1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22
23 /*
24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 * Copyright (c) 2013, 2016 by Delphix. All rights reserved.
26 * Copyright 2017 Nexenta Systems, Inc.
27 */
28
29 /*
30 * The 512-byte leaf is broken into 32 16-byte chunks.
31 * chunk number n means l_chunk[n], even though the header precedes it.
32 * the names are stored null-terminated.
33 */
34
35 #include <sys/zio.h>
36 #include <sys/spa.h>
37 #include <sys/dmu.h>
38 #include <sys/zfs_context.h>
39 #include <sys/fs/zfs.h>
40 #include <sys/zap.h>
41 #include <sys/zap_impl.h>
42 #include <sys/zap_leaf.h>
43 #include <sys/arc.h>
44
45 static uint16_t *zap_leaf_rehash_entry(zap_leaf_t *l, struct zap_leaf_entry *le,
46 uint16_t entry);
47
48 #define CHAIN_END 0xffff /* end of the chunk chain */
49
50 #define LEAF_HASH(l, h) \
51 ((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
52 ((h) >> \
53 (64 - ZAP_LEAF_HASH_SHIFT(l) - zap_leaf_phys(l)->l_hdr.lh_prefix_len)))
54
55 #define LEAF_HASH_ENTPTR(l, h) (&zap_leaf_phys(l)->l_hash[LEAF_HASH(l, h)])
56
57 static void
stv(int len,void * addr,uint64_t value)58 stv(int len, void *addr, uint64_t value)
59 {
60 switch (len) {
61 case 1:
62 *(uint8_t *)addr = value;
63 return;
64 case 2:
65 *(uint16_t *)addr = value;
66 return;
67 case 4:
68 *(uint32_t *)addr = value;
69 return;
70 case 8:
71 *(uint64_t *)addr = value;
72 return;
73 default:
74 PANIC("bad int len %d", len);
75 }
76 }
77
78 static uint64_t
ldv(int len,const void * addr)79 ldv(int len, const void *addr)
80 {
81 switch (len) {
82 case 1:
83 return (*(uint8_t *)addr);
84 case 2:
85 return (*(uint16_t *)addr);
86 case 4:
87 return (*(uint32_t *)addr);
88 case 8:
89 return (*(uint64_t *)addr);
90 default:
91 PANIC("bad int len %d", len);
92 }
93 return (0xFEEDFACEDEADBEEFULL);
94 }
95
96 void
zap_leaf_byteswap(zap_leaf_phys_t * buf,size_t size)97 zap_leaf_byteswap(zap_leaf_phys_t *buf, size_t size)
98 {
99 zap_leaf_t l;
100 dmu_buf_t l_dbuf;
101
102 l_dbuf.db_data = buf;
103 l.l_bs = highbit64(size) - 1;
104 l.l_dbuf = &l_dbuf;
105
106 buf->l_hdr.lh_block_type = BSWAP_64(buf->l_hdr.lh_block_type);
107 buf->l_hdr.lh_prefix = BSWAP_64(buf->l_hdr.lh_prefix);
108 buf->l_hdr.lh_magic = BSWAP_32(buf->l_hdr.lh_magic);
109 buf->l_hdr.lh_nfree = BSWAP_16(buf->l_hdr.lh_nfree);
110 buf->l_hdr.lh_nentries = BSWAP_16(buf->l_hdr.lh_nentries);
111 buf->l_hdr.lh_prefix_len = BSWAP_16(buf->l_hdr.lh_prefix_len);
112 buf->l_hdr.lh_freelist = BSWAP_16(buf->l_hdr.lh_freelist);
113
114 for (uint_t i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++)
115 buf->l_hash[i] = BSWAP_16(buf->l_hash[i]);
116
117 for (uint_t i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) {
118 zap_leaf_chunk_t *lc = &ZAP_LEAF_CHUNK(&l, i);
119 struct zap_leaf_entry *le;
120
121 switch (lc->l_free.lf_type) {
122 case ZAP_CHUNK_ENTRY:
123 le = &lc->l_entry;
124
125 le->le_type = BSWAP_8(le->le_type);
126 le->le_value_intlen = BSWAP_8(le->le_value_intlen);
127 le->le_next = BSWAP_16(le->le_next);
128 le->le_name_chunk = BSWAP_16(le->le_name_chunk);
129 le->le_name_numints = BSWAP_16(le->le_name_numints);
130 le->le_value_chunk = BSWAP_16(le->le_value_chunk);
131 le->le_value_numints = BSWAP_16(le->le_value_numints);
132 le->le_cd = BSWAP_32(le->le_cd);
133 le->le_hash = BSWAP_64(le->le_hash);
134 break;
135 case ZAP_CHUNK_FREE:
136 lc->l_free.lf_type = BSWAP_8(lc->l_free.lf_type);
137 lc->l_free.lf_next = BSWAP_16(lc->l_free.lf_next);
138 break;
139 case ZAP_CHUNK_ARRAY:
140 lc->l_array.la_type = BSWAP_8(lc->l_array.la_type);
141 lc->l_array.la_next = BSWAP_16(lc->l_array.la_next);
142 /* la_array doesn't need swapping */
143 break;
144 default:
145 cmn_err(CE_PANIC, "bad leaf type %d",
146 lc->l_free.lf_type);
147 }
148 }
149 }
150
151 void
zap_leaf_init(zap_leaf_t * l,boolean_t sort)152 zap_leaf_init(zap_leaf_t *l, boolean_t sort)
153 {
154 l->l_bs = highbit64(l->l_dbuf->db_size) - 1;
155 memset(&zap_leaf_phys(l)->l_hdr, 0,
156 sizeof (struct zap_leaf_header));
157 memset(zap_leaf_phys(l)->l_hash, CHAIN_END,
158 2*ZAP_LEAF_HASH_NUMENTRIES(l));
159 for (uint_t i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
160 ZAP_LEAF_CHUNK(l, i).l_free.lf_type = ZAP_CHUNK_FREE;
161 ZAP_LEAF_CHUNK(l, i).l_free.lf_next = i+1;
162 }
163 ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)-1).l_free.lf_next = CHAIN_END;
164 zap_leaf_phys(l)->l_hdr.lh_block_type = ZBT_LEAF;
165 zap_leaf_phys(l)->l_hdr.lh_magic = ZAP_LEAF_MAGIC;
166 zap_leaf_phys(l)->l_hdr.lh_nfree = ZAP_LEAF_NUMCHUNKS(l);
167 if (sort)
168 zap_leaf_phys(l)->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
169 }
170
171 /*
172 * Routines which manipulate leaf chunks (l_chunk[]).
173 */
174
175 static uint16_t
zap_leaf_chunk_alloc(zap_leaf_t * l)176 zap_leaf_chunk_alloc(zap_leaf_t *l)
177 {
178 ASSERT(zap_leaf_phys(l)->l_hdr.lh_nfree > 0);
179
180 uint_t chunk = zap_leaf_phys(l)->l_hdr.lh_freelist;
181 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
182 ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_free.lf_type, ==, ZAP_CHUNK_FREE);
183
184 zap_leaf_phys(l)->l_hdr.lh_freelist =
185 ZAP_LEAF_CHUNK(l, chunk).l_free.lf_next;
186
187 zap_leaf_phys(l)->l_hdr.lh_nfree--;
188
189 return (chunk);
190 }
191
192 static void
zap_leaf_chunk_free(zap_leaf_t * l,uint16_t chunk)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(zap_leaf_phys(l)->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 = zap_leaf_phys(l)->l_hdr.lh_freelist;
202 memset(zlf->lf_pad, 0, sizeof (zlf->lf_pad)); /* help it to compress */
203 zap_leaf_phys(l)->l_hdr.lh_freelist = chunk;
204
205 zap_leaf_phys(l)->l_hdr.lh_nfree++;
206 }
207
208 /*
209 * Routines which manipulate leaf arrays (zap_leaf_array type chunks).
210 */
211
212 static uint16_t
zap_leaf_array_create(zap_leaf_t * l,const char * buf,int integer_size,int num_integers)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 = integer_size;
219 uint64_t value = 0;
220 int shift = (integer_size - 1) * 8;
221 int len = num_integers;
222
223 ASSERT3U(num_integers * integer_size, <=, ZAP_MAXVALUELEN);
224
225 if (len > 0)
226 value = ldv(integer_size, buf);
227 while (len > 0) {
228 uint16_t chunk = zap_leaf_chunk_alloc(l);
229 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
230
231 la->la_type = ZAP_CHUNK_ARRAY;
232 for (int i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) {
233 la->la_array[i] = value >> shift;
234 value <<= 8;
235 if (--byten == 0) {
236 if (--len == 0)
237 break;
238 byten = integer_size;
239 buf += integer_size;
240 value = ldv(integer_size, buf);
241 }
242 }
243
244 *chunkp = chunk;
245 chunkp = &la->la_next;
246 }
247 *chunkp = CHAIN_END;
248
249 return (chunk_head);
250 }
251
252 /*
253 * Non-destructively copy array between leaves.
254 */
255 static uint16_t
zap_leaf_array_copy(zap_leaf_t * l,uint16_t chunk,zap_leaf_t * nl)256 zap_leaf_array_copy(zap_leaf_t *l, uint16_t chunk, zap_leaf_t *nl)
257 {
258 uint16_t new_chunk;
259 uint16_t *nchunkp = &new_chunk;
260
261 while (chunk != CHAIN_END) {
262 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
263 uint16_t nchunk = zap_leaf_chunk_alloc(nl);
264
265 struct zap_leaf_array *la =
266 &ZAP_LEAF_CHUNK(l, chunk).l_array;
267 struct zap_leaf_array *nla =
268 &ZAP_LEAF_CHUNK(nl, nchunk).l_array;
269 ASSERT3U(la->la_type, ==, ZAP_CHUNK_ARRAY);
270
271 *nla = *la; /* structure assignment */
272
273 chunk = la->la_next;
274 *nchunkp = nchunk;
275 nchunkp = &nla->la_next;
276 }
277 *nchunkp = CHAIN_END;
278 return (new_chunk);
279 }
280
281 /*
282 * Free array. Unlike trivial loop of zap_leaf_chunk_free() this does
283 * not reverse order of chunks in the free list, reducing fragmentation.
284 */
285 static void
zap_leaf_array_free(zap_leaf_t * l,uint16_t chunk)286 zap_leaf_array_free(zap_leaf_t *l, uint16_t chunk)
287 {
288 struct zap_leaf_header *hdr = &zap_leaf_phys(l)->l_hdr;
289 uint16_t *tailp = &hdr->lh_freelist;
290 uint16_t oldfree = *tailp;
291
292 while (chunk != CHAIN_END) {
293 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
294 zap_leaf_chunk_t *c = &ZAP_LEAF_CHUNK(l, chunk);
295 ASSERT3U(c->l_array.la_type, ==, ZAP_CHUNK_ARRAY);
296
297 *tailp = chunk;
298 chunk = c->l_array.la_next;
299
300 c->l_free.lf_type = ZAP_CHUNK_FREE;
301 memset(c->l_free.lf_pad, 0, sizeof (c->l_free.lf_pad));
302 tailp = &c->l_free.lf_next;
303
304 ASSERT3U(hdr->lh_nfree, <, ZAP_LEAF_NUMCHUNKS(l));
305 hdr->lh_nfree++;
306 }
307
308 *tailp = oldfree;
309 }
310
311 /* array_len and buf_len are in integers, not bytes */
312 static void
zap_leaf_array_read(zap_leaf_t * l,uint16_t chunk,int array_int_len,int array_len,int buf_int_len,uint64_t buf_len,void * buf)313 zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk,
314 int array_int_len, int array_len, int buf_int_len, uint64_t buf_len,
315 void *buf)
316 {
317 int len = MIN(array_len, buf_len);
318 int byten = 0;
319 uint64_t value = 0;
320 char *p = buf;
321
322 ASSERT3U(array_int_len, <=, buf_int_len);
323
324 /* Fast path for one 8-byte integer */
325 if (array_int_len == 8 && buf_int_len == 8 && len == 1) {
326 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
327 uint8_t *ip = la->la_array;
328 uint64_t *buf64 = buf;
329
330 *buf64 = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 |
331 (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 |
332 (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 |
333 (uint64_t)ip[6] << 8 | (uint64_t)ip[7];
334 return;
335 }
336
337 /* Fast path for an array of 1-byte integers (eg. the entry name) */
338 if (array_int_len == 1 && buf_int_len == 1 &&
339 buf_len > array_len + ZAP_LEAF_ARRAY_BYTES) {
340 while (chunk != CHAIN_END) {
341 struct zap_leaf_array *la =
342 &ZAP_LEAF_CHUNK(l, chunk).l_array;
343 memcpy(p, la->la_array, ZAP_LEAF_ARRAY_BYTES);
344 p += ZAP_LEAF_ARRAY_BYTES;
345 chunk = la->la_next;
346 }
347 return;
348 }
349
350 while (len > 0) {
351 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
352
353 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
354 for (int i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) {
355 value = (value << 8) | la->la_array[i];
356 byten++;
357 if (byten == array_int_len) {
358 stv(buf_int_len, p, value);
359 byten = 0;
360 len--;
361 if (len == 0)
362 return;
363 p += buf_int_len;
364 }
365 }
366 chunk = la->la_next;
367 }
368 }
369
370 static boolean_t
zap_leaf_array_match(zap_leaf_t * l,zap_name_t * zn,uint_t chunk,int array_numints)371 zap_leaf_array_match(zap_leaf_t *l, zap_name_t *zn,
372 uint_t chunk, int array_numints)
373 {
374 int bseen = 0;
375
376 if (zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY) {
377 uint64_t *thiskey =
378 kmem_alloc(array_numints * sizeof (*thiskey), KM_SLEEP);
379 ASSERT(zn->zn_key_intlen == sizeof (*thiskey));
380
381 zap_leaf_array_read(l, chunk, sizeof (*thiskey), array_numints,
382 sizeof (*thiskey), array_numints, thiskey);
383 boolean_t match = memcmp(thiskey, zn->zn_key_orig,
384 array_numints * sizeof (*thiskey)) == 0;
385 kmem_free(thiskey, array_numints * sizeof (*thiskey));
386 return (match);
387 }
388
389 ASSERT(zn->zn_key_intlen == 1);
390 if (zn->zn_matchtype & MT_NORMALIZE) {
391 char *thisname = kmem_alloc(array_numints, KM_SLEEP);
392
393 zap_leaf_array_read(l, chunk, sizeof (char), array_numints,
394 sizeof (char), array_numints, thisname);
395 boolean_t match = zap_match(zn, thisname);
396 kmem_free(thisname, array_numints);
397 return (match);
398 }
399
400 /*
401 * Fast path for exact matching.
402 * First check that the lengths match, so that we don't read
403 * past the end of the zn_key_orig array.
404 */
405 if (array_numints != zn->zn_key_orig_numints)
406 return (B_FALSE);
407 while (bseen < array_numints) {
408 struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array;
409 int toread = MIN(array_numints - bseen, ZAP_LEAF_ARRAY_BYTES);
410 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
411 if (memcmp(la->la_array, (char *)zn->zn_key_orig + bseen,
412 toread))
413 break;
414 chunk = la->la_next;
415 bseen += toread;
416 }
417 return (bseen == array_numints);
418 }
419
420 /*
421 * Routines which manipulate leaf entries.
422 */
423
424 int
zap_leaf_lookup(zap_leaf_t * l,zap_name_t * zn,zap_entry_handle_t * zeh)425 zap_leaf_lookup(zap_leaf_t *l, zap_name_t *zn, zap_entry_handle_t *zeh)
426 {
427 struct zap_leaf_entry *le;
428
429 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
430
431 for (uint16_t *chunkp = LEAF_HASH_ENTPTR(l, zn->zn_hash);
432 *chunkp != CHAIN_END; chunkp = &le->le_next) {
433 uint16_t chunk = *chunkp;
434 le = ZAP_LEAF_ENTRY(l, chunk);
435
436 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
437 ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
438
439 if (le->le_hash != zn->zn_hash)
440 continue;
441
442 /*
443 * NB: the entry chain is always sorted by cd on
444 * normalized zap objects, so this will find the
445 * lowest-cd match for MT_NORMALIZE.
446 */
447 ASSERT((zn->zn_matchtype == 0) ||
448 (zap_leaf_phys(l)->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED));
449 if (zap_leaf_array_match(l, zn, le->le_name_chunk,
450 le->le_name_numints)) {
451 zeh->zeh_num_integers = le->le_value_numints;
452 zeh->zeh_integer_size = le->le_value_intlen;
453 zeh->zeh_cd = le->le_cd;
454 zeh->zeh_hash = le->le_hash;
455 zeh->zeh_chunkp = chunkp;
456 zeh->zeh_leaf = l;
457 return (0);
458 }
459 }
460
461 return (SET_ERROR(ENOENT));
462 }
463
464 /* Return (h1,cd1 >= h2,cd2) */
465 #define HCD_GTEQ(h1, cd1, h2, cd2) \
466 ((h1 > h2) ? TRUE : ((h1 == h2 && cd1 >= cd2) ? TRUE : FALSE))
467
468 int
zap_leaf_lookup_closest(zap_leaf_t * l,uint64_t h,uint32_t cd,zap_entry_handle_t * zeh)469 zap_leaf_lookup_closest(zap_leaf_t *l,
470 uint64_t h, uint32_t cd, zap_entry_handle_t *zeh)
471 {
472 uint64_t besth = -1ULL;
473 uint32_t bestcd = -1U;
474 uint16_t bestlh = ZAP_LEAF_HASH_NUMENTRIES(l)-1;
475 struct zap_leaf_entry *le;
476
477 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
478
479 for (uint16_t lh = LEAF_HASH(l, h); lh <= bestlh; lh++) {
480 for (uint16_t chunk = zap_leaf_phys(l)->l_hash[lh];
481 chunk != CHAIN_END; chunk = le->le_next) {
482 le = ZAP_LEAF_ENTRY(l, chunk);
483
484 ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l));
485 ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
486
487 if (HCD_GTEQ(le->le_hash, le->le_cd, h, cd) &&
488 HCD_GTEQ(besth, bestcd, le->le_hash, le->le_cd)) {
489 ASSERT3U(bestlh, >=, lh);
490 bestlh = lh;
491 besth = le->le_hash;
492 bestcd = le->le_cd;
493
494 zeh->zeh_num_integers = le->le_value_numints;
495 zeh->zeh_integer_size = le->le_value_intlen;
496 zeh->zeh_cd = le->le_cd;
497 zeh->zeh_hash = le->le_hash;
498 zeh->zeh_fakechunk = chunk;
499 zeh->zeh_chunkp = &zeh->zeh_fakechunk;
500 zeh->zeh_leaf = l;
501 }
502 }
503 }
504
505 return (bestcd == -1U ? SET_ERROR(ENOENT) : 0);
506 }
507
508 int
zap_entry_read(const zap_entry_handle_t * zeh,uint8_t integer_size,uint64_t num_integers,void * buf)509 zap_entry_read(const zap_entry_handle_t *zeh,
510 uint8_t integer_size, uint64_t num_integers, void *buf)
511 {
512 struct zap_leaf_entry *le =
513 ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
514 ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
515
516 if (le->le_value_intlen > integer_size)
517 return (SET_ERROR(EINVAL));
518
519 zap_leaf_array_read(zeh->zeh_leaf, le->le_value_chunk,
520 le->le_value_intlen, le->le_value_numints,
521 integer_size, num_integers, buf);
522
523 if (zeh->zeh_num_integers > num_integers)
524 return (SET_ERROR(EOVERFLOW));
525 return (0);
526
527 }
528
529 int
zap_entry_read_name(zap_t * zap,const zap_entry_handle_t * zeh,uint16_t buflen,char * buf)530 zap_entry_read_name(zap_t *zap, const zap_entry_handle_t *zeh, uint16_t buflen,
531 char *buf)
532 {
533 struct zap_leaf_entry *le =
534 ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp);
535 ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
536
537 if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) {
538 zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 8,
539 le->le_name_numints, 8, buflen / 8, buf);
540 } else {
541 zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 1,
542 le->le_name_numints, 1, buflen, buf);
543 }
544 if (le->le_name_numints > buflen)
545 return (SET_ERROR(EOVERFLOW));
546 return (0);
547 }
548
549 int
zap_entry_update(zap_entry_handle_t * zeh,uint8_t integer_size,uint64_t num_integers,const void * buf)550 zap_entry_update(zap_entry_handle_t *zeh,
551 uint8_t integer_size, uint64_t num_integers, const void *buf)
552 {
553 zap_leaf_t *l = zeh->zeh_leaf;
554 struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, *zeh->zeh_chunkp);
555
556 int delta_chunks = ZAP_LEAF_ARRAY_NCHUNKS(num_integers * integer_size) -
557 ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_numints * le->le_value_intlen);
558
559 if ((int)zap_leaf_phys(l)->l_hdr.lh_nfree < delta_chunks)
560 return (SET_ERROR(EAGAIN));
561
562 zap_leaf_array_free(l, le->le_value_chunk);
563 le->le_value_chunk =
564 zap_leaf_array_create(l, buf, integer_size, num_integers);
565 le->le_value_numints = num_integers;
566 le->le_value_intlen = integer_size;
567 return (0);
568 }
569
570 void
zap_entry_remove(zap_entry_handle_t * zeh)571 zap_entry_remove(zap_entry_handle_t *zeh)
572 {
573 zap_leaf_t *l = zeh->zeh_leaf;
574
575 ASSERT3P(zeh->zeh_chunkp, !=, &zeh->zeh_fakechunk);
576
577 uint16_t entry_chunk = *zeh->zeh_chunkp;
578 struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry_chunk);
579 ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
580
581 *zeh->zeh_chunkp = le->le_next;
582
583 /* Free in opposite order to reduce fragmentation. */
584 zap_leaf_array_free(l, le->le_value_chunk);
585 zap_leaf_array_free(l, le->le_name_chunk);
586 zap_leaf_chunk_free(l, entry_chunk);
587
588 zap_leaf_phys(l)->l_hdr.lh_nentries--;
589 }
590
591 int
zap_entry_create(zap_leaf_t * l,zap_name_t * zn,uint32_t cd,uint8_t integer_size,uint64_t num_integers,const void * buf,zap_entry_handle_t * zeh)592 zap_entry_create(zap_leaf_t *l, zap_name_t *zn, uint32_t cd,
593 uint8_t integer_size, uint64_t num_integers, const void *buf,
594 zap_entry_handle_t *zeh)
595 {
596 uint16_t chunk;
597 struct zap_leaf_entry *le;
598 uint64_t h = zn->zn_hash;
599
600 uint64_t valuelen = integer_size * num_integers;
601
602 uint_t numchunks = 1 + ZAP_LEAF_ARRAY_NCHUNKS(zn->zn_key_orig_numints *
603 zn->zn_key_intlen) + ZAP_LEAF_ARRAY_NCHUNKS(valuelen);
604 if (numchunks > ZAP_LEAF_NUMCHUNKS(l))
605 return (SET_ERROR(E2BIG));
606
607 if (cd == ZAP_NEED_CD) {
608 /* find the lowest unused cd */
609 if (zap_leaf_phys(l)->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED) {
610 cd = 0;
611
612 for (chunk = *LEAF_HASH_ENTPTR(l, h);
613 chunk != CHAIN_END; chunk = le->le_next) {
614 le = ZAP_LEAF_ENTRY(l, chunk);
615 if (le->le_cd > cd)
616 break;
617 if (le->le_hash == h) {
618 ASSERT3U(cd, ==, le->le_cd);
619 cd++;
620 }
621 }
622 } else {
623 /* old unsorted format; do it the O(n^2) way */
624 for (cd = 0; ; cd++) {
625 for (chunk = *LEAF_HASH_ENTPTR(l, h);
626 chunk != CHAIN_END; chunk = le->le_next) {
627 le = ZAP_LEAF_ENTRY(l, chunk);
628 if (le->le_hash == h &&
629 le->le_cd == cd) {
630 break;
631 }
632 }
633 /* If this cd is not in use, we are good. */
634 if (chunk == CHAIN_END)
635 break;
636 }
637 }
638 /*
639 * We would run out of space in a block before we could
640 * store enough entries to run out of CD values.
641 */
642 ASSERT3U(cd, <, zap_maxcd(zn->zn_zap));
643 }
644
645 if (zap_leaf_phys(l)->l_hdr.lh_nfree < numchunks)
646 return (SET_ERROR(EAGAIN));
647
648 /* make the entry */
649 chunk = zap_leaf_chunk_alloc(l);
650 le = ZAP_LEAF_ENTRY(l, chunk);
651 le->le_type = ZAP_CHUNK_ENTRY;
652 le->le_name_chunk = zap_leaf_array_create(l, zn->zn_key_orig,
653 zn->zn_key_intlen, zn->zn_key_orig_numints);
654 le->le_name_numints = zn->zn_key_orig_numints;
655 le->le_value_chunk =
656 zap_leaf_array_create(l, buf, integer_size, num_integers);
657 le->le_value_numints = num_integers;
658 le->le_value_intlen = integer_size;
659 le->le_hash = h;
660 le->le_cd = cd;
661
662 /* link it into the hash chain */
663 /* XXX if we did the search above, we could just use that */
664 uint16_t *chunkp = zap_leaf_rehash_entry(l, le, chunk);
665
666 zap_leaf_phys(l)->l_hdr.lh_nentries++;
667
668 zeh->zeh_leaf = l;
669 zeh->zeh_num_integers = num_integers;
670 zeh->zeh_integer_size = le->le_value_intlen;
671 zeh->zeh_cd = le->le_cd;
672 zeh->zeh_hash = le->le_hash;
673 zeh->zeh_chunkp = chunkp;
674
675 return (0);
676 }
677
678 /*
679 * Determine if there is another entry with the same normalized form.
680 * For performance purposes, either zn or name must be provided (the
681 * other can be NULL). Note, there usually won't be any hash
682 * conflicts, in which case we don't need the concatenated/normalized
683 * form of the name. But all callers have one of these on hand anyway,
684 * so might as well take advantage. A cleaner but slower interface
685 * would accept neither argument, and compute the normalized name as
686 * needed (using zap_name_alloc_str(zap_entry_read_name(zeh))).
687 */
688 boolean_t
zap_entry_normalization_conflict(zap_entry_handle_t * zeh,zap_name_t * zn,const char * name,zap_t * zap)689 zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn,
690 const char *name, zap_t *zap)
691 {
692 struct zap_leaf_entry *le;
693 boolean_t allocdzn = B_FALSE;
694
695 if (zap->zap_normflags == 0)
696 return (B_FALSE);
697
698 for (uint16_t chunk = *LEAF_HASH_ENTPTR(zeh->zeh_leaf, zeh->zeh_hash);
699 chunk != CHAIN_END; chunk = le->le_next) {
700 le = ZAP_LEAF_ENTRY(zeh->zeh_leaf, chunk);
701 if (le->le_hash != zeh->zeh_hash)
702 continue;
703 if (le->le_cd == zeh->zeh_cd)
704 continue;
705
706 if (zn == NULL) {
707 zn = zap_name_alloc_str(zap, name, MT_NORMALIZE);
708 allocdzn = B_TRUE;
709 }
710 if (zap_leaf_array_match(zeh->zeh_leaf, zn,
711 le->le_name_chunk, le->le_name_numints)) {
712 if (allocdzn)
713 zap_name_free(zn);
714 return (B_TRUE);
715 }
716 }
717 if (allocdzn)
718 zap_name_free(zn);
719 return (B_FALSE);
720 }
721
722 /*
723 * Routines for transferring entries between leafs.
724 */
725
726 static uint16_t *
zap_leaf_rehash_entry(zap_leaf_t * l,struct zap_leaf_entry * le,uint16_t entry)727 zap_leaf_rehash_entry(zap_leaf_t *l, struct zap_leaf_entry *le, uint16_t entry)
728 {
729 struct zap_leaf_entry *le2;
730 uint16_t *chunkp;
731
732 /*
733 * keep the entry chain sorted by cd
734 * NB: this will not cause problems for unsorted leafs, though
735 * it is unnecessary there.
736 */
737 for (chunkp = LEAF_HASH_ENTPTR(l, le->le_hash);
738 *chunkp != CHAIN_END; chunkp = &le2->le_next) {
739 le2 = ZAP_LEAF_ENTRY(l, *chunkp);
740 if (le2->le_cd > le->le_cd)
741 break;
742 }
743
744 le->le_next = *chunkp;
745 *chunkp = entry;
746 return (chunkp);
747 }
748
749 static void
zap_leaf_transfer_entry(zap_leaf_t * l,uint_t entry,zap_leaf_t * nl)750 zap_leaf_transfer_entry(zap_leaf_t *l, uint_t entry, zap_leaf_t *nl)
751 {
752 struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry);
753 ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY);
754
755 uint16_t chunk = zap_leaf_chunk_alloc(nl);
756 struct zap_leaf_entry *nle = ZAP_LEAF_ENTRY(nl, chunk);
757 *nle = *le; /* structure assignment */
758
759 (void) zap_leaf_rehash_entry(nl, nle, chunk);
760
761 nle->le_name_chunk = zap_leaf_array_copy(l, le->le_name_chunk, nl);
762 nle->le_value_chunk = zap_leaf_array_copy(l, le->le_value_chunk, nl);
763
764 /* Free in opposite order to reduce fragmentation. */
765 zap_leaf_array_free(l, le->le_value_chunk);
766 zap_leaf_array_free(l, le->le_name_chunk);
767 zap_leaf_chunk_free(l, entry);
768
769 zap_leaf_phys(l)->l_hdr.lh_nentries--;
770 zap_leaf_phys(nl)->l_hdr.lh_nentries++;
771 }
772
773 /*
774 * Transfer the entries whose hash prefix ends in 1 to the new leaf.
775 */
776 void
zap_leaf_split(zap_leaf_t * l,zap_leaf_t * nl,boolean_t sort)777 zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort)
778 {
779 uint_t bit = 64 - 1 - zap_leaf_phys(l)->l_hdr.lh_prefix_len;
780
781 /* set new prefix and prefix_len */
782 zap_leaf_phys(l)->l_hdr.lh_prefix <<= 1;
783 zap_leaf_phys(l)->l_hdr.lh_prefix_len++;
784 zap_leaf_phys(nl)->l_hdr.lh_prefix =
785 zap_leaf_phys(l)->l_hdr.lh_prefix | 1;
786 zap_leaf_phys(nl)->l_hdr.lh_prefix_len =
787 zap_leaf_phys(l)->l_hdr.lh_prefix_len;
788
789 /* break existing hash chains */
790 memset(zap_leaf_phys(l)->l_hash, CHAIN_END,
791 2*ZAP_LEAF_HASH_NUMENTRIES(l));
792
793 if (sort)
794 zap_leaf_phys(l)->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED;
795
796 /*
797 * Transfer entries whose hash bit 'bit' is set to nl; rehash
798 * the remaining entries
799 *
800 * NB: We could find entries via the hashtable instead. That
801 * would be O(hashents+numents) rather than O(numblks+numents),
802 * but this accesses memory more sequentially, and when we're
803 * called, the block is usually pretty full.
804 */
805 for (uint_t i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
806 struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, i);
807 if (le->le_type != ZAP_CHUNK_ENTRY)
808 continue;
809
810 if (le->le_hash & (1ULL << bit))
811 zap_leaf_transfer_entry(l, i, nl);
812 else
813 (void) zap_leaf_rehash_entry(l, le, i);
814 }
815 }
816
817 void
zap_leaf_stats(zap_t * zap,zap_leaf_t * l,zap_stats_t * zs)818 zap_leaf_stats(zap_t *zap, zap_leaf_t *l, zap_stats_t *zs)
819 {
820 uint_t n = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
821 zap_leaf_phys(l)->l_hdr.lh_prefix_len;
822 n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
823 zs->zs_leafs_with_2n_pointers[n]++;
824
825
826 n = zap_leaf_phys(l)->l_hdr.lh_nentries/5;
827 n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
828 zs->zs_blocks_with_n5_entries[n]++;
829
830 n = ((1<<FZAP_BLOCK_SHIFT(zap)) -
831 zap_leaf_phys(l)->l_hdr.lh_nfree * (ZAP_LEAF_ARRAY_BYTES+1))*10 /
832 (1<<FZAP_BLOCK_SHIFT(zap));
833 n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
834 zs->zs_blocks_n_tenths_full[n]++;
835
836 for (uint_t i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(l); i++) {
837 uint_t nentries = 0;
838 uint_t chunk = zap_leaf_phys(l)->l_hash[i];
839
840 while (chunk != CHAIN_END) {
841 struct zap_leaf_entry *le =
842 ZAP_LEAF_ENTRY(l, chunk);
843
844 n = 1 + ZAP_LEAF_ARRAY_NCHUNKS(le->le_name_numints) +
845 ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_numints *
846 le->le_value_intlen);
847 n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
848 zs->zs_entries_using_n_chunks[n]++;
849
850 chunk = le->le_next;
851 nentries++;
852 }
853
854 n = nentries;
855 n = MIN(n, ZAP_HISTOGRAM_SIZE-1);
856 zs->zs_buckets_with_n_entries[n]++;
857 }
858 }
859