xref: /illumos-gate/usr/src/uts/common/fs/zfs/zap_leaf.c (revision a74909fd1b8a12f99797bbadb7a8af8bdca8596c)
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) 2013, 2014 by Delphix. All rights reserved.
24  */
25 
26 /*
27  * The 512-byte leaf is broken into 32 16-byte chunks.
28  * chunk number n means l_chunk[n], even though the header precedes it.
29  * the names are stored null-terminated.
30  */
31 
32 #include <sys/zio.h>
33 #include <sys/spa.h>
34 #include <sys/dmu.h>
35 #include <sys/zfs_context.h>
36 #include <sys/fs/zfs.h>
37 #include <sys/zap.h>
38 #include <sys/zap_impl.h>
39 #include <sys/zap_leaf.h>
40 #include <sys/arc.h>
41 
42 static uint16_t *zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry);
43 
44 #define	CHAIN_END 0xffff /* end of the chunk chain */
45 
46 /* half the (current) minimum block size */
47 #define	MAX_ARRAY_BYTES (8<<10)
48 
49 #define	LEAF_HASH(l, h) \
50 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
51 	((h) >> \
52 	(64 - ZAP_LEAF_HASH_SHIFT(l) - zap_leaf_phys(l)->l_hdr.lh_prefix_len)))
53 
54 #define	LEAF_HASH_ENTPTR(l, h) (&zap_leaf_phys(l)->l_hash[LEAF_HASH(l, h)])
55 
56 extern inline zap_leaf_phys_t *zap_leaf_phys(zap_leaf_t *l);
57 
58 static void
59 zap_memset(void *a, int c, size_t n)
60 {
61 	char *cp = a;
62 	char *cpend = cp + n;
63 
64 	while (cp < cpend)
65 		*cp++ = c;
66 }
67 
68 static void
69 stv(int len, void *addr, uint64_t value)
70 {
71 	switch (len) {
72 	case 1:
73 		*(uint8_t *)addr = value;
74 		return;
75 	case 2:
76 		*(uint16_t *)addr = value;
77 		return;
78 	case 4:
79 		*(uint32_t *)addr = value;
80 		return;
81 	case 8:
82 		*(uint64_t *)addr = value;
83 		return;
84 	}
85 	ASSERT(!"bad int len");
86 }
87 
88 static uint64_t
89 ldv(int len, const void *addr)
90 {
91 	switch (len) {
92 	case 1:
93 		return (*(uint8_t *)addr);
94 	case 2:
95 		return (*(uint16_t *)addr);
96 	case 4:
97 		return (*(uint32_t *)addr);
98 	case 8:
99 		return (*(uint64_t *)addr);
100 	}
101 	ASSERT(!"bad int len");
102 	return (0xFEEDFACEDEADBEEFULL);
103 }
104 
105 void
106 zap_leaf_byteswap(zap_leaf_phys_t *buf, int size)
107 {
108 	int i;
109 	zap_leaf_t l;
110 	dmu_buf_t l_dbuf;
111 
112 	l_dbuf.db_data = buf;
113 	l.l_bs = highbit64(size) - 1;
114 	l.l_dbuf = &l_dbuf;
115 
116 	buf->l_hdr.lh_block_type =	BSWAP_64(buf->l_hdr.lh_block_type);
117 	buf->l_hdr.lh_prefix =		BSWAP_64(buf->l_hdr.lh_prefix);
118 	buf->l_hdr.lh_magic =		BSWAP_32(buf->l_hdr.lh_magic);
119 	buf->l_hdr.lh_nfree =		BSWAP_16(buf->l_hdr.lh_nfree);
120 	buf->l_hdr.lh_nentries =	BSWAP_16(buf->l_hdr.lh_nentries);
121 	buf->l_hdr.lh_prefix_len =	BSWAP_16(buf->l_hdr.lh_prefix_len);
122 	buf->l_hdr.lh_freelist =	BSWAP_16(buf->l_hdr.lh_freelist);
123 
124 	for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++)
125 		buf->l_hash[i] = BSWAP_16(buf->l_hash[i]);
126 
127 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
128 		zap_leaf_chunk_t *lc = &ZAP_LEAF_CHUNK(&l, i);
129 		struct zap_leaf_entry *le;
130 
131 		switch (lc->l_free.lf_type) {
132 		case ZAP_CHUNK_ENTRY:
133 			le = &lc->l_entry;
134 
135 			le->le_type =		BSWAP_8(le->le_type);
136 			le->le_value_intlen =	BSWAP_8(le->le_value_intlen);
137 			le->le_next =		BSWAP_16(le->le_next);
138 			le->le_name_chunk =	BSWAP_16(le->le_name_chunk);
139 			le->le_name_numints =	BSWAP_16(le->le_name_numints);
140 			le->le_value_chunk =	BSWAP_16(le->le_value_chunk);
141 			le->le_value_numints =	BSWAP_16(le->le_value_numints);
142 			le->le_cd =		BSWAP_32(le->le_cd);
143 			le->le_hash =		BSWAP_64(le->le_hash);
144 			break;
145 		case ZAP_CHUNK_FREE:
146 			lc->l_free.lf_type =	BSWAP_8(lc->l_free.lf_type);
147 			lc->l_free.lf_next =	BSWAP_16(lc->l_free.lf_next);
148 			break;
149 		case ZAP_CHUNK_ARRAY:
150 			lc->l_array.la_type =	BSWAP_8(lc->l_array.la_type);
151 			lc->l_array.la_next =	BSWAP_16(lc->l_array.la_next);
152 			/* la_array doesn't need swapping */
153 			break;
154 		default:
155 			ASSERT(!"bad leaf type");
156 		}
157 	}
158 }
159 
160 void
161 zap_leaf_init(zap_leaf_t *l, boolean_t sort)
162 {
163 	int i;
164 
165 	l->l_bs = highbit64(l->l_dbuf->db_size) - 1;
166 	zap_memset(&zap_leaf_phys(l)->l_hdr, 0,
167 	    sizeof (struct zap_leaf_header));
168 	zap_memset(zap_leaf_phys(l)->l_hash, CHAIN_END,
169 	    2*ZAP_LEAF_HASH_NUMENTRIES(l));
170 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
171 		ZAP_LEAF_CHUNK(l, i).l_free.lf_type = ZAP_CHUNK_FREE;
172 		ZAP_LEAF_CHUNK(l, i).l_free.lf_next = i+1;
173 	}
174 	ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)-1).l_free.lf_next = CHAIN_END;
175 	zap_leaf_phys(l)->l_hdr.lh_block_type = ZBT_LEAF;
176 	zap_leaf_phys(l)->l_hdr.lh_magic = ZAP_LEAF_MAGIC;
177 	zap_leaf_phys(l)->l_hdr.lh_nfree = ZAP_LEAF_NUMCHUNKS(l);
178 	if (sort)
179 		zap_leaf_phys(l)->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
180 }
181 
182 /*
183  * Routines which manipulate leaf chunks (l_chunk[]).
184  */
185 
186 static uint16_t
187 zap_leaf_chunk_alloc(zap_leaf_t *l)
188 {
189 	int chunk;
190 
191 	ASSERT(zap_leaf_phys(l)->l_hdr.lh_nfree > 0);
192 
193 	chunk = zap_leaf_phys(l)->l_hdr.lh_freelist;
194 	ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
195 	ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_free.lf_type, ==, ZAP_CHUNK_FREE);
196 
197 	zap_leaf_phys(l)->l_hdr.lh_freelist =
198 	    ZAP_LEAF_CHUNK(l, chunk).l_free.lf_next;
199 
200 	zap_leaf_phys(l)->l_hdr.lh_nfree--;
201 
202 	return (chunk);
203 }
204 
205 static void
206 zap_leaf_chunk_free(zap_leaf_t *l, uint16_t chunk)
207 {
208 	struct zap_leaf_free *zlf = &ZAP_LEAF_CHUNK(l, chunk).l_free;
209 	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_nfree, <, ZAP_LEAF_NUMCHUNKS(l));
210 	ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
211 	ASSERT(zlf->lf_type != ZAP_CHUNK_FREE);
212 
213 	zlf->lf_type = ZAP_CHUNK_FREE;
214 	zlf->lf_next = zap_leaf_phys(l)->l_hdr.lh_freelist;
215 	bzero(zlf->lf_pad, sizeof (zlf->lf_pad)); /* help it to compress */
216 	zap_leaf_phys(l)->l_hdr.lh_freelist = chunk;
217 
218 	zap_leaf_phys(l)->l_hdr.lh_nfree++;
219 }
220 
221 /*
222  * Routines which manipulate leaf arrays (zap_leaf_array type chunks).
223  */
224 
225 static uint16_t
226 zap_leaf_array_create(zap_leaf_t *l, const char *buf,
227     int integer_size, int num_integers)
228 {
229 	uint16_t chunk_head;
230 	uint16_t *chunkp = &chunk_head;
231 	int byten = 0;
232 	uint64_t value = 0;
233 	int shift = (integer_size-1)*8;
234 	int len = num_integers;
235 
236 	ASSERT3U(num_integers * integer_size, <, MAX_ARRAY_BYTES);
237 
238 	while (len > 0) {
239 		uint16_t chunk = zap_leaf_chunk_alloc(l);
240 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
241 		int i;
242 
243 		la->la_type = ZAP_CHUNK_ARRAY;
244 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) {
245 			if (byten == 0)
246 				value = ldv(integer_size, buf);
247 			la->la_array[i] = value >> shift;
248 			value <<= 8;
249 			if (++byten == integer_size) {
250 				byten = 0;
251 				buf += integer_size;
252 				if (--len == 0)
253 					break;
254 			}
255 		}
256 
257 		*chunkp = chunk;
258 		chunkp = &la->la_next;
259 	}
260 	*chunkp = CHAIN_END;
261 
262 	return (chunk_head);
263 }
264 
265 static void
266 zap_leaf_array_free(zap_leaf_t *l, uint16_t *chunkp)
267 {
268 	uint16_t chunk = *chunkp;
269 
270 	*chunkp = CHAIN_END;
271 
272 	while (chunk != CHAIN_END) {
273 		int nextchunk = ZAP_LEAF_CHUNK(l, chunk).l_array.la_next;
274 		ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_array.la_type, ==,
275 		    ZAP_CHUNK_ARRAY);
276 		zap_leaf_chunk_free(l, chunk);
277 		chunk = nextchunk;
278 	}
279 }
280 
281 /* array_len and buf_len are in integers, not bytes */
282 static void
283 zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk,
284     int array_int_len, int array_len, int buf_int_len, uint64_t buf_len,
285     void *buf)
286 {
287 	int len = MIN(array_len, buf_len);
288 	int byten = 0;
289 	uint64_t value = 0;
290 	char *p = buf;
291 
292 	ASSERT3U(array_int_len, <=, buf_int_len);
293 
294 	/* Fast path for one 8-byte integer */
295 	if (array_int_len == 8 && buf_int_len == 8 && len == 1) {
296 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
297 		uint8_t *ip = la->la_array;
298 		uint64_t *buf64 = buf;
299 
300 		*buf64 = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 |
301 		    (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 |
302 		    (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 |
303 		    (uint64_t)ip[6] << 8 | (uint64_t)ip[7];
304 		return;
305 	}
306 
307 	/* Fast path for an array of 1-byte integers (eg. the entry name) */
308 	if (array_int_len == 1 && buf_int_len == 1 &&
309 	    buf_len > array_len + ZAP_LEAF_ARRAY_BYTES) {
310 		while (chunk != CHAIN_END) {
311 			struct zap_leaf_array *la =
312 			    &ZAP_LEAF_CHUNK(l, chunk).l_array;
313 			bcopy(la->la_array, p, ZAP_LEAF_ARRAY_BYTES);
314 			p += ZAP_LEAF_ARRAY_BYTES;
315 			chunk = la->la_next;
316 		}
317 		return;
318 	}
319 
320 	while (len > 0) {
321 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
322 		int i;
323 
324 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
325 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
326 			value = (value << 8) | la->la_array[i];
327 			byten++;
328 			if (byten == array_int_len) {
329 				stv(buf_int_len, p, value);
330 				byten = 0;
331 				len--;
332 				if (len == 0)
333 					return;
334 				p += buf_int_len;
335 			}
336 		}
337 		chunk = la->la_next;
338 	}
339 }
340 
341 static boolean_t
342 zap_leaf_array_match(zap_leaf_t *l, zap_name_t *zn,
343     int chunk, int array_numints)
344 {
345 	int bseen = 0;
346 
347 	if (zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY) {
348 		uint64_t *thiskey;
349 		boolean_t match;
350 
351 		ASSERT(zn->zn_key_intlen == sizeof (*thiskey));
352 		thiskey = kmem_alloc(array_numints * sizeof (*thiskey),
353 		    KM_SLEEP);
354 
355 		zap_leaf_array_read(l, chunk, sizeof (*thiskey), array_numints,
356 		    sizeof (*thiskey), array_numints, thiskey);
357 		match = bcmp(thiskey, zn->zn_key_orig,
358 		    array_numints * sizeof (*thiskey)) == 0;
359 		kmem_free(thiskey, array_numints * sizeof (*thiskey));
360 		return (match);
361 	}
362 
363 	ASSERT(zn->zn_key_intlen == 1);
364 	if (zn->zn_matchtype == MT_FIRST) {
365 		char *thisname = kmem_alloc(array_numints, KM_SLEEP);
366 		boolean_t match;
367 
368 		zap_leaf_array_read(l, chunk, sizeof (char), array_numints,
369 		    sizeof (char), array_numints, thisname);
370 		match = zap_match(zn, thisname);
371 		kmem_free(thisname, array_numints);
372 		return (match);
373 	}
374 
375 	/*
376 	 * Fast path for exact matching.
377 	 * First check that the lengths match, so that we don't read
378 	 * past the end of the zn_key_orig array.
379 	 */
380 	if (array_numints != zn->zn_key_orig_numints)
381 		return (B_FALSE);
382 	while (bseen < array_numints) {
383 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
384 		int toread = MIN(array_numints - bseen, ZAP_LEAF_ARRAY_BYTES);
385 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
386 		if (bcmp(la->la_array, (char *)zn->zn_key_orig + bseen, toread))
387 			break;
388 		chunk = la->la_next;
389 		bseen += toread;
390 	}
391 	return (bseen == array_numints);
392 }
393 
394 /*
395  * Routines which manipulate leaf entries.
396  */
397 
398 int
399 zap_leaf_lookup(zap_leaf_t *l, zap_name_t *zn, zap_entry_handle_t *zeh)
400 {
401 	uint16_t *chunkp;
402 	struct zap_leaf_entry *le;
403 
404 	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
405 
406 again:
407 	for (chunkp = LEAF_HASH_ENTPTR(l, zn->zn_hash);
408 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
409 		uint16_t chunk = *chunkp;
410 		le = ZAP_LEAF_ENTRY(l, chunk);
411 
412 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
413 		ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
414 
415 		if (le->le_hash != zn->zn_hash)
416 			continue;
417 
418 		/*
419 		 * NB: the entry chain is always sorted by cd on
420 		 * normalized zap objects, so this will find the
421 		 * lowest-cd match for MT_FIRST.
422 		 */
423 		ASSERT(zn->zn_matchtype == MT_EXACT ||
424 		    (zap_leaf_phys(l)->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED));
425 		if (zap_leaf_array_match(l, zn, le->le_name_chunk,
426 		    le->le_name_numints)) {
427 			zeh->zeh_num_integers = le->le_value_numints;
428 			zeh->zeh_integer_size = le->le_value_intlen;
429 			zeh->zeh_cd = le->le_cd;
430 			zeh->zeh_hash = le->le_hash;
431 			zeh->zeh_chunkp = chunkp;
432 			zeh->zeh_leaf = l;
433 			return (0);
434 		}
435 	}
436 
437 	/*
438 	 * NB: we could of course do this in one pass, but that would be
439 	 * a pain.  We'll see if MT_BEST is even used much.
440 	 */
441 	if (zn->zn_matchtype == MT_BEST) {
442 		zn->zn_matchtype = MT_FIRST;
443 		goto again;
444 	}
445 
446 	return (SET_ERROR(ENOENT));
447 }
448 
449 /* Return (h1,cd1 >= h2,cd2) */
450 #define	HCD_GTEQ(h1, cd1, h2, cd2) \
451 	((h1 > h2) ? TRUE : ((h1 == h2 && cd1 >= cd2) ? TRUE : FALSE))
452 
453 int
454 zap_leaf_lookup_closest(zap_leaf_t *l,
455     uint64_t h, uint32_t cd, zap_entry_handle_t *zeh)
456 {
457 	uint16_t chunk;
458 	uint64_t besth = -1ULL;
459 	uint32_t bestcd = -1U;
460 	uint16_t bestlh = ZAP_LEAF_HASH_NUMENTRIES(l)-1;
461 	uint16_t lh;
462 	struct zap_leaf_entry *le;
463 
464 	ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
465 
466 	for (lh = LEAF_HASH(l, h); lh <= bestlh; lh++) {
467 		for (chunk = zap_leaf_phys(l)->l_hash[lh];
468 		    chunk != CHAIN_END; chunk = le->le_next) {
469 			le = ZAP_LEAF_ENTRY(l, chunk);
470 
471 			ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
472 			ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
473 
474 			if (HCD_GTEQ(le->le_hash, le->le_cd, h, cd) &&
475 			    HCD_GTEQ(besth, bestcd, le->le_hash, le->le_cd)) {
476 				ASSERT3U(bestlh, >=, lh);
477 				bestlh = lh;
478 				besth = le->le_hash;
479 				bestcd = le->le_cd;
480 
481 				zeh->zeh_num_integers = le->le_value_numints;
482 				zeh->zeh_integer_size = le->le_value_intlen;
483 				zeh->zeh_cd = le->le_cd;
484 				zeh->zeh_hash = le->le_hash;
485 				zeh->zeh_fakechunk = chunk;
486 				zeh->zeh_chunkp = &zeh->zeh_fakechunk;
487 				zeh->zeh_leaf = l;
488 			}
489 		}
490 	}
491 
492 	return (bestcd == -1U ? ENOENT : 0);
493 }
494 
495 int
496 zap_entry_read(const zap_entry_handle_t *zeh,
497     uint8_t integer_size, uint64_t num_integers, void *buf)
498 {
499 	struct zap_leaf_entry *le =
500 	    ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
501 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
502 
503 	if (le->le_value_intlen > integer_size)
504 		return (SET_ERROR(EINVAL));
505 
506 	zap_leaf_array_read(zeh->zeh_leaf, le->le_value_chunk,
507 	    le->le_value_intlen, le->le_value_numints,
508 	    integer_size, num_integers, buf);
509 
510 	if (zeh->zeh_num_integers > num_integers)
511 		return (SET_ERROR(EOVERFLOW));
512 	return (0);
513 
514 }
515 
516 int
517 zap_entry_read_name(zap_t *zap, const zap_entry_handle_t *zeh, uint16_t buflen,
518     char *buf)
519 {
520 	struct zap_leaf_entry *le =
521 	    ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
522 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
523 
524 	if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
525 		zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 8,
526 		    le->le_name_numints, 8, buflen / 8, buf);
527 	} else {
528 		zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 1,
529 		    le->le_name_numints, 1, buflen, buf);
530 	}
531 	if (le->le_name_numints > buflen)
532 		return (SET_ERROR(EOVERFLOW));
533 	return (0);
534 }
535 
536 int
537 zap_entry_update(zap_entry_handle_t *zeh,
538 	uint8_t integer_size, uint64_t num_integers, const void *buf)
539 {
540 	int delta_chunks;
541 	zap_leaf_t *l = zeh->zeh_leaf;
542 	struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, *zeh->zeh_chunkp);
543 
544 	delta_chunks = ZAP_LEAF_ARRAY_NCHUNKS(num_integers * integer_size) -
545 	    ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_numints * le->le_value_intlen);
546 
547 	if ((int)zap_leaf_phys(l)->l_hdr.lh_nfree < delta_chunks)
548 		return (SET_ERROR(EAGAIN));
549 
550 	zap_leaf_array_free(l, &le->le_value_chunk);
551 	le->le_value_chunk =
552 	    zap_leaf_array_create(l, buf, integer_size, num_integers);
553 	le->le_value_numints = num_integers;
554 	le->le_value_intlen = integer_size;
555 	return (0);
556 }
557 
558 void
559 zap_entry_remove(zap_entry_handle_t *zeh)
560 {
561 	uint16_t entry_chunk;
562 	struct zap_leaf_entry *le;
563 	zap_leaf_t *l = zeh->zeh_leaf;
564 
565 	ASSERT3P(zeh->zeh_chunkp, !=, &zeh->zeh_fakechunk);
566 
567 	entry_chunk = *zeh->zeh_chunkp;
568 	le = ZAP_LEAF_ENTRY(l, entry_chunk);
569 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
570 
571 	zap_leaf_array_free(l, &le->le_name_chunk);
572 	zap_leaf_array_free(l, &le->le_value_chunk);
573 
574 	*zeh->zeh_chunkp = le->le_next;
575 	zap_leaf_chunk_free(l, entry_chunk);
576 
577 	zap_leaf_phys(l)->l_hdr.lh_nentries--;
578 }
579 
580 int
581 zap_entry_create(zap_leaf_t *l, zap_name_t *zn, uint32_t cd,
582     uint8_t integer_size, uint64_t num_integers, const void *buf,
583     zap_entry_handle_t *zeh)
584 {
585 	uint16_t chunk;
586 	uint16_t *chunkp;
587 	struct zap_leaf_entry *le;
588 	uint64_t valuelen;
589 	int numchunks;
590 	uint64_t h = zn->zn_hash;
591 
592 	valuelen = integer_size * num_integers;
593 
594 	numchunks = 1 + ZAP_LEAF_ARRAY_NCHUNKS(zn->zn_key_orig_numints *
595 	    zn->zn_key_intlen) + ZAP_LEAF_ARRAY_NCHUNKS(valuelen);
596 	if (numchunks > ZAP_LEAF_NUMCHUNKS(l))
597 		return (E2BIG);
598 
599 	if (cd == ZAP_NEED_CD) {
600 		/* find the lowest unused cd */
601 		if (zap_leaf_phys(l)->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED) {
602 			cd = 0;
603 
604 			for (chunk = *LEAF_HASH_ENTPTR(l, h);
605 			    chunk != CHAIN_END; chunk = le->le_next) {
606 				le = ZAP_LEAF_ENTRY(l, chunk);
607 				if (le->le_cd > cd)
608 					break;
609 				if (le->le_hash == h) {
610 					ASSERT3U(cd, ==, le->le_cd);
611 					cd++;
612 				}
613 			}
614 		} else {
615 			/* old unsorted format; do it the O(n^2) way */
616 			for (cd = 0; ; cd++) {
617 				for (chunk = *LEAF_HASH_ENTPTR(l, h);
618 				    chunk != CHAIN_END; chunk = le->le_next) {
619 					le = ZAP_LEAF_ENTRY(l, chunk);
620 					if (le->le_hash == h &&
621 					    le->le_cd == cd) {
622 						break;
623 					}
624 				}
625 				/* If this cd is not in use, we are good. */
626 				if (chunk == CHAIN_END)
627 					break;
628 			}
629 		}
630 		/*
631 		 * We would run out of space in a block before we could
632 		 * store enough entries to run out of CD values.
633 		 */
634 		ASSERT3U(cd, <, zap_maxcd(zn->zn_zap));
635 	}
636 
637 	if (zap_leaf_phys(l)->l_hdr.lh_nfree < numchunks)
638 		return (SET_ERROR(EAGAIN));
639 
640 	/* make the entry */
641 	chunk = zap_leaf_chunk_alloc(l);
642 	le = ZAP_LEAF_ENTRY(l, chunk);
643 	le->le_type = ZAP_CHUNK_ENTRY;
644 	le->le_name_chunk = zap_leaf_array_create(l, zn->zn_key_orig,
645 	    zn->zn_key_intlen, zn->zn_key_orig_numints);
646 	le->le_name_numints = zn->zn_key_orig_numints;
647 	le->le_value_chunk =
648 	    zap_leaf_array_create(l, buf, integer_size, num_integers);
649 	le->le_value_numints = num_integers;
650 	le->le_value_intlen = integer_size;
651 	le->le_hash = h;
652 	le->le_cd = cd;
653 
654 	/* link it into the hash chain */
655 	/* XXX if we did the search above, we could just use that */
656 	chunkp = zap_leaf_rehash_entry(l, chunk);
657 
658 	zap_leaf_phys(l)->l_hdr.lh_nentries++;
659 
660 	zeh->zeh_leaf = l;
661 	zeh->zeh_num_integers = num_integers;
662 	zeh->zeh_integer_size = le->le_value_intlen;
663 	zeh->zeh_cd = le->le_cd;
664 	zeh->zeh_hash = le->le_hash;
665 	zeh->zeh_chunkp = chunkp;
666 
667 	return (0);
668 }
669 
670 /*
671  * Determine if there is another entry with the same normalized form.
672  * For performance purposes, either zn or name must be provided (the
673  * other can be NULL).  Note, there usually won't be any hash
674  * conflicts, in which case we don't need the concatenated/normalized
675  * form of the name.  But all callers have one of these on hand anyway,
676  * so might as well take advantage.  A cleaner but slower interface
677  * would accept neither argument, and compute the normalized name as
678  * needed (using zap_name_alloc(zap_entry_read_name(zeh))).
679  */
680 boolean_t
681 zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn,
682     const char *name, zap_t *zap)
683 {
684 	uint64_t chunk;
685 	struct zap_leaf_entry *le;
686 	boolean_t allocdzn = B_FALSE;
687 
688 	if (zap->zap_normflags == 0)
689 		return (B_FALSE);
690 
691 	for (chunk = *LEAF_HASH_ENTPTR(zeh->zeh_leaf, zeh->zeh_hash);
692 	    chunk != CHAIN_END; chunk = le->le_next) {
693 		le = ZAP_LEAF_ENTRY(zeh->zeh_leaf, chunk);
694 		if (le->le_hash != zeh->zeh_hash)
695 			continue;
696 		if (le->le_cd == zeh->zeh_cd)
697 			continue;
698 
699 		if (zn == NULL) {
700 			zn = zap_name_alloc(zap, name, MT_FIRST);
701 			allocdzn = B_TRUE;
702 		}
703 		if (zap_leaf_array_match(zeh->zeh_leaf, zn,
704 		    le->le_name_chunk, le->le_name_numints)) {
705 			if (allocdzn)
706 				zap_name_free(zn);
707 			return (B_TRUE);
708 		}
709 	}
710 	if (allocdzn)
711 		zap_name_free(zn);
712 	return (B_FALSE);
713 }
714 
715 /*
716  * Routines for transferring entries between leafs.
717  */
718 
719 static uint16_t *
720 zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry)
721 {
722 	struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry);
723 	struct zap_leaf_entry *le2;
724 	uint16_t *chunkp;
725 
726 	/*
727 	 * keep the entry chain sorted by cd
728 	 * NB: this will not cause problems for unsorted leafs, though
729 	 * it is unnecessary there.
730 	 */
731 	for (chunkp = LEAF_HASH_ENTPTR(l, le->le_hash);
732 	    *chunkp != CHAIN_END; chunkp = &le2->le_next) {
733 		le2 = ZAP_LEAF_ENTRY(l, *chunkp);
734 		if (le2->le_cd > le->le_cd)
735 			break;
736 	}
737 
738 	le->le_next = *chunkp;
739 	*chunkp = entry;
740 	return (chunkp);
741 }
742 
743 static uint16_t
744 zap_leaf_transfer_array(zap_leaf_t *l, uint16_t chunk, zap_leaf_t *nl)
745 {
746 	uint16_t new_chunk;
747 	uint16_t *nchunkp = &new_chunk;
748 
749 	while (chunk != CHAIN_END) {
750 		uint16_t nchunk = zap_leaf_chunk_alloc(nl);
751 		struct zap_leaf_array *nla =
752 		    &ZAP_LEAF_CHUNK(nl, nchunk).l_array;
753 		struct zap_leaf_array *la =
754 		    &ZAP_LEAF_CHUNK(l, chunk).l_array;
755 		int nextchunk = la->la_next;
756 
757 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
758 		ASSERT3U(nchunk, <, ZAP_LEAF_NUMCHUNKS(l));
759 
760 		*nla = *la; /* structure assignment */
761 
762 		zap_leaf_chunk_free(l, chunk);
763 		chunk = nextchunk;
764 		*nchunkp = nchunk;
765 		nchunkp = &nla->la_next;
766 	}
767 	*nchunkp = CHAIN_END;
768 	return (new_chunk);
769 }
770 
771 static void
772 zap_leaf_transfer_entry(zap_leaf_t *l, int entry, zap_leaf_t *nl)
773 {
774 	struct zap_leaf_entry *le, *nle;
775 	uint16_t chunk;
776 
777 	le = ZAP_LEAF_ENTRY(l, entry);
778 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
779 
780 	chunk = zap_leaf_chunk_alloc(nl);
781 	nle = ZAP_LEAF_ENTRY(nl, chunk);
782 	*nle = *le; /* structure assignment */
783 
784 	(void) zap_leaf_rehash_entry(nl, chunk);
785 
786 	nle->le_name_chunk = zap_leaf_transfer_array(l, le->le_name_chunk, nl);
787 	nle->le_value_chunk =
788 	    zap_leaf_transfer_array(l, le->le_value_chunk, nl);
789 
790 	zap_leaf_chunk_free(l, entry);
791 
792 	zap_leaf_phys(l)->l_hdr.lh_nentries--;
793 	zap_leaf_phys(nl)->l_hdr.lh_nentries++;
794 }
795 
796 /*
797  * Transfer the entries whose hash prefix ends in 1 to the new leaf.
798  */
799 void
800 zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort)
801 {
802 	int i;
803 	int bit = 64 - 1 - zap_leaf_phys(l)->l_hdr.lh_prefix_len;
804 
805 	/* set new prefix and prefix_len */
806 	zap_leaf_phys(l)->l_hdr.lh_prefix <<= 1;
807 	zap_leaf_phys(l)->l_hdr.lh_prefix_len++;
808 	zap_leaf_phys(nl)->l_hdr.lh_prefix =
809 	    zap_leaf_phys(l)->l_hdr.lh_prefix | 1;
810 	zap_leaf_phys(nl)->l_hdr.lh_prefix_len =
811 	    zap_leaf_phys(l)->l_hdr.lh_prefix_len;
812 
813 	/* break existing hash chains */
814 	zap_memset(zap_leaf_phys(l)->l_hash, CHAIN_END,
815 	    2*ZAP_LEAF_HASH_NUMENTRIES(l));
816 
817 	if (sort)
818 		zap_leaf_phys(l)->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
819 
820 	/*
821 	 * Transfer entries whose hash bit 'bit' is set to nl; rehash
822 	 * the remaining entries
823 	 *
824 	 * NB: We could find entries via the hashtable instead. That
825 	 * would be O(hashents+numents) rather than O(numblks+numents),
826 	 * but this accesses memory more sequentially, and when we're
827 	 * called, the block is usually pretty full.
828 	 */
829 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
830 		struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, i);
831 		if (le->le_type != ZAP_CHUNK_ENTRY)
832 			continue;
833 
834 		if (le->le_hash & (1ULL << bit))
835 			zap_leaf_transfer_entry(l, i, nl);
836 		else
837 			(void) zap_leaf_rehash_entry(l, i);
838 	}
839 }
840 
841 void
842 zap_leaf_stats(zap_t *zap, zap_leaf_t *l, zap_stats_t *zs)
843 {
844 	int i, n;
845 
846 	n = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
847 	    zap_leaf_phys(l)->l_hdr.lh_prefix_len;
848 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
849 	zs->zs_leafs_with_2n_pointers[n]++;
850 
851 
852 	n = zap_leaf_phys(l)->l_hdr.lh_nentries/5;
853 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
854 	zs->zs_blocks_with_n5_entries[n]++;
855 
856 	n = ((1<<FZAP_BLOCK_SHIFT(zap)) -
857 	    zap_leaf_phys(l)->l_hdr.lh_nfree * (ZAP_LEAF_ARRAY_BYTES+1))*10 /
858 	    (1<<FZAP_BLOCK_SHIFT(zap));
859 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
860 	zs->zs_blocks_n_tenths_full[n]++;
861 
862 	for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(l); i++) {
863 		int nentries = 0;
864 		int chunk = zap_leaf_phys(l)->l_hash[i];
865 
866 		while (chunk != CHAIN_END) {
867 			struct zap_leaf_entry *le =
868 			    ZAP_LEAF_ENTRY(l, chunk);
869 
870 			n = 1 + ZAP_LEAF_ARRAY_NCHUNKS(le->le_name_numints) +
871 			    ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_numints *
872 			    le->le_value_intlen);
873 			n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
874 			zs->zs_entries_using_n_chunks[n]++;
875 
876 			chunk = le->le_next;
877 			nentries++;
878 		}
879 
880 		n = nentries;
881 		n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
882 		zs->zs_buckets_with_n_entries[n]++;
883 	}
884 }
885