xref: /freebsd/lib/libkvm/kvm.c (revision 61e21613) 
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
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 
37 #include <sys/param.h>
38 #include <sys/fnv_hash.h>
39 
40 #define	_WANT_VNET
41 
42 #include <sys/user.h>
43 #include <sys/linker.h>
44 #include <sys/pcpu.h>
45 #include <sys/stat.h>
46 #include <sys/sysctl.h>
47 #include <sys/mman.h>
48 
49 #include <stdbool.h>
50 #include <net/vnet.h>
51 
52 #include <fcntl.h>
53 #include <kvm.h>
54 #include <limits.h>
55 #include <paths.h>
56 #include <stdint.h>
57 #include <stdio.h>
58 #include <stdlib.h>
59 #include <string.h>
60 #include <unistd.h>
61 
62 #include "kvm_private.h"
63 
64 SET_DECLARE(kvm_arch, struct kvm_arch);
65 
66 static char _kd_is_null[] = "";
67 
68 char *
69 kvm_geterr(kvm_t *kd)
70 {
71 
72 	if (kd == NULL)
73 		return (_kd_is_null);
74 	return (kd->errbuf);
75 }
76 
77 static int
78 _kvm_read_kernel_ehdr(kvm_t *kd)
79 {
80 	Elf *elf;
81 
82 	if (elf_version(EV_CURRENT) == EV_NONE) {
83 		_kvm_err(kd, kd->program, "Unsupported libelf");
84 		return (-1);
85 	}
86 	elf = elf_begin(kd->nlfd, ELF_C_READ, NULL);
87 	if (elf == NULL) {
88 		_kvm_err(kd, kd->program, "%s", elf_errmsg(0));
89 		return (-1);
90 	}
91 	if (elf_kind(elf) != ELF_K_ELF) {
92 		_kvm_err(kd, kd->program, "kernel is not an ELF file");
93 		return (-1);
94 	}
95 	if (gelf_getehdr(elf, &kd->nlehdr) == NULL) {
96 		_kvm_err(kd, kd->program, "%s", elf_errmsg(0));
97 		elf_end(elf);
98 		return (-1);
99 	}
100 	elf_end(elf);
101 
102 	switch (kd->nlehdr.e_ident[EI_DATA]) {
103 	case ELFDATA2LSB:
104 	case ELFDATA2MSB:
105 		return (0);
106 	default:
107 		_kvm_err(kd, kd->program,
108 		    "unsupported ELF data encoding for kernel");
109 		return (-1);
110 	}
111 }
112 
113 static kvm_t *
114 _kvm_open(kvm_t *kd, const char *uf, const char *mf, int flag, char *errout)
115 {
116 	struct kvm_arch **parch;
117 	struct stat st;
118 
119 	kd->vmfd = -1;
120 	kd->pmfd = -1;
121 	kd->nlfd = -1;
122 	kd->vmst = NULL;
123 	kd->procbase = NULL;
124 	kd->argspc = NULL;
125 	kd->argv = NULL;
126 
127 	if (uf == NULL)
128 		uf = getbootfile();
129 	else if (strlen(uf) >= MAXPATHLEN) {
130 		_kvm_err(kd, kd->program, "exec file name too long");
131 		goto failed;
132 	}
133 	if (flag & ~O_RDWR) {
134 		_kvm_err(kd, kd->program, "bad flags arg");
135 		goto failed;
136 	}
137 	if (mf == NULL)
138 		mf = _PATH_MEM;
139 
140 	if ((kd->pmfd = open(mf, flag | O_CLOEXEC, 0)) < 0) {
141 		_kvm_syserr(kd, kd->program, "%s", mf);
142 		goto failed;
143 	}
144 	if (fstat(kd->pmfd, &st) < 0) {
145 		_kvm_syserr(kd, kd->program, "%s", mf);
146 		goto failed;
147 	}
148 	if (S_ISREG(st.st_mode) && st.st_size <= 0) {
149 		errno = EINVAL;
150 		_kvm_syserr(kd, kd->program, "empty file");
151 		goto failed;
152 	}
153 	if (S_ISCHR(st.st_mode)) {
154 		/*
155 		 * If this is a character special device, then check that
156 		 * it's /dev/mem.  If so, open kmem too.  (Maybe we should
157 		 * make it work for either /dev/mem or /dev/kmem -- in either
158 		 * case you're working with a live kernel.)
159 		 */
160 		if (strcmp(mf, _PATH_DEVNULL) == 0) {
161 			kd->vmfd = open(_PATH_DEVNULL, O_RDONLY | O_CLOEXEC);
162 			return (kd);
163 		} else if (strcmp(mf, _PATH_MEM) == 0) {
164 			if ((kd->vmfd = open(_PATH_KMEM, flag | O_CLOEXEC)) <
165 			    0) {
166 				_kvm_syserr(kd, kd->program, "%s", _PATH_KMEM);
167 				goto failed;
168 			}
169 			return (kd);
170 		}
171 	}
172 
173 	/*
174 	 * This is either a crash dump or a remote live system with its physical
175 	 * memory fully accessible via a special device.
176 	 * Open the namelist fd and determine the architecture.
177 	 */
178 	if ((kd->nlfd = open(uf, O_RDONLY | O_CLOEXEC, 0)) < 0) {
179 		_kvm_syserr(kd, kd->program, "%s", uf);
180 		goto failed;
181 	}
182 	if (_kvm_read_kernel_ehdr(kd) < 0)
183 		goto failed;
184 	if (strncmp(mf, _PATH_FWMEM, strlen(_PATH_FWMEM)) == 0 ||
185 	    strncmp(mf, _PATH_DEVVMM, strlen(_PATH_DEVVMM)) == 0) {
186 		kd->rawdump = 1;
187 		kd->writable = 1;
188 	}
189 	SET_FOREACH(parch, kvm_arch) {
190 		if ((*parch)->ka_probe(kd)) {
191 			kd->arch = *parch;
192 			break;
193 		}
194 	}
195 	if (kd->arch == NULL) {
196 		_kvm_err(kd, kd->program, "unsupported architecture");
197 		goto failed;
198 	}
199 
200 	/*
201 	 * Non-native kernels require a symbol resolver.
202 	 */
203 	if (!kd->arch->ka_native(kd) && kd->resolve_symbol == NULL) {
204 		_kvm_err(kd, kd->program,
205 		    "non-native kernel requires a symbol resolver");
206 		goto failed;
207 	}
208 
209 	/*
210 	 * Initialize the virtual address translation machinery.
211 	 */
212 	if (kd->arch->ka_initvtop(kd) < 0)
213 		goto failed;
214 	return (kd);
215 failed:
216 	/*
217 	 * Copy out the error if doing sane error semantics.
218 	 */
219 	if (errout != NULL)
220 		strlcpy(errout, kd->errbuf, _POSIX2_LINE_MAX);
221 	(void)kvm_close(kd);
222 	return (NULL);
223 }
224 
225 kvm_t *
226 kvm_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 		if (errout != NULL)
233 			(void)strlcpy(errout, strerror(errno),
234 			    _POSIX2_LINE_MAX);
235 		return (NULL);
236 	}
237 	return (_kvm_open(kd, uf, mf, flag, errout));
238 }
239 
240 kvm_t *
241 kvm_open(const char *uf, const char *mf, const char *sf __unused, int flag,
242     const char *errstr)
243 {
244 	kvm_t *kd;
245 
246 	if ((kd = calloc(1, sizeof(*kd))) == NULL) {
247 		if (errstr != NULL)
248 			(void)fprintf(stderr, "%s: %s\n",
249 				      errstr, strerror(errno));
250 		return (NULL);
251 	}
252 	kd->program = errstr;
253 	return (_kvm_open(kd, uf, mf, flag, NULL));
254 }
255 
256 kvm_t *
257 kvm_open2(const char *uf, const char *mf, int flag, char *errout,
258     int (*resolver)(const char *, kvaddr_t *))
259 {
260 	kvm_t *kd;
261 
262 	if ((kd = calloc(1, sizeof(*kd))) == NULL) {
263 		if (errout != NULL)
264 			(void)strlcpy(errout, strerror(errno),
265 			    _POSIX2_LINE_MAX);
266 		return (NULL);
267 	}
268 	kd->resolve_symbol = resolver;
269 	return (_kvm_open(kd, uf, mf, flag, errout));
270 }
271 
272 int
273 kvm_close(kvm_t *kd)
274 {
275 	int error = 0;
276 
277 	if (kd == NULL) {
278 		errno = EINVAL;
279 		return (-1);
280 	}
281 	if (kd->vmst != NULL)
282 		kd->arch->ka_freevtop(kd);
283 	if (kd->pmfd >= 0)
284 		error |= close(kd->pmfd);
285 	if (kd->vmfd >= 0)
286 		error |= close(kd->vmfd);
287 	if (kd->nlfd >= 0)
288 		error |= close(kd->nlfd);
289 	if (kd->procbase != 0)
290 		free((void *)kd->procbase);
291 	if (kd->argbuf != 0)
292 		free((void *) kd->argbuf);
293 	if (kd->argspc != 0)
294 		free((void *) kd->argspc);
295 	if (kd->argv != 0)
296 		free((void *)kd->argv);
297 	if (kd->dpcpu_initialized != 0)
298 		free(kd->dpcpu_off);
299 	if (kd->pt_map != NULL)
300 		free(kd->pt_map);
301 	if (kd->page_map != NULL)
302 		free(kd->page_map);
303 	if (kd->sparse_map != MAP_FAILED && kd->sparse_map != NULL)
304 		munmap(kd->sparse_map, kd->pt_sparse_size);
305 	free((void *)kd);
306 
307 	return (error);
308 }
309 
310 int
311 kvm_nlist2(kvm_t *kd, struct kvm_nlist *nl)
312 {
313 
314 	/*
315 	 * If called via the public interface, permit initialization of
316 	 * further virtualized modules on demand.
317 	 */
318 	return (_kvm_nlist(kd, nl, 1));
319 }
320 
321 int
322 kvm_nlist(kvm_t *kd, struct nlist *nl)
323 {
324 	struct kvm_nlist *kl;
325 	int count, i, nfail;
326 
327 	/*
328 	 * Avoid reporting truncated addresses by failing for non-native
329 	 * cores.
330 	 */
331 	if (!kvm_native(kd)) {
332 		_kvm_err(kd, kd->program, "kvm_nlist of non-native vmcore");
333 		return (-1);
334 	}
335 
336 	for (count = 0; nl[count].n_name != NULL && nl[count].n_name[0] != '\0';
337 	     count++)
338 		;
339 	if (count == 0)
340 		return (0);
341 	kl = calloc(count + 1, sizeof(*kl));
342 	if (kl == NULL) {
343 		_kvm_err(kd, kd->program, "cannot allocate memory");
344 		return (-1);
345 	}
346 	for (i = 0; i < count; i++)
347 		kl[i].n_name = nl[i].n_name;
348 	nfail = kvm_nlist2(kd, kl);
349 	for (i = 0; i < count; i++) {
350 		nl[i].n_type = kl[i].n_type;
351 		nl[i].n_other = 0;
352 		nl[i].n_desc = 0;
353 		nl[i].n_value = kl[i].n_value;
354 	}
355 	free(kl);
356 	return (nfail);
357 }
358 
359 ssize_t
360 kvm_read(kvm_t *kd, u_long kva, void *buf, size_t len)
361 {
362 
363 	return (kvm_read2(kd, kva, buf, len));
364 }
365 
366 ssize_t
367 kvm_read2(kvm_t *kd, kvaddr_t kva, void *buf, size_t len)
368 {
369 	int cc;
370 	ssize_t cr;
371 	off_t pa;
372 	char *cp;
373 
374 	if (ISALIVE(kd)) {
375 		/*
376 		 * We're using /dev/kmem.  Just read straight from the
377 		 * device and let the active kernel do the address translation.
378 		 */
379 		errno = 0;
380 		if (lseek(kd->vmfd, (off_t)kva, 0) == -1 && errno != 0) {
381 			_kvm_err(kd, 0, "invalid address (0x%jx)",
382 			    (uintmax_t)kva);
383 			return (-1);
384 		}
385 		cr = read(kd->vmfd, buf, len);
386 		if (cr < 0) {
387 			_kvm_syserr(kd, 0, "kvm_read");
388 			return (-1);
389 		} else if (cr < (ssize_t)len)
390 			_kvm_err(kd, kd->program, "short read");
391 		return (cr);
392 	}
393 
394 	cp = buf;
395 	while (len > 0) {
396 		cc = kd->arch->ka_kvatop(kd, kva, &pa);
397 		if (cc == 0)
398 			return (-1);
399 		if (cc > (ssize_t)len)
400 			cc = len;
401 		errno = 0;
402 		if (lseek(kd->pmfd, pa, 0) == -1 && errno != 0) {
403 			_kvm_syserr(kd, 0, _PATH_MEM);
404 			break;
405 		}
406 		cr = read(kd->pmfd, cp, cc);
407 		if (cr < 0) {
408 			_kvm_syserr(kd, kd->program, "kvm_read");
409 			break;
410 		}
411 		/*
412 		 * If ka_kvatop returns a bogus value or our core file is
413 		 * truncated, we might wind up seeking beyond the end of the
414 		 * core file in which case the read will return 0 (EOF).
415 		 */
416 		if (cr == 0)
417 			break;
418 		cp += cr;
419 		kva += cr;
420 		len -= cr;
421 	}
422 
423 	return (cp - (char *)buf);
424 }
425 
426 ssize_t
427 kvm_write(kvm_t *kd, u_long kva, const void *buf, size_t len)
428 {
429 	int cc;
430 	ssize_t cw;
431 	off_t pa;
432 	const char *cp;
433 
434 	if (!ISALIVE(kd) && !kd->writable) {
435 		_kvm_err(kd, kd->program,
436 		    "kvm_write not implemented for dead kernels");
437 		return (-1);
438 	}
439 
440 	if (ISALIVE(kd)) {
441 		/*
442 		 * Just like kvm_read, only we write.
443 		 */
444 		errno = 0;
445 		if (lseek(kd->vmfd, (off_t)kva, 0) == -1 && errno != 0) {
446 			_kvm_err(kd, 0, "invalid address (%lx)", kva);
447 			return (-1);
448 		}
449 		cc = write(kd->vmfd, buf, len);
450 		if (cc < 0) {
451 			_kvm_syserr(kd, 0, "kvm_write");
452 			return (-1);
453 		} else if ((size_t)cc < len)
454 			_kvm_err(kd, kd->program, "short write");
455 		return (cc);
456 	}
457 
458 	cp = buf;
459 	while (len > 0) {
460 		cc = kd->arch->ka_kvatop(kd, kva, &pa);
461 		if (cc == 0)
462 			return (-1);
463 		if (cc > (ssize_t)len)
464 			cc = len;
465 		errno = 0;
466 		if (lseek(kd->pmfd, pa, 0) == -1 && errno != 0) {
467 			_kvm_syserr(kd, 0, _PATH_MEM);
468 			break;
469 		}
470 		cw = write(kd->pmfd, cp, cc);
471 		if (cw < 0) {
472 			_kvm_syserr(kd, kd->program, "kvm_write");
473 			break;
474 		}
475 		/*
476 		 * If ka_kvatop returns a bogus value or our core file is
477 		 * truncated, we might wind up seeking beyond the end of the
478 		 * core file in which case the read will return 0 (EOF).
479 		 */
480 		if (cw == 0)
481 			break;
482 		cp += cw;
483 		kva += cw;
484 		len -= cw;
485 	}
486 
487 	return (cp - (const char *)buf);
488 }
489 
490 int
491 kvm_native(kvm_t *kd)
492 {
493 
494 	if (ISALIVE(kd))
495 		return (1);
496 	return (kd->arch->ka_native(kd));
497 }
498 
499 int
500 kvm_walk_pages(kvm_t *kd, kvm_walk_pages_cb_t *cb, void *closure)
501 {
502 
503 	if (kd->arch->ka_walk_pages == NULL)
504 		return (0);
505 
506 	return (kd->arch->ka_walk_pages(kd, cb, closure));
507 }
508 
509 kssize_t
510 kvm_kerndisp(kvm_t *kd)
511 {
512 	unsigned long kernbase, rel_kernbase;
513 	size_t kernbase_len = sizeof(kernbase);
514 	size_t rel_kernbase_len = sizeof(rel_kernbase);
515 
516 	if (ISALIVE(kd)) {
517 		if (sysctlbyname("kern.base_address", &kernbase,
518 		    &kernbase_len, NULL, 0) == -1) {
519 			_kvm_syserr(kd, kd->program,
520 				"failed to get kernel base address");
521 			return (0);
522 		}
523 		if (sysctlbyname("kern.relbase_address", &rel_kernbase,
524 		    &rel_kernbase_len, NULL, 0) == -1) {
525 			_kvm_syserr(kd, kd->program,
526 				"failed to get relocated kernel base address");
527 			return (0);
528 		}
529 		return (rel_kernbase - kernbase);
530 	}
531 
532 	if (kd->arch->ka_kerndisp == NULL)
533 		return (0);
534 
535 	return (kd->arch->ka_kerndisp(kd));
536 }
537