xfs_log_priv.h 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614
  1. /*
  2. * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  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
  7. * published by the Free Software Foundation.
  8. *
  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.
  13. *
  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
  17. */
  18. #ifndef __XFS_LOG_PRIV_H__
  19. #define __XFS_LOG_PRIV_H__
  20. struct xfs_buf;
  21. struct xlog;
  22. struct xlog_ticket;
  23. struct xfs_mount;
  24. struct xfs_log_callback;
  25. /*
  26. * Flags for log structure
  27. */
  28. #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */
  29. #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */
  30. #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being
  31. shutdown */
  32. #define XLOG_TAIL_WARN 0x10 /* log tail verify warning issued */
  33. /*
  34. * get client id from packed copy.
  35. *
  36. * this hack is here because the xlog_pack code copies four bytes
  37. * of xlog_op_header containing the fields oh_clientid, oh_flags
  38. * and oh_res2 into the packed copy.
  39. *
  40. * later on this four byte chunk is treated as an int and the
  41. * client id is pulled out.
  42. *
  43. * this has endian issues, of course.
  44. */
  45. static inline uint xlog_get_client_id(__be32 i)
  46. {
  47. return be32_to_cpu(i) >> 24;
  48. }
  49. /*
  50. * In core log state
  51. */
  52. #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */
  53. #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
  54. #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */
  55. #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
  56. #define XLOG_STATE_DO_CALLBACK \
  57. 0x0010 /* Process callback functions */
  58. #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */
  59. #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/
  60. #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */
  61. #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */
  62. #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */
  63. /*
  64. * Flags to log ticket
  65. */
  66. #define XLOG_TIC_INITED 0x1 /* has been initialized */
  67. #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */
  68. #define XLOG_TIC_FLAGS \
  69. { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \
  70. { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
  71. /*
  72. * Below are states for covering allocation transactions.
  73. * By covering, we mean changing the h_tail_lsn in the last on-disk
  74. * log write such that no allocation transactions will be re-done during
  75. * recovery after a system crash. Recovery starts at the last on-disk
  76. * log write.
  77. *
  78. * These states are used to insert dummy log entries to cover
  79. * space allocation transactions which can undo non-transactional changes
  80. * after a crash. Writes to a file with space
  81. * already allocated do not result in any transactions. Allocations
  82. * might include space beyond the EOF. So if we just push the EOF a
  83. * little, the last transaction for the file could contain the wrong
  84. * size. If there is no file system activity, after an allocation
  85. * transaction, and the system crashes, the allocation transaction
  86. * will get replayed and the file will be truncated. This could
  87. * be hours/days/... after the allocation occurred.
  88. *
  89. * The fix for this is to do two dummy transactions when the
  90. * system is idle. We need two dummy transaction because the h_tail_lsn
  91. * in the log record header needs to point beyond the last possible
  92. * non-dummy transaction. The first dummy changes the h_tail_lsn to
  93. * the first transaction before the dummy. The second dummy causes
  94. * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
  95. *
  96. * These dummy transactions get committed when everything
  97. * is idle (after there has been some activity).
  98. *
  99. * There are 5 states used to control this.
  100. *
  101. * IDLE -- no logging has been done on the file system or
  102. * we are done covering previous transactions.
  103. * NEED -- logging has occurred and we need a dummy transaction
  104. * when the log becomes idle.
  105. * DONE -- we were in the NEED state and have committed a dummy
  106. * transaction.
  107. * NEED2 -- we detected that a dummy transaction has gone to the
  108. * on disk log with no other transactions.
  109. * DONE2 -- we committed a dummy transaction when in the NEED2 state.
  110. *
  111. * There are two places where we switch states:
  112. *
  113. * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
  114. * We commit the dummy transaction and switch to DONE or DONE2,
  115. * respectively. In all other states, we don't do anything.
  116. *
  117. * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
  118. *
  119. * No matter what state we are in, if this isn't the dummy
  120. * transaction going out, the next state is NEED.
  121. * So, if we aren't in the DONE or DONE2 states, the next state
  122. * is NEED. We can't be finishing a write of the dummy record
  123. * unless it was committed and the state switched to DONE or DONE2.
  124. *
  125. * If we are in the DONE state and this was a write of the
  126. * dummy transaction, we move to NEED2.
  127. *
  128. * If we are in the DONE2 state and this was a write of the
  129. * dummy transaction, we move to IDLE.
  130. *
  131. *
  132. * Writing only one dummy transaction can get appended to
  133. * one file space allocation. When this happens, the log recovery
  134. * code replays the space allocation and a file could be truncated.
  135. * This is why we have the NEED2 and DONE2 states before going idle.
  136. */
  137. #define XLOG_STATE_COVER_IDLE 0
  138. #define XLOG_STATE_COVER_NEED 1
  139. #define XLOG_STATE_COVER_DONE 2
  140. #define XLOG_STATE_COVER_NEED2 3
  141. #define XLOG_STATE_COVER_DONE2 4
  142. #define XLOG_COVER_OPS 5
  143. /* Ticket reservation region accounting */
  144. #define XLOG_TIC_LEN_MAX 15
  145. /*
  146. * Reservation region
  147. * As would be stored in xfs_log_iovec but without the i_addr which
  148. * we don't care about.
  149. */
  150. typedef struct xlog_res {
  151. uint r_len; /* region length :4 */
  152. uint r_type; /* region's transaction type :4 */
  153. } xlog_res_t;
  154. typedef struct xlog_ticket {
  155. struct list_head t_queue; /* reserve/write queue */
  156. struct task_struct *t_task; /* task that owns this ticket */
  157. xlog_tid_t t_tid; /* transaction identifier : 4 */
  158. atomic_t t_ref; /* ticket reference count : 4 */
  159. int t_curr_res; /* current reservation in bytes : 4 */
  160. int t_unit_res; /* unit reservation in bytes : 4 */
  161. char t_ocnt; /* original count : 1 */
  162. char t_cnt; /* current count : 1 */
  163. char t_clientid; /* who does this belong to; : 1 */
  164. char t_flags; /* properties of reservation : 1 */
  165. uint t_trans_type; /* transaction type : 4 */
  166. /* reservation array fields */
  167. uint t_res_num; /* num in array : 4 */
  168. uint t_res_num_ophdrs; /* num op hdrs : 4 */
  169. uint t_res_arr_sum; /* array sum : 4 */
  170. uint t_res_o_flow; /* sum overflow : 4 */
  171. xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */
  172. } xlog_ticket_t;
  173. /*
  174. * - A log record header is 512 bytes. There is plenty of room to grow the
  175. * xlog_rec_header_t into the reserved space.
  176. * - ic_data follows, so a write to disk can start at the beginning of
  177. * the iclog.
  178. * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
  179. * - ic_next is the pointer to the next iclog in the ring.
  180. * - ic_bp is a pointer to the buffer used to write this incore log to disk.
  181. * - ic_log is a pointer back to the global log structure.
  182. * - ic_callback is a linked list of callback function/argument pairs to be
  183. * called after an iclog finishes writing.
  184. * - ic_size is the full size of the header plus data.
  185. * - ic_offset is the current number of bytes written to in this iclog.
  186. * - ic_refcnt is bumped when someone is writing to the log.
  187. * - ic_state is the state of the iclog.
  188. *
  189. * Because of cacheline contention on large machines, we need to separate
  190. * various resources onto different cachelines. To start with, make the
  191. * structure cacheline aligned. The following fields can be contended on
  192. * by independent processes:
  193. *
  194. * - ic_callback_*
  195. * - ic_refcnt
  196. * - fields protected by the global l_icloglock
  197. *
  198. * so we need to ensure that these fields are located in separate cachelines.
  199. * We'll put all the read-only and l_icloglock fields in the first cacheline,
  200. * and move everything else out to subsequent cachelines.
  201. */
  202. typedef struct xlog_in_core {
  203. wait_queue_head_t ic_force_wait;
  204. wait_queue_head_t ic_write_wait;
  205. struct xlog_in_core *ic_next;
  206. struct xlog_in_core *ic_prev;
  207. struct xfs_buf *ic_bp;
  208. struct xlog *ic_log;
  209. int ic_size;
  210. int ic_offset;
  211. int ic_bwritecnt;
  212. unsigned short ic_state;
  213. char *ic_datap; /* pointer to iclog data */
  214. /* Callback structures need their own cacheline */
  215. spinlock_t ic_callback_lock ____cacheline_aligned_in_smp;
  216. struct xfs_log_callback *ic_callback;
  217. struct xfs_log_callback **ic_callback_tail;
  218. /* reference counts need their own cacheline */
  219. atomic_t ic_refcnt ____cacheline_aligned_in_smp;
  220. xlog_in_core_2_t *ic_data;
  221. #define ic_header ic_data->hic_header
  222. } xlog_in_core_t;
  223. /*
  224. * The CIL context is used to aggregate per-transaction details as well be
  225. * passed to the iclog for checkpoint post-commit processing. After being
  226. * passed to the iclog, another context needs to be allocated for tracking the
  227. * next set of transactions to be aggregated into a checkpoint.
  228. */
  229. struct xfs_cil;
  230. struct xfs_cil_ctx {
  231. struct xfs_cil *cil;
  232. xfs_lsn_t sequence; /* chkpt sequence # */
  233. xfs_lsn_t start_lsn; /* first LSN of chkpt commit */
  234. xfs_lsn_t commit_lsn; /* chkpt commit record lsn */
  235. struct xlog_ticket *ticket; /* chkpt ticket */
  236. int nvecs; /* number of regions */
  237. int space_used; /* aggregate size of regions */
  238. struct list_head busy_extents; /* busy extents in chkpt */
  239. struct xfs_log_vec *lv_chain; /* logvecs being pushed */
  240. struct xfs_log_callback log_cb; /* completion callback hook. */
  241. struct list_head committing; /* ctx committing list */
  242. };
  243. /*
  244. * Committed Item List structure
  245. *
  246. * This structure is used to track log items that have been committed but not
  247. * yet written into the log. It is used only when the delayed logging mount
  248. * option is enabled.
  249. *
  250. * This structure tracks the list of committing checkpoint contexts so
  251. * we can avoid the problem of having to hold out new transactions during a
  252. * flush until we have a the commit record LSN of the checkpoint. We can
  253. * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
  254. * sequence match and extract the commit LSN directly from there. If the
  255. * checkpoint is still in the process of committing, we can block waiting for
  256. * the commit LSN to be determined as well. This should make synchronous
  257. * operations almost as efficient as the old logging methods.
  258. */
  259. struct xfs_cil {
  260. struct xlog *xc_log;
  261. struct list_head xc_cil;
  262. spinlock_t xc_cil_lock;
  263. struct rw_semaphore xc_ctx_lock ____cacheline_aligned_in_smp;
  264. struct xfs_cil_ctx *xc_ctx;
  265. spinlock_t xc_push_lock ____cacheline_aligned_in_smp;
  266. xfs_lsn_t xc_push_seq;
  267. struct list_head xc_committing;
  268. wait_queue_head_t xc_commit_wait;
  269. xfs_lsn_t xc_current_sequence;
  270. struct work_struct xc_push_work;
  271. } ____cacheline_aligned_in_smp;
  272. /*
  273. * The amount of log space we allow the CIL to aggregate is difficult to size.
  274. * Whatever we choose, we have to make sure we can get a reservation for the
  275. * log space effectively, that it is large enough to capture sufficient
  276. * relogging to reduce log buffer IO significantly, but it is not too large for
  277. * the log or induces too much latency when writing out through the iclogs. We
  278. * track both space consumed and the number of vectors in the checkpoint
  279. * context, so we need to decide which to use for limiting.
  280. *
  281. * Every log buffer we write out during a push needs a header reserved, which
  282. * is at least one sector and more for v2 logs. Hence we need a reservation of
  283. * at least 512 bytes per 32k of log space just for the LR headers. That means
  284. * 16KB of reservation per megabyte of delayed logging space we will consume,
  285. * plus various headers. The number of headers will vary based on the num of
  286. * io vectors, so limiting on a specific number of vectors is going to result
  287. * in transactions of varying size. IOWs, it is more consistent to track and
  288. * limit space consumed in the log rather than by the number of objects being
  289. * logged in order to prevent checkpoint ticket overruns.
  290. *
  291. * Further, use of static reservations through the log grant mechanism is
  292. * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
  293. * grant) and a significant deadlock potential because regranting write space
  294. * can block on log pushes. Hence if we have to regrant log space during a log
  295. * push, we can deadlock.
  296. *
  297. * However, we can avoid this by use of a dynamic "reservation stealing"
  298. * technique during transaction commit whereby unused reservation space in the
  299. * transaction ticket is transferred to the CIL ctx commit ticket to cover the
  300. * space needed by the checkpoint transaction. This means that we never need to
  301. * specifically reserve space for the CIL checkpoint transaction, nor do we
  302. * need to regrant space once the checkpoint completes. This also means the
  303. * checkpoint transaction ticket is specific to the checkpoint context, rather
  304. * than the CIL itself.
  305. *
  306. * With dynamic reservations, we can effectively make up arbitrary limits for
  307. * the checkpoint size so long as they don't violate any other size rules.
  308. * Recovery imposes a rule that no transaction exceed half the log, so we are
  309. * limited by that. Furthermore, the log transaction reservation subsystem
  310. * tries to keep 25% of the log free, so we need to keep below that limit or we
  311. * risk running out of free log space to start any new transactions.
  312. *
  313. * In order to keep background CIL push efficient, we will set a lower
  314. * threshold at which background pushing is attempted without blocking current
  315. * transaction commits. A separate, higher bound defines when CIL pushes are
  316. * enforced to ensure we stay within our maximum checkpoint size bounds.
  317. * threshold, yet give us plenty of space for aggregation on large logs.
  318. */
  319. #define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3)
  320. /*
  321. * ticket grant locks, queues and accounting have their own cachlines
  322. * as these are quite hot and can be operated on concurrently.
  323. */
  324. struct xlog_grant_head {
  325. spinlock_t lock ____cacheline_aligned_in_smp;
  326. struct list_head waiters;
  327. atomic64_t grant;
  328. };
  329. /*
  330. * The reservation head lsn is not made up of a cycle number and block number.
  331. * Instead, it uses a cycle number and byte number. Logs don't expect to
  332. * overflow 31 bits worth of byte offset, so using a byte number will mean
  333. * that round off problems won't occur when releasing partial reservations.
  334. */
  335. struct xlog {
  336. /* The following fields don't need locking */
  337. struct xfs_mount *l_mp; /* mount point */
  338. struct xfs_ail *l_ailp; /* AIL log is working with */
  339. struct xfs_cil *l_cilp; /* CIL log is working with */
  340. struct xfs_buf *l_xbuf; /* extra buffer for log
  341. * wrapping */
  342. struct xfs_buftarg *l_targ; /* buftarg of log */
  343. struct delayed_work l_work; /* background flush work */
  344. uint l_flags;
  345. uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
  346. struct list_head *l_buf_cancel_table;
  347. int l_iclog_hsize; /* size of iclog header */
  348. int l_iclog_heads; /* # of iclog header sectors */
  349. uint l_sectBBsize; /* sector size in BBs (2^n) */
  350. int l_iclog_size; /* size of log in bytes */
  351. int l_iclog_size_log; /* log power size of log */
  352. int l_iclog_bufs; /* number of iclog buffers */
  353. xfs_daddr_t l_logBBstart; /* start block of log */
  354. int l_logsize; /* size of log in bytes */
  355. int l_logBBsize; /* size of log in BB chunks */
  356. /* The following block of fields are changed while holding icloglock */
  357. wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp;
  358. /* waiting for iclog flush */
  359. int l_covered_state;/* state of "covering disk
  360. * log entries" */
  361. xlog_in_core_t *l_iclog; /* head log queue */
  362. spinlock_t l_icloglock; /* grab to change iclog state */
  363. int l_curr_cycle; /* Cycle number of log writes */
  364. int l_prev_cycle; /* Cycle number before last
  365. * block increment */
  366. int l_curr_block; /* current logical log block */
  367. int l_prev_block; /* previous logical log block */
  368. /*
  369. * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
  370. * read without needing to hold specific locks. To avoid operations
  371. * contending with other hot objects, place each of them on a separate
  372. * cacheline.
  373. */
  374. /* lsn of last LR on disk */
  375. atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp;
  376. /* lsn of 1st LR with unflushed * buffers */
  377. atomic64_t l_tail_lsn ____cacheline_aligned_in_smp;
  378. struct xlog_grant_head l_reserve_head;
  379. struct xlog_grant_head l_write_head;
  380. struct xfs_kobj l_kobj;
  381. /* The following field are used for debugging; need to hold icloglock */
  382. #ifdef DEBUG
  383. void *l_iclog_bak[XLOG_MAX_ICLOGS];
  384. #endif
  385. };
  386. #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
  387. ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE))
  388. #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR)
  389. /* common routines */
  390. extern int
  391. xlog_recover(
  392. struct xlog *log);
  393. extern int
  394. xlog_recover_finish(
  395. struct xlog *log);
  396. extern int
  397. xlog_recover_cancel(struct xlog *);
  398. extern __le32 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
  399. char *dp, int size);
  400. extern kmem_zone_t *xfs_log_ticket_zone;
  401. struct xlog_ticket *
  402. xlog_ticket_alloc(
  403. struct xlog *log,
  404. int unit_bytes,
  405. int count,
  406. char client,
  407. bool permanent,
  408. xfs_km_flags_t alloc_flags);
  409. static inline void
  410. xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
  411. {
  412. *ptr += bytes;
  413. *len -= bytes;
  414. *off += bytes;
  415. }
  416. void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
  417. int
  418. xlog_write(
  419. struct xlog *log,
  420. struct xfs_log_vec *log_vector,
  421. struct xlog_ticket *tic,
  422. xfs_lsn_t *start_lsn,
  423. struct xlog_in_core **commit_iclog,
  424. uint flags);
  425. /*
  426. * When we crack an atomic LSN, we sample it first so that the value will not
  427. * change while we are cracking it into the component values. This means we
  428. * will always get consistent component values to work from. This should always
  429. * be used to sample and crack LSNs that are stored and updated in atomic
  430. * variables.
  431. */
  432. static inline void
  433. xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
  434. {
  435. xfs_lsn_t val = atomic64_read(lsn);
  436. *cycle = CYCLE_LSN(val);
  437. *block = BLOCK_LSN(val);
  438. }
  439. /*
  440. * Calculate and assign a value to an atomic LSN variable from component pieces.
  441. */
  442. static inline void
  443. xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
  444. {
  445. atomic64_set(lsn, xlog_assign_lsn(cycle, block));
  446. }
  447. /*
  448. * When we crack the grant head, we sample it first so that the value will not
  449. * change while we are cracking it into the component values. This means we
  450. * will always get consistent component values to work from.
  451. */
  452. static inline void
  453. xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
  454. {
  455. *cycle = val >> 32;
  456. *space = val & 0xffffffff;
  457. }
  458. static inline void
  459. xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
  460. {
  461. xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
  462. }
  463. static inline int64_t
  464. xlog_assign_grant_head_val(int cycle, int space)
  465. {
  466. return ((int64_t)cycle << 32) | space;
  467. }
  468. static inline void
  469. xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
  470. {
  471. atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
  472. }
  473. /*
  474. * Committed Item List interfaces
  475. */
  476. int xlog_cil_init(struct xlog *log);
  477. void xlog_cil_init_post_recovery(struct xlog *log);
  478. void xlog_cil_destroy(struct xlog *log);
  479. bool xlog_cil_empty(struct xlog *log);
  480. /*
  481. * CIL force routines
  482. */
  483. xfs_lsn_t
  484. xlog_cil_force_lsn(
  485. struct xlog *log,
  486. xfs_lsn_t sequence);
  487. static inline void
  488. xlog_cil_force(struct xlog *log)
  489. {
  490. xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
  491. }
  492. /*
  493. * Unmount record type is used as a pseudo transaction type for the ticket.
  494. * It's value must be outside the range of XFS_TRANS_* values.
  495. */
  496. #define XLOG_UNMOUNT_REC_TYPE (-1U)
  497. /*
  498. * Wrapper function for waiting on a wait queue serialised against wakeups
  499. * by a spinlock. This matches the semantics of all the wait queues used in the
  500. * log code.
  501. */
  502. static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
  503. {
  504. DECLARE_WAITQUEUE(wait, current);
  505. add_wait_queue_exclusive(wq, &wait);
  506. __set_current_state(TASK_UNINTERRUPTIBLE);
  507. spin_unlock(lock);
  508. schedule();
  509. remove_wait_queue(wq, &wait);
  510. }
  511. /*
  512. * The LSN is valid so long as it is behind the current LSN. If it isn't, this
  513. * means that the next log record that includes this metadata could have a
  514. * smaller LSN. In turn, this means that the modification in the log would not
  515. * replay.
  516. */
  517. static inline bool
  518. xlog_valid_lsn(
  519. struct xlog *log,
  520. xfs_lsn_t lsn)
  521. {
  522. int cur_cycle;
  523. int cur_block;
  524. bool valid = true;
  525. /*
  526. * First, sample the current lsn without locking to avoid added
  527. * contention from metadata I/O. The current cycle and block are updated
  528. * (in xlog_state_switch_iclogs()) and read here in a particular order
  529. * to avoid false negatives (e.g., thinking the metadata LSN is valid
  530. * when it is not).
  531. *
  532. * The current block is always rewound before the cycle is bumped in
  533. * xlog_state_switch_iclogs() to ensure the current LSN is never seen in
  534. * a transiently forward state. Instead, we can see the LSN in a
  535. * transiently behind state if we happen to race with a cycle wrap.
  536. */
  537. cur_cycle = ACCESS_ONCE(log->l_curr_cycle);
  538. smp_rmb();
  539. cur_block = ACCESS_ONCE(log->l_curr_block);
  540. if ((CYCLE_LSN(lsn) > cur_cycle) ||
  541. (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) {
  542. /*
  543. * If the metadata LSN appears invalid, it's possible the check
  544. * above raced with a wrap to the next log cycle. Grab the lock
  545. * to check for sure.
  546. */
  547. spin_lock(&log->l_icloglock);
  548. cur_cycle = log->l_curr_cycle;
  549. cur_block = log->l_curr_block;
  550. spin_unlock(&log->l_icloglock);
  551. if ((CYCLE_LSN(lsn) > cur_cycle) ||
  552. (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block))
  553. valid = false;
  554. }
  555. return valid;
  556. }
  557. #endif /* __XFS_LOG_PRIV_H__ */