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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Integros [integros.com]
27 /* Portions Copyright 2010 Robert Milkowski */
32 #include <sys/types.h>
47 * Each objset has its own intent log. The log header (zil_header_t)
48 * for objset N's intent log is kept in the Nth object of the SPA's
49 * intent_log objset. The log header points to a chain of log blocks,
50 * each of which contains log records (i.e., transactions) followed by
51 * a log block trailer (zil_trailer_t). The format of a log record
52 * depends on the record (or transaction) type, but all records begin
53 * with a common structure that defines the type, length, and txg.
57 * Intent log header - this on disk structure holds fields to manage
58 * the log. All fields are 64 bit to easily handle cross architectures.
60 typedef struct zil_header {
61 uint64_t zh_claim_txg; /* txg in which log blocks were claimed */
62 uint64_t zh_replay_seq; /* highest replayed sequence number */
63 blkptr_t zh_log; /* log chain */
64 uint64_t zh_claim_blk_seq; /* highest claimed block sequence number */
65 uint64_t zh_flags; /* header flags */
66 uint64_t zh_claim_lr_seq; /* highest claimed lr sequence number */
71 * zh_flags bit settings
73 #define ZIL_REPLAY_NEEDED 0x1 /* replay needed - internal only */
74 #define ZIL_CLAIM_LR_SEQ_VALID 0x2 /* zh_claim_lr_seq field is valid */
79 * Log blocks are chained together. Originally they were chained at the
80 * end of the block. For performance reasons the chain was moved to the
81 * beginning of the block which allows writes for only the data being used.
82 * The older position is supported for backwards compatability.
84 * The zio_eck_t contains a zec_cksum which for the intent log is
85 * the sequence number of this log block. A seq of 0 is invalid.
86 * The zec_cksum is checked by the SPA against the sequence
87 * number passed in the blk_cksum field of the blkptr_t
89 typedef struct zil_chain {
91 blkptr_t zc_next_blk; /* next block in chain */
92 uint64_t zc_nused; /* bytes in log block used */
93 zio_eck_t zc_eck; /* block trailer */
96 #define ZIL_MIN_BLKSZ 4096ULL
99 * ziltest is by and large an ugly hack, but very useful in
100 * checking replay without tedious work.
101 * When running ziltest we want to keep all itx's and so maintain
102 * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG
103 * We subtract TXG_CONCURRENT_STATES to allow for common code.
105 #define ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES)
108 * The words of a log block checksum.
110 #define ZIL_ZC_GUID_0 0
111 #define ZIL_ZC_GUID_1 1
112 #define ZIL_ZC_OBJSET 2
115 typedef enum zil_create {
122 * size of xvattr log section.
123 * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
124 * for create time and a single 64 bit integer for all of the attributes,
125 * and 4 64 bit integers (32 bytes) for the scanstamp.
129 #define ZIL_XVAT_SIZE(mapsize) \
130 sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
131 (sizeof (uint64_t) * 7)
134 * Size of ACL in log. The ACE data is padded out to properly align
135 * on 8 byte boundary.
138 #define ZIL_ACE_LENGTH(x) (roundup(x, sizeof (uint64_t)))
141 * Intent log transaction types and record structures
143 #define TX_CREATE 1 /* Create file */
144 #define TX_MKDIR 2 /* Make directory */
145 #define TX_MKXATTR 3 /* Make XATTR directory */
146 #define TX_SYMLINK 4 /* Create symbolic link to a file */
147 #define TX_REMOVE 5 /* Remove file */
148 #define TX_RMDIR 6 /* Remove directory */
149 #define TX_LINK 7 /* Create hard link to a file */
150 #define TX_RENAME 8 /* Rename a file */
151 #define TX_WRITE 9 /* File write */
152 #define TX_TRUNCATE 10 /* Truncate a file */
153 #define TX_SETATTR 11 /* Set file attributes */
154 #define TX_ACL_V0 12 /* Set old formatted ACL */
155 #define TX_ACL 13 /* Set ACL */
156 #define TX_CREATE_ACL 14 /* create with ACL */
157 #define TX_CREATE_ATTR 15 /* create + attrs */
158 #define TX_CREATE_ACL_ATTR 16 /* create with ACL + attrs */
159 #define TX_MKDIR_ACL 17 /* mkdir with ACL */
160 #define TX_MKDIR_ATTR 18 /* mkdir with attr */
161 #define TX_MKDIR_ACL_ATTR 19 /* mkdir with ACL + attrs */
162 #define TX_WRITE2 20 /* dmu_sync EALREADY write */
163 #define TX_MAX_TYPE 21 /* Max transaction type */
166 * The transactions for mkdir, symlink, remove, rmdir, link, and rename
167 * may have the following bit set, indicating the original request
168 * specified case-insensitive handling of names.
170 #define TX_CI ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
173 * Transactions for write, truncate, setattr, acl_v0, and acl can be logged
174 * out of order. For convenience in the code, all such records must have
175 * lr_foid at the same offset.
177 #define TX_OOO(txtype) \
178 ((txtype) == TX_WRITE || \
179 (txtype) == TX_TRUNCATE || \
180 (txtype) == TX_SETATTR || \
181 (txtype) == TX_ACL_V0 || \
182 (txtype) == TX_ACL || \
183 (txtype) == TX_WRITE2)
186 * Format of log records.
187 * The fields are carefully defined to allow them to be aligned
188 * and sized the same on sparc & intel architectures.
189 * Each log record has a common structure at the beginning.
191 * The log record on disk (lrc_seq) holds the sequence number of all log
192 * records which is used to ensure we don't replay the same record.
194 typedef struct { /* common log record header */
195 uint64_t lrc_txtype; /* intent log transaction type */
196 uint64_t lrc_reclen; /* transaction record length */
197 uint64_t lrc_txg; /* dmu transaction group number */
198 uint64_t lrc_seq; /* see comment above */
202 * Common start of all out-of-order record types (TX_OOO() above).
205 lr_t lr_common; /* common portion of log record */
206 uint64_t lr_foid; /* object id */
210 * Handle option extended vattr attributes.
212 * Whenever new attributes are added the version number
213 * will need to be updated as will code in
214 * zfs_log.c and zfs_replay.c
217 uint32_t lr_attr_masksize; /* number of elements in array */
218 uint32_t lr_attr_bitmap; /* First entry of array */
219 /* remainder of array and any additional fields */
223 * log record for creates without optional ACL.
224 * This log record does support optional xvattr_t attributes.
227 lr_t lr_common; /* common portion of log record */
228 uint64_t lr_doid; /* object id of directory */
229 uint64_t lr_foid; /* object id of created file object */
230 uint64_t lr_mode; /* mode of object */
231 uint64_t lr_uid; /* uid of object */
232 uint64_t lr_gid; /* gid of object */
233 uint64_t lr_gen; /* generation (txg of creation) */
234 uint64_t lr_crtime[2]; /* creation time */
235 uint64_t lr_rdev; /* rdev of object to create */
236 /* name of object to create follows this */
237 /* for symlinks, link content follows name */
238 /* for creates with xvattr data, the name follows the xvattr info */
242 * FUID ACL record will be an array of ACEs from the original ACL.
243 * If this array includes ephemeral IDs, the record will also include
244 * an array of log-specific FUIDs to replace the ephemeral IDs.
245 * Only one copy of each unique domain will be present, so the log-specific
246 * FUIDs will use an index into a compressed domain table. On replay this
247 * information will be used to construct real FUIDs (and bypass idmap,
248 * since it may not be available).
252 * Log record for creates with optional ACL
253 * This log record is also used for recording any FUID
254 * information needed for replaying the create. If the
255 * file doesn't have any actual ACEs then the lr_aclcnt
258 * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
259 * If create is also setting xvattr's, then acl data follows xvattr.
260 * If ACE FUIDs are needed then they will follow the xvattr_t. Following
261 * the FUIDs will be the domain table information. The FUIDs for the owner
262 * and group will be in lr_create. Name follows ACL data.
265 lr_create_t lr_create; /* common create portion */
266 uint64_t lr_aclcnt; /* number of ACEs in ACL */
267 uint64_t lr_domcnt; /* number of unique domains */
268 uint64_t lr_fuidcnt; /* number of real fuids */
269 uint64_t lr_acl_bytes; /* number of bytes in ACL */
270 uint64_t lr_acl_flags; /* ACL flags */
274 lr_t lr_common; /* common portion of log record */
275 uint64_t lr_doid; /* obj id of directory */
276 /* name of object to remove follows this */
280 lr_t lr_common; /* common portion of log record */
281 uint64_t lr_doid; /* obj id of directory */
282 uint64_t lr_link_obj; /* obj id of link */
283 /* name of object to link follows this */
287 lr_t lr_common; /* common portion of log record */
288 uint64_t lr_sdoid; /* obj id of source directory */
289 uint64_t lr_tdoid; /* obj id of target directory */
290 /* 2 strings: names of source and destination follow this */
294 lr_t lr_common; /* common portion of log record */
295 uint64_t lr_foid; /* file object to write */
296 uint64_t lr_offset; /* offset to write to */
297 uint64_t lr_length; /* user data length to write */
298 uint64_t lr_blkoff; /* no longer used */
299 blkptr_t lr_blkptr; /* spa block pointer for replay */
300 /* write data will follow for small writes */
304 lr_t lr_common; /* common portion of log record */
305 uint64_t lr_foid; /* object id of file to truncate */
306 uint64_t lr_offset; /* offset to truncate from */
307 uint64_t lr_length; /* length to truncate */
311 lr_t lr_common; /* common portion of log record */
312 uint64_t lr_foid; /* file object to change attributes */
313 uint64_t lr_mask; /* mask of attributes to set */
314 uint64_t lr_mode; /* mode to set */
315 uint64_t lr_uid; /* uid to set */
316 uint64_t lr_gid; /* gid to set */
317 uint64_t lr_size; /* size to set */
318 uint64_t lr_atime[2]; /* access time */
319 uint64_t lr_mtime[2]; /* modification time */
320 /* optional attribute lr_attr_t may be here */
324 lr_t lr_common; /* common portion of log record */
325 uint64_t lr_foid; /* obj id of file */
326 uint64_t lr_aclcnt; /* number of acl entries */
327 /* lr_aclcnt number of ace_t entries follow this */
331 lr_t lr_common; /* common portion of log record */
332 uint64_t lr_foid; /* obj id of file */
333 uint64_t lr_aclcnt; /* number of ACEs in ACL */
334 uint64_t lr_domcnt; /* number of unique domains */
335 uint64_t lr_fuidcnt; /* number of real fuids */
336 uint64_t lr_acl_bytes; /* number of bytes in ACL */
337 uint64_t lr_acl_flags; /* ACL flags */
338 /* lr_acl_bytes number of variable sized ace's follows */
342 * ZIL structure definitions, interface function prototype and globals.
346 * Writes are handled in three different ways:
349 * In this mode, if we need to commit the write later, then the block
350 * is immediately written into the file system (using dmu_sync),
351 * and a pointer to the block is put into the log record.
352 * When the txg commits the block is linked in.
353 * This saves additionally writing the data into the log record.
354 * There are a few requirements for this to occur:
355 * - write is greater than zfs/zvol_immediate_write_sz
356 * - not using slogs (as slogs are assumed to always be faster
357 * than writing into the main pool)
358 * - the write occupies only one block
360 * If we know we'll immediately be committing the
361 * transaction (FSYNC or FDSYNC), the we allocate a larger
362 * log record here for the data and copy the data in.
364 * Otherwise we don't allocate a buffer, and *if* we need to
365 * flush the write later then a buffer is allocated and
366 * we retrieve the data using the dmu.
369 WR_INDIRECT, /* indirect - a large write (dmu_sync() data */
370 /* and put blkptr in log, rather than actual data) */
371 WR_COPIED, /* immediate - data is copied into lr_write_t */
372 WR_NEED_COPY, /* immediate - data needs to be copied if pushed */
373 WR_NUM_STATES /* number of states */
377 list_node_t itx_node; /* linkage on zl_itx_list */
378 void *itx_private; /* type-specific opaque data */
379 itx_wr_state_t itx_wr_state; /* write state */
380 uint8_t itx_sync; /* synchronous transaction */
381 uint64_t itx_oid; /* object id */
382 lr_t itx_lr; /* common part of log record */
383 /* followed by type-specific part of lr_xx_t and its immediate data */
386 typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
388 typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
390 typedef int zil_replay_func_t();
391 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio);
393 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
394 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg);
396 extern void zil_init(void);
397 extern void zil_fini(void);
399 extern zilog_t *zil_alloc(objset_t *os, zil_header_t *zh_phys);
400 extern void zil_free(zilog_t *zilog);
402 extern zilog_t *zil_open(objset_t *os, zil_get_data_t *get_data);
403 extern void zil_close(zilog_t *zilog);
405 extern void zil_replay(objset_t *os, void *arg,
406 zil_replay_func_t *replay_func[TX_MAX_TYPE]);
407 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
408 extern void zil_destroy(zilog_t *zilog, boolean_t keep_first);
409 extern void zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
410 extern void zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx);
412 extern itx_t *zil_itx_create(uint64_t txtype, size_t lrsize);
413 extern void zil_itx_destroy(itx_t *itx);
414 extern void zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
416 extern void zil_async_to_sync(zilog_t *zilog, uint64_t oid);
417 extern void zil_commit(zilog_t *zilog, uint64_t oid);
419 extern int zil_vdev_offline(const char *osname, void *txarg);
420 extern int zil_claim(struct dsl_pool *dp,
421 struct dsl_dataset *ds, void *txarg);
422 extern int zil_check_log_chain(struct dsl_pool *dp,
423 struct dsl_dataset *ds, void *tx);
424 extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx);
425 extern void zil_clean(zilog_t *zilog, uint64_t synced_txg);
427 extern int zil_suspend(const char *osname, void **cookiep);
428 extern void zil_resume(void *cookie);
430 extern void zil_add_block(zilog_t *zilog, const blkptr_t *bp);
431 extern int zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
433 extern void zil_set_sync(zilog_t *zilog, uint64_t syncval);
435 extern void zil_set_logbias(zilog_t *zilog, uint64_t slogval);
437 extern int zil_replay_disable;
443 #endif /* _SYS_ZIL_H */