bsd/vfs/vfs_journal.h

   1
   2 /*
   3  * Copyright (c) 2000-2007 Apple 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 _blk_info {
  47     int32_t    bsize;
  48     union {
  49         int32_t    cksum;
  50         uint32_t   sequence_num;
  51     } b;
  52 } _blk_info;
  53
  54 typedef struct block_info {
  55     off_t       bnum;                // block # on the file system device
  56     union {
  57         _blk_info   bi;
  58         struct buf *bp;
  59     } u;
  60 } __attribute__((__packed__)) block_info;
  61
  62 typedef struct block_list_header {
  63     u_int16_t   max_blocks;          // max number of blocks in this chunk
  64     u_int16_t   num_blocks;          // number of valid block numbers in block_nums
  65     int32_t     bytes_used;          // how many bytes of this tbuffer are used
  66     int32_t     checksum;            // on-disk: checksum of this header and binfo[0]
  67     int32_t     flags;               // check-checksums, initial blhdr, etc
  68     block_info  binfo[1];            // so we can reference them by name
  69 } block_list_header;
  70
  71 #define BLHDR_CHECK_CHECKSUMS   0x0001
  72 #define BLHDR_FIRST_HEADER      0x0002
  73
  74
  75 struct journal;
  76
  77 typedef struct transaction {
  78     int                 tbuffer_size;  // in bytes
  79     char               *tbuffer;       // memory copy of the transaction
  80     block_list_header  *blhdr;         // points to the first byte of tbuffer
  81     int                 num_blhdrs;    // how many buffers we've allocated
  82     int                 total_bytes;   // total # of bytes in transaction
  83     int                 num_flushed;   // how many bytes have been flushed
  84     int                 num_killed;    // how many bytes were "killed"
  85     off_t               journal_start; // where in the journal this transaction starts
  86     off_t               journal_end;   // where in the journal this transaction ends
  87     struct journal     *jnl;           // ptr back to the journal structure
  88     struct transaction *next;          // list of tr's (either completed or to be free'd)
  89     uint32_t            sequence_num;
  90 } transaction;
  91
  92
  93 /*
  94  * This is written to block zero of the journal and it
  95  * maintains overall state about the journal.
  96  */
  97 typedef struct journal_header {
  98     int32_t        magic;
  99     int32_t        endian;
 100     volatile off_t start;         // zero-based byte offset of the start of the first transaction
 101     volatile off_t end;           // zero-based byte offset of where free space begins
 102     off_t          size;          // size in bytes of the entire journal
 103     int32_t        blhdr_size;    // size in bytes of each block_list_header in the journal
 104     int32_t        checksum;
 105     int32_t        jhdr_size;     // block size (in bytes) of the journal header
 106     uint32_t       sequence_num;  // NEW FIELD: a monotonically increasing value assigned to all txn's
 107 } journal_header;
 108
 109 #define JOURNAL_HEADER_MAGIC  0x4a4e4c78   // 'JNLx'
 110 #define ENDIAN_MAGIC          0x12345678
 111
 112 //
 113 // we only checksum the original size of the journal_header to remain
 114 // backwards compatible.  the size of the original journal_heade is
 115 // everything up to the the sequence_num field, hence we use the
 116 // offsetof macro to calculate the size.
 117 //
 118 #define JOURNAL_HEADER_CKSUM_SIZE  (offsetof(struct journal_header, sequence_num))
 119
 120 #define OLD_JOURNAL_HEADER_MAGIC  0x4a484452   // 'JHDR'
 121
 122
 123 /*
 124  * In memory structure about the journal.
 125  */
 126 typedef struct journal {
 127     lck_mtx_t           jlock;             // protects the struct journal data
 128
 129     struct vnode       *jdev;              // vnode of the device where the journal lives
 130     off_t               jdev_offset;       // byte offset to the start of the journal
 131     const char         *jdev_name;
 132
 133     struct vnode       *fsdev;             // vnode of the file system device
 134
 135     void              (*flush)(void *arg); // fs callback to flush meta data blocks
 136     void               *flush_arg;         // arg that's passed to flush()
 137
 138     int32_t             flags;
 139     int32_t             tbuffer_size;      // default transaction buffer size
 140
 141     char               *header_buf;        // in-memory copy of the journal header
 142     int32_t             header_buf_size;
 143     journal_header     *jhdr;              // points to the first byte of header_buf
 144
 145     off_t               max_read_size;
 146     off_t               max_write_size;
 147
 148     transaction        *cur_tr;            // for group-commit
 149     transaction        *completed_trs;     // out-of-order transactions that completed
 150     transaction        *active_tr;         // for nested transactions
 151     int32_t             nested_count;      // for nested transactions
 152     void               *owner;             // a ptr that's unique to the calling process
 153
 154     transaction        *tr_freeme;         // transaction structs that need to be free'd
 155
 156     volatile off_t      active_start;      // the active start that we only keep in memory
 157     lck_mtx_t           old_start_lock;    // protects the old_start
 158     volatile off_t      old_start[16];     // this is how we do lazy start update
 159
 160     int                 last_flush_err;    // last error from flushing the cache
 161 } journal;
 162
 163 /* internal-only journal flags (top 16 bits) */
 164 #define JOURNAL_CLOSE_PENDING     0x00010000
 165 #define JOURNAL_INVALID           0x00020000
 166 #define JOURNAL_FLUSHCACHE_ERR    0x00040000   // means we already printed this err
 167 #define JOURNAL_NEED_SWAP         0x00080000   // swap any data read from disk
 168 #define JOURNAL_DO_FUA_WRITES     0x00100000   // do force-unit-access writes
 169
 170 /* journal_open/create options are always in the low-16 bits */
 171 #define JOURNAL_OPTION_FLAGS_MASK 0x0000ffff
 172
 173 __BEGIN_DECLS
 174 /*
 175  * Prototypes.
 176  */
 177
 178 /*
 179  * Call journal_init() to initialize the journaling code (sets up lock attributes)
 180  */
 181 void      journal_init(void) __attribute__((section("__TEXT, initcode")));
 182
 183 /*
 184  * Call journal_create() to create a new journal.  You only
 185  * call this once, typically at file system creation time.
 186  *
 187  * The "jvp" argument is the vnode where the journal is written.
 188  * The journal starts at "offset" and is "journal_size" bytes long.
 189  *
 190  * The "fsvp" argument is the vnode of your file system.  It may be
 191  * the same as "jvp".
 192  *
 193  * The "min_fs_block_size" argument is the minimum block size
 194  * (in bytes) that the file system will ever write.  Typically
 195  * this is the block size of the file system (1k, 4k, etc) but
 196  * on HFS+ it is the minimum block size of the underlying device.
 197  *
 198  * The flags argument lets you disable group commit if you
 199  * want tighter guarantees on transactions (in exchange for
 200  * lower performance).
 201  *
 202  * The tbuffer_size is the size of the transaction buffer
 203  * used by the journal. If you specify zero, the journal code
 204  * will use a reasonable defaults.  The tbuffer_size should
 205  * be an integer multiple of the min_fs_block_size.
 206  *
 207  * Returns a valid journal pointer or NULL if one could not
 208  * be created.
 209  */
 210 journal *journal_create(struct vnode *jvp,
 211                                                 off_t         offset,
 212                                                 off_t         journal_size,
 213                                                 struct vnode *fsvp,
 214                                                 size_t        min_fs_block_size,
 215                                                 int32_t       flags,
 216                                                 int32_t       tbuffer_size,
 217                                                 void        (*flush)(void *arg),
 218                                                 void         *arg);
 219
 220 /*
 221  * Call journal_open() when mounting an existing file system
 222  * that has a previously created journal.  It will take care
 223  * of validating the journal and replaying it if necessary.
 224  *
 225  * See journal_create() for a description of the arguments.
 226  *
 227  * Returns a valid journal pointer of NULL if it runs into
 228  * trouble reading/playing back the journal.
 229  */
 230 journal  *journal_open(struct vnode *jvp,
 231                                            off_t         offset,
 232                                            off_t         journal_size,
 233                                            struct vnode *fsvp,
 234                                            size_t        min_fs_block_size,
 235                                            int32_t       flags,
 236                                            int32_t       tbuffer_size,
 237                                            void        (*flush)(void *arg),
 238                                            void         *arg);
 239
 240 /*
 241  * Test whether the journal is clean or not.  This is intended
 242  * to be used when you're mounting read-only.  If the journal
 243  * is not clean for some reason then you should not mount the
 244  * volume as your data structures may be in an unknown state.
 245  */
 246 int journal_is_clean(struct vnode *jvp,
 247                      off_t         offset,
 248                      off_t         journal_size,
 249                      struct vnode *fsvp,
 250                      size_t        min_fs_block_size);
 251
 252
 253 /*
 254  * Call journal_close() just before your file system is unmounted.
 255  * It flushes any outstanding transactions and makes sure the
 256  * journal is in a consistent state.
 257  */
 258 void      journal_close(journal *journalp);
 259
 260 /*
 261  * flags for journal_create/open.  only can use
 262  * the low 16 bits for flags because internal
 263  * bits go in the high 16.
 264  */
 265 #define JOURNAL_NO_GROUP_COMMIT   0x00000001
 266 #define JOURNAL_RESET             0x00000002
 267
 268 /*
 269  * Transaction related functions.
 270  *
 271  * Before you start modifying file system meta data, you
 272  * should call journal_start_transaction().  Then before
 273  * you modify each block, call journal_modify_block_start()
 274  * and when you're done, journal_modify_block_end().  When
 275  * you've modified the last block as part of a transaction,
 276  * call journal_end_transaction() to commit the changes.
 277  *
 278  * If you decide to abort the modifications to a block you
 279  * should call journal_modify_block_abort().
 280  *
 281  * If as part of a transaction you need want to throw out
 282  * any previous copies of a block (because it got deleted)
 283  * then call journal_kill_block().  This will mark it so
 284  * that the journal does not play it back (effectively
 285  * dropping it).
 286  */
 287 int   journal_start_transaction(journal *jnl);
 288 int   journal_modify_block_start(journal *jnl, struct buf *bp);
 289 int   journal_modify_block_abort(journal *jnl, struct buf *bp);
 290 int   journal_modify_block_end(journal *jnl, struct buf *bp, void (*func)(struct buf *bp, void *arg), void *arg);
 291 int   journal_kill_block(journal *jnl, struct buf *bp);
 292 int   journal_end_transaction(journal *jnl);
 293
 294 int   journal_active(journal *jnl);
 295 int   journal_flush(journal *jnl);
 296 void *journal_owner(journal *jnl);    // compare against current_thread()
 297 int   journal_uses_fua(journal *jnl);
 298
 299
 300 /*
 301  * Relocate the journal.
 302  *
 303  * You provide the new starting offset and size for the journal. You may
 304  * optionally provide a new tbuffer_size; passing zero defaults to not
 305  * changing the tbuffer size except as needed to fit within the new journal
 306  * size.
 307  *
 308  * You must have already started a transaction. The transaction may contain
 309  * modified blocks (such as those needed to deallocate the old journal,
 310  * allocate the new journal, and update the location and size of the journal
 311  * in filesystem-private structures). Any transactions prior to the active
 312  * transaction will be flushed to the old journal. The new journal will be
 313  * initialized, and the blocks from the active transaction will be written to
 314  * the new journal. The caller will need to update the structures that
 315  * identify the location and size of the journal from the callback routine.
 316  */
 317 int journal_relocate(journal *jnl, off_t offset, off_t journal_size, int32_t tbuffer_size,
 318         errno_t (*callback)(void *), void *callback_arg);
 319
 320 __END_DECLS
 321
 322 #endif /* __APPLE_API_UNSTABLE */
 323 #endif /* !_SYS_VFS_JOURNAL_H_ */