_pid:
1496 * If core_pattern does not include a %p (as is the default)
1497 * and core_uses_pid is set, then .%pid will be appended to
1498 * the filename. Do not do this for piped commands. */
1499 if (!ispipe
&& !pid_in_pattern
1500 && (core_uses_pid
|| nr_threads
)) {
1501 rc
= snprintf(out_ptr
, out_end
- out_ptr
,
1502 ".%d", task_tgid_vnr(current
));
1503 if (rc
> out_end
- out_ptr
)
1512 static int zap_process(struct task_struct
*start
)
1514 struct task_struct
*t
;
1517 start
->signal
->flags
= SIGNAL_GROUP_EXIT
;
1518 start
->signal
->group_stop_count
= 0;
1522 if (t
!= current
&& t
->mm
) {
1523 sigaddset(&t
->pending
.signal
, SIGKILL
);
1524 signal_wake_up(t
, 1);
1527 } while_each_thread(start
, t
);
1532 static inline int zap_threads(struct task_struct
*tsk
, struct mm_struct
*mm
,
1533 struct core_state
*core_state
, int exit_code
)
1535 struct task_struct
*g
, *p
;
1536 unsigned long flags
;
1539 spin_lock_irq(&tsk
->sighand
->siglock
);
1540 if (!signal_group_exit(tsk
->signal
)) {
1541 mm
->core_state
= core_state
;
1542 tsk
->signal
->group_exit_code
= exit_code
;
1543 nr
= zap_process(tsk
);
1545 spin_unlock_irq(&tsk
->sighand
->siglock
);
1546 if (unlikely(nr
< 0))
1549 if (atomic_read(&mm
->mm_users
) == nr
+ 1)
1552 * We should find and kill all tasks which use this mm, and we should
1553 * count them correctly into ->nr_threads. We don't take tasklist
1554 * lock, but this is safe wrt:
1557 * None of sub-threads can fork after zap_process(leader). All
1558 * processes which were created before this point should be
1559 * visible to zap_threads() because copy_process() adds the new
1560 * process to the tail of init_task.tasks list, and lock/unlock
1561 * of ->siglock provides a memory barrier.
1564 * The caller holds mm->mmap_sem. This means that the task which
1565 * uses this mm can't pass exit_mm(), so it can't exit or clear
1569 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1570 * we must see either old or new leader, this does not matter.
1571 * However, it can change p->sighand, so lock_task_sighand(p)
1572 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1575 * Note also that "g" can be the old leader with ->mm == NULL
1576 * and already unhashed and thus removed from ->thread_group.
1577 * This is OK, __unhash_process()->list_del_rcu() does not
1578 * clear the ->next pointer, we will find the new leader via
1582 for_each_process(g
) {
1583 if (g
== tsk
->group_leader
)
1585 if (g
->flags
& PF_KTHREAD
)
1590 if (unlikely(p
->mm
== mm
)) {
1591 lock_task_sighand(p
, &flags
);
1592 nr
+= zap_process(p
);
1593 unlock_task_sighand(p
, &flags
);
1597 } while_each_thread(g
, p
);
1601 atomic_set(&core_state
->nr_threads
, nr
);
1605 static int coredump_wait(int exit_code
, struct core_state
*core_state
)
1607 struct task_struct
*tsk
= current
;
1608 struct mm_struct
*mm
= tsk
->mm
;
1609 struct completion
*vfork_done
;
1612 init_completion(&core_state
->startup
);
1613 core_state
->dumper
.task
= tsk
;
1614 core_state
->dumper
.next
= NULL
;
1615 core_waiters
= zap_threads(tsk
, mm
, core_state
, exit_code
);
1616 up_write(&mm
->mmap_sem
);
1618 if (unlikely(core_waiters
< 0))
1622 * Make sure nobody is waiting for us to release the VM,
1623 * otherwise we can deadlock when we wait on each other
1625 vfork_done
= tsk
->vfork_done
;
1627 tsk
->vfork_done
= NULL
;
1628 complete(vfork_done
);
1632 wait_for_completion(&core_state
->startup
);
1634 return core_waiters
;
1637 static void coredump_finish(struct mm_struct
*mm
)
1639 struct core_thread
*curr
, *next
;
1640 struct task_struct
*task
;
1642 next
= mm
->core_state
->dumper
.next
;
1643 while ((curr
= next
) != NULL
) {
1647 * see exit_mm(), curr->task must not see
1648 * ->task == NULL before we read ->next.
1652 wake_up_process(task
);
1655 mm
->core_state
= NULL
;
1659 * set_dumpable converts traditional three-value dumpable to two flags and
1660 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1661 * these bits are not changed atomically. So get_dumpable can observe the
1662 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1663 * return either old dumpable or new one by paying attention to the order of
1664 * modifying the bits.
1666 * dumpable | mm->flags (binary)
1667 * old new | initial interim final
1668 * ---------+-----------------------
1676 * (*) get_dumpable regards interim value of 10 as 11.
1678 void set_dumpable(struct mm_struct
*mm
, int value
)
1682 clear_bit(MMF_DUMPABLE
, &mm
->flags
);
1684 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1687 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1689 clear_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1692 set_bit(MMF_DUMP_SECURELY
, &mm
->flags
);
1694 set_bit(MMF_DUMPABLE
, &mm
->flags
);
1699 int get_dumpable(struct mm_struct
*mm
)
1703 ret
= mm
->flags
& 0x3;
1704 return (ret
>= 2) ? 2 : ret
;
1707 int do_coredump(long signr
, int exit_code
, struct pt_regs
* regs
)
1709 struct core_state core_state
;
1710 char corename
[CORENAME_MAX_SIZE
+ 1];
1711 struct mm_struct
*mm
= current
->mm
;
1712 struct linux_binfmt
* binfmt
;
1713 struct inode
* inode
;
1716 int fsuid
= current
->fsuid
;
1719 unsigned long core_limit
= current
->signal
->rlim
[RLIMIT_CORE
].rlim_cur
;
1720 char **helper_argv
= NULL
;
1721 int helper_argc
= 0;
1724 audit_core_dumps(signr
);
1726 binfmt
= current
->binfmt
;
1727 if (!binfmt
|| !binfmt
->core_dump
)
1729 down_write(&mm
->mmap_sem
);
1731 * If another thread got here first, or we are not dumpable, bail out.
1733 if (mm
->core_state
|| !get_dumpable(mm
)) {
1734 up_write(&mm
->mmap_sem
);
1739 * We cannot trust fsuid as being the "true" uid of the
1740 * process nor do we know its entire history. We only know it
1741 * was tainted so we dump it as root in mode 2.
1743 if (get_dumpable(mm
) == 2) { /* Setuid core dump mode */
1744 flag
= O_EXCL
; /* Stop rewrite attacks */
1745 current
->fsuid
= 0; /* Dump root private */
1748 retval
= coredump_wait(exit_code
, &core_state
);
1753 * Clear any false indication of pending signals that might
1754 * be seen by the filesystem code called to write the core file.
1756 clear_thread_flag(TIF_SIGPENDING
);
1759 * lock_kernel() because format_corename() is controlled by sysctl, which
1760 * uses lock_kernel()
1763 ispipe
= format_corename(corename
, retval
, signr
);
1766 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1767 * to a pipe. Since we're not writing directly to the filesystem
1768 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1769 * created unless the pipe reader choses to write out the core file
1770 * at which point file size limits and permissions will be imposed
1771 * as it does with any other process
1773 if ((!ispipe
) && (core_limit
< binfmt
->min_coredump
))
1777 helper_argv
= argv_split(GFP_KERNEL
, corename
+1, &helper_argc
);
1778 /* Terminate the string before the first option */
1779 delimit
= strchr(corename
, ' ');
1782 delimit
= strrchr(helper_argv
[0], '/');
1786 delimit
= helper_argv
[0];
1787 if (!strcmp(delimit
, current
->comm
)) {
1788 printk(KERN_NOTICE
"Recursive core dump detected, "
1793 core_limit
= RLIM_INFINITY
;
1795 /* SIGPIPE can happen, but it's just never processed */
1796 if (call_usermodehelper_pipe(corename
+1, helper_argv
, NULL
,
1798 printk(KERN_INFO
"Core dump to %s pipe failed\n",
1803 file
= filp_open(corename
,
1804 O_CREAT
| 2 | O_NOFOLLOW
| O_LARGEFILE
| flag
,
1808 inode
= file
->f_path
.dentry
->d_inode
;
1809 if (inode
->i_nlink
> 1)
1810 goto close_fail
; /* multiple links - don't dump */
1811 if (!ispipe
&& d_unhashed(file
->f_path
.dentry
))
1814 /* AK: actually i see no reason to not allow this for named pipes etc.,
1815 but keep the previous behaviour for now. */
1816 if (!ispipe
&& !S_ISREG(inode
->i_mode
))
1819 * Dont allow local users get cute and trick others to coredump
1820 * into their pre-created files:
1822 if (inode
->i_uid
!= current
->fsuid
)
1826 if (!file
->f_op
->write
)
1828 if (!ispipe
&& do_truncate(file
->f_path
.dentry
, 0, 0, file
) != 0)
1831 retval
= binfmt
->core_dump(signr
, regs
, file
, core_limit
);
1834 current
->signal
->group_exit_code
|= 0x80;
1836 filp_close(file
, NULL
);
1839 argv_free(helper_argv
);
1841 current
->fsuid
= fsuid
;
1842 coredump_finish(mm
);