1 /* Copyright (C) 2007-2010 The Android Open Source Project
2 **
3 ** This software is licensed under the terms of the GNU General Public
4 ** License version 2, as published by the Free Software Foundation, and
5 ** may be copied, distributed, and modified under those terms.
6 **
7 ** This program is distributed in the hope that it will be useful,
8 ** but WITHOUT ANY WARRANTY; without even the implied warranty of
9 ** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10 ** GNU General Public License for more details.
11 */
12 
13 /*
14  * Contains declaration of a class DwarfCU, that encapsulates a compilation
15  * unit in the .debug_info section of the mapped ELF file.
16  */
17 
18 #ifndef ELFF_DWARF_CU_H_
19 #define ELFF_DWARF_CU_H_
20 
21 #include "dwarf_defs.h"
22 #include "dwarf_die.h"
23 
24 /* Address information descriptor. */
25 typedef struct Dwarf_AddressInfo {
26   /* Routine DIE containing the address. */
27   const DIEObject*  die_obj;
28 
29   /* Source file name for the address. */
30   const char*       file_name;
31 
32   /* Source file directory path for the address. */
33   const char*       dir_name;
34 
35   /* Source file line number for the address. */
36   Elf_Word          line_number;
37 } Dwarf_AddressInfo;
38 
39 /* STMTL header cached by compilation unit. This header is contained in
40  * the .debug_line section of the ELF file. */
41 typedef struct Dwarf_STMTL_Hdr {
42   /* The size in bytes of the line number information for this compilation
43    * unit, not including the unit_length field itself. */
44   Elf_Xword                   unit_length;
45 
46   /* A version number. This number is specific to the line number information
47    * and is independent of the DWARF version number. */
48   Elf_Half                    version;
49 
50   /* The number of bytes following the header_length field to the beginning of
51    * the first byte of the line number program itself. In the 32-bit DWARF
52    * format, this is a 4-byte unsigned length; in the 64-bit DWARF format,
53    * this field is an 8-byte unsigned length. */
54   Elf_Xword                   header_length;
55 
56   /* The size in bytes of the smallest target machine instruction. Line number
57    * program opcodes that alter the address register first multiply their
58    * operands by this value. */
59   Elf_Byte                    min_instruction_len;
60 
61   /* The initial value of the is_stmt register. */
62   Elf_Byte                    default_is_stmt;
63 
64   /* This parameter affects the meaning of the special opcodes. */
65   Elf_Sbyte                   line_base;
66 
67   /* This parameter affects the meaning of the special opcodes. */
68   Elf_Byte                    line_range;
69 
70   /* The number assigned to the first special opcode. */
71   Elf_Byte                    opcode_base;
72 
73   /* Points to standard_opcode_lengths array in the actual STMTL header in
74    * the mapped .debug_line section. */
75   const Elf_Byte*             standard_opcode_lengths;
76 
77   /* Pointer to the beginning of the list of include directories in the mapped
78    * .debug_line section. */
79   const char*                 include_directories;
80 
81   /* Number of include directories in the list that begins with
82    * include_directories. */
83   Elf_Word                    inc_dir_num;
84 
85   /* Pointer to the beginning of the list of file information in the mapped
86    * .debug_line section. Each entry in this list begins with zero-terminated
87    * file name, followed by ULEB128 encoding directory index for the file,
88    * followed by ULEB128 encoding last modification time, followed by ULEB128
89    * encoding length of file in bytes. */
90   const Dwarf_STMTL_FileDesc* file_infos;
91 
92   /* Start of the "Line Number Program" in the mapped .debug_line section. */
93   const Elf_Byte*             start;
94 
95   /* End of the "Line Number Program" in the mapped .debug_line section. */
96   const Elf_Byte*             end;
97 } Dwarf_STMTL_Hdr;
98 
99 /* Encapsulates architecture-independent functionality of a
100  * compilation unit.
101  */
102 class DwarfCU : public DwarfAllocBase {
103 friend class ElfFile;
104  public:
105   /* Constructs DwarfCU instance.
106    * Param:
107    *  elf - Instance of ElfFile containing this compilation unit.
108    */
109   explicit DwarfCU(ElfFile* elf);
110 
111   /* Destructs DwarfCU instance. */
112   virtual ~DwarfCU();
113 
114   /* Creates DwarfCUImpl instance, depending on DWARF format.
115    * Param:
116    *  elf - Instance of ElfFile containing this compilation unit.
117    *  hdr - Pointer to compilation unit header inside mapped .debug_info
118    *    section of the ELF file. Actual data addressed by this pointer
119    *    must be Dwarf32_CUHdr for 32 bit DWARFs, or Dwarf64_CUHdr for
120    *    64 bit DWARFs.
121    * Return:
122    *  Created DwarfCUImpl instance (typecasted back to DwarfCU) on success,
123    *  or NULL on failure.
124    */
125   static DwarfCU* create_instance(ElfFile* elf, const void* hdr);
126 
127   /* Process a DIE attribute.
128    * Param:
129    *  attr - Attribute list inside the mapped .debug_info section of the ELF
130    *    file.
131    *  form - Attribute's form, definig representation of attribute value in the
132    *    mapped .debug_info section of the ELF file.
133    *  attr_value - Upon return contains attribute value.
134    * Return:
135    *  Pointer to the next DIE attribute inside the mapped .debug_info section
136    *  of the ELF file.
137    */
138   const Elf_Byte* process_attrib(const Elf_Byte* attr,
139                                  Dwarf_Form form,
140                                  Dwarf_Value* attr_value) const;
141 
142   /* Dumps this compilation unit to the stdout. */
143   void dump() const;
144 
145   /* Gets instance of ElfFile containing this compilation unit. */
elf_file()146   ElfFile* elf_file() const {
147     return elf_file_;
148   }
149 
150   /* Gets DIE object for this CU. */
cu_die()151   DIEObject* cu_die() const {
152     return cu_die_;
153   }
154 
155   /* Gets byte size of the pointer type for this compilation unit. */
addr_sizeof()156   Elf_Byte addr_sizeof() const {
157     return addr_sizeof_;
158   }
159 
160   /* Gets full path to the compilation directory (DW_AT_comp_dir attribute).
161    * Return:
162    *  Full path to the compilation directory (DW_AT_comp_dir attribute),
163    *  or NULL, if that attribute was missing in CU's attribute list.
164    */
comp_dir_path()165   const char* comp_dir_path() const {
166     DIEAttrib attr;
167     return cu_die()->get_attrib(DW_AT_comp_dir, &attr) ? attr.value()->str :
168                                                          NULL;
169   }
170 
171   /* Gets relative (from the compilation directory) path to the compiled file.
172    * (DW_AT_name attribute).
173    * Return:
174    *  Relative path to the compiled file (DW_AT_name attribute), or NULL, if
175    *  that attribute was missing in CU's attribute list.
176    */
rel_cu_path()177   const char* rel_cu_path() const {
178     DIEAttrib attr;
179     return cu_die()->get_attrib(DW_AT_name, &attr) ? attr.value()->str :
180                                                      NULL;
181   }
182 
183   /* Gets next compilation unit in the list. NULL indicates the last CU. */
prev_cu()184   DwarfCU* prev_cu() const {
185     return prev_cu_;
186   }
187 
188   /* Links this CU to the list of prevoiusly discovered CUs. */
set_prev_cu(DwarfCU * prev)189   void set_prev_cu(DwarfCU* prev) {
190     prev_cu_ = prev;
191   }
192 
193   /* Checks if DWARF version for this CU is higher than 2. */
is_DWARF3_or_higher()194   bool is_DWARF3_or_higher() const {
195     return version_ >= 3;
196   }
197 
198   /* Gets DIE abbreviation for given abbreviation number.
199    * See DwarfAbbrDieArray::get() */
get_die_abbr(Dwarf_AbbrNum num)200   const Dwarf_Abbr_DIE* get_die_abbr(Dwarf_AbbrNum num) const {
201     return abbrs_.get(num);
202   }
203 
204   /* Gets DIE object containing given address.
205    * DIE address ranges may overlap (for instance, address range for an inlined
206    * routine will be contained within the address range of a routine where it
207    * was inlined). This method will return a DIE object that is a "leaf" in
208    * that inlining tree. I.e the returned DIE object represents the last DIE
209    * in the branch of all DIEs containing given address.
210    * Param:
211    *  address - Address to get a DIE for. NOTE: for the sake of simplicity we
212    *    explicitly use 64-bit type for an address.
213    * Return:
214    *  Leaf DIE containing given address, or NULL if this CU doesn't contain
215    *  the given address.
216    */
get_leaf_die_for_address(Elf_Xword address)217   DIEObject* get_leaf_die_for_address(Elf_Xword address) const {
218     return cu_die_->get_leaf_for_address(address);
219   }
220 
221   /* Checks if this CU contains 64, or 32-bit addresses. */
is_CU_address_64()222   bool is_CU_address_64() const {
223     return addr_sizeof_ == 8;
224   }
is_CU_address_32()225   bool is_CU_address_32() const {
226     return addr_sizeof_ == 4;
227   }
228 
229 //=============================================================================
230 // DWARF format dependent methods
231 //=============================================================================
232 
233  public:
234   /* Parses this compilation unit in .debug_info section, collecting children
235    * DIEs of this compilation unit.
236    * Param:
237    *  parse_context - Parsing context that lists tags for DIEs that should be
238    *    collected during parsing. NULL passed in this parameter indicates DIEs
239    *    for all tags should be collected.
240    *  next_cu_die - Upon successful return contains pointer to the next
241    *    compilation unit descriptor inside mapped .debug_info section of
242    *    the ELF file.
243    * Return:
244    *  true on success, false on failure.
245    */
246   virtual bool parse(const DwarfParseContext* parse_context,
247                      const void** next_cu_die) = 0;
248 
249   /* Gets a DIE object referenced by an offset from the beginning of
250    * this CU in the mapped .debug_info section.
251    */
252   virtual DIEObject* get_referenced_die_object(Elf_Word ref) const = 0;
253 
254   /* Gets a reference to a DIE object (offset of the DIE from the
255    * beginning of this CU in the mapped .debug_info section.
256    */
257   virtual Elf_Word get_die_reference(const Dwarf_DIE* die) const = 0;
258 
259   /* Gets PC address information.
260    * Param:
261    *  address - PC address to get information for.
262    *  info - Upon success contains information about routine that belongs to
263    *    this compilation unit, containing the given address.
264    * Return:
265    *  true on success, or false if this CU doesn't contain the given address.
266    */
267   virtual bool get_pc_address_file_info(Elf_Xword address,
268                                         Dwarf_AddressInfo* info) = 0;
269 
270   /* Gets file descriptor in the mapped .debug_line section of ELF file for a
271    * given index in the file descriptor list.
272    * Param:
273    *  index - 1-based index of file descriptor in the file descriptor list.
274    * Return:
275    *  File descriptor for the given index, or NULL if index was too big.
276    *  NOTE: pointer returned from this method addressed mapped section of
277    *  ELF file.
278    */
279   virtual const Dwarf_STMTL_FileDesc* get_stmt_file_info(Elf_Word index) = 0;
280 
281   /* Gets directory name by an index in the mapped .debug_line section of
282    * ELF file.
283    * Param:
284    *  dir_index - Index of the directory in the file descriptor list. If this
285    *    parameter is zero, compilation directory (DW_AT_comp_dir) for this CU
286    *    will be returned.
287    * Return:
288    *  Directory name for the given index, or NULL if index was too big.
289    *  NOTE: pointer returned from this method addressed mapped section of
290    *  ELF file.
291    */
292   virtual const char* get_stmt_dir_name(Elf_Word dir_index) = 0;
293 
294  protected:
295   /* DIE abbreviation descriptors, cached for this compilation unit. */
296   DwarfAbbrDieArray   abbrs_;
297 
298   /* Instance of an ELF file that contains this compilation unit. */
299   ElfFile*            elf_file_;
300 
301   /* DIE object for this CU. */
302   DIEObject*          cu_die_;
303 
304   /* Next compilation unit in the list (previous in the order they've been
305    * discovered during ELF file parsing).
306    */
307   DwarfCU*            prev_cu_;
308 
309   /* DWARF version for this CU. */
310   Elf_Half            version_;
311 
312   /* Byte size of the pointer type for this compilation unit. */
313   Elf_Byte            addr_sizeof_;
314 };
315 
316 /* Encapsulates architecture-dependent functionality of a compilation unit.
317  * Template param:
318  *  Dwarf_CUHdr - type compilation unit header in the mapped .debug_info
319  *    section of ELF file. Must be:
320  *    - Dwarf32_CUHdr for 32-bit DWARF, or
321  *    - Dwarf64_CUHdr for 64-bit DWARF.
322  *  Dwarf_Off - type for an offset field in DWARF data format. Must be:
323  *    - Dwarf32_Off for 32-bit DWARF, or
324  *    - Dwarf64_Off for 64-bit DWARF.
325  */
326 template <typename Dwarf_CUHdr, typename Dwarf_Off>
327 class DwarfCUImpl : public DwarfCU {
328  public:
329   /* Constructs DwarfCU instance.
330    * Param:
331    *  elf - Instance of ElfFile containing this compilation unit.
332    *  hdr - Pointer to compilation unit header inside mapped .debug_info
333    *    section of the ELF file.
334    */
335   DwarfCUImpl(ElfFile* elf, const Dwarf_CUHdr* hdr);
336 
337   /* Destructs DwarfCU instance. */
~DwarfCUImpl()338   ~DwarfCUImpl() {
339   }
340 
341   /* Parses this compilation unit in .debug_info section, collecting children
342    * DIEs of this compilation unit. This is an implementation of DwarfCU's
343    * abstract metod.
344    * See DwarfCU::parse().
345    */
346   bool parse(const DwarfParseContext* parse_context,
347              const void** next_cu_die);
348 
349   /* Gets PC address information.
350    * This is an implementation of DwarfCU's abstract metod.
351    * See DwarfCU::get_pc_address_file_info().
352    */
353   bool get_pc_address_file_info(Elf_Xword address, Dwarf_AddressInfo* info);
354 
355   /* Gets file descriptor in the mapped .debug_line section of ELF file for a
356    * given index in the file descriptor list.
357    * This is an implementation of DwarfCU's abstract metod.
358    * See DwarfCU::get_stmt_file_info().
359    */
360   const Dwarf_STMTL_FileDesc* get_stmt_file_info(Elf_Word index);
361 
362   /* Gets directory name by an index in the mapped .debug_line section of
363    * ELF file.
364    * This is an implementation of DwarfCU's abstract metod.
365    * See DwarfCU::get_stmt_dir_name().
366    */
367   const char* get_stmt_dir_name(Elf_Word dir_index);
368 
369   /* Gets a DIE object referenced by an offset from the beginning of
370    * this CU. This is an implementation of DwarfCU's abstract metod.
371    */
get_referenced_die_object(Elf_Word ref)372   DIEObject* get_referenced_die_object(Elf_Word ref) const {
373     const Dwarf_DIE* die = get_referenced_die(ref);
374     return cu_die_->find_die_object(die);
375   }
376 
377   /* Gets a reference to a DIE object (offset of the DIE from the
378    * beginning of this CU in the mapped .debug_info section.
379    * This is an implementation of DwarfCU's abstract metod.
380    */
get_die_reference(const Dwarf_DIE * die)381   Elf_Word get_die_reference(const Dwarf_DIE* die) const {
382     return static_cast<Elf_Word>(diff_ptr(cu_header_, die));
383   }
384 
385  protected:
386   /* Process a child DIE (and all its children) in this compilation unit.
387    * Param:
388    *  parse_context - See DwarfCU::parse().
389    *  die - DIE descriptor of the child to process in this method.
390    *  parent_obj - Parent object of the child to process in this method.
391    *    NOTE: this parameter can be NULL only for a DIE that represents this
392    *    compilation unit itself.
393    * Return:
394    *  Pointer to the end of child's attribute list in the mapped .debug_info
395    *  section on success, or NULL on failure. Usually, pointer returned from
396    *  this method is simply discarded, since parent calculates address of the
397    *  next sibling's DIE based on DW_AT_sibling attribute of the DIE preceding
398    *  child's DIE.
399    */
400   const Elf_Byte* process_DIE(const DwarfParseContext* parse_context,
401                               const Dwarf_DIE* die,
402                               DIEObject* parent_obj);
403 
404   /* Creates a DIE object for the given DIE.
405    * Param:
406    *  parse_context See DwarfCU::parse().
407    *  die - DIE to create an object for.
408    *  parent - Parent DIE object for the one that's being created in this
409    *    method.
410    *  tag - Tag of the DIE object that's being created in this method.
411    * Return:
412    *  Created DIE object. This method may returns NULL in two cases:
413    *    - We're not interested in this DIE (decided by looking at 'tag'
414    *      parameter. In this case errno should be set to zero.
415    *    - Memory allocation has failed. In this case errno should be
416    *      set to ENOMEM.
417    */
418   DIEObject* create_die_object(const DwarfParseContext* parse_context,
419                                const Dwarf_DIE* die,
420                                DIEObject* parent,
421                                Dwarf_Tag tag);
422 
423   /* Initializes (caches) STMT lines header for this CU. */
424   bool init_stmtl();
425 
426   /* Saves current source file information, collected in the state machine by
427    * the "Line Number Program".
428    * Param:
429    *  state - State machine collected "Line Number Program" results.
430    *  info - Upon success contains source file information, copied over from
431    *    the state machine.
432    * Return:
433    *  true on success, or false on failure.
434    */
435   bool set_source_info(const DwarfStateMachine* state,
436                        Dwarf_AddressInfo* info);
437 
438   /* Gets pointer to the DIE descriptor for this CU. */
get_DIE()439   const Dwarf_DIE* get_DIE() const {
440     /* CU's DIE descriptor immediately follows CU header. */
441     return INC_CPTR_T(Dwarf_DIE, cu_header_, sizeof(Dwarf_CUHdr));
442   }
443 
444   /* Caches STMTL header from .debug_line section to stmtl_header_.
445    * Template param:
446    *  Dwarf_STMTL_Hdr - Dwarf_STMTL_Hdr32, or Dwarf_STMTL_Hdr64, depending
447    *    on the header type.
448    * Param:
449    *  stmtl_hdr - STMTL header in the mapped .debug_line section to cache.
450    */
451   template <typename Dwarf_STMTL_Hdr>
cache_stmtl(const Dwarf_STMTL_Hdr * stmtl_hdr)452   void cache_stmtl(const Dwarf_STMTL_Hdr* stmtl_hdr) {
453     stmtl_header_.unit_length = elf_file()->pull_val(stmtl_hdr->unit_length.size);
454     stmtl_header_.version = elf_file()->pull_val(stmtl_hdr->version);
455     stmtl_header_.header_length = elf_file()->pull_val(stmtl_hdr->header_length);
456     stmtl_header_.min_instruction_len = stmtl_hdr->min_instruction_len;
457     stmtl_header_.default_is_stmt = stmtl_hdr->default_is_stmt;
458     stmtl_header_.line_base = stmtl_hdr->line_base;
459     stmtl_header_.line_range = stmtl_hdr->line_range;
460     stmtl_header_.opcode_base = stmtl_hdr->opcode_base;
461     stmtl_header_.standard_opcode_lengths = &stmtl_hdr->standard_opcode_lengths;
462     stmtl_header_.start = INC_CPTR_T(Elf_Byte, &stmtl_hdr->min_instruction_len,
463                                      stmtl_header_.header_length);
464     stmtl_header_.end = INC_CPTR_T(Elf_Byte, &stmtl_hdr->version,
465                                    stmtl_header_.unit_length);
466     stmtl_header_.include_directories =
467         INC_CPTR_T(char, stmtl_header_.standard_opcode_lengths,
468                    stmtl_header_.opcode_base - 1);
469     const char* dir = stmtl_header_.include_directories;
470     while (*dir != '\0') {
471       dir += strlen(dir) + 1;
472       stmtl_header_.inc_dir_num++;
473     }
474     stmtl_header_.file_infos = INC_CPTR_T(Dwarf_STMTL_FileDesc, dir, 1);
475   }
476 
477   /* Gets a DIE referenced by an offset from the beginning of this CU
478    * in the mapped .debug_info section.
479    */
get_referenced_die(Elf_Word ref)480   const Dwarf_DIE* get_referenced_die(Elf_Word ref) const {
481     return INC_CPTR_T(Dwarf_DIE, cu_header_, ref);
482   }
483 
484   /* Checks if pointer to the DIE attribute is contained within the CU's area
485    * of the mapped .debug_info section.
486    * Param:
487    *  ptr - Pointer to the DIE attribute to check.
488    * Return:
489    *  true, if pointer to the DIE attribute is contained within the CU's area
490    *  of the mapped .debug_info section, or false if attribute pointer goes
491    *  beyond CU's area of the mapped .debug_info section.
492    */
is_attrib_ptr_valid(const void * ptr)493   bool is_attrib_ptr_valid(const void* ptr) const {
494     return diff_ptr(cu_header_, ptr) < cu_size_;
495   }
496 
497  protected:
498   /* Pointer to this compilation unit header inside the mapped .debug_info
499    * section of the ELF file.
500    */
501   const Dwarf_CUHdr*          cu_header_;
502 
503   /* Size of this compilation unit area in the mapped .debug_info section.
504    * This value has been cached off the CU header in order to avoid
505    * endianness conversions.
506    */
507   Dwarf_Off                   cu_size_;
508 
509   /* STMT lines header, cached off mapped .debug_line section. */
510   Dwarf_STMTL_Hdr             stmtl_header_;
511 };
512 
513 #endif  // ELFF_DWARF_CU_H_
514