10b61f8a4SDave Chinner // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 37b718769SNathan Scott * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc. 47b718769SNathan Scott * All Rights Reserved. 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds #ifndef __XFS_LOG_PRIV_H__ 71da177e4SLinus Torvalds #define __XFS_LOG_PRIV_H__ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds struct xfs_buf; 10ad223e60SMark Tinguely struct xlog; 11a844f451SNathan Scott struct xlog_ticket; 121da177e4SLinus Torvalds struct xfs_mount; 131da177e4SLinus Torvalds 141da177e4SLinus Torvalds /* 15fc06c6d0SDave Chinner * Flags for log structure 161da177e4SLinus Torvalds */ 17fc06c6d0SDave Chinner #define XLOG_ACTIVE_RECOVERY 0x2 /* in the middle of recovery */ 18fc06c6d0SDave Chinner #define XLOG_RECOVERY_NEEDED 0x4 /* log was recovered */ 19fc06c6d0SDave Chinner #define XLOG_IO_ERROR 0x8 /* log hit an I/O error, and being 20fc06c6d0SDave Chinner shutdown */ 21fc06c6d0SDave Chinner #define XLOG_TAIL_WARN 0x10 /* log tail verify warning issued */ 221da177e4SLinus Torvalds 231da177e4SLinus Torvalds /* 241da177e4SLinus Torvalds * get client id from packed copy. 251da177e4SLinus Torvalds * 261da177e4SLinus Torvalds * this hack is here because the xlog_pack code copies four bytes 271da177e4SLinus Torvalds * of xlog_op_header containing the fields oh_clientid, oh_flags 281da177e4SLinus Torvalds * and oh_res2 into the packed copy. 291da177e4SLinus Torvalds * 301da177e4SLinus Torvalds * later on this four byte chunk is treated as an int and the 311da177e4SLinus Torvalds * client id is pulled out. 321da177e4SLinus Torvalds * 331da177e4SLinus Torvalds * this has endian issues, of course. 341da177e4SLinus Torvalds */ 35b53e675dSChristoph Hellwig static inline uint xlog_get_client_id(__be32 i) 3603bea6feSChristoph Hellwig { 37b53e675dSChristoph Hellwig return be32_to_cpu(i) >> 24; 3803bea6feSChristoph Hellwig } 391da177e4SLinus Torvalds 401da177e4SLinus Torvalds /* 411da177e4SLinus Torvalds * In core log state 421da177e4SLinus Torvalds */ 431da177e4SLinus Torvalds #define XLOG_STATE_ACTIVE 0x0001 /* Current IC log being written to */ 441da177e4SLinus Torvalds #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */ 451da177e4SLinus Torvalds #define XLOG_STATE_SYNCING 0x0004 /* This IC log is syncing */ 461da177e4SLinus Torvalds #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */ 471da177e4SLinus Torvalds #define XLOG_STATE_DO_CALLBACK \ 481da177e4SLinus Torvalds 0x0010 /* Process callback functions */ 491da177e4SLinus Torvalds #define XLOG_STATE_CALLBACK 0x0020 /* Callback functions now */ 501da177e4SLinus Torvalds #define XLOG_STATE_DIRTY 0x0040 /* Dirty IC log, not ready for ACTIVE status*/ 511da177e4SLinus Torvalds #define XLOG_STATE_IOERROR 0x0080 /* IO error happened in sync'ing log */ 521da177e4SLinus Torvalds #define XLOG_STATE_ALL 0x7FFF /* All possible valid flags */ 531da177e4SLinus Torvalds #define XLOG_STATE_NOTUSED 0x8000 /* This IC log not being used */ 541da177e4SLinus Torvalds 551da177e4SLinus Torvalds /* 561da177e4SLinus Torvalds * Flags to log ticket 571da177e4SLinus Torvalds */ 581da177e4SLinus Torvalds #define XLOG_TIC_INITED 0x1 /* has been initialized */ 591da177e4SLinus Torvalds #define XLOG_TIC_PERM_RESERV 0x2 /* permanent reservation */ 600b1b213fSChristoph Hellwig 610b1b213fSChristoph Hellwig #define XLOG_TIC_FLAGS \ 620b1b213fSChristoph Hellwig { XLOG_TIC_INITED, "XLOG_TIC_INITED" }, \ 6310547941SDave Chinner { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" } 640b1b213fSChristoph Hellwig 651da177e4SLinus Torvalds /* 661da177e4SLinus Torvalds * Below are states for covering allocation transactions. 671da177e4SLinus Torvalds * By covering, we mean changing the h_tail_lsn in the last on-disk 681da177e4SLinus Torvalds * log write such that no allocation transactions will be re-done during 691da177e4SLinus Torvalds * recovery after a system crash. Recovery starts at the last on-disk 701da177e4SLinus Torvalds * log write. 711da177e4SLinus Torvalds * 721da177e4SLinus Torvalds * These states are used to insert dummy log entries to cover 731da177e4SLinus Torvalds * space allocation transactions which can undo non-transactional changes 741da177e4SLinus Torvalds * after a crash. Writes to a file with space 751da177e4SLinus Torvalds * already allocated do not result in any transactions. Allocations 761da177e4SLinus Torvalds * might include space beyond the EOF. So if we just push the EOF a 771da177e4SLinus Torvalds * little, the last transaction for the file could contain the wrong 781da177e4SLinus Torvalds * size. If there is no file system activity, after an allocation 791da177e4SLinus Torvalds * transaction, and the system crashes, the allocation transaction 801da177e4SLinus Torvalds * will get replayed and the file will be truncated. This could 811da177e4SLinus Torvalds * be hours/days/... after the allocation occurred. 821da177e4SLinus Torvalds * 831da177e4SLinus Torvalds * The fix for this is to do two dummy transactions when the 841da177e4SLinus Torvalds * system is idle. We need two dummy transaction because the h_tail_lsn 851da177e4SLinus Torvalds * in the log record header needs to point beyond the last possible 861da177e4SLinus Torvalds * non-dummy transaction. The first dummy changes the h_tail_lsn to 871da177e4SLinus Torvalds * the first transaction before the dummy. The second dummy causes 881da177e4SLinus Torvalds * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn. 891da177e4SLinus Torvalds * 901da177e4SLinus Torvalds * These dummy transactions get committed when everything 911da177e4SLinus Torvalds * is idle (after there has been some activity). 921da177e4SLinus Torvalds * 931da177e4SLinus Torvalds * There are 5 states used to control this. 941da177e4SLinus Torvalds * 951da177e4SLinus Torvalds * IDLE -- no logging has been done on the file system or 961da177e4SLinus Torvalds * we are done covering previous transactions. 971da177e4SLinus Torvalds * NEED -- logging has occurred and we need a dummy transaction 981da177e4SLinus Torvalds * when the log becomes idle. 991da177e4SLinus Torvalds * DONE -- we were in the NEED state and have committed a dummy 1001da177e4SLinus Torvalds * transaction. 1011da177e4SLinus Torvalds * NEED2 -- we detected that a dummy transaction has gone to the 1021da177e4SLinus Torvalds * on disk log with no other transactions. 1031da177e4SLinus Torvalds * DONE2 -- we committed a dummy transaction when in the NEED2 state. 1041da177e4SLinus Torvalds * 1051da177e4SLinus Torvalds * There are two places where we switch states: 1061da177e4SLinus Torvalds * 1071da177e4SLinus Torvalds * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2. 1081da177e4SLinus Torvalds * We commit the dummy transaction and switch to DONE or DONE2, 1091da177e4SLinus Torvalds * respectively. In all other states, we don't do anything. 1101da177e4SLinus Torvalds * 1111da177e4SLinus Torvalds * 2.) When we finish writing the on-disk log (xlog_state_clean_log). 1121da177e4SLinus Torvalds * 1131da177e4SLinus Torvalds * No matter what state we are in, if this isn't the dummy 1141da177e4SLinus Torvalds * transaction going out, the next state is NEED. 1151da177e4SLinus Torvalds * So, if we aren't in the DONE or DONE2 states, the next state 1161da177e4SLinus Torvalds * is NEED. We can't be finishing a write of the dummy record 1171da177e4SLinus Torvalds * unless it was committed and the state switched to DONE or DONE2. 1181da177e4SLinus Torvalds * 1191da177e4SLinus Torvalds * If we are in the DONE state and this was a write of the 1201da177e4SLinus Torvalds * dummy transaction, we move to NEED2. 1211da177e4SLinus Torvalds * 1221da177e4SLinus Torvalds * If we are in the DONE2 state and this was a write of the 1231da177e4SLinus Torvalds * dummy transaction, we move to IDLE. 1241da177e4SLinus Torvalds * 1251da177e4SLinus Torvalds * 1261da177e4SLinus Torvalds * Writing only one dummy transaction can get appended to 1271da177e4SLinus Torvalds * one file space allocation. When this happens, the log recovery 1281da177e4SLinus Torvalds * code replays the space allocation and a file could be truncated. 1291da177e4SLinus Torvalds * This is why we have the NEED2 and DONE2 states before going idle. 1301da177e4SLinus Torvalds */ 1311da177e4SLinus Torvalds 1321da177e4SLinus Torvalds #define XLOG_STATE_COVER_IDLE 0 1331da177e4SLinus Torvalds #define XLOG_STATE_COVER_NEED 1 1341da177e4SLinus Torvalds #define XLOG_STATE_COVER_DONE 2 1351da177e4SLinus Torvalds #define XLOG_STATE_COVER_NEED2 3 1361da177e4SLinus Torvalds #define XLOG_STATE_COVER_DONE2 4 1371da177e4SLinus Torvalds 1381da177e4SLinus Torvalds #define XLOG_COVER_OPS 5 1391da177e4SLinus Torvalds 1407e9c6396STim Shimmin /* Ticket reservation region accounting */ 1417e9c6396STim Shimmin #define XLOG_TIC_LEN_MAX 15 1427e9c6396STim Shimmin 1437e9c6396STim Shimmin /* 1447e9c6396STim Shimmin * Reservation region 1457e9c6396STim Shimmin * As would be stored in xfs_log_iovec but without the i_addr which 1467e9c6396STim Shimmin * we don't care about. 1477e9c6396STim Shimmin */ 1487e9c6396STim Shimmin typedef struct xlog_res { 1491259845dSTim Shimmin uint r_len; /* region length :4 */ 1501259845dSTim Shimmin uint r_type; /* region's transaction type :4 */ 1517e9c6396STim Shimmin } xlog_res_t; 1527e9c6396STim Shimmin 1531da177e4SLinus Torvalds typedef struct xlog_ticket { 15410547941SDave Chinner struct list_head t_queue; /* reserve/write queue */ 15514a7235fSChristoph Hellwig struct task_struct *t_task; /* task that owns this ticket */ 1561da177e4SLinus Torvalds xlog_tid_t t_tid; /* transaction identifier : 4 */ 157cc09c0dcSDave Chinner atomic_t t_ref; /* ticket reference count : 4 */ 1581da177e4SLinus Torvalds int t_curr_res; /* current reservation in bytes : 4 */ 1591da177e4SLinus Torvalds int t_unit_res; /* unit reservation in bytes : 4 */ 1607e9c6396STim Shimmin char t_ocnt; /* original count : 1 */ 1617e9c6396STim Shimmin char t_cnt; /* current count : 1 */ 1627e9c6396STim Shimmin char t_clientid; /* who does this belong to; : 1 */ 1637e9c6396STim Shimmin char t_flags; /* properties of reservation : 1 */ 1647e9c6396STim Shimmin 1657e9c6396STim Shimmin /* reservation array fields */ 1667e9c6396STim Shimmin uint t_res_num; /* num in array : 4 */ 1677e9c6396STim Shimmin uint t_res_num_ophdrs; /* num op hdrs : 4 */ 1687e9c6396STim Shimmin uint t_res_arr_sum; /* array sum : 4 */ 1697e9c6396STim Shimmin uint t_res_o_flow; /* sum overflow : 4 */ 1701259845dSTim Shimmin xlog_res_t t_res_arr[XLOG_TIC_LEN_MAX]; /* array of res : 8 * 15 */ 1711da177e4SLinus Torvalds } xlog_ticket_t; 1727e9c6396STim Shimmin 1731da177e4SLinus Torvalds /* 1741da177e4SLinus Torvalds * - A log record header is 512 bytes. There is plenty of room to grow the 1751da177e4SLinus Torvalds * xlog_rec_header_t into the reserved space. 1761da177e4SLinus Torvalds * - ic_data follows, so a write to disk can start at the beginning of 1771da177e4SLinus Torvalds * the iclog. 17812017fafSDavid Chinner * - ic_forcewait is used to implement synchronous forcing of the iclog to disk. 1791da177e4SLinus Torvalds * - ic_next is the pointer to the next iclog in the ring. 1801da177e4SLinus Torvalds * - ic_log is a pointer back to the global log structure. 18179b54d9bSChristoph Hellwig * - ic_size is the full size of the log buffer, minus the cycle headers. 18279b54d9bSChristoph Hellwig * - ic_io_size is the size of the currently pending log buffer write, which 18379b54d9bSChristoph Hellwig * might be smaller than ic_size 1841da177e4SLinus Torvalds * - ic_offset is the current number of bytes written to in this iclog. 1851da177e4SLinus Torvalds * - ic_refcnt is bumped when someone is writing to the log. 1861da177e4SLinus Torvalds * - ic_state is the state of the iclog. 187114d23aaSDavid Chinner * 188114d23aaSDavid Chinner * Because of cacheline contention on large machines, we need to separate 189114d23aaSDavid Chinner * various resources onto different cachelines. To start with, make the 190114d23aaSDavid Chinner * structure cacheline aligned. The following fields can be contended on 191114d23aaSDavid Chinner * by independent processes: 192114d23aaSDavid Chinner * 19389ae379dSChristoph Hellwig * - ic_callbacks 194114d23aaSDavid Chinner * - ic_refcnt 195114d23aaSDavid Chinner * - fields protected by the global l_icloglock 196114d23aaSDavid Chinner * 197114d23aaSDavid Chinner * so we need to ensure that these fields are located in separate cachelines. 198114d23aaSDavid Chinner * We'll put all the read-only and l_icloglock fields in the first cacheline, 199114d23aaSDavid Chinner * and move everything else out to subsequent cachelines. 2001da177e4SLinus Torvalds */ 201b28708d6SChristoph Hellwig typedef struct xlog_in_core { 202eb40a875SDave Chinner wait_queue_head_t ic_force_wait; 203eb40a875SDave Chinner wait_queue_head_t ic_write_wait; 2041da177e4SLinus Torvalds struct xlog_in_core *ic_next; 2051da177e4SLinus Torvalds struct xlog_in_core *ic_prev; 206ad223e60SMark Tinguely struct xlog *ic_log; 20779b54d9bSChristoph Hellwig u32 ic_size; 20879b54d9bSChristoph Hellwig u32 ic_io_size; 20979b54d9bSChristoph Hellwig u32 ic_offset; 210a5687787SChristoph Hellwig unsigned short ic_state; 2111da177e4SLinus Torvalds char *ic_datap; /* pointer to iclog data */ 212114d23aaSDavid Chinner 213114d23aaSDavid Chinner /* Callback structures need their own cacheline */ 214114d23aaSDavid Chinner spinlock_t ic_callback_lock ____cacheline_aligned_in_smp; 21589ae379dSChristoph Hellwig struct list_head ic_callbacks; 216114d23aaSDavid Chinner 217114d23aaSDavid Chinner /* reference counts need their own cacheline */ 218114d23aaSDavid Chinner atomic_t ic_refcnt ____cacheline_aligned_in_smp; 219b28708d6SChristoph Hellwig xlog_in_core_2_t *ic_data; 220b28708d6SChristoph Hellwig #define ic_header ic_data->hic_header 221366fc4b8SChristoph Hellwig #ifdef DEBUG 222366fc4b8SChristoph Hellwig bool ic_fail_crc : 1; 223366fc4b8SChristoph Hellwig #endif 22479b54d9bSChristoph Hellwig struct semaphore ic_sema; 22579b54d9bSChristoph Hellwig struct work_struct ic_end_io_work; 22679b54d9bSChristoph Hellwig struct bio ic_bio; 22779b54d9bSChristoph Hellwig struct bio_vec ic_bvec[]; 2281da177e4SLinus Torvalds } xlog_in_core_t; 2291da177e4SLinus Torvalds 2301da177e4SLinus Torvalds /* 23171e330b5SDave Chinner * The CIL context is used to aggregate per-transaction details as well be 23271e330b5SDave Chinner * passed to the iclog for checkpoint post-commit processing. After being 23371e330b5SDave Chinner * passed to the iclog, another context needs to be allocated for tracking the 23471e330b5SDave Chinner * next set of transactions to be aggregated into a checkpoint. 23571e330b5SDave Chinner */ 23671e330b5SDave Chinner struct xfs_cil; 23771e330b5SDave Chinner 23871e330b5SDave Chinner struct xfs_cil_ctx { 23971e330b5SDave Chinner struct xfs_cil *cil; 24071e330b5SDave Chinner xfs_lsn_t sequence; /* chkpt sequence # */ 24171e330b5SDave Chinner xfs_lsn_t start_lsn; /* first LSN of chkpt commit */ 24271e330b5SDave Chinner xfs_lsn_t commit_lsn; /* chkpt commit record lsn */ 24371e330b5SDave Chinner struct xlog_ticket *ticket; /* chkpt ticket */ 24471e330b5SDave Chinner int nvecs; /* number of regions */ 24571e330b5SDave Chinner int space_used; /* aggregate size of regions */ 24671e330b5SDave Chinner struct list_head busy_extents; /* busy extents in chkpt */ 24771e330b5SDave Chinner struct xfs_log_vec *lv_chain; /* logvecs being pushed */ 24889ae379dSChristoph Hellwig struct list_head iclog_entry; 24971e330b5SDave Chinner struct list_head committing; /* ctx committing list */ 2504560e78fSChristoph Hellwig struct work_struct discard_endio_work; 25171e330b5SDave Chinner }; 25271e330b5SDave Chinner 25371e330b5SDave Chinner /* 25471e330b5SDave Chinner * Committed Item List structure 25571e330b5SDave Chinner * 25671e330b5SDave Chinner * This structure is used to track log items that have been committed but not 25771e330b5SDave Chinner * yet written into the log. It is used only when the delayed logging mount 25871e330b5SDave Chinner * option is enabled. 25971e330b5SDave Chinner * 26071e330b5SDave Chinner * This structure tracks the list of committing checkpoint contexts so 26171e330b5SDave Chinner * we can avoid the problem of having to hold out new transactions during a 26271e330b5SDave Chinner * flush until we have a the commit record LSN of the checkpoint. We can 26371e330b5SDave Chinner * traverse the list of committing contexts in xlog_cil_push_lsn() to find a 26471e330b5SDave Chinner * sequence match and extract the commit LSN directly from there. If the 26571e330b5SDave Chinner * checkpoint is still in the process of committing, we can block waiting for 26671e330b5SDave Chinner * the commit LSN to be determined as well. This should make synchronous 26771e330b5SDave Chinner * operations almost as efficient as the old logging methods. 26871e330b5SDave Chinner */ 26971e330b5SDave Chinner struct xfs_cil { 270ad223e60SMark Tinguely struct xlog *xc_log; 27171e330b5SDave Chinner struct list_head xc_cil; 27271e330b5SDave Chinner spinlock_t xc_cil_lock; 2734bb928cdSDave Chinner 2744bb928cdSDave Chinner struct rw_semaphore xc_ctx_lock ____cacheline_aligned_in_smp; 27571e330b5SDave Chinner struct xfs_cil_ctx *xc_ctx; 2764bb928cdSDave Chinner 2774bb928cdSDave Chinner spinlock_t xc_push_lock ____cacheline_aligned_in_smp; 2784bb928cdSDave Chinner xfs_lsn_t xc_push_seq; 27971e330b5SDave Chinner struct list_head xc_committing; 280eb40a875SDave Chinner wait_queue_head_t xc_commit_wait; 281a44f13edSDave Chinner xfs_lsn_t xc_current_sequence; 2824c2d542fSDave Chinner struct work_struct xc_push_work; 2834bb928cdSDave Chinner } ____cacheline_aligned_in_smp; 28471e330b5SDave Chinner 28571e330b5SDave Chinner /* 28680168676SDave Chinner * The amount of log space we allow the CIL to aggregate is difficult to size. 28780168676SDave Chinner * Whatever we choose, we have to make sure we can get a reservation for the 28880168676SDave Chinner * log space effectively, that it is large enough to capture sufficient 28980168676SDave Chinner * relogging to reduce log buffer IO significantly, but it is not too large for 29080168676SDave Chinner * the log or induces too much latency when writing out through the iclogs. We 29180168676SDave Chinner * track both space consumed and the number of vectors in the checkpoint 29280168676SDave Chinner * context, so we need to decide which to use for limiting. 293df806158SDave Chinner * 294df806158SDave Chinner * Every log buffer we write out during a push needs a header reserved, which 295df806158SDave Chinner * is at least one sector and more for v2 logs. Hence we need a reservation of 296df806158SDave Chinner * at least 512 bytes per 32k of log space just for the LR headers. That means 297df806158SDave Chinner * 16KB of reservation per megabyte of delayed logging space we will consume, 298df806158SDave Chinner * plus various headers. The number of headers will vary based on the num of 299df806158SDave Chinner * io vectors, so limiting on a specific number of vectors is going to result 300df806158SDave Chinner * in transactions of varying size. IOWs, it is more consistent to track and 301df806158SDave Chinner * limit space consumed in the log rather than by the number of objects being 302df806158SDave Chinner * logged in order to prevent checkpoint ticket overruns. 303df806158SDave Chinner * 304df806158SDave Chinner * Further, use of static reservations through the log grant mechanism is 305df806158SDave Chinner * problematic. It introduces a lot of complexity (e.g. reserve grant vs write 306df806158SDave Chinner * grant) and a significant deadlock potential because regranting write space 307df806158SDave Chinner * can block on log pushes. Hence if we have to regrant log space during a log 308df806158SDave Chinner * push, we can deadlock. 309df806158SDave Chinner * 310df806158SDave Chinner * However, we can avoid this by use of a dynamic "reservation stealing" 311df806158SDave Chinner * technique during transaction commit whereby unused reservation space in the 312df806158SDave Chinner * transaction ticket is transferred to the CIL ctx commit ticket to cover the 313df806158SDave Chinner * space needed by the checkpoint transaction. This means that we never need to 314df806158SDave Chinner * specifically reserve space for the CIL checkpoint transaction, nor do we 315df806158SDave Chinner * need to regrant space once the checkpoint completes. This also means the 316df806158SDave Chinner * checkpoint transaction ticket is specific to the checkpoint context, rather 317df806158SDave Chinner * than the CIL itself. 318df806158SDave Chinner * 31980168676SDave Chinner * With dynamic reservations, we can effectively make up arbitrary limits for 32080168676SDave Chinner * the checkpoint size so long as they don't violate any other size rules. 32180168676SDave Chinner * Recovery imposes a rule that no transaction exceed half the log, so we are 32280168676SDave Chinner * limited by that. Furthermore, the log transaction reservation subsystem 32380168676SDave Chinner * tries to keep 25% of the log free, so we need to keep below that limit or we 32480168676SDave Chinner * risk running out of free log space to start any new transactions. 32580168676SDave Chinner * 32680168676SDave Chinner * In order to keep background CIL push efficient, we will set a lower 32780168676SDave Chinner * threshold at which background pushing is attempted without blocking current 32880168676SDave Chinner * transaction commits. A separate, higher bound defines when CIL pushes are 32980168676SDave Chinner * enforced to ensure we stay within our maximum checkpoint size bounds. 33080168676SDave Chinner * threshold, yet give us plenty of space for aggregation on large logs. 331df806158SDave Chinner */ 33280168676SDave Chinner #define XLOG_CIL_SPACE_LIMIT(log) (log->l_logsize >> 3) 333df806158SDave Chinner 334df806158SDave Chinner /* 33528496968SChristoph Hellwig * ticket grant locks, queues and accounting have their own cachlines 33628496968SChristoph Hellwig * as these are quite hot and can be operated on concurrently. 33728496968SChristoph Hellwig */ 33828496968SChristoph Hellwig struct xlog_grant_head { 33928496968SChristoph Hellwig spinlock_t lock ____cacheline_aligned_in_smp; 34028496968SChristoph Hellwig struct list_head waiters; 34128496968SChristoph Hellwig atomic64_t grant; 34228496968SChristoph Hellwig }; 34328496968SChristoph Hellwig 34428496968SChristoph Hellwig /* 3451da177e4SLinus Torvalds * The reservation head lsn is not made up of a cycle number and block number. 3461da177e4SLinus Torvalds * Instead, it uses a cycle number and byte number. Logs don't expect to 3471da177e4SLinus Torvalds * overflow 31 bits worth of byte offset, so using a byte number will mean 3481da177e4SLinus Torvalds * that round off problems won't occur when releasing partial reservations. 3491da177e4SLinus Torvalds */ 3509a8d2fdbSMark Tinguely struct xlog { 3514679b2d3SDavid Chinner /* The following fields don't need locking */ 3524679b2d3SDavid Chinner struct xfs_mount *l_mp; /* mount point */ 353a9c21c1bSDavid Chinner struct xfs_ail *l_ailp; /* AIL log is working with */ 35471e330b5SDave Chinner struct xfs_cil *l_cilp; /* CIL log is working with */ 3554679b2d3SDavid Chinner struct xfs_buftarg *l_targ; /* buftarg of log */ 3561058d0f5SChristoph Hellwig struct workqueue_struct *l_ioend_workqueue; /* for I/O completions */ 357f661f1e0SDave Chinner struct delayed_work l_work; /* background flush work */ 3584679b2d3SDavid Chinner uint l_flags; 3594679b2d3SDavid Chinner uint l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */ 360d5689eaaSChristoph Hellwig struct list_head *l_buf_cancel_table; 3614679b2d3SDavid Chinner int l_iclog_hsize; /* size of iclog header */ 3624679b2d3SDavid Chinner int l_iclog_heads; /* # of iclog header sectors */ 36348389ef1SAlex Elder uint l_sectBBsize; /* sector size in BBs (2^n) */ 3644679b2d3SDavid Chinner int l_iclog_size; /* size of log in bytes */ 3654679b2d3SDavid Chinner int l_iclog_bufs; /* number of iclog buffers */ 3664679b2d3SDavid Chinner xfs_daddr_t l_logBBstart; /* start block of log */ 3674679b2d3SDavid Chinner int l_logsize; /* size of log in bytes */ 3684679b2d3SDavid Chinner int l_logBBsize; /* size of log in BB chunks */ 3694679b2d3SDavid Chinner 3701da177e4SLinus Torvalds /* The following block of fields are changed while holding icloglock */ 371eb40a875SDave Chinner wait_queue_head_t l_flush_wait ____cacheline_aligned_in_smp; 372d748c623SMatthew Wilcox /* waiting for iclog flush */ 3731da177e4SLinus Torvalds int l_covered_state;/* state of "covering disk 3741da177e4SLinus Torvalds * log entries" */ 3751da177e4SLinus Torvalds xlog_in_core_t *l_iclog; /* head log queue */ 376b22cd72cSEric Sandeen spinlock_t l_icloglock; /* grab to change iclog state */ 3771da177e4SLinus Torvalds int l_curr_cycle; /* Cycle number of log writes */ 3781da177e4SLinus Torvalds int l_prev_cycle; /* Cycle number before last 3791da177e4SLinus Torvalds * block increment */ 3801da177e4SLinus Torvalds int l_curr_block; /* current logical log block */ 3811da177e4SLinus Torvalds int l_prev_block; /* previous logical log block */ 3821da177e4SLinus Torvalds 38384f3c683SDave Chinner /* 3841c3cb9ecSDave Chinner * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and 3851c3cb9ecSDave Chinner * read without needing to hold specific locks. To avoid operations 3861c3cb9ecSDave Chinner * contending with other hot objects, place each of them on a separate 3871c3cb9ecSDave Chinner * cacheline. 38884f3c683SDave Chinner */ 38984f3c683SDave Chinner /* lsn of last LR on disk */ 39084f3c683SDave Chinner atomic64_t l_last_sync_lsn ____cacheline_aligned_in_smp; 3911c3cb9ecSDave Chinner /* lsn of 1st LR with unflushed * buffers */ 3921c3cb9ecSDave Chinner atomic64_t l_tail_lsn ____cacheline_aligned_in_smp; 39384f3c683SDave Chinner 39428496968SChristoph Hellwig struct xlog_grant_head l_reserve_head; 39528496968SChristoph Hellwig struct xlog_grant_head l_write_head; 3963f16b985SDave Chinner 397baff4e44SBrian Foster struct xfs_kobj l_kobj; 398baff4e44SBrian Foster 3994679b2d3SDavid Chinner /* The following field are used for debugging; need to hold icloglock */ 4004679b2d3SDavid Chinner #ifdef DEBUG 4015809d5e0SChristoph Hellwig void *l_iclog_bak[XLOG_MAX_ICLOGS]; 402609adfc2SBrian Foster /* log record crc error injection factor */ 403609adfc2SBrian Foster uint32_t l_badcrc_factor; 4044679b2d3SDavid Chinner #endif 40512818d24SBrian Foster /* log recovery lsn tracking (for buffer submission */ 40612818d24SBrian Foster xfs_lsn_t l_recovery_lsn; 4079a8d2fdbSMark Tinguely }; 4081da177e4SLinus Torvalds 409d5689eaaSChristoph Hellwig #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \ 410c8ce540dSDarrick J. Wong ((log)->l_buf_cancel_table + ((uint64_t)blkno % XLOG_BC_TABLE_SIZE)) 411d5689eaaSChristoph Hellwig 412cfcbbbd0SNathan Scott #define XLOG_FORCED_SHUTDOWN(log) ((log)->l_flags & XLOG_IO_ERROR) 413cfcbbbd0SNathan Scott 4141da177e4SLinus Torvalds /* common routines */ 4159a8d2fdbSMark Tinguely extern int 4169a8d2fdbSMark Tinguely xlog_recover( 4179a8d2fdbSMark Tinguely struct xlog *log); 4189a8d2fdbSMark Tinguely extern int 4199a8d2fdbSMark Tinguely xlog_recover_finish( 4209a8d2fdbSMark Tinguely struct xlog *log); 421*a7a9250eSHariprasad Kelam extern void 422f0b2efadSBrian Foster xlog_recover_cancel(struct xlog *); 4230e446be4SChristoph Hellwig 424f9668a09SDave Chinner extern __le32 xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead, 4250e446be4SChristoph Hellwig char *dp, int size); 4261da177e4SLinus Torvalds 427eb01c9cdSDavid Chinner extern kmem_zone_t *xfs_log_ticket_zone; 428ad223e60SMark Tinguely struct xlog_ticket * 429ad223e60SMark Tinguely xlog_ticket_alloc( 430ad223e60SMark Tinguely struct xlog *log, 431ad223e60SMark Tinguely int unit_bytes, 432ad223e60SMark Tinguely int count, 433ad223e60SMark Tinguely char client, 434ad223e60SMark Tinguely bool permanent, 43577ba7877SAl Viro xfs_km_flags_t alloc_flags); 43671e330b5SDave Chinner 437eb01c9cdSDavid Chinner 438e6b1f273SChristoph Hellwig static inline void 439e6b1f273SChristoph Hellwig xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes) 440e6b1f273SChristoph Hellwig { 441e6b1f273SChristoph Hellwig *ptr += bytes; 442e6b1f273SChristoph Hellwig *len -= bytes; 443e6b1f273SChristoph Hellwig *off += bytes; 444e6b1f273SChristoph Hellwig } 445e6b1f273SChristoph Hellwig 44671e330b5SDave Chinner void xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket); 447d4ca1d55SBrian Foster void xlog_print_trans(struct xfs_trans *); 448ad223e60SMark Tinguely int 449ad223e60SMark Tinguely xlog_write( 450ad223e60SMark Tinguely struct xlog *log, 451ad223e60SMark Tinguely struct xfs_log_vec *log_vector, 452ad223e60SMark Tinguely struct xlog_ticket *tic, 453ad223e60SMark Tinguely xfs_lsn_t *start_lsn, 454ad223e60SMark Tinguely struct xlog_in_core **commit_iclog, 455ad223e60SMark Tinguely uint flags); 45671e330b5SDave Chinner 45771e330b5SDave Chinner /* 4581c3cb9ecSDave Chinner * When we crack an atomic LSN, we sample it first so that the value will not 4591c3cb9ecSDave Chinner * change while we are cracking it into the component values. This means we 4601c3cb9ecSDave Chinner * will always get consistent component values to work from. This should always 46125985edcSLucas De Marchi * be used to sample and crack LSNs that are stored and updated in atomic 4621c3cb9ecSDave Chinner * variables. 4631c3cb9ecSDave Chinner */ 4641c3cb9ecSDave Chinner static inline void 4651c3cb9ecSDave Chinner xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block) 4661c3cb9ecSDave Chinner { 4671c3cb9ecSDave Chinner xfs_lsn_t val = atomic64_read(lsn); 4681c3cb9ecSDave Chinner 4691c3cb9ecSDave Chinner *cycle = CYCLE_LSN(val); 4701c3cb9ecSDave Chinner *block = BLOCK_LSN(val); 4711c3cb9ecSDave Chinner } 4721c3cb9ecSDave Chinner 4731c3cb9ecSDave Chinner /* 4741c3cb9ecSDave Chinner * Calculate and assign a value to an atomic LSN variable from component pieces. 4751c3cb9ecSDave Chinner */ 4761c3cb9ecSDave Chinner static inline void 4771c3cb9ecSDave Chinner xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block) 4781c3cb9ecSDave Chinner { 4791c3cb9ecSDave Chinner atomic64_set(lsn, xlog_assign_lsn(cycle, block)); 4801c3cb9ecSDave Chinner } 4811c3cb9ecSDave Chinner 4821c3cb9ecSDave Chinner /* 483d0eb2f38SDave Chinner * When we crack the grant head, we sample it first so that the value will not 484a69ed03cSDave Chinner * change while we are cracking it into the component values. This means we 485a69ed03cSDave Chinner * will always get consistent component values to work from. 486a69ed03cSDave Chinner */ 487a69ed03cSDave Chinner static inline void 488d0eb2f38SDave Chinner xlog_crack_grant_head_val(int64_t val, int *cycle, int *space) 489a69ed03cSDave Chinner { 490a69ed03cSDave Chinner *cycle = val >> 32; 491a69ed03cSDave Chinner *space = val & 0xffffffff; 492a69ed03cSDave Chinner } 493a69ed03cSDave Chinner 494a69ed03cSDave Chinner static inline void 495d0eb2f38SDave Chinner xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space) 496d0eb2f38SDave Chinner { 497d0eb2f38SDave Chinner xlog_crack_grant_head_val(atomic64_read(head), cycle, space); 498d0eb2f38SDave Chinner } 499d0eb2f38SDave Chinner 500d0eb2f38SDave Chinner static inline int64_t 501d0eb2f38SDave Chinner xlog_assign_grant_head_val(int cycle, int space) 502d0eb2f38SDave Chinner { 503d0eb2f38SDave Chinner return ((int64_t)cycle << 32) | space; 504d0eb2f38SDave Chinner } 505d0eb2f38SDave Chinner 506d0eb2f38SDave Chinner static inline void 507c8a09ff8SDave Chinner xlog_assign_grant_head(atomic64_t *head, int cycle, int space) 508a69ed03cSDave Chinner { 509d0eb2f38SDave Chinner atomic64_set(head, xlog_assign_grant_head_val(cycle, space)); 510a69ed03cSDave Chinner } 511a69ed03cSDave Chinner 512a69ed03cSDave Chinner /* 51371e330b5SDave Chinner * Committed Item List interfaces 51471e330b5SDave Chinner */ 5152c6e24ceSDave Chinner int xlog_cil_init(struct xlog *log); 5162c6e24ceSDave Chinner void xlog_cil_init_post_recovery(struct xlog *log); 5172c6e24ceSDave Chinner void xlog_cil_destroy(struct xlog *log); 5182c6e24ceSDave Chinner bool xlog_cil_empty(struct xlog *log); 51971e330b5SDave Chinner 520a44f13edSDave Chinner /* 521a44f13edSDave Chinner * CIL force routines 522a44f13edSDave Chinner */ 523ad223e60SMark Tinguely xfs_lsn_t 524ad223e60SMark Tinguely xlog_cil_force_lsn( 525ad223e60SMark Tinguely struct xlog *log, 526ad223e60SMark Tinguely xfs_lsn_t sequence); 527a44f13edSDave Chinner 528a44f13edSDave Chinner static inline void 529ad223e60SMark Tinguely xlog_cil_force(struct xlog *log) 530a44f13edSDave Chinner { 531a44f13edSDave Chinner xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence); 532a44f13edSDave Chinner } 53371e330b5SDave Chinner 534955e47adSTim Shimmin /* 535955e47adSTim Shimmin * Unmount record type is used as a pseudo transaction type for the ticket. 536955e47adSTim Shimmin * It's value must be outside the range of XFS_TRANS_* values. 537955e47adSTim Shimmin */ 538955e47adSTim Shimmin #define XLOG_UNMOUNT_REC_TYPE (-1U) 539955e47adSTim Shimmin 540eb40a875SDave Chinner /* 541eb40a875SDave Chinner * Wrapper function for waiting on a wait queue serialised against wakeups 542eb40a875SDave Chinner * by a spinlock. This matches the semantics of all the wait queues used in the 543eb40a875SDave Chinner * log code. 544eb40a875SDave Chinner */ 545eb40a875SDave Chinner static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock) 546eb40a875SDave Chinner { 547eb40a875SDave Chinner DECLARE_WAITQUEUE(wait, current); 548eb40a875SDave Chinner 549eb40a875SDave Chinner add_wait_queue_exclusive(wq, &wait); 550eb40a875SDave Chinner __set_current_state(TASK_UNINTERRUPTIBLE); 551eb40a875SDave Chinner spin_unlock(lock); 552eb40a875SDave Chinner schedule(); 553eb40a875SDave Chinner remove_wait_queue(wq, &wait); 554eb40a875SDave Chinner } 5551da177e4SLinus Torvalds 556a45086e2SBrian Foster /* 557a45086e2SBrian Foster * The LSN is valid so long as it is behind the current LSN. If it isn't, this 558a45086e2SBrian Foster * means that the next log record that includes this metadata could have a 559a45086e2SBrian Foster * smaller LSN. In turn, this means that the modification in the log would not 560a45086e2SBrian Foster * replay. 561a45086e2SBrian Foster */ 562a45086e2SBrian Foster static inline bool 563a45086e2SBrian Foster xlog_valid_lsn( 564a45086e2SBrian Foster struct xlog *log, 565a45086e2SBrian Foster xfs_lsn_t lsn) 566a45086e2SBrian Foster { 567a45086e2SBrian Foster int cur_cycle; 568a45086e2SBrian Foster int cur_block; 569a45086e2SBrian Foster bool valid = true; 570a45086e2SBrian Foster 571a45086e2SBrian Foster /* 572a45086e2SBrian Foster * First, sample the current lsn without locking to avoid added 573a45086e2SBrian Foster * contention from metadata I/O. The current cycle and block are updated 574a45086e2SBrian Foster * (in xlog_state_switch_iclogs()) and read here in a particular order 575a45086e2SBrian Foster * to avoid false negatives (e.g., thinking the metadata LSN is valid 576a45086e2SBrian Foster * when it is not). 577a45086e2SBrian Foster * 578a45086e2SBrian Foster * The current block is always rewound before the cycle is bumped in 579a45086e2SBrian Foster * xlog_state_switch_iclogs() to ensure the current LSN is never seen in 580a45086e2SBrian Foster * a transiently forward state. Instead, we can see the LSN in a 581a45086e2SBrian Foster * transiently behind state if we happen to race with a cycle wrap. 582a45086e2SBrian Foster */ 5836aa7de05SMark Rutland cur_cycle = READ_ONCE(log->l_curr_cycle); 584a45086e2SBrian Foster smp_rmb(); 5856aa7de05SMark Rutland cur_block = READ_ONCE(log->l_curr_block); 586a45086e2SBrian Foster 587a45086e2SBrian Foster if ((CYCLE_LSN(lsn) > cur_cycle) || 588a45086e2SBrian Foster (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) { 589a45086e2SBrian Foster /* 590a45086e2SBrian Foster * If the metadata LSN appears invalid, it's possible the check 591a45086e2SBrian Foster * above raced with a wrap to the next log cycle. Grab the lock 592a45086e2SBrian Foster * to check for sure. 593a45086e2SBrian Foster */ 594a45086e2SBrian Foster spin_lock(&log->l_icloglock); 595a45086e2SBrian Foster cur_cycle = log->l_curr_cycle; 596a45086e2SBrian Foster cur_block = log->l_curr_block; 597a45086e2SBrian Foster spin_unlock(&log->l_icloglock); 598a45086e2SBrian Foster 599a45086e2SBrian Foster if ((CYCLE_LSN(lsn) > cur_cycle) || 600a45086e2SBrian Foster (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) 601a45086e2SBrian Foster valid = false; 602a45086e2SBrian Foster } 603a45086e2SBrian Foster 604a45086e2SBrian Foster return valid; 605a45086e2SBrian Foster } 606a45086e2SBrian Foster 6071da177e4SLinus Torvalds #endif /* __XFS_LOG_PRIV_H__ */ 608