bsd/vfs/vfs_journal.h

   1
   2 /*
   3  * Copyright (c) 2000-2002 Apple Computer, Inc. All rights reserved.
   4  *
   5  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   6  *
   7  * This file contains Original Code and/or Modifications of Original Code
   8  * as defined in and that are subject to the Apple Public Source License
   9  * Version 2.0 (the 'License'). You may not use this file except in
  10  * compliance with the License. The rights granted to you under the License
  11  * may not be used to create, or enable the creation or redistribution of,
  12  * unlawful or unlicensed copies of an Apple operating system, or to
  13  * circumvent, violate, or enable the circumvention or violation of, any
  14  * terms of an Apple operating system software license agreement.
  15  *
  16  * Please obtain a copy of the License at
  17  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  18  *
  19  * The Original Code and all software distributed under the License are
  20  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  21  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  22  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  23  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  24  * Please see the License for the specific language governing rights and
  25  * limitations under the License.
  26  *
  27  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  28  */
  29 /*
  30  * This header contains the structures and function prototypes
  31  * for the vfs journaling code.  The data types are not meant
  32  * to be modified by user code.  Just use the functions and do
  33  * not mess around with the structs.
  34  */
  35 #ifndef _SYS_VFS_JOURNAL_H_
  36 #define _SYS_VFS_JOURNAL_H_
  37
  38 #include <sys/appleapiopts.h>
  39 #include <sys/cdefs.h>
  40
  41 #ifdef __APPLE_API_UNSTABLE
  42
  43 #include <sys/types.h>
  44 #include <kern/locks.h>
  45
  46 typedef struct block_info {
  47     off_t       bnum;                // block # on the file system device
  48     size_t      bsize;               // in bytes
  49     struct buf *bp;
  50 } block_info;
  51
  52 typedef struct block_list_header {
  53     u_int16_t   max_blocks;          // max number of blocks in this chunk
  54     u_int16_t   num_blocks;          // number of valid block numbers in block_nums
  55     int32_t     bytes_used;          // how many bytes of this tbuffer are used
  56     int32_t     checksum;            // on-disk: checksum of this header and binfo[0]
  57     int32_t     pad;                 // pad out to 16 bytes
  58     block_info  binfo[1];            // so we can reference them by name
  59 } block_list_header;
  60
  61
  62 struct journal;
  63
  64 typedef struct transaction {
  65     int                 tbuffer_size;  // in bytes
  66     char               *tbuffer;       // memory copy of the transaction
  67     block_list_header  *blhdr;         // points to the first byte of tbuffer
  68     int                 num_blhdrs;    // how many buffers we've allocated
  69     int                 total_bytes;   // total # of bytes in transaction
  70     int                 num_flushed;   // how many bytes have been flushed
  71     int                 num_killed;    // how many bytes were "killed"
  72     off_t               journal_start; // where in the journal this transaction starts
  73     off_t               journal_end;   // where in the journal this transaction ends
  74     struct journal     *jnl;           // ptr back to the journal structure
  75     struct transaction *next;          // list of tr's (either completed or to be free'd)
  76 } transaction;
  77
  78
  79 /*
  80  * This is written to block zero of the journal and it
  81  * maintains overall state about the journal.
  82  */
  83 typedef struct journal_header {
  84     int32_t        magic;
  85     int32_t        endian;
  86     volatile off_t start;         // zero-based byte offset of the start of the first transaction
  87     volatile off_t end;           // zero-based byte offset of where free space begins
  88     off_t          size;          // size in bytes of the entire journal
  89     int32_t        blhdr_size;    // size in bytes of each block_list_header in the journal
  90     int32_t        checksum;
  91     int32_t        jhdr_size;     // block size (in bytes) of the journal header
  92 } journal_header;
  93
  94 #define JOURNAL_HEADER_MAGIC  0x4a4e4c78   // 'JNLx'
  95 #define ENDIAN_MAGIC          0x12345678
  96
  97 #define OLD_JOURNAL_HEADER_MAGIC  0x4a484452   // 'JHDR'
  98
  99
 100 /*
 101  * In memory structure about the journal.
 102  */
 103 typedef struct journal {
 104     lck_mtx_t           jlock;             // protects the struct journal data
 105
 106     struct vnode       *jdev;              // vnode of the device where the journal lives
 107     off_t               jdev_offset;       // byte offset to the start of the journal
 108
 109     struct vnode       *fsdev;             // vnode of the file system device
 110
 111     void              (*flush)(void *arg); // fs callback to flush meta data blocks
 112     void               *flush_arg;         // arg that's passed to flush()
 113
 114     int32_t             flags;
 115     int32_t             tbuffer_size;      // default transaction buffer size
 116
 117     char               *header_buf;        // in-memory copy of the journal header
 118     journal_header     *jhdr;              // points to the first byte of header_buf
 119
 120     transaction        *cur_tr;            // for group-commit
 121     transaction        *completed_trs;     // out-of-order transactions that completed
 122     transaction        *active_tr;         // for nested transactions
 123     int32_t             nested_count;      // for nested transactions
 124     void               *owner;             // a ptr that's unique to the calling process
 125
 126     transaction        *tr_freeme;         // transaction structs that need to be free'd
 127
 128     volatile off_t      active_start;      // the active start that we only keep in memory
 129     lck_mtx_t           old_start_lock;    // protects the old_start
 130     volatile off_t      old_start[16];     // this is how we do lazy start update
 131
 132     int                 last_flush_err;    // last error from flushing the cache
 133 } journal;
 134
 135 /* internal-only journal flags (top 16 bits) */
 136 #define JOURNAL_CLOSE_PENDING     0x00010000
 137 #define JOURNAL_INVALID           0x00020000
 138 #define JOURNAL_FLUSHCACHE_ERR    0x00040000   // means we already printed this err
 139 #define JOURNAL_NEED_SWAP         0x00080000   // swap any data read from disk
 140
 141 /* journal_open/create options are always in the low-16 bits */
 142 #define JOURNAL_OPTION_FLAGS_MASK 0x0000ffff
 143
 144 __BEGIN_DECLS
 145 /*
 146  * Prototypes.
 147  */
 148
 149 /*
 150  * Call journal_init() to initialize the journaling code (sets up lock attributes)
 151  */
 152 void      journal_init(void);
 153
 154 /*
 155  * Call journal_create() to create a new journal.  You only
 156  * call this once, typically at file system creation time.
 157  *
 158  * The "jvp" argument is the vnode where the journal is written.
 159  * The journal starts at "offset" and is "journal_size" bytes long.
 160  *
 161  * The "fsvp" argument is the vnode of your file system.  It may be
 162  * the same as "jvp".
 163  *
 164  * The "min_fs_block_size" argument is the minimum block size
 165  * (in bytes) that the file system will ever write.  Typically
 166  * this is the block size of the file system (1k, 4k, etc) but
 167  * on HFS+ it is the minimum block size of the underlying device.
 168  *
 169  * The flags argument lets you disable group commit if you
 170  * want tighter guarantees on transactions (in exchange for
 171  * lower performance).
 172  *
 173  * The tbuffer_size is the size of the transaction buffer
 174  * used by the journal. If you specify zero, the journal code
 175  * will use a reasonable defaults.  The tbuffer_size should
 176  * be an integer multiple of the min_fs_block_size.
 177  *
 178  * Returns a valid journal pointer or NULL if one could not
 179  * be created.
 180  */
 181 journal *journal_create(struct vnode *jvp,
 182                                                 off_t         offset,
 183                                                 off_t         journal_size,
 184                                                 struct vnode *fsvp,
 185                                                 size_t        min_fs_block_size,
 186                                                 int32_t       flags,
 187                                                 int32_t       tbuffer_size,
 188                                                 void        (*flush)(void *arg),
 189                                                 void         *arg);
 190
 191 /*
 192  * Call journal_open() when mounting an existing file system
 193  * that has a previously created journal.  It will take care
 194  * of validating the journal and replaying it if necessary.
 195  *
 196  * See journal_create() for a description of the arguments.
 197  *
 198  * Returns a valid journal pointer of NULL if it runs into
 199  * trouble reading/playing back the journal.
 200  */
 201 journal  *journal_open(struct vnode *jvp,
 202                                            off_t         offset,
 203                                            off_t         journal_size,
 204                                            struct vnode *fsvp,
 205                                            size_t        min_fs_block_size,
 206                                            int32_t       flags,
 207                                            int32_t       tbuffer_size,
 208                                            void        (*flush)(void *arg),
 209                                            void         *arg);
 210
 211 /*
 212  * Test whether the journal is clean or not.  This is intended
 213  * to be used when you're mounting read-only.  If the journal
 214  * is not clean for some reason then you should not mount the
 215  * volume as your data structures may be in an unknown state.
 216  */
 217 int journal_is_clean(struct vnode *jvp,
 218                      off_t         offset,
 219                      off_t         journal_size,
 220                      struct vnode *fsvp,
 221                      size_t        min_fs_block_size);
 222
 223
 224 /*
 225  * Call journal_close() just before your file system is unmounted.
 226  * It flushes any outstanding transactions and makes sure the
 227  * journal is in a consistent state.
 228  */
 229 void      journal_close(journal *journalp);
 230
 231 /*
 232  * flags for journal_create/open.  only can use
 233  * the low 16 bits for flags because internal
 234  * bits go in the high 16.
 235  */
 236 #define JOURNAL_NO_GROUP_COMMIT   0x00000001
 237 #define JOURNAL_RESET             0x00000002
 238
 239 /*
 240  * Transaction related functions.
 241  *
 242  * Before you start modifying file system meta data, you
 243  * should call journal_start_transaction().  Then before
 244  * you modify each block, call journal_modify_block_start()
 245  * and when you're done, journal_modify_block_end().  When
 246  * you've modified the last block as part of a transaction,
 247  * call journal_end_transaction() to commit the changes.
 248  *
 249  * If you decide to abort the modifications to a block you
 250  * should call journal_modify_block_abort().
 251  *
 252  * If as part of a transaction you need want to throw out
 253  * any previous copies of a block (because it got deleted)
 254  * then call journal_kill_block().  This will mark it so
 255  * that the journal does not play it back (effectively
 256  * dropping it).
 257  */
 258 int   journal_start_transaction(journal *jnl);
 259 int   journal_modify_block_start(journal *jnl, struct buf *bp);
 260 int   journal_modify_block_abort(journal *jnl, struct buf *bp);
 261 int   journal_modify_block_end(journal *jnl, struct buf *bp);
 262 int   journal_kill_block(journal *jnl, struct buf *bp);
 263 int   journal_end_transaction(journal *jnl);
 264
 265 int   journal_active(journal *jnl);
 266 int   journal_flush(journal *jnl);
 267 void *journal_owner(journal *jnl);    // compare against current_thread()
 268
 269 __END_DECLS
 270
 271 #endif /* __APPLE_API_UNSTABLE */
 272 #endif /* !_SYS_VFS_JOURNAL_H_ */