]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * 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. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * 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 | |
1da177e4 LT |
17 | */ |
18 | #ifndef __XFS_LOG_PRIV_H__ | |
19 | #define __XFS_LOG_PRIV_H__ | |
20 | ||
21 | struct xfs_buf; | |
ad223e60 | 22 | struct xlog; |
a844f451 | 23 | struct xlog_ticket; |
1da177e4 LT |
24 | struct xfs_mount; |
25 | ||
26 | /* | |
fc06c6d0 | 27 | * Flags for log structure |
1da177e4 | 28 | */ |
fc06c6d0 DC |
29 | #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 */ | |
1da177e4 LT |
34 | |
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 | */ | |
b53e675d | 47 | static inline uint xlog_get_client_id(__be32 i) |
03bea6fe | 48 | { |
b53e675d | 49 | return be32_to_cpu(i) >> 24; |
03bea6fe | 50 | } |
1da177e4 | 51 | |
1da177e4 LT |
52 | /* |
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 */ | |
1da177e4 | 66 | |
1da177e4 LT |
67 | /* |
68 | * Flags to log ticket | |
69 | */ | |
70 | #define XLOG_TIC_INITED 0x1 /* has been initialized */ | |
71 | #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */ | |
0b1b213f CH |
72 | |
73 | #define XLOG_TIC_FLAGS \ | |
74 | { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \ | |
10547941 | 75 | { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" } |
0b1b213f | 76 | |
1da177e4 LT |
77 | /* |
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 | ||
7e9c6396 | 152 | /* Ticket reservation region accounting */ |
7e9c6396 | 153 | #define XLOG_TIC_LEN_MAX 15 |
7e9c6396 TS |
154 | |
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 | */ | |
160 | typedef struct xlog_res { | |
1259845d TS |
161 | uint r_len; /* region length :4 */ |
162 | uint r_type; /* region's transaction type :4 */ | |
7e9c6396 | 163 | } xlog_res_t; |
7e9c6396 | 164 | |
1da177e4 | 165 | typedef struct xlog_ticket { |
10547941 | 166 | struct list_head t_queue; /* reserve/write queue */ |
14a7235f | 167 | struct task_struct *t_task; /* task that owns this ticket */ |
7e9c6396 | 168 | xlog_tid_t t_tid; /* transaction identifier : 4 */ |
cc09c0dc | 169 | atomic_t t_ref; /* ticket reference count : 4 */ |
7e9c6396 TS |
170 | 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 | ||
7e9c6396 TS |
178 | /* reservation array fields */ |
179 | uint t_res_num; /* num in array : 4 */ | |
7e9c6396 TS |
180 | 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 */ | |
1259845d | 183 | xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */ |
1da177e4 | 184 | } xlog_ticket_t; |
7e9c6396 | 185 | |
1da177e4 LT |
186 | /* |
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. | |
12017faf | 191 | * - ic_forcewait is used to implement synchronous forcing of the iclog to disk. |
1da177e4 LT |
192 | * - 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. | |
114d23aa DC |
201 | * |
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. | |
1da177e4 | 214 | */ |
b28708d6 | 215 | typedef struct xlog_in_core { |
eb40a875 DC |
216 | wait_queue_head_t ic_force_wait; |
217 | wait_queue_head_t ic_write_wait; | |
1da177e4 LT |
218 | struct xlog_in_core *ic_next; |
219 | struct xlog_in_core *ic_prev; | |
220 | struct xfs_buf *ic_bp; | |
ad223e60 | 221 | struct xlog *ic_log; |
1da177e4 LT |
222 | int ic_size; |
223 | int ic_offset; | |
1da177e4 | 224 | int ic_bwritecnt; |
a5687787 | 225 | unsigned short ic_state; |
1da177e4 | 226 | char *ic_datap; /* pointer to iclog data */ |
114d23aa DC |
227 | |
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; | |
b28708d6 CH |
235 | xlog_in_core_2_t *ic_data; |
236 | #define ic_header ic_data->hic_header | |
1da177e4 LT |
237 | } xlog_in_core_t; |
238 | ||
71e330b5 DC |
239 | /* |
240 | * 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 | */ | |
245 | struct xfs_cil; | |
246 | ||
247 | struct 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 | */ | |
277 | struct xfs_cil { | |
ad223e60 | 278 | struct xlog *xc_log; |
71e330b5 DC |
279 | struct list_head xc_cil; |
280 | spinlock_t xc_cil_lock; | |
4bb928cd DC |
281 | |
282 | struct rw_semaphore xc_ctx_lock ____cacheline_aligned_in_smp; | |
71e330b5 | 283 | struct xfs_cil_ctx *xc_ctx; |
4bb928cd DC |
284 | |
285 | spinlock_t xc_push_lock ____cacheline_aligned_in_smp; | |
286 | xfs_lsn_t xc_push_seq; | |
71e330b5 | 287 | struct list_head xc_committing; |
eb40a875 | 288 | wait_queue_head_t xc_commit_wait; |
a44f13ed | 289 | xfs_lsn_t xc_current_sequence; |
4c2d542f | 290 | struct work_struct xc_push_work; |
4bb928cd | 291 | } ____cacheline_aligned_in_smp; |
71e330b5 | 292 | |
df806158 | 293 | /* |
80168676 DC |
294 | * 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. | |
df806158 DC |
301 | * |
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 | * | |
80168676 DC |
327 | * 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. | |
df806158 | 339 | */ |
80168676 | 340 | #define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3) |
df806158 | 341 | |
28496968 CH |
342 | /* |
343 | * ticket grant locks, queues and accounting have their own cachlines | |
344 | * as these are quite hot and can be operated on concurrently. | |
345 | */ | |
346 | struct xlog_grant_head { | |
347 | spinlock_t lock ____cacheline_aligned_in_smp; | |
348 | struct list_head waiters; | |
349 | atomic64_t grant; | |
350 | }; | |
351 | ||
1da177e4 LT |
352 | /* |
353 | * 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 | */ | |
9a8d2fdb | 358 | struct xlog { |
4679b2d3 DC |
359 | /* The following fields don't need locking */ |
360 | struct xfs_mount *l_mp; /* mount point */ | |
a9c21c1b | 361 | struct xfs_ail *l_ailp; /* AIL log is working with */ |
71e330b5 | 362 | struct xfs_cil *l_cilp; /* CIL log is working with */ |
4679b2d3 DC |
363 | struct xfs_buf *l_xbuf; /* extra buffer for log |
364 | * wrapping */ | |
365 | struct xfs_buftarg *l_targ; /* buftarg of log */ | |
f661f1e0 | 366 | struct delayed_work l_work; /* background flush work */ |
4679b2d3 DC |
367 | uint l_flags; |
368 | uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */ | |
d5689eaa | 369 | struct list_head *l_buf_cancel_table; |
4679b2d3 DC |
370 | int l_iclog_hsize; /* size of iclog header */ |
371 | int l_iclog_heads; /* # of iclog header sectors */ | |
48389ef1 | 372 | uint l_sectBBsize; /* sector size in BBs (2^n) */ |
4679b2d3 DC |
373 | 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 | ||
1da177e4 | 380 | /* The following block of fields are changed while holding icloglock */ |
eb40a875 | 381 | wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp; |
d748c623 | 382 | /* waiting for iclog flush */ |
1da177e4 LT |
383 | int l_covered_state;/* state of "covering disk |
384 | * log entries" */ | |
1da177e4 | 385 | xlog_in_core_t *l_iclog; /* head log queue */ |
b22cd72c | 386 | spinlock_t l_icloglock; /* grab to change iclog state */ |
1da177e4 LT |
387 | 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 */ | |
1da177e4 | 392 | |
84f3c683 | 393 | /* |
1c3cb9ec DC |
394 | * 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. | |
84f3c683 DC |
398 | */ |
399 | /* lsn of last LR on disk */ | |
400 | atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp; | |
1c3cb9ec DC |
401 | /* lsn of 1st LR with unflushed * buffers */ |
402 | atomic64_t l_tail_lsn ____cacheline_aligned_in_smp; | |
84f3c683 | 403 | |
28496968 CH |
404 | struct xlog_grant_head l_reserve_head; |
405 | struct xlog_grant_head l_write_head; | |
3f16b985 | 406 | |
4679b2d3 DC |
407 | /* 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 | ||
9a8d2fdb | 412 | }; |
1da177e4 | 413 | |
d5689eaa CH |
414 | #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \ |
415 | ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE)) | |
416 | ||
cfcbbbd0 NS |
417 | #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR) |
418 | ||
1da177e4 | 419 | /* common routines */ |
9a8d2fdb MT |
420 | extern int |
421 | xlog_recover( | |
422 | struct xlog *log); | |
423 | extern int | |
424 | xlog_recover_finish( | |
425 | struct xlog *log); | |
0e446be4 | 426 | |
f9668a09 | 427 | extern __le32 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead, |
0e446be4 | 428 | char *dp, int size); |
1da177e4 | 429 | |
71e330b5 | 430 | extern kmem_zone_t *xfs_log_ticket_zone; |
ad223e60 MT |
431 | struct xlog_ticket * |
432 | xlog_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); | |
71e330b5 | 439 | |
eb01c9cd | 440 | |
e6b1f273 CH |
441 | static inline void |
442 | xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes) | |
443 | { | |
444 | *ptr += bytes; | |
445 | *len -= bytes; | |
446 | *off += bytes; | |
447 | } | |
448 | ||
71e330b5 | 449 | void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket); |
ad223e60 MT |
450 | int |
451 | xlog_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); | |
71e330b5 | 458 | |
1c3cb9ec DC |
459 | /* |
460 | * 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 | |
25985edc | 463 | * be used to sample and crack LSNs that are stored and updated in atomic |
1c3cb9ec DC |
464 | * variables. |
465 | */ | |
466 | static inline void | |
467 | xlog_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 | */ | |
478 | static inline void | |
479 | xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block) | |
480 | { | |
481 | atomic64_set(lsn, xlog_assign_lsn(cycle, block)); | |
482 | } | |
483 | ||
a69ed03c | 484 | /* |
d0eb2f38 | 485 | * When we crack the grant head, we sample it first so that the value will not |
a69ed03c DC |
486 | * change while we are cracking it into the component values. This means we |
487 | * will always get consistent component values to work from. | |
488 | */ | |
489 | static inline void | |
d0eb2f38 | 490 | xlog_crack_grant_head_val(int64_t val, int *cycle, int *space) |
a69ed03c | 491 | { |
a69ed03c DC |
492 | *cycle = val >> 32; |
493 | *space = val & 0xffffffff; | |
494 | } | |
495 | ||
d0eb2f38 DC |
496 | static inline void |
497 | xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space) | |
498 | { | |
499 | xlog_crack_grant_head_val(atomic64_read(head), cycle, space); | |
500 | } | |
501 | ||
502 | static inline int64_t | |
503 | xlog_assign_grant_head_val(int cycle, int space) | |
504 | { | |
505 | return ((int64_t)cycle << 32) | space; | |
506 | } | |
507 | ||
a69ed03c | 508 | static inline void |
c8a09ff8 | 509 | xlog_assign_grant_head(atomic64_t *head, int cycle, int space) |
a69ed03c | 510 | { |
d0eb2f38 | 511 | atomic64_set(head, xlog_assign_grant_head_val(cycle, space)); |
a69ed03c DC |
512 | } |
513 | ||
71e330b5 DC |
514 | /* |
515 | * Committed Item List interfaces | |
516 | */ | |
ad223e60 MT |
517 | int |
518 | xlog_cil_init(struct xlog *log); | |
519 | void | |
520 | xlog_cil_init_post_recovery(struct xlog *log); | |
521 | void | |
522 | xlog_cil_destroy(struct xlog *log); | |
71e330b5 | 523 | |
a44f13ed DC |
524 | /* |
525 | * CIL force routines | |
526 | */ | |
ad223e60 MT |
527 | xfs_lsn_t |
528 | xlog_cil_force_lsn( | |
529 | struct xlog *log, | |
530 | xfs_lsn_t sequence); | |
a44f13ed DC |
531 | |
532 | static inline void | |
ad223e60 | 533 | xlog_cil_force(struct xlog *log) |
a44f13ed DC |
534 | { |
535 | xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence); | |
536 | } | |
71e330b5 | 537 | |
955e47ad TS |
538 | /* |
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 | ||
eb40a875 DC |
544 | /* |
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 | */ | |
549 | static 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 | } | |
1da177e4 LT |
559 | |
560 | #endif /* __XFS_LOG_PRIV_H__ */ |