xref: /freebsd/sys/contrib/openzfs/include/sys/zil.h (revision 7a7741af18d6c8a804cc643cb7ecda9d730c6aa6)
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 	uint8_t		lr_attr_data[];
252 } lr_attr_t;
253 
254 /*
255  * log record for creates without optional ACL.
256  * This log record does support optional xvattr_t attributes.
257  */
258 typedef struct {
259 	lr_t		lr_common;	/* common portion of log record */
260 	uint64_t	lr_doid;	/* object id of directory */
261 	uint64_t	lr_foid;	/* object id of created file object */
262 	uint64_t	lr_mode;	/* mode of object */
263 	uint64_t	lr_uid;		/* uid of object */
264 	uint64_t	lr_gid;		/* gid of object */
265 	uint64_t	lr_gen;		/* generation (txg of creation) */
266 	uint64_t	lr_crtime[2];	/* creation time */
267 	uint64_t	lr_rdev;	/* rdev of object to create */
268 } _lr_create_t;
269 
270 typedef struct {
271 	_lr_create_t	lr_create;	/* common create portion */
272 	/* name of object to create follows this */
273 	/* for symlinks, link content follows name */
274 	/* for creates with xvattr data, the name follows the xvattr info */
275 	uint8_t		lr_data[];
276 } lr_create_t;
277 
278 /*
279  * FUID ACL record will be an array of ACEs from the original ACL.
280  * If this array includes ephemeral IDs, the record will also include
281  * an array of log-specific FUIDs to replace the ephemeral IDs.
282  * Only one copy of each unique domain will be present, so the log-specific
283  * FUIDs will use an index into a compressed domain table.  On replay this
284  * information will be used to construct real FUIDs (and bypass idmap,
285  * since it may not be available).
286  */
287 
288 /*
289  * Log record for creates with optional ACL
290  * This log record is also used for recording any FUID
291  * information needed for replaying the create.  If the
292  * file doesn't have any actual ACEs then the lr_aclcnt
293  * would be zero.
294  *
295  * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
296  * If create is also setting xvattr's, then acl data follows xvattr.
297  * If ACE FUIDs are needed then they will follow the xvattr_t.  Following
298  * the FUIDs will be the domain table information.  The FUIDs for the owner
299  * and group will be in lr_create.  Name follows ACL data.
300  */
301 typedef struct {
302 	_lr_create_t	lr_create;	/* common create portion */
303 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
304 	uint64_t	lr_domcnt;	/* number of unique domains */
305 	uint64_t	lr_fuidcnt;	/* number of real fuids */
306 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
307 	uint64_t	lr_acl_flags;	/* ACL flags */
308 	uint8_t		lr_data[];
309 } lr_acl_create_t;
310 
311 typedef struct {
312 	lr_t		lr_common;	/* common portion of log record */
313 	uint64_t	lr_doid;	/* obj id of directory */
314 	/* name of object to remove follows this */
315 	uint8_t		lr_data[];
316 } lr_remove_t;
317 
318 typedef struct {
319 	lr_t		lr_common;	/* common portion of log record */
320 	uint64_t	lr_doid;	/* obj id of directory */
321 	uint64_t	lr_link_obj;	/* obj id of link */
322 	/* name of object to link follows this */
323 	uint8_t		lr_data[];
324 } lr_link_t;
325 
326 typedef struct {
327 	lr_t		lr_common;	/* common portion of log record */
328 	uint64_t	lr_sdoid;	/* obj id of source directory */
329 	uint64_t	lr_tdoid;	/* obj id of target directory */
330 } _lr_rename_t;
331 
332 typedef struct {
333 	_lr_rename_t	lr_rename;	/* common rename portion */
334 	/* 2 strings: names of source and destination follow this */
335 	uint8_t		lr_data[];
336 } lr_rename_t;
337 
338 typedef struct {
339 	_lr_rename_t	lr_rename;	/* common rename portion */
340 	/* members related to the whiteout file (based on _lr_create_t) */
341 	uint64_t	lr_wfoid;	/* obj id of the new whiteout file */
342 	uint64_t	lr_wmode;	/* mode of object */
343 	uint64_t	lr_wuid;	/* uid of whiteout */
344 	uint64_t	lr_wgid;	/* gid of whiteout */
345 	uint64_t	lr_wgen;	/* generation (txg of creation) */
346 	uint64_t	lr_wcrtime[2];	/* creation time */
347 	uint64_t	lr_wrdev;	/* always makedev(0, 0) */
348 	/* 2 strings: names of source and destination follow this */
349 	uint8_t		lr_data[];
350 } lr_rename_whiteout_t;
351 
352 typedef struct {
353 	lr_t		lr_common;	/* common portion of log record */
354 	uint64_t	lr_foid;	/* file object to write */
355 	uint64_t	lr_offset;	/* offset to write to */
356 	uint64_t	lr_length;	/* user data length to write */
357 	uint64_t	lr_blkoff;	/* no longer used */
358 	blkptr_t	lr_blkptr;	/* spa block pointer for replay */
359 	/* write data will follow for small writes */
360 	uint8_t		lr_data[];
361 } lr_write_t;
362 
363 typedef struct {
364 	lr_t		lr_common;	/* common portion of log record */
365 	uint64_t	lr_foid;	/* object id of file to truncate */
366 	uint64_t	lr_offset;	/* offset to truncate from */
367 	uint64_t	lr_length;	/* length to truncate */
368 } lr_truncate_t;
369 
370 typedef struct {
371 	lr_t		lr_common;	/* common portion of log record */
372 	uint64_t	lr_foid;	/* file object to change attributes */
373 	uint64_t	lr_mask;	/* mask of attributes to set */
374 	uint64_t	lr_mode;	/* mode to set */
375 	uint64_t	lr_uid;		/* uid to set */
376 	uint64_t	lr_gid;		/* gid to set */
377 	uint64_t	lr_size;	/* size to set */
378 	uint64_t	lr_atime[2];	/* access time */
379 	uint64_t	lr_mtime[2];	/* modification time */
380 	/* optional attribute lr_attr_t may be here */
381 	uint8_t		lr_data[];
382 } lr_setattr_t;
383 
384 typedef struct {
385 	lr_t		lr_common;	/* common portion of log record */
386 	uint64_t	lr_foid;	/* file object to change attributes */
387 	uint64_t	lr_size;
388 	/* xattr name and value follows */
389 	uint8_t		lr_data[];
390 } lr_setsaxattr_t;
391 
392 typedef struct {
393 	lr_t		lr_common;	/* common portion of log record */
394 	uint64_t	lr_foid;	/* obj id of file */
395 	uint64_t	lr_aclcnt;	/* number of acl entries */
396 	/* lr_aclcnt number of ace_t entries follow this */
397 	uint8_t		lr_data[];
398 } lr_acl_v0_t;
399 
400 typedef struct {
401 	lr_t		lr_common;	/* common portion of log record */
402 	uint64_t	lr_foid;	/* obj id of file */
403 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
404 	uint64_t	lr_domcnt;	/* number of unique domains */
405 	uint64_t	lr_fuidcnt;	/* number of real fuids */
406 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
407 	uint64_t	lr_acl_flags;	/* ACL flags */
408 	/* lr_acl_bytes number of variable sized ace's follows */
409 	uint8_t		lr_data[];
410 } lr_acl_t;
411 
412 typedef struct {
413 	lr_t		lr_common;	/* common portion of log record */
414 	uint64_t	lr_foid;	/* file object to clone into */
415 	uint64_t	lr_offset;	/* offset to clone to */
416 	uint64_t	lr_length;	/* length of the blocks to clone */
417 	uint64_t	lr_blksz;	/* file's block size */
418 	uint64_t	lr_nbps;	/* number of block pointers */
419 	/* block pointers of the blocks to clone follows */
420 	blkptr_t	lr_bps[];
421 } lr_clone_range_t;
422 
423 /*
424  * ZIL structure definitions, interface function prototype and globals.
425  */
426 
427 /*
428  * Writes are handled in three different ways:
429  *
430  * WR_INDIRECT:
431  *    In this mode, if we need to commit the write later, then the block
432  *    is immediately written into the file system (using dmu_sync),
433  *    and a pointer to the block is put into the log record.
434  *    When the txg commits the block is linked in.
435  *    This saves additionally writing the data into the log record.
436  *    There are a few requirements for this to occur:
437  *	- write is greater than zfs/zvol_immediate_write_sz
438  *	- not using slogs (as slogs are assumed to always be faster
439  *	  than writing into the main pool)
440  *	- the write occupies only one block
441  * WR_COPIED:
442  *    If we know we'll immediately be committing the
443  *    transaction (O_SYNC or O_DSYNC), then we allocate a larger
444  *    log record here for the data and copy the data in.
445  * WR_NEED_COPY:
446  *    Otherwise we don't allocate a buffer, and *if* we need to
447  *    flush the write later then a buffer is allocated and
448  *    we retrieve the data using the dmu.
449  */
450 typedef enum {
451 	WR_INDIRECT,	/* indirect - a large write (dmu_sync() data */
452 			/* and put blkptr in log, rather than actual data) */
453 	WR_COPIED,	/* immediate - data is copied into lr_write_t */
454 	WR_NEED_COPY,	/* immediate - data needs to be copied if pushed */
455 	WR_NUM_STATES	/* number of states */
456 } itx_wr_state_t;
457 
458 typedef void (*zil_callback_t)(void *data);
459 
460 typedef struct itx {
461 	list_node_t	itx_node;	/* linkage on zl_itx_list */
462 	void		*itx_private;	/* type-specific opaque data */
463 	itx_wr_state_t	itx_wr_state;	/* write state */
464 	uint8_t		itx_sync;	/* synchronous transaction */
465 	zil_callback_t	itx_callback;   /* Called when the itx is persistent */
466 	void		*itx_callback_data; /* User data for the callback */
467 	size_t		itx_size;	/* allocated itx structure size */
468 	uint64_t	itx_oid;	/* object id */
469 	uint64_t	itx_gen;	/* gen number for zfs_get_data */
470 	lr_t		itx_lr;		/* common part of log record */
471 	uint8_t		itx_lr_data[];	/* type-specific part of lr_xx_t */
472 } itx_t;
473 
474 /*
475  * Used for zil kstat.
476  */
477 typedef struct zil_stats {
478 	/*
479 	 * Number of times a ZIL commit (e.g. fsync) has been requested.
480 	 */
481 	kstat_named_t zil_commit_count;
482 
483 	/*
484 	 * Number of times the ZIL has been flushed to stable storage.
485 	 * This is less than zil_commit_count when commits are "merged"
486 	 * (see the documentation above zil_commit()).
487 	 */
488 	kstat_named_t zil_commit_writer_count;
489 
490 	/*
491 	 * Number of times a ZIL commit failed and the ZIL was forced to fall
492 	 * back to txg_wait_synced(). The separate counts are for different
493 	 * reasons:
494 	 * -   error: ZIL IO (write/flush) returned an error
495 	 *            (see zil_commit_impl())
496 	 * -   stall: LWB block allocation failed, ZIL chain abandoned
497 	 *            (see zil_commit_writer_stall())
498 	 * - suspend: ZIL suspended
499 	 *            (see zil_commit(), zil_get_commit_list())
500 	 */
501 	kstat_named_t zil_commit_error_count;
502 	kstat_named_t zil_commit_stall_count;
503 	kstat_named_t zil_commit_suspend_count;
504 
505 	/*
506 	 * Number of transactions (reads, writes, renames, etc.)
507 	 * that have been committed.
508 	 */
509 	kstat_named_t zil_itx_count;
510 
511 	/*
512 	 * See the documentation for itx_wr_state_t above.
513 	 * Note that "bytes" accumulates the length of the transactions
514 	 * (i.e. data), not the actual log record sizes.
515 	 */
516 	kstat_named_t zil_itx_indirect_count;
517 	kstat_named_t zil_itx_indirect_bytes;
518 	kstat_named_t zil_itx_copied_count;
519 	kstat_named_t zil_itx_copied_bytes;
520 	kstat_named_t zil_itx_needcopy_count;
521 	kstat_named_t zil_itx_needcopy_bytes;
522 
523 	/*
524 	 * Transactions which have been allocated to the "normal"
525 	 * (i.e. not slog) storage pool. Note that "bytes" accumulate
526 	 * the actual log record sizes - which do not include the actual
527 	 * data in case of indirect writes.  bytes <= write <= alloc.
528 	 */
529 	kstat_named_t zil_itx_metaslab_normal_count;
530 	kstat_named_t zil_itx_metaslab_normal_bytes;
531 	kstat_named_t zil_itx_metaslab_normal_write;
532 	kstat_named_t zil_itx_metaslab_normal_alloc;
533 
534 	/*
535 	 * Transactions which have been allocated to the "slog" storage pool.
536 	 * If there are no separate log devices, this is the same as the
537 	 * "normal" pool.  bytes <= write <= alloc.
538 	 */
539 	kstat_named_t zil_itx_metaslab_slog_count;
540 	kstat_named_t zil_itx_metaslab_slog_bytes;
541 	kstat_named_t zil_itx_metaslab_slog_write;
542 	kstat_named_t zil_itx_metaslab_slog_alloc;
543 } zil_kstat_values_t;
544 
545 typedef struct zil_sums {
546 	wmsum_t zil_commit_count;
547 	wmsum_t zil_commit_writer_count;
548 	wmsum_t zil_commit_error_count;
549 	wmsum_t zil_commit_stall_count;
550 	wmsum_t zil_commit_suspend_count;
551 	wmsum_t zil_itx_count;
552 	wmsum_t zil_itx_indirect_count;
553 	wmsum_t zil_itx_indirect_bytes;
554 	wmsum_t zil_itx_copied_count;
555 	wmsum_t zil_itx_copied_bytes;
556 	wmsum_t zil_itx_needcopy_count;
557 	wmsum_t zil_itx_needcopy_bytes;
558 	wmsum_t zil_itx_metaslab_normal_count;
559 	wmsum_t zil_itx_metaslab_normal_bytes;
560 	wmsum_t zil_itx_metaslab_normal_write;
561 	wmsum_t zil_itx_metaslab_normal_alloc;
562 	wmsum_t zil_itx_metaslab_slog_count;
563 	wmsum_t zil_itx_metaslab_slog_bytes;
564 	wmsum_t zil_itx_metaslab_slog_write;
565 	wmsum_t zil_itx_metaslab_slog_alloc;
566 } zil_sums_t;
567 
568 #define	ZIL_STAT_INCR(zil, stat, val) \
569 	do { \
570 		int64_t tmpval = (val); \
571 		wmsum_add(&(zil_sums_global.stat), tmpval); \
572 		if ((zil)->zl_sums) \
573 			wmsum_add(&((zil)->zl_sums->stat), tmpval); \
574 	} while (0)
575 
576 #define	ZIL_STAT_BUMP(zil, stat) \
577     ZIL_STAT_INCR(zil, stat, 1);
578 
579 typedef int zil_parse_blk_func_t(zilog_t *zilog, const blkptr_t *bp, void *arg,
580     uint64_t txg);
581 typedef int zil_parse_lr_func_t(zilog_t *zilog, const lr_t *lr, void *arg,
582     uint64_t txg);
583 typedef int zil_replay_func_t(void *arg1, void *arg2, boolean_t byteswap);
584 typedef int zil_get_data_t(void *arg, uint64_t arg2, lr_write_t *lr, char *dbuf,
585     struct lwb *lwb, zio_t *zio);
586 
587 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
588     zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg,
589     boolean_t decrypt);
590 
591 extern void	zil_init(void);
592 extern void	zil_fini(void);
593 
594 extern zilog_t	*zil_alloc(objset_t *os, zil_header_t *zh_phys);
595 extern void	zil_free(zilog_t *zilog);
596 
597 extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data,
598     zil_sums_t *zil_sums);
599 extern void	zil_close(zilog_t *zilog);
600 
601 extern boolean_t zil_replay(objset_t *os, void *arg,
602     zil_replay_func_t *const replay_func[TX_MAX_TYPE]);
603 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
604 extern boolean_t zil_destroy(zilog_t *zilog, boolean_t keep_first);
605 extern void	zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
606 
607 extern itx_t	*zil_itx_create(uint64_t txtype, size_t lrsize);
608 extern void	zil_itx_destroy(itx_t *itx);
609 extern void	zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
610 
611 extern void	zil_async_to_sync(zilog_t *zilog, uint64_t oid);
612 extern void	zil_commit(zilog_t *zilog, uint64_t oid);
613 extern void	zil_commit_impl(zilog_t *zilog, uint64_t oid);
614 extern void	zil_remove_async(zilog_t *zilog, uint64_t oid);
615 
616 extern int	zil_reset(const char *osname, void *txarg);
617 extern int	zil_claim(struct dsl_pool *dp,
618     struct dsl_dataset *ds, void *txarg);
619 extern int 	zil_check_log_chain(struct dsl_pool *dp,
620     struct dsl_dataset *ds, void *tx);
621 extern void	zil_sync(zilog_t *zilog, dmu_tx_t *tx);
622 extern void	zil_clean(zilog_t *zilog, uint64_t synced_txg);
623 
624 extern int	zil_suspend(const char *osname, void **cookiep);
625 extern void	zil_resume(void *cookie);
626 
627 extern void	zil_lwb_add_block(struct lwb *lwb, const blkptr_t *bp);
628 extern void	zil_lwb_add_txg(struct lwb *lwb, uint64_t txg);
629 extern int	zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
630 
631 extern void	zil_set_sync(zilog_t *zilog, uint64_t syncval);
632 
633 extern void	zil_set_logbias(zilog_t *zilog, uint64_t slogval);
634 
635 extern uint64_t	zil_max_copied_data(zilog_t *zilog);
636 extern uint64_t	zil_max_log_data(zilog_t *zilog, size_t hdrsize);
637 
638 extern void zil_sums_init(zil_sums_t *zs);
639 extern void zil_sums_fini(zil_sums_t *zs);
640 extern void zil_kstat_values_update(zil_kstat_values_t *zs,
641     zil_sums_t *zil_sums);
642 
643 extern int zil_replay_disable;
644 
645 #ifdef	__cplusplus
646 }
647 #endif
648 
649 #endif	/* _SYS_ZIL_H */
650