bsd/vfs/vfs_journal.h

   1
   2 /*
   3  * Copyright (c) 2000-2011 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 #include <sys/disk.h>
  46
  47
  48 typedef struct _blk_info {
  49     int32_t    bsize;
  50     union {
  51         int32_t    cksum;
  52         uint32_t   sequence_num;
  53     } b;
  54 } _blk_info;
  55
  56 typedef struct block_info {
  57     off_t       bnum;                // block # on the file system device
  58     union {
  59         _blk_info   bi;
  60         struct buf *bp;
  61     } u;
  62 } __attribute__((__packed__)) block_info;
  63
  64 typedef struct block_list_header {
  65     u_int16_t   max_blocks;          // max number of blocks in this chunk
  66     u_int16_t   num_blocks;          // number of valid block numbers in block_nums
  67     int32_t     bytes_used;          // how many bytes of this tbuffer are used
  68     int32_t     checksum;            // on-disk: checksum of this header and binfo[0]
  69     int32_t     flags;               // check-checksums, initial blhdr, etc
  70     block_info  binfo[1];            // so we can reference them by name
  71 } block_list_header;
  72
  73 #define BLHDR_CHECK_CHECKSUMS   0x0001
  74 #define BLHDR_FIRST_HEADER      0x0002
  75
  76
  77 struct journal;
  78
  79 struct jnl_trim_list {
  80         uint32_t        allocated_count;
  81         uint32_t        extent_count;
  82         dk_extent_t *extents;
  83 };
  84
  85 typedef void (*jnl_trim_callback_t)(void *arg, uint32_t extent_count, const dk_extent_t *extents);
  86
  87 typedef struct transaction {
  88     int                 tbuffer_size;  // in bytes
  89     char               *tbuffer;       // memory copy of the transaction
  90     block_list_header  *blhdr;         // points to the first byte of tbuffer
  91     int                 num_blhdrs;    // how many buffers we've allocated
  92     int                 total_bytes;   // total # of bytes in transaction
  93     int                 num_flushed;   // how many bytes have been flushed
  94     int                 num_killed;    // how many bytes were "killed"
  95     off_t               journal_start; // where in the journal this transaction starts
  96     off_t               journal_end;   // where in the journal this transaction ends
  97     struct journal     *jnl;           // ptr back to the journal structure
  98     struct transaction *next;          // list of tr's (either completed or to be free'd)
  99     uint32_t            sequence_num;
 100         struct jnl_trim_list trim;
 101     boolean_t           delayed_header_write;
 102 } transaction;
 103
 104
 105 /*
 106  * This is written to block zero of the journal and it
 107  * maintains overall state about the journal.
 108  */
 109 typedef struct journal_header {
 110     int32_t        magic;
 111     int32_t        endian;
 112     volatile off_t start;         // zero-based byte offset of the start of the first transaction
 113     volatile off_t end;           // zero-based byte offset of where free space begins
 114     off_t          size;          // size in bytes of the entire journal
 115     int32_t        blhdr_size;    // size in bytes of each block_list_header in the journal
 116     int32_t        checksum;
 117     int32_t        jhdr_size;     // block size (in bytes) of the journal header
 118     uint32_t       sequence_num;  // NEW FIELD: a monotonically increasing value assigned to all txn's
 119 } journal_header;
 120
 121 #define JOURNAL_HEADER_MAGIC  0x4a4e4c78   // 'JNLx'
 122 #define ENDIAN_MAGIC          0x12345678
 123
 124 //
 125 // we only checksum the original size of the journal_header to remain
 126 // backwards compatible.  the size of the original journal_heade is
 127 // everything up to the the sequence_num field, hence we use the
 128 // offsetof macro to calculate the size.
 129 //
 130 #define JOURNAL_HEADER_CKSUM_SIZE  (offsetof(struct journal_header, sequence_num))
 131
 132 #define OLD_JOURNAL_HEADER_MAGIC  0x4a484452   // 'JHDR'
 133
 134
 135 /*
 136  * In memory structure about the journal.
 137  */
 138 typedef struct journal {
 139     lck_mtx_t           jlock;             // protects the struct journal data
 140     lck_mtx_t           flock;             // serializes flushing of journal
 141         lck_rw_t            trim_lock;         // protects the async_trim field, below
 142
 143
 144     struct vnode       *jdev;              // vnode of the device where the journal lives
 145     off_t               jdev_offset;       // byte offset to the start of the journal
 146     const char         *jdev_name;
 147
 148     struct vnode       *fsdev;             // vnode of the file system device
 149
 150     void              (*flush)(void *arg); // fs callback to flush meta data blocks
 151     void               *flush_arg;         // arg that's passed to flush()
 152
 153     int32_t             flags;
 154     int32_t             tbuffer_size;      // default transaction buffer size
 155     boolean_t           flush_aborted;
 156     boolean_t           flushing;
 157     boolean_t           asyncIO;
 158     boolean_t           writing_header;
 159     boolean_t           write_header_failed;
 160
 161     struct jnl_trim_list *async_trim;      // extents to be trimmed by transaction being asynchronously flushed
 162     jnl_trim_callback_t trim_callback;
 163     void                                *trim_callback_arg;
 164
 165     char               *header_buf;        // in-memory copy of the journal header
 166     int32_t             header_buf_size;
 167     journal_header     *jhdr;              // points to the first byte of header_buf
 168
 169         uint32_t                saved_sequence_num;
 170         uint32_t                sequence_num;
 171
 172     off_t               max_read_size;
 173     off_t               max_write_size;
 174
 175     transaction        *cur_tr;            // for group-commit
 176     transaction        *completed_trs;     // out-of-order transactions that completed
 177     transaction        *active_tr;         // for nested transactions
 178     int32_t             nested_count;      // for nested transactions
 179     void               *owner;             // a ptr that's unique to the calling process
 180
 181     transaction        *tr_freeme;         // transaction structs that need to be free'd
 182
 183     volatile off_t      active_start;      // the active start that we only keep in memory
 184     lck_mtx_t           old_start_lock;    // protects the old_start
 185     volatile off_t      old_start[16];     // this is how we do lazy start update
 186
 187     int                 last_flush_err;    // last error from flushing the cache
 188 } journal;
 189
 190 /* internal-only journal flags (top 16 bits) */
 191 #define JOURNAL_CLOSE_PENDING     0x00010000
 192 #define JOURNAL_INVALID           0x00020000
 193 #define JOURNAL_FLUSHCACHE_ERR    0x00040000   // means we already printed this err
 194 #define JOURNAL_NEED_SWAP         0x00080000   // swap any data read from disk
 195 #define JOURNAL_DO_FUA_WRITES     0x00100000   // do force-unit-access writes
 196 #define JOURNAL_USE_UNMAP         0x00200000   // device supports UNMAP (TRIM)
 197
 198
 199 /* journal_open/create options are always in the low-16 bits */
 200 #define JOURNAL_OPTION_FLAGS_MASK 0x0000ffff
 201
 202 __BEGIN_DECLS
 203 /*
 204  * Prototypes.
 205  */
 206
 207 /*
 208  * Call journal_init() to initialize the journaling code (sets up lock attributes)
 209  */
 210 void      journal_init(void) __attribute__((section("__TEXT, initcode")));
 211
 212 /*
 213  * Call journal_create() to create a new journal.  You only
 214  * call this once, typically at file system creation time.
 215  *
 216  * The "jvp" argument is the vnode where the journal is written.
 217  * The journal starts at "offset" and is "journal_size" bytes long.
 218  *
 219  * The "fsvp" argument is the vnode of your file system.  It may be
 220  * the same as "jvp".
 221  *
 222  * The "min_fs_block_size" argument is the minimum block size
 223  * (in bytes) that the file system will ever write.  Typically
 224  * this is the block size of the file system (1k, 4k, etc) but
 225  * on HFS+ it is the minimum block size of the underlying device.
 226  *
 227  * The flags argument lets you disable group commit if you
 228  * want tighter guarantees on transactions (in exchange for
 229  * lower performance).
 230  *
 231  * The tbuffer_size is the size of the transaction buffer
 232  * used by the journal. If you specify zero, the journal code
 233  * will use a reasonable defaults.  The tbuffer_size should
 234  * be an integer multiple of the min_fs_block_size.
 235  *
 236  * Returns a valid journal pointer or NULL if one could not
 237  * be created.
 238  */
 239 journal *journal_create(struct vnode *jvp,
 240                                                 off_t         offset,
 241                                                 off_t         journal_size,
 242                                                 struct vnode *fsvp,
 243                                                 size_t        min_fs_block_size,
 244                                                 int32_t       flags,
 245                                                 int32_t       tbuffer_size,
 246                                                 void        (*flush)(void *arg),
 247                                                 void         *arg);
 248
 249 /*
 250  * Call journal_open() when mounting an existing file system
 251  * that has a previously created journal.  It will take care
 252  * of validating the journal and replaying it if necessary.
 253  *
 254  * See journal_create() for a description of the arguments.
 255  *
 256  * Returns a valid journal pointer of NULL if it runs into
 257  * trouble reading/playing back the journal.
 258  */
 259 journal  *journal_open(struct vnode *jvp,
 260                                            off_t         offset,
 261                                            off_t         journal_size,
 262                                            struct vnode *fsvp,
 263                                            size_t        min_fs_block_size,
 264                                            int32_t       flags,
 265                                            int32_t       tbuffer_size,
 266                                            void        (*flush)(void *arg),
 267                                            void         *arg);
 268
 269 /*
 270  * Test whether the journal is clean or not.  This is intended
 271  * to be used when you're mounting read-only.  If the journal
 272  * is not clean for some reason then you should not mount the
 273  * volume as your data structures may be in an unknown state.
 274  */
 275 int journal_is_clean(struct vnode *jvp,
 276                      off_t         offset,
 277                      off_t         journal_size,
 278                      struct vnode *fsvp,
 279                      size_t        min_fs_block_size);
 280
 281
 282 /*
 283  * Call journal_close() just before your file system is unmounted.
 284  * It flushes any outstanding transactions and makes sure the
 285  * journal is in a consistent state.
 286  */
 287 void      journal_close(journal *journalp);
 288
 289 /*
 290  * flags for journal_create/open.  only can use
 291  * the low 16 bits for flags because internal
 292  * bits go in the high 16.
 293  */
 294 #define JOURNAL_NO_GROUP_COMMIT   0x00000001
 295 #define JOURNAL_RESET             0x00000002
 296
 297 /*
 298  * Transaction related functions.
 299  *
 300  * Before you start modifying file system meta data, you
 301  * should call journal_start_transaction().  Then before
 302  * you modify each block, call journal_modify_block_start()
 303  * and when you're done, journal_modify_block_end().  When
 304  * you've modified the last block as part of a transaction,
 305  * call journal_end_transaction() to commit the changes.
 306  *
 307  * If you decide to abort the modifications to a block you
 308  * should call journal_modify_block_abort().
 309  *
 310  * If as part of a transaction you need want to throw out
 311  * any previous copies of a block (because it got deleted)
 312  * then call journal_kill_block().  This will mark it so
 313  * that the journal does not play it back (effectively
 314  * dropping it).
 315  *
 316  * journal_trim_add_extent() marks a range of bytes on the device which should
 317  * be trimmed (invalidated, unmapped).  journal_trim_remove_extent() marks a
 318  * range of bytes which should no longer be trimmed.  Accumulated extents
 319  * will be trimmed when the transaction is flushed to the on-disk journal.
 320  */
 321 int   journal_start_transaction(journal *jnl);
 322 int   journal_modify_block_start(journal *jnl, struct buf *bp);
 323 int   journal_modify_block_abort(journal *jnl, struct buf *bp);
 324 int   journal_modify_block_end(journal *jnl, struct buf *bp, void (*func)(struct buf *bp, void *arg), void *arg);
 325 int   journal_kill_block(journal *jnl, struct buf *bp);
 326 #ifdef BSD_KERNEL_PRIVATE
 327 int   journal_trim_add_extent(journal *jnl, uint64_t offset, uint64_t length);
 328 int   journal_trim_remove_extent(journal *jnl, uint64_t offset, uint64_t length);
 329 void  journal_trim_set_callback(journal *jnl, jnl_trim_callback_t callback, void *arg);
 330 #endif
 331 int   journal_end_transaction(journal *jnl);
 332
 333 int   journal_active(journal *jnl);
 334 int   journal_flush(journal *jnl, boolean_t wait_for_IO);
 335 void *journal_owner(journal *jnl);    // compare against current_thread()
 336 int   journal_uses_fua(journal *jnl);
 337
 338
 339 /*
 340  * Relocate the journal.
 341  *
 342  * You provide the new starting offset and size for the journal. You may
 343  * optionally provide a new tbuffer_size; passing zero defaults to not
 344  * changing the tbuffer size except as needed to fit within the new journal
 345  * size.
 346  *
 347  * You must have already started a transaction. The transaction may contain
 348  * modified blocks (such as those needed to deallocate the old journal,
 349  * allocate the new journal, and update the location and size of the journal
 350  * in filesystem-private structures). Any transactions prior to the active
 351  * transaction will be flushed to the old journal. The new journal will be
 352  * initialized, and the blocks from the active transaction will be written to
 353  * the new journal. The caller will need to update the structures that
 354  * identify the location and size of the journal from the callback routine.
 355  */
 356 int journal_relocate(journal *jnl, off_t offset, off_t journal_size, int32_t tbuffer_size,
 357         errno_t (*callback)(void *), void *callback_arg);
 358
 359 __END_DECLS
 360
 361 #endif /* __APPLE_API_UNSTABLE */
 362 #endif /* !_SYS_VFS_JOURNAL_H_ */