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) 2013 by Delphix. All rights reserved.
24 */
25
26 /*
27 * This file contains the top half of the zfs directory structure
28 * implementation. The bottom half is in zap_leaf.c.
29 *
30 * The zdir is an extendable hash data structure. There is a table of
31 * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are
32 * each a constant size and hold a variable number of directory entries.
33 * The buckets (aka "leaf nodes") are implemented in zap_leaf.c.
34 *
35 * The pointer table holds a power of 2 number of pointers.
36 * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to
37 * by the pointer at index i in the table holds entries whose hash value
38 * has a zd_prefix_len - bit prefix
39 */
40
41 #include <sys/spa.h>
42 #include <sys/dmu.h>
43 #include <sys/zfs_context.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/fs/zfs.h>
46 #include <sys/zap.h>
47 #include <sys/refcount.h>
48 #include <sys/zap_impl.h>
49 #include <sys/zap_leaf.h>
50
51 int fzap_default_block_shift = 14; /* 16k blocksize */
52
53 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
54
55 void
56 fzap_byteswap(void *vbuf, size_t size)
57 {
58 uint64_t block_type;
59
60 block_type = *(uint64_t *)vbuf;
61
62 if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF))
63 zap_leaf_byteswap(vbuf, size);
64 else {
65 /* it's a ptrtbl block */
66 byteswap_uint64_array(vbuf, size);
67 }
68 }
69
70 void
71 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
72 {
73 dmu_buf_t *db;
74 zap_leaf_t *l;
75 int i;
76 zap_phys_t *zp;
77
78 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
79 zap->zap_ismicro = FALSE;
80
81 zap->db_evict.evict_func = zap_evict;
82
83 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
84 zap->zap_f.zap_block_shift = highbit(zap->zap_dbuf->db_size) - 1;
85
86 zp = zap->zap_f_phys;
87 /*
88 * explicitly zero it since it might be coming from an
89 * initialized microzap
90 */
91 bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
92 zp->zap_block_type = ZBT_HEADER;
93 zp->zap_magic = ZAP_MAGIC;
94
95 zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
96
97 zp->zap_freeblk = 2; /* block 1 will be the first leaf */
98 zp->zap_num_leafs = 1;
99 zp->zap_num_entries = 0;
100 zp->zap_salt = zap->zap_salt;
101 zp->zap_normflags = zap->zap_normflags;
102 zp->zap_flags = flags;
103
104 /* block 1 will be the first leaf */
105 for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
106 ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
107
108 /*
109 * set up block 1 - the first leaf
110 */
111 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
112 1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
113 dmu_buf_will_dirty(db, tx);
114
115 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
116 l->l_dbuf = db;
117
118 zap_leaf_init(l, zp->zap_normflags != 0);
119
120 kmem_free(l, sizeof (zap_leaf_t));
121 dmu_buf_rele(db, FTAG);
122 }
123
124 static int
125 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
126 {
127 if (RW_WRITE_HELD(&zap->zap_rwlock))
128 return (1);
129 if (rw_tryupgrade(&zap->zap_rwlock)) {
130 dmu_buf_will_dirty(zap->zap_dbuf, tx);
131 return (1);
132 }
133 return (0);
134 }
135
136 /*
137 * Generic routines for dealing with the pointer & cookie tables.
138 */
139
140 static int
141 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
142 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
143 dmu_tx_t *tx)
144 {
145 uint64_t b, newblk;
146 dmu_buf_t *db_old, *db_new;
147 int err;
148 int bs = FZAP_BLOCK_SHIFT(zap);
149 int hepb = 1<<(bs-4);
150 /* hepb = half the number of entries in a block */
151
152 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
153 ASSERT(tbl->zt_blk != 0);
154 ASSERT(tbl->zt_numblks > 0);
155
156 if (tbl->zt_nextblk != 0) {
157 newblk = tbl->zt_nextblk;
158 } else {
159 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
160 tbl->zt_nextblk = newblk;
161 ASSERT0(tbl->zt_blks_copied);
162 dmu_prefetch(zap->zap_objset, zap->zap_object,
163 tbl->zt_blk << bs, tbl->zt_numblks << bs);
164 }
165
166 /*
167 * Copy the ptrtbl from the old to new location.
168 */
169
170 b = tbl->zt_blks_copied;
171 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
172 (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
173 if (err)
174 return (err);
175
176 /* first half of entries in old[b] go to new[2*b+0] */
177 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
178 (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
179 dmu_buf_will_dirty(db_new, tx);
180 transfer_func(db_old->db_data, db_new->db_data, hepb);
181 dmu_buf_rele(db_new, FTAG);
182
183 /* second half of entries in old[b] go to new[2*b+1] */
184 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
185 (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
186 dmu_buf_will_dirty(db_new, tx);
187 transfer_func((uint64_t *)db_old->db_data + hepb,
188 db_new->db_data, hepb);
189 dmu_buf_rele(db_new, FTAG);
190
191 dmu_buf_rele(db_old, FTAG);
192
193 tbl->zt_blks_copied++;
194
195 dprintf("copied block %llu of %llu\n",
196 tbl->zt_blks_copied, tbl->zt_numblks);
197
198 if (tbl->zt_blks_copied == tbl->zt_numblks) {
199 (void) dmu_free_range(zap->zap_objset, zap->zap_object,
200 tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
201
202 tbl->zt_blk = newblk;
203 tbl->zt_numblks *= 2;
204 tbl->zt_shift++;
205 tbl->zt_nextblk = 0;
206 tbl->zt_blks_copied = 0;
207
208 dprintf("finished; numblocks now %llu (%lluk entries)\n",
209 tbl->zt_numblks, 1<<(tbl->zt_shift-10));
210 }
211
212 return (0);
213 }
214
215 static int
216 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
217 dmu_tx_t *tx)
218 {
219 int err;
220 uint64_t blk, off;
221 int bs = FZAP_BLOCK_SHIFT(zap);
222 dmu_buf_t *db;
223
224 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
225 ASSERT(tbl->zt_blk != 0);
226
227 dprintf("storing %llx at index %llx\n", val, idx);
228
229 blk = idx >> (bs-3);
230 off = idx & ((1<<(bs-3))-1);
231
232 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
233 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
234 if (err)
235 return (err);
236 dmu_buf_will_dirty(db, tx);
237
238 if (tbl->zt_nextblk != 0) {
239 uint64_t idx2 = idx * 2;
240 uint64_t blk2 = idx2 >> (bs-3);
241 uint64_t off2 = idx2 & ((1<<(bs-3))-1);
242 dmu_buf_t *db2;
243
244 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
245 (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
246 DMU_READ_NO_PREFETCH);
247 if (err) {
248 dmu_buf_rele(db, FTAG);
249 return (err);
250 }
251 dmu_buf_will_dirty(db2, tx);
252 ((uint64_t *)db2->db_data)[off2] = val;
253 ((uint64_t *)db2->db_data)[off2+1] = val;
254 dmu_buf_rele(db2, FTAG);
255 }
256
257 ((uint64_t *)db->db_data)[off] = val;
258 dmu_buf_rele(db, FTAG);
259
260 return (0);
261 }
262
263 static int
264 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
265 {
266 uint64_t blk, off;
267 int err;
268 dmu_buf_t *db;
269 int bs = FZAP_BLOCK_SHIFT(zap);
270
271 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
272
273 blk = idx >> (bs-3);
274 off = idx & ((1<<(bs-3))-1);
275
276 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
277 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
278 if (err)
279 return (err);
280 *valp = ((uint64_t *)db->db_data)[off];
281 dmu_buf_rele(db, FTAG);
282
283 if (tbl->zt_nextblk != 0) {
284 /*
285 * read the nextblk for the sake of i/o error checking,
286 * so that zap_table_load() will catch errors for
287 * zap_table_store.
288 */
289 blk = (idx*2) >> (bs-3);
290
291 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
292 (tbl->zt_nextblk + blk) << bs, FTAG, &db,
293 DMU_READ_NO_PREFETCH);
294 dmu_buf_rele(db, FTAG);
295 }
296 return (err);
297 }
298
299 /*
300 * Routines for growing the ptrtbl.
301 */
302
303 static void
304 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
305 {
306 int i;
307 for (i = 0; i < n; i++) {
308 uint64_t lb = src[i];
309 dst[2*i+0] = lb;
310 dst[2*i+1] = lb;
311 }
312 }
313
314 static int
315 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
316 {
317 /*
318 * The pointer table should never use more hash bits than we
319 * have (otherwise we'd be using useless zero bits to index it).
320 * If we are within 2 bits of running out, stop growing, since
321 * this is already an aberrant condition.
322 */
323 if (zap->zap_f_phys->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
324 return (SET_ERROR(ENOSPC));
325
326 if (zap->zap_f_phys->zap_ptrtbl.zt_numblks == 0) {
327 /*
328 * We are outgrowing the "embedded" ptrtbl (the one
329 * stored in the header block). Give it its own entire
330 * block, which will double the size of the ptrtbl.
331 */
332 uint64_t newblk;
333 dmu_buf_t *db_new;
334 int err;
335
336 ASSERT3U(zap->zap_f_phys->zap_ptrtbl.zt_shift, ==,
337 ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
338 ASSERT0(zap->zap_f_phys->zap_ptrtbl.zt_blk);
339
340 newblk = zap_allocate_blocks(zap, 1);
341 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
342 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
343 DMU_READ_NO_PREFETCH);
344 if (err)
345 return (err);
346 dmu_buf_will_dirty(db_new, tx);
347 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
348 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
349 dmu_buf_rele(db_new, FTAG);
350
351 zap->zap_f_phys->zap_ptrtbl.zt_blk = newblk;
352 zap->zap_f_phys->zap_ptrtbl.zt_numblks = 1;
353 zap->zap_f_phys->zap_ptrtbl.zt_shift++;
354
355 ASSERT3U(1ULL << zap->zap_f_phys->zap_ptrtbl.zt_shift, ==,
356 zap->zap_f_phys->zap_ptrtbl.zt_numblks <<
357 (FZAP_BLOCK_SHIFT(zap)-3));
358
359 return (0);
360 } else {
361 return (zap_table_grow(zap, &zap->zap_f_phys->zap_ptrtbl,
362 zap_ptrtbl_transfer, tx));
363 }
364 }
365
366 static void
367 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
368 {
369 dmu_buf_will_dirty(zap->zap_dbuf, tx);
370 mutex_enter(&zap->zap_f.zap_num_entries_mtx);
371 ASSERT(delta > 0 || zap->zap_f_phys->zap_num_entries >= -delta);
372 zap->zap_f_phys->zap_num_entries += delta;
373 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
374 }
375
376 static uint64_t
377 zap_allocate_blocks(zap_t *zap, int nblocks)
378 {
379 uint64_t newblk;
380 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
381 newblk = zap->zap_f_phys->zap_freeblk;
382 zap->zap_f_phys->zap_freeblk += nblocks;
383 return (newblk);
384 }
385
386 static void
387 zap_leaf_pageout(dmu_buf_user_t *dbu)
388 {
389 zap_leaf_t *l = (zap_leaf_t *)dbu;
390
391 rw_destroy(&l->l_rwlock);
392 kmem_free(l, sizeof (zap_leaf_t));
393 }
394
395 static zap_leaf_t *
396 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
397 {
398 void *winner;
399 zap_leaf_t *l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
400
401 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
402
403 rw_init(&l->l_rwlock, 0, 0, 0);
404 rw_enter(&l->l_rwlock, RW_WRITER);
405 l->l_blkid = zap_allocate_blocks(zap, 1);
406 l->l_dbuf = NULL;
407
408 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
409 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
410 DMU_READ_NO_PREFETCH));
411 dmu_buf_init_user(&l->db_evict, zap_leaf_pageout);
412 winner = (zap_leaf_t *)dmu_buf_set_user(l->l_dbuf, &l->db_evict);
413 ASSERT(winner == NULL);
414 dmu_buf_will_dirty(l->l_dbuf, tx);
415
416 zap_leaf_init(l, zap->zap_normflags != 0);
417
418 zap->zap_f_phys->zap_num_leafs++;
419
420 return (l);
421 }
422
423 int
424 fzap_count(zap_t *zap, uint64_t *count)
425 {
426 ASSERT(!zap->zap_ismicro);
427 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
428 *count = zap->zap_f_phys->zap_num_entries;
429 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
430 return (0);
431 }
432
433 /*
434 * Routines for obtaining zap_leaf_t's
435 */
436
437 void
438 zap_put_leaf(zap_leaf_t *l)
439 {
440 rw_exit(&l->l_rwlock);
441 dmu_buf_rele(l->l_dbuf, NULL);
442 }
443
444 static zap_leaf_t *
445 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
446 {
447 zap_leaf_t *l, *winner;
448
449 ASSERT(blkid != 0);
450
451 l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
452 rw_init(&l->l_rwlock, 0, 0, 0);
453 rw_enter(&l->l_rwlock, RW_WRITER);
454 l->l_blkid = blkid;
455 l->l_bs = highbit(db->db_size)-1;
456 l->l_dbuf = db;
457
458 dmu_buf_init_user(&l->db_evict, zap_leaf_pageout);
459 winner = (zap_leaf_t *)dmu_buf_set_user(db, &l->db_evict);
460
461 rw_exit(&l->l_rwlock);
462 if (winner != NULL) {
463 /* someone else set it first */
464 zap_leaf_pageout(&l->db_evict);
465 l = winner;
466 }
467
468 /*
469 * lhr_pad was previously used for the next leaf in the leaf
470 * chain. There should be no chained leafs (as we have removed
471 * support for them).
472 */
473 ASSERT0(l->l_phys->l_hdr.lh_pad1);
474
475 /*
476 * There should be more hash entries than there can be
477 * chunks to put in the hash table
478 */
479 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
480
481 /* The chunks should begin at the end of the hash table */
482 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
483 &l->l_phys->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
484
485 /* The chunks should end at the end of the block */
486 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
487 (uintptr_t)l->l_phys, ==, l->l_dbuf->db_size);
488
489 return (l);
490 }
491
492 static int
493 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
494 zap_leaf_t **lp)
495 {
496 dmu_buf_t *db;
497 zap_leaf_t *l;
498 int bs = FZAP_BLOCK_SHIFT(zap);
499 int err;
500
501 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
502
503 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
504 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
505 if (err)
506 return (err);
507
508 ASSERT3U(db->db_object, ==, zap->zap_object);
509 ASSERT3U(db->db_offset, ==, blkid << bs);
510 ASSERT3U(db->db_size, ==, 1 << bs);
511 ASSERT(blkid != 0);
512
513 l = (zap_leaf_t *)dmu_buf_get_user(db);
514
515 if (l == NULL)
516 l = zap_open_leaf(blkid, db);
517
518 rw_enter(&l->l_rwlock, lt);
519 /*
520 * Must lock before dirtying, otherwise l->l_phys could change,
521 * causing ASSERT below to fail.
522 */
523 if (lt == RW_WRITER)
524 dmu_buf_will_dirty(db, tx);
525 ASSERT3U(l->l_blkid, ==, blkid);
526 ASSERT3P(l->l_dbuf, ==, db);
527 ASSERT3P(l->l_phys, ==, l->l_dbuf->db_data);
528 ASSERT3U(l->l_phys->l_hdr.lh_block_type, ==, ZBT_LEAF);
529 ASSERT3U(l->l_phys->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
530
531 *lp = l;
532 return (0);
533 }
534
535 static int
536 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
537 {
538 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
539
540 if (zap->zap_f_phys->zap_ptrtbl.zt_numblks == 0) {
541 ASSERT3U(idx, <,
542 (1ULL << zap->zap_f_phys->zap_ptrtbl.zt_shift));
543 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
544 return (0);
545 } else {
546 return (zap_table_load(zap, &zap->zap_f_phys->zap_ptrtbl,
547 idx, valp));
548 }
549 }
550
551 static int
552 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
553 {
554 ASSERT(tx != NULL);
555 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
556
557 if (zap->zap_f_phys->zap_ptrtbl.zt_blk == 0) {
558 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
559 return (0);
560 } else {
561 return (zap_table_store(zap, &zap->zap_f_phys->zap_ptrtbl,
562 idx, blk, tx));
563 }
564 }
565
566 static int
567 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
568 {
569 uint64_t idx, blk;
570 int err;
571
572 ASSERT(zap->zap_dbuf == NULL ||
573 zap->zap_f_phys == zap->zap_dbuf->db_data);
574 ASSERT3U(zap->zap_f_phys->zap_magic, ==, ZAP_MAGIC);
575 idx = ZAP_HASH_IDX(h, zap->zap_f_phys->zap_ptrtbl.zt_shift);
576 err = zap_idx_to_blk(zap, idx, &blk);
577 if (err != 0)
578 return (err);
579 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
580
581 ASSERT(err || ZAP_HASH_IDX(h, (*lp)->l_phys->l_hdr.lh_prefix_len) ==
582 (*lp)->l_phys->l_hdr.lh_prefix);
583 return (err);
584 }
585
586 static int
587 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx, zap_leaf_t **lp)
588 {
589 zap_t *zap = zn->zn_zap;
590 uint64_t hash = zn->zn_hash;
591 zap_leaf_t *nl;
592 int prefix_diff, i, err;
593 uint64_t sibling;
594 int old_prefix_len = l->l_phys->l_hdr.lh_prefix_len;
595
596 ASSERT3U(old_prefix_len, <=, zap->zap_f_phys->zap_ptrtbl.zt_shift);
597 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
598
599 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
600 l->l_phys->l_hdr.lh_prefix);
601
602 if (zap_tryupgradedir(zap, tx) == 0 ||
603 old_prefix_len == zap->zap_f_phys->zap_ptrtbl.zt_shift) {
604 /* We failed to upgrade, or need to grow the pointer table */
605 objset_t *os = zap->zap_objset;
606 uint64_t object = zap->zap_object;
607
608 zap_put_leaf(l);
609 zap_unlockdir(zap);
610 err = zap_lockdir(os, object, tx, RW_WRITER,
611 FALSE, FALSE, &zn->zn_zap);
612 zap = zn->zn_zap;
613 if (err)
614 return (err);
615 ASSERT(!zap->zap_ismicro);
616
617 while (old_prefix_len ==
618 zap->zap_f_phys->zap_ptrtbl.zt_shift) {
619 err = zap_grow_ptrtbl(zap, tx);
620 if (err)
621 return (err);
622 }
623
624 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
625 if (err)
626 return (err);
627
628 if (l->l_phys->l_hdr.lh_prefix_len != old_prefix_len) {
629 /* it split while our locks were down */
630 *lp = l;
631 return (0);
632 }
633 }
634 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
635 ASSERT3U(old_prefix_len, <, zap->zap_f_phys->zap_ptrtbl.zt_shift);
636 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
637 l->l_phys->l_hdr.lh_prefix);
638
639 prefix_diff = zap->zap_f_phys->zap_ptrtbl.zt_shift -
640 (old_prefix_len + 1);
641 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
642
643 /* check for i/o errors before doing zap_leaf_split */
644 for (i = 0; i < (1ULL<<prefix_diff); i++) {
645 uint64_t blk;
646 err = zap_idx_to_blk(zap, sibling+i, &blk);
647 if (err)
648 return (err);
649 ASSERT3U(blk, ==, l->l_blkid);
650 }
651
652 nl = zap_create_leaf(zap, tx);
653 zap_leaf_split(l, nl, zap->zap_normflags != 0);
654
655 /* set sibling pointers */
656 for (i = 0; i < (1ULL << prefix_diff); i++) {
657 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
658 ASSERT0(err); /* we checked for i/o errors above */
659 }
660
661 if (hash & (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len))) {
662 /* we want the sibling */
663 zap_put_leaf(l);
664 *lp = nl;
665 } else {
666 zap_put_leaf(nl);
667 *lp = l;
668 }
669
670 return (0);
671 }
672
673 static void
674 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx)
675 {
676 zap_t *zap = zn->zn_zap;
677 int shift = zap->zap_f_phys->zap_ptrtbl.zt_shift;
678 int leaffull = (l->l_phys->l_hdr.lh_prefix_len == shift &&
679 l->l_phys->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
680
681 zap_put_leaf(l);
682
683 if (leaffull || zap->zap_f_phys->zap_ptrtbl.zt_nextblk) {
684 int err;
685
686 /*
687 * We are in the middle of growing the pointer table, or
688 * this leaf will soon make us grow it.
689 */
690 if (zap_tryupgradedir(zap, tx) == 0) {
691 objset_t *os = zap->zap_objset;
692 uint64_t zapobj = zap->zap_object;
693
694 zap_unlockdir(zap);
695 err = zap_lockdir(os, zapobj, tx,
696 RW_WRITER, FALSE, FALSE, &zn->zn_zap);
697 zap = zn->zn_zap;
698 if (err)
699 return;
700 }
701
702 /* could have finished growing while our locks were down */
703 if (zap->zap_f_phys->zap_ptrtbl.zt_shift == shift)
704 (void) zap_grow_ptrtbl(zap, tx);
705 }
706 }
707
708 static int
709 fzap_checkname(zap_name_t *zn)
710 {
711 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
712 return (SET_ERROR(ENAMETOOLONG));
713 return (0);
714 }
715
716 static int
717 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
718 {
719 /* Only integer sizes supported by C */
720 switch (integer_size) {
721 case 1:
722 case 2:
723 case 4:
724 case 8:
725 break;
726 default:
727 return (SET_ERROR(EINVAL));
728 }
729
730 if (integer_size * num_integers > ZAP_MAXVALUELEN)
731 return (E2BIG);
732
733 return (0);
734 }
735
736 static int
737 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
738 {
739 int err;
740
741 if ((err = fzap_checkname(zn)) != 0)
742 return (err);
743 return (fzap_checksize(integer_size, num_integers));
744 }
745
746 /*
747 * Routines for manipulating attributes.
748 */
749 int
750 fzap_lookup(zap_name_t *zn,
751 uint64_t integer_size, uint64_t num_integers, void *buf,
752 char *realname, int rn_len, boolean_t *ncp)
753 {
754 zap_leaf_t *l;
755 int err;
756 zap_entry_handle_t zeh;
757
758 if ((err = fzap_checkname(zn)) != 0)
759 return (err);
760
761 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
762 if (err != 0)
763 return (err);
764 err = zap_leaf_lookup(l, zn, &zeh);
765 if (err == 0) {
766 if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
767 zap_put_leaf(l);
768 return (err);
769 }
770
771 err = zap_entry_read(&zeh, integer_size, num_integers, buf);
772 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
773 if (ncp) {
774 *ncp = zap_entry_normalization_conflict(&zeh,
775 zn, NULL, zn->zn_zap);
776 }
777 }
778
779 zap_put_leaf(l);
780 return (err);
781 }
782
783 int
784 fzap_add_cd(zap_name_t *zn,
785 uint64_t integer_size, uint64_t num_integers,
786 const void *val, uint32_t cd, dmu_tx_t *tx)
787 {
788 zap_leaf_t *l;
789 int err;
790 zap_entry_handle_t zeh;
791 zap_t *zap = zn->zn_zap;
792
793 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
794 ASSERT(!zap->zap_ismicro);
795 ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
796
797 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
798 if (err != 0)
799 return (err);
800 retry:
801 err = zap_leaf_lookup(l, zn, &zeh);
802 if (err == 0) {
803 err = SET_ERROR(EEXIST);
804 goto out;
805 }
806 if (err != ENOENT)
807 goto out;
808
809 err = zap_entry_create(l, zn, cd,
810 integer_size, num_integers, val, &zeh);
811
812 if (err == 0) {
813 zap_increment_num_entries(zap, 1, tx);
814 } else if (err == EAGAIN) {
815 err = zap_expand_leaf(zn, l, tx, &l);
816 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
817 if (err == 0)
818 goto retry;
819 }
820
821 out:
822 if (zap != NULL)
823 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
824 return (err);
825 }
826
827 int
828 fzap_add(zap_name_t *zn,
829 uint64_t integer_size, uint64_t num_integers,
830 const void *val, dmu_tx_t *tx)
831 {
832 int err = fzap_check(zn, integer_size, num_integers);
833 if (err != 0)
834 return (err);
835
836 return (fzap_add_cd(zn, integer_size, num_integers,
837 val, ZAP_NEED_CD, tx));
838 }
839
840 int
841 fzap_update(zap_name_t *zn,
842 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
843 {
844 zap_leaf_t *l;
845 int err, create;
846 zap_entry_handle_t zeh;
847 zap_t *zap = zn->zn_zap;
848
849 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
850 err = fzap_check(zn, integer_size, num_integers);
851 if (err != 0)
852 return (err);
853
854 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
855 if (err != 0)
856 return (err);
857 retry:
858 err = zap_leaf_lookup(l, zn, &zeh);
859 create = (err == ENOENT);
860 ASSERT(err == 0 || err == ENOENT);
861
862 if (create) {
863 err = zap_entry_create(l, zn, ZAP_NEED_CD,
864 integer_size, num_integers, val, &zeh);
865 if (err == 0)
866 zap_increment_num_entries(zap, 1, tx);
867 } else {
868 err = zap_entry_update(&zeh, integer_size, num_integers, val);
869 }
870
871 if (err == EAGAIN) {
872 err = zap_expand_leaf(zn, l, tx, &l);
873 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
874 if (err == 0)
875 goto retry;
876 }
877
878 if (zap != NULL)
879 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
880 return (err);
881 }
882
883 int
884 fzap_length(zap_name_t *zn,
885 uint64_t *integer_size, uint64_t *num_integers)
886 {
887 zap_leaf_t *l;
888 int err;
889 zap_entry_handle_t zeh;
890
891 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
892 if (err != 0)
893 return (err);
894 err = zap_leaf_lookup(l, zn, &zeh);
895 if (err != 0)
896 goto out;
897
898 if (integer_size)
899 *integer_size = zeh.zeh_integer_size;
900 if (num_integers)
901 *num_integers = zeh.zeh_num_integers;
902 out:
903 zap_put_leaf(l);
904 return (err);
905 }
906
907 int
908 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
909 {
910 zap_leaf_t *l;
911 int err;
912 zap_entry_handle_t zeh;
913
914 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
915 if (err != 0)
916 return (err);
917 err = zap_leaf_lookup(l, zn, &zeh);
918 if (err == 0) {
919 zap_entry_remove(&zeh);
920 zap_increment_num_entries(zn->zn_zap, -1, tx);
921 }
922 zap_put_leaf(l);
923 return (err);
924 }
925
926 void
927 fzap_prefetch(zap_name_t *zn)
928 {
929 uint64_t idx, blk;
930 zap_t *zap = zn->zn_zap;
931 int bs;
932
933 idx = ZAP_HASH_IDX(zn->zn_hash,
934 zap->zap_f_phys->zap_ptrtbl.zt_shift);
935 if (zap_idx_to_blk(zap, idx, &blk) != 0)
936 return;
937 bs = FZAP_BLOCK_SHIFT(zap);
938 dmu_prefetch(zap->zap_objset, zap->zap_object, blk << bs, 1 << bs);
939 }
940
941 /*
942 * Helper functions for consumers.
943 */
944
945 uint64_t
946 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
947 const char *name, dmu_tx_t *tx)
948 {
949 uint64_t new_obj;
950
951 VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
952 VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
953 tx) == 0);
954
955 return (new_obj);
956 }
957
958 int
959 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
960 char *name)
961 {
962 zap_cursor_t zc;
963 zap_attribute_t *za;
964 int err;
965
966 if (mask == 0)
967 mask = -1ULL;
968
969 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
970 for (zap_cursor_init(&zc, os, zapobj);
971 (err = zap_cursor_retrieve(&zc, za)) == 0;
972 zap_cursor_advance(&zc)) {
973 if ((za->za_first_integer & mask) == (value & mask)) {
974 (void) strcpy(name, za->za_name);
975 break;
976 }
977 }
978 zap_cursor_fini(&zc);
979 kmem_free(za, sizeof (zap_attribute_t));
980 return (err);
981 }
982
983 int
984 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
985 {
986 zap_cursor_t zc;
987 zap_attribute_t za;
988 int err;
989
990 for (zap_cursor_init(&zc, os, fromobj);
991 zap_cursor_retrieve(&zc, &za) == 0;
992 (void) zap_cursor_advance(&zc)) {
993 if (za.za_integer_length != 8 || za.za_num_integers != 1)
994 return (SET_ERROR(EINVAL));
995 err = zap_add(os, intoobj, za.za_name,
996 8, 1, &za.za_first_integer, tx);
997 if (err)
998 return (err);
999 }
1000 zap_cursor_fini(&zc);
1001 return (0);
1002 }
1003
1004 int
1005 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1006 uint64_t value, dmu_tx_t *tx)
1007 {
1008 zap_cursor_t zc;
1009 zap_attribute_t za;
1010 int err;
1011
1012 for (zap_cursor_init(&zc, os, fromobj);
1013 zap_cursor_retrieve(&zc, &za) == 0;
1014 (void) zap_cursor_advance(&zc)) {
1015 if (za.za_integer_length != 8 || za.za_num_integers != 1)
1016 return (SET_ERROR(EINVAL));
1017 err = zap_add(os, intoobj, za.za_name,
1018 8, 1, &value, tx);
1019 if (err)
1020 return (err);
1021 }
1022 zap_cursor_fini(&zc);
1023 return (0);
1024 }
1025
1026 int
1027 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1028 dmu_tx_t *tx)
1029 {
1030 zap_cursor_t zc;
1031 zap_attribute_t za;
1032 int err;
1033
1034 for (zap_cursor_init(&zc, os, fromobj);
1035 zap_cursor_retrieve(&zc, &za) == 0;
1036 (void) zap_cursor_advance(&zc)) {
1037 uint64_t delta = 0;
1038
1039 if (za.za_integer_length != 8 || za.za_num_integers != 1)
1040 return (SET_ERROR(EINVAL));
1041
1042 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1043 if (err != 0 && err != ENOENT)
1044 return (err);
1045 delta += za.za_first_integer;
1046 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1047 if (err)
1048 return (err);
1049 }
1050 zap_cursor_fini(&zc);
1051 return (0);
1052 }
1053
1054 int
1055 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1056 {
1057 char name[20];
1058
1059 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1060 return (zap_add(os, obj, name, 8, 1, &value, tx));
1061 }
1062
1063 int
1064 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1065 {
1066 char name[20];
1067
1068 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1069 return (zap_remove(os, obj, name, tx));
1070 }
1071
1072 int
1073 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1074 {
1075 char name[20];
1076
1077 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1078 return (zap_lookup(os, obj, name, 8, 1, &value));
1079 }
1080
1081 int
1082 zap_add_int_key(objset_t *os, uint64_t obj,
1083 uint64_t key, uint64_t value, dmu_tx_t *tx)
1084 {
1085 char name[20];
1086
1087 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1088 return (zap_add(os, obj, name, 8, 1, &value, tx));
1089 }
1090
1091 int
1092 zap_update_int_key(objset_t *os, uint64_t obj,
1093 uint64_t key, uint64_t value, dmu_tx_t *tx)
1094 {
1095 char name[20];
1096
1097 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1098 return (zap_update(os, obj, name, 8, 1, &value, tx));
1099 }
1100
1101 int
1102 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1103 {
1104 char name[20];
1105
1106 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1107 return (zap_lookup(os, obj, name, 8, 1, valuep));
1108 }
1109
1110 int
1111 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1112 dmu_tx_t *tx)
1113 {
1114 uint64_t value = 0;
1115 int err;
1116
1117 if (delta == 0)
1118 return (0);
1119
1120 err = zap_lookup(os, obj, name, 8, 1, &value);
1121 if (err != 0 && err != ENOENT)
1122 return (err);
1123 value += delta;
1124 if (value == 0)
1125 err = zap_remove(os, obj, name, tx);
1126 else
1127 err = zap_update(os, obj, name, 8, 1, &value, tx);
1128 return (err);
1129 }
1130
1131 int
1132 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1133 dmu_tx_t *tx)
1134 {
1135 char name[20];
1136
1137 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1138 return (zap_increment(os, obj, name, delta, tx));
1139 }
1140
1141 /*
1142 * Routines for iterating over the attributes.
1143 */
1144
1145 int
1146 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1147 {
1148 int err = ENOENT;
1149 zap_entry_handle_t zeh;
1150 zap_leaf_t *l;
1151
1152 /* retrieve the next entry at or after zc_hash/zc_cd */
1153 /* if no entry, return ENOENT */
1154
1155 if (zc->zc_leaf &&
1156 (ZAP_HASH_IDX(zc->zc_hash,
1157 zc->zc_leaf->l_phys->l_hdr.lh_prefix_len) !=
1158 zc->zc_leaf->l_phys->l_hdr.lh_prefix)) {
1159 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1160 zap_put_leaf(zc->zc_leaf);
1161 zc->zc_leaf = NULL;
1162 }
1163
1164 again:
1165 if (zc->zc_leaf == NULL) {
1166 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1167 &zc->zc_leaf);
1168 if (err != 0)
1169 return (err);
1170 } else {
1171 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1172 }
1173 l = zc->zc_leaf;
1174
1175 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1176
1177 if (err == ENOENT) {
1178 uint64_t nocare =
1179 (1ULL << (64 - l->l_phys->l_hdr.lh_prefix_len)) - 1;
1180 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1181 zc->zc_cd = 0;
1182 if (l->l_phys->l_hdr.lh_prefix_len == 0 || zc->zc_hash == 0) {
1183 zc->zc_hash = -1ULL;
1184 } else {
1185 zap_put_leaf(zc->zc_leaf);
1186 zc->zc_leaf = NULL;
1187 goto again;
1188 }
1189 }
1190
1191 if (err == 0) {
1192 zc->zc_hash = zeh.zeh_hash;
1193 zc->zc_cd = zeh.zeh_cd;
1194 za->za_integer_length = zeh.zeh_integer_size;
1195 za->za_num_integers = zeh.zeh_num_integers;
1196 if (zeh.zeh_num_integers == 0) {
1197 za->za_first_integer = 0;
1198 } else {
1199 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1200 ASSERT(err == 0 || err == EOVERFLOW);
1201 }
1202 err = zap_entry_read_name(zap, &zeh,
1203 sizeof (za->za_name), za->za_name);
1204 ASSERT(err == 0);
1205
1206 za->za_normalization_conflict =
1207 zap_entry_normalization_conflict(&zeh,
1208 NULL, za->za_name, zap);
1209 }
1210 rw_exit(&zc->zc_leaf->l_rwlock);
1211 return (err);
1212 }
1213
1214 static void
1215 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1216 {
1217 int i, err;
1218 uint64_t lastblk = 0;
1219
1220 /*
1221 * NB: if a leaf has more pointers than an entire ptrtbl block
1222 * can hold, then it'll be accounted for more than once, since
1223 * we won't have lastblk.
1224 */
1225 for (i = 0; i < len; i++) {
1226 zap_leaf_t *l;
1227
1228 if (tbl[i] == lastblk)
1229 continue;
1230 lastblk = tbl[i];
1231
1232 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1233 if (err == 0) {
1234 zap_leaf_stats(zap, l, zs);
1235 zap_put_leaf(l);
1236 }
1237 }
1238 }
1239
1240 int
1241 fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn)
1242 {
1243 int err;
1244 zap_leaf_t *l;
1245 zap_entry_handle_t zeh;
1246
1247 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
1248 return (SET_ERROR(ENAMETOOLONG));
1249
1250 err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l);
1251 if (err != 0)
1252 return (err);
1253
1254 err = zap_leaf_lookup(l, zn, &zeh);
1255 if (err != 0)
1256 return (err);
1257
1258 zc->zc_leaf = l;
1259 zc->zc_hash = zeh.zeh_hash;
1260 zc->zc_cd = zeh.zeh_cd;
1261
1262 return (err);
1263 }
1264
1265 void
1266 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1267 {
1268 int bs = FZAP_BLOCK_SHIFT(zap);
1269 zs->zs_blocksize = 1ULL << bs;
1270
1271 /*
1272 * Set zap_phys_t fields
1273 */
1274 zs->zs_num_leafs = zap->zap_f_phys->zap_num_leafs;
1275 zs->zs_num_entries = zap->zap_f_phys->zap_num_entries;
1276 zs->zs_num_blocks = zap->zap_f_phys->zap_freeblk;
1277 zs->zs_block_type = zap->zap_f_phys->zap_block_type;
1278 zs->zs_magic = zap->zap_f_phys->zap_magic;
1279 zs->zs_salt = zap->zap_f_phys->zap_salt;
1280
1281 /*
1282 * Set zap_ptrtbl fields
1283 */
1284 zs->zs_ptrtbl_len = 1ULL << zap->zap_f_phys->zap_ptrtbl.zt_shift;
1285 zs->zs_ptrtbl_nextblk = zap->zap_f_phys->zap_ptrtbl.zt_nextblk;
1286 zs->zs_ptrtbl_blks_copied =
1287 zap->zap_f_phys->zap_ptrtbl.zt_blks_copied;
1288 zs->zs_ptrtbl_zt_blk = zap->zap_f_phys->zap_ptrtbl.zt_blk;
1289 zs->zs_ptrtbl_zt_numblks = zap->zap_f_phys->zap_ptrtbl.zt_numblks;
1290 zs->zs_ptrtbl_zt_shift = zap->zap_f_phys->zap_ptrtbl.zt_shift;
1291
1292 if (zap->zap_f_phys->zap_ptrtbl.zt_numblks == 0) {
1293 /* the ptrtbl is entirely in the header block. */
1294 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1295 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1296 } else {
1297 int b;
1298
1299 dmu_prefetch(zap->zap_objset, zap->zap_object,
1300 zap->zap_f_phys->zap_ptrtbl.zt_blk << bs,
1301 zap->zap_f_phys->zap_ptrtbl.zt_numblks << bs);
1302
1303 for (b = 0; b < zap->zap_f_phys->zap_ptrtbl.zt_numblks;
1304 b++) {
1305 dmu_buf_t *db;
1306 int err;
1307
1308 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1309 (zap->zap_f_phys->zap_ptrtbl.zt_blk + b) << bs,
1310 FTAG, &db, DMU_READ_NO_PREFETCH);
1311 if (err == 0) {
1312 zap_stats_ptrtbl(zap, db->db_data,
1313 1<<(bs-3), zs);
1314 dmu_buf_rele(db, FTAG);
1315 }
1316 }
1317 }
1318 }
1319
1320 int
1321 fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite,
1322 uint64_t *tooverwrite)
1323 {
1324 zap_t *zap = zn->zn_zap;
1325 zap_leaf_t *l;
1326 int err;
1327
1328 /*
1329 * Account for the header block of the fatzap.
1330 */
1331 if (!add && dmu_buf_freeable(zap->zap_dbuf)) {
1332 *tooverwrite += zap->zap_dbuf->db_size;
1333 } else {
1334 *towrite += zap->zap_dbuf->db_size;
1335 }
1336
1337 /*
1338 * Account for the pointer table blocks.
1339 * If we are adding we need to account for the following cases :
1340 * - If the pointer table is embedded, this operation could force an
1341 * external pointer table.
1342 * - If this already has an external pointer table this operation
1343 * could extend the table.
1344 */
1345 if (add) {
1346 if (zap->zap_f_phys->zap_ptrtbl.zt_blk == 0)
1347 *towrite += zap->zap_dbuf->db_size;
1348 else
1349 *towrite += (zap->zap_dbuf->db_size * 3);
1350 }
1351
1352 /*
1353 * Now, check if the block containing leaf is freeable
1354 * and account accordingly.
1355 */
1356 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l);
1357 if (err != 0) {
1358 return (err);
1359 }
1360
1361 if (!add && dmu_buf_freeable(l->l_dbuf)) {
1362 *tooverwrite += l->l_dbuf->db_size;
1363 } else {
1364 /*
1365 * If this an add operation, the leaf block could split.
1366 * Hence, we need to account for an additional leaf block.
1367 */
1368 *towrite += (add ? 2 : 1) * l->l_dbuf->db_size;
1369 }
1370
1371 zap_put_leaf(l);
1372 return (0);
1373 }