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 http://www.opensolaris.org/os/licensing.
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, 2017 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Integros [integros.com]
25 */
26
27/* Portions Copyright 2010 Robert Milkowski */
28
29#ifndef	_SYS_ZIL_H
30#define	_SYS_ZIL_H
31
32#include <sys/types.h>
33#include <sys/spa.h>
34#include <sys/zio.h>
35#include <sys/dmu.h>
36#include <sys/zio_crypt.h>
37
38#ifdef	__cplusplus
39extern "C" {
40#endif
41
42struct dsl_pool;
43struct dsl_dataset;
44struct 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 */
62typedef 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 compatability.
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 */
91typedef 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
117typedef 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_MAX_TYPE		21	/* Max transaction type */
167
168/*
169 * The transactions for mkdir, symlink, remove, rmdir, link, and rename
170 * may have the following bit set, indicating the original request
171 * specified case-insensitive handling of names.
172 */
173#define	TX_CI	((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
174
175/*
176 * Transactions for write, truncate, setattr, acl_v0, and acl can be logged
177 * out of order.  For convenience in the code, all such records must have
178 * lr_foid at the same offset.
179 */
180#define	TX_OOO(txtype)			\
181	((txtype) == TX_WRITE ||	\
182	(txtype) == TX_TRUNCATE ||	\
183	(txtype) == TX_SETATTR ||	\
184	(txtype) == TX_ACL_V0 ||	\
185	(txtype) == TX_ACL ||		\
186	(txtype) == TX_WRITE2)
187
188/*
189 * The number of dnode slots consumed by the object is stored in the 8
190 * unused upper bits of the object ID. We subtract 1 from the value
191 * stored on disk for compatibility with implementations that don't
192 * support large dnodes. The slot count for a single-slot dnode will
193 * contain 0 for those bits to preserve the log record format for
194 * "small" dnodes.
195 */
196#define	LR_FOID_GET_SLOTS(oid) (BF64_GET((oid), 56, 8) + 1)
197#define	LR_FOID_SET_SLOTS(oid, x) BF64_SET((oid), 56, 8, (x) - 1)
198#define	LR_FOID_GET_OBJ(oid) BF64_GET((oid), 0, DN_MAX_OBJECT_SHIFT)
199#define	LR_FOID_SET_OBJ(oid, x) BF64_SET((oid), 0, DN_MAX_OBJECT_SHIFT, (x))
200
201/*
202 * Format of log records.
203 * The fields are carefully defined to allow them to be aligned
204 * and sized the same on sparc & intel architectures.
205 * Each log record has a common structure at the beginning.
206 *
207 * The log record on disk (lrc_seq) holds the sequence number of all log
208 * records which is used to ensure we don't replay the same record.
209 */
210typedef struct {			/* common log record header */
211	uint64_t	lrc_txtype;	/* intent log transaction type */
212	uint64_t	lrc_reclen;	/* transaction record length */
213	uint64_t	lrc_txg;	/* dmu transaction group number */
214	uint64_t	lrc_seq;	/* see comment above */
215} lr_t;
216
217/*
218 * Common start of all out-of-order record types (TX_OOO() above).
219 */
220typedef struct {
221	lr_t		lr_common;	/* common portion of log record */
222	uint64_t	lr_foid;	/* object id */
223} lr_ooo_t;
224
225/*
226 * Handle option extended vattr attributes.
227 *
228 * Whenever new attributes are added the version number
229 * will need to be updated as will code in
230 * zfs_log.c and zfs_replay.c
231 */
232typedef struct {
233	uint32_t	lr_attr_masksize; /* number of elements in array */
234	uint32_t	lr_attr_bitmap; /* First entry of array */
235	/* remainder of array and any additional fields */
236} lr_attr_t;
237
238/*
239 * log record for creates without optional ACL.
240 * This log record does support optional xvattr_t attributes.
241 */
242typedef struct {
243	lr_t		lr_common;	/* common portion of log record */
244	uint64_t	lr_doid;	/* object id of directory */
245	uint64_t	lr_foid;	/* object id of created file object */
246	uint64_t	lr_mode;	/* mode of object */
247	uint64_t	lr_uid;		/* uid of object */
248	uint64_t	lr_gid;		/* gid of object */
249	uint64_t	lr_gen;		/* generation (txg of creation) */
250	uint64_t	lr_crtime[2];	/* creation time */
251	uint64_t	lr_rdev;	/* rdev of object to create */
252	/* name of object to create follows this */
253	/* for symlinks, link content follows name */
254	/* for creates with xvattr data, the name follows the xvattr info */
255} lr_create_t;
256
257/*
258 * FUID ACL record will be an array of ACEs from the original ACL.
259 * If this array includes ephemeral IDs, the record will also include
260 * an array of log-specific FUIDs to replace the ephemeral IDs.
261 * Only one copy of each unique domain will be present, so the log-specific
262 * FUIDs will use an index into a compressed domain table.  On replay this
263 * information will be used to construct real FUIDs (and bypass idmap,
264 * since it may not be available).
265 */
266
267/*
268 * Log record for creates with optional ACL
269 * This log record is also used for recording any FUID
270 * information needed for replaying the create.  If the
271 * file doesn't have any actual ACEs then the lr_aclcnt
272 * would be zero.
273 *
274 * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
275 * If create is also setting xvattr's, then acl data follows xvattr.
276 * If ACE FUIDs are needed then they will follow the xvattr_t.  Following
277 * the FUIDs will be the domain table information.  The FUIDs for the owner
278 * and group will be in lr_create.  Name follows ACL data.
279 */
280typedef struct {
281	lr_create_t	lr_create;	/* common create portion */
282	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
283	uint64_t	lr_domcnt;	/* number of unique domains */
284	uint64_t	lr_fuidcnt;	/* number of real fuids */
285	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
286	uint64_t	lr_acl_flags;	/* ACL flags */
287} lr_acl_create_t;
288
289typedef struct {
290	lr_t		lr_common;	/* common portion of log record */
291	uint64_t	lr_doid;	/* obj id of directory */
292	/* name of object to remove follows this */
293} lr_remove_t;
294
295typedef struct {
296	lr_t		lr_common;	/* common portion of log record */
297	uint64_t	lr_doid;	/* obj id of directory */
298	uint64_t	lr_link_obj;	/* obj id of link */
299	/* name of object to link follows this */
300} lr_link_t;
301
302typedef struct {
303	lr_t		lr_common;	/* common portion of log record */
304	uint64_t	lr_sdoid;	/* obj id of source directory */
305	uint64_t	lr_tdoid;	/* obj id of target directory */
306	/* 2 strings: names of source and destination follow this */
307} lr_rename_t;
308
309typedef struct {
310	lr_t		lr_common;	/* common portion of log record */
311	uint64_t	lr_foid;	/* file object to write */
312	uint64_t	lr_offset;	/* offset to write to */
313	uint64_t	lr_length;	/* user data length to write */
314	uint64_t	lr_blkoff;	/* no longer used */
315	blkptr_t	lr_blkptr;	/* spa block pointer for replay */
316	/* write data will follow for small writes */
317} lr_write_t;
318
319typedef struct {
320	lr_t		lr_common;	/* common portion of log record */
321	uint64_t	lr_foid;	/* object id of file to truncate */
322	uint64_t	lr_offset;	/* offset to truncate from */
323	uint64_t	lr_length;	/* length to truncate */
324} lr_truncate_t;
325
326typedef struct {
327	lr_t		lr_common;	/* common portion of log record */
328	uint64_t	lr_foid;	/* file object to change attributes */
329	uint64_t	lr_mask;	/* mask of attributes to set */
330	uint64_t	lr_mode;	/* mode to set */
331	uint64_t	lr_uid;		/* uid to set */
332	uint64_t	lr_gid;		/* gid to set */
333	uint64_t	lr_size;	/* size to set */
334	uint64_t	lr_atime[2];	/* access time */
335	uint64_t	lr_mtime[2];	/* modification time */
336	/* optional attribute lr_attr_t may be here */
337} lr_setattr_t;
338
339typedef struct {
340	lr_t		lr_common;	/* common portion of log record */
341	uint64_t	lr_foid;	/* obj id of file */
342	uint64_t	lr_aclcnt;	/* number of acl entries */
343	/* lr_aclcnt number of ace_t entries follow this */
344} lr_acl_v0_t;
345
346typedef struct {
347	lr_t		lr_common;	/* common portion of log record */
348	uint64_t	lr_foid;	/* obj id of file */
349	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
350	uint64_t	lr_domcnt;	/* number of unique domains */
351	uint64_t	lr_fuidcnt;	/* number of real fuids */
352	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
353	uint64_t	lr_acl_flags;	/* ACL flags */
354	/* lr_acl_bytes number of variable sized ace's follows */
355} lr_acl_t;
356
357/*
358 * ZIL structure definitions, interface function prototype and globals.
359 */
360
361/*
362 * Writes are handled in three different ways:
363 *
364 * WR_INDIRECT:
365 *    In this mode, if we need to commit the write later, then the block
366 *    is immediately written into the file system (using dmu_sync),
367 *    and a pointer to the block is put into the log record.
368 *    When the txg commits the block is linked in.
369 *    This saves additionally writing the data into the log record.
370 *    There are a few requirements for this to occur:
371 *	- write is greater than zfs/zvol_immediate_write_sz
372 *	- not using slogs (as slogs are assumed to always be faster
373 *	  than writing into the main pool)
374 *	- the write occupies only one block
375 * WR_COPIED:
376 *    If we know we'll immediately be committing the
377 *    transaction (FSYNC or FDSYNC), the we allocate a larger
378 *    log record here for the data and copy the data in.
379 * WR_NEED_COPY:
380 *    Otherwise we don't allocate a buffer, and *if* we need to
381 *    flush the write later then a buffer is allocated and
382 *    we retrieve the data using the dmu.
383 */
384typedef enum {
385	WR_INDIRECT,	/* indirect - a large write (dmu_sync() data */
386			/* and put blkptr in log, rather than actual data) */
387	WR_COPIED,	/* immediate - data is copied into lr_write_t */
388	WR_NEED_COPY,	/* immediate - data needs to be copied if pushed */
389	WR_NUM_STATES	/* number of states */
390} itx_wr_state_t;
391
392typedef struct itx {
393	list_node_t	itx_node;	/* linkage on zl_itx_list */
394	void		*itx_private;	/* type-specific opaque data */
395	itx_wr_state_t	itx_wr_state;	/* write state */
396	uint8_t		itx_sync;	/* synchronous transaction */
397	uint64_t	itx_oid;	/* object id */
398	lr_t		itx_lr;		/* common part of log record */
399	/* followed by type-specific part of lr_xx_t and its immediate data */
400} itx_t;
401
402typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
403    uint64_t txg);
404typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
405    uint64_t txg);
406typedef int zil_replay_func_t(void *arg1, void *arg2, boolean_t byteswap);
407typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf,
408    struct lwb *lwb, zio_t *zio);
409
410extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
411    zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg,
412    boolean_t decrypt);
413
414extern void	zil_init(void);
415extern void	zil_fini(void);
416
417extern zilog_t	*zil_alloc(objset_t *os, zil_header_t *zh_phys);
418extern void	zil_free(zilog_t *zilog);
419
420extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data);
421extern void	zil_close(zilog_t *zilog);
422
423extern void	zil_replay(objset_t *os, void *arg,
424    zil_replay_func_t *replay_func[TX_MAX_TYPE]);
425extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
426extern void	zil_destroy(zilog_t *zilog, boolean_t keep_first);
427extern void	zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
428extern void	zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx);
429
430extern itx_t	*zil_itx_create(uint64_t txtype, size_t lrsize);
431extern void	zil_itx_destroy(itx_t *itx);
432extern void	zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
433
434extern void	zil_commit(zilog_t *zilog, uint64_t oid);
435extern void	zil_commit_impl(zilog_t *zilog, uint64_t oid);
436
437extern int	zil_reset(const char *osname, void *txarg);
438extern int	zil_claim(struct dsl_pool *dp,
439    struct dsl_dataset *ds, void *txarg);
440extern int	zil_check_log_chain(struct dsl_pool *dp,
441    struct dsl_dataset *ds, void *tx);
442extern void	zil_sync(zilog_t *zilog, dmu_tx_t *tx);
443extern void	zil_clean(zilog_t *zilog, uint64_t synced_txg);
444
445extern int	zil_suspend(const char *osname, void **cookiep);
446extern void	zil_resume(void *cookie);
447
448extern void	zil_lwb_add_block(struct lwb *lwb, const blkptr_t *bp);
449extern void	zil_lwb_add_txg(struct lwb *lwb, uint64_t txg);
450extern int	zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
451
452extern void	zil_set_sync(zilog_t *zilog, uint64_t syncval);
453
454extern void	zil_set_logbias(zilog_t *zilog, uint64_t slogval);
455
456extern int zil_replay_disable;
457
458#ifdef	__cplusplus
459}
460#endif
461
462#endif	/* _SYS_ZIL_H */
463