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>
35 #include <asm/uaccess.h>
36 #include <asm/pgtable.h>
37 #include <asm/mmu_context.h>
40 #include <linux/kmod.h>
44 * Here are the actual binaries that will be accepted:
45 * add more with "register_binfmt()" if using modules...
47 * These are defined again for the 'real' modules if you are using a
48 * module definition for these routines.
51 static struct linux_binfmt
*formats
= (struct linux_binfmt
*) NULL
;
53 void __init
binfmt_setup(void)
55 #ifdef CONFIG_BINFMT_MISC
59 #ifdef CONFIG_BINFMT_ELF
63 #ifdef CONFIG_BINFMT_ELF32
67 #ifdef CONFIG_BINFMT_AOUT
71 #ifdef CONFIG_BINFMT_AOUT32
75 #ifdef CONFIG_BINFMT_JAVA
79 #ifdef CONFIG_BINFMT_EM86
83 /* This cannot be configured out of the kernel */
87 int register_binfmt(struct linux_binfmt
* fmt
)
89 struct linux_binfmt
** tmp
= &formats
;
105 #ifdef CONFIG_MODULES
106 int unregister_binfmt(struct linux_binfmt
* fmt
)
108 struct linux_binfmt
** tmp
= &formats
;
119 #endif /* CONFIG_MODULES */
121 /* N.B. Error returns must be < 0 */
122 int open_dentry(struct dentry
* dentry
, int mode
)
124 struct inode
* inode
= dentry
->d_inode
;
129 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
131 fd
= get_unused_fd();
134 f
= get_empty_filp();
138 f
->f_mode
= (mode
+1) & O_ACCMODE
;
139 f
->f_dentry
= dentry
;
142 f
->f_op
= inode
->i_op
->default_file_ops
;
144 error
= f
->f_op
->open(inode
,f
);
164 * Note that a shared library must be both readable and executable due to
167 * Also note that we take the address to load from from the file itself.
169 asmlinkage
int sys_uselib(const char * library
)
173 struct linux_binfmt
* fmt
;
176 fd
= sys_open(library
, 0, 0);
182 if (file
&& file
->f_dentry
&& file
->f_op
&& file
->f_op
->read
) {
183 for (fmt
= formats
; fmt
; fmt
= fmt
->next
) {
184 int (*fn
)(int) = fmt
->load_shlib
;
187 /* N.B. Should use file instead of fd */
189 if (retval
!= -ENOEXEC
)
201 * count() counts the number of arguments/envelopes
203 static int count(char ** argv
)
212 error
= get_user(p
,argv
);
225 * 'copy_string()' copies argument/envelope strings from user
226 * memory to free pages in kernel mem. These are in a format ready
227 * to be put directly into the top of new user memory.
229 * Modified by TYT, 11/24/91 to add the from_kmem argument, which specifies
230 * whether the string and the string array are from user or kernel segments:
232 * from_kmem argv * argv **
233 * 0 user space user space
234 * 1 kernel space user space
235 * 2 kernel space kernel space
237 * We do this by playing games with the fs segment register. Since it
238 * is expensive to load a segment register, we try to avoid calling
239 * set_fs() unless we absolutely have to.
241 unsigned long copy_strings(int argc
,char ** argv
,unsigned long *page
,
242 unsigned long p
, int from_kmem
)
248 return 0; /* bullet-proofing */
258 get_user(str
, argv
+argc
);
260 panic("VFS: argc is wrong");
263 len
= strlen_user(str
); /* includes the '\0' */
264 if (p
< len
) { /* this shouldn't happen - 128kB */
272 int offset
, bytes_to_copy
;
274 offset
= pos
% PAGE_SIZE
;
275 if (!(pag
= (char *) page
[pos
/PAGE_SIZE
]) &&
276 !(pag
= (char *) page
[pos
/PAGE_SIZE
] =
277 (unsigned long *) get_free_page(GFP_USER
))) {
282 bytes_to_copy
= PAGE_SIZE
- offset
;
283 if (bytes_to_copy
> len
)
285 copy_from_user(pag
+ offset
, str
, bytes_to_copy
);
286 pos
+= bytes_to_copy
;
287 str
+= bytes_to_copy
;
288 len
-= bytes_to_copy
;
296 unsigned long setup_arg_pages(unsigned long p
, struct linux_binprm
* bprm
)
298 unsigned long stack_base
;
299 struct vm_area_struct
*mpnt
;
302 stack_base
= STACK_TOP
- MAX_ARG_PAGES
*PAGE_SIZE
;
306 bprm
->loader
+= stack_base
;
307 bprm
->exec
+= stack_base
;
309 mpnt
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
311 mpnt
->vm_mm
= current
->mm
;
312 mpnt
->vm_start
= PAGE_MASK
& (unsigned long) p
;
313 mpnt
->vm_end
= STACK_TOP
;
314 mpnt
->vm_page_prot
= PAGE_COPY
;
315 mpnt
->vm_flags
= VM_STACK_FLAGS
;
318 mpnt
->vm_file
= NULL
;
320 insert_vm_struct(current
->mm
, mpnt
);
321 current
->mm
->total_vm
= (mpnt
->vm_end
- mpnt
->vm_start
) >> PAGE_SHIFT
;
324 for (i
= 0 ; i
< MAX_ARG_PAGES
; i
++) {
327 put_dirty_page(current
,bprm
->page
[i
],stack_base
);
329 stack_base
+= PAGE_SIZE
;
335 * Read in the complete executable. This is used for "-N" files
336 * that aren't on a block boundary, and for files on filesystems
337 * without bmap support.
339 int read_exec(struct dentry
*dentry
, unsigned long offset
,
340 char * addr
, unsigned long count
, int to_kmem
)
343 struct inode
* inode
= dentry
->d_inode
;
344 int result
= -ENOEXEC
;
346 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
348 if (init_private_file(&file
, dentry
, 1))
350 if (!file
.f_op
->read
)
352 if (file
.f_op
->llseek
) {
353 if (file
.f_op
->llseek(&file
,offset
,0) != offset
)
358 mm_segment_t old_fs
= get_fs();
360 result
= file
.f_op
->read(&file
, addr
, count
, &file
.f_pos
);
363 result
= verify_area(VERIFY_WRITE
, addr
, count
);
366 result
= file
.f_op
->read(&file
, addr
, count
, &file
.f_pos
);
369 if (file
.f_op
->release
)
370 file
.f_op
->release(inode
,&file
);
375 static int exec_mmap(void)
377 struct mm_struct
* mm
, * old_mm
;
380 if (atomic_read(¤t
->mm
->count
) == 1) {
381 flush_cache_mm(current
->mm
);
383 release_segments(current
->mm
);
384 exit_mmap(current
->mm
);
385 flush_tlb_mm(current
->mm
);
394 mm
->cpu_vm_mask
= (1UL << smp_processor_id());
398 * Make sure we have a private ldt if needed ...
400 nr
= current
->tarray_ptr
- &task
[0];
401 copy_segments(nr
, current
, mm
);
403 old_mm
= current
->mm
;
405 retval
= new_page_tables(current
);
408 activate_context(current
);
415 * Failure ... restore the prior mm_struct.
418 /* The pgd belongs to the parent ... don't free it! */
420 current
->mm
= old_mm
;
421 /* restore the ldt for this task */
422 copy_segments(nr
, current
, NULL
);
430 * This function makes sure the current process has its own signal table,
431 * so that flush_signal_handlers can later reset the handlers without
432 * disturbing other processes. (Other processes might share the signal
433 * table via the CLONE_SIGHAND option to clone().)
436 static inline int make_private_signals(void)
438 struct signal_struct
* newsig
;
440 if (atomic_read(¤t
->sig
->count
) <= 1)
442 newsig
= kmalloc(sizeof(*newsig
), GFP_KERNEL
);
445 spin_lock_init(&newsig
->siglock
);
446 atomic_set(&newsig
->count
, 1);
447 memcpy(newsig
->action
, current
->sig
->action
, sizeof(newsig
->action
));
448 current
->sig
= newsig
;
453 * If make_private_signals() made a copy of the signal table, decrement the
454 * refcount of the original table, and free it if necessary.
455 * We don't do that in make_private_signals() so that we can back off
456 * in flush_old_exec() if an error occurs after calling make_private_signals().
459 static inline void release_old_signals(struct signal_struct
* oldsig
)
461 if (current
->sig
== oldsig
)
463 if (atomic_dec_and_test(&oldsig
->count
))
468 * These functions flushes out all traces of the currently running executable
469 * so that a new one can be started
472 static inline void flush_old_files(struct files_struct
* files
)
478 unsigned long set
, i
;
481 if (i
>= files
->max_fds
)
483 set
= files
->close_on_exec
.fds_bits
[j
];
484 files
->close_on_exec
.fds_bits
[j
] = 0;
486 for ( ; set
; i
++,set
>>= 1) {
493 int flush_old_exec(struct linux_binprm
* bprm
)
497 struct signal_struct
* oldsig
;
500 * Make sure we have a private signal table
502 oldsig
= current
->sig
;
503 retval
= make_private_signals();
504 if (retval
) goto flush_failed
;
507 * Release all of the old mmap stuff
509 retval
= exec_mmap();
510 if (retval
) goto mmap_failed
;
512 /* This is the point of no return */
513 release_old_signals(oldsig
);
515 if (current
->euid
== current
->uid
&& current
->egid
== current
->gid
)
516 current
->dumpable
= 1;
517 name
= bprm
->filename
;
518 for (i
=0; (ch
= *(name
++)) != '\0';) {
523 current
->comm
[i
++] = ch
;
525 current
->comm
[i
] = '\0';
529 if (bprm
->e_uid
!= current
->euid
|| bprm
->e_gid
!= current
->egid
||
530 permission(bprm
->dentry
->d_inode
,MAY_READ
))
531 current
->dumpable
= 0;
533 flush_signal_handlers(current
);
534 flush_old_files(current
->files
);
539 if (current
->sig
!= oldsig
)
542 current
->sig
= oldsig
;
547 * We mustn't allow tracing of suid binaries, unless
548 * the tracer has the capability to trace anything..
550 static inline int must_not_trace_exec(struct task_struct
* p
)
552 return (p
->flags
& PF_PTRACED
) && !cap_raised(p
->p_pptr
->cap_effective
, CAP_SYS_PTRACE
);
556 * Fill the binprm structure from the inode.
557 * Check permissions, then read the first 512 bytes
559 int prepare_binprm(struct linux_binprm
*bprm
)
562 int retval
,id_change
,cap_raised
;
563 struct inode
* inode
= bprm
->dentry
->d_inode
;
565 mode
= inode
->i_mode
;
566 if (!S_ISREG(mode
)) /* must be regular file */
568 if (!(mode
& 0111)) /* with at least _one_ execute bit set */
570 if (IS_NOEXEC(inode
)) /* FS mustn't be mounted noexec */
574 if ((retval
= permission(inode
, MAY_EXEC
)) != 0)
576 /* better not execute files which are being written to */
577 if (inode
->i_writecount
> 0)
580 bprm
->e_uid
= current
->euid
;
581 bprm
->e_gid
= current
->egid
;
582 id_change
= cap_raised
= 0;
585 if (mode
& S_ISUID
) {
586 bprm
->e_uid
= inode
->i_uid
;
587 if (bprm
->e_uid
!= current
->euid
)
593 * If setgid is set but no group execute bit then this
594 * is a candidate for mandatory locking, not a setgid
597 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
598 bprm
->e_gid
= inode
->i_gid
;
599 if (!in_group_p(bprm
->e_gid
))
603 /* We don't have VFS support for capabilities yet */
604 cap_clear(bprm
->cap_inheritable
);
605 cap_clear(bprm
->cap_permitted
);
606 cap_clear(bprm
->cap_effective
);
608 /* To support inheritance of root-permissions and suid-root
609 * executables under compatibility mode, we raise the
610 * effective and inherited bitmasks of the executable file
611 * (translation: we set the executable "capability dumb" and
612 * set the allowed set to maximum). We don't set any forced
615 * If only the real uid is 0, we only raise the inheritable
616 * bitmask of the executable file (translation: we set the
617 * allowed set to maximum and the application to "capability
621 if (!issecure(SECURE_NOROOT
)) {
622 if (bprm
->e_uid
== 0 || current
->uid
== 0)
623 cap_set_full(bprm
->cap_inheritable
);
624 if (bprm
->e_uid
== 0)
625 cap_set_full(bprm
->cap_effective
);
628 /* Only if pP' is _not_ a subset of pP, do we consider there
629 * has been a capability related "change of capability". In
630 * such cases, we need to check that the elevation of
631 * privilege does not go against other system constraints.
632 * The new Permitted set is defined below -- see (***). */
634 kernel_cap_t working
=
635 cap_combine(bprm
->cap_permitted
,
636 cap_intersect(bprm
->cap_inheritable
,
637 current
->cap_inheritable
));
638 if (!cap_issubset(working
, current
->cap_permitted
)) {
643 if (id_change
|| cap_raised
) {
644 /* We can't suid-execute if we're sharing parts of the executable */
645 /* or if we're being traced (or if suid execs are not allowed) */
646 /* (current->mm->count > 1 is ok, as we'll get a new mm anyway) */
648 || must_not_trace_exec(current
)
649 || (atomic_read(¤t
->fs
->count
) > 1)
650 || (atomic_read(¤t
->sig
->count
) > 1)
651 || (atomic_read(¤t
->files
->count
) > 1)) {
652 if (id_change
&& !capable(CAP_SETUID
))
654 if (cap_raised
&& !capable(CAP_SETPCAP
))
659 memset(bprm
->buf
,0,sizeof(bprm
->buf
));
660 return read_exec(bprm
->dentry
,0,bprm
->buf
,128,1);
664 * This function is used to produce the new IDs and capabilities
665 * from the old ones and the file's capabilities.
667 * The formula used for evolving capabilities is:
670 * (***) pP' = fP | (fI & pI)
671 * pE' = pP' & fE [NB. fE is 0 or ~0]
673 * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
674 * ' indicates post-exec().
677 void compute_creds(struct linux_binprm
*bprm
)
679 int new_permitted
= cap_t(bprm
->cap_permitted
) |
680 (cap_t(bprm
->cap_inheritable
) &
681 cap_t(current
->cap_inheritable
));
683 /* For init, we want to retain the capabilities set
684 * in the init_task struct. Thus we skip the usual
685 * capability rules */
686 if (current
->pid
!= 1) {
687 cap_t(current
->cap_permitted
) = new_permitted
;
688 cap_t(current
->cap_effective
) = new_permitted
&
689 cap_t(bprm
->cap_effective
);
692 /* AUD: Audit candidate if current->cap_effective is set */
694 current
->suid
= current
->euid
= current
->fsuid
= bprm
->e_uid
;
695 current
->sgid
= current
->egid
= current
->fsgid
= bprm
->e_gid
;
696 if (current
->euid
!= current
->uid
|| current
->egid
!= current
->gid
||
697 !cap_issubset(new_permitted
, current
->cap_permitted
))
698 current
->dumpable
= 0;
702 void remove_arg_zero(struct linux_binprm
*bprm
)
705 unsigned long offset
;
707 offset
= bprm
->p
% PAGE_SIZE
;
708 page
= (char*)bprm
->page
[bprm
->p
/PAGE_SIZE
];
709 while(bprm
->p
++,*(page
+offset
++))
710 if(offset
==PAGE_SIZE
){
712 page
= (char*)bprm
->page
[bprm
->p
/PAGE_SIZE
];
719 * cycle the list of binary formats handler, until one recognizes the image
721 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
724 struct linux_binfmt
*fmt
;
726 /* handle /sbin/loader.. */
728 struct exec
* eh
= (struct exec
*) bprm
->buf
;
729 struct linux_binprm bprm_loader
;
731 if (!bprm
->loader
&& eh
->fh
.f_magic
== 0x183 &&
732 (eh
->fh
.f_flags
& 0x3000) == 0x3000)
735 char * dynloader
[] = { "/sbin/loader" };
736 struct dentry
* dentry
;
741 bprm_loader
.p
= PAGE_SIZE
*MAX_ARG_PAGES
-sizeof(void *);
742 for (i
=0 ; i
<MAX_ARG_PAGES
; i
++) /* clear page-table */
743 bprm_loader
.page
[i
] = 0;
745 dentry
= open_namei(dynloader
[0], 0, 0);
746 retval
= PTR_ERR(dentry
);
749 bprm
->dentry
= dentry
;
750 bprm
->loader
= bprm_loader
.p
;
751 retval
= prepare_binprm(bprm
);
754 /* should call search_binary_handler recursively here,
755 but it does not matter */
759 for (try=0; try<2; try++) {
760 for (fmt
= formats
; fmt
; fmt
= fmt
->next
) {
761 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
764 retval
= fn(bprm
, regs
);
769 current
->did_exec
= 1;
772 if (retval
!= -ENOEXEC
)
774 if (!bprm
->dentry
) /* We don't have the dentry anymore */
777 if (retval
!= -ENOEXEC
) {
781 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
783 if (printable(bprm
->buf
[0]) &&
784 printable(bprm
->buf
[1]) &&
785 printable(bprm
->buf
[2]) &&
786 printable(bprm
->buf
[3]))
787 break; /* -ENOEXEC */
788 sprintf(modname
, "binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
789 request_module(modname
);
798 * sys_execve() executes a new program.
800 int do_execve(char * filename
, char ** argv
, char ** envp
, struct pt_regs
* regs
)
802 struct linux_binprm bprm
;
803 struct dentry
* dentry
;
807 bprm
.p
= PAGE_SIZE
*MAX_ARG_PAGES
-sizeof(void *);
808 for (i
=0 ; i
<MAX_ARG_PAGES
; i
++) /* clear page-table */
811 dentry
= open_namei(filename
, 0, 0);
812 retval
= PTR_ERR(dentry
);
816 bprm
.dentry
= dentry
;
817 bprm
.filename
= filename
;
822 if ((bprm
.argc
= count(argv
)) < 0) {
827 if ((bprm
.envc
= count(envp
)) < 0) {
832 retval
= prepare_binprm(&bprm
);
835 bprm
.p
= copy_strings(1, &bprm
.filename
, bprm
.page
, bprm
.p
, 2);
837 bprm
.p
= copy_strings(bprm
.envc
,envp
,bprm
.page
,bprm
.p
,0);
838 bprm
.p
= copy_strings(bprm
.argc
,argv
,bprm
.page
,bprm
.p
,0);
844 retval
= search_binary_handler(&bprm
,regs
);
849 /* Something went wrong, return the inode and free the argument pages*/
853 for (i
=0 ; i
<MAX_ARG_PAGES
; i
++)
854 free_page(bprm
.page
[i
]);