1/*-
2 * Copyright (c) 1989, 1992, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software developed by the Computer Systems
6 * Engineering group at Lawrence Berkeley Laboratory under DARPA contract
7 * BG 91-66 and contributed to Berkeley.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 * 4. Neither the name of the University nor the names of its contributors
18 *    may be used to endorse or promote products derived from this software
19 *    without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD$");
36
37#if defined(LIBC_SCCS) && !defined(lint)
38#if 0
39static char sccsid[] = "@(#)kvm.c	8.2 (Berkeley) 2/13/94";
40#endif
41#endif /* LIBC_SCCS and not lint */
42
43#include <sys/param.h>
44
45#define	_WANT_VNET
46
47#include <sys/user.h>
48#include <sys/proc.h>
49#include <sys/ioctl.h>
50#include <sys/stat.h>
51#include <sys/sysctl.h>
52#include <sys/linker.h>
53#include <sys/pcpu.h>
54
55#include <net/vnet.h>
56
57#include <vm/vm.h>
58#include <vm/vm_param.h>
59
60#include <machine/vmparam.h>
61
62#include <ctype.h>
63#include <fcntl.h>
64#include <kvm.h>
65#include <limits.h>
66#include <nlist.h>
67#include <paths.h>
68#include <stdio.h>
69#include <stdlib.h>
70#include <string.h>
71#include <strings.h>
72#include <unistd.h>
73
74#include "kvm_private.h"
75
76/* from src/lib/libc/gen/nlist.c */
77int __fdnlist(int, struct nlist *);
78
79char *
80kvm_geterr(kvm_t *kd)
81{
82	return (kd->errbuf);
83}
84
85#include <stdarg.h>
86
87/*
88 * Report an error using printf style arguments.  "program" is kd->program
89 * on hard errors, and 0 on soft errors, so that under sun error emulation,
90 * only hard errors are printed out (otherwise, programs like gdb will
91 * generate tons of error messages when trying to access bogus pointers).
92 */
93void
94_kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
95{
96	va_list ap;
97
98	va_start(ap, fmt);
99	if (program != NULL) {
100		(void)fprintf(stderr, "%s: ", program);
101		(void)vfprintf(stderr, fmt, ap);
102		(void)fputc('\n', stderr);
103	} else
104		(void)vsnprintf(kd->errbuf,
105		    sizeof(kd->errbuf), fmt, ap);
106
107	va_end(ap);
108}
109
110void
111_kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
112{
113	va_list ap;
114	int n;
115
116	va_start(ap, fmt);
117	if (program != NULL) {
118		(void)fprintf(stderr, "%s: ", program);
119		(void)vfprintf(stderr, fmt, ap);
120		(void)fprintf(stderr, ": %s\n", strerror(errno));
121	} else {
122		char *cp = kd->errbuf;
123
124		(void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
125		n = strlen(cp);
126		(void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
127		    strerror(errno));
128	}
129	va_end(ap);
130}
131
132void *
133_kvm_malloc(kvm_t *kd, size_t n)
134{
135	void *p;
136
137	if ((p = calloc(n, sizeof(char))) == NULL)
138		_kvm_err(kd, kd->program, "can't allocate %zu bytes: %s",
139			 n, strerror(errno));
140	return (p);
141}
142
143static kvm_t *
144_kvm_open(kvm_t *kd, const char *uf, const char *mf, int flag, char *errout)
145{
146	struct stat st;
147
148	kd->vmfd = -1;
149	kd->pmfd = -1;
150	kd->nlfd = -1;
151	kd->vmst = 0;
152	kd->procbase = 0;
153	kd->argspc = 0;
154	kd->argv = 0;
155
156	if (uf == 0)
157		uf = getbootfile();
158	else if (strlen(uf) >= MAXPATHLEN) {
159		_kvm_err(kd, kd->program, "exec file name too long");
160		goto failed;
161	}
162	if (flag & ~O_RDWR) {
163		_kvm_err(kd, kd->program, "bad flags arg");
164		goto failed;
165	}
166	if (mf == 0)
167		mf = _PATH_MEM;
168
169	if ((kd->pmfd = open(mf, flag | O_CLOEXEC, 0)) < 0) {
170		_kvm_syserr(kd, kd->program, "%s", mf);
171		goto failed;
172	}
173	if (fstat(kd->pmfd, &st) < 0) {
174		_kvm_syserr(kd, kd->program, "%s", mf);
175		goto failed;
176	}
177	if (S_ISREG(st.st_mode) && st.st_size <= 0) {
178		errno = EINVAL;
179		_kvm_syserr(kd, kd->program, "empty file");
180		goto failed;
181	}
182	if (S_ISCHR(st.st_mode)) {
183		/*
184		 * If this is a character special device, then check that
185		 * it's /dev/mem.  If so, open kmem too.  (Maybe we should
186		 * make it work for either /dev/mem or /dev/kmem -- in either
187		 * case you're working with a live kernel.)
188		 */
189		if (strcmp(mf, _PATH_DEVNULL) == 0) {
190			kd->vmfd = open(_PATH_DEVNULL, O_RDONLY | O_CLOEXEC);
191			return (kd);
192		} else if (strcmp(mf, _PATH_MEM) == 0) {
193			if ((kd->vmfd = open(_PATH_KMEM, flag | O_CLOEXEC)) <
194			    0) {
195				_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
196				goto failed;
197			}
198			return (kd);
199		}
200	}
201	/*
202	 * This is a crash dump.
203	 * Initialize the virtual address translation machinery,
204	 * but first setup the namelist fd.
205	 */
206	if ((kd->nlfd = open(uf, O_RDONLY | O_CLOEXEC, 0)) < 0) {
207		_kvm_syserr(kd, kd->program, "%s", uf);
208		goto failed;
209	}
210	if (strncmp(mf, _PATH_FWMEM, strlen(_PATH_FWMEM)) == 0)
211		kd->rawdump = 1;
212	if (_kvm_initvtop(kd) < 0)
213		goto failed;
214	return (kd);
215failed:
216	/*
217	 * Copy out the error if doing sane error semantics.
218	 */
219	if (errout != 0)
220		strlcpy(errout, kd->errbuf, _POSIX2_LINE_MAX);
221	(void)kvm_close(kd);
222	return (0);
223}
224
225kvm_t *
226kvm_openfiles(const char *uf, const char *mf, const char *sf __unused, int flag,
227    char *errout)
228{
229	kvm_t *kd;
230
231	if ((kd = calloc(1, sizeof(*kd))) == NULL) {
232		(void)strlcpy(errout, strerror(errno), _POSIX2_LINE_MAX);
233		return (0);
234	}
235	kd->program = 0;
236	return (_kvm_open(kd, uf, mf, flag, errout));
237}
238
239kvm_t *
240kvm_open(const char *uf, const char *mf, const char *sf __unused, int flag,
241    const char *errstr)
242{
243	kvm_t *kd;
244
245	if ((kd = calloc(1, sizeof(*kd))) == NULL) {
246		if (errstr != NULL)
247			(void)fprintf(stderr, "%s: %s\n",
248				      errstr, strerror(errno));
249		return (0);
250	}
251	kd->program = errstr;
252	return (_kvm_open(kd, uf, mf, flag, NULL));
253}
254
255int
256kvm_close(kvm_t *kd)
257{
258	int error = 0;
259
260	if (kd->pmfd >= 0)
261		error |= close(kd->pmfd);
262	if (kd->vmfd >= 0)
263		error |= close(kd->vmfd);
264	if (kd->nlfd >= 0)
265		error |= close(kd->nlfd);
266	if (kd->vmst)
267		_kvm_freevtop(kd);
268	if (kd->procbase != 0)
269		free((void *)kd->procbase);
270	if (kd->argbuf != 0)
271		free((void *) kd->argbuf);
272	if (kd->argspc != 0)
273		free((void *) kd->argspc);
274	if (kd->argv != 0)
275		free((void *)kd->argv);
276	free((void *)kd);
277
278	return (0);
279}
280
281/*
282 * Walk the list of unresolved symbols, generate a new list and prefix the
283 * symbol names, try again, and merge back what we could resolve.
284 */
285static int
286kvm_fdnlist_prefix(kvm_t *kd, struct nlist *nl, int missing, const char *prefix,
287    uintptr_t (*validate_fn)(kvm_t *, uintptr_t))
288{
289	struct nlist *n, *np, *p;
290	char *cp, *ce;
291	const char *ccp;
292	size_t len;
293	int slen, unresolved;
294
295	/*
296	 * Calculate the space we need to malloc for nlist and names.
297	 * We are going to store the name twice for later lookups: once
298	 * with the prefix and once the unmodified name delmited by \0.
299	 */
300	len = 0;
301	unresolved = 0;
302	for (p = nl; p->n_name && p->n_name[0]; ++p) {
303		if (p->n_type != N_UNDF)
304			continue;
305		len += sizeof(struct nlist) + strlen(prefix) +
306		    2 * (strlen(p->n_name) + 1);
307		unresolved++;
308	}
309	if (unresolved == 0)
310		return (unresolved);
311	/* Add space for the terminating nlist entry. */
312	len += sizeof(struct nlist);
313	unresolved++;
314
315	/* Alloc one chunk for (nlist, [names]) and setup pointers. */
316	n = np = malloc(len);
317	bzero(n, len);
318	if (n == NULL)
319		return (missing);
320	cp = ce = (char *)np;
321	cp += unresolved * sizeof(struct nlist);
322	ce += len;
323
324	/* Generate shortened nlist with special prefix. */
325	unresolved = 0;
326	for (p = nl; p->n_name && p->n_name[0]; ++p) {
327		if (p->n_type != N_UNDF)
328			continue;
329		bcopy(p, np, sizeof(struct nlist));
330		/* Save the new\0orig. name so we can later match it again. */
331		slen = snprintf(cp, ce - cp, "%s%s%c%s", prefix,
332		    (prefix[0] != '\0' && p->n_name[0] == '_') ?
333			(p->n_name + 1) : p->n_name, '\0', p->n_name);
334		if (slen < 0 || slen >= ce - cp)
335			continue;
336		np->n_name = cp;
337		cp += slen + 1;
338		np++;
339		unresolved++;
340	}
341
342	/* Do lookup on the reduced list. */
343	np = n;
344	unresolved = __fdnlist(kd->nlfd, np);
345
346	/* Check if we could resolve further symbols and update the list. */
347	if (unresolved >= 0 && unresolved < missing) {
348		/* Find the first freshly resolved entry. */
349		for (; np->n_name && np->n_name[0]; np++)
350			if (np->n_type != N_UNDF)
351				break;
352		/*
353		 * The lists are both in the same order,
354		 * so we can walk them in parallel.
355		 */
356		for (p = nl; np->n_name && np->n_name[0] &&
357		    p->n_name && p->n_name[0]; ++p) {
358			if (p->n_type != N_UNDF)
359				continue;
360			/* Skip expanded name and compare to orig. one. */
361			ccp = np->n_name + strlen(np->n_name) + 1;
362			if (strcmp(ccp, p->n_name) != 0)
363				continue;
364			/* Update nlist with new, translated results. */
365			p->n_type = np->n_type;
366			p->n_other = np->n_other;
367			p->n_desc = np->n_desc;
368			if (validate_fn)
369				p->n_value = (*validate_fn)(kd, np->n_value);
370			else
371				p->n_value = np->n_value;
372			missing--;
373			/* Find next freshly resolved entry. */
374			for (np++; np->n_name && np->n_name[0]; np++)
375				if (np->n_type != N_UNDF)
376					break;
377		}
378	}
379	/* We could assert missing = unresolved here. */
380
381	free(n);
382	return (unresolved);
383}
384
385int
386_kvm_nlist(kvm_t *kd, struct nlist *nl, int initialize)
387{
388	struct nlist *p;
389	int nvalid;
390	struct kld_sym_lookup lookup;
391	int error;
392	const char *prefix = "";
393	char symname[1024]; /* XXX-BZ symbol name length limit? */
394	int tried_vnet, tried_dpcpu;
395
396	/*
397	 * If we can't use the kld symbol lookup, revert to the
398	 * slow library call.
399	 */
400	if (!ISALIVE(kd)) {
401		error = __fdnlist(kd->nlfd, nl);
402		if (error <= 0)			/* Hard error or success. */
403			return (error);
404
405		if (_kvm_vnet_initialized(kd, initialize))
406			error = kvm_fdnlist_prefix(kd, nl, error,
407			    VNET_SYMPREFIX, _kvm_vnet_validaddr);
408
409		if (error > 0 && _kvm_dpcpu_initialized(kd, initialize))
410			error = kvm_fdnlist_prefix(kd, nl, error,
411			    DPCPU_SYMPREFIX, _kvm_dpcpu_validaddr);
412
413		return (error);
414	}
415
416	/*
417	 * We can use the kld lookup syscall.  Go through each nlist entry
418	 * and look it up with a kldsym(2) syscall.
419	 */
420	nvalid = 0;
421	tried_vnet = 0;
422	tried_dpcpu = 0;
423again:
424	for (p = nl; p->n_name && p->n_name[0]; ++p) {
425		if (p->n_type != N_UNDF)
426			continue;
427
428		lookup.version = sizeof(lookup);
429		lookup.symvalue = 0;
430		lookup.symsize = 0;
431
432		error = snprintf(symname, sizeof(symname), "%s%s", prefix,
433		    (prefix[0] != '\0' && p->n_name[0] == '_') ?
434			(p->n_name + 1) : p->n_name);
435		if (error < 0 || error >= (int)sizeof(symname))
436			continue;
437		lookup.symname = symname;
438		if (lookup.symname[0] == '_')
439			lookup.symname++;
440
441		if (kldsym(0, KLDSYM_LOOKUP, &lookup) != -1) {
442			p->n_type = N_TEXT;
443			p->n_other = 0;
444			p->n_desc = 0;
445			if (_kvm_vnet_initialized(kd, initialize) &&
446			    strcmp(prefix, VNET_SYMPREFIX) == 0)
447				p->n_value =
448				    _kvm_vnet_validaddr(kd, lookup.symvalue);
449			else if (_kvm_dpcpu_initialized(kd, initialize) &&
450			    strcmp(prefix, DPCPU_SYMPREFIX) == 0)
451				p->n_value =
452				    _kvm_dpcpu_validaddr(kd, lookup.symvalue);
453			else
454				p->n_value = lookup.symvalue;
455			++nvalid;
456			/* lookup.symsize */
457		}
458	}
459
460	/*
461	 * Check the number of entries that weren't found. If they exist,
462	 * try again with a prefix for virtualized or DPCPU symbol names.
463	 */
464	error = ((p - nl) - nvalid);
465	if (error && _kvm_vnet_initialized(kd, initialize) && !tried_vnet) {
466		tried_vnet = 1;
467		prefix = VNET_SYMPREFIX;
468		goto again;
469	}
470	if (error && _kvm_dpcpu_initialized(kd, initialize) && !tried_dpcpu) {
471		tried_dpcpu = 1;
472		prefix = DPCPU_SYMPREFIX;
473		goto again;
474	}
475
476	/*
477	 * Return the number of entries that weren't found. If they exist,
478	 * also fill internal error buffer.
479	 */
480	error = ((p - nl) - nvalid);
481	if (error)
482		_kvm_syserr(kd, kd->program, "kvm_nlist");
483	return (error);
484}
485
486int
487kvm_nlist(kvm_t *kd, struct nlist *nl)
488{
489
490	/*
491	 * If called via the public interface, permit intialization of
492	 * further virtualized modules on demand.
493	 */
494	return (_kvm_nlist(kd, nl, 1));
495}
496
497ssize_t
498kvm_read(kvm_t *kd, u_long kva, void *buf, size_t len)
499{
500	int cc;
501	ssize_t cr;
502	off_t pa;
503	char *cp;
504
505	if (ISALIVE(kd)) {
506		/*
507		 * We're using /dev/kmem.  Just read straight from the
508		 * device and let the active kernel do the address translation.
509		 */
510		errno = 0;
511		if (lseek(kd->vmfd, (off_t)kva, 0) == -1 && errno != 0) {
512			_kvm_err(kd, 0, "invalid address (%lx)", kva);
513			return (-1);
514		}
515		cr = read(kd->vmfd, buf, len);
516		if (cr < 0) {
517			_kvm_syserr(kd, 0, "kvm_read");
518			return (-1);
519		} else if (cr < (ssize_t)len)
520			_kvm_err(kd, kd->program, "short read");
521		return (cr);
522	}
523
524	cp = buf;
525	while (len > 0) {
526		cc = _kvm_kvatop(kd, kva, &pa);
527		if (cc == 0)
528			return (-1);
529		if (cc > (ssize_t)len)
530			cc = len;
531		errno = 0;
532		if (lseek(kd->pmfd, pa, 0) == -1 && errno != 0) {
533			_kvm_syserr(kd, 0, _PATH_MEM);
534			break;
535		}
536		cr = read(kd->pmfd, cp, cc);
537		if (cr < 0) {
538			_kvm_syserr(kd, kd->program, "kvm_read");
539			break;
540		}
541		/*
542		 * If kvm_kvatop returns a bogus value or our core file is
543		 * truncated, we might wind up seeking beyond the end of the
544		 * core file in which case the read will return 0 (EOF).
545		 */
546		if (cr == 0)
547			break;
548		cp += cr;
549		kva += cr;
550		len -= cr;
551	}
552
553	return (cp - (char *)buf);
554}
555
556ssize_t
557kvm_write(kvm_t *kd, u_long kva, const void *buf, size_t len)
558{
559	int cc;
560
561	if (ISALIVE(kd)) {
562		/*
563		 * Just like kvm_read, only we write.
564		 */
565		errno = 0;
566		if (lseek(kd->vmfd, (off_t)kva, 0) == -1 && errno != 0) {
567			_kvm_err(kd, 0, "invalid address (%lx)", kva);
568			return (-1);
569		}
570		cc = write(kd->vmfd, buf, len);
571		if (cc < 0) {
572			_kvm_syserr(kd, 0, "kvm_write");
573			return (-1);
574		} else if ((size_t)cc < len)
575			_kvm_err(kd, kd->program, "short write");
576		return (cc);
577	} else {
578		_kvm_err(kd, kd->program,
579		    "kvm_write not implemented for dead kernels");
580		return (-1);
581	}
582	/* NOTREACHED */
583}
584