xref: /netbsd-src/lib/libkvm/kvm.c (revision 33881f779a77dce6440bdc44610d94de75bebefe)
1 /*	$NetBSD: kvm.c,v 1.105 2020/03/08 00:06:42 chs 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
39 static char sccsid[] = "@(#)kvm.c	8.2 (Berkeley) 2/13/94";
40 #else
41 __RCSID("$NetBSD: kvm.c,v 1.105 2020/03/08 00:06:42 chs Exp $");
42 #endif
43 #endif /* LIBC_SCCS and not lint */
44 
45 #include <sys/param.h>
46 #include <sys/lwp.h>
47 #include <sys/proc.h>
48 #include <sys/ioctl.h>
49 #include <sys/stat.h>
50 #include <sys/sysctl.h>
51 
52 #include <sys/core.h>
53 #include <sys/exec.h>
54 #include <sys/kcore.h>
55 #include <sys/ksyms.h>
56 #include <sys/types.h>
57 
58 #include <uvm/uvm_extern.h>
59 
60 #include <machine/cpu.h>
61 
62 #include <ctype.h>
63 #include <errno.h>
64 #include <fcntl.h>
65 #include <limits.h>
66 #include <nlist.h>
67 #include <paths.h>
68 #include <stdarg.h>
69 #include <stdio.h>
70 #include <stdlib.h>
71 #include <string.h>
72 #include <unistd.h>
73 #include <kvm.h>
74 
75 #include "kvm_private.h"
76 
77 static int	_kvm_get_header(kvm_t *);
78 static kvm_t	*_kvm_open(kvm_t *, const char *, const char *,
79 		    const char *, int, char *);
80 static int	clear_gap(kvm_t *, bool (*)(void *, const void *, size_t),
81 		    void *, size_t);
82 static off_t	Lseek(kvm_t *, int, off_t, int);
83 static ssize_t	Pread(kvm_t *, int, void *, size_t, off_t);
84 
85 char *
86 kvm_geterr(kvm_t *kd)
87 {
88 	return (kd->errbuf);
89 }
90 
91 const char *
92 kvm_getkernelname(kvm_t *kd)
93 {
94 	return kd->kernelname;
95 }
96 
97 /*
98  * Report an error using printf style arguments.  "program" is kd->program
99  * on hard errors, and 0 on soft errors, so that under sun error emulation,
100  * only hard errors are printed out (otherwise, programs like gdb will
101  * generate tons of error messages when trying to access bogus pointers).
102  */
103 void
104 _kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
105 {
106 	va_list ap;
107 
108 	va_start(ap, fmt);
109 	if (program != NULL) {
110 		(void)fprintf(stderr, "%s: ", program);
111 		(void)vfprintf(stderr, fmt, ap);
112 		(void)fputc('\n', stderr);
113 	} else
114 		(void)vsnprintf(kd->errbuf,
115 		    sizeof(kd->errbuf), fmt, ap);
116 
117 	va_end(ap);
118 }
119 
120 void
121 _kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
122 {
123 	va_list ap;
124 	size_t n;
125 
126 	va_start(ap, fmt);
127 	if (program != NULL) {
128 		(void)fprintf(stderr, "%s: ", program);
129 		(void)vfprintf(stderr, fmt, ap);
130 		(void)fprintf(stderr, ": %s\n", strerror(errno));
131 	} else {
132 		char *cp = kd->errbuf;
133 
134 		(void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
135 		n = strlen(cp);
136 		(void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
137 		    strerror(errno));
138 	}
139 	va_end(ap);
140 }
141 
142 void *
143 _kvm_malloc(kvm_t *kd, size_t n)
144 {
145 	void *p;
146 
147 	if ((p = malloc(n)) == NULL)
148 		_kvm_err(kd, kd->program, "%s", strerror(errno));
149 	return (p);
150 }
151 
152 /*
153  * Wrapper around the lseek(2) system call; calls _kvm_syserr() for us
154  * in the event of emergency.
155  */
156 static off_t
157 Lseek(kvm_t *kd, int fd, off_t offset, int whence)
158 {
159 	off_t off;
160 
161 	errno = 0;
162 
163 	if ((off = lseek(fd, offset, whence)) == -1 && errno != 0) {
164 		_kvm_syserr(kd, kd->program, "Lseek");
165 		return ((off_t)-1);
166 	}
167 	return (off);
168 }
169 
170 ssize_t
171 _kvm_pread(kvm_t *kd, int fd, void *buf, size_t size, off_t off)
172 {
173 	ptrdiff_t moff;
174 	void *newbuf;
175 	size_t dsize;
176 	ssize_t rv;
177 	off_t doff;
178 
179 	/* If aligned nothing to do. */
180  	if (((off % kd->fdalign) | (size % kd->fdalign)) == 0) {
181 		return pread(fd, buf, size, off);
182  	}
183 
184 	/*
185 	 * Otherwise must buffer.  We can't tolerate short reads in this
186 	 * case (lazy bum).
187 	 */
188 	moff = (ptrdiff_t)off % kd->fdalign;
189 	doff = off - moff;
190 	dsize = moff + size + kd->fdalign - 1;
191 	dsize -= dsize % kd->fdalign;
192 	if (kd->iobufsz < dsize) {
193 		newbuf = realloc(kd->iobuf, dsize);
194 		if (newbuf == NULL) {
195 			_kvm_syserr(kd, 0, "cannot allocate I/O buffer");
196 			return (-1);
197 		}
198 		kd->iobuf = newbuf;
199 		kd->iobufsz = dsize;
200 	}
201 	rv = pread(fd, kd->iobuf, dsize, doff);
202 	if (rv < size + moff)
203 		return -1;
204 	memcpy(buf, kd->iobuf + moff, size);
205 	return size;
206 }
207 
208 /*
209  * Wrapper around the pread(2) system call; calls _kvm_syserr() for us
210  * in the event of emergency.
211  */
212 static ssize_t
213 Pread(kvm_t *kd, int fd, void *buf, size_t nbytes, off_t offset)
214 {
215 	ssize_t rv;
216 
217 	errno = 0;
218 
219 	if ((rv = _kvm_pread(kd, fd, buf, nbytes, offset)) != nbytes &&
220 	    errno != 0)
221 		_kvm_syserr(kd, kd->program, "Pread");
222 	return (rv);
223 }
224 
225 static kvm_t *
226 _kvm_open(kvm_t *kd, const char *uf, const char *mf, const char *sf, int flag,
227     char *errout)
228 {
229 	struct stat st;
230 	int ufgiven;
231 
232 	kd->pmfd = -1;
233 	kd->vmfd = -1;
234 	kd->swfd = -1;
235 	kd->nlfd = -1;
236 	kd->alive = KVM_ALIVE_DEAD;
237 	kd->procbase = NULL;
238 	kd->procbase_len = 0;
239 	kd->procbase2 = NULL;
240 	kd->procbase2_len = 0;
241 	kd->lwpbase = NULL;
242 	kd->lwpbase_len = 0;
243 	kd->nbpg = getpagesize();
244 	kd->swapspc = NULL;
245 	kd->argspc = NULL;
246 	kd->argspc_len = 0;
247 	kd->argbuf = NULL;
248 	kd->argv = NULL;
249 	kd->vmst = NULL;
250 	kd->vm_page_buckets = NULL;
251 	kd->kcore_hdr = NULL;
252 	kd->cpu_dsize = 0;
253 	kd->cpu_data = NULL;
254 	kd->dump_off = 0;
255 	kd->fdalign = 1;
256 	kd->iobuf = NULL;
257 	kd->iobufsz = 0;
258 	kd->errbuf[0] = '\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 
328 	if ((kd->pmfd = open(mf, flag | O_CLOEXEC, 0)) < 0) {
329 		_kvm_syserr(kd, kd->program, "%s", mf);
330 		goto failed;
331 	}
332 	if (fstat(kd->pmfd, &st) < 0) {
333 		_kvm_syserr(kd, kd->program, "%s", mf);
334 		goto failed;
335 	}
336 	if (S_ISCHR(st.st_mode) && strcmp(mf, _PATH_MEM) == 0) {
337 		/*
338 		 * If this is /dev/mem, open kmem too.  (Maybe we should
339 		 * make it work for either /dev/mem or /dev/kmem -- in either
340 		 * case you're working with a live kernel.)
341 		 */
342 		if ((kd->vmfd = open(_PATH_KMEM, flag | O_CLOEXEC, 0)) < 0) {
343 			_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
344 			goto failed;
345 		}
346 		kd->alive = KVM_ALIVE_FILES;
347 		if ((kd->swfd = open(sf, flag | O_CLOEXEC, 0)) < 0) {
348 			if (errno != ENXIO) {
349 				_kvm_syserr(kd, kd->program, "%s", sf);
350 				goto failed;
351 			}
352 			/* swap is not configured?  not fatal */
353 		}
354 	} else {
355 		if (S_ISCHR(st.st_mode)) {
356 			kd->fdalign = DEV_BSIZE;
357 		} else {
358 			kd->fdalign = 1;
359 		}
360 
361 		/*
362 		 * This is a crash dump.
363 		 * Initialize the virtual address translation machinery.
364 		 *
365 		 * If there is no valid core header, fail silently here.
366 		 * The address translations however will fail without
367 		 * header. Things can be made to run by calling
368 		 * kvm_dump_mkheader() before doing any translation.
369 		 */
370 		if (_kvm_get_header(kd) == 0) {
371 			if (_kvm_initvtop(kd) < 0)
372 				goto failed;
373 		}
374 	}
375 	return (kd);
376 failed:
377 	/*
378 	 * Copy out the error if doing sane error semantics.
379 	 */
380 	if (errout != 0)
381 		(void)strlcpy(errout, kd->errbuf, _POSIX2_LINE_MAX);
382 	(void)kvm_close(kd);
383 	return (0);
384 }
385 
386 /*
387  * The kernel dump file (from savecore) contains:
388  *    kcore_hdr_t kcore_hdr;
389  *    kcore_seg_t cpu_hdr;
390  *    (opaque)    cpu_data; (size is cpu_hdr.c_size)
391  *	  kcore_seg_t mem_hdr;
392  *    (memory)    mem_data; (size is mem_hdr.c_size)
393  *
394  * Note: khdr is padded to khdr.c_hdrsize;
395  * cpu_hdr and mem_hdr are padded to khdr.c_seghdrsize
396  */
397 static int
398 _kvm_get_header(kvm_t *kd)
399 {
400 	kcore_hdr_t	kcore_hdr;
401 	kcore_seg_t	cpu_hdr;
402 	kcore_seg_t	mem_hdr;
403 	size_t		offset;
404 	ssize_t		sz;
405 
406 	/*
407 	 * Read the kcore_hdr_t
408 	 */
409 	sz = Pread(kd, kd->pmfd, &kcore_hdr, sizeof(kcore_hdr), (off_t)0);
410 	if (sz != sizeof(kcore_hdr))
411 		return (-1);
412 
413 	/*
414 	 * Currently, we only support dump-files made by the current
415 	 * architecture...
416 	 */
417 	if ((CORE_GETMAGIC(kcore_hdr) != KCORE_MAGIC) ||
418 	    (CORE_GETMID(kcore_hdr) != MID_MACHINE))
419 		return (-1);
420 
421 	/*
422 	 * Currently, we only support exactly 2 segments: cpu-segment
423 	 * and data-segment in exactly that order.
424 	 */
425 	if (kcore_hdr.c_nseg != 2)
426 		return (-1);
427 
428 	/*
429 	 * Save away the kcore_hdr.  All errors after this
430 	 * should do a to "goto fail" to deallocate things.
431 	 */
432 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr));
433 	memcpy(kd->kcore_hdr, &kcore_hdr, sizeof(kcore_hdr));
434 	offset = kcore_hdr.c_hdrsize;
435 
436 	/*
437 	 * Read the CPU segment header
438 	 */
439 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), (off_t)offset);
440 	if (sz != sizeof(cpu_hdr))
441 		goto fail;
442 	if ((CORE_GETMAGIC(cpu_hdr) != KCORESEG_MAGIC) ||
443 	    (CORE_GETFLAG(cpu_hdr) != CORE_CPU))
444 		goto fail;
445 	offset += kcore_hdr.c_seghdrsize;
446 
447 	/*
448 	 * Read the CPU segment DATA.
449 	 */
450 	kd->cpu_dsize = cpu_hdr.c_size;
451 	kd->cpu_data = _kvm_malloc(kd, cpu_hdr.c_size);
452 	if (kd->cpu_data == NULL)
453 		goto fail;
454 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size, (off_t)offset);
455 	if (sz != cpu_hdr.c_size)
456 		goto fail;
457 	offset += cpu_hdr.c_size;
458 
459 	/*
460 	 * Read the next segment header: data segment
461 	 */
462 	sz = Pread(kd, kd->pmfd, &mem_hdr, sizeof(mem_hdr), (off_t)offset);
463 	if (sz != sizeof(mem_hdr))
464 		goto fail;
465 	offset += kcore_hdr.c_seghdrsize;
466 
467 	if ((CORE_GETMAGIC(mem_hdr) != KCORESEG_MAGIC) ||
468 	    (CORE_GETFLAG(mem_hdr) != CORE_DATA))
469 		goto fail;
470 
471 	kd->dump_off = offset;
472 	return (0);
473 
474 fail:
475 	if (kd->kcore_hdr != NULL) {
476 		free(kd->kcore_hdr);
477 		kd->kcore_hdr = NULL;
478 	}
479 	if (kd->cpu_data != NULL) {
480 		free(kd->cpu_data);
481 		kd->cpu_data = NULL;
482 		kd->cpu_dsize = 0;
483 	}
484 	return (-1);
485 }
486 
487 /*
488  * The format while on the dump device is: (new format)
489  *	kcore_seg_t cpu_hdr;
490  *	(opaque)    cpu_data; (size is cpu_hdr.c_size)
491  *	kcore_seg_t mem_hdr;
492  *	(memory)    mem_data; (size is mem_hdr.c_size)
493  */
494 int
495 kvm_dump_mkheader(kvm_t *kd, off_t dump_off)
496 {
497 	kcore_seg_t	cpu_hdr;
498 	size_t hdr_size;
499 	ssize_t sz;
500 
501 	if (kd->kcore_hdr != NULL) {
502 	    _kvm_err(kd, kd->program, "already has a dump header");
503 	    return (-1);
504 	}
505 	if (ISALIVE(kd)) {
506 		_kvm_err(kd, kd->program, "don't use on live kernel");
507 		return (-1);
508 	}
509 
510 	/*
511 	 * Validate new format crash dump
512 	 */
513 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), dump_off);
514 	if (sz != sizeof(cpu_hdr)) {
515 		if (sz == -1)
516 			_kvm_err(kd, 0, "read %zx bytes at offset %"PRIx64
517 			    " for cpu_hdr failed: %s", sizeof(cpu_hdr),
518 			    dump_off, strerror(errno));
519 		else
520 			_kvm_err(kd, 0, "read %zx bytes at offset %"PRIx64
521 			    " for cpu_hdr instead of requested %zu",
522 			    sz, dump_off, sizeof(cpu_hdr));
523 		return (-1);
524 	}
525 	if ((CORE_GETMAGIC(cpu_hdr) != KCORE_MAGIC)
526 		|| (CORE_GETMID(cpu_hdr) != MID_MACHINE)) {
527 		_kvm_err(kd, 0, "invalid magic in cpu_hdr");
528 		return (0);
529 	}
530 	hdr_size = ALIGN(sizeof(cpu_hdr));
531 
532 	/*
533 	 * Read the CPU segment.
534 	 */
535 	kd->cpu_dsize = cpu_hdr.c_size;
536 	kd->cpu_data = _kvm_malloc(kd, kd->cpu_dsize);
537 	if (kd->cpu_data == NULL) {
538 		_kvm_err(kd, kd->program, "no cpu_data");
539 		goto fail;
540 	}
541 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size,
542 	    dump_off + hdr_size);
543 	if (sz != cpu_hdr.c_size) {
544 		_kvm_err(kd, kd->program, "size %zu != cpu_hdr.csize %"PRIu32,
545 		    sz, cpu_hdr.c_size);
546 		goto fail;
547 	}
548 	hdr_size += kd->cpu_dsize;
549 
550 	/*
551 	 * Leave phys mem pointer at beginning of memory data
552 	 */
553 	kd->dump_off = dump_off + hdr_size;
554 	if (Lseek(kd, kd->pmfd, kd->dump_off, SEEK_SET) == -1) {
555 		_kvm_err(kd, kd->program, "failed to seek to %" PRId64,
556 		    (int64_t)kd->dump_off);
557 		goto fail;
558 	}
559 
560 	/*
561 	 * Create a kcore_hdr.
562 	 */
563 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr_t));
564 	if (kd->kcore_hdr == NULL) {
565 		_kvm_err(kd, kd->program, "failed to allocate header");
566 		goto fail;
567 	}
568 
569 	kd->kcore_hdr->c_hdrsize    = ALIGN(sizeof(kcore_hdr_t));
570 	kd->kcore_hdr->c_seghdrsize = ALIGN(sizeof(kcore_seg_t));
571 	kd->kcore_hdr->c_nseg       = 2;
572 	CORE_SETMAGIC(*(kd->kcore_hdr), KCORE_MAGIC, MID_MACHINE,0);
573 
574 	/*
575 	 * Now that we have a valid header, enable translations.
576 	 */
577 	if (_kvm_initvtop(kd) == 0)
578 		/* Success */
579 		return (hdr_size);
580 
581 fail:
582 	if (kd->kcore_hdr != NULL) {
583 		free(kd->kcore_hdr);
584 		kd->kcore_hdr = NULL;
585 	}
586 	if (kd->cpu_data != NULL) {
587 		free(kd->cpu_data);
588 		kd->cpu_data = NULL;
589 		kd->cpu_dsize = 0;
590 	}
591 	return (-1);
592 }
593 
594 static int
595 clear_gap(kvm_t *kd, bool (*write_buf)(void *, const void *, size_t),
596     void *cookie, size_t size)
597 {
598 	char buf[1024];
599 	size_t len;
600 
601 	(void)memset(buf, 0, size > sizeof(buf) ? sizeof(buf) : size);
602 
603 	while (size > 0) {
604 		len = size > sizeof(buf) ? sizeof(buf) : size;
605 		if (!(*write_buf)(cookie, buf, len)) {
606 			_kvm_syserr(kd, kd->program, "clear_gap");
607 			return -1;
608 		}
609 		size -= len;
610 	}
611 
612 	return 0;
613 }
614 
615 /*
616  * Write the dump header by calling write_buf with cookie as first argument.
617  */
618 int
619 kvm_dump_header(kvm_t *kd, bool (*write_buf)(void *, const void *, size_t),
620     void *cookie, int dumpsize)
621 {
622 	kcore_seg_t	seghdr;
623 	long		offset;
624 	size_t		gap;
625 
626 	if (kd->kcore_hdr == NULL || kd->cpu_data == NULL) {
627 		_kvm_err(kd, kd->program, "no valid dump header(s)");
628 		return (-1);
629 	}
630 
631 	/*
632 	 * Write the generic header
633 	 */
634 	offset = 0;
635 	if (!(*write_buf)(cookie, kd->kcore_hdr, sizeof(kcore_hdr_t))) {
636 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
637 		return (-1);
638 	}
639 	offset += kd->kcore_hdr->c_hdrsize;
640 	gap     = kd->kcore_hdr->c_hdrsize - sizeof(kcore_hdr_t);
641 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
642 		return (-1);
643 
644 	/*
645 	 * Write the CPU header
646 	 */
647 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_CPU);
648 	seghdr.c_size = ALIGN(kd->cpu_dsize);
649 	if (!(*write_buf)(cookie, &seghdr, sizeof(seghdr))) {
650 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
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, write_buf, cookie, gap) == -1)
656 		return (-1);
657 
658 	if (!(*write_buf)(cookie, kd->cpu_data, kd->cpu_dsize)) {
659 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
660 		return (-1);
661 	}
662 	offset += seghdr.c_size;
663 	gap     = seghdr.c_size - kd->cpu_dsize;
664 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
665 		return (-1);
666 
667 	/*
668 	 * Write the actual dump data segment header
669 	 */
670 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_DATA);
671 	seghdr.c_size = dumpsize;
672 	if (!(*write_buf)(cookie, &seghdr, sizeof(seghdr))) {
673 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
674 		return (-1);
675 	}
676 	offset += kd->kcore_hdr->c_seghdrsize;
677 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
678 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
679 		return (-1);
680 
681 	return (int)offset;
682 }
683 
684 static bool
685 kvm_dump_header_stdio(void *cookie, const void *buf, size_t len)
686 {
687 	return fwrite(buf, len, 1, (FILE *)cookie) == 1;
688 }
689 
690 int
691 kvm_dump_wrtheader(kvm_t *kd, FILE *fp, int dumpsize)
692 {
693 	return kvm_dump_header(kd, kvm_dump_header_stdio, fp, dumpsize);
694 }
695 
696 kvm_t *
697 kvm_openfiles(const char *uf, const char *mf, const char *sf,
698     int flag, char *errout)
699 {
700 	kvm_t *kd;
701 
702 	if ((kd = malloc(sizeof(*kd))) == NULL) {
703 		(void)strlcpy(errout, strerror(errno), _POSIX2_LINE_MAX);
704 		return (0);
705 	}
706 	kd->program = 0;
707 	return (_kvm_open(kd, uf, mf, sf, flag, errout));
708 }
709 
710 kvm_t *
711 kvm_open(const char *uf, const char *mf, const char *sf, int flag,
712     const char *program)
713 {
714 	kvm_t *kd;
715 
716 	if ((kd = malloc(sizeof(*kd))) == NULL) {
717 		(void)fprintf(stderr, "%s: %s\n",
718 		    program ? program : getprogname(), strerror(errno));
719 		return (0);
720 	}
721 	kd->program = program;
722 	return (_kvm_open(kd, uf, mf, sf, flag, NULL));
723 }
724 
725 int
726 kvm_close(kvm_t *kd)
727 {
728 	int error = 0;
729 
730 	if (kd->pmfd >= 0)
731 		error |= close(kd->pmfd);
732 	if (kd->vmfd >= 0)
733 		error |= close(kd->vmfd);
734 	if (kd->nlfd >= 0)
735 		error |= close(kd->nlfd);
736 	if (kd->swfd >= 0)
737 		error |= close(kd->swfd);
738 	if (kd->vmst)
739 		_kvm_freevtop(kd);
740 	kd->cpu_dsize = 0;
741 	if (kd->cpu_data != NULL)
742 		free(kd->cpu_data);
743 	if (kd->kcore_hdr != NULL)
744 		free(kd->kcore_hdr);
745 	if (kd->procbase != 0)
746 		free(kd->procbase);
747 	if (kd->procbase2 != 0)
748 		free(kd->procbase2);
749 	if (kd->lwpbase != 0)
750 		free(kd->lwpbase);
751 	if (kd->swapspc != 0)
752 		free(kd->swapspc);
753 	if (kd->argspc != 0)
754 		free(kd->argspc);
755 	if (kd->argbuf != 0)
756 		free(kd->argbuf);
757 	if (kd->argv != 0)
758 		free(kd->argv);
759 	if (kd->iobuf != 0)
760 		free(kd->iobuf);
761 	free(kd);
762 
763 	return (error);
764 }
765 
766 int
767 kvm_nlist(kvm_t *kd, struct nlist *nl)
768 {
769 	int rv;
770 
771 	/*
772 	 * Call the nlist(3) routines to retrieve the given namelist.
773 	 */
774 	rv = __fdnlist(kd->nlfd, nl);
775 
776 	if (rv == -1)
777 		_kvm_err(kd, 0, "bad namelist");
778 
779 	return (rv);
780 }
781 
782 int
783 kvm_dump_inval(kvm_t *kd)
784 {
785 	struct nlist	nl[2];
786 	paddr_t		pa;
787 	size_t		dsize;
788 	off_t		doff;
789 	void		*newbuf;
790 
791 	if (ISALIVE(kd)) {
792 		_kvm_err(kd, kd->program, "clearing dump on live kernel");
793 		return (-1);
794 	}
795 	nl[0].n_name = "_dumpmag";
796 	nl[1].n_name = NULL;
797 
798 	if (kvm_nlist(kd, nl) == -1) {
799 		_kvm_err(kd, 0, "bad namelist");
800 		return (-1);
801 	}
802 	if (_kvm_kvatop(kd, (vaddr_t)nl[0].n_value, &pa) == 0)
803 		return (-1);
804 
805 	errno = 0;
806 	dsize = MAX(kd->fdalign, sizeof(u_long));
807 	if (kd->iobufsz < dsize) {
808 		newbuf = realloc(kd->iobuf, dsize);
809 		if (newbuf == NULL) {
810 			_kvm_syserr(kd, 0, "cannot allocate I/O buffer");
811 			return (-1);
812 		}
813 		kd->iobuf = newbuf;
814 		kd->iobufsz = dsize;
815 	}
816 	memset(kd->iobuf, 0, dsize);
817 	doff = _kvm_pa2off(kd, pa);
818 	doff -= doff % kd->fdalign;
819 	if (pwrite(kd->pmfd, kd->iobuf, dsize, doff) == -1) {
820 		_kvm_syserr(kd, 0, "cannot invalidate dump - pwrite");
821 		return (-1);
822 	}
823 	return (0);
824 }
825 
826 ssize_t
827 kvm_read(kvm_t *kd, u_long kva, void *buf, size_t len)
828 {
829 	int cc;
830 	void *cp;
831 
832 	if (ISKMEM(kd)) {
833 		/*
834 		 * We're using /dev/kmem.  Just read straight from the
835 		 * device and let the active kernel do the address translation.
836 		 */
837 		errno = 0;
838 		cc = _kvm_pread(kd, kd->vmfd, buf, len, (off_t)kva);
839 		if (cc < 0) {
840 			_kvm_syserr(kd, 0, "kvm_read");
841 			return (-1);
842 		} else if (cc < len)
843 			_kvm_err(kd, kd->program, "short read");
844 		return (cc);
845 	} else if (ISSYSCTL(kd)) {
846 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
847 		    "can't use kvm_read");
848 		return (-1);
849 	} else {
850 		if ((kd->kcore_hdr == NULL) || (kd->cpu_data == NULL)) {
851 			_kvm_err(kd, kd->program, "no valid dump header");
852 			return (-1);
853 		}
854 		cp = buf;
855 		while (len > 0) {
856 			paddr_t	pa;
857 			off_t	foff;
858 
859 			cc = _kvm_kvatop(kd, (vaddr_t)kva, &pa);
860 			if (cc == 0) {
861 				_kvm_err(kd, kd->program, "_kvm_kvatop(%lx)", kva);
862 				return (-1);
863 			}
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 
890 ssize_t
891 kvm_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