xref: /freebsd/sys/contrib/openzfs/include/sys/zil.h (revision 4e8d558c)
1 /*
2  * CDDL HEADER START
3  *
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.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
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]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
24  */
25 
26 /* Portions Copyright 2010 Robert Milkowski */
27 
28 #ifndef	_SYS_ZIL_H
29 #define	_SYS_ZIL_H
30 
31 #include <sys/types.h>
32 #include <sys/spa.h>
33 #include <sys/zio.h>
34 #include <sys/dmu.h>
35 #include <sys/zio_crypt.h>
36 #include <sys/wmsum.h>
37 
38 #ifdef	__cplusplus
39 extern "C" {
40 #endif
41 
42 struct dsl_pool;
43 struct dsl_dataset;
44 struct lwb;
45 
46 /*
47  * Intent log format:
48  *
49  * Each objset has its own intent log.  The log header (zil_header_t)
50  * for objset N's intent log is kept in the Nth object of the SPA's
51  * intent_log objset.  The log header points to a chain of log blocks,
52  * each of which contains log records (i.e., transactions) followed by
53  * a log block trailer (zil_trailer_t).  The format of a log record
54  * depends on the record (or transaction) type, but all records begin
55  * with a common structure that defines the type, length, and txg.
56  */
57 
58 /*
59  * Intent log header - this on disk structure holds fields to manage
60  * the log.  All fields are 64 bit to easily handle cross architectures.
61  */
62 typedef struct zil_header {
63 	uint64_t zh_claim_txg;	/* txg in which log blocks were claimed */
64 	uint64_t zh_replay_seq;	/* highest replayed sequence number */
65 	blkptr_t zh_log;	/* log chain */
66 	uint64_t zh_claim_blk_seq; /* highest claimed block sequence number */
67 	uint64_t zh_flags;	/* header flags */
68 	uint64_t zh_claim_lr_seq; /* highest claimed lr sequence number */
69 	uint64_t zh_pad[3];
70 } zil_header_t;
71 
72 /*
73  * zh_flags bit settings
74  */
75 #define	ZIL_REPLAY_NEEDED	0x1	/* replay needed - internal only */
76 #define	ZIL_CLAIM_LR_SEQ_VALID	0x2	/* zh_claim_lr_seq field is valid */
77 
78 /*
79  * Log block chaining.
80  *
81  * Log blocks are chained together. Originally they were chained at the
82  * end of the block. For performance reasons the chain was moved to the
83  * beginning of the block which allows writes for only the data being used.
84  * The older position is supported for backwards compatibility.
85  *
86  * The zio_eck_t contains a zec_cksum which for the intent log is
87  * the sequence number of this log block. A seq of 0 is invalid.
88  * The zec_cksum is checked by the SPA against the sequence
89  * number passed in the blk_cksum field of the blkptr_t
90  */
91 typedef struct zil_chain {
92 	uint64_t zc_pad;
93 	blkptr_t zc_next_blk;	/* next block in chain */
94 	uint64_t zc_nused;	/* bytes in log block used */
95 	zio_eck_t zc_eck;	/* block trailer */
96 } zil_chain_t;
97 
98 #define	ZIL_MIN_BLKSZ	4096ULL
99 
100 /*
101  * ziltest is by and large an ugly hack, but very useful in
102  * checking replay without tedious work.
103  * When running ziltest we want to keep all itx's and so maintain
104  * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG
105  * We subtract TXG_CONCURRENT_STATES to allow for common code.
106  */
107 #define	ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES)
108 
109 /*
110  * The words of a log block checksum.
111  */
112 #define	ZIL_ZC_GUID_0	0
113 #define	ZIL_ZC_GUID_1	1
114 #define	ZIL_ZC_OBJSET	2
115 #define	ZIL_ZC_SEQ	3
116 
117 typedef enum zil_create {
118 	Z_FILE,
119 	Z_DIR,
120 	Z_XATTRDIR,
121 } zil_create_t;
122 
123 /*
124  * size of xvattr log section.
125  * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
126  * for create time and a single 64 bit integer for all of the attributes,
127  * and 4 64 bit integers (32 bytes) for the scanstamp.
128  *
129  */
130 
131 #define	ZIL_XVAT_SIZE(mapsize) \
132 	sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
133 	(sizeof (uint64_t) * 7)
134 
135 /*
136  * Size of ACL in log.  The ACE data is padded out to properly align
137  * on 8 byte boundary.
138  */
139 
140 #define	ZIL_ACE_LENGTH(x)	(roundup(x, sizeof (uint64_t)))
141 
142 /*
143  * Intent log transaction types and record structures
144  */
145 #define	TX_COMMIT		0	/* Commit marker (no on-disk state) */
146 #define	TX_CREATE		1	/* Create file */
147 #define	TX_MKDIR		2	/* Make directory */
148 #define	TX_MKXATTR		3	/* Make XATTR directory */
149 #define	TX_SYMLINK		4	/* Create symbolic link to a file */
150 #define	TX_REMOVE		5	/* Remove file */
151 #define	TX_RMDIR		6	/* Remove directory */
152 #define	TX_LINK			7	/* Create hard link to a file */
153 #define	TX_RENAME		8	/* Rename a file */
154 #define	TX_WRITE		9	/* File write */
155 #define	TX_TRUNCATE		10	/* Truncate a file */
156 #define	TX_SETATTR		11	/* Set file attributes */
157 #define	TX_ACL_V0		12	/* Set old formatted ACL */
158 #define	TX_ACL			13	/* Set ACL */
159 #define	TX_CREATE_ACL		14	/* create with ACL */
160 #define	TX_CREATE_ATTR		15	/* create + attrs */
161 #define	TX_CREATE_ACL_ATTR 	16	/* create with ACL + attrs */
162 #define	TX_MKDIR_ACL		17	/* mkdir with ACL */
163 #define	TX_MKDIR_ATTR		18	/* mkdir with attr */
164 #define	TX_MKDIR_ACL_ATTR	19	/* mkdir with ACL + attrs */
165 #define	TX_WRITE2		20	/* dmu_sync EALREADY write */
166 #define	TX_SETSAXATTR		21	/* Set sa xattrs on file */
167 #define	TX_RENAME_EXCHANGE	22	/* Atomic swap via renameat2 */
168 #define	TX_RENAME_WHITEOUT	23	/* Atomic whiteout via renameat2 */
169 #define	TX_CLONE_RANGE		24	/* Clone a file range */
170 #define	TX_MAX_TYPE		25	/* Max transaction type */
171 
172 /*
173  * The transactions for mkdir, symlink, remove, rmdir, link, and rename
174  * may have the following bit set, indicating the original request
175  * specified case-insensitive handling of names.
176  */
177 #define	TX_CI	((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
178 
179 /*
180  * Transactions for operations below can be logged out of order.
181  * For convenience in the code, all such records must have lr_foid
182  * at the same offset.
183  */
184 #define	TX_OOO(txtype)			\
185 	((txtype) == TX_WRITE ||	\
186 	(txtype) == TX_TRUNCATE ||	\
187 	(txtype) == TX_SETATTR ||	\
188 	(txtype) == TX_ACL_V0 ||	\
189 	(txtype) == TX_ACL ||		\
190 	(txtype) == TX_WRITE2 ||	\
191 	(txtype) == TX_SETSAXATTR ||	\
192 	(txtype) == TX_CLONE_RANGE)
193 
194 /*
195  * The number of dnode slots consumed by the object is stored in the 8
196  * unused upper bits of the object ID. We subtract 1 from the value
197  * stored on disk for compatibility with implementations that don't
198  * support large dnodes. The slot count for a single-slot dnode will
199  * contain 0 for those bits to preserve the log record format for
200  * "small" dnodes.
201  */
202 #define	LR_FOID_GET_SLOTS(oid) (BF64_GET((oid), 56, 8) + 1)
203 #define	LR_FOID_SET_SLOTS(oid, x) BF64_SET((oid), 56, 8, (x) - 1)
204 #define	LR_FOID_GET_OBJ(oid) BF64_GET((oid), 0, DN_MAX_OBJECT_SHIFT)
205 #define	LR_FOID_SET_OBJ(oid, x) BF64_SET((oid), 0, DN_MAX_OBJECT_SHIFT, (x))
206 
207 /*
208  * Format of log records.
209  * The fields are carefully defined to allow them to be aligned
210  * and sized the same on sparc & intel architectures.
211  * Each log record has a common structure at the beginning.
212  *
213  * The log record on disk (lrc_seq) holds the sequence number of all log
214  * records which is used to ensure we don't replay the same record.
215  */
216 typedef struct {			/* common log record header */
217 	uint64_t	lrc_txtype;	/* intent log transaction type */
218 	uint64_t	lrc_reclen;	/* transaction record length */
219 	uint64_t	lrc_txg;	/* dmu transaction group number */
220 	uint64_t	lrc_seq;	/* see comment above */
221 } lr_t;
222 
223 /*
224  * Common start of all out-of-order record types (TX_OOO() above).
225  */
226 typedef struct {
227 	lr_t		lr_common;	/* common portion of log record */
228 	uint64_t	lr_foid;	/* object id */
229 } lr_ooo_t;
230 
231 /*
232  * Additional lr_attr_t fields.
233  */
234 typedef struct {
235 	uint64_t	lr_attr_attrs;		/* all of the attributes */
236 	uint64_t	lr_attr_crtime[2];	/* create time */
237 	uint8_t		lr_attr_scanstamp[32];
238 } lr_attr_end_t;
239 
240 /*
241  * Handle option extended vattr attributes.
242  *
243  * Whenever new attributes are added the version number
244  * will need to be updated as will code in
245  * zfs_log.c and zfs_replay.c
246  */
247 typedef struct {
248 	uint32_t	lr_attr_masksize; /* number of elements in array */
249 	uint32_t	lr_attr_bitmap; /* First entry of array */
250 	/* remainder of array and additional lr_attr_end_t fields */
251 } lr_attr_t;
252 
253 /*
254  * log record for creates without optional ACL.
255  * This log record does support optional xvattr_t attributes.
256  */
257 typedef struct {
258 	lr_t		lr_common;	/* common portion of log record */
259 	uint64_t	lr_doid;	/* object id of directory */
260 	uint64_t	lr_foid;	/* object id of created file object */
261 	uint64_t	lr_mode;	/* mode of object */
262 	uint64_t	lr_uid;		/* uid of object */
263 	uint64_t	lr_gid;		/* gid of object */
264 	uint64_t	lr_gen;		/* generation (txg of creation) */
265 	uint64_t	lr_crtime[2];	/* creation time */
266 	uint64_t	lr_rdev;	/* rdev of object to create */
267 	/* name of object to create follows this */
268 	/* for symlinks, link content follows name */
269 	/* for creates with xvattr data, the name follows the xvattr info */
270 } lr_create_t;
271 
272 /*
273  * FUID ACL record will be an array of ACEs from the original ACL.
274  * If this array includes ephemeral IDs, the record will also include
275  * an array of log-specific FUIDs to replace the ephemeral IDs.
276  * Only one copy of each unique domain will be present, so the log-specific
277  * FUIDs will use an index into a compressed domain table.  On replay this
278  * information will be used to construct real FUIDs (and bypass idmap,
279  * since it may not be available).
280  */
281 
282 /*
283  * Log record for creates with optional ACL
284  * This log record is also used for recording any FUID
285  * information needed for replaying the create.  If the
286  * file doesn't have any actual ACEs then the lr_aclcnt
287  * would be zero.
288  *
289  * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
290  * If create is also setting xvattr's, then acl data follows xvattr.
291  * If ACE FUIDs are needed then they will follow the xvattr_t.  Following
292  * the FUIDs will be the domain table information.  The FUIDs for the owner
293  * and group will be in lr_create.  Name follows ACL data.
294  */
295 typedef struct {
296 	lr_create_t	lr_create;	/* common create portion */
297 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
298 	uint64_t	lr_domcnt;	/* number of unique domains */
299 	uint64_t	lr_fuidcnt;	/* number of real fuids */
300 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
301 	uint64_t	lr_acl_flags;	/* ACL flags */
302 } lr_acl_create_t;
303 
304 typedef struct {
305 	lr_t		lr_common;	/* common portion of log record */
306 	uint64_t	lr_doid;	/* obj id of directory */
307 	/* name of object to remove follows this */
308 } lr_remove_t;
309 
310 typedef struct {
311 	lr_t		lr_common;	/* common portion of log record */
312 	uint64_t	lr_doid;	/* obj id of directory */
313 	uint64_t	lr_link_obj;	/* obj id of link */
314 	/* name of object to link follows this */
315 } lr_link_t;
316 
317 typedef struct {
318 	lr_t		lr_common;	/* common portion of log record */
319 	uint64_t	lr_sdoid;	/* obj id of source directory */
320 	uint64_t	lr_tdoid;	/* obj id of target directory */
321 	/* 2 strings: names of source and destination follow this */
322 } lr_rename_t;
323 
324 typedef struct {
325 	lr_rename_t	lr_rename;	/* common rename portion */
326 	/* members related to the whiteout file (based on lr_create_t) */
327 	uint64_t	lr_wfoid;	/* obj id of the new whiteout file */
328 	uint64_t	lr_wmode;	/* mode of object */
329 	uint64_t	lr_wuid;	/* uid of whiteout */
330 	uint64_t	lr_wgid;	/* gid of whiteout */
331 	uint64_t	lr_wgen;	/* generation (txg of creation) */
332 	uint64_t	lr_wcrtime[2];	/* creation time */
333 	uint64_t	lr_wrdev;	/* always makedev(0, 0) */
334 	/* 2 strings: names of source and destination follow this */
335 } lr_rename_whiteout_t;
336 
337 typedef struct {
338 	lr_t		lr_common;	/* common portion of log record */
339 	uint64_t	lr_foid;	/* file object to write */
340 	uint64_t	lr_offset;	/* offset to write to */
341 	uint64_t	lr_length;	/* user data length to write */
342 	uint64_t	lr_blkoff;	/* no longer used */
343 	blkptr_t	lr_blkptr;	/* spa block pointer for replay */
344 	/* write data will follow for small writes */
345 } lr_write_t;
346 
347 typedef struct {
348 	lr_t		lr_common;	/* common portion of log record */
349 	uint64_t	lr_foid;	/* object id of file to truncate */
350 	uint64_t	lr_offset;	/* offset to truncate from */
351 	uint64_t	lr_length;	/* length to truncate */
352 } lr_truncate_t;
353 
354 typedef struct {
355 	lr_t		lr_common;	/* common portion of log record */
356 	uint64_t	lr_foid;	/* file object to change attributes */
357 	uint64_t	lr_mask;	/* mask of attributes to set */
358 	uint64_t	lr_mode;	/* mode to set */
359 	uint64_t	lr_uid;		/* uid to set */
360 	uint64_t	lr_gid;		/* gid to set */
361 	uint64_t	lr_size;	/* size to set */
362 	uint64_t	lr_atime[2];	/* access time */
363 	uint64_t	lr_mtime[2];	/* modification time */
364 	/* optional attribute lr_attr_t may be here */
365 } lr_setattr_t;
366 
367 typedef struct {
368 	lr_t		lr_common;	/* common portion of log record */
369 	uint64_t	lr_foid;	/* file object to change attributes */
370 	uint64_t	lr_size;
371 	/* xattr name and value follows */
372 } lr_setsaxattr_t;
373 
374 typedef struct {
375 	lr_t		lr_common;	/* common portion of log record */
376 	uint64_t	lr_foid;	/* obj id of file */
377 	uint64_t	lr_aclcnt;	/* number of acl entries */
378 	/* lr_aclcnt number of ace_t entries follow this */
379 } lr_acl_v0_t;
380 
381 typedef struct {
382 	lr_t		lr_common;	/* common portion of log record */
383 	uint64_t	lr_foid;	/* obj id of file */
384 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
385 	uint64_t	lr_domcnt;	/* number of unique domains */
386 	uint64_t	lr_fuidcnt;	/* number of real fuids */
387 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
388 	uint64_t	lr_acl_flags;	/* ACL flags */
389 	/* lr_acl_bytes number of variable sized ace's follows */
390 } lr_acl_t;
391 
392 typedef struct {
393 	lr_t		lr_common;	/* common portion of log record */
394 	uint64_t	lr_foid;	/* file object to clone into */
395 	uint64_t	lr_offset;	/* offset to clone to */
396 	uint64_t	lr_length;	/* length of the blocks to clone */
397 	uint64_t	lr_blksz;	/* file's block size */
398 	uint64_t	lr_nbps;	/* number of block pointers */
399 	blkptr_t	lr_bps[];
400 	/* block pointers of the blocks to clone follows */
401 } lr_clone_range_t;
402 
403 /*
404  * ZIL structure definitions, interface function prototype and globals.
405  */
406 
407 /*
408  * Writes are handled in three different ways:
409  *
410  * WR_INDIRECT:
411  *    In this mode, if we need to commit the write later, then the block
412  *    is immediately written into the file system (using dmu_sync),
413  *    and a pointer to the block is put into the log record.
414  *    When the txg commits the block is linked in.
415  *    This saves additionally writing the data into the log record.
416  *    There are a few requirements for this to occur:
417  *	- write is greater than zfs/zvol_immediate_write_sz
418  *	- not using slogs (as slogs are assumed to always be faster
419  *	  than writing into the main pool)
420  *	- the write occupies only one block
421  * WR_COPIED:
422  *    If we know we'll immediately be committing the
423  *    transaction (O_SYNC or O_DSYNC), then we allocate a larger
424  *    log record here for the data and copy the data in.
425  * WR_NEED_COPY:
426  *    Otherwise we don't allocate a buffer, and *if* we need to
427  *    flush the write later then a buffer is allocated and
428  *    we retrieve the data using the dmu.
429  */
430 typedef enum {
431 	WR_INDIRECT,	/* indirect - a large write (dmu_sync() data */
432 			/* and put blkptr in log, rather than actual data) */
433 	WR_COPIED,	/* immediate - data is copied into lr_write_t */
434 	WR_NEED_COPY,	/* immediate - data needs to be copied if pushed */
435 	WR_NUM_STATES	/* number of states */
436 } itx_wr_state_t;
437 
438 typedef void (*zil_callback_t)(void *data);
439 
440 typedef struct itx {
441 	list_node_t	itx_node;	/* linkage on zl_itx_list */
442 	void		*itx_private;	/* type-specific opaque data */
443 	itx_wr_state_t	itx_wr_state;	/* write state */
444 	uint8_t		itx_sync;	/* synchronous transaction */
445 	zil_callback_t	itx_callback;   /* Called when the itx is persistent */
446 	void		*itx_callback_data; /* User data for the callback */
447 	size_t		itx_size;	/* allocated itx structure size */
448 	uint64_t	itx_oid;	/* object id */
449 	uint64_t	itx_gen;	/* gen number for zfs_get_data */
450 	lr_t		itx_lr;		/* common part of log record */
451 	/* followed by type-specific part of lr_xx_t and its immediate data */
452 } itx_t;
453 
454 /*
455  * Used for zil kstat.
456  */
457 typedef struct zil_stats {
458 	/*
459 	 * Number of times a ZIL commit (e.g. fsync) has been requested.
460 	 */
461 	kstat_named_t zil_commit_count;
462 
463 	/*
464 	 * Number of times the ZIL has been flushed to stable storage.
465 	 * This is less than zil_commit_count when commits are "merged"
466 	 * (see the documentation above zil_commit()).
467 	 */
468 	kstat_named_t zil_commit_writer_count;
469 
470 	/*
471 	 * Number of transactions (reads, writes, renames, etc.)
472 	 * that have been committed.
473 	 */
474 	kstat_named_t zil_itx_count;
475 
476 	/*
477 	 * See the documentation for itx_wr_state_t above.
478 	 * Note that "bytes" accumulates the length of the transactions
479 	 * (i.e. data), not the actual log record sizes.
480 	 */
481 	kstat_named_t zil_itx_indirect_count;
482 	kstat_named_t zil_itx_indirect_bytes;
483 	kstat_named_t zil_itx_copied_count;
484 	kstat_named_t zil_itx_copied_bytes;
485 	kstat_named_t zil_itx_needcopy_count;
486 	kstat_named_t zil_itx_needcopy_bytes;
487 
488 	/*
489 	 * Transactions which have been allocated to the "normal"
490 	 * (i.e. not slog) storage pool. Note that "bytes" accumulate
491 	 * the actual log record sizes - which do not include the actual
492 	 * data in case of indirect writes.  bytes <= write <= alloc.
493 	 */
494 	kstat_named_t zil_itx_metaslab_normal_count;
495 	kstat_named_t zil_itx_metaslab_normal_bytes;
496 	kstat_named_t zil_itx_metaslab_normal_write;
497 	kstat_named_t zil_itx_metaslab_normal_alloc;
498 
499 	/*
500 	 * Transactions which have been allocated to the "slog" storage pool.
501 	 * If there are no separate log devices, this is the same as the
502 	 * "normal" pool.  bytes <= write <= alloc.
503 	 */
504 	kstat_named_t zil_itx_metaslab_slog_count;
505 	kstat_named_t zil_itx_metaslab_slog_bytes;
506 	kstat_named_t zil_itx_metaslab_slog_write;
507 	kstat_named_t zil_itx_metaslab_slog_alloc;
508 } zil_kstat_values_t;
509 
510 typedef struct zil_sums {
511 	wmsum_t zil_commit_count;
512 	wmsum_t zil_commit_writer_count;
513 	wmsum_t zil_itx_count;
514 	wmsum_t zil_itx_indirect_count;
515 	wmsum_t zil_itx_indirect_bytes;
516 	wmsum_t zil_itx_copied_count;
517 	wmsum_t zil_itx_copied_bytes;
518 	wmsum_t zil_itx_needcopy_count;
519 	wmsum_t zil_itx_needcopy_bytes;
520 	wmsum_t zil_itx_metaslab_normal_count;
521 	wmsum_t zil_itx_metaslab_normal_bytes;
522 	wmsum_t zil_itx_metaslab_normal_write;
523 	wmsum_t zil_itx_metaslab_normal_alloc;
524 	wmsum_t zil_itx_metaslab_slog_count;
525 	wmsum_t zil_itx_metaslab_slog_bytes;
526 	wmsum_t zil_itx_metaslab_slog_write;
527 	wmsum_t zil_itx_metaslab_slog_alloc;
528 } zil_sums_t;
529 
530 #define	ZIL_STAT_INCR(zil, stat, val) \
531 	do { \
532 		int64_t tmpval = (val); \
533 		wmsum_add(&(zil_sums_global.stat), tmpval); \
534 		if ((zil)->zl_sums) \
535 			wmsum_add(&((zil)->zl_sums->stat), tmpval); \
536 	} while (0)
537 
538 #define	ZIL_STAT_BUMP(zil, stat) \
539     ZIL_STAT_INCR(zil, stat, 1);
540 
541 typedef int zil_parse_blk_func_t(zilog_t *zilog, const blkptr_t *bp, void *arg,
542     uint64_t txg);
543 typedef int zil_parse_lr_func_t(zilog_t *zilog, const lr_t *lr, void *arg,
544     uint64_t txg);
545 typedef int zil_replay_func_t(void *arg1, void *arg2, boolean_t byteswap);
546 typedef int zil_get_data_t(void *arg, uint64_t arg2, lr_write_t *lr, char *dbuf,
547     struct lwb *lwb, zio_t *zio);
548 
549 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
550     zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg,
551     boolean_t decrypt);
552 
553 extern void	zil_init(void);
554 extern void	zil_fini(void);
555 
556 extern zilog_t	*zil_alloc(objset_t *os, zil_header_t *zh_phys);
557 extern void	zil_free(zilog_t *zilog);
558 
559 extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data,
560     zil_sums_t *zil_sums);
561 extern void	zil_close(zilog_t *zilog);
562 
563 extern boolean_t zil_replay(objset_t *os, void *arg,
564     zil_replay_func_t *const replay_func[TX_MAX_TYPE]);
565 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
566 extern boolean_t zil_destroy(zilog_t *zilog, boolean_t keep_first);
567 extern void	zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
568 
569 extern itx_t	*zil_itx_create(uint64_t txtype, size_t lrsize);
570 extern void	zil_itx_destroy(itx_t *itx);
571 extern void	zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
572 
573 extern void	zil_async_to_sync(zilog_t *zilog, uint64_t oid);
574 extern void	zil_commit(zilog_t *zilog, uint64_t oid);
575 extern void	zil_commit_impl(zilog_t *zilog, uint64_t oid);
576 extern void	zil_remove_async(zilog_t *zilog, uint64_t oid);
577 
578 extern int	zil_reset(const char *osname, void *txarg);
579 extern int	zil_claim(struct dsl_pool *dp,
580     struct dsl_dataset *ds, void *txarg);
581 extern int 	zil_check_log_chain(struct dsl_pool *dp,
582     struct dsl_dataset *ds, void *tx);
583 extern void	zil_sync(zilog_t *zilog, dmu_tx_t *tx);
584 extern void	zil_clean(zilog_t *zilog, uint64_t synced_txg);
585 
586 extern int	zil_suspend(const char *osname, void **cookiep);
587 extern void	zil_resume(void *cookie);
588 
589 extern void	zil_lwb_add_block(struct lwb *lwb, const blkptr_t *bp);
590 extern void	zil_lwb_add_txg(struct lwb *lwb, uint64_t txg);
591 extern int	zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
592 
593 extern void	zil_set_sync(zilog_t *zilog, uint64_t syncval);
594 
595 extern void	zil_set_logbias(zilog_t *zilog, uint64_t slogval);
596 
597 extern uint64_t	zil_max_copied_data(zilog_t *zilog);
598 extern uint64_t	zil_max_log_data(zilog_t *zilog, size_t hdrsize);
599 
600 extern void zil_sums_init(zil_sums_t *zs);
601 extern void zil_sums_fini(zil_sums_t *zs);
602 extern void zil_kstat_values_update(zil_kstat_values_t *zs,
603     zil_sums_t *zil_sums);
604 
605 extern int zil_replay_disable;
606 
607 #ifdef	__cplusplus
608 }
609 #endif
610 
611 #endif	/* _SYS_ZIL_H */
612