Home | History | Annotate | Line # | Download | only in libkvm
kvm.c revision 1.76
      1 /*	$NetBSD: kvm.c,v 1.76 2002/09/17 20:34:08 atatat Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1989, 1992, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  *
      7  * This code is derived from software developed by the Computer Systems
      8  * Engineering group at Lawrence Berkeley Laboratory under DARPA contract
      9  * BG 91-66 and contributed to Berkeley.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  * 3. All advertising materials mentioning features or use of this software
     20  *    must display the following acknowledgement:
     21  *	This product includes software developed by the University of
     22  *	California, Berkeley and its contributors.
     23  * 4. Neither the name of the University nor the names of its contributors
     24  *    may be used to endorse or promote products derived from this software
     25  *    without specific prior written permission.
     26  *
     27  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     28  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     31  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     32  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     33  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     34  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     35  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     36  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     37  * SUCH DAMAGE.
     38  */
     39 
     40 #include <sys/cdefs.h>
     41 #if defined(LIBC_SCCS) && !defined(lint)
     42 #if 0
     43 static char sccsid[] = "@(#)kvm.c	8.2 (Berkeley) 2/13/94";
     44 #else
     45 __RCSID("$NetBSD: kvm.c,v 1.76 2002/09/17 20:34:08 atatat Exp $");
     46 #endif
     47 #endif /* LIBC_SCCS and not lint */
     48 
     49 #include <sys/param.h>
     50 #include <sys/user.h>
     51 #include <sys/proc.h>
     52 #include <sys/ioctl.h>
     53 #include <sys/stat.h>
     54 #include <sys/sysctl.h>
     55 
     56 #include <sys/core.h>
     57 #include <sys/exec_aout.h>
     58 #include <sys/kcore.h>
     59 
     60 #include <uvm/uvm_extern.h>
     61 
     62 #include <machine/cpu.h>
     63 
     64 #include <ctype.h>
     65 #include <db.h>
     66 #include <fcntl.h>
     67 #include <limits.h>
     68 #include <nlist.h>
     69 #include <paths.h>
     70 #include <stdarg.h>
     71 #include <stdio.h>
     72 #include <stdlib.h>
     73 #include <string.h>
     74 #include <unistd.h>
     75 #include <kvm.h>
     76 
     77 #include "kvm_private.h"
     78 
     79 static int	kvm_dbopen __P((kvm_t *));
     80 static int	_kvm_get_header __P((kvm_t *));
     81 static kvm_t	*_kvm_open __P((kvm_t *, const char *, const char *,
     82 		    const char *, int, char *));
     83 static int	clear_gap __P((kvm_t *, FILE *, int));
     84 static int	open_cloexec  __P((const char *, int, int));
     85 static off_t	Lseek __P((kvm_t *, int, off_t, int));
     86 static ssize_t	Pread __P((kvm_t *, int, void *, size_t, off_t));
     87 
     88 char *
     89 kvm_geterr(kd)
     90 	kvm_t *kd;
     91 {
     92 	return (kd->errbuf);
     93 }
     94 
     95 /*
     96  * Report an error using printf style arguments.  "program" is kd->program
     97  * on hard errors, and 0 on soft errors, so that under sun error emulation,
     98  * only hard errors are printed out (otherwise, programs like gdb will
     99  * generate tons of error messages when trying to access bogus pointers).
    100  */
    101 void
    102 _kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
    103 {
    104 	va_list ap;
    105 
    106 	va_start(ap, fmt);
    107 	if (program != NULL) {
    108 		(void)fprintf(stderr, "%s: ", program);
    109 		(void)vfprintf(stderr, fmt, ap);
    110 		(void)fputc('\n', stderr);
    111 	} else
    112 		(void)vsnprintf(kd->errbuf,
    113 		    sizeof(kd->errbuf), fmt, ap);
    114 
    115 	va_end(ap);
    116 }
    117 
    118 void
    119 _kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
    120 {
    121 	va_list ap;
    122 	size_t n;
    123 
    124 	va_start(ap, fmt);
    125 	if (program != NULL) {
    126 		(void)fprintf(stderr, "%s: ", program);
    127 		(void)vfprintf(stderr, fmt, ap);
    128 		(void)fprintf(stderr, ": %s\n", strerror(errno));
    129 	} else {
    130 		char *cp = kd->errbuf;
    131 
    132 		(void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
    133 		n = strlen(cp);
    134 		(void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
    135 		    strerror(errno));
    136 	}
    137 	va_end(ap);
    138 }
    139 
    140 void *
    141 _kvm_malloc(kd, n)
    142 	kvm_t *kd;
    143 	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  * Open a file setting the close on exec bit.
    154  */
    155 static int
    156 open_cloexec(fname, flags, mode)
    157 	const char *fname;
    158 	int flags, mode;
    159 {
    160 	int fd;
    161 
    162 	if ((fd = open(fname, flags, mode)) == -1)
    163 		return fd;
    164 	if (fcntl(fd, F_SETFD, (void *)1) == -1)
    165 		goto error;
    166 
    167 	return fd;
    168 error:
    169 	flags = errno;
    170 	(void)close(fd);
    171 	errno = flags;
    172 	return -1;
    173 }
    174 
    175 /*
    176  * Wrapper around the lseek(2) system call; calls _kvm_syserr() for us
    177  * in the event of emergency.
    178  */
    179 static off_t
    180 Lseek(kd, fd, offset, whence)
    181 	kvm_t *kd;
    182 	int fd;
    183 	off_t offset;
    184 	int whence;
    185 {
    186 	off_t off;
    187 
    188 	errno = 0;
    189 
    190 	if ((off = lseek(fd, offset, whence)) == -1 && errno != 0) {
    191 		_kvm_syserr(kd, kd->program, "Lseek");
    192 		return ((off_t)-1);
    193 	}
    194 	return (off);
    195 }
    196 
    197 /*
    198  * Wrapper around the pread(2) system call; calls _kvm_syserr() for us
    199  * in the event of emergency.
    200  */
    201 static ssize_t
    202 Pread(kd, fd, buf, nbytes, offset)
    203 	kvm_t *kd;
    204 	int fd;
    205 	void *buf;
    206 	size_t nbytes;
    207 	off_t offset;
    208 {
    209 	ssize_t rv;
    210 
    211 	errno = 0;
    212 
    213 	if ((rv = pread(fd, buf, nbytes, offset)) != nbytes &&
    214 	    errno != 0)
    215 		_kvm_syserr(kd, kd->program, "Pread");
    216 	return (rv);
    217 }
    218 
    219 static kvm_t *
    220 _kvm_open(kd, uf, mf, sf, flag, errout)
    221 	kvm_t *kd;
    222 	const char *uf;
    223 	const char *mf;
    224 	const char *sf;
    225 	int flag;
    226 	char *errout;
    227 {
    228 	struct stat st;
    229 	int ufgiven;
    230 
    231 	kd->db = 0;
    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 = 0;
    238 	kd->procbase2 = 0;
    239 	kd->nbpg = getpagesize();
    240 	kd->swapspc = 0;
    241 	kd->argspc = 0;
    242 	kd->arglen = 0;
    243 	kd->argbuf = 0;
    244 	kd->argv = 0;
    245 	kd->vmst = 0;
    246 	kd->vm_page_buckets = 0;
    247 	kd->kcore_hdr = 0;
    248 	kd->cpu_dsize = 0;
    249 	kd->cpu_data = 0;
    250 	kd->dump_off = 0;
    251 
    252 	if (flag & KVM_NO_FILES) {
    253 		kd->alive = KVM_ALIVE_SYSCTL;
    254 		return(kd);
    255 	}
    256 
    257 	/*
    258 	 * Call the MD open hook.  This sets:
    259 	 *	usrstack, min_uva, max_uva
    260 	 */
    261 	if (_kvm_mdopen(kd)) {
    262 		_kvm_err(kd, kd->program, "md init failed");
    263 		goto failed;
    264 	}
    265 
    266 	ufgiven = (uf != NULL);
    267 	if (!ufgiven) {
    268 #ifdef CPU_BOOTED_KERNEL
    269 		/* 130 is 128 + '/' + '\0' */
    270 		static char booted_kernel[130];
    271 		int mib[2], rc;
    272 		size_t len;
    273 
    274 		mib[0] = CTL_MACHDEP;
    275 		mib[1] = CPU_BOOTED_KERNEL;
    276 		booted_kernel[0] = '/';
    277 		booted_kernel[1] = '\0';
    278 		len = sizeof(booted_kernel) - 2;
    279 		rc = sysctl(&mib[0], 2, &booted_kernel[1], &len, NULL, 0);
    280 		booted_kernel[sizeof(booted_kernel) - 1] = '\0';
    281 		uf = (booted_kernel[1] == '/') ?
    282 		    &booted_kernel[1] : &booted_kernel[0];
    283 		if (rc != -1)
    284 			rc = stat(uf, &st);
    285 		if (rc != -1 && !S_ISREG(st.st_mode))
    286 			rc = -1;
    287 		if (rc == -1)
    288 #endif /* CPU_BOOTED_KERNEL */
    289 			uf = _PATH_UNIX;
    290 	}
    291 	else if (strlen(uf) >= MAXPATHLEN) {
    292 		_kvm_err(kd, kd->program, "exec file name too long");
    293 		goto failed;
    294 	}
    295 	if (flag & ~O_RDWR) {
    296 		_kvm_err(kd, kd->program, "bad flags arg");
    297 		goto failed;
    298 	}
    299 	if (mf == 0)
    300 		mf = _PATH_MEM;
    301 	if (sf == 0)
    302 		sf = _PATH_DRUM;
    303 
    304 	if ((kd->pmfd = open_cloexec(mf, flag, 0)) < 0) {
    305 		_kvm_syserr(kd, kd->program, "%s", mf);
    306 		goto failed;
    307 	}
    308 	if (fstat(kd->pmfd, &st) < 0) {
    309 		_kvm_syserr(kd, kd->program, "%s", mf);
    310 		goto failed;
    311 	}
    312 	if (S_ISCHR(st.st_mode)) {
    313 		/*
    314 		 * If this is a character special device, then check that
    315 		 * it's /dev/mem.  If so, open kmem too.  (Maybe we should
    316 		 * make it work for either /dev/mem or /dev/kmem -- in either
    317 		 * case you're working with a live kernel.)
    318 		 */
    319 		if (strcmp(mf, _PATH_MEM) != 0) {	/* XXX */
    320 			_kvm_err(kd, kd->program,
    321 				 "%s: not physical memory device", mf);
    322 			goto failed;
    323 		}
    324 		if ((kd->vmfd = open_cloexec(_PATH_KMEM, flag, 0)) < 0) {
    325 			_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
    326 			goto failed;
    327 		}
    328 		kd->alive = KVM_ALIVE_FILES;
    329 		if ((kd->swfd = open_cloexec(sf, flag, 0)) < 0) {
    330 			_kvm_syserr(kd, kd->program, "%s", sf);
    331 			goto failed;
    332 		}
    333 		/*
    334 		 * Open kvm nlist database.  We only try to use
    335 		 * the pre-built database if the namelist file name
    336 		 * pointer is NULL.  If the database cannot or should
    337 		 * not be opened, open the namelist argument so we
    338 		 * revert to slow nlist() calls.
    339 		 */
    340 		if ((ufgiven || kvm_dbopen(kd) < 0) &&
    341 		    (kd->nlfd = open_cloexec(uf, O_RDONLY, 0)) < 0) {
    342 			_kvm_syserr(kd, kd->program, "%s", uf);
    343 			goto failed;
    344 		}
    345 	} else {
    346 		/*
    347 		 * This is a crash dump.
    348 		 * Initialize the virtual address translation machinery,
    349 		 * but first setup the namelist fd.
    350 		 */
    351 		if ((kd->nlfd = open_cloexec(uf, O_RDONLY, 0)) < 0) {
    352 			_kvm_syserr(kd, kd->program, "%s", uf);
    353 			goto failed;
    354 		}
    355 
    356 		/*
    357 		 * If there is no valid core header, fail silently here.
    358 		 * The address translations however will fail without
    359 		 * header. Things can be made to run by calling
    360 		 * kvm_dump_mkheader() before doing any translation.
    361 		 */
    362 		if (_kvm_get_header(kd) == 0) {
    363 			if (_kvm_initvtop(kd) < 0)
    364 				goto failed;
    365 		}
    366 	}
    367 	return (kd);
    368 failed:
    369 	/*
    370 	 * Copy out the error if doing sane error semantics.
    371 	 */
    372 	if (errout != 0)
    373 		(void)strncpy(errout, kd->errbuf, _POSIX2_LINE_MAX - 1);
    374 	(void)kvm_close(kd);
    375 	return (0);
    376 }
    377 
    378 /*
    379  * The kernel dump file (from savecore) contains:
    380  *    kcore_hdr_t kcore_hdr;
    381  *    kcore_seg_t cpu_hdr;
    382  *    (opaque)    cpu_data; (size is cpu_hdr.c_size)
    383  *	  kcore_seg_t mem_hdr;
    384  *    (memory)    mem_data; (size is mem_hdr.c_size)
    385  *
    386  * Note: khdr is padded to khdr.c_hdrsize;
    387  * cpu_hdr and mem_hdr are padded to khdr.c_seghdrsize
    388  */
    389 static int
    390 _kvm_get_header(kd)
    391 	kvm_t	*kd;
    392 {
    393 	kcore_hdr_t	kcore_hdr;
    394 	kcore_seg_t	cpu_hdr;
    395 	kcore_seg_t	mem_hdr;
    396 	size_t		offset;
    397 	ssize_t		sz;
    398 
    399 	/*
    400 	 * Read the kcore_hdr_t
    401 	 */
    402 	sz = Pread(kd, kd->pmfd, &kcore_hdr, sizeof(kcore_hdr), (off_t)0);
    403 	if (sz != sizeof(kcore_hdr))
    404 		return (-1);
    405 
    406 	/*
    407 	 * Currently, we only support dump-files made by the current
    408 	 * architecture...
    409 	 */
    410 	if ((CORE_GETMAGIC(kcore_hdr) != KCORE_MAGIC) ||
    411 	    (CORE_GETMID(kcore_hdr) != MID_MACHINE))
    412 		return (-1);
    413 
    414 	/*
    415 	 * Currently, we only support exactly 2 segments: cpu-segment
    416 	 * and data-segment in exactly that order.
    417 	 */
    418 	if (kcore_hdr.c_nseg != 2)
    419 		return (-1);
    420 
    421 	/*
    422 	 * Save away the kcore_hdr.  All errors after this
    423 	 * should do a to "goto fail" to deallocate things.
    424 	 */
    425 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr));
    426 	memcpy(kd->kcore_hdr, &kcore_hdr, sizeof(kcore_hdr));
    427 	offset = kcore_hdr.c_hdrsize;
    428 
    429 	/*
    430 	 * Read the CPU segment header
    431 	 */
    432 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), (off_t)offset);
    433 	if (sz != sizeof(cpu_hdr))
    434 		goto fail;
    435 	if ((CORE_GETMAGIC(cpu_hdr) != KCORESEG_MAGIC) ||
    436 	    (CORE_GETFLAG(cpu_hdr) != CORE_CPU))
    437 		goto fail;
    438 	offset += kcore_hdr.c_seghdrsize;
    439 
    440 	/*
    441 	 * Read the CPU segment DATA.
    442 	 */
    443 	kd->cpu_dsize = cpu_hdr.c_size;
    444 	kd->cpu_data = _kvm_malloc(kd, cpu_hdr.c_size);
    445 	if (kd->cpu_data == NULL)
    446 		goto fail;
    447 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size, (off_t)offset);
    448 	if (sz != cpu_hdr.c_size)
    449 		goto fail;
    450 	offset += cpu_hdr.c_size;
    451 
    452 	/*
    453 	 * Read the next segment header: data segment
    454 	 */
    455 	sz = Pread(kd, kd->pmfd, &mem_hdr, sizeof(mem_hdr), (off_t)offset);
    456 	if (sz != sizeof(mem_hdr))
    457 		goto fail;
    458 	offset += kcore_hdr.c_seghdrsize;
    459 
    460 	if ((CORE_GETMAGIC(mem_hdr) != KCORESEG_MAGIC) ||
    461 	    (CORE_GETFLAG(mem_hdr) != CORE_DATA))
    462 		goto fail;
    463 
    464 	kd->dump_off = offset;
    465 	return (0);
    466 
    467 fail:
    468 	if (kd->kcore_hdr != NULL) {
    469 		free(kd->kcore_hdr);
    470 		kd->kcore_hdr = NULL;
    471 	}
    472 	if (kd->cpu_data != NULL) {
    473 		free(kd->cpu_data);
    474 		kd->cpu_data = NULL;
    475 		kd->cpu_dsize = 0;
    476 	}
    477 	return (-1);
    478 }
    479 
    480 /*
    481  * The format while on the dump device is: (new format)
    482  *	kcore_seg_t cpu_hdr;
    483  *	(opaque)    cpu_data; (size is cpu_hdr.c_size)
    484  *	kcore_seg_t mem_hdr;
    485  *	(memory)    mem_data; (size is mem_hdr.c_size)
    486  */
    487 int
    488 kvm_dump_mkheader(kd, dump_off)
    489 kvm_t	*kd;
    490 off_t	dump_off;
    491 {
    492 	kcore_seg_t	cpu_hdr;
    493 	size_t hdr_size;
    494 	ssize_t sz;
    495 
    496 	if (kd->kcore_hdr != NULL) {
    497 	    _kvm_err(kd, kd->program, "already has a dump header");
    498 	    return (-1);
    499 	}
    500 	if (ISALIVE(kd)) {
    501 		_kvm_err(kd, kd->program, "don't use on live kernel");
    502 		return (-1);
    503 	}
    504 
    505 	/*
    506 	 * Validate new format crash dump
    507 	 */
    508 	sz = Pread(kd, kd->pmfd, &cpu_hdr, sizeof(cpu_hdr), dump_off);
    509 	if (sz != sizeof(cpu_hdr))
    510 		return (-1);
    511 	if ((CORE_GETMAGIC(cpu_hdr) != KCORE_MAGIC)
    512 		|| (CORE_GETMID(cpu_hdr) != MID_MACHINE)) {
    513 		_kvm_err(kd, 0, "invalid magic in cpu_hdr");
    514 		return (0);
    515 	}
    516 	hdr_size = ALIGN(sizeof(cpu_hdr));
    517 
    518 	/*
    519 	 * Read the CPU segment.
    520 	 */
    521 	kd->cpu_dsize = cpu_hdr.c_size;
    522 	kd->cpu_data = _kvm_malloc(kd, kd->cpu_dsize);
    523 	if (kd->cpu_data == NULL)
    524 		goto fail;
    525 	sz = Pread(kd, kd->pmfd, kd->cpu_data, cpu_hdr.c_size,
    526 	    dump_off + hdr_size);
    527 	if (sz != cpu_hdr.c_size)
    528 		goto fail;
    529 	hdr_size += kd->cpu_dsize;
    530 
    531 	/*
    532 	 * Leave phys mem pointer at beginning of memory data
    533 	 */
    534 	kd->dump_off = dump_off + hdr_size;
    535 	if (Lseek(kd, kd->pmfd, kd->dump_off, SEEK_SET) == -1)
    536 		goto fail;
    537 
    538 	/*
    539 	 * Create a kcore_hdr.
    540 	 */
    541 	kd->kcore_hdr = _kvm_malloc(kd, sizeof(kcore_hdr_t));
    542 	if (kd->kcore_hdr == NULL)
    543 		goto fail;
    544 
    545 	kd->kcore_hdr->c_hdrsize    = ALIGN(sizeof(kcore_hdr_t));
    546 	kd->kcore_hdr->c_seghdrsize = ALIGN(sizeof(kcore_seg_t));
    547 	kd->kcore_hdr->c_nseg       = 2;
    548 	CORE_SETMAGIC(*(kd->kcore_hdr), KCORE_MAGIC, MID_MACHINE,0);
    549 
    550 	/*
    551 	 * Now that we have a valid header, enable translations.
    552 	 */
    553 	if (_kvm_initvtop(kd) == 0)
    554 		/* Success */
    555 		return (hdr_size);
    556 
    557 fail:
    558 	if (kd->kcore_hdr != NULL) {
    559 		free(kd->kcore_hdr);
    560 		kd->kcore_hdr = NULL;
    561 	}
    562 	if (kd->cpu_data != NULL) {
    563 		free(kd->cpu_data);
    564 		kd->cpu_data = NULL;
    565 		kd->cpu_dsize = 0;
    566 	}
    567 	return (-1);
    568 }
    569 
    570 static int
    571 clear_gap(kd, fp, size)
    572 kvm_t	*kd;
    573 FILE	*fp;
    574 int	size;
    575 {
    576 	if (size <= 0) /* XXX - < 0 should never happen */
    577 		return (0);
    578 	while (size-- > 0) {
    579 		if (fputc(0, fp) == EOF) {
    580 			_kvm_syserr(kd, kd->program, "clear_gap");
    581 			return (-1);
    582 		}
    583 	}
    584 	return (0);
    585 }
    586 
    587 /*
    588  * Write the dump header info to 'fp'. Note that we can't use fseek(3) here
    589  * because 'fp' might be a file pointer obtained by zopen().
    590  */
    591 int
    592 kvm_dump_wrtheader(kd, fp, dumpsize)
    593 kvm_t	*kd;
    594 FILE	*fp;
    595 int	dumpsize;
    596 {
    597 	kcore_seg_t	seghdr;
    598 	long		offset;
    599 	int		gap;
    600 
    601 	if (kd->kcore_hdr == NULL || kd->cpu_data == NULL) {
    602 		_kvm_err(kd, kd->program, "no valid dump header(s)");
    603 		return (-1);
    604 	}
    605 
    606 	/*
    607 	 * Write the generic header
    608 	 */
    609 	offset = 0;
    610 	if (fwrite((void*)kd->kcore_hdr, sizeof(kcore_hdr_t), 1, fp) == 0) {
    611 		_kvm_syserr(kd, kd->program, "kvm_dump_wrtheader");
    612 		return (-1);
    613 	}
    614 	offset += kd->kcore_hdr->c_hdrsize;
    615 	gap     = kd->kcore_hdr->c_hdrsize - sizeof(kcore_hdr_t);
    616 	if (clear_gap(kd, fp, gap) == -1)
    617 		return (-1);
    618 
    619 	/*
    620 	 * Write the cpu header
    621 	 */
    622 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_CPU);
    623 	seghdr.c_size = ALIGN(kd->cpu_dsize);
    624 	if (fwrite((void*)&seghdr, sizeof(seghdr), 1, fp) == 0) {
    625 		_kvm_syserr(kd, kd->program, "kvm_dump_wrtheader");
    626 		return (-1);
    627 	}
    628 	offset += kd->kcore_hdr->c_seghdrsize;
    629 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
    630 	if (clear_gap(kd, fp, gap) == -1)
    631 		return (-1);
    632 
    633 	if (fwrite((void*)kd->cpu_data, kd->cpu_dsize, 1, fp) == 0) {
    634 		_kvm_syserr(kd, kd->program, "kvm_dump_wrtheader");
    635 		return (-1);
    636 	}
    637 	offset += seghdr.c_size;
    638 	gap     = seghdr.c_size - kd->cpu_dsize;
    639 	if (clear_gap(kd, fp, gap) == -1)
    640 		return (-1);
    641 
    642 	/*
    643 	 * Write the actual dump data segment header
    644 	 */
    645 	CORE_SETMAGIC(seghdr, KCORESEG_MAGIC, 0, CORE_DATA);
    646 	seghdr.c_size = dumpsize;
    647 	if (fwrite((void*)&seghdr, sizeof(seghdr), 1, fp) == 0) {
    648 		_kvm_syserr(kd, kd->program, "kvm_dump_wrtheader");
    649 		return (-1);
    650 	}
    651 	offset += kd->kcore_hdr->c_seghdrsize;
    652 	gap     = kd->kcore_hdr->c_seghdrsize - sizeof(seghdr);
    653 	if (clear_gap(kd, fp, gap) == -1)
    654 		return (-1);
    655 
    656 	return (int)offset;
    657 }
    658 
    659 kvm_t *
    660 kvm_openfiles(uf, mf, sf, flag, errout)
    661 	const char *uf;
    662 	const char *mf;
    663 	const char *sf;
    664 	int flag;
    665 	char *errout;
    666 {
    667 	kvm_t *kd;
    668 
    669 	if ((kd = malloc(sizeof(*kd))) == NULL) {
    670 		(void)strncpy(errout, strerror(errno), _POSIX2_LINE_MAX - 1);
    671 		return (0);
    672 	}
    673 	kd->program = 0;
    674 	return (_kvm_open(kd, uf, mf, sf, flag, errout));
    675 }
    676 
    677 kvm_t *
    678 kvm_open(uf, mf, sf, flag, program)
    679 	const char *uf;
    680 	const char *mf;
    681 	const char *sf;
    682 	int flag;
    683 	const char *program;
    684 {
    685 	kvm_t *kd;
    686 
    687 	if ((kd = malloc(sizeof(*kd))) == NULL && program != NULL) {
    688 		(void)fprintf(stderr, "%s: %s\n", program, strerror(errno));
    689 		return (0);
    690 	}
    691 	kd->program = program;
    692 	return (_kvm_open(kd, uf, mf, sf, flag, NULL));
    693 }
    694 
    695 int
    696 kvm_close(kd)
    697 	kvm_t *kd;
    698 {
    699 	int error = 0;
    700 
    701 	if (kd->pmfd >= 0)
    702 		error |= close(kd->pmfd);
    703 	if (kd->vmfd >= 0)
    704 		error |= close(kd->vmfd);
    705 	if (kd->nlfd >= 0)
    706 		error |= close(kd->nlfd);
    707 	if (kd->swfd >= 0)
    708 		error |= close(kd->swfd);
    709 	if (kd->db != 0)
    710 		error |= (kd->db->close)(kd->db);
    711 	if (kd->vmst)
    712 		_kvm_freevtop(kd);
    713 	kd->cpu_dsize = 0;
    714 	if (kd->cpu_data != NULL)
    715 		free((void *)kd->cpu_data);
    716 	if (kd->kcore_hdr != NULL)
    717 		free((void *)kd->kcore_hdr);
    718 	if (kd->procbase != 0)
    719 		free((void *)kd->procbase);
    720 	if (kd->procbase2 != 0)
    721 		free((void *)kd->procbase2);
    722 	if (kd->swapspc != 0)
    723 		free((void *)kd->swapspc);
    724 	if (kd->argspc != 0)
    725 		free((void *)kd->argspc);
    726 	if (kd->argbuf != 0)
    727 		free((void *)kd->argbuf);
    728 	if (kd->argv != 0)
    729 		free((void *)kd->argv);
    730 	free((void *)kd);
    731 
    732 	return (0);
    733 }
    734 
    735 /*
    736  * Set up state necessary to do queries on the kernel namelist
    737  * data base.  If the data base is out-of-data/incompatible with
    738  * given executable, set up things so we revert to standard nlist call.
    739  * Only called for live kernels.  Return 0 on success, -1 on failure.
    740  */
    741 static int
    742 kvm_dbopen(kd)
    743 	kvm_t *kd;
    744 {
    745 	DBT rec;
    746 	size_t dbversionlen;
    747 	struct nlist nitem;
    748 	char dbversion[_POSIX2_LINE_MAX];
    749 	char kversion[_POSIX2_LINE_MAX];
    750 	int fd;
    751 
    752 	kd->db = dbopen(_PATH_KVMDB, O_RDONLY, 0, DB_HASH, NULL);
    753 	if (kd->db == 0)
    754 		return (-1);
    755 	if ((fd = (*kd->db->fd)(kd->db)) >= 0) {
    756 		if (fcntl(fd, F_SETFD, (void *)1) == -1) {
    757 		       (*kd->db->close)(kd->db);
    758 		       return (-1);
    759 		}
    760 	}
    761 	/*
    762 	 * read version out of database
    763 	 */
    764 	rec.data = VRS_KEY;
    765 	rec.size = sizeof(VRS_KEY) - 1;
    766 	if ((kd->db->get)(kd->db, (DBT *)&rec, (DBT *)&rec, 0))
    767 		goto close;
    768 	if (rec.data == 0 || rec.size > sizeof(dbversion))
    769 		goto close;
    770 
    771 	memcpy(dbversion, rec.data, rec.size);
    772 	dbversionlen = rec.size;
    773 	/*
    774 	 * Read version string from kernel memory.
    775 	 * Since we are dealing with a live kernel, we can call kvm_read()
    776 	 * at this point.
    777 	 */
    778 	rec.data = VRS_SYM;
    779 	rec.size = sizeof(VRS_SYM) - 1;
    780 	if ((kd->db->get)(kd->db, (DBT *)&rec, (DBT *)&rec, 0))
    781 		goto close;
    782 	if (rec.data == 0 || rec.size != sizeof(struct nlist))
    783 		goto close;
    784 	memcpy(&nitem, rec.data, sizeof(nitem));
    785 	if (kvm_read(kd, (u_long)nitem.n_value, kversion, dbversionlen) !=
    786 	    dbversionlen)
    787 		goto close;
    788 	/*
    789 	 * If they match, we win - otherwise clear out kd->db so
    790 	 * we revert to slow nlist().
    791 	 */
    792 	if (memcmp(dbversion, kversion, dbversionlen) == 0)
    793 		return (0);
    794 close:
    795 	(void)(kd->db->close)(kd->db);
    796 	kd->db = 0;
    797 
    798 	return (-1);
    799 }
    800 
    801 int
    802 kvm_nlist(kd, nl)
    803 	kvm_t *kd;
    804 	struct nlist *nl;
    805 {
    806 	struct nlist *p;
    807 	int nvalid, rv;
    808 
    809 	/*
    810 	 * If we can't use the data base, revert to the
    811 	 * slow library call.
    812 	 */
    813 	if (kd->db == 0) {
    814 		rv = __fdnlist(kd->nlfd, nl);
    815 		if (rv == -1)
    816 			_kvm_err(kd, 0, "bad namelist");
    817 		return (rv);
    818 	}
    819 
    820 	/*
    821 	 * We can use the kvm data base.  Go through each nlist entry
    822 	 * and look it up with a db query.
    823 	 */
    824 	nvalid = 0;
    825 	for (p = nl; p->n_name && p->n_name[0]; ++p) {
    826 		int len;
    827 		DBT rec;
    828 
    829 		if ((len = strlen(p->n_name)) > 4096) {
    830 			/* sanity */
    831 			_kvm_err(kd, kd->program, "symbol too large");
    832 			return (-1);
    833 		}
    834 		rec.data = (char *)p->n_name;
    835 		rec.size = len;
    836 
    837 		/*
    838 		 * Make sure that n_value = 0 when the symbol isn't found
    839 		 */
    840 		p->n_value = 0;
    841 
    842 		if ((kd->db->get)(kd->db, (DBT *)&rec, (DBT *)&rec, 0))
    843 			continue;
    844 		if (rec.data == 0 || rec.size != sizeof(struct nlist))
    845 			continue;
    846 		++nvalid;
    847 		/*
    848 		 * Avoid alignment issues.
    849 		 */
    850 		(void)memcpy(&p->n_type, &((struct nlist *)rec.data)->n_type,
    851 		      sizeof(p->n_type));
    852 		(void)memcpy(&p->n_value, &((struct nlist *)rec.data)->n_value,
    853 		      sizeof(p->n_value));
    854 	}
    855 	/*
    856 	 * Return the number of entries that weren't found.
    857 	 */
    858 	return ((p - nl) - nvalid);
    859 }
    860 
    861 int kvm_dump_inval(kd)
    862 kvm_t	*kd;
    863 {
    864 	struct nlist	nl[2];
    865 	u_long		pa, val;
    866 
    867 	if (ISALIVE(kd)) {
    868 		_kvm_err(kd, kd->program, "clearing dump on live kernel");
    869 		return (-1);
    870 	}
    871 	nl[0].n_name = "_dumpmag";
    872 	nl[1].n_name = NULL;
    873 
    874 	if (kvm_nlist(kd, nl) == -1) {
    875 		_kvm_err(kd, 0, "bad namelist");
    876 		return (-1);
    877 	}
    878 	if (_kvm_kvatop(kd, (u_long)nl[0].n_value, &pa) == 0)
    879 		return (-1);
    880 
    881 	errno = 0;
    882 	val = 0;
    883 	if (pwrite(kd->pmfd, (void *) &val, sizeof(val),
    884 	    _kvm_pa2off(kd, pa)) == -1) {
    885 		_kvm_syserr(kd, 0, "cannot invalidate dump - pwrite");
    886 		return (-1);
    887 	}
    888 	return (0);
    889 }
    890 
    891 ssize_t
    892 kvm_read(kd, kva, buf, len)
    893 	kvm_t *kd;
    894 	u_long kva;
    895 	void *buf;
    896 	size_t len;
    897 {
    898 	int cc;
    899 	void *cp;
    900 
    901 	if (ISKMEM(kd)) {
    902 		/*
    903 		 * We're using /dev/kmem.  Just read straight from the
    904 		 * device and let the active kernel do the address translation.
    905 		 */
    906 		errno = 0;
    907 		cc = pread(kd->vmfd, buf, len, (off_t)kva);
    908 		if (cc < 0) {
    909 			_kvm_syserr(kd, 0, "kvm_read");
    910 			return (-1);
    911 		} else if (cc < len)
    912 			_kvm_err(kd, kd->program, "short read");
    913 		return (cc);
    914 	} else if (ISSYSCTL(kd)) {
    915 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
    916 		    "can't use kvm_read");
    917 		return (-1);
    918 	} else {
    919 		if ((kd->kcore_hdr == NULL) || (kd->cpu_data == NULL)) {
    920 			_kvm_err(kd, kd->program, "no valid dump header");
    921 			return (-1);
    922 		}
    923 		cp = buf;
    924 		while (len > 0) {
    925 			u_long	pa;
    926 			off_t	foff;
    927 
    928 			cc = _kvm_kvatop(kd, kva, &pa);
    929 			if (cc == 0)
    930 				return (-1);
    931 			if (cc > len)
    932 				cc = len;
    933 			foff = _kvm_pa2off(kd, pa);
    934 			errno = 0;
    935 			cc = pread(kd->pmfd, cp, (size_t)cc, foff);
    936 			if (cc < 0) {
    937 				_kvm_syserr(kd, kd->program, "kvm_read");
    938 				break;
    939 			}
    940 			/*
    941 			 * If kvm_kvatop returns a bogus value or our core
    942 			 * file is truncated, we might wind up seeking beyond
    943 			 * the end of the core file in which case the read will
    944 			 * return 0 (EOF).
    945 			 */
    946 			if (cc == 0)
    947 				break;
    948 			cp = (char *)cp + cc;
    949 			kva += cc;
    950 			len -= cc;
    951 		}
    952 		return ((char *)cp - (char *)buf);
    953 	}
    954 	/* NOTREACHED */
    955 }
    956 
    957 ssize_t
    958 kvm_write(kd, kva, buf, len)
    959 	kvm_t *kd;
    960 	u_long kva;
    961 	const void *buf;
    962 	size_t len;
    963 {
    964 	int cc;
    965 
    966 	if (ISKMEM(kd)) {
    967 		/*
    968 		 * Just like kvm_read, only we write.
    969 		 */
    970 		errno = 0;
    971 		cc = pwrite(kd->vmfd, buf, len, (off_t)kva);
    972 		if (cc < 0) {
    973 			_kvm_syserr(kd, 0, "kvm_write");
    974 			return (-1);
    975 		} else if (cc < len)
    976 			_kvm_err(kd, kd->program, "short write");
    977 		return (cc);
    978 	} else if (ISSYSCTL(kd)) {
    979 		_kvm_err(kd, kd->program, "kvm_open called with KVM_NO_FILES, "
    980 		    "can't use kvm_write");
    981 		return (-1);
    982 	} else {
    983 		_kvm_err(kd, kd->program,
    984 		    "kvm_write not implemented for dead kernels");
    985 		return (-1);
    986 	}
    987 	/* NOTREACHED */
    988 }
    989