1 /* 2 * Copyright (c) 2003, 2012, 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 <elf.h> 32 #include <link.h> 33 #include "libproc_impl.h" 34 #include "salibelf.h" 35 36 // This file has the libproc implementation to read core files. 37 // For live processes, refer to ps_proc.c. Portions of this is adapted 38 // /modelled after Solaris libproc.so (in particular Pcore.c) 39 40 //---------------------------------------------------------------------- 41 // ps_prochandle cleanup helper functions 42 43 // close all file descriptors 44 static void close_elf_files(struct ps_prochandle* ph) { 45 lib_info* lib = NULL; 46 47 // close core file descriptor 48 if (ph->core->core_fd >= 0) 49 close(ph->core->core_fd); 50 51 // close exec file descriptor 52 if (ph->core->exec_fd >= 0) 53 close(ph->core->exec_fd); 54 55 // close interp file descriptor 56 if (ph->core->interp_fd >= 0) 57 close(ph->core->interp_fd); 58 59 // close class share archive file 60 if (ph->core->classes_jsa_fd >= 0) 61 close(ph->core->classes_jsa_fd); 62 63 // close all library file descriptors 64 lib = ph->libs; 65 while (lib) { 66 int fd = lib->fd; 67 if (fd >= 0 && fd != ph->core->exec_fd) close(fd); 68 lib = lib->next; 69 } 70 } 71 72 // clean all map_info stuff 73 static void destroy_map_info(struct ps_prochandle* ph) { 74 map_info* map = ph->core->maps; 75 while (map) { 76 map_info* next = map->next; 77 free(map); 78 map = next; 79 } 80 81 if (ph->core->map_array) { 82 free(ph->core->map_array); 83 } 84 85 // Part of the class sharing workaround 86 map = ph->core->class_share_maps; 87 while (map) { 88 map_info* next = map->next; 89 free(map); 90 map = next; 91 } 92 } 93 94 // ps_prochandle operations 95 static void core_release(struct ps_prochandle* ph) { 96 if (ph->core) { 97 close_elf_files(ph); 98 destroy_map_info(ph); 99 free(ph->core); 100 } 101 } 102 103 static map_info* allocate_init_map(int fd, off_t offset, uintptr_t vaddr, size_t memsz) { 104 map_info* map; 105 if ( (map = (map_info*) calloc(1, sizeof(map_info))) == NULL) { 106 print_debug("can't allocate memory for map_info\n"); 107 return NULL; 108 } 109 110 // initialize map 111 map->fd = fd; 112 map->offset = offset; 113 map->vaddr = vaddr; 114 map->memsz = memsz; 115 return map; 116 } 117 118 // add map info with given fd, offset, vaddr and memsz 119 static map_info* add_map_info(struct ps_prochandle* ph, int fd, off_t offset, 120 uintptr_t vaddr, size_t memsz) { 121 map_info* map; 122 if ((map = allocate_init_map(fd, offset, vaddr, memsz)) == NULL) { 123 return NULL; 124 } 125 126 // add this to map list 127 map->next = ph->core->maps; 128 ph->core->maps = map; 129 ph->core->num_maps++; 130 131 return map; 132 } 133 134 // Part of the class sharing workaround 135 static map_info* add_class_share_map_info(struct ps_prochandle* ph, off_t offset, 136 uintptr_t vaddr, size_t memsz) { 137 map_info* map; 138 if ((map = allocate_init_map(ph->core->classes_jsa_fd, 139 offset, vaddr, memsz)) == NULL) { 140 return NULL; 141 } 142 143 map->next = ph->core->class_share_maps; 144 ph->core->class_share_maps = map; 145 return map; 146 } 147 148 // Return the map_info for the given virtual address. We keep a sorted 149 // array of pointers in ph->map_array, so we can binary search. 150 static map_info* core_lookup(struct ps_prochandle *ph, uintptr_t addr) 151 { 152 int mid, lo = 0, hi = ph->core->num_maps - 1; 153 map_info *mp; 154 155 while (hi - lo > 1) { 156 mid = (lo + hi) / 2; 157 if (addr >= ph->core->map_array[mid]->vaddr) 158 lo = mid; 159 else 160 hi = mid; 161 } 162 163 if (addr < ph->core->map_array[hi]->vaddr) 164 mp = ph->core->map_array[lo]; 165 else 166 mp = ph->core->map_array[hi]; 167 168 if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) 169 return (mp); 170 171 172 // Part of the class sharing workaround 173 // Unfortunately, we have no way of detecting -Xshare state. 174 // Check out the share maps atlast, if we don't find anywhere. 175 // This is done this way so to avoid reading share pages 176 // ahead of other normal maps. For eg. with -Xshare:off we don't 177 // want to prefer class sharing data to data from core. 178 mp = ph->core->class_share_maps; 179 if (mp) { 180 print_debug("can't locate map_info at 0x%lx, trying class share maps\n", 181 addr); 182 } 183 while (mp) { 184 if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) { 185 print_debug("located map_info at 0x%lx from class share maps\n", 186 addr); 187 return (mp); 188 } 189 mp = mp->next; 190 } 191 192 print_debug("can't locate map_info at 0x%lx\n", addr); 193 return (NULL); 194 } 195 196 //--------------------------------------------------------------- 197 // Part of the class sharing workaround: 198 // 199 // With class sharing, pages are mapped from classes[_g].jsa file. 200 // The read-only class sharing pages are mapped as MAP_SHARED, 201 // PROT_READ pages. These pages are not dumped into core dump. 202 // With this workaround, these pages are read from classes[_g].jsa. 203 204 // FIXME: !HACK ALERT! 205 // The format of sharing achive file header is needed to read shared heap 206 // file mappings. For now, I am hard coding portion of FileMapHeader here. 207 // Refer to filemap.hpp. 208 209 // FileMapHeader describes the shared space data in the file to be 210 // mapped. This structure gets written to a file. It is not a class, 211 // so that the compilers don't add any compiler-private data to it. 212 213 #define NUM_SHARED_MAPS 4 214 215 // Refer to FileMapInfo::_current_version in filemap.hpp 216 #define CURRENT_ARCHIVE_VERSION 1 217 218 struct FileMapHeader { 219 int _magic; // identify file type. 220 int _version; // (from enum, above.) 221 size_t _alignment; // how shared archive should be aligned 222 223 struct space_info { 224 int _file_offset; // sizeof(this) rounded to vm page size 225 char* _base; // copy-on-write base address 226 size_t _capacity; // for validity checking 227 size_t _used; // for setting space top on read 228 229 // 4991491 NOTICE These are C++ bool's in filemap.hpp and must match up with 230 // the C type matching the C++ bool type on any given platform. For 231 // Hotspot on BSD we assume the corresponding C type is char but 232 // licensees on BSD versions may need to adjust the type of these fields. 233 char _read_only; // read only space? 234 char _allow_exec; // executable code in space? 235 236 } _space[NUM_SHARED_MAPS]; 237 238 // Ignore the rest of the FileMapHeader. We don't need those fields here. 239 }; 240 241 static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) { 242 jboolean i; 243 if (ps_pread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) { 244 *pvalue = i; 245 return true; 246 } else { 247 return false; 248 } 249 } 250 251 static bool read_pointer(struct ps_prochandle* ph, uintptr_t addr, uintptr_t* pvalue) { 252 uintptr_t uip; 253 if (ps_pread(ph, (psaddr_t) addr, &uip, sizeof(uip)) == PS_OK) { 254 *pvalue = uip; 255 return true; 256 } else { 257 return false; 258 } 259 } 260 261 // used to read strings from debuggee 262 static bool read_string(struct ps_prochandle* ph, uintptr_t addr, char* buf, size_t size) { 263 size_t i = 0; 264 char c = ' '; 265 266 while (c != '\0') { 267 if (ps_pread(ph, (psaddr_t) addr, &c, sizeof(char)) != PS_OK) 268 return false; 269 if (i < size - 1) 270 buf[i] = c; 271 else // smaller buffer 272 return false; 273 i++; addr++; 274 } 275 276 buf[i] = '\0'; 277 return true; 278 } 279 280 #define USE_SHARED_SPACES_SYM "UseSharedSpaces" 281 // mangled name of Arguments::SharedArchivePath 282 #define SHARED_ARCHIVE_PATH_SYM "_ZN9Arguments17SharedArchivePathE" 283 284 static bool init_classsharing_workaround(struct ps_prochandle* ph) { 285 lib_info* lib = ph->libs; 286 while (lib != NULL) { 287 // we are iterating over shared objects from the core dump. look for 288 // libjvm[_g].so. 289 const char *jvm_name = 0; 290 if ((jvm_name = strstr(lib->name, "/libjvm.so")) != 0 || 291 (jvm_name = strstr(lib->name, "/libjvm_g.so")) != 0) { 292 char classes_jsa[PATH_MAX]; 293 struct FileMapHeader header; 294 size_t n = 0; 295 int fd = -1, m = 0; 296 uintptr_t base = 0, useSharedSpacesAddr = 0; 297 uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0; 298 jboolean useSharedSpaces = 0; 299 300 memset(classes_jsa, 0, sizeof(classes_jsa)); 301 jvm_name = lib->name; 302 useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM); 303 if (useSharedSpacesAddr == 0) { 304 print_debug("can't lookup 'UseSharedSpaces' flag\n"); 305 return false; 306 } 307 308 // Hotspot vm types are not exported to build this library. So 309 // using equivalent type jboolean to read the value of 310 // UseSharedSpaces which is same as hotspot type "bool". 311 if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) { 312 print_debug("can't read the value of 'UseSharedSpaces' flag\n"); 313 return false; 314 } 315 316 if ((int)useSharedSpaces == 0) { 317 print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n"); 318 return true; 319 } 320 321 sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM); 322 if (sharedArchivePathAddrAddr == 0) { 323 print_debug("can't lookup shared archive path symbol\n"); 324 return false; 325 } 326 327 if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) { 328 print_debug("can't read shared archive path pointer\n"); 329 return false; 330 } 331 332 if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) { 333 print_debug("can't read shared archive path value\n"); 334 return false; 335 } 336 337 print_debug("looking for %s\n", classes_jsa); 338 // open the class sharing archive file 339 fd = pathmap_open(classes_jsa); 340 if (fd < 0) { 341 print_debug("can't open %s!\n", classes_jsa); 342 ph->core->classes_jsa_fd = -1; 343 return false; 344 } else { 345 print_debug("opened %s\n", classes_jsa); 346 } 347 348 // read FileMapHeader from the file 349 memset(&header, 0, sizeof(struct FileMapHeader)); 350 if ((n = read(fd, &header, sizeof(struct FileMapHeader))) 351 != sizeof(struct FileMapHeader)) { 352 print_debug("can't read shared archive file map header from %s\n", classes_jsa); 353 close(fd); 354 return false; 355 } 356 357 // check file magic 358 if (header._magic != 0xf00baba2) { 359 print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n", 360 classes_jsa, header._magic); 361 close(fd); 362 return false; 363 } 364 365 // check version 366 if (header._version != CURRENT_ARCHIVE_VERSION) { 367 print_debug("%s has wrong shared archive file version %d, expecting %d\n", 368 classes_jsa, header._version, CURRENT_ARCHIVE_VERSION); 369 close(fd); 370 return false; 371 } 372 373 ph->core->classes_jsa_fd = fd; 374 // add read-only maps from classes[_g].jsa to the list of maps 375 for (m = 0; m < NUM_SHARED_MAPS; m++) { 376 if (header._space[m]._read_only) { 377 base = (uintptr_t) header._space[m]._base; 378 // no need to worry about the fractional pages at-the-end. 379 // possible fractional pages are handled by core_read_data. 380 add_class_share_map_info(ph, (off_t) header._space[m]._file_offset, 381 base, (size_t) header._space[m]._used); 382 print_debug("added a share archive map at 0x%lx\n", base); 383 } 384 } 385 return true; 386 } 387 lib = lib->next; 388 } 389 return true; 390 } 391 392 393 //--------------------------------------------------------------------------- 394 // functions to handle map_info 395 396 // Order mappings based on virtual address. We use this function as the 397 // callback for sorting the array of map_info pointers. 398 static int core_cmp_mapping(const void *lhsp, const void *rhsp) 399 { 400 const map_info *lhs = *((const map_info **)lhsp); 401 const map_info *rhs = *((const map_info **)rhsp); 402 403 if (lhs->vaddr == rhs->vaddr) 404 return (0); 405 406 return (lhs->vaddr < rhs->vaddr ? -1 : 1); 407 } 408 409 // we sort map_info by starting virtual address so that we can do 410 // binary search to read from an address. 411 static bool sort_map_array(struct ps_prochandle* ph) { 412 size_t num_maps = ph->core->num_maps; 413 map_info* map = ph->core->maps; 414 int i = 0; 415 416 // allocate map_array 417 map_info** array; 418 if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) { 419 print_debug("can't allocate memory for map array\n"); 420 return false; 421 } 422 423 // add maps to array 424 while (map) { 425 array[i] = map; 426 i++; 427 map = map->next; 428 } 429 430 // sort is called twice. If this is second time, clear map array 431 if (ph->core->map_array) free(ph->core->map_array); 432 ph->core->map_array = array; 433 // sort the map_info array by base virtual address. 434 qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*), 435 core_cmp_mapping); 436 437 // print map 438 if (is_debug()) { 439 int j = 0; 440 print_debug("---- sorted virtual address map ----\n"); 441 for (j = 0; j < ph->core->num_maps; j++) { 442 print_debug("base = 0x%lx\tsize = %d\n", ph->core->map_array[j]->vaddr, 443 ph->core->map_array[j]->memsz); 444 } 445 } 446 447 return true; 448 } 449 450 #ifndef MIN 451 #define MIN(x, y) (((x) < (y))? (x): (y)) 452 #endif 453 454 static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) { 455 ssize_t resid = size; 456 int page_size=sysconf(_SC_PAGE_SIZE); 457 while (resid != 0) { 458 map_info *mp = core_lookup(ph, addr); 459 uintptr_t mapoff; 460 ssize_t len, rem; 461 off_t off; 462 int fd; 463 464 if (mp == NULL) 465 break; /* No mapping for this address */ 466 467 fd = mp->fd; 468 mapoff = addr - mp->vaddr; 469 len = MIN(resid, mp->memsz - mapoff); 470 off = mp->offset + mapoff; 471 472 if ((len = pread(fd, buf, len, off)) <= 0) 473 break; 474 475 resid -= len; 476 addr += len; 477 buf = (char *)buf + len; 478 479 // mappings always start at page boundary. But, may end in fractional 480 // page. fill zeros for possible fractional page at the end of a mapping. 481 rem = mp->memsz % page_size; 482 if (rem > 0) { 483 rem = page_size - rem; 484 len = MIN(resid, rem); 485 resid -= len; 486 addr += len; 487 // we are not assuming 'buf' to be zero initialized. 488 memset(buf, 0, len); 489 buf += len; 490 } 491 } 492 493 if (resid) { 494 print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n", 495 size, addr, resid); 496 return false; 497 } else { 498 return true; 499 } 500 } 501 502 // null implementation for write 503 static bool core_write_data(struct ps_prochandle* ph, 504 uintptr_t addr, const char *buf , size_t size) { 505 return false; 506 } 507 508 static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id, 509 struct reg* regs) { 510 // for core we have cached the lwp regs from NOTE section 511 thread_info* thr = ph->threads; 512 while (thr) { 513 if (thr->lwp_id == lwp_id) { 514 memcpy(regs, &thr->regs, sizeof(struct reg)); 515 return true; 516 } 517 thr = thr->next; 518 } 519 return false; 520 } 521 522 static bool core_get_lwp_info(struct ps_prochandle *ph, lwpid_t lwp_id, void *linfo) { 523 print_debug("core_get_lwp_info not implemented\n"); 524 return false; 525 } 526 527 static ps_prochandle_ops core_ops = { 528 .release= core_release, 529 .p_pread= core_read_data, 530 .p_pwrite= core_write_data, 531 .get_lwp_regs= core_get_lwp_regs, 532 .get_lwp_info= core_get_lwp_info 533 }; 534 535 // read regs and create thread from NT_PRSTATUS entries from core file 536 static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) { 537 // we have to read prstatus_t from buf 538 // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t"); 539 prstatus_t* prstat = (prstatus_t*) buf; 540 thread_info* newthr; 541 print_debug("got integer regset for lwp %d\n", prstat->pr_pid); 542 // we set pthread_t to -1 for core dump 543 if((newthr = add_thread_info(ph, (pthread_t) -1, prstat->pr_pid)) == NULL) 544 return false; 545 546 // copy regs 547 memcpy(&newthr->regs, &prstat->pr_reg, sizeof(struct reg)); 548 549 if (is_debug()) { 550 print_debug("integer regset\n"); 551 #ifdef i386 552 // print the regset 553 print_debug("\teax = 0x%x\n", newthr->regs.r_eax); 554 print_debug("\tebx = 0x%x\n", newthr->regs.r_ebx); 555 print_debug("\tecx = 0x%x\n", newthr->regs.r_ecx); 556 print_debug("\tedx = 0x%x\n", newthr->regs.r_edx); 557 print_debug("\tesp = 0x%x\n", newthr->regs.r_esp); 558 print_debug("\tebp = 0x%x\n", newthr->regs.r_ebp); 559 print_debug("\tesi = 0x%x\n", newthr->regs.r_esi); 560 print_debug("\tedi = 0x%x\n", newthr->regs.r_edi); 561 print_debug("\teip = 0x%x\n", newthr->regs.r_eip); 562 #endif 563 564 #if defined(amd64) || defined(x86_64) 565 // print the regset 566 print_debug("\tr15 = 0x%lx\n", newthr->regs.r_r15); 567 print_debug("\tr14 = 0x%lx\n", newthr->regs.r_r14); 568 print_debug("\tr13 = 0x%lx\n", newthr->regs.r_r13); 569 print_debug("\tr12 = 0x%lx\n", newthr->regs.r_r12); 570 print_debug("\trbp = 0x%lx\n", newthr->regs.r_rbp); 571 print_debug("\trbx = 0x%lx\n", newthr->regs.r_rbx); 572 print_debug("\tr11 = 0x%lx\n", newthr->regs.r_r11); 573 print_debug("\tr10 = 0x%lx\n", newthr->regs.r_r10); 574 print_debug("\tr9 = 0x%lx\n", newthr->regs.r_r9); 575 print_debug("\tr8 = 0x%lx\n", newthr->regs.r_r8); 576 print_debug("\trax = 0x%lx\n", newthr->regs.r_rax); 577 print_debug("\trcx = 0x%lx\n", newthr->regs.r_rcx); 578 print_debug("\trdx = 0x%lx\n", newthr->regs.r_rdx); 579 print_debug("\trsi = 0x%lx\n", newthr->regs.r_rsi); 580 print_debug("\trdi = 0x%lx\n", newthr->regs.r_rdi); 581 //print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax); 582 print_debug("\trip = 0x%lx\n", newthr->regs.r_rip); 583 print_debug("\tcs = 0x%lx\n", newthr->regs.r_cs); 584 //print_debug("\teflags = 0x%lx\n", newthr->regs.eflags); 585 print_debug("\trsp = 0x%lx\n", newthr->regs.r_rsp); 586 print_debug("\tss = 0x%lx\n", newthr->regs.r_ss); 587 //print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base); 588 //print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base); 589 //print_debug("\tds = 0x%lx\n", newthr->regs.ds); 590 //print_debug("\tes = 0x%lx\n", newthr->regs.es); 591 //print_debug("\tfs = 0x%lx\n", newthr->regs.fs); 592 //print_debug("\tgs = 0x%lx\n", newthr->regs.gs); 593 #endif 594 } 595 596 return true; 597 } 598 599 #define ROUNDUP(x, y) ((((x)+((y)-1))/(y))*(y)) 600 601 // read NT_PRSTATUS entries from core NOTE segment 602 static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) { 603 char* buf = NULL; 604 char* p = NULL; 605 size_t size = note_phdr->p_filesz; 606 607 // we are interested in just prstatus entries. we will ignore the rest. 608 // Advance the seek pointer to the start of the PT_NOTE data 609 if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) { 610 print_debug("failed to lseek to PT_NOTE data\n"); 611 return false; 612 } 613 614 // Now process the PT_NOTE structures. Each one is preceded by 615 // an Elf{32/64}_Nhdr structure describing its type and size. 616 if ( (buf = (char*) malloc(size)) == NULL) { 617 print_debug("can't allocate memory for reading core notes\n"); 618 goto err; 619 } 620 621 // read notes into buffer 622 if (read(ph->core->core_fd, buf, size) != size) { 623 print_debug("failed to read notes, core file must have been truncated\n"); 624 goto err; 625 } 626 627 p = buf; 628 while (p < buf + size) { 629 ELF_NHDR* notep = (ELF_NHDR*) p; 630 char* descdata = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4); 631 print_debug("Note header with n_type = %d and n_descsz = %u\n", 632 notep->n_type, notep->n_descsz); 633 634 if (notep->n_type == NT_PRSTATUS) { 635 if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) 636 return false; 637 } 638 p = descdata + ROUNDUP(notep->n_descsz, 4); 639 } 640 641 free(buf); 642 return true; 643 644 err: 645 if (buf) free(buf); 646 return false; 647 } 648 649 // read all segments from core file 650 static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) { 651 int i = 0; 652 ELF_PHDR* phbuf = NULL; 653 ELF_PHDR* core_php = NULL; 654 655 if ((phbuf = read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL) 656 return false; 657 658 /* 659 * Now iterate through the program headers in the core file. 660 * We're interested in two types of Phdrs: PT_NOTE (which 661 * contains a set of saved /proc structures), and PT_LOAD (which 662 * represents a memory mapping from the process's address space). 663 * 664 * Difference b/w Solaris PT_NOTE and BSD PT_NOTE: 665 * 666 * In Solaris there are two PT_NOTE segments the first PT_NOTE (if present) 667 * contains /proc structs in the pre-2.6 unstructured /proc format. the last 668 * PT_NOTE has data in new /proc format. 669 * 670 * In Solaris, there is only one pstatus (process status). pstatus contains 671 * integer register set among other stuff. For each LWP, we have one lwpstatus 672 * entry that has integer regset for that LWP. 673 * 674 * Linux threads are actually 'clone'd processes. To support core analysis 675 * of "multithreaded" process, Linux creates more than one pstatus (called 676 * "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one 677 * "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular 678 * function "elf_core_dump". 679 */ 680 681 for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) { 682 switch (core_php->p_type) { 683 case PT_NOTE: 684 if (core_handle_note(ph, core_php) != true) goto err; 685 break; 686 687 case PT_LOAD: { 688 if (core_php->p_filesz != 0) { 689 if (add_map_info(ph, ph->core->core_fd, core_php->p_offset, 690 core_php->p_vaddr, core_php->p_filesz) == NULL) goto err; 691 } 692 break; 693 } 694 } 695 696 core_php++; 697 } 698 699 free(phbuf); 700 return true; 701 err: 702 free(phbuf); 703 return false; 704 } 705 706 // read segments of a shared object 707 static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) { 708 int i = 0; 709 ELF_PHDR* phbuf; 710 ELF_PHDR* lib_php = NULL; 711 712 if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL) 713 return false; 714 715 // we want to process only PT_LOAD segments that are not writable. 716 // i.e., text segments. The read/write/exec (data) segments would 717 // have been already added from core file segments. 718 for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) { 719 if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) { 720 if (add_map_info(ph, lib_fd, lib_php->p_offset, lib_php->p_vaddr + lib_base, lib_php->p_filesz) == NULL) 721 goto err; 722 } 723 lib_php++; 724 } 725 726 free(phbuf); 727 return true; 728 err: 729 free(phbuf); 730 return false; 731 } 732 733 // process segments from interpreter (ld-elf.so.1) 734 static bool read_interp_segments(struct ps_prochandle* ph) { 735 ELF_EHDR interp_ehdr; 736 737 if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) { 738 print_debug("interpreter is not a valid ELF file\n"); 739 return false; 740 } 741 742 if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) { 743 print_debug("can't read segments of interpreter\n"); 744 return false; 745 } 746 747 return true; 748 } 749 750 // process segments of a a.out 751 static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) { 752 int i = 0; 753 ELF_PHDR* phbuf = NULL; 754 ELF_PHDR* exec_php = NULL; 755 756 if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL) 757 return false; 758 759 for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) { 760 switch (exec_php->p_type) { 761 762 // add mappings for PT_LOAD segments 763 case PT_LOAD: { 764 // add only non-writable segments of non-zero filesz 765 if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) { 766 if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err; 767 } 768 break; 769 } 770 771 // read the interpreter and it's segments 772 case PT_INTERP: { 773 char interp_name[BUF_SIZE]; 774 775 pread(ph->core->exec_fd, interp_name, MIN(exec_php->p_filesz, BUF_SIZE), exec_php->p_offset); 776 print_debug("ELF interpreter %s\n", interp_name); 777 // read interpreter segments as well 778 if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) { 779 print_debug("can't open runtime loader\n"); 780 goto err; 781 } 782 break; 783 } 784 785 // from PT_DYNAMIC we want to read address of first link_map addr 786 case PT_DYNAMIC: { 787 ph->core->dynamic_addr = exec_php->p_vaddr; 788 print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr); 789 break; 790 } 791 792 } // switch 793 exec_php++; 794 } // for 795 796 free(phbuf); 797 return true; 798 err: 799 free(phbuf); 800 return false; 801 } 802 803 804 #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug, r_map) 805 #define LD_BASE_OFFSET offsetof(struct r_debug, r_ldbase) 806 #define LINK_MAP_ADDR_OFFSET offsetof(struct link_map, l_addr) 807 #define LINK_MAP_NAME_OFFSET offsetof(struct link_map, l_name) 808 #define LINK_MAP_NEXT_OFFSET offsetof(struct link_map, l_next) 809 810 // read shared library info from runtime linker's data structures. 811 // This work is done by librtlb_db in Solaris 812 static bool read_shared_lib_info(struct ps_prochandle* ph) { 813 uintptr_t addr = ph->core->dynamic_addr; 814 uintptr_t debug_base; 815 uintptr_t first_link_map_addr; 816 uintptr_t ld_base_addr; 817 uintptr_t link_map_addr; 818 uintptr_t lib_base_diff; 819 uintptr_t lib_base; 820 uintptr_t lib_name_addr; 821 char lib_name[BUF_SIZE]; 822 ELF_DYN dyn; 823 ELF_EHDR elf_ehdr; 824 int lib_fd; 825 826 // _DYNAMIC has information of the form 827 // [tag] [data] [tag] [data] ..... 828 // Both tag and data are pointer sized. 829 // We look for dynamic info with DT_DEBUG. This has shared object info. 830 // refer to struct r_debug in link.h 831 832 dyn.d_tag = DT_NULL; 833 while (dyn.d_tag != DT_DEBUG) { 834 if (ps_pread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) { 835 print_debug("can't read debug info from _DYNAMIC\n"); 836 return false; 837 } 838 addr += sizeof(ELF_DYN); 839 } 840 841 // we have got Dyn entry with DT_DEBUG 842 debug_base = dyn.d_un.d_ptr; 843 // at debug_base we have struct r_debug. This has first link map in r_map field 844 if (ps_pread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET, 845 &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) { 846 print_debug("can't read first link map address\n"); 847 return false; 848 } 849 850 // read ld_base address from struct r_debug 851 // XXX: There is no r_ldbase member on BSD 852 /* 853 if (ps_pread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr, 854 sizeof(uintptr_t)) != PS_OK) { 855 print_debug("can't read ld base address\n"); 856 return false; 857 } 858 ph->core->ld_base_addr = ld_base_addr; 859 */ 860 ph->core->ld_base_addr = 0; 861 862 print_debug("interpreter base address is 0x%lx\n", ld_base_addr); 863 864 // now read segments from interp (i.e ld-elf.so.1) 865 if (read_interp_segments(ph) != true) 866 return false; 867 868 // after adding interpreter (ld.so) mappings sort again 869 if (sort_map_array(ph) != true) 870 return false; 871 872 print_debug("first link map is at 0x%lx\n", first_link_map_addr); 873 874 link_map_addr = first_link_map_addr; 875 while (link_map_addr != 0) { 876 // read library base address of the .so. Note that even though <sys/link.h> calls 877 // link_map->l_addr as "base address", this is * not * really base virtual 878 // address of the shared object. This is actually the difference b/w the virtual 879 // address mentioned in shared object and the actual virtual base where runtime 880 // linker loaded it. We use "base diff" in read_lib_segments call below. 881 882 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET, 883 &lib_base_diff, sizeof(uintptr_t)) != PS_OK) { 884 print_debug("can't read shared object base address diff\n"); 885 return false; 886 } 887 888 // read address of the name 889 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET, 890 &lib_name_addr, sizeof(uintptr_t)) != PS_OK) { 891 print_debug("can't read address of shared object name\n"); 892 return false; 893 } 894 895 // read name of the shared object 896 if (read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) { 897 print_debug("can't read shared object name\n"); 898 return false; 899 } 900 901 if (lib_name[0] != '\0') { 902 // ignore empty lib names 903 lib_fd = pathmap_open(lib_name); 904 905 if (lib_fd < 0) { 906 print_debug("can't open shared object %s\n", lib_name); 907 // continue with other libraries... 908 } else { 909 if (read_elf_header(lib_fd, &elf_ehdr)) { 910 lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr); 911 print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n", 912 lib_name, lib_base, lib_base_diff); 913 // while adding library mappings we need to use "base difference". 914 if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) { 915 print_debug("can't read shared object's segments\n"); 916 close(lib_fd); 917 return false; 918 } 919 add_lib_info_fd(ph, lib_name, lib_fd, lib_base); 920 // Map info is added for the library (lib_name) so 921 // we need to re-sort it before calling the p_pdread. 922 if (sort_map_array(ph) != true) 923 return false; 924 } else { 925 print_debug("can't read ELF header for shared object %s\n", lib_name); 926 close(lib_fd); 927 // continue with other libraries... 928 } 929 } 930 } 931 932 // read next link_map address 933 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET, 934 &link_map_addr, sizeof(uintptr_t)) != PS_OK) { 935 print_debug("can't read next link in link_map\n"); 936 return false; 937 } 938 } 939 940 return true; 941 } 942 943 // the one and only one exposed stuff from this file 944 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) { 945 ELF_EHDR core_ehdr; 946 ELF_EHDR exec_ehdr; 947 948 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle)); 949 if (ph == NULL) { 950 print_debug("can't allocate ps_prochandle\n"); 951 return NULL; 952 } 953 954 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) { 955 free(ph); 956 print_debug("can't allocate ps_prochandle\n"); 957 return NULL; 958 } 959 960 // initialize ph 961 ph->ops = &core_ops; 962 ph->core->core_fd = -1; 963 ph->core->exec_fd = -1; 964 ph->core->interp_fd = -1; 965 966 // open the core file 967 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) { 968 print_debug("can't open core file\n"); 969 goto err; 970 } 971 972 // read core file ELF header 973 if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) { 974 print_debug("core file is not a valid ELF ET_CORE file\n"); 975 goto err; 976 } 977 978 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) { 979 print_debug("can't open executable file\n"); 980 goto err; 981 } 982 983 if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) { 984 print_debug("executable file is not a valid ELF ET_EXEC file\n"); 985 goto err; 986 } 987 988 // process core file segments 989 if (read_core_segments(ph, &core_ehdr) != true) 990 goto err; 991 992 // process exec file segments 993 if (read_exec_segments(ph, &exec_ehdr) != true) 994 goto err; 995 996 // exec file is also treated like a shared object for symbol search 997 if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd, 998 (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL) 999 goto err; 1000 1001 // allocate and sort maps into map_array, we need to do this 1002 // here because read_shared_lib_info needs to read from debuggee 1003 // address space 1004 if (sort_map_array(ph) != true) 1005 goto err; 1006 1007 if (read_shared_lib_info(ph) != true) 1008 goto err; 1009 1010 // sort again because we have added more mappings from shared objects 1011 if (sort_map_array(ph) != true) 1012 goto err; 1013 1014 if (init_classsharing_workaround(ph) != true) 1015 goto err; 1016 1017 return ph; 1018 1019 err: 1020 Prelease(ph); 1021 return NULL; 1022 }