bsd/vfs/vfs_journal.h

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