2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
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.
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.
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
18 #ifndef __XFS_LOG_PRIV_H__
19 #define __XFS_LOG_PRIV_H__
25 struct xfs_buf_cancel
;
29 * Macros, structures, prototypes for internal log manager use.
32 #define XLOG_MIN_ICLOGS 2
33 #define XLOG_MED_ICLOGS 4
34 #define XLOG_MAX_ICLOGS 8
35 #define XLOG_CALLBACK_SIZE 10
36 #define XLOG_HEADER_MAGIC_NUM 0xFEEDbabe /* Invalid cycle number */
37 #define XLOG_VERSION_1 1
38 #define XLOG_VERSION_2 2 /* Large IClogs, Log sunit */
39 #define XLOG_VERSION_OKBITS (XLOG_VERSION_1 | XLOG_VERSION_2)
40 #define XLOG_RECORD_BSIZE (16*1024) /* eventually 32k */
41 #define XLOG_BIG_RECORD_BSIZE (32*1024) /* 32k buffers */
42 #define XLOG_MAX_RECORD_BSIZE (256*1024)
43 #define XLOG_HEADER_CYCLE_SIZE (32*1024) /* cycle data in header */
44 #define XLOG_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
45 #define XLOG_BIG_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
46 #define XLOG_MAX_RECORD_BSHIFT 18 /* 256k == 1 << 18 */
47 #define XLOG_BTOLSUNIT(log, b) (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
48 (log)->l_mp->m_sb.sb_logsunit)
49 #define XLOG_LSUNITTOB(log, su) ((su) * (log)->l_mp->m_sb.sb_logsunit)
51 #define XLOG_HEADER_SIZE 512
53 #define XLOG_REC_SHIFT(log) \
54 BTOBB(1 << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
55 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
56 #define XLOG_TOTAL_REC_SHIFT(log) \
57 BTOBB(XLOG_MAX_ICLOGS << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
58 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
64 #define ASSIGN_ANY_LSN_HOST(lsn,cycle,block) \
66 (lsn) = ((xfs_lsn_t)(cycle)<<32)|(block); \
68 #define ASSIGN_ANY_LSN_DISK(lsn,cycle,block) \
70 INT_SET(((uint *)&(lsn))[0], ARCH_CONVERT, (cycle)); \
71 INT_SET(((uint *)&(lsn))[1], ARCH_CONVERT, (block)); \
73 #define ASSIGN_LSN(lsn,log) \
74 ASSIGN_ANY_LSN_DISK(lsn,(log)->l_curr_cycle,(log)->l_curr_block);
76 #define XLOG_SET(f,b) (((f) & (b)) == (b))
78 #define GET_CYCLE(ptr, arch) \
79 (INT_GET(*(uint *)(ptr), arch) == XLOG_HEADER_MAGIC_NUM ? \
80 INT_GET(*((uint *)(ptr)+1), arch) : \
81 INT_GET(*(uint *)(ptr), arch) \
84 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
90 * get client id from packed copy.
92 * this hack is here because the xlog_pack code copies four bytes
93 * of xlog_op_header containing the fields oh_clientid, oh_flags
94 * and oh_res2 into the packed copy.
96 * later on this four byte chunk is treated as an int and the
97 * client id is pulled out.
99 * this has endian issues, of course.
102 #ifndef XFS_NATIVE_HOST
103 #define GET_CLIENT_ID(i,arch) \
106 #define GET_CLIENT_ID(i,arch) \
110 #define GRANT_LOCK(log) mutex_spinlock(&(log)->l_grant_lock)
111 #define GRANT_UNLOCK(log, s) mutex_spinunlock(&(log)->l_grant_lock, s)
112 #define LOG_LOCK(log) mutex_spinlock(&(log)->l_icloglock)
113 #define LOG_UNLOCK(log, s) mutex_spinunlock(&(log)->l_icloglock, s)
115 #define xlog_panic(args...) cmn_err(CE_PANIC, ## args)
116 #define xlog_exit(args...) cmn_err(CE_PANIC, ## args)
117 #define xlog_warn(args...) cmn_err(CE_WARN, ## args)
122 #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
123 #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
124 #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
125 #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
126 #define XLOG_STATE_DO_CALLBACK \
127 0x0010 /* Process callback functions */
128 #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
129 #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
130 #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
131 #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
132 #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
133 #endif /* __KERNEL__ */
136 * Flags to log operation header
138 * The first write of a new transaction will be preceded with a start
139 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
140 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
141 * the remainder of the current active in-core log, it is split up into
142 * multiple regions. Each partial region will be marked with a
143 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
146 #define XLOG_START_TRANS 0x01 /* Start a new transaction */
147 #define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
148 #define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
149 #define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
150 #define XLOG_END_TRANS 0x10 /* End a continued transaction */
151 #define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
152 #define XLOG_SKIP_TRANS (XLOG_COMMIT_TRANS | XLOG_CONTINUE_TRANS | \
153 XLOG_WAS_CONT_TRANS | XLOG_END_TRANS | \
158 * Flags to log ticket
160 #define XLOG_TIC_INITED 0x1 /* has been initialized */
161 #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
162 #define XLOG_TIC_IN_Q 0x4
163 #endif /* __KERNEL__ */
165 #define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
168 * Flags for log structure
170 #define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
171 #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
172 #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
173 #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
175 typedef __uint32_t xlog_tid_t
;
180 * Below are states for covering allocation transactions.
181 * By covering, we mean changing the h_tail_lsn in the last on-disk
182 * log write such that no allocation transactions will be re-done during
183 * recovery after a system crash. Recovery starts at the last on-disk
186 * These states are used to insert dummy log entries to cover
187 * space allocation transactions which can undo non-transactional changes
188 * after a crash. Writes to a file with space
189 * already allocated do not result in any transactions. Allocations
190 * might include space beyond the EOF. So if we just push the EOF a
191 * little, the last transaction for the file could contain the wrong
192 * size. If there is no file system activity, after an allocation
193 * transaction, and the system crashes, the allocation transaction
194 * will get replayed and the file will be truncated. This could
195 * be hours/days/... after the allocation occurred.
197 * The fix for this is to do two dummy transactions when the
198 * system is idle. We need two dummy transaction because the h_tail_lsn
199 * in the log record header needs to point beyond the last possible
200 * non-dummy transaction. The first dummy changes the h_tail_lsn to
201 * the first transaction before the dummy. The second dummy causes
202 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
204 * These dummy transactions get committed when everything
205 * is idle (after there has been some activity).
207 * There are 5 states used to control this.
209 * IDLE -- no logging has been done on the file system or
210 * we are done covering previous transactions.
211 * NEED -- logging has occurred and we need a dummy transaction
212 * when the log becomes idle.
213 * DONE -- we were in the NEED state and have committed a dummy
215 * NEED2 -- we detected that a dummy transaction has gone to the
216 * on disk log with no other transactions.
217 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
219 * There are two places where we switch states:
221 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
222 * We commit the dummy transaction and switch to DONE or DONE2,
223 * respectively. In all other states, we don't do anything.
225 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
227 * No matter what state we are in, if this isn't the dummy
228 * transaction going out, the next state is NEED.
229 * So, if we aren't in the DONE or DONE2 states, the next state
230 * is NEED. We can't be finishing a write of the dummy record
231 * unless it was committed and the state switched to DONE or DONE2.
233 * If we are in the DONE state and this was a write of the
234 * dummy transaction, we move to NEED2.
236 * If we are in the DONE2 state and this was a write of the
237 * dummy transaction, we move to IDLE.
240 * Writing only one dummy transaction can get appended to
241 * one file space allocation. When this happens, the log recovery
242 * code replays the space allocation and a file could be truncated.
243 * This is why we have the NEED2 and DONE2 states before going idle.
246 #define XLOG_STATE_COVER_IDLE 0
247 #define XLOG_STATE_COVER_NEED 1
248 #define XLOG_STATE_COVER_DONE 2
249 #define XLOG_STATE_COVER_NEED2 3
250 #define XLOG_STATE_COVER_DONE2 4
252 #define XLOG_COVER_OPS 5
255 /* Ticket reservation region accounting */
256 #define XLOG_TIC_LEN_MAX 15
257 #define XLOG_TIC_RESET_RES(t) ((t)->t_res_num = \
258 (t)->t_res_arr_sum = (t)->t_res_num_ophdrs = 0)
259 #define XLOG_TIC_ADD_OPHDR(t) ((t)->t_res_num_ophdrs++)
260 #define XLOG_TIC_ADD_REGION(t, len, type) \
262 if ((t)->t_res_num == XLOG_TIC_LEN_MAX) { \
263 /* add to overflow and start again */ \
264 (t)->t_res_o_flow += (t)->t_res_arr_sum; \
265 (t)->t_res_num = 0; \
266 (t)->t_res_arr_sum = 0; \
268 (t)->t_res_arr[(t)->t_res_num].r_len = (len); \
269 (t)->t_res_arr[(t)->t_res_num].r_type = (type); \
270 (t)->t_res_arr_sum += (len); \
276 * As would be stored in xfs_log_iovec but without the i_addr which
277 * we don't care about.
279 typedef struct xlog_res
{
280 uint r_len
; /* region length :4 */
281 uint r_type
; /* region's transaction type :4 */
284 typedef struct xlog_ticket
{
285 sv_t t_sema
; /* sleep on this semaphore : 20 */
286 struct xlog_ticket
*t_next
; /* :4|8 */
287 struct xlog_ticket
*t_prev
; /* :4|8 */
288 xlog_tid_t t_tid
; /* transaction identifier : 4 */
289 int t_curr_res
; /* current reservation in bytes : 4 */
290 int t_unit_res
; /* unit reservation in bytes : 4 */
291 char t_ocnt
; /* original count : 1 */
292 char t_cnt
; /* current count : 1 */
293 char t_clientid
; /* who does this belong to; : 1 */
294 char t_flags
; /* properties of reservation : 1 */
295 uint t_trans_type
; /* transaction type : 4 */
297 /* reservation array fields */
298 uint t_res_num
; /* num in array : 4 */
299 uint t_res_num_ophdrs
; /* num op hdrs : 4 */
300 uint t_res_arr_sum
; /* array sum : 4 */
301 uint t_res_o_flow
; /* sum overflow : 4 */
302 xlog_res_t t_res_arr
[XLOG_TIC_LEN_MAX
]; /* array of res : 8 * 15 */
308 typedef struct xlog_op_header
{
309 xlog_tid_t oh_tid
; /* transaction id of operation : 4 b */
310 int oh_len
; /* bytes in data region : 4 b */
311 __uint8_t oh_clientid
; /* who sent me this : 1 b */
312 __uint8_t oh_flags
; /* : 1 b */
313 ushort oh_res2
; /* 32 bit align : 2 b */
317 /* valid values for h_fmt */
318 #define XLOG_FMT_UNKNOWN 0
319 #define XLOG_FMT_LINUX_LE 1
320 #define XLOG_FMT_LINUX_BE 2
321 #define XLOG_FMT_IRIX_BE 3
324 #ifdef XFS_NATIVE_HOST
325 #define XLOG_FMT XLOG_FMT_LINUX_BE
327 #define XLOG_FMT XLOG_FMT_LINUX_LE
330 typedef struct xlog_rec_header
{
331 uint h_magicno
; /* log record (LR) identifier : 4 */
332 uint h_cycle
; /* write cycle of log : 4 */
333 int h_version
; /* LR version : 4 */
334 int h_len
; /* len in bytes; should be 64-bit aligned: 4 */
335 xfs_lsn_t h_lsn
; /* lsn of this LR : 8 */
336 xfs_lsn_t h_tail_lsn
; /* lsn of 1st LR w/ buffers not committed: 8 */
337 uint h_chksum
; /* may not be used; non-zero if used : 4 */
338 int h_prev_block
; /* block number to previous LR : 4 */
339 int h_num_logops
; /* number of log operations in this LR : 4 */
340 uint h_cycle_data
[XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
];
342 int h_fmt
; /* format of log record : 4 */
343 uuid_t h_fs_uuid
; /* uuid of FS : 16 */
344 int h_size
; /* iclog size : 4 */
347 typedef struct xlog_rec_ext_header
{
348 uint xh_cycle
; /* write cycle of log : 4 */
349 uint xh_cycle_data
[XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
]; /* : 256 */
350 } xlog_rec_ext_header_t
;
354 * - A log record header is 512 bytes. There is plenty of room to grow the
355 * xlog_rec_header_t into the reserved space.
356 * - ic_data follows, so a write to disk can start at the beginning of
358 * - ic_forcesema is used to implement synchronous forcing of the iclog to disk.
359 * - ic_next is the pointer to the next iclog in the ring.
360 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
361 * - ic_log is a pointer back to the global log structure.
362 * - ic_callback is a linked list of callback function/argument pairs to be
363 * called after an iclog finishes writing.
364 * - ic_size is the full size of the header plus data.
365 * - ic_offset is the current number of bytes written to in this iclog.
366 * - ic_refcnt is bumped when someone is writing to the log.
367 * - ic_state is the state of the iclog.
369 typedef struct xlog_iclog_fields
{
372 struct xlog_in_core
*ic_next
;
373 struct xlog_in_core
*ic_prev
;
374 struct xfs_buf
*ic_bp
;
376 xfs_log_callback_t
*ic_callback
;
377 xfs_log_callback_t
**ic_callback_tail
;
379 struct ktrace
*ic_trace
;
386 char *ic_datap
; /* pointer to iclog data */
387 } xlog_iclog_fields_t
;
389 typedef union xlog_in_core2
{
390 xlog_rec_header_t hic_header
;
391 xlog_rec_ext_header_t hic_xheader
;
392 char hic_sector
[XLOG_HEADER_SIZE
];
395 typedef struct xlog_in_core
{
396 xlog_iclog_fields_t hic_fields
;
397 xlog_in_core_2_t
*hic_data
;
401 * Defines to save our code from this glop.
403 #define ic_forcesema hic_fields.ic_forcesema
404 #define ic_writesema hic_fields.ic_writesema
405 #define ic_next hic_fields.ic_next
406 #define ic_prev hic_fields.ic_prev
407 #define ic_bp hic_fields.ic_bp
408 #define ic_log hic_fields.ic_log
409 #define ic_callback hic_fields.ic_callback
410 #define ic_callback_tail hic_fields.ic_callback_tail
411 #define ic_trace hic_fields.ic_trace
412 #define ic_size hic_fields.ic_size
413 #define ic_offset hic_fields.ic_offset
414 #define ic_refcnt hic_fields.ic_refcnt
415 #define ic_bwritecnt hic_fields.ic_bwritecnt
416 #define ic_state hic_fields.ic_state
417 #define ic_datap hic_fields.ic_datap
418 #define ic_header hic_data->hic_header
421 * The reservation head lsn is not made up of a cycle number and block number.
422 * Instead, it uses a cycle number and byte number. Logs don't expect to
423 * overflow 31 bits worth of byte offset, so using a byte number will mean
424 * that round off problems won't occur when releasing partial reservations.
427 /* The following block of fields are changed while holding icloglock */
428 sema_t l_flushsema
; /* iclog flushing semaphore */
429 int l_flushcnt
; /* # of procs waiting on this
431 int l_ticket_cnt
; /* free ticket count */
432 int l_ticket_tcnt
; /* total ticket count */
433 int l_covered_state
;/* state of "covering disk
435 xlog_ticket_t
*l_freelist
; /* free list of tickets */
436 xlog_ticket_t
*l_unmount_free
;/* kmem_free these addresses */
437 xlog_ticket_t
*l_tail
; /* free list of tickets */
438 xlog_in_core_t
*l_iclog
; /* head log queue */
439 lock_t l_icloglock
; /* grab to change iclog state */
440 xfs_lsn_t l_tail_lsn
; /* lsn of 1st LR with unflushed
442 xfs_lsn_t l_last_sync_lsn
;/* lsn of last LR on disk */
443 struct xfs_mount
*l_mp
; /* mount point */
444 struct xfs_buf
*l_xbuf
; /* extra buffer for log
446 struct xfs_buftarg
*l_targ
; /* buftarg of log */
447 xfs_daddr_t l_logBBstart
; /* start block of log */
448 int l_logsize
; /* size of log in bytes */
449 int l_logBBsize
; /* size of log in BB chunks */
450 int l_curr_cycle
; /* Cycle number of log writes */
451 int l_prev_cycle
; /* Cycle number before last
453 int l_curr_block
; /* current logical log block */
454 int l_prev_block
; /* previous logical log block */
455 int l_iclog_size
; /* size of log in bytes */
456 int l_iclog_size_log
; /* log power size of log */
457 int l_iclog_bufs
; /* number of iclog buffers */
459 /* The following field are used for debugging; need to hold icloglock */
460 char *l_iclog_bak
[XLOG_MAX_ICLOGS
];
462 /* The following block of fields are changed while holding grant_lock */
464 xlog_ticket_t
*l_reserve_headq
;
465 xlog_ticket_t
*l_write_headq
;
466 int l_grant_reserve_cycle
;
467 int l_grant_reserve_bytes
;
468 int l_grant_write_cycle
;
469 int l_grant_write_bytes
;
471 /* The following fields don't need locking */
473 struct ktrace
*l_trace
;
474 struct ktrace
*l_grant_trace
;
477 uint l_quotaoffs_flag
; /* XFS_DQ_*, for QUOTAOFFs */
478 struct xfs_buf_cancel
**l_buf_cancel_table
;
479 int l_iclog_hsize
; /* size of iclog header */
480 int l_iclog_heads
; /* # of iclog header sectors */
481 uint l_sectbb_log
; /* log2 of sector size in BBs */
482 uint l_sectbb_mask
; /* sector size (in BBs)
486 #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
489 /* common routines */
490 extern xfs_lsn_t
xlog_assign_tail_lsn(struct xfs_mount
*mp
);
491 extern int xlog_find_tail(xlog_t
*log
,
492 xfs_daddr_t
*head_blk
,
493 xfs_daddr_t
*tail_blk
);
494 extern int xlog_recover(xlog_t
*log
);
495 extern int xlog_recover_finish(xlog_t
*log
, int mfsi_flags
);
496 extern void xlog_pack_data(xlog_t
*log
, xlog_in_core_t
*iclog
, int);
497 extern void xlog_recover_process_iunlinks(xlog_t
*log
);
499 extern struct xfs_buf
*xlog_get_bp(xlog_t
*, int);
500 extern void xlog_put_bp(struct xfs_buf
*);
501 extern int xlog_bread(xlog_t
*, xfs_daddr_t
, int, struct xfs_buf
*);
504 #define XLOG_TRACE_GRAB_FLUSH 1
505 #define XLOG_TRACE_REL_FLUSH 2
506 #define XLOG_TRACE_SLEEP_FLUSH 3
507 #define XLOG_TRACE_WAKE_FLUSH 4
509 #endif /* __KERNEL__ */
511 #endif /* __XFS_LOG_PRIV_H__ */