blob: 18a801d76a4234e550427a83b6d6a65b7092d670 [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;
Mark Tinguelyad223e62012-06-14 09:22:15 -050022struct xlog;
Nathan Scotta844f452005-11-02 14:38:42 +110023struct xlog_ticket;
Linus Torvalds1da177e2005-04-16 15:20:36 -070024struct xfs_mount;
25
26/*
27 * Macros, structures, prototypes for internal log manager use.
28 */
29
30#define XLOG_MIN_ICLOGS 2
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#define XLOG_MAX_ICLOGS 8
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#define XLOG_HEADER_MAGIC_NUM 0xFEEDbabe /* Invalid cycle number */
33#define XLOG_VERSION_1 1
34#define XLOG_VERSION_2 2 /* Large IClogs, Log sunit */
35#define XLOG_VERSION_OKBITS (XLOG_VERSION_1 | XLOG_VERSION_2)
Eric Sandeendcb3b832007-08-16 16:25:33 +100036#define XLOG_MIN_RECORD_BSIZE (16*1024) /* eventually 32k */
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#define XLOG_BIG_RECORD_BSIZE (32*1024) /* 32k buffers */
38#define XLOG_MAX_RECORD_BSIZE (256*1024)
39#define XLOG_HEADER_CYCLE_SIZE (32*1024) /* cycle data in header */
Eric Sandeendcb3b832007-08-16 16:25:33 +100040#define XLOG_MIN_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#define XLOG_BIG_RECORD_BSHIFT 15 /* 32k == 1 << 15 */
42#define XLOG_MAX_RECORD_BSHIFT 18 /* 256k == 1 << 18 */
43#define XLOG_BTOLSUNIT(log, b) (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
44 (log)->l_mp->m_sb.sb_logsunit)
45#define XLOG_LSUNITTOB(log, su) ((su) * (log)->l_mp->m_sb.sb_logsunit)
46
47#define XLOG_HEADER_SIZE 512
48
49#define XLOG_REC_SHIFT(log) \
Eric Sandeen62118702008-03-06 13:44:28 +110050 BTOBB(1 << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
Linus Torvalds1da177e2005-04-16 15:20:36 -070051 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
52#define XLOG_TOTAL_REC_SHIFT(log) \
Eric Sandeen62118702008-03-06 13:44:28 +110053 BTOBB(XLOG_MAX_ICLOGS << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
55
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100056static inline xfs_lsn_t xlog_assign_lsn(uint cycle, uint block)
57{
58 return ((xfs_lsn_t)cycle << 32) | block;
59}
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100061static inline uint xlog_get_cycle(char *ptr)
62{
Christoph Hellwigb53e6752007-10-12 10:59:34 +100063 if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
64 return be32_to_cpu(*((__be32 *)ptr + 1));
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100065 else
Christoph Hellwigb53e6752007-10-12 10:59:34 +100066 return be32_to_cpu(*(__be32 *)ptr);
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100067}
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69#define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1)
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#ifdef __KERNEL__
72
73/*
74 * get client id from packed copy.
75 *
76 * this hack is here because the xlog_pack code copies four bytes
77 * of xlog_op_header containing the fields oh_clientid, oh_flags
78 * and oh_res2 into the packed copy.
79 *
80 * later on this four byte chunk is treated as an int and the
81 * client id is pulled out.
82 *
83 * this has endian issues, of course.
84 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +100085static inline uint xlog_get_client_id(__be32 i)
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100086{
Christoph Hellwigb53e6752007-10-12 10:59:34 +100087 return be32_to_cpu(i) >> 24;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100088}
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Linus Torvalds1da177e2005-04-16 15:20:36 -070090/*
91 * In core log state
92 */
93#define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
94#define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
95#define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
96#define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
97#define XLOG_STATE_DO_CALLBACK \
98 0x0010 /* Process callback functions */
99#define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
100#define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
101#define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
102#define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
103#define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
104#endif /* __KERNEL__ */
105
106/*
107 * Flags to log operation header
108 *
109 * The first write of a new transaction will be preceded with a start
110 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
111 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
112 * the remainder of the current active in-core log, it is split up into
113 * multiple regions. Each partial region will be marked with a
114 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
115 *
116 */
117#define XLOG_START_TRANS 0x01 /* Start a new transaction */
118#define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
119#define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
120#define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
121#define XLOG_END_TRANS 0x10 /* End a continued transaction */
122#define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124#ifdef __KERNEL__
125/*
126 * Flags to log ticket
127 */
128#define XLOG_TIC_INITED 0x1 /* has been initialized */
129#define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000130
131#define XLOG_TIC_FLAGS \
132 { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \
Dave Chinner10547942010-12-21 12:02:25 +1100133 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135#endif /* __KERNEL__ */
136
137#define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
138
139/*
140 * Flags for log structure
141 */
142#define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
143#define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
144#define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
145#define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
146 shutdown */
Dave Chinnerda8a1a42011-04-08 12:45:07 +1000147#define XLOG_TAIL_WARN 0x10 /* log tail verify warning issued */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Christoph Hellwig97d3ac72011-04-24 19:06:16 +0000149typedef __uint32_t xlog_tid_t;
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151#ifdef __KERNEL__
152/*
153 * Below are states for covering allocation transactions.
154 * By covering, we mean changing the h_tail_lsn in the last on-disk
155 * log write such that no allocation transactions will be re-done during
156 * recovery after a system crash. Recovery starts at the last on-disk
157 * log write.
158 *
159 * These states are used to insert dummy log entries to cover
160 * space allocation transactions which can undo non-transactional changes
161 * after a crash. Writes to a file with space
162 * already allocated do not result in any transactions. Allocations
163 * might include space beyond the EOF. So if we just push the EOF a
164 * little, the last transaction for the file could contain the wrong
165 * size. If there is no file system activity, after an allocation
166 * transaction, and the system crashes, the allocation transaction
167 * will get replayed and the file will be truncated. This could
168 * be hours/days/... after the allocation occurred.
169 *
170 * The fix for this is to do two dummy transactions when the
171 * system is idle. We need two dummy transaction because the h_tail_lsn
172 * in the log record header needs to point beyond the last possible
173 * non-dummy transaction. The first dummy changes the h_tail_lsn to
174 * the first transaction before the dummy. The second dummy causes
175 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
176 *
177 * These dummy transactions get committed when everything
178 * is idle (after there has been some activity).
179 *
180 * There are 5 states used to control this.
181 *
182 * IDLE -- no logging has been done on the file system or
183 * we are done covering previous transactions.
184 * NEED -- logging has occurred and we need a dummy transaction
185 * when the log becomes idle.
186 * DONE -- we were in the NEED state and have committed a dummy
187 * transaction.
188 * NEED2 -- we detected that a dummy transaction has gone to the
189 * on disk log with no other transactions.
190 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
191 *
192 * There are two places where we switch states:
193 *
194 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
195 * We commit the dummy transaction and switch to DONE or DONE2,
196 * respectively. In all other states, we don't do anything.
197 *
198 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
199 *
200 * No matter what state we are in, if this isn't the dummy
201 * transaction going out, the next state is NEED.
202 * So, if we aren't in the DONE or DONE2 states, the next state
203 * is NEED. We can't be finishing a write of the dummy record
204 * unless it was committed and the state switched to DONE or DONE2.
205 *
206 * If we are in the DONE state and this was a write of the
207 * dummy transaction, we move to NEED2.
208 *
209 * If we are in the DONE2 state and this was a write of the
210 * dummy transaction, we move to IDLE.
211 *
212 *
213 * Writing only one dummy transaction can get appended to
214 * one file space allocation. When this happens, the log recovery
215 * code replays the space allocation and a file could be truncated.
216 * This is why we have the NEED2 and DONE2 states before going idle.
217 */
218
219#define XLOG_STATE_COVER_IDLE 0
220#define XLOG_STATE_COVER_NEED 1
221#define XLOG_STATE_COVER_DONE 2
222#define XLOG_STATE_COVER_NEED2 3
223#define XLOG_STATE_COVER_DONE2 4
224
225#define XLOG_COVER_OPS 5
226
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000227
228/* Ticket reservation region accounting */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000229#define XLOG_TIC_LEN_MAX 15
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000230
231/*
232 * Reservation region
233 * As would be stored in xfs_log_iovec but without the i_addr which
234 * we don't care about.
235 */
236typedef struct xlog_res {
Tim Shimmin12598452006-01-11 21:02:47 +1100237 uint r_len; /* region length :4 */
238 uint r_type; /* region's transaction type :4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000239} xlog_res_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241typedef struct xlog_ticket {
Dave Chinner10547942010-12-21 12:02:25 +1100242 struct list_head t_queue; /* reserve/write queue */
Christoph Hellwig14a72352012-02-20 02:31:24 +0000243 struct task_struct *t_task; /* task that owns this ticket */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000244 xlog_tid_t t_tid; /* transaction identifier : 4 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100245 atomic_t t_ref; /* ticket reference count : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000246 int t_curr_res; /* current reservation in bytes : 4 */
247 int t_unit_res; /* unit reservation in bytes : 4 */
248 char t_ocnt; /* original count : 1 */
249 char t_cnt; /* current count : 1 */
250 char t_clientid; /* who does this belong to; : 1 */
251 char t_flags; /* properties of reservation : 1 */
252 uint t_trans_type; /* transaction type : 4 */
253
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000254 /* reservation array fields */
255 uint t_res_num; /* num in array : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000256 uint t_res_num_ophdrs; /* num op hdrs : 4 */
257 uint t_res_arr_sum; /* array sum : 4 */
258 uint t_res_o_flow; /* sum overflow : 4 */
Tim Shimmin12598452006-01-11 21:02:47 +1100259 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260} xlog_ticket_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262#endif
263
264
265typedef struct xlog_op_header {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +1000266 __be32 oh_tid; /* transaction id of operation : 4 b */
267 __be32 oh_len; /* bytes in data region : 4 b */
268 __u8 oh_clientid; /* who sent me this : 1 b */
269 __u8 oh_flags; /* : 1 b */
270 __u16 oh_res2; /* 32 bit align : 2 b */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271} xlog_op_header_t;
272
273
274/* valid values for h_fmt */
275#define XLOG_FMT_UNKNOWN 0
276#define XLOG_FMT_LINUX_LE 1
277#define XLOG_FMT_LINUX_BE 2
278#define XLOG_FMT_IRIX_BE 3
279
280/* our fmt */
Nathan Scottf016bad2005-09-08 15:30:05 +1000281#ifdef XFS_NATIVE_HOST
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282#define XLOG_FMT XLOG_FMT_LINUX_BE
283#else
Nathan Scottf016bad2005-09-08 15:30:05 +1000284#define XLOG_FMT XLOG_FMT_LINUX_LE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285#endif
286
287typedef struct xlog_rec_header {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000288 __be32 h_magicno; /* log record (LR) identifier : 4 */
289 __be32 h_cycle; /* write cycle of log : 4 */
290 __be32 h_version; /* LR version : 4 */
291 __be32 h_len; /* len in bytes; should be 64-bit aligned: 4 */
292 __be64 h_lsn; /* lsn of this LR : 8 */
293 __be64 h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
294 __be32 h_chksum; /* may not be used; non-zero if used : 4 */
295 __be32 h_prev_block; /* block number to previous LR : 4 */
296 __be32 h_num_logops; /* number of log operations in this LR : 4 */
297 __be32 h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000299 __be32 h_fmt; /* format of log record : 4 */
300 uuid_t h_fs_uuid; /* uuid of FS : 16 */
301 __be32 h_size; /* iclog size : 4 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302} xlog_rec_header_t;
303
304typedef struct xlog_rec_ext_header {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000305 __be32 xh_cycle; /* write cycle of log : 4 */
306 __be32 xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307} xlog_rec_ext_header_t;
308
309#ifdef __KERNEL__
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100310
311/*
312 * Quite misnamed, because this union lays out the actual on-disk log buffer.
313 */
314typedef union xlog_in_core2 {
315 xlog_rec_header_t hic_header;
316 xlog_rec_ext_header_t hic_xheader;
317 char hic_sector[XLOG_HEADER_SIZE];
318} xlog_in_core_2_t;
319
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320/*
321 * - A log record header is 512 bytes. There is plenty of room to grow the
322 * xlog_rec_header_t into the reserved space.
323 * - ic_data follows, so a write to disk can start at the beginning of
324 * the iclog.
David Chinner12017fa2008-08-13 16:34:31 +1000325 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 * - ic_next is the pointer to the next iclog in the ring.
327 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
328 * - ic_log is a pointer back to the global log structure.
329 * - ic_callback is a linked list of callback function/argument pairs to be
330 * called after an iclog finishes writing.
331 * - ic_size is the full size of the header plus data.
332 * - ic_offset is the current number of bytes written to in this iclog.
333 * - ic_refcnt is bumped when someone is writing to the log.
334 * - ic_state is the state of the iclog.
David Chinner114d23a2008-04-10 12:18:39 +1000335 *
336 * Because of cacheline contention on large machines, we need to separate
337 * various resources onto different cachelines. To start with, make the
338 * structure cacheline aligned. The following fields can be contended on
339 * by independent processes:
340 *
341 * - ic_callback_*
342 * - ic_refcnt
343 * - fields protected by the global l_icloglock
344 *
345 * so we need to ensure that these fields are located in separate cachelines.
346 * We'll put all the read-only and l_icloglock fields in the first cacheline,
347 * and move everything else out to subsequent cachelines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 */
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100349typedef struct xlog_in_core {
Dave Chinnereb40a872010-12-21 12:09:01 +1100350 wait_queue_head_t ic_force_wait;
351 wait_queue_head_t ic_write_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 struct xlog_in_core *ic_next;
353 struct xlog_in_core *ic_prev;
354 struct xfs_buf *ic_bp;
Mark Tinguelyad223e62012-06-14 09:22:15 -0500355 struct xlog *ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 int ic_size;
357 int ic_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 int ic_bwritecnt;
Christoph Hellwiga5687782009-02-09 08:37:39 +0100359 unsigned short ic_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 char *ic_datap; /* pointer to iclog data */
David Chinner114d23a2008-04-10 12:18:39 +1000361
362 /* Callback structures need their own cacheline */
363 spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
364 xfs_log_callback_t *ic_callback;
365 xfs_log_callback_t **ic_callback_tail;
366
367 /* reference counts need their own cacheline */
368 atomic_t ic_refcnt ____cacheline_aligned_in_smp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100369 xlog_in_core_2_t *ic_data;
370#define ic_header ic_data->hic_header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371} xlog_in_core_t;
372
373/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000374 * The CIL context is used to aggregate per-transaction details as well be
375 * passed to the iclog for checkpoint post-commit processing. After being
376 * passed to the iclog, another context needs to be allocated for tracking the
377 * next set of transactions to be aggregated into a checkpoint.
378 */
379struct xfs_cil;
380
381struct xfs_cil_ctx {
382 struct xfs_cil *cil;
383 xfs_lsn_t sequence; /* chkpt sequence # */
384 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
385 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
386 struct xlog_ticket *ticket; /* chkpt ticket */
387 int nvecs; /* number of regions */
388 int space_used; /* aggregate size of regions */
389 struct list_head busy_extents; /* busy extents in chkpt */
390 struct xfs_log_vec *lv_chain; /* logvecs being pushed */
391 xfs_log_callback_t log_cb; /* completion callback hook. */
392 struct list_head committing; /* ctx committing list */
393};
394
395/*
396 * Committed Item List structure
397 *
398 * This structure is used to track log items that have been committed but not
399 * yet written into the log. It is used only when the delayed logging mount
400 * option is enabled.
401 *
402 * This structure tracks the list of committing checkpoint contexts so
403 * we can avoid the problem of having to hold out new transactions during a
404 * flush until we have a the commit record LSN of the checkpoint. We can
405 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
406 * sequence match and extract the commit LSN directly from there. If the
407 * checkpoint is still in the process of committing, we can block waiting for
408 * the commit LSN to be determined as well. This should make synchronous
409 * operations almost as efficient as the old logging methods.
410 */
411struct xfs_cil {
Mark Tinguelyad223e62012-06-14 09:22:15 -0500412 struct xlog *xc_log;
Dave Chinner71e330b2010-05-21 14:37:18 +1000413 struct list_head xc_cil;
414 spinlock_t xc_cil_lock;
415 struct xfs_cil_ctx *xc_ctx;
416 struct rw_semaphore xc_ctx_lock;
417 struct list_head xc_committing;
Dave Chinnereb40a872010-12-21 12:09:01 +1100418 wait_queue_head_t xc_commit_wait;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000419 xfs_lsn_t xc_current_sequence;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000420 struct work_struct xc_push_work;
421 xfs_lsn_t xc_push_seq;
Dave Chinner71e330b2010-05-21 14:37:18 +1000422};
423
424/*
Dave Chinner80168672010-09-24 18:13:44 +1000425 * The amount of log space we allow the CIL to aggregate is difficult to size.
426 * Whatever we choose, we have to make sure we can get a reservation for the
427 * log space effectively, that it is large enough to capture sufficient
428 * relogging to reduce log buffer IO significantly, but it is not too large for
429 * the log or induces too much latency when writing out through the iclogs. We
430 * track both space consumed and the number of vectors in the checkpoint
431 * context, so we need to decide which to use for limiting.
Dave Chinnerdf806152010-05-17 15:52:13 +1000432 *
433 * Every log buffer we write out during a push needs a header reserved, which
434 * is at least one sector and more for v2 logs. Hence we need a reservation of
435 * at least 512 bytes per 32k of log space just for the LR headers. That means
436 * 16KB of reservation per megabyte of delayed logging space we will consume,
437 * plus various headers. The number of headers will vary based on the num of
438 * io vectors, so limiting on a specific number of vectors is going to result
439 * in transactions of varying size. IOWs, it is more consistent to track and
440 * limit space consumed in the log rather than by the number of objects being
441 * logged in order to prevent checkpoint ticket overruns.
442 *
443 * Further, use of static reservations through the log grant mechanism is
444 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
445 * grant) and a significant deadlock potential because regranting write space
446 * can block on log pushes. Hence if we have to regrant log space during a log
447 * push, we can deadlock.
448 *
449 * However, we can avoid this by use of a dynamic "reservation stealing"
450 * technique during transaction commit whereby unused reservation space in the
451 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
452 * space needed by the checkpoint transaction. This means that we never need to
453 * specifically reserve space for the CIL checkpoint transaction, nor do we
454 * need to regrant space once the checkpoint completes. This also means the
455 * checkpoint transaction ticket is specific to the checkpoint context, rather
456 * than the CIL itself.
457 *
Dave Chinner80168672010-09-24 18:13:44 +1000458 * With dynamic reservations, we can effectively make up arbitrary limits for
459 * the checkpoint size so long as they don't violate any other size rules.
460 * Recovery imposes a rule that no transaction exceed half the log, so we are
461 * limited by that. Furthermore, the log transaction reservation subsystem
462 * tries to keep 25% of the log free, so we need to keep below that limit or we
463 * risk running out of free log space to start any new transactions.
464 *
465 * In order to keep background CIL push efficient, we will set a lower
466 * threshold at which background pushing is attempted without blocking current
467 * transaction commits. A separate, higher bound defines when CIL pushes are
468 * enforced to ensure we stay within our maximum checkpoint size bounds.
469 * threshold, yet give us plenty of space for aggregation on large logs.
Dave Chinnerdf806152010-05-17 15:52:13 +1000470 */
Dave Chinner80168672010-09-24 18:13:44 +1000471#define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3)
472#define XLOG_CIL_HARD_SPACE_LIMIT(log) (3 * (log->l_logsize >> 4))
Dave Chinnerdf806152010-05-17 15:52:13 +1000473
474/*
Christoph Hellwig28496962012-02-20 02:31:25 +0000475 * ticket grant locks, queues and accounting have their own cachlines
476 * as these are quite hot and can be operated on concurrently.
477 */
478struct xlog_grant_head {
479 spinlock_t lock ____cacheline_aligned_in_smp;
480 struct list_head waiters;
481 atomic64_t grant;
482};
483
484/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 * The reservation head lsn is not made up of a cycle number and block number.
486 * Instead, it uses a cycle number and byte number. Logs don't expect to
487 * overflow 31 bits worth of byte offset, so using a byte number will mean
488 * that round off problems won't occur when releasing partial reservations.
489 */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500490struct xlog {
David Chinner4679b2d2008-04-10 12:18:54 +1000491 /* The following fields don't need locking */
492 struct xfs_mount *l_mp; /* mount point */
David Chinnera9c21c12008-10-30 17:39:35 +1100493 struct xfs_ail *l_ailp; /* AIL log is working with */
Dave Chinner71e330b2010-05-21 14:37:18 +1000494 struct xfs_cil *l_cilp; /* CIL log is working with */
David Chinner4679b2d2008-04-10 12:18:54 +1000495 struct xfs_buf *l_xbuf; /* extra buffer for log
496 * wrapping */
497 struct xfs_buftarg *l_targ; /* buftarg of log */
498 uint l_flags;
499 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000500 struct list_head *l_buf_cancel_table;
David Chinner4679b2d2008-04-10 12:18:54 +1000501 int l_iclog_hsize; /* size of iclog header */
502 int l_iclog_heads; /* # of iclog header sectors */
Alex Elder48389ef2010-04-20 17:10:21 +1000503 uint l_sectBBsize; /* sector size in BBs (2^n) */
David Chinner4679b2d2008-04-10 12:18:54 +1000504 int l_iclog_size; /* size of log in bytes */
505 int l_iclog_size_log; /* log power size of log */
506 int l_iclog_bufs; /* number of iclog buffers */
507 xfs_daddr_t l_logBBstart; /* start block of log */
508 int l_logsize; /* size of log in bytes */
509 int l_logBBsize; /* size of log in BB chunks */
510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 /* The following block of fields are changed while holding icloglock */
Dave Chinnereb40a872010-12-21 12:09:01 +1100512 wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp;
Matthew Wilcoxd748c622008-05-19 16:34:27 +1000513 /* waiting for iclog flush */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 int l_covered_state;/* state of "covering disk
515 * log entries" */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 xlog_in_core_t *l_iclog; /* head log queue */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000517 spinlock_t l_icloglock; /* grab to change iclog state */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 int l_curr_cycle; /* Cycle number of log writes */
519 int l_prev_cycle; /* Cycle number before last
520 * block increment */
521 int l_curr_block; /* current logical log block */
522 int l_prev_block; /* previous logical log block */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
Dave Chinner84f3c682010-12-03 22:11:29 +1100524 /*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100525 * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
526 * read without needing to hold specific locks. To avoid operations
527 * contending with other hot objects, place each of them on a separate
528 * cacheline.
Dave Chinner84f3c682010-12-03 22:11:29 +1100529 */
530 /* lsn of last LR on disk */
531 atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100532 /* lsn of 1st LR with unflushed * buffers */
533 atomic64_t l_tail_lsn ____cacheline_aligned_in_smp;
Dave Chinner84f3c682010-12-03 22:11:29 +1100534
Christoph Hellwig28496962012-02-20 02:31:25 +0000535 struct xlog_grant_head l_reserve_head;
536 struct xlog_grant_head l_write_head;
Dave Chinner3f16b982010-12-21 12:29:01 +1100537
David Chinner4679b2d2008-04-10 12:18:54 +1000538 /* The following field are used for debugging; need to hold icloglock */
539#ifdef DEBUG
540 char *l_iclog_bak[XLOG_MAX_ICLOGS];
541#endif
542
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500543};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000545#define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
546 ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE))
547
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100548#define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550/* common routines */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500551extern int
552xlog_recover(
553 struct xlog *log);
554extern int
555xlog_recover_finish(
556 struct xlog *log);
557extern void
558xlog_pack_data(
559 struct xlog *log,
560 struct xlog_in_core *iclog,
561 int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Dave Chinner71e330b2010-05-21 14:37:18 +1000563extern kmem_zone_t *xfs_log_ticket_zone;
Mark Tinguelyad223e62012-06-14 09:22:15 -0500564struct xlog_ticket *
565xlog_ticket_alloc(
566 struct xlog *log,
567 int unit_bytes,
568 int count,
569 char client,
570 bool permanent,
571 xfs_km_flags_t alloc_flags);
Dave Chinner71e330b2010-05-21 14:37:18 +1000572
David Chinnereb01c9c2008-04-10 12:18:46 +1000573
Christoph Hellwige6b1f272010-03-23 11:47:38 +1100574static inline void
575xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
576{
577 *ptr += bytes;
578 *len -= bytes;
579 *off += bytes;
580}
581
Dave Chinner71e330b2010-05-21 14:37:18 +1000582void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
Mark Tinguelyad223e62012-06-14 09:22:15 -0500583int
584xlog_write(
585 struct xlog *log,
586 struct xfs_log_vec *log_vector,
587 struct xlog_ticket *tic,
588 xfs_lsn_t *start_lsn,
589 struct xlog_in_core **commit_iclog,
590 uint flags);
Dave Chinner71e330b2010-05-21 14:37:18 +1000591
592/*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100593 * When we crack an atomic LSN, we sample it first so that the value will not
594 * change while we are cracking it into the component values. This means we
595 * will always get consistent component values to work from. This should always
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300596 * be used to sample and crack LSNs that are stored and updated in atomic
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100597 * variables.
598 */
599static inline void
600xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
601{
602 xfs_lsn_t val = atomic64_read(lsn);
603
604 *cycle = CYCLE_LSN(val);
605 *block = BLOCK_LSN(val);
606}
607
608/*
609 * Calculate and assign a value to an atomic LSN variable from component pieces.
610 */
611static inline void
612xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
613{
614 atomic64_set(lsn, xlog_assign_lsn(cycle, block));
615}
616
617/*
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100618 * When we crack the grant head, we sample it first so that the value will not
Dave Chinnera69ed032010-12-21 12:08:20 +1100619 * change while we are cracking it into the component values. This means we
620 * will always get consistent component values to work from.
621 */
622static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100623xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100624{
Dave Chinnera69ed032010-12-21 12:08:20 +1100625 *cycle = val >> 32;
626 *space = val & 0xffffffff;
627}
628
629static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100630xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
631{
632 xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
633}
634
635static inline int64_t
636xlog_assign_grant_head_val(int cycle, int space)
637{
638 return ((int64_t)cycle << 32) | space;
639}
640
641static inline void
Dave Chinnerc8a09ff2010-12-04 00:02:40 +1100642xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100643{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100644 atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
Dave Chinnera69ed032010-12-21 12:08:20 +1100645}
646
647/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000648 * Committed Item List interfaces
649 */
Mark Tinguelyad223e62012-06-14 09:22:15 -0500650int
651xlog_cil_init(struct xlog *log);
652void
653xlog_cil_init_post_recovery(struct xlog *log);
654void
655xlog_cil_destroy(struct xlog *log);
Dave Chinner71e330b2010-05-21 14:37:18 +1000656
Dave Chinnera44f13e2010-08-24 11:40:03 +1000657/*
658 * CIL force routines
659 */
Mark Tinguelyad223e62012-06-14 09:22:15 -0500660xfs_lsn_t
661xlog_cil_force_lsn(
662 struct xlog *log,
663 xfs_lsn_t sequence);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000664
665static inline void
Mark Tinguelyad223e62012-06-14 09:22:15 -0500666xlog_cil_force(struct xlog *log)
Dave Chinnera44f13e2010-08-24 11:40:03 +1000667{
668 xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
669}
Dave Chinner71e330b2010-05-21 14:37:18 +1000670
Tim Shimmin955e47a2006-09-28 11:04:16 +1000671/*
672 * Unmount record type is used as a pseudo transaction type for the ticket.
673 * It's value must be outside the range of XFS_TRANS_* values.
674 */
675#define XLOG_UNMOUNT_REC_TYPE (-1U)
676
Dave Chinnereb40a872010-12-21 12:09:01 +1100677/*
678 * Wrapper function for waiting on a wait queue serialised against wakeups
679 * by a spinlock. This matches the semantics of all the wait queues used in the
680 * log code.
681 */
682static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
683{
684 DECLARE_WAITQUEUE(wait, current);
685
686 add_wait_queue_exclusive(wq, &wait);
687 __set_current_state(TASK_UNINTERRUPTIBLE);
688 spin_unlock(lock);
689 schedule();
690 remove_wait_queue(wq, &wait);
691}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692#endif /* __KERNEL__ */
693
694#endif /* __XFS_LOG_PRIV_H__ */