xref: /netbsd-src/lib/libkvm/kvm.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: kvm.c,v 1.90 2007/11/08 21:58:26 joerg 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 1.90 2007/11/08 21:58:26 joerg 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_aout.h>
55 #include <sys/kcore.h>
56 #include <sys/ksyms.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 int	open_cloexec(const char *, int, int);
83 static off_t	Lseek(kvm_t *, int, off_t, int);
84 static ssize_t	Pread(kvm_t *, int, void *, size_t, off_t);
85 
86 char *
87 kvm_geterr(kvm_t *kd)
88 {
89 	return (kd->errbuf);
90 }
91 
92 /*
93  * Report an error using printf style arguments.  "program" is kd->program
94  * on hard errors, and 0 on soft errors, so that under sun error emulation,
95  * only hard errors are printed out (otherwise, programs like gdb will
96  * generate tons of error messages when trying to access bogus pointers).
97  */
98 void
99 _kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
100 {
101 	va_list ap;
102 
103 	va_start(ap, fmt);
104 	if (program != NULL) {
105 		(void)fprintf(stderr, "%s: ", program);
106 		(void)vfprintf(stderr, fmt, ap);
107 		(void)fputc('\n', stderr);
108 	} else
109 		(void)vsnprintf(kd->errbuf,
110 		    sizeof(kd->errbuf), fmt, ap);
111 
112 	va_end(ap);
113 }
114 
115 void
116 _kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
117 {
118 	va_list ap;
119 	size_t n;
120 
121 	va_start(ap, fmt);
122 	if (program != NULL) {
123 		(void)fprintf(stderr, "%s: ", program);
124 		(void)vfprintf(stderr, fmt, ap);
125 		(void)fprintf(stderr, ": %s\n", strerror(errno));
126 	} else {
127 		char *cp = kd->errbuf;
128 
129 		(void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
130 		n = strlen(cp);
131 		(void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
132 		    strerror(errno));
133 	}
134 	va_end(ap);
135 }
136 
137 void *
138 _kvm_malloc(kvm_t *kd, size_t n)
139 {
140 	void *p;
141 
142 	if ((p = malloc(n)) == NULL)
143 		_kvm_err(kd, kd->program, "%s", strerror(errno));
144 	return (p);
145 }
146 
147 /*
148  * Open a file setting the close on exec bit.
149  */
150 static int
151 open_cloexec(const char *fname, int flags, int mode)
152 {
153 	int fd;
154 
155 	if ((fd = open(fname, flags, mode)) == -1)
156 		return fd;
157 	if (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1)
158 		goto error;
159 
160 	return fd;
161 error:
162 	flags = errno;
163 	(void)close(fd);
164 	errno = flags;
165 	return -1;
166 }
167 
168 /*
169  * Wrapper around the lseek(2) system call; calls _kvm_syserr() for us
170  * in the event of emergency.
171  */
172 static off_t
173 Lseek(kvm_t *kd, int fd, off_t offset, int whence)
174 {
175 	off_t off;
176 
177 	errno = 0;
178 
179 	if ((off = lseek(fd, offset, whence)) == -1 && errno != 0) {
180 		_kvm_syserr(kd, kd->program, "Lseek");
181 		return ((off_t)-1);
182 	}
183 	return (off);
184 }
185 
186 /*
187  * Wrapper around the pread(2) system call; calls _kvm_syserr() for us
188  * in the event of emergency.
189  */
190 static ssize_t
191 Pread(kvm_t *kd, int fd, void *buf, size_t nbytes, off_t offset)
192 {
193 	ssize_t rv;
194 
195 	errno = 0;
196 
197 	if ((rv = pread(fd, buf, nbytes, offset)) != nbytes &&
198 	    errno != 0)
199 		_kvm_syserr(kd, kd->program, "Pread");
200 	return (rv);
201 }
202 
203 static kvm_t *
204 _kvm_open(kvm_t *kd, const char *uf, const char *mf, const char *sf, int flag,
205     char *errout)
206 {
207 	struct stat st;
208 	int ufgiven;
209 
210 	kd->pmfd = -1;
211 	kd->vmfd = -1;
212 	kd->swfd = -1;
213 	kd->nlfd = -1;
214 	kd->alive = KVM_ALIVE_DEAD;
215 	kd->procbase = NULL;
216 	kd->procbase_len = 0;
217 	kd->procbase2 = NULL;
218 	kd->procbase2_len = 0;
219 	kd->lwpbase = NULL;
220 	kd->lwpbase_len = 0;
221 	kd->nbpg = getpagesize();
222 	kd->swapspc = NULL;
223 	kd->argspc = NULL;
224 	kd->argspc_len = 0;
225 	kd->argbuf = NULL;
226 	kd->argv = NULL;
227 	kd->vmst = NULL;
228 	kd->vm_page_buckets = NULL;
229 	kd->kcore_hdr = NULL;
230 	kd->cpu_dsize = 0;
231 	kd->cpu_data = NULL;
232 	kd->dump_off = 0;
233 
234 	if (flag & KVM_NO_FILES) {
235 		kd->alive = KVM_ALIVE_SYSCTL;
236 		return(kd);
237 	}
238 
239 	/*
240 	 * Call the MD open hook.  This sets:
241 	 *	usrstack, min_uva, max_uva
242 	 */
243 	if (_kvm_mdopen(kd)) {
244 		_kvm_err(kd, kd->program, "md init failed");
245 		goto failed;
246 	}
247 
248 	ufgiven = (uf != NULL);
249 	if (!ufgiven) {
250 #ifdef CPU_BOOTED_KERNEL
251 		/* 130 is 128 + '/' + '\0' */
252 		static char booted_kernel[130];
253 		int mib[2], rc;
254 		size_t len;
255 
256 		mib[0] = CTL_MACHDEP;
257 		mib[1] = CPU_BOOTED_KERNEL;
258 		booted_kernel[0] = '/';
259 		booted_kernel[1] = '\0';
260 		len = sizeof(booted_kernel) - 2;
261 		rc = sysctl(&mib[0], 2, &booted_kernel[1], &len, NULL, 0);
262 		booted_kernel[sizeof(booted_kernel) - 1] = '\0';
263 		uf = (booted_kernel[1] == '/') ?
264 		    &booted_kernel[1] : &booted_kernel[0];
265 		if (rc != -1)
266 			rc = stat(uf, &st);
267 		if (rc != -1 && !S_ISREG(st.st_mode))
268 			rc = -1;
269 		if (rc == -1)
270 #endif /* CPU_BOOTED_KERNEL */
271 			uf = _PATH_UNIX;
272 	}
273 	else if (strlen(uf) >= MAXPATHLEN) {
274 		_kvm_err(kd, kd->program, "exec file name too long");
275 		goto failed;
276 	}
277 	if (flag & ~O_RDWR) {
278 		_kvm_err(kd, kd->program, "bad flags arg");
279 		goto failed;
280 	}
281 	if (mf == 0)
282 		mf = _PATH_MEM;
283 	if (sf == 0)
284 		sf = _PATH_DRUM;
285 
286 	if ((kd->pmfd = open_cloexec(mf, flag, 0)) < 0) {
287 		_kvm_syserr(kd, kd->program, "%s", mf);
288 		goto failed;
289 	}
290 	if (fstat(kd->pmfd, &st) < 0) {
291 		_kvm_syserr(kd, kd->program, "%s", mf);
292 		goto failed;
293 	}
294 	if (S_ISCHR(st.st_mode)) {
295 		/*
296 		 * If this is a character special device, then check that
297 		 * it's /dev/mem.  If so, open kmem too.  (Maybe we should
298 		 * make it work for either /dev/mem or /dev/kmem -- in either
299 		 * case you're working with a live kernel.)
300 		 */
301 		if (strcmp(mf, _PATH_MEM) != 0) {	/* XXX */
302 			_kvm_err(kd, kd->program,
303 				 "%s: not physical memory device", mf);
304 			goto failed;
305 		}
306 		if ((kd->vmfd = open_cloexec(_PATH_KMEM, flag, 0)) < 0) {
307 			_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
308 			goto failed;
309 		}
310 		kd->alive = KVM_ALIVE_FILES;
311 		if ((kd->swfd = open_cloexec(sf, flag, 0)) < 0) {
312 			if (errno != ENXIO) {
313 				_kvm_syserr(kd, kd->program, "%s", sf);
314 				goto failed;
315 			}
316 			/* swap is not configured?  not fatal */
317 		}
318 		/*
319 		 * Open the kernel namelist.  If /dev/ksyms doesn't
320 		 * exist, open the current kernel.
321 		 */
322 		if (ufgiven == 0)
323 			kd->nlfd = open_cloexec(_PATH_KSYMS, O_RDONLY, 0);
324 		if (kd->nlfd < 0) {
325 			if ((kd->nlfd = open_cloexec(uf, O_RDONLY, 0)) < 0) {
326 				_kvm_syserr(kd, kd->program, "%s", uf);
327 				goto failed;
328 			}
329 		} else {
330 			/*
331 			 * We're here because /dev/ksyms was opened
332 			 * successfully.  However, we don't want to keep it
333 			 * open, so we close it now.  Later, we will open
334 			 * it again, since it will be the only case where
335 			 * kd->nlfd is negative.
336 			 */
337 			close(kd->nlfd);
338 			kd->nlfd = -1;
339 		}
340 	} else {
341 		/*
342 		 * This is a crash dump.
343 		 * Initialize the virtual address translation machinery,
344 		 * but first setup the namelist fd.
345 		 */
346 		if ((kd->nlfd = open_cloexec(uf, O_RDONLY, 0)) < 0) {
347 			_kvm_syserr(kd, kd->program, "%s", uf);
348 			goto failed;
349 		}
350 
351 		/*
352 		 * If there is no valid core header, fail silently here.
353 		 * The address translations however will fail without
354 		 * header. Things can be made to run by calling
355 		 * kvm_dump_mkheader() before doing any translation.
356 		 */
357 		if (_kvm_get_header(kd) == 0) {
358 			if (_kvm_initvtop(kd) < 0)
359 				goto failed;
360 		}
361 	}
362 	return (kd);
363 failed:
364 	/*
365 	 * Copy out the error if doing sane error semantics.
366 	 */
367 	if (errout != 0)
368 		(void)strlcpy(errout, kd->errbuf, _POSIX2_LINE_MAX);
369 	(void)kvm_close(kd);
370 	return (0);
371 }
372 
373 /*
374  * The kernel dump file (from savecore) contains:
375  *    kcore_hdr_t kcore_hdr;
376  *    kcore_seg_t cpu_hdr;
377  *    (opaque)    cpu_data; (size is cpu_hdr.c_size)
378  *	  kcore_seg_t mem_hdr;
379  *    (memory)    mem_data; (size is mem_hdr.c_size)
380  *
381  * Note: khdr is padded to khdr.c_hdrsize;
382  * cpu_hdr and mem_hdr are padded to khdr.c_seghdrsize
383  */
384 static int
385 _kvm_get_header(kvm_t *kd)
386 {
387 	kcore_hdr_t	kcore_hdr;
388 	kcore_seg_t	cpu_hdr;
389 	kcore_seg_t	mem_hdr;
390 	size_t		offset;
391 	ssize_t		sz;
392 
393 	/*
394 	 * Read the kcore_hdr_t
395 	 */
396 	sz = Pread(kd, kd->pmfd, &kcore_hdr, sizeof(kcore_hdr), (off_t)0);
397 	if (sz != sizeof(kcore_hdr))
398 		return (-1);
399 
400 	/*
401 	 * Currently, we only support dump-files made by the current
402 	 * architecture...
403 	 */
404 	if ((CORE_GETMAGIC(kcore_hdr) != KCORE_MAGIC) ||
405 	    (CORE_GETMID(kcore_hdr) != MID_MACHINE))
406 		return (-1);
407 
408 	/*
409 	 * Currently, we only support exactly 2 segments: cpu-segment
410 	 * and data-segment in exactly that order.
411 	 */
412 	if (kcore_hdr.c_nseg != 2)
413 		return (-1);
414 
415 	/*
416 	 * Save away the kcore_hdr.  All errors after this
417 	 * should do a to "goto fail" to deallocate things.
418 	 */
419 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr));
420 	memcpy(kd->kcore_hdr, &kcore_hdr, sizeof(kcore_hdr));
421 	offset = kcore_hdr.c_hdrsize;
422 
423 	/*
424 	 * Read the CPU segment header
425 	 */
426 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), (off_t)offset);
427 	if (sz != sizeof(cpu_hdr))
428 		goto fail;
429 	if ((CORE_GETMAGIC(cpu_hdr) != KCORESEG_MAGIC) ||
430 	    (CORE_GETFLAG(cpu_hdr) != CORE_CPU))
431 		goto fail;
432 	offset += kcore_hdr.c_seghdrsize;
433 
434 	/*
435 	 * Read the CPU segment DATA.
436 	 */
437 	kd->cpu_dsize = cpu_hdr.c_size;
438 	kd->cpu_data = _kvm_malloc(kd, cpu_hdr.c_size);
439 	if (kd->cpu_data == NULL)
440 		goto fail;
441 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size, (off_t)offset);
442 	if (sz != cpu_hdr.c_size)
443 		goto fail;
444 	offset += cpu_hdr.c_size;
445 
446 	/*
447 	 * Read the next segment header: data segment
448 	 */
449 	sz = Pread(kd, kd->pmfd, &mem_hdr, sizeof(mem_hdr), (off_t)offset);
450 	if (sz != sizeof(mem_hdr))
451 		goto fail;
452 	offset += kcore_hdr.c_seghdrsize;
453 
454 	if ((CORE_GETMAGIC(mem_hdr) != KCORESEG_MAGIC) ||
455 	    (CORE_GETFLAG(mem_hdr) != CORE_DATA))
456 		goto fail;
457 
458 	kd->dump_off = offset;
459 	return (0);
460 
461 fail:
462 	if (kd->kcore_hdr != NULL) {
463 		free(kd->kcore_hdr);
464 		kd->kcore_hdr = NULL;
465 	}
466 	if (kd->cpu_data != NULL) {
467 		free(kd->cpu_data);
468 		kd->cpu_data = NULL;
469 		kd->cpu_dsize = 0;
470 	}
471 	return (-1);
472 }
473 
474 /*
475  * The format while on the dump device is: (new format)
476  *	kcore_seg_t cpu_hdr;
477  *	(opaque)    cpu_data; (size is cpu_hdr.c_size)
478  *	kcore_seg_t mem_hdr;
479  *	(memory)    mem_data; (size is mem_hdr.c_size)
480  */
481 int
482 kvm_dump_mkheader(kvm_t *kd, off_t dump_off)
483 {
484 	kcore_seg_t	cpu_hdr;
485 	size_t hdr_size;
486 	ssize_t sz;
487 
488 	if (kd->kcore_hdr != NULL) {
489 	    _kvm_err(kd, kd->program, "already has a dump header");
490 	    return (-1);
491 	}
492 	if (ISALIVE(kd)) {
493 		_kvm_err(kd, kd->program, "don't use on live kernel");
494 		return (-1);
495 	}
496 
497 	/*
498 	 * Validate new format crash dump
499 	 */
500 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), dump_off);
501 	if (sz != sizeof(cpu_hdr))
502 		return (-1);
503 	if ((CORE_GETMAGIC(cpu_hdr) != KCORE_MAGIC)
504 		|| (CORE_GETMID(cpu_hdr) != MID_MACHINE)) {
505 		_kvm_err(kd, 0, "invalid magic in cpu_hdr");
506 		return (0);
507 	}
508 	hdr_size = ALIGN(sizeof(cpu_hdr));
509 
510 	/*
511 	 * Read the CPU segment.
512 	 */
513 	kd->cpu_dsize = cpu_hdr.c_size;
514 	kd->cpu_data = _kvm_malloc(kd, kd->cpu_dsize);
515 	if (kd->cpu_data == NULL)
516 		goto fail;
517 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size,
518 	    dump_off + hdr_size);
519 	if (sz != cpu_hdr.c_size)
520 		goto fail;
521 	hdr_size += kd->cpu_dsize;
522 
523 	/*
524 	 * Leave phys mem pointer at beginning of memory data
525 	 */
526 	kd->dump_off = dump_off + hdr_size;
527 	if (Lseek(kd, kd->pmfd, kd->dump_off, SEEK_SET) == -1)
528 		goto fail;
529 
530 	/*
531 	 * Create a kcore_hdr.
532 	 */
533 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr_t));
534 	if (kd->kcore_hdr == NULL)
535 		goto fail;
536 
537 	kd->kcore_hdr->c_hdrsize    = ALIGN(sizeof(kcore_hdr_t));
538 	kd->kcore_hdr->c_seghdrsize = ALIGN(sizeof(kcore_seg_t));
539 	kd->kcore_hdr->c_nseg       = 2;
540 	CORE_SETMAGIC(*(kd->kcore_hdr), KCORE_MAGIC, MID_MACHINE,0);
541 
542 	/*
543 	 * Now that we have a valid header, enable translations.
544 	 */
545 	if (_kvm_initvtop(kd) == 0)
546 		/* Success */
547 		return (hdr_size);
548 
549 fail:
550 	if (kd->kcore_hdr != NULL) {
551 		free(kd->kcore_hdr);
552 		kd->kcore_hdr = NULL;
553 	}
554 	if (kd->cpu_data != NULL) {
555 		free(kd->cpu_data);
556 		kd->cpu_data = NULL;
557 		kd->cpu_dsize = 0;
558 	}
559 	return (-1);
560 }
561 
562 static int
563 clear_gap(kvm_t *kd, bool (*write_buf)(void *, const void *, size_t),
564     void *cookie, size_t size)
565 {
566 	char buf[1024];
567 	size_t len;
568 
569 	(void)memset(buf, 0, size > sizeof(buf) ? sizeof(buf) : size);
570 
571 	while (size > 0) {
572 		len = size > sizeof(buf) ? sizeof(buf) : size;
573 		if (!(*write_buf)(cookie, buf, len)) {
574 			_kvm_syserr(kd, kd->program, "clear_gap");
575 			return -1;
576 		}
577 		size -= len;
578 	}
579 
580 	return 0;
581 }
582 
583 /*
584  * Write the dump header by calling write_buf with cookie as first argument.
585  */
586 int
587 kvm_dump_header(kvm_t *kd, bool (*write_buf)(void *, const void *, size_t),
588     void *cookie, int dumpsize)
589 {
590 	kcore_seg_t	seghdr;
591 	long		offset;
592 	size_t		gap;
593 
594 	if (kd->kcore_hdr == NULL || kd->cpu_data == NULL) {
595 		_kvm_err(kd, kd->program, "no valid dump header(s)");
596 		return (-1);
597 	}
598 
599 	/*
600 	 * Write the generic header
601 	 */
602 	offset = 0;
603 	if (!(*write_buf)(cookie, kd->kcore_hdr, sizeof(kcore_hdr_t))) {
604 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
605 		return (-1);
606 	}
607 	offset += kd->kcore_hdr->c_hdrsize;
608 	gap     = kd->kcore_hdr->c_hdrsize - sizeof(kcore_hdr_t);
609 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
610 		return (-1);
611 
612 	/*
613 	 * Write the CPU header
614 	 */
615 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_CPU);
616 	seghdr.c_size = ALIGN(kd->cpu_dsize);
617 	if (!(*write_buf)(cookie, &seghdr, sizeof(seghdr))) {
618 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
619 		return (-1);
620 	}
621 	offset += kd->kcore_hdr->c_seghdrsize;
622 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
623 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
624 		return (-1);
625 
626 	if (!(*write_buf)(cookie, kd->cpu_data, kd->cpu_dsize)) {
627 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
628 		return (-1);
629 	}
630 	offset += seghdr.c_size;
631 	gap     = seghdr.c_size - kd->cpu_dsize;
632 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
633 		return (-1);
634 
635 	/*
636 	 * Write the actual dump data segment header
637 	 */
638 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_DATA);
639 	seghdr.c_size = dumpsize;
640 	if (!(*write_buf)(cookie, &seghdr, sizeof(seghdr))) {
641 		_kvm_syserr(kd, kd->program, "kvm_dump_header");
642 		return (-1);
643 	}
644 	offset += kd->kcore_hdr->c_seghdrsize;
645 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
646 	if (clear_gap(kd, write_buf, cookie, gap) == -1)
647 		return (-1);
648 
649 	return (int)offset;
650 }
651 
652 static bool
653 kvm_dump_header_stdio(void *cookie, const void *buf, size_t len)
654 {
655 	return fwrite(buf, len, 1, (FILE *)cookie) == 1;
656 }
657 
658 int
659 kvm_dump_wrtheader(kvm_t *kd, FILE *fp, int dumpsize)
660 {
661 	return kvm_dump_header(kd, kvm_dump_header_stdio, fp, dumpsize);
662 }
663 
664 kvm_t *
665 kvm_openfiles(const char *uf, const char *mf, const char *sf,
666     int flag, char *errout)
667 {
668 	kvm_t *kd;
669 
670 	if ((kd = malloc(sizeof(*kd))) == NULL) {
671 		(void)strlcpy(errout, strerror(errno), _POSIX2_LINE_MAX);
672 		return (0);
673 	}
674 	kd->program = 0;
675 	return (_kvm_open(kd, uf, mf, sf, flag, errout));
676 }
677 
678 kvm_t *
679 kvm_open(const char *uf, const char *mf, const char *sf, int flag,
680     const char *program)
681 {
682 	kvm_t *kd;
683 
684 	if ((kd = malloc(sizeof(*kd))) == NULL) {
685 		(void)fprintf(stderr, "%s: %s\n",
686 		    program ? program : getprogname(), strerror(errno));
687 		return (0);
688 	}
689 	kd->program = program;
690 	return (_kvm_open(kd, uf, mf, sf, flag, NULL));
691 }
692 
693 int
694 kvm_close(kvm_t *kd)
695 {
696 	int error = 0;
697 
698 	if (kd->pmfd >= 0)
699 		error |= close(kd->pmfd);
700 	if (kd->vmfd >= 0)
701 		error |= close(kd->vmfd);
702 	if (kd->nlfd >= 0)
703 		error |= close(kd->nlfd);
704 	if (kd->swfd >= 0)
705 		error |= close(kd->swfd);
706 	if (kd->vmst)
707 		_kvm_freevtop(kd);
708 	kd->cpu_dsize = 0;
709 	if (kd->cpu_data != NULL)
710 		free((void *)kd->cpu_data);
711 	if (kd->kcore_hdr != NULL)
712 		free((void *)kd->kcore_hdr);
713 	if (kd->procbase != 0)
714 		free((void *)kd->procbase);
715 	if (kd->procbase2 != 0)
716 		free((void *)kd->procbase2);
717 	if (kd->lwpbase != 0)
718 		free((void *)kd->lwpbase);
719 	if (kd->swapspc != 0)
720 		free((void *)kd->swapspc);
721 	if (kd->argspc != 0)
722 		free((void *)kd->argspc);
723 	if (kd->argbuf != 0)
724 		free((void *)kd->argbuf);
725 	if (kd->argv != 0)
726 		free((void *)kd->argv);
727 	free((void *)kd);
728 
729 	return (0);
730 }
731 
732 int
733 kvm_nlist(kvm_t *kd, struct nlist *nl)
734 {
735 	int rv, nlfd;
736 
737 	/*
738 	 * kd->nlfd might be negative when we get here, and in that
739 	 * case that means that we're using /dev/ksyms.
740 	 * So open it again, just for the time we retrieve the list.
741 	 */
742 	if (kd->nlfd < 0) {
743 		nlfd = open_cloexec(_PATH_KSYMS, O_RDONLY, 0);
744 		if (nlfd < 0) {
745 			_kvm_err(kd, 0, "failed to open %s", _PATH_KSYMS);
746 			return (nlfd);
747 		}
748 	} else
749 		nlfd = kd->nlfd;
750 
751 	/*
752 	 * Call the nlist(3) routines to retrieve the given namelist.
753 	 */
754 	rv = __fdnlist(nlfd, nl);
755 
756 	if (rv == -1)
757 		_kvm_err(kd, 0, "bad namelist");
758 
759 	if (kd->nlfd < 0)
760 		close(nlfd);
761 
762 	return (rv);
763 }
764 
765 int
766 kvm_dump_inval(kvm_t *kd)
767 {
768 	struct nlist	nl[2];
769 	u_long		pa, val;
770 
771 	if (ISALIVE(kd)) {
772 		_kvm_err(kd, kd->program, "clearing dump on live kernel");
773 		return (-1);
774 	}
775 	nl[0].n_name = "_dumpmag";
776 	nl[1].n_name = NULL;
777 
778 	if (kvm_nlist(kd, nl) == -1) {
779 		_kvm_err(kd, 0, "bad namelist");
780 		return (-1);
781 	}
782 	if (_kvm_kvatop(kd, (u_long)nl[0].n_value, &pa) == 0)
783 		return (-1);
784 
785 	errno = 0;
786 	val = 0;
787 	if (pwrite(kd->pmfd, (void *)&val, sizeof(val),
788 	    _kvm_pa2off(kd, pa)) == -1) {
789 		_kvm_syserr(kd, 0, "cannot invalidate dump - pwrite");
790 		return (-1);
791 	}
792 	return (0);
793 }
794 
795 ssize_t
796 kvm_read(kvm_t *kd, u_long kva, void *buf, size_t len)
797 {
798 	int cc;
799 	void *cp;
800 
801 	if (ISKMEM(kd)) {
802 		/*
803 		 * We're using /dev/kmem.  Just read straight from the
804 		 * device and let the active kernel do the address translation.
805 		 */
806 		errno = 0;
807 		cc = pread(kd->vmfd, buf, len, (off_t)kva);
808 		if (cc < 0) {
809 			_kvm_syserr(kd, 0, "kvm_read");
810 			return (-1);
811 		} else if (cc < len)
812 			_kvm_err(kd, kd->program, "short read");
813 		return (cc);
814 	} else if (ISSYSCTL(kd)) {
815 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
816 		    "can't use kvm_read");
817 		return (-1);
818 	} else {
819 		if ((kd->kcore_hdr == NULL) || (kd->cpu_data == NULL)) {
820 			_kvm_err(kd, kd->program, "no valid dump header");
821 			return (-1);
822 		}
823 		cp = buf;
824 		while (len > 0) {
825 			u_long	pa;
826 			off_t	foff;
827 
828 			cc = _kvm_kvatop(kd, kva, &pa);
829 			if (cc == 0)
830 				return (-1);
831 			if (cc > len)
832 				cc = len;
833 			foff = _kvm_pa2off(kd, pa);
834 			errno = 0;
835 			cc = pread(kd->pmfd, cp, (size_t)cc, foff);
836 			if (cc < 0) {
837 				_kvm_syserr(kd, kd->program, "kvm_read");
838 				break;
839 			}
840 			/*
841 			 * If kvm_kvatop returns a bogus value or our core
842 			 * file is truncated, we might wind up seeking beyond
843 			 * the end of the core file in which case the read will
844 			 * return 0 (EOF).
845 			 */
846 			if (cc == 0)
847 				break;
848 			cp = (char *)cp + cc;
849 			kva += cc;
850 			len -= cc;
851 		}
852 		return ((char *)cp - (char *)buf);
853 	}
854 	/* NOTREACHED */
855 }
856 
857 ssize_t
858 kvm_write(kvm_t *kd, u_long kva, const void *buf, size_t len)
859 {
860 	int cc;
861 
862 	if (ISKMEM(kd)) {
863 		/*
864 		 * Just like kvm_read, only we write.
865 		 */
866 		errno = 0;
867 		cc = pwrite(kd->vmfd, buf, len, (off_t)kva);
868 		if (cc < 0) {
869 			_kvm_syserr(kd, 0, "kvm_write");
870 			return (-1);
871 		} else if (cc < len)
872 			_kvm_err(kd, kd->program, "short write");
873 		return (cc);
874 	} else if (ISSYSCTL(kd)) {
875 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
876 		    "can't use kvm_write");
877 		return (-1);
878 	} else {
879 		_kvm_err(kd, kd->program,
880 		    "kvm_write not implemented for dead kernels");
881 		return (-1);
882 	}
883 	/* NOTREACHED */
884 }
885