blob: 136654b9400df9b28a40415b1fbca3c9b7d6a958 [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/*
Dave Chinnerfc06c6d2013-08-12 20:49:22 +100027 * Flags for log structure
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 */
Dave Chinnerfc06c6d2013-08-12 20:49:22 +100029#define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
30#define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
31#define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
32 shutdown */
33#define XLOG_TAIL_WARN 0x10 /* log tail verify warning issued */
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
35/*
36 * get client id from packed copy.
37 *
38 * this hack is here because the xlog_pack code copies four bytes
39 * of xlog_op_header containing the fields oh_clientid, oh_flags
40 * and oh_res2 into the packed copy.
41 *
42 * later on this four byte chunk is treated as an int and the
43 * client id is pulled out.
44 *
45 * this has endian issues, of course.
46 */
Christoph Hellwigb53e6752007-10-12 10:59:34 +100047static inline uint xlog_get_client_id(__be32 i)
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100048{
Christoph Hellwigb53e6752007-10-12 10:59:34 +100049 return be32_to_cpu(i) >> 24;
Christoph Hellwig03bea6f2007-10-12 10:58:05 +100050}
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Linus Torvalds1da177e2005-04-16 15:20:36 -070052/*
53 * In core log state
54 */
55#define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
56#define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
57#define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
58#define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
59#define XLOG_STATE_DO_CALLBACK \
60 0x0010 /* Process callback functions */
61#define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
62#define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
63#define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
64#define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
65#define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * Flags to log ticket
69 */
70#define XLOG_TIC_INITED 0x1 /* has been initialized */
71#define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000072
73#define XLOG_TIC_FLAGS \
74 { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \
Dave Chinner10547942010-12-21 12:02:25 +110075 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000076
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/*
78 * Below are states for covering allocation transactions.
79 * By covering, we mean changing the h_tail_lsn in the last on-disk
80 * log write such that no allocation transactions will be re-done during
81 * recovery after a system crash. Recovery starts at the last on-disk
82 * log write.
83 *
84 * These states are used to insert dummy log entries to cover
85 * space allocation transactions which can undo non-transactional changes
86 * after a crash. Writes to a file with space
87 * already allocated do not result in any transactions. Allocations
88 * might include space beyond the EOF. So if we just push the EOF a
89 * little, the last transaction for the file could contain the wrong
90 * size. If there is no file system activity, after an allocation
91 * transaction, and the system crashes, the allocation transaction
92 * will get replayed and the file will be truncated. This could
93 * be hours/days/... after the allocation occurred.
94 *
95 * The fix for this is to do two dummy transactions when the
96 * system is idle. We need two dummy transaction because the h_tail_lsn
97 * in the log record header needs to point beyond the last possible
98 * non-dummy transaction. The first dummy changes the h_tail_lsn to
99 * the first transaction before the dummy. The second dummy causes
100 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
101 *
102 * These dummy transactions get committed when everything
103 * is idle (after there has been some activity).
104 *
105 * There are 5 states used to control this.
106 *
107 * IDLE -- no logging has been done on the file system or
108 * we are done covering previous transactions.
109 * NEED -- logging has occurred and we need a dummy transaction
110 * when the log becomes idle.
111 * DONE -- we were in the NEED state and have committed a dummy
112 * transaction.
113 * NEED2 -- we detected that a dummy transaction has gone to the
114 * on disk log with no other transactions.
115 * DONE2 -- we committed a dummy transaction when in the NEED2 state.
116 *
117 * There are two places where we switch states:
118 *
119 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
120 * We commit the dummy transaction and switch to DONE or DONE2,
121 * respectively. In all other states, we don't do anything.
122 *
123 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
124 *
125 * No matter what state we are in, if this isn't the dummy
126 * transaction going out, the next state is NEED.
127 * So, if we aren't in the DONE or DONE2 states, the next state
128 * is NEED. We can't be finishing a write of the dummy record
129 * unless it was committed and the state switched to DONE or DONE2.
130 *
131 * If we are in the DONE state and this was a write of the
132 * dummy transaction, we move to NEED2.
133 *
134 * If we are in the DONE2 state and this was a write of the
135 * dummy transaction, we move to IDLE.
136 *
137 *
138 * Writing only one dummy transaction can get appended to
139 * one file space allocation. When this happens, the log recovery
140 * code replays the space allocation and a file could be truncated.
141 * This is why we have the NEED2 and DONE2 states before going idle.
142 */
143
144#define XLOG_STATE_COVER_IDLE 0
145#define XLOG_STATE_COVER_NEED 1
146#define XLOG_STATE_COVER_DONE 2
147#define XLOG_STATE_COVER_NEED2 3
148#define XLOG_STATE_COVER_DONE2 4
149
150#define XLOG_COVER_OPS 5
151
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000152/* Ticket reservation region accounting */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000153#define XLOG_TIC_LEN_MAX 15
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000154
155/*
156 * Reservation region
157 * As would be stored in xfs_log_iovec but without the i_addr which
158 * we don't care about.
159 */
160typedef struct xlog_res {
Tim Shimmin12598452006-01-11 21:02:47 +1100161 uint r_len; /* region length :4 */
162 uint r_type; /* region's transaction type :4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000163} xlog_res_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165typedef struct xlog_ticket {
Dave Chinner10547942010-12-21 12:02:25 +1100166 struct list_head t_queue; /* reserve/write queue */
Christoph Hellwig14a72352012-02-20 02:31:24 +0000167 struct task_struct *t_task; /* task that owns this ticket */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000168 xlog_tid_t t_tid; /* transaction identifier : 4 */
Dave Chinnercc09c0d2008-11-17 17:37:10 +1100169 atomic_t t_ref; /* ticket reference count : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000170 int t_curr_res; /* current reservation in bytes : 4 */
171 int t_unit_res; /* unit reservation in bytes : 4 */
172 char t_ocnt; /* original count : 1 */
173 char t_cnt; /* current count : 1 */
174 char t_clientid; /* who does this belong to; : 1 */
175 char t_flags; /* properties of reservation : 1 */
176 uint t_trans_type; /* transaction type : 4 */
177
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000178 /* reservation array fields */
179 uint t_res_num; /* num in array : 4 */
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000180 uint t_res_num_ophdrs; /* num op hdrs : 4 */
181 uint t_res_arr_sum; /* array sum : 4 */
182 uint t_res_o_flow; /* sum overflow : 4 */
Tim Shimmin12598452006-01-11 21:02:47 +1100183 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184} xlog_ticket_t;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186/*
187 * - A log record header is 512 bytes. There is plenty of room to grow the
188 * xlog_rec_header_t into the reserved space.
189 * - ic_data follows, so a write to disk can start at the beginning of
190 * the iclog.
David Chinner12017fa2008-08-13 16:34:31 +1000191 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 * - ic_next is the pointer to the next iclog in the ring.
193 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
194 * - ic_log is a pointer back to the global log structure.
195 * - ic_callback is a linked list of callback function/argument pairs to be
196 * called after an iclog finishes writing.
197 * - ic_size is the full size of the header plus data.
198 * - ic_offset is the current number of bytes written to in this iclog.
199 * - ic_refcnt is bumped when someone is writing to the log.
200 * - ic_state is the state of the iclog.
David Chinner114d23a2008-04-10 12:18:39 +1000201 *
202 * Because of cacheline contention on large machines, we need to separate
203 * various resources onto different cachelines. To start with, make the
204 * structure cacheline aligned. The following fields can be contended on
205 * by independent processes:
206 *
207 * - ic_callback_*
208 * - ic_refcnt
209 * - fields protected by the global l_icloglock
210 *
211 * so we need to ensure that these fields are located in separate cachelines.
212 * We'll put all the read-only and l_icloglock fields in the first cacheline,
213 * and move everything else out to subsequent cachelines.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 */
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100215typedef struct xlog_in_core {
Dave Chinnereb40a872010-12-21 12:09:01 +1100216 wait_queue_head_t ic_force_wait;
217 wait_queue_head_t ic_write_wait;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 struct xlog_in_core *ic_next;
219 struct xlog_in_core *ic_prev;
220 struct xfs_buf *ic_bp;
Mark Tinguelyad223e62012-06-14 09:22:15 -0500221 struct xlog *ic_log;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 int ic_size;
223 int ic_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 int ic_bwritecnt;
Christoph Hellwiga5687782009-02-09 08:37:39 +0100225 unsigned short ic_state;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 char *ic_datap; /* pointer to iclog data */
David Chinner114d23a2008-04-10 12:18:39 +1000227
228 /* Callback structures need their own cacheline */
229 spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
230 xfs_log_callback_t *ic_callback;
231 xfs_log_callback_t **ic_callback_tail;
232
233 /* reference counts need their own cacheline */
234 atomic_t ic_refcnt ____cacheline_aligned_in_smp;
Christoph Hellwigb28708d2008-11-28 14:23:38 +1100235 xlog_in_core_2_t *ic_data;
236#define ic_header ic_data->hic_header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237} xlog_in_core_t;
238
239/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000240 * The CIL context is used to aggregate per-transaction details as well be
241 * passed to the iclog for checkpoint post-commit processing. After being
242 * passed to the iclog, another context needs to be allocated for tracking the
243 * next set of transactions to be aggregated into a checkpoint.
244 */
245struct xfs_cil;
246
247struct xfs_cil_ctx {
248 struct xfs_cil *cil;
249 xfs_lsn_t sequence; /* chkpt sequence # */
250 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
251 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
252 struct xlog_ticket *ticket; /* chkpt ticket */
253 int nvecs; /* number of regions */
254 int space_used; /* aggregate size of regions */
255 struct list_head busy_extents; /* busy extents in chkpt */
256 struct xfs_log_vec *lv_chain; /* logvecs being pushed */
257 xfs_log_callback_t log_cb; /* completion callback hook. */
258 struct list_head committing; /* ctx committing list */
259};
260
261/*
262 * Committed Item List structure
263 *
264 * This structure is used to track log items that have been committed but not
265 * yet written into the log. It is used only when the delayed logging mount
266 * option is enabled.
267 *
268 * This structure tracks the list of committing checkpoint contexts so
269 * we can avoid the problem of having to hold out new transactions during a
270 * flush until we have a the commit record LSN of the checkpoint. We can
271 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
272 * sequence match and extract the commit LSN directly from there. If the
273 * checkpoint is still in the process of committing, we can block waiting for
274 * the commit LSN to be determined as well. This should make synchronous
275 * operations almost as efficient as the old logging methods.
276 */
277struct xfs_cil {
Mark Tinguelyad223e62012-06-14 09:22:15 -0500278 struct xlog *xc_log;
Dave Chinner71e330b2010-05-21 14:37:18 +1000279 struct list_head xc_cil;
280 spinlock_t xc_cil_lock;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000281
282 struct rw_semaphore xc_ctx_lock ____cacheline_aligned_in_smp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000283 struct xfs_cil_ctx *xc_ctx;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000284
285 spinlock_t xc_push_lock ____cacheline_aligned_in_smp;
286 xfs_lsn_t xc_push_seq;
Dave Chinner71e330b2010-05-21 14:37:18 +1000287 struct list_head xc_committing;
Dave Chinnereb40a872010-12-21 12:09:01 +1100288 wait_queue_head_t xc_commit_wait;
Dave Chinnera44f13e2010-08-24 11:40:03 +1000289 xfs_lsn_t xc_current_sequence;
Dave Chinner4c2d5422012-04-23 17:54:32 +1000290 struct work_struct xc_push_work;
Dave Chinner4bb928c2013-08-12 20:50:08 +1000291} ____cacheline_aligned_in_smp;
Dave Chinner71e330b2010-05-21 14:37:18 +1000292
293/*
Dave Chinner80168672010-09-24 18:13:44 +1000294 * The amount of log space we allow the CIL to aggregate is difficult to size.
295 * Whatever we choose, we have to make sure we can get a reservation for the
296 * log space effectively, that it is large enough to capture sufficient
297 * relogging to reduce log buffer IO significantly, but it is not too large for
298 * the log or induces too much latency when writing out through the iclogs. We
299 * track both space consumed and the number of vectors in the checkpoint
300 * context, so we need to decide which to use for limiting.
Dave Chinnerdf806152010-05-17 15:52:13 +1000301 *
302 * Every log buffer we write out during a push needs a header reserved, which
303 * is at least one sector and more for v2 logs. Hence we need a reservation of
304 * at least 512 bytes per 32k of log space just for the LR headers. That means
305 * 16KB of reservation per megabyte of delayed logging space we will consume,
306 * plus various headers. The number of headers will vary based on the num of
307 * io vectors, so limiting on a specific number of vectors is going to result
308 * in transactions of varying size. IOWs, it is more consistent to track and
309 * limit space consumed in the log rather than by the number of objects being
310 * logged in order to prevent checkpoint ticket overruns.
311 *
312 * Further, use of static reservations through the log grant mechanism is
313 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
314 * grant) and a significant deadlock potential because regranting write space
315 * can block on log pushes. Hence if we have to regrant log space during a log
316 * push, we can deadlock.
317 *
318 * However, we can avoid this by use of a dynamic "reservation stealing"
319 * technique during transaction commit whereby unused reservation space in the
320 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
321 * space needed by the checkpoint transaction. This means that we never need to
322 * specifically reserve space for the CIL checkpoint transaction, nor do we
323 * need to regrant space once the checkpoint completes. This also means the
324 * checkpoint transaction ticket is specific to the checkpoint context, rather
325 * than the CIL itself.
326 *
Dave Chinner80168672010-09-24 18:13:44 +1000327 * With dynamic reservations, we can effectively make up arbitrary limits for
328 * the checkpoint size so long as they don't violate any other size rules.
329 * Recovery imposes a rule that no transaction exceed half the log, so we are
330 * limited by that. Furthermore, the log transaction reservation subsystem
331 * tries to keep 25% of the log free, so we need to keep below that limit or we
332 * risk running out of free log space to start any new transactions.
333 *
334 * In order to keep background CIL push efficient, we will set a lower
335 * threshold at which background pushing is attempted without blocking current
336 * transaction commits. A separate, higher bound defines when CIL pushes are
337 * enforced to ensure we stay within our maximum checkpoint size bounds.
338 * threshold, yet give us plenty of space for aggregation on large logs.
Dave Chinnerdf806152010-05-17 15:52:13 +1000339 */
Dave Chinner80168672010-09-24 18:13:44 +1000340#define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3)
Dave Chinnerdf806152010-05-17 15:52:13 +1000341
342/*
Christoph Hellwig28496962012-02-20 02:31:25 +0000343 * ticket grant locks, queues and accounting have their own cachlines
344 * as these are quite hot and can be operated on concurrently.
345 */
346struct xlog_grant_head {
347 spinlock_t lock ____cacheline_aligned_in_smp;
348 struct list_head waiters;
349 atomic64_t grant;
350};
351
352/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 * The reservation head lsn is not made up of a cycle number and block number.
354 * Instead, it uses a cycle number and byte number. Logs don't expect to
355 * overflow 31 bits worth of byte offset, so using a byte number will mean
356 * that round off problems won't occur when releasing partial reservations.
357 */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500358struct xlog {
David Chinner4679b2d2008-04-10 12:18:54 +1000359 /* The following fields don't need locking */
360 struct xfs_mount *l_mp; /* mount point */
David Chinnera9c21c12008-10-30 17:39:35 +1100361 struct xfs_ail *l_ailp; /* AIL log is working with */
Dave Chinner71e330b2010-05-21 14:37:18 +1000362 struct xfs_cil *l_cilp; /* CIL log is working with */
David Chinner4679b2d2008-04-10 12:18:54 +1000363 struct xfs_buf *l_xbuf; /* extra buffer for log
364 * wrapping */
365 struct xfs_buftarg *l_targ; /* buftarg of log */
Dave Chinnerf661f1e2012-10-08 21:56:02 +1100366 struct delayed_work l_work; /* background flush work */
David Chinner4679b2d2008-04-10 12:18:54 +1000367 uint l_flags;
368 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000369 struct list_head *l_buf_cancel_table;
David Chinner4679b2d2008-04-10 12:18:54 +1000370 int l_iclog_hsize; /* size of iclog header */
371 int l_iclog_heads; /* # of iclog header sectors */
Alex Elder48389ef2010-04-20 17:10:21 +1000372 uint l_sectBBsize; /* sector size in BBs (2^n) */
David Chinner4679b2d2008-04-10 12:18:54 +1000373 int l_iclog_size; /* size of log in bytes */
374 int l_iclog_size_log; /* log power size of log */
375 int l_iclog_bufs; /* number of iclog buffers */
376 xfs_daddr_t l_logBBstart; /* start block of log */
377 int l_logsize; /* size of log in bytes */
378 int l_logBBsize; /* size of log in BB chunks */
379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 /* The following block of fields are changed while holding icloglock */
Dave Chinnereb40a872010-12-21 12:09:01 +1100381 wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp;
Matthew Wilcoxd748c622008-05-19 16:34:27 +1000382 /* waiting for iclog flush */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 int l_covered_state;/* state of "covering disk
384 * log entries" */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 xlog_in_core_t *l_iclog; /* head log queue */
Eric Sandeenb22cd72c2007-10-11 17:37:10 +1000386 spinlock_t l_icloglock; /* grab to change iclog state */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 int l_curr_cycle; /* Cycle number of log writes */
388 int l_prev_cycle; /* Cycle number before last
389 * block increment */
390 int l_curr_block; /* current logical log block */
391 int l_prev_block; /* previous logical log block */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Dave Chinner84f3c682010-12-03 22:11:29 +1100393 /*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100394 * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
395 * read without needing to hold specific locks. To avoid operations
396 * contending with other hot objects, place each of them on a separate
397 * cacheline.
Dave Chinner84f3c682010-12-03 22:11:29 +1100398 */
399 /* lsn of last LR on disk */
400 atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp;
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100401 /* lsn of 1st LR with unflushed * buffers */
402 atomic64_t l_tail_lsn ____cacheline_aligned_in_smp;
Dave Chinner84f3c682010-12-03 22:11:29 +1100403
Christoph Hellwig28496962012-02-20 02:31:25 +0000404 struct xlog_grant_head l_reserve_head;
405 struct xlog_grant_head l_write_head;
Dave Chinner3f16b982010-12-21 12:29:01 +1100406
David Chinner4679b2d2008-04-10 12:18:54 +1000407 /* The following field are used for debugging; need to hold icloglock */
408#ifdef DEBUG
409 char *l_iclog_bak[XLOG_MAX_ICLOGS];
410#endif
411
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500412};
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Christoph Hellwigd5689ea2010-12-01 22:06:22 +0000414#define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
415 ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE))
416
Nathan Scottcfcbbbd2005-11-02 15:12:04 +1100417#define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419/* common routines */
Mark Tinguely9a8d2fd2012-06-14 09:22:16 -0500420extern int
421xlog_recover(
422 struct xlog *log);
423extern int
424xlog_recover_finish(
425 struct xlog *log);
Christoph Hellwig0e446be2012-11-12 22:54:24 +1100426
Dave Chinnerf9668a02012-11-28 13:01:03 +1100427extern __le32 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
Christoph Hellwig0e446be2012-11-12 22:54:24 +1100428 char *dp, int size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
Dave Chinner71e330b2010-05-21 14:37:18 +1000430extern kmem_zone_t *xfs_log_ticket_zone;
Mark Tinguelyad223e62012-06-14 09:22:15 -0500431struct xlog_ticket *
432xlog_ticket_alloc(
433 struct xlog *log,
434 int unit_bytes,
435 int count,
436 char client,
437 bool permanent,
438 xfs_km_flags_t alloc_flags);
Dave Chinner71e330b2010-05-21 14:37:18 +1000439
David Chinnereb01c9c2008-04-10 12:18:46 +1000440
Christoph Hellwige6b1f272010-03-23 11:47:38 +1100441static inline void
442xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
443{
444 *ptr += bytes;
445 *len -= bytes;
446 *off += bytes;
447}
448
Dave Chinner71e330b2010-05-21 14:37:18 +1000449void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
Mark Tinguelyad223e62012-06-14 09:22:15 -0500450int
451xlog_write(
452 struct xlog *log,
453 struct xfs_log_vec *log_vector,
454 struct xlog_ticket *tic,
455 xfs_lsn_t *start_lsn,
456 struct xlog_in_core **commit_iclog,
457 uint flags);
Dave Chinner71e330b2010-05-21 14:37:18 +1000458
459/*
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100460 * When we crack an atomic LSN, we sample it first so that the value will not
461 * change while we are cracking it into the component values. This means we
462 * will always get consistent component values to work from. This should always
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300463 * be used to sample and crack LSNs that are stored and updated in atomic
Dave Chinner1c3cb9e2010-12-21 12:28:39 +1100464 * variables.
465 */
466static inline void
467xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
468{
469 xfs_lsn_t val = atomic64_read(lsn);
470
471 *cycle = CYCLE_LSN(val);
472 *block = BLOCK_LSN(val);
473}
474
475/*
476 * Calculate and assign a value to an atomic LSN variable from component pieces.
477 */
478static inline void
479xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
480{
481 atomic64_set(lsn, xlog_assign_lsn(cycle, block));
482}
483
484/*
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100485 * When we crack the grant head, we sample it first so that the value will not
Dave Chinnera69ed032010-12-21 12:08:20 +1100486 * change while we are cracking it into the component values. This means we
487 * will always get consistent component values to work from.
488 */
489static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100490xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100491{
Dave Chinnera69ed032010-12-21 12:08:20 +1100492 *cycle = val >> 32;
493 *space = val & 0xffffffff;
494}
495
496static inline void
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100497xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
498{
499 xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
500}
501
502static inline int64_t
503xlog_assign_grant_head_val(int cycle, int space)
504{
505 return ((int64_t)cycle << 32) | space;
506}
507
508static inline void
Dave Chinnerc8a09ff2010-12-04 00:02:40 +1100509xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
Dave Chinnera69ed032010-12-21 12:08:20 +1100510{
Dave Chinnerd0eb2f32010-12-21 12:29:14 +1100511 atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
Dave Chinnera69ed032010-12-21 12:08:20 +1100512}
513
514/*
Dave Chinner71e330b2010-05-21 14:37:18 +1000515 * Committed Item List interfaces
516 */
Mark Tinguelyad223e62012-06-14 09:22:15 -0500517int
518xlog_cil_init(struct xlog *log);
519void
520xlog_cil_init_post_recovery(struct xlog *log);
521void
522xlog_cil_destroy(struct xlog *log);
Dave Chinner71e330b2010-05-21 14:37:18 +1000523
Dave Chinnera44f13e2010-08-24 11:40:03 +1000524/*
525 * CIL force routines
526 */
Mark Tinguelyad223e62012-06-14 09:22:15 -0500527xfs_lsn_t
528xlog_cil_force_lsn(
529 struct xlog *log,
530 xfs_lsn_t sequence);
Dave Chinnera44f13e2010-08-24 11:40:03 +1000531
532static inline void
Mark Tinguelyad223e62012-06-14 09:22:15 -0500533xlog_cil_force(struct xlog *log)
Dave Chinnera44f13e2010-08-24 11:40:03 +1000534{
535 xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
536}
Dave Chinner71e330b2010-05-21 14:37:18 +1000537
Tim Shimmin955e47a2006-09-28 11:04:16 +1000538/*
539 * Unmount record type is used as a pseudo transaction type for the ticket.
540 * It's value must be outside the range of XFS_TRANS_* values.
541 */
542#define XLOG_UNMOUNT_REC_TYPE (-1U)
543
Dave Chinnereb40a872010-12-21 12:09:01 +1100544/*
545 * Wrapper function for waiting on a wait queue serialised against wakeups
546 * by a spinlock. This matches the semantics of all the wait queues used in the
547 * log code.
548 */
549static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
550{
551 DECLARE_WAITQUEUE(wait, current);
552
553 add_wait_queue_exclusive(wq, &wait);
554 __set_current_state(TASK_UNINTERRUPTIBLE);
555 spin_unlock(lock);
556 schedule();
557 remove_wait_queue(wq, &wait);
558}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559
560#endif /* __XFS_LOG_PRIV_H__ */