1/*
2 * Copyright (c) 2003, 2016, 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
45static void close_files(struct ps_prochandle* ph) {
46  lib_info* lib = NULL;
47
48  // close core file descriptor
49  if (ph->core->core_fd >= 0)
50    close(ph->core->core_fd);
51
52  // close exec file descriptor
53  if (ph->core->exec_fd >= 0)
54    close(ph->core->exec_fd);
55
56  // close interp file descriptor
57  if (ph->core->interp_fd >= 0)
58    close(ph->core->interp_fd);
59
60  // close class share archive file
61  if (ph->core->classes_jsa_fd >= 0)
62    close(ph->core->classes_jsa_fd);
63
64  // close all library file descriptors
65  lib = ph->libs;
66  while (lib) {
67    int fd = lib->fd;
68    if (fd >= 0 && fd != ph->core->exec_fd) {
69      close(fd);
70    }
71    lib = lib->next;
72  }
73}
74
75// clean all map_info stuff
76static void destroy_map_info(struct ps_prochandle* ph) {
77  map_info* map = ph->core->maps;
78  while (map) {
79    map_info* next = map->next;
80    free(map);
81    map = next;
82  }
83
84  if (ph->core->map_array) {
85    free(ph->core->map_array);
86  }
87
88  // Part of the class sharing workaround
89  map = ph->core->class_share_maps;
90  while (map) {
91    map_info* next = map->next;
92    free(map);
93    map = next;
94  }
95}
96
97// ps_prochandle operations
98static void core_release(struct ps_prochandle* ph) {
99  if (ph->core) {
100    close_files(ph);
101    destroy_map_info(ph);
102    free(ph->core);
103  }
104}
105
106static map_info* allocate_init_map(int fd, off_t offset, uintptr_t vaddr, size_t memsz) {
107  map_info* map;
108  if ( (map = (map_info*) calloc(1, sizeof(map_info))) == NULL) {
109    print_debug("can't allocate memory for map_info\n");
110    return NULL;
111  }
112
113  // initialize map
114  map->fd     = fd;
115  map->offset = offset;
116  map->vaddr  = vaddr;
117  map->memsz  = memsz;
118  return map;
119}
120
121// add map info with given fd, offset, vaddr and memsz
122static map_info* add_map_info(struct ps_prochandle* ph, int fd, off_t offset,
123                             uintptr_t vaddr, size_t memsz) {
124  map_info* map;
125  if ((map = allocate_init_map(fd, offset, vaddr, memsz)) == NULL) {
126    return NULL;
127  }
128
129  // add this to map list
130  map->next  = ph->core->maps;
131  ph->core->maps   = map;
132  ph->core->num_maps++;
133
134  return map;
135}
136
137// Part of the class sharing workaround
138static map_info* add_class_share_map_info(struct ps_prochandle* ph, off_t offset,
139                             uintptr_t vaddr, size_t memsz) {
140  map_info* map;
141  if ((map = allocate_init_map(ph->core->classes_jsa_fd,
142                               offset, vaddr, memsz)) == NULL) {
143    return NULL;
144  }
145
146  map->next = ph->core->class_share_maps;
147  ph->core->class_share_maps = map;
148  return map;
149}
150
151// Return the map_info for the given virtual address.  We keep a sorted
152// array of pointers in ph->map_array, so we can binary search.
153static map_info* core_lookup(struct ps_prochandle *ph, uintptr_t addr) {
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.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.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
221struct 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.
234    // We assume the corresponding C type is char but licensees
235    // 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
244static 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
254static 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
265static 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#define LIBJVM_NAME "/libjvm.dylib"
290#else
291#define USE_SHARED_SPACES_SYM "UseSharedSpaces"
292// mangled name of Arguments::SharedArchivePath
293#define SHARED_ARCHIVE_PATH_SYM "__ZN9Arguments17SharedArchivePathE"
294#define LIBJVM_NAME "/libjvm.so"
295#endif // __APPLE_
296
297static bool init_classsharing_workaround(struct ps_prochandle* ph) {
298  int m;
299  size_t n;
300  lib_info* lib = ph->libs;
301  while (lib != NULL) {
302    // we are iterating over shared objects from the core dump. look for
303    // libjvm.so.
304    const char *jvm_name = 0;
305    if ((jvm_name = strstr(lib->name, LIBJVM_NAME)) != 0) {
306      char classes_jsa[PATH_MAX];
307      struct FileMapHeader header;
308      int fd = -1;
309      uintptr_t base = 0, useSharedSpacesAddr = 0;
310      uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0;
311      jboolean useSharedSpaces = 0;
312
313      memset(classes_jsa, 0, sizeof(classes_jsa));
314      jvm_name = lib->name;
315      useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM);
316      if (useSharedSpacesAddr == 0) {
317        print_debug("can't lookup 'UseSharedSpaces' flag\n");
318        return false;
319      }
320
321      // Hotspot vm types are not exported to build this library. So
322      // using equivalent type jboolean to read the value of
323      // UseSharedSpaces which is same as hotspot type "bool".
324      if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) {
325        print_debug("can't read the value of 'UseSharedSpaces' flag\n");
326        return false;
327      }
328
329      if ((int)useSharedSpaces == 0) {
330        print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n");
331        return true;
332      }
333
334      sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM);
335      if (sharedArchivePathAddrAddr == 0) {
336        print_debug("can't lookup shared archive path symbol\n");
337        return false;
338      }
339
340      if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) {
341        print_debug("can't read shared archive path pointer\n");
342        return false;
343      }
344
345      if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) {
346        print_debug("can't read shared archive path value\n");
347        return false;
348      }
349
350      print_debug("looking for %s\n", classes_jsa);
351      // open the class sharing archive file
352      fd = pathmap_open(classes_jsa);
353      if (fd < 0) {
354        print_debug("can't open %s!\n", classes_jsa);
355        ph->core->classes_jsa_fd = -1;
356        return false;
357      } else {
358        print_debug("opened %s\n", classes_jsa);
359      }
360
361      // read FileMapHeader from the file
362      memset(&header, 0, sizeof(struct FileMapHeader));
363      if ((n = read(fd, &header, sizeof(struct FileMapHeader)))
364           != sizeof(struct FileMapHeader)) {
365        print_debug("can't read shared archive file map header from %s\n", classes_jsa);
366        close(fd);
367        return false;
368      }
369
370      // check file magic
371      if (header._magic != 0xf00baba2) {
372        print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n",
373                     classes_jsa, header._magic);
374        close(fd);
375        return false;
376      }
377
378      // check version
379      if (header._version != CURRENT_ARCHIVE_VERSION) {
380        print_debug("%s has wrong shared archive file version %d, expecting %d\n",
381                     classes_jsa, header._version, CURRENT_ARCHIVE_VERSION);
382        close(fd);
383        return false;
384      }
385
386      ph->core->classes_jsa_fd = fd;
387      // add read-only maps from classes.jsa to the list of maps
388      for (m = 0; m < NUM_SHARED_MAPS; m++) {
389        if (header._space[m]._read_only) {
390          base = (uintptr_t) header._space[m]._base;
391          // no need to worry about the fractional pages at-the-end.
392          // possible fractional pages are handled by core_read_data.
393          add_class_share_map_info(ph, (off_t) header._space[m]._file_offset,
394                                   base, (size_t) header._space[m]._used);
395          print_debug("added a share archive map at 0x%lx\n", base);
396        }
397      }
398      return true;
399   }
400   lib = lib->next;
401  }
402  return true;
403}
404
405//---------------------------------------------------------------------------
406// functions to handle map_info
407
408// Order mappings based on virtual address.  We use this function as the
409// callback for sorting the array of map_info pointers.
410static int core_cmp_mapping(const void *lhsp, const void *rhsp)
411{
412  const map_info *lhs = *((const map_info **)lhsp);
413  const map_info *rhs = *((const map_info **)rhsp);
414
415  if (lhs->vaddr == rhs->vaddr) {
416    return (0);
417  }
418
419  return (lhs->vaddr < rhs->vaddr ? -1 : 1);
420}
421
422// we sort map_info by starting virtual address so that we can do
423// binary search to read from an address.
424static bool sort_map_array(struct ps_prochandle* ph) {
425  size_t num_maps = ph->core->num_maps;
426  map_info* map = ph->core->maps;
427  int i = 0;
428
429  // allocate map_array
430  map_info** array;
431  if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) {
432    print_debug("can't allocate memory for map array\n");
433    return false;
434  }
435
436  // add maps to array
437  while (map) {
438    array[i] = map;
439    i++;
440    map = map->next;
441  }
442
443  // sort is called twice. If this is second time, clear map array
444  if (ph->core->map_array) {
445    free(ph->core->map_array);
446  }
447  ph->core->map_array = array;
448  // sort the map_info array by base virtual address.
449  qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*),
450        core_cmp_mapping);
451
452  // print map
453  if (is_debug()) {
454    int j = 0;
455    print_debug("---- sorted virtual address map ----\n");
456    for (j = 0; j < ph->core->num_maps; j++) {
457      print_debug("base = 0x%lx\tsize = %d\n", ph->core->map_array[j]->vaddr,
458                  ph->core->map_array[j]->memsz);
459    }
460  }
461
462  return true;
463}
464
465#ifndef MIN
466#define MIN(x, y) (((x) < (y))? (x): (y))
467#endif
468
469static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) {
470   ssize_t resid = size;
471   int page_size=sysconf(_SC_PAGE_SIZE);
472   while (resid != 0) {
473      map_info *mp = core_lookup(ph, addr);
474      uintptr_t mapoff;
475      ssize_t len, rem;
476      off_t off;
477      int fd;
478
479      if (mp == NULL) {
480         break;  /* No mapping for this address */
481      }
482
483      fd = mp->fd;
484      mapoff = addr - mp->vaddr;
485      len = MIN(resid, mp->memsz - mapoff);
486      off = mp->offset + mapoff;
487
488      if ((len = pread(fd, buf, len, off)) <= 0) {
489         break;
490      }
491
492      resid -= len;
493      addr += len;
494      buf = (char *)buf + len;
495
496      // mappings always start at page boundary. But, may end in fractional
497      // page. fill zeros for possible fractional page at the end of a mapping.
498      rem = mp->memsz % page_size;
499      if (rem > 0) {
500         rem = page_size - rem;
501         len = MIN(resid, rem);
502         resid -= len;
503         addr += len;
504         // we are not assuming 'buf' to be zero initialized.
505         memset(buf, 0, len);
506         buf += len;
507      }
508   }
509
510   if (resid) {
511      print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n",
512              size, addr, resid);
513      return false;
514   } else {
515      return true;
516   }
517}
518
519// null implementation for write
520static bool core_write_data(struct ps_prochandle* ph,
521                             uintptr_t addr, const char *buf , size_t size) {
522   return false;
523}
524
525static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id,
526                          struct reg* regs) {
527   // for core we have cached the lwp regs after segment parsed
528   sa_thread_info* thr = ph->threads;
529   while (thr) {
530     if (thr->lwp_id == lwp_id) {
531       memcpy(regs, &thr->regs, sizeof(struct reg));
532       return true;
533     }
534     thr = thr->next;
535   }
536   return false;
537}
538
539static bool core_get_lwp_info(struct ps_prochandle *ph, lwpid_t id, void *info) {
540   print_debug("core_get_lwp_info not implemented\n");
541   return false;
542}
543
544static ps_prochandle_ops core_ops = {
545   .release=  core_release,
546   .p_pread=  core_read_data,
547   .p_pwrite= core_write_data,
548   .get_lwp_regs= core_get_lwp_regs,
549   .get_lwp_info= core_get_lwp_info
550};
551
552// from this point, mainly two blocks divided by def __APPLE__
553// one for Macosx, the other for regular Bsd
554
555#ifdef __APPLE__
556
557void print_thread(sa_thread_info *threadinfo) {
558  print_debug("thread added: %d\n", threadinfo->lwp_id);
559  print_debug("registers:\n");
560  print_debug("  r_r15: 0x%" PRIx64 "\n", threadinfo->regs.r_r15);
561  print_debug("  r_r14: 0x%" PRIx64 "\n", threadinfo->regs.r_r14);
562  print_debug("  r_r13: 0x%" PRIx64 "\n", threadinfo->regs.r_r13);
563  print_debug("  r_r12: 0x%" PRIx64 "\n", threadinfo->regs.r_r12);
564  print_debug("  r_r11: 0x%" PRIx64 "\n", threadinfo->regs.r_r11);
565  print_debug("  r_r10: 0x%" PRIx64 "\n", threadinfo->regs.r_r10);
566  print_debug("  r_r9:  0x%" PRIx64 "\n", threadinfo->regs.r_r9);
567  print_debug("  r_r8:  0x%" PRIx64 "\n", threadinfo->regs.r_r8);
568  print_debug("  r_rdi: 0x%" PRIx64 "\n", threadinfo->regs.r_rdi);
569  print_debug("  r_rsi: 0x%" PRIx64 "\n", threadinfo->regs.r_rsi);
570  print_debug("  r_rbp: 0x%" PRIx64 "\n", threadinfo->regs.r_rbp);
571  print_debug("  r_rbx: 0x%" PRIx64 "\n", threadinfo->regs.r_rbx);
572  print_debug("  r_rdx: 0x%" PRIx64 "\n", threadinfo->regs.r_rdx);
573  print_debug("  r_rcx: 0x%" PRIx64 "\n", threadinfo->regs.r_rcx);
574  print_debug("  r_rax: 0x%" PRIx64 "\n", threadinfo->regs.r_rax);
575  print_debug("  r_fs:  0x%" PRIx32 "\n", threadinfo->regs.r_fs);
576  print_debug("  r_gs:  0x%" PRIx32 "\n", threadinfo->regs.r_gs);
577  print_debug("  r_rip  0x%" PRIx64 "\n", threadinfo->regs.r_rip);
578  print_debug("  r_cs:  0x%" PRIx64 "\n", threadinfo->regs.r_cs);
579  print_debug("  r_rsp: 0x%" PRIx64 "\n", threadinfo->regs.r_rsp);
580  print_debug("  r_rflags: 0x%" PRIx64 "\n", threadinfo->regs.r_rflags);
581}
582
583// read all segments64 commands from core file
584// read all thread commands from core file
585static bool read_core_segments(struct ps_prochandle* ph) {
586  int i = 0;
587  int num_threads = 0;
588  int fd = ph->core->core_fd;
589  off_t offset = 0;
590  mach_header_64      fhead;
591  load_command        lcmd;
592  segment_command_64  segcmd;
593  // thread_command      thrcmd;
594
595  lseek(fd, offset, SEEK_SET);
596  if(read(fd, (void *)&fhead, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
597     goto err;
598  }
599  print_debug("total commands: %d\n", fhead.ncmds);
600  offset += sizeof(mach_header_64);
601  for (i = 0; i < fhead.ncmds; i++) {
602    lseek(fd, offset, SEEK_SET);
603    if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
604      goto err;
605    }
606    offset += lcmd.cmdsize;    // next command position
607    if (lcmd.cmd == LC_SEGMENT_64) {
608      lseek(fd, -sizeof(load_command), SEEK_CUR);
609      if (read(fd, (void *)&segcmd, sizeof(segment_command_64)) != sizeof(segment_command_64)) {
610        print_debug("failed to read LC_SEGMENT_64 i = %d!\n", i);
611        goto err;
612      }
613      if (add_map_info(ph, fd, segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize) == NULL) {
614        print_debug("Failed to add map_info at i = %d\n", i);
615        goto err;
616      }
617      print_debug("segment added: %" PRIu64 " 0x%" PRIx64 " %d\n",
618                   segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize);
619    } else if (lcmd.cmd == LC_THREAD || lcmd.cmd == LC_UNIXTHREAD) {
620      typedef struct thread_fc {
621        uint32_t  flavor;
622        uint32_t  count;
623      } thread_fc;
624      thread_fc fc;
625      uint32_t size = sizeof(load_command);
626      while (size < lcmd.cmdsize) {
627        if (read(fd, (void *)&fc, sizeof(thread_fc)) != sizeof(thread_fc)) {
628          printf("Reading flavor, count failed.\n");
629          goto err;
630        }
631        size += sizeof(thread_fc);
632        if (fc.flavor == x86_THREAD_STATE) {
633          x86_thread_state_t thrstate;
634          if (read(fd, (void *)&thrstate, sizeof(x86_thread_state_t)) != sizeof(x86_thread_state_t)) {
635            printf("Reading flavor, count failed.\n");
636            goto err;
637          }
638          size += sizeof(x86_thread_state_t);
639          // create thread info list, update lwp_id later
640          sa_thread_info* newthr = add_thread_info(ph, (pthread_t) -1, (lwpid_t) num_threads++);
641          if (newthr == NULL) {
642            printf("create thread_info failed\n");
643            goto err;
644          }
645
646          // note __DARWIN_UNIX03 depengs on other definitions
647#if __DARWIN_UNIX03
648#define get_register_v(regst, regname) \
649  regst.uts.ts64.__##regname
650#else
651#define get_register_v(regst, regname) \
652  regst.uts.ts64.##regname
653#endif // __DARWIN_UNIX03
654          newthr->regs.r_rax = get_register_v(thrstate, rax);
655          newthr->regs.r_rbx = get_register_v(thrstate, rbx);
656          newthr->regs.r_rcx = get_register_v(thrstate, rcx);
657          newthr->regs.r_rdx = get_register_v(thrstate, rdx);
658          newthr->regs.r_rdi = get_register_v(thrstate, rdi);
659          newthr->regs.r_rsi = get_register_v(thrstate, rsi);
660          newthr->regs.r_rbp = get_register_v(thrstate, rbp);
661          newthr->regs.r_rsp = get_register_v(thrstate, rsp);
662          newthr->regs.r_r8  = get_register_v(thrstate, r8);
663          newthr->regs.r_r9  = get_register_v(thrstate, r9);
664          newthr->regs.r_r10 = get_register_v(thrstate, r10);
665          newthr->regs.r_r11 = get_register_v(thrstate, r11);
666          newthr->regs.r_r12 = get_register_v(thrstate, r12);
667          newthr->regs.r_r13 = get_register_v(thrstate, r13);
668          newthr->regs.r_r14 = get_register_v(thrstate, r14);
669          newthr->regs.r_r15 = get_register_v(thrstate, r15);
670          newthr->regs.r_rip = get_register_v(thrstate, rip);
671          newthr->regs.r_rflags = get_register_v(thrstate, rflags);
672          newthr->regs.r_cs  = get_register_v(thrstate, cs);
673          newthr->regs.r_fs  = get_register_v(thrstate, fs);
674          newthr->regs.r_gs  = get_register_v(thrstate, gs);
675          print_thread(newthr);
676        } else if (fc.flavor == x86_FLOAT_STATE) {
677          x86_float_state_t flstate;
678          if (read(fd, (void *)&flstate, sizeof(x86_float_state_t)) != sizeof(x86_float_state_t)) {
679            print_debug("Reading flavor, count failed.\n");
680            goto err;
681          }
682          size += sizeof(x86_float_state_t);
683        } else if (fc.flavor == x86_EXCEPTION_STATE) {
684          x86_exception_state_t excpstate;
685          if (read(fd, (void *)&excpstate, sizeof(x86_exception_state_t)) != sizeof(x86_exception_state_t)) {
686            printf("Reading flavor, count failed.\n");
687            goto err;
688          }
689          size += sizeof(x86_exception_state_t);
690        }
691      }
692    }
693  }
694  return true;
695err:
696  return false;
697}
698
699/**local function **/
700bool exists(const char *fname) {
701  return access(fname, F_OK) == 0;
702}
703
704// we check: 1. lib
705//           2. lib/server
706//           3. jre/lib
707//           4. jre/lib/server
708// from: 1. exe path
709//       2. JAVA_HOME
710//       3. DYLD_LIBRARY_PATH
711static bool get_real_path(struct ps_prochandle* ph, char *rpath) {
712  /** check if they exist in JAVA ***/
713  char* execname = ph->core->exec_path;
714  char  filepath[4096];
715  char* filename = strrchr(rpath, '/');               // like /libjvm.dylib
716  if (filename == NULL) {
717    return false;
718  }
719
720  char* posbin = strstr(execname, "/bin/java");
721  if (posbin != NULL) {
722    memcpy(filepath, execname, posbin - execname);    // not include trailing '/'
723    filepath[posbin - execname] = '\0';
724  } else {
725    char* java_home = getenv("JAVA_HOME");
726    if (java_home != NULL) {
727      strcpy(filepath, java_home);
728    } else {
729      char* dyldpath = getenv("DYLD_LIBRARY_PATH");
730      char* dypath = strtok(dyldpath, ":");
731      while (dypath != NULL) {
732        strcpy(filepath, dypath);
733        strcat(filepath, filename);
734        if (exists(filepath)) {
735           strcpy(rpath, filepath);
736           return true;
737        }
738        dypath = strtok(dyldpath, ":");
739      }
740      // not found
741      return false;
742    }
743  }
744  // for exec and java_home, jdkpath now is filepath
745  size_t filepath_base_size = strlen(filepath);
746
747  // first try /lib/ and /lib/server
748  strcat(filepath, "/lib");
749  strcat(filepath, filename);
750  if (exists(filepath)) {
751    strcpy(rpath, filepath);
752    return true;
753  }
754  char* pos = strstr(filepath, filename);    // like /libjvm.dylib
755  *pos = '\0';
756  strcat(filepath, "/server");
757  strcat(filepath, filename);
758  if (exists(filepath)) {
759    strcpy(rpath, filepath);
760    return true;
761  }
762
763  // then try /jre/lib/ and /jre/lib/server
764  filepath[filepath_base_size] = '\0';
765  strcat(filepath, "/jre/lib");
766  strcat(filepath, filename);
767  if (exists(filepath)) {
768    strcpy(rpath, filepath);
769    return true;
770  }
771  pos = strstr(filepath, filename);
772  *pos = '\0';
773  strcat(filepath, "/server");
774  strcat(filepath, filename);
775  if (exists(filepath)) {
776    strcpy(rpath, filepath);
777    return true;
778  }
779
780  return false;
781}
782
783static bool read_shared_lib_info(struct ps_prochandle* ph) {
784  static int pagesize = 0;
785  int fd = ph->core->core_fd;
786  int i = 0, j;
787  uint32_t  v;
788  mach_header_64 header;        // used to check if a file header in segment
789  load_command lcmd;
790  dylib_command dylibcmd;
791
792  char name[BUF_SIZE];  // use to store name
793
794  if (pagesize == 0) {
795    pagesize = getpagesize();
796    print_debug("page size is %d\n", pagesize);
797  }
798  for (j = 0; j < ph->core->num_maps; j++) {
799    map_info *iter = ph->core->map_array[j];   // head
800    off_t fpos = iter->offset;
801    if (iter->fd != fd) {
802      // only search core file!
803      continue;
804    }
805    print_debug("map_info %d: vmaddr = 0x%016" PRIx64 "  fileoff = %" PRIu64 "  vmsize = %" PRIu64 "\n",
806                           j, iter->vaddr, iter->offset, iter->memsz);
807    lseek(fd, fpos, SEEK_SET);
808    // we assume .dylib loaded at segment address --- which is true for JVM libraries
809    // multiple files may be loaded in one segment.
810    // if first word is not a magic word, means this segment does not contain lib file.
811    if (read(fd, (void *)&v, sizeof(uint32_t)) == sizeof(uint32_t)) {
812      if (v != MH_MAGIC_64) {
813        continue;
814      }
815    } else {
816      // may be encountered last map, which is not readable
817      continue;
818    }
819    while (ltell(fd) - iter->offset < iter->memsz) {
820      lseek(fd, fpos, SEEK_SET);
821      if (read(fd, (void *)&v, sizeof(uint32_t)) != sizeof(uint32_t)) {
822        break;
823      }
824      if (v != MH_MAGIC_64) {
825        fpos = (ltell(fd) + pagesize -1)/pagesize * pagesize;
826        continue;
827      }
828      lseek(fd, -sizeof(uint32_t), SEEK_CUR);
829      // this is the file begining to core file.
830      if (read(fd, (void *)&header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
831        goto err;
832      }
833      fpos = ltell(fd);
834
835      // found a mach-o file in this segment
836      for (i = 0; i < header.ncmds; i++) {
837        // read commands in this "file"
838        // LC_ID_DYLIB is the file itself for a .dylib
839        lseek(fd, fpos, SEEK_SET);
840        if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
841          return false;   // error
842        }
843        fpos += lcmd.cmdsize;  // next command position
844        // make sure still within seg size.
845        if (fpos  - lcmd.cmdsize - iter->offset > iter->memsz) {
846          print_debug("Warning: out of segement limit: %ld \n", fpos  - lcmd.cmdsize - iter->offset);
847          break;  // no need to iterate all commands
848        }
849        if (lcmd.cmd == LC_ID_DYLIB) {
850          lseek(fd, -sizeof(load_command), SEEK_CUR);
851          if (read(fd, (void *)&dylibcmd, sizeof(dylib_command)) != sizeof(dylib_command)) {
852            return false;
853          }
854          /**** name stored at dylib_command.dylib.name.offset, is a C string  */
855          lseek(fd, dylibcmd.dylib.name.offset - sizeof(dylib_command), SEEK_CUR);
856          int j = 0;
857          while (j < BUF_SIZE) {
858            read(fd, (void *)(name + j), sizeof(char));
859            if (name[j] == '\0') break;
860            j++;
861          }
862          print_debug("%s\n", name);
863          // changed name from @rpath/xxxx.dylib to real path
864          if (strrchr(name, '@')) {
865            get_real_path(ph, name);
866            print_debug("get_real_path returned: %s\n", name);
867          }
868          add_lib_info(ph, name, iter->vaddr);
869          break;
870        }
871      }
872      // done with the file, advanced to next page to search more files
873      fpos = (ltell(fd) + pagesize - 1) / pagesize * pagesize;
874    }
875  }
876  return true;
877err:
878  return false;
879}
880
881bool read_macho64_header(int fd, mach_header_64* core_header) {
882  bool is_macho = false;
883  if (fd < 0) return false;
884  off_t pos = ltell(fd);
885  lseek(fd, 0, SEEK_SET);
886  if (read(fd, (void *)core_header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
887    is_macho = false;
888  } else {
889    is_macho = (core_header->magic ==  MH_MAGIC_64 || core_header->magic ==  MH_CIGAM_64);
890  }
891  lseek(fd, pos, SEEK_SET);
892  return is_macho;
893}
894
895// the one and only one exposed stuff from this file
896struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
897  mach_header_64 core_header;
898  mach_header_64 exec_header;
899
900  struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
901  if (ph == NULL) {
902    print_debug("cant allocate ps_prochandle\n");
903    return NULL;
904  }
905
906  if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
907    free(ph);
908    print_debug("can't allocate ps_prochandle\n");
909    return NULL;
910  }
911
912  // initialize ph
913  ph->ops = &core_ops;
914  ph->core->core_fd   = -1;
915  ph->core->exec_fd   = -1;
916  ph->core->interp_fd = -1;
917
918  print_debug("exec: %s   core: %s", exec_file, core_file);
919
920  strncpy(ph->core->exec_path, exec_file, sizeof(ph->core->exec_path));
921
922  // open the core file
923  if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
924    print_error("can't open core file\n");
925    goto err;
926  }
927
928  // read core file header
929  if (read_macho64_header(ph->core->core_fd, &core_header) != true || core_header.filetype != MH_CORE) {
930    print_debug("core file is not a valid Mach-O file\n");
931    goto err;
932  }
933
934  if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
935    print_error("can't open executable file\n");
936    goto err;
937  }
938
939  if (read_macho64_header(ph->core->exec_fd, &exec_header) != true ||
940                          exec_header.filetype != MH_EXECUTE) {
941    print_error("executable file is not a valid Mach-O file\n");
942    goto err;
943  }
944
945  // process core file segments
946  if (read_core_segments(ph) != true) {
947    print_error("failed to read core segments\n");
948    goto err;
949  }
950
951  // allocate and sort maps into map_array, we need to do this
952  // here because read_shared_lib_info needs to read from debuggee
953  // address space
954  if (sort_map_array(ph) != true) {
955    print_error("failed to sort segment map array\n");
956    goto err;
957  }
958
959  if (read_shared_lib_info(ph) != true) {
960    print_error("failed to read libraries\n");
961    goto err;
962  }
963
964  // sort again because we have added more mappings from shared objects
965  if (sort_map_array(ph) != true) {
966    print_error("failed to sort segment map array\n");
967    goto err;
968  }
969
970  if (init_classsharing_workaround(ph) != true) {
971    print_error("failed to workaround classshareing\n");
972    goto err;
973  }
974
975  print_debug("Leave Pgrab_core\n");
976  return ph;
977
978err:
979  Prelease(ph);
980  return NULL;
981}
982
983#else // __APPLE__ (none macosx)
984
985// read regs and create thread from core file
986static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) {
987   // we have to read prstatus_t from buf
988   // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t");
989   prstatus_t* prstat = (prstatus_t*) buf;
990   sa_thread_info* newthr;
991   print_debug("got integer regset for lwp %d\n", prstat->pr_pid);
992   // we set pthread_t to -1 for core dump
993   if((newthr = add_thread_info(ph, (pthread_t) -1,  prstat->pr_pid)) == NULL)
994      return false;
995
996   // copy regs
997   memcpy(&newthr->regs, &prstat->pr_reg, sizeof(struct reg));
998
999   if (is_debug()) {
1000      print_debug("integer regset\n");
1001#ifdef i386
1002      // print the regset
1003      print_debug("\teax = 0x%x\n", newthr->regs.r_eax);
1004      print_debug("\tebx = 0x%x\n", newthr->regs.r_ebx);
1005      print_debug("\tecx = 0x%x\n", newthr->regs.r_ecx);
1006      print_debug("\tedx = 0x%x\n", newthr->regs.r_edx);
1007      print_debug("\tesp = 0x%x\n", newthr->regs.r_esp);
1008      print_debug("\tebp = 0x%x\n", newthr->regs.r_ebp);
1009      print_debug("\tesi = 0x%x\n", newthr->regs.r_esi);
1010      print_debug("\tedi = 0x%x\n", newthr->regs.r_edi);
1011      print_debug("\teip = 0x%x\n", newthr->regs.r_eip);
1012#endif
1013
1014#if defined(amd64) || defined(x86_64)
1015      // print the regset
1016      print_debug("\tr15 = 0x%lx\n", newthr->regs.r_r15);
1017      print_debug("\tr14 = 0x%lx\n", newthr->regs.r_r14);
1018      print_debug("\tr13 = 0x%lx\n", newthr->regs.r_r13);
1019      print_debug("\tr12 = 0x%lx\n", newthr->regs.r_r12);
1020      print_debug("\trbp = 0x%lx\n", newthr->regs.r_rbp);
1021      print_debug("\trbx = 0x%lx\n", newthr->regs.r_rbx);
1022      print_debug("\tr11 = 0x%lx\n", newthr->regs.r_r11);
1023      print_debug("\tr10 = 0x%lx\n", newthr->regs.r_r10);
1024      print_debug("\tr9 = 0x%lx\n", newthr->regs.r_r9);
1025      print_debug("\tr8 = 0x%lx\n", newthr->regs.r_r8);
1026      print_debug("\trax = 0x%lx\n", newthr->regs.r_rax);
1027      print_debug("\trcx = 0x%lx\n", newthr->regs.r_rcx);
1028      print_debug("\trdx = 0x%lx\n", newthr->regs.r_rdx);
1029      print_debug("\trsi = 0x%lx\n", newthr->regs.r_rsi);
1030      print_debug("\trdi = 0x%lx\n", newthr->regs.r_rdi);
1031      //print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax);
1032      print_debug("\trip = 0x%lx\n", newthr->regs.r_rip);
1033      print_debug("\tcs = 0x%lx\n", newthr->regs.r_cs);
1034      //print_debug("\teflags = 0x%lx\n", newthr->regs.eflags);
1035      print_debug("\trsp = 0x%lx\n", newthr->regs.r_rsp);
1036      print_debug("\tss = 0x%lx\n", newthr->regs.r_ss);
1037      //print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base);
1038      //print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base);
1039      //print_debug("\tds = 0x%lx\n", newthr->regs.ds);
1040      //print_debug("\tes = 0x%lx\n", newthr->regs.es);
1041      //print_debug("\tfs = 0x%lx\n", newthr->regs.fs);
1042      //print_debug("\tgs = 0x%lx\n", newthr->regs.gs);
1043#endif
1044   }
1045
1046   return true;
1047}
1048
1049#define ROUNDUP(x, y)  ((((x)+((y)-1))/(y))*(y))
1050
1051// read NT_PRSTATUS entries from core NOTE segment
1052static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) {
1053   char* buf = NULL;
1054   char* p = NULL;
1055   size_t size = note_phdr->p_filesz;
1056
1057   // we are interested in just prstatus entries. we will ignore the rest.
1058   // Advance the seek pointer to the start of the PT_NOTE data
1059   if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) {
1060      print_debug("failed to lseek to PT_NOTE data\n");
1061      return false;
1062   }
1063
1064   // Now process the PT_NOTE structures.  Each one is preceded by
1065   // an Elf{32/64}_Nhdr structure describing its type and size.
1066   if ( (buf = (char*) malloc(size)) == NULL) {
1067      print_debug("can't allocate memory for reading core notes\n");
1068      goto err;
1069   }
1070
1071   // read notes into buffer
1072   if (read(ph->core->core_fd, buf, size) != size) {
1073      print_debug("failed to read notes, core file must have been truncated\n");
1074      goto err;
1075   }
1076
1077   p = buf;
1078   while (p < buf + size) {
1079      ELF_NHDR* notep = (ELF_NHDR*) p;
1080      char* descdata  = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4);
1081      print_debug("Note header with n_type = %d and n_descsz = %u\n",
1082                                   notep->n_type, notep->n_descsz);
1083
1084      if (notep->n_type == NT_PRSTATUS) {
1085        if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) {
1086          return false;
1087        }
1088      }
1089      p = descdata + ROUNDUP(notep->n_descsz, 4);
1090   }
1091
1092   free(buf);
1093   return true;
1094
1095err:
1096   if (buf) free(buf);
1097   return false;
1098}
1099
1100// read all segments from core file
1101static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) {
1102   int i = 0;
1103   ELF_PHDR* phbuf = NULL;
1104   ELF_PHDR* core_php = NULL;
1105
1106   if ((phbuf =  read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL)
1107      return false;
1108
1109   /*
1110    * Now iterate through the program headers in the core file.
1111    * We're interested in two types of Phdrs: PT_NOTE (which
1112    * contains a set of saved /proc structures), and PT_LOAD (which
1113    * represents a memory mapping from the process's address space).
1114    *
1115    * Difference b/w Solaris PT_NOTE and Linux/BSD PT_NOTE:
1116    *
1117    *     In Solaris there are two PT_NOTE segments the first PT_NOTE (if present)
1118    *     contains /proc structs in the pre-2.6 unstructured /proc format. the last
1119    *     PT_NOTE has data in new /proc format.
1120    *
1121    *     In Solaris, there is only one pstatus (process status). pstatus contains
1122    *     integer register set among other stuff. For each LWP, we have one lwpstatus
1123    *     entry that has integer regset for that LWP.
1124    *
1125    *     Linux threads are actually 'clone'd processes. To support core analysis
1126    *     of "multithreaded" process, Linux creates more than one pstatus (called
1127    *     "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one
1128    *     "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular
1129    *     function "elf_core_dump".
1130    */
1131
1132    for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) {
1133      switch (core_php->p_type) {
1134         case PT_NOTE:
1135            if (core_handle_note(ph, core_php) != true) {
1136              goto err;
1137            }
1138            break;
1139
1140         case PT_LOAD: {
1141            if (core_php->p_filesz != 0) {
1142               if (add_map_info(ph, ph->core->core_fd, core_php->p_offset,
1143                  core_php->p_vaddr, core_php->p_filesz) == NULL) goto err;
1144            }
1145            break;
1146         }
1147      }
1148
1149      core_php++;
1150   }
1151
1152   free(phbuf);
1153   return true;
1154err:
1155   free(phbuf);
1156   return false;
1157}
1158
1159// read segments of a shared object
1160static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) {
1161  int i = 0;
1162  ELF_PHDR* phbuf;
1163  ELF_PHDR* lib_php = NULL;
1164
1165  int page_size=sysconf(_SC_PAGE_SIZE);
1166
1167  if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL) {
1168    return false;
1169  }
1170
1171  // we want to process only PT_LOAD segments that are not writable.
1172  // i.e., text segments. The read/write/exec (data) segments would
1173  // have been already added from core file segments.
1174  for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) {
1175    if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) {
1176
1177      uintptr_t target_vaddr = lib_php->p_vaddr + lib_base;
1178      map_info *existing_map = core_lookup(ph, target_vaddr);
1179
1180      if (existing_map == NULL){
1181        if (add_map_info(ph, lib_fd, lib_php->p_offset,
1182                          target_vaddr, lib_php->p_filesz) == NULL) {
1183          goto err;
1184        }
1185      } else {
1186        if ((existing_map->memsz != page_size) &&
1187            (existing_map->fd != lib_fd) &&
1188            (existing_map->memsz != lib_php->p_filesz)){
1189
1190          print_debug("address conflict @ 0x%lx (size = %ld, flags = %d\n)",
1191                        target_vaddr, lib_php->p_filesz, lib_php->p_flags);
1192          goto err;
1193        }
1194
1195        /* replace PT_LOAD segment with library segment */
1196        print_debug("overwrote with new address mapping (memsz %ld -> %ld)\n",
1197                     existing_map->memsz, lib_php->p_filesz);
1198
1199        existing_map->fd = lib_fd;
1200        existing_map->offset = lib_php->p_offset;
1201        existing_map->memsz = lib_php->p_filesz;
1202      }
1203    }
1204
1205    lib_php++;
1206  }
1207
1208  free(phbuf);
1209  return true;
1210err:
1211  free(phbuf);
1212  return false;
1213}
1214
1215// process segments from interpreter (ld.so or ld-linux.so or ld-elf.so)
1216static bool read_interp_segments(struct ps_prochandle* ph) {
1217   ELF_EHDR interp_ehdr;
1218
1219   if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) {
1220       print_debug("interpreter is not a valid ELF file\n");
1221       return false;
1222   }
1223
1224   if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) {
1225       print_debug("can't read segments of interpreter\n");
1226       return false;
1227   }
1228
1229   return true;
1230}
1231
1232// process segments of a a.out
1233static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) {
1234   int i = 0;
1235   ELF_PHDR* phbuf = NULL;
1236   ELF_PHDR* exec_php = NULL;
1237
1238   if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL)
1239      return false;
1240
1241   for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) {
1242      switch (exec_php->p_type) {
1243
1244         // add mappings for PT_LOAD segments
1245         case PT_LOAD: {
1246            // add only non-writable segments of non-zero filesz
1247            if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) {
1248               if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err;
1249            }
1250            break;
1251         }
1252
1253         // read the interpreter and it's segments
1254         case PT_INTERP: {
1255            char interp_name[BUF_SIZE];
1256
1257            pread(ph->core->exec_fd, interp_name, MIN(exec_php->p_filesz, BUF_SIZE), exec_php->p_offset);
1258            print_debug("ELF interpreter %s\n", interp_name);
1259            // read interpreter segments as well
1260            if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) {
1261               print_debug("can't open runtime loader\n");
1262               goto err;
1263            }
1264            break;
1265         }
1266
1267         // from PT_DYNAMIC we want to read address of first link_map addr
1268         case PT_DYNAMIC: {
1269            ph->core->dynamic_addr = exec_php->p_vaddr;
1270            print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr);
1271            break;
1272         }
1273
1274      } // switch
1275      exec_php++;
1276   } // for
1277
1278   free(phbuf);
1279   return true;
1280err:
1281   free(phbuf);
1282   return false;
1283}
1284
1285#define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug,  r_map)
1286#define LD_BASE_OFFSET        offsetof(struct r_debug,  r_ldbase)
1287#define LINK_MAP_ADDR_OFFSET  offsetof(struct link_map, l_addr)
1288#define LINK_MAP_NAME_OFFSET  offsetof(struct link_map, l_name)
1289#define LINK_MAP_NEXT_OFFSET  offsetof(struct link_map, l_next)
1290
1291// read shared library info from runtime linker's data structures.
1292// This work is done by librtlb_db in Solaris
1293static bool read_shared_lib_info(struct ps_prochandle* ph) {
1294  uintptr_t addr = ph->core->dynamic_addr;
1295  uintptr_t debug_base;
1296  uintptr_t first_link_map_addr;
1297  uintptr_t ld_base_addr;
1298  uintptr_t link_map_addr;
1299  uintptr_t lib_base_diff;
1300  uintptr_t lib_base;
1301  uintptr_t lib_name_addr;
1302  char lib_name[BUF_SIZE];
1303  ELF_DYN dyn;
1304  ELF_EHDR elf_ehdr;
1305  int lib_fd;
1306
1307  // _DYNAMIC has information of the form
1308  //         [tag] [data] [tag] [data] .....
1309  // Both tag and data are pointer sized.
1310  // We look for dynamic info with DT_DEBUG. This has shared object info.
1311  // refer to struct r_debug in link.h
1312
1313  dyn.d_tag = DT_NULL;
1314  while (dyn.d_tag != DT_DEBUG) {
1315    if (ps_pread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) {
1316      print_debug("can't read debug info from _DYNAMIC\n");
1317      return false;
1318    }
1319    addr += sizeof(ELF_DYN);
1320  }
1321
1322  // we have got Dyn entry with DT_DEBUG
1323  debug_base = dyn.d_un.d_ptr;
1324  // at debug_base we have struct r_debug. This has first link map in r_map field
1325  if (ps_pread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET,
1326                 &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1327    print_debug("can't read first link map address\n");
1328    return false;
1329  }
1330
1331  // read ld_base address from struct r_debug
1332#if 0  // There is no r_ldbase member on BSD
1333  if (ps_pread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr,
1334                  sizeof(uintptr_t)) != PS_OK) {
1335    print_debug("can't read ld base address\n");
1336    return false;
1337  }
1338  ph->core->ld_base_addr = ld_base_addr;
1339#else
1340  ph->core->ld_base_addr = 0;
1341#endif
1342
1343  print_debug("interpreter base address is 0x%lx\n", ld_base_addr);
1344
1345  // now read segments from interp (i.e ld.so or ld-linux.so or ld-elf.so)
1346  if (read_interp_segments(ph) != true) {
1347    return false;
1348  }
1349
1350  // after adding interpreter (ld.so) mappings sort again
1351  if (sort_map_array(ph) != true) {
1352    return false;
1353  }
1354
1355  print_debug("first link map is at 0x%lx\n", first_link_map_addr);
1356
1357  link_map_addr = first_link_map_addr;
1358  while (link_map_addr != 0) {
1359    // read library base address of the .so. Note that even though <sys/link.h> calls
1360    // link_map->l_addr as "base address",  this is * not * really base virtual
1361    // address of the shared object. This is actually the difference b/w the virtual
1362    // address mentioned in shared object and the actual virtual base where runtime
1363    // linker loaded it. We use "base diff" in read_lib_segments call below.
1364
1365    if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET,
1366                 &lib_base_diff, sizeof(uintptr_t)) != PS_OK) {
1367      print_debug("can't read shared object base address diff\n");
1368      return false;
1369    }
1370
1371    // read address of the name
1372    if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET,
1373                  &lib_name_addr, sizeof(uintptr_t)) != PS_OK) {
1374      print_debug("can't read address of shared object name\n");
1375      return false;
1376    }
1377
1378    // read name of the shared object
1379    if (read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) {
1380      print_debug("can't read shared object name\n");
1381      return false;
1382    }
1383
1384    if (lib_name[0] != '\0') {
1385      // ignore empty lib names
1386      lib_fd = pathmap_open(lib_name);
1387
1388      if (lib_fd < 0) {
1389        print_debug("can't open shared object %s\n", lib_name);
1390        // continue with other libraries...
1391      } else {
1392        if (read_elf_header(lib_fd, &elf_ehdr)) {
1393          lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr);
1394          print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n",
1395                       lib_name, lib_base, lib_base_diff);
1396          // while adding library mappings we need to use "base difference".
1397          if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) {
1398            print_debug("can't read shared object's segments\n");
1399            close(lib_fd);
1400            return false;
1401          }
1402          add_lib_info_fd(ph, lib_name, lib_fd, lib_base);
1403          // Map info is added for the library (lib_name) so
1404          // we need to re-sort it before calling the p_pdread.
1405          if (sort_map_array(ph) != true) {
1406            return false;
1407          }
1408        } else {
1409          print_debug("can't read ELF header for shared object %s\n", lib_name);
1410          close(lib_fd);
1411          // continue with other libraries...
1412        }
1413      }
1414    }
1415
1416    // read next link_map address
1417    if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET,
1418                  &link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1419      print_debug("can't read next link in link_map\n");
1420      return false;
1421    }
1422  }
1423
1424  return true;
1425}
1426
1427// the one and only one exposed stuff from this file
1428struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
1429  ELF_EHDR core_ehdr;
1430  ELF_EHDR exec_ehdr;
1431
1432  struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
1433  if (ph == NULL) {
1434    print_debug("can't allocate ps_prochandle\n");
1435    return NULL;
1436  }
1437
1438  if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
1439    free(ph);
1440    print_debug("can't allocate ps_prochandle\n");
1441    return NULL;
1442  }
1443
1444  // initialize ph
1445  ph->ops = &core_ops;
1446  ph->core->core_fd   = -1;
1447  ph->core->exec_fd   = -1;
1448  ph->core->interp_fd = -1;
1449
1450  print_debug("exec: %s   core: %s", exec_file, core_file);
1451
1452  // open the core file
1453  if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
1454    print_debug("can't open core file\n");
1455    goto err;
1456  }
1457
1458  // read core file ELF header
1459  if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) {
1460    print_debug("core file is not a valid ELF ET_CORE file\n");
1461    goto err;
1462  }
1463
1464  if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
1465    print_debug("can't open executable file\n");
1466    goto err;
1467  }
1468
1469  if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) {
1470    print_debug("executable file is not a valid ELF ET_EXEC file\n");
1471    goto err;
1472  }
1473
1474  // process core file segments
1475  if (read_core_segments(ph, &core_ehdr) != true) {
1476    goto err;
1477  }
1478
1479  // process exec file segments
1480  if (read_exec_segments(ph, &exec_ehdr) != true) {
1481    goto err;
1482  }
1483
1484  // exec file is also treated like a shared object for symbol search
1485  if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd,
1486                      (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL) {
1487    goto err;
1488  }
1489
1490  // allocate and sort maps into map_array, we need to do this
1491  // here because read_shared_lib_info needs to read from debuggee
1492  // address space
1493  if (sort_map_array(ph) != true) {
1494    goto err;
1495  }
1496
1497  if (read_shared_lib_info(ph) != true) {
1498    goto err;
1499  }
1500
1501  // sort again because we have added more mappings from shared objects
1502  if (sort_map_array(ph) != true) {
1503    goto err;
1504  }
1505
1506  if (init_classsharing_workaround(ph) != true) {
1507    goto err;
1508  }
1509
1510  print_debug("Leave Pgrab_core\n");
1511  return ph;
1512
1513err:
1514  Prelease(ph);
1515  return NULL;
1516}
1517
1518#endif // __APPLE__
1519