xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/zil.h (revision a6e57bd4c7a2bf9cc33be939d674d4c7d3e67cce)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #ifndef	_SYS_ZIL_H
27 #define	_SYS_ZIL_H
28 
29 #include <sys/types.h>
30 #include <sys/spa.h>
31 #include <sys/zio.h>
32 #include <sys/dmu.h>
33 
34 #ifdef	__cplusplus
35 extern "C" {
36 #endif
37 
38 /*
39  * Intent log format:
40  *
41  * Each objset has its own intent log.  The log header (zil_header_t)
42  * for objset N's intent log is kept in the Nth object of the SPA's
43  * intent_log objset.  The log header points to a chain of log blocks,
44  * each of which contains log records (i.e., transactions) followed by
45  * a log block trailer (zil_trailer_t).  The format of a log record
46  * depends on the record (or transaction) type, but all records begin
47  * with a common structure that defines the type, length, and txg.
48  */
49 
50 /*
51  * Intent log header - this on disk structure holds fields to manage
52  * the log.  All fields are 64 bit to easily handle cross architectures.
53  */
54 typedef struct zil_header {
55 	uint64_t zh_claim_txg;	/* txg in which log blocks were claimed */
56 	uint64_t zh_replay_seq;	/* highest replayed sequence number */
57 	blkptr_t zh_log;	/* log chain */
58 	uint64_t zh_claim_seq;	/* highest claimed sequence number */
59 	uint64_t zh_pad[5];
60 } zil_header_t;
61 
62 /*
63  * Log block trailer - structure at the end of the header and each log block
64  *
65  * The zit_bt contains a zbt_cksum which for the intent log is
66  * the sequence number of this log block. A seq of 0 is invalid.
67  * The zbt_cksum is checked by the SPA against the sequence
68  * number passed in the blk_cksum field of the blkptr_t
69  */
70 typedef struct zil_trailer {
71 	uint64_t zit_pad;
72 	blkptr_t zit_next_blk;	/* next block in chain */
73 	uint64_t zit_nused;	/* bytes in log block used */
74 	zio_block_tail_t zit_bt; /* block trailer */
75 } zil_trailer_t;
76 
77 #define	ZIL_MIN_BLKSZ	4096ULL
78 #define	ZIL_MAX_BLKSZ	SPA_MAXBLOCKSIZE
79 #define	ZIL_BLK_DATA_SZ(lwb)	((lwb)->lwb_sz - sizeof (zil_trailer_t))
80 
81 /*
82  * The words of a log block checksum.
83  */
84 #define	ZIL_ZC_GUID_0	0
85 #define	ZIL_ZC_GUID_1	1
86 #define	ZIL_ZC_OBJSET	2
87 #define	ZIL_ZC_SEQ	3
88 
89 typedef enum zil_create {
90 	Z_FILE,
91 	Z_DIR,
92 	Z_XATTRDIR,
93 } zil_create_t;
94 
95 /*
96  * size of xvattr log section.
97  * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
98  * for create time and a single 64 bit integer for all of the attributes,
99  * and 4 64 bit integers (32 bytes) for the scanstamp.
100  *
101  */
102 
103 #define	ZIL_XVAT_SIZE(mapsize) \
104 	sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
105 	(sizeof (uint64_t) * 7)
106 
107 /*
108  * Size of ACL in log.  The ACE data is padded out to properly align
109  * on 8 byte boundary.
110  */
111 
112 #define	ZIL_ACE_LENGTH(x)	(roundup(x, sizeof (uint64_t)))
113 
114 /*
115  * Intent log transaction types and record structures
116  */
117 #define	TX_CREATE		1	/* Create file */
118 #define	TX_MKDIR		2	/* Make directory */
119 #define	TX_MKXATTR		3	/* Make XATTR directory */
120 #define	TX_SYMLINK		4	/* Create symbolic link to a file */
121 #define	TX_REMOVE		5	/* Remove file */
122 #define	TX_RMDIR		6	/* Remove directory */
123 #define	TX_LINK			7	/* Create hard link to a file */
124 #define	TX_RENAME		8	/* Rename a file */
125 #define	TX_WRITE		9	/* File write */
126 #define	TX_TRUNCATE		10	/* Truncate a file */
127 #define	TX_SETATTR		11	/* Set file attributes */
128 #define	TX_ACL_V0		12	/* Set old formatted ACL */
129 #define	TX_ACL			13	/* Set ACL */
130 #define	TX_CREATE_ACL		14	/* create with ACL */
131 #define	TX_CREATE_ATTR		15	/* create + attrs */
132 #define	TX_CREATE_ACL_ATTR 	16	/* create with ACL + attrs */
133 #define	TX_MKDIR_ACL		17	/* mkdir with ACL */
134 #define	TX_MKDIR_ATTR		18	/* mkdir with attr */
135 #define	TX_MKDIR_ACL_ATTR	19	/* mkdir with ACL + attrs */
136 #define	TX_MAX_TYPE		20	/* Max transaction type */
137 
138 /*
139  * The transactions for mkdir, symlink, remove, rmdir, link, and rename
140  * may have the following bit set, indicating the original request
141  * specified case-insensitive handling of names.
142  */
143 #define	TX_CI	((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
144 
145 /*
146  * Format of log records.
147  * The fields are carefully defined to allow them to be aligned
148  * and sized the same on sparc & intel architectures.
149  * Each log record has a common structure at the beginning.
150  *
151  * Note, lrc_seq holds two different sequence numbers. Whilst in memory
152  * it contains the transaction sequence number.  The log record on
153  * disk holds the sequence number of all log records which is used to
154  * ensure we don't replay the same record.  The two sequence numbers are
155  * different because the transactions can now be pushed out of order.
156  */
157 typedef struct {			/* common log record header */
158 	uint64_t	lrc_txtype;	/* intent log transaction type */
159 	uint64_t	lrc_reclen;	/* transaction record length */
160 	uint64_t	lrc_txg;	/* dmu transaction group number */
161 	uint64_t	lrc_seq;	/* see comment above */
162 } lr_t;
163 
164 /*
165  * Handle option extended vattr attributes.
166  *
167  * Whenever new attributes are added the version number
168  * will need to be updated as will code in
169  * zfs_log.c and zfs_replay.c
170  */
171 typedef struct {
172 	uint32_t	lr_attr_masksize; /* number of elements in array */
173 	uint32_t	lr_attr_bitmap; /* First entry of array */
174 	/* remainder of array and any additional fields */
175 } lr_attr_t;
176 
177 /*
178  * log record for creates without optional ACL.
179  * This log record does support optional xvattr_t attributes.
180  */
181 typedef struct {
182 	lr_t		lr_common;	/* common portion of log record */
183 	uint64_t	lr_doid;	/* object id of directory */
184 	uint64_t	lr_foid;	/* object id of created file object */
185 	uint64_t	lr_mode;	/* mode of object */
186 	uint64_t	lr_uid;		/* uid of object */
187 	uint64_t	lr_gid;		/* gid of object */
188 	uint64_t	lr_gen;		/* generation (txg of creation) */
189 	uint64_t	lr_crtime[2];	/* creation time */
190 	uint64_t	lr_rdev;	/* rdev of object to create */
191 	/* name of object to create follows this */
192 	/* for symlinks, link content follows name */
193 	/* for creates with xvattr data, the name follows the xvattr info */
194 } lr_create_t;
195 
196 /*
197  * FUID ACL record will be an array of ACEs from the original ACL.
198  * If this array includes ephemeral IDs, the record will also include
199  * an array of log-specific FUIDs to replace the ephemeral IDs.
200  * Only one copy of each unique domain will be present, so the log-specific
201  * FUIDs will use an index into a compressed domain table.  On replay this
202  * information will be used to construct real FUIDs (and bypass idmap,
203  * since it may not be available).
204  */
205 
206 /*
207  * Log record for creates with optional ACL
208  * This log record is also used for recording any FUID
209  * information needed for replaying the create.  If the
210  * file doesn't have any actual ACEs then the lr_aclcnt
211  * would be zero.
212  */
213 typedef struct {
214 	lr_create_t	lr_create;	/* common create portion */
215 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
216 	uint64_t	lr_domcnt;	/* number of unique domains */
217 	uint64_t	lr_fuidcnt;	/* number of real fuids */
218 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
219 	uint64_t	lr_acl_flags;	/* ACL flags */
220 	/* lr_acl_bytes number of variable sized ace's follows */
221 	/* if create is also setting xvattr's, then acl data follows xvattr */
222 	/* if ACE FUIDs are needed then they will follow the xvattr_t */
223 	/* Following the FUIDs will be the domain table information. */
224 	/* The FUIDs for the owner and group will be in the lr_create */
225 	/* portion of the record. */
226 	/* name follows ACL data */
227 } lr_acl_create_t;
228 
229 typedef struct {
230 	lr_t		lr_common;	/* common portion of log record */
231 	uint64_t	lr_doid;	/* obj id of directory */
232 	/* name of object to remove follows this */
233 } lr_remove_t;
234 
235 typedef struct {
236 	lr_t		lr_common;	/* common portion of log record */
237 	uint64_t	lr_doid;	/* obj id of directory */
238 	uint64_t	lr_link_obj;	/* obj id of link */
239 	/* name of object to link follows this */
240 } lr_link_t;
241 
242 typedef struct {
243 	lr_t		lr_common;	/* common portion of log record */
244 	uint64_t	lr_sdoid;	/* obj id of source directory */
245 	uint64_t	lr_tdoid;	/* obj id of target directory */
246 	/* 2 strings: names of source and destination follow this */
247 } lr_rename_t;
248 
249 typedef struct {
250 	lr_t		lr_common;	/* common portion of log record */
251 	uint64_t	lr_foid;	/* file object to write */
252 	uint64_t	lr_offset;	/* offset to write to */
253 	uint64_t	lr_length;	/* user data length to write */
254 	uint64_t	lr_blkoff;	/* offset represented by lr_blkptr */
255 	blkptr_t	lr_blkptr;	/* spa block pointer for replay */
256 	/* write data will follow for small writes */
257 } lr_write_t;
258 
259 typedef struct {
260 	lr_t		lr_common;	/* common portion of log record */
261 	uint64_t	lr_foid;	/* object id of file to truncate */
262 	uint64_t	lr_offset;	/* offset to truncate from */
263 	uint64_t	lr_length;	/* length to truncate */
264 } lr_truncate_t;
265 
266 typedef struct {
267 	lr_t		lr_common;	/* common portion of log record */
268 	uint64_t	lr_foid;	/* file object to change attributes */
269 	uint64_t	lr_mask;	/* mask of attributes to set */
270 	uint64_t	lr_mode;	/* mode to set */
271 	uint64_t	lr_uid;		/* uid to set */
272 	uint64_t	lr_gid;		/* gid to set */
273 	uint64_t	lr_size;	/* size to set */
274 	uint64_t	lr_atime[2];	/* access time */
275 	uint64_t	lr_mtime[2];	/* modification time */
276 	/* optional attribute lr_attr_t may be here */
277 } lr_setattr_t;
278 
279 typedef struct {
280 	lr_t		lr_common;	/* common portion of log record */
281 	uint64_t	lr_foid;	/* obj id of file */
282 	uint64_t	lr_aclcnt;	/* number of acl entries */
283 	/* lr_aclcnt number of ace_t entries follow this */
284 } lr_acl_v0_t;
285 
286 typedef struct {
287 	lr_t		lr_common;	/* common portion of log record */
288 	uint64_t	lr_foid;	/* obj id of file */
289 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
290 	uint64_t	lr_domcnt;	/* number of unique domains */
291 	uint64_t	lr_fuidcnt;	/* number of real fuids */
292 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
293 	uint64_t	lr_acl_flags;	/* ACL flags */
294 	/* lr_acl_bytes number of variable sized ace's follows */
295 } lr_acl_t;
296 
297 /*
298  * ZIL structure definitions, interface function prototype and globals.
299  */
300 
301 /*
302  * ZFS intent log transaction structure
303  */
304 typedef enum {
305 	WR_INDIRECT,	/* indirect - a large write (dmu_sync() data */
306 			/* and put blkptr in log, rather than actual data) */
307 	WR_COPIED,	/* immediate - data is copied into lr_write_t */
308 	WR_NEED_COPY,	/* immediate - data needs to be copied if pushed */
309 } itx_wr_state_t;
310 
311 typedef struct itx {
312 	list_node_t	itx_node;	/* linkage on zl_itx_list */
313 	void		*itx_private;	/* type-specific opaque data */
314 	itx_wr_state_t	itx_wr_state;	/* write state */
315 	uint8_t		itx_sync;	/* synchronous transaction */
316 	uint64_t	itx_sod;	/* record size on disk */
317 	lr_t		itx_lr;		/* common part of log record */
318 	/* followed by type-specific part of lr_xx_t and its immediate data */
319 } itx_t;
320 
321 
322 /*
323  * zgd_t is passed through dmu_sync() to the callback routine zfs_get_done()
324  * to handle the cleanup of the dmu_sync() buffer write
325  */
326 typedef struct {
327 	zilog_t		*zgd_zilog;	/* zilog */
328 	blkptr_t	*zgd_bp;	/* block pointer */
329 	struct rl	*zgd_rl;	/* range lock */
330 } zgd_t;
331 
332 
333 typedef void zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
334     uint64_t txg);
335 typedef void zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
336     uint64_t txg);
337 typedef int zil_replay_func_t();
338 typedef void zil_replay_cleaner_t();
339 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio);
340 
341 extern uint64_t	zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
342     zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg);
343 
344 extern void	zil_init(void);
345 extern void	zil_fini(void);
346 
347 extern zilog_t	*zil_alloc(objset_t *os, zil_header_t *zh_phys);
348 extern void	zil_free(zilog_t *zilog);
349 
350 extern zilog_t	*zil_open(objset_t *os, zil_get_data_t *get_data);
351 extern void	zil_close(zilog_t *zilog);
352 
353 extern void	zil_replay(objset_t *os, void *arg, uint64_t *txgp,
354     zil_replay_func_t *replay_func[TX_MAX_TYPE],
355     zil_replay_cleaner_t *replay_cleaner);
356 extern void	zil_destroy(zilog_t *zilog, boolean_t keep_first);
357 extern void	zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx);
358 
359 extern itx_t	*zil_itx_create(uint64_t txtype, size_t lrsize);
360 extern uint64_t zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
361 
362 extern void	zil_commit(zilog_t *zilog, uint64_t seq, uint64_t oid);
363 
364 extern int	zil_claim(char *osname, void *txarg);
365 extern int	zil_check_log_chain(char *osname, void *txarg);
366 extern int	zil_clear_log_chain(char *osname, void *txarg);
367 extern void	zil_sync(zilog_t *zilog, dmu_tx_t *tx);
368 extern void	zil_clean(zilog_t *zilog);
369 extern int	zil_is_committed(zilog_t *zilog);
370 
371 extern int	zil_suspend(zilog_t *zilog);
372 extern void	zil_resume(zilog_t *zilog);
373 
374 extern void	zil_add_block(zilog_t *zilog, blkptr_t *bp);
375 
376 extern int zil_disable;
377 
378 #ifdef	__cplusplus
379 }
380 #endif
381 
382 #endif	/* _SYS_ZIL_H */
383