1 /* 2 * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include <jni.h> 26 #include <unistd.h> 27 #include <fcntl.h> 28 #include <string.h> 29 #include <stdlib.h> 30 #include <stddef.h> 31 #include "libproc_impl.h" 32 #include "../../../../hotspot/share/include/cds.h" 33 34 #ifdef __APPLE__ 35 #include "sun_jvm_hotspot_debugger_amd64_AMD64ThreadContext.h" 36 #endif 37 38 // This file has the libproc implementation to read core files. 39 // For live processes, refer to ps_proc.c. Portions of this is adapted 40 // /modelled after Solaris libproc.so (in particular Pcore.c) 41 42 //---------------------------------------------------------------------- 43 // ps_prochandle cleanup helper functions 44 45 // close all file descriptors 46 static void close_files(struct ps_prochandle* ph) { 47 lib_info* lib = NULL; 48 49 // close core file descriptor 50 if (ph->core->core_fd >= 0) 51 close(ph->core->core_fd); 52 53 // close exec file descriptor 54 if (ph->core->exec_fd >= 0) 55 close(ph->core->exec_fd); 56 57 // close interp file descriptor 58 if (ph->core->interp_fd >= 0) 59 close(ph->core->interp_fd); 60 61 // close class share archive file 62 if (ph->core->classes_jsa_fd >= 0) 63 close(ph->core->classes_jsa_fd); 64 65 // close all library file descriptors 66 lib = ph->libs; 67 while (lib) { 68 int fd = lib->fd; 69 if (fd >= 0 && fd != ph->core->exec_fd) { 70 close(fd); 71 } 72 lib = lib->next; 73 } 74 } 75 76 // clean all map_info stuff 77 static void destroy_map_info(struct ps_prochandle* ph) { 78 map_info* map = ph->core->maps; 79 while (map) { 80 map_info* next = map->next; 81 free(map); 82 map = next; 83 } 84 85 if (ph->core->map_array) { 86 free(ph->core->map_array); 87 } 88 89 // Part of the class sharing workaround 90 map = ph->core->class_share_maps; 91 while (map) { 92 map_info* next = map->next; 93 free(map); 94 map = next; 95 } 96 } 97 98 // ps_prochandle operations 99 static void core_release(struct ps_prochandle* ph) { 100 if (ph->core) { 101 close_files(ph); 102 destroy_map_info(ph); 103 free(ph->core); 104 } 105 } 106 107 static map_info* allocate_init_map(int fd, off_t offset, uintptr_t vaddr, size_t memsz) { 108 map_info* map; 109 if ( (map = (map_info*) calloc(1, sizeof(map_info))) == NULL) { 110 print_debug("can't allocate memory for map_info\n"); 111 return NULL; 112 } 113 114 // initialize map 115 map->fd = fd; 116 map->offset = offset; 117 map->vaddr = vaddr; 118 map->memsz = memsz; 119 return map; 120 } 121 122 // add map info with given fd, offset, vaddr and memsz 123 static map_info* add_map_info(struct ps_prochandle* ph, int fd, off_t offset, 124 uintptr_t vaddr, size_t memsz) { 125 map_info* map; 126 if ((map = allocate_init_map(fd, offset, vaddr, memsz)) == NULL) { 127 return NULL; 128 } 129 130 // add this to map list 131 map->next = ph->core->maps; 132 ph->core->maps = map; 133 ph->core->num_maps++; 134 135 return map; 136 } 137 138 // Part of the class sharing workaround 139 static map_info* add_class_share_map_info(struct ps_prochandle* ph, off_t offset, 140 uintptr_t vaddr, size_t memsz) { 141 map_info* map; 142 if ((map = allocate_init_map(ph->core->classes_jsa_fd, 143 offset, vaddr, memsz)) == NULL) { 144 return NULL; 145 } 146 147 map->next = ph->core->class_share_maps; 148 ph->core->class_share_maps = map; 149 return map; 150 } 151 152 // Return the map_info for the given virtual address. We keep a sorted 153 // array of pointers in ph->map_array, so we can binary search. 154 static map_info* core_lookup(struct ps_prochandle *ph, uintptr_t addr) { 155 int mid, lo = 0, hi = ph->core->num_maps - 1; 156 map_info *mp; 157 158 while (hi - lo > 1) { 159 mid = (lo + hi) / 2; 160 if (addr >= ph->core->map_array[mid]->vaddr) { 161 lo = mid; 162 } else { 163 hi = mid; 164 } 165 } 166 167 if (addr < ph->core->map_array[hi]->vaddr) { 168 mp = ph->core->map_array[lo]; 169 } else { 170 mp = ph->core->map_array[hi]; 171 } 172 173 if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) { 174 return (mp); 175 } 176 177 178 // Part of the class sharing workaround 179 // Unfortunately, we have no way of detecting -Xshare state. 180 // Check out the share maps atlast, if we don't find anywhere. 181 // This is done this way so to avoid reading share pages 182 // ahead of other normal maps. For eg. with -Xshare:off we don't 183 // want to prefer class sharing data to data from core. 184 mp = ph->core->class_share_maps; 185 if (mp) { 186 print_debug("can't locate map_info at 0x%lx, trying class share maps\n", addr); 187 } 188 while (mp) { 189 if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) { 190 print_debug("located map_info at 0x%lx from class share maps\n", addr); 191 return (mp); 192 } 193 mp = mp->next; 194 } 195 196 print_debug("can't locate map_info at 0x%lx\n", addr); 197 return (NULL); 198 } 199 200 //--------------------------------------------------------------- 201 // Part of the class sharing workaround: 202 // 203 // With class sharing, pages are mapped from classes.jsa file. 204 // The read-only class sharing pages are mapped as MAP_SHARED, 205 // PROT_READ pages. These pages are not dumped into core dump. 206 // With this workaround, these pages are read from classes.jsa. 207 208 typedef struct CDSFileMapHeaderBase FileMapHeader; 209 210 static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) { 211 jboolean i; 212 if (ps_pread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) { 213 *pvalue = i; 214 return true; 215 } else { 216 return false; 217 } 218 } 219 220 static bool read_pointer(struct ps_prochandle* ph, uintptr_t addr, uintptr_t* pvalue) { 221 uintptr_t uip; 222 if (ps_pread(ph, (psaddr_t) addr, (char *)&uip, sizeof(uip)) == PS_OK) { 223 *pvalue = uip; 224 return true; 225 } else { 226 return false; 227 } 228 } 229 230 // used to read strings from debuggee 231 static bool read_string(struct ps_prochandle* ph, uintptr_t addr, char* buf, size_t size) { 232 size_t i = 0; 233 char c = ' '; 234 235 while (c != '\0') { 236 if (ps_pread(ph, (psaddr_t) addr, &c, sizeof(char)) != PS_OK) { 237 return false; 238 } 239 if (i < size - 1) { 240 buf[i] = c; 241 } else { 242 // smaller buffer 243 return false; 244 } 245 i++; addr++; 246 } 247 buf[i] = '\0'; 248 return true; 249 } 250 251 // mangled name of Arguments::SharedArchivePath 252 #define SHARED_ARCHIVE_PATH_SYM "__ZN9Arguments17SharedArchivePathE" 253 254 #ifdef __APPLE__ 255 #define USE_SHARED_SPACES_SYM "_UseSharedSpaces" 256 #define LIBJVM_NAME "/libjvm.dylib" 257 #else 258 #define USE_SHARED_SPACES_SYM "UseSharedSpaces" 259 #define LIBJVM_NAME "/libjvm.so" 260 #endif // __APPLE_ 261 262 static bool init_classsharing_workaround(struct ps_prochandle* ph) { 263 int m; 264 size_t n; 265 lib_info* lib = ph->libs; 266 while (lib != NULL) { 267 // we are iterating over shared objects from the core dump. look for 268 // libjvm.so. 269 const char *jvm_name = 0; 270 if ((jvm_name = strstr(lib->name, LIBJVM_NAME)) != 0) { 271 char classes_jsa[PATH_MAX]; 272 FileMapHeader header; 273 int fd = -1; 274 uintptr_t base = 0, useSharedSpacesAddr = 0; 275 uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0; 276 jboolean useSharedSpaces = 0; 277 278 memset(classes_jsa, 0, sizeof(classes_jsa)); 279 jvm_name = lib->name; 280 useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM); 281 if (useSharedSpacesAddr == 0) { 282 print_debug("can't lookup 'UseSharedSpaces' flag\n"); 283 return false; 284 } 285 286 // Hotspot vm types are not exported to build this library. So 287 // using equivalent type jboolean to read the value of 288 // UseSharedSpaces which is same as hotspot type "bool". 289 if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) { 290 print_debug("can't read the value of 'UseSharedSpaces' flag\n"); 291 return false; 292 } 293 294 if ((int)useSharedSpaces == 0) { 295 print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n"); 296 return true; 297 } 298 299 sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM); 300 if (sharedArchivePathAddrAddr == 0) { 301 print_debug("can't lookup shared archive path symbol\n"); 302 return false; 303 } 304 305 if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) { 306 print_debug("can't read shared archive path pointer\n"); 307 return false; 308 } 309 310 if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) { 311 print_debug("can't read shared archive path value\n"); 312 return false; 313 } 314 315 print_debug("looking for %s\n", classes_jsa); 316 // open the class sharing archive file 317 fd = pathmap_open(classes_jsa); 318 if (fd < 0) { 319 print_debug("can't open %s!\n", classes_jsa); 320 ph->core->classes_jsa_fd = -1; 321 return false; 322 } else { 323 print_debug("opened %s\n", classes_jsa); 324 } 325 326 // read FileMapHeader from the file 327 memset(&header, 0, sizeof(FileMapHeader)); 328 if ((n = read(fd, &header, sizeof(FileMapHeader))) 329 != sizeof(FileMapHeader)) { 330 print_debug("can't read shared archive file map header from %s\n", classes_jsa); 331 close(fd); 332 return false; 333 } 334 335 // check file magic 336 if (header._magic != CDS_ARCHIVE_MAGIC) { 337 print_debug("%s has bad shared archive file magic number 0x%x, expecing 0x%x\n", 338 classes_jsa, header._magic, CDS_ARCHIVE_MAGIC); 339 close(fd); 340 return false; 341 } 342 343 // check version 344 if (header._version != CURRENT_CDS_ARCHIVE_VERSION) { 345 print_debug("%s has wrong shared archive file version %d, expecting %d\n", 346 classes_jsa, header._version, CURRENT_CDS_ARCHIVE_VERSION); 347 close(fd); 348 return false; 349 } 350 351 ph->core->classes_jsa_fd = fd; 352 // add read-only maps from classes.jsa to the list of maps 353 for (m = 0; m < NUM_CDS_REGIONS; m++) { 354 if (header._space[m]._read_only) { 355 base = (uintptr_t) header._space[m]._addr._base; 356 // no need to worry about the fractional pages at-the-end. 357 // possible fractional pages are handled by core_read_data. 358 add_class_share_map_info(ph, (off_t) header._space[m]._file_offset, 359 base, (size_t) header._space[m]._used); 360 print_debug("added a share archive map at 0x%lx\n", base); 361 } 362 } 363 return true; 364 } 365 lib = lib->next; 366 } 367 return true; 368 } 369 370 //--------------------------------------------------------------------------- 371 // functions to handle map_info 372 373 // Order mappings based on virtual address. We use this function as the 374 // callback for sorting the array of map_info pointers. 375 static int core_cmp_mapping(const void *lhsp, const void *rhsp) 376 { 377 const map_info *lhs = *((const map_info **)lhsp); 378 const map_info *rhs = *((const map_info **)rhsp); 379 380 if (lhs->vaddr == rhs->vaddr) { 381 return (0); 382 } 383 384 return (lhs->vaddr < rhs->vaddr ? -1 : 1); 385 } 386 387 // we sort map_info by starting virtual address so that we can do 388 // binary search to read from an address. 389 static bool sort_map_array(struct ps_prochandle* ph) { 390 size_t num_maps = ph->core->num_maps; 391 map_info* map = ph->core->maps; 392 int i = 0; 393 394 // allocate map_array 395 map_info** array; 396 if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) { 397 print_debug("can't allocate memory for map array\n"); 398 return false; 399 } 400 401 // add maps to array 402 while (map) { 403 array[i] = map; 404 i++; 405 map = map->next; 406 } 407 408 // sort is called twice. If this is second time, clear map array 409 if (ph->core->map_array) { 410 free(ph->core->map_array); 411 } 412 ph->core->map_array = array; 413 // sort the map_info array by base virtual address. 414 qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*), 415 core_cmp_mapping); 416 417 // print map 418 if (is_debug()) { 419 int j = 0; 420 print_debug("---- sorted virtual address map ----\n"); 421 for (j = 0; j < ph->core->num_maps; j++) { 422 print_debug("base = 0x%lx\tsize = %d\n", ph->core->map_array[j]->vaddr, 423 ph->core->map_array[j]->memsz); 424 } 425 } 426 427 return true; 428 } 429 430 #ifndef MIN 431 #define MIN(x, y) (((x) < (y))? (x): (y)) 432 #endif 433 434 static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) { 435 ssize_t resid = size; 436 int page_size=sysconf(_SC_PAGE_SIZE); 437 while (resid != 0) { 438 map_info *mp = core_lookup(ph, addr); 439 uintptr_t mapoff; 440 ssize_t len, rem; 441 off_t off; 442 int fd; 443 444 if (mp == NULL) { 445 break; /* No mapping for this address */ 446 } 447 448 fd = mp->fd; 449 mapoff = addr - mp->vaddr; 450 len = MIN(resid, mp->memsz - mapoff); 451 off = mp->offset + mapoff; 452 453 if ((len = pread(fd, buf, len, off)) <= 0) { 454 break; 455 } 456 457 resid -= len; 458 addr += len; 459 buf = (char *)buf + len; 460 461 // mappings always start at page boundary. But, may end in fractional 462 // page. fill zeros for possible fractional page at the end of a mapping. 463 rem = mp->memsz % page_size; 464 if (rem > 0) { 465 rem = page_size - rem; 466 len = MIN(resid, rem); 467 resid -= len; 468 addr += len; 469 // we are not assuming 'buf' to be zero initialized. 470 memset(buf, 0, len); 471 buf += len; 472 } 473 } 474 475 if (resid) { 476 print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n", 477 size, addr, resid); 478 return false; 479 } else { 480 return true; 481 } 482 } 483 484 // null implementation for write 485 static bool core_write_data(struct ps_prochandle* ph, 486 uintptr_t addr, const char *buf , size_t size) { 487 return false; 488 } 489 490 static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id, 491 struct reg* regs) { 492 // for core we have cached the lwp regs after segment parsed 493 sa_thread_info* thr = ph->threads; 494 while (thr) { 495 if (thr->lwp_id == lwp_id) { 496 memcpy(regs, &thr->regs, sizeof(struct reg)); 497 return true; 498 } 499 thr = thr->next; 500 } 501 return false; 502 } 503 504 static bool core_get_lwp_info(struct ps_prochandle *ph, lwpid_t id, void *info) { 505 print_debug("core_get_lwp_info not implemented\n"); 506 return false; 507 } 508 509 static ps_prochandle_ops core_ops = { 510 .release= core_release, 511 .p_pread= core_read_data, 512 .p_pwrite= core_write_data, 513 .get_lwp_regs= core_get_lwp_regs, 514 .get_lwp_info= core_get_lwp_info 515 }; 516 517 // from this point, mainly two blocks divided by def __APPLE__ 518 // one for Macosx, the other for regular Bsd 519 520 #ifdef __APPLE__ 521 522 void print_thread(sa_thread_info *threadinfo) { 523 print_debug("thread added: %d\n", threadinfo->lwp_id); 524 print_debug("registers:\n"); 525 print_debug(" r_r15: 0x%" PRIx64 "\n", threadinfo->regs.r_r15); 526 print_debug(" r_r14: 0x%" PRIx64 "\n", threadinfo->regs.r_r14); 527 print_debug(" r_r13: 0x%" PRIx64 "\n", threadinfo->regs.r_r13); 528 print_debug(" r_r12: 0x%" PRIx64 "\n", threadinfo->regs.r_r12); 529 print_debug(" r_r11: 0x%" PRIx64 "\n", threadinfo->regs.r_r11); 530 print_debug(" r_r10: 0x%" PRIx64 "\n", threadinfo->regs.r_r10); 531 print_debug(" r_r9: 0x%" PRIx64 "\n", threadinfo->regs.r_r9); 532 print_debug(" r_r8: 0x%" PRIx64 "\n", threadinfo->regs.r_r8); 533 print_debug(" r_rdi: 0x%" PRIx64 "\n", threadinfo->regs.r_rdi); 534 print_debug(" r_rsi: 0x%" PRIx64 "\n", threadinfo->regs.r_rsi); 535 print_debug(" r_rbp: 0x%" PRIx64 "\n", threadinfo->regs.r_rbp); 536 print_debug(" r_rbx: 0x%" PRIx64 "\n", threadinfo->regs.r_rbx); 537 print_debug(" r_rdx: 0x%" PRIx64 "\n", threadinfo->regs.r_rdx); 538 print_debug(" r_rcx: 0x%" PRIx64 "\n", threadinfo->regs.r_rcx); 539 print_debug(" r_rax: 0x%" PRIx64 "\n", threadinfo->regs.r_rax); 540 print_debug(" r_fs: 0x%" PRIx32 "\n", threadinfo->regs.r_fs); 541 print_debug(" r_gs: 0x%" PRIx32 "\n", threadinfo->regs.r_gs); 542 print_debug(" r_rip 0x%" PRIx64 "\n", threadinfo->regs.r_rip); 543 print_debug(" r_cs: 0x%" PRIx64 "\n", threadinfo->regs.r_cs); 544 print_debug(" r_rsp: 0x%" PRIx64 "\n", threadinfo->regs.r_rsp); 545 print_debug(" r_rflags: 0x%" PRIx64 "\n", threadinfo->regs.r_rflags); 546 } 547 548 // read all segments64 commands from core file 549 // read all thread commands from core file 550 static bool read_core_segments(struct ps_prochandle* ph) { 551 int i = 0; 552 int num_threads = 0; 553 int fd = ph->core->core_fd; 554 off_t offset = 0; 555 mach_header_64 fhead; 556 load_command lcmd; 557 segment_command_64 segcmd; 558 // thread_command thrcmd; 559 560 lseek(fd, offset, SEEK_SET); 561 if(read(fd, (void *)&fhead, sizeof(mach_header_64)) != sizeof(mach_header_64)) { 562 goto err; 563 } 564 print_debug("total commands: %d\n", fhead.ncmds); 565 offset += sizeof(mach_header_64); 566 for (i = 0; i < fhead.ncmds; i++) { 567 lseek(fd, offset, SEEK_SET); 568 if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) { 569 goto err; 570 } 571 offset += lcmd.cmdsize; // next command position 572 if (lcmd.cmd == LC_SEGMENT_64) { 573 lseek(fd, -sizeof(load_command), SEEK_CUR); 574 if (read(fd, (void *)&segcmd, sizeof(segment_command_64)) != sizeof(segment_command_64)) { 575 print_debug("failed to read LC_SEGMENT_64 i = %d!\n", i); 576 goto err; 577 } 578 if (add_map_info(ph, fd, segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize) == NULL) { 579 print_debug("Failed to add map_info at i = %d\n", i); 580 goto err; 581 } 582 print_debug("segment added: %" PRIu64 " 0x%" PRIx64 " %d\n", 583 segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize); 584 } else if (lcmd.cmd == LC_THREAD || lcmd.cmd == LC_UNIXTHREAD) { 585 typedef struct thread_fc { 586 uint32_t flavor; 587 uint32_t count; 588 } thread_fc; 589 thread_fc fc; 590 uint32_t size = sizeof(load_command); 591 while (size < lcmd.cmdsize) { 592 if (read(fd, (void *)&fc, sizeof(thread_fc)) != sizeof(thread_fc)) { 593 printf("Reading flavor, count failed.\n"); 594 goto err; 595 } 596 size += sizeof(thread_fc); 597 if (fc.flavor == x86_THREAD_STATE) { 598 x86_thread_state_t thrstate; 599 if (read(fd, (void *)&thrstate, sizeof(x86_thread_state_t)) != sizeof(x86_thread_state_t)) { 600 printf("Reading flavor, count failed.\n"); 601 goto err; 602 } 603 size += sizeof(x86_thread_state_t); 604 // create thread info list, update lwp_id later 605 sa_thread_info* newthr = add_thread_info(ph, (pthread_t) -1, (lwpid_t) num_threads++); 606 if (newthr == NULL) { 607 printf("create thread_info failed\n"); 608 goto err; 609 } 610 611 // note __DARWIN_UNIX03 depengs on other definitions 612 #if __DARWIN_UNIX03 613 #define get_register_v(regst, regname) \ 614 regst.uts.ts64.__##regname 615 #else 616 #define get_register_v(regst, regname) \ 617 regst.uts.ts64.##regname 618 #endif // __DARWIN_UNIX03 619 newthr->regs.r_rax = get_register_v(thrstate, rax); 620 newthr->regs.r_rbx = get_register_v(thrstate, rbx); 621 newthr->regs.r_rcx = get_register_v(thrstate, rcx); 622 newthr->regs.r_rdx = get_register_v(thrstate, rdx); 623 newthr->regs.r_rdi = get_register_v(thrstate, rdi); 624 newthr->regs.r_rsi = get_register_v(thrstate, rsi); 625 newthr->regs.r_rbp = get_register_v(thrstate, rbp); 626 newthr->regs.r_rsp = get_register_v(thrstate, rsp); 627 newthr->regs.r_r8 = get_register_v(thrstate, r8); 628 newthr->regs.r_r9 = get_register_v(thrstate, r9); 629 newthr->regs.r_r10 = get_register_v(thrstate, r10); 630 newthr->regs.r_r11 = get_register_v(thrstate, r11); 631 newthr->regs.r_r12 = get_register_v(thrstate, r12); 632 newthr->regs.r_r13 = get_register_v(thrstate, r13); 633 newthr->regs.r_r14 = get_register_v(thrstate, r14); 634 newthr->regs.r_r15 = get_register_v(thrstate, r15); 635 newthr->regs.r_rip = get_register_v(thrstate, rip); 636 newthr->regs.r_rflags = get_register_v(thrstate, rflags); 637 newthr->regs.r_cs = get_register_v(thrstate, cs); 638 newthr->regs.r_fs = get_register_v(thrstate, fs); 639 newthr->regs.r_gs = get_register_v(thrstate, gs); 640 print_thread(newthr); 641 } else if (fc.flavor == x86_FLOAT_STATE) { 642 x86_float_state_t flstate; 643 if (read(fd, (void *)&flstate, sizeof(x86_float_state_t)) != sizeof(x86_float_state_t)) { 644 print_debug("Reading flavor, count failed.\n"); 645 goto err; 646 } 647 size += sizeof(x86_float_state_t); 648 } else if (fc.flavor == x86_EXCEPTION_STATE) { 649 x86_exception_state_t excpstate; 650 if (read(fd, (void *)&excpstate, sizeof(x86_exception_state_t)) != sizeof(x86_exception_state_t)) { 651 printf("Reading flavor, count failed.\n"); 652 goto err; 653 } 654 size += sizeof(x86_exception_state_t); 655 } 656 } 657 } 658 } 659 return true; 660 err: 661 return false; 662 } 663 664 /**local function **/ 665 bool exists(const char *fname) { 666 return access(fname, F_OK) == 0; 667 } 668 669 // we check: 1. lib 670 // 2. lib/server 671 // 3. jre/lib 672 // 4. jre/lib/server 673 // from: 1. exe path 674 // 2. JAVA_HOME 675 // 3. DYLD_LIBRARY_PATH 676 static bool get_real_path(struct ps_prochandle* ph, char *rpath) { 677 /** check if they exist in JAVA ***/ 678 char* execname = ph->core->exec_path; 679 char filepath[4096]; 680 char* filename = strrchr(rpath, '/'); // like /libjvm.dylib 681 if (filename == NULL) { 682 return false; 683 } 684 685 char* posbin = strstr(execname, "/bin/java"); 686 if (posbin != NULL) { 687 memcpy(filepath, execname, posbin - execname); // not include trailing '/' 688 filepath[posbin - execname] = '\0'; 689 } else { 690 char* java_home = getenv("JAVA_HOME"); 691 if (java_home != NULL) { 692 strcpy(filepath, java_home); 693 } else { 694 char* dyldpath = getenv("DYLD_LIBRARY_PATH"); 695 char* dypath = strtok(dyldpath, ":"); 696 while (dypath != NULL) { 697 strcpy(filepath, dypath); 698 strcat(filepath, filename); 699 if (exists(filepath)) { 700 strcpy(rpath, filepath); 701 return true; 702 } 703 dypath = strtok(dyldpath, ":"); 704 } 705 // not found 706 return false; 707 } 708 } 709 // for exec and java_home, jdkpath now is filepath 710 size_t filepath_base_size = strlen(filepath); 711 712 // first try /lib/ and /lib/server 713 strcat(filepath, "/lib"); 714 strcat(filepath, filename); 715 if (exists(filepath)) { 716 strcpy(rpath, filepath); 717 return true; 718 } 719 char* pos = strstr(filepath, filename); // like /libjvm.dylib 720 *pos = '\0'; 721 strcat(filepath, "/server"); 722 strcat(filepath, filename); 723 if (exists(filepath)) { 724 strcpy(rpath, filepath); 725 return true; 726 } 727 728 // then try /jre/lib/ and /jre/lib/server 729 filepath[filepath_base_size] = '\0'; 730 strcat(filepath, "/jre/lib"); 731 strcat(filepath, filename); 732 if (exists(filepath)) { 733 strcpy(rpath, filepath); 734 return true; 735 } 736 pos = strstr(filepath, filename); 737 *pos = '\0'; 738 strcat(filepath, "/server"); 739 strcat(filepath, filename); 740 if (exists(filepath)) { 741 strcpy(rpath, filepath); 742 return true; 743 } 744 745 return false; 746 } 747 748 static bool read_shared_lib_info(struct ps_prochandle* ph) { 749 static int pagesize = 0; 750 int fd = ph->core->core_fd; 751 int i = 0, j; 752 uint32_t v; 753 mach_header_64 header; // used to check if a file header in segment 754 load_command lcmd; 755 dylib_command dylibcmd; 756 757 char name[BUF_SIZE]; // use to store name 758 759 if (pagesize == 0) { 760 pagesize = getpagesize(); 761 print_debug("page size is %d\n", pagesize); 762 } 763 for (j = 0; j < ph->core->num_maps; j++) { 764 map_info *iter = ph->core->map_array[j]; // head 765 off_t fpos = iter->offset; 766 if (iter->fd != fd) { 767 // only search core file! 768 continue; 769 } 770 print_debug("map_info %d: vmaddr = 0x%016" PRIx64 " fileoff = %" PRIu64 " vmsize = %" PRIu64 "\n", 771 j, iter->vaddr, iter->offset, iter->memsz); 772 lseek(fd, fpos, SEEK_SET); 773 // we assume .dylib loaded at segment address --- which is true for JVM libraries 774 // multiple files may be loaded in one segment. 775 // if first word is not a magic word, means this segment does not contain lib file. 776 if (read(fd, (void *)&v, sizeof(uint32_t)) == sizeof(uint32_t)) { 777 if (v != MH_MAGIC_64) { 778 continue; 779 } 780 } else { 781 // may be encountered last map, which is not readable 782 continue; 783 } 784 while (ltell(fd) - iter->offset < iter->memsz) { 785 lseek(fd, fpos, SEEK_SET); 786 if (read(fd, (void *)&v, sizeof(uint32_t)) != sizeof(uint32_t)) { 787 break; 788 } 789 if (v != MH_MAGIC_64) { 790 fpos = (ltell(fd) + pagesize -1)/pagesize * pagesize; 791 continue; 792 } 793 lseek(fd, -sizeof(uint32_t), SEEK_CUR); 794 // this is the file begining to core file. 795 if (read(fd, (void *)&header, sizeof(mach_header_64)) != sizeof(mach_header_64)) { 796 goto err; 797 } 798 fpos = ltell(fd); 799 800 // found a mach-o file in this segment 801 for (i = 0; i < header.ncmds; i++) { 802 // read commands in this "file" 803 // LC_ID_DYLIB is the file itself for a .dylib 804 lseek(fd, fpos, SEEK_SET); 805 if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) { 806 return false; // error 807 } 808 fpos += lcmd.cmdsize; // next command position 809 // make sure still within seg size. 810 if (fpos - lcmd.cmdsize - iter->offset > iter->memsz) { 811 print_debug("Warning: out of segement limit: %ld \n", fpos - lcmd.cmdsize - iter->offset); 812 break; // no need to iterate all commands 813 } 814 if (lcmd.cmd == LC_ID_DYLIB) { 815 lseek(fd, -sizeof(load_command), SEEK_CUR); 816 if (read(fd, (void *)&dylibcmd, sizeof(dylib_command)) != sizeof(dylib_command)) { 817 return false; 818 } 819 /**** name stored at dylib_command.dylib.name.offset, is a C string */ 820 lseek(fd, dylibcmd.dylib.name.offset - sizeof(dylib_command), SEEK_CUR); 821 int j = 0; 822 while (j < BUF_SIZE) { 823 read(fd, (void *)(name + j), sizeof(char)); 824 if (name[j] == '\0') break; 825 j++; 826 } 827 print_debug("%s\n", name); 828 // changed name from @rpath/xxxx.dylib to real path 829 if (strrchr(name, '@')) { 830 get_real_path(ph, name); 831 print_debug("get_real_path returned: %s\n", name); 832 } 833 add_lib_info(ph, name, iter->vaddr); 834 break; 835 } 836 } 837 // done with the file, advanced to next page to search more files 838 fpos = (ltell(fd) + pagesize - 1) / pagesize * pagesize; 839 } 840 } 841 return true; 842 err: 843 return false; 844 } 845 846 bool read_macho64_header(int fd, mach_header_64* core_header) { 847 bool is_macho = false; 848 if (fd < 0) return false; 849 off_t pos = ltell(fd); 850 lseek(fd, 0, SEEK_SET); 851 if (read(fd, (void *)core_header, sizeof(mach_header_64)) != sizeof(mach_header_64)) { 852 is_macho = false; 853 } else { 854 is_macho = (core_header->magic == MH_MAGIC_64 || core_header->magic == MH_CIGAM_64); 855 } 856 lseek(fd, pos, SEEK_SET); 857 return is_macho; 858 } 859 860 // the one and only one exposed stuff from this file 861 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) { 862 mach_header_64 core_header; 863 mach_header_64 exec_header; 864 865 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle)); 866 if (ph == NULL) { 867 print_debug("cant allocate ps_prochandle\n"); 868 return NULL; 869 } 870 871 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) { 872 free(ph); 873 print_debug("can't allocate ps_prochandle\n"); 874 return NULL; 875 } 876 877 // initialize ph 878 ph->ops = &core_ops; 879 ph->core->core_fd = -1; 880 ph->core->exec_fd = -1; 881 ph->core->interp_fd = -1; 882 883 print_debug("exec: %s core: %s", exec_file, core_file); 884 885 strncpy(ph->core->exec_path, exec_file, sizeof(ph->core->exec_path)); 886 887 // open the core file 888 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) { 889 print_error("can't open core file\n"); 890 goto err; 891 } 892 893 // read core file header 894 if (read_macho64_header(ph->core->core_fd, &core_header) != true || core_header.filetype != MH_CORE) { 895 print_debug("core file is not a valid Mach-O file\n"); 896 goto err; 897 } 898 899 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) { 900 print_error("can't open executable file\n"); 901 goto err; 902 } 903 904 if (read_macho64_header(ph->core->exec_fd, &exec_header) != true || 905 exec_header.filetype != MH_EXECUTE) { 906 print_error("executable file is not a valid Mach-O file\n"); 907 goto err; 908 } 909 910 // process core file segments 911 if (read_core_segments(ph) != true) { 912 print_error("failed to read core segments\n"); 913 goto err; 914 } 915 916 // allocate and sort maps into map_array, we need to do this 917 // here because read_shared_lib_info needs to read from debuggee 918 // address space 919 if (sort_map_array(ph) != true) { 920 print_error("failed to sort segment map array\n"); 921 goto err; 922 } 923 924 if (read_shared_lib_info(ph) != true) { 925 print_error("failed to read libraries\n"); 926 goto err; 927 } 928 929 // sort again because we have added more mappings from shared objects 930 if (sort_map_array(ph) != true) { 931 print_error("failed to sort segment map array\n"); 932 goto err; 933 } 934 935 if (init_classsharing_workaround(ph) != true) { 936 print_error("failed to workaround classshareing\n"); 937 goto err; 938 } 939 940 print_debug("Leave Pgrab_core\n"); 941 return ph; 942 943 err: 944 Prelease(ph); 945 return NULL; 946 } 947 948 #else // __APPLE__ (none macosx) 949 950 // read regs and create thread from core file 951 static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) { 952 // we have to read prstatus_t from buf 953 // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t"); 954 prstatus_t* prstat = (prstatus_t*) buf; 955 sa_thread_info* newthr; 956 print_debug("got integer regset for lwp %d\n", prstat->pr_pid); 957 // we set pthread_t to -1 for core dump 958 if((newthr = add_thread_info(ph, (pthread_t) -1, prstat->pr_pid)) == NULL) 959 return false; 960 961 // copy regs 962 memcpy(&newthr->regs, &prstat->pr_reg, sizeof(struct reg)); 963 964 if (is_debug()) { 965 print_debug("integer regset\n"); 966 #ifdef i386 967 // print the regset 968 print_debug("\teax = 0x%x\n", newthr->regs.r_eax); 969 print_debug("\tebx = 0x%x\n", newthr->regs.r_ebx); 970 print_debug("\tecx = 0x%x\n", newthr->regs.r_ecx); 971 print_debug("\tedx = 0x%x\n", newthr->regs.r_edx); 972 print_debug("\tesp = 0x%x\n", newthr->regs.r_esp); 973 print_debug("\tebp = 0x%x\n", newthr->regs.r_ebp); 974 print_debug("\tesi = 0x%x\n", newthr->regs.r_esi); 975 print_debug("\tedi = 0x%x\n", newthr->regs.r_edi); 976 print_debug("\teip = 0x%x\n", newthr->regs.r_eip); 977 #endif 978 979 #if defined(amd64) || defined(x86_64) 980 // print the regset 981 print_debug("\tr15 = 0x%lx\n", newthr->regs.r_r15); 982 print_debug("\tr14 = 0x%lx\n", newthr->regs.r_r14); 983 print_debug("\tr13 = 0x%lx\n", newthr->regs.r_r13); 984 print_debug("\tr12 = 0x%lx\n", newthr->regs.r_r12); 985 print_debug("\trbp = 0x%lx\n", newthr->regs.r_rbp); 986 print_debug("\trbx = 0x%lx\n", newthr->regs.r_rbx); 987 print_debug("\tr11 = 0x%lx\n", newthr->regs.r_r11); 988 print_debug("\tr10 = 0x%lx\n", newthr->regs.r_r10); 989 print_debug("\tr9 = 0x%lx\n", newthr->regs.r_r9); 990 print_debug("\tr8 = 0x%lx\n", newthr->regs.r_r8); 991 print_debug("\trax = 0x%lx\n", newthr->regs.r_rax); 992 print_debug("\trcx = 0x%lx\n", newthr->regs.r_rcx); 993 print_debug("\trdx = 0x%lx\n", newthr->regs.r_rdx); 994 print_debug("\trsi = 0x%lx\n", newthr->regs.r_rsi); 995 print_debug("\trdi = 0x%lx\n", newthr->regs.r_rdi); 996 //print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax); 997 print_debug("\trip = 0x%lx\n", newthr->regs.r_rip); 998 print_debug("\tcs = 0x%lx\n", newthr->regs.r_cs); 999 //print_debug("\teflags = 0x%lx\n", newthr->regs.eflags); 1000 print_debug("\trsp = 0x%lx\n", newthr->regs.r_rsp); 1001 print_debug("\tss = 0x%lx\n", newthr->regs.r_ss); 1002 //print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base); 1003 //print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base); 1004 //print_debug("\tds = 0x%lx\n", newthr->regs.ds); 1005 //print_debug("\tes = 0x%lx\n", newthr->regs.es); 1006 //print_debug("\tfs = 0x%lx\n", newthr->regs.fs); 1007 //print_debug("\tgs = 0x%lx\n", newthr->regs.gs); 1008 #endif 1009 } 1010 1011 return true; 1012 } 1013 1014 #define ROUNDUP(x, y) ((((x)+((y)-1))/(y))*(y)) 1015 1016 // read NT_PRSTATUS entries from core NOTE segment 1017 static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) { 1018 char* buf = NULL; 1019 char* p = NULL; 1020 size_t size = note_phdr->p_filesz; 1021 1022 // we are interested in just prstatus entries. we will ignore the rest. 1023 // Advance the seek pointer to the start of the PT_NOTE data 1024 if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) { 1025 print_debug("failed to lseek to PT_NOTE data\n"); 1026 return false; 1027 } 1028 1029 // Now process the PT_NOTE structures. Each one is preceded by 1030 // an Elf{32/64}_Nhdr structure describing its type and size. 1031 if ( (buf = (char*) malloc(size)) == NULL) { 1032 print_debug("can't allocate memory for reading core notes\n"); 1033 goto err; 1034 } 1035 1036 // read notes into buffer 1037 if (read(ph->core->core_fd, buf, size) != size) { 1038 print_debug("failed to read notes, core file must have been truncated\n"); 1039 goto err; 1040 } 1041 1042 p = buf; 1043 while (p < buf + size) { 1044 ELF_NHDR* notep = (ELF_NHDR*) p; 1045 char* descdata = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4); 1046 print_debug("Note header with n_type = %d and n_descsz = %u\n", 1047 notep->n_type, notep->n_descsz); 1048 1049 if (notep->n_type == NT_PRSTATUS) { 1050 if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) { 1051 return false; 1052 } 1053 } 1054 p = descdata + ROUNDUP(notep->n_descsz, 4); 1055 } 1056 1057 free(buf); 1058 return true; 1059 1060 err: 1061 if (buf) free(buf); 1062 return false; 1063 } 1064 1065 // read all segments from core file 1066 static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) { 1067 int i = 0; 1068 ELF_PHDR* phbuf = NULL; 1069 ELF_PHDR* core_php = NULL; 1070 1071 if ((phbuf = read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL) 1072 return false; 1073 1074 /* 1075 * Now iterate through the program headers in the core file. 1076 * We're interested in two types of Phdrs: PT_NOTE (which 1077 * contains a set of saved /proc structures), and PT_LOAD (which 1078 * represents a memory mapping from the process's address space). 1079 * 1080 * Difference b/w Solaris PT_NOTE and Linux/BSD PT_NOTE: 1081 * 1082 * In Solaris there are two PT_NOTE segments the first PT_NOTE (if present) 1083 * contains /proc structs in the pre-2.6 unstructured /proc format. the last 1084 * PT_NOTE has data in new /proc format. 1085 * 1086 * In Solaris, there is only one pstatus (process status). pstatus contains 1087 * integer register set among other stuff. For each LWP, we have one lwpstatus 1088 * entry that has integer regset for that LWP. 1089 * 1090 * Linux threads are actually 'clone'd processes. To support core analysis 1091 * of "multithreaded" process, Linux creates more than one pstatus (called 1092 * "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one 1093 * "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular 1094 * function "elf_core_dump". 1095 */ 1096 1097 for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) { 1098 switch (core_php->p_type) { 1099 case PT_NOTE: 1100 if (core_handle_note(ph, core_php) != true) { 1101 goto err; 1102 } 1103 break; 1104 1105 case PT_LOAD: { 1106 if (core_php->p_filesz != 0) { 1107 if (add_map_info(ph, ph->core->core_fd, core_php->p_offset, 1108 core_php->p_vaddr, core_php->p_filesz) == NULL) goto err; 1109 } 1110 break; 1111 } 1112 } 1113 1114 core_php++; 1115 } 1116 1117 free(phbuf); 1118 return true; 1119 err: 1120 free(phbuf); 1121 return false; 1122 } 1123 1124 // read segments of a shared object 1125 static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) { 1126 int i = 0; 1127 ELF_PHDR* phbuf; 1128 ELF_PHDR* lib_php = NULL; 1129 1130 int page_size=sysconf(_SC_PAGE_SIZE); 1131 1132 if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL) { 1133 return false; 1134 } 1135 1136 // we want to process only PT_LOAD segments that are not writable. 1137 // i.e., text segments. The read/write/exec (data) segments would 1138 // have been already added from core file segments. 1139 for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) { 1140 if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) { 1141 1142 uintptr_t target_vaddr = lib_php->p_vaddr + lib_base; 1143 map_info *existing_map = core_lookup(ph, target_vaddr); 1144 1145 if (existing_map == NULL){ 1146 if (add_map_info(ph, lib_fd, lib_php->p_offset, 1147 target_vaddr, lib_php->p_filesz) == NULL) { 1148 goto err; 1149 } 1150 } else { 1151 if ((existing_map->memsz != page_size) && 1152 (existing_map->fd != lib_fd) && 1153 (existing_map->memsz != lib_php->p_filesz)){ 1154 1155 print_debug("address conflict @ 0x%lx (size = %ld, flags = %d\n)", 1156 target_vaddr, lib_php->p_filesz, lib_php->p_flags); 1157 goto err; 1158 } 1159 1160 /* replace PT_LOAD segment with library segment */ 1161 print_debug("overwrote with new address mapping (memsz %ld -> %ld)\n", 1162 existing_map->memsz, lib_php->p_filesz); 1163 1164 existing_map->fd = lib_fd; 1165 existing_map->offset = lib_php->p_offset; 1166 existing_map->memsz = lib_php->p_filesz; 1167 } 1168 } 1169 1170 lib_php++; 1171 } 1172 1173 free(phbuf); 1174 return true; 1175 err: 1176 free(phbuf); 1177 return false; 1178 } 1179 1180 // process segments from interpreter (ld.so or ld-linux.so or ld-elf.so) 1181 static bool read_interp_segments(struct ps_prochandle* ph) { 1182 ELF_EHDR interp_ehdr; 1183 1184 if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) { 1185 print_debug("interpreter is not a valid ELF file\n"); 1186 return false; 1187 } 1188 1189 if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) { 1190 print_debug("can't read segments of interpreter\n"); 1191 return false; 1192 } 1193 1194 return true; 1195 } 1196 1197 // process segments of a a.out 1198 static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) { 1199 int i = 0; 1200 ELF_PHDR* phbuf = NULL; 1201 ELF_PHDR* exec_php = NULL; 1202 1203 if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL) 1204 return false; 1205 1206 for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) { 1207 switch (exec_php->p_type) { 1208 1209 // add mappings for PT_LOAD segments 1210 case PT_LOAD: { 1211 // add only non-writable segments of non-zero filesz 1212 if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) { 1213 if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err; 1214 } 1215 break; 1216 } 1217 1218 // read the interpreter and it's segments 1219 case PT_INTERP: { 1220 char interp_name[BUF_SIZE]; 1221 1222 pread(ph->core->exec_fd, interp_name, MIN(exec_php->p_filesz, BUF_SIZE), exec_php->p_offset); 1223 print_debug("ELF interpreter %s\n", interp_name); 1224 // read interpreter segments as well 1225 if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) { 1226 print_debug("can't open runtime loader\n"); 1227 goto err; 1228 } 1229 break; 1230 } 1231 1232 // from PT_DYNAMIC we want to read address of first link_map addr 1233 case PT_DYNAMIC: { 1234 ph->core->dynamic_addr = exec_php->p_vaddr; 1235 print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr); 1236 break; 1237 } 1238 1239 } // switch 1240 exec_php++; 1241 } // for 1242 1243 free(phbuf); 1244 return true; 1245 err: 1246 free(phbuf); 1247 return false; 1248 } 1249 1250 #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug, r_map) 1251 #define LD_BASE_OFFSET offsetof(struct r_debug, r_ldbase) 1252 #define LINK_MAP_ADDR_OFFSET offsetof(struct link_map, l_addr) 1253 #define LINK_MAP_NAME_OFFSET offsetof(struct link_map, l_name) 1254 #define LINK_MAP_NEXT_OFFSET offsetof(struct link_map, l_next) 1255 1256 // read shared library info from runtime linker's data structures. 1257 // This work is done by librtlb_db in Solaris 1258 static bool read_shared_lib_info(struct ps_prochandle* ph) { 1259 uintptr_t addr = ph->core->dynamic_addr; 1260 uintptr_t debug_base; 1261 uintptr_t first_link_map_addr; 1262 uintptr_t ld_base_addr; 1263 uintptr_t link_map_addr; 1264 uintptr_t lib_base_diff; 1265 uintptr_t lib_base; 1266 uintptr_t lib_name_addr; 1267 char lib_name[BUF_SIZE]; 1268 ELF_DYN dyn; 1269 ELF_EHDR elf_ehdr; 1270 int lib_fd; 1271 1272 // _DYNAMIC has information of the form 1273 // [tag] [data] [tag] [data] ..... 1274 // Both tag and data are pointer sized. 1275 // We look for dynamic info with DT_DEBUG. This has shared object info. 1276 // refer to struct r_debug in link.h 1277 1278 dyn.d_tag = DT_NULL; 1279 while (dyn.d_tag != DT_DEBUG) { 1280 if (ps_pread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) { 1281 print_debug("can't read debug info from _DYNAMIC\n"); 1282 return false; 1283 } 1284 addr += sizeof(ELF_DYN); 1285 } 1286 1287 // we have got Dyn entry with DT_DEBUG 1288 debug_base = dyn.d_un.d_ptr; 1289 // at debug_base we have struct r_debug. This has first link map in r_map field 1290 if (ps_pread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET, 1291 &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) { 1292 print_debug("can't read first link map address\n"); 1293 return false; 1294 } 1295 1296 // read ld_base address from struct r_debug 1297 #if 0 // There is no r_ldbase member on BSD 1298 if (ps_pread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr, 1299 sizeof(uintptr_t)) != PS_OK) { 1300 print_debug("can't read ld base address\n"); 1301 return false; 1302 } 1303 ph->core->ld_base_addr = ld_base_addr; 1304 #else 1305 ph->core->ld_base_addr = 0; 1306 #endif 1307 1308 print_debug("interpreter base address is 0x%lx\n", ld_base_addr); 1309 1310 // now read segments from interp (i.e ld.so or ld-linux.so or ld-elf.so) 1311 if (read_interp_segments(ph) != true) { 1312 return false; 1313 } 1314 1315 // after adding interpreter (ld.so) mappings sort again 1316 if (sort_map_array(ph) != true) { 1317 return false; 1318 } 1319 1320 print_debug("first link map is at 0x%lx\n", first_link_map_addr); 1321 1322 link_map_addr = first_link_map_addr; 1323 while (link_map_addr != 0) { 1324 // read library base address of the .so. Note that even though <sys/link.h> calls 1325 // link_map->l_addr as "base address", this is * not * really base virtual 1326 // address of the shared object. This is actually the difference b/w the virtual 1327 // address mentioned in shared object and the actual virtual base where runtime 1328 // linker loaded it. We use "base diff" in read_lib_segments call below. 1329 1330 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET, 1331 &lib_base_diff, sizeof(uintptr_t)) != PS_OK) { 1332 print_debug("can't read shared object base address diff\n"); 1333 return false; 1334 } 1335 1336 // read address of the name 1337 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET, 1338 &lib_name_addr, sizeof(uintptr_t)) != PS_OK) { 1339 print_debug("can't read address of shared object name\n"); 1340 return false; 1341 } 1342 1343 // read name of the shared object 1344 if (read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) { 1345 print_debug("can't read shared object name\n"); 1346 return false; 1347 } 1348 1349 if (lib_name[0] != '\0') { 1350 // ignore empty lib names 1351 lib_fd = pathmap_open(lib_name); 1352 1353 if (lib_fd < 0) { 1354 print_debug("can't open shared object %s\n", lib_name); 1355 // continue with other libraries... 1356 } else { 1357 if (read_elf_header(lib_fd, &elf_ehdr)) { 1358 lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr); 1359 print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n", 1360 lib_name, lib_base, lib_base_diff); 1361 // while adding library mappings we need to use "base difference". 1362 if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) { 1363 print_debug("can't read shared object's segments\n"); 1364 close(lib_fd); 1365 return false; 1366 } 1367 add_lib_info_fd(ph, lib_name, lib_fd, lib_base); 1368 // Map info is added for the library (lib_name) so 1369 // we need to re-sort it before calling the p_pdread. 1370 if (sort_map_array(ph) != true) { 1371 return false; 1372 } 1373 } else { 1374 print_debug("can't read ELF header for shared object %s\n", lib_name); 1375 close(lib_fd); 1376 // continue with other libraries... 1377 } 1378 } 1379 } 1380 1381 // read next link_map address 1382 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET, 1383 &link_map_addr, sizeof(uintptr_t)) != PS_OK) { 1384 print_debug("can't read next link in link_map\n"); 1385 return false; 1386 } 1387 } 1388 1389 return true; 1390 } 1391 1392 // the one and only one exposed stuff from this file 1393 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) { 1394 ELF_EHDR core_ehdr; 1395 ELF_EHDR exec_ehdr; 1396 1397 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle)); 1398 if (ph == NULL) { 1399 print_debug("can't allocate ps_prochandle\n"); 1400 return NULL; 1401 } 1402 1403 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) { 1404 free(ph); 1405 print_debug("can't allocate ps_prochandle\n"); 1406 return NULL; 1407 } 1408 1409 // initialize ph 1410 ph->ops = &core_ops; 1411 ph->core->core_fd = -1; 1412 ph->core->exec_fd = -1; 1413 ph->core->interp_fd = -1; 1414 1415 print_debug("exec: %s core: %s", exec_file, core_file); 1416 1417 // open the core file 1418 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) { 1419 print_debug("can't open core file\n"); 1420 goto err; 1421 } 1422 1423 // read core file ELF header 1424 if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) { 1425 print_debug("core file is not a valid ELF ET_CORE file\n"); 1426 goto err; 1427 } 1428 1429 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) { 1430 print_debug("can't open executable file\n"); 1431 goto err; 1432 } 1433 1434 if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) { 1435 print_debug("executable file is not a valid ELF ET_EXEC file\n"); 1436 goto err; 1437 } 1438 1439 // process core file segments 1440 if (read_core_segments(ph, &core_ehdr) != true) { 1441 goto err; 1442 } 1443 1444 // process exec file segments 1445 if (read_exec_segments(ph, &exec_ehdr) != true) { 1446 goto err; 1447 } 1448 1449 // exec file is also treated like a shared object for symbol search 1450 if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd, 1451 (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL) { 1452 goto err; 1453 } 1454 1455 // allocate and sort maps into map_array, we need to do this 1456 // here because read_shared_lib_info needs to read from debuggee 1457 // address space 1458 if (sort_map_array(ph) != true) { 1459 goto err; 1460 } 1461 1462 if (read_shared_lib_info(ph) != true) { 1463 goto err; 1464 } 1465 1466 // sort again because we have added more mappings from shared objects 1467 if (sort_map_array(ph) != true) { 1468 goto err; 1469 } 1470 1471 if (init_classsharing_workaround(ph) != true) { 1472 goto err; 1473 } 1474 1475 print_debug("Leave Pgrab_core\n"); 1476 return ph; 1477 1478 err: 1479 Prelease(ph); 1480 return NULL; 1481 } 1482 1483 #endif // __APPLE__