Deleted Added
sdiff udiff text old ( 296519 ) new ( 299439 )
full compact
zap_micro.c (296519) zap_micro.c (299439)
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 */
27
28#include <sys/zio.h>
29#include <sys/spa.h>
30#include <sys/dmu.h>
31#include <sys/zfs_context.h>
32#include <sys/zap.h>
33#include <sys/refcount.h>
34#include <sys/zap_impl.h>
35#include <sys/zap_leaf.h>
36#include <sys/avl.h>
37#include <sys/arc.h>
38#include <sys/dmu_objset.h>
39
40#ifdef _KERNEL
41#include <sys/sunddi.h>
42#endif
43
44extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
45
46static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
47
48uint64_t
49zap_getflags(zap_t *zap)
50{
51 if (zap->zap_ismicro)
52 return (0);
53 return (zap_f_phys(zap)->zap_flags);
54}
55
56int
57zap_hashbits(zap_t *zap)
58{
59 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
60 return (48);
61 else
62 return (28);
63}
64
65uint32_t
66zap_maxcd(zap_t *zap)
67{
68 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
69 return ((1<<16)-1);
70 else
71 return (-1U);
72}
73
74static uint64_t
75zap_hash(zap_name_t *zn)
76{
77 zap_t *zap = zn->zn_zap;
78 uint64_t h = 0;
79
80 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
81 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
82 h = *(uint64_t *)zn->zn_key_orig;
83 } else {
84 h = zap->zap_salt;
85 ASSERT(h != 0);
86 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
87
88 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
89 int i;
90 const uint64_t *wp = zn->zn_key_norm;
91
92 ASSERT(zn->zn_key_intlen == 8);
93 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
94 int j;
95 uint64_t word = *wp;
96
97 for (j = 0; j < zn->zn_key_intlen; j++) {
98 h = (h >> 8) ^
99 zfs_crc64_table[(h ^ word) & 0xFF];
100 word >>= NBBY;
101 }
102 }
103 } else {
104 int i, len;
105 const uint8_t *cp = zn->zn_key_norm;
106
107 /*
108 * We previously stored the terminating null on
109 * disk, but didn't hash it, so we need to
110 * continue to not hash it. (The
111 * zn_key_*_numints includes the terminating
112 * null for non-binary keys.)
113 */
114 len = zn->zn_key_norm_numints - 1;
115
116 ASSERT(zn->zn_key_intlen == 1);
117 for (i = 0; i < len; cp++, i++) {
118 h = (h >> 8) ^
119 zfs_crc64_table[(h ^ *cp) & 0xFF];
120 }
121 }
122 }
123 /*
124 * Don't use all 64 bits, since we need some in the cookie for
125 * the collision differentiator. We MUST use the high bits,
126 * since those are the ones that we first pay attention to when
127 * chosing the bucket.
128 */
129 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
130
131 return (h);
132}
133
134static int
135zap_normalize(zap_t *zap, const char *name, char *namenorm)
136{
137 size_t inlen, outlen;
138 int err;
139
140 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
141
142 inlen = strlen(name) + 1;
143 outlen = ZAP_MAXNAMELEN;
144
145 err = 0;
146 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
147 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
148 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
149
150 return (err);
151}
152
153boolean_t
154zap_match(zap_name_t *zn, const char *matchname)
155{
156 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
157
158 if (zn->zn_matchtype == MT_FIRST) {
159 char norm[ZAP_MAXNAMELEN];
160
161 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
162 return (B_FALSE);
163
164 return (strcmp(zn->zn_key_norm, norm) == 0);
165 } else {
166 /* MT_BEST or MT_EXACT */
167 return (strcmp(zn->zn_key_orig, matchname) == 0);
168 }
169}
170
171void
172zap_name_free(zap_name_t *zn)
173{
174 kmem_free(zn, sizeof (zap_name_t));
175}
176
177zap_name_t *
178zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
179{
180 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
181
182 zn->zn_zap = zap;
183 zn->zn_key_intlen = sizeof (*key);
184 zn->zn_key_orig = key;
185 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
186 zn->zn_matchtype = mt;
187 if (zap->zap_normflags) {
188 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
189 zap_name_free(zn);
190 return (NULL);
191 }
192 zn->zn_key_norm = zn->zn_normbuf;
193 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
194 } else {
195 if (mt != MT_EXACT) {
196 zap_name_free(zn);
197 return (NULL);
198 }
199 zn->zn_key_norm = zn->zn_key_orig;
200 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
201 }
202
203 zn->zn_hash = zap_hash(zn);
204 return (zn);
205}
206
207zap_name_t *
208zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
209{
210 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
211
212 ASSERT(zap->zap_normflags == 0);
213 zn->zn_zap = zap;
214 zn->zn_key_intlen = sizeof (*key);
215 zn->zn_key_orig = zn->zn_key_norm = key;
216 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
217 zn->zn_matchtype = MT_EXACT;
218
219 zn->zn_hash = zap_hash(zn);
220 return (zn);
221}
222
223static void
224mzap_byteswap(mzap_phys_t *buf, size_t size)
225{
226 int i, max;
227 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
228 buf->mz_salt = BSWAP_64(buf->mz_salt);
229 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
230 max = (size / MZAP_ENT_LEN) - 1;
231 for (i = 0; i < max; i++) {
232 buf->mz_chunk[i].mze_value =
233 BSWAP_64(buf->mz_chunk[i].mze_value);
234 buf->mz_chunk[i].mze_cd =
235 BSWAP_32(buf->mz_chunk[i].mze_cd);
236 }
237}
238
239void
240zap_byteswap(void *buf, size_t size)
241{
242 uint64_t block_type;
243
244 block_type = *(uint64_t *)buf;
245
246 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
247 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
248 mzap_byteswap(buf, size);
249 } else {
250 fzap_byteswap(buf, size);
251 }
252}
253
254static int
255mze_compare(const void *arg1, const void *arg2)
256{
257 const mzap_ent_t *mze1 = arg1;
258 const mzap_ent_t *mze2 = arg2;
259
260 if (mze1->mze_hash > mze2->mze_hash)
261 return (+1);
262 if (mze1->mze_hash < mze2->mze_hash)
263 return (-1);
264 if (mze1->mze_cd > mze2->mze_cd)
265 return (+1);
266 if (mze1->mze_cd < mze2->mze_cd)
267 return (-1);
268 return (0);
269}
270
271static int
272mze_insert(zap_t *zap, int chunkid, uint64_t hash)
273{
274 mzap_ent_t *mze;
275 avl_index_t idx;
276
277 ASSERT(zap->zap_ismicro);
278 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
279
280 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
281 mze->mze_chunkid = chunkid;
282 mze->mze_hash = hash;
283 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
284 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
285 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
286 kmem_free(mze, sizeof (mzap_ent_t));
287 return (EEXIST);
288 }
289 avl_insert(&zap->zap_m.zap_avl, mze, idx);
290 return (0);
291}
292
293static mzap_ent_t *
294mze_find(zap_name_t *zn)
295{
296 mzap_ent_t mze_tofind;
297 mzap_ent_t *mze;
298 avl_index_t idx;
299 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
300
301 ASSERT(zn->zn_zap->zap_ismicro);
302 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
303
304 mze_tofind.mze_hash = zn->zn_hash;
305 mze_tofind.mze_cd = 0;
306
307again:
308 mze = avl_find(avl, &mze_tofind, &idx);
309 if (mze == NULL)
310 mze = avl_nearest(avl, idx, AVL_AFTER);
311 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
312 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
313 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
314 return (mze);
315 }
316 if (zn->zn_matchtype == MT_BEST) {
317 zn->zn_matchtype = MT_FIRST;
318 goto again;
319 }
320 return (NULL);
321}
322
323static uint32_t
324mze_find_unused_cd(zap_t *zap, uint64_t hash)
325{
326 mzap_ent_t mze_tofind;
327 mzap_ent_t *mze;
328 avl_index_t idx;
329 avl_tree_t *avl = &zap->zap_m.zap_avl;
330 uint32_t cd;
331
332 ASSERT(zap->zap_ismicro);
333 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
334
335 mze_tofind.mze_hash = hash;
336 mze_tofind.mze_cd = 0;
337
338 cd = 0;
339 for (mze = avl_find(avl, &mze_tofind, &idx);
340 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
341 if (mze->mze_cd != cd)
342 break;
343 cd++;
344 }
345
346 return (cd);
347}
348
349static void
350mze_remove(zap_t *zap, mzap_ent_t *mze)
351{
352 ASSERT(zap->zap_ismicro);
353 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
354
355 avl_remove(&zap->zap_m.zap_avl, mze);
356 kmem_free(mze, sizeof (mzap_ent_t));
357}
358
359static void
360mze_destroy(zap_t *zap)
361{
362 mzap_ent_t *mze;
363 void *avlcookie = NULL;
364
365 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
366 kmem_free(mze, sizeof (mzap_ent_t));
367 avl_destroy(&zap->zap_m.zap_avl);
368}
369
370static zap_t *
371mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
372{
373 zap_t *winner;
374 zap_t *zap;
375 int i;
1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 */
27
28#include <sys/zio.h>
29#include <sys/spa.h>
30#include <sys/dmu.h>
31#include <sys/zfs_context.h>
32#include <sys/zap.h>
33#include <sys/refcount.h>
34#include <sys/zap_impl.h>
35#include <sys/zap_leaf.h>
36#include <sys/avl.h>
37#include <sys/arc.h>
38#include <sys/dmu_objset.h>
39
40#ifdef _KERNEL
41#include <sys/sunddi.h>
42#endif
43
44extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
45
46static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
47
48uint64_t
49zap_getflags(zap_t *zap)
50{
51 if (zap->zap_ismicro)
52 return (0);
53 return (zap_f_phys(zap)->zap_flags);
54}
55
56int
57zap_hashbits(zap_t *zap)
58{
59 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
60 return (48);
61 else
62 return (28);
63}
64
65uint32_t
66zap_maxcd(zap_t *zap)
67{
68 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
69 return ((1<<16)-1);
70 else
71 return (-1U);
72}
73
74static uint64_t
75zap_hash(zap_name_t *zn)
76{
77 zap_t *zap = zn->zn_zap;
78 uint64_t h = 0;
79
80 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
81 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
82 h = *(uint64_t *)zn->zn_key_orig;
83 } else {
84 h = zap->zap_salt;
85 ASSERT(h != 0);
86 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
87
88 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
89 int i;
90 const uint64_t *wp = zn->zn_key_norm;
91
92 ASSERT(zn->zn_key_intlen == 8);
93 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
94 int j;
95 uint64_t word = *wp;
96
97 for (j = 0; j < zn->zn_key_intlen; j++) {
98 h = (h >> 8) ^
99 zfs_crc64_table[(h ^ word) & 0xFF];
100 word >>= NBBY;
101 }
102 }
103 } else {
104 int i, len;
105 const uint8_t *cp = zn->zn_key_norm;
106
107 /*
108 * We previously stored the terminating null on
109 * disk, but didn't hash it, so we need to
110 * continue to not hash it. (The
111 * zn_key_*_numints includes the terminating
112 * null for non-binary keys.)
113 */
114 len = zn->zn_key_norm_numints - 1;
115
116 ASSERT(zn->zn_key_intlen == 1);
117 for (i = 0; i < len; cp++, i++) {
118 h = (h >> 8) ^
119 zfs_crc64_table[(h ^ *cp) & 0xFF];
120 }
121 }
122 }
123 /*
124 * Don't use all 64 bits, since we need some in the cookie for
125 * the collision differentiator. We MUST use the high bits,
126 * since those are the ones that we first pay attention to when
127 * chosing the bucket.
128 */
129 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
130
131 return (h);
132}
133
134static int
135zap_normalize(zap_t *zap, const char *name, char *namenorm)
136{
137 size_t inlen, outlen;
138 int err;
139
140 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
141
142 inlen = strlen(name) + 1;
143 outlen = ZAP_MAXNAMELEN;
144
145 err = 0;
146 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
147 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
148 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
149
150 return (err);
151}
152
153boolean_t
154zap_match(zap_name_t *zn, const char *matchname)
155{
156 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
157
158 if (zn->zn_matchtype == MT_FIRST) {
159 char norm[ZAP_MAXNAMELEN];
160
161 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
162 return (B_FALSE);
163
164 return (strcmp(zn->zn_key_norm, norm) == 0);
165 } else {
166 /* MT_BEST or MT_EXACT */
167 return (strcmp(zn->zn_key_orig, matchname) == 0);
168 }
169}
170
171void
172zap_name_free(zap_name_t *zn)
173{
174 kmem_free(zn, sizeof (zap_name_t));
175}
176
177zap_name_t *
178zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
179{
180 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
181
182 zn->zn_zap = zap;
183 zn->zn_key_intlen = sizeof (*key);
184 zn->zn_key_orig = key;
185 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
186 zn->zn_matchtype = mt;
187 if (zap->zap_normflags) {
188 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
189 zap_name_free(zn);
190 return (NULL);
191 }
192 zn->zn_key_norm = zn->zn_normbuf;
193 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
194 } else {
195 if (mt != MT_EXACT) {
196 zap_name_free(zn);
197 return (NULL);
198 }
199 zn->zn_key_norm = zn->zn_key_orig;
200 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
201 }
202
203 zn->zn_hash = zap_hash(zn);
204 return (zn);
205}
206
207zap_name_t *
208zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
209{
210 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
211
212 ASSERT(zap->zap_normflags == 0);
213 zn->zn_zap = zap;
214 zn->zn_key_intlen = sizeof (*key);
215 zn->zn_key_orig = zn->zn_key_norm = key;
216 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
217 zn->zn_matchtype = MT_EXACT;
218
219 zn->zn_hash = zap_hash(zn);
220 return (zn);
221}
222
223static void
224mzap_byteswap(mzap_phys_t *buf, size_t size)
225{
226 int i, max;
227 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
228 buf->mz_salt = BSWAP_64(buf->mz_salt);
229 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
230 max = (size / MZAP_ENT_LEN) - 1;
231 for (i = 0; i < max; i++) {
232 buf->mz_chunk[i].mze_value =
233 BSWAP_64(buf->mz_chunk[i].mze_value);
234 buf->mz_chunk[i].mze_cd =
235 BSWAP_32(buf->mz_chunk[i].mze_cd);
236 }
237}
238
239void
240zap_byteswap(void *buf, size_t size)
241{
242 uint64_t block_type;
243
244 block_type = *(uint64_t *)buf;
245
246 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
247 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
248 mzap_byteswap(buf, size);
249 } else {
250 fzap_byteswap(buf, size);
251 }
252}
253
254static int
255mze_compare(const void *arg1, const void *arg2)
256{
257 const mzap_ent_t *mze1 = arg1;
258 const mzap_ent_t *mze2 = arg2;
259
260 if (mze1->mze_hash > mze2->mze_hash)
261 return (+1);
262 if (mze1->mze_hash < mze2->mze_hash)
263 return (-1);
264 if (mze1->mze_cd > mze2->mze_cd)
265 return (+1);
266 if (mze1->mze_cd < mze2->mze_cd)
267 return (-1);
268 return (0);
269}
270
271static int
272mze_insert(zap_t *zap, int chunkid, uint64_t hash)
273{
274 mzap_ent_t *mze;
275 avl_index_t idx;
276
277 ASSERT(zap->zap_ismicro);
278 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
279
280 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
281 mze->mze_chunkid = chunkid;
282 mze->mze_hash = hash;
283 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
284 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
285 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
286 kmem_free(mze, sizeof (mzap_ent_t));
287 return (EEXIST);
288 }
289 avl_insert(&zap->zap_m.zap_avl, mze, idx);
290 return (0);
291}
292
293static mzap_ent_t *
294mze_find(zap_name_t *zn)
295{
296 mzap_ent_t mze_tofind;
297 mzap_ent_t *mze;
298 avl_index_t idx;
299 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
300
301 ASSERT(zn->zn_zap->zap_ismicro);
302 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
303
304 mze_tofind.mze_hash = zn->zn_hash;
305 mze_tofind.mze_cd = 0;
306
307again:
308 mze = avl_find(avl, &mze_tofind, &idx);
309 if (mze == NULL)
310 mze = avl_nearest(avl, idx, AVL_AFTER);
311 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
312 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
313 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
314 return (mze);
315 }
316 if (zn->zn_matchtype == MT_BEST) {
317 zn->zn_matchtype = MT_FIRST;
318 goto again;
319 }
320 return (NULL);
321}
322
323static uint32_t
324mze_find_unused_cd(zap_t *zap, uint64_t hash)
325{
326 mzap_ent_t mze_tofind;
327 mzap_ent_t *mze;
328 avl_index_t idx;
329 avl_tree_t *avl = &zap->zap_m.zap_avl;
330 uint32_t cd;
331
332 ASSERT(zap->zap_ismicro);
333 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
334
335 mze_tofind.mze_hash = hash;
336 mze_tofind.mze_cd = 0;
337
338 cd = 0;
339 for (mze = avl_find(avl, &mze_tofind, &idx);
340 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
341 if (mze->mze_cd != cd)
342 break;
343 cd++;
344 }
345
346 return (cd);
347}
348
349static void
350mze_remove(zap_t *zap, mzap_ent_t *mze)
351{
352 ASSERT(zap->zap_ismicro);
353 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
354
355 avl_remove(&zap->zap_m.zap_avl, mze);
356 kmem_free(mze, sizeof (mzap_ent_t));
357}
358
359static void
360mze_destroy(zap_t *zap)
361{
362 mzap_ent_t *mze;
363 void *avlcookie = NULL;
364
365 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
366 kmem_free(mze, sizeof (mzap_ent_t));
367 avl_destroy(&zap->zap_m.zap_avl);
368}
369
370static zap_t *
371mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
372{
373 zap_t *winner;
374 zap_t *zap;
375 int i;
376 uint64_t *zap_hdr = (uint64_t *)db->db_data;
377 uint64_t zap_block_type = zap_hdr[0];
378 uint64_t zap_magic = zap_hdr[1];
376
377 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
378
379 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
380 rw_init(&zap->zap_rwlock, 0, 0, 0);
381 rw_enter(&zap->zap_rwlock, RW_WRITER);
382 zap->zap_objset = os;
383 zap->zap_object = obj;
384 zap->zap_dbuf = db;
385
379
380 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
381
382 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
383 rw_init(&zap->zap_rwlock, 0, 0, 0);
384 rw_enter(&zap->zap_rwlock, RW_WRITER);
385 zap->zap_objset = os;
386 zap->zap_object = obj;
387 zap->zap_dbuf = db;
388
386 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
389 if (zap_block_type != ZBT_MICRO) {
387 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
388 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
390 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
391 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
392 if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
393 winner = NULL; /* No actual winner here... */
394 goto handle_winner;
395 }
389 } else {
390 zap->zap_ismicro = TRUE;
391 }
392
393 /*
394 * Make sure that zap_ismicro is set before we let others see
395 * it, because zap_lockdir() checks zap_ismicro without the lock
396 * held.
397 */
398 dmu_buf_init_user(&zap->zap_dbu, zap_evict, &zap->zap_dbuf);
399 winner = dmu_buf_set_user(db, &zap->zap_dbu);
400
396 } else {
397 zap->zap_ismicro = TRUE;
398 }
399
400 /*
401 * Make sure that zap_ismicro is set before we let others see
402 * it, because zap_lockdir() checks zap_ismicro without the lock
403 * held.
404 */
405 dmu_buf_init_user(&zap->zap_dbu, zap_evict, &zap->zap_dbuf);
406 winner = dmu_buf_set_user(db, &zap->zap_dbu);
407
401 if (winner != NULL) {
402 rw_exit(&zap->zap_rwlock);
403 rw_destroy(&zap->zap_rwlock);
404 if (!zap->zap_ismicro)
405 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
406 kmem_free(zap, sizeof (zap_t));
407 return (winner);
408 }
408 if (winner != NULL)
409 goto handle_winner;
409
410 if (zap->zap_ismicro) {
411 zap->zap_salt = zap_m_phys(zap)->mz_salt;
412 zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
413 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
414 avl_create(&zap->zap_m.zap_avl, mze_compare,
415 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
416
417 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
418 mzap_ent_phys_t *mze =
419 &zap_m_phys(zap)->mz_chunk[i];
420 if (mze->mze_name[0]) {
421 zap_name_t *zn;
422
423 zn = zap_name_alloc(zap, mze->mze_name,
424 MT_EXACT);
425 if (mze_insert(zap, i, zn->zn_hash) == 0)
426 zap->zap_m.zap_num_entries++;
427 else {
428 printf("ZFS WARNING: Duplicated ZAP "
429 "entry detected (%s).\n",
430 mze->mze_name);
431 }
432 zap_name_free(zn);
433 }
434 }
435 } else {
436 zap->zap_salt = zap_f_phys(zap)->zap_salt;
437 zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
438
439 ASSERT3U(sizeof (struct zap_leaf_header), ==,
440 2*ZAP_LEAF_CHUNKSIZE);
441
442 /*
443 * The embedded pointer table should not overlap the
444 * other members.
445 */
446 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
447 &zap_f_phys(zap)->zap_salt);
448
449 /*
450 * The embedded pointer table should end at the end of
451 * the block
452 */
453 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
454 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
455 (uintptr_t)zap_f_phys(zap), ==,
456 zap->zap_dbuf->db_size);
457 }
458 rw_exit(&zap->zap_rwlock);
459 return (zap);
410
411 if (zap->zap_ismicro) {
412 zap->zap_salt = zap_m_phys(zap)->mz_salt;
413 zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
414 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
415 avl_create(&zap->zap_m.zap_avl, mze_compare,
416 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
417
418 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
419 mzap_ent_phys_t *mze =
420 &zap_m_phys(zap)->mz_chunk[i];
421 if (mze->mze_name[0]) {
422 zap_name_t *zn;
423
424 zn = zap_name_alloc(zap, mze->mze_name,
425 MT_EXACT);
426 if (mze_insert(zap, i, zn->zn_hash) == 0)
427 zap->zap_m.zap_num_entries++;
428 else {
429 printf("ZFS WARNING: Duplicated ZAP "
430 "entry detected (%s).\n",
431 mze->mze_name);
432 }
433 zap_name_free(zn);
434 }
435 }
436 } else {
437 zap->zap_salt = zap_f_phys(zap)->zap_salt;
438 zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
439
440 ASSERT3U(sizeof (struct zap_leaf_header), ==,
441 2*ZAP_LEAF_CHUNKSIZE);
442
443 /*
444 * The embedded pointer table should not overlap the
445 * other members.
446 */
447 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
448 &zap_f_phys(zap)->zap_salt);
449
450 /*
451 * The embedded pointer table should end at the end of
452 * the block
453 */
454 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
455 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
456 (uintptr_t)zap_f_phys(zap), ==,
457 zap->zap_dbuf->db_size);
458 }
459 rw_exit(&zap->zap_rwlock);
460 return (zap);
461
462handle_winner:
463 rw_exit(&zap->zap_rwlock);
464 rw_destroy(&zap->zap_rwlock);
465 if (!zap->zap_ismicro)
466 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
467 kmem_free(zap, sizeof (zap_t));
468 return (winner);
460}
461
462int
463zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
464 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
465{
466 zap_t *zap;
467 dmu_buf_t *db;
468 krw_t lt;
469 int err;
470
471 *zapp = NULL;
472
473 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
474 if (err)
475 return (err);
476
477#ifdef ZFS_DEBUG
478 {
479 dmu_object_info_t doi;
480 dmu_object_info_from_db(db, &doi);
481 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
482 }
483#endif
484
485 zap = dmu_buf_get_user(db);
469}
470
471int
472zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
473 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
474{
475 zap_t *zap;
476 dmu_buf_t *db;
477 krw_t lt;
478 int err;
479
480 *zapp = NULL;
481
482 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
483 if (err)
484 return (err);
485
486#ifdef ZFS_DEBUG
487 {
488 dmu_object_info_t doi;
489 dmu_object_info_from_db(db, &doi);
490 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
491 }
492#endif
493
494 zap = dmu_buf_get_user(db);
486 if (zap == NULL)
495 if (zap == NULL) {
487 zap = mzap_open(os, obj, db);
496 zap = mzap_open(os, obj, db);
497 if (zap == NULL) {
498 /*
499 * mzap_open() didn't like what it saw on-disk.
500 * Check for corruption!
501 */
502 dmu_buf_rele(db, NULL);
503 return (SET_ERROR(EIO));
504 }
505 }
488
489 /*
490 * We're checking zap_ismicro without the lock held, in order to
491 * tell what type of lock we want. Once we have some sort of
492 * lock, see if it really is the right type. In practice this
493 * can only be different if it was upgraded from micro to fat,
494 * and micro wanted WRITER but fat only needs READER.
495 */
496 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
497 rw_enter(&zap->zap_rwlock, lt);
498 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
499 /* it was upgraded, now we only need reader */
500 ASSERT(lt == RW_WRITER);
501 ASSERT(RW_READER ==
502 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
503 rw_downgrade(&zap->zap_rwlock);
504 lt = RW_READER;
505 }
506
507 zap->zap_objset = os;
508
509 if (lt == RW_WRITER)
510 dmu_buf_will_dirty(db, tx);
511
512 ASSERT3P(zap->zap_dbuf, ==, db);
513
514 ASSERT(!zap->zap_ismicro ||
515 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
516 if (zap->zap_ismicro && tx && adding &&
517 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
518 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
519 if (newsz > MZAP_MAX_BLKSZ) {
520 dprintf("upgrading obj %llu: num_entries=%u\n",
521 obj, zap->zap_m.zap_num_entries);
522 *zapp = zap;
523 return (mzap_upgrade(zapp, tx, 0));
524 }
525 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
526 ASSERT0(err);
527 zap->zap_m.zap_num_chunks =
528 db->db_size / MZAP_ENT_LEN - 1;
529 }
530
531 *zapp = zap;
532 return (0);
533}
534
535void
536zap_unlockdir(zap_t *zap)
537{
538 rw_exit(&zap->zap_rwlock);
539 dmu_buf_rele(zap->zap_dbuf, NULL);
540}
541
542static int
543mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
544{
545 mzap_phys_t *mzp;
546 int i, sz, nchunks;
547 int err = 0;
548 zap_t *zap = *zapp;
549
550 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
551
552 sz = zap->zap_dbuf->db_size;
553 mzp = zio_buf_alloc(sz);
554 bcopy(zap->zap_dbuf->db_data, mzp, sz);
555 nchunks = zap->zap_m.zap_num_chunks;
556
557 if (!flags) {
558 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
559 1ULL << fzap_default_block_shift, 0, tx);
560 if (err) {
561 zio_buf_free(mzp, sz);
562 return (err);
563 }
564 }
565
566 dprintf("upgrading obj=%llu with %u chunks\n",
567 zap->zap_object, nchunks);
568 /* XXX destroy the avl later, so we can use the stored hash value */
569 mze_destroy(zap);
570
571 fzap_upgrade(zap, tx, flags);
572
573 for (i = 0; i < nchunks; i++) {
574 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
575 zap_name_t *zn;
576 if (mze->mze_name[0] == 0)
577 continue;
578 dprintf("adding %s=%llu\n",
579 mze->mze_name, mze->mze_value);
580 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
581 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
582 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
583 zap_name_free(zn);
584 if (err)
585 break;
586 }
587 zio_buf_free(mzp, sz);
588 *zapp = zap;
589 return (err);
590}
591
592void
593mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
594 dmu_tx_t *tx)
595{
596 dmu_buf_t *db;
597 mzap_phys_t *zp;
598
599 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
600
601#ifdef ZFS_DEBUG
602 {
603 dmu_object_info_t doi;
604 dmu_object_info_from_db(db, &doi);
605 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
606 }
607#endif
608
609 dmu_buf_will_dirty(db, tx);
610 zp = db->db_data;
611 zp->mz_block_type = ZBT_MICRO;
612 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
613 zp->mz_normflags = normflags;
614 dmu_buf_rele(db, FTAG);
615
616 if (flags != 0) {
617 zap_t *zap;
618 /* Only fat zap supports flags; upgrade immediately. */
619 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
620 B_FALSE, B_FALSE, &zap));
621 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
622 zap_unlockdir(zap);
623 }
624}
625
626int
627zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
628 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
629{
630 return (zap_create_claim_norm(os, obj,
631 0, ot, bonustype, bonuslen, tx));
632}
633
634int
635zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
636 dmu_object_type_t ot,
637 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
638{
639 int err;
640
641 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
642 if (err != 0)
643 return (err);
644 mzap_create_impl(os, obj, normflags, 0, tx);
645 return (0);
646}
647
648uint64_t
649zap_create(objset_t *os, dmu_object_type_t ot,
650 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
651{
652 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
653}
654
655uint64_t
656zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
657 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
658{
659 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
660
661 mzap_create_impl(os, obj, normflags, 0, tx);
662 return (obj);
663}
664
665uint64_t
666zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
667 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
668 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
669{
670 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
671
672 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
673 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
674 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
675 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
676
677 VERIFY(dmu_object_set_blocksize(os, obj,
678 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
679
680 mzap_create_impl(os, obj, normflags, flags, tx);
681 return (obj);
682}
683
684int
685zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
686{
687 /*
688 * dmu_object_free will free the object number and free the
689 * data. Freeing the data will cause our pageout function to be
690 * called, which will destroy our data (zap_leaf_t's and zap_t).
691 */
692
693 return (dmu_object_free(os, zapobj, tx));
694}
695
696void
697zap_evict(void *dbu)
698{
699 zap_t *zap = dbu;
700
701 rw_destroy(&zap->zap_rwlock);
702
703 if (zap->zap_ismicro)
704 mze_destroy(zap);
705 else
706 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
707
708 kmem_free(zap, sizeof (zap_t));
709}
710
711int
712zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
713{
714 zap_t *zap;
715 int err;
716
717 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
718 if (err)
719 return (err);
720 if (!zap->zap_ismicro) {
721 err = fzap_count(zap, count);
722 } else {
723 *count = zap->zap_m.zap_num_entries;
724 }
725 zap_unlockdir(zap);
726 return (err);
727}
728
729/*
730 * zn may be NULL; if not specified, it will be computed if needed.
731 * See also the comment above zap_entry_normalization_conflict().
732 */
733static boolean_t
734mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
735{
736 mzap_ent_t *other;
737 int direction = AVL_BEFORE;
738 boolean_t allocdzn = B_FALSE;
739
740 if (zap->zap_normflags == 0)
741 return (B_FALSE);
742
743again:
744 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
745 other && other->mze_hash == mze->mze_hash;
746 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
747
748 if (zn == NULL) {
749 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
750 MT_FIRST);
751 allocdzn = B_TRUE;
752 }
753 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
754 if (allocdzn)
755 zap_name_free(zn);
756 return (B_TRUE);
757 }
758 }
759
760 if (direction == AVL_BEFORE) {
761 direction = AVL_AFTER;
762 goto again;
763 }
764
765 if (allocdzn)
766 zap_name_free(zn);
767 return (B_FALSE);
768}
769
770/*
771 * Routines for manipulating attributes.
772 */
773
774int
775zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
776 uint64_t integer_size, uint64_t num_integers, void *buf)
777{
778 return (zap_lookup_norm(os, zapobj, name, integer_size,
779 num_integers, buf, MT_EXACT, NULL, 0, NULL));
780}
781
782int
783zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
784 uint64_t integer_size, uint64_t num_integers, void *buf,
785 matchtype_t mt, char *realname, int rn_len,
786 boolean_t *ncp)
787{
788 zap_t *zap;
789 int err;
790 mzap_ent_t *mze;
791 zap_name_t *zn;
792
793 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
794 if (err)
795 return (err);
796 zn = zap_name_alloc(zap, name, mt);
797 if (zn == NULL) {
798 zap_unlockdir(zap);
799 return (SET_ERROR(ENOTSUP));
800 }
801
802 if (!zap->zap_ismicro) {
803 err = fzap_lookup(zn, integer_size, num_integers, buf,
804 realname, rn_len, ncp);
805 } else {
806 mze = mze_find(zn);
807 if (mze == NULL) {
808 err = SET_ERROR(ENOENT);
809 } else {
810 if (num_integers < 1) {
811 err = SET_ERROR(EOVERFLOW);
812 } else if (integer_size != 8) {
813 err = SET_ERROR(EINVAL);
814 } else {
815 *(uint64_t *)buf =
816 MZE_PHYS(zap, mze)->mze_value;
817 (void) strlcpy(realname,
818 MZE_PHYS(zap, mze)->mze_name, rn_len);
819 if (ncp) {
820 *ncp = mzap_normalization_conflict(zap,
821 zn, mze);
822 }
823 }
824 }
825 }
826 zap_name_free(zn);
827 zap_unlockdir(zap);
828 return (err);
829}
830
831int
832zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
833 int key_numints)
834{
835 zap_t *zap;
836 int err;
837 zap_name_t *zn;
838
839 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
840 if (err)
841 return (err);
842 zn = zap_name_alloc_uint64(zap, key, key_numints);
843 if (zn == NULL) {
844 zap_unlockdir(zap);
845 return (SET_ERROR(ENOTSUP));
846 }
847
848 fzap_prefetch(zn);
849 zap_name_free(zn);
850 zap_unlockdir(zap);
851 return (err);
852}
853
854int
855zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
856 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
857{
858 zap_t *zap;
859 int err;
860 zap_name_t *zn;
861
862 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
863 if (err)
864 return (err);
865 zn = zap_name_alloc_uint64(zap, key, key_numints);
866 if (zn == NULL) {
867 zap_unlockdir(zap);
868 return (SET_ERROR(ENOTSUP));
869 }
870
871 err = fzap_lookup(zn, integer_size, num_integers, buf,
872 NULL, 0, NULL);
873 zap_name_free(zn);
874 zap_unlockdir(zap);
875 return (err);
876}
877
878int
879zap_contains(objset_t *os, uint64_t zapobj, const char *name)
880{
881 int err = zap_lookup_norm(os, zapobj, name, 0,
882 0, NULL, MT_EXACT, NULL, 0, NULL);
883 if (err == EOVERFLOW || err == EINVAL)
884 err = 0; /* found, but skipped reading the value */
885 return (err);
886}
887
888int
889zap_length(objset_t *os, uint64_t zapobj, const char *name,
890 uint64_t *integer_size, uint64_t *num_integers)
891{
892 zap_t *zap;
893 int err;
894 mzap_ent_t *mze;
895 zap_name_t *zn;
896
897 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
898 if (err)
899 return (err);
900 zn = zap_name_alloc(zap, name, MT_EXACT);
901 if (zn == NULL) {
902 zap_unlockdir(zap);
903 return (SET_ERROR(ENOTSUP));
904 }
905 if (!zap->zap_ismicro) {
906 err = fzap_length(zn, integer_size, num_integers);
907 } else {
908 mze = mze_find(zn);
909 if (mze == NULL) {
910 err = SET_ERROR(ENOENT);
911 } else {
912 if (integer_size)
913 *integer_size = 8;
914 if (num_integers)
915 *num_integers = 1;
916 }
917 }
918 zap_name_free(zn);
919 zap_unlockdir(zap);
920 return (err);
921}
922
923int
924zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
925 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
926{
927 zap_t *zap;
928 int err;
929 zap_name_t *zn;
930
931 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
932 if (err)
933 return (err);
934 zn = zap_name_alloc_uint64(zap, key, key_numints);
935 if (zn == NULL) {
936 zap_unlockdir(zap);
937 return (SET_ERROR(ENOTSUP));
938 }
939 err = fzap_length(zn, integer_size, num_integers);
940 zap_name_free(zn);
941 zap_unlockdir(zap);
942 return (err);
943}
944
945static void
946mzap_addent(zap_name_t *zn, uint64_t value)
947{
948 int i;
949 zap_t *zap = zn->zn_zap;
950 int start = zap->zap_m.zap_alloc_next;
951 uint32_t cd;
952
953 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
954
955#ifdef ZFS_DEBUG
956 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
957 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
958 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
959 }
960#endif
961
962 cd = mze_find_unused_cd(zap, zn->zn_hash);
963 /* given the limited size of the microzap, this can't happen */
964 ASSERT(cd < zap_maxcd(zap));
965
966again:
967 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
968 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
969 if (mze->mze_name[0] == 0) {
970 mze->mze_value = value;
971 mze->mze_cd = cd;
972 (void) strcpy(mze->mze_name, zn->zn_key_orig);
973 zap->zap_m.zap_num_entries++;
974 zap->zap_m.zap_alloc_next = i+1;
975 if (zap->zap_m.zap_alloc_next ==
976 zap->zap_m.zap_num_chunks)
977 zap->zap_m.zap_alloc_next = 0;
978 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
979 return;
980 }
981 }
982 if (start != 0) {
983 start = 0;
984 goto again;
985 }
986 ASSERT(!"out of entries!");
987}
988
989int
990zap_add(objset_t *os, uint64_t zapobj, const char *key,
991 int integer_size, uint64_t num_integers,
992 const void *val, dmu_tx_t *tx)
993{
994 zap_t *zap;
995 int err;
996 mzap_ent_t *mze;
997 const uint64_t *intval = val;
998 zap_name_t *zn;
999
1000 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1001 if (err)
1002 return (err);
1003 zn = zap_name_alloc(zap, key, MT_EXACT);
1004 if (zn == NULL) {
1005 zap_unlockdir(zap);
1006 return (SET_ERROR(ENOTSUP));
1007 }
1008 if (!zap->zap_ismicro) {
1009 err = fzap_add(zn, integer_size, num_integers, val, tx);
1010 zap = zn->zn_zap; /* fzap_add() may change zap */
1011 } else if (integer_size != 8 || num_integers != 1 ||
1012 strlen(key) >= MZAP_NAME_LEN) {
1013 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1014 if (err == 0)
1015 err = fzap_add(zn, integer_size, num_integers, val, tx);
1016 zap = zn->zn_zap; /* fzap_add() may change zap */
1017 } else {
1018 mze = mze_find(zn);
1019 if (mze != NULL) {
1020 err = SET_ERROR(EEXIST);
1021 } else {
1022 mzap_addent(zn, *intval);
1023 }
1024 }
1025 ASSERT(zap == zn->zn_zap);
1026 zap_name_free(zn);
1027 if (zap != NULL) /* may be NULL if fzap_add() failed */
1028 zap_unlockdir(zap);
1029 return (err);
1030}
1031
1032int
1033zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1034 int key_numints, int integer_size, uint64_t num_integers,
1035 const void *val, dmu_tx_t *tx)
1036{
1037 zap_t *zap;
1038 int err;
1039 zap_name_t *zn;
1040
1041 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1042 if (err)
1043 return (err);
1044 zn = zap_name_alloc_uint64(zap, key, key_numints);
1045 if (zn == NULL) {
1046 zap_unlockdir(zap);
1047 return (SET_ERROR(ENOTSUP));
1048 }
1049 err = fzap_add(zn, integer_size, num_integers, val, tx);
1050 zap = zn->zn_zap; /* fzap_add() may change zap */
1051 zap_name_free(zn);
1052 if (zap != NULL) /* may be NULL if fzap_add() failed */
1053 zap_unlockdir(zap);
1054 return (err);
1055}
1056
1057int
1058zap_update(objset_t *os, uint64_t zapobj, const char *name,
1059 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1060{
1061 zap_t *zap;
1062 mzap_ent_t *mze;
1063 uint64_t oldval;
1064 const uint64_t *intval = val;
1065 zap_name_t *zn;
1066 int err;
1067
1068#ifdef ZFS_DEBUG
1069 /*
1070 * If there is an old value, it shouldn't change across the
1071 * lockdir (eg, due to bprewrite's xlation).
1072 */
1073 if (integer_size == 8 && num_integers == 1)
1074 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1075#endif
1076
1077 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1078 if (err)
1079 return (err);
1080 zn = zap_name_alloc(zap, name, MT_EXACT);
1081 if (zn == NULL) {
1082 zap_unlockdir(zap);
1083 return (SET_ERROR(ENOTSUP));
1084 }
1085 if (!zap->zap_ismicro) {
1086 err = fzap_update(zn, integer_size, num_integers, val, tx);
1087 zap = zn->zn_zap; /* fzap_update() may change zap */
1088 } else if (integer_size != 8 || num_integers != 1 ||
1089 strlen(name) >= MZAP_NAME_LEN) {
1090 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1091 zapobj, integer_size, num_integers, name);
1092 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1093 if (err == 0)
1094 err = fzap_update(zn, integer_size, num_integers,
1095 val, tx);
1096 zap = zn->zn_zap; /* fzap_update() may change zap */
1097 } else {
1098 mze = mze_find(zn);
1099 if (mze != NULL) {
1100 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1101 MZE_PHYS(zap, mze)->mze_value = *intval;
1102 } else {
1103 mzap_addent(zn, *intval);
1104 }
1105 }
1106 ASSERT(zap == zn->zn_zap);
1107 zap_name_free(zn);
1108 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1109 zap_unlockdir(zap);
1110 return (err);
1111}
1112
1113int
1114zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1115 int key_numints,
1116 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1117{
1118 zap_t *zap;
1119 zap_name_t *zn;
1120 int err;
1121
1122 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1123 if (err)
1124 return (err);
1125 zn = zap_name_alloc_uint64(zap, key, key_numints);
1126 if (zn == NULL) {
1127 zap_unlockdir(zap);
1128 return (SET_ERROR(ENOTSUP));
1129 }
1130 err = fzap_update(zn, integer_size, num_integers, val, tx);
1131 zap = zn->zn_zap; /* fzap_update() may change zap */
1132 zap_name_free(zn);
1133 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1134 zap_unlockdir(zap);
1135 return (err);
1136}
1137
1138int
1139zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1140{
1141 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1142}
1143
1144int
1145zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1146 matchtype_t mt, dmu_tx_t *tx)
1147{
1148 zap_t *zap;
1149 int err;
1150 mzap_ent_t *mze;
1151 zap_name_t *zn;
1152
1153 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1154 if (err)
1155 return (err);
1156 zn = zap_name_alloc(zap, name, mt);
1157 if (zn == NULL) {
1158 zap_unlockdir(zap);
1159 return (SET_ERROR(ENOTSUP));
1160 }
1161 if (!zap->zap_ismicro) {
1162 err = fzap_remove(zn, tx);
1163 } else {
1164 mze = mze_find(zn);
1165 if (mze == NULL) {
1166 err = SET_ERROR(ENOENT);
1167 } else {
1168 zap->zap_m.zap_num_entries--;
1169 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1170 sizeof (mzap_ent_phys_t));
1171 mze_remove(zap, mze);
1172 }
1173 }
1174 zap_name_free(zn);
1175 zap_unlockdir(zap);
1176 return (err);
1177}
1178
1179int
1180zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1181 int key_numints, dmu_tx_t *tx)
1182{
1183 zap_t *zap;
1184 int err;
1185 zap_name_t *zn;
1186
1187 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1188 if (err)
1189 return (err);
1190 zn = zap_name_alloc_uint64(zap, key, key_numints);
1191 if (zn == NULL) {
1192 zap_unlockdir(zap);
1193 return (SET_ERROR(ENOTSUP));
1194 }
1195 err = fzap_remove(zn, tx);
1196 zap_name_free(zn);
1197 zap_unlockdir(zap);
1198 return (err);
1199}
1200
1201/*
1202 * Routines for iterating over the attributes.
1203 */
1204
1205void
1206zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1207 uint64_t serialized)
1208{
1209 zc->zc_objset = os;
1210 zc->zc_zap = NULL;
1211 zc->zc_leaf = NULL;
1212 zc->zc_zapobj = zapobj;
1213 zc->zc_serialized = serialized;
1214 zc->zc_hash = 0;
1215 zc->zc_cd = 0;
1216}
1217
1218void
1219zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1220{
1221 zap_cursor_init_serialized(zc, os, zapobj, 0);
1222}
1223
1224void
1225zap_cursor_fini(zap_cursor_t *zc)
1226{
1227 if (zc->zc_zap) {
1228 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1229 zap_unlockdir(zc->zc_zap);
1230 zc->zc_zap = NULL;
1231 }
1232 if (zc->zc_leaf) {
1233 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1234 zap_put_leaf(zc->zc_leaf);
1235 zc->zc_leaf = NULL;
1236 }
1237 zc->zc_objset = NULL;
1238}
1239
1240uint64_t
1241zap_cursor_serialize(zap_cursor_t *zc)
1242{
1243 if (zc->zc_hash == -1ULL)
1244 return (-1ULL);
1245 if (zc->zc_zap == NULL)
1246 return (zc->zc_serialized);
1247 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1248 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1249
1250 /*
1251 * We want to keep the high 32 bits of the cursor zero if we can, so
1252 * that 32-bit programs can access this. So usually use a small
1253 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1254 * of the cursor.
1255 *
1256 * [ collision differentiator | zap_hashbits()-bit hash value ]
1257 */
1258 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1259 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1260}
1261
1262int
1263zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1264{
1265 int err;
1266 avl_index_t idx;
1267 mzap_ent_t mze_tofind;
1268 mzap_ent_t *mze;
1269
1270 if (zc->zc_hash == -1ULL)
1271 return (SET_ERROR(ENOENT));
1272
1273 if (zc->zc_zap == NULL) {
1274 int hb;
1275 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1276 RW_READER, TRUE, FALSE, &zc->zc_zap);
1277 if (err)
1278 return (err);
1279
1280 /*
1281 * To support zap_cursor_init_serialized, advance, retrieve,
1282 * we must add to the existing zc_cd, which may already
1283 * be 1 due to the zap_cursor_advance.
1284 */
1285 ASSERT(zc->zc_hash == 0);
1286 hb = zap_hashbits(zc->zc_zap);
1287 zc->zc_hash = zc->zc_serialized << (64 - hb);
1288 zc->zc_cd += zc->zc_serialized >> hb;
1289 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1290 zc->zc_cd = 0;
1291 } else {
1292 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1293 }
1294 if (!zc->zc_zap->zap_ismicro) {
1295 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1296 } else {
1297 mze_tofind.mze_hash = zc->zc_hash;
1298 mze_tofind.mze_cd = zc->zc_cd;
1299
1300 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1301 if (mze == NULL) {
1302 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1303 idx, AVL_AFTER);
1304 }
1305 if (mze) {
1306 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1307 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1308 za->za_normalization_conflict =
1309 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1310 za->za_integer_length = 8;
1311 za->za_num_integers = 1;
1312 za->za_first_integer = mzep->mze_value;
1313 (void) strcpy(za->za_name, mzep->mze_name);
1314 zc->zc_hash = mze->mze_hash;
1315 zc->zc_cd = mze->mze_cd;
1316 err = 0;
1317 } else {
1318 zc->zc_hash = -1ULL;
1319 err = SET_ERROR(ENOENT);
1320 }
1321 }
1322 rw_exit(&zc->zc_zap->zap_rwlock);
1323 return (err);
1324}
1325
1326void
1327zap_cursor_advance(zap_cursor_t *zc)
1328{
1329 if (zc->zc_hash == -1ULL)
1330 return;
1331 zc->zc_cd++;
1332}
1333
1334int
1335zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1336{
1337 int err = 0;
1338 mzap_ent_t *mze;
1339 zap_name_t *zn;
1340
1341 if (zc->zc_zap == NULL) {
1342 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1343 RW_READER, TRUE, FALSE, &zc->zc_zap);
1344 if (err)
1345 return (err);
1346 } else {
1347 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1348 }
1349
1350 zn = zap_name_alloc(zc->zc_zap, name, mt);
1351 if (zn == NULL) {
1352 rw_exit(&zc->zc_zap->zap_rwlock);
1353 return (SET_ERROR(ENOTSUP));
1354 }
1355
1356 if (!zc->zc_zap->zap_ismicro) {
1357 err = fzap_cursor_move_to_key(zc, zn);
1358 } else {
1359 mze = mze_find(zn);
1360 if (mze == NULL) {
1361 err = SET_ERROR(ENOENT);
1362 goto out;
1363 }
1364 zc->zc_hash = mze->mze_hash;
1365 zc->zc_cd = mze->mze_cd;
1366 }
1367
1368out:
1369 zap_name_free(zn);
1370 rw_exit(&zc->zc_zap->zap_rwlock);
1371 return (err);
1372}
1373
1374int
1375zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1376{
1377 int err;
1378 zap_t *zap;
1379
1380 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1381 if (err)
1382 return (err);
1383
1384 bzero(zs, sizeof (zap_stats_t));
1385
1386 if (zap->zap_ismicro) {
1387 zs->zs_blocksize = zap->zap_dbuf->db_size;
1388 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1389 zs->zs_num_blocks = 1;
1390 } else {
1391 fzap_get_stats(zap, zs);
1392 }
1393 zap_unlockdir(zap);
1394 return (0);
1395}
1396
1397int
1398zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1399 uint64_t *towrite, uint64_t *tooverwrite)
1400{
1401 zap_t *zap;
1402 int err = 0;
1403
1404 /*
1405 * Since, we don't have a name, we cannot figure out which blocks will
1406 * be affected in this operation. So, account for the worst case :
1407 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1408 * - 4 new blocks written if adding:
1409 * - 2 blocks for possibly split leaves,
1410 * - 2 grown ptrtbl blocks
1411 *
1412 * This also accomodates the case where an add operation to a fairly
1413 * large microzap results in a promotion to fatzap.
1414 */
1415 if (name == NULL) {
1416 *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE;
1417 return (err);
1418 }
1419
1420 /*
1421 * We lock the zap with adding == FALSE. Because, if we pass
1422 * the actual value of add, it could trigger a mzap_upgrade().
1423 * At present we are just evaluating the possibility of this operation
1424 * and hence we donot want to trigger an upgrade.
1425 */
1426 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1427 if (err)
1428 return (err);
1429
1430 if (!zap->zap_ismicro) {
1431 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1432 if (zn) {
1433 err = fzap_count_write(zn, add, towrite,
1434 tooverwrite);
1435 zap_name_free(zn);
1436 } else {
1437 /*
1438 * We treat this case as similar to (name == NULL)
1439 */
1440 *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE;
1441 }
1442 } else {
1443 /*
1444 * We are here if (name != NULL) and this is a micro-zap.
1445 * We account for the header block depending on whether it
1446 * is freeable.
1447 *
1448 * Incase of an add-operation it is hard to find out
1449 * if this add will promote this microzap to fatzap.
1450 * Hence, we consider the worst case and account for the
1451 * blocks assuming this microzap would be promoted to a
1452 * fatzap.
1453 *
1454 * 1 block overwritten : header block
1455 * 4 new blocks written : 2 new split leaf, 2 grown
1456 * ptrtbl blocks
1457 */
1458 if (dmu_buf_freeable(zap->zap_dbuf))
1459 *tooverwrite += MZAP_MAX_BLKSZ;
1460 else
1461 *towrite += MZAP_MAX_BLKSZ;
1462
1463 if (add) {
1464 *towrite += 4 * MZAP_MAX_BLKSZ;
1465 }
1466 }
1467
1468 zap_unlockdir(zap);
1469 return (err);
1470}
506
507 /*
508 * We're checking zap_ismicro without the lock held, in order to
509 * tell what type of lock we want. Once we have some sort of
510 * lock, see if it really is the right type. In practice this
511 * can only be different if it was upgraded from micro to fat,
512 * and micro wanted WRITER but fat only needs READER.
513 */
514 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
515 rw_enter(&zap->zap_rwlock, lt);
516 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
517 /* it was upgraded, now we only need reader */
518 ASSERT(lt == RW_WRITER);
519 ASSERT(RW_READER ==
520 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
521 rw_downgrade(&zap->zap_rwlock);
522 lt = RW_READER;
523 }
524
525 zap->zap_objset = os;
526
527 if (lt == RW_WRITER)
528 dmu_buf_will_dirty(db, tx);
529
530 ASSERT3P(zap->zap_dbuf, ==, db);
531
532 ASSERT(!zap->zap_ismicro ||
533 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
534 if (zap->zap_ismicro && tx && adding &&
535 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
536 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
537 if (newsz > MZAP_MAX_BLKSZ) {
538 dprintf("upgrading obj %llu: num_entries=%u\n",
539 obj, zap->zap_m.zap_num_entries);
540 *zapp = zap;
541 return (mzap_upgrade(zapp, tx, 0));
542 }
543 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
544 ASSERT0(err);
545 zap->zap_m.zap_num_chunks =
546 db->db_size / MZAP_ENT_LEN - 1;
547 }
548
549 *zapp = zap;
550 return (0);
551}
552
553void
554zap_unlockdir(zap_t *zap)
555{
556 rw_exit(&zap->zap_rwlock);
557 dmu_buf_rele(zap->zap_dbuf, NULL);
558}
559
560static int
561mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
562{
563 mzap_phys_t *mzp;
564 int i, sz, nchunks;
565 int err = 0;
566 zap_t *zap = *zapp;
567
568 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
569
570 sz = zap->zap_dbuf->db_size;
571 mzp = zio_buf_alloc(sz);
572 bcopy(zap->zap_dbuf->db_data, mzp, sz);
573 nchunks = zap->zap_m.zap_num_chunks;
574
575 if (!flags) {
576 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
577 1ULL << fzap_default_block_shift, 0, tx);
578 if (err) {
579 zio_buf_free(mzp, sz);
580 return (err);
581 }
582 }
583
584 dprintf("upgrading obj=%llu with %u chunks\n",
585 zap->zap_object, nchunks);
586 /* XXX destroy the avl later, so we can use the stored hash value */
587 mze_destroy(zap);
588
589 fzap_upgrade(zap, tx, flags);
590
591 for (i = 0; i < nchunks; i++) {
592 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
593 zap_name_t *zn;
594 if (mze->mze_name[0] == 0)
595 continue;
596 dprintf("adding %s=%llu\n",
597 mze->mze_name, mze->mze_value);
598 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
599 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
600 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
601 zap_name_free(zn);
602 if (err)
603 break;
604 }
605 zio_buf_free(mzp, sz);
606 *zapp = zap;
607 return (err);
608}
609
610void
611mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
612 dmu_tx_t *tx)
613{
614 dmu_buf_t *db;
615 mzap_phys_t *zp;
616
617 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
618
619#ifdef ZFS_DEBUG
620 {
621 dmu_object_info_t doi;
622 dmu_object_info_from_db(db, &doi);
623 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
624 }
625#endif
626
627 dmu_buf_will_dirty(db, tx);
628 zp = db->db_data;
629 zp->mz_block_type = ZBT_MICRO;
630 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
631 zp->mz_normflags = normflags;
632 dmu_buf_rele(db, FTAG);
633
634 if (flags != 0) {
635 zap_t *zap;
636 /* Only fat zap supports flags; upgrade immediately. */
637 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
638 B_FALSE, B_FALSE, &zap));
639 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
640 zap_unlockdir(zap);
641 }
642}
643
644int
645zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
646 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
647{
648 return (zap_create_claim_norm(os, obj,
649 0, ot, bonustype, bonuslen, tx));
650}
651
652int
653zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
654 dmu_object_type_t ot,
655 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
656{
657 int err;
658
659 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
660 if (err != 0)
661 return (err);
662 mzap_create_impl(os, obj, normflags, 0, tx);
663 return (0);
664}
665
666uint64_t
667zap_create(objset_t *os, dmu_object_type_t ot,
668 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
669{
670 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
671}
672
673uint64_t
674zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
675 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
676{
677 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
678
679 mzap_create_impl(os, obj, normflags, 0, tx);
680 return (obj);
681}
682
683uint64_t
684zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
685 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
686 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
687{
688 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
689
690 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
691 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
692 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
693 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
694
695 VERIFY(dmu_object_set_blocksize(os, obj,
696 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
697
698 mzap_create_impl(os, obj, normflags, flags, tx);
699 return (obj);
700}
701
702int
703zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
704{
705 /*
706 * dmu_object_free will free the object number and free the
707 * data. Freeing the data will cause our pageout function to be
708 * called, which will destroy our data (zap_leaf_t's and zap_t).
709 */
710
711 return (dmu_object_free(os, zapobj, tx));
712}
713
714void
715zap_evict(void *dbu)
716{
717 zap_t *zap = dbu;
718
719 rw_destroy(&zap->zap_rwlock);
720
721 if (zap->zap_ismicro)
722 mze_destroy(zap);
723 else
724 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
725
726 kmem_free(zap, sizeof (zap_t));
727}
728
729int
730zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
731{
732 zap_t *zap;
733 int err;
734
735 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
736 if (err)
737 return (err);
738 if (!zap->zap_ismicro) {
739 err = fzap_count(zap, count);
740 } else {
741 *count = zap->zap_m.zap_num_entries;
742 }
743 zap_unlockdir(zap);
744 return (err);
745}
746
747/*
748 * zn may be NULL; if not specified, it will be computed if needed.
749 * See also the comment above zap_entry_normalization_conflict().
750 */
751static boolean_t
752mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
753{
754 mzap_ent_t *other;
755 int direction = AVL_BEFORE;
756 boolean_t allocdzn = B_FALSE;
757
758 if (zap->zap_normflags == 0)
759 return (B_FALSE);
760
761again:
762 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
763 other && other->mze_hash == mze->mze_hash;
764 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
765
766 if (zn == NULL) {
767 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
768 MT_FIRST);
769 allocdzn = B_TRUE;
770 }
771 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
772 if (allocdzn)
773 zap_name_free(zn);
774 return (B_TRUE);
775 }
776 }
777
778 if (direction == AVL_BEFORE) {
779 direction = AVL_AFTER;
780 goto again;
781 }
782
783 if (allocdzn)
784 zap_name_free(zn);
785 return (B_FALSE);
786}
787
788/*
789 * Routines for manipulating attributes.
790 */
791
792int
793zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
794 uint64_t integer_size, uint64_t num_integers, void *buf)
795{
796 return (zap_lookup_norm(os, zapobj, name, integer_size,
797 num_integers, buf, MT_EXACT, NULL, 0, NULL));
798}
799
800int
801zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
802 uint64_t integer_size, uint64_t num_integers, void *buf,
803 matchtype_t mt, char *realname, int rn_len,
804 boolean_t *ncp)
805{
806 zap_t *zap;
807 int err;
808 mzap_ent_t *mze;
809 zap_name_t *zn;
810
811 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
812 if (err)
813 return (err);
814 zn = zap_name_alloc(zap, name, mt);
815 if (zn == NULL) {
816 zap_unlockdir(zap);
817 return (SET_ERROR(ENOTSUP));
818 }
819
820 if (!zap->zap_ismicro) {
821 err = fzap_lookup(zn, integer_size, num_integers, buf,
822 realname, rn_len, ncp);
823 } else {
824 mze = mze_find(zn);
825 if (mze == NULL) {
826 err = SET_ERROR(ENOENT);
827 } else {
828 if (num_integers < 1) {
829 err = SET_ERROR(EOVERFLOW);
830 } else if (integer_size != 8) {
831 err = SET_ERROR(EINVAL);
832 } else {
833 *(uint64_t *)buf =
834 MZE_PHYS(zap, mze)->mze_value;
835 (void) strlcpy(realname,
836 MZE_PHYS(zap, mze)->mze_name, rn_len);
837 if (ncp) {
838 *ncp = mzap_normalization_conflict(zap,
839 zn, mze);
840 }
841 }
842 }
843 }
844 zap_name_free(zn);
845 zap_unlockdir(zap);
846 return (err);
847}
848
849int
850zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
851 int key_numints)
852{
853 zap_t *zap;
854 int err;
855 zap_name_t *zn;
856
857 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
858 if (err)
859 return (err);
860 zn = zap_name_alloc_uint64(zap, key, key_numints);
861 if (zn == NULL) {
862 zap_unlockdir(zap);
863 return (SET_ERROR(ENOTSUP));
864 }
865
866 fzap_prefetch(zn);
867 zap_name_free(zn);
868 zap_unlockdir(zap);
869 return (err);
870}
871
872int
873zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
874 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
875{
876 zap_t *zap;
877 int err;
878 zap_name_t *zn;
879
880 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
881 if (err)
882 return (err);
883 zn = zap_name_alloc_uint64(zap, key, key_numints);
884 if (zn == NULL) {
885 zap_unlockdir(zap);
886 return (SET_ERROR(ENOTSUP));
887 }
888
889 err = fzap_lookup(zn, integer_size, num_integers, buf,
890 NULL, 0, NULL);
891 zap_name_free(zn);
892 zap_unlockdir(zap);
893 return (err);
894}
895
896int
897zap_contains(objset_t *os, uint64_t zapobj, const char *name)
898{
899 int err = zap_lookup_norm(os, zapobj, name, 0,
900 0, NULL, MT_EXACT, NULL, 0, NULL);
901 if (err == EOVERFLOW || err == EINVAL)
902 err = 0; /* found, but skipped reading the value */
903 return (err);
904}
905
906int
907zap_length(objset_t *os, uint64_t zapobj, const char *name,
908 uint64_t *integer_size, uint64_t *num_integers)
909{
910 zap_t *zap;
911 int err;
912 mzap_ent_t *mze;
913 zap_name_t *zn;
914
915 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
916 if (err)
917 return (err);
918 zn = zap_name_alloc(zap, name, MT_EXACT);
919 if (zn == NULL) {
920 zap_unlockdir(zap);
921 return (SET_ERROR(ENOTSUP));
922 }
923 if (!zap->zap_ismicro) {
924 err = fzap_length(zn, integer_size, num_integers);
925 } else {
926 mze = mze_find(zn);
927 if (mze == NULL) {
928 err = SET_ERROR(ENOENT);
929 } else {
930 if (integer_size)
931 *integer_size = 8;
932 if (num_integers)
933 *num_integers = 1;
934 }
935 }
936 zap_name_free(zn);
937 zap_unlockdir(zap);
938 return (err);
939}
940
941int
942zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
943 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
944{
945 zap_t *zap;
946 int err;
947 zap_name_t *zn;
948
949 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
950 if (err)
951 return (err);
952 zn = zap_name_alloc_uint64(zap, key, key_numints);
953 if (zn == NULL) {
954 zap_unlockdir(zap);
955 return (SET_ERROR(ENOTSUP));
956 }
957 err = fzap_length(zn, integer_size, num_integers);
958 zap_name_free(zn);
959 zap_unlockdir(zap);
960 return (err);
961}
962
963static void
964mzap_addent(zap_name_t *zn, uint64_t value)
965{
966 int i;
967 zap_t *zap = zn->zn_zap;
968 int start = zap->zap_m.zap_alloc_next;
969 uint32_t cd;
970
971 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
972
973#ifdef ZFS_DEBUG
974 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
975 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
976 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
977 }
978#endif
979
980 cd = mze_find_unused_cd(zap, zn->zn_hash);
981 /* given the limited size of the microzap, this can't happen */
982 ASSERT(cd < zap_maxcd(zap));
983
984again:
985 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
986 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
987 if (mze->mze_name[0] == 0) {
988 mze->mze_value = value;
989 mze->mze_cd = cd;
990 (void) strcpy(mze->mze_name, zn->zn_key_orig);
991 zap->zap_m.zap_num_entries++;
992 zap->zap_m.zap_alloc_next = i+1;
993 if (zap->zap_m.zap_alloc_next ==
994 zap->zap_m.zap_num_chunks)
995 zap->zap_m.zap_alloc_next = 0;
996 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
997 return;
998 }
999 }
1000 if (start != 0) {
1001 start = 0;
1002 goto again;
1003 }
1004 ASSERT(!"out of entries!");
1005}
1006
1007int
1008zap_add(objset_t *os, uint64_t zapobj, const char *key,
1009 int integer_size, uint64_t num_integers,
1010 const void *val, dmu_tx_t *tx)
1011{
1012 zap_t *zap;
1013 int err;
1014 mzap_ent_t *mze;
1015 const uint64_t *intval = val;
1016 zap_name_t *zn;
1017
1018 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1019 if (err)
1020 return (err);
1021 zn = zap_name_alloc(zap, key, MT_EXACT);
1022 if (zn == NULL) {
1023 zap_unlockdir(zap);
1024 return (SET_ERROR(ENOTSUP));
1025 }
1026 if (!zap->zap_ismicro) {
1027 err = fzap_add(zn, integer_size, num_integers, val, tx);
1028 zap = zn->zn_zap; /* fzap_add() may change zap */
1029 } else if (integer_size != 8 || num_integers != 1 ||
1030 strlen(key) >= MZAP_NAME_LEN) {
1031 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1032 if (err == 0)
1033 err = fzap_add(zn, integer_size, num_integers, val, tx);
1034 zap = zn->zn_zap; /* fzap_add() may change zap */
1035 } else {
1036 mze = mze_find(zn);
1037 if (mze != NULL) {
1038 err = SET_ERROR(EEXIST);
1039 } else {
1040 mzap_addent(zn, *intval);
1041 }
1042 }
1043 ASSERT(zap == zn->zn_zap);
1044 zap_name_free(zn);
1045 if (zap != NULL) /* may be NULL if fzap_add() failed */
1046 zap_unlockdir(zap);
1047 return (err);
1048}
1049
1050int
1051zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1052 int key_numints, int integer_size, uint64_t num_integers,
1053 const void *val, dmu_tx_t *tx)
1054{
1055 zap_t *zap;
1056 int err;
1057 zap_name_t *zn;
1058
1059 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1060 if (err)
1061 return (err);
1062 zn = zap_name_alloc_uint64(zap, key, key_numints);
1063 if (zn == NULL) {
1064 zap_unlockdir(zap);
1065 return (SET_ERROR(ENOTSUP));
1066 }
1067 err = fzap_add(zn, integer_size, num_integers, val, tx);
1068 zap = zn->zn_zap; /* fzap_add() may change zap */
1069 zap_name_free(zn);
1070 if (zap != NULL) /* may be NULL if fzap_add() failed */
1071 zap_unlockdir(zap);
1072 return (err);
1073}
1074
1075int
1076zap_update(objset_t *os, uint64_t zapobj, const char *name,
1077 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1078{
1079 zap_t *zap;
1080 mzap_ent_t *mze;
1081 uint64_t oldval;
1082 const uint64_t *intval = val;
1083 zap_name_t *zn;
1084 int err;
1085
1086#ifdef ZFS_DEBUG
1087 /*
1088 * If there is an old value, it shouldn't change across the
1089 * lockdir (eg, due to bprewrite's xlation).
1090 */
1091 if (integer_size == 8 && num_integers == 1)
1092 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1093#endif
1094
1095 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1096 if (err)
1097 return (err);
1098 zn = zap_name_alloc(zap, name, MT_EXACT);
1099 if (zn == NULL) {
1100 zap_unlockdir(zap);
1101 return (SET_ERROR(ENOTSUP));
1102 }
1103 if (!zap->zap_ismicro) {
1104 err = fzap_update(zn, integer_size, num_integers, val, tx);
1105 zap = zn->zn_zap; /* fzap_update() may change zap */
1106 } else if (integer_size != 8 || num_integers != 1 ||
1107 strlen(name) >= MZAP_NAME_LEN) {
1108 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1109 zapobj, integer_size, num_integers, name);
1110 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1111 if (err == 0)
1112 err = fzap_update(zn, integer_size, num_integers,
1113 val, tx);
1114 zap = zn->zn_zap; /* fzap_update() may change zap */
1115 } else {
1116 mze = mze_find(zn);
1117 if (mze != NULL) {
1118 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1119 MZE_PHYS(zap, mze)->mze_value = *intval;
1120 } else {
1121 mzap_addent(zn, *intval);
1122 }
1123 }
1124 ASSERT(zap == zn->zn_zap);
1125 zap_name_free(zn);
1126 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1127 zap_unlockdir(zap);
1128 return (err);
1129}
1130
1131int
1132zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1133 int key_numints,
1134 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1135{
1136 zap_t *zap;
1137 zap_name_t *zn;
1138 int err;
1139
1140 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1141 if (err)
1142 return (err);
1143 zn = zap_name_alloc_uint64(zap, key, key_numints);
1144 if (zn == NULL) {
1145 zap_unlockdir(zap);
1146 return (SET_ERROR(ENOTSUP));
1147 }
1148 err = fzap_update(zn, integer_size, num_integers, val, tx);
1149 zap = zn->zn_zap; /* fzap_update() may change zap */
1150 zap_name_free(zn);
1151 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1152 zap_unlockdir(zap);
1153 return (err);
1154}
1155
1156int
1157zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1158{
1159 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1160}
1161
1162int
1163zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1164 matchtype_t mt, dmu_tx_t *tx)
1165{
1166 zap_t *zap;
1167 int err;
1168 mzap_ent_t *mze;
1169 zap_name_t *zn;
1170
1171 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1172 if (err)
1173 return (err);
1174 zn = zap_name_alloc(zap, name, mt);
1175 if (zn == NULL) {
1176 zap_unlockdir(zap);
1177 return (SET_ERROR(ENOTSUP));
1178 }
1179 if (!zap->zap_ismicro) {
1180 err = fzap_remove(zn, tx);
1181 } else {
1182 mze = mze_find(zn);
1183 if (mze == NULL) {
1184 err = SET_ERROR(ENOENT);
1185 } else {
1186 zap->zap_m.zap_num_entries--;
1187 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1188 sizeof (mzap_ent_phys_t));
1189 mze_remove(zap, mze);
1190 }
1191 }
1192 zap_name_free(zn);
1193 zap_unlockdir(zap);
1194 return (err);
1195}
1196
1197int
1198zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1199 int key_numints, dmu_tx_t *tx)
1200{
1201 zap_t *zap;
1202 int err;
1203 zap_name_t *zn;
1204
1205 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1206 if (err)
1207 return (err);
1208 zn = zap_name_alloc_uint64(zap, key, key_numints);
1209 if (zn == NULL) {
1210 zap_unlockdir(zap);
1211 return (SET_ERROR(ENOTSUP));
1212 }
1213 err = fzap_remove(zn, tx);
1214 zap_name_free(zn);
1215 zap_unlockdir(zap);
1216 return (err);
1217}
1218
1219/*
1220 * Routines for iterating over the attributes.
1221 */
1222
1223void
1224zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1225 uint64_t serialized)
1226{
1227 zc->zc_objset = os;
1228 zc->zc_zap = NULL;
1229 zc->zc_leaf = NULL;
1230 zc->zc_zapobj = zapobj;
1231 zc->zc_serialized = serialized;
1232 zc->zc_hash = 0;
1233 zc->zc_cd = 0;
1234}
1235
1236void
1237zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1238{
1239 zap_cursor_init_serialized(zc, os, zapobj, 0);
1240}
1241
1242void
1243zap_cursor_fini(zap_cursor_t *zc)
1244{
1245 if (zc->zc_zap) {
1246 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1247 zap_unlockdir(zc->zc_zap);
1248 zc->zc_zap = NULL;
1249 }
1250 if (zc->zc_leaf) {
1251 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1252 zap_put_leaf(zc->zc_leaf);
1253 zc->zc_leaf = NULL;
1254 }
1255 zc->zc_objset = NULL;
1256}
1257
1258uint64_t
1259zap_cursor_serialize(zap_cursor_t *zc)
1260{
1261 if (zc->zc_hash == -1ULL)
1262 return (-1ULL);
1263 if (zc->zc_zap == NULL)
1264 return (zc->zc_serialized);
1265 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1266 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1267
1268 /*
1269 * We want to keep the high 32 bits of the cursor zero if we can, so
1270 * that 32-bit programs can access this. So usually use a small
1271 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1272 * of the cursor.
1273 *
1274 * [ collision differentiator | zap_hashbits()-bit hash value ]
1275 */
1276 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1277 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1278}
1279
1280int
1281zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1282{
1283 int err;
1284 avl_index_t idx;
1285 mzap_ent_t mze_tofind;
1286 mzap_ent_t *mze;
1287
1288 if (zc->zc_hash == -1ULL)
1289 return (SET_ERROR(ENOENT));
1290
1291 if (zc->zc_zap == NULL) {
1292 int hb;
1293 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1294 RW_READER, TRUE, FALSE, &zc->zc_zap);
1295 if (err)
1296 return (err);
1297
1298 /*
1299 * To support zap_cursor_init_serialized, advance, retrieve,
1300 * we must add to the existing zc_cd, which may already
1301 * be 1 due to the zap_cursor_advance.
1302 */
1303 ASSERT(zc->zc_hash == 0);
1304 hb = zap_hashbits(zc->zc_zap);
1305 zc->zc_hash = zc->zc_serialized << (64 - hb);
1306 zc->zc_cd += zc->zc_serialized >> hb;
1307 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1308 zc->zc_cd = 0;
1309 } else {
1310 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1311 }
1312 if (!zc->zc_zap->zap_ismicro) {
1313 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1314 } else {
1315 mze_tofind.mze_hash = zc->zc_hash;
1316 mze_tofind.mze_cd = zc->zc_cd;
1317
1318 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1319 if (mze == NULL) {
1320 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1321 idx, AVL_AFTER);
1322 }
1323 if (mze) {
1324 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1325 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1326 za->za_normalization_conflict =
1327 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1328 za->za_integer_length = 8;
1329 za->za_num_integers = 1;
1330 za->za_first_integer = mzep->mze_value;
1331 (void) strcpy(za->za_name, mzep->mze_name);
1332 zc->zc_hash = mze->mze_hash;
1333 zc->zc_cd = mze->mze_cd;
1334 err = 0;
1335 } else {
1336 zc->zc_hash = -1ULL;
1337 err = SET_ERROR(ENOENT);
1338 }
1339 }
1340 rw_exit(&zc->zc_zap->zap_rwlock);
1341 return (err);
1342}
1343
1344void
1345zap_cursor_advance(zap_cursor_t *zc)
1346{
1347 if (zc->zc_hash == -1ULL)
1348 return;
1349 zc->zc_cd++;
1350}
1351
1352int
1353zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1354{
1355 int err = 0;
1356 mzap_ent_t *mze;
1357 zap_name_t *zn;
1358
1359 if (zc->zc_zap == NULL) {
1360 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1361 RW_READER, TRUE, FALSE, &zc->zc_zap);
1362 if (err)
1363 return (err);
1364 } else {
1365 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1366 }
1367
1368 zn = zap_name_alloc(zc->zc_zap, name, mt);
1369 if (zn == NULL) {
1370 rw_exit(&zc->zc_zap->zap_rwlock);
1371 return (SET_ERROR(ENOTSUP));
1372 }
1373
1374 if (!zc->zc_zap->zap_ismicro) {
1375 err = fzap_cursor_move_to_key(zc, zn);
1376 } else {
1377 mze = mze_find(zn);
1378 if (mze == NULL) {
1379 err = SET_ERROR(ENOENT);
1380 goto out;
1381 }
1382 zc->zc_hash = mze->mze_hash;
1383 zc->zc_cd = mze->mze_cd;
1384 }
1385
1386out:
1387 zap_name_free(zn);
1388 rw_exit(&zc->zc_zap->zap_rwlock);
1389 return (err);
1390}
1391
1392int
1393zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1394{
1395 int err;
1396 zap_t *zap;
1397
1398 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1399 if (err)
1400 return (err);
1401
1402 bzero(zs, sizeof (zap_stats_t));
1403
1404 if (zap->zap_ismicro) {
1405 zs->zs_blocksize = zap->zap_dbuf->db_size;
1406 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1407 zs->zs_num_blocks = 1;
1408 } else {
1409 fzap_get_stats(zap, zs);
1410 }
1411 zap_unlockdir(zap);
1412 return (0);
1413}
1414
1415int
1416zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1417 uint64_t *towrite, uint64_t *tooverwrite)
1418{
1419 zap_t *zap;
1420 int err = 0;
1421
1422 /*
1423 * Since, we don't have a name, we cannot figure out which blocks will
1424 * be affected in this operation. So, account for the worst case :
1425 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1426 * - 4 new blocks written if adding:
1427 * - 2 blocks for possibly split leaves,
1428 * - 2 grown ptrtbl blocks
1429 *
1430 * This also accomodates the case where an add operation to a fairly
1431 * large microzap results in a promotion to fatzap.
1432 */
1433 if (name == NULL) {
1434 *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE;
1435 return (err);
1436 }
1437
1438 /*
1439 * We lock the zap with adding == FALSE. Because, if we pass
1440 * the actual value of add, it could trigger a mzap_upgrade().
1441 * At present we are just evaluating the possibility of this operation
1442 * and hence we donot want to trigger an upgrade.
1443 */
1444 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1445 if (err)
1446 return (err);
1447
1448 if (!zap->zap_ismicro) {
1449 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1450 if (zn) {
1451 err = fzap_count_write(zn, add, towrite,
1452 tooverwrite);
1453 zap_name_free(zn);
1454 } else {
1455 /*
1456 * We treat this case as similar to (name == NULL)
1457 */
1458 *towrite += (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE;
1459 }
1460 } else {
1461 /*
1462 * We are here if (name != NULL) and this is a micro-zap.
1463 * We account for the header block depending on whether it
1464 * is freeable.
1465 *
1466 * Incase of an add-operation it is hard to find out
1467 * if this add will promote this microzap to fatzap.
1468 * Hence, we consider the worst case and account for the
1469 * blocks assuming this microzap would be promoted to a
1470 * fatzap.
1471 *
1472 * 1 block overwritten : header block
1473 * 4 new blocks written : 2 new split leaf, 2 grown
1474 * ptrtbl blocks
1475 */
1476 if (dmu_buf_freeable(zap->zap_dbuf))
1477 *tooverwrite += MZAP_MAX_BLKSZ;
1478 else
1479 *towrite += MZAP_MAX_BLKSZ;
1480
1481 if (add) {
1482 *towrite += 4 * MZAP_MAX_BLKSZ;
1483 }
1484 }
1485
1486 zap_unlockdir(zap);
1487 return (err);
1488}