]>
Commit | Line | Data |
---|---|---|
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 | */ | |
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 | ||
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_blk_seq; /* highest claimed block sequence number */ | |
61 | uint64_t zh_flags; /* header flags */ | |
62 | uint64_t zh_claim_lr_seq; /* highest claimed lr sequence number */ | |
63 | uint64_t zh_pad[3]; | |
64 | } zil_header_t; | |
65 | ||
66 | /* | |
67 | * zh_flags bit settings | |
68 | */ | |
69 | #define ZIL_REPLAY_NEEDED 0x1 /* replay needed - internal only */ | |
70 | #define ZIL_CLAIM_LR_SEQ_VALID 0x2 /* zh_claim_lr_seq field is valid */ | |
71 | ||
72 | /* | |
73 | * Log block chaining. | |
74 | * | |
75 | * Log blocks are chained together. Originally they were chained at the | |
76 | * end of the block. For performance reasons the chain was moved to the | |
77 | * beginning of the block which allows writes for only the data being used. | |
78 | * The older position is supported for backwards compatability. | |
79 | * | |
80 | * The zio_eck_t contains a zec_cksum which for the intent log is | |
81 | * the sequence number of this log block. A seq of 0 is invalid. | |
82 | * The zec_cksum is checked by the SPA against the sequence | |
83 | * number passed in the blk_cksum field of the blkptr_t | |
84 | */ | |
85 | typedef struct zil_chain { | |
86 | uint64_t zc_pad; | |
87 | blkptr_t zc_next_blk; /* next block in chain */ | |
88 | uint64_t zc_nused; /* bytes in log block used */ | |
89 | zio_eck_t zc_eck; /* block trailer */ | |
90 | } zil_chain_t; | |
91 | ||
92 | #define ZIL_MIN_BLKSZ 4096ULL | |
93 | #define ZIL_MAX_BLKSZ SPA_MAXBLOCKSIZE | |
94 | ||
95 | /* | |
96 | * The words of a log block checksum. | |
97 | */ | |
98 | #define ZIL_ZC_GUID_0 0 | |
99 | #define ZIL_ZC_GUID_1 1 | |
100 | #define ZIL_ZC_OBJSET 2 | |
101 | #define ZIL_ZC_SEQ 3 | |
102 | ||
103 | typedef enum zil_create { | |
104 | Z_FILE, | |
105 | Z_DIR, | |
106 | Z_XATTRDIR, | |
107 | } zil_create_t; | |
108 | ||
109 | /* | |
110 | * size of xvattr log section. | |
111 | * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps | |
112 | * for create time and a single 64 bit integer for all of the attributes, | |
113 | * and 4 64 bit integers (32 bytes) for the scanstamp. | |
114 | * | |
115 | */ | |
116 | ||
117 | #define ZIL_XVAT_SIZE(mapsize) \ | |
118 | sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \ | |
119 | (sizeof (uint64_t) * 7) | |
120 | ||
121 | /* | |
122 | * Size of ACL in log. The ACE data is padded out to properly align | |
123 | * on 8 byte boundary. | |
124 | */ | |
125 | ||
126 | #define ZIL_ACE_LENGTH(x) (roundup(x, sizeof (uint64_t))) | |
127 | ||
128 | /* | |
129 | * Intent log transaction types and record structures | |
130 | */ | |
131 | #define TX_CREATE 1 /* Create file */ | |
132 | #define TX_MKDIR 2 /* Make directory */ | |
133 | #define TX_MKXATTR 3 /* Make XATTR directory */ | |
134 | #define TX_SYMLINK 4 /* Create symbolic link to a file */ | |
135 | #define TX_REMOVE 5 /* Remove file */ | |
136 | #define TX_RMDIR 6 /* Remove directory */ | |
137 | #define TX_LINK 7 /* Create hard link to a file */ | |
138 | #define TX_RENAME 8 /* Rename a file */ | |
139 | #define TX_WRITE 9 /* File write */ | |
140 | #define TX_TRUNCATE 10 /* Truncate a file */ | |
141 | #define TX_SETATTR 11 /* Set file attributes */ | |
142 | #define TX_ACL_V0 12 /* Set old formatted ACL */ | |
143 | #define TX_ACL 13 /* Set ACL */ | |
144 | #define TX_CREATE_ACL 14 /* create with ACL */ | |
145 | #define TX_CREATE_ATTR 15 /* create + attrs */ | |
146 | #define TX_CREATE_ACL_ATTR 16 /* create with ACL + attrs */ | |
147 | #define TX_MKDIR_ACL 17 /* mkdir with ACL */ | |
148 | #define TX_MKDIR_ATTR 18 /* mkdir with attr */ | |
149 | #define TX_MKDIR_ACL_ATTR 19 /* mkdir with ACL + attrs */ | |
150 | #define TX_WRITE2 20 /* dmu_sync EALREADY write */ | |
151 | #define TX_MAX_TYPE 21 /* Max transaction type */ | |
152 | ||
153 | /* | |
154 | * The transactions for mkdir, symlink, remove, rmdir, link, and rename | |
155 | * may have the following bit set, indicating the original request | |
156 | * specified case-insensitive handling of names. | |
157 | */ | |
158 | #define TX_CI ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */ | |
159 | ||
160 | /* | |
161 | * Transactions for write, truncate, setattr, acl_v0, and acl can be logged | |
162 | * out of order. For convenience in the code, all such records must have | |
163 | * lr_foid at the same offset. | |
164 | */ | |
165 | #define TX_OOO(txtype) \ | |
166 | ((txtype) == TX_WRITE || \ | |
167 | (txtype) == TX_TRUNCATE || \ | |
168 | (txtype) == TX_SETATTR || \ | |
169 | (txtype) == TX_ACL_V0 || \ | |
170 | (txtype) == TX_ACL || \ | |
171 | (txtype) == TX_WRITE2) | |
172 | ||
173 | /* | |
174 | * Format of log records. | |
175 | * The fields are carefully defined to allow them to be aligned | |
176 | * and sized the same on sparc & intel architectures. | |
177 | * Each log record has a common structure at the beginning. | |
178 | * | |
179 | * The log record on disk (lrc_seq) holds the sequence number of all log | |
180 | * records which is used to ensure we don't replay the same record. | |
181 | */ | |
182 | typedef struct { /* common log record header */ | |
183 | uint64_t lrc_txtype; /* intent log transaction type */ | |
184 | uint64_t lrc_reclen; /* transaction record length */ | |
185 | uint64_t lrc_txg; /* dmu transaction group number */ | |
186 | uint64_t lrc_seq; /* see comment above */ | |
187 | } lr_t; | |
188 | ||
189 | /* | |
190 | * Common start of all out-of-order record types (TX_OOO() above). | |
191 | */ | |
192 | typedef struct { | |
193 | lr_t lr_common; /* common portion of log record */ | |
194 | uint64_t lr_foid; /* object id */ | |
195 | } lr_ooo_t; | |
196 | ||
197 | /* | |
198 | * Handle option extended vattr attributes. | |
199 | * | |
200 | * Whenever new attributes are added the version number | |
201 | * will need to be updated as will code in | |
202 | * zfs_log.c and zfs_replay.c | |
203 | */ | |
204 | typedef struct { | |
205 | uint32_t lr_attr_masksize; /* number of elements in array */ | |
206 | uint32_t lr_attr_bitmap; /* First entry of array */ | |
207 | /* remainder of array and any additional fields */ | |
208 | } lr_attr_t; | |
209 | ||
210 | /* | |
211 | * log record for creates without optional ACL. | |
212 | * This log record does support optional xvattr_t attributes. | |
213 | */ | |
214 | typedef struct { | |
215 | lr_t lr_common; /* common portion of log record */ | |
216 | uint64_t lr_doid; /* object id of directory */ | |
217 | uint64_t lr_foid; /* object id of created file object */ | |
218 | uint64_t lr_mode; /* mode of object */ | |
219 | uint64_t lr_uid; /* uid of object */ | |
220 | uint64_t lr_gid; /* gid of object */ | |
221 | uint64_t lr_gen; /* generation (txg of creation) */ | |
222 | uint64_t lr_crtime[2]; /* creation time */ | |
223 | uint64_t lr_rdev; /* rdev of object to create */ | |
224 | /* name of object to create follows this */ | |
225 | /* for symlinks, link content follows name */ | |
226 | /* for creates with xvattr data, the name follows the xvattr info */ | |
227 | } lr_create_t; | |
228 | ||
229 | /* | |
230 | * FUID ACL record will be an array of ACEs from the original ACL. | |
231 | * If this array includes ephemeral IDs, the record will also include | |
232 | * an array of log-specific FUIDs to replace the ephemeral IDs. | |
233 | * Only one copy of each unique domain will be present, so the log-specific | |
234 | * FUIDs will use an index into a compressed domain table. On replay this | |
235 | * information will be used to construct real FUIDs (and bypass idmap, | |
236 | * since it may not be available). | |
237 | */ | |
238 | ||
239 | /* | |
240 | * Log record for creates with optional ACL | |
241 | * This log record is also used for recording any FUID | |
242 | * information needed for replaying the create. If the | |
243 | * file doesn't have any actual ACEs then the lr_aclcnt | |
244 | * would be zero. | |
245 | * | |
246 | * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's. | |
247 | * If create is also setting xvattr's, then acl data follows xvattr. | |
248 | * If ACE FUIDs are needed then they will follow the xvattr_t. Following | |
249 | * the FUIDs will be the domain table information. The FUIDs for the owner | |
250 | * and group will be in lr_create. Name follows ACL data. | |
251 | */ | |
252 | typedef struct { | |
253 | lr_create_t lr_create; /* common create portion */ | |
254 | uint64_t lr_aclcnt; /* number of ACEs in ACL */ | |
255 | uint64_t lr_domcnt; /* number of unique domains */ | |
256 | uint64_t lr_fuidcnt; /* number of real fuids */ | |
257 | uint64_t lr_acl_bytes; /* number of bytes in ACL */ | |
258 | uint64_t lr_acl_flags; /* ACL flags */ | |
259 | } lr_acl_create_t; | |
260 | ||
261 | typedef struct { | |
262 | lr_t lr_common; /* common portion of log record */ | |
263 | uint64_t lr_doid; /* obj id of directory */ | |
264 | /* name of object to remove follows this */ | |
265 | } lr_remove_t; | |
266 | ||
267 | typedef struct { | |
268 | lr_t lr_common; /* common portion of log record */ | |
269 | uint64_t lr_doid; /* obj id of directory */ | |
270 | uint64_t lr_link_obj; /* obj id of link */ | |
271 | /* name of object to link follows this */ | |
272 | } lr_link_t; | |
273 | ||
274 | typedef struct { | |
275 | lr_t lr_common; /* common portion of log record */ | |
276 | uint64_t lr_sdoid; /* obj id of source directory */ | |
277 | uint64_t lr_tdoid; /* obj id of target directory */ | |
278 | /* 2 strings: names of source and destination follow this */ | |
279 | } lr_rename_t; | |
280 | ||
281 | typedef struct { | |
282 | lr_t lr_common; /* common portion of log record */ | |
283 | uint64_t lr_foid; /* file object to write */ | |
284 | uint64_t lr_offset; /* offset to write to */ | |
285 | uint64_t lr_length; /* user data length to write */ | |
286 | uint64_t lr_blkoff; /* no longer used */ | |
287 | blkptr_t lr_blkptr; /* spa block pointer for replay */ | |
288 | /* write data will follow for small writes */ | |
289 | } lr_write_t; | |
290 | ||
291 | typedef struct { | |
292 | lr_t lr_common; /* common portion of log record */ | |
293 | uint64_t lr_foid; /* object id of file to truncate */ | |
294 | uint64_t lr_offset; /* offset to truncate from */ | |
295 | uint64_t lr_length; /* length to truncate */ | |
296 | } lr_truncate_t; | |
297 | ||
298 | typedef struct { | |
299 | lr_t lr_common; /* common portion of log record */ | |
300 | uint64_t lr_foid; /* file object to change attributes */ | |
301 | uint64_t lr_mask; /* mask of attributes to set */ | |
302 | uint64_t lr_mode; /* mode to set */ | |
303 | uint64_t lr_uid; /* uid to set */ | |
304 | uint64_t lr_gid; /* gid to set */ | |
305 | uint64_t lr_size; /* size to set */ | |
306 | uint64_t lr_atime[2]; /* access time */ | |
307 | uint64_t lr_mtime[2]; /* modification time */ | |
308 | /* optional attribute lr_attr_t may be here */ | |
309 | } lr_setattr_t; | |
310 | ||
311 | typedef struct { | |
312 | lr_t lr_common; /* common portion of log record */ | |
313 | uint64_t lr_foid; /* obj id of file */ | |
314 | uint64_t lr_aclcnt; /* number of acl entries */ | |
315 | /* lr_aclcnt number of ace_t entries follow this */ | |
316 | } lr_acl_v0_t; | |
317 | ||
318 | typedef struct { | |
319 | lr_t lr_common; /* common portion of log record */ | |
320 | uint64_t lr_foid; /* obj id of file */ | |
321 | uint64_t lr_aclcnt; /* number of ACEs in ACL */ | |
322 | uint64_t lr_domcnt; /* number of unique domains */ | |
323 | uint64_t lr_fuidcnt; /* number of real fuids */ | |
324 | uint64_t lr_acl_bytes; /* number of bytes in ACL */ | |
325 | uint64_t lr_acl_flags; /* ACL flags */ | |
326 | /* lr_acl_bytes number of variable sized ace's follows */ | |
327 | } lr_acl_t; | |
328 | ||
329 | /* | |
330 | * ZIL structure definitions, interface function prototype and globals. | |
331 | */ | |
332 | ||
333 | /* | |
334 | * Writes are handled in three different ways: | |
335 | * | |
336 | * WR_INDIRECT: | |
337 | * In this mode, if we need to commit the write later, then the block | |
338 | * is immediately written into the file system (using dmu_sync), | |
339 | * and a pointer to the block is put into the log record. | |
340 | * When the txg commits the block is linked in. | |
341 | * This saves additionally writing the data into the log record. | |
342 | * There are a few requirements for this to occur: | |
343 | * - write is greater than zfs/zvol_immediate_write_sz | |
344 | * - not using slogs (as slogs are assumed to always be faster | |
345 | * than writing into the main pool) | |
346 | * - the write occupies only one block | |
347 | * WR_COPIED: | |
348 | * If we know we'll immediately be committing the | |
349 | * transaction (FSYNC or FDSYNC), the we allocate a larger | |
350 | * log record here for the data and copy the data in. | |
351 | * WR_NEED_COPY: | |
352 | * Otherwise we don't allocate a buffer, and *if* we need to | |
353 | * flush the write later then a buffer is allocated and | |
354 | * we retrieve the data using the dmu. | |
355 | */ | |
356 | typedef enum { | |
357 | WR_INDIRECT, /* indirect - a large write (dmu_sync() data */ | |
358 | /* and put blkptr in log, rather than actual data) */ | |
359 | WR_COPIED, /* immediate - data is copied into lr_write_t */ | |
360 | WR_NEED_COPY, /* immediate - data needs to be copied if pushed */ | |
361 | WR_NUM_STATES /* number of states */ | |
362 | } itx_wr_state_t; | |
363 | ||
364 | typedef void (*zil_callback_t)(void *data); | |
365 | ||
366 | typedef struct itx { | |
367 | list_node_t itx_node; /* linkage on zl_itx_list */ | |
368 | void *itx_private; /* type-specific opaque data */ | |
369 | itx_wr_state_t itx_wr_state; /* write state */ | |
370 | uint8_t itx_sync; /* synchronous transaction */ | |
371 | zil_callback_t itx_callback; /* Called when the itx is persistent */ | |
372 | void *itx_callback_data; /* User data for the callback */ | |
373 | uint64_t itx_sod; /* record size on disk */ | |
374 | uint64_t itx_oid; /* object id */ | |
375 | lr_t itx_lr; /* common part of log record */ | |
376 | /* followed by type-specific part of lr_xx_t and its immediate data */ | |
377 | } itx_t; | |
378 | ||
379 | /* | |
380 | * Used for zil kstat. | |
381 | */ | |
382 | typedef struct zil_stats { | |
383 | /* | |
384 | * Number of times a ZIL commit (e.g. fsync) has been requested. | |
385 | */ | |
386 | kstat_named_t zil_commit_count; | |
387 | ||
388 | /* | |
389 | * Number of times the ZIL has been flushed to stable storage. | |
390 | * This is less than zil_commit_count when commits are "merged" | |
391 | * (see the documentation above zil_commit()). | |
392 | */ | |
393 | kstat_named_t zil_commit_writer_count; | |
394 | ||
395 | /* | |
396 | * Number of transactions (reads, writes, renames, etc.) | |
397 | * that have been commited. | |
398 | */ | |
399 | kstat_named_t zil_itx_count; | |
400 | ||
401 | /* | |
402 | * See the documentation for itx_wr_state_t above. | |
403 | * Note that "bytes" accumulates the length of the transactions | |
404 | * (i.e. data), not the actual log record sizes. | |
405 | */ | |
406 | kstat_named_t zil_itx_indirect_count; | |
407 | kstat_named_t zil_itx_indirect_bytes; | |
408 | kstat_named_t zil_itx_copied_count; | |
409 | kstat_named_t zil_itx_copied_bytes; | |
410 | kstat_named_t zil_itx_needcopy_count; | |
411 | kstat_named_t zil_itx_needcopy_bytes; | |
412 | ||
413 | /* | |
414 | * Transactions which have been allocated to the "normal" | |
415 | * (i.e. not slog) storage pool. Note that "bytes" accumulate | |
416 | * the actual log record sizes - which do not include the actual | |
417 | * data in case of indirect writes. | |
418 | */ | |
419 | kstat_named_t zil_itx_metaslab_normal_count; | |
420 | kstat_named_t zil_itx_metaslab_normal_bytes; | |
421 | ||
422 | /* | |
423 | * Transactions which have been allocated to the "slog" storage pool. | |
424 | * If there are no separate log devices, this is the same as the | |
425 | * "normal" pool. | |
426 | */ | |
427 | kstat_named_t zil_itx_metaslab_slog_count; | |
428 | kstat_named_t zil_itx_metaslab_slog_bytes; | |
429 | } zil_stats_t; | |
430 | ||
431 | extern zil_stats_t zil_stats; | |
432 | ||
433 | #define ZIL_STAT_INCR(stat, val) \ | |
434 | atomic_add_64(&zil_stats.stat.value.ui64, (val)); | |
435 | #define ZIL_STAT_BUMP(stat) \ | |
436 | ZIL_STAT_INCR(stat, 1); | |
437 | ||
438 | typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg, | |
439 | uint64_t txg); | |
440 | typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg, | |
441 | uint64_t txg); | |
442 | typedef int (*const zil_replay_func_t)(void *, char *, boolean_t); | |
443 | typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio); | |
444 | ||
445 | extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, | |
446 | zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg); | |
447 | ||
448 | extern void zil_init(void); | |
449 | extern void zil_fini(void); | |
450 | ||
451 | extern zilog_t *zil_alloc(objset_t *os, zil_header_t *zh_phys); | |
452 | extern void zil_free(zilog_t *zilog); | |
453 | ||
454 | extern zilog_t *zil_open(objset_t *os, zil_get_data_t *get_data); | |
455 | extern void zil_close(zilog_t *zilog); | |
456 | ||
457 | extern void zil_replay(objset_t *os, void *arg, | |
458 | zil_replay_func_t replay_func[TX_MAX_TYPE]); | |
459 | extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx); | |
460 | extern void zil_destroy(zilog_t *zilog, boolean_t keep_first); | |
461 | extern void zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx); | |
462 | extern void zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx); | |
463 | ||
464 | extern itx_t *zil_itx_create(uint64_t txtype, size_t lrsize); | |
465 | extern void zil_itx_destroy(itx_t *itx); | |
466 | extern void zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx); | |
467 | ||
468 | extern void zil_commit(zilog_t *zilog, uint64_t oid); | |
469 | ||
470 | extern int zil_vdev_offline(const char *osname, void *txarg); | |
471 | extern int zil_claim(const char *osname, void *txarg); | |
472 | extern int zil_check_log_chain(const char *osname, void *txarg); | |
473 | extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx); | |
474 | extern void zil_clean(zilog_t *zilog, uint64_t synced_txg); | |
475 | ||
476 | extern int zil_suspend(const char *osname, void **cookiep); | |
477 | extern void zil_resume(void *cookie); | |
478 | ||
479 | extern void zil_add_block(zilog_t *zilog, const blkptr_t *bp); | |
480 | extern int zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp); | |
481 | ||
482 | extern void zil_set_sync(zilog_t *zilog, uint64_t syncval); | |
483 | ||
484 | extern void zil_set_logbias(zilog_t *zilog, uint64_t slogval); | |
485 | ||
486 | extern int zil_replay_disable; | |
487 | ||
488 | #ifdef __cplusplus | |
489 | } | |
490 | #endif | |
491 | ||
492 | #endif /* _SYS_ZIL_H */ |