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 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 | |
37 | #ifdef __cplusplus |
38 | extern "C" { |
39 | #endif |
40 | |
41 | struct dsl_pool; |
42 | struct dsl_dataset; |
43 | |
44 | /* |
45 | * Intent log format: |
46 | * |
47 | * Each objset has its own intent log. The log header (zil_header_t) |
48 | * for objset N's intent log is kept in the Nth object of the SPA's |
49 | * intent_log objset. The log header points to a chain of log blocks, |
50 | * each of which contains log records (i.e., transactions) followed by |
51 | * a log block trailer (zil_trailer_t). The format of a log record |
52 | * depends on the record (or transaction) type, but all records begin |
53 | * with a common structure that defines the type, length, and txg. |
54 | */ |
55 | |
56 | /* |
57 | * Intent log header - this on disk structure holds fields to manage |
58 | * the log. All fields are 64 bit to easily handle cross architectures. |
59 | */ |
60 | typedef struct { |
61 | uint64_t ; /* txg in which log blocks were claimed */ |
62 | uint64_t ; /* highest replayed sequence number */ |
63 | blkptr_t ; /* log chain */ |
64 | uint64_t ; /* highest claimed block sequence number */ |
65 | uint64_t ; /* header flags */ |
66 | uint64_t ; /* highest claimed lr sequence number */ |
67 | uint64_t [3]; |
68 | } ; |
69 | |
70 | /* |
71 | * zh_flags bit settings |
72 | */ |
73 | #define ZIL_REPLAY_NEEDED 0x1 /* replay needed - internal only */ |
74 | #define ZIL_CLAIM_LR_SEQ_VALID 0x2 /* zh_claim_lr_seq field is valid */ |
75 | |
76 | /* |
77 | * Log block chaining. |
78 | * |
79 | * Log blocks are chained together. Originally they were chained at the |
80 | * end of the block. For performance reasons the chain was moved to the |
81 | * beginning of the block which allows writes for only the data being used. |
82 | * The older position is supported for backwards compatability. |
83 | * |
84 | * The zio_eck_t contains a zec_cksum which for the intent log is |
85 | * the sequence number of this log block. A seq of 0 is invalid. |
86 | * The zec_cksum is checked by the SPA against the sequence |
87 | * number passed in the blk_cksum field of the blkptr_t |
88 | */ |
89 | typedef struct zil_chain { |
90 | uint64_t zc_pad; |
91 | blkptr_t zc_next_blk; /* next block in chain */ |
92 | uint64_t zc_nused; /* bytes in log block used */ |
93 | zio_eck_t zc_eck; /* block trailer */ |
94 | } zil_chain_t; |
95 | |
96 | #define ZIL_MIN_BLKSZ 4096ULL |
97 | |
98 | /* |
99 | * The words of a log block checksum. |
100 | */ |
101 | #define ZIL_ZC_GUID_0 0 |
102 | #define ZIL_ZC_GUID_1 1 |
103 | #define ZIL_ZC_OBJSET 2 |
104 | #define ZIL_ZC_SEQ 3 |
105 | |
106 | typedef enum zil_create { |
107 | Z_FILE, |
108 | Z_DIR, |
109 | Z_XATTRDIR, |
110 | } zil_create_t; |
111 | |
112 | /* |
113 | * size of xvattr log section. |
114 | * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps |
115 | * for create time and a single 64 bit integer for all of the attributes, |
116 | * and 4 64 bit integers (32 bytes) for the scanstamp. |
117 | * |
118 | */ |
119 | |
120 | #define ZIL_XVAT_SIZE(mapsize) \ |
121 | sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \ |
122 | (sizeof (uint64_t) * 7) |
123 | |
124 | /* |
125 | * Size of ACL in log. The ACE data is padded out to properly align |
126 | * on 8 byte boundary. |
127 | */ |
128 | |
129 | #define ZIL_ACE_LENGTH(x) (roundup(x, sizeof (uint64_t))) |
130 | |
131 | /* |
132 | * Intent log transaction types and record structures |
133 | */ |
134 | #define TX_CREATE 1 /* Create file */ |
135 | #define TX_MKDIR 2 /* Make directory */ |
136 | #define TX_MKXATTR 3 /* Make XATTR directory */ |
137 | #define TX_SYMLINK 4 /* Create symbolic link to a file */ |
138 | #define TX_REMOVE 5 /* Remove file */ |
139 | #define TX_RMDIR 6 /* Remove directory */ |
140 | #define TX_LINK 7 /* Create hard link to a file */ |
141 | #define TX_RENAME 8 /* Rename a file */ |
142 | #define TX_WRITE 9 /* File write */ |
143 | #define TX_TRUNCATE 10 /* Truncate a file */ |
144 | #define TX_SETATTR 11 /* Set file attributes */ |
145 | #define TX_ACL_V0 12 /* Set old formatted ACL */ |
146 | #define TX_ACL 13 /* Set ACL */ |
147 | #define TX_CREATE_ACL 14 /* create with ACL */ |
148 | #define TX_CREATE_ATTR 15 /* create + attrs */ |
149 | #define TX_CREATE_ACL_ATTR 16 /* create with ACL + attrs */ |
150 | #define TX_MKDIR_ACL 17 /* mkdir with ACL */ |
151 | #define TX_MKDIR_ATTR 18 /* mkdir with attr */ |
152 | #define TX_MKDIR_ACL_ATTR 19 /* mkdir with ACL + attrs */ |
153 | #define TX_WRITE2 20 /* dmu_sync EALREADY write */ |
154 | #define TX_MAX_TYPE 21 /* Max transaction type */ |
155 | |
156 | /* |
157 | * The transactions for mkdir, symlink, remove, rmdir, link, and rename |
158 | * may have the following bit set, indicating the original request |
159 | * specified case-insensitive handling of names. |
160 | */ |
161 | #define TX_CI ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */ |
162 | |
163 | /* |
164 | * Transactions for write, truncate, setattr, acl_v0, and acl can be logged |
165 | * out of order. For convenience in the code, all such records must have |
166 | * lr_foid at the same offset. |
167 | */ |
168 | #define TX_OOO(txtype) \ |
169 | ((txtype) == TX_WRITE || \ |
170 | (txtype) == TX_TRUNCATE || \ |
171 | (txtype) == TX_SETATTR || \ |
172 | (txtype) == TX_ACL_V0 || \ |
173 | (txtype) == TX_ACL || \ |
174 | (txtype) == TX_WRITE2) |
175 | |
176 | /* |
177 | * Format of log records. |
178 | * The fields are carefully defined to allow them to be aligned |
179 | * and sized the same on sparc & intel architectures. |
180 | * Each log record has a common structure at the beginning. |
181 | * |
182 | * The log record on disk (lrc_seq) holds the sequence number of all log |
183 | * records which is used to ensure we don't replay the same record. |
184 | */ |
185 | typedef struct { /* common log record header */ |
186 | uint64_t lrc_txtype; /* intent log transaction type */ |
187 | uint64_t lrc_reclen; /* transaction record length */ |
188 | uint64_t lrc_txg; /* dmu transaction group number */ |
189 | uint64_t lrc_seq; /* see comment above */ |
190 | } lr_t; |
191 | |
192 | /* |
193 | * Common start of all out-of-order record types (TX_OOO() above). |
194 | */ |
195 | typedef struct { |
196 | lr_t lr_common; /* common portion of log record */ |
197 | uint64_t lr_foid; /* object id */ |
198 | } lr_ooo_t; |
199 | |
200 | /* |
201 | * Handle option extended vattr attributes. |
202 | * |
203 | * Whenever new attributes are added the version number |
204 | * will need to be updated as will code in |
205 | * zfs_log.c and zfs_replay.c |
206 | */ |
207 | typedef struct { |
208 | uint32_t lr_attr_masksize; /* number of elements in array */ |
209 | uint32_t lr_attr_bitmap; /* First entry of array */ |
210 | /* remainder of array and any additional fields */ |
211 | } lr_attr_t; |
212 | |
213 | /* |
214 | * log record for creates without optional ACL. |
215 | * This log record does support optional xvattr_t attributes. |
216 | */ |
217 | typedef struct { |
218 | lr_t lr_common; /* common portion of log record */ |
219 | uint64_t lr_doid; /* object id of directory */ |
220 | uint64_t lr_foid; /* object id of created file object */ |
221 | uint64_t lr_mode; /* mode of object */ |
222 | uint64_t lr_uid; /* uid of object */ |
223 | uint64_t lr_gid; /* gid of object */ |
224 | uint64_t lr_gen; /* generation (txg of creation) */ |
225 | uint64_t lr_crtime[2]; /* creation time */ |
226 | uint64_t lr_rdev; /* rdev of object to create */ |
227 | /* name of object to create follows this */ |
228 | /* for symlinks, link content follows name */ |
229 | /* for creates with xvattr data, the name follows the xvattr info */ |
230 | } lr_create_t; |
231 | |
232 | /* |
233 | * FUID ACL record will be an array of ACEs from the original ACL. |
234 | * If this array includes ephemeral IDs, the record will also include |
235 | * an array of log-specific FUIDs to replace the ephemeral IDs. |
236 | * Only one copy of each unique domain will be present, so the log-specific |
237 | * FUIDs will use an index into a compressed domain table. On replay this |
238 | * information will be used to construct real FUIDs (and bypass idmap, |
239 | * since it may not be available). |
240 | */ |
241 | |
242 | /* |
243 | * Log record for creates with optional ACL |
244 | * This log record is also used for recording any FUID |
245 | * information needed for replaying the create. If the |
246 | * file doesn't have any actual ACEs then the lr_aclcnt |
247 | * would be zero. |
248 | * |
249 | * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's. |
250 | * If create is also setting xvattr's, then acl data follows xvattr. |
251 | * If ACE FUIDs are needed then they will follow the xvattr_t. Following |
252 | * the FUIDs will be the domain table information. The FUIDs for the owner |
253 | * and group will be in lr_create. Name follows ACL data. |
254 | */ |
255 | typedef struct { |
256 | lr_create_t lr_create; /* common create portion */ |
257 | uint64_t lr_aclcnt; /* number of ACEs in ACL */ |
258 | uint64_t lr_domcnt; /* number of unique domains */ |
259 | uint64_t lr_fuidcnt; /* number of real fuids */ |
260 | uint64_t lr_acl_bytes; /* number of bytes in ACL */ |
261 | uint64_t lr_acl_flags; /* ACL flags */ |
262 | } lr_acl_create_t; |
263 | |
264 | typedef struct { |
265 | lr_t lr_common; /* common portion of log record */ |
266 | uint64_t lr_doid; /* obj id of directory */ |
267 | /* name of object to remove follows this */ |
268 | } lr_remove_t; |
269 | |
270 | typedef struct { |
271 | lr_t lr_common; /* common portion of log record */ |
272 | uint64_t lr_doid; /* obj id of directory */ |
273 | uint64_t lr_link_obj; /* obj id of link */ |
274 | /* name of object to link follows this */ |
275 | } lr_link_t; |
276 | |
277 | typedef struct { |
278 | lr_t lr_common; /* common portion of log record */ |
279 | uint64_t lr_sdoid; /* obj id of source directory */ |
280 | uint64_t lr_tdoid; /* obj id of target directory */ |
281 | /* 2 strings: names of source and destination follow this */ |
282 | } lr_rename_t; |
283 | |
284 | typedef struct { |
285 | lr_t lr_common; /* common portion of log record */ |
286 | uint64_t lr_foid; /* file object to write */ |
287 | uint64_t lr_offset; /* offset to write to */ |
288 | uint64_t lr_length; /* user data length to write */ |
289 | uint64_t lr_blkoff; /* no longer used */ |
290 | blkptr_t lr_blkptr; /* spa block pointer for replay */ |
291 | /* write data will follow for small writes */ |
292 | } lr_write_t; |
293 | |
294 | typedef struct { |
295 | lr_t lr_common; /* common portion of log record */ |
296 | uint64_t lr_foid; /* object id of file to truncate */ |
297 | uint64_t lr_offset; /* offset to truncate from */ |
298 | uint64_t lr_length; /* length to truncate */ |
299 | } lr_truncate_t; |
300 | |
301 | typedef struct { |
302 | lr_t lr_common; /* common portion of log record */ |
303 | uint64_t lr_foid; /* file object to change attributes */ |
304 | uint64_t lr_mask; /* mask of attributes to set */ |
305 | uint64_t lr_mode; /* mode to set */ |
306 | uint64_t lr_uid; /* uid to set */ |
307 | uint64_t lr_gid; /* gid to set */ |
308 | uint64_t lr_size; /* size to set */ |
309 | uint64_t lr_atime[2]; /* access time */ |
310 | uint64_t lr_mtime[2]; /* modification time */ |
311 | /* optional attribute lr_attr_t may be here */ |
312 | } lr_setattr_t; |
313 | |
314 | typedef struct { |
315 | lr_t lr_common; /* common portion of log record */ |
316 | uint64_t lr_foid; /* obj id of file */ |
317 | uint64_t lr_aclcnt; /* number of acl entries */ |
318 | /* lr_aclcnt number of ace_t entries follow this */ |
319 | } lr_acl_v0_t; |
320 | |
321 | typedef struct { |
322 | lr_t lr_common; /* common portion of log record */ |
323 | uint64_t lr_foid; /* obj id of file */ |
324 | uint64_t lr_aclcnt; /* number of ACEs in ACL */ |
325 | uint64_t lr_domcnt; /* number of unique domains */ |
326 | uint64_t lr_fuidcnt; /* number of real fuids */ |
327 | uint64_t lr_acl_bytes; /* number of bytes in ACL */ |
328 | uint64_t lr_acl_flags; /* ACL flags */ |
329 | /* lr_acl_bytes number of variable sized ace's follows */ |
330 | } lr_acl_t; |
331 | |
332 | /* |
333 | * ZIL structure definitions, interface function prototype and globals. |
334 | */ |
335 | |
336 | /* |
337 | * Writes are handled in three different ways: |
338 | * |
339 | * WR_INDIRECT: |
340 | * In this mode, if we need to commit the write later, then the block |
341 | * is immediately written into the file system (using dmu_sync), |
342 | * and a pointer to the block is put into the log record. |
343 | * When the txg commits the block is linked in. |
344 | * This saves additionally writing the data into the log record. |
345 | * There are a few requirements for this to occur: |
346 | * - write is greater than zfs/zvol_immediate_write_sz |
347 | * - not using slogs (as slogs are assumed to always be faster |
348 | * than writing into the main pool) |
349 | * - the write occupies only one block |
350 | * WR_COPIED: |
351 | * If we know we'll immediately be committing the |
352 | * transaction (FSYNC or FDSYNC), the we allocate a larger |
353 | * log record here for the data and copy the data in. |
354 | * WR_NEED_COPY: |
355 | * Otherwise we don't allocate a buffer, and *if* we need to |
356 | * flush the write later then a buffer is allocated and |
357 | * we retrieve the data using the dmu. |
358 | */ |
359 | typedef enum { |
360 | WR_INDIRECT, /* indirect - a large write (dmu_sync() data */ |
361 | /* and put blkptr in log, rather than actual data) */ |
362 | WR_COPIED, /* immediate - data is copied into lr_write_t */ |
363 | WR_NEED_COPY, /* immediate - data needs to be copied if pushed */ |
364 | WR_NUM_STATES /* number of states */ |
365 | } itx_wr_state_t; |
366 | |
367 | typedef struct itx { |
368 | list_node_t itx_node; /* linkage on zl_itx_list */ |
369 | void *itx_private; /* type-specific opaque data */ |
370 | itx_wr_state_t itx_wr_state; /* write state */ |
371 | uint8_t itx_sync; /* synchronous transaction */ |
372 | uint64_t itx_oid; /* object id */ |
373 | lr_t itx_lr; /* common part of log record */ |
374 | /* followed by type-specific part of lr_xx_t and its immediate data */ |
375 | } itx_t; |
376 | |
377 | typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg, |
378 | uint64_t txg); |
379 | typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg, |
380 | uint64_t txg); |
381 | typedef int zil_replay_func_t(); |
382 | typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio); |
383 | |
384 | extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, |
385 | zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg); |
386 | |
387 | extern void zil_init(void); |
388 | extern void zil_fini(void); |
389 | |
390 | extern zilog_t *zil_alloc(objset_t *os, zil_header_t *zh_phys); |
391 | extern void zil_free(zilog_t *zilog); |
392 | |
393 | extern zilog_t *zil_open(objset_t *os, zil_get_data_t *get_data); |
394 | extern void zil_close(zilog_t *zilog); |
395 | |
396 | extern void zil_replay(objset_t *os, void *arg, |
397 | zil_replay_func_t *replay_func[TX_MAX_TYPE]); |
398 | extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx); |
399 | extern void zil_destroy(zilog_t *zilog, boolean_t keep_first); |
400 | extern void zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx); |
401 | extern void zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx); |
402 | |
403 | extern itx_t *zil_itx_create(uint64_t txtype, size_t lrsize); |
404 | extern void zil_itx_destroy(itx_t *itx); |
405 | extern void zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx); |
406 | |
407 | extern void zil_async_to_sync(zilog_t *zilog, uint64_t oid); |
408 | extern void zil_commit(zilog_t *zilog, uint64_t oid); |
409 | |
410 | extern int zil_vdev_offline(const char *osname, void *txarg); |
411 | extern int zil_claim(struct dsl_pool *dp, |
412 | struct dsl_dataset *ds, void *txarg); |
413 | extern int zil_check_log_chain(struct dsl_pool *dp, |
414 | struct dsl_dataset *ds, void *tx); |
415 | extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx); |
416 | extern void zil_clean(zilog_t *zilog, uint64_t synced_txg); |
417 | |
418 | extern int zil_suspend(const char *osname, void **cookiep); |
419 | extern void zil_resume(void *cookie); |
420 | |
421 | extern void zil_add_block(zilog_t *zilog, const blkptr_t *bp); |
422 | extern int zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp); |
423 | |
424 | extern void zil_set_sync(zilog_t *zilog, uint64_t syncval); |
425 | |
426 | extern void zil_set_logbias(zilog_t *zilog, uint64_t slogval); |
427 | |
428 | extern int zil_replay_disable; |
429 | |
430 | #ifdef __cplusplus |
431 | } |
432 | #endif |
433 | |
434 | #endif /* _SYS_ZIL_H */ |
435 | |