linux/fs/xfs/xfs_log_priv.h
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#ifndef __XFS_LOG_PRIV_H__
  19#define __XFS_LOG_PRIV_H__
  20
  21struct xfs_buf;
  22struct xlog;
  23struct xlog_ticket;
  24struct xfs_mount;
  25
  26/*
  27 * Macros, structures, prototypes for internal log manager use.
  28 */
  29
  30#define XLOG_MIN_ICLOGS         2
  31#define XLOG_MAX_ICLOGS         8
  32#define XLOG_HEADER_MAGIC_NUM   0xFEEDbabe      /* Invalid cycle number */
  33#define XLOG_VERSION_1          1
  34#define XLOG_VERSION_2          2               /* Large IClogs, Log sunit */
  35#define XLOG_VERSION_OKBITS     (XLOG_VERSION_1 | XLOG_VERSION_2)
  36#define XLOG_MIN_RECORD_BSIZE   (16*1024)       /* eventually 32k */
  37#define XLOG_BIG_RECORD_BSIZE   (32*1024)       /* 32k buffers */
  38#define XLOG_MAX_RECORD_BSIZE   (256*1024)
  39#define XLOG_HEADER_CYCLE_SIZE  (32*1024)       /* cycle data in header */
  40#define XLOG_MIN_RECORD_BSHIFT  14              /* 16384 == 1 << 14 */
  41#define XLOG_BIG_RECORD_BSHIFT  15              /* 32k == 1 << 15 */
  42#define XLOG_MAX_RECORD_BSHIFT  18              /* 256k == 1 << 18 */
  43#define XLOG_BTOLSUNIT(log, b)  (((b)+(log)->l_mp->m_sb.sb_logsunit-1) / \
  44                                 (log)->l_mp->m_sb.sb_logsunit)
  45#define XLOG_LSUNITTOB(log, su) ((su) * (log)->l_mp->m_sb.sb_logsunit)
  46
  47#define XLOG_HEADER_SIZE        512
  48
  49#define XLOG_REC_SHIFT(log) \
  50        BTOBB(1 << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
  51         XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
  52#define XLOG_TOTAL_REC_SHIFT(log) \
  53        BTOBB(XLOG_MAX_ICLOGS << (xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? \
  54         XLOG_MAX_RECORD_BSHIFT : XLOG_BIG_RECORD_BSHIFT))
  55
  56static inline xfs_lsn_t xlog_assign_lsn(uint cycle, uint block)
  57{
  58        return ((xfs_lsn_t)cycle << 32) | block;
  59}
  60
  61static inline uint xlog_get_cycle(char *ptr)
  62{
  63        if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
  64                return be32_to_cpu(*((__be32 *)ptr + 1));
  65        else
  66                return be32_to_cpu(*(__be32 *)ptr);
  67}
  68
  69#define BLK_AVG(blk1, blk2)     ((blk1+blk2) >> 1)
  70
  71#ifdef __KERNEL__
  72
  73/*
  74 * get client id from packed copy.
  75 *
  76 * this hack is here because the xlog_pack code copies four bytes
  77 * of xlog_op_header containing the fields oh_clientid, oh_flags
  78 * and oh_res2 into the packed copy.
  79 *
  80 * later on this four byte chunk is treated as an int and the
  81 * client id is pulled out.
  82 *
  83 * this has endian issues, of course.
  84 */
  85static inline uint xlog_get_client_id(__be32 i)
  86{
  87        return be32_to_cpu(i) >> 24;
  88}
  89
  90/*
  91 * In core log state
  92 */
  93#define XLOG_STATE_ACTIVE    0x0001 /* Current IC log being written to */
  94#define XLOG_STATE_WANT_SYNC 0x0002 /* Want to sync this iclog; no more writes */
  95#define XLOG_STATE_SYNCING   0x0004 /* This IC log is syncing */
  96#define XLOG_STATE_DONE_SYNC 0x0008 /* Done syncing to disk */
  97#define XLOG_STATE_DO_CALLBACK \
  98                             0x0010 /* Process callback functions */
  99#define XLOG_STATE_CALLBACK  0x0020 /* Callback functions now */
 100#define XLOG_STATE_DIRTY     0x0040 /* Dirty IC log, not ready for ACTIVE status*/
 101#define XLOG_STATE_IOERROR   0x0080 /* IO error happened in sync'ing log */
 102#define XLOG_STATE_ALL       0x7FFF /* All possible valid flags */
 103#define XLOG_STATE_NOTUSED   0x8000 /* This IC log not being used */
 104#endif  /* __KERNEL__ */
 105
 106/*
 107 * Flags to log operation header
 108 *
 109 * The first write of a new transaction will be preceded with a start
 110 * record, XLOG_START_TRANS.  Once a transaction is committed, a commit
 111 * record is written, XLOG_COMMIT_TRANS.  If a single region can not fit into
 112 * the remainder of the current active in-core log, it is split up into
 113 * multiple regions.  Each partial region will be marked with a
 114 * XLOG_CONTINUE_TRANS until the last one, which gets marked with XLOG_END_TRANS.
 115 *
 116 */
 117#define XLOG_START_TRANS        0x01    /* Start a new transaction */
 118#define XLOG_COMMIT_TRANS       0x02    /* Commit this transaction */
 119#define XLOG_CONTINUE_TRANS     0x04    /* Cont this trans into new region */
 120#define XLOG_WAS_CONT_TRANS     0x08    /* Cont this trans into new region */
 121#define XLOG_END_TRANS          0x10    /* End a continued transaction */
 122#define XLOG_UNMOUNT_TRANS      0x20    /* Unmount a filesystem transaction */
 123
 124#ifdef __KERNEL__
 125/*
 126 * Flags to log ticket
 127 */
 128#define XLOG_TIC_INITED         0x1     /* has been initialized */
 129#define XLOG_TIC_PERM_RESERV    0x2     /* permanent reservation */
 130
 131#define XLOG_TIC_FLAGS \
 132        { XLOG_TIC_INITED,      "XLOG_TIC_INITED" }, \
 133        { XLOG_TIC_PERM_RESERV, "XLOG_TIC_PERM_RESERV" }
 134
 135#endif  /* __KERNEL__ */
 136
 137#define XLOG_UNMOUNT_TYPE       0x556e  /* Un for Unmount */
 138
 139/*
 140 * Flags for log structure
 141 */
 142#define XLOG_ACTIVE_RECOVERY    0x2     /* in the middle of recovery */
 143#define XLOG_RECOVERY_NEEDED    0x4     /* log was recovered */
 144#define XLOG_IO_ERROR           0x8     /* log hit an I/O error, and being
 145                                           shutdown */
 146#define XLOG_TAIL_WARN          0x10    /* log tail verify warning issued */
 147
 148typedef __uint32_t xlog_tid_t;
 149
 150#ifdef __KERNEL__
 151/*
 152 * Below are states for covering allocation transactions.
 153 * By covering, we mean changing the h_tail_lsn in the last on-disk
 154 * log write such that no allocation transactions will be re-done during
 155 * recovery after a system crash. Recovery starts at the last on-disk
 156 * log write.
 157 *
 158 * These states are used to insert dummy log entries to cover
 159 * space allocation transactions which can undo non-transactional changes
 160 * after a crash. Writes to a file with space
 161 * already allocated do not result in any transactions. Allocations
 162 * might include space beyond the EOF. So if we just push the EOF a
 163 * little, the last transaction for the file could contain the wrong
 164 * size. If there is no file system activity, after an allocation
 165 * transaction, and the system crashes, the allocation transaction
 166 * will get replayed and the file will be truncated. This could
 167 * be hours/days/... after the allocation occurred.
 168 *
 169 * The fix for this is to do two dummy transactions when the
 170 * system is idle. We need two dummy transaction because the h_tail_lsn
 171 * in the log record header needs to point beyond the last possible
 172 * non-dummy transaction. The first dummy changes the h_tail_lsn to
 173 * the first transaction before the dummy. The second dummy causes
 174 * h_tail_lsn to point to the first dummy. Recovery starts at h_tail_lsn.
 175 *
 176 * These dummy transactions get committed when everything
 177 * is idle (after there has been some activity).
 178 *
 179 * There are 5 states used to control this.
 180 *
 181 *  IDLE -- no logging has been done on the file system or
 182 *              we are done covering previous transactions.
 183 *  NEED -- logging has occurred and we need a dummy transaction
 184 *              when the log becomes idle.
 185 *  DONE -- we were in the NEED state and have committed a dummy
 186 *              transaction.
 187 *  NEED2 -- we detected that a dummy transaction has gone to the
 188 *              on disk log with no other transactions.
 189 *  DONE2 -- we committed a dummy transaction when in the NEED2 state.
 190 *
 191 * There are two places where we switch states:
 192 *
 193 * 1.) In xfs_sync, when we detect an idle log and are in NEED or NEED2.
 194 *      We commit the dummy transaction and switch to DONE or DONE2,
 195 *      respectively. In all other states, we don't do anything.
 196 *
 197 * 2.) When we finish writing the on-disk log (xlog_state_clean_log).
 198 *
 199 *      No matter what state we are in, if this isn't the dummy
 200 *      transaction going out, the next state is NEED.
 201 *      So, if we aren't in the DONE or DONE2 states, the next state
 202 *      is NEED. We can't be finishing a write of the dummy record
 203 *      unless it was committed and the state switched to DONE or DONE2.
 204 *
 205 *      If we are in the DONE state and this was a write of the
 206 *              dummy transaction, we move to NEED2.
 207 *
 208 *      If we are in the DONE2 state and this was a write of the
 209 *              dummy transaction, we move to IDLE.
 210 *
 211 *
 212 * Writing only one dummy transaction can get appended to
 213 * one file space allocation. When this happens, the log recovery
 214 * code replays the space allocation and a file could be truncated.
 215 * This is why we have the NEED2 and DONE2 states before going idle.
 216 */
 217
 218#define XLOG_STATE_COVER_IDLE   0
 219#define XLOG_STATE_COVER_NEED   1
 220#define XLOG_STATE_COVER_DONE   2
 221#define XLOG_STATE_COVER_NEED2  3
 222#define XLOG_STATE_COVER_DONE2  4
 223
 224#define XLOG_COVER_OPS          5
 225
 226
 227/* Ticket reservation region accounting */ 
 228#define XLOG_TIC_LEN_MAX        15
 229
 230/*
 231 * Reservation region
 232 * As would be stored in xfs_log_iovec but without the i_addr which
 233 * we don't care about.
 234 */
 235typedef struct xlog_res {
 236        uint    r_len;  /* region length                :4 */
 237        uint    r_type; /* region's transaction type    :4 */
 238} xlog_res_t;
 239
 240typedef struct xlog_ticket {
 241        struct list_head   t_queue;      /* reserve/write queue */
 242        struct task_struct *t_task;      /* task that owns this ticket */
 243        xlog_tid_t         t_tid;        /* transaction identifier       : 4  */
 244        atomic_t           t_ref;        /* ticket reference count       : 4  */
 245        int                t_curr_res;   /* current reservation in bytes : 4  */
 246        int                t_unit_res;   /* unit reservation in bytes    : 4  */
 247        char               t_ocnt;       /* original count               : 1  */
 248        char               t_cnt;        /* current count                : 1  */
 249        char               t_clientid;   /* who does this belong to;     : 1  */
 250        char               t_flags;      /* properties of reservation    : 1  */
 251        uint               t_trans_type; /* transaction type             : 4  */
 252
 253        /* reservation array fields */
 254        uint               t_res_num;                    /* num in array : 4 */
 255        uint               t_res_num_ophdrs;             /* num op hdrs  : 4 */
 256        uint               t_res_arr_sum;                /* array sum    : 4 */
 257        uint               t_res_o_flow;                 /* sum overflow : 4 */
 258        xlog_res_t         t_res_arr[XLOG_TIC_LEN_MAX];  /* array of res : 8 * 15 */ 
 259} xlog_ticket_t;
 260
 261#endif
 262
 263
 264typedef struct xlog_op_header {
 265        __be32     oh_tid;      /* transaction id of operation  :  4 b */
 266        __be32     oh_len;      /* bytes in data region         :  4 b */
 267        __u8       oh_clientid; /* who sent me this             :  1 b */
 268        __u8       oh_flags;    /*                              :  1 b */
 269        __u16      oh_res2;     /* 32 bit align                 :  2 b */
 270} xlog_op_header_t;
 271
 272
 273/* valid values for h_fmt */
 274#define XLOG_FMT_UNKNOWN  0
 275#define XLOG_FMT_LINUX_LE 1
 276#define XLOG_FMT_LINUX_BE 2
 277#define XLOG_FMT_IRIX_BE  3
 278
 279/* our fmt */
 280#ifdef XFS_NATIVE_HOST
 281#define XLOG_FMT XLOG_FMT_LINUX_BE
 282#else
 283#define XLOG_FMT XLOG_FMT_LINUX_LE
 284#endif
 285
 286typedef struct xlog_rec_header {
 287        __be32    h_magicno;    /* log record (LR) identifier           :  4 */
 288        __be32    h_cycle;      /* write cycle of log                   :  4 */
 289        __be32    h_version;    /* LR version                           :  4 */
 290        __be32    h_len;        /* len in bytes; should be 64-bit aligned: 4 */
 291        __be64    h_lsn;        /* lsn of this LR                       :  8 */
 292        __be64    h_tail_lsn;   /* lsn of 1st LR w/ buffers not committed: 8 */
 293        __le32    h_crc;        /* crc of log record                    :  4 */
 294        __be32    h_prev_block; /* block number to previous LR          :  4 */
 295        __be32    h_num_logops; /* number of log operations in this LR  :  4 */
 296        __be32    h_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE];
 297        /* new fields */
 298        __be32    h_fmt;        /* format of log record                 :  4 */
 299        uuid_t    h_fs_uuid;    /* uuid of FS                           : 16 */
 300        __be32    h_size;       /* iclog size                           :  4 */
 301} xlog_rec_header_t;
 302
 303typedef struct xlog_rec_ext_header {
 304        __be32    xh_cycle;     /* write cycle of log                   : 4 */
 305        __be32    xh_cycle_data[XLOG_HEADER_CYCLE_SIZE / BBSIZE]; /*    : 256 */
 306} xlog_rec_ext_header_t;
 307
 308#ifdef __KERNEL__
 309
 310/*
 311 * Quite misnamed, because this union lays out the actual on-disk log buffer.
 312 */
 313typedef union xlog_in_core2 {
 314        xlog_rec_header_t       hic_header;
 315        xlog_rec_ext_header_t   hic_xheader;
 316        char                    hic_sector[XLOG_HEADER_SIZE];
 317} xlog_in_core_2_t;
 318
 319/*
 320 * - A log record header is 512 bytes.  There is plenty of room to grow the
 321 *      xlog_rec_header_t into the reserved space.
 322 * - ic_data follows, so a write to disk can start at the beginning of
 323 *      the iclog.
 324 * - ic_forcewait is used to implement synchronous forcing of the iclog to disk.
 325 * - ic_next is the pointer to the next iclog in the ring.
 326 * - ic_bp is a pointer to the buffer used to write this incore log to disk.
 327 * - ic_log is a pointer back to the global log structure.
 328 * - ic_callback is a linked list of callback function/argument pairs to be
 329 *      called after an iclog finishes writing.
 330 * - ic_size is the full size of the header plus data.
 331 * - ic_offset is the current number of bytes written to in this iclog.
 332 * - ic_refcnt is bumped when someone is writing to the log.
 333 * - ic_state is the state of the iclog.
 334 *
 335 * Because of cacheline contention on large machines, we need to separate
 336 * various resources onto different cachelines. To start with, make the
 337 * structure cacheline aligned. The following fields can be contended on
 338 * by independent processes:
 339 *
 340 *      - ic_callback_*
 341 *      - ic_refcnt
 342 *      - fields protected by the global l_icloglock
 343 *
 344 * so we need to ensure that these fields are located in separate cachelines.
 345 * We'll put all the read-only and l_icloglock fields in the first cacheline,
 346 * and move everything else out to subsequent cachelines.
 347 */
 348typedef struct xlog_in_core {
 349        wait_queue_head_t       ic_force_wait;
 350        wait_queue_head_t       ic_write_wait;
 351        struct xlog_in_core     *ic_next;
 352        struct xlog_in_core     *ic_prev;
 353        struct xfs_buf          *ic_bp;
 354        struct xlog             *ic_log;
 355        int                     ic_size;
 356        int                     ic_offset;
 357        int                     ic_bwritecnt;
 358        unsigned short          ic_state;
 359        char                    *ic_datap;      /* pointer to iclog data */
 360
 361        /* Callback structures need their own cacheline */
 362        spinlock_t              ic_callback_lock ____cacheline_aligned_in_smp;
 363        xfs_log_callback_t      *ic_callback;
 364        xfs_log_callback_t      **ic_callback_tail;
 365
 366        /* reference counts need their own cacheline */
 367        atomic_t                ic_refcnt ____cacheline_aligned_in_smp;
 368        xlog_in_core_2_t        *ic_data;
 369#define ic_header       ic_data->hic_header
 370} xlog_in_core_t;
 371
 372/*
 373 * The CIL context is used to aggregate per-transaction details as well be
 374 * passed to the iclog for checkpoint post-commit processing.  After being
 375 * passed to the iclog, another context needs to be allocated for tracking the
 376 * next set of transactions to be aggregated into a checkpoint.
 377 */
 378struct xfs_cil;
 379
 380struct xfs_cil_ctx {
 381        struct xfs_cil          *cil;
 382        xfs_lsn_t               sequence;       /* chkpt sequence # */
 383        xfs_lsn_t               start_lsn;      /* first LSN of chkpt commit */
 384        xfs_lsn_t               commit_lsn;     /* chkpt commit record lsn */
 385        struct xlog_ticket      *ticket;        /* chkpt ticket */
 386        int                     nvecs;          /* number of regions */
 387        int                     space_used;     /* aggregate size of regions */
 388        struct list_head        busy_extents;   /* busy extents in chkpt */
 389        struct xfs_log_vec      *lv_chain;      /* logvecs being pushed */
 390        xfs_log_callback_t      log_cb;         /* completion callback hook. */
 391        struct list_head        committing;     /* ctx committing list */
 392};
 393
 394/*
 395 * Committed Item List structure
 396 *
 397 * This structure is used to track log items that have been committed but not
 398 * yet written into the log. It is used only when the delayed logging mount
 399 * option is enabled.
 400 *
 401 * This structure tracks the list of committing checkpoint contexts so
 402 * we can avoid the problem of having to hold out new transactions during a
 403 * flush until we have a the commit record LSN of the checkpoint. We can
 404 * traverse the list of committing contexts in xlog_cil_push_lsn() to find a
 405 * sequence match and extract the commit LSN directly from there. If the
 406 * checkpoint is still in the process of committing, we can block waiting for
 407 * the commit LSN to be determined as well. This should make synchronous
 408 * operations almost as efficient as the old logging methods.
 409 */
 410struct xfs_cil {
 411        struct xlog             *xc_log;
 412        struct list_head        xc_cil;
 413        spinlock_t              xc_cil_lock;
 414        struct xfs_cil_ctx      *xc_ctx;
 415        struct rw_semaphore     xc_ctx_lock;
 416        struct list_head        xc_committing;
 417        wait_queue_head_t       xc_commit_wait;
 418        xfs_lsn_t               xc_current_sequence;
 419        struct work_struct      xc_push_work;
 420        xfs_lsn_t               xc_push_seq;
 421};
 422
 423/*
 424 * The amount of log space we allow the CIL to aggregate is difficult to size.
 425 * Whatever we choose, we have to make sure we can get a reservation for the
 426 * log space effectively, that it is large enough to capture sufficient
 427 * relogging to reduce log buffer IO significantly, but it is not too large for
 428 * the log or induces too much latency when writing out through the iclogs. We
 429 * track both space consumed and the number of vectors in the checkpoint
 430 * context, so we need to decide which to use for limiting.
 431 *
 432 * Every log buffer we write out during a push needs a header reserved, which
 433 * is at least one sector and more for v2 logs. Hence we need a reservation of
 434 * at least 512 bytes per 32k of log space just for the LR headers. That means
 435 * 16KB of reservation per megabyte of delayed logging space we will consume,
 436 * plus various headers.  The number of headers will vary based on the num of
 437 * io vectors, so limiting on a specific number of vectors is going to result
 438 * in transactions of varying size. IOWs, it is more consistent to track and
 439 * limit space consumed in the log rather than by the number of objects being
 440 * logged in order to prevent checkpoint ticket overruns.
 441 *
 442 * Further, use of static reservations through the log grant mechanism is
 443 * problematic. It introduces a lot of complexity (e.g. reserve grant vs write
 444 * grant) and a significant deadlock potential because regranting write space
 445 * can block on log pushes. Hence if we have to regrant log space during a log
 446 * push, we can deadlock.
 447 *
 448 * However, we can avoid this by use of a dynamic "reservation stealing"
 449 * technique during transaction commit whereby unused reservation space in the
 450 * transaction ticket is transferred to the CIL ctx commit ticket to cover the
 451 * space needed by the checkpoint transaction. This means that we never need to
 452 * specifically reserve space for the CIL checkpoint transaction, nor do we
 453 * need to regrant space once the checkpoint completes. This also means the
 454 * checkpoint transaction ticket is specific to the checkpoint context, rather
 455 * than the CIL itself.
 456 *
 457 * With dynamic reservations, we can effectively make up arbitrary limits for
 458 * the checkpoint size so long as they don't violate any other size rules.
 459 * Recovery imposes a rule that no transaction exceed half the log, so we are
 460 * limited by that.  Furthermore, the log transaction reservation subsystem
 461 * tries to keep 25% of the log free, so we need to keep below that limit or we
 462 * risk running out of free log space to start any new transactions.
 463 *
 464 * In order to keep background CIL push efficient, we will set a lower
 465 * threshold at which background pushing is attempted without blocking current
 466 * transaction commits.  A separate, higher bound defines when CIL pushes are
 467 * enforced to ensure we stay within our maximum checkpoint size bounds.
 468 * threshold, yet give us plenty of space for aggregation on large logs.
 469 */
 470#define XLOG_CIL_SPACE_LIMIT(log)       (log->l_logsize >> 3)
 471#define XLOG_CIL_HARD_SPACE_LIMIT(log)  (3 * (log->l_logsize >> 4))
 472
 473/*
 474 * ticket grant locks, queues and accounting have their own cachlines
 475 * as these are quite hot and can be operated on concurrently.
 476 */
 477struct xlog_grant_head {
 478        spinlock_t              lock ____cacheline_aligned_in_smp;
 479        struct list_head        waiters;
 480        atomic64_t              grant;
 481};
 482
 483/*
 484 * The reservation head lsn is not made up of a cycle number and block number.
 485 * Instead, it uses a cycle number and byte number.  Logs don't expect to
 486 * overflow 31 bits worth of byte offset, so using a byte number will mean
 487 * that round off problems won't occur when releasing partial reservations.
 488 */
 489struct xlog {
 490        /* The following fields don't need locking */
 491        struct xfs_mount        *l_mp;          /* mount point */
 492        struct xfs_ail          *l_ailp;        /* AIL log is working with */
 493        struct xfs_cil          *l_cilp;        /* CIL log is working with */
 494        struct xfs_buf          *l_xbuf;        /* extra buffer for log
 495                                                 * wrapping */
 496        struct xfs_buftarg      *l_targ;        /* buftarg of log */
 497        struct delayed_work     l_work;         /* background flush work */
 498        uint                    l_flags;
 499        uint                    l_quotaoffs_flag; /* XFS_DQ_*, for QUOTAOFFs */
 500        struct list_head        *l_buf_cancel_table;
 501        int                     l_iclog_hsize;  /* size of iclog header */
 502        int                     l_iclog_heads;  /* # of iclog header sectors */
 503        uint                    l_sectBBsize;   /* sector size in BBs (2^n) */
 504        int                     l_iclog_size;   /* size of log in bytes */
 505        int                     l_iclog_size_log; /* log power size of log */
 506        int                     l_iclog_bufs;   /* number of iclog buffers */
 507        xfs_daddr_t             l_logBBstart;   /* start block of log */
 508        int                     l_logsize;      /* size of log in bytes */
 509        int                     l_logBBsize;    /* size of log in BB chunks */
 510
 511        /* The following block of fields are changed while holding icloglock */
 512        wait_queue_head_t       l_flush_wait ____cacheline_aligned_in_smp;
 513                                                /* waiting for iclog flush */
 514        int                     l_covered_state;/* state of "covering disk
 515                                                 * log entries" */
 516        xlog_in_core_t          *l_iclog;       /* head log queue       */
 517        spinlock_t              l_icloglock;    /* grab to change iclog state */
 518        int                     l_curr_cycle;   /* Cycle number of log writes */
 519        int                     l_prev_cycle;   /* Cycle number before last
 520                                                 * block increment */
 521        int                     l_curr_block;   /* current logical log block */
 522        int                     l_prev_block;   /* previous logical log block */
 523
 524        /*
 525         * l_last_sync_lsn and l_tail_lsn are atomics so they can be set and
 526         * read without needing to hold specific locks. To avoid operations
 527         * contending with other hot objects, place each of them on a separate
 528         * cacheline.
 529         */
 530        /* lsn of last LR on disk */
 531        atomic64_t              l_last_sync_lsn ____cacheline_aligned_in_smp;
 532        /* lsn of 1st LR with unflushed * buffers */
 533        atomic64_t              l_tail_lsn ____cacheline_aligned_in_smp;
 534
 535        struct xlog_grant_head  l_reserve_head;
 536        struct xlog_grant_head  l_write_head;
 537
 538        /* The following field are used for debugging; need to hold icloglock */
 539#ifdef DEBUG
 540        char                    *l_iclog_bak[XLOG_MAX_ICLOGS];
 541#endif
 542
 543};
 544
 545#define XLOG_BUF_CANCEL_BUCKET(log, blkno) \
 546        ((log)->l_buf_cancel_table + ((__uint64_t)blkno % XLOG_BC_TABLE_SIZE))
 547
 548#define XLOG_FORCED_SHUTDOWN(log)       ((log)->l_flags & XLOG_IO_ERROR)
 549
 550/* common routines */
 551extern int
 552xlog_recover(
 553        struct xlog             *log);
 554extern int
 555xlog_recover_finish(
 556        struct xlog             *log);
 557
 558extern __le32    xlog_cksum(struct xlog *log, struct xlog_rec_header *rhead,
 559                            char *dp, int size);
 560
 561extern kmem_zone_t *xfs_log_ticket_zone;
 562struct xlog_ticket *
 563xlog_ticket_alloc(
 564        struct xlog     *log,
 565        int             unit_bytes,
 566        int             count,
 567        char            client,
 568        bool            permanent,
 569        xfs_km_flags_t  alloc_flags);
 570
 571
 572static inline void
 573xlog_write_adv_cnt(void **ptr, int *len, int *off, size_t bytes)
 574{
 575        *ptr += bytes;
 576        *len -= bytes;
 577        *off += bytes;
 578}
 579
 580void    xlog_print_tic_res(struct xfs_mount *mp, struct xlog_ticket *ticket);
 581int
 582xlog_write(
 583        struct xlog             *log,
 584        struct xfs_log_vec      *log_vector,
 585        struct xlog_ticket      *tic,
 586        xfs_lsn_t               *start_lsn,
 587        struct xlog_in_core     **commit_iclog,
 588        uint                    flags);
 589
 590/*
 591 * When we crack an atomic LSN, we sample it first so that the value will not
 592 * change while we are cracking it into the component values. This means we
 593 * will always get consistent component values to work from. This should always
 594 * be used to sample and crack LSNs that are stored and updated in atomic
 595 * variables.
 596 */
 597static inline void
 598xlog_crack_atomic_lsn(atomic64_t *lsn, uint *cycle, uint *block)
 599{
 600        xfs_lsn_t val = atomic64_read(lsn);
 601
 602        *cycle = CYCLE_LSN(val);
 603        *block = BLOCK_LSN(val);
 604}
 605
 606/*
 607 * Calculate and assign a value to an atomic LSN variable from component pieces.
 608 */
 609static inline void
 610xlog_assign_atomic_lsn(atomic64_t *lsn, uint cycle, uint block)
 611{
 612        atomic64_set(lsn, xlog_assign_lsn(cycle, block));
 613}
 614
 615/*
 616 * When we crack the grant head, we sample it first so that the value will not
 617 * change while we are cracking it into the component values. This means we
 618 * will always get consistent component values to work from.
 619 */
 620static inline void
 621xlog_crack_grant_head_val(int64_t val, int *cycle, int *space)
 622{
 623        *cycle = val >> 32;
 624        *space = val & 0xffffffff;
 625}
 626
 627static inline void
 628xlog_crack_grant_head(atomic64_t *head, int *cycle, int *space)
 629{
 630        xlog_crack_grant_head_val(atomic64_read(head), cycle, space);
 631}
 632
 633static inline int64_t
 634xlog_assign_grant_head_val(int cycle, int space)
 635{
 636        return ((int64_t)cycle << 32) | space;
 637}
 638
 639static inline void
 640xlog_assign_grant_head(atomic64_t *head, int cycle, int space)
 641{
 642        atomic64_set(head, xlog_assign_grant_head_val(cycle, space));
 643}
 644
 645/*
 646 * Committed Item List interfaces
 647 */
 648int
 649xlog_cil_init(struct xlog *log);
 650void
 651xlog_cil_init_post_recovery(struct xlog *log);
 652void
 653xlog_cil_destroy(struct xlog *log);
 654
 655/*
 656 * CIL force routines
 657 */
 658xfs_lsn_t
 659xlog_cil_force_lsn(
 660        struct xlog *log,
 661        xfs_lsn_t sequence);
 662
 663static inline void
 664xlog_cil_force(struct xlog *log)
 665{
 666        xlog_cil_force_lsn(log, log->l_cilp->xc_current_sequence);
 667}
 668
 669/*
 670 * Unmount record type is used as a pseudo transaction type for the ticket.
 671 * It's value must be outside the range of XFS_TRANS_* values.
 672 */
 673#define XLOG_UNMOUNT_REC_TYPE   (-1U)
 674
 675/*
 676 * Wrapper function for waiting on a wait queue serialised against wakeups
 677 * by a spinlock. This matches the semantics of all the wait queues used in the
 678 * log code.
 679 */
 680static inline void xlog_wait(wait_queue_head_t *wq, spinlock_t *lock)
 681{
 682        DECLARE_WAITQUEUE(wait, current);
 683
 684        add_wait_queue_exclusive(wq, &wait);
 685        __set_current_state(TASK_UNINTERRUPTIBLE);
 686        spin_unlock(lock);
 687        schedule();
 688        remove_wait_queue(wq, &wait);
 689}
 690#endif  /* __KERNEL__ */
 691
 692#endif  /* __XFS_LOG_PRIV_H__ */
 693
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.