root/fs/xfs/xfs_log_priv.h

/* [<][>][^][v][top][bottom][index][help] */

INCLUDED FROM


DEFINITIONS

This source file includes following definitions.
  1. xlog_get_client_id
  2. xlog_write_adv_cnt
  3. xlog_crack_atomic_lsn
  4. xlog_assign_atomic_lsn
  5. xlog_crack_grant_head_val
  6. xlog_crack_grant_head
  7. xlog_assign_grant_head_val
  8. xlog_assign_grant_head
  9. xlog_cil_force
  10. xlog_wait
  11. xlog_valid_lsn

   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 #define XLOG_STATE_ACTIVE    0x0001 /* Current IC log being written to */
  44 #define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
  45 #define XLOG_STATE_SYNCING   0x0004 /* This IC log is syncing */
  46 #define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
  47 #define XLOG_STATE_DO_CALLBACK \
  48                              0x0010 /* Process callback functions */
  49 #define XLOG_STATE_CALLBACK  0x0020 /* Callback functions now */
  50 #define XLOG_STATE_DIRTY     0x0040 /* Dirty IC log, not ready for ACTIVE status*/
  51 #define XLOG_STATE_IOERROR   0x0080 /* IO error happened in sync'ing log */
  52 #define XLOG_STATE_ALL       0x7FFF /* All possible valid flags */
  53 #define XLOG_STATE_NOTUSED   0x8000 /* This IC log not being used */
  54 
  55 /*
  56  * Flags to log ticket
  57  */
  58 #define XLOG_TIC_INITED         0x1     /* has been initialized */
  59 #define XLOG_TIC_PERM_RESERV    0x2     /* permanent reservation */
  60 
  61 #define XLOG_TIC_FLAGS \
  62         { XLOG_TIC_INITED,      "XLOG_TIC_INITED" }, \
  63         { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
  64 
  65 /*
  66  * Below are states for covering allocation transactions.
  67  * By covering, we mean changing the h_tail_lsn in the last on-disk
  68  * log write such that no allocation transactions will be re-done during
  69  * recovery after a system crash. Recovery starts at the last on-disk
  70  * log write.
  71  *
  72  * These states are used to insert dummy log entries to cover
  73  * space allocation transactions which can undo non-transactional changes
  74  * after a crash. Writes to a file with space
  75  * already allocated do not result in any transactions. Allocations
  76  * might include space beyond the EOF. So if we just push the EOF a
  77  * little, the last transaction for the file could contain the wrong
  78  * size. If there is no file system activity, after an allocation
  79  * transaction, and the system crashes, the allocation transaction
  80  * will get replayed and the file will be truncated. This could
  81  * be hours/days/... after the allocation occurred.
  82  *
  83  * The fix for this is to do two dummy transactions when the
  84  * system is idle. We need two dummy transaction because the h_tail_lsn
  85  * in the log record header needs to point beyond the last possible
  86  * non-dummy transaction. The first dummy changes the h_tail_lsn to
  87  * the first transaction before the dummy. The second dummy causes
  88  * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
  89  *
  90  * These dummy transactions get committed when everything
  91  * is idle (after there has been some activity).
  92  *
  93  * There are 5 states used to control this.
  94  *
  95  *  IDLE -- no logging has been done on the file system or
  96  *              we are done covering previous transactions.
  97  *  NEED -- logging has occurred and we need a dummy transaction
  98  *              when the log becomes idle.
  99  *  DONE -- we were in the NEED state and have committed a dummy
 100  *              transaction.
 101  *  NEED2 -- we detected that a dummy transaction has gone to the
 102  *              on disk log with no other transactions.
 103  *  DONE2 -- we committed a dummy transaction when in the NEED2 state.
 104  *
 105  * There are two places where we switch states:
 106  *
 107  * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
 108  *      We commit the dummy transaction and switch to DONE or DONE2,
 109  *      respectively. In all other states, we don't do anything.
 110  *
 111  * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
 112  *
 113  *      No matter what state we are in, if this isn't the dummy
 114  *      transaction going out, the next state is NEED.
 115  *      So, if we aren't in the DONE or DONE2 states, the next state
 116  *      is NEED. We can't be finishing a write of the dummy record
 117  *      unless it was committed and the state switched to DONE or DONE2.
 118  *
 119  *      If we are in the DONE state and this was a write of the
 120  *              dummy transaction, we move to NEED2.
 121  *
 122  *      If we are in the DONE2 state and this was a write of the
 123  *              dummy transaction, we move to IDLE.
 124  *
 125  *
 126  * Writing only one dummy transaction can get appended to
 127  * one file space allocation. When this happens, the log recovery
 128  * code replays the space allocation and a file could be truncated.
 129  * This is why we have the NEED2 and DONE2 states before going idle.
 130  */
 131 
 132 #define XLOG_STATE_COVER_IDLE   0
 133 #define XLOG_STATE_COVER_NEED   1
 134 #define XLOG_STATE_COVER_DONE   2
 135 #define XLOG_STATE_COVER_NEED2  3
 136 #define XLOG_STATE_COVER_DONE2  4
 137 
 138 #define XLOG_COVER_OPS          5
 139 
 140 /* Ticket reservation region accounting */ 
 141 #define XLOG_TIC_LEN_MAX        15
 142 
 143 /*
 144  * Reservation region
 145  * As would be stored in xfs_log_iovec but without the i_addr which
 146  * we don't care about.
 147  */
 148 typedef struct xlog_res {
 149         uint    r_len;  /* region length                :4 */
 150         uint    r_type; /* region's transaction type    :4 */
 151 } xlog_res_t;
 152 
 153 typedef struct xlog_ticket {
 154         struct list_head   t_queue;      /* reserve/write queue */
 155         struct task_struct *t_task;      /* task that owns this ticket */
 156         xlog_tid_t         t_tid;        /* transaction identifier       : 4  */
 157         atomic_t           t_ref;        /* ticket reference count       : 4  */
 158         int                t_curr_res;   /* current reservation in bytes : 4  */
 159         int                t_unit_res;   /* unit reservation in bytes    : 4  */
 160         char               t_ocnt;       /* original count               : 1  */
 161         char               t_cnt;        /* current count                : 1  */
 162         char               t_clientid;   /* who does this belong to;     : 1  */
 163         char               t_flags;      /* properties of reservation    : 1  */
 164 
 165         /* reservation array fields */
 166         uint               t_res_num;                    /* num in array : 4 */
 167         uint               t_res_num_ophdrs;             /* num op hdrs  : 4 */
 168         uint               t_res_arr_sum;                /* array sum    : 4 */
 169         uint               t_res_o_flow;                 /* sum overflow : 4 */
 170         xlog_res_t         t_res_arr[XLOG_TIC_LEN_MAX];  /* array of res : 8 * 15 */ 
 171 } xlog_ticket_t;
 172 
 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_log is a pointer back to the global log structure.
 181  * - ic_size is the full size of the log buffer, minus the cycle headers.
 182  * - ic_io_size is the size of the currently pending log buffer write, which
 183  *      might be smaller than ic_size
 184  * - ic_offset is the current number of bytes written to in this iclog.
 185  * - ic_refcnt is bumped when someone is writing to the log.
 186  * - ic_state is the state of the iclog.
 187  *
 188  * Because of cacheline contention on large machines, we need to separate
 189  * various resources onto different cachelines. To start with, make the
 190  * structure cacheline aligned. The following fields can be contended on
 191  * by independent processes:
 192  *
 193  *      - ic_callbacks
 194  *      - ic_refcnt
 195  *      - fields protected by the global l_icloglock
 196  *
 197  * so we need to ensure that these fields are located in separate cachelines.
 198  * We'll put all the read-only and l_icloglock fields in the first cacheline,
 199  * and move everything else out to subsequent cachelines.
 200  */
 201 typedef struct xlog_in_core {
 202         wait_queue_head_t       ic_force_wait;
 203         wait_queue_head_t       ic_write_wait;
 204         struct xlog_in_core     *ic_next;
 205         struct xlog_in_core     *ic_prev;
 206         struct xlog             *ic_log;
 207         u32                     ic_size;
 208         u32                     ic_io_size;
 209         u32                     ic_offset;
 210         unsigned short          ic_state;
 211         char                    *ic_datap;      /* pointer to iclog data */
 212 
 213         /* Callback structures need their own cacheline */
 214         spinlock_t              ic_callback_lock ____cacheline_aligned_in_smp;
 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_lsn_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_ticket      *ticket;        /* chkpt ticket */
 244         int                     nvecs;          /* number of regions */
 245         int                     space_used;     /* aggregate size of regions */
 246         struct list_head        busy_extents;   /* busy extents in chkpt */
 247         struct xfs_log_vec      *lv_chain;      /* logvecs being pushed */
 248         struct list_head        iclog_entry;
 249         struct list_head        committing;     /* ctx committing list */
 250         struct work_struct      discard_endio_work;
 251 };
 252 
 253 /*
 254  * Committed Item List structure
 255  *
 256  * This structure is used to track log items that have been committed but not
 257  * yet written into the log. It is used only when the delayed logging mount
 258  * option is enabled.
 259  *
 260  * This structure tracks the list of committing checkpoint contexts so
 261  * we can avoid the problem of having to hold out new transactions during a
 262  * flush until we have a the commit record LSN of the checkpoint. We can
 263  * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
 264  * sequence match and extract the commit LSN directly from there. If the
 265  * checkpoint is still in the process of committing, we can block waiting for
 266  * the commit LSN to be determined as well. This should make synchronous
 267  * operations almost as efficient as the old logging methods.
 268  */
 269 struct xfs_cil {
 270         struct xlog             *xc_log;
 271         struct list_head        xc_cil;
 272         spinlock_t              xc_cil_lock;
 273 
 274         struct rw_semaphore     xc_ctx_lock ____cacheline_aligned_in_smp;
 275         struct xfs_cil_ctx      *xc_ctx;
 276 
 277         spinlock_t              xc_push_lock ____cacheline_aligned_in_smp;
 278         xfs_lsn_t               xc_push_seq;
 279         struct list_head        xc_committing;
 280         wait_queue_head_t       xc_commit_wait;
 281         xfs_lsn_t               xc_current_sequence;
 282         struct work_struct      xc_push_work;
 283 } ____cacheline_aligned_in_smp;
 284 
 285 /*
 286  * The amount of log space we allow the CIL to aggregate is difficult to size.
 287  * Whatever we choose, we have to make sure we can get a reservation for the
 288  * log space effectively, that it is large enough to capture sufficient
 289  * relogging to reduce log buffer IO significantly, but it is not too large for
 290  * the log or induces too much latency when writing out through the iclogs. We
 291  * track both space consumed and the number of vectors in the checkpoint
 292  * context, so we need to decide which to use for limiting.
 293  *
 294  * Every log buffer we write out during a push needs a header reserved, which
 295  * is at least one sector and more for v2 logs. Hence we need a reservation of
 296  * at least 512 bytes per 32k of log space just for the LR headers. That means
 297  * 16KB of reservation per megabyte of delayed logging space we will consume,
 298  * plus various headers.  The number of headers will vary based on the num of
 299  * io vectors, so limiting on a specific number of vectors is going to result
 300  * in transactions of varying size. IOWs, it is more consistent to track and
 301  * limit space consumed in the log rather than by the number of objects being
 302  * logged in order to prevent checkpoint ticket overruns.
 303  *
 304  * Further, use of static reservations through the log grant mechanism is
 305  * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
 306  * grant) and a significant deadlock potential because regranting write space
 307  * can block on log pushes. Hence if we have to regrant log space during a log
 308  * push, we can deadlock.
 309  *
 310  * However, we can avoid this by use of a dynamic "reservation stealing"
 311  * technique during transaction commit whereby unused reservation space in the
 312  * transaction ticket is transferred to the CIL ctx commit ticket to cover the
 313  * space needed by the checkpoint transaction. This means that we never need to
 314  * specifically reserve space for the CIL checkpoint transaction, nor do we
 315  * need to regrant space once the checkpoint completes. This also means the
 316  * checkpoint transaction ticket is specific to the checkpoint context, rather
 317  * than the CIL itself.
 318  *
 319  * With dynamic reservations, we can effectively make up arbitrary limits for
 320  * the checkpoint size so long as they don't violate any other size rules.
 321  * Recovery imposes a rule that no transaction exceed half the log, so we are
 322  * limited by that.  Furthermore, the log transaction reservation subsystem
 323  * tries to keep 25% of the log free, so we need to keep below that limit or we
 324  * risk running out of free log space to start any new transactions.
 325  *
 326  * In order to keep background CIL push efficient, we will set a lower
 327  * threshold at which background pushing is attempted without blocking current
 328  * transaction commits.  A separate, higher bound defines when CIL pushes are
 329  * enforced to ensure we stay within our maximum checkpoint size bounds.
 330  * threshold, yet give us plenty of space for aggregation on large logs.
 331  */
 332 #define XLOG_CIL_SPACE_LIMIT(log)       (log->l_logsize >> 3)
 333 
 334 /*
 335  * ticket grant locks, queues and accounting have their own cachlines
 336  * as these are quite hot and can be operated on concurrently.
 337  */
 338 struct xlog_grant_head {
 339         spinlock_t              lock ____cacheline_aligned_in_smp;
 340         struct list_head        waiters;
 341         atomic64_t              grant;
 342 };
 343 
 344 /*
 345  * The reservation head lsn is not made up of a cycle number and block number.
 346  * Instead, it uses a cycle number and byte number.  Logs don't expect to
 347  * overflow 31 bits worth of byte offset, so using a byte number will mean
 348  * that round off problems won't occur when releasing partial reservations.
 349  */
 350 struct xlog {
 351         /* The following fields don't need locking */
 352         struct xfs_mount        *l_mp;          /* mount point */
 353         struct xfs_ail          *l_ailp;        /* AIL log is working with */
 354         struct xfs_cil          *l_cilp;        /* CIL log is working with */
 355         struct xfs_buftarg      *l_targ;        /* buftarg of log */
 356         struct workqueue_struct *l_ioend_workqueue; /* for I/O completions */
 357         struct delayed_work     l_work;         /* background flush work */
 358         uint                    l_flags;
 359         uint                    l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
 360         struct list_head        *l_buf_cancel_table;
 361         int                     l_iclog_hsize;  /* size of iclog header */
 362         int                     l_iclog_heads;  /* # of iclog header sectors */
 363         uint                    l_sectBBsize;   /* sector size in BBs (2^n) */
 364         int                     l_iclog_size;   /* size of log in bytes */
 365         int                     l_iclog_bufs;   /* number of iclog buffers */
 366         xfs_daddr_t             l_logBBstart;   /* start block of log */
 367         int                     l_logsize;      /* size of log in bytes */
 368         int                     l_logBBsize;    /* size of log in BB chunks */
 369 
 370         /* The following block of fields are changed while holding icloglock */
 371         wait_queue_head_t       l_flush_wait ____cacheline_aligned_in_smp;
 372                                                 /* waiting for iclog flush */
 373         int                     l_covered_state;/* state of "covering disk
 374                                                  * log entries" */
 375         xlog_in_core_t          *l_iclog;       /* head log queue       */
 376         spinlock_t              l_icloglock;    /* grab to change iclog state */
 377         int                     l_curr_cycle;   /* Cycle number of log writes */
 378         int                     l_prev_cycle;   /* Cycle number before last
 379                                                  * block increment */
 380         int                     l_curr_block;   /* current logical log block */
 381         int                     l_prev_block;   /* previous logical log block */
 382 
 383         /*
 384          * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
 385          * read without needing to hold specific locks. To avoid operations
 386          * contending with other hot objects, place each of them on a separate
 387          * cacheline.
 388          */
 389         /* lsn of last LR on disk */
 390         atomic64_t              l_last_sync_lsn ____cacheline_aligned_in_smp;
 391         /* lsn of 1st LR with unflushed * buffers */
 392         atomic64_t              l_tail_lsn ____cacheline_aligned_in_smp;
 393 
 394         struct xlog_grant_head  l_reserve_head;
 395         struct xlog_grant_head  l_write_head;
 396 
 397         struct xfs_kobj         l_kobj;
 398 
 399         /* The following field are used for debugging; need to hold icloglock */
 400 #ifdef DEBUG
 401         void                    *l_iclog_bak[XLOG_MAX_ICLOGS];
 402         /* log record crc error injection factor */
 403         uint32_t                l_badcrc_factor;
 404 #endif
 405         /* log recovery lsn tracking (for buffer submission */
 406         xfs_lsn_t               l_recovery_lsn;
 407 };
 408 
 409 #define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
 410         ((log)->l_buf_cancel_table + ((uint64_t)blkno % XLOG_BC_TABLE_SIZE))
 411 
 412 #define XLOG_FORCED_SHUTDOWN(log)       ((log)->l_flags & XLOG_IO_ERROR)
 413 
 414 /* common routines */
 415 extern int
 416 xlog_recover(
 417         struct xlog             *log);
 418 extern int
 419 xlog_recover_finish(
 420         struct xlog             *log);
 421 extern void
 422 xlog_recover_cancel(struct xlog *);
 423 
 424 extern __le32    xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
 425                             char *dp, int size);
 426 
 427 extern kmem_zone_t *xfs_log_ticket_zone;
 428 struct xlog_ticket *
 429 xlog_ticket_alloc(
 430         struct xlog     *log,
 431         int             unit_bytes,
 432         int             count,
 433         char            client,
 434         bool            permanent,
 435         xfs_km_flags_t  alloc_flags);
 436 
 437 
 438 static inline void
 439 xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
 440 {
 441         *ptr += bytes;
 442         *len -= bytes;
 443         *off += bytes;
 444 }
 445 
 446 void    xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
 447 void    xlog_print_trans(struct xfs_trans *);
 448 int
 449 xlog_write(
 450         struct xlog             *log,
 451         struct xfs_log_vec      *log_vector,
 452         struct xlog_ticket      *tic,
 453         xfs_lsn_t               *start_lsn,
 454         struct xlog_in_core     **commit_iclog,
 455         uint                    flags);
 456 
 457 /*
 458  * When we crack an atomic LSN, we sample it first so that the value will not
 459  * change while we are cracking it into the component values. This means we
 460  * will always get consistent component values to work from. This should always
 461  * be used to sample and crack LSNs that are stored and updated in atomic
 462  * variables.
 463  */
 464 static inline void
 465 xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
 466 {
 467         xfs_lsn_t val = atomic64_read(lsn);
 468 
 469         *cycle = CYCLE_LSN(val);
 470         *block = BLOCK_LSN(val);
 471 }
 472 
 473 /*
 474  * Calculate and assign a value to an atomic LSN variable from component pieces.
 475  */
 476 static inline void
 477 xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
 478 {
 479         atomic64_set(lsn, xlog_assign_lsn(cycle, block));
 480 }
 481 
 482 /*
 483  * When we crack the grant head, we sample it first so that the value will not
 484  * change while we are cracking it into the component values. This means we
 485  * will always get consistent component values to work from.
 486  */
 487 static inline void
 488 xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
 489 {
 490         *cycle = val >> 32;
 491         *space = val & 0xffffffff;
 492 }
 493 
 494 static inline void
 495 xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
 496 {
 497         xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
 498 }
 499 
 500 static inline int64_t
 501 xlog_assign_grant_head_val(int cycle, int space)
 502 {
 503         return ((int64_t)cycle << 32) | space;
 504 }
 505 
 506 static inline void
 507 xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
 508 {
 509         atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
 510 }
 511 
 512 /*
 513  * Committed Item List interfaces
 514  */
 515 int     xlog_cil_init(struct xlog *log);
 516 void    xlog_cil_init_post_recovery(struct xlog *log);
 517 void    xlog_cil_destroy(struct xlog *log);
 518 bool    xlog_cil_empty(struct xlog *log);
 519 
 520 /*
 521  * CIL force routines
 522  */
 523 xfs_lsn_t
 524 xlog_cil_force_lsn(
 525         struct xlog *log,
 526         xfs_lsn_t sequence);
 527 
 528 static inline void
 529 xlog_cil_force(struct xlog *log)
 530 {
 531         xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
 532 }
 533 
 534 /*
 535  * Unmount record type is used as a pseudo transaction type for the ticket.
 536  * It's value must be outside the range of XFS_TRANS_* values.
 537  */
 538 #define XLOG_UNMOUNT_REC_TYPE   (-1U)
 539 
 540 /*
 541  * Wrapper function for waiting on a wait queue serialised against wakeups
 542  * by a spinlock. This matches the semantics of all the wait queues used in the
 543  * log code.
 544  */
 545 static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
 546 {
 547         DECLARE_WAITQUEUE(wait, current);
 548 
 549         add_wait_queue_exclusive(wq, &wait);
 550         __set_current_state(TASK_UNINTERRUPTIBLE);
 551         spin_unlock(lock);
 552         schedule();
 553         remove_wait_queue(wq, &wait);
 554 }
 555 
 556 /*
 557  * The LSN is valid so long as it is behind the current LSN. If it isn't, this
 558  * means that the next log record that includes this metadata could have a
 559  * smaller LSN. In turn, this means that the modification in the log would not
 560  * replay.
 561  */
 562 static inline bool
 563 xlog_valid_lsn(
 564         struct xlog     *log,
 565         xfs_lsn_t       lsn)
 566 {
 567         int             cur_cycle;
 568         int             cur_block;
 569         bool            valid = true;
 570 
 571         /*
 572          * First, sample the current lsn without locking to avoid added
 573          * contention from metadata I/O. The current cycle and block are updated
 574          * (in xlog_state_switch_iclogs()) and read here in a particular order
 575          * to avoid false negatives (e.g., thinking the metadata LSN is valid
 576          * when it is not).
 577          *
 578          * The current block is always rewound before the cycle is bumped in
 579          * xlog_state_switch_iclogs() to ensure the current LSN is never seen in
 580          * a transiently forward state. Instead, we can see the LSN in a
 581          * transiently behind state if we happen to race with a cycle wrap.
 582          */
 583         cur_cycle = READ_ONCE(log->l_curr_cycle);
 584         smp_rmb();
 585         cur_block = READ_ONCE(log->l_curr_block);
 586 
 587         if ((CYCLE_LSN(lsn) > cur_cycle) ||
 588             (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block)) {
 589                 /*
 590                  * If the metadata LSN appears invalid, it's possible the check
 591                  * above raced with a wrap to the next log cycle. Grab the lock
 592                  * to check for sure.
 593                  */
 594                 spin_lock(&log->l_icloglock);
 595                 cur_cycle = log->l_curr_cycle;
 596                 cur_block = log->l_curr_block;
 597                 spin_unlock(&log->l_icloglock);
 598 
 599                 if ((CYCLE_LSN(lsn) > cur_cycle) ||
 600                     (CYCLE_LSN(lsn) == cur_cycle && BLOCK_LSN(lsn) > cur_block))
 601                         valid = false;
 602         }
 603 
 604         return valid;
 605 }
 606 
 607 #endif  /* __XFS_LOG_PRIV_H__ */

/* [<][>][^][v][top][bottom][index][help] */