xref: /illumos-gate/usr/src/uts/common/fs/zfs/zap_leaf.c (revision 9a5d73e03cd3312ddb571a748c40a63c58bd66e5)
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  * The 512-byte leaf is broken into 32 16-byte chunks.
30  * chunk number n means l_chunk[n], even though the header precedes it.
31  * the names are stored null-terminated.
32  */
33 
34 #include <sys/zfs_context.h>
35 #include <sys/zap.h>
36 #include <sys/zap_impl.h>
37 #include <sys/zap_leaf.h>
38 #include <sys/spa.h>
39 #include <sys/dmu.h>
40 
41 static uint16_t *zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry);
42 
43 #define	CHAIN_END 0xffff /* end of the chunk chain */
44 
45 /* half the (current) minimum block size */
46 #define	MAX_ARRAY_BYTES (8<<10)
47 
48 #define	LEAF_HASH(l, h) \
49 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
50 	((h) >> (64 - ZAP_LEAF_HASH_SHIFT(l)-(l)->l_phys->l_hdr.lh_prefix_len)))
51 
52 #define	LEAF_HASH_ENTPTR(l, h) (&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
53 
54 
55 static void
56 zap_memset(void *a, int c, size_t n)
57 {
58 	char *cp = a;
59 	char *cpend = cp + n;
60 
61 	while (cp < cpend)
62 		*cp++ = c;
63 }
64 
65 static void
66 stv(int len, void *addr, uint64_t value)
67 {
68 	switch (len) {
69 	case 1:
70 		*(uint8_t *)addr = value;
71 		return;
72 	case 2:
73 		*(uint16_t *)addr = value;
74 		return;
75 	case 4:
76 		*(uint32_t *)addr = value;
77 		return;
78 	case 8:
79 		*(uint64_t *)addr = value;
80 		return;
81 	}
82 	ASSERT(!"bad int len");
83 }
84 
85 static uint64_t
86 ldv(int len, const void *addr)
87 {
88 	switch (len) {
89 	case 1:
90 		return (*(uint8_t *)addr);
91 	case 2:
92 		return (*(uint16_t *)addr);
93 	case 4:
94 		return (*(uint32_t *)addr);
95 	case 8:
96 		return (*(uint64_t *)addr);
97 	}
98 	ASSERT(!"bad int len");
99 	return (0xFEEDFACEDEADBEEFULL);
100 }
101 
102 void
103 zap_leaf_byteswap(zap_leaf_phys_t *buf, int size)
104 {
105 	int i;
106 	zap_leaf_t l;
107 	l.l_bs = highbit(size)-1;
108 	l.l_phys = buf;
109 
110 	buf->l_hdr.lh_block_type = 	BSWAP_64(buf->l_hdr.lh_block_type);
111 	buf->l_hdr.lh_prefix = 		BSWAP_64(buf->l_hdr.lh_prefix);
112 	buf->l_hdr.lh_magic = 		BSWAP_32(buf->l_hdr.lh_magic);
113 	buf->l_hdr.lh_nfree = 		BSWAP_16(buf->l_hdr.lh_nfree);
114 	buf->l_hdr.lh_nentries = 	BSWAP_16(buf->l_hdr.lh_nentries);
115 	buf->l_hdr.lh_prefix_len = 	BSWAP_16(buf->l_hdr.lh_prefix_len);
116 	buf->l_hdr.lh_freelist = 	BSWAP_16(buf->l_hdr.lh_freelist);
117 
118 	for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++)
119 		buf->l_hash[i] = BSWAP_16(buf->l_hash[i]);
120 
121 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
122 		zap_leaf_chunk_t *lc = &ZAP_LEAF_CHUNK(&l, i);
123 		struct zap_leaf_entry *le;
124 
125 		switch (lc->l_free.lf_type) {
126 		case ZAP_CHUNK_ENTRY:
127 			le = &lc->l_entry;
128 
129 			le->le_type =		BSWAP_8(le->le_type);
130 			le->le_int_size =	BSWAP_8(le->le_int_size);
131 			le->le_next =		BSWAP_16(le->le_next);
132 			le->le_name_chunk =	BSWAP_16(le->le_name_chunk);
133 			le->le_name_length =	BSWAP_16(le->le_name_length);
134 			le->le_value_chunk =	BSWAP_16(le->le_value_chunk);
135 			le->le_value_length =	BSWAP_16(le->le_value_length);
136 			le->le_cd =		BSWAP_32(le->le_cd);
137 			le->le_hash =		BSWAP_64(le->le_hash);
138 			break;
139 		case ZAP_CHUNK_FREE:
140 			lc->l_free.lf_type =	BSWAP_8(lc->l_free.lf_type);
141 			lc->l_free.lf_next =	BSWAP_16(lc->l_free.lf_next);
142 			break;
143 		case ZAP_CHUNK_ARRAY:
144 			lc->l_array.la_type =	BSWAP_8(lc->l_array.la_type);
145 			lc->l_array.la_next =	BSWAP_16(lc->l_array.la_next);
146 			/* la_array doesn't need swapping */
147 			break;
148 		default:
149 			ASSERT(!"bad leaf type");
150 		}
151 	}
152 }
153 
154 void
155 zap_leaf_init(zap_leaf_t *l, boolean_t sort)
156 {
157 	int i;
158 
159 	l->l_bs = highbit(l->l_dbuf->db_size)-1;
160 	zap_memset(&l->l_phys->l_hdr, 0, sizeof (struct zap_leaf_header));
161 	zap_memset(l->l_phys->l_hash, CHAIN_END, 2*ZAP_LEAF_HASH_NUMENTRIES(l));
162 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
163 		ZAP_LEAF_CHUNK(l, i).l_free.lf_type = ZAP_CHUNK_FREE;
164 		ZAP_LEAF_CHUNK(l, i).l_free.lf_next = i+1;
165 	}
166 	ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)-1).l_free.lf_next = CHAIN_END;
167 	l->l_phys->l_hdr.lh_block_type = ZBT_LEAF;
168 	l->l_phys->l_hdr.lh_magic = ZAP_LEAF_MAGIC;
169 	l->l_phys->l_hdr.lh_nfree = ZAP_LEAF_NUMCHUNKS(l);
170 	if (sort)
171 		l->l_phys->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
172 }
173 
174 /*
175  * Routines which manipulate leaf chunks (l_chunk[]).
176  */
177 
178 static uint16_t
179 zap_leaf_chunk_alloc(zap_leaf_t *l)
180 {
181 	int chunk;
182 
183 	ASSERT(l->l_phys->l_hdr.lh_nfree > 0);
184 
185 	chunk = l->l_phys->l_hdr.lh_freelist;
186 	ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
187 	ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_free.lf_type, ==, ZAP_CHUNK_FREE);
188 
189 	l->l_phys->l_hdr.lh_freelist = ZAP_LEAF_CHUNK(l, chunk).l_free.lf_next;
190 
191 	l->l_phys->l_hdr.lh_nfree--;
192 
193 	return (chunk);
194 }
195 
196 static void
197 zap_leaf_chunk_free(zap_leaf_t *l, uint16_t chunk)
198 {
199 	struct zap_leaf_free *zlf = &ZAP_LEAF_CHUNK(l, chunk).l_free;
200 	ASSERT3U(l->l_phys->l_hdr.lh_nfree, <, ZAP_LEAF_NUMCHUNKS(l));
201 	ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
202 	ASSERT(zlf->lf_type != ZAP_CHUNK_FREE);
203 
204 	zlf->lf_type = ZAP_CHUNK_FREE;
205 	zlf->lf_next = l->l_phys->l_hdr.lh_freelist;
206 	bzero(zlf->lf_pad, sizeof (zlf->lf_pad)); /* help it to compress */
207 	l->l_phys->l_hdr.lh_freelist = chunk;
208 
209 	l->l_phys->l_hdr.lh_nfree++;
210 }
211 
212 /*
213  * Routines which manipulate leaf arrays (zap_leaf_array type chunks).
214  */
215 
216 static uint16_t
217 zap_leaf_array_create(zap_leaf_t *l, const char *buf,
218 	int integer_size, int num_integers)
219 {
220 	uint16_t chunk_head;
221 	uint16_t *chunkp = &chunk_head;
222 	int byten = 0;
223 	uint64_t value;
224 	int shift = (integer_size-1)*8;
225 	int len = num_integers;
226 
227 	ASSERT3U(num_integers * integer_size, <, MAX_ARRAY_BYTES);
228 
229 	while (len > 0) {
230 		uint16_t chunk = zap_leaf_chunk_alloc(l);
231 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
232 		int i;
233 
234 		la->la_type = ZAP_CHUNK_ARRAY;
235 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) {
236 			if (byten == 0)
237 				value = ldv(integer_size, buf);
238 			la->la_array[i] = value >> shift;
239 			value <<= 8;
240 			if (++byten == integer_size) {
241 				byten = 0;
242 				buf += integer_size;
243 				if (--len == 0)
244 					break;
245 			}
246 		}
247 
248 		*chunkp = chunk;
249 		chunkp = &la->la_next;
250 	}
251 	*chunkp = CHAIN_END;
252 
253 	return (chunk_head);
254 }
255 
256 static void
257 zap_leaf_array_free(zap_leaf_t *l, uint16_t *chunkp)
258 {
259 	uint16_t chunk = *chunkp;
260 
261 	*chunkp = CHAIN_END;
262 
263 	while (chunk != CHAIN_END) {
264 		int nextchunk = ZAP_LEAF_CHUNK(l, chunk).l_array.la_next;
265 		ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_array.la_type, ==,
266 		    ZAP_CHUNK_ARRAY);
267 		zap_leaf_chunk_free(l, chunk);
268 		chunk = nextchunk;
269 	}
270 }
271 
272 /* array_len and buf_len are in integers, not bytes */
273 static void
274 zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk,
275     int array_int_len, int array_len, int buf_int_len, uint64_t buf_len,
276     char *buf)
277 {
278 	int len = MIN(array_len, buf_len);
279 	int byten = 0;
280 	uint64_t value = 0;
281 
282 	ASSERT3U(array_int_len, <=, buf_int_len);
283 
284 	/* Fast path for one 8-byte integer */
285 	if (array_int_len == 8 && buf_int_len == 8 && len == 1) {
286 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
287 		uint8_t *ip = la->la_array;
288 		uint64_t *buf64 = (uint64_t *)buf;
289 
290 		*buf64 = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 |
291 		    (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 |
292 		    (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 |
293 		    (uint64_t)ip[6] << 8 | (uint64_t)ip[7];
294 		return;
295 	}
296 
297 	/* Fast path for an array of 1-byte integers (eg. the entry name) */
298 	if (array_int_len == 1 && buf_int_len == 1 &&
299 	    buf_len > array_len + ZAP_LEAF_ARRAY_BYTES) {
300 		while (chunk != CHAIN_END) {
301 			struct zap_leaf_array *la =
302 			    &ZAP_LEAF_CHUNK(l, chunk).l_array;
303 			bcopy(la->la_array, buf, ZAP_LEAF_ARRAY_BYTES);
304 			buf += ZAP_LEAF_ARRAY_BYTES;
305 			chunk = la->la_next;
306 		}
307 		return;
308 	}
309 
310 	while (len > 0) {
311 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
312 		int i;
313 
314 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
315 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
316 			value = (value << 8) | la->la_array[i];
317 			byten++;
318 			if (byten == array_int_len) {
319 				stv(buf_int_len, buf, value);
320 				byten = 0;
321 				len--;
322 				if (len == 0)
323 					return;
324 				buf += buf_int_len;
325 			}
326 		}
327 		chunk = la->la_next;
328 	}
329 }
330 
331 /*
332  * Only to be used on 8-bit arrays.
333  * array_len is actual len in bytes (not encoded le_value_length).
334  * namenorm is null-terminated.
335  */
336 static boolean_t
337 zap_leaf_array_match(zap_leaf_t *l, zap_name_t *zn, int chunk, int array_len)
338 {
339 	int bseen = 0;
340 
341 	if (zn->zn_matchtype == MT_FIRST) {
342 		char *thisname = kmem_alloc(array_len, KM_SLEEP);
343 		boolean_t match;
344 
345 		zap_leaf_array_read(l, chunk, 1, array_len, 1,
346 		    array_len, thisname);
347 		match = zap_match(zn, thisname);
348 		kmem_free(thisname, array_len);
349 		return (match);
350 	}
351 
352 	/* Fast path for exact matching */
353 	while (bseen < array_len) {
354 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
355 		int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);
356 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
357 		if (bcmp(la->la_array, zn->zn_name_orij + bseen, toread))
358 			break;
359 		chunk = la->la_next;
360 		bseen += toread;
361 	}
362 	return (bseen == array_len);
363 }
364 
365 /*
366  * Routines which manipulate leaf entries.
367  */
368 
369 int
370 zap_leaf_lookup(zap_leaf_t *l, zap_name_t *zn, zap_entry_handle_t *zeh)
371 {
372 	uint16_t *chunkp;
373 	struct zap_leaf_entry *le;
374 
375 	ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
376 
377 again:
378 	for (chunkp = LEAF_HASH_ENTPTR(l, zn->zn_hash);
379 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
380 		uint16_t chunk = *chunkp;
381 		le = ZAP_LEAF_ENTRY(l, chunk);
382 
383 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
384 		ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
385 
386 		if (le->le_hash != zn->zn_hash)
387 			continue;
388 
389 		/*
390 		 * NB: the entry chain is always sorted by cd on
391 		 * normalized zap objects, so this will find the
392 		 * lowest-cd match for MT_FIRST.
393 		 */
394 		ASSERT(zn->zn_matchtype == MT_EXACT ||
395 		    (l->l_phys->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED));
396 		if (zap_leaf_array_match(l, zn, le->le_name_chunk,
397 		    le->le_name_length)) {
398 			zeh->zeh_num_integers = le->le_value_length;
399 			zeh->zeh_integer_size = le->le_int_size;
400 			zeh->zeh_cd = le->le_cd;
401 			zeh->zeh_hash = le->le_hash;
402 			zeh->zeh_chunkp = chunkp;
403 			zeh->zeh_leaf = l;
404 			return (0);
405 		}
406 	}
407 
408 	/*
409 	 * NB: we could of course do this in one pass, but that would be
410 	 * a pain.  We'll see if MT_BEST is even used much.
411 	 */
412 	if (zn->zn_matchtype == MT_BEST) {
413 		zn->zn_matchtype = MT_FIRST;
414 		goto again;
415 	}
416 
417 	return (ENOENT);
418 }
419 
420 /* Return (h1,cd1 >= h2,cd2) */
421 #define	HCD_GTEQ(h1, cd1, h2, cd2) \
422 	((h1 > h2) ? TRUE : ((h1 == h2 && cd1 >= cd2) ? TRUE : FALSE))
423 
424 int
425 zap_leaf_lookup_closest(zap_leaf_t *l,
426     uint64_t h, uint32_t cd, zap_entry_handle_t *zeh)
427 {
428 	uint16_t chunk;
429 	uint64_t besth = -1ULL;
430 	uint32_t bestcd = ZAP_MAXCD;
431 	uint16_t bestlh = ZAP_LEAF_HASH_NUMENTRIES(l)-1;
432 	uint16_t lh;
433 	struct zap_leaf_entry *le;
434 
435 	ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
436 
437 	for (lh = LEAF_HASH(l, h); lh <= bestlh; lh++) {
438 		for (chunk = l->l_phys->l_hash[lh];
439 		    chunk != CHAIN_END; chunk = le->le_next) {
440 			le = ZAP_LEAF_ENTRY(l, chunk);
441 
442 			ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
443 			ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
444 
445 			if (HCD_GTEQ(le->le_hash, le->le_cd, h, cd) &&
446 			    HCD_GTEQ(besth, bestcd, le->le_hash, le->le_cd)) {
447 				ASSERT3U(bestlh, >=, lh);
448 				bestlh = lh;
449 				besth = le->le_hash;
450 				bestcd = le->le_cd;
451 
452 				zeh->zeh_num_integers = le->le_value_length;
453 				zeh->zeh_integer_size = le->le_int_size;
454 				zeh->zeh_cd = le->le_cd;
455 				zeh->zeh_hash = le->le_hash;
456 				zeh->zeh_fakechunk = chunk;
457 				zeh->zeh_chunkp = &zeh->zeh_fakechunk;
458 				zeh->zeh_leaf = l;
459 			}
460 		}
461 	}
462 
463 	return (bestcd == ZAP_MAXCD ? ENOENT : 0);
464 }
465 
466 int
467 zap_entry_read(const zap_entry_handle_t *zeh,
468     uint8_t integer_size, uint64_t num_integers, void *buf)
469 {
470 	struct zap_leaf_entry *le =
471 	    ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
472 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
473 
474 	if (le->le_int_size > integer_size)
475 		return (EINVAL);
476 
477 	zap_leaf_array_read(zeh->zeh_leaf, le->le_value_chunk, le->le_int_size,
478 	    le->le_value_length, integer_size, num_integers, buf);
479 
480 	if (zeh->zeh_num_integers > num_integers)
481 		return (EOVERFLOW);
482 	return (0);
483 
484 }
485 
486 int
487 zap_entry_read_name(const zap_entry_handle_t *zeh, uint16_t buflen, char *buf)
488 {
489 	struct zap_leaf_entry *le =
490 	    ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
491 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
492 
493 	zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 1,
494 	    le->le_name_length, 1, buflen, buf);
495 	if (le->le_name_length > buflen)
496 		return (EOVERFLOW);
497 	return (0);
498 }
499 
500 int
501 zap_entry_update(zap_entry_handle_t *zeh,
502 	uint8_t integer_size, uint64_t num_integers, const void *buf)
503 {
504 	int delta_chunks;
505 	zap_leaf_t *l = zeh->zeh_leaf;
506 	struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, *zeh->zeh_chunkp);
507 
508 	delta_chunks = ZAP_LEAF_ARRAY_NCHUNKS(num_integers * integer_size) -
509 	    ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_length * le->le_int_size);
510 
511 	if ((int)l->l_phys->l_hdr.lh_nfree < delta_chunks)
512 		return (EAGAIN);
513 
514 	/*
515 	 * We should search other chained leaves (via
516 	 * zap_entry_remove,create?) otherwise returning EAGAIN will
517 	 * just send us into an infinite loop if we have to chain
518 	 * another leaf block, rather than being able to split this
519 	 * block.
520 	 */
521 
522 	zap_leaf_array_free(l, &le->le_value_chunk);
523 	le->le_value_chunk =
524 	    zap_leaf_array_create(l, buf, integer_size, num_integers);
525 	le->le_value_length = num_integers;
526 	le->le_int_size = integer_size;
527 	return (0);
528 }
529 
530 void
531 zap_entry_remove(zap_entry_handle_t *zeh)
532 {
533 	uint16_t entry_chunk;
534 	struct zap_leaf_entry *le;
535 	zap_leaf_t *l = zeh->zeh_leaf;
536 
537 	ASSERT3P(zeh->zeh_chunkp, !=, &zeh->zeh_fakechunk);
538 
539 	entry_chunk = *zeh->zeh_chunkp;
540 	le = ZAP_LEAF_ENTRY(l, entry_chunk);
541 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
542 
543 	zap_leaf_array_free(l, &le->le_name_chunk);
544 	zap_leaf_array_free(l, &le->le_value_chunk);
545 
546 	*zeh->zeh_chunkp = le->le_next;
547 	zap_leaf_chunk_free(l, entry_chunk);
548 
549 	l->l_phys->l_hdr.lh_nentries--;
550 }
551 
552 int
553 zap_entry_create(zap_leaf_t *l, const char *name, uint64_t h, uint32_t cd,
554     uint8_t integer_size, uint64_t num_integers, const void *buf,
555     zap_entry_handle_t *zeh)
556 {
557 	uint16_t chunk;
558 	uint16_t *chunkp;
559 	struct zap_leaf_entry *le;
560 	uint64_t namelen, valuelen;
561 	int numchunks;
562 
563 	valuelen = integer_size * num_integers;
564 	namelen = strlen(name) + 1;
565 	ASSERT(namelen >= 2);
566 
567 	numchunks = 1 + ZAP_LEAF_ARRAY_NCHUNKS(namelen) +
568 	    ZAP_LEAF_ARRAY_NCHUNKS(valuelen);
569 	if (numchunks > ZAP_LEAF_NUMCHUNKS(l))
570 		return (E2BIG);
571 
572 	if (cd == ZAP_MAXCD) {
573 		/* find the lowest unused cd */
574 		if (l->l_phys->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED) {
575 			cd = 0;
576 
577 			for (chunk = *LEAF_HASH_ENTPTR(l, h);
578 			    chunk != CHAIN_END; chunk = le->le_next) {
579 				le = ZAP_LEAF_ENTRY(l, chunk);
580 				if (le->le_cd > cd)
581 					break;
582 				if (le->le_hash == h) {
583 					ASSERT3U(cd, ==, le->le_cd);
584 					cd++;
585 				}
586 			}
587 		} else {
588 			/* old unsorted format; do it the O(n^2) way */
589 			for (cd = 0; cd < ZAP_MAXCD; cd++) {
590 				for (chunk = *LEAF_HASH_ENTPTR(l, h);
591 				    chunk != CHAIN_END; chunk = le->le_next) {
592 					le = ZAP_LEAF_ENTRY(l, chunk);
593 					if (le->le_hash == h &&
594 					    le->le_cd == cd) {
595 						break;
596 					}
597 				}
598 				/* If this cd is not in use, we are good. */
599 				if (chunk == CHAIN_END)
600 					break;
601 			}
602 		}
603 		/*
604 		 * we would run out of space in a block before we could
605 		 * have ZAP_MAXCD entries
606 		 */
607 		ASSERT3U(cd, <, ZAP_MAXCD);
608 	}
609 
610 	if (l->l_phys->l_hdr.lh_nfree < numchunks)
611 		return (EAGAIN);
612 
613 	/* make the entry */
614 	chunk = zap_leaf_chunk_alloc(l);
615 	le = ZAP_LEAF_ENTRY(l, chunk);
616 	le->le_type = ZAP_CHUNK_ENTRY;
617 	le->le_name_chunk = zap_leaf_array_create(l, name, 1, namelen);
618 	le->le_name_length = namelen;
619 	le->le_value_chunk =
620 	    zap_leaf_array_create(l, buf, integer_size, num_integers);
621 	le->le_value_length = num_integers;
622 	le->le_int_size = integer_size;
623 	le->le_hash = h;
624 	le->le_cd = cd;
625 
626 	/* link it into the hash chain */
627 	/* XXX if we did the search above, we could just use that */
628 	chunkp = zap_leaf_rehash_entry(l, chunk);
629 
630 	l->l_phys->l_hdr.lh_nentries++;
631 
632 	zeh->zeh_leaf = l;
633 	zeh->zeh_num_integers = num_integers;
634 	zeh->zeh_integer_size = le->le_int_size;
635 	zeh->zeh_cd = le->le_cd;
636 	zeh->zeh_hash = le->le_hash;
637 	zeh->zeh_chunkp = chunkp;
638 
639 	return (0);
640 }
641 
642 /*
643  * Determine if there is another entry with the same normalized form.
644  * For performance purposes, either zn or name must be provided (the
645  * other can be NULL).  Note, there usually won't be any hash
646  * conflicts, in which case we don't need the concatenated/normalized
647  * form of the name.  But all callers have one of these on hand anyway,
648  * so might as well take advantage.  A cleaner but slower interface
649  * would accept neither argument, and compute the normalized name as
650  * needed (using zap_name_alloc(zap_entry_read_name(zeh))).
651  */
652 boolean_t
653 zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn,
654     const char *name, zap_t *zap)
655 {
656 	uint64_t chunk;
657 	struct zap_leaf_entry *le;
658 	boolean_t allocdzn = B_FALSE;
659 
660 	if (zap->zap_normflags == 0)
661 		return (B_FALSE);
662 
663 	for (chunk = *LEAF_HASH_ENTPTR(zeh->zeh_leaf, zeh->zeh_hash);
664 	    chunk != CHAIN_END; chunk = le->le_next) {
665 		le = ZAP_LEAF_ENTRY(zeh->zeh_leaf, chunk);
666 		if (le->le_hash != zeh->zeh_hash)
667 			continue;
668 		if (le->le_cd == zeh->zeh_cd)
669 			continue;
670 
671 		if (zn == NULL) {
672 			zn = zap_name_alloc(zap, name, MT_FIRST);
673 			allocdzn = B_TRUE;
674 		}
675 		if (zap_leaf_array_match(zeh->zeh_leaf, zn,
676 		    le->le_name_chunk, le->le_name_length)) {
677 			if (allocdzn)
678 				zap_name_free(zn);
679 			return (B_TRUE);
680 		}
681 	}
682 	if (allocdzn)
683 		zap_name_free(zn);
684 	return (B_FALSE);
685 }
686 
687 /*
688  * Routines for transferring entries between leafs.
689  */
690 
691 static uint16_t *
692 zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry)
693 {
694 	struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry);
695 	struct zap_leaf_entry *le2;
696 	uint16_t *chunkp;
697 
698 	/*
699 	 * keep the entry chain sorted by cd
700 	 * NB: this will not cause problems for unsorted leafs, though
701 	 * it is unnecessary there.
702 	 */
703 	for (chunkp = LEAF_HASH_ENTPTR(l, le->le_hash);
704 	    *chunkp != CHAIN_END; chunkp = &le2->le_next) {
705 		le2 = ZAP_LEAF_ENTRY(l, *chunkp);
706 		if (le2->le_cd > le->le_cd)
707 			break;
708 	}
709 
710 	le->le_next = *chunkp;
711 	*chunkp = entry;
712 	return (chunkp);
713 }
714 
715 static uint16_t
716 zap_leaf_transfer_array(zap_leaf_t *l, uint16_t chunk, zap_leaf_t *nl)
717 {
718 	uint16_t new_chunk;
719 	uint16_t *nchunkp = &new_chunk;
720 
721 	while (chunk != CHAIN_END) {
722 		uint16_t nchunk = zap_leaf_chunk_alloc(nl);
723 		struct zap_leaf_array *nla =
724 		    &ZAP_LEAF_CHUNK(nl, nchunk).l_array;
725 		struct zap_leaf_array *la =
726 		    &ZAP_LEAF_CHUNK(l, chunk).l_array;
727 		int nextchunk = la->la_next;
728 
729 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
730 		ASSERT3U(nchunk, <, ZAP_LEAF_NUMCHUNKS(l));
731 
732 		*nla = *la; /* structure assignment */
733 
734 		zap_leaf_chunk_free(l, chunk);
735 		chunk = nextchunk;
736 		*nchunkp = nchunk;
737 		nchunkp = &nla->la_next;
738 	}
739 	*nchunkp = CHAIN_END;
740 	return (new_chunk);
741 }
742 
743 static void
744 zap_leaf_transfer_entry(zap_leaf_t *l, int entry, zap_leaf_t *nl)
745 {
746 	struct zap_leaf_entry *le, *nle;
747 	uint16_t chunk;
748 
749 	le = ZAP_LEAF_ENTRY(l, entry);
750 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
751 
752 	chunk = zap_leaf_chunk_alloc(nl);
753 	nle = ZAP_LEAF_ENTRY(nl, chunk);
754 	*nle = *le; /* structure assignment */
755 
756 	(void) zap_leaf_rehash_entry(nl, chunk);
757 
758 	nle->le_name_chunk = zap_leaf_transfer_array(l, le->le_name_chunk, nl);
759 	nle->le_value_chunk =
760 	    zap_leaf_transfer_array(l, le->le_value_chunk, nl);
761 
762 	zap_leaf_chunk_free(l, entry);
763 
764 	l->l_phys->l_hdr.lh_nentries--;
765 	nl->l_phys->l_hdr.lh_nentries++;
766 }
767 
768 /*
769  * Transfer the entries whose hash prefix ends in 1 to the new leaf.
770  */
771 void
772 zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort)
773 {
774 	int i;
775 	int bit = 64 - 1 - l->l_phys->l_hdr.lh_prefix_len;
776 
777 	/* set new prefix and prefix_len */
778 	l->l_phys->l_hdr.lh_prefix <<= 1;
779 	l->l_phys->l_hdr.lh_prefix_len++;
780 	nl->l_phys->l_hdr.lh_prefix = l->l_phys->l_hdr.lh_prefix | 1;
781 	nl->l_phys->l_hdr.lh_prefix_len = l->l_phys->l_hdr.lh_prefix_len;
782 
783 	/* break existing hash chains */
784 	zap_memset(l->l_phys->l_hash, CHAIN_END, 2*ZAP_LEAF_HASH_NUMENTRIES(l));
785 
786 	if (sort)
787 		l->l_phys->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
788 
789 	/*
790 	 * Transfer entries whose hash bit 'bit' is set to nl; rehash
791 	 * the remaining entries
792 	 *
793 	 * NB: We could find entries via the hashtable instead. That
794 	 * would be O(hashents+numents) rather than O(numblks+numents),
795 	 * but this accesses memory more sequentially, and when we're
796 	 * called, the block is usually pretty full.
797 	 */
798 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
799 		struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, i);
800 		if (le->le_type != ZAP_CHUNK_ENTRY)
801 			continue;
802 
803 		if (le->le_hash & (1ULL << bit))
804 			zap_leaf_transfer_entry(l, i, nl);
805 		else
806 			(void) zap_leaf_rehash_entry(l, i);
807 	}
808 }
809 
810 void
811 zap_leaf_stats(zap_t *zap, zap_leaf_t *l, zap_stats_t *zs)
812 {
813 	int i, n;
814 
815 	n = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift -
816 	    l->l_phys->l_hdr.lh_prefix_len;
817 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
818 	zs->zs_leafs_with_2n_pointers[n]++;
819 
820 
821 	n = l->l_phys->l_hdr.lh_nentries/5;
822 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
823 	zs->zs_blocks_with_n5_entries[n]++;
824 
825 	n = ((1<<FZAP_BLOCK_SHIFT(zap)) -
826 	    l->l_phys->l_hdr.lh_nfree * (ZAP_LEAF_ARRAY_BYTES+1))*10 /
827 	    (1<<FZAP_BLOCK_SHIFT(zap));
828 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
829 	zs->zs_blocks_n_tenths_full[n]++;
830 
831 	for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(l); i++) {
832 		int nentries = 0;
833 		int chunk = l->l_phys->l_hash[i];
834 
835 		while (chunk != CHAIN_END) {
836 			struct zap_leaf_entry *le =
837 			    ZAP_LEAF_ENTRY(l, chunk);
838 
839 			n = 1 + ZAP_LEAF_ARRAY_NCHUNKS(le->le_name_length) +
840 			    ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_length *
841 			    le->le_int_size);
842 			n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
843 			zs->zs_entries_using_n_chunks[n]++;
844 
845 			chunk = le->le_next;
846 			nentries++;
847 		}
848 
849 		n = nentries;
850 		n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
851 		zs->zs_buckets_with_n_entries[n]++;
852 	}
853 }
854