?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ZIL: Call brt_pending_add() replaying TX_CLONE_RANGE
[zfs.git] / module / zfs / zap_micro.c
blob085d9cd8b4b6448c31a4e43e71a03a40b6e97c44
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 https://opensource.org/licenses/CDDL-1.0.
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 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 * Copyright 2017 Nexenta Systems, Inc.
27 */
28
29 #include <sys/zio.h>
30 #include <sys/spa.h>
31 #include <sys/dmu.h>
32 #include <sys/zfs_context.h>
33 #include <sys/zap.h>
34 #include <sys/zap_impl.h>
35 #include <sys/zap_leaf.h>
36 #include <sys/btree.h>
37 #include <sys/arc.h>
38 #include <sys/dmu_objset.h>
39
40 #ifdef _KERNEL
41 #include <sys/sunddi.h>
42 #endif
43
44 int zap_micro_max_size = MZAP_MAX_BLKSZ;
45
46 static int mzap_upgrade(zap_t **zapp,
47 const void *tag, dmu_tx_t *tx, zap_flags_t flags);
48
49 uint64_t
50 zap_getflags(zap_t *zap)
51 {
52 if (zap->zap_ismicro)
53 return (0);
54 return (zap_f_phys(zap)->zap_flags);
55 }
56
57 int
58 zap_hashbits(zap_t *zap)
59 {
60 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
61 return (48);
62 else
63 return (28);
64 }
65
66 uint32_t
67 zap_maxcd(zap_t *zap)
68 {
69 if (zap_getflags(zap) & ZAP_FLAG_HASH64)
70 return ((1<<16)-1);
71 else
72 return (-1U);
73 }
74
75 static uint64_t
76 zap_hash(zap_name_t *zn)
77 {
78 zap_t *zap = zn->zn_zap;
79 uint64_t h = 0;
80
81 if (zap_getflags(zap) & ZAP_FLAG_PRE_HASHED_KEY) {
82 ASSERT(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY);
83 h = *(uint64_t *)zn->zn_key_orig;
84 } else {
85 h = zap->zap_salt;
86 ASSERT(h != 0);
87 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
88
89 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
90 const uint64_t *wp = zn->zn_key_norm;
91
92 ASSERT(zn->zn_key_intlen == 8);
93 for (int i = 0; i < zn->zn_key_norm_numints;
94 wp++, i++) {
95 uint64_t word = *wp;
96
97 for (int j = 0; j < 8; j++) {
98 h = (h >> 8) ^
99 zfs_crc64_table[(h ^ word) & 0xFF];
100 word >>= NBBY;
101 }
102 }
103 } else {
104 const uint8_t *cp = zn->zn_key_norm;
105
106 /*
107 * We previously stored the terminating null on
108 * disk, but didn't hash it, so we need to
109 * continue to not hash it. (The
110 * zn_key_*_numints includes the terminating
111 * null for non-binary keys.)
112 */
113 int len = zn->zn_key_norm_numints - 1;
114
115 ASSERT(zn->zn_key_intlen == 1);
116 for (int i = 0; i < len; cp++, i++) {
117 h = (h >> 8) ^
118 zfs_crc64_table[(h ^ *cp) & 0xFF];
119 }
120 }
121 }
122 /*
123 * Don't use all 64 bits, since we need some in the cookie for
124 * the collision differentiator. We MUST use the high bits,
125 * since those are the ones that we first pay attention to when
126 * choosing the bucket.
127 */
128 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
129
130 return (h);
131 }
132
133 static int
134 zap_normalize(zap_t *zap, const char *name, char *namenorm, int normflags)
135 {
136 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
137
138 size_t inlen = strlen(name) + 1;
139 size_t outlen = ZAP_MAXNAMELEN;
140
141 int err = 0;
142 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
143 normflags | U8_TEXTPREP_IGNORE_NULL | U8_TEXTPREP_IGNORE_INVALID,
144 U8_UNICODE_LATEST, &err);
145
146 return (err);
147 }
148
149 boolean_t
150 zap_match(zap_name_t *zn, const char *matchname)
151 {
152 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
153
154 if (zn->zn_matchtype & MT_NORMALIZE) {
155 char norm[ZAP_MAXNAMELEN];
156
157 if (zap_normalize(zn->zn_zap, matchname, norm,
158 zn->zn_normflags) != 0)
159 return (B_FALSE);
160
161 return (strcmp(zn->zn_key_norm, norm) == 0);
162 } else {
163 return (strcmp(zn->zn_key_orig, matchname) == 0);
164 }
165 }
166
167 static zap_name_t *
168 zap_name_alloc(zap_t *zap)
169 {
170 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
171 zn->zn_zap = zap;
172 return (zn);
173 }
174
175 void
176 zap_name_free(zap_name_t *zn)
177 {
178 kmem_free(zn, sizeof (zap_name_t));
179 }
180
181 static int
182 zap_name_init_str(zap_name_t *zn, const char *key, matchtype_t mt)
183 {
184 zap_t *zap = zn->zn_zap;
185
186 zn->zn_key_intlen = sizeof (*key);
187 zn->zn_key_orig = key;
188 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
189 zn->zn_matchtype = mt;
190 zn->zn_normflags = zap->zap_normflags;
191
192 /*
193 * If we're dealing with a case sensitive lookup on a mixed or
194 * insensitive fs, remove U8_TEXTPREP_TOUPPER or the lookup
195 * will fold case to all caps overriding the lookup request.
196 */
197 if (mt & MT_MATCH_CASE)
198 zn->zn_normflags &= ~U8_TEXTPREP_TOUPPER;
199
200 if (zap->zap_normflags) {
201 /*
202 * We *must* use zap_normflags because this normalization is
203 * what the hash is computed from.
204 */
205 if (zap_normalize(zap, key, zn->zn_normbuf,
206 zap->zap_normflags) != 0)
207 return (SET_ERROR(ENOTSUP));
208 zn->zn_key_norm = zn->zn_normbuf;
209 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
210 } else {
211 if (mt != 0)
212 return (SET_ERROR(ENOTSUP));
213 zn->zn_key_norm = zn->zn_key_orig;
214 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
215 }
216
217 zn->zn_hash = zap_hash(zn);
218
219 if (zap->zap_normflags != zn->zn_normflags) {
220 /*
221 * We *must* use zn_normflags because this normalization is
222 * what the matching is based on. (Not the hash!)
223 */
224 if (zap_normalize(zap, key, zn->zn_normbuf,
225 zn->zn_normflags) != 0)
226 return (SET_ERROR(ENOTSUP));
227 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
228 }
229
230 return (0);
231 }
232
233 zap_name_t *
234 zap_name_alloc_str(zap_t *zap, const char *key, matchtype_t mt)
235 {
236 zap_name_t *zn = zap_name_alloc(zap);
237 if (zap_name_init_str(zn, key, mt) != 0) {
238 zap_name_free(zn);
239 return (NULL);
240 }
241 return (zn);
242 }
243
244 static zap_name_t *
245 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
246 {
247 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
248
249 ASSERT(zap->zap_normflags == 0);
250 zn->zn_zap = zap;
251 zn->zn_key_intlen = sizeof (*key);
252 zn->zn_key_orig = zn->zn_key_norm = key;
253 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
254 zn->zn_matchtype = 0;
255
256 zn->zn_hash = zap_hash(zn);
257 return (zn);
258 }
259
260 static void
261 mzap_byteswap(mzap_phys_t *buf, size_t size)
262 {
263 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
264 buf->mz_salt = BSWAP_64(buf->mz_salt);
265 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
266 int max = (size / MZAP_ENT_LEN) - 1;
267 for (int i = 0; i < max; i++) {
268 buf->mz_chunk[i].mze_value =
269 BSWAP_64(buf->mz_chunk[i].mze_value);
270 buf->mz_chunk[i].mze_cd =
271 BSWAP_32(buf->mz_chunk[i].mze_cd);
272 }
273 }
274
275 void
276 zap_byteswap(void *buf, size_t size)
277 {
278 uint64_t block_type = *(uint64_t *)buf;
279
280 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
281 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
282 mzap_byteswap(buf, size);
283 } else {
284 fzap_byteswap(buf, size);
285 }
286 }
287
288 __attribute__((always_inline)) inline
289 static int
290 mze_compare(const void *arg1, const void *arg2)
291 {
292 const mzap_ent_t *mze1 = arg1;
293 const mzap_ent_t *mze2 = arg2;
294
295 return (TREE_CMP((uint64_t)(mze1->mze_hash) << 32 | mze1->mze_cd,
296 (uint64_t)(mze2->mze_hash) << 32 | mze2->mze_cd));
297 }
298
299 ZFS_BTREE_FIND_IN_BUF_FUNC(mze_find_in_buf, mzap_ent_t,
300 mze_compare)
301
302 static void
303 mze_insert(zap_t *zap, uint16_t chunkid, uint64_t hash)
304 {
305 mzap_ent_t mze;
306
307 ASSERT(zap->zap_ismicro);
308 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
309
310 mze.mze_chunkid = chunkid;
311 ASSERT0(hash & 0xffffffff);
312 mze.mze_hash = hash >> 32;
313 ASSERT3U(MZE_PHYS(zap, &mze)->mze_cd, <=, 0xffff);
314 mze.mze_cd = (uint16_t)MZE_PHYS(zap, &mze)->mze_cd;
315 ASSERT(MZE_PHYS(zap, &mze)->mze_name[0] != 0);
316 zfs_btree_add(&zap->zap_m.zap_tree, &mze);
317 }
318
319 static mzap_ent_t *
320 mze_find(zap_name_t *zn, zfs_btree_index_t *idx)
321 {
322 mzap_ent_t mze_tofind;
323 mzap_ent_t *mze;
324 zfs_btree_t *tree = &zn->zn_zap->zap_m.zap_tree;
325
326 ASSERT(zn->zn_zap->zap_ismicro);
327 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
328
329 ASSERT0(zn->zn_hash & 0xffffffff);
330 mze_tofind.mze_hash = zn->zn_hash >> 32;
331 mze_tofind.mze_cd = 0;
332
333 mze = zfs_btree_find(tree, &mze_tofind, idx);
334 if (mze == NULL)
335 mze = zfs_btree_next(tree, idx, idx);
336 for (; mze && mze->mze_hash == mze_tofind.mze_hash;
337 mze = zfs_btree_next(tree, idx, idx)) {
338 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
339 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
340 return (mze);
341 }
342
343 return (NULL);
344 }
345
346 static uint32_t
347 mze_find_unused_cd(zap_t *zap, uint64_t hash)
348 {
349 mzap_ent_t mze_tofind;
350 zfs_btree_index_t idx;
351 zfs_btree_t *tree = &zap->zap_m.zap_tree;
352
353 ASSERT(zap->zap_ismicro);
354 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
355
356 ASSERT0(hash & 0xffffffff);
357 hash >>= 32;
358 mze_tofind.mze_hash = hash;
359 mze_tofind.mze_cd = 0;
360
361 uint32_t cd = 0;
362 for (mzap_ent_t *mze = zfs_btree_find(tree, &mze_tofind, &idx);
363 mze && mze->mze_hash == hash;
364 mze = zfs_btree_next(tree, &idx, &idx)) {
365 if (mze->mze_cd != cd)
366 break;
367 cd++;
368 }
369
370 return (cd);
371 }
372
373 /*
374 * Each mzap entry requires at max : 4 chunks
375 * 3 chunks for names + 1 chunk for value.
376 */
377 #define MZAP_ENT_CHUNKS (1 + ZAP_LEAF_ARRAY_NCHUNKS(MZAP_NAME_LEN) + \
378 ZAP_LEAF_ARRAY_NCHUNKS(sizeof (uint64_t)))
379
380 /*
381 * Check if the current entry keeps the colliding entries under the fatzap leaf
382 * size.
383 */
384 static boolean_t
385 mze_canfit_fzap_leaf(zap_name_t *zn, uint64_t hash)
386 {
387 zap_t *zap = zn->zn_zap;
388 mzap_ent_t mze_tofind;
389 zfs_btree_index_t idx;
390 zfs_btree_t *tree = &zap->zap_m.zap_tree;
391 uint32_t mzap_ents = 0;
392
393 ASSERT0(hash & 0xffffffff);
394 hash >>= 32;
395 mze_tofind.mze_hash = hash;
396 mze_tofind.mze_cd = 0;
397
398 for (mzap_ent_t *mze = zfs_btree_find(tree, &mze_tofind, &idx);
399 mze && mze->mze_hash == hash;
400 mze = zfs_btree_next(tree, &idx, &idx)) {
401 mzap_ents++;
402 }
403
404 /* Include the new entry being added */
405 mzap_ents++;
406
407 return (ZAP_LEAF_NUMCHUNKS_DEF > (mzap_ents * MZAP_ENT_CHUNKS));
408 }
409
410 static void
411 mze_destroy(zap_t *zap)
412 {
413 zfs_btree_clear(&zap->zap_m.zap_tree);
414 zfs_btree_destroy(&zap->zap_m.zap_tree);
415 }
416
417 static zap_t *
418 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
419 {
420 zap_t *winner;
421 uint64_t *zap_hdr = (uint64_t *)db->db_data;
422 uint64_t zap_block_type = zap_hdr[0];
423 uint64_t zap_magic = zap_hdr[1];
424
425 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
426
427 zap_t *zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
428 rw_init(&zap->zap_rwlock, NULL, RW_DEFAULT, NULL);
429 rw_enter(&zap->zap_rwlock, RW_WRITER);
430 zap->zap_objset = os;
431 zap->zap_object = obj;
432 zap->zap_dbuf = db;
433
434 if (zap_block_type != ZBT_MICRO) {
435 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, MUTEX_DEFAULT,
436 0);
437 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
438 if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
439 winner = NULL; /* No actual winner here... */
440 goto handle_winner;
441 }
442 } else {
443 zap->zap_ismicro = TRUE;
444 }
445
446 /*
447 * Make sure that zap_ismicro is set before we let others see
448 * it, because zap_lockdir() checks zap_ismicro without the lock
449 * held.
450 */
451 dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
452 winner = dmu_buf_set_user(db, &zap->zap_dbu);
453
454 if (winner != NULL)
455 goto handle_winner;
456
457 if (zap->zap_ismicro) {
458 zap->zap_salt = zap_m_phys(zap)->mz_salt;
459 zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
460 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
461
462 /*
463 * Reduce B-tree leaf from 4KB to 512 bytes to reduce memmove()
464 * overhead on massive inserts below. It still allows to store
465 * 62 entries before we have to add 2KB B-tree core node.
466 */
467 zfs_btree_create_custom(&zap->zap_m.zap_tree, mze_compare,
468 mze_find_in_buf, sizeof (mzap_ent_t), 512);
469
470 zap_name_t *zn = zap_name_alloc(zap);
471 for (uint16_t i = 0; i < zap->zap_m.zap_num_chunks; i++) {
472 mzap_ent_phys_t *mze =
473 &zap_m_phys(zap)->mz_chunk[i];
474 if (mze->mze_name[0]) {
475 zap->zap_m.zap_num_entries++;
476 zap_name_init_str(zn, mze->mze_name, 0);
477 mze_insert(zap, i, zn->zn_hash);
478 }
479 }
480 zap_name_free(zn);
481 } else {
482 zap->zap_salt = zap_f_phys(zap)->zap_salt;
483 zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
484
485 ASSERT3U(sizeof (struct zap_leaf_header), ==,
486 2*ZAP_LEAF_CHUNKSIZE);
487
488 /*
489 * The embedded pointer table should not overlap the
490 * other members.
491 */
492 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
493 &zap_f_phys(zap)->zap_salt);
494
495 /*
496 * The embedded pointer table should end at the end of
497 * the block
498 */
499 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
500 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
501 (uintptr_t)zap_f_phys(zap), ==,
502 zap->zap_dbuf->db_size);
503 }
504 rw_exit(&zap->zap_rwlock);
505 return (zap);
506
507 handle_winner:
508 rw_exit(&zap->zap_rwlock);
509 rw_destroy(&zap->zap_rwlock);
510 if (!zap->zap_ismicro)
511 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
512 kmem_free(zap, sizeof (zap_t));
513 return (winner);
514 }
515
516 /*
517 * This routine "consumes" the caller's hold on the dbuf, which must
518 * have the specified tag.
519 */
520 static int
521 zap_lockdir_impl(dmu_buf_t *db, const void *tag, dmu_tx_t *tx,
522 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
523 {
524 ASSERT0(db->db_offset);
525 objset_t *os = dmu_buf_get_objset(db);
526 uint64_t obj = db->db_object;
527 dmu_object_info_t doi;
528
529 *zapp = NULL;
530
531 dmu_object_info_from_db(db, &doi);
532 if (DMU_OT_BYTESWAP(doi.doi_type) != DMU_BSWAP_ZAP)
533 return (SET_ERROR(EINVAL));
534
535 zap_t *zap = dmu_buf_get_user(db);
536 if (zap == NULL) {
537 zap = mzap_open(os, obj, db);
538 if (zap == NULL) {
539 /*
540 * mzap_open() didn't like what it saw on-disk.
541 * Check for corruption!
542 */
543 return (SET_ERROR(EIO));
544 }
545 }
546
547 /*
548 * We're checking zap_ismicro without the lock held, in order to
549 * tell what type of lock we want. Once we have some sort of
550 * lock, see if it really is the right type. In practice this
551 * can only be different if it was upgraded from micro to fat,
552 * and micro wanted WRITER but fat only needs READER.
553 */
554 krw_t lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
555 rw_enter(&zap->zap_rwlock, lt);
556 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
557 /* it was upgraded, now we only need reader */
558 ASSERT(lt == RW_WRITER);
559 ASSERT(RW_READER ==
560 ((!zap->zap_ismicro && fatreader) ? RW_READER : lti));
561 rw_downgrade(&zap->zap_rwlock);
562 lt = RW_READER;
563 }
564
565 zap->zap_objset = os;
566
567 if (lt == RW_WRITER)
568 dmu_buf_will_dirty(db, tx);
569
570 ASSERT3P(zap->zap_dbuf, ==, db);
571
572 ASSERT(!zap->zap_ismicro ||
573 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
574 if (zap->zap_ismicro && tx && adding &&
575 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
576 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
577 if (newsz > zap_micro_max_size) {
578 dprintf("upgrading obj %llu: num_entries=%u\n",
579 (u_longlong_t)obj, zap->zap_m.zap_num_entries);
580 *zapp = zap;
581 int err = mzap_upgrade(zapp, tag, tx, 0);
582 if (err != 0)
583 rw_exit(&zap->zap_rwlock);
584 return (err);
585 }
586 VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
587 zap->zap_m.zap_num_chunks =
588 db->db_size / MZAP_ENT_LEN - 1;
589 }
590
591 *zapp = zap;
592 return (0);
593 }
594
595 static int
596 zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
597 krw_t lti, boolean_t fatreader, boolean_t adding, const void *tag,
598 zap_t **zapp)
599 {
600 dmu_buf_t *db;
601
602 int err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
603 if (err != 0) {
604 return (err);
605 }
606 #ifdef ZFS_DEBUG
607 {
608 dmu_object_info_t doi;
609 dmu_object_info_from_db(db, &doi);
610 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
611 }
612 #endif
613
614 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
615 if (err != 0) {
616 dmu_buf_rele(db, tag);
617 }
618 return (err);
619 }
620
621 int
622 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
623 krw_t lti, boolean_t fatreader, boolean_t adding, const void *tag,
624 zap_t **zapp)
625 {
626 dmu_buf_t *db;
627
628 int err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH);
629 if (err != 0)
630 return (err);
631 #ifdef ZFS_DEBUG
632 {
633 dmu_object_info_t doi;
634 dmu_object_info_from_db(db, &doi);
635 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
636 }
637 #endif
638 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
639 if (err != 0)
640 dmu_buf_rele(db, tag);
641 return (err);
642 }
643
644 void
645 zap_unlockdir(zap_t *zap, const void *tag)
646 {
647 rw_exit(&zap->zap_rwlock);
648 dmu_buf_rele(zap->zap_dbuf, tag);
649 }
650
651 static int
652 mzap_upgrade(zap_t **zapp, const void *tag, dmu_tx_t *tx, zap_flags_t flags)
653 {
654 int err = 0;
655 zap_t *zap = *zapp;
656
657 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
658
659 int sz = zap->zap_dbuf->db_size;
660 mzap_phys_t *mzp = vmem_alloc(sz, KM_SLEEP);
661 memcpy(mzp, zap->zap_dbuf->db_data, sz);
662 int nchunks = zap->zap_m.zap_num_chunks;
663
664 if (!flags) {
665 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
666 1ULL << fzap_default_block_shift, 0, tx);
667 if (err != 0) {
668 vmem_free(mzp, sz);
669 return (err);
670 }
671 }
672
673 dprintf("upgrading obj=%llu with %u chunks\n",
674 (u_longlong_t)zap->zap_object, nchunks);
675 /* XXX destroy the tree later, so we can use the stored hash value */
676 mze_destroy(zap);
677
678 fzap_upgrade(zap, tx, flags);
679
680 zap_name_t *zn = zap_name_alloc(zap);
681 for (int i = 0; i < nchunks; i++) {
682 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
683 if (mze->mze_name[0] == 0)
684 continue;
685 dprintf("adding %s=%llu\n",
686 mze->mze_name, (u_longlong_t)mze->mze_value);
687 zap_name_init_str(zn, mze->mze_name, 0);
688 /* If we fail here, we would end up losing entries */
689 VERIFY0(fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
690 tag, tx));
691 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
692 }
693 zap_name_free(zn);
694 vmem_free(mzp, sz);
695 *zapp = zap;
696 return (0);
697 }
698
699 /*
700 * The "normflags" determine the behavior of the matchtype_t which is
701 * passed to zap_lookup_norm(). Names which have the same normalized
702 * version will be stored with the same hash value, and therefore we can
703 * perform normalization-insensitive lookups. We can be Unicode form-
704 * insensitive and/or case-insensitive. The following flags are valid for
705 * "normflags":
706 *
707 * U8_TEXTPREP_NFC
708 * U8_TEXTPREP_NFD
709 * U8_TEXTPREP_NFKC
710 * U8_TEXTPREP_NFKD
711 * U8_TEXTPREP_TOUPPER
712 *
713 * The *_NF* (Normalization Form) flags are mutually exclusive; at most one
714 * of them may be supplied.
715 */
716 void
717 mzap_create_impl(dnode_t *dn, int normflags, zap_flags_t flags, dmu_tx_t *tx)
718 {
719 dmu_buf_t *db;
720
721 VERIFY0(dmu_buf_hold_by_dnode(dn, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
722
723 dmu_buf_will_dirty(db, tx);
724 mzap_phys_t *zp = db->db_data;
725 zp->mz_block_type = ZBT_MICRO;
726 zp->mz_salt =
727 ((uintptr_t)db ^ (uintptr_t)tx ^ (dn->dn_object << 1)) | 1ULL;
728 zp->mz_normflags = normflags;
729
730 if (flags != 0) {
731 zap_t *zap;
732 /* Only fat zap supports flags; upgrade immediately. */
733 VERIFY0(zap_lockdir_impl(db, FTAG, tx, RW_WRITER,
734 B_FALSE, B_FALSE, &zap));
735 VERIFY0(mzap_upgrade(&zap, FTAG, tx, flags));
736 zap_unlockdir(zap, FTAG);
737 } else {
738 dmu_buf_rele(db, FTAG);
739 }
740 }
741
742 static uint64_t
743 zap_create_impl(objset_t *os, int normflags, zap_flags_t flags,
744 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
745 dmu_object_type_t bonustype, int bonuslen, int dnodesize,
746 dnode_t **allocated_dnode, const void *tag, dmu_tx_t *tx)
747 {
748 uint64_t obj;
749
750 ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
751
752 if (allocated_dnode == NULL) {
753 dnode_t *dn;
754 obj = dmu_object_alloc_hold(os, ot, 1ULL << leaf_blockshift,
755 indirect_blockshift, bonustype, bonuslen, dnodesize,
756 &dn, FTAG, tx);
757 mzap_create_impl(dn, normflags, flags, tx);
758 dnode_rele(dn, FTAG);
759 } else {
760 obj = dmu_object_alloc_hold(os, ot, 1ULL << leaf_blockshift,
761 indirect_blockshift, bonustype, bonuslen, dnodesize,
762 allocated_dnode, tag, tx);
763 mzap_create_impl(*allocated_dnode, normflags, flags, tx);
764 }
765
766 return (obj);
767 }
768
769 int
770 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
771 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
772 {
773 return (zap_create_claim_dnsize(os, obj, ot, bonustype, bonuslen,
774 0, tx));
775 }
776
777 int
778 zap_create_claim_dnsize(objset_t *os, uint64_t obj, dmu_object_type_t ot,
779 dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
780 {
781 return (zap_create_claim_norm_dnsize(os, obj,
782 0, ot, bonustype, bonuslen, dnodesize, tx));
783 }
784
785 int
786 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
787 dmu_object_type_t ot,
788 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
789 {
790 return (zap_create_claim_norm_dnsize(os, obj, normflags, ot, bonustype,
791 bonuslen, 0, tx));
792 }
793
794 int
795 zap_create_claim_norm_dnsize(objset_t *os, uint64_t obj, int normflags,
796 dmu_object_type_t ot, dmu_object_type_t bonustype, int bonuslen,
797 int dnodesize, dmu_tx_t *tx)
798 {
799 dnode_t *dn;
800 int error;
801
802 ASSERT3U(DMU_OT_BYTESWAP(ot), ==, DMU_BSWAP_ZAP);
803 error = dmu_object_claim_dnsize(os, obj, ot, 0, bonustype, bonuslen,
804 dnodesize, tx);
805 if (error != 0)
806 return (error);
807
808 error = dnode_hold(os, obj, FTAG, &dn);
809 if (error != 0)
810 return (error);
811
812 mzap_create_impl(dn, normflags, 0, tx);
813
814 dnode_rele(dn, FTAG);
815
816 return (0);
817 }
818
819 uint64_t
820 zap_create(objset_t *os, dmu_object_type_t ot,
821 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
822 {
823 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
824 }
825
826 uint64_t
827 zap_create_dnsize(objset_t *os, dmu_object_type_t ot,
828 dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
829 {
830 return (zap_create_norm_dnsize(os, 0, ot, bonustype, bonuslen,
831 dnodesize, tx));
832 }
833
834 uint64_t
835 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
836 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
837 {
838 return (zap_create_norm_dnsize(os, normflags, ot, bonustype, bonuslen,
839 0, tx));
840 }
841
842 uint64_t
843 zap_create_norm_dnsize(objset_t *os, int normflags, dmu_object_type_t ot,
844 dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
845 {
846 return (zap_create_impl(os, normflags, 0, ot, 0, 0,
847 bonustype, bonuslen, dnodesize, NULL, NULL, tx));
848 }
849
850 uint64_t
851 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
852 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
853 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
854 {
855 return (zap_create_flags_dnsize(os, normflags, flags, ot,
856 leaf_blockshift, indirect_blockshift, bonustype, bonuslen, 0, tx));
857 }
858
859 uint64_t
860 zap_create_flags_dnsize(objset_t *os, int normflags, zap_flags_t flags,
861 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
862 dmu_object_type_t bonustype, int bonuslen, int dnodesize, dmu_tx_t *tx)
863 {
864 return (zap_create_impl(os, normflags, flags, ot, leaf_blockshift,
865 indirect_blockshift, bonustype, bonuslen, dnodesize, NULL, NULL,
866 tx));
867 }
868
869 /*
870 * Create a zap object and return a pointer to the newly allocated dnode via
871 * the allocated_dnode argument. The returned dnode will be held and the
872 * caller is responsible for releasing the hold by calling dnode_rele().
873 */
874 uint64_t
875 zap_create_hold(objset_t *os, int normflags, zap_flags_t flags,
876 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
877 dmu_object_type_t bonustype, int bonuslen, int dnodesize,
878 dnode_t **allocated_dnode, const void *tag, dmu_tx_t *tx)
879 {
880 return (zap_create_impl(os, normflags, flags, ot, leaf_blockshift,
881 indirect_blockshift, bonustype, bonuslen, dnodesize,
882 allocated_dnode, tag, tx));
883 }
884
885 int
886 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
887 {
888 /*
889 * dmu_object_free will free the object number and free the
890 * data. Freeing the data will cause our pageout function to be
891 * called, which will destroy our data (zap_leaf_t's and zap_t).
892 */
893
894 return (dmu_object_free(os, zapobj, tx));
895 }
896
897 void
898 zap_evict_sync(void *dbu)
899 {
900 zap_t *zap = dbu;
901
902 rw_destroy(&zap->zap_rwlock);
903
904 if (zap->zap_ismicro)
905 mze_destroy(zap);
906 else
907 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
908
909 kmem_free(zap, sizeof (zap_t));
910 }
911
912 int
913 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
914 {
915 zap_t *zap;
916
917 int err =
918 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
919 if (err != 0)
920 return (err);
921 if (!zap->zap_ismicro) {
922 err = fzap_count(zap, count);
923 } else {
924 *count = zap->zap_m.zap_num_entries;
925 }
926 zap_unlockdir(zap, FTAG);
927 return (err);
928 }
929
930 /*
931 * zn may be NULL; if not specified, it will be computed if needed.
932 * See also the comment above zap_entry_normalization_conflict().
933 */
934 static boolean_t
935 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze,
936 zfs_btree_index_t *idx)
937 {
938 boolean_t allocdzn = B_FALSE;
939 mzap_ent_t *other;
940 zfs_btree_index_t oidx;
941
942 if (zap->zap_normflags == 0)
943 return (B_FALSE);
944
945 for (other = zfs_btree_prev(&zap->zap_m.zap_tree, idx, &oidx);
946 other && other->mze_hash == mze->mze_hash;
947 other = zfs_btree_prev(&zap->zap_m.zap_tree, &oidx, &oidx)) {
948
949 if (zn == NULL) {
950 zn = zap_name_alloc_str(zap,
951 MZE_PHYS(zap, mze)->mze_name, MT_NORMALIZE);
952 allocdzn = B_TRUE;
953 }
954 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
955 if (allocdzn)
956 zap_name_free(zn);
957 return (B_TRUE);
958 }
959 }
960
961 for (other = zfs_btree_next(&zap->zap_m.zap_tree, idx, &oidx);
962 other && other->mze_hash == mze->mze_hash;
963 other = zfs_btree_next(&zap->zap_m.zap_tree, &oidx, &oidx)) {
964
965 if (zn == NULL) {
966 zn = zap_name_alloc_str(zap,
967 MZE_PHYS(zap, mze)->mze_name, MT_NORMALIZE);
968 allocdzn = B_TRUE;
969 }
970 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
971 if (allocdzn)
972 zap_name_free(zn);
973 return (B_TRUE);
974 }
975 }
976
977 if (allocdzn)
978 zap_name_free(zn);
979 return (B_FALSE);
980 }
981
982 /*
983 * Routines for manipulating attributes.
984 */
985
986 int
987 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
988 uint64_t integer_size, uint64_t num_integers, void *buf)
989 {
990 return (zap_lookup_norm(os, zapobj, name, integer_size,
991 num_integers, buf, 0, NULL, 0, NULL));
992 }
993
994 static int
995 zap_lookup_impl(zap_t *zap, const char *name,
996 uint64_t integer_size, uint64_t num_integers, void *buf,
997 matchtype_t mt, char *realname, int rn_len,
998 boolean_t *ncp)
999 {
1000 int err = 0;
1001
1002 zap_name_t *zn = zap_name_alloc_str(zap, name, mt);
1003 if (zn == NULL)
1004 return (SET_ERROR(ENOTSUP));
1005
1006 if (!zap->zap_ismicro) {
1007 err = fzap_lookup(zn, integer_size, num_integers, buf,
1008 realname, rn_len, ncp);
1009 } else {
1010 zfs_btree_index_t idx;
1011 mzap_ent_t *mze = mze_find(zn, &idx);
1012 if (mze == NULL) {
1013 err = SET_ERROR(ENOENT);
1014 } else {
1015 if (num_integers < 1) {
1016 err = SET_ERROR(EOVERFLOW);
1017 } else if (integer_size != 8) {
1018 err = SET_ERROR(EINVAL);
1019 } else {
1020 *(uint64_t *)buf =
1021 MZE_PHYS(zap, mze)->mze_value;
1022 if (realname != NULL)
1023 (void) strlcpy(realname,
1024 MZE_PHYS(zap, mze)->mze_name,
1025 rn_len);
1026 if (ncp) {
1027 *ncp = mzap_normalization_conflict(zap,
1028 zn, mze, &idx);
1029 }
1030 }
1031 }
1032 }
1033 zap_name_free(zn);
1034 return (err);
1035 }
1036
1037 int
1038 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
1039 uint64_t integer_size, uint64_t num_integers, void *buf,
1040 matchtype_t mt, char *realname, int rn_len,
1041 boolean_t *ncp)
1042 {
1043 zap_t *zap;
1044
1045 int err =
1046 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1047 if (err != 0)
1048 return (err);
1049 err = zap_lookup_impl(zap, name, integer_size,
1050 num_integers, buf, mt, realname, rn_len, ncp);
1051 zap_unlockdir(zap, FTAG);
1052 return (err);
1053 }
1054
1055 int
1056 zap_prefetch(objset_t *os, uint64_t zapobj, const char *name)
1057 {
1058 zap_t *zap;
1059 int err;
1060 zap_name_t *zn;
1061
1062 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1063 if (err)
1064 return (err);
1065 zn = zap_name_alloc_str(zap, name, 0);
1066 if (zn == NULL) {
1067 zap_unlockdir(zap, FTAG);
1068 return (SET_ERROR(ENOTSUP));
1069 }
1070
1071 fzap_prefetch(zn);
1072 zap_name_free(zn);
1073 zap_unlockdir(zap, FTAG);
1074 return (err);
1075 }
1076
1077 int
1078 zap_lookup_by_dnode(dnode_t *dn, const char *name,
1079 uint64_t integer_size, uint64_t num_integers, void *buf)
1080 {
1081 return (zap_lookup_norm_by_dnode(dn, name, integer_size,
1082 num_integers, buf, 0, NULL, 0, NULL));
1083 }
1084
1085 int
1086 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
1087 uint64_t integer_size, uint64_t num_integers, void *buf,
1088 matchtype_t mt, char *realname, int rn_len,
1089 boolean_t *ncp)
1090 {
1091 zap_t *zap;
1092
1093 int err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
1094 FTAG, &zap);
1095 if (err != 0)
1096 return (err);
1097 err = zap_lookup_impl(zap, name, integer_size,
1098 num_integers, buf, mt, realname, rn_len, ncp);
1099 zap_unlockdir(zap, FTAG);
1100 return (err);
1101 }
1102
1103 int
1104 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1105 int key_numints)
1106 {
1107 zap_t *zap;
1108
1109 int err =
1110 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1111 if (err != 0)
1112 return (err);
1113 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1114 if (zn == NULL) {
1115 zap_unlockdir(zap, FTAG);
1116 return (SET_ERROR(ENOTSUP));
1117 }
1118
1119 fzap_prefetch(zn);
1120 zap_name_free(zn);
1121 zap_unlockdir(zap, FTAG);
1122 return (err);
1123 }
1124
1125 int
1126 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1127 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
1128 {
1129 zap_t *zap;
1130
1131 int err =
1132 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1133 if (err != 0)
1134 return (err);
1135 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1136 if (zn == NULL) {
1137 zap_unlockdir(zap, FTAG);
1138 return (SET_ERROR(ENOTSUP));
1139 }
1140
1141 err = fzap_lookup(zn, integer_size, num_integers, buf,
1142 NULL, 0, NULL);
1143 zap_name_free(zn);
1144 zap_unlockdir(zap, FTAG);
1145 return (err);
1146 }
1147
1148 int
1149 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
1150 {
1151 int err = zap_lookup_norm(os, zapobj, name, 0,
1152 0, NULL, 0, NULL, 0, NULL);
1153 if (err == EOVERFLOW || err == EINVAL)
1154 err = 0; /* found, but skipped reading the value */
1155 return (err);
1156 }
1157
1158 int
1159 zap_length(objset_t *os, uint64_t zapobj, const char *name,
1160 uint64_t *integer_size, uint64_t *num_integers)
1161 {
1162 zap_t *zap;
1163
1164 int err =
1165 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1166 if (err != 0)
1167 return (err);
1168 zap_name_t *zn = zap_name_alloc_str(zap, name, 0);
1169 if (zn == NULL) {
1170 zap_unlockdir(zap, FTAG);
1171 return (SET_ERROR(ENOTSUP));
1172 }
1173 if (!zap->zap_ismicro) {
1174 err = fzap_length(zn, integer_size, num_integers);
1175 } else {
1176 zfs_btree_index_t idx;
1177 mzap_ent_t *mze = mze_find(zn, &idx);
1178 if (mze == NULL) {
1179 err = SET_ERROR(ENOENT);
1180 } else {
1181 if (integer_size)
1182 *integer_size = 8;
1183 if (num_integers)
1184 *num_integers = 1;
1185 }
1186 }
1187 zap_name_free(zn);
1188 zap_unlockdir(zap, FTAG);
1189 return (err);
1190 }
1191
1192 int
1193 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1194 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1195 {
1196 zap_t *zap;
1197
1198 int err =
1199 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1200 if (err != 0)
1201 return (err);
1202 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1203 if (zn == NULL) {
1204 zap_unlockdir(zap, FTAG);
1205 return (SET_ERROR(ENOTSUP));
1206 }
1207 err = fzap_length(zn, integer_size, num_integers);
1208 zap_name_free(zn);
1209 zap_unlockdir(zap, FTAG);
1210 return (err);
1211 }
1212
1213 static void
1214 mzap_addent(zap_name_t *zn, uint64_t value)
1215 {
1216 zap_t *zap = zn->zn_zap;
1217 uint16_t start = zap->zap_m.zap_alloc_next;
1218
1219 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1220
1221 #ifdef ZFS_DEBUG
1222 for (int i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1223 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1224 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1225 }
1226 #endif
1227
1228 uint32_t cd = mze_find_unused_cd(zap, zn->zn_hash);
1229 /* given the limited size of the microzap, this can't happen */
1230 ASSERT(cd < zap_maxcd(zap));
1231
1232 again:
1233 for (uint16_t i = start; i < zap->zap_m.zap_num_chunks; i++) {
1234 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1235 if (mze->mze_name[0] == 0) {
1236 mze->mze_value = value;
1237 mze->mze_cd = cd;
1238 (void) strlcpy(mze->mze_name, zn->zn_key_orig,
1239 sizeof (mze->mze_name));
1240 zap->zap_m.zap_num_entries++;
1241 zap->zap_m.zap_alloc_next = i+1;
1242 if (zap->zap_m.zap_alloc_next ==
1243 zap->zap_m.zap_num_chunks)
1244 zap->zap_m.zap_alloc_next = 0;
1245 mze_insert(zap, i, zn->zn_hash);
1246 return;
1247 }
1248 }
1249 if (start != 0) {
1250 start = 0;
1251 goto again;
1252 }
1253 cmn_err(CE_PANIC, "out of entries!");
1254 }
1255
1256 static int
1257 zap_add_impl(zap_t *zap, const char *key,
1258 int integer_size, uint64_t num_integers,
1259 const void *val, dmu_tx_t *tx, const void *tag)
1260 {
1261 const uint64_t *intval = val;
1262 int err = 0;
1263
1264 zap_name_t *zn = zap_name_alloc_str(zap, key, 0);
1265 if (zn == NULL) {
1266 zap_unlockdir(zap, tag);
1267 return (SET_ERROR(ENOTSUP));
1268 }
1269 if (!zap->zap_ismicro) {
1270 err = fzap_add(zn, integer_size, num_integers, val, tag, tx);
1271 zap = zn->zn_zap; /* fzap_add() may change zap */
1272 } else if (integer_size != 8 || num_integers != 1 ||
1273 strlen(key) >= MZAP_NAME_LEN ||
1274 !mze_canfit_fzap_leaf(zn, zn->zn_hash)) {
1275 err = mzap_upgrade(&zn->zn_zap, tag, tx, 0);
1276 if (err == 0) {
1277 err = fzap_add(zn, integer_size, num_integers, val,
1278 tag, tx);
1279 }
1280 zap = zn->zn_zap; /* fzap_add() may change zap */
1281 } else {
1282 zfs_btree_index_t idx;
1283 if (mze_find(zn, &idx) != NULL) {
1284 err = SET_ERROR(EEXIST);
1285 } else {
1286 mzap_addent(zn, *intval);
1287 }
1288 }
1289 ASSERT(zap == zn->zn_zap);
1290 zap_name_free(zn);
1291 if (zap != NULL) /* may be NULL if fzap_add() failed */
1292 zap_unlockdir(zap, tag);
1293 return (err);
1294 }
1295
1296 int
1297 zap_add(objset_t *os, uint64_t zapobj, const char *key,
1298 int integer_size, uint64_t num_integers,
1299 const void *val, dmu_tx_t *tx)
1300 {
1301 zap_t *zap;
1302 int err;
1303
1304 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1305 if (err != 0)
1306 return (err);
1307 err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1308 /* zap_add_impl() calls zap_unlockdir() */
1309 return (err);
1310 }
1311
1312 int
1313 zap_add_by_dnode(dnode_t *dn, const char *key,
1314 int integer_size, uint64_t num_integers,
1315 const void *val, dmu_tx_t *tx)
1316 {
1317 zap_t *zap;
1318 int err;
1319
1320 err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1321 if (err != 0)
1322 return (err);
1323 err = zap_add_impl(zap, key, integer_size, num_integers, val, tx, FTAG);
1324 /* zap_add_impl() calls zap_unlockdir() */
1325 return (err);
1326 }
1327
1328 int
1329 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1330 int key_numints, int integer_size, uint64_t num_integers,
1331 const void *val, dmu_tx_t *tx)
1332 {
1333 zap_t *zap;
1334
1335 int err =
1336 zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1337 if (err != 0)
1338 return (err);
1339 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1340 if (zn == NULL) {
1341 zap_unlockdir(zap, FTAG);
1342 return (SET_ERROR(ENOTSUP));
1343 }
1344 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1345 zap = zn->zn_zap; /* fzap_add() may change zap */
1346 zap_name_free(zn);
1347 if (zap != NULL) /* may be NULL if fzap_add() failed */
1348 zap_unlockdir(zap, FTAG);
1349 return (err);
1350 }
1351
1352 int
1353 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1354 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1355 {
1356 zap_t *zap;
1357 const uint64_t *intval = val;
1358
1359 int err =
1360 zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1361 if (err != 0)
1362 return (err);
1363 zap_name_t *zn = zap_name_alloc_str(zap, name, 0);
1364 if (zn == NULL) {
1365 zap_unlockdir(zap, FTAG);
1366 return (SET_ERROR(ENOTSUP));
1367 }
1368 if (!zap->zap_ismicro) {
1369 err = fzap_update(zn, integer_size, num_integers, val,
1370 FTAG, tx);
1371 zap = zn->zn_zap; /* fzap_update() may change zap */
1372 } else if (integer_size != 8 || num_integers != 1 ||
1373 strlen(name) >= MZAP_NAME_LEN) {
1374 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1375 (u_longlong_t)zapobj, integer_size,
1376 (u_longlong_t)num_integers, name);
1377 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1378 if (err == 0) {
1379 err = fzap_update(zn, integer_size, num_integers,
1380 val, FTAG, tx);
1381 }
1382 zap = zn->zn_zap; /* fzap_update() may change zap */
1383 } else {
1384 zfs_btree_index_t idx;
1385 mzap_ent_t *mze = mze_find(zn, &idx);
1386 if (mze != NULL) {
1387 MZE_PHYS(zap, mze)->mze_value = *intval;
1388 } else {
1389 mzap_addent(zn, *intval);
1390 }
1391 }
1392 ASSERT(zap == zn->zn_zap);
1393 zap_name_free(zn);
1394 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1395 zap_unlockdir(zap, FTAG);
1396 return (err);
1397 }
1398
1399 int
1400 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1401 int key_numints,
1402 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1403 {
1404 zap_t *zap;
1405
1406 int err =
1407 zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1408 if (err != 0)
1409 return (err);
1410 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1411 if (zn == NULL) {
1412 zap_unlockdir(zap, FTAG);
1413 return (SET_ERROR(ENOTSUP));
1414 }
1415 err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
1416 zap = zn->zn_zap; /* fzap_update() may change zap */
1417 zap_name_free(zn);
1418 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1419 zap_unlockdir(zap, FTAG);
1420 return (err);
1421 }
1422
1423 int
1424 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1425 {
1426 return (zap_remove_norm(os, zapobj, name, 0, tx));
1427 }
1428
1429 static int
1430 zap_remove_impl(zap_t *zap, const char *name,
1431 matchtype_t mt, dmu_tx_t *tx)
1432 {
1433 int err = 0;
1434
1435 zap_name_t *zn = zap_name_alloc_str(zap, name, mt);
1436 if (zn == NULL)
1437 return (SET_ERROR(ENOTSUP));
1438 if (!zap->zap_ismicro) {
1439 err = fzap_remove(zn, tx);
1440 } else {
1441 zfs_btree_index_t idx;
1442 mzap_ent_t *mze = mze_find(zn, &idx);
1443 if (mze == NULL) {
1444 err = SET_ERROR(ENOENT);
1445 } else {
1446 zap->zap_m.zap_num_entries--;
1447 memset(MZE_PHYS(zap, mze), 0, sizeof (mzap_ent_phys_t));
1448 zfs_btree_remove_idx(&zap->zap_m.zap_tree, &idx);
1449 }
1450 }
1451 zap_name_free(zn);
1452 return (err);
1453 }
1454
1455 int
1456 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1457 matchtype_t mt, dmu_tx_t *tx)
1458 {
1459 zap_t *zap;
1460 int err;
1461
1462 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1463 if (err)
1464 return (err);
1465 err = zap_remove_impl(zap, name, mt, tx);
1466 zap_unlockdir(zap, FTAG);
1467 return (err);
1468 }
1469
1470 int
1471 zap_remove_by_dnode(dnode_t *dn, const char *name, dmu_tx_t *tx)
1472 {
1473 zap_t *zap;
1474 int err;
1475
1476 err = zap_lockdir_by_dnode(dn, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1477 if (err)
1478 return (err);
1479 err = zap_remove_impl(zap, name, 0, tx);
1480 zap_unlockdir(zap, FTAG);
1481 return (err);
1482 }
1483
1484 int
1485 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1486 int key_numints, dmu_tx_t *tx)
1487 {
1488 zap_t *zap;
1489
1490 int err =
1491 zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1492 if (err != 0)
1493 return (err);
1494 zap_name_t *zn = zap_name_alloc_uint64(zap, key, key_numints);
1495 if (zn == NULL) {
1496 zap_unlockdir(zap, FTAG);
1497 return (SET_ERROR(ENOTSUP));
1498 }
1499 err = fzap_remove(zn, tx);
1500 zap_name_free(zn);
1501 zap_unlockdir(zap, FTAG);
1502 return (err);
1503 }
1504
1505 /*
1506 * Routines for iterating over the attributes.
1507 */
1508
1509 static void
1510 zap_cursor_init_impl(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1511 uint64_t serialized, boolean_t prefetch)
1512 {
1513 zc->zc_objset = os;
1514 zc->zc_zap = NULL;
1515 zc->zc_leaf = NULL;
1516 zc->zc_zapobj = zapobj;
1517 zc->zc_serialized = serialized;
1518 zc->zc_hash = 0;
1519 zc->zc_cd = 0;
1520 zc->zc_prefetch = prefetch;
1521 }
1522 void
1523 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1524 uint64_t serialized)
1525 {
1526 zap_cursor_init_impl(zc, os, zapobj, serialized, B_TRUE);
1527 }
1528
1529 /*
1530 * Initialize a cursor at the beginning of the ZAP object. The entire
1531 * ZAP object will be prefetched.
1532 */
1533 void
1534 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1535 {
1536 zap_cursor_init_impl(zc, os, zapobj, 0, B_TRUE);
1537 }
1538
1539 /*
1540 * Initialize a cursor at the beginning, but request that we not prefetch
1541 * the entire ZAP object.
1542 */
1543 void
1544 zap_cursor_init_noprefetch(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1545 {
1546 zap_cursor_init_impl(zc, os, zapobj, 0, B_FALSE);
1547 }
1548
1549 void
1550 zap_cursor_fini(zap_cursor_t *zc)
1551 {
1552 if (zc->zc_zap) {
1553 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1554 zap_unlockdir(zc->zc_zap, NULL);
1555 zc->zc_zap = NULL;
1556 }
1557 if (zc->zc_leaf) {
1558 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1559 zap_put_leaf(zc->zc_leaf);
1560 zc->zc_leaf = NULL;
1561 }
1562 zc->zc_objset = NULL;
1563 }
1564
1565 uint64_t
1566 zap_cursor_serialize(zap_cursor_t *zc)
1567 {
1568 if (zc->zc_hash == -1ULL)
1569 return (-1ULL);
1570 if (zc->zc_zap == NULL)
1571 return (zc->zc_serialized);
1572 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1573 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1574
1575 /*
1576 * We want to keep the high 32 bits of the cursor zero if we can, so
1577 * that 32-bit programs can access this. So usually use a small
1578 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1579 * of the cursor.
1580 *
1581 * [ collision differentiator | zap_hashbits()-bit hash value ]
1582 */
1583 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1584 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1585 }
1586
1587 int
1588 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1589 {
1590 int err;
1591
1592 if (zc->zc_hash == -1ULL)
1593 return (SET_ERROR(ENOENT));
1594
1595 if (zc->zc_zap == NULL) {
1596 int hb;
1597 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1598 RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1599 if (err != 0)
1600 return (err);
1601
1602 /*
1603 * To support zap_cursor_init_serialized, advance, retrieve,
1604 * we must add to the existing zc_cd, which may already
1605 * be 1 due to the zap_cursor_advance.
1606 */
1607 ASSERT(zc->zc_hash == 0);
1608 hb = zap_hashbits(zc->zc_zap);
1609 zc->zc_hash = zc->zc_serialized << (64 - hb);
1610 zc->zc_cd += zc->zc_serialized >> hb;
1611 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1612 zc->zc_cd = 0;
1613 } else {
1614 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1615 }
1616 if (!zc->zc_zap->zap_ismicro) {
1617 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1618 } else {
1619 zfs_btree_index_t idx;
1620 mzap_ent_t mze_tofind;
1621
1622 mze_tofind.mze_hash = zc->zc_hash >> 32;
1623 mze_tofind.mze_cd = zc->zc_cd;
1624
1625 mzap_ent_t *mze = zfs_btree_find(&zc->zc_zap->zap_m.zap_tree,
1626 &mze_tofind, &idx);
1627 if (mze == NULL) {
1628 mze = zfs_btree_next(&zc->zc_zap->zap_m.zap_tree,
1629 &idx, &idx);
1630 }
1631 if (mze) {
1632 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1633 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1634 za->za_normalization_conflict =
1635 mzap_normalization_conflict(zc->zc_zap, NULL,
1636 mze, &idx);
1637 za->za_integer_length = 8;
1638 za->za_num_integers = 1;
1639 za->za_first_integer = mzep->mze_value;
1640 (void) strlcpy(za->za_name, mzep->mze_name,
1641 sizeof (za->za_name));
1642 zc->zc_hash = (uint64_t)mze->mze_hash << 32;
1643 zc->zc_cd = mze->mze_cd;
1644 err = 0;
1645 } else {
1646 zc->zc_hash = -1ULL;
1647 err = SET_ERROR(ENOENT);
1648 }
1649 }
1650 rw_exit(&zc->zc_zap->zap_rwlock);
1651 return (err);
1652 }
1653
1654 void
1655 zap_cursor_advance(zap_cursor_t *zc)
1656 {
1657 if (zc->zc_hash == -1ULL)
1658 return;
1659 zc->zc_cd++;
1660 }
1661
1662 int
1663 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1664 {
1665 zap_t *zap;
1666
1667 int err =
1668 zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1669 if (err != 0)
1670 return (err);
1671
1672 memset(zs, 0, sizeof (zap_stats_t));
1673
1674 if (zap->zap_ismicro) {
1675 zs->zs_blocksize = zap->zap_dbuf->db_size;
1676 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1677 zs->zs_num_blocks = 1;
1678 } else {
1679 fzap_get_stats(zap, zs);
1680 }
1681 zap_unlockdir(zap, FTAG);
1682 return (0);
1683 }
1684
1685 #if defined(_KERNEL)
1686 EXPORT_SYMBOL(zap_create);
1687 EXPORT_SYMBOL(zap_create_dnsize);
1688 EXPORT_SYMBOL(zap_create_norm);
1689 EXPORT_SYMBOL(zap_create_norm_dnsize);
1690 EXPORT_SYMBOL(zap_create_flags);
1691 EXPORT_SYMBOL(zap_create_flags_dnsize);
1692 EXPORT_SYMBOL(zap_create_claim);
1693 EXPORT_SYMBOL(zap_create_claim_norm);
1694 EXPORT_SYMBOL(zap_create_claim_norm_dnsize);
1695 EXPORT_SYMBOL(zap_create_hold);
1696 EXPORT_SYMBOL(zap_destroy);
1697 EXPORT_SYMBOL(zap_lookup);
1698 EXPORT_SYMBOL(zap_lookup_by_dnode);
1699 EXPORT_SYMBOL(zap_lookup_norm);
1700 EXPORT_SYMBOL(zap_lookup_uint64);
1701 EXPORT_SYMBOL(zap_contains);
1702 EXPORT_SYMBOL(zap_prefetch);
1703 EXPORT_SYMBOL(zap_prefetch_uint64);
1704 EXPORT_SYMBOL(zap_add);
1705 EXPORT_SYMBOL(zap_add_by_dnode);
1706 EXPORT_SYMBOL(zap_add_uint64);
1707 EXPORT_SYMBOL(zap_update);
1708 EXPORT_SYMBOL(zap_update_uint64);
1709 EXPORT_SYMBOL(zap_length);
1710 EXPORT_SYMBOL(zap_length_uint64);
1711 EXPORT_SYMBOL(zap_remove);
1712 EXPORT_SYMBOL(zap_remove_by_dnode);
1713 EXPORT_SYMBOL(zap_remove_norm);
1714 EXPORT_SYMBOL(zap_remove_uint64);
1715 EXPORT_SYMBOL(zap_count);
1716 EXPORT_SYMBOL(zap_value_search);
1717 EXPORT_SYMBOL(zap_join);
1718 EXPORT_SYMBOL(zap_join_increment);
1719 EXPORT_SYMBOL(zap_add_int);
1720 EXPORT_SYMBOL(zap_remove_int);
1721 EXPORT_SYMBOL(zap_lookup_int);
1722 EXPORT_SYMBOL(zap_increment_int);
1723 EXPORT_SYMBOL(zap_add_int_key);
1724 EXPORT_SYMBOL(zap_lookup_int_key);
1725 EXPORT_SYMBOL(zap_increment);
1726 EXPORT_SYMBOL(zap_cursor_init);
1727 EXPORT_SYMBOL(zap_cursor_fini);
1728 EXPORT_SYMBOL(zap_cursor_retrieve);
1729 EXPORT_SYMBOL(zap_cursor_advance);
1730 EXPORT_SYMBOL(zap_cursor_serialize);
1731 EXPORT_SYMBOL(zap_cursor_init_serialized);
1732 EXPORT_SYMBOL(zap_get_stats);
1733
1734 /* CSTYLED */
1735 ZFS_MODULE_PARAM(zfs, , zap_micro_max_size, INT, ZMOD_RW,
1736 "Maximum micro ZAP size, before converting to a fat ZAP, in bytes");
1737 #endif
OpenZFS on Linux and FreeBSD