1 /*
2 * Copyright (C) 2008 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 /*
18 * Dalvik bytecode verifier.
19 */
20 #ifndef _DALVIK_CODEVERIFY
21 #define _DALVIK_CODEVERIFY
22
23 #include "analysis/VerifySubs.h"
24
25
26 /*
27 * Enumeration for register type values. The "hi" piece of a 64-bit value
28 * MUST immediately follow the "lo" piece in the enumeration, so we can check
29 * that hi==lo+1.
30 *
31 * Assignment of constants:
32 * [-MAXINT,-32768) : integer
33 * [-32768,-128) : short
34 * [-128,0) : byte
35 * 0 : zero
36 * 1 : one
37 * [2,128) : posbyte
38 * [128,32768) : posshort
39 * [32768,65536) : char
40 * [65536,MAXINT] : integer
41 *
42 * Allowed "implicit" widening conversions:
43 * zero -> boolean, posbyte, byte, posshort, short, char, integer, ref (null)
44 * one -> boolean, posbyte, byte, posshort, short, char, integer
45 * boolean -> posbyte, byte, posshort, short, char, integer
46 * posbyte -> posshort, short, integer, char
47 * byte -> short, integer
48 * posshort -> integer, char
49 * short -> integer
50 * char -> integer
51 *
52 * In addition, all of the above can convert to "float".
53 *
54 * We're more careful with integer values than the spec requires. The
55 * motivation is to restrict byte/char/short to the correct range of values.
56 * For example, if a method takes a byte argument, we don't want to allow
57 * the code to load the constant "1024" and pass it in.
58 */
59 enum {
60 kRegTypeUnknown = 0, /* initial state; use value=0 so calloc works */
61 kRegTypeUninit = 1, /* MUST be odd to distinguish from pointer */
62 kRegTypeConflict, /* merge clash makes this reg's type unknowable */
63
64 /*
65 * Category-1nr types. The order of these is chiseled into a couple
66 * of tables, so don't add, remove, or reorder if you can avoid it.
67 */
68 #define kRegType1nrSTART kRegTypeFloat
69 kRegTypeFloat,
70 kRegTypeZero, /* 32-bit 0, could be Boolean, Int, Float, or Ref */
71 kRegTypeOne, /* 32-bit 1, could be Boolean, Int, Float */
72 kRegTypeBoolean, /* must be 0 or 1 */
73 kRegTypePosByte, /* byte, known positive (can become char) */
74 kRegTypeByte,
75 kRegTypePosShort, /* short, known positive (can become char) */
76 kRegTypeShort,
77 kRegTypeChar,
78 kRegTypeInteger,
79 #define kRegType1nrEND kRegTypeInteger
80
81 kRegTypeLongLo, /* lower-numbered register; endian-independent */
82 kRegTypeLongHi,
83 kRegTypeDoubleLo,
84 kRegTypeDoubleHi,
85
86 /*
87 * Enumeration max; this is used with "full" (32-bit) RegType values.
88 *
89 * Anything larger than this is a ClassObject or uninit ref. Mask off
90 * all but the low 8 bits; if you're left with kRegTypeUninit, pull
91 * the uninit index out of the high 24. Because kRegTypeUninit has an
92 * odd value, there is no risk of a particular ClassObject pointer bit
93 * pattern being confused for it (assuming our class object allocator
94 * uses word alignment).
95 */
96 kRegTypeMAX
97 };
98 #define kRegTypeUninitMask 0xff
99 #define kRegTypeUninitShift 8
100
101 /*
102 * RegType holds information about the type of data held in a register.
103 * For most types it's a simple enum. For reference types it holds a
104 * pointer to the ClassObject, and for uninitialized references it holds
105 * an index into the UninitInstanceMap.
106 */
107 typedef u4 RegType;
108
109 /* table with merge logic for primitive types */
110 extern const char gDvmMergeTab[kRegTypeMAX][kRegTypeMAX];
111
112
113 /*
114 * Returns "true" if the flags indicate that this address holds the start
115 * of an instruction.
116 */
dvmInsnIsOpcode(const InsnFlags * insnFlags,int addr)117 INLINE bool dvmInsnIsOpcode(const InsnFlags* insnFlags, int addr) {
118 return (insnFlags[addr] & kInsnFlagWidthMask) != 0;
119 }
120
121 /*
122 * Extract the unsigned 16-bit instruction width from "flags".
123 */
dvmInsnGetWidth(const InsnFlags * insnFlags,int addr)124 INLINE int dvmInsnGetWidth(const InsnFlags* insnFlags, int addr) {
125 return insnFlags[addr] & kInsnFlagWidthMask;
126 }
127
128 /*
129 * Changed?
130 */
dvmInsnIsChanged(const InsnFlags * insnFlags,int addr)131 INLINE bool dvmInsnIsChanged(const InsnFlags* insnFlags, int addr) {
132 return (insnFlags[addr] & kInsnFlagChanged) != 0;
133 }
dvmInsnSetChanged(InsnFlags * insnFlags,int addr,bool changed)134 INLINE void dvmInsnSetChanged(InsnFlags* insnFlags, int addr, bool changed)
135 {
136 if (changed)
137 insnFlags[addr] |= kInsnFlagChanged;
138 else
139 insnFlags[addr] &= ~kInsnFlagChanged;
140 }
141
142 /*
143 * Visited?
144 */
dvmInsnIsVisited(const InsnFlags * insnFlags,int addr)145 INLINE bool dvmInsnIsVisited(const InsnFlags* insnFlags, int addr) {
146 return (insnFlags[addr] & kInsnFlagVisited) != 0;
147 }
dvmInsnSetVisited(InsnFlags * insnFlags,int addr,bool changed)148 INLINE void dvmInsnSetVisited(InsnFlags* insnFlags, int addr, bool changed)
149 {
150 if (changed)
151 insnFlags[addr] |= kInsnFlagVisited;
152 else
153 insnFlags[addr] &= ~kInsnFlagVisited;
154 }
155
156 /*
157 * Visited or changed?
158 */
dvmInsnIsVisitedOrChanged(const InsnFlags * insnFlags,int addr)159 INLINE bool dvmInsnIsVisitedOrChanged(const InsnFlags* insnFlags, int addr) {
160 return (insnFlags[addr] & (kInsnFlagVisited|kInsnFlagChanged)) != 0;
161 }
162
163 /*
164 * In a "try" block?
165 */
dvmInsnIsInTry(const InsnFlags * insnFlags,int addr)166 INLINE bool dvmInsnIsInTry(const InsnFlags* insnFlags, int addr) {
167 return (insnFlags[addr] & kInsnFlagInTry) != 0;
168 }
dvmInsnSetInTry(InsnFlags * insnFlags,int addr,bool inTry)169 INLINE void dvmInsnSetInTry(InsnFlags* insnFlags, int addr, bool inTry)
170 {
171 assert(inTry);
172 //if (inTry)
173 insnFlags[addr] |= kInsnFlagInTry;
174 //else
175 // insnFlags[addr] &= ~kInsnFlagInTry;
176 }
177
178 /*
179 * Instruction is a branch target or exception handler?
180 */
dvmInsnIsBranchTarget(const InsnFlags * insnFlags,int addr)181 INLINE bool dvmInsnIsBranchTarget(const InsnFlags* insnFlags, int addr) {
182 return (insnFlags[addr] & kInsnFlagBranchTarget) != 0;
183 }
dvmInsnSetBranchTarget(InsnFlags * insnFlags,int addr,bool isBranch)184 INLINE void dvmInsnSetBranchTarget(InsnFlags* insnFlags, int addr,
185 bool isBranch)
186 {
187 assert(isBranch);
188 //if (isBranch)
189 insnFlags[addr] |= kInsnFlagBranchTarget;
190 //else
191 // insnFlags[addr] &= ~kInsnFlagBranchTarget;
192 }
193
194 /*
195 * Instruction is a GC point?
196 */
dvmInsnIsGcPoint(const InsnFlags * insnFlags,int addr)197 INLINE bool dvmInsnIsGcPoint(const InsnFlags* insnFlags, int addr) {
198 return (insnFlags[addr] & kInsnFlagGcPoint) != 0;
199 }
dvmInsnSetGcPoint(InsnFlags * insnFlags,int addr,bool isBranch)200 INLINE void dvmInsnSetGcPoint(InsnFlags* insnFlags, int addr,
201 bool isBranch)
202 {
203 assert(isBranch);
204 //if (isBranch)
205 insnFlags[addr] |= kInsnFlagGcPoint;
206 //else
207 // insnFlags[addr] &= ~kInsnFlagGcPoint;
208 }
209
210
211 /*
212 * Table that maps uninitialized instances to classes, based on the
213 * address of the new-instance instruction.
214 */
215 typedef struct UninitInstanceMap {
216 int numEntries;
217 struct {
218 int addr; /* code offset, or -1 for method arg ("this") */
219 ClassObject* clazz; /* class created at this address */
220 } map[1];
221 } UninitInstanceMap;
222 #define kUninitThisArgAddr (-1)
223 #define kUninitThisArgSlot 0
224
225 /*
226 * Create a new UninitInstanceMap.
227 */
228 UninitInstanceMap* dvmCreateUninitInstanceMap(const Method* meth,
229 const InsnFlags* insnFlags, int newInstanceCount);
230
231 /*
232 * Release the storage associated with an UninitInstanceMap.
233 */
234 void dvmFreeUninitInstanceMap(UninitInstanceMap* uninitMap);
235
236 /*
237 * Associate a class with an address. Returns the map slot index, or -1
238 * if the address isn't listed in the map (shouldn't happen) or if a
239 * different class is already associated with the address (shouldn't
240 * happen either).
241 */
242 //int dvmSetUninitInstance(UninitInstanceMap* uninitMap, int addr,
243 // ClassObject* clazz);
244
245 /*
246 * Return the class associated with an uninitialized reference. Pass in
247 * the map index.
248 */
249 //ClassObject* dvmGetUninitInstance(const UninitInstanceMap* uninitMap, int idx);
250
251 /*
252 * Clear the class associated with an uninitialized reference. Pass in
253 * the map index.
254 */
255 //void dvmClearUninitInstance(UninitInstanceMap* uninitMap, int idx);
256
257
258 /*
259 * Verify bytecode in "meth". "insnFlags" should be populated with
260 * instruction widths and "in try" flags.
261 */
262 bool dvmVerifyCodeFlow(const Method* meth, InsnFlags* insnFlags,
263 UninitInstanceMap* uninitMap);
264
265 #endif /*_DALVIK_CODEVERIFY*/
266