5421 devzvol_readdir() needs to be more careful with strchr
[illumos-gate.git] / usr / src / uts / common / fs / zfs / zap.c
blobde2d4deb00ebe665939bbbb64088c9eedacaac68
1 /*
2 * CDDL HEADER START
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.
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.
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]
19 * CDDL HEADER END
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
27 * This file contains the top half of the zfs directory structure
28 * implementation. The bottom half is in zap_leaf.c.
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.
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
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>
51 int fzap_default_block_shift = 14; /* 16k blocksize */
53 extern inline zap_phys_t *zap_f_phys(zap_t *zap);
55 static void zap_leaf_pageout(dmu_buf_t *db, void *vl);
56 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
58 void
59 fzap_byteswap(void *vbuf, size_t size)
61 uint64_t block_type;
63 block_type = *(uint64_t *)vbuf;
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);
73 void
74 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
76 dmu_buf_t *db;
77 zap_leaf_t *l;
78 int i;
79 zap_phys_t *zp;
81 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
82 zap->zap_ismicro = FALSE;
84 (void) dmu_buf_update_user(zap->zap_dbuf, zap, zap, zap_evict);
86 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
87 zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;
89 zp = zap_f_phys(zap);
91 * explicitly zero it since it might be coming from an
92 * initialized microzap
94 bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
95 zp->zap_block_type = ZBT_HEADER;
96 zp->zap_magic = ZAP_MAGIC;
98 zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
100 zp->zap_freeblk = 2; /* block 1 will be the first leaf */
101 zp->zap_num_leafs = 1;
102 zp->zap_num_entries = 0;
103 zp->zap_salt = zap->zap_salt;
104 zp->zap_normflags = zap->zap_normflags;
105 zp->zap_flags = flags;
107 /* block 1 will be the first leaf */
108 for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
109 ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
112 * set up block 1 - the first leaf
114 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
115 1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
116 dmu_buf_will_dirty(db, tx);
118 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
119 l->l_dbuf = db;
121 zap_leaf_init(l, zp->zap_normflags != 0);
123 kmem_free(l, sizeof (zap_leaf_t));
124 dmu_buf_rele(db, FTAG);
127 static int
128 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
130 if (RW_WRITE_HELD(&zap->zap_rwlock))
131 return (1);
132 if (rw_tryupgrade(&zap->zap_rwlock)) {
133 dmu_buf_will_dirty(zap->zap_dbuf, tx);
134 return (1);
136 return (0);
140 * Generic routines for dealing with the pointer & cookie tables.
143 static int
144 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
145 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
146 dmu_tx_t *tx)
148 uint64_t b, newblk;
149 dmu_buf_t *db_old, *db_new;
150 int err;
151 int bs = FZAP_BLOCK_SHIFT(zap);
152 int hepb = 1<<(bs-4);
153 /* hepb = half the number of entries in a block */
155 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
156 ASSERT(tbl->zt_blk != 0);
157 ASSERT(tbl->zt_numblks > 0);
159 if (tbl->zt_nextblk != 0) {
160 newblk = tbl->zt_nextblk;
161 } else {
162 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
163 tbl->zt_nextblk = newblk;
164 ASSERT0(tbl->zt_blks_copied);
165 dmu_prefetch(zap->zap_objset, zap->zap_object,
166 tbl->zt_blk << bs, tbl->zt_numblks << bs);
170 * Copy the ptrtbl from the old to new location.
173 b = tbl->zt_blks_copied;
174 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
175 (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
176 if (err)
177 return (err);
179 /* first half of entries in old[b] go to new[2*b+0] */
180 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
181 (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
182 dmu_buf_will_dirty(db_new, tx);
183 transfer_func(db_old->db_data, db_new->db_data, hepb);
184 dmu_buf_rele(db_new, FTAG);
186 /* second half of entries in old[b] go to new[2*b+1] */
187 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
188 (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
189 dmu_buf_will_dirty(db_new, tx);
190 transfer_func((uint64_t *)db_old->db_data + hepb,
191 db_new->db_data, hepb);
192 dmu_buf_rele(db_new, FTAG);
194 dmu_buf_rele(db_old, FTAG);
196 tbl->zt_blks_copied++;
198 dprintf("copied block %llu of %llu\n",
199 tbl->zt_blks_copied, tbl->zt_numblks);
201 if (tbl->zt_blks_copied == tbl->zt_numblks) {
202 (void) dmu_free_range(zap->zap_objset, zap->zap_object,
203 tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
205 tbl->zt_blk = newblk;
206 tbl->zt_numblks *= 2;
207 tbl->zt_shift++;
208 tbl->zt_nextblk = 0;
209 tbl->zt_blks_copied = 0;
211 dprintf("finished; numblocks now %llu (%lluk entries)\n",
212 tbl->zt_numblks, 1<<(tbl->zt_shift-10));
215 return (0);
218 static int
219 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
220 dmu_tx_t *tx)
222 int err;
223 uint64_t blk, off;
224 int bs = FZAP_BLOCK_SHIFT(zap);
225 dmu_buf_t *db;
227 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
228 ASSERT(tbl->zt_blk != 0);
230 dprintf("storing %llx at index %llx\n", val, idx);
232 blk = idx >> (bs-3);
233 off = idx & ((1<<(bs-3))-1);
235 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
236 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
237 if (err)
238 return (err);
239 dmu_buf_will_dirty(db, tx);
241 if (tbl->zt_nextblk != 0) {
242 uint64_t idx2 = idx * 2;
243 uint64_t blk2 = idx2 >> (bs-3);
244 uint64_t off2 = idx2 & ((1<<(bs-3))-1);
245 dmu_buf_t *db2;
247 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
248 (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
249 DMU_READ_NO_PREFETCH);
250 if (err) {
251 dmu_buf_rele(db, FTAG);
252 return (err);
254 dmu_buf_will_dirty(db2, tx);
255 ((uint64_t *)db2->db_data)[off2] = val;
256 ((uint64_t *)db2->db_data)[off2+1] = val;
257 dmu_buf_rele(db2, FTAG);
260 ((uint64_t *)db->db_data)[off] = val;
261 dmu_buf_rele(db, FTAG);
263 return (0);
266 static int
267 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
269 uint64_t blk, off;
270 int err;
271 dmu_buf_t *db;
272 int bs = FZAP_BLOCK_SHIFT(zap);
274 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
276 blk = idx >> (bs-3);
277 off = idx & ((1<<(bs-3))-1);
279 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
280 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
281 if (err)
282 return (err);
283 *valp = ((uint64_t *)db->db_data)[off];
284 dmu_buf_rele(db, FTAG);
286 if (tbl->zt_nextblk != 0) {
288 * read the nextblk for the sake of i/o error checking,
289 * so that zap_table_load() will catch errors for
290 * zap_table_store.
292 blk = (idx*2) >> (bs-3);
294 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
295 (tbl->zt_nextblk + blk) << bs, FTAG, &db,
296 DMU_READ_NO_PREFETCH);
297 if (err == 0)
298 dmu_buf_rele(db, FTAG);
300 return (err);
304 * Routines for growing the ptrtbl.
307 static void
308 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
310 int i;
311 for (i = 0; i < n; i++) {
312 uint64_t lb = src[i];
313 dst[2*i+0] = lb;
314 dst[2*i+1] = lb;
318 static int
319 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
322 * The pointer table should never use more hash bits than we
323 * have (otherwise we'd be using useless zero bits to index it).
324 * If we are within 2 bits of running out, stop growing, since
325 * this is already an aberrant condition.
327 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
328 return (SET_ERROR(ENOSPC));
330 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
332 * We are outgrowing the "embedded" ptrtbl (the one
333 * stored in the header block). Give it its own entire
334 * block, which will double the size of the ptrtbl.
336 uint64_t newblk;
337 dmu_buf_t *db_new;
338 int err;
340 ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
341 ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
342 ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk);
344 newblk = zap_allocate_blocks(zap, 1);
345 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
346 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
347 DMU_READ_NO_PREFETCH);
348 if (err)
349 return (err);
350 dmu_buf_will_dirty(db_new, tx);
351 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
352 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
353 dmu_buf_rele(db_new, FTAG);
355 zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk;
356 zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1;
357 zap_f_phys(zap)->zap_ptrtbl.zt_shift++;
359 ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
360 zap_f_phys(zap)->zap_ptrtbl.zt_numblks <<
361 (FZAP_BLOCK_SHIFT(zap)-3));
363 return (0);
364 } else {
365 return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl,
366 zap_ptrtbl_transfer, tx));
370 static void
371 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
373 dmu_buf_will_dirty(zap->zap_dbuf, tx);
374 mutex_enter(&zap->zap_f.zap_num_entries_mtx);
375 ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta);
376 zap_f_phys(zap)->zap_num_entries += delta;
377 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
380 static uint64_t
381 zap_allocate_blocks(zap_t *zap, int nblocks)
383 uint64_t newblk;
384 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
385 newblk = zap_f_phys(zap)->zap_freeblk;
386 zap_f_phys(zap)->zap_freeblk += nblocks;
387 return (newblk);
390 static zap_leaf_t *
391 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
393 void *winner;
394 zap_leaf_t *l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
396 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
398 rw_init(&l->l_rwlock, 0, 0, 0);
399 rw_enter(&l->l_rwlock, RW_WRITER);
400 l->l_blkid = zap_allocate_blocks(zap, 1);
401 l->l_dbuf = NULL;
403 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
404 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
405 DMU_READ_NO_PREFETCH));
406 winner = dmu_buf_set_user(l->l_dbuf, l, zap_leaf_pageout);
407 ASSERT(winner == NULL);
408 dmu_buf_will_dirty(l->l_dbuf, tx);
410 zap_leaf_init(l, zap->zap_normflags != 0);
412 zap_f_phys(zap)->zap_num_leafs++;
414 return (l);
418 fzap_count(zap_t *zap, uint64_t *count)
420 ASSERT(!zap->zap_ismicro);
421 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
422 *count = zap_f_phys(zap)->zap_num_entries;
423 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
424 return (0);
428 * Routines for obtaining zap_leaf_t's
431 void
432 zap_put_leaf(zap_leaf_t *l)
434 rw_exit(&l->l_rwlock);
435 dmu_buf_rele(l->l_dbuf, NULL);
438 _NOTE(ARGSUSED(0))
439 static void
440 zap_leaf_pageout(dmu_buf_t *db, void *vl)
442 zap_leaf_t *l = vl;
444 rw_destroy(&l->l_rwlock);
445 kmem_free(l, sizeof (zap_leaf_t));
448 static zap_leaf_t *
449 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
451 zap_leaf_t *l, *winner;
453 ASSERT(blkid != 0);
455 l = kmem_alloc(sizeof (zap_leaf_t), KM_SLEEP);
456 rw_init(&l->l_rwlock, 0, 0, 0);
457 rw_enter(&l->l_rwlock, RW_WRITER);
458 l->l_blkid = blkid;
459 l->l_bs = highbit64(db->db_size) - 1;
460 l->l_dbuf = db;
462 winner = dmu_buf_set_user(db, l, zap_leaf_pageout);
464 rw_exit(&l->l_rwlock);
465 if (winner != NULL) {
466 /* someone else set it first */
467 zap_leaf_pageout(NULL, l);
468 l = winner;
472 * lhr_pad was previously used for the next leaf in the leaf
473 * chain. There should be no chained leafs (as we have removed
474 * support for them).
476 ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1);
479 * There should be more hash entries than there can be
480 * chunks to put in the hash table
482 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
484 /* The chunks should begin at the end of the hash table */
485 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
486 &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
488 /* The chunks should end at the end of the block */
489 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
490 (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size);
492 return (l);
495 static int
496 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
497 zap_leaf_t **lp)
499 dmu_buf_t *db;
500 zap_leaf_t *l;
501 int bs = FZAP_BLOCK_SHIFT(zap);
502 int err;
504 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
506 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
507 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
508 if (err)
509 return (err);
511 ASSERT3U(db->db_object, ==, zap->zap_object);
512 ASSERT3U(db->db_offset, ==, blkid << bs);
513 ASSERT3U(db->db_size, ==, 1 << bs);
514 ASSERT(blkid != 0);
516 l = dmu_buf_get_user(db);
518 if (l == NULL)
519 l = zap_open_leaf(blkid, db);
521 rw_enter(&l->l_rwlock, lt);
523 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
524 * causing ASSERT below to fail.
526 if (lt == RW_WRITER)
527 dmu_buf_will_dirty(db, tx);
528 ASSERT3U(l->l_blkid, ==, blkid);
529 ASSERT3P(l->l_dbuf, ==, db);
530 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
531 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
533 *lp = l;
534 return (0);
537 static int
538 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
540 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
542 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
543 ASSERT3U(idx, <,
544 (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
545 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
546 return (0);
547 } else {
548 return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
549 idx, valp));
553 static int
554 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
556 ASSERT(tx != NULL);
557 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
559 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
560 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
561 return (0);
562 } else {
563 return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
564 idx, blk, tx));
568 static int
569 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
571 uint64_t idx, blk;
572 int err;
574 ASSERT(zap->zap_dbuf == NULL ||
575 zap_f_phys(zap) == zap->zap_dbuf->db_data);
576 ASSERT3U(zap_f_phys(zap)->zap_magic, ==, ZAP_MAGIC);
577 idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
578 err = zap_idx_to_blk(zap, idx, &blk);
579 if (err != 0)
580 return (err);
581 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
583 ASSERT(err ||
584 ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
585 zap_leaf_phys(*lp)->l_hdr.lh_prefix);
586 return (err);
589 static int
590 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx, zap_leaf_t **lp)
592 zap_t *zap = zn->zn_zap;
593 uint64_t hash = zn->zn_hash;
594 zap_leaf_t *nl;
595 int prefix_diff, i, err;
596 uint64_t sibling;
597 int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len;
599 ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
600 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
602 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
603 zap_leaf_phys(l)->l_hdr.lh_prefix);
605 if (zap_tryupgradedir(zap, tx) == 0 ||
606 old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
607 /* We failed to upgrade, or need to grow the pointer table */
608 objset_t *os = zap->zap_objset;
609 uint64_t object = zap->zap_object;
611 zap_put_leaf(l);
612 zap_unlockdir(zap);
613 err = zap_lockdir(os, object, tx, RW_WRITER,
614 FALSE, FALSE, &zn->zn_zap);
615 zap = zn->zn_zap;
616 if (err)
617 return (err);
618 ASSERT(!zap->zap_ismicro);
620 while (old_prefix_len ==
621 zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
622 err = zap_grow_ptrtbl(zap, tx);
623 if (err)
624 return (err);
627 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
628 if (err)
629 return (err);
631 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) {
632 /* it split while our locks were down */
633 *lp = l;
634 return (0);
637 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
638 ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
639 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
640 zap_leaf_phys(l)->l_hdr.lh_prefix);
642 prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
643 (old_prefix_len + 1);
644 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
646 /* check for i/o errors before doing zap_leaf_split */
647 for (i = 0; i < (1ULL<<prefix_diff); i++) {
648 uint64_t blk;
649 err = zap_idx_to_blk(zap, sibling+i, &blk);
650 if (err)
651 return (err);
652 ASSERT3U(blk, ==, l->l_blkid);
655 nl = zap_create_leaf(zap, tx);
656 zap_leaf_split(l, nl, zap->zap_normflags != 0);
658 /* set sibling pointers */
659 for (i = 0; i < (1ULL << prefix_diff); i++) {
660 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
661 ASSERT0(err); /* we checked for i/o errors above */
664 if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) {
665 /* we want the sibling */
666 zap_put_leaf(l);
667 *lp = nl;
668 } else {
669 zap_put_leaf(nl);
670 *lp = l;
673 return (0);
676 static void
677 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx)
679 zap_t *zap = zn->zn_zap;
680 int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
681 int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
682 zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
684 zap_put_leaf(l);
686 if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
687 int err;
690 * We are in the middle of growing the pointer table, or
691 * this leaf will soon make us grow it.
693 if (zap_tryupgradedir(zap, tx) == 0) {
694 objset_t *os = zap->zap_objset;
695 uint64_t zapobj = zap->zap_object;
697 zap_unlockdir(zap);
698 err = zap_lockdir(os, zapobj, tx,
699 RW_WRITER, FALSE, FALSE, &zn->zn_zap);
700 zap = zn->zn_zap;
701 if (err)
702 return;
705 /* could have finished growing while our locks were down */
706 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
707 (void) zap_grow_ptrtbl(zap, tx);
711 static int
712 fzap_checkname(zap_name_t *zn)
714 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
715 return (SET_ERROR(ENAMETOOLONG));
716 return (0);
719 static int
720 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
722 /* Only integer sizes supported by C */
723 switch (integer_size) {
724 case 1:
725 case 2:
726 case 4:
727 case 8:
728 break;
729 default:
730 return (SET_ERROR(EINVAL));
733 if (integer_size * num_integers > ZAP_MAXVALUELEN)
734 return (E2BIG);
736 return (0);
739 static int
740 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
742 int err;
744 if ((err = fzap_checkname(zn)) != 0)
745 return (err);
746 return (fzap_checksize(integer_size, num_integers));
750 * Routines for manipulating attributes.
753 fzap_lookup(zap_name_t *zn,
754 uint64_t integer_size, uint64_t num_integers, void *buf,
755 char *realname, int rn_len, boolean_t *ncp)
757 zap_leaf_t *l;
758 int err;
759 zap_entry_handle_t zeh;
761 if ((err = fzap_checkname(zn)) != 0)
762 return (err);
764 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
765 if (err != 0)
766 return (err);
767 err = zap_leaf_lookup(l, zn, &zeh);
768 if (err == 0) {
769 if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
770 zap_put_leaf(l);
771 return (err);
774 err = zap_entry_read(&zeh, integer_size, num_integers, buf);
775 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
776 if (ncp) {
777 *ncp = zap_entry_normalization_conflict(&zeh,
778 zn, NULL, zn->zn_zap);
782 zap_put_leaf(l);
783 return (err);
787 fzap_add_cd(zap_name_t *zn,
788 uint64_t integer_size, uint64_t num_integers,
789 const void *val, uint32_t cd, dmu_tx_t *tx)
791 zap_leaf_t *l;
792 int err;
793 zap_entry_handle_t zeh;
794 zap_t *zap = zn->zn_zap;
796 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
797 ASSERT(!zap->zap_ismicro);
798 ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
800 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
801 if (err != 0)
802 return (err);
803 retry:
804 err = zap_leaf_lookup(l, zn, &zeh);
805 if (err == 0) {
806 err = SET_ERROR(EEXIST);
807 goto out;
809 if (err != ENOENT)
810 goto out;
812 err = zap_entry_create(l, zn, cd,
813 integer_size, num_integers, val, &zeh);
815 if (err == 0) {
816 zap_increment_num_entries(zap, 1, tx);
817 } else if (err == EAGAIN) {
818 err = zap_expand_leaf(zn, l, tx, &l);
819 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
820 if (err == 0)
821 goto retry;
824 out:
825 if (zap != NULL)
826 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
827 return (err);
831 fzap_add(zap_name_t *zn,
832 uint64_t integer_size, uint64_t num_integers,
833 const void *val, dmu_tx_t *tx)
835 int err = fzap_check(zn, integer_size, num_integers);
836 if (err != 0)
837 return (err);
839 return (fzap_add_cd(zn, integer_size, num_integers,
840 val, ZAP_NEED_CD, tx));
844 fzap_update(zap_name_t *zn,
845 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
847 zap_leaf_t *l;
848 int err, create;
849 zap_entry_handle_t zeh;
850 zap_t *zap = zn->zn_zap;
852 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
853 err = fzap_check(zn, integer_size, num_integers);
854 if (err != 0)
855 return (err);
857 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
858 if (err != 0)
859 return (err);
860 retry:
861 err = zap_leaf_lookup(l, zn, &zeh);
862 create = (err == ENOENT);
863 ASSERT(err == 0 || err == ENOENT);
865 if (create) {
866 err = zap_entry_create(l, zn, ZAP_NEED_CD,
867 integer_size, num_integers, val, &zeh);
868 if (err == 0)
869 zap_increment_num_entries(zap, 1, tx);
870 } else {
871 err = zap_entry_update(&zeh, integer_size, num_integers, val);
874 if (err == EAGAIN) {
875 err = zap_expand_leaf(zn, l, tx, &l);
876 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
877 if (err == 0)
878 goto retry;
881 if (zap != NULL)
882 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
883 return (err);
887 fzap_length(zap_name_t *zn,
888 uint64_t *integer_size, uint64_t *num_integers)
890 zap_leaf_t *l;
891 int err;
892 zap_entry_handle_t zeh;
894 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
895 if (err != 0)
896 return (err);
897 err = zap_leaf_lookup(l, zn, &zeh);
898 if (err != 0)
899 goto out;
901 if (integer_size)
902 *integer_size = zeh.zeh_integer_size;
903 if (num_integers)
904 *num_integers = zeh.zeh_num_integers;
905 out:
906 zap_put_leaf(l);
907 return (err);
911 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
913 zap_leaf_t *l;
914 int err;
915 zap_entry_handle_t zeh;
917 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
918 if (err != 0)
919 return (err);
920 err = zap_leaf_lookup(l, zn, &zeh);
921 if (err == 0) {
922 zap_entry_remove(&zeh);
923 zap_increment_num_entries(zn->zn_zap, -1, tx);
925 zap_put_leaf(l);
926 return (err);
929 void
930 fzap_prefetch(zap_name_t *zn)
932 uint64_t idx, blk;
933 zap_t *zap = zn->zn_zap;
934 int bs;
936 idx = ZAP_HASH_IDX(zn->zn_hash,
937 zap_f_phys(zap)->zap_ptrtbl.zt_shift);
938 if (zap_idx_to_blk(zap, idx, &blk) != 0)
939 return;
940 bs = FZAP_BLOCK_SHIFT(zap);
941 dmu_prefetch(zap->zap_objset, zap->zap_object, blk << bs, 1 << bs);
945 * Helper functions for consumers.
948 uint64_t
949 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
950 const char *name, dmu_tx_t *tx)
952 uint64_t new_obj;
954 VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
955 VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
956 tx) == 0);
958 return (new_obj);
962 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
963 char *name)
965 zap_cursor_t zc;
966 zap_attribute_t *za;
967 int err;
969 if (mask == 0)
970 mask = -1ULL;
972 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
973 for (zap_cursor_init(&zc, os, zapobj);
974 (err = zap_cursor_retrieve(&zc, za)) == 0;
975 zap_cursor_advance(&zc)) {
976 if ((za->za_first_integer & mask) == (value & mask)) {
977 (void) strcpy(name, za->za_name);
978 break;
981 zap_cursor_fini(&zc);
982 kmem_free(za, sizeof (zap_attribute_t));
983 return (err);
987 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
989 zap_cursor_t zc;
990 zap_attribute_t za;
991 int err;
993 err = 0;
994 for (zap_cursor_init(&zc, os, fromobj);
995 zap_cursor_retrieve(&zc, &za) == 0;
996 (void) zap_cursor_advance(&zc)) {
997 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
998 err = SET_ERROR(EINVAL);
999 break;
1001 err = zap_add(os, intoobj, za.za_name,
1002 8, 1, &za.za_first_integer, tx);
1003 if (err)
1004 break;
1006 zap_cursor_fini(&zc);
1007 return (err);
1011 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1012 uint64_t value, dmu_tx_t *tx)
1014 zap_cursor_t zc;
1015 zap_attribute_t za;
1016 int err;
1018 err = 0;
1019 for (zap_cursor_init(&zc, os, fromobj);
1020 zap_cursor_retrieve(&zc, &za) == 0;
1021 (void) zap_cursor_advance(&zc)) {
1022 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1023 err = SET_ERROR(EINVAL);
1024 break;
1026 err = zap_add(os, intoobj, za.za_name,
1027 8, 1, &value, tx);
1028 if (err)
1029 break;
1031 zap_cursor_fini(&zc);
1032 return (err);
1036 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1037 dmu_tx_t *tx)
1039 zap_cursor_t zc;
1040 zap_attribute_t za;
1041 int err;
1043 err = 0;
1044 for (zap_cursor_init(&zc, os, fromobj);
1045 zap_cursor_retrieve(&zc, &za) == 0;
1046 (void) zap_cursor_advance(&zc)) {
1047 uint64_t delta = 0;
1049 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1050 err = SET_ERROR(EINVAL);
1051 break;
1054 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1055 if (err != 0 && err != ENOENT)
1056 break;
1057 delta += za.za_first_integer;
1058 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1059 if (err)
1060 break;
1062 zap_cursor_fini(&zc);
1063 return (err);
1067 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1069 char name[20];
1071 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1072 return (zap_add(os, obj, name, 8, 1, &value, tx));
1076 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1078 char name[20];
1080 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1081 return (zap_remove(os, obj, name, tx));
1085 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1087 char name[20];
1089 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1090 return (zap_lookup(os, obj, name, 8, 1, &value));
1094 zap_add_int_key(objset_t *os, uint64_t obj,
1095 uint64_t key, uint64_t value, dmu_tx_t *tx)
1097 char name[20];
1099 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1100 return (zap_add(os, obj, name, 8, 1, &value, tx));
1104 zap_update_int_key(objset_t *os, uint64_t obj,
1105 uint64_t key, uint64_t value, dmu_tx_t *tx)
1107 char name[20];
1109 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1110 return (zap_update(os, obj, name, 8, 1, &value, tx));
1114 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1116 char name[20];
1118 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1119 return (zap_lookup(os, obj, name, 8, 1, valuep));
1123 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1124 dmu_tx_t *tx)
1126 uint64_t value = 0;
1127 int err;
1129 if (delta == 0)
1130 return (0);
1132 err = zap_lookup(os, obj, name, 8, 1, &value);
1133 if (err != 0 && err != ENOENT)
1134 return (err);
1135 value += delta;
1136 if (value == 0)
1137 err = zap_remove(os, obj, name, tx);
1138 else
1139 err = zap_update(os, obj, name, 8, 1, &value, tx);
1140 return (err);
1144 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1145 dmu_tx_t *tx)
1147 char name[20];
1149 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1150 return (zap_increment(os, obj, name, delta, tx));
1154 * Routines for iterating over the attributes.
1158 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1160 int err = ENOENT;
1161 zap_entry_handle_t zeh;
1162 zap_leaf_t *l;
1164 /* retrieve the next entry at or after zc_hash/zc_cd */
1165 /* if no entry, return ENOENT */
1167 if (zc->zc_leaf &&
1168 (ZAP_HASH_IDX(zc->zc_hash,
1169 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1170 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1171 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1172 zap_put_leaf(zc->zc_leaf);
1173 zc->zc_leaf = NULL;
1176 again:
1177 if (zc->zc_leaf == NULL) {
1178 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1179 &zc->zc_leaf);
1180 if (err != 0)
1181 return (err);
1182 } else {
1183 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1185 l = zc->zc_leaf;
1187 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1189 if (err == ENOENT) {
1190 uint64_t nocare =
1191 (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1192 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1193 zc->zc_cd = 0;
1194 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 ||
1195 zc->zc_hash == 0) {
1196 zc->zc_hash = -1ULL;
1197 } else {
1198 zap_put_leaf(zc->zc_leaf);
1199 zc->zc_leaf = NULL;
1200 goto again;
1204 if (err == 0) {
1205 zc->zc_hash = zeh.zeh_hash;
1206 zc->zc_cd = zeh.zeh_cd;
1207 za->za_integer_length = zeh.zeh_integer_size;
1208 za->za_num_integers = zeh.zeh_num_integers;
1209 if (zeh.zeh_num_integers == 0) {
1210 za->za_first_integer = 0;
1211 } else {
1212 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1213 ASSERT(err == 0 || err == EOVERFLOW);
1215 err = zap_entry_read_name(zap, &zeh,
1216 sizeof (za->za_name), za->za_name);
1217 ASSERT(err == 0);
1219 za->za_normalization_conflict =
1220 zap_entry_normalization_conflict(&zeh,
1221 NULL, za->za_name, zap);
1223 rw_exit(&zc->zc_leaf->l_rwlock);
1224 return (err);
1227 static void
1228 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1230 int i, err;
1231 uint64_t lastblk = 0;
1234 * NB: if a leaf has more pointers than an entire ptrtbl block
1235 * can hold, then it'll be accounted for more than once, since
1236 * we won't have lastblk.
1238 for (i = 0; i < len; i++) {
1239 zap_leaf_t *l;
1241 if (tbl[i] == lastblk)
1242 continue;
1243 lastblk = tbl[i];
1245 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1246 if (err == 0) {
1247 zap_leaf_stats(zap, l, zs);
1248 zap_put_leaf(l);
1253 void
1254 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1256 int bs = FZAP_BLOCK_SHIFT(zap);
1257 zs->zs_blocksize = 1ULL << bs;
1260 * Set zap_phys_t fields
1262 zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1263 zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1264 zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1265 zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1266 zs->zs_magic = zap_f_phys(zap)->zap_magic;
1267 zs->zs_salt = zap_f_phys(zap)->zap_salt;
1270 * Set zap_ptrtbl fields
1272 zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1273 zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1274 zs->zs_ptrtbl_blks_copied =
1275 zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1276 zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1277 zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1278 zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1280 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1281 /* the ptrtbl is entirely in the header block. */
1282 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1283 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1284 } else {
1285 int b;
1287 dmu_prefetch(zap->zap_objset, zap->zap_object,
1288 zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1289 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs);
1291 for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1292 b++) {
1293 dmu_buf_t *db;
1294 int err;
1296 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1297 (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1298 FTAG, &db, DMU_READ_NO_PREFETCH);
1299 if (err == 0) {
1300 zap_stats_ptrtbl(zap, db->db_data,
1301 1<<(bs-3), zs);
1302 dmu_buf_rele(db, FTAG);
1309 fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite,
1310 uint64_t *tooverwrite)
1312 zap_t *zap = zn->zn_zap;
1313 zap_leaf_t *l;
1314 int err;
1317 * Account for the header block of the fatzap.
1319 if (!add && dmu_buf_freeable(zap->zap_dbuf)) {
1320 *tooverwrite += zap->zap_dbuf->db_size;
1321 } else {
1322 *towrite += zap->zap_dbuf->db_size;
1326 * Account for the pointer table blocks.
1327 * If we are adding we need to account for the following cases :
1328 * - If the pointer table is embedded, this operation could force an
1329 * external pointer table.
1330 * - If this already has an external pointer table this operation
1331 * could extend the table.
1333 if (add) {
1334 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0)
1335 *towrite += zap->zap_dbuf->db_size;
1336 else
1337 *towrite += (zap->zap_dbuf->db_size * 3);
1341 * Now, check if the block containing leaf is freeable
1342 * and account accordingly.
1344 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l);
1345 if (err != 0) {
1346 return (err);
1349 if (!add && dmu_buf_freeable(l->l_dbuf)) {
1350 *tooverwrite += l->l_dbuf->db_size;
1351 } else {
1353 * If this an add operation, the leaf block could split.
1354 * Hence, we need to account for an additional leaf block.
1356 *towrite += (add ? 2 : 1) * l->l_dbuf->db_size;
1359 zap_put_leaf(l);
1360 return (0);