1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 */
27
28 #include <sys/zio.h>
29 #include <sys/spa.h>
30 #include <sys/dmu.h>
31 #include <sys/zfs_context.h>
32 #include <sys/zap.h>
33 #include <sys/refcount.h>
34 #include <sys/zap_impl.h>
35 #include <sys/zap_leaf.h>
36 #include <sys/avl.h>
37 #include <sys/arc.h>
38 #include <sys/dmu_objset.h>
39
40 #ifdef _KERNEL
41 #include <sys/sunddi.h>
42 #endif
43
44 extern inline mzap_phys_t *zap_m_phys(zap_t *zap);
45
46 static int mzap_upgrade(zap_t **zapp,
47 void *tag, dmu_tx_t *tx, zap_flags_t flags);
48
49 uint64_t
zap_getflags(zap_t * zap)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
zap_hashbits(zap_t * zap)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
zap_maxcd(zap_t * zap)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
zap_hash(zap_name_t * zn)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 int i;
91 const uint64_t *wp = zn->zn_key_norm;
92
93 ASSERT(zn->zn_key_intlen == 8);
94 for (i = 0; i < zn->zn_key_norm_numints; wp++, i++) {
95 int j;
96 uint64_t word = *wp;
97
98 for (j = 0; j < zn->zn_key_intlen; j++) {
99 h = (h >> 8) ^
100 zfs_crc64_table[(h ^ word) & 0xFF];
101 word >>= NBBY;
102 }
103 }
104 } else {
105 int i, len;
106 const uint8_t *cp = zn->zn_key_norm;
107
108 /*
109 * We previously stored the terminating null on
110 * disk, but didn't hash it, so we need to
111 * continue to not hash it. (The
112 * zn_key_*_numints includes the terminating
113 * null for non-binary keys.)
114 */
115 len = zn->zn_key_norm_numints - 1;
116
117 ASSERT(zn->zn_key_intlen == 1);
118 for (i = 0; i < len; cp++, i++) {
119 h = (h >> 8) ^
120 zfs_crc64_table[(h ^ *cp) & 0xFF];
121 }
122 }
123 }
124 /*
125 * Don't use all 64 bits, since we need some in the cookie for
126 * the collision differentiator. We MUST use the high bits,
127 * since those are the ones that we first pay attention to when
128 * chosing the bucket.
129 */
130 h &= ~((1ULL << (64 - zap_hashbits(zap))) - 1);
131
132 return (h);
133 }
134
135 static int
zap_normalize(zap_t * zap,const char * name,char * namenorm)136 zap_normalize(zap_t *zap, const char *name, char *namenorm)
137 {
138 size_t inlen, outlen;
139 int err;
140
141 ASSERT(!(zap_getflags(zap) & ZAP_FLAG_UINT64_KEY));
142
143 inlen = strlen(name) + 1;
144 outlen = ZAP_MAXNAMELEN;
145
146 err = 0;
147 (void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
148 zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL |
149 U8_TEXTPREP_IGNORE_INVALID, U8_UNICODE_LATEST, &err);
150
151 return (err);
152 }
153
154 boolean_t
zap_match(zap_name_t * zn,const char * matchname)155 zap_match(zap_name_t *zn, const char *matchname)
156 {
157 ASSERT(!(zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY));
158
159 if (zn->zn_matchtype == MT_FIRST) {
160 char norm[ZAP_MAXNAMELEN];
161
162 if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
163 return (B_FALSE);
164
165 return (strcmp(zn->zn_key_norm, norm) == 0);
166 } else {
167 /* MT_BEST or MT_EXACT */
168 return (strcmp(zn->zn_key_orig, matchname) == 0);
169 }
170 }
171
172 void
zap_name_free(zap_name_t * zn)173 zap_name_free(zap_name_t *zn)
174 {
175 kmem_free(zn, sizeof (zap_name_t));
176 }
177
178 zap_name_t *
zap_name_alloc(zap_t * zap,const char * key,matchtype_t mt)179 zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt)
180 {
181 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
182
183 zn->zn_zap = zap;
184 zn->zn_key_intlen = sizeof (*key);
185 zn->zn_key_orig = key;
186 zn->zn_key_orig_numints = strlen(zn->zn_key_orig) + 1;
187 zn->zn_matchtype = mt;
188 if (zap->zap_normflags) {
189 if (zap_normalize(zap, key, zn->zn_normbuf) != 0) {
190 zap_name_free(zn);
191 return (NULL);
192 }
193 zn->zn_key_norm = zn->zn_normbuf;
194 zn->zn_key_norm_numints = strlen(zn->zn_key_norm) + 1;
195 } else {
196 if (mt != MT_EXACT) {
197 zap_name_free(zn);
198 return (NULL);
199 }
200 zn->zn_key_norm = zn->zn_key_orig;
201 zn->zn_key_norm_numints = zn->zn_key_orig_numints;
202 }
203
204 zn->zn_hash = zap_hash(zn);
205 return (zn);
206 }
207
208 zap_name_t *
zap_name_alloc_uint64(zap_t * zap,const uint64_t * key,int numints)209 zap_name_alloc_uint64(zap_t *zap, const uint64_t *key, int numints)
210 {
211 zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
212
213 ASSERT(zap->zap_normflags == 0);
214 zn->zn_zap = zap;
215 zn->zn_key_intlen = sizeof (*key);
216 zn->zn_key_orig = zn->zn_key_norm = key;
217 zn->zn_key_orig_numints = zn->zn_key_norm_numints = numints;
218 zn->zn_matchtype = MT_EXACT;
219
220 zn->zn_hash = zap_hash(zn);
221 return (zn);
222 }
223
224 static void
mzap_byteswap(mzap_phys_t * buf,size_t size)225 mzap_byteswap(mzap_phys_t *buf, size_t size)
226 {
227 int i, max;
228 buf->mz_block_type = BSWAP_64(buf->mz_block_type);
229 buf->mz_salt = BSWAP_64(buf->mz_salt);
230 buf->mz_normflags = BSWAP_64(buf->mz_normflags);
231 max = (size / MZAP_ENT_LEN) - 1;
232 for (i = 0; i < max; i++) {
233 buf->mz_chunk[i].mze_value =
234 BSWAP_64(buf->mz_chunk[i].mze_value);
235 buf->mz_chunk[i].mze_cd =
236 BSWAP_32(buf->mz_chunk[i].mze_cd);
237 }
238 }
239
240 void
zap_byteswap(void * buf,size_t size)241 zap_byteswap(void *buf, size_t size)
242 {
243 uint64_t block_type;
244
245 block_type = *(uint64_t *)buf;
246
247 if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
248 /* ASSERT(magic == ZAP_LEAF_MAGIC); */
249 mzap_byteswap(buf, size);
250 } else {
251 fzap_byteswap(buf, size);
252 }
253 }
254
255 static int
mze_compare(const void * arg1,const void * arg2)256 mze_compare(const void *arg1, const void *arg2)
257 {
258 const mzap_ent_t *mze1 = arg1;
259 const mzap_ent_t *mze2 = arg2;
260
261 if (mze1->mze_hash > mze2->mze_hash)
262 return (+1);
263 if (mze1->mze_hash < mze2->mze_hash)
264 return (-1);
265 if (mze1->mze_cd > mze2->mze_cd)
266 return (+1);
267 if (mze1->mze_cd < mze2->mze_cd)
268 return (-1);
269 return (0);
270 }
271
272 static int
mze_insert(zap_t * zap,int chunkid,uint64_t hash)273 mze_insert(zap_t *zap, int chunkid, uint64_t hash)
274 {
275 mzap_ent_t *mze;
276 avl_index_t idx;
277
278 ASSERT(zap->zap_ismicro);
279 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
280
281 mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
282 mze->mze_chunkid = chunkid;
283 mze->mze_hash = hash;
284 mze->mze_cd = MZE_PHYS(zap, mze)->mze_cd;
285 ASSERT(MZE_PHYS(zap, mze)->mze_name[0] != 0);
286 if (avl_find(&zap->zap_m.zap_avl, mze, &idx) != NULL) {
287 kmem_free(mze, sizeof (mzap_ent_t));
288 return (EEXIST);
289 }
290 avl_insert(&zap->zap_m.zap_avl, mze, idx);
291 return (0);
292 }
293
294 static mzap_ent_t *
mze_find(zap_name_t * zn)295 mze_find(zap_name_t *zn)
296 {
297 mzap_ent_t mze_tofind;
298 mzap_ent_t *mze;
299 avl_index_t idx;
300 avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
301
302 ASSERT(zn->zn_zap->zap_ismicro);
303 ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
304
305 mze_tofind.mze_hash = zn->zn_hash;
306 mze_tofind.mze_cd = 0;
307
308 again:
309 mze = avl_find(avl, &mze_tofind, &idx);
310 if (mze == NULL)
311 mze = avl_nearest(avl, idx, AVL_AFTER);
312 for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
313 ASSERT3U(mze->mze_cd, ==, MZE_PHYS(zn->zn_zap, mze)->mze_cd);
314 if (zap_match(zn, MZE_PHYS(zn->zn_zap, mze)->mze_name))
315 return (mze);
316 }
317 if (zn->zn_matchtype == MT_BEST) {
318 zn->zn_matchtype = MT_FIRST;
319 goto again;
320 }
321 return (NULL);
322 }
323
324 static uint32_t
mze_find_unused_cd(zap_t * zap,uint64_t hash)325 mze_find_unused_cd(zap_t *zap, uint64_t hash)
326 {
327 mzap_ent_t mze_tofind;
328 mzap_ent_t *mze;
329 avl_index_t idx;
330 avl_tree_t *avl = &zap->zap_m.zap_avl;
331 uint32_t cd;
332
333 ASSERT(zap->zap_ismicro);
334 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
335
336 mze_tofind.mze_hash = hash;
337 mze_tofind.mze_cd = 0;
338
339 cd = 0;
340 for (mze = avl_find(avl, &mze_tofind, &idx);
341 mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
342 if (mze->mze_cd != cd)
343 break;
344 cd++;
345 }
346
347 return (cd);
348 }
349
350 static void
mze_remove(zap_t * zap,mzap_ent_t * mze)351 mze_remove(zap_t *zap, mzap_ent_t *mze)
352 {
353 ASSERT(zap->zap_ismicro);
354 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
355
356 avl_remove(&zap->zap_m.zap_avl, mze);
357 kmem_free(mze, sizeof (mzap_ent_t));
358 }
359
360 static void
mze_destroy(zap_t * zap)361 mze_destroy(zap_t *zap)
362 {
363 mzap_ent_t *mze;
364 void *avlcookie = NULL;
365
366 while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
367 kmem_free(mze, sizeof (mzap_ent_t));
368 avl_destroy(&zap->zap_m.zap_avl);
369 }
370
371 static zap_t *
mzap_open(objset_t * os,uint64_t obj,dmu_buf_t * db)372 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
373 {
374 zap_t *winner;
375 zap_t *zap;
376 int i;
377 uint64_t *zap_hdr = (uint64_t *)db->db_data;
378 uint64_t zap_block_type = zap_hdr[0];
379 uint64_t zap_magic = zap_hdr[1];
380
381 ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
382
383 zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
384 rw_init(&zap->zap_rwlock, 0, 0, 0);
385 rw_enter(&zap->zap_rwlock, RW_WRITER);
386 zap->zap_objset = os;
387 zap->zap_object = obj;
388 zap->zap_dbuf = db;
389
390 if (zap_block_type != ZBT_MICRO) {
391 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
392 zap->zap_f.zap_block_shift = highbit64(db->db_size) - 1;
393 if (zap_block_type != ZBT_HEADER || zap_magic != ZAP_MAGIC) {
394 winner = NULL; /* No actual winner here... */
395 goto handle_winner;
396 }
397 } else {
398 zap->zap_ismicro = TRUE;
399 }
400
401 /*
402 * Make sure that zap_ismicro is set before we let others see
403 * it, because zap_lockdir() checks zap_ismicro without the lock
404 * held.
405 */
406 dmu_buf_init_user(&zap->zap_dbu, zap_evict_sync, NULL, &zap->zap_dbuf);
407 winner = dmu_buf_set_user(db, &zap->zap_dbu);
408
409 if (winner != NULL)
410 goto handle_winner;
411
412 if (zap->zap_ismicro) {
413 zap->zap_salt = zap_m_phys(zap)->mz_salt;
414 zap->zap_normflags = zap_m_phys(zap)->mz_normflags;
415 zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
416 avl_create(&zap->zap_m.zap_avl, mze_compare,
417 sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
418
419 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
420 mzap_ent_phys_t *mze =
421 &zap_m_phys(zap)->mz_chunk[i];
422 if (mze->mze_name[0]) {
423 zap_name_t *zn;
424
425 zn = zap_name_alloc(zap, mze->mze_name,
426 MT_EXACT);
427 if (mze_insert(zap, i, zn->zn_hash) == 0)
428 zap->zap_m.zap_num_entries++;
429 else {
430 printf("ZFS WARNING: Duplicated ZAP "
431 "entry detected (%s).\n",
432 mze->mze_name);
433 }
434 zap_name_free(zn);
435 }
436 }
437 } else {
438 zap->zap_salt = zap_f_phys(zap)->zap_salt;
439 zap->zap_normflags = zap_f_phys(zap)->zap_normflags;
440
441 ASSERT3U(sizeof (struct zap_leaf_header), ==,
442 2*ZAP_LEAF_CHUNKSIZE);
443
444 /*
445 * The embedded pointer table should not overlap the
446 * other members.
447 */
448 ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
449 &zap_f_phys(zap)->zap_salt);
450
451 /*
452 * The embedded pointer table should end at the end of
453 * the block
454 */
455 ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
456 1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
457 (uintptr_t)zap_f_phys(zap), ==,
458 zap->zap_dbuf->db_size);
459 }
460 rw_exit(&zap->zap_rwlock);
461 return (zap);
462
463 handle_winner:
464 rw_exit(&zap->zap_rwlock);
465 rw_destroy(&zap->zap_rwlock);
466 if (!zap->zap_ismicro)
467 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
468 kmem_free(zap, sizeof (zap_t));
469 return (winner);
470 }
471
472 static int
zap_lockdir_impl(dmu_buf_t * db,void * tag,dmu_tx_t * tx,krw_t lti,boolean_t fatreader,boolean_t adding,zap_t ** zapp)473 zap_lockdir_impl(dmu_buf_t *db, void *tag, dmu_tx_t *tx,
474 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
475 {
476 zap_t *zap;
477 krw_t lt;
478
479 ASSERT0(db->db_offset);
480 objset_t *os = dmu_buf_get_objset(db);
481 uint64_t obj = db->db_object;
482
483 *zapp = NULL;
484
485 #ifdef ZFS_DEBUG
486 {
487 dmu_object_info_t doi;
488 dmu_object_info_from_db(db, &doi);
489 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
490 }
491 #endif
492
493 zap = dmu_buf_get_user(db);
494 if (zap == NULL) {
495 zap = mzap_open(os, obj, db);
496 if (zap == NULL) {
497 /*
498 * mzap_open() didn't like what it saw on-disk.
499 * Check for corruption!
500 */
501 return (SET_ERROR(EIO));
502 }
503 }
504
505 /*
506 * We're checking zap_ismicro without the lock held, in order to
507 * tell what type of lock we want. Once we have some sort of
508 * lock, see if it really is the right type. In practice this
509 * can only be different if it was upgraded from micro to fat,
510 * and micro wanted WRITER but fat only needs READER.
511 */
512 lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
513 rw_enter(&zap->zap_rwlock, lt);
514 if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
515 /* it was upgraded, now we only need reader */
516 ASSERT(lt == RW_WRITER);
517 ASSERT(RW_READER ==
518 (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
519 rw_downgrade(&zap->zap_rwlock);
520 lt = RW_READER;
521 }
522
523 zap->zap_objset = os;
524
525 if (lt == RW_WRITER)
526 dmu_buf_will_dirty(db, tx);
527
528 ASSERT3P(zap->zap_dbuf, ==, db);
529
530 ASSERT(!zap->zap_ismicro ||
531 zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
532 if (zap->zap_ismicro && tx && adding &&
533 zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
534 uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
535 if (newsz > MZAP_MAX_BLKSZ) {
536 dprintf("upgrading obj %llu: num_entries=%u\n",
537 obj, zap->zap_m.zap_num_entries);
538 *zapp = zap;
539 int err = mzap_upgrade(zapp, tag, tx, 0);
540 if (err != 0)
541 rw_exit(&zap->zap_rwlock);
542 return (err);
543 }
544 VERIFY0(dmu_object_set_blocksize(os, obj, newsz, 0, tx));
545 zap->zap_m.zap_num_chunks =
546 db->db_size / MZAP_ENT_LEN - 1;
547 }
548
549 *zapp = zap;
550 return (0);
551 }
552
553 static int
zap_lockdir_by_dnode(dnode_t * dn,dmu_tx_t * tx,krw_t lti,boolean_t fatreader,boolean_t adding,void * tag,zap_t ** zapp)554 zap_lockdir_by_dnode(dnode_t *dn, dmu_tx_t *tx,
555 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
556 {
557 dmu_buf_t *db;
558 int err;
559
560 err = dmu_buf_hold_by_dnode(dn, 0, tag, &db, DMU_READ_NO_PREFETCH);
561 if (err != 0) {
562 return (err);
563 }
564 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
565 if (err != 0) {
566 dmu_buf_rele(db, tag);
567 }
568 return (err);
569 }
570
571 int
zap_lockdir(objset_t * os,uint64_t obj,dmu_tx_t * tx,krw_t lti,boolean_t fatreader,boolean_t adding,void * tag,zap_t ** zapp)572 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
573 krw_t lti, boolean_t fatreader, boolean_t adding, void *tag, zap_t **zapp)
574 {
575 dmu_buf_t *db;
576 int err;
577
578 err = dmu_buf_hold(os, obj, 0, tag, &db, DMU_READ_NO_PREFETCH);
579 if (err != 0)
580 return (err);
581 err = zap_lockdir_impl(db, tag, tx, lti, fatreader, adding, zapp);
582 if (err != 0)
583 dmu_buf_rele(db, tag);
584 return (err);
585 }
586
587 void
zap_unlockdir(zap_t * zap,void * tag)588 zap_unlockdir(zap_t *zap, void *tag)
589 {
590 rw_exit(&zap->zap_rwlock);
591 dmu_buf_rele(zap->zap_dbuf, tag);
592 }
593
594 static int
mzap_upgrade(zap_t ** zapp,void * tag,dmu_tx_t * tx,zap_flags_t flags)595 mzap_upgrade(zap_t **zapp, void *tag, dmu_tx_t *tx, zap_flags_t flags)
596 {
597 mzap_phys_t *mzp;
598 int i, sz, nchunks;
599 int err = 0;
600 zap_t *zap = *zapp;
601
602 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
603
604 sz = zap->zap_dbuf->db_size;
605 mzp = zio_buf_alloc(sz);
606 bcopy(zap->zap_dbuf->db_data, mzp, sz);
607 nchunks = zap->zap_m.zap_num_chunks;
608
609 if (!flags) {
610 err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
611 1ULL << fzap_default_block_shift, 0, tx);
612 if (err) {
613 zio_buf_free(mzp, sz);
614 return (err);
615 }
616 }
617
618 dprintf("upgrading obj=%llu with %u chunks\n",
619 zap->zap_object, nchunks);
620 /* XXX destroy the avl later, so we can use the stored hash value */
621 mze_destroy(zap);
622
623 fzap_upgrade(zap, tx, flags);
624
625 for (i = 0; i < nchunks; i++) {
626 mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
627 zap_name_t *zn;
628 if (mze->mze_name[0] == 0)
629 continue;
630 dprintf("adding %s=%llu\n",
631 mze->mze_name, mze->mze_value);
632 zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
633 err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd,
634 tag, tx);
635 zap = zn->zn_zap; /* fzap_add_cd() may change zap */
636 zap_name_free(zn);
637 if (err)
638 break;
639 }
640 zio_buf_free(mzp, sz);
641 *zapp = zap;
642 return (err);
643 }
644
645 void
mzap_create_impl(objset_t * os,uint64_t obj,int normflags,zap_flags_t flags,dmu_tx_t * tx)646 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, zap_flags_t flags,
647 dmu_tx_t *tx)
648 {
649 dmu_buf_t *db;
650 mzap_phys_t *zp;
651
652 VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db, DMU_READ_NO_PREFETCH));
653
654 #ifdef ZFS_DEBUG
655 {
656 dmu_object_info_t doi;
657 dmu_object_info_from_db(db, &doi);
658 ASSERT3U(DMU_OT_BYTESWAP(doi.doi_type), ==, DMU_BSWAP_ZAP);
659 }
660 #endif
661
662 dmu_buf_will_dirty(db, tx);
663 zp = db->db_data;
664 zp->mz_block_type = ZBT_MICRO;
665 zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
666 zp->mz_normflags = normflags;
667 dmu_buf_rele(db, FTAG);
668
669 if (flags != 0) {
670 zap_t *zap;
671 /* Only fat zap supports flags; upgrade immediately. */
672 VERIFY(0 == zap_lockdir(os, obj, tx, RW_WRITER,
673 B_FALSE, B_FALSE, FTAG, &zap));
674 VERIFY3U(0, ==, mzap_upgrade(&zap, FTAG, tx, flags));
675 zap_unlockdir(zap, FTAG);
676 }
677 }
678
679 int
zap_create_claim(objset_t * os,uint64_t obj,dmu_object_type_t ot,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)680 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
681 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
682 {
683 return (zap_create_claim_norm(os, obj,
684 0, ot, bonustype, bonuslen, tx));
685 }
686
687 int
zap_create_claim_norm(objset_t * os,uint64_t obj,int normflags,dmu_object_type_t ot,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)688 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
689 dmu_object_type_t ot,
690 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
691 {
692 int err;
693
694 err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
695 if (err != 0)
696 return (err);
697 mzap_create_impl(os, obj, normflags, 0, tx);
698 return (0);
699 }
700
701 uint64_t
zap_create(objset_t * os,dmu_object_type_t ot,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)702 zap_create(objset_t *os, dmu_object_type_t ot,
703 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
704 {
705 return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
706 }
707
708 uint64_t
zap_create_norm(objset_t * os,int normflags,dmu_object_type_t ot,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)709 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
710 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
711 {
712 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
713
714 mzap_create_impl(os, obj, normflags, 0, tx);
715 return (obj);
716 }
717
718 uint64_t
zap_create_flags(objset_t * os,int normflags,zap_flags_t flags,dmu_object_type_t ot,int leaf_blockshift,int indirect_blockshift,dmu_object_type_t bonustype,int bonuslen,dmu_tx_t * tx)719 zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
720 dmu_object_type_t ot, int leaf_blockshift, int indirect_blockshift,
721 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
722 {
723 uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
724
725 ASSERT(leaf_blockshift >= SPA_MINBLOCKSHIFT &&
726 leaf_blockshift <= SPA_OLD_MAXBLOCKSHIFT &&
727 indirect_blockshift >= SPA_MINBLOCKSHIFT &&
728 indirect_blockshift <= SPA_OLD_MAXBLOCKSHIFT);
729
730 VERIFY(dmu_object_set_blocksize(os, obj,
731 1ULL << leaf_blockshift, indirect_blockshift, tx) == 0);
732
733 mzap_create_impl(os, obj, normflags, flags, tx);
734 return (obj);
735 }
736
737 int
zap_destroy(objset_t * os,uint64_t zapobj,dmu_tx_t * tx)738 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
739 {
740 /*
741 * dmu_object_free will free the object number and free the
742 * data. Freeing the data will cause our pageout function to be
743 * called, which will destroy our data (zap_leaf_t's and zap_t).
744 */
745
746 return (dmu_object_free(os, zapobj, tx));
747 }
748
749 void
zap_evict_sync(void * dbu)750 zap_evict_sync(void *dbu)
751 {
752 zap_t *zap = dbu;
753
754 rw_destroy(&zap->zap_rwlock);
755
756 if (zap->zap_ismicro)
757 mze_destroy(zap);
758 else
759 mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
760
761 kmem_free(zap, sizeof (zap_t));
762 }
763
764 int
zap_count(objset_t * os,uint64_t zapobj,uint64_t * count)765 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
766 {
767 zap_t *zap;
768 int err;
769
770 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
771 if (err)
772 return (err);
773 if (!zap->zap_ismicro) {
774 err = fzap_count(zap, count);
775 } else {
776 *count = zap->zap_m.zap_num_entries;
777 }
778 zap_unlockdir(zap, FTAG);
779 return (err);
780 }
781
782 /*
783 * zn may be NULL; if not specified, it will be computed if needed.
784 * See also the comment above zap_entry_normalization_conflict().
785 */
786 static boolean_t
mzap_normalization_conflict(zap_t * zap,zap_name_t * zn,mzap_ent_t * mze)787 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
788 {
789 mzap_ent_t *other;
790 int direction = AVL_BEFORE;
791 boolean_t allocdzn = B_FALSE;
792
793 if (zap->zap_normflags == 0)
794 return (B_FALSE);
795
796 again:
797 for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
798 other && other->mze_hash == mze->mze_hash;
799 other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
800
801 if (zn == NULL) {
802 zn = zap_name_alloc(zap, MZE_PHYS(zap, mze)->mze_name,
803 MT_FIRST);
804 allocdzn = B_TRUE;
805 }
806 if (zap_match(zn, MZE_PHYS(zap, other)->mze_name)) {
807 if (allocdzn)
808 zap_name_free(zn);
809 return (B_TRUE);
810 }
811 }
812
813 if (direction == AVL_BEFORE) {
814 direction = AVL_AFTER;
815 goto again;
816 }
817
818 if (allocdzn)
819 zap_name_free(zn);
820 return (B_FALSE);
821 }
822
823 /*
824 * Routines for manipulating attributes.
825 */
826
827 int
zap_lookup(objset_t * os,uint64_t zapobj,const char * name,uint64_t integer_size,uint64_t num_integers,void * buf)828 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
829 uint64_t integer_size, uint64_t num_integers, void *buf)
830 {
831 return (zap_lookup_norm(os, zapobj, name, integer_size,
832 num_integers, buf, MT_EXACT, NULL, 0, NULL));
833 }
834
835 static int
zap_lookup_impl(zap_t * zap,const char * name,uint64_t integer_size,uint64_t num_integers,void * buf,matchtype_t mt,char * realname,int rn_len,boolean_t * ncp)836 zap_lookup_impl(zap_t *zap, const char *name,
837 uint64_t integer_size, uint64_t num_integers, void *buf,
838 matchtype_t mt, char *realname, int rn_len,
839 boolean_t *ncp)
840 {
841 int err = 0;
842 mzap_ent_t *mze;
843 zap_name_t *zn;
844
845 zn = zap_name_alloc(zap, name, mt);
846 if (zn == NULL)
847 return (SET_ERROR(ENOTSUP));
848
849 if (!zap->zap_ismicro) {
850 err = fzap_lookup(zn, integer_size, num_integers, buf,
851 realname, rn_len, ncp);
852 } else {
853 mze = mze_find(zn);
854 if (mze == NULL) {
855 err = SET_ERROR(ENOENT);
856 } else {
857 if (num_integers < 1) {
858 err = SET_ERROR(EOVERFLOW);
859 } else if (integer_size != 8) {
860 err = SET_ERROR(EINVAL);
861 } else {
862 *(uint64_t *)buf =
863 MZE_PHYS(zap, mze)->mze_value;
864 if (realname != NULL)
865 (void) strlcpy(realname,
866 MZE_PHYS(zap, mze)->mze_name, rn_len);
867 if (ncp) {
868 *ncp = mzap_normalization_conflict(zap,
869 zn, mze);
870 }
871 }
872 }
873 }
874 zap_name_free(zn);
875 return (err);
876 }
877
878 int
zap_lookup_norm(objset_t * os,uint64_t zapobj,const char * name,uint64_t integer_size,uint64_t num_integers,void * buf,matchtype_t mt,char * realname,int rn_len,boolean_t * ncp)879 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
880 uint64_t integer_size, uint64_t num_integers, void *buf,
881 matchtype_t mt, char *realname, int rn_len,
882 boolean_t *ncp)
883 {
884 zap_t *zap;
885 int err;
886
887 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
888 if (err != 0)
889 return (err);
890 err = zap_lookup_impl(zap, name, integer_size,
891 num_integers, buf, mt, realname, rn_len, ncp);
892 zap_unlockdir(zap, FTAG);
893 return (err);
894 }
895
896 int
zap_lookup_by_dnode(dnode_t * dn,const char * name,uint64_t integer_size,uint64_t num_integers,void * buf)897 zap_lookup_by_dnode(dnode_t *dn, const char *name,
898 uint64_t integer_size, uint64_t num_integers, void *buf)
899 {
900 return (zap_lookup_norm_by_dnode(dn, name, integer_size,
901 num_integers, buf, MT_EXACT, NULL, 0, NULL));
902 }
903
904 int
zap_lookup_norm_by_dnode(dnode_t * dn,const char * name,uint64_t integer_size,uint64_t num_integers,void * buf,matchtype_t mt,char * realname,int rn_len,boolean_t * ncp)905 zap_lookup_norm_by_dnode(dnode_t *dn, const char *name,
906 uint64_t integer_size, uint64_t num_integers, void *buf,
907 matchtype_t mt, char *realname, int rn_len,
908 boolean_t *ncp)
909 {
910 zap_t *zap;
911 int err;
912
913 err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
914 FTAG, &zap);
915 if (err != 0)
916 return (err);
917 err = zap_lookup_impl(zap, name, integer_size,
918 num_integers, buf, mt, realname, rn_len, ncp);
919 zap_unlockdir(zap, FTAG);
920 return (err);
921 }
922
923 int
zap_prefetch_uint64(objset_t * os,uint64_t zapobj,const uint64_t * key,int key_numints)924 zap_prefetch_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
925 int key_numints)
926 {
927 zap_t *zap;
928 int err;
929 zap_name_t *zn;
930
931 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
932 if (err)
933 return (err);
934 zn = zap_name_alloc_uint64(zap, key, key_numints);
935 if (zn == NULL) {
936 zap_unlockdir(zap, FTAG);
937 return (SET_ERROR(ENOTSUP));
938 }
939
940 fzap_prefetch(zn);
941 zap_name_free(zn);
942 zap_unlockdir(zap, FTAG);
943 return (err);
944 }
945
946 int
zap_lookup_uint64(objset_t * os,uint64_t zapobj,const uint64_t * key,int key_numints,uint64_t integer_size,uint64_t num_integers,void * buf)947 zap_lookup_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
948 int key_numints, uint64_t integer_size, uint64_t num_integers, void *buf)
949 {
950 zap_t *zap;
951 int err;
952 zap_name_t *zn;
953
954 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
955 if (err)
956 return (err);
957 zn = zap_name_alloc_uint64(zap, key, key_numints);
958 if (zn == NULL) {
959 zap_unlockdir(zap, FTAG);
960 return (SET_ERROR(ENOTSUP));
961 }
962
963 err = fzap_lookup(zn, integer_size, num_integers, buf,
964 NULL, 0, NULL);
965 zap_name_free(zn);
966 zap_unlockdir(zap, FTAG);
967 return (err);
968 }
969
970 int
zap_contains(objset_t * os,uint64_t zapobj,const char * name)971 zap_contains(objset_t *os, uint64_t zapobj, const char *name)
972 {
973 int err = zap_lookup_norm(os, zapobj, name, 0,
974 0, NULL, MT_EXACT, NULL, 0, NULL);
975 if (err == EOVERFLOW || err == EINVAL)
976 err = 0; /* found, but skipped reading the value */
977 return (err);
978 }
979
980 int
zap_length(objset_t * os,uint64_t zapobj,const char * name,uint64_t * integer_size,uint64_t * num_integers)981 zap_length(objset_t *os, uint64_t zapobj, const char *name,
982 uint64_t *integer_size, uint64_t *num_integers)
983 {
984 zap_t *zap;
985 int err;
986 mzap_ent_t *mze;
987 zap_name_t *zn;
988
989 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
990 if (err)
991 return (err);
992 zn = zap_name_alloc(zap, name, MT_EXACT);
993 if (zn == NULL) {
994 zap_unlockdir(zap, FTAG);
995 return (SET_ERROR(ENOTSUP));
996 }
997 if (!zap->zap_ismicro) {
998 err = fzap_length(zn, integer_size, num_integers);
999 } else {
1000 mze = mze_find(zn);
1001 if (mze == NULL) {
1002 err = SET_ERROR(ENOENT);
1003 } else {
1004 if (integer_size)
1005 *integer_size = 8;
1006 if (num_integers)
1007 *num_integers = 1;
1008 }
1009 }
1010 zap_name_free(zn);
1011 zap_unlockdir(zap, FTAG);
1012 return (err);
1013 }
1014
1015 int
zap_length_uint64(objset_t * os,uint64_t zapobj,const uint64_t * key,int key_numints,uint64_t * integer_size,uint64_t * num_integers)1016 zap_length_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1017 int key_numints, uint64_t *integer_size, uint64_t *num_integers)
1018 {
1019 zap_t *zap;
1020 int err;
1021 zap_name_t *zn;
1022
1023 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1024 if (err)
1025 return (err);
1026 zn = zap_name_alloc_uint64(zap, key, key_numints);
1027 if (zn == NULL) {
1028 zap_unlockdir(zap, FTAG);
1029 return (SET_ERROR(ENOTSUP));
1030 }
1031 err = fzap_length(zn, integer_size, num_integers);
1032 zap_name_free(zn);
1033 zap_unlockdir(zap, FTAG);
1034 return (err);
1035 }
1036
1037 static void
mzap_addent(zap_name_t * zn,uint64_t value)1038 mzap_addent(zap_name_t *zn, uint64_t value)
1039 {
1040 int i;
1041 zap_t *zap = zn->zn_zap;
1042 int start = zap->zap_m.zap_alloc_next;
1043 uint32_t cd;
1044
1045 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
1046
1047 #ifdef ZFS_DEBUG
1048 for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
1049 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1050 ASSERT(strcmp(zn->zn_key_orig, mze->mze_name) != 0);
1051 }
1052 #endif
1053
1054 cd = mze_find_unused_cd(zap, zn->zn_hash);
1055 /* given the limited size of the microzap, this can't happen */
1056 ASSERT(cd < zap_maxcd(zap));
1057
1058 again:
1059 for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
1060 mzap_ent_phys_t *mze = &zap_m_phys(zap)->mz_chunk[i];
1061 if (mze->mze_name[0] == 0) {
1062 mze->mze_value = value;
1063 mze->mze_cd = cd;
1064 (void) strcpy(mze->mze_name, zn->zn_key_orig);
1065 zap->zap_m.zap_num_entries++;
1066 zap->zap_m.zap_alloc_next = i+1;
1067 if (zap->zap_m.zap_alloc_next ==
1068 zap->zap_m.zap_num_chunks)
1069 zap->zap_m.zap_alloc_next = 0;
1070 VERIFY(0 == mze_insert(zap, i, zn->zn_hash));
1071 return;
1072 }
1073 }
1074 if (start != 0) {
1075 start = 0;
1076 goto again;
1077 }
1078 ASSERT(!"out of entries!");
1079 }
1080
1081 int
zap_add(objset_t * os,uint64_t zapobj,const char * key,int integer_size,uint64_t num_integers,const void * val,dmu_tx_t * tx)1082 zap_add(objset_t *os, uint64_t zapobj, const char *key,
1083 int integer_size, uint64_t num_integers,
1084 const void *val, dmu_tx_t *tx)
1085 {
1086 zap_t *zap;
1087 int err;
1088 mzap_ent_t *mze;
1089 const uint64_t *intval = val;
1090 zap_name_t *zn;
1091
1092 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1093 if (err)
1094 return (err);
1095 zn = zap_name_alloc(zap, key, MT_EXACT);
1096 if (zn == NULL) {
1097 zap_unlockdir(zap, FTAG);
1098 return (SET_ERROR(ENOTSUP));
1099 }
1100 if (!zap->zap_ismicro) {
1101 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1102 zap = zn->zn_zap; /* fzap_add() may change zap */
1103 } else if (integer_size != 8 || num_integers != 1 ||
1104 strlen(key) >= MZAP_NAME_LEN) {
1105 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1106 if (err == 0) {
1107 err = fzap_add(zn, integer_size, num_integers, val,
1108 FTAG, tx);
1109 }
1110 zap = zn->zn_zap; /* fzap_add() may change zap */
1111 } else {
1112 mze = mze_find(zn);
1113 if (mze != NULL) {
1114 err = SET_ERROR(EEXIST);
1115 } else {
1116 mzap_addent(zn, *intval);
1117 }
1118 }
1119 ASSERT(zap == zn->zn_zap);
1120 zap_name_free(zn);
1121 if (zap != NULL) /* may be NULL if fzap_add() failed */
1122 zap_unlockdir(zap, FTAG);
1123 return (err);
1124 }
1125
1126 int
zap_add_uint64(objset_t * os,uint64_t zapobj,const uint64_t * key,int key_numints,int integer_size,uint64_t num_integers,const void * val,dmu_tx_t * tx)1127 zap_add_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1128 int key_numints, int integer_size, uint64_t num_integers,
1129 const void *val, dmu_tx_t *tx)
1130 {
1131 zap_t *zap;
1132 int err;
1133 zap_name_t *zn;
1134
1135 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1136 if (err)
1137 return (err);
1138 zn = zap_name_alloc_uint64(zap, key, key_numints);
1139 if (zn == NULL) {
1140 zap_unlockdir(zap, FTAG);
1141 return (SET_ERROR(ENOTSUP));
1142 }
1143 err = fzap_add(zn, integer_size, num_integers, val, FTAG, tx);
1144 zap = zn->zn_zap; /* fzap_add() may change zap */
1145 zap_name_free(zn);
1146 if (zap != NULL) /* may be NULL if fzap_add() failed */
1147 zap_unlockdir(zap, FTAG);
1148 return (err);
1149 }
1150
1151 int
zap_update(objset_t * os,uint64_t zapobj,const char * name,int integer_size,uint64_t num_integers,const void * val,dmu_tx_t * tx)1152 zap_update(objset_t *os, uint64_t zapobj, const char *name,
1153 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1154 {
1155 zap_t *zap;
1156 mzap_ent_t *mze;
1157 uint64_t oldval;
1158 const uint64_t *intval = val;
1159 zap_name_t *zn;
1160 int err;
1161
1162 #ifdef ZFS_DEBUG
1163 /*
1164 * If there is an old value, it shouldn't change across the
1165 * lockdir (eg, due to bprewrite's xlation).
1166 */
1167 if (integer_size == 8 && num_integers == 1)
1168 (void) zap_lookup(os, zapobj, name, 8, 1, &oldval);
1169 #endif
1170
1171 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1172 if (err)
1173 return (err);
1174 zn = zap_name_alloc(zap, name, MT_EXACT);
1175 if (zn == NULL) {
1176 zap_unlockdir(zap, FTAG);
1177 return (SET_ERROR(ENOTSUP));
1178 }
1179 if (!zap->zap_ismicro) {
1180 err = fzap_update(zn, integer_size, num_integers, val,
1181 FTAG, tx);
1182 zap = zn->zn_zap; /* fzap_update() may change zap */
1183 } else if (integer_size != 8 || num_integers != 1 ||
1184 strlen(name) >= MZAP_NAME_LEN) {
1185 dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
1186 zapobj, integer_size, num_integers, name);
1187 err = mzap_upgrade(&zn->zn_zap, FTAG, tx, 0);
1188 if (err == 0) {
1189 err = fzap_update(zn, integer_size, num_integers,
1190 val, FTAG, tx);
1191 }
1192 zap = zn->zn_zap; /* fzap_update() may change zap */
1193 } else {
1194 mze = mze_find(zn);
1195 if (mze != NULL) {
1196 ASSERT3U(MZE_PHYS(zap, mze)->mze_value, ==, oldval);
1197 MZE_PHYS(zap, mze)->mze_value = *intval;
1198 } else {
1199 mzap_addent(zn, *intval);
1200 }
1201 }
1202 ASSERT(zap == zn->zn_zap);
1203 zap_name_free(zn);
1204 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1205 zap_unlockdir(zap, FTAG);
1206 return (err);
1207 }
1208
1209 int
zap_update_uint64(objset_t * os,uint64_t zapobj,const uint64_t * key,int key_numints,int integer_size,uint64_t num_integers,const void * val,dmu_tx_t * tx)1210 zap_update_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1211 int key_numints,
1212 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
1213 {
1214 zap_t *zap;
1215 zap_name_t *zn;
1216 int err;
1217
1218 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, FTAG, &zap);
1219 if (err)
1220 return (err);
1221 zn = zap_name_alloc_uint64(zap, key, key_numints);
1222 if (zn == NULL) {
1223 zap_unlockdir(zap, FTAG);
1224 return (SET_ERROR(ENOTSUP));
1225 }
1226 err = fzap_update(zn, integer_size, num_integers, val, FTAG, tx);
1227 zap = zn->zn_zap; /* fzap_update() may change zap */
1228 zap_name_free(zn);
1229 if (zap != NULL) /* may be NULL if fzap_upgrade() failed */
1230 zap_unlockdir(zap, FTAG);
1231 return (err);
1232 }
1233
1234 int
zap_remove(objset_t * os,uint64_t zapobj,const char * name,dmu_tx_t * tx)1235 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
1236 {
1237 return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
1238 }
1239
1240 int
zap_remove_norm(objset_t * os,uint64_t zapobj,const char * name,matchtype_t mt,dmu_tx_t * tx)1241 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
1242 matchtype_t mt, dmu_tx_t *tx)
1243 {
1244 zap_t *zap;
1245 int err;
1246 mzap_ent_t *mze;
1247 zap_name_t *zn;
1248
1249 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1250 if (err)
1251 return (err);
1252 zn = zap_name_alloc(zap, name, mt);
1253 if (zn == NULL) {
1254 zap_unlockdir(zap, FTAG);
1255 return (SET_ERROR(ENOTSUP));
1256 }
1257 if (!zap->zap_ismicro) {
1258 err = fzap_remove(zn, tx);
1259 } else {
1260 mze = mze_find(zn);
1261 if (mze == NULL) {
1262 err = SET_ERROR(ENOENT);
1263 } else {
1264 zap->zap_m.zap_num_entries--;
1265 bzero(&zap_m_phys(zap)->mz_chunk[mze->mze_chunkid],
1266 sizeof (mzap_ent_phys_t));
1267 mze_remove(zap, mze);
1268 }
1269 }
1270 zap_name_free(zn);
1271 zap_unlockdir(zap, FTAG);
1272 return (err);
1273 }
1274
1275 int
zap_remove_uint64(objset_t * os,uint64_t zapobj,const uint64_t * key,int key_numints,dmu_tx_t * tx)1276 zap_remove_uint64(objset_t *os, uint64_t zapobj, const uint64_t *key,
1277 int key_numints, dmu_tx_t *tx)
1278 {
1279 zap_t *zap;
1280 int err;
1281 zap_name_t *zn;
1282
1283 err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, FTAG, &zap);
1284 if (err)
1285 return (err);
1286 zn = zap_name_alloc_uint64(zap, key, key_numints);
1287 if (zn == NULL) {
1288 zap_unlockdir(zap, FTAG);
1289 return (SET_ERROR(ENOTSUP));
1290 }
1291 err = fzap_remove(zn, tx);
1292 zap_name_free(zn);
1293 zap_unlockdir(zap, FTAG);
1294 return (err);
1295 }
1296
1297 /*
1298 * Routines for iterating over the attributes.
1299 */
1300
1301 void
zap_cursor_init_serialized(zap_cursor_t * zc,objset_t * os,uint64_t zapobj,uint64_t serialized)1302 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
1303 uint64_t serialized)
1304 {
1305 zc->zc_objset = os;
1306 zc->zc_zap = NULL;
1307 zc->zc_leaf = NULL;
1308 zc->zc_zapobj = zapobj;
1309 zc->zc_serialized = serialized;
1310 zc->zc_hash = 0;
1311 zc->zc_cd = 0;
1312 }
1313
1314 void
zap_cursor_init(zap_cursor_t * zc,objset_t * os,uint64_t zapobj)1315 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
1316 {
1317 zap_cursor_init_serialized(zc, os, zapobj, 0);
1318 }
1319
1320 void
zap_cursor_fini(zap_cursor_t * zc)1321 zap_cursor_fini(zap_cursor_t *zc)
1322 {
1323 if (zc->zc_zap) {
1324 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1325 zap_unlockdir(zc->zc_zap, NULL);
1326 zc->zc_zap = NULL;
1327 }
1328 if (zc->zc_leaf) {
1329 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1330 zap_put_leaf(zc->zc_leaf);
1331 zc->zc_leaf = NULL;
1332 }
1333 zc->zc_objset = NULL;
1334 }
1335
1336 uint64_t
zap_cursor_serialize(zap_cursor_t * zc)1337 zap_cursor_serialize(zap_cursor_t *zc)
1338 {
1339 if (zc->zc_hash == -1ULL)
1340 return (-1ULL);
1341 if (zc->zc_zap == NULL)
1342 return (zc->zc_serialized);
1343 ASSERT((zc->zc_hash & zap_maxcd(zc->zc_zap)) == 0);
1344 ASSERT(zc->zc_cd < zap_maxcd(zc->zc_zap));
1345
1346 /*
1347 * We want to keep the high 32 bits of the cursor zero if we can, so
1348 * that 32-bit programs can access this. So usually use a small
1349 * (28-bit) hash value so we can fit 4 bits of cd into the low 32-bits
1350 * of the cursor.
1351 *
1352 * [ collision differentiator | zap_hashbits()-bit hash value ]
1353 */
1354 return ((zc->zc_hash >> (64 - zap_hashbits(zc->zc_zap))) |
1355 ((uint64_t)zc->zc_cd << zap_hashbits(zc->zc_zap)));
1356 }
1357
1358 int
zap_cursor_retrieve(zap_cursor_t * zc,zap_attribute_t * za)1359 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
1360 {
1361 int err;
1362 avl_index_t idx;
1363 mzap_ent_t mze_tofind;
1364 mzap_ent_t *mze;
1365
1366 if (zc->zc_hash == -1ULL)
1367 return (SET_ERROR(ENOENT));
1368
1369 if (zc->zc_zap == NULL) {
1370 int hb;
1371 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1372 RW_READER, TRUE, FALSE, NULL, &zc->zc_zap);
1373 if (err)
1374 return (err);
1375
1376 /*
1377 * To support zap_cursor_init_serialized, advance, retrieve,
1378 * we must add to the existing zc_cd, which may already
1379 * be 1 due to the zap_cursor_advance.
1380 */
1381 ASSERT(zc->zc_hash == 0);
1382 hb = zap_hashbits(zc->zc_zap);
1383 zc->zc_hash = zc->zc_serialized << (64 - hb);
1384 zc->zc_cd += zc->zc_serialized >> hb;
1385 if (zc->zc_cd >= zap_maxcd(zc->zc_zap)) /* corrupt serialized */
1386 zc->zc_cd = 0;
1387 } else {
1388 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1389 }
1390 if (!zc->zc_zap->zap_ismicro) {
1391 err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1392 } else {
1393 mze_tofind.mze_hash = zc->zc_hash;
1394 mze_tofind.mze_cd = zc->zc_cd;
1395
1396 mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1397 if (mze == NULL) {
1398 mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1399 idx, AVL_AFTER);
1400 }
1401 if (mze) {
1402 mzap_ent_phys_t *mzep = MZE_PHYS(zc->zc_zap, mze);
1403 ASSERT3U(mze->mze_cd, ==, mzep->mze_cd);
1404 za->za_normalization_conflict =
1405 mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1406 za->za_integer_length = 8;
1407 za->za_num_integers = 1;
1408 za->za_first_integer = mzep->mze_value;
1409 (void) strcpy(za->za_name, mzep->mze_name);
1410 zc->zc_hash = mze->mze_hash;
1411 zc->zc_cd = mze->mze_cd;
1412 err = 0;
1413 } else {
1414 zc->zc_hash = -1ULL;
1415 err = SET_ERROR(ENOENT);
1416 }
1417 }
1418 rw_exit(&zc->zc_zap->zap_rwlock);
1419 return (err);
1420 }
1421
1422 void
zap_cursor_advance(zap_cursor_t * zc)1423 zap_cursor_advance(zap_cursor_t *zc)
1424 {
1425 if (zc->zc_hash == -1ULL)
1426 return;
1427 zc->zc_cd++;
1428 }
1429
1430 int
zap_cursor_move_to_key(zap_cursor_t * zc,const char * name,matchtype_t mt)1431 zap_cursor_move_to_key(zap_cursor_t *zc, const char *name, matchtype_t mt)
1432 {
1433 int err = 0;
1434 mzap_ent_t *mze;
1435 zap_name_t *zn;
1436
1437 if (zc->zc_zap == NULL) {
1438 err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
1439 RW_READER, TRUE, FALSE, FTAG, &zc->zc_zap);
1440 if (err)
1441 return (err);
1442 } else {
1443 rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
1444 }
1445
1446 zn = zap_name_alloc(zc->zc_zap, name, mt);
1447 if (zn == NULL) {
1448 rw_exit(&zc->zc_zap->zap_rwlock);
1449 return (SET_ERROR(ENOTSUP));
1450 }
1451
1452 if (!zc->zc_zap->zap_ismicro) {
1453 err = fzap_cursor_move_to_key(zc, zn);
1454 } else {
1455 mze = mze_find(zn);
1456 if (mze == NULL) {
1457 err = SET_ERROR(ENOENT);
1458 goto out;
1459 }
1460 zc->zc_hash = mze->mze_hash;
1461 zc->zc_cd = mze->mze_cd;
1462 }
1463
1464 out:
1465 zap_name_free(zn);
1466 rw_exit(&zc->zc_zap->zap_rwlock);
1467 return (err);
1468 }
1469
1470 int
zap_get_stats(objset_t * os,uint64_t zapobj,zap_stats_t * zs)1471 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1472 {
1473 int err;
1474 zap_t *zap;
1475
1476 err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, FTAG, &zap);
1477 if (err)
1478 return (err);
1479
1480 bzero(zs, sizeof (zap_stats_t));
1481
1482 if (zap->zap_ismicro) {
1483 zs->zs_blocksize = zap->zap_dbuf->db_size;
1484 zs->zs_num_entries = zap->zap_m.zap_num_entries;
1485 zs->zs_num_blocks = 1;
1486 } else {
1487 fzap_get_stats(zap, zs);
1488 }
1489 zap_unlockdir(zap, FTAG);
1490 return (0);
1491 }
1492
1493 int
zap_count_write_by_dnode(dnode_t * dn,const char * name,int add,refcount_t * towrite,refcount_t * tooverwrite)1494 zap_count_write_by_dnode(dnode_t *dn, const char *name, int add,
1495 refcount_t *towrite, refcount_t *tooverwrite)
1496 {
1497 zap_t *zap;
1498 int err = 0;
1499
1500 /*
1501 * Since, we don't have a name, we cannot figure out which blocks will
1502 * be affected in this operation. So, account for the worst case :
1503 * - 3 blocks overwritten: target leaf, ptrtbl block, header block
1504 * - 4 new blocks written if adding:
1505 * - 2 blocks for possibly split leaves,
1506 * - 2 grown ptrtbl blocks
1507 *
1508 * This also accommodates the case where an add operation to a fairly
1509 * large microzap results in a promotion to fatzap.
1510 */
1511 if (name == NULL) {
1512 (void) refcount_add_many(towrite,
1513 (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE, FTAG);
1514 return (err);
1515 }
1516
1517 /*
1518 * We lock the zap with adding == FALSE. Because, if we pass
1519 * the actual value of add, it could trigger a mzap_upgrade().
1520 * At present we are just evaluating the possibility of this operation
1521 * and hence we do not want to trigger an upgrade.
1522 */
1523 err = zap_lockdir_by_dnode(dn, NULL, RW_READER, TRUE, FALSE,
1524 FTAG, &zap);
1525 if (err != 0)
1526 return (err);
1527
1528 if (!zap->zap_ismicro) {
1529 zap_name_t *zn = zap_name_alloc(zap, name, MT_EXACT);
1530 if (zn) {
1531 err = fzap_count_write(zn, add, towrite,
1532 tooverwrite);
1533 zap_name_free(zn);
1534 } else {
1535 /*
1536 * We treat this case as similar to (name == NULL)
1537 */
1538 (void) refcount_add_many(towrite,
1539 (3 + (add ? 4 : 0)) * SPA_OLD_MAXBLOCKSIZE, FTAG);
1540 }
1541 } else {
1542 /*
1543 * We are here if (name != NULL) and this is a micro-zap.
1544 * We account for the header block depending on whether it
1545 * is freeable.
1546 *
1547 * Incase of an add-operation it is hard to find out
1548 * if this add will promote this microzap to fatzap.
1549 * Hence, we consider the worst case and account for the
1550 * blocks assuming this microzap would be promoted to a
1551 * fatzap.
1552 *
1553 * 1 block overwritten : header block
1554 * 4 new blocks written : 2 new split leaf, 2 grown
1555 * ptrtbl blocks
1556 */
1557 if (dmu_buf_freeable(zap->zap_dbuf)) {
1558 (void) refcount_add_many(tooverwrite,
1559 MZAP_MAX_BLKSZ, FTAG);
1560 } else {
1561 (void) refcount_add_many(towrite,
1562 MZAP_MAX_BLKSZ, FTAG);
1563 }
1564
1565 if (add) {
1566 (void) refcount_add_many(towrite,
1567 4 * MZAP_MAX_BLKSZ, FTAG);
1568 }
1569 }
1570
1571 zap_unlockdir(zap, FTAG);
1572 return (err);
1573 }
1574