xref: /titanic_51/usr/src/boot/sys/boot/zfs/zfsimpl.c (revision 376c1df2259df18a52137f7451f98953ffd01484)
1 /*-
2  * Copyright (c) 2007 Doug Rabson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 
29 /*
30  *	Stand-alone ZFS file reader.
31  */
32 
33 #include <sys/stat.h>
34 #include <sys/stdint.h>
35 
36 #include "zfsimpl.h"
37 #include "zfssubr.c"
38 
39 
40 struct zfsmount {
41 	const spa_t	*spa;
42 	objset_phys_t	objset;
43 	uint64_t	rootobj;
44 };
45 
46 /*
47  * List of all vdevs, chained through v_alllink.
48  */
49 static vdev_list_t zfs_vdevs;
50 
51  /*
52  * List of ZFS features supported for read
53  */
54 static const char *features_for_read[] = {
55 	"org.illumos:lz4_compress",
56 	"com.delphix:hole_birth",
57 	"com.delphix:extensible_dataset",
58 	"com.delphix:embedded_data",
59 	"org.open-zfs:large_blocks",
60 	"org.illumos:sha512",
61 	"org.zfsonlinux:large_dnode",
62 	NULL
63 };
64 
65 /*
66  * List of all pools, chained through spa_link.
67  */
68 static spa_list_t zfs_pools;
69 
70 static const dnode_phys_t *dnode_cache_obj;
71 static uint64_t dnode_cache_bn;
72 static char *dnode_cache_buf;
73 static char *zap_scratch;
74 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
75 
76 #define TEMP_SIZE	(1024 * 1024)
77 
78 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
79 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
80 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
81 
82 static void
83 zfs_init(void)
84 {
85 	STAILQ_INIT(&zfs_vdevs);
86 	STAILQ_INIT(&zfs_pools);
87 
88 	zfs_temp_buf = malloc(TEMP_SIZE);
89 	zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
90 	zfs_temp_ptr = zfs_temp_buf;
91 	dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
92 	zap_scratch = malloc(SPA_MAXBLOCKSIZE);
93 
94 	zfs_init_crc();
95 }
96 
97 static void *
98 zfs_alloc(size_t size)
99 {
100 	char *ptr;
101 
102 	if (zfs_temp_ptr + size > zfs_temp_end) {
103 		printf("ZFS: out of temporary buffer space\n");
104 		for (;;) ;
105 	}
106 	ptr = zfs_temp_ptr;
107 	zfs_temp_ptr += size;
108 
109 	return (ptr);
110 }
111 
112 static void
113 zfs_free(void *ptr, size_t size)
114 {
115 
116 	zfs_temp_ptr -= size;
117 	if (zfs_temp_ptr != ptr) {
118 		printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
119 		for (;;) ;
120 	}
121 }
122 
123 static int
124 xdr_int(const unsigned char **xdr, int *ip)
125 {
126 	*ip = ((*xdr)[0] << 24)
127 		| ((*xdr)[1] << 16)
128 		| ((*xdr)[2] << 8)
129 		| ((*xdr)[3] << 0);
130 	(*xdr) += 4;
131 	return (0);
132 }
133 
134 static int
135 xdr_u_int(const unsigned char **xdr, u_int *ip)
136 {
137 	*ip = ((*xdr)[0] << 24)
138 		| ((*xdr)[1] << 16)
139 		| ((*xdr)[2] << 8)
140 		| ((*xdr)[3] << 0);
141 	(*xdr) += 4;
142 	return (0);
143 }
144 
145 static int
146 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
147 {
148 	u_int hi, lo;
149 
150 	xdr_u_int(xdr, &hi);
151 	xdr_u_int(xdr, &lo);
152 	*lp = (((uint64_t) hi) << 32) | lo;
153 	return (0);
154 }
155 
156 static int
157 nvlist_find(const unsigned char *nvlist, const char *name, int type,
158 	    int* elementsp, void *valuep)
159 {
160 	const unsigned char *p, *pair;
161 	int junk;
162 	int encoded_size, decoded_size;
163 
164 	p = nvlist;
165 	xdr_int(&p, &junk);
166 	xdr_int(&p, &junk);
167 
168 	pair = p;
169 	xdr_int(&p, &encoded_size);
170 	xdr_int(&p, &decoded_size);
171 	while (encoded_size && decoded_size) {
172 		int namelen, pairtype, elements;
173 		const char *pairname;
174 
175 		xdr_int(&p, &namelen);
176 		pairname = (const char*) p;
177 		p += roundup(namelen, 4);
178 		xdr_int(&p, &pairtype);
179 
180 		if (!memcmp(name, pairname, namelen) && type == pairtype) {
181 			xdr_int(&p, &elements);
182 			if (elementsp)
183 				*elementsp = elements;
184 			if (type == DATA_TYPE_UINT64) {
185 				xdr_uint64_t(&p, (uint64_t *) valuep);
186 				return (0);
187 			} else if (type == DATA_TYPE_STRING) {
188 				int len;
189 				xdr_int(&p, &len);
190 				(*(const char**) valuep) = (const char*) p;
191 				return (0);
192 			} else if (type == DATA_TYPE_NVLIST
193 				   || type == DATA_TYPE_NVLIST_ARRAY) {
194 				(*(const unsigned char**) valuep) =
195 					 (const unsigned char*) p;
196 				return (0);
197 			} else {
198 				return (EIO);
199 			}
200 		} else {
201 			/*
202 			 * Not the pair we are looking for, skip to the next one.
203 			 */
204 			p = pair + encoded_size;
205 		}
206 
207 		pair = p;
208 		xdr_int(&p, &encoded_size);
209 		xdr_int(&p, &decoded_size);
210 	}
211 
212 	return (EIO);
213 }
214 
215 static int
216 nvlist_check_features_for_read(const unsigned char *nvlist)
217 {
218 	const unsigned char *p, *pair;
219 	int junk;
220 	int encoded_size, decoded_size;
221 	int rc;
222 
223 	rc = 0;
224 
225 	p = nvlist;
226 	xdr_int(&p, &junk);
227 	xdr_int(&p, &junk);
228 
229 	pair = p;
230 	xdr_int(&p, &encoded_size);
231 	xdr_int(&p, &decoded_size);
232 	while (encoded_size && decoded_size) {
233 		int namelen, pairtype;
234 		const char *pairname;
235 		int i, found;
236 
237 		found = 0;
238 
239 		xdr_int(&p, &namelen);
240 		pairname = (const char*) p;
241 		p += roundup(namelen, 4);
242 		xdr_int(&p, &pairtype);
243 
244 		for (i = 0; features_for_read[i] != NULL; i++) {
245 			if (!memcmp(pairname, features_for_read[i], namelen)) {
246 				found = 1;
247 				break;
248 			}
249 		}
250 
251 		if (!found) {
252 			printf("ZFS: unsupported feature: %s\n", pairname);
253 			rc = EIO;
254 		}
255 
256 		p = pair + encoded_size;
257 
258 		pair = p;
259 		xdr_int(&p, &encoded_size);
260 		xdr_int(&p, &decoded_size);
261 	}
262 
263 	return (rc);
264 }
265 
266 /*
267  * Return the next nvlist in an nvlist array.
268  */
269 static const unsigned char *
270 nvlist_next(const unsigned char *nvlist)
271 {
272 	const unsigned char *p, *pair;
273 	int junk;
274 	int encoded_size, decoded_size;
275 
276 	p = nvlist;
277 	xdr_int(&p, &junk);
278 	xdr_int(&p, &junk);
279 
280 	pair = p;
281 	xdr_int(&p, &encoded_size);
282 	xdr_int(&p, &decoded_size);
283 	while (encoded_size && decoded_size) {
284 		p = pair + encoded_size;
285 
286 		pair = p;
287 		xdr_int(&p, &encoded_size);
288 		xdr_int(&p, &decoded_size);
289 	}
290 
291 	return p;
292 }
293 
294 #ifdef TEST
295 
296 static const unsigned char *
297 nvlist_print(const unsigned char *nvlist, unsigned int indent)
298 {
299 	static const char* typenames[] = {
300 		"DATA_TYPE_UNKNOWN",
301 		"DATA_TYPE_BOOLEAN",
302 		"DATA_TYPE_BYTE",
303 		"DATA_TYPE_INT16",
304 		"DATA_TYPE_UINT16",
305 		"DATA_TYPE_INT32",
306 		"DATA_TYPE_UINT32",
307 		"DATA_TYPE_INT64",
308 		"DATA_TYPE_UINT64",
309 		"DATA_TYPE_STRING",
310 		"DATA_TYPE_BYTE_ARRAY",
311 		"DATA_TYPE_INT16_ARRAY",
312 		"DATA_TYPE_UINT16_ARRAY",
313 		"DATA_TYPE_INT32_ARRAY",
314 		"DATA_TYPE_UINT32_ARRAY",
315 		"DATA_TYPE_INT64_ARRAY",
316 		"DATA_TYPE_UINT64_ARRAY",
317 		"DATA_TYPE_STRING_ARRAY",
318 		"DATA_TYPE_HRTIME",
319 		"DATA_TYPE_NVLIST",
320 		"DATA_TYPE_NVLIST_ARRAY",
321 		"DATA_TYPE_BOOLEAN_VALUE",
322 		"DATA_TYPE_INT8",
323 		"DATA_TYPE_UINT8",
324 		"DATA_TYPE_BOOLEAN_ARRAY",
325 		"DATA_TYPE_INT8_ARRAY",
326 		"DATA_TYPE_UINT8_ARRAY"
327 	};
328 
329 	unsigned int i, j;
330 	const unsigned char *p, *pair;
331 	int junk;
332 	int encoded_size, decoded_size;
333 
334 	p = nvlist;
335 	xdr_int(&p, &junk);
336 	xdr_int(&p, &junk);
337 
338 	pair = p;
339 	xdr_int(&p, &encoded_size);
340 	xdr_int(&p, &decoded_size);
341 	while (encoded_size && decoded_size) {
342 		int namelen, pairtype, elements;
343 		const char *pairname;
344 
345 		xdr_int(&p, &namelen);
346 		pairname = (const char*) p;
347 		p += roundup(namelen, 4);
348 		xdr_int(&p, &pairtype);
349 
350 		for (i = 0; i < indent; i++)
351 			printf(" ");
352 		printf("%s %s", typenames[pairtype], pairname);
353 
354 		xdr_int(&p, &elements);
355 		switch (pairtype) {
356 		case DATA_TYPE_UINT64: {
357 			uint64_t val;
358 			xdr_uint64_t(&p, &val);
359 			printf(" = 0x%jx\n", (uintmax_t)val);
360 			break;
361 		}
362 
363 		case DATA_TYPE_STRING: {
364 			int len;
365 			xdr_int(&p, &len);
366 			printf(" = \"%s\"\n", p);
367 			break;
368 		}
369 
370 		case DATA_TYPE_NVLIST:
371 			printf("\n");
372 			nvlist_print(p, indent + 1);
373 			break;
374 
375 		case DATA_TYPE_NVLIST_ARRAY:
376 			for (j = 0; j < elements; j++) {
377 				printf("[%d]\n", j);
378 				p = nvlist_print(p, indent + 1);
379 				if (j != elements - 1) {
380 					for (i = 0; i < indent; i++)
381 						printf(" ");
382 					printf("%s %s", typenames[pairtype], pairname);
383 				}
384 			}
385 			break;
386 
387 		default:
388 			printf("\n");
389 		}
390 
391 		p = pair + encoded_size;
392 
393 		pair = p;
394 		xdr_int(&p, &encoded_size);
395 		xdr_int(&p, &decoded_size);
396 	}
397 
398 	return p;
399 }
400 
401 #endif
402 
403 static int
404 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
405     off_t offset, size_t size)
406 {
407 	size_t psize;
408 	int rc;
409 
410 	if (!vdev->v_phys_read)
411 		return (EIO);
412 
413 	if (bp) {
414 		psize = BP_GET_PSIZE(bp);
415 	} else {
416 		psize = size;
417 	}
418 
419 	/*printf("ZFS: reading %zu bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
420 	rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
421 	if (rc)
422 		return (rc);
423 	if (bp && zio_checksum_verify(bp, buf))
424 		return (EIO);
425 
426 	return (0);
427 }
428 
429 static int
430 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
431     off_t offset, size_t bytes)
432 {
433 
434 	return (vdev_read_phys(vdev, bp, buf,
435 		offset + VDEV_LABEL_START_SIZE, bytes));
436 }
437 
438 
439 static int
440 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
441     off_t offset, size_t bytes)
442 {
443 	vdev_t *kid;
444 	int rc;
445 
446 	rc = EIO;
447 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
448 		if (kid->v_state != VDEV_STATE_HEALTHY)
449 			continue;
450 		rc = kid->v_read(kid, bp, buf, offset, bytes);
451 		if (!rc)
452 			return (0);
453 	}
454 
455 	return (rc);
456 }
457 
458 static int
459 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
460     off_t offset, size_t bytes)
461 {
462 	vdev_t *kid;
463 
464 	/*
465 	 * Here we should have two kids:
466 	 * First one which is the one we are replacing and we can trust
467 	 * only this one to have valid data, but it might not be present.
468 	 * Second one is that one we are replacing with. It is most likely
469 	 * healthy, but we can't trust it has needed data, so we won't use it.
470 	 */
471 	kid = STAILQ_FIRST(&vdev->v_children);
472 	if (kid == NULL)
473 		return (EIO);
474 	if (kid->v_state != VDEV_STATE_HEALTHY)
475 		return (EIO);
476 	return (kid->v_read(kid, bp, buf, offset, bytes));
477 }
478 
479 static vdev_t *
480 vdev_find(uint64_t guid)
481 {
482 	vdev_t *vdev;
483 
484 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
485 		if (vdev->v_guid == guid)
486 			return (vdev);
487 
488 	return (0);
489 }
490 
491 static vdev_t *
492 vdev_create(uint64_t guid, vdev_read_t *vdev_read)
493 {
494 	vdev_t *vdev;
495 
496 	vdev = malloc(sizeof(vdev_t));
497 	memset(vdev, 0, sizeof(vdev_t));
498 	STAILQ_INIT(&vdev->v_children);
499 	vdev->v_guid = guid;
500 	vdev->v_state = VDEV_STATE_OFFLINE;
501 	vdev->v_read = vdev_read;
502 	vdev->v_phys_read = 0;
503 	vdev->v_read_priv = 0;
504 	STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
505 
506 	return (vdev);
507 }
508 
509 static int
510 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
511     vdev_t **vdevp, int is_newer)
512 {
513 	int rc;
514 	uint64_t guid, id, ashift, nparity;
515 	const char *type;
516 	const char *path;
517 	vdev_t *vdev, *kid;
518 	const unsigned char *kids;
519 	int nkids, i, is_new;
520 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
521 
522 	if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
523 	    NULL, &guid) ||
524 	    nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id) ||
525 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
526 	    NULL, &type)) {
527 		printf("ZFS: can't find vdev details\n");
528 		return (ENOENT);
529 	}
530 
531 	if (strcmp(type, VDEV_TYPE_MIRROR)
532 	    && strcmp(type, VDEV_TYPE_DISK)
533 #ifdef ZFS_TEST
534 	    && strcmp(type, VDEV_TYPE_FILE)
535 #endif
536 	    && strcmp(type, VDEV_TYPE_RAIDZ)
537 	    && strcmp(type, VDEV_TYPE_REPLACING)) {
538 		printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
539 		return (EIO);
540 	}
541 
542 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
543 
544 	nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
545 	    &is_offline);
546 	nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
547 	    &is_removed);
548 	nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
549 	    &is_faulted);
550 	nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, NULL,
551 	    &is_degraded);
552 	nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, NULL,
553 	    &isnt_present);
554 
555 	vdev = vdev_find(guid);
556 	if (!vdev) {
557 		is_new = 1;
558 
559 		if (!strcmp(type, VDEV_TYPE_MIRROR))
560 			vdev = vdev_create(guid, vdev_mirror_read);
561 		else if (!strcmp(type, VDEV_TYPE_RAIDZ))
562 			vdev = vdev_create(guid, vdev_raidz_read);
563 		else if (!strcmp(type, VDEV_TYPE_REPLACING))
564 			vdev = vdev_create(guid, vdev_replacing_read);
565 		else
566 			vdev = vdev_create(guid, vdev_disk_read);
567 
568 		vdev->v_id = id;
569 		vdev->v_top = pvdev != NULL ? pvdev : vdev;
570 		if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
571 		    DATA_TYPE_UINT64, NULL, &ashift) == 0) {
572 			vdev->v_ashift = ashift;
573 		} else {
574 			vdev->v_ashift = 0;
575 		}
576 		if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
577 		    DATA_TYPE_UINT64, NULL, &nparity) == 0) {
578 			vdev->v_nparity = nparity;
579 		} else {
580 			vdev->v_nparity = 0;
581 		}
582 		if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
583 		    DATA_TYPE_STRING, NULL, &path) == 0) {
584 			if (strncmp(path, "/dev/dsk/", 9) == 0)
585 				path += 9;
586 			vdev->v_name = strdup(path);
587 			if (nvlist_find(nvlist, ZPOOL_CONFIG_PHYS_PATH,
588 			    DATA_TYPE_STRING, NULL, &path) == 0) {
589 				vdev->v_phys_path = strdup(path);
590 			} else {
591 				vdev->v_phys_path = NULL;
592 			}
593 			if (nvlist_find(nvlist, ZPOOL_CONFIG_DEVID,
594 			    DATA_TYPE_STRING, NULL, &path) == 0) {
595 				vdev->v_devid = strdup(path);
596 			} else {
597 				vdev->v_devid = NULL;
598 			}
599 		} else {
600 			if (!strcmp(type, "raidz")) {
601 				if (vdev->v_nparity == 1)
602 					vdev->v_name = "raidz1";
603 				else if (vdev->v_nparity == 2)
604 					vdev->v_name = "raidz2";
605 				else if (vdev->v_nparity == 3)
606 					vdev->v_name = "raidz3";
607 				else {
608 					printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
609 					return (EIO);
610 				}
611 			} else {
612 				vdev->v_name = strdup(type);
613 			}
614 		}
615 	} else {
616 		is_new = 0;
617 	}
618 
619 	if (is_new || is_newer) {
620 		/*
621 		 * This is either new vdev or we've already seen this vdev,
622 		 * but from an older vdev label, so let's refresh its state
623 		 * from the newer label.
624 		 */
625 		if (is_offline)
626 			vdev->v_state = VDEV_STATE_OFFLINE;
627 		else if (is_removed)
628 			vdev->v_state = VDEV_STATE_REMOVED;
629 		else if (is_faulted)
630 			vdev->v_state = VDEV_STATE_FAULTED;
631 		else if (is_degraded)
632 			vdev->v_state = VDEV_STATE_DEGRADED;
633 		else if (isnt_present)
634 			vdev->v_state = VDEV_STATE_CANT_OPEN;
635 	}
636 
637 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
638 	    &nkids, &kids);
639 	/*
640 	 * Its ok if we don't have any kids.
641 	 */
642 	if (rc == 0) {
643 		vdev->v_nchildren = nkids;
644 		for (i = 0; i < nkids; i++) {
645 			rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
646 			if (rc)
647 				return (rc);
648 			if (is_new)
649 				STAILQ_INSERT_TAIL(&vdev->v_children, kid,
650 						   v_childlink);
651 			kids = nvlist_next(kids);
652 		}
653 	} else {
654 		vdev->v_nchildren = 0;
655 	}
656 
657 	if (vdevp)
658 		*vdevp = vdev;
659 	return (0);
660 }
661 
662 static void
663 vdev_set_state(vdev_t *vdev)
664 {
665 	vdev_t *kid;
666 	int good_kids;
667 	int bad_kids;
668 
669 	/*
670 	 * A mirror or raidz is healthy if all its kids are healthy. A
671 	 * mirror is degraded if any of its kids is healthy; a raidz
672 	 * is degraded if at most nparity kids are offline.
673 	 */
674 	if (STAILQ_FIRST(&vdev->v_children)) {
675 		good_kids = 0;
676 		bad_kids = 0;
677 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
678 			if (kid->v_state == VDEV_STATE_HEALTHY)
679 				good_kids++;
680 			else
681 				bad_kids++;
682 		}
683 		if (bad_kids == 0) {
684 			vdev->v_state = VDEV_STATE_HEALTHY;
685 		} else {
686 			if (vdev->v_read == vdev_mirror_read) {
687 				if (good_kids) {
688 					vdev->v_state = VDEV_STATE_DEGRADED;
689 				} else {
690 					vdev->v_state = VDEV_STATE_OFFLINE;
691 				}
692 			} else if (vdev->v_read == vdev_raidz_read) {
693 				if (bad_kids > vdev->v_nparity) {
694 					vdev->v_state = VDEV_STATE_OFFLINE;
695 				} else {
696 					vdev->v_state = VDEV_STATE_DEGRADED;
697 				}
698 			}
699 		}
700 	}
701 }
702 
703 static spa_t *
704 spa_find_by_guid(uint64_t guid)
705 {
706 	spa_t *spa;
707 
708 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
709 		if (spa->spa_guid == guid)
710 			return (spa);
711 
712 	return (0);
713 }
714 
715 static spa_t *
716 spa_find_by_name(const char *name)
717 {
718 	spa_t *spa;
719 
720 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
721 		if (!strcmp(spa->spa_name, name))
722 			return (spa);
723 
724 	return (0);
725 }
726 
727 spa_t *
728 spa_get_primary(void)
729 {
730 	return (STAILQ_FIRST(&zfs_pools));
731 }
732 
733 vdev_t *
734 spa_get_primary_vdev(const spa_t *spa)
735 {
736 	vdev_t *vdev;
737 	vdev_t *kid;
738 
739 	if (spa == NULL)
740 		spa = spa_get_primary();
741 	if (spa == NULL)
742 		return (NULL);
743 	vdev = STAILQ_FIRST(&spa->spa_vdevs);
744 	if (vdev == NULL)
745 		return (NULL);
746 	for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
747 	     kid = STAILQ_FIRST(&vdev->v_children))
748 		vdev = kid;
749 	return (vdev);
750 }
751 
752 static spa_t *
753 spa_create(uint64_t guid, const char *name)
754 {
755 	spa_t *spa;
756 
757 	if ((spa = malloc(sizeof(spa_t))) == NULL)
758 		return (NULL);
759 	memset(spa, 0, sizeof(spa_t));
760 	if ((spa->spa_name = strdup(name)) == NULL) {
761 		free(spa);
762 		return (NULL);
763 	}
764 	STAILQ_INIT(&spa->spa_vdevs);
765 	spa->spa_guid = guid;
766 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
767 
768 	return (spa);
769 }
770 
771 static const char *
772 state_name(vdev_state_t state)
773 {
774 	static const char* names[] = {
775 		"UNKNOWN",
776 		"CLOSED",
777 		"OFFLINE",
778 		"REMOVED",
779 		"CANT_OPEN",
780 		"FAULTED",
781 		"DEGRADED",
782 		"ONLINE"
783 	};
784 	return names[state];
785 }
786 
787 static int
788 pager_printf(const char *fmt, ...)
789 {
790 	char line[80];
791 	va_list args;
792 
793 	va_start(args, fmt);
794 	vsnprintf(line, sizeof (line), fmt, args);
795 	va_end(args);
796 	return (pager_output(line));
797 }
798 
799 #define STATUS_FORMAT	"        %s %s\n"
800 
801 static int
802 print_state(int indent, const char *name, vdev_state_t state)
803 {
804 	int i;
805 	char buf[512];
806 
807 	buf[0] = 0;
808 	for (i = 0; i < indent; i++)
809 		strcat(buf, "  ");
810 	strcat(buf, name);
811 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
812 }
813 
814 static int
815 vdev_status(vdev_t *vdev, int indent)
816 {
817 	vdev_t *kid;
818 	int ret;
819 	ret = print_state(indent, vdev->v_name, vdev->v_state);
820 	if (ret != 0)
821 		return (ret);
822 
823 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
824 		ret = vdev_status(kid, indent + 1);
825 		if (ret != 0)
826 			return (ret);
827 	}
828 	return (ret);
829 }
830 
831 static int
832 spa_status(spa_t *spa)
833 {
834 	static char bootfs[ZFS_MAXNAMELEN];
835 	uint64_t rootid;
836 	vdev_t *vdev;
837 	int good_kids, bad_kids, degraded_kids, ret;
838 	vdev_state_t state;
839 
840 	ret = pager_printf("  pool: %s\n", spa->spa_name);
841 	if (ret != 0)
842 		return (ret);
843 
844 	if (zfs_get_root(spa, &rootid) == 0 &&
845 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
846 		if (bootfs[0] == '\0')
847 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
848 		else
849 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
850 			    bootfs);
851 		if (ret != 0)
852 			return (ret);
853 	}
854 	ret = pager_printf("config:\n\n");
855 	if (ret != 0)
856 		return (ret);
857 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
858 	if (ret != 0)
859 		return (ret);
860 
861 	good_kids = 0;
862 	degraded_kids = 0;
863 	bad_kids = 0;
864 	STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
865 		if (vdev->v_state == VDEV_STATE_HEALTHY)
866 			good_kids++;
867 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
868 			degraded_kids++;
869 		else
870 			bad_kids++;
871 	}
872 
873 	state = VDEV_STATE_CLOSED;
874 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
875 		state = VDEV_STATE_HEALTHY;
876 	else if ((good_kids + degraded_kids) > 0)
877 		state = VDEV_STATE_DEGRADED;
878 
879 	ret = print_state(0, spa->spa_name, state);
880 	if (ret != 0)
881 		return (ret);
882 	STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
883 		ret = vdev_status(vdev, 1);
884 		if (ret != 0)
885 			return (ret);
886 	}
887 	return (ret);
888 }
889 
890 int
891 spa_all_status(void)
892 {
893 	spa_t *spa;
894 	int first = 1, ret = 0;
895 
896 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
897 		if (!first) {
898 			ret = pager_printf("\n");
899 			if (ret != 0)
900 				return (ret);
901 		}
902 		first = 0;
903 		ret = spa_status(spa);
904 		if (ret != 0)
905 			return (ret);
906 	}
907 	return (ret);
908 }
909 
910 uint64_t
911 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
912 {
913 	uint64_t label_offset;
914 
915 	if (l < VDEV_LABELS / 2)
916 		label_offset = 0;
917 	else
918 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
919 
920 	return (offset + l * sizeof (vdev_label_t) + label_offset);
921 }
922 
923 static int
924 vdev_probe(vdev_phys_read_t *phys_read, void *read_priv, spa_t **spap)
925 {
926 	vdev_t vtmp;
927 	vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
928 	vdev_phys_t *tmp_label;
929 	spa_t *spa;
930 	vdev_t *vdev, *top_vdev, *pool_vdev;
931 	off_t off;
932 	blkptr_t bp;
933 	const unsigned char *nvlist = NULL;
934 	uint64_t val;
935 	uint64_t guid;
936 	uint64_t best_txg = 0;
937 	uint64_t pool_txg, pool_guid;
938 	uint64_t psize;
939 	const char *pool_name;
940 	const unsigned char *vdevs;
941 	const unsigned char *features;
942 	int i, l, rc, is_newer;
943 	char *upbuf;
944 	const struct uberblock *up;
945 
946 	/*
947 	 * Load the vdev label and figure out which
948 	 * uberblock is most current.
949 	 */
950 	memset(&vtmp, 0, sizeof(vtmp));
951 	vtmp.v_phys_read = phys_read;
952 	vtmp.v_read_priv = read_priv;
953 	psize = P2ALIGN(ldi_get_size(read_priv),
954 	    (uint64_t)sizeof (vdev_label_t));
955 
956 	/* Test for minimum device size. */
957 	if (psize < SPA_MINDEVSIZE)
958 		return (EIO);
959 
960 	tmp_label = zfs_alloc(sizeof (vdev_phys_t));
961 
962 	for (l = 0; l < VDEV_LABELS; l++) {
963 		off = vdev_label_offset(psize, l,
964 		    offsetof(vdev_label_t, vl_vdev_phys));
965 
966 		BP_ZERO(&bp);
967 		BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
968 		BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
969 		BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
970 		BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
971 		DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
972 		ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
973 
974 		if (vdev_read_phys(&vtmp, &bp, tmp_label, off, 0))
975 			continue;
976 
977 		if (tmp_label->vp_nvlist[0] != NV_ENCODE_XDR)
978 			continue;
979 
980 		nvlist = (const unsigned char *) tmp_label->vp_nvlist + 4;
981 		if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
982 		    DATA_TYPE_UINT64, NULL, &pool_txg) != 0)
983 			continue;
984 
985 		if (best_txg <= pool_txg) {
986 			best_txg = pool_txg;
987 			memcpy(vdev_label, tmp_label, sizeof (vdev_phys_t));
988 		}
989 	}
990 
991 	zfs_free(tmp_label, sizeof (vdev_phys_t));
992 
993 	if (best_txg == 0)
994 		return (EIO);
995 
996 	if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR)
997 		return (EIO);
998 
999 	nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
1000 
1001 	if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1002 	    NULL, &val) != 0) {
1003 		return (EIO);
1004 	}
1005 
1006 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
1007 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1008 		    (unsigned) val, (unsigned) SPA_VERSION);
1009 		return (EIO);
1010 	}
1011 
1012 	/* Check ZFS features for read */
1013 	if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1014 	    DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1015 	    nvlist_check_features_for_read(features) != 0) {
1016 		return (EIO);
1017 	}
1018 
1019 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1020 	    NULL, &val) != 0) {
1021 		return (EIO);
1022 	}
1023 
1024 	if (val == POOL_STATE_DESTROYED) {
1025 		/* We don't boot only from destroyed pools. */
1026 		return (EIO);
1027 	}
1028 
1029 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1030 	    NULL, &pool_txg) != 0 ||
1031 	    nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1032 	    NULL, &pool_guid) != 0 ||
1033 	    nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1034 	    NULL, &pool_name) != 0) {
1035 		/*
1036 		 * Cache and spare devices end up here - just ignore
1037 		 * them.
1038 		 */
1039 		/*printf("ZFS: can't find pool details\n");*/
1040 		return (EIO);
1041 	}
1042 
1043 	if (nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64,
1044 	    NULL, &val) == 0 && val != 0) {
1045 		return (EIO);
1046 	}
1047 
1048 	/*
1049 	 * Create the pool if this is the first time we've seen it.
1050 	 */
1051 	spa = spa_find_by_guid(pool_guid);
1052 	if (spa == NULL) {
1053 		spa = spa_create(pool_guid, pool_name);
1054 		if (spa == NULL)
1055 			return (ENOMEM);
1056 	}
1057 	if (pool_txg > spa->spa_txg) {
1058 		spa->spa_txg = pool_txg;
1059 		is_newer = 1;
1060 	} else {
1061 		is_newer = 0;
1062 	}
1063 
1064 	/*
1065 	 * Get the vdev tree and create our in-core copy of it.
1066 	 * If we already have a vdev with this guid, this must
1067 	 * be some kind of alias (overlapping slices, dangerously dedicated
1068 	 * disks etc).
1069 	 */
1070 	if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1071 	    NULL, &guid) != 0) {
1072 		return (EIO);
1073 	}
1074 	vdev = vdev_find(guid);
1075 	if (vdev && vdev->v_phys_read)	/* Has this vdev already been inited? */
1076 		return (EIO);
1077 
1078 	if (nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1079 	    NULL, &vdevs)) {
1080 		return (EIO);
1081 	}
1082 
1083 	rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1084 	if (rc != 0)
1085 		return (rc);
1086 
1087 	/*
1088 	 * Add the toplevel vdev to the pool if its not already there.
1089 	 */
1090 	STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1091 		if (top_vdev == pool_vdev)
1092 			break;
1093 	if (!pool_vdev && top_vdev)
1094 		STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1095 
1096 	/*
1097 	 * We should already have created an incomplete vdev for this
1098 	 * vdev. Find it and initialise it with our read proc.
1099 	 */
1100 	vdev = vdev_find(guid);
1101 	if (vdev) {
1102 		vdev->v_phys_read = phys_read;
1103 		vdev->v_read_priv = read_priv;
1104 		vdev->v_state = VDEV_STATE_HEALTHY;
1105 	} else {
1106 		printf("ZFS: inconsistent nvlist contents\n");
1107 		return (EIO);
1108 	}
1109 
1110 	/*
1111 	 * Re-evaluate top-level vdev state.
1112 	 */
1113 	vdev_set_state(top_vdev);
1114 
1115 	/*
1116 	 * Ok, we are happy with the pool so far. Lets find
1117 	 * the best uberblock and then we can actually access
1118 	 * the contents of the pool.
1119 	 */
1120 	upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
1121 	up = (const struct uberblock *)upbuf;
1122 	for (l = 0; l < VDEV_LABELS; l++) {
1123 		for (i = 0; i < VDEV_UBERBLOCK_COUNT(vdev); i++) {
1124 			off = vdev_label_offset(psize, l,
1125 			    VDEV_UBERBLOCK_OFFSET(vdev, i));
1126 			BP_ZERO(&bp);
1127 			DVA_SET_OFFSET(&bp.blk_dva[0], off);
1128 			BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1129 			BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1130 			BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1131 			BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1132 			ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1133 
1134 			if (vdev_read_phys(vdev, &bp, upbuf, off, 0) != 0)
1135 				continue;
1136 
1137 			if (up->ub_magic != UBERBLOCK_MAGIC)
1138 				continue;
1139 			if (up->ub_txg < spa->spa_txg)
1140 				continue;
1141 			if (up->ub_txg > spa->spa_uberblock.ub_txg ||
1142 			    (up->ub_txg == spa->spa_uberblock.ub_txg &&
1143 			    up->ub_timestamp >
1144 			    spa->spa_uberblock.ub_timestamp)) {
1145 				spa->spa_uberblock = *up;
1146 			}
1147 		}
1148 	}
1149 	zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
1150 
1151 	if (spap != NULL)
1152 		*spap = spa;
1153 	return (0);
1154 }
1155 
1156 static int
1157 ilog2(int n)
1158 {
1159 	int v;
1160 
1161 	for (v = 0; v < 32; v++)
1162 		if (n == (1 << v))
1163 			return v;
1164 	return -1;
1165 }
1166 
1167 static int
1168 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1169 {
1170 	blkptr_t gbh_bp;
1171 	zio_gbh_phys_t zio_gb;
1172 	char *pbuf;
1173 	int i;
1174 
1175 	/* Artificial BP for gang block header. */
1176 	gbh_bp = *bp;
1177 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1178 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1179 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1180 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1181 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
1182 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1183 
1184 	/* Read gang header block using the artificial BP. */
1185 	if (zio_read(spa, &gbh_bp, &zio_gb))
1186 		return (EIO);
1187 
1188 	pbuf = buf;
1189 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1190 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1191 
1192 		if (BP_IS_HOLE(gbp))
1193 			continue;
1194 		if (zio_read(spa, gbp, pbuf))
1195 			return (EIO);
1196 		pbuf += BP_GET_PSIZE(gbp);
1197 	}
1198 
1199 	if (zio_checksum_verify(bp, buf))
1200 		return (EIO);
1201 	return (0);
1202 }
1203 
1204 static int
1205 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1206 {
1207 	int cpfunc = BP_GET_COMPRESS(bp);
1208 	uint64_t align, size;
1209 	void *pbuf;
1210 	int i, error;
1211 
1212 	/*
1213 	 * Process data embedded in block pointer
1214 	 */
1215 	if (BP_IS_EMBEDDED(bp)) {
1216 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1217 
1218 		size = BPE_GET_PSIZE(bp);
1219 		ASSERT(size <= BPE_PAYLOAD_SIZE);
1220 
1221 		if (cpfunc != ZIO_COMPRESS_OFF)
1222 			pbuf = zfs_alloc(size);
1223 		else
1224 			pbuf = buf;
1225 
1226 		decode_embedded_bp_compressed(bp, pbuf);
1227 		error = 0;
1228 
1229 		if (cpfunc != ZIO_COMPRESS_OFF) {
1230 			error = zio_decompress_data(cpfunc, pbuf,
1231 			    size, buf, BP_GET_LSIZE(bp));
1232 			zfs_free(pbuf, size);
1233 		}
1234 		if (error != 0)
1235 			printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
1236 			    error);
1237 		return (error);
1238 	}
1239 
1240 	error = EIO;
1241 
1242 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1243 		const dva_t *dva = &bp->blk_dva[i];
1244 		vdev_t *vdev;
1245 		int vdevid;
1246 		off_t offset;
1247 
1248 		if (!dva->dva_word[0] && !dva->dva_word[1])
1249 			continue;
1250 
1251 		vdevid = DVA_GET_VDEV(dva);
1252 		offset = DVA_GET_OFFSET(dva);
1253 		STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1254 			if (vdev->v_id == vdevid)
1255 				break;
1256 		}
1257 		if (!vdev || !vdev->v_read)
1258 			continue;
1259 
1260 		size = BP_GET_PSIZE(bp);
1261 		if (vdev->v_read == vdev_raidz_read) {
1262 			align = 1ULL << vdev->v_top->v_ashift;
1263 			if (P2PHASE(size, align) != 0)
1264 				size = P2ROUNDUP(size, align);
1265 		}
1266 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1267 			pbuf = zfs_alloc(size);
1268 		else
1269 			pbuf = buf;
1270 
1271 		if (DVA_GET_GANG(dva))
1272 			error = zio_read_gang(spa, bp, pbuf);
1273 		else
1274 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
1275 		if (error == 0) {
1276 			if (cpfunc != ZIO_COMPRESS_OFF)
1277 				error = zio_decompress_data(cpfunc, pbuf,
1278 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1279 			else if (size != BP_GET_PSIZE(bp))
1280 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1281 		}
1282 		if (buf != pbuf)
1283 			zfs_free(pbuf, size);
1284 		if (error == 0)
1285 			break;
1286 	}
1287 	if (error != 0)
1288 		printf("ZFS: i/o error - all block copies unavailable\n");
1289 	return (error);
1290 }
1291 
1292 static int
1293 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1294 {
1295 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1296 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1297 	int nlevels = dnode->dn_nlevels;
1298 	int i, rc;
1299 
1300 	if (bsize > SPA_MAXBLOCKSIZE) {
1301 		printf("ZFS: I/O error - blocks larger than %llu are not "
1302 		    "supported\n", SPA_MAXBLOCKSIZE);
1303 		return (EIO);
1304 	}
1305 
1306 	/*
1307 	 * Note: bsize may not be a power of two here so we need to do an
1308 	 * actual divide rather than a bitshift.
1309 	 */
1310 	while (buflen > 0) {
1311 		uint64_t bn = offset / bsize;
1312 		int boff = offset % bsize;
1313 		int ibn;
1314 		const blkptr_t *indbp;
1315 		blkptr_t bp;
1316 
1317 		if (bn > dnode->dn_maxblkid) {
1318 			printf("warning: zfs bug: bn %llx > dn_maxblkid %llx\n",
1319 			    (unsigned long long)bn,
1320 			    (unsigned long long)dnode->dn_maxblkid);
1321 			/*
1322 			 * zfs bug, will not return error
1323 			 * return (EIO);
1324 			 */
1325 		}
1326 
1327 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1328 			goto cached;
1329 
1330 		indbp = dnode->dn_blkptr;
1331 		for (i = 0; i < nlevels; i++) {
1332 			/*
1333 			 * Copy the bp from the indirect array so that
1334 			 * we can re-use the scratch buffer for multi-level
1335 			 * objects.
1336 			 */
1337 			ibn = bn >> ((nlevels - i - 1) * ibshift);
1338 			ibn &= ((1 << ibshift) - 1);
1339 			bp = indbp[ibn];
1340 			if (BP_IS_HOLE(&bp)) {
1341 				memset(dnode_cache_buf, 0, bsize);
1342 				break;
1343 			}
1344 			rc = zio_read(spa, &bp, dnode_cache_buf);
1345 			if (rc)
1346 				return (rc);
1347 			indbp = (const blkptr_t *) dnode_cache_buf;
1348 		}
1349 		dnode_cache_obj = dnode;
1350 		dnode_cache_bn = bn;
1351 	cached:
1352 
1353 		/*
1354 		 * The buffer contains our data block. Copy what we
1355 		 * need from it and loop.
1356 		 */
1357 		i = bsize - boff;
1358 		if (i > buflen) i = buflen;
1359 		memcpy(buf, &dnode_cache_buf[boff], i);
1360 		buf = ((char*) buf) + i;
1361 		offset += i;
1362 		buflen -= i;
1363 	}
1364 
1365 	return (0);
1366 }
1367 
1368 /*
1369  * Lookup a value in a microzap directory. Assumes that the zap
1370  * scratch buffer contains the directory contents.
1371  */
1372 static int
1373 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1374 {
1375 	const mzap_phys_t *mz;
1376 	const mzap_ent_phys_t *mze;
1377 	size_t size;
1378 	int chunks, i;
1379 
1380 	/*
1381 	 * Microzap objects use exactly one block. Read the whole
1382 	 * thing.
1383 	 */
1384 	size = dnode->dn_datablkszsec * 512;
1385 
1386 	mz = (const mzap_phys_t *) zap_scratch;
1387 	chunks = size / MZAP_ENT_LEN - 1;
1388 
1389 	for (i = 0; i < chunks; i++) {
1390 		mze = &mz->mz_chunk[i];
1391 		if (!strcmp(mze->mze_name, name)) {
1392 			*value = mze->mze_value;
1393 			return (0);
1394 		}
1395 	}
1396 
1397 	return (ENOENT);
1398 }
1399 
1400 /*
1401  * Compare a name with a zap leaf entry. Return non-zero if the name
1402  * matches.
1403  */
1404 static int
1405 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1406 {
1407 	size_t namelen;
1408 	const zap_leaf_chunk_t *nc;
1409 	const char *p;
1410 
1411 	namelen = zc->l_entry.le_name_numints;
1412 
1413 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1414 	p = name;
1415 	while (namelen > 0) {
1416 		size_t len;
1417 		len = namelen;
1418 		if (len > ZAP_LEAF_ARRAY_BYTES)
1419 			len = ZAP_LEAF_ARRAY_BYTES;
1420 		if (memcmp(p, nc->l_array.la_array, len))
1421 			return (0);
1422 		p += len;
1423 		namelen -= len;
1424 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1425 	}
1426 
1427 	return 1;
1428 }
1429 
1430 /*
1431  * Extract a uint64_t value from a zap leaf entry.
1432  */
1433 static uint64_t
1434 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1435 {
1436 	const zap_leaf_chunk_t *vc;
1437 	int i;
1438 	uint64_t value;
1439 	const uint8_t *p;
1440 
1441 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1442 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1443 		value = (value << 8) | p[i];
1444 	}
1445 
1446 	return value;
1447 }
1448 
1449 /*
1450  * Lookup a value in a fatzap directory. Assumes that the zap scratch
1451  * buffer contains the directory header.
1452  */
1453 static int
1454 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1455 {
1456 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1457 	zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1458 	fat_zap_t z;
1459 	uint64_t *ptrtbl;
1460 	uint64_t hash;
1461 	int rc;
1462 
1463 	if (zh.zap_magic != ZAP_MAGIC)
1464 		return (EIO);
1465 
1466 	z.zap_block_shift = ilog2(bsize);
1467 	z.zap_phys = (zap_phys_t *) zap_scratch;
1468 
1469 	/*
1470 	 * Figure out where the pointer table is and read it in if necessary.
1471 	 */
1472 	if (zh.zap_ptrtbl.zt_blk) {
1473 		rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1474 			       zap_scratch, bsize);
1475 		if (rc)
1476 			return (rc);
1477 		ptrtbl = (uint64_t *) zap_scratch;
1478 	} else {
1479 		ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1480 	}
1481 
1482 	hash = zap_hash(zh.zap_salt, name);
1483 
1484 	zap_leaf_t zl;
1485 	zl.l_bs = z.zap_block_shift;
1486 
1487 	off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1488 	zap_leaf_chunk_t *zc;
1489 
1490 	rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1491 	if (rc)
1492 		return (rc);
1493 
1494 	zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1495 
1496 	/*
1497 	 * Make sure this chunk matches our hash.
1498 	 */
1499 	if (zl.l_phys->l_hdr.lh_prefix_len > 0
1500 	    && zl.l_phys->l_hdr.lh_prefix
1501 	    != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1502 		return (ENOENT);
1503 
1504 	/*
1505 	 * Hash within the chunk to find our entry.
1506 	 */
1507 	int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1508 	int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1509 	h = zl.l_phys->l_hash[h];
1510 	if (h == 0xffff)
1511 		return (ENOENT);
1512 	zc = &ZAP_LEAF_CHUNK(&zl, h);
1513 	while (zc->l_entry.le_hash != hash) {
1514 		if (zc->l_entry.le_next == 0xffff) {
1515 			zc = 0;
1516 			break;
1517 		}
1518 		zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1519 	}
1520 	if (fzap_name_equal(&zl, zc, name)) {
1521 		if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints > 8)
1522 			return (E2BIG);
1523 		*value = fzap_leaf_value(&zl, zc);
1524 		return (0);
1525 	}
1526 
1527 	return (ENOENT);
1528 }
1529 
1530 /*
1531  * Lookup a name in a zap object and return its value as a uint64_t.
1532  */
1533 static int
1534 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1535 {
1536 	int rc;
1537 	uint64_t zap_type;
1538 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1539 
1540 	rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1541 	if (rc)
1542 		return (rc);
1543 
1544 	zap_type = *(uint64_t *) zap_scratch;
1545 	if (zap_type == ZBT_MICRO)
1546 		return mzap_lookup(dnode, name, value);
1547 	else if (zap_type == ZBT_HEADER)
1548 		return fzap_lookup(spa, dnode, name, value);
1549 	printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
1550 	return (EIO);
1551 }
1552 
1553 /*
1554  * List a microzap directory. Assumes that the zap scratch buffer contains
1555  * the directory contents.
1556  */
1557 static int
1558 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1559 {
1560 	const mzap_phys_t *mz;
1561 	const mzap_ent_phys_t *mze;
1562 	size_t size;
1563 	int chunks, i, rc;
1564 
1565 	/*
1566 	 * Microzap objects use exactly one block. Read the whole
1567 	 * thing.
1568 	 */
1569 	size = dnode->dn_datablkszsec * 512;
1570 	mz = (const mzap_phys_t *) zap_scratch;
1571 	chunks = size / MZAP_ENT_LEN - 1;
1572 
1573 	for (i = 0; i < chunks; i++) {
1574 		mze = &mz->mz_chunk[i];
1575 		if (mze->mze_name[0]) {
1576 			rc = callback(mze->mze_name, mze->mze_value);
1577 			if (rc != 0)
1578 				return (rc);
1579 		}
1580 	}
1581 
1582 	return (0);
1583 }
1584 
1585 /*
1586  * List a fatzap directory. Assumes that the zap scratch buffer contains
1587  * the directory header.
1588  */
1589 static int
1590 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1591 {
1592 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1593 	zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1594 	fat_zap_t z;
1595 	int i, j, rc;
1596 
1597 	if (zh.zap_magic != ZAP_MAGIC)
1598 		return (EIO);
1599 
1600 	z.zap_block_shift = ilog2(bsize);
1601 	z.zap_phys = (zap_phys_t *) zap_scratch;
1602 
1603 	/*
1604 	 * This assumes that the leaf blocks start at block 1. The
1605 	 * documentation isn't exactly clear on this.
1606 	 */
1607 	zap_leaf_t zl;
1608 	zl.l_bs = z.zap_block_shift;
1609 	for (i = 0; i < zh.zap_num_leafs; i++) {
1610 		off_t off = (i + 1) << zl.l_bs;
1611 		char name[256], *p;
1612 		uint64_t value;
1613 
1614 		if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1615 			return (EIO);
1616 
1617 		zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1618 
1619 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1620 			zap_leaf_chunk_t *zc, *nc;
1621 			int namelen;
1622 
1623 			zc = &ZAP_LEAF_CHUNK(&zl, j);
1624 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1625 				continue;
1626 			namelen = zc->l_entry.le_name_numints;
1627 			if (namelen > sizeof(name))
1628 				namelen = sizeof(name);
1629 
1630 			/*
1631 			 * Paste the name back together.
1632 			 */
1633 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1634 			p = name;
1635 			while (namelen > 0) {
1636 				int len;
1637 				len = namelen;
1638 				if (len > ZAP_LEAF_ARRAY_BYTES)
1639 					len = ZAP_LEAF_ARRAY_BYTES;
1640 				memcpy(p, nc->l_array.la_array, len);
1641 				p += len;
1642 				namelen -= len;
1643 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1644 			}
1645 
1646 			/*
1647 			 * Assume the first eight bytes of the value are
1648 			 * a uint64_t.
1649 			 */
1650 			value = fzap_leaf_value(&zl, zc);
1651 
1652 			//printf("%s 0x%jx\n", name, (uintmax_t)value);
1653 			rc = callback((const char *)name, value);
1654 			if (rc != 0)
1655 				return (rc);
1656 		}
1657 	}
1658 
1659 	return (0);
1660 }
1661 
1662 static int zfs_printf(const char *name, uint64_t value __unused)
1663 {
1664 
1665 	printf("%s\n", name);
1666 
1667 	return (0);
1668 }
1669 
1670 /*
1671  * List a zap directory.
1672  */
1673 static int
1674 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
1675 {
1676 	uint64_t zap_type;
1677 	size_t size = dnode->dn_datablkszsec * 512;
1678 
1679 	if (dnode_read(spa, dnode, 0, zap_scratch, size))
1680 		return (EIO);
1681 
1682 	zap_type = *(uint64_t *) zap_scratch;
1683 	if (zap_type == ZBT_MICRO)
1684 		return mzap_list(dnode, zfs_printf);
1685 	else
1686 		return fzap_list(spa, dnode, zfs_printf);
1687 }
1688 
1689 static int
1690 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1691 {
1692 	off_t offset;
1693 
1694 	offset = objnum * sizeof(dnode_phys_t);
1695 	return dnode_read(spa, &os->os_meta_dnode, offset,
1696 		dnode, sizeof(dnode_phys_t));
1697 }
1698 
1699 static int
1700 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1701 {
1702 	const mzap_phys_t *mz;
1703 	const mzap_ent_phys_t *mze;
1704 	size_t size;
1705 	int chunks, i;
1706 
1707 	/*
1708 	 * Microzap objects use exactly one block. Read the whole
1709 	 * thing.
1710 	 */
1711 	size = dnode->dn_datablkszsec * 512;
1712 
1713 	mz = (const mzap_phys_t *) zap_scratch;
1714 	chunks = size / MZAP_ENT_LEN - 1;
1715 
1716 	for (i = 0; i < chunks; i++) {
1717 		mze = &mz->mz_chunk[i];
1718 		if (value == mze->mze_value) {
1719 			strcpy(name, mze->mze_name);
1720 			return (0);
1721 		}
1722 	}
1723 
1724 	return (ENOENT);
1725 }
1726 
1727 static void
1728 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
1729 {
1730 	size_t namelen;
1731 	const zap_leaf_chunk_t *nc;
1732 	char *p;
1733 
1734 	namelen = zc->l_entry.le_name_numints;
1735 
1736 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1737 	p = name;
1738 	while (namelen > 0) {
1739 		size_t len;
1740 		len = namelen;
1741 		if (len > ZAP_LEAF_ARRAY_BYTES)
1742 			len = ZAP_LEAF_ARRAY_BYTES;
1743 		memcpy(p, nc->l_array.la_array, len);
1744 		p += len;
1745 		namelen -= len;
1746 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1747 	}
1748 
1749 	*p = '\0';
1750 }
1751 
1752 static int
1753 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1754 {
1755 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1756 	zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1757 	fat_zap_t z;
1758 	int i, j;
1759 
1760 	if (zh.zap_magic != ZAP_MAGIC)
1761 		return (EIO);
1762 
1763 	z.zap_block_shift = ilog2(bsize);
1764 	z.zap_phys = (zap_phys_t *) zap_scratch;
1765 
1766 	/*
1767 	 * This assumes that the leaf blocks start at block 1. The
1768 	 * documentation isn't exactly clear on this.
1769 	 */
1770 	zap_leaf_t zl;
1771 	zl.l_bs = z.zap_block_shift;
1772 	for (i = 0; i < zh.zap_num_leafs; i++) {
1773 		off_t off = (i + 1) << zl.l_bs;
1774 
1775 		if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1776 			return (EIO);
1777 
1778 		zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1779 
1780 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1781 			zap_leaf_chunk_t *zc;
1782 
1783 			zc = &ZAP_LEAF_CHUNK(&zl, j);
1784 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1785 				continue;
1786 			if (zc->l_entry.le_value_intlen != 8 ||
1787 			    zc->l_entry.le_value_numints != 1)
1788 				continue;
1789 
1790 			if (fzap_leaf_value(&zl, zc) == value) {
1791 				fzap_name_copy(&zl, zc, name);
1792 				return (0);
1793 			}
1794 		}
1795 	}
1796 
1797 	return (ENOENT);
1798 }
1799 
1800 static int
1801 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1802 {
1803 	int rc;
1804 	uint64_t zap_type;
1805 	size_t size = dnode->dn_datablkszsec * 512;
1806 
1807 	rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1808 	if (rc)
1809 		return (rc);
1810 
1811 	zap_type = *(uint64_t *) zap_scratch;
1812 	if (zap_type == ZBT_MICRO)
1813 		return mzap_rlookup(spa, dnode, name, value);
1814 	else
1815 		return fzap_rlookup(spa, dnode, name, value);
1816 }
1817 
1818 static int
1819 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
1820 {
1821 	char name[256];
1822 	char component[256];
1823 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
1824 	dnode_phys_t child_dir_zap, dataset, dir, parent;
1825 	dsl_dir_phys_t *dd;
1826 	dsl_dataset_phys_t *ds;
1827 	char *p;
1828 	int len;
1829 
1830 	p = &name[sizeof(name) - 1];
1831 	*p = '\0';
1832 
1833 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1834 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1835 		return (EIO);
1836 	}
1837 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
1838 	dir_obj = ds->ds_dir_obj;
1839 
1840 	for (;;) {
1841 		if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
1842 			return (EIO);
1843 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1844 
1845 		/* Actual loop condition. */
1846 		parent_obj  = dd->dd_parent_obj;
1847 		if (parent_obj == 0)
1848 			break;
1849 
1850 		if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
1851 			return (EIO);
1852 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
1853 		child_dir_zapobj = dd->dd_child_dir_zapobj;
1854 		if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1855 			return (EIO);
1856 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
1857 			return (EIO);
1858 
1859 		len = strlen(component);
1860 		p -= len;
1861 		memcpy(p, component, len);
1862 		--p;
1863 		*p = '/';
1864 
1865 		/* Actual loop iteration. */
1866 		dir_obj = parent_obj;
1867 	}
1868 
1869 	if (*p != '\0')
1870 		++p;
1871 	strcpy(result, p);
1872 
1873 	return (0);
1874 }
1875 
1876 static int
1877 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
1878 {
1879 	char element[256];
1880 	uint64_t dir_obj, child_dir_zapobj;
1881 	dnode_phys_t child_dir_zap, dir;
1882 	dsl_dir_phys_t *dd;
1883 	const char *p, *q;
1884 
1885 	if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
1886 		return (EIO);
1887 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &dir_obj))
1888 		return (EIO);
1889 
1890 	p = name;
1891 	for (;;) {
1892 		if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
1893 			return (EIO);
1894 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1895 
1896 		while (*p == '/')
1897 			p++;
1898 		/* Actual loop condition #1. */
1899 		if (*p == '\0')
1900 			break;
1901 
1902 		q = strchr(p, '/');
1903 		if (q) {
1904 			memcpy(element, p, q - p);
1905 			element[q - p] = '\0';
1906 			p = q + 1;
1907 		} else {
1908 			strcpy(element, p);
1909 			p += strlen(p);
1910 		}
1911 
1912 		child_dir_zapobj = dd->dd_child_dir_zapobj;
1913 		if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1914 			return (EIO);
1915 
1916 		/* Actual loop condition #2. */
1917 		if (zap_lookup(spa, &child_dir_zap, element, &dir_obj) != 0)
1918 			return (ENOENT);
1919 	}
1920 
1921 	*objnum = dd->dd_head_dataset_obj;
1922 	return (0);
1923 }
1924 
1925 #pragma GCC diagnostic ignored "-Wstrict-aliasing"
1926 static int
1927 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
1928 {
1929 	uint64_t dir_obj, child_dir_zapobj;
1930 	dnode_phys_t child_dir_zap, dir, dataset;
1931 	dsl_dataset_phys_t *ds;
1932 	dsl_dir_phys_t *dd;
1933 
1934 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1935 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1936 		return (EIO);
1937 	}
1938 	ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1939 	dir_obj = ds->ds_dir_obj;
1940 
1941 	if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
1942 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
1943 		return (EIO);
1944 	}
1945 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1946 
1947 	child_dir_zapobj = dd->dd_child_dir_zapobj;
1948 	if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
1949 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
1950 		return (EIO);
1951 	}
1952 
1953 	return (zap_list(spa, &child_dir_zap) != 0);
1954 }
1955 
1956 int
1957 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *, uint64_t))
1958 {
1959 	uint64_t dir_obj, child_dir_zapobj, zap_type;
1960 	dnode_phys_t child_dir_zap, dir, dataset;
1961 	dsl_dataset_phys_t *ds;
1962 	dsl_dir_phys_t *dd;
1963 	int err;
1964 
1965 	err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
1966 	if (err != 0) {
1967 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1968 		return (err);
1969 	}
1970 	ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1971 	dir_obj = ds->ds_dir_obj;
1972 
1973 	err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
1974 	if (err != 0) {
1975 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
1976 		return (err);
1977 	}
1978 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1979 
1980 	child_dir_zapobj = dd->dd_child_dir_zapobj;
1981 	err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
1982 	if (err != 0) {
1983 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
1984 		return (err);
1985 	}
1986 
1987 	err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
1988 	if (err != 0)
1989 		return (err);
1990 
1991 	zap_type = *(uint64_t *) zap_scratch;
1992 	if (zap_type == ZBT_MICRO)
1993 		return mzap_list(&child_dir_zap, callback);
1994 	else
1995 		return fzap_list(spa, &child_dir_zap, callback);
1996 }
1997 
1998 /*
1999  * Find the object set given the object number of its dataset object
2000  * and return its details in *objset
2001  */
2002 static int
2003 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2004 {
2005 	dnode_phys_t dataset;
2006 	dsl_dataset_phys_t *ds;
2007 
2008 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2009 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2010 		return (EIO);
2011 	}
2012 
2013 	ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2014 	if (zio_read(spa, &ds->ds_bp, objset)) {
2015 		printf("ZFS: can't read object set for dataset %ju\n",
2016 		    (uintmax_t)objnum);
2017 		return (EIO);
2018 	}
2019 
2020 	return (0);
2021 }
2022 
2023 /*
2024  * Find the object set pointed to by the BOOTFS property or the root
2025  * dataset if there is none and return its details in *objset
2026  */
2027 static int
2028 zfs_get_root(const spa_t *spa, uint64_t *objid)
2029 {
2030 	dnode_phys_t dir, propdir;
2031 	uint64_t props, bootfs, root;
2032 
2033 	*objid = 0;
2034 
2035 	/*
2036 	 * Start with the MOS directory object.
2037 	 */
2038 	if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
2039 		printf("ZFS: can't read MOS object directory\n");
2040 		return (EIO);
2041 	}
2042 
2043 	/*
2044 	 * Lookup the pool_props and see if we can find a bootfs.
2045 	 */
2046 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS, &props) == 0
2047 	     && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
2048 	     && zap_lookup(spa, &propdir, "bootfs", &bootfs) == 0
2049 	     && bootfs != 0)
2050 	{
2051 		*objid = bootfs;
2052 		return (0);
2053 	}
2054 	/*
2055 	 * Lookup the root dataset directory
2056 	 */
2057 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &root)
2058 	    || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
2059 		printf("ZFS: can't find root dsl_dir\n");
2060 		return (EIO);
2061 	}
2062 
2063 	/*
2064 	 * Use the information from the dataset directory's bonus buffer
2065 	 * to find the dataset object and from that the object set itself.
2066 	 */
2067 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
2068 	*objid = dd->dd_head_dataset_obj;
2069 	return (0);
2070 }
2071 
2072 static int
2073 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mnt)
2074 {
2075 
2076 	mnt->spa = spa;
2077 
2078 	/*
2079 	 * Find the root object set if not explicitly provided
2080 	 */
2081 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
2082 		printf("ZFS: can't find root filesystem\n");
2083 		return (EIO);
2084 	}
2085 
2086 	if (zfs_mount_dataset(spa, rootobj, &mnt->objset)) {
2087 		printf("ZFS: can't open root filesystem\n");
2088 		return (EIO);
2089 	}
2090 
2091 	mnt->rootobj = rootobj;
2092 
2093 	return (0);
2094 }
2095 
2096 /*
2097  * callback function for feature name checks.
2098  */
2099 static int
2100 check_feature(const char *name, uint64_t value)
2101 {
2102 	int i;
2103 
2104 	if (value == 0)
2105 		return (0);
2106 	if (name[0] == '\0')
2107 		return (0);
2108 
2109 	for (i = 0; features_for_read[i] != NULL; i++) {
2110 		if (strcmp(name, features_for_read[i]) == 0)
2111 			return (0);
2112 	}
2113 	printf("ZFS: unsupported feature: %s\n", name);
2114 	return (EIO);
2115 }
2116 
2117 /*
2118  * Checks whether the MOS features that are active are supported.
2119  */
2120 static int
2121 check_mos_features(const spa_t *spa)
2122 {
2123 	dnode_phys_t dir;
2124 	uint64_t objnum, zap_type;
2125 	size_t size;
2126 	int rc;
2127 
2128 	if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
2129 	    &dir)) != 0)
2130 		return (rc);
2131 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
2132 	    &objnum)) != 0) {
2133 		/*
2134 		 * It is older pool without features. As we have already
2135 		 * tested the label, just return without raising the error.
2136 		 */
2137 		if (rc == ENOENT)
2138 			rc = 0;
2139 		return (rc);
2140 	}
2141 
2142 	if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
2143 		return (rc);
2144 
2145 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
2146 		return (EIO);
2147 
2148 	size = dir.dn_datablkszsec * 512;
2149 	if (dnode_read(spa, &dir, 0, zap_scratch, size))
2150 		return (EIO);
2151 
2152 	zap_type = *(uint64_t *) zap_scratch;
2153 	if (zap_type == ZBT_MICRO)
2154 		rc = mzap_list(&dir, check_feature);
2155 	else
2156 		rc = fzap_list(spa, &dir, check_feature);
2157 
2158 	return (rc);
2159 }
2160 
2161 static int
2162 zfs_spa_init(spa_t *spa)
2163 {
2164 	int rc;
2165 
2166 	if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
2167 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
2168 		return (EIO);
2169 	}
2170 	if (spa->spa_mos.os_type != DMU_OST_META) {
2171 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
2172 		return (EIO);
2173 	}
2174 
2175 	rc = check_mos_features(spa);
2176 	if (rc != 0) {
2177 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
2178 	}
2179 
2180 	return (rc);
2181 }
2182 
2183 static int
2184 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
2185 {
2186 
2187 	if (dn->dn_bonustype != DMU_OT_SA) {
2188 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
2189 
2190 		sb->st_mode = zp->zp_mode;
2191 		sb->st_uid = zp->zp_uid;
2192 		sb->st_gid = zp->zp_gid;
2193 		sb->st_size = zp->zp_size;
2194 	} else {
2195 		sa_hdr_phys_t *sahdrp;
2196 		int hdrsize;
2197 		size_t size = 0;
2198 		void *buf = NULL;
2199 
2200 		if (dn->dn_bonuslen != 0)
2201 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2202 		else {
2203 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
2204 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
2205 				int error;
2206 
2207 				size = BP_GET_LSIZE(bp);
2208 				buf = zfs_alloc(size);
2209 				error = zio_read(spa, bp, buf);
2210 				if (error != 0) {
2211 					zfs_free(buf, size);
2212 					return (error);
2213 				}
2214 				sahdrp = buf;
2215 			} else {
2216 				return (EIO);
2217 			}
2218 		}
2219 		hdrsize = SA_HDR_SIZE(sahdrp);
2220 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
2221 		    SA_MODE_OFFSET);
2222 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
2223 		    SA_UID_OFFSET);
2224 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
2225 		    SA_GID_OFFSET);
2226 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
2227 		    SA_SIZE_OFFSET);
2228 		if (buf != NULL)
2229 			zfs_free(buf, size);
2230 	}
2231 
2232 	return (0);
2233 }
2234 
2235 static int
2236 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
2237 {
2238 	int rc = 0;
2239 
2240 	if (dn->dn_bonustype == DMU_OT_SA) {
2241 		sa_hdr_phys_t *sahdrp = NULL;
2242 		size_t size = 0;
2243 		void *buf = NULL;
2244 		int hdrsize;
2245 		char *p;
2246 
2247 		if (dn->dn_bonuslen != 0)
2248 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2249 		else {
2250 			blkptr_t *bp;
2251 
2252 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
2253 				return (EIO);
2254 			bp = DN_SPILL_BLKPTR(dn);
2255 
2256 			size = BP_GET_LSIZE(bp);
2257 			buf = zfs_alloc(size);
2258 			rc = zio_read(spa, bp, buf);
2259 			if (rc != 0) {
2260 				zfs_free(buf, size);
2261 				return (rc);
2262 			}
2263 			sahdrp = buf;
2264 		}
2265 		hdrsize = SA_HDR_SIZE(sahdrp);
2266 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
2267 		memcpy(path, p, psize);
2268 		if (buf != NULL)
2269 			zfs_free(buf, size);
2270 		return (0);
2271 	}
2272 	/*
2273 	 * Second test is purely to silence bogus compiler
2274 	 * warning about accessing past the end of dn_bonus.
2275 	 */
2276 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
2277 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
2278 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
2279 	} else {
2280 		rc = dnode_read(spa, dn, 0, path, psize);
2281 	}
2282 	return (rc);
2283 }
2284 
2285 struct obj_list {
2286 	uint64_t		objnum;
2287 	STAILQ_ENTRY(obj_list)	entry;
2288 };
2289 
2290 /*
2291  * Lookup a file and return its dnode.
2292  */
2293 static int
2294 zfs_lookup(const struct zfsmount *mnt, const char *upath, dnode_phys_t *dnode)
2295 {
2296 	int rc;
2297 	uint64_t objnum;
2298 	const spa_t *spa;
2299 	dnode_phys_t dn;
2300 	const char *p, *q;
2301 	char element[256];
2302 	char path[1024];
2303 	int symlinks_followed = 0;
2304 	struct stat sb;
2305 	struct obj_list *entry, *tentry;
2306 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
2307 
2308 	spa = mnt->spa;
2309 	if (mnt->objset.os_type != DMU_OST_ZFS) {
2310 		printf("ZFS: unexpected object set type %ju\n",
2311 		    (uintmax_t)mnt->objset.os_type);
2312 		return (EIO);
2313 	}
2314 
2315 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
2316 		return (ENOMEM);
2317 
2318 	/*
2319 	 * Get the root directory dnode.
2320 	 */
2321 	rc = objset_get_dnode(spa, &mnt->objset, MASTER_NODE_OBJ, &dn);
2322 	if (rc) {
2323 		free(entry);
2324 		return (rc);
2325 	}
2326 
2327 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, &objnum);
2328 	if (rc) {
2329 		free(entry);
2330 		return (rc);
2331 	}
2332 	entry->objnum = objnum;
2333 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2334 
2335 	rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2336 	if (rc != 0)
2337 		goto done;
2338 
2339 	p = upath;
2340 	while (p && *p) {
2341 		rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2342 		if (rc != 0)
2343 			goto done;
2344 
2345 		while (*p == '/')
2346 			p++;
2347 		if (*p == '\0')
2348 			break;
2349 		q = p;
2350 		while (*q != '\0' && *q != '/')
2351 			q++;
2352 
2353 		/* skip dot */
2354 		if (p + 1 == q && p[0] == '.') {
2355 			p++;
2356 			continue;
2357 		}
2358 		/* double dot */
2359 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
2360 			p += 2;
2361 			if (STAILQ_FIRST(&on_cache) ==
2362 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
2363 				rc = ENOENT;
2364 				goto done;
2365 			}
2366 			entry = STAILQ_FIRST(&on_cache);
2367 			STAILQ_REMOVE_HEAD(&on_cache, entry);
2368 			free(entry);
2369 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
2370 			continue;
2371 		}
2372 		if (q - p + 1 > sizeof(element)) {
2373 			rc = ENAMETOOLONG;
2374 			goto done;
2375 		}
2376 		memcpy(element, p, q - p);
2377 		element[q - p] = 0;
2378 		p = q;
2379 
2380 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
2381 			goto done;
2382 		if (!S_ISDIR(sb.st_mode)) {
2383 			rc = ENOTDIR;
2384 			goto done;
2385 		}
2386 
2387 		rc = zap_lookup(spa, &dn, element, &objnum);
2388 		if (rc)
2389 			goto done;
2390 		objnum = ZFS_DIRENT_OBJ(objnum);
2391 
2392 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
2393 			rc = ENOMEM;
2394 			goto done;
2395 		}
2396 		entry->objnum = objnum;
2397 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2398 		rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2399 		if (rc)
2400 			goto done;
2401 
2402 		/*
2403 		 * Check for symlink.
2404 		 */
2405 		rc = zfs_dnode_stat(spa, &dn, &sb);
2406 		if (rc)
2407 			goto done;
2408 		if (S_ISLNK(sb.st_mode)) {
2409 			if (symlinks_followed > 10) {
2410 				rc = EMLINK;
2411 				goto done;
2412 			}
2413 			symlinks_followed++;
2414 
2415 			/*
2416 			 * Read the link value and copy the tail of our
2417 			 * current path onto the end.
2418 			 */
2419 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
2420 				rc = ENAMETOOLONG;
2421 				goto done;
2422 			}
2423 			strcpy(&path[sb.st_size], p);
2424 
2425 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
2426 			if (rc != 0)
2427 				goto done;
2428 
2429 			/*
2430 			 * Restart with the new path, starting either at
2431 			 * the root or at the parent depending whether or
2432 			 * not the link is relative.
2433 			 */
2434 			p = path;
2435 			if (*p == '/') {
2436 				while (STAILQ_FIRST(&on_cache) !=
2437 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
2438 					entry = STAILQ_FIRST(&on_cache);
2439 					STAILQ_REMOVE_HEAD(&on_cache, entry);
2440 					free(entry);
2441 				}
2442 			} else {
2443 				entry = STAILQ_FIRST(&on_cache);
2444 				STAILQ_REMOVE_HEAD(&on_cache, entry);
2445 				free(entry);
2446 			}
2447 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
2448 		}
2449 	}
2450 
2451 	*dnode = dn;
2452 done:
2453 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
2454 		free(entry);
2455 	return (rc);
2456 }
2457