xref: /titanic_50/usr/src/uts/common/fs/zfs/zap_leaf.c (revision 2df1fe9ca32bb227b9158c67f5c00b54c20b10fd)
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 2006 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 #define	CHAIN_END 0xffff /* end of the chunk chain */
42 
43 /* half the (current) minimum block size */
44 #define	MAX_ARRAY_BYTES (8<<10)
45 
46 #define	LEAF_HASH(l, h) \
47 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
48 	((h) >> (64 - ZAP_LEAF_HASH_SHIFT(l)-(l)->l_phys->l_hdr.lh_prefix_len)))
49 
50 #define	LEAF_HASH_ENTPTR(l, h) (&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
51 
52 
53 static void
54 zap_memset(void *a, int c, size_t n)
55 {
56 	char *cp = a;
57 	char *cpend = cp + n;
58 
59 	while (cp < cpend)
60 		*cp++ = c;
61 }
62 
63 static void
64 stv(int len, void *addr, uint64_t value)
65 {
66 	switch (len) {
67 	case 1:
68 		*(uint8_t *)addr = value;
69 		return;
70 	case 2:
71 		*(uint16_t *)addr = value;
72 		return;
73 	case 4:
74 		*(uint32_t *)addr = value;
75 		return;
76 	case 8:
77 		*(uint64_t *)addr = value;
78 		return;
79 	}
80 	ASSERT(!"bad int len");
81 }
82 
83 static uint64_t
84 ldv(int len, const void *addr)
85 {
86 	switch (len) {
87 	case 1:
88 		return (*(uint8_t *)addr);
89 	case 2:
90 		return (*(uint16_t *)addr);
91 	case 4:
92 		return (*(uint32_t *)addr);
93 	case 8:
94 		return (*(uint64_t *)addr);
95 	}
96 	ASSERT(!"bad int len");
97 	return (0xFEEDFACEDEADBEEFULL);
98 }
99 
100 void
101 zap_leaf_byteswap(zap_leaf_phys_t *buf, int size)
102 {
103 	int i;
104 	zap_leaf_t l;
105 	l.l_bs = highbit(size)-1;
106 	l.l_phys = buf;
107 
108 	buf->l_hdr.lh_block_type = 	BSWAP_64(buf->l_hdr.lh_block_type);
109 	buf->l_hdr.lh_prefix = 		BSWAP_64(buf->l_hdr.lh_prefix);
110 	buf->l_hdr.lh_magic = 		BSWAP_32(buf->l_hdr.lh_magic);
111 	buf->l_hdr.lh_nfree = 		BSWAP_16(buf->l_hdr.lh_nfree);
112 	buf->l_hdr.lh_nentries = 	BSWAP_16(buf->l_hdr.lh_nentries);
113 	buf->l_hdr.lh_prefix_len = 	BSWAP_16(buf->l_hdr.lh_prefix_len);
114 	buf->l_hdr.lh_freelist = 	BSWAP_16(buf->l_hdr.lh_freelist);
115 
116 	for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++)
117 		buf->l_hash[i] = BSWAP_16(buf->l_hash[i]);
118 
119 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
120 		zap_leaf_chunk_t *lc = &ZAP_LEAF_CHUNK(&l, i);
121 		struct zap_leaf_entry *le;
122 
123 		switch (lc->l_free.lf_type) {
124 		case ZAP_CHUNK_ENTRY:
125 			le = &lc->l_entry;
126 
127 			le->le_type =		BSWAP_8(le->le_type);
128 			le->le_int_size =	BSWAP_8(le->le_int_size);
129 			le->le_next =		BSWAP_16(le->le_next);
130 			le->le_name_chunk =	BSWAP_16(le->le_name_chunk);
131 			le->le_name_length =	BSWAP_16(le->le_name_length);
132 			le->le_value_chunk =	BSWAP_16(le->le_value_chunk);
133 			le->le_value_length =	BSWAP_16(le->le_value_length);
134 			le->le_cd =		BSWAP_32(le->le_cd);
135 			le->le_hash =		BSWAP_64(le->le_hash);
136 			break;
137 		case ZAP_CHUNK_FREE:
138 			lc->l_free.lf_type =	BSWAP_8(lc->l_free.lf_type);
139 			lc->l_free.lf_next =	BSWAP_16(lc->l_free.lf_next);
140 			break;
141 		case ZAP_CHUNK_ARRAY:
142 			lc->l_array.la_type =	BSWAP_8(lc->l_array.la_type);
143 			lc->l_array.la_next =	BSWAP_16(lc->l_array.la_next);
144 			/* la_array doesn't need swapping */
145 			break;
146 		default:
147 			ASSERT(!"bad leaf type");
148 		}
149 	}
150 }
151 
152 void
153 zap_leaf_init(zap_leaf_t *l)
154 {
155 	int i;
156 
157 	l->l_bs = highbit(l->l_dbuf->db_size)-1;
158 	zap_memset(&l->l_phys->l_hdr, 0, sizeof (struct zap_leaf_header));
159 	zap_memset(l->l_phys->l_hash, CHAIN_END, 2*ZAP_LEAF_HASH_NUMENTRIES(l));
160 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
161 		ZAP_LEAF_CHUNK(l, i).l_free.lf_type = ZAP_CHUNK_FREE;
162 		ZAP_LEAF_CHUNK(l, i).l_free.lf_next = i+1;
163 	}
164 	ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)-1).l_free.lf_next = CHAIN_END;
165 	l->l_phys->l_hdr.lh_block_type = ZBT_LEAF;
166 	l->l_phys->l_hdr.lh_magic = ZAP_LEAF_MAGIC;
167 	l->l_phys->l_hdr.lh_nfree = ZAP_LEAF_NUMCHUNKS(l);
168 }
169 
170 /*
171  * Routines which manipulate leaf chunks (l_chunk[]).
172  */
173 
174 static uint16_t
175 zap_leaf_chunk_alloc(zap_leaf_t *l)
176 {
177 	int chunk;
178 
179 	ASSERT(l->l_phys->l_hdr.lh_nfree > 0);
180 
181 	chunk = l->l_phys->l_hdr.lh_freelist;
182 	ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
183 	ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_free.lf_type, ==, ZAP_CHUNK_FREE);
184 
185 	l->l_phys->l_hdr.lh_freelist = ZAP_LEAF_CHUNK(l, chunk).l_free.lf_next;
186 
187 	l->l_phys->l_hdr.lh_nfree--;
188 
189 	return (chunk);
190 }
191 
192 static void
193 zap_leaf_chunk_free(zap_leaf_t *l, uint16_t chunk)
194 {
195 	struct zap_leaf_free *zlf = &ZAP_LEAF_CHUNK(l, chunk).l_free;
196 	ASSERT3U(l->l_phys->l_hdr.lh_nfree, <, ZAP_LEAF_NUMCHUNKS(l));
197 	ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
198 	ASSERT(zlf->lf_type != ZAP_CHUNK_FREE);
199 
200 	zlf->lf_type = ZAP_CHUNK_FREE;
201 	zlf->lf_next = l->l_phys->l_hdr.lh_freelist;
202 	bzero(zlf->lf_pad, sizeof (zlf->lf_pad)); /* help it to compress */
203 	l->l_phys->l_hdr.lh_freelist = chunk;
204 
205 	l->l_phys->l_hdr.lh_nfree++;
206 }
207 
208 /*
209  * Routines which manipulate leaf arrays (zap_leaf_array type chunks).
210  */
211 
212 static uint16_t
213 zap_leaf_array_create(zap_leaf_t *l, const char *buf,
214 	int integer_size, int num_integers)
215 {
216 	uint16_t chunk_head;
217 	uint16_t *chunkp = &chunk_head;
218 	int byten = 0;
219 	uint64_t value;
220 	int shift = (integer_size-1)*8;
221 	int len = num_integers;
222 
223 	ASSERT3U(num_integers * integer_size, <, MAX_ARRAY_BYTES);
224 
225 	while (len > 0) {
226 		uint16_t chunk = zap_leaf_chunk_alloc(l);
227 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
228 		int i;
229 
230 		la->la_type = ZAP_CHUNK_ARRAY;
231 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) {
232 			if (byten == 0)
233 				value = ldv(integer_size, buf);
234 			la->la_array[i] = value >> shift;
235 			value <<= 8;
236 			if (++byten == integer_size) {
237 				byten = 0;
238 				buf += integer_size;
239 				if (--len == 0)
240 					break;
241 			}
242 		}
243 
244 		*chunkp = chunk;
245 		chunkp = &la->la_next;
246 	}
247 	*chunkp = CHAIN_END;
248 
249 	return (chunk_head);
250 }
251 
252 static void
253 zap_leaf_array_free(zap_leaf_t *l, uint16_t *chunkp)
254 {
255 	uint16_t chunk = *chunkp;
256 
257 	*chunkp = CHAIN_END;
258 
259 	while (chunk != CHAIN_END) {
260 		int nextchunk = ZAP_LEAF_CHUNK(l, chunk).l_array.la_next;
261 		ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_array.la_type, ==,
262 		    ZAP_CHUNK_ARRAY);
263 		zap_leaf_chunk_free(l, chunk);
264 		chunk = nextchunk;
265 	}
266 }
267 
268 /* array_len and buf_len are in integers, not bytes */
269 static void
270 zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk,
271     int array_int_len, int array_len, int buf_int_len, uint64_t buf_len,
272     char *buf)
273 {
274 	int len = MIN(array_len, buf_len);
275 	int byten = 0;
276 	uint64_t value = 0;
277 
278 	ASSERT3U(array_int_len, <=, buf_int_len);
279 
280 	/* Fast path for one 8-byte integer */
281 	if (array_int_len == 8 && buf_int_len == 8 && len == 1) {
282 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
283 		uint8_t *ip = la->la_array;
284 		uint64_t *buf64 = (uint64_t *)buf;
285 
286 		*buf64 = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 |
287 		    (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 |
288 		    (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 |
289 		    (uint64_t)ip[6] << 8 | (uint64_t)ip[7];
290 		return;
291 	}
292 
293 	/* Fast path for an array of 1-byte integers (eg. the entry name) */
294 	if (array_int_len == 1 && buf_int_len == 1 &&
295 	    buf_len > array_len + ZAP_LEAF_ARRAY_BYTES) {
296 		while (chunk != CHAIN_END) {
297 			struct zap_leaf_array *la =
298 			    &ZAP_LEAF_CHUNK(l, chunk).l_array;
299 			bcopy(la->la_array, buf, ZAP_LEAF_ARRAY_BYTES);
300 			buf += ZAP_LEAF_ARRAY_BYTES;
301 			chunk = la->la_next;
302 		}
303 		return;
304 	}
305 
306 	while (len > 0) {
307 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
308 		int i;
309 
310 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
311 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
312 			value = (value << 8) | la->la_array[i];
313 			byten++;
314 			if (byten == array_int_len) {
315 				stv(buf_int_len, buf, value);
316 				byten = 0;
317 				len--;
318 				if (len == 0)
319 					return;
320 				buf += buf_int_len;
321 			}
322 		}
323 		chunk = la->la_next;
324 	}
325 }
326 
327 /*
328  * Only to be used on 8-bit arrays.
329  * array_len is actual len in bytes (not encoded le_value_length).
330  * buf is null-terminated.
331  */
332 static int
333 zap_leaf_array_equal(zap_leaf_t *l, int chunk,
334     int array_len, const char *buf)
335 {
336 	int bseen = 0;
337 
338 	while (bseen < array_len) {
339 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
340 		int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);
341 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
342 		if (bcmp(la->la_array, buf + bseen, toread))
343 			break;
344 		chunk = la->la_next;
345 		bseen += toread;
346 	}
347 	return (bseen == array_len);
348 }
349 
350 /*
351  * Routines which manipulate leaf entries.
352  */
353 
354 int
355 zap_leaf_lookup(zap_leaf_t *l,
356     const char *name, uint64_t h, zap_entry_handle_t *zeh)
357 {
358 	uint16_t *chunkp;
359 	struct zap_leaf_entry *le;
360 
361 	ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
362 
363 	for (chunkp = LEAF_HASH_ENTPTR(l, h);
364 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
365 		uint16_t chunk = *chunkp;
366 		le = ZAP_LEAF_ENTRY(l, chunk);
367 
368 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
369 		ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
370 
371 		if (le->le_hash != h)
372 			continue;
373 
374 		if (zap_leaf_array_equal(l, le->le_name_chunk,
375 		    le->le_name_length, name)) {
376 			zeh->zeh_num_integers = le->le_value_length;
377 			zeh->zeh_integer_size = le->le_int_size;
378 			zeh->zeh_cd = le->le_cd;
379 			zeh->zeh_hash = le->le_hash;
380 			zeh->zeh_chunkp = chunkp;
381 			zeh->zeh_leaf = l;
382 			return (0);
383 		}
384 	}
385 
386 	return (ENOENT);
387 }
388 
389 /* Return (h1,cd1 >= h2,cd2) */
390 #define	HCD_GTEQ(h1, cd1, h2, cd2) \
391 	((h1 > h2) ? TRUE : ((h1 == h2 && cd1 >= cd2) ? TRUE : FALSE))
392 
393 int
394 zap_leaf_lookup_closest(zap_leaf_t *l,
395     uint64_t h, uint32_t cd, zap_entry_handle_t *zeh)
396 {
397 	uint16_t chunk;
398 	uint64_t besth = -1ULL;
399 	uint32_t bestcd = ZAP_MAXCD;
400 	uint16_t bestlh = ZAP_LEAF_HASH_NUMENTRIES(l)-1;
401 	uint16_t lh;
402 	struct zap_leaf_entry *le;
403 
404 	ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
405 
406 	for (lh = LEAF_HASH(l, h); lh <= bestlh; lh++) {
407 		for (chunk = l->l_phys->l_hash[lh];
408 		    chunk != CHAIN_END; chunk = le->le_next) {
409 			le = ZAP_LEAF_ENTRY(l, chunk);
410 
411 			ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
412 			ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
413 
414 			if (HCD_GTEQ(le->le_hash, le->le_cd, h, cd) &&
415 			    HCD_GTEQ(besth, bestcd, le->le_hash, le->le_cd)) {
416 				ASSERT3U(bestlh, >=, lh);
417 				bestlh = lh;
418 				besth = le->le_hash;
419 				bestcd = le->le_cd;
420 
421 				zeh->zeh_num_integers = le->le_value_length;
422 				zeh->zeh_integer_size = le->le_int_size;
423 				zeh->zeh_cd = le->le_cd;
424 				zeh->zeh_hash = le->le_hash;
425 				zeh->zeh_fakechunk = chunk;
426 				zeh->zeh_chunkp = &zeh->zeh_fakechunk;
427 				zeh->zeh_leaf = l;
428 			}
429 		}
430 	}
431 
432 	return (bestcd == ZAP_MAXCD ? ENOENT : 0);
433 }
434 
435 int
436 zap_entry_read(const zap_entry_handle_t *zeh,
437     uint8_t integer_size, uint64_t num_integers, void *buf)
438 {
439 	struct zap_leaf_entry *le =
440 	    ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
441 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
442 
443 	if (le->le_int_size > integer_size)
444 		return (EINVAL);
445 
446 	zap_leaf_array_read(zeh->zeh_leaf, le->le_value_chunk, le->le_int_size,
447 	    le->le_value_length, integer_size, num_integers, buf);
448 
449 	if (zeh->zeh_num_integers > num_integers)
450 		return (EOVERFLOW);
451 	return (0);
452 
453 }
454 
455 int
456 zap_entry_read_name(const zap_entry_handle_t *zeh, uint16_t buflen, char *buf)
457 {
458 	struct zap_leaf_entry *le =
459 	    ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
460 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
461 
462 	zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 1,
463 	    le->le_name_length, 1, buflen, buf);
464 	if (le->le_name_length > buflen)
465 		return (EOVERFLOW);
466 	return (0);
467 }
468 
469 int
470 zap_entry_update(zap_entry_handle_t *zeh,
471 	uint8_t integer_size, uint64_t num_integers, const void *buf)
472 {
473 	int delta_chunks;
474 	zap_leaf_t *l = zeh->zeh_leaf;
475 	struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, *zeh->zeh_chunkp);
476 
477 	delta_chunks = ZAP_LEAF_ARRAY_NCHUNKS(num_integers * integer_size) -
478 	    ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_length * le->le_int_size);
479 
480 	if ((int)l->l_phys->l_hdr.lh_nfree < delta_chunks)
481 		return (EAGAIN);
482 
483 	/*
484 	 * We should search other chained leaves (via
485 	 * zap_entry_remove,create?) otherwise returning EAGAIN will
486 	 * just send us into an infinite loop if we have to chain
487 	 * another leaf block, rather than being able to split this
488 	 * block.
489 	 */
490 
491 	zap_leaf_array_free(l, &le->le_value_chunk);
492 	le->le_value_chunk =
493 	    zap_leaf_array_create(l, buf, integer_size, num_integers);
494 	le->le_value_length = num_integers;
495 	le->le_int_size = integer_size;
496 	return (0);
497 }
498 
499 void
500 zap_entry_remove(zap_entry_handle_t *zeh)
501 {
502 	uint16_t entry_chunk;
503 	struct zap_leaf_entry *le;
504 	zap_leaf_t *l = zeh->zeh_leaf;
505 
506 	ASSERT3P(zeh->zeh_chunkp, !=, &zeh->zeh_fakechunk);
507 
508 	entry_chunk = *zeh->zeh_chunkp;
509 	le = ZAP_LEAF_ENTRY(l, entry_chunk);
510 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
511 
512 	zap_leaf_array_free(l, &le->le_name_chunk);
513 	zap_leaf_array_free(l, &le->le_value_chunk);
514 
515 	*zeh->zeh_chunkp = le->le_next;
516 	zap_leaf_chunk_free(l, entry_chunk);
517 
518 	l->l_phys->l_hdr.lh_nentries--;
519 }
520 
521 int
522 zap_entry_create(zap_leaf_t *l, const char *name, uint64_t h, uint32_t cd,
523     uint8_t integer_size, uint64_t num_integers, const void *buf,
524     zap_entry_handle_t *zeh)
525 {
526 	uint16_t chunk;
527 	uint16_t *chunkp;
528 	struct zap_leaf_entry *le;
529 	uint64_t namelen, valuelen;
530 	int numchunks;
531 
532 	valuelen = integer_size * num_integers;
533 	namelen = strlen(name) + 1;
534 	ASSERT(namelen >= 2);
535 
536 	numchunks = 1 + ZAP_LEAF_ARRAY_NCHUNKS(namelen) +
537 	    ZAP_LEAF_ARRAY_NCHUNKS(valuelen);
538 	if (numchunks > ZAP_LEAF_NUMCHUNKS(l))
539 		return (E2BIG);
540 
541 	if (cd == ZAP_MAXCD) {
542 		for (cd = 0; cd < ZAP_MAXCD; cd++) {
543 			for (chunk = *LEAF_HASH_ENTPTR(l, h);
544 			    chunk != CHAIN_END; chunk = le->le_next) {
545 				le = ZAP_LEAF_ENTRY(l, chunk);
546 				if (le->le_hash == h &&
547 				    le->le_cd == cd) {
548 					break;
549 				}
550 			}
551 			/* If this cd is not in use, we are good. */
552 			if (chunk == CHAIN_END)
553 				break;
554 		}
555 		/* If we tried all the cd's, we lose. */
556 		if (cd == ZAP_MAXCD)
557 			return (ENOSPC);
558 	}
559 
560 	if (l->l_phys->l_hdr.lh_nfree < numchunks)
561 		return (EAGAIN);
562 
563 	/* make the entry */
564 	chunk = zap_leaf_chunk_alloc(l);
565 	le = ZAP_LEAF_ENTRY(l, chunk);
566 	le->le_type = ZAP_CHUNK_ENTRY;
567 	le->le_name_chunk = zap_leaf_array_create(l, name, 1, namelen);
568 	le->le_name_length = namelen;
569 	le->le_value_chunk =
570 	    zap_leaf_array_create(l, buf, integer_size, num_integers);
571 	le->le_value_length = num_integers;
572 	le->le_int_size = integer_size;
573 	le->le_hash = h;
574 	le->le_cd = cd;
575 
576 	/* link it into the hash chain */
577 	chunkp = LEAF_HASH_ENTPTR(l, h);
578 	le->le_next = *chunkp;
579 	*chunkp = chunk;
580 
581 	l->l_phys->l_hdr.lh_nentries++;
582 
583 	zeh->zeh_leaf = l;
584 	zeh->zeh_num_integers = num_integers;
585 	zeh->zeh_integer_size = le->le_int_size;
586 	zeh->zeh_cd = le->le_cd;
587 	zeh->zeh_hash = le->le_hash;
588 	zeh->zeh_chunkp = chunkp;
589 
590 	return (0);
591 }
592 
593 /*
594  * Routines for transferring entries between leafs.
595  */
596 
597 static void
598 zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry)
599 {
600 	struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry);
601 	uint16_t *ptr = LEAF_HASH_ENTPTR(l, le->le_hash);
602 	le->le_next = *ptr;
603 	*ptr = entry;
604 }
605 
606 static uint16_t
607 zap_leaf_transfer_array(zap_leaf_t *l, uint16_t chunk, zap_leaf_t *nl)
608 {
609 	uint16_t new_chunk;
610 	uint16_t *nchunkp = &new_chunk;
611 
612 	while (chunk != CHAIN_END) {
613 		uint16_t nchunk = zap_leaf_chunk_alloc(nl);
614 		struct zap_leaf_array *nla =
615 		    &ZAP_LEAF_CHUNK(nl, nchunk).l_array;
616 		struct zap_leaf_array *la =
617 		    &ZAP_LEAF_CHUNK(l, chunk).l_array;
618 		int nextchunk = la->la_next;
619 
620 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
621 		ASSERT3U(nchunk, <, ZAP_LEAF_NUMCHUNKS(l));
622 
623 		*nla = *la; /* structure assignment */
624 
625 		zap_leaf_chunk_free(l, chunk);
626 		chunk = nextchunk;
627 		*nchunkp = nchunk;
628 		nchunkp = &nla->la_next;
629 	}
630 	*nchunkp = CHAIN_END;
631 	return (new_chunk);
632 }
633 
634 static void
635 zap_leaf_transfer_entry(zap_leaf_t *l, int entry, zap_leaf_t *nl)
636 {
637 	struct zap_leaf_entry *le, *nle;
638 	uint16_t chunk;
639 
640 	le = ZAP_LEAF_ENTRY(l, entry);
641 	ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
642 
643 	chunk = zap_leaf_chunk_alloc(nl);
644 	nle = ZAP_LEAF_ENTRY(nl, chunk);
645 	*nle = *le; /* structure assignment */
646 
647 	zap_leaf_rehash_entry(nl, chunk);
648 
649 	nle->le_name_chunk = zap_leaf_transfer_array(l, le->le_name_chunk, nl);
650 	nle->le_value_chunk =
651 	    zap_leaf_transfer_array(l, le->le_value_chunk, nl);
652 
653 	zap_leaf_chunk_free(l, entry);
654 
655 	l->l_phys->l_hdr.lh_nentries--;
656 	nl->l_phys->l_hdr.lh_nentries++;
657 }
658 
659 /*
660  * Transfer the entries whose hash prefix ends in 1 to the new leaf.
661  */
662 void
663 zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl)
664 {
665 	int i;
666 	int bit = 64 - 1 - l->l_phys->l_hdr.lh_prefix_len;
667 
668 	/* set new prefix and prefix_len */
669 	l->l_phys->l_hdr.lh_prefix <<= 1;
670 	l->l_phys->l_hdr.lh_prefix_len++;
671 	nl->l_phys->l_hdr.lh_prefix = l->l_phys->l_hdr.lh_prefix | 1;
672 	nl->l_phys->l_hdr.lh_prefix_len = l->l_phys->l_hdr.lh_prefix_len;
673 
674 	/* break existing hash chains */
675 	zap_memset(l->l_phys->l_hash, CHAIN_END, 2*ZAP_LEAF_HASH_NUMENTRIES(l));
676 
677 	/*
678 	 * Transfer entries whose hash bit 'bit' is set to nl; rehash
679 	 * the remaining entries
680 	 *
681 	 * NB: We could find entries via the hashtable instead. That
682 	 * would be O(hashents+numents) rather than O(numblks+numents),
683 	 * but this accesses memory more sequentially, and when we're
684 	 * called, the block is usually pretty full.
685 	 */
686 	for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
687 		struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, i);
688 		if (le->le_type != ZAP_CHUNK_ENTRY)
689 			continue;
690 
691 		if (le->le_hash & (1ULL << bit))
692 			zap_leaf_transfer_entry(l, i, nl);
693 		else
694 			zap_leaf_rehash_entry(l, i);
695 	}
696 }
697 
698 void
699 zap_leaf_stats(zap_t *zap, zap_leaf_t *l, zap_stats_t *zs)
700 {
701 	int i, n;
702 
703 	n = zap->zap_f.zap_phys->zap_ptrtbl.zt_shift -
704 	    l->l_phys->l_hdr.lh_prefix_len;
705 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
706 	zs->zs_leafs_with_2n_pointers[n]++;
707 
708 
709 	n = l->l_phys->l_hdr.lh_nentries/5;
710 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
711 	zs->zs_blocks_with_n5_entries[n]++;
712 
713 	n = ((1<<FZAP_BLOCK_SHIFT(zap)) -
714 	    l->l_phys->l_hdr.lh_nfree * (ZAP_LEAF_ARRAY_BYTES+1))*10 /
715 	    (1<<FZAP_BLOCK_SHIFT(zap));
716 	n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
717 	zs->zs_blocks_n_tenths_full[n]++;
718 
719 	for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(l); i++) {
720 		int nentries = 0;
721 		int chunk = l->l_phys->l_hash[i];
722 
723 		while (chunk != CHAIN_END) {
724 			struct zap_leaf_entry *le =
725 			    ZAP_LEAF_ENTRY(l, chunk);
726 
727 			n = 1 + ZAP_LEAF_ARRAY_NCHUNKS(le->le_name_length) +
728 			    ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_length *
729 				le->le_int_size);
730 			n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
731 			zs->zs_entries_using_n_chunks[n]++;
732 
733 			chunk = le->le_next;
734 			nentries++;
735 		}
736 
737 		n = nentries;
738 		n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
739 		zs->zs_buckets_with_n_entries[n]++;
740 	}
741 }
742