xref: /titanic_50/usr/src/uts/common/fs/zfs/sys/zil.h (revision 5e4df02aae6a2053aeb96be5e3ca1cad2de4ee66)
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 2007 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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
30 
31 #include <sys/types.h>
32 #include <sys/spa.h>
33 #include <sys/zio.h>
34 #include <sys/dmu.h>
35 
36 #ifdef	__cplusplus
37 extern "C" {
38 #endif
39 
40 /*
41  * Intent log format:
42  *
43  * Each objset has its own intent log.  The log header (zil_header_t)
44  * for objset N's intent log is kept in the Nth object of the SPA's
45  * intent_log objset.  The log header points to a chain of log blocks,
46  * each of which contains log records (i.e., transactions) followed by
47  * a log block trailer (zil_trailer_t).  The format of a log record
48  * depends on the record (or transaction) type, but all records begin
49  * with a common structure that defines the type, length, and txg.
50  */
51 
52 /*
53  * Intent log header - this on disk structure holds fields to manage
54  * the log.  All fields are 64 bit to easily handle cross architectures.
55  */
56 typedef struct zil_header {
57 	uint64_t zh_claim_txg;	/* txg in which log blocks were claimed */
58 	uint64_t zh_replay_seq;	/* highest replayed sequence number */
59 	blkptr_t zh_log;	/* log chain */
60 	uint64_t zh_claim_seq;	/* highest claimed sequence number */
61 	uint64_t zh_pad[5];
62 } zil_header_t;
63 
64 /*
65  * Log block trailer - structure at the end of the header and each log block
66  *
67  * The zit_bt contains a zbt_cksum which for the intent log is
68  * the sequence number of this log block. A seq of 0 is invalid.
69  * The zbt_cksum is checked by the SPA against the sequence
70  * number passed in the blk_cksum field of the blkptr_t
71  */
72 typedef struct zil_trailer {
73 	uint64_t zit_pad;
74 	blkptr_t zit_next_blk;	/* next block in chain */
75 	uint64_t zit_nused;	/* bytes in log block used */
76 	zio_block_tail_t zit_bt; /* block trailer */
77 } zil_trailer_t;
78 
79 #define	ZIL_MIN_BLKSZ	4096ULL
80 #define	ZIL_MAX_BLKSZ	SPA_MAXBLOCKSIZE
81 #define	ZIL_BLK_DATA_SZ(lwb)	((lwb)->lwb_sz - sizeof (zil_trailer_t))
82 
83 /*
84  * The words of a log block checksum.
85  */
86 #define	ZIL_ZC_GUID_0	0
87 #define	ZIL_ZC_GUID_1	1
88 #define	ZIL_ZC_OBJSET	2
89 #define	ZIL_ZC_SEQ	3
90 
91 typedef enum zil_create {
92 	Z_FILE,
93 	Z_DIR,
94 	Z_XATTRDIR,
95 } zil_create_t;
96 
97 /*
98  * size of xvattr log section.
99  * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
100  * for create time and a single 64 bit integer for all of the attributes,
101  * and 4 64 bit integers (32 bytes) for the scanstamp.
102  *
103  */
104 
105 #define	ZIL_XVAT_SIZE(mapsize) \
106 	sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
107 	(sizeof (uint64_t) * 7)
108 
109 /*
110  * Size of ACL in log.  The ACE data is padded out to properly align
111  * on 8 byte boundary.
112  */
113 
114 #define	ZIL_ACE_LENGTH(x)	(roundup(x, sizeof (uint64_t)))
115 
116 /*
117  * Intent log transaction types and record structures
118  */
119 #define	TX_CREATE		1	/* Create file */
120 #define	TX_MKDIR		2	/* Make directory */
121 #define	TX_MKXATTR		3	/* Make XATTR directory */
122 #define	TX_SYMLINK		4	/* Create symbolic link to a file */
123 #define	TX_REMOVE		5	/* Remove file */
124 #define	TX_RMDIR		6	/* Remove directory */
125 #define	TX_LINK			7	/* Create hard link to a file */
126 #define	TX_RENAME		8	/* Rename a file */
127 #define	TX_WRITE		9	/* File write */
128 #define	TX_TRUNCATE		10	/* Truncate a file */
129 #define	TX_SETATTR		11	/* Set file attributes */
130 #define	TX_ACL_V0		12	/* Set old formatted ACL */
131 #define	TX_ACL			13	/* Set ACL */
132 #define	TX_CREATE_ACL		14	/* create with ACL */
133 #define	TX_CREATE_ATTR		15	/* create + attrs */
134 #define	TX_CREATE_ACL_ATTR 	16	/* create with ACL + attrs */
135 #define	TX_MKDIR_ACL		17	/* mkdir with ACL */
136 #define	TX_MKDIR_ATTR		18	/* mkdir with attr */
137 #define	TX_MKDIR_ACL_ATTR	19	/* mkdir with ACL + attrs */
138 #define	TX_MAX_TYPE		20	/* Max transaction type */
139 
140 /*
141  * The transactions for mkdir, symlink, remove, rmdir, link, and rename
142  * may have the following bit set, indicating the original request
143  * specified case-insensitive handling of names.
144  */
145 #define	TX_CI	((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
146 
147 /*
148  * Format of log records.
149  * The fields are carefully defined to allow them to be aligned
150  * and sized the same on sparc & intel architectures.
151  * Each log record has a common structure at the beginning.
152  *
153  * Note, lrc_seq holds two different sequence numbers. Whilst in memory
154  * it contains the transaction sequence number.  The log record on
155  * disk holds the sequence number of all log records which is used to
156  * ensure we don't replay the same record.  The two sequence numbers are
157  * different because the transactions can now be pushed out of order.
158  */
159 typedef struct {			/* common log record header */
160 	uint64_t	lrc_txtype;	/* intent log transaction type */
161 	uint64_t	lrc_reclen;	/* transaction record length */
162 	uint64_t	lrc_txg;	/* dmu transaction group number */
163 	uint64_t	lrc_seq;	/* see comment above */
164 } lr_t;
165 
166 /*
167  * Handle option extended vattr attributes.
168  *
169  * Whenever new attributes are added the version number
170  * will need to be updated as will code in
171  * zfs_log.c and zfs_replay.c
172  */
173 typedef struct {
174 	uint32_t	lr_attr_masksize; /* number of elements in array */
175 	uint32_t	lr_attr_bitmap; /* First entry of array */
176 	/* remainder of array and any additional fields */
177 } lr_attr_t;
178 
179 /*
180  * log record for creates without optional ACL.
181  * This log record does support optional xvattr_t attributes.
182  */
183 typedef struct {
184 	lr_t		lr_common;	/* common portion of log record */
185 	uint64_t	lr_doid;	/* object id of directory */
186 	uint64_t	lr_foid;	/* object id of created file object */
187 	uint64_t	lr_mode;	/* mode of object */
188 	uint64_t	lr_uid;		/* uid of object */
189 	uint64_t	lr_gid;		/* gid of object */
190 	uint64_t	lr_gen;		/* generation (txg of creation) */
191 	uint64_t	lr_crtime[2];	/* creation time */
192 	uint64_t	lr_rdev;	/* rdev of object to create */
193 	/* name of object to create follows this */
194 	/* for symlinks, link content follows name */
195 	/* for creates with xvattr data, the name follows the xvattr info */
196 } lr_create_t;
197 
198 /*
199  * FUID ACL record will be an array of ACEs from the original ACL.
200  * If this array includes ephemeral IDs, the record will also include
201  * an array of log-specific FUIDs to replace the ephemeral IDs.
202  * Only one copy of each unique domain will be present, so the log-specific
203  * FUIDs will use an index into a compressed domain table.  On replay this
204  * information will be used to construct real FUIDs (and bypass idmap,
205  * since it may not be available).
206  */
207 
208 /*
209  * Log record for creates with optional ACL
210  * This log record is also used for recording any FUID
211  * information needed for replaying the create.  If the
212  * file doesn't have any actual ACEs then the lr_aclcnt
213  * would be zero.
214  */
215 typedef struct {
216 	lr_create_t	lr_create;	/* common create portion */
217 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
218 	uint64_t	lr_domcnt;	/* number of unique domains */
219 	uint64_t	lr_fuidcnt;	/* number of real fuids */
220 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
221 	uint64_t	lr_acl_flags;	/* ACL flags */
222 	/* lr_acl_bytes number of variable sized ace's follows */
223 	/* if create is also setting xvattr's, then acl data follows xvattr */
224 	/* if ACE FUIDs are needed then they will follow the xvattr_t */
225 	/* Following the FUIDs will be the domain table information. */
226 	/* The FUIDs for the owner and group will be in the lr_create */
227 	/* portion of the record. */
228 	/* name follows ACL data */
229 } lr_acl_create_t;
230 
231 typedef struct {
232 	lr_t		lr_common;	/* common portion of log record */
233 	uint64_t	lr_doid;	/* obj id of directory */
234 	/* name of object to remove follows this */
235 } lr_remove_t;
236 
237 typedef struct {
238 	lr_t		lr_common;	/* common portion of log record */
239 	uint64_t	lr_doid;	/* obj id of directory */
240 	uint64_t	lr_link_obj;	/* obj id of link */
241 	/* name of object to link follows this */
242 } lr_link_t;
243 
244 typedef struct {
245 	lr_t		lr_common;	/* common portion of log record */
246 	uint64_t	lr_sdoid;	/* obj id of source directory */
247 	uint64_t	lr_tdoid;	/* obj id of target directory */
248 	/* 2 strings: names of source and destination follow this */
249 } lr_rename_t;
250 
251 typedef struct {
252 	lr_t		lr_common;	/* common portion of log record */
253 	uint64_t	lr_foid;	/* file object to write */
254 	uint64_t	lr_offset;	/* offset to write to */
255 	uint64_t	lr_length;	/* user data length to write */
256 	uint64_t	lr_blkoff;	/* offset represented by lr_blkptr */
257 	blkptr_t	lr_blkptr;	/* spa block pointer for replay */
258 	/* write data will follow for small writes */
259 } lr_write_t;
260 
261 typedef struct {
262 	lr_t		lr_common;	/* common portion of log record */
263 	uint64_t	lr_foid;	/* object id of file to truncate */
264 	uint64_t	lr_offset;	/* offset to truncate from */
265 	uint64_t	lr_length;	/* length to truncate */
266 } lr_truncate_t;
267 
268 typedef struct {
269 	lr_t		lr_common;	/* common portion of log record */
270 	uint64_t	lr_foid;	/* file object to change attributes */
271 	uint64_t	lr_mask;	/* mask of attributes to set */
272 	uint64_t	lr_mode;	/* mode to set */
273 	uint64_t	lr_uid;		/* uid to set */
274 	uint64_t	lr_gid;		/* gid to set */
275 	uint64_t	lr_size;	/* size to set */
276 	uint64_t	lr_atime[2];	/* access time */
277 	uint64_t	lr_mtime[2];	/* modification time */
278 	/* optional attribute lr_attr_t may be here */
279 } lr_setattr_t;
280 
281 typedef struct {
282 	lr_t		lr_common;	/* common portion of log record */
283 	uint64_t	lr_foid;	/* obj id of file */
284 	uint64_t	lr_aclcnt;	/* number of acl entries */
285 	/* lr_aclcnt number of ace_t entries follow this */
286 } lr_acl_v0_t;
287 
288 typedef struct {
289 	lr_t		lr_common;	/* common portion of log record */
290 	uint64_t	lr_foid;	/* obj id of file */
291 	uint64_t	lr_aclcnt;	/* number of ACEs in ACL */
292 	uint64_t	lr_domcnt;	/* number of unique domains */
293 	uint64_t	lr_fuidcnt;	/* number of real fuids */
294 	uint64_t	lr_acl_bytes;	/* number of bytes in ACL */
295 	uint64_t	lr_acl_flags;	/* ACL flags */
296 	/* lr_acl_bytes number of variable sized ace's follows */
297 } lr_acl_t;
298 
299 /*
300  * ZIL structure definitions, interface function prototype and globals.
301  */
302 
303 /*
304  * ZFS intent log transaction structure
305  */
306 typedef enum {
307 	WR_INDIRECT,	/* indirect - a large write (dmu_sync() data */
308 			/* and put blkptr in log, rather than actual data) */
309 	WR_COPIED,	/* immediate - data is copied into lr_write_t */
310 	WR_NEED_COPY,	/* immediate - data needs to be copied if pushed */
311 } itx_wr_state_t;
312 
313 typedef struct itx {
314 	list_node_t	itx_node;	/* linkage on zl_itx_list */
315 	void		*itx_private;	/* type-specific opaque data */
316 	itx_wr_state_t	itx_wr_state;	/* write state */
317 	uint8_t		itx_sync;	/* synchronous transaction */
318 	lr_t		itx_lr;		/* common part of log record */
319 	/* followed by type-specific part of lr_xx_t and its immediate data */
320 } itx_t;
321 
322 
323 /*
324  * zgd_t is passed through dmu_sync() to the callback routine zfs_get_done()
325  * to handle the cleanup of the dmu_sync() buffer write
326  */
327 typedef struct {
328 	zilog_t		*zgd_zilog;	/* zilog */
329 	blkptr_t	*zgd_bp;	/* block pointer */
330 	struct rl	*zgd_rl;	/* range lock */
331 } zgd_t;
332 
333 
334 typedef void zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
335     uint64_t txg);
336 typedef void zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
337     uint64_t txg);
338 typedef int zil_replay_func_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 extern void	zil_destroy(zilog_t *zilog, boolean_t keep_first);
356 extern void	zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx);
357 
358 extern itx_t	*zil_itx_create(uint64_t txtype, size_t lrsize);
359 extern uint64_t zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
360 
361 extern void	zil_commit(zilog_t *zilog, uint64_t seq, uint64_t oid);
362 
363 extern int	zil_claim(char *osname, void *txarg);
364 extern void	zil_sync(zilog_t *zilog, dmu_tx_t *tx);
365 extern void	zil_clean(zilog_t *zilog);
366 extern int	zil_is_committed(zilog_t *zilog);
367 
368 extern int	zil_suspend(zilog_t *zilog);
369 extern void	zil_resume(zilog_t *zilog);
370 
371 extern void	zil_add_vdev(zilog_t *zilog, uint64_t vdev);
372 
373 extern int zil_disable;
374 
375 #ifdef	__cplusplus
376 }
377 #endif
378 
379 #endif	/* _SYS_ZIL_H */
380