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 #include <sys/dsl_pool.h>
  36 #include <sys/dsl_dataset.h>
  37 
  38 #ifdef  __cplusplus
  39 extern "C" {
  40 #endif
  41 
  42 /*
  43  * Intent log format:
  44  *
  45  * Each objset has its own intent log.  The log header (zil_header_t)
  46  * for objset N's intent log is kept in the Nth object of the SPA's
  47  * intent_log objset.  The log header points to a chain of log blocks,
  48  * each of which contains log records (i.e., transactions) followed by
  49  * a log block trailer (zil_trailer_t).  The format of a log record
  50  * depends on the record (or transaction) type, but all records begin
  51  * with a common structure that defines the type, length, and txg.
  52  */
  53 
  54 /*
  55  * Intent log header - this on disk structure holds fields to manage
  56  * the log.  All fields are 64 bit to easily handle cross architectures.
  57  */
  58 typedef struct zil_header {
  59         uint64_t zh_claim_txg;  /* txg in which log blocks were claimed */
  60         uint64_t zh_replay_seq; /* highest replayed sequence number */
  61         blkptr_t zh_log;        /* log chain */
  62         uint64_t zh_claim_blk_seq; /* highest claimed block sequence number */
  63         uint64_t zh_flags;      /* header flags */
  64         uint64_t zh_claim_lr_seq; /* highest claimed lr sequence number */
  65         uint64_t zh_pad[3];
  66 } zil_header_t;
  67 
  68 /*
  69  * zh_flags bit settings
  70  */
  71 #define ZIL_REPLAY_NEEDED       0x1     /* replay needed - internal only */
  72 #define ZIL_CLAIM_LR_SEQ_VALID  0x2     /* zh_claim_lr_seq field is valid */
  73 
  74 /*
  75  * Log block chaining.
  76  *
  77  * Log blocks are chained together. Originally they were chained at the
  78  * end of the block. For performance reasons the chain was moved to the
  79  * beginning of the block which allows writes for only the data being used.
  80  * The older position is supported for backwards compatability.
  81  *
  82  * The zio_eck_t contains a zec_cksum which for the intent log is
  83  * the sequence number of this log block. A seq of 0 is invalid.
  84  * The zec_cksum is checked by the SPA against the sequence
  85  * number passed in the blk_cksum field of the blkptr_t
  86  */
  87 typedef struct zil_chain {
  88         uint64_t zc_pad;
  89         blkptr_t zc_next_blk;   /* next block in chain */
  90         uint64_t zc_nused;      /* bytes in log block used */
  91         zio_eck_t zc_eck;       /* block trailer */
  92 } zil_chain_t;
  93 
  94 #define ZIL_MIN_BLKSZ   4096ULL
  95 #define ZIL_MAX_BLKSZ   SPA_MAXBLOCKSIZE
  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  * Format of log records.
 177  * The fields are carefully defined to allow them to be aligned
 178  * and sized the same on sparc & intel architectures.
 179  * Each log record has a common structure at the beginning.
 180  *
 181  * The log record on disk (lrc_seq) holds the sequence number of all log
 182  * records which is used to ensure we don't replay the same record.
 183  */
 184 typedef struct {                        /* common log record header */
 185         uint64_t        lrc_txtype;     /* intent log transaction type */
 186         uint64_t        lrc_reclen;     /* transaction record length */
 187         uint64_t        lrc_txg;        /* dmu transaction group number */
 188         uint64_t        lrc_seq;        /* see comment above */
 189 } lr_t;
 190 
 191 /*
 192  * Common start of all out-of-order record types (TX_OOO() above).
 193  */
 194 typedef struct {
 195         lr_t            lr_common;      /* common portion of log record */
 196         uint64_t        lr_foid;        /* object id */
 197 } lr_ooo_t;
 198 
 199 /*
 200  * Handle option extended vattr attributes.
 201  *
 202  * Whenever new attributes are added the version number
 203  * will need to be updated as will code in
 204  * zfs_log.c and zfs_replay.c
 205  */
 206 typedef struct {
 207         uint32_t        lr_attr_masksize; /* number of elements in array */
 208         uint32_t        lr_attr_bitmap; /* First entry of array */
 209         /* remainder of array and any additional fields */
 210 } lr_attr_t;
 211 
 212 /*
 213  * log record for creates without optional ACL.
 214  * This log record does support optional xvattr_t attributes.
 215  */
 216 typedef struct {
 217         lr_t            lr_common;      /* common portion of log record */
 218         uint64_t        lr_doid;        /* object id of directory */
 219         uint64_t        lr_foid;        /* object id of created file object */
 220         uint64_t        lr_mode;        /* mode of object */
 221         uint64_t        lr_uid;         /* uid of object */
 222         uint64_t        lr_gid;         /* gid of object */
 223         uint64_t        lr_gen;         /* generation (txg of creation) */
 224         uint64_t        lr_crtime[2];   /* creation time */
 225         uint64_t        lr_rdev;        /* rdev of object to create */
 226         /* name of object to create follows this */
 227         /* for symlinks, link content follows name */
 228         /* for creates with xvattr data, the name follows the xvattr info */
 229 } lr_create_t;
 230 
 231 /*
 232  * FUID ACL record will be an array of ACEs from the original ACL.
 233  * If this array includes ephemeral IDs, the record will also include
 234  * an array of log-specific FUIDs to replace the ephemeral IDs.
 235  * Only one copy of each unique domain will be present, so the log-specific
 236  * FUIDs will use an index into a compressed domain table.  On replay this
 237  * information will be used to construct real FUIDs (and bypass idmap,
 238  * since it may not be available).
 239  */
 240 
 241 /*
 242  * Log record for creates with optional ACL
 243  * This log record is also used for recording any FUID
 244  * information needed for replaying the create.  If the
 245  * file doesn't have any actual ACEs then the lr_aclcnt
 246  * would be zero.
 247  *
 248  * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's.
 249  * If create is also setting xvattr's, then acl data follows xvattr.
 250  * If ACE FUIDs are needed then they will follow the xvattr_t.  Following
 251  * the FUIDs will be the domain table information.  The FUIDs for the owner
 252  * and group will be in lr_create.  Name follows ACL data.
 253  */
 254 typedef struct {
 255         lr_create_t     lr_create;      /* common create portion */
 256         uint64_t        lr_aclcnt;      /* number of ACEs in ACL */
 257         uint64_t        lr_domcnt;      /* number of unique domains */
 258         uint64_t        lr_fuidcnt;     /* number of real fuids */
 259         uint64_t        lr_acl_bytes;   /* number of bytes in ACL */
 260         uint64_t        lr_acl_flags;   /* ACL flags */
 261 } lr_acl_create_t;
 262 
 263 typedef struct {
 264         lr_t            lr_common;      /* common portion of log record */
 265         uint64_t        lr_doid;        /* obj id of directory */
 266         /* name of object to remove follows this */
 267 } lr_remove_t;
 268 
 269 typedef struct {
 270         lr_t            lr_common;      /* common portion of log record */
 271         uint64_t        lr_doid;        /* obj id of directory */
 272         uint64_t        lr_link_obj;    /* obj id of link */
 273         /* name of object to link follows this */
 274 } lr_link_t;
 275 
 276 typedef struct {
 277         lr_t            lr_common;      /* common portion of log record */
 278         uint64_t        lr_sdoid;       /* obj id of source directory */
 279         uint64_t        lr_tdoid;       /* obj id of target directory */
 280         /* 2 strings: names of source and destination follow this */
 281 } lr_rename_t;
 282 
 283 typedef struct {
 284         lr_t            lr_common;      /* common portion of log record */
 285         uint64_t        lr_foid;        /* file object to write */
 286         uint64_t        lr_offset;      /* offset to write to */
 287         uint64_t        lr_length;      /* user data length to write */
 288         uint64_t        lr_blkoff;      /* no longer used */
 289         blkptr_t        lr_blkptr;      /* spa block pointer for replay */
 290         /* write data will follow for small writes */
 291 } lr_write_t;
 292 
 293 typedef struct {
 294         lr_t            lr_common;      /* common portion of log record */
 295         uint64_t        lr_foid;        /* object id of file to truncate */
 296         uint64_t        lr_offset;      /* offset to truncate from */
 297         uint64_t        lr_length;      /* length to truncate */
 298 } lr_truncate_t;
 299 
 300 typedef struct {
 301         lr_t            lr_common;      /* common portion of log record */
 302         uint64_t        lr_foid;        /* file object to change attributes */
 303         uint64_t        lr_mask;        /* mask of attributes to set */
 304         uint64_t        lr_mode;        /* mode to set */
 305         uint64_t        lr_uid;         /* uid to set */
 306         uint64_t        lr_gid;         /* gid to set */
 307         uint64_t        lr_size;        /* size to set */
 308         uint64_t        lr_atime[2];    /* access time */
 309         uint64_t        lr_mtime[2];    /* modification time */
 310         /* optional attribute lr_attr_t may be here */
 311 } lr_setattr_t;
 312 
 313 typedef struct {
 314         lr_t            lr_common;      /* common portion of log record */
 315         uint64_t        lr_foid;        /* obj id of file */
 316         uint64_t        lr_aclcnt;      /* number of acl entries */
 317         /* lr_aclcnt number of ace_t entries follow this */
 318 } lr_acl_v0_t;
 319 
 320 typedef struct {
 321         lr_t            lr_common;      /* common portion of log record */
 322         uint64_t        lr_foid;        /* obj id of file */
 323         uint64_t        lr_aclcnt;      /* number of ACEs in ACL */
 324         uint64_t        lr_domcnt;      /* number of unique domains */
 325         uint64_t        lr_fuidcnt;     /* number of real fuids */
 326         uint64_t        lr_acl_bytes;   /* number of bytes in ACL */
 327         uint64_t        lr_acl_flags;   /* ACL flags */
 328         /* lr_acl_bytes number of variable sized ace's follows */
 329 } lr_acl_t;
 330 
 331 /*
 332  * ZIL structure definitions, interface function prototype and globals.
 333  */
 334 
 335 /*
 336  * Writes are handled in three different ways:
 337  *
 338  * WR_INDIRECT:
 339  *    In this mode, if we need to commit the write later, then the block
 340  *    is immediately written into the file system (using dmu_sync),
 341  *    and a pointer to the block is put into the log record.
 342  *    When the txg commits the block is linked in.
 343  *    This saves additionally writing the data into the log record.
 344  *    There are a few requirements for this to occur:
 345  *      - write is greater than zfs/zvol_immediate_write_sz
 346  *      - not using slogs (as slogs are assumed to always be faster
 347  *        than writing into the main pool)
 348  *      - the write occupies only one block
 349  * WR_COPIED:
 350  *    If we know we'll immediately be committing the
 351  *    transaction (FSYNC or FDSYNC), the we allocate a larger
 352  *    log record here for the data and copy the data in.
 353  * WR_NEED_COPY:
 354  *    Otherwise we don't allocate a buffer, and *if* we need to
 355  *    flush the write later then a buffer is allocated and
 356  *    we retrieve the data using the dmu.
 357  */
 358 typedef enum {
 359         WR_INDIRECT,    /* indirect - a large write (dmu_sync() data */
 360                         /* and put blkptr in log, rather than actual data) */
 361         WR_COPIED,      /* immediate - data is copied into lr_write_t */
 362         WR_NEED_COPY,   /* immediate - data needs to be copied if pushed */
 363         WR_NUM_STATES   /* number of states */
 364 } itx_wr_state_t;
 365 
 366 typedef struct itx {
 367         list_node_t     itx_node;       /* linkage on zl_itx_list */
 368         void            *itx_private;   /* type-specific opaque data */
 369         itx_wr_state_t  itx_wr_state;   /* write state */
 370         uint8_t         itx_sync;       /* synchronous transaction */
 371         uint64_t        itx_sod;        /* record size on disk */
 372         uint64_t        itx_oid;        /* object id */
 373         lr_t            itx_lr;         /* common part of log record */
 374         /* followed by type-specific part of lr_xx_t and its immediate data */
 375 } itx_t;
 376 
 377 typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg,
 378     uint64_t txg);
 379 typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg,
 380     uint64_t txg);
 381 typedef int zil_replay_func_t();
 382 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio);
 383 
 384 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
 385     zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg);
 386 
 387 extern void     zil_init(void);
 388 extern void     zil_fini(void);
 389 
 390 extern zilog_t  *zil_alloc(objset_t *os, zil_header_t *zh_phys);
 391 extern void     zil_free(zilog_t *zilog);
 392 
 393 extern zilog_t  *zil_open(objset_t *os, zil_get_data_t *get_data);
 394 extern void     zil_close(zilog_t *zilog);
 395 
 396 extern void     zil_replay(objset_t *os, void *arg,
 397     zil_replay_func_t *replay_func[TX_MAX_TYPE]);
 398 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx);
 399 extern void     zil_destroy(zilog_t *zilog, boolean_t keep_first);
 400 extern void     zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx);
 401 extern void     zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx);
 402 
 403 extern itx_t    *zil_itx_create(uint64_t txtype, size_t lrsize);
 404 extern void     zil_itx_destroy(itx_t *itx);
 405 extern void     zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx);
 406 
 407 extern void     zil_commit(zilog_t *zilog, uint64_t oid);
 408 
 409 extern int      zil_vdev_offline(const char *osname, void *txarg);
 410 extern int      zil_claim(dsl_pool_t *dp, dsl_dataset_t *ds, void *txarg);
 411 extern int      zil_check_log_chain(dsl_pool_t *dp, dsl_dataset_t *ds,
 412     void *tx);
 413 extern void     zil_sync(zilog_t *zilog, dmu_tx_t *tx);
 414 extern void     zil_clean(zilog_t *zilog, uint64_t synced_txg);
 415 
 416 extern int      zil_suspend(const char *osname, void **cookiep);
 417 extern void     zil_resume(void *cookie);
 418 
 419 extern void     zil_add_block(zilog_t *zilog, const blkptr_t *bp);
 420 extern int      zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);
 421 
 422 extern void     zil_set_sync(zilog_t *zilog, uint64_t syncval);
 423 
 424 extern void     zil_set_logbias(zilog_t *zilog, uint64_t slogval);
 425 
 426 extern int zil_replay_disable;
 427 
 428 #ifdef  __cplusplus
 429 }
 430 #endif
 431 
 432 #endif  /* _SYS_ZIL_H */