uml: speed up page table walking
[pv_ops_mirror.git] / fs / xfs / xfs_log_priv.h
blob9bd3cdf11a87068db1b773b5afabb2ea0a5bc913
1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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__
21 struct xfs_buf;
22 struct ktrace;
23 struct log;
24 struct xlog_ticket;
25 struct xfs_buf_cancel;
26 struct xfs_mount;
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_HEADER_MAGIC_NUM 0xFEEDbabe /* Invalid cycle number */
36 #define XLOG_VERSION_1 1
37 #define XLOG_VERSION_2 2 /* Large IClogs, Log sunit */
38 #define XLOG_VERSION_OKBITS (XLOG_VERSION_1 | XLOG_VERSION_2)
39 #define XLOG_RECORD_BSIZE (16*1024) /* eventually 32k */
40 #define XLOG_BIG_RECORD_BSIZE (32*1024) /* 32k buffers */
41 #define XLOG_MAX_RECORD_BSIZE (256*1024)
42 #define XLOG_HEADER_CYCLE_SIZE (32*1024) /* cycle data in header */
43 #define XLOG_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
44 #define XLOG_BIG_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
45 #define XLOG_MAX_RECORD_BSHIFT 18 /* 256k == 1 << 18 */
46 #define XLOG_BTOLSUNIT(log, b) (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
47 (log)->l_mp->m_sb.sb_logsunit)
48 #define XLOG_LSUNITTOB(log, su) ((su) * (log)->l_mp->m_sb.sb_logsunit)
50 #define XLOG_HEADER_SIZE 512
52 #define XLOG_REC_SHIFT(log) \
53 BTOBB(1 << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
54 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
55 #define XLOG_TOTAL_REC_SHIFT(log) \
56 BTOBB(XLOG_MAX_ICLOGS << (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? \
57 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
60 * set lsns
63 #define ASSIGN_ANY_LSN_HOST(lsn,cycle,block) \
64 { \
65 (lsn) = ((xfs_lsn_t)(cycle)<<32)|(block); \
67 #define ASSIGN_ANY_LSN_DISK(lsn,cycle,block) \
68 { \
69 INT_SET(((uint *)&(lsn))[0], ARCH_CONVERT, (cycle)); \
70 INT_SET(((uint *)&(lsn))[1], ARCH_CONVERT, (block)); \
72 #define ASSIGN_LSN(lsn,log) \
73 ASSIGN_ANY_LSN_DISK(lsn,(log)->l_curr_cycle,(log)->l_curr_block);
75 #define XLOG_SET(f,b) (((f) & (b)) == (b))
77 #define GET_CYCLE(ptr, arch) \
78 (INT_GET(*(uint *)(ptr), arch) == XLOG_HEADER_MAGIC_NUM ? \
79 INT_GET(*((uint *)(ptr)+1), arch) : \
80 INT_GET(*(uint *)(ptr), arch) \
83 #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
86 #ifdef __KERNEL__
89 * get client id from packed copy.
91 * this hack is here because the xlog_pack code copies four bytes
92 * of xlog_op_header containing the fields oh_clientid, oh_flags
93 * and oh_res2 into the packed copy.
95 * later on this four byte chunk is treated as an int and the
96 * client id is pulled out.
98 * this has endian issues, of course.
101 #ifndef XFS_NATIVE_HOST
102 #define GET_CLIENT_ID(i,arch) \
103 ((i) & 0xff)
104 #else
105 #define GET_CLIENT_ID(i,arch) \
106 ((i) >> 24)
107 #endif
109 #define GRANT_LOCK(log) mutex_spinlock(&(log)->l_grant_lock)
110 #define GRANT_UNLOCK(log, s) mutex_spinunlock(&(log)->l_grant_lock, s)
111 #define LOG_LOCK(log) mutex_spinlock(&(log)->l_icloglock)
112 #define LOG_UNLOCK(log, s) mutex_spinunlock(&(log)->l_icloglock, s)
114 #define xlog_panic(args...) cmn_err(CE_PANIC, ## args)
115 #define xlog_exit(args...) cmn_err(CE_PANIC, ## args)
116 #define xlog_warn(args...) cmn_err(CE_WARN, ## args)
119 * In core log state
121 #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
122 #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
123 #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
124 #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
125 #define XLOG_STATE_DO_CALLBACK \
126 0x0010 /* Process callback functions */
127 #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
128 #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
129 #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
130 #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
131 #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
132 #endif /* __KERNEL__ */
135 * Flags to log operation header
137 * The first write of a new transaction will be preceded with a start
138 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
139 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
140 * the remainder of the current active in-core log, it is split up into
141 * multiple regions. Each partial region will be marked with a
142 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
145 #define XLOG_START_TRANS 0x01 /* Start a new transaction */
146 #define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
147 #define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
148 #define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
149 #define XLOG_END_TRANS 0x10 /* End a continued transaction */
150 #define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
152 #ifdef __KERNEL__
154 * Flags to log ticket
156 #define XLOG_TIC_INITED 0x1 /* has been initialized */
157 #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
158 #define XLOG_TIC_IN_Q 0x4
159 #endif /* __KERNEL__ */
161 #define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
164 * Flags for log structure
166 #define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
167 #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
168 #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
169 #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
170 shutdown */
171 typedef __uint32_t xlog_tid_t;
174 #ifdef __KERNEL__
176 * Below are states for covering allocation transactions.
177 * By covering, we mean changing the h_tail_lsn in the last on-disk
178 * log write such that no allocation transactions will be re-done during
179 * recovery after a system crash. Recovery starts at the last on-disk
180 * log write.
182 * These states are used to insert dummy log entries to cover
183 * space allocation transactions which can undo non-transactional changes
184 * after a crash. Writes to a file with space
185 * already allocated do not result in any transactions. Allocations
186 * might include space beyond the EOF. So if we just push the EOF a
187 * little, the last transaction for the file could contain the wrong
188 * size. If there is no file system activity, after an allocation
189 * transaction, and the system crashes, the allocation transaction
190 * will get replayed and the file will be truncated. This could
191 * be hours/days/... after the allocation occurred.
193 * The fix for this is to do two dummy transactions when the
194 * system is idle. We need two dummy transaction because the h_tail_lsn
195 * in the log record header needs to point beyond the last possible
196 * non-dummy transaction. The first dummy changes the h_tail_lsn to
197 * the first transaction before the dummy. The second dummy causes
198 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
200 * These dummy transactions get committed when everything
201 * is idle (after there has been some activity).
203 * There are 5 states used to control this.
205 * IDLE -- no logging has been done on the file system or
206 * we are done covering previous transactions.
207 * NEED -- logging has occurred and we need a dummy transaction
208 * when the log becomes idle.
209 * DONE -- we were in the NEED state and have committed a dummy
210 * transaction.
211 * NEED2 -- we detected that a dummy transaction has gone to the
212 * on disk log with no other transactions.
213 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
215 * There are two places where we switch states:
217 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
218 * We commit the dummy transaction and switch to DONE or DONE2,
219 * respectively. In all other states, we don't do anything.
221 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
223 * No matter what state we are in, if this isn't the dummy
224 * transaction going out, the next state is NEED.
225 * So, if we aren't in the DONE or DONE2 states, the next state
226 * is NEED. We can't be finishing a write of the dummy record
227 * unless it was committed and the state switched to DONE or DONE2.
229 * If we are in the DONE state and this was a write of the
230 * dummy transaction, we move to NEED2.
232 * If we are in the DONE2 state and this was a write of the
233 * dummy transaction, we move to IDLE.
236 * Writing only one dummy transaction can get appended to
237 * one file space allocation. When this happens, the log recovery
238 * code replays the space allocation and a file could be truncated.
239 * This is why we have the NEED2 and DONE2 states before going idle.
242 #define XLOG_STATE_COVER_IDLE 0
243 #define XLOG_STATE_COVER_NEED 1
244 #define XLOG_STATE_COVER_DONE 2
245 #define XLOG_STATE_COVER_NEED2 3
246 #define XLOG_STATE_COVER_DONE2 4
248 #define XLOG_COVER_OPS 5
251 /* Ticket reservation region accounting */
252 #define XLOG_TIC_LEN_MAX 15
253 #define XLOG_TIC_RESET_RES(t) ((t)->t_res_num = \
254 (t)->t_res_arr_sum = (t)->t_res_num_ophdrs = 0)
255 #define XLOG_TIC_ADD_OPHDR(t) ((t)->t_res_num_ophdrs++)
256 #define XLOG_TIC_ADD_REGION(t, len, type) \
257 do { \
258 if ((t)->t_res_num == XLOG_TIC_LEN_MAX) { \
259 /* add to overflow and start again */ \
260 (t)->t_res_o_flow += (t)->t_res_arr_sum; \
261 (t)->t_res_num = 0; \
262 (t)->t_res_arr_sum = 0; \
264 (t)->t_res_arr[(t)->t_res_num].r_len = (len); \
265 (t)->t_res_arr[(t)->t_res_num].r_type = (type); \
266 (t)->t_res_arr_sum += (len); \
267 (t)->t_res_num++; \
268 } while (0)
271 * Reservation region
272 * As would be stored in xfs_log_iovec but without the i_addr which
273 * we don't care about.
275 typedef struct xlog_res {
276 uint r_len; /* region length :4 */
277 uint r_type; /* region's transaction type :4 */
278 } xlog_res_t;
280 typedef struct xlog_ticket {
281 sv_t t_sema; /* sleep on this semaphore : 20 */
282 struct xlog_ticket *t_next; /* :4|8 */
283 struct xlog_ticket *t_prev; /* :4|8 */
284 xlog_tid_t t_tid; /* transaction identifier : 4 */
285 int t_curr_res; /* current reservation in bytes : 4 */
286 int t_unit_res; /* unit reservation in bytes : 4 */
287 char t_ocnt; /* original count : 1 */
288 char t_cnt; /* current count : 1 */
289 char t_clientid; /* who does this belong to; : 1 */
290 char t_flags; /* properties of reservation : 1 */
291 uint t_trans_type; /* transaction type : 4 */
293 /* reservation array fields */
294 uint t_res_num; /* num in array : 4 */
295 uint t_res_num_ophdrs; /* num op hdrs : 4 */
296 uint t_res_arr_sum; /* array sum : 4 */
297 uint t_res_o_flow; /* sum overflow : 4 */
298 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
299 } xlog_ticket_t;
301 #endif
304 typedef struct xlog_op_header {
305 xlog_tid_t oh_tid; /* transaction id of operation : 4 b */
306 int oh_len; /* bytes in data region : 4 b */
307 __uint8_t oh_clientid; /* who sent me this : 1 b */
308 __uint8_t oh_flags; /* : 1 b */
309 ushort oh_res2; /* 32 bit align : 2 b */
310 } xlog_op_header_t;
313 /* valid values for h_fmt */
314 #define XLOG_FMT_UNKNOWN 0
315 #define XLOG_FMT_LINUX_LE 1
316 #define XLOG_FMT_LINUX_BE 2
317 #define XLOG_FMT_IRIX_BE 3
319 /* our fmt */
320 #ifdef XFS_NATIVE_HOST
321 #define XLOG_FMT XLOG_FMT_LINUX_BE
322 #else
323 #define XLOG_FMT XLOG_FMT_LINUX_LE
324 #endif
326 typedef struct xlog_rec_header {
327 uint h_magicno; /* log record (LR) identifier : 4 */
328 uint h_cycle; /* write cycle of log : 4 */
329 int h_version; /* LR version : 4 */
330 int h_len; /* len in bytes; should be 64-bit aligned: 4 */
331 xfs_lsn_t h_lsn; /* lsn of this LR : 8 */
332 xfs_lsn_t h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
333 uint h_chksum; /* may not be used; non-zero if used : 4 */
334 int h_prev_block; /* block number to previous LR : 4 */
335 int h_num_logops; /* number of log operations in this LR : 4 */
336 uint h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
337 /* new fields */
338 int h_fmt; /* format of log record : 4 */
339 uuid_t h_fs_uuid; /* uuid of FS : 16 */
340 int h_size; /* iclog size : 4 */
341 } xlog_rec_header_t;
343 typedef struct xlog_rec_ext_header {
344 uint xh_cycle; /* write cycle of log : 4 */
345 uint xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
346 } xlog_rec_ext_header_t;
348 #ifdef __KERNEL__
350 * - A log record header is 512 bytes. There is plenty of room to grow the
351 * xlog_rec_header_t into the reserved space.
352 * - ic_data follows, so a write to disk can start at the beginning of
353 * the iclog.
354 * - ic_forcesema is used to implement synchronous forcing of the iclog to disk.
355 * - ic_next is the pointer to the next iclog in the ring.
356 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
357 * - ic_log is a pointer back to the global log structure.
358 * - ic_callback is a linked list of callback function/argument pairs to be
359 * called after an iclog finishes writing.
360 * - ic_size is the full size of the header plus data.
361 * - ic_offset is the current number of bytes written to in this iclog.
362 * - ic_refcnt is bumped when someone is writing to the log.
363 * - ic_state is the state of the iclog.
365 typedef struct xlog_iclog_fields {
366 sv_t ic_forcesema;
367 sv_t ic_writesema;
368 struct xlog_in_core *ic_next;
369 struct xlog_in_core *ic_prev;
370 struct xfs_buf *ic_bp;
371 struct log *ic_log;
372 xfs_log_callback_t *ic_callback;
373 xfs_log_callback_t **ic_callback_tail;
374 #ifdef XFS_LOG_TRACE
375 struct ktrace *ic_trace;
376 #endif
377 int ic_size;
378 int ic_offset;
379 int ic_refcnt;
380 int ic_bwritecnt;
381 ushort_t ic_state;
382 char *ic_datap; /* pointer to iclog data */
383 } xlog_iclog_fields_t;
385 typedef union xlog_in_core2 {
386 xlog_rec_header_t hic_header;
387 xlog_rec_ext_header_t hic_xheader;
388 char hic_sector[XLOG_HEADER_SIZE];
389 } xlog_in_core_2_t;
391 typedef struct xlog_in_core {
392 xlog_iclog_fields_t hic_fields;
393 xlog_in_core_2_t *hic_data;
394 } xlog_in_core_t;
397 * Defines to save our code from this glop.
399 #define ic_forcesema hic_fields.ic_forcesema
400 #define ic_writesema hic_fields.ic_writesema
401 #define ic_next hic_fields.ic_next
402 #define ic_prev hic_fields.ic_prev
403 #define ic_bp hic_fields.ic_bp
404 #define ic_log hic_fields.ic_log
405 #define ic_callback hic_fields.ic_callback
406 #define ic_callback_tail hic_fields.ic_callback_tail
407 #define ic_trace hic_fields.ic_trace
408 #define ic_size hic_fields.ic_size
409 #define ic_offset hic_fields.ic_offset
410 #define ic_refcnt hic_fields.ic_refcnt
411 #define ic_bwritecnt hic_fields.ic_bwritecnt
412 #define ic_state hic_fields.ic_state
413 #define ic_datap hic_fields.ic_datap
414 #define ic_header hic_data->hic_header
417 * The reservation head lsn is not made up of a cycle number and block number.
418 * Instead, it uses a cycle number and byte number. Logs don't expect to
419 * overflow 31 bits worth of byte offset, so using a byte number will mean
420 * that round off problems won't occur when releasing partial reservations.
422 typedef struct log {
423 /* The following block of fields are changed while holding icloglock */
424 sema_t l_flushsema; /* iclog flushing semaphore */
425 int l_flushcnt; /* # of procs waiting on this
426 * sema */
427 int l_ticket_cnt; /* free ticket count */
428 int l_ticket_tcnt; /* total ticket count */
429 int l_covered_state;/* state of "covering disk
430 * log entries" */
431 xlog_ticket_t *l_freelist; /* free list of tickets */
432 xlog_ticket_t *l_unmount_free;/* kmem_free these addresses */
433 xlog_ticket_t *l_tail; /* free list of tickets */
434 xlog_in_core_t *l_iclog; /* head log queue */
435 lock_t l_icloglock; /* grab to change iclog state */
436 xfs_lsn_t l_tail_lsn; /* lsn of 1st LR with unflushed
437 * buffers */
438 xfs_lsn_t l_last_sync_lsn;/* lsn of last LR on disk */
439 struct xfs_mount *l_mp; /* mount point */
440 struct xfs_buf *l_xbuf; /* extra buffer for log
441 * wrapping */
442 struct xfs_buftarg *l_targ; /* buftarg of log */
443 xfs_daddr_t l_logBBstart; /* start block of log */
444 int l_logsize; /* size of log in bytes */
445 int l_logBBsize; /* size of log in BB chunks */
446 int l_curr_cycle; /* Cycle number of log writes */
447 int l_prev_cycle; /* Cycle number before last
448 * block increment */
449 int l_curr_block; /* current logical log block */
450 int l_prev_block; /* previous logical log block */
451 int l_iclog_size; /* size of log in bytes */
452 int l_iclog_size_log; /* log power size of log */
453 int l_iclog_bufs; /* number of iclog buffers */
455 /* The following field are used for debugging; need to hold icloglock */
456 char *l_iclog_bak[XLOG_MAX_ICLOGS];
458 /* The following block of fields are changed while holding grant_lock */
459 lock_t l_grant_lock;
460 xlog_ticket_t *l_reserve_headq;
461 xlog_ticket_t *l_write_headq;
462 int l_grant_reserve_cycle;
463 int l_grant_reserve_bytes;
464 int l_grant_write_cycle;
465 int l_grant_write_bytes;
467 /* The following fields don't need locking */
468 #ifdef XFS_LOG_TRACE
469 struct ktrace *l_trace;
470 struct ktrace *l_grant_trace;
471 #endif
472 uint l_flags;
473 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
474 struct xfs_buf_cancel **l_buf_cancel_table;
475 int l_iclog_hsize; /* size of iclog header */
476 int l_iclog_heads; /* # of iclog header sectors */
477 uint l_sectbb_log; /* log2 of sector size in BBs */
478 uint l_sectbb_mask; /* sector size (in BBs)
479 * alignment mask */
480 } xlog_t;
482 #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
485 /* common routines */
486 extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp);
487 extern int xlog_find_tail(xlog_t *log,
488 xfs_daddr_t *head_blk,
489 xfs_daddr_t *tail_blk);
490 extern int xlog_recover(xlog_t *log);
491 extern int xlog_recover_finish(xlog_t *log, int mfsi_flags);
492 extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
493 extern void xlog_recover_process_iunlinks(xlog_t *log);
495 extern struct xfs_buf *xlog_get_bp(xlog_t *, int);
496 extern void xlog_put_bp(struct xfs_buf *);
497 extern int xlog_bread(xlog_t *, xfs_daddr_t, int, struct xfs_buf *);
499 /* iclog tracing */
500 #define XLOG_TRACE_GRAB_FLUSH 1
501 #define XLOG_TRACE_REL_FLUSH 2
502 #define XLOG_TRACE_SLEEP_FLUSH 3
503 #define XLOG_TRACE_WAKE_FLUSH 4
506 * Unmount record type is used as a pseudo transaction type for the ticket.
507 * It's value must be outside the range of XFS_TRANS_* values.
509 #define XLOG_UNMOUNT_REC_TYPE (-1U)
511 #endif /* __KERNEL__ */
513 #endif /* __XFS_LOG_PRIV_H__ */