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