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 <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_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) {
68 close(fd);
69 }
70 lib = lib->next;
71 }
72 }
73
74 // clean all map_info stuff
75 static void destroy_map_info(struct ps_prochandle* ph) {
76 map_info* map = ph->core->maps;
77 while (map) {
78 map_info* next = map->next;
79 free(map);
80 map = next;
81 }
82
83 if (ph->core->map_array) {
84 free(ph->core->map_array);
85 }
86
87 // Part of the class sharing workaround
88 map = ph->core->class_share_maps;
89 while (map) {
90 map_info* next = map->next;
91 free(map);
92 map = next;
93 }
94 }
95
96 // ps_prochandle operations
97 static void core_release(struct ps_prochandle* ph) {
98 if (ph->core) {
99 close_files(ph);
100 destroy_map_info(ph);
101 free(ph->core);
102 }
103 }
104
105 static map_info* allocate_init_map(int fd, off_t offset, uintptr_t vaddr, size_t memsz) {
106 map_info* map;
107 if ( (map = (map_info*) calloc(1, sizeof(map_info))) == NULL) {
108 print_debug("can't allocate memory for map_info\n");
109 return NULL;
110 }
111
112 // initialize map
113 map->fd = fd;
114 map->offset = offset;
115 map->vaddr = vaddr;
116 map->memsz = memsz;
117 return map;
118 }
119
120 // add map info with given fd, offset, vaddr and memsz
121 static map_info* add_map_info(struct ps_prochandle* ph, int fd, off_t offset,
122 uintptr_t vaddr, size_t memsz) {
123 map_info* map;
124 if ((map = allocate_init_map(fd, offset, vaddr, memsz)) == NULL) {
125 return NULL;
126 }
127
128 // add this to map list
129 map->next = ph->core->maps;
130 ph->core->maps = map;
131 ph->core->num_maps++;
132
133 return map;
134 }
135
136 // Part of the class sharing workaround
137 static map_info* add_class_share_map_info(struct ps_prochandle* ph, off_t offset,
138 uintptr_t vaddr, size_t memsz) {
139 map_info* map;
140 if ((map = allocate_init_map(ph->core->classes_jsa_fd,
141 offset, vaddr, memsz)) == NULL) {
142 return NULL;
143 }
144
145 map->next = ph->core->class_share_maps;
146 ph->core->class_share_maps = map;
147 return map;
148 }
149
150 // Return the map_info for the given virtual address. We keep a sorted
151 // array of pointers in ph->map_array, so we can binary search.
152 static map_info* core_lookup(struct ps_prochandle *ph, uintptr_t addr) {
153 int mid, lo = 0, hi = ph->core->num_maps - 1;
154 map_info *mp;
155
156 while (hi - lo > 1) {
157 mid = (lo + hi) / 2;
158 if (addr >= ph->core->map_array[mid]->vaddr) {
159 lo = mid;
160 } else {
161 hi = mid;
162 }
163 }
164
165 if (addr < ph->core->map_array[hi]->vaddr) {
166 mp = ph->core->map_array[lo];
167 } else {
168 mp = ph->core->map_array[hi];
169 }
170
171 if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) {
172 return (mp);
173 }
174
175
176 // Part of the class sharing workaround
177 // Unfortunately, we have no way of detecting -Xshare state.
178 // Check out the share maps atlast, if we don't find anywhere.
179 // This is done this way so to avoid reading share pages
180 // ahead of other normal maps. For eg. with -Xshare:off we don't
181 // want to prefer class sharing data to data from core.
182 mp = ph->core->class_share_maps;
183 if (mp) {
184 print_debug("can't locate map_info at 0x%lx, trying class share maps\n", addr);
185 }
186 while (mp) {
187 if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) {
188 print_debug("located map_info at 0x%lx from class share maps\n", addr);
189 return (mp);
190 }
191 mp = mp->next;
192 }
193
194 print_debug("can't locate map_info at 0x%lx\n", addr);
195 return (NULL);
196 }
197
198 //---------------------------------------------------------------
199 // Part of the class sharing workaround:
200 //
201 // With class sharing, pages are mapped from classes.jsa file.
202 // The read-only class sharing pages are mapped as MAP_SHARED,
203 // PROT_READ pages. These pages are not dumped into core dump.
204 // With this workaround, these pages are read from classes.jsa.
205
206 // FIXME: !HACK ALERT!
207 // The format of sharing achive file header is needed to read shared heap
208 // file mappings. For now, I am hard coding portion of FileMapHeader here.
209 // Refer to filemap.hpp.
210
211 // FileMapHeader describes the shared space data in the file to be
212 // mapped. This structure gets written to a file. It is not a class,
213 // so that the compilers don't add any compiler-private data to it.
214
215 #define NUM_SHARED_MAPS 9
216
217 // Refer to FileMapInfo::_current_version in filemap.hpp
218 #define CURRENT_ARCHIVE_VERSION 4
219
220 typedef unsigned char* address;
221 typedef uintptr_t uintx;
222 typedef intptr_t intx;
223
224 struct FileMapHeader {
225 int _magic; // identify file type.
226 int _crc; // header crc checksum.
227 int _version; // (from enum, above.)
228 size_t _alignment; // how shared archive should be aligned
229 int _obj_alignment; // value of ObjectAlignmentInBytes
230 address _narrow_oop_base; // compressed oop encoding base
231 int _narrow_oop_shift; // compressed oop encoding shift
232 bool _compact_strings; // value of CompactStrings
233 uintx _max_heap_size; // java max heap size during dumping
234 int _narrow_oop_mode; // compressed oop encoding mode
235 int _narrow_klass_shift; // save narrow klass base and shift
236 address _narrow_klass_base;
237 char* _misc_data_patching_start;
238 char* _read_only_tables_start;
239 address _cds_i2i_entry_code_buffers;
240 size_t _cds_i2i_entry_code_buffers_size;
241 size_t _core_spaces_size; // number of bytes allocated by the core spaces
242 // (mc, md, ro, rw and od).
243 struct MemRegion {
244 address _start;
245 size_t _word_size;
246 } _g1_reserved; // reserved region at dump time.
247
248 struct space_info {
249 int _crc; // crc checksum of the current space
250 size_t _file_offset; // sizeof(this) rounded to vm page size
251 union {
252 char* _base; // copy-on-write base address
253 intx _offset; // offset from the compressed oop encoding base, only used
254 // by archive heap space
255 } _addr;
256 size_t _used; // for setting space top on read
257 // 4991491 NOTICE These are C++ bool's in filemap.hpp and must match up with
258 // the C type matching the C++ bool type on any given platform.
259 // We assume the corresponding C type is char but licensees
260 // may need to adjust the type of these fields.
261 char _read_only; // read only space?
262 char _allow_exec; // executable code in space?
263 address _oopmap; // bitmap for relocating embedded oops
264 size_t _oopmap_size_in_bits;
265 } _space[NUM_SHARED_MAPS];
266
267 // Ignore the rest of the FileMapHeader. We don't need those fields here.
268 };
269
270 static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) {
271 jboolean i;
272 if (ps_pdread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) {
273 *pvalue = i;
274 return true;
275 } else {
276 return false;
277 }
278 }
279
280 static bool read_pointer(struct ps_prochandle* ph, uintptr_t addr, uintptr_t* pvalue) {
281 uintptr_t uip;
282 if (ps_pdread(ph, (psaddr_t) addr, (char *)&uip, sizeof(uip)) == PS_OK) {
283 *pvalue = uip;
284 return true;
285 } else {
286 return false;
287 }
288 }
289
290 // used to read strings from debuggee
291 static bool read_string(struct ps_prochandle* ph, uintptr_t addr, char* buf, size_t size) {
292 size_t i = 0;
293 char c = ' ';
294
295 while (c != '\0') {
296 if (ps_pdread(ph, (psaddr_t) addr, &c, sizeof(char)) != PS_OK) {
297 return false;
298 }
299 if (i < size - 1) {
300 buf[i] = c;
301 } else {
302 // smaller buffer
303 return false;
304 }
305 i++; addr++;
306 }
307
308 buf[i] = '\0';
309 return true;
310 }
311
312 #define USE_SHARED_SPACES_SYM "UseSharedSpaces"
313 // mangled name of Arguments::SharedArchivePath
314 #define SHARED_ARCHIVE_PATH_SYM "_ZN9Arguments17SharedArchivePathE"
315 #define LIBJVM_NAME "/libjvm.so"
316
317 static bool init_classsharing_workaround(struct ps_prochandle* ph) {
318 lib_info* lib = ph->libs;
319 while (lib != NULL) {
320 // we are iterating over shared objects from the core dump. look for
321 // libjvm.so.
322 const char *jvm_name = 0;
323 if ((jvm_name = strstr(lib->name, LIBJVM_NAME)) != 0) {
324 char classes_jsa[PATH_MAX];
325 struct FileMapHeader header;
326 int fd = -1;
327 int m = 0;
328 size_t n = 0;
329 uintptr_t base = 0, useSharedSpacesAddr = 0;
330 uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0;
331 jboolean useSharedSpaces = 0;
332 map_info* mi = 0;
333
334 memset(classes_jsa, 0, sizeof(classes_jsa));
335 jvm_name = lib->name;
336 useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM);
337 if (useSharedSpacesAddr == 0) {
338 print_debug("can't lookup 'UseSharedSpaces' flag\n");
339 return false;
340 }
341
342 // Hotspot vm types are not exported to build this library. So
343 // using equivalent type jboolean to read the value of
344 // UseSharedSpaces which is same as hotspot type "bool".
345 if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) {
346 print_debug("can't read the value of 'UseSharedSpaces' flag\n");
347 return false;
348 }
349
350 if ((int)useSharedSpaces == 0) {
351 print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n");
352 return true;
353 }
354
355 sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM);
356 if (sharedArchivePathAddrAddr == 0) {
357 print_debug("can't lookup shared archive path symbol\n");
358 return false;
359 }
360
361 if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) {
362 print_debug("can't read shared archive path pointer\n");
363 return false;
364 }
365
366 if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) {
367 print_debug("can't read shared archive path value\n");
368 return false;
369 }
370
371 print_debug("looking for %s\n", classes_jsa);
372 // open the class sharing archive file
373 fd = pathmap_open(classes_jsa);
374 if (fd < 0) {
375 print_debug("can't open %s!\n", classes_jsa);
376 ph->core->classes_jsa_fd = -1;
377 return false;
378 } else {
379 print_debug("opened %s\n", classes_jsa);
380 }
381
382 // read FileMapHeader from the file
383 memset(&header, 0, sizeof(struct FileMapHeader));
384 if ((n = read(fd, &header, sizeof(struct FileMapHeader)))
385 != sizeof(struct FileMapHeader)) {
386 print_debug("can't read shared archive file map header from %s\n", classes_jsa);
387 close(fd);
388 return false;
389 }
390
391 // check file magic
392 if (header._magic != 0xf00baba2) {
393 print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n",
394 classes_jsa, header._magic);
395 close(fd);
396 return false;
397 }
398
399 // check version
400 if (header._version != CURRENT_ARCHIVE_VERSION) {
401 print_debug("%s has wrong shared archive file version %d, expecting %d\n",
402 classes_jsa, header._version, CURRENT_ARCHIVE_VERSION);
403 close(fd);
404 return false;
405 }
406
407 ph->core->classes_jsa_fd = fd;
408 // add read-only maps from classes.jsa to the list of maps
409 for (m = 0; m < NUM_SHARED_MAPS; m++) {
410 if (header._space[m]._read_only) {
411 base = (uintptr_t) header._space[m]._addr._base;
412 // no need to worry about the fractional pages at-the-end.
413 // possible fractional pages are handled by core_read_data.
414 add_class_share_map_info(ph, (off_t) header._space[m]._file_offset,
415 base, (size_t) header._space[m]._used);
416 print_debug("added a share archive map at 0x%lx\n", base);
417 }
418 }
419 return true;
420 }
421 lib = lib->next;
422 }
423 return true;
424 }
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
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 = %zu\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 user_regs_struct* regs) {
550 // for core we have cached the lwp regs from NOTE section
551 thread_info* thr = ph->threads;
552 while (thr) {
553 if (thr->lwp_id == lwp_id) {
554 memcpy(regs, &thr->regs, sizeof(struct user_regs_struct));
555 return true;
556 }
557 thr = thr->next;
558 }
559 return false;
560 }
561
562 static ps_prochandle_ops core_ops = {
563 .release= core_release,
564 .p_pread= core_read_data,
565 .p_pwrite= core_write_data,
566 .get_lwp_regs= core_get_lwp_regs
567 };
568
569 // read regs and create thread from NT_PRSTATUS entries from core file
570 static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) {
571 // we have to read prstatus_t from buf
572 // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t");
573 prstatus_t* prstat = (prstatus_t*) buf;
574 thread_info* newthr;
575 print_debug("got integer regset for lwp %d\n", prstat->pr_pid);
576 // we set pthread_t to -1 for core dump
577 if((newthr = add_thread_info(ph, (pthread_t) -1, prstat->pr_pid)) == NULL)
578 return false;
579
580 // copy regs
581 memcpy(&newthr->regs, prstat->pr_reg, sizeof(struct user_regs_struct));
582
583 if (is_debug()) {
584 print_debug("integer regset\n");
585 #ifdef i386
586 // print the regset
587 print_debug("\teax = 0x%x\n", newthr->regs.eax);
588 print_debug("\tebx = 0x%x\n", newthr->regs.ebx);
589 print_debug("\tecx = 0x%x\n", newthr->regs.ecx);
590 print_debug("\tedx = 0x%x\n", newthr->regs.edx);
591 print_debug("\tesp = 0x%x\n", newthr->regs.esp);
592 print_debug("\tebp = 0x%x\n", newthr->regs.ebp);
593 print_debug("\tesi = 0x%x\n", newthr->regs.esi);
594 print_debug("\tedi = 0x%x\n", newthr->regs.edi);
595 print_debug("\teip = 0x%x\n", newthr->regs.eip);
596 #endif
597
598 #if defined(amd64) || defined(x86_64)
599 // print the regset
600 print_debug("\tr15 = 0x%lx\n", newthr->regs.r15);
601 print_debug("\tr14 = 0x%lx\n", newthr->regs.r14);
602 print_debug("\tr13 = 0x%lx\n", newthr->regs.r13);
603 print_debug("\tr12 = 0x%lx\n", newthr->regs.r12);
604 print_debug("\trbp = 0x%lx\n", newthr->regs.rbp);
605 print_debug("\trbx = 0x%lx\n", newthr->regs.rbx);
606 print_debug("\tr11 = 0x%lx\n", newthr->regs.r11);
607 print_debug("\tr10 = 0x%lx\n", newthr->regs.r10);
608 print_debug("\tr9 = 0x%lx\n", newthr->regs.r9);
609 print_debug("\tr8 = 0x%lx\n", newthr->regs.r8);
610 print_debug("\trax = 0x%lx\n", newthr->regs.rax);
611 print_debug("\trcx = 0x%lx\n", newthr->regs.rcx);
612 print_debug("\trdx = 0x%lx\n", newthr->regs.rdx);
613 print_debug("\trsi = 0x%lx\n", newthr->regs.rsi);
614 print_debug("\trdi = 0x%lx\n", newthr->regs.rdi);
615 print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax);
616 print_debug("\trip = 0x%lx\n", newthr->regs.rip);
617 print_debug("\tcs = 0x%lx\n", newthr->regs.cs);
618 print_debug("\teflags = 0x%lx\n", newthr->regs.eflags);
619 print_debug("\trsp = 0x%lx\n", newthr->regs.rsp);
620 print_debug("\tss = 0x%lx\n", newthr->regs.ss);
621 print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base);
622 print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base);
623 print_debug("\tds = 0x%lx\n", newthr->regs.ds);
624 print_debug("\tes = 0x%lx\n", newthr->regs.es);
625 print_debug("\tfs = 0x%lx\n", newthr->regs.fs);
626 print_debug("\tgs = 0x%lx\n", newthr->regs.gs);
627 #endif
628 }
629
630 return true;
631 }
632
633 #define ROUNDUP(x, y) ((((x)+((y)-1))/(y))*(y))
634
635 // read NT_PRSTATUS entries from core NOTE segment
636 static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) {
637 char* buf = NULL;
638 char* p = NULL;
639 size_t size = note_phdr->p_filesz;
640
641 // we are interested in just prstatus entries. we will ignore the rest.
642 // Advance the seek pointer to the start of the PT_NOTE data
643 if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) {
644 print_debug("failed to lseek to PT_NOTE data\n");
645 return false;
646 }
647
648 // Now process the PT_NOTE structures. Each one is preceded by
649 // an Elf{32/64}_Nhdr structure describing its type and size.
650 if ( (buf = (char*) malloc(size)) == NULL) {
651 print_debug("can't allocate memory for reading core notes\n");
652 goto err;
653 }
654
655 // read notes into buffer
656 if (read(ph->core->core_fd, buf, size) != size) {
657 print_debug("failed to read notes, core file must have been truncated\n");
658 goto err;
659 }
660
661 p = buf;
662 while (p < buf + size) {
663 ELF_NHDR* notep = (ELF_NHDR*) p;
664 char* descdata = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4);
665 print_debug("Note header with n_type = %d and n_descsz = %u\n",
666 notep->n_type, notep->n_descsz);
667
668 if (notep->n_type == NT_PRSTATUS) {
669 if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) {
670 return false;
671 }
672 } else if (notep->n_type == NT_AUXV) {
673 // Get first segment from entry point
674 ELF_AUXV *auxv = (ELF_AUXV *)descdata;
675 while (auxv->a_type != AT_NULL) {
676 if (auxv->a_type == AT_ENTRY) {
677 // Set entry point address to address of dynamic section.
678 // We will adjust it in read_exec_segments().
679 ph->core->dynamic_addr = auxv->a_un.a_val;
680 break;
681 }
682 auxv++;
683 }
684 }
685 p = descdata + ROUNDUP(notep->n_descsz, 4);
686 }
687
688 free(buf);
689 return true;
690
691 err:
692 if (buf) free(buf);
693 return false;
694 }
695
696 // read all segments from core file
697 static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) {
698 int i = 0;
699 ELF_PHDR* phbuf = NULL;
700 ELF_PHDR* core_php = NULL;
701
702 if ((phbuf = read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL)
703 return false;
704
705 /*
706 * Now iterate through the program headers in the core file.
707 * We're interested in two types of Phdrs: PT_NOTE (which
708 * contains a set of saved /proc structures), and PT_LOAD (which
709 * represents a memory mapping from the process's address space).
710 *
711 * Difference b/w Solaris PT_NOTE and Linux/BSD PT_NOTE:
712 *
713 * In Solaris there are two PT_NOTE segments the first PT_NOTE (if present)
714 * contains /proc structs in the pre-2.6 unstructured /proc format. the last
715 * PT_NOTE has data in new /proc format.
716 *
717 * In Solaris, there is only one pstatus (process status). pstatus contains
718 * integer register set among other stuff. For each LWP, we have one lwpstatus
719 * entry that has integer regset for that LWP.
720 *
721 * Linux threads are actually 'clone'd processes. To support core analysis
722 * of "multithreaded" process, Linux creates more than one pstatus (called
723 * "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one
724 * "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular
725 * function "elf_core_dump".
726 */
727
728 for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) {
729 switch (core_php->p_type) {
730 case PT_NOTE:
731 if (core_handle_note(ph, core_php) != true) {
732 goto err;
733 }
734 break;
735
736 case PT_LOAD: {
737 if (core_php->p_filesz != 0) {
738 if (add_map_info(ph, ph->core->core_fd, core_php->p_offset,
739 core_php->p_vaddr, core_php->p_filesz) == NULL) goto err;
740 }
741 break;
742 }
743 }
744
745 core_php++;
746 }
747
748 free(phbuf);
749 return true;
750 err:
751 free(phbuf);
752 return false;
753 }
754
755 // read segments of a shared object
756 static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) {
757 int i = 0;
758 ELF_PHDR* phbuf;
759 ELF_PHDR* lib_php = NULL;
760
761 int page_size = sysconf(_SC_PAGE_SIZE);
762
763 if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL) {
764 return false;
765 }
766
767 // we want to process only PT_LOAD segments that are not writable.
768 // i.e., text segments. The read/write/exec (data) segments would
769 // have been already added from core file segments.
770 for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) {
771 if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) {
772
773 uintptr_t target_vaddr = lib_php->p_vaddr + lib_base;
774 map_info *existing_map = core_lookup(ph, target_vaddr);
775
776 if (existing_map == NULL){
777 if (add_map_info(ph, lib_fd, lib_php->p_offset,
778 target_vaddr, lib_php->p_memsz) == NULL) {
779 goto err;
780 }
781 } else {
782 // Coredump stores value of p_memsz elf field
783 // rounded up to page boundary.
784
785 if ((existing_map->memsz != page_size) &&
786 (existing_map->fd != lib_fd) &&
787 (ROUNDUP(existing_map->memsz, page_size) != ROUNDUP(lib_php->p_memsz, page_size))) {
788
789 print_debug("address conflict @ 0x%lx (existing map size = %ld, size = %ld, flags = %d)\n",
790 target_vaddr, existing_map->memsz, lib_php->p_memsz, lib_php->p_flags);
791 goto err;
792 }
793
794 /* replace PT_LOAD segment with library segment */
795 print_debug("overwrote with new address mapping (memsz %ld -> %ld)\n",
796 existing_map->memsz, ROUNDUP(lib_php->p_memsz, page_size));
797
798 existing_map->fd = lib_fd;
799 existing_map->offset = lib_php->p_offset;
800 existing_map->memsz = ROUNDUP(lib_php->p_memsz, page_size);
801 }
802 }
803
804 lib_php++;
805 }
806
807 free(phbuf);
808 return true;
809 err:
810 free(phbuf);
811 return false;
812 }
813
814 // process segments from interpreter (ld.so or ld-linux.so)
815 static bool read_interp_segments(struct ps_prochandle* ph) {
816 ELF_EHDR interp_ehdr;
817
818 if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) {
819 print_debug("interpreter is not a valid ELF file\n");
820 return false;
821 }
822
823 if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) {
824 print_debug("can't read segments of interpreter\n");
825 return false;
826 }
827
828 return true;
829 }
830
831 // process segments of a a.out
832 static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) {
833 int i = 0;
834 ELF_PHDR* phbuf = NULL;
835 ELF_PHDR* exec_php = NULL;
836
837 if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL) {
838 return false;
839 }
840
841 for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) {
842 switch (exec_php->p_type) {
843
844 // add mappings for PT_LOAD segments
845 case PT_LOAD: {
846 // add only non-writable segments of non-zero filesz
847 if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) {
848 if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err;
849 }
850 break;
851 }
852
853 // read the interpreter and it's segments
854 case PT_INTERP: {
855 char interp_name[BUF_SIZE + 1];
856
857 // BUF_SIZE is PATH_MAX + NAME_MAX + 1.
858 if (exec_php->p_filesz > BUF_SIZE) {
859 goto err;
860 }
861 pread(ph->core->exec_fd, interp_name, exec_php->p_filesz, exec_php->p_offset);
862 interp_name[exec_php->p_filesz] = '\0';
863 print_debug("ELF interpreter %s\n", interp_name);
864 // read interpreter segments as well
865 if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) {
866 print_debug("can't open runtime loader\n");
867 goto err;
868 }
869 break;
870 }
871
872 // from PT_DYNAMIC we want to read address of first link_map addr
873 case PT_DYNAMIC: {
874 if (exec_ehdr->e_type == ET_EXEC) {
875 ph->core->dynamic_addr = exec_php->p_vaddr;
876 } else { // ET_DYN
877 // dynamic_addr has entry point of executable.
878 // Thus we should substract it.
879 ph->core->dynamic_addr += exec_php->p_vaddr - exec_ehdr->e_entry;
880 }
881 print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr);
882 break;
883 }
884
885 } // switch
886 exec_php++;
887 } // for
888
889 free(phbuf);
890 return true;
891 err:
892 free(phbuf);
893 return false;
894 }
895
896
897 #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug, r_map)
898 #define LD_BASE_OFFSET offsetof(struct r_debug, r_ldbase)
899 #define LINK_MAP_ADDR_OFFSET offsetof(struct link_map, l_addr)
900 #define LINK_MAP_NAME_OFFSET offsetof(struct link_map, l_name)
901 #define LINK_MAP_NEXT_OFFSET offsetof(struct link_map, l_next)
902
903 // read shared library info from runtime linker's data structures.
904 // This work is done by librtlb_db in Solaris
905 static bool read_shared_lib_info(struct ps_prochandle* ph) {
906 uintptr_t addr = ph->core->dynamic_addr;
907 uintptr_t debug_base;
908 uintptr_t first_link_map_addr;
909 uintptr_t ld_base_addr;
910 uintptr_t link_map_addr;
911 uintptr_t lib_base_diff;
912 uintptr_t lib_base;
913 uintptr_t lib_name_addr;
914 char lib_name[BUF_SIZE];
915 ELF_DYN dyn;
916 ELF_EHDR elf_ehdr;
917 int lib_fd;
918
919 // _DYNAMIC has information of the form
920 // [tag] [data] [tag] [data] .....
921 // Both tag and data are pointer sized.
922 // We look for dynamic info with DT_DEBUG. This has shared object info.
923 // refer to struct r_debug in link.h
924
925 dyn.d_tag = DT_NULL;
926 while (dyn.d_tag != DT_DEBUG) {
927 if (ps_pdread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) {
928 print_debug("can't read debug info from _DYNAMIC\n");
929 return false;
930 }
931 addr += sizeof(ELF_DYN);
932 }
933
934 // we have got Dyn entry with DT_DEBUG
935 debug_base = dyn.d_un.d_ptr;
936 // at debug_base we have struct r_debug. This has first link map in r_map field
937 if (ps_pdread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET,
938 &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) {
939 print_debug("can't read first link map address\n");
940 return false;
941 }
942
943 // read ld_base address from struct r_debug
944 if (ps_pdread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr,
945 sizeof(uintptr_t)) != PS_OK) {
946 print_debug("can't read ld base address\n");
947 return false;
948 }
949 ph->core->ld_base_addr = ld_base_addr;
950
951 print_debug("interpreter base address is 0x%lx\n", ld_base_addr);
952
953 // now read segments from interp (i.e ld.so or ld-linux.so or ld-elf.so)
954 if (read_interp_segments(ph) != true) {
955 return false;
956 }
957
958 // after adding interpreter (ld.so) mappings sort again
959 if (sort_map_array(ph) != true) {
960 return false;
961 }
962
963 print_debug("first link map is at 0x%lx\n", first_link_map_addr);
964
965 link_map_addr = first_link_map_addr;
966 while (link_map_addr != 0) {
967 // read library base address of the .so. Note that even though <sys/link.h> calls
968 // link_map->l_addr as "base address", this is * not * really base virtual
969 // address of the shared object. This is actually the difference b/w the virtual
970 // address mentioned in shared object and the actual virtual base where runtime
971 // linker loaded it. We use "base diff" in read_lib_segments call below.
972
973 if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET,
974 &lib_base_diff, sizeof(uintptr_t)) != PS_OK) {
975 print_debug("can't read shared object base address diff\n");
976 return false;
977 }
978
979 // read address of the name
980 if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET,
981 &lib_name_addr, sizeof(uintptr_t)) != PS_OK) {
982 print_debug("can't read address of shared object name\n");
983 return false;
984 }
985
986 // read name of the shared object
987 lib_name[0] = '\0';
988 if (lib_name_addr != 0 &&
989 read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) {
990 print_debug("can't read shared object name\n");
991 // don't let failure to read the name stop opening the file. If something is really wrong
992 // it will fail later.
993 }
994
995 if (lib_name[0] != '\0') {
996 // ignore empty lib names
997 lib_fd = pathmap_open(lib_name);
998
999 if (lib_fd < 0) {
1000 print_debug("can't open shared object %s\n", lib_name);
1001 // continue with other libraries...
1002 } else {
1003 if (read_elf_header(lib_fd, &elf_ehdr)) {
1004 lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr);
1005 print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n",
1006 lib_name, lib_base, lib_base_diff);
1007 // while adding library mappings we need to use "base difference".
1008 if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) {
1009 print_debug("can't read shared object's segments\n");
1010 close(lib_fd);
1011 return false;
1012 }
1013 add_lib_info_fd(ph, lib_name, lib_fd, lib_base);
1014 // Map info is added for the library (lib_name) so
1015 // we need to re-sort it before calling the p_pdread.
1016 if (sort_map_array(ph) != true)
1017 return false;
1018 } else {
1019 print_debug("can't read ELF header for shared object %s\n", lib_name);
1020 close(lib_fd);
1021 // continue with other libraries...
1022 }
1023 }
1024 }
1025
1026 // read next link_map address
1027 if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET,
1028 &link_map_addr, sizeof(uintptr_t)) != PS_OK) {
1029 print_debug("can't read next link in link_map\n");
1030 return false;
1031 }
1032 }
1033
1034 return true;
1035 }
1036
1037 // the one and only one exposed stuff from this file
1038 JNIEXPORT struct ps_prochandle* JNICALL
1039 Pgrab_core(const char* exec_file, const char* core_file) {
1040 ELF_EHDR core_ehdr;
1041 ELF_EHDR exec_ehdr;
1042 ELF_EHDR lib_ehdr;
1043
1044 struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle));
1045 if (ph == NULL) {
1046 print_debug("can't allocate ps_prochandle\n");
1047 return NULL;
1048 }
1049
1050 if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) {
1051 free(ph);
1052 print_debug("can't allocate ps_prochandle\n");
1053 return NULL;
1054 }
1055
1056 // initialize ph
1057 ph->ops = &core_ops;
1058 ph->core->core_fd = -1;
1059 ph->core->exec_fd = -1;
1060 ph->core->interp_fd = -1;
1061
1062 // open the core file
1063 if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) {
1064 print_debug("can't open core file\n");
1065 goto err;
1066 }
1067
1068 // read core file ELF header
1069 if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) {
1070 print_debug("core file is not a valid ELF ET_CORE file\n");
1071 goto err;
1072 }
1073
1074 if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) {
1075 print_debug("can't open executable file\n");
1076 goto err;
1077 }
1078
1079 if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true ||
1080 ((exec_ehdr.e_type != ET_EXEC) && (exec_ehdr.e_type != ET_DYN))) {
1081 print_debug("executable file is not a valid ELF file\n");
1082 goto err;
1083 }
1084
1085 // process core file segments
1086 if (read_core_segments(ph, &core_ehdr) != true) {
1087 goto err;
1088 }
1089
1090 // process exec file segments
1091 if (read_exec_segments(ph, &exec_ehdr) != true) {
1092 goto err;
1093 }
1094
1095 // exec file is also treated like a shared object for symbol search
1096 if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd,
1097 (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL) {
1098 goto err;
1099 }
1100
1101 // allocate and sort maps into map_array, we need to do this
1102 // here because read_shared_lib_info needs to read from debuggee
1103 // address space
1104 if (sort_map_array(ph) != true) {
1105 goto err;
1106 }
1107
1108 if (read_shared_lib_info(ph) != true) {
1109 goto err;
1110 }
1111
1112 // sort again because we have added more mappings from shared objects
1113 if (sort_map_array(ph) != true) {
1114 goto err;
1115 }
1116
1117 if (init_classsharing_workaround(ph) != true) {
1118 goto err;
1119 }
1120
1121 return ph;
1122
1123 err:
1124 Prelease(ph);
1125 return NULL;
1126 }