Mark ChangeLog
[official-gcc.git] / gcc / asan.c
blobd68579b2e5ea28dc12fee2e65cb02a6754cf5e09
1 /* AddressSanitizer, a fast memory error detector.
2 Copyright (C) 2012-2013 Free Software Foundation, Inc.
3 Contributed by Kostya Serebryany <kcc@google.com>
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 3, or (at your option) any later
10 version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING3. If not see
19 <http://www.gnu.org/licenses/>. */
22 #include "config.h"
23 #include "system.h"
24 #include "coretypes.h"
25 #include "gimple.h"
26 #include "tree-iterator.h"
27 #include "tree-flow.h"
28 #include "tree-pass.h"
29 #include "asan.h"
30 #include "gimple-pretty-print.h"
31 #include "target.h"
32 #include "expr.h"
33 #include "optabs.h"
34 #include "output.h"
35 #include "tm_p.h"
36 #include "langhooks.h"
37 #include "hash-table.h"
38 #include "alloc-pool.h"
40 /* AddressSanitizer finds out-of-bounds and use-after-free bugs
41 with <2x slowdown on average.
43 The tool consists of two parts:
44 instrumentation module (this file) and a run-time library.
45 The instrumentation module adds a run-time check before every memory insn.
46 For a 8- or 16- byte load accessing address X:
47 ShadowAddr = (X >> 3) + Offset
48 ShadowValue = *(char*)ShadowAddr; // *(short*) for 16-byte access.
49 if (ShadowValue)
50 __asan_report_load8(X);
51 For a load of N bytes (N=1, 2 or 4) from address X:
52 ShadowAddr = (X >> 3) + Offset
53 ShadowValue = *(char*)ShadowAddr;
54 if (ShadowValue)
55 if ((X & 7) + N - 1 > ShadowValue)
56 __asan_report_loadN(X);
57 Stores are instrumented similarly, but using __asan_report_storeN functions.
58 A call too __asan_init() is inserted to the list of module CTORs.
60 The run-time library redefines malloc (so that redzone are inserted around
61 the allocated memory) and free (so that reuse of free-ed memory is delayed),
62 provides __asan_report* and __asan_init functions.
64 Read more:
65 http://code.google.com/p/address-sanitizer/wiki/AddressSanitizerAlgorithm
67 The current implementation supports detection of out-of-bounds and
68 use-after-free in the heap, on the stack and for global variables.
70 [Protection of stack variables]
72 To understand how detection of out-of-bounds and use-after-free works
73 for stack variables, lets look at this example on x86_64 where the
74 stack grows downward:
76 int
77 foo ()
79 char a[23] = {0};
80 int b[2] = {0};
82 a[5] = 1;
83 b[1] = 2;
85 return a[5] + b[1];
88 For this function, the stack protected by asan will be organized as
89 follows, from the top of the stack to the bottom:
91 Slot 1/ [red zone of 32 bytes called 'RIGHT RedZone']
93 Slot 2/ [8 bytes of red zone, that adds up to the space of 'a' to make
94 the next slot be 32 bytes aligned; this one is called Partial
95 Redzone; this 32 bytes alignment is an asan constraint]
97 Slot 3/ [24 bytes for variable 'a']
99 Slot 4/ [red zone of 32 bytes called 'Middle RedZone']
101 Slot 5/ [24 bytes of Partial Red Zone (similar to slot 2]
103 Slot 6/ [8 bytes for variable 'b']
105 Slot 7/ [32 bytes of Red Zone at the bottom of the stack, called
106 'LEFT RedZone']
108 The 32 bytes of LEFT red zone at the bottom of the stack can be
109 decomposed as such:
111 1/ The first 8 bytes contain a magical asan number that is always
112 0x41B58AB3.
114 2/ The following 8 bytes contains a pointer to a string (to be
115 parsed at runtime by the runtime asan library), which format is
116 the following:
118 "<function-name> <space> <num-of-variables-on-the-stack>
119 (<32-bytes-aligned-offset-in-bytes-of-variable> <space>
120 <length-of-var-in-bytes> ){n} "
122 where '(...){n}' means the content inside the parenthesis occurs 'n'
123 times, with 'n' being the number of variables on the stack.
125 3/ The following 16 bytes of the red zone have no particular
126 format.
128 The shadow memory for that stack layout is going to look like this:
130 - content of shadow memory 8 bytes for slot 7: 0xF1F1F1F1.
131 The F1 byte pattern is a magic number called
132 ASAN_STACK_MAGIC_LEFT and is a way for the runtime to know that
133 the memory for that shadow byte is part of a the LEFT red zone
134 intended to seat at the bottom of the variables on the stack.
136 - content of shadow memory 8 bytes for slots 6 and 5:
137 0xF4F4F400. The F4 byte pattern is a magic number
138 called ASAN_STACK_MAGIC_PARTIAL. It flags the fact that the
139 memory region for this shadow byte is a PARTIAL red zone
140 intended to pad a variable A, so that the slot following
141 {A,padding} is 32 bytes aligned.
143 Note that the fact that the least significant byte of this
144 shadow memory content is 00 means that 8 bytes of its
145 corresponding memory (which corresponds to the memory of
146 variable 'b') is addressable.
148 - content of shadow memory 8 bytes for slot 4: 0xF2F2F2F2.
149 The F2 byte pattern is a magic number called
150 ASAN_STACK_MAGIC_MIDDLE. It flags the fact that the memory
151 region for this shadow byte is a MIDDLE red zone intended to
152 seat between two 32 aligned slots of {variable,padding}.
154 - content of shadow memory 8 bytes for slot 3 and 2:
155 0xF4000000. This represents is the concatenation of
156 variable 'a' and the partial red zone following it, like what we
157 had for variable 'b'. The least significant 3 bytes being 00
158 means that the 3 bytes of variable 'a' are addressable.
160 - content of shadow memory 8 bytes for slot 1: 0xF3F3F3F3.
161 The F3 byte pattern is a magic number called
162 ASAN_STACK_MAGIC_RIGHT. It flags the fact that the memory
163 region for this shadow byte is a RIGHT red zone intended to seat
164 at the top of the variables of the stack.
166 Note that the real variable layout is done in expand_used_vars in
167 cfgexpand.c. As far as Address Sanitizer is concerned, it lays out
168 stack variables as well as the different red zones, emits some
169 prologue code to populate the shadow memory as to poison (mark as
170 non-accessible) the regions of the red zones and mark the regions of
171 stack variables as accessible, and emit some epilogue code to
172 un-poison (mark as accessible) the regions of red zones right before
173 the function exits.
175 [Protection of global variables]
177 The basic idea is to insert a red zone between two global variables
178 and install a constructor function that calls the asan runtime to do
179 the populating of the relevant shadow memory regions at load time.
181 So the global variables are laid out as to insert a red zone between
182 them. The size of the red zones is so that each variable starts on a
183 32 bytes boundary.
185 Then a constructor function is installed so that, for each global
186 variable, it calls the runtime asan library function
187 __asan_register_globals_with an instance of this type:
189 struct __asan_global
191 // Address of the beginning of the global variable.
192 const void *__beg;
194 // Initial size of the global variable.
195 uptr __size;
197 // Size of the global variable + size of the red zone. This
198 // size is 32 bytes aligned.
199 uptr __size_with_redzone;
201 // Name of the global variable.
202 const void *__name;
204 // This is always set to NULL for now.
205 uptr __has_dynamic_init;
208 A destructor function that calls the runtime asan library function
209 _asan_unregister_globals is also installed. */
211 alias_set_type asan_shadow_set = -1;
213 /* Pointer types to 1 resp. 2 byte integers in shadow memory. A separate
214 alias set is used for all shadow memory accesses. */
215 static GTY(()) tree shadow_ptr_types[2];
217 /* Hashtable support for memory references used by gimple
218 statements. */
220 /* This type represents a reference to a memory region. */
221 struct asan_mem_ref
223 /* The expression of the begining of the memory region. */
224 tree start;
226 /* The size of the access (can be 1, 2, 4, 8, 16 for now). */
227 char access_size;
230 static alloc_pool asan_mem_ref_alloc_pool;
232 /* This creates the alloc pool used to store the instances of
233 asan_mem_ref that are stored in the hash table asan_mem_ref_ht. */
235 static alloc_pool
236 asan_mem_ref_get_alloc_pool ()
238 if (asan_mem_ref_alloc_pool == NULL)
239 asan_mem_ref_alloc_pool = create_alloc_pool ("asan_mem_ref",
240 sizeof (asan_mem_ref),
241 10);
242 return asan_mem_ref_alloc_pool;
246 /* Initializes an instance of asan_mem_ref. */
248 static void
249 asan_mem_ref_init (asan_mem_ref *ref, tree start, char access_size)
251 ref->start = start;
252 ref->access_size = access_size;
255 /* Allocates memory for an instance of asan_mem_ref into the memory
256 pool returned by asan_mem_ref_get_alloc_pool and initialize it.
257 START is the address of (or the expression pointing to) the
258 beginning of memory reference. ACCESS_SIZE is the size of the
259 access to the referenced memory. */
261 static asan_mem_ref*
262 asan_mem_ref_new (tree start, char access_size)
264 asan_mem_ref *ref =
265 (asan_mem_ref *) pool_alloc (asan_mem_ref_get_alloc_pool ());
267 asan_mem_ref_init (ref, start, access_size);
268 return ref;
271 /* This builds and returns a pointer to the end of the memory region
272 that starts at START and of length LEN. */
274 tree
275 asan_mem_ref_get_end (tree start, tree len)
277 if (len == NULL_TREE || integer_zerop (len))
278 return start;
280 return fold_build2 (POINTER_PLUS_EXPR, TREE_TYPE (start), start, len);
283 /* Return a tree expression that represents the end of the referenced
284 memory region. Beware that this function can actually build a new
285 tree expression. */
287 tree
288 asan_mem_ref_get_end (const asan_mem_ref *ref, tree len)
290 return asan_mem_ref_get_end (ref->start, len);
293 struct asan_mem_ref_hasher
294 : typed_noop_remove <asan_mem_ref>
296 typedef asan_mem_ref value_type;
297 typedef asan_mem_ref compare_type;
299 static inline hashval_t hash (const value_type *);
300 static inline bool equal (const value_type *, const compare_type *);
303 /* Hash a memory reference. */
305 inline hashval_t
306 asan_mem_ref_hasher::hash (const asan_mem_ref *mem_ref)
308 hashval_t h = iterative_hash_expr (mem_ref->start, 0);
309 h = iterative_hash_hashval_t (h, mem_ref->access_size);
310 return h;
313 /* Compare two memory references. We accept the length of either
314 memory references to be NULL_TREE. */
316 inline bool
317 asan_mem_ref_hasher::equal (const asan_mem_ref *m1,
318 const asan_mem_ref *m2)
320 return (m1->access_size == m2->access_size
321 && operand_equal_p (m1->start, m2->start, 0));
324 static hash_table <asan_mem_ref_hasher> asan_mem_ref_ht;
326 /* Returns a reference to the hash table containing memory references.
327 This function ensures that the hash table is created. Note that
328 this hash table is updated by the function
329 update_mem_ref_hash_table. */
331 static hash_table <asan_mem_ref_hasher> &
332 get_mem_ref_hash_table ()
334 if (!asan_mem_ref_ht.is_created ())
335 asan_mem_ref_ht.create (10);
337 return asan_mem_ref_ht;
340 /* Clear all entries from the memory references hash table. */
342 static void
343 empty_mem_ref_hash_table ()
345 if (asan_mem_ref_ht.is_created ())
346 asan_mem_ref_ht.empty ();
349 /* Free the memory references hash table. */
351 static void
352 free_mem_ref_resources ()
354 if (asan_mem_ref_ht.is_created ())
355 asan_mem_ref_ht.dispose ();
357 if (asan_mem_ref_alloc_pool)
359 free_alloc_pool (asan_mem_ref_alloc_pool);
360 asan_mem_ref_alloc_pool = NULL;
364 /* Return true iff the memory reference REF has been instrumented. */
366 static bool
367 has_mem_ref_been_instrumented (tree ref, char access_size)
369 asan_mem_ref r;
370 asan_mem_ref_init (&r, ref, access_size);
372 return (get_mem_ref_hash_table ().find (&r) != NULL);
375 /* Return true iff the memory reference REF has been instrumented. */
377 static bool
378 has_mem_ref_been_instrumented (const asan_mem_ref *ref)
380 return has_mem_ref_been_instrumented (ref->start, ref->access_size);
383 /* Return true iff access to memory region starting at REF and of
384 length LEN has been instrumented. */
386 static bool
387 has_mem_ref_been_instrumented (const asan_mem_ref *ref, tree len)
389 /* First let's see if the address of the beginning of REF has been
390 instrumented. */
391 if (!has_mem_ref_been_instrumented (ref))
392 return false;
394 if (len != 0)
396 /* Let's see if the end of the region has been instrumented. */
397 if (!has_mem_ref_been_instrumented (asan_mem_ref_get_end (ref, len),
398 ref->access_size))
399 return false;
401 return true;
404 /* Set REF to the memory reference present in a gimple assignment
405 ASSIGNMENT. Return true upon successful completion, false
406 otherwise. */
408 static bool
409 get_mem_ref_of_assignment (const gimple assignment,
410 asan_mem_ref *ref,
411 bool *ref_is_store)
413 gcc_assert (gimple_assign_single_p (assignment));
415 if (gimple_store_p (assignment))
417 ref->start = gimple_assign_lhs (assignment);
418 *ref_is_store = true;
420 else if (gimple_assign_load_p (assignment))
422 ref->start = gimple_assign_rhs1 (assignment);
423 *ref_is_store = false;
425 else
426 return false;
428 ref->access_size = int_size_in_bytes (TREE_TYPE (ref->start));
429 return true;
432 /* Return the memory references contained in a gimple statement
433 representing a builtin call that has to do with memory access. */
435 static bool
436 get_mem_refs_of_builtin_call (const gimple call,
437 asan_mem_ref *src0,
438 tree *src0_len,
439 bool *src0_is_store,
440 asan_mem_ref *src1,
441 tree *src1_len,
442 bool *src1_is_store,
443 asan_mem_ref *dst,
444 tree *dst_len,
445 bool *dst_is_store,
446 bool *dest_is_deref)
448 gcc_checking_assert (gimple_call_builtin_p (call, BUILT_IN_NORMAL));
450 tree callee = gimple_call_fndecl (call);
451 tree source0 = NULL_TREE, source1 = NULL_TREE,
452 dest = NULL_TREE, len = NULL_TREE;
453 bool is_store = true, got_reference_p = false;
454 char access_size = 1;
456 switch (DECL_FUNCTION_CODE (callee))
458 /* (s, s, n) style memops. */
459 case BUILT_IN_BCMP:
460 case BUILT_IN_MEMCMP:
461 source0 = gimple_call_arg (call, 0);
462 source1 = gimple_call_arg (call, 1);
463 len = gimple_call_arg (call, 2);
464 break;
466 /* (src, dest, n) style memops. */
467 case BUILT_IN_BCOPY:
468 source0 = gimple_call_arg (call, 0);
469 dest = gimple_call_arg (call, 1);
470 len = gimple_call_arg (call, 2);
471 break;
473 /* (dest, src, n) style memops. */
474 case BUILT_IN_MEMCPY:
475 case BUILT_IN_MEMCPY_CHK:
476 case BUILT_IN_MEMMOVE:
477 case BUILT_IN_MEMMOVE_CHK:
478 case BUILT_IN_MEMPCPY:
479 case BUILT_IN_MEMPCPY_CHK:
480 dest = gimple_call_arg (call, 0);
481 source0 = gimple_call_arg (call, 1);
482 len = gimple_call_arg (call, 2);
483 break;
485 /* (dest, n) style memops. */
486 case BUILT_IN_BZERO:
487 dest = gimple_call_arg (call, 0);
488 len = gimple_call_arg (call, 1);
489 break;
491 /* (dest, x, n) style memops*/
492 case BUILT_IN_MEMSET:
493 case BUILT_IN_MEMSET_CHK:
494 dest = gimple_call_arg (call, 0);
495 len = gimple_call_arg (call, 2);
496 break;
498 case BUILT_IN_STRLEN:
499 source0 = gimple_call_arg (call, 0);
500 len = gimple_call_lhs (call);
501 break ;
503 /* And now the __atomic* and __sync builtins.
504 These are handled differently from the classical memory memory
505 access builtins above. */
507 case BUILT_IN_ATOMIC_LOAD_1:
508 case BUILT_IN_ATOMIC_LOAD_2:
509 case BUILT_IN_ATOMIC_LOAD_4:
510 case BUILT_IN_ATOMIC_LOAD_8:
511 case BUILT_IN_ATOMIC_LOAD_16:
512 is_store = false;
513 /* fall through. */
515 case BUILT_IN_SYNC_FETCH_AND_ADD_1:
516 case BUILT_IN_SYNC_FETCH_AND_ADD_2:
517 case BUILT_IN_SYNC_FETCH_AND_ADD_4:
518 case BUILT_IN_SYNC_FETCH_AND_ADD_8:
519 case BUILT_IN_SYNC_FETCH_AND_ADD_16:
521 case BUILT_IN_SYNC_FETCH_AND_SUB_1:
522 case BUILT_IN_SYNC_FETCH_AND_SUB_2:
523 case BUILT_IN_SYNC_FETCH_AND_SUB_4:
524 case BUILT_IN_SYNC_FETCH_AND_SUB_8:
525 case BUILT_IN_SYNC_FETCH_AND_SUB_16:
527 case BUILT_IN_SYNC_FETCH_AND_OR_1:
528 case BUILT_IN_SYNC_FETCH_AND_OR_2:
529 case BUILT_IN_SYNC_FETCH_AND_OR_4:
530 case BUILT_IN_SYNC_FETCH_AND_OR_8:
531 case BUILT_IN_SYNC_FETCH_AND_OR_16:
533 case BUILT_IN_SYNC_FETCH_AND_AND_1:
534 case BUILT_IN_SYNC_FETCH_AND_AND_2:
535 case BUILT_IN_SYNC_FETCH_AND_AND_4:
536 case BUILT_IN_SYNC_FETCH_AND_AND_8:
537 case BUILT_IN_SYNC_FETCH_AND_AND_16:
539 case BUILT_IN_SYNC_FETCH_AND_XOR_1:
540 case BUILT_IN_SYNC_FETCH_AND_XOR_2:
541 case BUILT_IN_SYNC_FETCH_AND_XOR_4:
542 case BUILT_IN_SYNC_FETCH_AND_XOR_8:
543 case BUILT_IN_SYNC_FETCH_AND_XOR_16:
545 case BUILT_IN_SYNC_FETCH_AND_NAND_1:
546 case BUILT_IN_SYNC_FETCH_AND_NAND_2:
547 case BUILT_IN_SYNC_FETCH_AND_NAND_4:
548 case BUILT_IN_SYNC_FETCH_AND_NAND_8:
550 case BUILT_IN_SYNC_ADD_AND_FETCH_1:
551 case BUILT_IN_SYNC_ADD_AND_FETCH_2:
552 case BUILT_IN_SYNC_ADD_AND_FETCH_4:
553 case BUILT_IN_SYNC_ADD_AND_FETCH_8:
554 case BUILT_IN_SYNC_ADD_AND_FETCH_16:
556 case BUILT_IN_SYNC_SUB_AND_FETCH_1:
557 case BUILT_IN_SYNC_SUB_AND_FETCH_2:
558 case BUILT_IN_SYNC_SUB_AND_FETCH_4:
559 case BUILT_IN_SYNC_SUB_AND_FETCH_8:
560 case BUILT_IN_SYNC_SUB_AND_FETCH_16:
562 case BUILT_IN_SYNC_OR_AND_FETCH_1:
563 case BUILT_IN_SYNC_OR_AND_FETCH_2:
564 case BUILT_IN_SYNC_OR_AND_FETCH_4:
565 case BUILT_IN_SYNC_OR_AND_FETCH_8:
566 case BUILT_IN_SYNC_OR_AND_FETCH_16:
568 case BUILT_IN_SYNC_AND_AND_FETCH_1:
569 case BUILT_IN_SYNC_AND_AND_FETCH_2:
570 case BUILT_IN_SYNC_AND_AND_FETCH_4:
571 case BUILT_IN_SYNC_AND_AND_FETCH_8:
572 case BUILT_IN_SYNC_AND_AND_FETCH_16:
574 case BUILT_IN_SYNC_XOR_AND_FETCH_1:
575 case BUILT_IN_SYNC_XOR_AND_FETCH_2:
576 case BUILT_IN_SYNC_XOR_AND_FETCH_4:
577 case BUILT_IN_SYNC_XOR_AND_FETCH_8:
578 case BUILT_IN_SYNC_XOR_AND_FETCH_16:
580 case BUILT_IN_SYNC_NAND_AND_FETCH_1:
581 case BUILT_IN_SYNC_NAND_AND_FETCH_2:
582 case BUILT_IN_SYNC_NAND_AND_FETCH_4:
583 case BUILT_IN_SYNC_NAND_AND_FETCH_8:
585 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_1:
586 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_2:
587 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_4:
588 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_8:
589 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_16:
591 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_1:
592 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_2:
593 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_4:
594 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_8:
595 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_16:
597 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_1:
598 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_2:
599 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_4:
600 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_8:
601 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_16:
603 case BUILT_IN_SYNC_LOCK_RELEASE_1:
604 case BUILT_IN_SYNC_LOCK_RELEASE_2:
605 case BUILT_IN_SYNC_LOCK_RELEASE_4:
606 case BUILT_IN_SYNC_LOCK_RELEASE_8:
607 case BUILT_IN_SYNC_LOCK_RELEASE_16:
609 case BUILT_IN_ATOMIC_EXCHANGE_1:
610 case BUILT_IN_ATOMIC_EXCHANGE_2:
611 case BUILT_IN_ATOMIC_EXCHANGE_4:
612 case BUILT_IN_ATOMIC_EXCHANGE_8:
613 case BUILT_IN_ATOMIC_EXCHANGE_16:
615 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_1:
616 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_2:
617 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_4:
618 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_8:
619 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_16:
621 case BUILT_IN_ATOMIC_STORE_1:
622 case BUILT_IN_ATOMIC_STORE_2:
623 case BUILT_IN_ATOMIC_STORE_4:
624 case BUILT_IN_ATOMIC_STORE_8:
625 case BUILT_IN_ATOMIC_STORE_16:
627 case BUILT_IN_ATOMIC_ADD_FETCH_1:
628 case BUILT_IN_ATOMIC_ADD_FETCH_2:
629 case BUILT_IN_ATOMIC_ADD_FETCH_4:
630 case BUILT_IN_ATOMIC_ADD_FETCH_8:
631 case BUILT_IN_ATOMIC_ADD_FETCH_16:
633 case BUILT_IN_ATOMIC_SUB_FETCH_1:
634 case BUILT_IN_ATOMIC_SUB_FETCH_2:
635 case BUILT_IN_ATOMIC_SUB_FETCH_4:
636 case BUILT_IN_ATOMIC_SUB_FETCH_8:
637 case BUILT_IN_ATOMIC_SUB_FETCH_16:
639 case BUILT_IN_ATOMIC_AND_FETCH_1:
640 case BUILT_IN_ATOMIC_AND_FETCH_2:
641 case BUILT_IN_ATOMIC_AND_FETCH_4:
642 case BUILT_IN_ATOMIC_AND_FETCH_8:
643 case BUILT_IN_ATOMIC_AND_FETCH_16:
645 case BUILT_IN_ATOMIC_NAND_FETCH_1:
646 case BUILT_IN_ATOMIC_NAND_FETCH_2:
647 case BUILT_IN_ATOMIC_NAND_FETCH_4:
648 case BUILT_IN_ATOMIC_NAND_FETCH_8:
649 case BUILT_IN_ATOMIC_NAND_FETCH_16:
651 case BUILT_IN_ATOMIC_XOR_FETCH_1:
652 case BUILT_IN_ATOMIC_XOR_FETCH_2:
653 case BUILT_IN_ATOMIC_XOR_FETCH_4:
654 case BUILT_IN_ATOMIC_XOR_FETCH_8:
655 case BUILT_IN_ATOMIC_XOR_FETCH_16:
657 case BUILT_IN_ATOMIC_OR_FETCH_1:
658 case BUILT_IN_ATOMIC_OR_FETCH_2:
659 case BUILT_IN_ATOMIC_OR_FETCH_4:
660 case BUILT_IN_ATOMIC_OR_FETCH_8:
661 case BUILT_IN_ATOMIC_OR_FETCH_16:
663 case BUILT_IN_ATOMIC_FETCH_ADD_1:
664 case BUILT_IN_ATOMIC_FETCH_ADD_2:
665 case BUILT_IN_ATOMIC_FETCH_ADD_4:
666 case BUILT_IN_ATOMIC_FETCH_ADD_8:
667 case BUILT_IN_ATOMIC_FETCH_ADD_16:
669 case BUILT_IN_ATOMIC_FETCH_SUB_1:
670 case BUILT_IN_ATOMIC_FETCH_SUB_2:
671 case BUILT_IN_ATOMIC_FETCH_SUB_4:
672 case BUILT_IN_ATOMIC_FETCH_SUB_8:
673 case BUILT_IN_ATOMIC_FETCH_SUB_16:
675 case BUILT_IN_ATOMIC_FETCH_AND_1:
676 case BUILT_IN_ATOMIC_FETCH_AND_2:
677 case BUILT_IN_ATOMIC_FETCH_AND_4:
678 case BUILT_IN_ATOMIC_FETCH_AND_8:
679 case BUILT_IN_ATOMIC_FETCH_AND_16:
681 case BUILT_IN_ATOMIC_FETCH_NAND_1:
682 case BUILT_IN_ATOMIC_FETCH_NAND_2:
683 case BUILT_IN_ATOMIC_FETCH_NAND_4:
684 case BUILT_IN_ATOMIC_FETCH_NAND_8:
685 case BUILT_IN_ATOMIC_FETCH_NAND_16:
687 case BUILT_IN_ATOMIC_FETCH_XOR_1:
688 case BUILT_IN_ATOMIC_FETCH_XOR_2:
689 case BUILT_IN_ATOMIC_FETCH_XOR_4:
690 case BUILT_IN_ATOMIC_FETCH_XOR_8:
691 case BUILT_IN_ATOMIC_FETCH_XOR_16:
693 case BUILT_IN_ATOMIC_FETCH_OR_1:
694 case BUILT_IN_ATOMIC_FETCH_OR_2:
695 case BUILT_IN_ATOMIC_FETCH_OR_4:
696 case BUILT_IN_ATOMIC_FETCH_OR_8:
697 case BUILT_IN_ATOMIC_FETCH_OR_16:
699 dest = gimple_call_arg (call, 0);
700 /* DEST represents the address of a memory location.
701 instrument_derefs wants the memory location, so lets
702 dereference the address DEST before handing it to
703 instrument_derefs. */
704 if (TREE_CODE (dest) == ADDR_EXPR)
705 dest = TREE_OPERAND (dest, 0);
706 else if (TREE_CODE (dest) == SSA_NAME)
707 dest = build2 (MEM_REF, TREE_TYPE (TREE_TYPE (dest)),
708 dest, build_int_cst (TREE_TYPE (dest), 0));
709 else
710 gcc_unreachable ();
712 access_size = int_size_in_bytes (TREE_TYPE (dest));
715 default:
716 /* The other builtins memory access are not instrumented in this
717 function because they either don't have any length parameter,
718 or their length parameter is just a limit. */
719 break;
722 if (len != NULL_TREE)
724 if (source0 != NULL_TREE)
726 src0->start = source0;
727 src0->access_size = access_size;
728 *src0_len = len;
729 *src0_is_store = false;
732 if (source1 != NULL_TREE)
734 src1->start = source1;
735 src1->access_size = access_size;
736 *src1_len = len;
737 *src1_is_store = false;
740 if (dest != NULL_TREE)
742 dst->start = dest;
743 dst->access_size = access_size;
744 *dst_len = len;
745 *dst_is_store = true;
748 got_reference_p = true;
750 else if (dest)
752 dst->start = dest;
753 dst->access_size = access_size;
754 *dst_len = NULL_TREE;
755 *dst_is_store = is_store;
756 *dest_is_deref = true;
757 got_reference_p = true;
760 return got_reference_p;
763 /* Return true iff a given gimple statement has been instrumented.
764 Note that the statement is "defined" by the memory references it
765 contains. */
767 static bool
768 has_stmt_been_instrumented_p (gimple stmt)
770 if (gimple_assign_single_p (stmt))
772 bool r_is_store;
773 asan_mem_ref r;
774 asan_mem_ref_init (&r, NULL, 1);
776 if (get_mem_ref_of_assignment (stmt, &r, &r_is_store))
777 return has_mem_ref_been_instrumented (&r);
779 else if (gimple_call_builtin_p (stmt, BUILT_IN_NORMAL))
781 asan_mem_ref src0, src1, dest;
782 asan_mem_ref_init (&src0, NULL, 1);
783 asan_mem_ref_init (&src1, NULL, 1);
784 asan_mem_ref_init (&dest, NULL, 1);
786 tree src0_len = NULL_TREE, src1_len = NULL_TREE, dest_len = NULL_TREE;
787 bool src0_is_store = false, src1_is_store = false,
788 dest_is_store = false, dest_is_deref = false;
789 if (get_mem_refs_of_builtin_call (stmt,
790 &src0, &src0_len, &src0_is_store,
791 &src1, &src1_len, &src1_is_store,
792 &dest, &dest_len, &dest_is_store,
793 &dest_is_deref))
795 if (src0.start != NULL_TREE
796 && !has_mem_ref_been_instrumented (&src0, src0_len))
797 return false;
799 if (src1.start != NULL_TREE
800 && !has_mem_ref_been_instrumented (&src1, src1_len))
801 return false;
803 if (dest.start != NULL_TREE
804 && !has_mem_ref_been_instrumented (&dest, dest_len))
805 return false;
807 return true;
810 return false;
813 /* Insert a memory reference into the hash table. */
815 static void
816 update_mem_ref_hash_table (tree ref, char access_size)
818 hash_table <asan_mem_ref_hasher> ht = get_mem_ref_hash_table ();
820 asan_mem_ref r;
821 asan_mem_ref_init (&r, ref, access_size);
823 asan_mem_ref **slot = ht.find_slot (&r, INSERT);
824 if (*slot == NULL)
825 *slot = asan_mem_ref_new (ref, access_size);
828 /* Initialize shadow_ptr_types array. */
830 static void
831 asan_init_shadow_ptr_types (void)
833 asan_shadow_set = new_alias_set ();
834 shadow_ptr_types[0] = build_distinct_type_copy (signed_char_type_node);
835 TYPE_ALIAS_SET (shadow_ptr_types[0]) = asan_shadow_set;
836 shadow_ptr_types[0] = build_pointer_type (shadow_ptr_types[0]);
837 shadow_ptr_types[1] = build_distinct_type_copy (short_integer_type_node);
838 TYPE_ALIAS_SET (shadow_ptr_types[1]) = asan_shadow_set;
839 shadow_ptr_types[1] = build_pointer_type (shadow_ptr_types[1]);
840 initialize_sanitizer_builtins ();
843 /* Asan pretty-printer, used for buidling of the description STRING_CSTs. */
844 static pretty_printer asan_pp;
845 static bool asan_pp_initialized;
847 /* Initialize asan_pp. */
849 static void
850 asan_pp_initialize (void)
852 pp_construct (&asan_pp, /* prefix */NULL, /* line-width */0);
853 asan_pp_initialized = true;
856 /* Create ADDR_EXPR of STRING_CST with asan_pp text. */
858 static tree
859 asan_pp_string (void)
861 const char *buf = pp_base_formatted_text (&asan_pp);
862 size_t len = strlen (buf);
863 tree ret = build_string (len + 1, buf);
864 TREE_TYPE (ret)
865 = build_array_type (TREE_TYPE (shadow_ptr_types[0]),
866 build_index_type (size_int (len)));
867 TREE_READONLY (ret) = 1;
868 TREE_STATIC (ret) = 1;
869 return build1 (ADDR_EXPR, shadow_ptr_types[0], ret);
872 /* Return a CONST_INT representing 4 subsequent shadow memory bytes. */
874 static rtx
875 asan_shadow_cst (unsigned char shadow_bytes[4])
877 int i;
878 unsigned HOST_WIDE_INT val = 0;
879 gcc_assert (WORDS_BIG_ENDIAN == BYTES_BIG_ENDIAN);
880 for (i = 0; i < 4; i++)
881 val |= (unsigned HOST_WIDE_INT) shadow_bytes[BYTES_BIG_ENDIAN ? 3 - i : i]
882 << (BITS_PER_UNIT * i);
883 return GEN_INT (trunc_int_for_mode (val, SImode));
886 /* Clear shadow memory at SHADOW_MEM, LEN bytes. Can't call a library call here
887 though. */
889 static void
890 asan_clear_shadow (rtx shadow_mem, HOST_WIDE_INT len)
892 rtx insn, insns, top_label, end, addr, tmp, jump;
894 start_sequence ();
895 clear_storage (shadow_mem, GEN_INT (len), BLOCK_OP_NORMAL);
896 insns = get_insns ();
897 end_sequence ();
898 for (insn = insns; insn; insn = NEXT_INSN (insn))
899 if (CALL_P (insn))
900 break;
901 if (insn == NULL_RTX)
903 emit_insn (insns);
904 return;
907 gcc_assert ((len & 3) == 0);
908 top_label = gen_label_rtx ();
909 addr = force_reg (Pmode, XEXP (shadow_mem, 0));
910 shadow_mem = adjust_automodify_address (shadow_mem, SImode, addr, 0);
911 end = force_reg (Pmode, plus_constant (Pmode, addr, len));
912 emit_label (top_label);
914 emit_move_insn (shadow_mem, const0_rtx);
915 tmp = expand_simple_binop (Pmode, PLUS, addr, GEN_INT (4), addr,
916 true, OPTAB_LIB_WIDEN);
917 if (tmp != addr)
918 emit_move_insn (addr, tmp);
919 emit_cmp_and_jump_insns (addr, end, LT, NULL_RTX, Pmode, true, top_label);
920 jump = get_last_insn ();
921 gcc_assert (JUMP_P (jump));
922 add_reg_note (jump, REG_BR_PROB, GEN_INT (REG_BR_PROB_BASE * 80 / 100));
925 /* Insert code to protect stack vars. The prologue sequence should be emitted
926 directly, epilogue sequence returned. BASE is the register holding the
927 stack base, against which OFFSETS array offsets are relative to, OFFSETS
928 array contains pairs of offsets in reverse order, always the end offset
929 of some gap that needs protection followed by starting offset,
930 and DECLS is an array of representative decls for each var partition.
931 LENGTH is the length of the OFFSETS array, DECLS array is LENGTH / 2 - 1
932 elements long (OFFSETS include gap before the first variable as well
933 as gaps after each stack variable). */
936 asan_emit_stack_protection (rtx base, HOST_WIDE_INT *offsets, tree *decls,
937 int length)
939 rtx shadow_base, shadow_mem, ret, mem;
940 unsigned char shadow_bytes[4];
941 HOST_WIDE_INT base_offset = offsets[length - 1], offset, prev_offset;
942 HOST_WIDE_INT last_offset, last_size;
943 int l;
944 unsigned char cur_shadow_byte = ASAN_STACK_MAGIC_LEFT;
945 tree str_cst;
947 if (shadow_ptr_types[0] == NULL_TREE)
948 asan_init_shadow_ptr_types ();
950 /* First of all, prepare the description string. */
951 if (!asan_pp_initialized)
952 asan_pp_initialize ();
954 pp_clear_output_area (&asan_pp);
955 if (DECL_NAME (current_function_decl))
956 pp_base_tree_identifier (&asan_pp, DECL_NAME (current_function_decl));
957 else
958 pp_string (&asan_pp, "<unknown>");
959 pp_space (&asan_pp);
960 pp_decimal_int (&asan_pp, length / 2 - 1);
961 pp_space (&asan_pp);
962 for (l = length - 2; l; l -= 2)
964 tree decl = decls[l / 2 - 1];
965 pp_wide_integer (&asan_pp, offsets[l] - base_offset);
966 pp_space (&asan_pp);
967 pp_wide_integer (&asan_pp, offsets[l - 1] - offsets[l]);
968 pp_space (&asan_pp);
969 if (DECL_P (decl) && DECL_NAME (decl))
971 pp_decimal_int (&asan_pp, IDENTIFIER_LENGTH (DECL_NAME (decl)));
972 pp_space (&asan_pp);
973 pp_base_tree_identifier (&asan_pp, DECL_NAME (decl));
975 else
976 pp_string (&asan_pp, "9 <unknown>");
977 pp_space (&asan_pp);
979 str_cst = asan_pp_string ();
981 /* Emit the prologue sequence. */
982 base = expand_binop (Pmode, add_optab, base, GEN_INT (base_offset),
983 NULL_RTX, 1, OPTAB_DIRECT);
984 mem = gen_rtx_MEM (ptr_mode, base);
985 emit_move_insn (mem, GEN_INT (ASAN_STACK_FRAME_MAGIC));
986 mem = adjust_address (mem, VOIDmode, GET_MODE_SIZE (ptr_mode));
987 emit_move_insn (mem, expand_normal (str_cst));
988 shadow_base = expand_binop (Pmode, lshr_optab, base,
989 GEN_INT (ASAN_SHADOW_SHIFT),
990 NULL_RTX, 1, OPTAB_DIRECT);
991 shadow_base = expand_binop (Pmode, add_optab, shadow_base,
992 GEN_INT (targetm.asan_shadow_offset ()),
993 NULL_RTX, 1, OPTAB_DIRECT);
994 gcc_assert (asan_shadow_set != -1
995 && (ASAN_RED_ZONE_SIZE >> ASAN_SHADOW_SHIFT) == 4);
996 shadow_mem = gen_rtx_MEM (SImode, shadow_base);
997 set_mem_alias_set (shadow_mem, asan_shadow_set);
998 prev_offset = base_offset;
999 for (l = length; l; l -= 2)
1001 if (l == 2)
1002 cur_shadow_byte = ASAN_STACK_MAGIC_RIGHT;
1003 offset = offsets[l - 1];
1004 if ((offset - base_offset) & (ASAN_RED_ZONE_SIZE - 1))
1006 int i;
1007 HOST_WIDE_INT aoff
1008 = base_offset + ((offset - base_offset)
1009 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1));
1010 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1011 (aoff - prev_offset)
1012 >> ASAN_SHADOW_SHIFT);
1013 prev_offset = aoff;
1014 for (i = 0; i < 4; i++, aoff += (1 << ASAN_SHADOW_SHIFT))
1015 if (aoff < offset)
1017 if (aoff < offset - (1 << ASAN_SHADOW_SHIFT) + 1)
1018 shadow_bytes[i] = 0;
1019 else
1020 shadow_bytes[i] = offset - aoff;
1022 else
1023 shadow_bytes[i] = ASAN_STACK_MAGIC_PARTIAL;
1024 emit_move_insn (shadow_mem, asan_shadow_cst (shadow_bytes));
1025 offset = aoff;
1027 while (offset <= offsets[l - 2] - ASAN_RED_ZONE_SIZE)
1029 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1030 (offset - prev_offset)
1031 >> ASAN_SHADOW_SHIFT);
1032 prev_offset = offset;
1033 memset (shadow_bytes, cur_shadow_byte, 4);
1034 emit_move_insn (shadow_mem, asan_shadow_cst (shadow_bytes));
1035 offset += ASAN_RED_ZONE_SIZE;
1037 cur_shadow_byte = ASAN_STACK_MAGIC_MIDDLE;
1039 do_pending_stack_adjust ();
1041 /* Construct epilogue sequence. */
1042 start_sequence ();
1044 shadow_mem = gen_rtx_MEM (BLKmode, shadow_base);
1045 set_mem_alias_set (shadow_mem, asan_shadow_set);
1046 prev_offset = base_offset;
1047 last_offset = base_offset;
1048 last_size = 0;
1049 for (l = length; l; l -= 2)
1051 offset = base_offset + ((offsets[l - 1] - base_offset)
1052 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1));
1053 if (last_offset + last_size != offset)
1055 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1056 (last_offset - prev_offset)
1057 >> ASAN_SHADOW_SHIFT);
1058 prev_offset = last_offset;
1059 asan_clear_shadow (shadow_mem, last_size >> ASAN_SHADOW_SHIFT);
1060 last_offset = offset;
1061 last_size = 0;
1063 last_size += base_offset + ((offsets[l - 2] - base_offset)
1064 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1))
1065 - offset;
1067 if (last_size)
1069 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1070 (last_offset - prev_offset)
1071 >> ASAN_SHADOW_SHIFT);
1072 asan_clear_shadow (shadow_mem, last_size >> ASAN_SHADOW_SHIFT);
1075 do_pending_stack_adjust ();
1077 ret = get_insns ();
1078 end_sequence ();
1079 return ret;
1082 /* Return true if DECL, a global var, might be overridden and needs
1083 therefore a local alias. */
1085 static bool
1086 asan_needs_local_alias (tree decl)
1088 return DECL_WEAK (decl) || !targetm.binds_local_p (decl);
1091 /* Return true if DECL is a VAR_DECL that should be protected
1092 by Address Sanitizer, by appending a red zone with protected
1093 shadow memory after it and aligning it to at least
1094 ASAN_RED_ZONE_SIZE bytes. */
1096 bool
1097 asan_protect_global (tree decl)
1099 rtx rtl, symbol;
1101 if (TREE_CODE (decl) == STRING_CST)
1103 /* Instrument all STRING_CSTs except those created
1104 by asan_pp_string here. */
1105 if (shadow_ptr_types[0] != NULL_TREE
1106 && TREE_CODE (TREE_TYPE (decl)) == ARRAY_TYPE
1107 && TREE_TYPE (TREE_TYPE (decl)) == TREE_TYPE (shadow_ptr_types[0]))
1108 return false;
1109 return true;
1111 if (TREE_CODE (decl) != VAR_DECL
1112 /* TLS vars aren't statically protectable. */
1113 || DECL_THREAD_LOCAL_P (decl)
1114 /* Externs will be protected elsewhere. */
1115 || DECL_EXTERNAL (decl)
1116 || !DECL_RTL_SET_P (decl)
1117 /* Comdat vars pose an ABI problem, we can't know if
1118 the var that is selected by the linker will have
1119 padding or not. */
1120 || DECL_ONE_ONLY (decl)
1121 /* Similarly for common vars. People can use -fno-common. */
1122 || (DECL_COMMON (decl) && TREE_PUBLIC (decl))
1123 /* Don't protect if using user section, often vars placed
1124 into user section from multiple TUs are then assumed
1125 to be an array of such vars, putting padding in there
1126 breaks this assumption. */
1127 || (DECL_SECTION_NAME (decl) != NULL_TREE
1128 && !DECL_HAS_IMPLICIT_SECTION_NAME_P (decl))
1129 || DECL_SIZE (decl) == 0
1130 || ASAN_RED_ZONE_SIZE * BITS_PER_UNIT > MAX_OFILE_ALIGNMENT
1131 || !valid_constant_size_p (DECL_SIZE_UNIT (decl))
1132 || DECL_ALIGN_UNIT (decl) > 2 * ASAN_RED_ZONE_SIZE)
1133 return false;
1135 rtl = DECL_RTL (decl);
1136 if (!MEM_P (rtl) || GET_CODE (XEXP (rtl, 0)) != SYMBOL_REF)
1137 return false;
1138 symbol = XEXP (rtl, 0);
1140 if (CONSTANT_POOL_ADDRESS_P (symbol)
1141 || TREE_CONSTANT_POOL_ADDRESS_P (symbol))
1142 return false;
1144 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl)))
1145 return false;
1147 #ifndef ASM_OUTPUT_DEF
1148 if (asan_needs_local_alias (decl))
1149 return false;
1150 #endif
1152 return true;
1155 /* Construct a function tree for __asan_report_{load,store}{1,2,4,8,16}.
1156 IS_STORE is either 1 (for a store) or 0 (for a load).
1157 SIZE_IN_BYTES is one of 1, 2, 4, 8, 16. */
1159 static tree
1160 report_error_func (bool is_store, int size_in_bytes)
1162 static enum built_in_function report[2][5]
1163 = { { BUILT_IN_ASAN_REPORT_LOAD1, BUILT_IN_ASAN_REPORT_LOAD2,
1164 BUILT_IN_ASAN_REPORT_LOAD4, BUILT_IN_ASAN_REPORT_LOAD8,
1165 BUILT_IN_ASAN_REPORT_LOAD16 },
1166 { BUILT_IN_ASAN_REPORT_STORE1, BUILT_IN_ASAN_REPORT_STORE2,
1167 BUILT_IN_ASAN_REPORT_STORE4, BUILT_IN_ASAN_REPORT_STORE8,
1168 BUILT_IN_ASAN_REPORT_STORE16 } };
1169 return builtin_decl_implicit (report[is_store][exact_log2 (size_in_bytes)]);
1172 #define PROB_VERY_UNLIKELY (REG_BR_PROB_BASE / 2000 - 1)
1173 #define PROB_ALWAYS (REG_BR_PROB_BASE)
1175 /* Split the current basic block and create a condition statement
1176 insertion point right before or after the statement pointed to by
1177 ITER. Return an iterator to the point at which the caller might
1178 safely insert the condition statement.
1180 THEN_BLOCK must be set to the address of an uninitialized instance
1181 of basic_block. The function will then set *THEN_BLOCK to the
1182 'then block' of the condition statement to be inserted by the
1183 caller.
1185 If CREATE_THEN_FALLTHRU_EDGE is false, no edge will be created from
1186 *THEN_BLOCK to *FALLTHROUGH_BLOCK.
1188 Similarly, the function will set *FALLTRHOUGH_BLOCK to the 'else
1189 block' of the condition statement to be inserted by the caller.
1191 Note that *FALLTHROUGH_BLOCK is a new block that contains the
1192 statements starting from *ITER, and *THEN_BLOCK is a new empty
1193 block.
1195 *ITER is adjusted to point to always point to the first statement
1196 of the basic block * FALLTHROUGH_BLOCK. That statement is the
1197 same as what ITER was pointing to prior to calling this function,
1198 if BEFORE_P is true; otherwise, it is its following statement. */
1200 static gimple_stmt_iterator
1201 create_cond_insert_point (gimple_stmt_iterator *iter,
1202 bool before_p,
1203 bool then_more_likely_p,
1204 bool create_then_fallthru_edge,
1205 basic_block *then_block,
1206 basic_block *fallthrough_block)
1208 gimple_stmt_iterator gsi = *iter;
1210 if (!gsi_end_p (gsi) && before_p)
1211 gsi_prev (&gsi);
1213 basic_block cur_bb = gsi_bb (*iter);
1215 edge e = split_block (cur_bb, gsi_stmt (gsi));
1217 /* Get a hold on the 'condition block', the 'then block' and the
1218 'else block'. */
1219 basic_block cond_bb = e->src;
1220 basic_block fallthru_bb = e->dest;
1221 basic_block then_bb = create_empty_bb (cond_bb);
1223 /* Set up the newly created 'then block'. */
1224 e = make_edge (cond_bb, then_bb, EDGE_TRUE_VALUE);
1225 int fallthrough_probability
1226 = then_more_likely_p
1227 ? PROB_VERY_UNLIKELY
1228 : PROB_ALWAYS - PROB_VERY_UNLIKELY;
1229 e->probability = PROB_ALWAYS - fallthrough_probability;
1230 if (create_then_fallthru_edge)
1231 make_single_succ_edge (then_bb, fallthru_bb, EDGE_FALLTHRU);
1233 /* Set up the fallthrough basic block. */
1234 e = find_edge (cond_bb, fallthru_bb);
1235 e->flags = EDGE_FALSE_VALUE;
1236 e->count = cond_bb->count;
1237 e->probability = fallthrough_probability;
1239 /* Update dominance info for the newly created then_bb; note that
1240 fallthru_bb's dominance info has already been updated by
1241 split_bock. */
1242 if (dom_info_available_p (CDI_DOMINATORS))
1243 set_immediate_dominator (CDI_DOMINATORS, then_bb, cond_bb);
1245 *then_block = then_bb;
1246 *fallthrough_block = fallthru_bb;
1247 *iter = gsi_start_bb (fallthru_bb);
1249 return gsi_last_bb (cond_bb);
1252 /* Insert an if condition followed by a 'then block' right before the
1253 statement pointed to by ITER. The fallthrough block -- which is the
1254 else block of the condition as well as the destination of the
1255 outcoming edge of the 'then block' -- starts with the statement
1256 pointed to by ITER.
1258 COND is the condition of the if.
1260 If THEN_MORE_LIKELY_P is true, the probability of the edge to the
1261 'then block' is higher than the probability of the edge to the
1262 fallthrough block.
1264 Upon completion of the function, *THEN_BB is set to the newly
1265 inserted 'then block' and similarly, *FALLTHROUGH_BB is set to the
1266 fallthrough block.
1268 *ITER is adjusted to still point to the same statement it was
1269 pointing to initially. */
1271 static void
1272 insert_if_then_before_iter (gimple cond,
1273 gimple_stmt_iterator *iter,
1274 bool then_more_likely_p,
1275 basic_block *then_bb,
1276 basic_block *fallthrough_bb)
1278 gimple_stmt_iterator cond_insert_point =
1279 create_cond_insert_point (iter,
1280 /*before_p=*/true,
1281 then_more_likely_p,
1282 /*create_then_fallthru_edge=*/true,
1283 then_bb,
1284 fallthrough_bb);
1285 gsi_insert_after (&cond_insert_point, cond, GSI_NEW_STMT);
1288 /* Instrument the memory access instruction BASE. Insert new
1289 statements before or after ITER.
1291 Note that the memory access represented by BASE can be either an
1292 SSA_NAME, or a non-SSA expression. LOCATION is the source code
1293 location. IS_STORE is TRUE for a store, FALSE for a load.
1294 BEFORE_P is TRUE for inserting the instrumentation code before
1295 ITER, FALSE for inserting it after ITER. SIZE_IN_BYTES is one of
1296 1, 2, 4, 8, 16.
1298 If BEFORE_P is TRUE, *ITER is arranged to still point to the
1299 statement it was pointing to prior to calling this function,
1300 otherwise, it points to the statement logically following it. */
1302 static void
1303 build_check_stmt (location_t location, tree base, gimple_stmt_iterator *iter,
1304 bool before_p, bool is_store, int size_in_bytes)
1306 gimple_stmt_iterator gsi;
1307 basic_block then_bb, else_bb;
1308 tree t, base_addr, shadow;
1309 gimple g;
1310 tree shadow_ptr_type = shadow_ptr_types[size_in_bytes == 16 ? 1 : 0];
1311 tree shadow_type = TREE_TYPE (shadow_ptr_type);
1312 tree uintptr_type
1313 = build_nonstandard_integer_type (TYPE_PRECISION (TREE_TYPE (base)), 1);
1314 tree base_ssa = base;
1316 /* Get an iterator on the point where we can add the condition
1317 statement for the instrumentation. */
1318 gsi = create_cond_insert_point (iter, before_p,
1319 /*then_more_likely_p=*/false,
1320 /*create_then_fallthru_edge=*/false,
1321 &then_bb,
1322 &else_bb);
1324 base = unshare_expr (base);
1326 /* BASE can already be an SSA_NAME; in that case, do not create a
1327 new SSA_NAME for it. */
1328 if (TREE_CODE (base) != SSA_NAME)
1330 g = gimple_build_assign_with_ops (TREE_CODE (base),
1331 make_ssa_name (TREE_TYPE (base), NULL),
1332 base, NULL_TREE);
1333 gimple_set_location (g, location);
1334 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1335 base_ssa = gimple_assign_lhs (g);
1338 g = gimple_build_assign_with_ops (NOP_EXPR,
1339 make_ssa_name (uintptr_type, NULL),
1340 base_ssa, NULL_TREE);
1341 gimple_set_location (g, location);
1342 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1343 base_addr = gimple_assign_lhs (g);
1345 /* Build
1346 (base_addr >> ASAN_SHADOW_SHIFT) + targetm.asan_shadow_offset (). */
1348 t = build_int_cst (uintptr_type, ASAN_SHADOW_SHIFT);
1349 g = gimple_build_assign_with_ops (RSHIFT_EXPR,
1350 make_ssa_name (uintptr_type, NULL),
1351 base_addr, t);
1352 gimple_set_location (g, location);
1353 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1355 t = build_int_cst (uintptr_type, targetm.asan_shadow_offset ());
1356 g = gimple_build_assign_with_ops (PLUS_EXPR,
1357 make_ssa_name (uintptr_type, NULL),
1358 gimple_assign_lhs (g), t);
1359 gimple_set_location (g, location);
1360 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1362 g = gimple_build_assign_with_ops (NOP_EXPR,
1363 make_ssa_name (shadow_ptr_type, NULL),
1364 gimple_assign_lhs (g), NULL_TREE);
1365 gimple_set_location (g, location);
1366 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1368 t = build2 (MEM_REF, shadow_type, gimple_assign_lhs (g),
1369 build_int_cst (shadow_ptr_type, 0));
1370 g = gimple_build_assign_with_ops (MEM_REF,
1371 make_ssa_name (shadow_type, NULL),
1372 t, NULL_TREE);
1373 gimple_set_location (g, location);
1374 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1375 shadow = gimple_assign_lhs (g);
1377 if (size_in_bytes < 8)
1379 /* Slow path for 1, 2 and 4 byte accesses.
1380 Test (shadow != 0)
1381 & ((base_addr & 7) + (size_in_bytes - 1)) >= shadow). */
1382 g = gimple_build_assign_with_ops (NE_EXPR,
1383 make_ssa_name (boolean_type_node,
1384 NULL),
1385 shadow,
1386 build_int_cst (shadow_type, 0));
1387 gimple_set_location (g, location);
1388 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1389 t = gimple_assign_lhs (g);
1391 g = gimple_build_assign_with_ops (BIT_AND_EXPR,
1392 make_ssa_name (uintptr_type,
1393 NULL),
1394 base_addr,
1395 build_int_cst (uintptr_type, 7));
1396 gimple_set_location (g, location);
1397 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1399 g = gimple_build_assign_with_ops (NOP_EXPR,
1400 make_ssa_name (shadow_type,
1401 NULL),
1402 gimple_assign_lhs (g), NULL_TREE);
1403 gimple_set_location (g, location);
1404 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1406 if (size_in_bytes > 1)
1408 g = gimple_build_assign_with_ops (PLUS_EXPR,
1409 make_ssa_name (shadow_type,
1410 NULL),
1411 gimple_assign_lhs (g),
1412 build_int_cst (shadow_type,
1413 size_in_bytes - 1));
1414 gimple_set_location (g, location);
1415 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1418 g = gimple_build_assign_with_ops (GE_EXPR,
1419 make_ssa_name (boolean_type_node,
1420 NULL),
1421 gimple_assign_lhs (g),
1422 shadow);
1423 gimple_set_location (g, location);
1424 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1426 g = gimple_build_assign_with_ops (BIT_AND_EXPR,
1427 make_ssa_name (boolean_type_node,
1428 NULL),
1429 t, gimple_assign_lhs (g));
1430 gimple_set_location (g, location);
1431 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1432 t = gimple_assign_lhs (g);
1434 else
1435 t = shadow;
1437 g = gimple_build_cond (NE_EXPR, t, build_int_cst (TREE_TYPE (t), 0),
1438 NULL_TREE, NULL_TREE);
1439 gimple_set_location (g, location);
1440 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1442 /* Generate call to the run-time library (e.g. __asan_report_load8). */
1443 gsi = gsi_start_bb (then_bb);
1444 g = gimple_build_call (report_error_func (is_store, size_in_bytes),
1445 1, base_addr);
1446 gimple_set_location (g, location);
1447 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1449 *iter = gsi_start_bb (else_bb);
1452 /* If T represents a memory access, add instrumentation code before ITER.
1453 LOCATION is source code location.
1454 IS_STORE is either TRUE (for a store) or FALSE (for a load). */
1456 static void
1457 instrument_derefs (gimple_stmt_iterator *iter, tree t,
1458 location_t location, bool is_store)
1460 tree type, base;
1461 HOST_WIDE_INT size_in_bytes;
1463 type = TREE_TYPE (t);
1464 switch (TREE_CODE (t))
1466 case ARRAY_REF:
1467 case COMPONENT_REF:
1468 case INDIRECT_REF:
1469 case MEM_REF:
1470 break;
1471 default:
1472 return;
1475 size_in_bytes = int_size_in_bytes (type);
1476 if ((size_in_bytes & (size_in_bytes - 1)) != 0
1477 || (unsigned HOST_WIDE_INT) size_in_bytes - 1 >= 16)
1478 return;
1480 HOST_WIDE_INT bitsize, bitpos;
1481 tree offset;
1482 enum machine_mode mode;
1483 int volatilep = 0, unsignedp = 0;
1484 get_inner_reference (t, &bitsize, &bitpos, &offset,
1485 &mode, &unsignedp, &volatilep, false);
1486 if (bitpos % (size_in_bytes * BITS_PER_UNIT)
1487 || bitsize != size_in_bytes * BITS_PER_UNIT)
1489 if (TREE_CODE (t) == COMPONENT_REF
1490 && DECL_BIT_FIELD_REPRESENTATIVE (TREE_OPERAND (t, 1)) != NULL_TREE)
1492 tree repr = DECL_BIT_FIELD_REPRESENTATIVE (TREE_OPERAND (t, 1));
1493 instrument_derefs (iter, build3 (COMPONENT_REF, TREE_TYPE (repr),
1494 TREE_OPERAND (t, 0), repr,
1495 NULL_TREE), location, is_store);
1497 return;
1500 base = build_fold_addr_expr (t);
1501 if (!has_mem_ref_been_instrumented (base, size_in_bytes))
1503 build_check_stmt (location, base, iter, /*before_p=*/true,
1504 is_store, size_in_bytes);
1505 update_mem_ref_hash_table (base, size_in_bytes);
1506 update_mem_ref_hash_table (t, size_in_bytes);
1511 /* Instrument an access to a contiguous memory region that starts at
1512 the address pointed to by BASE, over a length of LEN (expressed in
1513 the sizeof (*BASE) bytes). ITER points to the instruction before
1514 which the instrumentation instructions must be inserted. LOCATION
1515 is the source location that the instrumentation instructions must
1516 have. If IS_STORE is true, then the memory access is a store;
1517 otherwise, it's a load. */
1519 static void
1520 instrument_mem_region_access (tree base, tree len,
1521 gimple_stmt_iterator *iter,
1522 location_t location, bool is_store)
1524 if (!POINTER_TYPE_P (TREE_TYPE (base))
1525 || !INTEGRAL_TYPE_P (TREE_TYPE (len))
1526 || integer_zerop (len))
1527 return;
1529 gimple_stmt_iterator gsi = *iter;
1531 basic_block fallthrough_bb = NULL, then_bb = NULL;
1533 /* If the beginning of the memory region has already been
1534 instrumented, do not instrument it. */
1535 bool start_instrumented = has_mem_ref_been_instrumented (base, 1);
1537 /* If the end of the memory region has already been instrumented, do
1538 not instrument it. */
1539 tree end = asan_mem_ref_get_end (base, len);
1540 bool end_instrumented = has_mem_ref_been_instrumented (end, 1);
1542 if (start_instrumented && end_instrumented)
1543 return;
1545 if (!is_gimple_constant (len))
1547 /* So, the length of the memory area to asan-protect is
1548 non-constant. Let's guard the generated instrumentation code
1549 like:
1551 if (len != 0)
1553 //asan instrumentation code goes here.
1555 // falltrough instructions, starting with *ITER. */
1557 gimple g = gimple_build_cond (NE_EXPR,
1558 len,
1559 build_int_cst (TREE_TYPE (len), 0),
1560 NULL_TREE, NULL_TREE);
1561 gimple_set_location (g, location);
1562 insert_if_then_before_iter (g, iter, /*then_more_likely_p=*/true,
1563 &then_bb, &fallthrough_bb);
1564 /* Note that fallthrough_bb starts with the statement that was
1565 pointed to by ITER. */
1567 /* The 'then block' of the 'if (len != 0) condition is where
1568 we'll generate the asan instrumentation code now. */
1569 gsi = gsi_last_bb (then_bb);
1572 if (!start_instrumented)
1574 /* Instrument the beginning of the memory region to be accessed,
1575 and arrange for the rest of the intrumentation code to be
1576 inserted in the then block *after* the current gsi. */
1577 build_check_stmt (location, base, &gsi, /*before_p=*/true, is_store, 1);
1579 if (then_bb)
1580 /* We are in the case where the length of the region is not
1581 constant; so instrumentation code is being generated in the
1582 'then block' of the 'if (len != 0) condition. Let's arrange
1583 for the subsequent instrumentation statements to go in the
1584 'then block'. */
1585 gsi = gsi_last_bb (then_bb);
1586 else
1588 *iter = gsi;
1589 /* Don't remember this access as instrumented, if length
1590 is unknown. It might be zero and not being actually
1591 instrumented, so we can't rely on it being instrumented. */
1592 update_mem_ref_hash_table (base, 1);
1596 if (end_instrumented)
1597 return;
1599 /* We want to instrument the access at the end of the memory region,
1600 which is at (base + len - 1). */
1602 /* offset = len - 1; */
1603 len = unshare_expr (len);
1604 tree offset;
1605 gimple_seq seq = NULL;
1606 if (TREE_CODE (len) == INTEGER_CST)
1607 offset = fold_build2 (MINUS_EXPR, size_type_node,
1608 fold_convert (size_type_node, len),
1609 build_int_cst (size_type_node, 1));
1610 else
1612 gimple g;
1613 tree t;
1615 if (TREE_CODE (len) != SSA_NAME)
1617 t = make_ssa_name (TREE_TYPE (len), NULL);
1618 g = gimple_build_assign_with_ops (TREE_CODE (len), t, len, NULL);
1619 gimple_set_location (g, location);
1620 gimple_seq_add_stmt_without_update (&seq, g);
1621 len = t;
1623 if (!useless_type_conversion_p (size_type_node, TREE_TYPE (len)))
1625 t = make_ssa_name (size_type_node, NULL);
1626 g = gimple_build_assign_with_ops (NOP_EXPR, t, len, NULL);
1627 gimple_set_location (g, location);
1628 gimple_seq_add_stmt_without_update (&seq, g);
1629 len = t;
1632 t = make_ssa_name (size_type_node, NULL);
1633 g = gimple_build_assign_with_ops (MINUS_EXPR, t, len,
1634 build_int_cst (size_type_node, 1));
1635 gimple_set_location (g, location);
1636 gimple_seq_add_stmt_without_update (&seq, g);
1637 offset = gimple_assign_lhs (g);
1640 /* _1 = base; */
1641 base = unshare_expr (base);
1642 gimple region_end =
1643 gimple_build_assign_with_ops (TREE_CODE (base),
1644 make_ssa_name (TREE_TYPE (base), NULL),
1645 base, NULL);
1646 gimple_set_location (region_end, location);
1647 gimple_seq_add_stmt_without_update (&seq, region_end);
1649 /* _2 = _1 + offset; */
1650 region_end =
1651 gimple_build_assign_with_ops (POINTER_PLUS_EXPR,
1652 make_ssa_name (TREE_TYPE (base), NULL),
1653 gimple_assign_lhs (region_end),
1654 offset);
1655 gimple_set_location (region_end, location);
1656 gimple_seq_add_stmt_without_update (&seq, region_end);
1657 gsi_insert_seq_before (&gsi, seq, GSI_SAME_STMT);
1659 /* instrument access at _2; */
1660 gsi = gsi_for_stmt (region_end);
1661 build_check_stmt (location, gimple_assign_lhs (region_end),
1662 &gsi, /*before_p=*/false, is_store, 1);
1664 if (then_bb == NULL)
1665 update_mem_ref_hash_table (end, 1);
1667 *iter = gsi_for_stmt (gsi_stmt (*iter));
1670 /* Instrument the call (to the builtin strlen function) pointed to by
1671 ITER.
1673 This function instruments the access to the first byte of the
1674 argument, right before the call. After the call it instruments the
1675 access to the last byte of the argument; it uses the result of the
1676 call to deduce the offset of that last byte.
1678 Upon completion, iff the call has actually been instrumented, this
1679 function returns TRUE and *ITER points to the statement logically
1680 following the built-in strlen function call *ITER was initially
1681 pointing to. Otherwise, the function returns FALSE and *ITER
1682 remains unchanged. */
1684 static bool
1685 instrument_strlen_call (gimple_stmt_iterator *iter)
1687 gimple call = gsi_stmt (*iter);
1688 gcc_assert (is_gimple_call (call));
1690 tree callee = gimple_call_fndecl (call);
1691 gcc_assert (is_builtin_fn (callee)
1692 && DECL_BUILT_IN_CLASS (callee) == BUILT_IN_NORMAL
1693 && DECL_FUNCTION_CODE (callee) == BUILT_IN_STRLEN);
1695 tree len = gimple_call_lhs (call);
1696 if (len == NULL)
1697 /* Some passes might clear the return value of the strlen call;
1698 bail out in that case. Return FALSE as we are not advancing
1699 *ITER. */
1700 return false;
1701 gcc_assert (INTEGRAL_TYPE_P (TREE_TYPE (len)));
1703 location_t loc = gimple_location (call);
1704 tree str_arg = gimple_call_arg (call, 0);
1706 /* Instrument the access to the first byte of str_arg. i.e:
1708 _1 = str_arg; instrument (_1); */
1709 tree cptr_type = build_pointer_type (char_type_node);
1710 gimple str_arg_ssa =
1711 gimple_build_assign_with_ops (NOP_EXPR,
1712 make_ssa_name (cptr_type, NULL),
1713 str_arg, NULL);
1714 gimple_set_location (str_arg_ssa, loc);
1715 gimple_stmt_iterator gsi = *iter;
1716 gsi_insert_before (&gsi, str_arg_ssa, GSI_NEW_STMT);
1717 build_check_stmt (loc, gimple_assign_lhs (str_arg_ssa), &gsi,
1718 /*before_p=*/false, /*is_store=*/false, 1);
1720 /* If we initially had an instruction like:
1722 int n = strlen (str)
1724 we now want to instrument the access to str[n], after the
1725 instruction above.*/
1727 /* So let's build the access to str[n] that is, access through the
1728 pointer_plus expr: (_1 + len). */
1729 gimple stmt =
1730 gimple_build_assign_with_ops (POINTER_PLUS_EXPR,
1731 make_ssa_name (cptr_type, NULL),
1732 gimple_assign_lhs (str_arg_ssa),
1733 len);
1734 gimple_set_location (stmt, loc);
1735 gsi_insert_after (&gsi, stmt, GSI_NEW_STMT);
1737 build_check_stmt (loc, gimple_assign_lhs (stmt), &gsi,
1738 /*before_p=*/false, /*is_store=*/false, 1);
1740 /* Ensure that iter points to the statement logically following the
1741 one it was initially pointing to. */
1742 *iter = gsi;
1743 /* As *ITER has been advanced to point to the next statement, let's
1744 return true to inform transform_statements that it shouldn't
1745 advance *ITER anymore; otherwises it will skip that next
1746 statement, which wouldn't be instrumented. */
1747 return true;
1750 /* Instrument the call to a built-in memory access function that is
1751 pointed to by the iterator ITER.
1753 Upon completion, return TRUE iff *ITER has been advanced to the
1754 statement following the one it was originally pointing to. */
1756 static bool
1757 instrument_builtin_call (gimple_stmt_iterator *iter)
1759 bool iter_advanced_p = false;
1760 gimple call = gsi_stmt (*iter);
1762 gcc_checking_assert (gimple_call_builtin_p (call, BUILT_IN_NORMAL));
1764 tree callee = gimple_call_fndecl (call);
1765 location_t loc = gimple_location (call);
1767 if (DECL_FUNCTION_CODE (callee) == BUILT_IN_STRLEN)
1768 iter_advanced_p = instrument_strlen_call (iter);
1769 else
1771 asan_mem_ref src0, src1, dest;
1772 asan_mem_ref_init (&src0, NULL, 1);
1773 asan_mem_ref_init (&src1, NULL, 1);
1774 asan_mem_ref_init (&dest, NULL, 1);
1776 tree src0_len = NULL_TREE, src1_len = NULL_TREE, dest_len = NULL_TREE;
1777 bool src0_is_store = false, src1_is_store = false,
1778 dest_is_store = false, dest_is_deref = false;
1780 if (get_mem_refs_of_builtin_call (call,
1781 &src0, &src0_len, &src0_is_store,
1782 &src1, &src1_len, &src1_is_store,
1783 &dest, &dest_len, &dest_is_store,
1784 &dest_is_deref))
1786 if (dest_is_deref)
1788 instrument_derefs (iter, dest.start, loc, dest_is_store);
1789 gsi_next (iter);
1790 iter_advanced_p = true;
1792 else if (src0_len || src1_len || dest_len)
1794 if (src0.start != NULL_TREE)
1795 instrument_mem_region_access (src0.start, src0_len,
1796 iter, loc, /*is_store=*/false);
1797 if (src1.start != NULL_TREE)
1798 instrument_mem_region_access (src1.start, src1_len,
1799 iter, loc, /*is_store=*/false);
1800 if (dest.start != NULL_TREE)
1801 instrument_mem_region_access (dest.start, dest_len,
1802 iter, loc, /*is_store=*/true);
1803 *iter = gsi_for_stmt (call);
1804 gsi_next (iter);
1805 iter_advanced_p = true;
1809 return iter_advanced_p;
1812 /* Instrument the assignment statement ITER if it is subject to
1813 instrumentation. Return TRUE iff instrumentation actually
1814 happened. In that case, the iterator ITER is advanced to the next
1815 logical expression following the one initially pointed to by ITER,
1816 and the relevant memory reference that which access has been
1817 instrumented is added to the memory references hash table. */
1819 static bool
1820 maybe_instrument_assignment (gimple_stmt_iterator *iter)
1822 gimple s = gsi_stmt (*iter);
1824 gcc_assert (gimple_assign_single_p (s));
1826 tree ref_expr = NULL_TREE;
1827 bool is_store, is_instrumented = false;
1829 if (gimple_store_p (s))
1831 ref_expr = gimple_assign_lhs (s);
1832 is_store = true;
1833 instrument_derefs (iter, ref_expr,
1834 gimple_location (s),
1835 is_store);
1836 is_instrumented = true;
1839 if (gimple_assign_load_p (s))
1841 ref_expr = gimple_assign_rhs1 (s);
1842 is_store = false;
1843 instrument_derefs (iter, ref_expr,
1844 gimple_location (s),
1845 is_store);
1846 is_instrumented = true;
1849 if (is_instrumented)
1850 gsi_next (iter);
1852 return is_instrumented;
1855 /* Instrument the function call pointed to by the iterator ITER, if it
1856 is subject to instrumentation. At the moment, the only function
1857 calls that are instrumented are some built-in functions that access
1858 memory. Look at instrument_builtin_call to learn more.
1860 Upon completion return TRUE iff *ITER was advanced to the statement
1861 following the one it was originally pointing to. */
1863 static bool
1864 maybe_instrument_call (gimple_stmt_iterator *iter)
1866 gimple stmt = gsi_stmt (*iter);
1867 bool is_builtin = gimple_call_builtin_p (stmt, BUILT_IN_NORMAL);
1869 if (is_builtin && instrument_builtin_call (iter))
1870 return true;
1872 if (gimple_call_noreturn_p (stmt))
1874 if (is_builtin)
1876 tree callee = gimple_call_fndecl (stmt);
1877 switch (DECL_FUNCTION_CODE (callee))
1879 case BUILT_IN_UNREACHABLE:
1880 case BUILT_IN_TRAP:
1881 /* Don't instrument these. */
1882 return false;
1885 tree decl = builtin_decl_implicit (BUILT_IN_ASAN_HANDLE_NO_RETURN);
1886 gimple g = gimple_build_call (decl, 0);
1887 gimple_set_location (g, gimple_location (stmt));
1888 gsi_insert_before (iter, g, GSI_SAME_STMT);
1890 return false;
1893 /* Walk each instruction of all basic block and instrument those that
1894 represent memory references: loads, stores, or function calls.
1895 In a given basic block, this function avoids instrumenting memory
1896 references that have already been instrumented. */
1898 static void
1899 transform_statements (void)
1901 basic_block bb, last_bb = NULL;
1902 gimple_stmt_iterator i;
1903 int saved_last_basic_block = last_basic_block;
1905 FOR_EACH_BB (bb)
1907 basic_block prev_bb = bb;
1909 if (bb->index >= saved_last_basic_block) continue;
1911 /* Flush the mem ref hash table, if current bb doesn't have
1912 exactly one predecessor, or if that predecessor (skipping
1913 over asan created basic blocks) isn't the last processed
1914 basic block. Thus we effectively flush on extended basic
1915 block boundaries. */
1916 while (single_pred_p (prev_bb))
1918 prev_bb = single_pred (prev_bb);
1919 if (prev_bb->index < saved_last_basic_block)
1920 break;
1922 if (prev_bb != last_bb)
1923 empty_mem_ref_hash_table ();
1924 last_bb = bb;
1926 for (i = gsi_start_bb (bb); !gsi_end_p (i);)
1928 gimple s = gsi_stmt (i);
1930 if (has_stmt_been_instrumented_p (s))
1931 gsi_next (&i);
1932 else if (gimple_assign_single_p (s)
1933 && maybe_instrument_assignment (&i))
1934 /* Nothing to do as maybe_instrument_assignment advanced
1935 the iterator I. */;
1936 else if (is_gimple_call (s) && maybe_instrument_call (&i))
1937 /* Nothing to do as maybe_instrument_call
1938 advanced the iterator I. */;
1939 else
1941 /* No instrumentation happened.
1943 If the current instruction is a function call that
1944 might free something, let's forget about the memory
1945 references that got instrumented. Otherwise we might
1946 miss some instrumentation opportunities. */
1947 if (is_gimple_call (s) && !nonfreeing_call_p (s))
1948 empty_mem_ref_hash_table ();
1950 gsi_next (&i);
1954 free_mem_ref_resources ();
1957 /* Build
1958 struct __asan_global
1960 const void *__beg;
1961 uptr __size;
1962 uptr __size_with_redzone;
1963 const void *__name;
1964 uptr __has_dynamic_init;
1965 } type. */
1967 static tree
1968 asan_global_struct (void)
1970 static const char *field_names[5]
1971 = { "__beg", "__size", "__size_with_redzone",
1972 "__name", "__has_dynamic_init" };
1973 tree fields[5], ret;
1974 int i;
1976 ret = make_node (RECORD_TYPE);
1977 for (i = 0; i < 5; i++)
1979 fields[i]
1980 = build_decl (UNKNOWN_LOCATION, FIELD_DECL,
1981 get_identifier (field_names[i]),
1982 (i == 0 || i == 3) ? const_ptr_type_node
1983 : build_nonstandard_integer_type (POINTER_SIZE, 1));
1984 DECL_CONTEXT (fields[i]) = ret;
1985 if (i)
1986 DECL_CHAIN (fields[i - 1]) = fields[i];
1988 TYPE_FIELDS (ret) = fields[0];
1989 TYPE_NAME (ret) = get_identifier ("__asan_global");
1990 layout_type (ret);
1991 return ret;
1994 /* Append description of a single global DECL into vector V.
1995 TYPE is __asan_global struct type as returned by asan_global_struct. */
1997 static void
1998 asan_add_global (tree decl, tree type, vec<constructor_elt, va_gc> *v)
2000 tree init, uptr = TREE_TYPE (DECL_CHAIN (TYPE_FIELDS (type)));
2001 unsigned HOST_WIDE_INT size;
2002 tree str_cst, refdecl = decl;
2003 vec<constructor_elt, va_gc> *vinner = NULL;
2005 if (!asan_pp_initialized)
2006 asan_pp_initialize ();
2008 pp_clear_output_area (&asan_pp);
2009 if (DECL_NAME (decl))
2010 pp_base_tree_identifier (&asan_pp, DECL_NAME (decl));
2011 else
2012 pp_string (&asan_pp, "<unknown>");
2013 pp_space (&asan_pp);
2014 pp_left_paren (&asan_pp);
2015 pp_string (&asan_pp, main_input_filename);
2016 pp_right_paren (&asan_pp);
2017 str_cst = asan_pp_string ();
2019 if (asan_needs_local_alias (decl))
2021 char buf[20];
2022 ASM_GENERATE_INTERNAL_LABEL (buf, "LASAN", vec_safe_length (v) + 1);
2023 refdecl = build_decl (DECL_SOURCE_LOCATION (decl),
2024 VAR_DECL, get_identifier (buf), TREE_TYPE (decl));
2025 TREE_ADDRESSABLE (refdecl) = TREE_ADDRESSABLE (decl);
2026 TREE_READONLY (refdecl) = TREE_READONLY (decl);
2027 TREE_THIS_VOLATILE (refdecl) = TREE_THIS_VOLATILE (decl);
2028 DECL_GIMPLE_REG_P (refdecl) = DECL_GIMPLE_REG_P (decl);
2029 DECL_ARTIFICIAL (refdecl) = DECL_ARTIFICIAL (decl);
2030 DECL_IGNORED_P (refdecl) = DECL_IGNORED_P (decl);
2031 TREE_STATIC (refdecl) = 1;
2032 TREE_PUBLIC (refdecl) = 0;
2033 TREE_USED (refdecl) = 1;
2034 assemble_alias (refdecl, DECL_ASSEMBLER_NAME (decl));
2037 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
2038 fold_convert (const_ptr_type_node,
2039 build_fold_addr_expr (refdecl)));
2040 size = tree_low_cst (DECL_SIZE_UNIT (decl), 1);
2041 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, size));
2042 size += asan_red_zone_size (size);
2043 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, size));
2044 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
2045 fold_convert (const_ptr_type_node, str_cst));
2046 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, 0));
2047 init = build_constructor (type, vinner);
2048 CONSTRUCTOR_APPEND_ELT (v, NULL_TREE, init);
2051 /* Initialize sanitizer.def builtins if the FE hasn't initialized them. */
2052 void
2053 initialize_sanitizer_builtins (void)
2055 tree decl;
2057 if (builtin_decl_implicit_p (BUILT_IN_ASAN_INIT))
2058 return;
2060 tree BT_FN_VOID = build_function_type_list (void_type_node, NULL_TREE);
2061 tree BT_FN_VOID_PTR
2062 = build_function_type_list (void_type_node, ptr_type_node, NULL_TREE);
2063 tree BT_FN_VOID_PTR_PTRMODE
2064 = build_function_type_list (void_type_node, ptr_type_node,
2065 build_nonstandard_integer_type (POINTER_SIZE,
2066 1), NULL_TREE);
2067 tree BT_FN_VOID_INT
2068 = build_function_type_list (void_type_node, integer_type_node, NULL_TREE);
2069 tree BT_FN_BOOL_VPTR_PTR_IX_INT_INT[5];
2070 tree BT_FN_IX_CONST_VPTR_INT[5];
2071 tree BT_FN_IX_VPTR_IX_INT[5];
2072 tree BT_FN_VOID_VPTR_IX_INT[5];
2073 tree vptr
2074 = build_pointer_type (build_qualified_type (void_type_node,
2075 TYPE_QUAL_VOLATILE));
2076 tree cvptr
2077 = build_pointer_type (build_qualified_type (void_type_node,
2078 TYPE_QUAL_VOLATILE
2079 |TYPE_QUAL_CONST));
2080 tree boolt
2081 = lang_hooks.types.type_for_size (BOOL_TYPE_SIZE, 1);
2082 int i;
2083 for (i = 0; i < 5; i++)
2085 tree ix = build_nonstandard_integer_type (BITS_PER_UNIT * (1 << i), 1);
2086 BT_FN_BOOL_VPTR_PTR_IX_INT_INT[i]
2087 = build_function_type_list (boolt, vptr, ptr_type_node, ix,
2088 integer_type_node, integer_type_node,
2089 NULL_TREE);
2090 BT_FN_IX_CONST_VPTR_INT[i]
2091 = build_function_type_list (ix, cvptr, integer_type_node, NULL_TREE);
2092 BT_FN_IX_VPTR_IX_INT[i]
2093 = build_function_type_list (ix, vptr, ix, integer_type_node,
2094 NULL_TREE);
2095 BT_FN_VOID_VPTR_IX_INT[i]
2096 = build_function_type_list (void_type_node, vptr, ix,
2097 integer_type_node, NULL_TREE);
2099 #define BT_FN_BOOL_VPTR_PTR_I1_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[0]
2100 #define BT_FN_I1_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[0]
2101 #define BT_FN_I1_VPTR_I1_INT BT_FN_IX_VPTR_IX_INT[0]
2102 #define BT_FN_VOID_VPTR_I1_INT BT_FN_VOID_VPTR_IX_INT[0]
2103 #define BT_FN_BOOL_VPTR_PTR_I2_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[1]
2104 #define BT_FN_I2_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[1]
2105 #define BT_FN_I2_VPTR_I2_INT BT_FN_IX_VPTR_IX_INT[1]
2106 #define BT_FN_VOID_VPTR_I2_INT BT_FN_VOID_VPTR_IX_INT[1]
2107 #define BT_FN_BOOL_VPTR_PTR_I4_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[2]
2108 #define BT_FN_I4_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[2]
2109 #define BT_FN_I4_VPTR_I4_INT BT_FN_IX_VPTR_IX_INT[2]
2110 #define BT_FN_VOID_VPTR_I4_INT BT_FN_VOID_VPTR_IX_INT[2]
2111 #define BT_FN_BOOL_VPTR_PTR_I8_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[3]
2112 #define BT_FN_I8_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[3]
2113 #define BT_FN_I8_VPTR_I8_INT BT_FN_IX_VPTR_IX_INT[3]
2114 #define BT_FN_VOID_VPTR_I8_INT BT_FN_VOID_VPTR_IX_INT[3]
2115 #define BT_FN_BOOL_VPTR_PTR_I16_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[4]
2116 #define BT_FN_I16_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[4]
2117 #define BT_FN_I16_VPTR_I16_INT BT_FN_IX_VPTR_IX_INT[4]
2118 #define BT_FN_VOID_VPTR_I16_INT BT_FN_VOID_VPTR_IX_INT[4]
2119 #undef ATTR_NOTHROW_LEAF_LIST
2120 #define ATTR_NOTHROW_LEAF_LIST ECF_NOTHROW | ECF_LEAF
2121 #undef ATTR_TMPURE_NOTHROW_LEAF_LIST
2122 #define ATTR_TMPURE_NOTHROW_LEAF_LIST ECF_TM_PURE | ATTR_NOTHROW_LEAF_LIST
2123 #undef ATTR_NORETURN_NOTHROW_LEAF_LIST
2124 #define ATTR_NORETURN_NOTHROW_LEAF_LIST ECF_NORETURN | ATTR_NOTHROW_LEAF_LIST
2125 #undef ATTR_TMPURE_NORETURN_NOTHROW_LEAF_LIST
2126 #define ATTR_TMPURE_NORETURN_NOTHROW_LEAF_LIST \
2127 ECF_TM_PURE | ATTR_NORETURN_NOTHROW_LEAF_LIST
2128 #undef DEF_SANITIZER_BUILTIN
2129 #define DEF_SANITIZER_BUILTIN(ENUM, NAME, TYPE, ATTRS) \
2130 decl = add_builtin_function ("__builtin_" NAME, TYPE, ENUM, \
2131 BUILT_IN_NORMAL, NAME, NULL_TREE); \
2132 set_call_expr_flags (decl, ATTRS); \
2133 set_builtin_decl (ENUM, decl, true);
2135 #include "sanitizer.def"
2137 #undef DEF_SANITIZER_BUILTIN
2140 /* Called via htab_traverse. Count number of emitted
2141 STRING_CSTs in the constant hash table. */
2143 static int
2144 count_string_csts (void **slot, void *data)
2146 struct constant_descriptor_tree *desc
2147 = (struct constant_descriptor_tree *) *slot;
2148 if (TREE_CODE (desc->value) == STRING_CST
2149 && TREE_ASM_WRITTEN (desc->value)
2150 && asan_protect_global (desc->value))
2151 ++*((unsigned HOST_WIDE_INT *) data);
2152 return 1;
2155 /* Helper structure to pass two parameters to
2156 add_string_csts. */
2158 struct asan_add_string_csts_data
2160 tree type;
2161 vec<constructor_elt, va_gc> *v;
2164 /* Called via htab_traverse. Call asan_add_global
2165 on emitted STRING_CSTs from the constant hash table. */
2167 static int
2168 add_string_csts (void **slot, void *data)
2170 struct constant_descriptor_tree *desc
2171 = (struct constant_descriptor_tree *) *slot;
2172 if (TREE_CODE (desc->value) == STRING_CST
2173 && TREE_ASM_WRITTEN (desc->value)
2174 && asan_protect_global (desc->value))
2176 struct asan_add_string_csts_data *aascd
2177 = (struct asan_add_string_csts_data *) data;
2178 asan_add_global (SYMBOL_REF_DECL (XEXP (desc->rtl, 0)),
2179 aascd->type, aascd->v);
2181 return 1;
2184 /* Needs to be GTY(()), because cgraph_build_static_cdtor may
2185 invoke ggc_collect. */
2186 static GTY(()) tree asan_ctor_statements;
2188 /* Module-level instrumentation.
2189 - Insert __asan_init() into the list of CTORs.
2190 - TODO: insert redzones around globals.
2193 void
2194 asan_finish_file (void)
2196 struct varpool_node *vnode;
2197 unsigned HOST_WIDE_INT gcount = 0;
2199 if (shadow_ptr_types[0] == NULL_TREE)
2200 asan_init_shadow_ptr_types ();
2201 /* Avoid instrumenting code in the asan ctors/dtors.
2202 We don't need to insert padding after the description strings,
2203 nor after .LASAN* array. */
2204 flag_asan = 0;
2206 tree fn = builtin_decl_implicit (BUILT_IN_ASAN_INIT);
2207 append_to_statement_list (build_call_expr (fn, 0), &asan_ctor_statements);
2208 FOR_EACH_DEFINED_VARIABLE (vnode)
2209 if (TREE_ASM_WRITTEN (vnode->symbol.decl)
2210 && asan_protect_global (vnode->symbol.decl))
2211 ++gcount;
2212 htab_t const_desc_htab = constant_pool_htab ();
2213 htab_traverse (const_desc_htab, count_string_csts, &gcount);
2214 if (gcount)
2216 tree type = asan_global_struct (), var, ctor;
2217 tree uptr = build_nonstandard_integer_type (POINTER_SIZE, 1);
2218 tree dtor_statements = NULL_TREE;
2219 vec<constructor_elt, va_gc> *v;
2220 char buf[20];
2222 type = build_array_type_nelts (type, gcount);
2223 ASM_GENERATE_INTERNAL_LABEL (buf, "LASAN", 0);
2224 var = build_decl (UNKNOWN_LOCATION, VAR_DECL, get_identifier (buf),
2225 type);
2226 TREE_STATIC (var) = 1;
2227 TREE_PUBLIC (var) = 0;
2228 DECL_ARTIFICIAL (var) = 1;
2229 DECL_IGNORED_P (var) = 1;
2230 vec_alloc (v, gcount);
2231 FOR_EACH_DEFINED_VARIABLE (vnode)
2232 if (TREE_ASM_WRITTEN (vnode->symbol.decl)
2233 && asan_protect_global (vnode->symbol.decl))
2234 asan_add_global (vnode->symbol.decl, TREE_TYPE (type), v);
2235 struct asan_add_string_csts_data aascd;
2236 aascd.type = TREE_TYPE (type);
2237 aascd.v = v;
2238 htab_traverse (const_desc_htab, add_string_csts, &aascd);
2239 ctor = build_constructor (type, v);
2240 TREE_CONSTANT (ctor) = 1;
2241 TREE_STATIC (ctor) = 1;
2242 DECL_INITIAL (var) = ctor;
2243 varpool_assemble_decl (varpool_node_for_decl (var));
2245 fn = builtin_decl_implicit (BUILT_IN_ASAN_REGISTER_GLOBALS);
2246 append_to_statement_list (build_call_expr (fn, 2,
2247 build_fold_addr_expr (var),
2248 build_int_cst (uptr, gcount)),
2249 &asan_ctor_statements);
2251 fn = builtin_decl_implicit (BUILT_IN_ASAN_UNREGISTER_GLOBALS);
2252 append_to_statement_list (build_call_expr (fn, 2,
2253 build_fold_addr_expr (var),
2254 build_int_cst (uptr, gcount)),
2255 &dtor_statements);
2256 cgraph_build_static_cdtor ('D', dtor_statements,
2257 MAX_RESERVED_INIT_PRIORITY - 1);
2259 cgraph_build_static_cdtor ('I', asan_ctor_statements,
2260 MAX_RESERVED_INIT_PRIORITY - 1);
2261 flag_asan = 1;
2264 /* Instrument the current function. */
2266 static unsigned int
2267 asan_instrument (void)
2269 if (shadow_ptr_types[0] == NULL_TREE)
2270 asan_init_shadow_ptr_types ();
2271 transform_statements ();
2272 return 0;
2275 static bool
2276 gate_asan (void)
2278 return flag_asan != 0
2279 && !lookup_attribute ("no_sanitize_address",
2280 DECL_ATTRIBUTES (current_function_decl));
2283 struct gimple_opt_pass pass_asan =
2286 GIMPLE_PASS,
2287 "asan", /* name */
2288 OPTGROUP_NONE, /* optinfo_flags */
2289 gate_asan, /* gate */
2290 asan_instrument, /* execute */
2291 NULL, /* sub */
2292 NULL, /* next */
2293 0, /* static_pass_number */
2294 TV_NONE, /* tv_id */
2295 PROP_ssa | PROP_cfg | PROP_gimple_leh,/* properties_required */
2296 0, /* properties_provided */
2297 0, /* properties_destroyed */
2298 0, /* todo_flags_start */
2299 TODO_verify_flow | TODO_verify_stmts
2300 | TODO_update_ssa /* todo_flags_finish */
2304 static bool
2305 gate_asan_O0 (void)
2307 return !optimize && gate_asan ();
2310 struct gimple_opt_pass pass_asan_O0 =
2313 GIMPLE_PASS,
2314 "asan0", /* name */
2315 OPTGROUP_NONE, /* optinfo_flags */
2316 gate_asan_O0, /* gate */
2317 asan_instrument, /* execute */
2318 NULL, /* sub */
2319 NULL, /* next */
2320 0, /* static_pass_number */
2321 TV_NONE, /* tv_id */
2322 PROP_ssa | PROP_cfg | PROP_gimple_leh,/* properties_required */
2323 0, /* properties_provided */
2324 0, /* properties_destroyed */
2325 0, /* todo_flags_start */
2326 TODO_verify_flow | TODO_verify_stmts
2327 | TODO_update_ssa /* todo_flags_finish */
2331 #include "gt-asan.h"