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 #ifdef __APPLE__
309 #define USE_SHARED_SPACES_SYM "_UseSharedSpaces"
310 // mangled name of Arguments::SharedArchivePath
311 #define SHARED_ARCHIVE_PATH_SYM "_ZN9Arguments17SharedArchivePathE"
312 #define LIBJVM_NAME "/libjvm.dylib"
313 #else
314 #define USE_SHARED_SPACES_SYM "UseSharedSpaces"
315 // mangled name of Arguments::SharedArchivePath
316 #define SHARED_ARCHIVE_PATH_SYM "__ZN9Arguments17SharedArchivePathE"
317 #define LIBJVM_NAME "/libjvm.so"
318 #endif // __APPLE_
319
320 static bool init_classsharing_workaround(struct ps_prochandle* ph) {
321 int m;
322 size_t n;
323 lib_info* lib = ph->libs;
324 while (lib != NULL) {
325 // we are iterating over shared objects from the core dump. look for
326 // libjvm.so.
327 const char *jvm_name = 0;
328 if ((jvm_name = strstr(lib->name, LIBJVM_NAME)) != 0) {
329 char classes_jsa[PATH_MAX];
330 struct FileMapHeader header;
331 int fd = -1;
332 uintptr_t base = 0, useSharedSpacesAddr = 0;
333 uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0;
334 jboolean useSharedSpaces = 0;
335
336 memset(classes_jsa, 0, sizeof(classes_jsa));
337 jvm_name = lib->name;
338 useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM);
339 if (useSharedSpacesAddr == 0) {
340 print_debug("can't lookup 'UseSharedSpaces' flag\n");
341 return false;
342 }
343
344 // Hotspot vm types are not exported to build this library. So
345 // using equivalent type jboolean to read the value of
346 // UseSharedSpaces which is same as hotspot type "bool".
347 if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) {
348 print_debug("can't read the value of 'UseSharedSpaces' flag\n");
349 return false;
350 }
351
352 if ((int)useSharedSpaces == 0) {
353 print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n");
354 return true;
355 }
356
357 sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM);
358 if (sharedArchivePathAddrAddr == 0) {
359 print_debug("can't lookup shared archive path symbol\n");
360 return false;
361 }
362
363 if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) {
364 print_debug("can't read shared archive path pointer\n");
365 return false;
366 }
367
368 if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) {
369 print_debug("can't read shared archive path value\n");
370 return false;
371 }
372
373 print_debug("looking for %s\n", classes_jsa);
374 // open the class sharing archive file
375 fd = pathmap_open(classes_jsa);
376 if (fd < 0) {
377 print_debug("can't open %s!\n", classes_jsa);
378 ph->core->classes_jsa_fd = -1;
379 return false;
380 } else {
381 print_debug("opened %s\n", classes_jsa);
382 }
383
384 // read FileMapHeader from the file
385 memset(&header, 0, sizeof(struct FileMapHeader));
386 if ((n = read(fd, &header, sizeof(struct FileMapHeader)))
387 != sizeof(struct FileMapHeader)) {
388 print_debug("can't read shared archive file map header from %s\n", classes_jsa);
389 close(fd);
390 return false;
391 }
392
393 // check file magic
394 if (header._magic != 0xf00baba2) {
395 print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n",
396 classes_jsa, header._magic);
397 close(fd);
398 return false;
399 }
400
401 // check version
402 if (header._version != CURRENT_ARCHIVE_VERSION) {
403 print_debug("%s has wrong shared archive file version %d, expecting %d\n",
404 classes_jsa, header._version, CURRENT_ARCHIVE_VERSION);
405 close(fd);
406 return false;
407 }
408
409 ph->core->classes_jsa_fd = fd;
410 // add read-only maps from classes.jsa to the list of maps
411 for (m = 0; m < NUM_SHARED_MAPS; m++) {
412 if (header._space[m]._read_only) {
413 base = (uintptr_t) header._space[m]._addr._base;
414 // no need to worry about the fractional pages at-the-end.
415 // possible fractional pages are handled by core_read_data.
416 add_class_share_map_info(ph, (off_t) header._space[m]._file_offset,
417 base, (size_t) header._space[m]._used);
418 print_debug("added a share archive map at 0x%lx\n", base);
419 }
420 }
421 return true;
422 }
423 lib = lib->next;
424 }
425 return true;
426 }
427
428 //---------------------------------------------------------------------------
429 // functions to handle map_info
430
431 // Order mappings based on virtual address. We use this function as the
432 // callback for sorting the array of map_info pointers.
433 static int core_cmp_mapping(const void *lhsp, const void *rhsp)
434 {
435 const map_info *lhs = *((const map_info **)lhsp);
436 const map_info *rhs = *((const map_info **)rhsp);
437
438 if (lhs->vaddr == rhs->vaddr) {
439 return (0);
440 }
441
442 return (lhs->vaddr < rhs->vaddr ? -1 : 1);
443 }
444
445 // we sort map_info by starting virtual address so that we can do
446 // binary search to read from an address.
447 static bool sort_map_array(struct ps_prochandle* ph) {
448 size_t num_maps = ph->core->num_maps;
449 map_info* map = ph->core->maps;
450 int i = 0;
451
452 // allocate map_array
453 map_info** array;
454 if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) {
455 print_debug("can't allocate memory for map array\n");
456 return false;
457 }
458
459 // add maps to array
460 while (map) {
461 array[i] = map;
462 i++;
463 map = map->next;
464 }
465
466 // sort is called twice. If this is second time, clear map array
467 if (ph->core->map_array) {
468 free(ph->core->map_array);
469 }
470 ph->core->map_array = array;
471 // sort the map_info array by base virtual address.
472 qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*),
473 core_cmp_mapping);
474
475 // print map
476 if (is_debug()) {
477 int j = 0;
478 print_debug("---- sorted virtual address map ----\n");
479 for (j = 0; j < ph->core->num_maps; j++) {
480 print_debug("base = 0x%lx\tsize = %d\n", ph->core->map_array[j]->vaddr,
481 ph->core->map_array[j]->memsz);
482 }
483 }
484
485 return true;
486 }
487
488 #ifndef MIN
489 #define MIN(x, y) (((x) < (y))? (x): (y))
490 #endif
491
492 static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) {
493 ssize_t resid = size;
494 int page_size=sysconf(_SC_PAGE_SIZE);
495 while (resid != 0) {
496 map_info *mp = core_lookup(ph, addr);
497 uintptr_t mapoff;
498 ssize_t len, rem;
499 off_t off;
500 int fd;
501
502 if (mp == NULL) {
503 break; /* No mapping for this address */
504 }
505
506 fd = mp->fd;
507 mapoff = addr - mp->vaddr;
508 len = MIN(resid, mp->memsz - mapoff);
509 off = mp->offset + mapoff;
510
511 if ((len = pread(fd, buf, len, off)) <= 0) {
512 break;
513 }
514
515 resid -= len;
516 addr += len;
517 buf = (char *)buf + len;
518
519 // mappings always start at page boundary. But, may end in fractional
520 // page. fill zeros for possible fractional page at the end of a mapping.
521 rem = mp->memsz % page_size;
522 if (rem > 0) {
523 rem = page_size - rem;
524 len = MIN(resid, rem);
525 resid -= len;
526 addr += len;
527 // we are not assuming 'buf' to be zero initialized.
528 memset(buf, 0, len);
529 buf += len;
530 }
531 }
532
533 if (resid) {
534 print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n",
535 size, addr, resid);
536 return false;
537 } else {
538 return true;
539 }
540 }
541
542 // null implementation for write
543 static bool core_write_data(struct ps_prochandle* ph,
544 uintptr_t addr, const char *buf , size_t size) {
545 return false;
546 }
547
548 static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id,
549 struct reg* regs) {
550 // for core we have cached the lwp regs after segment parsed
551 sa_thread_info* thr = ph->threads;
552 while (thr) {
553 if (thr->lwp_id == lwp_id) {
554 memcpy(regs, &thr->regs, sizeof(struct reg));
555 return true;
556 }
557 thr = thr->next;
558 }
559 return false;
560 }
561
562 static bool core_get_lwp_info(struct ps_prochandle *ph, lwpid_t id, void *info) {
563 print_debug("core_get_lwp_info not implemented\n");
564 return false;
565 }
566
567 static ps_prochandle_ops core_ops = {
568 .release= core_release,
569 .p_pread= core_read_data,
570 .p_pwrite= core_write_data,
571 .get_lwp_regs= core_get_lwp_regs,
572 .get_lwp_info= core_get_lwp_info
573 };
574
575 // from this point, mainly two blocks divided by def __APPLE__
576 // one for Macosx, the other for regular Bsd
577
578 #ifdef __APPLE__
579
580 void print_thread(sa_thread_info *threadinfo) {
581 print_debug("thread added: %d\n", threadinfo->lwp_id);
582 print_debug("registers:\n");
583 print_debug(" r_r15: 0x%" PRIx64 "\n", threadinfo->regs.r_r15);
584 print_debug(" r_r14: 0x%" PRIx64 "\n", threadinfo->regs.r_r14);
585 print_debug(" r_r13: 0x%" PRIx64 "\n", threadinfo->regs.r_r13);
586 print_debug(" r_r12: 0x%" PRIx64 "\n", threadinfo->regs.r_r12);
587 print_debug(" r_r11: 0x%" PRIx64 "\n", threadinfo->regs.r_r11);
588 print_debug(" r_r10: 0x%" PRIx64 "\n", threadinfo->regs.r_r10);
589 print_debug(" r_r9: 0x%" PRIx64 "\n", threadinfo->regs.r_r9);
590 print_debug(" r_r8: 0x%" PRIx64 "\n", threadinfo->regs.r_r8);
591 print_debug(" r_rdi: 0x%" PRIx64 "\n", threadinfo->regs.r_rdi);
592 print_debug(" r_rsi: 0x%" PRIx64 "\n", threadinfo->regs.r_rsi);
593 print_debug(" r_rbp: 0x%" PRIx64 "\n", threadinfo->regs.r_rbp);
594 print_debug(" r_rbx: 0x%" PRIx64 "\n", threadinfo->regs.r_rbx);
595 print_debug(" r_rdx: 0x%" PRIx64 "\n", threadinfo->regs.r_rdx);
596 print_debug(" r_rcx: 0x%" PRIx64 "\n", threadinfo->regs.r_rcx);
597 print_debug(" r_rax: 0x%" PRIx64 "\n", threadinfo->regs.r_rax);
598 print_debug(" r_fs: 0x%" PRIx32 "\n", threadinfo->regs.r_fs);
599 print_debug(" r_gs: 0x%" PRIx32 "\n", threadinfo->regs.r_gs);
600 print_debug(" r_rip 0x%" PRIx64 "\n", threadinfo->regs.r_rip);
601 print_debug(" r_cs: 0x%" PRIx64 "\n", threadinfo->regs.r_cs);
602 print_debug(" r_rsp: 0x%" PRIx64 "\n", threadinfo->regs.r_rsp);
603 print_debug(" r_rflags: 0x%" PRIx64 "\n", threadinfo->regs.r_rflags);
604 }
605
606 // read all segments64 commands from core file
607 // read all thread commands from core file
608 static bool read_core_segments(struct ps_prochandle* ph) {
609 int i = 0;
610 int num_threads = 0;
611 int fd = ph->core->core_fd;
612 off_t offset = 0;
613 mach_header_64 fhead;
614 load_command lcmd;
615 segment_command_64 segcmd;
616 // thread_command thrcmd;
617
618 lseek(fd, offset, SEEK_SET);
619 if(read(fd, (void *)&fhead, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
620 goto err;
621 }
622 print_debug("total commands: %d\n", fhead.ncmds);
623 offset += sizeof(mach_header_64);
624 for (i = 0; i < fhead.ncmds; i++) {
625 lseek(fd, offset, SEEK_SET);
626 if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
627 goto err;
628 }
629 offset += lcmd.cmdsize; // next command position
630 if (lcmd.cmd == LC_SEGMENT_64) {
631 lseek(fd, -sizeof(load_command), SEEK_CUR);
632 if (read(fd, (void *)&segcmd, sizeof(segment_command_64)) != sizeof(segment_command_64)) {
633 print_debug("failed to read LC_SEGMENT_64 i = %d!\n", i);
634 goto err;
635 }
636 if (add_map_info(ph, fd, segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize) == NULL) {
637 print_debug("Failed to add map_info at i = %d\n", i);
638 goto err;
639 }
640 print_debug("segment added: %" PRIu64 " 0x%" PRIx64 " %d\n",
641 segcmd.fileoff, segcmd.vmaddr, segcmd.vmsize);
642 } else if (lcmd.cmd == LC_THREAD || lcmd.cmd == LC_UNIXTHREAD) {
643 typedef struct thread_fc {
644 uint32_t flavor;
645 uint32_t count;
646 } thread_fc;
647 thread_fc fc;
648 uint32_t size = sizeof(load_command);
649 while (size < lcmd.cmdsize) {
650 if (read(fd, (void *)&fc, sizeof(thread_fc)) != sizeof(thread_fc)) {
651 printf("Reading flavor, count failed.\n");
652 goto err;
653 }
654 size += sizeof(thread_fc);
655 if (fc.flavor == x86_THREAD_STATE) {
656 x86_thread_state_t thrstate;
657 if (read(fd, (void *)&thrstate, sizeof(x86_thread_state_t)) != sizeof(x86_thread_state_t)) {
658 printf("Reading flavor, count failed.\n");
659 goto err;
660 }
661 size += sizeof(x86_thread_state_t);
662 // create thread info list, update lwp_id later
663 sa_thread_info* newthr = add_thread_info(ph, (pthread_t) -1, (lwpid_t) num_threads++);
664 if (newthr == NULL) {
665 printf("create thread_info failed\n");
666 goto err;
667 }
668
669 // note __DARWIN_UNIX03 depengs on other definitions
670 #if __DARWIN_UNIX03
671 #define get_register_v(regst, regname) \
672 regst.uts.ts64.__##regname
673 #else
674 #define get_register_v(regst, regname) \
675 regst.uts.ts64.##regname
676 #endif // __DARWIN_UNIX03
677 newthr->regs.r_rax = get_register_v(thrstate, rax);
678 newthr->regs.r_rbx = get_register_v(thrstate, rbx);
679 newthr->regs.r_rcx = get_register_v(thrstate, rcx);
680 newthr->regs.r_rdx = get_register_v(thrstate, rdx);
681 newthr->regs.r_rdi = get_register_v(thrstate, rdi);
682 newthr->regs.r_rsi = get_register_v(thrstate, rsi);
683 newthr->regs.r_rbp = get_register_v(thrstate, rbp);
684 newthr->regs.r_rsp = get_register_v(thrstate, rsp);
685 newthr->regs.r_r8 = get_register_v(thrstate, r8);
686 newthr->regs.r_r9 = get_register_v(thrstate, r9);
687 newthr->regs.r_r10 = get_register_v(thrstate, r10);
688 newthr->regs.r_r11 = get_register_v(thrstate, r11);
689 newthr->regs.r_r12 = get_register_v(thrstate, r12);
690 newthr->regs.r_r13 = get_register_v(thrstate, r13);
691 newthr->regs.r_r14 = get_register_v(thrstate, r14);
692 newthr->regs.r_r15 = get_register_v(thrstate, r15);
693 newthr->regs.r_rip = get_register_v(thrstate, rip);
694 newthr->regs.r_rflags = get_register_v(thrstate, rflags);
695 newthr->regs.r_cs = get_register_v(thrstate, cs);
696 newthr->regs.r_fs = get_register_v(thrstate, fs);
697 newthr->regs.r_gs = get_register_v(thrstate, gs);
698 print_thread(newthr);
699 } else if (fc.flavor == x86_FLOAT_STATE) {
700 x86_float_state_t flstate;
701 if (read(fd, (void *)&flstate, sizeof(x86_float_state_t)) != sizeof(x86_float_state_t)) {
702 print_debug("Reading flavor, count failed.\n");
703 goto err;
704 }
705 size += sizeof(x86_float_state_t);
706 } else if (fc.flavor == x86_EXCEPTION_STATE) {
707 x86_exception_state_t excpstate;
708 if (read(fd, (void *)&excpstate, sizeof(x86_exception_state_t)) != sizeof(x86_exception_state_t)) {
709 printf("Reading flavor, count failed.\n");
710 goto err;
711 }
712 size += sizeof(x86_exception_state_t);
713 }
714 }
715 }
716 }
717 return true;
718 err:
719 return false;
720 }
721
722 /**local function **/
723 bool exists(const char *fname) {
724 return access(fname, F_OK) == 0;
725 }
726
727 // we check: 1. lib
728 // 2. lib/server
729 // 3. jre/lib
730 // 4. jre/lib/server
731 // from: 1. exe path
732 // 2. JAVA_HOME
733 // 3. DYLD_LIBRARY_PATH
734 static bool get_real_path(struct ps_prochandle* ph, char *rpath) {
735 /** check if they exist in JAVA ***/
736 char* execname = ph->core->exec_path;
737 char filepath[4096];
738 char* filename = strrchr(rpath, '/'); // like /libjvm.dylib
739 if (filename == NULL) {
740 return false;
741 }
742
743 char* posbin = strstr(execname, "/bin/java");
744 if (posbin != NULL) {
745 memcpy(filepath, execname, posbin - execname); // not include trailing '/'
746 filepath[posbin - execname] = '\0';
747 } else {
748 char* java_home = getenv("JAVA_HOME");
749 if (java_home != NULL) {
750 strcpy(filepath, java_home);
751 } else {
752 char* dyldpath = getenv("DYLD_LIBRARY_PATH");
753 char* dypath = strtok(dyldpath, ":");
754 while (dypath != NULL) {
755 strcpy(filepath, dypath);
756 strcat(filepath, filename);
757 if (exists(filepath)) {
758 strcpy(rpath, filepath);
759 return true;
760 }
761 dypath = strtok(dyldpath, ":");
762 }
763 // not found
764 return false;
765 }
766 }
767 // for exec and java_home, jdkpath now is filepath
768 size_t filepath_base_size = strlen(filepath);
769
770 // first try /lib/ and /lib/server
771 strcat(filepath, "/lib");
772 strcat(filepath, filename);
773 if (exists(filepath)) {
774 strcpy(rpath, filepath);
775 return true;
776 }
777 char* pos = strstr(filepath, filename); // like /libjvm.dylib
778 *pos = '\0';
779 strcat(filepath, "/server");
780 strcat(filepath, filename);
781 if (exists(filepath)) {
782 strcpy(rpath, filepath);
783 return true;
784 }
785
786 // then try /jre/lib/ and /jre/lib/server
787 filepath[filepath_base_size] = '\0';
788 strcat(filepath, "/jre/lib");
789 strcat(filepath, filename);
790 if (exists(filepath)) {
791 strcpy(rpath, filepath);
792 return true;
793 }
794 pos = strstr(filepath, filename);
795 *pos = '\0';
796 strcat(filepath, "/server");
797 strcat(filepath, filename);
798 if (exists(filepath)) {
799 strcpy(rpath, filepath);
800 return true;
801 }
802
803 return false;
804 }
805
806 static bool read_shared_lib_info(struct ps_prochandle* ph) {
807 static int pagesize = 0;
808 int fd = ph->core->core_fd;
809 int i = 0, j;
810 uint32_t v;
811 mach_header_64 header; // used to check if a file header in segment
812 load_command lcmd;
813 dylib_command dylibcmd;
814
815 char name[BUF_SIZE]; // use to store name
816
817 if (pagesize == 0) {
818 pagesize = getpagesize();
819 print_debug("page size is %d\n", pagesize);
820 }
821 for (j = 0; j < ph->core->num_maps; j++) {
822 map_info *iter = ph->core->map_array[j]; // head
823 off_t fpos = iter->offset;
824 if (iter->fd != fd) {
825 // only search core file!
826 continue;
827 }
828 print_debug("map_info %d: vmaddr = 0x%016" PRIx64 " fileoff = %" PRIu64 " vmsize = %" PRIu64 "\n",
829 j, iter->vaddr, iter->offset, iter->memsz);
830 lseek(fd, fpos, SEEK_SET);
831 // we assume .dylib loaded at segment address --- which is true for JVM libraries
832 // multiple files may be loaded in one segment.
833 // if first word is not a magic word, means this segment does not contain lib file.
834 if (read(fd, (void *)&v, sizeof(uint32_t)) == sizeof(uint32_t)) {
835 if (v != MH_MAGIC_64) {
836 continue;
837 }
838 } else {
839 // may be encountered last map, which is not readable
840 continue;
841 }
842 while (ltell(fd) - iter->offset < iter->memsz) {
843 lseek(fd, fpos, SEEK_SET);
844 if (read(fd, (void *)&v, sizeof(uint32_t)) != sizeof(uint32_t)) {
845 break;
846 }
847 if (v != MH_MAGIC_64) {
848 fpos = (ltell(fd) + pagesize -1)/pagesize * pagesize;
849 continue;
850 }
851 lseek(fd, -sizeof(uint32_t), SEEK_CUR);
852 // this is the file begining to core file.
853 if (read(fd, (void *)&header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
854 goto err;
855 }
856 fpos = ltell(fd);
857
858 // found a mach-o file in this segment
859 for (i = 0; i < header.ncmds; i++) {
860 // read commands in this "file"
861 // LC_ID_DYLIB is the file itself for a .dylib
862 lseek(fd, fpos, SEEK_SET);
863 if (read(fd, (void *)&lcmd, sizeof(load_command)) != sizeof(load_command)) {
864 return false; // error
865 }
866 fpos += lcmd.cmdsize; // next command position
867 // make sure still within seg size.
868 if (fpos - lcmd.cmdsize - iter->offset > iter->memsz) {
869 print_debug("Warning: out of segement limit: %ld \n", fpos - lcmd.cmdsize - iter->offset);
870 break; // no need to iterate all commands
871 }
872 if (lcmd.cmd == LC_ID_DYLIB) {
873 lseek(fd, -sizeof(load_command), SEEK_CUR);
874 if (read(fd, (void *)&dylibcmd, sizeof(dylib_command)) != sizeof(dylib_command)) {
875 return false;
876 }
877 /**** name stored at dylib_command.dylib.name.offset, is a C string */
878 lseek(fd, dylibcmd.dylib.name.offset - sizeof(dylib_command), SEEK_CUR);
879 int j = 0;
880 while (j < BUF_SIZE) {
881 read(fd, (void *)(name + j), sizeof(char));
882 if (name[j] == '\0') break;
883 j++;
884 }
885 print_debug("%s\n", name);
886 // changed name from @rpath/xxxx.dylib to real path
887 if (strrchr(name, '@')) {
888 get_real_path(ph, name);
889 print_debug("get_real_path returned: %s\n", name);
890 }
891 add_lib_info(ph, name, iter->vaddr);
892 break;
893 }
894 }
895 // done with the file, advanced to next page to search more files
896 fpos = (ltell(fd) + pagesize - 1) / pagesize * pagesize;
897 }
898 }
899 return true;
900 err:
901 return false;
902 }
903
904 bool read_macho64_header(int fd, mach_header_64* core_header) {
905 bool is_macho = false;
906 if (fd < 0) return false;
907 off_t pos = ltell(fd);
908 lseek(fd, 0, SEEK_SET);
909 if (read(fd, (void *)core_header, sizeof(mach_header_64)) != sizeof(mach_header_64)) {
910 is_macho = false;
911 } else {
912 is_macho = (core_header->magic == MH_MAGIC_64 || core_header->magic == MH_CIGAM_64);
913 }
914 lseek(fd, pos, SEEK_SET);
915 return is_macho;
916 }
917
918 // the one and only one exposed stuff from this file
919 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
920 mach_header_64 core_header;
921 mach_header_64 exec_header;
922
923 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
924 if (ph == NULL) {
925 print_debug("cant allocate ps_prochandle\n");
926 return NULL;
927 }
928
929 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
930 free(ph);
931 print_debug("can't allocate ps_prochandle\n");
932 return NULL;
933 }
934
935 // initialize ph
936 ph->ops = &core_ops;
937 ph->core->core_fd = -1;
938 ph->core->exec_fd = -1;
939 ph->core->interp_fd = -1;
940
941 print_debug("exec: %s core: %s", exec_file, core_file);
942
943 strncpy(ph->core->exec_path, exec_file, sizeof(ph->core->exec_path));
944
945 // open the core file
946 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
947 print_error("can't open core file\n");
948 goto err;
949 }
950
951 // read core file header
952 if (read_macho64_header(ph->core->core_fd, &core_header) != true || core_header.filetype != MH_CORE) {
953 print_debug("core file is not a valid Mach-O file\n");
954 goto err;
955 }
956
957 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
958 print_error("can't open executable file\n");
959 goto err;
960 }
961
962 if (read_macho64_header(ph->core->exec_fd, &exec_header) != true ||
963 exec_header.filetype != MH_EXECUTE) {
964 print_error("executable file is not a valid Mach-O file\n");
965 goto err;
966 }
967
968 // process core file segments
969 if (read_core_segments(ph) != true) {
970 print_error("failed to read core segments\n");
971 goto err;
972 }
973
974 // allocate and sort maps into map_array, we need to do this
975 // here because read_shared_lib_info needs to read from debuggee
976 // address space
977 if (sort_map_array(ph) != true) {
978 print_error("failed to sort segment map array\n");
979 goto err;
980 }
981
982 if (read_shared_lib_info(ph) != true) {
983 print_error("failed to read libraries\n");
984 goto err;
985 }
986
987 // sort again because we have added more mappings from shared objects
988 if (sort_map_array(ph) != true) {
989 print_error("failed to sort segment map array\n");
990 goto err;
991 }
992
993 if (init_classsharing_workaround(ph) != true) {
994 print_error("failed to workaround classshareing\n");
995 goto err;
996 }
997
998 print_debug("Leave Pgrab_core\n");
999 return ph;
1000
1001 err:
1002 Prelease(ph);
1003 return NULL;
1004 }
1005
1006 #else // __APPLE__ (none macosx)
1007
1008 // read regs and create thread from core file
1009 static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) {
1010 // we have to read prstatus_t from buf
1011 // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t");
1012 prstatus_t* prstat = (prstatus_t*) buf;
1013 sa_thread_info* newthr;
1014 print_debug("got integer regset for lwp %d\n", prstat->pr_pid);
1015 // we set pthread_t to -1 for core dump
1016 if((newthr = add_thread_info(ph, (pthread_t) -1, prstat->pr_pid)) == NULL)
1017 return false;
1018
1019 // copy regs
1020 memcpy(&newthr->regs, &prstat->pr_reg, sizeof(struct reg));
1021
1022 if (is_debug()) {
1023 print_debug("integer regset\n");
1024 #ifdef i386
1025 // print the regset
1026 print_debug("\teax = 0x%x\n", newthr->regs.r_eax);
1027 print_debug("\tebx = 0x%x\n", newthr->regs.r_ebx);
1028 print_debug("\tecx = 0x%x\n", newthr->regs.r_ecx);
1029 print_debug("\tedx = 0x%x\n", newthr->regs.r_edx);
1030 print_debug("\tesp = 0x%x\n", newthr->regs.r_esp);
1031 print_debug("\tebp = 0x%x\n", newthr->regs.r_ebp);
1032 print_debug("\tesi = 0x%x\n", newthr->regs.r_esi);
1033 print_debug("\tedi = 0x%x\n", newthr->regs.r_edi);
1034 print_debug("\teip = 0x%x\n", newthr->regs.r_eip);
1035 #endif
1036
1037 #if defined(amd64) || defined(x86_64)
1038 // print the regset
1039 print_debug("\tr15 = 0x%lx\n", newthr->regs.r_r15);
1040 print_debug("\tr14 = 0x%lx\n", newthr->regs.r_r14);
1041 print_debug("\tr13 = 0x%lx\n", newthr->regs.r_r13);
1042 print_debug("\tr12 = 0x%lx\n", newthr->regs.r_r12);
1043 print_debug("\trbp = 0x%lx\n", newthr->regs.r_rbp);
1044 print_debug("\trbx = 0x%lx\n", newthr->regs.r_rbx);
1045 print_debug("\tr11 = 0x%lx\n", newthr->regs.r_r11);
1046 print_debug("\tr10 = 0x%lx\n", newthr->regs.r_r10);
1047 print_debug("\tr9 = 0x%lx\n", newthr->regs.r_r9);
1048 print_debug("\tr8 = 0x%lx\n", newthr->regs.r_r8);
1049 print_debug("\trax = 0x%lx\n", newthr->regs.r_rax);
1050 print_debug("\trcx = 0x%lx\n", newthr->regs.r_rcx);
1051 print_debug("\trdx = 0x%lx\n", newthr->regs.r_rdx);
1052 print_debug("\trsi = 0x%lx\n", newthr->regs.r_rsi);
1053 print_debug("\trdi = 0x%lx\n", newthr->regs.r_rdi);
1054 //print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax);
1055 print_debug("\trip = 0x%lx\n", newthr->regs.r_rip);
1056 print_debug("\tcs = 0x%lx\n", newthr->regs.r_cs);
1057 //print_debug("\teflags = 0x%lx\n", newthr->regs.eflags);
1058 print_debug("\trsp = 0x%lx\n", newthr->regs.r_rsp);
1059 print_debug("\tss = 0x%lx\n", newthr->regs.r_ss);
1060 //print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base);
1061 //print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base);
1062 //print_debug("\tds = 0x%lx\n", newthr->regs.ds);
1063 //print_debug("\tes = 0x%lx\n", newthr->regs.es);
1064 //print_debug("\tfs = 0x%lx\n", newthr->regs.fs);
1065 //print_debug("\tgs = 0x%lx\n", newthr->regs.gs);
1066 #endif
1067 }
1068
1069 return true;
1070 }
1071
1072 #define ROUNDUP(x, y) ((((x)+((y)-1))/(y))*(y))
1073
1074 // read NT_PRSTATUS entries from core NOTE segment
1075 static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) {
1076 char* buf = NULL;
1077 char* p = NULL;
1078 size_t size = note_phdr->p_filesz;
1079
1080 // we are interested in just prstatus entries. we will ignore the rest.
1081 // Advance the seek pointer to the start of the PT_NOTE data
1082 if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) {
1083 print_debug("failed to lseek to PT_NOTE data\n");
1084 return false;
1085 }
1086
1087 // Now process the PT_NOTE structures. Each one is preceded by
1088 // an Elf{32/64}_Nhdr structure describing its type and size.
1089 if ( (buf = (char*) malloc(size)) == NULL) {
1090 print_debug("can't allocate memory for reading core notes\n");
1091 goto err;
1092 }
1093
1094 // read notes into buffer
1095 if (read(ph->core->core_fd, buf, size) != size) {
1096 print_debug("failed to read notes, core file must have been truncated\n");
1097 goto err;
1098 }
1099
1100 p = buf;
1101 while (p < buf + size) {
1102 ELF_NHDR* notep = (ELF_NHDR*) p;
1103 char* descdata = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4);
1104 print_debug("Note header with n_type = %d and n_descsz = %u\n",
1105 notep->n_type, notep->n_descsz);
1106
1107 if (notep->n_type == NT_PRSTATUS) {
1108 if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) {
1109 return false;
1110 }
1111 }
1112 p = descdata + ROUNDUP(notep->n_descsz, 4);
1113 }
1114
1115 free(buf);
1116 return true;
1117
1118 err:
1119 if (buf) free(buf);
1120 return false;
1121 }
1122
1123 // read all segments from core file
1124 static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) {
1125 int i = 0;
1126 ELF_PHDR* phbuf = NULL;
1127 ELF_PHDR* core_php = NULL;
1128
1129 if ((phbuf = read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL)
1130 return false;
1131
1132 /*
1133 * Now iterate through the program headers in the core file.
1134 * We're interested in two types of Phdrs: PT_NOTE (which
1135 * contains a set of saved /proc structures), and PT_LOAD (which
1136 * represents a memory mapping from the process's address space).
1137 *
1138 * Difference b/w Solaris PT_NOTE and Linux/BSD PT_NOTE:
1139 *
1140 * In Solaris there are two PT_NOTE segments the first PT_NOTE (if present)
1141 * contains /proc structs in the pre-2.6 unstructured /proc format. the last
1142 * PT_NOTE has data in new /proc format.
1143 *
1144 * In Solaris, there is only one pstatus (process status). pstatus contains
1145 * integer register set among other stuff. For each LWP, we have one lwpstatus
1146 * entry that has integer regset for that LWP.
1147 *
1148 * Linux threads are actually 'clone'd processes. To support core analysis
1149 * of "multithreaded" process, Linux creates more than one pstatus (called
1150 * "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one
1151 * "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular
1152 * function "elf_core_dump".
1153 */
1154
1155 for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) {
1156 switch (core_php->p_type) {
1157 case PT_NOTE:
1158 if (core_handle_note(ph, core_php) != true) {
1159 goto err;
1160 }
1161 break;
1162
1163 case PT_LOAD: {
1164 if (core_php->p_filesz != 0) {
1165 if (add_map_info(ph, ph->core->core_fd, core_php->p_offset,
1166 core_php->p_vaddr, core_php->p_filesz) == NULL) goto err;
1167 }
1168 break;
1169 }
1170 }
1171
1172 core_php++;
1173 }
1174
1175 free(phbuf);
1176 return true;
1177 err:
1178 free(phbuf);
1179 return false;
1180 }
1181
1182 // read segments of a shared object
1183 static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) {
1184 int i = 0;
1185 ELF_PHDR* phbuf;
1186 ELF_PHDR* lib_php = NULL;
1187
1188 int page_size=sysconf(_SC_PAGE_SIZE);
1189
1190 if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL) {
1191 return false;
1192 }
1193
1194 // we want to process only PT_LOAD segments that are not writable.
1195 // i.e., text segments. The read/write/exec (data) segments would
1196 // have been already added from core file segments.
1197 for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) {
1198 if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) {
1199
1200 uintptr_t target_vaddr = lib_php->p_vaddr + lib_base;
1201 map_info *existing_map = core_lookup(ph, target_vaddr);
1202
1203 if (existing_map == NULL){
1204 if (add_map_info(ph, lib_fd, lib_php->p_offset,
1205 target_vaddr, lib_php->p_filesz) == NULL) {
1206 goto err;
1207 }
1208 } else {
1209 if ((existing_map->memsz != page_size) &&
1210 (existing_map->fd != lib_fd) &&
1211 (existing_map->memsz != lib_php->p_filesz)){
1212
1213 print_debug("address conflict @ 0x%lx (size = %ld, flags = %d\n)",
1214 target_vaddr, lib_php->p_filesz, lib_php->p_flags);
1215 goto err;
1216 }
1217
1218 /* replace PT_LOAD segment with library segment */
1219 print_debug("overwrote with new address mapping (memsz %ld -> %ld)\n",
1220 existing_map->memsz, lib_php->p_filesz);
1221
1222 existing_map->fd = lib_fd;
1223 existing_map->offset = lib_php->p_offset;
1224 existing_map->memsz = lib_php->p_filesz;
1225 }
1226 }
1227
1228 lib_php++;
1229 }
1230
1231 free(phbuf);
1232 return true;
1233 err:
1234 free(phbuf);
1235 return false;
1236 }
1237
1238 // process segments from interpreter (ld.so or ld-linux.so or ld-elf.so)
1239 static bool read_interp_segments(struct ps_prochandle* ph) {
1240 ELF_EHDR interp_ehdr;
1241
1242 if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) {
1243 print_debug("interpreter is not a valid ELF file\n");
1244 return false;
1245 }
1246
1247 if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) {
1248 print_debug("can't read segments of interpreter\n");
1249 return false;
1250 }
1251
1252 return true;
1253 }
1254
1255 // process segments of a a.out
1256 static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) {
1257 int i = 0;
1258 ELF_PHDR* phbuf = NULL;
1259 ELF_PHDR* exec_php = NULL;
1260
1261 if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL)
1262 return false;
1263
1264 for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) {
1265 switch (exec_php->p_type) {
1266
1267 // add mappings for PT_LOAD segments
1268 case PT_LOAD: {
1269 // add only non-writable segments of non-zero filesz
1270 if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) {
1271 if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err;
1272 }
1273 break;
1274 }
1275
1276 // read the interpreter and it's segments
1277 case PT_INTERP: {
1278 char interp_name[BUF_SIZE];
1279
1280 pread(ph->core->exec_fd, interp_name, MIN(exec_php->p_filesz, BUF_SIZE), exec_php->p_offset);
1281 print_debug("ELF interpreter %s\n", interp_name);
1282 // read interpreter segments as well
1283 if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) {
1284 print_debug("can't open runtime loader\n");
1285 goto err;
1286 }
1287 break;
1288 }
1289
1290 // from PT_DYNAMIC we want to read address of first link_map addr
1291 case PT_DYNAMIC: {
1292 ph->core->dynamic_addr = exec_php->p_vaddr;
1293 print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr);
1294 break;
1295 }
1296
1297 } // switch
1298 exec_php++;
1299 } // for
1300
1301 free(phbuf);
1302 return true;
1303 err:
1304 free(phbuf);
1305 return false;
1306 }
1307
1308 #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug, r_map)
1309 #define LD_BASE_OFFSET offsetof(struct r_debug, r_ldbase)
1310 #define LINK_MAP_ADDR_OFFSET offsetof(struct link_map, l_addr)
1311 #define LINK_MAP_NAME_OFFSET offsetof(struct link_map, l_name)
1312 #define LINK_MAP_NEXT_OFFSET offsetof(struct link_map, l_next)
1313
1314 // read shared library info from runtime linker's data structures.
1315 // This work is done by librtlb_db in Solaris
1316 static bool read_shared_lib_info(struct ps_prochandle* ph) {
1317 uintptr_t addr = ph->core->dynamic_addr;
1318 uintptr_t debug_base;
1319 uintptr_t first_link_map_addr;
1320 uintptr_t ld_base_addr;
1321 uintptr_t link_map_addr;
1322 uintptr_t lib_base_diff;
1323 uintptr_t lib_base;
1324 uintptr_t lib_name_addr;
1325 char lib_name[BUF_SIZE];
1326 ELF_DYN dyn;
1327 ELF_EHDR elf_ehdr;
1328 int lib_fd;
1329
1330 // _DYNAMIC has information of the form
1331 // [tag] [data] [tag] [data] .....
1332 // Both tag and data are pointer sized.
1333 // We look for dynamic info with DT_DEBUG. This has shared object info.
1334 // refer to struct r_debug in link.h
1335
1336 dyn.d_tag = DT_NULL;
1337 while (dyn.d_tag != DT_DEBUG) {
1338 if (ps_pread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) {
1339 print_debug("can't read debug info from _DYNAMIC\n");
1340 return false;
1341 }
1342 addr += sizeof(ELF_DYN);
1343 }
1344
1345 // we have got Dyn entry with DT_DEBUG
1346 debug_base = dyn.d_un.d_ptr;
1347 // at debug_base we have struct r_debug. This has first link map in r_map field
1348 if (ps_pread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET,
1349 &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1350 print_debug("can't read first link map address\n");
1351 return false;
1352 }
1353
1354 // read ld_base address from struct r_debug
1355 #if 0 // There is no r_ldbase member on BSD
1356 if (ps_pread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr,
1357 sizeof(uintptr_t)) != PS_OK) {
1358 print_debug("can't read ld base address\n");
1359 return false;
1360 }
1361 ph->core->ld_base_addr = ld_base_addr;
1362 #else
1363 ph->core->ld_base_addr = 0;
1364 #endif
1365
1366 print_debug("interpreter base address is 0x%lx\n", ld_base_addr);
1367
1368 // now read segments from interp (i.e ld.so or ld-linux.so or ld-elf.so)
1369 if (read_interp_segments(ph) != true) {
1370 return false;
1371 }
1372
1373 // after adding interpreter (ld.so) mappings sort again
1374 if (sort_map_array(ph) != true) {
1375 return false;
1376 }
1377
1378 print_debug("first link map is at 0x%lx\n", first_link_map_addr);
1379
1380 link_map_addr = first_link_map_addr;
1381 while (link_map_addr != 0) {
1382 // read library base address of the .so. Note that even though <sys/link.h> calls
1383 // link_map->l_addr as "base address", this is * not * really base virtual
1384 // address of the shared object. This is actually the difference b/w the virtual
1385 // address mentioned in shared object and the actual virtual base where runtime
1386 // linker loaded it. We use "base diff" in read_lib_segments call below.
1387
1388 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET,
1389 &lib_base_diff, sizeof(uintptr_t)) != PS_OK) {
1390 print_debug("can't read shared object base address diff\n");
1391 return false;
1392 }
1393
1394 // read address of the name
1395 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET,
1396 &lib_name_addr, sizeof(uintptr_t)) != PS_OK) {
1397 print_debug("can't read address of shared object name\n");
1398 return false;
1399 }
1400
1401 // read name of the shared object
1402 if (read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) {
1403 print_debug("can't read shared object name\n");
1404 return false;
1405 }
1406
1407 if (lib_name[0] != '\0') {
1408 // ignore empty lib names
1409 lib_fd = pathmap_open(lib_name);
1410
1411 if (lib_fd < 0) {
1412 print_debug("can't open shared object %s\n", lib_name);
1413 // continue with other libraries...
1414 } else {
1415 if (read_elf_header(lib_fd, &elf_ehdr)) {
1416 lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr);
1417 print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n",
1418 lib_name, lib_base, lib_base_diff);
1419 // while adding library mappings we need to use "base difference".
1420 if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) {
1421 print_debug("can't read shared object's segments\n");
1422 close(lib_fd);
1423 return false;
1424 }
1425 add_lib_info_fd(ph, lib_name, lib_fd, lib_base);
1426 // Map info is added for the library (lib_name) so
1427 // we need to re-sort it before calling the p_pdread.
1428 if (sort_map_array(ph) != true) {
1429 return false;
1430 }
1431 } else {
1432 print_debug("can't read ELF header for shared object %s\n", lib_name);
1433 close(lib_fd);
1434 // continue with other libraries...
1435 }
1436 }
1437 }
1438
1439 // read next link_map address
1440 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET,
1441 &link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1442 print_debug("can't read next link in link_map\n");
1443 return false;
1444 }
1445 }
1446
1447 return true;
1448 }
1449
1450 // the one and only one exposed stuff from this file
1451 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
1452 ELF_EHDR core_ehdr;
1453 ELF_EHDR exec_ehdr;
1454
1455 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
1456 if (ph == NULL) {
1457 print_debug("can't allocate ps_prochandle\n");
1458 return NULL;
1459 }
1460
1461 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
1462 free(ph);
1463 print_debug("can't allocate ps_prochandle\n");
1464 return NULL;
1465 }
1466
1467 // initialize ph
1468 ph->ops = &core_ops;
1469 ph->core->core_fd = -1;
1470 ph->core->exec_fd = -1;
1471 ph->core->interp_fd = -1;
1472
1473 print_debug("exec: %s core: %s", exec_file, core_file);
1474
1475 // open the core file
1476 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
1477 print_debug("can't open core file\n");
1478 goto err;
1479 }
1480
1481 // read core file ELF header
1482 if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) {
1483 print_debug("core file is not a valid ELF ET_CORE file\n");
1484 goto err;
1485 }
1486
1487 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
1488 print_debug("can't open executable file\n");
1489 goto err;
1490 }
1491
1492 if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) {
1493 print_debug("executable file is not a valid ELF ET_EXEC file\n");
1494 goto err;
1495 }
1496
1497 // process core file segments
1498 if (read_core_segments(ph, &core_ehdr) != true) {
1499 goto err;
1500 }
1501
1502 // process exec file segments
1503 if (read_exec_segments(ph, &exec_ehdr) != true) {
1504 goto err;
1505 }
1506
1507 // exec file is also treated like a shared object for symbol search
1508 if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd,
1509 (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL) {
1510 goto err;
1511 }
1512
1513 // allocate and sort maps into map_array, we need to do this
1514 // here because read_shared_lib_info needs to read from debuggee
1515 // address space
1516 if (sort_map_array(ph) != true) {
1517 goto err;
1518 }
1519
1520 if (read_shared_lib_info(ph) != true) {
1521 goto err;
1522 }
1523
1524 // sort again because we have added more mappings from shared objects
1525 if (sort_map_array(ph) != true) {
1526 goto err;
1527 }
1528
1529 if (init_classsharing_workaround(ph) != true) {
1530 goto err;
1531 }
1532
1533 print_debug("Leave Pgrab_core\n");
1534 return ph;
1535
1536 err:
1537 Prelease(ph);
1538 return NULL;
1539 }
1540
1541 #endif // __APPLE__