]>
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; | |
1da177e4 | 22 | struct log; |
a844f451 | 23 | struct xlog_ticket; |
1da177e4 LT |
24 | struct xfs_mount; |
25 | ||
26 | /* | |
27 | * Macros, structures, prototypes for internal log manager use. | |
28 | */ | |
29 | ||
30 | #define XLOG_MIN_ICLOGS 2 | |
1da177e4 | 31 | #define XLOG_MAX_ICLOGS 8 |
1da177e4 LT |
32 | #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) | |
dcb3b83f | 36 | #define XLOG_MIN_RECORD_BSIZE (16*1024) /* eventually 32k */ |
1da177e4 LT |
37 | #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 */ | |
dcb3b83f | 40 | #define XLOG_MIN_RECORD_BSHIFT 14 /* 16384 == 1 << 14 */ |
1da177e4 LT |
41 | #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) \ | |
62118709 | 50 | BTOBB(1 << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \ |
1da177e4 LT |
51 | XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT)) |
52 | #define XLOG_TOTAL_REC_SHIFT(log) \ | |
62118709 | 53 | BTOBB(XLOG_MAX_ICLOGS << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \ |
1da177e4 LT |
54 | XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT)) |
55 | ||
03bea6fe CH |
56 | static inline xfs_lsn_t xlog_assign_lsn(uint cycle, uint block) |
57 | { | |
58 | return ((xfs_lsn_t)cycle << 32) | block; | |
59 | } | |
1da177e4 | 60 | |
03bea6fe CH |
61 | static inline uint xlog_get_cycle(char *ptr) |
62 | { | |
b53e675d CH |
63 | if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM) |
64 | return be32_to_cpu(*((__be32 *)ptr + 1)); | |
03bea6fe | 65 | else |
b53e675d | 66 | return be32_to_cpu(*(__be32 *)ptr); |
03bea6fe | 67 | } |
1da177e4 LT |
68 | |
69 | #define BLK_AVG(blk1, blk2) ((blk1+blk2) >> 1) | |
70 | ||
1da177e4 LT |
71 | #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 | */ | |
b53e675d | 85 | static inline uint xlog_get_client_id(__be32 i) |
03bea6fe | 86 | { |
b53e675d | 87 | return be32_to_cpu(i) >> 24; |
03bea6fe | 88 | } |
1da177e4 | 89 | |
1da177e4 LT |
90 | /* |
91 | * In core log state | |
92 | */ | |
93 | #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */ | |
94 | #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */ | |
95 | #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */ | |
96 | #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */ | |
97 | #define XLOG_STATE_DO_CALLBACK \ | |
98 | 0x0010 /* Process callback functions */ | |
99 | #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */ | |
100 | #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/ | |
101 | #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */ | |
102 | #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */ | |
103 | #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */ | |
104 | #endif /* __KERNEL__ */ | |
105 | ||
106 | /* | |
107 | * Flags to log operation header | |
108 | * | |
109 | * The first write of a new transaction will be preceded with a start | |
110 | * record, XLOG_START_TRANS. Once a transaction is committed, a commit | |
111 | * record is written, XLOG_COMMIT_TRANS. If a single region can not fit into | |
112 | * the remainder of the current active in-core log, it is split up into | |
113 | * multiple regions. Each partial region will be marked with a | |
114 | * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS. | |
115 | * | |
116 | */ | |
117 | #define XLOG_START_TRANS 0x01 /* Start a new transaction */ | |
118 | #define XLOG_COMMIT_TRANS 0x02 /* Commit this transaction */ | |
119 | #define XLOG_CONTINUE_TRANS 0x04 /* Cont this trans into new region */ | |
120 | #define XLOG_WAS_CONT_TRANS 0x08 /* Cont this trans into new region */ | |
121 | #define XLOG_END_TRANS 0x10 /* End a continued transaction */ | |
122 | #define XLOG_UNMOUNT_TRANS 0x20 /* Unmount a filesystem transaction */ | |
1da177e4 LT |
123 | |
124 | #ifdef __KERNEL__ | |
125 | /* | |
126 | * Flags to log ticket | |
127 | */ | |
128 | #define XLOG_TIC_INITED 0x1 /* has been initialized */ | |
129 | #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */ | |
0b1b213f CH |
130 | |
131 | #define XLOG_TIC_FLAGS \ | |
132 | { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \ | |
10547941 | 133 | { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" } |
0b1b213f | 134 | |
1da177e4 LT |
135 | #endif /* __KERNEL__ */ |
136 | ||
137 | #define XLOG_UNMOUNT_TYPE 0x556e /* Un for Unmount */ | |
138 | ||
139 | /* | |
140 | * Flags for log structure | |
141 | */ | |
142 | #define XLOG_CHKSUM_MISMATCH 0x1 /* used only during recovery */ | |
143 | #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */ | |
144 | #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */ | |
145 | #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being | |
146 | shutdown */ | |
da8a1a4a | 147 | #define XLOG_TAIL_WARN 0x10 /* log tail verify warning issued */ |
1da177e4 | 148 | |
97d3ac75 CH |
149 | typedef __uint32_t xlog_tid_t; |
150 | ||
1da177e4 LT |
151 | #ifdef __KERNEL__ |
152 | /* | |
153 | * Below are states for covering allocation transactions. | |
154 | * By covering, we mean changing the h_tail_lsn in the last on-disk | |
155 | * log write such that no allocation transactions will be re-done during | |
156 | * recovery after a system crash. Recovery starts at the last on-disk | |
157 | * log write. | |
158 | * | |
159 | * These states are used to insert dummy log entries to cover | |
160 | * space allocation transactions which can undo non-transactional changes | |
161 | * after a crash. Writes to a file with space | |
162 | * already allocated do not result in any transactions. Allocations | |
163 | * might include space beyond the EOF. So if we just push the EOF a | |
164 | * little, the last transaction for the file could contain the wrong | |
165 | * size. If there is no file system activity, after an allocation | |
166 | * transaction, and the system crashes, the allocation transaction | |
167 | * will get replayed and the file will be truncated. This could | |
168 | * be hours/days/... after the allocation occurred. | |
169 | * | |
170 | * The fix for this is to do two dummy transactions when the | |
171 | * system is idle. We need two dummy transaction because the h_tail_lsn | |
172 | * in the log record header needs to point beyond the last possible | |
173 | * non-dummy transaction. The first dummy changes the h_tail_lsn to | |
174 | * the first transaction before the dummy. The second dummy causes | |
175 | * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn. | |
176 | * | |
177 | * These dummy transactions get committed when everything | |
178 | * is idle (after there has been some activity). | |
179 | * | |
180 | * There are 5 states used to control this. | |
181 | * | |
182 | * IDLE -- no logging has been done on the file system or | |
183 | * we are done covering previous transactions. | |
184 | * NEED -- logging has occurred and we need a dummy transaction | |
185 | * when the log becomes idle. | |
186 | * DONE -- we were in the NEED state and have committed a dummy | |
187 | * transaction. | |
188 | * NEED2 -- we detected that a dummy transaction has gone to the | |
189 | * on disk log with no other transactions. | |
190 | * DONE2 -- we committed a dummy transaction when in the NEED2 state. | |
191 | * | |
192 | * There are two places where we switch states: | |
193 | * | |
194 | * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2. | |
195 | * We commit the dummy transaction and switch to DONE or DONE2, | |
196 | * respectively. In all other states, we don't do anything. | |
197 | * | |
198 | * 2.) When we finish writing the on-disk log (xlog_state_clean_log). | |
199 | * | |
200 | * No matter what state we are in, if this isn't the dummy | |
201 | * transaction going out, the next state is NEED. | |
202 | * So, if we aren't in the DONE or DONE2 states, the next state | |
203 | * is NEED. We can't be finishing a write of the dummy record | |
204 | * unless it was committed and the state switched to DONE or DONE2. | |
205 | * | |
206 | * If we are in the DONE state and this was a write of the | |
207 | * dummy transaction, we move to NEED2. | |
208 | * | |
209 | * If we are in the DONE2 state and this was a write of the | |
210 | * dummy transaction, we move to IDLE. | |
211 | * | |
212 | * | |
213 | * Writing only one dummy transaction can get appended to | |
214 | * one file space allocation. When this happens, the log recovery | |
215 | * code replays the space allocation and a file could be truncated. | |
216 | * This is why we have the NEED2 and DONE2 states before going idle. | |
217 | */ | |
218 | ||
219 | #define XLOG_STATE_COVER_IDLE 0 | |
220 | #define XLOG_STATE_COVER_NEED 1 | |
221 | #define XLOG_STATE_COVER_DONE 2 | |
222 | #define XLOG_STATE_COVER_NEED2 3 | |
223 | #define XLOG_STATE_COVER_DONE2 4 | |
224 | ||
225 | #define XLOG_COVER_OPS 5 | |
226 | ||
7e9c6396 TS |
227 | |
228 | /* Ticket reservation region accounting */ | |
7e9c6396 | 229 | #define XLOG_TIC_LEN_MAX 15 |
7e9c6396 TS |
230 | |
231 | /* | |
232 | * Reservation region | |
233 | * As would be stored in xfs_log_iovec but without the i_addr which | |
234 | * we don't care about. | |
235 | */ | |
236 | typedef struct xlog_res { | |
1259845d TS |
237 | uint r_len; /* region length :4 */ |
238 | uint r_type; /* region's transaction type :4 */ | |
7e9c6396 | 239 | } xlog_res_t; |
7e9c6396 | 240 | |
1da177e4 | 241 | typedef struct xlog_ticket { |
eb40a875 | 242 | wait_queue_head_t t_wait; /* ticket wait queue */ |
10547941 | 243 | struct list_head t_queue; /* reserve/write queue */ |
7e9c6396 | 244 | xlog_tid_t t_tid; /* transaction identifier : 4 */ |
cc09c0dc | 245 | atomic_t t_ref; /* ticket reference count : 4 */ |
7e9c6396 TS |
246 | int t_curr_res; /* current reservation in bytes : 4 */ |
247 | int t_unit_res; /* unit reservation in bytes : 4 */ | |
248 | char t_ocnt; /* original count : 1 */ | |
249 | char t_cnt; /* current count : 1 */ | |
250 | char t_clientid; /* who does this belong to; : 1 */ | |
251 | char t_flags; /* properties of reservation : 1 */ | |
252 | uint t_trans_type; /* transaction type : 4 */ | |
253 | ||
7e9c6396 TS |
254 | /* reservation array fields */ |
255 | uint t_res_num; /* num in array : 4 */ | |
7e9c6396 TS |
256 | uint t_res_num_ophdrs; /* num op hdrs : 4 */ |
257 | uint t_res_arr_sum; /* array sum : 4 */ | |
258 | uint t_res_o_flow; /* sum overflow : 4 */ | |
1259845d | 259 | xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */ |
1da177e4 | 260 | } xlog_ticket_t; |
7e9c6396 | 261 | |
1da177e4 LT |
262 | #endif |
263 | ||
264 | ||
265 | typedef struct xlog_op_header { | |
67fcb7bf CH |
266 | __be32 oh_tid; /* transaction id of operation : 4 b */ |
267 | __be32 oh_len; /* bytes in data region : 4 b */ | |
268 | __u8 oh_clientid; /* who sent me this : 1 b */ | |
269 | __u8 oh_flags; /* : 1 b */ | |
270 | __u16 oh_res2; /* 32 bit align : 2 b */ | |
1da177e4 LT |
271 | } xlog_op_header_t; |
272 | ||
273 | ||
274 | /* valid values for h_fmt */ | |
275 | #define XLOG_FMT_UNKNOWN 0 | |
276 | #define XLOG_FMT_LINUX_LE 1 | |
277 | #define XLOG_FMT_LINUX_BE 2 | |
278 | #define XLOG_FMT_IRIX_BE 3 | |
279 | ||
280 | /* our fmt */ | |
f016bad6 | 281 | #ifdef XFS_NATIVE_HOST |
1da177e4 LT |
282 | #define XLOG_FMT XLOG_FMT_LINUX_BE |
283 | #else | |
f016bad6 | 284 | #define XLOG_FMT XLOG_FMT_LINUX_LE |
1da177e4 LT |
285 | #endif |
286 | ||
287 | typedef struct xlog_rec_header { | |
b53e675d CH |
288 | __be32 h_magicno; /* log record (LR) identifier : 4 */ |
289 | __be32 h_cycle; /* write cycle of log : 4 */ | |
290 | __be32 h_version; /* LR version : 4 */ | |
291 | __be32 h_len; /* len in bytes; should be 64-bit aligned: 4 */ | |
292 | __be64 h_lsn; /* lsn of this LR : 8 */ | |
293 | __be64 h_tail_lsn; /* lsn of 1st LR w/ buffers not committed: 8 */ | |
294 | __be32 h_chksum; /* may not be used; non-zero if used : 4 */ | |
295 | __be32 h_prev_block; /* block number to previous LR : 4 */ | |
296 | __be32 h_num_logops; /* number of log operations in this LR : 4 */ | |
297 | __be32 h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; | |
1da177e4 | 298 | /* new fields */ |
b53e675d CH |
299 | __be32 h_fmt; /* format of log record : 4 */ |
300 | uuid_t h_fs_uuid; /* uuid of FS : 16 */ | |
301 | __be32 h_size; /* iclog size : 4 */ | |
1da177e4 LT |
302 | } xlog_rec_header_t; |
303 | ||
304 | typedef struct xlog_rec_ext_header { | |
b53e675d CH |
305 | __be32 xh_cycle; /* write cycle of log : 4 */ |
306 | __be32 xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /* : 256 */ | |
1da177e4 LT |
307 | } xlog_rec_ext_header_t; |
308 | ||
309 | #ifdef __KERNEL__ | |
b28708d6 CH |
310 | |
311 | /* | |
312 | * Quite misnamed, because this union lays out the actual on-disk log buffer. | |
313 | */ | |
314 | typedef union xlog_in_core2 { | |
315 | xlog_rec_header_t hic_header; | |
316 | xlog_rec_ext_header_t hic_xheader; | |
317 | char hic_sector[XLOG_HEADER_SIZE]; | |
318 | } xlog_in_core_2_t; | |
319 | ||
1da177e4 LT |
320 | /* |
321 | * - A log record header is 512 bytes. There is plenty of room to grow the | |
322 | * xlog_rec_header_t into the reserved space. | |
323 | * - ic_data follows, so a write to disk can start at the beginning of | |
324 | * the iclog. | |
12017faf | 325 | * - ic_forcewait is used to implement synchronous forcing of the iclog to disk. |
1da177e4 LT |
326 | * - ic_next is the pointer to the next iclog in the ring. |
327 | * - ic_bp is a pointer to the buffer used to write this incore log to disk. | |
328 | * - ic_log is a pointer back to the global log structure. | |
329 | * - ic_callback is a linked list of callback function/argument pairs to be | |
330 | * called after an iclog finishes writing. | |
331 | * - ic_size is the full size of the header plus data. | |
332 | * - ic_offset is the current number of bytes written to in this iclog. | |
333 | * - ic_refcnt is bumped when someone is writing to the log. | |
334 | * - ic_state is the state of the iclog. | |
114d23aa DC |
335 | * |
336 | * Because of cacheline contention on large machines, we need to separate | |
337 | * various resources onto different cachelines. To start with, make the | |
338 | * structure cacheline aligned. The following fields can be contended on | |
339 | * by independent processes: | |
340 | * | |
341 | * - ic_callback_* | |
342 | * - ic_refcnt | |
343 | * - fields protected by the global l_icloglock | |
344 | * | |
345 | * so we need to ensure that these fields are located in separate cachelines. | |
346 | * We'll put all the read-only and l_icloglock fields in the first cacheline, | |
347 | * and move everything else out to subsequent cachelines. | |
1da177e4 | 348 | */ |
b28708d6 | 349 | typedef struct xlog_in_core { |
eb40a875 DC |
350 | wait_queue_head_t ic_force_wait; |
351 | wait_queue_head_t ic_write_wait; | |
1da177e4 LT |
352 | struct xlog_in_core *ic_next; |
353 | struct xlog_in_core *ic_prev; | |
354 | struct xfs_buf *ic_bp; | |
355 | struct log *ic_log; | |
1da177e4 LT |
356 | int ic_size; |
357 | int ic_offset; | |
1da177e4 | 358 | int ic_bwritecnt; |
a5687787 | 359 | unsigned short ic_state; |
1da177e4 | 360 | char *ic_datap; /* pointer to iclog data */ |
114d23aa DC |
361 | |
362 | /* Callback structures need their own cacheline */ | |
363 | spinlock_t ic_callback_lock ____cacheline_aligned_in_smp; | |
364 | xfs_log_callback_t *ic_callback; | |
365 | xfs_log_callback_t **ic_callback_tail; | |
366 | ||
367 | /* reference counts need their own cacheline */ | |
368 | atomic_t ic_refcnt ____cacheline_aligned_in_smp; | |
b28708d6 CH |
369 | xlog_in_core_2_t *ic_data; |
370 | #define ic_header ic_data->hic_header | |
1da177e4 LT |
371 | } xlog_in_core_t; |
372 | ||
71e330b5 DC |
373 | /* |
374 | * The CIL context is used to aggregate per-transaction details as well be | |
375 | * passed to the iclog for checkpoint post-commit processing. After being | |
376 | * passed to the iclog, another context needs to be allocated for tracking the | |
377 | * next set of transactions to be aggregated into a checkpoint. | |
378 | */ | |
379 | struct xfs_cil; | |
380 | ||
381 | struct xfs_cil_ctx { | |
382 | struct xfs_cil *cil; | |
383 | xfs_lsn_t sequence; /* chkpt sequence # */ | |
384 | xfs_lsn_t start_lsn; /* first LSN of chkpt commit */ | |
385 | xfs_lsn_t commit_lsn; /* chkpt commit record lsn */ | |
386 | struct xlog_ticket *ticket; /* chkpt ticket */ | |
387 | int nvecs; /* number of regions */ | |
388 | int space_used; /* aggregate size of regions */ | |
389 | struct list_head busy_extents; /* busy extents in chkpt */ | |
390 | struct xfs_log_vec *lv_chain; /* logvecs being pushed */ | |
391 | xfs_log_callback_t log_cb; /* completion callback hook. */ | |
392 | struct list_head committing; /* ctx committing list */ | |
393 | }; | |
394 | ||
395 | /* | |
396 | * Committed Item List structure | |
397 | * | |
398 | * This structure is used to track log items that have been committed but not | |
399 | * yet written into the log. It is used only when the delayed logging mount | |
400 | * option is enabled. | |
401 | * | |
402 | * This structure tracks the list of committing checkpoint contexts so | |
403 | * we can avoid the problem of having to hold out new transactions during a | |
404 | * flush until we have a the commit record LSN of the checkpoint. We can | |
405 | * traverse the list of committing contexts in xlog_cil_push_lsn() to find a | |
406 | * sequence match and extract the commit LSN directly from there. If the | |
407 | * checkpoint is still in the process of committing, we can block waiting for | |
408 | * the commit LSN to be determined as well. This should make synchronous | |
409 | * operations almost as efficient as the old logging methods. | |
410 | */ | |
411 | struct xfs_cil { | |
412 | struct log *xc_log; | |
413 | struct list_head xc_cil; | |
414 | spinlock_t xc_cil_lock; | |
415 | struct xfs_cil_ctx *xc_ctx; | |
416 | struct rw_semaphore xc_ctx_lock; | |
417 | struct list_head xc_committing; | |
eb40a875 | 418 | wait_queue_head_t xc_commit_wait; |
a44f13ed | 419 | xfs_lsn_t xc_current_sequence; |
71e330b5 DC |
420 | }; |
421 | ||
df806158 | 422 | /* |
80168676 DC |
423 | * The amount of log space we allow the CIL to aggregate is difficult to size. |
424 | * Whatever we choose, we have to make sure we can get a reservation for the | |
425 | * log space effectively, that it is large enough to capture sufficient | |
426 | * relogging to reduce log buffer IO significantly, but it is not too large for | |
427 | * the log or induces too much latency when writing out through the iclogs. We | |
428 | * track both space consumed and the number of vectors in the checkpoint | |
429 | * context, so we need to decide which to use for limiting. | |
df806158 DC |
430 | * |
431 | * Every log buffer we write out during a push needs a header reserved, which | |
432 | * is at least one sector and more for v2 logs. Hence we need a reservation of | |
433 | * at least 512 bytes per 32k of log space just for the LR headers. That means | |
434 | * 16KB of reservation per megabyte of delayed logging space we will consume, | |
435 | * plus various headers. The number of headers will vary based on the num of | |
436 | * io vectors, so limiting on a specific number of vectors is going to result | |
437 | * in transactions of varying size. IOWs, it is more consistent to track and | |
438 | * limit space consumed in the log rather than by the number of objects being | |
439 | * logged in order to prevent checkpoint ticket overruns. | |
440 | * | |
441 | * Further, use of static reservations through the log grant mechanism is | |
442 | * problematic. It introduces a lot of complexity (e.g. reserve grant vs write | |
443 | * grant) and a significant deadlock potential because regranting write space | |
444 | * can block on log pushes. Hence if we have to regrant log space during a log | |
445 | * push, we can deadlock. | |
446 | * | |
447 | * However, we can avoid this by use of a dynamic "reservation stealing" | |
448 | * technique during transaction commit whereby unused reservation space in the | |
449 | * transaction ticket is transferred to the CIL ctx commit ticket to cover the | |
450 | * space needed by the checkpoint transaction. This means that we never need to | |
451 | * specifically reserve space for the CIL checkpoint transaction, nor do we | |
452 | * need to regrant space once the checkpoint completes. This also means the | |
453 | * checkpoint transaction ticket is specific to the checkpoint context, rather | |
454 | * than the CIL itself. | |
455 | * | |
80168676 DC |
456 | * With dynamic reservations, we can effectively make up arbitrary limits for |
457 | * the checkpoint size so long as they don't violate any other size rules. | |
458 | * Recovery imposes a rule that no transaction exceed half the log, so we are | |
459 | * limited by that. Furthermore, the log transaction reservation subsystem | |
460 | * tries to keep 25% of the log free, so we need to keep below that limit or we | |
461 | * risk running out of free log space to start any new transactions. | |
462 | * | |
463 | * In order to keep background CIL push efficient, we will set a lower | |
464 | * threshold at which background pushing is attempted without blocking current | |
465 | * transaction commits. A separate, higher bound defines when CIL pushes are | |
466 | * enforced to ensure we stay within our maximum checkpoint size bounds. | |
467 | * threshold, yet give us plenty of space for aggregation on large logs. | |
df806158 | 468 | */ |
80168676 DC |
469 | #define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3) |
470 | #define XLOG_CIL_HARD_SPACE_LIMIT(log) (3 * (log->l_logsize >> 4)) | |
df806158 | 471 | |
1da177e4 LT |
472 | /* |
473 | * The reservation head lsn is not made up of a cycle number and block number. | |
474 | * Instead, it uses a cycle number and byte number. Logs don't expect to | |
475 | * overflow 31 bits worth of byte offset, so using a byte number will mean | |
476 | * that round off problems won't occur when releasing partial reservations. | |
477 | */ | |
478 | typedef struct log { | |
4679b2d3 DC |
479 | /* The following fields don't need locking */ |
480 | struct xfs_mount *l_mp; /* mount point */ | |
a9c21c1b | 481 | struct xfs_ail *l_ailp; /* AIL log is working with */ |
71e330b5 | 482 | struct xfs_cil *l_cilp; /* CIL log is working with */ |
4679b2d3 DC |
483 | struct xfs_buf *l_xbuf; /* extra buffer for log |
484 | * wrapping */ | |
485 | struct xfs_buftarg *l_targ; /* buftarg of log */ | |
486 | uint l_flags; | |
487 | uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */ | |
d5689eaa | 488 | struct list_head *l_buf_cancel_table; |
4679b2d3 DC |
489 | int l_iclog_hsize; /* size of iclog header */ |
490 | int l_iclog_heads; /* # of iclog header sectors */ | |
48389ef1 | 491 | uint l_sectBBsize; /* sector size in BBs (2^n) */ |
4679b2d3 DC |
492 | int l_iclog_size; /* size of log in bytes */ |
493 | int l_iclog_size_log; /* log power size of log */ | |
494 | int l_iclog_bufs; /* number of iclog buffers */ | |
495 | xfs_daddr_t l_logBBstart; /* start block of log */ | |
496 | int l_logsize; /* size of log in bytes */ | |
497 | int l_logBBsize; /* size of log in BB chunks */ | |
498 | ||
1da177e4 | 499 | /* The following block of fields are changed while holding icloglock */ |
eb40a875 | 500 | wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp; |
d748c623 | 501 | /* waiting for iclog flush */ |
1da177e4 LT |
502 | int l_covered_state;/* state of "covering disk |
503 | * log entries" */ | |
1da177e4 | 504 | xlog_in_core_t *l_iclog; /* head log queue */ |
b22cd72c | 505 | spinlock_t l_icloglock; /* grab to change iclog state */ |
1da177e4 LT |
506 | int l_curr_cycle; /* Cycle number of log writes */ |
507 | int l_prev_cycle; /* Cycle number before last | |
508 | * block increment */ | |
509 | int l_curr_block; /* current logical log block */ | |
510 | int l_prev_block; /* previous logical log block */ | |
1da177e4 | 511 | |
84f3c683 | 512 | /* |
1c3cb9ec DC |
513 | * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and |
514 | * read without needing to hold specific locks. To avoid operations | |
515 | * contending with other hot objects, place each of them on a separate | |
516 | * cacheline. | |
84f3c683 DC |
517 | */ |
518 | /* lsn of last LR on disk */ | |
519 | atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp; | |
1c3cb9ec DC |
520 | /* lsn of 1st LR with unflushed * buffers */ |
521 | atomic64_t l_tail_lsn ____cacheline_aligned_in_smp; | |
84f3c683 | 522 | |
3f16b985 DC |
523 | /* |
524 | * ticket grant locks, queues and accounting have their own cachlines | |
525 | * as these are quite hot and can be operated on concurrently. | |
526 | */ | |
527 | spinlock_t l_grant_reserve_lock ____cacheline_aligned_in_smp; | |
528 | struct list_head l_reserveq; | |
529 | atomic64_t l_grant_reserve_head; | |
530 | ||
531 | spinlock_t l_grant_write_lock ____cacheline_aligned_in_smp; | |
532 | struct list_head l_writeq; | |
533 | atomic64_t l_grant_write_head; | |
534 | ||
4679b2d3 DC |
535 | /* The following field are used for debugging; need to hold icloglock */ |
536 | #ifdef DEBUG | |
537 | char *l_iclog_bak[XLOG_MAX_ICLOGS]; | |
538 | #endif | |
539 | ||
1da177e4 LT |
540 | } xlog_t; |
541 | ||
d5689eaa CH |
542 | #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \ |
543 | ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE)) | |
544 | ||
cfcbbbd0 NS |
545 | #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR) |
546 | ||
1da177e4 LT |
547 | /* common routines */ |
548 | extern xfs_lsn_t xlog_assign_tail_lsn(struct xfs_mount *mp); | |
65be6054 | 549 | extern int xlog_recover(xlog_t *log); |
4249023a | 550 | extern int xlog_recover_finish(xlog_t *log); |
1da177e4 | 551 | extern void xlog_pack_data(xlog_t *log, xlog_in_core_t *iclog, int); |
1da177e4 | 552 | |
71e330b5 DC |
553 | extern kmem_zone_t *xfs_log_ticket_zone; |
554 | struct xlog_ticket *xlog_ticket_alloc(struct log *log, int unit_bytes, | |
555 | int count, char client, uint xflags, | |
556 | int alloc_flags); | |
557 | ||
eb01c9cd | 558 | |
e6b1f273 CH |
559 | static inline void |
560 | xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes) | |
561 | { | |
562 | *ptr += bytes; | |
563 | *len -= bytes; | |
564 | *off += bytes; | |
565 | } | |
566 | ||
71e330b5 DC |
567 | void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket); |
568 | int xlog_write(struct log *log, struct xfs_log_vec *log_vector, | |
569 | struct xlog_ticket *tic, xfs_lsn_t *start_lsn, | |
570 | xlog_in_core_t **commit_iclog, uint flags); | |
571 | ||
1c3cb9ec DC |
572 | /* |
573 | * When we crack an atomic LSN, we sample it first so that the value will not | |
574 | * change while we are cracking it into the component values. This means we | |
575 | * will always get consistent component values to work from. This should always | |
25985edc | 576 | * be used to sample and crack LSNs that are stored and updated in atomic |
1c3cb9ec DC |
577 | * variables. |
578 | */ | |
579 | static inline void | |
580 | xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block) | |
581 | { | |
582 | xfs_lsn_t val = atomic64_read(lsn); | |
583 | ||
584 | *cycle = CYCLE_LSN(val); | |
585 | *block = BLOCK_LSN(val); | |
586 | } | |
587 | ||
588 | /* | |
589 | * Calculate and assign a value to an atomic LSN variable from component pieces. | |
590 | */ | |
591 | static inline void | |
592 | xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block) | |
593 | { | |
594 | atomic64_set(lsn, xlog_assign_lsn(cycle, block)); | |
595 | } | |
596 | ||
a69ed03c | 597 | /* |
d0eb2f38 | 598 | * When we crack the grant head, we sample it first so that the value will not |
a69ed03c DC |
599 | * change while we are cracking it into the component values. This means we |
600 | * will always get consistent component values to work from. | |
601 | */ | |
602 | static inline void | |
d0eb2f38 | 603 | xlog_crack_grant_head_val(int64_t val, int *cycle, int *space) |
a69ed03c | 604 | { |
a69ed03c DC |
605 | *cycle = val >> 32; |
606 | *space = val & 0xffffffff; | |
607 | } | |
608 | ||
d0eb2f38 DC |
609 | static inline void |
610 | xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space) | |
611 | { | |
612 | xlog_crack_grant_head_val(atomic64_read(head), cycle, space); | |
613 | } | |
614 | ||
615 | static inline int64_t | |
616 | xlog_assign_grant_head_val(int cycle, int space) | |
617 | { | |
618 | return ((int64_t)cycle << 32) | space; | |
619 | } | |
620 | ||
a69ed03c | 621 | static inline void |
c8a09ff8 | 622 | xlog_assign_grant_head(atomic64_t *head, int cycle, int space) |
a69ed03c | 623 | { |
d0eb2f38 | 624 | atomic64_set(head, xlog_assign_grant_head_val(cycle, space)); |
a69ed03c DC |
625 | } |
626 | ||
71e330b5 DC |
627 | /* |
628 | * Committed Item List interfaces | |
629 | */ | |
630 | int xlog_cil_init(struct log *log); | |
631 | void xlog_cil_init_post_recovery(struct log *log); | |
632 | void xlog_cil_destroy(struct log *log); | |
633 | ||
a44f13ed DC |
634 | /* |
635 | * CIL force routines | |
636 | */ | |
637 | xfs_lsn_t xlog_cil_force_lsn(struct log *log, xfs_lsn_t sequence); | |
638 | ||
639 | static inline void | |
640 | xlog_cil_force(struct log *log) | |
641 | { | |
642 | xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence); | |
643 | } | |
71e330b5 | 644 | |
955e47ad TS |
645 | /* |
646 | * Unmount record type is used as a pseudo transaction type for the ticket. | |
647 | * It's value must be outside the range of XFS_TRANS_* values. | |
648 | */ | |
649 | #define XLOG_UNMOUNT_REC_TYPE (-1U) | |
650 | ||
eb40a875 DC |
651 | /* |
652 | * Wrapper function for waiting on a wait queue serialised against wakeups | |
653 | * by a spinlock. This matches the semantics of all the wait queues used in the | |
654 | * log code. | |
655 | */ | |
656 | static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock) | |
657 | { | |
658 | DECLARE_WAITQUEUE(wait, current); | |
659 | ||
660 | add_wait_queue_exclusive(wq, &wait); | |
661 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
662 | spin_unlock(lock); | |
663 | schedule(); | |
664 | remove_wait_queue(wq, &wait); | |
665 | } | |
1da177e4 LT |
666 | #endif /* __KERNEL__ */ |
667 | ||
668 | #endif /* __XFS_LOG_PRIV_H__ */ |