xref: /freebsd/lib/libkvm/kvm_private.c (revision 879e0604ee6ac0ded87f6754141e7af25f4b7d5a)
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  * 3. 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 #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/mman.h>
47 
48 #include <net/vnet.h>
49 
50 #include <assert.h>
51 #include <fcntl.h>
52 #include <vm/vm.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 #include <stdarg.h>
62 #include <inttypes.h>
63 
64 #include "kvm_private.h"
65 
66 /*
67  * Routines private to libkvm.
68  */
69 
70 /* from src/lib/libc/gen/nlist.c */
71 int __fdnlist(int, struct nlist *);
72 
73 /*
74  * Report an error using printf style arguments.  "program" is kd->program
75  * on hard errors, and 0 on soft errors, so that under sun error emulation,
76  * only hard errors are printed out (otherwise, programs like gdb will
77  * generate tons of error messages when trying to access bogus pointers).
78  */
79 void
80 _kvm_err(kvm_t *kd, const char *program, const char *fmt, ...)
81 {
82 	va_list ap;
83 
84 	va_start(ap, fmt);
85 	if (program != NULL) {
86 		(void)fprintf(stderr, "%s: ", program);
87 		(void)vfprintf(stderr, fmt, ap);
88 		(void)fputc('\n', stderr);
89 	} else
90 		(void)vsnprintf(kd->errbuf,
91 		    sizeof(kd->errbuf), fmt, ap);
92 
93 	va_end(ap);
94 }
95 
96 void
97 _kvm_syserr(kvm_t *kd, const char *program, const char *fmt, ...)
98 {
99 	va_list ap;
100 	int n;
101 
102 	va_start(ap, fmt);
103 	if (program != NULL) {
104 		(void)fprintf(stderr, "%s: ", program);
105 		(void)vfprintf(stderr, fmt, ap);
106 		(void)fprintf(stderr, ": %s\n", strerror(errno));
107 	} else {
108 		char *cp = kd->errbuf;
109 
110 		(void)vsnprintf(cp, sizeof(kd->errbuf), fmt, ap);
111 		n = strlen(cp);
112 		(void)snprintf(&cp[n], sizeof(kd->errbuf) - n, ": %s",
113 		    strerror(errno));
114 	}
115 	va_end(ap);
116 }
117 
118 void *
119 _kvm_malloc(kvm_t *kd, size_t n)
120 {
121 	void *p;
122 
123 	if ((p = calloc(n, sizeof(char))) == NULL)
124 		_kvm_err(kd, kd->program, "can't allocate %zu bytes: %s",
125 			 n, strerror(errno));
126 	return (p);
127 }
128 
129 int
130 _kvm_probe_elf_kernel(kvm_t *kd, int class, int machine)
131 {
132 
133 	return (kd->nlehdr.e_ident[EI_CLASS] == class &&
134 	    ((machine == EM_PPC || machine == EM_PPC64) ?
135 	     kd->nlehdr.e_type == ET_DYN : kd->nlehdr.e_type == ET_EXEC) &&
136 	    kd->nlehdr.e_machine == machine);
137 }
138 
139 int
140 _kvm_is_minidump(kvm_t *kd)
141 {
142 	char minihdr[8];
143 
144 	if (kd->rawdump)
145 		return (0);
146 	if (pread(kd->pmfd, &minihdr, 8, 0) == 8 &&
147 	    memcmp(&minihdr, "minidump", 8) == 0)
148 		return (1);
149 	return (0);
150 }
151 
152 /*
153  * The powerpc backend has a hack to strip a leading kerneldump
154  * header from the core before treating it as an ELF header.
155  *
156  * We can add that here if we can get a change to libelf to support
157  * an initial offset into the file.  Alternatively we could patch
158  * savecore to extract cores from a regular file instead.
159  */
160 int
161 _kvm_read_core_phdrs(kvm_t *kd, size_t *phnump, GElf_Phdr **phdrp)
162 {
163 	GElf_Ehdr ehdr;
164 	GElf_Phdr *phdr;
165 	Elf *elf;
166 	size_t i, phnum;
167 
168 	elf = elf_begin(kd->pmfd, ELF_C_READ, NULL);
169 	if (elf == NULL) {
170 		_kvm_err(kd, kd->program, "%s", elf_errmsg(0));
171 		return (-1);
172 	}
173 	if (elf_kind(elf) != ELF_K_ELF) {
174 		_kvm_err(kd, kd->program, "invalid core");
175 		goto bad;
176 	}
177 	if (gelf_getclass(elf) != kd->nlehdr.e_ident[EI_CLASS]) {
178 		_kvm_err(kd, kd->program, "invalid core");
179 		goto bad;
180 	}
181 	if (gelf_getehdr(elf, &ehdr) == NULL) {
182 		_kvm_err(kd, kd->program, "%s", elf_errmsg(0));
183 		goto bad;
184 	}
185 	if (ehdr.e_type != ET_CORE) {
186 		_kvm_err(kd, kd->program, "invalid core");
187 		goto bad;
188 	}
189 	if (ehdr.e_machine != kd->nlehdr.e_machine) {
190 		_kvm_err(kd, kd->program, "invalid core");
191 		goto bad;
192 	}
193 
194 	if (elf_getphdrnum(elf, &phnum) == -1) {
195 		_kvm_err(kd, kd->program, "%s", elf_errmsg(0));
196 		goto bad;
197 	}
198 
199 	phdr = calloc(phnum, sizeof(*phdr));
200 	if (phdr == NULL) {
201 		_kvm_err(kd, kd->program, "failed to allocate phdrs");
202 		goto bad;
203 	}
204 
205 	for (i = 0; i < phnum; i++) {
206 		if (gelf_getphdr(elf, i, &phdr[i]) == NULL) {
207 			free(phdr);
208 			_kvm_err(kd, kd->program, "%s", elf_errmsg(0));
209 			goto bad;
210 		}
211 	}
212 	elf_end(elf);
213 	*phnump = phnum;
214 	*phdrp = phdr;
215 	return (0);
216 
217 bad:
218 	elf_end(elf);
219 	return (-1);
220 }
221 
222 /*
223  * Transform v such that only bits [bit0, bitN) may be set.  Generates a
224  * bitmask covering the number of bits, then shifts so +bit0+ is the first.
225  */
226 static uint64_t
227 bitmask_range(uint64_t v, uint64_t bit0, uint64_t bitN)
228 {
229 	if (bit0 == 0 && bitN == BITS_IN(v))
230 		return (v);
231 
232 	return (v & (((1ULL << (bitN - bit0)) - 1ULL) << bit0));
233 }
234 
235 /*
236  * Returns the number of bits in a given byte array range starting at a
237  * given base, from bit0 to bitN.  bit0 may be non-zero in the case of
238  * counting backwards from bitN.
239  */
240 static uint64_t
241 popcount_bytes(uint64_t *addr, uint32_t bit0, uint32_t bitN)
242 {
243 	uint32_t res = bitN - bit0;
244 	uint64_t count = 0;
245 	uint32_t bound;
246 
247 	/* Align to 64-bit boundary on the left side if needed. */
248 	if ((bit0 % BITS_IN(*addr)) != 0) {
249 		bound = MIN(bitN, roundup2(bit0, BITS_IN(*addr)));
250 		count += __bitcount64(bitmask_range(*addr, bit0, bound));
251 		res -= (bound - bit0);
252 		addr++;
253 	}
254 
255 	while (res > 0) {
256 		bound = MIN(res, BITS_IN(*addr));
257 		count += __bitcount64(bitmask_range(*addr, 0, bound));
258 		res -= bound;
259 		addr++;
260 	}
261 
262 	return (count);
263 }
264 
265 void *
266 _kvm_pmap_get(kvm_t *kd, u_long idx, size_t len)
267 {
268 	uintptr_t off = idx * len;
269 
270 	if ((off_t)off >= kd->pt_sparse_off)
271 		return (NULL);
272 	return (void *)((uintptr_t)kd->page_map + off);
273 }
274 
275 void *
276 _kvm_map_get(kvm_t *kd, u_long pa, unsigned int page_size)
277 {
278 	off_t off;
279 	uintptr_t addr;
280 
281 	off = _kvm_pt_find(kd, pa, page_size);
282 	if (off == -1)
283 		return NULL;
284 
285 	addr = (uintptr_t)kd->page_map + off;
286 	if (off >= kd->pt_sparse_off)
287 		addr = (uintptr_t)kd->sparse_map + (off - kd->pt_sparse_off);
288 	return (void *)addr;
289 }
290 
291 int
292 _kvm_pt_init(kvm_t *kd, size_t map_len, off_t map_off, off_t sparse_off,
293     int page_size, int word_size)
294 {
295 	uint64_t *addr;
296 	uint32_t *popcount_bin;
297 	int bin_popcounts = 0;
298 	uint64_t pc_bins, res;
299 	ssize_t rd;
300 
301 	/*
302 	 * Map the bitmap specified by the arguments.
303 	 */
304 	kd->pt_map = _kvm_malloc(kd, map_len);
305 	if (kd->pt_map == NULL) {
306 		_kvm_err(kd, kd->program, "cannot allocate %zu bytes for bitmap",
307 		    map_len);
308 		return (-1);
309 	}
310 	rd = pread(kd->pmfd, kd->pt_map, map_len, map_off);
311 	if (rd < 0 || rd != (ssize_t)map_len) {
312 		_kvm_err(kd, kd->program, "cannot read %zu bytes for bitmap",
313 		    map_len);
314 		return (-1);
315 	}
316 	kd->pt_map_size = map_len;
317 
318 	/*
319 	 * Generate a popcount cache for every POPCOUNT_BITS in the bitmap,
320 	 * so lookups only have to calculate the number of bits set between
321 	 * a cache point and their bit.  This reduces lookups to O(1),
322 	 * without significantly increasing memory requirements.
323 	 *
324 	 * Round up the number of bins so that 'upper half' lookups work for
325 	 * the final bin, if needed.  The first popcount is 0, since no bits
326 	 * precede bit 0, so add 1 for that also.  Without this, extra work
327 	 * would be needed to handle the first PTEs in _kvm_pt_find().
328 	 */
329 	addr = kd->pt_map;
330 	res = map_len;
331 	pc_bins = 1 + (res * NBBY + POPCOUNT_BITS / 2) / POPCOUNT_BITS;
332 	kd->pt_popcounts = calloc(pc_bins, sizeof(uint32_t));
333 	if (kd->pt_popcounts == NULL) {
334 		_kvm_err(kd, kd->program, "cannot allocate popcount bins");
335 		return (-1);
336 	}
337 
338 	for (popcount_bin = &kd->pt_popcounts[1]; res > 0;
339 	    addr++, res -= sizeof(*addr)) {
340 		*popcount_bin += popcount_bytes(addr, 0,
341 		    MIN(res * NBBY, BITS_IN(*addr)));
342 		if (++bin_popcounts == POPCOUNTS_IN(*addr)) {
343 			popcount_bin++;
344 			*popcount_bin = *(popcount_bin - 1);
345 			bin_popcounts = 0;
346 		}
347 	}
348 
349 	assert(pc_bins * sizeof(*popcount_bin) ==
350 	    ((uintptr_t)popcount_bin - (uintptr_t)kd->pt_popcounts));
351 
352 	kd->pt_sparse_off = sparse_off;
353 	kd->pt_sparse_size = (uint64_t)*popcount_bin * page_size;
354 	kd->pt_page_size = page_size;
355 	kd->pt_word_size = word_size;
356 
357 	/*
358 	 * Map the sparse page array.  This is useful for performing point
359 	 * lookups of specific pages, e.g. for kvm_walk_pages.  Generally,
360 	 * this is much larger than is reasonable to read in up front, so
361 	 * mmap it in instead.
362 	 */
363 	kd->sparse_map = mmap(NULL, kd->pt_sparse_size, PROT_READ,
364 	    MAP_PRIVATE, kd->pmfd, kd->pt_sparse_off);
365 	if (kd->sparse_map == MAP_FAILED) {
366 		_kvm_err(kd, kd->program, "cannot map %" PRIu64
367 		    " bytes from fd %d offset %jd for sparse map: %s",
368 		    kd->pt_sparse_size, kd->pmfd,
369 		    (intmax_t)kd->pt_sparse_off, strerror(errno));
370 		return (-1);
371 	}
372 	return (0);
373 }
374 
375 int
376 _kvm_pmap_init(kvm_t *kd, uint32_t pmap_size, off_t pmap_off)
377 {
378 	ssize_t exp_len = pmap_size;
379 
380 	kd->page_map_size = pmap_size;
381 	kd->page_map_off = pmap_off;
382 	kd->page_map = _kvm_malloc(kd, pmap_size);
383 	if (kd->page_map == NULL) {
384 		_kvm_err(kd, kd->program, "cannot allocate %u bytes "
385 		    "for page map", pmap_size);
386 		return (-1);
387 	}
388 	if (pread(kd->pmfd, kd->page_map, pmap_size, pmap_off) != exp_len) {
389 		_kvm_err(kd, kd->program, "cannot read %d bytes from "
390 		    "offset %jd for page map", pmap_size, (intmax_t)pmap_off);
391 		return (-1);
392 	}
393 	return (0);
394 }
395 
396 /*
397  * Find the offset for the given physical page address; returns -1 otherwise.
398  *
399  * A page's offset is represented by the sparse page base offset plus the
400  * number of bits set before its bit multiplied by page size.  This means
401  * that if a page exists in the dump, it's necessary to know how many pages
402  * in the dump precede it.  Reduce this O(n) counting to O(1) by caching the
403  * number of bits set at POPCOUNT_BITS intervals.
404  *
405  * Then to find the number of pages before the requested address, simply
406  * index into the cache and count the number of bits set between that cache
407  * bin and the page's bit.  Halve the number of bytes that have to be
408  * checked by also counting down from the next higher bin if it's closer.
409  */
410 off_t
411 _kvm_pt_find(kvm_t *kd, uint64_t pa, unsigned int page_size)
412 {
413 	uint64_t *bitmap = kd->pt_map;
414 	uint64_t pte_bit_id = pa / page_size;
415 	uint64_t pte_u64 = pte_bit_id / BITS_IN(*bitmap);
416 	uint64_t popcount_id = pte_bit_id / POPCOUNT_BITS;
417 	uint64_t pte_mask = 1ULL << (pte_bit_id % BITS_IN(*bitmap));
418 	uint64_t bitN;
419 	uint32_t count;
420 
421 	/* Check whether the page address requested is in the dump. */
422 	if (pte_bit_id >= (kd->pt_map_size * NBBY) ||
423 	    (bitmap[pte_u64] & pte_mask) == 0)
424 		return (-1);
425 
426 	/*
427 	 * Add/sub popcounts from the bitmap until the PTE's bit is reached.
428 	 * For bits that are in the upper half between the calculated
429 	 * popcount id and the next one, use the next one and subtract to
430 	 * minimize the number of popcounts required.
431 	 */
432 	if ((pte_bit_id % POPCOUNT_BITS) < (POPCOUNT_BITS / 2)) {
433 		count = kd->pt_popcounts[popcount_id] + popcount_bytes(
434 		    bitmap + popcount_id * POPCOUNTS_IN(*bitmap),
435 		    0, pte_bit_id - popcount_id * POPCOUNT_BITS);
436 	} else {
437 		/*
438 		 * Counting in reverse is trickier, since we must avoid
439 		 * reading from bytes that are not in range, and invert.
440 		 */
441 		uint64_t pte_u64_bit_off = pte_u64 * BITS_IN(*bitmap);
442 
443 		popcount_id++;
444 		bitN = MIN(popcount_id * POPCOUNT_BITS,
445 		    kd->pt_map_size * BITS_IN(uint8_t));
446 		count = kd->pt_popcounts[popcount_id] - popcount_bytes(
447 		    bitmap + pte_u64,
448 		    pte_bit_id - pte_u64_bit_off, bitN - pte_u64_bit_off);
449 	}
450 
451 	/*
452 	 * This can only happen if the core is truncated.  Treat these
453 	 * entries as if they don't exist, since their backing doesn't.
454 	 */
455 	if (count >= (kd->pt_sparse_size / page_size))
456 		return (-1);
457 
458 	return (kd->pt_sparse_off + (uint64_t)count * page_size);
459 }
460 
461 static int
462 kvm_fdnlist(kvm_t *kd, struct kvm_nlist *list)
463 {
464 	kvaddr_t addr;
465 	int error, nfail;
466 
467 	if (kd->resolve_symbol == NULL) {
468 		struct nlist *nl;
469 		int count, i;
470 
471 		for (count = 0; list[count].n_name != NULL &&
472 		     list[count].n_name[0] != '\0'; count++)
473 			;
474 		nl = calloc(count + 1, sizeof(*nl));
475 		for (i = 0; i < count; i++)
476 			nl[i].n_name = list[i].n_name;
477 		nfail = __fdnlist(kd->nlfd, nl);
478 		for (i = 0; i < count; i++) {
479 			list[i].n_type = nl[i].n_type;
480 			list[i].n_value = nl[i].n_value;
481 		}
482 		free(nl);
483 		return (nfail);
484 	}
485 
486 	nfail = 0;
487 	while (list->n_name != NULL && list->n_name[0] != '\0') {
488 		error = kd->resolve_symbol(list->n_name, &addr);
489 		if (error != 0) {
490 			nfail++;
491 			list->n_value = 0;
492 			list->n_type = 0;
493 		} else {
494 			list->n_value = addr;
495 			list->n_type = N_DATA | N_EXT;
496 		}
497 		list++;
498 	}
499 	return (nfail);
500 }
501 
502 /*
503  * Walk the list of unresolved symbols, generate a new list and prefix the
504  * symbol names, try again, and merge back what we could resolve.
505  */
506 static int
507 kvm_fdnlist_prefix(kvm_t *kd, struct kvm_nlist *nl, int missing,
508     const char *prefix, kvaddr_t (*validate_fn)(kvm_t *, kvaddr_t))
509 {
510 	struct kvm_nlist *n, *np, *p;
511 	char *cp, *ce;
512 	const char *ccp;
513 	size_t len;
514 	int slen, unresolved;
515 
516 	/*
517 	 * Calculate the space we need to malloc for nlist and names.
518 	 * We are going to store the name twice for later lookups: once
519 	 * with the prefix and once the unmodified name delmited by \0.
520 	 */
521 	len = 0;
522 	unresolved = 0;
523 	for (p = nl; p->n_name && p->n_name[0]; ++p) {
524 		if (p->n_type != N_UNDF)
525 			continue;
526 		len += sizeof(struct kvm_nlist) + strlen(prefix) +
527 		    2 * (strlen(p->n_name) + 1);
528 		unresolved++;
529 	}
530 	if (unresolved == 0)
531 		return (unresolved);
532 	/* Add space for the terminating nlist entry. */
533 	len += sizeof(struct kvm_nlist);
534 	unresolved++;
535 
536 	/* Alloc one chunk for (nlist, [names]) and setup pointers. */
537 	n = np = malloc(len);
538 	bzero(n, len);
539 	if (n == NULL)
540 		return (missing);
541 	cp = ce = (char *)np;
542 	cp += unresolved * sizeof(struct kvm_nlist);
543 	ce += len;
544 
545 	/* Generate shortened nlist with special prefix. */
546 	unresolved = 0;
547 	for (p = nl; p->n_name && p->n_name[0]; ++p) {
548 		if (p->n_type != N_UNDF)
549 			continue;
550 		*np = *p;
551 		/* Save the new\0orig. name so we can later match it again. */
552 		slen = snprintf(cp, ce - cp, "%s%s%c%s", prefix,
553 		    (prefix[0] != '\0' && p->n_name[0] == '_') ?
554 			(p->n_name + 1) : p->n_name, '\0', p->n_name);
555 		if (slen < 0 || slen >= ce - cp)
556 			continue;
557 		np->n_name = cp;
558 		cp += slen + 1;
559 		np++;
560 		unresolved++;
561 	}
562 
563 	/* Do lookup on the reduced list. */
564 	np = n;
565 	unresolved = kvm_fdnlist(kd, np);
566 
567 	/* Check if we could resolve further symbols and update the list. */
568 	if (unresolved >= 0 && unresolved < missing) {
569 		/* Find the first freshly resolved entry. */
570 		for (; np->n_name && np->n_name[0]; np++)
571 			if (np->n_type != N_UNDF)
572 				break;
573 		/*
574 		 * The lists are both in the same order,
575 		 * so we can walk them in parallel.
576 		 */
577 		for (p = nl; np->n_name && np->n_name[0] &&
578 		    p->n_name && p->n_name[0]; ++p) {
579 			if (p->n_type != N_UNDF)
580 				continue;
581 			/* Skip expanded name and compare to orig. one. */
582 			ccp = np->n_name + strlen(np->n_name) + 1;
583 			if (strcmp(ccp, p->n_name) != 0)
584 				continue;
585 			/* Update nlist with new, translated results. */
586 			p->n_type = np->n_type;
587 			if (validate_fn)
588 				p->n_value = (*validate_fn)(kd, np->n_value);
589 			else
590 				p->n_value = np->n_value;
591 			missing--;
592 			/* Find next freshly resolved entry. */
593 			for (np++; np->n_name && np->n_name[0]; np++)
594 				if (np->n_type != N_UNDF)
595 					break;
596 		}
597 	}
598 	/* We could assert missing = unresolved here. */
599 
600 	free(n);
601 	return (unresolved);
602 }
603 
604 int
605 _kvm_nlist(kvm_t *kd, struct kvm_nlist *nl, int initialize)
606 {
607 	struct kvm_nlist *p;
608 	int nvalid;
609 	struct kld_sym_lookup lookup;
610 	int error;
611 	const char *prefix = "";
612 	char symname[1024]; /* XXX-BZ symbol name length limit? */
613 	int tried_vnet, tried_dpcpu;
614 
615 	/*
616 	 * If we can't use the kld symbol lookup, revert to the
617 	 * slow library call.
618 	 */
619 	if (!ISALIVE(kd)) {
620 		error = kvm_fdnlist(kd, nl);
621 		if (error <= 0)			/* Hard error or success. */
622 			return (error);
623 
624 		if (_kvm_vnet_initialized(kd, initialize))
625 			error = kvm_fdnlist_prefix(kd, nl, error,
626 			    VNET_SYMPREFIX, _kvm_vnet_validaddr);
627 
628 		if (error > 0 && _kvm_dpcpu_initialized(kd, initialize))
629 			error = kvm_fdnlist_prefix(kd, nl, error,
630 			    DPCPU_SYMPREFIX, _kvm_dpcpu_validaddr);
631 
632 		return (error);
633 	}
634 
635 	/*
636 	 * We can use the kld lookup syscall.  Go through each nlist entry
637 	 * and look it up with a kldsym(2) syscall.
638 	 */
639 	nvalid = 0;
640 	tried_vnet = 0;
641 	tried_dpcpu = 0;
642 again:
643 	for (p = nl; p->n_name && p->n_name[0]; ++p) {
644 		if (p->n_type != N_UNDF)
645 			continue;
646 
647 		lookup.version = sizeof(lookup);
648 		lookup.symvalue = 0;
649 		lookup.symsize = 0;
650 
651 		error = snprintf(symname, sizeof(symname), "%s%s", prefix,
652 		    (prefix[0] != '\0' && p->n_name[0] == '_') ?
653 			(p->n_name + 1) : p->n_name);
654 		if (error < 0 || error >= (int)sizeof(symname))
655 			continue;
656 		lookup.symname = symname;
657 		if (lookup.symname[0] == '_')
658 			lookup.symname++;
659 
660 		if (kldsym(0, KLDSYM_LOOKUP, &lookup) != -1) {
661 			p->n_type = N_TEXT;
662 			if (_kvm_vnet_initialized(kd, initialize) &&
663 			    strcmp(prefix, VNET_SYMPREFIX) == 0)
664 				p->n_value =
665 				    _kvm_vnet_validaddr(kd, lookup.symvalue);
666 			else if (_kvm_dpcpu_initialized(kd, initialize) &&
667 			    strcmp(prefix, DPCPU_SYMPREFIX) == 0)
668 				p->n_value =
669 				    _kvm_dpcpu_validaddr(kd, lookup.symvalue);
670 			else
671 				p->n_value = lookup.symvalue;
672 			++nvalid;
673 			/* lookup.symsize */
674 		}
675 	}
676 
677 	/*
678 	 * Check the number of entries that weren't found. If they exist,
679 	 * try again with a prefix for virtualized or DPCPU symbol names.
680 	 */
681 	error = ((p - nl) - nvalid);
682 	if (error && _kvm_vnet_initialized(kd, initialize) && !tried_vnet) {
683 		tried_vnet = 1;
684 		prefix = VNET_SYMPREFIX;
685 		goto again;
686 	}
687 	if (error && _kvm_dpcpu_initialized(kd, initialize) && !tried_dpcpu) {
688 		tried_dpcpu = 1;
689 		prefix = DPCPU_SYMPREFIX;
690 		goto again;
691 	}
692 
693 	/*
694 	 * Return the number of entries that weren't found. If they exist,
695 	 * also fill internal error buffer.
696 	 */
697 	error = ((p - nl) - nvalid);
698 	if (error)
699 		_kvm_syserr(kd, kd->program, "kvm_nlist");
700 	return (error);
701 }
702 
703 int
704 _kvm_bitmap_init(struct kvm_bitmap *bm, u_long bitmapsize, u_long *idx)
705 {
706 
707 	*idx = ULONG_MAX;
708 	bm->map = calloc(bitmapsize, sizeof *bm->map);
709 	if (bm->map == NULL)
710 		return (0);
711 	bm->size = bitmapsize;
712 	return (1);
713 }
714 
715 void
716 _kvm_bitmap_set(struct kvm_bitmap *bm, u_long pa, unsigned int page_size)
717 {
718 	u_long bm_index = pa / page_size;
719 	uint8_t *byte = &bm->map[bm_index / 8];
720 
721 	*byte |= (1UL << (bm_index % 8));
722 }
723 
724 int
725 _kvm_bitmap_next(struct kvm_bitmap *bm, u_long *idx)
726 {
727 	u_long first_invalid = bm->size * CHAR_BIT;
728 
729 	if (*idx == ULONG_MAX)
730 		*idx = 0;
731 	else
732 		(*idx)++;
733 
734 	/* Find the next valid idx. */
735 	for (; *idx < first_invalid; (*idx)++) {
736 		unsigned int mask = *idx % CHAR_BIT;
737 		if ((bm->map[*idx * CHAR_BIT] & mask) == 0)
738 			break;
739 	}
740 
741 	return (*idx < first_invalid);
742 }
743 
744 void
745 _kvm_bitmap_deinit(struct kvm_bitmap *bm)
746 {
747 
748 	free(bm->map);
749 }
750 
751 int
752 _kvm_visit_cb(kvm_t *kd, kvm_walk_pages_cb_t *cb, void *arg, u_long pa,
753     u_long kmap_vaddr, u_long dmap_vaddr, vm_prot_t prot, size_t len,
754     unsigned int page_size)
755 {
756 	unsigned int pgsz = page_size ? page_size : len;
757 	struct kvm_page p = {
758 		.kp_version = LIBKVM_WALK_PAGES_VERSION,
759 		.kp_paddr = pa,
760 		.kp_kmap_vaddr = kmap_vaddr,
761 		.kp_dmap_vaddr = dmap_vaddr,
762 		.kp_prot = prot,
763 		.kp_offset = _kvm_pt_find(kd, pa, pgsz),
764 		.kp_len = len,
765 	};
766 
767 	return cb(&p, arg);
768 }
769