Print this page
5269 zfs: zpool import slow
PORTING: this code relies on the property of taskq_wait to wait
until no more tasks are queued and no more tasks are active. As
we always queue new tasks from within other tasks, task_wait
reliably waits for the full recursion to finish, even though we
enqueue new tasks after taskq_wait has been called.
On platforms other than illumos, taskq_wait may not have this
property.
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/sys/zil.h
+++ new/usr/src/uts/common/fs/zfs/sys/zil.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 24 */
25 25
26 26 /* Portions Copyright 2010 Robert Milkowski */
27 27
28 28 #ifndef _SYS_ZIL_H
29 29 #define _SYS_ZIL_H
|
↓ open down ↓ |
29 lines elided |
↑ open up ↑ |
30 30
31 31 #include <sys/types.h>
32 32 #include <sys/spa.h>
33 33 #include <sys/zio.h>
34 34 #include <sys/dmu.h>
35 35
36 36 #ifdef __cplusplus
37 37 extern "C" {
38 38 #endif
39 39
40 +struct dsl_pool;
41 +struct dsl_dataset;
42 +
40 43 /*
41 44 * Intent log format:
42 45 *
43 46 * Each objset has its own intent log. The log header (zil_header_t)
44 47 * for objset N's intent log is kept in the Nth object of the SPA's
45 48 * intent_log objset. The log header points to a chain of log blocks,
46 49 * each of which contains log records (i.e., transactions) followed by
47 50 * a log block trailer (zil_trailer_t). The format of a log record
48 51 * depends on the record (or transaction) type, but all records begin
49 52 * with a common structure that defines the type, length, and txg.
50 53 */
51 54
52 55 /*
53 56 * Intent log header - this on disk structure holds fields to manage
54 57 * the log. All fields are 64 bit to easily handle cross architectures.
55 58 */
56 59 typedef struct zil_header {
57 60 uint64_t zh_claim_txg; /* txg in which log blocks were claimed */
58 61 uint64_t zh_replay_seq; /* highest replayed sequence number */
59 62 blkptr_t zh_log; /* log chain */
60 63 uint64_t zh_claim_blk_seq; /* highest claimed block sequence number */
61 64 uint64_t zh_flags; /* header flags */
62 65 uint64_t zh_claim_lr_seq; /* highest claimed lr sequence number */
63 66 uint64_t zh_pad[3];
64 67 } zil_header_t;
65 68
66 69 /*
67 70 * zh_flags bit settings
68 71 */
69 72 #define ZIL_REPLAY_NEEDED 0x1 /* replay needed - internal only */
70 73 #define ZIL_CLAIM_LR_SEQ_VALID 0x2 /* zh_claim_lr_seq field is valid */
71 74
72 75 /*
73 76 * Log block chaining.
74 77 *
75 78 * Log blocks are chained together. Originally they were chained at the
76 79 * end of the block. For performance reasons the chain was moved to the
77 80 * beginning of the block which allows writes for only the data being used.
78 81 * The older position is supported for backwards compatability.
79 82 *
80 83 * The zio_eck_t contains a zec_cksum which for the intent log is
81 84 * the sequence number of this log block. A seq of 0 is invalid.
82 85 * The zec_cksum is checked by the SPA against the sequence
83 86 * number passed in the blk_cksum field of the blkptr_t
84 87 */
85 88 typedef struct zil_chain {
86 89 uint64_t zc_pad;
87 90 blkptr_t zc_next_blk; /* next block in chain */
88 91 uint64_t zc_nused; /* bytes in log block used */
89 92 zio_eck_t zc_eck; /* block trailer */
90 93 } zil_chain_t;
91 94
92 95 #define ZIL_MIN_BLKSZ 4096ULL
93 96
94 97 /*
95 98 * The words of a log block checksum.
96 99 */
97 100 #define ZIL_ZC_GUID_0 0
98 101 #define ZIL_ZC_GUID_1 1
99 102 #define ZIL_ZC_OBJSET 2
100 103 #define ZIL_ZC_SEQ 3
101 104
102 105 typedef enum zil_create {
103 106 Z_FILE,
104 107 Z_DIR,
105 108 Z_XATTRDIR,
106 109 } zil_create_t;
107 110
108 111 /*
109 112 * size of xvattr log section.
110 113 * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
111 114 * for create time and a single 64 bit integer for all of the attributes,
112 115 * and 4 64 bit integers (32 bytes) for the scanstamp.
113 116 *
114 117 */
115 118
116 119 #define ZIL_XVAT_SIZE(mapsize) \
117 120 sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
118 121 (sizeof (uint64_t) * 7)
119 122
120 123 /*
121 124 * Size of ACL in log. The ACE data is padded out to properly align
122 125 * on 8 byte boundary.
123 126 */
124 127
125 128 #define ZIL_ACE_LENGTH(x) (roundup(x, sizeof (uint64_t)))
126 129
127 130 /*
128 131 * Intent log transaction types and record structures
129 132 */
130 133 #define TX_CREATE 1 /* Create file */
131 134 #define TX_MKDIR 2 /* Make directory */
132 135 #define TX_MKXATTR 3 /* Make XATTR directory */
133 136 #define TX_SYMLINK 4 /* Create symbolic link to a file */
134 137 #define TX_REMOVE 5 /* Remove file */
135 138 #define TX_RMDIR 6 /* Remove directory */
136 139 #define TX_LINK 7 /* Create hard link to a file */
137 140 #define TX_RENAME 8 /* Rename a file */
138 141 #define TX_WRITE 9 /* File write */
139 142 #define TX_TRUNCATE 10 /* Truncate a file */
140 143 #define TX_SETATTR 11 /* Set file attributes */
141 144 #define TX_ACL_V0 12 /* Set old formatted ACL */
142 145 #define TX_ACL 13 /* Set ACL */
143 146 #define TX_CREATE_ACL 14 /* create with ACL */
144 147 #define TX_CREATE_ATTR 15 /* create + attrs */
145 148 #define TX_CREATE_ACL_ATTR 16 /* create with ACL + attrs */
146 149 #define TX_MKDIR_ACL 17 /* mkdir with ACL */
147 150 #define TX_MKDIR_ATTR 18 /* mkdir with attr */
148 151 #define TX_MKDIR_ACL_ATTR 19 /* mkdir with ACL + attrs */
149 152 #define TX_WRITE2 20 /* dmu_sync EALREADY write */
150 153 #define TX_MAX_TYPE 21 /* Max transaction type */
151 154
152 155 /*
153 156 * The transactions for mkdir, symlink, remove, rmdir, link, and rename
154 157 * may have the following bit set, indicating the original request
155 158 * specified case-insensitive handling of names.
156 159 */
157 160 #define TX_CI ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
158 161
159 162 /*
160 163 * Transactions for write, truncate, setattr, acl_v0, and acl can be logged
161 164 * out of order. For convenience in the code, all such records must have
162 165 * lr_foid at the same offset.
163 166 */
164 167 #define TX_OOO(txtype) \
165 168 ((txtype) == TX_WRITE || \
166 169 (txtype) == TX_TRUNCATE || \
167 170 (txtype) == TX_SETATTR || \
168 171 (txtype) == TX_ACL_V0 || \
169 172 (txtype) == TX_ACL || \
170 173 (txtype) == TX_WRITE2)
171 174
172 175 /*
173 176 * Format of log records.
174 177 * The fields are carefully defined to allow them to be aligned
175 178 * and sized the same on sparc & intel architectures.
176 179 * Each log record has a common structure at the beginning.
177 180 *
178 181 * The log record on disk (lrc_seq) holds the sequence number of all log
179 182 * records which is used to ensure we don't replay the same record.
180 183 */
181 184 typedef struct { /* common log record header */
182 185 uint64_t lrc_txtype; /* intent log transaction type */
183 186 uint64_t lrc_reclen; /* transaction record length */
184 187 uint64_t lrc_txg; /* dmu transaction group number */
185 188 uint64_t lrc_seq; /* see comment above */
186 189 } lr_t;
187 190
188 191 /*
189 192 * Common start of all out-of-order record types (TX_OOO() above).
190 193 */
191 194 typedef struct {
192 195 lr_t lr_common; /* common portion of log record */
193 196 uint64_t lr_foid; /* object id */
194 197 } lr_ooo_t;
195 198
196 199 /*
197 200 * Handle option extended vattr attributes.
198 201 *
199 202 * Whenever new attributes are added the version number
200 203 * will need to be updated as will code in
201 204 * zfs_log.c and zfs_replay.c
202 205 */
203 206 typedef struct {
204 207 uint32_t lr_attr_masksize; /* number of elements in array */
205 208 uint32_t lr_attr_bitmap; /* First entry of array */
206 209 /* remainder of array and any additional fields */
207 210 } lr_attr_t;
208 211
209 212 /*
210 213 * log record for creates without optional ACL.
211 214 * This log record does support optional xvattr_t attributes.
212 215 */
213 216 typedef struct {
214 217 lr_t lr_common; /* common portion of log record */
215 218 uint64_t lr_doid; /* object id of directory */
216 219 uint64_t lr_foid; /* object id of created file object */
217 220 uint64_t lr_mode; /* mode of object */
218 221 uint64_t lr_uid; /* uid of object */
219 222 uint64_t lr_gid; /* gid of object */
220 223 uint64_t lr_gen; /* generation (txg of creation) */
221 224 uint64_t lr_crtime[2]; /* creation time */
222 225 uint64_t lr_rdev; /* rdev of object to create */
223 226 /* name of object to create follows this */
224 227 /* for symlinks, link content follows name */
225 228 /* for creates with xvattr data, the name follows the xvattr info */
226 229 } lr_create_t;
227 230
228 231 /*
229 232 * FUID ACL record will be an array of ACEs from the original ACL.
230 233 * If this array includes ephemeral IDs, the record will also include
231 234 * an array of log-specific FUIDs to replace the ephemeral IDs.
232 235 * Only one copy of each unique domain will be present, so the log-specific
233 236 * FUIDs will use an index into a compressed domain table. On replay this
234 237 * information will be used to construct real FUIDs (and bypass idmap,
235 238 * since it may not be available).
236 239 */
237 240
238 241 /*
239 242 * Log record for creates with optional ACL
240 243 * This log record is also used for recording any FUID
241 244 * information needed for replaying the create. If the
242 245 * file doesn't have any actual ACEs then the lr_aclcnt
243 246 * would be zero.
244 247 *
245 248 * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
246 249 * If create is also setting xvattr's, then acl data follows xvattr.
247 250 * If ACE FUIDs are needed then they will follow the xvattr_t. Following
248 251 * the FUIDs will be the domain table information. The FUIDs for the owner
249 252 * and group will be in lr_create. Name follows ACL data.
250 253 */
251 254 typedef struct {
252 255 lr_create_t lr_create; /* common create portion */
253 256 uint64_t lr_aclcnt; /* number of ACEs in ACL */
254 257 uint64_t lr_domcnt; /* number of unique domains */
255 258 uint64_t lr_fuidcnt; /* number of real fuids */
256 259 uint64_t lr_acl_bytes; /* number of bytes in ACL */
257 260 uint64_t lr_acl_flags; /* ACL flags */
258 261 } lr_acl_create_t;
259 262
260 263 typedef struct {
261 264 lr_t lr_common; /* common portion of log record */
262 265 uint64_t lr_doid; /* obj id of directory */
263 266 /* name of object to remove follows this */
264 267 } lr_remove_t;
265 268
266 269 typedef struct {
267 270 lr_t lr_common; /* common portion of log record */
268 271 uint64_t lr_doid; /* obj id of directory */
269 272 uint64_t lr_link_obj; /* obj id of link */
270 273 /* name of object to link follows this */
271 274 } lr_link_t;
272 275
273 276 typedef struct {
274 277 lr_t lr_common; /* common portion of log record */
275 278 uint64_t lr_sdoid; /* obj id of source directory */
276 279 uint64_t lr_tdoid; /* obj id of target directory */
277 280 /* 2 strings: names of source and destination follow this */
278 281 } lr_rename_t;
279 282
280 283 typedef struct {
281 284 lr_t lr_common; /* common portion of log record */
282 285 uint64_t lr_foid; /* file object to write */
283 286 uint64_t lr_offset; /* offset to write to */
284 287 uint64_t lr_length; /* user data length to write */
285 288 uint64_t lr_blkoff; /* no longer used */
286 289 blkptr_t lr_blkptr; /* spa block pointer for replay */
287 290 /* write data will follow for small writes */
288 291 } lr_write_t;
289 292
290 293 typedef struct {
291 294 lr_t lr_common; /* common portion of log record */
292 295 uint64_t lr_foid; /* object id of file to truncate */
293 296 uint64_t lr_offset; /* offset to truncate from */
294 297 uint64_t lr_length; /* length to truncate */
295 298 } lr_truncate_t;
296 299
297 300 typedef struct {
298 301 lr_t lr_common; /* common portion of log record */
299 302 uint64_t lr_foid; /* file object to change attributes */
300 303 uint64_t lr_mask; /* mask of attributes to set */
301 304 uint64_t lr_mode; /* mode to set */
302 305 uint64_t lr_uid; /* uid to set */
303 306 uint64_t lr_gid; /* gid to set */
304 307 uint64_t lr_size; /* size to set */
305 308 uint64_t lr_atime[2]; /* access time */
306 309 uint64_t lr_mtime[2]; /* modification time */
307 310 /* optional attribute lr_attr_t may be here */
308 311 } lr_setattr_t;
309 312
310 313 typedef struct {
311 314 lr_t lr_common; /* common portion of log record */
312 315 uint64_t lr_foid; /* obj id of file */
313 316 uint64_t lr_aclcnt; /* number of acl entries */
314 317 /* lr_aclcnt number of ace_t entries follow this */
315 318 } lr_acl_v0_t;
316 319
317 320 typedef struct {
318 321 lr_t lr_common; /* common portion of log record */
319 322 uint64_t lr_foid; /* obj id of file */
320 323 uint64_t lr_aclcnt; /* number of ACEs in ACL */
321 324 uint64_t lr_domcnt; /* number of unique domains */
322 325 uint64_t lr_fuidcnt; /* number of real fuids */
323 326 uint64_t lr_acl_bytes; /* number of bytes in ACL */
324 327 uint64_t lr_acl_flags; /* ACL flags */
325 328 /* lr_acl_bytes number of variable sized ace's follows */
326 329 } lr_acl_t;
327 330
328 331 /*
329 332 * ZIL structure definitions, interface function prototype and globals.
330 333 */
331 334
332 335 /*
333 336 * Writes are handled in three different ways:
334 337 *
335 338 * WR_INDIRECT:
336 339 * In this mode, if we need to commit the write later, then the block
337 340 * is immediately written into the file system (using dmu_sync),
338 341 * and a pointer to the block is put into the log record.
339 342 * When the txg commits the block is linked in.
340 343 * This saves additionally writing the data into the log record.
341 344 * There are a few requirements for this to occur:
342 345 * - write is greater than zfs/zvol_immediate_write_sz
343 346 * - not using slogs (as slogs are assumed to always be faster
344 347 * than writing into the main pool)
345 348 * - the write occupies only one block
346 349 * WR_COPIED:
347 350 * If we know we'll immediately be committing the
348 351 * transaction (FSYNC or FDSYNC), the we allocate a larger
349 352 * log record here for the data and copy the data in.
350 353 * WR_NEED_COPY:
351 354 * Otherwise we don't allocate a buffer, and *if* we need to
352 355 * flush the write later then a buffer is allocated and
353 356 * we retrieve the data using the dmu.
354 357 */
355 358 typedef enum {
356 359 WR_INDIRECT, /* indirect - a large write (dmu_sync() data */
357 360 /* and put blkptr in log, rather than actual data) */
358 361 WR_COPIED, /* immediate - data is copied into lr_write_t */
359 362 WR_NEED_COPY, /* immediate - data needs to be copied if pushed */
360 363 WR_NUM_STATES /* number of states */
361 364 } itx_wr_state_t;
362 365
363 366 typedef struct itx {
364 367 list_node_t itx_node; /* linkage on zl_itx_list */
365 368 void *itx_private; /* type-specific opaque data */
366 369 itx_wr_state_t itx_wr_state; /* write state */
367 370 uint8_t itx_sync; /* synchronous transaction */
368 371 uint64_t itx_sod; /* record size on disk */
369 372 uint64_t itx_oid; /* object id */
370 373 lr_t itx_lr; /* common part of log record */
371 374 /* followed by type-specific part of lr_xx_t and its immediate data */
372 375 } itx_t;
373 376
374 377 typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
375 378 uint64_t txg);
376 379 typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
377 380 uint64_t txg);
378 381 typedef int zil_replay_func_t();
379 382 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio);
380 383
381 384 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
382 385 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg);
383 386
384 387 extern void zil_init(void);
385 388 extern void zil_fini(void);
386 389
387 390 extern zilog_t *zil_alloc(objset_t *os, zil_header_t *zh_phys);
388 391 extern void zil_free(zilog_t *zilog);
389 392
390 393 extern zilog_t *zil_open(objset_t *os, zil_get_data_t *get_data);
391 394 extern void zil_close(zilog_t *zilog);
392 395
393 396 extern void zil_replay(objset_t *os, void *arg,
394 397 zil_replay_func_t *replay_func[TX_MAX_TYPE]);
395 398 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
396 399 extern void zil_destroy(zilog_t *zilog, boolean_t keep_first);
|
↓ open down ↓ |
347 lines elided |
↑ open up ↑ |
397 400 extern void zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
398 401 extern void zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx);
399 402
400 403 extern itx_t *zil_itx_create(uint64_t txtype, size_t lrsize);
401 404 extern void zil_itx_destroy(itx_t *itx);
402 405 extern void zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
403 406
404 407 extern void zil_commit(zilog_t *zilog, uint64_t oid);
405 408
406 409 extern int zil_vdev_offline(const char *osname, void *txarg);
407 -extern int zil_claim(const char *osname, void *txarg);
408 -extern int zil_check_log_chain(const char *osname, void *txarg);
410 +extern int zil_claim(struct dsl_pool *dp,
411 + struct dsl_dataset *ds, void *txarg);
412 +extern int zil_check_log_chain(struct dsl_pool *dp,
413 + struct dsl_dataset *ds, void *tx);
409 414 extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx);
410 415 extern void zil_clean(zilog_t *zilog, uint64_t synced_txg);
411 416
412 417 extern int zil_suspend(const char *osname, void **cookiep);
413 418 extern void zil_resume(void *cookie);
414 419
415 420 extern void zil_add_block(zilog_t *zilog, const blkptr_t *bp);
416 421 extern int zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
417 422
418 423 extern void zil_set_sync(zilog_t *zilog, uint64_t syncval);
419 424
420 425 extern void zil_set_logbias(zilog_t *zilog, uint64_t slogval);
421 426
422 427 extern int zil_replay_disable;
423 428
424 429 #ifdef __cplusplus
425 430 }
426 431 #endif
427 432
428 433 #endif /* _SYS_ZIL_H */
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX