Benjamin Kramer | 665a8dc | 2012-01-15 15:26:07 +0000 | [diff] [blame] | 1 | <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" |
| 2 | "http://www.w3.org/TR/html4/strict.dtd"> |
Ted Kremenek | d76e0a6 | 2009-04-03 01:38:55 +0000 | [diff] [blame] | 3 | <html> |
| 4 | <head> |
| 5 | <title>Static Analyzer Design Document: Memory Regions</title> |
| 6 | </head> |
| 7 | <body> |
| 8 | |
| 9 | <h1>Static Analyzer Design Document: Memory Regions</h1> |
| 10 | |
| 11 | <h3>Authors</h3> |
| 12 | |
| 13 | <p>Ted Kremenek, <tt>kremenek at apple</tt><br> |
| 14 | Zhongxing Xu, <tt>xuzhongzhing at gmail</tt></p> |
| 15 | |
| 16 | <h2 id="intro">Introduction</h2> |
| 17 | |
| 18 | <p>The path-sensitive analysis engine in libAnalysis employs an extensible API |
| 19 | for abstractly modeling the memory of an analyzed program. This API employs the |
| 20 | concept of "memory regions" to abstractly model chunks of program memory such as |
| 21 | program variables and dynamically allocated memory such as those returned from |
| 22 | 'malloc' and 'alloca'. Regions are hierarchical, with subregions modeling |
| 23 | subtyping relationships, field and array offsets into larger chunks of memory, |
| 24 | and so on.</p> |
| 25 | |
| 26 | <p>The region API consists of two components:</p> |
| 27 | |
| 28 | <ul> <li>A taxonomy and representation of regions themselves within the analyzer |
| 29 | engine. The primary definitions and interfaces are described in <tt><a |
| 30 | href="http://clang.llvm.org/doxygen/MemRegion_8h-source.html">MemRegion.h</a></tt>. |
| 31 | At the root of the region hierarchy is the class <tt>MemRegion</tt> with |
| 32 | specific subclasses refining the region concept for variables, heap allocated |
| 33 | memory, and so forth.</li> <li>The modeling of binding of values to regions. For |
| 34 | example, modeling the value stored to a local variable <tt>x</tt> consists of |
| 35 | recording the binding between the region for <tt>x</tt> (which represents the |
| 36 | raw memory associated with <tt>x</tt>) and the value stored to <tt>x</tt>. This |
| 37 | binding relationship is captured with the notion of "symbolic |
| 38 | stores."</li> </ul> |
| 39 | |
| 40 | <p>Symbolic stores, which can be thought of as representing the relation |
| 41 | <tt>regions -> values</tt>, are implemented by subclasses of the |
| 42 | <tt>StoreManager</tt> class (<tt><a |
| 43 | href="http://clang.llvm.org/doxygen/Store_8h-source.html">Store.h</a></tt>). A |
| 44 | particular StoreManager implementation has complete flexibility concerning the |
| 45 | following: |
| 46 | |
| 47 | <ul> |
| 48 | <li><em>How</em> to model the binding between regions and values</li> |
| 49 | <li><em>What</em> bindings are recorded |
| 50 | </ul> |
| 51 | |
| 52 | <p>Together, both points allow different StoreManagers to tradeoff between |
| 53 | different levels of analysis precision and scalability concerning the reasoning |
| 54 | of program memory. Meanwhile, the core path-sensitive engine makes no |
| 55 | assumptions about either points, and queries a StoreManager about the bindings |
| 56 | to a memory region through a generic interface that all StoreManagers share. If |
| 57 | a particular StoreManager cannot reason about the potential bindings of a given |
| 58 | memory region (e.g., '<tt>BasicStoreManager</tt>' does not reason about fields |
| 59 | of structures) then the StoreManager can simply return 'unknown' (represented by |
| 60 | '<tt>UnknownVal</tt>') for a particular region-binding. This separation of |
| 61 | concerns not only isolates the core analysis engine from the details of |
| 62 | reasoning about program memory but also facilities the option of a client of the |
| 63 | path-sensitive engine to easily swap in different StoreManager implementations |
Benjamin Kramer | 665a8dc | 2012-01-15 15:26:07 +0000 | [diff] [blame] | 64 | that internally reason about program memory in very different ways.</p> |
Ted Kremenek | d76e0a6 | 2009-04-03 01:38:55 +0000 | [diff] [blame] | 65 | |
| 66 | <p>The rest of this document is divided into two parts. We first discuss region |
| 67 | taxonomy and the semantics of regions. We then discuss the StoreManager |
| 68 | interface, and details of how the currently available StoreManager classes |
| 69 | implement region bindings.</p> |
| 70 | |
| 71 | <h2 id="regions">Memory Regions and Region Taxonomy</h2> |
| 72 | |
| 73 | <h3>Pointers</h3> |
| 74 | |
| 75 | <p>Before talking about the memory regions, we would talk about the pointers |
| 76 | since memory regions are essentially used to represent pointer values.</p> |
| 77 | |
| 78 | <p>The pointer is a type of values. Pointer values have two semantic aspects. |
| 79 | One is its physical value, which is an address or location. The other is the |
| 80 | type of the memory object residing in the address.</p> |
| 81 | |
| 82 | <p>Memory regions are designed to abstract these two properties of the pointer. |
| 83 | The physical value of a pointer is represented by MemRegion pointers. The rvalue |
| 84 | type of the region corresponds to the type of the pointee object.</p> |
| 85 | |
| 86 | <p>One complication is that we could have different view regions on the same |
| 87 | memory chunk. They represent the same memory location, but have different |
| 88 | abstract location, i.e., MemRegion pointers. Thus we need to canonicalize the |
| 89 | abstract locations to get a unique abstract location for one physical |
| 90 | location.</p> |
| 91 | |
| 92 | <p>Furthermore, these different view regions may or may not represent memory |
| 93 | objects of different types. Some different types are semantically the same, |
| 94 | for example, 'struct s' and 'my_type' are the same type.</p> |
| 95 | |
| 96 | <pre> |
| 97 | struct s; |
| 98 | typedef struct s my_type; |
| 99 | </pre> |
| 100 | |
| 101 | <p>But <tt>char</tt> and <tt>int</tt> are not the same type in the code below:</p> |
| 102 | |
| 103 | <pre> |
| 104 | void *p; |
| 105 | int *q = (int*) p; |
| 106 | char *r = (char*) p; |
Benjamin Kramer | 665a8dc | 2012-01-15 15:26:07 +0000 | [diff] [blame] | 107 | </pre> |
Ted Kremenek | d76e0a6 | 2009-04-03 01:38:55 +0000 | [diff] [blame] | 108 | |
| 109 | <p>Thus we need to canonicalize the MemRegion which is used in binding and |
| 110 | retrieving.</p> |
| 111 | |
Zhongxing Xu | cf665e1 | 2009-04-10 06:52:49 +0000 | [diff] [blame] | 112 | <h3>Regions</h3> |
| 113 | <p>Region is the entity used to model pointer values. A Region has the following |
| 114 | properties:</p> |
| 115 | |
| 116 | <ul> |
| 117 | <li>Kind</li> |
| 118 | |
| 119 | <li>ObjectType: the type of the object residing on the region.</li> |
| 120 | |
| 121 | <li>LocationType: the type of the pointer value that the region corresponds to. |
| 122 | Usually this is the pointer to the ObjectType. But sometimes we want to cache |
| 123 | this type explicitly, for example, for a CodeTextRegion.</li> |
| 124 | |
| 125 | <li>StartLocation</li> |
| 126 | |
| 127 | <li>EndLocation</li> |
| 128 | </ul> |
| 129 | |
Ted Kremenek | d76e0a6 | 2009-04-03 01:38:55 +0000 | [diff] [blame] | 130 | <h3>Symbolic Regions</h3> |
| 131 | |
| 132 | <p>A symbolic region is a map of the concept of symbolic values into the domain |
| 133 | of regions. It is the way that we represent symbolic pointers. Whenever a |
| 134 | symbolic pointer value is needed, a symbolic region is created to represent |
| 135 | it.</p> |
| 136 | |
| 137 | <p>A symbolic region has no type. It wraps a SymbolData. But sometimes we have |
| 138 | type information associated with a symbolic region. For this case, a |
| 139 | TypedViewRegion is created to layer the type information on top of the symbolic |
| 140 | region. The reason we do not carry type information with the symbolic region is |
| 141 | that the symbolic regions can have no type. To be consistent, we don't let them |
| 142 | to carry type information.</p> |
| 143 | |
| 144 | <p>Like a symbolic pointer, a symbolic region may be NULL, has unknown extent, |
| 145 | and represents a generic chunk of memory.</p> |
| 146 | |
| 147 | <p><em><b>NOTE</b>: We plan not to use loc::SymbolVal in RegionStore and remove it |
| 148 | gradually.</em></p> |
| 149 | |
| 150 | <p>Symbolic regions get their rvalue types through the following ways:</p> |
| 151 | |
| 152 | <ul> |
| 153 | <li>Through the parameter or global variable that points to it, e.g.: |
| 154 | <pre> |
| 155 | void f(struct s* p) { |
| 156 | ... |
| 157 | } |
| 158 | </pre> |
| 159 | |
| 160 | <p>The symbolic region pointed to by <tt>p</tt> has type <tt>struct |
| 161 | s</tt>.</p></li> |
| 162 | |
| 163 | <li>Through explicit or implicit casts, e.g.: |
| 164 | <pre> |
| 165 | void f(void* p) { |
| 166 | struct s* q = (struct s*) p; |
| 167 | ... |
| 168 | } |
| 169 | </pre> |
| 170 | </li> |
| 171 | </ul> |
| 172 | |
| 173 | <p>We attach the type information to the symbolic region lazily. For the first |
| 174 | case above, we create the <tt>TypedViewRegion</tt> only when the pointer is |
| 175 | actually used to access the pointee memory object, that is when the element or |
| 176 | field region is created. For the cast case, the <tt>TypedViewRegion</tt> is |
| 177 | created when visiting the <tt>CastExpr</tt>.</p> |
| 178 | |
| 179 | <p>The reason for doing lazy typing is that symbolic regions are sometimes only |
| 180 | used to do location comparison.</p> |
| 181 | |
| 182 | <h3>Pointer Casts</h3> |
| 183 | |
| 184 | <p>Pointer casts allow people to impose different 'views' onto a chunk of |
| 185 | memory.</p> |
| 186 | |
| 187 | <p>Usually we have two kinds of casts. One kind of casts cast down with in the |
| 188 | type hierarchy. It imposes more specific views onto more generic memory regions. |
| 189 | The other kind of casts cast up with in the type hierarchy. It strips away more |
| 190 | specific views on top of the more generic memory regions.</p> |
| 191 | |
| 192 | <p>We simulate the down casts by layering another <tt>TypedViewRegion</tt> on |
| 193 | top of the original region. We simulate the up casts by striping away the top |
| 194 | <tt>TypedViewRegion</tt>. Down casts is usually simple. For up casts, if the |
| 195 | there is no <tt>TypedViewRegion</tt> to be stripped, we return the original |
| 196 | region. If the underlying region is of the different type than the cast-to type, |
| 197 | we flag an error state.</p> |
| 198 | |
| 199 | <p>For toll-free bridging casts, we return the original region.</p> |
| 200 | |
| 201 | <p>We can set up a partial order for pointer types, with the most general type |
| 202 | <tt>void*</tt> at the top. The partial order forms a tree with <tt>void*</tt> as |
| 203 | its root node.</p> |
| 204 | |
| 205 | <p>Every <tt>MemRegion</tt> has a root position in the type tree. For example, |
| 206 | the pointee region of <tt>void *p</tt> has its root position at the root node of |
| 207 | the tree. <tt>VarRegion</tt> of <tt>int x</tt> has its root position at the 'int |
| 208 | type' node.</p> |
| 209 | |
| 210 | <p><tt>TypedViewRegion</tt> is used to move the region down or up in the tree. |
| 211 | Moving down in the tree adds a <tt>TypedViewRegion</tt>. Moving up in the tree |
| 212 | removes a <Tt>TypedViewRegion</tt>.</p> |
| 213 | |
| 214 | <p>Do we want to allow moving up beyond the root position? This happens |
| 215 | when:</p> <pre> int x; void *p = &x; </pre> |
| 216 | |
| 217 | <p>The region of <tt>x</tt> has its root position at 'int*' node. the cast to |
| 218 | void* moves that region up to the 'void*' node. I propose to not allow such |
Zhongxing Xu | 4a11551 | 2009-04-20 10:09:10 +0000 | [diff] [blame] | 219 | casts, and assign the region of <tt>x</tt> for <tt>p</tt>.</p> |
| 220 | |
| 221 | <p>Another non-ideal case is that people might cast to a non-generic pointer |
| 222 | from another non-generic pointer instead of first casting it back to the generic |
| 223 | pointer. Direct handling of this case would result in multiple layers of |
| 224 | TypedViewRegions. This enforces an incorrect semantic view to the region, |
| 225 | because we can only have one typed view on a region at a time. To avoid this |
| 226 | inconsistency, before casting the region, we strip the TypedViewRegion, then do |
| 227 | the cast. In summary, we only allow one layer of TypedViewRegion.</p> |
Ted Kremenek | d76e0a6 | 2009-04-03 01:38:55 +0000 | [diff] [blame] | 228 | |
| 229 | <h3>Region Bindings</h3> |
| 230 | |
| 231 | <p>The following region kinds are boundable: VarRegion, CompoundLiteralRegion, |
| 232 | StringRegion, ElementRegion, FieldRegion, and ObjCIvarRegion.</p> |
| 233 | |
| 234 | <p>When binding regions, we perform canonicalization on element regions and field |
| 235 | regions. This is because we can have different views on the same region, some |
| 236 | of which are essentially the same view with different sugar type names.</p> |
| 237 | |
| 238 | <p>To canonicalize a region, we get the canonical types for all TypedViewRegions |
| 239 | along the way up to the root region, and make new TypedViewRegions with those |
| 240 | canonical types.</p> |
| 241 | |
| 242 | <p>For Objective-C and C++, perhaps another canonicalization rule should be |
| 243 | added: for FieldRegion, the least derived class that has the field is used as |
| 244 | the type of the super region of the FieldRegion.</p> |
| 245 | |
| 246 | <p>All bindings and retrievings are done on the canonicalized regions.</p> |
| 247 | |
| 248 | <p>Canonicalization is transparent outside the region store manager, and more |
| 249 | specifically, unaware outside the Bind() and Retrieve() method. We don't need to |
| 250 | consider region canonicalization when doing pointer cast.</p> |
| 251 | |
| 252 | <h3>Constraint Manager</h3> |
| 253 | |
| 254 | <p>The constraint manager reasons about the abstract location of memory objects. |
| 255 | We can have different views on a region, but none of these views changes the |
| 256 | location of that object. Thus we should get the same abstract location for those |
| 257 | regions.</p> |
| 258 | |
| 259 | </body> |
| 260 | </html> |