1/*	$NetBSD: kvm.c,v 1.98 2011/09/12 21:11:32 christos 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. Neither the name of the University nor the names of its contributors
20 *    may be used to endorse or promote products derived from this software
21 *    without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 */
35
36#include <sys/cdefs.h>
37#if defined(LIBC_SCCS) && !defined(lint)
38#if 0
39static char sccsid[] = "@(#)kvm.c	8.2 (Berkeley) 2/13/94";
40#else
41__RCSID("$NetBSD: kvm.c,v 1.98 2011/09/12 21:11:32 christos Exp $");
42#endif
43#endif /* LIBC_SCCS and not lint */
44
45#include <sys/param.h>
46#include <sys/user.h>
47#include <sys/lwp.h>
48#include <sys/proc.h>
49#include <sys/ioctl.h>
50#include <sys/stat.h>
51#include <sys/sysctl.h>
52
53#include <sys/core.h>
54#include <sys/exec.h>
55#include <sys/kcore.h>
56#include <sys/ksyms.h>
57#include <sys/types.h>
58
59#include <uvm/uvm_extern.h>
60
61#include <machine/cpu.h>
62
63#include <ctype.h>
64#include <errno.h>
65#include <fcntl.h>
66#include <limits.h>
67#include <nlist.h>
68#include <paths.h>
69#include <stdarg.h>
70#include <stdio.h>
71#include <stdlib.h>
72#include <string.h>
73#include <unistd.h>
74#include <kvm.h>
75
76#include "kvm_private.h"
77
78static int	_kvm_get_header(kvm_t *);
79static kvm_t	*_kvm_open(kvm_t *, const char *, const char *,
80		    const char *, int, char *);
81static int	clear_gap(kvm_t *, bool (*)(void *, const void *, size_t),
82		    void *, size_t);
83static off_t	Lseek(kvm_t *, int, off_t, int);
84static ssize_t	Pread(kvm_t *, int, void *, size_t, off_t);
85
86char *
87kvm_geterr(kvm_t *kd)
88{
89	return (kd->errbuf);
90}
91
92const char *
93kvm_getkernelname(kvm_t *kd)
94{
95	return kd->kernelname;
96}
97
98/*
99 * Report an error using printf style arguments.  "program" is kd->program
100 * on hard errors, and 0 on soft errors, so that under sun error emulation,
101 * only hard errors are printed out (otherwise, programs like gdb will
102 * generate tons of error messages when trying to access bogus pointers).
103 */
104void
105_kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
106{
107	va_list ap;
108
109	va_start(ap, fmt);
110	if (program != NULL) {
111		(void)fprintf(stderr, "%s: ", program);
112		(void)vfprintf(stderr, fmt, ap);
113		(void)fputc('\n', stderr);
114	} else
115		(void)vsnprintf(kd->errbuf,
116		    sizeof(kd->errbuf), fmt, ap);
117
118	va_end(ap);
119}
120
121void
122_kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
123{
124	va_list ap;
125	size_t n;
126
127	va_start(ap, fmt);
128	if (program != NULL) {
129		(void)fprintf(stderr, "%s: ", program);
130		(void)vfprintf(stderr, fmt, ap);
131		(void)fprintf(stderr, ": %s\n", strerror(errno));
132	} else {
133		char *cp = kd->errbuf;
134
135		(void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
136		n = strlen(cp);
137		(void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
138		    strerror(errno));
139	}
140	va_end(ap);
141}
142
143void *
144_kvm_malloc(kvm_t *kd, 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 * Wrapper around the lseek(2) system call; calls _kvm_syserr() for us
155 * in the event of emergency.
156 */
157static off_t
158Lseek(kvm_t *kd, int fd, off_t offset, int whence)
159{
160	off_t off;
161
162	errno = 0;
163
164	if ((off = lseek(fd, offset, whence)) == -1 && errno != 0) {
165		_kvm_syserr(kd, kd->program, "Lseek");
166		return ((off_t)-1);
167	}
168	return (off);
169}
170
171ssize_t
172_kvm_pread(kvm_t *kd, int fd, void *buf, size_t size, off_t off)
173{
174	ptrdiff_t moff;
175	void *newbuf;
176	size_t dsize;
177	ssize_t rv;
178	off_t doff;
179
180	/* If aligned nothing to do. */
181 	if (((off % kd->fdalign) | (size % kd->fdalign)) == 0) {
182		return pread(fd, buf, size, off);
183 	}
184
185	/*
186	 * Otherwise must buffer.  We can't tolerate short reads in this
187	 * case (lazy bum).
188	 */
189	moff = (ptrdiff_t)off % kd->fdalign;
190	doff = off - moff;
191	dsize = moff + size + kd->fdalign - 1;
192	dsize -= dsize % kd->fdalign;
193	if (kd->iobufsz < dsize) {
194		newbuf = realloc(kd->iobuf, dsize);
195		if (newbuf == NULL) {
196			_kvm_syserr(kd, 0, "cannot allocate I/O buffer");
197			return (-1);
198		}
199		kd->iobuf = newbuf;
200		kd->iobufsz = dsize;
201	}
202	rv = pread(fd, kd->iobuf, dsize, doff);
203	if (rv < size + moff)
204		return -1;
205	memcpy(buf, kd->iobuf + moff, size);
206	return size;
207}
208
209/*
210 * Wrapper around the pread(2) system call; calls _kvm_syserr() for us
211 * in the event of emergency.
212 */
213static ssize_t
214Pread(kvm_t *kd, int fd, void *buf, size_t nbytes, off_t offset)
215{
216	ssize_t rv;
217
218	errno = 0;
219
220	if ((rv = _kvm_pread(kd, fd, buf, nbytes, offset)) != nbytes &&
221	    errno != 0)
222		_kvm_syserr(kd, kd->program, "Pread");
223	return (rv);
224}
225
226static kvm_t *
227_kvm_open(kvm_t *kd, const char *uf, const char *mf, const char *sf, int flag,
228    char *errout)
229{
230	struct stat st;
231	int ufgiven;
232
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 = NULL;
239	kd->procbase_len = 0;
240	kd->procbase2 = NULL;
241	kd->procbase2_len = 0;
242	kd->lwpbase = NULL;
243	kd->lwpbase_len = 0;
244	kd->nbpg = getpagesize();
245	kd->swapspc = NULL;
246	kd->argspc = NULL;
247	kd->argspc_len = 0;
248	kd->argbuf = NULL;
249	kd->argv = NULL;
250	kd->vmst = NULL;
251	kd->vm_page_buckets = NULL;
252	kd->kcore_hdr = NULL;
253	kd->cpu_dsize = 0;
254	kd->cpu_data = NULL;
255	kd->dump_off = 0;
256	kd->fdalign = 1;
257	kd->iobuf = NULL;
258	kd->iobufsz = 0;
259
260	if (flag & KVM_NO_FILES) {
261		kd->alive = KVM_ALIVE_SYSCTL;
262		return(kd);
263	}
264
265	/*
266	 * Call the MD open hook.  This sets:
267	 *	usrstack, min_uva, max_uva
268	 */
269	if (_kvm_mdopen(kd)) {
270		_kvm_err(kd, kd->program, "md init failed");
271		goto failed;
272	}
273
274	ufgiven = (uf != NULL);
275	if (!ufgiven) {
276#ifdef CPU_BOOTED_KERNEL
277		/* 130 is 128 + '/' + '\0' */
278		static char booted_kernel[130];
279		int mib[2], rc;
280		size_t len;
281
282		mib[0] = CTL_MACHDEP;
283		mib[1] = CPU_BOOTED_KERNEL;
284		booted_kernel[0] = '/';
285		booted_kernel[1] = '\0';
286		len = sizeof(booted_kernel) - 2;
287		rc = sysctl(&mib[0], 2, &booted_kernel[1], &len, NULL, 0);
288		booted_kernel[sizeof(booted_kernel) - 1] = '\0';
289		uf = (booted_kernel[1] == '/') ?
290		    &booted_kernel[1] : &booted_kernel[0];
291		if (rc != -1)
292			rc = stat(uf, &st);
293		if (rc != -1 && !S_ISREG(st.st_mode))
294			rc = -1;
295		if (rc == -1)
296#endif /* CPU_BOOTED_KERNEL */
297			uf = _PATH_UNIX;
298	}
299	else if (strlen(uf) >= MAXPATHLEN) {
300		_kvm_err(kd, kd->program, "exec file name too long");
301		goto failed;
302	}
303	if (flag & ~O_RDWR) {
304		_kvm_err(kd, kd->program, "bad flags arg");
305		goto failed;
306	}
307	if (mf == 0)
308		mf = _PATH_MEM;
309	if (sf == 0)
310		sf = _PATH_DRUM;
311
312	/*
313	 * Open the kernel namelist.  If /dev/ksyms doesn't
314	 * exist, open the current kernel.
315	 */
316	if (ufgiven == 0)
317		kd->nlfd = open(_PATH_KSYMS, O_RDONLY | O_CLOEXEC, 0);
318	if (kd->nlfd < 0) {
319		if ((kd->nlfd = open(uf, O_RDONLY | O_CLOEXEC, 0)) < 0) {
320			_kvm_syserr(kd, kd->program, "%s", uf);
321			goto failed;
322		}
323		strlcpy(kd->kernelname, uf, sizeof(kd->kernelname));
324	} else {
325		strlcpy(kd->kernelname, _PATH_KSYMS, sizeof(kd->kernelname));
326		/*
327		 * We're here because /dev/ksyms was opened
328		 * successfully.  However, we don't want to keep it
329		 * open, so we close it now.  Later, we will open
330		 * it again, since it will be the only case where
331		 * kd->nlfd is negative.
332		 */
333		close(kd->nlfd);
334		kd->nlfd = -1;
335	}
336
337	if ((kd->pmfd = open(mf, flag | O_CLOEXEC, 0)) < 0) {
338		_kvm_syserr(kd, kd->program, "%s", mf);
339		goto failed;
340	}
341	if (fstat(kd->pmfd, &st) < 0) {
342		_kvm_syserr(kd, kd->program, "%s", mf);
343		goto failed;
344	}
345	if (S_ISCHR(st.st_mode) && strcmp(mf, _PATH_MEM) == 0) {
346		/*
347		 * If this is /dev/mem, open kmem too.  (Maybe we should
348		 * make it work for either /dev/mem or /dev/kmem -- in either
349		 * case you're working with a live kernel.)
350		 */
351		if ((kd->vmfd = open(_PATH_KMEM, flag | O_CLOEXEC, 0)) < 0) {
352			_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
353			goto failed;
354		}
355		kd->alive = KVM_ALIVE_FILES;
356		if ((kd->swfd = open(sf, flag | O_CLOEXEC, 0)) < 0) {
357			if (errno != ENXIO) {
358				_kvm_syserr(kd, kd->program, "%s", sf);
359				goto failed;
360			}
361			/* swap is not configured?  not fatal */
362		}
363	} else {
364		kd->fdalign = DEV_BSIZE;	/* XXX */
365		/*
366		 * This is a crash dump.
367		 * Initialize the virtual address translation machinery.
368		 *
369		 * If there is no valid core header, fail silently here.
370		 * The address translations however will fail without
371		 * header. Things can be made to run by calling
372		 * kvm_dump_mkheader() before doing any translation.
373		 */
374		if (_kvm_get_header(kd) == 0) {
375			if (_kvm_initvtop(kd) < 0)
376				goto failed;
377		}
378	}
379	return (kd);
380failed:
381	/*
382	 * Copy out the error if doing sane error semantics.
383	 */
384	if (errout != 0)
385		(void)strlcpy(errout, kd->errbuf, _POSIX2_LINE_MAX);
386	(void)kvm_close(kd);
387	return (0);
388}
389
390/*
391 * The kernel dump file (from savecore) contains:
392 *    kcore_hdr_t kcore_hdr;
393 *    kcore_seg_t cpu_hdr;
394 *    (opaque)    cpu_data; (size is cpu_hdr.c_size)
395 *	  kcore_seg_t mem_hdr;
396 *    (memory)    mem_data; (size is mem_hdr.c_size)
397 *
398 * Note: khdr is padded to khdr.c_hdrsize;
399 * cpu_hdr and mem_hdr are padded to khdr.c_seghdrsize
400 */
401static int
402_kvm_get_header(kvm_t *kd)
403{
404	kcore_hdr_t	kcore_hdr;
405	kcore_seg_t	cpu_hdr;
406	kcore_seg_t	mem_hdr;
407	size_t		offset;
408	ssize_t		sz;
409
410	/*
411	 * Read the kcore_hdr_t
412	 */
413	sz = Pread(kd, kd->pmfd, &kcore_hdr, sizeof(kcore_hdr), (off_t)0);
414	if (sz != sizeof(kcore_hdr))
415		return (-1);
416
417	/*
418	 * Currently, we only support dump-files made by the current
419	 * architecture...
420	 */
421	if ((CORE_GETMAGIC(kcore_hdr) != KCORE_MAGIC) ||
422	    (CORE_GETMID(kcore_hdr) != MID_MACHINE))
423		return (-1);
424
425	/*
426	 * Currently, we only support exactly 2 segments: cpu-segment
427	 * and data-segment in exactly that order.
428	 */
429	if (kcore_hdr.c_nseg != 2)
430		return (-1);
431
432	/*
433	 * Save away the kcore_hdr.  All errors after this
434	 * should do a to "goto fail" to deallocate things.
435	 */
436	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr));
437	memcpy(kd->kcore_hdr, &kcore_hdr, sizeof(kcore_hdr));
438	offset = kcore_hdr.c_hdrsize;
439
440	/*
441	 * Read the CPU segment header
442	 */
443	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), (off_t)offset);
444	if (sz != sizeof(cpu_hdr))
445		goto fail;
446	if ((CORE_GETMAGIC(cpu_hdr) != KCORESEG_MAGIC) ||
447	    (CORE_GETFLAG(cpu_hdr) != CORE_CPU))
448		goto fail;
449	offset += kcore_hdr.c_seghdrsize;
450
451	/*
452	 * Read the CPU segment DATA.
453	 */
454	kd->cpu_dsize = cpu_hdr.c_size;
455	kd->cpu_data = _kvm_malloc(kd, cpu_hdr.c_size);
456	if (kd->cpu_data == NULL)
457		goto fail;
458	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size, (off_t)offset);
459	if (sz != cpu_hdr.c_size)
460		goto fail;
461	offset += cpu_hdr.c_size;
462
463	/*
464	 * Read the next segment header: data segment
465	 */
466	sz = Pread(kd, kd->pmfd, &mem_hdr, sizeof(mem_hdr), (off_t)offset);
467	if (sz != sizeof(mem_hdr))
468		goto fail;
469	offset += kcore_hdr.c_seghdrsize;
470
471	if ((CORE_GETMAGIC(mem_hdr) != KCORESEG_MAGIC) ||
472	    (CORE_GETFLAG(mem_hdr) != CORE_DATA))
473		goto fail;
474
475	kd->dump_off = offset;
476	return (0);
477
478fail:
479	if (kd->kcore_hdr != NULL) {
480		free(kd->kcore_hdr);
481		kd->kcore_hdr = NULL;
482	}
483	if (kd->cpu_data != NULL) {
484		free(kd->cpu_data);
485		kd->cpu_data = NULL;
486		kd->cpu_dsize = 0;
487	}
488	return (-1);
489}
490
491/*
492 * The format while on the dump device is: (new format)
493 *	kcore_seg_t cpu_hdr;
494 *	(opaque)    cpu_data; (size is cpu_hdr.c_size)
495 *	kcore_seg_t mem_hdr;
496 *	(memory)    mem_data; (size is mem_hdr.c_size)
497 */
498int
499kvm_dump_mkheader(kvm_t *kd, off_t dump_off)
500{
501	kcore_seg_t	cpu_hdr;
502	size_t hdr_size;
503	ssize_t sz;
504
505	if (kd->kcore_hdr != NULL) {
506	    _kvm_err(kd, kd->program, "already has a dump header");
507	    return (-1);
508	}
509	if (ISALIVE(kd)) {
510		_kvm_err(kd, kd->program, "don't use on live kernel");
511		return (-1);
512	}
513
514	/*
515	 * Validate new format crash dump
516	 */
517	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), dump_off);
518	if (sz != sizeof(cpu_hdr))
519		return (-1);
520	if ((CORE_GETMAGIC(cpu_hdr) != KCORE_MAGIC)
521		|| (CORE_GETMID(cpu_hdr) != MID_MACHINE)) {
522		_kvm_err(kd, 0, "invalid magic in cpu_hdr");
523		return (0);
524	}
525	hdr_size = ALIGN(sizeof(cpu_hdr));
526
527	/*
528	 * Read the CPU segment.
529	 */
530	kd->cpu_dsize = cpu_hdr.c_size;
531	kd->cpu_data = _kvm_malloc(kd, kd->cpu_dsize);
532	if (kd->cpu_data == NULL)
533		goto fail;
534	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size,
535	    dump_off + hdr_size);
536	if (sz != cpu_hdr.c_size)
537		goto fail;
538	hdr_size += kd->cpu_dsize;
539
540	/*
541	 * Leave phys mem pointer at beginning of memory data
542	 */
543	kd->dump_off = dump_off + hdr_size;
544	if (Lseek(kd, kd->pmfd, kd->dump_off, SEEK_SET) == -1)
545		goto fail;
546
547	/*
548	 * Create a kcore_hdr.
549	 */
550	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr_t));
551	if (kd->kcore_hdr == NULL)
552		goto fail;
553
554	kd->kcore_hdr->c_hdrsize    = ALIGN(sizeof(kcore_hdr_t));
555	kd->kcore_hdr->c_seghdrsize = ALIGN(sizeof(kcore_seg_t));
556	kd->kcore_hdr->c_nseg       = 2;
557	CORE_SETMAGIC(*(kd->kcore_hdr), KCORE_MAGIC, MID_MACHINE,0);
558
559	/*
560	 * Now that we have a valid header, enable translations.
561	 */
562	if (_kvm_initvtop(kd) == 0)
563		/* Success */
564		return (hdr_size);
565
566fail:
567	if (kd->kcore_hdr != NULL) {
568		free(kd->kcore_hdr);
569		kd->kcore_hdr = NULL;
570	}
571	if (kd->cpu_data != NULL) {
572		free(kd->cpu_data);
573		kd->cpu_data = NULL;
574		kd->cpu_dsize = 0;
575	}
576	return (-1);
577}
578
579static int
580clear_gap(kvm_t *kd, bool (*write_buf)(void *, const void *, size_t),
581    void *cookie, size_t size)
582{
583	char buf[1024];
584	size_t len;
585
586	(void)memset(buf, 0, size > sizeof(buf) ? sizeof(buf) : size);
587
588	while (size > 0) {
589		len = size > sizeof(buf) ? sizeof(buf) : size;
590		if (!(*write_buf)(cookie, buf, len)) {
591			_kvm_syserr(kd, kd->program, "clear_gap");
592			return -1;
593		}
594		size -= len;
595	}
596
597	return 0;
598}
599
600/*
601 * Write the dump header by calling write_buf with cookie as first argument.
602 */
603int
604kvm_dump_header(kvm_t *kd, bool (*write_buf)(void *, const void *, size_t),
605    void *cookie, int dumpsize)
606{
607	kcore_seg_t	seghdr;
608	long		offset;
609	size_t		gap;
610
611	if (kd->kcore_hdr == NULL || kd->cpu_data == NULL) {
612		_kvm_err(kd, kd->program, "no valid dump header(s)");
613		return (-1);
614	}
615
616	/*
617	 * Write the generic header
618	 */
619	offset = 0;
620	if (!(*write_buf)(cookie, kd->kcore_hdr, sizeof(kcore_hdr_t))) {
621		_kvm_syserr(kd, kd->program, "kvm_dump_header");
622		return (-1);
623	}
624	offset += kd->kcore_hdr->c_hdrsize;
625	gap     = kd->kcore_hdr->c_hdrsize - sizeof(kcore_hdr_t);
626	if (clear_gap(kd, write_buf, cookie, gap) == -1)
627		return (-1);
628
629	/*
630	 * Write the CPU header
631	 */
632	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_CPU);
633	seghdr.c_size = ALIGN(kd->cpu_dsize);
634	if (!(*write_buf)(cookie, &seghdr, sizeof(seghdr))) {
635		_kvm_syserr(kd, kd->program, "kvm_dump_header");
636		return (-1);
637	}
638	offset += kd->kcore_hdr->c_seghdrsize;
639	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
640	if (clear_gap(kd, write_buf, cookie, gap) == -1)
641		return (-1);
642
643	if (!(*write_buf)(cookie, kd->cpu_data, kd->cpu_dsize)) {
644		_kvm_syserr(kd, kd->program, "kvm_dump_header");
645		return (-1);
646	}
647	offset += seghdr.c_size;
648	gap     = seghdr.c_size - kd->cpu_dsize;
649	if (clear_gap(kd, write_buf, cookie, gap) == -1)
650		return (-1);
651
652	/*
653	 * Write the actual dump data segment header
654	 */
655	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_DATA);
656	seghdr.c_size = dumpsize;
657	if (!(*write_buf)(cookie, &seghdr, sizeof(seghdr))) {
658		_kvm_syserr(kd, kd->program, "kvm_dump_header");
659		return (-1);
660	}
661	offset += kd->kcore_hdr->c_seghdrsize;
662	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
663	if (clear_gap(kd, write_buf, cookie, gap) == -1)
664		return (-1);
665
666	return (int)offset;
667}
668
669static bool
670kvm_dump_header_stdio(void *cookie, const void *buf, size_t len)
671{
672	return fwrite(buf, len, 1, (FILE *)cookie) == 1;
673}
674
675int
676kvm_dump_wrtheader(kvm_t *kd, FILE *fp, int dumpsize)
677{
678	return kvm_dump_header(kd, kvm_dump_header_stdio, fp, dumpsize);
679}
680
681kvm_t *
682kvm_openfiles(const char *uf, const char *mf, const char *sf,
683    int flag, char *errout)
684{
685	kvm_t *kd;
686
687	if ((kd = malloc(sizeof(*kd))) == NULL) {
688		(void)strlcpy(errout, strerror(errno), _POSIX2_LINE_MAX);
689		return (0);
690	}
691	kd->program = 0;
692	return (_kvm_open(kd, uf, mf, sf, flag, errout));
693}
694
695kvm_t *
696kvm_open(const char *uf, const char *mf, const char *sf, int flag,
697    const char *program)
698{
699	kvm_t *kd;
700
701	if ((kd = malloc(sizeof(*kd))) == NULL) {
702		(void)fprintf(stderr, "%s: %s\n",
703		    program ? program : getprogname(), strerror(errno));
704		return (0);
705	}
706	kd->program = program;
707	return (_kvm_open(kd, uf, mf, sf, flag, NULL));
708}
709
710int
711kvm_close(kvm_t *kd)
712{
713	int error = 0;
714
715	if (kd->pmfd >= 0)
716		error |= close(kd->pmfd);
717	if (kd->vmfd >= 0)
718		error |= close(kd->vmfd);
719	if (kd->nlfd >= 0)
720		error |= close(kd->nlfd);
721	if (kd->swfd >= 0)
722		error |= close(kd->swfd);
723	if (kd->vmst)
724		_kvm_freevtop(kd);
725	kd->cpu_dsize = 0;
726	if (kd->cpu_data != NULL)
727		free(kd->cpu_data);
728	if (kd->kcore_hdr != NULL)
729		free(kd->kcore_hdr);
730	if (kd->procbase != 0)
731		free(kd->procbase);
732	if (kd->procbase2 != 0)
733		free(kd->procbase2);
734	if (kd->lwpbase != 0)
735		free(kd->lwpbase);
736	if (kd->swapspc != 0)
737		free(kd->swapspc);
738	if (kd->argspc != 0)
739		free(kd->argspc);
740	if (kd->argbuf != 0)
741		free(kd->argbuf);
742	if (kd->argv != 0)
743		free(kd->argv);
744	if (kd->iobuf != 0)
745		free(kd->iobuf);
746	free(kd);
747
748	return (error);
749}
750
751int
752kvm_nlist(kvm_t *kd, struct nlist *nl)
753{
754	int rv, nlfd;
755
756	/*
757	 * kd->nlfd might be negative when we get here, and in that
758	 * case that means that we're using /dev/ksyms.
759	 * So open it again, just for the time we retrieve the list.
760	 */
761	if (kd->nlfd < 0) {
762		nlfd = open(_PATH_KSYMS, O_RDONLY | O_CLOEXEC, 0);
763		if (nlfd < 0) {
764			_kvm_err(kd, 0, "failed to open %s", _PATH_KSYMS);
765			return (nlfd);
766		}
767	} else
768		nlfd = kd->nlfd;
769
770	/*
771	 * Call the nlist(3) routines to retrieve the given namelist.
772	 */
773	rv = __fdnlist(nlfd, nl);
774
775	if (rv == -1)
776		_kvm_err(kd, 0, "bad namelist");
777
778	if (kd->nlfd < 0)
779		close(nlfd);
780
781	return (rv);
782}
783
784int
785kvm_dump_inval(kvm_t *kd)
786{
787	struct nlist	nl[2];
788	paddr_t		pa;
789	size_t		dsize;
790	off_t		doff;
791	void		*newbuf;
792
793	if (ISALIVE(kd)) {
794		_kvm_err(kd, kd->program, "clearing dump on live kernel");
795		return (-1);
796	}
797	nl[0].n_name = "_dumpmag";
798	nl[1].n_name = NULL;
799
800	if (kvm_nlist(kd, nl) == -1) {
801		_kvm_err(kd, 0, "bad namelist");
802		return (-1);
803	}
804	if (_kvm_kvatop(kd, (vaddr_t)nl[0].n_value, &pa) == 0)
805		return (-1);
806
807	errno = 0;
808	dsize = MAX(kd->fdalign, sizeof(u_long));
809	if (kd->iobufsz < dsize) {
810		newbuf = realloc(kd->iobuf, dsize);
811		if (newbuf == NULL) {
812			_kvm_syserr(kd, 0, "cannot allocate I/O buffer");
813			return (-1);
814		}
815		kd->iobuf = newbuf;
816		kd->iobufsz = dsize;
817	}
818	memset(kd->iobuf, 0, dsize);
819	doff = _kvm_pa2off(kd, pa);
820	doff -= doff % kd->fdalign;
821	if (pwrite(kd->pmfd, kd->iobuf, dsize, doff) == -1) {
822		_kvm_syserr(kd, 0, "cannot invalidate dump - pwrite");
823		return (-1);
824	}
825	return (0);
826}
827
828ssize_t
829kvm_read(kvm_t *kd, u_long kva, void *buf, size_t len)
830{
831	int cc;
832	void *cp;
833
834	if (ISKMEM(kd)) {
835		/*
836		 * We're using /dev/kmem.  Just read straight from the
837		 * device and let the active kernel do the address translation.
838		 */
839		errno = 0;
840		cc = _kvm_pread(kd, kd->vmfd, buf, len, (off_t)kva);
841		if (cc < 0) {
842			_kvm_syserr(kd, 0, "kvm_read");
843			return (-1);
844		} else if (cc < len)
845			_kvm_err(kd, kd->program, "short read");
846		return (cc);
847	} else if (ISSYSCTL(kd)) {
848		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
849		    "can't use kvm_read");
850		return (-1);
851	} else {
852		if ((kd->kcore_hdr == NULL) || (kd->cpu_data == NULL)) {
853			_kvm_err(kd, kd->program, "no valid dump header");
854			return (-1);
855		}
856		cp = buf;
857		while (len > 0) {
858			paddr_t	pa;
859			off_t	foff;
860
861			cc = _kvm_kvatop(kd, (vaddr_t)kva, &pa);
862			if (cc == 0)
863				return (-1);
864			if (cc > len)
865				cc = len;
866			foff = _kvm_pa2off(kd, pa);
867			errno = 0;
868			cc = _kvm_pread(kd, kd->pmfd, cp, (size_t)cc, foff);
869			if (cc < 0) {
870				_kvm_syserr(kd, kd->program, "kvm_read");
871				break;
872			}
873			/*
874			 * If kvm_kvatop returns a bogus value or our core
875			 * file is truncated, we might wind up seeking beyond
876			 * the end of the core file in which case the read will
877			 * return 0 (EOF).
878			 */
879			if (cc == 0)
880				break;
881			cp = (char *)cp + cc;
882			kva += cc;
883			len -= cc;
884		}
885		return ((char *)cp - (char *)buf);
886	}
887	/* NOTREACHED */
888}
889
890ssize_t
891kvm_write(kvm_t *kd, u_long kva, const void *buf, size_t len)
892{
893	int cc;
894
895	if (ISKMEM(kd)) {
896		/*
897		 * Just like kvm_read, only we write.
898		 */
899		errno = 0;
900		cc = pwrite(kd->vmfd, buf, len, (off_t)kva);
901		if (cc < 0) {
902			_kvm_syserr(kd, 0, "kvm_write");
903			return (-1);
904		} else if (cc < len)
905			_kvm_err(kd, kd->program, "short write");
906		return (cc);
907	} else if (ISSYSCTL(kd)) {
908		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
909		    "can't use kvm_write");
910		return (-1);
911	} else {
912		_kvm_err(kd, kd->program,
913		    "kvm_write not implemented for dead kernels");
914		return (-1);
915	}
916	/* NOTREACHED */
917}
918