xref: /freebsd/lib/libkvm/kvm.c (revision efe3b0de1438e7a8473d92f2be57072394559e3c)
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
39 static 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 #include <sys/fnv_hash.h>
45 
46 #define	_WANT_VNET
47 
48 #include <sys/user.h>
49 #include <sys/linker.h>
50 #include <sys/pcpu.h>
51 #include <sys/stat.h>
52 
53 #include <net/vnet.h>
54 
55 #include <fcntl.h>
56 #include <kvm.h>
57 #include <limits.h>
58 #include <paths.h>
59 #include <stdint.h>
60 #include <stdio.h>
61 #include <stdlib.h>
62 #include <string.h>
63 #include <unistd.h>
64 
65 #include "kvm_private.h"
66 
67 SET_DECLARE(kvm_arch, struct kvm_arch);
68 
69 char *
70 kvm_geterr(kvm_t *kd)
71 {
72 	return (kd->errbuf);
73 }
74 
75 static int
76 _kvm_read_kernel_ehdr(kvm_t *kd)
77 {
78 	Elf *elf;
79 
80 	if (elf_version(EV_CURRENT) == EV_NONE) {
81 		_kvm_err(kd, kd->program, "Unsupported libelf");
82 		return (-1);
83 	}
84 	elf = elf_begin(kd->nlfd, ELF_C_READ, NULL);
85 	if (elf == NULL) {
86 		_kvm_err(kd, kd->program, "%s", elf_errmsg(0));
87 		return (-1);
88 	}
89 	if (elf_kind(elf) != ELF_K_ELF) {
90 		_kvm_err(kd, kd->program, "kernel is not an ELF file");
91 		return (-1);
92 	}
93 	if (gelf_getehdr(elf, &kd->nlehdr) == NULL) {
94 		_kvm_err(kd, kd->program, "%s", elf_errmsg(0));
95 		elf_end(elf);
96 		return (-1);
97 	}
98 	elf_end(elf);
99 
100 	switch (kd->nlehdr.e_ident[EI_DATA]) {
101 	case ELFDATA2LSB:
102 	case ELFDATA2MSB:
103 		return (0);
104 	default:
105 		_kvm_err(kd, kd->program,
106 		    "unsupported ELF data encoding for kernel");
107 		return (-1);
108 	}
109 }
110 
111 static kvm_t *
112 _kvm_open(kvm_t *kd, const char *uf, const char *mf, int flag, char *errout)
113 {
114 	struct kvm_arch **parch;
115 	struct stat st;
116 
117 	kd->vmfd = -1;
118 	kd->pmfd = -1;
119 	kd->nlfd = -1;
120 	kd->vmst = NULL;
121 	kd->procbase = NULL;
122 	kd->argspc = NULL;
123 	kd->argv = NULL;
124 
125 	if (uf == NULL)
126 		uf = getbootfile();
127 	else if (strlen(uf) >= MAXPATHLEN) {
128 		_kvm_err(kd, kd->program, "exec file name too long");
129 		goto failed;
130 	}
131 	if (flag & ~O_RDWR) {
132 		_kvm_err(kd, kd->program, "bad flags arg");
133 		goto failed;
134 	}
135 	if (mf == NULL)
136 		mf = _PATH_MEM;
137 
138 	if ((kd->pmfd = open(mf, flag | O_CLOEXEC, 0)) < 0) {
139 		_kvm_syserr(kd, kd->program, "%s", mf);
140 		goto failed;
141 	}
142 	if (fstat(kd->pmfd, &st) < 0) {
143 		_kvm_syserr(kd, kd->program, "%s", mf);
144 		goto failed;
145 	}
146 	if (S_ISREG(st.st_mode) && st.st_size <= 0) {
147 		errno = EINVAL;
148 		_kvm_syserr(kd, kd->program, "empty file");
149 		goto failed;
150 	}
151 	if (S_ISCHR(st.st_mode)) {
152 		/*
153 		 * If this is a character special device, then check that
154 		 * it's /dev/mem.  If so, open kmem too.  (Maybe we should
155 		 * make it work for either /dev/mem or /dev/kmem -- in either
156 		 * case you're working with a live kernel.)
157 		 */
158 		if (strcmp(mf, _PATH_DEVNULL) == 0) {
159 			kd->vmfd = open(_PATH_DEVNULL, O_RDONLY | O_CLOEXEC);
160 			return (kd);
161 		} else if (strcmp(mf, _PATH_MEM) == 0) {
162 			if ((kd->vmfd = open(_PATH_KMEM, flag | O_CLOEXEC)) <
163 			    0) {
164 				_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
165 				goto failed;
166 			}
167 			return (kd);
168 		}
169 	}
170 
171 	/*
172 	 * This is either a crash dump or a remote live system with its physical
173 	 * memory fully accessible via a special device.
174 	 * Open the namelist fd and determine the architecture.
175 	 */
176 	if ((kd->nlfd = open(uf, O_RDONLY | O_CLOEXEC, 0)) < 0) {
177 		_kvm_syserr(kd, kd->program, "%s", uf);
178 		goto failed;
179 	}
180 	if (_kvm_read_kernel_ehdr(kd) < 0)
181 		goto failed;
182 	if (strncmp(mf, _PATH_FWMEM, strlen(_PATH_FWMEM)) == 0 ||
183 	    strncmp(mf, _PATH_DEVVMM, strlen(_PATH_DEVVMM)) == 0) {
184 		kd->rawdump = 1;
185 		kd->writable = 1;
186 	}
187 	SET_FOREACH(parch, kvm_arch) {
188 		if ((*parch)->ka_probe(kd)) {
189 			kd->arch = *parch;
190 			break;
191 		}
192 	}
193 	if (kd->arch == NULL) {
194 		_kvm_err(kd, kd->program, "unsupported architecture");
195 		goto failed;
196 	}
197 
198 	/*
199 	 * Non-native kernels require a symbol resolver.
200 	 */
201 	if (!kd->arch->ka_native(kd) && kd->resolve_symbol == NULL) {
202 		_kvm_err(kd, kd->program,
203 		    "non-native kernel requires a symbol resolver");
204 		goto failed;
205 	}
206 
207 	/*
208 	 * Initialize the virtual address translation machinery.
209 	 */
210 	if (kd->arch->ka_initvtop(kd) < 0)
211 		goto failed;
212 	return (kd);
213 failed:
214 	/*
215 	 * Copy out the error if doing sane error semantics.
216 	 */
217 	if (errout != NULL)
218 		strlcpy(errout, kd->errbuf, _POSIX2_LINE_MAX);
219 	(void)kvm_close(kd);
220 	return (0);
221 }
222 
223 kvm_t *
224 kvm_openfiles(const char *uf, const char *mf, const char *sf __unused, int flag,
225     char *errout)
226 {
227 	kvm_t *kd;
228 
229 	if ((kd = calloc(1, sizeof(*kd))) == NULL) {
230 		if (errout != NULL)
231 			(void)strlcpy(errout, strerror(errno),
232 			    _POSIX2_LINE_MAX);
233 		return (0);
234 	}
235 	return (_kvm_open(kd, uf, mf, flag, errout));
236 }
237 
238 kvm_t *
239 kvm_open(const char *uf, const char *mf, const char *sf __unused, int flag,
240     const char *errstr)
241 {
242 	kvm_t *kd;
243 
244 	if ((kd = calloc(1, sizeof(*kd))) == NULL) {
245 		if (errstr != NULL)
246 			(void)fprintf(stderr, "%s: %s\n",
247 				      errstr, strerror(errno));
248 		return (0);
249 	}
250 	kd->program = errstr;
251 	return (_kvm_open(kd, uf, mf, flag, NULL));
252 }
253 
254 kvm_t *
255 kvm_open2(const char *uf, const char *mf, int flag, char *errout,
256     int (*resolver)(const char *, kvaddr_t *))
257 {
258 	kvm_t *kd;
259 
260 	if ((kd = calloc(1, sizeof(*kd))) == NULL) {
261 		if (errout != NULL)
262 			(void)strlcpy(errout, strerror(errno),
263 			    _POSIX2_LINE_MAX);
264 		return (0);
265 	}
266 	kd->resolve_symbol = resolver;
267 	return (_kvm_open(kd, uf, mf, flag, errout));
268 }
269 
270 int
271 kvm_close(kvm_t *kd)
272 {
273 	int error = 0;
274 
275 	if (kd->vmst != NULL)
276 		kd->arch->ka_freevtop(kd);
277 	if (kd->pmfd >= 0)
278 		error |= close(kd->pmfd);
279 	if (kd->vmfd >= 0)
280 		error |= close(kd->vmfd);
281 	if (kd->nlfd >= 0)
282 		error |= close(kd->nlfd);
283 	if (kd->procbase != 0)
284 		free((void *)kd->procbase);
285 	if (kd->argbuf != 0)
286 		free((void *) kd->argbuf);
287 	if (kd->argspc != 0)
288 		free((void *) kd->argspc);
289 	if (kd->argv != 0)
290 		free((void *)kd->argv);
291 	if (kd->pt_map != NULL)
292 		free(kd->pt_map);
293 	free((void *)kd);
294 
295 	return (0);
296 }
297 
298 int
299 kvm_nlist2(kvm_t *kd, struct kvm_nlist *nl)
300 {
301 
302 	/*
303 	 * If called via the public interface, permit initialization of
304 	 * further virtualized modules on demand.
305 	 */
306 	return (_kvm_nlist(kd, nl, 1));
307 }
308 
309 int
310 kvm_nlist(kvm_t *kd, struct nlist *nl)
311 {
312 	struct kvm_nlist *kl;
313 	int count, i, nfail;
314 
315 	/*
316 	 * Avoid reporting truncated addresses by failing for non-native
317 	 * cores.
318 	 */
319 	if (!kvm_native(kd)) {
320 		_kvm_err(kd, kd->program, "kvm_nlist of non-native vmcore");
321 		return (-1);
322 	}
323 
324 	for (count = 0; nl[count].n_name != NULL && nl[count].n_name[0] != '\0';
325 	     count++)
326 		;
327 	if (count == 0)
328 		return (0);
329 	kl = calloc(count + 1, sizeof(*kl));
330 	for (i = 0; i < count; i++)
331 		kl[i].n_name = nl[i].n_name;
332 	nfail = kvm_nlist2(kd, kl);
333 	for (i = 0; i < count; i++) {
334 		nl[i].n_type = kl[i].n_type;
335 		nl[i].n_other = 0;
336 		nl[i].n_desc = 0;
337 		nl[i].n_value = kl[i].n_value;
338 	}
339 	return (nfail);
340 }
341 
342 ssize_t
343 kvm_read(kvm_t *kd, u_long kva, void *buf, size_t len)
344 {
345 
346 	return (kvm_read2(kd, kva, buf, len));
347 }
348 
349 ssize_t
350 kvm_read2(kvm_t *kd, kvaddr_t kva, void *buf, size_t len)
351 {
352 	int cc;
353 	ssize_t cr;
354 	off_t pa;
355 	char *cp;
356 
357 	if (ISALIVE(kd)) {
358 		/*
359 		 * We're using /dev/kmem.  Just read straight from the
360 		 * device and let the active kernel do the address translation.
361 		 */
362 		errno = 0;
363 		if (lseek(kd->vmfd, (off_t)kva, 0) == -1 && errno != 0) {
364 			_kvm_err(kd, 0, "invalid address (0x%jx)",
365 			    (uintmax_t)kva);
366 			return (-1);
367 		}
368 		cr = read(kd->vmfd, buf, len);
369 		if (cr < 0) {
370 			_kvm_syserr(kd, 0, "kvm_read");
371 			return (-1);
372 		} else if (cr < (ssize_t)len)
373 			_kvm_err(kd, kd->program, "short read");
374 		return (cr);
375 	}
376 
377 	cp = buf;
378 	while (len > 0) {
379 		cc = kd->arch->ka_kvatop(kd, kva, &pa);
380 		if (cc == 0)
381 			return (-1);
382 		if (cc > (ssize_t)len)
383 			cc = len;
384 		errno = 0;
385 		if (lseek(kd->pmfd, pa, 0) == -1 && errno != 0) {
386 			_kvm_syserr(kd, 0, _PATH_MEM);
387 			break;
388 		}
389 		cr = read(kd->pmfd, cp, cc);
390 		if (cr < 0) {
391 			_kvm_syserr(kd, kd->program, "kvm_read");
392 			break;
393 		}
394 		/*
395 		 * If ka_kvatop returns a bogus value or our core file is
396 		 * truncated, we might wind up seeking beyond the end of the
397 		 * core file in which case the read will return 0 (EOF).
398 		 */
399 		if (cr == 0)
400 			break;
401 		cp += cr;
402 		kva += cr;
403 		len -= cr;
404 	}
405 
406 	return (cp - (char *)buf);
407 }
408 
409 ssize_t
410 kvm_write(kvm_t *kd, u_long kva, const void *buf, size_t len)
411 {
412 	int cc;
413 	ssize_t cw;
414 	off_t pa;
415 	const char *cp;
416 
417 	if (!ISALIVE(kd) && !kd->writable) {
418 		_kvm_err(kd, kd->program,
419 		    "kvm_write not implemented for dead kernels");
420 		return (-1);
421 	}
422 
423 	if (ISALIVE(kd)) {
424 		/*
425 		 * Just like kvm_read, only we write.
426 		 */
427 		errno = 0;
428 		if (lseek(kd->vmfd, (off_t)kva, 0) == -1 && errno != 0) {
429 			_kvm_err(kd, 0, "invalid address (%lx)", kva);
430 			return (-1);
431 		}
432 		cc = write(kd->vmfd, buf, len);
433 		if (cc < 0) {
434 			_kvm_syserr(kd, 0, "kvm_write");
435 			return (-1);
436 		} else if ((size_t)cc < len)
437 			_kvm_err(kd, kd->program, "short write");
438 		return (cc);
439 	}
440 
441 	cp = buf;
442 	while (len > 0) {
443 		cc = kd->arch->ka_kvatop(kd, kva, &pa);
444 		if (cc == 0)
445 			return (-1);
446 		if (cc > (ssize_t)len)
447 			cc = len;
448 		errno = 0;
449 		if (lseek(kd->pmfd, pa, 0) == -1 && errno != 0) {
450 			_kvm_syserr(kd, 0, _PATH_MEM);
451 			break;
452 		}
453 		cw = write(kd->pmfd, cp, cc);
454 		if (cw < 0) {
455 			_kvm_syserr(kd, kd->program, "kvm_write");
456 			break;
457 		}
458 		/*
459 		 * If ka_kvatop returns a bogus value or our core file is
460 		 * truncated, we might wind up seeking beyond the end of the
461 		 * core file in which case the read will return 0 (EOF).
462 		 */
463 		if (cw == 0)
464 			break;
465 		cp += cw;
466 		kva += cw;
467 		len -= cw;
468 	}
469 
470 	return (cp - (char *)buf);
471 }
472 
473 int
474 kvm_native(kvm_t *kd)
475 {
476 
477 	if (ISALIVE(kd))
478 		return (1);
479 	return (kd->arch->ka_native(kd));
480 }
481