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__)
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(_WIN32)
45 #warning Bound checking does not support malloc (etc.) in this environment.
46 #undef CONFIG_TCC_MALLOC_HOOKS
50 #define BOUND_T1_BITS 13
51 #define BOUND_T2_BITS 11
52 #define BOUND_T3_BITS (32 - BOUND_T1_BITS - BOUND_T2_BITS)
54 #define BOUND_T1_SIZE (1 << BOUND_T1_BITS)
55 #define BOUND_T2_SIZE (1 << BOUND_T2_BITS)
56 #define BOUND_T3_SIZE (1 << BOUND_T3_BITS)
57 #define BOUND_E_BITS 4
59 #define BOUND_T23_BITS (BOUND_T2_BITS + BOUND_T3_BITS)
60 #define BOUND_T23_SIZE (1 << BOUND_T23_BITS)
63 /* this pointer is generated when bound check is incorrect */
64 #define INVALID_POINTER ((void *)(-2))
65 /* size of an empty region */
66 #define EMPTY_SIZE 0xffffffff
67 /* size of an invalid region */
68 #define INVALID_SIZE 0
70 typedef struct BoundEntry
{
73 struct BoundEntry
*next
;
74 unsigned long is_invalid
; /* true if pointers outside region are invalid */
77 /* external interface */
78 void __bound_init(void);
79 void __bound_new_region(void *p
, unsigned long size
);
80 int __bound_delete_region(void *p
);
82 #define FASTCALL __attribute__((regparm(3)))
84 void *__bound_malloc(size_t size
, const void *caller
);
85 void *__bound_memalign(size_t size
, size_t align
, const void *caller
);
86 void __bound_free(void *ptr
, const void *caller
);
87 void *__bound_realloc(void *ptr
, size_t size
, const void *caller
);
88 static void *libc_malloc(size_t size
);
89 static void libc_free(void *ptr
);
90 static void install_malloc_hooks(void);
91 static void restore_malloc_hooks(void);
93 #ifdef CONFIG_TCC_MALLOC_HOOKS
94 static void *saved_malloc_hook
;
95 static void *saved_free_hook
;
96 static void *saved_realloc_hook
;
97 static void *saved_memalign_hook
;
100 /* TCC definitions */
101 extern char __bounds_start
; /* start of static bounds table */
102 /* error message, just for TCC */
103 const char *__bound_error_msg
;
105 /* runtime error output */
106 extern void rt_error(unsigned long pc
, const char *fmt
, ...);
109 static BoundEntry
*__bound_t1
[BOUND_T1_SIZE
]; /* page table */
111 static BoundEntry
**__bound_t1
; /* page table */
113 static BoundEntry
*__bound_empty_t2
; /* empty page, for unused pages */
114 static BoundEntry
*__bound_invalid_t2
; /* invalid page, for invalid pointers */
116 static BoundEntry
*__bound_find_region(BoundEntry
*e1
, void *p
)
118 unsigned long addr
, tmp
;
123 addr
= (unsigned long)p
;
125 if (addr
<= e
->size
) {
126 /* put region at the head */
128 e1
->start
= e
->start
;
137 /* no entry found: return empty entry or invalid entry */
139 return __bound_invalid_t2
;
141 return __bound_empty_t2
;
144 /* print a bound error message */
145 static void bound_error(const char *fmt
, ...)
147 __bound_error_msg
= fmt
;
148 *(int *)0 = 0; /* force a runtime error */
151 static void bound_alloc_error(void)
153 bound_error("not enough memory for bound checking code");
156 /* return '(p + offset)' for pointer arithmetic (a pointer can reach
157 the end of a region in this case */
158 void * FASTCALL
__bound_ptr_add(void *p
, int offset
)
160 unsigned long addr
= (unsigned long)p
;
162 #if defined(BOUND_DEBUG)
163 printf("add: 0x%x %d\n", (int)p
, offset
);
166 e
= __bound_t1
[addr
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
)];
167 e
= (BoundEntry
*)((char *)e
+
168 ((addr
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
169 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
)));
171 if (addr
> e
->size
) {
172 e
= __bound_find_region(e
, p
);
173 addr
= (unsigned long)p
- e
->start
;
177 return INVALID_POINTER
; /* return an invalid pointer */
181 /* return '(p + offset)' for pointer indirection (the resulting must
182 be strictly inside the region */
183 #define BOUND_PTR_INDIR(dsize) \
184 void * FASTCALL __bound_ptr_indir ## dsize (void *p, int offset) \
186 unsigned long addr = (unsigned long)p; \
189 e = __bound_t1[addr >> (BOUND_T2_BITS + BOUND_T3_BITS)]; \
190 e = (BoundEntry *)((char *)e + \
191 ((addr >> (BOUND_T3_BITS - BOUND_E_BITS)) & \
192 ((BOUND_T2_SIZE - 1) << BOUND_E_BITS))); \
194 if (addr > e->size) { \
195 e = __bound_find_region(e, p); \
196 addr = (unsigned long)p - e->start; \
198 addr += offset + dsize; \
199 if (addr > e->size) \
200 return INVALID_POINTER; /* return an invalid pointer */ \
211 /* return the frame pointer of the caller */
212 #define GET_CALLER_FP(fp)\
214 fp = (unsigned long)__builtin_frame_address(1);\
217 /* called when entering a function to add all the local regions */
218 void FASTCALL
__bound_local_new(void *p1
)
220 unsigned long addr
, size
, fp
, *p
= p1
;
229 __bound_new_region((void *)addr
, size
);
233 /* called when leaving a function to delete all the local regions */
234 void FASTCALL
__bound_local_delete(void *p1
)
236 unsigned long addr
, fp
, *p
= p1
;
244 __bound_delete_region((void *)addr
);
248 static BoundEntry
*__bound_new_page(void)
253 page
= libc_malloc(sizeof(BoundEntry
) * BOUND_T2_SIZE
);
256 for(i
=0;i
<BOUND_T2_SIZE
;i
++) {
257 /* put empty entries */
259 page
[i
].size
= EMPTY_SIZE
;
261 page
[i
].is_invalid
= 0;
266 /* currently we use malloc(). Should use bound_new_page() */
267 static BoundEntry
*bound_new_entry(void)
270 e
= libc_malloc(sizeof(BoundEntry
));
274 static void bound_free_entry(BoundEntry
*e
)
279 static inline BoundEntry
*get_page(int index
)
282 page
= __bound_t1
[index
];
283 if (page
== __bound_empty_t2
|| page
== __bound_invalid_t2
) {
284 /* create a new page if necessary */
285 page
= __bound_new_page();
286 __bound_t1
[index
] = page
;
291 /* mark a region as being invalid (can only be used during init) */
292 static void mark_invalid(unsigned long addr
, unsigned long size
)
294 unsigned long start
, end
;
296 int t1_start
, t1_end
, i
, j
, t2_start
, t2_end
;
301 t2_start
= (start
+ BOUND_T3_SIZE
- 1) >> BOUND_T3_BITS
;
303 t2_end
= end
>> BOUND_T3_BITS
;
305 t2_end
= 1 << (BOUND_T1_BITS
+ BOUND_T2_BITS
);
308 printf("mark_invalid: start = %x %x\n", t2_start
, t2_end
);
311 /* first we handle full pages */
312 t1_start
= (t2_start
+ BOUND_T2_SIZE
- 1) >> BOUND_T2_BITS
;
313 t1_end
= t2_end
>> BOUND_T2_BITS
;
315 i
= t2_start
& (BOUND_T2_SIZE
- 1);
316 j
= t2_end
& (BOUND_T2_SIZE
- 1);
318 if (t1_start
== t1_end
) {
319 page
= get_page(t2_start
>> BOUND_T2_BITS
);
321 page
[i
].size
= INVALID_SIZE
;
322 page
[i
].is_invalid
= 1;
326 page
= get_page(t2_start
>> BOUND_T2_BITS
);
327 for(; i
< BOUND_T2_SIZE
; i
++) {
328 page
[i
].size
= INVALID_SIZE
;
329 page
[i
].is_invalid
= 1;
332 for(i
= t1_start
; i
< t1_end
; i
++) {
333 __bound_t1
[i
] = __bound_invalid_t2
;
336 page
= get_page(t1_end
);
337 for(i
= 0; i
< j
; i
++) {
338 page
[i
].size
= INVALID_SIZE
;
339 page
[i
].is_invalid
= 1;
345 void __bound_init(void)
349 unsigned long start
, size
;
352 /* save malloc hooks and install bound check hooks */
353 install_malloc_hooks();
356 __bound_t1
= libc_malloc(BOUND_T1_SIZE
* sizeof(BoundEntry
*));
360 __bound_empty_t2
= __bound_new_page();
361 for(i
=0;i
<BOUND_T1_SIZE
;i
++) {
362 __bound_t1
[i
] = __bound_empty_t2
;
365 page
= __bound_new_page();
366 for(i
=0;i
<BOUND_T2_SIZE
;i
++) {
367 /* put invalid entries */
369 page
[i
].size
= INVALID_SIZE
;
371 page
[i
].is_invalid
= 1;
373 __bound_invalid_t2
= page
;
375 /* invalid pointer zone */
376 start
= (unsigned long)INVALID_POINTER
& ~(BOUND_T23_SIZE
- 1);
377 size
= BOUND_T23_SIZE
;
378 mark_invalid(start
, size
);
380 #if !defined(__TINYC__) && defined(CONFIG_TCC_MALLOC_HOOKS)
381 /* malloc zone is also marked invalid. can only use that with
382 * hooks because all libs should use the same malloc. The solution
383 * would be to build a new malloc for tcc.
385 * usually heap (= malloc zone) comes right after bss, i.e. after _end, but
386 * not always - either if we are running from under `tcc -b -run`, or if
387 * address space randomization is turned on(a), heap start will be separated
390 * So sbrk(0) will be a good approximation for start_brk:
392 * - if we are a separately compiled program, __bound_init() runs early,
393 * and sbrk(0) should be equal or very near to start_brk(b) (in case other
394 * constructors malloc something), or
396 * - if we are running from under `tcc -b -run`, sbrk(0) will return
397 * start of heap portion which is under this program control, and not
398 * mark as invalid earlier allocated memory.
401 * (a) /proc/sys/kernel/randomize_va_space = 2, on Linux;
402 * usually turned on by default.
404 * (b) on Linux >= v3.3, the alternative is to read
405 * start_brk from /proc/self/stat
407 start
= (unsigned long)sbrk(0);
408 size
= 128 * 0x100000;
409 mark_invalid(start
, size
);
412 /* add all static bound check values */
413 p
= (int *)&__bounds_start
;
415 __bound_new_region((void *)p
[0], p
[1]);
420 void __bound_exit(void)
422 restore_malloc_hooks();
425 static inline void add_region(BoundEntry
*e
,
426 unsigned long start
, unsigned long size
)
430 /* no region : add it */
434 /* already regions in the list: add it at the head */
435 e1
= bound_new_entry();
436 e1
->start
= e
->start
;
445 /* create a new region. It should not already exist in the region list */
446 void __bound_new_region(void *p
, unsigned long size
)
448 unsigned long start
, end
;
449 BoundEntry
*page
, *e
, *e2
;
450 int t1_start
, t1_end
, i
, t2_start
, t2_end
;
452 start
= (unsigned long)p
;
454 t1_start
= start
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
);
455 t1_end
= end
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
);
458 page
= get_page(t1_start
);
459 t2_start
= (start
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
460 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
);
461 t2_end
= (end
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
462 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
);
464 printf("new %lx %lx %x %x %x %x\n",
465 start
, end
, t1_start
, t1_end
, t2_start
, t2_end
);
468 e
= (BoundEntry
*)((char *)page
+ t2_start
);
469 add_region(e
, start
, size
);
471 if (t1_end
== t1_start
) {
472 /* same ending page */
473 e2
= (BoundEntry
*)((char *)page
+ t2_end
);
480 add_region(e
, start
, size
);
483 /* mark until end of page */
484 e2
= page
+ BOUND_T2_SIZE
;
490 /* mark intermediate pages, if any */
491 for(i
=t1_start
+1;i
<t1_end
;i
++) {
493 e2
= page
+ BOUND_T2_SIZE
;
494 for(e
=page
;e
<e2
;e
++) {
500 page
= get_page(t1_end
);
501 e2
= (BoundEntry
*)((char *)page
+ t2_end
);
502 for(e
=page
;e
<e2
;e
++) {
506 add_region(e
, start
, size
);
510 /* delete a region */
511 static inline void delete_region(BoundEntry
*e
,
512 void *p
, unsigned long empty_size
)
517 addr
= (unsigned long)p
;
519 if (addr
<= e
->size
) {
520 /* region found is first one */
523 /* no more region: mark it empty */
525 e
->size
= empty_size
;
527 /* copy next region in head */
528 e
->start
= e1
->start
;
531 bound_free_entry(e1
);
534 /* find the matching region */
538 /* region not found: do nothing */
541 addr
= (unsigned long)p
- e
->start
;
542 if (addr
<= e
->size
) {
543 /* found: remove entry */
552 /* WARNING: 'p' must be the starting point of the region. */
553 /* return non zero if error */
554 int __bound_delete_region(void *p
)
556 unsigned long start
, end
, addr
, size
, empty_size
;
557 BoundEntry
*page
, *e
, *e2
;
558 int t1_start
, t1_end
, t2_start
, t2_end
, i
;
560 start
= (unsigned long)p
;
561 t1_start
= start
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
);
562 t2_start
= (start
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
563 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
);
565 /* find region size */
566 page
= __bound_t1
[t1_start
];
567 e
= (BoundEntry
*)((char *)page
+ t2_start
);
568 addr
= start
- e
->start
;
570 e
= __bound_find_region(e
, p
);
571 /* test if invalid region */
572 if (e
->size
== EMPTY_SIZE
|| (unsigned long)p
!= e
->start
)
574 /* compute the size we put in invalid regions */
576 empty_size
= INVALID_SIZE
;
578 empty_size
= EMPTY_SIZE
;
582 /* now we can free each entry */
583 t1_end
= end
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
);
584 t2_end
= (end
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
585 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
);
587 delete_region(e
, p
, empty_size
);
588 if (t1_end
== t1_start
) {
589 /* same ending page */
590 e2
= (BoundEntry
*)((char *)page
+ t2_end
);
595 e
->size
= empty_size
;
597 delete_region(e
, p
, empty_size
);
600 /* mark until end of page */
601 e2
= page
+ BOUND_T2_SIZE
;
605 e
->size
= empty_size
;
607 /* mark intermediate pages, if any */
608 /* XXX: should free them */
609 for(i
=t1_start
+1;i
<t1_end
;i
++) {
611 e2
= page
+ BOUND_T2_SIZE
;
612 for(e
=page
;e
<e2
;e
++) {
614 e
->size
= empty_size
;
618 page
= get_page(t1_end
);
619 e2
= (BoundEntry
*)((char *)page
+ t2_end
);
620 for(e
=page
;e
<e2
;e
++) {
622 e
->size
= empty_size
;
624 delete_region(e
, p
, empty_size
);
629 /* return the size of the region starting at p, or EMPTY_SIZE if non
631 static unsigned long get_region_size(void *p
)
633 unsigned long addr
= (unsigned long)p
;
636 e
= __bound_t1
[addr
>> (BOUND_T2_BITS
+ BOUND_T3_BITS
)];
637 e
= (BoundEntry
*)((char *)e
+
638 ((addr
>> (BOUND_T3_BITS
- BOUND_E_BITS
)) &
639 ((BOUND_T2_SIZE
- 1) << BOUND_E_BITS
)));
642 e
= __bound_find_region(e
, p
);
643 if (e
->start
!= (unsigned long)p
)
648 /* patched memory functions */
650 /* force compiler to perform stores coded up to this point */
651 #define barrier() __asm__ __volatile__ ("": : : "memory")
653 static void install_malloc_hooks(void)
655 #ifdef CONFIG_TCC_MALLOC_HOOKS
656 saved_malloc_hook
= __malloc_hook
;
657 saved_free_hook
= __free_hook
;
658 saved_realloc_hook
= __realloc_hook
;
659 saved_memalign_hook
= __memalign_hook
;
660 __malloc_hook
= __bound_malloc
;
661 __free_hook
= __bound_free
;
662 __realloc_hook
= __bound_realloc
;
663 __memalign_hook
= __bound_memalign
;
669 static void restore_malloc_hooks(void)
671 #ifdef CONFIG_TCC_MALLOC_HOOKS
672 __malloc_hook
= saved_malloc_hook
;
673 __free_hook
= saved_free_hook
;
674 __realloc_hook
= saved_realloc_hook
;
675 __memalign_hook
= saved_memalign_hook
;
681 static void *libc_malloc(size_t size
)
684 restore_malloc_hooks();
686 install_malloc_hooks();
690 static void libc_free(void *ptr
)
692 restore_malloc_hooks();
694 install_malloc_hooks();
697 /* XXX: we should use a malloc which ensure that it is unlikely that
698 two malloc'ed data have the same address if 'free' are made in
700 void *__bound_malloc(size_t size
, const void *caller
)
704 /* we allocate one more byte to ensure the regions will be
705 separated by at least one byte. With the glibc malloc, it may
706 be in fact not necessary */
707 ptr
= libc_malloc(size
+ 1);
711 __bound_new_region(ptr
, size
);
715 void *__bound_memalign(size_t size
, size_t align
, const void *caller
)
719 restore_malloc_hooks();
721 #ifndef HAVE_MEMALIGN
723 /* XXX: handle it ? */
726 /* we suppose that malloc aligns to at least four bytes */
727 ptr
= malloc(size
+ 1);
730 /* we allocate one more byte to ensure the regions will be
731 separated by at least one byte. With the glibc malloc, it may
732 be in fact not necessary */
733 ptr
= memalign(size
+ 1, align
);
736 install_malloc_hooks();
740 __bound_new_region(ptr
, size
);
744 void __bound_free(void *ptr
, const void *caller
)
748 if (__bound_delete_region(ptr
) != 0)
749 bound_error("freeing invalid region");
754 void *__bound_realloc(void *ptr
, size_t size
, const void *caller
)
760 __bound_free(ptr
, caller
);
763 ptr1
= __bound_malloc(size
, caller
);
764 if (ptr
== NULL
|| ptr1
== NULL
)
766 old_size
= get_region_size(ptr
);
767 if (old_size
== EMPTY_SIZE
)
768 bound_error("realloc'ing invalid pointer");
769 memcpy(ptr1
, ptr
, old_size
);
770 __bound_free(ptr
, caller
);
775 #ifndef CONFIG_TCC_MALLOC_HOOKS
776 void *__bound_calloc(size_t nmemb
, size_t size
)
780 ptr
= __bound_malloc(size
, NULL
);
783 memset(ptr
, 0, size
);
789 static void bound_dump(void)
791 BoundEntry
*page
, *e
;
794 printf("region dump:\n");
795 for(i
=0;i
<BOUND_T1_SIZE
;i
++) {
796 page
= __bound_t1
[i
];
797 for(j
=0;j
<BOUND_T2_SIZE
;j
++) {
799 /* do not print invalid or empty entries */
800 if (e
->size
!= EMPTY_SIZE
&& e
->start
!= 0) {
802 (i
<< (BOUND_T2_BITS
+ BOUND_T3_BITS
)) +
803 (j
<< BOUND_T3_BITS
));
805 printf(" %08lx:%08lx", e
->start
, e
->start
+ e
->size
);
815 /* some useful checked functions */
817 /* check that (p ... p + size - 1) lies inside 'p' region, if any */
818 static void __bound_check(const void *p
, size_t size
)
822 p
= __bound_ptr_add((void *)p
, size
);
823 if (p
== INVALID_POINTER
)
824 bound_error("invalid pointer");
827 void *__bound_memcpy(void *dst
, const void *src
, size_t size
)
829 __bound_check(dst
, size
);
830 __bound_check(src
, size
);
831 /* check also region overlap */
832 if (src
>= dst
&& src
< dst
+ size
)
833 bound_error("overlapping regions in memcpy()");
834 return memcpy(dst
, src
, size
);
837 void *__bound_memmove(void *dst
, const void *src
, size_t size
)
839 __bound_check(dst
, size
);
840 __bound_check(src
, size
);
841 return memmove(dst
, src
, size
);
844 void *__bound_memset(void *dst
, int c
, size_t size
)
846 __bound_check(dst
, size
);
847 return memset(dst
, c
, size
);
850 /* XXX: could be optimized */
851 int __bound_strlen(const char *s
)
858 p
= __bound_ptr_indir1((char *)s
, len
);
859 if (p
== INVALID_POINTER
)
860 bound_error("bad pointer in strlen()");
868 char *__bound_strcpy(char *dst
, const char *src
)
871 len
= __bound_strlen(src
);
872 return __bound_memcpy(dst
, src
, len
+ 1);