bsd/vfs/vfs_journal.h

   1
   2 /*
   3  * Copyright (c) 2000-2002 Apple Computer, Inc. All rights reserved.
   4  *
   5  * @APPLE_LICENSE_HEADER_START@
   6  *
   7  * The contents of this file constitute Original Code as defined in and
   8  * are subject to the Apple Public Source License Version 1.1 (the
   9  * "License").  You may not use this file except in compliance with the
  10  * License.  Please obtain a copy of the License at
  11  * http://www.apple.com/publicsource and read it before using this file.
  12  *
  13  * This Original Code and all software distributed under the License are
  14  * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  15  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  16  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
  18  * License for the specific language governing rights and limitations
  19  * under the License.
  20  *
  21  * @APPLE_LICENSE_HEADER_END@
  22  */
  23 /*
  24  * This header contains the structures and function prototypes
  25  * for the vfs journaling code.  The data types are not meant
  26  * to be modified by user code.  Just use the functions and do
  27  * not mess around with the structs.
  28  */
  29 #ifndef _SYS_VFS_JOURNAL_H_
  30 #define _SYS_VFS_JOURNAL_H_
  31
  32 #include <sys/appleapiopts.h>
  33 #include <sys/cdefs.h>
  34
  35 #ifdef __APPLE_API_UNSTABLE
  36
  37 #include <sys/types.h>
  38 #include <kern/locks.h>
  39
  40 typedef struct block_info {
  41     off_t       bnum;                // block # on the file system device
  42     size_t      bsize;               // in bytes
  43     struct buf *bp;
  44 } block_info;
  45
  46 typedef struct block_list_header {
  47     u_int16_t   max_blocks;          // max number of blocks in this chunk
  48     u_int16_t   num_blocks;          // number of valid block numbers in block_nums
  49     int32_t     bytes_used;          // how many bytes of this tbuffer are used
  50     int32_t     checksum;            // on-disk: checksum of this header and binfo[0]
  51     int32_t     pad;                 // pad out to 16 bytes
  52     block_info  binfo[1];            // so we can reference them by name
  53 } block_list_header;
  54
  55
  56 struct journal;
  57
  58 typedef struct transaction {
  59     int                 tbuffer_size;  // in bytes
  60     char               *tbuffer;       // memory copy of the transaction
  61     block_list_header  *blhdr;         // points to the first byte of tbuffer
  62     int                 num_blhdrs;    // how many buffers we've allocated
  63     int                 total_bytes;   // total # of bytes in transaction
  64     int                 num_flushed;   // how many bytes have been flushed
  65     int                 num_killed;    // how many bytes were "killed"
  66     off_t               journal_start; // where in the journal this transaction starts
  67     off_t               journal_end;   // where in the journal this transaction ends
  68     struct journal     *jnl;           // ptr back to the journal structure
  69     struct transaction *next;          // list of tr's (either completed or to be free'd)
  70 } transaction;
  71
  72
  73 /*
  74  * This is written to block zero of the journal and it
  75  * maintains overall state about the journal.
  76  */
  77 typedef struct journal_header {
  78     int32_t        magic;
  79     int32_t        endian;
  80     volatile off_t start;         // zero-based byte offset of the start of the first transaction
  81     volatile off_t end;           // zero-based byte offset of where free space begins
  82     off_t          size;          // size in bytes of the entire journal
  83     int32_t        blhdr_size;    // size in bytes of each block_list_header in the journal
  84     int32_t        checksum;
  85     int32_t        jhdr_size;     // block size (in bytes) of the journal header
  86 } journal_header;
  87
  88 #define JOURNAL_HEADER_MAGIC  0x4a4e4c78   // 'JNLx'
  89 #define ENDIAN_MAGIC          0x12345678
  90
  91 #define OLD_JOURNAL_HEADER_MAGIC  0x4a484452   // 'JHDR'
  92
  93
  94 /*
  95  * In memory structure about the journal.
  96  */
  97 typedef struct journal {
  98     lck_mtx_t           jlock;             // protects the struct journal data
  99
 100     struct vnode       *jdev;              // vnode of the device where the journal lives
 101     off_t               jdev_offset;       // byte offset to the start of the journal
 102
 103     struct vnode       *fsdev;             // vnode of the file system device
 104
 105     void              (*flush)(void *arg); // fs callback to flush meta data blocks
 106     void               *flush_arg;         // arg that's passed to flush()
 107
 108     int32_t             flags;
 109     int32_t             tbuffer_size;      // default transaction buffer size
 110
 111     char               *header_buf;        // in-memory copy of the journal header
 112     journal_header     *jhdr;              // points to the first byte of header_buf
 113
 114     transaction        *cur_tr;            // for group-commit
 115     transaction        *completed_trs;     // out-of-order transactions that completed
 116     transaction        *active_tr;         // for nested transactions
 117     int32_t             nested_count;      // for nested transactions
 118     void               *owner;             // a ptr that's unique to the calling process
 119
 120     transaction        *tr_freeme;         // transaction structs that need to be free'd
 121
 122     volatile off_t      active_start;      // the active start that we only keep in memory
 123     lck_mtx_t           old_start_lock;    // protects the old_start
 124     volatile off_t      old_start[16];     // this is how we do lazy start update
 125
 126     int                 last_flush_err;    // last error from flushing the cache
 127 } journal;
 128
 129 /* internal-only journal flags (top 16 bits) */
 130 #define JOURNAL_CLOSE_PENDING     0x00010000
 131 #define JOURNAL_INVALID           0x00020000
 132 #define JOURNAL_FLUSHCACHE_ERR    0x00040000   // means we already printed this err
 133 #define JOURNAL_NEED_SWAP         0x00080000   // swap any data read from disk
 134
 135 /* journal_open/create options are always in the low-16 bits */
 136 #define JOURNAL_OPTION_FLAGS_MASK 0x0000ffff
 137
 138 __BEGIN_DECLS
 139 /*
 140  * Prototypes.
 141  */
 142
 143 /*
 144  * Call journal_init() to initialize the journaling code (sets up lock attributes)
 145  */
 146 void      journal_init(void);
 147
 148 /*
 149  * Call journal_create() to create a new journal.  You only
 150  * call this once, typically at file system creation time.
 151  *
 152  * The "jvp" argument is the vnode where the journal is written.
 153  * The journal starts at "offset" and is "journal_size" bytes long.
 154  *
 155  * The "fsvp" argument is the vnode of your file system.  It may be
 156  * the same as "jvp".
 157  *
 158  * The "min_fs_block_size" argument is the minimum block size
 159  * (in bytes) that the file system will ever write.  Typically
 160  * this is the block size of the file system (1k, 4k, etc) but
 161  * on HFS+ it is the minimum block size of the underlying device.
 162  *
 163  * The flags argument lets you disable group commit if you
 164  * want tighter guarantees on transactions (in exchange for
 165  * lower performance).
 166  *
 167  * The tbuffer_size is the size of the transaction buffer
 168  * used by the journal. If you specify zero, the journal code
 169  * will use a reasonable defaults.  The tbuffer_size should
 170  * be an integer multiple of the min_fs_block_size.
 171  *
 172  * Returns a valid journal pointer or NULL if one could not
 173  * be created.
 174  */
 175 journal *journal_create(struct vnode *jvp,
 176                                                 off_t         offset,
 177                                                 off_t         journal_size,
 178                                                 struct vnode *fsvp,
 179                                                 size_t        min_fs_block_size,
 180                                                 int32_t       flags,
 181                                                 int32_t       tbuffer_size,
 182                                                 void        (*flush)(void *arg),
 183                                                 void         *arg);
 184
 185 /*
 186  * Call journal_open() when mounting an existing file system
 187  * that has a previously created journal.  It will take care
 188  * of validating the journal and replaying it if necessary.
 189  *
 190  * See journal_create() for a description of the arguments.
 191  *
 192  * Returns a valid journal pointer of NULL if it runs into
 193  * trouble reading/playing back the journal.
 194  */
 195 journal  *journal_open(struct vnode *jvp,
 196                                            off_t         offset,
 197                                            off_t         journal_size,
 198                                            struct vnode *fsvp,
 199                                            size_t        min_fs_block_size,
 200                                            int32_t       flags,
 201                                            int32_t       tbuffer_size,
 202                                            void        (*flush)(void *arg),
 203                                            void         *arg);
 204
 205 /*
 206  * Test whether the journal is clean or not.  This is intended
 207  * to be used when you're mounting read-only.  If the journal
 208  * is not clean for some reason then you should not mount the
 209  * volume as your data structures may be in an unknown state.
 210  */
 211 int journal_is_clean(struct vnode *jvp,
 212                      off_t         offset,
 213                      off_t         journal_size,
 214                      struct vnode *fsvp,
 215                      size_t        min_fs_block_size);
 216
 217
 218 /*
 219  * Call journal_close() just before your file system is unmounted.
 220  * It flushes any outstanding transactions and makes sure the
 221  * journal is in a consistent state.
 222  */
 223 void      journal_close(journal *journalp);
 224
 225 /*
 226  * flags for journal_create/open.  only can use
 227  * the low 16 bits for flags because internal
 228  * bits go in the high 16.
 229  */
 230 #define JOURNAL_NO_GROUP_COMMIT   0x00000001
 231 #define JOURNAL_RESET             0x00000002
 232
 233 /*
 234  * Transaction related functions.
 235  *
 236  * Before you start modifying file system meta data, you
 237  * should call journal_start_transaction().  Then before
 238  * you modify each block, call journal_modify_block_start()
 239  * and when you're done, journal_modify_block_end().  When
 240  * you've modified the last block as part of a transaction,
 241  * call journal_end_transaction() to commit the changes.
 242  *
 243  * If you decide to abort the modifications to a block you
 244  * should call journal_modify_block_abort().
 245  *
 246  * If as part of a transaction you need want to throw out
 247  * any previous copies of a block (because it got deleted)
 248  * then call journal_kill_block().  This will mark it so
 249  * that the journal does not play it back (effectively
 250  * dropping it).
 251  */
 252 int   journal_start_transaction(journal *jnl);
 253 int   journal_modify_block_start(journal *jnl, struct buf *bp);
 254 int   journal_modify_block_abort(journal *jnl, struct buf *bp);
 255 int   journal_modify_block_end(journal *jnl, struct buf *bp);
 256 int   journal_kill_block(journal *jnl, struct buf *bp);
 257 int   journal_end_transaction(journal *jnl);
 258
 259 int   journal_active(journal *jnl);
 260 int   journal_flush(journal *jnl);
 261 void *journal_owner(journal *jnl);    // compare against current_thread()
 262
 263 __END_DECLS
 264
 265 #endif /* __APPLE_API_UNSTABLE */
 266 #endif /* !_SYS_VFS_JOURNAL_H_ */