xref: /titanic_52/usr/src/uts/common/fs/zfs/zap.c (revision b02e9a2d4d2071d770e5aa9ae8f83f2bbe1f2ced)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 
29 /*
30  * This file contains the top half of the zfs directory structure
31  * implementation. The bottom half is in zap_leaf.c.
32  *
33  * The zdir is an extendable hash data structure. There is a table of
34  * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are
35  * each a constant size and hold a variable number of directory entries.
36  * The buckets (aka "leaf nodes") are implemented in zap_leaf.c.
37  *
38  * The pointer table holds a power of 2 number of pointers.
39  * (1<<zap_t->zd_data->zd_phys->zd_prefix_len).  The bucket pointed to
40  * by the pointer at index i in the table holds entries whose hash value
41  * has a zd_prefix_len - bit prefix
42  */
43 
44 #include <sys/spa.h>
45 #include <sys/dmu.h>
46 #include <sys/zfs_context.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 static void zap_leaf_pageout(dmu_buf_t *db, void *vl);
55 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
56 
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)
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 	(void) dmu_buf_update_user(zap->zap_dbuf, zap, zap,
85 	    &zap->zap_f.zap_phys, zap_evict);
86 
87 	mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
88 	zap->zap_f.zap_block_shift = highbit(zap->zap_dbuf->db_size) - 1;
89 
90 	zp = zap->zap_f.zap_phys;
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 
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));
116 	dmu_buf_will_dirty(db, tx);
117 
118 	l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
119 	l->l_dbuf = db;
120 	l->l_phys = db->db_data;
121 
122 	zap_leaf_init(l, spa_version(dmu_objset_spa(zap->zap_objset)));
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 		ASSERT3U(tbl->zt_blks_copied, ==, 0);
166 		dmu_prefetch(zap->zap_objset, zap->zap_object,
167 		    tbl->zt_blk << bs, tbl->zt_numblks << bs);
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);
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));
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));
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 
219 static int
220 zap_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);
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 		if (err) {
251 			dmu_buf_rele(db, FTAG);
252 			return (err);
253 		}
254 		dmu_buf_will_dirty(db2, tx);
255 		((uint64_t *)db2->db_data)[off2] = val;
256 		((uint64_t *)db2->db_data)[off2+1] = val;
257 		dmu_buf_rele(db2, FTAG);
258 	}
259 
260 	((uint64_t *)db->db_data)[off] = val;
261 	dmu_buf_rele(db, FTAG);
262 
263 	return (0);
264 }
265 
266 static int
267 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
268 {
269 	uint64_t blk, off;
270 	int err;
271 	dmu_buf_t *db;
272 	int bs = FZAP_BLOCK_SHIFT(zap);
273 
274 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
275 
276 	blk = idx >> (bs-3);
277 	off = idx & ((1<<(bs-3))-1);
278 
279 	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
280 	    (tbl->zt_blk + blk) << bs, FTAG, &db);
281 	if (err)
282 		return (err);
283 	*valp = ((uint64_t *)db->db_data)[off];
284 	dmu_buf_rele(db, FTAG);
285 
286 	if (tbl->zt_nextblk != 0) {
287 		/*
288 		 * read the nextblk for the sake of i/o error checking,
289 		 * so that zap_table_load() will catch errors for
290 		 * zap_table_store.
291 		 */
292 		blk = (idx*2) >> (bs-3);
293 
294 		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
295 		    (tbl->zt_nextblk + blk) << bs, FTAG, &db);
296 		dmu_buf_rele(db, FTAG);
297 	}
298 	return (err);
299 }
300 
301 /*
302  * Routines for growing the ptrtbl.
303  */
304 
305 static void
306 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
307 {
308 	int i;
309 	for (i = 0; i < n; i++) {
310 		uint64_t lb = src[i];
311 		dst[2*i+0] = lb;
312 		dst[2*i+1] = lb;
313 	}
314 }
315 
316 static int
317 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
318 {
319 	/* In case things go horribly wrong. */
320 	if (zap->zap_f.zap_phys->zap_ptrtbl.zt_shift >= ZAP_HASHBITS-2)
321 		return (ENOSPC);
322 
323 	if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) {
324 		/*
325 		 * We are outgrowing the "embedded" ptrtbl (the one
326 		 * stored in the header block).  Give it its own entire
327 		 * block, which will double the size of the ptrtbl.
328 		 */
329 		uint64_t newblk;
330 		dmu_buf_t *db_new;
331 		int err;
332 
333 		ASSERT3U(zap->zap_f.zap_phys->zap_ptrtbl.zt_shift, ==,
334 		    ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
335 		ASSERT3U(zap->zap_f.zap_phys->zap_ptrtbl.zt_blk, ==, 0);
336 
337 		newblk = zap_allocate_blocks(zap, 1);
338 		err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
339 		    newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new);
340 		if (err)
341 			return (err);
342 		dmu_buf_will_dirty(db_new, tx);
343 		zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
344 		    db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
345 		dmu_buf_rele(db_new, FTAG);
346 
347 		zap->zap_f.zap_phys->zap_ptrtbl.zt_blk = newblk;
348 		zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks = 1;
349 		zap->zap_f.zap_phys->zap_ptrtbl.zt_shift++;
350 
351 		ASSERT3U(1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift, ==,
352 		    zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks <<
353 		    (FZAP_BLOCK_SHIFT(zap)-3));
354 
355 		return (0);
356 	} else {
357 		return (zap_table_grow(zap, &zap->zap_f.zap_phys->zap_ptrtbl,
358 		    zap_ptrtbl_transfer, tx));
359 	}
360 }
361 
362 static void
363 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
364 {
365 	dmu_buf_will_dirty(zap->zap_dbuf, tx);
366 	mutex_enter(&zap->zap_f.zap_num_entries_mtx);
367 	ASSERT(delta > 0 || zap->zap_f.zap_phys->zap_num_entries >= -delta);
368 	zap->zap_f.zap_phys->zap_num_entries += delta;
369 	mutex_exit(&zap->zap_f.zap_num_entries_mtx);
370 }
371 
372 static uint64_t
373 zap_allocate_blocks(zap_t *zap, int nblocks)
374 {
375 	uint64_t newblk;
376 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
377 	newblk = zap->zap_f.zap_phys->zap_freeblk;
378 	zap->zap_f.zap_phys->zap_freeblk += nblocks;
379 	return (newblk);
380 }
381 
382 static zap_leaf_t *
383 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
384 {
385 	void *winner;
386 	zap_leaf_t *l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
387 
388 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
389 
390 	rw_init(&l->l_rwlock, 0, 0, 0);
391 	rw_enter(&l->l_rwlock, RW_WRITER);
392 	l->l_blkid = zap_allocate_blocks(zap, 1);
393 	l->l_dbuf = NULL;
394 	l->l_phys = NULL;
395 
396 	VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
397 	    l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf));
398 	winner = dmu_buf_set_user(l->l_dbuf, l, &l->l_phys, zap_leaf_pageout);
399 	ASSERT(winner == NULL);
400 	dmu_buf_will_dirty(l->l_dbuf, tx);
401 
402 	zap_leaf_init(l, spa_version(dmu_objset_spa(zap->zap_objset)));
403 
404 	zap->zap_f.zap_phys->zap_num_leafs++;
405 
406 	return (l);
407 }
408 
409 int
410 fzap_count(zap_t *zap, uint64_t *count)
411 {
412 	ASSERT(!zap->zap_ismicro);
413 	mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
414 	*count = zap->zap_f.zap_phys->zap_num_entries;
415 	mutex_exit(&zap->zap_f.zap_num_entries_mtx);
416 	return (0);
417 }
418 
419 /*
420  * Routines for obtaining zap_leaf_t's
421  */
422 
423 void
424 zap_put_leaf(zap_leaf_t *l)
425 {
426 	rw_exit(&l->l_rwlock);
427 	dmu_buf_rele(l->l_dbuf, NULL);
428 }
429 
430 _NOTE(ARGSUSED(0))
431 static void
432 zap_leaf_pageout(dmu_buf_t *db, void *vl)
433 {
434 	zap_leaf_t *l = vl;
435 
436 	rw_destroy(&l->l_rwlock);
437 	kmem_free(l, sizeof (zap_leaf_t));
438 }
439 
440 static zap_leaf_t *
441 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
442 {
443 	zap_leaf_t *l, *winner;
444 
445 	ASSERT(blkid != 0);
446 
447 	l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
448 	rw_init(&l->l_rwlock, 0, 0, 0);
449 	rw_enter(&l->l_rwlock, RW_WRITER);
450 	l->l_blkid = blkid;
451 	l->l_bs = highbit(db->db_size)-1;
452 	l->l_dbuf = db;
453 	l->l_phys = NULL;
454 
455 	winner = dmu_buf_set_user(db, l, &l->l_phys, zap_leaf_pageout);
456 
457 	rw_exit(&l->l_rwlock);
458 	if (winner != NULL) {
459 		/* someone else set it first */
460 		zap_leaf_pageout(NULL, l);
461 		l = winner;
462 	}
463 
464 	/*
465 	 * lhr_pad was previously used for the next leaf in the leaf
466 	 * chain.  There should be no chained leafs (as we have removed
467 	 * support for them).
468 	 */
469 	ASSERT3U(l->l_phys->l_hdr.lh_pad1, ==, 0);
470 
471 	/*
472 	 * There should be more hash entries than there can be
473 	 * chunks to put in the hash table
474 	 */
475 	ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
476 
477 	/* The chunks should begin at the end of the hash table */
478 	ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
479 	    &l->l_phys->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
480 
481 	/* The chunks should end at the end of the block */
482 	ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
483 	    (uintptr_t)l->l_phys, ==, l->l_dbuf->db_size);
484 
485 	return (l);
486 }
487 
488 static int
489 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
490     zap_leaf_t **lp)
491 {
492 	dmu_buf_t *db;
493 	zap_leaf_t *l;
494 	int bs = FZAP_BLOCK_SHIFT(zap);
495 	int err;
496 
497 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
498 
499 	err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
500 	    blkid << bs, NULL, &db);
501 	if (err)
502 		return (err);
503 
504 	ASSERT3U(db->db_object, ==, zap->zap_object);
505 	ASSERT3U(db->db_offset, ==, blkid << bs);
506 	ASSERT3U(db->db_size, ==, 1 << bs);
507 	ASSERT(blkid != 0);
508 
509 	l = dmu_buf_get_user(db);
510 
511 	if (l == NULL)
512 		l = zap_open_leaf(blkid, db);
513 
514 	rw_enter(&l->l_rwlock, lt);
515 	/*
516 	 * Must lock before dirtying, otherwise l->l_phys could change,
517 	 * causing ASSERT below to fail.
518 	 */
519 	if (lt == RW_WRITER)
520 		dmu_buf_will_dirty(db, tx);
521 	ASSERT3U(l->l_blkid, ==, blkid);
522 	ASSERT3P(l->l_dbuf, ==, db);
523 	ASSERT3P(l->l_phys, ==, l->l_dbuf->db_data);
524 	ASSERT3U(l->l_phys->l_hdr.lh_block_type, ==, ZBT_LEAF);
525 	ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
526 
527 	*lp = l;
528 	return (0);
529 }
530 
531 static int
532 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
533 {
534 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
535 
536 	if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) {
537 		ASSERT3U(idx, <,
538 		    (1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift));
539 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
540 		return (0);
541 	} else {
542 		return (zap_table_load(zap, &zap->zap_f.zap_phys->zap_ptrtbl,
543 		    idx, valp));
544 	}
545 }
546 
547 static int
548 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
549 {
550 	ASSERT(tx != NULL);
551 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
552 
553 	if (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk == 0) {
554 		ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
555 		return (0);
556 	} else {
557 		return (zap_table_store(zap, &zap->zap_f.zap_phys->zap_ptrtbl,
558 		    idx, blk, tx));
559 	}
560 }
561 
562 static int
563 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
564 {
565 	uint64_t idx, blk;
566 	int err;
567 
568 	ASSERT(zap->zap_dbuf == NULL ||
569 	    zap->zap_f.zap_phys == zap->zap_dbuf->db_data);
570 	ASSERT3U(zap->zap_f.zap_phys->zap_magic, ==, ZAP_MAGIC);
571 	idx = ZAP_HASH_IDX(h, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift);
572 	err = zap_idx_to_blk(zap, idx, &blk);
573 	if (err != 0)
574 		return (err);
575 	err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
576 
577 	ASSERT(err || ZAP_HASH_IDX(h, (*lp)->l_phys->l_hdr.lh_prefix_len) ==
578 	    (*lp)->l_phys->l_hdr.lh_prefix);
579 	return (err);
580 }
581 
582 static int
583 zap_expand_leaf(zap_t *zap, zap_leaf_t *l, uint64_t hash, dmu_tx_t *tx,
584     zap_leaf_t **lp)
585 {
586 	zap_leaf_t *nl;
587 	int prefix_diff, i, err;
588 	uint64_t sibling;
589 	int old_prefix_len = l->l_phys->l_hdr.lh_prefix_len;
590 
591 	ASSERT3U(old_prefix_len, <=, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift);
592 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
593 
594 	ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
595 	    l->l_phys->l_hdr.lh_prefix);
596 
597 	if (zap_tryupgradedir(zap, tx) == 0 ||
598 	    old_prefix_len == zap->zap_f.zap_phys->zap_ptrtbl.zt_shift) {
599 		/* We failed to upgrade, or need to grow the pointer table */
600 		objset_t *os = zap->zap_objset;
601 		uint64_t object = zap->zap_object;
602 
603 		zap_put_leaf(l);
604 		zap_unlockdir(zap);
605 		err = zap_lockdir(os, object, tx, RW_WRITER, FALSE, &zap);
606 		if (err)
607 			return (err);
608 		ASSERT(!zap->zap_ismicro);
609 
610 		while (old_prefix_len ==
611 		    zap->zap_f.zap_phys->zap_ptrtbl.zt_shift) {
612 			err = zap_grow_ptrtbl(zap, tx);
613 			if (err)
614 				return (err);
615 		}
616 
617 		err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
618 		if (err)
619 			return (err);
620 
621 		if (l->l_phys->l_hdr.lh_prefix_len != old_prefix_len) {
622 			/* it split while our locks were down */
623 			*lp = l;
624 			return (0);
625 		}
626 	}
627 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
628 	ASSERT3U(old_prefix_len, <, zap->zap_f.zap_phys->zap_ptrtbl.zt_shift);
629 	ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
630 	    l->l_phys->l_hdr.lh_prefix);
631 
632 	prefix_diff = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift -
633 	    (old_prefix_len + 1);
634 	sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
635 
636 	/* check for i/o errors before doing zap_leaf_split */
637 	for (i = 0; i < (1ULL<<prefix_diff); i++) {
638 		uint64_t blk;
639 		err = zap_idx_to_blk(zap, sibling+i, &blk);
640 		if (err)
641 			return (err);
642 		ASSERT3U(blk, ==, l->l_blkid);
643 	}
644 
645 	nl = zap_create_leaf(zap, tx);
646 	zap_leaf_split(l, nl, spa_version(dmu_objset_spa(zap->zap_objset)));
647 
648 	/* set sibling pointers */
649 	for (i = 0; i < (1ULL<<prefix_diff); i++) {
650 		err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
651 		ASSERT3U(err, ==, 0); /* we checked for i/o errors above */
652 	}
653 
654 	if (hash & (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len))) {
655 		/* we want the sibling */
656 		zap_put_leaf(l);
657 		*lp = nl;
658 	} else {
659 		zap_put_leaf(nl);
660 		*lp = l;
661 	}
662 
663 	return (0);
664 }
665 
666 static void
667 zap_put_leaf_maybe_grow_ptrtbl(zap_t *zap, zap_leaf_t *l, dmu_tx_t *tx)
668 {
669 	int shift = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift;
670 	int leaffull = (l->l_phys->l_hdr.lh_prefix_len == shift &&
671 	    l->l_phys->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
672 
673 	zap_put_leaf(l);
674 
675 	if (leaffull || zap->zap_f.zap_phys->zap_ptrtbl.zt_nextblk) {
676 		int err;
677 
678 		/*
679 		 * We are in the middle of growing the pointer table, or
680 		 * this leaf will soon make us grow it.
681 		 */
682 		if (zap_tryupgradedir(zap, tx) == 0) {
683 			objset_t *os = zap->zap_objset;
684 			uint64_t zapobj = zap->zap_object;
685 
686 			zap_unlockdir(zap);
687 			err = zap_lockdir(os, zapobj, tx,
688 			    RW_WRITER, FALSE, &zap);
689 			if (err)
690 				return;
691 		}
692 
693 		/* could have finished growing while our locks were down */
694 		if (zap->zap_f.zap_phys->zap_ptrtbl.zt_shift == shift)
695 			(void) zap_grow_ptrtbl(zap, tx);
696 	}
697 }
698 
699 
700 static int
701 fzap_checksize(const char *name, uint64_t integer_size, uint64_t num_integers)
702 {
703 	if (name && strlen(name) > ZAP_MAXNAMELEN)
704 		return (E2BIG);
705 
706 	/* Only integer sizes supported by C */
707 	switch (integer_size) {
708 	case 1:
709 	case 2:
710 	case 4:
711 	case 8:
712 		break;
713 	default:
714 		return (EINVAL);
715 	}
716 
717 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
718 		return (E2BIG);
719 
720 	return (0);
721 }
722 
723 /*
724  * Routines for manipulating attributes.
725  */
726 int
727 fzap_lookup(zap_name_t *zn,
728     uint64_t integer_size, uint64_t num_integers, void *buf,
729     char *realname, int rn_len, boolean_t *ncp)
730 {
731 	zap_leaf_t *l;
732 	int err;
733 	zap_entry_handle_t zeh;
734 
735 	err = fzap_checksize(zn->zn_name_orij, integer_size, num_integers);
736 	if (err != 0)
737 		return (err);
738 
739 	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
740 	if (err != 0)
741 		return (err);
742 	err = zap_leaf_lookup(l, zn, &zeh);
743 	if (err == 0) {
744 		err = zap_entry_read(&zeh, integer_size, num_integers, buf);
745 		(void) zap_entry_read_name(&zeh, rn_len, realname);
746 		if (ncp) {
747 			*ncp = zap_entry_normalization_conflict(&zeh,
748 			    zn, NULL, zn->zn_zap);
749 		}
750 	}
751 
752 	zap_put_leaf(l);
753 	return (err);
754 }
755 
756 int
757 fzap_add_cd(zap_name_t *zn,
758     uint64_t integer_size, uint64_t num_integers,
759     const void *val, uint32_t cd, dmu_tx_t *tx)
760 {
761 	zap_leaf_t *l;
762 	int err;
763 	zap_entry_handle_t zeh;
764 	zap_t *zap = zn->zn_zap;
765 
766 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
767 	ASSERT(!zap->zap_ismicro);
768 	ASSERT(fzap_checksize(zn->zn_name_orij,
769 	    integer_size, num_integers) == 0);
770 
771 	err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
772 	if (err != 0)
773 		return (err);
774 retry:
775 	err = zap_leaf_lookup(l, zn, &zeh);
776 	if (err == 0) {
777 		err = EEXIST;
778 		goto out;
779 	}
780 	if (err != ENOENT)
781 		goto out;
782 
783 	err = zap_entry_create(l, zn->zn_name_orij, zn->zn_hash, cd,
784 	    integer_size, num_integers, val, &zeh);
785 
786 	if (err == 0) {
787 		zap_increment_num_entries(zap, 1, tx);
788 	} else if (err == EAGAIN) {
789 		err = zap_expand_leaf(zap, l, zn->zn_hash, tx, &l);
790 		if (err == 0)
791 			goto retry;
792 	}
793 
794 out:
795 	zap_put_leaf_maybe_grow_ptrtbl(zap, l, tx);
796 	return (err);
797 }
798 
799 int
800 fzap_add(zap_name_t *zn,
801     uint64_t integer_size, uint64_t num_integers,
802     const void *val, dmu_tx_t *tx)
803 {
804 	int err = fzap_checksize(zn->zn_name_orij, integer_size, num_integers);
805 	if (err != 0)
806 		return (err);
807 
808 	return (fzap_add_cd(zn, integer_size, num_integers,
809 	    val, ZAP_MAXCD, tx));
810 }
811 
812 int
813 fzap_update(zap_name_t *zn,
814     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
815 {
816 	zap_leaf_t *l;
817 	int err, create;
818 	zap_entry_handle_t zeh;
819 	zap_t *zap = zn->zn_zap;
820 
821 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
822 	err = fzap_checksize(zn->zn_name_orij, integer_size, num_integers);
823 	if (err != 0)
824 		return (err);
825 
826 	err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
827 	if (err != 0)
828 		return (err);
829 retry:
830 	err = zap_leaf_lookup(l, zn, &zeh);
831 	create = (err == ENOENT);
832 	ASSERT(err == 0 || err == ENOENT);
833 
834 	if (create) {
835 		err = zap_entry_create(l, zn->zn_name_orij, zn->zn_hash,
836 		    ZAP_MAXCD, integer_size, num_integers, val, &zeh);
837 		if (err == 0)
838 			zap_increment_num_entries(zap, 1, tx);
839 	} else {
840 		err = zap_entry_update(&zeh, integer_size, num_integers, val);
841 	}
842 
843 	if (err == EAGAIN) {
844 		err = zap_expand_leaf(zap, l, zn->zn_hash, tx, &l);
845 		if (err == 0)
846 			goto retry;
847 	}
848 
849 	zap_put_leaf_maybe_grow_ptrtbl(zap, l, tx);
850 	return (err);
851 }
852 
853 int
854 fzap_length(zap_name_t *zn,
855     uint64_t *integer_size, uint64_t *num_integers)
856 {
857 	zap_leaf_t *l;
858 	int err;
859 	zap_entry_handle_t zeh;
860 
861 	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
862 	if (err != 0)
863 		return (err);
864 	err = zap_leaf_lookup(l, zn, &zeh);
865 	if (err != 0)
866 		goto out;
867 
868 	if (integer_size)
869 		*integer_size = zeh.zeh_integer_size;
870 	if (num_integers)
871 		*num_integers = zeh.zeh_num_integers;
872 out:
873 	zap_put_leaf(l);
874 	return (err);
875 }
876 
877 int
878 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
879 {
880 	zap_leaf_t *l;
881 	int err;
882 	zap_entry_handle_t zeh;
883 
884 	err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
885 	if (err != 0)
886 		return (err);
887 	err = zap_leaf_lookup(l, zn, &zeh);
888 	if (err == 0) {
889 		zap_entry_remove(&zeh);
890 		zap_increment_num_entries(zn->zn_zap, -1, tx);
891 	}
892 	zap_put_leaf(l);
893 	return (err);
894 }
895 
896 int
897 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
898     char *name)
899 {
900 	zap_cursor_t zc;
901 	zap_attribute_t *za;
902 	int err;
903 
904 	if (mask == 0)
905 		mask = -1ULL;
906 
907 	za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
908 	for (zap_cursor_init(&zc, os, zapobj);
909 	    (err = zap_cursor_retrieve(&zc, za)) == 0;
910 	    zap_cursor_advance(&zc)) {
911 		if ((za->za_first_integer & mask) == (value & mask)) {
912 			(void) strcpy(name, za->za_name);
913 			break;
914 		}
915 	}
916 	zap_cursor_fini(&zc);
917 	kmem_free(za, sizeof (zap_attribute_t));
918 	return (err);
919 }
920 
921 
922 /*
923  * Routines for iterating over the attributes.
924  */
925 
926 int
927 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
928 {
929 	int err = ENOENT;
930 	zap_entry_handle_t zeh;
931 	zap_leaf_t *l;
932 
933 	/* retrieve the next entry at or after zc_hash/zc_cd */
934 	/* if no entry, return ENOENT */
935 
936 	if (zc->zc_leaf &&
937 	    (ZAP_HASH_IDX(zc->zc_hash,
938 	    zc->zc_leaf->l_phys->l_hdr.lh_prefix_len) !=
939 	    zc->zc_leaf->l_phys->l_hdr.lh_prefix)) {
940 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
941 		zap_put_leaf(zc->zc_leaf);
942 		zc->zc_leaf = NULL;
943 	}
944 
945 again:
946 	if (zc->zc_leaf == NULL) {
947 		err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
948 		    &zc->zc_leaf);
949 		if (err != 0)
950 			return (err);
951 	} else {
952 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
953 	}
954 	l = zc->zc_leaf;
955 
956 	err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
957 
958 	if (err == ENOENT) {
959 		uint64_t nocare =
960 		    (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len)) - 1;
961 		zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
962 		zc->zc_cd = 0;
963 		if (l->l_phys->l_hdr.lh_prefix_len == 0 || zc->zc_hash == 0) {
964 			zc->zc_hash = -1ULL;
965 		} else {
966 			zap_put_leaf(zc->zc_leaf);
967 			zc->zc_leaf = NULL;
968 			goto again;
969 		}
970 	}
971 
972 	if (err == 0) {
973 		zc->zc_hash = zeh.zeh_hash;
974 		zc->zc_cd = zeh.zeh_cd;
975 		za->za_integer_length = zeh.zeh_integer_size;
976 		za->za_num_integers = zeh.zeh_num_integers;
977 		if (zeh.zeh_num_integers == 0) {
978 			za->za_first_integer = 0;
979 		} else {
980 			err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
981 			ASSERT(err == 0 || err == EOVERFLOW);
982 		}
983 		err = zap_entry_read_name(&zeh,
984 		    sizeof (za->za_name), za->za_name);
985 		ASSERT(err == 0);
986 
987 		za->za_normalization_conflict =
988 		    zap_entry_normalization_conflict(&zeh,
989 		    NULL, za->za_name, zap);
990 	}
991 	rw_exit(&zc->zc_leaf->l_rwlock);
992 	return (err);
993 }
994 
995 
996 static void
997 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
998 {
999 	int i, err;
1000 	uint64_t lastblk = 0;
1001 
1002 	/*
1003 	 * NB: if a leaf has more pointers than an entire ptrtbl block
1004 	 * can hold, then it'll be accounted for more than once, since
1005 	 * we won't have lastblk.
1006 	 */
1007 	for (i = 0; i < len; i++) {
1008 		zap_leaf_t *l;
1009 
1010 		if (tbl[i] == lastblk)
1011 			continue;
1012 		lastblk = tbl[i];
1013 
1014 		err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1015 		if (err == 0) {
1016 			zap_leaf_stats(zap, l, zs);
1017 			zap_put_leaf(l);
1018 		}
1019 	}
1020 }
1021 
1022 void
1023 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1024 {
1025 	int bs = FZAP_BLOCK_SHIFT(zap);
1026 	zs->zs_blocksize = 1ULL << bs;
1027 
1028 	/*
1029 	 * Set zap_phys_t fields
1030 	 */
1031 	zs->zs_num_leafs = zap->zap_f.zap_phys->zap_num_leafs;
1032 	zs->zs_num_entries = zap->zap_f.zap_phys->zap_num_entries;
1033 	zs->zs_num_blocks = zap->zap_f.zap_phys->zap_freeblk;
1034 	zs->zs_block_type = zap->zap_f.zap_phys->zap_block_type;
1035 	zs->zs_magic = zap->zap_f.zap_phys->zap_magic;
1036 	zs->zs_salt = zap->zap_f.zap_phys->zap_salt;
1037 
1038 	/*
1039 	 * Set zap_ptrtbl fields
1040 	 */
1041 	zs->zs_ptrtbl_len = 1ULL << zap->zap_f.zap_phys->zap_ptrtbl.zt_shift;
1042 	zs->zs_ptrtbl_nextblk = zap->zap_f.zap_phys->zap_ptrtbl.zt_nextblk;
1043 	zs->zs_ptrtbl_blks_copied =
1044 	    zap->zap_f.zap_phys->zap_ptrtbl.zt_blks_copied;
1045 	zs->zs_ptrtbl_zt_blk = zap->zap_f.zap_phys->zap_ptrtbl.zt_blk;
1046 	zs->zs_ptrtbl_zt_numblks = zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks;
1047 	zs->zs_ptrtbl_zt_shift = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift;
1048 
1049 	if (zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks == 0) {
1050 		/* the ptrtbl is entirely in the header block. */
1051 		zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1052 		    1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1053 	} else {
1054 		int b;
1055 
1056 		dmu_prefetch(zap->zap_objset, zap->zap_object,
1057 		    zap->zap_f.zap_phys->zap_ptrtbl.zt_blk << bs,
1058 		    zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks << bs);
1059 
1060 		for (b = 0; b < zap->zap_f.zap_phys->zap_ptrtbl.zt_numblks;
1061 		    b++) {
1062 			dmu_buf_t *db;
1063 			int err;
1064 
1065 			err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1066 			    (zap->zap_f.zap_phys->zap_ptrtbl.zt_blk + b) << bs,
1067 			    FTAG, &db);
1068 			if (err == 0) {
1069 				zap_stats_ptrtbl(zap, db->db_data,
1070 				    1<<(bs-3), zs);
1071 				dmu_buf_rele(db, FTAG);
1072 			}
1073 		}
1074 	}
1075 }
1076