xref: /netbsd-src/lib/libkvm/kvm.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: kvm.c,v 1.103 2018/02/07 14:03:18 maxv 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 1.103 2018/02/07 14:03:18 maxv 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 
259 	if (flag & KVM_NO_FILES) {
260 		kd->alive = KVM_ALIVE_SYSCTL;
261 		return(kd);
262 	}
263 
264 	/*
265 	 * Call the MD open hook.  This sets:
266 	 *	usrstack, min_uva, max_uva
267 	 */
268 	if (_kvm_mdopen(kd)) {
269 		_kvm_err(kd, kd->program, "md init failed");
270 		goto failed;
271 	}
272 
273 	ufgiven = (uf != NULL);
274 	if (!ufgiven) {
275 #ifdef CPU_BOOTED_KERNEL
276 		/* 130 is 128 + '/' + '\0' */
277 		static char booted_kernel[130];
278 		int mib[2], rc;
279 		size_t len;
280 
281 		mib[0] = CTL_MACHDEP;
282 		mib[1] = CPU_BOOTED_KERNEL;
283 		booted_kernel[0] = '/';
284 		booted_kernel[1] = '\0';
285 		len = sizeof(booted_kernel) - 2;
286 		rc = sysctl(&mib[0], 2, &booted_kernel[1], &len, NULL, 0);
287 		booted_kernel[sizeof(booted_kernel) - 1] = '\0';
288 		uf = (booted_kernel[1] == '/') ?
289 		    &booted_kernel[1] : &booted_kernel[0];
290 		if (rc != -1)
291 			rc = stat(uf, &st);
292 		if (rc != -1 && !S_ISREG(st.st_mode))
293 			rc = -1;
294 		if (rc == -1)
295 #endif /* CPU_BOOTED_KERNEL */
296 			uf = _PATH_UNIX;
297 	}
298 	else if (strlen(uf) >= MAXPATHLEN) {
299 		_kvm_err(kd, kd->program, "exec file name too long");
300 		goto failed;
301 	}
302 	if (flag & ~O_RDWR) {
303 		_kvm_err(kd, kd->program, "bad flags arg");
304 		goto failed;
305 	}
306 	if (mf == 0)
307 		mf = _PATH_MEM;
308 	if (sf == 0)
309 		sf = _PATH_DRUM;
310 
311 	/*
312 	 * Open the kernel namelist.  If /dev/ksyms doesn't
313 	 * exist, open the current kernel.
314 	 */
315 	if (ufgiven == 0)
316 		kd->nlfd = open(_PATH_KSYMS, O_RDONLY | O_CLOEXEC, 0);
317 	if (kd->nlfd < 0) {
318 		if ((kd->nlfd = open(uf, O_RDONLY | O_CLOEXEC, 0)) < 0) {
319 			_kvm_syserr(kd, kd->program, "%s", uf);
320 			goto failed;
321 		}
322 		strlcpy(kd->kernelname, uf, sizeof(kd->kernelname));
323 	} else {
324 		strlcpy(kd->kernelname, _PATH_KSYMS, sizeof(kd->kernelname));
325 	}
326 
327 	if ((kd->pmfd = open(mf, flag | O_CLOEXEC, 0)) < 0) {
328 		_kvm_syserr(kd, kd->program, "%s", mf);
329 		goto failed;
330 	}
331 	if (fstat(kd->pmfd, &st) < 0) {
332 		_kvm_syserr(kd, kd->program, "%s", mf);
333 		goto failed;
334 	}
335 	if (S_ISCHR(st.st_mode) && strcmp(mf, _PATH_MEM) == 0) {
336 		/*
337 		 * If this is /dev/mem, open kmem too.  (Maybe we should
338 		 * make it work for either /dev/mem or /dev/kmem -- in either
339 		 * case you're working with a live kernel.)
340 		 */
341 		if ((kd->vmfd = open(_PATH_KMEM, flag | O_CLOEXEC, 0)) < 0) {
342 			_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
343 			goto failed;
344 		}
345 		kd->alive = KVM_ALIVE_FILES;
346 		if ((kd->swfd = open(sf, flag | O_CLOEXEC, 0)) < 0) {
347 			if (errno != ENXIO) {
348 				_kvm_syserr(kd, kd->program, "%s", sf);
349 				goto failed;
350 			}
351 			/* swap is not configured?  not fatal */
352 		}
353 	} else {
354 		kd->fdalign = DEV_BSIZE;	/* XXX */
355 		/*
356 		 * This is a crash dump.
357 		 * Initialize the virtual address translation machinery.
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(kvm_t *kd)
393 {
394 	kcore_hdr_t	kcore_hdr;
395 	kcore_seg_t	cpu_hdr;
396 	kcore_seg_t	mem_hdr;
397 	size_t		offset;
398 	ssize_t		sz;
399 
400 	/*
401 	 * Read the kcore_hdr_t
402 	 */
403 	sz = Pread(kd, kd->pmfd, &kcore_hdr, sizeof(kcore_hdr), (off_t)0);
404 	if (sz != sizeof(kcore_hdr))
405 		return (-1);
406 
407 	/*
408 	 * Currently, we only support dump-files made by the current
409 	 * architecture...
410 	 */
411 	if ((CORE_GETMAGIC(kcore_hdr) != KCORE_MAGIC) ||
412 	    (CORE_GETMID(kcore_hdr) != MID_MACHINE))
413 		return (-1);
414 
415 	/*
416 	 * Currently, we only support exactly 2 segments: cpu-segment
417 	 * and data-segment in exactly that order.
418 	 */
419 	if (kcore_hdr.c_nseg != 2)
420 		return (-1);
421 
422 	/*
423 	 * Save away the kcore_hdr.  All errors after this
424 	 * should do a to "goto fail" to deallocate things.
425 	 */
426 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr));
427 	memcpy(kd->kcore_hdr, &kcore_hdr, sizeof(kcore_hdr));
428 	offset = kcore_hdr.c_hdrsize;
429 
430 	/*
431 	 * Read the CPU segment header
432 	 */
433 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), (off_t)offset);
434 	if (sz != sizeof(cpu_hdr))
435 		goto fail;
436 	if ((CORE_GETMAGIC(cpu_hdr) != KCORESEG_MAGIC) ||
437 	    (CORE_GETFLAG(cpu_hdr) != CORE_CPU))
438 		goto fail;
439 	offset += kcore_hdr.c_seghdrsize;
440 
441 	/*
442 	 * Read the CPU segment DATA.
443 	 */
444 	kd->cpu_dsize = cpu_hdr.c_size;
445 	kd->cpu_data = _kvm_malloc(kd, cpu_hdr.c_size);
446 	if (kd->cpu_data == NULL)
447 		goto fail;
448 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size, (off_t)offset);
449 	if (sz != cpu_hdr.c_size)
450 		goto fail;
451 	offset += cpu_hdr.c_size;
452 
453 	/*
454 	 * Read the next segment header: data segment
455 	 */
456 	sz = Pread(kd, kd->pmfd, &mem_hdr, sizeof(mem_hdr), (off_t)offset);
457 	if (sz != sizeof(mem_hdr))
458 		goto fail;
459 	offset += kcore_hdr.c_seghdrsize;
460 
461 	if ((CORE_GETMAGIC(mem_hdr) != KCORESEG_MAGIC) ||
462 	    (CORE_GETFLAG(mem_hdr) != CORE_DATA))
463 		goto fail;
464 
465 	kd->dump_off = offset;
466 	return (0);
467 
468 fail:
469 	if (kd->kcore_hdr != NULL) {
470 		free(kd->kcore_hdr);
471 		kd->kcore_hdr = NULL;
472 	}
473 	if (kd->cpu_data != NULL) {
474 		free(kd->cpu_data);
475 		kd->cpu_data = NULL;
476 		kd->cpu_dsize = 0;
477 	}
478 	return (-1);
479 }
480 
481 /*
482  * The format while on the dump device is: (new format)
483  *	kcore_seg_t cpu_hdr;
484  *	(opaque)    cpu_data; (size is cpu_hdr.c_size)
485  *	kcore_seg_t mem_hdr;
486  *	(memory)    mem_data; (size is mem_hdr.c_size)
487  */
488 int
489 kvm_dump_mkheader(kvm_t *kd, off_t dump_off)
490 {
491 	kcore_seg_t	cpu_hdr;
492 	size_t hdr_size;
493 	ssize_t sz;
494 
495 	if (kd->kcore_hdr != NULL) {
496 	    _kvm_err(kd, kd->program, "already has a dump header");
497 	    return (-1);
498 	}
499 	if (ISALIVE(kd)) {
500 		_kvm_err(kd, kd->program, "don't use on live kernel");
501 		return (-1);
502 	}
503 
504 	/*
505 	 * Validate new format crash dump
506 	 */
507 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), dump_off);
508 	if (sz != sizeof(cpu_hdr)) {
509 		if (sz == -1)
510 			_kvm_err(kd, 0, "read %zx bytes at offset %"PRIx64
511 			    " for cpu_hdr failed: %s", sizeof(cpu_hdr),
512 			    dump_off, strerror(errno));
513 		else
514 			_kvm_err(kd, 0, "read %zx bytes at offset %"PRIx64
515 			    " for cpu_hdr instead of requested %zu",
516 			    sz, dump_off, sizeof(cpu_hdr));
517 		return (-1);
518 	}
519 	if ((CORE_GETMAGIC(cpu_hdr) != KCORE_MAGIC)
520 		|| (CORE_GETMID(cpu_hdr) != MID_MACHINE)) {
521 		_kvm_err(kd, 0, "invalid magic in cpu_hdr");
522 		return (0);
523 	}
524 	hdr_size = ALIGN(sizeof(cpu_hdr));
525 
526 	/*
527 	 * Read the CPU segment.
528 	 */
529 	kd->cpu_dsize = cpu_hdr.c_size;
530 	kd->cpu_data = _kvm_malloc(kd, kd->cpu_dsize);
531 	if (kd->cpu_data == NULL) {
532 		_kvm_err(kd, kd->program, "no cpu_data");
533 		goto fail;
534 	}
535 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size,
536 	    dump_off + hdr_size);
537 	if (sz != cpu_hdr.c_size) {
538 		_kvm_err(kd, kd->program, "size %zu != cpu_hdr.csize %"PRIu32,
539 		    sz, cpu_hdr.c_size);
540 		goto fail;
541 	}
542 	hdr_size += kd->cpu_dsize;
543 
544 	/*
545 	 * Leave phys mem pointer at beginning of memory data
546 	 */
547 	kd->dump_off = dump_off + hdr_size;
548 	if (Lseek(kd, kd->pmfd, kd->dump_off, SEEK_SET) == -1) {
549 		_kvm_err(kd, kd->program, "failed to seek to %" PRId64,
550 		    (int64_t)kd->dump_off);
551 		goto fail;
552 	}
553 
554 	/*
555 	 * Create a kcore_hdr.
556 	 */
557 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr_t));
558 	if (kd->kcore_hdr == NULL) {
559 		_kvm_err(kd, kd->program, "failed to allocate header");
560 		goto fail;
561 	}
562 
563 	kd->kcore_hdr->c_hdrsize    = ALIGN(sizeof(kcore_hdr_t));
564 	kd->kcore_hdr->c_seghdrsize = ALIGN(sizeof(kcore_seg_t));
565 	kd->kcore_hdr->c_nseg       = 2;
566 	CORE_SETMAGIC(*(kd->kcore_hdr), KCORE_MAGIC, MID_MACHINE,0);
567 
568 	/*
569 	 * Now that we have a valid header, enable translations.
570 	 */
571 	if (_kvm_initvtop(kd) == 0)
572 		/* Success */
573 		return (hdr_size);
574 
575 fail:
576 	if (kd->kcore_hdr != NULL) {
577 		free(kd->kcore_hdr);
578 		kd->kcore_hdr = NULL;
579 	}
580 	if (kd->cpu_data != NULL) {
581 		free(kd->cpu_data);
582 		kd->cpu_data = NULL;
583 		kd->cpu_dsize = 0;
584 	}
585 	return (-1);
586 }
587 
588 static int
589 clear_gap(kvm_t *kd, bool (*write_buf)(void *, const void *, size_t),
590     void *cookie, size_t size)
591 {
592 	char buf[1024];
593 	size_t len;
594 
595 	(void)memset(buf, 0, size > sizeof(buf) ? sizeof(buf) : size);
596 
597 	while (size > 0) {
598 		len = size > sizeof(buf) ? sizeof(buf) : size;
599 		if (!(*write_buf)(cookie, buf, len)) {
600 			_kvm_syserr(kd, kd->program, "clear_gap");
601 			return -1;
602 		}
603 		size -= len;
604 	}
605 
606 	return 0;
607 }
608 
609 /*
610  * Write the dump header by calling write_buf with cookie as first argument.
611  */
612 int
613 kvm_dump_header(kvm_t *kd, bool (*write_buf)(void *, const void *, size_t),
614     void *cookie, int dumpsize)
615 {
616 	kcore_seg_t	seghdr;
617 	long		offset;
618 	size_t		gap;
619 
620 	if (kd->kcore_hdr == NULL || kd->cpu_data == NULL) {
621 		_kvm_err(kd, kd->program, "no valid dump header(s)");
622 		return (-1);
623 	}
624 
625 	/*
626 	 * Write the generic header
627 	 */
628 	offset = 0;
629 	if (!(*write_buf)(cookie, kd->kcore_hdr, sizeof(kcore_hdr_t))) {
630 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
631 		return (-1);
632 	}
633 	offset += kd->kcore_hdr->c_hdrsize;
634 	gap     = kd->kcore_hdr->c_hdrsize - sizeof(kcore_hdr_t);
635 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
636 		return (-1);
637 
638 	/*
639 	 * Write the CPU header
640 	 */
641 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_CPU);
642 	seghdr.c_size = ALIGN(kd->cpu_dsize);
643 	if (!(*write_buf)(cookie, &seghdr, sizeof(seghdr))) {
644 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
645 		return (-1);
646 	}
647 	offset += kd->kcore_hdr->c_seghdrsize;
648 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
649 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
650 		return (-1);
651 
652 	if (!(*write_buf)(cookie, kd->cpu_data, kd->cpu_dsize)) {
653 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
654 		return (-1);
655 	}
656 	offset += seghdr.c_size;
657 	gap     = seghdr.c_size - kd->cpu_dsize;
658 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
659 		return (-1);
660 
661 	/*
662 	 * Write the actual dump data segment header
663 	 */
664 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_DATA);
665 	seghdr.c_size = dumpsize;
666 	if (!(*write_buf)(cookie, &seghdr, sizeof(seghdr))) {
667 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
668 		return (-1);
669 	}
670 	offset += kd->kcore_hdr->c_seghdrsize;
671 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
672 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
673 		return (-1);
674 
675 	return (int)offset;
676 }
677 
678 static bool
679 kvm_dump_header_stdio(void *cookie, const void *buf, size_t len)
680 {
681 	return fwrite(buf, len, 1, (FILE *)cookie) == 1;
682 }
683 
684 int
685 kvm_dump_wrtheader(kvm_t *kd, FILE *fp, int dumpsize)
686 {
687 	return kvm_dump_header(kd, kvm_dump_header_stdio, fp, dumpsize);
688 }
689 
690 kvm_t *
691 kvm_openfiles(const char *uf, const char *mf, const char *sf,
692     int flag, char *errout)
693 {
694 	kvm_t *kd;
695 
696 	if ((kd = malloc(sizeof(*kd))) == NULL) {
697 		(void)strlcpy(errout, strerror(errno), _POSIX2_LINE_MAX);
698 		return (0);
699 	}
700 	kd->program = 0;
701 	return (_kvm_open(kd, uf, mf, sf, flag, errout));
702 }
703 
704 kvm_t *
705 kvm_open(const char *uf, const char *mf, const char *sf, int flag,
706     const char *program)
707 {
708 	kvm_t *kd;
709 
710 	if ((kd = malloc(sizeof(*kd))) == NULL) {
711 		(void)fprintf(stderr, "%s: %s\n",
712 		    program ? program : getprogname(), strerror(errno));
713 		return (0);
714 	}
715 	kd->program = program;
716 	return (_kvm_open(kd, uf, mf, sf, flag, NULL));
717 }
718 
719 int
720 kvm_close(kvm_t *kd)
721 {
722 	int error = 0;
723 
724 	if (kd->pmfd >= 0)
725 		error |= close(kd->pmfd);
726 	if (kd->vmfd >= 0)
727 		error |= close(kd->vmfd);
728 	if (kd->nlfd >= 0)
729 		error |= close(kd->nlfd);
730 	if (kd->swfd >= 0)
731 		error |= close(kd->swfd);
732 	if (kd->vmst)
733 		_kvm_freevtop(kd);
734 	kd->cpu_dsize = 0;
735 	if (kd->cpu_data != NULL)
736 		free(kd->cpu_data);
737 	if (kd->kcore_hdr != NULL)
738 		free(kd->kcore_hdr);
739 	if (kd->procbase != 0)
740 		free(kd->procbase);
741 	if (kd->procbase2 != 0)
742 		free(kd->procbase2);
743 	if (kd->lwpbase != 0)
744 		free(kd->lwpbase);
745 	if (kd->swapspc != 0)
746 		free(kd->swapspc);
747 	if (kd->argspc != 0)
748 		free(kd->argspc);
749 	if (kd->argbuf != 0)
750 		free(kd->argbuf);
751 	if (kd->argv != 0)
752 		free(kd->argv);
753 	if (kd->iobuf != 0)
754 		free(kd->iobuf);
755 	free(kd);
756 
757 	return (error);
758 }
759 
760 int
761 kvm_nlist(kvm_t *kd, struct nlist *nl)
762 {
763 	int rv;
764 
765 	/*
766 	 * Call the nlist(3) routines to retrieve the given namelist.
767 	 */
768 	rv = __fdnlist(kd->nlfd, nl);
769 
770 	if (rv == -1)
771 		_kvm_err(kd, 0, "bad namelist");
772 
773 	return (rv);
774 }
775 
776 int
777 kvm_dump_inval(kvm_t *kd)
778 {
779 	struct nlist	nl[2];
780 	paddr_t		pa;
781 	size_t		dsize;
782 	off_t		doff;
783 	void		*newbuf;
784 
785 	if (ISALIVE(kd)) {
786 		_kvm_err(kd, kd->program, "clearing dump on live kernel");
787 		return (-1);
788 	}
789 	nl[0].n_name = "_dumpmag";
790 	nl[1].n_name = NULL;
791 
792 	if (kvm_nlist(kd, nl) == -1) {
793 		_kvm_err(kd, 0, "bad namelist");
794 		return (-1);
795 	}
796 	if (_kvm_kvatop(kd, (vaddr_t)nl[0].n_value, &pa) == 0)
797 		return (-1);
798 
799 	errno = 0;
800 	dsize = MAX(kd->fdalign, sizeof(u_long));
801 	if (kd->iobufsz < dsize) {
802 		newbuf = realloc(kd->iobuf, dsize);
803 		if (newbuf == NULL) {
804 			_kvm_syserr(kd, 0, "cannot allocate I/O buffer");
805 			return (-1);
806 		}
807 		kd->iobuf = newbuf;
808 		kd->iobufsz = dsize;
809 	}
810 	memset(kd->iobuf, 0, dsize);
811 	doff = _kvm_pa2off(kd, pa);
812 	doff -= doff % kd->fdalign;
813 	if (pwrite(kd->pmfd, kd->iobuf, dsize, doff) == -1) {
814 		_kvm_syserr(kd, 0, "cannot invalidate dump - pwrite");
815 		return (-1);
816 	}
817 	return (0);
818 }
819 
820 ssize_t
821 kvm_read(kvm_t *kd, u_long kva, void *buf, size_t len)
822 {
823 	int cc;
824 	void *cp;
825 
826 	if (ISKMEM(kd)) {
827 		/*
828 		 * We're using /dev/kmem.  Just read straight from the
829 		 * device and let the active kernel do the address translation.
830 		 */
831 		errno = 0;
832 		cc = _kvm_pread(kd, kd->vmfd, buf, len, (off_t)kva);
833 		if (cc < 0) {
834 			_kvm_syserr(kd, 0, "kvm_read");
835 			return (-1);
836 		} else if (cc < len)
837 			_kvm_err(kd, kd->program, "short read");
838 		return (cc);
839 	} else if (ISSYSCTL(kd)) {
840 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
841 		    "can't use kvm_read");
842 		return (-1);
843 	} else {
844 		if ((kd->kcore_hdr == NULL) || (kd->cpu_data == NULL)) {
845 			_kvm_err(kd, kd->program, "no valid dump header");
846 			return (-1);
847 		}
848 		cp = buf;
849 		while (len > 0) {
850 			paddr_t	pa;
851 			off_t	foff;
852 
853 			cc = _kvm_kvatop(kd, (vaddr_t)kva, &pa);
854 			if (cc == 0)
855 				return (-1);
856 			if (cc > len)
857 				cc = len;
858 			foff = _kvm_pa2off(kd, pa);
859 			errno = 0;
860 			cc = _kvm_pread(kd, kd->pmfd, cp, (size_t)cc, foff);
861 			if (cc < 0) {
862 				_kvm_syserr(kd, kd->program, "kvm_read");
863 				break;
864 			}
865 			/*
866 			 * If kvm_kvatop returns a bogus value or our core
867 			 * file is truncated, we might wind up seeking beyond
868 			 * the end of the core file in which case the read will
869 			 * return 0 (EOF).
870 			 */
871 			if (cc == 0)
872 				break;
873 			cp = (char *)cp + cc;
874 			kva += cc;
875 			len -= cc;
876 		}
877 		return ((char *)cp - (char *)buf);
878 	}
879 	/* NOTREACHED */
880 }
881 
882 ssize_t
883 kvm_write(kvm_t *kd, u_long kva, const void *buf, size_t len)
884 {
885 	int cc;
886 
887 	if (ISKMEM(kd)) {
888 		/*
889 		 * Just like kvm_read, only we write.
890 		 */
891 		errno = 0;
892 		cc = pwrite(kd->vmfd, buf, len, (off_t)kva);
893 		if (cc < 0) {
894 			_kvm_syserr(kd, 0, "kvm_write");
895 			return (-1);
896 		} else if (cc < len)
897 			_kvm_err(kd, kd->program, "short write");
898 		return (cc);
899 	} else if (ISSYSCTL(kd)) {
900 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
901 		    "can't use kvm_write");
902 		return (-1);
903 	} else {
904 		_kvm_err(kd, kd->program,
905 		    "kvm_write not implemented for dead kernels");
906 		return (-1);
907 	}
908 	/* NOTREACHED */
909 }
910