]>
Commit | Line | Data |
---|---|---|
70e083d2 TG |
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 | */ | |
86e3c28a | 21 | |
70e083d2 TG |
22 | /* |
23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
86e3c28a CIK |
24 | * Copyright (c) 2013, 2015 by Delphix. All rights reserved. |
25 | * Copyright 2017 Nexenta Systems, Inc. | |
70e083d2 TG |
26 | */ |
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/zio.h> | |
35 | #include <sys/spa.h> | |
36 | #include <sys/dmu.h> | |
37 | #include <sys/zfs_context.h> | |
38 | #include <sys/fs/zfs.h> | |
39 | #include <sys/zap.h> | |
40 | #include <sys/zap_impl.h> | |
41 | #include <sys/zap_leaf.h> | |
42 | #include <sys/arc.h> | |
43 | ||
44 | static uint16_t *zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry); | |
45 | ||
46 | #define CHAIN_END 0xffff /* end of the chunk chain */ | |
47 | ||
48 | /* half the (current) minimum block size */ | |
49 | #define MAX_ARRAY_BYTES (8<<10) | |
50 | ||
51 | #define LEAF_HASH(l, h) \ | |
52 | ((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \ | |
53 | ((h) >> \ | |
54 | (64 - ZAP_LEAF_HASH_SHIFT(l) - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) | |
55 | ||
56 | #define LEAF_HASH_ENTPTR(l, h) (&zap_leaf_phys(l)->l_hash[LEAF_HASH(l, h)]) | |
57 | ||
58 | extern inline zap_leaf_phys_t *zap_leaf_phys(zap_leaf_t *l); | |
59 | ||
60 | static void | |
61 | zap_memset(void *a, int c, size_t n) | |
62 | { | |
63 | char *cp = a; | |
64 | char *cpend = cp + n; | |
65 | ||
66 | while (cp < cpend) | |
67 | *cp++ = c; | |
68 | } | |
69 | ||
70 | static void | |
71 | stv(int len, void *addr, uint64_t value) | |
72 | { | |
73 | switch (len) { | |
74 | case 1: | |
75 | *(uint8_t *)addr = value; | |
76 | return; | |
77 | case 2: | |
78 | *(uint16_t *)addr = value; | |
79 | return; | |
80 | case 4: | |
81 | *(uint32_t *)addr = value; | |
82 | return; | |
83 | case 8: | |
84 | *(uint64_t *)addr = value; | |
85 | return; | |
86 | default: | |
87 | cmn_err(CE_PANIC, "bad int len %d", len); | |
88 | } | |
89 | } | |
90 | ||
91 | static uint64_t | |
92 | ldv(int len, const void *addr) | |
93 | { | |
94 | switch (len) { | |
95 | case 1: | |
96 | return (*(uint8_t *)addr); | |
97 | case 2: | |
98 | return (*(uint16_t *)addr); | |
99 | case 4: | |
100 | return (*(uint32_t *)addr); | |
101 | case 8: | |
102 | return (*(uint64_t *)addr); | |
103 | default: | |
104 | cmn_err(CE_PANIC, "bad int len %d", len); | |
105 | } | |
106 | return (0xFEEDFACEDEADBEEFULL); | |
107 | } | |
108 | ||
109 | void | |
110 | zap_leaf_byteswap(zap_leaf_phys_t *buf, int size) | |
111 | { | |
112 | int i; | |
113 | zap_leaf_t l; | |
114 | dmu_buf_t l_dbuf; | |
115 | ||
116 | l_dbuf.db_data = buf; | |
117 | l.l_bs = highbit64(size) - 1; | |
118 | l.l_dbuf = &l_dbuf; | |
119 | ||
120 | buf->l_hdr.lh_block_type = BSWAP_64(buf->l_hdr.lh_block_type); | |
121 | buf->l_hdr.lh_prefix = BSWAP_64(buf->l_hdr.lh_prefix); | |
122 | buf->l_hdr.lh_magic = BSWAP_32(buf->l_hdr.lh_magic); | |
123 | buf->l_hdr.lh_nfree = BSWAP_16(buf->l_hdr.lh_nfree); | |
124 | buf->l_hdr.lh_nentries = BSWAP_16(buf->l_hdr.lh_nentries); | |
125 | buf->l_hdr.lh_prefix_len = BSWAP_16(buf->l_hdr.lh_prefix_len); | |
126 | buf->l_hdr.lh_freelist = BSWAP_16(buf->l_hdr.lh_freelist); | |
127 | ||
128 | for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(&l); i++) | |
129 | buf->l_hash[i] = BSWAP_16(buf->l_hash[i]); | |
130 | ||
131 | for (i = 0; i < ZAP_LEAF_NUMCHUNKS(&l); i++) { | |
132 | zap_leaf_chunk_t *lc = &ZAP_LEAF_CHUNK(&l, i); | |
133 | struct zap_leaf_entry *le; | |
134 | ||
135 | switch (lc->l_free.lf_type) { | |
136 | case ZAP_CHUNK_ENTRY: | |
137 | le = &lc->l_entry; | |
138 | ||
139 | le->le_type = BSWAP_8(le->le_type); | |
140 | le->le_value_intlen = BSWAP_8(le->le_value_intlen); | |
141 | le->le_next = BSWAP_16(le->le_next); | |
142 | le->le_name_chunk = BSWAP_16(le->le_name_chunk); | |
143 | le->le_name_numints = BSWAP_16(le->le_name_numints); | |
144 | le->le_value_chunk = BSWAP_16(le->le_value_chunk); | |
145 | le->le_value_numints = BSWAP_16(le->le_value_numints); | |
146 | le->le_cd = BSWAP_32(le->le_cd); | |
147 | le->le_hash = BSWAP_64(le->le_hash); | |
148 | break; | |
149 | case ZAP_CHUNK_FREE: | |
150 | lc->l_free.lf_type = BSWAP_8(lc->l_free.lf_type); | |
151 | lc->l_free.lf_next = BSWAP_16(lc->l_free.lf_next); | |
152 | break; | |
153 | case ZAP_CHUNK_ARRAY: | |
154 | lc->l_array.la_type = BSWAP_8(lc->l_array.la_type); | |
155 | lc->l_array.la_next = BSWAP_16(lc->l_array.la_next); | |
156 | /* la_array doesn't need swapping */ | |
157 | break; | |
158 | default: | |
159 | cmn_err(CE_PANIC, "bad leaf type %d", | |
160 | lc->l_free.lf_type); | |
161 | } | |
162 | } | |
163 | } | |
164 | ||
165 | void | |
166 | zap_leaf_init(zap_leaf_t *l, boolean_t sort) | |
167 | { | |
168 | int i; | |
169 | ||
170 | l->l_bs = highbit64(l->l_dbuf->db_size) - 1; | |
171 | zap_memset(&zap_leaf_phys(l)->l_hdr, 0, | |
172 | sizeof (struct zap_leaf_header)); | |
173 | zap_memset(zap_leaf_phys(l)->l_hash, CHAIN_END, | |
174 | 2*ZAP_LEAF_HASH_NUMENTRIES(l)); | |
175 | for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) { | |
176 | ZAP_LEAF_CHUNK(l, i).l_free.lf_type = ZAP_CHUNK_FREE; | |
177 | ZAP_LEAF_CHUNK(l, i).l_free.lf_next = i+1; | |
178 | } | |
179 | ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)-1).l_free.lf_next = CHAIN_END; | |
180 | zap_leaf_phys(l)->l_hdr.lh_block_type = ZBT_LEAF; | |
181 | zap_leaf_phys(l)->l_hdr.lh_magic = ZAP_LEAF_MAGIC; | |
182 | zap_leaf_phys(l)->l_hdr.lh_nfree = ZAP_LEAF_NUMCHUNKS(l); | |
183 | if (sort) | |
184 | zap_leaf_phys(l)->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED; | |
185 | } | |
186 | ||
187 | /* | |
188 | * Routines which manipulate leaf chunks (l_chunk[]). | |
189 | */ | |
190 | ||
191 | static uint16_t | |
192 | zap_leaf_chunk_alloc(zap_leaf_t *l) | |
193 | { | |
194 | int chunk; | |
195 | ||
196 | ASSERT(zap_leaf_phys(l)->l_hdr.lh_nfree > 0); | |
197 | ||
198 | chunk = zap_leaf_phys(l)->l_hdr.lh_freelist; | |
199 | ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l)); | |
200 | ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_free.lf_type, ==, ZAP_CHUNK_FREE); | |
201 | ||
202 | zap_leaf_phys(l)->l_hdr.lh_freelist = | |
203 | ZAP_LEAF_CHUNK(l, chunk).l_free.lf_next; | |
204 | ||
205 | zap_leaf_phys(l)->l_hdr.lh_nfree--; | |
206 | ||
207 | return (chunk); | |
208 | } | |
209 | ||
210 | static void | |
211 | zap_leaf_chunk_free(zap_leaf_t *l, uint16_t chunk) | |
212 | { | |
213 | struct zap_leaf_free *zlf = &ZAP_LEAF_CHUNK(l, chunk).l_free; | |
214 | ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_nfree, <, ZAP_LEAF_NUMCHUNKS(l)); | |
215 | ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l)); | |
216 | ASSERT(zlf->lf_type != ZAP_CHUNK_FREE); | |
217 | ||
218 | zlf->lf_type = ZAP_CHUNK_FREE; | |
219 | zlf->lf_next = zap_leaf_phys(l)->l_hdr.lh_freelist; | |
220 | bzero(zlf->lf_pad, sizeof (zlf->lf_pad)); /* help it to compress */ | |
221 | zap_leaf_phys(l)->l_hdr.lh_freelist = chunk; | |
222 | ||
223 | zap_leaf_phys(l)->l_hdr.lh_nfree++; | |
224 | } | |
225 | ||
226 | /* | |
227 | * Routines which manipulate leaf arrays (zap_leaf_array type chunks). | |
228 | */ | |
229 | ||
230 | static uint16_t | |
231 | zap_leaf_array_create(zap_leaf_t *l, const char *buf, | |
232 | int integer_size, int num_integers) | |
233 | { | |
234 | uint16_t chunk_head; | |
235 | uint16_t *chunkp = &chunk_head; | |
236 | int byten = 0; | |
237 | uint64_t value = 0; | |
238 | int shift = (integer_size-1)*8; | |
239 | int len = num_integers; | |
240 | ||
241 | ASSERT3U(num_integers * integer_size, <, MAX_ARRAY_BYTES); | |
242 | ||
243 | while (len > 0) { | |
244 | uint16_t chunk = zap_leaf_chunk_alloc(l); | |
245 | struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array; | |
246 | int i; | |
247 | ||
248 | la->la_type = ZAP_CHUNK_ARRAY; | |
249 | for (i = 0; i < ZAP_LEAF_ARRAY_BYTES; i++) { | |
250 | if (byten == 0) | |
251 | value = ldv(integer_size, buf); | |
252 | la->la_array[i] = value >> shift; | |
253 | value <<= 8; | |
254 | if (++byten == integer_size) { | |
255 | byten = 0; | |
256 | buf += integer_size; | |
257 | if (--len == 0) | |
258 | break; | |
259 | } | |
260 | } | |
261 | ||
262 | *chunkp = chunk; | |
263 | chunkp = &la->la_next; | |
264 | } | |
265 | *chunkp = CHAIN_END; | |
266 | ||
267 | return (chunk_head); | |
268 | } | |
269 | ||
270 | static void | |
271 | zap_leaf_array_free(zap_leaf_t *l, uint16_t *chunkp) | |
272 | { | |
273 | uint16_t chunk = *chunkp; | |
274 | ||
275 | *chunkp = CHAIN_END; | |
276 | ||
277 | while (chunk != CHAIN_END) { | |
278 | int nextchunk = ZAP_LEAF_CHUNK(l, chunk).l_array.la_next; | |
279 | ASSERT3U(ZAP_LEAF_CHUNK(l, chunk).l_array.la_type, ==, | |
280 | ZAP_CHUNK_ARRAY); | |
281 | zap_leaf_chunk_free(l, chunk); | |
282 | chunk = nextchunk; | |
283 | } | |
284 | } | |
285 | ||
286 | /* array_len and buf_len are in integers, not bytes */ | |
287 | static void | |
288 | zap_leaf_array_read(zap_leaf_t *l, uint16_t chunk, | |
289 | int array_int_len, int array_len, int buf_int_len, uint64_t buf_len, | |
290 | void *buf) | |
291 | { | |
292 | int len = MIN(array_len, buf_len); | |
293 | int byten = 0; | |
294 | uint64_t value = 0; | |
295 | char *p = buf; | |
296 | ||
297 | ASSERT3U(array_int_len, <=, buf_int_len); | |
298 | ||
299 | /* Fast path for one 8-byte integer */ | |
300 | if (array_int_len == 8 && buf_int_len == 8 && len == 1) { | |
301 | struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array; | |
302 | uint8_t *ip = la->la_array; | |
303 | uint64_t *buf64 = buf; | |
304 | ||
305 | *buf64 = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 | | |
306 | (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 | | |
307 | (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 | | |
308 | (uint64_t)ip[6] << 8 | (uint64_t)ip[7]; | |
309 | return; | |
310 | } | |
311 | ||
312 | /* Fast path for an array of 1-byte integers (eg. the entry name) */ | |
313 | if (array_int_len == 1 && buf_int_len == 1 && | |
314 | buf_len > array_len + ZAP_LEAF_ARRAY_BYTES) { | |
315 | while (chunk != CHAIN_END) { | |
316 | struct zap_leaf_array *la = | |
317 | &ZAP_LEAF_CHUNK(l, chunk).l_array; | |
318 | bcopy(la->la_array, p, ZAP_LEAF_ARRAY_BYTES); | |
319 | p += ZAP_LEAF_ARRAY_BYTES; | |
320 | chunk = la->la_next; | |
321 | } | |
322 | return; | |
323 | } | |
324 | ||
325 | while (len > 0) { | |
326 | struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array; | |
327 | int i; | |
328 | ||
329 | ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l)); | |
330 | for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) { | |
331 | value = (value << 8) | la->la_array[i]; | |
332 | byten++; | |
333 | if (byten == array_int_len) { | |
334 | stv(buf_int_len, p, value); | |
335 | byten = 0; | |
336 | len--; | |
337 | if (len == 0) | |
338 | return; | |
339 | p += buf_int_len; | |
340 | } | |
341 | } | |
342 | chunk = la->la_next; | |
343 | } | |
344 | } | |
345 | ||
346 | static boolean_t | |
347 | zap_leaf_array_match(zap_leaf_t *l, zap_name_t *zn, | |
348 | int chunk, int array_numints) | |
349 | { | |
350 | int bseen = 0; | |
351 | ||
352 | if (zap_getflags(zn->zn_zap) & ZAP_FLAG_UINT64_KEY) { | |
353 | uint64_t *thiskey; | |
354 | boolean_t match; | |
355 | ||
356 | ASSERT(zn->zn_key_intlen == sizeof (*thiskey)); | |
357 | thiskey = kmem_alloc(array_numints * sizeof (*thiskey), | |
358 | KM_SLEEP); | |
359 | ||
360 | zap_leaf_array_read(l, chunk, sizeof (*thiskey), array_numints, | |
361 | sizeof (*thiskey), array_numints, thiskey); | |
362 | match = bcmp(thiskey, zn->zn_key_orig, | |
363 | array_numints * sizeof (*thiskey)) == 0; | |
364 | kmem_free(thiskey, array_numints * sizeof (*thiskey)); | |
365 | return (match); | |
366 | } | |
367 | ||
368 | ASSERT(zn->zn_key_intlen == 1); | |
86e3c28a | 369 | if (zn->zn_matchtype & MT_NORMALIZE) { |
70e083d2 TG |
370 | char *thisname = kmem_alloc(array_numints, KM_SLEEP); |
371 | boolean_t match; | |
372 | ||
373 | zap_leaf_array_read(l, chunk, sizeof (char), array_numints, | |
374 | sizeof (char), array_numints, thisname); | |
375 | match = zap_match(zn, thisname); | |
376 | kmem_free(thisname, array_numints); | |
377 | return (match); | |
378 | } | |
379 | ||
380 | /* | |
381 | * Fast path for exact matching. | |
382 | * First check that the lengths match, so that we don't read | |
383 | * past the end of the zn_key_orig array. | |
384 | */ | |
385 | if (array_numints != zn->zn_key_orig_numints) | |
386 | return (B_FALSE); | |
387 | while (bseen < array_numints) { | |
388 | struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, chunk).l_array; | |
389 | int toread = MIN(array_numints - bseen, ZAP_LEAF_ARRAY_BYTES); | |
390 | ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l)); | |
391 | if (bcmp(la->la_array, (char *)zn->zn_key_orig + bseen, toread)) | |
392 | break; | |
393 | chunk = la->la_next; | |
394 | bseen += toread; | |
395 | } | |
396 | return (bseen == array_numints); | |
397 | } | |
398 | ||
399 | /* | |
400 | * Routines which manipulate leaf entries. | |
401 | */ | |
402 | ||
403 | int | |
404 | zap_leaf_lookup(zap_leaf_t *l, zap_name_t *zn, zap_entry_handle_t *zeh) | |
405 | { | |
406 | uint16_t *chunkp; | |
407 | struct zap_leaf_entry *le; | |
408 | ||
409 | ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC); | |
410 | ||
70e083d2 TG |
411 | for (chunkp = LEAF_HASH_ENTPTR(l, zn->zn_hash); |
412 | *chunkp != CHAIN_END; chunkp = &le->le_next) { | |
413 | uint16_t chunk = *chunkp; | |
414 | le = ZAP_LEAF_ENTRY(l, chunk); | |
415 | ||
416 | ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l)); | |
417 | ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY); | |
418 | ||
419 | if (le->le_hash != zn->zn_hash) | |
420 | continue; | |
421 | ||
422 | /* | |
423 | * NB: the entry chain is always sorted by cd on | |
424 | * normalized zap objects, so this will find the | |
86e3c28a | 425 | * lowest-cd match for MT_NORMALIZE. |
70e083d2 | 426 | */ |
86e3c28a | 427 | ASSERT((zn->zn_matchtype == 0) || |
70e083d2 TG |
428 | (zap_leaf_phys(l)->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED)); |
429 | if (zap_leaf_array_match(l, zn, le->le_name_chunk, | |
430 | le->le_name_numints)) { | |
431 | zeh->zeh_num_integers = le->le_value_numints; | |
432 | zeh->zeh_integer_size = le->le_value_intlen; | |
433 | zeh->zeh_cd = le->le_cd; | |
434 | zeh->zeh_hash = le->le_hash; | |
435 | zeh->zeh_chunkp = chunkp; | |
436 | zeh->zeh_leaf = l; | |
437 | return (0); | |
438 | } | |
439 | } | |
440 | ||
70e083d2 TG |
441 | return (SET_ERROR(ENOENT)); |
442 | } | |
443 | ||
444 | /* Return (h1,cd1 >= h2,cd2) */ | |
445 | #define HCD_GTEQ(h1, cd1, h2, cd2) \ | |
446 | ((h1 > h2) ? TRUE : ((h1 == h2 && cd1 >= cd2) ? TRUE : FALSE)) | |
447 | ||
448 | int | |
449 | zap_leaf_lookup_closest(zap_leaf_t *l, | |
450 | uint64_t h, uint32_t cd, zap_entry_handle_t *zeh) | |
451 | { | |
452 | uint16_t chunk; | |
453 | uint64_t besth = -1ULL; | |
454 | uint32_t bestcd = -1U; | |
455 | uint16_t bestlh = ZAP_LEAF_HASH_NUMENTRIES(l)-1; | |
456 | uint16_t lh; | |
457 | struct zap_leaf_entry *le; | |
458 | ||
459 | ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC); | |
460 | ||
461 | for (lh = LEAF_HASH(l, h); lh <= bestlh; lh++) { | |
462 | for (chunk = zap_leaf_phys(l)->l_hash[lh]; | |
463 | chunk != CHAIN_END; chunk = le->le_next) { | |
464 | le = ZAP_LEAF_ENTRY(l, chunk); | |
465 | ||
466 | ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l)); | |
467 | ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY); | |
468 | ||
469 | if (HCD_GTEQ(le->le_hash, le->le_cd, h, cd) && | |
470 | HCD_GTEQ(besth, bestcd, le->le_hash, le->le_cd)) { | |
471 | ASSERT3U(bestlh, >=, lh); | |
472 | bestlh = lh; | |
473 | besth = le->le_hash; | |
474 | bestcd = le->le_cd; | |
475 | ||
476 | zeh->zeh_num_integers = le->le_value_numints; | |
477 | zeh->zeh_integer_size = le->le_value_intlen; | |
478 | zeh->zeh_cd = le->le_cd; | |
479 | zeh->zeh_hash = le->le_hash; | |
480 | zeh->zeh_fakechunk = chunk; | |
481 | zeh->zeh_chunkp = &zeh->zeh_fakechunk; | |
482 | zeh->zeh_leaf = l; | |
483 | } | |
484 | } | |
485 | } | |
486 | ||
487 | return (bestcd == -1U ? ENOENT : 0); | |
488 | } | |
489 | ||
490 | int | |
491 | zap_entry_read(const zap_entry_handle_t *zeh, | |
492 | uint8_t integer_size, uint64_t num_integers, void *buf) | |
493 | { | |
494 | struct zap_leaf_entry *le = | |
495 | ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp); | |
496 | ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY); | |
497 | ||
498 | if (le->le_value_intlen > integer_size) | |
499 | return (SET_ERROR(EINVAL)); | |
500 | ||
501 | zap_leaf_array_read(zeh->zeh_leaf, le->le_value_chunk, | |
502 | le->le_value_intlen, le->le_value_numints, | |
503 | integer_size, num_integers, buf); | |
504 | ||
505 | if (zeh->zeh_num_integers > num_integers) | |
506 | return (SET_ERROR(EOVERFLOW)); | |
507 | return (0); | |
508 | ||
509 | } | |
510 | ||
511 | int | |
512 | zap_entry_read_name(zap_t *zap, const zap_entry_handle_t *zeh, uint16_t buflen, | |
513 | char *buf) | |
514 | { | |
515 | struct zap_leaf_entry *le = | |
516 | ZAP_LEAF_ENTRY(zeh->zeh_leaf, *zeh->zeh_chunkp); | |
517 | ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY); | |
518 | ||
519 | if (zap_getflags(zap) & ZAP_FLAG_UINT64_KEY) { | |
520 | zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 8, | |
521 | le->le_name_numints, 8, buflen / 8, buf); | |
522 | } else { | |
523 | zap_leaf_array_read(zeh->zeh_leaf, le->le_name_chunk, 1, | |
524 | le->le_name_numints, 1, buflen, buf); | |
525 | } | |
526 | if (le->le_name_numints > buflen) | |
527 | return (SET_ERROR(EOVERFLOW)); | |
528 | return (0); | |
529 | } | |
530 | ||
531 | int | |
532 | zap_entry_update(zap_entry_handle_t *zeh, | |
86e3c28a | 533 | uint8_t integer_size, uint64_t num_integers, const void *buf) |
70e083d2 TG |
534 | { |
535 | int delta_chunks; | |
536 | zap_leaf_t *l = zeh->zeh_leaf; | |
537 | struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, *zeh->zeh_chunkp); | |
538 | ||
539 | delta_chunks = ZAP_LEAF_ARRAY_NCHUNKS(num_integers * integer_size) - | |
540 | ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_numints * le->le_value_intlen); | |
541 | ||
542 | if ((int)zap_leaf_phys(l)->l_hdr.lh_nfree < delta_chunks) | |
543 | return (SET_ERROR(EAGAIN)); | |
544 | ||
545 | zap_leaf_array_free(l, &le->le_value_chunk); | |
546 | le->le_value_chunk = | |
547 | zap_leaf_array_create(l, buf, integer_size, num_integers); | |
548 | le->le_value_numints = num_integers; | |
549 | le->le_value_intlen = integer_size; | |
550 | return (0); | |
551 | } | |
552 | ||
553 | void | |
554 | zap_entry_remove(zap_entry_handle_t *zeh) | |
555 | { | |
556 | uint16_t entry_chunk; | |
557 | struct zap_leaf_entry *le; | |
558 | zap_leaf_t *l = zeh->zeh_leaf; | |
559 | ||
560 | ASSERT3P(zeh->zeh_chunkp, !=, &zeh->zeh_fakechunk); | |
561 | ||
562 | entry_chunk = *zeh->zeh_chunkp; | |
563 | le = ZAP_LEAF_ENTRY(l, entry_chunk); | |
564 | ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY); | |
565 | ||
566 | zap_leaf_array_free(l, &le->le_name_chunk); | |
567 | zap_leaf_array_free(l, &le->le_value_chunk); | |
568 | ||
569 | *zeh->zeh_chunkp = le->le_next; | |
570 | zap_leaf_chunk_free(l, entry_chunk); | |
571 | ||
572 | zap_leaf_phys(l)->l_hdr.lh_nentries--; | |
573 | } | |
574 | ||
575 | int | |
576 | zap_entry_create(zap_leaf_t *l, zap_name_t *zn, uint32_t cd, | |
577 | uint8_t integer_size, uint64_t num_integers, const void *buf, | |
578 | zap_entry_handle_t *zeh) | |
579 | { | |
580 | uint16_t chunk; | |
581 | uint16_t *chunkp; | |
582 | struct zap_leaf_entry *le; | |
583 | uint64_t valuelen; | |
584 | int numchunks; | |
585 | uint64_t h = zn->zn_hash; | |
586 | ||
587 | valuelen = integer_size * num_integers; | |
588 | ||
589 | numchunks = 1 + ZAP_LEAF_ARRAY_NCHUNKS(zn->zn_key_orig_numints * | |
590 | zn->zn_key_intlen) + ZAP_LEAF_ARRAY_NCHUNKS(valuelen); | |
591 | if (numchunks > ZAP_LEAF_NUMCHUNKS(l)) | |
592 | return (E2BIG); | |
593 | ||
594 | if (cd == ZAP_NEED_CD) { | |
595 | /* find the lowest unused cd */ | |
596 | if (zap_leaf_phys(l)->l_hdr.lh_flags & ZLF_ENTRIES_CDSORTED) { | |
597 | cd = 0; | |
598 | ||
599 | for (chunk = *LEAF_HASH_ENTPTR(l, h); | |
600 | chunk != CHAIN_END; chunk = le->le_next) { | |
601 | le = ZAP_LEAF_ENTRY(l, chunk); | |
602 | if (le->le_cd > cd) | |
603 | break; | |
604 | if (le->le_hash == h) { | |
605 | ASSERT3U(cd, ==, le->le_cd); | |
606 | cd++; | |
607 | } | |
608 | } | |
609 | } else { | |
610 | /* old unsorted format; do it the O(n^2) way */ | |
611 | for (cd = 0; ; cd++) { | |
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_hash == h && | |
616 | le->le_cd == cd) { | |
617 | break; | |
618 | } | |
619 | } | |
620 | /* If this cd is not in use, we are good. */ | |
621 | if (chunk == CHAIN_END) | |
622 | break; | |
623 | } | |
624 | } | |
625 | /* | |
626 | * We would run out of space in a block before we could | |
627 | * store enough entries to run out of CD values. | |
628 | */ | |
629 | ASSERT3U(cd, <, zap_maxcd(zn->zn_zap)); | |
630 | } | |
631 | ||
632 | if (zap_leaf_phys(l)->l_hdr.lh_nfree < numchunks) | |
633 | return (SET_ERROR(EAGAIN)); | |
634 | ||
635 | /* make the entry */ | |
636 | chunk = zap_leaf_chunk_alloc(l); | |
637 | le = ZAP_LEAF_ENTRY(l, chunk); | |
638 | le->le_type = ZAP_CHUNK_ENTRY; | |
639 | le->le_name_chunk = zap_leaf_array_create(l, zn->zn_key_orig, | |
640 | zn->zn_key_intlen, zn->zn_key_orig_numints); | |
641 | le->le_name_numints = zn->zn_key_orig_numints; | |
642 | le->le_value_chunk = | |
643 | zap_leaf_array_create(l, buf, integer_size, num_integers); | |
644 | le->le_value_numints = num_integers; | |
645 | le->le_value_intlen = integer_size; | |
646 | le->le_hash = h; | |
647 | le->le_cd = cd; | |
648 | ||
649 | /* link it into the hash chain */ | |
650 | /* XXX if we did the search above, we could just use that */ | |
651 | chunkp = zap_leaf_rehash_entry(l, chunk); | |
652 | ||
653 | zap_leaf_phys(l)->l_hdr.lh_nentries++; | |
654 | ||
655 | zeh->zeh_leaf = l; | |
656 | zeh->zeh_num_integers = num_integers; | |
657 | zeh->zeh_integer_size = le->le_value_intlen; | |
658 | zeh->zeh_cd = le->le_cd; | |
659 | zeh->zeh_hash = le->le_hash; | |
660 | zeh->zeh_chunkp = chunkp; | |
661 | ||
662 | return (0); | |
663 | } | |
664 | ||
665 | /* | |
666 | * Determine if there is another entry with the same normalized form. | |
667 | * For performance purposes, either zn or name must be provided (the | |
668 | * other can be NULL). Note, there usually won't be any hash | |
669 | * conflicts, in which case we don't need the concatenated/normalized | |
670 | * form of the name. But all callers have one of these on hand anyway, | |
671 | * so might as well take advantage. A cleaner but slower interface | |
672 | * would accept neither argument, and compute the normalized name as | |
673 | * needed (using zap_name_alloc(zap_entry_read_name(zeh))). | |
674 | */ | |
675 | boolean_t | |
676 | zap_entry_normalization_conflict(zap_entry_handle_t *zeh, zap_name_t *zn, | |
677 | const char *name, zap_t *zap) | |
678 | { | |
679 | uint64_t chunk; | |
680 | struct zap_leaf_entry *le; | |
681 | boolean_t allocdzn = B_FALSE; | |
682 | ||
683 | if (zap->zap_normflags == 0) | |
684 | return (B_FALSE); | |
685 | ||
686 | for (chunk = *LEAF_HASH_ENTPTR(zeh->zeh_leaf, zeh->zeh_hash); | |
687 | chunk != CHAIN_END; chunk = le->le_next) { | |
688 | le = ZAP_LEAF_ENTRY(zeh->zeh_leaf, chunk); | |
689 | if (le->le_hash != zeh->zeh_hash) | |
690 | continue; | |
691 | if (le->le_cd == zeh->zeh_cd) | |
692 | continue; | |
693 | ||
694 | if (zn == NULL) { | |
86e3c28a | 695 | zn = zap_name_alloc(zap, name, MT_NORMALIZE); |
70e083d2 TG |
696 | allocdzn = B_TRUE; |
697 | } | |
698 | if (zap_leaf_array_match(zeh->zeh_leaf, zn, | |
699 | le->le_name_chunk, le->le_name_numints)) { | |
700 | if (allocdzn) | |
701 | zap_name_free(zn); | |
702 | return (B_TRUE); | |
703 | } | |
704 | } | |
705 | if (allocdzn) | |
706 | zap_name_free(zn); | |
707 | return (B_FALSE); | |
708 | } | |
709 | ||
710 | /* | |
711 | * Routines for transferring entries between leafs. | |
712 | */ | |
713 | ||
714 | static uint16_t * | |
715 | zap_leaf_rehash_entry(zap_leaf_t *l, uint16_t entry) | |
716 | { | |
717 | struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, entry); | |
718 | struct zap_leaf_entry *le2; | |
719 | uint16_t *chunkp; | |
720 | ||
721 | /* | |
722 | * keep the entry chain sorted by cd | |
723 | * NB: this will not cause problems for unsorted leafs, though | |
724 | * it is unnecessary there. | |
725 | */ | |
726 | for (chunkp = LEAF_HASH_ENTPTR(l, le->le_hash); | |
727 | *chunkp != CHAIN_END; chunkp = &le2->le_next) { | |
728 | le2 = ZAP_LEAF_ENTRY(l, *chunkp); | |
729 | if (le2->le_cd > le->le_cd) | |
730 | break; | |
731 | } | |
732 | ||
733 | le->le_next = *chunkp; | |
734 | *chunkp = entry; | |
735 | return (chunkp); | |
736 | } | |
737 | ||
738 | static uint16_t | |
739 | zap_leaf_transfer_array(zap_leaf_t *l, uint16_t chunk, zap_leaf_t *nl) | |
740 | { | |
741 | uint16_t new_chunk; | |
742 | uint16_t *nchunkp = &new_chunk; | |
743 | ||
744 | while (chunk != CHAIN_END) { | |
745 | uint16_t nchunk = zap_leaf_chunk_alloc(nl); | |
746 | struct zap_leaf_array *nla = | |
747 | &ZAP_LEAF_CHUNK(nl, nchunk).l_array; | |
748 | struct zap_leaf_array *la = | |
749 | &ZAP_LEAF_CHUNK(l, chunk).l_array; | |
750 | int nextchunk = la->la_next; | |
751 | ||
752 | ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(l)); | |
753 | ASSERT3U(nchunk, <, ZAP_LEAF_NUMCHUNKS(l)); | |
754 | ||
755 | *nla = *la; /* structure assignment */ | |
756 | ||
757 | zap_leaf_chunk_free(l, chunk); | |
758 | chunk = nextchunk; | |
759 | *nchunkp = nchunk; | |
760 | nchunkp = &nla->la_next; | |
761 | } | |
762 | *nchunkp = CHAIN_END; | |
763 | return (new_chunk); | |
764 | } | |
765 | ||
766 | static void | |
767 | zap_leaf_transfer_entry(zap_leaf_t *l, int entry, zap_leaf_t *nl) | |
768 | { | |
769 | struct zap_leaf_entry *le, *nle; | |
770 | uint16_t chunk; | |
771 | ||
772 | le = ZAP_LEAF_ENTRY(l, entry); | |
773 | ASSERT3U(le->le_type, ==, ZAP_CHUNK_ENTRY); | |
774 | ||
775 | chunk = zap_leaf_chunk_alloc(nl); | |
776 | nle = ZAP_LEAF_ENTRY(nl, chunk); | |
777 | *nle = *le; /* structure assignment */ | |
778 | ||
779 | (void) zap_leaf_rehash_entry(nl, chunk); | |
780 | ||
781 | nle->le_name_chunk = zap_leaf_transfer_array(l, le->le_name_chunk, nl); | |
782 | nle->le_value_chunk = | |
783 | zap_leaf_transfer_array(l, le->le_value_chunk, nl); | |
784 | ||
785 | zap_leaf_chunk_free(l, entry); | |
786 | ||
787 | zap_leaf_phys(l)->l_hdr.lh_nentries--; | |
788 | zap_leaf_phys(nl)->l_hdr.lh_nentries++; | |
789 | } | |
790 | ||
791 | /* | |
792 | * Transfer the entries whose hash prefix ends in 1 to the new leaf. | |
793 | */ | |
794 | void | |
795 | zap_leaf_split(zap_leaf_t *l, zap_leaf_t *nl, boolean_t sort) | |
796 | { | |
797 | int i; | |
798 | int bit = 64 - 1 - zap_leaf_phys(l)->l_hdr.lh_prefix_len; | |
799 | ||
800 | /* set new prefix and prefix_len */ | |
801 | zap_leaf_phys(l)->l_hdr.lh_prefix <<= 1; | |
802 | zap_leaf_phys(l)->l_hdr.lh_prefix_len++; | |
803 | zap_leaf_phys(nl)->l_hdr.lh_prefix = | |
804 | zap_leaf_phys(l)->l_hdr.lh_prefix | 1; | |
805 | zap_leaf_phys(nl)->l_hdr.lh_prefix_len = | |
806 | zap_leaf_phys(l)->l_hdr.lh_prefix_len; | |
807 | ||
808 | /* break existing hash chains */ | |
809 | zap_memset(zap_leaf_phys(l)->l_hash, CHAIN_END, | |
810 | 2*ZAP_LEAF_HASH_NUMENTRIES(l)); | |
811 | ||
812 | if (sort) | |
813 | zap_leaf_phys(l)->l_hdr.lh_flags |= ZLF_ENTRIES_CDSORTED; | |
814 | ||
815 | /* | |
816 | * Transfer entries whose hash bit 'bit' is set to nl; rehash | |
817 | * the remaining entries | |
818 | * | |
819 | * NB: We could find entries via the hashtable instead. That | |
820 | * would be O(hashents+numents) rather than O(numblks+numents), | |
821 | * but this accesses memory more sequentially, and when we're | |
822 | * called, the block is usually pretty full. | |
823 | */ | |
824 | for (i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) { | |
825 | struct zap_leaf_entry *le = ZAP_LEAF_ENTRY(l, i); | |
826 | if (le->le_type != ZAP_CHUNK_ENTRY) | |
827 | continue; | |
828 | ||
829 | if (le->le_hash & (1ULL << bit)) | |
830 | zap_leaf_transfer_entry(l, i, nl); | |
831 | else | |
832 | (void) zap_leaf_rehash_entry(l, i); | |
833 | } | |
834 | } | |
835 | ||
836 | void | |
837 | zap_leaf_stats(zap_t *zap, zap_leaf_t *l, zap_stats_t *zs) | |
838 | { | |
839 | int i, n; | |
840 | ||
841 | n = zap_f_phys(zap)->zap_ptrtbl.zt_shift - | |
842 | zap_leaf_phys(l)->l_hdr.lh_prefix_len; | |
843 | n = MIN(n, ZAP_HISTOGRAM_SIZE-1); | |
844 | zs->zs_leafs_with_2n_pointers[n]++; | |
845 | ||
846 | ||
847 | n = zap_leaf_phys(l)->l_hdr.lh_nentries/5; | |
848 | n = MIN(n, ZAP_HISTOGRAM_SIZE-1); | |
849 | zs->zs_blocks_with_n5_entries[n]++; | |
850 | ||
851 | n = ((1<<FZAP_BLOCK_SHIFT(zap)) - | |
852 | zap_leaf_phys(l)->l_hdr.lh_nfree * (ZAP_LEAF_ARRAY_BYTES+1))*10 / | |
853 | (1<<FZAP_BLOCK_SHIFT(zap)); | |
854 | n = MIN(n, ZAP_HISTOGRAM_SIZE-1); | |
855 | zs->zs_blocks_n_tenths_full[n]++; | |
856 | ||
857 | for (i = 0; i < ZAP_LEAF_HASH_NUMENTRIES(l); i++) { | |
858 | int nentries = 0; | |
859 | int chunk = zap_leaf_phys(l)->l_hash[i]; | |
860 | ||
861 | while (chunk != CHAIN_END) { | |
862 | struct zap_leaf_entry *le = | |
863 | ZAP_LEAF_ENTRY(l, chunk); | |
864 | ||
865 | n = 1 + ZAP_LEAF_ARRAY_NCHUNKS(le->le_name_numints) + | |
866 | ZAP_LEAF_ARRAY_NCHUNKS(le->le_value_numints * | |
867 | le->le_value_intlen); | |
868 | n = MIN(n, ZAP_HISTOGRAM_SIZE-1); | |
869 | zs->zs_entries_using_n_chunks[n]++; | |
870 | ||
871 | chunk = le->le_next; | |
872 | nentries++; | |
873 | } | |
874 | ||
875 | n = nentries; | |
876 | n = MIN(n, ZAP_HISTOGRAM_SIZE-1); | |
877 | zs->zs_buckets_with_n_entries[n]++; | |
878 | } | |
879 | } |