1 /* Get macro information. 2 Copyright (C) 2002-2009, 2014, 2017, 2018 Red Hat, Inc. 3 This file is part of elfutils. 4 5 This file is free software; you can redistribute it and/or modify 6 it under the terms of either 7 8 * the GNU Lesser General Public License as published by the Free 9 Software Foundation; either version 3 of the License, or (at 10 your option) any later version 11 12 or 13 14 * the GNU General Public License as published by the Free 15 Software Foundation; either version 2 of the License, or (at 16 your option) any later version 17 18 or both in parallel, as here. 19 20 elfutils is distributed in the hope that it will be useful, but 21 WITHOUT ANY WARRANTY; without even the implied warranty of 22 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 23 General Public License for more details. 24 25 You should have received copies of the GNU General Public License and 26 the GNU Lesser General Public License along with this program. If 27 not, see <http://www.gnu.org/licenses/>. */ 28 29 #ifdef HAVE_CONFIG_H 30 # include <config.h> 31 #endif 32 33 #include <assert.h> 34 #include <dwarf.h> 35 #include <search.h> 36 #include <stdlib.h> 37 #include <string.h> 38 39 #include <libdwP.h> 40 41 static int 42 get_offset_from (Dwarf_Die *die, int name, Dwarf_Word *retp) 43 { 44 /* Get the appropriate attribute. */ 45 Dwarf_Attribute attr; 46 if (INTUSE(dwarf_attr) (die, name, &attr) == NULL) 47 return -1; 48 49 /* Offset into the corresponding section. */ 50 return INTUSE(dwarf_formudata) (&attr, retp); 51 } 52 53 static int 54 macro_op_compare (const void *p1, const void *p2) 55 { 56 const Dwarf_Macro_Op_Table *t1 = (const Dwarf_Macro_Op_Table *) p1; 57 const Dwarf_Macro_Op_Table *t2 = (const Dwarf_Macro_Op_Table *) p2; 58 59 if (t1->offset < t2->offset) 60 return -1; 61 if (t1->offset > t2->offset) 62 return 1; 63 64 if (t1->sec_index < t2->sec_index) 65 return -1; 66 if (t1->sec_index > t2->sec_index) 67 return 1; 68 69 return 0; 70 } 71 72 static void 73 build_table (Dwarf_Macro_Op_Table *table, 74 Dwarf_Macro_Op_Proto op_protos[static 255]) 75 { 76 unsigned ct = 0; 77 for (unsigned i = 1; i < 256; ++i) 78 if (op_protos[i - 1].forms != NULL) 79 table->table[table->opcodes[i - 1] = ct++] = op_protos[i - 1]; 80 else 81 table->opcodes[i - 1] = 0xff; 82 } 83 84 #define MACRO_PROTO(NAME, ...) \ 85 Dwarf_Macro_Op_Proto NAME = ({ \ 86 static const uint8_t proto[] = {__VA_ARGS__}; \ 87 (Dwarf_Macro_Op_Proto) {sizeof proto, proto}; \ 88 }) 89 90 enum { macinfo_data_size = offsetof (Dwarf_Macro_Op_Table, table[5]) }; 91 static unsigned char macinfo_data[macinfo_data_size] 92 __attribute__ ((aligned (__alignof (Dwarf_Macro_Op_Table)))); 93 94 static __attribute__ ((constructor)) void 95 init_macinfo_table (void) 96 { 97 MACRO_PROTO (p_udata_str, DW_FORM_udata, DW_FORM_string); 98 MACRO_PROTO (p_udata_udata, DW_FORM_udata, DW_FORM_udata); 99 MACRO_PROTO (p_none); 100 101 Dwarf_Macro_Op_Proto op_protos[255] = 102 { 103 [DW_MACINFO_define - 1] = p_udata_str, 104 [DW_MACINFO_undef - 1] = p_udata_str, 105 [DW_MACINFO_vendor_ext - 1] = p_udata_str, 106 [DW_MACINFO_start_file - 1] = p_udata_udata, 107 [DW_MACINFO_end_file - 1] = p_none, 108 /* If you are adding more elements to this array, increase 109 MACINFO_DATA_SIZE above. */ 110 }; 111 112 Dwarf_Macro_Op_Table *macinfo_table = (void *) macinfo_data; 113 memset (macinfo_table, 0, sizeof macinfo_data); 114 build_table (macinfo_table, op_protos); 115 macinfo_table->sec_index = IDX_debug_macinfo; 116 } 117 118 static Dwarf_Macro_Op_Table * 119 get_macinfo_table (Dwarf *dbg, Dwarf_Word macoff, Dwarf_Die *cudie) 120 { 121 assert (cudie != NULL); 122 123 Dwarf_Attribute attr_mem, *attr 124 = INTUSE(dwarf_attr) (cudie, DW_AT_stmt_list, &attr_mem); 125 Dwarf_Off line_offset = (Dwarf_Off) -1; 126 if (attr != NULL) 127 if (unlikely (INTUSE(dwarf_formudata) (attr, &line_offset) != 0)) 128 return NULL; 129 130 Dwarf_Macro_Op_Table *table = libdw_alloc (dbg, Dwarf_Macro_Op_Table, 131 macinfo_data_size, 1); 132 memcpy (table, macinfo_data, macinfo_data_size); 133 134 table->offset = macoff; 135 table->sec_index = IDX_debug_macinfo; 136 table->line_offset = line_offset; 137 table->is_64bit = cudie->cu->address_size == 8; 138 table->comp_dir = __libdw_getcompdir (cudie); 139 140 return table; 141 } 142 143 static Dwarf_Macro_Op_Table * 144 get_table_for_offset (Dwarf *dbg, Dwarf_Word macoff, 145 const unsigned char *readp, 146 const unsigned char *const endp, 147 Dwarf_Die *cudie) 148 { 149 const unsigned char *startp = readp; 150 151 /* Request at least 3 bytes for header. */ 152 if (readp + 3 > endp) 153 { 154 invalid_dwarf: 155 __libdw_seterrno (DWARF_E_INVALID_DWARF); 156 return NULL; 157 } 158 159 uint16_t version = read_2ubyte_unaligned_inc (dbg, readp); 160 if (version != 4 && version != 5) 161 { 162 __libdw_seterrno (DWARF_E_INVALID_VERSION); 163 return NULL; 164 } 165 166 uint8_t flags = *readp++; 167 bool is_64bit = (flags & 0x1) != 0; 168 169 Dwarf_Off line_offset = (Dwarf_Off) -1; 170 if ((flags & 0x2) != 0) 171 { 172 line_offset = read_addr_unaligned_inc (is_64bit ? 8 : 4, dbg, readp); 173 if (readp > endp) 174 goto invalid_dwarf; 175 } 176 else if (cudie != NULL) 177 { 178 Dwarf_Attribute attr_mem, *attr 179 = INTUSE(dwarf_attr) (cudie, DW_AT_stmt_list, &attr_mem); 180 if (attr != NULL) 181 if (unlikely (INTUSE(dwarf_formudata) (attr, &line_offset) != 0)) 182 return NULL; 183 } 184 185 /* """The macinfo entry types defined in this standard may, but 186 might not, be described in the table""". 187 188 I.e. these may be present. It's tempting to simply skip them, 189 but it's probably more correct to tolerate that a producer tweaks 190 the way certain opcodes are encoded, for whatever reasons. */ 191 192 MACRO_PROTO (p_udata_str, DW_FORM_udata, DW_FORM_string); 193 MACRO_PROTO (p_udata_strp, DW_FORM_udata, DW_FORM_strp); 194 MACRO_PROTO (p_udata_strsup, DW_FORM_udata, DW_FORM_strp_sup); 195 MACRO_PROTO (p_udata_strx, DW_FORM_udata, DW_FORM_strx); 196 MACRO_PROTO (p_udata_udata, DW_FORM_udata, DW_FORM_udata); 197 MACRO_PROTO (p_secoffset, DW_FORM_sec_offset); 198 MACRO_PROTO (p_none); 199 200 Dwarf_Macro_Op_Proto op_protos[255] = 201 { 202 [DW_MACRO_define - 1] = p_udata_str, 203 [DW_MACRO_undef - 1] = p_udata_str, 204 [DW_MACRO_define_strp - 1] = p_udata_strp, 205 [DW_MACRO_undef_strp - 1] = p_udata_strp, 206 [DW_MACRO_start_file - 1] = p_udata_udata, 207 [DW_MACRO_end_file - 1] = p_none, 208 [DW_MACRO_import - 1] = p_secoffset, 209 [DW_MACRO_define_sup - 1] = p_udata_strsup, 210 [DW_MACRO_undef_sup - 1] = p_udata_strsup, 211 [DW_MACRO_import_sup - 1] = p_secoffset, /* XXX - but in sup!. */ 212 [DW_MACRO_define_strx - 1] = p_udata_strx, 213 [DW_MACRO_undef_strx - 1] = p_udata_strx, 214 }; 215 216 if ((flags & 0x4) != 0) 217 { 218 unsigned count = *readp++; 219 for (unsigned i = 0; i < count; ++i) 220 { 221 unsigned opcode = *readp++; 222 223 Dwarf_Macro_Op_Proto e; 224 if (readp >= endp) 225 goto invalid; 226 get_uleb128 (e.nforms, readp, endp); 227 e.forms = readp; 228 op_protos[opcode - 1] = e; 229 230 readp += e.nforms; 231 if (readp > endp) 232 { 233 invalid: 234 __libdw_seterrno (DWARF_E_INVALID_DWARF); 235 return NULL; 236 } 237 } 238 } 239 240 size_t ct = 0; 241 for (unsigned i = 1; i < 256; ++i) 242 if (op_protos[i - 1].forms != NULL) 243 ++ct; 244 245 /* We support at most 0xfe opcodes defined in the table, as 0xff is 246 a value that means that given opcode is not stored at all. But 247 that should be fine, as opcode 0 is not allocated. */ 248 assert (ct < 0xff); 249 250 size_t macop_table_size = offsetof (Dwarf_Macro_Op_Table, table[ct]); 251 252 Dwarf_Macro_Op_Table *table = libdw_alloc (dbg, Dwarf_Macro_Op_Table, 253 macop_table_size, 1); 254 255 *table = (Dwarf_Macro_Op_Table) { 256 .offset = macoff, 257 .sec_index = IDX_debug_macro, 258 .line_offset = line_offset, 259 .header_len = readp - startp, 260 .version = version, 261 .is_64bit = is_64bit, 262 263 /* NULL if CUDIE is NULL or DW_AT_comp_dir is absent. */ 264 .comp_dir = __libdw_getcompdir (cudie), 265 }; 266 build_table (table, op_protos); 267 268 return table; 269 } 270 271 static Dwarf_Macro_Op_Table * 272 cache_op_table (Dwarf *dbg, int sec_index, Dwarf_Off macoff, 273 const unsigned char *startp, 274 const unsigned char *const endp, 275 Dwarf_Die *cudie) 276 { 277 Dwarf_Macro_Op_Table fake = { .offset = macoff, .sec_index = sec_index }; 278 Dwarf_Macro_Op_Table **found = tfind (&fake, &dbg->macro_ops, 279 macro_op_compare); 280 if (found != NULL) 281 return *found; 282 283 Dwarf_Macro_Op_Table *table = sec_index == IDX_debug_macro 284 ? get_table_for_offset (dbg, macoff, startp, endp, cudie) 285 : get_macinfo_table (dbg, macoff, cudie); 286 287 if (table == NULL) 288 return NULL; 289 290 Dwarf_Macro_Op_Table **ret = tsearch (table, &dbg->macro_ops, 291 macro_op_compare); 292 if (unlikely (ret == NULL)) 293 { 294 __libdw_seterrno (DWARF_E_NOMEM); 295 return NULL; 296 } 297 298 return *ret; 299 } 300 301 static ptrdiff_t 302 read_macros (Dwarf *dbg, int sec_index, 303 Dwarf_Off macoff, int (*callback) (Dwarf_Macro *, void *), 304 void *arg, ptrdiff_t offset, bool accept_0xff, 305 Dwarf_Die *cudie) 306 { 307 Elf_Data *d = dbg->sectiondata[sec_index]; 308 if (unlikely (d == NULL || d->d_buf == NULL)) 309 { 310 __libdw_seterrno (DWARF_E_NO_ENTRY); 311 return -1; 312 } 313 314 if (unlikely (macoff >= d->d_size)) 315 { 316 __libdw_seterrno (DWARF_E_INVALID_DWARF); 317 return -1; 318 } 319 320 const unsigned char *const startp = d->d_buf + macoff; 321 const unsigned char *const endp = d->d_buf + d->d_size; 322 323 Dwarf_Macro_Op_Table *table = cache_op_table (dbg, sec_index, macoff, 324 startp, endp, cudie); 325 if (table == NULL) 326 return -1; 327 328 if (offset == 0) 329 offset = table->header_len; 330 331 assert (offset >= 0); 332 assert (offset < endp - startp); 333 const unsigned char *readp = startp + offset; 334 335 while (readp < endp) 336 { 337 unsigned int opcode = *readp++; 338 if (opcode == 0) 339 /* Nothing more to do. */ 340 return 0; 341 342 if (unlikely (opcode == 0xff && ! accept_0xff)) 343 { 344 /* See comment below at dwarf_getmacros for explanation of 345 why we are doing this. */ 346 __libdw_seterrno (DWARF_E_INVALID_OPCODE); 347 return -1; 348 } 349 350 unsigned int idx = table->opcodes[opcode - 1]; 351 if (idx == 0xff) 352 { 353 __libdw_seterrno (DWARF_E_INVALID_OPCODE); 354 return -1; 355 } 356 357 Dwarf_Macro_Op_Proto *proto = &table->table[idx]; 358 359 /* A fake CU with bare minimum data to fool dwarf_formX into 360 doing the right thing with the attributes that we put out. 361 We pretend it is the same version as the actual table. 362 Version 4 for the old GNU extension, version 5 for DWARF5. 363 To handle DW_FORM_strx[1234] we set the .str_offsets_base 364 from the given CU. 365 XXX We will need to deal with DW_MACRO_import_sup and change 366 out the dbg somehow for the DW_FORM_sec_offset to make sense. */ 367 Dwarf_CU fake_cu = { 368 .dbg = dbg, 369 .sec_idx = sec_index, 370 .version = table->version, 371 .offset_size = table->is_64bit ? 8 : 4, 372 .str_off_base = str_offsets_base_off (dbg, (cudie != NULL 373 ? cudie->cu: NULL)), 374 .startp = (void *) startp + offset, 375 .endp = (void *) endp, 376 }; 377 378 Dwarf_Attribute *attributes; 379 Dwarf_Attribute *attributesp = NULL; 380 Dwarf_Attribute nattributes[8]; 381 if (unlikely (proto->nforms > 8)) 382 { 383 attributesp = malloc (sizeof (Dwarf_Attribute) * proto->nforms); 384 if (attributesp == NULL) 385 { 386 __libdw_seterrno (DWARF_E_NOMEM); 387 return -1; 388 } 389 attributes = attributesp; 390 } 391 else 392 attributes = &nattributes[0]; 393 394 for (Dwarf_Word i = 0; i < proto->nforms; ++i) 395 { 396 /* We pretend this is a DW_AT[_GNU]_macros attribute so that 397 DW_FORM_sec_offset forms get correctly interpreted as 398 offset into .debug_macro. XXX Deal with DW_MACRO_import_sup 399 (swap .dbg) for DW_FORM_sec_offset? */ 400 attributes[i].code = (fake_cu.version == 4 ? DW_AT_GNU_macros 401 : DW_AT_macros); 402 attributes[i].form = proto->forms[i]; 403 attributes[i].valp = (void *) readp; 404 attributes[i].cu = &fake_cu; 405 406 /* We don't want forms that aren't allowed because they could 407 read from the "abbrev" like DW_FORM_implicit_const. */ 408 if (! libdw_valid_user_form (attributes[i].form)) 409 { 410 __libdw_seterrno (DWARF_E_INVALID_DWARF); 411 free (attributesp); 412 return -1; 413 } 414 415 size_t len = __libdw_form_val_len (&fake_cu, proto->forms[i], readp); 416 if (unlikely (len == (size_t) -1)) 417 { 418 free (attributesp); 419 return -1; 420 } 421 422 readp += len; 423 } 424 425 Dwarf_Macro macro = { 426 .table = table, 427 .opcode = opcode, 428 .attributes = attributes, 429 }; 430 431 int res = callback (¯o, arg); 432 if (unlikely (attributesp != NULL)) 433 free (attributesp); 434 435 if (res != DWARF_CB_OK) 436 return readp - startp; 437 } 438 439 return 0; 440 } 441 442 /* Token layout: 443 444 - The highest bit is used for distinguishing between callers that 445 know that opcode 0xff may have one of two incompatible meanings. 446 The mask that we use for selecting this bit is 447 DWARF_GETMACROS_START. 448 449 - The rest of the token (31 or 63 bits) encodes address inside the 450 macro unit. 451 452 Besides, token value of 0 signals end of iteration and -1 is 453 reserved for signaling errors. That means it's impossible to 454 represent maximum offset of a .debug_macro unit to new-style 455 callers (which in practice decreases the permissible macro unit 456 size by another 1 byte). */ 457 458 static ptrdiff_t 459 token_from_offset (ptrdiff_t offset, bool accept_0xff) 460 { 461 if (offset == -1 || offset == 0) 462 return offset; 463 464 /* Make sure the offset didn't overflow into the flag bit. */ 465 if ((offset & DWARF_GETMACROS_START) != 0) 466 { 467 __libdw_seterrno (DWARF_E_TOO_BIG); 468 return -1; 469 } 470 471 if (accept_0xff) 472 offset |= DWARF_GETMACROS_START; 473 474 return offset; 475 } 476 477 static ptrdiff_t 478 offset_from_token (ptrdiff_t token, bool *accept_0xffp) 479 { 480 *accept_0xffp = (token & DWARF_GETMACROS_START) != 0; 481 token &= ~DWARF_GETMACROS_START; 482 483 return token; 484 } 485 486 static ptrdiff_t 487 gnu_macros_getmacros_off (Dwarf *dbg, Dwarf_Off macoff, 488 int (*callback) (Dwarf_Macro *, void *), 489 void *arg, ptrdiff_t offset, bool accept_0xff, 490 Dwarf_Die *cudie) 491 { 492 assert (offset >= 0); 493 494 if (macoff >= dbg->sectiondata[IDX_debug_macro]->d_size) 495 { 496 __libdw_seterrno (DWARF_E_INVALID_OFFSET); 497 return -1; 498 } 499 500 return read_macros (dbg, IDX_debug_macro, macoff, 501 callback, arg, offset, accept_0xff, cudie); 502 } 503 504 static ptrdiff_t 505 macro_info_getmacros_off (Dwarf *dbg, Dwarf_Off macoff, 506 int (*callback) (Dwarf_Macro *, void *), 507 void *arg, ptrdiff_t offset, Dwarf_Die *cudie) 508 { 509 assert (offset >= 0); 510 511 return read_macros (dbg, IDX_debug_macinfo, macoff, 512 callback, arg, offset, true, cudie); 513 } 514 515 ptrdiff_t 516 dwarf_getmacros_off (Dwarf *dbg, Dwarf_Off macoff, 517 int (*callback) (Dwarf_Macro *, void *), 518 void *arg, ptrdiff_t token) 519 { 520 if (dbg == NULL) 521 { 522 __libdw_seterrno (DWARF_E_NO_DWARF); 523 return -1; 524 } 525 526 bool accept_0xff; 527 ptrdiff_t offset = offset_from_token (token, &accept_0xff); 528 assert (accept_0xff); 529 530 offset = gnu_macros_getmacros_off (dbg, macoff, callback, arg, offset, 531 accept_0xff, NULL); 532 533 return token_from_offset (offset, accept_0xff); 534 } 535 536 ptrdiff_t 537 dwarf_getmacros (Dwarf_Die *cudie, int (*callback) (Dwarf_Macro *, void *), 538 void *arg, ptrdiff_t token) 539 { 540 if (cudie == NULL) 541 { 542 __libdw_seterrno (DWARF_E_NO_DWARF); 543 return -1; 544 } 545 546 /* This function might be called from a code that expects to see 547 DW_MACINFO_* opcodes, not DW_MACRO_{GNU_,}* ones. It is fine to 548 serve most DW_MACRO_{GNU_,}* opcodes to such code, because those 549 whose values are the same as DW_MACINFO_* ones also have the same 550 behavior. It is not very likely that a .debug_macro section 551 would only use the part of opcode space that it shares with 552 .debug_macinfo, but it is possible. Serving the opcodes that are 553 only valid in DW_MACRO_{GNU_,}* domain is OK as well, because 554 clients in general need to be ready that newer standards define 555 more opcodes, and have coping mechanisms for unfamiliar opcodes. 556 557 The one exception to the above rule is opcode 0xff, which has 558 concrete semantics in .debug_macinfo, but falls into vendor block 559 in .debug_macro, and can be assigned to do whatever. There is 560 some small probability that the two opcodes would look 561 superficially similar enough that a client would be confused and 562 misbehave as a result. For this reason, we refuse to serve 563 through this interface 0xff's originating from .debug_macro 564 unless the TOKEN that we obtained indicates the call originates 565 from a new-style caller. See above for details on what 566 information is encoded into tokens. */ 567 568 bool accept_0xff; 569 ptrdiff_t offset = offset_from_token (token, &accept_0xff); 570 571 /* DW_AT_macro_info */ 572 if (dwarf_hasattr (cudie, DW_AT_macro_info)) 573 { 574 Dwarf_Word macoff; 575 if (get_offset_from (cudie, DW_AT_macro_info, &macoff) != 0) 576 return -1; 577 offset = macro_info_getmacros_off (cudie->cu->dbg, macoff, 578 callback, arg, offset, cudie); 579 } 580 else 581 { 582 /* DW_AT_GNU_macros, DW_AT_macros */ 583 Dwarf_Word macoff; 584 if (get_offset_from (cudie, DW_AT_GNU_macros, &macoff) != 0 585 && get_offset_from (cudie, DW_AT_macros, &macoff) != 0) 586 return -1; 587 offset = gnu_macros_getmacros_off (cudie->cu->dbg, macoff, 588 callback, arg, offset, accept_0xff, 589 cudie); 590 } 591 592 return token_from_offset (offset, accept_0xff); 593 } 594