5056 ZFS deadlock on db_mtx and dn_holds
[illumos-gate.git] / usr / src / uts / common / fs / zfs / zap.c
blob20f349651000c7149bee85e7a98b308c176799ac
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.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * This file contains the top half of the zfs directory structure
29 * implementation. The bottom half is in zap_leaf.c.
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.
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
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>
52 int fzap_default_block_shift = 14; /* 16k blocksize */
54 extern inline zap_phys_t *zap_f_phys(zap_t *zap);
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 zap->zap_dbu.dbu_evict_func = 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 void
391 zap_leaf_pageout(void *dbu)
393 zap_leaf_t *l = dbu;
395 rw_destroy(&l->l_rwlock);
396 kmem_free(l, sizeof (zap_leaf_t));
399 static zap_leaf_t *
400 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
402 void *winner;
403 zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
405 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
407 rw_init(&l->l_rwlock, 0, 0, 0);
408 rw_enter(&l->l_rwlock, RW_WRITER);
409 l->l_blkid = zap_allocate_blocks(zap, 1);
410 l->l_dbuf = NULL;
412 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
413 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
414 DMU_READ_NO_PREFETCH));
415 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
416 winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu);
417 ASSERT(winner == NULL);
418 dmu_buf_will_dirty(l->l_dbuf, tx);
420 zap_leaf_init(l, zap->zap_normflags != 0);
422 zap_f_phys(zap)->zap_num_leafs++;
424 return (l);
428 fzap_count(zap_t *zap, uint64_t *count)
430 ASSERT(!zap->zap_ismicro);
431 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
432 *count = zap_f_phys(zap)->zap_num_entries;
433 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
434 return (0);
438 * Routines for obtaining zap_leaf_t's
441 void
442 zap_put_leaf(zap_leaf_t *l)
444 rw_exit(&l->l_rwlock);
445 dmu_buf_rele(l->l_dbuf, NULL);
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_zalloc(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 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
463 winner = dmu_buf_set_user(db, &l->l_dbu);
465 rw_exit(&l->l_rwlock);
466 if (winner != NULL) {
467 /* someone else set it first */
468 zap_leaf_pageout(&l->l_dbu);
469 l = winner;
473 * lhr_pad was previously used for the next leaf in the leaf
474 * chain. There should be no chained leafs (as we have removed
475 * support for them).
477 ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1);
480 * There should be more hash entries than there can be
481 * chunks to put in the hash table
483 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
485 /* The chunks should begin at the end of the hash table */
486 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
487 &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
489 /* The chunks should end at the end of the block */
490 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
491 (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size);
493 return (l);
496 static int
497 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
498 zap_leaf_t **lp)
500 dmu_buf_t *db;
501 zap_leaf_t *l;
502 int bs = FZAP_BLOCK_SHIFT(zap);
503 int err;
505 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
507 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
508 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
509 if (err)
510 return (err);
512 ASSERT3U(db->db_object, ==, zap->zap_object);
513 ASSERT3U(db->db_offset, ==, blkid << bs);
514 ASSERT3U(db->db_size, ==, 1 << bs);
515 ASSERT(blkid != 0);
517 l = dmu_buf_get_user(db);
519 if (l == NULL)
520 l = zap_open_leaf(blkid, db);
522 rw_enter(&l->l_rwlock, lt);
524 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
525 * causing ASSERT below to fail.
527 if (lt == RW_WRITER)
528 dmu_buf_will_dirty(db, tx);
529 ASSERT3U(l->l_blkid, ==, blkid);
530 ASSERT3P(l->l_dbuf, ==, db);
531 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
532 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
534 *lp = l;
535 return (0);
538 static int
539 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
541 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
543 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
544 ASSERT3U(idx, <,
545 (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
546 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
547 return (0);
548 } else {
549 return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
550 idx, valp));
554 static int
555 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
557 ASSERT(tx != NULL);
558 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
560 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
561 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
562 return (0);
563 } else {
564 return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
565 idx, blk, tx));
569 static int
570 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
572 uint64_t idx, blk;
573 int err;
575 ASSERT(zap->zap_dbuf == NULL ||
576 zap_f_phys(zap) == zap->zap_dbuf->db_data);
577 ASSERT3U(zap_f_phys(zap)->zap_magic, ==, ZAP_MAGIC);
578 idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
579 err = zap_idx_to_blk(zap, idx, &blk);
580 if (err != 0)
581 return (err);
582 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
584 ASSERT(err ||
585 ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
586 zap_leaf_phys(*lp)->l_hdr.lh_prefix);
587 return (err);
590 static int
591 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx, zap_leaf_t **lp)
593 zap_t *zap = zn->zn_zap;
594 uint64_t hash = zn->zn_hash;
595 zap_leaf_t *nl;
596 int prefix_diff, i, err;
597 uint64_t sibling;
598 int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len;
600 ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
601 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
603 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
604 zap_leaf_phys(l)->l_hdr.lh_prefix);
606 if (zap_tryupgradedir(zap, tx) == 0 ||
607 old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
608 /* We failed to upgrade, or need to grow the pointer table */
609 objset_t *os = zap->zap_objset;
610 uint64_t object = zap->zap_object;
612 zap_put_leaf(l);
613 zap_unlockdir(zap);
614 err = zap_lockdir(os, object, tx, RW_WRITER,
615 FALSE, FALSE, &zn->zn_zap);
616 zap = zn->zn_zap;
617 if (err)
618 return (err);
619 ASSERT(!zap->zap_ismicro);
621 while (old_prefix_len ==
622 zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
623 err = zap_grow_ptrtbl(zap, tx);
624 if (err)
625 return (err);
628 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
629 if (err)
630 return (err);
632 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) {
633 /* it split while our locks were down */
634 *lp = l;
635 return (0);
638 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
639 ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
640 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
641 zap_leaf_phys(l)->l_hdr.lh_prefix);
643 prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
644 (old_prefix_len + 1);
645 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
647 /* check for i/o errors before doing zap_leaf_split */
648 for (i = 0; i < (1ULL<<prefix_diff); i++) {
649 uint64_t blk;
650 err = zap_idx_to_blk(zap, sibling+i, &blk);
651 if (err)
652 return (err);
653 ASSERT3U(blk, ==, l->l_blkid);
656 nl = zap_create_leaf(zap, tx);
657 zap_leaf_split(l, nl, zap->zap_normflags != 0);
659 /* set sibling pointers */
660 for (i = 0; i < (1ULL << prefix_diff); i++) {
661 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
662 ASSERT0(err); /* we checked for i/o errors above */
665 if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) {
666 /* we want the sibling */
667 zap_put_leaf(l);
668 *lp = nl;
669 } else {
670 zap_put_leaf(nl);
671 *lp = l;
674 return (0);
677 static void
678 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx)
680 zap_t *zap = zn->zn_zap;
681 int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
682 int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
683 zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
685 zap_put_leaf(l);
687 if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
688 int err;
691 * We are in the middle of growing the pointer table, or
692 * this leaf will soon make us grow it.
694 if (zap_tryupgradedir(zap, tx) == 0) {
695 objset_t *os = zap->zap_objset;
696 uint64_t zapobj = zap->zap_object;
698 zap_unlockdir(zap);
699 err = zap_lockdir(os, zapobj, tx,
700 RW_WRITER, FALSE, FALSE, &zn->zn_zap);
701 zap = zn->zn_zap;
702 if (err)
703 return;
706 /* could have finished growing while our locks were down */
707 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
708 (void) zap_grow_ptrtbl(zap, tx);
712 static int
713 fzap_checkname(zap_name_t *zn)
715 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
716 return (SET_ERROR(ENAMETOOLONG));
717 return (0);
720 static int
721 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
723 /* Only integer sizes supported by C */
724 switch (integer_size) {
725 case 1:
726 case 2:
727 case 4:
728 case 8:
729 break;
730 default:
731 return (SET_ERROR(EINVAL));
734 if (integer_size * num_integers > ZAP_MAXVALUELEN)
735 return (E2BIG);
737 return (0);
740 static int
741 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
743 int err;
745 if ((err = fzap_checkname(zn)) != 0)
746 return (err);
747 return (fzap_checksize(integer_size, num_integers));
751 * Routines for manipulating attributes.
754 fzap_lookup(zap_name_t *zn,
755 uint64_t integer_size, uint64_t num_integers, void *buf,
756 char *realname, int rn_len, boolean_t *ncp)
758 zap_leaf_t *l;
759 int err;
760 zap_entry_handle_t zeh;
762 if ((err = fzap_checkname(zn)) != 0)
763 return (err);
765 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
766 if (err != 0)
767 return (err);
768 err = zap_leaf_lookup(l, zn, &zeh);
769 if (err == 0) {
770 if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
771 zap_put_leaf(l);
772 return (err);
775 err = zap_entry_read(&zeh, integer_size, num_integers, buf);
776 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
777 if (ncp) {
778 *ncp = zap_entry_normalization_conflict(&zeh,
779 zn, NULL, zn->zn_zap);
783 zap_put_leaf(l);
784 return (err);
788 fzap_add_cd(zap_name_t *zn,
789 uint64_t integer_size, uint64_t num_integers,
790 const void *val, uint32_t cd, dmu_tx_t *tx)
792 zap_leaf_t *l;
793 int err;
794 zap_entry_handle_t zeh;
795 zap_t *zap = zn->zn_zap;
797 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
798 ASSERT(!zap->zap_ismicro);
799 ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
801 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
802 if (err != 0)
803 return (err);
804 retry:
805 err = zap_leaf_lookup(l, zn, &zeh);
806 if (err == 0) {
807 err = SET_ERROR(EEXIST);
808 goto out;
810 if (err != ENOENT)
811 goto out;
813 err = zap_entry_create(l, zn, cd,
814 integer_size, num_integers, val, &zeh);
816 if (err == 0) {
817 zap_increment_num_entries(zap, 1, tx);
818 } else if (err == EAGAIN) {
819 err = zap_expand_leaf(zn, l, tx, &l);
820 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
821 if (err == 0)
822 goto retry;
825 out:
826 if (zap != NULL)
827 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
828 return (err);
832 fzap_add(zap_name_t *zn,
833 uint64_t integer_size, uint64_t num_integers,
834 const void *val, dmu_tx_t *tx)
836 int err = fzap_check(zn, integer_size, num_integers);
837 if (err != 0)
838 return (err);
840 return (fzap_add_cd(zn, integer_size, num_integers,
841 val, ZAP_NEED_CD, tx));
845 fzap_update(zap_name_t *zn,
846 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
848 zap_leaf_t *l;
849 int err, create;
850 zap_entry_handle_t zeh;
851 zap_t *zap = zn->zn_zap;
853 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
854 err = fzap_check(zn, integer_size, num_integers);
855 if (err != 0)
856 return (err);
858 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
859 if (err != 0)
860 return (err);
861 retry:
862 err = zap_leaf_lookup(l, zn, &zeh);
863 create = (err == ENOENT);
864 ASSERT(err == 0 || err == ENOENT);
866 if (create) {
867 err = zap_entry_create(l, zn, ZAP_NEED_CD,
868 integer_size, num_integers, val, &zeh);
869 if (err == 0)
870 zap_increment_num_entries(zap, 1, tx);
871 } else {
872 err = zap_entry_update(&zeh, integer_size, num_integers, val);
875 if (err == EAGAIN) {
876 err = zap_expand_leaf(zn, l, tx, &l);
877 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
878 if (err == 0)
879 goto retry;
882 if (zap != NULL)
883 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
884 return (err);
888 fzap_length(zap_name_t *zn,
889 uint64_t *integer_size, uint64_t *num_integers)
891 zap_leaf_t *l;
892 int err;
893 zap_entry_handle_t zeh;
895 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
896 if (err != 0)
897 return (err);
898 err = zap_leaf_lookup(l, zn, &zeh);
899 if (err != 0)
900 goto out;
902 if (integer_size)
903 *integer_size = zeh.zeh_integer_size;
904 if (num_integers)
905 *num_integers = zeh.zeh_num_integers;
906 out:
907 zap_put_leaf(l);
908 return (err);
912 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
914 zap_leaf_t *l;
915 int err;
916 zap_entry_handle_t zeh;
918 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
919 if (err != 0)
920 return (err);
921 err = zap_leaf_lookup(l, zn, &zeh);
922 if (err == 0) {
923 zap_entry_remove(&zeh);
924 zap_increment_num_entries(zn->zn_zap, -1, tx);
926 zap_put_leaf(l);
927 return (err);
930 void
931 fzap_prefetch(zap_name_t *zn)
933 uint64_t idx, blk;
934 zap_t *zap = zn->zn_zap;
935 int bs;
937 idx = ZAP_HASH_IDX(zn->zn_hash,
938 zap_f_phys(zap)->zap_ptrtbl.zt_shift);
939 if (zap_idx_to_blk(zap, idx, &blk) != 0)
940 return;
941 bs = FZAP_BLOCK_SHIFT(zap);
942 dmu_prefetch(zap->zap_objset, zap->zap_object, blk << bs, 1 << bs);
946 * Helper functions for consumers.
949 uint64_t
950 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
951 const char *name, dmu_tx_t *tx)
953 uint64_t new_obj;
955 VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
956 VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
957 tx) == 0);
959 return (new_obj);
963 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
964 char *name)
966 zap_cursor_t zc;
967 zap_attribute_t *za;
968 int err;
970 if (mask == 0)
971 mask = -1ULL;
973 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
974 for (zap_cursor_init(&zc, os, zapobj);
975 (err = zap_cursor_retrieve(&zc, za)) == 0;
976 zap_cursor_advance(&zc)) {
977 if ((za->za_first_integer & mask) == (value & mask)) {
978 (void) strcpy(name, za->za_name);
979 break;
982 zap_cursor_fini(&zc);
983 kmem_free(za, sizeof (zap_attribute_t));
984 return (err);
988 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
990 zap_cursor_t zc;
991 zap_attribute_t za;
992 int err;
994 err = 0;
995 for (zap_cursor_init(&zc, os, fromobj);
996 zap_cursor_retrieve(&zc, &za) == 0;
997 (void) zap_cursor_advance(&zc)) {
998 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
999 err = SET_ERROR(EINVAL);
1000 break;
1002 err = zap_add(os, intoobj, za.za_name,
1003 8, 1, &za.za_first_integer, tx);
1004 if (err)
1005 break;
1007 zap_cursor_fini(&zc);
1008 return (err);
1012 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1013 uint64_t value, dmu_tx_t *tx)
1015 zap_cursor_t zc;
1016 zap_attribute_t za;
1017 int err;
1019 err = 0;
1020 for (zap_cursor_init(&zc, os, fromobj);
1021 zap_cursor_retrieve(&zc, &za) == 0;
1022 (void) zap_cursor_advance(&zc)) {
1023 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1024 err = SET_ERROR(EINVAL);
1025 break;
1027 err = zap_add(os, intoobj, za.za_name,
1028 8, 1, &value, tx);
1029 if (err)
1030 break;
1032 zap_cursor_fini(&zc);
1033 return (err);
1037 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1038 dmu_tx_t *tx)
1040 zap_cursor_t zc;
1041 zap_attribute_t za;
1042 int err;
1044 err = 0;
1045 for (zap_cursor_init(&zc, os, fromobj);
1046 zap_cursor_retrieve(&zc, &za) == 0;
1047 (void) zap_cursor_advance(&zc)) {
1048 uint64_t delta = 0;
1050 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1051 err = SET_ERROR(EINVAL);
1052 break;
1055 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1056 if (err != 0 && err != ENOENT)
1057 break;
1058 delta += za.za_first_integer;
1059 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1060 if (err)
1061 break;
1063 zap_cursor_fini(&zc);
1064 return (err);
1068 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1070 char name[20];
1072 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1073 return (zap_add(os, obj, name, 8, 1, &value, tx));
1077 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1079 char name[20];
1081 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1082 return (zap_remove(os, obj, name, tx));
1086 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1088 char name[20];
1090 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1091 return (zap_lookup(os, obj, name, 8, 1, &value));
1095 zap_add_int_key(objset_t *os, uint64_t obj,
1096 uint64_t key, uint64_t value, dmu_tx_t *tx)
1098 char name[20];
1100 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1101 return (zap_add(os, obj, name, 8, 1, &value, tx));
1105 zap_update_int_key(objset_t *os, uint64_t obj,
1106 uint64_t key, uint64_t value, dmu_tx_t *tx)
1108 char name[20];
1110 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1111 return (zap_update(os, obj, name, 8, 1, &value, tx));
1115 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1117 char name[20];
1119 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1120 return (zap_lookup(os, obj, name, 8, 1, valuep));
1124 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1125 dmu_tx_t *tx)
1127 uint64_t value = 0;
1128 int err;
1130 if (delta == 0)
1131 return (0);
1133 err = zap_lookup(os, obj, name, 8, 1, &value);
1134 if (err != 0 && err != ENOENT)
1135 return (err);
1136 value += delta;
1137 if (value == 0)
1138 err = zap_remove(os, obj, name, tx);
1139 else
1140 err = zap_update(os, obj, name, 8, 1, &value, tx);
1141 return (err);
1145 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1146 dmu_tx_t *tx)
1148 char name[20];
1150 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1151 return (zap_increment(os, obj, name, delta, tx));
1155 * Routines for iterating over the attributes.
1159 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1161 int err = ENOENT;
1162 zap_entry_handle_t zeh;
1163 zap_leaf_t *l;
1165 /* retrieve the next entry at or after zc_hash/zc_cd */
1166 /* if no entry, return ENOENT */
1168 if (zc->zc_leaf &&
1169 (ZAP_HASH_IDX(zc->zc_hash,
1170 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1171 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1172 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1173 zap_put_leaf(zc->zc_leaf);
1174 zc->zc_leaf = NULL;
1177 again:
1178 if (zc->zc_leaf == NULL) {
1179 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1180 &zc->zc_leaf);
1181 if (err != 0)
1182 return (err);
1183 } else {
1184 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1186 l = zc->zc_leaf;
1188 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1190 if (err == ENOENT) {
1191 uint64_t nocare =
1192 (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1193 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1194 zc->zc_cd = 0;
1195 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 ||
1196 zc->zc_hash == 0) {
1197 zc->zc_hash = -1ULL;
1198 } else {
1199 zap_put_leaf(zc->zc_leaf);
1200 zc->zc_leaf = NULL;
1201 goto again;
1205 if (err == 0) {
1206 zc->zc_hash = zeh.zeh_hash;
1207 zc->zc_cd = zeh.zeh_cd;
1208 za->za_integer_length = zeh.zeh_integer_size;
1209 za->za_num_integers = zeh.zeh_num_integers;
1210 if (zeh.zeh_num_integers == 0) {
1211 za->za_first_integer = 0;
1212 } else {
1213 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1214 ASSERT(err == 0 || err == EOVERFLOW);
1216 err = zap_entry_read_name(zap, &zeh,
1217 sizeof (za->za_name), za->za_name);
1218 ASSERT(err == 0);
1220 za->za_normalization_conflict =
1221 zap_entry_normalization_conflict(&zeh,
1222 NULL, za->za_name, zap);
1224 rw_exit(&zc->zc_leaf->l_rwlock);
1225 return (err);
1228 static void
1229 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1231 int i, err;
1232 uint64_t lastblk = 0;
1235 * NB: if a leaf has more pointers than an entire ptrtbl block
1236 * can hold, then it'll be accounted for more than once, since
1237 * we won't have lastblk.
1239 for (i = 0; i < len; i++) {
1240 zap_leaf_t *l;
1242 if (tbl[i] == lastblk)
1243 continue;
1244 lastblk = tbl[i];
1246 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1247 if (err == 0) {
1248 zap_leaf_stats(zap, l, zs);
1249 zap_put_leaf(l);
1254 void
1255 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1257 int bs = FZAP_BLOCK_SHIFT(zap);
1258 zs->zs_blocksize = 1ULL << bs;
1261 * Set zap_phys_t fields
1263 zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1264 zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1265 zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1266 zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1267 zs->zs_magic = zap_f_phys(zap)->zap_magic;
1268 zs->zs_salt = zap_f_phys(zap)->zap_salt;
1271 * Set zap_ptrtbl fields
1273 zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1274 zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1275 zs->zs_ptrtbl_blks_copied =
1276 zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1277 zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1278 zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1279 zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1281 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1282 /* the ptrtbl is entirely in the header block. */
1283 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1284 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1285 } else {
1286 int b;
1288 dmu_prefetch(zap->zap_objset, zap->zap_object,
1289 zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1290 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs);
1292 for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1293 b++) {
1294 dmu_buf_t *db;
1295 int err;
1297 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1298 (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1299 FTAG, &db, DMU_READ_NO_PREFETCH);
1300 if (err == 0) {
1301 zap_stats_ptrtbl(zap, db->db_data,
1302 1<<(bs-3), zs);
1303 dmu_buf_rele(db, FTAG);
1310 fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite,
1311 uint64_t *tooverwrite)
1313 zap_t *zap = zn->zn_zap;
1314 zap_leaf_t *l;
1315 int err;
1318 * Account for the header block of the fatzap.
1320 if (!add && dmu_buf_freeable(zap->zap_dbuf)) {
1321 *tooverwrite += zap->zap_dbuf->db_size;
1322 } else {
1323 *towrite += zap->zap_dbuf->db_size;
1327 * Account for the pointer table blocks.
1328 * If we are adding we need to account for the following cases :
1329 * - If the pointer table is embedded, this operation could force an
1330 * external pointer table.
1331 * - If this already has an external pointer table this operation
1332 * could extend the table.
1334 if (add) {
1335 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0)
1336 *towrite += zap->zap_dbuf->db_size;
1337 else
1338 *towrite += (zap->zap_dbuf->db_size * 3);
1342 * Now, check if the block containing leaf is freeable
1343 * and account accordingly.
1345 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l);
1346 if (err != 0) {
1347 return (err);
1350 if (!add && dmu_buf_freeable(l->l_dbuf)) {
1351 *tooverwrite += l->l_dbuf->db_size;
1352 } else {
1354 * If this an add operation, the leaf block could split.
1355 * Hence, we need to account for an additional leaf block.
1357 *towrite += (add ? 2 : 1) * l->l_dbuf->db_size;
1360 zap_put_leaf(l);
1361 return (0);