Fix undefined behavior, don't pass NULL to fwrite
[deliverable/binutils-gdb.git] / gdb / dwarf-index-write.c
1 /* DWARF index writing support for GDB.
2
3 Copyright (C) 1994-2018 Free Software Foundation, Inc.
4
5 This file is part of GDB.
6
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
11
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
16
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19
20 #include "defs.h"
21
22 #include "addrmap.h"
23 #include "cli/cli-decode.h"
24 #include "common/byte-vector.h"
25 #include "common/filestuff.h"
26 #include "common/gdb_unlinker.h"
27 #include "common/scoped_fd.h"
28 #include "complaints.h"
29 #include "dwarf-index-common.h"
30 #include "dwarf2.h"
31 #include "dwarf2read.h"
32 #include "gdb/gdb-index.h"
33 #include "gdbcmd.h"
34 #include "objfiles.h"
35 #include "psympriv.h"
36
37 #include <algorithm>
38 #include <cmath>
39 #include <set>
40 #include <unordered_map>
41 #include <unordered_set>
42
43 /* Ensure only legit values are used. */
44 #define DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE(cu_index, value) \
45 do { \
46 gdb_assert ((unsigned int) (value) <= 1); \
47 GDB_INDEX_SYMBOL_STATIC_SET_VALUE((cu_index), (value)); \
48 } while (0)
49
50 /* Ensure only legit values are used. */
51 #define DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE(cu_index, value) \
52 do { \
53 gdb_assert ((value) >= GDB_INDEX_SYMBOL_KIND_TYPE \
54 && (value) <= GDB_INDEX_SYMBOL_KIND_OTHER); \
55 GDB_INDEX_SYMBOL_KIND_SET_VALUE((cu_index), (value)); \
56 } while (0)
57
58 /* Ensure we don't use more than the alloted nuber of bits for the CU. */
59 #define DW2_GDB_INDEX_CU_SET_VALUE(cu_index, value) \
60 do { \
61 gdb_assert (((value) & ~GDB_INDEX_CU_MASK) == 0); \
62 GDB_INDEX_CU_SET_VALUE((cu_index), (value)); \
63 } while (0)
64
65 /* The "save gdb-index" command. */
66
67 /* Write SIZE bytes from the buffer pointed to by DATA to FILE, with
68 error checking. */
69
70 static void
71 file_write (FILE *file, const void *data, size_t size)
72 {
73 if (fwrite (data, 1, size, file) != size)
74 error (_("couldn't data write to file"));
75 }
76
77 /* Write the contents of VEC to FILE, with error checking. */
78
79 template<typename Elem, typename Alloc>
80 static void
81 file_write (FILE *file, const std::vector<Elem, Alloc> &vec)
82 {
83 if (!vec.empty ())
84 file_write (file, vec.data (), vec.size () * sizeof (vec[0]));
85 }
86
87 /* In-memory buffer to prepare data to be written later to a file. */
88 class data_buf
89 {
90 public:
91 /* Copy DATA to the end of the buffer. */
92 template<typename T>
93 void append_data (const T &data)
94 {
95 std::copy (reinterpret_cast<const gdb_byte *> (&data),
96 reinterpret_cast<const gdb_byte *> (&data + 1),
97 grow (sizeof (data)));
98 }
99
100 /* Copy CSTR (a zero-terminated string) to the end of buffer. The
101 terminating zero is appended too. */
102 void append_cstr0 (const char *cstr)
103 {
104 const size_t size = strlen (cstr) + 1;
105 std::copy (cstr, cstr + size, grow (size));
106 }
107
108 /* Store INPUT as ULEB128 to the end of buffer. */
109 void append_unsigned_leb128 (ULONGEST input)
110 {
111 for (;;)
112 {
113 gdb_byte output = input & 0x7f;
114 input >>= 7;
115 if (input)
116 output |= 0x80;
117 append_data (output);
118 if (input == 0)
119 break;
120 }
121 }
122
123 /* Accept a host-format integer in VAL and append it to the buffer
124 as a target-format integer which is LEN bytes long. */
125 void append_uint (size_t len, bfd_endian byte_order, ULONGEST val)
126 {
127 ::store_unsigned_integer (grow (len), len, byte_order, val);
128 }
129
130 /* Return the size of the buffer. */
131 size_t size () const
132 {
133 return m_vec.size ();
134 }
135
136 /* Return true iff the buffer is empty. */
137 bool empty () const
138 {
139 return m_vec.empty ();
140 }
141
142 /* Write the buffer to FILE. */
143 void file_write (FILE *file) const
144 {
145 ::file_write (file, m_vec);
146 }
147
148 private:
149 /* Grow SIZE bytes at the end of the buffer. Returns a pointer to
150 the start of the new block. */
151 gdb_byte *grow (size_t size)
152 {
153 m_vec.resize (m_vec.size () + size);
154 return &*(m_vec.end () - size);
155 }
156
157 gdb::byte_vector m_vec;
158 };
159
160 /* An entry in the symbol table. */
161 struct symtab_index_entry
162 {
163 /* The name of the symbol. */
164 const char *name;
165 /* The offset of the name in the constant pool. */
166 offset_type index_offset;
167 /* A sorted vector of the indices of all the CUs that hold an object
168 of this name. */
169 std::vector<offset_type> cu_indices;
170 };
171
172 /* The symbol table. This is a power-of-2-sized hash table. */
173 struct mapped_symtab
174 {
175 mapped_symtab ()
176 {
177 data.resize (1024);
178 }
179
180 offset_type n_elements = 0;
181 std::vector<symtab_index_entry> data;
182 };
183
184 /* Find a slot in SYMTAB for the symbol NAME. Returns a reference to
185 the slot.
186
187 Function is used only during write_hash_table so no index format backward
188 compatibility is needed. */
189
190 static symtab_index_entry &
191 find_slot (struct mapped_symtab *symtab, const char *name)
192 {
193 offset_type index, step, hash = mapped_index_string_hash (INT_MAX, name);
194
195 index = hash & (symtab->data.size () - 1);
196 step = ((hash * 17) & (symtab->data.size () - 1)) | 1;
197
198 for (;;)
199 {
200 if (symtab->data[index].name == NULL
201 || strcmp (name, symtab->data[index].name) == 0)
202 return symtab->data[index];
203 index = (index + step) & (symtab->data.size () - 1);
204 }
205 }
206
207 /* Expand SYMTAB's hash table. */
208
209 static void
210 hash_expand (struct mapped_symtab *symtab)
211 {
212 auto old_entries = std::move (symtab->data);
213
214 symtab->data.clear ();
215 symtab->data.resize (old_entries.size () * 2);
216
217 for (auto &it : old_entries)
218 if (it.name != NULL)
219 {
220 auto &ref = find_slot (symtab, it.name);
221 ref = std::move (it);
222 }
223 }
224
225 /* Add an entry to SYMTAB. NAME is the name of the symbol.
226 CU_INDEX is the index of the CU in which the symbol appears.
227 IS_STATIC is one if the symbol is static, otherwise zero (global). */
228
229 static void
230 add_index_entry (struct mapped_symtab *symtab, const char *name,
231 int is_static, gdb_index_symbol_kind kind,
232 offset_type cu_index)
233 {
234 offset_type cu_index_and_attrs;
235
236 ++symtab->n_elements;
237 if (4 * symtab->n_elements / 3 >= symtab->data.size ())
238 hash_expand (symtab);
239
240 symtab_index_entry &slot = find_slot (symtab, name);
241 if (slot.name == NULL)
242 {
243 slot.name = name;
244 /* index_offset is set later. */
245 }
246
247 cu_index_and_attrs = 0;
248 DW2_GDB_INDEX_CU_SET_VALUE (cu_index_and_attrs, cu_index);
249 DW2_GDB_INDEX_SYMBOL_STATIC_SET_VALUE (cu_index_and_attrs, is_static);
250 DW2_GDB_INDEX_SYMBOL_KIND_SET_VALUE (cu_index_and_attrs, kind);
251
252 /* We don't want to record an index value twice as we want to avoid the
253 duplication.
254 We process all global symbols and then all static symbols
255 (which would allow us to avoid the duplication by only having to check
256 the last entry pushed), but a symbol could have multiple kinds in one CU.
257 To keep things simple we don't worry about the duplication here and
258 sort and uniqufy the list after we've processed all symbols. */
259 slot.cu_indices.push_back (cu_index_and_attrs);
260 }
261
262 /* Sort and remove duplicates of all symbols' cu_indices lists. */
263
264 static void
265 uniquify_cu_indices (struct mapped_symtab *symtab)
266 {
267 for (auto &entry : symtab->data)
268 {
269 if (entry.name != NULL && !entry.cu_indices.empty ())
270 {
271 auto &cu_indices = entry.cu_indices;
272 std::sort (cu_indices.begin (), cu_indices.end ());
273 auto from = std::unique (cu_indices.begin (), cu_indices.end ());
274 cu_indices.erase (from, cu_indices.end ());
275 }
276 }
277 }
278
279 /* A form of 'const char *' suitable for container keys. Only the
280 pointer is stored. The strings themselves are compared, not the
281 pointers. */
282 class c_str_view
283 {
284 public:
285 c_str_view (const char *cstr)
286 : m_cstr (cstr)
287 {}
288
289 bool operator== (const c_str_view &other) const
290 {
291 return strcmp (m_cstr, other.m_cstr) == 0;
292 }
293
294 /* Return the underlying C string. Note, the returned string is
295 only a reference with lifetime of this object. */
296 const char *c_str () const
297 {
298 return m_cstr;
299 }
300
301 private:
302 friend class c_str_view_hasher;
303 const char *const m_cstr;
304 };
305
306 /* A std::unordered_map::hasher for c_str_view that uses the right
307 hash function for strings in a mapped index. */
308 class c_str_view_hasher
309 {
310 public:
311 size_t operator () (const c_str_view &x) const
312 {
313 return mapped_index_string_hash (INT_MAX, x.m_cstr);
314 }
315 };
316
317 /* A std::unordered_map::hasher for std::vector<>. */
318 template<typename T>
319 class vector_hasher
320 {
321 public:
322 size_t operator () (const std::vector<T> &key) const
323 {
324 return iterative_hash (key.data (),
325 sizeof (key.front ()) * key.size (), 0);
326 }
327 };
328
329 /* Write the mapped hash table SYMTAB to the data buffer OUTPUT, with
330 constant pool entries going into the data buffer CPOOL. */
331
332 static void
333 write_hash_table (mapped_symtab *symtab, data_buf &output, data_buf &cpool)
334 {
335 {
336 /* Elements are sorted vectors of the indices of all the CUs that
337 hold an object of this name. */
338 std::unordered_map<std::vector<offset_type>, offset_type,
339 vector_hasher<offset_type>>
340 symbol_hash_table;
341
342 /* We add all the index vectors to the constant pool first, to
343 ensure alignment is ok. */
344 for (symtab_index_entry &entry : symtab->data)
345 {
346 if (entry.name == NULL)
347 continue;
348 gdb_assert (entry.index_offset == 0);
349
350 /* Finding before inserting is faster than always trying to
351 insert, because inserting always allocates a node, does the
352 lookup, and then destroys the new node if another node
353 already had the same key. C++17 try_emplace will avoid
354 this. */
355 const auto found
356 = symbol_hash_table.find (entry.cu_indices);
357 if (found != symbol_hash_table.end ())
358 {
359 entry.index_offset = found->second;
360 continue;
361 }
362
363 symbol_hash_table.emplace (entry.cu_indices, cpool.size ());
364 entry.index_offset = cpool.size ();
365 cpool.append_data (MAYBE_SWAP (entry.cu_indices.size ()));
366 for (const auto index : entry.cu_indices)
367 cpool.append_data (MAYBE_SWAP (index));
368 }
369 }
370
371 /* Now write out the hash table. */
372 std::unordered_map<c_str_view, offset_type, c_str_view_hasher> str_table;
373 for (const auto &entry : symtab->data)
374 {
375 offset_type str_off, vec_off;
376
377 if (entry.name != NULL)
378 {
379 const auto insertpair = str_table.emplace (entry.name, cpool.size ());
380 if (insertpair.second)
381 cpool.append_cstr0 (entry.name);
382 str_off = insertpair.first->second;
383 vec_off = entry.index_offset;
384 }
385 else
386 {
387 /* While 0 is a valid constant pool index, it is not valid
388 to have 0 for both offsets. */
389 str_off = 0;
390 vec_off = 0;
391 }
392
393 output.append_data (MAYBE_SWAP (str_off));
394 output.append_data (MAYBE_SWAP (vec_off));
395 }
396 }
397
398 typedef std::unordered_map<partial_symtab *, unsigned int> psym_index_map;
399
400 /* Helper struct for building the address table. */
401 struct addrmap_index_data
402 {
403 addrmap_index_data (data_buf &addr_vec_, psym_index_map &cu_index_htab_)
404 : addr_vec (addr_vec_), cu_index_htab (cu_index_htab_)
405 {}
406
407 struct objfile *objfile;
408 data_buf &addr_vec;
409 psym_index_map &cu_index_htab;
410
411 /* Non-zero if the previous_* fields are valid.
412 We can't write an entry until we see the next entry (since it is only then
413 that we know the end of the entry). */
414 int previous_valid;
415 /* Index of the CU in the table of all CUs in the index file. */
416 unsigned int previous_cu_index;
417 /* Start address of the CU. */
418 CORE_ADDR previous_cu_start;
419 };
420
421 /* Write an address entry to ADDR_VEC. */
422
423 static void
424 add_address_entry (struct objfile *objfile, data_buf &addr_vec,
425 CORE_ADDR start, CORE_ADDR end, unsigned int cu_index)
426 {
427 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, start);
428 addr_vec.append_uint (8, BFD_ENDIAN_LITTLE, end);
429 addr_vec.append_data (MAYBE_SWAP (cu_index));
430 }
431
432 /* Worker function for traversing an addrmap to build the address table. */
433
434 static int
435 add_address_entry_worker (void *datap, CORE_ADDR start_addr, void *obj)
436 {
437 struct addrmap_index_data *data = (struct addrmap_index_data *) datap;
438 struct partial_symtab *pst = (struct partial_symtab *) obj;
439
440 if (data->previous_valid)
441 add_address_entry (data->objfile, data->addr_vec,
442 data->previous_cu_start, start_addr,
443 data->previous_cu_index);
444
445 data->previous_cu_start = start_addr;
446 if (pst != NULL)
447 {
448 const auto it = data->cu_index_htab.find (pst);
449 gdb_assert (it != data->cu_index_htab.cend ());
450 data->previous_cu_index = it->second;
451 data->previous_valid = 1;
452 }
453 else
454 data->previous_valid = 0;
455
456 return 0;
457 }
458
459 /* Write OBJFILE's address map to ADDR_VEC.
460 CU_INDEX_HTAB is used to map addrmap entries to their CU indices
461 in the index file. */
462
463 static void
464 write_address_map (struct objfile *objfile, data_buf &addr_vec,
465 psym_index_map &cu_index_htab)
466 {
467 struct addrmap_index_data addrmap_index_data (addr_vec, cu_index_htab);
468
469 /* When writing the address table, we have to cope with the fact that
470 the addrmap iterator only provides the start of a region; we have to
471 wait until the next invocation to get the start of the next region. */
472
473 addrmap_index_data.objfile = objfile;
474 addrmap_index_data.previous_valid = 0;
475
476 addrmap_foreach (objfile->psymtabs_addrmap, add_address_entry_worker,
477 &addrmap_index_data);
478
479 /* It's highly unlikely the last entry (end address = 0xff...ff)
480 is valid, but we should still handle it.
481 The end address is recorded as the start of the next region, but that
482 doesn't work here. To cope we pass 0xff...ff, this is a rare situation
483 anyway. */
484 if (addrmap_index_data.previous_valid)
485 add_address_entry (objfile, addr_vec,
486 addrmap_index_data.previous_cu_start, (CORE_ADDR) -1,
487 addrmap_index_data.previous_cu_index);
488 }
489
490 /* Return the symbol kind of PSYM. */
491
492 static gdb_index_symbol_kind
493 symbol_kind (struct partial_symbol *psym)
494 {
495 domain_enum domain = psym->domain;
496 enum address_class aclass = psym->aclass;
497
498 switch (domain)
499 {
500 case VAR_DOMAIN:
501 switch (aclass)
502 {
503 case LOC_BLOCK:
504 return GDB_INDEX_SYMBOL_KIND_FUNCTION;
505 case LOC_TYPEDEF:
506 return GDB_INDEX_SYMBOL_KIND_TYPE;
507 case LOC_COMPUTED:
508 case LOC_CONST_BYTES:
509 case LOC_OPTIMIZED_OUT:
510 case LOC_STATIC:
511 return GDB_INDEX_SYMBOL_KIND_VARIABLE;
512 case LOC_CONST:
513 /* Note: It's currently impossible to recognize psyms as enum values
514 short of reading the type info. For now punt. */
515 return GDB_INDEX_SYMBOL_KIND_VARIABLE;
516 default:
517 /* There are other LOC_FOO values that one might want to classify
518 as variables, but dwarf2read.c doesn't currently use them. */
519 return GDB_INDEX_SYMBOL_KIND_OTHER;
520 }
521 case STRUCT_DOMAIN:
522 return GDB_INDEX_SYMBOL_KIND_TYPE;
523 default:
524 return GDB_INDEX_SYMBOL_KIND_OTHER;
525 }
526 }
527
528 /* Add a list of partial symbols to SYMTAB. */
529
530 static void
531 write_psymbols (struct mapped_symtab *symtab,
532 std::unordered_set<partial_symbol *> &psyms_seen,
533 struct partial_symbol **psymp,
534 int count,
535 offset_type cu_index,
536 int is_static)
537 {
538 for (; count-- > 0; ++psymp)
539 {
540 struct partial_symbol *psym = *psymp;
541
542 if (psym->language == language_ada)
543 error (_("Ada is not currently supported by the index"));
544
545 /* Only add a given psymbol once. */
546 if (psyms_seen.insert (psym).second)
547 {
548 gdb_index_symbol_kind kind = symbol_kind (psym);
549
550 add_index_entry (symtab, symbol_search_name (psym),
551 is_static, kind, cu_index);
552 }
553 }
554 }
555
556 /* A helper struct used when iterating over debug_types. */
557 struct signatured_type_index_data
558 {
559 signatured_type_index_data (data_buf &types_list_,
560 std::unordered_set<partial_symbol *> &psyms_seen_)
561 : types_list (types_list_), psyms_seen (psyms_seen_)
562 {}
563
564 struct objfile *objfile;
565 struct mapped_symtab *symtab;
566 data_buf &types_list;
567 std::unordered_set<partial_symbol *> &psyms_seen;
568 int cu_index;
569 };
570
571 /* A helper function that writes a single signatured_type to an
572 obstack. */
573
574 static int
575 write_one_signatured_type (void **slot, void *d)
576 {
577 struct signatured_type_index_data *info
578 = (struct signatured_type_index_data *) d;
579 struct signatured_type *entry = (struct signatured_type *) *slot;
580 struct partial_symtab *psymtab = entry->per_cu.v.psymtab;
581
582 write_psymbols (info->symtab,
583 info->psyms_seen,
584 (info->objfile->global_psymbols.data ()
585 + psymtab->globals_offset),
586 psymtab->n_global_syms, info->cu_index,
587 0);
588 write_psymbols (info->symtab,
589 info->psyms_seen,
590 (info->objfile->static_psymbols.data ()
591 + psymtab->statics_offset),
592 psymtab->n_static_syms, info->cu_index,
593 1);
594
595 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE,
596 to_underlying (entry->per_cu.sect_off));
597 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE,
598 to_underlying (entry->type_offset_in_tu));
599 info->types_list.append_uint (8, BFD_ENDIAN_LITTLE, entry->signature);
600
601 ++info->cu_index;
602
603 return 1;
604 }
605
606 /* Recurse into all "included" dependencies and count their symbols as
607 if they appeared in this psymtab. */
608
609 static void
610 recursively_count_psymbols (struct partial_symtab *psymtab,
611 size_t &psyms_seen)
612 {
613 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
614 if (psymtab->dependencies[i]->user != NULL)
615 recursively_count_psymbols (psymtab->dependencies[i],
616 psyms_seen);
617
618 psyms_seen += psymtab->n_global_syms;
619 psyms_seen += psymtab->n_static_syms;
620 }
621
622 /* Recurse into all "included" dependencies and write their symbols as
623 if they appeared in this psymtab. */
624
625 static void
626 recursively_write_psymbols (struct objfile *objfile,
627 struct partial_symtab *psymtab,
628 struct mapped_symtab *symtab,
629 std::unordered_set<partial_symbol *> &psyms_seen,
630 offset_type cu_index)
631 {
632 int i;
633
634 for (i = 0; i < psymtab->number_of_dependencies; ++i)
635 if (psymtab->dependencies[i]->user != NULL)
636 recursively_write_psymbols (objfile, psymtab->dependencies[i],
637 symtab, psyms_seen, cu_index);
638
639 write_psymbols (symtab,
640 psyms_seen,
641 objfile->global_psymbols.data () + psymtab->globals_offset,
642 psymtab->n_global_syms, cu_index,
643 0);
644 write_psymbols (symtab,
645 psyms_seen,
646 objfile->static_psymbols.data () + psymtab->statics_offset,
647 psymtab->n_static_syms, cu_index,
648 1);
649 }
650
651 /* DWARF-5 .debug_names builder. */
652 class debug_names
653 {
654 public:
655 debug_names (struct dwarf2_per_objfile *dwarf2_per_objfile, bool is_dwarf64,
656 bfd_endian dwarf5_byte_order)
657 : m_dwarf5_byte_order (dwarf5_byte_order),
658 m_dwarf32 (dwarf5_byte_order),
659 m_dwarf64 (dwarf5_byte_order),
660 m_dwarf (is_dwarf64
661 ? static_cast<dwarf &> (m_dwarf64)
662 : static_cast<dwarf &> (m_dwarf32)),
663 m_name_table_string_offs (m_dwarf.name_table_string_offs),
664 m_name_table_entry_offs (m_dwarf.name_table_entry_offs),
665 m_debugstrlookup (dwarf2_per_objfile)
666 {}
667
668 int dwarf5_offset_size () const
669 {
670 const bool dwarf5_is_dwarf64 = &m_dwarf == &m_dwarf64;
671 return dwarf5_is_dwarf64 ? 8 : 4;
672 }
673
674 /* Is this symbol from DW_TAG_compile_unit or DW_TAG_type_unit? */
675 enum class unit_kind { cu, tu };
676
677 /* Insert one symbol. */
678 void insert (const partial_symbol *psym, int cu_index, bool is_static,
679 unit_kind kind)
680 {
681 const int dwarf_tag = psymbol_tag (psym);
682 if (dwarf_tag == 0)
683 return;
684 const char *const name = symbol_search_name (psym);
685 const auto insertpair
686 = m_name_to_value_set.emplace (c_str_view (name),
687 std::set<symbol_value> ());
688 std::set<symbol_value> &value_set = insertpair.first->second;
689 value_set.emplace (symbol_value (dwarf_tag, cu_index, is_static, kind));
690 }
691
692 /* Build all the tables. All symbols must be already inserted.
693 This function does not call file_write, caller has to do it
694 afterwards. */
695 void build ()
696 {
697 /* Verify the build method has not be called twice. */
698 gdb_assert (m_abbrev_table.empty ());
699 const size_t name_count = m_name_to_value_set.size ();
700 m_bucket_table.resize
701 (std::pow (2, std::ceil (std::log2 (name_count * 4 / 3))));
702 m_hash_table.reserve (name_count);
703 m_name_table_string_offs.reserve (name_count);
704 m_name_table_entry_offs.reserve (name_count);
705
706 /* Map each hash of symbol to its name and value. */
707 struct hash_it_pair
708 {
709 uint32_t hash;
710 decltype (m_name_to_value_set)::const_iterator it;
711 };
712 std::vector<std::forward_list<hash_it_pair>> bucket_hash;
713 bucket_hash.resize (m_bucket_table.size ());
714 for (decltype (m_name_to_value_set)::const_iterator it
715 = m_name_to_value_set.cbegin ();
716 it != m_name_to_value_set.cend ();
717 ++it)
718 {
719 const char *const name = it->first.c_str ();
720 const uint32_t hash = dwarf5_djb_hash (name);
721 hash_it_pair hashitpair;
722 hashitpair.hash = hash;
723 hashitpair.it = it;
724 auto &slot = bucket_hash[hash % bucket_hash.size()];
725 slot.push_front (std::move (hashitpair));
726 }
727 for (size_t bucket_ix = 0; bucket_ix < bucket_hash.size (); ++bucket_ix)
728 {
729 const std::forward_list<hash_it_pair> &hashitlist
730 = bucket_hash[bucket_ix];
731 if (hashitlist.empty ())
732 continue;
733 uint32_t &bucket_slot = m_bucket_table[bucket_ix];
734 /* The hashes array is indexed starting at 1. */
735 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&bucket_slot),
736 sizeof (bucket_slot), m_dwarf5_byte_order,
737 m_hash_table.size () + 1);
738 for (const hash_it_pair &hashitpair : hashitlist)
739 {
740 m_hash_table.push_back (0);
741 store_unsigned_integer (reinterpret_cast<gdb_byte *>
742 (&m_hash_table.back ()),
743 sizeof (m_hash_table.back ()),
744 m_dwarf5_byte_order, hashitpair.hash);
745 const c_str_view &name = hashitpair.it->first;
746 const std::set<symbol_value> &value_set = hashitpair.it->second;
747 m_name_table_string_offs.push_back_reorder
748 (m_debugstrlookup.lookup (name.c_str ()));
749 m_name_table_entry_offs.push_back_reorder (m_entry_pool.size ());
750 gdb_assert (!value_set.empty ());
751 for (const symbol_value &value : value_set)
752 {
753 int &idx = m_indexkey_to_idx[index_key (value.dwarf_tag,
754 value.is_static,
755 value.kind)];
756 if (idx == 0)
757 {
758 idx = m_idx_next++;
759 m_abbrev_table.append_unsigned_leb128 (idx);
760 m_abbrev_table.append_unsigned_leb128 (value.dwarf_tag);
761 m_abbrev_table.append_unsigned_leb128
762 (value.kind == unit_kind::cu ? DW_IDX_compile_unit
763 : DW_IDX_type_unit);
764 m_abbrev_table.append_unsigned_leb128 (DW_FORM_udata);
765 m_abbrev_table.append_unsigned_leb128 (value.is_static
766 ? DW_IDX_GNU_internal
767 : DW_IDX_GNU_external);
768 m_abbrev_table.append_unsigned_leb128 (DW_FORM_flag_present);
769
770 /* Terminate attributes list. */
771 m_abbrev_table.append_unsigned_leb128 (0);
772 m_abbrev_table.append_unsigned_leb128 (0);
773 }
774
775 m_entry_pool.append_unsigned_leb128 (idx);
776 m_entry_pool.append_unsigned_leb128 (value.cu_index);
777 }
778
779 /* Terminate the list of CUs. */
780 m_entry_pool.append_unsigned_leb128 (0);
781 }
782 }
783 gdb_assert (m_hash_table.size () == name_count);
784
785 /* Terminate tags list. */
786 m_abbrev_table.append_unsigned_leb128 (0);
787 }
788
789 /* Return .debug_names bucket count. This must be called only after
790 calling the build method. */
791 uint32_t bucket_count () const
792 {
793 /* Verify the build method has been already called. */
794 gdb_assert (!m_abbrev_table.empty ());
795 const uint32_t retval = m_bucket_table.size ();
796
797 /* Check for overflow. */
798 gdb_assert (retval == m_bucket_table.size ());
799 return retval;
800 }
801
802 /* Return .debug_names names count. This must be called only after
803 calling the build method. */
804 uint32_t name_count () const
805 {
806 /* Verify the build method has been already called. */
807 gdb_assert (!m_abbrev_table.empty ());
808 const uint32_t retval = m_hash_table.size ();
809
810 /* Check for overflow. */
811 gdb_assert (retval == m_hash_table.size ());
812 return retval;
813 }
814
815 /* Return number of bytes of .debug_names abbreviation table. This
816 must be called only after calling the build method. */
817 uint32_t abbrev_table_bytes () const
818 {
819 gdb_assert (!m_abbrev_table.empty ());
820 return m_abbrev_table.size ();
821 }
822
823 /* Recurse into all "included" dependencies and store their symbols
824 as if they appeared in this psymtab. */
825 void recursively_write_psymbols
826 (struct objfile *objfile,
827 struct partial_symtab *psymtab,
828 std::unordered_set<partial_symbol *> &psyms_seen,
829 int cu_index)
830 {
831 for (int i = 0; i < psymtab->number_of_dependencies; ++i)
832 if (psymtab->dependencies[i]->user != NULL)
833 recursively_write_psymbols (objfile, psymtab->dependencies[i],
834 psyms_seen, cu_index);
835
836 write_psymbols (psyms_seen,
837 objfile->global_psymbols.data () + psymtab->globals_offset,
838 psymtab->n_global_syms, cu_index, false, unit_kind::cu);
839 write_psymbols (psyms_seen,
840 objfile->static_psymbols.data () + psymtab->statics_offset,
841 psymtab->n_static_syms, cu_index, true, unit_kind::cu);
842 }
843
844 /* Return number of bytes the .debug_names section will have. This
845 must be called only after calling the build method. */
846 size_t bytes () const
847 {
848 /* Verify the build method has been already called. */
849 gdb_assert (!m_abbrev_table.empty ());
850 size_t expected_bytes = 0;
851 expected_bytes += m_bucket_table.size () * sizeof (m_bucket_table[0]);
852 expected_bytes += m_hash_table.size () * sizeof (m_hash_table[0]);
853 expected_bytes += m_name_table_string_offs.bytes ();
854 expected_bytes += m_name_table_entry_offs.bytes ();
855 expected_bytes += m_abbrev_table.size ();
856 expected_bytes += m_entry_pool.size ();
857 return expected_bytes;
858 }
859
860 /* Write .debug_names to FILE_NAMES and .debug_str addition to
861 FILE_STR. This must be called only after calling the build
862 method. */
863 void file_write (FILE *file_names, FILE *file_str) const
864 {
865 /* Verify the build method has been already called. */
866 gdb_assert (!m_abbrev_table.empty ());
867 ::file_write (file_names, m_bucket_table);
868 ::file_write (file_names, m_hash_table);
869 m_name_table_string_offs.file_write (file_names);
870 m_name_table_entry_offs.file_write (file_names);
871 m_abbrev_table.file_write (file_names);
872 m_entry_pool.file_write (file_names);
873 m_debugstrlookup.file_write (file_str);
874 }
875
876 /* A helper user data for write_one_signatured_type. */
877 class write_one_signatured_type_data
878 {
879 public:
880 write_one_signatured_type_data (debug_names &nametable_,
881 signatured_type_index_data &&info_)
882 : nametable (nametable_), info (std::move (info_))
883 {}
884 debug_names &nametable;
885 struct signatured_type_index_data info;
886 };
887
888 /* A helper function to pass write_one_signatured_type to
889 htab_traverse_noresize. */
890 static int
891 write_one_signatured_type (void **slot, void *d)
892 {
893 write_one_signatured_type_data *data = (write_one_signatured_type_data *) d;
894 struct signatured_type_index_data *info = &data->info;
895 struct signatured_type *entry = (struct signatured_type *) *slot;
896
897 data->nametable.write_one_signatured_type (entry, info);
898
899 return 1;
900 }
901
902 private:
903
904 /* Storage for symbol names mapping them to their .debug_str section
905 offsets. */
906 class debug_str_lookup
907 {
908 public:
909
910 /* Object costructor to be called for current DWARF2_PER_OBJFILE.
911 All .debug_str section strings are automatically stored. */
912 debug_str_lookup (struct dwarf2_per_objfile *dwarf2_per_objfile)
913 : m_abfd (dwarf2_per_objfile->objfile->obfd),
914 m_dwarf2_per_objfile (dwarf2_per_objfile)
915 {
916 dwarf2_read_section (dwarf2_per_objfile->objfile,
917 &dwarf2_per_objfile->str);
918 if (dwarf2_per_objfile->str.buffer == NULL)
919 return;
920 for (const gdb_byte *data = dwarf2_per_objfile->str.buffer;
921 data < (dwarf2_per_objfile->str.buffer
922 + dwarf2_per_objfile->str.size);)
923 {
924 const char *const s = reinterpret_cast<const char *> (data);
925 const auto insertpair
926 = m_str_table.emplace (c_str_view (s),
927 data - dwarf2_per_objfile->str.buffer);
928 if (!insertpair.second)
929 complaint (_("Duplicate string \"%s\" in "
930 ".debug_str section [in module %s]"),
931 s, bfd_get_filename (m_abfd));
932 data += strlen (s) + 1;
933 }
934 }
935
936 /* Return offset of symbol name S in the .debug_str section. Add
937 such symbol to the section's end if it does not exist there
938 yet. */
939 size_t lookup (const char *s)
940 {
941 const auto it = m_str_table.find (c_str_view (s));
942 if (it != m_str_table.end ())
943 return it->second;
944 const size_t offset = (m_dwarf2_per_objfile->str.size
945 + m_str_add_buf.size ());
946 m_str_table.emplace (c_str_view (s), offset);
947 m_str_add_buf.append_cstr0 (s);
948 return offset;
949 }
950
951 /* Append the end of the .debug_str section to FILE. */
952 void file_write (FILE *file) const
953 {
954 m_str_add_buf.file_write (file);
955 }
956
957 private:
958 std::unordered_map<c_str_view, size_t, c_str_view_hasher> m_str_table;
959 bfd *const m_abfd;
960 struct dwarf2_per_objfile *m_dwarf2_per_objfile;
961
962 /* Data to add at the end of .debug_str for new needed symbol names. */
963 data_buf m_str_add_buf;
964 };
965
966 /* Container to map used DWARF tags to their .debug_names abbreviation
967 tags. */
968 class index_key
969 {
970 public:
971 index_key (int dwarf_tag_, bool is_static_, unit_kind kind_)
972 : dwarf_tag (dwarf_tag_), is_static (is_static_), kind (kind_)
973 {
974 }
975
976 bool
977 operator== (const index_key &other) const
978 {
979 return (dwarf_tag == other.dwarf_tag && is_static == other.is_static
980 && kind == other.kind);
981 }
982
983 const int dwarf_tag;
984 const bool is_static;
985 const unit_kind kind;
986 };
987
988 /* Provide std::unordered_map::hasher for index_key. */
989 class index_key_hasher
990 {
991 public:
992 size_t
993 operator () (const index_key &key) const
994 {
995 return (std::hash<int>() (key.dwarf_tag) << 1) | key.is_static;
996 }
997 };
998
999 /* Parameters of one symbol entry. */
1000 class symbol_value
1001 {
1002 public:
1003 const int dwarf_tag, cu_index;
1004 const bool is_static;
1005 const unit_kind kind;
1006
1007 symbol_value (int dwarf_tag_, int cu_index_, bool is_static_,
1008 unit_kind kind_)
1009 : dwarf_tag (dwarf_tag_), cu_index (cu_index_), is_static (is_static_),
1010 kind (kind_)
1011 {}
1012
1013 bool
1014 operator< (const symbol_value &other) const
1015 {
1016 #define X(n) \
1017 do \
1018 { \
1019 if (n < other.n) \
1020 return true; \
1021 if (n > other.n) \
1022 return false; \
1023 } \
1024 while (0)
1025 X (dwarf_tag);
1026 X (is_static);
1027 X (kind);
1028 X (cu_index);
1029 #undef X
1030 return false;
1031 }
1032 };
1033
1034 /* Abstract base class to unify DWARF-32 and DWARF-64 name table
1035 output. */
1036 class offset_vec
1037 {
1038 protected:
1039 const bfd_endian dwarf5_byte_order;
1040 public:
1041 explicit offset_vec (bfd_endian dwarf5_byte_order_)
1042 : dwarf5_byte_order (dwarf5_byte_order_)
1043 {}
1044
1045 /* Call std::vector::reserve for NELEM elements. */
1046 virtual void reserve (size_t nelem) = 0;
1047
1048 /* Call std::vector::push_back with store_unsigned_integer byte
1049 reordering for ELEM. */
1050 virtual void push_back_reorder (size_t elem) = 0;
1051
1052 /* Return expected output size in bytes. */
1053 virtual size_t bytes () const = 0;
1054
1055 /* Write name table to FILE. */
1056 virtual void file_write (FILE *file) const = 0;
1057 };
1058
1059 /* Template to unify DWARF-32 and DWARF-64 output. */
1060 template<typename OffsetSize>
1061 class offset_vec_tmpl : public offset_vec
1062 {
1063 public:
1064 explicit offset_vec_tmpl (bfd_endian dwarf5_byte_order_)
1065 : offset_vec (dwarf5_byte_order_)
1066 {}
1067
1068 /* Implement offset_vec::reserve. */
1069 void reserve (size_t nelem) override
1070 {
1071 m_vec.reserve (nelem);
1072 }
1073
1074 /* Implement offset_vec::push_back_reorder. */
1075 void push_back_reorder (size_t elem) override
1076 {
1077 m_vec.push_back (elem);
1078 /* Check for overflow. */
1079 gdb_assert (m_vec.back () == elem);
1080 store_unsigned_integer (reinterpret_cast<gdb_byte *> (&m_vec.back ()),
1081 sizeof (m_vec.back ()), dwarf5_byte_order, elem);
1082 }
1083
1084 /* Implement offset_vec::bytes. */
1085 size_t bytes () const override
1086 {
1087 return m_vec.size () * sizeof (m_vec[0]);
1088 }
1089
1090 /* Implement offset_vec::file_write. */
1091 void file_write (FILE *file) const override
1092 {
1093 ::file_write (file, m_vec);
1094 }
1095
1096 private:
1097 std::vector<OffsetSize> m_vec;
1098 };
1099
1100 /* Base class to unify DWARF-32 and DWARF-64 .debug_names output
1101 respecting name table width. */
1102 class dwarf
1103 {
1104 public:
1105 offset_vec &name_table_string_offs, &name_table_entry_offs;
1106
1107 dwarf (offset_vec &name_table_string_offs_,
1108 offset_vec &name_table_entry_offs_)
1109 : name_table_string_offs (name_table_string_offs_),
1110 name_table_entry_offs (name_table_entry_offs_)
1111 {
1112 }
1113 };
1114
1115 /* Template to unify DWARF-32 and DWARF-64 .debug_names output
1116 respecting name table width. */
1117 template<typename OffsetSize>
1118 class dwarf_tmpl : public dwarf
1119 {
1120 public:
1121 explicit dwarf_tmpl (bfd_endian dwarf5_byte_order_)
1122 : dwarf (m_name_table_string_offs, m_name_table_entry_offs),
1123 m_name_table_string_offs (dwarf5_byte_order_),
1124 m_name_table_entry_offs (dwarf5_byte_order_)
1125 {}
1126
1127 private:
1128 offset_vec_tmpl<OffsetSize> m_name_table_string_offs;
1129 offset_vec_tmpl<OffsetSize> m_name_table_entry_offs;
1130 };
1131
1132 /* Try to reconstruct original DWARF tag for given partial_symbol.
1133 This function is not DWARF-5 compliant but it is sufficient for
1134 GDB as a DWARF-5 index consumer. */
1135 static int psymbol_tag (const struct partial_symbol *psym)
1136 {
1137 domain_enum domain = psym->domain;
1138 enum address_class aclass = psym->aclass;
1139
1140 switch (domain)
1141 {
1142 case VAR_DOMAIN:
1143 switch (aclass)
1144 {
1145 case LOC_BLOCK:
1146 return DW_TAG_subprogram;
1147 case LOC_TYPEDEF:
1148 return DW_TAG_typedef;
1149 case LOC_COMPUTED:
1150 case LOC_CONST_BYTES:
1151 case LOC_OPTIMIZED_OUT:
1152 case LOC_STATIC:
1153 return DW_TAG_variable;
1154 case LOC_CONST:
1155 /* Note: It's currently impossible to recognize psyms as enum values
1156 short of reading the type info. For now punt. */
1157 return DW_TAG_variable;
1158 default:
1159 /* There are other LOC_FOO values that one might want to classify
1160 as variables, but dwarf2read.c doesn't currently use them. */
1161 return DW_TAG_variable;
1162 }
1163 case STRUCT_DOMAIN:
1164 return DW_TAG_structure_type;
1165 default:
1166 return 0;
1167 }
1168 }
1169
1170 /* Call insert for all partial symbols and mark them in PSYMS_SEEN. */
1171 void write_psymbols (std::unordered_set<partial_symbol *> &psyms_seen,
1172 struct partial_symbol **psymp, int count, int cu_index,
1173 bool is_static, unit_kind kind)
1174 {
1175 for (; count-- > 0; ++psymp)
1176 {
1177 struct partial_symbol *psym = *psymp;
1178
1179 if (psym->language == language_ada)
1180 error (_("Ada is not currently supported by the index"));
1181
1182 /* Only add a given psymbol once. */
1183 if (psyms_seen.insert (psym).second)
1184 insert (psym, cu_index, is_static, kind);
1185 }
1186 }
1187
1188 /* A helper function that writes a single signatured_type
1189 to a debug_names. */
1190 void
1191 write_one_signatured_type (struct signatured_type *entry,
1192 struct signatured_type_index_data *info)
1193 {
1194 struct partial_symtab *psymtab = entry->per_cu.v.psymtab;
1195
1196 write_psymbols (info->psyms_seen,
1197 (info->objfile->global_psymbols.data ()
1198 + psymtab->globals_offset),
1199 psymtab->n_global_syms, info->cu_index, false,
1200 unit_kind::tu);
1201 write_psymbols (info->psyms_seen,
1202 (info->objfile->static_psymbols.data ()
1203 + psymtab->statics_offset),
1204 psymtab->n_static_syms, info->cu_index, true,
1205 unit_kind::tu);
1206
1207 info->types_list.append_uint (dwarf5_offset_size (), m_dwarf5_byte_order,
1208 to_underlying (entry->per_cu.sect_off));
1209
1210 ++info->cu_index;
1211 }
1212
1213 /* Store value of each symbol. */
1214 std::unordered_map<c_str_view, std::set<symbol_value>, c_str_view_hasher>
1215 m_name_to_value_set;
1216
1217 /* Tables of DWARF-5 .debug_names. They are in object file byte
1218 order. */
1219 std::vector<uint32_t> m_bucket_table;
1220 std::vector<uint32_t> m_hash_table;
1221
1222 const bfd_endian m_dwarf5_byte_order;
1223 dwarf_tmpl<uint32_t> m_dwarf32;
1224 dwarf_tmpl<uint64_t> m_dwarf64;
1225 dwarf &m_dwarf;
1226 offset_vec &m_name_table_string_offs, &m_name_table_entry_offs;
1227 debug_str_lookup m_debugstrlookup;
1228
1229 /* Map each used .debug_names abbreviation tag parameter to its
1230 index value. */
1231 std::unordered_map<index_key, int, index_key_hasher> m_indexkey_to_idx;
1232
1233 /* Next unused .debug_names abbreviation tag for
1234 m_indexkey_to_idx. */
1235 int m_idx_next = 1;
1236
1237 /* .debug_names abbreviation table. */
1238 data_buf m_abbrev_table;
1239
1240 /* .debug_names entry pool. */
1241 data_buf m_entry_pool;
1242 };
1243
1244 /* Return iff any of the needed offsets does not fit into 32-bit
1245 .debug_names section. */
1246
1247 static bool
1248 check_dwarf64_offsets (struct dwarf2_per_objfile *dwarf2_per_objfile)
1249 {
1250 for (dwarf2_per_cu_data *per_cu : dwarf2_per_objfile->all_comp_units)
1251 {
1252 if (to_underlying (per_cu->sect_off) >= (static_cast<uint64_t> (1) << 32))
1253 return true;
1254 }
1255 for (const signatured_type *sigtype : dwarf2_per_objfile->all_type_units)
1256 {
1257 const dwarf2_per_cu_data &per_cu = sigtype->per_cu;
1258
1259 if (to_underlying (per_cu.sect_off) >= (static_cast<uint64_t> (1) << 32))
1260 return true;
1261 }
1262 return false;
1263 }
1264
1265 /* The psyms_seen set is potentially going to be largish (~40k
1266 elements when indexing a -g3 build of GDB itself). Estimate the
1267 number of elements in order to avoid too many rehashes, which
1268 require rebuilding buckets and thus many trips to
1269 malloc/free. */
1270
1271 static size_t
1272 psyms_seen_size (struct dwarf2_per_objfile *dwarf2_per_objfile)
1273 {
1274 size_t psyms_count = 0;
1275 for (dwarf2_per_cu_data *per_cu : dwarf2_per_objfile->all_comp_units)
1276 {
1277 struct partial_symtab *psymtab = per_cu->v.psymtab;
1278
1279 if (psymtab != NULL && psymtab->user == NULL)
1280 recursively_count_psymbols (psymtab, psyms_count);
1281 }
1282 /* Generating an index for gdb itself shows a ratio of
1283 TOTAL_SEEN_SYMS/UNIQUE_SYMS or ~5. 4 seems like a good bet. */
1284 return psyms_count / 4;
1285 }
1286
1287 /* Write new .gdb_index section for OBJFILE into OUT_FILE.
1288 Return how many bytes were expected to be written into OUT_FILE. */
1289
1290 static size_t
1291 write_gdbindex (struct dwarf2_per_objfile *dwarf2_per_objfile, FILE *out_file)
1292 {
1293 struct objfile *objfile = dwarf2_per_objfile->objfile;
1294 mapped_symtab symtab;
1295 data_buf cu_list;
1296
1297 /* While we're scanning CU's create a table that maps a psymtab pointer
1298 (which is what addrmap records) to its index (which is what is recorded
1299 in the index file). This will later be needed to write the address
1300 table. */
1301 psym_index_map cu_index_htab;
1302 cu_index_htab.reserve (dwarf2_per_objfile->all_comp_units.size ());
1303
1304 /* The CU list is already sorted, so we don't need to do additional
1305 work here. Also, the debug_types entries do not appear in
1306 all_comp_units, but only in their own hash table. */
1307
1308 std::unordered_set<partial_symbol *> psyms_seen
1309 (psyms_seen_size (dwarf2_per_objfile));
1310 for (int i = 0; i < dwarf2_per_objfile->all_comp_units.size (); ++i)
1311 {
1312 struct dwarf2_per_cu_data *per_cu
1313 = dwarf2_per_objfile->all_comp_units[i];
1314 struct partial_symtab *psymtab = per_cu->v.psymtab;
1315
1316 /* CU of a shared file from 'dwz -m' may be unused by this main file.
1317 It may be referenced from a local scope but in such case it does not
1318 need to be present in .gdb_index. */
1319 if (psymtab == NULL)
1320 continue;
1321
1322 if (psymtab->user == NULL)
1323 recursively_write_psymbols (objfile, psymtab, &symtab,
1324 psyms_seen, i);
1325
1326 const auto insertpair = cu_index_htab.emplace (psymtab, i);
1327 gdb_assert (insertpair.second);
1328
1329 cu_list.append_uint (8, BFD_ENDIAN_LITTLE,
1330 to_underlying (per_cu->sect_off));
1331 cu_list.append_uint (8, BFD_ENDIAN_LITTLE, per_cu->length);
1332 }
1333
1334 /* Dump the address map. */
1335 data_buf addr_vec;
1336 write_address_map (objfile, addr_vec, cu_index_htab);
1337
1338 /* Write out the .debug_type entries, if any. */
1339 data_buf types_cu_list;
1340 if (dwarf2_per_objfile->signatured_types)
1341 {
1342 signatured_type_index_data sig_data (types_cu_list,
1343 psyms_seen);
1344
1345 sig_data.objfile = objfile;
1346 sig_data.symtab = &symtab;
1347 sig_data.cu_index = dwarf2_per_objfile->all_comp_units.size ();
1348 htab_traverse_noresize (dwarf2_per_objfile->signatured_types,
1349 write_one_signatured_type, &sig_data);
1350 }
1351
1352 /* Now that we've processed all symbols we can shrink their cu_indices
1353 lists. */
1354 uniquify_cu_indices (&symtab);
1355
1356 data_buf symtab_vec, constant_pool;
1357 write_hash_table (&symtab, symtab_vec, constant_pool);
1358
1359 data_buf contents;
1360 const offset_type size_of_contents = 6 * sizeof (offset_type);
1361 offset_type total_len = size_of_contents;
1362
1363 /* The version number. */
1364 contents.append_data (MAYBE_SWAP (8));
1365
1366 /* The offset of the CU list from the start of the file. */
1367 contents.append_data (MAYBE_SWAP (total_len));
1368 total_len += cu_list.size ();
1369
1370 /* The offset of the types CU list from the start of the file. */
1371 contents.append_data (MAYBE_SWAP (total_len));
1372 total_len += types_cu_list.size ();
1373
1374 /* The offset of the address table from the start of the file. */
1375 contents.append_data (MAYBE_SWAP (total_len));
1376 total_len += addr_vec.size ();
1377
1378 /* The offset of the symbol table from the start of the file. */
1379 contents.append_data (MAYBE_SWAP (total_len));
1380 total_len += symtab_vec.size ();
1381
1382 /* The offset of the constant pool from the start of the file. */
1383 contents.append_data (MAYBE_SWAP (total_len));
1384 total_len += constant_pool.size ();
1385
1386 gdb_assert (contents.size () == size_of_contents);
1387
1388 contents.file_write (out_file);
1389 cu_list.file_write (out_file);
1390 types_cu_list.file_write (out_file);
1391 addr_vec.file_write (out_file);
1392 symtab_vec.file_write (out_file);
1393 constant_pool.file_write (out_file);
1394
1395 return total_len;
1396 }
1397
1398 /* DWARF-5 augmentation string for GDB's DW_IDX_GNU_* extension. */
1399 static const gdb_byte dwarf5_gdb_augmentation[] = { 'G', 'D', 'B', 0 };
1400
1401 /* Write a new .debug_names section for OBJFILE into OUT_FILE, write
1402 needed addition to .debug_str section to OUT_FILE_STR. Return how
1403 many bytes were expected to be written into OUT_FILE. */
1404
1405 static size_t
1406 write_debug_names (struct dwarf2_per_objfile *dwarf2_per_objfile,
1407 FILE *out_file, FILE *out_file_str)
1408 {
1409 const bool dwarf5_is_dwarf64 = check_dwarf64_offsets (dwarf2_per_objfile);
1410 struct objfile *objfile = dwarf2_per_objfile->objfile;
1411 const enum bfd_endian dwarf5_byte_order
1412 = gdbarch_byte_order (get_objfile_arch (objfile));
1413
1414 /* The CU list is already sorted, so we don't need to do additional
1415 work here. Also, the debug_types entries do not appear in
1416 all_comp_units, but only in their own hash table. */
1417 data_buf cu_list;
1418 debug_names nametable (dwarf2_per_objfile, dwarf5_is_dwarf64,
1419 dwarf5_byte_order);
1420 std::unordered_set<partial_symbol *>
1421 psyms_seen (psyms_seen_size (dwarf2_per_objfile));
1422 for (int i = 0; i < dwarf2_per_objfile->all_comp_units.size (); ++i)
1423 {
1424 const dwarf2_per_cu_data *per_cu = dwarf2_per_objfile->all_comp_units[i];
1425 partial_symtab *psymtab = per_cu->v.psymtab;
1426
1427 /* CU of a shared file from 'dwz -m' may be unused by this main
1428 file. It may be referenced from a local scope but in such
1429 case it does not need to be present in .debug_names. */
1430 if (psymtab == NULL)
1431 continue;
1432
1433 if (psymtab->user == NULL)
1434 nametable.recursively_write_psymbols (objfile, psymtab, psyms_seen, i);
1435
1436 cu_list.append_uint (nametable.dwarf5_offset_size (), dwarf5_byte_order,
1437 to_underlying (per_cu->sect_off));
1438 }
1439
1440 /* Write out the .debug_type entries, if any. */
1441 data_buf types_cu_list;
1442 if (dwarf2_per_objfile->signatured_types)
1443 {
1444 debug_names::write_one_signatured_type_data sig_data (nametable,
1445 signatured_type_index_data (types_cu_list, psyms_seen));
1446
1447 sig_data.info.objfile = objfile;
1448 /* It is used only for gdb_index. */
1449 sig_data.info.symtab = nullptr;
1450 sig_data.info.cu_index = 0;
1451 htab_traverse_noresize (dwarf2_per_objfile->signatured_types,
1452 debug_names::write_one_signatured_type,
1453 &sig_data);
1454 }
1455
1456 nametable.build ();
1457
1458 /* No addr_vec - DWARF-5 uses .debug_aranges generated by GCC. */
1459
1460 const offset_type bytes_of_header
1461 = ((dwarf5_is_dwarf64 ? 12 : 4)
1462 + 2 + 2 + 7 * 4
1463 + sizeof (dwarf5_gdb_augmentation));
1464 size_t expected_bytes = 0;
1465 expected_bytes += bytes_of_header;
1466 expected_bytes += cu_list.size ();
1467 expected_bytes += types_cu_list.size ();
1468 expected_bytes += nametable.bytes ();
1469 data_buf header;
1470
1471 if (!dwarf5_is_dwarf64)
1472 {
1473 const uint64_t size64 = expected_bytes - 4;
1474 gdb_assert (size64 < 0xfffffff0);
1475 header.append_uint (4, dwarf5_byte_order, size64);
1476 }
1477 else
1478 {
1479 header.append_uint (4, dwarf5_byte_order, 0xffffffff);
1480 header.append_uint (8, dwarf5_byte_order, expected_bytes - 12);
1481 }
1482
1483 /* The version number. */
1484 header.append_uint (2, dwarf5_byte_order, 5);
1485
1486 /* Padding. */
1487 header.append_uint (2, dwarf5_byte_order, 0);
1488
1489 /* comp_unit_count - The number of CUs in the CU list. */
1490 header.append_uint (4, dwarf5_byte_order,
1491 dwarf2_per_objfile->all_comp_units.size ());
1492
1493 /* local_type_unit_count - The number of TUs in the local TU
1494 list. */
1495 header.append_uint (4, dwarf5_byte_order,
1496 dwarf2_per_objfile->all_type_units.size ());
1497
1498 /* foreign_type_unit_count - The number of TUs in the foreign TU
1499 list. */
1500 header.append_uint (4, dwarf5_byte_order, 0);
1501
1502 /* bucket_count - The number of hash buckets in the hash lookup
1503 table. */
1504 header.append_uint (4, dwarf5_byte_order, nametable.bucket_count ());
1505
1506 /* name_count - The number of unique names in the index. */
1507 header.append_uint (4, dwarf5_byte_order, nametable.name_count ());
1508
1509 /* abbrev_table_size - The size in bytes of the abbreviations
1510 table. */
1511 header.append_uint (4, dwarf5_byte_order, nametable.abbrev_table_bytes ());
1512
1513 /* augmentation_string_size - The size in bytes of the augmentation
1514 string. This value is rounded up to a multiple of 4. */
1515 static_assert (sizeof (dwarf5_gdb_augmentation) % 4 == 0, "");
1516 header.append_uint (4, dwarf5_byte_order, sizeof (dwarf5_gdb_augmentation));
1517 header.append_data (dwarf5_gdb_augmentation);
1518
1519 gdb_assert (header.size () == bytes_of_header);
1520
1521 header.file_write (out_file);
1522 cu_list.file_write (out_file);
1523 types_cu_list.file_write (out_file);
1524 nametable.file_write (out_file, out_file_str);
1525
1526 return expected_bytes;
1527 }
1528
1529 /* Assert that FILE's size is EXPECTED_SIZE. Assumes file's seek
1530 position is at the end of the file. */
1531
1532 static void
1533 assert_file_size (FILE *file, const char *filename, size_t expected_size)
1534 {
1535 const auto file_size = ftell (file);
1536 if (file_size == -1)
1537 error (_("Can't get `%s' size"), filename);
1538 gdb_assert (file_size == expected_size);
1539 }
1540
1541 /* See dwarf-index-write.h. */
1542
1543 void
1544 write_psymtabs_to_index (struct dwarf2_per_objfile *dwarf2_per_objfile,
1545 const char *dir, const char *basename,
1546 dw_index_kind index_kind)
1547 {
1548 struct objfile *objfile = dwarf2_per_objfile->objfile;
1549
1550 if (dwarf2_per_objfile->using_index)
1551 error (_("Cannot use an index to create the index"));
1552
1553 if (VEC_length (dwarf2_section_info_def, dwarf2_per_objfile->types) > 1)
1554 error (_("Cannot make an index when the file has multiple .debug_types sections"));
1555
1556 if (!objfile->psymtabs || !objfile->psymtabs_addrmap)
1557 return;
1558
1559 struct stat st;
1560 if (stat (objfile_name (objfile), &st) < 0)
1561 perror_with_name (objfile_name (objfile));
1562
1563 /* Make a filename suitable to pass to mkstemp based on F (e.g.
1564 /tmp/foo -> /tmp/foo-XXXXXX). */
1565 auto make_temp_filename = [] (const std::string &f) -> gdb::char_vector
1566 {
1567 gdb::char_vector filename_temp (f.length () + 8);
1568 strcpy (filename_temp.data (), f.c_str ());
1569 strcat (filename_temp.data () + f.size (), "-XXXXXX");
1570 return filename_temp;
1571 };
1572
1573 std::string filename (std::string (dir) + SLASH_STRING + basename
1574 + (index_kind == dw_index_kind::DEBUG_NAMES
1575 ? INDEX5_SUFFIX : INDEX4_SUFFIX));
1576 gdb::char_vector filename_temp = make_temp_filename (filename);
1577
1578 gdb::optional<scoped_fd> out_file_fd
1579 (gdb::in_place, mkstemp (filename_temp.data ()));
1580 if (out_file_fd->get () == -1)
1581 perror_with_name (("mkstemp"));
1582
1583 FILE *out_file = gdb_fopen_cloexec (filename_temp.data (), "wb").release ();
1584 if (out_file == nullptr)
1585 error (_("Can't open `%s' for writing"), filename_temp.data ());
1586
1587 /* Order matters here; we want FILE to be closed before FILENAME_TEMP is
1588 unlinked, because on MS-Windows one cannot delete a file that is
1589 still open. (Don't call anything here that might throw until
1590 file_closer is created.) We don't need OUT_FILE_FD anymore, so we might
1591 as well close it now. */
1592 out_file_fd.reset ();
1593 gdb::unlinker unlink_file (filename_temp.data ());
1594 gdb_file_up close_out_file (out_file);
1595
1596 if (index_kind == dw_index_kind::DEBUG_NAMES)
1597 {
1598 std::string filename_str (std::string (dir) + SLASH_STRING
1599 + basename + DEBUG_STR_SUFFIX);
1600 gdb::char_vector filename_str_temp = make_temp_filename (filename_str);
1601
1602 gdb::optional<scoped_fd> out_file_str_fd
1603 (gdb::in_place, mkstemp (filename_str_temp.data ()));
1604 if (out_file_str_fd->get () == -1)
1605 perror_with_name (("mkstemp"));
1606
1607 FILE *out_file_str
1608 = gdb_fopen_cloexec (filename_str_temp.data (), "wb").release ();
1609 if (out_file_str == nullptr)
1610 error (_("Can't open `%s' for writing"), filename_str_temp.data ());
1611
1612 out_file_str_fd.reset ();
1613 gdb::unlinker unlink_file_str (filename_str_temp.data ());
1614 gdb_file_up close_out_file_str (out_file_str);
1615
1616 const size_t total_len
1617 = write_debug_names (dwarf2_per_objfile, out_file, out_file_str);
1618 assert_file_size (out_file, filename_temp.data (), total_len);
1619
1620 /* We want to keep the file .debug_str file too. */
1621 unlink_file_str.keep ();
1622
1623 /* Close and move the str file in place. */
1624 close_out_file_str.reset ();
1625 if (rename (filename_str_temp.data (), filename_str.c_str ()) != 0)
1626 perror_with_name (("rename"));
1627 }
1628 else
1629 {
1630 const size_t total_len
1631 = write_gdbindex (dwarf2_per_objfile, out_file);
1632 assert_file_size (out_file, filename_temp.data (), total_len);
1633 }
1634
1635 /* We want to keep the file. */
1636 unlink_file.keep ();
1637
1638 /* Close and move the file in place. */
1639 close_out_file.reset ();
1640 if (rename (filename_temp.data (), filename.c_str ()) != 0)
1641 perror_with_name (("rename"));
1642 }
1643
1644 /* Implementation of the `save gdb-index' command.
1645
1646 Note that the .gdb_index file format used by this command is
1647 documented in the GDB manual. Any changes here must be documented
1648 there. */
1649
1650 static void
1651 save_gdb_index_command (const char *arg, int from_tty)
1652 {
1653 struct objfile *objfile;
1654 const char dwarf5space[] = "-dwarf-5 ";
1655 dw_index_kind index_kind = dw_index_kind::GDB_INDEX;
1656
1657 if (!arg)
1658 arg = "";
1659
1660 arg = skip_spaces (arg);
1661 if (strncmp (arg, dwarf5space, strlen (dwarf5space)) == 0)
1662 {
1663 index_kind = dw_index_kind::DEBUG_NAMES;
1664 arg += strlen (dwarf5space);
1665 arg = skip_spaces (arg);
1666 }
1667
1668 if (!*arg)
1669 error (_("usage: save gdb-index [-dwarf-5] DIRECTORY"));
1670
1671 ALL_OBJFILES (objfile)
1672 {
1673 struct stat st;
1674
1675 /* If the objfile does not correspond to an actual file, skip it. */
1676 if (stat (objfile_name (objfile), &st) < 0)
1677 continue;
1678
1679 struct dwarf2_per_objfile *dwarf2_per_objfile
1680 = get_dwarf2_per_objfile (objfile);
1681
1682 if (dwarf2_per_objfile != NULL)
1683 {
1684 TRY
1685 {
1686 const char *basename = lbasename (objfile_name (objfile));
1687 write_psymtabs_to_index (dwarf2_per_objfile, arg, basename,
1688 index_kind);
1689 }
1690 CATCH (except, RETURN_MASK_ERROR)
1691 {
1692 exception_fprintf (gdb_stderr, except,
1693 _("Error while writing index for `%s': "),
1694 objfile_name (objfile));
1695 }
1696 END_CATCH
1697 }
1698
1699 }
1700 }
1701
1702 void
1703 _initialize_dwarf_index_write ()
1704 {
1705 cmd_list_element *c = add_cmd ("gdb-index", class_files,
1706 save_gdb_index_command, _("\
1707 Save a gdb-index file.\n\
1708 Usage: save gdb-index [-dwarf-5] DIRECTORY\n\
1709 \n\
1710 No options create one file with .gdb-index extension for pre-DWARF-5\n\
1711 compatible .gdb_index section. With -dwarf-5 creates two files with\n\
1712 extension .debug_names and .debug_str for DWARF-5 .debug_names section."),
1713 &save_cmdlist);
1714 set_cmd_completer (c, filename_completer);
1715 }
This page took 0.083354 seconds and 5 git commands to generate.