[asan] Avoid instrumenting duplicated memory access in the same basic block
[official-gcc.git] / gcc / asan.c
blob3cb9511232993b10d7a30a3f69ca3d0a27af8f93
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
752 if (dest)
754 dst->start = dest;
755 dst->access_size = access_size;
756 *dst_len = NULL_TREE;
757 *dst_is_store = is_store;
758 *dest_is_deref = true;
759 got_reference_p = true;
763 return got_reference_p;
766 /* Return true iff a given gimple statement has been instrumented.
767 Note that the statement is "defined" by the memory references it
768 contains. */
770 static bool
771 has_stmt_been_instrumented_p (gimple stmt)
773 if (gimple_assign_single_p (stmt))
775 bool r_is_store;
776 asan_mem_ref r;
777 asan_mem_ref_init (&r, NULL, 1);
779 if (get_mem_ref_of_assignment (stmt, &r, &r_is_store))
780 return has_mem_ref_been_instrumented (&r);
782 else if (gimple_call_builtin_p (stmt, BUILT_IN_NORMAL))
784 asan_mem_ref src0, src1, dest;
785 asan_mem_ref_init (&src0, NULL, 1);
786 asan_mem_ref_init (&src1, NULL, 1);
787 asan_mem_ref_init (&dest, NULL, 1);
789 tree src0_len = NULL_TREE, src1_len = NULL_TREE, dest_len = NULL_TREE;
790 bool src0_is_store = false, src1_is_store = false,
791 dest_is_store = false, dest_is_deref = false;
792 if (get_mem_refs_of_builtin_call (stmt,
793 &src0, &src0_len, &src0_is_store,
794 &src1, &src1_len, &src1_is_store,
795 &dest, &dest_len, &dest_is_store,
796 &dest_is_deref))
798 if (src0.start != NULL_TREE
799 && !has_mem_ref_been_instrumented (&src0, src0_len))
800 return false;
802 if (src1.start != NULL_TREE
803 && !has_mem_ref_been_instrumented (&src1, src1_len))
804 return false;
806 if (dest.start != NULL_TREE
807 && !has_mem_ref_been_instrumented (&dest, dest_len))
808 return false;
810 return true;
813 return false;
816 /* Insert a memory reference into the hash table. */
818 static void
819 update_mem_ref_hash_table (tree ref, char access_size)
821 hash_table <asan_mem_ref_hasher> ht = get_mem_ref_hash_table ();
823 asan_mem_ref r;
824 asan_mem_ref_init (&r, ref, access_size);
826 asan_mem_ref **slot = ht.find_slot (&r, INSERT);
827 if (*slot == NULL)
828 *slot = asan_mem_ref_new (ref, access_size);
831 /* Initialize shadow_ptr_types array. */
833 static void
834 asan_init_shadow_ptr_types (void)
836 asan_shadow_set = new_alias_set ();
837 shadow_ptr_types[0] = build_distinct_type_copy (signed_char_type_node);
838 TYPE_ALIAS_SET (shadow_ptr_types[0]) = asan_shadow_set;
839 shadow_ptr_types[0] = build_pointer_type (shadow_ptr_types[0]);
840 shadow_ptr_types[1] = build_distinct_type_copy (short_integer_type_node);
841 TYPE_ALIAS_SET (shadow_ptr_types[1]) = asan_shadow_set;
842 shadow_ptr_types[1] = build_pointer_type (shadow_ptr_types[1]);
843 initialize_sanitizer_builtins ();
846 /* Asan pretty-printer, used for buidling of the description STRING_CSTs. */
847 static pretty_printer asan_pp;
848 static bool asan_pp_initialized;
850 /* Initialize asan_pp. */
852 static void
853 asan_pp_initialize (void)
855 pp_construct (&asan_pp, /* prefix */NULL, /* line-width */0);
856 asan_pp_initialized = true;
859 /* Create ADDR_EXPR of STRING_CST with asan_pp text. */
861 static tree
862 asan_pp_string (void)
864 const char *buf = pp_base_formatted_text (&asan_pp);
865 size_t len = strlen (buf);
866 tree ret = build_string (len + 1, buf);
867 TREE_TYPE (ret)
868 = build_array_type (TREE_TYPE (shadow_ptr_types[0]),
869 build_index_type (size_int (len)));
870 TREE_READONLY (ret) = 1;
871 TREE_STATIC (ret) = 1;
872 return build1 (ADDR_EXPR, shadow_ptr_types[0], ret);
875 /* Return a CONST_INT representing 4 subsequent shadow memory bytes. */
877 static rtx
878 asan_shadow_cst (unsigned char shadow_bytes[4])
880 int i;
881 unsigned HOST_WIDE_INT val = 0;
882 gcc_assert (WORDS_BIG_ENDIAN == BYTES_BIG_ENDIAN);
883 for (i = 0; i < 4; i++)
884 val |= (unsigned HOST_WIDE_INT) shadow_bytes[BYTES_BIG_ENDIAN ? 3 - i : i]
885 << (BITS_PER_UNIT * i);
886 return GEN_INT (trunc_int_for_mode (val, SImode));
889 /* Clear shadow memory at SHADOW_MEM, LEN bytes. Can't call a library call here
890 though. */
892 static void
893 asan_clear_shadow (rtx shadow_mem, HOST_WIDE_INT len)
895 rtx insn, insns, top_label, end, addr, tmp, jump;
897 start_sequence ();
898 clear_storage (shadow_mem, GEN_INT (len), BLOCK_OP_NORMAL);
899 insns = get_insns ();
900 end_sequence ();
901 for (insn = insns; insn; insn = NEXT_INSN (insn))
902 if (CALL_P (insn))
903 break;
904 if (insn == NULL_RTX)
906 emit_insn (insns);
907 return;
910 gcc_assert ((len & 3) == 0);
911 top_label = gen_label_rtx ();
912 addr = force_reg (Pmode, XEXP (shadow_mem, 0));
913 shadow_mem = adjust_automodify_address (shadow_mem, SImode, addr, 0);
914 end = force_reg (Pmode, plus_constant (Pmode, addr, len));
915 emit_label (top_label);
917 emit_move_insn (shadow_mem, const0_rtx);
918 tmp = expand_simple_binop (Pmode, PLUS, addr, GEN_INT (4), addr,
919 true, OPTAB_LIB_WIDEN);
920 if (tmp != addr)
921 emit_move_insn (addr, tmp);
922 emit_cmp_and_jump_insns (addr, end, LT, NULL_RTX, Pmode, true, top_label);
923 jump = get_last_insn ();
924 gcc_assert (JUMP_P (jump));
925 add_reg_note (jump, REG_BR_PROB, GEN_INT (REG_BR_PROB_BASE * 80 / 100));
928 /* Insert code to protect stack vars. The prologue sequence should be emitted
929 directly, epilogue sequence returned. BASE is the register holding the
930 stack base, against which OFFSETS array offsets are relative to, OFFSETS
931 array contains pairs of offsets in reverse order, always the end offset
932 of some gap that needs protection followed by starting offset,
933 and DECLS is an array of representative decls for each var partition.
934 LENGTH is the length of the OFFSETS array, DECLS array is LENGTH / 2 - 1
935 elements long (OFFSETS include gap before the first variable as well
936 as gaps after each stack variable). */
939 asan_emit_stack_protection (rtx base, HOST_WIDE_INT *offsets, tree *decls,
940 int length)
942 rtx shadow_base, shadow_mem, ret, mem;
943 unsigned char shadow_bytes[4];
944 HOST_WIDE_INT base_offset = offsets[length - 1], offset, prev_offset;
945 HOST_WIDE_INT last_offset, last_size;
946 int l;
947 unsigned char cur_shadow_byte = ASAN_STACK_MAGIC_LEFT;
948 tree str_cst;
950 if (shadow_ptr_types[0] == NULL_TREE)
951 asan_init_shadow_ptr_types ();
953 /* First of all, prepare the description string. */
954 if (!asan_pp_initialized)
955 asan_pp_initialize ();
957 pp_clear_output_area (&asan_pp);
958 if (DECL_NAME (current_function_decl))
959 pp_base_tree_identifier (&asan_pp, DECL_NAME (current_function_decl));
960 else
961 pp_string (&asan_pp, "<unknown>");
962 pp_space (&asan_pp);
963 pp_decimal_int (&asan_pp, length / 2 - 1);
964 pp_space (&asan_pp);
965 for (l = length - 2; l; l -= 2)
967 tree decl = decls[l / 2 - 1];
968 pp_wide_integer (&asan_pp, offsets[l] - base_offset);
969 pp_space (&asan_pp);
970 pp_wide_integer (&asan_pp, offsets[l - 1] - offsets[l]);
971 pp_space (&asan_pp);
972 if (DECL_P (decl) && DECL_NAME (decl))
974 pp_decimal_int (&asan_pp, IDENTIFIER_LENGTH (DECL_NAME (decl)));
975 pp_space (&asan_pp);
976 pp_base_tree_identifier (&asan_pp, DECL_NAME (decl));
978 else
979 pp_string (&asan_pp, "9 <unknown>");
980 pp_space (&asan_pp);
982 str_cst = asan_pp_string ();
984 /* Emit the prologue sequence. */
985 base = expand_binop (Pmode, add_optab, base, GEN_INT (base_offset),
986 NULL_RTX, 1, OPTAB_DIRECT);
987 mem = gen_rtx_MEM (ptr_mode, base);
988 emit_move_insn (mem, GEN_INT (ASAN_STACK_FRAME_MAGIC));
989 mem = adjust_address (mem, VOIDmode, GET_MODE_SIZE (ptr_mode));
990 emit_move_insn (mem, expand_normal (str_cst));
991 shadow_base = expand_binop (Pmode, lshr_optab, base,
992 GEN_INT (ASAN_SHADOW_SHIFT),
993 NULL_RTX, 1, OPTAB_DIRECT);
994 shadow_base = expand_binop (Pmode, add_optab, shadow_base,
995 GEN_INT (targetm.asan_shadow_offset ()),
996 NULL_RTX, 1, OPTAB_DIRECT);
997 gcc_assert (asan_shadow_set != -1
998 && (ASAN_RED_ZONE_SIZE >> ASAN_SHADOW_SHIFT) == 4);
999 shadow_mem = gen_rtx_MEM (SImode, shadow_base);
1000 set_mem_alias_set (shadow_mem, asan_shadow_set);
1001 prev_offset = base_offset;
1002 for (l = length; l; l -= 2)
1004 if (l == 2)
1005 cur_shadow_byte = ASAN_STACK_MAGIC_RIGHT;
1006 offset = offsets[l - 1];
1007 if ((offset - base_offset) & (ASAN_RED_ZONE_SIZE - 1))
1009 int i;
1010 HOST_WIDE_INT aoff
1011 = base_offset + ((offset - base_offset)
1012 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1));
1013 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1014 (aoff - prev_offset)
1015 >> ASAN_SHADOW_SHIFT);
1016 prev_offset = aoff;
1017 for (i = 0; i < 4; i++, aoff += (1 << ASAN_SHADOW_SHIFT))
1018 if (aoff < offset)
1020 if (aoff < offset - (1 << ASAN_SHADOW_SHIFT) + 1)
1021 shadow_bytes[i] = 0;
1022 else
1023 shadow_bytes[i] = offset - aoff;
1025 else
1026 shadow_bytes[i] = ASAN_STACK_MAGIC_PARTIAL;
1027 emit_move_insn (shadow_mem, asan_shadow_cst (shadow_bytes));
1028 offset = aoff;
1030 while (offset <= offsets[l - 2] - ASAN_RED_ZONE_SIZE)
1032 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1033 (offset - prev_offset)
1034 >> ASAN_SHADOW_SHIFT);
1035 prev_offset = offset;
1036 memset (shadow_bytes, cur_shadow_byte, 4);
1037 emit_move_insn (shadow_mem, asan_shadow_cst (shadow_bytes));
1038 offset += ASAN_RED_ZONE_SIZE;
1040 cur_shadow_byte = ASAN_STACK_MAGIC_MIDDLE;
1042 do_pending_stack_adjust ();
1044 /* Construct epilogue sequence. */
1045 start_sequence ();
1047 shadow_mem = gen_rtx_MEM (BLKmode, shadow_base);
1048 set_mem_alias_set (shadow_mem, asan_shadow_set);
1049 prev_offset = base_offset;
1050 last_offset = base_offset;
1051 last_size = 0;
1052 for (l = length; l; l -= 2)
1054 offset = base_offset + ((offsets[l - 1] - base_offset)
1055 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1));
1056 if (last_offset + last_size != offset)
1058 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1059 (last_offset - prev_offset)
1060 >> ASAN_SHADOW_SHIFT);
1061 prev_offset = last_offset;
1062 asan_clear_shadow (shadow_mem, last_size >> ASAN_SHADOW_SHIFT);
1063 last_offset = offset;
1064 last_size = 0;
1066 last_size += base_offset + ((offsets[l - 2] - base_offset)
1067 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1))
1068 - offset;
1070 if (last_size)
1072 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1073 (last_offset - prev_offset)
1074 >> ASAN_SHADOW_SHIFT);
1075 asan_clear_shadow (shadow_mem, last_size >> ASAN_SHADOW_SHIFT);
1078 do_pending_stack_adjust ();
1080 ret = get_insns ();
1081 end_sequence ();
1082 return ret;
1085 /* Return true if DECL, a global var, might be overridden and needs
1086 therefore a local alias. */
1088 static bool
1089 asan_needs_local_alias (tree decl)
1091 return DECL_WEAK (decl) || !targetm.binds_local_p (decl);
1094 /* Return true if DECL is a VAR_DECL that should be protected
1095 by Address Sanitizer, by appending a red zone with protected
1096 shadow memory after it and aligning it to at least
1097 ASAN_RED_ZONE_SIZE bytes. */
1099 bool
1100 asan_protect_global (tree decl)
1102 rtx rtl, symbol;
1104 if (TREE_CODE (decl) == STRING_CST)
1106 /* Instrument all STRING_CSTs except those created
1107 by asan_pp_string here. */
1108 if (shadow_ptr_types[0] != NULL_TREE
1109 && TREE_CODE (TREE_TYPE (decl)) == ARRAY_TYPE
1110 && TREE_TYPE (TREE_TYPE (decl)) == TREE_TYPE (shadow_ptr_types[0]))
1111 return false;
1112 return true;
1114 if (TREE_CODE (decl) != VAR_DECL
1115 /* TLS vars aren't statically protectable. */
1116 || DECL_THREAD_LOCAL_P (decl)
1117 /* Externs will be protected elsewhere. */
1118 || DECL_EXTERNAL (decl)
1119 || !DECL_RTL_SET_P (decl)
1120 /* Comdat vars pose an ABI problem, we can't know if
1121 the var that is selected by the linker will have
1122 padding or not. */
1123 || DECL_ONE_ONLY (decl)
1124 /* Similarly for common vars. People can use -fno-common. */
1125 || (DECL_COMMON (decl) && TREE_PUBLIC (decl))
1126 /* Don't protect if using user section, often vars placed
1127 into user section from multiple TUs are then assumed
1128 to be an array of such vars, putting padding in there
1129 breaks this assumption. */
1130 || (DECL_SECTION_NAME (decl) != NULL_TREE
1131 && !DECL_HAS_IMPLICIT_SECTION_NAME_P (decl))
1132 || DECL_SIZE (decl) == 0
1133 || ASAN_RED_ZONE_SIZE * BITS_PER_UNIT > MAX_OFILE_ALIGNMENT
1134 || !valid_constant_size_p (DECL_SIZE_UNIT (decl))
1135 || DECL_ALIGN_UNIT (decl) > 2 * ASAN_RED_ZONE_SIZE)
1136 return false;
1138 rtl = DECL_RTL (decl);
1139 if (!MEM_P (rtl) || GET_CODE (XEXP (rtl, 0)) != SYMBOL_REF)
1140 return false;
1141 symbol = XEXP (rtl, 0);
1143 if (CONSTANT_POOL_ADDRESS_P (symbol)
1144 || TREE_CONSTANT_POOL_ADDRESS_P (symbol))
1145 return false;
1147 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl)))
1148 return false;
1150 #ifndef ASM_OUTPUT_DEF
1151 if (asan_needs_local_alias (decl))
1152 return false;
1153 #endif
1155 return true;
1158 /* Construct a function tree for __asan_report_{load,store}{1,2,4,8,16}.
1159 IS_STORE is either 1 (for a store) or 0 (for a load).
1160 SIZE_IN_BYTES is one of 1, 2, 4, 8, 16. */
1162 static tree
1163 report_error_func (bool is_store, int size_in_bytes)
1165 static enum built_in_function report[2][5]
1166 = { { BUILT_IN_ASAN_REPORT_LOAD1, BUILT_IN_ASAN_REPORT_LOAD2,
1167 BUILT_IN_ASAN_REPORT_LOAD4, BUILT_IN_ASAN_REPORT_LOAD8,
1168 BUILT_IN_ASAN_REPORT_LOAD16 },
1169 { BUILT_IN_ASAN_REPORT_STORE1, BUILT_IN_ASAN_REPORT_STORE2,
1170 BUILT_IN_ASAN_REPORT_STORE4, BUILT_IN_ASAN_REPORT_STORE8,
1171 BUILT_IN_ASAN_REPORT_STORE16 } };
1172 return builtin_decl_implicit (report[is_store][exact_log2 (size_in_bytes)]);
1175 #define PROB_VERY_UNLIKELY (REG_BR_PROB_BASE / 2000 - 1)
1176 #define PROB_ALWAYS (REG_BR_PROB_BASE)
1178 /* Split the current basic block and create a condition statement
1179 insertion point right before or after the statement pointed to by
1180 ITER. Return an iterator to the point at which the caller might
1181 safely insert the condition statement.
1183 THEN_BLOCK must be set to the address of an uninitialized instance
1184 of basic_block. The function will then set *THEN_BLOCK to the
1185 'then block' of the condition statement to be inserted by the
1186 caller.
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 basic_block *then_block,
1205 basic_block *fallthrough_block)
1207 gimple_stmt_iterator gsi = *iter;
1209 if (!gsi_end_p (gsi) && before_p)
1210 gsi_prev (&gsi);
1212 basic_block cur_bb = gsi_bb (*iter);
1214 edge e = split_block (cur_bb, gsi_stmt (gsi));
1216 /* Get a hold on the 'condition block', the 'then block' and the
1217 'else block'. */
1218 basic_block cond_bb = e->src;
1219 basic_block fallthru_bb = e->dest;
1220 basic_block then_bb = create_empty_bb (cond_bb);
1222 /* Set up the newly created 'then block'. */
1223 e = make_edge (cond_bb, then_bb, EDGE_TRUE_VALUE);
1224 int fallthrough_probability
1225 = then_more_likely_p
1226 ? PROB_VERY_UNLIKELY
1227 : PROB_ALWAYS - PROB_VERY_UNLIKELY;
1228 e->probability = PROB_ALWAYS - fallthrough_probability;
1229 make_single_succ_edge (then_bb, fallthru_bb, EDGE_FALLTHRU);
1231 /* Set up the fallthrough basic block. */
1232 e = find_edge (cond_bb, fallthru_bb);
1233 e->flags = EDGE_FALSE_VALUE;
1234 e->count = cond_bb->count;
1235 e->probability = fallthrough_probability;
1237 /* Update dominance info for the newly created then_bb; note that
1238 fallthru_bb's dominance info has already been updated by
1239 split_bock. */
1240 if (dom_info_available_p (CDI_DOMINATORS))
1241 set_immediate_dominator (CDI_DOMINATORS, then_bb, cond_bb);
1243 *then_block = then_bb;
1244 *fallthrough_block = fallthru_bb;
1245 *iter = gsi_start_bb (fallthru_bb);
1247 return gsi_last_bb (cond_bb);
1250 /* Insert an if condition followed by a 'then block' right before the
1251 statement pointed to by ITER. The fallthrough block -- which is the
1252 else block of the condition as well as the destination of the
1253 outcoming edge of the 'then block' -- starts with the statement
1254 pointed to by ITER.
1256 COND is the condition of the if.
1258 If THEN_MORE_LIKELY_P is true, the probability of the edge to the
1259 'then block' is higher than the probability of the edge to the
1260 fallthrough block.
1262 Upon completion of the function, *THEN_BB is set to the newly
1263 inserted 'then block' and similarly, *FALLTHROUGH_BB is set to the
1264 fallthrough block.
1266 *ITER is adjusted to still point to the same statement it was
1267 pointing to initially. */
1269 static void
1270 insert_if_then_before_iter (gimple cond,
1271 gimple_stmt_iterator *iter,
1272 bool then_more_likely_p,
1273 basic_block *then_bb,
1274 basic_block *fallthrough_bb)
1276 gimple_stmt_iterator cond_insert_point =
1277 create_cond_insert_point (iter,
1278 /*before_p=*/true,
1279 then_more_likely_p,
1280 then_bb,
1281 fallthrough_bb);
1282 gsi_insert_after (&cond_insert_point, cond, GSI_NEW_STMT);
1285 /* Instrument the memory access instruction BASE. Insert new
1286 statements before or after ITER.
1288 Note that the memory access represented by BASE can be either an
1289 SSA_NAME, or a non-SSA expression. LOCATION is the source code
1290 location. IS_STORE is TRUE for a store, FALSE for a load.
1291 BEFORE_P is TRUE for inserting the instrumentation code before
1292 ITER, FALSE for inserting it after ITER. SIZE_IN_BYTES is one of
1293 1, 2, 4, 8, 16.
1295 If BEFORE_P is TRUE, *ITER is arranged to still point to the
1296 statement it was pointing to prior to calling this function,
1297 otherwise, it points to the statement logically following it. */
1299 static void
1300 build_check_stmt (location_t location, tree base, gimple_stmt_iterator *iter,
1301 bool before_p, bool is_store, int size_in_bytes)
1303 gimple_stmt_iterator gsi;
1304 basic_block then_bb, else_bb;
1305 tree t, base_addr, shadow;
1306 gimple g;
1307 tree shadow_ptr_type = shadow_ptr_types[size_in_bytes == 16 ? 1 : 0];
1308 tree shadow_type = TREE_TYPE (shadow_ptr_type);
1309 tree uintptr_type
1310 = build_nonstandard_integer_type (TYPE_PRECISION (TREE_TYPE (base)), 1);
1311 tree base_ssa = base;
1313 /* Get an iterator on the point where we can add the condition
1314 statement for the instrumentation. */
1315 gsi = create_cond_insert_point (iter, before_p,
1316 /*then_more_likely_p=*/false,
1317 &then_bb,
1318 &else_bb);
1320 base = unshare_expr (base);
1322 /* BASE can already be an SSA_NAME; in that case, do not create a
1323 new SSA_NAME for it. */
1324 if (TREE_CODE (base) != SSA_NAME)
1326 g = gimple_build_assign_with_ops (TREE_CODE (base),
1327 make_ssa_name (TREE_TYPE (base), NULL),
1328 base, NULL_TREE);
1329 gimple_set_location (g, location);
1330 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1331 base_ssa = gimple_assign_lhs (g);
1334 g = gimple_build_assign_with_ops (NOP_EXPR,
1335 make_ssa_name (uintptr_type, NULL),
1336 base_ssa, NULL_TREE);
1337 gimple_set_location (g, location);
1338 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1339 base_addr = gimple_assign_lhs (g);
1341 /* Build
1342 (base_addr >> ASAN_SHADOW_SHIFT) + targetm.asan_shadow_offset (). */
1344 t = build_int_cst (uintptr_type, ASAN_SHADOW_SHIFT);
1345 g = gimple_build_assign_with_ops (RSHIFT_EXPR,
1346 make_ssa_name (uintptr_type, NULL),
1347 base_addr, t);
1348 gimple_set_location (g, location);
1349 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1351 t = build_int_cst (uintptr_type, targetm.asan_shadow_offset ());
1352 g = gimple_build_assign_with_ops (PLUS_EXPR,
1353 make_ssa_name (uintptr_type, NULL),
1354 gimple_assign_lhs (g), t);
1355 gimple_set_location (g, location);
1356 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1358 g = gimple_build_assign_with_ops (NOP_EXPR,
1359 make_ssa_name (shadow_ptr_type, NULL),
1360 gimple_assign_lhs (g), NULL_TREE);
1361 gimple_set_location (g, location);
1362 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1364 t = build2 (MEM_REF, shadow_type, gimple_assign_lhs (g),
1365 build_int_cst (shadow_ptr_type, 0));
1366 g = gimple_build_assign_with_ops (MEM_REF,
1367 make_ssa_name (shadow_type, NULL),
1368 t, NULL_TREE);
1369 gimple_set_location (g, location);
1370 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1371 shadow = gimple_assign_lhs (g);
1373 if (size_in_bytes < 8)
1375 /* Slow path for 1, 2 and 4 byte accesses.
1376 Test (shadow != 0)
1377 & ((base_addr & 7) + (size_in_bytes - 1)) >= shadow). */
1378 g = gimple_build_assign_with_ops (NE_EXPR,
1379 make_ssa_name (boolean_type_node,
1380 NULL),
1381 shadow,
1382 build_int_cst (shadow_type, 0));
1383 gimple_set_location (g, location);
1384 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1385 t = gimple_assign_lhs (g);
1387 g = gimple_build_assign_with_ops (BIT_AND_EXPR,
1388 make_ssa_name (uintptr_type,
1389 NULL),
1390 base_addr,
1391 build_int_cst (uintptr_type, 7));
1392 gimple_set_location (g, location);
1393 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1395 g = gimple_build_assign_with_ops (NOP_EXPR,
1396 make_ssa_name (shadow_type,
1397 NULL),
1398 gimple_assign_lhs (g), NULL_TREE);
1399 gimple_set_location (g, location);
1400 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1402 if (size_in_bytes > 1)
1404 g = gimple_build_assign_with_ops (PLUS_EXPR,
1405 make_ssa_name (shadow_type,
1406 NULL),
1407 gimple_assign_lhs (g),
1408 build_int_cst (shadow_type,
1409 size_in_bytes - 1));
1410 gimple_set_location (g, location);
1411 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1414 g = gimple_build_assign_with_ops (GE_EXPR,
1415 make_ssa_name (boolean_type_node,
1416 NULL),
1417 gimple_assign_lhs (g),
1418 shadow);
1419 gimple_set_location (g, location);
1420 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1422 g = gimple_build_assign_with_ops (BIT_AND_EXPR,
1423 make_ssa_name (boolean_type_node,
1424 NULL),
1425 t, gimple_assign_lhs (g));
1426 gimple_set_location (g, location);
1427 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1428 t = gimple_assign_lhs (g);
1430 else
1431 t = shadow;
1433 g = gimple_build_cond (NE_EXPR, t, build_int_cst (TREE_TYPE (t), 0),
1434 NULL_TREE, NULL_TREE);
1435 gimple_set_location (g, location);
1436 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1438 /* Generate call to the run-time library (e.g. __asan_report_load8). */
1439 gsi = gsi_start_bb (then_bb);
1440 g = gimple_build_call (report_error_func (is_store, size_in_bytes),
1441 1, base_addr);
1442 gimple_set_location (g, location);
1443 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1445 *iter = gsi_start_bb (else_bb);
1448 /* If T represents a memory access, add instrumentation code before ITER.
1449 LOCATION is source code location.
1450 IS_STORE is either TRUE (for a store) or FALSE (for a load). */
1452 static void
1453 instrument_derefs (gimple_stmt_iterator *iter, tree t,
1454 location_t location, bool is_store)
1456 tree type, base;
1457 HOST_WIDE_INT size_in_bytes;
1459 type = TREE_TYPE (t);
1460 switch (TREE_CODE (t))
1462 case ARRAY_REF:
1463 case COMPONENT_REF:
1464 case INDIRECT_REF:
1465 case MEM_REF:
1466 break;
1467 default:
1468 return;
1471 size_in_bytes = int_size_in_bytes (type);
1472 if ((size_in_bytes & (size_in_bytes - 1)) != 0
1473 || (unsigned HOST_WIDE_INT) size_in_bytes - 1 >= 16)
1474 return;
1476 HOST_WIDE_INT bitsize, bitpos;
1477 tree offset;
1478 enum machine_mode mode;
1479 int volatilep = 0, unsignedp = 0;
1480 get_inner_reference (t, &bitsize, &bitpos, &offset,
1481 &mode, &unsignedp, &volatilep, false);
1482 if (bitpos % (size_in_bytes * BITS_PER_UNIT)
1483 || bitsize != size_in_bytes * BITS_PER_UNIT)
1485 if (TREE_CODE (t) == COMPONENT_REF
1486 && DECL_BIT_FIELD_REPRESENTATIVE (TREE_OPERAND (t, 1)) != NULL_TREE)
1488 tree repr = DECL_BIT_FIELD_REPRESENTATIVE (TREE_OPERAND (t, 1));
1489 instrument_derefs (iter, build3 (COMPONENT_REF, TREE_TYPE (repr),
1490 TREE_OPERAND (t, 0), repr,
1491 NULL_TREE), location, is_store);
1493 return;
1496 base = build_fold_addr_expr (t);
1497 if (!has_mem_ref_been_instrumented (base, size_in_bytes))
1499 build_check_stmt (location, base, iter, /*before_p=*/true,
1500 is_store, size_in_bytes);
1501 update_mem_ref_hash_table (base, size_in_bytes);
1502 update_mem_ref_hash_table (t, size_in_bytes);
1507 /* Instrument an access to a contiguous memory region that starts at
1508 the address pointed to by BASE, over a length of LEN (expressed in
1509 the sizeof (*BASE) bytes). ITER points to the instruction before
1510 which the instrumentation instructions must be inserted. LOCATION
1511 is the source location that the instrumentation instructions must
1512 have. If IS_STORE is true, then the memory access is a store;
1513 otherwise, it's a load. */
1515 static void
1516 instrument_mem_region_access (tree base, tree len,
1517 gimple_stmt_iterator *iter,
1518 location_t location, bool is_store)
1520 if (!POINTER_TYPE_P (TREE_TYPE (base))
1521 || !INTEGRAL_TYPE_P (TREE_TYPE (len))
1522 || integer_zerop (len))
1523 return;
1525 gimple_stmt_iterator gsi = *iter;
1527 basic_block fallthrough_bb = NULL, then_bb = NULL;
1529 /* If the beginning of the memory region has already been
1530 instrumented, do not instrument it. */
1531 if (has_mem_ref_been_instrumented (base, 1))
1532 goto after_first_instrumentation;
1534 if (!is_gimple_constant (len))
1536 /* So, the length of the memory area to asan-protect is
1537 non-constant. Let's guard the generated instrumentation code
1538 like:
1540 if (len != 0)
1542 //asan instrumentation code goes here.
1544 // falltrough instructions, starting with *ITER. */
1546 gimple g = gimple_build_cond (NE_EXPR,
1547 len,
1548 build_int_cst (TREE_TYPE (len), 0),
1549 NULL_TREE, NULL_TREE);
1550 gimple_set_location (g, location);
1551 insert_if_then_before_iter (g, iter, /*then_more_likely_p=*/true,
1552 &then_bb, &fallthrough_bb);
1553 /* Note that fallthrough_bb starts with the statement that was
1554 pointed to by ITER. */
1556 /* The 'then block' of the 'if (len != 0) condition is where
1557 we'll generate the asan instrumentation code now. */
1558 gsi = gsi_start_bb (then_bb);
1561 /* Instrument the beginning of the memory region to be accessed,
1562 and arrange for the rest of the intrumentation code to be
1563 inserted in the then block *after* the current gsi. */
1564 build_check_stmt (location, base, &gsi, /*before_p=*/true, is_store, 1);
1566 if (then_bb)
1567 /* We are in the case where the length of the region is not
1568 constant; so instrumentation code is being generated in the
1569 'then block' of the 'if (len != 0) condition. Let's arrange
1570 for the subsequent instrumentation statements to go in the
1571 'then block'. */
1572 gsi = gsi_last_bb (then_bb);
1573 else
1574 *iter = gsi;
1576 update_mem_ref_hash_table (base, 1);
1578 after_first_instrumentation:
1580 /* We want to instrument the access at the end of the memory region,
1581 which is at (base + len - 1). */
1583 /* If the end of the memory region has already been instrumented, do
1584 not instrument it. */
1585 tree end = asan_mem_ref_get_end (base, len);
1586 if (has_mem_ref_been_instrumented (end, 1))
1587 return;
1589 /* offset = len - 1; */
1590 len = unshare_expr (len);
1591 tree offset;
1592 gimple_seq seq = NULL;
1593 if (TREE_CODE (len) == INTEGER_CST)
1594 offset = fold_build2 (MINUS_EXPR, size_type_node,
1595 fold_convert (size_type_node, len),
1596 build_int_cst (size_type_node, 1));
1597 else
1599 gimple g;
1600 tree t;
1602 if (TREE_CODE (len) != SSA_NAME)
1604 t = make_ssa_name (TREE_TYPE (len), NULL);
1605 g = gimple_build_assign_with_ops (TREE_CODE (len), t, len, NULL);
1606 gimple_set_location (g, location);
1607 gimple_seq_add_stmt_without_update (&seq, g);
1608 len = t;
1610 if (!useless_type_conversion_p (size_type_node, TREE_TYPE (len)))
1612 t = make_ssa_name (size_type_node, NULL);
1613 g = gimple_build_assign_with_ops (NOP_EXPR, t, len, NULL);
1614 gimple_set_location (g, location);
1615 gimple_seq_add_stmt_without_update (&seq, g);
1616 len = t;
1619 t = make_ssa_name (size_type_node, NULL);
1620 g = gimple_build_assign_with_ops (MINUS_EXPR, t, len,
1621 build_int_cst (size_type_node, 1));
1622 gimple_set_location (g, location);
1623 gimple_seq_add_stmt_without_update (&seq, g);
1624 offset = gimple_assign_lhs (g);
1627 /* _1 = base; */
1628 base = unshare_expr (base);
1629 gimple region_end =
1630 gimple_build_assign_with_ops (TREE_CODE (base),
1631 make_ssa_name (TREE_TYPE (base), NULL),
1632 base, NULL);
1633 gimple_set_location (region_end, location);
1634 gimple_seq_add_stmt_without_update (&seq, region_end);
1635 gsi_insert_seq_before (&gsi, seq, GSI_SAME_STMT);
1636 gsi_prev (&gsi);
1638 /* _2 = _1 + offset; */
1639 region_end =
1640 gimple_build_assign_with_ops (POINTER_PLUS_EXPR,
1641 make_ssa_name (TREE_TYPE (base), NULL),
1642 gimple_assign_lhs (region_end),
1643 offset);
1644 gimple_set_location (region_end, location);
1645 gsi_insert_after (&gsi, region_end, GSI_NEW_STMT);
1647 /* instrument access at _2; */
1648 build_check_stmt (location, gimple_assign_lhs (region_end),
1649 &gsi, /*before_p=*/false, is_store, 1);
1651 update_mem_ref_hash_table (end, 1);
1654 /* Instrument the call (to the builtin strlen function) pointed to by
1655 ITER.
1657 This function instruments the access to the first byte of the
1658 argument, right before the call. After the call it instruments the
1659 access to the last byte of the argument; it uses the result of the
1660 call to deduce the offset of that last byte.
1662 Upon completion, iff the call has actullay been instrumented, this
1663 function returns TRUE and *ITER points to the statement logically
1664 following the built-in strlen function call *ITER was initially
1665 pointing to. Otherwise, the function returns FALSE and *ITER
1666 remains unchanged. */
1668 static bool
1669 instrument_strlen_call (gimple_stmt_iterator *iter)
1671 gimple call = gsi_stmt (*iter);
1672 gcc_assert (is_gimple_call (call));
1674 tree callee = gimple_call_fndecl (call);
1675 gcc_assert (is_builtin_fn (callee)
1676 && DECL_BUILT_IN_CLASS (callee) == BUILT_IN_NORMAL
1677 && DECL_FUNCTION_CODE (callee) == BUILT_IN_STRLEN);
1679 tree len = gimple_call_lhs (call);
1680 if (len == NULL)
1681 /* Some passes might clear the return value of the strlen call;
1682 bail out in that case. Return FALSE as we are not advancing
1683 *ITER. */
1684 return false;
1685 gcc_assert (INTEGRAL_TYPE_P (TREE_TYPE (len)));
1687 location_t loc = gimple_location (call);
1688 tree str_arg = gimple_call_arg (call, 0);
1690 /* Instrument the access to the first byte of str_arg. i.e:
1692 _1 = str_arg; instrument (_1); */
1693 gimple str_arg_ssa =
1694 gimple_build_assign_with_ops (NOP_EXPR,
1695 make_ssa_name (build_pointer_type
1696 (char_type_node), NULL),
1697 str_arg, NULL);
1698 gimple_set_location (str_arg_ssa, loc);
1699 gimple_stmt_iterator gsi = *iter;
1700 gsi_insert_before (&gsi, str_arg_ssa, GSI_NEW_STMT);
1701 build_check_stmt (loc, gimple_assign_lhs (str_arg_ssa), &gsi,
1702 /*before_p=*/false, /*is_store=*/false, 1);
1704 /* If we initially had an instruction like:
1706 int n = strlen (str)
1708 we now want to instrument the access to str[n], after the
1709 instruction above.*/
1711 /* So let's build the access to str[n] that is, access through the
1712 pointer_plus expr: (_1 + len). */
1713 gimple stmt =
1714 gimple_build_assign_with_ops (POINTER_PLUS_EXPR,
1715 make_ssa_name (TREE_TYPE (str_arg),
1716 NULL),
1717 gimple_assign_lhs (str_arg_ssa),
1718 len);
1719 gimple_set_location (stmt, loc);
1720 gsi_insert_after (&gsi, stmt, GSI_NEW_STMT);
1722 build_check_stmt (loc, gimple_assign_lhs (stmt), &gsi,
1723 /*before_p=*/false, /*is_store=*/false, 1);
1725 /* Ensure that iter points to the statement logically following the
1726 one it was initially pointing to. */
1727 *iter = gsi;
1728 /* As *ITER has been advanced to point to the next statement, let's
1729 return true to inform transform_statements that it shouldn't
1730 advance *ITER anymore; otherwises it will skip that next
1731 statement, which wouldn't be instrumented. */
1732 return true;
1735 /* Instrument the call to a built-in memory access function that is
1736 pointed to by the iterator ITER.
1738 Upon completion, return TRUE iff *ITER has been advanced to the
1739 statement following the one it was originally pointing to. */
1741 static bool
1742 instrument_builtin_call (gimple_stmt_iterator *iter)
1744 bool iter_advanced_p = false;
1745 gimple call = gsi_stmt (*iter);
1747 gcc_checking_assert (gimple_call_builtin_p (call, BUILT_IN_NORMAL));
1749 tree callee = gimple_call_fndecl (call);
1750 location_t loc = gimple_location (call);
1752 if (DECL_FUNCTION_CODE (callee) == BUILT_IN_STRLEN)
1753 iter_advanced_p = instrument_strlen_call (iter);
1754 else
1756 asan_mem_ref src0, src1, dest;
1757 asan_mem_ref_init (&src0, NULL, 1);
1758 asan_mem_ref_init (&src1, NULL, 1);
1759 asan_mem_ref_init (&dest, NULL, 1);
1761 tree src0_len = NULL_TREE, src1_len = NULL_TREE, dest_len = NULL_TREE;
1762 bool src0_is_store = false, src1_is_store = false,
1763 dest_is_store = false, dest_is_deref = false;
1765 if (get_mem_refs_of_builtin_call (call,
1766 &src0, &src0_len, &src0_is_store,
1767 &src1, &src0_len, &src1_is_store,
1768 &dest, &dest_len, &dest_is_store,
1769 &dest_is_deref))
1771 if (dest_is_deref)
1773 instrument_derefs (iter, dest.start, loc, dest_is_store);
1774 gsi_next (iter);
1775 iter_advanced_p = true;
1777 else if (src0_len || src1_len || dest_len)
1779 if (src0.start)
1780 instrument_mem_region_access (src0.start, src0_len,
1781 iter, loc, /*is_store=*/false);
1782 if (src1.start != NULL_TREE)
1783 instrument_mem_region_access (src1.start, src1_len,
1784 iter, loc, /*is_store=*/false);
1785 if (dest.start != NULL_TREE)
1786 instrument_mem_region_access (dest.start, dest_len,
1787 iter, loc, /*is_store=*/true);
1788 *iter = gsi_for_stmt (call);
1789 gsi_next (iter);
1790 iter_advanced_p = true;
1794 return iter_advanced_p;
1797 /* Instrument the assignment statement ITER if it is subject to
1798 instrumentation. Return TRUE iff instrumentation actually
1799 happened. In that case, the iterator ITER is advanced to the next
1800 logical expression following the one initially pointed to by ITER,
1801 and the relevant memory reference that which access has been
1802 instrumented is added to the memory references hash table. */
1804 static bool
1805 maybe_instrument_assignment (gimple_stmt_iterator *iter)
1807 gimple s = gsi_stmt (*iter);
1809 gcc_assert (gimple_assign_single_p (s));
1811 tree ref_expr = NULL_TREE;
1812 bool is_store, is_instrumented = false;
1814 if (gimple_store_p (s))
1816 ref_expr = gimple_assign_lhs (s);
1817 is_store = true;
1818 instrument_derefs (iter, ref_expr,
1819 gimple_location (s),
1820 is_store);
1821 is_instrumented = true;
1824 if (gimple_assign_load_p (s))
1826 ref_expr = gimple_assign_rhs1 (s);
1827 is_store = false;
1828 instrument_derefs (iter, ref_expr,
1829 gimple_location (s),
1830 is_store);
1831 is_instrumented = true;
1834 if (is_instrumented)
1835 gsi_next (iter);
1837 return is_instrumented;
1840 /* Instrument the function call pointed to by the iterator ITER, if it
1841 is subject to instrumentation. At the moment, the only function
1842 calls that are instrumented are some built-in functions that access
1843 memory. Look at instrument_builtin_call to learn more.
1845 Upon completion return TRUE iff *ITER was advanced to the statement
1846 following the one it was originally pointing to. */
1848 static bool
1849 maybe_instrument_call (gimple_stmt_iterator *iter)
1851 gimple stmt = gsi_stmt (*iter);
1852 bool is_builtin = gimple_call_builtin_p (stmt, BUILT_IN_NORMAL);
1854 if (is_builtin && instrument_builtin_call (iter))
1855 return true;
1857 if (gimple_call_noreturn_p (stmt))
1859 if (is_builtin)
1861 tree callee = gimple_call_fndecl (stmt);
1862 switch (DECL_FUNCTION_CODE (callee))
1864 case BUILT_IN_UNREACHABLE:
1865 case BUILT_IN_TRAP:
1866 /* Don't instrument these. */
1867 return false;
1870 tree decl = builtin_decl_implicit (BUILT_IN_ASAN_HANDLE_NO_RETURN);
1871 gimple g = gimple_build_call (decl, 0);
1872 gimple_set_location (g, gimple_location (stmt));
1873 gsi_insert_before (iter, g, GSI_SAME_STMT);
1875 return false;
1878 /* Walk each instruction of all basic block and instrument those that
1879 represent memory references: loads, stores, or function calls.
1880 In a given basic block, this function avoids instrumenting memory
1881 references that have already been instrumented. */
1883 static void
1884 transform_statements (void)
1886 basic_block bb;
1887 gimple_stmt_iterator i;
1888 int saved_last_basic_block = last_basic_block;
1890 FOR_EACH_BB (bb)
1892 empty_mem_ref_hash_table ();
1894 if (bb->index >= saved_last_basic_block) continue;
1895 for (i = gsi_start_bb (bb); !gsi_end_p (i);)
1897 gimple s = gsi_stmt (i);
1899 if (has_stmt_been_instrumented_p (s))
1900 gsi_next (&i);
1901 else if (gimple_assign_single_p (s)
1902 && maybe_instrument_assignment (&i))
1903 /* Nothing to do as maybe_instrument_assignment advanced
1904 the iterator I. */;
1905 else if (is_gimple_call (s) && maybe_instrument_call (&i))
1906 /* Nothing to do as maybe_instrument_call
1907 advanced the iterator I. */;
1908 else
1910 /* No instrumentation happened.
1912 If the current instruction is a function call, let's
1913 forget about the memory references that got
1914 instrumented. Otherwise we might miss some
1915 instrumentation opportunities. */
1916 if (is_gimple_call (s))
1917 empty_mem_ref_hash_table ();
1919 gsi_next (&i);
1923 free_mem_ref_resources ();
1926 /* Build
1927 struct __asan_global
1929 const void *__beg;
1930 uptr __size;
1931 uptr __size_with_redzone;
1932 const void *__name;
1933 uptr __has_dynamic_init;
1934 } type. */
1936 static tree
1937 asan_global_struct (void)
1939 static const char *field_names[5]
1940 = { "__beg", "__size", "__size_with_redzone",
1941 "__name", "__has_dynamic_init" };
1942 tree fields[5], ret;
1943 int i;
1945 ret = make_node (RECORD_TYPE);
1946 for (i = 0; i < 5; i++)
1948 fields[i]
1949 = build_decl (UNKNOWN_LOCATION, FIELD_DECL,
1950 get_identifier (field_names[i]),
1951 (i == 0 || i == 3) ? const_ptr_type_node
1952 : build_nonstandard_integer_type (POINTER_SIZE, 1));
1953 DECL_CONTEXT (fields[i]) = ret;
1954 if (i)
1955 DECL_CHAIN (fields[i - 1]) = fields[i];
1957 TYPE_FIELDS (ret) = fields[0];
1958 TYPE_NAME (ret) = get_identifier ("__asan_global");
1959 layout_type (ret);
1960 return ret;
1963 /* Append description of a single global DECL into vector V.
1964 TYPE is __asan_global struct type as returned by asan_global_struct. */
1966 static void
1967 asan_add_global (tree decl, tree type, vec<constructor_elt, va_gc> *v)
1969 tree init, uptr = TREE_TYPE (DECL_CHAIN (TYPE_FIELDS (type)));
1970 unsigned HOST_WIDE_INT size;
1971 tree str_cst, refdecl = decl;
1972 vec<constructor_elt, va_gc> *vinner = NULL;
1974 if (!asan_pp_initialized)
1975 asan_pp_initialize ();
1977 pp_clear_output_area (&asan_pp);
1978 if (DECL_NAME (decl))
1979 pp_base_tree_identifier (&asan_pp, DECL_NAME (decl));
1980 else
1981 pp_string (&asan_pp, "<unknown>");
1982 pp_space (&asan_pp);
1983 pp_left_paren (&asan_pp);
1984 pp_string (&asan_pp, main_input_filename);
1985 pp_right_paren (&asan_pp);
1986 str_cst = asan_pp_string ();
1988 if (asan_needs_local_alias (decl))
1990 char buf[20];
1991 ASM_GENERATE_INTERNAL_LABEL (buf, "LASAN", vec_safe_length (v) + 1);
1992 refdecl = build_decl (DECL_SOURCE_LOCATION (decl),
1993 VAR_DECL, get_identifier (buf), TREE_TYPE (decl));
1994 TREE_ADDRESSABLE (refdecl) = TREE_ADDRESSABLE (decl);
1995 TREE_READONLY (refdecl) = TREE_READONLY (decl);
1996 TREE_THIS_VOLATILE (refdecl) = TREE_THIS_VOLATILE (decl);
1997 DECL_GIMPLE_REG_P (refdecl) = DECL_GIMPLE_REG_P (decl);
1998 DECL_ARTIFICIAL (refdecl) = DECL_ARTIFICIAL (decl);
1999 DECL_IGNORED_P (refdecl) = DECL_IGNORED_P (decl);
2000 TREE_STATIC (refdecl) = 1;
2001 TREE_PUBLIC (refdecl) = 0;
2002 TREE_USED (refdecl) = 1;
2003 assemble_alias (refdecl, DECL_ASSEMBLER_NAME (decl));
2006 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
2007 fold_convert (const_ptr_type_node,
2008 build_fold_addr_expr (refdecl)));
2009 size = tree_low_cst (DECL_SIZE_UNIT (decl), 1);
2010 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, size));
2011 size += asan_red_zone_size (size);
2012 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, size));
2013 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
2014 fold_convert (const_ptr_type_node, str_cst));
2015 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, 0));
2016 init = build_constructor (type, vinner);
2017 CONSTRUCTOR_APPEND_ELT (v, NULL_TREE, init);
2020 /* Initialize sanitizer.def builtins if the FE hasn't initialized them. */
2021 void
2022 initialize_sanitizer_builtins (void)
2024 tree decl;
2026 if (builtin_decl_implicit_p (BUILT_IN_ASAN_INIT))
2027 return;
2029 tree BT_FN_VOID = build_function_type_list (void_type_node, NULL_TREE);
2030 tree BT_FN_VOID_PTR
2031 = build_function_type_list (void_type_node, ptr_type_node, NULL_TREE);
2032 tree BT_FN_VOID_PTR_PTRMODE
2033 = build_function_type_list (void_type_node, ptr_type_node,
2034 build_nonstandard_integer_type (POINTER_SIZE,
2035 1), NULL_TREE);
2036 tree BT_FN_VOID_INT
2037 = build_function_type_list (void_type_node, integer_type_node, NULL_TREE);
2038 tree BT_FN_BOOL_VPTR_PTR_IX_INT_INT[5];
2039 tree BT_FN_IX_CONST_VPTR_INT[5];
2040 tree BT_FN_IX_VPTR_IX_INT[5];
2041 tree BT_FN_VOID_VPTR_IX_INT[5];
2042 tree vptr
2043 = build_pointer_type (build_qualified_type (void_type_node,
2044 TYPE_QUAL_VOLATILE));
2045 tree cvptr
2046 = build_pointer_type (build_qualified_type (void_type_node,
2047 TYPE_QUAL_VOLATILE
2048 |TYPE_QUAL_CONST));
2049 tree boolt
2050 = lang_hooks.types.type_for_size (BOOL_TYPE_SIZE, 1);
2051 int i;
2052 for (i = 0; i < 5; i++)
2054 tree ix = build_nonstandard_integer_type (BITS_PER_UNIT * (1 << i), 1);
2055 BT_FN_BOOL_VPTR_PTR_IX_INT_INT[i]
2056 = build_function_type_list (boolt, vptr, ptr_type_node, ix,
2057 integer_type_node, integer_type_node,
2058 NULL_TREE);
2059 BT_FN_IX_CONST_VPTR_INT[i]
2060 = build_function_type_list (ix, cvptr, integer_type_node, NULL_TREE);
2061 BT_FN_IX_VPTR_IX_INT[i]
2062 = build_function_type_list (ix, vptr, ix, integer_type_node,
2063 NULL_TREE);
2064 BT_FN_VOID_VPTR_IX_INT[i]
2065 = build_function_type_list (void_type_node, vptr, ix,
2066 integer_type_node, NULL_TREE);
2068 #define BT_FN_BOOL_VPTR_PTR_I1_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[0]
2069 #define BT_FN_I1_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[0]
2070 #define BT_FN_I1_VPTR_I1_INT BT_FN_IX_VPTR_IX_INT[0]
2071 #define BT_FN_VOID_VPTR_I1_INT BT_FN_VOID_VPTR_IX_INT[0]
2072 #define BT_FN_BOOL_VPTR_PTR_I2_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[1]
2073 #define BT_FN_I2_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[1]
2074 #define BT_FN_I2_VPTR_I2_INT BT_FN_IX_VPTR_IX_INT[1]
2075 #define BT_FN_VOID_VPTR_I2_INT BT_FN_VOID_VPTR_IX_INT[1]
2076 #define BT_FN_BOOL_VPTR_PTR_I4_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[2]
2077 #define BT_FN_I4_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[2]
2078 #define BT_FN_I4_VPTR_I4_INT BT_FN_IX_VPTR_IX_INT[2]
2079 #define BT_FN_VOID_VPTR_I4_INT BT_FN_VOID_VPTR_IX_INT[2]
2080 #define BT_FN_BOOL_VPTR_PTR_I8_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[3]
2081 #define BT_FN_I8_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[3]
2082 #define BT_FN_I8_VPTR_I8_INT BT_FN_IX_VPTR_IX_INT[3]
2083 #define BT_FN_VOID_VPTR_I8_INT BT_FN_VOID_VPTR_IX_INT[3]
2084 #define BT_FN_BOOL_VPTR_PTR_I16_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[4]
2085 #define BT_FN_I16_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[4]
2086 #define BT_FN_I16_VPTR_I16_INT BT_FN_IX_VPTR_IX_INT[4]
2087 #define BT_FN_VOID_VPTR_I16_INT BT_FN_VOID_VPTR_IX_INT[4]
2088 #undef ATTR_NOTHROW_LEAF_LIST
2089 #define ATTR_NOTHROW_LEAF_LIST ECF_NOTHROW | ECF_LEAF
2090 #undef ATTR_TMPURE_NOTHROW_LEAF_LIST
2091 #define ATTR_TMPURE_NOTHROW_LEAF_LIST ECF_TM_PURE | ATTR_NOTHROW_LEAF_LIST
2092 #undef ATTR_NORETURN_NOTHROW_LEAF_LIST
2093 #define ATTR_NORETURN_NOTHROW_LEAF_LIST ECF_NORETURN | ATTR_NOTHROW_LEAF_LIST
2094 #undef ATTR_TMPURE_NORETURN_NOTHROW_LEAF_LIST
2095 #define ATTR_TMPURE_NORETURN_NOTHROW_LEAF_LIST \
2096 ECF_TM_PURE | ATTR_NORETURN_NOTHROW_LEAF_LIST
2097 #undef DEF_SANITIZER_BUILTIN
2098 #define DEF_SANITIZER_BUILTIN(ENUM, NAME, TYPE, ATTRS) \
2099 decl = add_builtin_function ("__builtin_" NAME, TYPE, ENUM, \
2100 BUILT_IN_NORMAL, NAME, NULL_TREE); \
2101 set_call_expr_flags (decl, ATTRS); \
2102 set_builtin_decl (ENUM, decl, true);
2104 #include "sanitizer.def"
2106 #undef DEF_SANITIZER_BUILTIN
2109 /* Called via htab_traverse. Count number of emitted
2110 STRING_CSTs in the constant hash table. */
2112 static int
2113 count_string_csts (void **slot, void *data)
2115 struct constant_descriptor_tree *desc
2116 = (struct constant_descriptor_tree *) *slot;
2117 if (TREE_CODE (desc->value) == STRING_CST
2118 && TREE_ASM_WRITTEN (desc->value)
2119 && asan_protect_global (desc->value))
2120 ++*((unsigned HOST_WIDE_INT *) data);
2121 return 1;
2124 /* Helper structure to pass two parameters to
2125 add_string_csts. */
2127 struct asan_add_string_csts_data
2129 tree type;
2130 vec<constructor_elt, va_gc> *v;
2133 /* Called via htab_traverse. Call asan_add_global
2134 on emitted STRING_CSTs from the constant hash table. */
2136 static int
2137 add_string_csts (void **slot, void *data)
2139 struct constant_descriptor_tree *desc
2140 = (struct constant_descriptor_tree *) *slot;
2141 if (TREE_CODE (desc->value) == STRING_CST
2142 && TREE_ASM_WRITTEN (desc->value)
2143 && asan_protect_global (desc->value))
2145 struct asan_add_string_csts_data *aascd
2146 = (struct asan_add_string_csts_data *) data;
2147 asan_add_global (SYMBOL_REF_DECL (XEXP (desc->rtl, 0)),
2148 aascd->type, aascd->v);
2150 return 1;
2153 /* Needs to be GTY(()), because cgraph_build_static_cdtor may
2154 invoke ggc_collect. */
2155 static GTY(()) tree asan_ctor_statements;
2157 /* Module-level instrumentation.
2158 - Insert __asan_init() into the list of CTORs.
2159 - TODO: insert redzones around globals.
2162 void
2163 asan_finish_file (void)
2165 struct varpool_node *vnode;
2166 unsigned HOST_WIDE_INT gcount = 0;
2168 if (shadow_ptr_types[0] == NULL_TREE)
2169 asan_init_shadow_ptr_types ();
2170 /* Avoid instrumenting code in the asan ctors/dtors.
2171 We don't need to insert padding after the description strings,
2172 nor after .LASAN* array. */
2173 flag_asan = 0;
2175 tree fn = builtin_decl_implicit (BUILT_IN_ASAN_INIT);
2176 append_to_statement_list (build_call_expr (fn, 0), &asan_ctor_statements);
2177 FOR_EACH_DEFINED_VARIABLE (vnode)
2178 if (TREE_ASM_WRITTEN (vnode->symbol.decl)
2179 && asan_protect_global (vnode->symbol.decl))
2180 ++gcount;
2181 htab_t const_desc_htab = constant_pool_htab ();
2182 htab_traverse (const_desc_htab, count_string_csts, &gcount);
2183 if (gcount)
2185 tree type = asan_global_struct (), var, ctor;
2186 tree uptr = build_nonstandard_integer_type (POINTER_SIZE, 1);
2187 tree dtor_statements = NULL_TREE;
2188 vec<constructor_elt, va_gc> *v;
2189 char buf[20];
2191 type = build_array_type_nelts (type, gcount);
2192 ASM_GENERATE_INTERNAL_LABEL (buf, "LASAN", 0);
2193 var = build_decl (UNKNOWN_LOCATION, VAR_DECL, get_identifier (buf),
2194 type);
2195 TREE_STATIC (var) = 1;
2196 TREE_PUBLIC (var) = 0;
2197 DECL_ARTIFICIAL (var) = 1;
2198 DECL_IGNORED_P (var) = 1;
2199 vec_alloc (v, gcount);
2200 FOR_EACH_DEFINED_VARIABLE (vnode)
2201 if (TREE_ASM_WRITTEN (vnode->symbol.decl)
2202 && asan_protect_global (vnode->symbol.decl))
2203 asan_add_global (vnode->symbol.decl, TREE_TYPE (type), v);
2204 struct asan_add_string_csts_data aascd;
2205 aascd.type = TREE_TYPE (type);
2206 aascd.v = v;
2207 htab_traverse (const_desc_htab, add_string_csts, &aascd);
2208 ctor = build_constructor (type, v);
2209 TREE_CONSTANT (ctor) = 1;
2210 TREE_STATIC (ctor) = 1;
2211 DECL_INITIAL (var) = ctor;
2212 varpool_assemble_decl (varpool_node_for_decl (var));
2214 fn = builtin_decl_implicit (BUILT_IN_ASAN_REGISTER_GLOBALS);
2215 append_to_statement_list (build_call_expr (fn, 2,
2216 build_fold_addr_expr (var),
2217 build_int_cst (uptr, gcount)),
2218 &asan_ctor_statements);
2220 fn = builtin_decl_implicit (BUILT_IN_ASAN_UNREGISTER_GLOBALS);
2221 append_to_statement_list (build_call_expr (fn, 2,
2222 build_fold_addr_expr (var),
2223 build_int_cst (uptr, gcount)),
2224 &dtor_statements);
2225 cgraph_build_static_cdtor ('D', dtor_statements,
2226 MAX_RESERVED_INIT_PRIORITY - 1);
2228 cgraph_build_static_cdtor ('I', asan_ctor_statements,
2229 MAX_RESERVED_INIT_PRIORITY - 1);
2230 flag_asan = 1;
2233 /* Instrument the current function. */
2235 static unsigned int
2236 asan_instrument (void)
2238 if (shadow_ptr_types[0] == NULL_TREE)
2239 asan_init_shadow_ptr_types ();
2240 transform_statements ();
2241 return 0;
2244 static bool
2245 gate_asan (void)
2247 return flag_asan != 0
2248 && !lookup_attribute ("no_address_safety_analysis",
2249 DECL_ATTRIBUTES (current_function_decl));
2252 struct gimple_opt_pass pass_asan =
2255 GIMPLE_PASS,
2256 "asan", /* name */
2257 OPTGROUP_NONE, /* optinfo_flags */
2258 gate_asan, /* gate */
2259 asan_instrument, /* execute */
2260 NULL, /* sub */
2261 NULL, /* next */
2262 0, /* static_pass_number */
2263 TV_NONE, /* tv_id */
2264 PROP_ssa | PROP_cfg | PROP_gimple_leh,/* properties_required */
2265 0, /* properties_provided */
2266 0, /* properties_destroyed */
2267 0, /* todo_flags_start */
2268 TODO_verify_flow | TODO_verify_stmts
2269 | TODO_update_ssa /* todo_flags_finish */
2273 static bool
2274 gate_asan_O0 (void)
2276 return !optimize && gate_asan ();
2279 struct gimple_opt_pass pass_asan_O0 =
2282 GIMPLE_PASS,
2283 "asan0", /* name */
2284 OPTGROUP_NONE, /* optinfo_flags */
2285 gate_asan_O0, /* gate */
2286 asan_instrument, /* execute */
2287 NULL, /* sub */
2288 NULL, /* next */
2289 0, /* static_pass_number */
2290 TV_NONE, /* tv_id */
2291 PROP_ssa | PROP_cfg | PROP_gimple_leh,/* properties_required */
2292 0, /* properties_provided */
2293 0, /* properties_destroyed */
2294 0, /* todo_flags_start */
2295 TODO_verify_flow | TODO_verify_stmts
2296 | TODO_update_ssa /* todo_flags_finish */
2300 #include "gt-asan.h"