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 by Delphix. All rights reserved.
26 /* Portions Copyright 2010 Robert Milkowski */
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_blk_seq
; /* highest claimed block sequence number */
61 uint64_t zh_flags
; /* header flags */
62 uint64_t zh_claim_lr_seq
; /* highest claimed lr sequence number */
67 * zh_flags bit settings
69 #define ZIL_REPLAY_NEEDED 0x1 /* replay needed - internal only */
70 #define ZIL_CLAIM_LR_SEQ_VALID 0x2 /* zh_claim_lr_seq field is valid */
75 * Log blocks are chained together. Originally they were chained at the
76 * end of the block. For performance reasons the chain was moved to the
77 * beginning of the block which allows writes for only the data being used.
78 * The older position is supported for backwards compatability.
80 * The zio_eck_t contains a zec_cksum which for the intent log is
81 * the sequence number of this log block. A seq of 0 is invalid.
82 * The zec_cksum is checked by the SPA against the sequence
83 * number passed in the blk_cksum field of the blkptr_t
85 typedef struct zil_chain
{
87 blkptr_t zc_next_blk
; /* next block in chain */
88 uint64_t zc_nused
; /* bytes in log block used */
89 zio_eck_t zc_eck
; /* block trailer */
92 #define ZIL_MIN_BLKSZ 4096ULL
93 #define ZIL_MAX_BLKSZ SPA_MAXBLOCKSIZE
96 * The words of a log block checksum.
98 #define ZIL_ZC_GUID_0 0
99 #define ZIL_ZC_GUID_1 1
100 #define ZIL_ZC_OBJSET 2
103 typedef enum zil_create
{
110 * size of xvattr log section.
111 * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
112 * for create time and a single 64 bit integer for all of the attributes,
113 * and 4 64 bit integers (32 bytes) for the scanstamp.
117 #define ZIL_XVAT_SIZE(mapsize) \
118 sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
119 (sizeof (uint64_t) * 7)
122 * Size of ACL in log. The ACE data is padded out to properly align
123 * on 8 byte boundary.
126 #define ZIL_ACE_LENGTH(x) (roundup(x, sizeof (uint64_t)))
129 * Intent log transaction types and record structures
131 #define TX_CREATE 1 /* Create file */
132 #define TX_MKDIR 2 /* Make directory */
133 #define TX_MKXATTR 3 /* Make XATTR directory */
134 #define TX_SYMLINK 4 /* Create symbolic link to a file */
135 #define TX_REMOVE 5 /* Remove file */
136 #define TX_RMDIR 6 /* Remove directory */
137 #define TX_LINK 7 /* Create hard link to a file */
138 #define TX_RENAME 8 /* Rename a file */
139 #define TX_WRITE 9 /* File write */
140 #define TX_TRUNCATE 10 /* Truncate a file */
141 #define TX_SETATTR 11 /* Set file attributes */
142 #define TX_ACL_V0 12 /* Set old formatted ACL */
143 #define TX_ACL 13 /* Set ACL */
144 #define TX_CREATE_ACL 14 /* create with ACL */
145 #define TX_CREATE_ATTR 15 /* create + attrs */
146 #define TX_CREATE_ACL_ATTR 16 /* create with ACL + attrs */
147 #define TX_MKDIR_ACL 17 /* mkdir with ACL */
148 #define TX_MKDIR_ATTR 18 /* mkdir with attr */
149 #define TX_MKDIR_ACL_ATTR 19 /* mkdir with ACL + attrs */
150 #define TX_WRITE2 20 /* dmu_sync EALREADY write */
151 #define TX_MAX_TYPE 21 /* Max transaction type */
154 * The transactions for mkdir, symlink, remove, rmdir, link, and rename
155 * may have the following bit set, indicating the original request
156 * specified case-insensitive handling of names.
158 #define TX_CI ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
161 * Transactions for write, truncate, setattr, acl_v0, and acl can be logged
162 * out of order. For convenience in the code, all such records must have
163 * lr_foid at the same offset.
165 #define TX_OOO(txtype) \
166 ((txtype) == TX_WRITE || \
167 (txtype) == TX_TRUNCATE || \
168 (txtype) == TX_SETATTR || \
169 (txtype) == TX_ACL_V0 || \
170 (txtype) == TX_ACL || \
171 (txtype) == TX_WRITE2)
174 * Format of log records.
175 * The fields are carefully defined to allow them to be aligned
176 * and sized the same on sparc & intel architectures.
177 * Each log record has a common structure at the beginning.
179 * The log record on disk (lrc_seq) holds the sequence number of all log
180 * records which is used to ensure we don't replay the same record.
182 typedef struct { /* common log record header */
183 uint64_t lrc_txtype
; /* intent log transaction type */
184 uint64_t lrc_reclen
; /* transaction record length */
185 uint64_t lrc_txg
; /* dmu transaction group number */
186 uint64_t lrc_seq
; /* see comment above */
190 * Common start of all out-of-order record types (TX_OOO() above).
193 lr_t lr_common
; /* common portion of log record */
194 uint64_t lr_foid
; /* object id */
198 * Handle option extended vattr attributes.
200 * Whenever new attributes are added the version number
201 * will need to be updated as will code in
202 * zfs_log.c and zfs_replay.c
205 uint32_t lr_attr_masksize
; /* number of elements in array */
206 uint32_t lr_attr_bitmap
; /* First entry of array */
207 /* remainder of array and any additional fields */
211 * log record for creates without optional ACL.
212 * This log record does support optional xvattr_t attributes.
215 lr_t lr_common
; /* common portion of log record */
216 uint64_t lr_doid
; /* object id of directory */
217 uint64_t lr_foid
; /* object id of created file object */
218 uint64_t lr_mode
; /* mode of object */
219 uint64_t lr_uid
; /* uid of object */
220 uint64_t lr_gid
; /* gid of object */
221 uint64_t lr_gen
; /* generation (txg of creation) */
222 uint64_t lr_crtime
[2]; /* creation time */
223 uint64_t lr_rdev
; /* rdev of object to create */
224 /* name of object to create follows this */
225 /* for symlinks, link content follows name */
226 /* for creates with xvattr data, the name follows the xvattr info */
230 * FUID ACL record will be an array of ACEs from the original ACL.
231 * If this array includes ephemeral IDs, the record will also include
232 * an array of log-specific FUIDs to replace the ephemeral IDs.
233 * Only one copy of each unique domain will be present, so the log-specific
234 * FUIDs will use an index into a compressed domain table. On replay this
235 * information will be used to construct real FUIDs (and bypass idmap,
236 * since it may not be available).
240 * Log record for creates with optional ACL
241 * This log record is also used for recording any FUID
242 * information needed for replaying the create. If the
243 * file doesn't have any actual ACEs then the lr_aclcnt
247 lr_create_t lr_create
; /* common create portion */
248 uint64_t lr_aclcnt
; /* number of ACEs in ACL */
249 uint64_t lr_domcnt
; /* number of unique domains */
250 uint64_t lr_fuidcnt
; /* number of real fuids */
251 uint64_t lr_acl_bytes
; /* number of bytes in ACL */
252 uint64_t lr_acl_flags
; /* ACL flags */
253 /* lr_acl_bytes number of variable sized ace's follows */
254 /* if create is also setting xvattr's, then acl data follows xvattr */
255 /* if ACE FUIDs are needed then they will follow the xvattr_t */
256 /* Following the FUIDs will be the domain table information. */
257 /* The FUIDs for the owner and group will be in the lr_create */
258 /* portion of the record. */
259 /* name follows ACL data */
263 lr_t lr_common
; /* common portion of log record */
264 uint64_t lr_doid
; /* obj id of directory */
265 /* name of object to remove follows this */
269 lr_t lr_common
; /* common portion of log record */
270 uint64_t lr_doid
; /* obj id of directory */
271 uint64_t lr_link_obj
; /* obj id of link */
272 /* name of object to link follows this */
276 lr_t lr_common
; /* common portion of log record */
277 uint64_t lr_sdoid
; /* obj id of source directory */
278 uint64_t lr_tdoid
; /* obj id of target directory */
279 /* 2 strings: names of source and destination follow this */
283 lr_t lr_common
; /* common portion of log record */
284 uint64_t lr_foid
; /* file object to write */
285 uint64_t lr_offset
; /* offset to write to */
286 uint64_t lr_length
; /* user data length to write */
287 uint64_t lr_blkoff
; /* no longer used */
288 blkptr_t lr_blkptr
; /* spa block pointer for replay */
289 /* write data will follow for small writes */
293 lr_t lr_common
; /* common portion of log record */
294 uint64_t lr_foid
; /* object id of file to truncate */
295 uint64_t lr_offset
; /* offset to truncate from */
296 uint64_t lr_length
; /* length to truncate */
300 lr_t lr_common
; /* common portion of log record */
301 uint64_t lr_foid
; /* file object to change attributes */
302 uint64_t lr_mask
; /* mask of attributes to set */
303 uint64_t lr_mode
; /* mode to set */
304 uint64_t lr_uid
; /* uid to set */
305 uint64_t lr_gid
; /* gid to set */
306 uint64_t lr_size
; /* size to set */
307 uint64_t lr_atime
[2]; /* access time */
308 uint64_t lr_mtime
[2]; /* modification time */
309 /* optional attribute lr_attr_t may be here */
313 lr_t lr_common
; /* common portion of log record */
314 uint64_t lr_foid
; /* obj id of file */
315 uint64_t lr_aclcnt
; /* number of acl entries */
316 /* lr_aclcnt number of ace_t entries follow this */
320 lr_t lr_common
; /* common portion of log record */
321 uint64_t lr_foid
; /* obj id of file */
322 uint64_t lr_aclcnt
; /* number of ACEs in ACL */
323 uint64_t lr_domcnt
; /* number of unique domains */
324 uint64_t lr_fuidcnt
; /* number of real fuids */
325 uint64_t lr_acl_bytes
; /* number of bytes in ACL */
326 uint64_t lr_acl_flags
; /* ACL flags */
327 /* lr_acl_bytes number of variable sized ace's follows */
331 * ZIL structure definitions, interface function prototype and globals.
335 * Writes are handled in three different ways:
338 * In this mode, if we need to commit the write later, then the block
339 * is immediately written into the file system (using dmu_sync),
340 * and a pointer to the block is put into the log record.
341 * When the txg commits the block is linked in.
342 * This saves additionally writing the data into the log record.
343 * There are a few requirements for this to occur:
344 * - write is greater than zfs/zvol_immediate_write_sz
345 * - not using slogs (as slogs are assumed to always be faster
346 * than writing into the main pool)
347 * - the write occupies only one block
349 * If we know we'll immediately be committing the
350 * transaction (FSYNC or FDSYNC), the we allocate a larger
351 * log record here for the data and copy the data in.
353 * Otherwise we don't allocate a buffer, and *if* we need to
354 * flush the write later then a buffer is allocated and
355 * we retrieve the data using the dmu.
358 WR_INDIRECT
, /* indirect - a large write (dmu_sync() data */
359 /* and put blkptr in log, rather than actual data) */
360 WR_COPIED
, /* immediate - data is copied into lr_write_t */
361 WR_NEED_COPY
, /* immediate - data needs to be copied if pushed */
362 WR_NUM_STATES
/* number of states */
366 list_node_t itx_node
; /* linkage on zl_itx_list */
367 void *itx_private
; /* type-specific opaque data */
368 itx_wr_state_t itx_wr_state
; /* write state */
369 uint8_t itx_sync
; /* synchronous transaction */
370 uint64_t itx_sod
; /* record size on disk */
371 uint64_t itx_oid
; /* object id */
372 lr_t itx_lr
; /* common part of log record */
373 /* followed by type-specific part of lr_xx_t and its immediate data */
377 * Used for zil kstat.
379 typedef struct zil_stats
{
381 * Number of times a ZIL commit (e.g. fsync) has been requested.
383 kstat_named_t zil_commit_count
;
386 * Number of times the ZIL has been flushed to stable storage.
387 * This is less than zil_commit_count when commits are "merged"
388 * (see the documentation above zil_commit()).
390 kstat_named_t zil_commit_writer_count
;
393 * Number of transactions (reads, writes, renames, etc.)
394 * that have been commited.
396 kstat_named_t zil_itx_count
;
399 * See the documentation for itx_wr_state_t above.
400 * Note that "bytes" accumulates the length of the transactions
401 * (i.e. data), not the actual log record sizes.
403 kstat_named_t zil_itx_indirect_count
;
404 kstat_named_t zil_itx_indirect_bytes
;
405 kstat_named_t zil_itx_copied_count
;
406 kstat_named_t zil_itx_copied_bytes
;
407 kstat_named_t zil_itx_needcopy_count
;
408 kstat_named_t zil_itx_needcopy_bytes
;
411 * Transactions which have been allocated to the "normal"
412 * (i.e. not slog) storage pool. Note that "bytes" accumulate
413 * the actual log record sizes - which do not include the actual
414 * data in case of indirect writes.
416 kstat_named_t zil_itx_metaslab_normal_count
;
417 kstat_named_t zil_itx_metaslab_normal_bytes
;
420 * Transactions which have been allocated to the "slog" storage pool.
421 * If there are no separate log devices, this is the same as the
424 kstat_named_t zil_itx_metaslab_slog_count
;
425 kstat_named_t zil_itx_metaslab_slog_bytes
;
428 extern zil_stats_t zil_stats
;
430 #define ZIL_STAT_INCR(stat, val) \
431 atomic_add_64(&zil_stats.stat.value.ui64, (val));
432 #define ZIL_STAT_BUMP(stat) \
433 ZIL_STAT_INCR(stat, 1);
435 typedef int zil_parse_blk_func_t(zilog_t
*zilog
, blkptr_t
*bp
, void *arg
,
437 typedef int zil_parse_lr_func_t(zilog_t
*zilog
, lr_t
*lr
, void *arg
,
439 typedef int (*const zil_replay_func_t
)(void *, char *, boolean_t
);
440 typedef int zil_get_data_t(void *arg
, lr_write_t
*lr
, char *dbuf
, zio_t
*zio
);
442 extern int zil_parse(zilog_t
*zilog
, zil_parse_blk_func_t
*parse_blk_func
,
443 zil_parse_lr_func_t
*parse_lr_func
, void *arg
, uint64_t txg
);
445 extern void zil_init(void);
446 extern void zil_fini(void);
448 extern zilog_t
*zil_alloc(objset_t
*os
, zil_header_t
*zh_phys
);
449 extern void zil_free(zilog_t
*zilog
);
451 extern zilog_t
*zil_open(objset_t
*os
, zil_get_data_t
*get_data
);
452 extern void zil_close(zilog_t
*zilog
);
454 extern void zil_replay(objset_t
*os
, void *arg
,
455 zil_replay_func_t replay_func
[TX_MAX_TYPE
]);
456 extern boolean_t
zil_replaying(zilog_t
*zilog
, dmu_tx_t
*tx
);
457 extern void zil_destroy(zilog_t
*zilog
, boolean_t keep_first
);
458 extern void zil_destroy_sync(zilog_t
*zilog
, dmu_tx_t
*tx
);
459 extern void zil_rollback_destroy(zilog_t
*zilog
, dmu_tx_t
*tx
);
461 extern itx_t
*zil_itx_create(uint64_t txtype
, size_t lrsize
);
462 extern void zil_itx_destroy(itx_t
*itx
);
463 extern void zil_itx_assign(zilog_t
*zilog
, itx_t
*itx
, dmu_tx_t
*tx
);
465 extern void zil_commit(zilog_t
*zilog
, uint64_t oid
);
467 extern int zil_vdev_offline(const char *osname
, void *txarg
);
468 extern int zil_claim(const char *osname
, void *txarg
);
469 extern int zil_check_log_chain(const char *osname
, void *txarg
);
470 extern void zil_sync(zilog_t
*zilog
, dmu_tx_t
*tx
);
471 extern void zil_clean(zilog_t
*zilog
, uint64_t synced_txg
);
473 extern int zil_suspend(zilog_t
*zilog
);
474 extern void zil_resume(zilog_t
*zilog
);
476 extern void zil_add_block(zilog_t
*zilog
, const blkptr_t
*bp
);
477 extern int zil_bp_tree_add(zilog_t
*zilog
, const blkptr_t
*bp
);
479 extern void zil_set_sync(zilog_t
*zilog
, uint64_t syncval
);
481 extern void zil_set_logbias(zilog_t
*zilog
, uint64_t slogval
);
483 extern int zil_replay_disable
;
489 #endif /* _SYS_ZIL_H */