1 : // Copyright 2012 Google Inc. All Rights Reserved.
2 : //
3 : // Licensed under the Apache License, Version 2.0 (the "License");
4 : // you may not use this file except in compliance with the License.
5 : // You may obtain a copy of the License at
6 : //
7 : // http://www.apache.org/licenses/LICENSE-2.0
8 : //
9 : // Unless required by applicable law or agreed to in writing, software
10 : // distributed under the License is distributed on an "AS IS" BASIS,
11 : // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 : // See the License for the specific language governing permissions and
13 : // limitations under the License.
14 : //
15 : // A block graph is an abstract graph of blocks, each of which has an ID, a
16 : // type, a size and a few other properties. Each block represents either code or
17 : // data, and blocks can reference one another through references of various
18 : // types.
19 : //
20 : // The BlockGraph also stores minimum knowledge of sections (names and
21 : // characteristics), and each block belongs to at most one section. In this
22 : // sense, a BlockGraph acts as top-level division of blocks.
23 :
24 : #ifndef SYZYGY_BLOCK_GRAPH_BLOCK_GRAPH_H_
25 : #define SYZYGY_BLOCK_GRAPH_BLOCK_GRAPH_H_
26 :
27 : #include <hash_map>
28 : #include <map>
29 : #include <set>
30 : #include <string>
31 : #include <vector>
32 :
33 : #include "base/basictypes.h"
34 : #include "base/file_util.h"
35 : #include "base/strings/string_piece.h"
36 : #include "syzygy/common/align.h"
37 : #include "syzygy/core/address.h"
38 : #include "syzygy/core/address_space.h"
39 : #include "syzygy/core/string_table.h"
40 :
41 : namespace block_graph {
42 :
43 : // Forward declaration.
44 : class BlockGraphSerializer;
45 :
46 : // NOTE: When adding attributes be sure to update any uses of them in
47 : // block_graph.cc, for example in MergeIntersectingBlocks.
48 : #define BLOCK_ATTRIBUTE_ENUM(F) \
49 : /* Set for functions declared non-returning. */ \
50 : F(NON_RETURN_FUNCTION) \
51 : /* Set for blocks that are inferred by the decomposer. */ \
52 : F(GAP_BLOCK) \
53 : /* Set for blocks that are parsed by the PEFileParser. These */ \
54 : /* blocks are unmovable, indivisible, etc, and have to be treated */ \
55 : /* specially. */ \
56 : F(PE_PARSED) \
57 : /* Set for blocks that are created from section contribution */ \
58 : /* information or directly from COFF sections. */ \
59 : F(SECTION_CONTRIB) \
60 : /* This is used to indicate that a block consists purely of padding */ \
61 : /* data. */ \
62 : F(PADDING_BLOCK) \
63 : /* Indicates blocks that contain inline assembly. */ \
64 : F(HAS_INLINE_ASSEMBLY) \
65 : /* Indicates that the block was built by a compiler whose precise */ \
66 : /* behaviour and semantics we are unfamiliar with. */ \
67 : F(BUILT_BY_UNSUPPORTED_COMPILER) \
68 : /* Indicates that the block has been built by the Syzygy toolchain, and */ \
69 : /* thus is inherently safe for basic-block decomposition without having */ \
70 : /* to perform the myriad of safety checks we do otherwise. */ \
71 : F(BUILT_BY_SYZYGY) \
72 : /* Deprecated: used by old decomposer. */ \
73 : F(RESERVED_ATTRIBUTE1) \
74 : /* Deprecated: used by old decomposer. */ \
75 : F(RESERVED_ATTRIBUTE2) \
76 : /* This is set for functions that have exception handling enabled. */ \
77 : /* Without delving far deeper into the specifics, it is unsafe to basic */ \
78 : /* block decompose these blocks. */ \
79 : F(HAS_EXCEPTION_HANDLING) \
80 : /* Deprecated: used by old decomposer. */ \
81 : F(RESERVED_ATTRIBUTE3) \
82 : /* This is set for blocks that have a thunk symbol pointing to them. */ \
83 : /* Typically thunk blocks are compiler or linker-generated, such as */ \
84 : /* e.g. import thunks, delay load import thunks, etc. */ \
85 : F(THUNK) \
86 : /* This is set for blocks that have been parsed as COFF groups. The */ \
87 : /* contents of these blocks are semantically indivisible. */ \
88 : F(COFF_GROUP) \
89 : /* COFF headers block; not set for PE. */ \
90 : F(COFF_HEADERS) \
91 : /* COFF symbol table. */ \
92 : F(COFF_SYMBOL_TABLE) \
93 : /* COFF string table. */ \
94 : F(COFF_STRING_TABLE) \
95 : /* COFF relocation table; these should be ignored when dealing with */ \
96 : /* block graphs, as all the information is represented as references. */ \
97 : F(COFF_RELOC_DATA) \
98 : /* COFF BSS (unmapped) block; has size but no data. */ \
99 : F(COFF_BSS) \
100 : /* Contains unsupported instructions. */ \
101 : F(UNSUPPORTED_INSTRUCTIONS) \
102 : /* This always needs to be set to the next available attribute bit. */ \
103 : F(BLOCK_ATTRIBUTES_MAX)
104 :
105 : // The BlockGraph is a top-level container for Blocks.
106 : class BlockGraph {
107 : public:
108 : typedef core::RelativeAddress RelativeAddress;
109 :
110 : typedef size_t SectionId;
111 : typedef size_t BlockId;
112 : typedef size_t Size;
113 : typedef ptrdiff_t Offset;
114 : typedef uint32 BlockAttributes;
115 : typedef uint32 LabelAttributes;
116 :
117 : // The BlockGraph maintains a list of sections, and each block belongs
118 : // to one of them. This is the set of information we keep regarding them.
119 : struct Section;
120 : // The section map contains all sections, indexed by id.
121 : typedef std::map<SectionId, Section> SectionMap;
122 :
123 : static const SectionId kInvalidSectionId;
124 :
125 : // The types of image formats that we can currently represent in a
126 : // BlockGraph.
127 : enum ImageFormat {
128 : UNKNOWN_IMAGE_FORMAT,
129 : PE_IMAGE,
130 : COFF_IMAGE,
131 :
132 : // This must always be the last entry, and kImageFormat must be kept in
133 : // sync with this enum.
134 : IMAGE_FORMAT_MAX
135 : };
136 :
137 : static const char* ImageFormatToString(ImageFormat format);
138 :
139 : // Assign distinct bit IDs to each attribute constant.
140 : enum BlockAttributeIdEnum {
141 : #define DECLARE_ENUM_BIT(name) name##_BIT,
142 : BLOCK_ATTRIBUTE_ENUM(DECLARE_ENUM_BIT)
143 : #undef DECLARE_ENUM_BIT
144 : };
145 :
146 : enum BlockAttributeEnum {
147 : #define DECLARE_ENUM(name) name = (1 << name##_BIT),
148 : BLOCK_ATTRIBUTE_ENUM(DECLARE_ENUM)
149 : #undef DECLARE_ENUM
150 : };
151 :
152 : // Returns a string containing the names of the supplied attributes.
153 : static std::string BlockAttributesToString(BlockAttributes attrs);
154 :
155 : enum BlockType {
156 : CODE_BLOCK,
157 : DATA_BLOCK,
158 :
159 : // NOTE: This must always be last, and kBlockType must be kept in sync
160 : // with this enum.
161 : BLOCK_TYPE_MAX
162 : };
163 :
164 : static const char* BlockTypeToString(BlockType type);
165 :
166 : // Label attributes. Attributes of the form _END_LABEL type actually
167 : // point to the first byte past the range they delineate. To make the
168 : // semantics of moving labels easier, we shift these labels left by one and
169 : // make them follow the last byte of the delineated range.
170 : enum LabelAttributesEnum {
171 : // The label points to an entry-point in a code block.
172 : CODE_LABEL = (1 << 0),
173 :
174 : // Mark the start and end of the debuggable portion of a code block.
175 : DEBUG_START_LABEL = (1 << 1),
176 : DEBUG_END_LABEL = (1 << 2),
177 :
178 : // Mark the start and end of an embedded scope in a code block.
179 : SCOPE_START_LABEL = (1 << 3),
180 : SCOPE_END_LABEL = (1 << 4),
181 :
182 : // Marks the location of a (virtual table?) call.
183 : CALL_SITE_LABEL = (1 << 5),
184 :
185 : // The label points to the start of a jump table. The length is inferred
186 : // by the location of the next label, or the end of the block. This will
187 : // also have DATA_LABEL set.
188 : JUMP_TABLE_LABEL = (1 << 6),
189 : // The label points to the start of a case table. The length is inferred
190 : // by the location of the next label, or the end of the block. This will
191 : // also have DATA_LABEL set.
192 : CASE_TABLE_LABEL = (1 << 7),
193 : // The label originated from a data symbol. The length is inferred by the
194 : // location of the next label, or the end of the block. The type of data
195 : // is unknown.
196 : DATA_LABEL = (1 << 8),
197 :
198 : // Used to mark a label that was derived from a public symbol. These are
199 : // usually actually pointing to code and global data symbols, but we can't
200 : // always tell (there may be public symbols pointing to data in a code
201 : // block).
202 : PUBLIC_SYMBOL_LABEL = (1 << 9),
203 :
204 : // This always needs to be the most significant bit.
205 : LABEL_ATTRIBUTES_MAX = (1 << 10),
206 : };
207 :
208 : static std::string LabelAttributesToString(LabelAttributes label_attributes);
209 :
210 : enum ReferenceType {
211 : // Common reference types.
212 : PC_RELATIVE_REF,
213 : ABSOLUTE_REF,
214 : RELATIVE_REF,
215 : FILE_OFFSET_REF,
216 :
217 : // Object-file reference types.
218 : SECTION_REF,
219 : SECTION_OFFSET_REF,
220 :
221 : // Relocation reference types.
222 : // TODO(chrisha): Do we need a separate bit, and different ABS and REL
223 : // reference types for COFF files?
224 : RELOC_REF_BIT = 1u << 3,
225 : RELOC_PC_RELATIVE_REF = PC_RELATIVE_REF | RELOC_REF_BIT,
226 : RELOC_ABSOLUTE_REF = ABSOLUTE_REF | RELOC_REF_BIT,
227 : RELOC_RELATIVE_REF = RELATIVE_REF | RELOC_REF_BIT,
228 : RELOC_SECTION_REF = SECTION_REF | RELOC_REF_BIT,
229 : RELOC_SECTION_OFFSET_REF = SECTION_OFFSET_REF | RELOC_REF_BIT,
230 :
231 : // Must be last!
232 : REFERENCE_TYPE_MAX,
233 : };
234 :
235 : // Forward declarations.
236 : class AddressSpace;
237 : class Block;
238 : class Label;
239 : class Reference;
240 :
241 : // The block map contains all blocks, indexed by id.
242 : typedef std::map<BlockId, Block> BlockMap;
243 :
244 : BlockGraph();
245 : ~BlockGraph();
246 :
247 : // Adds a section with the given name.
248 : //
249 : // @param name The section name.
250 : // @param characteristics The section characteristics.
251 : // @returns the newly created section.
252 : Section* AddSection(const base::StringPiece& name, uint32 characteristics);
253 :
254 : // Finds a section with the given name, returning NULL if no such section
255 : // exists.
256 : //
257 : // @param name The section name.
258 : // @returns the section if one is found, NULL otherwise.
259 : Section* FindSection(const base::StringPiece& name);
260 : const Section* FindSection(const base::StringPiece& name) const;
261 :
262 : // Find or adds section with the given name.
263 : //
264 : // If a section with the given name already exists, updates its
265 : // characteristics and returns it. Otherwise, creates a new section and
266 : // returns it. If multiple sections exist with the given name, the first
267 : // one encountered is returned.
268 : //
269 : // TODO(chrisha): The semantics of this function are a little odd. It would
270 : // make more sense for it to return only if a section with matching name
271 : // AND characteristics is found, otherwise to create a new one.
272 : //
273 : // @param name The section name.
274 : // @param characteristics The section characteristics.
275 : // @returns the new or found section.
276 : Section* FindOrAddSection(const base::StringPiece& name,
277 : uint32 characteristics);
278 :
279 : // Removes the given section from the BlockGraph.
280 : //
281 : // The section must belong to this block graph. Be aware that this can leave
282 : // Blocks with dangling section_ids.
283 : //
284 : // @param section The section to remove.
285 : // @returns true on success, false otherwise.
286 : bool RemoveSection(Section* section);
287 :
288 : // Removes the section with the given id from the BlockGraph.
289 : //
290 : // @param id The id of the section to remove.
291 : // @returns true on success, false otherwise.
292 : bool RemoveSectionById(SectionId id);
293 :
294 : // Add @p block of type @p type and @p size and
295 : // return the new block.
296 : // @returns the new block.
297 : Block* AddBlock(BlockType type, Size size, const base::StringPiece& name);
298 :
299 : // Deletes the given block from the BlockGraph. The block must belong to this
300 : // block graph, and have no references or referrers. Returns true on success,
301 : // false otherwise. On failure, the BlockGraph has not been changed.
302 : bool RemoveBlock(Block* block);
303 :
304 : // Deletes the block with the given @p id from the block graph. The block id
305 : // must be valid, and the block must have no references or referrers. Returns
306 : // true on success, false otherwise. On failure, the BlockGraph has not been
307 : // changed.
308 : bool RemoveBlockById(BlockId id);
309 :
310 : // Accessors.
311 E : const SectionMap& sections() const { return sections_; }
312 E : SectionMap& sections_mutable() { return sections_; }
313 E : const BlockMap& blocks() const { return blocks_; }
314 E : BlockMap& blocks_mutable() { return blocks_; }
315 :
316 : // @{
317 : // Retrieve the section with the given id.
318 : //
319 : // @param id The id of the section to retrieve.
320 : // @returns the section in question or NULL if no such section.
321 : Section* GetSectionById(SectionId id);
322 : const Section* GetSectionById(SectionId id) const;
323 : // @}
324 :
325 : // @{
326 : // Retrieve the block with id.
327 : // @returns the block in question or NULL if no such block.
328 : Block* GetBlockById(BlockId id);
329 : const Block* GetBlockById(BlockId id) const;
330 : // @}
331 :
332 : // Get the string table.
333 : // @returns the string table of this BlockGraph.
334 E : core::StringTable& string_table() { return string_table_; }
335 :
336 : // Sets the image format.
337 : // @param image_format The format of the image.
338 E : void set_image_format(ImageFormat image_format) {
339 E : image_format_ = image_format;
340 E : }
341 : // @returns the image format.
342 E : ImageFormat image_format() const { return image_format_; }
343 :
344 : private:
345 : // Give BlockGraphSerializer access to our innards for serialization.
346 : friend BlockGraphSerializer;
347 :
348 : // Removes a block by the iterator to it. The iterator must be valid.
349 : bool RemoveBlockByIterator(BlockMap::iterator it);
350 :
351 : // All sections we contain.
352 : SectionMap sections_;
353 :
354 : // Our section ID allocator.
355 : SectionId next_section_id_;
356 :
357 : // All blocks we contain.
358 : BlockMap blocks_;
359 :
360 : // Our block ID allocator.
361 : BlockId next_block_id_;
362 :
363 : // A string table used to intern strings.
364 : core::StringTable string_table_;
365 :
366 : // The format of the image represented by this block graph. Defaults to
367 : // UNKNOWN_IMAGE_FORMAT. Usually initialized by the appropriate decomposer.
368 : ImageFormat image_format_;
369 :
370 : DISALLOW_COPY_AND_ASSIGN(BlockGraph);
371 : };
372 :
373 : // The BlockGraph maintains a list of sections, and each block belongs
374 : // to one of them. This is the set of information we keep regarding them.
375 : struct BlockGraph::Section {
376 : // Default constructor. Required for serialization.
377 E : Section() : id_(kInvalidSectionId), characteristics_(0) {
378 E : }
379 :
380 : // Full constructor.
381 : //
382 : // @param id The section id. This must not be kInvalidSectionId.
383 : // @param name The name of the section. Must not be empty or NULL.
384 : // @param characteristics The characteristics of the section.
385 E : Section(SectionId id, const base::StringPiece& name, uint32 characteristics)
386 : : id_(id), name_(), characteristics_(characteristics) {
387 E : DCHECK_NE(kInvalidSectionId, id);
388 E : DCHECK(name != NULL);
389 E : name.CopyToString(&name_);
390 E : DCHECK(!name_.empty());
391 E : }
392 :
393 : // Get the id of this section.
394 : //
395 : // @returns the id of the section.
396 E : SectionId id() const { return id_; }
397 :
398 : // Get the name of this section.
399 : //
400 : // @returns the section name.
401 E : const std::string& name() const { return name_; }
402 :
403 : // Sets the name for this section.
404 : //
405 : // @param name The name of the section. If NULL or empty, this will fail.
406 : // @returns true if the name is set, false otherwise.
407 : bool set_name(const base::StringPiece& name);
408 :
409 : // Get the characteristics of this section.
410 : //
411 : // @returns the section characteristics.
412 E : uint32 characteristics() const { return characteristics_; }
413 :
414 : // Sets the characteristics for this section.
415 : //
416 : // @param characteristics The new characteristics to set.
417 E : void set_characteristics(uint32 characteristics) {
418 E : characteristics_ = characteristics;
419 E : }
420 :
421 : // Sets a one or more additional characteristics for this section.
422 : //
423 : // @param characteristic The new characteristic(s) to set for this section.
424 E : void set_characteristic(uint32 characteristic) {
425 E : characteristics_ |= characteristic;
426 E : }
427 :
428 : // Clears one or more characteristics for this section.
429 : //
430 : // @param characteristic The characteristic(s) to clear for this section.
431 E : void clear_characteristic(uint32 characteristic) {
432 E : characteristics_ &= ~characteristic;
433 E : }
434 :
435 : // @name Serialization functions.
436 : // @{
437 : bool Save(core::OutArchive* out_archive) const;
438 : bool Load(core::InArchive* in_archive);
439 : // @}
440 :
441 : // A simple comparison operator for serialization tests.
442 E : bool operator==(const Section& other) const {
443 : return id_ == other.id_ && name_ == other.name_ &&
444 E : characteristics_ == other.characteristics_;
445 E : }
446 :
447 : // A not-equal comparison operator.
448 E : bool operator!=(const Section& other) const {
449 E : return !operator==(other);
450 E : }
451 :
452 : private:
453 : // The id of the section. This has no particular meaning other than as a way
454 : // to identify sections uniquely.
455 : SectionId id_;
456 : // The name of the section. This will be truncated to a max of 8 characters
457 : // on output.
458 : std::string name_;
459 : // The section characteristics, a bitmask of IMAGE_SCN_* values.
460 : uint32 characteristics_;
461 : };
462 :
463 : // A label denotes the beginning (or end) of a sub-region within a (code)
464 : // block. In particular, a code label represents an instruction boundary
465 : // at which disassembly can begin and a data label represents the beginning
466 : // of embedded data.
467 : class BlockGraph::Label {
468 : public:
469 : // Default constructor.
470 E : Label() : attributes_(0) {
471 E : }
472 :
473 : // Full constructor.
474 E : Label(const base::StringPiece& name, LabelAttributes attributes)
475 : : name_(name.begin(), name.end()), attributes_(attributes) {
476 E : }
477 :
478 : // Copy construction.
479 E : Label(const Label& other)
480 : : name_(other.name_), attributes_(other.attributes_) {
481 E : }
482 :
483 : // @name Accessors.
484 : // @{
485 E : const std::string& name() const { return name_; }
486 : // @}
487 :
488 : // A helper function for logging and debugging.
489 : std::string ToString() const;
490 :
491 : // Equality comparator for unittesting.
492 E : bool operator==(const Label& other) const {
493 E : return name_ == other.name_ && attributes_ == other.attributes_;
494 E : }
495 :
496 : // The label attributes are a bitmask. You can set them wholesale,
497 : // or set and clear them individually by bitmasking.
498 E : LabelAttributes attributes() const { return attributes_; }
499 E : void set_attributes(LabelAttributes attributes) { attributes_ = attributes; }
500 :
501 : // Set or clear one or more attributes.
502 E : void set_attribute(LabelAttributes attribute) { attributes_ |= attribute; }
503 E : void clear_attribute(LabelAttributes attribute) { attributes_ &= ~attribute; }
504 :
505 : // Determines if all or any of the given attributes are set.
506 E : bool has_attributes(LabelAttributes attributes) const {
507 E : return (attributes_ & attributes) == attributes;
508 E : }
509 E : bool has_any_attributes(LabelAttributes attributes) const {
510 E : return (attributes_ & attributes) != 0;
511 E : }
512 :
513 : // @returns true if this label is valid, false otherwise.
514 : bool IsValid() const;
515 :
516 : // Tests a set of label attributes for validity.
517 : // @param attributes the attributes to test.
518 : // @returns true if the provided attributes are valid, false otherwise.
519 : static bool AreValidAttributes(LabelAttributes attributes);
520 :
521 : private:
522 : // The name by which this label is known.
523 : std::string name_;
524 :
525 : // The disposition of the bytes found at this label.
526 : LabelAttributes attributes_;
527 : };
528 :
529 : // A block represents a block of either code or data.
530 : //
531 : // Since blocks may be split and up and glued together in arbitrary ways, each
532 : // block maintains an address-space over its data, associating ranges of block
533 : // data to ranges of bytes in the original image. This effectively encodes OMAP
534 : // data, allowing the PDB file to be updated.
535 : //
536 : // Each block also stores references to other blocks in the graph, their
537 : // relative location within the block and their type and size.
538 : //
539 : // Each block has a set of attributes, including a size, a name and a
540 : // "current" address. Most of those attributes are mutable, and are set in the
541 : // process of creating and manipulating images and graph address spaces.
542 : class BlockGraph::Block {
543 : public:
544 : // Set of the blocks that have a reference to this block.
545 : // This is keyed on block and source offset (not destination offset),
546 : // to allow one to easily locate and remove the backreferences on change or
547 : // deletion.
548 : typedef std::pair<Block*, Offset> Referrer;
549 : typedef std::set<Referrer> ReferrerSet;
550 :
551 : // Map of references that this block makes to other blocks.
552 : typedef std::map<Offset, Reference> ReferenceMap;
553 :
554 : // Represents a range of data in this block.
555 : typedef core::AddressRange<Offset, Size> DataRange;
556 :
557 : // Represents a range of data in the original image.
558 : typedef core::AddressRange<RelativeAddress, Size> SourceRange;
559 :
560 : // A map between bytes in this block and bytes in the original image.
561 : typedef core::AddressRangeMap<DataRange, SourceRange> SourceRanges;
562 :
563 : // The flags that can be passed to the TransferReferrers function.
564 : enum TransferReferrersFlags {
565 : kSkipInternalReferences = (1 << 0),
566 : kTransferInternalReferences = (1 << 1)
567 : };
568 :
569 : // Typed labels associated with various offsets in the block. Some of these
570 : // labels (of type CODE_LABEL) represent code start points for disassembly
571 : // while others (of type DATA_LABEL) represent the start of embedded data
572 : // within the block. Note that, while possible, it is NOT guaranteed that
573 : // all basic blocks are marked with a label. Basic block decomposition should
574 : // disassemble from the code labels to discover all basic blocks.
575 : typedef std::map<Offset, Label> LabelMap;
576 :
577 : ~Block();
578 :
579 : // Accessors.
580 E : BlockId id() const { return id_; }
581 E : BlockType type() const { return type_; }
582 E : void set_type(BlockType type) { type_ = type; }
583 :
584 E : Size size() const { return size_; }
585 :
586 : // Set the total size of the block. Note that allocated data_size_ must
587 : // always be less than or equal to the total size.
588 E : void set_size(Size size) {
589 E : DCHECK_LE(data_size_, size);
590 E : size_ = size;
591 E : }
592 :
593 E : const std::string& name() const {
594 E : DCHECK(name_ != NULL);
595 E : return *name_;
596 E : }
597 : void set_name(const base::StringPiece& name);
598 :
599 : const std::string& compiland_name() const;
600 : void set_compiland_name(const base::StringPiece& name);
601 :
602 : // Gets the alignment of the block. Note that the alignment is applied to
603 : // the byte at the alignment offset.
604 : // @returns The current alignment of the block.
605 E : Size alignment() const { return alignment_; }
606 : // Sets the alignment that applies to the block when building the layout.
607 : // By default, the first byte of the block will be aligned, but that can
608 : // be overridden by setting the alignment offset.
609 : // @param alignment The new alignment. The alignment should be a power of two.
610 : // The value of zero is invalid, use 1 to signal no alignment requirements.
611 E : void set_alignment(Size alignment) {
612 : // Ensure that alignment is a non-zero power of two.
613 E : DCHECK(common::IsPowerOfTwo(alignment));
614 E : alignment_ = alignment;
615 E : }
616 :
617 : // Gets the offset the alignment is applied to.
618 : // @returns The offset that will be aligned.
619 E : Offset alignment_offset() const { return alignment_offset_; }
620 : // Sets the offset the alignment refers to. The layout builder makes sure the
621 : // byte at this position is aligned. By default this value is zero which
622 : // means that the first byte of the block should be aligned.
623 : // @param alignment_offset The new offset of the alignment.
624 E : void set_alignment_offset(Offset alignment_offset) {
625 E : alignment_offset_ = alignment_offset;
626 E : }
627 :
628 : // Returns the minimum amount of padding bytes that are inserted before the
629 : // block when building the layout.
630 : // @returns The minimum amount of padding.
631 E : Size padding_before() const { return padding_before_; }
632 : // Sets the minium amount of padding bytes before the block when building the
633 : // layout.
634 : // @param padding_before At least this amount of bytes are inserted before
635 : // the block when building the layout.
636 E : void set_padding_before(Size padding_before) {
637 E : padding_before_ = padding_before;
638 E : }
639 :
640 : // The address of the block is set any time the block is assigned
641 : // an address in an address space.
642 E : RelativeAddress addr() const { return addr_; }
643 E : void set_addr(RelativeAddress addr) { addr_ = addr; }
644 :
645 : // The section ID for the block. These IDs are wrt to the SectionMap in the
646 : // parent BlockGraph.
647 E : SectionId section() const { return section_; }
648 E : void set_section(SectionId section) { section_ = section; }
649 :
650 : // The block attributes are a bitmask. You can set them wholesale,
651 : // or set and clear them individually by bitmasking.
652 E : BlockAttributes attributes() const { return attributes_; }
653 E : void set_attributes(BlockAttributes attributes) { attributes_ = attributes; }
654 :
655 : // Set or clear one or more attributes.
656 E : void set_attribute(BlockAttributes attribute) { attributes_ |= attribute; }
657 E : void clear_attribute(BlockAttributes attribute) {
658 E : attributes_ &= ~attribute;
659 E : }
660 :
661 : // This is true iff data_ is in the ownership of the block.
662 : // Iff true, the block will delete [] data_ on destruction or when
663 : // data is overwritten.
664 E : bool owns_data() const { return owns_data_; }
665 :
666 : // Makes room for the given amount of data at the given offset. This is
667 : // special in that it will patch up any labels, source ranges and referrers
668 : // that land beyond the newly created data, shifting them to the right by
669 : // @p size. If the data for this block is actually allocated it will also
670 : // patch up the allocated data by zeroing the newly allocate range of data,
671 : // and shifting the tail by @p size. If the new data is strictly implicit
672 : // (offset > data_size), then the allocated data is not affected in any way
673 : // unless @p always_allocate_data is true.
674 : //
675 : // @param offset the offset at which to insert the new data.
676 : // @param size the size of the new data to be inserted.
677 : // @param always_allocate_data if true, then data_size will be grown if
678 : // necessary to ensure that the newly created data can be written.
679 : // @pre 0 <= offset <= size()
680 : void InsertData(Offset offset, Size size, bool always_allocate_data);
681 :
682 : // Removes the data in the given range. This will refuse to remove labels,
683 : // references and referrers that land in the range, and will fail if any
684 : // exist. It will also shift any labels, references and referrers that land
685 : // beyond the end of the removed range. Source ranges will also be fixed. If
686 : // the removed range lies within the initialized data then the data will also
687 : // be truncated/shifted as necessary.
688 : //
689 : // @param offset the offset at which to remove data.
690 : // @param size the size of the data to remove, in bytes.
691 : // @returns true on success, false otherwise.
692 : // @pre 0 <= offset <= size
693 : bool RemoveData(Offset offset, Size size);
694 :
695 : // Performs an inline resize of data in a BlockGraph. If the data is shrinking
696 : // this equates to a RemoveData operation. If it is growing it equates to an
697 : // InsertData operation.
698 : //
699 : // @param offset the offset of the data to resize.
700 : // @param current_size the current size of the data to resize.
701 : // @param new_size the desired size of the data.
702 : // @param always_allocate_data if true, then data_size will be grown if
703 : // necessary to ensure that the resized data can be written.
704 : // @returns true on success, false otherwise.
705 : // @pre 0 <= offset <= size
706 : bool InsertOrRemoveData(Offset offset, Size current_size, Size new_size,
707 : bool always_allocate_data);
708 :
709 : // Set the data the block refers to.
710 : // @param data NULL or the data this block refers to.
711 : // The underlying data must outlive this block.
712 : // @param data_size the size of data, or zero if data == NULL.
713 : // @pre data_size <= size().
714 : void SetData(const uint8* data, size_t data_size);
715 :
716 : // Allocates and returns a new data buffer of the given size. The returned
717 : // data will have been initialized to zero.
718 : // @pre data_size > 0.
719 : // @pre data_size <= size().
720 : uint8* AllocateData(size_t data_size);
721 :
722 : // Makes a copy of data, returns a pointer to the copy.
723 : // @pre data_size <= size().
724 : uint8* CopyData(size_t data_size, const void* data);
725 :
726 : // Resizes data to new_size by truncating or zero-extending the current data.
727 : // @pre new_size <= size().
728 : const uint8* ResizeData(size_t new_size);
729 :
730 : // Returns a mutable copy of the block's data. If the block doesn't own
731 : // the data on entry, it'll be copied and the copy returned to the caller.
732 : uint8* GetMutableData();
733 :
734 : // The data bytes the block refers to.
735 E : const uint8* data() const { return data_; }
736 :
737 : // The data size may be smaller than the block size (see size()),
738 : // when the block e.g. refers to data that's all or part
739 : // zero-initialized by the linker/loader.
740 E : size_t data_size() const { return data_size_; }
741 :
742 E : const ReferenceMap& references() const { return references_; }
743 E : const ReferrerSet& referrers() const { return referrers_; }
744 E : const SourceRanges& source_ranges() const { return source_ranges_; }
745 E : SourceRanges& source_ranges() { return source_ranges_; }
746 E : const LabelMap& labels() const { return labels_; }
747 :
748 : // Returns true if there are any other blocks holding a reference to this one.
749 : bool HasExternalReferrers() const;
750 :
751 : // Set the reference at @p offset to @p ref.
752 : // If there's a pre-existing reference at @p offset, this overrides it.
753 : // @param offset offset of the reference into this block.
754 : // @param ref the reference to add.
755 : // @returns true iff this inserts a new reference.
756 : bool SetReference(Offset offset, const Reference& ref);
757 :
758 : // Retrieve the reference at @p offset if one exists.
759 : // @param reference on success returns the reference @p offset.
760 : // @returns true iff there was a reference at @p offset.
761 : bool GetReference(Offset offset, Reference* reference) const;
762 :
763 : // Remove the reference at @p offset.
764 : // @returns true iff there was a reference at @p offset.
765 : bool RemoveReference(Offset offset);
766 :
767 : // Remove all references from this block. This is handy when removing a block
768 : // from the block graph.
769 : bool RemoveAllReferences();
770 :
771 : // Set a label to @p offset.
772 : // A label in code marks the location of the start of an instruction -
773 : // e.g. a location where disassembly can usefully commence. Labels
774 : // appear to be inserted by the VS tool chain where e.g. a switch
775 : // statement is implemented with a jump table, to note the location
776 : // of the jump destinations.
777 : // @param offset the offset of the label to set.
778 : // @param name the name of the label.
779 : // @param attributes the attributes of the label.
780 : // @returns true iff a new label is inserted.
781 : // @note that only one label can exist at each offset, and the first
782 : // label set at any offset will stay there.
783 : // @{
784 : bool SetLabel(Offset offset, const Label& label);
785 : bool SetLabel(Offset offset,
786 : const base::StringPiece& name,
787 E : LabelAttributes attributes) {
788 E : return SetLabel(offset, Label(name, attributes));
789 E : }
790 : // @}
791 :
792 : // Gets the label at the given @p offset.
793 : // @param offset the offset of the label to get.
794 : // @param label the string to receive the label.
795 : // @returns true if the label exists, false otherwise.
796 : bool GetLabel(Offset offset, Label* label) const;
797 :
798 : // Removes the label at the given @p offset.
799 : // @param offset the offset of the label to remove.
800 : // @returns true if the label existed and was removed, false it it did not
801 : // exist.
802 : bool RemoveLabel(Offset offset);
803 :
804 : // Returns true iff the block has a label at @p offset.
805 : // @param offset the offset of the label to search for.
806 : bool HasLabel(Offset offset) const;
807 :
808 : // Change all references to this block to refer to @p new_block instead,
809 : // while offsetting each reference by @p offset.
810 : // @param offset The offset that we should apply to each reference.
811 : // @param new_block The block the reference should point to.
812 : // @param flags The flags that control some parameters of this function.
813 : // @note this fails if any of the transferred references end up with offsets
814 : // less than zero, or greater than new_block->size().
815 : // @returns true iff all references were transferred successfully.
816 : bool TransferReferrers(Offset offset,
817 : Block* new_block,
818 : TransferReferrersFlags flags);
819 :
820 : // Returns true if this block contains the given range of bytes.
821 : bool Contains(RelativeAddress address, size_t size) const;
822 :
823 : protected:
824 : // Give BlockGraph access to our innards for serialization.
825 : friend class BlockGraph;
826 : // Give BlockGraphSerializer access to our innards for serialization.
827 : friend class BlockGraphSerializer;
828 :
829 : // Full constructor.
830 : // @note This is protected so that blocks may only be created via the
831 : // BlockGraph factory.
832 : Block(BlockId id,
833 : BlockType type,
834 : Size size,
835 : const base::StringPiece& name,
836 : BlockGraph* block_graph);
837 :
838 : // This constructor is used by serialization.
839 : explicit Block(BlockGraph* block_graph);
840 :
841 : // Allocates and returns a new data buffer of the given size. The returned
842 : // data buffer will not have been initialized in any way.
843 : uint8* AllocateRawData(size_t size);
844 :
845 : BlockId id_;
846 : BlockType type_;
847 : Size size_;
848 : Size alignment_;
849 : Offset alignment_offset_;
850 : Size padding_before_;
851 : const std::string* name_;
852 : const std::string* compiland_name_;
853 : RelativeAddress addr_;
854 :
855 : // BlockGraph to which belongs this Block. A block can only belongs to one
856 : // Block Graph.
857 : BlockGraph* block_graph_;
858 :
859 : SectionId section_;
860 : BlockAttributes attributes_;
861 :
862 : ReferenceMap references_;
863 : ReferrerSet referrers_;
864 : SourceRanges source_ranges_;
865 : LabelMap labels_;
866 :
867 : // True iff data_ is ours to deallocate with delete [].
868 : // If this is false, data_ must be guaranteed to outlive the block.
869 : bool owns_data_;
870 : // A pointer to the code or data we represent.
871 : const uint8* data_;
872 : // Size of the above.
873 : size_t data_size_;
874 : };
875 :
876 : // A graph address space endows a graph with a non-overlapping ordering
877 : // on blocks, where each block occupies zero or one address ranges in the
878 : // address space. No two blocks may overlap in an address space. Empty blocks
879 : // are not stored in the underlying address-space implementation itself, but do
880 : // have associated addresses and are stored in the block-address map.
881 : class BlockGraph::AddressSpace {
882 : public:
883 : typedef core::AddressSpace<RelativeAddress, BlockGraph::Size, Block*>
884 : AddressSpaceImpl;
885 : typedef AddressSpaceImpl::Range Range;
886 : typedef AddressSpaceImpl::RangeMap RangeMap;
887 : typedef AddressSpaceImpl::RangeMapIter RangeMapIter;
888 : typedef AddressSpaceImpl::RangeMapConstIter RangeMapConstIter;
889 : typedef AddressSpaceImpl::RangeMapIterPair RangeMapIterPair;
890 : typedef AddressSpaceImpl::RangeMapConstIterPair RangeMapConstIterPair;
891 : typedef stdext::hash_map<const Block*, RelativeAddress> BlockAddressMap;
892 :
893 : // Constructs a new empty address space.
894 : // @p start to @p start + @p size on @p graph.
895 : explicit AddressSpace(BlockGraph* graph);
896 :
897 : // Add a block of type @p type and @p size at @p address to our associated
898 : // graph, and return the new block.
899 : // @returns the new block, or NULL if the new block would overlap
900 : // an existing block.
901 : Block* AddBlock(BlockType type,
902 : RelativeAddress addr,
903 : Size size,
904 : const base::StringPiece& name);
905 :
906 : // Resizes a block in the address-space by extending to the right, or
907 : // trimming. Updates the block size but does not udpate its contents. This
908 : // invalidates any RangeMap iterators to the block in question.
909 : // @param block The block whose size is to change.
910 : // @param size The new size of the block. Must be > 0.
911 : // @returns true on success, false if not possible due to a conflict.
912 : bool ResizeBlock(Block* block, size_t size);
913 :
914 : // Merges all blocks that intersect @p range into a single block.
915 : // Moves labels and references from the intersecting blocks to the
916 : // merged block, and changes referring blocks to refer to the new,
917 : // merged block. Removes the original blocks from the BlockGraph.
918 : // @returns the new, merged block if there was at least one intersecting
919 : // block in @p range, or NULL otherwise.
920 : Block* MergeIntersectingBlocks(const Range& range);
921 :
922 : // Insert existing block @p block at @p address.
923 : // @returns true on success, or false if the @p block would overlap
924 : // an existing block.
925 : bool InsertBlock(RelativeAddress addr, Block* block);
926 :
927 : // Returns a pointer to the block containing address, or NULL
928 : // if no block contains address.
929 : Block* GetBlockByAddress(RelativeAddress addr) const;
930 :
931 : // Returns a pointer to the block containing the address range
932 : // [address, address + size), or NULL if no block contains that
933 : // range.
934 : Block* GetContainingBlock(RelativeAddress addr, Size size) const;
935 :
936 : // Finds the first block, if any that intersects
937 : // [@p address, @p address + @p size).
938 : Block* GetFirstIntersectingBlock(RelativeAddress address, Size size);
939 :
940 : // Check whether the address space contains @p block.
941 : // @param block the block in question.
942 : // @returns true if the block is in the address space, false otherwise.
943 : bool ContainsBlock(const Block* block);
944 :
945 : // Locates all blocks that intersect [@p address, @p address + @p size).
946 : // @returns a pair of iterators that iterate over the found blocks.
947 : RangeMapConstIterPair GetIntersectingBlocks(RelativeAddress address,
948 : Size size) const;
949 : RangeMapIterPair GetIntersectingBlocks(RelativeAddress address, Size size);
950 :
951 : // Retrieve the address off @p block.
952 : // @param block the block in question.
953 : // @param addr on success, returns the address of @p block in this
954 : // address space.
955 : // @returns true on success, false if @p block is not in this
956 : // address space.
957 : bool GetAddressOf(const Block* block, RelativeAddress* addr) const;
958 :
959 : // Accessor.
960 E : BlockGraph* graph() { return graph_; }
961 E : const BlockGraph* graph() const { return graph_; }
962 :
963 E : RangeMapConstIter begin() const {
964 E : return address_space_.ranges().begin();
965 E : }
966 :
967 E : RangeMapConstIter end() const {
968 E : return address_space_.ranges().end();
969 E : }
970 :
971 : // @returns the number of blocks in the address-space. This includes empty
972 : // blocks, which won't actually be visited by iteration.
973 E : size_t size() const {
974 : // We use the size of the map of addresses, as zero sized blocks only live
975 : // there.
976 E : return block_addresses_.size();
977 E : }
978 :
979 : // @returns a reference to the underlaying address-space implementation. Only
980 : // non-empty blocks are inserted in the address space.
981 E : const AddressSpaceImpl& address_space_impl() const {
982 E : return address_space_;
983 E : }
984 :
985 : // @returne a reference to the map of blocks addresses, by block pointer.
986 E : const BlockAddressMap& block_addresses() const {
987 E : return block_addresses_;
988 E : }
989 :
990 : protected:
991 : bool InsertImpl(RelativeAddress addr, Block* block);
992 :
993 : AddressSpaceImpl address_space_;
994 : BlockAddressMap block_addresses_;
995 : BlockGraph* graph_;
996 : };
997 :
998 : // Represents a reference from one block to another. References may be offset.
999 : // That is, they may refer to an object at a given location, but actually point
1000 : // to a location that is some fixed distance away from that object. This allows,
1001 : // for example, non-zero based indexing into a table. The object that is
1002 : // intended to be dereferenced is called the 'base' of the offset.
1003 : //
1004 : // BlockGraph references are from a location (offset) in one block, to some
1005 : // location in another block. The referenced block itself plays the role of the
1006 : // 'base' of the reference, with the offset of the reference being stored as
1007 : // an integer from the beginning of the block. However, basic block
1008 : // decomposition requires breaking the block into smaller pieces and thus we
1009 : // need to carry around an explicit base value, indicating which byte in the
1010 : // block is intended to be referenced.
1011 : //
1012 : // A direct reference to a location will have the same value for 'base' and
1013 : // 'offset'.
1014 : //
1015 : // Here is an example:
1016 : //
1017 : // /----------\
1018 : // +---------------------------+
1019 : // O | B | <--- Referenced block
1020 : // +---------------------------+ B = base
1021 : // \-----/ O = offset
1022 : //
1023 : class BlockGraph::Reference {
1024 : public:
1025 : Reference() :
1026 E : type_(RELATIVE_REF), size_(0), referenced_(NULL), offset_(0), base_(0) {
1027 E : }
1028 :
1029 : // @param type type of reference.
1030 : // @param size size of reference.
1031 : // @param referenced the referenced block.
1032 : // @param offset offset from the beginning of the block of the location to be
1033 : // explicitly referred to.
1034 : // @param base offset into the block of the location actually being
1035 : // referenced. This must be strictly within @p referenced.
1036 : Reference(ReferenceType type,
1037 : Size size,
1038 : Block* referenced,
1039 : Offset offset,
1040 : Offset base)
1041 : : type_(type),
1042 : size_(size),
1043 : referenced_(referenced),
1044 : offset_(offset),
1045 E : base_(base) {
1046 E : DCHECK(IsValid());
1047 E : }
1048 :
1049 : // Copy constructor.
1050 : Reference(const Reference& other)
1051 : : type_(other.type_),
1052 : size_(other.size_),
1053 : referenced_(other.referenced_),
1054 : offset_(other.offset_),
1055 E : base_(other.base_) {
1056 E : }
1057 :
1058 : // Accessors.
1059 E : ReferenceType type() const { return type_; }
1060 E : Size size() const { return size_; }
1061 E : Block* referenced() const { return referenced_; }
1062 E : Offset offset() const { return offset_; }
1063 E : Offset base() const { return base_; }
1064 :
1065 : // Determines if this is a direct reference. That is, if the actual location
1066 : // being referenced (offset) and the intended location being referenced (base)
1067 : // are the same.
1068 : //
1069 : // @returns true if the reference is direct, false otherwise.
1070 E : bool IsDirect() const { return base_ == offset_; }
1071 :
1072 : // Determines if this is a valid reference, by imposing size constraints on
1073 : // reference types, and determining if the base address of the reference is
1074 : // strictly contained within the referenced block.
1075 : //
1076 : // @returns true if valid, false otherwise.
1077 : bool IsValid() const;
1078 :
1079 E : bool operator==(const Reference& other) const {
1080 : return type_ == other.type_ &&
1081 : size_ == other.size_ &&
1082 : referenced_ == other.referenced_ &&
1083 : offset_ == other.offset_ &&
1084 E : base_ == other.base_;
1085 E : }
1086 :
1087 : // The maximum size that a reference may have. This needs to be kept in sync
1088 : // with the expectations of IsValid().
1089 : static const size_t kMaximumSize = 4;
1090 :
1091 : // Returns true if the given reference type and size combination is valid.
1092 : static bool IsValidTypeSize(ReferenceType type, Size size);
1093 :
1094 : private:
1095 : // Type of this reference.
1096 : ReferenceType type_;
1097 :
1098 : // Size of this reference.
1099 : // Absolute references are always pointer wide, but PC-relative
1100 : // references can be 1, 2 or 4 bytes wide, which affects their range.
1101 : Size size_;
1102 :
1103 : // The block referenced.
1104 : Block* referenced_;
1105 :
1106 : // Offset into the referenced block.
1107 : Offset offset_;
1108 :
1109 : // The base of the reference, as in offset in the block. This must be a
1110 : // location strictly within the block.
1111 : Offset base_;
1112 : };
1113 :
1114 : // Commonly used container types.
1115 : typedef std::vector<BlockGraph::Block*> BlockVector;
1116 : typedef std::vector<const BlockGraph::Block*> ConstBlockVector;
1117 :
1118 : // A small helper struct for dumping block information to log messages.
1119 : struct BlockInfo {
1120 : enum AddressType {
1121 : kNoAddress,
1122 : kAbsoluteAddress,
1123 : kFileOffsetAddress,
1124 : kRelativeAddress,
1125 : };
1126 :
1127 i : explicit BlockInfo(const BlockGraph::Block* block)
1128 : : block(block), type(kNoAddress) {
1129 i : DCHECK_NE(static_cast<BlockGraph::Block*>(nullptr), block);
1130 i : }
1131 :
1132 : BlockInfo(const BlockGraph::Block* block,
1133 : core::AbsoluteAddress address)
1134 : : block(block), type(kAbsoluteAddress), abs_addr(address) {
1135 : DCHECK_NE(static_cast<BlockGraph::Block*>(nullptr), block);
1136 : }
1137 : BlockInfo(const BlockGraph::Block* block,
1138 : core::FileOffsetAddress address)
1139 : : block(block), type(kFileOffsetAddress), file_addr(address) {
1140 : DCHECK_NE(static_cast<BlockGraph::Block*>(nullptr), block);
1141 : }
1142 i : BlockInfo(const BlockGraph::Block* block,
1143 : core::RelativeAddress address)
1144 : : block(block), type(kRelativeAddress), rel_addr(address) {
1145 i : DCHECK_NE(static_cast<BlockGraph::Block*>(nullptr), block);
1146 i : }
1147 :
1148 : const BlockGraph::Block* block;
1149 : AddressType type;
1150 :
1151 : // Ideally these would be in a union but because they have non-trivial
1152 : // constructors they are not allowed.
1153 : core::AbsoluteAddress abs_addr;
1154 : core::FileOffsetAddress file_addr;
1155 : core::RelativeAddress rel_addr;
1156 :
1157 : private:
1158 : DISALLOW_COPY_AND_ASSIGN(BlockInfo);
1159 : };
1160 :
1161 : } // namespace block_graph
1162 :
1163 : // Pretty prints a BlockInfo to an ostream. This has to be outside of any
1164 : // namespaces so that operator<< is found properly.
1165 : std::ostream& operator<<(std::ostream& os, const block_graph::BlockInfo& bi);
1166 :
1167 : #endif // SYZYGY_BLOCK_GRAPH_BLOCK_GRAPH_H_
|