blob: d5f8be8f4bf603cac8c3f0394759e5fefdd8d42e [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_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#define xlog_panic(args...) cmn_err(CE_PANIC, ## args)
91#define xlog_exit(args...) cmn_err(CE_PANIC, ## args)
92#define xlog_warn(args...) cmn_err(CE_WARN, ## args)
93
94/*
95 * In core log state
96 */
97#define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
98#define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
99#define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
100#define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
101#define XLOG_STATE_DO_CALLBACK \
102 0x0010 /* Process callback functions */
103#define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
104#define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
105#define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
106#define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
107#define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
108#endif /* __KERNEL__ */
109
110/*
111 * Flags to log operation header
112 *
113 * The first write of a new transaction will be preceded with a start
114 * record, XLOG_START_TRANS. Once a transaction is committed, a commit
115 * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into
116 * the remainder of the current active in-core log, it is split up into
117 * multiple regions. Each partial region will be marked with a
118 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
119 *
120 */
121#define XLOG_START_TRANS 0x01 /* Start a new transaction */
122#define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */
123#define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */
124#define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */
125#define XLOG_END_TRANS 0x10 /* End a continued transaction */
126#define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
128#ifdef __KERNEL__
129/*
130 * Flags to log ticket
131 */
132#define XLOG_TIC_INITED 0x1 /* has been initialized */
133#define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000134
135#define XLOG_TIC_FLAGS \
136 { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \
Dave Chinner10547942010-12-21 12:02:25 +1100137 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139#endif /* __KERNEL__ */
140
141#define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */
142
143/*
144 * Flags for log structure
145 */
146#define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */
147#define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
148#define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
149#define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
150 shutdown */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
152#ifdef __KERNEL__
153/*
154 * Below are states for covering allocation transactions.
155 * By covering, we mean changing the h_tail_lsn in the last on-disk
156 * log write such that no allocation transactions will be re-done during
157 * recovery after a system crash. Recovery starts at the last on-disk
158 * log write.
159 *
160 * These states are used to insert dummy log entries to cover
161 * space allocation transactions which can undo non-transactional changes
162 * after a crash. Writes to a file with space
163 * already allocated do not result in any transactions. Allocations
164 * might include space beyond the EOF. So if we just push the EOF a
165 * little, the last transaction for the file could contain the wrong
166 * size. If there is no file system activity, after an allocation
167 * transaction, and the system crashes, the allocation transaction
168 * will get replayed and the file will be truncated. This could
169 * be hours/days/... after the allocation occurred.
170 *
171 * The fix for this is to do two dummy transactions when the
172 * system is idle. We need two dummy transaction because the h_tail_lsn
173 * in the log record header needs to point beyond the last possible
174 * non-dummy transaction. The first dummy changes the h_tail_lsn to
175 * the first transaction before the dummy. The second dummy causes
176 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
177 *
178 * These dummy transactions get committed when everything
179 * is idle (after there has been some activity).
180 *
181 * There are 5 states used to control this.
182 *
183 * IDLE -- no logging has been done on the file system or
184 * we are done covering previous transactions.
185 * NEED -- logging has occurred and we need a dummy transaction
186 * when the log becomes idle.
187 * DONE -- we were in the NEED state and have committed a dummy
188 * transaction.
189 * NEED2 -- we detected that a dummy transaction has gone to the
190 * on disk log with no other transactions.
191 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
192 *
193 * There are two places where we switch states:
194 *
195 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
196 * We commit the dummy transaction and switch to DONE or DONE2,
197 * respectively. In all other states, we don't do anything.
198 *
199 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
200 *
201 * No matter what state we are in, if this isn't the dummy
202 * transaction going out, the next state is NEED.
203 * So, if we aren't in the DONE or DONE2 states, the next state
204 * is NEED. We can't be finishing a write of the dummy record
205 * unless it was committed and the state switched to DONE or DONE2.
206 *
207 * If we are in the DONE state and this was a write of the
208 * dummy transaction, we move to NEED2.
209 *
210 * If we are in the DONE2 state and this was a write of the
211 * dummy transaction, we move to IDLE.
212 *
213 *
214 * Writing only one dummy transaction can get appended to
215 * one file space allocation. When this happens, the log recovery
216 * code replays the space allocation and a file could be truncated.
217 * This is why we have the NEED2 and DONE2 states before going idle.
218 */
219
220#define XLOG_STATE_COVER_IDLE 0
221#define XLOG_STATE_COVER_NEED 1
222#define XLOG_STATE_COVER_DONE 2
223#define XLOG_STATE_COVER_NEED2 3
224#define XLOG_STATE_COVER_DONE2 4
225
226#define XLOG_COVER_OPS 5
227
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000228
229/* Ticket reservation region accounting */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000230#define XLOG_TIC_LEN_MAX 15
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000231
232/*
233 * Reservation region
234 * As would be stored in xfs_log_iovec but without the i_addr which
235 * we don't care about.
236 */
237typedef struct xlog_res {
Tim Shimmin12598452006-01-11 21:02:47 +1100238 uint r_len; /* region length :4 */
239 uint r_type; /* region's transaction type :4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000240} xlog_res_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242typedef struct xlog_ticket {
Dave Chinnereb40a872010-12-21 12:09:01 +1100243 wait_queue_head_t t_wait; /* ticket wait queue */
Dave Chinner10547942010-12-21 12:02:25 +1100244 struct list_head t_queue; /* reserve/write queue */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000245 xlog_tid_t t_tid; /* transaction identifier : 4 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100246 atomic_t t_ref; /* ticket reference count : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000247 int t_curr_res; /* current reservation in bytes : 4 */
248 int t_unit_res; /* unit reservation in bytes : 4 */
249 char t_ocnt; /* original count : 1 */
250 char t_cnt; /* current count : 1 */
251 char t_clientid; /* who does this belong to; : 1 */
252 char t_flags; /* properties of reservation : 1 */
253 uint t_trans_type; /* transaction type : 4 */
254
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000255 /* reservation array fields */
256 uint t_res_num; /* num in array : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000257 uint t_res_num_ophdrs; /* num op hdrs : 4 */
258 uint t_res_arr_sum; /* array sum : 4 */
259 uint t_res_o_flow; /* sum overflow : 4 */
Tim Shimmin12598452006-01-11 21:02:47 +1100260 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261} xlog_ticket_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000262
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263#endif
264
265
266typedef struct xlog_op_header {
Christoph Hellwig67fcb7b2007-10-12 10:58:59 +1000267 __be32 oh_tid; /* transaction id of operation : 4 b */
268 __be32 oh_len; /* bytes in data region : 4 b */
269 __u8 oh_clientid; /* who sent me this : 1 b */
270 __u8 oh_flags; /* : 1 b */
271 __u16 oh_res2; /* 32 bit align : 2 b */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272} xlog_op_header_t;
273
274
275/* valid values for h_fmt */
276#define XLOG_FMT_UNKNOWN 0
277#define XLOG_FMT_LINUX_LE 1
278#define XLOG_FMT_LINUX_BE 2
279#define XLOG_FMT_IRIX_BE 3
280
281/* our fmt */
Nathan Scottf016bad2005-09-08 15:30:05 +1000282#ifdef XFS_NATIVE_HOST
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283#define XLOG_FMT XLOG_FMT_LINUX_BE
284#else
Nathan Scottf016bad2005-09-08 15:30:05 +1000285#define XLOG_FMT XLOG_FMT_LINUX_LE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286#endif
287
288typedef struct xlog_rec_header {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000289 __be32 h_magicno; /* log record (LR) identifier : 4 */
290 __be32 h_cycle; /* write cycle of log : 4 */
291 __be32 h_version; /* LR version : 4 */
292 __be32 h_len; /* len in bytes; should be 64-bit aligned: 4 */
293 __be64 h_lsn; /* lsn of this LR : 8 */
294 __be64 h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */
295 __be32 h_chksum; /* may not be used; non-zero if used : 4 */
296 __be32 h_prev_block; /* block number to previous LR : 4 */
297 __be32 h_num_logops; /* number of log operations in this LR : 4 */
298 __be32 h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 /* new fields */
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000300 __be32 h_fmt; /* format of log record : 4 */
301 uuid_t h_fs_uuid; /* uuid of FS : 16 */
302 __be32 h_size; /* iclog size : 4 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303} xlog_rec_header_t;
304
305typedef struct xlog_rec_ext_header {
Christoph Hellwigb53e6752007-10-12 10:59:34 +1000306 __be32 xh_cycle; /* write cycle of log : 4 */
307 __be32 xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308} xlog_rec_ext_header_t;
309
310#ifdef __KERNEL__
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100311
312/*
313 * Quite misnamed, because this union lays out the actual on-disk log buffer.
314 */
315typedef union xlog_in_core2 {
316 xlog_rec_header_t hic_header;
317 xlog_rec_ext_header_t hic_xheader;
318 char hic_sector[XLOG_HEADER_SIZE];
319} xlog_in_core_2_t;
320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321/*
322 * - A log record header is 512 bytes. There is plenty of room to grow the
323 * xlog_rec_header_t into the reserved space.
324 * - ic_data follows, so a write to disk can start at the beginning of
325 * the iclog.
David Chinner12017fa2008-08-13 16:34:31 +1000326 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 * - ic_next is the pointer to the next iclog in the ring.
328 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
329 * - ic_log is a pointer back to the global log structure.
330 * - ic_callback is a linked list of callback function/argument pairs to be
331 * called after an iclog finishes writing.
332 * - ic_size is the full size of the header plus data.
333 * - ic_offset is the current number of bytes written to in this iclog.
334 * - ic_refcnt is bumped when someone is writing to the log.
335 * - ic_state is the state of the iclog.
David Chinner114d23a2008-04-10 12:18:39 +1000336 *
337 * Because of cacheline contention on large machines, we need to separate
338 * various resources onto different cachelines. To start with, make the
339 * structure cacheline aligned. The following fields can be contended on
340 * by independent processes:
341 *
342 * - ic_callback_*
343 * - ic_refcnt
344 * - fields protected by the global l_icloglock
345 *
346 * so we need to ensure that these fields are located in separate cachelines.
347 * We'll put all the read-only and l_icloglock fields in the first cacheline,
348 * and move everything else out to subsequent cachelines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 */
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100350typedef struct xlog_in_core {
Dave Chinnereb40a872010-12-21 12:09:01 +1100351 wait_queue_head_t ic_force_wait;
352 wait_queue_head_t ic_write_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 struct xlog_in_core *ic_next;
354 struct xlog_in_core *ic_prev;
355 struct xfs_buf *ic_bp;
356 struct log *ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 int ic_size;
358 int ic_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 int ic_bwritecnt;
Christoph Hellwiga5687782009-02-09 08:37:39 +0100360 unsigned short ic_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 char *ic_datap; /* pointer to iclog data */
David Chinner114d23a2008-04-10 12:18:39 +1000362
363 /* Callback structures need their own cacheline */
364 spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
365 xfs_log_callback_t *ic_callback;
366 xfs_log_callback_t **ic_callback_tail;
367
368 /* reference counts need their own cacheline */
369 atomic_t ic_refcnt ____cacheline_aligned_in_smp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100370 xlog_in_core_2_t *ic_data;
371#define ic_header ic_data->hic_header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372} xlog_in_core_t;
373
374/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000375 * The CIL context is used to aggregate per-transaction details as well be
376 * passed to the iclog for checkpoint post-commit processing. After being
377 * passed to the iclog, another context needs to be allocated for tracking the
378 * next set of transactions to be aggregated into a checkpoint.
379 */
380struct xfs_cil;
381
382struct xfs_cil_ctx {
383 struct xfs_cil *cil;
384 xfs_lsn_t sequence; /* chkpt sequence # */
385 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
386 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
387 struct xlog_ticket *ticket; /* chkpt ticket */
388 int nvecs; /* number of regions */
389 int space_used; /* aggregate size of regions */
390 struct list_head busy_extents; /* busy extents in chkpt */
391 struct xfs_log_vec *lv_chain; /* logvecs being pushed */
392 xfs_log_callback_t log_cb; /* completion callback hook. */
393 struct list_head committing; /* ctx committing list */
394};
395
396/*
397 * Committed Item List structure
398 *
399 * This structure is used to track log items that have been committed but not
400 * yet written into the log. It is used only when the delayed logging mount
401 * option is enabled.
402 *
403 * This structure tracks the list of committing checkpoint contexts so
404 * we can avoid the problem of having to hold out new transactions during a
405 * flush until we have a the commit record LSN of the checkpoint. We can
406 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
407 * sequence match and extract the commit LSN directly from there. If the
408 * checkpoint is still in the process of committing, we can block waiting for
409 * the commit LSN to be determined as well. This should make synchronous
410 * operations almost as efficient as the old logging methods.
411 */
412struct xfs_cil {
413 struct log *xc_log;
414 struct list_head xc_cil;
415 spinlock_t xc_cil_lock;
416 struct xfs_cil_ctx *xc_ctx;
417 struct rw_semaphore xc_ctx_lock;
418 struct list_head xc_committing;
Dave Chinnereb40a872010-12-21 12:09:01 +1100419 wait_queue_head_t xc_commit_wait;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000420 xfs_lsn_t xc_current_sequence;
Dave Chinner71e330b2010-05-21 14:37:18 +1000421};
422
423/*
Dave Chinner80168672010-09-24 18:13:44 +1000424 * The amount of log space we allow the CIL to aggregate is difficult to size.
425 * Whatever we choose, we have to make sure we can get a reservation for the
426 * log space effectively, that it is large enough to capture sufficient
427 * relogging to reduce log buffer IO significantly, but it is not too large for
428 * the log or induces too much latency when writing out through the iclogs. We
429 * track both space consumed and the number of vectors in the checkpoint
430 * context, so we need to decide which to use for limiting.
Dave Chinnerdf806152010-05-17 15:52:13 +1000431 *
432 * Every log buffer we write out during a push needs a header reserved, which
433 * is at least one sector and more for v2 logs. Hence we need a reservation of
434 * at least 512 bytes per 32k of log space just for the LR headers. That means
435 * 16KB of reservation per megabyte of delayed logging space we will consume,
436 * plus various headers. The number of headers will vary based on the num of
437 * io vectors, so limiting on a specific number of vectors is going to result
438 * in transactions of varying size. IOWs, it is more consistent to track and
439 * limit space consumed in the log rather than by the number of objects being
440 * logged in order to prevent checkpoint ticket overruns.
441 *
442 * Further, use of static reservations through the log grant mechanism is
443 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
444 * grant) and a significant deadlock potential because regranting write space
445 * can block on log pushes. Hence if we have to regrant log space during a log
446 * push, we can deadlock.
447 *
448 * However, we can avoid this by use of a dynamic "reservation stealing"
449 * technique during transaction commit whereby unused reservation space in the
450 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
451 * space needed by the checkpoint transaction. This means that we never need to
452 * specifically reserve space for the CIL checkpoint transaction, nor do we
453 * need to regrant space once the checkpoint completes. This also means the
454 * checkpoint transaction ticket is specific to the checkpoint context, rather
455 * than the CIL itself.
456 *
Dave Chinner80168672010-09-24 18:13:44 +1000457 * With dynamic reservations, we can effectively make up arbitrary limits for
458 * the checkpoint size so long as they don't violate any other size rules.
459 * Recovery imposes a rule that no transaction exceed half the log, so we are
460 * limited by that. Furthermore, the log transaction reservation subsystem
461 * tries to keep 25% of the log free, so we need to keep below that limit or we
462 * risk running out of free log space to start any new transactions.
463 *
464 * In order to keep background CIL push efficient, we will set a lower
465 * threshold at which background pushing is attempted without blocking current
466 * transaction commits. A separate, higher bound defines when CIL pushes are
467 * enforced to ensure we stay within our maximum checkpoint size bounds.
468 * threshold, yet give us plenty of space for aggregation on large logs.
Dave Chinnerdf806152010-05-17 15:52:13 +1000469 */
Dave Chinner80168672010-09-24 18:13:44 +1000470#define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3)
471#define XLOG_CIL_HARD_SPACE_LIMIT(log) (3 * (log->l_logsize >> 4))
Dave Chinnerdf806152010-05-17 15:52:13 +1000472
473/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 * The reservation head lsn is not made up of a cycle number and block number.
475 * Instead, it uses a cycle number and byte number. Logs don't expect to
476 * overflow 31 bits worth of byte offset, so using a byte number will mean
477 * that round off problems won't occur when releasing partial reservations.
478 */
479typedef struct log {
David Chinner4679b2d2008-04-10 12:18:54 +1000480 /* The following fields don't need locking */
481 struct xfs_mount *l_mp; /* mount point */
David Chinnera9c21c12008-10-30 17:39:35 +1100482 struct xfs_ail *l_ailp; /* AIL log is working with */
Dave Chinner71e330b2010-05-21 14:37:18 +1000483 struct xfs_cil *l_cilp; /* CIL log is working with */
David Chinner4679b2d2008-04-10 12:18:54 +1000484 struct xfs_buf *l_xbuf; /* extra buffer for log
485 * wrapping */
486 struct xfs_buftarg *l_targ; /* buftarg of log */
487 uint l_flags;
488 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000489 struct list_head *l_buf_cancel_table;
David Chinner4679b2d2008-04-10 12:18:54 +1000490 int l_iclog_hsize; /* size of iclog header */
491 int l_iclog_heads; /* # of iclog header sectors */
Alex Elder48389ef2010-04-20 17:10:21 +1000492 uint l_sectBBsize; /* sector size in BBs (2^n) */
David Chinner4679b2d2008-04-10 12:18:54 +1000493 int l_iclog_size; /* size of log in bytes */
494 int l_iclog_size_log; /* log power size of log */
495 int l_iclog_bufs; /* number of iclog buffers */
496 xfs_daddr_t l_logBBstart; /* start block of log */
497 int l_logsize; /* size of log in bytes */
498 int l_logBBsize; /* size of log in BB chunks */
499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 /* The following block of fields are changed while holding icloglock */
Dave Chinnereb40a872010-12-21 12:09:01 +1100501 wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp;
Matthew Wilcoxd748c622008-05-19 16:34:27 +1000502 /* waiting for iclog flush */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 int l_covered_state;/* state of "covering disk
504 * log entries" */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 xlog_in_core_t *l_iclog; /* head log queue */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000506 spinlock_t l_icloglock; /* grab to change iclog state */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 int l_curr_cycle; /* Cycle number of log writes */
508 int l_prev_cycle; /* Cycle number before last
509 * block increment */
510 int l_curr_block; /* current logical log block */
511 int l_prev_block; /* previous logical log block */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Dave Chinner84f3c682010-12-03 22:11:29 +1100513 /*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100514 * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
515 * read without needing to hold specific locks. To avoid operations
516 * contending with other hot objects, place each of them on a separate
517 * cacheline.
Dave Chinner84f3c682010-12-03 22:11:29 +1100518 */
519 /* lsn of last LR on disk */
520 atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100521 /* lsn of 1st LR with unflushed * buffers */
522 atomic64_t l_tail_lsn ____cacheline_aligned_in_smp;
Dave Chinner84f3c682010-12-03 22:11:29 +1100523
Dave Chinner3f16b982010-12-21 12:29:01 +1100524 /*
525 * ticket grant locks, queues and accounting have their own cachlines
526 * as these are quite hot and can be operated on concurrently.
527 */
528 spinlock_t l_grant_reserve_lock ____cacheline_aligned_in_smp;
529 struct list_head l_reserveq;
530 atomic64_t l_grant_reserve_head;
531
532 spinlock_t l_grant_write_lock ____cacheline_aligned_in_smp;
533 struct list_head l_writeq;
534 atomic64_t l_grant_write_head;
535
David Chinner4679b2d2008-04-10 12:18:54 +1000536 /* The following field are used for debugging; need to hold icloglock */
537#ifdef DEBUG
538 char *l_iclog_bak[XLOG_MAX_ICLOGS];
539#endif
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541} xlog_t;
542
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000543#define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
544 ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE))
545
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100546#define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548/* common routines */
549extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp);
Eric Sandeen65be6052006-01-11 15:34:19 +1100550extern int xlog_recover(xlog_t *log);
Christoph Hellwig42490232008-08-13 16:49:32 +1000551extern int xlog_recover_finish(xlog_t *log);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
Dave Chinner71e330b2010-05-21 14:37:18 +1000554extern kmem_zone_t *xfs_log_ticket_zone;
555struct xlog_ticket *xlog_ticket_alloc(struct log *log, int unit_bytes,
556 int count, char client, uint xflags,
557 int alloc_flags);
558
David Chinnereb01c9c2008-04-10 12:18:46 +1000559
Christoph Hellwige6b1f272010-03-23 11:47:38 +1100560static inline void
561xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
562{
563 *ptr += bytes;
564 *len -= bytes;
565 *off += bytes;
566}
567
Dave Chinner71e330b2010-05-21 14:37:18 +1000568void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
569int xlog_write(struct log *log, struct xfs_log_vec *log_vector,
570 struct xlog_ticket *tic, xfs_lsn_t *start_lsn,
571 xlog_in_core_t **commit_iclog, uint flags);
572
573/*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100574 * When we crack an atomic LSN, we sample it first so that the value will not
575 * change while we are cracking it into the component values. This means we
576 * will always get consistent component values to work from. This should always
577 * be used to smaple and crack LSNs taht are stored and updated in atomic
578 * variables.
579 */
580static inline void
581xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
582{
583 xfs_lsn_t val = atomic64_read(lsn);
584
585 *cycle = CYCLE_LSN(val);
586 *block = BLOCK_LSN(val);
587}
588
589/*
590 * Calculate and assign a value to an atomic LSN variable from component pieces.
591 */
592static inline void
593xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
594{
595 atomic64_set(lsn, xlog_assign_lsn(cycle, block));
596}
597
598/*
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100599 * When we crack the grant head, we sample it first so that the value will not
Dave Chinnera69ed032010-12-21 12:08:20 +1100600 * change while we are cracking it into the component values. This means we
601 * will always get consistent component values to work from.
602 */
603static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100604xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100605{
Dave Chinnera69ed032010-12-21 12:08:20 +1100606 *cycle = val >> 32;
607 *space = val & 0xffffffff;
608}
609
610static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100611xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
612{
613 xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
614}
615
616static inline int64_t
617xlog_assign_grant_head_val(int cycle, int space)
618{
619 return ((int64_t)cycle << 32) | space;
620}
621
622static inline void
Dave Chinnerc8a09ff2010-12-04 00:02:40 +1100623xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100624{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100625 atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
Dave Chinnera69ed032010-12-21 12:08:20 +1100626}
627
628/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000629 * Committed Item List interfaces
630 */
631int xlog_cil_init(struct log *log);
632void xlog_cil_init_post_recovery(struct log *log);
633void xlog_cil_destroy(struct log *log);
634
Dave Chinnera44f13e2010-08-24 11:40:03 +1000635/*
636 * CIL force routines
637 */
638xfs_lsn_t xlog_cil_force_lsn(struct log *log, xfs_lsn_t sequence);
639
640static inline void
641xlog_cil_force(struct log *log)
642{
643 xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
644}
Dave Chinner71e330b2010-05-21 14:37:18 +1000645
Tim Shimmin955e47a2006-09-28 11:04:16 +1000646/*
647 * Unmount record type is used as a pseudo transaction type for the ticket.
648 * It's value must be outside the range of XFS_TRANS_* values.
649 */
650#define XLOG_UNMOUNT_REC_TYPE (-1U)
651
Dave Chinnereb40a872010-12-21 12:09:01 +1100652/*
653 * Wrapper function for waiting on a wait queue serialised against wakeups
654 * by a spinlock. This matches the semantics of all the wait queues used in the
655 * log code.
656 */
657static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
658{
659 DECLARE_WAITQUEUE(wait, current);
660
661 add_wait_queue_exclusive(wq, &wait);
662 __set_current_state(TASK_UNINTERRUPTIBLE);
663 spin_unlock(lock);
664 schedule();
665 remove_wait_queue(wq, &wait);
666}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667#endif /* __KERNEL__ */
668
669#endif /* __XFS_LOG_PRIV_H__ */