xref: /freebsd/sys/contrib/openzfs/include/sys/zil.h (revision dbd5678d)
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_MAX_TYPE		24	/* Max transaction type */
170 
171 /*
172  * The transactions for mkdir, symlink, remove, rmdir, link, and rename
173  * may have the following bit set, indicating the original request
174  * specified case-insensitive handling of names.
175  */
176 #define	TX_CI	((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
177 
178 /*
179  * Transactions for write, truncate, setattr, acl_v0, and acl can be logged
180  * out of order.  For convenience in the code, all such records must have
181  * lr_foid at the same offset.
182  */
183 #define	TX_OOO(txtype)			\
184 	((txtype) == TX_WRITE ||	\
185 	(txtype) == TX_TRUNCATE ||	\
186 	(txtype) == TX_SETATTR ||	\
187 	(txtype) == TX_ACL_V0 ||	\
188 	(txtype) == TX_ACL ||		\
189 	(txtype) == TX_WRITE2 ||	\
190 	(txtype) == TX_SETSAXATTR)
191 
192 /*
193  * The number of dnode slots consumed by the object is stored in the 8
194  * unused upper bits of the object ID. We subtract 1 from the value
195  * stored on disk for compatibility with implementations that don't
196  * support large dnodes. The slot count for a single-slot dnode will
197  * contain 0 for those bits to preserve the log record format for
198  * "small" dnodes.
199  */
200 #define	LR_FOID_GET_SLOTS(oid) (BF64_GET((oid), 56, 8) + 1)
201 #define	LR_FOID_SET_SLOTS(oid, x) BF64_SET((oid), 56, 8, (x) - 1)
202 #define	LR_FOID_GET_OBJ(oid) BF64_GET((oid), 0, DN_MAX_OBJECT_SHIFT)
203 #define	LR_FOID_SET_OBJ(oid, x) BF64_SET((oid), 0, DN_MAX_OBJECT_SHIFT, (x))
204 
205 /*
206  * Format of log records.
207  * The fields are carefully defined to allow them to be aligned
208  * and sized the same on sparc & intel architectures.
209  * Each log record has a common structure at the beginning.
210  *
211  * The log record on disk (lrc_seq) holds the sequence number of all log
212  * records which is used to ensure we don't replay the same record.
213  */
214 typedef struct {			/* common log record header */
215 	uint64_t	lrc_txtype;	/* intent log transaction type */
216 	uint64_t	lrc_reclen;	/* transaction record length */
217 	uint64_t	lrc_txg;	/* dmu transaction group number */
218 	uint64_t	lrc_seq;	/* see comment above */
219 } lr_t;
220 
221 /*
222  * Common start of all out-of-order record types (TX_OOO() above).
223  */
224 typedef struct {
225 	lr_t		lr_common;	/* common portion of log record */
226 	uint64_t	lr_foid;	/* object id */
227 } lr_ooo_t;
228 
229 /*
230  * Additional lr_attr_t fields.
231  */
232 typedef struct {
233 	uint64_t	lr_attr_attrs;		/* all of the attributes */
234 	uint64_t	lr_attr_crtime[2];	/* create time */
235 	uint8_t		lr_attr_scanstamp[32];
236 } lr_attr_end_t;
237 
238 /*
239  * Handle option extended vattr attributes.
240  *
241  * Whenever new attributes are added the version number
242  * will need to be updated as will code in
243  * zfs_log.c and zfs_replay.c
244  */
245 typedef struct {
246 	uint32_t	lr_attr_masksize; /* number of elements in array */
247 	uint32_t	lr_attr_bitmap; /* First entry of array */
248 	/* remainder of array and additional lr_attr_end_t fields */
249 } lr_attr_t;
250 
251 /*
252  * log record for creates without optional ACL.
253  * This log record does support optional xvattr_t attributes.
254  */
255 typedef struct {
256 	lr_t		lr_common;	/* common portion of log record */
257 	uint64_t	lr_doid;	/* object id of directory */
258 	uint64_t	lr_foid;	/* object id of created file object */
259 	uint64_t	lr_mode;	/* mode of object */
260 	uint64_t	lr_uid;		/* uid of object */
261 	uint64_t	lr_gid;		/* gid of object */
262 	uint64_t	lr_gen;		/* generation (txg of creation) */
263 	uint64_t	lr_crtime[2];	/* creation time */
264 	uint64_t	lr_rdev;	/* rdev of object to create */
265 	/* name of object to create follows this */
266 	/* for symlinks, link content follows name */
267 	/* for creates with xvattr data, the name follows the xvattr info */
268 } lr_create_t;
269 
270 /*
271  * FUID ACL record will be an array of ACEs from the original ACL.
272  * If this array includes ephemeral IDs, the record will also include
273  * an array of log-specific FUIDs to replace the ephemeral IDs.
274  * Only one copy of each unique domain will be present, so the log-specific
275  * FUIDs will use an index into a compressed domain table.  On replay this
276  * information will be used to construct real FUIDs (and bypass idmap,
277  * since it may not be available).
278  */
279 
280 /*
281  * Log record for creates with optional ACL
282  * This log record is also used for recording any FUID
283  * information needed for replaying the create.  If the
284  * file doesn't have any actual ACEs then the lr_aclcnt
285  * would be zero.
286  *
287  * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
288  * If create is also setting xvattr's, then acl data follows xvattr.
289  * If ACE FUIDs are needed then they will follow the xvattr_t.  Following
290  * the FUIDs will be the domain table information.  The FUIDs for the owner
291  * and group will be in lr_create.  Name follows ACL data.
292  */
293 typedef struct {
294 	lr_create_t	lr_create;	/* common create portion */
295 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
296 	uint64_t	lr_domcnt;	/* number of unique domains */
297 	uint64_t	lr_fuidcnt;	/* number of real fuids */
298 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
299 	uint64_t	lr_acl_flags;	/* ACL flags */
300 } lr_acl_create_t;
301 
302 typedef struct {
303 	lr_t		lr_common;	/* common portion of log record */
304 	uint64_t	lr_doid;	/* obj id of directory */
305 	/* name of object to remove follows this */
306 } lr_remove_t;
307 
308 typedef struct {
309 	lr_t		lr_common;	/* common portion of log record */
310 	uint64_t	lr_doid;	/* obj id of directory */
311 	uint64_t	lr_link_obj;	/* obj id of link */
312 	/* name of object to link follows this */
313 } lr_link_t;
314 
315 typedef struct {
316 	lr_t		lr_common;	/* common portion of log record */
317 	uint64_t	lr_sdoid;	/* obj id of source directory */
318 	uint64_t	lr_tdoid;	/* obj id of target directory */
319 	/* 2 strings: names of source and destination follow this */
320 } lr_rename_t;
321 
322 typedef struct {
323 	lr_rename_t	lr_rename;	/* common rename portion */
324 	/* members related to the whiteout file (based on lr_create_t) */
325 	uint64_t	lr_wfoid;	/* obj id of the new whiteout file */
326 	uint64_t	lr_wmode;	/* mode of object */
327 	uint64_t	lr_wuid;	/* uid of whiteout */
328 	uint64_t	lr_wgid;	/* gid of whiteout */
329 	uint64_t	lr_wgen;	/* generation (txg of creation) */
330 	uint64_t	lr_wcrtime[2];	/* creation time */
331 	uint64_t	lr_wrdev;	/* always makedev(0, 0) */
332 	/* 2 strings: names of source and destination follow this */
333 } lr_rename_whiteout_t;
334 
335 typedef struct {
336 	lr_t		lr_common;	/* common portion of log record */
337 	uint64_t	lr_foid;	/* file object to write */
338 	uint64_t	lr_offset;	/* offset to write to */
339 	uint64_t	lr_length;	/* user data length to write */
340 	uint64_t	lr_blkoff;	/* no longer used */
341 	blkptr_t	lr_blkptr;	/* spa block pointer for replay */
342 	/* write data will follow for small writes */
343 } lr_write_t;
344 
345 typedef struct {
346 	lr_t		lr_common;	/* common portion of log record */
347 	uint64_t	lr_foid;	/* object id of file to truncate */
348 	uint64_t	lr_offset;	/* offset to truncate from */
349 	uint64_t	lr_length;	/* length to truncate */
350 } lr_truncate_t;
351 
352 typedef struct {
353 	lr_t		lr_common;	/* common portion of log record */
354 	uint64_t	lr_foid;	/* file object to change attributes */
355 	uint64_t	lr_mask;	/* mask of attributes to set */
356 	uint64_t	lr_mode;	/* mode to set */
357 	uint64_t	lr_uid;		/* uid to set */
358 	uint64_t	lr_gid;		/* gid to set */
359 	uint64_t	lr_size;	/* size to set */
360 	uint64_t	lr_atime[2];	/* access time */
361 	uint64_t	lr_mtime[2];	/* modification time */
362 	/* optional attribute lr_attr_t may be here */
363 } lr_setattr_t;
364 
365 typedef struct {
366 	lr_t		lr_common;	/* common portion of log record */
367 	uint64_t	lr_foid;	/* file object to change attributes */
368 	uint64_t	lr_size;
369 	/* xattr name and value follows */
370 } lr_setsaxattr_t;
371 
372 typedef struct {
373 	lr_t		lr_common;	/* common portion of log record */
374 	uint64_t	lr_foid;	/* obj id of file */
375 	uint64_t	lr_aclcnt;	/* number of acl entries */
376 	/* lr_aclcnt number of ace_t entries follow this */
377 } lr_acl_v0_t;
378 
379 typedef struct {
380 	lr_t		lr_common;	/* common portion of log record */
381 	uint64_t	lr_foid;	/* obj id of file */
382 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
383 	uint64_t	lr_domcnt;	/* number of unique domains */
384 	uint64_t	lr_fuidcnt;	/* number of real fuids */
385 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
386 	uint64_t	lr_acl_flags;	/* ACL flags */
387 	/* lr_acl_bytes number of variable sized ace's follows */
388 } lr_acl_t;
389 
390 /*
391  * ZIL structure definitions, interface function prototype and globals.
392  */
393 
394 /*
395  * Writes are handled in three different ways:
396  *
397  * WR_INDIRECT:
398  *    In this mode, if we need to commit the write later, then the block
399  *    is immediately written into the file system (using dmu_sync),
400  *    and a pointer to the block is put into the log record.
401  *    When the txg commits the block is linked in.
402  *    This saves additionally writing the data into the log record.
403  *    There are a few requirements for this to occur:
404  *	- write is greater than zfs/zvol_immediate_write_sz
405  *	- not using slogs (as slogs are assumed to always be faster
406  *	  than writing into the main pool)
407  *	- the write occupies only one block
408  * WR_COPIED:
409  *    If we know we'll immediately be committing the
410  *    transaction (O_SYNC or O_DSYNC), then we allocate a larger
411  *    log record here for the data and copy the data in.
412  * WR_NEED_COPY:
413  *    Otherwise we don't allocate a buffer, and *if* we need to
414  *    flush the write later then a buffer is allocated and
415  *    we retrieve the data using the dmu.
416  */
417 typedef enum {
418 	WR_INDIRECT,	/* indirect - a large write (dmu_sync() data */
419 			/* and put blkptr in log, rather than actual data) */
420 	WR_COPIED,	/* immediate - data is copied into lr_write_t */
421 	WR_NEED_COPY,	/* immediate - data needs to be copied if pushed */
422 	WR_NUM_STATES	/* number of states */
423 } itx_wr_state_t;
424 
425 typedef void (*zil_callback_t)(void *data);
426 
427 typedef struct itx {
428 	list_node_t	itx_node;	/* linkage on zl_itx_list */
429 	void		*itx_private;	/* type-specific opaque data */
430 	itx_wr_state_t	itx_wr_state;	/* write state */
431 	uint8_t		itx_sync;	/* synchronous transaction */
432 	zil_callback_t	itx_callback;   /* Called when the itx is persistent */
433 	void		*itx_callback_data; /* User data for the callback */
434 	size_t		itx_size;	/* allocated itx structure size */
435 	uint64_t	itx_oid;	/* object id */
436 	uint64_t	itx_gen;	/* gen number for zfs_get_data */
437 	lr_t		itx_lr;		/* common part of log record */
438 	/* followed by type-specific part of lr_xx_t and its immediate data */
439 } itx_t;
440 
441 /*
442  * Used for zil kstat.
443  */
444 typedef struct zil_stats {
445 	/*
446 	 * Number of times a ZIL commit (e.g. fsync) has been requested.
447 	 */
448 	kstat_named_t zil_commit_count;
449 
450 	/*
451 	 * Number of times the ZIL has been flushed to stable storage.
452 	 * This is less than zil_commit_count when commits are "merged"
453 	 * (see the documentation above zil_commit()).
454 	 */
455 	kstat_named_t zil_commit_writer_count;
456 
457 	/*
458 	 * Number of transactions (reads, writes, renames, etc.)
459 	 * that have been committed.
460 	 */
461 	kstat_named_t zil_itx_count;
462 
463 	/*
464 	 * See the documentation for itx_wr_state_t above.
465 	 * Note that "bytes" accumulates the length of the transactions
466 	 * (i.e. data), not the actual log record sizes.
467 	 */
468 	kstat_named_t zil_itx_indirect_count;
469 	kstat_named_t zil_itx_indirect_bytes;
470 	kstat_named_t zil_itx_copied_count;
471 	kstat_named_t zil_itx_copied_bytes;
472 	kstat_named_t zil_itx_needcopy_count;
473 	kstat_named_t zil_itx_needcopy_bytes;
474 
475 	/*
476 	 * Transactions which have been allocated to the "normal"
477 	 * (i.e. not slog) storage pool. Note that "bytes" accumulate
478 	 * the actual log record sizes - which do not include the actual
479 	 * data in case of indirect writes.
480 	 */
481 	kstat_named_t zil_itx_metaslab_normal_count;
482 	kstat_named_t zil_itx_metaslab_normal_bytes;
483 
484 	/*
485 	 * Transactions which have been allocated to the "slog" storage pool.
486 	 * If there are no separate log devices, this is the same as the
487 	 * "normal" pool.
488 	 */
489 	kstat_named_t zil_itx_metaslab_slog_count;
490 	kstat_named_t zil_itx_metaslab_slog_bytes;
491 } zil_kstat_values_t;
492 
493 typedef struct zil_sums {
494 	wmsum_t zil_commit_count;
495 	wmsum_t zil_commit_writer_count;
496 	wmsum_t zil_itx_count;
497 	wmsum_t zil_itx_indirect_count;
498 	wmsum_t zil_itx_indirect_bytes;
499 	wmsum_t zil_itx_copied_count;
500 	wmsum_t zil_itx_copied_bytes;
501 	wmsum_t zil_itx_needcopy_count;
502 	wmsum_t zil_itx_needcopy_bytes;
503 	wmsum_t zil_itx_metaslab_normal_count;
504 	wmsum_t zil_itx_metaslab_normal_bytes;
505 	wmsum_t zil_itx_metaslab_slog_count;
506 	wmsum_t zil_itx_metaslab_slog_bytes;
507 } zil_sums_t;
508 
509 #define	ZIL_STAT_INCR(zil, stat, val) \
510 	do { \
511 		int64_t tmpval = (val); \
512 		wmsum_add(&(zil_sums_global.stat), tmpval); \
513 		if ((zil)->zl_sums) \
514 			wmsum_add(&((zil)->zl_sums->stat), tmpval); \
515 	} while (0)
516 
517 #define	ZIL_STAT_BUMP(zil, stat) \
518     ZIL_STAT_INCR(zil, stat, 1);
519 
520 typedef int zil_parse_blk_func_t(zilog_t *zilog, const blkptr_t *bp, void *arg,
521     uint64_t txg);
522 typedef int zil_parse_lr_func_t(zilog_t *zilog, const lr_t *lr, void *arg,
523     uint64_t txg);
524 typedef int zil_replay_func_t(void *arg1, void *arg2, boolean_t byteswap);
525 typedef int zil_get_data_t(void *arg, uint64_t arg2, lr_write_t *lr, char *dbuf,
526     struct lwb *lwb, zio_t *zio);
527 
528 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
529     zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg,
530     boolean_t decrypt);
531 
532 extern void	zil_init(void);
533 extern void	zil_fini(void);
534 
535 extern zilog_t	*zil_alloc(objset_t *os, zil_header_t *zh_phys);
536 extern void	zil_free(zilog_t *zilog);
537 
538 extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data,
539     zil_sums_t *zil_sums);
540 extern void	zil_close(zilog_t *zilog);
541 
542 extern boolean_t zil_replay(objset_t *os, void *arg,
543     zil_replay_func_t *const replay_func[TX_MAX_TYPE]);
544 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
545 extern boolean_t zil_destroy(zilog_t *zilog, boolean_t keep_first);
546 extern void	zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
547 
548 extern itx_t	*zil_itx_create(uint64_t txtype, size_t lrsize);
549 extern void	zil_itx_destroy(itx_t *itx);
550 extern void	zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
551 
552 extern void	zil_async_to_sync(zilog_t *zilog, uint64_t oid);
553 extern void	zil_commit(zilog_t *zilog, uint64_t oid);
554 extern void	zil_commit_impl(zilog_t *zilog, uint64_t oid);
555 extern void	zil_remove_async(zilog_t *zilog, uint64_t oid);
556 
557 extern int	zil_reset(const char *osname, void *txarg);
558 extern int	zil_claim(struct dsl_pool *dp,
559     struct dsl_dataset *ds, void *txarg);
560 extern int 	zil_check_log_chain(struct dsl_pool *dp,
561     struct dsl_dataset *ds, void *tx);
562 extern void	zil_sync(zilog_t *zilog, dmu_tx_t *tx);
563 extern void	zil_clean(zilog_t *zilog, uint64_t synced_txg);
564 
565 extern int	zil_suspend(const char *osname, void **cookiep);
566 extern void	zil_resume(void *cookie);
567 
568 extern void	zil_lwb_add_block(struct lwb *lwb, const blkptr_t *bp);
569 extern void	zil_lwb_add_txg(struct lwb *lwb, uint64_t txg);
570 extern int	zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
571 
572 extern void	zil_set_sync(zilog_t *zilog, uint64_t syncval);
573 
574 extern void	zil_set_logbias(zilog_t *zilog, uint64_t slogval);
575 
576 extern uint64_t	zil_max_copied_data(zilog_t *zilog);
577 extern uint64_t	zil_max_log_data(zilog_t *zilog);
578 
579 extern void zil_sums_init(zil_sums_t *zs);
580 extern void zil_sums_fini(zil_sums_t *zs);
581 extern void zil_kstat_values_update(zil_kstat_values_t *zs,
582     zil_sums_t *zil_sums);
583 
584 extern int zil_replay_disable;
585 
586 #ifdef	__cplusplus
587 }
588 #endif
589 
590 #endif	/* _SYS_ZIL_H */
591