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  * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
   8  *
   9  * This file contains Original Code and/or Modifications of Original Code
  10  * as defined in and that are subject to the Apple Public Source License
  11  * Version 2.0 (the 'License'). You may not use this file except in
  12  * compliance with the License. Please obtain a copy of the License at
  13  * http://www.opensource.apple.com/apsl/ and read it before using this
  14  * file.
  15  *
  16  * The Original Code and all software distributed under the License are
  17  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  18  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  19  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  21  * Please see the License for the specific language governing rights and
  22  * limitations under the License.
  23  *
  24  * @APPLE_LICENSE_HEADER_END@
  25  */
  26 /*
  27  * This header contains the structures and function prototypes
  28  * for the vfs journaling code.  The data types are not meant
  29  * to be modified by user code.  Just use the functions and do
  30  * not mess around with the structs.
  31  */
  32 #ifndef _SYS_VFS_JOURNAL_H_
  33 #define _SYS_VFS_JOURNAL_H_
  34
  35 #include <sys/appleapiopts.h>
  36
  37 #ifdef __APPLE_API_UNSTABLE
  38
  39 #include <sys/types.h>
  40 #include <sys/lock.h>
  41
  42 typedef struct block_info {
  43     off_t       bnum;                // block # on the file system device
  44     size_t      bsize;               // in bytes
  45     struct buf *bp;
  46 } block_info;
  47
  48 typedef struct block_list_header {
  49     u_int16_t   max_blocks;          // max number of blocks in this chunk
  50     u_int16_t   num_blocks;          // number of valid block numbers in block_nums
  51     int32_t     bytes_used;          // how many bytes of this tbuffer are used
  52     int32_t     checksum;            // on-disk: checksum of this header and binfo[0]
  53     int32_t     pad;                 // pad out to 16 bytes
  54     block_info  binfo[1];            // so we can reference them by name
  55 } block_list_header;
  56
  57
  58 struct journal;
  59
  60 typedef struct transaction {
  61     int                 tbuffer_size;  // in bytes
  62     char               *tbuffer;       // memory copy of the transaction
  63     block_list_header  *blhdr;         // points to the first byte of tbuffer
  64     int                 num_blhdrs;    // how many buffers we've allocated
  65     int                 total_bytes;   // total # of bytes in transaction
  66     int                 num_flushed;   // how many bytes have been flushed
  67     int                 num_killed;    // how many bytes were "killed"
  68     off_t               journal_start; // where in the journal this transaction starts
  69     off_t               journal_end;   // where in the journal this transaction ends
  70     struct journal     *jnl;           // ptr back to the journal structure
  71     struct transaction *next;          // list of tr's (either completed or to be free'd)
  72 } transaction;
  73
  74
  75 /*
  76  * This is written to block zero of the journal and it
  77  * maintains overall state about the journal.
  78  */
  79 typedef struct journal_header {
  80     int32_t        magic;
  81     int32_t        endian;
  82     volatile off_t start;         // zero-based byte offset of the start of the first transaction
  83     volatile off_t end;           // zero-based byte offset of where free space begins
  84     off_t          size;          // size in bytes of the entire journal
  85     int32_t        blhdr_size;    // size in bytes of each block_list_header in the journal
  86     int32_t        checksum;
  87     int32_t        jhdr_size;     // block size (in bytes) of the journal header
  88 } journal_header;
  89
  90 #define JOURNAL_HEADER_MAGIC  0x4a4e4c78   // 'JNLx'
  91 #define ENDIAN_MAGIC          0x12345678
  92
  93 #define OLD_JOURNAL_HEADER_MAGIC  0x4a484452   // 'JHDR'
  94
  95
  96 /*
  97  * In memory structure about the journal.
  98  */
  99 typedef struct journal {
 100         struct lock__bsd__      jlock;
 101
 102     struct vnode       *jdev;              // vnode of the device where the journal lives
 103     off_t               jdev_offset;       // byte offset to the start of the journal
 104
 105     struct vnode       *fsdev;             // vnode of the file system device
 106
 107     void              (*flush)(void *arg); // fs callback to flush meta data blocks
 108     void               *flush_arg;         // arg that's passed to flush()
 109
 110     int32_t             flags;
 111     int32_t             tbuffer_size;      // default transaction buffer size
 112
 113     char               *header_buf;        // in-memory copy of the journal header
 114     journal_header     *jhdr;              // points to the first byte of header_buf
 115
 116     transaction        *cur_tr;            // for group-commit
 117     transaction        *completed_trs;     // out-of-order transactions that completed
 118     transaction        *active_tr;         // for nested transactions
 119     int32_t             nested_count;      // for nested transactions
 120     void               *owner;             // a ptr that's unique to the calling process
 121
 122     transaction        *tr_freeme;         // transaction structs that need to be free'd
 123
 124         volatile off_t      active_start;      // the active start that we only keep in memory
 125         simple_lock_data_t  old_start_lock;    // guard access
 126         volatile off_t      old_start[16];     // this is how we do lazy start update
 127
 128         int                 last_flush_err;    // last error from flushing the cache
 129 } journal;
 130
 131 /* internal-only journal flags (top 16 bits) */
 132 #define JOURNAL_CLOSE_PENDING     0x00010000
 133 #define JOURNAL_INVALID           0x00020000
 134 #define JOURNAL_FLUSHCACHE_ERR    0x00040000   // means we already printed this err
 135 #define JOURNAL_NEED_SWAP         0x00080000   // swap any data read from disk
 136
 137 /* journal_open/create options are always in the low-16 bits */
 138 #define JOURNAL_OPTION_FLAGS_MASK 0x0000ffff
 139
 140 /*
 141  * Prototypes.
 142  */
 143
 144 /*
 145  * Call journal_create() to create a new journal.  You only
 146  * call this once, typically at file system creation time.
 147  *
 148  * The "jvp" argument is the vnode where the journal is written.
 149  * The journal starts at "offset" and is "journal_size" bytes long.
 150  *
 151  * The "fsvp" argument is the vnode of your file system.  It may be
 152  * the same as "jvp".
 153  *
 154  * The "min_fs_block_size" argument is the minimum block size
 155  * (in bytes) that the file system will ever write.  Typically
 156  * this is the block size of the file system (1k, 4k, etc) but
 157  * on HFS+ it is the minimum block size of the underlying device.
 158  *
 159  * The flags argument lets you disable group commit if you
 160  * want tighter guarantees on transactions (in exchange for
 161  * lower performance).
 162  *
 163  * The tbuffer_size is the size of the transaction buffer
 164  * used by the journal. If you specify zero, the journal code
 165  * will use a reasonable defaults.  The tbuffer_size should
 166  * be an integer multiple of the min_fs_block_size.
 167  *
 168  * Returns a valid journal pointer or NULL if one could not
 169  * be created.
 170  */
 171 journal *journal_create(struct vnode *jvp,
 172                                                 off_t         offset,
 173                                                 off_t         journal_size,
 174                                                 struct vnode *fsvp,
 175                                                 size_t        min_fs_block_size,
 176                                                 int32_t       flags,
 177                                                 int32_t       tbuffer_size,
 178                                                 void        (*flush)(void *arg),
 179                                                 void         *arg);
 180
 181 /*
 182  * Call journal_open() when mounting an existing file system
 183  * that has a previously created journal.  It will take care
 184  * of validating the journal and replaying it if necessary.
 185  *
 186  * See journal_create() for a description of the arguments.
 187  *
 188  * Returns a valid journal pointer of NULL if it runs into
 189  * trouble reading/playing back the journal.
 190  */
 191 journal  *journal_open(struct vnode *jvp,
 192                                            off_t         offset,
 193                                            off_t         journal_size,
 194                                            struct vnode *fsvp,
 195                                            size_t        min_fs_block_size,
 196                                            int32_t       flags,
 197                                            int32_t       tbuffer_size,
 198                                            void        (*flush)(void *arg),
 199                                            void         *arg);
 200
 201 /*
 202  * Call journal_close() just before your file system is unmounted.
 203  * It flushes any outstanding transactions and makes sure the
 204  * journal is in a consistent state.
 205  */
 206 void      journal_close(journal *journal);
 207
 208 /*
 209  * flags for journal_create/open.  only can use
 210  * the low 16 bits for flags because internal
 211  * bits go in the high 16.
 212  */
 213 #define JOURNAL_NO_GROUP_COMMIT   0x00000001
 214 #define JOURNAL_RESET             0x00000002
 215
 216 /*
 217  * Transaction related functions.
 218  *
 219  * Before you start modifying file system meta data, you
 220  * should call journal_start_transaction().  Then before
 221  * you modify each block, call journal_modify_block_start()
 222  * and when you're done, journal_modify_block_end().  When
 223  * you've modified the last block as part of a transaction,
 224  * call journal_end_transaction() to commit the changes.
 225  *
 226  * If you decide to abort the modifications to a block you
 227  * should call journal_modify_block_abort().
 228  *
 229  * If as part of a transaction you need want to throw out
 230  * any previous copies of a block (because it got deleted)
 231  * then call journal_kill_block().  This will mark it so
 232  * that the journal does not play it back (effectively
 233  * dropping it).
 234  */
 235 int   journal_start_transaction(journal *jnl);
 236 int   journal_modify_block_start(journal *jnl, struct buf *bp);
 237 int   journal_modify_block_abort(journal *jnl, struct buf *bp);
 238 int   journal_modify_block_end(journal *jnl, struct buf *bp);
 239 int   journal_kill_block(journal *jnl, struct buf *bp);
 240 int   journal_end_transaction(journal *jnl);
 241
 242 int   journal_active(journal *jnl);
 243 int   journal_flush(journal *jnl);
 244
 245 #endif /* __APPLE_API_UNSTABLE */
 246 #endif /* !_SYS_VFS_JOURNAL_H_ */