2 * linux/fs/binfmt_elf.c
4 * These are the functions used to load ELF format executables as used
5 * on SVr4 machines. Information on the format may be found in the book
6 * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
9 * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
12 #include <linux/module.h>
13 #include <linux/kernel.h>
15 #include <linux/stat.h>
16 #include <linux/time.h>
18 #include <linux/mman.h>
19 #include <linux/a.out.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/ptrace.h>
27 #include <linux/slab.h>
28 #include <linux/shm.h>
29 #include <linux/personality.h>
30 #include <linux/elfcore.h>
31 #include <linux/init.h>
32 #include <linux/highuid.h>
33 #include <linux/smp.h>
34 #include <linux/compiler.h>
35 #include <linux/highmem.h>
36 #include <linux/pagemap.h>
37 #include <linux/security.h>
38 #include <linux/syscalls.h>
39 #include <linux/random.h>
40 #include <linux/elf.h>
41 #include <linux/utsname.h>
42 #include <asm/uaccess.h>
43 #include <asm/param.h>
46 static int load_elf_binary(struct linux_binprm
*bprm
, struct pt_regs
*regs
);
47 static int load_elf_library(struct file
*);
48 static unsigned long elf_map (struct file
*, unsigned long, struct elf_phdr
*, int, int, unsigned long);
51 * If we don't support core dumping, then supply a NULL so we
54 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
55 static int elf_core_dump(long signr
, struct pt_regs
*regs
, struct file
*file
, unsigned long limit
);
57 #define elf_core_dump NULL
60 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
61 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
63 #define ELF_MIN_ALIGN PAGE_SIZE
66 #ifndef ELF_CORE_EFLAGS
67 #define ELF_CORE_EFLAGS 0
70 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
71 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
72 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
74 static struct linux_binfmt elf_format
= {
75 .module
= THIS_MODULE
,
76 .load_binary
= load_elf_binary
,
77 .load_shlib
= load_elf_library
,
78 .core_dump
= elf_core_dump
,
79 .min_coredump
= ELF_EXEC_PAGESIZE
,
83 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
85 static int set_brk(unsigned long start
, unsigned long end
)
87 start
= ELF_PAGEALIGN(start
);
88 end
= ELF_PAGEALIGN(end
);
91 down_write(¤t
->mm
->mmap_sem
);
92 addr
= do_brk(start
, end
- start
);
93 up_write(¤t
->mm
->mmap_sem
);
97 current
->mm
->start_brk
= current
->mm
->brk
= end
;
101 /* We need to explicitly zero any fractional pages
102 after the data section (i.e. bss). This would
103 contain the junk from the file that should not
106 static int padzero(unsigned long elf_bss
)
110 nbyte
= ELF_PAGEOFFSET(elf_bss
);
112 nbyte
= ELF_MIN_ALIGN
- nbyte
;
113 if (clear_user((void __user
*) elf_bss
, nbyte
))
119 /* Let's use some macros to make this stack manipulation a litle clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130 (((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
135 create_elf_tables(struct linux_binprm
*bprm
, struct elfhdr
*exec
,
136 int interp_aout
, unsigned long load_addr
,
137 unsigned long interp_load_addr
)
139 unsigned long p
= bprm
->p
;
140 int argc
= bprm
->argc
;
141 int envc
= bprm
->envc
;
142 elf_addr_t __user
*argv
;
143 elf_addr_t __user
*envp
;
144 elf_addr_t __user
*sp
;
145 elf_addr_t __user
*u_platform
;
146 const char *k_platform
= ELF_PLATFORM
;
148 elf_addr_t
*elf_info
;
150 struct task_struct
*tsk
= current
;
151 struct vm_area_struct
*vma
;
154 * In some cases (e.g. Hyper-Threading), we want to avoid L1
155 * evictions by the processes running on the same package. One
156 * thing we can do is to shuffle the initial stack for them.
159 p
= arch_align_stack(p
);
162 * If this architecture has a platform capability string, copy it
163 * to userspace. In some cases (Sparc), this info is impossible
164 * for userspace to get any other way, in others (i386) it is
169 size_t len
= strlen(k_platform
) + 1;
171 u_platform
= (elf_addr_t __user
*)STACK_ALLOC(p
, len
);
172 if (__copy_to_user(u_platform
, k_platform
, len
))
176 /* Create the ELF interpreter info */
177 elf_info
= (elf_addr_t
*)current
->mm
->saved_auxv
;
178 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
179 #define NEW_AUX_ENT(id, val) \
181 elf_info[ei_index++] = id; \
182 elf_info[ei_index++] = val; \
187 * ARCH_DLINFO must come first so PPC can do its special alignment of
189 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
190 * ARCH_DLINFO changes
194 NEW_AUX_ENT(AT_HWCAP
, ELF_HWCAP
);
195 NEW_AUX_ENT(AT_PAGESZ
, ELF_EXEC_PAGESIZE
);
196 NEW_AUX_ENT(AT_CLKTCK
, CLOCKS_PER_SEC
);
197 NEW_AUX_ENT(AT_PHDR
, load_addr
+ exec
->e_phoff
);
198 NEW_AUX_ENT(AT_PHENT
, sizeof(struct elf_phdr
));
199 NEW_AUX_ENT(AT_PHNUM
, exec
->e_phnum
);
200 NEW_AUX_ENT(AT_BASE
, interp_load_addr
);
201 NEW_AUX_ENT(AT_FLAGS
, 0);
202 NEW_AUX_ENT(AT_ENTRY
, exec
->e_entry
);
203 NEW_AUX_ENT(AT_UID
, tsk
->uid
);
204 NEW_AUX_ENT(AT_EUID
, tsk
->euid
);
205 NEW_AUX_ENT(AT_GID
, tsk
->gid
);
206 NEW_AUX_ENT(AT_EGID
, tsk
->egid
);
207 NEW_AUX_ENT(AT_SECURE
, security_bprm_secureexec(bprm
));
209 NEW_AUX_ENT(AT_PLATFORM
,
210 (elf_addr_t
)(unsigned long)u_platform
);
212 if (bprm
->interp_flags
& BINPRM_FLAGS_EXECFD
) {
213 NEW_AUX_ENT(AT_EXECFD
, bprm
->interp_data
);
216 /* AT_NULL is zero; clear the rest too */
217 memset(&elf_info
[ei_index
], 0,
218 sizeof current
->mm
->saved_auxv
- ei_index
* sizeof elf_info
[0]);
220 /* And advance past the AT_NULL entry. */
223 sp
= STACK_ADD(p
, ei_index
);
225 items
= (argc
+ 1) + (envc
+ 1);
227 items
+= 3; /* a.out interpreters require argv & envp too */
229 items
+= 1; /* ELF interpreters only put argc on the stack */
231 bprm
->p
= STACK_ROUND(sp
, items
);
233 /* Point sp at the lowest address on the stack */
234 #ifdef CONFIG_STACK_GROWSUP
235 sp
= (elf_addr_t __user
*)bprm
->p
- items
- ei_index
;
236 bprm
->exec
= (unsigned long)sp
; /* XXX: PARISC HACK */
238 sp
= (elf_addr_t __user
*)bprm
->p
;
243 * Grow the stack manually; some architectures have a limit on how
244 * far ahead a user-space access may be in order to grow the stack.
246 vma
= find_extend_vma(current
->mm
, bprm
->p
);
250 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
251 if (__put_user(argc
, sp
++))
255 envp
= argv
+ argc
+ 1;
256 if (__put_user((elf_addr_t
)(unsigned long)argv
, sp
++) ||
257 __put_user((elf_addr_t
)(unsigned long)envp
, sp
++))
261 envp
= argv
+ argc
+ 1;
264 /* Populate argv and envp */
265 p
= current
->mm
->arg_end
= current
->mm
->arg_start
;
268 if (__put_user((elf_addr_t
)p
, argv
++))
270 len
= strnlen_user((void __user
*)p
, MAX_ARG_STRLEN
);
271 if (!len
|| len
> MAX_ARG_STRLEN
)
275 if (__put_user(0, argv
))
277 current
->mm
->arg_end
= current
->mm
->env_start
= p
;
280 if (__put_user((elf_addr_t
)p
, envp
++))
282 len
= strnlen_user((void __user
*)p
, MAX_ARG_STRLEN
);
283 if (!len
|| len
> MAX_ARG_STRLEN
)
287 if (__put_user(0, envp
))
289 current
->mm
->env_end
= p
;
291 /* Put the elf_info on the stack in the right place. */
292 sp
= (elf_addr_t __user
*)envp
+ 1;
293 if (copy_to_user(sp
, elf_info
, ei_index
* sizeof(elf_addr_t
)))
300 static unsigned long elf_map(struct file
*filep
, unsigned long addr
,
301 struct elf_phdr
*eppnt
, int prot
, int type
,
302 unsigned long total_size
)
304 unsigned long map_addr
;
305 unsigned long size
= eppnt
->p_filesz
+ ELF_PAGEOFFSET(eppnt
->p_vaddr
);
306 unsigned long off
= eppnt
->p_offset
- ELF_PAGEOFFSET(eppnt
->p_vaddr
);
307 addr
= ELF_PAGESTART(addr
);
308 size
= ELF_PAGEALIGN(size
);
310 /* mmap() will return -EINVAL if given a zero size, but a
311 * segment with zero filesize is perfectly valid */
315 down_write(¤t
->mm
->mmap_sem
);
317 * total_size is the size of the ELF (interpreter) image.
318 * The _first_ mmap needs to know the full size, otherwise
319 * randomization might put this image into an overlapping
320 * position with the ELF binary image. (since size < total_size)
321 * So we first map the 'big' image - and unmap the remainder at
322 * the end. (which unmap is needed for ELF images with holes.)
325 total_size
= ELF_PAGEALIGN(total_size
);
326 map_addr
= do_mmap(filep
, addr
, total_size
, prot
, type
, off
);
327 if (!BAD_ADDR(map_addr
))
328 do_munmap(current
->mm
, map_addr
+size
, total_size
-size
);
330 map_addr
= do_mmap(filep
, addr
, size
, prot
, type
, off
);
332 up_write(¤t
->mm
->mmap_sem
);
336 #endif /* !elf_map */
338 static unsigned long total_mapping_size(struct elf_phdr
*cmds
, int nr
)
340 int i
, first_idx
= -1, last_idx
= -1;
342 for (i
= 0; i
< nr
; i
++) {
343 if (cmds
[i
].p_type
== PT_LOAD
) {
352 return cmds
[last_idx
].p_vaddr
+ cmds
[last_idx
].p_memsz
-
353 ELF_PAGESTART(cmds
[first_idx
].p_vaddr
);
357 /* This is much more generalized than the library routine read function,
358 so we keep this separate. Technically the library read function
359 is only provided so that we can read a.out libraries that have
362 static unsigned long load_elf_interp(struct elfhdr
*interp_elf_ex
,
363 struct file
*interpreter
, unsigned long *interp_map_addr
,
364 unsigned long no_base
)
366 struct elf_phdr
*elf_phdata
;
367 struct elf_phdr
*eppnt
;
368 unsigned long load_addr
= 0;
369 int load_addr_set
= 0;
370 unsigned long last_bss
= 0, elf_bss
= 0;
371 unsigned long error
= ~0UL;
372 unsigned long total_size
;
375 /* First of all, some simple consistency checks */
376 if (interp_elf_ex
->e_type
!= ET_EXEC
&&
377 interp_elf_ex
->e_type
!= ET_DYN
)
379 if (!elf_check_arch(interp_elf_ex
))
381 if (!interpreter
->f_op
|| !interpreter
->f_op
->mmap
)
385 * If the size of this structure has changed, then punt, since
386 * we will be doing the wrong thing.
388 if (interp_elf_ex
->e_phentsize
!= sizeof(struct elf_phdr
))
390 if (interp_elf_ex
->e_phnum
< 1 ||
391 interp_elf_ex
->e_phnum
> 65536U / sizeof(struct elf_phdr
))
394 /* Now read in all of the header information */
395 size
= sizeof(struct elf_phdr
) * interp_elf_ex
->e_phnum
;
396 if (size
> ELF_MIN_ALIGN
)
398 elf_phdata
= kmalloc(size
, GFP_KERNEL
);
402 retval
= kernel_read(interpreter
, interp_elf_ex
->e_phoff
,
403 (char *)elf_phdata
,size
);
405 if (retval
!= size
) {
411 total_size
= total_mapping_size(elf_phdata
, interp_elf_ex
->e_phnum
);
418 for (i
= 0; i
< interp_elf_ex
->e_phnum
; i
++, eppnt
++) {
419 if (eppnt
->p_type
== PT_LOAD
) {
420 int elf_type
= MAP_PRIVATE
| MAP_DENYWRITE
;
422 unsigned long vaddr
= 0;
423 unsigned long k
, map_addr
;
425 if (eppnt
->p_flags
& PF_R
)
426 elf_prot
= PROT_READ
;
427 if (eppnt
->p_flags
& PF_W
)
428 elf_prot
|= PROT_WRITE
;
429 if (eppnt
->p_flags
& PF_X
)
430 elf_prot
|= PROT_EXEC
;
431 vaddr
= eppnt
->p_vaddr
;
432 if (interp_elf_ex
->e_type
== ET_EXEC
|| load_addr_set
)
433 elf_type
|= MAP_FIXED
;
434 else if (no_base
&& interp_elf_ex
->e_type
== ET_DYN
)
437 map_addr
= elf_map(interpreter
, load_addr
+ vaddr
,
438 eppnt
, elf_prot
, elf_type
, total_size
);
440 if (!*interp_map_addr
)
441 *interp_map_addr
= map_addr
;
443 if (BAD_ADDR(map_addr
))
446 if (!load_addr_set
&&
447 interp_elf_ex
->e_type
== ET_DYN
) {
448 load_addr
= map_addr
- ELF_PAGESTART(vaddr
);
453 * Check to see if the section's size will overflow the
454 * allowed task size. Note that p_filesz must always be
455 * <= p_memsize so it's only necessary to check p_memsz.
457 k
= load_addr
+ eppnt
->p_vaddr
;
459 eppnt
->p_filesz
> eppnt
->p_memsz
||
460 eppnt
->p_memsz
> TASK_SIZE
||
461 TASK_SIZE
- eppnt
->p_memsz
< k
) {
467 * Find the end of the file mapping for this phdr, and
468 * keep track of the largest address we see for this.
470 k
= load_addr
+ eppnt
->p_vaddr
+ eppnt
->p_filesz
;
475 * Do the same thing for the memory mapping - between
476 * elf_bss and last_bss is the bss section.
478 k
= load_addr
+ eppnt
->p_memsz
+ eppnt
->p_vaddr
;
485 * Now fill out the bss section. First pad the last page up
486 * to the page boundary, and then perform a mmap to make sure
487 * that there are zero-mapped pages up to and including the
490 if (padzero(elf_bss
)) {
495 /* What we have mapped so far */
496 elf_bss
= ELF_PAGESTART(elf_bss
+ ELF_MIN_ALIGN
- 1);
498 /* Map the last of the bss segment */
499 if (last_bss
> elf_bss
) {
500 down_write(¤t
->mm
->mmap_sem
);
501 error
= do_brk(elf_bss
, last_bss
- elf_bss
);
502 up_write(¤t
->mm
->mmap_sem
);
515 static unsigned long load_aout_interp(struct exec
*interp_ex
,
516 struct file
*interpreter
)
518 unsigned long text_data
, elf_entry
= ~0UL;
522 current
->mm
->end_code
= interp_ex
->a_text
;
523 text_data
= interp_ex
->a_text
+ interp_ex
->a_data
;
524 current
->mm
->end_data
= text_data
;
525 current
->mm
->brk
= interp_ex
->a_bss
+ text_data
;
527 switch (N_MAGIC(*interp_ex
)) {
530 addr
= (char __user
*)0;
534 offset
= N_TXTOFF(*interp_ex
);
535 addr
= (char __user
*)N_TXTADDR(*interp_ex
);
541 down_write(¤t
->mm
->mmap_sem
);
542 do_brk(0, text_data
);
543 up_write(¤t
->mm
->mmap_sem
);
544 if (!interpreter
->f_op
|| !interpreter
->f_op
->read
)
546 if (interpreter
->f_op
->read(interpreter
, addr
, text_data
, &offset
) < 0)
548 flush_icache_range((unsigned long)addr
,
549 (unsigned long)addr
+ text_data
);
551 down_write(¤t
->mm
->mmap_sem
);
552 do_brk(ELF_PAGESTART(text_data
+ ELF_MIN_ALIGN
- 1),
554 up_write(¤t
->mm
->mmap_sem
);
555 elf_entry
= interp_ex
->a_entry
;
562 * These are the functions used to load ELF style executables and shared
563 * libraries. There is no binary dependent code anywhere else.
566 #define INTERPRETER_NONE 0
567 #define INTERPRETER_AOUT 1
568 #define INTERPRETER_ELF 2
570 #ifndef STACK_RND_MASK
571 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
574 static unsigned long randomize_stack_top(unsigned long stack_top
)
576 unsigned int random_variable
= 0;
578 if ((current
->flags
& PF_RANDOMIZE
) &&
579 !(current
->personality
& ADDR_NO_RANDOMIZE
)) {
580 random_variable
= get_random_int() & STACK_RND_MASK
;
581 random_variable
<<= PAGE_SHIFT
;
583 #ifdef CONFIG_STACK_GROWSUP
584 return PAGE_ALIGN(stack_top
) + random_variable
;
586 return PAGE_ALIGN(stack_top
) - random_variable
;
590 static int load_elf_binary(struct linux_binprm
*bprm
, struct pt_regs
*regs
)
592 struct file
*interpreter
= NULL
; /* to shut gcc up */
593 unsigned long load_addr
= 0, load_bias
= 0;
594 int load_addr_set
= 0;
595 char * elf_interpreter
= NULL
;
596 unsigned int interpreter_type
= INTERPRETER_NONE
;
597 unsigned char ibcs2_interpreter
= 0;
599 struct elf_phdr
*elf_ppnt
, *elf_phdata
;
600 unsigned long elf_bss
, elf_brk
;
604 unsigned long elf_entry
;
605 unsigned long interp_load_addr
= 0;
606 unsigned long start_code
, end_code
, start_data
, end_data
;
607 unsigned long reloc_func_desc
= 0;
608 char passed_fileno
[6];
609 struct files_struct
*files
;
610 int executable_stack
= EXSTACK_DEFAULT
;
611 unsigned long def_flags
= 0;
613 struct elfhdr elf_ex
;
614 struct elfhdr interp_elf_ex
;
615 struct exec interp_ex
;
618 loc
= kmalloc(sizeof(*loc
), GFP_KERNEL
);
624 /* Get the exec-header */
625 loc
->elf_ex
= *((struct elfhdr
*)bprm
->buf
);
628 /* First of all, some simple consistency checks */
629 if (memcmp(loc
->elf_ex
.e_ident
, ELFMAG
, SELFMAG
) != 0)
632 if (loc
->elf_ex
.e_type
!= ET_EXEC
&& loc
->elf_ex
.e_type
!= ET_DYN
)
634 if (!elf_check_arch(&loc
->elf_ex
))
636 if (!bprm
->file
->f_op
||!bprm
->file
->f_op
->mmap
)
639 /* Now read in all of the header information */
640 if (loc
->elf_ex
.e_phentsize
!= sizeof(struct elf_phdr
))
642 if (loc
->elf_ex
.e_phnum
< 1 ||
643 loc
->elf_ex
.e_phnum
> 65536U / sizeof(struct elf_phdr
))
645 size
= loc
->elf_ex
.e_phnum
* sizeof(struct elf_phdr
);
647 elf_phdata
= kmalloc(size
, GFP_KERNEL
);
651 retval
= kernel_read(bprm
->file
, loc
->elf_ex
.e_phoff
,
652 (char *)elf_phdata
, size
);
653 if (retval
!= size
) {
659 files
= current
->files
; /* Refcounted so ok */
660 retval
= unshare_files();
663 if (files
== current
->files
) {
664 put_files_struct(files
);
668 /* exec will make our files private anyway, but for the a.out
669 loader stuff we need to do it earlier */
670 retval
= get_unused_fd();
673 get_file(bprm
->file
);
674 fd_install(elf_exec_fileno
= retval
, bprm
->file
);
676 elf_ppnt
= elf_phdata
;
685 for (i
= 0; i
< loc
->elf_ex
.e_phnum
; i
++) {
686 if (elf_ppnt
->p_type
== PT_INTERP
) {
687 /* This is the program interpreter used for
688 * shared libraries - for now assume that this
689 * is an a.out format binary
692 if (elf_ppnt
->p_filesz
> PATH_MAX
||
693 elf_ppnt
->p_filesz
< 2)
697 elf_interpreter
= kmalloc(elf_ppnt
->p_filesz
,
699 if (!elf_interpreter
)
702 retval
= kernel_read(bprm
->file
, elf_ppnt
->p_offset
,
705 if (retval
!= elf_ppnt
->p_filesz
) {
708 goto out_free_interp
;
710 /* make sure path is NULL terminated */
712 if (elf_interpreter
[elf_ppnt
->p_filesz
- 1] != '\0')
713 goto out_free_interp
;
715 /* If the program interpreter is one of these two,
716 * then assume an iBCS2 image. Otherwise assume
717 * a native linux image.
719 if (strcmp(elf_interpreter
,"/usr/lib/libc.so.1") == 0 ||
720 strcmp(elf_interpreter
,"/usr/lib/ld.so.1") == 0)
721 ibcs2_interpreter
= 1;
724 * The early SET_PERSONALITY here is so that the lookup
725 * for the interpreter happens in the namespace of the
726 * to-be-execed image. SET_PERSONALITY can select an
729 * However, SET_PERSONALITY is NOT allowed to switch
730 * this task into the new images's memory mapping
731 * policy - that is, TASK_SIZE must still evaluate to
732 * that which is appropriate to the execing application.
733 * This is because exit_mmap() needs to have TASK_SIZE
734 * evaluate to the size of the old image.
736 * So if (say) a 64-bit application is execing a 32-bit
737 * application it is the architecture's responsibility
738 * to defer changing the value of TASK_SIZE until the
739 * switch really is going to happen - do this in
740 * flush_thread(). - akpm
742 SET_PERSONALITY(loc
->elf_ex
, ibcs2_interpreter
);
744 interpreter
= open_exec(elf_interpreter
);
745 retval
= PTR_ERR(interpreter
);
746 if (IS_ERR(interpreter
))
747 goto out_free_interp
;
750 * If the binary is not readable then enforce
751 * mm->dumpable = 0 regardless of the interpreter's
754 if (file_permission(interpreter
, MAY_READ
) < 0)
755 bprm
->interp_flags
|= BINPRM_FLAGS_ENFORCE_NONDUMP
;
757 retval
= kernel_read(interpreter
, 0, bprm
->buf
,
759 if (retval
!= BINPRM_BUF_SIZE
) {
762 goto out_free_dentry
;
765 /* Get the exec headers */
766 loc
->interp_ex
= *((struct exec
*)bprm
->buf
);
767 loc
->interp_elf_ex
= *((struct elfhdr
*)bprm
->buf
);
773 elf_ppnt
= elf_phdata
;
774 for (i
= 0; i
< loc
->elf_ex
.e_phnum
; i
++, elf_ppnt
++)
775 if (elf_ppnt
->p_type
== PT_GNU_STACK
) {
776 if (elf_ppnt
->p_flags
& PF_X
)
777 executable_stack
= EXSTACK_ENABLE_X
;
779 executable_stack
= EXSTACK_DISABLE_X
;
783 /* Some simple consistency checks for the interpreter */
784 if (elf_interpreter
) {
786 interpreter_type
= INTERPRETER_ELF
| INTERPRETER_AOUT
;
788 /* Now figure out which format our binary is */
789 if ((N_MAGIC(loc
->interp_ex
) != OMAGIC
) &&
790 (N_MAGIC(loc
->interp_ex
) != ZMAGIC
) &&
791 (N_MAGIC(loc
->interp_ex
) != QMAGIC
))
792 interpreter_type
= INTERPRETER_ELF
;
794 if (memcmp(loc
->interp_elf_ex
.e_ident
, ELFMAG
, SELFMAG
) != 0)
795 interpreter_type
&= ~INTERPRETER_ELF
;
797 if (interpreter_type
== INTERPRETER_AOUT
&& warn
< 10) {
798 printk(KERN_WARNING
"a.out ELF interpreter %s is "
799 "deprecated and will not be supported "
800 "after Linux 2.6.25\n", elf_interpreter
);
805 if (!interpreter_type
)
806 goto out_free_dentry
;
808 /* Make sure only one type was selected */
809 if ((interpreter_type
& INTERPRETER_ELF
) &&
810 interpreter_type
!= INTERPRETER_ELF
) {
811 // FIXME - ratelimit this before re-enabling
812 // printk(KERN_WARNING "ELF: Ambiguous type, using ELF\n");
813 interpreter_type
= INTERPRETER_ELF
;
815 /* Verify the interpreter has a valid arch */
816 if ((interpreter_type
== INTERPRETER_ELF
) &&
817 !elf_check_arch(&loc
->interp_elf_ex
))
818 goto out_free_dentry
;
820 /* Executables without an interpreter also need a personality */
821 SET_PERSONALITY(loc
->elf_ex
, ibcs2_interpreter
);
824 /* OK, we are done with that, now set up the arg stuff,
825 and then start this sucker up */
826 if ((!bprm
->sh_bang
) && (interpreter_type
== INTERPRETER_AOUT
)) {
827 char *passed_p
= passed_fileno
;
828 sprintf(passed_fileno
, "%d", elf_exec_fileno
);
830 if (elf_interpreter
) {
831 retval
= copy_strings_kernel(1, &passed_p
, bprm
);
833 goto out_free_dentry
;
838 /* Flush all traces of the currently running executable */
839 retval
= flush_old_exec(bprm
);
841 goto out_free_dentry
;
843 /* Discard our unneeded old files struct */
845 put_files_struct(files
);
849 /* OK, This is the point of no return */
850 current
->flags
&= ~PF_FORKNOEXEC
;
851 current
->mm
->def_flags
= def_flags
;
853 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
854 may depend on the personality. */
855 SET_PERSONALITY(loc
->elf_ex
, ibcs2_interpreter
);
856 if (elf_read_implies_exec(loc
->elf_ex
, executable_stack
))
857 current
->personality
|= READ_IMPLIES_EXEC
;
859 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
860 current
->flags
|= PF_RANDOMIZE
;
861 arch_pick_mmap_layout(current
->mm
);
863 /* Do this so that we can load the interpreter, if need be. We will
864 change some of these later */
865 current
->mm
->free_area_cache
= current
->mm
->mmap_base
;
866 current
->mm
->cached_hole_size
= 0;
867 retval
= setup_arg_pages(bprm
, randomize_stack_top(STACK_TOP
),
870 send_sig(SIGKILL
, current
, 0);
871 goto out_free_dentry
;
874 current
->mm
->start_stack
= bprm
->p
;
876 /* Now we do a little grungy work by mmaping the ELF image into
877 the correct location in memory. */
878 for(i
= 0, elf_ppnt
= elf_phdata
;
879 i
< loc
->elf_ex
.e_phnum
; i
++, elf_ppnt
++) {
880 int elf_prot
= 0, elf_flags
;
881 unsigned long k
, vaddr
;
883 if (elf_ppnt
->p_type
!= PT_LOAD
)
886 if (unlikely (elf_brk
> elf_bss
)) {
889 /* There was a PT_LOAD segment with p_memsz > p_filesz
890 before this one. Map anonymous pages, if needed,
891 and clear the area. */
892 retval
= set_brk (elf_bss
+ load_bias
,
893 elf_brk
+ load_bias
);
895 send_sig(SIGKILL
, current
, 0);
896 goto out_free_dentry
;
898 nbyte
= ELF_PAGEOFFSET(elf_bss
);
900 nbyte
= ELF_MIN_ALIGN
- nbyte
;
901 if (nbyte
> elf_brk
- elf_bss
)
902 nbyte
= elf_brk
- elf_bss
;
903 if (clear_user((void __user
*)elf_bss
+
906 * This bss-zeroing can fail if the ELF
907 * file specifies odd protections. So
908 * we don't check the return value
914 if (elf_ppnt
->p_flags
& PF_R
)
915 elf_prot
|= PROT_READ
;
916 if (elf_ppnt
->p_flags
& PF_W
)
917 elf_prot
|= PROT_WRITE
;
918 if (elf_ppnt
->p_flags
& PF_X
)
919 elf_prot
|= PROT_EXEC
;
921 elf_flags
= MAP_PRIVATE
| MAP_DENYWRITE
| MAP_EXECUTABLE
;
923 vaddr
= elf_ppnt
->p_vaddr
;
924 if (loc
->elf_ex
.e_type
== ET_EXEC
|| load_addr_set
) {
925 elf_flags
|= MAP_FIXED
;
926 } else if (loc
->elf_ex
.e_type
== ET_DYN
) {
927 /* Try and get dynamic programs out of the way of the
928 * default mmap base, as well as whatever program they
929 * might try to exec. This is because the brk will
930 * follow the loader, and is not movable. */
934 load_bias
= ELF_PAGESTART(ELF_ET_DYN_BASE
- vaddr
);
938 error
= elf_map(bprm
->file
, load_bias
+ vaddr
, elf_ppnt
,
939 elf_prot
, elf_flags
,0);
940 if (BAD_ADDR(error
)) {
941 send_sig(SIGKILL
, current
, 0);
942 retval
= IS_ERR((void *)error
) ?
943 PTR_ERR((void*)error
) : -EINVAL
;
944 goto out_free_dentry
;
947 if (!load_addr_set
) {
949 load_addr
= (elf_ppnt
->p_vaddr
- elf_ppnt
->p_offset
);
950 if (loc
->elf_ex
.e_type
== ET_DYN
) {
952 ELF_PAGESTART(load_bias
+ vaddr
);
953 load_addr
+= load_bias
;
954 reloc_func_desc
= load_bias
;
957 k
= elf_ppnt
->p_vaddr
;
964 * Check to see if the section's size will overflow the
965 * allowed task size. Note that p_filesz must always be
966 * <= p_memsz so it is only necessary to check p_memsz.
968 if (BAD_ADDR(k
) || elf_ppnt
->p_filesz
> elf_ppnt
->p_memsz
||
969 elf_ppnt
->p_memsz
> TASK_SIZE
||
970 TASK_SIZE
- elf_ppnt
->p_memsz
< k
) {
971 /* set_brk can never work. Avoid overflows. */
972 send_sig(SIGKILL
, current
, 0);
974 goto out_free_dentry
;
977 k
= elf_ppnt
->p_vaddr
+ elf_ppnt
->p_filesz
;
981 if ((elf_ppnt
->p_flags
& PF_X
) && end_code
< k
)
985 k
= elf_ppnt
->p_vaddr
+ elf_ppnt
->p_memsz
;
990 loc
->elf_ex
.e_entry
+= load_bias
;
991 elf_bss
+= load_bias
;
992 elf_brk
+= load_bias
;
993 start_code
+= load_bias
;
994 end_code
+= load_bias
;
995 start_data
+= load_bias
;
996 end_data
+= load_bias
;
998 /* Calling set_brk effectively mmaps the pages that we need
999 * for the bss and break sections. We must do this before
1000 * mapping in the interpreter, to make sure it doesn't wind
1001 * up getting placed where the bss needs to go.
1003 retval
= set_brk(elf_bss
, elf_brk
);
1005 send_sig(SIGKILL
, current
, 0);
1006 goto out_free_dentry
;
1008 if (likely(elf_bss
!= elf_brk
) && unlikely(padzero(elf_bss
))) {
1009 send_sig(SIGSEGV
, current
, 0);
1010 retval
= -EFAULT
; /* Nobody gets to see this, but.. */
1011 goto out_free_dentry
;
1014 if (elf_interpreter
) {
1015 if (interpreter_type
== INTERPRETER_AOUT
) {
1016 elf_entry
= load_aout_interp(&loc
->interp_ex
,
1019 unsigned long uninitialized_var(interp_map_addr
);
1021 elf_entry
= load_elf_interp(&loc
->interp_elf_ex
,
1025 if (!IS_ERR((void *)elf_entry
)) {
1027 * load_elf_interp() returns relocation
1030 interp_load_addr
= elf_entry
;
1031 elf_entry
+= loc
->interp_elf_ex
.e_entry
;
1034 if (BAD_ADDR(elf_entry
)) {
1035 force_sig(SIGSEGV
, current
);
1036 retval
= IS_ERR((void *)elf_entry
) ?
1037 (int)elf_entry
: -EINVAL
;
1038 goto out_free_dentry
;
1040 reloc_func_desc
= interp_load_addr
;
1042 allow_write_access(interpreter
);
1044 kfree(elf_interpreter
);
1046 elf_entry
= loc
->elf_ex
.e_entry
;
1047 if (BAD_ADDR(elf_entry
)) {
1048 force_sig(SIGSEGV
, current
);
1050 goto out_free_dentry
;
1056 if (interpreter_type
!= INTERPRETER_AOUT
)
1057 sys_close(elf_exec_fileno
);
1059 set_binfmt(&elf_format
);
1061 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
1062 retval
= arch_setup_additional_pages(bprm
, executable_stack
);
1064 send_sig(SIGKILL
, current
, 0);
1067 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1069 compute_creds(bprm
);
1070 current
->flags
&= ~PF_FORKNOEXEC
;
1071 retval
= create_elf_tables(bprm
, &loc
->elf_ex
,
1072 (interpreter_type
== INTERPRETER_AOUT
),
1073 load_addr
, interp_load_addr
);
1075 send_sig(SIGKILL
, current
, 0);
1078 /* N.B. passed_fileno might not be initialized? */
1079 if (interpreter_type
== INTERPRETER_AOUT
)
1080 current
->mm
->arg_start
+= strlen(passed_fileno
) + 1;
1081 current
->mm
->end_code
= end_code
;
1082 current
->mm
->start_code
= start_code
;
1083 current
->mm
->start_data
= start_data
;
1084 current
->mm
->end_data
= end_data
;
1085 current
->mm
->start_stack
= bprm
->p
;
1087 #ifdef arch_randomize_brk
1088 if (current
->flags
& PF_RANDOMIZE
)
1089 current
->mm
->brk
= current
->mm
->start_brk
=
1090 arch_randomize_brk(current
->mm
);
1093 if (current
->personality
& MMAP_PAGE_ZERO
) {
1094 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1095 and some applications "depend" upon this behavior.
1096 Since we do not have the power to recompile these, we
1097 emulate the SVr4 behavior. Sigh. */
1098 down_write(¤t
->mm
->mmap_sem
);
1099 error
= do_mmap(NULL
, 0, PAGE_SIZE
, PROT_READ
| PROT_EXEC
,
1100 MAP_FIXED
| MAP_PRIVATE
, 0);
1101 up_write(¤t
->mm
->mmap_sem
);
1104 #ifdef ELF_PLAT_INIT
1106 * The ABI may specify that certain registers be set up in special
1107 * ways (on i386 %edx is the address of a DT_FINI function, for
1108 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1109 * that the e_entry field is the address of the function descriptor
1110 * for the startup routine, rather than the address of the startup
1111 * routine itself. This macro performs whatever initialization to
1112 * the regs structure is required as well as any relocations to the
1113 * function descriptor entries when executing dynamically links apps.
1115 ELF_PLAT_INIT(regs
, reloc_func_desc
);
1118 start_thread(regs
, elf_entry
, bprm
->p
);
1119 if (unlikely(current
->ptrace
& PT_PTRACED
)) {
1120 if (current
->ptrace
& PT_TRACE_EXEC
)
1121 ptrace_notify ((PTRACE_EVENT_EXEC
<< 8) | SIGTRAP
);
1123 send_sig(SIGTRAP
, current
, 0);
1133 allow_write_access(interpreter
);
1137 kfree(elf_interpreter
);
1139 sys_close(elf_exec_fileno
);
1142 reset_files_struct(current
, files
);
1148 /* This is really simpleminded and specialized - we are loading an
1149 a.out library that is given an ELF header. */
1150 static int load_elf_library(struct file
*file
)
1152 struct elf_phdr
*elf_phdata
;
1153 struct elf_phdr
*eppnt
;
1154 unsigned long elf_bss
, bss
, len
;
1155 int retval
, error
, i
, j
;
1156 struct elfhdr elf_ex
;
1159 retval
= kernel_read(file
, 0, (char *)&elf_ex
, sizeof(elf_ex
));
1160 if (retval
!= sizeof(elf_ex
))
1163 if (memcmp(elf_ex
.e_ident
, ELFMAG
, SELFMAG
) != 0)
1166 /* First of all, some simple consistency checks */
1167 if (elf_ex
.e_type
!= ET_EXEC
|| elf_ex
.e_phnum
> 2 ||
1168 !elf_check_arch(&elf_ex
) || !file
->f_op
|| !file
->f_op
->mmap
)
1171 /* Now read in all of the header information */
1173 j
= sizeof(struct elf_phdr
) * elf_ex
.e_phnum
;
1174 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1177 elf_phdata
= kmalloc(j
, GFP_KERNEL
);
1183 retval
= kernel_read(file
, elf_ex
.e_phoff
, (char *)eppnt
, j
);
1187 for (j
= 0, i
= 0; i
<elf_ex
.e_phnum
; i
++)
1188 if ((eppnt
+ i
)->p_type
== PT_LOAD
)
1193 while (eppnt
->p_type
!= PT_LOAD
)
1196 /* Now use mmap to map the library into memory. */
1197 down_write(¤t
->mm
->mmap_sem
);
1198 error
= do_mmap(file
,
1199 ELF_PAGESTART(eppnt
->p_vaddr
),
1201 ELF_PAGEOFFSET(eppnt
->p_vaddr
)),
1202 PROT_READ
| PROT_WRITE
| PROT_EXEC
,
1203 MAP_FIXED
| MAP_PRIVATE
| MAP_DENYWRITE
,
1205 ELF_PAGEOFFSET(eppnt
->p_vaddr
)));
1206 up_write(¤t
->mm
->mmap_sem
);
1207 if (error
!= ELF_PAGESTART(eppnt
->p_vaddr
))
1210 elf_bss
= eppnt
->p_vaddr
+ eppnt
->p_filesz
;
1211 if (padzero(elf_bss
)) {
1216 len
= ELF_PAGESTART(eppnt
->p_filesz
+ eppnt
->p_vaddr
+
1218 bss
= eppnt
->p_memsz
+ eppnt
->p_vaddr
;
1220 down_write(¤t
->mm
->mmap_sem
);
1221 do_brk(len
, bss
- len
);
1222 up_write(¤t
->mm
->mmap_sem
);
1233 * Note that some platforms still use traditional core dumps and not
1234 * the ELF core dump. Each platform can select it as appropriate.
1236 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1241 * Modelled on fs/exec.c:aout_core_dump()
1242 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1245 * These are the only things you should do on a core-file: use only these
1246 * functions to write out all the necessary info.
1248 static int dump_write(struct file
*file
, const void *addr
, int nr
)
1250 return file
->f_op
->write(file
, addr
, nr
, &file
->f_pos
) == nr
;
1253 static int dump_seek(struct file
*file
, loff_t off
)
1255 if (file
->f_op
->llseek
&& file
->f_op
->llseek
!= no_llseek
) {
1256 if (file
->f_op
->llseek(file
, off
, SEEK_CUR
) < 0)
1259 char *buf
= (char *)get_zeroed_page(GFP_KERNEL
);
1263 unsigned long n
= off
;
1266 if (!dump_write(file
, buf
, n
))
1270 free_page((unsigned long)buf
);
1276 * Decide what to dump of a segment, part, all or none.
1278 static unsigned long vma_dump_size(struct vm_area_struct
*vma
,
1279 unsigned long mm_flags
)
1281 /* The vma can be set up to tell us the answer directly. */
1282 if (vma
->vm_flags
& VM_ALWAYSDUMP
)
1285 /* Do not dump I/O mapped devices or special mappings */
1286 if (vma
->vm_flags
& (VM_IO
| VM_RESERVED
))
1289 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1291 /* By default, dump shared memory if mapped from an anonymous file. */
1292 if (vma
->vm_flags
& VM_SHARED
) {
1293 if (vma
->vm_file
->f_path
.dentry
->d_inode
->i_nlink
== 0 ?
1294 FILTER(ANON_SHARED
) : FILTER(MAPPED_SHARED
))
1299 /* Dump segments that have been written to. */
1300 if (vma
->anon_vma
&& FILTER(ANON_PRIVATE
))
1302 if (vma
->vm_file
== NULL
)
1305 if (FILTER(MAPPED_PRIVATE
))
1309 * If this looks like the beginning of a DSO or executable mapping,
1310 * check for an ELF header. If we find one, dump the first page to
1311 * aid in determining what was mapped here.
1313 if (FILTER(ELF_HEADERS
) && vma
->vm_file
!= NULL
&& vma
->vm_pgoff
== 0) {
1314 u32 __user
*header
= (u32 __user
*) vma
->vm_start
;
1317 * Doing it this way gets the constant folded by GCC.
1321 char elfmag
[SELFMAG
];
1323 BUILD_BUG_ON(SELFMAG
!= sizeof word
);
1324 magic
.elfmag
[EI_MAG0
] = ELFMAG0
;
1325 magic
.elfmag
[EI_MAG1
] = ELFMAG1
;
1326 magic
.elfmag
[EI_MAG2
] = ELFMAG2
;
1327 magic
.elfmag
[EI_MAG3
] = ELFMAG3
;
1328 if (get_user(word
, header
) == 0 && word
== magic
.cmp
)
1337 return vma
->vm_end
- vma
->vm_start
;
1340 /* An ELF note in memory */
1345 unsigned int datasz
;
1349 static int notesize(struct memelfnote
*en
)
1353 sz
= sizeof(struct elf_note
);
1354 sz
+= roundup(strlen(en
->name
) + 1, 4);
1355 sz
+= roundup(en
->datasz
, 4);
1360 #define DUMP_WRITE(addr, nr, foffset) \
1361 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1363 static int alignfile(struct file
*file
, loff_t
*foffset
)
1365 static const char buf
[4] = { 0, };
1366 DUMP_WRITE(buf
, roundup(*foffset
, 4) - *foffset
, foffset
);
1370 static int writenote(struct memelfnote
*men
, struct file
*file
,
1374 en
.n_namesz
= strlen(men
->name
) + 1;
1375 en
.n_descsz
= men
->datasz
;
1376 en
.n_type
= men
->type
;
1378 DUMP_WRITE(&en
, sizeof(en
), foffset
);
1379 DUMP_WRITE(men
->name
, en
.n_namesz
, foffset
);
1380 if (!alignfile(file
, foffset
))
1382 DUMP_WRITE(men
->data
, men
->datasz
, foffset
);
1383 if (!alignfile(file
, foffset
))
1390 #define DUMP_WRITE(addr, nr) \
1391 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1393 #define DUMP_SEEK(off) \
1394 if (!dump_seek(file, (off))) \
1397 static void fill_elf_header(struct elfhdr
*elf
, int segs
)
1399 memcpy(elf
->e_ident
, ELFMAG
, SELFMAG
);
1400 elf
->e_ident
[EI_CLASS
] = ELF_CLASS
;
1401 elf
->e_ident
[EI_DATA
] = ELF_DATA
;
1402 elf
->e_ident
[EI_VERSION
] = EV_CURRENT
;
1403 elf
->e_ident
[EI_OSABI
] = ELF_OSABI
;
1404 memset(elf
->e_ident
+EI_PAD
, 0, EI_NIDENT
-EI_PAD
);
1406 elf
->e_type
= ET_CORE
;
1407 elf
->e_machine
= ELF_ARCH
;
1408 elf
->e_version
= EV_CURRENT
;
1410 elf
->e_phoff
= sizeof(struct elfhdr
);
1412 elf
->e_flags
= ELF_CORE_EFLAGS
;
1413 elf
->e_ehsize
= sizeof(struct elfhdr
);
1414 elf
->e_phentsize
= sizeof(struct elf_phdr
);
1415 elf
->e_phnum
= segs
;
1416 elf
->e_shentsize
= 0;
1418 elf
->e_shstrndx
= 0;
1422 static void fill_elf_note_phdr(struct elf_phdr
*phdr
, int sz
, loff_t offset
)
1424 phdr
->p_type
= PT_NOTE
;
1425 phdr
->p_offset
= offset
;
1428 phdr
->p_filesz
= sz
;
1435 static void fill_note(struct memelfnote
*note
, const char *name
, int type
,
1436 unsigned int sz
, void *data
)
1446 * fill up all the fields in prstatus from the given task struct, except
1447 * registers which need to be filled up separately.
1449 static void fill_prstatus(struct elf_prstatus
*prstatus
,
1450 struct task_struct
*p
, long signr
)
1452 prstatus
->pr_info
.si_signo
= prstatus
->pr_cursig
= signr
;
1453 prstatus
->pr_sigpend
= p
->pending
.signal
.sig
[0];
1454 prstatus
->pr_sighold
= p
->blocked
.sig
[0];
1455 prstatus
->pr_pid
= task_pid_vnr(p
);
1456 prstatus
->pr_ppid
= task_pid_vnr(p
->real_parent
);
1457 prstatus
->pr_pgrp
= task_pgrp_vnr(p
);
1458 prstatus
->pr_sid
= task_session_vnr(p
);
1459 if (thread_group_leader(p
)) {
1461 * This is the record for the group leader. Add in the
1462 * cumulative times of previous dead threads. This total
1463 * won't include the time of each live thread whose state
1464 * is included in the core dump. The final total reported
1465 * to our parent process when it calls wait4 will include
1466 * those sums as well as the little bit more time it takes
1467 * this and each other thread to finish dying after the
1468 * core dump synchronization phase.
1470 cputime_to_timeval(cputime_add(p
->utime
, p
->signal
->utime
),
1471 &prstatus
->pr_utime
);
1472 cputime_to_timeval(cputime_add(p
->stime
, p
->signal
->stime
),
1473 &prstatus
->pr_stime
);
1475 cputime_to_timeval(p
->utime
, &prstatus
->pr_utime
);
1476 cputime_to_timeval(p
->stime
, &prstatus
->pr_stime
);
1478 cputime_to_timeval(p
->signal
->cutime
, &prstatus
->pr_cutime
);
1479 cputime_to_timeval(p
->signal
->cstime
, &prstatus
->pr_cstime
);
1482 static int fill_psinfo(struct elf_prpsinfo
*psinfo
, struct task_struct
*p
,
1483 struct mm_struct
*mm
)
1485 unsigned int i
, len
;
1487 /* first copy the parameters from user space */
1488 memset(psinfo
, 0, sizeof(struct elf_prpsinfo
));
1490 len
= mm
->arg_end
- mm
->arg_start
;
1491 if (len
>= ELF_PRARGSZ
)
1492 len
= ELF_PRARGSZ
-1;
1493 if (copy_from_user(&psinfo
->pr_psargs
,
1494 (const char __user
*)mm
->arg_start
, len
))
1496 for(i
= 0; i
< len
; i
++)
1497 if (psinfo
->pr_psargs
[i
] == 0)
1498 psinfo
->pr_psargs
[i
] = ' ';
1499 psinfo
->pr_psargs
[len
] = 0;
1501 psinfo
->pr_pid
= task_pid_vnr(p
);
1502 psinfo
->pr_ppid
= task_pid_vnr(p
->real_parent
);
1503 psinfo
->pr_pgrp
= task_pgrp_vnr(p
);
1504 psinfo
->pr_sid
= task_session_vnr(p
);
1506 i
= p
->state
? ffz(~p
->state
) + 1 : 0;
1507 psinfo
->pr_state
= i
;
1508 psinfo
->pr_sname
= (i
> 5) ? '.' : "RSDTZW"[i
];
1509 psinfo
->pr_zomb
= psinfo
->pr_sname
== 'Z';
1510 psinfo
->pr_nice
= task_nice(p
);
1511 psinfo
->pr_flag
= p
->flags
;
1512 SET_UID(psinfo
->pr_uid
, p
->uid
);
1513 SET_GID(psinfo
->pr_gid
, p
->gid
);
1514 strncpy(psinfo
->pr_fname
, p
->comm
, sizeof(psinfo
->pr_fname
));
1519 /* Here is the structure in which status of each thread is captured. */
1520 struct elf_thread_status
1522 struct list_head list
;
1523 struct elf_prstatus prstatus
; /* NT_PRSTATUS */
1524 elf_fpregset_t fpu
; /* NT_PRFPREG */
1525 struct task_struct
*thread
;
1526 #ifdef ELF_CORE_COPY_XFPREGS
1527 elf_fpxregset_t xfpu
; /* ELF_CORE_XFPREG_TYPE */
1529 struct memelfnote notes
[3];
1534 * In order to add the specific thread information for the elf file format,
1535 * we need to keep a linked list of every threads pr_status and then create
1536 * a single section for them in the final core file.
1538 static int elf_dump_thread_status(long signr
, struct elf_thread_status
*t
)
1541 struct task_struct
*p
= t
->thread
;
1544 fill_prstatus(&t
->prstatus
, p
, signr
);
1545 elf_core_copy_task_regs(p
, &t
->prstatus
.pr_reg
);
1547 fill_note(&t
->notes
[0], "CORE", NT_PRSTATUS
, sizeof(t
->prstatus
),
1550 sz
+= notesize(&t
->notes
[0]);
1552 if ((t
->prstatus
.pr_fpvalid
= elf_core_copy_task_fpregs(p
, NULL
,
1554 fill_note(&t
->notes
[1], "CORE", NT_PRFPREG
, sizeof(t
->fpu
),
1557 sz
+= notesize(&t
->notes
[1]);
1560 #ifdef ELF_CORE_COPY_XFPREGS
1561 if (elf_core_copy_task_xfpregs(p
, &t
->xfpu
)) {
1562 fill_note(&t
->notes
[2], "LINUX", ELF_CORE_XFPREG_TYPE
,
1563 sizeof(t
->xfpu
), &t
->xfpu
);
1565 sz
+= notesize(&t
->notes
[2]);
1571 static struct vm_area_struct
*first_vma(struct task_struct
*tsk
,
1572 struct vm_area_struct
*gate_vma
)
1574 struct vm_area_struct
*ret
= tsk
->mm
->mmap
;
1581 * Helper function for iterating across a vma list. It ensures that the caller
1582 * will visit `gate_vma' prior to terminating the search.
1584 static struct vm_area_struct
*next_vma(struct vm_area_struct
*this_vma
,
1585 struct vm_area_struct
*gate_vma
)
1587 struct vm_area_struct
*ret
;
1589 ret
= this_vma
->vm_next
;
1592 if (this_vma
== gate_vma
)
1600 * This is a two-pass process; first we find the offsets of the bits,
1601 * and then they are actually written out. If we run out of core limit
1604 static int elf_core_dump(long signr
, struct pt_regs
*regs
, struct file
*file
, unsigned long limit
)
1612 struct vm_area_struct
*vma
, *gate_vma
;
1613 struct elfhdr
*elf
= NULL
;
1614 loff_t offset
= 0, dataoff
, foffset
;
1616 struct memelfnote
*notes
= NULL
;
1617 struct elf_prstatus
*prstatus
= NULL
; /* NT_PRSTATUS */
1618 struct elf_prpsinfo
*psinfo
= NULL
; /* NT_PRPSINFO */
1619 struct task_struct
*g
, *p
;
1620 LIST_HEAD(thread_list
);
1621 struct list_head
*t
;
1622 elf_fpregset_t
*fpu
= NULL
;
1623 #ifdef ELF_CORE_COPY_XFPREGS
1624 elf_fpxregset_t
*xfpu
= NULL
;
1626 int thread_status_size
= 0;
1628 unsigned long mm_flags
;
1631 * We no longer stop all VM operations.
1633 * This is because those proceses that could possibly change map_count
1634 * or the mmap / vma pages are now blocked in do_exit on current
1635 * finishing this core dump.
1637 * Only ptrace can touch these memory addresses, but it doesn't change
1638 * the map_count or the pages allocated. So no possibility of crashing
1639 * exists while dumping the mm->vm_next areas to the core file.
1642 /* alloc memory for large data structures: too large to be on stack */
1643 elf
= kmalloc(sizeof(*elf
), GFP_KERNEL
);
1646 prstatus
= kmalloc(sizeof(*prstatus
), GFP_KERNEL
);
1649 psinfo
= kmalloc(sizeof(*psinfo
), GFP_KERNEL
);
1652 notes
= kmalloc(NUM_NOTES
* sizeof(struct memelfnote
), GFP_KERNEL
);
1655 fpu
= kmalloc(sizeof(*fpu
), GFP_KERNEL
);
1658 #ifdef ELF_CORE_COPY_XFPREGS
1659 xfpu
= kmalloc(sizeof(*xfpu
), GFP_KERNEL
);
1665 struct elf_thread_status
*tmp
;
1668 if (current
->mm
== p
->mm
&& current
!= p
) {
1669 tmp
= kzalloc(sizeof(*tmp
), GFP_ATOMIC
);
1675 list_add(&tmp
->list
, &thread_list
);
1677 while_each_thread(g
,p
);
1679 list_for_each(t
, &thread_list
) {
1680 struct elf_thread_status
*tmp
;
1683 tmp
= list_entry(t
, struct elf_thread_status
, list
);
1684 sz
= elf_dump_thread_status(signr
, tmp
);
1685 thread_status_size
+= sz
;
1688 /* now collect the dump for the current */
1689 memset(prstatus
, 0, sizeof(*prstatus
));
1690 fill_prstatus(prstatus
, current
, signr
);
1691 elf_core_copy_regs(&prstatus
->pr_reg
, regs
);
1693 segs
= current
->mm
->map_count
;
1694 #ifdef ELF_CORE_EXTRA_PHDRS
1695 segs
+= ELF_CORE_EXTRA_PHDRS
;
1698 gate_vma
= get_gate_vma(current
);
1699 if (gate_vma
!= NULL
)
1703 fill_elf_header(elf
, segs
+ 1); /* including notes section */
1706 current
->flags
|= PF_DUMPCORE
;
1709 * Set up the notes in similar form to SVR4 core dumps made
1710 * with info from their /proc.
1713 fill_note(notes
+ 0, "CORE", NT_PRSTATUS
, sizeof(*prstatus
), prstatus
);
1714 fill_psinfo(psinfo
, current
->group_leader
, current
->mm
);
1715 fill_note(notes
+ 1, "CORE", NT_PRPSINFO
, sizeof(*psinfo
), psinfo
);
1719 auxv
= (elf_addr_t
*)current
->mm
->saved_auxv
;
1724 while (auxv
[i
- 2] != AT_NULL
);
1725 fill_note(¬es
[numnote
++], "CORE", NT_AUXV
,
1726 i
* sizeof(elf_addr_t
), auxv
);
1728 /* Try to dump the FPU. */
1729 if ((prstatus
->pr_fpvalid
=
1730 elf_core_copy_task_fpregs(current
, regs
, fpu
)))
1731 fill_note(notes
+ numnote
++,
1732 "CORE", NT_PRFPREG
, sizeof(*fpu
), fpu
);
1733 #ifdef ELF_CORE_COPY_XFPREGS
1734 if (elf_core_copy_task_xfpregs(current
, xfpu
))
1735 fill_note(notes
+ numnote
++,
1736 "LINUX", ELF_CORE_XFPREG_TYPE
, sizeof(*xfpu
), xfpu
);
1742 DUMP_WRITE(elf
, sizeof(*elf
));
1743 offset
+= sizeof(*elf
); /* Elf header */
1744 offset
+= (segs
+ 1) * sizeof(struct elf_phdr
); /* Program headers */
1747 /* Write notes phdr entry */
1749 struct elf_phdr phdr
;
1752 for (i
= 0; i
< numnote
; i
++)
1753 sz
+= notesize(notes
+ i
);
1755 sz
+= thread_status_size
;
1757 sz
+= elf_coredump_extra_notes_size();
1759 fill_elf_note_phdr(&phdr
, sz
, offset
);
1761 DUMP_WRITE(&phdr
, sizeof(phdr
));
1764 dataoff
= offset
= roundup(offset
, ELF_EXEC_PAGESIZE
);
1767 * We must use the same mm->flags while dumping core to avoid
1768 * inconsistency between the program headers and bodies, otherwise an
1769 * unusable core file can be generated.
1771 mm_flags
= current
->mm
->flags
;
1773 /* Write program headers for segments dump */
1774 for (vma
= first_vma(current
, gate_vma
); vma
!= NULL
;
1775 vma
= next_vma(vma
, gate_vma
)) {
1776 struct elf_phdr phdr
;
1778 phdr
.p_type
= PT_LOAD
;
1779 phdr
.p_offset
= offset
;
1780 phdr
.p_vaddr
= vma
->vm_start
;
1782 phdr
.p_filesz
= vma_dump_size(vma
, mm_flags
);
1783 phdr
.p_memsz
= vma
->vm_end
- vma
->vm_start
;
1784 offset
+= phdr
.p_filesz
;
1785 phdr
.p_flags
= vma
->vm_flags
& VM_READ
? PF_R
: 0;
1786 if (vma
->vm_flags
& VM_WRITE
)
1787 phdr
.p_flags
|= PF_W
;
1788 if (vma
->vm_flags
& VM_EXEC
)
1789 phdr
.p_flags
|= PF_X
;
1790 phdr
.p_align
= ELF_EXEC_PAGESIZE
;
1792 DUMP_WRITE(&phdr
, sizeof(phdr
));
1795 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1796 ELF_CORE_WRITE_EXTRA_PHDRS
;
1799 /* write out the notes section */
1800 for (i
= 0; i
< numnote
; i
++)
1801 if (!writenote(notes
+ i
, file
, &foffset
))
1804 if (elf_coredump_extra_notes_write(file
, &foffset
))
1807 /* write out the thread status notes section */
1808 list_for_each(t
, &thread_list
) {
1809 struct elf_thread_status
*tmp
=
1810 list_entry(t
, struct elf_thread_status
, list
);
1812 for (i
= 0; i
< tmp
->num_notes
; i
++)
1813 if (!writenote(&tmp
->notes
[i
], file
, &foffset
))
1818 DUMP_SEEK(dataoff
- foffset
);
1820 for (vma
= first_vma(current
, gate_vma
); vma
!= NULL
;
1821 vma
= next_vma(vma
, gate_vma
)) {
1825 end
= vma
->vm_start
+ vma_dump_size(vma
, mm_flags
);
1827 for (addr
= vma
->vm_start
; addr
< end
; addr
+= PAGE_SIZE
) {
1829 struct vm_area_struct
*vma
;
1831 if (get_user_pages(current
, current
->mm
, addr
, 1, 0, 1,
1832 &page
, &vma
) <= 0) {
1833 DUMP_SEEK(PAGE_SIZE
);
1835 if (page
== ZERO_PAGE(0)) {
1836 if (!dump_seek(file
, PAGE_SIZE
)) {
1837 page_cache_release(page
);
1842 flush_cache_page(vma
, addr
,
1845 if ((size
+= PAGE_SIZE
) > limit
||
1846 !dump_write(file
, kaddr
,
1849 page_cache_release(page
);
1854 page_cache_release(page
);
1859 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1860 ELF_CORE_WRITE_EXTRA_DATA
;
1867 while (!list_empty(&thread_list
)) {
1868 struct list_head
*tmp
= thread_list
.next
;
1870 kfree(list_entry(tmp
, struct elf_thread_status
, list
));
1878 #ifdef ELF_CORE_COPY_XFPREGS
1885 #endif /* USE_ELF_CORE_DUMP */
1887 static int __init
init_elf_binfmt(void)
1889 return register_binfmt(&elf_format
);
1892 static void __exit
exit_elf_binfmt(void)
1894 /* Remove the COFF and ELF loaders. */
1895 unregister_binfmt(&elf_format
);
1898 core_initcall(init_elf_binfmt
);
1899 module_exit(exit_elf_binfmt
);
1900 MODULE_LICENSE("GPL");