blob: 1c87c8abfbed0aef749760ac794b102d217d30eb [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;
Dave Chinner239880e2013-10-23 10:50:10 +110025struct xfs_log_callback;
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
27/*
Dave Chinnerfc06c6d2013-08-12 20:49:22 +100028 * Flags for log structure
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 */
Dave Chinnerfc06c6d2013-08-12 20:49:22 +100030#define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
31#define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
32#define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
33 shutdown */
34#define XLOG_TAIL_WARN 0x10 /* log tail verify warning issued */
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36/*
37 * get client id from packed copy.
38 *
39 * this hack is here because the xlog_pack code copies four bytes
40 * of xlog_op_header containing the fields oh_clientid, oh_flags
41 * and oh_res2 into the packed copy.
42 *
43 * later on this four byte chunk is treated as an int and the
44 * client id is pulled out.
45 *
46 * this has endian issues, of course.
47 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +100048static inline uint xlog_get_client_id(__be32 i)
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100049{
Christoph Hellwigb53e6752007-10-12 10:59:34 +100050 return be32_to_cpu(i) >> 24;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100051}
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Linus Torvalds1da177e2005-04-16 15:20:36 -070053/*
54 * In core log state
55 */
56#define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
57#define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
58#define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
59#define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
60#define XLOG_STATE_DO_CALLBACK \
61 0x0010 /* Process callback functions */
62#define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
63#define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
64#define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
65#define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
66#define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 * Flags to log ticket
70 */
71#define XLOG_TIC_INITED 0x1 /* has been initialized */
72#define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000073
74#define XLOG_TIC_FLAGS \
75 { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \
Dave Chinner10547942010-12-21 12:02:25 +110076 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000077
Linus Torvalds1da177e2005-04-16 15:20:36 -070078/*
79 * Below are states for covering allocation transactions.
80 * By covering, we mean changing the h_tail_lsn in the last on-disk
81 * log write such that no allocation transactions will be re-done during
82 * recovery after a system crash. Recovery starts at the last on-disk
83 * log write.
84 *
85 * These states are used to insert dummy log entries to cover
86 * space allocation transactions which can undo non-transactional changes
87 * after a crash. Writes to a file with space
88 * already allocated do not result in any transactions. Allocations
89 * might include space beyond the EOF. So if we just push the EOF a
90 * little, the last transaction for the file could contain the wrong
91 * size. If there is no file system activity, after an allocation
92 * transaction, and the system crashes, the allocation transaction
93 * will get replayed and the file will be truncated. This could
94 * be hours/days/... after the allocation occurred.
95 *
96 * The fix for this is to do two dummy transactions when the
97 * system is idle. We need two dummy transaction because the h_tail_lsn
98 * in the log record header needs to point beyond the last possible
99 * non-dummy transaction. The first dummy changes the h_tail_lsn to
100 * the first transaction before the dummy. The second dummy causes
101 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
102 *
103 * These dummy transactions get committed when everything
104 * is idle (after there has been some activity).
105 *
106 * There are 5 states used to control this.
107 *
108 * IDLE -- no logging has been done on the file system or
109 * we are done covering previous transactions.
110 * NEED -- logging has occurred and we need a dummy transaction
111 * when the log becomes idle.
112 * DONE -- we were in the NEED state and have committed a dummy
113 * transaction.
114 * NEED2 -- we detected that a dummy transaction has gone to the
115 * on disk log with no other transactions.
116 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
117 *
118 * There are two places where we switch states:
119 *
120 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
121 * We commit the dummy transaction and switch to DONE or DONE2,
122 * respectively. In all other states, we don't do anything.
123 *
124 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
125 *
126 * No matter what state we are in, if this isn't the dummy
127 * transaction going out, the next state is NEED.
128 * So, if we aren't in the DONE or DONE2 states, the next state
129 * is NEED. We can't be finishing a write of the dummy record
130 * unless it was committed and the state switched to DONE or DONE2.
131 *
132 * If we are in the DONE state and this was a write of the
133 * dummy transaction, we move to NEED2.
134 *
135 * If we are in the DONE2 state and this was a write of the
136 * dummy transaction, we move to IDLE.
137 *
138 *
139 * Writing only one dummy transaction can get appended to
140 * one file space allocation. When this happens, the log recovery
141 * code replays the space allocation and a file could be truncated.
142 * This is why we have the NEED2 and DONE2 states before going idle.
143 */
144
145#define XLOG_STATE_COVER_IDLE 0
146#define XLOG_STATE_COVER_NEED 1
147#define XLOG_STATE_COVER_DONE 2
148#define XLOG_STATE_COVER_NEED2 3
149#define XLOG_STATE_COVER_DONE2 4
150
151#define XLOG_COVER_OPS 5
152
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000153/* Ticket reservation region accounting */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000154#define XLOG_TIC_LEN_MAX 15
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000155
156/*
157 * Reservation region
158 * As would be stored in xfs_log_iovec but without the i_addr which
159 * we don't care about.
160 */
161typedef struct xlog_res {
Tim Shimmin12598452006-01-11 21:02:47 +1100162 uint r_len; /* region length :4 */
163 uint r_type; /* region's transaction type :4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000164} xlog_res_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166typedef struct xlog_ticket {
Dave Chinner10547942010-12-21 12:02:25 +1100167 struct list_head t_queue; /* reserve/write queue */
Christoph Hellwig14a72352012-02-20 02:31:24 +0000168 struct task_struct *t_task; /* task that owns this ticket */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000169 xlog_tid_t t_tid; /* transaction identifier : 4 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100170 atomic_t t_ref; /* ticket reference count : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000171 int t_curr_res; /* current reservation in bytes : 4 */
172 int t_unit_res; /* unit reservation in bytes : 4 */
173 char t_ocnt; /* original count : 1 */
174 char t_cnt; /* current count : 1 */
175 char t_clientid; /* who does this belong to; : 1 */
176 char t_flags; /* properties of reservation : 1 */
177 uint t_trans_type; /* transaction type : 4 */
178
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000179 /* reservation array fields */
180 uint t_res_num; /* num in array : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000181 uint t_res_num_ophdrs; /* num op hdrs : 4 */
182 uint t_res_arr_sum; /* array sum : 4 */
183 uint t_res_o_flow; /* sum overflow : 4 */
Tim Shimmin12598452006-01-11 21:02:47 +1100184 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185} xlog_ticket_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187/*
188 * - A log record header is 512 bytes. There is plenty of room to grow the
189 * xlog_rec_header_t into the reserved space.
190 * - ic_data follows, so a write to disk can start at the beginning of
191 * the iclog.
David Chinner12017fa2008-08-13 16:34:31 +1000192 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 * - ic_next is the pointer to the next iclog in the ring.
194 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
195 * - ic_log is a pointer back to the global log structure.
196 * - ic_callback is a linked list of callback function/argument pairs to be
197 * called after an iclog finishes writing.
198 * - ic_size is the full size of the header plus data.
199 * - ic_offset is the current number of bytes written to in this iclog.
200 * - ic_refcnt is bumped when someone is writing to the log.
201 * - ic_state is the state of the iclog.
David Chinner114d23a2008-04-10 12:18:39 +1000202 *
203 * Because of cacheline contention on large machines, we need to separate
204 * various resources onto different cachelines. To start with, make the
205 * structure cacheline aligned. The following fields can be contended on
206 * by independent processes:
207 *
208 * - ic_callback_*
209 * - ic_refcnt
210 * - fields protected by the global l_icloglock
211 *
212 * so we need to ensure that these fields are located in separate cachelines.
213 * We'll put all the read-only and l_icloglock fields in the first cacheline,
214 * and move everything else out to subsequent cachelines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 */
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100216typedef struct xlog_in_core {
Dave Chinnereb40a872010-12-21 12:09:01 +1100217 wait_queue_head_t ic_force_wait;
218 wait_queue_head_t ic_write_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 struct xlog_in_core *ic_next;
220 struct xlog_in_core *ic_prev;
221 struct xfs_buf *ic_bp;
Mark Tinguelyad223e62012-06-14 09:22:15 -0500222 struct xlog *ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 int ic_size;
224 int ic_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 int ic_bwritecnt;
Christoph Hellwiga5687782009-02-09 08:37:39 +0100226 unsigned short ic_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 char *ic_datap; /* pointer to iclog data */
David Chinner114d23a2008-04-10 12:18:39 +1000228
229 /* Callback structures need their own cacheline */
230 spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
Dave Chinner239880e2013-10-23 10:50:10 +1100231 struct xfs_log_callback *ic_callback;
232 struct xfs_log_callback **ic_callback_tail;
David Chinner114d23a2008-04-10 12:18:39 +1000233
234 /* reference counts need their own cacheline */
235 atomic_t ic_refcnt ____cacheline_aligned_in_smp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100236 xlog_in_core_2_t *ic_data;
237#define ic_header ic_data->hic_header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238} xlog_in_core_t;
239
240/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000241 * The CIL context is used to aggregate per-transaction details as well be
242 * passed to the iclog for checkpoint post-commit processing. After being
243 * passed to the iclog, another context needs to be allocated for tracking the
244 * next set of transactions to be aggregated into a checkpoint.
245 */
246struct xfs_cil;
247
248struct xfs_cil_ctx {
249 struct xfs_cil *cil;
250 xfs_lsn_t sequence; /* chkpt sequence # */
251 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
252 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
253 struct xlog_ticket *ticket; /* chkpt ticket */
254 int nvecs; /* number of regions */
255 int space_used; /* aggregate size of regions */
256 struct list_head busy_extents; /* busy extents in chkpt */
257 struct xfs_log_vec *lv_chain; /* logvecs being pushed */
Dave Chinner239880e2013-10-23 10:50:10 +1100258 struct xfs_log_callback log_cb; /* completion callback hook. */
Dave Chinner71e330b2010-05-21 14:37:18 +1000259 struct list_head committing; /* ctx committing list */
260};
261
262/*
263 * Committed Item List structure
264 *
265 * This structure is used to track log items that have been committed but not
266 * yet written into the log. It is used only when the delayed logging mount
267 * option is enabled.
268 *
269 * This structure tracks the list of committing checkpoint contexts so
270 * we can avoid the problem of having to hold out new transactions during a
271 * flush until we have a the commit record LSN of the checkpoint. We can
272 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
273 * sequence match and extract the commit LSN directly from there. If the
274 * checkpoint is still in the process of committing, we can block waiting for
275 * the commit LSN to be determined as well. This should make synchronous
276 * operations almost as efficient as the old logging methods.
277 */
278struct xfs_cil {
Mark Tinguelyad223e62012-06-14 09:22:15 -0500279 struct xlog *xc_log;
Dave Chinner71e330b2010-05-21 14:37:18 +1000280 struct list_head xc_cil;
281 spinlock_t xc_cil_lock;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000282
283 struct rw_semaphore xc_ctx_lock ____cacheline_aligned_in_smp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000284 struct xfs_cil_ctx *xc_ctx;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000285
286 spinlock_t xc_push_lock ____cacheline_aligned_in_smp;
287 xfs_lsn_t xc_push_seq;
Dave Chinner71e330b2010-05-21 14:37:18 +1000288 struct list_head xc_committing;
Dave Chinnereb40a872010-12-21 12:09:01 +1100289 wait_queue_head_t xc_commit_wait;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000290 xfs_lsn_t xc_current_sequence;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000291 struct work_struct xc_push_work;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000292} ____cacheline_aligned_in_smp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000293
294/*
Dave Chinner80168672010-09-24 18:13:44 +1000295 * The amount of log space we allow the CIL to aggregate is difficult to size.
296 * Whatever we choose, we have to make sure we can get a reservation for the
297 * log space effectively, that it is large enough to capture sufficient
298 * relogging to reduce log buffer IO significantly, but it is not too large for
299 * the log or induces too much latency when writing out through the iclogs. We
300 * track both space consumed and the number of vectors in the checkpoint
301 * context, so we need to decide which to use for limiting.
Dave Chinnerdf806152010-05-17 15:52:13 +1000302 *
303 * Every log buffer we write out during a push needs a header reserved, which
304 * is at least one sector and more for v2 logs. Hence we need a reservation of
305 * at least 512 bytes per 32k of log space just for the LR headers. That means
306 * 16KB of reservation per megabyte of delayed logging space we will consume,
307 * plus various headers. The number of headers will vary based on the num of
308 * io vectors, so limiting on a specific number of vectors is going to result
309 * in transactions of varying size. IOWs, it is more consistent to track and
310 * limit space consumed in the log rather than by the number of objects being
311 * logged in order to prevent checkpoint ticket overruns.
312 *
313 * Further, use of static reservations through the log grant mechanism is
314 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
315 * grant) and a significant deadlock potential because regranting write space
316 * can block on log pushes. Hence if we have to regrant log space during a log
317 * push, we can deadlock.
318 *
319 * However, we can avoid this by use of a dynamic "reservation stealing"
320 * technique during transaction commit whereby unused reservation space in the
321 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
322 * space needed by the checkpoint transaction. This means that we never need to
323 * specifically reserve space for the CIL checkpoint transaction, nor do we
324 * need to regrant space once the checkpoint completes. This also means the
325 * checkpoint transaction ticket is specific to the checkpoint context, rather
326 * than the CIL itself.
327 *
Dave Chinner80168672010-09-24 18:13:44 +1000328 * With dynamic reservations, we can effectively make up arbitrary limits for
329 * the checkpoint size so long as they don't violate any other size rules.
330 * Recovery imposes a rule that no transaction exceed half the log, so we are
331 * limited by that. Furthermore, the log transaction reservation subsystem
332 * tries to keep 25% of the log free, so we need to keep below that limit or we
333 * risk running out of free log space to start any new transactions.
334 *
335 * In order to keep background CIL push efficient, we will set a lower
336 * threshold at which background pushing is attempted without blocking current
337 * transaction commits. A separate, higher bound defines when CIL pushes are
338 * enforced to ensure we stay within our maximum checkpoint size bounds.
339 * threshold, yet give us plenty of space for aggregation on large logs.
Dave Chinnerdf806152010-05-17 15:52:13 +1000340 */
Dave Chinner80168672010-09-24 18:13:44 +1000341#define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3)
Dave Chinnerdf806152010-05-17 15:52:13 +1000342
343/*
Christoph Hellwig28496962012-02-20 02:31:25 +0000344 * ticket grant locks, queues and accounting have their own cachlines
345 * as these are quite hot and can be operated on concurrently.
346 */
347struct xlog_grant_head {
348 spinlock_t lock ____cacheline_aligned_in_smp;
349 struct list_head waiters;
350 atomic64_t grant;
351};
352
353/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 * The reservation head lsn is not made up of a cycle number and block number.
355 * Instead, it uses a cycle number and byte number. Logs don't expect to
356 * overflow 31 bits worth of byte offset, so using a byte number will mean
357 * that round off problems won't occur when releasing partial reservations.
358 */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500359struct xlog {
David Chinner4679b2d2008-04-10 12:18:54 +1000360 /* The following fields don't need locking */
361 struct xfs_mount *l_mp; /* mount point */
David Chinnera9c21c12008-10-30 17:39:35 +1100362 struct xfs_ail *l_ailp; /* AIL log is working with */
Dave Chinner71e330b2010-05-21 14:37:18 +1000363 struct xfs_cil *l_cilp; /* CIL log is working with */
David Chinner4679b2d2008-04-10 12:18:54 +1000364 struct xfs_buf *l_xbuf; /* extra buffer for log
365 * wrapping */
366 struct xfs_buftarg *l_targ; /* buftarg of log */
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100367 struct delayed_work l_work; /* background flush work */
David Chinner4679b2d2008-04-10 12:18:54 +1000368 uint l_flags;
369 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000370 struct list_head *l_buf_cancel_table;
David Chinner4679b2d2008-04-10 12:18:54 +1000371 int l_iclog_hsize; /* size of iclog header */
372 int l_iclog_heads; /* # of iclog header sectors */
Alex Elder48389ef2010-04-20 17:10:21 +1000373 uint l_sectBBsize; /* sector size in BBs (2^n) */
David Chinner4679b2d2008-04-10 12:18:54 +1000374 int l_iclog_size; /* size of log in bytes */
375 int l_iclog_size_log; /* log power size of log */
376 int l_iclog_bufs; /* number of iclog buffers */
377 xfs_daddr_t l_logBBstart; /* start block of log */
378 int l_logsize; /* size of log in bytes */
379 int l_logBBsize; /* size of log in BB chunks */
380
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 /* The following block of fields are changed while holding icloglock */
Dave Chinnereb40a872010-12-21 12:09:01 +1100382 wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp;
Matthew Wilcoxd748c622008-05-19 16:34:27 +1000383 /* waiting for iclog flush */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 int l_covered_state;/* state of "covering disk
385 * log entries" */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 xlog_in_core_t *l_iclog; /* head log queue */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000387 spinlock_t l_icloglock; /* grab to change iclog state */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 int l_curr_cycle; /* Cycle number of log writes */
389 int l_prev_cycle; /* Cycle number before last
390 * block increment */
391 int l_curr_block; /* current logical log block */
392 int l_prev_block; /* previous logical log block */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
Dave Chinner84f3c682010-12-03 22:11:29 +1100394 /*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100395 * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
396 * read without needing to hold specific locks. To avoid operations
397 * contending with other hot objects, place each of them on a separate
398 * cacheline.
Dave Chinner84f3c682010-12-03 22:11:29 +1100399 */
400 /* lsn of last LR on disk */
401 atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100402 /* lsn of 1st LR with unflushed * buffers */
403 atomic64_t l_tail_lsn ____cacheline_aligned_in_smp;
Dave Chinner84f3c682010-12-03 22:11:29 +1100404
Christoph Hellwig28496962012-02-20 02:31:25 +0000405 struct xlog_grant_head l_reserve_head;
406 struct xlog_grant_head l_write_head;
Dave Chinner3f16b982010-12-21 12:29:01 +1100407
Brian Fosterbaff4e42014-07-15 08:07:29 +1000408 struct xfs_kobj l_kobj;
409
David Chinner4679b2d2008-04-10 12:18:54 +1000410 /* The following field are used for debugging; need to hold icloglock */
411#ifdef DEBUG
Christoph Hellwig5809d5e2015-06-22 09:44:47 +1000412 void *l_iclog_bak[XLOG_MAX_ICLOGS];
David Chinner4679b2d2008-04-10 12:18:54 +1000413#endif
414
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500415};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000417#define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
418 ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE))
419
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100420#define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
421
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422/* common routines */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500423extern int
424xlog_recover(
425 struct xlog *log);
426extern int
427xlog_recover_finish(
428 struct xlog *log);
Christoph Hellwig0e446be2012-11-12 22:54:24 +1100429
Dave Chinnerf9668a02012-11-28 13:01:03 +1100430extern __le32 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
Christoph Hellwig0e446be2012-11-12 22:54:24 +1100431 char *dp, int size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Dave Chinner71e330b2010-05-21 14:37:18 +1000433extern kmem_zone_t *xfs_log_ticket_zone;
Mark Tinguelyad223e62012-06-14 09:22:15 -0500434struct xlog_ticket *
435xlog_ticket_alloc(
436 struct xlog *log,
437 int unit_bytes,
438 int count,
439 char client,
440 bool permanent,
441 xfs_km_flags_t alloc_flags);
Dave Chinner71e330b2010-05-21 14:37:18 +1000442
David Chinnereb01c9c2008-04-10 12:18:46 +1000443
Christoph Hellwige6b1f272010-03-23 11:47:38 +1100444static inline void
445xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
446{
447 *ptr += bytes;
448 *len -= bytes;
449 *off += bytes;
450}
451
Dave Chinner71e330b2010-05-21 14:37:18 +1000452void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
Mark Tinguelyad223e62012-06-14 09:22:15 -0500453int
454xlog_write(
455 struct xlog *log,
456 struct xfs_log_vec *log_vector,
457 struct xlog_ticket *tic,
458 xfs_lsn_t *start_lsn,
459 struct xlog_in_core **commit_iclog,
460 uint flags);
Dave Chinner71e330b2010-05-21 14:37:18 +1000461
462/*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100463 * When we crack an atomic LSN, we sample it first so that the value will not
464 * change while we are cracking it into the component values. This means we
465 * will always get consistent component values to work from. This should always
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300466 * be used to sample and crack LSNs that are stored and updated in atomic
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100467 * variables.
468 */
469static inline void
470xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
471{
472 xfs_lsn_t val = atomic64_read(lsn);
473
474 *cycle = CYCLE_LSN(val);
475 *block = BLOCK_LSN(val);
476}
477
478/*
479 * Calculate and assign a value to an atomic LSN variable from component pieces.
480 */
481static inline void
482xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
483{
484 atomic64_set(lsn, xlog_assign_lsn(cycle, block));
485}
486
487/*
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100488 * When we crack the grant head, we sample it first so that the value will not
Dave Chinnera69ed032010-12-21 12:08:20 +1100489 * change while we are cracking it into the component values. This means we
490 * will always get consistent component values to work from.
491 */
492static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100493xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100494{
Dave Chinnera69ed032010-12-21 12:08:20 +1100495 *cycle = val >> 32;
496 *space = val & 0xffffffff;
497}
498
499static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100500xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
501{
502 xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
503}
504
505static inline int64_t
506xlog_assign_grant_head_val(int cycle, int space)
507{
508 return ((int64_t)cycle << 32) | space;
509}
510
511static inline void
Dave Chinnerc8a09ff2010-12-04 00:02:40 +1100512xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100513{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100514 atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
Dave Chinnera69ed032010-12-21 12:08:20 +1100515}
516
517/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000518 * Committed Item List interfaces
519 */
Dave Chinner2c6e24c2013-10-15 09:17:49 +1100520int xlog_cil_init(struct xlog *log);
521void xlog_cil_init_post_recovery(struct xlog *log);
522void xlog_cil_destroy(struct xlog *log);
523bool xlog_cil_empty(struct xlog *log);
Dave Chinner71e330b2010-05-21 14:37:18 +1000524
Dave Chinnera44f13e2010-08-24 11:40:03 +1000525/*
526 * CIL force routines
527 */
Mark Tinguelyad223e62012-06-14 09:22:15 -0500528xfs_lsn_t
529xlog_cil_force_lsn(
530 struct xlog *log,
531 xfs_lsn_t sequence);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000532
533static inline void
Mark Tinguelyad223e62012-06-14 09:22:15 -0500534xlog_cil_force(struct xlog *log)
Dave Chinnera44f13e2010-08-24 11:40:03 +1000535{
536 xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
537}
Dave Chinner71e330b2010-05-21 14:37:18 +1000538
Tim Shimmin955e47a2006-09-28 11:04:16 +1000539/*
540 * Unmount record type is used as a pseudo transaction type for the ticket.
541 * It's value must be outside the range of XFS_TRANS_* values.
542 */
543#define XLOG_UNMOUNT_REC_TYPE (-1U)
544
Dave Chinnereb40a872010-12-21 12:09:01 +1100545/*
546 * Wrapper function for waiting on a wait queue serialised against wakeups
547 * by a spinlock. This matches the semantics of all the wait queues used in the
548 * log code.
549 */
550static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
551{
552 DECLARE_WAITQUEUE(wait, current);
553
554 add_wait_queue_exclusive(wq, &wait);
555 __set_current_state(TASK_UNINTERRUPTIBLE);
556 spin_unlock(lock);
557 schedule();
558 remove_wait_queue(wq, &wait);
559}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561#endif /* __XFS_LOG_PRIV_H__ */