xref: /titanic_44/usr/src/uts/common/fs/zfs/zap.c (revision 8e22b25c6d6e25bc389a391b5a406e206fcfeec1)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25  */
26 
27 /*
28  * This file contains the top half of the zfs directory structure
29  * implementation. The bottom half is in zap_leaf.c.
30  *
31  * The zdir is an extendable hash data structure. There is a table of
32  * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are
33  * each a constant size and hold a variable number of directory entries.
34  * The buckets (aka "leaf nodes") are implemented in zap_leaf.c.
35  *
36  * The pointer table holds a power of 2 number of pointers.
37  * (1<<zap_t->zd_data->zd_phys->zd_prefix_len).  The bucket pointed to
38  * by the pointer at index i in the table holds entries whose hash value
39  * has a zd_prefix_len - bit prefix
40  */
41 
42 #include <sys/spa.h>
43 #include <sys/dmu.h>
44 #include <sys/zfs_context.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/fs/zfs.h>
47 #include <sys/zap.h>
48 #include <sys/refcount.h>
49 #include <sys/zap_impl.h>
50 #include <sys/zap_leaf.h>
51 
52 int fzap_default_block_shift = 14; /* 16k blocksize */
53 
54 extern inline zap_phys_t *zap_f_phys(zap_t *zap);
55 
56 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
57 
58 void
59 fzap_byteswap(void *vbuf, size_t size)
60 {
61 	uint64_t block_type;
62 
63 	block_type = *(uint64_t *)vbuf;
64 
65 	if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF))
66 		zap_leaf_byteswap(vbuf, size);
67 	else {
68 		/* it's a ptrtbl block */
69 		byteswap_uint64_array(vbuf, size);
70 	}
71 }
72 
73 void
74 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
75 {
76 	dmu_buf_t *db;
77 	zap_leaf_t *l;
78 	int i;
79 	zap_phys_t *zp;
80 
81 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
82 	zap->zap_ismicro = FALSE;
83 
84 	zap->zap_dbu.dbu_evict_func_sync = zap_evict_sync;
85 	zap->zap_dbu.dbu_evict_func_async = NULL;
86 
87 	mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
88 	zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;
89 
90 	zp = zap_f_phys(zap);
91 	/*
92 	 * explicitly zero it since it might be coming from an
93 	 * initialized microzap
94 	 */
95 	bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
96 	zp->zap_block_type = ZBT_HEADER;
97 	zp->zap_magic = ZAP_MAGIC;
98 
99 	zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
100 
101 	zp->zap_freeblk = 2;		/* block 1 will be the first leaf */
102 	zp->zap_num_leafs = 1;
103 	zp->zap_num_entries = 0;
104 	zp->zap_salt = zap->zap_salt;
105 	zp->zap_normflags = zap->zap_normflags;
106 	zp->zap_flags = flags;
107 
108 	/* block 1 will be the first leaf */
109 	for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
110 		ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
111 
112 	/*
113 	 * set up block 1 - the first leaf
114 	 */
115 	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
116 	    1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
117 	dmu_buf_will_dirty(db, tx);
118 
119 	l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
120 	l->l_dbuf = db;
121 
122 	zap_leaf_init(l, zp->zap_normflags != 0);
123 
124 	kmem_free(l, sizeof (zap_leaf_t));
125 	dmu_buf_rele(db, FTAG);
126 }
127 
128 static int
129 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
130 {
131 	if (RW_WRITE_HELD(&zap->zap_rwlock))
132 		return (1);
133 	if (rw_tryupgrade(&zap->zap_rwlock)) {
134 		dmu_buf_will_dirty(zap->zap_dbuf, tx);
135 		return (1);
136 	}
137 	return (0);
138 }
139 
140 /*
141  * Generic routines for dealing with the pointer & cookie tables.
142  */
143 
144 static int
145 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
146     void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
147     dmu_tx_t *tx)
148 {
149 	uint64_t b, newblk;
150 	dmu_buf_t *db_old, *db_new;
151 	int err;
152 	int bs = FZAP_BLOCK_SHIFT(zap);
153 	int hepb = 1<<(bs-4);
154 	/* hepb = half the number of entries in a block */
155 
156 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
157 	ASSERT(tbl->zt_blk != 0);
158 	ASSERT(tbl->zt_numblks > 0);
159 
160 	if (tbl->zt_nextblk != 0) {
161 		newblk = tbl->zt_nextblk;
162 	} else {
163 		newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
164 		tbl->zt_nextblk = newblk;
165 		ASSERT0(tbl->zt_blks_copied);
166 		dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
167 		    tbl->zt_blk << bs, tbl->zt_numblks << bs,
168 		    ZIO_PRIORITY_SYNC_READ);
169 	}
170 
171 	/*
172 	 * Copy the ptrtbl from the old to new location.
173 	 */
174 
175 	b = tbl->zt_blks_copied;
176 	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
177 	    (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
178 	if (err)
179 		return (err);
180 
181 	/* first half of entries in old[b] go to new[2*b+0] */
182 	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
183 	    (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
184 	dmu_buf_will_dirty(db_new, tx);
185 	transfer_func(db_old->db_data, db_new->db_data, hepb);
186 	dmu_buf_rele(db_new, FTAG);
187 
188 	/* second half of entries in old[b] go to new[2*b+1] */
189 	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
190 	    (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
191 	dmu_buf_will_dirty(db_new, tx);
192 	transfer_func((uint64_t *)db_old->db_data + hepb,
193 	    db_new->db_data, hepb);
194 	dmu_buf_rele(db_new, FTAG);
195 
196 	dmu_buf_rele(db_old, FTAG);
197 
198 	tbl->zt_blks_copied++;
199 
200 	dprintf("copied block %llu of %llu\n",
201 	    tbl->zt_blks_copied, tbl->zt_numblks);
202 
203 	if (tbl->zt_blks_copied == tbl->zt_numblks) {
204 		(void) dmu_free_range(zap->zap_objset, zap->zap_object,
205 		    tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
206 
207 		tbl->zt_blk = newblk;
208 		tbl->zt_numblks *= 2;
209 		tbl->zt_shift++;
210 		tbl->zt_nextblk = 0;
211 		tbl->zt_blks_copied = 0;
212 
213 		dprintf("finished; numblocks now %llu (%lluk entries)\n",
214 		    tbl->zt_numblks, 1<<(tbl->zt_shift-10));
215 	}
216 
217 	return (0);
218 }
219 
220 static int
221 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
222     dmu_tx_t *tx)
223 {
224 	int err;
225 	uint64_t blk, off;
226 	int bs = FZAP_BLOCK_SHIFT(zap);
227 	dmu_buf_t *db;
228 
229 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
230 	ASSERT(tbl->zt_blk != 0);
231 
232 	dprintf("storing %llx at index %llx\n", val, idx);
233 
234 	blk = idx >> (bs-3);
235 	off = idx & ((1<<(bs-3))-1);
236 
237 	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
238 	    (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
239 	if (err)
240 		return (err);
241 	dmu_buf_will_dirty(db, tx);
242 
243 	if (tbl->zt_nextblk != 0) {
244 		uint64_t idx2 = idx * 2;
245 		uint64_t blk2 = idx2 >> (bs-3);
246 		uint64_t off2 = idx2 & ((1<<(bs-3))-1);
247 		dmu_buf_t *db2;
248 
249 		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
250 		    (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
251 		    DMU_READ_NO_PREFETCH);
252 		if (err) {
253 			dmu_buf_rele(db, FTAG);
254 			return (err);
255 		}
256 		dmu_buf_will_dirty(db2, tx);
257 		((uint64_t *)db2->db_data)[off2] = val;
258 		((uint64_t *)db2->db_data)[off2+1] = val;
259 		dmu_buf_rele(db2, FTAG);
260 	}
261 
262 	((uint64_t *)db->db_data)[off] = val;
263 	dmu_buf_rele(db, FTAG);
264 
265 	return (0);
266 }
267 
268 static int
269 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
270 {
271 	uint64_t blk, off;
272 	int err;
273 	dmu_buf_t *db;
274 	int bs = FZAP_BLOCK_SHIFT(zap);
275 
276 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
277 
278 	blk = idx >> (bs-3);
279 	off = idx & ((1<<(bs-3))-1);
280 
281 	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
282 	    (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
283 	if (err)
284 		return (err);
285 	*valp = ((uint64_t *)db->db_data)[off];
286 	dmu_buf_rele(db, FTAG);
287 
288 	if (tbl->zt_nextblk != 0) {
289 		/*
290 		 * read the nextblk for the sake of i/o error checking,
291 		 * so that zap_table_load() will catch errors for
292 		 * zap_table_store.
293 		 */
294 		blk = (idx*2) >> (bs-3);
295 
296 		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
297 		    (tbl->zt_nextblk + blk) << bs, FTAG, &db,
298 		    DMU_READ_NO_PREFETCH);
299 		if (err == 0)
300 			dmu_buf_rele(db, FTAG);
301 	}
302 	return (err);
303 }
304 
305 /*
306  * Routines for growing the ptrtbl.
307  */
308 
309 static void
310 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
311 {
312 	int i;
313 	for (i = 0; i < n; i++) {
314 		uint64_t lb = src[i];
315 		dst[2*i+0] = lb;
316 		dst[2*i+1] = lb;
317 	}
318 }
319 
320 static int
321 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
322 {
323 	/*
324 	 * The pointer table should never use more hash bits than we
325 	 * have (otherwise we'd be using useless zero bits to index it).
326 	 * If we are within 2 bits of running out, stop growing, since
327 	 * this is already an aberrant condition.
328 	 */
329 	if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
330 		return (SET_ERROR(ENOSPC));
331 
332 	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
333 		/*
334 		 * We are outgrowing the "embedded" ptrtbl (the one
335 		 * stored in the header block).  Give it its own entire
336 		 * block, which will double the size of the ptrtbl.
337 		 */
338 		uint64_t newblk;
339 		dmu_buf_t *db_new;
340 		int err;
341 
342 		ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
343 		    ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
344 		ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk);
345 
346 		newblk = zap_allocate_blocks(zap, 1);
347 		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
348 		    newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
349 		    DMU_READ_NO_PREFETCH);
350 		if (err)
351 			return (err);
352 		dmu_buf_will_dirty(db_new, tx);
353 		zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
354 		    db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
355 		dmu_buf_rele(db_new, FTAG);
356 
357 		zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk;
358 		zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1;
359 		zap_f_phys(zap)->zap_ptrtbl.zt_shift++;
360 
361 		ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
362 		    zap_f_phys(zap)->zap_ptrtbl.zt_numblks <<
363 		    (FZAP_BLOCK_SHIFT(zap)-3));
364 
365 		return (0);
366 	} else {
367 		return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl,
368 		    zap_ptrtbl_transfer, tx));
369 	}
370 }
371 
372 static void
373 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
374 {
375 	dmu_buf_will_dirty(zap->zap_dbuf, tx);
376 	mutex_enter(&zap->zap_f.zap_num_entries_mtx);
377 	ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta);
378 	zap_f_phys(zap)->zap_num_entries += delta;
379 	mutex_exit(&zap->zap_f.zap_num_entries_mtx);
380 }
381 
382 static uint64_t
383 zap_allocate_blocks(zap_t *zap, int nblocks)
384 {
385 	uint64_t newblk;
386 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
387 	newblk = zap_f_phys(zap)->zap_freeblk;
388 	zap_f_phys(zap)->zap_freeblk += nblocks;
389 	return (newblk);
390 }
391 
392 static void
393 zap_leaf_evict_sync(void *dbu)
394 {
395 	zap_leaf_t *l = dbu;
396 
397 	rw_destroy(&l->l_rwlock);
398 	kmem_free(l, sizeof (zap_leaf_t));
399 }
400 
401 static zap_leaf_t *
402 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
403 {
404 	void *winner;
405 	zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
406 
407 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
408 
409 	rw_init(&l->l_rwlock, 0, 0, 0);
410 	rw_enter(&l->l_rwlock, RW_WRITER);
411 	l->l_blkid = zap_allocate_blocks(zap, 1);
412 	l->l_dbuf = NULL;
413 
414 	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
415 	    l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
416 	    DMU_READ_NO_PREFETCH));
417 	dmu_buf_init_user(&l->l_dbu, zap_leaf_evict_sync, NULL, &l->l_dbuf);
418 	winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu);
419 	ASSERT(winner == NULL);
420 	dmu_buf_will_dirty(l->l_dbuf, tx);
421 
422 	zap_leaf_init(l, zap->zap_normflags != 0);
423 
424 	zap_f_phys(zap)->zap_num_leafs++;
425 
426 	return (l);
427 }
428 
429 int
430 fzap_count(zap_t *zap, uint64_t *count)
431 {
432 	ASSERT(!zap->zap_ismicro);
433 	mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
434 	*count = zap_f_phys(zap)->zap_num_entries;
435 	mutex_exit(&zap->zap_f.zap_num_entries_mtx);
436 	return (0);
437 }
438 
439 /*
440  * Routines for obtaining zap_leaf_t's
441  */
442 
443 void
444 zap_put_leaf(zap_leaf_t *l)
445 {
446 	rw_exit(&l->l_rwlock);
447 	dmu_buf_rele(l->l_dbuf, NULL);
448 }
449 
450 static zap_leaf_t *
451 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
452 {
453 	zap_leaf_t *l, *winner;
454 
455 	ASSERT(blkid != 0);
456 
457 	l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
458 	rw_init(&l->l_rwlock, 0, 0, 0);
459 	rw_enter(&l->l_rwlock, RW_WRITER);
460 	l->l_blkid = blkid;
461 	l->l_bs = highbit64(db->db_size) - 1;
462 	l->l_dbuf = db;
463 
464 	dmu_buf_init_user(&l->l_dbu, zap_leaf_evict_sync, NULL, &l->l_dbuf);
465 	winner = dmu_buf_set_user(db, &l->l_dbu);
466 
467 	rw_exit(&l->l_rwlock);
468 	if (winner != NULL) {
469 		/* someone else set it first */
470 		zap_leaf_evict_sync(&l->l_dbu);
471 		l = winner;
472 	}
473 
474 	/*
475 	 * lhr_pad was previously used for the next leaf in the leaf
476 	 * chain.  There should be no chained leafs (as we have removed
477 	 * support for them).
478 	 */
479 	ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1);
480 
481 	/*
482 	 * There should be more hash entries than there can be
483 	 * chunks to put in the hash table
484 	 */
485 	ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
486 
487 	/* The chunks should begin at the end of the hash table */
488 	ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
489 	    &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
490 
491 	/* The chunks should end at the end of the block */
492 	ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
493 	    (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size);
494 
495 	return (l);
496 }
497 
498 static int
499 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
500     zap_leaf_t **lp)
501 {
502 	dmu_buf_t *db;
503 	zap_leaf_t *l;
504 	int bs = FZAP_BLOCK_SHIFT(zap);
505 	int err;
506 
507 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
508 
509 	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
510 	    blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
511 	if (err)
512 		return (err);
513 
514 	ASSERT3U(db->db_object, ==, zap->zap_object);
515 	ASSERT3U(db->db_offset, ==, blkid << bs);
516 	ASSERT3U(db->db_size, ==, 1 << bs);
517 	ASSERT(blkid != 0);
518 
519 	l = dmu_buf_get_user(db);
520 
521 	if (l == NULL)
522 		l = zap_open_leaf(blkid, db);
523 
524 	rw_enter(&l->l_rwlock, lt);
525 	/*
526 	 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
527 	 * causing ASSERT below to fail.
528 	 */
529 	if (lt == RW_WRITER)
530 		dmu_buf_will_dirty(db, tx);
531 	ASSERT3U(l->l_blkid, ==, blkid);
532 	ASSERT3P(l->l_dbuf, ==, db);
533 	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
534 	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
535 
536 	*lp = l;
537 	return (0);
538 }
539 
540 static int
541 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
542 {
543 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
544 
545 	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
546 		ASSERT3U(idx, <,
547 		    (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
548 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
549 		return (0);
550 	} else {
551 		return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
552 		    idx, valp));
553 	}
554 }
555 
556 static int
557 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
558 {
559 	ASSERT(tx != NULL);
560 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
561 
562 	if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
563 		ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
564 		return (0);
565 	} else {
566 		return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
567 		    idx, blk, tx));
568 	}
569 }
570 
571 static int
572 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
573 {
574 	uint64_t idx, blk;
575 	int err;
576 
577 	ASSERT(zap->zap_dbuf == NULL ||
578 	    zap_f_phys(zap) == zap->zap_dbuf->db_data);
579 	ASSERT3U(zap_f_phys(zap)->zap_magic, ==, ZAP_MAGIC);
580 	idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
581 	err = zap_idx_to_blk(zap, idx, &blk);
582 	if (err != 0)
583 		return (err);
584 	err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
585 
586 	ASSERT(err ||
587 	    ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
588 	    zap_leaf_phys(*lp)->l_hdr.lh_prefix);
589 	return (err);
590 }
591 
592 static int
593 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx, zap_leaf_t **lp)
594 {
595 	zap_t *zap = zn->zn_zap;
596 	uint64_t hash = zn->zn_hash;
597 	zap_leaf_t *nl;
598 	int prefix_diff, i, err;
599 	uint64_t sibling;
600 	int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len;
601 
602 	ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
603 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
604 
605 	ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
606 	    zap_leaf_phys(l)->l_hdr.lh_prefix);
607 
608 	if (zap_tryupgradedir(zap, tx) == 0 ||
609 	    old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
610 		/* We failed to upgrade, or need to grow the pointer table */
611 		objset_t *os = zap->zap_objset;
612 		uint64_t object = zap->zap_object;
613 
614 		zap_put_leaf(l);
615 		zap_unlockdir(zap);
616 		err = zap_lockdir(os, object, tx, RW_WRITER,
617 		    FALSE, FALSE, &zn->zn_zap);
618 		zap = zn->zn_zap;
619 		if (err)
620 			return (err);
621 		ASSERT(!zap->zap_ismicro);
622 
623 		while (old_prefix_len ==
624 		    zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
625 			err = zap_grow_ptrtbl(zap, tx);
626 			if (err)
627 				return (err);
628 		}
629 
630 		err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
631 		if (err)
632 			return (err);
633 
634 		if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) {
635 			/* it split while our locks were down */
636 			*lp = l;
637 			return (0);
638 		}
639 	}
640 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
641 	ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
642 	ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
643 	    zap_leaf_phys(l)->l_hdr.lh_prefix);
644 
645 	prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
646 	    (old_prefix_len + 1);
647 	sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
648 
649 	/* check for i/o errors before doing zap_leaf_split */
650 	for (i = 0; i < (1ULL<<prefix_diff); i++) {
651 		uint64_t blk;
652 		err = zap_idx_to_blk(zap, sibling+i, &blk);
653 		if (err)
654 			return (err);
655 		ASSERT3U(blk, ==, l->l_blkid);
656 	}
657 
658 	nl = zap_create_leaf(zap, tx);
659 	zap_leaf_split(l, nl, zap->zap_normflags != 0);
660 
661 	/* set sibling pointers */
662 	for (i = 0; i < (1ULL << prefix_diff); i++) {
663 		err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
664 		ASSERT0(err); /* we checked for i/o errors above */
665 	}
666 
667 	if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) {
668 		/* we want the sibling */
669 		zap_put_leaf(l);
670 		*lp = nl;
671 	} else {
672 		zap_put_leaf(nl);
673 		*lp = l;
674 	}
675 
676 	return (0);
677 }
678 
679 static void
680 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx)
681 {
682 	zap_t *zap = zn->zn_zap;
683 	int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
684 	int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
685 	    zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
686 
687 	zap_put_leaf(l);
688 
689 	if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
690 		int err;
691 
692 		/*
693 		 * We are in the middle of growing the pointer table, or
694 		 * this leaf will soon make us grow it.
695 		 */
696 		if (zap_tryupgradedir(zap, tx) == 0) {
697 			objset_t *os = zap->zap_objset;
698 			uint64_t zapobj = zap->zap_object;
699 
700 			zap_unlockdir(zap);
701 			err = zap_lockdir(os, zapobj, tx,
702 			    RW_WRITER, FALSE, FALSE, &zn->zn_zap);
703 			zap = zn->zn_zap;
704 			if (err)
705 				return;
706 		}
707 
708 		/* could have finished growing while our locks were down */
709 		if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
710 			(void) zap_grow_ptrtbl(zap, tx);
711 	}
712 }
713 
714 static int
715 fzap_checkname(zap_name_t *zn)
716 {
717 	if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
718 		return (SET_ERROR(ENAMETOOLONG));
719 	return (0);
720 }
721 
722 static int
723 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
724 {
725 	/* Only integer sizes supported by C */
726 	switch (integer_size) {
727 	case 1:
728 	case 2:
729 	case 4:
730 	case 8:
731 		break;
732 	default:
733 		return (SET_ERROR(EINVAL));
734 	}
735 
736 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
737 		return (E2BIG);
738 
739 	return (0);
740 }
741 
742 static int
743 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
744 {
745 	int err;
746 
747 	if ((err = fzap_checkname(zn)) != 0)
748 		return (err);
749 	return (fzap_checksize(integer_size, num_integers));
750 }
751 
752 /*
753  * Routines for manipulating attributes.
754  */
755 int
756 fzap_lookup(zap_name_t *zn,
757     uint64_t integer_size, uint64_t num_integers, void *buf,
758     char *realname, int rn_len, boolean_t *ncp)
759 {
760 	zap_leaf_t *l;
761 	int err;
762 	zap_entry_handle_t zeh;
763 
764 	if ((err = fzap_checkname(zn)) != 0)
765 		return (err);
766 
767 	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
768 	if (err != 0)
769 		return (err);
770 	err = zap_leaf_lookup(l, zn, &zeh);
771 	if (err == 0) {
772 		if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
773 			zap_put_leaf(l);
774 			return (err);
775 		}
776 
777 		err = zap_entry_read(&zeh, integer_size, num_integers, buf);
778 		(void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
779 		if (ncp) {
780 			*ncp = zap_entry_normalization_conflict(&zeh,
781 			    zn, NULL, zn->zn_zap);
782 		}
783 	}
784 
785 	zap_put_leaf(l);
786 	return (err);
787 }
788 
789 int
790 fzap_add_cd(zap_name_t *zn,
791     uint64_t integer_size, uint64_t num_integers,
792     const void *val, uint32_t cd, dmu_tx_t *tx)
793 {
794 	zap_leaf_t *l;
795 	int err;
796 	zap_entry_handle_t zeh;
797 	zap_t *zap = zn->zn_zap;
798 
799 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
800 	ASSERT(!zap->zap_ismicro);
801 	ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
802 
803 	err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
804 	if (err != 0)
805 		return (err);
806 retry:
807 	err = zap_leaf_lookup(l, zn, &zeh);
808 	if (err == 0) {
809 		err = SET_ERROR(EEXIST);
810 		goto out;
811 	}
812 	if (err != ENOENT)
813 		goto out;
814 
815 	err = zap_entry_create(l, zn, cd,
816 	    integer_size, num_integers, val, &zeh);
817 
818 	if (err == 0) {
819 		zap_increment_num_entries(zap, 1, tx);
820 	} else if (err == EAGAIN) {
821 		err = zap_expand_leaf(zn, l, tx, &l);
822 		zap = zn->zn_zap;	/* zap_expand_leaf() may change zap */
823 		if (err == 0)
824 			goto retry;
825 	}
826 
827 out:
828 	if (zap != NULL)
829 		zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
830 	return (err);
831 }
832 
833 int
834 fzap_add(zap_name_t *zn,
835     uint64_t integer_size, uint64_t num_integers,
836     const void *val, dmu_tx_t *tx)
837 {
838 	int err = fzap_check(zn, integer_size, num_integers);
839 	if (err != 0)
840 		return (err);
841 
842 	return (fzap_add_cd(zn, integer_size, num_integers,
843 	    val, ZAP_NEED_CD, tx));
844 }
845 
846 int
847 fzap_update(zap_name_t *zn,
848     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
849 {
850 	zap_leaf_t *l;
851 	int err, create;
852 	zap_entry_handle_t zeh;
853 	zap_t *zap = zn->zn_zap;
854 
855 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
856 	err = fzap_check(zn, integer_size, num_integers);
857 	if (err != 0)
858 		return (err);
859 
860 	err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
861 	if (err != 0)
862 		return (err);
863 retry:
864 	err = zap_leaf_lookup(l, zn, &zeh);
865 	create = (err == ENOENT);
866 	ASSERT(err == 0 || err == ENOENT);
867 
868 	if (create) {
869 		err = zap_entry_create(l, zn, ZAP_NEED_CD,
870 		    integer_size, num_integers, val, &zeh);
871 		if (err == 0)
872 			zap_increment_num_entries(zap, 1, tx);
873 	} else {
874 		err = zap_entry_update(&zeh, integer_size, num_integers, val);
875 	}
876 
877 	if (err == EAGAIN) {
878 		err = zap_expand_leaf(zn, l, tx, &l);
879 		zap = zn->zn_zap;	/* zap_expand_leaf() may change zap */
880 		if (err == 0)
881 			goto retry;
882 	}
883 
884 	if (zap != NULL)
885 		zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
886 	return (err);
887 }
888 
889 int
890 fzap_length(zap_name_t *zn,
891     uint64_t *integer_size, uint64_t *num_integers)
892 {
893 	zap_leaf_t *l;
894 	int err;
895 	zap_entry_handle_t zeh;
896 
897 	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
898 	if (err != 0)
899 		return (err);
900 	err = zap_leaf_lookup(l, zn, &zeh);
901 	if (err != 0)
902 		goto out;
903 
904 	if (integer_size)
905 		*integer_size = zeh.zeh_integer_size;
906 	if (num_integers)
907 		*num_integers = zeh.zeh_num_integers;
908 out:
909 	zap_put_leaf(l);
910 	return (err);
911 }
912 
913 int
914 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
915 {
916 	zap_leaf_t *l;
917 	int err;
918 	zap_entry_handle_t zeh;
919 
920 	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
921 	if (err != 0)
922 		return (err);
923 	err = zap_leaf_lookup(l, zn, &zeh);
924 	if (err == 0) {
925 		zap_entry_remove(&zeh);
926 		zap_increment_num_entries(zn->zn_zap, -1, tx);
927 	}
928 	zap_put_leaf(l);
929 	return (err);
930 }
931 
932 void
933 fzap_prefetch(zap_name_t *zn)
934 {
935 	uint64_t idx, blk;
936 	zap_t *zap = zn->zn_zap;
937 	int bs;
938 
939 	idx = ZAP_HASH_IDX(zn->zn_hash,
940 	    zap_f_phys(zap)->zap_ptrtbl.zt_shift);
941 	if (zap_idx_to_blk(zap, idx, &blk) != 0)
942 		return;
943 	bs = FZAP_BLOCK_SHIFT(zap);
944 	dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs,
945 	    ZIO_PRIORITY_SYNC_READ);
946 }
947 
948 /*
949  * Helper functions for consumers.
950  */
951 
952 uint64_t
953 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
954     const char *name, dmu_tx_t *tx)
955 {
956 	uint64_t new_obj;
957 
958 	VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
959 	VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
960 	    tx) == 0);
961 
962 	return (new_obj);
963 }
964 
965 int
966 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
967     char *name)
968 {
969 	zap_cursor_t zc;
970 	zap_attribute_t *za;
971 	int err;
972 
973 	if (mask == 0)
974 		mask = -1ULL;
975 
976 	za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
977 	for (zap_cursor_init(&zc, os, zapobj);
978 	    (err = zap_cursor_retrieve(&zc, za)) == 0;
979 	    zap_cursor_advance(&zc)) {
980 		if ((za->za_first_integer & mask) == (value & mask)) {
981 			(void) strcpy(name, za->za_name);
982 			break;
983 		}
984 	}
985 	zap_cursor_fini(&zc);
986 	kmem_free(za, sizeof (zap_attribute_t));
987 	return (err);
988 }
989 
990 int
991 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
992 {
993 	zap_cursor_t zc;
994 	zap_attribute_t za;
995 	int err;
996 
997 	err = 0;
998 	for (zap_cursor_init(&zc, os, fromobj);
999 	    zap_cursor_retrieve(&zc, &za) == 0;
1000 	    (void) zap_cursor_advance(&zc)) {
1001 		if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1002 			err = SET_ERROR(EINVAL);
1003 			break;
1004 		}
1005 		err = zap_add(os, intoobj, za.za_name,
1006 		    8, 1, &za.za_first_integer, tx);
1007 		if (err)
1008 			break;
1009 	}
1010 	zap_cursor_fini(&zc);
1011 	return (err);
1012 }
1013 
1014 int
1015 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1016     uint64_t value, dmu_tx_t *tx)
1017 {
1018 	zap_cursor_t zc;
1019 	zap_attribute_t za;
1020 	int err;
1021 
1022 	err = 0;
1023 	for (zap_cursor_init(&zc, os, fromobj);
1024 	    zap_cursor_retrieve(&zc, &za) == 0;
1025 	    (void) zap_cursor_advance(&zc)) {
1026 		if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1027 			err = SET_ERROR(EINVAL);
1028 			break;
1029 		}
1030 		err = zap_add(os, intoobj, za.za_name,
1031 		    8, 1, &value, tx);
1032 		if (err)
1033 			break;
1034 	}
1035 	zap_cursor_fini(&zc);
1036 	return (err);
1037 }
1038 
1039 int
1040 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1041     dmu_tx_t *tx)
1042 {
1043 	zap_cursor_t zc;
1044 	zap_attribute_t za;
1045 	int err;
1046 
1047 	err = 0;
1048 	for (zap_cursor_init(&zc, os, fromobj);
1049 	    zap_cursor_retrieve(&zc, &za) == 0;
1050 	    (void) zap_cursor_advance(&zc)) {
1051 		uint64_t delta = 0;
1052 
1053 		if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1054 			err = SET_ERROR(EINVAL);
1055 			break;
1056 		}
1057 
1058 		err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1059 		if (err != 0 && err != ENOENT)
1060 			break;
1061 		delta += za.za_first_integer;
1062 		err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1063 		if (err)
1064 			break;
1065 	}
1066 	zap_cursor_fini(&zc);
1067 	return (err);
1068 }
1069 
1070 int
1071 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1072 {
1073 	char name[20];
1074 
1075 	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1076 	return (zap_add(os, obj, name, 8, 1, &value, tx));
1077 }
1078 
1079 int
1080 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1081 {
1082 	char name[20];
1083 
1084 	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1085 	return (zap_remove(os, obj, name, tx));
1086 }
1087 
1088 int
1089 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1090 {
1091 	char name[20];
1092 
1093 	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1094 	return (zap_lookup(os, obj, name, 8, 1, &value));
1095 }
1096 
1097 int
1098 zap_add_int_key(objset_t *os, uint64_t obj,
1099     uint64_t key, uint64_t value, dmu_tx_t *tx)
1100 {
1101 	char name[20];
1102 
1103 	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1104 	return (zap_add(os, obj, name, 8, 1, &value, tx));
1105 }
1106 
1107 int
1108 zap_update_int_key(objset_t *os, uint64_t obj,
1109     uint64_t key, uint64_t value, dmu_tx_t *tx)
1110 {
1111 	char name[20];
1112 
1113 	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1114 	return (zap_update(os, obj, name, 8, 1, &value, tx));
1115 }
1116 
1117 int
1118 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1119 {
1120 	char name[20];
1121 
1122 	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1123 	return (zap_lookup(os, obj, name, 8, 1, valuep));
1124 }
1125 
1126 int
1127 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1128     dmu_tx_t *tx)
1129 {
1130 	uint64_t value = 0;
1131 	int err;
1132 
1133 	if (delta == 0)
1134 		return (0);
1135 
1136 	err = zap_lookup(os, obj, name, 8, 1, &value);
1137 	if (err != 0 && err != ENOENT)
1138 		return (err);
1139 	value += delta;
1140 	if (value == 0)
1141 		err = zap_remove(os, obj, name, tx);
1142 	else
1143 		err = zap_update(os, obj, name, 8, 1, &value, tx);
1144 	return (err);
1145 }
1146 
1147 int
1148 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1149     dmu_tx_t *tx)
1150 {
1151 	char name[20];
1152 
1153 	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1154 	return (zap_increment(os, obj, name, delta, tx));
1155 }
1156 
1157 /*
1158  * Routines for iterating over the attributes.
1159  */
1160 
1161 int
1162 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1163 {
1164 	int err = ENOENT;
1165 	zap_entry_handle_t zeh;
1166 	zap_leaf_t *l;
1167 
1168 	/* retrieve the next entry at or after zc_hash/zc_cd */
1169 	/* if no entry, return ENOENT */
1170 
1171 	if (zc->zc_leaf &&
1172 	    (ZAP_HASH_IDX(zc->zc_hash,
1173 	    zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1174 	    zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1175 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1176 		zap_put_leaf(zc->zc_leaf);
1177 		zc->zc_leaf = NULL;
1178 	}
1179 
1180 again:
1181 	if (zc->zc_leaf == NULL) {
1182 		err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1183 		    &zc->zc_leaf);
1184 		if (err != 0)
1185 			return (err);
1186 	} else {
1187 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1188 	}
1189 	l = zc->zc_leaf;
1190 
1191 	err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1192 
1193 	if (err == ENOENT) {
1194 		uint64_t nocare =
1195 		    (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1196 		zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1197 		zc->zc_cd = 0;
1198 		if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 ||
1199 		    zc->zc_hash == 0) {
1200 			zc->zc_hash = -1ULL;
1201 		} else {
1202 			zap_put_leaf(zc->zc_leaf);
1203 			zc->zc_leaf = NULL;
1204 			goto again;
1205 		}
1206 	}
1207 
1208 	if (err == 0) {
1209 		zc->zc_hash = zeh.zeh_hash;
1210 		zc->zc_cd = zeh.zeh_cd;
1211 		za->za_integer_length = zeh.zeh_integer_size;
1212 		za->za_num_integers = zeh.zeh_num_integers;
1213 		if (zeh.zeh_num_integers == 0) {
1214 			za->za_first_integer = 0;
1215 		} else {
1216 			err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1217 			ASSERT(err == 0 || err == EOVERFLOW);
1218 		}
1219 		err = zap_entry_read_name(zap, &zeh,
1220 		    sizeof (za->za_name), za->za_name);
1221 		ASSERT(err == 0);
1222 
1223 		za->za_normalization_conflict =
1224 		    zap_entry_normalization_conflict(&zeh,
1225 		    NULL, za->za_name, zap);
1226 	}
1227 	rw_exit(&zc->zc_leaf->l_rwlock);
1228 	return (err);
1229 }
1230 
1231 static void
1232 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1233 {
1234 	int i, err;
1235 	uint64_t lastblk = 0;
1236 
1237 	/*
1238 	 * NB: if a leaf has more pointers than an entire ptrtbl block
1239 	 * can hold, then it'll be accounted for more than once, since
1240 	 * we won't have lastblk.
1241 	 */
1242 	for (i = 0; i < len; i++) {
1243 		zap_leaf_t *l;
1244 
1245 		if (tbl[i] == lastblk)
1246 			continue;
1247 		lastblk = tbl[i];
1248 
1249 		err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1250 		if (err == 0) {
1251 			zap_leaf_stats(zap, l, zs);
1252 			zap_put_leaf(l);
1253 		}
1254 	}
1255 }
1256 
1257 void
1258 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1259 {
1260 	int bs = FZAP_BLOCK_SHIFT(zap);
1261 	zs->zs_blocksize = 1ULL << bs;
1262 
1263 	/*
1264 	 * Set zap_phys_t fields
1265 	 */
1266 	zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1267 	zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1268 	zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1269 	zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1270 	zs->zs_magic = zap_f_phys(zap)->zap_magic;
1271 	zs->zs_salt = zap_f_phys(zap)->zap_salt;
1272 
1273 	/*
1274 	 * Set zap_ptrtbl fields
1275 	 */
1276 	zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1277 	zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1278 	zs->zs_ptrtbl_blks_copied =
1279 	    zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1280 	zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1281 	zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1282 	zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1283 
1284 	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1285 		/* the ptrtbl is entirely in the header block. */
1286 		zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1287 		    1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1288 	} else {
1289 		int b;
1290 
1291 		dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
1292 		    zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1293 		    zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs,
1294 		    ZIO_PRIORITY_SYNC_READ);
1295 
1296 		for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1297 		    b++) {
1298 			dmu_buf_t *db;
1299 			int err;
1300 
1301 			err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1302 			    (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1303 			    FTAG, &db, DMU_READ_NO_PREFETCH);
1304 			if (err == 0) {
1305 				zap_stats_ptrtbl(zap, db->db_data,
1306 				    1<<(bs-3), zs);
1307 				dmu_buf_rele(db, FTAG);
1308 			}
1309 		}
1310 	}
1311 }
1312 
1313 int
1314 fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite,
1315     uint64_t *tooverwrite)
1316 {
1317 	zap_t *zap = zn->zn_zap;
1318 	zap_leaf_t *l;
1319 	int err;
1320 
1321 	/*
1322 	 * Account for the header block of the fatzap.
1323 	 */
1324 	if (!add && dmu_buf_freeable(zap->zap_dbuf)) {
1325 		*tooverwrite += zap->zap_dbuf->db_size;
1326 	} else {
1327 		*towrite += zap->zap_dbuf->db_size;
1328 	}
1329 
1330 	/*
1331 	 * Account for the pointer table blocks.
1332 	 * If we are adding we need to account for the following cases :
1333 	 * - If the pointer table is embedded, this operation could force an
1334 	 *   external pointer table.
1335 	 * - If this already has an external pointer table this operation
1336 	 *   could extend the table.
1337 	 */
1338 	if (add) {
1339 		if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0)
1340 			*towrite += zap->zap_dbuf->db_size;
1341 		else
1342 			*towrite += (zap->zap_dbuf->db_size * 3);
1343 	}
1344 
1345 	/*
1346 	 * Now, check if the block containing leaf is freeable
1347 	 * and account accordingly.
1348 	 */
1349 	err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l);
1350 	if (err != 0) {
1351 		return (err);
1352 	}
1353 
1354 	if (!add && dmu_buf_freeable(l->l_dbuf)) {
1355 		*tooverwrite += l->l_dbuf->db_size;
1356 	} else {
1357 		/*
1358 		 * If this an add operation, the leaf block could split.
1359 		 * Hence, we need to account for an additional leaf block.
1360 		 */
1361 		*towrite += (add ? 2 : 1) * l->l_dbuf->db_size;
1362 	}
1363 
1364 	zap_put_leaf(l);
1365 	return (0);
1366 }
1367