gcc/c-family/
[official-gcc.git] / gcc / asan.c
blob4353db6316016acc3af18431c4cf0c1f3c9c1133
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 "tree.h"
26 #include "gimple.h"
27 #include "gimplify.h"
28 #include "gimple-iterator.h"
29 #include "tree-iterator.h"
30 #include "cgraph.h"
31 #include "tree-ssanames.h"
32 #include "tree-pass.h"
33 #include "asan.h"
34 #include "gimple-pretty-print.h"
35 #include "target.h"
36 #include "expr.h"
37 #include "optabs.h"
38 #include "output.h"
39 #include "tm_p.h"
40 #include "langhooks.h"
41 #include "hash-table.h"
42 #include "alloc-pool.h"
43 #include "cfgloop.h"
44 #include "gimple-builder.h"
46 /* AddressSanitizer finds out-of-bounds and use-after-free bugs
47 with <2x slowdown on average.
49 The tool consists of two parts:
50 instrumentation module (this file) and a run-time library.
51 The instrumentation module adds a run-time check before every memory insn.
52 For a 8- or 16- byte load accessing address X:
53 ShadowAddr = (X >> 3) + Offset
54 ShadowValue = *(char*)ShadowAddr; // *(short*) for 16-byte access.
55 if (ShadowValue)
56 __asan_report_load8(X);
57 For a load of N bytes (N=1, 2 or 4) from address X:
58 ShadowAddr = (X >> 3) + Offset
59 ShadowValue = *(char*)ShadowAddr;
60 if (ShadowValue)
61 if ((X & 7) + N - 1 > ShadowValue)
62 __asan_report_loadN(X);
63 Stores are instrumented similarly, but using __asan_report_storeN functions.
64 A call too __asan_init_vN() is inserted to the list of module CTORs.
65 N is the version number of the AddressSanitizer API. The changes between the
66 API versions are listed in libsanitizer/asan/asan_interface_internal.h.
68 The run-time library redefines malloc (so that redzone are inserted around
69 the allocated memory) and free (so that reuse of free-ed memory is delayed),
70 provides __asan_report* and __asan_init_vN functions.
72 Read more:
73 http://code.google.com/p/address-sanitizer/wiki/AddressSanitizerAlgorithm
75 The current implementation supports detection of out-of-bounds and
76 use-after-free in the heap, on the stack and for global variables.
78 [Protection of stack variables]
80 To understand how detection of out-of-bounds and use-after-free works
81 for stack variables, lets look at this example on x86_64 where the
82 stack grows downward:
84 int
85 foo ()
87 char a[23] = {0};
88 int b[2] = {0};
90 a[5] = 1;
91 b[1] = 2;
93 return a[5] + b[1];
96 For this function, the stack protected by asan will be organized as
97 follows, from the top of the stack to the bottom:
99 Slot 1/ [red zone of 32 bytes called 'RIGHT RedZone']
101 Slot 2/ [8 bytes of red zone, that adds up to the space of 'a' to make
102 the next slot be 32 bytes aligned; this one is called Partial
103 Redzone; this 32 bytes alignment is an asan constraint]
105 Slot 3/ [24 bytes for variable 'a']
107 Slot 4/ [red zone of 32 bytes called 'Middle RedZone']
109 Slot 5/ [24 bytes of Partial Red Zone (similar to slot 2]
111 Slot 6/ [8 bytes for variable 'b']
113 Slot 7/ [32 bytes of Red Zone at the bottom of the stack, called
114 'LEFT RedZone']
116 The 32 bytes of LEFT red zone at the bottom of the stack can be
117 decomposed as such:
119 1/ The first 8 bytes contain a magical asan number that is always
120 0x41B58AB3.
122 2/ The following 8 bytes contains a pointer to a string (to be
123 parsed at runtime by the runtime asan library), which format is
124 the following:
126 "<function-name> <space> <num-of-variables-on-the-stack>
127 (<32-bytes-aligned-offset-in-bytes-of-variable> <space>
128 <length-of-var-in-bytes> ){n} "
130 where '(...){n}' means the content inside the parenthesis occurs 'n'
131 times, with 'n' being the number of variables on the stack.
133 3/ The following 8 bytes contain the PC of the current function which
134 will be used by the run-time library to print an error message.
136 4/ The following 8 bytes are reserved for internal use by the run-time.
138 The shadow memory for that stack layout is going to look like this:
140 - content of shadow memory 8 bytes for slot 7: 0xF1F1F1F1.
141 The F1 byte pattern is a magic number called
142 ASAN_STACK_MAGIC_LEFT and is a way for the runtime to know that
143 the memory for that shadow byte is part of a the LEFT red zone
144 intended to seat at the bottom of the variables on the stack.
146 - content of shadow memory 8 bytes for slots 6 and 5:
147 0xF4F4F400. The F4 byte pattern is a magic number
148 called ASAN_STACK_MAGIC_PARTIAL. It flags the fact that the
149 memory region for this shadow byte is a PARTIAL red zone
150 intended to pad a variable A, so that the slot following
151 {A,padding} is 32 bytes aligned.
153 Note that the fact that the least significant byte of this
154 shadow memory content is 00 means that 8 bytes of its
155 corresponding memory (which corresponds to the memory of
156 variable 'b') is addressable.
158 - content of shadow memory 8 bytes for slot 4: 0xF2F2F2F2.
159 The F2 byte pattern is a magic number called
160 ASAN_STACK_MAGIC_MIDDLE. It flags the fact that the memory
161 region for this shadow byte is a MIDDLE red zone intended to
162 seat between two 32 aligned slots of {variable,padding}.
164 - content of shadow memory 8 bytes for slot 3 and 2:
165 0xF4000000. This represents is the concatenation of
166 variable 'a' and the partial red zone following it, like what we
167 had for variable 'b'. The least significant 3 bytes being 00
168 means that the 3 bytes of variable 'a' are addressable.
170 - content of shadow memory 8 bytes for slot 1: 0xF3F3F3F3.
171 The F3 byte pattern is a magic number called
172 ASAN_STACK_MAGIC_RIGHT. It flags the fact that the memory
173 region for this shadow byte is a RIGHT red zone intended to seat
174 at the top of the variables of the stack.
176 Note that the real variable layout is done in expand_used_vars in
177 cfgexpand.c. As far as Address Sanitizer is concerned, it lays out
178 stack variables as well as the different red zones, emits some
179 prologue code to populate the shadow memory as to poison (mark as
180 non-accessible) the regions of the red zones and mark the regions of
181 stack variables as accessible, and emit some epilogue code to
182 un-poison (mark as accessible) the regions of red zones right before
183 the function exits.
185 [Protection of global variables]
187 The basic idea is to insert a red zone between two global variables
188 and install a constructor function that calls the asan runtime to do
189 the populating of the relevant shadow memory regions at load time.
191 So the global variables are laid out as to insert a red zone between
192 them. The size of the red zones is so that each variable starts on a
193 32 bytes boundary.
195 Then a constructor function is installed so that, for each global
196 variable, it calls the runtime asan library function
197 __asan_register_globals_with an instance of this type:
199 struct __asan_global
201 // Address of the beginning of the global variable.
202 const void *__beg;
204 // Initial size of the global variable.
205 uptr __size;
207 // Size of the global variable + size of the red zone. This
208 // size is 32 bytes aligned.
209 uptr __size_with_redzone;
211 // Name of the global variable.
212 const void *__name;
214 // Name of the module where the global variable is declared.
215 const void *__module_name;
217 // This is always set to NULL for now.
218 uptr __has_dynamic_init;
221 A destructor function that calls the runtime asan library function
222 _asan_unregister_globals is also installed. */
224 alias_set_type asan_shadow_set = -1;
226 /* Pointer types to 1 resp. 2 byte integers in shadow memory. A separate
227 alias set is used for all shadow memory accesses. */
228 static GTY(()) tree shadow_ptr_types[2];
230 /* Hashtable support for memory references used by gimple
231 statements. */
233 /* This type represents a reference to a memory region. */
234 struct asan_mem_ref
236 /* The expression of the beginning of the memory region. */
237 tree start;
239 /* The size of the access (can be 1, 2, 4, 8, 16 for now). */
240 char access_size;
243 static alloc_pool asan_mem_ref_alloc_pool;
245 /* This creates the alloc pool used to store the instances of
246 asan_mem_ref that are stored in the hash table asan_mem_ref_ht. */
248 static alloc_pool
249 asan_mem_ref_get_alloc_pool ()
251 if (asan_mem_ref_alloc_pool == NULL)
252 asan_mem_ref_alloc_pool = create_alloc_pool ("asan_mem_ref",
253 sizeof (asan_mem_ref),
254 10);
255 return asan_mem_ref_alloc_pool;
259 /* Initializes an instance of asan_mem_ref. */
261 static void
262 asan_mem_ref_init (asan_mem_ref *ref, tree start, char access_size)
264 ref->start = start;
265 ref->access_size = access_size;
268 /* Allocates memory for an instance of asan_mem_ref into the memory
269 pool returned by asan_mem_ref_get_alloc_pool and initialize it.
270 START is the address of (or the expression pointing to) the
271 beginning of memory reference. ACCESS_SIZE is the size of the
272 access to the referenced memory. */
274 static asan_mem_ref*
275 asan_mem_ref_new (tree start, char access_size)
277 asan_mem_ref *ref =
278 (asan_mem_ref *) pool_alloc (asan_mem_ref_get_alloc_pool ());
280 asan_mem_ref_init (ref, start, access_size);
281 return ref;
284 /* This builds and returns a pointer to the end of the memory region
285 that starts at START and of length LEN. */
287 tree
288 asan_mem_ref_get_end (tree start, tree len)
290 if (len == NULL_TREE || integer_zerop (len))
291 return start;
293 return fold_build2 (POINTER_PLUS_EXPR, TREE_TYPE (start), start, len);
296 /* Return a tree expression that represents the end of the referenced
297 memory region. Beware that this function can actually build a new
298 tree expression. */
300 tree
301 asan_mem_ref_get_end (const asan_mem_ref *ref, tree len)
303 return asan_mem_ref_get_end (ref->start, len);
306 struct asan_mem_ref_hasher
307 : typed_noop_remove <asan_mem_ref>
309 typedef asan_mem_ref value_type;
310 typedef asan_mem_ref compare_type;
312 static inline hashval_t hash (const value_type *);
313 static inline bool equal (const value_type *, const compare_type *);
316 /* Hash a memory reference. */
318 inline hashval_t
319 asan_mem_ref_hasher::hash (const asan_mem_ref *mem_ref)
321 hashval_t h = iterative_hash_expr (mem_ref->start, 0);
322 h = iterative_hash_hashval_t (h, mem_ref->access_size);
323 return h;
326 /* Compare two memory references. We accept the length of either
327 memory references to be NULL_TREE. */
329 inline bool
330 asan_mem_ref_hasher::equal (const asan_mem_ref *m1,
331 const asan_mem_ref *m2)
333 return (m1->access_size == m2->access_size
334 && operand_equal_p (m1->start, m2->start, 0));
337 static hash_table <asan_mem_ref_hasher> asan_mem_ref_ht;
339 /* Returns a reference to the hash table containing memory references.
340 This function ensures that the hash table is created. Note that
341 this hash table is updated by the function
342 update_mem_ref_hash_table. */
344 static hash_table <asan_mem_ref_hasher> &
345 get_mem_ref_hash_table ()
347 if (!asan_mem_ref_ht.is_created ())
348 asan_mem_ref_ht.create (10);
350 return asan_mem_ref_ht;
353 /* Clear all entries from the memory references hash table. */
355 static void
356 empty_mem_ref_hash_table ()
358 if (asan_mem_ref_ht.is_created ())
359 asan_mem_ref_ht.empty ();
362 /* Free the memory references hash table. */
364 static void
365 free_mem_ref_resources ()
367 if (asan_mem_ref_ht.is_created ())
368 asan_mem_ref_ht.dispose ();
370 if (asan_mem_ref_alloc_pool)
372 free_alloc_pool (asan_mem_ref_alloc_pool);
373 asan_mem_ref_alloc_pool = NULL;
377 /* Return true iff the memory reference REF has been instrumented. */
379 static bool
380 has_mem_ref_been_instrumented (tree ref, char access_size)
382 asan_mem_ref r;
383 asan_mem_ref_init (&r, ref, access_size);
385 return (get_mem_ref_hash_table ().find (&r) != NULL);
388 /* Return true iff the memory reference REF has been instrumented. */
390 static bool
391 has_mem_ref_been_instrumented (const asan_mem_ref *ref)
393 return has_mem_ref_been_instrumented (ref->start, ref->access_size);
396 /* Return true iff access to memory region starting at REF and of
397 length LEN has been instrumented. */
399 static bool
400 has_mem_ref_been_instrumented (const asan_mem_ref *ref, tree len)
402 /* First let's see if the address of the beginning of REF has been
403 instrumented. */
404 if (!has_mem_ref_been_instrumented (ref))
405 return false;
407 if (len != 0)
409 /* Let's see if the end of the region has been instrumented. */
410 if (!has_mem_ref_been_instrumented (asan_mem_ref_get_end (ref, len),
411 ref->access_size))
412 return false;
414 return true;
417 /* Set REF to the memory reference present in a gimple assignment
418 ASSIGNMENT. Return true upon successful completion, false
419 otherwise. */
421 static bool
422 get_mem_ref_of_assignment (const gimple assignment,
423 asan_mem_ref *ref,
424 bool *ref_is_store)
426 gcc_assert (gimple_assign_single_p (assignment));
428 if (gimple_store_p (assignment)
429 && !gimple_clobber_p (assignment))
431 ref->start = gimple_assign_lhs (assignment);
432 *ref_is_store = true;
434 else if (gimple_assign_load_p (assignment))
436 ref->start = gimple_assign_rhs1 (assignment);
437 *ref_is_store = false;
439 else
440 return false;
442 ref->access_size = int_size_in_bytes (TREE_TYPE (ref->start));
443 return true;
446 /* Return the memory references contained in a gimple statement
447 representing a builtin call that has to do with memory access. */
449 static bool
450 get_mem_refs_of_builtin_call (const gimple call,
451 asan_mem_ref *src0,
452 tree *src0_len,
453 bool *src0_is_store,
454 asan_mem_ref *src1,
455 tree *src1_len,
456 bool *src1_is_store,
457 asan_mem_ref *dst,
458 tree *dst_len,
459 bool *dst_is_store,
460 bool *dest_is_deref)
462 gcc_checking_assert (gimple_call_builtin_p (call, BUILT_IN_NORMAL));
464 tree callee = gimple_call_fndecl (call);
465 tree source0 = NULL_TREE, source1 = NULL_TREE,
466 dest = NULL_TREE, len = NULL_TREE;
467 bool is_store = true, got_reference_p = false;
468 char access_size = 1;
470 switch (DECL_FUNCTION_CODE (callee))
472 /* (s, s, n) style memops. */
473 case BUILT_IN_BCMP:
474 case BUILT_IN_MEMCMP:
475 source0 = gimple_call_arg (call, 0);
476 source1 = gimple_call_arg (call, 1);
477 len = gimple_call_arg (call, 2);
478 break;
480 /* (src, dest, n) style memops. */
481 case BUILT_IN_BCOPY:
482 source0 = gimple_call_arg (call, 0);
483 dest = gimple_call_arg (call, 1);
484 len = gimple_call_arg (call, 2);
485 break;
487 /* (dest, src, n) style memops. */
488 case BUILT_IN_MEMCPY:
489 case BUILT_IN_MEMCPY_CHK:
490 case BUILT_IN_MEMMOVE:
491 case BUILT_IN_MEMMOVE_CHK:
492 case BUILT_IN_MEMPCPY:
493 case BUILT_IN_MEMPCPY_CHK:
494 dest = gimple_call_arg (call, 0);
495 source0 = gimple_call_arg (call, 1);
496 len = gimple_call_arg (call, 2);
497 break;
499 /* (dest, n) style memops. */
500 case BUILT_IN_BZERO:
501 dest = gimple_call_arg (call, 0);
502 len = gimple_call_arg (call, 1);
503 break;
505 /* (dest, x, n) style memops*/
506 case BUILT_IN_MEMSET:
507 case BUILT_IN_MEMSET_CHK:
508 dest = gimple_call_arg (call, 0);
509 len = gimple_call_arg (call, 2);
510 break;
512 case BUILT_IN_STRLEN:
513 source0 = gimple_call_arg (call, 0);
514 len = gimple_call_lhs (call);
515 break ;
517 /* And now the __atomic* and __sync builtins.
518 These are handled differently from the classical memory memory
519 access builtins above. */
521 case BUILT_IN_ATOMIC_LOAD_1:
522 case BUILT_IN_ATOMIC_LOAD_2:
523 case BUILT_IN_ATOMIC_LOAD_4:
524 case BUILT_IN_ATOMIC_LOAD_8:
525 case BUILT_IN_ATOMIC_LOAD_16:
526 is_store = false;
527 /* fall through. */
529 case BUILT_IN_SYNC_FETCH_AND_ADD_1:
530 case BUILT_IN_SYNC_FETCH_AND_ADD_2:
531 case BUILT_IN_SYNC_FETCH_AND_ADD_4:
532 case BUILT_IN_SYNC_FETCH_AND_ADD_8:
533 case BUILT_IN_SYNC_FETCH_AND_ADD_16:
535 case BUILT_IN_SYNC_FETCH_AND_SUB_1:
536 case BUILT_IN_SYNC_FETCH_AND_SUB_2:
537 case BUILT_IN_SYNC_FETCH_AND_SUB_4:
538 case BUILT_IN_SYNC_FETCH_AND_SUB_8:
539 case BUILT_IN_SYNC_FETCH_AND_SUB_16:
541 case BUILT_IN_SYNC_FETCH_AND_OR_1:
542 case BUILT_IN_SYNC_FETCH_AND_OR_2:
543 case BUILT_IN_SYNC_FETCH_AND_OR_4:
544 case BUILT_IN_SYNC_FETCH_AND_OR_8:
545 case BUILT_IN_SYNC_FETCH_AND_OR_16:
547 case BUILT_IN_SYNC_FETCH_AND_AND_1:
548 case BUILT_IN_SYNC_FETCH_AND_AND_2:
549 case BUILT_IN_SYNC_FETCH_AND_AND_4:
550 case BUILT_IN_SYNC_FETCH_AND_AND_8:
551 case BUILT_IN_SYNC_FETCH_AND_AND_16:
553 case BUILT_IN_SYNC_FETCH_AND_XOR_1:
554 case BUILT_IN_SYNC_FETCH_AND_XOR_2:
555 case BUILT_IN_SYNC_FETCH_AND_XOR_4:
556 case BUILT_IN_SYNC_FETCH_AND_XOR_8:
557 case BUILT_IN_SYNC_FETCH_AND_XOR_16:
559 case BUILT_IN_SYNC_FETCH_AND_NAND_1:
560 case BUILT_IN_SYNC_FETCH_AND_NAND_2:
561 case BUILT_IN_SYNC_FETCH_AND_NAND_4:
562 case BUILT_IN_SYNC_FETCH_AND_NAND_8:
564 case BUILT_IN_SYNC_ADD_AND_FETCH_1:
565 case BUILT_IN_SYNC_ADD_AND_FETCH_2:
566 case BUILT_IN_SYNC_ADD_AND_FETCH_4:
567 case BUILT_IN_SYNC_ADD_AND_FETCH_8:
568 case BUILT_IN_SYNC_ADD_AND_FETCH_16:
570 case BUILT_IN_SYNC_SUB_AND_FETCH_1:
571 case BUILT_IN_SYNC_SUB_AND_FETCH_2:
572 case BUILT_IN_SYNC_SUB_AND_FETCH_4:
573 case BUILT_IN_SYNC_SUB_AND_FETCH_8:
574 case BUILT_IN_SYNC_SUB_AND_FETCH_16:
576 case BUILT_IN_SYNC_OR_AND_FETCH_1:
577 case BUILT_IN_SYNC_OR_AND_FETCH_2:
578 case BUILT_IN_SYNC_OR_AND_FETCH_4:
579 case BUILT_IN_SYNC_OR_AND_FETCH_8:
580 case BUILT_IN_SYNC_OR_AND_FETCH_16:
582 case BUILT_IN_SYNC_AND_AND_FETCH_1:
583 case BUILT_IN_SYNC_AND_AND_FETCH_2:
584 case BUILT_IN_SYNC_AND_AND_FETCH_4:
585 case BUILT_IN_SYNC_AND_AND_FETCH_8:
586 case BUILT_IN_SYNC_AND_AND_FETCH_16:
588 case BUILT_IN_SYNC_XOR_AND_FETCH_1:
589 case BUILT_IN_SYNC_XOR_AND_FETCH_2:
590 case BUILT_IN_SYNC_XOR_AND_FETCH_4:
591 case BUILT_IN_SYNC_XOR_AND_FETCH_8:
592 case BUILT_IN_SYNC_XOR_AND_FETCH_16:
594 case BUILT_IN_SYNC_NAND_AND_FETCH_1:
595 case BUILT_IN_SYNC_NAND_AND_FETCH_2:
596 case BUILT_IN_SYNC_NAND_AND_FETCH_4:
597 case BUILT_IN_SYNC_NAND_AND_FETCH_8:
599 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_1:
600 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_2:
601 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_4:
602 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_8:
603 case BUILT_IN_SYNC_BOOL_COMPARE_AND_SWAP_16:
605 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_1:
606 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_2:
607 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_4:
608 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_8:
609 case BUILT_IN_SYNC_VAL_COMPARE_AND_SWAP_16:
611 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_1:
612 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_2:
613 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_4:
614 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_8:
615 case BUILT_IN_SYNC_LOCK_TEST_AND_SET_16:
617 case BUILT_IN_SYNC_LOCK_RELEASE_1:
618 case BUILT_IN_SYNC_LOCK_RELEASE_2:
619 case BUILT_IN_SYNC_LOCK_RELEASE_4:
620 case BUILT_IN_SYNC_LOCK_RELEASE_8:
621 case BUILT_IN_SYNC_LOCK_RELEASE_16:
623 case BUILT_IN_ATOMIC_EXCHANGE_1:
624 case BUILT_IN_ATOMIC_EXCHANGE_2:
625 case BUILT_IN_ATOMIC_EXCHANGE_4:
626 case BUILT_IN_ATOMIC_EXCHANGE_8:
627 case BUILT_IN_ATOMIC_EXCHANGE_16:
629 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_1:
630 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_2:
631 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_4:
632 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_8:
633 case BUILT_IN_ATOMIC_COMPARE_EXCHANGE_16:
635 case BUILT_IN_ATOMIC_STORE_1:
636 case BUILT_IN_ATOMIC_STORE_2:
637 case BUILT_IN_ATOMIC_STORE_4:
638 case BUILT_IN_ATOMIC_STORE_8:
639 case BUILT_IN_ATOMIC_STORE_16:
641 case BUILT_IN_ATOMIC_ADD_FETCH_1:
642 case BUILT_IN_ATOMIC_ADD_FETCH_2:
643 case BUILT_IN_ATOMIC_ADD_FETCH_4:
644 case BUILT_IN_ATOMIC_ADD_FETCH_8:
645 case BUILT_IN_ATOMIC_ADD_FETCH_16:
647 case BUILT_IN_ATOMIC_SUB_FETCH_1:
648 case BUILT_IN_ATOMIC_SUB_FETCH_2:
649 case BUILT_IN_ATOMIC_SUB_FETCH_4:
650 case BUILT_IN_ATOMIC_SUB_FETCH_8:
651 case BUILT_IN_ATOMIC_SUB_FETCH_16:
653 case BUILT_IN_ATOMIC_AND_FETCH_1:
654 case BUILT_IN_ATOMIC_AND_FETCH_2:
655 case BUILT_IN_ATOMIC_AND_FETCH_4:
656 case BUILT_IN_ATOMIC_AND_FETCH_8:
657 case BUILT_IN_ATOMIC_AND_FETCH_16:
659 case BUILT_IN_ATOMIC_NAND_FETCH_1:
660 case BUILT_IN_ATOMIC_NAND_FETCH_2:
661 case BUILT_IN_ATOMIC_NAND_FETCH_4:
662 case BUILT_IN_ATOMIC_NAND_FETCH_8:
663 case BUILT_IN_ATOMIC_NAND_FETCH_16:
665 case BUILT_IN_ATOMIC_XOR_FETCH_1:
666 case BUILT_IN_ATOMIC_XOR_FETCH_2:
667 case BUILT_IN_ATOMIC_XOR_FETCH_4:
668 case BUILT_IN_ATOMIC_XOR_FETCH_8:
669 case BUILT_IN_ATOMIC_XOR_FETCH_16:
671 case BUILT_IN_ATOMIC_OR_FETCH_1:
672 case BUILT_IN_ATOMIC_OR_FETCH_2:
673 case BUILT_IN_ATOMIC_OR_FETCH_4:
674 case BUILT_IN_ATOMIC_OR_FETCH_8:
675 case BUILT_IN_ATOMIC_OR_FETCH_16:
677 case BUILT_IN_ATOMIC_FETCH_ADD_1:
678 case BUILT_IN_ATOMIC_FETCH_ADD_2:
679 case BUILT_IN_ATOMIC_FETCH_ADD_4:
680 case BUILT_IN_ATOMIC_FETCH_ADD_8:
681 case BUILT_IN_ATOMIC_FETCH_ADD_16:
683 case BUILT_IN_ATOMIC_FETCH_SUB_1:
684 case BUILT_IN_ATOMIC_FETCH_SUB_2:
685 case BUILT_IN_ATOMIC_FETCH_SUB_4:
686 case BUILT_IN_ATOMIC_FETCH_SUB_8:
687 case BUILT_IN_ATOMIC_FETCH_SUB_16:
689 case BUILT_IN_ATOMIC_FETCH_AND_1:
690 case BUILT_IN_ATOMIC_FETCH_AND_2:
691 case BUILT_IN_ATOMIC_FETCH_AND_4:
692 case BUILT_IN_ATOMIC_FETCH_AND_8:
693 case BUILT_IN_ATOMIC_FETCH_AND_16:
695 case BUILT_IN_ATOMIC_FETCH_NAND_1:
696 case BUILT_IN_ATOMIC_FETCH_NAND_2:
697 case BUILT_IN_ATOMIC_FETCH_NAND_4:
698 case BUILT_IN_ATOMIC_FETCH_NAND_8:
699 case BUILT_IN_ATOMIC_FETCH_NAND_16:
701 case BUILT_IN_ATOMIC_FETCH_XOR_1:
702 case BUILT_IN_ATOMIC_FETCH_XOR_2:
703 case BUILT_IN_ATOMIC_FETCH_XOR_4:
704 case BUILT_IN_ATOMIC_FETCH_XOR_8:
705 case BUILT_IN_ATOMIC_FETCH_XOR_16:
707 case BUILT_IN_ATOMIC_FETCH_OR_1:
708 case BUILT_IN_ATOMIC_FETCH_OR_2:
709 case BUILT_IN_ATOMIC_FETCH_OR_4:
710 case BUILT_IN_ATOMIC_FETCH_OR_8:
711 case BUILT_IN_ATOMIC_FETCH_OR_16:
713 dest = gimple_call_arg (call, 0);
714 /* DEST represents the address of a memory location.
715 instrument_derefs wants the memory location, so lets
716 dereference the address DEST before handing it to
717 instrument_derefs. */
718 if (TREE_CODE (dest) == ADDR_EXPR)
719 dest = TREE_OPERAND (dest, 0);
720 else if (TREE_CODE (dest) == SSA_NAME || TREE_CODE (dest) == INTEGER_CST)
721 dest = build2 (MEM_REF, TREE_TYPE (TREE_TYPE (dest)),
722 dest, build_int_cst (TREE_TYPE (dest), 0));
723 else
724 gcc_unreachable ();
726 access_size = int_size_in_bytes (TREE_TYPE (dest));
729 default:
730 /* The other builtins memory access are not instrumented in this
731 function because they either don't have any length parameter,
732 or their length parameter is just a limit. */
733 break;
736 if (len != NULL_TREE)
738 if (source0 != NULL_TREE)
740 src0->start = source0;
741 src0->access_size = access_size;
742 *src0_len = len;
743 *src0_is_store = false;
746 if (source1 != NULL_TREE)
748 src1->start = source1;
749 src1->access_size = access_size;
750 *src1_len = len;
751 *src1_is_store = false;
754 if (dest != NULL_TREE)
756 dst->start = dest;
757 dst->access_size = access_size;
758 *dst_len = len;
759 *dst_is_store = true;
762 got_reference_p = true;
764 else if (dest)
766 dst->start = dest;
767 dst->access_size = access_size;
768 *dst_len = NULL_TREE;
769 *dst_is_store = is_store;
770 *dest_is_deref = true;
771 got_reference_p = true;
774 return got_reference_p;
777 /* Return true iff a given gimple statement has been instrumented.
778 Note that the statement is "defined" by the memory references it
779 contains. */
781 static bool
782 has_stmt_been_instrumented_p (gimple stmt)
784 if (gimple_assign_single_p (stmt))
786 bool r_is_store;
787 asan_mem_ref r;
788 asan_mem_ref_init (&r, NULL, 1);
790 if (get_mem_ref_of_assignment (stmt, &r, &r_is_store))
791 return has_mem_ref_been_instrumented (&r);
793 else if (gimple_call_builtin_p (stmt, BUILT_IN_NORMAL))
795 asan_mem_ref src0, src1, dest;
796 asan_mem_ref_init (&src0, NULL, 1);
797 asan_mem_ref_init (&src1, NULL, 1);
798 asan_mem_ref_init (&dest, NULL, 1);
800 tree src0_len = NULL_TREE, src1_len = NULL_TREE, dest_len = NULL_TREE;
801 bool src0_is_store = false, src1_is_store = false,
802 dest_is_store = false, dest_is_deref = false;
803 if (get_mem_refs_of_builtin_call (stmt,
804 &src0, &src0_len, &src0_is_store,
805 &src1, &src1_len, &src1_is_store,
806 &dest, &dest_len, &dest_is_store,
807 &dest_is_deref))
809 if (src0.start != NULL_TREE
810 && !has_mem_ref_been_instrumented (&src0, src0_len))
811 return false;
813 if (src1.start != NULL_TREE
814 && !has_mem_ref_been_instrumented (&src1, src1_len))
815 return false;
817 if (dest.start != NULL_TREE
818 && !has_mem_ref_been_instrumented (&dest, dest_len))
819 return false;
821 return true;
824 return false;
827 /* Insert a memory reference into the hash table. */
829 static void
830 update_mem_ref_hash_table (tree ref, char access_size)
832 hash_table <asan_mem_ref_hasher> ht = get_mem_ref_hash_table ();
834 asan_mem_ref r;
835 asan_mem_ref_init (&r, ref, access_size);
837 asan_mem_ref **slot = ht.find_slot (&r, INSERT);
838 if (*slot == NULL)
839 *slot = asan_mem_ref_new (ref, access_size);
842 /* Initialize shadow_ptr_types array. */
844 static void
845 asan_init_shadow_ptr_types (void)
847 asan_shadow_set = new_alias_set ();
848 shadow_ptr_types[0] = build_distinct_type_copy (signed_char_type_node);
849 TYPE_ALIAS_SET (shadow_ptr_types[0]) = asan_shadow_set;
850 shadow_ptr_types[0] = build_pointer_type (shadow_ptr_types[0]);
851 shadow_ptr_types[1] = build_distinct_type_copy (short_integer_type_node);
852 TYPE_ALIAS_SET (shadow_ptr_types[1]) = asan_shadow_set;
853 shadow_ptr_types[1] = build_pointer_type (shadow_ptr_types[1]);
854 initialize_sanitizer_builtins ();
857 /* Create ADDR_EXPR of STRING_CST with the PP pretty printer text. */
859 static tree
860 asan_pp_string (pretty_printer *pp)
862 const char *buf = pp_formatted_text (pp);
863 size_t len = strlen (buf);
864 tree ret = build_string (len + 1, buf);
865 TREE_TYPE (ret)
866 = build_array_type (TREE_TYPE (shadow_ptr_types[0]),
867 build_index_type (size_int (len)));
868 TREE_READONLY (ret) = 1;
869 TREE_STATIC (ret) = 1;
870 return build1 (ADDR_EXPR, shadow_ptr_types[0], ret);
873 /* Return a CONST_INT representing 4 subsequent shadow memory bytes. */
875 static rtx
876 asan_shadow_cst (unsigned char shadow_bytes[4])
878 int i;
879 unsigned HOST_WIDE_INT val = 0;
880 gcc_assert (WORDS_BIG_ENDIAN == BYTES_BIG_ENDIAN);
881 for (i = 0; i < 4; i++)
882 val |= (unsigned HOST_WIDE_INT) shadow_bytes[BYTES_BIG_ENDIAN ? 3 - i : i]
883 << (BITS_PER_UNIT * i);
884 return gen_int_mode (val, SImode);
887 /* Clear shadow memory at SHADOW_MEM, LEN bytes. Can't call a library call here
888 though. */
890 static void
891 asan_clear_shadow (rtx shadow_mem, HOST_WIDE_INT len)
893 rtx insn, insns, top_label, end, addr, tmp, jump;
895 start_sequence ();
896 clear_storage (shadow_mem, GEN_INT (len), BLOCK_OP_NORMAL);
897 insns = get_insns ();
898 end_sequence ();
899 for (insn = insns; insn; insn = NEXT_INSN (insn))
900 if (CALL_P (insn))
901 break;
902 if (insn == NULL_RTX)
904 emit_insn (insns);
905 return;
908 gcc_assert ((len & 3) == 0);
909 top_label = gen_label_rtx ();
910 addr = copy_to_mode_reg (Pmode, XEXP (shadow_mem, 0));
911 shadow_mem = adjust_automodify_address (shadow_mem, SImode, addr, 0);
912 end = force_reg (Pmode, plus_constant (Pmode, addr, len));
913 emit_label (top_label);
915 emit_move_insn (shadow_mem, const0_rtx);
916 tmp = expand_simple_binop (Pmode, PLUS, addr, gen_int_mode (4, Pmode), addr,
917 true, OPTAB_LIB_WIDEN);
918 if (tmp != addr)
919 emit_move_insn (addr, tmp);
920 emit_cmp_and_jump_insns (addr, end, LT, NULL_RTX, Pmode, true, top_label);
921 jump = get_last_insn ();
922 gcc_assert (JUMP_P (jump));
923 add_int_reg_note (jump, REG_BR_PROB, REG_BR_PROB_BASE * 80 / 100);
926 void
927 asan_function_start (void)
929 section *fnsec = function_section (current_function_decl);
930 switch_to_section (fnsec);
931 ASM_OUTPUT_DEBUG_LABEL (asm_out_file, "LASANPC",
932 current_function_funcdef_no);
935 /* Insert code to protect stack vars. The prologue sequence should be emitted
936 directly, epilogue sequence returned. BASE is the register holding the
937 stack base, against which OFFSETS array offsets are relative to, OFFSETS
938 array contains pairs of offsets in reverse order, always the end offset
939 of some gap that needs protection followed by starting offset,
940 and DECLS is an array of representative decls for each var partition.
941 LENGTH is the length of the OFFSETS array, DECLS array is LENGTH / 2 - 1
942 elements long (OFFSETS include gap before the first variable as well
943 as gaps after each stack variable). */
946 asan_emit_stack_protection (rtx base, HOST_WIDE_INT *offsets, tree *decls,
947 int length)
949 rtx shadow_base, shadow_mem, ret, mem;
950 char buf[30];
951 unsigned char shadow_bytes[4];
952 HOST_WIDE_INT base_offset = offsets[length - 1], offset, prev_offset;
953 HOST_WIDE_INT last_offset, last_size;
954 int l;
955 unsigned char cur_shadow_byte = ASAN_STACK_MAGIC_LEFT;
956 tree str_cst, decl, id;
958 if (shadow_ptr_types[0] == NULL_TREE)
959 asan_init_shadow_ptr_types ();
961 /* First of all, prepare the description string. */
962 pretty_printer asan_pp;
964 pp_decimal_int (&asan_pp, length / 2 - 1);
965 pp_space (&asan_pp);
966 for (l = length - 2; l; l -= 2)
968 tree decl = decls[l / 2 - 1];
969 pp_wide_integer (&asan_pp, offsets[l] - base_offset);
970 pp_space (&asan_pp);
971 pp_wide_integer (&asan_pp, offsets[l - 1] - offsets[l]);
972 pp_space (&asan_pp);
973 if (DECL_P (decl) && DECL_NAME (decl))
975 pp_decimal_int (&asan_pp, IDENTIFIER_LENGTH (DECL_NAME (decl)));
976 pp_space (&asan_pp);
977 pp_tree_identifier (&asan_pp, DECL_NAME (decl));
979 else
980 pp_string (&asan_pp, "9 <unknown>");
981 pp_space (&asan_pp);
983 str_cst = asan_pp_string (&asan_pp);
985 /* Emit the prologue sequence. */
986 base = expand_binop (Pmode, add_optab, base,
987 gen_int_mode (base_offset, Pmode),
988 NULL_RTX, 1, OPTAB_DIRECT);
989 mem = gen_rtx_MEM (ptr_mode, base);
990 emit_move_insn (mem, gen_int_mode (ASAN_STACK_FRAME_MAGIC, ptr_mode));
991 mem = adjust_address (mem, VOIDmode, GET_MODE_SIZE (ptr_mode));
992 emit_move_insn (mem, expand_normal (str_cst));
993 mem = adjust_address (mem, VOIDmode, GET_MODE_SIZE (ptr_mode));
994 ASM_GENERATE_INTERNAL_LABEL (buf, "LASANPC", current_function_funcdef_no);
995 id = get_identifier (buf);
996 decl = build_decl (DECL_SOURCE_LOCATION (current_function_decl),
997 VAR_DECL, id, char_type_node);
998 SET_DECL_ASSEMBLER_NAME (decl, id);
999 TREE_ADDRESSABLE (decl) = 1;
1000 TREE_READONLY (decl) = 1;
1001 DECL_ARTIFICIAL (decl) = 1;
1002 DECL_IGNORED_P (decl) = 1;
1003 TREE_STATIC (decl) = 1;
1004 TREE_PUBLIC (decl) = 0;
1005 TREE_USED (decl) = 1;
1006 DECL_INITIAL (decl) = decl;
1007 TREE_ASM_WRITTEN (decl) = 1;
1008 TREE_ASM_WRITTEN (id) = 1;
1009 emit_move_insn (mem, expand_normal (build_fold_addr_expr (decl)));
1010 shadow_base = expand_binop (Pmode, lshr_optab, base,
1011 GEN_INT (ASAN_SHADOW_SHIFT),
1012 NULL_RTX, 1, OPTAB_DIRECT);
1013 shadow_base = expand_binop (Pmode, add_optab, shadow_base,
1014 gen_int_mode (targetm.asan_shadow_offset (),
1015 Pmode),
1016 NULL_RTX, 1, OPTAB_DIRECT);
1017 gcc_assert (asan_shadow_set != -1
1018 && (ASAN_RED_ZONE_SIZE >> ASAN_SHADOW_SHIFT) == 4);
1019 shadow_mem = gen_rtx_MEM (SImode, shadow_base);
1020 set_mem_alias_set (shadow_mem, asan_shadow_set);
1021 prev_offset = base_offset;
1022 for (l = length; l; l -= 2)
1024 if (l == 2)
1025 cur_shadow_byte = ASAN_STACK_MAGIC_RIGHT;
1026 offset = offsets[l - 1];
1027 if ((offset - base_offset) & (ASAN_RED_ZONE_SIZE - 1))
1029 int i;
1030 HOST_WIDE_INT aoff
1031 = base_offset + ((offset - base_offset)
1032 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1));
1033 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1034 (aoff - prev_offset)
1035 >> ASAN_SHADOW_SHIFT);
1036 prev_offset = aoff;
1037 for (i = 0; i < 4; i++, aoff += (1 << ASAN_SHADOW_SHIFT))
1038 if (aoff < offset)
1040 if (aoff < offset - (1 << ASAN_SHADOW_SHIFT) + 1)
1041 shadow_bytes[i] = 0;
1042 else
1043 shadow_bytes[i] = offset - aoff;
1045 else
1046 shadow_bytes[i] = ASAN_STACK_MAGIC_PARTIAL;
1047 emit_move_insn (shadow_mem, asan_shadow_cst (shadow_bytes));
1048 offset = aoff;
1050 while (offset <= offsets[l - 2] - ASAN_RED_ZONE_SIZE)
1052 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1053 (offset - prev_offset)
1054 >> ASAN_SHADOW_SHIFT);
1055 prev_offset = offset;
1056 memset (shadow_bytes, cur_shadow_byte, 4);
1057 emit_move_insn (shadow_mem, asan_shadow_cst (shadow_bytes));
1058 offset += ASAN_RED_ZONE_SIZE;
1060 cur_shadow_byte = ASAN_STACK_MAGIC_MIDDLE;
1062 do_pending_stack_adjust ();
1064 /* Construct epilogue sequence. */
1065 start_sequence ();
1067 shadow_mem = gen_rtx_MEM (BLKmode, shadow_base);
1068 set_mem_alias_set (shadow_mem, asan_shadow_set);
1069 prev_offset = base_offset;
1070 last_offset = base_offset;
1071 last_size = 0;
1072 for (l = length; l; l -= 2)
1074 offset = base_offset + ((offsets[l - 1] - base_offset)
1075 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1));
1076 if (last_offset + last_size != offset)
1078 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1079 (last_offset - prev_offset)
1080 >> ASAN_SHADOW_SHIFT);
1081 prev_offset = last_offset;
1082 asan_clear_shadow (shadow_mem, last_size >> ASAN_SHADOW_SHIFT);
1083 last_offset = offset;
1084 last_size = 0;
1086 last_size += base_offset + ((offsets[l - 2] - base_offset)
1087 & ~(ASAN_RED_ZONE_SIZE - HOST_WIDE_INT_1))
1088 - offset;
1090 if (last_size)
1092 shadow_mem = adjust_address (shadow_mem, VOIDmode,
1093 (last_offset - prev_offset)
1094 >> ASAN_SHADOW_SHIFT);
1095 asan_clear_shadow (shadow_mem, last_size >> ASAN_SHADOW_SHIFT);
1098 do_pending_stack_adjust ();
1100 ret = get_insns ();
1101 end_sequence ();
1102 return ret;
1105 /* Return true if DECL, a global var, might be overridden and needs
1106 therefore a local alias. */
1108 static bool
1109 asan_needs_local_alias (tree decl)
1111 return DECL_WEAK (decl) || !targetm.binds_local_p (decl);
1114 /* Return true if DECL is a VAR_DECL that should be protected
1115 by Address Sanitizer, by appending a red zone with protected
1116 shadow memory after it and aligning it to at least
1117 ASAN_RED_ZONE_SIZE bytes. */
1119 bool
1120 asan_protect_global (tree decl)
1122 rtx rtl, symbol;
1124 if (TREE_CODE (decl) == STRING_CST)
1126 /* Instrument all STRING_CSTs except those created
1127 by asan_pp_string here. */
1128 if (shadow_ptr_types[0] != NULL_TREE
1129 && TREE_CODE (TREE_TYPE (decl)) == ARRAY_TYPE
1130 && TREE_TYPE (TREE_TYPE (decl)) == TREE_TYPE (shadow_ptr_types[0]))
1131 return false;
1132 return true;
1134 if (TREE_CODE (decl) != VAR_DECL
1135 /* TLS vars aren't statically protectable. */
1136 || DECL_THREAD_LOCAL_P (decl)
1137 /* Externs will be protected elsewhere. */
1138 || DECL_EXTERNAL (decl)
1139 || !DECL_RTL_SET_P (decl)
1140 /* Comdat vars pose an ABI problem, we can't know if
1141 the var that is selected by the linker will have
1142 padding or not. */
1143 || DECL_ONE_ONLY (decl)
1144 /* Similarly for common vars. People can use -fno-common. */
1145 || (DECL_COMMON (decl) && TREE_PUBLIC (decl))
1146 /* Don't protect if using user section, often vars placed
1147 into user section from multiple TUs are then assumed
1148 to be an array of such vars, putting padding in there
1149 breaks this assumption. */
1150 || (DECL_SECTION_NAME (decl) != NULL_TREE
1151 && !DECL_HAS_IMPLICIT_SECTION_NAME_P (decl))
1152 || DECL_SIZE (decl) == 0
1153 || ASAN_RED_ZONE_SIZE * BITS_PER_UNIT > MAX_OFILE_ALIGNMENT
1154 || !valid_constant_size_p (DECL_SIZE_UNIT (decl))
1155 || DECL_ALIGN_UNIT (decl) > 2 * ASAN_RED_ZONE_SIZE)
1156 return false;
1158 rtl = DECL_RTL (decl);
1159 if (!MEM_P (rtl) || GET_CODE (XEXP (rtl, 0)) != SYMBOL_REF)
1160 return false;
1161 symbol = XEXP (rtl, 0);
1163 if (CONSTANT_POOL_ADDRESS_P (symbol)
1164 || TREE_CONSTANT_POOL_ADDRESS_P (symbol))
1165 return false;
1167 if (lookup_attribute ("weakref", DECL_ATTRIBUTES (decl)))
1168 return false;
1170 #ifndef ASM_OUTPUT_DEF
1171 if (asan_needs_local_alias (decl))
1172 return false;
1173 #endif
1175 return true;
1178 /* Construct a function tree for __asan_report_{load,store}{1,2,4,8,16}.
1179 IS_STORE is either 1 (for a store) or 0 (for a load).
1180 SIZE_IN_BYTES is one of 1, 2, 4, 8, 16. */
1182 static tree
1183 report_error_func (bool is_store, int size_in_bytes)
1185 static enum built_in_function report[2][5]
1186 = { { BUILT_IN_ASAN_REPORT_LOAD1, BUILT_IN_ASAN_REPORT_LOAD2,
1187 BUILT_IN_ASAN_REPORT_LOAD4, BUILT_IN_ASAN_REPORT_LOAD8,
1188 BUILT_IN_ASAN_REPORT_LOAD16 },
1189 { BUILT_IN_ASAN_REPORT_STORE1, BUILT_IN_ASAN_REPORT_STORE2,
1190 BUILT_IN_ASAN_REPORT_STORE4, BUILT_IN_ASAN_REPORT_STORE8,
1191 BUILT_IN_ASAN_REPORT_STORE16 } };
1192 return builtin_decl_implicit (report[is_store][exact_log2 (size_in_bytes)]);
1195 #define PROB_VERY_UNLIKELY (REG_BR_PROB_BASE / 2000 - 1)
1196 #define PROB_ALWAYS (REG_BR_PROB_BASE)
1198 /* Split the current basic block and create a condition statement
1199 insertion point right before or after the statement pointed to by
1200 ITER. Return an iterator to the point at which the caller might
1201 safely insert the condition statement.
1203 THEN_BLOCK must be set to the address of an uninitialized instance
1204 of basic_block. The function will then set *THEN_BLOCK to the
1205 'then block' of the condition statement to be inserted by the
1206 caller.
1208 If CREATE_THEN_FALLTHRU_EDGE is false, no edge will be created from
1209 *THEN_BLOCK to *FALLTHROUGH_BLOCK.
1211 Similarly, the function will set *FALLTRHOUGH_BLOCK to the 'else
1212 block' of the condition statement to be inserted by the caller.
1214 Note that *FALLTHROUGH_BLOCK is a new block that contains the
1215 statements starting from *ITER, and *THEN_BLOCK is a new empty
1216 block.
1218 *ITER is adjusted to point to always point to the first statement
1219 of the basic block * FALLTHROUGH_BLOCK. That statement is the
1220 same as what ITER was pointing to prior to calling this function,
1221 if BEFORE_P is true; otherwise, it is its following statement. */
1223 static gimple_stmt_iterator
1224 create_cond_insert_point (gimple_stmt_iterator *iter,
1225 bool before_p,
1226 bool then_more_likely_p,
1227 bool create_then_fallthru_edge,
1228 basic_block *then_block,
1229 basic_block *fallthrough_block)
1231 gimple_stmt_iterator gsi = *iter;
1233 if (!gsi_end_p (gsi) && before_p)
1234 gsi_prev (&gsi);
1236 basic_block cur_bb = gsi_bb (*iter);
1238 edge e = split_block (cur_bb, gsi_stmt (gsi));
1240 /* Get a hold on the 'condition block', the 'then block' and the
1241 'else block'. */
1242 basic_block cond_bb = e->src;
1243 basic_block fallthru_bb = e->dest;
1244 basic_block then_bb = create_empty_bb (cond_bb);
1245 if (current_loops)
1247 add_bb_to_loop (then_bb, cond_bb->loop_father);
1248 loops_state_set (LOOPS_NEED_FIXUP);
1251 /* Set up the newly created 'then block'. */
1252 e = make_edge (cond_bb, then_bb, EDGE_TRUE_VALUE);
1253 int fallthrough_probability
1254 = then_more_likely_p
1255 ? PROB_VERY_UNLIKELY
1256 : PROB_ALWAYS - PROB_VERY_UNLIKELY;
1257 e->probability = PROB_ALWAYS - fallthrough_probability;
1258 if (create_then_fallthru_edge)
1259 make_single_succ_edge (then_bb, fallthru_bb, EDGE_FALLTHRU);
1261 /* Set up the fallthrough basic block. */
1262 e = find_edge (cond_bb, fallthru_bb);
1263 e->flags = EDGE_FALSE_VALUE;
1264 e->count = cond_bb->count;
1265 e->probability = fallthrough_probability;
1267 /* Update dominance info for the newly created then_bb; note that
1268 fallthru_bb's dominance info has already been updated by
1269 split_bock. */
1270 if (dom_info_available_p (CDI_DOMINATORS))
1271 set_immediate_dominator (CDI_DOMINATORS, then_bb, cond_bb);
1273 *then_block = then_bb;
1274 *fallthrough_block = fallthru_bb;
1275 *iter = gsi_start_bb (fallthru_bb);
1277 return gsi_last_bb (cond_bb);
1280 /* Insert an if condition followed by a 'then block' right before the
1281 statement pointed to by ITER. The fallthrough block -- which is the
1282 else block of the condition as well as the destination of the
1283 outcoming edge of the 'then block' -- starts with the statement
1284 pointed to by ITER.
1286 COND is the condition of the if.
1288 If THEN_MORE_LIKELY_P is true, the probability of the edge to the
1289 'then block' is higher than the probability of the edge to the
1290 fallthrough block.
1292 Upon completion of the function, *THEN_BB is set to the newly
1293 inserted 'then block' and similarly, *FALLTHROUGH_BB is set to the
1294 fallthrough block.
1296 *ITER is adjusted to still point to the same statement it was
1297 pointing to initially. */
1299 static void
1300 insert_if_then_before_iter (gimple cond,
1301 gimple_stmt_iterator *iter,
1302 bool then_more_likely_p,
1303 basic_block *then_bb,
1304 basic_block *fallthrough_bb)
1306 gimple_stmt_iterator cond_insert_point =
1307 create_cond_insert_point (iter,
1308 /*before_p=*/true,
1309 then_more_likely_p,
1310 /*create_then_fallthru_edge=*/true,
1311 then_bb,
1312 fallthrough_bb);
1313 gsi_insert_after (&cond_insert_point, cond, GSI_NEW_STMT);
1316 /* Instrument the memory access instruction BASE. Insert new
1317 statements before or after ITER.
1319 Note that the memory access represented by BASE can be either an
1320 SSA_NAME, or a non-SSA expression. LOCATION is the source code
1321 location. IS_STORE is TRUE for a store, FALSE for a load.
1322 BEFORE_P is TRUE for inserting the instrumentation code before
1323 ITER, FALSE for inserting it after ITER. SIZE_IN_BYTES is one of
1324 1, 2, 4, 8, 16.
1326 If BEFORE_P is TRUE, *ITER is arranged to still point to the
1327 statement it was pointing to prior to calling this function,
1328 otherwise, it points to the statement logically following it. */
1330 static void
1331 build_check_stmt (location_t location, tree base, gimple_stmt_iterator *iter,
1332 bool before_p, bool is_store, int size_in_bytes)
1334 gimple_stmt_iterator gsi;
1335 basic_block then_bb, else_bb;
1336 tree t, base_addr, shadow;
1337 gimple g;
1338 tree shadow_ptr_type = shadow_ptr_types[size_in_bytes == 16 ? 1 : 0];
1339 tree shadow_type = TREE_TYPE (shadow_ptr_type);
1340 tree uintptr_type
1341 = build_nonstandard_integer_type (TYPE_PRECISION (TREE_TYPE (base)), 1);
1342 tree base_ssa = base;
1344 /* Get an iterator on the point where we can add the condition
1345 statement for the instrumentation. */
1346 gsi = create_cond_insert_point (iter, before_p,
1347 /*then_more_likely_p=*/false,
1348 /*create_then_fallthru_edge=*/false,
1349 &then_bb,
1350 &else_bb);
1352 base = unshare_expr (base);
1354 /* BASE can already be an SSA_NAME; in that case, do not create a
1355 new SSA_NAME for it. */
1356 if (TREE_CODE (base) != SSA_NAME)
1358 g = gimple_build_assign_with_ops (TREE_CODE (base),
1359 make_ssa_name (TREE_TYPE (base), NULL),
1360 base, NULL_TREE);
1361 gimple_set_location (g, location);
1362 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1363 base_ssa = gimple_assign_lhs (g);
1366 g = gimple_build_assign_with_ops (NOP_EXPR,
1367 make_ssa_name (uintptr_type, NULL),
1368 base_ssa, NULL_TREE);
1369 gimple_set_location (g, location);
1370 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1371 base_addr = gimple_assign_lhs (g);
1373 /* Build
1374 (base_addr >> ASAN_SHADOW_SHIFT) + targetm.asan_shadow_offset (). */
1376 t = build_int_cst (uintptr_type, ASAN_SHADOW_SHIFT);
1377 g = gimple_build_assign_with_ops (RSHIFT_EXPR,
1378 make_ssa_name (uintptr_type, NULL),
1379 base_addr, t);
1380 gimple_set_location (g, location);
1381 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1383 t = build_int_cst (uintptr_type, targetm.asan_shadow_offset ());
1384 g = gimple_build_assign_with_ops (PLUS_EXPR,
1385 make_ssa_name (uintptr_type, NULL),
1386 gimple_assign_lhs (g), t);
1387 gimple_set_location (g, location);
1388 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1390 g = gimple_build_assign_with_ops (NOP_EXPR,
1391 make_ssa_name (shadow_ptr_type, NULL),
1392 gimple_assign_lhs (g), NULL_TREE);
1393 gimple_set_location (g, location);
1394 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1396 t = build2 (MEM_REF, shadow_type, gimple_assign_lhs (g),
1397 build_int_cst (shadow_ptr_type, 0));
1398 g = gimple_build_assign_with_ops (MEM_REF,
1399 make_ssa_name (shadow_type, NULL),
1400 t, NULL_TREE);
1401 gimple_set_location (g, location);
1402 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1403 shadow = gimple_assign_lhs (g);
1405 if (size_in_bytes < 8)
1407 /* Slow path for 1, 2 and 4 byte accesses.
1408 Test (shadow != 0)
1409 & ((base_addr & 7) + (size_in_bytes - 1)) >= shadow). */
1410 gimple_seq seq = NULL;
1411 gimple shadow_test = build_assign (NE_EXPR, shadow, 0);
1412 gimple_seq_add_stmt (&seq, shadow_test);
1413 gimple_seq_add_stmt (&seq, build_assign (BIT_AND_EXPR, base_addr, 7));
1414 gimple_seq_add_stmt (&seq, build_type_cast (shadow_type,
1415 gimple_seq_last (seq)));
1416 if (size_in_bytes > 1)
1417 gimple_seq_add_stmt (&seq,
1418 build_assign (PLUS_EXPR, gimple_seq_last (seq),
1419 size_in_bytes - 1));
1420 gimple_seq_add_stmt (&seq, build_assign (GE_EXPR, gimple_seq_last (seq),
1421 shadow));
1422 gimple_seq_add_stmt (&seq, build_assign (BIT_AND_EXPR, shadow_test,
1423 gimple_seq_last (seq)));
1424 t = gimple_assign_lhs (gimple_seq_last (seq));
1425 gimple_seq_set_location (seq, location);
1426 gsi_insert_seq_after (&gsi, seq, GSI_CONTINUE_LINKING);
1428 else
1429 t = shadow;
1431 g = gimple_build_cond (NE_EXPR, t, build_int_cst (TREE_TYPE (t), 0),
1432 NULL_TREE, NULL_TREE);
1433 gimple_set_location (g, location);
1434 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1436 /* Generate call to the run-time library (e.g. __asan_report_load8). */
1437 gsi = gsi_start_bb (then_bb);
1438 g = gimple_build_call (report_error_func (is_store, size_in_bytes),
1439 1, base_addr);
1440 gimple_set_location (g, location);
1441 gsi_insert_after (&gsi, g, GSI_NEW_STMT);
1443 *iter = gsi_start_bb (else_bb);
1446 /* If T represents a memory access, add instrumentation code before ITER.
1447 LOCATION is source code location.
1448 IS_STORE is either TRUE (for a store) or FALSE (for a load). */
1450 static void
1451 instrument_derefs (gimple_stmt_iterator *iter, tree t,
1452 location_t location, bool is_store)
1454 tree type, base;
1455 HOST_WIDE_INT size_in_bytes;
1457 type = TREE_TYPE (t);
1458 switch (TREE_CODE (t))
1460 case ARRAY_REF:
1461 case COMPONENT_REF:
1462 case INDIRECT_REF:
1463 case MEM_REF:
1464 break;
1465 default:
1466 return;
1469 size_in_bytes = int_size_in_bytes (type);
1470 if ((size_in_bytes & (size_in_bytes - 1)) != 0
1471 || (unsigned HOST_WIDE_INT) size_in_bytes - 1 >= 16)
1472 return;
1474 HOST_WIDE_INT bitsize, bitpos;
1475 tree offset;
1476 enum machine_mode mode;
1477 int volatilep = 0, unsignedp = 0;
1478 get_inner_reference (t, &bitsize, &bitpos, &offset,
1479 &mode, &unsignedp, &volatilep, false);
1480 if (bitpos % (size_in_bytes * BITS_PER_UNIT)
1481 || bitsize != size_in_bytes * BITS_PER_UNIT)
1483 if (TREE_CODE (t) == COMPONENT_REF
1484 && DECL_BIT_FIELD_REPRESENTATIVE (TREE_OPERAND (t, 1)) != NULL_TREE)
1486 tree repr = DECL_BIT_FIELD_REPRESENTATIVE (TREE_OPERAND (t, 1));
1487 instrument_derefs (iter, build3 (COMPONENT_REF, TREE_TYPE (repr),
1488 TREE_OPERAND (t, 0), repr,
1489 NULL_TREE), location, is_store);
1491 return;
1494 base = build_fold_addr_expr (t);
1495 if (!has_mem_ref_been_instrumented (base, size_in_bytes))
1497 build_check_stmt (location, base, iter, /*before_p=*/true,
1498 is_store, size_in_bytes);
1499 update_mem_ref_hash_table (base, size_in_bytes);
1500 update_mem_ref_hash_table (t, size_in_bytes);
1505 /* Instrument an access to a contiguous memory region that starts at
1506 the address pointed to by BASE, over a length of LEN (expressed in
1507 the sizeof (*BASE) bytes). ITER points to the instruction before
1508 which the instrumentation instructions must be inserted. LOCATION
1509 is the source location that the instrumentation instructions must
1510 have. If IS_STORE is true, then the memory access is a store;
1511 otherwise, it's a load. */
1513 static void
1514 instrument_mem_region_access (tree base, tree len,
1515 gimple_stmt_iterator *iter,
1516 location_t location, bool is_store)
1518 if (!POINTER_TYPE_P (TREE_TYPE (base))
1519 || !INTEGRAL_TYPE_P (TREE_TYPE (len))
1520 || integer_zerop (len))
1521 return;
1523 gimple_stmt_iterator gsi = *iter;
1525 basic_block fallthrough_bb = NULL, then_bb = NULL;
1527 /* If the beginning of the memory region has already been
1528 instrumented, do not instrument it. */
1529 bool start_instrumented = has_mem_ref_been_instrumented (base, 1);
1531 /* If the end of the memory region has already been instrumented, do
1532 not instrument it. */
1533 tree end = asan_mem_ref_get_end (base, len);
1534 bool end_instrumented = has_mem_ref_been_instrumented (end, 1);
1536 if (start_instrumented && end_instrumented)
1537 return;
1539 if (!is_gimple_constant (len))
1541 /* So, the length of the memory area to asan-protect is
1542 non-constant. Let's guard the generated instrumentation code
1543 like:
1545 if (len != 0)
1547 //asan instrumentation code goes here.
1549 // falltrough instructions, starting with *ITER. */
1551 gimple g = gimple_build_cond (NE_EXPR,
1552 len,
1553 build_int_cst (TREE_TYPE (len), 0),
1554 NULL_TREE, NULL_TREE);
1555 gimple_set_location (g, location);
1556 insert_if_then_before_iter (g, iter, /*then_more_likely_p=*/true,
1557 &then_bb, &fallthrough_bb);
1558 /* Note that fallthrough_bb starts with the statement that was
1559 pointed to by ITER. */
1561 /* The 'then block' of the 'if (len != 0) condition is where
1562 we'll generate the asan instrumentation code now. */
1563 gsi = gsi_last_bb (then_bb);
1566 if (!start_instrumented)
1568 /* Instrument the beginning of the memory region to be accessed,
1569 and arrange for the rest of the intrumentation code to be
1570 inserted in the then block *after* the current gsi. */
1571 build_check_stmt (location, base, &gsi, /*before_p=*/true, is_store, 1);
1573 if (then_bb)
1574 /* We are in the case where the length of the region is not
1575 constant; so instrumentation code is being generated in the
1576 'then block' of the 'if (len != 0) condition. Let's arrange
1577 for the subsequent instrumentation statements to go in the
1578 'then block'. */
1579 gsi = gsi_last_bb (then_bb);
1580 else
1582 *iter = gsi;
1583 /* Don't remember this access as instrumented, if length
1584 is unknown. It might be zero and not being actually
1585 instrumented, so we can't rely on it being instrumented. */
1586 update_mem_ref_hash_table (base, 1);
1590 if (end_instrumented)
1591 return;
1593 /* We want to instrument the access at the end of the memory region,
1594 which is at (base + len - 1). */
1596 /* offset = len - 1; */
1597 len = unshare_expr (len);
1598 tree offset;
1599 gimple_seq seq = NULL;
1600 if (TREE_CODE (len) == INTEGER_CST)
1601 offset = fold_build2 (MINUS_EXPR, size_type_node,
1602 fold_convert (size_type_node, len),
1603 build_int_cst (size_type_node, 1));
1604 else
1606 gimple g;
1607 tree t;
1609 if (TREE_CODE (len) != SSA_NAME)
1611 t = make_ssa_name (TREE_TYPE (len), NULL);
1612 g = gimple_build_assign_with_ops (TREE_CODE (len), t, len, NULL);
1613 gimple_set_location (g, location);
1614 gimple_seq_add_stmt_without_update (&seq, g);
1615 len = t;
1617 if (!useless_type_conversion_p (size_type_node, TREE_TYPE (len)))
1619 t = make_ssa_name (size_type_node, NULL);
1620 g = gimple_build_assign_with_ops (NOP_EXPR, t, len, NULL);
1621 gimple_set_location (g, location);
1622 gimple_seq_add_stmt_without_update (&seq, g);
1623 len = t;
1626 t = make_ssa_name (size_type_node, NULL);
1627 g = gimple_build_assign_with_ops (MINUS_EXPR, t, len,
1628 build_int_cst (size_type_node, 1));
1629 gimple_set_location (g, location);
1630 gimple_seq_add_stmt_without_update (&seq, g);
1631 offset = gimple_assign_lhs (g);
1634 /* _1 = base; */
1635 base = unshare_expr (base);
1636 gimple region_end =
1637 gimple_build_assign_with_ops (TREE_CODE (base),
1638 make_ssa_name (TREE_TYPE (base), NULL),
1639 base, NULL);
1640 gimple_set_location (region_end, location);
1641 gimple_seq_add_stmt_without_update (&seq, region_end);
1643 /* _2 = _1 + offset; */
1644 region_end =
1645 gimple_build_assign_with_ops (POINTER_PLUS_EXPR,
1646 make_ssa_name (TREE_TYPE (base), NULL),
1647 gimple_assign_lhs (region_end),
1648 offset);
1649 gimple_set_location (region_end, location);
1650 gimple_seq_add_stmt_without_update (&seq, region_end);
1651 gsi_insert_seq_before (&gsi, seq, GSI_SAME_STMT);
1653 /* instrument access at _2; */
1654 gsi = gsi_for_stmt (region_end);
1655 build_check_stmt (location, gimple_assign_lhs (region_end),
1656 &gsi, /*before_p=*/false, is_store, 1);
1658 if (then_bb == NULL)
1659 update_mem_ref_hash_table (end, 1);
1661 *iter = gsi_for_stmt (gsi_stmt (*iter));
1664 /* Instrument the call (to the builtin strlen function) pointed to by
1665 ITER.
1667 This function instruments the access to the first byte of the
1668 argument, right before the call. After the call it instruments the
1669 access to the last byte of the argument; it uses the result of the
1670 call to deduce the offset of that last byte.
1672 Upon completion, iff the call has actually been instrumented, this
1673 function returns TRUE and *ITER points to the statement logically
1674 following the built-in strlen function call *ITER was initially
1675 pointing to. Otherwise, the function returns FALSE and *ITER
1676 remains unchanged. */
1678 static bool
1679 instrument_strlen_call (gimple_stmt_iterator *iter)
1681 gimple call = gsi_stmt (*iter);
1682 gcc_assert (is_gimple_call (call));
1684 tree callee = gimple_call_fndecl (call);
1685 gcc_assert (is_builtin_fn (callee)
1686 && DECL_BUILT_IN_CLASS (callee) == BUILT_IN_NORMAL
1687 && DECL_FUNCTION_CODE (callee) == BUILT_IN_STRLEN);
1689 tree len = gimple_call_lhs (call);
1690 if (len == NULL)
1691 /* Some passes might clear the return value of the strlen call;
1692 bail out in that case. Return FALSE as we are not advancing
1693 *ITER. */
1694 return false;
1695 gcc_assert (INTEGRAL_TYPE_P (TREE_TYPE (len)));
1697 location_t loc = gimple_location (call);
1698 tree str_arg = gimple_call_arg (call, 0);
1700 /* Instrument the access to the first byte of str_arg. i.e:
1702 _1 = str_arg; instrument (_1); */
1703 tree cptr_type = build_pointer_type (char_type_node);
1704 gimple str_arg_ssa =
1705 gimple_build_assign_with_ops (NOP_EXPR,
1706 make_ssa_name (cptr_type, NULL),
1707 str_arg, NULL);
1708 gimple_set_location (str_arg_ssa, loc);
1709 gimple_stmt_iterator gsi = *iter;
1710 gsi_insert_before (&gsi, str_arg_ssa, GSI_NEW_STMT);
1711 build_check_stmt (loc, gimple_assign_lhs (str_arg_ssa), &gsi,
1712 /*before_p=*/false, /*is_store=*/false, 1);
1714 /* If we initially had an instruction like:
1716 int n = strlen (str)
1718 we now want to instrument the access to str[n], after the
1719 instruction above.*/
1721 /* So let's build the access to str[n] that is, access through the
1722 pointer_plus expr: (_1 + len). */
1723 gimple stmt =
1724 gimple_build_assign_with_ops (POINTER_PLUS_EXPR,
1725 make_ssa_name (cptr_type, NULL),
1726 gimple_assign_lhs (str_arg_ssa),
1727 len);
1728 gimple_set_location (stmt, loc);
1729 gsi_insert_after (&gsi, stmt, GSI_NEW_STMT);
1731 build_check_stmt (loc, gimple_assign_lhs (stmt), &gsi,
1732 /*before_p=*/false, /*is_store=*/false, 1);
1734 /* Ensure that iter points to the statement logically following the
1735 one it was initially pointing to. */
1736 *iter = gsi;
1737 /* As *ITER has been advanced to point to the next statement, let's
1738 return true to inform transform_statements that it shouldn't
1739 advance *ITER anymore; otherwises it will skip that next
1740 statement, which wouldn't be instrumented. */
1741 return true;
1744 /* Instrument the call to a built-in memory access function that is
1745 pointed to by the iterator ITER.
1747 Upon completion, return TRUE iff *ITER has been advanced to the
1748 statement following the one it was originally pointing to. */
1750 static bool
1751 instrument_builtin_call (gimple_stmt_iterator *iter)
1753 bool iter_advanced_p = false;
1754 gimple call = gsi_stmt (*iter);
1756 gcc_checking_assert (gimple_call_builtin_p (call, BUILT_IN_NORMAL));
1758 tree callee = gimple_call_fndecl (call);
1759 location_t loc = gimple_location (call);
1761 if (DECL_FUNCTION_CODE (callee) == BUILT_IN_STRLEN)
1762 iter_advanced_p = instrument_strlen_call (iter);
1763 else
1765 asan_mem_ref src0, src1, dest;
1766 asan_mem_ref_init (&src0, NULL, 1);
1767 asan_mem_ref_init (&src1, NULL, 1);
1768 asan_mem_ref_init (&dest, NULL, 1);
1770 tree src0_len = NULL_TREE, src1_len = NULL_TREE, dest_len = NULL_TREE;
1771 bool src0_is_store = false, src1_is_store = false,
1772 dest_is_store = false, dest_is_deref = false;
1774 if (get_mem_refs_of_builtin_call (call,
1775 &src0, &src0_len, &src0_is_store,
1776 &src1, &src1_len, &src1_is_store,
1777 &dest, &dest_len, &dest_is_store,
1778 &dest_is_deref))
1780 if (dest_is_deref)
1782 instrument_derefs (iter, dest.start, loc, dest_is_store);
1783 gsi_next (iter);
1784 iter_advanced_p = true;
1786 else if (src0_len || src1_len || dest_len)
1788 if (src0.start != NULL_TREE)
1789 instrument_mem_region_access (src0.start, src0_len,
1790 iter, loc, /*is_store=*/false);
1791 if (src1.start != NULL_TREE)
1792 instrument_mem_region_access (src1.start, src1_len,
1793 iter, loc, /*is_store=*/false);
1794 if (dest.start != NULL_TREE)
1795 instrument_mem_region_access (dest.start, dest_len,
1796 iter, loc, /*is_store=*/true);
1797 *iter = gsi_for_stmt (call);
1798 gsi_next (iter);
1799 iter_advanced_p = true;
1803 return iter_advanced_p;
1806 /* Instrument the assignment statement ITER if it is subject to
1807 instrumentation. Return TRUE iff instrumentation actually
1808 happened. In that case, the iterator ITER is advanced to the next
1809 logical expression following the one initially pointed to by ITER,
1810 and the relevant memory reference that which access has been
1811 instrumented is added to the memory references hash table. */
1813 static bool
1814 maybe_instrument_assignment (gimple_stmt_iterator *iter)
1816 gimple s = gsi_stmt (*iter);
1818 gcc_assert (gimple_assign_single_p (s));
1820 tree ref_expr = NULL_TREE;
1821 bool is_store, is_instrumented = false;
1823 if (gimple_store_p (s))
1825 ref_expr = gimple_assign_lhs (s);
1826 is_store = true;
1827 instrument_derefs (iter, ref_expr,
1828 gimple_location (s),
1829 is_store);
1830 is_instrumented = true;
1833 if (gimple_assign_load_p (s))
1835 ref_expr = gimple_assign_rhs1 (s);
1836 is_store = false;
1837 instrument_derefs (iter, ref_expr,
1838 gimple_location (s),
1839 is_store);
1840 is_instrumented = true;
1843 if (is_instrumented)
1844 gsi_next (iter);
1846 return is_instrumented;
1849 /* Instrument the function call pointed to by the iterator ITER, if it
1850 is subject to instrumentation. At the moment, the only function
1851 calls that are instrumented are some built-in functions that access
1852 memory. Look at instrument_builtin_call to learn more.
1854 Upon completion return TRUE iff *ITER was advanced to the statement
1855 following the one it was originally pointing to. */
1857 static bool
1858 maybe_instrument_call (gimple_stmt_iterator *iter)
1860 gimple stmt = gsi_stmt (*iter);
1861 bool is_builtin = gimple_call_builtin_p (stmt, BUILT_IN_NORMAL);
1863 if (is_builtin && instrument_builtin_call (iter))
1864 return true;
1866 if (gimple_call_noreturn_p (stmt))
1868 if (is_builtin)
1870 tree callee = gimple_call_fndecl (stmt);
1871 switch (DECL_FUNCTION_CODE (callee))
1873 case BUILT_IN_UNREACHABLE:
1874 case BUILT_IN_TRAP:
1875 /* Don't instrument these. */
1876 return false;
1879 tree decl = builtin_decl_implicit (BUILT_IN_ASAN_HANDLE_NO_RETURN);
1880 gimple g = gimple_build_call (decl, 0);
1881 gimple_set_location (g, gimple_location (stmt));
1882 gsi_insert_before (iter, g, GSI_SAME_STMT);
1884 return false;
1887 /* Walk each instruction of all basic block and instrument those that
1888 represent memory references: loads, stores, or function calls.
1889 In a given basic block, this function avoids instrumenting memory
1890 references that have already been instrumented. */
1892 static void
1893 transform_statements (void)
1895 basic_block bb, last_bb = NULL;
1896 gimple_stmt_iterator i;
1897 int saved_last_basic_block = last_basic_block;
1899 FOR_EACH_BB (bb)
1901 basic_block prev_bb = bb;
1903 if (bb->index >= saved_last_basic_block) continue;
1905 /* Flush the mem ref hash table, if current bb doesn't have
1906 exactly one predecessor, or if that predecessor (skipping
1907 over asan created basic blocks) isn't the last processed
1908 basic block. Thus we effectively flush on extended basic
1909 block boundaries. */
1910 while (single_pred_p (prev_bb))
1912 prev_bb = single_pred (prev_bb);
1913 if (prev_bb->index < saved_last_basic_block)
1914 break;
1916 if (prev_bb != last_bb)
1917 empty_mem_ref_hash_table ();
1918 last_bb = bb;
1920 for (i = gsi_start_bb (bb); !gsi_end_p (i);)
1922 gimple s = gsi_stmt (i);
1924 if (has_stmt_been_instrumented_p (s))
1925 gsi_next (&i);
1926 else if (gimple_assign_single_p (s)
1927 && maybe_instrument_assignment (&i))
1928 /* Nothing to do as maybe_instrument_assignment advanced
1929 the iterator I. */;
1930 else if (is_gimple_call (s) && maybe_instrument_call (&i))
1931 /* Nothing to do as maybe_instrument_call
1932 advanced the iterator I. */;
1933 else
1935 /* No instrumentation happened.
1937 If the current instruction is a function call that
1938 might free something, let's forget about the memory
1939 references that got instrumented. Otherwise we might
1940 miss some instrumentation opportunities. */
1941 if (is_gimple_call (s) && !nonfreeing_call_p (s))
1942 empty_mem_ref_hash_table ();
1944 gsi_next (&i);
1948 free_mem_ref_resources ();
1951 /* Build
1952 struct __asan_global
1954 const void *__beg;
1955 uptr __size;
1956 uptr __size_with_redzone;
1957 const void *__name;
1958 const void *__module_name;
1959 uptr __has_dynamic_init;
1960 } type. */
1962 static tree
1963 asan_global_struct (void)
1965 static const char *field_names[6]
1966 = { "__beg", "__size", "__size_with_redzone",
1967 "__name", "__module_name", "__has_dynamic_init" };
1968 tree fields[6], ret;
1969 int i;
1971 ret = make_node (RECORD_TYPE);
1972 for (i = 0; i < 6; i++)
1974 fields[i]
1975 = build_decl (UNKNOWN_LOCATION, FIELD_DECL,
1976 get_identifier (field_names[i]),
1977 (i == 0 || i == 3) ? const_ptr_type_node
1978 : pointer_sized_int_node);
1979 DECL_CONTEXT (fields[i]) = ret;
1980 if (i)
1981 DECL_CHAIN (fields[i - 1]) = fields[i];
1983 TYPE_FIELDS (ret) = fields[0];
1984 TYPE_NAME (ret) = get_identifier ("__asan_global");
1985 layout_type (ret);
1986 return ret;
1989 /* Append description of a single global DECL into vector V.
1990 TYPE is __asan_global struct type as returned by asan_global_struct. */
1992 static void
1993 asan_add_global (tree decl, tree type, vec<constructor_elt, va_gc> *v)
1995 tree init, uptr = TREE_TYPE (DECL_CHAIN (TYPE_FIELDS (type)));
1996 unsigned HOST_WIDE_INT size;
1997 tree str_cst, module_name_cst, refdecl = decl;
1998 vec<constructor_elt, va_gc> *vinner = NULL;
2000 pretty_printer asan_pp, module_name_pp;
2002 if (DECL_NAME (decl))
2003 pp_tree_identifier (&asan_pp, DECL_NAME (decl));
2004 else
2005 pp_string (&asan_pp, "<unknown>");
2006 str_cst = asan_pp_string (&asan_pp);
2008 pp_string (&module_name_pp, main_input_filename);
2009 module_name_cst = asan_pp_string (&module_name_pp);
2011 if (asan_needs_local_alias (decl))
2013 char buf[20];
2014 ASM_GENERATE_INTERNAL_LABEL (buf, "LASAN", vec_safe_length (v) + 1);
2015 refdecl = build_decl (DECL_SOURCE_LOCATION (decl),
2016 VAR_DECL, get_identifier (buf), TREE_TYPE (decl));
2017 TREE_ADDRESSABLE (refdecl) = TREE_ADDRESSABLE (decl);
2018 TREE_READONLY (refdecl) = TREE_READONLY (decl);
2019 TREE_THIS_VOLATILE (refdecl) = TREE_THIS_VOLATILE (decl);
2020 DECL_GIMPLE_REG_P (refdecl) = DECL_GIMPLE_REG_P (decl);
2021 DECL_ARTIFICIAL (refdecl) = DECL_ARTIFICIAL (decl);
2022 DECL_IGNORED_P (refdecl) = DECL_IGNORED_P (decl);
2023 TREE_STATIC (refdecl) = 1;
2024 TREE_PUBLIC (refdecl) = 0;
2025 TREE_USED (refdecl) = 1;
2026 assemble_alias (refdecl, DECL_ASSEMBLER_NAME (decl));
2029 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
2030 fold_convert (const_ptr_type_node,
2031 build_fold_addr_expr (refdecl)));
2032 size = tree_to_uhwi (DECL_SIZE_UNIT (decl));
2033 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, size));
2034 size += asan_red_zone_size (size);
2035 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, size));
2036 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
2037 fold_convert (const_ptr_type_node, str_cst));
2038 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE,
2039 fold_convert (const_ptr_type_node, module_name_cst));
2040 CONSTRUCTOR_APPEND_ELT (vinner, NULL_TREE, build_int_cst (uptr, 0));
2041 init = build_constructor (type, vinner);
2042 CONSTRUCTOR_APPEND_ELT (v, NULL_TREE, init);
2045 /* Initialize sanitizer.def builtins if the FE hasn't initialized them. */
2046 void
2047 initialize_sanitizer_builtins (void)
2049 tree decl;
2051 if (builtin_decl_implicit_p (BUILT_IN_ASAN_INIT))
2052 return;
2054 tree BT_FN_VOID = build_function_type_list (void_type_node, NULL_TREE);
2055 tree BT_FN_VOID_PTR
2056 = build_function_type_list (void_type_node, ptr_type_node, NULL_TREE);
2057 tree BT_FN_VOID_PTR_PTR
2058 = build_function_type_list (void_type_node, ptr_type_node,
2059 ptr_type_node, NULL_TREE);
2060 tree BT_FN_VOID_PTR_PTR_PTR
2061 = build_function_type_list (void_type_node, ptr_type_node,
2062 ptr_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 pointer_sized_int_node, NULL_TREE);
2066 tree BT_FN_VOID_INT
2067 = build_function_type_list (void_type_node, integer_type_node, NULL_TREE);
2068 tree BT_FN_BOOL_VPTR_PTR_IX_INT_INT[5];
2069 tree BT_FN_IX_CONST_VPTR_INT[5];
2070 tree BT_FN_IX_VPTR_IX_INT[5];
2071 tree BT_FN_VOID_VPTR_IX_INT[5];
2072 tree vptr
2073 = build_pointer_type (build_qualified_type (void_type_node,
2074 TYPE_QUAL_VOLATILE));
2075 tree cvptr
2076 = build_pointer_type (build_qualified_type (void_type_node,
2077 TYPE_QUAL_VOLATILE
2078 |TYPE_QUAL_CONST));
2079 tree boolt
2080 = lang_hooks.types.type_for_size (BOOL_TYPE_SIZE, 1);
2081 int i;
2082 for (i = 0; i < 5; i++)
2084 tree ix = build_nonstandard_integer_type (BITS_PER_UNIT * (1 << i), 1);
2085 BT_FN_BOOL_VPTR_PTR_IX_INT_INT[i]
2086 = build_function_type_list (boolt, vptr, ptr_type_node, ix,
2087 integer_type_node, integer_type_node,
2088 NULL_TREE);
2089 BT_FN_IX_CONST_VPTR_INT[i]
2090 = build_function_type_list (ix, cvptr, integer_type_node, NULL_TREE);
2091 BT_FN_IX_VPTR_IX_INT[i]
2092 = build_function_type_list (ix, vptr, ix, integer_type_node,
2093 NULL_TREE);
2094 BT_FN_VOID_VPTR_IX_INT[i]
2095 = build_function_type_list (void_type_node, vptr, ix,
2096 integer_type_node, NULL_TREE);
2098 #define BT_FN_BOOL_VPTR_PTR_I1_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[0]
2099 #define BT_FN_I1_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[0]
2100 #define BT_FN_I1_VPTR_I1_INT BT_FN_IX_VPTR_IX_INT[0]
2101 #define BT_FN_VOID_VPTR_I1_INT BT_FN_VOID_VPTR_IX_INT[0]
2102 #define BT_FN_BOOL_VPTR_PTR_I2_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[1]
2103 #define BT_FN_I2_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[1]
2104 #define BT_FN_I2_VPTR_I2_INT BT_FN_IX_VPTR_IX_INT[1]
2105 #define BT_FN_VOID_VPTR_I2_INT BT_FN_VOID_VPTR_IX_INT[1]
2106 #define BT_FN_BOOL_VPTR_PTR_I4_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[2]
2107 #define BT_FN_I4_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[2]
2108 #define BT_FN_I4_VPTR_I4_INT BT_FN_IX_VPTR_IX_INT[2]
2109 #define BT_FN_VOID_VPTR_I4_INT BT_FN_VOID_VPTR_IX_INT[2]
2110 #define BT_FN_BOOL_VPTR_PTR_I8_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[3]
2111 #define BT_FN_I8_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[3]
2112 #define BT_FN_I8_VPTR_I8_INT BT_FN_IX_VPTR_IX_INT[3]
2113 #define BT_FN_VOID_VPTR_I8_INT BT_FN_VOID_VPTR_IX_INT[3]
2114 #define BT_FN_BOOL_VPTR_PTR_I16_INT_INT BT_FN_BOOL_VPTR_PTR_IX_INT_INT[4]
2115 #define BT_FN_I16_CONST_VPTR_INT BT_FN_IX_CONST_VPTR_INT[4]
2116 #define BT_FN_I16_VPTR_I16_INT BT_FN_IX_VPTR_IX_INT[4]
2117 #define BT_FN_VOID_VPTR_I16_INT BT_FN_VOID_VPTR_IX_INT[4]
2118 #undef ATTR_NOTHROW_LEAF_LIST
2119 #define ATTR_NOTHROW_LEAF_LIST ECF_NOTHROW | ECF_LEAF
2120 #undef ATTR_TMPURE_NOTHROW_LEAF_LIST
2121 #define ATTR_TMPURE_NOTHROW_LEAF_LIST ECF_TM_PURE | ATTR_NOTHROW_LEAF_LIST
2122 #undef ATTR_NORETURN_NOTHROW_LEAF_LIST
2123 #define ATTR_NORETURN_NOTHROW_LEAF_LIST ECF_NORETURN | ATTR_NOTHROW_LEAF_LIST
2124 #undef ATTR_TMPURE_NORETURN_NOTHROW_LEAF_LIST
2125 #define ATTR_TMPURE_NORETURN_NOTHROW_LEAF_LIST \
2126 ECF_TM_PURE | ATTR_NORETURN_NOTHROW_LEAF_LIST
2127 #undef ATTR_COLD_NOTHROW_LEAF_LIST
2128 #define ATTR_COLD_NOTHROW_LEAF_LIST \
2129 /* ECF_COLD missing */ ATTR_NOTHROW_LEAF_LIST
2130 #undef ATTR_COLD_NORETURN_NOTHROW_LEAF_LIST
2131 #define ATTR_COLD_NORETURN_NOTHROW_LEAF_LIST \
2132 /* ECF_COLD missing */ ATTR_NORETURN_NOTHROW_LEAF_LIST
2133 #undef DEF_SANITIZER_BUILTIN
2134 #define DEF_SANITIZER_BUILTIN(ENUM, NAME, TYPE, ATTRS) \
2135 decl = add_builtin_function ("__builtin_" NAME, TYPE, ENUM, \
2136 BUILT_IN_NORMAL, NAME, NULL_TREE); \
2137 set_call_expr_flags (decl, ATTRS); \
2138 set_builtin_decl (ENUM, decl, true);
2140 #include "sanitizer.def"
2142 #undef DEF_SANITIZER_BUILTIN
2145 /* Called via htab_traverse. Count number of emitted
2146 STRING_CSTs in the constant hash table. */
2148 static int
2149 count_string_csts (void **slot, void *data)
2151 struct constant_descriptor_tree *desc
2152 = (struct constant_descriptor_tree *) *slot;
2153 if (TREE_CODE (desc->value) == STRING_CST
2154 && TREE_ASM_WRITTEN (desc->value)
2155 && asan_protect_global (desc->value))
2156 ++*((unsigned HOST_WIDE_INT *) data);
2157 return 1;
2160 /* Helper structure to pass two parameters to
2161 add_string_csts. */
2163 struct asan_add_string_csts_data
2165 tree type;
2166 vec<constructor_elt, va_gc> *v;
2169 /* Called via htab_traverse. Call asan_add_global
2170 on emitted STRING_CSTs from the constant hash table. */
2172 static int
2173 add_string_csts (void **slot, void *data)
2175 struct constant_descriptor_tree *desc
2176 = (struct constant_descriptor_tree *) *slot;
2177 if (TREE_CODE (desc->value) == STRING_CST
2178 && TREE_ASM_WRITTEN (desc->value)
2179 && asan_protect_global (desc->value))
2181 struct asan_add_string_csts_data *aascd
2182 = (struct asan_add_string_csts_data *) data;
2183 asan_add_global (SYMBOL_REF_DECL (XEXP (desc->rtl, 0)),
2184 aascd->type, aascd->v);
2186 return 1;
2189 /* Needs to be GTY(()), because cgraph_build_static_cdtor may
2190 invoke ggc_collect. */
2191 static GTY(()) tree asan_ctor_statements;
2193 /* Module-level instrumentation.
2194 - Insert __asan_init_vN() into the list of CTORs.
2195 - TODO: insert redzones around globals.
2198 void
2199 asan_finish_file (void)
2201 struct varpool_node *vnode;
2202 unsigned HOST_WIDE_INT gcount = 0;
2204 if (shadow_ptr_types[0] == NULL_TREE)
2205 asan_init_shadow_ptr_types ();
2206 /* Avoid instrumenting code in the asan ctors/dtors.
2207 We don't need to insert padding after the description strings,
2208 nor after .LASAN* array. */
2209 flag_sanitize &= ~SANITIZE_ADDRESS;
2211 tree fn = builtin_decl_implicit (BUILT_IN_ASAN_INIT);
2212 append_to_statement_list (build_call_expr (fn, 0), &asan_ctor_statements);
2213 FOR_EACH_DEFINED_VARIABLE (vnode)
2214 if (TREE_ASM_WRITTEN (vnode->decl)
2215 && asan_protect_global (vnode->decl))
2216 ++gcount;
2217 htab_t const_desc_htab = constant_pool_htab ();
2218 htab_traverse (const_desc_htab, count_string_csts, &gcount);
2219 if (gcount)
2221 tree type = asan_global_struct (), var, ctor;
2222 tree dtor_statements = NULL_TREE;
2223 vec<constructor_elt, va_gc> *v;
2224 char buf[20];
2226 type = build_array_type_nelts (type, gcount);
2227 ASM_GENERATE_INTERNAL_LABEL (buf, "LASAN", 0);
2228 var = build_decl (UNKNOWN_LOCATION, VAR_DECL, get_identifier (buf),
2229 type);
2230 TREE_STATIC (var) = 1;
2231 TREE_PUBLIC (var) = 0;
2232 DECL_ARTIFICIAL (var) = 1;
2233 DECL_IGNORED_P (var) = 1;
2234 vec_alloc (v, gcount);
2235 FOR_EACH_DEFINED_VARIABLE (vnode)
2236 if (TREE_ASM_WRITTEN (vnode->decl)
2237 && asan_protect_global (vnode->decl))
2238 asan_add_global (vnode->decl, TREE_TYPE (type), v);
2239 struct asan_add_string_csts_data aascd;
2240 aascd.type = TREE_TYPE (type);
2241 aascd.v = v;
2242 htab_traverse (const_desc_htab, add_string_csts, &aascd);
2243 ctor = build_constructor (type, v);
2244 TREE_CONSTANT (ctor) = 1;
2245 TREE_STATIC (ctor) = 1;
2246 DECL_INITIAL (var) = ctor;
2247 varpool_assemble_decl (varpool_node_for_decl (var));
2249 fn = builtin_decl_implicit (BUILT_IN_ASAN_REGISTER_GLOBALS);
2250 tree gcount_tree = build_int_cst (pointer_sized_int_node, gcount);
2251 append_to_statement_list (build_call_expr (fn, 2,
2252 build_fold_addr_expr (var),
2253 gcount_tree),
2254 &asan_ctor_statements);
2256 fn = builtin_decl_implicit (BUILT_IN_ASAN_UNREGISTER_GLOBALS);
2257 append_to_statement_list (build_call_expr (fn, 2,
2258 build_fold_addr_expr (var),
2259 gcount_tree),
2260 &dtor_statements);
2261 cgraph_build_static_cdtor ('D', dtor_statements,
2262 MAX_RESERVED_INIT_PRIORITY - 1);
2264 cgraph_build_static_cdtor ('I', asan_ctor_statements,
2265 MAX_RESERVED_INIT_PRIORITY - 1);
2266 flag_sanitize |= SANITIZE_ADDRESS;
2269 /* Instrument the current function. */
2271 static unsigned int
2272 asan_instrument (void)
2274 if (shadow_ptr_types[0] == NULL_TREE)
2275 asan_init_shadow_ptr_types ();
2276 transform_statements ();
2277 return 0;
2280 static bool
2281 gate_asan (void)
2283 return (flag_sanitize & SANITIZE_ADDRESS) != 0
2284 && !lookup_attribute ("no_sanitize_address",
2285 DECL_ATTRIBUTES (current_function_decl));
2288 namespace {
2290 const pass_data pass_data_asan =
2292 GIMPLE_PASS, /* type */
2293 "asan", /* name */
2294 OPTGROUP_NONE, /* optinfo_flags */
2295 true, /* has_gate */
2296 true, /* has_execute */
2297 TV_NONE, /* tv_id */
2298 ( PROP_ssa | PROP_cfg | PROP_gimple_leh ), /* properties_required */
2299 0, /* properties_provided */
2300 0, /* properties_destroyed */
2301 0, /* todo_flags_start */
2302 ( TODO_verify_flow | TODO_verify_stmts
2303 | TODO_update_ssa ), /* todo_flags_finish */
2306 class pass_asan : public gimple_opt_pass
2308 public:
2309 pass_asan (gcc::context *ctxt)
2310 : gimple_opt_pass (pass_data_asan, ctxt)
2313 /* opt_pass methods: */
2314 opt_pass * clone () { return new pass_asan (m_ctxt); }
2315 bool gate () { return gate_asan (); }
2316 unsigned int execute () { return asan_instrument (); }
2318 }; // class pass_asan
2320 } // anon namespace
2322 gimple_opt_pass *
2323 make_pass_asan (gcc::context *ctxt)
2325 return new pass_asan (ctxt);
2328 static bool
2329 gate_asan_O0 (void)
2331 return !optimize && gate_asan ();
2334 namespace {
2336 const pass_data pass_data_asan_O0 =
2338 GIMPLE_PASS, /* type */
2339 "asan0", /* name */
2340 OPTGROUP_NONE, /* optinfo_flags */
2341 true, /* has_gate */
2342 true, /* has_execute */
2343 TV_NONE, /* tv_id */
2344 ( PROP_ssa | PROP_cfg | PROP_gimple_leh ), /* properties_required */
2345 0, /* properties_provided */
2346 0, /* properties_destroyed */
2347 0, /* todo_flags_start */
2348 ( TODO_verify_flow | TODO_verify_stmts
2349 | TODO_update_ssa ), /* todo_flags_finish */
2352 class pass_asan_O0 : public gimple_opt_pass
2354 public:
2355 pass_asan_O0 (gcc::context *ctxt)
2356 : gimple_opt_pass (pass_data_asan_O0, ctxt)
2359 /* opt_pass methods: */
2360 bool gate () { return gate_asan_O0 (); }
2361 unsigned int execute () { return asan_instrument (); }
2363 }; // class pass_asan_O0
2365 } // anon namespace
2367 gimple_opt_pass *
2368 make_pass_asan_O0 (gcc::context *ctxt)
2370 return new pass_asan_O0 (ctxt);
2373 #include "gt-asan.h"