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>
48 * Each objset has its own intent log. The log header (zil_header_t)
49 * for objset N's intent log is kept in the Nth object of the SPA's
50 * intent_log objset. The log header points to a chain of log blocks,
51 * each of which contains log records (i.e., transactions) followed by
52 * a log block trailer (zil_trailer_t). The format of a log record
53 * depends on the record (or transaction) type, but all records begin
54 * with a common structure that defines the type, length, and txg.
58 * Intent log header - this on disk structure holds fields to manage
59 * the log. All fields are 64 bit to easily handle cross architectures.
61 typedef struct zil_header
{
62 uint64_t zh_claim_txg
; /* txg in which log blocks were claimed */
63 uint64_t zh_replay_seq
; /* highest replayed sequence number */
64 blkptr_t zh_log
; /* log chain */
65 uint64_t zh_claim_blk_seq
; /* highest claimed block sequence number */
66 uint64_t zh_flags
; /* header flags */
67 uint64_t zh_claim_lr_seq
; /* highest claimed lr sequence number */
72 * zh_flags bit settings
74 #define ZIL_REPLAY_NEEDED 0x1 /* replay needed - internal only */
75 #define ZIL_CLAIM_LR_SEQ_VALID 0x2 /* zh_claim_lr_seq field is valid */
80 * Log blocks are chained together. Originally they were chained at the
81 * end of the block. For performance reasons the chain was moved to the
82 * beginning of the block which allows writes for only the data being used.
83 * The older position is supported for backwards compatability.
85 * The zio_eck_t contains a zec_cksum which for the intent log is
86 * the sequence number of this log block. A seq of 0 is invalid.
87 * The zec_cksum is checked by the SPA against the sequence
88 * number passed in the blk_cksum field of the blkptr_t
90 typedef struct zil_chain
{
92 blkptr_t zc_next_blk
; /* next block in chain */
93 uint64_t zc_nused
; /* bytes in log block used */
94 zio_eck_t zc_eck
; /* block trailer */
97 #define ZIL_MIN_BLKSZ 4096ULL
100 * ziltest is by and large an ugly hack, but very useful in
101 * checking replay without tedious work.
102 * When running ziltest we want to keep all itx's and so maintain
103 * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG
104 * We subtract TXG_CONCURRENT_STATES to allow for common code.
106 #define ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES)
109 * The words of a log block checksum.
111 #define ZIL_ZC_GUID_0 0
112 #define ZIL_ZC_GUID_1 1
113 #define ZIL_ZC_OBJSET 2
116 typedef enum zil_create
{
123 * size of xvattr log section.
124 * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
125 * for create time and a single 64 bit integer for all of the attributes,
126 * and 4 64 bit integers (32 bytes) for the scanstamp.
130 #define ZIL_XVAT_SIZE(mapsize) \
131 sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
132 (sizeof (uint64_t) * 7)
135 * Size of ACL in log. The ACE data is padded out to properly align
136 * on 8 byte boundary.
139 #define ZIL_ACE_LENGTH(x) (roundup(x, sizeof (uint64_t)))
142 * Intent log transaction types and record structures
144 #define TX_COMMIT 0 /* Commit marker (no on-disk state) */
145 #define TX_CREATE 1 /* Create file */
146 #define TX_MKDIR 2 /* Make directory */
147 #define TX_MKXATTR 3 /* Make XATTR directory */
148 #define TX_SYMLINK 4 /* Create symbolic link to a file */
149 #define TX_REMOVE 5 /* Remove file */
150 #define TX_RMDIR 6 /* Remove directory */
151 #define TX_LINK 7 /* Create hard link to a file */
152 #define TX_RENAME 8 /* Rename a file */
153 #define TX_WRITE 9 /* File write */
154 #define TX_TRUNCATE 10 /* Truncate a file */
155 #define TX_SETATTR 11 /* Set file attributes */
156 #define TX_ACL_V0 12 /* Set old formatted ACL */
157 #define TX_ACL 13 /* Set ACL */
158 #define TX_CREATE_ACL 14 /* create with ACL */
159 #define TX_CREATE_ATTR 15 /* create + attrs */
160 #define TX_CREATE_ACL_ATTR 16 /* create with ACL + attrs */
161 #define TX_MKDIR_ACL 17 /* mkdir with ACL */
162 #define TX_MKDIR_ATTR 18 /* mkdir with attr */
163 #define TX_MKDIR_ACL_ATTR 19 /* mkdir with ACL + attrs */
164 #define TX_WRITE2 20 /* dmu_sync EALREADY write */
165 #define TX_MAX_TYPE 21 /* Max transaction type */
168 * The transactions for mkdir, symlink, remove, rmdir, link, and rename
169 * may have the following bit set, indicating the original request
170 * specified case-insensitive handling of names.
172 #define TX_CI ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
175 * Transactions for write, truncate, setattr, acl_v0, and acl can be logged
176 * out of order. For convenience in the code, all such records must have
177 * lr_foid at the same offset.
179 #define TX_OOO(txtype) \
180 ((txtype) == TX_WRITE || \
181 (txtype) == TX_TRUNCATE || \
182 (txtype) == TX_SETATTR || \
183 (txtype) == TX_ACL_V0 || \
184 (txtype) == TX_ACL || \
185 (txtype) == TX_WRITE2)
188 * Format of log records.
189 * The fields are carefully defined to allow them to be aligned
190 * and sized the same on sparc & intel architectures.
191 * Each log record has a common structure at the beginning.
193 * The log record on disk (lrc_seq) holds the sequence number of all log
194 * records which is used to ensure we don't replay the same record.
196 typedef struct { /* common log record header */
197 uint64_t lrc_txtype
; /* intent log transaction type */
198 uint64_t lrc_reclen
; /* transaction record length */
199 uint64_t lrc_txg
; /* dmu transaction group number */
200 uint64_t lrc_seq
; /* see comment above */
204 * Common start of all out-of-order record types (TX_OOO() above).
207 lr_t lr_common
; /* common portion of log record */
208 uint64_t lr_foid
; /* object id */
212 * Handle option extended vattr attributes.
214 * Whenever new attributes are added the version number
215 * will need to be updated as will code in
216 * zfs_log.c and zfs_replay.c
219 uint32_t lr_attr_masksize
; /* number of elements in array */
220 uint32_t lr_attr_bitmap
; /* First entry of array */
221 /* remainder of array and any additional fields */
225 * log record for creates without optional ACL.
226 * This log record does support optional xvattr_t attributes.
229 lr_t lr_common
; /* common portion of log record */
230 uint64_t lr_doid
; /* object id of directory */
231 uint64_t lr_foid
; /* object id of created file object */
232 uint64_t lr_mode
; /* mode of object */
233 uint64_t lr_uid
; /* uid of object */
234 uint64_t lr_gid
; /* gid of object */
235 uint64_t lr_gen
; /* generation (txg of creation) */
236 uint64_t lr_crtime
[2]; /* creation time */
237 uint64_t lr_rdev
; /* rdev of object to create */
238 /* name of object to create follows this */
239 /* for symlinks, link content follows name */
240 /* for creates with xvattr data, the name follows the xvattr info */
244 * FUID ACL record will be an array of ACEs from the original ACL.
245 * If this array includes ephemeral IDs, the record will also include
246 * an array of log-specific FUIDs to replace the ephemeral IDs.
247 * Only one copy of each unique domain will be present, so the log-specific
248 * FUIDs will use an index into a compressed domain table. On replay this
249 * information will be used to construct real FUIDs (and bypass idmap,
250 * since it may not be available).
254 * Log record for creates with optional ACL
255 * This log record is also used for recording any FUID
256 * information needed for replaying the create. If the
257 * file doesn't have any actual ACEs then the lr_aclcnt
260 * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
261 * If create is also setting xvattr's, then acl data follows xvattr.
262 * If ACE FUIDs are needed then they will follow the xvattr_t. Following
263 * the FUIDs will be the domain table information. The FUIDs for the owner
264 * and group will be in lr_create. Name follows ACL data.
267 lr_create_t lr_create
; /* common create portion */
268 uint64_t lr_aclcnt
; /* number of ACEs in ACL */
269 uint64_t lr_domcnt
; /* number of unique domains */
270 uint64_t lr_fuidcnt
; /* number of real fuids */
271 uint64_t lr_acl_bytes
; /* number of bytes in ACL */
272 uint64_t lr_acl_flags
; /* ACL flags */
276 lr_t lr_common
; /* common portion of log record */
277 uint64_t lr_doid
; /* obj id of directory */
278 /* name of object to remove follows this */
282 lr_t lr_common
; /* common portion of log record */
283 uint64_t lr_doid
; /* obj id of directory */
284 uint64_t lr_link_obj
; /* obj id of link */
285 /* name of object to link follows this */
289 lr_t lr_common
; /* common portion of log record */
290 uint64_t lr_sdoid
; /* obj id of source directory */
291 uint64_t lr_tdoid
; /* obj id of target directory */
292 /* 2 strings: names of source and destination follow this */
296 lr_t lr_common
; /* common portion of log record */
297 uint64_t lr_foid
; /* file object to write */
298 uint64_t lr_offset
; /* offset to write to */
299 uint64_t lr_length
; /* user data length to write */
300 uint64_t lr_blkoff
; /* no longer used */
301 blkptr_t lr_blkptr
; /* spa block pointer for replay */
302 /* write data will follow for small writes */
306 lr_t lr_common
; /* common portion of log record */
307 uint64_t lr_foid
; /* object id of file to truncate */
308 uint64_t lr_offset
; /* offset to truncate from */
309 uint64_t lr_length
; /* length to truncate */
313 lr_t lr_common
; /* common portion of log record */
314 uint64_t lr_foid
; /* file object to change attributes */
315 uint64_t lr_mask
; /* mask of attributes to set */
316 uint64_t lr_mode
; /* mode to set */
317 uint64_t lr_uid
; /* uid to set */
318 uint64_t lr_gid
; /* gid to set */
319 uint64_t lr_size
; /* size to set */
320 uint64_t lr_atime
[2]; /* access time */
321 uint64_t lr_mtime
[2]; /* modification time */
322 /* optional attribute lr_attr_t may be here */
326 lr_t lr_common
; /* common portion of log record */
327 uint64_t lr_foid
; /* obj id of file */
328 uint64_t lr_aclcnt
; /* number of acl entries */
329 /* lr_aclcnt number of ace_t entries follow this */
333 lr_t lr_common
; /* common portion of log record */
334 uint64_t lr_foid
; /* obj id of file */
335 uint64_t lr_aclcnt
; /* number of ACEs in ACL */
336 uint64_t lr_domcnt
; /* number of unique domains */
337 uint64_t lr_fuidcnt
; /* number of real fuids */
338 uint64_t lr_acl_bytes
; /* number of bytes in ACL */
339 uint64_t lr_acl_flags
; /* ACL flags */
340 /* lr_acl_bytes number of variable sized ace's follows */
344 * ZIL structure definitions, interface function prototype and globals.
348 * Writes are handled in three different ways:
351 * In this mode, if we need to commit the write later, then the block
352 * is immediately written into the file system (using dmu_sync),
353 * and a pointer to the block is put into the log record.
354 * When the txg commits the block is linked in.
355 * This saves additionally writing the data into the log record.
356 * There are a few requirements for this to occur:
357 * - write is greater than zfs/zvol_immediate_write_sz
358 * - not using slogs (as slogs are assumed to always be faster
359 * than writing into the main pool)
360 * - the write occupies only one block
362 * If we know we'll immediately be committing the
363 * transaction (FSYNC or FDSYNC), the we allocate a larger
364 * log record here for the data and copy the data in.
366 * Otherwise we don't allocate a buffer, and *if* we need to
367 * flush the write later then a buffer is allocated and
368 * we retrieve the data using the dmu.
371 WR_INDIRECT
, /* indirect - a large write (dmu_sync() data */
372 /* and put blkptr in log, rather than actual data) */
373 WR_COPIED
, /* immediate - data is copied into lr_write_t */
374 WR_NEED_COPY
, /* immediate - data needs to be copied if pushed */
375 WR_NUM_STATES
/* number of states */
379 list_node_t itx_node
; /* linkage on zl_itx_list */
380 void *itx_private
; /* type-specific opaque data */
381 itx_wr_state_t itx_wr_state
; /* write state */
382 uint8_t itx_sync
; /* synchronous transaction */
383 uint64_t itx_oid
; /* object id */
384 lr_t itx_lr
; /* common part of log record */
385 /* followed by type-specific part of lr_xx_t and its immediate data */
388 typedef int zil_parse_blk_func_t(zilog_t
*zilog
, blkptr_t
*bp
, void *arg
,
390 typedef int zil_parse_lr_func_t(zilog_t
*zilog
, lr_t
*lr
, void *arg
,
392 typedef int zil_replay_func_t(void *arg1
, void *arg2
, boolean_t byteswap
);
393 typedef int zil_get_data_t(void *arg
, lr_write_t
*lr
, char *dbuf
,
394 struct lwb
*lwb
, zio_t
*zio
);
396 extern int zil_parse(zilog_t
*zilog
, zil_parse_blk_func_t
*parse_blk_func
,
397 zil_parse_lr_func_t
*parse_lr_func
, void *arg
, uint64_t txg
);
399 extern void zil_init(void);
400 extern void zil_fini(void);
402 extern zilog_t
*zil_alloc(objset_t
*os
, zil_header_t
*zh_phys
);
403 extern void zil_free(zilog_t
*zilog
);
405 extern zilog_t
*zil_open(objset_t
*os
, zil_get_data_t
*get_data
);
406 extern void zil_close(zilog_t
*zilog
);
408 extern void zil_replay(objset_t
*os
, void *arg
,
409 zil_replay_func_t
*replay_func
[TX_MAX_TYPE
]);
410 extern boolean_t
zil_replaying(zilog_t
*zilog
, dmu_tx_t
*tx
);
411 extern void zil_destroy(zilog_t
*zilog
, boolean_t keep_first
);
412 extern void zil_destroy_sync(zilog_t
*zilog
, dmu_tx_t
*tx
);
413 extern void zil_rollback_destroy(zilog_t
*zilog
, dmu_tx_t
*tx
);
415 extern itx_t
*zil_itx_create(uint64_t txtype
, size_t lrsize
);
416 extern void zil_itx_destroy(itx_t
*itx
);
417 extern void zil_itx_assign(zilog_t
*zilog
, itx_t
*itx
, dmu_tx_t
*tx
);
419 extern void zil_commit(zilog_t
*zilog
, uint64_t oid
);
421 extern int zil_vdev_offline(const char *osname
, void *txarg
);
422 extern int zil_claim(struct dsl_pool
*dp
,
423 struct dsl_dataset
*ds
, void *txarg
);
424 extern int zil_check_log_chain(struct dsl_pool
*dp
,
425 struct dsl_dataset
*ds
, void *tx
);
426 extern void zil_sync(zilog_t
*zilog
, dmu_tx_t
*tx
);
427 extern void zil_clean(zilog_t
*zilog
, uint64_t synced_txg
);
429 extern int zil_suspend(const char *osname
, void **cookiep
);
430 extern void zil_resume(void *cookie
);
432 extern void zil_lwb_add_block(struct lwb
*lwb
, const blkptr_t
*bp
);
433 extern void zil_lwb_add_txg(struct lwb
*lwb
, uint64_t txg
);
434 extern int zil_bp_tree_add(zilog_t
*zilog
, const blkptr_t
*bp
);
436 extern void zil_set_sync(zilog_t
*zilog
, uint64_t syncval
);
438 extern void zil_set_logbias(zilog_t
*zilog
, uint64_t slogval
);
440 extern int zil_replay_disable
;
446 #endif /* _SYS_ZIL_H */