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_EM86
79 /* This cannot be configured out of the kernel */
83 int register_binfmt(struct linux_binfmt
* fmt
)
85 struct linux_binfmt
** tmp
= &formats
;
101 #ifdef CONFIG_MODULES
102 int unregister_binfmt(struct linux_binfmt
* fmt
)
104 struct linux_binfmt
** tmp
= &formats
;
115 #endif /* CONFIG_MODULES */
117 /* N.B. Error returns must be < 0 */
118 int open_dentry(struct dentry
* dentry
, int mode
)
120 struct inode
* inode
= dentry
->d_inode
;
122 struct list_head
* l
= NULL
;
126 l
= &inode
->i_sb
->s_files
;
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
);
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_strings()' 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 int copy_strings(int argc
,char ** argv
, struct linux_binprm
*bprm
)
237 if (get_user(str
, argv
+argc
) || !str
|| !(len
= strlen_user(str
)))
243 /* XXX: add architecture specific overflow check here. */
248 int offset
, bytes_to_copy
;
250 offset
= pos
% PAGE_SIZE
;
251 if (!(pag
= (char *) bprm
->page
[pos
/PAGE_SIZE
]) &&
252 !(pag
= (char *) bprm
->page
[pos
/PAGE_SIZE
] =
253 (unsigned long *) get_free_page(GFP_USER
)))
256 bytes_to_copy
= PAGE_SIZE
- offset
;
257 if (bytes_to_copy
> len
)
259 if (copy_from_user(pag
+ offset
, str
, bytes_to_copy
))
262 pos
+= bytes_to_copy
;
263 str
+= bytes_to_copy
;
264 len
-= bytes_to_copy
;
271 * Like copy_strings, but get argv and its values from kernel memory.
273 int copy_strings_kernel(int argc
,char ** argv
, struct linux_binprm
*bprm
)
276 mm_segment_t oldfs
= get_fs();
278 r
= copy_strings(argc
, argv
, bprm
);
283 int setup_arg_pages(struct linux_binprm
*bprm
)
285 unsigned long stack_base
;
286 struct vm_area_struct
*mpnt
;
289 stack_base
= STACK_TOP
- MAX_ARG_PAGES
*PAGE_SIZE
;
291 bprm
->p
+= stack_base
;
293 bprm
->loader
+= stack_base
;
294 bprm
->exec
+= stack_base
;
296 mpnt
= kmem_cache_alloc(vm_area_cachep
, SLAB_KERNEL
);
301 mpnt
->vm_mm
= current
->mm
;
302 mpnt
->vm_start
= PAGE_MASK
& (unsigned long) bprm
->p
;
303 mpnt
->vm_end
= STACK_TOP
;
304 mpnt
->vm_page_prot
= PAGE_COPY
;
305 mpnt
->vm_flags
= VM_STACK_FLAGS
;
308 mpnt
->vm_file
= NULL
;
310 insert_vm_struct(current
->mm
, mpnt
);
311 current
->mm
->total_vm
= (mpnt
->vm_end
- mpnt
->vm_start
) >> PAGE_SHIFT
;
314 for (i
= 0 ; i
< MAX_ARG_PAGES
; i
++) {
317 put_dirty_page(current
,bprm
->page
[i
],stack_base
);
319 stack_base
+= PAGE_SIZE
;
326 * Read in the complete executable. This is used for "-N" files
327 * that aren't on a block boundary, and for files on filesystems
328 * without get_block support.
330 int read_exec(struct dentry
*dentry
, unsigned long offset
,
331 char * addr
, unsigned long count
, int to_kmem
)
334 struct inode
* inode
= dentry
->d_inode
;
335 int result
= -ENOEXEC
;
337 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
339 if (init_private_file(&file
, dentry
, 1))
341 if (!file
.f_op
->read
)
343 if (file
.f_op
->llseek
) {
344 if (file
.f_op
->llseek(&file
,offset
,0) != offset
)
349 mm_segment_t old_fs
= get_fs();
351 result
= file
.f_op
->read(&file
, addr
, count
, &file
.f_pos
);
354 result
= verify_area(VERIFY_WRITE
, addr
, count
);
357 result
= file
.f_op
->read(&file
, addr
, count
, &file
.f_pos
);
360 if (file
.f_op
->release
)
361 file
.f_op
->release(inode
,&file
);
366 static int exec_mmap(void)
368 struct mm_struct
* mm
, * old_mm
;
370 old_mm
= current
->mm
;
371 if (old_mm
&& atomic_read(&old_mm
->mm_users
) == 1) {
372 flush_cache_mm(old_mm
);
374 release_segments(old_mm
);
376 flush_tlb_mm(old_mm
);
382 mm
->cpu_vm_mask
= (1UL << smp_processor_id());
385 mm
->pgd
= pgd_alloc();
387 struct mm_struct
*active_mm
= current
->active_mm
;
390 current
->active_mm
= mm
;
400 kmem_cache_free(mm_cachep
, mm
);
406 * This function makes sure the current process has its own signal table,
407 * so that flush_signal_handlers can later reset the handlers without
408 * disturbing other processes. (Other processes might share the signal
409 * table via the CLONE_SIGHAND option to clone().)
412 static inline int make_private_signals(void)
414 struct signal_struct
* newsig
;
416 if (atomic_read(¤t
->sig
->count
) <= 1)
418 newsig
= kmalloc(sizeof(*newsig
), GFP_KERNEL
);
421 spin_lock_init(&newsig
->siglock
);
422 atomic_set(&newsig
->count
, 1);
423 memcpy(newsig
->action
, current
->sig
->action
, sizeof(newsig
->action
));
424 current
->sig
= newsig
;
429 * If make_private_signals() made a copy of the signal table, decrement the
430 * refcount of the original table, and free it if necessary.
431 * We don't do that in make_private_signals() so that we can back off
432 * in flush_old_exec() if an error occurs after calling make_private_signals().
435 static inline void release_old_signals(struct signal_struct
* oldsig
)
437 if (current
->sig
== oldsig
)
439 if (atomic_dec_and_test(&oldsig
->count
))
444 * These functions flushes out all traces of the currently running executable
445 * so that a new one can be started
448 static inline void flush_old_files(struct files_struct
* files
)
454 unsigned long set
, i
;
457 if (i
>= files
->max_fds
)
459 set
= xchg(&files
->close_on_exec
.fds_bits
[j
], 0);
461 for ( ; set
; i
++,set
>>= 1) {
468 int flush_old_exec(struct linux_binprm
* bprm
)
472 struct signal_struct
* oldsig
;
475 * Make sure we have a private signal table
477 oldsig
= current
->sig
;
478 retval
= make_private_signals();
479 if (retval
) goto flush_failed
;
482 * Release all of the old mmap stuff
484 retval
= exec_mmap();
485 if (retval
) goto mmap_failed
;
487 /* This is the point of no return */
488 release_old_signals(oldsig
);
490 if (current
->euid
== current
->uid
&& current
->egid
== current
->gid
)
491 current
->dumpable
= 1;
492 name
= bprm
->filename
;
493 for (i
=0; (ch
= *(name
++)) != '\0';) {
498 current
->comm
[i
++] = ch
;
500 current
->comm
[i
] = '\0';
504 if (bprm
->e_uid
!= current
->euid
|| bprm
->e_gid
!= current
->egid
||
505 permission(bprm
->dentry
->d_inode
,MAY_READ
))
506 current
->dumpable
= 0;
508 flush_signal_handlers(current
);
509 flush_old_files(current
->files
);
514 if (current
->sig
!= oldsig
)
517 current
->sig
= oldsig
;
522 * We mustn't allow tracing of suid binaries, unless
523 * the tracer has the capability to trace anything..
525 static inline int must_not_trace_exec(struct task_struct
* p
)
527 return (p
->flags
& PF_PTRACED
) && !cap_raised(p
->p_pptr
->cap_effective
, CAP_SYS_PTRACE
);
531 * Fill the binprm structure from the inode.
532 * Check permissions, then read the first 512 bytes
534 int prepare_binprm(struct linux_binprm
*bprm
)
537 int retval
,id_change
,cap_raised
;
538 struct inode
* inode
= bprm
->dentry
->d_inode
;
540 mode
= inode
->i_mode
;
541 if (!S_ISREG(mode
)) /* must be regular file */
543 if (!(mode
& 0111)) /* with at least _one_ execute bit set */
545 if (IS_NOEXEC(inode
)) /* FS mustn't be mounted noexec */
549 if ((retval
= permission(inode
, MAY_EXEC
)) != 0)
551 /* better not execute files which are being written to */
552 if (atomic_read(&inode
->i_writecount
) > 0)
555 bprm
->e_uid
= current
->euid
;
556 bprm
->e_gid
= current
->egid
;
557 id_change
= cap_raised
= 0;
560 if (mode
& S_ISUID
) {
561 bprm
->e_uid
= inode
->i_uid
;
562 if (bprm
->e_uid
!= current
->euid
)
568 * If setgid is set but no group execute bit then this
569 * is a candidate for mandatory locking, not a setgid
572 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
)) {
573 bprm
->e_gid
= inode
->i_gid
;
574 if (!in_group_p(bprm
->e_gid
))
578 /* We don't have VFS support for capabilities yet */
579 cap_clear(bprm
->cap_inheritable
);
580 cap_clear(bprm
->cap_permitted
);
581 cap_clear(bprm
->cap_effective
);
583 /* To support inheritance of root-permissions and suid-root
584 * executables under compatibility mode, we raise the
585 * effective and inherited bitmasks of the executable file
586 * (translation: we set the executable "capability dumb" and
587 * set the allowed set to maximum). We don't set any forced
590 * If only the real uid is 0, we only raise the inheritable
591 * bitmask of the executable file (translation: we set the
592 * allowed set to maximum and the application to "capability
596 if (!issecure(SECURE_NOROOT
)) {
597 if (bprm
->e_uid
== 0 || current
->uid
== 0)
598 cap_set_full(bprm
->cap_inheritable
);
599 if (bprm
->e_uid
== 0)
600 cap_set_full(bprm
->cap_effective
);
603 /* Only if pP' is _not_ a subset of pP, do we consider there
604 * has been a capability related "change of capability". In
605 * such cases, we need to check that the elevation of
606 * privilege does not go against other system constraints.
607 * The new Permitted set is defined below -- see (***). */
609 kernel_cap_t working
=
610 cap_combine(bprm
->cap_permitted
,
611 cap_intersect(bprm
->cap_inheritable
,
612 current
->cap_inheritable
));
613 if (!cap_issubset(working
, current
->cap_permitted
)) {
618 if (id_change
|| cap_raised
) {
619 /* We can't suid-execute if we're sharing parts of the executable */
620 /* or if we're being traced (or if suid execs are not allowed) */
621 /* (current->mm->mm_users > 1 is ok, as we'll get a new mm anyway) */
623 || must_not_trace_exec(current
)
624 || (atomic_read(¤t
->fs
->count
) > 1)
625 || (atomic_read(¤t
->sig
->count
) > 1)
626 || (atomic_read(¤t
->files
->count
) > 1)) {
627 if (id_change
&& !capable(CAP_SETUID
))
629 if (cap_raised
&& !capable(CAP_SETPCAP
))
634 memset(bprm
->buf
,0,sizeof(bprm
->buf
));
635 return read_exec(bprm
->dentry
,0,bprm
->buf
,128,1);
639 * This function is used to produce the new IDs and capabilities
640 * from the old ones and the file's capabilities.
642 * The formula used for evolving capabilities is:
645 * (***) pP' = fP | (fI & pI)
646 * pE' = pP' & fE [NB. fE is 0 or ~0]
648 * I=Inheritable, P=Permitted, E=Effective // p=process, f=file
649 * ' indicates post-exec().
652 void compute_creds(struct linux_binprm
*bprm
)
654 int new_permitted
= cap_t(bprm
->cap_permitted
) |
655 (cap_t(bprm
->cap_inheritable
) &
656 cap_t(current
->cap_inheritable
));
658 /* For init, we want to retain the capabilities set
659 * in the init_task struct. Thus we skip the usual
660 * capability rules */
661 if (current
->pid
!= 1) {
662 cap_t(current
->cap_permitted
) = new_permitted
;
663 cap_t(current
->cap_effective
) = new_permitted
&
664 cap_t(bprm
->cap_effective
);
667 /* AUD: Audit candidate if current->cap_effective is set */
669 current
->suid
= current
->euid
= current
->fsuid
= bprm
->e_uid
;
670 current
->sgid
= current
->egid
= current
->fsgid
= bprm
->e_gid
;
671 if (current
->euid
!= current
->uid
|| current
->egid
!= current
->gid
||
672 !cap_issubset(new_permitted
, current
->cap_permitted
))
673 current
->dumpable
= 0;
677 void remove_arg_zero(struct linux_binprm
*bprm
)
680 unsigned long offset
;
682 offset
= bprm
->p
% PAGE_SIZE
;
683 page
= (char*)bprm
->page
[bprm
->p
/PAGE_SIZE
];
684 while(bprm
->p
++,*(page
+offset
++))
685 if(offset
==PAGE_SIZE
){
687 page
= (char*)bprm
->page
[bprm
->p
/PAGE_SIZE
];
694 * cycle the list of binary formats handler, until one recognizes the image
696 int search_binary_handler(struct linux_binprm
*bprm
,struct pt_regs
*regs
)
699 struct linux_binfmt
*fmt
;
701 /* handle /sbin/loader.. */
703 struct exec
* eh
= (struct exec
*) bprm
->buf
;
704 struct linux_binprm bprm_loader
;
706 if (!bprm
->loader
&& eh
->fh
.f_magic
== 0x183 &&
707 (eh
->fh
.f_flags
& 0x3000) == 0x3000)
710 char * dynloader
[] = { "/sbin/loader" };
711 struct dentry
* dentry
;
716 bprm_loader
.p
= PAGE_SIZE
*MAX_ARG_PAGES
-sizeof(void *);
717 for (i
=0 ; i
<MAX_ARG_PAGES
; i
++) /* clear page-table */
718 bprm_loader
.page
[i
] = 0;
720 dentry
= open_namei(dynloader
[0], 0, 0);
721 retval
= PTR_ERR(dentry
);
724 bprm
->dentry
= dentry
;
725 bprm
->loader
= bprm_loader
.p
;
726 retval
= prepare_binprm(bprm
);
729 /* should call search_binary_handler recursively here,
730 but it does not matter */
734 for (try=0; try<2; try++) {
735 for (fmt
= formats
; fmt
; fmt
= fmt
->next
) {
736 int (*fn
)(struct linux_binprm
*, struct pt_regs
*) = fmt
->load_binary
;
739 retval
= fn(bprm
, regs
);
744 current
->did_exec
= 1;
747 if (retval
!= -ENOEXEC
)
749 if (!bprm
->dentry
) /* We don't have the dentry anymore */
752 if (retval
!= -ENOEXEC
) {
756 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
758 if (printable(bprm
->buf
[0]) &&
759 printable(bprm
->buf
[1]) &&
760 printable(bprm
->buf
[2]) &&
761 printable(bprm
->buf
[3]))
762 break; /* -ENOEXEC */
763 sprintf(modname
, "binfmt-%04x", *(unsigned short *)(&bprm
->buf
[2]));
764 request_module(modname
);
773 * sys_execve() executes a new program.
775 int do_execve(char * filename
, char ** argv
, char ** envp
, struct pt_regs
* regs
)
777 struct linux_binprm bprm
;
778 struct dentry
* dentry
;
782 bprm
.p
= PAGE_SIZE
*MAX_ARG_PAGES
-sizeof(void *);
783 memset(bprm
.page
, 0, MAX_ARG_PAGES
*sizeof(bprm
.page
[0]));
785 dentry
= open_namei(filename
, 0, 0);
786 retval
= PTR_ERR(dentry
);
790 bprm
.dentry
= dentry
;
791 bprm
.filename
= filename
;
795 if ((bprm
.argc
= count(argv
)) < 0) {
800 if ((bprm
.envc
= count(envp
)) < 0) {
805 retval
= prepare_binprm(&bprm
);
809 retval
= copy_strings_kernel(1, &bprm
.filename
, &bprm
);
814 retval
= copy_strings(bprm
.envc
, envp
, &bprm
);
818 retval
= copy_strings(bprm
.argc
, argv
, &bprm
);
822 retval
= search_binary_handler(&bprm
,regs
);
828 /* Something went wrong, return the inode and free the argument pages*/
832 /* Assumes that free_page() can take a NULL argument. */
833 /* I hope this is ok for all architectures */
834 for (i
=0 ; i
<MAX_ARG_PAGES
; i
++)
835 free_page(bprm
.page
[i
]);
840 int do_coredump(long signr
, struct pt_regs
* regs
)
842 struct linux_binfmt
* binfmt
;
843 char corename
[6+sizeof(current
->comm
)];
845 struct dentry
* dentry
;
846 struct inode
* inode
;
849 binfmt
= current
->binfmt
;
850 if (!binfmt
|| !binfmt
->core_dump
)
852 if (!current
->dumpable
|| atomic_read(¤t
->mm
->mm_users
) != 1)
853 current
->dumpable
= 0;
854 if (current
->rlim
[RLIMIT_CORE
].rlim_cur
< binfmt
->min_coredump
)
857 memcpy(corename
,"core.", 5);
859 memcpy(corename
+5,current
->comm
,sizeof(current
->comm
));
863 file
= filp_open(corename
, O_CREAT
| 2 | O_TRUNC
| O_NOFOLLOW
, 0600);
866 dentry
= file
->f_dentry
;
867 inode
= dentry
->d_inode
;
868 if (inode
->i_nlink
> 1)
869 goto close_fail
; /* multiple links - don't dump */
871 if (!S_ISREG(inode
->i_mode
))
873 if (!inode
->i_op
|| !inode
->i_op
->default_file_ops
)
875 if (!file
->f_op
->write
)
877 if (!binfmt
->core_dump(signr
, regs
, file
))
879 filp_close(file
, NULL
);
884 filp_close(file
, NULL
);