include/sys/zil.h

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2012 by Delphix. All rights reserved.
  24  */
  25
  26 /* Portions Copyright 2010 Robert Milkowski */
  27
  28 #ifndef _SYS_ZIL_H
  29 #define _SYS_ZIL_H
  30
  31 #include <sys/types.h>
  32 #include <sys/spa.h>
  33 #include <sys/zio.h>
  34 #include <sys/dmu.h>
  35
  36 #ifdef  __cplusplus
  37 extern "C" {
  38 #endif
  39
  40 struct dsl_pool;
  41 struct dsl_dataset;
  42
  43 /*
  44  * Intent log format:
  45  *
  46  * Each objset has its own intent log.  The log header (zil_header_t)
  47  * for objset N's intent log is kept in the Nth object of the SPA's
  48  * intent_log objset.  The log header points to a chain of log blocks,
  49  * each of which contains log records (i.e., transactions) followed by
  50  * a log block trailer (zil_trailer_t).  The format of a log record
  51  * depends on the record (or transaction) type, but all records begin
  52  * with a common structure that defines the type, length, and txg.
  53  */
  54
  55 /*
  56  * Intent log header - this on disk structure holds fields to manage
  57  * the log.  All fields are 64 bit to easily handle cross architectures.
  58  */
  59 typedef struct zil_header {
  60         uint64_t zh_claim_txg;  /* txg in which log blocks were claimed */
  61         uint64_t zh_replay_seq; /* highest replayed sequence number */
  62         blkptr_t zh_log;        /* log chain */
  63         uint64_t zh_claim_blk_seq; /* highest claimed block sequence number */
  64         uint64_t zh_flags;      /* header flags */
  65         uint64_t zh_claim_lr_seq; /* highest claimed lr sequence number */
  66         uint64_t zh_pad[3];
  67 } zil_header_t;
  68
  69 /*
  70  * zh_flags bit settings
  71  */
  72 #define ZIL_REPLAY_NEEDED       0x1     /* replay needed - internal only */
  73 #define ZIL_CLAIM_LR_SEQ_VALID  0x2     /* zh_claim_lr_seq field is valid */
  74
  75 /*
  76  * Log block chaining.
  77  *
  78  * Log blocks are chained together. Originally they were chained at the
  79  * end of the block. For performance reasons the chain was moved to the
  80  * beginning of the block which allows writes for only the data being used.
  81  * The older position is supported for backwards compatability.
  82  *
  83  * The zio_eck_t contains a zec_cksum which for the intent log is
  84  * the sequence number of this log block. A seq of 0 is invalid.
  85  * The zec_cksum is checked by the SPA against the sequence
  86  * number passed in the blk_cksum field of the blkptr_t
  87  */
  88 typedef struct zil_chain {
  89         uint64_t zc_pad;
  90         blkptr_t zc_next_blk;   /* next block in chain */
  91         uint64_t zc_nused;      /* bytes in log block used */
  92         zio_eck_t zc_eck;       /* block trailer */
  93 } zil_chain_t;
  94
  95 #define ZIL_MIN_BLKSZ   4096ULL
  96
  97 /*
  98  * The words of a log block checksum.
  99  */
 100 #define ZIL_ZC_GUID_0   0
 101 #define ZIL_ZC_GUID_1   1
 102 #define ZIL_ZC_OBJSET   2
 103 #define ZIL_ZC_SEQ      3
 104
 105 typedef enum zil_create {
 106         Z_FILE,
 107         Z_DIR,
 108         Z_XATTRDIR,
 109 } zil_create_t;
 110
 111 /*
 112  * size of xvattr log section.
 113  * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
 114  * for create time and a single 64 bit integer for all of the attributes,
 115  * and 4 64 bit integers (32 bytes) for the scanstamp.
 116  *
 117  */
 118
 119 #define ZIL_XVAT_SIZE(mapsize) \
 120         sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
 121         (sizeof (uint64_t) * 7)
 122
 123 /*
 124  * Size of ACL in log.  The ACE data is padded out to properly align
 125  * on 8 byte boundary.
 126  */
 127
 128 #define ZIL_ACE_LENGTH(x)       (roundup(x, sizeof (uint64_t)))
 129
 130 /*
 131  * Intent log transaction types and record structures
 132  */
 133 #define TX_CREATE               1       /* Create file */
 134 #define TX_MKDIR                2       /* Make directory */
 135 #define TX_MKXATTR              3       /* Make XATTR directory */
 136 #define TX_SYMLINK              4       /* Create symbolic link to a file */
 137 #define TX_REMOVE               5       /* Remove file */
 138 #define TX_RMDIR                6       /* Remove directory */
 139 #define TX_LINK                 7       /* Create hard link to a file */
 140 #define TX_RENAME               8       /* Rename a file */
 141 #define TX_WRITE                9       /* File write */
 142 #define TX_TRUNCATE             10      /* Truncate a file */
 143 #define TX_SETATTR              11      /* Set file attributes */
 144 #define TX_ACL_V0               12      /* Set old formatted ACL */
 145 #define TX_ACL                  13      /* Set ACL */
 146 #define TX_CREATE_ACL           14      /* create with ACL */
 147 #define TX_CREATE_ATTR          15      /* create + attrs */
 148 #define TX_CREATE_ACL_ATTR      16      /* create with ACL + attrs */
 149 #define TX_MKDIR_ACL            17      /* mkdir with ACL */
 150 #define TX_MKDIR_ATTR           18      /* mkdir with attr */
 151 #define TX_MKDIR_ACL_ATTR       19      /* mkdir with ACL + attrs */
 152 #define TX_WRITE2               20      /* dmu_sync EALREADY write */
 153 #define TX_MAX_TYPE             21      /* Max transaction type */
 154
 155 /*
 156  * The transactions for mkdir, symlink, remove, rmdir, link, and rename
 157  * may have the following bit set, indicating the original request
 158  * specified case-insensitive handling of names.
 159  */
 160 #define TX_CI   ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
 161
 162 /*
 163  * Transactions for write, truncate, setattr, acl_v0, and acl can be logged
 164  * out of order.  For convenience in the code, all such records must have
 165  * lr_foid at the same offset.
 166  */
 167 #define TX_OOO(txtype)                  \
 168         ((txtype) == TX_WRITE ||        \
 169         (txtype) == TX_TRUNCATE ||      \
 170         (txtype) == TX_SETATTR ||       \
 171         (txtype) == TX_ACL_V0 ||        \
 172         (txtype) == TX_ACL ||           \
 173         (txtype) == TX_WRITE2)
 174
 175 /*
 176  * The number of dnode slots consumed by the object is stored in the 8
 177  * unused upper bits of the object ID. We subtract 1 from the value
 178  * stored on disk for compatibility with implementations that don't
 179  * support large dnodes. The slot count for a single-slot dnode will
 180  * contain 0 for those bits to preserve the log record format for
 181  * "small" dnodes.
 182  */
 183 #define LR_FOID_GET_SLOTS(oid) (BF64_GET((oid), 56, 8) + 1)
 184 #define LR_FOID_SET_SLOTS(oid, x) BF64_SET((oid), 56, 8, (x) - 1)
 185 #define LR_FOID_GET_OBJ(oid) BF64_GET((oid), 0, DN_MAX_OBJECT_SHIFT)
 186 #define LR_FOID_SET_OBJ(oid, x) BF64_SET((oid), 0, DN_MAX_OBJECT_SHIFT, (x))
 187
 188 /*
 189  * Format of log records.
 190  * The fields are carefully defined to allow them to be aligned
 191  * and sized the same on sparc & intel architectures.
 192  * Each log record has a common structure at the beginning.
 193  *
 194  * The log record on disk (lrc_seq) holds the sequence number of all log
 195  * records which is used to ensure we don't replay the same record.
 196  */
 197 typedef struct {                        /* common log record header */
 198         uint64_t        lrc_txtype;     /* intent log transaction type */
 199         uint64_t        lrc_reclen;     /* transaction record length */
 200         uint64_t        lrc_txg;        /* dmu transaction group number */
 201         uint64_t        lrc_seq;        /* see comment above */
 202 } lr_t;
 203
 204 /*
 205  * Common start of all out-of-order record types (TX_OOO() above).
 206  */
 207 typedef struct {
 208         lr_t            lr_common;      /* common portion of log record */
 209         uint64_t        lr_foid;        /* object id */
 210 } lr_ooo_t;
 211
 212 /*
 213  * Handle option extended vattr attributes.
 214  *
 215  * Whenever new attributes are added the version number
 216  * will need to be updated as will code in
 217  * zfs_log.c and zfs_replay.c
 218  */
 219 typedef struct {
 220         uint32_t        lr_attr_masksize; /* number of elements in array */
 221         uint32_t        lr_attr_bitmap; /* First entry of array */
 222         /* remainder of array and any additional fields */
 223 } lr_attr_t;
 224
 225 /*
 226  * log record for creates without optional ACL.
 227  * This log record does support optional xvattr_t attributes.
 228  */
 229 typedef struct {
 230         lr_t            lr_common;      /* common portion of log record */
 231         uint64_t        lr_doid;        /* object id of directory */
 232         uint64_t        lr_foid;        /* object id of created file object */
 233         uint64_t        lr_mode;        /* mode of object */
 234         uint64_t        lr_uid;         /* uid of object */
 235         uint64_t        lr_gid;         /* gid of object */
 236         uint64_t        lr_gen;         /* generation (txg of creation) */
 237         uint64_t        lr_crtime[2];   /* creation time */
 238         uint64_t        lr_rdev;        /* rdev of object to create */
 239         /* name of object to create follows this */
 240         /* for symlinks, link content follows name */
 241         /* for creates with xvattr data, the name follows the xvattr info */
 242 } lr_create_t;
 243
 244 /*
 245  * FUID ACL record will be an array of ACEs from the original ACL.
 246  * If this array includes ephemeral IDs, the record will also include
 247  * an array of log-specific FUIDs to replace the ephemeral IDs.
 248  * Only one copy of each unique domain will be present, so the log-specific
 249  * FUIDs will use an index into a compressed domain table.  On replay this
 250  * information will be used to construct real FUIDs (and bypass idmap,
 251  * since it may not be available).
 252  */
 253
 254 /*
 255  * Log record for creates with optional ACL
 256  * This log record is also used for recording any FUID
 257  * information needed for replaying the create.  If the
 258  * file doesn't have any actual ACEs then the lr_aclcnt
 259  * would be zero.
 260  *
 261  * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
 262  * If create is also setting xvattr's, then acl data follows xvattr.
 263  * If ACE FUIDs are needed then they will follow the xvattr_t.  Following
 264  * the FUIDs will be the domain table information.  The FUIDs for the owner
 265  * and group will be in lr_create.  Name follows ACL data.
 266  */
 267 typedef struct {
 268         lr_create_t     lr_create;      /* common create portion */
 269         uint64_t        lr_aclcnt;      /* number of ACEs in ACL */
 270         uint64_t        lr_domcnt;      /* number of unique domains */
 271         uint64_t        lr_fuidcnt;     /* number of real fuids */
 272         uint64_t        lr_acl_bytes;   /* number of bytes in ACL */
 273         uint64_t        lr_acl_flags;   /* ACL flags */
 274 } lr_acl_create_t;
 275
 276 typedef struct {
 277         lr_t            lr_common;      /* common portion of log record */
 278         uint64_t        lr_doid;        /* obj id of directory */
 279         /* name of object to remove follows this */
 280 } lr_remove_t;
 281
 282 typedef struct {
 283         lr_t            lr_common;      /* common portion of log record */
 284         uint64_t        lr_doid;        /* obj id of directory */
 285         uint64_t        lr_link_obj;    /* obj id of link */
 286         /* name of object to link follows this */
 287 } lr_link_t;
 288
 289 typedef struct {
 290         lr_t            lr_common;      /* common portion of log record */
 291         uint64_t        lr_sdoid;       /* obj id of source directory */
 292         uint64_t        lr_tdoid;       /* obj id of target directory */
 293         /* 2 strings: names of source and destination follow this */
 294 } lr_rename_t;
 295
 296 typedef struct {
 297         lr_t            lr_common;      /* common portion of log record */
 298         uint64_t        lr_foid;        /* file object to write */
 299         uint64_t        lr_offset;      /* offset to write to */
 300         uint64_t        lr_length;      /* user data length to write */
 301         uint64_t        lr_blkoff;      /* no longer used */
 302         blkptr_t        lr_blkptr;      /* spa block pointer for replay */
 303         /* write data will follow for small writes */
 304 } lr_write_t;
 305
 306 typedef struct {
 307         lr_t            lr_common;      /* common portion of log record */
 308         uint64_t        lr_foid;        /* object id of file to truncate */
 309         uint64_t        lr_offset;      /* offset to truncate from */
 310         uint64_t        lr_length;      /* length to truncate */
 311 } lr_truncate_t;
 312
 313 typedef struct {
 314         lr_t            lr_common;      /* common portion of log record */
 315         uint64_t        lr_foid;        /* file object to change attributes */
 316         uint64_t        lr_mask;        /* mask of attributes to set */
 317         uint64_t        lr_mode;        /* mode to set */
 318         uint64_t        lr_uid;         /* uid to set */
 319         uint64_t        lr_gid;         /* gid to set */
 320         uint64_t        lr_size;        /* size to set */
 321         uint64_t        lr_atime[2];    /* access time */
 322         uint64_t        lr_mtime[2];    /* modification time */
 323         /* optional attribute lr_attr_t may be here */
 324 } lr_setattr_t;
 325
 326 typedef struct {
 327         lr_t            lr_common;      /* common portion of log record */
 328         uint64_t        lr_foid;        /* obj id of file */
 329         uint64_t        lr_aclcnt;      /* number of acl entries */
 330         /* lr_aclcnt number of ace_t entries follow this */
 331 } lr_acl_v0_t;
 332
 333 typedef struct {
 334         lr_t            lr_common;      /* common portion of log record */
 335         uint64_t        lr_foid;        /* obj id of file */
 336         uint64_t        lr_aclcnt;      /* number of ACEs in ACL */
 337         uint64_t        lr_domcnt;      /* number of unique domains */
 338         uint64_t        lr_fuidcnt;     /* number of real fuids */
 339         uint64_t        lr_acl_bytes;   /* number of bytes in ACL */
 340         uint64_t        lr_acl_flags;   /* ACL flags */
 341         /* lr_acl_bytes number of variable sized ace's follows */
 342 } lr_acl_t;
 343
 344 /*
 345  * ZIL structure definitions, interface function prototype and globals.
 346  */
 347
 348 /*
 349  * Writes are handled in three different ways:
 350  *
 351  * WR_INDIRECT:
 352  *    In this mode, if we need to commit the write later, then the block
 353  *    is immediately written into the file system (using dmu_sync),
 354  *    and a pointer to the block is put into the log record.
 355  *    When the txg commits the block is linked in.
 356  *    This saves additionally writing the data into the log record.
 357  *    There are a few requirements for this to occur:
 358  *      - write is greater than zfs/zvol_immediate_write_sz
 359  *      - not using slogs (as slogs are assumed to always be faster
 360  *        than writing into the main pool)
 361  *      - the write occupies only one block
 362  * WR_COPIED:
 363  *    If we know we'll immediately be committing the
 364  *    transaction (FSYNC or FDSYNC), the we allocate a larger
 365  *    log record here for the data and copy the data in.
 366  * WR_NEED_COPY:
 367  *    Otherwise we don't allocate a buffer, and *if* we need to
 368  *    flush the write later then a buffer is allocated and
 369  *    we retrieve the data using the dmu.
 370  */
 371 typedef enum {
 372         WR_INDIRECT,    /* indirect - a large write (dmu_sync() data */
 373                         /* and put blkptr in log, rather than actual data) */
 374         WR_COPIED,      /* immediate - data is copied into lr_write_t */
 375         WR_NEED_COPY,   /* immediate - data needs to be copied if pushed */
 376         WR_NUM_STATES   /* number of states */
 377 } itx_wr_state_t;
 378
 379 typedef void (*zil_callback_t)(void *data);
 380
 381 typedef struct itx {
 382         list_node_t     itx_node;       /* linkage on zl_itx_list */
 383         void            *itx_private;   /* type-specific opaque data */
 384         itx_wr_state_t  itx_wr_state;   /* write state */
 385         uint8_t         itx_sync;       /* synchronous transaction */
 386         zil_callback_t  itx_callback;   /* Called when the itx is persistent */
 387         void            *itx_callback_data; /* User data for the callback */
 388         uint64_t        itx_sod;        /* record size on disk */
 389         uint64_t        itx_oid;        /* object id */
 390         lr_t            itx_lr;         /* common part of log record */
 391         /* followed by type-specific part of lr_xx_t and its immediate data */
 392 } itx_t;
 393
 394 /*
 395  * Used for zil kstat.
 396  */
 397 typedef struct zil_stats {
 398         /*
 399          * Number of times a ZIL commit (e.g. fsync) has been requested.
 400          */
 401         kstat_named_t zil_commit_count;
 402
 403         /*
 404          * Number of times the ZIL has been flushed to stable storage.
 405          * This is less than zil_commit_count when commits are "merged"
 406          * (see the documentation above zil_commit()).
 407          */
 408         kstat_named_t zil_commit_writer_count;
 409
 410         /*
 411          * Number of transactions (reads, writes, renames, etc.)
 412          * that have been commited.
 413          */
 414         kstat_named_t zil_itx_count;
 415
 416         /*
 417          * See the documentation for itx_wr_state_t above.
 418          * Note that "bytes" accumulates the length of the transactions
 419          * (i.e. data), not the actual log record sizes.
 420          */
 421         kstat_named_t zil_itx_indirect_count;
 422         kstat_named_t zil_itx_indirect_bytes;
 423         kstat_named_t zil_itx_copied_count;
 424         kstat_named_t zil_itx_copied_bytes;
 425         kstat_named_t zil_itx_needcopy_count;
 426         kstat_named_t zil_itx_needcopy_bytes;
 427
 428         /*
 429          * Transactions which have been allocated to the "normal"
 430          * (i.e. not slog) storage pool. Note that "bytes" accumulate
 431          * the actual log record sizes - which do not include the actual
 432          * data in case of indirect writes.
 433          */
 434         kstat_named_t zil_itx_metaslab_normal_count;
 435         kstat_named_t zil_itx_metaslab_normal_bytes;
 436
 437         /*
 438          * Transactions which have been allocated to the "slog" storage pool.
 439          * If there are no separate log devices, this is the same as the
 440          * "normal" pool.
 441          */
 442         kstat_named_t zil_itx_metaslab_slog_count;
 443         kstat_named_t zil_itx_metaslab_slog_bytes;
 444 } zil_stats_t;
 445
 446 extern zil_stats_t zil_stats;
 447
 448 #define ZIL_STAT_INCR(stat, val) \
 449     atomic_add_64(&zil_stats.stat.value.ui64, (val));
 450 #define ZIL_STAT_BUMP(stat) \
 451     ZIL_STAT_INCR(stat, 1);
 452
 453 typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
 454     uint64_t txg);
 455 typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
 456     uint64_t txg);
 457 typedef int (*const zil_replay_func_t)(void *, char *, boolean_t);
 458 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio);
 459
 460 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
 461     zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg);
 462
 463 extern void     zil_init(void);
 464 extern void     zil_fini(void);
 465
 466 extern zilog_t  *zil_alloc(objset_t *os, zil_header_t *zh_phys);
 467 extern void     zil_free(zilog_t *zilog);
 468
 469 extern zilog_t  *zil_open(objset_t *os, zil_get_data_t *get_data);
 470 extern void     zil_close(zilog_t *zilog);
 471
 472 extern void     zil_replay(objset_t *os, void *arg,
 473     zil_replay_func_t replay_func[TX_MAX_TYPE]);
 474 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
 475 extern void     zil_destroy(zilog_t *zilog, boolean_t keep_first);
 476 extern void     zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
 477
 478 extern itx_t    *zil_itx_create(uint64_t txtype, size_t lrsize);
 479 extern void     zil_itx_destroy(itx_t *itx);
 480 extern void     zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
 481
 482 extern void     zil_commit(zilog_t *zilog, uint64_t oid);
 483
 484 extern int      zil_vdev_offline(const char *osname, void *txarg);
 485 extern int      zil_claim(struct dsl_pool *dp,
 486     struct dsl_dataset *ds, void *txarg);
 487 extern int      zil_check_log_chain(struct dsl_pool *dp,
 488     struct dsl_dataset *ds, void *tx);
 489 extern void     zil_sync(zilog_t *zilog, dmu_tx_t *tx);
 490 extern void     zil_clean(zilog_t *zilog, uint64_t synced_txg);
 491
 492 extern int      zil_suspend(const char *osname, void **cookiep);
 493 extern void     zil_resume(void *cookie);
 494
 495 extern void     zil_add_block(zilog_t *zilog, const blkptr_t *bp);
 496 extern int      zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
 497
 498 extern void     zil_set_sync(zilog_t *zilog, uint64_t syncval);
 499
 500 extern void     zil_set_logbias(zilog_t *zilog, uint64_t slogval);
 501
 502 extern int zil_replay_disable;
 503
 504 #ifdef  __cplusplus
 505 }
 506 #endif
 507
 508 #endif  /* _SYS_ZIL_H */