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/user.h>
33 #include <linux/smp_lock.h>
34 #include <linux/init.h>
36 #include <asm/uaccess.h>
37 #include <asm/pgtable.h>
38 #include <asm/mmu_context.h>
41 #include <linux/kmod.h>
45 * Here are the actual binaries that will be accepted:
46 * add more with "register_binfmt()" if using modules...
48 * These are defined again for the 'real' modules if you are using a
49 * module definition for these routines.
52 static struct linux_binfmt
*formats
= (struct linux_binfmt
*) NULL
;
54 void __init
binfmt_setup(void)
56 #ifdef CONFIG_BINFMT_MISC
60 #ifdef CONFIG_BINFMT_ELF
64 #ifdef CONFIG_BINFMT_ELF32
68 #ifdef CONFIG_BINFMT_AOUT
72 #ifdef CONFIG_BINFMT_AOUT32
76 #ifdef CONFIG_BINFMT_JAVA
80 #ifdef CONFIG_BINFMT_EM86
84 /* This cannot be configured out of the kernel */
88 int register_binfmt(struct linux_binfmt
* fmt
)
90 struct linux_binfmt
** tmp
= &formats
;
106 #ifdef CONFIG_MODULES
107 int unregister_binfmt(struct linux_binfmt
* fmt
)
109 struct linux_binfmt
** tmp
= &formats
;
120 #endif /* CONFIG_MODULES */
122 /* N.B. Error returns must be < 0 */
123 int open_dentry(struct dentry
* dentry
, int mode
)
125 struct inode
* inode
= dentry
->d_inode
;
130 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
132 fd
= get_unused_fd();
135 f
= get_empty_filp();
139 f
->f_mode
= (mode
+1) & O_ACCMODE
;
140 f
->f_dentry
= dentry
;
143 f
->f_op
= inode
->i_op
->default_file_ops
;
145 error
= f
->f_op
->open(inode
,f
);
165 * Note that a shared library must be both readable and executable due to
168 * Also note that we take the address to load from from the file itself.
170 asmlinkage
int sys_uselib(const char * library
)
174 struct linux_binfmt
* fmt
;
177 fd
= sys_open(library
, 0, 0);
183 if (file
&& file
->f_dentry
&& file
->f_op
&& file
->f_op
->read
) {
184 for (fmt
= formats
; fmt
; fmt
= fmt
->next
) {
185 int (*fn
)(int) = fmt
->load_shlib
;
188 /* N.B. Should use file instead of fd */
190 if (retval
!= -ENOEXEC
)
202 * count() counts the number of arguments/envelopes
204 static int count(char ** argv
)
213 error
= get_user(p
,argv
);
226 * 'copy_string()' copies argument/envelope strings from user
227 * memory to free pages in kernel mem. These are in a format ready
228 * to be put directly into the top of new user memory.
230 * Modified by TYT, 11/24/91 to add the from_kmem argument, which specifies
231 * whether the string and the string array are from user or kernel segments:
233 * from_kmem argv * argv **
234 * 0 user space user space
235 * 1 kernel space user space
236 * 2 kernel space kernel space
238 * We do this by playing games with the fs segment register. Since it
239 * is expensive to load a segment register, we try to avoid calling
240 * set_fs() unless we absolutely have to.
242 unsigned long copy_strings(int argc
,char ** argv
,unsigned long *page
,
243 unsigned long p
, int from_kmem
)
249 return 0; /* bullet-proofing */
259 get_user(str
, argv
+argc
);
261 panic("VFS: argc is wrong");
264 len
= strlen_user(str
); /* includes the '\0' */
265 if (p
< len
) { /* this shouldn't happen - 128kB */
273 int offset
, bytes_to_copy
;
275 offset
= pos
% PAGE_SIZE
;
276 if (!(pag
= (char *) page
[pos
/PAGE_SIZE
]) &&
277 !(pag
= (char *) page
[pos
/PAGE_SIZE
] =
278 (unsigned long *) get_free_page(GFP_USER
))) {
283 bytes_to_copy
= PAGE_SIZE
- offset
;
284 if (bytes_to_copy
> len
)
286 copy_from_user(pag
+ offset
, str
, bytes_to_copy
);
287 pos
+= bytes_to_copy
;
288 str
+= bytes_to_copy
;
289 len
-= bytes_to_copy
;
297 unsigned long setup_arg_pages(unsigned long p
, struct linux_binprm
* bprm
)
299 unsigned long stack_base
;
300 struct vm_area_struct
*mpnt
;
303 stack_base
= STACK_TOP
- MAX_ARG_PAGES
*PAGE_SIZE
;
307 bprm
->loader
+= stack_base
;
308 bprm
->exec
+= stack_base
;
310 mpnt
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
312 mpnt
->vm_mm
= current
->mm
;
313 mpnt
->vm_start
= PAGE_MASK
& (unsigned long) p
;
314 mpnt
->vm_end
= STACK_TOP
;
315 mpnt
->vm_page_prot
= PAGE_COPY
;
316 mpnt
->vm_flags
= VM_STACK_FLAGS
;
319 mpnt
->vm_file
= NULL
;
321 insert_vm_struct(current
->mm
, mpnt
);
322 current
->mm
->total_vm
= (mpnt
->vm_end
- mpnt
->vm_start
) >> PAGE_SHIFT
;
325 for (i
= 0 ; i
< MAX_ARG_PAGES
; i
++) {
328 put_dirty_page(current
,bprm
->page
[i
],stack_base
);
330 stack_base
+= PAGE_SIZE
;
336 * Read in the complete executable. This is used for "-N" files
337 * that aren't on a block boundary, and for files on filesystems
338 * without bmap support.
340 int read_exec(struct dentry
*dentry
, unsigned long offset
,
341 char * addr
, unsigned long count
, int to_kmem
)
344 struct inode
* inode
= dentry
->d_inode
;
345 int result
= -ENOEXEC
;
347 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
349 if (init_private_file(&file
, dentry
, 1))
351 if (!file
.f_op
->read
)
353 if (file
.f_op
->llseek
) {
354 if (file
.f_op
->llseek(&file
,offset
,0) != offset
)
359 mm_segment_t old_fs
= get_fs();
361 result
= file
.f_op
->read(&file
, addr
, count
, &file
.f_pos
);
364 result
= verify_area(VERIFY_WRITE
, addr
, count
);
367 result
= file
.f_op
->read(&file
, addr
, count
, &file
.f_pos
);
370 if (file
.f_op
->release
)
371 file
.f_op
->release(inode
,&file
);
376 static int exec_mmap(void)
378 struct mm_struct
* mm
, * old_mm
;
381 if (atomic_read(¤t
->mm
->count
) == 1) {
382 flush_cache_mm(current
->mm
);
384 release_segments(current
->mm
);
385 exit_mmap(current
->mm
);
386 flush_tlb_mm(current
->mm
);
395 mm
->cpu_vm_mask
= (1UL << smp_processor_id());
399 * Make sure we have a private ldt if needed ...
401 nr
= current
->tarray_ptr
- &task
[0];
402 copy_segments(nr
, current
, mm
);
404 old_mm
= current
->mm
;
406 retval
= new_page_tables(current
);
409 activate_context(current
);
416 * Failure ... restore the prior mm_struct.
419 /* The pgd belongs to the parent ... don't free it! */
421 current
->mm
= old_mm
;
422 /* restore the ldt for this task */
423 copy_segments(nr
, current
, NULL
);
431 * This function makes sure the current process has its own signal table,
432 * so that flush_signal_handlers can later reset the handlers without
433 * disturbing other processes. (Other processes might share the signal
434 * table via the CLONE_SIGHAND option to clone().)
437 static inline int make_private_signals(void)
439 struct signal_struct
* newsig
;
441 if (atomic_read(¤t
->sig
->count
) <= 1)
443 newsig
= kmalloc(sizeof(*newsig
), GFP_KERNEL
);
446 spin_lock_init(&newsig
->siglock
);
447 atomic_set(&newsig
->count
, 1);
448 memcpy(newsig
->action
, current
->sig
->action
, sizeof(newsig
->action
));
449 current
->sig
= newsig
;
454 * If make_private_signals() made a copy of the signal table, decrement the
455 * refcount of the original table, and free it if necessary.
456 * We don't do that in make_private_signals() so that we can back off
457 * in flush_old_exec() if an error occurs after calling make_private_signals().
460 static inline void release_old_signals(struct signal_struct
* oldsig
)
462 if (current
->sig
== oldsig
)
464 if (atomic_dec_and_test(&oldsig
->count
))
469 * These functions flushes out all traces of the currently running executable
470 * so that a new one can be started
473 static inline void flush_old_files(struct files_struct
* files
)
479 unsigned long set
, i
;
482 if (i
>= files
->max_fds
)
484 set
= files
->close_on_exec
.fds_bits
[j
];
485 files
->close_on_exec
.fds_bits
[j
] = 0;
487 for ( ; set
; i
++,set
>>= 1) {
494 int flush_old_exec(struct linux_binprm
* bprm
)
498 struct signal_struct
* oldsig
;
501 * Make sure we have a private signal table
503 oldsig
= current
->sig
;
504 retval
= make_private_signals();
505 if (retval
) goto flush_failed
;
508 * Release all of the old mmap stuff
510 retval
= exec_mmap();
511 if (retval
) goto mmap_failed
;
513 /* This is the point of no return */
514 release_old_signals(oldsig
);
516 if (current
->euid
== current
->uid
&& current
->egid
== current
->gid
)
517 current
->dumpable
= 1;
518 name
= bprm
->filename
;
519 for (i
=0; (ch
= *(name
++)) != '\0';) {
524 current
->comm
[i
++] = ch
;
526 current
->comm
[i
] = '\0';
530 if (bprm
->e_uid
!= current
->euid
|| bprm
->e_gid
!= current
->egid
||
531 permission(bprm
->dentry
->d_inode
,MAY_READ
))
532 current
->dumpable
= 0;
534 flush_signal_handlers(current
);
535 flush_old_files(current
->files
);
540 if (current
->sig
!= oldsig
)
543 current
->sig
= oldsig
;
548 * We mustn't allow tracing of suid binaries, unless
549 * the tracer has the capability to trace anything..
551 static inline int must_not_trace_exec(struct task_struct
* p
)
553 return (p
->flags
& PF_PTRACED
) && !cap_raised(p
->p_pptr
->cap_effective
, CAP_SYS_PTRACE
);
557 * Fill the binprm structure from the inode.
558 * Check permissions, then read the first 512 bytes
560 int prepare_binprm(struct linux_binprm
*bprm
)
563 int retval
,id_change
,cap_raised
;
564 struct inode
* inode
= bprm
->dentry
->d_inode
;
566 mode
= inode
->i_mode
;
567 if (!S_ISREG(mode
)) /* must be regular file */
569 if (!(mode
& 0111)) /* with at least _one_ execute bit set */
571 if (IS_NOEXEC(inode
)) /* FS mustn't be mounted noexec */
575 if ((retval
= permission(inode
, MAY_EXEC
)) != 0)
577 /* better not execute files which are being written to */
578 if (inode
->i_writecount
> 0)
581 bprm
->e_uid
= current
->euid
;
582 bprm
->e_gid
= current
->egid
;
583 id_change
= cap_raised
= 0;
586 if (mode
& S_ISUID
) {
587 bprm
->e_uid
= inode
->i_uid
;
588 if (bprm
->e_uid
!= current
->euid
)
594 * If setgid is set but no group execute bit then this
595 * is a candidate for mandatory locking, not a setgid
598 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
599 bprm
->e_gid
= inode
->i_gid
;
600 if (!in_group_p(bprm
->e_gid
))
604 /* We don't have VFS support for capabilities yet */
605 cap_clear(bprm
->cap_inheritable
);
606 cap_clear(bprm
->cap_permitted
);
607 cap_clear(bprm
->cap_effective
);
609 /* To support inheritance of root-permissions and suid-root
610 * executables under compatibility mode, we raise the
611 * effective and inherited bitmasks of the executable file
612 * (translation: we set the executable "capability dumb" and
613 * set the allowed set to maximum). We don't set any forced
616 * If only the real uid is 0, we only raise the inheritable
617 * bitmask of the executable file (translation: we set the
618 * allowed set to maximum and the application to "capability
622 if (!issecure(SECURE_NOROOT
)) {
623 if (bprm
->e_uid
== 0 || current
->uid
== 0)
624 cap_set_full(bprm
->cap_inheritable
);
625 if (bprm
->e_uid
== 0)
626 cap_set_full(bprm
->cap_effective
);
629 /* Only if pP' is _not_ a subset of pP, do we consider there
630 * has been a capability related "change of capability". In
631 * such cases, we need to check that the elevation of
632 * privilege does not go against other system constraints.
633 * The new Permitted set is defined below -- see (***). */
635 kernel_cap_t working
=
636 cap_combine(bprm
->cap_permitted
,
637 cap_intersect(bprm
->cap_inheritable
,
638 current
->cap_inheritable
));
639 if (!cap_issubset(working
, current
->cap_permitted
)) {
644 if (id_change
|| cap_raised
) {
645 /* We can't suid-execute if we're sharing parts of the executable */
646 /* or if we're being traced (or if suid execs are not allowed) */
647 /* (current->mm->count > 1 is ok, as we'll get a new mm anyway) */
649 || must_not_trace_exec(current
)
650 || (atomic_read(¤t
->fs
->count
) > 1)
651 || (atomic_read(¤t
->sig
->count
) > 1)
652 || (atomic_read(¤t
->files
->count
) > 1)) {
653 if (id_change
&& !capable(CAP_SETUID
))
655 if (cap_raised
&& !capable(CAP_SETPCAP
))
660 memset(bprm
->buf
,0,sizeof(bprm
->buf
));
661 return read_exec(bprm
->dentry
,0,bprm
->buf
,128,1);
665 * This function is used to produce the new IDs and capabilities
666 * from the old ones and the file's capabilities.
668 * The formula used for evolving capabilities is:
671 * (***) pP' = fP | (fI & pI)
672 * pE' = pP' & fE [NB. fE is 0 or ~0]
674 * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
675 * ' indicates post-exec().
678 void compute_creds(struct linux_binprm
*bprm
)
680 int new_permitted
= cap_t(bprm
->cap_permitted
) |
681 (cap_t(bprm
->cap_inheritable
) &
682 cap_t(current
->cap_inheritable
));
684 /* For init, we want to retain the capabilities set
685 * in the init_task struct. Thus we skip the usual
686 * capability rules */
687 if (current
->pid
!= 1) {
688 cap_t(current
->cap_permitted
) = new_permitted
;
689 cap_t(current
->cap_effective
) = new_permitted
&
690 cap_t(bprm
->cap_effective
);
693 /* AUD: Audit candidate if current->cap_effective is set */
695 current
->suid
= current
->euid
= current
->fsuid
= bprm
->e_uid
;
696 current
->sgid
= current
->egid
= current
->fsgid
= bprm
->e_gid
;
697 if (current
->euid
!= current
->uid
|| current
->egid
!= current
->gid
||
698 !cap_issubset(new_permitted
, current
->cap_permitted
))
699 current
->dumpable
= 0;
703 void remove_arg_zero(struct linux_binprm
*bprm
)
706 unsigned long offset
;
708 offset
= bprm
->p
% PAGE_SIZE
;
709 page
= (char*)bprm
->page
[bprm
->p
/PAGE_SIZE
];
710 while(bprm
->p
++,*(page
+offset
++))
711 if(offset
==PAGE_SIZE
){
713 page
= (char*)bprm
->page
[bprm
->p
/PAGE_SIZE
];
720 * cycle the list of binary formats handler, until one recognizes the image
722 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
725 struct linux_binfmt
*fmt
;
727 /* handle /sbin/loader.. */
729 struct exec
* eh
= (struct exec
*) bprm
->buf
;
730 struct linux_binprm bprm_loader
;
732 if (!bprm
->loader
&& eh
->fh
.f_magic
== 0x183 &&
733 (eh
->fh
.f_flags
& 0x3000) == 0x3000)
736 char * dynloader
[] = { "/sbin/loader" };
737 struct dentry
* dentry
;
742 bprm_loader
.p
= PAGE_SIZE
*MAX_ARG_PAGES
-sizeof(void *);
743 for (i
=0 ; i
<MAX_ARG_PAGES
; i
++) /* clear page-table */
744 bprm_loader
.page
[i
] = 0;
746 dentry
= open_namei(dynloader
[0], 0, 0);
747 retval
= PTR_ERR(dentry
);
750 bprm
->dentry
= dentry
;
751 bprm
->loader
= bprm_loader
.p
;
752 retval
= prepare_binprm(bprm
);
755 /* should call search_binary_handler recursively here,
756 but it does not matter */
760 for (try=0; try<2; try++) {
761 for (fmt
= formats
; fmt
; fmt
= fmt
->next
) {
762 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
765 retval
= fn(bprm
, regs
);
770 current
->did_exec
= 1;
773 if (retval
!= -ENOEXEC
)
775 if (!bprm
->dentry
) /* We don't have the dentry anymore */
778 if (retval
!= -ENOEXEC
) {
782 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
784 if (printable(bprm
->buf
[0]) &&
785 printable(bprm
->buf
[1]) &&
786 printable(bprm
->buf
[2]) &&
787 printable(bprm
->buf
[3]))
788 break; /* -ENOEXEC */
789 sprintf(modname
, "binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
790 request_module(modname
);
799 * sys_execve() executes a new program.
801 int do_execve(char * filename
, char ** argv
, char ** envp
, struct pt_regs
* regs
)
803 struct linux_binprm bprm
;
804 struct dentry
* dentry
;
808 bprm
.p
= PAGE_SIZE
*MAX_ARG_PAGES
-sizeof(void *);
809 for (i
=0 ; i
<MAX_ARG_PAGES
; i
++) /* clear page-table */
812 dentry
= open_namei(filename
, 0, 0);
813 retval
= PTR_ERR(dentry
);
817 bprm
.dentry
= dentry
;
818 bprm
.filename
= filename
;
823 if ((bprm
.argc
= count(argv
)) < 0) {
828 if ((bprm
.envc
= count(envp
)) < 0) {
833 retval
= prepare_binprm(&bprm
);
836 bprm
.p
= copy_strings(1, &bprm
.filename
, bprm
.page
, bprm
.p
, 2);
838 bprm
.p
= copy_strings(bprm
.envc
,envp
,bprm
.page
,bprm
.p
,0);
839 bprm
.p
= copy_strings(bprm
.argc
,argv
,bprm
.page
,bprm
.p
,0);
845 retval
= search_binary_handler(&bprm
,regs
);
850 /* Something went wrong, return the inode and free the argument pages*/
854 for (i
=0 ; i
<MAX_ARG_PAGES
; i
++)
855 free_page(bprm
.page
[i
]);