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 * get client id from packed copy. 16 * 17 * this hack is here because the xlog_pack code copies four bytes 18 * of xlog_op_header containing the fields oh_clientid, oh_flags 19 * and oh_res2 into the packed copy. 20 * 21 * later on this four byte chunk is treated as an int and the 22 * client id is pulled out. 23 * 24 * this has endian issues, of course. 25 */ 26 static inline uint xlog_get_client_id(__be32 i) 27 { 28 return be32_to_cpu(i) >> 24; 29 } 30 31 /* 32 * In core log state 33 */ 34 enum xlog_iclog_state { 35 XLOG_STATE_ACTIVE, /* Current IC log being written to */ 36 XLOG_STATE_WANT_SYNC, /* Want to sync this iclog; no more writes */ 37 XLOG_STATE_SYNCING, /* This IC log is syncing */ 38 XLOG_STATE_DONE_SYNC, /* Done syncing to disk */ 39 XLOG_STATE_CALLBACK, /* Callback functions now */ 40 XLOG_STATE_DIRTY, /* Dirty IC log, not ready for ACTIVE status */ 41 }; 42 43 #define XLOG_STATE_STRINGS \ 44 { XLOG_STATE_ACTIVE, "XLOG_STATE_ACTIVE" }, \ 45 { XLOG_STATE_WANT_SYNC, "XLOG_STATE_WANT_SYNC" }, \ 46 { XLOG_STATE_SYNCING, "XLOG_STATE_SYNCING" }, \ 47 { XLOG_STATE_DONE_SYNC, "XLOG_STATE_DONE_SYNC" }, \ 48 { XLOG_STATE_CALLBACK, "XLOG_STATE_CALLBACK" }, \ 49 { XLOG_STATE_DIRTY, "XLOG_STATE_DIRTY" } 50 51 /* 52 * In core log flags 53 */ 54 #define XLOG_ICL_NEED_FLUSH (1 << 0) /* iclog needs REQ_PREFLUSH */ 55 #define XLOG_ICL_NEED_FUA (1 << 1) /* iclog needs REQ_FUA */ 56 57 #define XLOG_ICL_STRINGS \ 58 { XLOG_ICL_NEED_FLUSH, "XLOG_ICL_NEED_FLUSH" }, \ 59 { XLOG_ICL_NEED_FUA, "XLOG_ICL_NEED_FUA" } 60 61 62 /* 63 * Log ticket flags 64 */ 65 #define XLOG_TIC_PERM_RESERV 0x1 /* permanent reservation */ 66 67 #define XLOG_TIC_FLAGS \ 68 { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" } 69 70 /* 71 * Below are states for covering allocation transactions. 72 * By covering, we mean changing the h_tail_lsn in the last on-disk 73 * log write such that no allocation transactions will be re-done during 74 * recovery after a system crash. Recovery starts at the last on-disk 75 * log write. 76 * 77 * These states are used to insert dummy log entries to cover 78 * space allocation transactions which can undo non-transactional changes 79 * after a crash. Writes to a file with space 80 * already allocated do not result in any transactions. Allocations 81 * might include space beyond the EOF. So if we just push the EOF a 82 * little, the last transaction for the file could contain the wrong 83 * size. If there is no file system activity, after an allocation 84 * transaction, and the system crashes, the allocation transaction 85 * will get replayed and the file will be truncated. This could 86 * be hours/days/... after the allocation occurred. 87 * 88 * The fix for this is to do two dummy transactions when the 89 * system is idle. We need two dummy transaction because the h_tail_lsn 90 * in the log record header needs to point beyond the last possible 91 * non-dummy transaction. The first dummy changes the h_tail_lsn to 92 * the first transaction before the dummy. The second dummy causes 93 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn. 94 * 95 * These dummy transactions get committed when everything 96 * is idle (after there has been some activity). 97 * 98 * There are 5 states used to control this. 99 * 100 * IDLE -- no logging has been done on the file system or 101 * we are done covering previous transactions. 102 * NEED -- logging has occurred and we need a dummy transaction 103 * when the log becomes idle. 104 * DONE -- we were in the NEED state and have committed a dummy 105 * transaction. 106 * NEED2 -- we detected that a dummy transaction has gone to the 107 * on disk log with no other transactions. 108 * DONE2 -- we committed a dummy transaction when in the NEED2 state. 109 * 110 * There are two places where we switch states: 111 * 112 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2. 113 * We commit the dummy transaction and switch to DONE or DONE2, 114 * respectively. In all other states, we don't do anything. 115 * 116 * 2.) When we finish writing the on-disk log (xlog_state_clean_log). 117 * 118 * No matter what state we are in, if this isn't the dummy 119 * transaction going out, the next state is NEED. 120 * So, if we aren't in the DONE or DONE2 states, the next state 121 * is NEED. We can't be finishing a write of the dummy record 122 * unless it was committed and the state switched to DONE or DONE2. 123 * 124 * If we are in the DONE state and this was a write of the 125 * dummy transaction, we move to NEED2. 126 * 127 * If we are in the DONE2 state and this was a write of the 128 * dummy transaction, we move to IDLE. 129 * 130 * 131 * Writing only one dummy transaction can get appended to 132 * one file space allocation. When this happens, the log recovery 133 * code replays the space allocation and a file could be truncated. 134 * This is why we have the NEED2 and DONE2 states before going idle. 135 */ 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 4 142 143 #define XLOG_COVER_OPS 5 144 145 /* Ticket reservation region accounting */ 146 #define XLOG_TIC_LEN_MAX 15 147 148 /* 149 * Reservation region 150 * As would be stored in xfs_log_iovec but without the i_addr which 151 * we don't care about. 152 */ 153 typedef struct xlog_res { 154 uint r_len; /* region length :4 */ 155 uint r_type; /* region's transaction type :4 */ 156 } xlog_res_t; 157 158 typedef struct xlog_ticket { 159 struct list_head t_queue; /* reserve/write queue */ 160 struct task_struct *t_task; /* task that owns this ticket */ 161 xlog_tid_t t_tid; /* transaction identifier : 4 */ 162 atomic_t t_ref; /* ticket reference count : 4 */ 163 int t_curr_res; /* current reservation in bytes : 4 */ 164 int t_unit_res; /* unit reservation in bytes : 4 */ 165 char t_ocnt; /* original count : 1 */ 166 char t_cnt; /* current count : 1 */ 167 char t_clientid; /* who does this belong to; : 1 */ 168 char t_flags; /* properties of reservation : 1 */ 169 170 /* reservation array fields */ 171 uint t_res_num; /* num in array : 4 */ 172 uint t_res_num_ophdrs; /* num op hdrs : 4 */ 173 uint t_res_arr_sum; /* array sum : 4 */ 174 uint t_res_o_flow; /* sum overflow : 4 */ 175 xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */ 176 } xlog_ticket_t; 177 178 /* 179 * - A log record header is 512 bytes. There is plenty of room to grow the 180 * xlog_rec_header_t into the reserved space. 181 * - ic_data follows, so a write to disk can start at the beginning of 182 * the iclog. 183 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk. 184 * - ic_next is the pointer to the next iclog in the ring. 185 * - ic_log is a pointer back to the global log structure. 186 * - ic_size is the full size of the log buffer, minus the cycle headers. 187 * - ic_offset is the current number of bytes written to in this iclog. 188 * - ic_refcnt is bumped when someone is writing to the log. 189 * - ic_state is the state of the iclog. 190 * 191 * Because of cacheline contention on large machines, we need to separate 192 * various resources onto different cachelines. To start with, make the 193 * structure cacheline aligned. The following fields can be contended on 194 * by independent processes: 195 * 196 * - ic_callbacks 197 * - ic_refcnt 198 * - fields protected by the global l_icloglock 199 * 200 * so we need to ensure that these fields are located in separate cachelines. 201 * We'll put all the read-only and l_icloglock fields in the first cacheline, 202 * and move everything else out to subsequent cachelines. 203 */ 204 typedef struct xlog_in_core { 205 wait_queue_head_t ic_force_wait; 206 wait_queue_head_t ic_write_wait; 207 struct xlog_in_core *ic_next; 208 struct xlog_in_core *ic_prev; 209 struct xlog *ic_log; 210 u32 ic_size; 211 u32 ic_offset; 212 enum xlog_iclog_state ic_state; 213 unsigned int ic_flags; 214 char *ic_datap; /* pointer to iclog data */ 215 struct list_head ic_callbacks; 216 217 /* reference counts need their own cacheline */ 218 atomic_t ic_refcnt ____cacheline_aligned_in_smp; 219 xlog_in_core_2_t *ic_data; 220 #define ic_header ic_data->hic_header 221 #ifdef DEBUG 222 bool ic_fail_crc : 1; 223 #endif 224 struct semaphore ic_sema; 225 struct work_struct ic_end_io_work; 226 struct bio ic_bio; 227 struct bio_vec ic_bvec[]; 228 } xlog_in_core_t; 229 230 /* 231 * The CIL context is used to aggregate per-transaction details as well be 232 * passed to the iclog for checkpoint post-commit processing. After being 233 * passed to the iclog, another context needs to be allocated for tracking the 234 * next set of transactions to be aggregated into a checkpoint. 235 */ 236 struct xfs_cil; 237 238 struct xfs_cil_ctx { 239 struct xfs_cil *cil; 240 xfs_csn_t sequence; /* chkpt sequence # */ 241 xfs_lsn_t start_lsn; /* first LSN of chkpt commit */ 242 xfs_lsn_t commit_lsn; /* chkpt commit record lsn */ 243 struct xlog_in_core *commit_iclog; 244 struct xlog_ticket *ticket; /* chkpt ticket */ 245 int nvecs; /* number of regions */ 246 int space_used; /* aggregate size of regions */ 247 struct list_head busy_extents; /* busy extents in chkpt */ 248 struct xfs_log_vec *lv_chain; /* logvecs being pushed */ 249 struct list_head iclog_entry; 250 struct list_head committing; /* ctx committing list */ 251 struct work_struct discard_endio_work; 252 struct work_struct push_work; 253 }; 254 255 /* 256 * Committed Item List structure 257 * 258 * This structure is used to track log items that have been committed but not 259 * yet written into the log. It is used only when the delayed logging mount 260 * option is enabled. 261 * 262 * This structure tracks the list of committing checkpoint contexts so 263 * we can avoid the problem of having to hold out new transactions during a 264 * flush until we have a the commit record LSN of the checkpoint. We can 265 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a 266 * sequence match and extract the commit LSN directly from there. If the 267 * checkpoint is still in the process of committing, we can block waiting for 268 * the commit LSN to be determined as well. This should make synchronous 269 * operations almost as efficient as the old logging methods. 270 */ 271 struct xfs_cil { 272 struct xlog *xc_log; 273 struct list_head xc_cil; 274 spinlock_t xc_cil_lock; 275 struct workqueue_struct *xc_push_wq; 276 277 struct rw_semaphore xc_ctx_lock ____cacheline_aligned_in_smp; 278 struct xfs_cil_ctx *xc_ctx; 279 280 spinlock_t xc_push_lock ____cacheline_aligned_in_smp; 281 xfs_csn_t xc_push_seq; 282 bool xc_push_commit_stable; 283 struct list_head xc_committing; 284 wait_queue_head_t xc_commit_wait; 285 wait_queue_head_t xc_start_wait; 286 xfs_csn_t xc_current_sequence; 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 long l_opstate; /* operational state */ 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 /* Users of log incompat features should take a read lock. */ 454 struct rw_semaphore l_incompat_users; 455 }; 456 457 #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \ 458 ((log)->l_buf_cancel_table + ((uint64_t)blkno % XLOG_BC_TABLE_SIZE)) 459 460 /* 461 * Bits for operational state 462 */ 463 #define XLOG_ACTIVE_RECOVERY 0 /* in the middle of recovery */ 464 #define XLOG_RECOVERY_NEEDED 1 /* log was recovered */ 465 #define XLOG_IO_ERROR 2 /* log hit an I/O error, and being 466 shutdown */ 467 #define XLOG_TAIL_WARN 3 /* log tail verify warning issued */ 468 469 static inline bool 470 xlog_recovery_needed(struct xlog *log) 471 { 472 return test_bit(XLOG_RECOVERY_NEEDED, &log->l_opstate); 473 } 474 475 static inline bool 476 xlog_in_recovery(struct xlog *log) 477 { 478 return test_bit(XLOG_ACTIVE_RECOVERY, &log->l_opstate); 479 } 480 481 static inline bool 482 xlog_is_shutdown(struct xlog *log) 483 { 484 return test_bit(XLOG_IO_ERROR, &log->l_opstate); 485 } 486 487 /* 488 * Wait until the xlog_force_shutdown() has marked the log as shut down 489 * so xlog_is_shutdown() will always return true. 490 */ 491 static inline void 492 xlog_shutdown_wait( 493 struct xlog *log) 494 { 495 wait_var_event(&log->l_opstate, xlog_is_shutdown(log)); 496 } 497 498 /* common routines */ 499 extern int 500 xlog_recover( 501 struct xlog *log); 502 extern int 503 xlog_recover_finish( 504 struct xlog *log); 505 extern void 506 xlog_recover_cancel(struct xlog *); 507 508 extern __le32 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead, 509 char *dp, int size); 510 511 extern struct kmem_cache *xfs_log_ticket_cache; 512 struct xlog_ticket * 513 xlog_ticket_alloc( 514 struct xlog *log, 515 int unit_bytes, 516 int count, 517 char client, 518 bool permanent); 519 520 static inline void 521 xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes) 522 { 523 *ptr += bytes; 524 *len -= bytes; 525 *off += bytes; 526 } 527 528 void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket); 529 void xlog_print_trans(struct xfs_trans *); 530 int xlog_write(struct xlog *log, struct xfs_cil_ctx *ctx, 531 struct xfs_log_vec *log_vector, struct xlog_ticket *tic, 532 uint optype); 533 void xfs_log_ticket_ungrant(struct xlog *log, struct xlog_ticket *ticket); 534 void xfs_log_ticket_regrant(struct xlog *log, struct xlog_ticket *ticket); 535 536 void xlog_state_switch_iclogs(struct xlog *log, struct xlog_in_core *iclog, 537 int eventual_size); 538 int xlog_state_release_iclog(struct xlog *log, struct xlog_in_core *iclog); 539 540 /* 541 * When we crack an atomic LSN, we sample it first so that the value will not 542 * change while we are cracking it into the component values. This means we 543 * will always get consistent component values to work from. This should always 544 * be used to sample and crack LSNs that are stored and updated in atomic 545 * variables. 546 */ 547 static inline void 548 xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block) 549 { 550 xfs_lsn_t val = atomic64_read(lsn); 551 552 *cycle = CYCLE_LSN(val); 553 *block = BLOCK_LSN(val); 554 } 555 556 /* 557 * Calculate and assign a value to an atomic LSN variable from component pieces. 558 */ 559 static inline void 560 xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block) 561 { 562 atomic64_set(lsn, xlog_assign_lsn(cycle, block)); 563 } 564 565 /* 566 * When we crack the grant head, we sample it first so that the value will not 567 * change while we are cracking it into the component values. This means we 568 * will always get consistent component values to work from. 569 */ 570 static inline void 571 xlog_crack_grant_head_val(int64_t val, int *cycle, int *space) 572 { 573 *cycle = val >> 32; 574 *space = val & 0xffffffff; 575 } 576 577 static inline void 578 xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space) 579 { 580 xlog_crack_grant_head_val(atomic64_read(head), cycle, space); 581 } 582 583 static inline int64_t 584 xlog_assign_grant_head_val(int cycle, int space) 585 { 586 return ((int64_t)cycle << 32) | space; 587 } 588 589 static inline void 590 xlog_assign_grant_head(atomic64_t *head, int cycle, int space) 591 { 592 atomic64_set(head, xlog_assign_grant_head_val(cycle, space)); 593 } 594 595 /* 596 * Committed Item List interfaces 597 */ 598 int xlog_cil_init(struct xlog *log); 599 void xlog_cil_init_post_recovery(struct xlog *log); 600 void xlog_cil_destroy(struct xlog *log); 601 bool xlog_cil_empty(struct xlog *log); 602 void xlog_cil_commit(struct xlog *log, struct xfs_trans *tp, 603 xfs_csn_t *commit_seq, bool regrant); 604 void xlog_cil_set_ctx_write_state(struct xfs_cil_ctx *ctx, 605 struct xlog_in_core *iclog); 606 607 608 /* 609 * CIL force routines 610 */ 611 void xlog_cil_flush(struct xlog *log); 612 xfs_lsn_t xlog_cil_force_seq(struct xlog *log, xfs_csn_t sequence); 613 614 static inline void 615 xlog_cil_force(struct xlog *log) 616 { 617 xlog_cil_force_seq(log, log->l_cilp->xc_current_sequence); 618 } 619 620 /* 621 * Wrapper function for waiting on a wait queue serialised against wakeups 622 * by a spinlock. This matches the semantics of all the wait queues used in the 623 * log code. 624 */ 625 static inline void 626 xlog_wait( 627 struct wait_queue_head *wq, 628 struct spinlock *lock) 629 __releases(lock) 630 { 631 DECLARE_WAITQUEUE(wait, current); 632 633 add_wait_queue_exclusive(wq, &wait); 634 __set_current_state(TASK_UNINTERRUPTIBLE); 635 spin_unlock(lock); 636 schedule(); 637 remove_wait_queue(wq, &wait); 638 } 639 640 int xlog_wait_on_iclog(struct xlog_in_core *iclog); 641 642 /* 643 * The LSN is valid so long as it is behind the current LSN. If it isn't, this 644 * means that the next log record that includes this metadata could have a 645 * smaller LSN. In turn, this means that the modification in the log would not 646 * replay. 647 */ 648 static inline bool 649 xlog_valid_lsn( 650 struct xlog *log, 651 xfs_lsn_t lsn) 652 { 653 int cur_cycle; 654 int cur_block; 655 bool valid = true; 656 657 /* 658 * First, sample the current lsn without locking to avoid added 659 * contention from metadata I/O. The current cycle and block are updated 660 * (in xlog_state_switch_iclogs()) and read here in a particular order 661 * to avoid false negatives (e.g., thinking the metadata LSN is valid 662 * when it is not). 663 * 664 * The current block is always rewound before the cycle is bumped in 665 * xlog_state_switch_iclogs() to ensure the current LSN is never seen in 666 * a transiently forward state. Instead, we can see the LSN in a 667 * transiently behind state if we happen to race with a cycle wrap. 668 */ 669 cur_cycle = READ_ONCE(log->l_curr_cycle); 670 smp_rmb(); 671 cur_block = READ_ONCE(log->l_curr_block); 672 673 if ((CYCLE_LSN(lsn) > cur_cycle) || 674 (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) { 675 /* 676 * If the metadata LSN appears invalid, it's possible the check 677 * above raced with a wrap to the next log cycle. Grab the lock 678 * to check for sure. 679 */ 680 spin_lock(&log->l_icloglock); 681 cur_cycle = log->l_curr_cycle; 682 cur_block = log->l_curr_block; 683 spin_unlock(&log->l_icloglock); 684 685 if ((CYCLE_LSN(lsn) > cur_cycle) || 686 (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) 687 valid = false; 688 } 689 690 return valid; 691 } 692 693 #endif /* __XFS_LOG_PRIV_H__ */ 694