zap.c revision 307287
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, 2016 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
52int fzap_default_block_shift = 14; /* 16k blocksize */
53
54extern inline zap_phys_t *zap_f_phys(zap_t *zap);
55
56static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
57
58void
59fzap_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
73void
74fzap_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 = zap_evict;
85
86	mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
87	zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;
88
89	zp = zap_f_phys(zap);
90	/*
91	 * explicitly zero it since it might be coming from an
92	 * initialized microzap
93	 */
94	bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
95	zp->zap_block_type = ZBT_HEADER;
96	zp->zap_magic = ZAP_MAGIC;
97
98	zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
99
100	zp->zap_freeblk = 2;		/* block 1 will be the first leaf */
101	zp->zap_num_leafs = 1;
102	zp->zap_num_entries = 0;
103	zp->zap_salt = zap->zap_salt;
104	zp->zap_normflags = zap->zap_normflags;
105	zp->zap_flags = flags;
106
107	/* block 1 will be the first leaf */
108	for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
109		ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
110
111	/*
112	 * set up block 1 - the first leaf
113	 */
114	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
115	    1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
116	dmu_buf_will_dirty(db, tx);
117
118	l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
119	l->l_dbuf = db;
120
121	zap_leaf_init(l, zp->zap_normflags != 0);
122
123	kmem_free(l, sizeof (zap_leaf_t));
124	dmu_buf_rele(db, FTAG);
125}
126
127static int
128zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
129{
130	if (RW_WRITE_HELD(&zap->zap_rwlock))
131		return (1);
132	if (rw_tryupgrade(&zap->zap_rwlock)) {
133		dmu_buf_will_dirty(zap->zap_dbuf, tx);
134		return (1);
135	}
136	return (0);
137}
138
139/*
140 * Generic routines for dealing with the pointer & cookie tables.
141 */
142
143static int
144zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
145    void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
146    dmu_tx_t *tx)
147{
148	uint64_t b, newblk;
149	dmu_buf_t *db_old, *db_new;
150	int err;
151	int bs = FZAP_BLOCK_SHIFT(zap);
152	int hepb = 1<<(bs-4);
153	/* hepb = half the number of entries in a block */
154
155	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
156	ASSERT(tbl->zt_blk != 0);
157	ASSERT(tbl->zt_numblks > 0);
158
159	if (tbl->zt_nextblk != 0) {
160		newblk = tbl->zt_nextblk;
161	} else {
162		newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
163		tbl->zt_nextblk = newblk;
164		ASSERT0(tbl->zt_blks_copied);
165		dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
166		    tbl->zt_blk << bs, tbl->zt_numblks << bs,
167		    ZIO_PRIORITY_SYNC_READ);
168	}
169
170	/*
171	 * Copy the ptrtbl from the old to new location.
172	 */
173
174	b = tbl->zt_blks_copied;
175	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
176	    (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
177	if (err)
178		return (err);
179
180	/* first half of entries in old[b] go to new[2*b+0] */
181	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
182	    (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
183	dmu_buf_will_dirty(db_new, tx);
184	transfer_func(db_old->db_data, db_new->db_data, hepb);
185	dmu_buf_rele(db_new, FTAG);
186
187	/* second half of entries in old[b] go to new[2*b+1] */
188	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
189	    (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
190	dmu_buf_will_dirty(db_new, tx);
191	transfer_func((uint64_t *)db_old->db_data + hepb,
192	    db_new->db_data, hepb);
193	dmu_buf_rele(db_new, FTAG);
194
195	dmu_buf_rele(db_old, FTAG);
196
197	tbl->zt_blks_copied++;
198
199	dprintf("copied block %llu of %llu\n",
200	    tbl->zt_blks_copied, tbl->zt_numblks);
201
202	if (tbl->zt_blks_copied == tbl->zt_numblks) {
203		(void) dmu_free_range(zap->zap_objset, zap->zap_object,
204		    tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
205
206		tbl->zt_blk = newblk;
207		tbl->zt_numblks *= 2;
208		tbl->zt_shift++;
209		tbl->zt_nextblk = 0;
210		tbl->zt_blks_copied = 0;
211
212		dprintf("finished; numblocks now %llu (%lluk entries)\n",
213		    tbl->zt_numblks, 1<<(tbl->zt_shift-10));
214	}
215
216	return (0);
217}
218
219static int
220zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
221    dmu_tx_t *tx)
222{
223	int err;
224	uint64_t blk, off;
225	int bs = FZAP_BLOCK_SHIFT(zap);
226	dmu_buf_t *db;
227
228	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
229	ASSERT(tbl->zt_blk != 0);
230
231	dprintf("storing %llx at index %llx\n", val, idx);
232
233	blk = idx >> (bs-3);
234	off = idx & ((1<<(bs-3))-1);
235
236	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
237	    (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
238	if (err)
239		return (err);
240	dmu_buf_will_dirty(db, tx);
241
242	if (tbl->zt_nextblk != 0) {
243		uint64_t idx2 = idx * 2;
244		uint64_t blk2 = idx2 >> (bs-3);
245		uint64_t off2 = idx2 & ((1<<(bs-3))-1);
246		dmu_buf_t *db2;
247
248		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
249		    (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
250		    DMU_READ_NO_PREFETCH);
251		if (err) {
252			dmu_buf_rele(db, FTAG);
253			return (err);
254		}
255		dmu_buf_will_dirty(db2, tx);
256		((uint64_t *)db2->db_data)[off2] = val;
257		((uint64_t *)db2->db_data)[off2+1] = val;
258		dmu_buf_rele(db2, FTAG);
259	}
260
261	((uint64_t *)db->db_data)[off] = val;
262	dmu_buf_rele(db, FTAG);
263
264	return (0);
265}
266
267static int
268zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
269{
270	uint64_t blk, off;
271	int err;
272	dmu_buf_t *db;
273	int bs = FZAP_BLOCK_SHIFT(zap);
274
275	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
276
277	blk = idx >> (bs-3);
278	off = idx & ((1<<(bs-3))-1);
279
280	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
281	    (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
282	if (err)
283		return (err);
284	*valp = ((uint64_t *)db->db_data)[off];
285	dmu_buf_rele(db, FTAG);
286
287	if (tbl->zt_nextblk != 0) {
288		/*
289		 * read the nextblk for the sake of i/o error checking,
290		 * so that zap_table_load() will catch errors for
291		 * zap_table_store.
292		 */
293		blk = (idx*2) >> (bs-3);
294
295		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
296		    (tbl->zt_nextblk + blk) << bs, FTAG, &db,
297		    DMU_READ_NO_PREFETCH);
298		if (err == 0)
299			dmu_buf_rele(db, FTAG);
300	}
301	return (err);
302}
303
304/*
305 * Routines for growing the ptrtbl.
306 */
307
308static void
309zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
310{
311	int i;
312	for (i = 0; i < n; i++) {
313		uint64_t lb = src[i];
314		dst[2*i+0] = lb;
315		dst[2*i+1] = lb;
316	}
317}
318
319static int
320zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
321{
322	/*
323	 * The pointer table should never use more hash bits than we
324	 * have (otherwise we'd be using useless zero bits to index it).
325	 * If we are within 2 bits of running out, stop growing, since
326	 * this is already an aberrant condition.
327	 */
328	if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
329		return (SET_ERROR(ENOSPC));
330
331	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
332		/*
333		 * We are outgrowing the "embedded" ptrtbl (the one
334		 * stored in the header block).  Give it its own entire
335		 * block, which will double the size of the ptrtbl.
336		 */
337		uint64_t newblk;
338		dmu_buf_t *db_new;
339		int err;
340
341		ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
342		    ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
343		ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk);
344
345		newblk = zap_allocate_blocks(zap, 1);
346		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
347		    newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
348		    DMU_READ_NO_PREFETCH);
349		if (err)
350			return (err);
351		dmu_buf_will_dirty(db_new, tx);
352		zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
353		    db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
354		dmu_buf_rele(db_new, FTAG);
355
356		zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk;
357		zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1;
358		zap_f_phys(zap)->zap_ptrtbl.zt_shift++;
359
360		ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
361		    zap_f_phys(zap)->zap_ptrtbl.zt_numblks <<
362		    (FZAP_BLOCK_SHIFT(zap)-3));
363
364		return (0);
365	} else {
366		return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl,
367		    zap_ptrtbl_transfer, tx));
368	}
369}
370
371static void
372zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
373{
374	dmu_buf_will_dirty(zap->zap_dbuf, tx);
375	mutex_enter(&zap->zap_f.zap_num_entries_mtx);
376	ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta);
377	zap_f_phys(zap)->zap_num_entries += delta;
378	mutex_exit(&zap->zap_f.zap_num_entries_mtx);
379}
380
381static uint64_t
382zap_allocate_blocks(zap_t *zap, int nblocks)
383{
384	uint64_t newblk;
385	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
386	newblk = zap_f_phys(zap)->zap_freeblk;
387	zap_f_phys(zap)->zap_freeblk += nblocks;
388	return (newblk);
389}
390
391static void
392zap_leaf_pageout(void *dbu)
393{
394	zap_leaf_t *l = dbu;
395
396	rw_destroy(&l->l_rwlock);
397	kmem_free(l, sizeof (zap_leaf_t));
398}
399
400static zap_leaf_t *
401zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
402{
403	void *winner;
404	zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
405
406	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
407
408	rw_init(&l->l_rwlock, 0, 0, 0);
409	rw_enter(&l->l_rwlock, RW_WRITER);
410	l->l_blkid = zap_allocate_blocks(zap, 1);
411	l->l_dbuf = NULL;
412
413	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
414	    l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
415	    DMU_READ_NO_PREFETCH));
416	dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
417	winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu);
418	ASSERT(winner == NULL);
419	dmu_buf_will_dirty(l->l_dbuf, tx);
420
421	zap_leaf_init(l, zap->zap_normflags != 0);
422
423	zap_f_phys(zap)->zap_num_leafs++;
424
425	return (l);
426}
427
428int
429fzap_count(zap_t *zap, uint64_t *count)
430{
431	ASSERT(!zap->zap_ismicro);
432	mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
433	*count = zap_f_phys(zap)->zap_num_entries;
434	mutex_exit(&zap->zap_f.zap_num_entries_mtx);
435	return (0);
436}
437
438/*
439 * Routines for obtaining zap_leaf_t's
440 */
441
442void
443zap_put_leaf(zap_leaf_t *l)
444{
445	rw_exit(&l->l_rwlock);
446	dmu_buf_rele(l->l_dbuf, NULL);
447}
448
449static zap_leaf_t *
450zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
451{
452	zap_leaf_t *l, *winner;
453
454	ASSERT(blkid != 0);
455
456	l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
457	rw_init(&l->l_rwlock, 0, 0, 0);
458	rw_enter(&l->l_rwlock, RW_WRITER);
459	l->l_blkid = blkid;
460	l->l_bs = highbit64(db->db_size) - 1;
461	l->l_dbuf = db;
462
463	dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
464	winner = dmu_buf_set_user(db, &l->l_dbu);
465
466	rw_exit(&l->l_rwlock);
467	if (winner != NULL) {
468		/* someone else set it first */
469		zap_leaf_pageout(&l->l_dbu);
470		l = winner;
471	}
472
473	/*
474	 * lhr_pad was previously used for the next leaf in the leaf
475	 * chain.  There should be no chained leafs (as we have removed
476	 * support for them).
477	 */
478	ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1);
479
480	/*
481	 * There should be more hash entries than there can be
482	 * chunks to put in the hash table
483	 */
484	ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
485
486	/* The chunks should begin at the end of the hash table */
487	ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
488	    &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
489
490	/* The chunks should end at the end of the block */
491	ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
492	    (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size);
493
494	return (l);
495}
496
497static int
498zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
499    zap_leaf_t **lp)
500{
501	dmu_buf_t *db;
502	zap_leaf_t *l;
503	int bs = FZAP_BLOCK_SHIFT(zap);
504	int err;
505
506	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
507
508	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
509	    blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
510	if (err)
511		return (err);
512
513	ASSERT3U(db->db_object, ==, zap->zap_object);
514	ASSERT3U(db->db_offset, ==, blkid << bs);
515	ASSERT3U(db->db_size, ==, 1 << bs);
516	ASSERT(blkid != 0);
517
518	l = dmu_buf_get_user(db);
519
520	if (l == NULL)
521		l = zap_open_leaf(blkid, db);
522
523	rw_enter(&l->l_rwlock, lt);
524	/*
525	 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
526	 * causing ASSERT below to fail.
527	 */
528	if (lt == RW_WRITER)
529		dmu_buf_will_dirty(db, tx);
530	ASSERT3U(l->l_blkid, ==, blkid);
531	ASSERT3P(l->l_dbuf, ==, db);
532	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
533	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
534
535	*lp = l;
536	return (0);
537}
538
539static int
540zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
541{
542	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
543
544	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
545		ASSERT3U(idx, <,
546		    (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
547		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
548		return (0);
549	} else {
550		return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
551		    idx, valp));
552	}
553}
554
555static int
556zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
557{
558	ASSERT(tx != NULL);
559	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
560
561	if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
562		ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
563		return (0);
564	} else {
565		return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
566		    idx, blk, tx));
567	}
568}
569
570static int
571zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
572{
573	uint64_t idx, blk;
574	int err;
575
576	ASSERT(zap->zap_dbuf == NULL ||
577	    zap_f_phys(zap) == zap->zap_dbuf->db_data);
578	ASSERT3U(zap_f_phys(zap)->zap_magic, ==, ZAP_MAGIC);
579	idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
580	err = zap_idx_to_blk(zap, idx, &blk);
581	if (err != 0)
582		return (err);
583	err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
584
585	ASSERT(err ||
586	    ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
587	    zap_leaf_phys(*lp)->l_hdr.lh_prefix);
588	return (err);
589}
590
591static int
592zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l,
593    void *tag, 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, tag);
616		err = zap_lockdir(os, object, tx, RW_WRITER,
617		    FALSE, FALSE, tag, &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
679static void
680zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l,
681    void *tag, dmu_tx_t *tx)
682{
683	zap_t *zap = zn->zn_zap;
684	int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
685	int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
686	    zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
687
688	zap_put_leaf(l);
689
690	if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
691		int err;
692
693		/*
694		 * We are in the middle of growing the pointer table, or
695		 * this leaf will soon make us grow it.
696		 */
697		if (zap_tryupgradedir(zap, tx) == 0) {
698			objset_t *os = zap->zap_objset;
699			uint64_t zapobj = zap->zap_object;
700
701			zap_unlockdir(zap, tag);
702			err = zap_lockdir(os, zapobj, tx,
703			    RW_WRITER, FALSE, FALSE, tag, &zn->zn_zap);
704			zap = zn->zn_zap;
705			if (err)
706				return;
707		}
708
709		/* could have finished growing while our locks were down */
710		if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
711			(void) zap_grow_ptrtbl(zap, tx);
712	}
713}
714
715static int
716fzap_checkname(zap_name_t *zn)
717{
718	if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
719		return (SET_ERROR(ENAMETOOLONG));
720	return (0);
721}
722
723static int
724fzap_checksize(uint64_t integer_size, uint64_t num_integers)
725{
726	/* Only integer sizes supported by C */
727	switch (integer_size) {
728	case 1:
729	case 2:
730	case 4:
731	case 8:
732		break;
733	default:
734		return (SET_ERROR(EINVAL));
735	}
736
737	if (integer_size * num_integers > ZAP_MAXVALUELEN)
738		return (E2BIG);
739
740	return (0);
741}
742
743static int
744fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
745{
746	int err;
747
748	if ((err = fzap_checkname(zn)) != 0)
749		return (err);
750	return (fzap_checksize(integer_size, num_integers));
751}
752
753/*
754 * Routines for manipulating attributes.
755 */
756int
757fzap_lookup(zap_name_t *zn,
758    uint64_t integer_size, uint64_t num_integers, void *buf,
759    char *realname, int rn_len, boolean_t *ncp)
760{
761	zap_leaf_t *l;
762	int err;
763	zap_entry_handle_t zeh;
764
765	if ((err = fzap_checkname(zn)) != 0)
766		return (err);
767
768	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
769	if (err != 0)
770		return (err);
771	err = zap_leaf_lookup(l, zn, &zeh);
772	if (err == 0) {
773		if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
774			zap_put_leaf(l);
775			return (err);
776		}
777
778		err = zap_entry_read(&zeh, integer_size, num_integers, buf);
779		(void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
780		if (ncp) {
781			*ncp = zap_entry_normalization_conflict(&zeh,
782			    zn, NULL, zn->zn_zap);
783		}
784	}
785
786	zap_put_leaf(l);
787	return (err);
788}
789
790int
791fzap_add_cd(zap_name_t *zn,
792    uint64_t integer_size, uint64_t num_integers,
793    const void *val, uint32_t cd, void *tag, dmu_tx_t *tx)
794{
795	zap_leaf_t *l;
796	int err;
797	zap_entry_handle_t zeh;
798	zap_t *zap = zn->zn_zap;
799
800	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
801	ASSERT(!zap->zap_ismicro);
802	ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
803
804	err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
805	if (err != 0)
806		return (err);
807retry:
808	err = zap_leaf_lookup(l, zn, &zeh);
809	if (err == 0) {
810		err = SET_ERROR(EEXIST);
811		goto out;
812	}
813	if (err != ENOENT)
814		goto out;
815
816	err = zap_entry_create(l, zn, cd,
817	    integer_size, num_integers, val, &zeh);
818
819	if (err == 0) {
820		zap_increment_num_entries(zap, 1, tx);
821	} else if (err == EAGAIN) {
822		err = zap_expand_leaf(zn, l, tag, tx, &l);
823		zap = zn->zn_zap;	/* zap_expand_leaf() may change zap */
824		if (err == 0)
825			goto retry;
826	}
827
828out:
829	if (zap != NULL)
830		zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
831	return (err);
832}
833
834int
835fzap_add(zap_name_t *zn,
836    uint64_t integer_size, uint64_t num_integers,
837    const void *val, void *tag, dmu_tx_t *tx)
838{
839	int err = fzap_check(zn, integer_size, num_integers);
840	if (err != 0)
841		return (err);
842
843	return (fzap_add_cd(zn, integer_size, num_integers,
844	    val, ZAP_NEED_CD, tag, tx));
845}
846
847int
848fzap_update(zap_name_t *zn,
849    int integer_size, uint64_t num_integers, const void *val,
850    void *tag, dmu_tx_t *tx)
851{
852	zap_leaf_t *l;
853	int err, create;
854	zap_entry_handle_t zeh;
855	zap_t *zap = zn->zn_zap;
856
857	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
858	err = fzap_check(zn, integer_size, num_integers);
859	if (err != 0)
860		return (err);
861
862	err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
863	if (err != 0)
864		return (err);
865retry:
866	err = zap_leaf_lookup(l, zn, &zeh);
867	create = (err == ENOENT);
868	ASSERT(err == 0 || err == ENOENT);
869
870	if (create) {
871		err = zap_entry_create(l, zn, ZAP_NEED_CD,
872		    integer_size, num_integers, val, &zeh);
873		if (err == 0)
874			zap_increment_num_entries(zap, 1, tx);
875	} else {
876		err = zap_entry_update(&zeh, integer_size, num_integers, val);
877	}
878
879	if (err == EAGAIN) {
880		err = zap_expand_leaf(zn, l, tag, tx, &l);
881		zap = zn->zn_zap;	/* zap_expand_leaf() may change zap */
882		if (err == 0)
883			goto retry;
884	}
885
886	if (zap != NULL)
887		zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
888	return (err);
889}
890
891int
892fzap_length(zap_name_t *zn,
893    uint64_t *integer_size, uint64_t *num_integers)
894{
895	zap_leaf_t *l;
896	int err;
897	zap_entry_handle_t zeh;
898
899	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
900	if (err != 0)
901		return (err);
902	err = zap_leaf_lookup(l, zn, &zeh);
903	if (err != 0)
904		goto out;
905
906	if (integer_size)
907		*integer_size = zeh.zeh_integer_size;
908	if (num_integers)
909		*num_integers = zeh.zeh_num_integers;
910out:
911	zap_put_leaf(l);
912	return (err);
913}
914
915int
916fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
917{
918	zap_leaf_t *l;
919	int err;
920	zap_entry_handle_t zeh;
921
922	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
923	if (err != 0)
924		return (err);
925	err = zap_leaf_lookup(l, zn, &zeh);
926	if (err == 0) {
927		zap_entry_remove(&zeh);
928		zap_increment_num_entries(zn->zn_zap, -1, tx);
929	}
930	zap_put_leaf(l);
931	return (err);
932}
933
934void
935fzap_prefetch(zap_name_t *zn)
936{
937	uint64_t idx, blk;
938	zap_t *zap = zn->zn_zap;
939	int bs;
940
941	idx = ZAP_HASH_IDX(zn->zn_hash,
942	    zap_f_phys(zap)->zap_ptrtbl.zt_shift);
943	if (zap_idx_to_blk(zap, idx, &blk) != 0)
944		return;
945	bs = FZAP_BLOCK_SHIFT(zap);
946	dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs,
947	    ZIO_PRIORITY_SYNC_READ);
948}
949
950/*
951 * Helper functions for consumers.
952 */
953
954uint64_t
955zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
956    const char *name, dmu_tx_t *tx)
957{
958	uint64_t new_obj;
959
960	VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
961	VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
962	    tx) == 0);
963
964	return (new_obj);
965}
966
967int
968zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
969    char *name)
970{
971	zap_cursor_t zc;
972	zap_attribute_t *za;
973	int err;
974
975	if (mask == 0)
976		mask = -1ULL;
977
978	za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
979	for (zap_cursor_init(&zc, os, zapobj);
980	    (err = zap_cursor_retrieve(&zc, za)) == 0;
981	    zap_cursor_advance(&zc)) {
982		if ((za->za_first_integer & mask) == (value & mask)) {
983			(void) strcpy(name, za->za_name);
984			break;
985		}
986	}
987	zap_cursor_fini(&zc);
988	kmem_free(za, sizeof (zap_attribute_t));
989	return (err);
990}
991
992int
993zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
994{
995	zap_cursor_t zc;
996	zap_attribute_t za;
997	int err;
998
999	err = 0;
1000	for (zap_cursor_init(&zc, os, fromobj);
1001	    zap_cursor_retrieve(&zc, &za) == 0;
1002	    (void) zap_cursor_advance(&zc)) {
1003		if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1004			err = SET_ERROR(EINVAL);
1005			break;
1006		}
1007		err = zap_add(os, intoobj, za.za_name,
1008		    8, 1, &za.za_first_integer, tx);
1009		if (err)
1010			break;
1011	}
1012	zap_cursor_fini(&zc);
1013	return (err);
1014}
1015
1016int
1017zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1018    uint64_t value, dmu_tx_t *tx)
1019{
1020	zap_cursor_t zc;
1021	zap_attribute_t za;
1022	int err;
1023
1024	err = 0;
1025	for (zap_cursor_init(&zc, os, fromobj);
1026	    zap_cursor_retrieve(&zc, &za) == 0;
1027	    (void) zap_cursor_advance(&zc)) {
1028		if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1029			err = SET_ERROR(EINVAL);
1030			break;
1031		}
1032		err = zap_add(os, intoobj, za.za_name,
1033		    8, 1, &value, tx);
1034		if (err)
1035			break;
1036	}
1037	zap_cursor_fini(&zc);
1038	return (err);
1039}
1040
1041int
1042zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1043    dmu_tx_t *tx)
1044{
1045	zap_cursor_t zc;
1046	zap_attribute_t za;
1047	int err;
1048
1049	err = 0;
1050	for (zap_cursor_init(&zc, os, fromobj);
1051	    zap_cursor_retrieve(&zc, &za) == 0;
1052	    (void) zap_cursor_advance(&zc)) {
1053		uint64_t delta = 0;
1054
1055		if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1056			err = SET_ERROR(EINVAL);
1057			break;
1058		}
1059
1060		err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1061		if (err != 0 && err != ENOENT)
1062			break;
1063		delta += za.za_first_integer;
1064		err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1065		if (err)
1066			break;
1067	}
1068	zap_cursor_fini(&zc);
1069	return (err);
1070}
1071
1072int
1073zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1074{
1075	char name[20];
1076
1077	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1078	return (zap_add(os, obj, name, 8, 1, &value, tx));
1079}
1080
1081int
1082zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1083{
1084	char name[20];
1085
1086	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1087	return (zap_remove(os, obj, name, tx));
1088}
1089
1090int
1091zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1092{
1093	char name[20];
1094
1095	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1096	return (zap_lookup(os, obj, name, 8, 1, &value));
1097}
1098
1099int
1100zap_add_int_key(objset_t *os, uint64_t obj,
1101    uint64_t key, uint64_t value, dmu_tx_t *tx)
1102{
1103	char name[20];
1104
1105	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1106	return (zap_add(os, obj, name, 8, 1, &value, tx));
1107}
1108
1109int
1110zap_update_int_key(objset_t *os, uint64_t obj,
1111    uint64_t key, uint64_t value, dmu_tx_t *tx)
1112{
1113	char name[20];
1114
1115	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1116	return (zap_update(os, obj, name, 8, 1, &value, tx));
1117}
1118
1119int
1120zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1121{
1122	char name[20];
1123
1124	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1125	return (zap_lookup(os, obj, name, 8, 1, valuep));
1126}
1127
1128int
1129zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1130    dmu_tx_t *tx)
1131{
1132	uint64_t value = 0;
1133	int err;
1134
1135	if (delta == 0)
1136		return (0);
1137
1138	err = zap_lookup(os, obj, name, 8, 1, &value);
1139	if (err != 0 && err != ENOENT)
1140		return (err);
1141	value += delta;
1142	if (value == 0)
1143		err = zap_remove(os, obj, name, tx);
1144	else
1145		err = zap_update(os, obj, name, 8, 1, &value, tx);
1146	return (err);
1147}
1148
1149int
1150zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1151    dmu_tx_t *tx)
1152{
1153	char name[20];
1154
1155	(void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1156	return (zap_increment(os, obj, name, delta, tx));
1157}
1158
1159/*
1160 * Routines for iterating over the attributes.
1161 */
1162
1163int
1164fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1165{
1166	int err = ENOENT;
1167	zap_entry_handle_t zeh;
1168	zap_leaf_t *l;
1169
1170	/* retrieve the next entry at or after zc_hash/zc_cd */
1171	/* if no entry, return ENOENT */
1172
1173	if (zc->zc_leaf &&
1174	    (ZAP_HASH_IDX(zc->zc_hash,
1175	    zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1176	    zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1177		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1178		zap_put_leaf(zc->zc_leaf);
1179		zc->zc_leaf = NULL;
1180	}
1181
1182again:
1183	if (zc->zc_leaf == NULL) {
1184		err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1185		    &zc->zc_leaf);
1186		if (err != 0)
1187			return (err);
1188	} else {
1189		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1190	}
1191	l = zc->zc_leaf;
1192
1193	err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1194
1195	if (err == ENOENT) {
1196		uint64_t nocare =
1197		    (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1198		zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1199		zc->zc_cd = 0;
1200		if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 ||
1201		    zc->zc_hash == 0) {
1202			zc->zc_hash = -1ULL;
1203		} else {
1204			zap_put_leaf(zc->zc_leaf);
1205			zc->zc_leaf = NULL;
1206			goto again;
1207		}
1208	}
1209
1210	if (err == 0) {
1211		zc->zc_hash = zeh.zeh_hash;
1212		zc->zc_cd = zeh.zeh_cd;
1213		za->za_integer_length = zeh.zeh_integer_size;
1214		za->za_num_integers = zeh.zeh_num_integers;
1215		if (zeh.zeh_num_integers == 0) {
1216			za->za_first_integer = 0;
1217		} else {
1218			err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1219			ASSERT(err == 0 || err == EOVERFLOW);
1220		}
1221		err = zap_entry_read_name(zap, &zeh,
1222		    sizeof (za->za_name), za->za_name);
1223		ASSERT(err == 0);
1224
1225		za->za_normalization_conflict =
1226		    zap_entry_normalization_conflict(&zeh,
1227		    NULL, za->za_name, zap);
1228	}
1229	rw_exit(&zc->zc_leaf->l_rwlock);
1230	return (err);
1231}
1232
1233static void
1234zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1235{
1236	int i, err;
1237	uint64_t lastblk = 0;
1238
1239	/*
1240	 * NB: if a leaf has more pointers than an entire ptrtbl block
1241	 * can hold, then it'll be accounted for more than once, since
1242	 * we won't have lastblk.
1243	 */
1244	for (i = 0; i < len; i++) {
1245		zap_leaf_t *l;
1246
1247		if (tbl[i] == lastblk)
1248			continue;
1249		lastblk = tbl[i];
1250
1251		err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1252		if (err == 0) {
1253			zap_leaf_stats(zap, l, zs);
1254			zap_put_leaf(l);
1255		}
1256	}
1257}
1258
1259int
1260fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn)
1261{
1262	int err;
1263	zap_leaf_t *l;
1264	zap_entry_handle_t zeh;
1265
1266	if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
1267		return (SET_ERROR(ENAMETOOLONG));
1268
1269	err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l);
1270	if (err != 0)
1271		return (err);
1272
1273	err = zap_leaf_lookup(l, zn, &zeh);
1274	if (err != 0)
1275		return (err);
1276
1277	zc->zc_leaf = l;
1278	zc->zc_hash = zeh.zeh_hash;
1279	zc->zc_cd = zeh.zeh_cd;
1280
1281	return (err);
1282}
1283
1284void
1285fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1286{
1287	int bs = FZAP_BLOCK_SHIFT(zap);
1288	zs->zs_blocksize = 1ULL << bs;
1289
1290	/*
1291	 * Set zap_phys_t fields
1292	 */
1293	zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1294	zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1295	zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1296	zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1297	zs->zs_magic = zap_f_phys(zap)->zap_magic;
1298	zs->zs_salt = zap_f_phys(zap)->zap_salt;
1299
1300	/*
1301	 * Set zap_ptrtbl fields
1302	 */
1303	zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1304	zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1305	zs->zs_ptrtbl_blks_copied =
1306	    zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1307	zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1308	zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1309	zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1310
1311	if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1312		/* the ptrtbl is entirely in the header block. */
1313		zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1314		    1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1315	} else {
1316		int b;
1317
1318		dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
1319		    zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1320		    zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs,
1321		    ZIO_PRIORITY_SYNC_READ);
1322
1323		for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1324		    b++) {
1325			dmu_buf_t *db;
1326			int err;
1327
1328			err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1329			    (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1330			    FTAG, &db, DMU_READ_NO_PREFETCH);
1331			if (err == 0) {
1332				zap_stats_ptrtbl(zap, db->db_data,
1333				    1<<(bs-3), zs);
1334				dmu_buf_rele(db, FTAG);
1335			}
1336		}
1337	}
1338}
1339
1340int
1341fzap_count_write(zap_name_t *zn, int add, refcount_t *towrite,
1342    refcount_t *tooverwrite)
1343{
1344	zap_t *zap = zn->zn_zap;
1345	zap_leaf_t *l;
1346	int err;
1347
1348	/*
1349	 * Account for the header block of the fatzap.
1350	 */
1351	if (!add && dmu_buf_freeable(zap->zap_dbuf)) {
1352		(void) refcount_add_many(tooverwrite,
1353		    zap->zap_dbuf->db_size, FTAG);
1354	} else {
1355		(void) refcount_add_many(towrite,
1356		    zap->zap_dbuf->db_size, FTAG);
1357	}
1358
1359	/*
1360	 * Account for the pointer table blocks.
1361	 * If we are adding we need to account for the following cases :
1362	 * - If the pointer table is embedded, this operation could force an
1363	 *   external pointer table.
1364	 * - If this already has an external pointer table this operation
1365	 *   could extend the table.
1366	 */
1367	if (add) {
1368		if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
1369			(void) refcount_add_many(towrite,
1370			    zap->zap_dbuf->db_size, FTAG);
1371		} else {
1372			(void) refcount_add_many(towrite,
1373			    zap->zap_dbuf->db_size * 3, FTAG);
1374		}
1375	}
1376
1377	/*
1378	 * Now, check if the block containing leaf is freeable
1379	 * and account accordingly.
1380	 */
1381	err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l);
1382	if (err != 0) {
1383		return (err);
1384	}
1385
1386	if (!add && dmu_buf_freeable(l->l_dbuf)) {
1387		(void) refcount_add_many(tooverwrite, l->l_dbuf->db_size, FTAG);
1388	} else {
1389		/*
1390		 * If this an add operation, the leaf block could split.
1391		 * Hence, we need to account for an additional leaf block.
1392		 */
1393		(void) refcount_add_many(towrite,
1394		    (add ? 2 : 1) * l->l_dbuf->db_size, FTAG);
1395	}
1396
1397	zap_put_leaf(l);
1398	return (0);
1399}
1400