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]
22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
29 #pragma ident "%Z%%M% %I% %E% SMI"
31 #include <sys/types.h>
43 * Each objset has its own intent log. The log header (zil_header_t)
44 * for objset N's intent log is kept in the Nth object of the SPA's
45 * intent_log objset. The log header points to a chain of log blocks,
46 * each of which contains log records (i.e., transactions) followed by
47 * a log block trailer (zil_trailer_t). The format of a log record
48 * depends on the record (or transaction) type, but all records begin
49 * with a common structure that defines the type, length, and txg.
53 * Intent log header - this on disk structure holds fields to manage
54 * the log. All fields are 64 bit to easily handle cross architectures.
56 typedef struct zil_header {
57 uint64_t zh_claim_txg; /* txg in which log blocks were claimed */
58 uint64_t zh_replay_seq; /* highest replayed sequence number */
59 blkptr_t zh_log; /* log chain */
60 uint64_t zh_claim_seq; /* highest claimed sequence number */
65 * Log block trailer - structure at the end of the header and each log block
67 * The zit_bt contains a zbt_cksum which for the intent log is
68 * the sequence number of this log block. A seq of 0 is invalid.
69 * The zbt_cksum is checked by the SPA against the sequence
70 * number passed in the blk_cksum field of the blkptr_t
72 typedef struct zil_trailer {
74 blkptr_t zit_next_blk; /* next block in chain */
75 uint64_t zit_nused; /* bytes in log block used */
76 zio_block_tail_t zit_bt; /* block trailer */
79 #define ZIL_MIN_BLKSZ 4096ULL
80 #define ZIL_MAX_BLKSZ SPA_MAXBLOCKSIZE
81 #define ZIL_BLK_DATA_SZ(lwb) ((lwb)->lwb_sz - sizeof (zil_trailer_t))
84 * The words of a log block checksum.
86 #define ZIL_ZC_GUID_0 0
87 #define ZIL_ZC_GUID_1 1
88 #define ZIL_ZC_OBJSET 2
92 * Intent log transaction types and record structures
94 #define TX_CREATE 1 /* Create file */
95 #define TX_MKDIR 2 /* Make directory */
96 #define TX_MKXATTR 3 /* Make XATTR directory */
97 #define TX_SYMLINK 4 /* Create symbolic link to a file */
98 #define TX_REMOVE 5 /* Remove file */
99 #define TX_RMDIR 6 /* Remove directory */
100 #define TX_LINK 7 /* Create hard link to a file */
101 #define TX_RENAME 8 /* Rename a file */
102 #define TX_WRITE 9 /* File write */
103 #define TX_TRUNCATE 10 /* Truncate a file */
104 #define TX_SETATTR 11 /* Set file attributes */
105 #define TX_ACL 12 /* Set acl */
106 #define TX_MAX_TYPE 13 /* Max transaction type */
109 * Format of log records.
110 * The fields are carefully defined to allow them to be aligned
111 * and sized the same on sparc & intel architectures.
112 * Each log record has a common structure at the beginning.
114 * Note, lrc_seq holds two different sequence numbers. Whilst in memory
115 * it contains the transaction sequence number. The log record on
116 * disk holds the sequence number of all log records which is used to
117 * ensure we don't replay the same record. The two sequence numbers are
118 * different because the transactions can now be pushed out of order.
120 typedef struct { /* common log record header */
121 uint64_t lrc_txtype; /* intent log transaction type */
122 uint64_t lrc_reclen; /* transaction record length */
123 uint64_t lrc_txg; /* dmu transaction group number */
124 uint64_t lrc_seq; /* see comment above */
128 lr_t lr_common; /* common portion of log record */
129 uint64_t lr_doid; /* object id of directory */
130 uint64_t lr_foid; /* object id of created file object */
131 uint64_t lr_mode; /* mode of object */
132 uint64_t lr_uid; /* uid of object */
133 uint64_t lr_gid; /* gid of object */
134 uint64_t lr_gen; /* generation (txg of creation) */
135 uint64_t lr_crtime[2]; /* creation time */
136 uint64_t lr_rdev; /* rdev of object to create */
137 /* name of object to create follows this */
138 /* for symlinks, link content follows name */
142 lr_t lr_common; /* common portion of log record */
143 uint64_t lr_doid; /* obj id of directory */
144 /* name of object to remove follows this */
148 lr_t lr_common; /* common portion of log record */
149 uint64_t lr_doid; /* obj id of directory */
150 uint64_t lr_link_obj; /* obj id of link */
151 /* name of object to link follows this */
155 lr_t lr_common; /* common portion of log record */
156 uint64_t lr_sdoid; /* obj id of source directory */
157 uint64_t lr_tdoid; /* obj id of target directory */
158 /* 2 strings: names of source and destination follow this */
162 lr_t lr_common; /* common portion of log record */
163 uint64_t lr_foid; /* file object to write */
164 uint64_t lr_offset; /* offset to write to */
165 uint64_t lr_length; /* user data length to write */
166 uint64_t lr_blkoff; /* offset represented by lr_blkptr */
167 blkptr_t lr_blkptr; /* spa block pointer for replay */
168 /* write data will follow for small writes */
172 lr_t lr_common; /* common portion of log record */
173 uint64_t lr_foid; /* object id of file to truncate */
174 uint64_t lr_offset; /* offset to truncate from */
175 uint64_t lr_length; /* length to truncate */
179 lr_t lr_common; /* common portion of log record */
180 uint64_t lr_foid; /* file object to change attributes */
181 uint64_t lr_mask; /* mask of attributes to set */
182 uint64_t lr_mode; /* mode to set */
183 uint64_t lr_uid; /* uid to set */
184 uint64_t lr_gid; /* gid to set */
185 uint64_t lr_size; /* size to set */
186 uint64_t lr_atime[2]; /* access time */
187 uint64_t lr_mtime[2]; /* modification time */
191 lr_t lr_common; /* common portion of log record */
192 uint64_t lr_foid; /* obj id of file */
193 uint64_t lr_aclcnt; /* number of acl entries */
194 /* lr_aclcnt number of ace_t entries follow this */
198 * ZIL structure definitions, interface function prototype and globals.
202 * ZFS intent log transaction structure
205 WR_INDIRECT, /* indirect - a large write (dmu_sync() data */
206 /* and put blkptr in log, rather than actual data) */
207 WR_COPIED, /* immediate - data is copied into lr_write_t */
208 WR_NEED_COPY, /* immediate - data needs to be copied if pushed */
212 list_node_t itx_node; /* linkage on zl_itx_list */
213 void *itx_private; /* type-specific opaque data */
214 itx_wr_state_t itx_wr_state; /* write state */
215 uint8_t itx_sync; /* synchronous transaction */
216 lr_t itx_lr; /* common part of log record */
217 /* followed by type-specific part of lr_xx_t and its immediate data */
222 * zgd_t is passed through dmu_sync() to the callback routine zfs_get_done()
223 * to handle the cleanup of the dmu_sync() buffer write
226 zilog_t *zgd_zilog; /* zilog */
227 blkptr_t *zgd_bp; /* block pointer */
228 struct rl *zgd_rl; /* range lock */
232 typedef void zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
234 typedef void zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
236 typedef int zil_replay_func_t();
237 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio);
239 extern uint64_t zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
240 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg);
242 extern void zil_init(void);
243 extern void zil_fini(void);
245 extern zilog_t *zil_alloc(objset_t *os, zil_header_t *zh_phys);
246 extern void zil_free(zilog_t *zilog);
248 extern zilog_t *zil_open(objset_t *os, zil_get_data_t *get_data);
249 extern void zil_close(zilog_t *zilog);
251 extern void zil_replay(objset_t *os, void *arg, uint64_t *txgp,
252 zil_replay_func_t *replay_func[TX_MAX_TYPE]);
253 extern void zil_destroy(zilog_t *zilog, boolean_t keep_first);
255 extern itx_t *zil_itx_create(int txtype, size_t lrsize);
256 extern uint64_t zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
258 extern void zil_commit(zilog_t *zilog, uint64_t seq, uint64_t oid);
260 extern int zil_claim(char *osname, void *txarg);
261 extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx);
262 extern void zil_clean(zilog_t *zilog);
263 extern int zil_is_committed(zilog_t *zilog);
265 extern int zil_suspend(zilog_t *zilog);
266 extern void zil_resume(zilog_t *zilog);
268 extern void zil_add_vdev(zilog_t *zilog, uint64_t vdev);
270 extern int zil_disable;
276 #endif /* _SYS_ZIL_H */