xref: /titanic_41/usr/src/uts/common/fs/zfs/zap_micro.c (revision 7c64d3750da7fda7e450b8f9b0b963905ded6379)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/spa.h>
29 #include <sys/dmu.h>
30 #include <sys/zfs_context.h>
31 #include <sys/zap.h>
32 #include <sys/refcount.h>
33 #include <sys/zap_impl.h>
34 #include <sys/zap_leaf.h>
35 #include <sys/avl.h>
36 
37 #ifdef _KERNEL
38 #include <sys/sunddi.h>
39 #endif
40 
41 static int mzap_upgrade(zap_t **zapp, dmu_tx_t *tx);
42 
43 
44 static uint64_t
45 zap_hash(zap_t *zap, const char *normname)
46 {
47 	const uint8_t *cp;
48 	uint8_t c;
49 	uint64_t crc = zap->zap_salt;
50 
51 	/* NB: name must already be normalized, if necessary */
52 
53 	ASSERT(crc != 0);
54 	ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
55 	for (cp = (const uint8_t *)normname; (c = *cp) != '\0'; cp++) {
56 		crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ c) & 0xFF];
57 	}
58 
59 	/*
60 	 * Only use 28 bits, since we need 4 bits in the cookie for the
61 	 * collision differentiator.  We MUST use the high bits, since
62 	 * those are the ones that we first pay attention to when
63 	 * chosing the bucket.
64 	 */
65 	crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
66 
67 	return (crc);
68 }
69 
70 static int
71 zap_normalize(zap_t *zap, const char *name, char *namenorm)
72 {
73 	size_t inlen, outlen;
74 	int err;
75 
76 	inlen = strlen(name) + 1;
77 	outlen = ZAP_MAXNAMELEN;
78 
79 	err = 0;
80 	(void) u8_textprep_str((char *)name, &inlen, namenorm, &outlen,
81 	    zap->zap_normflags | U8_TEXTPREP_IGNORE_NULL, U8_UNICODE_LATEST,
82 	    &err);
83 
84 	return (err);
85 }
86 
87 boolean_t
88 zap_match(zap_name_t *zn, const char *matchname)
89 {
90 	if (zn->zn_matchtype == MT_FIRST) {
91 		char norm[ZAP_MAXNAMELEN];
92 
93 		if (zap_normalize(zn->zn_zap, matchname, norm) != 0)
94 			return (B_FALSE);
95 
96 		return (strcmp(zn->zn_name_norm, norm) == 0);
97 	} else {
98 		/* MT_BEST or MT_EXACT */
99 		return (strcmp(zn->zn_name_orij, matchname) == 0);
100 	}
101 }
102 
103 void
104 zap_name_free(zap_name_t *zn)
105 {
106 	kmem_free(zn, sizeof (zap_name_t));
107 }
108 
109 /* XXX combine this with zap_lockdir()? */
110 zap_name_t *
111 zap_name_alloc(zap_t *zap, const char *name, matchtype_t mt)
112 {
113 	zap_name_t *zn = kmem_alloc(sizeof (zap_name_t), KM_SLEEP);
114 
115 	zn->zn_zap = zap;
116 	zn->zn_name_orij = name;
117 	zn->zn_matchtype = mt;
118 	if (zap->zap_normflags) {
119 		if (zap_normalize(zap, name, zn->zn_normbuf) != 0) {
120 			zap_name_free(zn);
121 			return (NULL);
122 		}
123 		zn->zn_name_norm = zn->zn_normbuf;
124 	} else {
125 		if (mt != MT_EXACT) {
126 			zap_name_free(zn);
127 			return (NULL);
128 		}
129 		zn->zn_name_norm = zn->zn_name_orij;
130 	}
131 
132 	zn->zn_hash = zap_hash(zap, zn->zn_name_norm);
133 	return (zn);
134 }
135 
136 static void
137 mzap_byteswap(mzap_phys_t *buf, size_t size)
138 {
139 	int i, max;
140 	buf->mz_block_type = BSWAP_64(buf->mz_block_type);
141 	buf->mz_salt = BSWAP_64(buf->mz_salt);
142 	max = (size / MZAP_ENT_LEN) - 1;
143 	for (i = 0; i < max; i++) {
144 		buf->mz_chunk[i].mze_value =
145 		    BSWAP_64(buf->mz_chunk[i].mze_value);
146 		buf->mz_chunk[i].mze_cd =
147 		    BSWAP_32(buf->mz_chunk[i].mze_cd);
148 	}
149 }
150 
151 void
152 zap_byteswap(void *buf, size_t size)
153 {
154 	uint64_t block_type;
155 
156 	block_type = *(uint64_t *)buf;
157 
158 	if (block_type == ZBT_MICRO || block_type == BSWAP_64(ZBT_MICRO)) {
159 		/* ASSERT(magic == ZAP_LEAF_MAGIC); */
160 		mzap_byteswap(buf, size);
161 	} else {
162 		fzap_byteswap(buf, size);
163 	}
164 }
165 
166 static int
167 mze_compare(const void *arg1, const void *arg2)
168 {
169 	const mzap_ent_t *mze1 = arg1;
170 	const mzap_ent_t *mze2 = arg2;
171 
172 	if (mze1->mze_hash > mze2->mze_hash)
173 		return (+1);
174 	if (mze1->mze_hash < mze2->mze_hash)
175 		return (-1);
176 	if (mze1->mze_phys.mze_cd > mze2->mze_phys.mze_cd)
177 		return (+1);
178 	if (mze1->mze_phys.mze_cd < mze2->mze_phys.mze_cd)
179 		return (-1);
180 	return (0);
181 }
182 
183 static void
184 mze_insert(zap_t *zap, int chunkid, uint64_t hash, mzap_ent_phys_t *mzep)
185 {
186 	mzap_ent_t *mze;
187 
188 	ASSERT(zap->zap_ismicro);
189 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
190 	ASSERT(mzep->mze_cd < ZAP_MAXCD);
191 
192 	mze = kmem_alloc(sizeof (mzap_ent_t), KM_SLEEP);
193 	mze->mze_chunkid = chunkid;
194 	mze->mze_hash = hash;
195 	mze->mze_phys = *mzep;
196 	avl_add(&zap->zap_m.zap_avl, mze);
197 }
198 
199 static mzap_ent_t *
200 mze_find(zap_name_t *zn)
201 {
202 	mzap_ent_t mze_tofind;
203 	mzap_ent_t *mze;
204 	avl_index_t idx;
205 	avl_tree_t *avl = &zn->zn_zap->zap_m.zap_avl;
206 
207 	ASSERT(zn->zn_zap->zap_ismicro);
208 	ASSERT(RW_LOCK_HELD(&zn->zn_zap->zap_rwlock));
209 
210 	if (strlen(zn->zn_name_norm) >= sizeof (mze_tofind.mze_phys.mze_name))
211 		return (NULL);
212 
213 	mze_tofind.mze_hash = zn->zn_hash;
214 	mze_tofind.mze_phys.mze_cd = 0;
215 
216 again:
217 	mze = avl_find(avl, &mze_tofind, &idx);
218 	if (mze == NULL)
219 		mze = avl_nearest(avl, idx, AVL_AFTER);
220 	for (; mze && mze->mze_hash == zn->zn_hash; mze = AVL_NEXT(avl, mze)) {
221 		if (zap_match(zn, mze->mze_phys.mze_name))
222 			return (mze);
223 	}
224 	if (zn->zn_matchtype == MT_BEST) {
225 		zn->zn_matchtype = MT_FIRST;
226 		goto again;
227 	}
228 	return (NULL);
229 }
230 
231 static uint32_t
232 mze_find_unused_cd(zap_t *zap, uint64_t hash)
233 {
234 	mzap_ent_t mze_tofind;
235 	mzap_ent_t *mze;
236 	avl_index_t idx;
237 	avl_tree_t *avl = &zap->zap_m.zap_avl;
238 	uint32_t cd;
239 
240 	ASSERT(zap->zap_ismicro);
241 	ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
242 
243 	mze_tofind.mze_hash = hash;
244 	mze_tofind.mze_phys.mze_cd = 0;
245 
246 	cd = 0;
247 	for (mze = avl_find(avl, &mze_tofind, &idx);
248 	    mze && mze->mze_hash == hash; mze = AVL_NEXT(avl, mze)) {
249 		if (mze->mze_phys.mze_cd != cd)
250 			break;
251 		cd++;
252 	}
253 
254 	return (cd);
255 }
256 
257 static void
258 mze_remove(zap_t *zap, mzap_ent_t *mze)
259 {
260 	ASSERT(zap->zap_ismicro);
261 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
262 
263 	avl_remove(&zap->zap_m.zap_avl, mze);
264 	kmem_free(mze, sizeof (mzap_ent_t));
265 }
266 
267 static void
268 mze_destroy(zap_t *zap)
269 {
270 	mzap_ent_t *mze;
271 	void *avlcookie = NULL;
272 
273 	while (mze = avl_destroy_nodes(&zap->zap_m.zap_avl, &avlcookie))
274 		kmem_free(mze, sizeof (mzap_ent_t));
275 	avl_destroy(&zap->zap_m.zap_avl);
276 }
277 
278 static zap_t *
279 mzap_open(objset_t *os, uint64_t obj, dmu_buf_t *db)
280 {
281 	zap_t *winner;
282 	zap_t *zap;
283 	int i;
284 
285 	ASSERT3U(MZAP_ENT_LEN, ==, sizeof (mzap_ent_phys_t));
286 
287 	zap = kmem_zalloc(sizeof (zap_t), KM_SLEEP);
288 	rw_init(&zap->zap_rwlock, 0, 0, 0);
289 	rw_enter(&zap->zap_rwlock, RW_WRITER);
290 	zap->zap_objset = os;
291 	zap->zap_object = obj;
292 	zap->zap_dbuf = db;
293 
294 	if (*(uint64_t *)db->db_data != ZBT_MICRO) {
295 		mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
296 		zap->zap_f.zap_block_shift = highbit(db->db_size) - 1;
297 	} else {
298 		zap->zap_ismicro = TRUE;
299 	}
300 
301 	/*
302 	 * Make sure that zap_ismicro is set before we let others see
303 	 * it, because zap_lockdir() checks zap_ismicro without the lock
304 	 * held.
305 	 */
306 	winner = dmu_buf_set_user(db, zap, &zap->zap_m.zap_phys, zap_evict);
307 
308 	if (winner != NULL) {
309 		rw_exit(&zap->zap_rwlock);
310 		rw_destroy(&zap->zap_rwlock);
311 		if (!zap->zap_ismicro)
312 			mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
313 		kmem_free(zap, sizeof (zap_t));
314 		return (winner);
315 	}
316 
317 	if (zap->zap_ismicro) {
318 		zap->zap_salt = zap->zap_m.zap_phys->mz_salt;
319 		zap->zap_normflags = zap->zap_m.zap_phys->mz_normflags;
320 		zap->zap_m.zap_num_chunks = db->db_size / MZAP_ENT_LEN - 1;
321 		avl_create(&zap->zap_m.zap_avl, mze_compare,
322 		    sizeof (mzap_ent_t), offsetof(mzap_ent_t, mze_node));
323 
324 		for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
325 			mzap_ent_phys_t *mze =
326 			    &zap->zap_m.zap_phys->mz_chunk[i];
327 			if (mze->mze_name[0]) {
328 				zap_name_t *zn;
329 
330 				zap->zap_m.zap_num_entries++;
331 				zn = zap_name_alloc(zap, mze->mze_name,
332 				    MT_EXACT);
333 				mze_insert(zap, i, zn->zn_hash, mze);
334 				zap_name_free(zn);
335 			}
336 		}
337 	} else {
338 		zap->zap_salt = zap->zap_f.zap_phys->zap_salt;
339 		zap->zap_normflags = zap->zap_f.zap_phys->zap_normflags;
340 
341 		ASSERT3U(sizeof (struct zap_leaf_header), ==,
342 		    2*ZAP_LEAF_CHUNKSIZE);
343 
344 		/*
345 		 * The embedded pointer table should not overlap the
346 		 * other members.
347 		 */
348 		ASSERT3P(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0), >,
349 		    &zap->zap_f.zap_phys->zap_salt);
350 
351 		/*
352 		 * The embedded pointer table should end at the end of
353 		 * the block
354 		 */
355 		ASSERT3U((uintptr_t)&ZAP_EMBEDDED_PTRTBL_ENT(zap,
356 		    1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)) -
357 		    (uintptr_t)zap->zap_f.zap_phys, ==,
358 		    zap->zap_dbuf->db_size);
359 	}
360 	rw_exit(&zap->zap_rwlock);
361 	return (zap);
362 }
363 
364 int
365 zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
366     krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp)
367 {
368 	zap_t *zap;
369 	dmu_buf_t *db;
370 	krw_t lt;
371 	int err;
372 
373 	*zapp = NULL;
374 
375 	err = dmu_buf_hold(os, obj, 0, NULL, &db);
376 	if (err)
377 		return (err);
378 
379 #ifdef ZFS_DEBUG
380 	{
381 		dmu_object_info_t doi;
382 		dmu_object_info_from_db(db, &doi);
383 		ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
384 	}
385 #endif
386 
387 	zap = dmu_buf_get_user(db);
388 	if (zap == NULL)
389 		zap = mzap_open(os, obj, db);
390 
391 	/*
392 	 * We're checking zap_ismicro without the lock held, in order to
393 	 * tell what type of lock we want.  Once we have some sort of
394 	 * lock, see if it really is the right type.  In practice this
395 	 * can only be different if it was upgraded from micro to fat,
396 	 * and micro wanted WRITER but fat only needs READER.
397 	 */
398 	lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
399 	rw_enter(&zap->zap_rwlock, lt);
400 	if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
401 		/* it was upgraded, now we only need reader */
402 		ASSERT(lt == RW_WRITER);
403 		ASSERT(RW_READER ==
404 		    (!zap->zap_ismicro && fatreader) ? RW_READER : lti);
405 		rw_downgrade(&zap->zap_rwlock);
406 		lt = RW_READER;
407 	}
408 
409 	zap->zap_objset = os;
410 
411 	if (lt == RW_WRITER)
412 		dmu_buf_will_dirty(db, tx);
413 
414 	ASSERT3P(zap->zap_dbuf, ==, db);
415 
416 	ASSERT(!zap->zap_ismicro ||
417 	    zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
418 	if (zap->zap_ismicro && tx && adding &&
419 	    zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
420 		uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
421 		if (newsz > MZAP_MAX_BLKSZ) {
422 			dprintf("upgrading obj %llu: num_entries=%u\n",
423 			    obj, zap->zap_m.zap_num_entries);
424 			*zapp = zap;
425 			return (mzap_upgrade(zapp, tx));
426 		}
427 		err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
428 		ASSERT3U(err, ==, 0);
429 		zap->zap_m.zap_num_chunks =
430 		    db->db_size / MZAP_ENT_LEN - 1;
431 	}
432 
433 	*zapp = zap;
434 	return (0);
435 }
436 
437 void
438 zap_unlockdir(zap_t *zap)
439 {
440 	rw_exit(&zap->zap_rwlock);
441 	dmu_buf_rele(zap->zap_dbuf, NULL);
442 }
443 
444 static int
445 mzap_upgrade(zap_t **zapp, dmu_tx_t *tx)
446 {
447 	mzap_phys_t *mzp;
448 	int i, sz, nchunks, err;
449 	zap_t *zap = *zapp;
450 
451 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
452 
453 	sz = zap->zap_dbuf->db_size;
454 	mzp = kmem_alloc(sz, KM_SLEEP);
455 	bcopy(zap->zap_dbuf->db_data, mzp, sz);
456 	nchunks = zap->zap_m.zap_num_chunks;
457 
458 	err = dmu_object_set_blocksize(zap->zap_objset, zap->zap_object,
459 	    1ULL << fzap_default_block_shift, 0, tx);
460 	if (err) {
461 		kmem_free(mzp, sz);
462 		return (err);
463 	}
464 
465 	dprintf("upgrading obj=%llu with %u chunks\n",
466 	    zap->zap_object, nchunks);
467 	/* XXX destroy the avl later, so we can use the stored hash value */
468 	mze_destroy(zap);
469 
470 	fzap_upgrade(zap, tx);
471 
472 	for (i = 0; i < nchunks; i++) {
473 		int err;
474 		mzap_ent_phys_t *mze = &mzp->mz_chunk[i];
475 		zap_name_t *zn;
476 		if (mze->mze_name[0] == 0)
477 			continue;
478 		dprintf("adding %s=%llu\n",
479 		    mze->mze_name, mze->mze_value);
480 		zn = zap_name_alloc(zap, mze->mze_name, MT_EXACT);
481 		err = fzap_add_cd(zn, 8, 1, &mze->mze_value, mze->mze_cd, tx);
482 		zap = zn->zn_zap;	/* fzap_add_cd() may change zap */
483 		zap_name_free(zn);
484 		if (err)
485 			break;
486 	}
487 	kmem_free(mzp, sz);
488 	*zapp = zap;
489 	return (err);
490 }
491 
492 static void
493 mzap_create_impl(objset_t *os, uint64_t obj, int normflags, dmu_tx_t *tx)
494 {
495 	dmu_buf_t *db;
496 	mzap_phys_t *zp;
497 
498 	VERIFY(0 == dmu_buf_hold(os, obj, 0, FTAG, &db));
499 
500 #ifdef ZFS_DEBUG
501 	{
502 		dmu_object_info_t doi;
503 		dmu_object_info_from_db(db, &doi);
504 		ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
505 	}
506 #endif
507 
508 	dmu_buf_will_dirty(db, tx);
509 	zp = db->db_data;
510 	zp->mz_block_type = ZBT_MICRO;
511 	zp->mz_salt = ((uintptr_t)db ^ (uintptr_t)tx ^ (obj << 1)) | 1ULL;
512 	zp->mz_normflags = normflags;
513 	dmu_buf_rele(db, FTAG);
514 }
515 
516 int
517 zap_create_claim(objset_t *os, uint64_t obj, dmu_object_type_t ot,
518     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
519 {
520 	return (zap_create_claim_norm(os, obj,
521 	    0, ot, bonustype, bonuslen, tx));
522 }
523 
524 int
525 zap_create_claim_norm(objset_t *os, uint64_t obj, int normflags,
526     dmu_object_type_t ot,
527     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
528 {
529 	int err;
530 
531 	err = dmu_object_claim(os, obj, ot, 0, bonustype, bonuslen, tx);
532 	if (err != 0)
533 		return (err);
534 	mzap_create_impl(os, obj, normflags, tx);
535 	return (0);
536 }
537 
538 uint64_t
539 zap_create(objset_t *os, dmu_object_type_t ot,
540     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
541 {
542 	return (zap_create_norm(os, 0, ot, bonustype, bonuslen, tx));
543 }
544 
545 uint64_t
546 zap_create_norm(objset_t *os, int normflags, dmu_object_type_t ot,
547     dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
548 {
549 	uint64_t obj = dmu_object_alloc(os, ot, 0, bonustype, bonuslen, tx);
550 
551 	mzap_create_impl(os, obj, normflags, tx);
552 	return (obj);
553 }
554 
555 int
556 zap_destroy(objset_t *os, uint64_t zapobj, dmu_tx_t *tx)
557 {
558 	/*
559 	 * dmu_object_free will free the object number and free the
560 	 * data.  Freeing the data will cause our pageout function to be
561 	 * called, which will destroy our data (zap_leaf_t's and zap_t).
562 	 */
563 
564 	return (dmu_object_free(os, zapobj, tx));
565 }
566 
567 _NOTE(ARGSUSED(0))
568 void
569 zap_evict(dmu_buf_t *db, void *vzap)
570 {
571 	zap_t *zap = vzap;
572 
573 	rw_destroy(&zap->zap_rwlock);
574 
575 	if (zap->zap_ismicro)
576 		mze_destroy(zap);
577 	else
578 		mutex_destroy(&zap->zap_f.zap_num_entries_mtx);
579 
580 	kmem_free(zap, sizeof (zap_t));
581 }
582 
583 int
584 zap_count(objset_t *os, uint64_t zapobj, uint64_t *count)
585 {
586 	zap_t *zap;
587 	int err;
588 
589 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
590 	if (err)
591 		return (err);
592 	if (!zap->zap_ismicro) {
593 		err = fzap_count(zap, count);
594 	} else {
595 		*count = zap->zap_m.zap_num_entries;
596 	}
597 	zap_unlockdir(zap);
598 	return (err);
599 }
600 
601 /*
602  * zn may be NULL; if not specified, it will be computed if needed.
603  * See also the comment above zap_entry_normalization_conflict().
604  */
605 static boolean_t
606 mzap_normalization_conflict(zap_t *zap, zap_name_t *zn, mzap_ent_t *mze)
607 {
608 	mzap_ent_t *other;
609 	int direction = AVL_BEFORE;
610 	boolean_t allocdzn = B_FALSE;
611 
612 	if (zap->zap_normflags == 0)
613 		return (B_FALSE);
614 
615 again:
616 	for (other = avl_walk(&zap->zap_m.zap_avl, mze, direction);
617 	    other && other->mze_hash == mze->mze_hash;
618 	    other = avl_walk(&zap->zap_m.zap_avl, other, direction)) {
619 
620 		if (zn == NULL) {
621 			zn = zap_name_alloc(zap, mze->mze_phys.mze_name,
622 			    MT_FIRST);
623 			allocdzn = B_TRUE;
624 		}
625 		if (zap_match(zn, other->mze_phys.mze_name)) {
626 			if (allocdzn)
627 				zap_name_free(zn);
628 			return (B_TRUE);
629 		}
630 	}
631 
632 	if (direction == AVL_BEFORE) {
633 		direction = AVL_AFTER;
634 		goto again;
635 	}
636 
637 	if (allocdzn)
638 		zap_name_free(zn);
639 	return (B_FALSE);
640 }
641 
642 /*
643  * Routines for manipulating attributes.
644  */
645 
646 int
647 zap_lookup(objset_t *os, uint64_t zapobj, const char *name,
648     uint64_t integer_size, uint64_t num_integers, void *buf)
649 {
650 	return (zap_lookup_norm(os, zapobj, name, integer_size,
651 	    num_integers, buf, MT_EXACT, NULL, 0, NULL));
652 }
653 
654 int
655 zap_lookup_norm(objset_t *os, uint64_t zapobj, const char *name,
656     uint64_t integer_size, uint64_t num_integers, void *buf,
657     matchtype_t mt, char *realname, int rn_len,
658     boolean_t *ncp)
659 {
660 	zap_t *zap;
661 	int err;
662 	mzap_ent_t *mze;
663 	zap_name_t *zn;
664 
665 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
666 	if (err)
667 		return (err);
668 	zn = zap_name_alloc(zap, name, mt);
669 	if (zn == NULL) {
670 		zap_unlockdir(zap);
671 		return (ENOTSUP);
672 	}
673 
674 	if (!zap->zap_ismicro) {
675 		err = fzap_lookup(zn, integer_size, num_integers, buf,
676 		    realname, rn_len, ncp);
677 	} else {
678 		mze = mze_find(zn);
679 		if (mze == NULL) {
680 			err = ENOENT;
681 		} else {
682 			if (num_integers < 1) {
683 				err = EOVERFLOW;
684 			} else if (integer_size != 8) {
685 				err = EINVAL;
686 			} else {
687 				*(uint64_t *)buf = mze->mze_phys.mze_value;
688 				(void) strlcpy(realname,
689 				    mze->mze_phys.mze_name, rn_len);
690 				if (ncp) {
691 					*ncp = mzap_normalization_conflict(zap,
692 					    zn, mze);
693 				}
694 			}
695 		}
696 	}
697 	zap_name_free(zn);
698 	zap_unlockdir(zap);
699 	return (err);
700 }
701 
702 int
703 zap_length(objset_t *os, uint64_t zapobj, const char *name,
704     uint64_t *integer_size, uint64_t *num_integers)
705 {
706 	zap_t *zap;
707 	int err;
708 	mzap_ent_t *mze;
709 	zap_name_t *zn;
710 
711 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
712 	if (err)
713 		return (err);
714 	zn = zap_name_alloc(zap, name, MT_EXACT);
715 	if (zn == NULL) {
716 		zap_unlockdir(zap);
717 		return (ENOTSUP);
718 	}
719 	if (!zap->zap_ismicro) {
720 		err = fzap_length(zn, integer_size, num_integers);
721 	} else {
722 		mze = mze_find(zn);
723 		if (mze == NULL) {
724 			err = ENOENT;
725 		} else {
726 			if (integer_size)
727 				*integer_size = 8;
728 			if (num_integers)
729 				*num_integers = 1;
730 		}
731 	}
732 	zap_name_free(zn);
733 	zap_unlockdir(zap);
734 	return (err);
735 }
736 
737 static void
738 mzap_addent(zap_name_t *zn, uint64_t value)
739 {
740 	int i;
741 	zap_t *zap = zn->zn_zap;
742 	int start = zap->zap_m.zap_alloc_next;
743 	uint32_t cd;
744 
745 	dprintf("obj=%llu %s=%llu\n", zap->zap_object,
746 	    zn->zn_name_orij, value);
747 	ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
748 
749 #ifdef ZFS_DEBUG
750 	for (i = 0; i < zap->zap_m.zap_num_chunks; i++) {
751 		mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
752 		ASSERT(strcmp(zn->zn_name_orij, mze->mze_name) != 0);
753 	}
754 #endif
755 
756 	cd = mze_find_unused_cd(zap, zn->zn_hash);
757 	/* given the limited size of the microzap, this can't happen */
758 	ASSERT(cd != ZAP_MAXCD);
759 
760 again:
761 	for (i = start; i < zap->zap_m.zap_num_chunks; i++) {
762 		mzap_ent_phys_t *mze = &zap->zap_m.zap_phys->mz_chunk[i];
763 		if (mze->mze_name[0] == 0) {
764 			mze->mze_value = value;
765 			mze->mze_cd = cd;
766 			(void) strcpy(mze->mze_name, zn->zn_name_orij);
767 			zap->zap_m.zap_num_entries++;
768 			zap->zap_m.zap_alloc_next = i+1;
769 			if (zap->zap_m.zap_alloc_next ==
770 			    zap->zap_m.zap_num_chunks)
771 				zap->zap_m.zap_alloc_next = 0;
772 			mze_insert(zap, i, zn->zn_hash, mze);
773 			return;
774 		}
775 	}
776 	if (start != 0) {
777 		start = 0;
778 		goto again;
779 	}
780 	ASSERT(!"out of entries!");
781 }
782 
783 int
784 zap_add(objset_t *os, uint64_t zapobj, const char *name,
785     int integer_size, uint64_t num_integers,
786     const void *val, dmu_tx_t *tx)
787 {
788 	zap_t *zap;
789 	int err;
790 	mzap_ent_t *mze;
791 	const uint64_t *intval = val;
792 	zap_name_t *zn;
793 
794 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
795 	if (err)
796 		return (err);
797 	zn = zap_name_alloc(zap, name, MT_EXACT);
798 	if (zn == NULL) {
799 		zap_unlockdir(zap);
800 		return (ENOTSUP);
801 	}
802 	if (!zap->zap_ismicro) {
803 		err = fzap_add(zn, integer_size, num_integers, val, tx);
804 		zap = zn->zn_zap;	/* fzap_add() may change zap */
805 	} else if (integer_size != 8 || num_integers != 1 ||
806 	    strlen(name) >= MZAP_NAME_LEN) {
807 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
808 		    zapobj, integer_size, num_integers, name);
809 		err = mzap_upgrade(&zn->zn_zap, tx);
810 		if (err == 0)
811 			err = fzap_add(zn, integer_size, num_integers, val, tx);
812 		zap = zn->zn_zap;	/* fzap_add() may change zap */
813 	} else {
814 		mze = mze_find(zn);
815 		if (mze != NULL) {
816 			err = EEXIST;
817 		} else {
818 			mzap_addent(zn, *intval);
819 		}
820 	}
821 	ASSERT(zap == zn->zn_zap);
822 	zap_name_free(zn);
823 	if (zap != NULL)	/* may be NULL if fzap_add() failed */
824 		zap_unlockdir(zap);
825 	return (err);
826 }
827 
828 int
829 zap_update(objset_t *os, uint64_t zapobj, const char *name,
830     int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
831 {
832 	zap_t *zap;
833 	mzap_ent_t *mze;
834 	const uint64_t *intval = val;
835 	zap_name_t *zn;
836 	int err;
837 
838 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, TRUE, &zap);
839 	if (err)
840 		return (err);
841 	zn = zap_name_alloc(zap, name, MT_EXACT);
842 	if (zn == NULL) {
843 		zap_unlockdir(zap);
844 		return (ENOTSUP);
845 	}
846 	if (!zap->zap_ismicro) {
847 		err = fzap_update(zn, integer_size, num_integers, val, tx);
848 		zap = zn->zn_zap;	/* fzap_update() may change zap */
849 	} else if (integer_size != 8 || num_integers != 1 ||
850 	    strlen(name) >= MZAP_NAME_LEN) {
851 		dprintf("upgrading obj %llu: intsz=%u numint=%llu name=%s\n",
852 		    zapobj, integer_size, num_integers, name);
853 		err = mzap_upgrade(&zn->zn_zap, tx);
854 		if (err == 0)
855 			err = fzap_update(zn, integer_size, num_integers,
856 			    val, tx);
857 		zap = zn->zn_zap;	/* fzap_update() may change zap */
858 	} else {
859 		mze = mze_find(zn);
860 		if (mze != NULL) {
861 			mze->mze_phys.mze_value = *intval;
862 			zap->zap_m.zap_phys->mz_chunk
863 			    [mze->mze_chunkid].mze_value = *intval;
864 		} else {
865 			mzap_addent(zn, *intval);
866 		}
867 	}
868 	ASSERT(zap == zn->zn_zap);
869 	zap_name_free(zn);
870 	if (zap != NULL)	/* may be NULL if fzap_upgrade() failed */
871 		zap_unlockdir(zap);
872 	return (err);
873 }
874 
875 int
876 zap_remove(objset_t *os, uint64_t zapobj, const char *name, dmu_tx_t *tx)
877 {
878 	return (zap_remove_norm(os, zapobj, name, MT_EXACT, tx));
879 }
880 
881 int
882 zap_remove_norm(objset_t *os, uint64_t zapobj, const char *name,
883     matchtype_t mt, dmu_tx_t *tx)
884 {
885 	zap_t *zap;
886 	int err;
887 	mzap_ent_t *mze;
888 	zap_name_t *zn;
889 
890 	err = zap_lockdir(os, zapobj, tx, RW_WRITER, TRUE, FALSE, &zap);
891 	if (err)
892 		return (err);
893 	zn = zap_name_alloc(zap, name, mt);
894 	if (zn == NULL) {
895 		zap_unlockdir(zap);
896 		return (ENOTSUP);
897 	}
898 	if (!zap->zap_ismicro) {
899 		err = fzap_remove(zn, tx);
900 	} else {
901 		mze = mze_find(zn);
902 		if (mze == NULL) {
903 			err = ENOENT;
904 		} else {
905 			zap->zap_m.zap_num_entries--;
906 			bzero(&zap->zap_m.zap_phys->mz_chunk[mze->mze_chunkid],
907 			    sizeof (mzap_ent_phys_t));
908 			mze_remove(zap, mze);
909 		}
910 	}
911 	zap_name_free(zn);
912 	zap_unlockdir(zap);
913 	return (err);
914 }
915 
916 /*
917  * Routines for iterating over the attributes.
918  */
919 
920 /*
921  * We want to keep the high 32 bits of the cursor zero if we can, so
922  * that 32-bit programs can access this.  So use a small hash value so
923  * we can fit 4 bits of cd into the 32-bit cursor.
924  *
925  * [ 4 zero bits | 32-bit collision differentiator | 28-bit hash value ]
926  */
927 void
928 zap_cursor_init_serialized(zap_cursor_t *zc, objset_t *os, uint64_t zapobj,
929     uint64_t serialized)
930 {
931 	zc->zc_objset = os;
932 	zc->zc_zap = NULL;
933 	zc->zc_leaf = NULL;
934 	zc->zc_zapobj = zapobj;
935 	if (serialized == -1ULL) {
936 		zc->zc_hash = -1ULL;
937 		zc->zc_cd = 0;
938 	} else {
939 		zc->zc_hash = serialized << (64-ZAP_HASHBITS);
940 		zc->zc_cd = serialized >> ZAP_HASHBITS;
941 		if (zc->zc_cd >= ZAP_MAXCD) /* corrupt serialized */
942 			zc->zc_cd = 0;
943 	}
944 }
945 
946 void
947 zap_cursor_init(zap_cursor_t *zc, objset_t *os, uint64_t zapobj)
948 {
949 	zap_cursor_init_serialized(zc, os, zapobj, 0);
950 }
951 
952 void
953 zap_cursor_fini(zap_cursor_t *zc)
954 {
955 	if (zc->zc_zap) {
956 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
957 		zap_unlockdir(zc->zc_zap);
958 		zc->zc_zap = NULL;
959 	}
960 	if (zc->zc_leaf) {
961 		rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
962 		zap_put_leaf(zc->zc_leaf);
963 		zc->zc_leaf = NULL;
964 	}
965 	zc->zc_objset = NULL;
966 }
967 
968 uint64_t
969 zap_cursor_serialize(zap_cursor_t *zc)
970 {
971 	if (zc->zc_hash == -1ULL)
972 		return (-1ULL);
973 	ASSERT((zc->zc_hash & (ZAP_MAXCD-1)) == 0);
974 	ASSERT(zc->zc_cd < ZAP_MAXCD);
975 	return ((zc->zc_hash >> (64-ZAP_HASHBITS)) |
976 	    ((uint64_t)zc->zc_cd << ZAP_HASHBITS));
977 }
978 
979 int
980 zap_cursor_retrieve(zap_cursor_t *zc, zap_attribute_t *za)
981 {
982 	int err;
983 	avl_index_t idx;
984 	mzap_ent_t mze_tofind;
985 	mzap_ent_t *mze;
986 
987 	if (zc->zc_hash == -1ULL)
988 		return (ENOENT);
989 
990 	if (zc->zc_zap == NULL) {
991 		err = zap_lockdir(zc->zc_objset, zc->zc_zapobj, NULL,
992 		    RW_READER, TRUE, FALSE, &zc->zc_zap);
993 		if (err)
994 			return (err);
995 	} else {
996 		rw_enter(&zc->zc_zap->zap_rwlock, RW_READER);
997 	}
998 	if (!zc->zc_zap->zap_ismicro) {
999 		err = fzap_cursor_retrieve(zc->zc_zap, zc, za);
1000 	} else {
1001 		err = ENOENT;
1002 
1003 		mze_tofind.mze_hash = zc->zc_hash;
1004 		mze_tofind.mze_phys.mze_cd = zc->zc_cd;
1005 
1006 		mze = avl_find(&zc->zc_zap->zap_m.zap_avl, &mze_tofind, &idx);
1007 		if (mze == NULL) {
1008 			mze = avl_nearest(&zc->zc_zap->zap_m.zap_avl,
1009 			    idx, AVL_AFTER);
1010 		}
1011 		if (mze) {
1012 			ASSERT(0 == bcmp(&mze->mze_phys,
1013 			    &zc->zc_zap->zap_m.zap_phys->mz_chunk
1014 			    [mze->mze_chunkid], sizeof (mze->mze_phys)));
1015 
1016 			za->za_normalization_conflict =
1017 			    mzap_normalization_conflict(zc->zc_zap, NULL, mze);
1018 			za->za_integer_length = 8;
1019 			za->za_num_integers = 1;
1020 			za->za_first_integer = mze->mze_phys.mze_value;
1021 			(void) strcpy(za->za_name, mze->mze_phys.mze_name);
1022 			zc->zc_hash = mze->mze_hash;
1023 			zc->zc_cd = mze->mze_phys.mze_cd;
1024 			err = 0;
1025 		} else {
1026 			zc->zc_hash = -1ULL;
1027 		}
1028 	}
1029 	rw_exit(&zc->zc_zap->zap_rwlock);
1030 	return (err);
1031 }
1032 
1033 void
1034 zap_cursor_advance(zap_cursor_t *zc)
1035 {
1036 	if (zc->zc_hash == -1ULL)
1037 		return;
1038 	zc->zc_cd++;
1039 	if (zc->zc_cd >= ZAP_MAXCD) {
1040 		zc->zc_cd = 0;
1041 		zc->zc_hash += 1ULL<<(64-ZAP_HASHBITS);
1042 		if (zc->zc_hash == 0) /* EOF */
1043 			zc->zc_hash = -1ULL;
1044 	}
1045 }
1046 
1047 int
1048 zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
1049 {
1050 	int err;
1051 	zap_t *zap;
1052 
1053 	err = zap_lockdir(os, zapobj, NULL, RW_READER, TRUE, FALSE, &zap);
1054 	if (err)
1055 		return (err);
1056 
1057 	bzero(zs, sizeof (zap_stats_t));
1058 
1059 	if (zap->zap_ismicro) {
1060 		zs->zs_blocksize = zap->zap_dbuf->db_size;
1061 		zs->zs_num_entries = zap->zap_m.zap_num_entries;
1062 		zs->zs_num_blocks = 1;
1063 	} else {
1064 		fzap_get_stats(zap, zs);
1065 	}
1066 	zap_unlockdir(zap);
1067 	return (0);
1068 }
1069