1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 1983, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <stdarg.h>
28 #include <unistd.h>
29 #include <fcntl.h>
30 #include <errno.h>
31 #include <string.h>
32 #include <deflt.h>
33 #include <time.h>
34 #include <syslog.h>
35 #include <stropts.h>
36 #include <pthread.h>
37 #include <limits.h>
38 #include <atomic.h>
39 #include <libnvpair.h>
40 #include <libintl.h>
41 #include <sys/mem.h>
42 #include <sys/statvfs.h>
43 #include <sys/dumphdr.h>
44 #include <sys/dumpadm.h>
45 #include <sys/compress.h>
46 #include <sys/panic.h>
47 #include <sys/sysmacros.h>
48 #include <sys/stat.h>
49 #include <sys/resource.h>
50 #include <bzip2/bzlib.h>
51 #include <sys/fm/util.h>
52 #include <fm/libfmevent.h>
53 #include <sys/int_fmtio.h>
54
55
56 /* fread/fwrite buffer size */
57 #define FBUFSIZE (1ULL << 20)
58
59 /* minimum size for output buffering */
60 #define MINCOREBLKSIZE (1ULL << 17)
61
62 /* create this file if metrics collection is enabled in the kernel */
63 #define METRICSFILE "METRICS.csv"
64
65 static char progname[9] = "savecore";
66 static char *savedir; /* savecore directory */
67 static char *dumpfile; /* source of raw crash dump */
68 static long bounds = -1; /* numeric suffix */
69 static long pagesize; /* dump pagesize */
70 static int dumpfd = -1; /* dumpfile descriptor */
71 static dumphdr_t corehdr, dumphdr; /* initial and terminal dumphdrs */
72 static boolean_t dump_incomplete; /* dumphdr indicates incomplete */
73 static boolean_t fm_panic; /* dump is the result of fm_panic */
74 static offset_t endoff; /* offset of end-of-dump header */
75 static int verbose; /* chatty mode */
76 static int disregard_valid_flag; /* disregard valid flag */
77 static int livedump; /* dump the current running system */
78 static int interactive; /* user invoked; no syslog */
79 static int csave; /* save dump compressed */
80 static int filemode; /* processing file, not dump device */
81 static int percent_done; /* progress indicator */
82 static hrtime_t startts; /* timestamp at start */
83 static volatile uint64_t saved; /* count of pages written */
84 static volatile uint64_t zpages; /* count of zero pages not written */
85 static dumpdatahdr_t datahdr; /* compression info */
86 static long coreblksize; /* preferred write size (st_blksize) */
87 static int cflag; /* run as savecore -c */
88 static int mflag; /* run as savecore -m */
89
90 /*
91 * Payload information for the events we raise. These are used
92 * in raise_event to determine what payload to include.
93 */
94 #define SC_PAYLOAD_SAVEDIR 0x0001 /* Include savedir in event */
95 #define SC_PAYLOAD_INSTANCE 0x0002 /* Include bounds instance number */
96 #define SC_PAYLOAD_IMAGEUUID 0x0004 /* Include dump OS instance uuid */
97 #define SC_PAYLOAD_CRASHTIME 0x0008 /* Include epoch crashtime */
98 #define SC_PAYLOAD_PANICSTR 0x0010 /* Include panic string */
99 #define SC_PAYLOAD_PANICSTACK 0x0020 /* Include panic string */
100 #define SC_PAYLOAD_FAILREASON 0x0040 /* Include failure reason */
101 #define SC_PAYLOAD_DUMPCOMPLETE 0x0080 /* Include completeness indicator */
102 #define SC_PAYLOAD_ISCOMPRESSED 0x0100 /* Dump is in vmdump.N form */
103 #define SC_PAYLOAD_DUMPADM_EN 0x0200 /* Is dumpadm enabled or not? */
104 #define SC_PAYLOAD_FM_PANIC 0x0400 /* Panic initiated by FMA */
105 #define SC_PAYLOAD_JUSTCHECKING 0x0800 /* Run with -c flag? */
106
107 enum sc_event_type {
108 SC_EVENT_DUMP_PENDING,
109 SC_EVENT_SAVECORE_FAILURE,
110 SC_EVENT_DUMP_AVAILABLE
111 };
112
113 /*
114 * Common payload
115 */
116 #define _SC_PAYLOAD_CMN \
117 SC_PAYLOAD_IMAGEUUID | \
118 SC_PAYLOAD_CRASHTIME | \
119 SC_PAYLOAD_PANICSTR | \
120 SC_PAYLOAD_PANICSTACK | \
121 SC_PAYLOAD_DUMPCOMPLETE | \
122 SC_PAYLOAD_FM_PANIC | \
123 SC_PAYLOAD_SAVEDIR
124
125 static const struct {
126 const char *sce_subclass;
127 uint32_t sce_payload;
128 } sc_event[] = {
129 /*
130 * SC_EVENT_DUMP_PENDING
131 */
132 {
133 "dump_pending_on_device",
134 _SC_PAYLOAD_CMN | SC_PAYLOAD_DUMPADM_EN |
135 SC_PAYLOAD_JUSTCHECKING
136 },
137
138 /*
139 * SC_EVENT_SAVECORE_FAILURE
140 */
141 {
142 "savecore_failure",
143 _SC_PAYLOAD_CMN | SC_PAYLOAD_INSTANCE | SC_PAYLOAD_FAILREASON
144 },
145
146 /*
147 * SC_EVENT_DUMP_AVAILABLE
148 */
149 {
150 "dump_available",
151 _SC_PAYLOAD_CMN | SC_PAYLOAD_INSTANCE | SC_PAYLOAD_ISCOMPRESSED
152 },
153 };
154
155 static void raise_event(enum sc_event_type, char *);
156
157 static void
158 usage(void)
159 {
160 (void) fprintf(stderr,
161 "usage: %s [-Lvd] [-f dumpfile] [dirname]\n", progname);
162 exit(1);
163 }
164
165 #define SC_SL_NONE 0x0001 /* no syslog */
166 #define SC_SL_ERR 0x0002 /* syslog if !interactive, LOG_ERR */
167 #define SC_SL_WARN 0x0004 /* syslog if !interactive, LOG_WARNING */
168 #define SC_IF_VERBOSE 0x0008 /* message only if -v */
169 #define SC_IF_ISATTY 0x0010 /* message only if interactive */
170 #define SC_EXIT_OK 0x0020 /* exit(0) */
171 #define SC_EXIT_ERR 0x0040 /* exit(1) */
172 #define SC_EXIT_PEND 0x0080 /* exit(2) */
173 #define SC_EXIT_FM 0x0100 /* exit(3) */
174
175 #define _SC_ALLEXIT (SC_EXIT_OK | SC_EXIT_ERR | SC_EXIT_PEND | SC_EXIT_FM)
176
177 static void
178 logprint(uint32_t flags, char *message, ...)
179 {
180 va_list args;
181 char buf[1024];
182 int do_always = ((flags & (SC_IF_VERBOSE | SC_IF_ISATTY)) == 0);
183 int do_ifverb = (flags & SC_IF_VERBOSE) && verbose;
184 int do_ifisatty = (flags & SC_IF_ISATTY) && interactive;
185 int code;
186 static int logprint_raised = 0;
187
188 if (do_always || do_ifverb || do_ifisatty) {
189 va_start(args, message);
190 /*LINTED: E_SEC_PRINTF_VAR_FMT*/
191 (void) vsnprintf(buf, sizeof (buf), message, args);
192 (void) fprintf(stderr, "%s: %s\n", progname, buf);
193 if (!interactive) {
194 switch (flags & (SC_SL_NONE | SC_SL_ERR | SC_SL_WARN)) {
195 case SC_SL_ERR:
196 /*LINTED: E_SEC_PRINTF_VAR_FMT*/
197 syslog(LOG_ERR, buf);
198 break;
199
200 case SC_SL_WARN:
201 /*LINTED: E_SEC_PRINTF_VAR_FMT*/
202 syslog(LOG_WARNING, buf);
203 break;
204
205 default:
206 break;
207 }
208 }
209 va_end(args);
210 }
211
212 switch (flags & _SC_ALLEXIT) {
213 case 0:
214 return;
215
216 case SC_EXIT_OK:
217 code = 0;
218 break;
219
220 case SC_EXIT_PEND:
221 code = 2;
222 break;
223
224 case SC_EXIT_FM:
225 code = 3;
226 break;
227
228 case SC_EXIT_ERR:
229 default:
230 /*
231 * Raise an ireport saying why we are exiting. Do not
232 * raise if run as savecore -m. If something in the
233 * raise_event codepath calls logprint avoid recursion.
234 */
235 if (!mflag && logprint_raised++ == 0)
236 raise_event(SC_EVENT_SAVECORE_FAILURE, buf);
237 code = 1;
238 break;
239 }
240
241 exit(code);
242 }
243
244 /*
245 * System call / libc wrappers that exit on error.
246 */
247 static int
248 Open(const char *name, int oflags, mode_t mode)
249 {
250 int fd;
251
252 if ((fd = open64(name, oflags, mode)) == -1)
253 logprint(SC_SL_ERR | SC_EXIT_ERR, "open(\"%s\"): %s",
254 name, strerror(errno));
255 return (fd);
256 }
257
258 static void
259 Fread(void *buf, size_t size, FILE *f)
260 {
261 if (fread(buf, size, 1, f) != 1)
262 logprint(SC_SL_ERR | SC_EXIT_ERR, "fread: ferror %d feof %d",
263 ferror(f), feof(f));
264 }
265
266 static void
267 Fwrite(void *buf, size_t size, FILE *f)
268 {
269 if (fwrite(buf, size, 1, f) != 1)
270 logprint(SC_SL_ERR | SC_EXIT_ERR, "fwrite: %s",
271 strerror(errno));
272 }
273
274 static void
275 Fseek(offset_t off, FILE *f)
276 {
277 if (fseeko64(f, off, SEEK_SET) != 0)
278 logprint(SC_SL_ERR | SC_EXIT_ERR, "fseeko64: %s",
279 strerror(errno));
280 }
281
282 typedef struct stat64 Stat_t;
283
284 static void
285 Fstat(int fd, Stat_t *sb, const char *fname)
286 {
287 if (fstat64(fd, sb) != 0)
288 logprint(SC_SL_ERR | SC_EXIT_ERR, "fstat(\"%s\"): %s", fname,
289 strerror(errno));
290 }
291
292 static void
293 Stat(const char *fname, Stat_t *sb)
294 {
295 if (stat64(fname, sb) != 0)
296 logprint(SC_SL_ERR | SC_EXIT_ERR, "stat(\"%s\"): %s", fname,
297 strerror(errno));
298 }
299
300 static void
301 Pread(int fd, void *buf, size_t size, offset_t off)
302 {
303 ssize_t sz = pread64(fd, buf, size, off);
304
305 if (sz < 0)
306 logprint(SC_SL_ERR | SC_EXIT_ERR,
307 "pread: %s", strerror(errno));
308 else if (sz != size)
309 logprint(SC_SL_ERR | SC_EXIT_ERR,
310 "pread: size %ld != %ld", sz, size);
311 }
312
313 static void
314 Pwrite(int fd, void *buf, size_t size, off64_t off)
315 {
316 if (pwrite64(fd, buf, size, off) != size)
317 logprint(SC_SL_ERR | SC_EXIT_ERR, "pwrite: %s",
318 strerror(errno));
319 }
320
321 static void *
322 Zalloc(size_t size)
323 {
324 void *buf;
325
326 if ((buf = calloc(size, 1)) == NULL)
327 logprint(SC_SL_ERR | SC_EXIT_ERR, "calloc: %s",
328 strerror(errno));
329 return (buf);
330 }
331
332 static long
333 read_number_from_file(const char *filename, long default_value)
334 {
335 long file_value = -1;
336 FILE *fp;
337
338 if ((fp = fopen(filename, "r")) != NULL) {
339 (void) fscanf(fp, "%ld", &file_value);
340 (void) fclose(fp);
341 }
342 return (file_value < 0 ? default_value : file_value);
343 }
344
345 static void
346 read_dumphdr(void)
347 {
348 if (filemode)
349 dumpfd = Open(dumpfile, O_RDONLY, 0644);
350 else
351 dumpfd = Open(dumpfile, O_RDWR | O_DSYNC, 0644);
352 endoff = llseek(dumpfd, -DUMP_OFFSET, SEEK_END) & -DUMP_OFFSET;
353 Pread(dumpfd, &dumphdr, sizeof (dumphdr), endoff);
354 Pread(dumpfd, &datahdr, sizeof (datahdr), endoff + sizeof (dumphdr));
355
356 pagesize = dumphdr.dump_pagesize;
357
358 if (dumphdr.dump_magic != DUMP_MAGIC)
359 logprint(SC_SL_NONE | SC_EXIT_OK, "bad magic number %x",
360 dumphdr.dump_magic);
361
362 if ((dumphdr.dump_flags & DF_VALID) == 0 && !disregard_valid_flag)
363 logprint(SC_SL_NONE | SC_IF_VERBOSE | SC_EXIT_OK,
364 "dump already processed");
365
366 if (dumphdr.dump_version != DUMP_VERSION)
367 logprint(SC_SL_NONE | SC_IF_VERBOSE | SC_EXIT_OK,
368 "dump version (%d) != %s version (%d)",
369 dumphdr.dump_version, progname, DUMP_VERSION);
370
371 if (dumphdr.dump_wordsize != DUMP_WORDSIZE)
372 logprint(SC_SL_NONE | SC_EXIT_OK,
373 "dump is from %u-bit kernel - cannot save on %u-bit kernel",
374 dumphdr.dump_wordsize, DUMP_WORDSIZE);
375
376 if (datahdr.dump_datahdr_magic == DUMP_DATAHDR_MAGIC) {
377 if (datahdr.dump_datahdr_version != DUMP_DATAHDR_VERSION)
378 logprint(SC_SL_NONE | SC_IF_VERBOSE | SC_EXIT_OK,
379 "dump data version (%d) != %s data version (%d)",
380 datahdr.dump_datahdr_version, progname,
381 DUMP_DATAHDR_VERSION);
382 } else {
383 (void) memset(&datahdr, 0, sizeof (datahdr));
384 datahdr.dump_maxcsize = pagesize;
385 }
386
387 /*
388 * Read the initial header, clear the valid bits, and compare headers.
389 * The main header may have been overwritten by swapping if we're
390 * using a swap partition as the dump device, in which case we bail.
391 */
392 Pread(dumpfd, &corehdr, sizeof (dumphdr_t), dumphdr.dump_start);
393
394 corehdr.dump_flags &= ~DF_VALID;
395 dumphdr.dump_flags &= ~DF_VALID;
396
397 if (memcmp(&corehdr, &dumphdr, sizeof (dumphdr_t)) != 0) {
398 /*
399 * Clear valid bit so we don't complain on every invocation.
400 */
401 if (!filemode)
402 Pwrite(dumpfd, &dumphdr, sizeof (dumphdr), endoff);
403 logprint(SC_SL_ERR | SC_EXIT_ERR,
404 "initial dump header corrupt");
405 }
406 }
407
408 static void
409 check_space(int csave)
410 {
411 struct statvfs fsb;
412 int64_t spacefree, dumpsize, minfree, datasize;
413
414 if (statvfs(".", &fsb) < 0)
415 logprint(SC_SL_ERR | SC_EXIT_ERR, "statvfs: %s",
416 strerror(errno));
417
418 dumpsize = dumphdr.dump_data - dumphdr.dump_start;
419 datasize = dumphdr.dump_npages * pagesize;
420 if (!csave)
421 dumpsize += datasize;
422 else
423 dumpsize += datahdr.dump_data_csize;
424
425 spacefree = (int64_t)fsb.f_bavail * fsb.f_frsize;
426 minfree = 1024LL * read_number_from_file("minfree", 1024);
427 if (spacefree < minfree + dumpsize) {
428 logprint(SC_SL_ERR | SC_EXIT_ERR,
429 "not enough space in %s (%lld MB avail, %lld MB needed)",
430 savedir, spacefree >> 20, (minfree + dumpsize) >> 20);
431 }
432 }
433
434 static void
435 build_dump_map(int corefd, const pfn_t *pfn_table)
436 {
437 long i;
438 static long misses = 0;
439 size_t dump_mapsize = (corehdr.dump_hashmask + 1) * sizeof (dump_map_t);
440 mem_vtop_t vtop;
441 dump_map_t *dmp = Zalloc(dump_mapsize);
442 char *inbuf = Zalloc(FBUFSIZE);
443 FILE *in = fdopen(dup(dumpfd), "rb");
444
445 (void) setvbuf(in, inbuf, _IOFBF, FBUFSIZE);
446 Fseek(dumphdr.dump_map, in);
447
448 corehdr.dump_data = corehdr.dump_map + roundup(dump_mapsize, pagesize);
449
450 for (i = 0; i < corehdr.dump_nvtop; i++) {
451 long first = 0;
452 long last = corehdr.dump_npages - 1;
453 long middle;
454 pfn_t pfn;
455 uintptr_t h;
456
457 Fread(&vtop, sizeof (mem_vtop_t), in);
458 while (last >= first) {
459 middle = (first + last) / 2;
460 pfn = pfn_table[middle];
461 if (pfn == vtop.m_pfn)
462 break;
463 if (pfn < vtop.m_pfn)
464 first = middle + 1;
465 else
466 last = middle - 1;
467 }
468 if (pfn != vtop.m_pfn) {
469 if (++misses <= 10)
470 (void) fprintf(stderr,
471 "pfn %ld not found for as=%p, va=%p\n",
472 vtop.m_pfn, (void *)vtop.m_as, vtop.m_va);
473 continue;
474 }
475
476 dmp[i].dm_as = vtop.m_as;
477 dmp[i].dm_va = (uintptr_t)vtop.m_va;
478 dmp[i].dm_data = corehdr.dump_data +
479 ((uint64_t)middle << corehdr.dump_pageshift);
480
481 h = DUMP_HASH(&corehdr, dmp[i].dm_as, dmp[i].dm_va);
482 dmp[i].dm_next = dmp[h].dm_first;
483 dmp[h].dm_first = corehdr.dump_map + i * sizeof (dump_map_t);
484 }
485
486 Pwrite(corefd, dmp, dump_mapsize, corehdr.dump_map);
487 free(dmp);
488 (void) fclose(in);
489 free(inbuf);
490 }
491
492 /*
493 * Copy whole sections of the dump device to the file.
494 */
495 static void
496 Copy(offset_t dumpoff, len_t nb, offset_t *offp, int fd, char *buf,
497 size_t sz)
498 {
499 size_t nr;
500 offset_t off = *offp;
501
502 while (nb > 0) {
503 nr = sz < nb ? sz : (size_t)nb;
504 Pread(dumpfd, buf, nr, dumpoff);
505 Pwrite(fd, buf, nr, off);
506 off += nr;
507 dumpoff += nr;
508 nb -= nr;
509 }
510 *offp = off;
511 }
512
513 /*
514 * Copy pages when the dump data header is missing.
515 * This supports older kernels with latest savecore.
516 */
517 static void
518 CopyPages(offset_t *offp, int fd, char *buf, size_t sz)
519 {
520 uint32_t csize;
521 FILE *in = fdopen(dup(dumpfd), "rb");
522 FILE *out = fdopen(dup(fd), "wb");
523 char *cbuf = Zalloc(pagesize);
524 char *outbuf = Zalloc(FBUFSIZE);
525 pgcnt_t np = dumphdr.dump_npages;
526
527 (void) setvbuf(out, outbuf, _IOFBF, FBUFSIZE);
528 (void) setvbuf(in, buf, _IOFBF, sz);
529 Fseek(dumphdr.dump_data, in);
530
531 Fseek(*offp, out);
532 while (np > 0) {
533 Fread(&csize, sizeof (uint32_t), in);
534 Fwrite(&csize, sizeof (uint32_t), out);
535 *offp += sizeof (uint32_t);
536 if (csize > pagesize || csize == 0) {
537 logprint(SC_SL_ERR,
538 "CopyPages: page %lu csize %d (0x%x) pagesize %d",
539 dumphdr.dump_npages - np, csize, csize,
540 pagesize);
541 break;
542 }
543 Fread(cbuf, csize, in);
544 Fwrite(cbuf, csize, out);
545 *offp += csize;
546 np--;
547 }
548 (void) fclose(in);
549 (void) fclose(out);
550 free(outbuf);
551 free(buf);
552 }
553
554 /*
555 * Concatenate dump contents into a new file.
556 * Update corehdr with new offsets.
557 */
558 static void
559 copy_crashfile(const char *corefile)
560 {
561 int corefd = Open(corefile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
562 size_t bufsz = FBUFSIZE;
563 char *inbuf = Zalloc(bufsz);
564 offset_t coreoff;
565 size_t nb;
566
567 logprint(SC_SL_ERR | SC_IF_VERBOSE,
568 "Copying %s to %s/%s\n", dumpfile, savedir, corefile);
569
570 /*
571 * This dump file is still compressed
572 */
573 corehdr.dump_flags |= DF_COMPRESSED | DF_VALID;
574
575 /*
576 * Leave room for corehdr, it is updated and written last
577 */
578 corehdr.dump_start = 0;
579 coreoff = sizeof (corehdr);
580
581 /*
582 * Read in the compressed symbol table, copy it to corefile.
583 */
584 coreoff = roundup(coreoff, pagesize);
585 corehdr.dump_ksyms = coreoff;
586 Copy(dumphdr.dump_ksyms, dumphdr.dump_ksyms_csize, &coreoff, corefd,
587 inbuf, bufsz);
588
589 /*
590 * Save the pfn table.
591 */
592 coreoff = roundup(coreoff, pagesize);
593 corehdr.dump_pfn = coreoff;
594 Copy(dumphdr.dump_pfn, dumphdr.dump_npages * sizeof (pfn_t), &coreoff,
595 corefd, inbuf, bufsz);
596
597 /*
598 * Save the dump map.
599 */
600 coreoff = roundup(coreoff, pagesize);
601 corehdr.dump_map = coreoff;
602 Copy(dumphdr.dump_map, dumphdr.dump_nvtop * sizeof (mem_vtop_t),
603 &coreoff, corefd, inbuf, bufsz);
604
605 /*
606 * Save the data pages.
607 */
608 coreoff = roundup(coreoff, pagesize);
609 corehdr.dump_data = coreoff;
610 if (datahdr.dump_data_csize != 0)
611 Copy(dumphdr.dump_data, datahdr.dump_data_csize, &coreoff,
612 corefd, inbuf, bufsz);
613 else
614 CopyPages(&coreoff, corefd, inbuf, bufsz);
615
616 /*
617 * Now write the modified dump header to front and end of the copy.
618 * Make it look like a valid dump device.
619 *
620 * From dumphdr.h: Two headers are written out: one at the
621 * beginning of the dump, and the other at the very end of the
622 * dump device. The terminal header is at a known location
623 * (end of device) so we can always find it.
624 *
625 * Pad with zeros to each DUMP_OFFSET boundary.
626 */
627 (void) memset(inbuf, 0, DUMP_OFFSET);
628
629 nb = DUMP_OFFSET - (coreoff & (DUMP_OFFSET - 1));
630 if (nb > 0) {
631 Pwrite(corefd, inbuf, nb, coreoff);
632 coreoff += nb;
633 }
634
635 Pwrite(corefd, &corehdr, sizeof (corehdr), coreoff);
636 coreoff += sizeof (corehdr);
637
638 Pwrite(corefd, &datahdr, sizeof (datahdr), coreoff);
639 coreoff += sizeof (datahdr);
640
641 nb = DUMP_OFFSET - (coreoff & (DUMP_OFFSET - 1));
642 if (nb > 0) {
643 Pwrite(corefd, inbuf, nb, coreoff);
644 }
645
646 free(inbuf);
647 Pwrite(corefd, &corehdr, sizeof (corehdr), corehdr.dump_start);
648
649 /*
650 * Write out the modified dump header to the dump device.
651 * The dump device has been processed, so DF_VALID is clear.
652 */
653 if (!filemode)
654 Pwrite(dumpfd, &dumphdr, sizeof (dumphdr), endoff);
655
656 (void) close(corefd);
657 }
658
659 /*
660 * compressed streams
661 */
662 typedef struct blockhdr blockhdr_t;
663 typedef struct block block_t;
664
665 struct blockhdr {
666 block_t *head;
667 block_t *tail;
668 };
669
670 struct block {
671 block_t *next;
672 char *block;
673 int size;
674 };
675
676 typedef enum streamstate {
677 STREAMSTART,
678 STREAMPAGES
679 } streamstate_t;
680
681 typedef struct stream {
682 streamstate_t state;
683 int init;
684 int tag;
685 int bound;
686 int nout;
687 char *blkbuf;
688 blockhdr_t blocks;
689 pgcnt_t pagenum;
690 pgcnt_t curpage;
691 pgcnt_t npages;
692 pgcnt_t done;
693 bz_stream strm;
694 dumpcsize_t sc;
695 dumpstreamhdr_t sh;
696 } stream_t;
697
698 static stream_t *streams;
699 static stream_t *endstreams;
700
701 const int cs = sizeof (dumpcsize_t);
702
703 typedef struct tinfo {
704 pthread_t tid;
705 int corefd;
706 } tinfo_t;
707
708 static int threads_stop;
709 static int threads_active;
710 static tinfo_t *tinfo;
711 static tinfo_t *endtinfo;
712
713 static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;
714 static pthread_cond_t cvfree = PTHREAD_COND_INITIALIZER;
715 static pthread_cond_t cvwork = PTHREAD_COND_INITIALIZER;
716 static pthread_cond_t cvbarrier = PTHREAD_COND_INITIALIZER;
717
718 static blockhdr_t freeblocks;
719
720 static void
721 enqt(blockhdr_t *h, block_t *b)
722 {
723 b->next = NULL;
724 if (h->tail == NULL)
725 h->head = b;
726 else
727 h->tail->next = b;
728 h->tail = b;
729 }
730
731 static block_t *
732 deqh(blockhdr_t *h)
733 {
734 block_t *b = h->head;
735
736 if (b != NULL) {
737 h->head = b->next;
738 if (h->head == NULL)
739 h->tail = NULL;
740 }
741 return (b);
742 }
743
744 static void *runstreams(void *arg);
745
746 static void
747 initstreams(int corefd, int nstreams, int maxcsize)
748 {
749 int nthreads;
750 int nblocks;
751 int i;
752 block_t *b;
753 tinfo_t *t;
754
755 nthreads = sysconf(_SC_NPROCESSORS_ONLN);
756 if (nstreams < nthreads)
757 nthreads = nstreams;
758 if (nthreads < 1)
759 nthreads = 1;
760 nblocks = nthreads * 2;
761
762 tinfo = Zalloc(nthreads * sizeof (tinfo_t));
763 endtinfo = &tinfo[nthreads];
764
765 /* init streams */
766 streams = Zalloc(nstreams * sizeof (stream_t));
767 endstreams = &streams[nstreams];
768
769 /* init stream block buffers */
770 for (i = 0; i < nblocks; i++) {
771 b = Zalloc(sizeof (block_t));
772 b->block = Zalloc(maxcsize);
773 enqt(&freeblocks, b);
774 }
775
776 /* init worker threads */
777 (void) pthread_mutex_lock(&lock);
778 threads_active = 1;
779 threads_stop = 0;
780 for (t = tinfo; t != endtinfo; t++) {
781 t->corefd = dup(corefd);
782 if (t->corefd < 0) {
783 nthreads = t - tinfo;
784 endtinfo = t;
785 break;
786 }
787 if (pthread_create(&t->tid, NULL, runstreams, t) != 0)
788 logprint(SC_SL_ERR | SC_EXIT_ERR, "pthread_create: %s",
789 strerror(errno));
790 }
791 (void) pthread_mutex_unlock(&lock);
792 }
793
794 static void
795 sbarrier()
796 {
797 stream_t *s;
798
799 (void) pthread_mutex_lock(&lock);
800 for (s = streams; s != endstreams; s++) {
801 while (s->bound || s->blocks.head != NULL)
802 (void) pthread_cond_wait(&cvbarrier, &lock);
803 }
804 (void) pthread_mutex_unlock(&lock);
805 }
806
807 static void
808 stopstreams()
809 {
810 tinfo_t *t;
811
812 if (threads_active) {
813 sbarrier();
814 (void) pthread_mutex_lock(&lock);
815 threads_stop = 1;
816 (void) pthread_cond_signal(&cvwork);
817 (void) pthread_mutex_unlock(&lock);
818 for (t = tinfo; t != endtinfo; t++)
819 (void) pthread_join(t->tid, NULL);
820 free(tinfo);
821 tinfo = NULL;
822 threads_active = 0;
823 }
824 }
825
826 static block_t *
827 getfreeblock()
828 {
829 block_t *b;
830
831 (void) pthread_mutex_lock(&lock);
832 while ((b = deqh(&freeblocks)) == NULL)
833 (void) pthread_cond_wait(&cvfree, &lock);
834 (void) pthread_mutex_unlock(&lock);
835 return (b);
836 }
837
838 /* data page offset from page number */
839 #define BTOP(b) ((b) >> dumphdr.dump_pageshift)
840 #define PTOB(p) ((p) << dumphdr.dump_pageshift)
841 #define DATAOFF(p) (corehdr.dump_data + PTOB(p))
842
843 /* check for coreblksize boundary */
844 static int
845 isblkbnd(pgcnt_t pgnum)
846 {
847 return (P2PHASE(DATAOFF(pgnum), coreblksize) == 0);
848 }
849
850 static int
851 iszpage(char *buf)
852 {
853 size_t sz;
854 uint64_t *pl;
855
856 /*LINTED:E_BAD_PTR_CAST_ALIGN*/
857 pl = (uint64_t *)(buf);
858 for (sz = 0; sz < pagesize; sz += sizeof (*pl))
859 if (*pl++ != 0)
860 return (0);
861 return (1);
862 }
863
864 volatile uint_t *hist;
865
866 /* write pages to the core file */
867 static void
868 putpage(int corefd, char *buf, pgcnt_t pgnum, pgcnt_t np)
869 {
870 atomic_inc_uint(&hist[np]);
871 if (np > 0)
872 Pwrite(corefd, buf, PTOB(np), DATAOFF(pgnum));
873 }
874
875 /*
876 * Process one lzjb block.
877 * No object (stream header or page) will be split over a block boundary.
878 */
879 static void
880 lzjbblock(int corefd, stream_t *s, char *block, size_t blocksz)
881 {
882 int in = 0;
883 int csize;
884 int doflush;
885 char *out;
886 size_t dsize;
887 dumpcsize_t sc;
888 dumpstreamhdr_t sh;
889
890 if (!s->init) {
891 s->init = 1;
892 if (s->blkbuf == NULL)
893 s->blkbuf = Zalloc(coreblksize);
894 s->state = STREAMSTART;
895 }
896 while (in < blocksz) {
897 switch (s->state) {
898 case STREAMSTART:
899 (void) memcpy(&sh, block + in, sizeof (sh));
900 in += sizeof (sh);
901 if (strcmp(DUMP_STREAM_MAGIC, sh.stream_magic) != 0)
902 logprint(SC_SL_ERR | SC_EXIT_ERR,
903 "LZJB STREAMSTART: bad stream header");
904 if (sh.stream_npages > datahdr.dump_maxrange)
905 logprint(SC_SL_ERR | SC_EXIT_ERR,
906 "LZJB STREAMSTART: bad range: %d > %d",
907 sh.stream_npages, datahdr.dump_maxrange);
908 s->pagenum = sh.stream_pagenum;
909 s->npages = sh.stream_npages;
910 s->curpage = s->pagenum;
911 s->nout = 0;
912 s->done = 0;
913 s->state = STREAMPAGES;
914 break;
915 case STREAMPAGES:
916 (void) memcpy(&sc, block + in, cs);
917 in += cs;
918 csize = DUMP_GET_CSIZE(sc);
919 if (csize > pagesize)
920 logprint(SC_SL_ERR | SC_EXIT_ERR,
921 "LZJB STREAMPAGES: bad csize=%d", csize);
922
923 out = s->blkbuf + PTOB(s->nout);
924 dsize = decompress(block + in, out, csize, pagesize);
925
926 if (dsize != pagesize)
927 logprint(SC_SL_ERR | SC_EXIT_ERR,
928 "LZJB STREAMPAGES: dsize %d != pagesize %d",
929 dsize, pagesize);
930
931 in += csize;
932 atomic_inc_64(&saved);
933
934 doflush = 0;
935 if (s->nout == 0 && iszpage(out)) {
936 doflush = 1;
937 atomic_inc_64(&zpages);
938 } else if (++s->nout >= BTOP(coreblksize) ||
939 isblkbnd(s->curpage + s->nout)) {
940 doflush = 1;
941 }
942 if (++s->done >= s->npages) {
943 s->state = STREAMSTART;
944 doflush = 1;
945 }
946 if (doflush) {
947 putpage(corefd, s->blkbuf, s->curpage, s->nout);
948 s->nout = 0;
949 s->curpage = s->pagenum + s->done;
950 }
951 break;
952 }
953 }
954 }
955
956 /* bzlib library reports errors with this callback */
957 void
958 bz_internal_error(int errcode)
959 {
960 logprint(SC_SL_ERR | SC_EXIT_ERR, "bz_internal_error: err %s\n",
961 BZ2_bzErrorString(errcode));
962 }
963
964 /*
965 * Return one object in the stream.
966 *
967 * An object (stream header or page) will likely span an input block
968 * of compression data. Return non-zero when an entire object has been
969 * retrieved from the stream.
970 */
971 static int
972 bz2decompress(stream_t *s, void *buf, size_t size)
973 {
974 int rc;
975
976 if (s->strm.avail_out == 0) {
977 s->strm.next_out = buf;
978 s->strm.avail_out = size;
979 }
980 while (s->strm.avail_in > 0) {
981 rc = BZ2_bzDecompress(&s->strm);
982 if (rc == BZ_STREAM_END) {
983 rc = BZ2_bzDecompressReset(&s->strm);
984 if (rc != BZ_OK)
985 logprint(SC_SL_ERR | SC_EXIT_ERR,
986 "BZ2_bzDecompressReset: %s",
987 BZ2_bzErrorString(rc));
988 continue;
989 }
990
991 if (s->strm.avail_out == 0)
992 break;
993 }
994 return (s->strm.avail_out == 0);
995 }
996
997 /*
998 * Process one bzip2 block.
999 * The interface is documented here:
1000 * http://www.bzip.org/1.0.5/bzip2-manual-1.0.5.html
1001 */
1002 static void
1003 bz2block(int corefd, stream_t *s, char *block, size_t blocksz)
1004 {
1005 int rc = 0;
1006 int doflush;
1007 char *out;
1008
1009 if (!s->init) {
1010 s->init = 1;
1011 rc = BZ2_bzDecompressInit(&s->strm, 0, 0);
1012 if (rc != BZ_OK)
1013 logprint(SC_SL_ERR | SC_EXIT_ERR,
1014 "BZ2_bzDecompressInit: %s", BZ2_bzErrorString(rc));
1015 if (s->blkbuf == NULL)
1016 s->blkbuf = Zalloc(coreblksize);
1017 s->strm.avail_out = 0;
1018 s->state = STREAMSTART;
1019 }
1020 s->strm.next_in = block;
1021 s->strm.avail_in = blocksz;
1022
1023 while (s->strm.avail_in > 0) {
1024 switch (s->state) {
1025 case STREAMSTART:
1026 if (!bz2decompress(s, &s->sh, sizeof (s->sh)))
1027 return;
1028 if (strcmp(DUMP_STREAM_MAGIC, s->sh.stream_magic) != 0)
1029 logprint(SC_SL_ERR | SC_EXIT_ERR,
1030 "BZ2 STREAMSTART: bad stream header");
1031 if (s->sh.stream_npages > datahdr.dump_maxrange)
1032 logprint(SC_SL_ERR | SC_EXIT_ERR,
1033 "BZ2 STREAMSTART: bad range: %d > %d",
1034 s->sh.stream_npages, datahdr.dump_maxrange);
1035 s->pagenum = s->sh.stream_pagenum;
1036 s->npages = s->sh.stream_npages;
1037 s->curpage = s->pagenum;
1038 s->nout = 0;
1039 s->done = 0;
1040 s->state = STREAMPAGES;
1041 break;
1042 case STREAMPAGES:
1043 out = s->blkbuf + PTOB(s->nout);
1044 if (!bz2decompress(s, out, pagesize))
1045 return;
1046
1047 atomic_inc_64(&saved);
1048
1049 doflush = 0;
1050 if (s->nout == 0 && iszpage(out)) {
1051 doflush = 1;
1052 atomic_inc_64(&zpages);
1053 } else if (++s->nout >= BTOP(coreblksize) ||
1054 isblkbnd(s->curpage + s->nout)) {
1055 doflush = 1;
1056 }
1057 if (++s->done >= s->npages) {
1058 s->state = STREAMSTART;
1059 doflush = 1;
1060 }
1061 if (doflush) {
1062 putpage(corefd, s->blkbuf, s->curpage, s->nout);
1063 s->nout = 0;
1064 s->curpage = s->pagenum + s->done;
1065 }
1066 break;
1067 }
1068 }
1069 }
1070
1071 /* report progress */
1072 static void
1073 report_progress()
1074 {
1075 int sec, percent;
1076
1077 if (!interactive)
1078 return;
1079
1080 percent = saved * 100LL / corehdr.dump_npages;
1081 if (percent > percent_done) {
1082 sec = (gethrtime() - startts) / 1000 / 1000 / 1000;
1083 (void) printf("\r%2d:%02d %3d%% done", sec / 60, sec % 60,
1084 percent);
1085 (void) fflush(stdout);
1086 percent_done = percent;
1087 }
1088 }
1089
1090 /* thread body */
1091 static void *
1092 runstreams(void *arg)
1093 {
1094 tinfo_t *t = arg;
1095 stream_t *s;
1096 block_t *b;
1097 int bound;
1098
1099 (void) pthread_mutex_lock(&lock);
1100 while (!threads_stop) {
1101 bound = 0;
1102 for (s = streams; s != endstreams; s++) {
1103 if (s->bound || s->blocks.head == NULL)
1104 continue;
1105 s->bound = 1;
1106 bound = 1;
1107 (void) pthread_cond_signal(&cvwork);
1108 while (s->blocks.head != NULL) {
1109 b = deqh(&s->blocks);
1110 (void) pthread_mutex_unlock(&lock);
1111
1112 if (datahdr.dump_clevel < DUMP_CLEVEL_BZIP2)
1113 lzjbblock(t->corefd, s, b->block,
1114 b->size);
1115 else
1116 bz2block(t->corefd, s, b->block,
1117 b->size);
1118
1119 (void) pthread_mutex_lock(&lock);
1120 enqt(&freeblocks, b);
1121 (void) pthread_cond_signal(&cvfree);
1122
1123 report_progress();
1124 }
1125 s->bound = 0;
1126 (void) pthread_cond_signal(&cvbarrier);
1127 }
1128 if (!bound && !threads_stop)
1129 (void) pthread_cond_wait(&cvwork, &lock);
1130 }
1131 (void) close(t->corefd);
1132 (void) pthread_cond_signal(&cvwork);
1133 (void) pthread_mutex_unlock(&lock);
1134 return (arg);
1135 }
1136
1137 /*
1138 * Process compressed pages.
1139 *
1140 * The old format, now called single-threaded lzjb, is a 32-bit size
1141 * word followed by 'size' bytes of lzjb compression data for one
1142 * page. The new format extends this by storing a 12-bit "tag" in the
1143 * upper bits of the size word. When the size word is pagesize or
1144 * less, it is assumed to be one lzjb page. When the size word is
1145 * greater than pagesize, it is assumed to be a "stream block",
1146 * belonging to up to 4095 streams. In practice, the number of streams
1147 * is set to one less than the number of CPUs running at crash
1148 * time. One CPU processes the crash dump, the remaining CPUs
1149 * separately process groups of data pages.
1150 *
1151 * savecore creates a thread per stream, but never more threads than
1152 * the number of CPUs running savecore. This is because savecore can
1153 * be processing a crash file from a remote machine, which may have
1154 * more CPUs.
1155 *
1156 * When the kernel uses parallel lzjb or parallel bzip2, we expect a
1157 * series of 128KB blocks of compression data. In this case, each
1158 * block has a "tag", in the range 1-4095. Each block is handed off to
1159 * to the threads running "runstreams". The dump format is either lzjb
1160 * or bzip2, never a mixture. These threads, in turn, process the
1161 * compression data for groups of pages. Groups of pages are delimited
1162 * by a "stream header", which indicates a starting pfn and number of
1163 * pages. When a stream block has been read, the condition variable
1164 * "cvwork" is signalled, which causes one of the avaiable threads to
1165 * wake up and process the stream.
1166 *
1167 * In the parallel case there will be streams blocks encoding all data
1168 * pages. The stream of blocks is terminated by a zero size
1169 * word. There can be a few lzjb pages tacked on the end, depending on
1170 * the architecture. The sbarrier function ensures that all stream
1171 * blocks have been processed so that the page number for the few
1172 * single pages at the end can be known.
1173 */
1174 static void
1175 decompress_pages(int corefd)
1176 {
1177 char *cpage = NULL;
1178 char *dpage = NULL;
1179 char *out;
1180 pgcnt_t curpage;
1181 block_t *b;
1182 FILE *dumpf;
1183 FILE *tracef = NULL;
1184 stream_t *s;
1185 size_t dsize;
1186 size_t insz = FBUFSIZE;
1187 char *inbuf = Zalloc(insz);
1188 uint32_t csize;
1189 dumpcsize_t dcsize;
1190 int nstreams = datahdr.dump_nstreams;
1191 int maxcsize = datahdr.dump_maxcsize;
1192 int nout, tag, doflush;
1193
1194 dumpf = fdopen(dup(dumpfd), "rb");
1195 if (dumpf == NULL)
1196 logprint(SC_SL_ERR | SC_EXIT_ERR, "fdopen: %s",
1197 strerror(errno));
1198
1199 (void) setvbuf(dumpf, inbuf, _IOFBF, insz);
1200 Fseek(dumphdr.dump_data, dumpf);
1201
1202 /*LINTED: E_CONSTANT_CONDITION*/
1203 while (1) {
1204
1205 /*
1206 * The csize word delimits stream blocks.
1207 * See dumphdr.h for a description.
1208 */
1209 Fread(&dcsize, sizeof (dcsize), dumpf);
1210
1211 tag = DUMP_GET_TAG(dcsize);
1212 csize = DUMP_GET_CSIZE(dcsize);
1213
1214 if (tag != 0) { /* a stream block */
1215
1216 if (nstreams == 0)
1217 logprint(SC_SL_ERR | SC_EXIT_ERR,
1218 "starting data header is missing");
1219
1220 if (tag > nstreams)
1221 logprint(SC_SL_ERR | SC_EXIT_ERR,
1222 "stream tag %d not in range 1..%d",
1223 tag, nstreams);
1224
1225 if (csize > maxcsize)
1226 logprint(SC_SL_ERR | SC_EXIT_ERR,
1227 "block size 0x%x > max csize 0x%x",
1228 csize, maxcsize);
1229
1230 if (streams == NULL)
1231 initstreams(corefd, nstreams, maxcsize);
1232 s = &streams[tag - 1];
1233 s->tag = tag;
1234
1235 b = getfreeblock();
1236 b->size = csize;
1237 Fread(b->block, csize, dumpf);
1238
1239 (void) pthread_mutex_lock(&lock);
1240 enqt(&s->blocks, b);
1241 if (!s->bound)
1242 (void) pthread_cond_signal(&cvwork);
1243 (void) pthread_mutex_unlock(&lock);
1244
1245 } else if (csize > 0) { /* one lzjb page */
1246
1247 if (csize > pagesize)
1248 logprint(SC_SL_ERR | SC_EXIT_ERR,
1249 "csize 0x%x > pagesize 0x%x",
1250 csize, pagesize);
1251
1252 if (cpage == NULL)
1253 cpage = Zalloc(pagesize);
1254 if (dpage == NULL) {
1255 dpage = Zalloc(coreblksize);
1256 nout = 0;
1257 }
1258
1259 Fread(cpage, csize, dumpf);
1260
1261 out = dpage + PTOB(nout);
1262 dsize = decompress(cpage, out, csize, pagesize);
1263
1264 if (dsize != pagesize)
1265 logprint(SC_SL_ERR | SC_EXIT_ERR,
1266 "dsize 0x%x != pagesize 0x%x",
1267 dsize, pagesize);
1268
1269 /*
1270 * wait for streams to flush so that 'saved' is correct
1271 */
1272 if (threads_active)
1273 sbarrier();
1274
1275 doflush = 0;
1276 if (nout == 0)
1277 curpage = saved;
1278
1279 atomic_inc_64(&saved);
1280
1281 if (nout == 0 && iszpage(dpage)) {
1282 doflush = 1;
1283 atomic_inc_64(&zpages);
1284 } else if (++nout >= BTOP(coreblksize) ||
1285 isblkbnd(curpage + nout) ||
1286 saved >= dumphdr.dump_npages) {
1287 doflush = 1;
1288 }
1289
1290 if (doflush) {
1291 putpage(corefd, dpage, curpage, nout);
1292 nout = 0;
1293 }
1294
1295 report_progress();
1296
1297 /*
1298 * Non-streams lzjb does not use blocks. Stop
1299 * here if all the pages have been decompressed.
1300 */
1301 if (saved >= dumphdr.dump_npages)
1302 break;
1303
1304 } else {
1305 break; /* end of data */
1306 }
1307 }
1308
1309 stopstreams();
1310 if (tracef != NULL)
1311 (void) fclose(tracef);
1312 (void) fclose(dumpf);
1313 if (inbuf)
1314 free(inbuf);
1315 if (cpage)
1316 free(cpage);
1317 if (dpage)
1318 free(dpage);
1319 if (streams)
1320 free(streams);
1321 }
1322
1323 static void
1324 build_corefile(const char *namelist, const char *corefile)
1325 {
1326 size_t pfn_table_size = dumphdr.dump_npages * sizeof (pfn_t);
1327 size_t ksyms_size = dumphdr.dump_ksyms_size;
1328 size_t ksyms_csize = dumphdr.dump_ksyms_csize;
1329 pfn_t *pfn_table;
1330 char *ksyms_base = Zalloc(ksyms_size);
1331 char *ksyms_cbase = Zalloc(ksyms_csize);
1332 size_t ksyms_dsize;
1333 Stat_t st;
1334 int corefd = Open(corefile, O_WRONLY | O_CREAT | O_TRUNC, 0644);
1335 int namefd = Open(namelist, O_WRONLY | O_CREAT | O_TRUNC, 0644);
1336
1337 (void) printf("Constructing namelist %s/%s\n", savedir, namelist);
1338
1339 /*
1340 * Determine the optimum write size for the core file
1341 */
1342 Fstat(corefd, &st, corefile);
1343
1344 if (verbose > 1)
1345 (void) printf("%s: %ld block size\n", corefile,
1346 (long)st.st_blksize);
1347 coreblksize = st.st_blksize;
1348 if (coreblksize < MINCOREBLKSIZE || !ISP2(coreblksize))
1349 coreblksize = MINCOREBLKSIZE;
1350
1351 hist = Zalloc((sizeof (uint64_t) * BTOP(coreblksize)) + 1);
1352
1353 /*
1354 * This dump file is now uncompressed
1355 */
1356 corehdr.dump_flags &= ~DF_COMPRESSED;
1357
1358 /*
1359 * Read in the compressed symbol table, copy it to corefile,
1360 * decompress it, and write the result to namelist.
1361 */
1362 corehdr.dump_ksyms = pagesize;
1363 Pread(dumpfd, ksyms_cbase, ksyms_csize, dumphdr.dump_ksyms);
1364 Pwrite(corefd, ksyms_cbase, ksyms_csize, corehdr.dump_ksyms);
1365
1366 ksyms_dsize = decompress(ksyms_cbase, ksyms_base, ksyms_csize,
1367 ksyms_size);
1368 if (ksyms_dsize != ksyms_size)
1369 logprint(SC_SL_WARN,
1370 "bad data in symbol table, %lu of %lu bytes saved",
1371 ksyms_dsize, ksyms_size);
1372
1373 Pwrite(namefd, ksyms_base, ksyms_size, 0);
1374 (void) close(namefd);
1375 free(ksyms_cbase);
1376 free(ksyms_base);
1377
1378 (void) printf("Constructing corefile %s/%s\n", savedir, corefile);
1379
1380 /*
1381 * Read in and write out the pfn table.
1382 */
1383 pfn_table = Zalloc(pfn_table_size);
1384 corehdr.dump_pfn = corehdr.dump_ksyms + roundup(ksyms_size, pagesize);
1385 Pread(dumpfd, pfn_table, pfn_table_size, dumphdr.dump_pfn);
1386 Pwrite(corefd, pfn_table, pfn_table_size, corehdr.dump_pfn);
1387
1388 /*
1389 * Convert the raw translation data into a hashed dump map.
1390 */
1391 corehdr.dump_map = corehdr.dump_pfn + roundup(pfn_table_size, pagesize);
1392 build_dump_map(corefd, pfn_table);
1393 free(pfn_table);
1394
1395 /*
1396 * Decompress the pages
1397 */
1398 decompress_pages(corefd);
1399 (void) printf(": %ld of %ld pages saved\n", (pgcnt_t)saved,
1400 dumphdr.dump_npages);
1401
1402 if (verbose)
1403 (void) printf("%ld (%ld%%) zero pages were not written\n",
1404 (pgcnt_t)zpages, (pgcnt_t)zpages * 100 /
1405 dumphdr.dump_npages);
1406
1407 if (saved != dumphdr.dump_npages)
1408 logprint(SC_SL_WARN, "bad data after page %ld", saved);
1409
1410 /*
1411 * Write out the modified dump headers.
1412 */
1413 Pwrite(corefd, &corehdr, sizeof (corehdr), 0);
1414 if (!filemode)
1415 Pwrite(dumpfd, &dumphdr, sizeof (dumphdr), endoff);
1416
1417 (void) close(corefd);
1418 }
1419
1420 /*
1421 * When the system panics, the kernel saves all undelivered messages (messages
1422 * that never made it out to syslogd(1M)) in the dump. At a mimimum, the
1423 * panic message itself will always fall into this category. Upon reboot,
1424 * the syslog startup script runs savecore -m to recover these messages.
1425 *
1426 * To do this, we read the unsent messages from the dump and send them to
1427 * /dev/conslog on priority band 1. This has the effect of prepending them
1428 * to any already-accumulated messages in the console backlog, thus preserving
1429 * temporal ordering across the reboot.
1430 *
1431 * Note: since savecore -m is used *only* for this purpose, it does *not*
1432 * attempt to save the crash dump. The dump will be saved later, after
1433 * syslogd(1M) starts, by the savecore startup script.
1434 */
1435 static int
1436 message_save(void)
1437 {
1438 offset_t dumpoff = -(DUMP_OFFSET + DUMP_LOGSIZE);
1439 offset_t ldoff;
1440 log_dump_t ld;
1441 log_ctl_t lc;
1442 struct strbuf ctl, dat;
1443 int logfd;
1444
1445 logfd = Open("/dev/conslog", O_WRONLY, 0644);
1446 dumpfd = Open(dumpfile, O_RDWR | O_DSYNC, 0644);
1447 dumpoff = llseek(dumpfd, dumpoff, SEEK_END) & -DUMP_OFFSET;
1448
1449 ctl.buf = (void *)&lc;
1450 ctl.len = sizeof (log_ctl_t);
1451
1452 dat.buf = Zalloc(DUMP_LOGSIZE);
1453
1454 for (;;) {
1455 ldoff = dumpoff;
1456
1457 Pread(dumpfd, &ld, sizeof (log_dump_t), dumpoff);
1458 dumpoff += sizeof (log_dump_t);
1459 dat.len = ld.ld_msgsize;
1460
1461 if (ld.ld_magic == 0)
1462 break;
1463
1464 if (ld.ld_magic != LOG_MAGIC)
1465 logprint(SC_SL_ERR | SC_IF_VERBOSE | SC_EXIT_ERR,
1466 "bad magic %x", ld.ld_magic);
1467
1468 if (dat.len >= DUMP_LOGSIZE)
1469 logprint(SC_SL_ERR | SC_IF_VERBOSE | SC_EXIT_ERR,
1470 "bad size %d", ld.ld_msgsize);
1471
1472 Pread(dumpfd, ctl.buf, ctl.len, dumpoff);
1473 dumpoff += ctl.len;
1474
1475 if (ld.ld_csum != checksum32(ctl.buf, ctl.len))
1476 logprint(SC_SL_ERR | SC_IF_VERBOSE | SC_EXIT_OK,
1477 "bad log_ctl checksum");
1478
1479 lc.flags |= SL_LOGONLY;
1480
1481 Pread(dumpfd, dat.buf, dat.len, dumpoff);
1482 dumpoff += dat.len;
1483
1484 if (ld.ld_msum != checksum32(dat.buf, dat.len))
1485 logprint(SC_SL_ERR | SC_IF_VERBOSE | SC_EXIT_OK,
1486 "bad message checksum");
1487
1488 if (putpmsg(logfd, &ctl, &dat, 1, MSG_BAND) == -1)
1489 logprint(SC_SL_ERR | SC_EXIT_ERR, "putpmsg: %s",
1490 strerror(errno));
1491
1492 ld.ld_magic = 0; /* clear magic so we never save twice */
1493 Pwrite(dumpfd, &ld, sizeof (log_dump_t), ldoff);
1494 }
1495 return (0);
1496 }
1497
1498 static long
1499 getbounds(const char *f)
1500 {
1501 long b = -1;
1502 const char *p = strrchr(f, '/');
1503
1504 (void) sscanf(p ? p + 1 : f, "vmdump.%ld", &b);
1505 return (b);
1506 }
1507
1508 static void
1509 stack_retrieve(char *stack)
1510 {
1511 summary_dump_t sd;
1512 offset_t dumpoff = -(DUMP_OFFSET + DUMP_LOGSIZE +
1513 DUMP_ERPTSIZE);
1514 dumpoff -= DUMP_SUMMARYSIZE;
1515
1516 dumpfd = Open(dumpfile, O_RDWR | O_DSYNC, 0644);
1517 dumpoff = llseek(dumpfd, dumpoff, SEEK_END) & -DUMP_OFFSET;
1518
1519 Pread(dumpfd, &sd, sizeof (summary_dump_t), dumpoff);
1520 dumpoff += sizeof (summary_dump_t);
1521
1522 if (sd.sd_magic == 0) {
1523 *stack = '\0';
1524 return;
1525 }
1526
1527 if (sd.sd_magic != SUMMARY_MAGIC) {
1528 *stack = '\0';
1529 logprint(SC_SL_NONE | SC_IF_VERBOSE,
1530 "bad summary magic %x", sd.sd_magic);
1531 return;
1532 }
1533 Pread(dumpfd, stack, STACK_BUF_SIZE, dumpoff);
1534 if (sd.sd_ssum != checksum32(stack, STACK_BUF_SIZE))
1535 logprint(SC_SL_NONE | SC_IF_VERBOSE, "bad stack checksum");
1536 }
1537
1538 static void
1539 raise_event(enum sc_event_type evidx, char *warn_string)
1540 {
1541 uint32_t pl = sc_event[evidx].sce_payload;
1542 char panic_stack[STACK_BUF_SIZE];
1543 nvlist_t *attr = NULL;
1544 char uuidbuf[36 + 1];
1545 int err = 0;
1546
1547 if (nvlist_alloc(&attr, NV_UNIQUE_NAME, 0) != 0)
1548 goto publish; /* try to send payload-free event */
1549
1550 if (pl & SC_PAYLOAD_SAVEDIR && savedir != NULL)
1551 err |= nvlist_add_string(attr, "dumpdir", savedir);
1552
1553 if (pl & SC_PAYLOAD_INSTANCE && bounds != -1)
1554 err |= nvlist_add_int64(attr, "instance", bounds);
1555
1556 if (pl & SC_PAYLOAD_ISCOMPRESSED) {
1557 err |= nvlist_add_boolean_value(attr, "compressed",
1558 csave ? B_TRUE : B_FALSE);
1559 }
1560
1561 if (pl & SC_PAYLOAD_DUMPADM_EN) {
1562 char *disabled = defread("DUMPADM_ENABLE=no");
1563
1564 err |= nvlist_add_boolean_value(attr, "savecore-enabled",
1565 disabled ? B_FALSE : B_TRUE);
1566 }
1567
1568 if (pl & SC_PAYLOAD_IMAGEUUID) {
1569 (void) strncpy(uuidbuf, corehdr.dump_uuid, 36);
1570 uuidbuf[36] = '\0';
1571 err |= nvlist_add_string(attr, "os-instance-uuid", uuidbuf);
1572 }
1573
1574 if (pl & SC_PAYLOAD_CRASHTIME) {
1575 err |= nvlist_add_int64(attr, "crashtime",
1576 (int64_t)corehdr.dump_crashtime);
1577 }
1578
1579 if (pl & SC_PAYLOAD_PANICSTR && corehdr.dump_panicstring[0] != '\0') {
1580 err |= nvlist_add_string(attr, "panicstr",
1581 corehdr.dump_panicstring);
1582 }
1583
1584 if (pl & SC_PAYLOAD_PANICSTACK) {
1585 stack_retrieve(panic_stack);
1586
1587 if (panic_stack[0] != '\0') {
1588 /*
1589 * The summary page may not be present if the dump
1590 * was previously recorded compressed.
1591 */
1592 (void) nvlist_add_string(attr, "panicstack",
1593 panic_stack);
1594 }
1595 }
1596
1597 /* add warning string if this is an ireport for dump failure */
1598 if (pl & SC_PAYLOAD_FAILREASON && warn_string != NULL)
1599 (void) nvlist_add_string(attr, "failure-reason", warn_string);
1600
1601 if (pl & SC_PAYLOAD_DUMPCOMPLETE)
1602 err |= nvlist_add_boolean_value(attr, "dump-incomplete",
1603 dump_incomplete ? B_TRUE : B_FALSE);
1604
1605 if (pl & SC_PAYLOAD_FM_PANIC) {
1606 err |= nvlist_add_boolean_value(attr, "fm-panic",
1607 fm_panic ? B_TRUE : B_FALSE);
1608 }
1609
1610 if (pl & SC_PAYLOAD_JUSTCHECKING) {
1611 err |= nvlist_add_boolean_value(attr, "will-attempt-savecore",
1612 cflag ? B_FALSE : B_TRUE);
1613 }
1614
1615 if (err)
1616 logprint(SC_SL_WARN, "Errors while constructing '%s' "
1617 "event payload; will try to publish anyway.");
1618 publish:
1619 if (fmev_rspublish_nvl(FMEV_RULESET_ON_SUNOS,
1620 "panic", sc_event[evidx].sce_subclass, FMEV_HIPRI,
1621 attr) != FMEV_SUCCESS) {
1622 logprint(SC_SL_ERR, "failed to publish '%s' event: %s",
1623 sc_event[evidx].sce_subclass, fmev_strerror(fmev_errno));
1624 nvlist_free(attr);
1625 }
1626
1627 }
1628
1629
1630 int
1631 main(int argc, char *argv[])
1632 {
1633 int i, c, bfd;
1634 Stat_t st;
1635 struct rlimit rl;
1636 long filebounds = -1;
1637 char namelist[30], corefile[30], boundstr[30];
1638
1639 startts = gethrtime();
1640
1641 (void) getrlimit(RLIMIT_NOFILE, &rl);
1642 rl.rlim_cur = rl.rlim_max;
1643 (void) setrlimit(RLIMIT_NOFILE, &rl);
1644
1645 openlog(progname, LOG_ODELAY, LOG_AUTH);
1646
1647 (void) defopen("/etc/dumpadm.conf");
1648 savedir = defread("DUMPADM_SAVDIR=");
1649 if (savedir != NULL)
1650 savedir = strdup(savedir);
1651
1652 while ((c = getopt(argc, argv, "Lvcdmf:")) != EOF) {
1653 switch (c) {
1654 case 'L':
1655 livedump++;
1656 break;
1657 case 'v':
1658 verbose++;
1659 break;
1660 case 'c':
1661 cflag++;
1662 break;
1663 case 'd':
1664 disregard_valid_flag++;
1665 break;
1666 case 'm':
1667 mflag++;
1668 break;
1669 case 'f':
1670 dumpfile = optarg;
1671 filebounds = getbounds(dumpfile);
1672 break;
1673 case '?':
1674 usage();
1675 }
1676 }
1677
1678 if (geteuid() != 0 && filebounds < 0) {
1679 (void) fprintf(stderr, "%s: %s %s\n", progname,
1680 gettext("you must be root to use"), progname);
1681 exit(1);
1682 }
1683
1684 interactive = isatty(STDOUT_FILENO);
1685
1686 if (cflag && livedump)
1687 usage();
1688
1689 if (dumpfile == NULL || livedump)
1690 dumpfd = Open("/dev/dump", O_RDONLY, 0444);
1691
1692 if (dumpfile == NULL) {
1693 dumpfile = Zalloc(MAXPATHLEN);
1694 if (ioctl(dumpfd, DIOCGETDEV, dumpfile) == -1)
1695 logprint(SC_SL_NONE | SC_IF_ISATTY | SC_EXIT_ERR,
1696 "no dump device configured");
1697 }
1698
1699 if (mflag)
1700 return (message_save());
1701
1702 if (optind == argc - 1)
1703 savedir = argv[optind];
1704
1705 if (savedir == NULL || optind < argc - 1)
1706 usage();
1707
1708 if (livedump && ioctl(dumpfd, DIOCDUMP, NULL) == -1)
1709 logprint(SC_SL_NONE | SC_EXIT_ERR,
1710 "dedicated dump device required");
1711
1712 (void) close(dumpfd);
1713 dumpfd = -1;
1714
1715 Stat(dumpfile, &st);
1716
1717 filemode = S_ISREG(st.st_mode);
1718
1719 if (!filemode && defread("DUMPADM_CSAVE=off") == NULL)
1720 csave = 1;
1721
1722 read_dumphdr();
1723
1724 /*
1725 * We want this message to go to the log file, but not the console.
1726 * There's no good way to do that with the existing syslog facility.
1727 * We could extend it to handle this, but there doesn't seem to be
1728 * a general need for it, so we isolate the complexity here instead.
1729 */
1730 if (dumphdr.dump_panicstring[0] != '\0') {
1731 int logfd = Open("/dev/conslog", O_WRONLY, 0644);
1732 log_ctl_t lc;
1733 struct strbuf ctl, dat;
1734 char msg[DUMP_PANICSIZE + 100];
1735 char fmt[] = "reboot after panic: %s";
1736 uint32_t msgid;
1737
1738 STRLOG_MAKE_MSGID(fmt, msgid);
1739
1740 /* LINTED: E_SEC_SPRINTF_UNBOUNDED_COPY */
1741 (void) sprintf(msg, "%s: [ID %u FACILITY_AND_PRIORITY] ",
1742 progname, msgid);
1743 /* LINTED: E_SEC_PRINTF_VAR_FMT */
1744 (void) sprintf(msg + strlen(msg), fmt,
1745 dumphdr.dump_panicstring);
1746
1747 lc.pri = LOG_AUTH | LOG_ERR;
1748 lc.flags = SL_CONSOLE | SL_LOGONLY;
1749 lc.level = 0;
1750
1751 ctl.buf = (void *)&lc;
1752 ctl.len = sizeof (log_ctl_t);
1753
1754 dat.buf = (void *)msg;
1755 dat.len = strlen(msg) + 1;
1756
1757 (void) putmsg(logfd, &ctl, &dat, 0);
1758 (void) close(logfd);
1759 }
1760
1761 if ((dumphdr.dump_flags & DF_COMPLETE) == 0) {
1762 logprint(SC_SL_WARN, "incomplete dump on dump device");
1763 dump_incomplete = B_TRUE;
1764 }
1765
1766 if (dumphdr.dump_fm_panic)
1767 fm_panic = B_TRUE;
1768
1769 /*
1770 * We have a valid dump on a dump device and know as much about
1771 * it as we're going to at this stage. Raise an event for
1772 * logging and so that FMA can open a case for this panic.
1773 * Avoid this step for FMA-initiated panics - FMA will replay
1774 * ereports off the dump device independently of savecore and
1775 * will make a diagnosis, so we don't want to open two cases
1776 * for the same event. Also avoid raising an event for a
1777 * livedump, or when we inflating a compressed dump.
1778 */
1779 if (!fm_panic && !livedump && !filemode)
1780 raise_event(SC_EVENT_DUMP_PENDING, NULL);
1781
1782 logprint(SC_SL_WARN, "System dump time: %s",
1783 ctime(&dumphdr.dump_crashtime));
1784
1785 /*
1786 * Option -c is designed for use from svc-dumpadm where we know
1787 * that dumpadm -n is in effect but run savecore -c just to
1788 * get the above dump_pending_on_device event raised. If it is run
1789 * interactively then just print further panic details.
1790 */
1791 if (cflag) {
1792 char *disabled = defread("DUMPADM_ENABLE=no");
1793 int lvl = interactive ? SC_SL_WARN : SC_SL_ERR;
1794 int ec = fm_panic ? SC_EXIT_FM : SC_EXIT_PEND;
1795
1796 logprint(lvl | ec,
1797 "Panic crashdump pending on dump device%s "
1798 "run savecore(1M) manually to extract. "
1799 "Image UUID %s%s.",
1800 disabled ? " but dumpadm -n in effect;" : ";",
1801 corehdr.dump_uuid,
1802 fm_panic ? "(fault-management initiated)" : "");
1803 /*NOTREACHED*/
1804 }
1805
1806 if (chdir(savedir) == -1)
1807 logprint(SC_SL_ERR | SC_EXIT_ERR, "chdir(\"%s\"): %s",
1808 savedir, strerror(errno));
1809
1810 check_space(csave);
1811
1812 if (filebounds < 0)
1813 bounds = read_number_from_file("bounds", 0);
1814 else
1815 bounds = filebounds;
1816
1817 if (csave) {
1818 size_t metrics_size = datahdr.dump_metrics;
1819
1820 (void) sprintf(corefile, "vmdump.%ld", bounds);
1821
1822 datahdr.dump_metrics = 0;
1823
1824 logprint(SC_SL_ERR,
1825 "Saving compressed system crash dump in %s/%s",
1826 savedir, corefile);
1827
1828 copy_crashfile(corefile);
1829
1830 /*
1831 * Raise a fault management event that indicates the system
1832 * has panicked. We know a reasonable amount about the
1833 * condition at this time, but the dump is still compressed.
1834 */
1835 if (!livedump && !fm_panic)
1836 raise_event(SC_EVENT_DUMP_AVAILABLE, NULL);
1837
1838 if (metrics_size > 0) {
1839 int sec = (gethrtime() - startts) / 1000 / 1000 / 1000;
1840 FILE *mfile = fopen(METRICSFILE, "a");
1841 char *metrics = Zalloc(metrics_size + 1);
1842
1843 Pread(dumpfd, metrics, metrics_size, endoff +
1844 sizeof (dumphdr) + sizeof (datahdr));
1845
1846 if (sec < 1)
1847 sec = 1;
1848
1849 if (mfile == NULL) {
1850 logprint(SC_SL_WARN,
1851 "Can't create %s:\n%s",
1852 METRICSFILE, metrics);
1853 } else {
1854 (void) fprintf(mfile, "[[[[,,,");
1855 for (i = 0; i < argc; i++)
1856 (void) fprintf(mfile, "%s ", argv[i]);
1857 (void) fprintf(mfile, "\n");
1858 (void) fprintf(mfile, ",,,%s %s %s %s %s\n",
1859 dumphdr.dump_utsname.sysname,
1860 dumphdr.dump_utsname.nodename,
1861 dumphdr.dump_utsname.release,
1862 dumphdr.dump_utsname.version,
1863 dumphdr.dump_utsname.machine);
1864 (void) fprintf(mfile, ",,,%s dump time %s\n",
1865 dumphdr.dump_flags & DF_LIVE ? "Live" :
1866 "Crash", ctime(&dumphdr.dump_crashtime));
1867 (void) fprintf(mfile, ",,,%s/%s\n", savedir,
1868 corefile);
1869 (void) fprintf(mfile, "Metrics:\n%s\n",
1870 metrics);
1871 (void) fprintf(mfile, "Copy pages,%ld\n",
1872 dumphdr. dump_npages);
1873 (void) fprintf(mfile, "Copy time,%d\n", sec);
1874 (void) fprintf(mfile, "Copy pages/sec,%ld\n",
1875 dumphdr.dump_npages / sec);
1876 (void) fprintf(mfile, "]]]]\n");
1877 (void) fclose(mfile);
1878 }
1879 free(metrics);
1880 }
1881
1882 logprint(SC_SL_ERR,
1883 "Decompress the crash dump with "
1884 "\n'savecore -vf %s/%s'",
1885 savedir, corefile);
1886
1887 } else {
1888 (void) sprintf(namelist, "unix.%ld", bounds);
1889 (void) sprintf(corefile, "vmcore.%ld", bounds);
1890
1891 if (interactive && filebounds >= 0 && access(corefile, F_OK)
1892 == 0)
1893 logprint(SC_SL_NONE | SC_EXIT_ERR,
1894 "%s already exists: remove with "
1895 "'rm -f %s/{unix,vmcore}.%ld'",
1896 corefile, savedir, bounds);
1897
1898 logprint(SC_SL_ERR,
1899 "saving system crash dump in %s/{unix,vmcore}.%ld",
1900 savedir, bounds);
1901
1902 build_corefile(namelist, corefile);
1903
1904 if (!livedump && !filemode && !fm_panic)
1905 raise_event(SC_EVENT_DUMP_AVAILABLE, NULL);
1906
1907 if (access(METRICSFILE, F_OK) == 0) {
1908 int sec = (gethrtime() - startts) / 1000 / 1000 / 1000;
1909 FILE *mfile = fopen(METRICSFILE, "a");
1910
1911 if (sec < 1)
1912 sec = 1;
1913
1914 (void) fprintf(mfile, "[[[[,,,");
1915 for (i = 0; i < argc; i++)
1916 (void) fprintf(mfile, "%s ", argv[i]);
1917 (void) fprintf(mfile, "\n");
1918 (void) fprintf(mfile, ",,,%s/%s\n", savedir, corefile);
1919 (void) fprintf(mfile, ",,,%s %s %s %s %s\n",
1920 dumphdr.dump_utsname.sysname,
1921 dumphdr.dump_utsname.nodename,
1922 dumphdr.dump_utsname.release,
1923 dumphdr.dump_utsname.version,
1924 dumphdr.dump_utsname.machine);
1925 (void) fprintf(mfile, "Uncompress pages,%"PRIu64"\n",
1926 saved);
1927 (void) fprintf(mfile, "Uncompress time,%d\n", sec);
1928 (void) fprintf(mfile, "Uncompress pages/sec,%"
1929 PRIu64"\n", saved / sec);
1930 (void) fprintf(mfile, "]]]]\n");
1931 (void) fclose(mfile);
1932 }
1933 }
1934
1935 if (filebounds < 0) {
1936 (void) sprintf(boundstr, "%ld\n", bounds + 1);
1937 bfd = Open("bounds", O_WRONLY | O_CREAT | O_TRUNC, 0644);
1938 Pwrite(bfd, boundstr, strlen(boundstr), 0);
1939 (void) close(bfd);
1940 }
1941
1942 if (verbose) {
1943 int sec = (gethrtime() - startts) / 1000 / 1000 / 1000;
1944
1945 (void) printf("%d:%02d dump %s is done\n",
1946 sec / 60, sec % 60,
1947 csave ? "copy" : "decompress");
1948 }
1949
1950 if (verbose > 1 && hist != NULL) {
1951 int i, nw;
1952
1953 for (i = 1, nw = 0; i <= BTOP(coreblksize); ++i)
1954 nw += hist[i] * i;
1955 (void) printf("pages count %%\n");
1956 for (i = 0; i <= BTOP(coreblksize); ++i) {
1957 if (hist[i] == 0)
1958 continue;
1959 (void) printf("%3d %5u %6.2f\n",
1960 i, hist[i], 100.0 * hist[i] * i / nw);
1961 }
1962 }
1963
1964 (void) close(dumpfd);
1965 dumpfd = -1;
1966
1967 return (0);
1968 }