2 * Tiny C Memory and bounds checker
4 * Copyright (c) 2002 Fabrice Bellard
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
24 #if !defined(__FreeBSD__) && !defined(__FreeBSD_kernel__) \
25 && !defined(__DragonFly__) && !defined(__OpenBSD__) && !defined(__NetBSD__)
32 /* #define BOUND_DEBUG */
34 /* define so that bound array is static (faster, but use memory if
35 bound checking not used) */
36 /* #define BOUND_STATIC */
38 /* use malloc hooks. Currently the code cannot be reliable if no hooks */
39 #define CONFIG_TCC_MALLOC_HOOKS
42 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) \
43 || defined(__DragonFly__) || defined(__dietlibc__) \
44 || defined(__UCLIBC__) || defined(__OpenBSD__) || defined(__NetBSD__) \
45 || defined(_WIN32) || defined(TCC_UCLIBC)
46 #warning Bound checking does not support malloc (etc.) in this environment.
47 #undef CONFIG_TCC_MALLOC_HOOKS
51 #define BOUND_T1_BITS 13
52 #define BOUND_T2_BITS 11
53 #define BOUND_T3_BITS (32 - BOUND_T1_BITS - BOUND_T2_BITS)
55 #define BOUND_T1_SIZE (1 << BOUND_T1_BITS)
56 #define BOUND_T2_SIZE (1 << BOUND_T2_BITS)
57 #define BOUND_T3_SIZE (1 << BOUND_T3_BITS)
58 #define BOUND_E_BITS 4
60 #define BOUND_T23_BITS (BOUND_T2_BITS + BOUND_T3_BITS)
61 #define BOUND_T23_SIZE (1 << BOUND_T23_BITS)
64 /* this pointer is generated when bound check is incorrect */
65 #define INVALID_POINTER ((void *)(-2))
66 /* size of an empty region */
67 #define EMPTY_SIZE 0xffffffff
68 /* size of an invalid region */
69 #define INVALID_SIZE 0
71 typedef struct BoundEntry
{
74 struct BoundEntry
*next
;
75 size_t is_invalid
; /* true if pointers outside region are invalid */
78 /* external interface */
79 void __bound_init(void);
80 void __bound_new_region(void *p
, size_t size
);
81 int __bound_delete_region(void *p
);
84 /* an __attribute__ macro is defined in the system headers */
87 #define FASTCALL __attribute__((regparm(3)))
89 void *__bound_malloc(size_t size
, const void *caller
);
90 void *__bound_memalign(size_t size
, size_t align
, const void *caller
);
91 void __bound_free(void *ptr
, const void *caller
);
92 void *__bound_realloc(void *ptr
, size_t size
, const void *caller
);
93 static void *libc_malloc(size_t size
);
94 static void libc_free(void *ptr
);
95 static void install_malloc_hooks(void);
96 static void restore_malloc_hooks(void);
98 #ifdef CONFIG_TCC_MALLOC_HOOKS
99 static void *saved_malloc_hook
;
100 static void *saved_free_hook
;
101 static void *saved_realloc_hook
;
102 static void *saved_memalign_hook
;
105 /* TCC definitions */
106 extern char __bounds_start
; /* start of static bounds table */
107 /* error message, just for TCC */
108 const char *__bound_error_msg
;
110 /* runtime error output */
111 extern void rt_error(size_t pc
, const char *fmt
, ...);
114 static BoundEntry
*__bound_t1
[BOUND_T1_SIZE
]; /* page table */
116 static BoundEntry
**__bound_t1
; /* page table */
118 static BoundEntry
*__bound_empty_t2
; /* empty page, for unused pages */
119 static BoundEntry
*__bound_invalid_t2
; /* invalid page, for invalid pointers */
121 static BoundEntry
*__bound_find_region(BoundEntry
*e1
, void *p
)
130 if (addr
<= e
->size
) {
131 /* put region at the head */
133 e1
->start
= e
->start
;
142 /* no entry found: return empty entry or invalid entry */
144 return __bound_invalid_t2
;
146 return __bound_empty_t2
;
149 /* print a bound error message */
150 static void bound_error(const char *fmt
, ...)
152 __bound_error_msg
= fmt
;
153 *(int *)0 = 0; /* force a runtime error */
156 static void bound_alloc_error(void)
158 bound_error("not enough memory for bound checking code");
161 /* return '(p + offset)' for pointer arithmetic (a pointer can reach
162 the end of a region in this case */
163 void * FASTCALL
__bound_ptr_add(void *p
, size_t offset
)
165 size_t addr
= (size_t)p
;
167 #if defined(BOUND_DEBUG)
168 printf("%s %s: 0x%x %d\n", __FILE__
, __FUNCTION__
, (int)p
, offset
);
171 e
= __bound_t1
[addr
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
)];
172 e
= (BoundEntry
*)((char *)e
+
173 ((addr
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
174 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
)));
176 if (addr
> e
->size
) {
177 e
= __bound_find_region(e
, p
);
178 addr
= (size_t)p
- e
->start
;
181 if (addr
>= e
->size
) {
182 #if defined(BOUND_DEBUG)
183 printf("%s %s: 0x%zx is outside of the region\n", __FILE__
, __FUNCTION__
, p
+ offset
);
185 return INVALID_POINTER
; /* return an invalid pointer */
190 /* return '(p + offset)' for pointer indirection (the resulting must
191 be strictly inside the region */
192 #define BOUND_PTR_INDIR(dsize) \
193 void * FASTCALL __bound_ptr_indir ## dsize (void *p, size_t offset) \
195 size_t addr = (size_t)p; \
198 e = __bound_t1[addr >> (BOUND_T2_BITS + BOUND_T3_BITS)]; \
199 e = (BoundEntry *)((char *)e + \
200 ((addr >> (BOUND_T3_BITS - BOUND_E_BITS)) & \
201 ((BOUND_T2_SIZE - 1) << BOUND_E_BITS))); \
203 if (addr > e->size) { \
204 e = __bound_find_region(e, p); \
205 addr = (size_t)p - e->start; \
207 addr += offset + dsize; \
208 if (addr > e->size) \
209 return INVALID_POINTER; /* return an invalid pointer */ \
220 /* return the frame pointer of the caller */
221 #define GET_CALLER_FP(fp)\
223 fp = (size_t)__builtin_frame_address(1);\
226 /* called when entering a function to add all the local regions */
227 void FASTCALL
__bound_local_new(void *p1
)
229 size_t addr
, size
, fp
, *p
= p1
;
231 fprintf(stderr
, "%s, %s start p1=%zx *p1=%zx\n", __FILE__
, __FUNCTION__
, p
, *p
);
241 __bound_new_region((void *)addr
, size
);
244 fprintf(stderr
, "%s, %s end\n", __FILE__
, __FUNCTION__
);
248 /* called when leaving a function to delete all the local regions */
249 void FASTCALL
__bound_local_delete(void *p1
)
251 size_t addr
, fp
, *p
= p1
;
259 __bound_delete_region((void *)addr
);
263 static BoundEntry
*__bound_new_page(void)
268 page
= libc_malloc(sizeof(BoundEntry
) * BOUND_T2_SIZE
);
271 for(i
=0;i
<BOUND_T2_SIZE
;i
++) {
272 /* put empty entries */
274 page
[i
].size
= EMPTY_SIZE
;
276 page
[i
].is_invalid
= 0;
281 /* currently we use malloc(). Should use bound_new_page() */
282 static BoundEntry
*bound_new_entry(void)
285 e
= libc_malloc(sizeof(BoundEntry
));
289 static void bound_free_entry(BoundEntry
*e
)
294 static inline BoundEntry
*get_page(int index
)
297 page
= __bound_t1
[index
];
298 if (page
== __bound_empty_t2
|| page
== __bound_invalid_t2
) {
299 /* create a new page if necessary */
300 page
= __bound_new_page();
301 __bound_t1
[index
] = page
;
306 /* mark a region as being invalid (can only be used during init) */
307 static void mark_invalid(size_t addr
, size_t size
)
311 size_t t1_start
, t1_end
, i
, j
, t2_start
, t2_end
;
316 t2_start
= (start
+ BOUND_T3_SIZE
- 1) >> BOUND_T3_BITS
;
318 t2_end
= end
>> BOUND_T3_BITS
;
320 t2_end
= 1 << (BOUND_T1_BITS
+ BOUND_T2_BITS
);
323 printf("mark_invalid: start = %x %x\n", t2_start
, t2_end
);
326 /* first we handle full pages */
327 t1_start
= (t2_start
+ BOUND_T2_SIZE
- 1) >> BOUND_T2_BITS
;
328 t1_end
= t2_end
>> BOUND_T2_BITS
;
330 i
= t2_start
& (BOUND_T2_SIZE
- 1);
331 j
= t2_end
& (BOUND_T2_SIZE
- 1);
333 if (t1_start
== t1_end
) {
334 page
= get_page(t2_start
>> BOUND_T2_BITS
);
336 page
[i
].size
= INVALID_SIZE
;
337 page
[i
].is_invalid
= 1;
341 page
= get_page(t2_start
>> BOUND_T2_BITS
);
342 for(; i
< BOUND_T2_SIZE
; i
++) {
343 page
[i
].size
= INVALID_SIZE
;
344 page
[i
].is_invalid
= 1;
347 for(i
= t1_start
; i
< t1_end
; i
++) {
348 __bound_t1
[i
] = __bound_invalid_t2
;
351 page
= get_page(t1_end
);
352 for(i
= 0; i
< j
; i
++) {
353 page
[i
].size
= INVALID_SIZE
;
354 page
[i
].is_invalid
= 1;
360 void __bound_init(void)
368 fprintf(stderr
, "%s, %s() start\n", __FILE__
, __FUNCTION__
);
371 /* save malloc hooks and install bound check hooks */
372 install_malloc_hooks();
375 __bound_t1
= libc_malloc(BOUND_T1_SIZE
* sizeof(BoundEntry
*));
379 __bound_empty_t2
= __bound_new_page();
380 for(i
=0;i
<BOUND_T1_SIZE
;i
++) {
381 __bound_t1
[i
] = __bound_empty_t2
;
384 page
= __bound_new_page();
385 for(i
=0;i
<BOUND_T2_SIZE
;i
++) {
386 /* put invalid entries */
388 page
[i
].size
= INVALID_SIZE
;
390 page
[i
].is_invalid
= 1;
392 __bound_invalid_t2
= page
;
394 /* invalid pointer zone */
395 start
= (size_t)INVALID_POINTER
& ~(BOUND_T23_SIZE
- 1);
396 size
= BOUND_T23_SIZE
;
397 mark_invalid(start
, size
);
399 #if defined(CONFIG_TCC_MALLOC_HOOKS)
400 /* malloc zone is also marked invalid. can only use that with
401 * hooks because all libs should use the same malloc. The solution
402 * would be to build a new malloc for tcc.
404 * usually heap (= malloc zone) comes right after bss, i.e. after _end, but
405 * not always - either if we are running from under `tcc -b -run`, or if
406 * address space randomization is turned on(a), heap start will be separated
409 * So sbrk(0) will be a good approximation for start_brk:
411 * - if we are a separately compiled program, __bound_init() runs early,
412 * and sbrk(0) should be equal or very near to start_brk(b) (in case other
413 * constructors malloc something), or
415 * - if we are running from under `tcc -b -run`, sbrk(0) will return
416 * start of heap portion which is under this program control, and not
417 * mark as invalid earlier allocated memory.
420 * (a) /proc/sys/kernel/randomize_va_space = 2, on Linux;
421 * usually turned on by default.
423 * (b) on Linux >= v3.3, the alternative is to read
424 * start_brk from /proc/self/stat
426 start
= (size_t)sbrk(0);
427 size
= 128 * 0x100000;
428 mark_invalid(start
, size
);
431 /* add all static bound check values */
432 p
= (size_t *)&__bounds_start
;
434 __bound_new_region((void *)p
[0], p
[1]);
438 fprintf(stderr
, "%s, %s() end\n\n", __FILE__
, __FUNCTION__
);
442 void __bound_main_arg(void **p
)
447 fprintf(stderr
, "%s, %s calling __bound_new_region(%p, %llx)\n",
448 __FILE__
, __FUNCTION__
, (void *) p
- start
);
450 __bound_new_region(start
, (void *) p
- start
);
453 void __bound_exit(void)
455 restore_malloc_hooks();
458 static inline void add_region(BoundEntry
*e
,
459 size_t start
, size_t size
)
463 /* no region : add it */
467 /* already regions in the list: add it at the head */
468 e1
= bound_new_entry();
469 e1
->start
= e
->start
;
478 /* create a new region. It should not already exist in the region list */
479 void __bound_new_region(void *p
, size_t size
)
482 BoundEntry
*page
, *e
, *e2
;
483 size_t t1_start
, t1_end
, i
, t2_start
, t2_end
;
486 fprintf(stderr
, "%s, %s(%p, %zx) start\n",
487 __FILE__
, __FUNCTION__
, p
, size
);
492 t1_start
= start
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
);
493 t1_end
= end
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
);
496 page
= get_page(t1_start
);
497 t2_start
= (start
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
498 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
);
499 t2_end
= (end
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
500 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
);
503 e
= (BoundEntry
*)((char *)page
+ t2_start
);
504 add_region(e
, start
, size
);
506 if (t1_end
== t1_start
) {
507 /* same ending page */
508 e2
= (BoundEntry
*)((char *)page
+ t2_end
);
515 add_region(e
, start
, size
);
518 /* mark until end of page */
519 e2
= page
+ BOUND_T2_SIZE
;
525 /* mark intermediate pages, if any */
526 for(i
=t1_start
+1;i
<t1_end
;i
++) {
528 e2
= page
+ BOUND_T2_SIZE
;
529 for(e
=page
;e
<e2
;e
++) {
535 page
= get_page(t1_end
);
536 e2
= (BoundEntry
*)((char *)page
+ t2_end
);
537 for(e
=page
;e
<e2
;e
++) {
541 add_region(e
, start
, size
);
544 fprintf(stderr
, "%s, %s end\n", __FILE__
, __FUNCTION__
);
548 /* delete a region */
549 static inline void delete_region(BoundEntry
*e
,
550 void *p
, size_t empty_size
)
557 if (addr
<= e
->size
) {
558 /* region found is first one */
561 /* no more region: mark it empty */
563 e
->size
= empty_size
;
565 /* copy next region in head */
566 e
->start
= e1
->start
;
569 bound_free_entry(e1
);
572 /* find the matching region */
576 /* region not found: do nothing */
579 addr
= (size_t)p
- e
->start
;
580 if (addr
<= e
->size
) {
581 /* found: remove entry */
590 /* WARNING: 'p' must be the starting point of the region. */
591 /* return non zero if error */
592 int __bound_delete_region(void *p
)
594 size_t start
, end
, addr
, size
, empty_size
;
595 BoundEntry
*page
, *e
, *e2
;
596 size_t t1_start
, t1_end
, t2_start
, t2_end
, i
;
599 fprintf(stderr
, "%s %s() start\n", __FILE__
, __FUNCTION__
);
603 t1_start
= start
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
);
604 t2_start
= (start
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
605 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
);
607 /* find region size */
608 page
= __bound_t1
[t1_start
];
609 e
= (BoundEntry
*)((char *)page
+ t2_start
);
610 addr
= start
- e
->start
;
612 e
= __bound_find_region(e
, p
);
613 /* test if invalid region */
614 if (e
->size
== EMPTY_SIZE
|| (size_t)p
!= e
->start
)
616 /* compute the size we put in invalid regions */
618 empty_size
= INVALID_SIZE
;
620 empty_size
= EMPTY_SIZE
;
624 /* now we can free each entry */
625 t1_end
= end
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
);
626 t2_end
= (end
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
627 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
);
629 delete_region(e
, p
, empty_size
);
630 if (t1_end
== t1_start
) {
631 /* same ending page */
632 e2
= (BoundEntry
*)((char *)page
+ t2_end
);
637 e
->size
= empty_size
;
639 delete_region(e
, p
, empty_size
);
642 /* mark until end of page */
643 e2
= page
+ BOUND_T2_SIZE
;
647 e
->size
= empty_size
;
649 /* mark intermediate pages, if any */
650 /* XXX: should free them */
651 for(i
=t1_start
+1;i
<t1_end
;i
++) {
653 e2
= page
+ BOUND_T2_SIZE
;
654 for(e
=page
;e
<e2
;e
++) {
656 e
->size
= empty_size
;
660 page
= get_page(t1_end
);
661 e2
= (BoundEntry
*)((char *)page
+ t2_end
);
662 for(e
=page
;e
<e2
;e
++) {
664 e
->size
= empty_size
;
666 delete_region(e
, p
, empty_size
);
670 fprintf(stderr
, "%s %s() end\n", __FILE__
, __FUNCTION__
);
676 /* return the size of the region starting at p, or EMPTY_SIZE if non
678 static size_t get_region_size(void *p
)
680 size_t addr
= (size_t)p
;
683 e
= __bound_t1
[addr
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
)];
684 e
= (BoundEntry
*)((char *)e
+
685 ((addr
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
686 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
)));
689 e
= __bound_find_region(e
, p
);
690 if (e
->start
!= (size_t)p
)
695 /* patched memory functions */
697 /* force compiler to perform stores coded up to this point */
698 #define barrier() __asm__ __volatile__ ("": : : "memory")
700 static void install_malloc_hooks(void)
702 #ifdef CONFIG_TCC_MALLOC_HOOKS
703 saved_malloc_hook
= __malloc_hook
;
704 saved_free_hook
= __free_hook
;
705 saved_realloc_hook
= __realloc_hook
;
706 saved_memalign_hook
= __memalign_hook
;
707 __malloc_hook
= __bound_malloc
;
708 __free_hook
= __bound_free
;
709 __realloc_hook
= __bound_realloc
;
710 __memalign_hook
= __bound_memalign
;
716 static void restore_malloc_hooks(void)
718 #ifdef CONFIG_TCC_MALLOC_HOOKS
719 __malloc_hook
= saved_malloc_hook
;
720 __free_hook
= saved_free_hook
;
721 __realloc_hook
= saved_realloc_hook
;
722 __memalign_hook
= saved_memalign_hook
;
728 static void *libc_malloc(size_t size
)
731 restore_malloc_hooks();
733 install_malloc_hooks();
737 static void libc_free(void *ptr
)
739 restore_malloc_hooks();
741 install_malloc_hooks();
744 /* XXX: we should use a malloc which ensure that it is unlikely that
745 two malloc'ed data have the same address if 'free' are made in
747 void *__bound_malloc(size_t size
, const void *caller
)
751 /* we allocate one more byte to ensure the regions will be
752 separated by at least one byte. With the glibc malloc, it may
753 be in fact not necessary */
754 ptr
= libc_malloc(size
+ 1);
760 fprintf(stderr
, "%s, %s calling __bound_new_region(%p, %llx)\n",
761 __FILE__
, __FUNCTION__
, ptr
, size
);
763 __bound_new_region(ptr
, size
);
767 void *__bound_memalign(size_t size
, size_t align
, const void *caller
)
771 restore_malloc_hooks();
773 #ifndef HAVE_MEMALIGN
775 /* XXX: handle it ? */
778 /* we suppose that malloc aligns to at least four bytes */
779 ptr
= malloc(size
+ 1);
782 /* we allocate one more byte to ensure the regions will be
783 separated by at least one byte. With the glibc malloc, it may
784 be in fact not necessary */
785 ptr
= memalign(size
+ 1, align
);
788 install_malloc_hooks();
794 fprintf(stderr
, "%s, %s calling __bound_new_region(%p, %llx)\n",
795 __FILE__
, __FUNCTION__
, ptr
, size
);
797 __bound_new_region(ptr
, size
);
801 void __bound_free(void *ptr
, const void *caller
)
805 if (__bound_delete_region(ptr
) != 0)
806 bound_error("freeing invalid region");
811 void *__bound_realloc(void *ptr
, size_t size
, const void *caller
)
817 __bound_free(ptr
, caller
);
820 ptr1
= __bound_malloc(size
, caller
);
821 if (ptr
== NULL
|| ptr1
== NULL
)
823 old_size
= get_region_size(ptr
);
824 if (old_size
== EMPTY_SIZE
)
825 bound_error("realloc'ing invalid pointer");
826 memcpy(ptr1
, ptr
, old_size
);
827 __bound_free(ptr
, caller
);
832 #ifndef CONFIG_TCC_MALLOC_HOOKS
833 void *__bound_calloc(size_t nmemb
, size_t size
)
837 ptr
= __bound_malloc(size
, NULL
);
840 memset(ptr
, 0, size
);
846 static void bound_dump(void)
848 BoundEntry
*page
, *e
;
851 printf("region dump:\n");
852 for(i
=0;i
<BOUND_T1_SIZE
;i
++) {
853 page
= __bound_t1
[i
];
854 for(j
=0;j
<BOUND_T2_SIZE
;j
++) {
856 /* do not print invalid or empty entries */
857 if (e
->size
!= EMPTY_SIZE
&& e
->start
!= 0) {
859 (i
<< (BOUND_T2_BITS
+ BOUND_T3_BITS
)) +
860 (j
<< BOUND_T3_BITS
));
862 printf(" %08lx:%08lx", e
->start
, e
->start
+ e
->size
);
872 /* some useful checked functions */
874 /* check that (p ... p + size - 1) lies inside 'p' region, if any */
875 static void __bound_check(const void *p
, size_t size
)
879 p
= __bound_ptr_add((void *)p
, size
);
880 if (p
== INVALID_POINTER
)
881 bound_error("invalid pointer");
884 void *__bound_memcpy(void *dst
, const void *src
, size_t size
)
886 __bound_check(dst
, size
);
887 __bound_check(src
, size
);
888 /* check also region overlap */
889 if (src
>= dst
&& src
< dst
+ size
)
890 bound_error("overlapping regions in memcpy()");
891 return memcpy(dst
, src
, size
);
894 void *__bound_memmove(void *dst
, const void *src
, size_t size
)
896 __bound_check(dst
, size
);
897 __bound_check(src
, size
);
898 return memmove(dst
, src
, size
);
901 void *__bound_memset(void *dst
, int c
, size_t size
)
903 __bound_check(dst
, size
);
904 return memset(dst
, c
, size
);
907 /* XXX: could be optimized */
908 int __bound_strlen(const char *s
)
915 p
= __bound_ptr_indir1((char *)s
, len
);
916 if (p
== INVALID_POINTER
)
917 bound_error("bad pointer in strlen()");
925 char *__bound_strcpy(char *dst
, const char *src
)
928 len
= __bound_strlen(src
);
929 return __bound_memcpy(dst
, src
, len
+ 1);