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_MAX_ICLOGS 8
34 #define XLOG_HEADER_MAGIC_NUM 0xFEEDbabe /* Invalid cycle number */
35 #define XLOG_VERSION_1 1
36 #define XLOG_VERSION_2 2 /* Large IClogs, Log sunit */
37 #define XLOG_VERSION_OKBITS (XLOG_VERSION_1 | XLOG_VERSION_2)
38 #define XLOG_MIN_RECORD_BSIZE (16*1024) /* eventually 32k */
39 #define XLOG_BIG_RECORD_BSIZE (32*1024) /* 32k buffers */
40 #define XLOG_MAX_RECORD_BSIZE (256*1024)
41 #define XLOG_HEADER_CYCLE_SIZE (32*1024) /* cycle data in header */
42 #define XLOG_MIN_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
43 #define XLOG_BIG_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
44 #define XLOG_MAX_RECORD_BSHIFT 18 /* 256k == 1 << 18 */
45 #define XLOG_BTOLSUNIT(log, b) (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
46 (log)->l_mp->m_sb.sb_logsunit)
47 #define XLOG_LSUNITTOB(log, su) ((su) * (log)->l_mp->m_sb.sb_logsunit)
49 #define XLOG_HEADER_SIZE 512
51 #define XLOG_REC_SHIFT(log) \
52 BTOBB(1 << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
53 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
54 #define XLOG_TOTAL_REC_SHIFT(log) \
55 BTOBB(XLOG_MAX_ICLOGS << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
56 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
62 #define ASSIGN_ANY_LSN_HOST(lsn,cycle,block) \
64 (lsn) = ((xfs_lsn_t)(cycle)<<32)|(block); \
66 #define ASSIGN_ANY_LSN_DISK(lsn,cycle,block) \
68 INT_SET(((uint *)&(lsn))[0], ARCH_CONVERT, (cycle)); \
69 INT_SET(((uint *)&(lsn))[1], ARCH_CONVERT, (block)); \
71 #define ASSIGN_LSN(lsn,log) \
72 ASSIGN_ANY_LSN_DISK(lsn,(log)->l_curr_cycle,(log)->l_curr_block);
74 #define XLOG_SET(f,b) (((f) & (b)) == (b))
76 #define GET_CYCLE(ptr, arch) \
77 (INT_GET(*(uint *)(ptr), arch) == XLOG_HEADER_MAGIC_NUM ? \
78 INT_GET(*((uint *)(ptr)+1), arch) : \
79 INT_GET(*(uint *)(ptr), arch) \
82 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
88 * get client id from packed copy.
90 * this hack is here because the xlog_pack code copies four bytes
91 * of xlog_op_header containing the fields oh_clientid, oh_flags
92 * and oh_res2 into the packed copy.
94 * later on this four byte chunk is treated as an int and the
95 * client id is pulled out.
97 * this has endian issues, of course.
100 #ifndef XFS_NATIVE_HOST
101 #define GET_CLIENT_ID(i,arch) \
104 #define GET_CLIENT_ID(i,arch) \
108 #define GRANT_LOCK(log) mutex_spinlock(&(log)->l_grant_lock)
109 #define GRANT_UNLOCK(log, s) mutex_spinunlock(&(log)->l_grant_lock, s)
110 #define LOG_LOCK(log) mutex_spinlock(&(log)->l_icloglock)
111 #define LOG_UNLOCK(log, s) mutex_spinunlock(&(log)->l_icloglock, s)
113 #define xlog_panic(args...) cmn_err(CE_PANIC, ## args)
114 #define xlog_exit(args...) cmn_err(CE_PANIC, ## args)
115 #define xlog_warn(args...) cmn_err(CE_WARN, ## args)
120 #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
121 #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
122 #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
123 #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
124 #define XLOG_STATE_DO_CALLBACK \
125 0x0010 /* Process callback functions */
126 #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
127 #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
128 #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
129 #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
130 #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
131 #endif /* __KERNEL__ */
134 * Flags to log operation header
136 * The first write of a new transaction will be preceded with a start
137 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
138 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
139 * the remainder of the current active in-core log, it is split up into
140 * multiple regions. Each partial region will be marked with a
141 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
144 #define XLOG_START_TRANS 0x01 /* Start a new transaction */
145 #define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
146 #define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
147 #define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
148 #define XLOG_END_TRANS 0x10 /* End a continued transaction */
149 #define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
153 * Flags to log ticket
155 #define XLOG_TIC_INITED 0x1 /* has been initialized */
156 #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
157 #define XLOG_TIC_IN_Q 0x4
158 #endif /* __KERNEL__ */
160 #define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
163 * Flags for log structure
165 #define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
166 #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
167 #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
168 #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
170 typedef __uint32_t xlog_tid_t;
175 * Below are states for covering allocation transactions.
176 * By covering, we mean changing the h_tail_lsn in the last on-disk
177 * log write such that no allocation transactions will be re-done during
178 * recovery after a system crash. Recovery starts at the last on-disk
181 * These states are used to insert dummy log entries to cover
182 * space allocation transactions which can undo non-transactional changes
183 * after a crash. Writes to a file with space
184 * already allocated do not result in any transactions. Allocations
185 * might include space beyond the EOF. So if we just push the EOF a
186 * little, the last transaction for the file could contain the wrong
187 * size. If there is no file system activity, after an allocation
188 * transaction, and the system crashes, the allocation transaction
189 * will get replayed and the file will be truncated. This could
190 * be hours/days/... after the allocation occurred.
192 * The fix for this is to do two dummy transactions when the
193 * system is idle. We need two dummy transaction because the h_tail_lsn
194 * in the log record header needs to point beyond the last possible
195 * non-dummy transaction. The first dummy changes the h_tail_lsn to
196 * the first transaction before the dummy. The second dummy causes
197 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
199 * These dummy transactions get committed when everything
200 * is idle (after there has been some activity).
202 * There are 5 states used to control this.
204 * IDLE -- no logging has been done on the file system or
205 * we are done covering previous transactions.
206 * NEED -- logging has occurred and we need a dummy transaction
207 * when the log becomes idle.
208 * DONE -- we were in the NEED state and have committed a dummy
210 * NEED2 -- we detected that a dummy transaction has gone to the
211 * on disk log with no other transactions.
212 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
214 * There are two places where we switch states:
216 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
217 * We commit the dummy transaction and switch to DONE or DONE2,
218 * respectively. In all other states, we don't do anything.
220 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
222 * No matter what state we are in, if this isn't the dummy
223 * transaction going out, the next state is NEED.
224 * So, if we aren't in the DONE or DONE2 states, the next state
225 * is NEED. We can't be finishing a write of the dummy record
226 * unless it was committed and the state switched to DONE or DONE2.
228 * If we are in the DONE state and this was a write of the
229 * dummy transaction, we move to NEED2.
231 * If we are in the DONE2 state and this was a write of the
232 * dummy transaction, we move to IDLE.
235 * Writing only one dummy transaction can get appended to
236 * one file space allocation. When this happens, the log recovery
237 * code replays the space allocation and a file could be truncated.
238 * This is why we have the NEED2 and DONE2 states before going idle.
241 #define XLOG_STATE_COVER_IDLE 0
242 #define XLOG_STATE_COVER_NEED 1
243 #define XLOG_STATE_COVER_DONE 2
244 #define XLOG_STATE_COVER_NEED2 3
245 #define XLOG_STATE_COVER_DONE2 4
247 #define XLOG_COVER_OPS 5
250 /* Ticket reservation region accounting */
251 #define XLOG_TIC_LEN_MAX 15
255 * As would be stored in xfs_log_iovec but without the i_addr which
256 * we don't care about.
258 typedef struct xlog_res {
259 uint r_len; /* region length :4 */
260 uint r_type; /* region's transaction type :4 */
263 typedef struct xlog_ticket {
264 sv_t t_sema; /* sleep on this semaphore : 20 */
265 struct xlog_ticket *t_next; /* :4|8 */
266 struct xlog_ticket *t_prev; /* :4|8 */
267 xlog_tid_t t_tid; /* transaction identifier : 4 */
268 int t_curr_res; /* current reservation in bytes : 4 */
269 int t_unit_res; /* unit reservation in bytes : 4 */
270 char t_ocnt; /* original count : 1 */
271 char t_cnt; /* current count : 1 */
272 char t_clientid; /* who does this belong to; : 1 */
273 char t_flags; /* properties of reservation : 1 */
274 uint t_trans_type; /* transaction type : 4 */
276 /* reservation array fields */
277 uint t_res_num; /* num in array : 4 */
278 uint t_res_num_ophdrs; /* num op hdrs : 4 */
279 uint t_res_arr_sum; /* array sum : 4 */
280 uint t_res_o_flow; /* sum overflow : 4 */
281 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
287 typedef struct xlog_op_header {
288 xlog_tid_t oh_tid; /* transaction id of operation : 4 b */
289 int oh_len; /* bytes in data region : 4 b */
290 __uint8_t oh_clientid; /* who sent me this : 1 b */
291 __uint8_t oh_flags; /* : 1 b */
292 ushort oh_res2; /* 32 bit align : 2 b */
296 /* valid values for h_fmt */
297 #define XLOG_FMT_UNKNOWN 0
298 #define XLOG_FMT_LINUX_LE 1
299 #define XLOG_FMT_LINUX_BE 2
300 #define XLOG_FMT_IRIX_BE 3
303 #ifdef XFS_NATIVE_HOST
304 #define XLOG_FMT XLOG_FMT_LINUX_BE
306 #define XLOG_FMT XLOG_FMT_LINUX_LE
309 typedef struct xlog_rec_header {
310 uint h_magicno; /* log record (LR) identifier : 4 */
311 uint h_cycle; /* write cycle of log : 4 */
312 int h_version; /* LR version : 4 */
313 int h_len; /* len in bytes; should be 64-bit aligned: 4 */
314 xfs_lsn_t h_lsn; /* lsn of this LR : 8 */
315 xfs_lsn_t h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
316 uint h_chksum; /* may not be used; non-zero if used : 4 */
317 int h_prev_block; /* block number to previous LR : 4 */
318 int h_num_logops; /* number of log operations in this LR : 4 */
319 uint h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
321 int h_fmt; /* format of log record : 4 */
322 uuid_t h_fs_uuid; /* uuid of FS : 16 */
323 int h_size; /* iclog size : 4 */
326 typedef struct xlog_rec_ext_header {
327 uint xh_cycle; /* write cycle of log : 4 */
328 uint xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
329 } xlog_rec_ext_header_t;
333 * - A log record header is 512 bytes. There is plenty of room to grow the
334 * xlog_rec_header_t into the reserved space.
335 * - ic_data follows, so a write to disk can start at the beginning of
337 * - ic_forcesema is used to implement synchronous forcing of the iclog to disk.
338 * - ic_next is the pointer to the next iclog in the ring.
339 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
340 * - ic_log is a pointer back to the global log structure.
341 * - ic_callback is a linked list of callback function/argument pairs to be
342 * called after an iclog finishes writing.
343 * - ic_size is the full size of the header plus data.
344 * - ic_offset is the current number of bytes written to in this iclog.
345 * - ic_refcnt is bumped when someone is writing to the log.
346 * - ic_state is the state of the iclog.
348 typedef struct xlog_iclog_fields {
351 struct xlog_in_core *ic_next;
352 struct xlog_in_core *ic_prev;
353 struct xfs_buf *ic_bp;
355 xfs_log_callback_t *ic_callback;
356 xfs_log_callback_t **ic_callback_tail;
358 struct ktrace *ic_trace;
365 char *ic_datap; /* pointer to iclog data */
366 } xlog_iclog_fields_t;
368 typedef union xlog_in_core2 {
369 xlog_rec_header_t hic_header;
370 xlog_rec_ext_header_t hic_xheader;
371 char hic_sector[XLOG_HEADER_SIZE];
374 typedef struct xlog_in_core {
375 xlog_iclog_fields_t hic_fields;
376 xlog_in_core_2_t *hic_data;
380 * Defines to save our code from this glop.
382 #define ic_forcesema hic_fields.ic_forcesema
383 #define ic_writesema hic_fields.ic_writesema
384 #define ic_next hic_fields.ic_next
385 #define ic_prev hic_fields.ic_prev
386 #define ic_bp hic_fields.ic_bp
387 #define ic_log hic_fields.ic_log
388 #define ic_callback hic_fields.ic_callback
389 #define ic_callback_tail hic_fields.ic_callback_tail
390 #define ic_trace hic_fields.ic_trace
391 #define ic_size hic_fields.ic_size
392 #define ic_offset hic_fields.ic_offset
393 #define ic_refcnt hic_fields.ic_refcnt
394 #define ic_bwritecnt hic_fields.ic_bwritecnt
395 #define ic_state hic_fields.ic_state
396 #define ic_datap hic_fields.ic_datap
397 #define ic_header hic_data->hic_header
400 * The reservation head lsn is not made up of a cycle number and block number.
401 * Instead, it uses a cycle number and byte number. Logs don't expect to
402 * overflow 31 bits worth of byte offset, so using a byte number will mean
403 * that round off problems won't occur when releasing partial reservations.
406 /* The following block of fields are changed while holding icloglock */
407 sema_t l_flushsema; /* iclog flushing semaphore */
408 int l_flushcnt; /* # of procs waiting on this
410 int l_ticket_cnt; /* free ticket count */
411 int l_ticket_tcnt; /* total ticket count */
412 int l_covered_state;/* state of "covering disk
414 xlog_ticket_t *l_freelist; /* free list of tickets */
415 xlog_ticket_t *l_unmount_free;/* kmem_free these addresses */
416 xlog_ticket_t *l_tail; /* free list of tickets */
417 xlog_in_core_t *l_iclog; /* head log queue */
418 lock_t l_icloglock; /* grab to change iclog state */
419 xfs_lsn_t l_tail_lsn; /* lsn of 1st LR with unflushed
421 xfs_lsn_t l_last_sync_lsn;/* lsn of last LR on disk */
422 struct xfs_mount *l_mp; /* mount point */
423 struct xfs_buf *l_xbuf; /* extra buffer for log
425 struct xfs_buftarg *l_targ; /* buftarg of log */
426 xfs_daddr_t l_logBBstart; /* start block of log */
427 int l_logsize; /* size of log in bytes */
428 int l_logBBsize; /* size of log in BB chunks */
429 int l_curr_cycle; /* Cycle number of log writes */
430 int l_prev_cycle; /* Cycle number before last
432 int l_curr_block; /* current logical log block */
433 int l_prev_block; /* previous logical log block */
434 int l_iclog_size; /* size of log in bytes */
435 int l_iclog_size_log; /* log power size of log */
436 int l_iclog_bufs; /* number of iclog buffers */
438 /* The following field are used for debugging; need to hold icloglock */
439 char *l_iclog_bak[XLOG_MAX_ICLOGS];
441 /* The following block of fields are changed while holding grant_lock */
443 xlog_ticket_t *l_reserve_headq;
444 xlog_ticket_t *l_write_headq;
445 int l_grant_reserve_cycle;
446 int l_grant_reserve_bytes;
447 int l_grant_write_cycle;
448 int l_grant_write_bytes;
450 /* The following fields don't need locking */
452 struct ktrace *l_trace;
453 struct ktrace *l_grant_trace;
456 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
457 struct xfs_buf_cancel **l_buf_cancel_table;
458 int l_iclog_hsize; /* size of iclog header */
459 int l_iclog_heads; /* # of iclog header sectors */
460 uint l_sectbb_log; /* log2 of sector size in BBs */
461 uint l_sectbb_mask; /* sector size (in BBs)
465 #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
468 /* common routines */
469 extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp);
470 extern int xlog_find_tail(xlog_t *log,
471 xfs_daddr_t *head_blk,
472 xfs_daddr_t *tail_blk);
473 extern int xlog_recover(xlog_t *log);
474 extern int xlog_recover_finish(xlog_t *log, int mfsi_flags);
475 extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
476 extern void xlog_recover_process_iunlinks(xlog_t *log);
478 extern struct xfs_buf *xlog_get_bp(xlog_t *, int);
479 extern void xlog_put_bp(struct xfs_buf *);
480 extern int xlog_bread(xlog_t *, xfs_daddr_t, int, struct xfs_buf *);
483 #define XLOG_TRACE_GRAB_FLUSH 1
484 #define XLOG_TRACE_REL_FLUSH 2
485 #define XLOG_TRACE_SLEEP_FLUSH 3
486 #define XLOG_TRACE_WAKE_FLUSH 4
489 * Unmount record type is used as a pseudo transaction type for the ticket.
490 * It's value must be outside the range of XFS_TRANS_* values.
492 #define XLOG_UNMOUNT_REC_TYPE (-1U)
494 #endif /* __KERNEL__ */
496 #endif /* __XFS_LOG_PRIV_H__ */