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/config.h>
26 #include <linux/slab.h>
27 #include <linux/file.h>
28 #include <linux/mman.h>
29 #include <linux/a.out.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/smp_lock.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/highmem.h>
37 #include <asm/uaccess.h>
38 #include <asm/pgtable.h>
39 #include <asm/mmu_context.h>
42 #include <linux/kmod.h>
46 * Here are the actual binaries that will be accepted:
47 * add more with "register_binfmt()" if using modules...
49 * These are defined again for the 'real' modules if you are using a
50 * module definition for these routines.
53 static struct linux_binfmt
*formats
= (struct linux_binfmt
*) NULL
;
55 int register_binfmt(struct linux_binfmt
* fmt
)
57 struct linux_binfmt
** tmp
= &formats
;
73 int unregister_binfmt(struct linux_binfmt
* fmt
)
75 struct linux_binfmt
** tmp
= &formats
;
87 /* N.B. Error returns must be < 0 */
88 int open_dentry(struct dentry
* dentry
, int mode
)
90 struct inode
* inode
= dentry
->d_inode
;
92 struct list_head
* l
= NULL
;
96 l
= &inode
->i_sb
->s_files
;
99 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
101 fd
= get_unused_fd();
104 f
= get_empty_filp();
108 f
->f_mode
= (mode
+1) & O_ACCMODE
;
109 f
->f_dentry
= dentry
;
112 f
->f_op
= inode
->i_op
->default_file_ops
;
114 error
= f
->f_op
->open(inode
,f
);
135 * Note that a shared library must be both readable and executable due to
138 * Also note that we take the address to load from from the file itself.
140 asmlinkage
long sys_uselib(const char * library
)
144 struct linux_binfmt
* fmt
;
147 fd
= sys_open(library
, 0, 0);
153 if (file
&& file
->f_dentry
&& file
->f_op
&& file
->f_op
->read
) {
154 for (fmt
= formats
; fmt
; fmt
= fmt
->next
) {
155 int (*fn
)(int) = fmt
->load_shlib
;
158 /* N.B. Should use file instead of fd */
160 if (retval
!= -ENOEXEC
)
172 * count() counts the number of arguments/envelopes
174 static int count(char ** argv
, int max
)
183 error
= get_user(p
,argv
);
197 * 'copy_strings()' copies argument/envelope strings from user
198 * memory to free pages in kernel mem. These are in a format ready
199 * to be put directly into the top of new user memory.
201 int copy_strings(int argc
,char ** argv
, struct linux_binprm
*bprm
)
208 if (get_user(str
, argv
+argc
) || !str
|| !(len
= strnlen_user(str
, bprm
->p
)))
214 /* XXX: add architecture specific overflow check here. */
221 int offset
, bytes_to_copy
;
223 offset
= pos
% PAGE_SIZE
;
225 page
= bprm
->page
[i
];
229 * Cannot yet use highmem page because
230 * we cannot sleep with a kmap held.
232 page
= __get_pages(GFP_USER
, 0);
233 bprm
->page
[i
] = page
;
238 kaddr
= (char *)kmap(page
, KM_WRITE
);
241 memset(kaddr
, 0, offset
);
242 bytes_to_copy
= PAGE_SIZE
- offset
;
243 if (bytes_to_copy
> len
) {
246 memset(kaddr
+offset
+len
, 0, PAGE_SIZE
-offset
-len
);
248 err
= copy_from_user(kaddr
+ offset
, str
, bytes_to_copy
);
249 flush_page_to_ram(page
);
250 kunmap((unsigned long)kaddr
, KM_WRITE
);
255 pos
+= bytes_to_copy
;
256 str
+= bytes_to_copy
;
257 len
-= bytes_to_copy
;
264 * Like copy_strings, but get argv and its values from kernel memory.
266 int copy_strings_kernel(int argc
,char ** argv
, struct linux_binprm
*bprm
)
269 mm_segment_t oldfs
= get_fs();
271 r
= copy_strings(argc
, argv
, bprm
);
276 int setup_arg_pages(struct linux_binprm
*bprm
)
278 unsigned long stack_base
;
279 struct vm_area_struct
*mpnt
;
282 stack_base
= STACK_TOP
- MAX_ARG_PAGES
*PAGE_SIZE
;
284 bprm
->p
+= stack_base
;
286 bprm
->loader
+= stack_base
;
287 bprm
->exec
+= stack_base
;
289 mpnt
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
294 mpnt
->vm_mm
= current
->mm
;
295 mpnt
->vm_start
= PAGE_MASK
& (unsigned long) bprm
->p
;
296 mpnt
->vm_end
= STACK_TOP
;
297 mpnt
->vm_page_prot
= PAGE_COPY
;
298 mpnt
->vm_flags
= VM_STACK_FLAGS
;
301 mpnt
->vm_file
= NULL
;
302 mpnt
->vm_private_data
= (void *) 0;
303 vmlist_modify_lock(current
->mm
);
304 insert_vm_struct(current
->mm
, mpnt
);
305 vmlist_modify_unlock(current
->mm
);
306 current
->mm
->total_vm
= (mpnt
->vm_end
- mpnt
->vm_start
) >> PAGE_SHIFT
;
309 for (i
= 0 ; i
< MAX_ARG_PAGES
; i
++) {
312 put_dirty_page(current
,bprm
->page
[i
],stack_base
);
314 stack_base
+= PAGE_SIZE
;
321 * Read in the complete executable. This is used for "-N" files
322 * that aren't on a block boundary, and for files on filesystems
323 * without get_block support.
325 int read_exec(struct dentry
*dentry
, unsigned long offset
,
326 char * addr
, unsigned long count
, int to_kmem
)
329 struct inode
* inode
= dentry
->d_inode
;
330 int result
= -ENOEXEC
;
332 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
334 if (init_private_file(&file
, dentry
, 1))
336 if (!file
.f_op
->read
)
338 if (file
.f_op
->llseek
) {
339 if (file
.f_op
->llseek(&file
,offset
,0) != offset
)
344 mm_segment_t old_fs
= get_fs();
346 result
= file
.f_op
->read(&file
, addr
, count
, &file
.f_pos
);
349 result
= verify_area(VERIFY_WRITE
, addr
, count
);
352 result
= file
.f_op
->read(&file
, addr
, count
, &file
.f_pos
);
355 if (file
.f_op
->release
)
356 file
.f_op
->release(inode
,&file
);
361 static int exec_mmap(void)
363 struct mm_struct
* mm
, * old_mm
;
365 old_mm
= current
->mm
;
366 if (old_mm
&& atomic_read(&old_mm
->mm_users
) == 1) {
367 flush_cache_mm(old_mm
);
370 flush_tlb_mm(old_mm
);
376 struct mm_struct
*active_mm
= current
->active_mm
;
379 current
->active_mm
= mm
;
380 activate_mm(active_mm
, mm
);
383 if (active_mm
!= old_mm
) BUG();
394 * This function makes sure the current process has its own signal table,
395 * so that flush_signal_handlers can later reset the handlers without
396 * disturbing other processes. (Other processes might share the signal
397 * table via the CLONE_SIGHAND option to clone().)
400 static inline int make_private_signals(void)
402 struct signal_struct
* newsig
;
404 if (atomic_read(¤t
->sig
->count
) <= 1)
406 newsig
= kmalloc(sizeof(*newsig
), GFP_KERNEL
);
409 spin_lock_init(&newsig
->siglock
);
410 atomic_set(&newsig
->count
, 1);
411 memcpy(newsig
->action
, current
->sig
->action
, sizeof(newsig
->action
));
412 current
->sig
= newsig
;
417 * If make_private_signals() made a copy of the signal table, decrement the
418 * refcount of the original table, and free it if necessary.
419 * We don't do that in make_private_signals() so that we can back off
420 * in flush_old_exec() if an error occurs after calling make_private_signals().
423 static inline void release_old_signals(struct signal_struct
* oldsig
)
425 if (current
->sig
== oldsig
)
427 if (atomic_dec_and_test(&oldsig
->count
))
432 * These functions flushes out all traces of the currently running executable
433 * so that a new one can be started
436 static inline void flush_old_files(struct files_struct
* files
)
442 unsigned long set
, i
;
445 if (i
>= files
->max_fds
|| i
>= files
->max_fdset
)
447 set
= xchg(&files
->close_on_exec
->fds_bits
[j
], 0);
449 for ( ; set
; i
++,set
>>= 1) {
456 int flush_old_exec(struct linux_binprm
* bprm
)
460 struct signal_struct
* oldsig
;
463 * Make sure we have a private signal table
465 oldsig
= current
->sig
;
466 retval
= make_private_signals();
467 if (retval
) goto flush_failed
;
470 * Release all of the old mmap stuff
472 retval
= exec_mmap();
473 if (retval
) goto mmap_failed
;
475 /* This is the point of no return */
476 release_old_signals(oldsig
);
478 if (current
->euid
== current
->uid
&& current
->egid
== current
->gid
)
479 current
->dumpable
= 1;
480 name
= bprm
->filename
;
481 for (i
=0; (ch
= *(name
++)) != '\0';) {
486 current
->comm
[i
++] = ch
;
488 current
->comm
[i
] = '\0';
492 if (bprm
->e_uid
!= current
->euid
|| bprm
->e_gid
!= current
->egid
||
493 permission(bprm
->dentry
->d_inode
,MAY_READ
))
494 current
->dumpable
= 0;
496 /* An exec changes our domain. We are no longer part of the thread
499 current
->self_exec_id
++;
501 flush_signal_handlers(current
);
502 flush_old_files(current
->files
);
507 if (current
->sig
!= oldsig
)
510 current
->sig
= oldsig
;
515 * We mustn't allow tracing of suid binaries, unless
516 * the tracer has the capability to trace anything..
518 static inline int must_not_trace_exec(struct task_struct
* p
)
520 return (p
->flags
& PF_PTRACED
) && !cap_raised(p
->p_pptr
->cap_effective
, CAP_SYS_PTRACE
);
524 * Fill the binprm structure from the inode.
525 * Check permissions, then read the first 512 bytes
527 int prepare_binprm(struct linux_binprm
*bprm
)
530 int retval
,id_change
,cap_raised
;
531 struct inode
* inode
= bprm
->dentry
->d_inode
;
533 mode
= inode
->i_mode
;
534 if (!S_ISREG(mode
)) /* must be regular file */
536 if (!(mode
& 0111)) /* with at least _one_ execute bit set */
538 if (IS_NOEXEC(inode
)) /* FS mustn't be mounted noexec */
542 if ((retval
= permission(inode
, MAY_EXEC
)) != 0)
544 /* better not execute files which are being written to */
545 if (atomic_read(&inode
->i_writecount
) > 0)
548 bprm
->e_uid
= current
->euid
;
549 bprm
->e_gid
= current
->egid
;
550 id_change
= cap_raised
= 0;
553 if (mode
& S_ISUID
) {
554 bprm
->e_uid
= inode
->i_uid
;
555 if (bprm
->e_uid
!= current
->euid
)
561 * If setgid is set but no group execute bit then this
562 * is a candidate for mandatory locking, not a setgid
565 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
566 bprm
->e_gid
= inode
->i_gid
;
567 if (!in_group_p(bprm
->e_gid
))
571 /* We don't have VFS support for capabilities yet */
572 cap_clear(bprm
->cap_inheritable
);
573 cap_clear(bprm
->cap_permitted
);
574 cap_clear(bprm
->cap_effective
);
576 /* To support inheritance of root-permissions and suid-root
577 * executables under compatibility mode, we raise the
578 * effective and inherited bitmasks of the executable file
579 * (translation: we set the executable "capability dumb" and
580 * set the allowed set to maximum). We don't set any forced
583 * If only the real uid is 0, we only raise the inheritable
584 * bitmask of the executable file (translation: we set the
585 * allowed set to maximum and the application to "capability
589 if (!issecure(SECURE_NOROOT
)) {
590 if (bprm
->e_uid
== 0 || current
->uid
== 0)
591 cap_set_full(bprm
->cap_inheritable
);
592 if (bprm
->e_uid
== 0)
593 cap_set_full(bprm
->cap_effective
);
596 /* Only if pP' is _not_ a subset of pP, do we consider there
597 * has been a capability related "change of capability". In
598 * such cases, we need to check that the elevation of
599 * privilege does not go against other system constraints.
600 * The new Permitted set is defined below -- see (***). */
602 kernel_cap_t working
=
603 cap_combine(bprm
->cap_permitted
,
604 cap_intersect(bprm
->cap_inheritable
,
605 current
->cap_inheritable
));
606 if (!cap_issubset(working
, current
->cap_permitted
)) {
611 if (id_change
|| cap_raised
) {
612 /* We can't suid-execute if we're sharing parts of the executable */
613 /* or if we're being traced (or if suid execs are not allowed) */
614 /* (current->mm->mm_users > 1 is ok, as we'll get a new mm anyway) */
616 || must_not_trace_exec(current
)
617 || (atomic_read(¤t
->fs
->count
) > 1)
618 || (atomic_read(¤t
->sig
->count
) > 1)
619 || (atomic_read(¤t
->files
->count
) > 1)) {
620 if (id_change
&& !capable(CAP_SETUID
))
622 if (cap_raised
&& !capable(CAP_SETPCAP
))
627 memset(bprm
->buf
,0,sizeof(bprm
->buf
));
628 return read_exec(bprm
->dentry
,0,bprm
->buf
,128,1);
632 * This function is used to produce the new IDs and capabilities
633 * from the old ones and the file's capabilities.
635 * The formula used for evolving capabilities is:
638 * (***) pP' = fP | (fI & pI)
639 * pE' = pP' & fE [NB. fE is 0 or ~0]
641 * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
642 * ' indicates post-exec().
645 void compute_creds(struct linux_binprm
*bprm
)
647 int new_permitted
= cap_t(bprm
->cap_permitted
) |
648 (cap_t(bprm
->cap_inheritable
) &
649 cap_t(current
->cap_inheritable
));
651 /* For init, we want to retain the capabilities set
652 * in the init_task struct. Thus we skip the usual
653 * capability rules */
654 if (current
->pid
!= 1) {
655 cap_t(current
->cap_permitted
) = new_permitted
;
656 cap_t(current
->cap_effective
) = new_permitted
&
657 cap_t(bprm
->cap_effective
);
660 /* AUD: Audit candidate if current->cap_effective is set */
662 current
->suid
= current
->euid
= current
->fsuid
= bprm
->e_uid
;
663 current
->sgid
= current
->egid
= current
->fsgid
= bprm
->e_gid
;
664 if (current
->euid
!= current
->uid
|| current
->egid
!= current
->gid
||
665 !cap_issubset(new_permitted
, current
->cap_permitted
))
666 current
->dumpable
= 0;
670 void remove_arg_zero(struct linux_binprm
*bprm
)
673 unsigned long offset
;
677 offset
= bprm
->p
% PAGE_SIZE
;
680 while (bprm
->p
++, *(kaddr
+offset
++)) {
681 if (offset
!= PAGE_SIZE
)
684 kunmap((unsigned long)kaddr
, KM_WRITE
);
686 page
= bprm
->page
[bprm
->p
/PAGE_SIZE
];
687 kaddr
= (char *)kmap(page
, KM_WRITE
);
689 kunmap((unsigned long)kaddr
, KM_WRITE
);
695 * cycle the list of binary formats handler, until one recognizes the image
697 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
700 struct linux_binfmt
*fmt
;
702 /* handle /sbin/loader.. */
704 struct exec
* eh
= (struct exec
*) bprm
->buf
;
705 struct linux_binprm bprm_loader
;
707 if (!bprm
->loader
&& eh
->fh
.f_magic
== 0x183 &&
708 (eh
->fh
.f_flags
& 0x3000) == 0x3000)
711 char * dynloader
[] = { "/sbin/loader" };
712 struct dentry
* dentry
;
717 bprm_loader
.p
= PAGE_SIZE
*MAX_ARG_PAGES
-sizeof(void *);
718 for (i
= 0 ; i
< MAX_ARG_PAGES
; i
++) /* clear page-table */
719 bprm_loader
.page
[i
] = NULL
;
721 dentry
= open_namei(dynloader
[0], 0, 0);
722 retval
= PTR_ERR(dentry
);
725 bprm
->dentry
= dentry
;
726 bprm
->loader
= bprm_loader
.p
;
727 retval
= prepare_binprm(bprm
);
730 /* should call search_binary_handler recursively here,
731 but it does not matter */
735 for (try=0; try<2; try++) {
736 for (fmt
= formats
; fmt
; fmt
= fmt
->next
) {
737 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
740 retval
= fn(bprm
, regs
);
745 current
->did_exec
= 1;
748 if (retval
!= -ENOEXEC
)
750 if (!bprm
->dentry
) /* We don't have the dentry anymore */
753 if (retval
!= -ENOEXEC
) {
757 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
759 if (printable(bprm
->buf
[0]) &&
760 printable(bprm
->buf
[1]) &&
761 printable(bprm
->buf
[2]) &&
762 printable(bprm
->buf
[3]))
763 break; /* -ENOEXEC */
764 sprintf(modname
, "binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
765 request_module(modname
);
774 * sys_execve() executes a new program.
776 int do_execve(char * filename
, char ** argv
, char ** envp
, struct pt_regs
* regs
)
778 struct linux_binprm bprm
;
779 struct dentry
* dentry
;
783 bprm
.p
= PAGE_SIZE
*MAX_ARG_PAGES
-sizeof(void *);
784 memset(bprm
.page
, 0, MAX_ARG_PAGES
*sizeof(bprm
.page
[0]));
786 dentry
= open_namei(filename
, 0, 0);
787 retval
= PTR_ERR(dentry
);
791 bprm
.dentry
= dentry
;
792 bprm
.filename
= filename
;
796 if ((bprm
.argc
= count(argv
, bprm
.p
/ sizeof(void *))) < 0) {
801 if ((bprm
.envc
= count(envp
, bprm
.p
/ sizeof(void *))) < 0) {
806 retval
= prepare_binprm(&bprm
);
810 retval
= copy_strings_kernel(1, &bprm
.filename
, &bprm
);
815 retval
= copy_strings(bprm
.envc
, envp
, &bprm
);
819 retval
= copy_strings(bprm
.argc
, argv
, &bprm
);
823 retval
= search_binary_handler(&bprm
,regs
);
829 /* Something went wrong, return the inode and free the argument pages*/
833 /* Assumes that free_page() can take a NULL argument. */
834 /* I hope this is ok for all architectures */
835 for (i
= 0 ; i
< MAX_ARG_PAGES
; i
++)
837 __free_page(bprm
.page
[i
]);
842 int do_coredump(long signr
, struct pt_regs
* regs
)
844 struct linux_binfmt
* binfmt
;
845 char corename
[6+sizeof(current
->comm
)];
847 struct dentry
* dentry
;
848 struct inode
* inode
;
851 binfmt
= current
->binfmt
;
852 if (!binfmt
|| !binfmt
->core_dump
)
854 if (!current
->dumpable
|| atomic_read(¤t
->mm
->mm_users
) != 1)
856 current
->dumpable
= 0;
857 if (current
->rlim
[RLIMIT_CORE
].rlim_cur
< binfmt
->min_coredump
)
860 memcpy(corename
,"core.", 5);
862 memcpy(corename
+5,current
->comm
,sizeof(current
->comm
));
866 file
= filp_open(corename
, O_CREAT
| 2 | O_TRUNC
| O_NOFOLLOW
, 0600);
869 dentry
= file
->f_dentry
;
870 inode
= dentry
->d_inode
;
871 if (inode
->i_nlink
> 1)
872 goto close_fail
; /* multiple links - don't dump */
874 if (!S_ISREG(inode
->i_mode
))
876 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
878 if (!file
->f_op
->write
)
880 if (!binfmt
->core_dump(signr
, regs
, file
))
882 filp_close(file
, NULL
);
887 filp_close(file
, NULL
);