xref: /netbsd-src/lib/libkvm/kvm.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: kvm.c,v 1.79 2003/01/18 10:40:41 thorpej Exp $	*/
2 
3 /*-
4  * Copyright (c) 1989, 1992, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software developed by the Computer Systems
8  * Engineering group at Lawrence Berkeley Laboratory under DARPA contract
9  * BG 91-66 and contributed to Berkeley.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the University of
22  *	California, Berkeley and its contributors.
23  * 4. Neither the name of the University nor the names of its contributors
24  *    may be used to endorse or promote products derived from this software
25  *    without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37  * SUCH DAMAGE.
38  */
39 
40 #include <sys/cdefs.h>
41 #if defined(LIBC_SCCS) && !defined(lint)
42 #if 0
43 static char sccsid[] = "@(#)kvm.c	8.2 (Berkeley) 2/13/94";
44 #else
45 __RCSID("$NetBSD: kvm.c,v 1.79 2003/01/18 10:40:41 thorpej Exp $");
46 #endif
47 #endif /* LIBC_SCCS and not lint */
48 
49 #include <sys/param.h>
50 #include <sys/user.h>
51 #include <sys/lwp.h>
52 #include <sys/proc.h>
53 #include <sys/ioctl.h>
54 #include <sys/stat.h>
55 #include <sys/sysctl.h>
56 
57 #include <sys/core.h>
58 #include <sys/exec_aout.h>
59 #include <sys/kcore.h>
60 
61 #include <uvm/uvm_extern.h>
62 
63 #include <machine/cpu.h>
64 
65 #include <ctype.h>
66 #include <db.h>
67 #include <fcntl.h>
68 #include <limits.h>
69 #include <nlist.h>
70 #include <paths.h>
71 #include <stdarg.h>
72 #include <stdio.h>
73 #include <stdlib.h>
74 #include <string.h>
75 #include <unistd.h>
76 #include <kvm.h>
77 
78 #include "kvm_private.h"
79 
80 static int	kvm_dbopen __P((kvm_t *));
81 static int	_kvm_get_header __P((kvm_t *));
82 static kvm_t	*_kvm_open __P((kvm_t *, const char *, const char *,
83 		    const char *, int, char *));
84 static int	clear_gap __P((kvm_t *, FILE *, int));
85 static int	open_cloexec  __P((const char *, int, int));
86 static off_t	Lseek __P((kvm_t *, int, off_t, int));
87 static ssize_t	Pread __P((kvm_t *, int, void *, size_t, off_t));
88 
89 char *
90 kvm_geterr(kd)
91 	kvm_t *kd;
92 {
93 	return (kd->errbuf);
94 }
95 
96 /*
97  * Report an error using printf style arguments.  "program" is kd->program
98  * on hard errors, and 0 on soft errors, so that under sun error emulation,
99  * only hard errors are printed out (otherwise, programs like gdb will
100  * generate tons of error messages when trying to access bogus pointers).
101  */
102 void
103 _kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
104 {
105 	va_list ap;
106 
107 	va_start(ap, fmt);
108 	if (program != NULL) {
109 		(void)fprintf(stderr, "%s: ", program);
110 		(void)vfprintf(stderr, fmt, ap);
111 		(void)fputc('\n', stderr);
112 	} else
113 		(void)vsnprintf(kd->errbuf,
114 		    sizeof(kd->errbuf), fmt, ap);
115 
116 	va_end(ap);
117 }
118 
119 void
120 _kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
121 {
122 	va_list ap;
123 	size_t n;
124 
125 	va_start(ap, fmt);
126 	if (program != NULL) {
127 		(void)fprintf(stderr, "%s: ", program);
128 		(void)vfprintf(stderr, fmt, ap);
129 		(void)fprintf(stderr, ": %s\n", strerror(errno));
130 	} else {
131 		char *cp = kd->errbuf;
132 
133 		(void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
134 		n = strlen(cp);
135 		(void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
136 		    strerror(errno));
137 	}
138 	va_end(ap);
139 }
140 
141 void *
142 _kvm_malloc(kd, n)
143 	kvm_t *kd;
144 	size_t n;
145 {
146 	void *p;
147 
148 	if ((p = malloc(n)) == NULL)
149 		_kvm_err(kd, kd->program, "%s", strerror(errno));
150 	return (p);
151 }
152 
153 /*
154  * Open a file setting the close on exec bit.
155  */
156 static int
157 open_cloexec(fname, flags, mode)
158 	const char *fname;
159 	int flags, mode;
160 {
161 	int fd;
162 
163 	if ((fd = open(fname, flags, mode)) == -1)
164 		return fd;
165 	if (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1)
166 		goto error;
167 
168 	return fd;
169 error:
170 	flags = errno;
171 	(void)close(fd);
172 	errno = flags;
173 	return -1;
174 }
175 
176 /*
177  * Wrapper around the lseek(2) system call; calls _kvm_syserr() for us
178  * in the event of emergency.
179  */
180 static off_t
181 Lseek(kd, fd, offset, whence)
182 	kvm_t *kd;
183 	int fd;
184 	off_t offset;
185 	int whence;
186 {
187 	off_t off;
188 
189 	errno = 0;
190 
191 	if ((off = lseek(fd, offset, whence)) == -1 && errno != 0) {
192 		_kvm_syserr(kd, kd->program, "Lseek");
193 		return ((off_t)-1);
194 	}
195 	return (off);
196 }
197 
198 /*
199  * Wrapper around the pread(2) system call; calls _kvm_syserr() for us
200  * in the event of emergency.
201  */
202 static ssize_t
203 Pread(kd, fd, buf, nbytes, offset)
204 	kvm_t *kd;
205 	int fd;
206 	void *buf;
207 	size_t nbytes;
208 	off_t offset;
209 {
210 	ssize_t rv;
211 
212 	errno = 0;
213 
214 	if ((rv = pread(fd, buf, nbytes, offset)) != nbytes &&
215 	    errno != 0)
216 		_kvm_syserr(kd, kd->program, "Pread");
217 	return (rv);
218 }
219 
220 static kvm_t *
221 _kvm_open(kd, uf, mf, sf, flag, errout)
222 	kvm_t *kd;
223 	const char *uf;
224 	const char *mf;
225 	const char *sf;
226 	int flag;
227 	char *errout;
228 {
229 	struct stat st;
230 	int ufgiven;
231 
232 	kd->db = 0;
233 	kd->pmfd = -1;
234 	kd->vmfd = -1;
235 	kd->swfd = -1;
236 	kd->nlfd = -1;
237 	kd->alive = KVM_ALIVE_DEAD;
238 	kd->procbase = 0;
239 	kd->procbase2 = 0;
240 	kd->lwpbase = 0;
241 	kd->nbpg = getpagesize();
242 	kd->swapspc = 0;
243 	kd->argspc = 0;
244 	kd->arglen = 0;
245 	kd->argbuf = 0;
246 	kd->argv = 0;
247 	kd->vmst = 0;
248 	kd->vm_page_buckets = 0;
249 	kd->kcore_hdr = 0;
250 	kd->cpu_dsize = 0;
251 	kd->cpu_data = 0;
252 	kd->dump_off = 0;
253 
254 	if (flag & KVM_NO_FILES) {
255 		kd->alive = KVM_ALIVE_SYSCTL;
256 		return(kd);
257 	}
258 
259 	/*
260 	 * Call the MD open hook.  This sets:
261 	 *	usrstack, min_uva, max_uva
262 	 */
263 	if (_kvm_mdopen(kd)) {
264 		_kvm_err(kd, kd->program, "md init failed");
265 		goto failed;
266 	}
267 
268 	ufgiven = (uf != NULL);
269 	if (!ufgiven) {
270 #ifdef CPU_BOOTED_KERNEL
271 		/* 130 is 128 + '/' + '\0' */
272 		static char booted_kernel[130];
273 		int mib[2], rc;
274 		size_t len;
275 
276 		mib[0] = CTL_MACHDEP;
277 		mib[1] = CPU_BOOTED_KERNEL;
278 		booted_kernel[0] = '/';
279 		booted_kernel[1] = '\0';
280 		len = sizeof(booted_kernel) - 2;
281 		rc = sysctl(&mib[0], 2, &booted_kernel[1], &len, NULL, 0);
282 		booted_kernel[sizeof(booted_kernel) - 1] = '\0';
283 		uf = (booted_kernel[1] == '/') ?
284 		    &booted_kernel[1] : &booted_kernel[0];
285 		if (rc != -1)
286 			rc = stat(uf, &st);
287 		if (rc != -1 && !S_ISREG(st.st_mode))
288 			rc = -1;
289 		if (rc == -1)
290 #endif /* CPU_BOOTED_KERNEL */
291 			uf = _PATH_UNIX;
292 	}
293 	else if (strlen(uf) >= MAXPATHLEN) {
294 		_kvm_err(kd, kd->program, "exec file name too long");
295 		goto failed;
296 	}
297 	if (flag & ~O_RDWR) {
298 		_kvm_err(kd, kd->program, "bad flags arg");
299 		goto failed;
300 	}
301 	if (mf == 0)
302 		mf = _PATH_MEM;
303 	if (sf == 0)
304 		sf = _PATH_DRUM;
305 
306 	if ((kd->pmfd = open_cloexec(mf, flag, 0)) < 0) {
307 		_kvm_syserr(kd, kd->program, "%s", mf);
308 		goto failed;
309 	}
310 	if (fstat(kd->pmfd, &st) < 0) {
311 		_kvm_syserr(kd, kd->program, "%s", mf);
312 		goto failed;
313 	}
314 	if (S_ISCHR(st.st_mode)) {
315 		/*
316 		 * If this is a character special device, then check that
317 		 * it's /dev/mem.  If so, open kmem too.  (Maybe we should
318 		 * make it work for either /dev/mem or /dev/kmem -- in either
319 		 * case you're working with a live kernel.)
320 		 */
321 		if (strcmp(mf, _PATH_MEM) != 0) {	/* XXX */
322 			_kvm_err(kd, kd->program,
323 				 "%s: not physical memory device", mf);
324 			goto failed;
325 		}
326 		if ((kd->vmfd = open_cloexec(_PATH_KMEM, flag, 0)) < 0) {
327 			_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
328 			goto failed;
329 		}
330 		kd->alive = KVM_ALIVE_FILES;
331 		if ((kd->swfd = open_cloexec(sf, flag, 0)) < 0) {
332 			_kvm_syserr(kd, kd->program, "%s", sf);
333 			goto failed;
334 		}
335 		/*
336 		 * Open kvm nlist database.  We only try to use
337 		 * the pre-built database if the namelist file name
338 		 * pointer is NULL.  If the database cannot or should
339 		 * not be opened, open the namelist argument so we
340 		 * revert to slow nlist() calls.
341 		 */
342 		if ((ufgiven || kvm_dbopen(kd) < 0) &&
343 		    (kd->nlfd = open_cloexec(uf, O_RDONLY, 0)) < 0) {
344 			_kvm_syserr(kd, kd->program, "%s", uf);
345 			goto failed;
346 		}
347 	} else {
348 		/*
349 		 * This is a crash dump.
350 		 * Initialize the virtual address translation machinery,
351 		 * but first setup the namelist fd.
352 		 */
353 		if ((kd->nlfd = open_cloexec(uf, O_RDONLY, 0)) < 0) {
354 			_kvm_syserr(kd, kd->program, "%s", uf);
355 			goto failed;
356 		}
357 
358 		/*
359 		 * If there is no valid core header, fail silently here.
360 		 * The address translations however will fail without
361 		 * header. Things can be made to run by calling
362 		 * kvm_dump_mkheader() before doing any translation.
363 		 */
364 		if (_kvm_get_header(kd) == 0) {
365 			if (_kvm_initvtop(kd) < 0)
366 				goto failed;
367 		}
368 	}
369 	return (kd);
370 failed:
371 	/*
372 	 * Copy out the error if doing sane error semantics.
373 	 */
374 	if (errout != 0)
375 		(void)strlcpy(errout, kd->errbuf, _POSIX2_LINE_MAX);
376 	(void)kvm_close(kd);
377 	return (0);
378 }
379 
380 /*
381  * The kernel dump file (from savecore) contains:
382  *    kcore_hdr_t kcore_hdr;
383  *    kcore_seg_t cpu_hdr;
384  *    (opaque)    cpu_data; (size is cpu_hdr.c_size)
385  *	  kcore_seg_t mem_hdr;
386  *    (memory)    mem_data; (size is mem_hdr.c_size)
387  *
388  * Note: khdr is padded to khdr.c_hdrsize;
389  * cpu_hdr and mem_hdr are padded to khdr.c_seghdrsize
390  */
391 static int
392 _kvm_get_header(kd)
393 	kvm_t	*kd;
394 {
395 	kcore_hdr_t	kcore_hdr;
396 	kcore_seg_t	cpu_hdr;
397 	kcore_seg_t	mem_hdr;
398 	size_t		offset;
399 	ssize_t		sz;
400 
401 	/*
402 	 * Read the kcore_hdr_t
403 	 */
404 	sz = Pread(kd, kd->pmfd, &kcore_hdr, sizeof(kcore_hdr), (off_t)0);
405 	if (sz != sizeof(kcore_hdr))
406 		return (-1);
407 
408 	/*
409 	 * Currently, we only support dump-files made by the current
410 	 * architecture...
411 	 */
412 	if ((CORE_GETMAGIC(kcore_hdr) != KCORE_MAGIC) ||
413 	    (CORE_GETMID(kcore_hdr) != MID_MACHINE))
414 		return (-1);
415 
416 	/*
417 	 * Currently, we only support exactly 2 segments: cpu-segment
418 	 * and data-segment in exactly that order.
419 	 */
420 	if (kcore_hdr.c_nseg != 2)
421 		return (-1);
422 
423 	/*
424 	 * Save away the kcore_hdr.  All errors after this
425 	 * should do a to "goto fail" to deallocate things.
426 	 */
427 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr));
428 	memcpy(kd->kcore_hdr, &kcore_hdr, sizeof(kcore_hdr));
429 	offset = kcore_hdr.c_hdrsize;
430 
431 	/*
432 	 * Read the CPU segment header
433 	 */
434 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), (off_t)offset);
435 	if (sz != sizeof(cpu_hdr))
436 		goto fail;
437 	if ((CORE_GETMAGIC(cpu_hdr) != KCORESEG_MAGIC) ||
438 	    (CORE_GETFLAG(cpu_hdr) != CORE_CPU))
439 		goto fail;
440 	offset += kcore_hdr.c_seghdrsize;
441 
442 	/*
443 	 * Read the CPU segment DATA.
444 	 */
445 	kd->cpu_dsize = cpu_hdr.c_size;
446 	kd->cpu_data = _kvm_malloc(kd, cpu_hdr.c_size);
447 	if (kd->cpu_data == NULL)
448 		goto fail;
449 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size, (off_t)offset);
450 	if (sz != cpu_hdr.c_size)
451 		goto fail;
452 	offset += cpu_hdr.c_size;
453 
454 	/*
455 	 * Read the next segment header: data segment
456 	 */
457 	sz = Pread(kd, kd->pmfd, &mem_hdr, sizeof(mem_hdr), (off_t)offset);
458 	if (sz != sizeof(mem_hdr))
459 		goto fail;
460 	offset += kcore_hdr.c_seghdrsize;
461 
462 	if ((CORE_GETMAGIC(mem_hdr) != KCORESEG_MAGIC) ||
463 	    (CORE_GETFLAG(mem_hdr) != CORE_DATA))
464 		goto fail;
465 
466 	kd->dump_off = offset;
467 	return (0);
468 
469 fail:
470 	if (kd->kcore_hdr != NULL) {
471 		free(kd->kcore_hdr);
472 		kd->kcore_hdr = NULL;
473 	}
474 	if (kd->cpu_data != NULL) {
475 		free(kd->cpu_data);
476 		kd->cpu_data = NULL;
477 		kd->cpu_dsize = 0;
478 	}
479 	return (-1);
480 }
481 
482 /*
483  * The format while on the dump device is: (new format)
484  *	kcore_seg_t cpu_hdr;
485  *	(opaque)    cpu_data; (size is cpu_hdr.c_size)
486  *	kcore_seg_t mem_hdr;
487  *	(memory)    mem_data; (size is mem_hdr.c_size)
488  */
489 int
490 kvm_dump_mkheader(kd, dump_off)
491 kvm_t	*kd;
492 off_t	dump_off;
493 {
494 	kcore_seg_t	cpu_hdr;
495 	size_t hdr_size;
496 	ssize_t sz;
497 
498 	if (kd->kcore_hdr != NULL) {
499 	    _kvm_err(kd, kd->program, "already has a dump header");
500 	    return (-1);
501 	}
502 	if (ISALIVE(kd)) {
503 		_kvm_err(kd, kd->program, "don't use on live kernel");
504 		return (-1);
505 	}
506 
507 	/*
508 	 * Validate new format crash dump
509 	 */
510 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), dump_off);
511 	if (sz != sizeof(cpu_hdr))
512 		return (-1);
513 	if ((CORE_GETMAGIC(cpu_hdr) != KCORE_MAGIC)
514 		|| (CORE_GETMID(cpu_hdr) != MID_MACHINE)) {
515 		_kvm_err(kd, 0, "invalid magic in cpu_hdr");
516 		return (0);
517 	}
518 	hdr_size = ALIGN(sizeof(cpu_hdr));
519 
520 	/*
521 	 * Read the CPU segment.
522 	 */
523 	kd->cpu_dsize = cpu_hdr.c_size;
524 	kd->cpu_data = _kvm_malloc(kd, kd->cpu_dsize);
525 	if (kd->cpu_data == NULL)
526 		goto fail;
527 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size,
528 	    dump_off + hdr_size);
529 	if (sz != cpu_hdr.c_size)
530 		goto fail;
531 	hdr_size += kd->cpu_dsize;
532 
533 	/*
534 	 * Leave phys mem pointer at beginning of memory data
535 	 */
536 	kd->dump_off = dump_off + hdr_size;
537 	if (Lseek(kd, kd->pmfd, kd->dump_off, SEEK_SET) == -1)
538 		goto fail;
539 
540 	/*
541 	 * Create a kcore_hdr.
542 	 */
543 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr_t));
544 	if (kd->kcore_hdr == NULL)
545 		goto fail;
546 
547 	kd->kcore_hdr->c_hdrsize    = ALIGN(sizeof(kcore_hdr_t));
548 	kd->kcore_hdr->c_seghdrsize = ALIGN(sizeof(kcore_seg_t));
549 	kd->kcore_hdr->c_nseg       = 2;
550 	CORE_SETMAGIC(*(kd->kcore_hdr), KCORE_MAGIC, MID_MACHINE,0);
551 
552 	/*
553 	 * Now that we have a valid header, enable translations.
554 	 */
555 	if (_kvm_initvtop(kd) == 0)
556 		/* Success */
557 		return (hdr_size);
558 
559 fail:
560 	if (kd->kcore_hdr != NULL) {
561 		free(kd->kcore_hdr);
562 		kd->kcore_hdr = NULL;
563 	}
564 	if (kd->cpu_data != NULL) {
565 		free(kd->cpu_data);
566 		kd->cpu_data = NULL;
567 		kd->cpu_dsize = 0;
568 	}
569 	return (-1);
570 }
571 
572 static int
573 clear_gap(kd, fp, size)
574 kvm_t	*kd;
575 FILE	*fp;
576 int	size;
577 {
578 	if (size <= 0) /* XXX - < 0 should never happen */
579 		return (0);
580 	while (size-- > 0) {
581 		if (fputc(0, fp) == EOF) {
582 			_kvm_syserr(kd, kd->program, "clear_gap");
583 			return (-1);
584 		}
585 	}
586 	return (0);
587 }
588 
589 /*
590  * Write the dump header info to 'fp'. Note that we can't use fseek(3) here
591  * because 'fp' might be a file pointer obtained by zopen().
592  */
593 int
594 kvm_dump_wrtheader(kd, fp, dumpsize)
595 kvm_t	*kd;
596 FILE	*fp;
597 int	dumpsize;
598 {
599 	kcore_seg_t	seghdr;
600 	long		offset;
601 	int		gap;
602 
603 	if (kd->kcore_hdr == NULL || kd->cpu_data == NULL) {
604 		_kvm_err(kd, kd->program, "no valid dump header(s)");
605 		return (-1);
606 	}
607 
608 	/*
609 	 * Write the generic header
610 	 */
611 	offset = 0;
612 	if (fwrite((void*)kd->kcore_hdr, sizeof(kcore_hdr_t), 1, fp) == 0) {
613 		_kvm_syserr(kd, kd->program, "kvm_dump_wrtheader");
614 		return (-1);
615 	}
616 	offset += kd->kcore_hdr->c_hdrsize;
617 	gap     = kd->kcore_hdr->c_hdrsize - sizeof(kcore_hdr_t);
618 	if (clear_gap(kd, fp, gap) == -1)
619 		return (-1);
620 
621 	/*
622 	 * Write the cpu header
623 	 */
624 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_CPU);
625 	seghdr.c_size = ALIGN(kd->cpu_dsize);
626 	if (fwrite((void*)&seghdr, sizeof(seghdr), 1, fp) == 0) {
627 		_kvm_syserr(kd, kd->program, "kvm_dump_wrtheader");
628 		return (-1);
629 	}
630 	offset += kd->kcore_hdr->c_seghdrsize;
631 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
632 	if (clear_gap(kd, fp, gap) == -1)
633 		return (-1);
634 
635 	if (fwrite((void*)kd->cpu_data, kd->cpu_dsize, 1, fp) == 0) {
636 		_kvm_syserr(kd, kd->program, "kvm_dump_wrtheader");
637 		return (-1);
638 	}
639 	offset += seghdr.c_size;
640 	gap     = seghdr.c_size - kd->cpu_dsize;
641 	if (clear_gap(kd, fp, gap) == -1)
642 		return (-1);
643 
644 	/*
645 	 * Write the actual dump data segment header
646 	 */
647 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_DATA);
648 	seghdr.c_size = dumpsize;
649 	if (fwrite((void*)&seghdr, sizeof(seghdr), 1, fp) == 0) {
650 		_kvm_syserr(kd, kd->program, "kvm_dump_wrtheader");
651 		return (-1);
652 	}
653 	offset += kd->kcore_hdr->c_seghdrsize;
654 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
655 	if (clear_gap(kd, fp, gap) == -1)
656 		return (-1);
657 
658 	return (int)offset;
659 }
660 
661 kvm_t *
662 kvm_openfiles(uf, mf, sf, flag, errout)
663 	const char *uf;
664 	const char *mf;
665 	const char *sf;
666 	int flag;
667 	char *errout;
668 {
669 	kvm_t *kd;
670 
671 	if ((kd = malloc(sizeof(*kd))) == NULL) {
672 		(void)strlcpy(errout, strerror(errno), _POSIX2_LINE_MAX);
673 		return (0);
674 	}
675 	kd->program = 0;
676 	return (_kvm_open(kd, uf, mf, sf, flag, errout));
677 }
678 
679 kvm_t *
680 kvm_open(uf, mf, sf, flag, program)
681 	const char *uf;
682 	const char *mf;
683 	const char *sf;
684 	int flag;
685 	const char *program;
686 {
687 	kvm_t *kd;
688 
689 	if ((kd = malloc(sizeof(*kd))) == NULL && program != NULL) {
690 		(void)fprintf(stderr, "%s: %s\n", program, strerror(errno));
691 		return (0);
692 	}
693 	kd->program = program;
694 	return (_kvm_open(kd, uf, mf, sf, flag, NULL));
695 }
696 
697 int
698 kvm_close(kd)
699 	kvm_t *kd;
700 {
701 	int error = 0;
702 
703 	if (kd->pmfd >= 0)
704 		error |= close(kd->pmfd);
705 	if (kd->vmfd >= 0)
706 		error |= close(kd->vmfd);
707 	if (kd->nlfd >= 0)
708 		error |= close(kd->nlfd);
709 	if (kd->swfd >= 0)
710 		error |= close(kd->swfd);
711 	if (kd->db != 0)
712 		error |= (kd->db->close)(kd->db);
713 	if (kd->vmst)
714 		_kvm_freevtop(kd);
715 	kd->cpu_dsize = 0;
716 	if (kd->cpu_data != NULL)
717 		free((void *)kd->cpu_data);
718 	if (kd->kcore_hdr != NULL)
719 		free((void *)kd->kcore_hdr);
720 	if (kd->procbase != 0)
721 		free((void *)kd->procbase);
722 	if (kd->procbase2 != 0)
723 		free((void *)kd->procbase2);
724 	if (kd->lwpbase != 0)
725 		free((void *)kd->lwpbase);
726 	if (kd->swapspc != 0)
727 		free((void *)kd->swapspc);
728 	if (kd->argspc != 0)
729 		free((void *)kd->argspc);
730 	if (kd->argbuf != 0)
731 		free((void *)kd->argbuf);
732 	if (kd->argv != 0)
733 		free((void *)kd->argv);
734 	free((void *)kd);
735 
736 	return (0);
737 }
738 
739 /*
740  * Set up state necessary to do queries on the kernel namelist
741  * data base.  If the data base is out-of-data/incompatible with
742  * given executable, set up things so we revert to standard nlist call.
743  * Only called for live kernels.  Return 0 on success, -1 on failure.
744  */
745 static int
746 kvm_dbopen(kd)
747 	kvm_t *kd;
748 {
749 	DBT rec;
750 	size_t dbversionlen;
751 	struct nlist nitem;
752 	char dbversion[_POSIX2_LINE_MAX];
753 	char kversion[_POSIX2_LINE_MAX];
754 	int fd;
755 
756 	kd->db = dbopen(_PATH_KVMDB, O_RDONLY, 0, DB_HASH, NULL);
757 	if (kd->db == 0)
758 		return (-1);
759 	if ((fd = (*kd->db->fd)(kd->db)) >= 0) {
760 		if (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1) {
761 		       (*kd->db->close)(kd->db);
762 		       return (-1);
763 		}
764 	}
765 	/*
766 	 * read version out of database
767 	 */
768 	rec.data = VRS_KEY;
769 	rec.size = sizeof(VRS_KEY) - 1;
770 	if ((kd->db->get)(kd->db, (DBT *)&rec, (DBT *)&rec, 0))
771 		goto close;
772 	if (rec.data == 0 || rec.size > sizeof(dbversion))
773 		goto close;
774 
775 	memcpy(dbversion, rec.data, rec.size);
776 	dbversionlen = rec.size;
777 	/*
778 	 * Read version string from kernel memory.
779 	 * Since we are dealing with a live kernel, we can call kvm_read()
780 	 * at this point.
781 	 */
782 	rec.data = VRS_SYM;
783 	rec.size = sizeof(VRS_SYM) - 1;
784 	if ((kd->db->get)(kd->db, (DBT *)&rec, (DBT *)&rec, 0))
785 		goto close;
786 	if (rec.data == 0 || rec.size != sizeof(struct nlist))
787 		goto close;
788 	memcpy(&nitem, rec.data, sizeof(nitem));
789 	if (kvm_read(kd, (u_long)nitem.n_value, kversion, dbversionlen) !=
790 	    dbversionlen)
791 		goto close;
792 	/*
793 	 * If they match, we win - otherwise clear out kd->db so
794 	 * we revert to slow nlist().
795 	 */
796 	if (memcmp(dbversion, kversion, dbversionlen) == 0)
797 		return (0);
798 close:
799 	(void)(kd->db->close)(kd->db);
800 	kd->db = 0;
801 
802 	return (-1);
803 }
804 
805 int
806 kvm_nlist(kd, nl)
807 	kvm_t *kd;
808 	struct nlist *nl;
809 {
810 	struct nlist *p;
811 	int nvalid, rv;
812 
813 	/*
814 	 * If we can't use the data base, revert to the
815 	 * slow library call.
816 	 */
817 	if (kd->db == 0) {
818 		rv = __fdnlist(kd->nlfd, nl);
819 		if (rv == -1)
820 			_kvm_err(kd, 0, "bad namelist");
821 		return (rv);
822 	}
823 
824 	/*
825 	 * We can use the kvm data base.  Go through each nlist entry
826 	 * and look it up with a db query.
827 	 */
828 	nvalid = 0;
829 	for (p = nl; p->n_name && p->n_name[0]; ++p) {
830 		int len;
831 		DBT rec;
832 
833 		if ((len = strlen(p->n_name)) > 4096) {
834 			/* sanity */
835 			_kvm_err(kd, kd->program, "symbol too large");
836 			return (-1);
837 		}
838 		rec.data = (char *)p->n_name;
839 		rec.size = len;
840 
841 		/*
842 		 * Make sure that n_value = 0 when the symbol isn't found
843 		 */
844 		p->n_value = 0;
845 
846 		if ((kd->db->get)(kd->db, (DBT *)&rec, (DBT *)&rec, 0))
847 			continue;
848 		if (rec.data == 0 || rec.size != sizeof(struct nlist))
849 			continue;
850 		++nvalid;
851 		/*
852 		 * Avoid alignment issues.
853 		 */
854 		(void)memcpy(&p->n_type, &((struct nlist *)rec.data)->n_type,
855 		      sizeof(p->n_type));
856 		(void)memcpy(&p->n_value, &((struct nlist *)rec.data)->n_value,
857 		      sizeof(p->n_value));
858 	}
859 	/*
860 	 * Return the number of entries that weren't found.
861 	 */
862 	return ((p - nl) - nvalid);
863 }
864 
865 int kvm_dump_inval(kd)
866 kvm_t	*kd;
867 {
868 	struct nlist	nl[2];
869 	u_long		pa, val;
870 
871 	if (ISALIVE(kd)) {
872 		_kvm_err(kd, kd->program, "clearing dump on live kernel");
873 		return (-1);
874 	}
875 	nl[0].n_name = "_dumpmag";
876 	nl[1].n_name = NULL;
877 
878 	if (kvm_nlist(kd, nl) == -1) {
879 		_kvm_err(kd, 0, "bad namelist");
880 		return (-1);
881 	}
882 	if (_kvm_kvatop(kd, (u_long)nl[0].n_value, &pa) == 0)
883 		return (-1);
884 
885 	errno = 0;
886 	val = 0;
887 	if (pwrite(kd->pmfd, (void *)&val, sizeof(val),
888 	    _kvm_pa2off(kd, pa)) == -1) {
889 		_kvm_syserr(kd, 0, "cannot invalidate dump - pwrite");
890 		return (-1);
891 	}
892 	return (0);
893 }
894 
895 ssize_t
896 kvm_read(kd, kva, buf, len)
897 	kvm_t *kd;
898 	u_long kva;
899 	void *buf;
900 	size_t len;
901 {
902 	int cc;
903 	void *cp;
904 
905 	if (ISKMEM(kd)) {
906 		/*
907 		 * We're using /dev/kmem.  Just read straight from the
908 		 * device and let the active kernel do the address translation.
909 		 */
910 		errno = 0;
911 		cc = pread(kd->vmfd, buf, len, (off_t)kva);
912 		if (cc < 0) {
913 			_kvm_syserr(kd, 0, "kvm_read");
914 			return (-1);
915 		} else if (cc < len)
916 			_kvm_err(kd, kd->program, "short read");
917 		return (cc);
918 	} else if (ISSYSCTL(kd)) {
919 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
920 		    "can't use kvm_read");
921 		return (-1);
922 	} else {
923 		if ((kd->kcore_hdr == NULL) || (kd->cpu_data == NULL)) {
924 			_kvm_err(kd, kd->program, "no valid dump header");
925 			return (-1);
926 		}
927 		cp = buf;
928 		while (len > 0) {
929 			u_long	pa;
930 			off_t	foff;
931 
932 			cc = _kvm_kvatop(kd, kva, &pa);
933 			if (cc == 0)
934 				return (-1);
935 			if (cc > len)
936 				cc = len;
937 			foff = _kvm_pa2off(kd, pa);
938 			errno = 0;
939 			cc = pread(kd->pmfd, cp, (size_t)cc, foff);
940 			if (cc < 0) {
941 				_kvm_syserr(kd, kd->program, "kvm_read");
942 				break;
943 			}
944 			/*
945 			 * If kvm_kvatop returns a bogus value or our core
946 			 * file is truncated, we might wind up seeking beyond
947 			 * the end of the core file in which case the read will
948 			 * return 0 (EOF).
949 			 */
950 			if (cc == 0)
951 				break;
952 			cp = (char *)cp + cc;
953 			kva += cc;
954 			len -= cc;
955 		}
956 		return ((char *)cp - (char *)buf);
957 	}
958 	/* NOTREACHED */
959 }
960 
961 ssize_t
962 kvm_write(kd, kva, buf, len)
963 	kvm_t *kd;
964 	u_long kva;
965 	const void *buf;
966 	size_t len;
967 {
968 	int cc;
969 
970 	if (ISKMEM(kd)) {
971 		/*
972 		 * Just like kvm_read, only we write.
973 		 */
974 		errno = 0;
975 		cc = pwrite(kd->vmfd, buf, len, (off_t)kva);
976 		if (cc < 0) {
977 			_kvm_syserr(kd, 0, "kvm_write");
978 			return (-1);
979 		} else if (cc < len)
980 			_kvm_err(kd, kd->program, "short write");
981 		return (cc);
982 	} else if (ISSYSCTL(kd)) {
983 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
984 		    "can't use kvm_write");
985 		return (-1);
986 	} else {
987 		_kvm_err(kd, kd->program,
988 		    "kvm_write not implemented for dead kernels");
989 		return (-1);
990 	}
991 	/* NOTREACHED */
992 }
993