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) 2012 by Delphix. All rights reserved.
24 */
25
26#include <sys/zio.h>
27#include <sys/spa.h>
28#include <sys/dmu.h>
29#include <sys/zfs_context.h>
30#include <sys/zap.h>
31#include <sys/refcount.h>
32#include <sys/zap_impl.h>
33#include <sys/zap_leaf.h>
34#include <sys/avl.h>
35#include <sys/arc.h>
36
37#ifdef _KERNEL
38#include <sys/sunddi.h>
39#endif
40
41static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
42
43uint64_t
44zap_getflags(zap_t *zap)
45{
46 if (zap->zap_ismicro)
47 return (0);
48 return (zap->zap_u.zap_fat.zap_phys->zap_flags);
49}
50
51int
52zap_hashbits(zap_t *zap)
53{
54 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
55 return (48);
56 else
57 return (28);
58}
59
60uint32_t
61zap_maxcd(zap_t *zap)
62{
63 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
64 return ((1<<16)-1);
65 else
66 return (-1U);
67}
68
69static uint64_t
70zap_hash(zap_name_t *zn)
71{
72 zap_t *zap = zn->zn_zap;
73 uint64_t h = 0;
74
75 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
76 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
77 h = *(uint64_t *)zn->zn_key_orig;
78 } else {
79 h = zap->zap_salt;
80 ASSERT(h != 0);
81 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
82
83 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
84 int i;
85 const uint64_t *wp = zn->zn_key_norm;
86
87 ASSERT(zn->zn_key_intlen == 8);
88 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
89 int j;
90 uint64_t word = *wp;
91
92 for (j = 0; j < zn->zn_key_intlen; j++) {
93 h = (h >> 8) ^
94 zfs_crc64_table[(h ^ word) & 0xFF];
95 word >>= NBBY;
96 }
97 }
98 } else {
99 int i, len;
100 const uint8_t *cp = zn->zn_key_norm;
101
102 /*
103 * We previously stored the terminating null on
104 * disk, but didn't hash it, so we need to
105 * continue to not hash it. (The
106 * zn_key_*_numints includes the terminating
107 * null for non-binary keys.)
108 */
109 len = zn->zn_key_norm_numints - 1;
110
111 ASSERT(zn->zn_key_intlen == 1);
112 for (i = 0; i < len; cp++, i++) {
113 h = (h >> 8) ^
114 zfs_crc64_table[(h ^ *cp) & 0xFF];
115 }
116 }
117 }
118 /*
119 * Don't use all 64 bits, since we need some in the cookie for
120 * the collision differentiator. We MUST use the high bits,
121 * since those are the ones that we first pay attention to when
122 * chosing the bucket.
123 */
124 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
125
126 return (h);
127}
128
129static int
130zap_normalize(zap_t *zap, const char *name, char *namenorm)
131{
132 size_t inlen, outlen;
133 int err;
134
135 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
136
137 inlen = strlen(name) + 1;
138 outlen = ZAP_MAXNAMELEN;
139
140 err = 0;
141 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
142 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
143 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
144
145 return (err);
146}
147
148boolean_t
149zap_match(zap_name_t *zn, const char *matchname)
150{
151 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
152
153 if (zn->zn_matchtype == MT_FIRST) {
154 char norm[ZAP_MAXNAMELEN];
155
156 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
157 return (B_FALSE);
158
159 return (strcmp(zn->zn_key_norm, norm) == 0);
160 } else {
161 /* MT_BEST or MT_EXACT */
162 return (strcmp(zn->zn_key_orig, matchname) == 0);
163 }
164}
165
166void
167zap_name_free(zap_name_t *zn)
168{
169 kmem_free(zn, sizeof (zap_name_t));
170}
171
172zap_name_t *
173zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
174{
175 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
176
177 zn->zn_zap = zap;
178 zn->zn_key_intlen = sizeof (*key);
179 zn->zn_key_orig = key;
180 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
181 zn->zn_matchtype = mt;
182 if (zap->zap_normflags) {
183 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
184 zap_name_free(zn);
185 return (NULL);
186 }
187 zn->zn_key_norm = zn->zn_normbuf;
188 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
189 } else {
190 if (mt != MT_EXACT) {
191 zap_name_free(zn);
192 return (NULL);
193 }
194 zn->zn_key_norm = zn->zn_key_orig;
195 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
196 }
197
198 zn->zn_hash = zap_hash(zn);
199 return (zn);
200}
201
202zap_name_t *
203zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
204{
205 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
206
207 ASSERT(zap->zap_normflags == 0);
208 zn->zn_zap = zap;
209 zn->zn_key_intlen = sizeof (*key);
210 zn->zn_key_orig = zn->zn_key_norm = key;
211 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
212 zn->zn_matchtype = MT_EXACT;
213
214 zn->zn_hash = zap_hash(zn);
215 return (zn);
216}
217
218static void
219mzap_byteswap(mzap_phys_t *buf, size_t size)
220{
221 int i, max;
222 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
223 buf->mz_salt = BSWAP_64(buf->mz_salt);
224 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
225 max = (size / MZAP_ENT_LEN) - 1;
226 for (i = 0; i < max; i++) {
227 buf->mz_chunk[i].mze_value =
228 BSWAP_64(buf->mz_chunk[i].mze_value);
229 buf->mz_chunk[i].mze_cd =
230 BSWAP_32(buf->mz_chunk[i].mze_cd);
231 }
232}
233
234void
235zap_byteswap(void *buf, size_t size)
236{
237 uint64_t block_type;
238
239 block_type = *(uint64_t *)buf;
240
241 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
242 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
243 mzap_byteswap(buf, size);
244 } else {
245 fzap_byteswap(buf, size);
246 }
247}
248
249static int
250mze_compare(const void *arg1, const void *arg2)
251{
252 const mzap_ent_t *mze1 = arg1;
253 const mzap_ent_t *mze2 = arg2;
254
255 if (mze1->mze_hash > mze2->mze_hash)
256 return (+1);
257 if (mze1->mze_hash < mze2->mze_hash)
258 return (-1);
259 if (mze1->mze_cd > mze2->mze_cd)
260 return (+1);
261 if (mze1->mze_cd < mze2->mze_cd)
262 return (-1);
263 return (0);
264}
265
266static int
267mze_insert(zap_t *zap, int chunkid, uint64_t hash)
268{
269 mzap_ent_t *mze;
270 avl_index_t idx;
271
272 ASSERT(zap->zap_ismicro);
273 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
274
275 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
276 mze->mze_chunkid = chunkid;
277 mze->mze_hash = hash;
278 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
279 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
280 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
281 kmem_free(mze, sizeof (mzap_ent_t));
282 return (EEXIST);
283 }
284 avl_insert(&zap->zap_m.zap_avl, mze, idx);
285 return (0);
286}
287
288static mzap_ent_t *
289mze_find(zap_name_t *zn)
290{
291 mzap_ent_t mze_tofind;
292 mzap_ent_t *mze;
293 avl_index_t idx;
294 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
295
296 ASSERT(zn->zn_zap->zap_ismicro);
297 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
298
299 mze_tofind.mze_hash = zn->zn_hash;
300 mze_tofind.mze_cd = 0;
301
302again:
303 mze = avl_find(avl, &mze_tofind, &idx);
304 if (mze == NULL)
305 mze = avl_nearest(avl, idx, AVL_AFTER);
306 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
307 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
308 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
309 return (mze);
310 }
311 if (zn->zn_matchtype == MT_BEST) {
312 zn->zn_matchtype = MT_FIRST;
313 goto again;
314 }
315 return (NULL);
316}
317
318static uint32_t
319mze_find_unused_cd(zap_t *zap, uint64_t hash)
320{
321 mzap_ent_t mze_tofind;
322 mzap_ent_t *mze;
323 avl_index_t idx;
324 avl_tree_t *avl = &zap->zap_m.zap_avl;
325 uint32_t cd;
326
327 ASSERT(zap->zap_ismicro);
328 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
329
330 mze_tofind.mze_hash = hash;
331 mze_tofind.mze_cd = 0;
332
333 cd = 0;
334 for (mze = avl_find(avl, &mze_tofind, &idx);
335 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
336 if (mze->mze_cd != cd)
337 break;
338 cd++;
339 }
340
341 return (cd);
342}
343
344static void
345mze_remove(zap_t *zap, mzap_ent_t *mze)
346{
347 ASSERT(zap->zap_ismicro);
348 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
349
350 avl_remove(&zap->zap_m.zap_avl, mze);
351 kmem_free(mze, sizeof (mzap_ent_t));
352}
353
354static void
355mze_destroy(zap_t *zap)
356{
357 mzap_ent_t *mze;
358 void *avlcookie = NULL;
359
360 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
361 kmem_free(mze, sizeof (mzap_ent_t));
362 avl_destroy(&zap->zap_m.zap_avl);
363}
364
365static zap_t *
366mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
367{
368 zap_t *winner;
369 zap_t *zap;
370 int i;
371
372 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
373
374 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
375 rw_init(&zap->zap_rwlock, 0, 0, 0);
376 rw_enter(&zap->zap_rwlock, RW_WRITER);
377 zap->zap_objset = os;
378 zap->zap_object = obj;
379 zap->zap_dbuf = db;
380
381 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
382 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
383 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
384 } else {
385 zap->zap_ismicro = TRUE;
386 }
387
388 /*
389 * Make sure that zap_ismicro is set before we let others see
390 * it, because zap_lockdir() checks zap_ismicro without the lock
391 * held.
392 */
393 winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
394
395 if (winner != NULL) {
396 rw_exit(&zap->zap_rwlock);
397 rw_destroy(&zap->zap_rwlock);
398 if (!zap->zap_ismicro)
399 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
400 kmem_free(zap, sizeof (zap_t));
401 return (winner);
402 }
403
404 if (zap->zap_ismicro) {
405 zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
406 zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
407 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
408 avl_create(&zap->zap_m.zap_avl, mze_compare,
409 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
410
411 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
412 mzap_ent_phys_t *mze =
413 &zap->zap_m.zap_phys->mz_chunk[i];
414 if (mze->mze_name[0]) {
415 zap_name_t *zn;
416
417 zn = zap_name_alloc(zap, mze->mze_name,
418 MT_EXACT);
419 if (mze_insert(zap, i, zn->zn_hash) == 0)
420 zap->zap_m.zap_num_entries++;
421 else {
422 printf("ZFS WARNING: Duplicated ZAP "
423 "entry detected (%s).\n",
424 mze->mze_name);
425 }
426 zap_name_free(zn);
427 }
428 }
429 } else {
430 zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
431 zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
432
433 ASSERT3U(sizeof (struct zap_leaf_header), ==,
434 2*ZAP_LEAF_CHUNKSIZE);
435
436 /*
437 * The embedded pointer table should not overlap the
438 * other members.
439 */
440 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
441 &zap->zap_f.zap_phys->zap_salt);
442
443 /*
444 * The embedded pointer table should end at the end of
445 * the block
446 */
447 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
448 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
449 (uintptr_t)zap->zap_f.zap_phys, ==,
450 zap->zap_dbuf->db_size);
451 }
452 rw_exit(&zap->zap_rwlock);
453 return (zap);
454}
455
456int
457zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
458 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
459{
460 zap_t *zap;
461 dmu_buf_t *db;
462 krw_t lt;
463 int err;
464
465 *zapp = NULL;
466
467 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
468 if (err)
469 return (err);
470
471#ifdef ZFS_DEBUG
472 {
473 dmu_object_info_t doi;
474 dmu_object_info_from_db(db, &doi);
475 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
476 }
477#endif
478
479 zap = dmu_buf_get_user(db);
480 if (zap == NULL)
481 zap = mzap_open(os, obj, db);
482
483 /*
484 * We're checking zap_ismicro without the lock held, in order to
485 * tell what type of lock we want. Once we have some sort of
486 * lock, see if it really is the right type. In practice this
487 * can only be different if it was upgraded from micro to fat,
488 * and micro wanted WRITER but fat only needs READER.
489 */
490 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
491 rw_enter(&zap->zap_rwlock, lt);
492 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
493 /* it was upgraded, now we only need reader */
494 ASSERT(lt == RW_WRITER);
495 ASSERT(RW_READER ==
496 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
497 rw_downgrade(&zap->zap_rwlock);
498 lt = RW_READER;
499 }
500
501 zap->zap_objset = os;
502
503 if (lt == RW_WRITER)
504 dmu_buf_will_dirty(db, tx);
505
506 ASSERT3P(zap->zap_dbuf, ==, db);
507
508 ASSERT(!zap->zap_ismicro ||
509 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
510 if (zap->zap_ismicro && tx && adding &&
511 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
512 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
513 if (newsz > MZAP_MAX_BLKSZ) {
514 dprintf("upgrading obj %llu: num_entries=%u\n",
515 obj, zap->zap_m.zap_num_entries);
516 *zapp = zap;
517 return (mzap_upgrade(zapp, tx, 0));
518 }
519 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
520 ASSERT0(err);
521 zap->zap_m.zap_num_chunks =
522 db->db_size / MZAP_ENT_LEN - 1;
523 }
524
525 *zapp = zap;
526 return (0);
527}
528
529void
530zap_unlockdir(zap_t *zap)
531{
532 rw_exit(&zap->zap_rwlock);
533 dmu_buf_rele(zap->zap_dbuf, NULL);
534}
535
536static int
537mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
538{
539 mzap_phys_t *mzp;
540 int i, sz, nchunks;
541 int err = 0;
542 zap_t *zap = *zapp;
543
544 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
545
546 sz = zap->zap_dbuf->db_size;
547 mzp = kmem_alloc(sz, KM_SLEEP);
548 bcopy(zap->zap_dbuf->db_data, mzp, sz);
549 nchunks = zap->zap_m.zap_num_chunks;
550
551 if (!flags) {
552 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
553 1ULL << fzap_default_block_shift, 0, tx);
554 if (err) {
555 kmem_free(mzp, sz);
556 return (err);
557 }
558 }
559
560 dprintf("upgrading obj=%llu with %u chunks\n",
561 zap->zap_object, nchunks);
562 /* XXX destroy the avl later, so we can use the stored hash value */
563 mze_destroy(zap);
564
565 fzap_upgrade(zap, tx, flags);
566
567 for (i = 0; i < nchunks; i++) {
568 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
569 zap_name_t *zn;
570 if (mze->mze_name[0] == 0)
571 continue;
572 dprintf("adding %s=%llu\n",
573 mze->mze_name, mze->mze_value);
574 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
575 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
576 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
577 zap_name_free(zn);
578 if (err)
579 break;
580 }
581 kmem_free(mzp, sz);
582 *zapp = zap;
583 return (err);
584}
585
586static void
587mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
588 dmu_tx_t *tx)
589{
590 dmu_buf_t *db;
591 mzap_phys_t *zp;
592
593 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
594
595#ifdef ZFS_DEBUG
596 {
597 dmu_object_info_t doi;
598 dmu_object_info_from_db(db, &doi);
599 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
600 }
601#endif
602
603 dmu_buf_will_dirty(db, tx);
604 zp = db->db_data;
605 zp->mz_block_type = ZBT_MICRO;
606 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
607 zp->mz_normflags = normflags;
608 dmu_buf_rele(db, FTAG);
609
610 if (flags != 0) {
611 zap_t *zap;
612 /* Only fat zap supports flags; upgrade immediately. */
613 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
614 B_FALSE, B_FALSE, &zap));
615 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
616 zap_unlockdir(zap);
617 }
618}
619
620int
621zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
622 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
623{
624 return (zap_create_claim_norm(os, obj,
625 0, ot, bonustype, bonuslen, tx));
626}
627
628int
629zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
630 dmu_object_type_t ot,
631 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
632{
633 int err;
634
635 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
636 if (err != 0)
637 return (err);
638 mzap_create_impl(os, obj, normflags, 0, tx);
639 return (0);
640}
641
642uint64_t
643zap_create(objset_t *os, dmu_object_type_t ot,
644 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
645{
646 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
647}
648
649uint64_t
650zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
651 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
652{
653 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
654
655 mzap_create_impl(os, obj, normflags, 0, tx);
656 return (obj);
657}
658
659uint64_t
660zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
661 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
662 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
663{
664 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
665
666 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
667 leaf_blockshift <= SPA_MAXBLOCKSHIFT &&
668 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
669 indirect_blockshift <= SPA_MAXBLOCKSHIFT);
670
671 VERIFY(dmu_object_set_blocksize(os, obj,
672 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
673
674 mzap_create_impl(os, obj, normflags, flags, tx);
675 return (obj);
676}
677
678int
679zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
680{
681 /*
682 * dmu_object_free will free the object number and free the
683 * data. Freeing the data will cause our pageout function to be
684 * called, which will destroy our data (zap_leaf_t's and zap_t).
685 */
686
687 return (dmu_object_free(os, zapobj, tx));
688}
689
690_NOTE(ARGSUSED(0))
691void
692zap_evict(dmu_buf_t *db, void *vzap)
693{
694 zap_t *zap = vzap;
695
696 rw_destroy(&zap->zap_rwlock);
697
698 if (zap->zap_ismicro)
699 mze_destroy(zap);
700 else
701 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
702
703 kmem_free(zap, sizeof (zap_t));
704}
705
706int
707zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
708{
709 zap_t *zap;
710 int err;
711
712 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
713 if (err)
714 return (err);
715 if (!zap->zap_ismicro) {
716 err = fzap_count(zap, count);
717 } else {
718 *count = zap->zap_m.zap_num_entries;
719 }
720 zap_unlockdir(zap);
721 return (err);
722}
723
724/*
725 * zn may be NULL; if not specified, it will be computed if needed.
726 * See also the comment above zap_entry_normalization_conflict().
727 */
728static boolean_t
729mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
730{
731 mzap_ent_t *other;
732 int direction = AVL_BEFORE;
733 boolean_t allocdzn = B_FALSE;
734
735 if (zap->zap_normflags == 0)
736 return (B_FALSE);
737
738again:
739 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
740 other && other->mze_hash == mze->mze_hash;
741 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
742
743 if (zn == NULL) {
744 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
745 MT_FIRST);
746 allocdzn = B_TRUE;
747 }
748 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
749 if (allocdzn)
750 zap_name_free(zn);
751 return (B_TRUE);
752 }
753 }
754
755 if (direction == AVL_BEFORE) {
756 direction = AVL_AFTER;
757 goto again;
758 }
759
760 if (allocdzn)
761 zap_name_free(zn);
762 return (B_FALSE);
763}
764
765/*
766 * Routines for manipulating attributes.
767 */
768
769int
770zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
771 uint64_t integer_size, uint64_t num_integers, void *buf)
772{
773 return (zap_lookup_norm(os, zapobj, name, integer_size,
774 num_integers, buf, MT_EXACT, NULL, 0, NULL));
775}
776
777int
778zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
779 uint64_t integer_size, uint64_t num_integers, void *buf,
780 matchtype_t mt, char *realname, int rn_len,
781 boolean_t *ncp)
782{
783 zap_t *zap;
784 int err;
785 mzap_ent_t *mze;
786 zap_name_t *zn;
787
788 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
789 if (err)
790 return (err);
791 zn = zap_name_alloc(zap, name, mt);
792 if (zn == NULL) {
793 zap_unlockdir(zap);
794 return (ENOTSUP);
795 }
796
797 if (!zap->zap_ismicro) {
798 err = fzap_lookup(zn, integer_size, num_integers, buf,
799 realname, rn_len, ncp);
800 } else {
801 mze = mze_find(zn);
802 if (mze == NULL) {
803 err = ENOENT;
804 } else {
805 if (num_integers < 1) {
806 err = EOVERFLOW;
807 } else if (integer_size != 8) {
808 err = EINVAL;
809 } else {
810 *(uint64_t *)buf =
811 MZE_PHYS(zap, mze)->mze_value;
812 (void) strlcpy(realname,
813 MZE_PHYS(zap, mze)->mze_name, rn_len);
814 if (ncp) {
815 *ncp = mzap_normalization_conflict(zap,
816 zn, mze);
817 }
818 }
819 }
820 }
821 zap_name_free(zn);
822 zap_unlockdir(zap);
823 return (err);
824}
825
826int
827zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
828 int key_numints)
829{
830 zap_t *zap;
831 int err;
832 zap_name_t *zn;
833
834 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
835 if (err)
836 return (err);
837 zn = zap_name_alloc_uint64(zap, key, key_numints);
838 if (zn == NULL) {
839 zap_unlockdir(zap);
840 return (ENOTSUP);
841 }
842
843 fzap_prefetch(zn);
844 zap_name_free(zn);
845 zap_unlockdir(zap);
846 return (err);
847}
848
849int
850zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
851 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
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 (ENOTSUP);
864 }
865
866 err = fzap_lookup(zn, integer_size, num_integers, buf,
867 NULL, 0, NULL);
868 zap_name_free(zn);
869 zap_unlockdir(zap);
870 return (err);
871}
872
873int
874zap_contains(objset_t *os, uint64_t zapobj, const char *name)
875{
876 int err = (zap_lookup_norm(os, zapobj, name, 0,
877 0, NULL, MT_EXACT, NULL, 0, NULL));
878 if (err == EOVERFLOW || err == EINVAL)
879 err = 0; /* found, but skipped reading the value */
880 return (err);
881}
882
883int
884zap_length(objset_t *os, uint64_t zapobj, const char *name,
885 uint64_t *integer_size, uint64_t *num_integers)
886{
887 zap_t *zap;
888 int err;
889 mzap_ent_t *mze;
890 zap_name_t *zn;
891
892 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
893 if (err)
894 return (err);
895 zn = zap_name_alloc(zap, name, MT_EXACT);
896 if (zn == NULL) {
897 zap_unlockdir(zap);
898 return (ENOTSUP);
899 }
900 if (!zap->zap_ismicro) {
901 err = fzap_length(zn, integer_size, num_integers);
902 } else {
903 mze = mze_find(zn);
904 if (mze == NULL) {
905 err = ENOENT;
906 } else {
907 if (integer_size)
908 *integer_size = 8;
909 if (num_integers)
910 *num_integers = 1;
911 }
912 }
913 zap_name_free(zn);
914 zap_unlockdir(zap);
915 return (err);
916}
917
918int
919zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
920 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
921{
922 zap_t *zap;
923 int err;
924 zap_name_t *zn;
925
926 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
927 if (err)
928 return (err);
929 zn = zap_name_alloc_uint64(zap, key, key_numints);
930 if (zn == NULL) {
931 zap_unlockdir(zap);
932 return (ENOTSUP);
933 }
934 err = fzap_length(zn, integer_size, num_integers);
935 zap_name_free(zn);
936 zap_unlockdir(zap);
937 return (err);
938}
939
940static void
941mzap_addent(zap_name_t *zn, uint64_t value)
942{
943 int i;
944 zap_t *zap = zn->zn_zap;
945 int start = zap->zap_m.zap_alloc_next;
946 uint32_t cd;
947
948 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
949
950#ifdef ZFS_DEBUG
951 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
952 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
953 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
954 }
955#endif
956
957 cd = mze_find_unused_cd(zap, zn->zn_hash);
958 /* given the limited size of the microzap, this can't happen */
959 ASSERT(cd < zap_maxcd(zap));
960
961again:
962 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
963 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
964 if (mze->mze_name[0] == 0) {
965 mze->mze_value = value;
966 mze->mze_cd = cd;
967 (void) strcpy(mze->mze_name, zn->zn_key_orig);
968 zap->zap_m.zap_num_entries++;
969 zap->zap_m.zap_alloc_next = i+1;
970 if (zap->zap_m.zap_alloc_next ==
971 zap->zap_m.zap_num_chunks)
972 zap->zap_m.zap_alloc_next = 0;
973 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
974 return;
975 }
976 }
977 if (start != 0) {
978 start = 0;
979 goto again;
980 }
981 ASSERT(!"out of entries!");
982}
983
984int
985zap_add(objset_t *os, uint64_t zapobj, const char *key,
986 int integer_size, uint64_t num_integers,
987 const void *val, dmu_tx_t *tx)
988{
989 zap_t *zap;
990 int err;
991 mzap_ent_t *mze;
992 const uint64_t *intval = val;
993 zap_name_t *zn;
994
995 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
996 if (err)
997 return (err);
998 zn = zap_name_alloc(zap, key, MT_EXACT);
999 if (zn == NULL) {
1000 zap_unlockdir(zap);
1001 return (ENOTSUP);
1002 }
1003 if (!zap->zap_ismicro) {
1004 err = fzap_add(zn, integer_size, num_integers, val, tx);
1005 zap = zn->zn_zap; /* fzap_add() may change zap */
1006 } else if (integer_size != 8 || num_integers != 1 ||
1007 strlen(key) >= MZAP_NAME_LEN) {
1008 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1009 if (err == 0)
1010 err = fzap_add(zn, integer_size, num_integers, val, tx);
1011 zap = zn->zn_zap; /* fzap_add() may change zap */
1012 } else {
1013 mze = mze_find(zn);
1014 if (mze != NULL) {
1015 err = EEXIST;
1016 } else {
1017 mzap_addent(zn, *intval);
1018 }
1019 }
1020 ASSERT(zap == zn->zn_zap);
1021 zap_name_free(zn);
1022 if (zap != NULL) /* may be NULL if fzap_add() failed */
1023 zap_unlockdir(zap);
1024 return (err);
1025}
1026
1027int
1028zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1029 int key_numints, int integer_size, uint64_t num_integers,
1030 const void *val, dmu_tx_t *tx)
1031{
1032 zap_t *zap;
1033 int err;
1034 zap_name_t *zn;
1035
1036 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1037 if (err)
1038 return (err);
1039 zn = zap_name_alloc_uint64(zap, key, key_numints);
1040 if (zn == NULL) {
1041 zap_unlockdir(zap);
1042 return (ENOTSUP);
1043 }
1044 err = fzap_add(zn, integer_size, num_integers, val, tx);
1045 zap = zn->zn_zap; /* fzap_add() may change zap */
1046 zap_name_free(zn);
1047 if (zap != NULL) /* may be NULL if fzap_add() failed */
1048 zap_unlockdir(zap);
1049 return (err);
1050}
1051
1052int
1053zap_update(objset_t *os, uint64_t zapobj, const char *name,
1054 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1055{
1056 zap_t *zap;
1057 mzap_ent_t *mze;
1058 uint64_t oldval;
1059 const uint64_t *intval = val;
1060 zap_name_t *zn;
1061 int err;
1062
1063#ifdef ZFS_DEBUG
1064 /*
1065 * If there is an old value, it shouldn't change across the
1066 * lockdir (eg, due to bprewrite's xlation).
1067 */
1068 if (integer_size == 8 && num_integers == 1)
1069 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1070#endif
1071
1072 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1073 if (err)
1074 return (err);
1075 zn = zap_name_alloc(zap, name, MT_EXACT);
1076 if (zn == NULL) {
1077 zap_unlockdir(zap);
1078 return (ENOTSUP);
1079 }
1080 if (!zap->zap_ismicro) {
1081 err = fzap_update(zn, integer_size, num_integers, val, tx);
1082 zap = zn->zn_zap; /* fzap_update() may change zap */
1083 } else if (integer_size != 8 || num_integers != 1 ||
1084 strlen(name) >= MZAP_NAME_LEN) {
1085 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1086 zapobj, integer_size, num_integers, name);
1087 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1088 if (err == 0)
1089 err = fzap_update(zn, integer_size, num_integers,
1090 val, tx);
1091 zap = zn->zn_zap; /* fzap_update() may change zap */
1092 } else {
1093 mze = mze_find(zn);
1094 if (mze != NULL) {
1095 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1096 MZE_PHYS(zap, mze)->mze_value = *intval;
1097 } else {
1098 mzap_addent(zn, *intval);
1099 }
1100 }
1101 ASSERT(zap == zn->zn_zap);
1102 zap_name_free(zn);
1103 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1104 zap_unlockdir(zap);
1105 return (err);
1106}
1107
1108int
1109zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1110 int key_numints,
1111 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1112{
1113 zap_t *zap;
1114 zap_name_t *zn;
1115 int err;
1116
1117 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1118 if (err)
1119 return (err);
1120 zn = zap_name_alloc_uint64(zap, key, key_numints);
1121 if (zn == NULL) {
1122 zap_unlockdir(zap);
1123 return (ENOTSUP);
1124 }
1125 err = fzap_update(zn, integer_size, num_integers, val, tx);
1126 zap = zn->zn_zap; /* fzap_update() may change zap */
1127 zap_name_free(zn);
1128 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1129 zap_unlockdir(zap);
1130 return (err);
1131}
1132
1133int
1134zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1135{
1136 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1137}
1138
1139int
1140zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1141 matchtype_t mt, dmu_tx_t *tx)
1142{
1143 zap_t *zap;
1144 int err;
1145 mzap_ent_t *mze;
1146 zap_name_t *zn;
1147
1148 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1149 if (err)
1150 return (err);
1151 zn = zap_name_alloc(zap, name, mt);
1152 if (zn == NULL) {
1153 zap_unlockdir(zap);
1154 return (ENOTSUP);
1155 }
1156 if (!zap->zap_ismicro) {
1157 err = fzap_remove(zn, tx);
1158 } else {
1159 mze = mze_find(zn);
1160 if (mze == NULL) {
1161 err = ENOENT;
1162 } else {
1163 zap->zap_m.zap_num_entries--;
1164 bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
1165 sizeof (mzap_ent_phys_t));
1166 mze_remove(zap, mze);
1167 }
1168 }
1169 zap_name_free(zn);
1170 zap_unlockdir(zap);
1171 return (err);
1172}
1173
1174int
1175zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1176 int key_numints, dmu_tx_t *tx)
1177{
1178 zap_t *zap;
1179 int err;
1180 zap_name_t *zn;
1181
1182 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1183 if (err)
1184 return (err);
1185 zn = zap_name_alloc_uint64(zap, key, key_numints);
1186 if (zn == NULL) {
1187 zap_unlockdir(zap);
1188 return (ENOTSUP);
1189 }
1190 err = fzap_remove(zn, tx);
1191 zap_name_free(zn);
1192 zap_unlockdir(zap);
1193 return (err);
1194}
1195
1196/*
1197 * Routines for iterating over the attributes.
1198 */
1199
1200void
1201zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1202 uint64_t serialized)
1203{
1204 zc->zc_objset = os;
1205 zc->zc_zap = NULL;
1206 zc->zc_leaf = NULL;
1207 zc->zc_zapobj = zapobj;
1208 zc->zc_serialized = serialized;
1209 zc->zc_hash = 0;
1210 zc->zc_cd = 0;
1211}
1212
1213void
1214zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1215{
1216 zap_cursor_init_serialized(zc, os, zapobj, 0);
1217}
1218
1219void
1220zap_cursor_fini(zap_cursor_t *zc)
1221{
1222 if (zc->zc_zap) {
1223 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1224 zap_unlockdir(zc->zc_zap);
1225 zc->zc_zap = NULL;
1226 }
1227 if (zc->zc_leaf) {
1228 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1229 zap_put_leaf(zc->zc_leaf);
1230 zc->zc_leaf = NULL;
1231 }
1232 zc->zc_objset = NULL;
1233}
1234
1235uint64_t
1236zap_cursor_serialize(zap_cursor_t *zc)
1237{
1238 if (zc->zc_hash == -1ULL)
1239 return (-1ULL);
1240 if (zc->zc_zap == NULL)
1241 return (zc->zc_serialized);
1242 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1243 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1244
1245 /*
1246 * We want to keep the high 32 bits of the cursor zero if we can, so
1247 * that 32-bit programs can access this. So usually use a small
1248 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1249 * of the cursor.
1250 *
1251 * [ collision differentiator | zap_hashbits()-bit hash value ]
1252 */
1253 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1254 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1255}
1256
1257int
1258zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1259{
1260 int err;
1261 avl_index_t idx;
1262 mzap_ent_t mze_tofind;
1263 mzap_ent_t *mze;
1264
1265 if (zc->zc_hash == -1ULL)
1266 return (ENOENT);
1267
1268 if (zc->zc_zap == NULL) {
1269 int hb;
1270 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1271 RW_READER, TRUE, FALSE, &zc->zc_zap);
1272 if (err)
1273 return (err);
1274
1275 /*
1276 * To support zap_cursor_init_serialized, advance, retrieve,
1277 * we must add to the existing zc_cd, which may already
1278 * be 1 due to the zap_cursor_advance.
1279 */
1280 ASSERT(zc->zc_hash == 0);
1281 hb = zap_hashbits(zc->zc_zap);
1282 zc->zc_hash = zc->zc_serialized << (64 - hb);
1283 zc->zc_cd += zc->zc_serialized >> hb;
1284 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1285 zc->zc_cd = 0;
1286 } else {
1287 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1288 }
1289 if (!zc->zc_zap->zap_ismicro) {
1290 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1291 } else {
1292 err = ENOENT;
1293
1294 mze_tofind.mze_hash = zc->zc_hash;
1295 mze_tofind.mze_cd = zc->zc_cd;
1296
1297 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1298 if (mze == NULL) {
1299 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1300 idx, AVL_AFTER);
1301 }
1302 if (mze) {
1303 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1304 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1305 za->za_normalization_conflict =
1306 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1307 za->za_integer_length = 8;
1308 za->za_num_integers = 1;
1309 za->za_first_integer = mzep->mze_value;
1310 (void) strcpy(za->za_name, mzep->mze_name);
1311 zc->zc_hash = mze->mze_hash;
1312 zc->zc_cd = mze->mze_cd;
1313 err = 0;
1314 } else {
1315 zc->zc_hash = -1ULL;
1316 }
1317 }
1318 rw_exit(&zc->zc_zap->zap_rwlock);
1319 return (err);
1320}
1321
1322void
1323zap_cursor_advance(zap_cursor_t *zc)
1324{
1325 if (zc->zc_hash == -1ULL)
1326 return;
1327 zc->zc_cd++;
1328}
1329
1330int
1331zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1332{
1333 int err = 0;
1334 mzap_ent_t *mze;
1335 zap_name_t *zn;
1336
1337 if (zc->zc_zap == NULL) {
1338 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1339 RW_READER, TRUE, FALSE, &zc->zc_zap);
1340 if (err)
1341 return (err);
1342 } else {
1343 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1344 }
1345
1346 zn = zap_name_alloc(zc->zc_zap, name, mt);
1347 if (zn == NULL) {
1348 rw_exit(&zc->zc_zap->zap_rwlock);
1349 return (ENOTSUP);
1350 }
1351
1352 if (!zc->zc_zap->zap_ismicro) {
1353 err = fzap_cursor_move_to_key(zc, zn);
1354 } else {
1355 mze = mze_find(zn);
1356 if (mze == NULL) {
1357 err = ENOENT;
1358 goto out;
1359 }
1360 zc->zc_hash = mze->mze_hash;
1361 zc->zc_cd = mze->mze_cd;
1362 }
1363
1364out:
1365 zap_name_free(zn);
1366 rw_exit(&zc->zc_zap->zap_rwlock);
1367 return (err);
1368}
1369
1370int
1371zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1372{
1373 int err;
1374 zap_t *zap;
1375
1376 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1377 if (err)
1378 return (err);
1379
1380 bzero(zs, sizeof (zap_stats_t));
1381
1382 if (zap->zap_ismicro) {
1383 zs->zs_blocksize = zap->zap_dbuf->db_size;
1384 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1385 zs->zs_num_blocks = 1;
1386 } else {
1387 fzap_get_stats(zap, zs);
1388 }
1389 zap_unlockdir(zap);
1390 return (0);
1391}
1392
1393int
1394zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1395 uint64_t *towrite, uint64_t *tooverwrite)
1396{
1397 zap_t *zap;
1398 int err = 0;
1399
1400
1401 /*
1402 * Since, we don't have a name, we cannot figure out which blocks will
1403 * be affected in this operation. So, account for the worst case :
1404 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1405 * - 4 new blocks written if adding:
1406 * - 2 blocks for possibly split leaves,
1407 * - 2 grown ptrtbl blocks
1408 *
1409 * This also accomodates the case where an add operation to a fairly
1410 * large microzap results in a promotion to fatzap.
1411 */
1412 if (name == NULL) {
1413 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1414 return (err);
1415 }
1416
1417 /*
1418 * We lock the zap with adding == FALSE. Because, if we pass
1419 * the actual value of add, it could trigger a mzap_upgrade().
1420 * At present we are just evaluating the possibility of this operation
1421 * and hence we donot want to trigger an upgrade.
1422 */
1423 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1424 if (err)
1425 return (err);
1426
1427 if (!zap->zap_ismicro) {
1428 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1429 if (zn) {
1430 err = fzap_count_write(zn, add, towrite,
1431 tooverwrite);
1432 zap_name_free(zn);
1433 } else {
1434 /*
1435 * We treat this case as similar to (name == NULL)
1436 */
1437 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1438 }
1439 } else {
1440 /*
1441 * We are here if (name != NULL) and this is a micro-zap.
1442 * We account for the header block depending on whether it
1443 * is freeable.
1444 *
1445 * Incase of an add-operation it is hard to find out
1446 * if this add will promote this microzap to fatzap.
1447 * Hence, we consider the worst case and account for the
1448 * blocks assuming this microzap would be promoted to a
1449 * fatzap.
1450 *
1451 * 1 block overwritten : header block
1452 * 4 new blocks written : 2 new split leaf, 2 grown
1453 * ptrtbl blocks
1454 */
1455 if (dmu_buf_freeable(zap->zap_dbuf))
1456 *tooverwrite += SPA_MAXBLOCKSIZE;
1457 else
1458 *towrite += SPA_MAXBLOCKSIZE;
1459
1460 if (add) {
1461 *towrite += 4 * SPA_MAXBLOCKSIZE;
1462 }
1463 }
1464
1465 zap_unlockdir(zap);
1466 return (err);
1467}
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) 2012 by Delphix. All rights reserved.
24 */
25
26#include <sys/zio.h>
27#include <sys/spa.h>
28#include <sys/dmu.h>
29#include <sys/zfs_context.h>
30#include <sys/zap.h>
31#include <sys/refcount.h>
32#include <sys/zap_impl.h>
33#include <sys/zap_leaf.h>
34#include <sys/avl.h>
35#include <sys/arc.h>
36
37#ifdef _KERNEL
38#include <sys/sunddi.h>
39#endif
40
41static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags);
42
43uint64_t
44zap_getflags(zap_t *zap)
45{
46 if (zap->zap_ismicro)
47 return (0);
48 return (zap->zap_u.zap_fat.zap_phys->zap_flags);
49}
50
51int
52zap_hashbits(zap_t *zap)
53{
54 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
55 return (48);
56 else
57 return (28);
58}
59
60uint32_t
61zap_maxcd(zap_t *zap)
62{
63 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
64 return ((1<<16)-1);
65 else
66 return (-1U);
67}
68
69static uint64_t
70zap_hash(zap_name_t *zn)
71{
72 zap_t *zap = zn->zn_zap;
73 uint64_t h = 0;
74
75 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
76 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
77 h = *(uint64_t *)zn->zn_key_orig;
78 } else {
79 h = zap->zap_salt;
80 ASSERT(h != 0);
81 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
82
83 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
84 int i;
85 const uint64_t *wp = zn->zn_key_norm;
86
87 ASSERT(zn->zn_key_intlen == 8);
88 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
89 int j;
90 uint64_t word = *wp;
91
92 for (j = 0; j < zn->zn_key_intlen; j++) {
93 h = (h >> 8) ^
94 zfs_crc64_table[(h ^ word) & 0xFF];
95 word >>= NBBY;
96 }
97 }
98 } else {
99 int i, len;
100 const uint8_t *cp = zn->zn_key_norm;
101
102 /*
103 * We previously stored the terminating null on
104 * disk, but didn't hash it, so we need to
105 * continue to not hash it. (The
106 * zn_key_*_numints includes the terminating
107 * null for non-binary keys.)
108 */
109 len = zn->zn_key_norm_numints - 1;
110
111 ASSERT(zn->zn_key_intlen == 1);
112 for (i = 0; i < len; cp++, i++) {
113 h = (h >> 8) ^
114 zfs_crc64_table[(h ^ *cp) & 0xFF];
115 }
116 }
117 }
118 /*
119 * Don't use all 64 bits, since we need some in the cookie for
120 * the collision differentiator. We MUST use the high bits,
121 * since those are the ones that we first pay attention to when
122 * chosing the bucket.
123 */
124 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
125
126 return (h);
127}
128
129static int
130zap_normalize(zap_t *zap, const char *name, char *namenorm)
131{
132 size_t inlen, outlen;
133 int err;
134
135 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
136
137 inlen = strlen(name) + 1;
138 outlen = ZAP_MAXNAMELEN;
139
140 err = 0;
141 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
142 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
143 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
144
145 return (err);
146}
147
148boolean_t
149zap_match(zap_name_t *zn, const char *matchname)
150{
151 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
152
153 if (zn->zn_matchtype == MT_FIRST) {
154 char norm[ZAP_MAXNAMELEN];
155
156 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
157 return (B_FALSE);
158
159 return (strcmp(zn->zn_key_norm, norm) == 0);
160 } else {
161 /* MT_BEST or MT_EXACT */
162 return (strcmp(zn->zn_key_orig, matchname) == 0);
163 }
164}
165
166void
167zap_name_free(zap_name_t *zn)
168{
169 kmem_free(zn, sizeof (zap_name_t));
170}
171
172zap_name_t *
173zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
174{
175 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
176
177 zn->zn_zap = zap;
178 zn->zn_key_intlen = sizeof (*key);
179 zn->zn_key_orig = key;
180 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
181 zn->zn_matchtype = mt;
182 if (zap->zap_normflags) {
183 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
184 zap_name_free(zn);
185 return (NULL);
186 }
187 zn->zn_key_norm = zn->zn_normbuf;
188 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
189 } else {
190 if (mt != MT_EXACT) {
191 zap_name_free(zn);
192 return (NULL);
193 }
194 zn->zn_key_norm = zn->zn_key_orig;
195 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
196 }
197
198 zn->zn_hash = zap_hash(zn);
199 return (zn);
200}
201
202zap_name_t *
203zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
204{
205 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
206
207 ASSERT(zap->zap_normflags == 0);
208 zn->zn_zap = zap;
209 zn->zn_key_intlen = sizeof (*key);
210 zn->zn_key_orig = zn->zn_key_norm = key;
211 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
212 zn->zn_matchtype = MT_EXACT;
213
214 zn->zn_hash = zap_hash(zn);
215 return (zn);
216}
217
218static void
219mzap_byteswap(mzap_phys_t *buf, size_t size)
220{
221 int i, max;
222 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
223 buf->mz_salt = BSWAP_64(buf->mz_salt);
224 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
225 max = (size / MZAP_ENT_LEN) - 1;
226 for (i = 0; i < max; i++) {
227 buf->mz_chunk[i].mze_value =
228 BSWAP_64(buf->mz_chunk[i].mze_value);
229 buf->mz_chunk[i].mze_cd =
230 BSWAP_32(buf->mz_chunk[i].mze_cd);
231 }
232}
233
234void
235zap_byteswap(void *buf, size_t size)
236{
237 uint64_t block_type;
238
239 block_type = *(uint64_t *)buf;
240
241 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
242 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
243 mzap_byteswap(buf, size);
244 } else {
245 fzap_byteswap(buf, size);
246 }
247}
248
249static int
250mze_compare(const void *arg1, const void *arg2)
251{
252 const mzap_ent_t *mze1 = arg1;
253 const mzap_ent_t *mze2 = arg2;
254
255 if (mze1->mze_hash > mze2->mze_hash)
256 return (+1);
257 if (mze1->mze_hash < mze2->mze_hash)
258 return (-1);
259 if (mze1->mze_cd > mze2->mze_cd)
260 return (+1);
261 if (mze1->mze_cd < mze2->mze_cd)
262 return (-1);
263 return (0);
264}
265
266static int
267mze_insert(zap_t *zap, int chunkid, uint64_t hash)
268{
269 mzap_ent_t *mze;
270 avl_index_t idx;
271
272 ASSERT(zap->zap_ismicro);
273 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
274
275 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
276 mze->mze_chunkid = chunkid;
277 mze->mze_hash = hash;
278 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
279 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
280 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
281 kmem_free(mze, sizeof (mzap_ent_t));
282 return (EEXIST);
283 }
284 avl_insert(&zap->zap_m.zap_avl, mze, idx);
285 return (0);
286}
287
288static mzap_ent_t *
289mze_find(zap_name_t *zn)
290{
291 mzap_ent_t mze_tofind;
292 mzap_ent_t *mze;
293 avl_index_t idx;
294 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
295
296 ASSERT(zn->zn_zap->zap_ismicro);
297 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
298
299 mze_tofind.mze_hash = zn->zn_hash;
300 mze_tofind.mze_cd = 0;
301
302again:
303 mze = avl_find(avl, &mze_tofind, &idx);
304 if (mze == NULL)
305 mze = avl_nearest(avl, idx, AVL_AFTER);
306 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
307 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
308 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
309 return (mze);
310 }
311 if (zn->zn_matchtype == MT_BEST) {
312 zn->zn_matchtype = MT_FIRST;
313 goto again;
314 }
315 return (NULL);
316}
317
318static uint32_t
319mze_find_unused_cd(zap_t *zap, uint64_t hash)
320{
321 mzap_ent_t mze_tofind;
322 mzap_ent_t *mze;
323 avl_index_t idx;
324 avl_tree_t *avl = &zap->zap_m.zap_avl;
325 uint32_t cd;
326
327 ASSERT(zap->zap_ismicro);
328 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
329
330 mze_tofind.mze_hash = hash;
331 mze_tofind.mze_cd = 0;
332
333 cd = 0;
334 for (mze = avl_find(avl, &mze_tofind, &idx);
335 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
336 if (mze->mze_cd != cd)
337 break;
338 cd++;
339 }
340
341 return (cd);
342}
343
344static void
345mze_remove(zap_t *zap, mzap_ent_t *mze)
346{
347 ASSERT(zap->zap_ismicro);
348 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
349
350 avl_remove(&zap->zap_m.zap_avl, mze);
351 kmem_free(mze, sizeof (mzap_ent_t));
352}
353
354static void
355mze_destroy(zap_t *zap)
356{
357 mzap_ent_t *mze;
358 void *avlcookie = NULL;
359
360 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
361 kmem_free(mze, sizeof (mzap_ent_t));
362 avl_destroy(&zap->zap_m.zap_avl);
363}
364
365static zap_t *
366mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
367{
368 zap_t *winner;
369 zap_t *zap;
370 int i;
371
372 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
373
374 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
375 rw_init(&zap->zap_rwlock, 0, 0, 0);
376 rw_enter(&zap->zap_rwlock, RW_WRITER);
377 zap->zap_objset = os;
378 zap->zap_object = obj;
379 zap->zap_dbuf = db;
380
381 if (*(uint64_t *)db->db_data != ZBT_MICRO) {
382 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
383 zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
384 } else {
385 zap->zap_ismicro = TRUE;
386 }
387
388 /*
389 * Make sure that zap_ismicro is set before we let others see
390 * it, because zap_lockdir() checks zap_ismicro without the lock
391 * held.
392 */
393 winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
394
395 if (winner != NULL) {
396 rw_exit(&zap->zap_rwlock);
397 rw_destroy(&zap->zap_rwlock);
398 if (!zap->zap_ismicro)
399 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
400 kmem_free(zap, sizeof (zap_t));
401 return (winner);
402 }
403
404 if (zap->zap_ismicro) {
405 zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
406 zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
407 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
408 avl_create(&zap->zap_m.zap_avl, mze_compare,
409 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
410
411 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
412 mzap_ent_phys_t *mze =
413 &zap->zap_m.zap_phys->mz_chunk[i];
414 if (mze->mze_name[0]) {
415 zap_name_t *zn;
416
417 zn = zap_name_alloc(zap, mze->mze_name,
418 MT_EXACT);
419 if (mze_insert(zap, i, zn->zn_hash) == 0)
420 zap->zap_m.zap_num_entries++;
421 else {
422 printf("ZFS WARNING: Duplicated ZAP "
423 "entry detected (%s).\n",
424 mze->mze_name);
425 }
426 zap_name_free(zn);
427 }
428 }
429 } else {
430 zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
431 zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
432
433 ASSERT3U(sizeof (struct zap_leaf_header), ==,
434 2*ZAP_LEAF_CHUNKSIZE);
435
436 /*
437 * The embedded pointer table should not overlap the
438 * other members.
439 */
440 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
441 &zap->zap_f.zap_phys->zap_salt);
442
443 /*
444 * The embedded pointer table should end at the end of
445 * the block
446 */
447 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
448 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
449 (uintptr_t)zap->zap_f.zap_phys, ==,
450 zap->zap_dbuf->db_size);
451 }
452 rw_exit(&zap->zap_rwlock);
453 return (zap);
454}
455
456int
457zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
458 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
459{
460 zap_t *zap;
461 dmu_buf_t *db;
462 krw_t lt;
463 int err;
464
465 *zapp = NULL;
466
467 err = dmu_buf_hold(os, obj, 0, NULL, &db, DMU_READ_NO_PREFETCH);
468 if (err)
469 return (err);
470
471#ifdef ZFS_DEBUG
472 {
473 dmu_object_info_t doi;
474 dmu_object_info_from_db(db, &doi);
475 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
476 }
477#endif
478
479 zap = dmu_buf_get_user(db);
480 if (zap == NULL)
481 zap = mzap_open(os, obj, db);
482
483 /*
484 * We're checking zap_ismicro without the lock held, in order to
485 * tell what type of lock we want. Once we have some sort of
486 * lock, see if it really is the right type. In practice this
487 * can only be different if it was upgraded from micro to fat,
488 * and micro wanted WRITER but fat only needs READER.
489 */
490 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
491 rw_enter(&zap->zap_rwlock, lt);
492 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
493 /* it was upgraded, now we only need reader */
494 ASSERT(lt == RW_WRITER);
495 ASSERT(RW_READER ==
496 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
497 rw_downgrade(&zap->zap_rwlock);
498 lt = RW_READER;
499 }
500
501 zap->zap_objset = os;
502
503 if (lt == RW_WRITER)
504 dmu_buf_will_dirty(db, tx);
505
506 ASSERT3P(zap->zap_dbuf, ==, db);
507
508 ASSERT(!zap->zap_ismicro ||
509 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
510 if (zap->zap_ismicro && tx && adding &&
511 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
512 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
513 if (newsz > MZAP_MAX_BLKSZ) {
514 dprintf("upgrading obj %llu: num_entries=%u\n",
515 obj, zap->zap_m.zap_num_entries);
516 *zapp = zap;
517 return (mzap_upgrade(zapp, tx, 0));
518 }
519 err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
520 ASSERT0(err);
521 zap->zap_m.zap_num_chunks =
522 db->db_size / MZAP_ENT_LEN - 1;
523 }
524
525 *zapp = zap;
526 return (0);
527}
528
529void
530zap_unlockdir(zap_t *zap)
531{
532 rw_exit(&zap->zap_rwlock);
533 dmu_buf_rele(zap->zap_dbuf, NULL);
534}
535
536static int
537mzap_upgrade(zap_t **zapp, dmu_tx_t *tx, zap_flags_t flags)
538{
539 mzap_phys_t *mzp;
540 int i, sz, nchunks;
541 int err = 0;
542 zap_t *zap = *zapp;
543
544 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
545
546 sz = zap->zap_dbuf->db_size;
547 mzp = kmem_alloc(sz, KM_SLEEP);
548 bcopy(zap->zap_dbuf->db_data, mzp, sz);
549 nchunks = zap->zap_m.zap_num_chunks;
550
551 if (!flags) {
552 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
553 1ULL << fzap_default_block_shift, 0, tx);
554 if (err) {
555 kmem_free(mzp, sz);
556 return (err);
557 }
558 }
559
560 dprintf("upgrading obj=%llu with %u chunks\n",
561 zap->zap_object, nchunks);
562 /* XXX destroy the avl later, so we can use the stored hash value */
563 mze_destroy(zap);
564
565 fzap_upgrade(zap, tx, flags);
566
567 for (i = 0; i < nchunks; i++) {
568 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
569 zap_name_t *zn;
570 if (mze->mze_name[0] == 0)
571 continue;
572 dprintf("adding %s=%llu\n",
573 mze->mze_name, mze->mze_value);
574 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
575 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
576 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
577 zap_name_free(zn);
578 if (err)
579 break;
580 }
581 kmem_free(mzp, sz);
582 *zapp = zap;
583 return (err);
584}
585
586static void
587mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
588 dmu_tx_t *tx)
589{
590 dmu_buf_t *db;
591 mzap_phys_t *zp;
592
593 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
594
595#ifdef ZFS_DEBUG
596 {
597 dmu_object_info_t doi;
598 dmu_object_info_from_db(db, &doi);
599 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
600 }
601#endif
602
603 dmu_buf_will_dirty(db, tx);
604 zp = db->db_data;
605 zp->mz_block_type = ZBT_MICRO;
606 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
607 zp->mz_normflags = normflags;
608 dmu_buf_rele(db, FTAG);
609
610 if (flags != 0) {
611 zap_t *zap;
612 /* Only fat zap supports flags; upgrade immediately. */
613 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
614 B_FALSE, B_FALSE, &zap));
615 VERIFY3U(0, ==, mzap_upgrade(&zap, tx, flags));
616 zap_unlockdir(zap);
617 }
618}
619
620int
621zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
622 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
623{
624 return (zap_create_claim_norm(os, obj,
625 0, ot, bonustype, bonuslen, tx));
626}
627
628int
629zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
630 dmu_object_type_t ot,
631 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
632{
633 int err;
634
635 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
636 if (err != 0)
637 return (err);
638 mzap_create_impl(os, obj, normflags, 0, tx);
639 return (0);
640}
641
642uint64_t
643zap_create(objset_t *os, dmu_object_type_t ot,
644 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
645{
646 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
647}
648
649uint64_t
650zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
651 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
652{
653 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
654
655 mzap_create_impl(os, obj, normflags, 0, tx);
656 return (obj);
657}
658
659uint64_t
660zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
661 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
662 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
663{
664 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
665
666 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
667 leaf_blockshift <= SPA_MAXBLOCKSHIFT &&
668 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
669 indirect_blockshift <= SPA_MAXBLOCKSHIFT);
670
671 VERIFY(dmu_object_set_blocksize(os, obj,
672 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
673
674 mzap_create_impl(os, obj, normflags, flags, tx);
675 return (obj);
676}
677
678int
679zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
680{
681 /*
682 * dmu_object_free will free the object number and free the
683 * data. Freeing the data will cause our pageout function to be
684 * called, which will destroy our data (zap_leaf_t's and zap_t).
685 */
686
687 return (dmu_object_free(os, zapobj, tx));
688}
689
690_NOTE(ARGSUSED(0))
691void
692zap_evict(dmu_buf_t *db, void *vzap)
693{
694 zap_t *zap = vzap;
695
696 rw_destroy(&zap->zap_rwlock);
697
698 if (zap->zap_ismicro)
699 mze_destroy(zap);
700 else
701 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
702
703 kmem_free(zap, sizeof (zap_t));
704}
705
706int
707zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
708{
709 zap_t *zap;
710 int err;
711
712 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
713 if (err)
714 return (err);
715 if (!zap->zap_ismicro) {
716 err = fzap_count(zap, count);
717 } else {
718 *count = zap->zap_m.zap_num_entries;
719 }
720 zap_unlockdir(zap);
721 return (err);
722}
723
724/*
725 * zn may be NULL; if not specified, it will be computed if needed.
726 * See also the comment above zap_entry_normalization_conflict().
727 */
728static boolean_t
729mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
730{
731 mzap_ent_t *other;
732 int direction = AVL_BEFORE;
733 boolean_t allocdzn = B_FALSE;
734
735 if (zap->zap_normflags == 0)
736 return (B_FALSE);
737
738again:
739 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
740 other && other->mze_hash == mze->mze_hash;
741 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
742
743 if (zn == NULL) {
744 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
745 MT_FIRST);
746 allocdzn = B_TRUE;
747 }
748 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
749 if (allocdzn)
750 zap_name_free(zn);
751 return (B_TRUE);
752 }
753 }
754
755 if (direction == AVL_BEFORE) {
756 direction = AVL_AFTER;
757 goto again;
758 }
759
760 if (allocdzn)
761 zap_name_free(zn);
762 return (B_FALSE);
763}
764
765/*
766 * Routines for manipulating attributes.
767 */
768
769int
770zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
771 uint64_t integer_size, uint64_t num_integers, void *buf)
772{
773 return (zap_lookup_norm(os, zapobj, name, integer_size,
774 num_integers, buf, MT_EXACT, NULL, 0, NULL));
775}
776
777int
778zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
779 uint64_t integer_size, uint64_t num_integers, void *buf,
780 matchtype_t mt, char *realname, int rn_len,
781 boolean_t *ncp)
782{
783 zap_t *zap;
784 int err;
785 mzap_ent_t *mze;
786 zap_name_t *zn;
787
788 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
789 if (err)
790 return (err);
791 zn = zap_name_alloc(zap, name, mt);
792 if (zn == NULL) {
793 zap_unlockdir(zap);
794 return (ENOTSUP);
795 }
796
797 if (!zap->zap_ismicro) {
798 err = fzap_lookup(zn, integer_size, num_integers, buf,
799 realname, rn_len, ncp);
800 } else {
801 mze = mze_find(zn);
802 if (mze == NULL) {
803 err = ENOENT;
804 } else {
805 if (num_integers < 1) {
806 err = EOVERFLOW;
807 } else if (integer_size != 8) {
808 err = EINVAL;
809 } else {
810 *(uint64_t *)buf =
811 MZE_PHYS(zap, mze)->mze_value;
812 (void) strlcpy(realname,
813 MZE_PHYS(zap, mze)->mze_name, rn_len);
814 if (ncp) {
815 *ncp = mzap_normalization_conflict(zap,
816 zn, mze);
817 }
818 }
819 }
820 }
821 zap_name_free(zn);
822 zap_unlockdir(zap);
823 return (err);
824}
825
826int
827zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
828 int key_numints)
829{
830 zap_t *zap;
831 int err;
832 zap_name_t *zn;
833
834 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
835 if (err)
836 return (err);
837 zn = zap_name_alloc_uint64(zap, key, key_numints);
838 if (zn == NULL) {
839 zap_unlockdir(zap);
840 return (ENOTSUP);
841 }
842
843 fzap_prefetch(zn);
844 zap_name_free(zn);
845 zap_unlockdir(zap);
846 return (err);
847}
848
849int
850zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
851 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
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 (ENOTSUP);
864 }
865
866 err = fzap_lookup(zn, integer_size, num_integers, buf,
867 NULL, 0, NULL);
868 zap_name_free(zn);
869 zap_unlockdir(zap);
870 return (err);
871}
872
873int
874zap_contains(objset_t *os, uint64_t zapobj, const char *name)
875{
876 int err = (zap_lookup_norm(os, zapobj, name, 0,
877 0, NULL, MT_EXACT, NULL, 0, NULL));
878 if (err == EOVERFLOW || err == EINVAL)
879 err = 0; /* found, but skipped reading the value */
880 return (err);
881}
882
883int
884zap_length(objset_t *os, uint64_t zapobj, const char *name,
885 uint64_t *integer_size, uint64_t *num_integers)
886{
887 zap_t *zap;
888 int err;
889 mzap_ent_t *mze;
890 zap_name_t *zn;
891
892 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
893 if (err)
894 return (err);
895 zn = zap_name_alloc(zap, name, MT_EXACT);
896 if (zn == NULL) {
897 zap_unlockdir(zap);
898 return (ENOTSUP);
899 }
900 if (!zap->zap_ismicro) {
901 err = fzap_length(zn, integer_size, num_integers);
902 } else {
903 mze = mze_find(zn);
904 if (mze == NULL) {
905 err = ENOENT;
906 } else {
907 if (integer_size)
908 *integer_size = 8;
909 if (num_integers)
910 *num_integers = 1;
911 }
912 }
913 zap_name_free(zn);
914 zap_unlockdir(zap);
915 return (err);
916}
917
918int
919zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
920 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
921{
922 zap_t *zap;
923 int err;
924 zap_name_t *zn;
925
926 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
927 if (err)
928 return (err);
929 zn = zap_name_alloc_uint64(zap, key, key_numints);
930 if (zn == NULL) {
931 zap_unlockdir(zap);
932 return (ENOTSUP);
933 }
934 err = fzap_length(zn, integer_size, num_integers);
935 zap_name_free(zn);
936 zap_unlockdir(zap);
937 return (err);
938}
939
940static void
941mzap_addent(zap_name_t *zn, uint64_t value)
942{
943 int i;
944 zap_t *zap = zn->zn_zap;
945 int start = zap->zap_m.zap_alloc_next;
946 uint32_t cd;
947
948 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
949
950#ifdef ZFS_DEBUG
951 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
952 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
953 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
954 }
955#endif
956
957 cd = mze_find_unused_cd(zap, zn->zn_hash);
958 /* given the limited size of the microzap, this can't happen */
959 ASSERT(cd < zap_maxcd(zap));
960
961again:
962 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
963 mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
964 if (mze->mze_name[0] == 0) {
965 mze->mze_value = value;
966 mze->mze_cd = cd;
967 (void) strcpy(mze->mze_name, zn->zn_key_orig);
968 zap->zap_m.zap_num_entries++;
969 zap->zap_m.zap_alloc_next = i+1;
970 if (zap->zap_m.zap_alloc_next ==
971 zap->zap_m.zap_num_chunks)
972 zap->zap_m.zap_alloc_next = 0;
973 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
974 return;
975 }
976 }
977 if (start != 0) {
978 start = 0;
979 goto again;
980 }
981 ASSERT(!"out of entries!");
982}
983
984int
985zap_add(objset_t *os, uint64_t zapobj, const char *key,
986 int integer_size, uint64_t num_integers,
987 const void *val, dmu_tx_t *tx)
988{
989 zap_t *zap;
990 int err;
991 mzap_ent_t *mze;
992 const uint64_t *intval = val;
993 zap_name_t *zn;
994
995 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
996 if (err)
997 return (err);
998 zn = zap_name_alloc(zap, key, MT_EXACT);
999 if (zn == NULL) {
1000 zap_unlockdir(zap);
1001 return (ENOTSUP);
1002 }
1003 if (!zap->zap_ismicro) {
1004 err = fzap_add(zn, integer_size, num_integers, val, tx);
1005 zap = zn->zn_zap; /* fzap_add() may change zap */
1006 } else if (integer_size != 8 || num_integers != 1 ||
1007 strlen(key) >= MZAP_NAME_LEN) {
1008 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1009 if (err == 0)
1010 err = fzap_add(zn, integer_size, num_integers, val, tx);
1011 zap = zn->zn_zap; /* fzap_add() may change zap */
1012 } else {
1013 mze = mze_find(zn);
1014 if (mze != NULL) {
1015 err = EEXIST;
1016 } else {
1017 mzap_addent(zn, *intval);
1018 }
1019 }
1020 ASSERT(zap == zn->zn_zap);
1021 zap_name_free(zn);
1022 if (zap != NULL) /* may be NULL if fzap_add() failed */
1023 zap_unlockdir(zap);
1024 return (err);
1025}
1026
1027int
1028zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1029 int key_numints, int integer_size, uint64_t num_integers,
1030 const void *val, dmu_tx_t *tx)
1031{
1032 zap_t *zap;
1033 int err;
1034 zap_name_t *zn;
1035
1036 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1037 if (err)
1038 return (err);
1039 zn = zap_name_alloc_uint64(zap, key, key_numints);
1040 if (zn == NULL) {
1041 zap_unlockdir(zap);
1042 return (ENOTSUP);
1043 }
1044 err = fzap_add(zn, integer_size, num_integers, val, tx);
1045 zap = zn->zn_zap; /* fzap_add() may change zap */
1046 zap_name_free(zn);
1047 if (zap != NULL) /* may be NULL if fzap_add() failed */
1048 zap_unlockdir(zap);
1049 return (err);
1050}
1051
1052int
1053zap_update(objset_t *os, uint64_t zapobj, const char *name,
1054 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1055{
1056 zap_t *zap;
1057 mzap_ent_t *mze;
1058 uint64_t oldval;
1059 const uint64_t *intval = val;
1060 zap_name_t *zn;
1061 int err;
1062
1063#ifdef ZFS_DEBUG
1064 /*
1065 * If there is an old value, it shouldn't change across the
1066 * lockdir (eg, due to bprewrite's xlation).
1067 */
1068 if (integer_size == 8 && num_integers == 1)
1069 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1070#endif
1071
1072 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1073 if (err)
1074 return (err);
1075 zn = zap_name_alloc(zap, name, MT_EXACT);
1076 if (zn == NULL) {
1077 zap_unlockdir(zap);
1078 return (ENOTSUP);
1079 }
1080 if (!zap->zap_ismicro) {
1081 err = fzap_update(zn, integer_size, num_integers, val, tx);
1082 zap = zn->zn_zap; /* fzap_update() may change zap */
1083 } else if (integer_size != 8 || num_integers != 1 ||
1084 strlen(name) >= MZAP_NAME_LEN) {
1085 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1086 zapobj, integer_size, num_integers, name);
1087 err = mzap_upgrade(&zn->zn_zap, tx, 0);
1088 if (err == 0)
1089 err = fzap_update(zn, integer_size, num_integers,
1090 val, tx);
1091 zap = zn->zn_zap; /* fzap_update() may change zap */
1092 } else {
1093 mze = mze_find(zn);
1094 if (mze != NULL) {
1095 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1096 MZE_PHYS(zap, mze)->mze_value = *intval;
1097 } else {
1098 mzap_addent(zn, *intval);
1099 }
1100 }
1101 ASSERT(zap == zn->zn_zap);
1102 zap_name_free(zn);
1103 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1104 zap_unlockdir(zap);
1105 return (err);
1106}
1107
1108int
1109zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1110 int key_numints,
1111 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1112{
1113 zap_t *zap;
1114 zap_name_t *zn;
1115 int err;
1116
1117 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
1118 if (err)
1119 return (err);
1120 zn = zap_name_alloc_uint64(zap, key, key_numints);
1121 if (zn == NULL) {
1122 zap_unlockdir(zap);
1123 return (ENOTSUP);
1124 }
1125 err = fzap_update(zn, integer_size, num_integers, val, tx);
1126 zap = zn->zn_zap; /* fzap_update() may change zap */
1127 zap_name_free(zn);
1128 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1129 zap_unlockdir(zap);
1130 return (err);
1131}
1132
1133int
1134zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1135{
1136 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1137}
1138
1139int
1140zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1141 matchtype_t mt, dmu_tx_t *tx)
1142{
1143 zap_t *zap;
1144 int err;
1145 mzap_ent_t *mze;
1146 zap_name_t *zn;
1147
1148 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1149 if (err)
1150 return (err);
1151 zn = zap_name_alloc(zap, name, mt);
1152 if (zn == NULL) {
1153 zap_unlockdir(zap);
1154 return (ENOTSUP);
1155 }
1156 if (!zap->zap_ismicro) {
1157 err = fzap_remove(zn, tx);
1158 } else {
1159 mze = mze_find(zn);
1160 if (mze == NULL) {
1161 err = ENOENT;
1162 } else {
1163 zap->zap_m.zap_num_entries--;
1164 bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
1165 sizeof (mzap_ent_phys_t));
1166 mze_remove(zap, mze);
1167 }
1168 }
1169 zap_name_free(zn);
1170 zap_unlockdir(zap);
1171 return (err);
1172}
1173
1174int
1175zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1176 int key_numints, dmu_tx_t *tx)
1177{
1178 zap_t *zap;
1179 int err;
1180 zap_name_t *zn;
1181
1182 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
1183 if (err)
1184 return (err);
1185 zn = zap_name_alloc_uint64(zap, key, key_numints);
1186 if (zn == NULL) {
1187 zap_unlockdir(zap);
1188 return (ENOTSUP);
1189 }
1190 err = fzap_remove(zn, tx);
1191 zap_name_free(zn);
1192 zap_unlockdir(zap);
1193 return (err);
1194}
1195
1196/*
1197 * Routines for iterating over the attributes.
1198 */
1199
1200void
1201zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1202 uint64_t serialized)
1203{
1204 zc->zc_objset = os;
1205 zc->zc_zap = NULL;
1206 zc->zc_leaf = NULL;
1207 zc->zc_zapobj = zapobj;
1208 zc->zc_serialized = serialized;
1209 zc->zc_hash = 0;
1210 zc->zc_cd = 0;
1211}
1212
1213void
1214zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1215{
1216 zap_cursor_init_serialized(zc, os, zapobj, 0);
1217}
1218
1219void
1220zap_cursor_fini(zap_cursor_t *zc)
1221{
1222 if (zc->zc_zap) {
1223 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1224 zap_unlockdir(zc->zc_zap);
1225 zc->zc_zap = NULL;
1226 }
1227 if (zc->zc_leaf) {
1228 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1229 zap_put_leaf(zc->zc_leaf);
1230 zc->zc_leaf = NULL;
1231 }
1232 zc->zc_objset = NULL;
1233}
1234
1235uint64_t
1236zap_cursor_serialize(zap_cursor_t *zc)
1237{
1238 if (zc->zc_hash == -1ULL)
1239 return (-1ULL);
1240 if (zc->zc_zap == NULL)
1241 return (zc->zc_serialized);
1242 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1243 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1244
1245 /*
1246 * We want to keep the high 32 bits of the cursor zero if we can, so
1247 * that 32-bit programs can access this. So usually use a small
1248 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1249 * of the cursor.
1250 *
1251 * [ collision differentiator | zap_hashbits()-bit hash value ]
1252 */
1253 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1254 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1255}
1256
1257int
1258zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1259{
1260 int err;
1261 avl_index_t idx;
1262 mzap_ent_t mze_tofind;
1263 mzap_ent_t *mze;
1264
1265 if (zc->zc_hash == -1ULL)
1266 return (ENOENT);
1267
1268 if (zc->zc_zap == NULL) {
1269 int hb;
1270 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1271 RW_READER, TRUE, FALSE, &zc->zc_zap);
1272 if (err)
1273 return (err);
1274
1275 /*
1276 * To support zap_cursor_init_serialized, advance, retrieve,
1277 * we must add to the existing zc_cd, which may already
1278 * be 1 due to the zap_cursor_advance.
1279 */
1280 ASSERT(zc->zc_hash == 0);
1281 hb = zap_hashbits(zc->zc_zap);
1282 zc->zc_hash = zc->zc_serialized << (64 - hb);
1283 zc->zc_cd += zc->zc_serialized >> hb;
1284 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1285 zc->zc_cd = 0;
1286 } else {
1287 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1288 }
1289 if (!zc->zc_zap->zap_ismicro) {
1290 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1291 } else {
1292 err = ENOENT;
1293
1294 mze_tofind.mze_hash = zc->zc_hash;
1295 mze_tofind.mze_cd = zc->zc_cd;
1296
1297 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1298 if (mze == NULL) {
1299 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1300 idx, AVL_AFTER);
1301 }
1302 if (mze) {
1303 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1304 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1305 za->za_normalization_conflict =
1306 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1307 za->za_integer_length = 8;
1308 za->za_num_integers = 1;
1309 za->za_first_integer = mzep->mze_value;
1310 (void) strcpy(za->za_name, mzep->mze_name);
1311 zc->zc_hash = mze->mze_hash;
1312 zc->zc_cd = mze->mze_cd;
1313 err = 0;
1314 } else {
1315 zc->zc_hash = -1ULL;
1316 }
1317 }
1318 rw_exit(&zc->zc_zap->zap_rwlock);
1319 return (err);
1320}
1321
1322void
1323zap_cursor_advance(zap_cursor_t *zc)
1324{
1325 if (zc->zc_hash == -1ULL)
1326 return;
1327 zc->zc_cd++;
1328}
1329
1330int
1331zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1332{
1333 int err = 0;
1334 mzap_ent_t *mze;
1335 zap_name_t *zn;
1336
1337 if (zc->zc_zap == NULL) {
1338 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1339 RW_READER, TRUE, FALSE, &zc->zc_zap);
1340 if (err)
1341 return (err);
1342 } else {
1343 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1344 }
1345
1346 zn = zap_name_alloc(zc->zc_zap, name, mt);
1347 if (zn == NULL) {
1348 rw_exit(&zc->zc_zap->zap_rwlock);
1349 return (ENOTSUP);
1350 }
1351
1352 if (!zc->zc_zap->zap_ismicro) {
1353 err = fzap_cursor_move_to_key(zc, zn);
1354 } else {
1355 mze = mze_find(zn);
1356 if (mze == NULL) {
1357 err = ENOENT;
1358 goto out;
1359 }
1360 zc->zc_hash = mze->mze_hash;
1361 zc->zc_cd = mze->mze_cd;
1362 }
1363
1364out:
1365 zap_name_free(zn);
1366 rw_exit(&zc->zc_zap->zap_rwlock);
1367 return (err);
1368}
1369
1370int
1371zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1372{
1373 int err;
1374 zap_t *zap;
1375
1376 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1377 if (err)
1378 return (err);
1379
1380 bzero(zs, sizeof (zap_stats_t));
1381
1382 if (zap->zap_ismicro) {
1383 zs->zs_blocksize = zap->zap_dbuf->db_size;
1384 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1385 zs->zs_num_blocks = 1;
1386 } else {
1387 fzap_get_stats(zap, zs);
1388 }
1389 zap_unlockdir(zap);
1390 return (0);
1391}
1392
1393int
1394zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
1395 uint64_t *towrite, uint64_t *tooverwrite)
1396{
1397 zap_t *zap;
1398 int err = 0;
1399
1400
1401 /*
1402 * Since, we don't have a name, we cannot figure out which blocks will
1403 * be affected in this operation. So, account for the worst case :
1404 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1405 * - 4 new blocks written if adding:
1406 * - 2 blocks for possibly split leaves,
1407 * - 2 grown ptrtbl blocks
1408 *
1409 * This also accomodates the case where an add operation to a fairly
1410 * large microzap results in a promotion to fatzap.
1411 */
1412 if (name == NULL) {
1413 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1414 return (err);
1415 }
1416
1417 /*
1418 * We lock the zap with adding == FALSE. Because, if we pass
1419 * the actual value of add, it could trigger a mzap_upgrade().
1420 * At present we are just evaluating the possibility of this operation
1421 * and hence we donot want to trigger an upgrade.
1422 */
1423 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1424 if (err)
1425 return (err);
1426
1427 if (!zap->zap_ismicro) {
1428 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1429 if (zn) {
1430 err = fzap_count_write(zn, add, towrite,
1431 tooverwrite);
1432 zap_name_free(zn);
1433 } else {
1434 /*
1435 * We treat this case as similar to (name == NULL)
1436 */
1437 *towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
1438 }
1439 } else {
1440 /*
1441 * We are here if (name != NULL) and this is a micro-zap.
1442 * We account for the header block depending on whether it
1443 * is freeable.
1444 *
1445 * Incase of an add-operation it is hard to find out
1446 * if this add will promote this microzap to fatzap.
1447 * Hence, we consider the worst case and account for the
1448 * blocks assuming this microzap would be promoted to a
1449 * fatzap.
1450 *
1451 * 1 block overwritten : header block
1452 * 4 new blocks written : 2 new split leaf, 2 grown
1453 * ptrtbl blocks
1454 */
1455 if (dmu_buf_freeable(zap->zap_dbuf))
1456 *tooverwrite += SPA_MAXBLOCKSIZE;
1457 else
1458 *towrite += SPA_MAXBLOCKSIZE;
1459
1460 if (add) {
1461 *towrite += 4 * SPA_MAXBLOCKSIZE;
1462 }
1463 }
1464
1465 zap_unlockdir(zap);
1466 return (err);
1467}