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, 2017 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 #include <sys/zio_crypt.h>
  36
  37 #ifdef  __cplusplus
  38 extern "C" {
  39 #endif
  40
  41 struct dsl_pool;
  42 struct dsl_dataset;
  43
  44 /*
  45  * Intent log format:
  46  *
  47  * Each objset has its own intent log.  The log header (zil_header_t)
  48  * for objset N's intent log is kept in the Nth object of the SPA's
  49  * intent_log objset.  The log header points to a chain of log blocks,
  50  * each of which contains log records (i.e., transactions) followed by
  51  * a log block trailer (zil_trailer_t).  The format of a log record
  52  * depends on the record (or transaction) type, but all records begin
  53  * with a common structure that defines the type, length, and txg.
  54  */
  55
  56 /*
  57  * Intent log header - this on disk structure holds fields to manage
  58  * the log.  All fields are 64 bit to easily handle cross architectures.
  59  */
  60 typedef struct zil_header {
  61         uint64_t zh_claim_txg;  /* txg in which log blocks were claimed */
  62         uint64_t zh_replay_seq; /* highest replayed sequence number */
  63         blkptr_t zh_log;        /* log chain */
  64         uint64_t zh_claim_blk_seq; /* highest claimed block sequence number */
  65         uint64_t zh_flags;      /* header flags */
  66         uint64_t zh_claim_lr_seq; /* highest claimed lr sequence number */
  67         uint64_t zh_pad[3];
  68 } zil_header_t;
  69
  70 /*
  71  * zh_flags bit settings
  72  */
  73 #define ZIL_REPLAY_NEEDED       0x1     /* replay needed - internal only */
  74 #define ZIL_CLAIM_LR_SEQ_VALID  0x2     /* zh_claim_lr_seq field is valid */
  75
  76 /*
  77  * Log block chaining.
  78  *
  79  * Log blocks are chained together. Originally they were chained at the
  80  * end of the block. For performance reasons the chain was moved to the
  81  * beginning of the block which allows writes for only the data being used.
  82  * The older position is supported for backwards compatability.
  83  *
  84  * The zio_eck_t contains a zec_cksum which for the intent log is
  85  * the sequence number of this log block. A seq of 0 is invalid.
  86  * The zec_cksum is checked by the SPA against the sequence
  87  * number passed in the blk_cksum field of the blkptr_t
  88  */
  89 typedef struct zil_chain {
  90         uint64_t zc_pad;
  91         blkptr_t zc_next_blk;   /* next block in chain */
  92         uint64_t zc_nused;      /* bytes in log block used */
  93         zio_eck_t zc_eck;       /* block trailer */
  94 } zil_chain_t;
  95
  96 #define ZIL_MIN_BLKSZ   4096ULL
  97
  98 /*
  99  * ziltest is by and large an ugly hack, but very useful in
 100  * checking replay without tedious work.
 101  * When running ziltest we want to keep all itx's and so maintain
 102  * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG
 103  * We subtract TXG_CONCURRENT_STATES to allow for common code.
 104  */
 105 #define ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES)
 106
 107 /*
 108  * The words of a log block checksum.
 109  */
 110 #define ZIL_ZC_GUID_0   0
 111 #define ZIL_ZC_GUID_1   1
 112 #define ZIL_ZC_OBJSET   2
 113 #define ZIL_ZC_SEQ      3
 114
 115 typedef enum zil_create {
 116         Z_FILE,
 117         Z_DIR,
 118         Z_XATTRDIR,
 119 } zil_create_t;
 120
 121 /*
 122  * size of xvattr log section.
 123  * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps
 124  * for create time and a single 64 bit integer for all of the attributes,
 125  * and 4 64 bit integers (32 bytes) for the scanstamp.
 126  *
 127  */
 128
 129 #define ZIL_XVAT_SIZE(mapsize) \
 130         sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \
 131         (sizeof (uint64_t) * 7)
 132
 133 /*
 134  * Size of ACL in log.  The ACE data is padded out to properly align
 135  * on 8 byte boundary.
 136  */
 137
 138 #define ZIL_ACE_LENGTH(x)       (roundup(x, sizeof (uint64_t)))
 139
 140 /*
 141  * Intent log transaction types and record structures
 142  */
 143 #define TX_CREATE               1       /* Create file */
 144 #define TX_MKDIR                2       /* Make directory */
 145 #define TX_MKXATTR              3       /* Make XATTR directory */
 146 #define TX_SYMLINK              4       /* Create symbolic link to a file */
 147 #define TX_REMOVE               5       /* Remove file */
 148 #define TX_RMDIR                6       /* Remove directory */
 149 #define TX_LINK                 7       /* Create hard link to a file */
 150 #define TX_RENAME               8       /* Rename a file */
 151 #define TX_WRITE                9       /* File write */
 152 #define TX_TRUNCATE             10      /* Truncate a file */
 153 #define TX_SETATTR              11      /* Set file attributes */
 154 #define TX_ACL_V0               12      /* Set old formatted ACL */
 155 #define TX_ACL                  13      /* Set ACL */
 156 #define TX_CREATE_ACL           14      /* create with ACL */
 157 #define TX_CREATE_ATTR          15      /* create + attrs */
 158 #define TX_CREATE_ACL_ATTR      16      /* create with ACL + attrs */
 159 #define TX_MKDIR_ACL            17      /* mkdir with ACL */
 160 #define TX_MKDIR_ATTR           18      /* mkdir with attr */
 161 #define TX_MKDIR_ACL_ATTR       19      /* mkdir with ACL + attrs */
 162 #define TX_WRITE2               20      /* dmu_sync EALREADY write */
 163 #define TX_MAX_TYPE             21      /* Max transaction type */
 164
 165 /*
 166  * The transactions for mkdir, symlink, remove, rmdir, link, and rename
 167  * may have the following bit set, indicating the original request
 168  * specified case-insensitive handling of names.
 169  */
 170 #define TX_CI   ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */
 171
 172 /*
 173  * Transactions for write, truncate, setattr, acl_v0, and acl can be logged
 174  * out of order.  For convenience in the code, all such records must have
 175  * lr_foid at the same offset.
 176  */
 177 #define TX_OOO(txtype)                  \
 178         ((txtype) == TX_WRITE ||        \
 179         (txtype) == TX_TRUNCATE ||      \
 180         (txtype) == TX_SETATTR ||       \
 181         (txtype) == TX_ACL_V0 ||        \
 182         (txtype) == TX_ACL ||           \
 183         (txtype) == TX_WRITE2)
 184
 185 /*
 186  * The number of dnode slots consumed by the object is stored in the 8
 187  * unused upper bits of the object ID. We subtract 1 from the value
 188  * stored on disk for compatibility with implementations that don't
 189  * support large dnodes. The slot count for a single-slot dnode will
 190  * contain 0 for those bits to preserve the log record format for
 191  * "small" dnodes.
 192  */
 193 #define LR_FOID_GET_SLOTS(oid) (BF64_GET((oid), 56, 8) + 1)
 194 #define LR_FOID_SET_SLOTS(oid, x) BF64_SET((oid), 56, 8, (x) - 1)
 195 #define LR_FOID_GET_OBJ(oid) BF64_GET((oid), 0, DN_MAX_OBJECT_SHIFT)
 196 #define LR_FOID_SET_OBJ(oid, x) BF64_SET((oid), 0, DN_MAX_OBJECT_SHIFT, (x))
 197
 198 /*
 199  * Format of log records.
 200  * The fields are carefully defined to allow them to be aligned
 201  * and sized the same on sparc & intel architectures.
 202  * Each log record has a common structure at the beginning.
 203  *
 204  * The log record on disk (lrc_seq) holds the sequence number of all log
 205  * records which is used to ensure we don't replay the same record.
 206  */
 207 typedef struct {                        /* common log record header */
 208         uint64_t        lrc_txtype;     /* intent log transaction type */
 209         uint64_t        lrc_reclen;     /* transaction record length */
 210         uint64_t        lrc_txg;        /* dmu transaction group number */
 211         uint64_t        lrc_seq;        /* see comment above */
 212 } lr_t;
 213
 214 /*
 215  * Common start of all out-of-order record types (TX_OOO() above).
 216  */
 217 typedef struct {
 218         lr_t            lr_common;      /* common portion of log record */
 219         uint64_t        lr_foid;        /* object id */
 220 } lr_ooo_t;
 221
 222 /*
 223  * Handle option extended vattr attributes.
 224  *
 225  * Whenever new attributes are added the version number
 226  * will need to be updated as will code in
 227  * zfs_log.c and zfs_replay.c
 228  */
 229 typedef struct {
 230         uint32_t        lr_attr_masksize; /* number of elements in array */
 231         uint32_t        lr_attr_bitmap; /* First entry of array */
 232         /* remainder of array and any additional fields */
 233 } lr_attr_t;
 234
 235 /*
 236  * log record for creates without optional ACL.
 237  * This log record does support optional xvattr_t attributes.
 238  */
 239 typedef struct {
 240         lr_t            lr_common;      /* common portion of log record */
 241         uint64_t        lr_doid;        /* object id of directory */
 242         uint64_t        lr_foid;        /* object id of created file object */
 243         uint64_t        lr_mode;        /* mode of object */
 244         uint64_t        lr_uid;         /* uid of object */
 245         uint64_t        lr_gid;         /* gid of object */
 246         uint64_t        lr_gen;         /* generation (txg of creation) */
 247         uint64_t        lr_crtime[2];   /* creation time */
 248         uint64_t        lr_rdev;        /* rdev of object to create */
 249         /* name of object to create follows this */
 250         /* for symlinks, link content follows name */
 251         /* for creates with xvattr data, the name follows the xvattr info */
 252 } lr_create_t;
 253
 254 /*
 255  * FUID ACL record will be an array of ACEs from the original ACL.
 256  * If this array includes ephemeral IDs, the record will also include
 257  * an array of log-specific FUIDs to replace the ephemeral IDs.
 258  * Only one copy of each unique domain will be present, so the log-specific
 259  * FUIDs will use an index into a compressed domain table.  On replay this
 260  * information will be used to construct real FUIDs (and bypass idmap,
 261  * since it may not be available).
 262  */
 263
 264 /*
 265  * Log record for creates with optional ACL
 266  * This log record is also used for recording any FUID
 267  * information needed for replaying the create.  If the
 268  * file doesn't have any actual ACEs then the lr_aclcnt
 269  * would be zero.
 270  *
 271  * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
 272  * If create is also setting xvattr's, then acl data follows xvattr.
 273  * If ACE FUIDs are needed then they will follow the xvattr_t.  Following
 274  * the FUIDs will be the domain table information.  The FUIDs for the owner
 275  * and group will be in lr_create.  Name follows ACL data.
 276  */
 277 typedef struct {
 278         lr_create_t     lr_create;      /* common create portion */
 279         uint64_t        lr_aclcnt;      /* number of ACEs in ACL */
 280         uint64_t        lr_domcnt;      /* number of unique domains */
 281         uint64_t        lr_fuidcnt;     /* number of real fuids */
 282         uint64_t        lr_acl_bytes;   /* number of bytes in ACL */
 283         uint64_t        lr_acl_flags;   /* ACL flags */
 284 } lr_acl_create_t;
 285
 286 typedef struct {
 287         lr_t            lr_common;      /* common portion of log record */
 288         uint64_t        lr_doid;        /* obj id of directory */
 289         /* name of object to remove follows this */
 290 } lr_remove_t;
 291
 292 typedef struct {
 293         lr_t            lr_common;      /* common portion of log record */
 294         uint64_t        lr_doid;        /* obj id of directory */
 295         uint64_t        lr_link_obj;    /* obj id of link */
 296         /* name of object to link follows this */
 297 } lr_link_t;
 298
 299 typedef struct {
 300         lr_t            lr_common;      /* common portion of log record */
 301         uint64_t        lr_sdoid;       /* obj id of source directory */
 302         uint64_t        lr_tdoid;       /* obj id of target directory */
 303         /* 2 strings: names of source and destination follow this */
 304 } lr_rename_t;
 305
 306 typedef struct {
 307         lr_t            lr_common;      /* common portion of log record */
 308         uint64_t        lr_foid;        /* file object to write */
 309         uint64_t        lr_offset;      /* offset to write to */
 310         uint64_t        lr_length;      /* user data length to write */
 311         uint64_t        lr_blkoff;      /* no longer used */
 312         blkptr_t        lr_blkptr;      /* spa block pointer for replay */
 313         /* write data will follow for small writes */
 314 } lr_write_t;
 315
 316 typedef struct {
 317         lr_t            lr_common;      /* common portion of log record */
 318         uint64_t        lr_foid;        /* object id of file to truncate */
 319         uint64_t        lr_offset;      /* offset to truncate from */
 320         uint64_t        lr_length;      /* length to truncate */
 321 } lr_truncate_t;
 322
 323 typedef struct {
 324         lr_t            lr_common;      /* common portion of log record */
 325         uint64_t        lr_foid;        /* file object to change attributes */
 326         uint64_t        lr_mask;        /* mask of attributes to set */
 327         uint64_t        lr_mode;        /* mode to set */
 328         uint64_t        lr_uid;         /* uid to set */
 329         uint64_t        lr_gid;         /* gid to set */
 330         uint64_t        lr_size;        /* size to set */
 331         uint64_t        lr_atime[2];    /* access time */
 332         uint64_t        lr_mtime[2];    /* modification time */
 333         /* optional attribute lr_attr_t may be here */
 334 } lr_setattr_t;
 335
 336 typedef struct {
 337         lr_t            lr_common;      /* common portion of log record */
 338         uint64_t        lr_foid;        /* obj id of file */
 339         uint64_t        lr_aclcnt;      /* number of acl entries */
 340         /* lr_aclcnt number of ace_t entries follow this */
 341 } lr_acl_v0_t;
 342
 343 typedef struct {
 344         lr_t            lr_common;      /* common portion of log record */
 345         uint64_t        lr_foid;        /* obj id of file */
 346         uint64_t        lr_aclcnt;      /* number of ACEs in ACL */
 347         uint64_t        lr_domcnt;      /* number of unique domains */
 348         uint64_t        lr_fuidcnt;     /* number of real fuids */
 349         uint64_t        lr_acl_bytes;   /* number of bytes in ACL */
 350         uint64_t        lr_acl_flags;   /* ACL flags */
 351         /* lr_acl_bytes number of variable sized ace's follows */
 352 } lr_acl_t;
 353
 354 /*
 355  * ZIL structure definitions, interface function prototype and globals.
 356  */
 357
 358 /*
 359  * Writes are handled in three different ways:
 360  *
 361  * WR_INDIRECT:
 362  *    In this mode, if we need to commit the write later, then the block
 363  *    is immediately written into the file system (using dmu_sync),
 364  *    and a pointer to the block is put into the log record.
 365  *    When the txg commits the block is linked in.
 366  *    This saves additionally writing the data into the log record.
 367  *    There are a few requirements for this to occur:
 368  *      - write is greater than zfs/zvol_immediate_write_sz
 369  *      - not using slogs (as slogs are assumed to always be faster
 370  *        than writing into the main pool)
 371  *      - the write occupies only one block
 372  * WR_COPIED:
 373  *    If we know we'll immediately be committing the
 374  *    transaction (FSYNC or FDSYNC), then we allocate a larger
 375  *    log record here for the data and copy the data in.
 376  * WR_NEED_COPY:
 377  *    Otherwise we don't allocate a buffer, and *if* we need to
 378  *    flush the write later then a buffer is allocated and
 379  *    we retrieve the data using the dmu.
 380  */
 381 typedef enum {
 382         WR_INDIRECT,    /* indirect - a large write (dmu_sync() data */
 383                         /* and put blkptr in log, rather than actual data) */
 384         WR_COPIED,      /* immediate - data is copied into lr_write_t */
 385         WR_NEED_COPY,   /* immediate - data needs to be copied if pushed */
 386         WR_NUM_STATES   /* number of states */
 387 } itx_wr_state_t;
 388
 389 typedef void (*zil_callback_t)(void *data);
 390
 391 typedef struct itx {
 392         list_node_t     itx_node;       /* linkage on zl_itx_list */
 393         void            *itx_private;   /* type-specific opaque data */
 394         itx_wr_state_t  itx_wr_state;   /* write state */
 395         uint8_t         itx_sync;       /* synchronous transaction */
 396         zil_callback_t  itx_callback;   /* Called when the itx is persistent */
 397         void            *itx_callback_data; /* User data for the callback */
 398         uint64_t        itx_oid;        /* object id */
 399         lr_t            itx_lr;         /* common part of log record */
 400         /* followed by type-specific part of lr_xx_t and its immediate data */
 401 } itx_t;
 402
 403 /*
 404  * Used for zil kstat.
 405  */
 406 typedef struct zil_stats {
 407         /*
 408          * Number of times a ZIL commit (e.g. fsync) has been requested.
 409          */
 410         kstat_named_t zil_commit_count;
 411
 412         /*
 413          * Number of times the ZIL has been flushed to stable storage.
 414          * This is less than zil_commit_count when commits are "merged"
 415          * (see the documentation above zil_commit()).
 416          */
 417         kstat_named_t zil_commit_writer_count;
 418
 419         /*
 420          * Number of transactions (reads, writes, renames, etc.)
 421          * that have been commited.
 422          */
 423         kstat_named_t zil_itx_count;
 424
 425         /*
 426          * See the documentation for itx_wr_state_t above.
 427          * Note that "bytes" accumulates the length of the transactions
 428          * (i.e. data), not the actual log record sizes.
 429          */
 430         kstat_named_t zil_itx_indirect_count;
 431         kstat_named_t zil_itx_indirect_bytes;
 432         kstat_named_t zil_itx_copied_count;
 433         kstat_named_t zil_itx_copied_bytes;
 434         kstat_named_t zil_itx_needcopy_count;
 435         kstat_named_t zil_itx_needcopy_bytes;
 436
 437         /*
 438          * Transactions which have been allocated to the "normal"
 439          * (i.e. not slog) storage pool. Note that "bytes" accumulate
 440          * the actual log record sizes - which do not include the actual
 441          * data in case of indirect writes.
 442          */
 443         kstat_named_t zil_itx_metaslab_normal_count;
 444         kstat_named_t zil_itx_metaslab_normal_bytes;
 445
 446         /*
 447          * Transactions which have been allocated to the "slog" storage pool.
 448          * If there are no separate log devices, this is the same as the
 449          * "normal" pool.
 450          */
 451         kstat_named_t zil_itx_metaslab_slog_count;
 452         kstat_named_t zil_itx_metaslab_slog_bytes;
 453 } zil_stats_t;
 454
 455 extern zil_stats_t zil_stats;
 456
 457 #define ZIL_STAT_INCR(stat, val) \
 458     atomic_add_64(&zil_stats.stat.value.ui64, (val));
 459 #define ZIL_STAT_BUMP(stat) \
 460     ZIL_STAT_INCR(stat, 1);
 461
 462 typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
 463     uint64_t txg);
 464 typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
 465     uint64_t txg);
 466 typedef int (*const zil_replay_func_t)(void *, char *, boolean_t);
 467 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio);
 468
 469 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
 470     zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg,
 471     boolean_t decrypt);
 472
 473 extern void     zil_init(void);
 474 extern void     zil_fini(void);
 475
 476 extern zilog_t  *zil_alloc(objset_t *os, zil_header_t *zh_phys);
 477 extern void     zil_free(zilog_t *zilog);
 478
 479 extern zilog_t  *zil_open(objset_t *os, zil_get_data_t *get_data);
 480 extern void     zil_close(zilog_t *zilog);
 481
 482 extern void     zil_replay(objset_t *os, void *arg,
 483     zil_replay_func_t replay_func[TX_MAX_TYPE]);
 484 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
 485 extern void     zil_destroy(zilog_t *zilog, boolean_t keep_first);
 486 extern void     zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
 487
 488 extern itx_t    *zil_itx_create(uint64_t txtype, size_t lrsize);
 489 extern void     zil_itx_destroy(itx_t *itx);
 490 extern void     zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
 491
 492 extern void     zil_commit(zilog_t *zilog, uint64_t oid);
 493
 494 extern int      zil_vdev_offline(const char *osname, void *txarg);
 495 extern int      zil_claim(struct dsl_pool *dp,
 496     struct dsl_dataset *ds, void *txarg);
 497 extern int      zil_check_log_chain(struct dsl_pool *dp,
 498     struct dsl_dataset *ds, void *tx);
 499 extern void     zil_sync(zilog_t *zilog, dmu_tx_t *tx);
 500 extern void     zil_clean(zilog_t *zilog, uint64_t synced_txg);
 501
 502 extern int      zil_suspend(const char *osname, void **cookiep);
 503 extern void     zil_resume(void *cookie);
 504
 505 extern void     zil_add_block(zilog_t *zilog, const blkptr_t *bp);
 506 extern int      zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
 507
 508 extern void     zil_set_sync(zilog_t *zilog, uint64_t syncval);
 509
 510 extern void     zil_set_logbias(zilog_t *zilog, uint64_t slogval);
 511
 512 extern int zil_replay_disable;
 513
 514 #ifdef  __cplusplus
 515 }
 516 #endif
 517
 518 #endif  /* _SYS_ZIL_H */