5056 ZFS deadlock on db_mtx and dn_holds
[illumos-gate.git] / usr / src / uts / common / fs / zfs / sys / zap_impl.h
blobacc795cbc44eea590d56c2a3c4be4a6b320513bf
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) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 #ifndef _SYS_ZAP_IMPL_H
28 #define _SYS_ZAP_IMPL_H
30 #include <sys/zap.h>
31 #include <sys/zfs_context.h>
32 #include <sys/avl.h>
34 #ifdef __cplusplus
35 extern "C" {
36 #endif
38 extern int fzap_default_block_shift;
40 #define ZAP_MAGIC 0x2F52AB2ABULL
42 #define FZAP_BLOCK_SHIFT(zap) ((zap)->zap_f.zap_block_shift)
44 #define MZAP_ENT_LEN 64
45 #define MZAP_NAME_LEN (MZAP_ENT_LEN - 8 - 4 - 2)
46 #define MZAP_MAX_BLKSZ SPA_OLD_MAXBLOCKSIZE
48 #define ZAP_NEED_CD (-1U)
50 typedef struct mzap_ent_phys {
51 uint64_t mze_value;
52 uint32_t mze_cd;
53 uint16_t mze_pad; /* in case we want to chain them someday */
54 char mze_name[MZAP_NAME_LEN];
55 } mzap_ent_phys_t;
57 typedef struct mzap_phys {
58 uint64_t mz_block_type; /* ZBT_MICRO */
59 uint64_t mz_salt;
60 uint64_t mz_normflags;
61 uint64_t mz_pad[5];
62 mzap_ent_phys_t mz_chunk[1];
63 /* actually variable size depending on block size */
64 } mzap_phys_t;
66 typedef struct mzap_ent {
67 avl_node_t mze_node;
68 int mze_chunkid;
69 uint64_t mze_hash;
70 uint32_t mze_cd; /* copy from mze_phys->mze_cd */
71 } mzap_ent_t;
73 #define MZE_PHYS(zap, mze) \
74 (&zap_m_phys(zap)->mz_chunk[(mze)->mze_chunkid])
77 * The (fat) zap is stored in one object. It is an array of
78 * 1<<FZAP_BLOCK_SHIFT byte blocks. The layout looks like one of:
80 * ptrtbl fits in first block:
81 * [zap_phys_t zap_ptrtbl_shift < 6] [zap_leaf_t] ...
83 * ptrtbl too big for first block:
84 * [zap_phys_t zap_ptrtbl_shift >= 6] [zap_leaf_t] [ptrtbl] ...
88 struct dmu_buf;
89 struct zap_leaf;
91 #define ZBT_LEAF ((1ULL << 63) + 0)
92 #define ZBT_HEADER ((1ULL << 63) + 1)
93 #define ZBT_MICRO ((1ULL << 63) + 3)
94 /* any other values are ptrtbl blocks */
97 * the embedded pointer table takes up half a block:
98 * block size / entry size (2^3) / 2
100 #define ZAP_EMBEDDED_PTRTBL_SHIFT(zap) (FZAP_BLOCK_SHIFT(zap) - 3 - 1)
103 * The embedded pointer table starts half-way through the block. Since
104 * the pointer table itself is half the block, it starts at (64-bit)
105 * word number (1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap)).
107 #define ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) \
108 ((uint64_t *)zap_f_phys(zap)) \
109 [(idx) + (1<<ZAP_EMBEDDED_PTRTBL_SHIFT(zap))]
112 * TAKE NOTE:
113 * If zap_phys_t is modified, zap_byteswap() must be modified.
115 typedef struct zap_phys {
116 uint64_t zap_block_type; /* ZBT_HEADER */
117 uint64_t zap_magic; /* ZAP_MAGIC */
119 struct zap_table_phys {
120 uint64_t zt_blk; /* starting block number */
121 uint64_t zt_numblks; /* number of blocks */
122 uint64_t zt_shift; /* bits to index it */
123 uint64_t zt_nextblk; /* next (larger) copy start block */
124 uint64_t zt_blks_copied; /* number source blocks copied */
125 } zap_ptrtbl;
127 uint64_t zap_freeblk; /* the next free block */
128 uint64_t zap_num_leafs; /* number of leafs */
129 uint64_t zap_num_entries; /* number of entries */
130 uint64_t zap_salt; /* salt to stir into hash function */
131 uint64_t zap_normflags; /* flags for u8_textprep_str() */
132 uint64_t zap_flags; /* zap_flags_t */
134 * This structure is followed by padding, and then the embedded
135 * pointer table. The embedded pointer table takes up second
136 * half of the block. It is accessed using the
137 * ZAP_EMBEDDED_PTRTBL_ENT() macro.
139 } zap_phys_t;
141 typedef struct zap_table_phys zap_table_phys_t;
143 typedef struct zap {
144 dmu_buf_user_t zap_dbu;
145 objset_t *zap_objset;
146 uint64_t zap_object;
147 struct dmu_buf *zap_dbuf;
148 krwlock_t zap_rwlock;
149 boolean_t zap_ismicro;
150 int zap_normflags;
151 uint64_t zap_salt;
152 union {
153 struct {
155 * zap_num_entries_mtx protects
156 * zap_num_entries
158 kmutex_t zap_num_entries_mtx;
159 int zap_block_shift;
160 } zap_fat;
161 struct {
162 int16_t zap_num_entries;
163 int16_t zap_num_chunks;
164 int16_t zap_alloc_next;
165 avl_tree_t zap_avl;
166 } zap_micro;
167 } zap_u;
168 } zap_t;
170 inline zap_phys_t *
171 zap_f_phys(zap_t *zap)
173 return (zap->zap_dbuf->db_data);
176 inline mzap_phys_t *
177 zap_m_phys(zap_t *zap)
179 return (zap->zap_dbuf->db_data);
182 typedef struct zap_name {
183 zap_t *zn_zap;
184 int zn_key_intlen;
185 const void *zn_key_orig;
186 int zn_key_orig_numints;
187 const void *zn_key_norm;
188 int zn_key_norm_numints;
189 uint64_t zn_hash;
190 matchtype_t zn_matchtype;
191 char zn_normbuf[ZAP_MAXNAMELEN];
192 } zap_name_t;
194 #define zap_f zap_u.zap_fat
195 #define zap_m zap_u.zap_micro
197 boolean_t zap_match(zap_name_t *zn, const char *matchname);
198 int zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
199 krw_t lti, boolean_t fatreader, boolean_t adding, zap_t **zapp);
200 void zap_unlockdir(zap_t *zap);
201 void zap_evict(void *dbu);
202 zap_name_t *zap_name_alloc(zap_t *zap, const char *key, matchtype_t mt);
203 void zap_name_free(zap_name_t *zn);
204 int zap_hashbits(zap_t *zap);
205 uint32_t zap_maxcd(zap_t *zap);
206 uint64_t zap_getflags(zap_t *zap);
208 #define ZAP_HASH_IDX(hash, n) (((n) == 0) ? 0 : ((hash) >> (64 - (n))))
210 void fzap_byteswap(void *buf, size_t size);
211 int fzap_count(zap_t *zap, uint64_t *count);
212 int fzap_lookup(zap_name_t *zn,
213 uint64_t integer_size, uint64_t num_integers, void *buf,
214 char *realname, int rn_len, boolean_t *normalization_conflictp);
215 void fzap_prefetch(zap_name_t *zn);
216 int fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite,
217 uint64_t *tooverwrite);
218 int fzap_add(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers,
219 const void *val, dmu_tx_t *tx);
220 int fzap_update(zap_name_t *zn,
221 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx);
222 int fzap_length(zap_name_t *zn,
223 uint64_t *integer_size, uint64_t *num_integers);
224 int fzap_remove(zap_name_t *zn, dmu_tx_t *tx);
225 int fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za);
226 void fzap_get_stats(zap_t *zap, zap_stats_t *zs);
227 void zap_put_leaf(struct zap_leaf *l);
229 int fzap_add_cd(zap_name_t *zn,
230 uint64_t integer_size, uint64_t num_integers,
231 const void *val, uint32_t cd, dmu_tx_t *tx);
232 void fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags);
234 #ifdef __cplusplus
236 #endif
238 #endif /* _SYS_ZAP_IMPL_H */