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))
59 static inline xfs_lsn_t xlog_assign_lsn(uint cycle, uint block)
61 return ((xfs_lsn_t)cycle << 32) | block;
64 static inline uint xlog_get_cycle(char *ptr)
66 if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
67 return be32_to_cpu(*((__be32 *)ptr + 1));
69 return be32_to_cpu(*(__be32 *)ptr);
72 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
77 * get client id from packed copy.
79 * this hack is here because the xlog_pack code copies four bytes
80 * of xlog_op_header containing the fields oh_clientid, oh_flags
81 * and oh_res2 into the packed copy.
83 * later on this four byte chunk is treated as an int and the
84 * client id is pulled out.
86 * this has endian issues, of course.
88 static inline uint xlog_get_client_id(__be32 i)
90 return be32_to_cpu(i) >> 24;
93 #define xlog_panic(args...) cmn_err(CE_PANIC, ## args)
94 #define xlog_exit(args...) cmn_err(CE_PANIC, ## args)
95 #define xlog_warn(args...) cmn_err(CE_WARN, ## args)
100 #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
101 #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
102 #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
103 #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
104 #define XLOG_STATE_DO_CALLBACK \
105 0x0010 /* Process callback functions */
106 #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
107 #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
108 #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
109 #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
110 #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
111 #endif /* __KERNEL__ */
114 * Flags to log operation header
116 * The first write of a new transaction will be preceded with a start
117 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
118 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
119 * the remainder of the current active in-core log, it is split up into
120 * multiple regions. Each partial region will be marked with a
121 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
124 #define XLOG_START_TRANS 0x01 /* Start a new transaction */
125 #define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
126 #define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
127 #define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
128 #define XLOG_END_TRANS 0x10 /* End a continued transaction */
129 #define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
133 * Flags to log ticket
135 #define XLOG_TIC_INITED 0x1 /* has been initialized */
136 #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
137 #define XLOG_TIC_IN_Q 0x4
138 #endif /* __KERNEL__ */
140 #define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
143 * Flags for log structure
145 #define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
146 #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
147 #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
148 #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
150 typedef __uint32_t xlog_tid_t;
155 * Below are states for covering allocation transactions.
156 * By covering, we mean changing the h_tail_lsn in the last on-disk
157 * log write such that no allocation transactions will be re-done during
158 * recovery after a system crash. Recovery starts at the last on-disk
161 * These states are used to insert dummy log entries to cover
162 * space allocation transactions which can undo non-transactional changes
163 * after a crash. Writes to a file with space
164 * already allocated do not result in any transactions. Allocations
165 * might include space beyond the EOF. So if we just push the EOF a
166 * little, the last transaction for the file could contain the wrong
167 * size. If there is no file system activity, after an allocation
168 * transaction, and the system crashes, the allocation transaction
169 * will get replayed and the file will be truncated. This could
170 * be hours/days/... after the allocation occurred.
172 * The fix for this is to do two dummy transactions when the
173 * system is idle. We need two dummy transaction because the h_tail_lsn
174 * in the log record header needs to point beyond the last possible
175 * non-dummy transaction. The first dummy changes the h_tail_lsn to
176 * the first transaction before the dummy. The second dummy causes
177 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
179 * These dummy transactions get committed when everything
180 * is idle (after there has been some activity).
182 * There are 5 states used to control this.
184 * IDLE -- no logging has been done on the file system or
185 * we are done covering previous transactions.
186 * NEED -- logging has occurred and we need a dummy transaction
187 * when the log becomes idle.
188 * DONE -- we were in the NEED state and have committed a dummy
190 * NEED2 -- we detected that a dummy transaction has gone to the
191 * on disk log with no other transactions.
192 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
194 * There are two places where we switch states:
196 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
197 * We commit the dummy transaction and switch to DONE or DONE2,
198 * respectively. In all other states, we don't do anything.
200 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
202 * No matter what state we are in, if this isn't the dummy
203 * transaction going out, the next state is NEED.
204 * So, if we aren't in the DONE or DONE2 states, the next state
205 * is NEED. We can't be finishing a write of the dummy record
206 * unless it was committed and the state switched to DONE or DONE2.
208 * If we are in the DONE state and this was a write of the
209 * dummy transaction, we move to NEED2.
211 * If we are in the DONE2 state and this was a write of the
212 * dummy transaction, we move to IDLE.
215 * Writing only one dummy transaction can get appended to
216 * one file space allocation. When this happens, the log recovery
217 * code replays the space allocation and a file could be truncated.
218 * This is why we have the NEED2 and DONE2 states before going idle.
221 #define XLOG_STATE_COVER_IDLE 0
222 #define XLOG_STATE_COVER_NEED 1
223 #define XLOG_STATE_COVER_DONE 2
224 #define XLOG_STATE_COVER_NEED2 3
225 #define XLOG_STATE_COVER_DONE2 4
227 #define XLOG_COVER_OPS 5
230 /* Ticket reservation region accounting */
231 #define XLOG_TIC_LEN_MAX 15
235 * As would be stored in xfs_log_iovec but without the i_addr which
236 * we don't care about.
238 typedef struct xlog_res {
239 uint r_len; /* region length :4 */
240 uint r_type; /* region's transaction type :4 */
243 typedef struct xlog_ticket {
244 sv_t t_sema; /* sleep on this semaphore : 20 */
245 struct xlog_ticket *t_next; /* :4|8 */
246 struct xlog_ticket *t_prev; /* :4|8 */
247 xlog_tid_t t_tid; /* transaction identifier : 4 */
248 int t_curr_res; /* current reservation in bytes : 4 */
249 int t_unit_res; /* unit reservation in bytes : 4 */
250 char t_ocnt; /* original count : 1 */
251 char t_cnt; /* current count : 1 */
252 char t_clientid; /* who does this belong to; : 1 */
253 char t_flags; /* properties of reservation : 1 */
254 uint t_trans_type; /* transaction type : 4 */
256 /* reservation array fields */
257 uint t_res_num; /* num in array : 4 */
258 uint t_res_num_ophdrs; /* num op hdrs : 4 */
259 uint t_res_arr_sum; /* array sum : 4 */
260 uint t_res_o_flow; /* sum overflow : 4 */
261 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
267 typedef struct xlog_op_header {
268 __be32 oh_tid; /* transaction id of operation : 4 b */
269 __be32 oh_len; /* bytes in data region : 4 b */
270 __u8 oh_clientid; /* who sent me this : 1 b */
271 __u8 oh_flags; /* : 1 b */
272 __u16 oh_res2; /* 32 bit align : 2 b */
276 /* valid values for h_fmt */
277 #define XLOG_FMT_UNKNOWN 0
278 #define XLOG_FMT_LINUX_LE 1
279 #define XLOG_FMT_LINUX_BE 2
280 #define XLOG_FMT_IRIX_BE 3
283 #ifdef XFS_NATIVE_HOST
284 #define XLOG_FMT XLOG_FMT_LINUX_BE
286 #define XLOG_FMT XLOG_FMT_LINUX_LE
289 typedef struct xlog_rec_header {
290 __be32 h_magicno; /* log record (LR) identifier : 4 */
291 __be32 h_cycle; /* write cycle of log : 4 */
292 __be32 h_version; /* LR version : 4 */
293 __be32 h_len; /* len in bytes; should be 64-bit aligned: 4 */
294 __be64 h_lsn; /* lsn of this LR : 8 */
295 __be64 h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
296 __be32 h_chksum; /* may not be used; non-zero if used : 4 */
297 __be32 h_prev_block; /* block number to previous LR : 4 */
298 __be32 h_num_logops; /* number of log operations in this LR : 4 */
299 __be32 h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
301 __be32 h_fmt; /* format of log record : 4 */
302 uuid_t h_fs_uuid; /* uuid of FS : 16 */
303 __be32 h_size; /* iclog size : 4 */
306 typedef struct xlog_rec_ext_header {
307 __be32 xh_cycle; /* write cycle of log : 4 */
308 __be32 xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
309 } xlog_rec_ext_header_t;
313 * - A log record header is 512 bytes. There is plenty of room to grow the
314 * xlog_rec_header_t into the reserved space.
315 * - ic_data follows, so a write to disk can start at the beginning of
317 * - ic_forcesema is used to implement synchronous forcing of the iclog to disk.
318 * - ic_next is the pointer to the next iclog in the ring.
319 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
320 * - ic_log is a pointer back to the global log structure.
321 * - ic_callback is a linked list of callback function/argument pairs to be
322 * called after an iclog finishes writing.
323 * - ic_size is the full size of the header plus data.
324 * - ic_offset is the current number of bytes written to in this iclog.
325 * - ic_refcnt is bumped when someone is writing to the log.
326 * - ic_state is the state of the iclog.
328 * Because of cacheline contention on large machines, we need to separate
329 * various resources onto different cachelines. To start with, make the
330 * structure cacheline aligned. The following fields can be contended on
331 * by independent processes:
335 * - fields protected by the global l_icloglock
337 * so we need to ensure that these fields are located in separate cachelines.
338 * We'll put all the read-only and l_icloglock fields in the first cacheline,
339 * and move everything else out to subsequent cachelines.
341 typedef struct xlog_iclog_fields {
344 struct xlog_in_core *ic_next;
345 struct xlog_in_core *ic_prev;
346 struct xfs_buf *ic_bp;
352 char *ic_datap; /* pointer to iclog data */
354 struct ktrace *ic_trace;
357 /* Callback structures need their own cacheline */
358 spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
359 xfs_log_callback_t *ic_callback;
360 xfs_log_callback_t **ic_callback_tail;
362 /* reference counts need their own cacheline */
363 atomic_t ic_refcnt ____cacheline_aligned_in_smp;
364 } xlog_iclog_fields_t;
366 typedef union xlog_in_core2 {
367 xlog_rec_header_t hic_header;
368 xlog_rec_ext_header_t hic_xheader;
369 char hic_sector[XLOG_HEADER_SIZE];
372 typedef struct xlog_in_core {
373 xlog_iclog_fields_t hic_fields;
374 xlog_in_core_2_t *hic_data;
378 * Defines to save our code from this glop.
380 #define ic_forcesema hic_fields.ic_forcesema
381 #define ic_writesema hic_fields.ic_writesema
382 #define ic_next hic_fields.ic_next
383 #define ic_prev hic_fields.ic_prev
384 #define ic_bp hic_fields.ic_bp
385 #define ic_log hic_fields.ic_log
386 #define ic_callback hic_fields.ic_callback
387 #define ic_callback_lock hic_fields.ic_callback_lock
388 #define ic_callback_tail hic_fields.ic_callback_tail
389 #define ic_trace hic_fields.ic_trace
390 #define ic_size hic_fields.ic_size
391 #define ic_offset hic_fields.ic_offset
392 #define ic_refcnt hic_fields.ic_refcnt
393 #define ic_bwritecnt hic_fields.ic_bwritecnt
394 #define ic_state hic_fields.ic_state
395 #define ic_datap hic_fields.ic_datap
396 #define ic_header hic_data->hic_header
399 * The reservation head lsn is not made up of a cycle number and block number.
400 * Instead, it uses a cycle number and byte number. Logs don't expect to
401 * overflow 31 bits worth of byte offset, so using a byte number will mean
402 * that round off problems won't occur when releasing partial reservations.
405 /* The following fields don't need locking */
406 struct xfs_mount *l_mp; /* mount point */
407 struct xfs_buf *l_xbuf; /* extra buffer for log
409 struct xfs_buftarg *l_targ; /* buftarg of log */
411 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
412 struct xfs_buf_cancel **l_buf_cancel_table;
413 int l_iclog_hsize; /* size of iclog header */
414 int l_iclog_heads; /* # of iclog header sectors */
415 uint l_sectbb_log; /* log2 of sector size in BBs */
416 uint l_sectbb_mask; /* sector size (in BBs)
418 int l_iclog_size; /* size of log in bytes */
419 int l_iclog_size_log; /* log power size of log */
420 int l_iclog_bufs; /* number of iclog buffers */
421 xfs_daddr_t l_logBBstart; /* start block of log */
422 int l_logsize; /* size of log in bytes */
423 int l_logBBsize; /* size of log in BB chunks */
425 /* The following block of fields are changed while holding icloglock */
426 sv_t l_flush_wait ____cacheline_aligned_in_smp;
427 /* waiting for iclog flush */
428 int l_covered_state;/* state of "covering disk
430 xlog_in_core_t *l_iclog; /* head log queue */
431 spinlock_t l_icloglock; /* grab to change iclog state */
432 xfs_lsn_t l_tail_lsn; /* lsn of 1st LR with unflushed
434 xfs_lsn_t l_last_sync_lsn;/* lsn of last LR on disk */
435 int l_curr_cycle; /* Cycle number of log writes */
436 int l_prev_cycle; /* Cycle number before last
438 int l_curr_block; /* current logical log block */
439 int l_prev_block; /* previous logical log block */
441 /* The following block of fields are changed while holding grant_lock */
442 spinlock_t l_grant_lock ____cacheline_aligned_in_smp;
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;
451 struct ktrace *l_trace;
452 struct ktrace *l_grant_trace;
455 /* The following field are used for debugging; need to hold icloglock */
457 char *l_iclog_bak[XLOG_MAX_ICLOGS];
462 #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
465 /* common routines */
466 extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp);
467 extern int xlog_find_tail(xlog_t *log,
468 xfs_daddr_t *head_blk,
469 xfs_daddr_t *tail_blk);
470 extern int xlog_recover(xlog_t *log);
471 extern int xlog_recover_finish(xlog_t *log, int mfsi_flags);
472 extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
473 extern void xlog_recover_process_iunlinks(xlog_t *log);
475 extern struct xfs_buf *xlog_get_bp(xlog_t *, int);
476 extern void xlog_put_bp(struct xfs_buf *);
477 extern int xlog_bread(xlog_t *, xfs_daddr_t, int, struct xfs_buf *);
479 extern kmem_zone_t *xfs_log_ticket_zone;
482 #define XLOG_TRACE_GRAB_FLUSH 1
483 #define XLOG_TRACE_REL_FLUSH 2
484 #define XLOG_TRACE_SLEEP_FLUSH 3
485 #define XLOG_TRACE_WAKE_FLUSH 4
488 * Unmount record type is used as a pseudo transaction type for the ticket.
489 * It's value must be outside the range of XFS_TRANS_* values.
491 #define XLOG_UNMOUNT_REC_TYPE (-1U)
493 #endif /* __KERNEL__ */
495 #endif /* __XFS_LOG_PRIV_H__ */