1 /*
2 * Copyright (c) 2003, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include <jni.h>
26 #include <unistd.h>
27 #include <fcntl.h>
28 #include <string.h>
29 #include <stdlib.h>
30 #include <stddef.h>
31 #include "libproc_impl.h"
32
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
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
76 static 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
98 static 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
106 static 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
122 static 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
138 static 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.
153 static 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 9
217
218 // Refer to FileMapInfo::_current_version in filemap.hpp
219 #define CURRENT_ARCHIVE_VERSION 3
220
221 typedef unsigned char* address;
222 typedef uintptr_t uintx;
223 typedef intptr_t intx;
224
225
226 struct FileMapHeader {
227 int _magic; // identify file type.
228 int _crc; // header crc checksum.
229 int _version; // (from enum, above.)
230 size_t _alignment; // how shared archive should be aligned
231 int _obj_alignment; // value of ObjectAlignmentInBytes
232 address _narrow_oop_base; // compressed oop encoding base
233 int _narrow_oop_shift; // compressed oop encoding shift
234 bool _compact_strings; // value of CompactStrings
235 uintx _max_heap_size; // java max heap size during dumping
236 int _narrow_oop_mode; // compressed oop encoding mode
237 int _narrow_klass_shift; // save narrow klass base and shift
238 address _narrow_klass_base;
239 char* _misc_data_patching_start;
240 char* _read_only_tables_start;
241 address _cds_i2i_entry_code_buffers;
242 size_t _cds_i2i_entry_code_buffers_size;
243 size_t _core_spaces_size; // number of bytes allocated by the core spaces
244 // (mc, md, ro, rw and od).
245
246
247 struct space_info {
248 int _crc; // crc checksum of the current space
249 size_t _file_offset; // sizeof(this) rounded to vm page size
250 union {
251 char* _base; // copy-on-write base address
252 intx _offset; // offset from the compressed oop encoding base, only used
253 // by archive heap space
254 } _addr;
255 size_t _used; // for setting space top on read
256 // 4991491 NOTICE These are C++ bool's in filemap.hpp and must match up with
257 // the C type matching the C++ bool type on any given platform.
258 // We assume the corresponding C type is char but licensees
259 // may need to adjust the type of these fields.
260 char _read_only; // read only space?
261 char _allow_exec; // executable code in space?
262 } _space[NUM_SHARED_MAPS];
263
264 // Ignore the rest of the FileMapHeader. We don't need those fields here.
265 };
266
267 static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) {
268 jboolean i;
269 if (ps_pread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) {
270 *pvalue = i;
271 return true;
272 } else {
273 return false;
274 }
275 }
276
277 static bool read_pointer(struct ps_prochandle* ph, uintptr_t addr, uintptr_t* pvalue) {
278 uintptr_t uip;
279 if (ps_pread(ph, (psaddr_t) addr, (char *)&uip, sizeof(uip)) == PS_OK) {
280 *pvalue = uip;
281 return true;
282 } else {
283 return false;
284 }
285 }
286
287 // used to read strings from debuggee
288 static bool read_string(struct ps_prochandle* ph, uintptr_t addr, char* buf, size_t size) {
289 size_t i = 0;
290 char c = ' ';
291
292 while (c != '\0') {
293 if (ps_pread(ph, (psaddr_t) addr, &c, sizeof(char)) != PS_OK) {
294 return false;
295 }
296 if (i < size - 1) {
297 buf[i] = c;
298 } else {
299 // smaller buffer
300 return false;
301 }
302 i++; addr++;
303 }
304 buf[i] = '\0';
305 return true;
306 }
307
308 // mangled name of Arguments::SharedArchivePath
309 #define SHARED_ARCHIVE_PATH_SYM "__ZN9Arguments17SharedArchivePathE"
310
311 #ifdef __APPLE__
312 #define USE_SHARED_SPACES_SYM "_UseSharedSpaces"
313 #define LIBJVM_NAME "/libjvm.dylib"
314 #else
315 #define USE_SHARED_SPACES_SYM "UseSharedSpaces"
316 #define LIBJVM_NAME "/libjvm.so"
317 #endif // __APPLE_
318
319 static bool init_classsharing_workaround(struct ps_prochandle* ph) {
320 int m;
321 size_t n;
322 lib_info* lib = ph->libs;
323 while (lib != NULL) {
324 // we are iterating over shared objects from the core dump. look for
325 // libjvm.so.
326 const char *jvm_name = 0;
327 if ((jvm_name = strstr(lib->name, LIBJVM_NAME)) != 0) {
328 char classes_jsa[PATH_MAX];
329 struct FileMapHeader header;
330 int fd = -1;
331 uintptr_t base = 0, useSharedSpacesAddr = 0;
332 uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0;
333 jboolean useSharedSpaces = 0;
334
335 memset(classes_jsa, 0, sizeof(classes_jsa));
336 jvm_name = lib->name;
337 useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM);
338 if (useSharedSpacesAddr == 0) {
339 print_debug("can't lookup 'UseSharedSpaces' flag\n");
340 return false;
341 }
342
343 // Hotspot vm types are not exported to build this library. So
344 // using equivalent type jboolean to read the value of
345 // UseSharedSpaces which is same as hotspot type "bool".
346 if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) {
347 print_debug("can't read the value of 'UseSharedSpaces' flag\n");
348 return false;
349 }
350
351 if ((int)useSharedSpaces == 0) {
352 print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n");
353 return true;
354 }
355
356 sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM);
357 if (sharedArchivePathAddrAddr == 0) {
358 print_debug("can't lookup shared archive path symbol\n");
359 return false;
360 }
361
362 if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) {
363 print_debug("can't read shared archive path pointer\n");
364 return false;
365 }
366
367 if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) {
368 print_debug("can't read shared archive path value\n");
369 return false;
370 }
371
372 print_debug("looking for %s\n", classes_jsa);
373 // open the class sharing archive file
374 fd = pathmap_open(classes_jsa);
375 if (fd < 0) {
376 print_debug("can't open %s!\n", classes_jsa);
377 ph->core->classes_jsa_fd = -1;
378 return false;
379 } else {
380 print_debug("opened %s\n", classes_jsa);
381 }
382
383 // read FileMapHeader from the file
384 memset(&header, 0, sizeof(struct FileMapHeader));
385 if ((n = read(fd, &header, sizeof(struct FileMapHeader)))
386 != sizeof(struct FileMapHeader)) {
387 print_debug("can't read shared archive file map header from %s\n", classes_jsa);
388 close(fd);
389 return false;
390 }
391
392 // check file magic
393 if (header._magic != 0xf00baba2) {
394 print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n",
395 classes_jsa, header._magic);
396 close(fd);
397 return false;
398 }
399
400 // check version
401 if (header._version != CURRENT_ARCHIVE_VERSION) {
402 print_debug("%s has wrong shared archive file version %d, expecting %d\n",
403 classes_jsa, header._version, CURRENT_ARCHIVE_VERSION);
404 close(fd);
405 return false;
406 }
407
408 ph->core->classes_jsa_fd = fd;
409 // add read-only maps from classes.jsa to the list of maps
410 for (m = 0; m < NUM_SHARED_MAPS; m++) {
411 if (header._space[m]._read_only) {
412 base = (uintptr_t) header._space[m]._addr._base;
413 // no need to worry about the fractional pages at-the-end.
414 // possible fractional pages are handled by core_read_data.
415 add_class_share_map_info(ph, (off_t) header._space[m]._file_offset,
416 base, (size_t) header._space[m]._used);
417 print_debug("added a share archive map at 0x%lx\n", base);
418 }
419 }
420 return true;
421 }
422 lib = lib->next;
423 }
424 return true;
425 }
426
427 //---------------------------------------------------------------------------
428 // functions to handle map_info
429
430 // Order mappings based on virtual address. We use this function as the
431 // callback for sorting the array of map_info pointers.
432 static int core_cmp_mapping(const void *lhsp, const void *rhsp)
433 {
434 const map_info *lhs = *((const map_info **)lhsp);
435 const map_info *rhs = *((const map_info **)rhsp);
436
437 if (lhs->vaddr == rhs->vaddr) {
438 return (0);
439 }
440
441 return (lhs->vaddr < rhs->vaddr ? -1 : 1);
442 }
443
444 // we sort map_info by starting virtual address so that we can do
445 // binary search to read from an address.
446 static bool sort_map_array(struct ps_prochandle* ph) {
447 size_t num_maps = ph->core->num_maps;
448 map_info* map = ph->core->maps;
449 int i = 0;
450
451 // allocate map_array
452 map_info** array;
453 if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) {
454 print_debug("can't allocate memory for map array\n");
455 return false;
456 }
457
458 // add maps to array
459 while (map) {
460 array[i] = map;
461 i++;
462 map = map->next;
463 }
464
465 // sort is called twice. If this is second time, clear map array
466 if (ph->core->map_array) {
467 free(ph->core->map_array);
468 }
469 ph->core->map_array = array;
470 // sort the map_info array by base virtual address.
471 qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*),
472 core_cmp_mapping);
473
474 // print map
475 if (is_debug()) {
476 int j = 0;
477 print_debug("---- sorted virtual address map ----\n");
478 for (j = 0; j < ph->core->num_maps; j++) {
479 print_debug("base = 0x%lx\tsize = %d\n", ph->core->map_array[j]->vaddr,
480 ph->core->map_array[j]->memsz);
481 }
482 }
483
484 return true;
485 }
486
487 #ifndef MIN
488 #define MIN(x, y) (((x) < (y))? (x): (y))
489 #endif
490
491 static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) {
492 ssize_t resid = size;
493 int page_size=sysconf(_SC_PAGE_SIZE);
494 while (resid != 0) {
495 map_info *mp = core_lookup(ph, addr);
496 uintptr_t mapoff;
497 ssize_t len, rem;
498 off_t off;
499 int fd;
500
501 if (mp == NULL) {
502 break; /* No mapping for this address */
503 }
504
505 fd = mp->fd;
506 mapoff = addr - mp->vaddr;
507 len = MIN(resid, mp->memsz - mapoff);
508 off = mp->offset + mapoff;
509
510 if ((len = pread(fd, buf, len, off)) <= 0) {
511 break;
512 }
513
514 resid -= len;
515 addr += len;
516 buf = (char *)buf + len;
517
518 // mappings always start at page boundary. But, may end in fractional
519 // page. fill zeros for possible fractional page at the end of a mapping.
520 rem = mp->memsz % page_size;
521 if (rem > 0) {
522 rem = page_size - rem;
523 len = MIN(resid, rem);
524 resid -= len;
525 addr += len;
526 // we are not assuming 'buf' to be zero initialized.
527 memset(buf, 0, len);
528 buf += len;
529 }
530 }
531
532 if (resid) {
533 print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n",
534 size, addr, resid);
535 return false;
536 } else {
537 return true;
538 }
539 }
540
541 // null implementation for write
542 static bool core_write_data(struct ps_prochandle* ph,
543 uintptr_t addr, const char *buf , size_t size) {
544 return false;
545 }
546
547 static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id,
548 struct reg* regs) {
549 // for core we have cached the lwp regs after segment parsed
550 sa_thread_info* thr = ph->threads;
551 while (thr) {
552 if (thr->lwp_id == lwp_id) {
553 memcpy(regs, &thr->regs, sizeof(struct reg));
554 return true;
555 }
556 thr = thr->next;
557 }
558 return false;
559 }
560
561 static bool core_get_lwp_info(struct ps_prochandle *ph, lwpid_t id, void *info) {
562 print_debug("core_get_lwp_info not implemented\n");
563 return false;
564 }
565
566 static ps_prochandle_ops core_ops = {
567 .release= core_release,
568 .p_pread= core_read_data,
569 .p_pwrite= core_write_data,
570 .get_lwp_regs= core_get_lwp_regs,
571 .get_lwp_info= core_get_lwp_info
572 };
573
574 // from this point, mainly two blocks divided by def __APPLE__
575 // one for Macosx, the other for regular Bsd
576
577 #ifdef __APPLE__
578
579 void print_thread(sa_thread_info *threadinfo) {
580 print_debug("thread added: %d\n", threadinfo->lwp_id);
581 print_debug("registers:\n");
582 print_debug(" r_r15: 0x%" PRIx64 "\n", threadinfo->regs.r_r15);
583 print_debug(" r_r14: 0x%" PRIx64 "\n", threadinfo->regs.r_r14);
584 print_debug(" r_r13: 0x%" PRIx64 "\n", threadinfo->regs.r_r13);
585 print_debug(" r_r12: 0x%" PRIx64 "\n", threadinfo->regs.r_r12);
586 print_debug(" r_r11: 0x%" PRIx64 "\n", threadinfo->regs.r_r11);
587 print_debug(" r_r10: 0x%" PRIx64 "\n", threadinfo->regs.r_r10);
588 print_debug(" r_r9: 0x%" PRIx64 "\n", threadinfo->regs.r_r9);
589 print_debug(" r_r8: 0x%" PRIx64 "\n", threadinfo->regs.r_r8);
590 print_debug(" r_rdi: 0x%" PRIx64 "\n", threadinfo->regs.r_rdi);
591 print_debug(" r_rsi: 0x%" PRIx64 "\n", threadinfo->regs.r_rsi);
592 print_debug(" r_rbp: 0x%" PRIx64 "\n", threadinfo->regs.r_rbp);
593 print_debug(" r_rbx: 0x%" PRIx64 "\n", threadinfo->regs.r_rbx);
594 print_debug(" r_rdx: 0x%" PRIx64 "\n", threadinfo->regs.r_rdx);
595 print_debug(" r_rcx: 0x%" PRIx64 "\n", threadinfo->regs.r_rcx);
596 print_debug(" r_rax: 0x%" PRIx64 "\n", threadinfo->regs.r_rax);
597 print_debug(" r_fs: 0x%" PRIx32 "\n", threadinfo->regs.r_fs);
598 print_debug(" r_gs: 0x%" PRIx32 "\n", threadinfo->regs.r_gs);
599 print_debug(" r_rip 0x%" PRIx64 "\n", threadinfo->regs.r_rip);
600 print_debug(" r_cs: 0x%" PRIx64 "\n", threadinfo->regs.r_cs);
601 print_debug(" r_rsp: 0x%" PRIx64 "\n", threadinfo->regs.r_rsp);
602 print_debug(" r_rflags: 0x%" PRIx64 "\n", threadinfo->regs.r_rflags);
603 }
604
605 // read all segments64 commands from core file
606 // read all thread commands from core file
607 static bool read_core_segments(struct ps_prochandle* ph) {
608 int i = 0;
609 int num_threads = 0;
610 int fd = ph->core->core_fd;
611 off_t offset = 0;
612 mach_header_64 fhead;
613 load_command lcmd;
614 segment_command_64 segcmd;
615 // thread_command thrcmd;
616
617 lseek(fd, offset, SEEK_SET);
618 if(read(fd, (void *)&fhead, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
619 goto err;
620 }
621 print_debug("total commands: %d\n", fhead.ncmds);
622 offset += sizeof(mach_header_64);
623 for (i = 0; i < fhead.ncmds; i++) {
624 lseek(fd, offset, SEEK_SET);
625 if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
626 goto err;
627 }
628 offset += lcmd.cmdsize; // next command position
629 if (lcmd.cmd == LC_SEGMENT_64) {
630 lseek(fd, -sizeof(load_command), SEEK_CUR);
631 if (read(fd, (void *)&segcmd, sizeof(segment_command_64)) != sizeof(segment_command_64)) {
632 print_debug("failed to read LC_SEGMENT_64 i = %d!\n", i);
633 goto err;
634 }
635 if (add_map_info(ph, fd, segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize) == NULL) {
636 print_debug("Failed to add map_info at i = %d\n", i);
637 goto err;
638 }
639 print_debug("segment added: %" PRIu64 " 0x%" PRIx64 " %d\n",
640 segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize);
641 } else if (lcmd.cmd == LC_THREAD || lcmd.cmd == LC_UNIXTHREAD) {
642 typedef struct thread_fc {
643 uint32_t flavor;
644 uint32_t count;
645 } thread_fc;
646 thread_fc fc;
647 uint32_t size = sizeof(load_command);
648 while (size < lcmd.cmdsize) {
649 if (read(fd, (void *)&fc, sizeof(thread_fc)) != sizeof(thread_fc)) {
650 printf("Reading flavor, count failed.\n");
651 goto err;
652 }
653 size += sizeof(thread_fc);
654 if (fc.flavor == x86_THREAD_STATE) {
655 x86_thread_state_t thrstate;
656 if (read(fd, (void *)&thrstate, sizeof(x86_thread_state_t)) != sizeof(x86_thread_state_t)) {
657 printf("Reading flavor, count failed.\n");
658 goto err;
659 }
660 size += sizeof(x86_thread_state_t);
661 // create thread info list, update lwp_id later
662 sa_thread_info* newthr = add_thread_info(ph, (pthread_t) -1, (lwpid_t) num_threads++);
663 if (newthr == NULL) {
664 printf("create thread_info failed\n");
665 goto err;
666 }
667
668 // note __DARWIN_UNIX03 depengs on other definitions
669 #if __DARWIN_UNIX03
670 #define get_register_v(regst, regname) \
671 regst.uts.ts64.__##regname
672 #else
673 #define get_register_v(regst, regname) \
674 regst.uts.ts64.##regname
675 #endif // __DARWIN_UNIX03
676 newthr->regs.r_rax = get_register_v(thrstate, rax);
677 newthr->regs.r_rbx = get_register_v(thrstate, rbx);
678 newthr->regs.r_rcx = get_register_v(thrstate, rcx);
679 newthr->regs.r_rdx = get_register_v(thrstate, rdx);
680 newthr->regs.r_rdi = get_register_v(thrstate, rdi);
681 newthr->regs.r_rsi = get_register_v(thrstate, rsi);
682 newthr->regs.r_rbp = get_register_v(thrstate, rbp);
683 newthr->regs.r_rsp = get_register_v(thrstate, rsp);
684 newthr->regs.r_r8 = get_register_v(thrstate, r8);
685 newthr->regs.r_r9 = get_register_v(thrstate, r9);
686 newthr->regs.r_r10 = get_register_v(thrstate, r10);
687 newthr->regs.r_r11 = get_register_v(thrstate, r11);
688 newthr->regs.r_r12 = get_register_v(thrstate, r12);
689 newthr->regs.r_r13 = get_register_v(thrstate, r13);
690 newthr->regs.r_r14 = get_register_v(thrstate, r14);
691 newthr->regs.r_r15 = get_register_v(thrstate, r15);
692 newthr->regs.r_rip = get_register_v(thrstate, rip);
693 newthr->regs.r_rflags = get_register_v(thrstate, rflags);
694 newthr->regs.r_cs = get_register_v(thrstate, cs);
695 newthr->regs.r_fs = get_register_v(thrstate, fs);
696 newthr->regs.r_gs = get_register_v(thrstate, gs);
697 print_thread(newthr);
698 } else if (fc.flavor == x86_FLOAT_STATE) {
699 x86_float_state_t flstate;
700 if (read(fd, (void *)&flstate, sizeof(x86_float_state_t)) != sizeof(x86_float_state_t)) {
701 print_debug("Reading flavor, count failed.\n");
702 goto err;
703 }
704 size += sizeof(x86_float_state_t);
705 } else if (fc.flavor == x86_EXCEPTION_STATE) {
706 x86_exception_state_t excpstate;
707 if (read(fd, (void *)&excpstate, sizeof(x86_exception_state_t)) != sizeof(x86_exception_state_t)) {
708 printf("Reading flavor, count failed.\n");
709 goto err;
710 }
711 size += sizeof(x86_exception_state_t);
712 }
713 }
714 }
715 }
716 return true;
717 err:
718 return false;
719 }
720
721 /**local function **/
722 bool exists(const char *fname) {
723 return access(fname, F_OK) == 0;
724 }
725
726 // we check: 1. lib
727 // 2. lib/server
728 // 3. jre/lib
729 // 4. jre/lib/server
730 // from: 1. exe path
731 // 2. JAVA_HOME
732 // 3. DYLD_LIBRARY_PATH
733 static bool get_real_path(struct ps_prochandle* ph, char *rpath) {
734 /** check if they exist in JAVA ***/
735 char* execname = ph->core->exec_path;
736 char filepath[4096];
737 char* filename = strrchr(rpath, '/'); // like /libjvm.dylib
738 if (filename == NULL) {
739 return false;
740 }
741
742 char* posbin = strstr(execname, "/bin/java");
743 if (posbin != NULL) {
744 memcpy(filepath, execname, posbin - execname); // not include trailing '/'
745 filepath[posbin - execname] = '\0';
746 } else {
747 char* java_home = getenv("JAVA_HOME");
748 if (java_home != NULL) {
749 strcpy(filepath, java_home);
750 } else {
751 char* dyldpath = getenv("DYLD_LIBRARY_PATH");
752 char* dypath = strtok(dyldpath, ":");
753 while (dypath != NULL) {
754 strcpy(filepath, dypath);
755 strcat(filepath, filename);
756 if (exists(filepath)) {
757 strcpy(rpath, filepath);
758 return true;
759 }
760 dypath = strtok(dyldpath, ":");
761 }
762 // not found
763 return false;
764 }
765 }
766 // for exec and java_home, jdkpath now is filepath
767 size_t filepath_base_size = strlen(filepath);
768
769 // first try /lib/ and /lib/server
770 strcat(filepath, "/lib");
771 strcat(filepath, filename);
772 if (exists(filepath)) {
773 strcpy(rpath, filepath);
774 return true;
775 }
776 char* pos = strstr(filepath, filename); // like /libjvm.dylib
777 *pos = '\0';
778 strcat(filepath, "/server");
779 strcat(filepath, filename);
780 if (exists(filepath)) {
781 strcpy(rpath, filepath);
782 return true;
783 }
784
785 // then try /jre/lib/ and /jre/lib/server
786 filepath[filepath_base_size] = '\0';
787 strcat(filepath, "/jre/lib");
788 strcat(filepath, filename);
789 if (exists(filepath)) {
790 strcpy(rpath, filepath);
791 return true;
792 }
793 pos = strstr(filepath, filename);
794 *pos = '\0';
795 strcat(filepath, "/server");
796 strcat(filepath, filename);
797 if (exists(filepath)) {
798 strcpy(rpath, filepath);
799 return true;
800 }
801
802 return false;
803 }
804
805 static bool read_shared_lib_info(struct ps_prochandle* ph) {
806 static int pagesize = 0;
807 int fd = ph->core->core_fd;
808 int i = 0, j;
809 uint32_t v;
810 mach_header_64 header; // used to check if a file header in segment
811 load_command lcmd;
812 dylib_command dylibcmd;
813
814 char name[BUF_SIZE]; // use to store name
815
816 if (pagesize == 0) {
817 pagesize = getpagesize();
818 print_debug("page size is %d\n", pagesize);
819 }
820 for (j = 0; j < ph->core->num_maps; j++) {
821 map_info *iter = ph->core->map_array[j]; // head
822 off_t fpos = iter->offset;
823 if (iter->fd != fd) {
824 // only search core file!
825 continue;
826 }
827 print_debug("map_info %d: vmaddr = 0x%016" PRIx64 " fileoff = %" PRIu64 " vmsize = %" PRIu64 "\n",
828 j, iter->vaddr, iter->offset, iter->memsz);
829 lseek(fd, fpos, SEEK_SET);
830 // we assume .dylib loaded at segment address --- which is true for JVM libraries
831 // multiple files may be loaded in one segment.
832 // if first word is not a magic word, means this segment does not contain lib file.
833 if (read(fd, (void *)&v, sizeof(uint32_t)) == sizeof(uint32_t)) {
834 if (v != MH_MAGIC_64) {
835 continue;
836 }
837 } else {
838 // may be encountered last map, which is not readable
839 continue;
840 }
841 while (ltell(fd) - iter->offset < iter->memsz) {
842 lseek(fd, fpos, SEEK_SET);
843 if (read(fd, (void *)&v, sizeof(uint32_t)) != sizeof(uint32_t)) {
844 break;
845 }
846 if (v != MH_MAGIC_64) {
847 fpos = (ltell(fd) + pagesize -1)/pagesize * pagesize;
848 continue;
849 }
850 lseek(fd, -sizeof(uint32_t), SEEK_CUR);
851 // this is the file begining to core file.
852 if (read(fd, (void *)&header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
853 goto err;
854 }
855 fpos = ltell(fd);
856
857 // found a mach-o file in this segment
858 for (i = 0; i < header.ncmds; i++) {
859 // read commands in this "file"
860 // LC_ID_DYLIB is the file itself for a .dylib
861 lseek(fd, fpos, SEEK_SET);
862 if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
863 return false; // error
864 }
865 fpos += lcmd.cmdsize; // next command position
866 // make sure still within seg size.
867 if (fpos - lcmd.cmdsize - iter->offset > iter->memsz) {
868 print_debug("Warning: out of segement limit: %ld \n", fpos - lcmd.cmdsize - iter->offset);
869 break; // no need to iterate all commands
870 }
871 if (lcmd.cmd == LC_ID_DYLIB) {
872 lseek(fd, -sizeof(load_command), SEEK_CUR);
873 if (read(fd, (void *)&dylibcmd, sizeof(dylib_command)) != sizeof(dylib_command)) {
874 return false;
875 }
876 /**** name stored at dylib_command.dylib.name.offset, is a C string */
877 lseek(fd, dylibcmd.dylib.name.offset - sizeof(dylib_command), SEEK_CUR);
878 int j = 0;
879 while (j < BUF_SIZE) {
880 read(fd, (void *)(name + j), sizeof(char));
881 if (name[j] == '\0') break;
882 j++;
883 }
884 print_debug("%s\n", name);
885 // changed name from @rpath/xxxx.dylib to real path
886 if (strrchr(name, '@')) {
887 get_real_path(ph, name);
888 print_debug("get_real_path returned: %s\n", name);
889 }
890 add_lib_info(ph, name, iter->vaddr);
891 break;
892 }
893 }
894 // done with the file, advanced to next page to search more files
895 fpos = (ltell(fd) + pagesize - 1) / pagesize * pagesize;
896 }
897 }
898 return true;
899 err:
900 return false;
901 }
902
903 bool read_macho64_header(int fd, mach_header_64* core_header) {
904 bool is_macho = false;
905 if (fd < 0) return false;
906 off_t pos = ltell(fd);
907 lseek(fd, 0, SEEK_SET);
908 if (read(fd, (void *)core_header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
909 is_macho = false;
910 } else {
911 is_macho = (core_header->magic == MH_MAGIC_64 || core_header->magic == MH_CIGAM_64);
912 }
913 lseek(fd, pos, SEEK_SET);
914 return is_macho;
915 }
916
917 // the one and only one exposed stuff from this file
918 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
919 mach_header_64 core_header;
920 mach_header_64 exec_header;
921
922 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
923 if (ph == NULL) {
924 print_debug("cant allocate ps_prochandle\n");
925 return NULL;
926 }
927
928 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
929 free(ph);
930 print_debug("can't allocate ps_prochandle\n");
931 return NULL;
932 }
933
934 // initialize ph
935 ph->ops = &core_ops;
936 ph->core->core_fd = -1;
937 ph->core->exec_fd = -1;
938 ph->core->interp_fd = -1;
939
940 print_debug("exec: %s core: %s", exec_file, core_file);
941
942 strncpy(ph->core->exec_path, exec_file, sizeof(ph->core->exec_path));
943
944 // open the core file
945 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
946 print_error("can't open core file\n");
947 goto err;
948 }
949
950 // read core file header
951 if (read_macho64_header(ph->core->core_fd, &core_header) != true || core_header.filetype != MH_CORE) {
952 print_debug("core file is not a valid Mach-O file\n");
953 goto err;
954 }
955
956 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
957 print_error("can't open executable file\n");
958 goto err;
959 }
960
961 if (read_macho64_header(ph->core->exec_fd, &exec_header) != true ||
962 exec_header.filetype != MH_EXECUTE) {
963 print_error("executable file is not a valid Mach-O file\n");
964 goto err;
965 }
966
967 // process core file segments
968 if (read_core_segments(ph) != true) {
969 print_error("failed to read core segments\n");
970 goto err;
971 }
972
973 // allocate and sort maps into map_array, we need to do this
974 // here because read_shared_lib_info needs to read from debuggee
975 // address space
976 if (sort_map_array(ph) != true) {
977 print_error("failed to sort segment map array\n");
978 goto err;
979 }
980
981 if (read_shared_lib_info(ph) != true) {
982 print_error("failed to read libraries\n");
983 goto err;
984 }
985
986 // sort again because we have added more mappings from shared objects
987 if (sort_map_array(ph) != true) {
988 print_error("failed to sort segment map array\n");
989 goto err;
990 }
991
992 if (init_classsharing_workaround(ph) != true) {
993 print_error("failed to workaround classshareing\n");
994 goto err;
995 }
996
997 print_debug("Leave Pgrab_core\n");
998 return ph;
999
1000 err:
1001 Prelease(ph);
1002 return NULL;
1003 }
1004
1005 #else // __APPLE__ (none macosx)
1006
1007 // read regs and create thread from core file
1008 static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) {
1009 // we have to read prstatus_t from buf
1010 // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t");
1011 prstatus_t* prstat = (prstatus_t*) buf;
1012 sa_thread_info* newthr;
1013 print_debug("got integer regset for lwp %d\n", prstat->pr_pid);
1014 // we set pthread_t to -1 for core dump
1015 if((newthr = add_thread_info(ph, (pthread_t) -1, prstat->pr_pid)) == NULL)
1016 return false;
1017
1018 // copy regs
1019 memcpy(&newthr->regs, &prstat->pr_reg, sizeof(struct reg));
1020
1021 if (is_debug()) {
1022 print_debug("integer regset\n");
1023 #ifdef i386
1024 // print the regset
1025 print_debug("\teax = 0x%x\n", newthr->regs.r_eax);
1026 print_debug("\tebx = 0x%x\n", newthr->regs.r_ebx);
1027 print_debug("\tecx = 0x%x\n", newthr->regs.r_ecx);
1028 print_debug("\tedx = 0x%x\n", newthr->regs.r_edx);
1029 print_debug("\tesp = 0x%x\n", newthr->regs.r_esp);
1030 print_debug("\tebp = 0x%x\n", newthr->regs.r_ebp);
1031 print_debug("\tesi = 0x%x\n", newthr->regs.r_esi);
1032 print_debug("\tedi = 0x%x\n", newthr->regs.r_edi);
1033 print_debug("\teip = 0x%x\n", newthr->regs.r_eip);
1034 #endif
1035
1036 #if defined(amd64) || defined(x86_64)
1037 // print the regset
1038 print_debug("\tr15 = 0x%lx\n", newthr->regs.r_r15);
1039 print_debug("\tr14 = 0x%lx\n", newthr->regs.r_r14);
1040 print_debug("\tr13 = 0x%lx\n", newthr->regs.r_r13);
1041 print_debug("\tr12 = 0x%lx\n", newthr->regs.r_r12);
1042 print_debug("\trbp = 0x%lx\n", newthr->regs.r_rbp);
1043 print_debug("\trbx = 0x%lx\n", newthr->regs.r_rbx);
1044 print_debug("\tr11 = 0x%lx\n", newthr->regs.r_r11);
1045 print_debug("\tr10 = 0x%lx\n", newthr->regs.r_r10);
1046 print_debug("\tr9 = 0x%lx\n", newthr->regs.r_r9);
1047 print_debug("\tr8 = 0x%lx\n", newthr->regs.r_r8);
1048 print_debug("\trax = 0x%lx\n", newthr->regs.r_rax);
1049 print_debug("\trcx = 0x%lx\n", newthr->regs.r_rcx);
1050 print_debug("\trdx = 0x%lx\n", newthr->regs.r_rdx);
1051 print_debug("\trsi = 0x%lx\n", newthr->regs.r_rsi);
1052 print_debug("\trdi = 0x%lx\n", newthr->regs.r_rdi);
1053 //print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax);
1054 print_debug("\trip = 0x%lx\n", newthr->regs.r_rip);
1055 print_debug("\tcs = 0x%lx\n", newthr->regs.r_cs);
1056 //print_debug("\teflags = 0x%lx\n", newthr->regs.eflags);
1057 print_debug("\trsp = 0x%lx\n", newthr->regs.r_rsp);
1058 print_debug("\tss = 0x%lx\n", newthr->regs.r_ss);
1059 //print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base);
1060 //print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base);
1061 //print_debug("\tds = 0x%lx\n", newthr->regs.ds);
1062 //print_debug("\tes = 0x%lx\n", newthr->regs.es);
1063 //print_debug("\tfs = 0x%lx\n", newthr->regs.fs);
1064 //print_debug("\tgs = 0x%lx\n", newthr->regs.gs);
1065 #endif
1066 }
1067
1068 return true;
1069 }
1070
1071 #define ROUNDUP(x, y) ((((x)+((y)-1))/(y))*(y))
1072
1073 // read NT_PRSTATUS entries from core NOTE segment
1074 static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) {
1075 char* buf = NULL;
1076 char* p = NULL;
1077 size_t size = note_phdr->p_filesz;
1078
1079 // we are interested in just prstatus entries. we will ignore the rest.
1080 // Advance the seek pointer to the start of the PT_NOTE data
1081 if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) {
1082 print_debug("failed to lseek to PT_NOTE data\n");
1083 return false;
1084 }
1085
1086 // Now process the PT_NOTE structures. Each one is preceded by
1087 // an Elf{32/64}_Nhdr structure describing its type and size.
1088 if ( (buf = (char*) malloc(size)) == NULL) {
1089 print_debug("can't allocate memory for reading core notes\n");
1090 goto err;
1091 }
1092
1093 // read notes into buffer
1094 if (read(ph->core->core_fd, buf, size) != size) {
1095 print_debug("failed to read notes, core file must have been truncated\n");
1096 goto err;
1097 }
1098
1099 p = buf;
1100 while (p < buf + size) {
1101 ELF_NHDR* notep = (ELF_NHDR*) p;
1102 char* descdata = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4);
1103 print_debug("Note header with n_type = %d and n_descsz = %u\n",
1104 notep->n_type, notep->n_descsz);
1105
1106 if (notep->n_type == NT_PRSTATUS) {
1107 if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) {
1108 return false;
1109 }
1110 }
1111 p = descdata + ROUNDUP(notep->n_descsz, 4);
1112 }
1113
1114 free(buf);
1115 return true;
1116
1117 err:
1118 if (buf) free(buf);
1119 return false;
1120 }
1121
1122 // read all segments from core file
1123 static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) {
1124 int i = 0;
1125 ELF_PHDR* phbuf = NULL;
1126 ELF_PHDR* core_php = NULL;
1127
1128 if ((phbuf = read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL)
1129 return false;
1130
1131 /*
1132 * Now iterate through the program headers in the core file.
1133 * We're interested in two types of Phdrs: PT_NOTE (which
1134 * contains a set of saved /proc structures), and PT_LOAD (which
1135 * represents a memory mapping from the process's address space).
1136 *
1137 * Difference b/w Solaris PT_NOTE and Linux/BSD PT_NOTE:
1138 *
1139 * In Solaris there are two PT_NOTE segments the first PT_NOTE (if present)
1140 * contains /proc structs in the pre-2.6 unstructured /proc format. the last
1141 * PT_NOTE has data in new /proc format.
1142 *
1143 * In Solaris, there is only one pstatus (process status). pstatus contains
1144 * integer register set among other stuff. For each LWP, we have one lwpstatus
1145 * entry that has integer regset for that LWP.
1146 *
1147 * Linux threads are actually 'clone'd processes. To support core analysis
1148 * of "multithreaded" process, Linux creates more than one pstatus (called
1149 * "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one
1150 * "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular
1151 * function "elf_core_dump".
1152 */
1153
1154 for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) {
1155 switch (core_php->p_type) {
1156 case PT_NOTE:
1157 if (core_handle_note(ph, core_php) != true) {
1158 goto err;
1159 }
1160 break;
1161
1162 case PT_LOAD: {
1163 if (core_php->p_filesz != 0) {
1164 if (add_map_info(ph, ph->core->core_fd, core_php->p_offset,
1165 core_php->p_vaddr, core_php->p_filesz) == NULL) goto err;
1166 }
1167 break;
1168 }
1169 }
1170
1171 core_php++;
1172 }
1173
1174 free(phbuf);
1175 return true;
1176 err:
1177 free(phbuf);
1178 return false;
1179 }
1180
1181 // read segments of a shared object
1182 static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) {
1183 int i = 0;
1184 ELF_PHDR* phbuf;
1185 ELF_PHDR* lib_php = NULL;
1186
1187 int page_size=sysconf(_SC_PAGE_SIZE);
1188
1189 if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL) {
1190 return false;
1191 }
1192
1193 // we want to process only PT_LOAD segments that are not writable.
1194 // i.e., text segments. The read/write/exec (data) segments would
1195 // have been already added from core file segments.
1196 for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) {
1197 if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) {
1198
1199 uintptr_t target_vaddr = lib_php->p_vaddr + lib_base;
1200 map_info *existing_map = core_lookup(ph, target_vaddr);
1201
1202 if (existing_map == NULL){
1203 if (add_map_info(ph, lib_fd, lib_php->p_offset,
1204 target_vaddr, lib_php->p_filesz) == NULL) {
1205 goto err;
1206 }
1207 } else {
1208 if ((existing_map->memsz != page_size) &&
1209 (existing_map->fd != lib_fd) &&
1210 (existing_map->memsz != lib_php->p_filesz)){
1211
1212 print_debug("address conflict @ 0x%lx (size = %ld, flags = %d\n)",
1213 target_vaddr, lib_php->p_filesz, lib_php->p_flags);
1214 goto err;
1215 }
1216
1217 /* replace PT_LOAD segment with library segment */
1218 print_debug("overwrote with new address mapping (memsz %ld -> %ld)\n",
1219 existing_map->memsz, lib_php->p_filesz);
1220
1221 existing_map->fd = lib_fd;
1222 existing_map->offset = lib_php->p_offset;
1223 existing_map->memsz = lib_php->p_filesz;
1224 }
1225 }
1226
1227 lib_php++;
1228 }
1229
1230 free(phbuf);
1231 return true;
1232 err:
1233 free(phbuf);
1234 return false;
1235 }
1236
1237 // process segments from interpreter (ld.so or ld-linux.so or ld-elf.so)
1238 static bool read_interp_segments(struct ps_prochandle* ph) {
1239 ELF_EHDR interp_ehdr;
1240
1241 if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) {
1242 print_debug("interpreter is not a valid ELF file\n");
1243 return false;
1244 }
1245
1246 if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) {
1247 print_debug("can't read segments of interpreter\n");
1248 return false;
1249 }
1250
1251 return true;
1252 }
1253
1254 // process segments of a a.out
1255 static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) {
1256 int i = 0;
1257 ELF_PHDR* phbuf = NULL;
1258 ELF_PHDR* exec_php = NULL;
1259
1260 if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL)
1261 return false;
1262
1263 for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) {
1264 switch (exec_php->p_type) {
1265
1266 // add mappings for PT_LOAD segments
1267 case PT_LOAD: {
1268 // add only non-writable segments of non-zero filesz
1269 if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) {
1270 if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err;
1271 }
1272 break;
1273 }
1274
1275 // read the interpreter and it's segments
1276 case PT_INTERP: {
1277 char interp_name[BUF_SIZE];
1278
1279 pread(ph->core->exec_fd, interp_name, MIN(exec_php->p_filesz, BUF_SIZE), exec_php->p_offset);
1280 print_debug("ELF interpreter %s\n", interp_name);
1281 // read interpreter segments as well
1282 if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) {
1283 print_debug("can't open runtime loader\n");
1284 goto err;
1285 }
1286 break;
1287 }
1288
1289 // from PT_DYNAMIC we want to read address of first link_map addr
1290 case PT_DYNAMIC: {
1291 ph->core->dynamic_addr = exec_php->p_vaddr;
1292 print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr);
1293 break;
1294 }
1295
1296 } // switch
1297 exec_php++;
1298 } // for
1299
1300 free(phbuf);
1301 return true;
1302 err:
1303 free(phbuf);
1304 return false;
1305 }
1306
1307 #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug, r_map)
1308 #define LD_BASE_OFFSET offsetof(struct r_debug, r_ldbase)
1309 #define LINK_MAP_ADDR_OFFSET offsetof(struct link_map, l_addr)
1310 #define LINK_MAP_NAME_OFFSET offsetof(struct link_map, l_name)
1311 #define LINK_MAP_NEXT_OFFSET offsetof(struct link_map, l_next)
1312
1313 // read shared library info from runtime linker's data structures.
1314 // This work is done by librtlb_db in Solaris
1315 static bool read_shared_lib_info(struct ps_prochandle* ph) {
1316 uintptr_t addr = ph->core->dynamic_addr;
1317 uintptr_t debug_base;
1318 uintptr_t first_link_map_addr;
1319 uintptr_t ld_base_addr;
1320 uintptr_t link_map_addr;
1321 uintptr_t lib_base_diff;
1322 uintptr_t lib_base;
1323 uintptr_t lib_name_addr;
1324 char lib_name[BUF_SIZE];
1325 ELF_DYN dyn;
1326 ELF_EHDR elf_ehdr;
1327 int lib_fd;
1328
1329 // _DYNAMIC has information of the form
1330 // [tag] [data] [tag] [data] .....
1331 // Both tag and data are pointer sized.
1332 // We look for dynamic info with DT_DEBUG. This has shared object info.
1333 // refer to struct r_debug in link.h
1334
1335 dyn.d_tag = DT_NULL;
1336 while (dyn.d_tag != DT_DEBUG) {
1337 if (ps_pread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) {
1338 print_debug("can't read debug info from _DYNAMIC\n");
1339 return false;
1340 }
1341 addr += sizeof(ELF_DYN);
1342 }
1343
1344 // we have got Dyn entry with DT_DEBUG
1345 debug_base = dyn.d_un.d_ptr;
1346 // at debug_base we have struct r_debug. This has first link map in r_map field
1347 if (ps_pread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET,
1348 &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1349 print_debug("can't read first link map address\n");
1350 return false;
1351 }
1352
1353 // read ld_base address from struct r_debug
1354 #if 0 // There is no r_ldbase member on BSD
1355 if (ps_pread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr,
1356 sizeof(uintptr_t)) != PS_OK) {
1357 print_debug("can't read ld base address\n");
1358 return false;
1359 }
1360 ph->core->ld_base_addr = ld_base_addr;
1361 #else
1362 ph->core->ld_base_addr = 0;
1363 #endif
1364
1365 print_debug("interpreter base address is 0x%lx\n", ld_base_addr);
1366
1367 // now read segments from interp (i.e ld.so or ld-linux.so or ld-elf.so)
1368 if (read_interp_segments(ph) != true) {
1369 return false;
1370 }
1371
1372 // after adding interpreter (ld.so) mappings sort again
1373 if (sort_map_array(ph) != true) {
1374 return false;
1375 }
1376
1377 print_debug("first link map is at 0x%lx\n", first_link_map_addr);
1378
1379 link_map_addr = first_link_map_addr;
1380 while (link_map_addr != 0) {
1381 // read library base address of the .so. Note that even though <sys/link.h> calls
1382 // link_map->l_addr as "base address", this is * not * really base virtual
1383 // address of the shared object. This is actually the difference b/w the virtual
1384 // address mentioned in shared object and the actual virtual base where runtime
1385 // linker loaded it. We use "base diff" in read_lib_segments call below.
1386
1387 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET,
1388 &lib_base_diff, sizeof(uintptr_t)) != PS_OK) {
1389 print_debug("can't read shared object base address diff\n");
1390 return false;
1391 }
1392
1393 // read address of the name
1394 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET,
1395 &lib_name_addr, sizeof(uintptr_t)) != PS_OK) {
1396 print_debug("can't read address of shared object name\n");
1397 return false;
1398 }
1399
1400 // read name of the shared object
1401 if (read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) {
1402 print_debug("can't read shared object name\n");
1403 return false;
1404 }
1405
1406 if (lib_name[0] != '\0') {
1407 // ignore empty lib names
1408 lib_fd = pathmap_open(lib_name);
1409
1410 if (lib_fd < 0) {
1411 print_debug("can't open shared object %s\n", lib_name);
1412 // continue with other libraries...
1413 } else {
1414 if (read_elf_header(lib_fd, &elf_ehdr)) {
1415 lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr);
1416 print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n",
1417 lib_name, lib_base, lib_base_diff);
1418 // while adding library mappings we need to use "base difference".
1419 if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) {
1420 print_debug("can't read shared object's segments\n");
1421 close(lib_fd);
1422 return false;
1423 }
1424 add_lib_info_fd(ph, lib_name, lib_fd, lib_base);
1425 // Map info is added for the library (lib_name) so
1426 // we need to re-sort it before calling the p_pdread.
1427 if (sort_map_array(ph) != true) {
1428 return false;
1429 }
1430 } else {
1431 print_debug("can't read ELF header for shared object %s\n", lib_name);
1432 close(lib_fd);
1433 // continue with other libraries...
1434 }
1435 }
1436 }
1437
1438 // read next link_map address
1439 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET,
1440 &link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1441 print_debug("can't read next link in link_map\n");
1442 return false;
1443 }
1444 }
1445
1446 return true;
1447 }
1448
1449 // the one and only one exposed stuff from this file
1450 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
1451 ELF_EHDR core_ehdr;
1452 ELF_EHDR exec_ehdr;
1453
1454 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
1455 if (ph == NULL) {
1456 print_debug("can't allocate ps_prochandle\n");
1457 return NULL;
1458 }
1459
1460 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
1461 free(ph);
1462 print_debug("can't allocate ps_prochandle\n");
1463 return NULL;
1464 }
1465
1466 // initialize ph
1467 ph->ops = &core_ops;
1468 ph->core->core_fd = -1;
1469 ph->core->exec_fd = -1;
1470 ph->core->interp_fd = -1;
1471
1472 print_debug("exec: %s core: %s", exec_file, core_file);
1473
1474 // open the core file
1475 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
1476 print_debug("can't open core file\n");
1477 goto err;
1478 }
1479
1480 // read core file ELF header
1481 if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) {
1482 print_debug("core file is not a valid ELF ET_CORE file\n");
1483 goto err;
1484 }
1485
1486 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
1487 print_debug("can't open executable file\n");
1488 goto err;
1489 }
1490
1491 if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) {
1492 print_debug("executable file is not a valid ELF ET_EXEC file\n");
1493 goto err;
1494 }
1495
1496 // process core file segments
1497 if (read_core_segments(ph, &core_ehdr) != true) {
1498 goto err;
1499 }
1500
1501 // process exec file segments
1502 if (read_exec_segments(ph, &exec_ehdr) != true) {
1503 goto err;
1504 }
1505
1506 // exec file is also treated like a shared object for symbol search
1507 if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd,
1508 (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL) {
1509 goto err;
1510 }
1511
1512 // allocate and sort maps into map_array, we need to do this
1513 // here because read_shared_lib_info needs to read from debuggee
1514 // address space
1515 if (sort_map_array(ph) != true) {
1516 goto err;
1517 }
1518
1519 if (read_shared_lib_info(ph) != true) {
1520 goto err;
1521 }
1522
1523 // sort again because we have added more mappings from shared objects
1524 if (sort_map_array(ph) != true) {
1525 goto err;
1526 }
1527
1528 if (init_classsharing_workaround(ph) != true) {
1529 goto err;
1530 }
1531
1532 print_debug("Leave Pgrab_core\n");
1533 return ph;
1534
1535 err:
1536 Prelease(ph);
1537 return NULL;
1538 }
1539
1540 #endif // __APPLE__