1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18 #ifndef __XFS_LOG_PRIV_H__
19 #define __XFS_LOG_PRIV_H__
20
21 struct xfs_buf;
22 struct log;
23 struct xlog_ticket;
24 struct xfs_mount;
25
26 /*
27 * Macros, structures, prototypes for internal log manager use.
28 */
29
30 #define XLOG_MIN_ICLOGS 2
31 #define XLOG_MAX_ICLOGS 8
32 #define XLOG_HEADER_MAGIC_NUM 0xFEEDbabe /* Invalid cycle number */
33 #define XLOG_VERSION_1 1
34 #define XLOG_VERSION_2 2 /* Large IClogs, Log sunit */
35 #define XLOG_VERSION_OKBITS (XLOG_VERSION_1 | XLOG_VERSION_2)
36 #define XLOG_MIN_RECORD_BSIZE (16*1024) /* eventually 32k */
37 #define XLOG_BIG_RECORD_BSIZE (32*1024) /* 32k buffers */
38 #define XLOG_MAX_RECORD_BSIZE (256*1024)
39 #define XLOG_HEADER_CYCLE_SIZE (32*1024) /* cycle data in header */
40 #define XLOG_MIN_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
41 #define XLOG_BIG_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
42 #define XLOG_MAX_RECORD_BSHIFT 18 /* 256k == 1 << 18 */
43 #define XLOG_BTOLSUNIT(log, b) (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
44 (log)->l_mp->m_sb.sb_logsunit)
45 #define XLOG_LSUNITTOB(log, su) ((su) * (log)->l_mp->m_sb.sb_logsunit)
46
47 #define XLOG_HEADER_SIZE 512
48
49 #define XLOG_REC_SHIFT(log) \
50 BTOBB(1 << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
51 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
52 #define XLOG_TOTAL_REC_SHIFT(log) \
53 BTOBB(XLOG_MAX_ICLOGS << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
54 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
55
xlog_assign_lsn(uint cycle,uint block)56 static inline xfs_lsn_t xlog_assign_lsn(uint cycle, uint block)
57 {
58 return ((xfs_lsn_t)cycle << 32) | block;
59 }
60
xlog_get_cycle(char * ptr)61 static inline uint xlog_get_cycle(char *ptr)
62 {
63 if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
64 return be32_to_cpu(*((__be32 *)ptr + 1));
65 else
66 return be32_to_cpu(*(__be32 *)ptr);
67 }
68
69 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
70
71 #ifdef __KERNEL__
72
73 /*
74 * get client id from packed copy.
75 *
76 * this hack is here because the xlog_pack code copies four bytes
77 * of xlog_op_header containing the fields oh_clientid, oh_flags
78 * and oh_res2 into the packed copy.
79 *
80 * later on this four byte chunk is treated as an int and the
81 * client id is pulled out.
82 *
83 * this has endian issues, of course.
84 */
xlog_get_client_id(__be32 i)85 static inline uint xlog_get_client_id(__be32 i)
86 {
87 return be32_to_cpu(i) >> 24;
88 }
89
90 /*
91 * In core log state
92 */
93 #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
94 #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
95 #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
96 #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
97 #define XLOG_STATE_DO_CALLBACK \
98 0x0010 /* Process callback functions */
99 #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
100 #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
101 #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
102 #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
103 #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
104 #endif /* __KERNEL__ */
105
106 /*
107 * Flags to log operation header
108 *
109 * The first write of a new transaction will be preceded with a start
110 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
111 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
112 * the remainder of the current active in-core log, it is split up into
113 * multiple regions. Each partial region will be marked with a
114 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
115 *
116 */
117 #define XLOG_START_TRANS 0x01 /* Start a new transaction */
118 #define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
119 #define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
120 #define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
121 #define XLOG_END_TRANS 0x10 /* End a continued transaction */
122 #define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
123
124 #ifdef __KERNEL__
125 /*
126 * Flags to log ticket
127 */
128 #define XLOG_TIC_INITED 0x1 /* has been initialized */
129 #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
130
131 #define XLOG_TIC_FLAGS \
132 { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \
133 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
134
135 #endif /* __KERNEL__ */
136
137 #define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
138
139 /*
140 * Flags for log structure
141 */
142 #define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
143 #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
144 #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
145 #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
146 shutdown */
147 #define XLOG_TAIL_WARN 0x10 /* log tail verify warning issued */
148
149 #ifdef __KERNEL__
150 /*
151 * Below are states for covering allocation transactions.
152 * By covering, we mean changing the h_tail_lsn in the last on-disk
153 * log write such that no allocation transactions will be re-done during
154 * recovery after a system crash. Recovery starts at the last on-disk
155 * log write.
156 *
157 * These states are used to insert dummy log entries to cover
158 * space allocation transactions which can undo non-transactional changes
159 * after a crash. Writes to a file with space
160 * already allocated do not result in any transactions. Allocations
161 * might include space beyond the EOF. So if we just push the EOF a
162 * little, the last transaction for the file could contain the wrong
163 * size. If there is no file system activity, after an allocation
164 * transaction, and the system crashes, the allocation transaction
165 * will get replayed and the file will be truncated. This could
166 * be hours/days/... after the allocation occurred.
167 *
168 * The fix for this is to do two dummy transactions when the
169 * system is idle. We need two dummy transaction because the h_tail_lsn
170 * in the log record header needs to point beyond the last possible
171 * non-dummy transaction. The first dummy changes the h_tail_lsn to
172 * the first transaction before the dummy. The second dummy causes
173 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
174 *
175 * These dummy transactions get committed when everything
176 * is idle (after there has been some activity).
177 *
178 * There are 5 states used to control this.
179 *
180 * IDLE -- no logging has been done on the file system or
181 * we are done covering previous transactions.
182 * NEED -- logging has occurred and we need a dummy transaction
183 * when the log becomes idle.
184 * DONE -- we were in the NEED state and have committed a dummy
185 * transaction.
186 * NEED2 -- we detected that a dummy transaction has gone to the
187 * on disk log with no other transactions.
188 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
189 *
190 * There are two places where we switch states:
191 *
192 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
193 * We commit the dummy transaction and switch to DONE or DONE2,
194 * respectively. In all other states, we don't do anything.
195 *
196 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
197 *
198 * No matter what state we are in, if this isn't the dummy
199 * transaction going out, the next state is NEED.
200 * So, if we aren't in the DONE or DONE2 states, the next state
201 * is NEED. We can't be finishing a write of the dummy record
202 * unless it was committed and the state switched to DONE or DONE2.
203 *
204 * If we are in the DONE state and this was a write of the
205 * dummy transaction, we move to NEED2.
206 *
207 * If we are in the DONE2 state and this was a write of the
208 * dummy transaction, we move to IDLE.
209 *
210 *
211 * Writing only one dummy transaction can get appended to
212 * one file space allocation. When this happens, the log recovery
213 * code replays the space allocation and a file could be truncated.
214 * This is why we have the NEED2 and DONE2 states before going idle.
215 */
216
217 #define XLOG_STATE_COVER_IDLE 0
218 #define XLOG_STATE_COVER_NEED 1
219 #define XLOG_STATE_COVER_DONE 2
220 #define XLOG_STATE_COVER_NEED2 3
221 #define XLOG_STATE_COVER_DONE2 4
222
223 #define XLOG_COVER_OPS 5
224
225
226 /* Ticket reservation region accounting */
227 #define XLOG_TIC_LEN_MAX 15
228
229 /*
230 * Reservation region
231 * As would be stored in xfs_log_iovec but without the i_addr which
232 * we don't care about.
233 */
234 typedef struct xlog_res {
235 uint r_len; /* region length :4 */
236 uint r_type; /* region's transaction type :4 */
237 } xlog_res_t;
238
239 typedef struct xlog_ticket {
240 wait_queue_head_t t_wait; /* ticket wait queue */
241 struct list_head t_queue; /* reserve/write queue */
242 xlog_tid_t t_tid; /* transaction identifier : 4 */
243 atomic_t t_ref; /* ticket reference count : 4 */
244 int t_curr_res; /* current reservation in bytes : 4 */
245 int t_unit_res; /* unit reservation in bytes : 4 */
246 char t_ocnt; /* original count : 1 */
247 char t_cnt; /* current count : 1 */
248 char t_clientid; /* who does this belong to; : 1 */
249 char t_flags; /* properties of reservation : 1 */
250 uint t_trans_type; /* transaction type : 4 */
251
252 /* reservation array fields */
253 uint t_res_num; /* num in array : 4 */
254 uint t_res_num_ophdrs; /* num op hdrs : 4 */
255 uint t_res_arr_sum; /* array sum : 4 */
256 uint t_res_o_flow; /* sum overflow : 4 */
257 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
258 } xlog_ticket_t;
259
260 #endif
261
262
263 typedef struct xlog_op_header {
264 __be32 oh_tid; /* transaction id of operation : 4 b */
265 __be32 oh_len; /* bytes in data region : 4 b */
266 __u8 oh_clientid; /* who sent me this : 1 b */
267 __u8 oh_flags; /* : 1 b */
268 __u16 oh_res2; /* 32 bit align : 2 b */
269 } xlog_op_header_t;
270
271
272 /* valid values for h_fmt */
273 #define XLOG_FMT_UNKNOWN 0
274 #define XLOG_FMT_LINUX_LE 1
275 #define XLOG_FMT_LINUX_BE 2
276 #define XLOG_FMT_IRIX_BE 3
277
278 /* our fmt */
279 #ifdef XFS_NATIVE_HOST
280 #define XLOG_FMT XLOG_FMT_LINUX_BE
281 #else
282 #define XLOG_FMT XLOG_FMT_LINUX_LE
283 #endif
284
285 typedef struct xlog_rec_header {
286 __be32 h_magicno; /* log record (LR) identifier : 4 */
287 __be32 h_cycle; /* write cycle of log : 4 */
288 __be32 h_version; /* LR version : 4 */
289 __be32 h_len; /* len in bytes; should be 64-bit aligned: 4 */
290 __be64 h_lsn; /* lsn of this LR : 8 */
291 __be64 h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
292 __be32 h_chksum; /* may not be used; non-zero if used : 4 */
293 __be32 h_prev_block; /* block number to previous LR : 4 */
294 __be32 h_num_logops; /* number of log operations in this LR : 4 */
295 __be32 h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
296 /* new fields */
297 __be32 h_fmt; /* format of log record : 4 */
298 uuid_t h_fs_uuid; /* uuid of FS : 16 */
299 __be32 h_size; /* iclog size : 4 */
300 } xlog_rec_header_t;
301
302 typedef struct xlog_rec_ext_header {
303 __be32 xh_cycle; /* write cycle of log : 4 */
304 __be32 xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
305 } xlog_rec_ext_header_t;
306
307 #ifdef __KERNEL__
308
309 /*
310 * Quite misnamed, because this union lays out the actual on-disk log buffer.
311 */
312 typedef union xlog_in_core2 {
313 xlog_rec_header_t hic_header;
314 xlog_rec_ext_header_t hic_xheader;
315 char hic_sector[XLOG_HEADER_SIZE];
316 } xlog_in_core_2_t;
317
318 /*
319 * - A log record header is 512 bytes. There is plenty of room to grow the
320 * xlog_rec_header_t into the reserved space.
321 * - ic_data follows, so a write to disk can start at the beginning of
322 * the iclog.
323 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
324 * - ic_next is the pointer to the next iclog in the ring.
325 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
326 * - ic_log is a pointer back to the global log structure.
327 * - ic_callback is a linked list of callback function/argument pairs to be
328 * called after an iclog finishes writing.
329 * - ic_size is the full size of the header plus data.
330 * - ic_offset is the current number of bytes written to in this iclog.
331 * - ic_refcnt is bumped when someone is writing to the log.
332 * - ic_state is the state of the iclog.
333 *
334 * Because of cacheline contention on large machines, we need to separate
335 * various resources onto different cachelines. To start with, make the
336 * structure cacheline aligned. The following fields can be contended on
337 * by independent processes:
338 *
339 * - ic_callback_*
340 * - ic_refcnt
341 * - fields protected by the global l_icloglock
342 *
343 * so we need to ensure that these fields are located in separate cachelines.
344 * We'll put all the read-only and l_icloglock fields in the first cacheline,
345 * and move everything else out to subsequent cachelines.
346 */
347 typedef struct xlog_in_core {
348 wait_queue_head_t ic_force_wait;
349 wait_queue_head_t ic_write_wait;
350 struct xlog_in_core *ic_next;
351 struct xlog_in_core *ic_prev;
352 struct xfs_buf *ic_bp;
353 struct log *ic_log;
354 int ic_size;
355 int ic_offset;
356 int ic_bwritecnt;
357 unsigned short ic_state;
358 char *ic_datap; /* pointer to iclog data */
359
360 /* Callback structures need their own cacheline */
361 spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
362 xfs_log_callback_t *ic_callback;
363 xfs_log_callback_t **ic_callback_tail;
364
365 /* reference counts need their own cacheline */
366 atomic_t ic_refcnt ____cacheline_aligned_in_smp;
367 xlog_in_core_2_t *ic_data;
368 #define ic_header ic_data->hic_header
369 } xlog_in_core_t;
370
371 /*
372 * The CIL context is used to aggregate per-transaction details as well be
373 * passed to the iclog for checkpoint post-commit processing. After being
374 * passed to the iclog, another context needs to be allocated for tracking the
375 * next set of transactions to be aggregated into a checkpoint.
376 */
377 struct xfs_cil;
378
379 struct xfs_cil_ctx {
380 struct xfs_cil *cil;
381 xfs_lsn_t sequence; /* chkpt sequence # */
382 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
383 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
384 struct xlog_ticket *ticket; /* chkpt ticket */
385 int nvecs; /* number of regions */
386 int space_used; /* aggregate size of regions */
387 struct list_head busy_extents; /* busy extents in chkpt */
388 struct xfs_log_vec *lv_chain; /* logvecs being pushed */
389 xfs_log_callback_t log_cb; /* completion callback hook. */
390 struct list_head committing; /* ctx committing list */
391 };
392
393 /*
394 * Committed Item List structure
395 *
396 * This structure is used to track log items that have been committed but not
397 * yet written into the log. It is used only when the delayed logging mount
398 * option is enabled.
399 *
400 * This structure tracks the list of committing checkpoint contexts so
401 * we can avoid the problem of having to hold out new transactions during a
402 * flush until we have a the commit record LSN of the checkpoint. We can
403 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
404 * sequence match and extract the commit LSN directly from there. If the
405 * checkpoint is still in the process of committing, we can block waiting for
406 * the commit LSN to be determined as well. This should make synchronous
407 * operations almost as efficient as the old logging methods.
408 */
409 struct xfs_cil {
410 struct log *xc_log;
411 struct list_head xc_cil;
412 spinlock_t xc_cil_lock;
413 struct xfs_cil_ctx *xc_ctx;
414 struct rw_semaphore xc_ctx_lock;
415 struct list_head xc_committing;
416 wait_queue_head_t xc_commit_wait;
417 xfs_lsn_t xc_current_sequence;
418 };
419
420 /*
421 * The amount of log space we allow the CIL to aggregate is difficult to size.
422 * Whatever we choose, we have to make sure we can get a reservation for the
423 * log space effectively, that it is large enough to capture sufficient
424 * relogging to reduce log buffer IO significantly, but it is not too large for
425 * the log or induces too much latency when writing out through the iclogs. We
426 * track both space consumed and the number of vectors in the checkpoint
427 * context, so we need to decide which to use for limiting.
428 *
429 * Every log buffer we write out during a push needs a header reserved, which
430 * is at least one sector and more for v2 logs. Hence we need a reservation of
431 * at least 512 bytes per 32k of log space just for the LR headers. That means
432 * 16KB of reservation per megabyte of delayed logging space we will consume,
433 * plus various headers. The number of headers will vary based on the num of
434 * io vectors, so limiting on a specific number of vectors is going to result
435 * in transactions of varying size. IOWs, it is more consistent to track and
436 * limit space consumed in the log rather than by the number of objects being
437 * logged in order to prevent checkpoint ticket overruns.
438 *
439 * Further, use of static reservations through the log grant mechanism is
440 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
441 * grant) and a significant deadlock potential because regranting write space
442 * can block on log pushes. Hence if we have to regrant log space during a log
443 * push, we can deadlock.
444 *
445 * However, we can avoid this by use of a dynamic "reservation stealing"
446 * technique during transaction commit whereby unused reservation space in the
447 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
448 * space needed by the checkpoint transaction. This means that we never need to
449 * specifically reserve space for the CIL checkpoint transaction, nor do we
450 * need to regrant space once the checkpoint completes. This also means the
451 * checkpoint transaction ticket is specific to the checkpoint context, rather
452 * than the CIL itself.
453 *
454 * With dynamic reservations, we can effectively make up arbitrary limits for
455 * the checkpoint size so long as they don't violate any other size rules.
456 * Recovery imposes a rule that no transaction exceed half the log, so we are
457 * limited by that. Furthermore, the log transaction reservation subsystem
458 * tries to keep 25% of the log free, so we need to keep below that limit or we
459 * risk running out of free log space to start any new transactions.
460 *
461 * In order to keep background CIL push efficient, we will set a lower
462 * threshold at which background pushing is attempted without blocking current
463 * transaction commits. A separate, higher bound defines when CIL pushes are
464 * enforced to ensure we stay within our maximum checkpoint size bounds.
465 * threshold, yet give us plenty of space for aggregation on large logs.
466 */
467 #define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3)
468 #define XLOG_CIL_HARD_SPACE_LIMIT(log) (3 * (log->l_logsize >> 4))
469
470 /*
471 * The reservation head lsn is not made up of a cycle number and block number.
472 * Instead, it uses a cycle number and byte number. Logs don't expect to
473 * overflow 31 bits worth of byte offset, so using a byte number will mean
474 * that round off problems won't occur when releasing partial reservations.
475 */
476 typedef struct log {
477 /* The following fields don't need locking */
478 struct xfs_mount *l_mp; /* mount point */
479 struct xfs_ail *l_ailp; /* AIL log is working with */
480 struct xfs_cil *l_cilp; /* CIL log is working with */
481 struct xfs_buf *l_xbuf; /* extra buffer for log
482 * wrapping */
483 struct xfs_buftarg *l_targ; /* buftarg of log */
484 uint l_flags;
485 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
486 struct list_head *l_buf_cancel_table;
487 int l_iclog_hsize; /* size of iclog header */
488 int l_iclog_heads; /* # of iclog header sectors */
489 uint l_sectBBsize; /* sector size in BBs (2^n) */
490 int l_iclog_size; /* size of log in bytes */
491 int l_iclog_size_log; /* log power size of log */
492 int l_iclog_bufs; /* number of iclog buffers */
493 xfs_daddr_t l_logBBstart; /* start block of log */
494 int l_logsize; /* size of log in bytes */
495 int l_logBBsize; /* size of log in BB chunks */
496
497 /* The following block of fields are changed while holding icloglock */
498 wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp;
499 /* waiting for iclog flush */
500 int l_covered_state;/* state of "covering disk
501 * log entries" */
502 xlog_in_core_t *l_iclog; /* head log queue */
503 spinlock_t l_icloglock; /* grab to change iclog state */
504 int l_curr_cycle; /* Cycle number of log writes */
505 int l_prev_cycle; /* Cycle number before last
506 * block increment */
507 int l_curr_block; /* current logical log block */
508 int l_prev_block; /* previous logical log block */
509
510 /*
511 * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
512 * read without needing to hold specific locks. To avoid operations
513 * contending with other hot objects, place each of them on a separate
514 * cacheline.
515 */
516 /* lsn of last LR on disk */
517 atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp;
518 /* lsn of 1st LR with unflushed * buffers */
519 atomic64_t l_tail_lsn ____cacheline_aligned_in_smp;
520
521 /*
522 * ticket grant locks, queues and accounting have their own cachlines
523 * as these are quite hot and can be operated on concurrently.
524 */
525 spinlock_t l_grant_reserve_lock ____cacheline_aligned_in_smp;
526 struct list_head l_reserveq;
527 atomic64_t l_grant_reserve_head;
528
529 spinlock_t l_grant_write_lock ____cacheline_aligned_in_smp;
530 struct list_head l_writeq;
531 atomic64_t l_grant_write_head;
532
533 /* The following field are used for debugging; need to hold icloglock */
534 #ifdef DEBUG
535 char *l_iclog_bak[XLOG_MAX_ICLOGS];
536 #endif
537
538 } xlog_t;
539
540 #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
541 ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE))
542
543 #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
544
545 /* common routines */
546 extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp);
547 extern int xlog_recover(xlog_t *log);
548 extern int xlog_recover_finish(xlog_t *log);
549 extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
550
551 extern kmem_zone_t *xfs_log_ticket_zone;
552 struct xlog_ticket *xlog_ticket_alloc(struct log *log, int unit_bytes,
553 int count, char client, uint xflags,
554 int alloc_flags);
555
556
557 static inline void
xlog_write_adv_cnt(void ** ptr,int * len,int * off,size_t bytes)558 xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
559 {
560 *ptr += bytes;
561 *len -= bytes;
562 *off += bytes;
563 }
564
565 void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
566 int xlog_write(struct log *log, struct xfs_log_vec *log_vector,
567 struct xlog_ticket *tic, xfs_lsn_t *start_lsn,
568 xlog_in_core_t **commit_iclog, uint flags);
569
570 /*
571 * When we crack an atomic LSN, we sample it first so that the value will not
572 * change while we are cracking it into the component values. This means we
573 * will always get consistent component values to work from. This should always
574 * be used to sample and crack LSNs that are stored and updated in atomic
575 * variables.
576 */
577 static inline void
xlog_crack_atomic_lsn(atomic64_t * lsn,uint * cycle,uint * block)578 xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
579 {
580 xfs_lsn_t val = atomic64_read(lsn);
581
582 *cycle = CYCLE_LSN(val);
583 *block = BLOCK_LSN(val);
584 }
585
586 /*
587 * Calculate and assign a value to an atomic LSN variable from component pieces.
588 */
589 static inline void
xlog_assign_atomic_lsn(atomic64_t * lsn,uint cycle,uint block)590 xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
591 {
592 atomic64_set(lsn, xlog_assign_lsn(cycle, block));
593 }
594
595 /*
596 * When we crack the grant head, we sample it first so that the value will not
597 * change while we are cracking it into the component values. This means we
598 * will always get consistent component values to work from.
599 */
600 static inline void
xlog_crack_grant_head_val(int64_t val,int * cycle,int * space)601 xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
602 {
603 *cycle = val >> 32;
604 *space = val & 0xffffffff;
605 }
606
607 static inline void
xlog_crack_grant_head(atomic64_t * head,int * cycle,int * space)608 xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
609 {
610 xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
611 }
612
613 static inline int64_t
xlog_assign_grant_head_val(int cycle,int space)614 xlog_assign_grant_head_val(int cycle, int space)
615 {
616 return ((int64_t)cycle << 32) | space;
617 }
618
619 static inline void
xlog_assign_grant_head(atomic64_t * head,int cycle,int space)620 xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
621 {
622 atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
623 }
624
625 /*
626 * Committed Item List interfaces
627 */
628 int xlog_cil_init(struct log *log);
629 void xlog_cil_init_post_recovery(struct log *log);
630 void xlog_cil_destroy(struct log *log);
631
632 /*
633 * CIL force routines
634 */
635 xfs_lsn_t xlog_cil_force_lsn(struct log *log, xfs_lsn_t sequence);
636
637 static inline void
xlog_cil_force(struct log * log)638 xlog_cil_force(struct log *log)
639 {
640 xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
641 }
642
643 /*
644 * Unmount record type is used as a pseudo transaction type for the ticket.
645 * It's value must be outside the range of XFS_TRANS_* values.
646 */
647 #define XLOG_UNMOUNT_REC_TYPE (-1U)
648
649 /*
650 * Wrapper function for waiting on a wait queue serialised against wakeups
651 * by a spinlock. This matches the semantics of all the wait queues used in the
652 * log code.
653 */
xlog_wait(wait_queue_head_t * wq,spinlock_t * lock)654 static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
655 {
656 DECLARE_WAITQUEUE(wait, current);
657
658 add_wait_queue_exclusive(wq, &wait);
659 __set_current_state(TASK_UNINTERRUPTIBLE);
660 spin_unlock(lock);
661 schedule();
662 remove_wait_queue(wq, &wait);
663 }
664 #endif /* __KERNEL__ */
665
666 #endif /* __XFS_LOG_PRIV_H__ */
667