1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2022 The FreeBSD Foundation
5 *
6 * This software was developed by Mark Johnston under sponsorship from
7 * the FreeBSD Foundation.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions are
11 * met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31 #include <sys/types.h>
32 #include <sys/endian.h>
33
34 #include <assert.h>
35 #include <stddef.h>
36 #include <stdlib.h>
37 #include <string.h>
38
39 #include <util.h>
40
41 #include "makefs.h"
42 #include "zfs.h"
43
44 typedef struct zfs_zap_entry {
45 char *name; /* entry key, private copy */
46 uint64_t hash; /* key hash */
47 union {
48 uint8_t *valp;
49 uint16_t *val16p;
50 uint32_t *val32p;
51 uint64_t *val64p;
52 }; /* entry value, an integer array */
53 uint64_t val64; /* embedded value for a common case */
54 size_t intsz; /* array element size; 1, 2, 4 or 8 */
55 size_t intcnt; /* array size */
56 STAILQ_ENTRY(zfs_zap_entry) next;
57 } zfs_zap_entry_t;
58
59 struct zfs_zap {
60 STAILQ_HEAD(, zfs_zap_entry) kvps;
61 uint64_t hashsalt; /* key hash input */
62 unsigned long kvpcnt; /* number of key-value pairs */
63 unsigned long chunks; /* count of chunks needed for fat ZAP */
64 bool micro; /* can this be a micro ZAP? */
65
66 dnode_phys_t *dnode; /* backpointer */
67 zfs_objset_t *os; /* backpointer */
68 };
69
70 static uint16_t
zap_entry_chunks(zfs_zap_entry_t * ent)71 zap_entry_chunks(zfs_zap_entry_t *ent)
72 {
73 return (1 + howmany(strlen(ent->name) + 1, ZAP_LEAF_ARRAY_BYTES) +
74 howmany(ent->intsz * ent->intcnt, ZAP_LEAF_ARRAY_BYTES));
75 }
76
77 static uint64_t
zap_hash(uint64_t salt,const char * name)78 zap_hash(uint64_t salt, const char *name)
79 {
80 static uint64_t crc64_table[256];
81 const uint64_t crc64_poly = 0xC96C5795D7870F42UL;
82 const uint8_t *cp;
83 uint64_t crc;
84 uint8_t c;
85
86 assert(salt != 0);
87 if (crc64_table[128] == 0) {
88 for (int i = 0; i < 256; i++) {
89 uint64_t *t;
90
91 t = crc64_table + i;
92 *t = i;
93 for (int j = 8; j > 0; j--)
94 *t = (*t >> 1) ^ (-(*t & 1) & crc64_poly);
95 }
96 }
97 assert(crc64_table[128] == crc64_poly);
98
99 for (cp = (const uint8_t *)name, crc = salt; (c = *cp) != '\0'; cp++)
100 crc = (crc >> 8) ^ crc64_table[(crc ^ c) & 0xFF];
101
102 /*
103 * Only use 28 bits, since we need 4 bits in the cookie for the
104 * collision differentiator. We MUST use the high bits, since
105 * those are the ones that we first pay attention to when
106 * choosing the bucket.
107 */
108 crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
109
110 return (crc);
111 }
112
113 zfs_zap_t *
zap_alloc(zfs_objset_t * os,dnode_phys_t * dnode)114 zap_alloc(zfs_objset_t *os, dnode_phys_t *dnode)
115 {
116 zfs_zap_t *zap;
117
118 zap = ecalloc(1, sizeof(*zap));
119 STAILQ_INIT(&zap->kvps);
120 zap->hashsalt = ((uint64_t)random() << 32) | random();
121 zap->micro = true;
122 zap->kvpcnt = 0;
123 zap->chunks = 0;
124 zap->dnode = dnode;
125 zap->os = os;
126 return (zap);
127 }
128
129 void
zap_add(zfs_zap_t * zap,const char * name,size_t intsz,size_t intcnt,const uint8_t * val)130 zap_add(zfs_zap_t *zap, const char *name, size_t intsz, size_t intcnt,
131 const uint8_t *val)
132 {
133 zfs_zap_entry_t *ent;
134
135 assert(intsz == 1 || intsz == 2 || intsz == 4 || intsz == 8);
136 assert(strlen(name) + 1 <= ZAP_MAXNAMELEN);
137 assert(intcnt <= ZAP_MAXVALUELEN && intcnt * intsz <= ZAP_MAXVALUELEN);
138
139 ent = ecalloc(1, sizeof(*ent));
140 ent->name = estrdup(name);
141 ent->hash = zap_hash(zap->hashsalt, ent->name);
142 ent->intsz = intsz;
143 ent->intcnt = intcnt;
144 if (intsz == sizeof(uint64_t) && intcnt == 1) {
145 /*
146 * Micro-optimization to elide a memory allocation in that most
147 * common case where this is a directory entry.
148 */
149 ent->val64p = &ent->val64;
150 } else {
151 ent->valp = ecalloc(intcnt, intsz);
152 }
153 memcpy(ent->valp, val, intcnt * intsz);
154 zap->kvpcnt++;
155 zap->chunks += zap_entry_chunks(ent);
156 STAILQ_INSERT_TAIL(&zap->kvps, ent, next);
157
158 if (zap->micro && (intcnt != 1 || intsz != sizeof(uint64_t) ||
159 strlen(name) + 1 > MZAP_NAME_LEN || zap->kvpcnt > MZAP_ENT_MAX))
160 zap->micro = false;
161 }
162
163 void
zap_add_uint64(zfs_zap_t * zap,const char * name,uint64_t val)164 zap_add_uint64(zfs_zap_t *zap, const char *name, uint64_t val)
165 {
166 zap_add(zap, name, sizeof(uint64_t), 1, (uint8_t *)&val);
167 }
168
169 void
zap_add_uint64_self(zfs_zap_t * zap,uint64_t val)170 zap_add_uint64_self(zfs_zap_t *zap, uint64_t val)
171 {
172 char name[32];
173
174 snprintf(name, sizeof(name), "%jx", (uintmax_t)val);
175 zap_add(zap, name, sizeof(uint64_t), 1, (uint8_t *)&val);
176 }
177
178 void
zap_add_string(zfs_zap_t * zap,const char * name,const char * val)179 zap_add_string(zfs_zap_t *zap, const char *name, const char *val)
180 {
181 zap_add(zap, name, 1, strlen(val) + 1, val);
182 }
183
184 bool
zap_entry_exists(zfs_zap_t * zap,const char * name)185 zap_entry_exists(zfs_zap_t *zap, const char *name)
186 {
187 zfs_zap_entry_t *ent;
188
189 STAILQ_FOREACH(ent, &zap->kvps, next) {
190 if (strcmp(ent->name, name) == 0)
191 return (true);
192 }
193 return (false);
194 }
195
196 static void
zap_micro_write(zfs_opt_t * zfs,zfs_zap_t * zap)197 zap_micro_write(zfs_opt_t *zfs, zfs_zap_t *zap)
198 {
199 dnode_phys_t *dnode;
200 zfs_zap_entry_t *ent;
201 mzap_phys_t *mzap;
202 mzap_ent_phys_t *ment;
203 off_t bytes, loc;
204
205 memset(zfs->filebuf, 0, sizeof(zfs->filebuf));
206 mzap = (mzap_phys_t *)&zfs->filebuf[0];
207 mzap->mz_block_type = ZBT_MICRO;
208 mzap->mz_salt = zap->hashsalt;
209 mzap->mz_normflags = 0;
210
211 bytes = sizeof(*mzap) + (zap->kvpcnt - 1) * sizeof(*ment);
212 assert(bytes <= (off_t)MZAP_MAX_BLKSZ);
213
214 ment = &mzap->mz_chunk[0];
215 STAILQ_FOREACH(ent, &zap->kvps, next) {
216 memcpy(&ment->mze_value, ent->valp, ent->intsz * ent->intcnt);
217 ment->mze_cd = 0; /* XXX-MJ */
218 strlcpy(ment->mze_name, ent->name, sizeof(ment->mze_name));
219 ment++;
220 }
221
222 loc = objset_space_alloc(zfs, zap->os, &bytes);
223
224 dnode = zap->dnode;
225 dnode->dn_maxblkid = 0;
226 dnode->dn_datablkszsec = bytes >> MINBLOCKSHIFT;
227
228 vdev_pwrite_dnode_data(zfs, dnode, zfs->filebuf, bytes, loc);
229 }
230
231 /*
232 * Write some data to the fat ZAP leaf chunk starting at index "li".
233 *
234 * Note that individual integers in the value may be split among consecutive
235 * leaves.
236 */
237 static void
zap_fat_write_array_chunk(zap_leaf_t * l,uint16_t li,size_t sz,const uint8_t * val)238 zap_fat_write_array_chunk(zap_leaf_t *l, uint16_t li, size_t sz,
239 const uint8_t *val)
240 {
241 struct zap_leaf_array *la;
242
243 assert(sz <= ZAP_MAXVALUELEN);
244
245 for (uint16_t n, resid = sz; resid > 0; resid -= n, val += n, li++) {
246 n = MIN(resid, ZAP_LEAF_ARRAY_BYTES);
247
248 la = &ZAP_LEAF_CHUNK(l, li).l_array;
249 assert(la->la_type == ZAP_CHUNK_FREE);
250 la->la_type = ZAP_CHUNK_ARRAY;
251 memcpy(la->la_array, val, n);
252 la->la_next = li + 1;
253 }
254 la->la_next = 0xffff;
255 }
256
257 /*
258 * Find the shortest hash prefix length which lets us distribute keys without
259 * overflowing a leaf block. This is not (space) optimal, but is simple, and
260 * directories large enough to overflow a single 128KB leaf block are uncommon.
261 */
262 static unsigned int
zap_fat_write_prefixlen(zfs_zap_t * zap,zap_leaf_t * l)263 zap_fat_write_prefixlen(zfs_zap_t *zap, zap_leaf_t *l)
264 {
265 zfs_zap_entry_t *ent;
266 unsigned int prefixlen;
267
268 if (zap->chunks <= ZAP_LEAF_NUMCHUNKS(l)) {
269 /*
270 * All chunks will fit in a single leaf block.
271 */
272 return (0);
273 }
274
275 for (prefixlen = 1; prefixlen < (unsigned int)l->l_bs; prefixlen++) {
276 uint32_t *leafchunks;
277
278 leafchunks = ecalloc(1u << prefixlen, sizeof(*leafchunks));
279 STAILQ_FOREACH(ent, &zap->kvps, next) {
280 uint64_t li;
281 uint16_t chunks;
282
283 li = ZAP_HASH_IDX(ent->hash, prefixlen);
284
285 chunks = zap_entry_chunks(ent);
286 if (ZAP_LEAF_NUMCHUNKS(l) - leafchunks[li] < chunks) {
287 /*
288 * Not enough space, grow the prefix and retry.
289 */
290 break;
291 }
292 leafchunks[li] += chunks;
293 }
294 free(leafchunks);
295
296 if (ent == NULL) {
297 /*
298 * Everything fits, we're done.
299 */
300 break;
301 }
302 }
303
304 /*
305 * If this fails, then we need to expand the pointer table. For now
306 * this situation is unhandled since it is hard to trigger.
307 */
308 assert(prefixlen < (unsigned int)l->l_bs);
309
310 return (prefixlen);
311 }
312
313 /*
314 * Initialize a fat ZAP leaf block.
315 */
316 static void
zap_fat_write_leaf_init(zap_leaf_t * l,uint64_t prefix,int prefixlen)317 zap_fat_write_leaf_init(zap_leaf_t *l, uint64_t prefix, int prefixlen)
318 {
319 zap_leaf_phys_t *leaf;
320
321 leaf = l->l_phys;
322
323 leaf->l_hdr.lh_block_type = ZBT_LEAF;
324 leaf->l_hdr.lh_magic = ZAP_LEAF_MAGIC;
325 leaf->l_hdr.lh_nfree = ZAP_LEAF_NUMCHUNKS(l);
326 leaf->l_hdr.lh_prefix = prefix;
327 leaf->l_hdr.lh_prefix_len = prefixlen;
328
329 /* Initialize the leaf hash table. */
330 assert(leaf->l_hdr.lh_nfree < 0xffff);
331 memset(leaf->l_hash, 0xff,
332 ZAP_LEAF_HASH_NUMENTRIES(l) * sizeof(*leaf->l_hash));
333
334 /* Initialize the leaf chunks. */
335 for (uint16_t i = 0; i < ZAP_LEAF_NUMCHUNKS(l); i++) {
336 struct zap_leaf_free *lf;
337
338 lf = &ZAP_LEAF_CHUNK(l, i).l_free;
339 lf->lf_type = ZAP_CHUNK_FREE;
340 if (i + 1 == ZAP_LEAF_NUMCHUNKS(l))
341 lf->lf_next = 0xffff;
342 else
343 lf->lf_next = i + 1;
344 }
345 }
346
347 static void
zap_fat_write(zfs_opt_t * zfs,zfs_zap_t * zap)348 zap_fat_write(zfs_opt_t *zfs, zfs_zap_t *zap)
349 {
350 struct dnode_cursor *c;
351 zap_leaf_t l;
352 zap_phys_t *zaphdr;
353 struct zap_table_phys *zt;
354 zfs_zap_entry_t *ent;
355 dnode_phys_t *dnode;
356 uint8_t *leafblks;
357 uint64_t lblkcnt, *ptrhasht;
358 off_t loc, blksz;
359 size_t blkshift;
360 unsigned int prefixlen;
361 int ptrcnt;
362
363 /*
364 * For simplicity, always use the largest block size. This should be ok
365 * since most directories will be micro ZAPs, but it's space inefficient
366 * for small ZAPs and might need to be revisited.
367 */
368 blkshift = MAXBLOCKSHIFT;
369 blksz = (off_t)1 << blkshift;
370
371 /*
372 * Embedded pointer tables give up to 8192 entries. This ought to be
373 * enough for anything except massive directories.
374 */
375 ptrcnt = (blksz / 2) / sizeof(uint64_t);
376
377 memset(zfs->filebuf, 0, sizeof(zfs->filebuf));
378 zaphdr = (zap_phys_t *)&zfs->filebuf[0];
379 zaphdr->zap_block_type = ZBT_HEADER;
380 zaphdr->zap_magic = ZAP_MAGIC;
381 zaphdr->zap_num_entries = zap->kvpcnt;
382 zaphdr->zap_salt = zap->hashsalt;
383
384 l.l_bs = blkshift;
385 l.l_phys = NULL;
386
387 zt = &zaphdr->zap_ptrtbl;
388 zt->zt_blk = 0;
389 zt->zt_numblks = 0;
390 zt->zt_shift = flsll(ptrcnt) - 1;
391 zt->zt_nextblk = 0;
392 zt->zt_blks_copied = 0;
393
394 /*
395 * How many leaf blocks do we need? Initialize them and update the
396 * header.
397 */
398 prefixlen = zap_fat_write_prefixlen(zap, &l);
399 lblkcnt = (uint64_t)1 << prefixlen;
400 leafblks = ecalloc(lblkcnt, blksz);
401 for (unsigned int li = 0; li < lblkcnt; li++) {
402 l.l_phys = (zap_leaf_phys_t *)(leafblks + li * blksz);
403 zap_fat_write_leaf_init(&l, li, prefixlen);
404 }
405 zaphdr->zap_num_leafs = lblkcnt;
406 zaphdr->zap_freeblk = lblkcnt + 1;
407
408 /*
409 * For each entry, figure out which leaf block it belongs to based on
410 * the upper bits of its hash, allocate chunks from that leaf, and fill
411 * them out.
412 */
413 ptrhasht = (uint64_t *)(&zfs->filebuf[0] + blksz / 2);
414 STAILQ_FOREACH(ent, &zap->kvps, next) {
415 struct zap_leaf_entry *le;
416 uint16_t *lptr;
417 uint64_t hi, li;
418 uint16_t namelen, nchunks, nnamechunks, nvalchunks;
419
420 hi = ZAP_HASH_IDX(ent->hash, zt->zt_shift);
421 li = ZAP_HASH_IDX(ent->hash, prefixlen);
422 assert(ptrhasht[hi] == 0 || ptrhasht[hi] == li + 1);
423 ptrhasht[hi] = li + 1;
424 l.l_phys = (zap_leaf_phys_t *)(leafblks + li * blksz);
425
426 namelen = strlen(ent->name) + 1;
427
428 /*
429 * How many leaf chunks do we need for this entry?
430 */
431 nnamechunks = howmany(namelen, ZAP_LEAF_ARRAY_BYTES);
432 nvalchunks = howmany(ent->intcnt,
433 ZAP_LEAF_ARRAY_BYTES / ent->intsz);
434 nchunks = 1 + nnamechunks + nvalchunks;
435
436 /*
437 * Allocate a run of free leaf chunks for this entry,
438 * potentially extending a hash chain.
439 */
440 assert(l.l_phys->l_hdr.lh_nfree >= nchunks);
441 l.l_phys->l_hdr.lh_nfree -= nchunks;
442 l.l_phys->l_hdr.lh_nentries++;
443 lptr = ZAP_LEAF_HASH_ENTPTR(&l, ent->hash);
444 while (*lptr != 0xffff) {
445 assert(*lptr < ZAP_LEAF_NUMCHUNKS(&l));
446 le = ZAP_LEAF_ENTRY(&l, *lptr);
447 assert(le->le_type == ZAP_CHUNK_ENTRY);
448 le->le_cd++;
449 lptr = &le->le_next;
450 }
451 *lptr = l.l_phys->l_hdr.lh_freelist;
452 l.l_phys->l_hdr.lh_freelist += nchunks;
453 assert(l.l_phys->l_hdr.lh_freelist <=
454 ZAP_LEAF_NUMCHUNKS(&l));
455 if (l.l_phys->l_hdr.lh_freelist ==
456 ZAP_LEAF_NUMCHUNKS(&l))
457 l.l_phys->l_hdr.lh_freelist = 0xffff;
458
459 /*
460 * Integer values must be stored in big-endian format.
461 */
462 switch (ent->intsz) {
463 case 1:
464 break;
465 case 2:
466 for (uint16_t *v = ent->val16p;
467 v - ent->val16p < (ptrdiff_t)ent->intcnt;
468 v++)
469 *v = htobe16(*v);
470 break;
471 case 4:
472 for (uint32_t *v = ent->val32p;
473 v - ent->val32p < (ptrdiff_t)ent->intcnt;
474 v++)
475 *v = htobe32(*v);
476 break;
477 case 8:
478 for (uint64_t *v = ent->val64p;
479 v - ent->val64p < (ptrdiff_t)ent->intcnt;
480 v++)
481 *v = htobe64(*v);
482 break;
483 default:
484 assert(0);
485 }
486
487 /*
488 * Finally, write out the leaf chunks for this entry.
489 */
490 le = ZAP_LEAF_ENTRY(&l, *lptr);
491 assert(le->le_type == ZAP_CHUNK_FREE);
492 le->le_type = ZAP_CHUNK_ENTRY;
493 le->le_next = 0xffff;
494 le->le_name_chunk = *lptr + 1;
495 le->le_name_numints = namelen;
496 le->le_value_chunk = *lptr + 1 + nnamechunks;
497 le->le_value_intlen = ent->intsz;
498 le->le_value_numints = ent->intcnt;
499 le->le_hash = ent->hash;
500 zap_fat_write_array_chunk(&l, *lptr + 1, namelen, ent->name);
501 zap_fat_write_array_chunk(&l, *lptr + 1 + nnamechunks,
502 ent->intcnt * ent->intsz, ent->valp);
503 }
504
505 /*
506 * Initialize unused slots of the pointer table.
507 */
508 for (int i = 0; i < ptrcnt; i++)
509 if (ptrhasht[i] == 0)
510 ptrhasht[i] = (i >> (zt->zt_shift - prefixlen)) + 1;
511
512 /*
513 * Write the whole thing to disk.
514 */
515 dnode = zap->dnode;
516 dnode->dn_datablkszsec = blksz >> MINBLOCKSHIFT;
517 dnode->dn_maxblkid = lblkcnt + 1;
518
519 c = dnode_cursor_init(zfs, zap->os, zap->dnode,
520 (lblkcnt + 1) * blksz, blksz);
521
522 loc = objset_space_alloc(zfs, zap->os, &blksz);
523 vdev_pwrite_dnode_indir(zfs, dnode, 0, 1, zfs->filebuf, blksz, loc,
524 dnode_cursor_next(zfs, c, 0));
525
526 for (uint64_t i = 0; i < lblkcnt; i++) {
527 loc = objset_space_alloc(zfs, zap->os, &blksz);
528 vdev_pwrite_dnode_indir(zfs, dnode, 0, 1, leafblks + i * blksz,
529 blksz, loc, dnode_cursor_next(zfs, c, (i + 1) * blksz));
530 }
531
532 dnode_cursor_finish(zfs, c);
533
534 free(leafblks);
535 }
536
537 void
zap_write(zfs_opt_t * zfs,zfs_zap_t * zap)538 zap_write(zfs_opt_t *zfs, zfs_zap_t *zap)
539 {
540 zfs_zap_entry_t *ent;
541
542 if (zap->micro) {
543 zap_micro_write(zfs, zap);
544 } else {
545 assert(!STAILQ_EMPTY(&zap->kvps));
546 assert(zap->kvpcnt > 0);
547 zap_fat_write(zfs, zap);
548 }
549
550 while ((ent = STAILQ_FIRST(&zap->kvps)) != NULL) {
551 STAILQ_REMOVE_HEAD(&zap->kvps, next);
552 if (ent->val64p != &ent->val64)
553 free(ent->valp);
554 free(ent->name);
555 free(ent);
556 }
557 free(zap);
558 }
559