?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
loader: add skein/edonr support
[unleashed.git] / usr / src / boot / sys / boot / zfs / zfsimpl.c
blobb3edf4653a3c075e3bdadc27eca72dd882165d9a
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.illumos:skein",
62 "org.illumos:edonr",
63 "org.zfsonlinux:large_dnode",
64 "com.joyent:multi_vdev_crash_dump",
65 NULL
66 };
67
68 /*
69 * List of all pools, chained through spa_link.
70 */
71 static spa_list_t zfs_pools;
72
73 static const dnode_phys_t *dnode_cache_obj;
74 static uint64_t dnode_cache_bn;
75 static char *dnode_cache_buf;
76 static char *zap_scratch;
77 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
78
79 #define TEMP_SIZE (1024 * 1024)
80
81 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
82 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
83 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
84 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
85 const char *name, uint64_t integer_size, uint64_t num_integers,
86 void *value);
87
88 static void
89 zfs_init(void)
90 {
91 STAILQ_INIT(&zfs_vdevs);
92 STAILQ_INIT(&zfs_pools);
93
94 zfs_temp_buf = malloc(TEMP_SIZE);
95 zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
96 zfs_temp_ptr = zfs_temp_buf;
97 dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
98 zap_scratch = malloc(SPA_MAXBLOCKSIZE);
99
100 zfs_init_crc();
101 }
102
103 static void *
104 zfs_alloc(size_t size)
105 {
106 char *ptr;
107
108 if (zfs_temp_ptr + size > zfs_temp_end) {
109 printf("ZFS: out of temporary buffer space\n");
110 for (;;) ;
111 }
112 ptr = zfs_temp_ptr;
113 zfs_temp_ptr += size;
114
115 return (ptr);
116 }
117
118 static void
119 zfs_free(void *ptr, size_t size)
120 {
121
122 zfs_temp_ptr -= size;
123 if (zfs_temp_ptr != ptr) {
124 printf("ZFS: zfs_alloc()/zfs_free() mismatch\n");
125 for (;;) ;
126 }
127 }
128
129 static int
130 xdr_int(const unsigned char **xdr, int *ip)
131 {
132 *ip = ((*xdr)[0] << 24)
133 | ((*xdr)[1] << 16)
134 | ((*xdr)[2] << 8)
135 | ((*xdr)[3] << 0);
136 (*xdr) += 4;
137 return (0);
138 }
139
140 static int
141 xdr_u_int(const unsigned char **xdr, u_int *ip)
142 {
143 *ip = ((*xdr)[0] << 24)
144 | ((*xdr)[1] << 16)
145 | ((*xdr)[2] << 8)
146 | ((*xdr)[3] << 0);
147 (*xdr) += 4;
148 return (0);
149 }
150
151 static int
152 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
153 {
154 u_int hi, lo;
155
156 xdr_u_int(xdr, &hi);
157 xdr_u_int(xdr, &lo);
158 *lp = (((uint64_t) hi) << 32) | lo;
159 return (0);
160 }
161
162 static int
163 nvlist_find(const unsigned char *nvlist, const char *name, int type,
164 int* elementsp, void *valuep)
165 {
166 const unsigned char *p, *pair;
167 int junk;
168 int encoded_size, decoded_size;
169
170 p = nvlist;
171 xdr_int(&p, &junk);
172 xdr_int(&p, &junk);
173
174 pair = p;
175 xdr_int(&p, &encoded_size);
176 xdr_int(&p, &decoded_size);
177 while (encoded_size && decoded_size) {
178 int namelen, pairtype, elements;
179 const char *pairname;
180
181 xdr_int(&p, &namelen);
182 pairname = (const char*) p;
183 p += roundup(namelen, 4);
184 xdr_int(&p, &pairtype);
185
186 if (!memcmp(name, pairname, namelen) && type == pairtype) {
187 xdr_int(&p, &elements);
188 if (elementsp)
189 *elementsp = elements;
190 if (type == DATA_TYPE_UINT64) {
191 xdr_uint64_t(&p, (uint64_t *) valuep);
192 return (0);
193 } else if (type == DATA_TYPE_STRING) {
194 int len;
195 xdr_int(&p, &len);
196 (*(const char**) valuep) = (const char*) p;
197 return (0);
198 } else if (type == DATA_TYPE_NVLIST
199 || type == DATA_TYPE_NVLIST_ARRAY) {
200 (*(const unsigned char**) valuep) =
201 (const unsigned char*) p;
202 return (0);
203 } else {
204 return (EIO);
205 }
206 } else {
207 /*
208 * Not the pair we are looking for, skip to the next one.
209 */
210 p = pair + encoded_size;
211 }
212
213 pair = p;
214 xdr_int(&p, &encoded_size);
215 xdr_int(&p, &decoded_size);
216 }
217
218 return (EIO);
219 }
220
221 static int
222 nvlist_check_features_for_read(const unsigned char *nvlist)
223 {
224 const unsigned char *p, *pair;
225 int junk;
226 int encoded_size, decoded_size;
227 int rc;
228
229 rc = 0;
230
231 p = nvlist;
232 xdr_int(&p, &junk);
233 xdr_int(&p, &junk);
234
235 pair = p;
236 xdr_int(&p, &encoded_size);
237 xdr_int(&p, &decoded_size);
238 while (encoded_size && decoded_size) {
239 int namelen, pairtype;
240 const char *pairname;
241 int i, found;
242
243 found = 0;
244
245 xdr_int(&p, &namelen);
246 pairname = (const char*) p;
247 p += roundup(namelen, 4);
248 xdr_int(&p, &pairtype);
249
250 for (i = 0; features_for_read[i] != NULL; i++) {
251 if (!memcmp(pairname, features_for_read[i], namelen)) {
252 found = 1;
253 break;
254 }
255 }
256
257 if (!found) {
258 printf("ZFS: unsupported feature: %s\n", pairname);
259 rc = EIO;
260 }
261
262 p = pair + encoded_size;
263
264 pair = p;
265 xdr_int(&p, &encoded_size);
266 xdr_int(&p, &decoded_size);
267 }
268
269 return (rc);
270 }
271
272 /*
273 * Return the next nvlist in an nvlist array.
274 */
275 static const unsigned char *
276 nvlist_next(const unsigned char *nvlist)
277 {
278 const unsigned char *p, *pair;
279 int junk;
280 int encoded_size, decoded_size;
281
282 p = nvlist;
283 xdr_int(&p, &junk);
284 xdr_int(&p, &junk);
285
286 pair = p;
287 xdr_int(&p, &encoded_size);
288 xdr_int(&p, &decoded_size);
289 while (encoded_size && decoded_size) {
290 p = pair + encoded_size;
291
292 pair = p;
293 xdr_int(&p, &encoded_size);
294 xdr_int(&p, &decoded_size);
295 }
296
297 return p;
298 }
299
300 #ifdef TEST
301
302 static const unsigned char *
303 nvlist_print(const unsigned char *nvlist, unsigned int indent)
304 {
305 static const char* typenames[] = {
306 "DATA_TYPE_UNKNOWN",
307 "DATA_TYPE_BOOLEAN",
308 "DATA_TYPE_BYTE",
309 "DATA_TYPE_INT16",
310 "DATA_TYPE_UINT16",
311 "DATA_TYPE_INT32",
312 "DATA_TYPE_UINT32",
313 "DATA_TYPE_INT64",
314 "DATA_TYPE_UINT64",
315 "DATA_TYPE_STRING",
316 "DATA_TYPE_BYTE_ARRAY",
317 "DATA_TYPE_INT16_ARRAY",
318 "DATA_TYPE_UINT16_ARRAY",
319 "DATA_TYPE_INT32_ARRAY",
320 "DATA_TYPE_UINT32_ARRAY",
321 "DATA_TYPE_INT64_ARRAY",
322 "DATA_TYPE_UINT64_ARRAY",
323 "DATA_TYPE_STRING_ARRAY",
324 "DATA_TYPE_HRTIME",
325 "DATA_TYPE_NVLIST",
326 "DATA_TYPE_NVLIST_ARRAY",
327 "DATA_TYPE_BOOLEAN_VALUE",
328 "DATA_TYPE_INT8",
329 "DATA_TYPE_UINT8",
330 "DATA_TYPE_BOOLEAN_ARRAY",
331 "DATA_TYPE_INT8_ARRAY",
332 "DATA_TYPE_UINT8_ARRAY"
333 };
334
335 unsigned int i, j;
336 const unsigned char *p, *pair;
337 int junk;
338 int encoded_size, decoded_size;
339
340 p = nvlist;
341 xdr_int(&p, &junk);
342 xdr_int(&p, &junk);
343
344 pair = p;
345 xdr_int(&p, &encoded_size);
346 xdr_int(&p, &decoded_size);
347 while (encoded_size && decoded_size) {
348 int namelen, pairtype, elements;
349 const char *pairname;
350
351 xdr_int(&p, &namelen);
352 pairname = (const char*) p;
353 p += roundup(namelen, 4);
354 xdr_int(&p, &pairtype);
355
356 for (i = 0; i < indent; i++)
357 printf(" ");
358 printf("%s %s", typenames[pairtype], pairname);
359
360 xdr_int(&p, &elements);
361 switch (pairtype) {
362 case DATA_TYPE_UINT64: {
363 uint64_t val;
364 xdr_uint64_t(&p, &val);
365 printf(" = 0x%jx\n", (uintmax_t)val);
366 break;
367 }
368
369 case DATA_TYPE_STRING: {
370 int len;
371 xdr_int(&p, &len);
372 printf(" = \"%s\"\n", p);
373 break;
374 }
375
376 case DATA_TYPE_NVLIST:
377 printf("\n");
378 nvlist_print(p, indent + 1);
379 break;
380
381 case DATA_TYPE_NVLIST_ARRAY:
382 for (j = 0; j < elements; j++) {
383 printf("[%d]\n", j);
384 p = nvlist_print(p, indent + 1);
385 if (j != elements - 1) {
386 for (i = 0; i < indent; i++)
387 printf(" ");
388 printf("%s %s", typenames[pairtype], pairname);
389 }
390 }
391 break;
392
393 default:
394 printf("\n");
395 }
396
397 p = pair + encoded_size;
398
399 pair = p;
400 xdr_int(&p, &encoded_size);
401 xdr_int(&p, &decoded_size);
402 }
403
404 return p;
405 }
406
407 #endif
408
409 static int
410 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
411 off_t offset, size_t size)
412 {
413 size_t psize;
414 int rc;
415
416 if (!vdev->v_phys_read)
417 return (EIO);
418
419 if (bp) {
420 psize = BP_GET_PSIZE(bp);
421 } else {
422 psize = size;
423 }
424
425 /*printf("ZFS: reading %zu bytes at 0x%jx to %p\n", psize, (uintmax_t)offset, buf);*/
426 rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
427 if (rc)
428 return (rc);
429 if (bp && zio_checksum_verify(vdev->spa, bp, buf))
430 return (EIO);
431
432 return (0);
433 }
434
435 static int
436 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
437 off_t offset, size_t bytes)
438 {
439
440 return (vdev_read_phys(vdev, bp, buf,
441 offset + VDEV_LABEL_START_SIZE, bytes));
442 }
443
444
445 static int
446 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
447 off_t offset, size_t bytes)
448 {
449 vdev_t *kid;
450 int rc;
451
452 rc = EIO;
453 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
454 if (kid->v_state != VDEV_STATE_HEALTHY)
455 continue;
456 rc = kid->v_read(kid, bp, buf, offset, bytes);
457 if (!rc)
458 return (0);
459 }
460
461 return (rc);
462 }
463
464 static int
465 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
466 off_t offset, size_t bytes)
467 {
468 vdev_t *kid;
469
470 /*
471 * Here we should have two kids:
472 * First one which is the one we are replacing and we can trust
473 * only this one to have valid data, but it might not be present.
474 * Second one is that one we are replacing with. It is most likely
475 * healthy, but we can't trust it has needed data, so we won't use it.
476 */
477 kid = STAILQ_FIRST(&vdev->v_children);
478 if (kid == NULL)
479 return (EIO);
480 if (kid->v_state != VDEV_STATE_HEALTHY)
481 return (EIO);
482 return (kid->v_read(kid, bp, buf, offset, bytes));
483 }
484
485 static vdev_t *
486 vdev_find(uint64_t guid)
487 {
488 vdev_t *vdev;
489
490 STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
491 if (vdev->v_guid == guid)
492 return (vdev);
493
494 return (0);
495 }
496
497 static vdev_t *
498 vdev_create(uint64_t guid, vdev_read_t *vdev_read)
499 {
500 vdev_t *vdev;
501
502 vdev = malloc(sizeof(vdev_t));
503 memset(vdev, 0, sizeof(vdev_t));
504 STAILQ_INIT(&vdev->v_children);
505 vdev->v_guid = guid;
506 vdev->v_state = VDEV_STATE_OFFLINE;
507 vdev->v_read = vdev_read;
508 vdev->v_phys_read = 0;
509 vdev->v_read_priv = 0;
510 STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
511
512 return (vdev);
513 }
514
515 static int
516 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t *pvdev,
517 vdev_t **vdevp, int is_newer)
518 {
519 int rc;
520 uint64_t guid, id, ashift, nparity;
521 const char *type;
522 const char *path;
523 vdev_t *vdev, *kid;
524 const unsigned char *kids;
525 int nkids, i, is_new;
526 uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
527
528 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
529 NULL, &guid) ||
530 nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id) ||
531 nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
532 NULL, &type)) {
533 printf("ZFS: can't find vdev details\n");
534 return (ENOENT);
535 }
536
537 if (strcmp(type, VDEV_TYPE_MIRROR)
538 && strcmp(type, VDEV_TYPE_DISK)
539 #ifdef ZFS_TEST
540 && strcmp(type, VDEV_TYPE_FILE)
541 #endif
542 && strcmp(type, VDEV_TYPE_RAIDZ)
543 && strcmp(type, VDEV_TYPE_REPLACING)) {
544 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
545 return (EIO);
546 }
547
548 is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
549
550 nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
551 &is_offline);
552 nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
553 &is_removed);
554 nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
555 &is_faulted);
556 nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, NULL,
557 &is_degraded);
558 nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64, NULL,
559 &isnt_present);
560
561 vdev = vdev_find(guid);
562 if (!vdev) {
563 is_new = 1;
564
565 if (!strcmp(type, VDEV_TYPE_MIRROR))
566 vdev = vdev_create(guid, vdev_mirror_read);
567 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
568 vdev = vdev_create(guid, vdev_raidz_read);
569 else if (!strcmp(type, VDEV_TYPE_REPLACING))
570 vdev = vdev_create(guid, vdev_replacing_read);
571 else
572 vdev = vdev_create(guid, vdev_disk_read);
573
574 vdev->v_id = id;
575 vdev->v_top = pvdev != NULL ? pvdev : vdev;
576 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
577 DATA_TYPE_UINT64, NULL, &ashift) == 0) {
578 vdev->v_ashift = ashift;
579 } else {
580 vdev->v_ashift = 0;
581 }
582 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
583 DATA_TYPE_UINT64, NULL, &nparity) == 0) {
584 vdev->v_nparity = nparity;
585 } else {
586 vdev->v_nparity = 0;
587 }
588 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
589 DATA_TYPE_STRING, NULL, &path) == 0) {
590 if (strncmp(path, "/dev/dsk/", 9) == 0)
591 path += 9;
592 vdev->v_name = strdup(path);
593 if (nvlist_find(nvlist, ZPOOL_CONFIG_PHYS_PATH,
594 DATA_TYPE_STRING, NULL, &path) == 0) {
595 vdev->v_phys_path = strdup(path);
596 } else {
597 vdev->v_phys_path = NULL;
598 }
599 if (nvlist_find(nvlist, ZPOOL_CONFIG_DEVID,
600 DATA_TYPE_STRING, NULL, &path) == 0) {
601 vdev->v_devid = strdup(path);
602 } else {
603 vdev->v_devid = NULL;
604 }
605 } else {
606 if (!strcmp(type, "raidz")) {
607 if (vdev->v_nparity == 1)
608 vdev->v_name = "raidz1";
609 else if (vdev->v_nparity == 2)
610 vdev->v_name = "raidz2";
611 else if (vdev->v_nparity == 3)
612 vdev->v_name = "raidz3";
613 else {
614 printf("ZFS: can only boot from disk, mirror, raidz1, raidz2 and raidz3 vdevs\n");
615 return (EIO);
616 }
617 } else {
618 vdev->v_name = strdup(type);
619 }
620 }
621 } else {
622 is_new = 0;
623 }
624
625 if (is_new || is_newer) {
626 /*
627 * This is either new vdev or we've already seen this vdev,
628 * but from an older vdev label, so let's refresh its state
629 * from the newer label.
630 */
631 if (is_offline)
632 vdev->v_state = VDEV_STATE_OFFLINE;
633 else if (is_removed)
634 vdev->v_state = VDEV_STATE_REMOVED;
635 else if (is_faulted)
636 vdev->v_state = VDEV_STATE_FAULTED;
637 else if (is_degraded)
638 vdev->v_state = VDEV_STATE_DEGRADED;
639 else if (isnt_present)
640 vdev->v_state = VDEV_STATE_CANT_OPEN;
641 }
642
643 rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
644 &nkids, &kids);
645 /*
646 * Its ok if we don't have any kids.
647 */
648 if (rc == 0) {
649 vdev->v_nchildren = nkids;
650 for (i = 0; i < nkids; i++) {
651 rc = vdev_init_from_nvlist(kids, vdev, &kid, is_newer);
652 if (rc)
653 return (rc);
654 if (is_new)
655 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
656 v_childlink);
657 kids = nvlist_next(kids);
658 }
659 } else {
660 vdev->v_nchildren = 0;
661 }
662
663 if (vdevp)
664 *vdevp = vdev;
665 return (0);
666 }
667
668 static void
669 vdev_set_state(vdev_t *vdev)
670 {
671 vdev_t *kid;
672 int good_kids;
673 int bad_kids;
674
675 /*
676 * A mirror or raidz is healthy if all its kids are healthy. A
677 * mirror is degraded if any of its kids is healthy; a raidz
678 * is degraded if at most nparity kids are offline.
679 */
680 if (STAILQ_FIRST(&vdev->v_children)) {
681 good_kids = 0;
682 bad_kids = 0;
683 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
684 if (kid->v_state == VDEV_STATE_HEALTHY)
685 good_kids++;
686 else
687 bad_kids++;
688 }
689 if (bad_kids == 0) {
690 vdev->v_state = VDEV_STATE_HEALTHY;
691 } else {
692 if (vdev->v_read == vdev_mirror_read) {
693 if (good_kids) {
694 vdev->v_state = VDEV_STATE_DEGRADED;
695 } else {
696 vdev->v_state = VDEV_STATE_OFFLINE;
697 }
698 } else if (vdev->v_read == vdev_raidz_read) {
699 if (bad_kids > vdev->v_nparity) {
700 vdev->v_state = VDEV_STATE_OFFLINE;
701 } else {
702 vdev->v_state = VDEV_STATE_DEGRADED;
703 }
704 }
705 }
706 }
707 }
708
709 static spa_t *
710 spa_find_by_guid(uint64_t guid)
711 {
712 spa_t *spa;
713
714 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
715 if (spa->spa_guid == guid)
716 return (spa);
717
718 return (0);
719 }
720
721 static spa_t *
722 spa_find_by_name(const char *name)
723 {
724 spa_t *spa;
725
726 STAILQ_FOREACH(spa, &zfs_pools, spa_link)
727 if (!strcmp(spa->spa_name, name))
728 return (spa);
729
730 return (0);
731 }
732
733 spa_t *
734 spa_get_primary(void)
735 {
736 return (STAILQ_FIRST(&zfs_pools));
737 }
738
739 vdev_t *
740 spa_get_primary_vdev(const spa_t *spa)
741 {
742 vdev_t *vdev;
743 vdev_t *kid;
744
745 if (spa == NULL)
746 spa = spa_get_primary();
747 if (spa == NULL)
748 return (NULL);
749 vdev = STAILQ_FIRST(&spa->spa_vdevs);
750 if (vdev == NULL)
751 return (NULL);
752 for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
753 kid = STAILQ_FIRST(&vdev->v_children))
754 vdev = kid;
755 return (vdev);
756 }
757
758 static spa_t *
759 spa_create(uint64_t guid, const char *name)
760 {
761 spa_t *spa;
762
763 if ((spa = malloc(sizeof(spa_t))) == NULL)
764 return (NULL);
765 memset(spa, 0, sizeof(spa_t));
766 if ((spa->spa_name = strdup(name)) == NULL) {
767 free(spa);
768 return (NULL);
769 }
770 STAILQ_INIT(&spa->spa_vdevs);
771 spa->spa_guid = guid;
772 STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
773
774 return (spa);
775 }
776
777 static const char *
778 state_name(vdev_state_t state)
779 {
780 static const char* names[] = {
781 "UNKNOWN",
782 "CLOSED",
783 "OFFLINE",
784 "REMOVED",
785 "CANT_OPEN",
786 "FAULTED",
787 "DEGRADED",
788 "ONLINE"
789 };
790 return names[state];
791 }
792
793 static int
794 pager_printf(const char *fmt, ...)
795 {
796 char line[80];
797 va_list args;
798
799 va_start(args, fmt);
800 vsnprintf(line, sizeof (line), fmt, args);
801 va_end(args);
802 return (pager_output(line));
803 }
804
805 #define STATUS_FORMAT " %s %s\n"
806
807 static int
808 print_state(int indent, const char *name, vdev_state_t state)
809 {
810 int i;
811 char buf[512];
812
813 buf[0] = 0;
814 for (i = 0; i < indent; i++)
815 strcat(buf, " ");
816 strcat(buf, name);
817 return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
818 }
819
820 static int
821 vdev_status(vdev_t *vdev, int indent)
822 {
823 vdev_t *kid;
824 int ret;
825 ret = print_state(indent, vdev->v_name, vdev->v_state);
826 if (ret != 0)
827 return (ret);
828
829 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
830 ret = vdev_status(kid, indent + 1);
831 if (ret != 0)
832 return (ret);
833 }
834 return (ret);
835 }
836
837 static int
838 spa_status(spa_t *spa)
839 {
840 static char bootfs[ZFS_MAXNAMELEN];
841 uint64_t rootid;
842 vdev_t *vdev;
843 int good_kids, bad_kids, degraded_kids, ret;
844 vdev_state_t state;
845
846 ret = pager_printf(" pool: %s\n", spa->spa_name);
847 if (ret != 0)
848 return (ret);
849
850 if (zfs_get_root(spa, &rootid) == 0 &&
851 zfs_rlookup(spa, rootid, bootfs) == 0) {
852 if (bootfs[0] == '\0')
853 ret = pager_printf("bootfs: %s\n", spa->spa_name);
854 else
855 ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
856 bootfs);
857 if (ret != 0)
858 return (ret);
859 }
860 ret = pager_printf("config:\n\n");
861 if (ret != 0)
862 return (ret);
863 ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
864 if (ret != 0)
865 return (ret);
866
867 good_kids = 0;
868 degraded_kids = 0;
869 bad_kids = 0;
870 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
871 if (vdev->v_state == VDEV_STATE_HEALTHY)
872 good_kids++;
873 else if (vdev->v_state == VDEV_STATE_DEGRADED)
874 degraded_kids++;
875 else
876 bad_kids++;
877 }
878
879 state = VDEV_STATE_CLOSED;
880 if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
881 state = VDEV_STATE_HEALTHY;
882 else if ((good_kids + degraded_kids) > 0)
883 state = VDEV_STATE_DEGRADED;
884
885 ret = print_state(0, spa->spa_name, state);
886 if (ret != 0)
887 return (ret);
888 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
889 ret = vdev_status(vdev, 1);
890 if (ret != 0)
891 return (ret);
892 }
893 return (ret);
894 }
895
896 int
897 spa_all_status(void)
898 {
899 spa_t *spa;
900 int first = 1, ret = 0;
901
902 STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
903 if (!first) {
904 ret = pager_printf("\n");
905 if (ret != 0)
906 return (ret);
907 }
908 first = 0;
909 ret = spa_status(spa);
910 if (ret != 0)
911 return (ret);
912 }
913 return (ret);
914 }
915
916 uint64_t
917 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
918 {
919 uint64_t label_offset;
920
921 if (l < VDEV_LABELS / 2)
922 label_offset = 0;
923 else
924 label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
925
926 return (offset + l * sizeof (vdev_label_t) + label_offset);
927 }
928
929 static int
930 vdev_probe(vdev_phys_read_t *phys_read, void *read_priv, spa_t **spap)
931 {
932 vdev_t vtmp;
933 vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
934 vdev_phys_t *tmp_label;
935 spa_t *spa;
936 vdev_t *vdev, *top_vdev, *pool_vdev;
937 off_t off;
938 blkptr_t bp;
939 const unsigned char *nvlist = NULL;
940 uint64_t val;
941 uint64_t guid;
942 uint64_t best_txg = 0;
943 uint64_t pool_txg, pool_guid;
944 uint64_t psize;
945 const char *pool_name;
946 const unsigned char *vdevs;
947 const unsigned char *features;
948 int i, l, rc, is_newer;
949 char *upbuf;
950 const struct uberblock *up;
951
952 /*
953 * Load the vdev label and figure out which
954 * uberblock is most current.
955 */
956 memset(&vtmp, 0, sizeof(vtmp));
957 vtmp.v_phys_read = phys_read;
958 vtmp.v_read_priv = read_priv;
959 psize = P2ALIGN(ldi_get_size(read_priv),
960 (uint64_t)sizeof (vdev_label_t));
961
962 /* Test for minimum device size. */
963 if (psize < SPA_MINDEVSIZE)
964 return (EIO);
965
966 tmp_label = zfs_alloc(sizeof (vdev_phys_t));
967
968 for (l = 0; l < VDEV_LABELS; l++) {
969 off = vdev_label_offset(psize, l,
970 offsetof(vdev_label_t, vl_vdev_phys));
971
972 BP_ZERO(&bp);
973 BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
974 BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
975 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
976 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
977 DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
978 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
979
980 if (vdev_read_phys(&vtmp, &bp, tmp_label, off, 0))
981 continue;
982
983 if (tmp_label->vp_nvlist[0] != NV_ENCODE_XDR)
984 continue;
985
986 nvlist = (const unsigned char *) tmp_label->vp_nvlist + 4;
987 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
988 DATA_TYPE_UINT64, NULL, &pool_txg) != 0)
989 continue;
990
991 if (best_txg <= pool_txg) {
992 best_txg = pool_txg;
993 memcpy(vdev_label, tmp_label, sizeof (vdev_phys_t));
994 }
995 }
996
997 zfs_free(tmp_label, sizeof (vdev_phys_t));
998
999 if (best_txg == 0)
1000 return (EIO);
1001
1002 if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR)
1003 return (EIO);
1004
1005 nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
1006
1007 if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1008 NULL, &val) != 0) {
1009 return (EIO);
1010 }
1011
1012 if (!SPA_VERSION_IS_SUPPORTED(val)) {
1013 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1014 (unsigned) val, (unsigned) SPA_VERSION);
1015 return (EIO);
1016 }
1017
1018 /* Check ZFS features for read */
1019 if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1020 DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1021 nvlist_check_features_for_read(features) != 0) {
1022 return (EIO);
1023 }
1024
1025 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1026 NULL, &val) != 0) {
1027 return (EIO);
1028 }
1029
1030 if (val == POOL_STATE_DESTROYED) {
1031 /* We don't boot only from destroyed pools. */
1032 return (EIO);
1033 }
1034
1035 if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1036 NULL, &pool_txg) != 0 ||
1037 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1038 NULL, &pool_guid) != 0 ||
1039 nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1040 NULL, &pool_name) != 0) {
1041 /*
1042 * Cache and spare devices end up here - just ignore
1043 * them.
1044 */
1045 /*printf("ZFS: can't find pool details\n");*/
1046 return (EIO);
1047 }
1048
1049 if (nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64,
1050 NULL, &val) == 0 && val != 0) {
1051 return (EIO);
1052 }
1053
1054 /*
1055 * Create the pool if this is the first time we've seen it.
1056 */
1057 spa = spa_find_by_guid(pool_guid);
1058 if (spa == NULL) {
1059 spa = spa_create(pool_guid, pool_name);
1060 if (spa == NULL)
1061 return (ENOMEM);
1062 }
1063 if (pool_txg > spa->spa_txg) {
1064 spa->spa_txg = pool_txg;
1065 is_newer = 1;
1066 } else {
1067 is_newer = 0;
1068 }
1069
1070 /*
1071 * Get the vdev tree and create our in-core copy of it.
1072 * If we already have a vdev with this guid, this must
1073 * be some kind of alias (overlapping slices, dangerously dedicated
1074 * disks etc).
1075 */
1076 if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1077 NULL, &guid) != 0) {
1078 return (EIO);
1079 }
1080 vdev = vdev_find(guid);
1081 if (vdev && vdev->v_phys_read) /* Has this vdev already been inited? */
1082 return (EIO);
1083
1084 if (nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1085 NULL, &vdevs)) {
1086 return (EIO);
1087 }
1088
1089 rc = vdev_init_from_nvlist(vdevs, NULL, &top_vdev, is_newer);
1090 if (rc != 0)
1091 return (rc);
1092
1093 /*
1094 * Add the toplevel vdev to the pool if its not already there.
1095 */
1096 STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
1097 if (top_vdev == pool_vdev)
1098 break;
1099 if (!pool_vdev && top_vdev) {
1100 top_vdev->spa = spa;
1101 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
1102 }
1103
1104 /*
1105 * We should already have created an incomplete vdev for this
1106 * vdev. Find it and initialise it with our read proc.
1107 */
1108 vdev = vdev_find(guid);
1109 if (vdev) {
1110 vdev->v_phys_read = phys_read;
1111 vdev->v_read_priv = read_priv;
1112 vdev->v_state = VDEV_STATE_HEALTHY;
1113 } else {
1114 printf("ZFS: inconsistent nvlist contents\n");
1115 return (EIO);
1116 }
1117
1118 /* Record boot vdev for spa. */
1119 if (is_newer == 1)
1120 spa->spa_boot_vdev = vdev;
1121
1122 /*
1123 * Re-evaluate top-level vdev state.
1124 */
1125 vdev_set_state(top_vdev);
1126
1127 /*
1128 * Ok, we are happy with the pool so far. Lets find
1129 * the best uberblock and then we can actually access
1130 * the contents of the pool.
1131 */
1132 upbuf = zfs_alloc(VDEV_UBERBLOCK_SIZE(vdev));
1133 up = (const struct uberblock *)upbuf;
1134 for (l = 0; l < VDEV_LABELS; l++) {
1135 for (i = 0; i < VDEV_UBERBLOCK_COUNT(vdev); i++) {
1136 off = vdev_label_offset(psize, l,
1137 VDEV_UBERBLOCK_OFFSET(vdev, i));
1138 BP_ZERO(&bp);
1139 DVA_SET_OFFSET(&bp.blk_dva[0], off);
1140 BP_SET_LSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1141 BP_SET_PSIZE(&bp, VDEV_UBERBLOCK_SIZE(vdev));
1142 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1143 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1144 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1145
1146 if (vdev_read_phys(vdev, &bp, upbuf, off, 0) != 0)
1147 continue;
1148
1149 if (up->ub_magic != UBERBLOCK_MAGIC)
1150 continue;
1151 if (up->ub_txg < spa->spa_txg)
1152 continue;
1153 if (up->ub_txg > spa->spa_uberblock.ub_txg ||
1154 (up->ub_txg == spa->spa_uberblock.ub_txg &&
1155 up->ub_timestamp >
1156 spa->spa_uberblock.ub_timestamp)) {
1157 spa->spa_uberblock = *up;
1158 }
1159 }
1160 }
1161 zfs_free(upbuf, VDEV_UBERBLOCK_SIZE(vdev));
1162
1163 vdev->spa = spa;
1164 if (spap != NULL)
1165 *spap = spa;
1166 return (0);
1167 }
1168
1169 static int
1170 ilog2(int n)
1171 {
1172 int v;
1173
1174 for (v = 0; v < 32; v++)
1175 if (n == (1 << v))
1176 return v;
1177 return -1;
1178 }
1179
1180 static int
1181 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1182 {
1183 blkptr_t gbh_bp;
1184 zio_gbh_phys_t zio_gb;
1185 char *pbuf;
1186 int i;
1187
1188 /* Artificial BP for gang block header. */
1189 gbh_bp = *bp;
1190 BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1191 BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1192 BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1193 BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1194 for (i = 0; i < SPA_DVAS_PER_BP; i++)
1195 DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1196
1197 /* Read gang header block using the artificial BP. */
1198 if (zio_read(spa, &gbh_bp, &zio_gb))
1199 return (EIO);
1200
1201 pbuf = buf;
1202 for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1203 blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1204
1205 if (BP_IS_HOLE(gbp))
1206 continue;
1207 if (zio_read(spa, gbp, pbuf))
1208 return (EIO);
1209 pbuf += BP_GET_PSIZE(gbp);
1210 }
1211
1212 if (zio_checksum_verify(spa, bp, buf))
1213 return (EIO);
1214 return (0);
1215 }
1216
1217 static int
1218 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1219 {
1220 int cpfunc = BP_GET_COMPRESS(bp);
1221 uint64_t align, size;
1222 void *pbuf;
1223 int i, error;
1224
1225 /*
1226 * Process data embedded in block pointer
1227 */
1228 if (BP_IS_EMBEDDED(bp)) {
1229 ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1230
1231 size = BPE_GET_PSIZE(bp);
1232 ASSERT(size <= BPE_PAYLOAD_SIZE);
1233
1234 if (cpfunc != ZIO_COMPRESS_OFF)
1235 pbuf = zfs_alloc(size);
1236 else
1237 pbuf = buf;
1238
1239 decode_embedded_bp_compressed(bp, pbuf);
1240 error = 0;
1241
1242 if (cpfunc != ZIO_COMPRESS_OFF) {
1243 error = zio_decompress_data(cpfunc, pbuf,
1244 size, buf, BP_GET_LSIZE(bp));
1245 zfs_free(pbuf, size);
1246 }
1247 if (error != 0)
1248 printf("ZFS: i/o error - unable to decompress block pointer data, error %d\n",
1249 error);
1250 return (error);
1251 }
1252
1253 error = EIO;
1254
1255 for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1256 const dva_t *dva = &bp->blk_dva[i];
1257 vdev_t *vdev;
1258 int vdevid;
1259 off_t offset;
1260
1261 if (!dva->dva_word[0] && !dva->dva_word[1])
1262 continue;
1263
1264 vdevid = DVA_GET_VDEV(dva);
1265 offset = DVA_GET_OFFSET(dva);
1266 STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
1267 if (vdev->v_id == vdevid)
1268 break;
1269 }
1270 if (!vdev || !vdev->v_read)
1271 continue;
1272
1273 size = BP_GET_PSIZE(bp);
1274 if (vdev->v_read == vdev_raidz_read) {
1275 align = 1ULL << vdev->v_top->v_ashift;
1276 if (P2PHASE(size, align) != 0)
1277 size = P2ROUNDUP(size, align);
1278 }
1279 if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1280 pbuf = zfs_alloc(size);
1281 else
1282 pbuf = buf;
1283
1284 if (DVA_GET_GANG(dva))
1285 error = zio_read_gang(spa, bp, pbuf);
1286 else
1287 error = vdev->v_read(vdev, bp, pbuf, offset, size);
1288 if (error == 0) {
1289 if (cpfunc != ZIO_COMPRESS_OFF)
1290 error = zio_decompress_data(cpfunc, pbuf,
1291 BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1292 else if (size != BP_GET_PSIZE(bp))
1293 bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1294 }
1295 if (buf != pbuf)
1296 zfs_free(pbuf, size);
1297 if (error == 0)
1298 break;
1299 }
1300 if (error != 0)
1301 printf("ZFS: i/o error - all block copies unavailable\n");
1302 return (error);
1303 }
1304
1305 static int
1306 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1307 {
1308 int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1309 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1310 int nlevels = dnode->dn_nlevels;
1311 int i, rc;
1312
1313 if (bsize > SPA_MAXBLOCKSIZE) {
1314 printf("ZFS: I/O error - blocks larger than %llu are not "
1315 "supported\n", SPA_MAXBLOCKSIZE);
1316 return (EIO);
1317 }
1318
1319 /*
1320 * Note: bsize may not be a power of two here so we need to do an
1321 * actual divide rather than a bitshift.
1322 */
1323 while (buflen > 0) {
1324 uint64_t bn = offset / bsize;
1325 int boff = offset % bsize;
1326 int ibn;
1327 const blkptr_t *indbp;
1328 blkptr_t bp;
1329
1330 if (bn > dnode->dn_maxblkid) {
1331 printf("warning: zfs bug: bn %llx > dn_maxblkid %llx\n",
1332 (unsigned long long)bn,
1333 (unsigned long long)dnode->dn_maxblkid);
1334 /*
1335 * zfs bug, will not return error
1336 * return (EIO);
1337 */
1338 }
1339
1340 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1341 goto cached;
1342
1343 indbp = dnode->dn_blkptr;
1344 for (i = 0; i < nlevels; i++) {
1345 /*
1346 * Copy the bp from the indirect array so that
1347 * we can re-use the scratch buffer for multi-level
1348 * objects.
1349 */
1350 ibn = bn >> ((nlevels - i - 1) * ibshift);
1351 ibn &= ((1 << ibshift) - 1);
1352 bp = indbp[ibn];
1353 if (BP_IS_HOLE(&bp)) {
1354 memset(dnode_cache_buf, 0, bsize);
1355 break;
1356 }
1357 rc = zio_read(spa, &bp, dnode_cache_buf);
1358 if (rc)
1359 return (rc);
1360 indbp = (const blkptr_t *) dnode_cache_buf;
1361 }
1362 dnode_cache_obj = dnode;
1363 dnode_cache_bn = bn;
1364 cached:
1365
1366 /*
1367 * The buffer contains our data block. Copy what we
1368 * need from it and loop.
1369 */
1370 i = bsize - boff;
1371 if (i > buflen) i = buflen;
1372 memcpy(buf, &dnode_cache_buf[boff], i);
1373 buf = ((char*) buf) + i;
1374 offset += i;
1375 buflen -= i;
1376 }
1377
1378 return (0);
1379 }
1380
1381 /*
1382 * Lookup a value in a microzap directory. Assumes that the zap
1383 * scratch buffer contains the directory contents.
1384 */
1385 static int
1386 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
1387 {
1388 const mzap_phys_t *mz;
1389 const mzap_ent_phys_t *mze;
1390 size_t size;
1391 int chunks, i;
1392
1393 /*
1394 * Microzap objects use exactly one block. Read the whole
1395 * thing.
1396 */
1397 size = dnode->dn_datablkszsec * 512;
1398
1399 mz = (const mzap_phys_t *) zap_scratch;
1400 chunks = size / MZAP_ENT_LEN - 1;
1401
1402 for (i = 0; i < chunks; i++) {
1403 mze = &mz->mz_chunk[i];
1404 if (!strcmp(mze->mze_name, name)) {
1405 *value = mze->mze_value;
1406 return (0);
1407 }
1408 }
1409
1410 return (ENOENT);
1411 }
1412
1413 /*
1414 * Compare a name with a zap leaf entry. Return non-zero if the name
1415 * matches.
1416 */
1417 static int
1418 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1419 {
1420 size_t namelen;
1421 const zap_leaf_chunk_t *nc;
1422 const char *p;
1423
1424 namelen = zc->l_entry.le_name_numints;
1425
1426 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1427 p = name;
1428 while (namelen > 0) {
1429 size_t len;
1430 len = namelen;
1431 if (len > ZAP_LEAF_ARRAY_BYTES)
1432 len = ZAP_LEAF_ARRAY_BYTES;
1433 if (memcmp(p, nc->l_array.la_array, len))
1434 return (0);
1435 p += len;
1436 namelen -= len;
1437 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1438 }
1439
1440 return 1;
1441 }
1442
1443 /*
1444 * Extract a uint64_t value from a zap leaf entry.
1445 */
1446 static uint64_t
1447 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1448 {
1449 const zap_leaf_chunk_t *vc;
1450 int i;
1451 uint64_t value;
1452 const uint8_t *p;
1453
1454 vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1455 for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1456 value = (value << 8) | p[i];
1457 }
1458
1459 return value;
1460 }
1461
1462 static void
1463 stv(int len, void *addr, uint64_t value)
1464 {
1465 switch (len) {
1466 case 1:
1467 *(uint8_t *)addr = value;
1468 return;
1469 case 2:
1470 *(uint16_t *)addr = value;
1471 return;
1472 case 4:
1473 *(uint32_t *)addr = value;
1474 return;
1475 case 8:
1476 *(uint64_t *)addr = value;
1477 return;
1478 }
1479 }
1480
1481 /*
1482 * Extract a array from a zap leaf entry.
1483 */
1484 static void
1485 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
1486 uint64_t integer_size, uint64_t num_integers, void *buf)
1487 {
1488 uint64_t array_int_len = zc->l_entry.le_value_intlen;
1489 uint64_t value = 0;
1490 uint64_t *u64 = buf;
1491 char *p = buf;
1492 int len = MIN(zc->l_entry.le_value_numints, num_integers);
1493 int chunk = zc->l_entry.le_value_chunk;
1494 int byten = 0;
1495
1496 if (integer_size == 8 && len == 1) {
1497 *u64 = fzap_leaf_value(zl, zc);
1498 return;
1499 }
1500
1501 while (len > 0) {
1502 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
1503 int i;
1504
1505 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
1506 for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
1507 value = (value << 8) | la->la_array[i];
1508 byten++;
1509 if (byten == array_int_len) {
1510 stv(integer_size, p, value);
1511 byten = 0;
1512 len--;
1513 if (len == 0)
1514 return;
1515 p += integer_size;
1516 }
1517 }
1518 chunk = la->la_next;
1519 }
1520 }
1521
1522 static int
1523 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
1524 {
1525
1526 switch (integer_size) {
1527 case 1:
1528 case 2:
1529 case 4:
1530 case 8:
1531 break;
1532 default:
1533 return (EINVAL);
1534 }
1535
1536 if (integer_size * num_integers > ZAP_MAXVALUELEN)
1537 return (E2BIG);
1538
1539 return (0);
1540 }
1541
1542 /*
1543 * Lookup a value in a fatzap directory. Assumes that the zap scratch
1544 * buffer contains the directory header.
1545 */
1546 static int
1547 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
1548 uint64_t integer_size, uint64_t num_integers, void *value)
1549 {
1550 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1551 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1552 fat_zap_t z;
1553 uint64_t *ptrtbl;
1554 uint64_t hash;
1555 int rc;
1556
1557 if (zh.zap_magic != ZAP_MAGIC)
1558 return (EIO);
1559
1560 if ((rc = fzap_check_size(integer_size, num_integers)) != 0)
1561 return (rc);
1562
1563 z.zap_block_shift = ilog2(bsize);
1564 z.zap_phys = (zap_phys_t *) zap_scratch;
1565
1566 /*
1567 * Figure out where the pointer table is and read it in if necessary.
1568 */
1569 if (zh.zap_ptrtbl.zt_blk) {
1570 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1571 zap_scratch, bsize);
1572 if (rc)
1573 return (rc);
1574 ptrtbl = (uint64_t *) zap_scratch;
1575 } else {
1576 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1577 }
1578
1579 hash = zap_hash(zh.zap_salt, name);
1580
1581 zap_leaf_t zl;
1582 zl.l_bs = z.zap_block_shift;
1583
1584 off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1585 zap_leaf_chunk_t *zc;
1586
1587 rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1588 if (rc)
1589 return (rc);
1590
1591 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1592
1593 /*
1594 * Make sure this chunk matches our hash.
1595 */
1596 if (zl.l_phys->l_hdr.lh_prefix_len > 0
1597 && zl.l_phys->l_hdr.lh_prefix
1598 != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1599 return (ENOENT);
1600
1601 /*
1602 * Hash within the chunk to find our entry.
1603 */
1604 int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1605 int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1606 h = zl.l_phys->l_hash[h];
1607 if (h == 0xffff)
1608 return (ENOENT);
1609 zc = &ZAP_LEAF_CHUNK(&zl, h);
1610 while (zc->l_entry.le_hash != hash) {
1611 if (zc->l_entry.le_next == 0xffff) {
1612 zc = 0;
1613 break;
1614 }
1615 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1616 }
1617 if (fzap_name_equal(&zl, zc, name)) {
1618 if (zc->l_entry.le_value_intlen > integer_size)
1619 return (EINVAL);
1620
1621 fzap_leaf_array(&zl, zc, integer_size, num_integers, value);
1622 return (0);
1623 }
1624
1625 return (ENOENT);
1626 }
1627
1628 /*
1629 * Lookup a name in a zap object and return its value as a uint64_t.
1630 */
1631 static int
1632 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
1633 uint64_t integer_size, uint64_t num_integers, void *value)
1634 {
1635 int rc;
1636 uint64_t zap_type;
1637 size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1638
1639 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1640 if (rc)
1641 return (rc);
1642
1643 zap_type = *(uint64_t *) zap_scratch;
1644 if (zap_type == ZBT_MICRO)
1645 return mzap_lookup(dnode, name, value);
1646 else if (zap_type == ZBT_HEADER) {
1647 return fzap_lookup(spa, dnode, name, integer_size,
1648 num_integers, value);
1649 }
1650 printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
1651 return (EIO);
1652 }
1653
1654 /*
1655 * List a microzap directory. Assumes that the zap scratch buffer contains
1656 * the directory contents.
1657 */
1658 static int
1659 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1660 {
1661 const mzap_phys_t *mz;
1662 const mzap_ent_phys_t *mze;
1663 size_t size;
1664 int chunks, i, rc;
1665
1666 /*
1667 * Microzap objects use exactly one block. Read the whole
1668 * thing.
1669 */
1670 size = dnode->dn_datablkszsec * 512;
1671 mz = (const mzap_phys_t *) zap_scratch;
1672 chunks = size / MZAP_ENT_LEN - 1;
1673
1674 for (i = 0; i < chunks; i++) {
1675 mze = &mz->mz_chunk[i];
1676 if (mze->mze_name[0]) {
1677 rc = callback(mze->mze_name, mze->mze_value);
1678 if (rc != 0)
1679 return (rc);
1680 }
1681 }
1682
1683 return (0);
1684 }
1685
1686 /*
1687 * List a fatzap directory. Assumes that the zap scratch buffer contains
1688 * the directory header.
1689 */
1690 static int
1691 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
1692 {
1693 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1694 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1695 fat_zap_t z;
1696 int i, j, rc;
1697
1698 if (zh.zap_magic != ZAP_MAGIC)
1699 return (EIO);
1700
1701 z.zap_block_shift = ilog2(bsize);
1702 z.zap_phys = (zap_phys_t *) zap_scratch;
1703
1704 /*
1705 * This assumes that the leaf blocks start at block 1. The
1706 * documentation isn't exactly clear on this.
1707 */
1708 zap_leaf_t zl;
1709 zl.l_bs = z.zap_block_shift;
1710 for (i = 0; i < zh.zap_num_leafs; i++) {
1711 off_t off = (i + 1) << zl.l_bs;
1712 char name[256], *p;
1713 uint64_t value;
1714
1715 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1716 return (EIO);
1717
1718 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1719
1720 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1721 zap_leaf_chunk_t *zc, *nc;
1722 int namelen;
1723
1724 zc = &ZAP_LEAF_CHUNK(&zl, j);
1725 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1726 continue;
1727 namelen = zc->l_entry.le_name_numints;
1728 if (namelen > sizeof(name))
1729 namelen = sizeof(name);
1730
1731 /*
1732 * Paste the name back together.
1733 */
1734 nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1735 p = name;
1736 while (namelen > 0) {
1737 int len;
1738 len = namelen;
1739 if (len > ZAP_LEAF_ARRAY_BYTES)
1740 len = ZAP_LEAF_ARRAY_BYTES;
1741 memcpy(p, nc->l_array.la_array, len);
1742 p += len;
1743 namelen -= len;
1744 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1745 }
1746
1747 /*
1748 * Assume the first eight bytes of the value are
1749 * a uint64_t.
1750 */
1751 value = fzap_leaf_value(&zl, zc);
1752
1753 //printf("%s 0x%jx\n", name, (uintmax_t)value);
1754 rc = callback((const char *)name, value);
1755 if (rc != 0)
1756 return (rc);
1757 }
1758 }
1759
1760 return (0);
1761 }
1762
1763 static int zfs_printf(const char *name, uint64_t value __unused)
1764 {
1765
1766 printf("%s\n", name);
1767
1768 return (0);
1769 }
1770
1771 /*
1772 * List a zap directory.
1773 */
1774 static int
1775 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
1776 {
1777 uint64_t zap_type;
1778 size_t size = dnode->dn_datablkszsec * 512;
1779
1780 if (dnode_read(spa, dnode, 0, zap_scratch, size))
1781 return (EIO);
1782
1783 zap_type = *(uint64_t *) zap_scratch;
1784 if (zap_type == ZBT_MICRO)
1785 return mzap_list(dnode, zfs_printf);
1786 else
1787 return fzap_list(spa, dnode, zfs_printf);
1788 }
1789
1790 static int
1791 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1792 {
1793 off_t offset;
1794
1795 offset = objnum * sizeof(dnode_phys_t);
1796 return dnode_read(spa, &os->os_meta_dnode, offset,
1797 dnode, sizeof(dnode_phys_t));
1798 }
1799
1800 static int
1801 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1802 {
1803 const mzap_phys_t *mz;
1804 const mzap_ent_phys_t *mze;
1805 size_t size;
1806 int chunks, i;
1807
1808 /*
1809 * Microzap objects use exactly one block. Read the whole
1810 * thing.
1811 */
1812 size = dnode->dn_datablkszsec * 512;
1813
1814 mz = (const mzap_phys_t *) zap_scratch;
1815 chunks = size / MZAP_ENT_LEN - 1;
1816
1817 for (i = 0; i < chunks; i++) {
1818 mze = &mz->mz_chunk[i];
1819 if (value == mze->mze_value) {
1820 strcpy(name, mze->mze_name);
1821 return (0);
1822 }
1823 }
1824
1825 return (ENOENT);
1826 }
1827
1828 static void
1829 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
1830 {
1831 size_t namelen;
1832 const zap_leaf_chunk_t *nc;
1833 char *p;
1834
1835 namelen = zc->l_entry.le_name_numints;
1836
1837 nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1838 p = name;
1839 while (namelen > 0) {
1840 size_t len;
1841 len = namelen;
1842 if (len > ZAP_LEAF_ARRAY_BYTES)
1843 len = ZAP_LEAF_ARRAY_BYTES;
1844 memcpy(p, nc->l_array.la_array, len);
1845 p += len;
1846 namelen -= len;
1847 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1848 }
1849
1850 *p = '\0';
1851 }
1852
1853 static int
1854 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1855 {
1856 int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1857 zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1858 fat_zap_t z;
1859 int i, j;
1860
1861 if (zh.zap_magic != ZAP_MAGIC)
1862 return (EIO);
1863
1864 z.zap_block_shift = ilog2(bsize);
1865 z.zap_phys = (zap_phys_t *) zap_scratch;
1866
1867 /*
1868 * This assumes that the leaf blocks start at block 1. The
1869 * documentation isn't exactly clear on this.
1870 */
1871 zap_leaf_t zl;
1872 zl.l_bs = z.zap_block_shift;
1873 for (i = 0; i < zh.zap_num_leafs; i++) {
1874 off_t off = (i + 1) << zl.l_bs;
1875
1876 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1877 return (EIO);
1878
1879 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1880
1881 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1882 zap_leaf_chunk_t *zc;
1883
1884 zc = &ZAP_LEAF_CHUNK(&zl, j);
1885 if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1886 continue;
1887 if (zc->l_entry.le_value_intlen != 8 ||
1888 zc->l_entry.le_value_numints != 1)
1889 continue;
1890
1891 if (fzap_leaf_value(&zl, zc) == value) {
1892 fzap_name_copy(&zl, zc, name);
1893 return (0);
1894 }
1895 }
1896 }
1897
1898 return (ENOENT);
1899 }
1900
1901 static int
1902 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name, uint64_t value)
1903 {
1904 int rc;
1905 uint64_t zap_type;
1906 size_t size = dnode->dn_datablkszsec * 512;
1907
1908 rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1909 if (rc)
1910 return (rc);
1911
1912 zap_type = *(uint64_t *) zap_scratch;
1913 if (zap_type == ZBT_MICRO)
1914 return mzap_rlookup(spa, dnode, name, value);
1915 else
1916 return fzap_rlookup(spa, dnode, name, value);
1917 }
1918
1919 static int
1920 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
1921 {
1922 char name[256];
1923 char component[256];
1924 uint64_t dir_obj, parent_obj, child_dir_zapobj;
1925 dnode_phys_t child_dir_zap, dataset, dir, parent;
1926 dsl_dir_phys_t *dd;
1927 dsl_dataset_phys_t *ds;
1928 char *p;
1929 int len;
1930
1931 p = &name[sizeof(name) - 1];
1932 *p = '\0';
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 for (;;) {
1942 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
1943 return (EIO);
1944 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1945
1946 /* Actual loop condition. */
1947 parent_obj = dd->dd_parent_obj;
1948 if (parent_obj == 0)
1949 break;
1950
1951 if (objset_get_dnode(spa, &spa->spa_mos, parent_obj, &parent) != 0)
1952 return (EIO);
1953 dd = (dsl_dir_phys_t *)&parent.dn_bonus;
1954 child_dir_zapobj = dd->dd_child_dir_zapobj;
1955 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
1956 return (EIO);
1957 if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
1958 return (EIO);
1959
1960 len = strlen(component);
1961 p -= len;
1962 memcpy(p, component, len);
1963 --p;
1964 *p = '/';
1965
1966 /* Actual loop iteration. */
1967 dir_obj = parent_obj;
1968 }
1969
1970 if (*p != '\0')
1971 ++p;
1972 strcpy(result, p);
1973
1974 return (0);
1975 }
1976
1977 static int
1978 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
1979 {
1980 char element[256];
1981 uint64_t dir_obj, child_dir_zapobj;
1982 dnode_phys_t child_dir_zap, dir;
1983 dsl_dir_phys_t *dd;
1984 const char *p, *q;
1985
1986 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir))
1987 return (EIO);
1988 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
1989 1, &dir_obj))
1990 return (EIO);
1991
1992 p = name;
1993 for (;;) {
1994 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
1995 return (EIO);
1996 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
1997
1998 while (*p == '/')
1999 p++;
2000 /* Actual loop condition #1. */
2001 if (*p == '\0')
2002 break;
2003
2004 q = strchr(p, '/');
2005 if (q) {
2006 memcpy(element, p, q - p);
2007 element[q - p] = '\0';
2008 p = q + 1;
2009 } else {
2010 strcpy(element, p);
2011 p += strlen(p);
2012 }
2013
2014 child_dir_zapobj = dd->dd_child_dir_zapobj;
2015 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0)
2016 return (EIO);
2017
2018 /* Actual loop condition #2. */
2019 if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2020 1, &dir_obj) != 0)
2021 return (ENOENT);
2022 }
2023
2024 *objnum = dd->dd_head_dataset_obj;
2025 return (0);
2026 }
2027
2028 #pragma GCC diagnostic ignored "-Wstrict-aliasing"
2029 static int
2030 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2031 {
2032 uint64_t dir_obj, child_dir_zapobj;
2033 dnode_phys_t child_dir_zap, dir, dataset;
2034 dsl_dataset_phys_t *ds;
2035 dsl_dir_phys_t *dd;
2036
2037 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2038 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2039 return (EIO);
2040 }
2041 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2042 dir_obj = ds->ds_dir_obj;
2043
2044 if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
2045 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2046 return (EIO);
2047 }
2048 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2049
2050 child_dir_zapobj = dd->dd_child_dir_zapobj;
2051 if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap) != 0) {
2052 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2053 return (EIO);
2054 }
2055
2056 return (zap_list(spa, &child_dir_zap) != 0);
2057 }
2058
2059 int
2060 zfs_callback_dataset(const spa_t *spa, uint64_t objnum, int (*callback)(const char *, uint64_t))
2061 {
2062 uint64_t dir_obj, child_dir_zapobj, zap_type;
2063 dnode_phys_t child_dir_zap, dir, dataset;
2064 dsl_dataset_phys_t *ds;
2065 dsl_dir_phys_t *dd;
2066 int err;
2067
2068 err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
2069 if (err != 0) {
2070 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2071 return (err);
2072 }
2073 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2074 dir_obj = ds->ds_dir_obj;
2075
2076 err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
2077 if (err != 0) {
2078 printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2079 return (err);
2080 }
2081 dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2082
2083 child_dir_zapobj = dd->dd_child_dir_zapobj;
2084 err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj, &child_dir_zap);
2085 if (err != 0) {
2086 printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2087 return (err);
2088 }
2089
2090 err = dnode_read(spa, &child_dir_zap, 0, zap_scratch, child_dir_zap.dn_datablkszsec * 512);
2091 if (err != 0)
2092 return (err);
2093
2094 zap_type = *(uint64_t *) zap_scratch;
2095 if (zap_type == ZBT_MICRO)
2096 return mzap_list(&child_dir_zap, callback);
2097 else
2098 return fzap_list(spa, &child_dir_zap, callback);
2099 }
2100
2101 /*
2102 * Find the object set given the object number of its dataset object
2103 * and return its details in *objset
2104 */
2105 static int
2106 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2107 {
2108 dnode_phys_t dataset;
2109 dsl_dataset_phys_t *ds;
2110
2111 if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2112 printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2113 return (EIO);
2114 }
2115
2116 ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
2117 if (zio_read(spa, &ds->ds_bp, objset)) {
2118 printf("ZFS: can't read object set for dataset %ju\n",
2119 (uintmax_t)objnum);
2120 return (EIO);
2121 }
2122
2123 return (0);
2124 }
2125
2126 /*
2127 * Find the object set pointed to by the BOOTFS property or the root
2128 * dataset if there is none and return its details in *objset
2129 */
2130 static int
2131 zfs_get_root(const spa_t *spa, uint64_t *objid)
2132 {
2133 dnode_phys_t dir, propdir;
2134 uint64_t props, bootfs, root;
2135
2136 *objid = 0;
2137
2138 /*
2139 * Start with the MOS directory object.
2140 */
2141 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
2142 printf("ZFS: can't read MOS object directory\n");
2143 return (EIO);
2144 }
2145
2146 /*
2147 * Lookup the pool_props and see if we can find a bootfs.
2148 */
2149 if (zap_lookup(spa, &dir, DMU_POOL_PROPS, sizeof (props), 1, &props) == 0
2150 && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
2151 && zap_lookup(spa, &propdir, "bootfs", sizeof (bootfs), 1, &bootfs) == 0
2152 && bootfs != 0)
2153 {
2154 *objid = bootfs;
2155 return (0);
2156 }
2157 /*
2158 * Lookup the root dataset directory
2159 */
2160 if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (root), 1, &root)
2161 || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
2162 printf("ZFS: can't find root dsl_dir\n");
2163 return (EIO);
2164 }
2165
2166 /*
2167 * Use the information from the dataset directory's bonus buffer
2168 * to find the dataset object and from that the object set itself.
2169 */
2170 dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
2171 *objid = dd->dd_head_dataset_obj;
2172 return (0);
2173 }
2174
2175 static int
2176 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mnt)
2177 {
2178
2179 mnt->spa = spa;
2180
2181 /*
2182 * Find the root object set if not explicitly provided
2183 */
2184 if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
2185 printf("ZFS: can't find root filesystem\n");
2186 return (EIO);
2187 }
2188
2189 if (zfs_mount_dataset(spa, rootobj, &mnt->objset)) {
2190 printf("ZFS: can't open root filesystem\n");
2191 return (EIO);
2192 }
2193
2194 mnt->rootobj = rootobj;
2195
2196 return (0);
2197 }
2198
2199 /*
2200 * callback function for feature name checks.
2201 */
2202 static int
2203 check_feature(const char *name, uint64_t value)
2204 {
2205 int i;
2206
2207 if (value == 0)
2208 return (0);
2209 if (name[0] == '\0')
2210 return (0);
2211
2212 for (i = 0; features_for_read[i] != NULL; i++) {
2213 if (strcmp(name, features_for_read[i]) == 0)
2214 return (0);
2215 }
2216 printf("ZFS: unsupported feature: %s\n", name);
2217 return (EIO);
2218 }
2219
2220 /*
2221 * Checks whether the MOS features that are active are supported.
2222 */
2223 static int
2224 check_mos_features(const spa_t *spa)
2225 {
2226 dnode_phys_t dir;
2227 uint64_t objnum, zap_type;
2228 size_t size;
2229 int rc;
2230
2231 if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
2232 &dir)) != 0)
2233 return (rc);
2234 if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
2235 sizeof (objnum), 1, &objnum)) != 0) {
2236 /*
2237 * It is older pool without features. As we have already
2238 * tested the label, just return without raising the error.
2239 */
2240 if (rc == ENOENT)
2241 rc = 0;
2242 return (rc);
2243 }
2244
2245 if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
2246 return (rc);
2247
2248 if (dir.dn_type != DMU_OTN_ZAP_METADATA)
2249 return (EIO);
2250
2251 size = dir.dn_datablkszsec * 512;
2252 if (dnode_read(spa, &dir, 0, zap_scratch, size))
2253 return (EIO);
2254
2255 zap_type = *(uint64_t *) zap_scratch;
2256 if (zap_type == ZBT_MICRO)
2257 rc = mzap_list(&dir, check_feature);
2258 else
2259 rc = fzap_list(spa, &dir, check_feature);
2260
2261 return (rc);
2262 }
2263
2264 static int
2265 zfs_spa_init(spa_t *spa)
2266 {
2267 dnode_phys_t dir;
2268 int rc;
2269
2270 if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
2271 printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
2272 return (EIO);
2273 }
2274 if (spa->spa_mos.os_type != DMU_OST_META) {
2275 printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
2276 return (EIO);
2277 }
2278
2279 if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT,
2280 &dir)) {
2281 printf("ZFS: failed to read pool %s directory object\n",
2282 spa->spa_name);
2283 return (EIO);
2284 }
2285 /* this is allowed to fail, older pools do not have salt */
2286 rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
2287 sizeof (spa->spa_cksum_salt.zcs_bytes),
2288 spa->spa_cksum_salt.zcs_bytes);
2289
2290 rc = check_mos_features(spa);
2291 if (rc != 0) {
2292 printf("ZFS: pool %s is not supported\n", spa->spa_name);
2293 }
2294
2295 return (rc);
2296 }
2297
2298 static int
2299 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
2300 {
2301
2302 if (dn->dn_bonustype != DMU_OT_SA) {
2303 znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
2304
2305 sb->st_mode = zp->zp_mode;
2306 sb->st_uid = zp->zp_uid;
2307 sb->st_gid = zp->zp_gid;
2308 sb->st_size = zp->zp_size;
2309 } else {
2310 sa_hdr_phys_t *sahdrp;
2311 int hdrsize;
2312 size_t size = 0;
2313 void *buf = NULL;
2314
2315 if (dn->dn_bonuslen != 0)
2316 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2317 else {
2318 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
2319 blkptr_t *bp = DN_SPILL_BLKPTR(dn);
2320 int error;
2321
2322 size = BP_GET_LSIZE(bp);
2323 buf = zfs_alloc(size);
2324 error = zio_read(spa, bp, buf);
2325 if (error != 0) {
2326 zfs_free(buf, size);
2327 return (error);
2328 }
2329 sahdrp = buf;
2330 } else {
2331 return (EIO);
2332 }
2333 }
2334 hdrsize = SA_HDR_SIZE(sahdrp);
2335 sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
2336 SA_MODE_OFFSET);
2337 sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
2338 SA_UID_OFFSET);
2339 sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
2340 SA_GID_OFFSET);
2341 sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
2342 SA_SIZE_OFFSET);
2343 if (buf != NULL)
2344 zfs_free(buf, size);
2345 }
2346
2347 return (0);
2348 }
2349
2350 static int
2351 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
2352 {
2353 int rc = 0;
2354
2355 if (dn->dn_bonustype == DMU_OT_SA) {
2356 sa_hdr_phys_t *sahdrp = NULL;
2357 size_t size = 0;
2358 void *buf = NULL;
2359 int hdrsize;
2360 char *p;
2361
2362 if (dn->dn_bonuslen != 0)
2363 sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
2364 else {
2365 blkptr_t *bp;
2366
2367 if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
2368 return (EIO);
2369 bp = DN_SPILL_BLKPTR(dn);
2370
2371 size = BP_GET_LSIZE(bp);
2372 buf = zfs_alloc(size);
2373 rc = zio_read(spa, bp, buf);
2374 if (rc != 0) {
2375 zfs_free(buf, size);
2376 return (rc);
2377 }
2378 sahdrp = buf;
2379 }
2380 hdrsize = SA_HDR_SIZE(sahdrp);
2381 p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
2382 memcpy(path, p, psize);
2383 if (buf != NULL)
2384 zfs_free(buf, size);
2385 return (0);
2386 }
2387 /*
2388 * Second test is purely to silence bogus compiler
2389 * warning about accessing past the end of dn_bonus.
2390 */
2391 if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
2392 sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
2393 memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
2394 } else {
2395 rc = dnode_read(spa, dn, 0, path, psize);
2396 }
2397 return (rc);
2398 }
2399
2400 struct obj_list {
2401 uint64_t objnum;
2402 STAILQ_ENTRY(obj_list) entry;
2403 };
2404
2405 /*
2406 * Lookup a file and return its dnode.
2407 */
2408 static int
2409 zfs_lookup(const struct zfsmount *mnt, const char *upath, dnode_phys_t *dnode)
2410 {
2411 int rc;
2412 uint64_t objnum;
2413 const spa_t *spa;
2414 dnode_phys_t dn;
2415 const char *p, *q;
2416 char element[256];
2417 char path[1024];
2418 int symlinks_followed = 0;
2419 struct stat sb;
2420 struct obj_list *entry, *tentry;
2421 STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
2422
2423 spa = mnt->spa;
2424 if (mnt->objset.os_type != DMU_OST_ZFS) {
2425 printf("ZFS: unexpected object set type %ju\n",
2426 (uintmax_t)mnt->objset.os_type);
2427 return (EIO);
2428 }
2429
2430 if ((entry = malloc(sizeof(struct obj_list))) == NULL)
2431 return (ENOMEM);
2432
2433 /*
2434 * Get the root directory dnode.
2435 */
2436 rc = objset_get_dnode(spa, &mnt->objset, MASTER_NODE_OBJ, &dn);
2437 if (rc) {
2438 free(entry);
2439 return (rc);
2440 }
2441
2442 rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
2443 if (rc) {
2444 free(entry);
2445 return (rc);
2446 }
2447 entry->objnum = objnum;
2448 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2449
2450 rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2451 if (rc != 0)
2452 goto done;
2453
2454 p = upath;
2455 while (p && *p) {
2456 rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2457 if (rc != 0)
2458 goto done;
2459
2460 while (*p == '/')
2461 p++;
2462 if (*p == '\0')
2463 break;
2464 q = p;
2465 while (*q != '\0' && *q != '/')
2466 q++;
2467
2468 /* skip dot */
2469 if (p + 1 == q && p[0] == '.') {
2470 p++;
2471 continue;
2472 }
2473 /* double dot */
2474 if (p + 2 == q && p[0] == '.' && p[1] == '.') {
2475 p += 2;
2476 if (STAILQ_FIRST(&on_cache) ==
2477 STAILQ_LAST(&on_cache, obj_list, entry)) {
2478 rc = ENOENT;
2479 goto done;
2480 }
2481 entry = STAILQ_FIRST(&on_cache);
2482 STAILQ_REMOVE_HEAD(&on_cache, entry);
2483 free(entry);
2484 objnum = (STAILQ_FIRST(&on_cache))->objnum;
2485 continue;
2486 }
2487 if (q - p + 1 > sizeof(element)) {
2488 rc = ENAMETOOLONG;
2489 goto done;
2490 }
2491 memcpy(element, p, q - p);
2492 element[q - p] = 0;
2493 p = q;
2494
2495 if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
2496 goto done;
2497 if (!S_ISDIR(sb.st_mode)) {
2498 rc = ENOTDIR;
2499 goto done;
2500 }
2501
2502 rc = zap_lookup(spa, &dn, element, sizeof(objnum), 1, &objnum);
2503 if (rc)
2504 goto done;
2505 objnum = ZFS_DIRENT_OBJ(objnum);
2506
2507 if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
2508 rc = ENOMEM;
2509 goto done;
2510 }
2511 entry->objnum = objnum;
2512 STAILQ_INSERT_HEAD(&on_cache, entry, entry);
2513 rc = objset_get_dnode(spa, &mnt->objset, objnum, &dn);
2514 if (rc)
2515 goto done;
2516
2517 /*
2518 * Check for symlink.
2519 */
2520 rc = zfs_dnode_stat(spa, &dn, &sb);
2521 if (rc)
2522 goto done;
2523 if (S_ISLNK(sb.st_mode)) {
2524 if (symlinks_followed > 10) {
2525 rc = EMLINK;
2526 goto done;
2527 }
2528 symlinks_followed++;
2529
2530 /*
2531 * Read the link value and copy the tail of our
2532 * current path onto the end.
2533 */
2534 if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
2535 rc = ENAMETOOLONG;
2536 goto done;
2537 }
2538 strcpy(&path[sb.st_size], p);
2539
2540 rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
2541 if (rc != 0)
2542 goto done;
2543
2544 /*
2545 * Restart with the new path, starting either at
2546 * the root or at the parent depending whether or
2547 * not the link is relative.
2548 */
2549 p = path;
2550 if (*p == '/') {
2551 while (STAILQ_FIRST(&on_cache) !=
2552 STAILQ_LAST(&on_cache, obj_list, entry)) {
2553 entry = STAILQ_FIRST(&on_cache);
2554 STAILQ_REMOVE_HEAD(&on_cache, entry);
2555 free(entry);
2556 }
2557 } else {
2558 entry = STAILQ_FIRST(&on_cache);
2559 STAILQ_REMOVE_HEAD(&on_cache, entry);
2560 free(entry);
2561 }
2562 objnum = (STAILQ_FIRST(&on_cache))->objnum;
2563 }
2564 }
2565
2566 *dnode = dn;
2567 done:
2568 STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
2569 free(entry);
2570 return (rc);
2571 }
Unleashed OS