blob: edcdfe01617f673bc1047caca3acc243f66a0658 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
18#ifndef __XFS_LOG_PRIV_H__
19#define __XFS_LOG_PRIV_H__
20
21struct xfs_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -070022struct log;
Nathan Scotta844f452005-11-02 14:38:42 +110023struct xlog_ticket;
Linus Torvalds1da177e2005-04-16 15:20:36 -070024struct xfs_buf_cancel;
25struct xfs_mount;
26
27/*
28 * Macros, structures, prototypes for internal log manager use.
29 */
30
31#define XLOG_MIN_ICLOGS 2
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#define XLOG_MAX_ICLOGS 8
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#define XLOG_HEADER_MAGIC_NUM 0xFEEDbabe /* Invalid cycle number */
34#define XLOG_VERSION_1 1
35#define XLOG_VERSION_2 2 /* Large IClogs, Log sunit */
36#define XLOG_VERSION_OKBITS (XLOG_VERSION_1 | XLOG_VERSION_2)
Eric Sandeendcb3b832007-08-16 16:25:33 +100037#define XLOG_MIN_RECORD_BSIZE (16*1024) /* eventually 32k */
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#define XLOG_BIG_RECORD_BSIZE (32*1024) /* 32k buffers */
39#define XLOG_MAX_RECORD_BSIZE (256*1024)
40#define XLOG_HEADER_CYCLE_SIZE (32*1024) /* cycle data in header */
Eric Sandeendcb3b832007-08-16 16:25:33 +100041#define XLOG_MIN_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#define XLOG_BIG_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
43#define XLOG_MAX_RECORD_BSHIFT 18 /* 256k == 1 << 18 */
44#define XLOG_BTOLSUNIT(log, b) (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
45 (log)->l_mp->m_sb.sb_logsunit)
46#define XLOG_LSUNITTOB(log, su) ((su) * (log)->l_mp->m_sb.sb_logsunit)
47
48#define XLOG_HEADER_SIZE 512
49
50#define XLOG_REC_SHIFT(log) \
Eric Sandeen62118702008-03-06 13:44:28 +110051 BTOBB(1 << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
53#define XLOG_TOTAL_REC_SHIFT(log) \
Eric Sandeen62118702008-03-06 13:44:28 +110054 BTOBB(XLOG_MAX_ICLOGS << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
56
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100058static inline xfs_lsn_t xlog_assign_lsn(uint cycle, uint block)
59{
60 return ((xfs_lsn_t)cycle << 32) | block;
61}
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100063static inline uint xlog_get_cycle(char *ptr)
64{
Christoph Hellwigb53e6752007-10-12 10:59:34 +100065 if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
66 return be32_to_cpu(*((__be32 *)ptr + 1));
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100067 else
Christoph Hellwigb53e6752007-10-12 10:59:34 +100068 return be32_to_cpu(*(__be32 *)ptr);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100069}
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
71#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#ifdef __KERNEL__
74
75/*
76 * get client id from packed copy.
77 *
78 * this hack is here because the xlog_pack code copies four bytes
79 * of xlog_op_header containing the fields oh_clientid, oh_flags
80 * and oh_res2 into the packed copy.
81 *
82 * later on this four byte chunk is treated as an int and the
83 * client id is pulled out.
84 *
85 * this has endian issues, of course.
86 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +100087static inline uint xlog_get_client_id(__be32 i)
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100088{
Christoph Hellwigb53e6752007-10-12 10:59:34 +100089 return be32_to_cpu(i) >> 24;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100090}
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Linus Torvalds1da177e2005-04-16 15:20:36 -070092#define xlog_panic(args...) cmn_err(CE_PANIC, ## args)
93#define xlog_exit(args...) cmn_err(CE_PANIC, ## args)
94#define xlog_warn(args...) cmn_err(CE_WARN, ## args)
95
96/*
97 * In core log state
98 */
99#define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
100#define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
101#define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
102#define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
103#define XLOG_STATE_DO_CALLBACK \
104 0x0010 /* Process callback functions */
105#define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
106#define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
107#define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
108#define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
109#define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
110#endif /* __KERNEL__ */
111
112/*
113 * Flags to log operation header
114 *
115 * The first write of a new transaction will be preceded with a start
116 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
117 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
118 * the remainder of the current active in-core log, it is split up into
119 * multiple regions. Each partial region will be marked with a
120 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
121 *
122 */
123#define XLOG_START_TRANS 0x01 /* Start a new transaction */
124#define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
125#define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
126#define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
127#define XLOG_END_TRANS 0x10 /* End a continued transaction */
128#define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130#ifdef __KERNEL__
131/*
132 * Flags to log ticket
133 */
134#define XLOG_TIC_INITED 0x1 /* has been initialized */
135#define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
136#define XLOG_TIC_IN_Q 0x4
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000137
138#define XLOG_TIC_FLAGS \
139 { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \
140 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }, \
141 { XLOG_TIC_IN_Q, "XLOG_TIC_IN_Q" }
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143#endif /* __KERNEL__ */
144
145#define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
146
147/*
148 * Flags for log structure
149 */
150#define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
151#define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
152#define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
153#define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
154 shutdown */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
156#ifdef __KERNEL__
157/*
158 * Below are states for covering allocation transactions.
159 * By covering, we mean changing the h_tail_lsn in the last on-disk
160 * log write such that no allocation transactions will be re-done during
161 * recovery after a system crash. Recovery starts at the last on-disk
162 * log write.
163 *
164 * These states are used to insert dummy log entries to cover
165 * space allocation transactions which can undo non-transactional changes
166 * after a crash. Writes to a file with space
167 * already allocated do not result in any transactions. Allocations
168 * might include space beyond the EOF. So if we just push the EOF a
169 * little, the last transaction for the file could contain the wrong
170 * size. If there is no file system activity, after an allocation
171 * transaction, and the system crashes, the allocation transaction
172 * will get replayed and the file will be truncated. This could
173 * be hours/days/... after the allocation occurred.
174 *
175 * The fix for this is to do two dummy transactions when the
176 * system is idle. We need two dummy transaction because the h_tail_lsn
177 * in the log record header needs to point beyond the last possible
178 * non-dummy transaction. The first dummy changes the h_tail_lsn to
179 * the first transaction before the dummy. The second dummy causes
180 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
181 *
182 * These dummy transactions get committed when everything
183 * is idle (after there has been some activity).
184 *
185 * There are 5 states used to control this.
186 *
187 * IDLE -- no logging has been done on the file system or
188 * we are done covering previous transactions.
189 * NEED -- logging has occurred and we need a dummy transaction
190 * when the log becomes idle.
191 * DONE -- we were in the NEED state and have committed a dummy
192 * transaction.
193 * NEED2 -- we detected that a dummy transaction has gone to the
194 * on disk log with no other transactions.
195 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
196 *
197 * There are two places where we switch states:
198 *
199 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
200 * We commit the dummy transaction and switch to DONE or DONE2,
201 * respectively. In all other states, we don't do anything.
202 *
203 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
204 *
205 * No matter what state we are in, if this isn't the dummy
206 * transaction going out, the next state is NEED.
207 * So, if we aren't in the DONE or DONE2 states, the next state
208 * is NEED. We can't be finishing a write of the dummy record
209 * unless it was committed and the state switched to DONE or DONE2.
210 *
211 * If we are in the DONE state and this was a write of the
212 * dummy transaction, we move to NEED2.
213 *
214 * If we are in the DONE2 state and this was a write of the
215 * dummy transaction, we move to IDLE.
216 *
217 *
218 * Writing only one dummy transaction can get appended to
219 * one file space allocation. When this happens, the log recovery
220 * code replays the space allocation and a file could be truncated.
221 * This is why we have the NEED2 and DONE2 states before going idle.
222 */
223
224#define XLOG_STATE_COVER_IDLE 0
225#define XLOG_STATE_COVER_NEED 1
226#define XLOG_STATE_COVER_DONE 2
227#define XLOG_STATE_COVER_NEED2 3
228#define XLOG_STATE_COVER_DONE2 4
229
230#define XLOG_COVER_OPS 5
231
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000232
233/* Ticket reservation region accounting */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000234#define XLOG_TIC_LEN_MAX 15
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000235
236/*
237 * Reservation region
238 * As would be stored in xfs_log_iovec but without the i_addr which
239 * we don't care about.
240 */
241typedef struct xlog_res {
Tim Shimmin12598452006-01-11 21:02:47 +1100242 uint r_len; /* region length :4 */
243 uint r_type; /* region's transaction type :4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000244} xlog_res_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000245
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246typedef struct xlog_ticket {
David Chinner12017fa2008-08-13 16:34:31 +1000247 sv_t t_wait; /* ticket wait queue : 20 */
David Chinnereb01c9c2008-04-10 12:18:46 +1000248 struct xlog_ticket *t_next; /* :4|8 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000249 struct xlog_ticket *t_prev; /* :4|8 */
250 xlog_tid_t t_tid; /* transaction identifier : 4 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100251 atomic_t t_ref; /* ticket reference count : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000252 int t_curr_res; /* current reservation in bytes : 4 */
253 int t_unit_res; /* unit reservation in bytes : 4 */
254 char t_ocnt; /* original count : 1 */
255 char t_cnt; /* current count : 1 */
256 char t_clientid; /* who does this belong to; : 1 */
257 char t_flags; /* properties of reservation : 1 */
258 uint t_trans_type; /* transaction type : 4 */
259
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000260 /* reservation array fields */
261 uint t_res_num; /* num in array : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000262 uint t_res_num_ophdrs; /* num op hdrs : 4 */
263 uint t_res_arr_sum; /* array sum : 4 */
264 uint t_res_o_flow; /* sum overflow : 4 */
Tim Shimmin12598452006-01-11 21:02:47 +1100265 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266} xlog_ticket_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268#endif
269
270
271typedef struct xlog_op_header {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +1000272 __be32 oh_tid; /* transaction id of operation : 4 b */
273 __be32 oh_len; /* bytes in data region : 4 b */
274 __u8 oh_clientid; /* who sent me this : 1 b */
275 __u8 oh_flags; /* : 1 b */
276 __u16 oh_res2; /* 32 bit align : 2 b */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277} xlog_op_header_t;
278
279
280/* valid values for h_fmt */
281#define XLOG_FMT_UNKNOWN 0
282#define XLOG_FMT_LINUX_LE 1
283#define XLOG_FMT_LINUX_BE 2
284#define XLOG_FMT_IRIX_BE 3
285
286/* our fmt */
Nathan Scottf016bad2005-09-08 15:30:05 +1000287#ifdef XFS_NATIVE_HOST
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288#define XLOG_FMT XLOG_FMT_LINUX_BE
289#else
Nathan Scottf016bad2005-09-08 15:30:05 +1000290#define XLOG_FMT XLOG_FMT_LINUX_LE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291#endif
292
293typedef struct xlog_rec_header {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000294 __be32 h_magicno; /* log record (LR) identifier : 4 */
295 __be32 h_cycle; /* write cycle of log : 4 */
296 __be32 h_version; /* LR version : 4 */
297 __be32 h_len; /* len in bytes; should be 64-bit aligned: 4 */
298 __be64 h_lsn; /* lsn of this LR : 8 */
299 __be64 h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
300 __be32 h_chksum; /* may not be used; non-zero if used : 4 */
301 __be32 h_prev_block; /* block number to previous LR : 4 */
302 __be32 h_num_logops; /* number of log operations in this LR : 4 */
303 __be32 h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000305 __be32 h_fmt; /* format of log record : 4 */
306 uuid_t h_fs_uuid; /* uuid of FS : 16 */
307 __be32 h_size; /* iclog size : 4 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308} xlog_rec_header_t;
309
310typedef struct xlog_rec_ext_header {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000311 __be32 xh_cycle; /* write cycle of log : 4 */
312 __be32 xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313} xlog_rec_ext_header_t;
314
315#ifdef __KERNEL__
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100316
317/*
318 * Quite misnamed, because this union lays out the actual on-disk log buffer.
319 */
320typedef union xlog_in_core2 {
321 xlog_rec_header_t hic_header;
322 xlog_rec_ext_header_t hic_xheader;
323 char hic_sector[XLOG_HEADER_SIZE];
324} xlog_in_core_2_t;
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326/*
327 * - A log record header is 512 bytes. There is plenty of room to grow the
328 * xlog_rec_header_t into the reserved space.
329 * - ic_data follows, so a write to disk can start at the beginning of
330 * the iclog.
David Chinner12017fa2008-08-13 16:34:31 +1000331 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 * - ic_next is the pointer to the next iclog in the ring.
333 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
334 * - ic_log is a pointer back to the global log structure.
335 * - ic_callback is a linked list of callback function/argument pairs to be
336 * called after an iclog finishes writing.
337 * - ic_size is the full size of the header plus data.
338 * - ic_offset is the current number of bytes written to in this iclog.
339 * - ic_refcnt is bumped when someone is writing to the log.
340 * - ic_state is the state of the iclog.
David Chinner114d23a2008-04-10 12:18:39 +1000341 *
342 * Because of cacheline contention on large machines, we need to separate
343 * various resources onto different cachelines. To start with, make the
344 * structure cacheline aligned. The following fields can be contended on
345 * by independent processes:
346 *
347 * - ic_callback_*
348 * - ic_refcnt
349 * - fields protected by the global l_icloglock
350 *
351 * so we need to ensure that these fields are located in separate cachelines.
352 * We'll put all the read-only and l_icloglock fields in the first cacheline,
353 * and move everything else out to subsequent cachelines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 */
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100355typedef struct xlog_in_core {
David Chinner12017fa2008-08-13 16:34:31 +1000356 sv_t ic_force_wait;
357 sv_t ic_write_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 struct xlog_in_core *ic_next;
359 struct xlog_in_core *ic_prev;
360 struct xfs_buf *ic_bp;
361 struct log *ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 int ic_size;
363 int ic_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 int ic_bwritecnt;
Christoph Hellwiga5687782009-02-09 08:37:39 +0100365 unsigned short ic_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 char *ic_datap; /* pointer to iclog data */
David Chinner114d23a2008-04-10 12:18:39 +1000367
368 /* Callback structures need their own cacheline */
369 spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
370 xfs_log_callback_t *ic_callback;
371 xfs_log_callback_t **ic_callback_tail;
372
373 /* reference counts need their own cacheline */
374 atomic_t ic_refcnt ____cacheline_aligned_in_smp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100375 xlog_in_core_2_t *ic_data;
376#define ic_header ic_data->hic_header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377} xlog_in_core_t;
378
379/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000380 * The CIL context is used to aggregate per-transaction details as well be
381 * passed to the iclog for checkpoint post-commit processing. After being
382 * passed to the iclog, another context needs to be allocated for tracking the
383 * next set of transactions to be aggregated into a checkpoint.
384 */
385struct xfs_cil;
386
387struct xfs_cil_ctx {
388 struct xfs_cil *cil;
389 xfs_lsn_t sequence; /* chkpt sequence # */
390 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
391 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
392 struct xlog_ticket *ticket; /* chkpt ticket */
393 int nvecs; /* number of regions */
394 int space_used; /* aggregate size of regions */
395 struct list_head busy_extents; /* busy extents in chkpt */
396 struct xfs_log_vec *lv_chain; /* logvecs being pushed */
397 xfs_log_callback_t log_cb; /* completion callback hook. */
398 struct list_head committing; /* ctx committing list */
399};
400
401/*
402 * Committed Item List structure
403 *
404 * This structure is used to track log items that have been committed but not
405 * yet written into the log. It is used only when the delayed logging mount
406 * option is enabled.
407 *
408 * This structure tracks the list of committing checkpoint contexts so
409 * we can avoid the problem of having to hold out new transactions during a
410 * flush until we have a the commit record LSN of the checkpoint. We can
411 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
412 * sequence match and extract the commit LSN directly from there. If the
413 * checkpoint is still in the process of committing, we can block waiting for
414 * the commit LSN to be determined as well. This should make synchronous
415 * operations almost as efficient as the old logging methods.
416 */
417struct xfs_cil {
418 struct log *xc_log;
419 struct list_head xc_cil;
420 spinlock_t xc_cil_lock;
421 struct xfs_cil_ctx *xc_ctx;
422 struct rw_semaphore xc_ctx_lock;
423 struct list_head xc_committing;
424 sv_t xc_commit_wait;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000425 xfs_lsn_t xc_current_sequence;
Dave Chinner71e330b2010-05-21 14:37:18 +1000426};
427
428/*
Dave Chinner80168672010-09-24 18:13:44 +1000429 * The amount of log space we allow the CIL to aggregate is difficult to size.
430 * Whatever we choose, we have to make sure we can get a reservation for the
431 * log space effectively, that it is large enough to capture sufficient
432 * relogging to reduce log buffer IO significantly, but it is not too large for
433 * the log or induces too much latency when writing out through the iclogs. We
434 * track both space consumed and the number of vectors in the checkpoint
435 * context, so we need to decide which to use for limiting.
Dave Chinnerdf806152010-05-17 15:52:13 +1000436 *
437 * Every log buffer we write out during a push needs a header reserved, which
438 * is at least one sector and more for v2 logs. Hence we need a reservation of
439 * at least 512 bytes per 32k of log space just for the LR headers. That means
440 * 16KB of reservation per megabyte of delayed logging space we will consume,
441 * plus various headers. The number of headers will vary based on the num of
442 * io vectors, so limiting on a specific number of vectors is going to result
443 * in transactions of varying size. IOWs, it is more consistent to track and
444 * limit space consumed in the log rather than by the number of objects being
445 * logged in order to prevent checkpoint ticket overruns.
446 *
447 * Further, use of static reservations through the log grant mechanism is
448 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
449 * grant) and a significant deadlock potential because regranting write space
450 * can block on log pushes. Hence if we have to regrant log space during a log
451 * push, we can deadlock.
452 *
453 * However, we can avoid this by use of a dynamic "reservation stealing"
454 * technique during transaction commit whereby unused reservation space in the
455 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
456 * space needed by the checkpoint transaction. This means that we never need to
457 * specifically reserve space for the CIL checkpoint transaction, nor do we
458 * need to regrant space once the checkpoint completes. This also means the
459 * checkpoint transaction ticket is specific to the checkpoint context, rather
460 * than the CIL itself.
461 *
Dave Chinner80168672010-09-24 18:13:44 +1000462 * With dynamic reservations, we can effectively make up arbitrary limits for
463 * the checkpoint size so long as they don't violate any other size rules.
464 * Recovery imposes a rule that no transaction exceed half the log, so we are
465 * limited by that. Furthermore, the log transaction reservation subsystem
466 * tries to keep 25% of the log free, so we need to keep below that limit or we
467 * risk running out of free log space to start any new transactions.
468 *
469 * In order to keep background CIL push efficient, we will set a lower
470 * threshold at which background pushing is attempted without blocking current
471 * transaction commits. A separate, higher bound defines when CIL pushes are
472 * enforced to ensure we stay within our maximum checkpoint size bounds.
473 * threshold, yet give us plenty of space for aggregation on large logs.
Dave Chinnerdf806152010-05-17 15:52:13 +1000474 */
Dave Chinner80168672010-09-24 18:13:44 +1000475#define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3)
476#define XLOG_CIL_HARD_SPACE_LIMIT(log) (3 * (log->l_logsize >> 4))
Dave Chinnerdf806152010-05-17 15:52:13 +1000477
478/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 * The reservation head lsn is not made up of a cycle number and block number.
480 * Instead, it uses a cycle number and byte number. Logs don't expect to
481 * overflow 31 bits worth of byte offset, so using a byte number will mean
482 * that round off problems won't occur when releasing partial reservations.
483 */
484typedef struct log {
David Chinner4679b2d2008-04-10 12:18:54 +1000485 /* The following fields don't need locking */
486 struct xfs_mount *l_mp; /* mount point */
David Chinnera9c21c12008-10-30 17:39:35 +1100487 struct xfs_ail *l_ailp; /* AIL log is working with */
Dave Chinner71e330b2010-05-21 14:37:18 +1000488 struct xfs_cil *l_cilp; /* CIL log is working with */
David Chinner4679b2d2008-04-10 12:18:54 +1000489 struct xfs_buf *l_xbuf; /* extra buffer for log
490 * wrapping */
491 struct xfs_buftarg *l_targ; /* buftarg of log */
492 uint l_flags;
493 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
494 struct xfs_buf_cancel **l_buf_cancel_table;
495 int l_iclog_hsize; /* size of iclog header */
496 int l_iclog_heads; /* # of iclog header sectors */
Alex Elder48389ef2010-04-20 17:10:21 +1000497 uint l_sectBBsize; /* sector size in BBs (2^n) */
David Chinner4679b2d2008-04-10 12:18:54 +1000498 int l_iclog_size; /* size of log in bytes */
499 int l_iclog_size_log; /* log power size of log */
500 int l_iclog_bufs; /* number of iclog buffers */
501 xfs_daddr_t l_logBBstart; /* start block of log */
502 int l_logsize; /* size of log in bytes */
503 int l_logBBsize; /* size of log in BB chunks */
504
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 /* The following block of fields are changed while holding icloglock */
Matthew Wilcoxd748c622008-05-19 16:34:27 +1000506 sv_t l_flush_wait ____cacheline_aligned_in_smp;
507 /* waiting for iclog flush */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 int l_covered_state;/* state of "covering disk
509 * log entries" */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 xlog_in_core_t *l_iclog; /* head log queue */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000511 spinlock_t l_icloglock; /* grab to change iclog state */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 xfs_lsn_t l_tail_lsn; /* lsn of 1st LR with unflushed
513 * buffers */
514 xfs_lsn_t l_last_sync_lsn;/* lsn of last LR on disk */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 int l_curr_cycle; /* Cycle number of log writes */
516 int l_prev_cycle; /* Cycle number before last
517 * block increment */
518 int l_curr_block; /* current logical log block */
519 int l_prev_block; /* previous logical log block */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521 /* The following block of fields are changed while holding grant_lock */
David Chinner4679b2d2008-04-10 12:18:54 +1000522 spinlock_t l_grant_lock ____cacheline_aligned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 xlog_ticket_t *l_reserve_headq;
524 xlog_ticket_t *l_write_headq;
525 int l_grant_reserve_cycle;
526 int l_grant_reserve_bytes;
527 int l_grant_write_cycle;
528 int l_grant_write_bytes;
529
David Chinner4679b2d2008-04-10 12:18:54 +1000530 /* The following field are used for debugging; need to hold icloglock */
531#ifdef DEBUG
532 char *l_iclog_bak[XLOG_MAX_ICLOGS];
533#endif
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535} xlog_t;
536
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100537#define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
538
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539/* common routines */
540extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp);
Eric Sandeen65be6052006-01-11 15:34:19 +1100541extern int xlog_recover(xlog_t *log);
Christoph Hellwig42490232008-08-13 16:49:32 +1000542extern int xlog_recover_finish(xlog_t *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Dave Chinner71e330b2010-05-21 14:37:18 +1000545extern kmem_zone_t *xfs_log_ticket_zone;
546struct xlog_ticket *xlog_ticket_alloc(struct log *log, int unit_bytes,
547 int count, char client, uint xflags,
548 int alloc_flags);
549
David Chinnereb01c9c2008-04-10 12:18:46 +1000550
Christoph Hellwige6b1f272010-03-23 11:47:38 +1100551static inline void
552xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
553{
554 *ptr += bytes;
555 *len -= bytes;
556 *off += bytes;
557}
558
Dave Chinner71e330b2010-05-21 14:37:18 +1000559void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
560int xlog_write(struct log *log, struct xfs_log_vec *log_vector,
561 struct xlog_ticket *tic, xfs_lsn_t *start_lsn,
562 xlog_in_core_t **commit_iclog, uint flags);
563
564/*
565 * Committed Item List interfaces
566 */
567int xlog_cil_init(struct log *log);
568void xlog_cil_init_post_recovery(struct log *log);
569void xlog_cil_destroy(struct log *log);
570
Dave Chinnera44f13e2010-08-24 11:40:03 +1000571/*
572 * CIL force routines
573 */
574xfs_lsn_t xlog_cil_force_lsn(struct log *log, xfs_lsn_t sequence);
575
576static inline void
577xlog_cil_force(struct log *log)
578{
579 xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
580}
Dave Chinner71e330b2010-05-21 14:37:18 +1000581
Tim Shimmin955e47a2006-09-28 11:04:16 +1000582/*
583 * Unmount record type is used as a pseudo transaction type for the ticket.
584 * It's value must be outside the range of XFS_TRANS_* values.
585 */
586#define XLOG_UNMOUNT_REC_TYPE (-1U)
587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588#endif /* __KERNEL__ */
589
590#endif /* __XFS_LOG_PRIV_H__ */