initial commit
[freebsd-arm:freebsd-arm.git] / boot / zfs / zfsimpl.c
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 __FBSDID("$FreeBSD$");
29
30 /*
31  *      Stand-alone ZFS file reader.
32  */
33
34 #include "zfsimpl.h"
35 #include "zfssubr.c"
36
37 /*
38  * List of all vdevs, chained through v_alllink.
39  */
40 static vdev_list_t zfs_vdevs;
41
42 /*
43  * List of all pools, chained through spa_link.
44  */
45 static spa_list_t zfs_pools;
46
47 static uint64_t zfs_crc64_table[256];
48 static const dnode_phys_t *dnode_cache_obj = 0;
49 static uint64_t dnode_cache_bn;
50 static char *dnode_cache_buf;
51 static char *zap_scratch;
52 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
53
54 #define TEMP_SIZE       (1024 * 1024)
55
56 static int zio_read(spa_t *spa, const blkptr_t *bp, void *buf);
57
58 static void
59 zfs_init(void)
60 {
61         STAILQ_INIT(&zfs_vdevs);
62         STAILQ_INIT(&zfs_pools);
63
64         zfs_temp_buf = malloc(TEMP_SIZE);
65         zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
66         zfs_temp_ptr = zfs_temp_buf;
67         dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
68         zap_scratch = malloc(SPA_MAXBLOCKSIZE);
69
70         zfs_init_crc();
71 }
72
73 static char *
74 zfs_alloc_temp(size_t sz)
75 {
76         char *p;
77
78         if (zfs_temp_ptr + sz > zfs_temp_end) {
79                 printf("ZFS: out of temporary buffer space\n");
80                 for (;;) ;
81         }
82         p = zfs_temp_ptr;
83         zfs_temp_ptr += sz;
84
85         return (p);
86 }
87
88 static void
89 zfs_reset_temp(void)
90 {
91
92         zfs_temp_ptr = zfs_temp_buf;
93 }
94
95 static int
96 xdr_int(const unsigned char **xdr, int *ip)
97 {
98         *ip = ((*xdr)[0] << 24)
99                 | ((*xdr)[1] << 16)
100                 | ((*xdr)[2] << 8)
101                 | ((*xdr)[3] << 0);
102         (*xdr) += 4;
103         return (0);
104 }
105
106 static int
107 xdr_u_int(const unsigned char **xdr, u_int *ip)
108 {
109         *ip = ((*xdr)[0] << 24)
110                 | ((*xdr)[1] << 16)
111                 | ((*xdr)[2] << 8)
112                 | ((*xdr)[3] << 0);
113         (*xdr) += 4;
114         return (0);
115 }
116
117 static int
118 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
119 {
120         u_int hi, lo;
121
122         xdr_u_int(xdr, &hi);
123         xdr_u_int(xdr, &lo);
124         *lp = (((uint64_t) hi) << 32) | lo;
125         return (0);
126 }
127
128 static int
129 nvlist_find(const unsigned char *nvlist, const char *name, int type,
130             int* elementsp, void *valuep)
131 {
132         const unsigned char *p, *pair;
133         int junk;
134         int encoded_size, decoded_size;
135
136         p = nvlist;
137         xdr_int(&p, &junk);
138         xdr_int(&p, &junk);
139
140         pair = p;
141         xdr_int(&p, &encoded_size);
142         xdr_int(&p, &decoded_size);
143         while (encoded_size && decoded_size) {
144                 int namelen, pairtype, elements;
145                 const char *pairname;
146
147                 xdr_int(&p, &namelen);
148                 pairname = (const char*) p;
149                 p += roundup(namelen, 4);
150                 xdr_int(&p, &pairtype);
151
152                 if (!memcmp(name, pairname, namelen) && type == pairtype) {
153                         xdr_int(&p, &elements);
154                         if (elementsp)
155                                 *elementsp = elements;
156                         if (type == DATA_TYPE_UINT64) {
157                                 xdr_uint64_t(&p, (uint64_t *) valuep);
158                                 return (0);
159                         } else if (type == DATA_TYPE_STRING) {
160                                 int len;
161                                 xdr_int(&p, &len);
162                                 (*(const char**) valuep) = (const char*) p;
163                                 return (0);
164                         } else if (type == DATA_TYPE_NVLIST
165                                    || type == DATA_TYPE_NVLIST_ARRAY) {
166                                 (*(const unsigned char**) valuep) =
167                                          (const unsigned char*) p;
168                                 return (0);
169                         } else {
170                                 return (EIO);
171                         }
172                 } else {
173                         /*
174                          * Not the pair we are looking for, skip to the next one.
175                          */
176                         p = pair + encoded_size;
177                 }
178
179                 pair = p;
180                 xdr_int(&p, &encoded_size);
181                 xdr_int(&p, &decoded_size);
182         }
183
184         return (EIO);
185 }
186
187 /*
188  * Return the next nvlist in an nvlist array.
189  */
190 static const unsigned char *
191 nvlist_next(const unsigned char *nvlist)
192 {
193         const unsigned char *p, *pair;
194         int junk;
195         int encoded_size, decoded_size;
196
197         p = nvlist;
198         xdr_int(&p, &junk);
199         xdr_int(&p, &junk);
200
201         pair = p;
202         xdr_int(&p, &encoded_size);
203         xdr_int(&p, &decoded_size);
204         while (encoded_size && decoded_size) {
205                 p = pair + encoded_size;
206
207                 pair = p;
208                 xdr_int(&p, &encoded_size);
209                 xdr_int(&p, &decoded_size);
210         }
211
212         return p;
213 }
214
215 #ifdef TEST
216
217 static const unsigned char *
218 nvlist_print(const unsigned char *nvlist, unsigned int indent)
219 {
220         static const char* typenames[] = {
221                 "DATA_TYPE_UNKNOWN",
222                 "DATA_TYPE_BOOLEAN",
223                 "DATA_TYPE_BYTE",
224                 "DATA_TYPE_INT16",
225                 "DATA_TYPE_UINT16",
226                 "DATA_TYPE_INT32",
227                 "DATA_TYPE_UINT32",
228                 "DATA_TYPE_INT64",
229                 "DATA_TYPE_UINT64",
230                 "DATA_TYPE_STRING",
231                 "DATA_TYPE_BYTE_ARRAY",
232                 "DATA_TYPE_INT16_ARRAY",
233                 "DATA_TYPE_UINT16_ARRAY",
234                 "DATA_TYPE_INT32_ARRAY",
235                 "DATA_TYPE_UINT32_ARRAY",
236                 "DATA_TYPE_INT64_ARRAY",
237                 "DATA_TYPE_UINT64_ARRAY",
238                 "DATA_TYPE_STRING_ARRAY",
239                 "DATA_TYPE_HRTIME",
240                 "DATA_TYPE_NVLIST",
241                 "DATA_TYPE_NVLIST_ARRAY",
242                 "DATA_TYPE_BOOLEAN_VALUE",
243                 "DATA_TYPE_INT8",
244                 "DATA_TYPE_UINT8",
245                 "DATA_TYPE_BOOLEAN_ARRAY",
246                 "DATA_TYPE_INT8_ARRAY",
247                 "DATA_TYPE_UINT8_ARRAY"
248         };
249
250         unsigned int i, j;
251         const unsigned char *p, *pair;
252         int junk;
253         int encoded_size, decoded_size;
254
255         p = nvlist;
256         xdr_int(&p, &junk);
257         xdr_int(&p, &junk);
258
259         pair = p;
260         xdr_int(&p, &encoded_size);
261         xdr_int(&p, &decoded_size);
262         while (encoded_size && decoded_size) {
263                 int namelen, pairtype, elements;
264                 const char *pairname;
265
266                 xdr_int(&p, &namelen);
267                 pairname = (const char*) p;
268                 p += roundup(namelen, 4);
269                 xdr_int(&p, &pairtype);
270
271                 for (i = 0; i < indent; i++)
272                         printf(" ");
273                 printf("%s %s", typenames[pairtype], pairname);
274
275                 xdr_int(&p, &elements);
276                 switch (pairtype) {
277                 case DATA_TYPE_UINT64: {
278                         uint64_t val;
279                         xdr_uint64_t(&p, &val);
280                         printf(" = 0x%llx\n", val);
281                         break;
282                 }
283
284                 case DATA_TYPE_STRING: {
285                         int len;
286                         xdr_int(&p, &len);
287                         printf(" = \"%s\"\n", p);
288                         break;
289                 }
290
291                 case DATA_TYPE_NVLIST:
292                         printf("\n");
293                         nvlist_print(p, indent + 1);
294                         break;
295
296                 case DATA_TYPE_NVLIST_ARRAY:
297                         for (j = 0; j < elements; j++) {
298                                 printf("[%d]\n", j);
299                                 p = nvlist_print(p, indent + 1);
300                                 if (j != elements - 1) {
301                                         for (i = 0; i < indent; i++)
302                                                 printf(" ");
303                                         printf("%s %s", typenames[pairtype], pairname);
304                                 }
305                         }
306                         break;
307
308                 default:
309                         printf("\n");
310                 }
311
312                 p = pair + encoded_size;
313
314                 pair = p;
315                 xdr_int(&p, &encoded_size);
316                 xdr_int(&p, &decoded_size);
317         }
318
319         return p;
320 }
321
322 #endif
323
324 static int
325 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
326     off_t offset, size_t size)
327 {
328         size_t psize;
329         int rc;
330
331         if (bp) {
332                 psize = BP_GET_PSIZE(bp);
333         } else {
334                 psize = size;
335         }
336
337         /*printf("ZFS: reading %d bytes at 0x%llx to %p\n", psize, offset, buf);*/
338         rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
339         if (rc)
340                 return (rc);
341         if (bp && zio_checksum_error(bp, buf))
342                 return (EIO);
343
344         return (0);
345 }
346
347 static int
348 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
349     off_t offset, size_t bytes)
350 {
351
352         return (vdev_read_phys(vdev, bp, buf,
353                 offset + VDEV_LABEL_START_SIZE, bytes));
354 }
355
356
357 static int
358 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
359     off_t offset, size_t bytes)
360 {
361         vdev_t *kid;
362         int rc;
363
364         rc = EIO;
365         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
366                 if (kid->v_state != VDEV_STATE_HEALTHY)
367                         continue;
368                 rc = kid->v_read(kid, bp, buf, offset, bytes);
369                 if (!rc)
370                         return (0);
371         }
372
373         return (rc);
374 }
375
376 static vdev_t *
377 vdev_find(uint64_t guid)
378 {
379         vdev_t *vdev;
380
381         STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
382                 if (vdev->v_guid == guid)
383                         return (vdev);
384
385         return (0);
386 }
387
388 static vdev_t *
389 vdev_create(uint64_t guid, vdev_read_t *read)
390 {
391         vdev_t *vdev;
392
393         vdev = malloc(sizeof(vdev_t));
394         memset(vdev, 0, sizeof(vdev_t));
395         STAILQ_INIT(&vdev->v_children);
396         vdev->v_guid = guid;
397         vdev->v_state = VDEV_STATE_OFFLINE;
398         vdev->v_read = read;
399         vdev->v_phys_read = 0;
400         vdev->v_read_priv = 0;
401         STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
402
403         return (vdev);
404 }
405
406 static int
407 vdev_init_from_nvlist(const unsigned char *nvlist, vdev_t **vdevp, int is_newer)
408 {
409         int rc;
410         uint64_t guid, id, ashift, nparity;
411         const char *type;
412         const char *path;
413         vdev_t *vdev, *kid;
414         const unsigned char *kids;
415         int nkids, i, is_new;
416         uint64_t is_offline, is_faulted, is_degraded, is_removed;
417
418         if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID,
419                         DATA_TYPE_UINT64, 0, &guid)
420             || nvlist_find(nvlist, ZPOOL_CONFIG_ID,
421                            DATA_TYPE_UINT64, 0, &id)
422             || nvlist_find(nvlist, ZPOOL_CONFIG_TYPE,
423                            DATA_TYPE_STRING, 0, &type)) {
424                 printf("ZFS: can't find vdev details\n");
425                 return (ENOENT);
426         }
427
428         if (strcmp(type, VDEV_TYPE_MIRROR)
429             && strcmp(type, VDEV_TYPE_DISK)
430             && strcmp(type, VDEV_TYPE_RAIDZ)) {
431                 printf("ZFS: can only boot from disk, mirror or raidz vdevs\n");
432                 return (EIO);
433         }
434
435         is_offline = is_removed = is_faulted = is_degraded = 0;
436
437         nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, 0,
438                         &is_offline);
439         nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, 0,
440                         &is_removed);
441         nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, 0,
442                         &is_faulted);
443         nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64, 0,
444                         &is_degraded);
445
446         vdev = vdev_find(guid);
447         if (!vdev) {
448                 is_new = 1;
449
450                 if (!strcmp(type, VDEV_TYPE_MIRROR))
451                         vdev = vdev_create(guid, vdev_mirror_read);
452                 else if (!strcmp(type, VDEV_TYPE_RAIDZ))
453                         vdev = vdev_create(guid, vdev_raidz_read);
454                 else
455                         vdev = vdev_create(guid, vdev_disk_read);
456
457                 vdev->v_id = id;
458                 if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
459                         DATA_TYPE_UINT64, 0, &ashift) == 0)
460                         vdev->v_ashift = ashift;
461                 else
462                         vdev->v_ashift = 0;
463                 if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
464                         DATA_TYPE_UINT64, 0, &nparity) == 0)
465                         vdev->v_nparity = nparity;
466                 else
467                         vdev->v_nparity = 0;
468                 if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
469                                 DATA_TYPE_STRING, 0, &path) == 0) {
470                         if (strlen(path) > 5
471                             && path[0] == '/'
472                             && path[1] == 'd'
473                             && path[2] == 'e'
474                             && path[3] == 'v'
475                             && path[4] == '/')
476                                 path += 5;
477                         vdev->v_name = strdup(path);
478                 } else {
479                         if (!strcmp(type, "raidz")) {
480                                 if (vdev->v_nparity == 1)
481                                         vdev->v_name = "raidz1";
482                                 else
483                                         vdev->v_name = "raidz2";
484                         } else {
485                                 vdev->v_name = strdup(type);
486                         }
487                 }
488
489                 if (is_offline)
490                         vdev->v_state = VDEV_STATE_OFFLINE;
491                 else if (is_removed)
492                         vdev->v_state = VDEV_STATE_REMOVED;
493                 else if (is_faulted)
494                         vdev->v_state = VDEV_STATE_FAULTED;
495                 else if (is_degraded)
496                         vdev->v_state = VDEV_STATE_DEGRADED;
497                 else
498                         vdev->v_state = VDEV_STATE_HEALTHY;
499         } else {
500                 is_new = 0;
501
502                 if (is_newer) {
503                         /*
504                          * We've already seen this vdev, but from an older
505                          * vdev label, so let's refresh its state from the
506                          * newer label.
507                          */
508                         if (is_offline)
509                                 vdev->v_state = VDEV_STATE_OFFLINE;
510                         else if (is_removed)
511                                 vdev->v_state = VDEV_STATE_REMOVED;
512                         else if (is_faulted)
513                                 vdev->v_state = VDEV_STATE_FAULTED;
514                         else if (is_degraded)
515                                 vdev->v_state = VDEV_STATE_DEGRADED;
516                         else
517                                 vdev->v_state = VDEV_STATE_HEALTHY;
518                 }
519         }
520
521         rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN,
522                          DATA_TYPE_NVLIST_ARRAY, &nkids, &kids);
523         /*
524          * Its ok if we don't have any kids.
525          */
526         if (rc == 0) {
527                 vdev->v_nchildren = nkids;
528                 for (i = 0; i < nkids; i++) {
529                         rc = vdev_init_from_nvlist(kids, &kid, is_newer);
530                         if (rc)
531                                 return (rc);
532                         if (is_new)
533                                 STAILQ_INSERT_TAIL(&vdev->v_children, kid,
534                                                    v_childlink);
535                         kids = nvlist_next(kids);
536                 }
537         } else {
538                 vdev->v_nchildren = 0;
539         }
540
541         if (vdevp)
542                 *vdevp = vdev;
543         return (0);
544 }
545
546 static void
547 vdev_set_state(vdev_t *vdev)
548 {
549         vdev_t *kid;
550         int good_kids;
551         int bad_kids;
552
553         /*
554          * A mirror or raidz is healthy if all its kids are healthy. A
555          * mirror is degraded if any of its kids is healthy; a raidz
556          * is degraded if at most nparity kids are offline.
557          */
558         if (STAILQ_FIRST(&vdev->v_children)) {
559                 good_kids = 0;
560                 bad_kids = 0;
561                 STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
562                         if (kid->v_state == VDEV_STATE_HEALTHY)
563                                 good_kids++;
564                         else
565                                 bad_kids++;
566                 }
567                 if (bad_kids == 0) {
568                         vdev->v_state = VDEV_STATE_HEALTHY;
569                 } else {
570                         if (vdev->v_read == vdev_mirror_read) {
571                                 if (good_kids) {
572                                         vdev->v_state = VDEV_STATE_DEGRADED;
573                                 } else {
574                                         vdev->v_state = VDEV_STATE_OFFLINE;
575                                 }
576                         } else if (vdev->v_read == vdev_raidz_read) {
577                                 if (bad_kids > vdev->v_nparity) {
578                                         vdev->v_state = VDEV_STATE_OFFLINE;
579                                 } else {
580                                         vdev->v_state = VDEV_STATE_DEGRADED;
581                                 }
582                         }
583                 }
584         }
585 }
586
587 static spa_t *
588 spa_find_by_guid(uint64_t guid)
589 {
590         spa_t *spa;
591
592         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
593                 if (spa->spa_guid == guid)
594                         return (spa);
595
596         return (0);
597 }
598
599 #ifdef BOOT2
600
601 static spa_t *
602 spa_find_by_name(const char *name)
603 {
604         spa_t *spa;
605
606         STAILQ_FOREACH(spa, &zfs_pools, spa_link)
607                 if (!strcmp(spa->spa_name, name))
608                         return (spa);
609
610         return (0);
611 }
612
613 #endif
614
615 static spa_t *
616 spa_create(uint64_t guid)
617 {
618         spa_t *spa;
619
620         spa = malloc(sizeof(spa_t));
621         memset(spa, 0, sizeof(spa_t));
622         STAILQ_INIT(&spa->spa_vdevs);
623         spa->spa_guid = guid;
624         STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
625
626         return (spa);
627 }
628
629 static const char *
630 state_name(vdev_state_t state)
631 {
632         static const char* names[] = {
633                 "UNKNOWN",
634                 "CLOSED",
635                 "OFFLINE",
636                 "REMOVED",
637                 "CANT_OPEN",
638                 "FAULTED",
639                 "DEGRADED",
640                 "ONLINE"
641         };
642         return names[state];
643 }
644
645 #ifdef BOOT2
646
647 #define pager_printf printf
648
649 #else
650
651 static void
652 pager_printf(const char *fmt, ...)
653 {
654         char line[80];
655         va_list args;
656
657         va_start(args, fmt);
658         vsprintf(line, fmt, args);
659         va_end(args);
660         pager_output(line);
661 }
662
663 #endif
664
665 #define STATUS_FORMAT   "        %-16s %-10s\n"
666
667 static void
668 print_state(int indent, const char *name, vdev_state_t state)
669 {
670         int i;
671         char buf[512];
672
673         buf[0] = 0;
674         for (i = 0; i < indent; i++)
675                 strcat(buf, "  ");
676         strcat(buf, name);
677         pager_printf(STATUS_FORMAT, buf, state_name(state));
678         
679 }
680
681 static void
682 vdev_status(vdev_t *vdev, int indent)
683 {
684         vdev_t *kid;
685         print_state(indent, vdev->v_name, vdev->v_state);
686
687         STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
688                 vdev_status(kid, indent + 1);
689         }
690 }
691
692 static void
693 spa_status(spa_t *spa)
694 {
695         vdev_t *vdev;
696         int good_kids, bad_kids, degraded_kids;
697         vdev_state_t state;
698
699         pager_printf("  pool: %s\n", spa->spa_name);
700         pager_printf("config:\n\n");
701         pager_printf(STATUS_FORMAT, "NAME", "STATE");
702
703         good_kids = 0;
704         degraded_kids = 0;
705         bad_kids = 0;
706         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
707                 if (vdev->v_state == VDEV_STATE_HEALTHY)
708                         good_kids++;
709                 else if (vdev->v_state == VDEV_STATE_DEGRADED)
710                         degraded_kids++;
711                 else
712                         bad_kids++;
713         }
714
715         state = VDEV_STATE_CLOSED;
716         if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
717                 state = VDEV_STATE_HEALTHY;
718         else if ((good_kids + degraded_kids) > 0)
719                 state = VDEV_STATE_DEGRADED;
720
721         print_state(0, spa->spa_name, state);
722         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink) {
723                 vdev_status(vdev, 1);
724         }
725 }
726
727 static void
728 spa_all_status(void)
729 {
730         spa_t *spa;
731         int first = 1;
732
733         STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
734                 if (!first)
735                         pager_printf("\n");
736                 first = 0;
737                 spa_status(spa);
738         }
739 }
740
741 static int
742 vdev_probe(vdev_phys_read_t *read, void *read_priv, spa_t **spap)
743 {
744         vdev_t vtmp;
745         vdev_phys_t *vdev_label = (vdev_phys_t *) zap_scratch;
746         spa_t *spa;
747         vdev_t *vdev, *top_vdev, *pool_vdev;
748         off_t off;
749         blkptr_t bp;
750         const unsigned char *nvlist;
751         uint64_t val;
752         uint64_t guid;
753         uint64_t pool_txg, pool_guid;
754         const char *pool_name;
755         const unsigned char *vdevs;
756         int i, rc, is_newer;
757         char upbuf[1024];
758         const struct uberblock *up;
759
760         /*
761          * Load the vdev label and figure out which
762          * uberblock is most current.
763          */
764         memset(&vtmp, 0, sizeof(vtmp));
765         vtmp.v_phys_read = read;
766         vtmp.v_read_priv = read_priv;
767         off = offsetof(vdev_label_t, vl_vdev_phys);
768         BP_ZERO(&bp);
769         BP_SET_LSIZE(&bp, sizeof(vdev_phys_t));
770         BP_SET_PSIZE(&bp, sizeof(vdev_phys_t));
771         BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
772         BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
773         ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
774         if (vdev_read_phys(&vtmp, &bp, vdev_label, off, 0))
775                 return (EIO);
776
777         if (vdev_label->vp_nvlist[0] != NV_ENCODE_XDR) {
778                 return (EIO);
779         }
780
781         nvlist = (const unsigned char *) vdev_label->vp_nvlist + 4;
782
783         if (nvlist_find(nvlist,
784                         ZPOOL_CONFIG_VERSION,
785                         DATA_TYPE_UINT64, 0, &val)) {
786                 return (EIO);
787         }
788
789         if (val > SPA_VERSION) {
790                 printf("ZFS: unsupported ZFS version %u (should be %u)\n",
791                     (unsigned) val, (unsigned) SPA_VERSION);
792                 return (EIO);
793         }
794
795         if (nvlist_find(nvlist,
796                         ZPOOL_CONFIG_POOL_STATE,
797                         DATA_TYPE_UINT64, 0, &val)) {
798                 return (EIO);
799         }
800
801 #ifndef TEST
802         if (val != POOL_STATE_ACTIVE) {
803                 /*
804                  * Don't print a message here. If we happen to reboot
805                  * while where is an exported pool around, we don't
806                  * need a cascade of confusing messages during boot.
807                  */
808                 /*printf("ZFS: pool is not active\n");*/
809                 return (EIO);
810         }
811 #endif
812
813         if (nvlist_find(nvlist,
814                         ZPOOL_CONFIG_POOL_TXG,
815                         DATA_TYPE_UINT64, 0, &pool_txg)
816             || nvlist_find(nvlist,
817                            ZPOOL_CONFIG_POOL_GUID,
818                            DATA_TYPE_UINT64, 0, &pool_guid)
819             || nvlist_find(nvlist,
820                            ZPOOL_CONFIG_POOL_NAME,
821                            DATA_TYPE_STRING, 0, &pool_name)) {
822                 /*
823                  * Cache and spare devices end up here - just ignore
824                  * them.
825                  */
826                 /*printf("ZFS: can't find pool details\n");*/
827                 return (EIO);
828         }
829
830         /*
831          * Create the pool if this is the first time we've seen it.
832          */
833         spa = spa_find_by_guid(pool_guid);
834         if (!spa) {
835                 spa = spa_create(pool_guid);
836                 spa->spa_name = strdup(pool_name);
837         }
838         if (pool_txg > spa->spa_txg) {
839                 spa->spa_txg = pool_txg;
840                 is_newer = 1;
841         } else
842                 is_newer = 0;
843
844         /*
845          * Get the vdev tree and create our in-core copy of it.
846          * If we already have a vdev with this guid, this must
847          * be some kind of alias (overlapping slices, dangerously dedicated
848          * disks etc).
849          */
850         if (nvlist_find(nvlist,
851                         ZPOOL_CONFIG_GUID,
852                         DATA_TYPE_UINT64, 0, &guid)) {
853                 return (EIO);
854         }
855         vdev = vdev_find(guid);
856         if (vdev && vdev->v_phys_read)  /* Has this vdev already been inited? */
857                 return (EIO);
858
859         if (nvlist_find(nvlist,
860                         ZPOOL_CONFIG_VDEV_TREE,
861                         DATA_TYPE_NVLIST, 0, &vdevs)) {
862                 return (EIO);
863         }
864
865         rc = vdev_init_from_nvlist(vdevs, &top_vdev, is_newer);
866         if (rc)
867                 return (rc);
868
869         /*
870          * Add the toplevel vdev to the pool if its not already there.
871          */
872         STAILQ_FOREACH(pool_vdev, &spa->spa_vdevs, v_childlink)
873                 if (top_vdev == pool_vdev)
874                         break;
875         if (!pool_vdev && top_vdev)
876                 STAILQ_INSERT_TAIL(&spa->spa_vdevs, top_vdev, v_childlink);
877
878         /*
879          * We should already have created an incomplete vdev for this
880          * vdev. Find it and initialise it with our read proc.
881          */
882         vdev = vdev_find(guid);
883         if (vdev) {
884                 vdev->v_phys_read = read;
885                 vdev->v_read_priv = read_priv;
886         } else {
887                 printf("ZFS: inconsistent nvlist contents\n");
888                 return (EIO);
889         }
890
891         /*
892          * Re-evaluate top-level vdev state.
893          */
894         vdev_set_state(top_vdev);
895
896         /*
897          * Ok, we are happy with the pool so far. Lets find
898          * the best uberblock and then we can actually access
899          * the contents of the pool.
900          */
901         for (i = 0;
902              i < VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT;
903              i++) {
904                 off = offsetof(vdev_label_t, vl_uberblock);
905                 off += i << UBERBLOCK_SHIFT;
906                 BP_ZERO(&bp);
907                 DVA_SET_OFFSET(&bp.blk_dva[0], off);
908                 BP_SET_LSIZE(&bp, 1 << UBERBLOCK_SHIFT);
909                 BP_SET_PSIZE(&bp, 1 << UBERBLOCK_SHIFT);
910                 BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
911                 BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
912                 ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
913                 if (vdev_read_phys(vdev, &bp, upbuf, off, 0))
914                         continue;
915
916                 up = (const struct uberblock *) upbuf;
917                 if (up->ub_magic != UBERBLOCK_MAGIC)
918                         continue;
919                 if (up->ub_txg < spa->spa_txg)
920                         continue;
921                 if (up->ub_txg > spa->spa_uberblock.ub_txg) {
922                         spa->spa_uberblock = *up;
923                 } else if (up->ub_txg == spa->spa_uberblock.ub_txg) {
924                         if (up->ub_timestamp > spa->spa_uberblock.ub_timestamp)
925                                 spa->spa_uberblock = *up;
926                 }
927         }
928
929         if (spap)
930                 *spap = spa;
931         return (0);
932 }
933
934 static int
935 ilog2(int n)
936 {
937         int v;
938
939         for (v = 0; v < 32; v++)
940                 if (n == (1 << v))
941                         return v;
942         return -1;
943 }
944
945 static int
946 zio_read_gang(spa_t *spa, const blkptr_t *bp, const dva_t *dva, void *buf)
947 {
948         zio_gbh_phys_t zio_gb;
949         vdev_t *vdev;
950         int vdevid;
951         off_t offset;
952         int i;
953
954         vdevid = DVA_GET_VDEV(dva);
955         offset = DVA_GET_OFFSET(dva);
956         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink)
957                 if (vdev->v_id == vdevid)
958                         break;
959         if (!vdev || !vdev->v_read)
960                 return (EIO);
961         if (vdev->v_read(vdev, bp, &zio_gb, offset, SPA_GANGBLOCKSIZE))
962                 return (EIO);
963
964         for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
965                 if (zio_read(spa, &zio_gb.zg_blkptr[i], buf))
966                         return (EIO);
967         }
968  
969         return (0);
970 }
971
972 static int
973 zio_read(spa_t *spa, const blkptr_t *bp, void *buf)
974 {
975         int cpfunc = BP_GET_COMPRESS(bp);
976         size_t lsize = BP_GET_LSIZE(bp);
977         size_t psize = BP_GET_PSIZE(bp);
978         void *pbuf;
979         int i;
980
981         zfs_reset_temp();
982         if (cpfunc != ZIO_COMPRESS_OFF)
983                 pbuf = zfs_alloc_temp(psize);
984         else
985                 pbuf = buf;
986
987         for (i = 0; i < SPA_DVAS_PER_BP; i++) {
988                 const dva_t *dva = &bp->blk_dva[i];
989                 vdev_t *vdev;
990                 int vdevid;
991                 off_t offset;
992
993                 if (!dva->dva_word[0] && !dva->dva_word[1])
994                         continue;
995
996                 if (DVA_GET_GANG(dva)) {
997                         printf("ZFS: gang block detected!\n");
998                         if (zio_read_gang(spa, bp, dva, buf))
999                                 return (EIO); 
1000                 } else {
1001                         vdevid = DVA_GET_VDEV(dva);
1002                         offset = DVA_GET_OFFSET(dva);
1003                         STAILQ_FOREACH(vdev, &spa->spa_vdevs, v_childlink)
1004                                 if (vdev->v_id == vdevid)
1005                                         break;
1006                         if (!vdev || !vdev->v_read) {
1007                                 continue;
1008                         }
1009                         if (vdev->v_read(vdev, bp, pbuf, offset, psize))
1010                                 continue;
1011
1012                         if (cpfunc != ZIO_COMPRESS_OFF) {
1013                                 if (zio_decompress_data(cpfunc, pbuf, psize,
1014                                     buf, lsize))
1015                                         return (EIO);
1016                         }
1017                 }
1018
1019                 return (0);
1020         }
1021         printf("ZFS: i/o error - all block copies unavailable\n");
1022
1023         return (EIO);
1024 }
1025
1026 static int
1027 dnode_read(spa_t *spa, const dnode_phys_t *dnode, off_t offset, void *buf, size_t buflen)
1028 {
1029         int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
1030         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1031         int nlevels = dnode->dn_nlevels;
1032         int i, rc;
1033
1034         /*
1035          * Note: bsize may not be a power of two here so we need to do an
1036          * actual divide rather than a bitshift.
1037          */
1038         while (buflen > 0) {
1039                 uint64_t bn = offset / bsize;
1040                 int boff = offset % bsize;
1041                 int ibn;
1042                 const blkptr_t *indbp;
1043                 blkptr_t bp;
1044
1045                 if (bn > dnode->dn_maxblkid)
1046                         return (EIO);
1047
1048                 if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
1049                         goto cached;
1050
1051                 indbp = dnode->dn_blkptr;
1052                 for (i = 0; i < nlevels; i++) {
1053                         /*
1054                          * Copy the bp from the indirect array so that
1055                          * we can re-use the scratch buffer for multi-level
1056                          * objects.
1057                          */
1058                         ibn = bn >> ((nlevels - i - 1) * ibshift);
1059                         ibn &= ((1 << ibshift) - 1);
1060                         bp = indbp[ibn];
1061                         rc = zio_read(spa, &bp, dnode_cache_buf);
1062                         if (rc)
1063                                 return (rc);
1064                         indbp = (const blkptr_t *) dnode_cache_buf;
1065                 }
1066                 dnode_cache_obj = dnode;
1067                 dnode_cache_bn = bn;
1068         cached:
1069
1070                 /*
1071                  * The buffer contains our data block. Copy what we
1072                  * need from it and loop.
1073                  */ 
1074                 i = bsize - boff;
1075                 if (i > buflen) i = buflen;
1076                 memcpy(buf, &dnode_cache_buf[boff], i);
1077                 buf = ((char*) buf) + i;
1078                 offset += i;
1079                 buflen -= i;
1080         }
1081
1082         return (0);
1083 }
1084
1085 /*
1086  * Lookup a value in a microzap directory. Assumes that the zap
1087  * scratch buffer contains the directory contents.
1088  */
1089 static int
1090 mzap_lookup(spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1091 {
1092         const mzap_phys_t *mz;
1093         const mzap_ent_phys_t *mze;
1094         size_t size;
1095         int chunks, i;
1096
1097         /*
1098          * Microzap objects use exactly one block. Read the whole
1099          * thing.
1100          */
1101         size = dnode->dn_datablkszsec * 512;
1102
1103         mz = (const mzap_phys_t *) zap_scratch;
1104         chunks = size / MZAP_ENT_LEN - 1;
1105
1106         for (i = 0; i < chunks; i++) {
1107                 mze = &mz->mz_chunk[i];
1108                 if (!strcmp(mze->mze_name, name)) {
1109                         *value = mze->mze_value;
1110                         return (0);
1111                 }
1112         }
1113
1114         return (ENOENT);
1115 }
1116
1117 /*
1118  * Compare a name with a zap leaf entry. Return non-zero if the name
1119  * matches.
1120  */
1121 static int
1122 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, const char *name)
1123 {
1124         size_t namelen;
1125         const zap_leaf_chunk_t *nc;
1126         const char *p;
1127
1128         namelen = zc->l_entry.le_name_length;
1129                         
1130         nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
1131         p = name;
1132         while (namelen > 0) {
1133                 size_t len;
1134                 len = namelen;
1135                 if (len > ZAP_LEAF_ARRAY_BYTES)
1136                         len = ZAP_LEAF_ARRAY_BYTES;
1137                 if (memcmp(p, nc->l_array.la_array, len))
1138                         return (0);
1139                 p += len;
1140                 namelen -= len;
1141                 nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
1142         }
1143
1144         return 1;
1145 }
1146
1147 /*
1148  * Extract a uint64_t value from a zap leaf entry.
1149  */
1150 static uint64_t
1151 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
1152 {
1153         const zap_leaf_chunk_t *vc;
1154         int i;
1155         uint64_t value;
1156         const uint8_t *p;
1157
1158         vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
1159         for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
1160                 value = (value << 8) | p[i];
1161         }
1162
1163         return value;
1164 }
1165
1166 /*
1167  * Lookup a value in a fatzap directory. Assumes that the zap scratch
1168  * buffer contains the directory header.
1169  */
1170 static int
1171 fzap_lookup(spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1172 {
1173         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1174         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1175         fat_zap_t z;
1176         uint64_t *ptrtbl;
1177         uint64_t hash;
1178         int rc;
1179
1180         if (zh.zap_magic != ZAP_MAGIC)
1181                 return (EIO);
1182
1183         z.zap_block_shift = ilog2(bsize);
1184         z.zap_phys = (zap_phys_t *) zap_scratch;
1185
1186         /*
1187          * Figure out where the pointer table is and read it in if necessary.
1188          */
1189         if (zh.zap_ptrtbl.zt_blk) {
1190                 rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
1191                                zap_scratch, bsize);
1192                 if (rc)
1193                         return (rc);
1194                 ptrtbl = (uint64_t *) zap_scratch;
1195         } else {
1196                 ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
1197         }
1198
1199         hash = zap_hash(zh.zap_salt, name);
1200
1201         zap_leaf_t zl;
1202         zl.l_bs = z.zap_block_shift;
1203
1204         off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
1205         zap_leaf_chunk_t *zc;
1206
1207         rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
1208         if (rc)
1209                 return (rc);
1210
1211         zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1212
1213         /*
1214          * Make sure this chunk matches our hash.
1215          */
1216         if (zl.l_phys->l_hdr.lh_prefix_len > 0
1217             && zl.l_phys->l_hdr.lh_prefix
1218             != hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
1219                 return (ENOENT);
1220
1221         /*
1222          * Hash within the chunk to find our entry.
1223          */
1224         int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) - zl.l_phys->l_hdr.lh_prefix_len);
1225         int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
1226         h = zl.l_phys->l_hash[h];
1227         if (h == 0xffff)
1228                 return (ENOENT);
1229         zc = &ZAP_LEAF_CHUNK(&zl, h);
1230         while (zc->l_entry.le_hash != hash) {
1231                 if (zc->l_entry.le_next == 0xffff) {
1232                         zc = 0;
1233                         break;
1234                 }
1235                 zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
1236         }
1237         if (fzap_name_equal(&zl, zc, name)) {
1238                 *value = fzap_leaf_value(&zl, zc);
1239                 return (0);
1240         }
1241
1242         return (ENOENT);
1243 }
1244
1245 /*
1246  * Lookup a name in a zap object and return its value as a uint64_t.
1247  */
1248 static int
1249 zap_lookup(spa_t *spa, const dnode_phys_t *dnode, const char *name, uint64_t *value)
1250 {
1251         int rc;
1252         uint64_t zap_type;
1253         size_t size = dnode->dn_datablkszsec * 512;
1254
1255         rc = dnode_read(spa, dnode, 0, zap_scratch, size);
1256         if (rc)
1257                 return (rc);
1258
1259         zap_type = *(uint64_t *) zap_scratch;
1260         if (zap_type == ZBT_MICRO)
1261                 return mzap_lookup(spa, dnode, name, value);
1262         else
1263                 return fzap_lookup(spa, dnode, name, value);
1264 }
1265
1266 #ifdef BOOT2
1267
1268 /*
1269  * List a microzap directory. Assumes that the zap scratch buffer contains
1270  * the directory contents.
1271  */
1272 static int
1273 mzap_list(spa_t *spa, const dnode_phys_t *dnode)
1274 {
1275         const mzap_phys_t *mz;
1276         const mzap_ent_phys_t *mze;
1277         size_t size;
1278         int chunks, i;
1279
1280         /*
1281          * Microzap objects use exactly one block. Read the whole
1282          * thing.
1283          */
1284         size = dnode->dn_datablkszsec * 512;
1285         mz = (const mzap_phys_t *) zap_scratch;
1286         chunks = size / MZAP_ENT_LEN - 1;
1287
1288         for (i = 0; i < chunks; i++) {
1289                 mze = &mz->mz_chunk[i];
1290                 if (mze->mze_name[0])
1291                         //printf("%-32s 0x%llx\n", mze->mze_name, mze->mze_value);
1292                         printf("%s\n", mze->mze_name);
1293         }
1294
1295         return (0);
1296 }
1297
1298 /*
1299  * List a fatzap directory. Assumes that the zap scratch buffer contains
1300  * the directory header.
1301  */
1302 static int
1303 fzap_list(spa_t *spa, const dnode_phys_t *dnode)
1304 {
1305         int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1306         zap_phys_t zh = *(zap_phys_t *) zap_scratch;
1307         fat_zap_t z;
1308         int i, j;
1309
1310         if (zh.zap_magic != ZAP_MAGIC)
1311                 return (EIO);
1312
1313         z.zap_block_shift = ilog2(bsize);
1314         z.zap_phys = (zap_phys_t *) zap_scratch;
1315
1316         /*
1317          * This assumes that the leaf blocks start at block 1. The
1318          * documentation isn't exactly clear on this.
1319          */
1320         zap_leaf_t zl;
1321         zl.l_bs = z.zap_block_shift;
1322         for (i = 0; i < zh.zap_num_leafs; i++) {
1323                 off_t off = (i + 1) << zl.l_bs;
1324                 char name[256], *p;
1325                 uint64_t value;
1326
1327                 if (dnode_read(spa, dnode, off, zap_scratch, bsize))
1328                         return (EIO);
1329
1330                 zl.l_phys = (zap_leaf_phys_t *) zap_scratch;
1331
1332                 for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
1333                         zap_leaf_chunk_t *zc, *nc;
1334                         int namelen;
1335
1336                         zc = &ZAP_LEAF_CHUNK(&zl, j);
1337                         if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
1338                                 continue;
1339                         namelen = zc->l_entry.le_name_length;
1340                         if (namelen > sizeof(name))
1341                                 namelen = sizeof(name);
1342                         
1343                         /*
1344                          * Paste the name back together.
1345                          */
1346                         nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
1347                         p = name;
1348                         while (namelen > 0) {
1349                                 int len;
1350                                 len = namelen;
1351                                 if (len > ZAP_LEAF_ARRAY_BYTES)
1352                                         len = ZAP_LEAF_ARRAY_BYTES;
1353                                 memcpy(p, nc->l_array.la_array, len);
1354                                 p += len;
1355                                 namelen -= len;
1356                                 nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
1357                         }
1358
1359                         /*
1360                          * Assume the first eight bytes of the value are
1361                          * a uint64_t.
1362                          */
1363                         value = fzap_leaf_value(&zl, zc);
1364
1365                         printf("%-32s 0x%llx\n", name, value);
1366                 }
1367         }
1368
1369         return (0);
1370 }
1371
1372 /*
1373  * List a zap directory.
1374  */
1375 static int
1376 zap_list(spa_t *spa, const dnode_phys_t *dnode)
1377 {
1378         uint64_t zap_type;
1379         size_t size = dnode->dn_datablkszsec * 512;
1380
1381         if (dnode_read(spa, dnode, 0, zap_scratch, size))
1382                 return (EIO);
1383
1384         zap_type = *(uint64_t *) zap_scratch;
1385         if (zap_type == ZBT_MICRO)
1386                 return mzap_list(spa, dnode);
1387         else
1388                 return fzap_list(spa, dnode);
1389 }
1390
1391 #endif
1392
1393 static int
1394 objset_get_dnode(spa_t *spa, const objset_phys_t *os, uint64_t objnum, dnode_phys_t *dnode)
1395 {
1396         off_t offset;
1397
1398         offset = objnum * sizeof(dnode_phys_t);
1399         return dnode_read(spa, &os->os_meta_dnode, offset,
1400                 dnode, sizeof(dnode_phys_t));
1401 }
1402
1403 /*
1404  * Find the object set given the object number of its dataset object
1405  * and return its details in *objset
1406  */
1407 static int
1408 zfs_mount_dataset(spa_t *spa, uint64_t objnum, objset_phys_t *objset)
1409 {
1410         dnode_phys_t dataset;
1411         dsl_dataset_phys_t *ds;
1412
1413         if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
1414                 printf("ZFS: can't find dataset %llu\n", objnum);
1415                 return (EIO);
1416         }
1417
1418         ds = (dsl_dataset_phys_t *) &dataset.dn_bonus;
1419         if (zio_read(spa, &ds->ds_bp, objset)) {
1420                 printf("ZFS: can't read object set for dataset %llu\n", objnum);
1421                 return (EIO);
1422         }
1423
1424         return (0);
1425 }
1426
1427 /*
1428  * Find the object set pointed to by the BOOTFS property or the root
1429  * dataset if there is none and return its details in *objset
1430  */
1431 static int
1432 zfs_mount_root(spa_t *spa, objset_phys_t *objset)
1433 {
1434         dnode_phys_t dir, propdir;
1435         uint64_t props, bootfs, root;
1436
1437         /*
1438          * Start with the MOS directory object.
1439          */
1440         if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT, &dir)) {
1441                 printf("ZFS: can't read MOS object directory\n");
1442                 return (EIO);
1443         }
1444
1445         /*
1446          * Lookup the pool_props and see if we can find a bootfs.
1447          */
1448         if (zap_lookup(spa, &dir, DMU_POOL_PROPS, &props) == 0
1449              && objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0
1450              && zap_lookup(spa, &propdir, "bootfs", &bootfs) == 0
1451              && bootfs != 0)
1452                 return zfs_mount_dataset(spa, bootfs, objset);
1453
1454         /*
1455          * Lookup the root dataset directory
1456          */
1457         if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, &root)
1458             || objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
1459                 printf("ZFS: can't find root dsl_dir\n");
1460                 return (EIO);
1461         }
1462
1463         /*
1464          * Use the information from the dataset directory's bonus buffer
1465          * to find the dataset object and from that the object set itself.
1466          */
1467         dsl_dir_phys_t *dd = (dsl_dir_phys_t *) &dir.dn_bonus;
1468         return zfs_mount_dataset(spa, dd->dd_head_dataset_obj, objset);
1469 }
1470
1471 static int
1472 zfs_mount_pool(spa_t *spa)
1473 {
1474         /*
1475          * Find the MOS and work our way in from there.
1476          */
1477         if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
1478                 printf("ZFS: can't read MOS\n");
1479                 return (EIO);
1480         }
1481
1482         /*
1483          * Find the root object set
1484          */
1485         if (zfs_mount_root(spa, &spa->spa_root_objset)) {
1486                 printf("Can't find root filesystem - giving up\n");
1487                 return (EIO);
1488         }
1489
1490         return (0);
1491 }
1492
1493 /*
1494  * Lookup a file and return its dnode.
1495  */
1496 static int
1497 zfs_lookup(spa_t *spa, const char *upath, dnode_phys_t *dnode)
1498 {
1499         int rc;
1500         uint64_t objnum, rootnum, parentnum;
1501         dnode_phys_t dn;
1502         const znode_phys_t *zp = (const znode_phys_t *) dn.dn_bonus;
1503         const char *p, *q;
1504         char element[256];
1505         char path[1024];
1506         int symlinks_followed = 0;
1507
1508         if (spa->spa_root_objset.os_type != DMU_OST_ZFS) {
1509                 printf("ZFS: unexpected object set type %llu\n",
1510                        spa->spa_root_objset.os_type);
1511                 return (EIO);
1512         }
1513
1514         /*
1515          * Get the root directory dnode.
1516          */
1517         rc = objset_get_dnode(spa, &spa->spa_root_objset, MASTER_NODE_OBJ, &dn);
1518         if (rc)
1519                 return (rc);
1520
1521         rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, &rootnum);
1522         if (rc)
1523                 return (rc);
1524
1525         rc = objset_get_dnode(spa, &spa->spa_root_objset, rootnum, &dn);
1526         if (rc)
1527                 return (rc);
1528
1529         objnum = rootnum;
1530         p = upath;
1531         while (p && *p) {
1532                 while (*p == '/')
1533                         p++;
1534                 if (!*p)
1535                         break;
1536                 q = strchr(p, '/');
1537                 if (q) {
1538                         memcpy(element, p, q - p);
1539                         element[q - p] = 0;
1540                         p = q;
1541                 } else {
1542                         strcpy(element, p);
1543                         p = 0;
1544                 }
1545
1546                 if ((zp->zp_mode >> 12) != 0x4) {
1547                         return (ENOTDIR);
1548                 }
1549
1550                 parentnum = objnum;
1551                 rc = zap_lookup(spa, &dn, element, &objnum);
1552                 if (rc)
1553                         return (rc);
1554                 objnum = ZFS_DIRENT_OBJ(objnum);
1555
1556                 rc = objset_get_dnode(spa, &spa->spa_root_objset, objnum, &dn);
1557                 if (rc)
1558                         return (rc);
1559
1560                 /*
1561                  * Check for symlink.
1562                  */
1563                 if ((zp->zp_mode >> 12) == 0xa) {
1564                         if (symlinks_followed > 10)
1565                                 return (EMLINK);
1566                         symlinks_followed++;
1567
1568                         /*
1569                          * Read the link value and copy the tail of our
1570                          * current path onto the end.
1571                          */
1572                         if (p)
1573                                 strcpy(&path[zp->zp_size], p);
1574                         else
1575                                 path[zp->zp_size] = 0;
1576                         if (zp->zp_size + sizeof(znode_phys_t) <= dn.dn_bonuslen) {
1577                                 memcpy(path, &dn.dn_bonus[sizeof(znode_phys_t)],
1578                                         zp->zp_size);
1579                         } else {
1580                                 rc = dnode_read(spa, &dn, 0, path, zp->zp_size);
1581                                 if (rc)
1582                                         return (rc);
1583                         }
1584
1585                         /*
1586                          * Restart with the new path, starting either at
1587                          * the root or at the parent depending whether or
1588                          * not the link is relative.
1589                          */
1590                         p = path;
1591                         if (*p == '/')
1592                                 objnum = rootnum;
1593                         else
1594                                 objnum = parentnum;
1595                         objset_get_dnode(spa, &spa->spa_root_objset, objnum, &dn);
1596                 }
1597         }
1598
1599         *dnode = dn;
1600         return (0);
1601 }