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