?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
9051 loader should ignore active multi_vdev_crash_dump feature on zpool
[unleashed.git] / usr / src / boot / sys / boot / zfs / zfsimpl.c
blob26654c046970d8a998ee40dd69ad5e3e0d589a8c
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 "com.joyent:multi_vdev_crash_dump",
63 NULL
64 };
65
66 /*
67 * List of all pools, chained through spa_link.
68 */
69 static spa_list_t zfs_pools;
70
71 static const dnode_phys_t *dnode_cache_obj;
72 static uint64_t dnode_cache_bn;
73 static char *dnode_cache_buf;
74 static char *zap_scratch;
75 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
76
77 #define TEMP_SIZE (1024 * 1024)
78
79 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
80 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
81 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
82
83 static void
84 zfs_init(void)
85 {
86 STAILQ_INIT(&zfs_vdevs);
87 STAILQ_INIT(&zfs_pools);
88
89 zfs_temp_buf = malloc(TEMP_SIZE);
90 zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
91 zfs_temp_ptr = zfs_temp_buf;
92 dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
93 zap_scratch = malloc(SPA_MAXBLOCKSIZE);
94
95 zfs_init_crc();
96 }
97
98 static void *
99 zfs_alloc(size_t size)
100 {
101 char *ptr;
102
103 if (zfs_temp_ptr + size > zfs_temp_end) {
104 printf("ZFS: out of temporary buffer space\n");
105 for (;;) ;
106 }
107 ptr = zfs_temp_ptr;
108 zfs_temp_ptr += size;
109
110 return (ptr);
111 }
112
113 static void
114 zfs_free(void *ptr, size_t size)
115 {
116
117 zfs_temp_ptr -= size;
118 if (zfs_temp_ptr != ptr) {
119 printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
120 for (;;) ;
121 }
122 }
123
124 static int
125 xdr_int(const unsigned char **xdr, int *ip)
126 {
127 *ip = ((*xdr)[0] << 24)
128 | ((*xdr)[1] << 16)
129 | ((*xdr)[2] << 8)
130 | ((*xdr)[3] << 0);
131 (*xdr) += 4;
132 return (0);
133 }
134
135 static int
136 xdr_u_int(const unsigned char **xdr, u_int *ip)
137 {
138 *ip = ((*xdr)[0] << 24)
139 | ((*xdr)[1] << 16)
140 | ((*xdr)[2] << 8)
141 | ((*xdr)[3] << 0);
142 (*xdr) += 4;
143 return (0);
144 }
145
146 static int
147 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
148 {
149 u_int hi, lo;
150
151 xdr_u_int(xdr, &hi);
152 xdr_u_int(xdr, &lo);
153 *lp = (((uint64_t) hi) << 32) | lo;
154 return (0);
155 }
156
157 static int
158 nvlist_find(const unsigned char *nvlist, const char *name, int type,
159 int* elementsp, void *valuep)
160 {
161 const unsigned char *p, *pair;
162 int junk;
163 int encoded_size, decoded_size;
164
165 p = nvlist;
166 xdr_int(&p, &junk);
167 xdr_int(&p, &junk);
168
169 pair = p;
170 xdr_int(&p, &encoded_size);
171 xdr_int(&p, &decoded_size);
172 while (encoded_size && decoded_size) {
173 int namelen, pairtype, elements;
174 const char *pairname;
175
176 xdr_int(&p, &namelen);
177 pairname = (const char*) p;
178 p += roundup(namelen, 4);
179 xdr_int(&p, &pairtype);
180
181 if (!memcmp(name, pairname, namelen) && type == pairtype) {
182 xdr_int(&p, &elements);
183 if (elementsp)
184 *elementsp = elements;
185 if (type == DATA_TYPE_UINT64) {
186 xdr_uint64_t(&p, (uint64_t *) valuep);
187 return (0);
188 } else if (type == DATA_TYPE_STRING) {
189 int len;
190 xdr_int(&p, &len);
191 (*(const char**) valuep) = (const char*) p;
192 return (0);
193 } else if (type == DATA_TYPE_NVLIST
194 || type == DATA_TYPE_NVLIST_ARRAY) {
195 (*(const unsigned char**) valuep) =
196 (const unsigned char*) p;
197 return (0);
198 } else {
199 return (EIO);
200 }
201 } else {
202 /*
203 * Not the pair we are looking for, skip to the next one.
204 */
205 p = pair + encoded_size;
206 }
207
208 pair = p;
209 xdr_int(&p, &encoded_size);
210 xdr_int(&p, &decoded_size);
211 }
212
213 return (EIO);
214 }
215
216 static int
217 nvlist_check_features_for_read(const unsigned char *nvlist)
218 {
219 const unsigned char *p, *pair;
220 int junk;
221 int encoded_size, decoded_size;
222 int rc;
223
224 rc = 0;
225
226 p = nvlist;
227 xdr_int(&p, &junk);
228 xdr_int(&p, &junk);
229
230 pair = p;
231 xdr_int(&p, &encoded_size);
232 xdr_int(&p, &decoded_size);
233 while (encoded_size && decoded_size) {
234 int namelen, pairtype;
235 const char *pairname;
236 int i, found;
237
238 found = 0;
239
240 xdr_int(&p, &namelen);
241 pairname = (const char*) p;
242 p += roundup(namelen, 4);
243 xdr_int(&p, &pairtype);
244
245 for (i = 0; features_for_read[i] != NULL; i++) {
246 if (!memcmp(pairname, features_for_read[i], namelen)) {
247 found = 1;
248 break;
249 }
250 }
251
252 if (!found) {
253 printf("ZFS: unsupported feature: %s\n", pairname);
254 rc = EIO;
255 }
256
257 p = pair + encoded_size;
258
259 pair = p;
260 xdr_int(&p, &encoded_size);
261 xdr_int(&p, &decoded_size);
262 }
263
264 return (rc);
265 }
266
267 /*
268 * Return the next nvlist in an nvlist array.
269 */
270 static const unsigned char *
271 nvlist_next(const unsigned char *nvlist)
272 {
273 const unsigned char *p, *pair;
274 int junk;
275 int encoded_size, decoded_size;
276
277 p = nvlist;
278 xdr_int(&p, &junk);
279 xdr_int(&p, &junk);
280
281 pair = p;
282 xdr_int(&p, &encoded_size);
283 xdr_int(&p, &decoded_size);
284 while (encoded_size && decoded_size) {
285 p = pair + encoded_size;
286
287 pair = p;
288 xdr_int(&p, &encoded_size);
289 xdr_int(&p, &decoded_size);
290 }
291
292 return p;
293 }
294
295 #ifdef TEST
296
297 static const unsigned char *
298 nvlist_print(const unsigned char *nvlist, unsigned int indent)
299 {
300 static const char* typenames[] = {
301 "DATA_TYPE_UNKNOWN",
302 "DATA_TYPE_BOOLEAN",
303 "DATA_TYPE_BYTE",
304 "DATA_TYPE_INT16",
305 "DATA_TYPE_UINT16",
306 "DATA_TYPE_INT32",
307 "DATA_TYPE_UINT32",
308 "DATA_TYPE_INT64",
309 "DATA_TYPE_UINT64",
310 "DATA_TYPE_STRING",
311 "DATA_TYPE_BYTE_ARRAY",
312 "DATA_TYPE_INT16_ARRAY",
313 "DATA_TYPE_UINT16_ARRAY",
314 "DATA_TYPE_INT32_ARRAY",
315 "DATA_TYPE_UINT32_ARRAY",
316 "DATA_TYPE_INT64_ARRAY",
317 "DATA_TYPE_UINT64_ARRAY",
318 "DATA_TYPE_STRING_ARRAY",
319 "DATA_TYPE_HRTIME",
320 "DATA_TYPE_NVLIST",
321 "DATA_TYPE_NVLIST_ARRAY",
322 "DATA_TYPE_BOOLEAN_VALUE",
323 "DATA_TYPE_INT8",
324 "DATA_TYPE_UINT8",
325 "DATA_TYPE_BOOLEAN_ARRAY",
326 "DATA_TYPE_INT8_ARRAY",
327 "DATA_TYPE_UINT8_ARRAY"
328 };
329
330 unsigned int i, j;
331 const unsigned char *p, *pair;
332 int junk;
333 int encoded_size, decoded_size;
334
335 p = nvlist;
336 xdr_int(&p, &junk);
337 xdr_int(&p, &junk);
338
339 pair = p;
340 xdr_int(&p, &encoded_size);
341 xdr_int(&p, &decoded_size);
342 while (encoded_size && decoded_size) {
343 int namelen, pairtype, elements;
344 const char *pairname;
345
346 xdr_int(&p, &namelen);
347 pairname = (const char*) p;
348 p += roundup(namelen, 4);
349 xdr_int(&p, &pairtype);
350
351 for (i = 0; i < indent; i++)
352 printf(" ");
353 printf("%s %s", typenames[pairtype], pairname);
354
355 xdr_int(&p, &elements);
356 switch (pairtype) {
357 case DATA_TYPE_UINT64: {
358 uint64_t val;
359 xdr_uint64_t(&p, &val);
360 printf(" = 0x%jx\n", (uintmax_t)val);
361 break;
362 }
363
364 case DATA_TYPE_STRING: {
365 int len;
366 xdr_int(&p, &len);
367 printf(" = \"%s\"\n", p);
368 break;
369 }
370
371 case DATA_TYPE_NVLIST:
372 printf("\n");
373 nvlist_print(p, indent + 1);
374 break;
375
376 case DATA_TYPE_NVLIST_ARRAY:
377 for (j = 0; j < elements; j++) {
378 printf("[%d]\n", j);
379 p = nvlist_print(p, indent + 1);
380 if (j != elements - 1) {
381 for (i = 0; i < indent; i++)
382 printf(" ");
383 printf("%s %s", typenames[pairtype], pairname);
384 }
385 }
386 break;
387
388 default:
389 printf("\n");
390 }
391
392 p = pair + encoded_size;
393
394 pair = p;
395 xdr_int(&p, &encoded_size);
396 xdr_int(&p, &decoded_size);
397 }
398
399 return p;
400 }
401
402 #endif
403
404 static int
405 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
406 off_t offset, size_t size)
407 {
408 size_t psize;
409 int rc;
410
411 if (!vdev->v_phys_read)
412 return (EIO);
413
414 if (bp) {
415 psize = BP_GET_PSIZE(bp);
416 } else {
417 psize = size;
418 }
419
420 /*printf("ZFS: reading %zu bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
421 rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
422 if (rc)
423 return (rc);
424 if (bp && zio_checksum_verify(bp, buf))
425 return (EIO);
426
427 return (0);
428 }
429
430 static int
431 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
432 off_t offset, size_t bytes)
433 {
434
435 return (vdev_read_phys(vdev, bp, buf,
436 offset + VDEV_LABEL_START_SIZE, bytes));
437 }
438
439
440 static int
441 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
442 off_t offset, size_t bytes)
443 {
444 vdev_t *kid;
445 int rc;
446
447 rc = EIO;
448 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
449 if (kid->v_state != VDEV_STATE_HEALTHY)
450 continue;
451 rc = kid->v_read(kid, bp, buf, offset, bytes);
452 if (!rc)
453 return (0);
454 }
455
456 return (rc);
457 }
458
459 static int
460 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
461 off_t offset, size_t bytes)
462 {
463 vdev_t *kid;
464
465 /*
466 * Here we should have two kids:
467 * First one which is the one we are replacing and we can trust
468 * only this one to have valid data, but it might not be present.
469 * Second one is that one we are replacing with. It is most likely
470 * healthy, but we can't trust it has needed data, so we won't use it.
471 */
472 kid = STAILQ_FIRST(&vdev->v_children);
473 if (kid == NULL)
474 return (EIO);
475 if (kid->v_state != VDEV_STATE_HEALTHY)
476 return (EIO);
477 return (kid->v_read(kid, bp, buf, offset, bytes));
478 }
479
480 static vdev_t *
481 vdev_find(uint64_t guid)
482 {
483 vdev_t *vdev;
484
485 STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
486 if (vdev->v_guid == guid)
487 return (vdev);
488
489 return (0);
490 }
491
492 static vdev_t *
493 vdev_create(uint64_t guid, vdev_read_t *vdev_read)
494 {
495 vdev_t *vdev;
496
497 vdev = malloc(sizeof(vdev_t));
498 memset(vdev, 0, sizeof(vdev_t));
499 STAILQ_INIT(&vdev->v_children);
500 vdev->v_guid = guid;
501 vdev->v_state = VDEV_STATE_OFFLINE;
502 vdev->v_read = vdev_read;
503 vdev->v_phys_read = 0;
504 vdev->v_read_priv = 0;
505 STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
506
507 return (vdev);
508 }
509
510 static int
511 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
512 vdev_t **vdevp, int is_newer)
513 {
514 int rc;
515 uint64_t guid, id, ashift, nparity;
516 const char *type;
517 const char *path;
518 vdev_t *vdev, *kid;
519 const unsigned char *kids;
520 int nkids, i, is_new;
521 uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
522
523 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
524 NULL, &guid) ||
525 nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id) ||
526 nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
527 NULL, &type)) {
528 printf("ZFS: can't find vdev details\n");
529 return (ENOENT);
530 }
531
532 if (strcmp(type, VDEV_TYPE_MIRROR)
533 && strcmp(type, VDEV_TYPE_DISK)
534 #ifdef ZFS_TEST
535 && strcmp(type, VDEV_TYPE_FILE)
536 #endif
537 && strcmp(type, VDEV_TYPE_RAIDZ)
538 && strcmp(type, VDEV_TYPE_REPLACING)) {
539 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
540 return (EIO);
541 }
542
543 is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
544
545 nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
546 &is_offline);
547 nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
548 &is_removed);
549 nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
550 &is_faulted);
551 nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, NULL,
552 &is_degraded);
553 nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, NULL,
554 &isnt_present);
555
556 vdev = vdev_find(guid);
557 if (!vdev) {
558 is_new = 1;
559
560 if (!strcmp(type, VDEV_TYPE_MIRROR))
561 vdev = vdev_create(guid, vdev_mirror_read);
562 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
563 vdev = vdev_create(guid, vdev_raidz_read);
564 else if (!strcmp(type, VDEV_TYPE_REPLACING))
565 vdev = vdev_create(guid, vdev_replacing_read);
566 else
567 vdev = vdev_create(guid, vdev_disk_read);
568
569 vdev->v_id = id;
570 vdev->v_top = pvdev != NULL ? pvdev : vdev;
571 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
572 DATA_TYPE_UINT64, NULL, &ashift) == 0) {
573 vdev->v_ashift = ashift;
574 } else {
575 vdev->v_ashift = 0;
576 }
577 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
578 DATA_TYPE_UINT64, NULL, &nparity) == 0) {
579 vdev->v_nparity = nparity;
580 } else {
581 vdev->v_nparity = 0;
582 }
583 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
584 DATA_TYPE_STRING, NULL, &path) == 0) {
585 if (strncmp(path, "/dev/dsk/", 9) == 0)
586 path += 9;
587 vdev->v_name = strdup(path);
588 if (nvlist_find(nvlist, ZPOOL_CONFIG_PHYS_PATH,
589 DATA_TYPE_STRING, NULL, &path) == 0) {
590 vdev->v_phys_path = strdup(path);
591 } else {
592 vdev->v_phys_path = NULL;
593 }
594 if (nvlist_find(nvlist, ZPOOL_CONFIG_DEVID,
595 DATA_TYPE_STRING, NULL, &path) == 0) {
596 vdev->v_devid = strdup(path);
597 } else {
598 vdev->v_devid = NULL;
599 }
600 } else {
601 if (!strcmp(type, "raidz")) {
602 if (vdev->v_nparity == 1)
603 vdev->v_name = "raidz1";
604 else if (vdev->v_nparity == 2)
605 vdev->v_name = "raidz2";
606 else if (vdev->v_nparity == 3)
607 vdev->v_name = "raidz3";
608 else {
609 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
610 return (EIO);
611 }
612 } else {
613 vdev->v_name = strdup(type);
614 }
615 }
616 } else {
617 is_new = 0;
618 }
619
620 if (is_new || is_newer) {
621 /*
622 * This is either new vdev or we've already seen this vdev,
623 * but from an older vdev label, so let's refresh its state
624 * from the newer label.
625 */
626 if (is_offline)
627 vdev->v_state = VDEV_STATE_OFFLINE;
628 else if (is_removed)
629 vdev->v_state = VDEV_STATE_REMOVED;
630 else if (is_faulted)
631 vdev->v_state = VDEV_STATE_FAULTED;
632 else if (is_degraded)
633 vdev->v_state = VDEV_STATE_DEGRADED;
634 else if (isnt_present)
635 vdev->v_state = VDEV_STATE_CANT_OPEN;
636 }
637
638 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
639 &nkids, &kids);
640 /*
641 * Its ok if we don't have any kids.
642 */
643 if (rc == 0) {
644 vdev->v_nchildren = nkids;
645 for (i = 0; i < nkids; i++) {
646 rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
647 if (rc)
648 return (rc);
649 if (is_new)
650 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
651 v_childlink);
652 kids = nvlist_next(kids);
653 }
654 } else {
655 vdev->v_nchildren = 0;
656 }
657
658 if (vdevp)
659 *vdevp = vdev;
660 return (0);
661 }
662
663 static void
664 vdev_set_state(vdev_t *vdev)
665 {
666 vdev_t *kid;
667 int good_kids;
668 int bad_kids;
669
670 /*
671 * A mirror or raidz is healthy if all its kids are healthy. A
672 * mirror is degraded if any of its kids is healthy; a raidz
673 * is degraded if at most nparity kids are offline.
674 */
675 if (STAILQ_FIRST(&vdev->v_children)) {
676 good_kids = 0;
677 bad_kids = 0;
678 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
679 if (kid->v_state == VDEV_STATE_HEALTHY)
680 good_kids++;
681 else
682 bad_kids++;
683 }
684 if (bad_kids == 0) {
685 vdev->v_state = VDEV_STATE_HEALTHY;
686 } else {
687 if (vdev->v_read == vdev_mirror_read) {
688 if (good_kids) {
689 vdev->v_state = VDEV_STATE_DEGRADED;
690 } else {
691 vdev->v_state = VDEV_STATE_OFFLINE;
692 }
693 } else if (vdev->v_read == vdev_raidz_read) {
694 if (bad_kids > vdev->v_nparity) {
695 vdev->v_state = VDEV_STATE_OFFLINE;
696 } else {
697 vdev->v_state = VDEV_STATE_DEGRADED;
698 }
699 }
700 }
701 }
702 }
703
704 static spa_t *
705 spa_find_by_guid(uint64_t guid)
706 {
707 spa_t *spa;
708
709 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
710 if (spa->spa_guid == guid)
711 return (spa);
712
713 return (0);
714 }
715
716 static spa_t *
717 spa_find_by_name(const char *name)
718 {
719 spa_t *spa;
720
721 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
722 if (!strcmp(spa->spa_name, name))
723 return (spa);
724
725 return (0);
726 }
727
728 spa_t *
729 spa_get_primary(void)
730 {
731 return (STAILQ_FIRST(&zfs_pools));
732 }
733
734 vdev_t *
735 spa_get_primary_vdev(const spa_t *spa)
736 {
737 vdev_t *vdev;
738 vdev_t *kid;
739
740 if (spa == NULL)
741 spa = spa_get_primary();
742 if (spa == NULL)
743 return (NULL);
744 vdev = STAILQ_FIRST(&spa->spa_vdevs);
745 if (vdev == NULL)
746 return (NULL);
747 for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
748 kid = STAILQ_FIRST(&vdev->v_children))
749 vdev = kid;
750 return (vdev);
751 }
752
753 static spa_t *
754 spa_create(uint64_t guid, const char *name)
755 {
756 spa_t *spa;
757
758 if ((spa = malloc(sizeof(spa_t))) == NULL)
759 return (NULL);
760 memset(spa, 0, sizeof(spa_t));
761 if ((spa->spa_name = strdup(name)) == NULL) {
762 free(spa);
763 return (NULL);
764 }
765 STAILQ_INIT(&spa->spa_vdevs);
766 spa->spa_guid = guid;
767 STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
768
769 return (spa);
770 }
771
772 static const char *
773 state_name(vdev_state_t state)
774 {
775 static const char* names[] = {
776 "UNKNOWN",
777 "CLOSED",
778 "OFFLINE",
779 "REMOVED",
780 "CANT_OPEN",
781 "FAULTED",
782 "DEGRADED",
783 "ONLINE"
784 };
785 return names[state];
786 }
787
788 static int
789 pager_printf(const char *fmt, ...)
790 {
791 char line[80];
792 va_list args;
793
794 va_start(args, fmt);
795 vsnprintf(line, sizeof (line), fmt, args);
796 va_end(args);
797 return (pager_output(line));
798 }
799
800 #define STATUS_FORMAT " %s %s\n"
801
802 static int
803 print_state(int indent, const char *name, vdev_state_t state)
804 {
805 int i;
806 char buf[512];
807
808 buf[0] = 0;
809 for (i = 0; i < indent; i++)
810 strcat(buf, " ");
811 strcat(buf, name);
812 return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
813 }
814
815 static int
816 vdev_status(vdev_t *vdev, int indent)
817 {
818 vdev_t *kid;
819 int ret;
820 ret = print_state(indent, vdev->v_name, vdev->v_state);
821 if (ret != 0)
822 return (ret);
823
824 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
825 ret = vdev_status(kid, indent + 1);
826 if (ret != 0)
827 return (ret);
828 }
829 return (ret);
830 }
831
832 static int
833 spa_status(spa_t *spa)
834 {
835 static char bootfs[ZFS_MAXNAMELEN];
836 uint64_t rootid;
837 vdev_t *vdev;
838 int good_kids, bad_kids, degraded_kids, ret;
839 vdev_state_t state;
840
841 ret = pager_printf(" pool: %s\n", spa->spa_name);
842 if (ret != 0)
843 return (ret);
844
845 if (zfs_get_root(spa, &rootid) == 0 &&
846 zfs_rlookup(spa, rootid, bootfs) == 0) {
847 if (bootfs[0] == '\0')
848 ret = pager_printf("bootfs: %s\n", spa->spa_name);
849 else
850 ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
851 bootfs);
852 if (ret != 0)
853 return (ret);
854 }
855 ret = pager_printf("config:\n\n");
856 if (ret != 0)
857 return (ret);
858 ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
859 if (ret != 0)
860 return (ret);
861
862 good_kids = 0;
863 degraded_kids = 0;
864 bad_kids = 0;
865 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
866 if (vdev->v_state == VDEV_STATE_HEALTHY)
867 good_kids++;
868 else if (vdev->v_state == VDEV_STATE_DEGRADED)
869 degraded_kids++;
870 else
871 bad_kids++;
872 }
873
874 state = VDEV_STATE_CLOSED;
875 if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
876 state = VDEV_STATE_HEALTHY;
877 else if ((good_kids + degraded_kids) > 0)
878 state = VDEV_STATE_DEGRADED;
879
880 ret = print_state(0, spa->spa_name, state);
881 if (ret != 0)
882 return (ret);
883 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
884 ret = vdev_status(vdev, 1);
885 if (ret != 0)
886 return (ret);
887 }
888 return (ret);
889 }
890
891 int
892 spa_all_status(void)
893 {
894 spa_t *spa;
895 int first = 1, ret = 0;
896
897 STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
898 if (!first) {
899 ret = pager_printf("\n");
900 if (ret != 0)
901 return (ret);
902 }
903 first = 0;
904 ret = spa_status(spa);
905 if (ret != 0)
906 return (ret);
907 }
908 return (ret);
909 }
910
911 uint64_t
912 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
913 {
914 uint64_t label_offset;
915
916 if (l < VDEV_LABELS / 2)
917 label_offset = 0;
918 else
919 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
920
921 return (offset + l * sizeof (vdev_label_t) + label_offset);
922 }
923
924 static int
925 vdev_probe(vdev_phys_read_t *phys_read, void *read_priv, spa_t **spap)
926 {
927 vdev_t vtmp;
928 vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
929 vdev_phys_t *tmp_label;
930 spa_t *spa;
931 vdev_t *vdev, *top_vdev, *pool_vdev;
932 off_t off;
933 blkptr_t bp;
934 const unsigned char *nvlist = NULL;
935 uint64_t val;
936 uint64_t guid;
937 uint64_t best_txg = 0;
938 uint64_t pool_txg, pool_guid;
939 uint64_t psize;
940 const char *pool_name;
941 const unsigned char *vdevs;
942 const unsigned char *features;
943 int i, l, rc, is_newer;
944 char *upbuf;
945 const struct uberblock *up;
946
947 /*
948 * Load the vdev label and figure out which
949 * uberblock is most current.
950 */
951 memset(&vtmp, 0, sizeof(vtmp));
952 vtmp.v_phys_read = phys_read;
953 vtmp.v_read_priv = read_priv;
954 psize = P2ALIGN(ldi_get_size(read_priv),
955 (uint64_t)sizeof (vdev_label_t));
956
957 /* Test for minimum device size. */
958 if (psize < SPA_MINDEVSIZE)
959 return (EIO);
960
961 tmp_label = zfs_alloc(sizeof (vdev_phys_t));
962
963 for (l = 0; l < VDEV_LABELS; l++) {
964 off = vdev_label_offset(psize, l,
965 offsetof(vdev_label_t, vl_vdev_phys));
966
967 BP_ZERO(&bp);
968 BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
969 BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
970 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
971 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
972 DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
973 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
974
975 if (vdev_read_phys(&vtmp, &bp, tmp_label, off, 0))
976 continue;
977
978 if (tmp_label->vp_nvlist[0] != NV_ENCODE_XDR)
979 continue;
980
981 nvlist = (const unsigned char *) tmp_label->vp_nvlist + 4;
982 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
983 DATA_TYPE_UINT64, NULL, &pool_txg) != 0)
984 continue;
985
986 if (best_txg <= pool_txg) {
987 best_txg = pool_txg;
988 memcpy(vdev_label, tmp_label, sizeof (vdev_phys_t));
989 }
990 }
991
992 zfs_free(tmp_label, sizeof (vdev_phys_t));
993
994 if (best_txg == 0)
995 return (EIO);
996
997 if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR)
998 return (EIO);
999
1000 nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
1001
1002 if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1003 NULL, &val) != 0) {
1004 return (EIO);
1005 }
1006
1007 if (!SPA_VERSION_IS_SUPPORTED(val)) {
1008 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1009 (unsigned) val, (unsigned) SPA_VERSION);
1010 return (EIO);
1011 }
1012
1013 /* Check ZFS features for read */
1014 if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1015 DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1016 nvlist_check_features_for_read(features) != 0) {
1017 return (EIO);
1018 }
1019
1020 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1021 NULL, &val) != 0) {
1022 return (EIO);
1023 }
1024
1025 if (val == POOL_STATE_DESTROYED) {
1026 /* We don't boot only from destroyed pools. */
1027 return (EIO);
1028 }
1029
1030 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1031 NULL, &pool_txg) != 0 ||
1032 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1033 NULL, &pool_guid) != 0 ||
1034 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1035 NULL, &pool_name) != 0) {
1036 /*
1037 * Cache and spare devices end up here - just ignore
1038 * them.
1039 */
1040 /*printf("ZFS: can't find pool details\n");*/
1041 return (EIO);
1042 }
1043
1044 if (nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64,
1045 NULL, &val) == 0 && val != 0) {
1046 return (EIO);
1047 }
1048
1049 /*
1050 * Create the pool if this is the first time we've seen it.
1051 */
1052 spa = spa_find_by_guid(pool_guid);
1053 if (spa == NULL) {
1054 spa = spa_create(pool_guid, pool_name);
1055 if (spa == NULL)
1056 return (ENOMEM);
1057 }
1058 if (pool_txg > spa->spa_txg) {
1059 spa->spa_txg = pool_txg;
1060 is_newer = 1;
1061 } else {
1062 is_newer = 0;
1063 }
1064
1065 /*
1066 * Get the vdev tree and create our in-core copy of it.
1067 * If we already have a vdev with this guid, this must
1068 * be some kind of alias (overlapping slices, dangerously dedicated
1069 * disks etc).
1070 */
1071 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1072 NULL, &guid) != 0) {
1073 return (EIO);
1074 }
1075 vdev = vdev_find(guid);
1076 if (vdev && vdev->v_phys_read) /* Has this vdev already been inited? */
1077 return (EIO);
1078
1079 if (nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1080 NULL, &vdevs)) {
1081 return (EIO);
1082 }
1083
1084 rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1085 if (rc != 0)
1086 return (rc);
1087
1088 /*
1089 * Add the toplevel vdev to the pool if its not already there.
1090 */
1091 STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1092 if (top_vdev == pool_vdev)
1093 break;
1094 if (!pool_vdev && top_vdev)
1095 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1096
1097 /*
1098 * We should already have created an incomplete vdev for this
1099 * vdev. Find it and initialise it with our read proc.
1100 */
1101 vdev = vdev_find(guid);
1102 if (vdev) {
1103 vdev->v_phys_read = phys_read;
1104 vdev->v_read_priv = read_priv;
1105 vdev->v_state = VDEV_STATE_HEALTHY;
1106 } else {
1107 printf("ZFS: inconsistent nvlist contents\n");
1108 return (EIO);
1109 }
1110
1111 /*
1112 * Re-evaluate top-level vdev state.
1113 */
1114 vdev_set_state(top_vdev);
1115
1116 /*
1117 * Ok, we are happy with the pool so far. Lets find
1118 * the best uberblock and then we can actually access
1119 * the contents of the pool.
1120 */
1121 upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
1122 up = (const struct uberblock *)upbuf;
1123 for (l = 0; l < VDEV_LABELS; l++) {
1124 for (i = 0; i < VDEV_UBERBLOCK_COUNT(vdev); i++) {
1125 off = vdev_label_offset(psize, l,
1126 VDEV_UBERBLOCK_OFFSET(vdev, i));
1127 BP_ZERO(&bp);
1128 DVA_SET_OFFSET(&bp.blk_dva[0], off);
1129 BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1130 BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1131 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1132 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1133 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1134
1135 if (vdev_read_phys(vdev, &bp, upbuf, off, 0) != 0)
1136 continue;
1137
1138 if (up->ub_magic != UBERBLOCK_MAGIC)
1139 continue;
1140 if (up->ub_txg < spa->spa_txg)
1141 continue;
1142 if (up->ub_txg > spa->spa_uberblock.ub_txg ||
1143 (up->ub_txg == spa->spa_uberblock.ub_txg &&
1144 up->ub_timestamp >
1145 spa->spa_uberblock.ub_timestamp)) {
1146 spa->spa_uberblock = *up;
1147 }
1148 }
1149 }
1150 zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
1151
1152 if (spap != NULL)
1153 *spap = spa;
1154 return (0);
1155 }
1156
1157 static int
1158 ilog2(int n)
1159 {
1160 int v;
1161
1162 for (v = 0; v < 32; v++)
1163 if (n == (1 << v))
1164 return v;
1165 return -1;
1166 }
1167
1168 static int
1169 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1170 {
1171 blkptr_t gbh_bp;
1172 zio_gbh_phys_t zio_gb;
1173 char *pbuf;
1174 int i;
1175
1176 /* Artificial BP for gang block header. */
1177 gbh_bp = *bp;
1178 BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1179 BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1180 BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1181 BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1182 for (i = 0; i < SPA_DVAS_PER_BP; i++)
1183 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1184
1185 /* Read gang header block using the artificial BP. */
1186 if (zio_read(spa, &gbh_bp, &zio_gb))
1187 return (EIO);
1188
1189 pbuf = buf;
1190 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1191 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1192
1193 if (BP_IS_HOLE(gbp))
1194 continue;
1195 if (zio_read(spa, gbp, pbuf))
1196 return (EIO);
1197 pbuf += BP_GET_PSIZE(gbp);
1198 }
1199
1200 if (zio_checksum_verify(bp, buf))
1201 return (EIO);
1202 return (0);
1203 }
1204
1205 static int
1206 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1207 {
1208 int cpfunc = BP_GET_COMPRESS(bp);
1209 uint64_t align, size;
1210 void *pbuf;
1211 int i, error;
1212
1213 /*
1214 * Process data embedded in block pointer
1215 */
1216 if (BP_IS_EMBEDDED(bp)) {
1217 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1218
1219 size = BPE_GET_PSIZE(bp);
1220 ASSERT(size <= BPE_PAYLOAD_SIZE);
1221
1222 if (cpfunc != ZIO_COMPRESS_OFF)
1223 pbuf = zfs_alloc(size);
1224 else
1225 pbuf = buf;
1226
1227 decode_embedded_bp_compressed(bp, pbuf);
1228 error = 0;
1229
1230 if (cpfunc != ZIO_COMPRESS_OFF) {
1231 error = zio_decompress_data(cpfunc, pbuf,
1232 size, buf, BP_GET_LSIZE(bp));
1233 zfs_free(pbuf, size);
1234 }
1235 if (error != 0)
1236 printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
1237 error);
1238 return (error);
1239 }
1240
1241 error = EIO;
1242
1243 for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1244 const dva_t *dva = &bp->blk_dva[i];
1245 vdev_t *vdev;
1246 int vdevid;
1247 off_t offset;
1248
1249 if (!dva->dva_word[0] && !dva->dva_word[1])
1250 continue;
1251
1252 vdevid = DVA_GET_VDEV(dva);
1253 offset = DVA_GET_OFFSET(dva);
1254 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1255 if (vdev->v_id == vdevid)
1256 break;
1257 }
1258 if (!vdev || !vdev->v_read)
1259 continue;
1260
1261 size = BP_GET_PSIZE(bp);
1262 if (vdev->v_read == vdev_raidz_read) {
1263 align = 1ULL << vdev->v_top->v_ashift;
1264 if (P2PHASE(size, align) != 0)
1265 size = P2ROUNDUP(size, align);
1266 }
1267 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1268 pbuf = zfs_alloc(size);
1269 else
1270 pbuf = buf;
1271
1272 if (DVA_GET_GANG(dva))
1273 error = zio_read_gang(spa, bp, pbuf);
1274 else
1275 error = vdev->v_read(vdev, bp, pbuf, offset, size);
1276 if (error == 0) {
1277 if (cpfunc != ZIO_COMPRESS_OFF)
1278 error = zio_decompress_data(cpfunc, pbuf,
1279 BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1280 else if (size != BP_GET_PSIZE(bp))
1281 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1282 }
1283 if (buf != pbuf)
1284 zfs_free(pbuf, size);
1285 if (error == 0)
1286 break;
1287 }
1288 if (error != 0)
1289 printf("ZFS: i/o error - all block copies unavailable\n");
1290 return (error);
1291 }
1292
1293 static int
1294 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1295 {
1296 int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1297 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1298 int nlevels = dnode->dn_nlevels;
1299 int i, rc;
1300
1301 if (bsize > SPA_MAXBLOCKSIZE) {
1302 printf("ZFS: I/O error - blocks larger than %llu are not "
1303 "supported\n", SPA_MAXBLOCKSIZE);
1304 return (EIO);
1305 }
1306
1307 /*
1308 * Note: bsize may not be a power of two here so we need to do an
1309 * actual divide rather than a bitshift.
1310 */
1311 while (buflen > 0) {
1312 uint64_t bn = offset / bsize;
1313 int boff = offset % bsize;
1314 int ibn;
1315 const blkptr_t *indbp;
1316 blkptr_t bp;
1317
1318 if (bn > dnode->dn_maxblkid) {
1319 printf("warning: zfs bug: bn %llx > dn_maxblkid %llx\n",
1320 (unsigned long long)bn,
1321 (unsigned long long)dnode->dn_maxblkid);
1322 /*
1323 * zfs bug, will not return error
1324 * return (EIO);
1325 */
1326 }
1327
1328 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1329 goto cached;
1330
1331 indbp = dnode->dn_blkptr;
1332 for (i = 0; i < nlevels; i++) {
1333 /*
1334 * Copy the bp from the indirect array so that
1335 * we can re-use the scratch buffer for multi-level
1336 * objects.
1337 */
1338 ibn = bn >> ((nlevels - i - 1) * ibshift);
1339 ibn &= ((1 << ibshift) - 1);
1340 bp = indbp[ibn];
1341 if (BP_IS_HOLE(&bp)) {
1342 memset(dnode_cache_buf, 0, bsize);
1343 break;
1344 }
1345 rc = zio_read(spa, &bp, dnode_cache_buf);
1346 if (rc)
1347 return (rc);
1348 indbp = (const blkptr_t *) dnode_cache_buf;
1349 }
1350 dnode_cache_obj = dnode;
1351 dnode_cache_bn = bn;
1352 cached:
1353
1354 /*
1355 * The buffer contains our data block. Copy what we
1356 * need from it and loop.
1357 */
1358 i = bsize - boff;
1359 if (i > buflen) i = buflen;
1360 memcpy(buf, &dnode_cache_buf[boff], i);
1361 buf = ((char*) buf) + i;
1362 offset += i;
1363 buflen -= i;
1364 }
1365
1366 return (0);
1367 }
1368
1369 /*
1370 * Lookup a value in a microzap directory. Assumes that the zap
1371 * scratch buffer contains the directory contents.
1372 */
1373 static int
1374 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1375 {
1376 const mzap_phys_t *mz;
1377 const mzap_ent_phys_t *mze;
1378 size_t size;
1379 int chunks, i;
1380
1381 /*
1382 * Microzap objects use exactly one block. Read the whole
1383 * thing.
1384 */
1385 size = dnode->dn_datablkszsec * 512;
1386
1387 mz = (const mzap_phys_t *) zap_scratch;
1388 chunks = size / MZAP_ENT_LEN - 1;
1389
1390 for (i = 0; i < chunks; i++) {
1391 mze = &mz->mz_chunk[i];
1392 if (!strcmp(mze->mze_name, name)) {
1393 *value = mze->mze_value;
1394 return (0);
1395 }
1396 }
1397
1398 return (ENOENT);
1399 }
1400
1401 /*
1402 * Compare a name with a zap leaf entry. Return non-zero if the name
1403 * matches.
1404 */
1405 static int
1406 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1407 {
1408 size_t namelen;
1409 const zap_leaf_chunk_t *nc;
1410 const char *p;
1411
1412 namelen = zc->l_entry.le_name_numints;
1413
1414 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1415 p = name;
1416 while (namelen > 0) {
1417 size_t len;
1418 len = namelen;
1419 if (len > ZAP_LEAF_ARRAY_BYTES)
1420 len = ZAP_LEAF_ARRAY_BYTES;
1421 if (memcmp(p, nc->l_array.la_array, len))
1422 return (0);
1423 p += len;
1424 namelen -= len;
1425 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1426 }
1427
1428 return 1;
1429 }
1430
1431 /*
1432 * Extract a uint64_t value from a zap leaf entry.
1433 */
1434 static uint64_t
1435 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1436 {
1437 const zap_leaf_chunk_t *vc;
1438 int i;
1439 uint64_t value;
1440 const uint8_t *p;
1441
1442 vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1443 for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1444 value = (value << 8) | p[i];
1445 }
1446
1447 return value;
1448 }
1449
1450 /*
1451 * Lookup a value in a fatzap directory. Assumes that the zap scratch
1452 * buffer contains the directory header.
1453 */
1454 static int
1455 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1456 {
1457 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1458 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1459 fat_zap_t z;
1460 uint64_t *ptrtbl;
1461 uint64_t hash;
1462 int rc;
1463
1464 if (zh.zap_magic != ZAP_MAGIC)
1465 return (EIO);
1466
1467 z.zap_block_shift = ilog2(bsize);
1468 z.zap_phys = (zap_phys_t *) zap_scratch;
1469
1470 /*
1471 * Figure out where the pointer table is and read it in if necessary.
1472 */
1473 if (zh.zap_ptrtbl.zt_blk) {
1474 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1475 zap_scratch, bsize);
1476 if (rc)
1477 return (rc);
1478 ptrtbl = (uint64_t *) zap_scratch;
1479 } else {
1480 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1481 }
1482
1483 hash = zap_hash(zh.zap_salt, name);
1484
1485 zap_leaf_t zl;
1486 zl.l_bs = z.zap_block_shift;
1487
1488 off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1489 zap_leaf_chunk_t *zc;
1490
1491 rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1492 if (rc)
1493 return (rc);
1494
1495 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1496
1497 /*
1498 * Make sure this chunk matches our hash.
1499 */
1500 if (zl.l_phys->l_hdr.lh_prefix_len > 0
1501 && zl.l_phys->l_hdr.lh_prefix
1502 != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1503 return (ENOENT);
1504
1505 /*
1506 * Hash within the chunk to find our entry.
1507 */
1508 int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1509 int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1510 h = zl.l_phys->l_hash[h];
1511 if (h == 0xffff)
1512 return (ENOENT);
1513 zc = &ZAP_LEAF_CHUNK(&zl, h);
1514 while (zc->l_entry.le_hash != hash) {
1515 if (zc->l_entry.le_next == 0xffff) {
1516 zc = 0;
1517 break;
1518 }
1519 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1520 }
1521 if (fzap_name_equal(&zl, zc, name)) {
1522 if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints > 8)
1523 return (E2BIG);
1524 *value = fzap_leaf_value(&zl, zc);
1525 return (0);
1526 }
1527
1528 return (ENOENT);
1529 }
1530
1531 /*
1532 * Lookup a name in a zap object and return its value as a uint64_t.
1533 */
1534 static int
1535 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1536 {
1537 int rc;
1538 uint64_t zap_type;
1539 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1540
1541 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1542 if (rc)
1543 return (rc);
1544
1545 zap_type = *(uint64_t *) zap_scratch;
1546 if (zap_type == ZBT_MICRO)
1547 return mzap_lookup(dnode, name, value);
1548 else if (zap_type == ZBT_HEADER)
1549 return fzap_lookup(spa, dnode, name, value);
1550 printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
1551 return (EIO);
1552 }
1553
1554 /*
1555 * List a microzap directory. Assumes that the zap scratch buffer contains
1556 * the directory contents.
1557 */
1558 static int
1559 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1560 {
1561 const mzap_phys_t *mz;
1562 const mzap_ent_phys_t *mze;
1563 size_t size;
1564 int chunks, i, rc;
1565
1566 /*
1567 * Microzap objects use exactly one block. Read the whole
1568 * thing.
1569 */
1570 size = dnode->dn_datablkszsec * 512;
1571 mz = (const mzap_phys_t *) zap_scratch;
1572 chunks = size / MZAP_ENT_LEN - 1;
1573
1574 for (i = 0; i < chunks; i++) {
1575 mze = &mz->mz_chunk[i];
1576 if (mze->mze_name[0]) {
1577 rc = callback(mze->mze_name, mze->mze_value);
1578 if (rc != 0)
1579 return (rc);
1580 }
1581 }
1582
1583 return (0);
1584 }
1585
1586 /*
1587 * List a fatzap directory. Assumes that the zap scratch buffer contains
1588 * the directory header.
1589 */
1590 static int
1591 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1592 {
1593 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1594 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1595 fat_zap_t z;
1596 int i, j, rc;
1597
1598 if (zh.zap_magic != ZAP_MAGIC)
1599 return (EIO);
1600
1601 z.zap_block_shift = ilog2(bsize);
1602 z.zap_phys = (zap_phys_t *) zap_scratch;
1603
1604 /*
1605 * This assumes that the leaf blocks start at block 1. The
1606 * documentation isn't exactly clear on this.
1607 */
1608 zap_leaf_t zl;
1609 zl.l_bs = z.zap_block_shift;
1610 for (i = 0; i < zh.zap_num_leafs; i++) {
1611 off_t off = (i + 1) << zl.l_bs;
1612 char name[256], *p;
1613 uint64_t value;
1614
1615 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1616 return (EIO);
1617
1618 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1619
1620 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1621 zap_leaf_chunk_t *zc, *nc;
1622 int namelen;
1623
1624 zc = &ZAP_LEAF_CHUNK(&zl, j);
1625 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1626 continue;
1627 namelen = zc->l_entry.le_name_numints;
1628 if (namelen > sizeof(name))
1629 namelen = sizeof(name);
1630
1631 /*
1632 * Paste the name back together.
1633 */
1634 nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1635 p = name;
1636 while (namelen > 0) {
1637 int len;
1638 len = namelen;
1639 if (len > ZAP_LEAF_ARRAY_BYTES)
1640 len = ZAP_LEAF_ARRAY_BYTES;
1641 memcpy(p, nc->l_array.la_array, len);
1642 p += len;
1643 namelen -= len;
1644 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1645 }
1646
1647 /*
1648 * Assume the first eight bytes of the value are
1649 * a uint64_t.
1650 */
1651 value = fzap_leaf_value(&zl, zc);
1652
1653 //printf("%s 0x%jx\n", name, (uintmax_t)value);
1654 rc = callback((const char *)name, value);
1655 if (rc != 0)
1656 return (rc);
1657 }
1658 }
1659
1660 return (0);
1661 }
1662
1663 static int zfs_printf(const char *name, uint64_t value __unused)
1664 {
1665
1666 printf("%s\n", name);
1667
1668 return (0);
1669 }
1670
1671 /*
1672 * List a zap directory.
1673 */
1674 static int
1675 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
1676 {
1677 uint64_t zap_type;
1678 size_t size = dnode->dn_datablkszsec * 512;
1679
1680 if (dnode_read(spa, dnode, 0, zap_scratch, size))
1681 return (EIO);
1682
1683 zap_type = *(uint64_t *) zap_scratch;
1684 if (zap_type == ZBT_MICRO)
1685 return mzap_list(dnode, zfs_printf);
1686 else
1687 return fzap_list(spa, dnode, zfs_printf);
1688 }
1689
1690 static int
1691 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1692 {
1693 off_t offset;
1694
1695 offset = objnum * sizeof(dnode_phys_t);
1696 return dnode_read(spa, &os->os_meta_dnode, offset,
1697 dnode, sizeof(dnode_phys_t));
1698 }
1699
1700 static int
1701 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1702 {
1703 const mzap_phys_t *mz;
1704 const mzap_ent_phys_t *mze;
1705 size_t size;
1706 int chunks, i;
1707
1708 /*
1709 * Microzap objects use exactly one block. Read the whole
1710 * thing.
1711 */
1712 size = dnode->dn_datablkszsec * 512;
1713
1714 mz = (const mzap_phys_t *) zap_scratch;
1715 chunks = size / MZAP_ENT_LEN - 1;
1716
1717 for (i = 0; i < chunks; i++) {
1718 mze = &mz->mz_chunk[i];
1719 if (value == mze->mze_value) {
1720 strcpy(name, mze->mze_name);
1721 return (0);
1722 }
1723 }
1724
1725 return (ENOENT);
1726 }
1727
1728 static void
1729 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
1730 {
1731 size_t namelen;
1732 const zap_leaf_chunk_t *nc;
1733 char *p;
1734
1735 namelen = zc->l_entry.le_name_numints;
1736
1737 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1738 p = name;
1739 while (namelen > 0) {
1740 size_t len;
1741 len = namelen;
1742 if (len > ZAP_LEAF_ARRAY_BYTES)
1743 len = ZAP_LEAF_ARRAY_BYTES;
1744 memcpy(p, nc->l_array.la_array, len);
1745 p += len;
1746 namelen -= len;
1747 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1748 }
1749
1750 *p = '\0';
1751 }
1752
1753 static int
1754 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1755 {
1756 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1757 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1758 fat_zap_t z;
1759 int i, j;
1760
1761 if (zh.zap_magic != ZAP_MAGIC)
1762 return (EIO);
1763
1764 z.zap_block_shift = ilog2(bsize);
1765 z.zap_phys = (zap_phys_t *) zap_scratch;
1766
1767 /*
1768 * This assumes that the leaf blocks start at block 1. The
1769 * documentation isn't exactly clear on this.
1770 */
1771 zap_leaf_t zl;
1772 zl.l_bs = z.zap_block_shift;
1773 for (i = 0; i < zh.zap_num_leafs; i++) {
1774 off_t off = (i + 1) << zl.l_bs;
1775
1776 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1777 return (EIO);
1778
1779 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1780
1781 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1782 zap_leaf_chunk_t *zc;
1783
1784 zc = &ZAP_LEAF_CHUNK(&zl, j);
1785 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1786 continue;
1787 if (zc->l_entry.le_value_intlen != 8 ||
1788 zc->l_entry.le_value_numints != 1)
1789 continue;
1790
1791 if (fzap_leaf_value(&zl, zc) == value) {
1792 fzap_name_copy(&zl, zc, name);
1793 return (0);
1794 }
1795 }
1796 }
1797
1798 return (ENOENT);
1799 }
1800
1801 static int
1802 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1803 {
1804 int rc;
1805 uint64_t zap_type;
1806 size_t size = dnode->dn_datablkszsec * 512;
1807
1808 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1809 if (rc)
1810 return (rc);
1811
1812 zap_type = *(uint64_t *) zap_scratch;
1813 if (zap_type == ZBT_MICRO)
1814 return mzap_rlookup(spa, dnode, name, value);
1815 else
1816 return fzap_rlookup(spa, dnode, name, value);
1817 }
1818
1819 static int
1820 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
1821 {
1822 char name[256];
1823 char component[256];
1824 uint64_t dir_obj, parent_obj, child_dir_zapobj;
1825 dnode_phys_t child_dir_zap, dataset, dir, parent;
1826 dsl_dir_phys_t *dd;
1827 dsl_dataset_phys_t *ds;
1828 char *p;
1829 int len;
1830
1831 p = &name[sizeof(name) - 1];
1832 *p = '\0';
1833
1834 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1835 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1836 return (EIO);
1837 }
1838 ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
1839 dir_obj = ds->ds_dir_obj;
1840
1841 for (;;) {
1842 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
1843 return (EIO);
1844 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1845
1846 /* Actual loop condition. */
1847 parent_obj = dd->dd_parent_obj;
1848 if (parent_obj == 0)
1849 break;
1850
1851 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
1852 return (EIO);
1853 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
1854 child_dir_zapobj = dd->dd_child_dir_zapobj;
1855 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1856 return (EIO);
1857 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
1858 return (EIO);
1859
1860 len = strlen(component);
1861 p -= len;
1862 memcpy(p, component, len);
1863 --p;
1864 *p = '/';
1865
1866 /* Actual loop iteration. */
1867 dir_obj = parent_obj;
1868 }
1869
1870 if (*p != '\0')
1871 ++p;
1872 strcpy(result, p);
1873
1874 return (0);
1875 }
1876
1877 static int
1878 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
1879 {
1880 char element[256];
1881 uint64_t dir_obj, child_dir_zapobj;
1882 dnode_phys_t child_dir_zap, dir;
1883 dsl_dir_phys_t *dd;
1884 const char *p, *q;
1885
1886 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
1887 return (EIO);
1888 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &dir_obj))
1889 return (EIO);
1890
1891 p = name;
1892 for (;;) {
1893 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
1894 return (EIO);
1895 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1896
1897 while (*p == '/')
1898 p++;
1899 /* Actual loop condition #1. */
1900 if (*p == '\0')
1901 break;
1902
1903 q = strchr(p, '/');
1904 if (q) {
1905 memcpy(element, p, q - p);
1906 element[q - p] = '\0';
1907 p = q + 1;
1908 } else {
1909 strcpy(element, p);
1910 p += strlen(p);
1911 }
1912
1913 child_dir_zapobj = dd->dd_child_dir_zapobj;
1914 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1915 return (EIO);
1916
1917 /* Actual loop condition #2. */
1918 if (zap_lookup(spa, &child_dir_zap, element, &dir_obj) != 0)
1919 return (ENOENT);
1920 }
1921
1922 *objnum = dd->dd_head_dataset_obj;
1923 return (0);
1924 }
1925
1926 #pragma GCC diagnostic ignored "-Wstrict-aliasing"
1927 static int
1928 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
1929 {
1930 uint64_t dir_obj, child_dir_zapobj;
1931 dnode_phys_t child_dir_zap, dir, dataset;
1932 dsl_dataset_phys_t *ds;
1933 dsl_dir_phys_t *dd;
1934
1935 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1936 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1937 return (EIO);
1938 }
1939 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1940 dir_obj = ds->ds_dir_obj;
1941
1942 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
1943 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
1944 return (EIO);
1945 }
1946 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1947
1948 child_dir_zapobj = dd->dd_child_dir_zapobj;
1949 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
1950 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
1951 return (EIO);
1952 }
1953
1954 return (zap_list(spa, &child_dir_zap) != 0);
1955 }
1956
1957 int
1958 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *, uint64_t))
1959 {
1960 uint64_t dir_obj, child_dir_zapobj, zap_type;
1961 dnode_phys_t child_dir_zap, dir, dataset;
1962 dsl_dataset_phys_t *ds;
1963 dsl_dir_phys_t *dd;
1964 int err;
1965
1966 err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
1967 if (err != 0) {
1968 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
1969 return (err);
1970 }
1971 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1972 dir_obj = ds->ds_dir_obj;
1973
1974 err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
1975 if (err != 0) {
1976 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
1977 return (err);
1978 }
1979 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1980
1981 child_dir_zapobj = dd->dd_child_dir_zapobj;
1982 err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
1983 if (err != 0) {
1984 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
1985 return (err);
1986 }
1987
1988 err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
1989 if (err != 0)
1990 return (err);
1991
1992 zap_type = *(uint64_t *) zap_scratch;
1993 if (zap_type == ZBT_MICRO)
1994 return mzap_list(&child_dir_zap, callback);
1995 else
1996 return fzap_list(spa, &child_dir_zap, callback);
1997 }
1998
1999 /*
2000 * Find the object set given the object number of its dataset object
2001 * and return its details in *objset
2002 */
2003 static int
2004 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2005 {
2006 dnode_phys_t dataset;
2007 dsl_dataset_phys_t *ds;
2008
2009 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2010 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2011 return (EIO);
2012 }
2013
2014 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2015 if (zio_read(spa, &ds->ds_bp, objset)) {
2016 printf("ZFS: can't read object set for dataset %ju\n",
2017 (uintmax_t)objnum);
2018 return (EIO);
2019 }
2020
2021 return (0);
2022 }
2023
2024 /*
2025 * Find the object set pointed to by the BOOTFS property or the root
2026 * dataset if there is none and return its details in *objset
2027 */
2028 static int
2029 zfs_get_root(const spa_t *spa, uint64_t *objid)
2030 {
2031 dnode_phys_t dir, propdir;
2032 uint64_t props, bootfs, root;
2033
2034 *objid = 0;
2035
2036 /*
2037 * Start with the MOS directory object.
2038 */
2039 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
2040 printf("ZFS: can't read MOS object directory\n");
2041 return (EIO);
2042 }
2043
2044 /*
2045 * Lookup the pool_props and see if we can find a bootfs.
2046 */
2047 if (zap_lookup(spa, &dir, DMU_POOL_PROPS, &props) == 0
2048 && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
2049 && zap_lookup(spa, &propdir, "bootfs", &bootfs) == 0
2050 && bootfs != 0)
2051 {
2052 *objid = bootfs;
2053 return (0);
2054 }
2055 /*
2056 * Lookup the root dataset directory
2057 */
2058 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &root)
2059 || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
2060 printf("ZFS: can't find root dsl_dir\n");
2061 return (EIO);
2062 }
2063
2064 /*
2065 * Use the information from the dataset directory's bonus buffer
2066 * to find the dataset object and from that the object set itself.
2067 */
2068 dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
2069 *objid = dd->dd_head_dataset_obj;
2070 return (0);
2071 }
2072
2073 static int
2074 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mnt)
2075 {
2076
2077 mnt->spa = spa;
2078
2079 /*
2080 * Find the root object set if not explicitly provided
2081 */
2082 if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
2083 printf("ZFS: can't find root filesystem\n");
2084 return (EIO);
2085 }
2086
2087 if (zfs_mount_dataset(spa, rootobj, &mnt->objset)) {
2088 printf("ZFS: can't open root filesystem\n");
2089 return (EIO);
2090 }
2091
2092 mnt->rootobj = rootobj;
2093
2094 return (0);
2095 }
2096
2097 /*
2098 * callback function for feature name checks.
2099 */
2100 static int
2101 check_feature(const char *name, uint64_t value)
2102 {
2103 int i;
2104
2105 if (value == 0)
2106 return (0);
2107 if (name[0] == '\0')
2108 return (0);
2109
2110 for (i = 0; features_for_read[i] != NULL; i++) {
2111 if (strcmp(name, features_for_read[i]) == 0)
2112 return (0);
2113 }
2114 printf("ZFS: unsupported feature: %s\n", name);
2115 return (EIO);
2116 }
2117
2118 /*
2119 * Checks whether the MOS features that are active are supported.
2120 */
2121 static int
2122 check_mos_features(const spa_t *spa)
2123 {
2124 dnode_phys_t dir;
2125 uint64_t objnum, zap_type;
2126 size_t size;
2127 int rc;
2128
2129 if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
2130 &dir)) != 0)
2131 return (rc);
2132 if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
2133 &objnum)) != 0) {
2134 /*
2135 * It is older pool without features. As we have already
2136 * tested the label, just return without raising the error.
2137 */
2138 if (rc == ENOENT)
2139 rc = 0;
2140 return (rc);
2141 }
2142
2143 if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
2144 return (rc);
2145
2146 if (dir.dn_type != DMU_OTN_ZAP_METADATA)
2147 return (EIO);
2148
2149 size = dir.dn_datablkszsec * 512;
2150 if (dnode_read(spa, &dir, 0, zap_scratch, size))
2151 return (EIO);
2152
2153 zap_type = *(uint64_t *) zap_scratch;
2154 if (zap_type == ZBT_MICRO)
2155 rc = mzap_list(&dir, check_feature);
2156 else
2157 rc = fzap_list(spa, &dir, check_feature);
2158
2159 return (rc);
2160 }
2161
2162 static int
2163 zfs_spa_init(spa_t *spa)
2164 {
2165 int rc;
2166
2167 if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
2168 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
2169 return (EIO);
2170 }
2171 if (spa->spa_mos.os_type != DMU_OST_META) {
2172 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
2173 return (EIO);
2174 }
2175
2176 rc = check_mos_features(spa);
2177 if (rc != 0) {
2178 printf("ZFS: pool %s is not supported\n", spa->spa_name);
2179 }
2180
2181 return (rc);
2182 }
2183
2184 static int
2185 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
2186 {
2187
2188 if (dn->dn_bonustype != DMU_OT_SA) {
2189 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
2190
2191 sb->st_mode = zp->zp_mode;
2192 sb->st_uid = zp->zp_uid;
2193 sb->st_gid = zp->zp_gid;
2194 sb->st_size = zp->zp_size;
2195 } else {
2196 sa_hdr_phys_t *sahdrp;
2197 int hdrsize;
2198 size_t size = 0;
2199 void *buf = NULL;
2200
2201 if (dn->dn_bonuslen != 0)
2202 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2203 else {
2204 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
2205 blkptr_t *bp = DN_SPILL_BLKPTR(dn);
2206 int error;
2207
2208 size = BP_GET_LSIZE(bp);
2209 buf = zfs_alloc(size);
2210 error = zio_read(spa, bp, buf);
2211 if (error != 0) {
2212 zfs_free(buf, size);
2213 return (error);
2214 }
2215 sahdrp = buf;
2216 } else {
2217 return (EIO);
2218 }
2219 }
2220 hdrsize = SA_HDR_SIZE(sahdrp);
2221 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
2222 SA_MODE_OFFSET);
2223 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
2224 SA_UID_OFFSET);
2225 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
2226 SA_GID_OFFSET);
2227 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
2228 SA_SIZE_OFFSET);
2229 if (buf != NULL)
2230 zfs_free(buf, size);
2231 }
2232
2233 return (0);
2234 }
2235
2236 static int
2237 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
2238 {
2239 int rc = 0;
2240
2241 if (dn->dn_bonustype == DMU_OT_SA) {
2242 sa_hdr_phys_t *sahdrp = NULL;
2243 size_t size = 0;
2244 void *buf = NULL;
2245 int hdrsize;
2246 char *p;
2247
2248 if (dn->dn_bonuslen != 0)
2249 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2250 else {
2251 blkptr_t *bp;
2252
2253 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
2254 return (EIO);
2255 bp = DN_SPILL_BLKPTR(dn);
2256
2257 size = BP_GET_LSIZE(bp);
2258 buf = zfs_alloc(size);
2259 rc = zio_read(spa, bp, buf);
2260 if (rc != 0) {
2261 zfs_free(buf, size);
2262 return (rc);
2263 }
2264 sahdrp = buf;
2265 }
2266 hdrsize = SA_HDR_SIZE(sahdrp);
2267 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
2268 memcpy(path, p, psize);
2269 if (buf != NULL)
2270 zfs_free(buf, size);
2271 return (0);
2272 }
2273 /*
2274 * Second test is purely to silence bogus compiler
2275 * warning about accessing past the end of dn_bonus.
2276 */
2277 if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
2278 sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
2279 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
2280 } else {
2281 rc = dnode_read(spa, dn, 0, path, psize);
2282 }
2283 return (rc);
2284 }
2285
2286 struct obj_list {
2287 uint64_t objnum;
2288 STAILQ_ENTRY(obj_list) entry;
2289 };
2290
2291 /*
2292 * Lookup a file and return its dnode.
2293 */
2294 static int
2295 zfs_lookup(const struct zfsmount *mnt, const char *upath, dnode_phys_t *dnode)
2296 {
2297 int rc;
2298 uint64_t objnum;
2299 const spa_t *spa;
2300 dnode_phys_t dn;
2301 const char *p, *q;
2302 char element[256];
2303 char path[1024];
2304 int symlinks_followed = 0;
2305 struct stat sb;
2306 struct obj_list *entry, *tentry;
2307 STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
2308
2309 spa = mnt->spa;
2310 if (mnt->objset.os_type != DMU_OST_ZFS) {
2311 printf("ZFS: unexpected object set type %ju\n",
2312 (uintmax_t)mnt->objset.os_type);
2313 return (EIO);
2314 }
2315
2316 if ((entry = malloc(sizeof(struct obj_list))) == NULL)
2317 return (ENOMEM);
2318
2319 /*
2320 * Get the root directory dnode.
2321 */
2322 rc = objset_get_dnode(spa, &mnt->objset, MASTER_NODE_OBJ, &dn);
2323 if (rc) {
2324 free(entry);
2325 return (rc);
2326 }
2327
2328 rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, &objnum);
2329 if (rc) {
2330 free(entry);
2331 return (rc);
2332 }
2333 entry->objnum = objnum;
2334 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2335
2336 rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2337 if (rc != 0)
2338 goto done;
2339
2340 p = upath;
2341 while (p && *p) {
2342 rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2343 if (rc != 0)
2344 goto done;
2345
2346 while (*p == '/')
2347 p++;
2348 if (*p == '\0')
2349 break;
2350 q = p;
2351 while (*q != '\0' && *q != '/')
2352 q++;
2353
2354 /* skip dot */
2355 if (p + 1 == q && p[0] == '.') {
2356 p++;
2357 continue;
2358 }
2359 /* double dot */
2360 if (p + 2 == q && p[0] == '.' && p[1] == '.') {
2361 p += 2;
2362 if (STAILQ_FIRST(&on_cache) ==
2363 STAILQ_LAST(&on_cache, obj_list, entry)) {
2364 rc = ENOENT;
2365 goto done;
2366 }
2367 entry = STAILQ_FIRST(&on_cache);
2368 STAILQ_REMOVE_HEAD(&on_cache, entry);
2369 free(entry);
2370 objnum = (STAILQ_FIRST(&on_cache))->objnum;
2371 continue;
2372 }
2373 if (q - p + 1 > sizeof(element)) {
2374 rc = ENAMETOOLONG;
2375 goto done;
2376 }
2377 memcpy(element, p, q - p);
2378 element[q - p] = 0;
2379 p = q;
2380
2381 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
2382 goto done;
2383 if (!S_ISDIR(sb.st_mode)) {
2384 rc = ENOTDIR;
2385 goto done;
2386 }
2387
2388 rc = zap_lookup(spa, &dn, element, &objnum);
2389 if (rc)
2390 goto done;
2391 objnum = ZFS_DIRENT_OBJ(objnum);
2392
2393 if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
2394 rc = ENOMEM;
2395 goto done;
2396 }
2397 entry->objnum = objnum;
2398 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2399 rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2400 if (rc)
2401 goto done;
2402
2403 /*
2404 * Check for symlink.
2405 */
2406 rc = zfs_dnode_stat(spa, &dn, &sb);
2407 if (rc)
2408 goto done;
2409 if (S_ISLNK(sb.st_mode)) {
2410 if (symlinks_followed > 10) {
2411 rc = EMLINK;
2412 goto done;
2413 }
2414 symlinks_followed++;
2415
2416 /*
2417 * Read the link value and copy the tail of our
2418 * current path onto the end.
2419 */
2420 if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
2421 rc = ENAMETOOLONG;
2422 goto done;
2423 }
2424 strcpy(&path[sb.st_size], p);
2425
2426 rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
2427 if (rc != 0)
2428 goto done;
2429
2430 /*
2431 * Restart with the new path, starting either at
2432 * the root or at the parent depending whether or
2433 * not the link is relative.
2434 */
2435 p = path;
2436 if (*p == '/') {
2437 while (STAILQ_FIRST(&on_cache) !=
2438 STAILQ_LAST(&on_cache, obj_list, entry)) {
2439 entry = STAILQ_FIRST(&on_cache);
2440 STAILQ_REMOVE_HEAD(&on_cache, entry);
2441 free(entry);
2442 }
2443 } else {
2444 entry = STAILQ_FIRST(&on_cache);
2445 STAILQ_REMOVE_HEAD(&on_cache, entry);
2446 free(entry);
2447 }
2448 objnum = (STAILQ_FIRST(&on_cache))->objnum;
2449 }
2450 }
2451
2452 *dnode = dn;
2453 done:
2454 STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
2455 free(entry);
2456 return (rc);
2457 }
Unleashed OS