1 /*
2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include <jni.h>
26 #include <unistd.h>
27 #include <fcntl.h>
28 #include <string.h>
29 #include <stdlib.h>
30 #include <stddef.h>
31 #include <elf.h>
32 #include <link.h>
33 #include "libproc_impl.h"
34 #include "salibelf.h"
35
36 // This file has the libproc implementation to read core files.
37 // For live processes, refer to ps_proc.c. Portions of this is adapted
38 // /modelled after Solaris libproc.so (in particular Pcore.c)
39
40 //----------------------------------------------------------------------
41 // ps_prochandle cleanup helper functions
42
43 // close all file descriptors
44 static void close_elf_files(struct ps_prochandle* ph) {
45 lib_info* lib = NULL;
46
47 // close core file descriptor
48 if (ph->core->core_fd >= 0)
49 close(ph->core->core_fd);
50
51 // close exec file descriptor
52 if (ph->core->exec_fd >= 0)
53 close(ph->core->exec_fd);
54
55 // close interp file descriptor
56 if (ph->core->interp_fd >= 0)
57 close(ph->core->interp_fd);
58
59 // close class share archive file
60 if (ph->core->classes_jsa_fd >= 0)
61 close(ph->core->classes_jsa_fd);
62
63 // close all library file descriptors
64 lib = ph->libs;
65 while (lib) {
66 int fd = lib->fd;
67 if (fd >= 0 && fd != ph->core->exec_fd) close(fd);
68 lib = lib->next;
69 }
70 }
71
72 // clean all map_info stuff
73 static void destroy_map_info(struct ps_prochandle* ph) {
74 map_info* map = ph->core->maps;
75 while (map) {
76 map_info* next = map->next;
77 free(map);
78 map = next;
79 }
80
81 if (ph->core->map_array) {
82 free(ph->core->map_array);
83 }
84
85 // Part of the class sharing workaround
86 map = ph->core->class_share_maps;
87 while (map) {
88 map_info* next = map->next;
89 free(map);
90 map = next;
91 }
92 }
93
94 // ps_prochandle operations
95 static void core_release(struct ps_prochandle* ph) {
96 if (ph->core) {
97 close_elf_files(ph);
98 destroy_map_info(ph);
99 free(ph->core);
100 }
101 }
102
103 static map_info* allocate_init_map(int fd, off_t offset, uintptr_t vaddr, size_t memsz) {
104 map_info* map;
105 if ( (map = (map_info*) calloc(1, sizeof(map_info))) == NULL) {
106 print_debug("can't allocate memory for map_info\n");
107 return NULL;
108 }
109
110 // initialize map
111 map->fd = fd;
112 map->offset = offset;
113 map->vaddr = vaddr;
114 map->memsz = memsz;
115 return map;
116 }
117
118 // add map info with given fd, offset, vaddr and memsz
119 static map_info* add_map_info(struct ps_prochandle* ph, int fd, off_t offset,
120 uintptr_t vaddr, size_t memsz) {
121 map_info* map;
122 if ((map = allocate_init_map(fd, offset, vaddr, memsz)) == NULL) {
123 return NULL;
124 }
125
126 // add this to map list
127 map->next = ph->core->maps;
128 ph->core->maps = map;
129 ph->core->num_maps++;
130
131 return map;
132 }
133
134 // Part of the class sharing workaround
135 static map_info* add_class_share_map_info(struct ps_prochandle* ph, off_t offset,
136 uintptr_t vaddr, size_t memsz) {
137 map_info* map;
138 if ((map = allocate_init_map(ph->core->classes_jsa_fd,
139 offset, vaddr, memsz)) == NULL) {
140 return NULL;
141 }
142
143 map->next = ph->core->class_share_maps;
144 ph->core->class_share_maps = map;
145 return map;
146 }
147
148 // Return the map_info for the given virtual address. We keep a sorted
149 // array of pointers in ph->map_array, so we can binary search.
150 static map_info* core_lookup(struct ps_prochandle *ph, uintptr_t addr)
151 {
152 int mid, lo = 0, hi = ph->core->num_maps - 1;
153 map_info *mp;
154
155 while (hi - lo > 1) {
156 mid = (lo + hi) / 2;
157 if (addr >= ph->core->map_array[mid]->vaddr)
158 lo = mid;
159 else
160 hi = mid;
161 }
162
163 if (addr < ph->core->map_array[hi]->vaddr)
164 mp = ph->core->map_array[lo];
165 else
166 mp = ph->core->map_array[hi];
167
168 if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz)
169 return (mp);
170
171
172 // Part of the class sharing workaround
173 // Unfortunately, we have no way of detecting -Xshare state.
174 // Check out the share maps atlast, if we don't find anywhere.
175 // This is done this way so to avoid reading share pages
176 // ahead of other normal maps. For eg. with -Xshare:off we don't
177 // want to prefer class sharing data to data from core.
178 mp = ph->core->class_share_maps;
179 if (mp) {
180 print_debug("can't locate map_info at 0x%lx, trying class share maps\n",
181 addr);
182 }
183 while (mp) {
184 if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) {
185 print_debug("located map_info at 0x%lx from class share maps\n",
186 addr);
187 return (mp);
188 }
189 mp = mp->next;
190 }
191
192 print_debug("can't locate map_info at 0x%lx\n", addr);
193 return (NULL);
194 }
195
196 //---------------------------------------------------------------
197 // Part of the class sharing workaround:
198 //
199 // With class sharing, pages are mapped from classes[_g].jsa file.
200 // The read-only class sharing pages are mapped as MAP_SHARED,
201 // PROT_READ pages. These pages are not dumped into core dump.
202 // With this workaround, these pages are read from classes[_g].jsa.
203
204 // FIXME: !HACK ALERT!
205 // The format of sharing achive file header is needed to read shared heap
206 // file mappings. For now, I am hard coding portion of FileMapHeader here.
207 // Refer to filemap.hpp.
208
209 // FileMapHeader describes the shared space data in the file to be
210 // mapped. This structure gets written to a file. It is not a class,
211 // so that the compilers don't add any compiler-private data to it.
212
213 // Refer to CompactingPermGenGen::n_regions in compactingPermGenGen.hpp
214 #define NUM_SHARED_MAPS 4
215
216 // Refer to FileMapInfo::_current_version in filemap.hpp
217 #define CURRENT_ARCHIVE_VERSION 1
218
219 struct FileMapHeader {
220 int _magic; // identify file type.
221 int _version; // (from enum, above.)
222 size_t _alignment; // how shared archive should be aligned
223
224 struct space_info {
225 int _file_offset; // sizeof(this) rounded to vm page size
226 char* _base; // copy-on-write base address
227 size_t _capacity; // for validity checking
228 size_t _used; // for setting space top on read
229
230 // 4991491 NOTICE These are C++ bool's in filemap.hpp and must match up with
231 // the C type matching the C++ bool type on any given platform. For
232 // Hotspot on BSD we assume the corresponding C type is char but
233 // licensees on BSD versions may need to adjust the type of these fields.
234 char _read_only; // read only space?
235 char _allow_exec; // executable code in space?
236
237 } _space[NUM_SHARED_MAPS]; // was _space[CompactingPermGenGen::n_regions];
238
239 // Ignore the rest of the FileMapHeader. We don't need those fields here.
240 };
241
242 static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) {
243 jboolean i;
244 if (ps_pread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) {
245 *pvalue = i;
246 return true;
247 } else {
248 return false;
249 }
250 }
251
252 static bool read_pointer(struct ps_prochandle* ph, uintptr_t addr, uintptr_t* pvalue) {
253 uintptr_t uip;
254 if (ps_pread(ph, (psaddr_t) addr, &uip, sizeof(uip)) == PS_OK) {
255 *pvalue = uip;
256 return true;
257 } else {
258 return false;
259 }
260 }
261
262 // used to read strings from debuggee
263 static bool read_string(struct ps_prochandle* ph, uintptr_t addr, char* buf, size_t size) {
264 size_t i = 0;
265 char c = ' ';
266
267 while (c != '\0') {
268 if (ps_pread(ph, (psaddr_t) addr, &c, sizeof(char)) != PS_OK)
269 return false;
270 if (i < size - 1)
271 buf[i] = c;
272 else // smaller buffer
273 return false;
274 i++; addr++;
275 }
276
277 buf[i] = '\0';
278 return true;
279 }
280
281 #define USE_SHARED_SPACES_SYM "UseSharedSpaces"
282 // mangled name of Arguments::SharedArchivePath
283 #define SHARED_ARCHIVE_PATH_SYM "_ZN9Arguments17SharedArchivePathE"
284
285 static bool init_classsharing_workaround(struct ps_prochandle* ph) {
286 lib_info* lib = ph->libs;
287 while (lib != NULL) {
288 // we are iterating over shared objects from the core dump. look for
289 // libjvm[_g].so.
290 const char *jvm_name = 0;
291 if ((jvm_name = strstr(lib->name, "/libjvm.so")) != 0 ||
292 (jvm_name = strstr(lib->name, "/libjvm_g.so")) != 0) {
293 char classes_jsa[PATH_MAX];
294 struct FileMapHeader header;
295 size_t n = 0;
296 int fd = -1, m = 0;
297 uintptr_t base = 0, useSharedSpacesAddr = 0;
298 uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0;
299 jboolean useSharedSpaces = 0;
300
301 memset(classes_jsa, 0, sizeof(classes_jsa));
302 jvm_name = lib->name;
303 useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM);
304 if (useSharedSpacesAddr == 0) {
305 print_debug("can't lookup 'UseSharedSpaces' flag\n");
306 return false;
307 }
308
309 // Hotspot vm types are not exported to build this library. So
310 // using equivalent type jboolean to read the value of
311 // UseSharedSpaces which is same as hotspot type "bool".
312 if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) {
313 print_debug("can't read the value of 'UseSharedSpaces' flag\n");
314 return false;
315 }
316
317 if ((int)useSharedSpaces == 0) {
318 print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n");
319 return true;
320 }
321
322 sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM);
323 if (sharedArchivePathAddrAddr == 0) {
324 print_debug("can't lookup shared archive path symbol\n");
325 return false;
326 }
327
328 if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) {
329 print_debug("can't read shared archive path pointer\n");
330 return false;
331 }
332
333 if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) {
334 print_debug("can't read shared archive path value\n");
335 return false;
336 }
337
338 print_debug("looking for %s\n", classes_jsa);
339 // open the class sharing archive file
340 fd = pathmap_open(classes_jsa);
341 if (fd < 0) {
342 print_debug("can't open %s!\n", classes_jsa);
343 ph->core->classes_jsa_fd = -1;
344 return false;
345 } else {
346 print_debug("opened %s\n", classes_jsa);
347 }
348
349 // read FileMapHeader from the file
350 memset(&header, 0, sizeof(struct FileMapHeader));
351 if ((n = read(fd, &header, sizeof(struct FileMapHeader)))
352 != sizeof(struct FileMapHeader)) {
353 print_debug("can't read shared archive file map header from %s\n", classes_jsa);
354 close(fd);
355 return false;
356 }
357
358 // check file magic
359 if (header._magic != 0xf00baba2) {
360 print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n",
361 classes_jsa, header._magic);
362 close(fd);
363 return false;
364 }
365
366 // check version
367 if (header._version != CURRENT_ARCHIVE_VERSION) {
368 print_debug("%s has wrong shared archive file version %d, expecting %d\n",
369 classes_jsa, header._version, CURRENT_ARCHIVE_VERSION);
370 close(fd);
371 return false;
372 }
373
374 ph->core->classes_jsa_fd = fd;
375 // add read-only maps from classes[_g].jsa to the list of maps
376 for (m = 0; m < NUM_SHARED_MAPS; m++) {
377 if (header._space[m]._read_only) {
378 base = (uintptr_t) header._space[m]._base;
379 // no need to worry about the fractional pages at-the-end.
380 // possible fractional pages are handled by core_read_data.
381 add_class_share_map_info(ph, (off_t) header._space[m]._file_offset,
382 base, (size_t) header._space[m]._used);
383 print_debug("added a share archive map at 0x%lx\n", base);
384 }
385 }
386 return true;
387 }
388 lib = lib->next;
389 }
390 return true;
391 }
392
393
394 //---------------------------------------------------------------------------
395 // functions to handle map_info
396
397 // Order mappings based on virtual address. We use this function as the
398 // callback for sorting the array of map_info pointers.
399 static int core_cmp_mapping(const void *lhsp, const void *rhsp)
400 {
401 const map_info *lhs = *((const map_info **)lhsp);
402 const map_info *rhs = *((const map_info **)rhsp);
403
404 if (lhs->vaddr == rhs->vaddr)
405 return (0);
406
407 return (lhs->vaddr < rhs->vaddr ? -1 : 1);
408 }
409
410 // we sort map_info by starting virtual address so that we can do
411 // binary search to read from an address.
412 static bool sort_map_array(struct ps_prochandle* ph) {
413 size_t num_maps = ph->core->num_maps;
414 map_info* map = ph->core->maps;
415 int i = 0;
416
417 // allocate map_array
418 map_info** array;
419 if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) {
420 print_debug("can't allocate memory for map array\n");
421 return false;
422 }
423
424 // add maps to array
425 while (map) {
426 array[i] = map;
427 i++;
428 map = map->next;
429 }
430
431 // sort is called twice. If this is second time, clear map array
432 if (ph->core->map_array) free(ph->core->map_array);
433 ph->core->map_array = array;
434 // sort the map_info array by base virtual address.
435 qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*),
436 core_cmp_mapping);
437
438 // print map
439 if (is_debug()) {
440 int j = 0;
441 print_debug("---- sorted virtual address map ----\n");
442 for (j = 0; j < ph->core->num_maps; j++) {
443 print_debug("base = 0x%lx\tsize = %d\n", ph->core->map_array[j]->vaddr,
444 ph->core->map_array[j]->memsz);
445 }
446 }
447
448 return true;
449 }
450
451 #ifndef MIN
452 #define MIN(x, y) (((x) < (y))? (x): (y))
453 #endif
454
455 static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) {
456 ssize_t resid = size;
457 int page_size=sysconf(_SC_PAGE_SIZE);
458 while (resid != 0) {
459 map_info *mp = core_lookup(ph, addr);
460 uintptr_t mapoff;
461 ssize_t len, rem;
462 off_t off;
463 int fd;
464
465 if (mp == NULL)
466 break; /* No mapping for this address */
467
468 fd = mp->fd;
469 mapoff = addr - mp->vaddr;
470 len = MIN(resid, mp->memsz - mapoff);
471 off = mp->offset + mapoff;
472
473 if ((len = pread(fd, buf, len, off)) <= 0)
474 break;
475
476 resid -= len;
477 addr += len;
478 buf = (char *)buf + len;
479
480 // mappings always start at page boundary. But, may end in fractional
481 // page. fill zeros for possible fractional page at the end of a mapping.
482 rem = mp->memsz % page_size;
483 if (rem > 0) {
484 rem = page_size - rem;
485 len = MIN(resid, rem);
486 resid -= len;
487 addr += len;
488 // we are not assuming 'buf' to be zero initialized.
489 memset(buf, 0, len);
490 buf += len;
491 }
492 }
493
494 if (resid) {
495 print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n",
496 size, addr, resid);
497 return false;
498 } else {
499 return true;
500 }
501 }
502
503 // null implementation for write
504 static bool core_write_data(struct ps_prochandle* ph,
505 uintptr_t addr, const char *buf , size_t size) {
506 return false;
507 }
508
509 static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id,
510 struct reg* regs) {
511 // for core we have cached the lwp regs from NOTE section
512 thread_info* thr = ph->threads;
513 while (thr) {
514 if (thr->lwp_id == lwp_id) {
515 memcpy(regs, &thr->regs, sizeof(struct reg));
516 return true;
517 }
518 thr = thr->next;
519 }
520 return false;
521 }
522
523 static bool core_get_lwp_info(struct ps_prochandle *ph, lwpid_t lwp_id, void *linfo) {
524 print_debug("core_get_lwp_info not implemented\n");
525 return false;
526 }
527
528 static ps_prochandle_ops core_ops = {
529 .release= core_release,
530 .p_pread= core_read_data,
531 .p_pwrite= core_write_data,
532 .get_lwp_regs= core_get_lwp_regs,
533 .get_lwp_info= core_get_lwp_info
534 };
535
536 // read regs and create thread from NT_PRSTATUS entries from core file
537 static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) {
538 // we have to read prstatus_t from buf
539 // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t");
540 prstatus_t* prstat = (prstatus_t*) buf;
541 thread_info* newthr;
542 print_debug("got integer regset for lwp %d\n", prstat->pr_pid);
543 // we set pthread_t to -1 for core dump
544 if((newthr = add_thread_info(ph, (pthread_t) -1, prstat->pr_pid)) == NULL)
545 return false;
546
547 // copy regs
548 memcpy(&newthr->regs, &prstat->pr_reg, sizeof(struct reg));
549
550 if (is_debug()) {
551 print_debug("integer regset\n");
552 #ifdef i386
553 // print the regset
554 print_debug("\teax = 0x%x\n", newthr->regs.r_eax);
555 print_debug("\tebx = 0x%x\n", newthr->regs.r_ebx);
556 print_debug("\tecx = 0x%x\n", newthr->regs.r_ecx);
557 print_debug("\tedx = 0x%x\n", newthr->regs.r_edx);
558 print_debug("\tesp = 0x%x\n", newthr->regs.r_esp);
559 print_debug("\tebp = 0x%x\n", newthr->regs.r_ebp);
560 print_debug("\tesi = 0x%x\n", newthr->regs.r_esi);
561 print_debug("\tedi = 0x%x\n", newthr->regs.r_edi);
562 print_debug("\teip = 0x%x\n", newthr->regs.r_eip);
563 #endif
564
565 #if defined(amd64) || defined(x86_64)
566 // print the regset
567 print_debug("\tr15 = 0x%lx\n", newthr->regs.r_r15);
568 print_debug("\tr14 = 0x%lx\n", newthr->regs.r_r14);
569 print_debug("\tr13 = 0x%lx\n", newthr->regs.r_r13);
570 print_debug("\tr12 = 0x%lx\n", newthr->regs.r_r12);
571 print_debug("\trbp = 0x%lx\n", newthr->regs.r_rbp);
572 print_debug("\trbx = 0x%lx\n", newthr->regs.r_rbx);
573 print_debug("\tr11 = 0x%lx\n", newthr->regs.r_r11);
574 print_debug("\tr10 = 0x%lx\n", newthr->regs.r_r10);
575 print_debug("\tr9 = 0x%lx\n", newthr->regs.r_r9);
576 print_debug("\tr8 = 0x%lx\n", newthr->regs.r_r8);
577 print_debug("\trax = 0x%lx\n", newthr->regs.r_rax);
578 print_debug("\trcx = 0x%lx\n", newthr->regs.r_rcx);
579 print_debug("\trdx = 0x%lx\n", newthr->regs.r_rdx);
580 print_debug("\trsi = 0x%lx\n", newthr->regs.r_rsi);
581 print_debug("\trdi = 0x%lx\n", newthr->regs.r_rdi);
582 //print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax);
583 print_debug("\trip = 0x%lx\n", newthr->regs.r_rip);
584 print_debug("\tcs = 0x%lx\n", newthr->regs.r_cs);
585 //print_debug("\teflags = 0x%lx\n", newthr->regs.eflags);
586 print_debug("\trsp = 0x%lx\n", newthr->regs.r_rsp);
587 print_debug("\tss = 0x%lx\n", newthr->regs.r_ss);
588 //print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base);
589 //print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base);
590 //print_debug("\tds = 0x%lx\n", newthr->regs.ds);
591 //print_debug("\tes = 0x%lx\n", newthr->regs.es);
592 //print_debug("\tfs = 0x%lx\n", newthr->regs.fs);
593 //print_debug("\tgs = 0x%lx\n", newthr->regs.gs);
594 #endif
595 }
596
597 return true;
598 }
599
600 #define ROUNDUP(x, y) ((((x)+((y)-1))/(y))*(y))
601
602 // read NT_PRSTATUS entries from core NOTE segment
603 static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) {
604 char* buf = NULL;
605 char* p = NULL;
606 size_t size = note_phdr->p_filesz;
607
608 // we are interested in just prstatus entries. we will ignore the rest.
609 // Advance the seek pointer to the start of the PT_NOTE data
610 if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) {
611 print_debug("failed to lseek to PT_NOTE data\n");
612 return false;
613 }
614
615 // Now process the PT_NOTE structures. Each one is preceded by
616 // an Elf{32/64}_Nhdr structure describing its type and size.
617 if ( (buf = (char*) malloc(size)) == NULL) {
618 print_debug("can't allocate memory for reading core notes\n");
619 goto err;
620 }
621
622 // read notes into buffer
623 if (read(ph->core->core_fd, buf, size) != size) {
624 print_debug("failed to read notes, core file must have been truncated\n");
625 goto err;
626 }
627
628 p = buf;
629 while (p < buf + size) {
630 ELF_NHDR* notep = (ELF_NHDR*) p;
631 char* descdata = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4);
632 print_debug("Note header with n_type = %d and n_descsz = %u\n",
633 notep->n_type, notep->n_descsz);
634
635 if (notep->n_type == NT_PRSTATUS) {
636 if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true)
637 return false;
638 }
639 p = descdata + ROUNDUP(notep->n_descsz, 4);
640 }
641
642 free(buf);
643 return true;
644
645 err:
646 if (buf) free(buf);
647 return false;
648 }
649
650 // read all segments from core file
651 static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) {
652 int i = 0;
653 ELF_PHDR* phbuf = NULL;
654 ELF_PHDR* core_php = NULL;
655
656 if ((phbuf = read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL)
657 return false;
658
659 /*
660 * Now iterate through the program headers in the core file.
661 * We're interested in two types of Phdrs: PT_NOTE (which
662 * contains a set of saved /proc structures), and PT_LOAD (which
663 * represents a memory mapping from the process's address space).
664 *
665 * Difference b/w Solaris PT_NOTE and BSD PT_NOTE:
666 *
667 * In Solaris there are two PT_NOTE segments the first PT_NOTE (if present)
668 * contains /proc structs in the pre-2.6 unstructured /proc format. the last
669 * PT_NOTE has data in new /proc format.
670 *
671 * In Solaris, there is only one pstatus (process status). pstatus contains
672 * integer register set among other stuff. For each LWP, we have one lwpstatus
673 * entry that has integer regset for that LWP.
674 *
675 * Linux threads are actually 'clone'd processes. To support core analysis
676 * of "multithreaded" process, Linux creates more than one pstatus (called
677 * "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one
678 * "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular
679 * function "elf_core_dump".
680 */
681
682 for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) {
683 switch (core_php->p_type) {
684 case PT_NOTE:
685 if (core_handle_note(ph, core_php) != true) goto err;
686 break;
687
688 case PT_LOAD: {
689 if (core_php->p_filesz != 0) {
690 if (add_map_info(ph, ph->core->core_fd, core_php->p_offset,
691 core_php->p_vaddr, core_php->p_filesz) == NULL) goto err;
692 }
693 break;
694 }
695 }
696
697 core_php++;
698 }
699
700 free(phbuf);
701 return true;
702 err:
703 free(phbuf);
704 return false;
705 }
706
707 // read segments of a shared object
708 static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) {
709 int i = 0;
710 ELF_PHDR* phbuf;
711 ELF_PHDR* lib_php = NULL;
712
713 if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL)
714 return false;
715
716 // we want to process only PT_LOAD segments that are not writable.
717 // i.e., text segments. The read/write/exec (data) segments would
718 // have been already added from core file segments.
719 for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) {
720 if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) {
721 if (add_map_info(ph, lib_fd, lib_php->p_offset, lib_php->p_vaddr + lib_base, lib_php->p_filesz) == NULL)
722 goto err;
723 }
724 lib_php++;
725 }
726
727 free(phbuf);
728 return true;
729 err:
730 free(phbuf);
731 return false;
732 }
733
734 // process segments from interpreter (ld-elf.so.1)
735 static bool read_interp_segments(struct ps_prochandle* ph) {
736 ELF_EHDR interp_ehdr;
737
738 if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) {
739 print_debug("interpreter is not a valid ELF file\n");
740 return false;
741 }
742
743 if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) {
744 print_debug("can't read segments of interpreter\n");
745 return false;
746 }
747
748 return true;
749 }
750
751 // process segments of a a.out
752 static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) {
753 int i = 0;
754 ELF_PHDR* phbuf = NULL;
755 ELF_PHDR* exec_php = NULL;
756
757 if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL)
758 return false;
759
760 for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) {
761 switch (exec_php->p_type) {
762
763 // add mappings for PT_LOAD segments
764 case PT_LOAD: {
765 // add only non-writable segments of non-zero filesz
766 if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) {
767 if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err;
768 }
769 break;
770 }
771
772 // read the interpreter and it's segments
773 case PT_INTERP: {
774 char interp_name[BUF_SIZE];
775
776 pread(ph->core->exec_fd, interp_name, MIN(exec_php->p_filesz, BUF_SIZE), exec_php->p_offset);
777 print_debug("ELF interpreter %s\n", interp_name);
778 // read interpreter segments as well
779 if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) {
780 print_debug("can't open runtime loader\n");
781 goto err;
782 }
783 break;
784 }
785
786 // from PT_DYNAMIC we want to read address of first link_map addr
787 case PT_DYNAMIC: {
788 ph->core->dynamic_addr = exec_php->p_vaddr;
789 print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr);
790 break;
791 }
792
793 } // switch
794 exec_php++;
795 } // for
796
797 free(phbuf);
798 return true;
799 err:
800 free(phbuf);
801 return false;
802 }
803
804
805 #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug, r_map)
806 #define LD_BASE_OFFSET offsetof(struct r_debug, r_ldbase)
807 #define LINK_MAP_ADDR_OFFSET offsetof(struct link_map, l_addr)
808 #define LINK_MAP_NAME_OFFSET offsetof(struct link_map, l_name)
809 #define LINK_MAP_NEXT_OFFSET offsetof(struct link_map, l_next)
810
811 // read shared library info from runtime linker's data structures.
812 // This work is done by librtlb_db in Solaris
813 static bool read_shared_lib_info(struct ps_prochandle* ph) {
814 uintptr_t addr = ph->core->dynamic_addr;
815 uintptr_t debug_base;
816 uintptr_t first_link_map_addr;
817 uintptr_t ld_base_addr;
818 uintptr_t link_map_addr;
819 uintptr_t lib_base_diff;
820 uintptr_t lib_base;
821 uintptr_t lib_name_addr;
822 char lib_name[BUF_SIZE];
823 ELF_DYN dyn;
824 ELF_EHDR elf_ehdr;
825 int lib_fd;
826
827 // _DYNAMIC has information of the form
828 // [tag] [data] [tag] [data] .....
829 // Both tag and data are pointer sized.
830 // We look for dynamic info with DT_DEBUG. This has shared object info.
831 // refer to struct r_debug in link.h
832
833 dyn.d_tag = DT_NULL;
834 while (dyn.d_tag != DT_DEBUG) {
835 if (ps_pread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) {
836 print_debug("can't read debug info from _DYNAMIC\n");
837 return false;
838 }
839 addr += sizeof(ELF_DYN);
840 }
841
842 // we have got Dyn entry with DT_DEBUG
843 debug_base = dyn.d_un.d_ptr;
844 // at debug_base we have struct r_debug. This has first link map in r_map field
845 if (ps_pread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET,
846 &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) {
847 print_debug("can't read first link map address\n");
848 return false;
849 }
850
851 // read ld_base address from struct r_debug
852 // XXX: There is no r_ldbase member on BSD
853 /*
854 if (ps_pread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr,
855 sizeof(uintptr_t)) != PS_OK) {
856 print_debug("can't read ld base address\n");
857 return false;
858 }
859 ph->core->ld_base_addr = ld_base_addr;
860 */
861 ph->core->ld_base_addr = 0;
862
863 print_debug("interpreter base address is 0x%lx\n", ld_base_addr);
864
865 // now read segments from interp (i.e ld-elf.so.1)
866 if (read_interp_segments(ph) != true)
867 return false;
868
869 // after adding interpreter (ld.so) mappings sort again
870 if (sort_map_array(ph) != true)
871 return false;
872
873 print_debug("first link map is at 0x%lx\n", first_link_map_addr);
874
875 link_map_addr = first_link_map_addr;
876 while (link_map_addr != 0) {
877 // read library base address of the .so. Note that even though <sys/link.h> calls
878 // link_map->l_addr as "base address", this is * not * really base virtual
879 // address of the shared object. This is actually the difference b/w the virtual
880 // address mentioned in shared object and the actual virtual base where runtime
881 // linker loaded it. We use "base diff" in read_lib_segments call below.
882
883 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET,
884 &lib_base_diff, sizeof(uintptr_t)) != PS_OK) {
885 print_debug("can't read shared object base address diff\n");
886 return false;
887 }
888
889 // read address of the name
890 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET,
891 &lib_name_addr, sizeof(uintptr_t)) != PS_OK) {
892 print_debug("can't read address of shared object name\n");
893 return false;
894 }
895
896 // read name of the shared object
897 if (read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) {
898 print_debug("can't read shared object name\n");
899 return false;
900 }
901
902 if (lib_name[0] != '\0') {
903 // ignore empty lib names
904 lib_fd = pathmap_open(lib_name);
905
906 if (lib_fd < 0) {
907 print_debug("can't open shared object %s\n", lib_name);
908 // continue with other libraries...
909 } else {
910 if (read_elf_header(lib_fd, &elf_ehdr)) {
911 lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr);
912 print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n",
913 lib_name, lib_base, lib_base_diff);
914 // while adding library mappings we need to use "base difference".
915 if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) {
916 print_debug("can't read shared object's segments\n");
917 close(lib_fd);
918 return false;
919 }
920 add_lib_info_fd(ph, lib_name, lib_fd, lib_base);
921 // Map info is added for the library (lib_name) so
922 // we need to re-sort it before calling the p_pdread.
923 if (sort_map_array(ph) != true)
924 return false;
925 } else {
926 print_debug("can't read ELF header for shared object %s\n", lib_name);
927 close(lib_fd);
928 // continue with other libraries...
929 }
930 }
931 }
932
933 // read next link_map address
934 if (ps_pread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET,
935 &link_map_addr, sizeof(uintptr_t)) != PS_OK) {
936 print_debug("can't read next link in link_map\n");
937 return false;
938 }
939 }
940
941 return true;
942 }
943
944 // the one and only one exposed stuff from this file
945 struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) {
946 ELF_EHDR core_ehdr;
947 ELF_EHDR exec_ehdr;
948
949 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
950 if (ph == NULL) {
951 print_debug("can't allocate ps_prochandle\n");
952 return NULL;
953 }
954
955 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
956 free(ph);
957 print_debug("can't allocate ps_prochandle\n");
958 return NULL;
959 }
960
961 // initialize ph
962 ph->ops = &core_ops;
963 ph->core->core_fd = -1;
964 ph->core->exec_fd = -1;
965 ph->core->interp_fd = -1;
966
967 // open the core file
968 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
969 print_debug("can't open core file\n");
970 goto err;
971 }
972
973 // read core file ELF header
974 if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) {
975 print_debug("core file is not a valid ELF ET_CORE file\n");
976 goto err;
977 }
978
979 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
980 print_debug("can't open executable file\n");
981 goto err;
982 }
983
984 if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) {
985 print_debug("executable file is not a valid ELF ET_EXEC file\n");
986 goto err;
987 }
988
989 // process core file segments
990 if (read_core_segments(ph, &core_ehdr) != true)
991 goto err;
992
993 // process exec file segments
994 if (read_exec_segments(ph, &exec_ehdr) != true)
995 goto err;
996
997 // exec file is also treated like a shared object for symbol search
998 if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd,
999 (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL)
1000 goto err;
1001
1002 // allocate and sort maps into map_array, we need to do this
1003 // here because read_shared_lib_info needs to read from debuggee
1004 // address space
1005 if (sort_map_array(ph) != true)
1006 goto err;
1007
1008 if (read_shared_lib_info(ph) != true)
1009 goto err;
1010
1011 // sort again because we have added more mappings from shared objects
1012 if (sort_map_array(ph) != true)
1013 goto err;
1014
1015 if (init_classsharing_workaround(ph) != true)
1016 goto err;
1017
1018 return ph;
1019
1020 err:
1021 Prelease(ph);
1022 return NULL;
1023 }