1 /*
   2  * Copyright (c) 2003, 2013, 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 
  33 #ifdef __APPLE__
  34 #include "sun_jvm_hotspot_debugger_amd64_AMD64ThreadContext.h"
  35 #endif 
  36 
  37 // This file has the libproc implementation to read core files.
  38 // For live processes, refer to ps_proc.c. Portions of this is adapted
  39 // /modelled after Solaris libproc.so (in particular Pcore.c)
  40 
  41 //----------------------------------------------------------------------
  42 // ps_prochandle cleanup helper functions
  43 
  44 // close all file descriptors
  45 static void close_files(struct ps_prochandle* ph) {
  46   lib_info* lib = NULL;
  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) {
  68       close(fd);
  69     }
  70     lib = lib->next;
  71   }
  72 }
  73 
  74 // clean all map_info stuff
  75 static void destroy_map_info(struct ps_prochandle* ph) {
  76   map_info* map = ph->core->maps;
  77   while (map) {
  78     map_info* next = map->next;
  79     free(map);
  80     map = next;
  81   }
  82 
  83   if (ph->core->map_array) {
  84     free(ph->core->map_array);
  85   }
  86 
  87   // Part of the class sharing workaround
  88   map = ph->core->class_share_maps;
  89   while (map) {
  90     map_info* next = map->next;
  91     free(map);
  92     map = next;
  93   }
  94 }
  95 
  96 // ps_prochandle operations
  97 static void core_release(struct ps_prochandle* ph) {
  98   if (ph->core) {
  99     close_files(ph);
 100     destroy_map_info(ph);
 101     free(ph->core);
 102   }
 103 }
 104 
 105 static map_info* allocate_init_map(int fd, off_t offset, uintptr_t vaddr, size_t memsz) {
 106   map_info* map;
 107   if ( (map = (map_info*) calloc(1, sizeof(map_info))) == NULL) {
 108     print_debug("can't allocate memory for map_info\n");
 109     return NULL;
 110   }
 111 
 112   // initialize map
 113   map->fd     = fd;
 114   map->offset = offset;
 115   map->vaddr  = vaddr;
 116   map->memsz  = memsz;
 117   return map;
 118 }
 119 
 120 // add map info with given fd, offset, vaddr and memsz
 121 static map_info* add_map_info(struct ps_prochandle* ph, int fd, off_t offset,
 122                              uintptr_t vaddr, size_t memsz) {
 123   map_info* map;
 124   if ((map = allocate_init_map(fd, offset, vaddr, memsz)) == NULL) {
 125     return NULL;
 126   }
 127 
 128   // add this to map list
 129   map->next  = ph->core->maps;
 130   ph->core->maps   = map;
 131   ph->core->num_maps++;
 132 
 133   return map;
 134 }
 135 
 136 // Part of the class sharing workaround
 137 static map_info* add_class_share_map_info(struct ps_prochandle* ph, off_t offset,
 138                              uintptr_t vaddr, size_t memsz) {
 139   map_info* map;
 140   if ((map = allocate_init_map(ph->core->classes_jsa_fd,
 141                                offset, vaddr, memsz)) == NULL) {
 142     return NULL;
 143   }
 144 
 145   map->next = ph->core->class_share_maps;
 146   ph->core->class_share_maps = map;
 147   return map;
 148 }
 149 
 150 // Return the map_info for the given virtual address.  We keep a sorted
 151 // array of pointers in ph->map_array, so we can binary search.
 152 static map_info* core_lookup(struct ps_prochandle *ph, uintptr_t addr)
 153 {
 154   int mid, lo = 0, hi = ph->core->num_maps - 1;
 155   map_info *mp;
 156 
 157   while (hi - lo > 1) {
 158     mid = (lo + hi) / 2;
 159     if (addr >= ph->core->map_array[mid]->vaddr) {
 160       lo = mid;
 161     } else {
 162       hi = mid;
 163     }
 164   }
 165 
 166   if (addr < ph->core->map_array[hi]->vaddr) {
 167     mp = ph->core->map_array[lo];
 168   } else {
 169     mp = ph->core->map_array[hi];
 170   }
 171 
 172   if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) {
 173     return (mp); 
 174   }
 175 
 176 
 177   // Part of the class sharing workaround
 178   // Unfortunately, we have no way of detecting -Xshare state.
 179   // Check out the share maps atlast, if we don't find anywhere.
 180   // This is done this way so to avoid reading share pages
 181   // ahead of other normal maps. For eg. with -Xshare:off we don't
 182   // want to prefer class sharing data to data from core.
 183   mp = ph->core->class_share_maps;
 184   if (mp) {
 185     print_debug("can't locate map_info at 0x%lx, trying class share maps\n", addr);
 186   }
 187   while (mp) {
 188     if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) {
 189       print_debug("located map_info at 0x%lx from class share maps\n", addr);
 190       return (mp);
 191     }
 192     mp = mp->next;
 193   }
 194 
 195   print_debug("can't locate map_info at 0x%lx\n", addr);
 196   return (NULL);
 197 }
 198 
 199 //---------------------------------------------------------------
 200 // Part of the class sharing workaround:
 201 //
 202 // With class sharing, pages are mapped from classes[_g].jsa file.
 203 // The read-only class sharing pages are mapped as MAP_SHARED,
 204 // PROT_READ pages. These pages are not dumped into core dump.
 205 // With this workaround, these pages are read from classes[_g].jsa.
 206 
 207 // FIXME: !HACK ALERT!
 208 // The format of sharing achive file header is needed to read shared heap
 209 // file mappings. For now, I am hard coding portion of FileMapHeader here.
 210 // Refer to filemap.hpp.
 211 
 212 // FileMapHeader describes the shared space data in the file to be
 213 // mapped.  This structure gets written to a file.  It is not a class,
 214 // so that the compilers don't add any compiler-private data to it.
 215 
 216 #define NUM_SHARED_MAPS 4
 217 
 218 // Refer to FileMapInfo::_current_version in filemap.hpp
 219 #define CURRENT_ARCHIVE_VERSION 1
 220 
 221 struct FileMapHeader {
 222   int   _magic;              // identify file type.
 223   int   _version;            // (from enum, above.)
 224   size_t _alignment;         // how shared archive should be aligned
 225 
 226   struct space_info {
 227     int    _file_offset;     // sizeof(this) rounded to vm page size
 228     char*  _base;            // copy-on-write base address
 229     size_t _capacity;        // for validity checking
 230     size_t _used;            // for setting space top on read
 231 
 232     // 4991491 NOTICE These are C++ bool's in filemap.hpp and must match up with
 233     // the C type matching the C++ bool type on any given platform. For
 234     // Hotspot on BSD we assume the corresponding C type is char but
 235     // licensees on BSD versions may need to adjust the type of these fields.
 236     char   _read_only;       // read only space?
 237     char   _allow_exec;      // executable code in space?
 238 
 239   } _space[NUM_SHARED_MAPS];
 240 
 241   // Ignore the rest of the FileMapHeader. We don't need those fields here.
 242 };
 243 
 244 static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) {
 245   jboolean i;
 246   if (ps_pread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) {
 247     *pvalue = i;
 248     return true;
 249   } else {
 250     return false;
 251   }
 252 }
 253 
 254 static bool read_pointer(struct ps_prochandle* ph, uintptr_t addr, uintptr_t* pvalue) {
 255   uintptr_t uip;
 256   if (ps_pread(ph, (psaddr_t) addr, (char *)&uip, sizeof(uip)) == PS_OK) {
 257     *pvalue = uip;
 258     return true;
 259   } else {
 260     return false;
 261   }
 262 }
 263 
 264 // used to read strings from debuggee
 265 static bool read_string(struct ps_prochandle* ph, uintptr_t addr, char* buf, size_t size) {
 266   size_t i = 0;
 267   char  c = ' ';
 268 
 269   while (c != '\0') {
 270     if (ps_pread(ph, (psaddr_t) addr, &c, sizeof(char)) != PS_OK) {
 271       return false; 
 272     }
 273     if (i < size - 1) {
 274       buf[i] = c; 
 275     } else {
 276       // smaller buffer
 277       return false;
 278     }
 279     i++; addr++;
 280   }
 281   buf[i] = '\0';
 282   return true;
 283 }
 284 
 285 #ifdef __APPLE__
 286 #define USE_SHARED_SPACES_SYM "_UseSharedSpaces"
 287 // mangled name of Arguments::SharedArchivePath
 288 #define SHARED_ARCHIVE_PATH_SYM "_ZN9Arguments17SharedArchivePathE"
 289 #else
 290 #define USE_SHARED_SPACES_SYM "UseSharedSpaces"
 291 // mangled name of Arguments::SharedArchivePath
 292 #define SHARED_ARCHIVE_PATH_SYM "__ZN9Arguments17SharedArchivePathE"
 293 #endif // __APPLE_
 294 
 295 static bool init_classsharing_workaround(struct ps_prochandle* ph) {
 296   int m;
 297   size_t n;
 298   lib_info* lib = ph->libs;
 299   while (lib != NULL) {
 300     // we are iterating over shared objects from the core dump. look for
 301     // libjvm[_g].so.
 302     const char *jvm_name = 0;
 303 #ifdef __APPLE__
 304     if ((jvm_name = strstr(lib->name, "/libjvm.dylib")) != 0 ||
 305         (jvm_name = strstr(lib->name, "/libjvm_g.dylib")) != 0)
 306 #else
 307     if ((jvm_name = strstr(lib->name, "/libjvm.so")) != 0 ||
 308         (jvm_name = strstr(lib->name, "/libjvm_g.so")) != 0)
 309 #endif // __APPLE__
 310     {
 311       char classes_jsa[PATH_MAX];
 312       struct FileMapHeader header;
 313       int fd = -1;
 314       uintptr_t base = 0, useSharedSpacesAddr = 0;
 315       uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0;
 316       jboolean useSharedSpaces = 0;
 317 
 318       memset(classes_jsa, 0, sizeof(classes_jsa));
 319       jvm_name = lib->name;
 320       useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM);
 321       if (useSharedSpacesAddr == 0) {
 322         print_debug("can't lookup 'UseSharedSpaces' flag\n");
 323         return false;
 324       }
 325 
 326       // Hotspot vm types are not exported to build this library. So
 327       // using equivalent type jboolean to read the value of
 328       // UseSharedSpaces which is same as hotspot type "bool".
 329       if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) {
 330         print_debug("can't read the value of 'UseSharedSpaces' flag\n");
 331         return false;
 332       }
 333 
 334       if ((int)useSharedSpaces == 0) {
 335         print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n");
 336         return true;
 337       }
 338 
 339       sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM);
 340       if (sharedArchivePathAddrAddr == 0) {
 341         print_debug("can't lookup shared archive path symbol\n");
 342         return false;
 343       }
 344 
 345       if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) {
 346         print_debug("can't read shared archive path pointer\n");
 347         return false;
 348       }
 349 
 350       if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) {
 351         print_debug("can't read shared archive path value\n");
 352         return false;
 353       }
 354 
 355       print_debug("looking for %s\n", classes_jsa);
 356       // open the class sharing archive file
 357       fd = pathmap_open(classes_jsa);
 358       if (fd < 0) {
 359         print_debug("can't open %s!\n", classes_jsa);
 360         ph->core->classes_jsa_fd = -1;
 361         return false;
 362       } else {
 363         print_debug("opened %s\n", classes_jsa);
 364       }
 365 
 366       // read FileMapHeader from the file
 367       memset(&header, 0, sizeof(struct FileMapHeader));
 368       if ((n = read(fd, &header, sizeof(struct FileMapHeader)))
 369            != sizeof(struct FileMapHeader)) {
 370         print_debug("can't read shared archive file map header from %s\n", classes_jsa);
 371         close(fd);
 372         return false;
 373       }
 374 
 375       // check file magic
 376       if (header._magic != 0xf00baba2) {
 377         print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n",
 378                      classes_jsa, header._magic);
 379         close(fd);
 380         return false;
 381       }
 382 
 383       // check version
 384       if (header._version != CURRENT_ARCHIVE_VERSION) {
 385         print_debug("%s has wrong shared archive file version %d, expecting %d\n",
 386                      classes_jsa, header._version, CURRENT_ARCHIVE_VERSION);
 387         close(fd);
 388         return false;
 389       }
 390 
 391       ph->core->classes_jsa_fd = fd;
 392       // add read-only maps from classes[_g].jsa to the list of maps
 393       for (m = 0; m < NUM_SHARED_MAPS; m++) {
 394         if (header._space[m]._read_only) {
 395           base = (uintptr_t) header._space[m]._base;
 396           // no need to worry about the fractional pages at-the-end.
 397           // possible fractional pages are handled by core_read_data.
 398           add_class_share_map_info(ph, (off_t) header._space[m]._file_offset,
 399                                    base, (size_t) header._space[m]._used);
 400           print_debug("added a share archive map at 0x%lx\n", base);
 401         }
 402       }
 403       return true;
 404     }
 405     lib = lib->next;
 406   }
 407   return true;
 408 }
 409 
 410 //---------------------------------------------------------------------------
 411 // functions to handle map_info
 412 
 413 // Order mappings based on virtual address.  We use this function as the
 414 // callback for sorting the array of map_info pointers.
 415 static int core_cmp_mapping(const void *lhsp, const void *rhsp)
 416 {
 417   const map_info *lhs = *((const map_info **)lhsp);
 418   const map_info *rhs = *((const map_info **)rhsp);
 419 
 420   if (lhs->vaddr == rhs->vaddr) {
 421     return (0);
 422   }
 423 
 424   return (lhs->vaddr < rhs->vaddr ? -1 : 1);
 425 }
 426 
 427 // we sort map_info by starting virtual address so that we can do
 428 // binary search to read from an address.
 429 static bool sort_map_array(struct ps_prochandle* ph) {
 430   size_t num_maps = ph->core->num_maps;
 431   map_info* map = ph->core->maps;
 432   int i = 0;
 433 
 434   // allocate map_array
 435   map_info** array;
 436   if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) {
 437      print_debug("can't allocate memory for map array\n");
 438      return false;
 439   }
 440 
 441   // add maps to array
 442   while (map) {
 443     array[i] = map;
 444     i++;
 445     map = map->next;
 446   }
 447 
 448   // sort is called twice. If this is second time, clear map array
 449   if (ph->core->map_array) {
 450     free(ph->core->map_array);
 451   }
 452   ph->core->map_array = array;
 453   // sort the map_info array by base virtual address.
 454   qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*),
 455            core_cmp_mapping);
 456 
 457   // print map
 458   if (is_debug()) {
 459     int j = 0;
 460     print_debug("---- sorted virtual address map ----\n");
 461     for (j = 0; j < ph->core->num_maps; j++) {
 462       print_debug("base = 0x%lx\tsize = %d\n", ph->core->map_array[j]->vaddr,
 463                                        ph->core->map_array[j]->memsz);
 464     }
 465   }
 466 
 467   return true;
 468 }
 469 
 470 #ifndef MIN
 471 #define MIN(x, y) (((x) < (y))? (x): (y))
 472 #endif
 473 
 474 static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) {
 475    ssize_t resid = size;
 476    int page_size=sysconf(_SC_PAGE_SIZE);
 477    while (resid != 0) {
 478       map_info *mp = core_lookup(ph, addr);
 479       uintptr_t mapoff;
 480       ssize_t len, rem;
 481       off_t off;
 482       int fd;
 483 
 484       if (mp == NULL) {
 485          break;  /* No mapping for this address */
 486       }
 487 
 488       fd = mp->fd;
 489       mapoff = addr - mp->vaddr;
 490       len = MIN(resid, mp->memsz - mapoff);
 491       off = mp->offset + mapoff;
 492 
 493       if ((len = pread(fd, buf, len, off)) <= 0) {
 494          break;
 495       }
 496 
 497       resid -= len;
 498       addr += len;
 499       buf = (char *)buf + len;
 500 
 501       // mappings always start at page boundary. But, may end in fractional
 502       // page. fill zeros for possible fractional page at the end of a mapping.
 503       rem = mp->memsz % page_size;
 504       if (rem > 0) {
 505          rem = page_size - rem;
 506          len = MIN(resid, rem);
 507          resid -= len;
 508          addr += len;
 509          // we are not assuming 'buf' to be zero initialized.
 510          memset(buf, 0, len);
 511          buf += len;
 512       }
 513    }
 514 
 515    if (resid) {
 516       print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n",
 517               size, addr, resid);
 518       return false;
 519    } else {
 520       return true;
 521    }
 522 }
 523 
 524 // null implementation for write
 525 static bool core_write_data(struct ps_prochandle* ph,
 526                              uintptr_t addr, const char *buf , size_t size) {
 527    return false;
 528 }
 529 
 530 static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id,
 531                           struct reg* regs) {
 532    // for core we have cached the lwp regs after segment parsed
 533    sa_thread_info* thr = ph->threads;
 534    while (thr) {
 535      if (thr->lwp_id == lwp_id) {
 536        memcpy(regs, &thr->regs, sizeof(struct reg));
 537        return true;
 538      }
 539      thr = thr->next;
 540    }
 541    return false;
 542 }
 543 
 544 static bool core_get_lwp_info(struct ps_prochandle *ph, lwpid_t id, void *info) {
 545    print_debug("core_get_lwp_info not implemented\n");
 546    return false;
 547 }
 548 
 549 static ps_prochandle_ops core_ops = {
 550    .release=  core_release,
 551    .p_pread=  core_read_data,
 552    .p_pwrite= core_write_data,
 553    .get_lwp_regs= core_get_lwp_regs,
 554    .get_lwp_info= core_get_lwp_info
 555 };
 556 
 557 // from this point, mainly two blocks divided by def __APPLE__
 558 // one for Macosx, the other for regular Bsd
 559 
 560 #ifdef __APPLE__
 561 
 562 void print_thread(sa_thread_info *threadinfo) {
 563   print_debug("thread added: %d\n", threadinfo->lwp_id);
 564   print_debug("registers:\n");
 565   print_debug("  r_r15: 0x%" PRIx64 "\n", threadinfo->regs.r_r15);
 566   print_debug("  r_r14: 0x%" PRIx64 "\n", threadinfo->regs.r_r14);
 567   print_debug("  r_r13: 0x%" PRIx64 "\n", threadinfo->regs.r_r13);
 568   print_debug("  r_r12: 0x%" PRIx64 "\n", threadinfo->regs.r_r12);
 569   print_debug("  r_r11: 0x%" PRIx64 "\n", threadinfo->regs.r_r11);
 570   print_debug("  r_r10: 0x%" PRIx64 "\n", threadinfo->regs.r_r10);
 571   print_debug("  r_r9:  0x%" PRIx64 "\n", threadinfo->regs.r_r9);
 572   print_debug("  r_r8:  0x%" PRIx64 "\n", threadinfo->regs.r_r8);
 573   print_debug("  r_rdi: 0x%" PRIx64 "\n", threadinfo->regs.r_rdi);
 574   print_debug("  r_rsi: 0x%" PRIx64 "\n", threadinfo->regs.r_rsi);
 575   print_debug("  r_rbp: 0x%" PRIx64 "\n", threadinfo->regs.r_rbp);
 576   print_debug("  r_rbx: 0x%" PRIx64 "\n", threadinfo->regs.r_rbx);
 577   print_debug("  r_rdx: 0x%" PRIx64 "\n", threadinfo->regs.r_rdx);
 578   print_debug("  r_rcx: 0x%" PRIx64 "\n", threadinfo->regs.r_rcx);
 579   print_debug("  r_rax: 0x%" PRIx64 "\n", threadinfo->regs.r_rax);
 580   print_debug("  r_fs:  0x%" PRIx32 "\n", threadinfo->regs.r_fs);
 581   print_debug("  r_gs:  0x%" PRIx32 "\n", threadinfo->regs.r_gs);
 582   print_debug("  r_rip  0x%" PRIx64 "\n", threadinfo->regs.r_rip);
 583   print_debug("  r_cs:  0x%" PRIx64 "\n", threadinfo->regs.r_cs);
 584   print_debug("  r_rsp: 0x%" PRIx64 "\n", threadinfo->regs.r_rsp);
 585   print_debug("  r_rflags: 0x%" PRIx64 "\n", threadinfo->regs.r_rflags);
 586 }
 587 
 588 // read all segments64 commands from core file
 589 // read all thread commands from core file
 590 static bool read_core_segments(struct ps_prochandle* ph) {
 591   int i = 0;
 592   int num_threads = 0;
 593   int fd = ph->core->core_fd;
 594   off_t offset = 0;
 595   mach_header_64      fhead;
 596   load_command        lcmd;
 597   segment_command_64  segcmd;
 598   // thread_command      thrcmd;
 599    
 600   lseek(fd, offset, SEEK_SET);
 601   if(read(fd, (void *)&fhead, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
 602      goto err;
 603   }
 604   print_debug("total commands: %d\n", fhead.ncmds);
 605   offset += sizeof(mach_header_64);
 606   for (i = 0; i < fhead.ncmds; i++) {
 607     lseek(fd, offset, SEEK_SET);
 608     if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
 609       goto err;
 610     }
 611     offset += lcmd.cmdsize;    // next command position 
 612     if (lcmd.cmd == LC_SEGMENT_64) {
 613       lseek(fd, -sizeof(load_command), SEEK_CUR);
 614       if (read(fd, (void *)&segcmd, sizeof(segment_command_64)) != sizeof(segment_command_64)) {
 615         print_debug("failed to read LC_SEGMENT_64 i = %d!\n", i);
 616         goto err;
 617       }
 618       if (add_map_info(ph, fd, segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize) == NULL) {
 619         print_debug("Failed to add map_info at i = %d\n", i);
 620         goto err;
 621       }
 622       print_debug("segment added: %" PRIu64 " 0x%" PRIx64 " %d\n",
 623                    segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize); 
 624     } else if (lcmd.cmd == LC_THREAD || lcmd.cmd == LC_UNIXTHREAD) {
 625       typedef struct thread_fc {
 626         uint32_t  flavor;
 627         uint32_t  count;
 628       } thread_fc;
 629       thread_fc fc;
 630       uint32_t size = sizeof(load_command);
 631       while (size < lcmd.cmdsize) {
 632         if (read(fd, (void *)&fc, sizeof(thread_fc)) != sizeof(thread_fc)) {
 633           printf("Reading flavor, count failed.\n");
 634           goto err;
 635         }
 636         size += sizeof(thread_fc);
 637         if (fc.flavor == x86_THREAD_STATE) {
 638           x86_thread_state_t thrstate;
 639           if (read(fd, (void *)&thrstate, sizeof(x86_thread_state_t)) != sizeof(x86_thread_state_t)) {
 640             printf("Reading flavor, count failed.\n");
 641             goto err;
 642           }
 643           size += sizeof(x86_thread_state_t);
 644           // create thread info list, update lwp_id later
 645           sa_thread_info* newthr = add_thread_info(ph, (pthread_t) -1, (lwpid_t) num_threads++);
 646           if (newthr == NULL) {
 647             printf("create thread_info failed\n");
 648             goto err;
 649           }
 650 
 651           // note __DARWIN_UNIX03 depengs on other definitions
 652 #if __DARWIN_UNIX03
 653 #define get_register_v(regst, regname) \
 654   regst.uts.ts64.__##regname
 655 #else
 656 #define get_register_v(regst, regname) \
 657   regst.uts.ts64.##regname
 658 #endif // __DARWIN_UNIX03
 659           newthr->regs.r_rax = get_register_v(thrstate, rax);
 660           newthr->regs.r_rbx = get_register_v(thrstate, rbx);
 661           newthr->regs.r_rcx = get_register_v(thrstate, rcx);
 662           newthr->regs.r_rdx = get_register_v(thrstate, rdx);
 663           newthr->regs.r_rdi = get_register_v(thrstate, rdi);
 664           newthr->regs.r_rsi = get_register_v(thrstate, rsi);
 665           newthr->regs.r_rbp = get_register_v(thrstate, rbp);
 666           newthr->regs.r_rsp = get_register_v(thrstate, rsp);
 667           newthr->regs.r_r8  = get_register_v(thrstate, r8);
 668           newthr->regs.r_r9  = get_register_v(thrstate, r9);
 669           newthr->regs.r_r10 = get_register_v(thrstate, r10);
 670           newthr->regs.r_r11 = get_register_v(thrstate, r11);
 671           newthr->regs.r_r12 = get_register_v(thrstate, r12);
 672           newthr->regs.r_r13 = get_register_v(thrstate, r13);
 673           newthr->regs.r_r14 = get_register_v(thrstate, r14);
 674           newthr->regs.r_r15 = get_register_v(thrstate, r15);
 675           newthr->regs.r_rip = get_register_v(thrstate, rip);
 676           newthr->regs.r_rflags = get_register_v(thrstate, rflags);
 677           newthr->regs.r_cs  = get_register_v(thrstate, cs);
 678           newthr->regs.r_fs  = get_register_v(thrstate, fs);
 679           newthr->regs.r_gs  = get_register_v(thrstate, gs);
 680           print_thread(newthr);
 681         } else if (fc.flavor == x86_FLOAT_STATE) {
 682           x86_float_state_t flstate;
 683           if (read(fd, (void *)&flstate, sizeof(x86_float_state_t)) != sizeof(x86_float_state_t)) {
 684             print_debug("Reading flavor, count failed.\n");
 685             goto err;
 686           } 
 687           size += sizeof(x86_float_state_t);
 688         } else if (fc.flavor == x86_EXCEPTION_STATE) {
 689           x86_exception_state_t excpstate;
 690           if (read(fd, (void *)&excpstate, sizeof(x86_exception_state_t)) != sizeof(x86_exception_state_t)) {
 691             printf("Reading flavor, count failed.\n");
 692             goto err;
 693           }
 694           size += sizeof(x86_exception_state_t);
 695         }
 696       }
 697     }
 698   }
 699   return true;
 700 err:
 701   return false;
 702 }
 703 
 704 /**local function **/
 705 bool exists(const char *fname)
 706 {
 707   int fd;
 708   if ((fd = open(fname, O_RDONLY)) > 0) {
 709     close(fd);
 710     return true;
 711   }
 712   return false;
 713 }
 714 
 715 // we check: 1. lib
 716 //           2. lib/server
 717 //           3. jre/lib
 718 //           4. jre/lib/server
 719 // from: 1. exe path
 720 //       2. JAVA_HOME
 721 //       3. DYLD_LIBRARY_PATH
 722 static bool get_real_path(struct ps_prochandle* ph, char *rpath) {
 723   /** check if they exist in JAVA ***/
 724   char* execname = ph->core->exec_path;
 725   char  filepath[4096];
 726   char* filename = strrchr(rpath, '/');               // like /libjvm.dylib
 727   if (filename == NULL) {
 728     return false;
 729   }
 730 
 731   char* posbin = strstr(execname, "/bin/java");
 732   if (posbin != NULL) {
 733     memcpy(filepath, execname, posbin - execname);    // not include trailing '/'
 734     filepath[posbin - execname] = '\0';
 735   } else {
 736     char* java_home = getenv("JAVA_HOME");
 737     if (java_home != NULL) {
 738       strcpy(filepath, java_home);
 739     } else {
 740       char* dyldpath = getenv("DYLD_LIBRARY_PATH");
 741       char* dypath = strtok(dyldpath, ":");
 742       while (dypath != NULL) {
 743         strcpy(filepath, dypath);
 744         strcat(filepath, filename);
 745         if (exists(filepath)) {
 746            strcpy(rpath, filepath);
 747            return true;
 748         }
 749         dypath = strtok(dyldpath, ":");
 750       }
 751       // not found 
 752       return false;
 753     }
 754   }
 755   // for exec and java_home, jdkpath now is filepath
 756   size_t filepath_base_size = strlen(filepath);
 757 
 758   // first try /lib/ and /lib/server
 759   strcat(filepath, "/lib");
 760   strcat(filepath, filename);
 761   if (exists(filepath)) {
 762     strcpy(rpath, filepath);
 763     return true;
 764   }
 765   char* pos = strstr(filepath, filename);    // like /libjvm.dylib
 766   *pos = '\0';
 767   strcat(filepath, "/server");
 768   strcat(filepath, filename);
 769   if (exists(filepath)) {
 770     strcpy(rpath, filepath);
 771     return true;
 772   }
 773 
 774 
 775   // then try /jre/lib/ and /jre/lib/server
 776   filepath[filepath_base_size] = '\0';
 777   strcat(filepath, "/jre/lib");
 778   strcat(filepath, filename);
 779   if (exists(filepath)) {
 780     strcpy(rpath, filepath);
 781     return true;
 782   }
 783   pos = strstr(filepath, filename);
 784   *pos = '\0';
 785   strcat(filepath, "/server");
 786   strcat(filepath, filename);
 787   if (exists(filepath)) {
 788     strcpy(rpath, filepath);
 789     return true;
 790   }
 791 
 792   return false;
 793 }
 794 
 795 static bool read_shared_lib_info(struct ps_prochandle* ph) {
 796   static int pagesize = 0;
 797   int fd = ph->core->core_fd;
 798   int i = 0, j;
 799   uint32_t  v;
 800   mach_header_64 header;        // used to check if a file header in segment
 801   load_command lcmd;
 802   dylib_command dylibcmd;
 803   
 804   char name[BUF_SIZE];  // use to store name
 805 
 806   if (pagesize == 0) {
 807     pagesize = getpagesize();
 808     print_debug("page size is %d\n", pagesize);
 809   }
 810   for (j = 0; j < ph->core->num_maps; j++) {
 811     map_info *iter = ph->core->map_array[j];   // head
 812     off_t fpos = iter->offset;
 813     if (iter->fd != fd) {
 814       // only search core file!
 815       continue;
 816     }
 817     print_debug("map_info %d: vmaddr = 0x%016" PRIx64 "  fileoff = %" PRIu64 "  vmsize = %" PRIu64 "\n",
 818                            j, iter->vaddr, iter->offset, iter->memsz);
 819     lseek(fd, fpos, SEEK_SET);
 820     // we assume .dylib loaded at segment address --- which is true for JVM libraries
 821     // multiple files may be loaded in one segment.
 822     // if first word is not a magic word, means this segment does not contain lib file.
 823     if (read(fd, (void *)&v, sizeof(uint32_t)) == sizeof(uint32_t)) {
 824       if (v != MH_MAGIC_64) {
 825         continue;
 826       } 
 827     } else {
 828       // may be encountered last map, which is not readable
 829       continue;   
 830     }
 831     while (ltell(fd) - iter->offset < iter->memsz) {
 832       lseek(fd, fpos, SEEK_SET);
 833       if (read(fd, (void *)&v, sizeof(uint32_t)) != sizeof(uint32_t)) {
 834         break;
 835       } 
 836       if (v != MH_MAGIC_64) {
 837         fpos = (ltell(fd) + pagesize -1)/pagesize * pagesize;
 838         continue;
 839       }
 840       lseek(fd, -sizeof(uint32_t), SEEK_CUR);
 841       // this is the file begining to core file.
 842       if (read(fd, (void *)&header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
 843         goto err;
 844       }
 845       fpos = ltell(fd);
 846       
 847       // found a mach-o file in this segment
 848       for (i = 0; i < header.ncmds; i++) {
 849         // read commands in this "file"
 850         // LC_ID_DYLIB is the file itself for a .dylib
 851         lseek(fd, fpos, SEEK_SET);
 852         if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
 853           return false;   // error
 854         }
 855         fpos += lcmd.cmdsize;  // next command position
 856         // make sure still within seg size.
 857         if (fpos  - lcmd.cmdsize - iter->offset > iter->memsz) {
 858           print_debug("Warning: out of segement limit: %ld \n", fpos  - lcmd.cmdsize - iter->offset);
 859           break;  // no need to iterate all commands
 860         }
 861         if (lcmd.cmd == LC_ID_DYLIB) {
 862           lseek(fd, -sizeof(load_command), SEEK_CUR);
 863           if (read(fd, (void *)&dylibcmd, sizeof(dylib_command)) != sizeof(dylib_command)) {
 864             return false;
 865           }
 866           /**** name stored at dylib_command.dylib.name.offset, is a C string  */
 867           lseek(fd, dylibcmd.dylib.name.offset - sizeof(dylib_command), SEEK_CUR);
 868           int j = 0;
 869           while (j < BUF_SIZE) {
 870             read(fd, (void *)(name + j), sizeof(char));
 871             if (name[j] == '\0') break;
 872             j++;
 873           }
 874           print_debug("%s\n", name);
 875           // changed name from @rpath/xxxx.dylib to real path
 876           if (strrchr(name, '@')) {
 877             get_real_path(ph, name);
 878             print_debug("get_real_path returned: %s\n", name);
 879           }    
 880           add_lib_info(ph, name, iter->vaddr);
 881           break; 
 882         }
 883       }
 884       // done with the file, advanced to next page to search more files
 885       fpos = (ltell(fd) + pagesize - 1) / pagesize * pagesize;
 886     }
 887   }
 888   return true;
 889 err:
 890   return false;
 891 }
 892 
 893 bool read_macho64_header(int fd, mach_header_64* core_header) {
 894   bool is_macho = false;
 895   if (fd < 0) return false;
 896   off_t pos = ltell(fd);
 897   lseek(fd, 0, SEEK_SET);
 898   if (read(fd, (void *)core_header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
 899     is_macho = false;
 900   } else {
 901     is_macho = (core_header->magic ==  MH_MAGIC_64 || core_header->magic ==  MH_CIGAM_64);
 902   }
 903   lseek(fd, pos, SEEK_SET);
 904   return is_macho;
 905 }
 906 
 907 // the one and only one exposed stuff from this file
 908 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
 909   mach_header_64 core_header;
 910   mach_header_64 exec_header;
 911 
 912   struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
 913   if (ph == NULL) {
 914     print_debug("cant allocate ps_prochandle\n");
 915     return NULL;
 916   }
 917 
 918   if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
 919     free(ph);
 920     print_debug("can't allocate ps_prochandle\n");
 921     return NULL;
 922   }
 923 
 924   // initialize ph
 925   ph->ops = &core_ops;
 926   ph->core->core_fd   = -1;
 927   ph->core->exec_fd   = -1;
 928   ph->core->interp_fd = -1;
 929 
 930   print_debug("exec: %s   core: %s", exec_file, core_file);
 931 
 932   strncpy(ph->core->exec_path, exec_file, sizeof(ph->core->exec_path));
 933 
 934   // open the core file
 935   if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
 936     print_error("can't open core file\n");
 937     goto err;
 938   }
 939 
 940   // read core file header
 941   if (read_macho64_header(ph->core->core_fd, &core_header) != true || core_header.filetype != MH_CORE) {
 942     print_debug("core file is not a valid Mach-O file\n");
 943     goto err;
 944   }
 945 
 946   if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
 947     print_error("can't open executable file\n");
 948     goto err;
 949   }
 950 
 951   if (read_macho64_header(ph->core->exec_fd, &exec_header) != true ||
 952                           exec_header.filetype != MH_EXECUTE) {
 953     print_error("executable file is not a valid Mach-O file\n");
 954     goto err;
 955   }
 956 
 957   // process core file segments
 958   if (read_core_segments(ph) != true) {
 959     print_error("failed to read core segments\n");
 960     goto err;
 961   }
 962 
 963   // allocate and sort maps into map_array, we need to do this
 964   // here because read_shared_lib_info needs to read from debuggee
 965   // address space
 966   if (sort_map_array(ph) != true) {
 967     print_error("failed to sort segment map array\n");
 968     goto err;
 969   }
 970 
 971   if (read_shared_lib_info(ph) != true) {
 972     print_error("failed to read libraries\n");
 973     goto err;
 974   }
 975 
 976   // sort again because we have added more mappings from shared objects
 977   if (sort_map_array(ph) != true) {
 978     print_error("failed to sort segment map array\n");
 979     goto err;
 980   }
 981 
 982   if (init_classsharing_workaround(ph) != true) {
 983     print_error("failed to workaround classshareing\n");
 984     goto err;
 985   }
 986 
 987   print_debug("Leave Pgrab_core\n"); 
 988   return ph;
 989 
 990 err:
 991   Prelease(ph);
 992   return NULL;
 993 }
 994 
 995 #else // __APPLE__ (none macosx)  
 996 
 997 // read regs and create thread from core file
 998 static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) {
 999    // we have to read prstatus_t from buf
1000    // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t");
1001    prstatus_t* prstat = (prstatus_t*) buf;
1002    sa_thread_info* newthr;
1003    print_debug("got integer regset for lwp %d\n", prstat->pr_pid);
1004    // we set pthread_t to -1 for core dump
1005    if((newthr = add_thread_info(ph, (pthread_t) -1,  prstat->pr_pid)) == NULL)
1006       return false;
1007 
1008    // copy regs
1009    memcpy(&newthr->regs, &prstat->pr_reg, sizeof(struct reg));
1010 
1011    if (is_debug()) {
1012       print_debug("integer regset\n");
1013 #ifdef i386
1014       // print the regset
1015       print_debug("\teax = 0x%x\n", newthr->regs.r_eax);
1016       print_debug("\tebx = 0x%x\n", newthr->regs.r_ebx);
1017       print_debug("\tecx = 0x%x\n", newthr->regs.r_ecx);
1018       print_debug("\tedx = 0x%x\n", newthr->regs.r_edx);
1019       print_debug("\tesp = 0x%x\n", newthr->regs.r_esp);
1020       print_debug("\tebp = 0x%x\n", newthr->regs.r_ebp);
1021       print_debug("\tesi = 0x%x\n", newthr->regs.r_esi);
1022       print_debug("\tedi = 0x%x\n", newthr->regs.r_edi);
1023       print_debug("\teip = 0x%x\n", newthr->regs.r_eip);
1024 #endif
1025 
1026 #if defined(amd64) || defined(x86_64)
1027       // print the regset
1028       print_debug("\tr15 = 0x%lx\n", newthr->regs.r_r15);
1029       print_debug("\tr14 = 0x%lx\n", newthr->regs.r_r14);
1030       print_debug("\tr13 = 0x%lx\n", newthr->regs.r_r13);
1031       print_debug("\tr12 = 0x%lx\n", newthr->regs.r_r12);
1032       print_debug("\trbp = 0x%lx\n", newthr->regs.r_rbp);
1033       print_debug("\trbx = 0x%lx\n", newthr->regs.r_rbx);
1034       print_debug("\tr11 = 0x%lx\n", newthr->regs.r_r11);
1035       print_debug("\tr10 = 0x%lx\n", newthr->regs.r_r10);
1036       print_debug("\tr9 = 0x%lx\n", newthr->regs.r_r9);
1037       print_debug("\tr8 = 0x%lx\n", newthr->regs.r_r8);
1038       print_debug("\trax = 0x%lx\n", newthr->regs.r_rax);
1039       print_debug("\trcx = 0x%lx\n", newthr->regs.r_rcx);
1040       print_debug("\trdx = 0x%lx\n", newthr->regs.r_rdx);
1041       print_debug("\trsi = 0x%lx\n", newthr->regs.r_rsi);
1042       print_debug("\trdi = 0x%lx\n", newthr->regs.r_rdi);
1043       //print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax);
1044       print_debug("\trip = 0x%lx\n", newthr->regs.r_rip);
1045       print_debug("\tcs = 0x%lx\n", newthr->regs.r_cs);
1046       //print_debug("\teflags = 0x%lx\n", newthr->regs.eflags);
1047       print_debug("\trsp = 0x%lx\n", newthr->regs.r_rsp);
1048       print_debug("\tss = 0x%lx\n", newthr->regs.r_ss);
1049       //print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base);
1050       //print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base);
1051       //print_debug("\tds = 0x%lx\n", newthr->regs.ds);
1052       //print_debug("\tes = 0x%lx\n", newthr->regs.es);
1053       //print_debug("\tfs = 0x%lx\n", newthr->regs.fs);
1054       //print_debug("\tgs = 0x%lx\n", newthr->regs.gs);
1055 #endif
1056    }
1057 
1058    return true;
1059 }
1060 
1061 #define ROUNDUP(x, y)  ((((x)+((y)-1))/(y))*(y))
1062 
1063 // read NT_PRSTATUS entries from core NOTE segment
1064 static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) {
1065    char* buf = NULL;
1066    char* p = NULL;
1067    size_t size = note_phdr->p_filesz;
1068 
1069    // we are interested in just prstatus entries. we will ignore the rest.
1070    // Advance the seek pointer to the start of the PT_NOTE data
1071    if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) {
1072       print_debug("failed to lseek to PT_NOTE data\n");
1073       return false;
1074    }
1075 
1076    // Now process the PT_NOTE structures.  Each one is preceded by
1077    // an Elf{32/64}_Nhdr structure describing its type and size.
1078    if ( (buf = (char*) malloc(size)) == NULL) {
1079       print_debug("can't allocate memory for reading core notes\n");
1080       goto err;
1081    }
1082 
1083    // read notes into buffer
1084    if (read(ph->core->core_fd, buf, size) != size) {
1085       print_debug("failed to read notes, core file must have been truncated\n");
1086       goto err;
1087    }
1088 
1089    p = buf;
1090    while (p < buf + size) {
1091       ELF_NHDR* notep = (ELF_NHDR*) p;
1092       char* descdata  = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4);
1093       print_debug("Note header with n_type = %d and n_descsz = %u\n",
1094                                    notep->n_type, notep->n_descsz);
1095 
1096       if (notep->n_type == NT_PRSTATUS) {
1097          if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) {
1098             return false;
1099          }
1100       }
1101       p = descdata + ROUNDUP(notep->n_descsz, 4);
1102    }
1103 
1104    free(buf);
1105    return true;
1106 
1107 err:
1108    if (buf) free(buf);
1109    return false;
1110 }
1111 
1112 // read all segments from core file
1113 static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) {
1114    int i = 0;
1115    ELF_PHDR* phbuf = NULL;
1116    ELF_PHDR* core_php = NULL;
1117 
1118    if ((phbuf =  read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL)
1119       return false;
1120 
1121    /*
1122     * Now iterate through the program headers in the core file.
1123     * We're interested in two types of Phdrs: PT_NOTE (which
1124     * contains a set of saved /proc structures), and PT_LOAD (which
1125     * represents a memory mapping from the process's address space).
1126     *
1127     * Difference b/w Solaris PT_NOTE and BSD PT_NOTE:
1128     *
1129     *     In Solaris there are two PT_NOTE segments the first PT_NOTE (if present)
1130     *     contains /proc structs in the pre-2.6 unstructured /proc format. the last
1131     *     PT_NOTE has data in new /proc format.
1132     *
1133     *     In Solaris, there is only one pstatus (process status). pstatus contains
1134     *     integer register set among other stuff. For each LWP, we have one lwpstatus
1135     *     entry that has integer regset for that LWP.
1136     *
1137     *     Linux threads are actually 'clone'd processes. To support core analysis
1138     *     of "multithreaded" process, Linux creates more than one pstatus (called
1139     *     "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one
1140     *     "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular
1141     *     function "elf_core_dump".
1142     */
1143 
1144     for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) {
1145       switch (core_php->p_type) {
1146          case PT_NOTE:
1147             if (core_handle_note(ph, core_php) != true) {
1148               goto err;
1149             }
1150             break;
1151 
1152          case PT_LOAD: {
1153             if (core_php->p_filesz != 0) {
1154                if (add_map_info(ph, ph->core->core_fd, core_php->p_offset,
1155                   core_php->p_vaddr, core_php->p_filesz) == NULL) goto err;
1156             }
1157             break;
1158          }
1159       }
1160 
1161       core_php++;
1162    }
1163 
1164    free(phbuf);
1165    return true;
1166 err:
1167    free(phbuf);
1168    return false;
1169 }
1170 
1171 // read segments of a shared object
1172 static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) {
1173    int i = 0;
1174    ELF_PHDR* phbuf;
1175    ELF_PHDR* lib_php = NULL;
1176 
1177    if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL)
1178       return false;
1179 
1180    // we want to process only PT_LOAD segments that are not writable.
1181    // i.e., text segments. The read/write/exec (data) segments would
1182    // have been already added from core file segments.
1183    for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) {
1184       if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) {
1185          if (add_map_info(ph, lib_fd, lib_php->p_offset, lib_php->p_vaddr + lib_base, lib_php->p_filesz) == NULL)
1186             goto err;
1187       }
1188       lib_php++;
1189    }
1190 
1191    free(phbuf);
1192    return true;
1193 err:
1194    free(phbuf);
1195    return false;
1196 }
1197 
1198 // process segments from interpreter (ld-elf.so.1)
1199 static bool read_interp_segments(struct ps_prochandle* ph) {
1200    ELF_EHDR interp_ehdr;
1201 
1202    if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) {
1203        print_debug("interpreter is not a valid ELF file\n");
1204        return false;
1205    }
1206 
1207    if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) {
1208        print_debug("can't read segments of interpreter\n");
1209        return false;
1210    }
1211 
1212    return true;
1213 }
1214 
1215 // process segments of a a.out
1216 static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) {
1217    int i = 0;
1218    ELF_PHDR* phbuf = NULL;
1219    ELF_PHDR* exec_php = NULL;
1220 
1221    if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL)
1222       return false;
1223 
1224    for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) {
1225       switch (exec_php->p_type) {
1226 
1227          // add mappings for PT_LOAD segments
1228          case PT_LOAD: {
1229             // add only non-writable segments of non-zero filesz
1230             if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) {
1231                if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err;
1232             }
1233             break;
1234          }
1235 
1236          // read the interpreter and it's segments
1237          case PT_INTERP: {
1238             char interp_name[BUF_SIZE];
1239 
1240             pread(ph->core->exec_fd, interp_name, MIN(exec_php->p_filesz, BUF_SIZE), exec_php->p_offset);
1241             print_debug("ELF interpreter %s\n", interp_name);
1242             // read interpreter segments as well
1243             if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) {
1244                print_debug("can't open runtime loader\n");
1245                goto err;
1246             }
1247             break;
1248          }
1249 
1250          // from PT_DYNAMIC we want to read address of first link_map addr
1251          case PT_DYNAMIC: {
1252             ph->core->dynamic_addr = exec_php->p_vaddr;
1253             print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr);
1254             break;
1255          }
1256 
1257       } // switch
1258       exec_php++;
1259    } // for
1260 
1261    free(phbuf);
1262    return true;
1263 err:
1264    free(phbuf);
1265    return false;
1266 }
1267 
1268 #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug,  r_map)
1269 #define LD_BASE_OFFSET        offsetof(struct r_debug,  r_ldbase)
1270 #define LINK_MAP_ADDR_OFFSET  offsetof(struct link_map, l_addr)
1271 #define LINK_MAP_NAME_OFFSET  offsetof(struct link_map, l_name)
1272 #define LINK_MAP_NEXT_OFFSET  offsetof(struct link_map, l_next)
1273 
1274 // read shared library info from runtime linker's data structures.
1275 // This work is done by librtlb_db in Solaris
1276 static bool read_shared_lib_info(struct ps_prochandle* ph) {
1277   uintptr_t addr = ph->core->dynamic_addr;
1278   uintptr_t debug_base;
1279   uintptr_t first_link_map_addr;
1280   uintptr_t ld_base_addr;
1281   uintptr_t link_map_addr;
1282   uintptr_t lib_base_diff;
1283   uintptr_t lib_base;
1284   uintptr_t lib_name_addr;
1285   char lib_name[BUF_SIZE];
1286   ELF_DYN dyn;
1287   ELF_EHDR elf_ehdr;
1288   int lib_fd;
1289 
1290   // _DYNAMIC has information of the form
1291   //         [tag] [data] [tag] [data] .....
1292   // Both tag and data are pointer sized.
1293   // We look for dynamic info with DT_DEBUG. This has shared object info.
1294   // refer to struct r_debug in link.h
1295 
1296   dyn.d_tag = DT_NULL;
1297   while (dyn.d_tag != DT_DEBUG) {
1298     if (ps_pread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) {
1299       print_debug("can't read debug info from _DYNAMIC\n");
1300       return false;
1301     }
1302     addr += sizeof(ELF_DYN);
1303   }
1304 
1305   // we have got Dyn entry with DT_DEBUG
1306   debug_base = dyn.d_un.d_ptr;
1307   // at debug_base we have struct r_debug. This has first link map in r_map field
1308   if (ps_pread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET,
1309                   &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1310     print_debug("can't read first link map address\n");
1311     return false;
1312   }
1313  
1314   // read ld_base address from struct r_debug
1315   // XXX: There is no r_ldbase member on BSD
1316   /*
1317   if (ps_pread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr,
1318                   sizeof(uintptr_t)) != PS_OK) {
1319     print_debug("can't read ld base address\n");
1320     return false;
1321   }
1322   ph->core->ld_base_addr = ld_base_addr;
1323   */
1324   ph->core->ld_base_addr = 0;
1325  
1326   print_debug("interpreter base address is 0x%lx\n", ld_base_addr);
1327  
1328   // now read segments from interp (i.e ld-elf.so.1)
1329   if (read_interp_segments(ph) != true)
1330     return false;
1331  
1332   // after adding interpreter (ld.so) mappings sort again
1333   if (sort_map_array(ph) != true)
1334     return false;
1335  
1336   print_debug("first link map is at 0x%lx\n", first_link_map_addr);
1337 
1338   link_map_addr = first_link_map_addr;
1339   while (link_map_addr != 0) {
1340     // read library base address of the .so. Note that even though <sys/link.h> calls
1341     // link_map->l_addr as "base address",  this is * not * really base virtual
1342     // address of the shared object. This is actually the difference b/w the virtual
1343     // address mentioned in shared object and the actual virtual base where runtime
1344     // linker loaded it. We use "base diff" in read_lib_segments call below.
1345 
1346     if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET,
1347                  &lib_base_diff, sizeof(uintptr_t)) != PS_OK) {
1348       print_debug("can't read shared object base address diff\n");
1349       return false;
1350     }
1351  
1352     // read address of the name
1353     if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET,
1354                   &lib_name_addr, sizeof(uintptr_t)) != PS_OK) {
1355       print_debug("can't read address of shared object name\n");
1356       return false;
1357     }
1358  
1359     // read name of the shared object
1360     if (read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) {
1361       print_debug("can't read shared object name\n");
1362       return false;
1363     }
1364  
1365     if (lib_name[0] != '\0') {
1366       // ignore empty lib names
1367       lib_fd = pathmap_open(lib_name);
1368  
1369       if (lib_fd < 0) {
1370         print_debug("can't open shared object %s\n", lib_name);
1371         // continue with other libraries...
1372       } else {
1373         if (read_elf_header(lib_fd, &elf_ehdr)) {
1374           lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr);
1375           print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n",
1376                        lib_name, lib_base, lib_base_diff);
1377           // while adding library mappings we need to use "base difference".
1378           if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) {
1379             print_debug("can't read shared object's segments\n");
1380             close(lib_fd);
1381             return false;
1382           }
1383           add_lib_info_fd(ph, lib_name, lib_fd, lib_base);
1384           // Map info is added for the library (lib_name) so
1385           // we need to re-sort it before calling the p_pdread.
1386           if (sort_map_array(ph) != true)
1387             return false;
1388         } else {
1389           print_debug("can't read ELF header for shared object %s\n", lib_name);
1390           close(lib_fd);
1391           // continue with other libraries...
1392         }
1393       }
1394     }
1395  
1396     // read next link_map address
1397     if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET,
1398                  &link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1399       print_debug("can't read next link in link_map\n");
1400       return false;
1401     }
1402   }
1403  
1404   return true;
1405 }
1406  
1407 // the one and only one exposed stuff from this file
1408 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
1409   ELF_EHDR core_ehdr;
1410   ELF_EHDR exec_ehdr;
1411 
1412   struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
1413   if (ph == NULL) {
1414     print_debug("cant allocate ps_prochandle\n");
1415     return NULL;
1416   }
1417 
1418   if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
1419     free(ph);
1420     print_debug("can't allocate ps_prochandle\n");
1421     return NULL;
1422   }
1423 
1424   // initialize ph
1425   ph->ops = &core_ops;
1426   ph->core->core_fd   = -1;
1427   ph->core->exec_fd   = -1;
1428   ph->core->interp_fd = -1;
1429 
1430   print_debug("exec: %s   core: %s", exec_file, core_file);
1431 
1432   // open the core file
1433   if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
1434     print_debug("can't open core file\n");
1435     goto err;
1436   }
1437 
1438   // read core file ELF header
1439   if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) {
1440     print_debug("core file is not a valid ELF ET_CORE file\n");
1441     goto err;
1442   }
1443 
1444   if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
1445     print_debug("can't open executable file\n");
1446     goto err;
1447   }
1448 
1449   if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) {
1450      print_debug("executable file is not a valid ELF ET_EXEC file\n");
1451      goto err;
1452   }
1453 
1454   // process core file segments
1455   if (read_core_segments(ph, &core_ehdr) != true)
1456      goto err;
1457 
1458   // process exec file segments
1459   if (read_exec_segments(ph, &exec_ehdr) != true)
1460      goto err;
1461 
1462   // exec file is also treated like a shared object for symbol search
1463   if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd,
1464                       (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL)
1465      goto err;
1466   
1467   // allocate and sort maps into map_array, we need to do this
1468   // here because read_shared_lib_info needs to read from debuggee
1469   // address space
1470   if (sort_map_array(ph) != true)
1471     goto err;
1472 
1473   if (read_shared_lib_info(ph) != true)
1474     goto err;
1475 
1476   // sort again because we have added more mappings from shared objects
1477   if (sort_map_array(ph) != true)
1478     goto err;
1479 
1480   if (init_classsharing_workaround(ph) != true)
1481     goto err;
1482 
1483   print_debug("Leave Pgrab_core\n"); 
1484   return ph;
1485 
1486 err:
1487   Prelease(ph);
1488   return NULL;
1489 }
1490 
1491 #endif // __APPLE__