1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. 4 * All Rights Reserved. 5 */ 6 #ifndef __XFS_LOG_PRIV_H__ 7 #define __XFS_LOG_PRIV_H__ 8 9 struct xfs_buf; 10 struct xlog; 11 struct xlog_ticket; 12 struct xfs_mount; 13 14 /* 15 * Flags for log structure 16 */ 17 #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */ 18 #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */ 19 #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being 20 shutdown */ 21 #define XLOG_TAIL_WARN 0x10 /* log tail verify warning issued */ 22 23 /* 24 * get client id from packed copy. 25 * 26 * this hack is here because the xlog_pack code copies four bytes 27 * of xlog_op_header containing the fields oh_clientid, oh_flags 28 * and oh_res2 into the packed copy. 29 * 30 * later on this four byte chunk is treated as an int and the 31 * client id is pulled out. 32 * 33 * this has endian issues, of course. 34 */ 35 static inline uint xlog_get_client_id(__be32 i) 36 { 37 return be32_to_cpu(i) >> 24; 38 } 39 40 /* 41 * In core log state 42 */ 43 enum xlog_iclog_state { 44 XLOG_STATE_ACTIVE, /* Current IC log being written to */ 45 XLOG_STATE_WANT_SYNC, /* Want to sync this iclog; no more writes */ 46 XLOG_STATE_SYNCING, /* This IC log is syncing */ 47 XLOG_STATE_DONE_SYNC, /* Done syncing to disk */ 48 XLOG_STATE_CALLBACK, /* Callback functions now */ 49 XLOG_STATE_DIRTY, /* Dirty IC log, not ready for ACTIVE status */ 50 XLOG_STATE_IOERROR, /* IO error happened in sync'ing log */ 51 }; 52 53 #define XLOG_STATE_STRINGS \ 54 { XLOG_STATE_ACTIVE, "XLOG_STATE_ACTIVE" }, \ 55 { XLOG_STATE_WANT_SYNC, "XLOG_STATE_WANT_SYNC" }, \ 56 { XLOG_STATE_SYNCING, "XLOG_STATE_SYNCING" }, \ 57 { XLOG_STATE_DONE_SYNC, "XLOG_STATE_DONE_SYNC" }, \ 58 { XLOG_STATE_CALLBACK, "XLOG_STATE_CALLBACK" }, \ 59 { XLOG_STATE_DIRTY, "XLOG_STATE_DIRTY" }, \ 60 { XLOG_STATE_IOERROR, "XLOG_STATE_IOERROR" } 61 62 63 /* 64 * Log ticket flags 65 */ 66 #define XLOG_TIC_PERM_RESERV 0x1 /* permanent reservation */ 67 68 #define XLOG_TIC_FLAGS \ 69 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" } 70 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 */ 137 138 #define XLOG_STATE_COVER_IDLE 0 139 #define XLOG_STATE_COVER_NEED 1 140 #define XLOG_STATE_COVER_DONE 2 141 #define XLOG_STATE_COVER_NEED2 3 142 #define XLOG_STATE_COVER_DONE2 4 143 144 #define XLOG_COVER_OPS 5 145 146 #define XLOG_ICL_NEED_FLUSH (1 << 0) /* iclog needs REQ_PREFLUSH */ 147 #define XLOG_ICL_NEED_FUA (1 << 1) /* iclog needs REQ_FUA */ 148 149 /* Ticket reservation region accounting */ 150 #define XLOG_TIC_LEN_MAX 15 151 152 /* 153 * Reservation region 154 * As would be stored in xfs_log_iovec but without the i_addr which 155 * we don't care about. 156 */ 157 typedef struct xlog_res { 158 uint r_len; /* region length :4 */ 159 uint r_type; /* region's transaction type :4 */ 160 } xlog_res_t; 161 162 typedef struct xlog_ticket { 163 struct list_head t_queue; /* reserve/write queue */ 164 struct task_struct *t_task; /* task that owns this ticket */ 165 xlog_tid_t t_tid; /* transaction identifier : 4 */ 166 atomic_t t_ref; /* ticket reference count : 4 */ 167 int t_curr_res; /* current reservation in bytes : 4 */ 168 int t_unit_res; /* unit reservation in bytes : 4 */ 169 char t_ocnt; /* original count : 1 */ 170 char t_cnt; /* current count : 1 */ 171 char t_clientid; /* who does this belong to; : 1 */ 172 char t_flags; /* properties of reservation : 1 */ 173 174 /* reservation array fields */ 175 uint t_res_num; /* num in array : 4 */ 176 uint t_res_num_ophdrs; /* num op hdrs : 4 */ 177 uint t_res_arr_sum; /* array sum : 4 */ 178 uint t_res_o_flow; /* sum overflow : 4 */ 179 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */ 180 } xlog_ticket_t; 181 182 /* 183 * - A log record header is 512 bytes. There is plenty of room to grow the 184 * xlog_rec_header_t into the reserved space. 185 * - ic_data follows, so a write to disk can start at the beginning of 186 * the iclog. 187 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk. 188 * - ic_next is the pointer to the next iclog in the ring. 189 * - ic_log is a pointer back to the global log structure. 190 * - ic_size is the full size of the log buffer, minus the cycle headers. 191 * - ic_offset is the current number of bytes written to in this iclog. 192 * - ic_refcnt is bumped when someone is writing to the log. 193 * - ic_state is the state of the iclog. 194 * 195 * Because of cacheline contention on large machines, we need to separate 196 * various resources onto different cachelines. To start with, make the 197 * structure cacheline aligned. The following fields can be contended on 198 * by independent processes: 199 * 200 * - ic_callbacks 201 * - ic_refcnt 202 * - fields protected by the global l_icloglock 203 * 204 * so we need to ensure that these fields are located in separate cachelines. 205 * We'll put all the read-only and l_icloglock fields in the first cacheline, 206 * and move everything else out to subsequent cachelines. 207 */ 208 typedef struct xlog_in_core { 209 wait_queue_head_t ic_force_wait; 210 wait_queue_head_t ic_write_wait; 211 struct xlog_in_core *ic_next; 212 struct xlog_in_core *ic_prev; 213 struct xlog *ic_log; 214 u32 ic_size; 215 u32 ic_offset; 216 enum xlog_iclog_state ic_state; 217 unsigned int ic_flags; 218 char *ic_datap; /* pointer to iclog data */ 219 struct list_head ic_callbacks; 220 221 /* reference counts need their own cacheline */ 222 atomic_t ic_refcnt ____cacheline_aligned_in_smp; 223 xlog_in_core_2_t *ic_data; 224 #define ic_header ic_data->hic_header 225 #ifdef DEBUG 226 bool ic_fail_crc : 1; 227 #endif 228 struct semaphore ic_sema; 229 struct work_struct ic_end_io_work; 230 struct bio ic_bio; 231 struct bio_vec ic_bvec[]; 232 } xlog_in_core_t; 233 234 /* 235 * The CIL context is used to aggregate per-transaction details as well be 236 * passed to the iclog for checkpoint post-commit processing. After being 237 * passed to the iclog, another context needs to be allocated for tracking the 238 * next set of transactions to be aggregated into a checkpoint. 239 */ 240 struct xfs_cil; 241 242 struct xfs_cil_ctx { 243 struct xfs_cil *cil; 244 xfs_csn_t sequence; /* chkpt sequence # */ 245 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */ 246 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */ 247 struct xlog_ticket *ticket; /* chkpt ticket */ 248 int nvecs; /* number of regions */ 249 int space_used; /* aggregate size of regions */ 250 struct list_head busy_extents; /* busy extents in chkpt */ 251 struct xfs_log_vec *lv_chain; /* logvecs being pushed */ 252 struct list_head iclog_entry; 253 struct list_head committing; /* ctx committing list */ 254 struct work_struct discard_endio_work; 255 }; 256 257 /* 258 * Committed Item List structure 259 * 260 * This structure is used to track log items that have been committed but not 261 * yet written into the log. It is used only when the delayed logging mount 262 * option is enabled. 263 * 264 * This structure tracks the list of committing checkpoint contexts so 265 * we can avoid the problem of having to hold out new transactions during a 266 * flush until we have a the commit record LSN of the checkpoint. We can 267 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a 268 * sequence match and extract the commit LSN directly from there. If the 269 * checkpoint is still in the process of committing, we can block waiting for 270 * the commit LSN to be determined as well. This should make synchronous 271 * operations almost as efficient as the old logging methods. 272 */ 273 struct xfs_cil { 274 struct xlog *xc_log; 275 struct list_head xc_cil; 276 spinlock_t xc_cil_lock; 277 278 struct rw_semaphore xc_ctx_lock ____cacheline_aligned_in_smp; 279 struct xfs_cil_ctx *xc_ctx; 280 281 spinlock_t xc_push_lock ____cacheline_aligned_in_smp; 282 xfs_csn_t xc_push_seq; 283 struct list_head xc_committing; 284 wait_queue_head_t xc_commit_wait; 285 xfs_csn_t xc_current_sequence; 286 struct work_struct xc_push_work; 287 wait_queue_head_t xc_push_wait; /* background push throttle */ 288 } ____cacheline_aligned_in_smp; 289 290 /* 291 * The amount of log space we allow the CIL to aggregate is difficult to size. 292 * Whatever we choose, we have to make sure we can get a reservation for the 293 * log space effectively, that it is large enough to capture sufficient 294 * relogging to reduce log buffer IO significantly, but it is not too large for 295 * the log or induces too much latency when writing out through the iclogs. We 296 * track both space consumed and the number of vectors in the checkpoint 297 * context, so we need to decide which to use for limiting. 298 * 299 * Every log buffer we write out during a push needs a header reserved, which 300 * is at least one sector and more for v2 logs. Hence we need a reservation of 301 * at least 512 bytes per 32k of log space just for the LR headers. That means 302 * 16KB of reservation per megabyte of delayed logging space we will consume, 303 * plus various headers. The number of headers will vary based on the num of 304 * io vectors, so limiting on a specific number of vectors is going to result 305 * in transactions of varying size. IOWs, it is more consistent to track and 306 * limit space consumed in the log rather than by the number of objects being 307 * logged in order to prevent checkpoint ticket overruns. 308 * 309 * Further, use of static reservations through the log grant mechanism is 310 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write 311 * grant) and a significant deadlock potential because regranting write space 312 * can block on log pushes. Hence if we have to regrant log space during a log 313 * push, we can deadlock. 314 * 315 * However, we can avoid this by use of a dynamic "reservation stealing" 316 * technique during transaction commit whereby unused reservation space in the 317 * transaction ticket is transferred to the CIL ctx commit ticket to cover the 318 * space needed by the checkpoint transaction. This means that we never need to 319 * specifically reserve space for the CIL checkpoint transaction, nor do we 320 * need to regrant space once the checkpoint completes. This also means the 321 * checkpoint transaction ticket is specific to the checkpoint context, rather 322 * than the CIL itself. 323 * 324 * With dynamic reservations, we can effectively make up arbitrary limits for 325 * the checkpoint size so long as they don't violate any other size rules. 326 * Recovery imposes a rule that no transaction exceed half the log, so we are 327 * limited by that. Furthermore, the log transaction reservation subsystem 328 * tries to keep 25% of the log free, so we need to keep below that limit or we 329 * risk running out of free log space to start any new transactions. 330 * 331 * In order to keep background CIL push efficient, we only need to ensure the 332 * CIL is large enough to maintain sufficient in-memory relogging to avoid 333 * repeated physical writes of frequently modified metadata. If we allow the CIL 334 * to grow to a substantial fraction of the log, then we may be pinning hundreds 335 * of megabytes of metadata in memory until the CIL flushes. This can cause 336 * issues when we are running low on memory - pinned memory cannot be reclaimed, 337 * and the CIL consumes a lot of memory. Hence we need to set an upper physical 338 * size limit for the CIL that limits the maximum amount of memory pinned by the 339 * CIL but does not limit performance by reducing relogging efficiency 340 * significantly. 341 * 342 * As such, the CIL push threshold ends up being the smaller of two thresholds: 343 * - a threshold large enough that it allows CIL to be pushed and progress to be 344 * made without excessive blocking of incoming transaction commits. This is 345 * defined to be 12.5% of the log space - half the 25% push threshold of the 346 * AIL. 347 * - small enough that it doesn't pin excessive amounts of memory but maintains 348 * close to peak relogging efficiency. This is defined to be 16x the iclog 349 * buffer window (32MB) as measurements have shown this to be roughly the 350 * point of diminishing performance increases under highly concurrent 351 * modification workloads. 352 * 353 * To prevent the CIL from overflowing upper commit size bounds, we introduce a 354 * new threshold at which we block committing transactions until the background 355 * CIL commit commences and switches to a new context. While this is not a hard 356 * limit, it forces the process committing a transaction to the CIL to block and 357 * yeild the CPU, giving the CIL push work a chance to be scheduled and start 358 * work. This prevents a process running lots of transactions from overfilling 359 * the CIL because it is not yielding the CPU. We set the blocking limit at 360 * twice the background push space threshold so we keep in line with the AIL 361 * push thresholds. 362 * 363 * Note: this is not a -hard- limit as blocking is applied after the transaction 364 * is inserted into the CIL and the push has been triggered. It is largely a 365 * throttling mechanism that allows the CIL push to be scheduled and run. A hard 366 * limit will be difficult to implement without introducing global serialisation 367 * in the CIL commit fast path, and it's not at all clear that we actually need 368 * such hard limits given the ~7 years we've run without a hard limit before 369 * finding the first situation where a checkpoint size overflow actually 370 * occurred. Hence the simple throttle, and an ASSERT check to tell us that 371 * we've overrun the max size. 372 */ 373 #define XLOG_CIL_SPACE_LIMIT(log) \ 374 min_t(int, (log)->l_logsize >> 3, BBTOB(XLOG_TOTAL_REC_SHIFT(log)) << 4) 375 376 #define XLOG_CIL_BLOCKING_SPACE_LIMIT(log) \ 377 (XLOG_CIL_SPACE_LIMIT(log) * 2) 378 379 /* 380 * ticket grant locks, queues and accounting have their own cachlines 381 * as these are quite hot and can be operated on concurrently. 382 */ 383 struct xlog_grant_head { 384 spinlock_t lock ____cacheline_aligned_in_smp; 385 struct list_head waiters; 386 atomic64_t grant; 387 }; 388 389 /* 390 * The reservation head lsn is not made up of a cycle number and block number. 391 * Instead, it uses a cycle number and byte number. Logs don't expect to 392 * overflow 31 bits worth of byte offset, so using a byte number will mean 393 * that round off problems won't occur when releasing partial reservations. 394 */ 395 struct xlog { 396 /* The following fields don't need locking */ 397 struct xfs_mount *l_mp; /* mount point */ 398 struct xfs_ail *l_ailp; /* AIL log is working with */ 399 struct xfs_cil *l_cilp; /* CIL log is working with */ 400 struct xfs_buftarg *l_targ; /* buftarg of log */ 401 struct workqueue_struct *l_ioend_workqueue; /* for I/O completions */ 402 struct delayed_work l_work; /* background flush work */ 403 uint l_flags; 404 uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */ 405 struct list_head *l_buf_cancel_table; 406 int l_iclog_hsize; /* size of iclog header */ 407 int l_iclog_heads; /* # of iclog header sectors */ 408 uint l_sectBBsize; /* sector size in BBs (2^n) */ 409 int l_iclog_size; /* size of log in bytes */ 410 int l_iclog_bufs; /* number of iclog buffers */ 411 xfs_daddr_t l_logBBstart; /* start block of log */ 412 int l_logsize; /* size of log in bytes */ 413 int l_logBBsize; /* size of log in BB chunks */ 414 415 /* The following block of fields are changed while holding icloglock */ 416 wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp; 417 /* waiting for iclog flush */ 418 int l_covered_state;/* state of "covering disk 419 * log entries" */ 420 xlog_in_core_t *l_iclog; /* head log queue */ 421 spinlock_t l_icloglock; /* grab to change iclog state */ 422 int l_curr_cycle; /* Cycle number of log writes */ 423 int l_prev_cycle; /* Cycle number before last 424 * block increment */ 425 int l_curr_block; /* current logical log block */ 426 int l_prev_block; /* previous logical log block */ 427 428 /* 429 * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and 430 * read without needing to hold specific locks. To avoid operations 431 * contending with other hot objects, place each of them on a separate 432 * cacheline. 433 */ 434 /* lsn of last LR on disk */ 435 atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp; 436 /* lsn of 1st LR with unflushed * buffers */ 437 atomic64_t l_tail_lsn ____cacheline_aligned_in_smp; 438 439 struct xlog_grant_head l_reserve_head; 440 struct xlog_grant_head l_write_head; 441 442 struct xfs_kobj l_kobj; 443 444 /* The following field are used for debugging; need to hold icloglock */ 445 #ifdef DEBUG 446 void *l_iclog_bak[XLOG_MAX_ICLOGS]; 447 #endif 448 /* log recovery lsn tracking (for buffer submission */ 449 xfs_lsn_t l_recovery_lsn; 450 451 uint32_t l_iclog_roundoff;/* padding roundoff */ 452 }; 453 454 #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \ 455 ((log)->l_buf_cancel_table + ((uint64_t)blkno % XLOG_BC_TABLE_SIZE)) 456 457 #define XLOG_FORCED_SHUTDOWN(log) \ 458 (unlikely((log)->l_flags & XLOG_IO_ERROR)) 459 460 /* common routines */ 461 extern int 462 xlog_recover( 463 struct xlog *log); 464 extern int 465 xlog_recover_finish( 466 struct xlog *log); 467 extern void 468 xlog_recover_cancel(struct xlog *); 469 470 extern __le32 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead, 471 char *dp, int size); 472 473 extern kmem_zone_t *xfs_log_ticket_zone; 474 struct xlog_ticket * 475 xlog_ticket_alloc( 476 struct xlog *log, 477 int unit_bytes, 478 int count, 479 char client, 480 bool permanent); 481 482 static inline void 483 xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes) 484 { 485 *ptr += bytes; 486 *len -= bytes; 487 *off += bytes; 488 } 489 490 void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket); 491 void xlog_print_trans(struct xfs_trans *); 492 int xlog_write(struct xlog *log, struct xfs_log_vec *log_vector, 493 struct xlog_ticket *tic, xfs_lsn_t *start_lsn, 494 struct xlog_in_core **commit_iclog, uint optype); 495 int xlog_commit_record(struct xlog *log, struct xlog_ticket *ticket, 496 struct xlog_in_core **iclog, xfs_lsn_t *lsn); 497 void xfs_log_ticket_ungrant(struct xlog *log, struct xlog_ticket *ticket); 498 void xfs_log_ticket_regrant(struct xlog *log, struct xlog_ticket *ticket); 499 500 int xlog_state_release_iclog(struct xlog *log, struct xlog_in_core *iclog); 501 502 /* 503 * When we crack an atomic LSN, we sample it first so that the value will not 504 * change while we are cracking it into the component values. This means we 505 * will always get consistent component values to work from. This should always 506 * be used to sample and crack LSNs that are stored and updated in atomic 507 * variables. 508 */ 509 static inline void 510 xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block) 511 { 512 xfs_lsn_t val = atomic64_read(lsn); 513 514 *cycle = CYCLE_LSN(val); 515 *block = BLOCK_LSN(val); 516 } 517 518 /* 519 * Calculate and assign a value to an atomic LSN variable from component pieces. 520 */ 521 static inline void 522 xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block) 523 { 524 atomic64_set(lsn, xlog_assign_lsn(cycle, block)); 525 } 526 527 /* 528 * When we crack the grant head, we sample it first so that the value will not 529 * change while we are cracking it into the component values. This means we 530 * will always get consistent component values to work from. 531 */ 532 static inline void 533 xlog_crack_grant_head_val(int64_t val, int *cycle, int *space) 534 { 535 *cycle = val >> 32; 536 *space = val & 0xffffffff; 537 } 538 539 static inline void 540 xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space) 541 { 542 xlog_crack_grant_head_val(atomic64_read(head), cycle, space); 543 } 544 545 static inline int64_t 546 xlog_assign_grant_head_val(int cycle, int space) 547 { 548 return ((int64_t)cycle << 32) | space; 549 } 550 551 static inline void 552 xlog_assign_grant_head(atomic64_t *head, int cycle, int space) 553 { 554 atomic64_set(head, xlog_assign_grant_head_val(cycle, space)); 555 } 556 557 /* 558 * Committed Item List interfaces 559 */ 560 int xlog_cil_init(struct xlog *log); 561 void xlog_cil_init_post_recovery(struct xlog *log); 562 void xlog_cil_destroy(struct xlog *log); 563 bool xlog_cil_empty(struct xlog *log); 564 void xlog_cil_commit(struct xlog *log, struct xfs_trans *tp, 565 xfs_csn_t *commit_seq, bool regrant); 566 567 /* 568 * CIL force routines 569 */ 570 xfs_lsn_t xlog_cil_force_seq(struct xlog *log, xfs_csn_t sequence); 571 572 static inline void 573 xlog_cil_force(struct xlog *log) 574 { 575 xlog_cil_force_seq(log, log->l_cilp->xc_current_sequence); 576 } 577 578 /* 579 * Wrapper function for waiting on a wait queue serialised against wakeups 580 * by a spinlock. This matches the semantics of all the wait queues used in the 581 * log code. 582 */ 583 static inline void 584 xlog_wait( 585 struct wait_queue_head *wq, 586 struct spinlock *lock) 587 __releases(lock) 588 { 589 DECLARE_WAITQUEUE(wait, current); 590 591 add_wait_queue_exclusive(wq, &wait); 592 __set_current_state(TASK_UNINTERRUPTIBLE); 593 spin_unlock(lock); 594 schedule(); 595 remove_wait_queue(wq, &wait); 596 } 597 598 int xlog_wait_on_iclog(struct xlog_in_core *iclog); 599 600 /* 601 * The LSN is valid so long as it is behind the current LSN. If it isn't, this 602 * means that the next log record that includes this metadata could have a 603 * smaller LSN. In turn, this means that the modification in the log would not 604 * replay. 605 */ 606 static inline bool 607 xlog_valid_lsn( 608 struct xlog *log, 609 xfs_lsn_t lsn) 610 { 611 int cur_cycle; 612 int cur_block; 613 bool valid = true; 614 615 /* 616 * First, sample the current lsn without locking to avoid added 617 * contention from metadata I/O. The current cycle and block are updated 618 * (in xlog_state_switch_iclogs()) and read here in a particular order 619 * to avoid false negatives (e.g., thinking the metadata LSN is valid 620 * when it is not). 621 * 622 * The current block is always rewound before the cycle is bumped in 623 * xlog_state_switch_iclogs() to ensure the current LSN is never seen in 624 * a transiently forward state. Instead, we can see the LSN in a 625 * transiently behind state if we happen to race with a cycle wrap. 626 */ 627 cur_cycle = READ_ONCE(log->l_curr_cycle); 628 smp_rmb(); 629 cur_block = READ_ONCE(log->l_curr_block); 630 631 if ((CYCLE_LSN(lsn) > cur_cycle) || 632 (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) { 633 /* 634 * If the metadata LSN appears invalid, it's possible the check 635 * above raced with a wrap to the next log cycle. Grab the lock 636 * to check for sure. 637 */ 638 spin_lock(&log->l_icloglock); 639 cur_cycle = log->l_curr_cycle; 640 cur_block = log->l_curr_block; 641 spin_unlock(&log->l_icloglock); 642 643 if ((CYCLE_LSN(lsn) > cur_cycle) || 644 (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) 645 valid = false; 646 } 647 648 return valid; 649 } 650 651 #endif /* __XFS_LOG_PRIV_H__ */ 652