]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/zap.c
Undo c89 workarounds to match with upstream
[mirror_zfs.git] / module / zfs / zap.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 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 */
26
27 /*
28 * This file contains the top half of the zfs directory structure
29 * implementation. The bottom half is in zap_leaf.c.
30 *
31 * The zdir is an extendable hash data structure. There is a table of
32 * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are
33 * each a constant size and hold a variable number of directory entries.
34 * The buckets (aka "leaf nodes") are implemented in zap_leaf.c.
35 *
36 * The pointer table holds a power of 2 number of pointers.
37 * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to
38 * by the pointer at index i in the table holds entries whose hash value
39 * has a zd_prefix_len - bit prefix
40 */
41
42 #include <sys/spa.h>
43 #include <sys/dmu.h>
44 #include <sys/zfs_context.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/fs/zfs.h>
47 #include <sys/zap.h>
48 #include <sys/refcount.h>
49 #include <sys/zap_impl.h>
50 #include <sys/zap_leaf.h>
51
52 int fzap_default_block_shift = 14; /* 16k blocksize */
53
54 extern inline zap_phys_t *zap_f_phys(zap_t *zap);
55
56 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
57
58 void
59 fzap_byteswap(void *vbuf, size_t size)
60 {
61 uint64_t block_type;
62
63 block_type = *(uint64_t *)vbuf;
64
65 if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF))
66 zap_leaf_byteswap(vbuf, size);
67 else {
68 /* it's a ptrtbl block */
69 byteswap_uint64_array(vbuf, size);
70 }
71 }
72
73 void
74 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
75 {
76 dmu_buf_t *db;
77 zap_leaf_t *l;
78 int i;
79 zap_phys_t *zp;
80
81 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
82 zap->zap_ismicro = FALSE;
83
84 zap->zap_dbu.dbu_evict_func_sync = zap_evict_sync;
85 zap->zap_dbu.dbu_evict_func_async = NULL;
86
87 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, MUTEX_DEFAULT, 0);
88 zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;
89
90 zp = zap_f_phys(zap);
91 /*
92 * explicitly zero it since it might be coming from an
93 * initialized microzap
94 */
95 bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
96 zp->zap_block_type = ZBT_HEADER;
97 zp->zap_magic = ZAP_MAGIC;
98
99 zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
100
101 zp->zap_freeblk = 2; /* block 1 will be the first leaf */
102 zp->zap_num_leafs = 1;
103 zp->zap_num_entries = 0;
104 zp->zap_salt = zap->zap_salt;
105 zp->zap_normflags = zap->zap_normflags;
106 zp->zap_flags = flags;
107
108 /* block 1 will be the first leaf */
109 for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
110 ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
111
112 /*
113 * set up block 1 - the first leaf
114 */
115 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
116 1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
117 dmu_buf_will_dirty(db, tx);
118
119 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
120 l->l_dbuf = db;
121
122 zap_leaf_init(l, zp->zap_normflags != 0);
123
124 kmem_free(l, sizeof (zap_leaf_t));
125 dmu_buf_rele(db, FTAG);
126 }
127
128 static int
129 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
130 {
131 if (RW_WRITE_HELD(&zap->zap_rwlock))
132 return (1);
133 if (rw_tryupgrade(&zap->zap_rwlock)) {
134 dmu_buf_will_dirty(zap->zap_dbuf, tx);
135 return (1);
136 }
137 return (0);
138 }
139
140 /*
141 * Generic routines for dealing with the pointer & cookie tables.
142 */
143
144 static int
145 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
146 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
147 dmu_tx_t *tx)
148 {
149 uint64_t b, newblk;
150 dmu_buf_t *db_old, *db_new;
151 int err;
152 int bs = FZAP_BLOCK_SHIFT(zap);
153 int hepb = 1<<(bs-4);
154 /* hepb = half the number of entries in a block */
155
156 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
157 ASSERT(tbl->zt_blk != 0);
158 ASSERT(tbl->zt_numblks > 0);
159
160 if (tbl->zt_nextblk != 0) {
161 newblk = tbl->zt_nextblk;
162 } else {
163 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
164 tbl->zt_nextblk = newblk;
165 ASSERT0(tbl->zt_blks_copied);
166 dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
167 tbl->zt_blk << bs, tbl->zt_numblks << bs,
168 ZIO_PRIORITY_SYNC_READ);
169 }
170
171 /*
172 * Copy the ptrtbl from the old to new location.
173 */
174
175 b = tbl->zt_blks_copied;
176 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
177 (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
178 if (err)
179 return (err);
180
181 /* first half of entries in old[b] go to new[2*b+0] */
182 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
183 (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
184 dmu_buf_will_dirty(db_new, tx);
185 transfer_func(db_old->db_data, db_new->db_data, hepb);
186 dmu_buf_rele(db_new, FTAG);
187
188 /* second half of entries in old[b] go to new[2*b+1] */
189 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
190 (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
191 dmu_buf_will_dirty(db_new, tx);
192 transfer_func((uint64_t *)db_old->db_data + hepb,
193 db_new->db_data, hepb);
194 dmu_buf_rele(db_new, FTAG);
195
196 dmu_buf_rele(db_old, FTAG);
197
198 tbl->zt_blks_copied++;
199
200 dprintf("copied block %llu of %llu\n",
201 tbl->zt_blks_copied, tbl->zt_numblks);
202
203 if (tbl->zt_blks_copied == tbl->zt_numblks) {
204 (void) dmu_free_range(zap->zap_objset, zap->zap_object,
205 tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
206
207 tbl->zt_blk = newblk;
208 tbl->zt_numblks *= 2;
209 tbl->zt_shift++;
210 tbl->zt_nextblk = 0;
211 tbl->zt_blks_copied = 0;
212
213 dprintf("finished; numblocks now %llu (%uk entries)\n",
214 tbl->zt_numblks, 1<<(tbl->zt_shift-10));
215 }
216
217 return (0);
218 }
219
220 static int
221 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
222 dmu_tx_t *tx)
223 {
224 int err;
225 uint64_t blk, off;
226 int bs = FZAP_BLOCK_SHIFT(zap);
227 dmu_buf_t *db;
228
229 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
230 ASSERT(tbl->zt_blk != 0);
231
232 dprintf("storing %llx at index %llx\n", val, idx);
233
234 blk = idx >> (bs-3);
235 off = idx & ((1<<(bs-3))-1);
236
237 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
238 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
239 if (err)
240 return (err);
241 dmu_buf_will_dirty(db, tx);
242
243 if (tbl->zt_nextblk != 0) {
244 uint64_t idx2 = idx * 2;
245 uint64_t blk2 = idx2 >> (bs-3);
246 uint64_t off2 = idx2 & ((1<<(bs-3))-1);
247 dmu_buf_t *db2;
248
249 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
250 (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
251 DMU_READ_NO_PREFETCH);
252 if (err) {
253 dmu_buf_rele(db, FTAG);
254 return (err);
255 }
256 dmu_buf_will_dirty(db2, tx);
257 ((uint64_t *)db2->db_data)[off2] = val;
258 ((uint64_t *)db2->db_data)[off2+1] = val;
259 dmu_buf_rele(db2, FTAG);
260 }
261
262 ((uint64_t *)db->db_data)[off] = val;
263 dmu_buf_rele(db, FTAG);
264
265 return (0);
266 }
267
268 static int
269 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
270 {
271 uint64_t blk, off;
272 int err;
273 dmu_buf_t *db;
274 dnode_t *dn;
275 int bs = FZAP_BLOCK_SHIFT(zap);
276
277 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
278
279 blk = idx >> (bs-3);
280 off = idx & ((1<<(bs-3))-1);
281
282 /*
283 * Note: this is equivalent to dmu_buf_hold(), but we use
284 * _dnode_enter / _by_dnode because it's faster because we don't
285 * have to hold the dnode.
286 */
287 dn = dmu_buf_dnode_enter(zap->zap_dbuf);
288 err = dmu_buf_hold_by_dnode(dn,
289 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
290 dmu_buf_dnode_exit(zap->zap_dbuf);
291 if (err)
292 return (err);
293 *valp = ((uint64_t *)db->db_data)[off];
294 dmu_buf_rele(db, FTAG);
295
296 if (tbl->zt_nextblk != 0) {
297 /*
298 * read the nextblk for the sake of i/o error checking,
299 * so that zap_table_load() will catch errors for
300 * zap_table_store.
301 */
302 blk = (idx*2) >> (bs-3);
303
304 dn = dmu_buf_dnode_enter(zap->zap_dbuf);
305 err = dmu_buf_hold_by_dnode(dn,
306 (tbl->zt_nextblk + blk) << bs, FTAG, &db,
307 DMU_READ_NO_PREFETCH);
308 dmu_buf_dnode_exit(zap->zap_dbuf);
309 if (err == 0)
310 dmu_buf_rele(db, FTAG);
311 }
312 return (err);
313 }
314
315 /*
316 * Routines for growing the ptrtbl.
317 */
318
319 static void
320 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
321 {
322 int i;
323 for (i = 0; i < n; i++) {
324 uint64_t lb = src[i];
325 dst[2*i+0] = lb;
326 dst[2*i+1] = lb;
327 }
328 }
329
330 static int
331 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
332 {
333 /*
334 * The pointer table should never use more hash bits than we
335 * have (otherwise we'd be using useless zero bits to index it).
336 * If we are within 2 bits of running out, stop growing, since
337 * this is already an aberrant condition.
338 */
339 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
340 return (SET_ERROR(ENOSPC));
341
342 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
343 /*
344 * We are outgrowing the "embedded" ptrtbl (the one
345 * stored in the header block). Give it its own entire
346 * block, which will double the size of the ptrtbl.
347 */
348 uint64_t newblk;
349 dmu_buf_t *db_new;
350 int err;
351
352 ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
353 ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
354 ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk);
355
356 newblk = zap_allocate_blocks(zap, 1);
357 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
358 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
359 DMU_READ_NO_PREFETCH);
360 if (err)
361 return (err);
362 dmu_buf_will_dirty(db_new, tx);
363 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
364 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
365 dmu_buf_rele(db_new, FTAG);
366
367 zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk;
368 zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1;
369 zap_f_phys(zap)->zap_ptrtbl.zt_shift++;
370
371 ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
372 zap_f_phys(zap)->zap_ptrtbl.zt_numblks <<
373 (FZAP_BLOCK_SHIFT(zap)-3));
374
375 return (0);
376 } else {
377 return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl,
378 zap_ptrtbl_transfer, tx));
379 }
380 }
381
382 static void
383 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
384 {
385 dmu_buf_will_dirty(zap->zap_dbuf, tx);
386 mutex_enter(&zap->zap_f.zap_num_entries_mtx);
387 ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta);
388 zap_f_phys(zap)->zap_num_entries += delta;
389 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
390 }
391
392 static uint64_t
393 zap_allocate_blocks(zap_t *zap, int nblocks)
394 {
395 uint64_t newblk;
396 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
397 newblk = zap_f_phys(zap)->zap_freeblk;
398 zap_f_phys(zap)->zap_freeblk += nblocks;
399 return (newblk);
400 }
401
402 static void
403 zap_leaf_evict_sync(void *dbu)
404 {
405 zap_leaf_t *l = dbu;
406
407 rw_destroy(&l->l_rwlock);
408 kmem_free(l, sizeof (zap_leaf_t));
409 }
410
411 static zap_leaf_t *
412 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
413 {
414 void *winner;
415 zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
416
417 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
418
419 rw_init(&l->l_rwlock, NULL, RW_NOLOCKDEP, NULL);
420 rw_enter(&l->l_rwlock, RW_WRITER);
421 l->l_blkid = zap_allocate_blocks(zap, 1);
422 l->l_dbuf = NULL;
423
424 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
425 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
426 DMU_READ_NO_PREFETCH));
427 dmu_buf_init_user(&l->l_dbu, zap_leaf_evict_sync, NULL, &l->l_dbuf);
428 winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu);
429 ASSERT(winner == NULL);
430 dmu_buf_will_dirty(l->l_dbuf, tx);
431
432 zap_leaf_init(l, zap->zap_normflags != 0);
433
434 zap_f_phys(zap)->zap_num_leafs++;
435
436 return (l);
437 }
438
439 int
440 fzap_count(zap_t *zap, uint64_t *count)
441 {
442 ASSERT(!zap->zap_ismicro);
443 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
444 *count = zap_f_phys(zap)->zap_num_entries;
445 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
446 return (0);
447 }
448
449 /*
450 * Routines for obtaining zap_leaf_t's
451 */
452
453 void
454 zap_put_leaf(zap_leaf_t *l)
455 {
456 rw_exit(&l->l_rwlock);
457 dmu_buf_rele(l->l_dbuf, NULL);
458 }
459
460 static zap_leaf_t *
461 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
462 {
463 zap_leaf_t *l, *winner;
464
465 ASSERT(blkid != 0);
466
467 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
468 rw_init(&l->l_rwlock, NULL, RW_DEFAULT, NULL);
469 rw_enter(&l->l_rwlock, RW_WRITER);
470 l->l_blkid = blkid;
471 l->l_bs = highbit64(db->db_size) - 1;
472 l->l_dbuf = db;
473
474 dmu_buf_init_user(&l->l_dbu, zap_leaf_evict_sync, NULL, &l->l_dbuf);
475 winner = dmu_buf_set_user(db, &l->l_dbu);
476
477 rw_exit(&l->l_rwlock);
478 if (winner != NULL) {
479 /* someone else set it first */
480 zap_leaf_evict_sync(&l->l_dbu);
481 l = winner;
482 }
483
484 /*
485 * lhr_pad was previously used for the next leaf in the leaf
486 * chain. There should be no chained leafs (as we have removed
487 * support for them).
488 */
489 ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1);
490
491 /*
492 * There should be more hash entries than there can be
493 * chunks to put in the hash table
494 */
495 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
496
497 /* The chunks should begin at the end of the hash table */
498 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==, (zap_leaf_chunk_t *)
499 &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
500
501 /* The chunks should end at the end of the block */
502 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
503 (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size);
504
505 return (l);
506 }
507
508 static int
509 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
510 zap_leaf_t **lp)
511 {
512 dmu_buf_t *db;
513 zap_leaf_t *l;
514 int bs = FZAP_BLOCK_SHIFT(zap);
515 int err;
516
517 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
518
519 /*
520 * If system crashed just after dmu_free_long_range in zfs_rmnode, we
521 * would be left with an empty xattr dir in delete queue. blkid=0
522 * would be passed in when doing zfs_purgedir. If that's the case we
523 * should just return immediately. The underlying objects should
524 * already be freed, so this should be perfectly fine.
525 */
526 if (blkid == 0)
527 return (SET_ERROR(ENOENT));
528
529 dnode_t *dn = dmu_buf_dnode_enter(zap->zap_dbuf);
530 err = dmu_buf_hold_by_dnode(dn,
531 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
532 dmu_buf_dnode_exit(zap->zap_dbuf);
533 if (err)
534 return (err);
535
536 ASSERT3U(db->db_object, ==, zap->zap_object);
537 ASSERT3U(db->db_offset, ==, blkid << bs);
538 ASSERT3U(db->db_size, ==, 1 << bs);
539 ASSERT(blkid != 0);
540
541 l = dmu_buf_get_user(db);
542
543 if (l == NULL)
544 l = zap_open_leaf(blkid, db);
545
546 rw_enter(&l->l_rwlock, lt);
547 /*
548 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
549 * causing ASSERT below to fail.
550 */
551 if (lt == RW_WRITER)
552 dmu_buf_will_dirty(db, tx);
553 ASSERT3U(l->l_blkid, ==, blkid);
554 ASSERT3P(l->l_dbuf, ==, db);
555 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
556 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
557
558 *lp = l;
559 return (0);
560 }
561
562 static int
563 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
564 {
565 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
566
567 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
568 ASSERT3U(idx, <,
569 (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
570 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
571 return (0);
572 } else {
573 return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
574 idx, valp));
575 }
576 }
577
578 static int
579 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
580 {
581 ASSERT(tx != NULL);
582 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
583
584 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
585 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
586 return (0);
587 } else {
588 return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
589 idx, blk, tx));
590 }
591 }
592
593 static int
594 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
595 {
596 uint64_t idx, blk;
597 int err;
598
599 ASSERT(zap->zap_dbuf == NULL ||
600 zap_f_phys(zap) == zap->zap_dbuf->db_data);
601
602 /* Reality check for corrupt zap objects (leaf or header). */
603 if ((zap_f_phys(zap)->zap_block_type != ZBT_LEAF &&
604 zap_f_phys(zap)->zap_block_type != ZBT_HEADER) ||
605 zap_f_phys(zap)->zap_magic != ZAP_MAGIC) {
606 return (SET_ERROR(EIO));
607 }
608 idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
609 err = zap_idx_to_blk(zap, idx, &blk);
610 if (err != 0)
611 return (err);
612 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
613
614 ASSERT(err ||
615 ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
616 zap_leaf_phys(*lp)->l_hdr.lh_prefix);
617 return (err);
618 }
619
620 static int
621 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l,
622 void *tag, dmu_tx_t *tx, zap_leaf_t **lp)
623 {
624 zap_t *zap = zn->zn_zap;
625 uint64_t hash = zn->zn_hash;
626 zap_leaf_t *nl;
627 int prefix_diff, i, err;
628 uint64_t sibling;
629 int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len;
630
631 ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
632 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
633
634 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
635 zap_leaf_phys(l)->l_hdr.lh_prefix);
636
637 if (zap_tryupgradedir(zap, tx) == 0 ||
638 old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
639 /* We failed to upgrade, or need to grow the pointer table */
640 objset_t *os = zap->zap_objset;
641 uint64_t object = zap->zap_object;
642
643 zap_put_leaf(l);
644 zap_unlockdir(zap, tag);
645 err = zap_lockdir(os, object, tx, RW_WRITER,
646 FALSE, FALSE, tag, &zn->zn_zap);
647 zap = zn->zn_zap;
648 if (err)
649 return (err);
650 ASSERT(!zap->zap_ismicro);
651
652 while (old_prefix_len ==
653 zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
654 err = zap_grow_ptrtbl(zap, tx);
655 if (err)
656 return (err);
657 }
658
659 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
660 if (err)
661 return (err);
662
663 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) {
664 /* it split while our locks were down */
665 *lp = l;
666 return (0);
667 }
668 }
669 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
670 ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
671 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
672 zap_leaf_phys(l)->l_hdr.lh_prefix);
673
674 prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
675 (old_prefix_len + 1);
676 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
677
678 /* check for i/o errors before doing zap_leaf_split */
679 for (i = 0; i < (1ULL<<prefix_diff); i++) {
680 uint64_t blk;
681 err = zap_idx_to_blk(zap, sibling+i, &blk);
682 if (err)
683 return (err);
684 ASSERT3U(blk, ==, l->l_blkid);
685 }
686
687 nl = zap_create_leaf(zap, tx);
688 zap_leaf_split(l, nl, zap->zap_normflags != 0);
689
690 /* set sibling pointers */
691 for (i = 0; i < (1ULL << prefix_diff); i++) {
692 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
693 ASSERT0(err); /* we checked for i/o errors above */
694 }
695
696 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_prefix_len, >, 0);
697
698 if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) {
699 /* we want the sibling */
700 zap_put_leaf(l);
701 *lp = nl;
702 } else {
703 zap_put_leaf(nl);
704 *lp = l;
705 }
706
707 return (0);
708 }
709
710 static void
711 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l,
712 void *tag, dmu_tx_t *tx)
713 {
714 zap_t *zap = zn->zn_zap;
715 int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
716 int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
717 zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
718
719 zap_put_leaf(l);
720
721 if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
722 int err;
723
724 /*
725 * We are in the middle of growing the pointer table, or
726 * this leaf will soon make us grow it.
727 */
728 if (zap_tryupgradedir(zap, tx) == 0) {
729 objset_t *os = zap->zap_objset;
730 uint64_t zapobj = zap->zap_object;
731
732 zap_unlockdir(zap, tag);
733 err = zap_lockdir(os, zapobj, tx,
734 RW_WRITER, FALSE, FALSE, tag, &zn->zn_zap);
735 zap = zn->zn_zap;
736 if (err)
737 return;
738 }
739
740 /* could have finished growing while our locks were down */
741 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
742 (void) zap_grow_ptrtbl(zap, tx);
743 }
744 }
745
746 static int
747 fzap_checkname(zap_name_t *zn)
748 {
749 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
750 return (SET_ERROR(ENAMETOOLONG));
751 return (0);
752 }
753
754 static int
755 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
756 {
757 /* Only integer sizes supported by C */
758 switch (integer_size) {
759 case 1:
760 case 2:
761 case 4:
762 case 8:
763 break;
764 default:
765 return (SET_ERROR(EINVAL));
766 }
767
768 if (integer_size * num_integers > ZAP_MAXVALUELEN)
769 return (SET_ERROR(E2BIG));
770
771 return (0);
772 }
773
774 static int
775 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
776 {
777 int err;
778
779 if ((err = fzap_checkname(zn)) != 0)
780 return (err);
781 return (fzap_checksize(integer_size, num_integers));
782 }
783
784 /*
785 * Routines for manipulating attributes.
786 */
787 int
788 fzap_lookup(zap_name_t *zn,
789 uint64_t integer_size, uint64_t num_integers, void *buf,
790 char *realname, int rn_len, boolean_t *ncp)
791 {
792 zap_leaf_t *l;
793 int err;
794 zap_entry_handle_t zeh;
795
796 if ((err = fzap_checkname(zn)) != 0)
797 return (err);
798
799 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
800 if (err != 0)
801 return (err);
802 err = zap_leaf_lookup(l, zn, &zeh);
803 if (err == 0) {
804 if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
805 zap_put_leaf(l);
806 return (err);
807 }
808
809 err = zap_entry_read(&zeh, integer_size, num_integers, buf);
810 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
811 if (ncp) {
812 *ncp = zap_entry_normalization_conflict(&zeh,
813 zn, NULL, zn->zn_zap);
814 }
815 }
816
817 zap_put_leaf(l);
818 return (err);
819 }
820
821 int
822 fzap_add_cd(zap_name_t *zn,
823 uint64_t integer_size, uint64_t num_integers,
824 const void *val, uint32_t cd, void *tag, dmu_tx_t *tx)
825 {
826 zap_leaf_t *l;
827 int err;
828 zap_entry_handle_t zeh;
829 zap_t *zap = zn->zn_zap;
830
831 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
832 ASSERT(!zap->zap_ismicro);
833 ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
834
835 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
836 if (err != 0)
837 return (err);
838 retry:
839 err = zap_leaf_lookup(l, zn, &zeh);
840 if (err == 0) {
841 err = SET_ERROR(EEXIST);
842 goto out;
843 }
844 if (err != ENOENT)
845 goto out;
846
847 err = zap_entry_create(l, zn, cd,
848 integer_size, num_integers, val, &zeh);
849
850 if (err == 0) {
851 zap_increment_num_entries(zap, 1, tx);
852 } else if (err == EAGAIN) {
853 err = zap_expand_leaf(zn, l, tag, tx, &l);
854 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
855 if (err == 0)
856 goto retry;
857 }
858
859 out:
860 if (zap != NULL)
861 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
862 return (err);
863 }
864
865 int
866 fzap_add(zap_name_t *zn,
867 uint64_t integer_size, uint64_t num_integers,
868 const void *val, void *tag, dmu_tx_t *tx)
869 {
870 int err = fzap_check(zn, integer_size, num_integers);
871 if (err != 0)
872 return (err);
873
874 return (fzap_add_cd(zn, integer_size, num_integers,
875 val, ZAP_NEED_CD, tag, tx));
876 }
877
878 int
879 fzap_update(zap_name_t *zn,
880 int integer_size, uint64_t num_integers, const void *val,
881 void *tag, dmu_tx_t *tx)
882 {
883 zap_leaf_t *l;
884 int err, create;
885 zap_entry_handle_t zeh;
886 zap_t *zap = zn->zn_zap;
887
888 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
889 err = fzap_check(zn, integer_size, num_integers);
890 if (err != 0)
891 return (err);
892
893 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
894 if (err != 0)
895 return (err);
896 retry:
897 err = zap_leaf_lookup(l, zn, &zeh);
898 create = (err == ENOENT);
899 ASSERT(err == 0 || err == ENOENT);
900
901 if (create) {
902 err = zap_entry_create(l, zn, ZAP_NEED_CD,
903 integer_size, num_integers, val, &zeh);
904 if (err == 0)
905 zap_increment_num_entries(zap, 1, tx);
906 } else {
907 err = zap_entry_update(&zeh, integer_size, num_integers, val);
908 }
909
910 if (err == EAGAIN) {
911 err = zap_expand_leaf(zn, l, tag, tx, &l);
912 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
913 if (err == 0)
914 goto retry;
915 }
916
917 if (zap != NULL)
918 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tag, tx);
919 return (err);
920 }
921
922 int
923 fzap_length(zap_name_t *zn,
924 uint64_t *integer_size, uint64_t *num_integers)
925 {
926 zap_leaf_t *l;
927 int err;
928 zap_entry_handle_t zeh;
929
930 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
931 if (err != 0)
932 return (err);
933 err = zap_leaf_lookup(l, zn, &zeh);
934 if (err != 0)
935 goto out;
936
937 if (integer_size)
938 *integer_size = zeh.zeh_integer_size;
939 if (num_integers)
940 *num_integers = zeh.zeh_num_integers;
941 out:
942 zap_put_leaf(l);
943 return (err);
944 }
945
946 int
947 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
948 {
949 zap_leaf_t *l;
950 int err;
951 zap_entry_handle_t zeh;
952
953 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
954 if (err != 0)
955 return (err);
956 err = zap_leaf_lookup(l, zn, &zeh);
957 if (err == 0) {
958 zap_entry_remove(&zeh);
959 zap_increment_num_entries(zn->zn_zap, -1, tx);
960 }
961 zap_put_leaf(l);
962 return (err);
963 }
964
965 void
966 fzap_prefetch(zap_name_t *zn)
967 {
968 uint64_t idx, blk;
969 zap_t *zap = zn->zn_zap;
970 int bs;
971
972 idx = ZAP_HASH_IDX(zn->zn_hash,
973 zap_f_phys(zap)->zap_ptrtbl.zt_shift);
974 if (zap_idx_to_blk(zap, idx, &blk) != 0)
975 return;
976 bs = FZAP_BLOCK_SHIFT(zap);
977 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs,
978 ZIO_PRIORITY_SYNC_READ);
979 }
980
981 /*
982 * Helper functions for consumers.
983 */
984
985 uint64_t
986 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
987 const char *name, dmu_tx_t *tx)
988 {
989 return (zap_create_link_dnsize(os, ot, parent_obj, name, 0, tx));
990 }
991
992 uint64_t
993 zap_create_link_dnsize(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
994 const char *name, int dnodesize, dmu_tx_t *tx)
995 {
996 uint64_t new_obj;
997
998 VERIFY((new_obj = zap_create_dnsize(os, ot, DMU_OT_NONE, 0,
999 dnodesize, tx)) > 0);
1000 VERIFY0(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
1001 tx));
1002
1003 return (new_obj);
1004 }
1005
1006 int
1007 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
1008 char *name)
1009 {
1010 zap_cursor_t zc;
1011 zap_attribute_t *za;
1012 int err;
1013
1014 if (mask == 0)
1015 mask = -1ULL;
1016
1017 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1018 for (zap_cursor_init(&zc, os, zapobj);
1019 (err = zap_cursor_retrieve(&zc, za)) == 0;
1020 zap_cursor_advance(&zc)) {
1021 if ((za->za_first_integer & mask) == (value & mask)) {
1022 (void) strcpy(name, za->za_name);
1023 break;
1024 }
1025 }
1026 zap_cursor_fini(&zc);
1027 kmem_free(za, sizeof (zap_attribute_t));
1028 return (err);
1029 }
1030
1031 int
1032 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
1033 {
1034 zap_cursor_t zc;
1035 zap_attribute_t za;
1036 int err;
1037
1038 err = 0;
1039 for (zap_cursor_init(&zc, os, fromobj);
1040 zap_cursor_retrieve(&zc, &za) == 0;
1041 (void) zap_cursor_advance(&zc)) {
1042 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1043 err = SET_ERROR(EINVAL);
1044 break;
1045 }
1046 err = zap_add(os, intoobj, za.za_name,
1047 8, 1, &za.za_first_integer, tx);
1048 if (err)
1049 break;
1050 }
1051 zap_cursor_fini(&zc);
1052 return (err);
1053 }
1054
1055 int
1056 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1057 uint64_t value, dmu_tx_t *tx)
1058 {
1059 zap_cursor_t zc;
1060 zap_attribute_t za;
1061 int err;
1062
1063 err = 0;
1064 for (zap_cursor_init(&zc, os, fromobj);
1065 zap_cursor_retrieve(&zc, &za) == 0;
1066 (void) zap_cursor_advance(&zc)) {
1067 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1068 err = SET_ERROR(EINVAL);
1069 break;
1070 }
1071 err = zap_add(os, intoobj, za.za_name,
1072 8, 1, &value, tx);
1073 if (err)
1074 break;
1075 }
1076 zap_cursor_fini(&zc);
1077 return (err);
1078 }
1079
1080 int
1081 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1082 dmu_tx_t *tx)
1083 {
1084 zap_cursor_t zc;
1085 zap_attribute_t za;
1086 int err;
1087
1088 err = 0;
1089 for (zap_cursor_init(&zc, os, fromobj);
1090 zap_cursor_retrieve(&zc, &za) == 0;
1091 (void) zap_cursor_advance(&zc)) {
1092 uint64_t delta = 0;
1093
1094 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1095 err = SET_ERROR(EINVAL);
1096 break;
1097 }
1098
1099 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1100 if (err != 0 && err != ENOENT)
1101 break;
1102 delta += za.za_first_integer;
1103 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1104 if (err)
1105 break;
1106 }
1107 zap_cursor_fini(&zc);
1108 return (err);
1109 }
1110
1111 int
1112 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1113 {
1114 char name[20];
1115
1116 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1117 return (zap_add(os, obj, name, 8, 1, &value, tx));
1118 }
1119
1120 int
1121 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1122 {
1123 char name[20];
1124
1125 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1126 return (zap_remove(os, obj, name, tx));
1127 }
1128
1129 int
1130 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1131 {
1132 char name[20];
1133
1134 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1135 return (zap_lookup(os, obj, name, 8, 1, &value));
1136 }
1137
1138 int
1139 zap_add_int_key(objset_t *os, uint64_t obj,
1140 uint64_t key, uint64_t value, dmu_tx_t *tx)
1141 {
1142 char name[20];
1143
1144 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1145 return (zap_add(os, obj, name, 8, 1, &value, tx));
1146 }
1147
1148 int
1149 zap_update_int_key(objset_t *os, uint64_t obj,
1150 uint64_t key, uint64_t value, dmu_tx_t *tx)
1151 {
1152 char name[20];
1153
1154 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1155 return (zap_update(os, obj, name, 8, 1, &value, tx));
1156 }
1157
1158 int
1159 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1160 {
1161 char name[20];
1162
1163 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1164 return (zap_lookup(os, obj, name, 8, 1, valuep));
1165 }
1166
1167 int
1168 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1169 dmu_tx_t *tx)
1170 {
1171 uint64_t value = 0;
1172 int err;
1173
1174 if (delta == 0)
1175 return (0);
1176
1177 err = zap_lookup(os, obj, name, 8, 1, &value);
1178 if (err != 0 && err != ENOENT)
1179 return (err);
1180 value += delta;
1181 if (value == 0)
1182 err = zap_remove(os, obj, name, tx);
1183 else
1184 err = zap_update(os, obj, name, 8, 1, &value, tx);
1185 return (err);
1186 }
1187
1188 int
1189 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1190 dmu_tx_t *tx)
1191 {
1192 char name[20];
1193
1194 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1195 return (zap_increment(os, obj, name, delta, tx));
1196 }
1197
1198 /*
1199 * Routines for iterating over the attributes.
1200 */
1201
1202 int
1203 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1204 {
1205 int err = ENOENT;
1206 zap_entry_handle_t zeh;
1207 zap_leaf_t *l;
1208
1209 /* retrieve the next entry at or after zc_hash/zc_cd */
1210 /* if no entry, return ENOENT */
1211
1212 if (zc->zc_leaf &&
1213 (ZAP_HASH_IDX(zc->zc_hash,
1214 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1215 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1216 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1217 zap_put_leaf(zc->zc_leaf);
1218 zc->zc_leaf = NULL;
1219 }
1220
1221 again:
1222 if (zc->zc_leaf == NULL) {
1223 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1224 &zc->zc_leaf);
1225 if (err != 0)
1226 return (err);
1227 } else {
1228 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1229 }
1230 l = zc->zc_leaf;
1231
1232 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1233
1234 if (err == ENOENT) {
1235 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0) {
1236 zc->zc_hash = -1ULL;
1237 zc->zc_cd = 0;
1238 } else {
1239 uint64_t nocare = (1ULL <<
1240 (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1241
1242 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1243 zc->zc_cd = 0;
1244
1245 if (zc->zc_hash == 0) {
1246 zc->zc_hash = -1ULL;
1247 } else {
1248 zap_put_leaf(zc->zc_leaf);
1249 zc->zc_leaf = NULL;
1250 goto again;
1251 }
1252 }
1253 }
1254
1255 if (err == 0) {
1256 zc->zc_hash = zeh.zeh_hash;
1257 zc->zc_cd = zeh.zeh_cd;
1258 za->za_integer_length = zeh.zeh_integer_size;
1259 za->za_num_integers = zeh.zeh_num_integers;
1260 if (zeh.zeh_num_integers == 0) {
1261 za->za_first_integer = 0;
1262 } else {
1263 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1264 ASSERT(err == 0 || err == EOVERFLOW);
1265 }
1266 err = zap_entry_read_name(zap, &zeh,
1267 sizeof (za->za_name), za->za_name);
1268 ASSERT(err == 0);
1269
1270 za->za_normalization_conflict =
1271 zap_entry_normalization_conflict(&zeh,
1272 NULL, za->za_name, zap);
1273 }
1274 rw_exit(&zc->zc_leaf->l_rwlock);
1275 return (err);
1276 }
1277
1278 static void
1279 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1280 {
1281 int i, err;
1282 uint64_t lastblk = 0;
1283
1284 /*
1285 * NB: if a leaf has more pointers than an entire ptrtbl block
1286 * can hold, then it'll be accounted for more than once, since
1287 * we won't have lastblk.
1288 */
1289 for (i = 0; i < len; i++) {
1290 zap_leaf_t *l;
1291
1292 if (tbl[i] == lastblk)
1293 continue;
1294 lastblk = tbl[i];
1295
1296 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1297 if (err == 0) {
1298 zap_leaf_stats(zap, l, zs);
1299 zap_put_leaf(l);
1300 }
1301 }
1302 }
1303
1304 void
1305 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1306 {
1307 int bs = FZAP_BLOCK_SHIFT(zap);
1308 zs->zs_blocksize = 1ULL << bs;
1309
1310 /*
1311 * Set zap_phys_t fields
1312 */
1313 zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1314 zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1315 zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1316 zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1317 zs->zs_magic = zap_f_phys(zap)->zap_magic;
1318 zs->zs_salt = zap_f_phys(zap)->zap_salt;
1319
1320 /*
1321 * Set zap_ptrtbl fields
1322 */
1323 zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1324 zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1325 zs->zs_ptrtbl_blks_copied =
1326 zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1327 zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1328 zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1329 zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1330
1331 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1332 /* the ptrtbl is entirely in the header block. */
1333 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1334 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1335 } else {
1336 int b;
1337
1338 dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
1339 zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1340 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs,
1341 ZIO_PRIORITY_SYNC_READ);
1342
1343 for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1344 b++) {
1345 dmu_buf_t *db;
1346 int err;
1347
1348 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1349 (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1350 FTAG, &db, DMU_READ_NO_PREFETCH);
1351 if (err == 0) {
1352 zap_stats_ptrtbl(zap, db->db_data,
1353 1<<(bs-3), zs);
1354 dmu_buf_rele(db, FTAG);
1355 }
1356 }
1357 }
1358 }