bsd/vfs/vfs_journal.h

   1
   2 /*
   3  * Copyright (c) 2000-2014 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     uint32_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         boolean_t       flush_on_completion; //flush transaction immediately upon txn end.
 103 } transaction;
 104
 105
 106 /*
 107  * This is written to block zero of the journal and it
 108  * maintains overall state about the journal.
 109  */
 110 typedef struct journal_header {
 111     int32_t        magic;
 112     int32_t        endian;
 113     volatile off_t start;         // zero-based byte offset of the start of the first transaction
 114     volatile off_t end;           // zero-based byte offset of where free space begins
 115     off_t          size;          // size in bytes of the entire journal
 116     int32_t        blhdr_size;    // size in bytes of each block_list_header in the journal
 117     uint32_t        checksum;
 118     int32_t        jhdr_size;     // block size (in bytes) of the journal header
 119     uint32_t       sequence_num;  // NEW FIELD: a monotonically increasing value assigned to all txn's
 120 } journal_header;
 121
 122 #define JOURNAL_HEADER_MAGIC  0x4a4e4c78   // 'JNLx'
 123 #define ENDIAN_MAGIC          0x12345678
 124
 125 //
 126 // we only checksum the original size of the journal_header to remain
 127 // backwards compatible.  the size of the original journal_heade is
 128 // everything up to the the sequence_num field, hence we use the
 129 // offsetof macro to calculate the size.
 130 //
 131 #define JOURNAL_HEADER_CKSUM_SIZE  (offsetof(struct journal_header, sequence_num))
 132
 133 #define OLD_JOURNAL_HEADER_MAGIC  0x4a484452   // 'JHDR'
 134
 135
 136 /*
 137  * In memory structure about the journal.
 138  */
 139 typedef struct journal {
 140     lck_mtx_t           jlock;             // protects the struct journal data
 141     lck_mtx_t           flock;             // serializes flushing of journal
 142         lck_rw_t            trim_lock;         // protects the async_trim field, below
 143
 144
 145     struct vnode       *jdev;              // vnode of the device where the journal lives
 146     off_t               jdev_offset;       // byte offset to the start of the journal
 147     const char         *jdev_name;
 148
 149     struct vnode       *fsdev;             // vnode of the file system device
 150     struct mount       *fsmount;           // mount of the file system
 151
 152     void              (*flush)(void *arg); // fs callback to flush meta data blocks
 153     void               *flush_arg;         // arg that's passed to flush()
 154
 155     int32_t             flags;
 156     int32_t             tbuffer_size;      // default transaction buffer size
 157     boolean_t           flush_aborted;
 158     boolean_t           flushing;
 159     boolean_t           asyncIO;
 160     boolean_t           writing_header;
 161     boolean_t           write_header_failed;
 162
 163     struct jnl_trim_list *async_trim;      // extents to be trimmed by transaction being asynchronously flushed
 164     jnl_trim_callback_t trim_callback;
 165     void                                *trim_callback_arg;
 166
 167     char               *header_buf;        // in-memory copy of the journal header
 168     int32_t             header_buf_size;
 169     journal_header     *jhdr;              // points to the first byte of header_buf
 170
 171         uint32_t                saved_sequence_num;
 172         uint32_t                sequence_num;
 173
 174     off_t               max_read_size;
 175     off_t               max_write_size;
 176
 177     transaction        *cur_tr;            // for group-commit
 178     transaction        *completed_trs;     // out-of-order transactions that completed
 179     transaction        *active_tr;         // for nested transactions
 180     int32_t             nested_count;      // for nested transactions
 181     void               *owner;             // a ptr that's unique to the calling process
 182
 183     transaction        *tr_freeme;         // transaction structs that need to be free'd
 184
 185     volatile off_t      active_start;      // the active start that we only keep in memory
 186     lck_mtx_t           old_start_lock;    // protects the old_start
 187     volatile off_t      old_start[16];     // this is how we do lazy start update
 188
 189     int                 last_flush_err;    // last error from flushing the cache
 190     uint32_t            flush_counter;     // a monotonically increasing value assigned on track cache flush
 191 } journal;
 192
 193 /* internal-only journal flags (top 16 bits) */
 194 #define JOURNAL_CLOSE_PENDING     0x00010000
 195 #define JOURNAL_INVALID           0x00020000
 196 #define JOURNAL_FLUSHCACHE_ERR    0x00040000   // means we already printed this err
 197 #define JOURNAL_NEED_SWAP         0x00080000   // swap any data read from disk
 198 #define JOURNAL_DO_FUA_WRITES     0x00100000   // do force-unit-access writes
 199 #define JOURNAL_USE_UNMAP         0x00200000   // device supports UNMAP (TRIM)
 200 #define JOURNAL_FEATURE_BARRIER   0x00400000   // device supports barrier-only flush
 201
 202
 203 /* journal_open/create options are always in the low-16 bits */
 204 #define JOURNAL_OPTION_FLAGS_MASK 0x0000ffff
 205
 206 __BEGIN_DECLS
 207 /*
 208  * Prototypes.
 209  */
 210
 211 /*
 212  * Call journal_init() to initialize the journaling code (sets up lock attributes)
 213  */
 214 void      journal_init(void);
 215
 216 /*
 217  * Call journal_create() to create a new journal.  You only
 218  * call this once, typically at file system creation time.
 219  *
 220  * The "jvp" argument is the vnode where the journal is written.
 221  * The journal starts at "offset" and is "journal_size" bytes long.
 222  *
 223  * The "fsvp" argument is the vnode of your file system.  It may be
 224  * the same as "jvp".
 225  *
 226  * The "min_fs_block_size" argument is the minimum block size
 227  * (in bytes) that the file system will ever write.  Typically
 228  * this is the block size of the file system (1k, 4k, etc) but
 229  * on HFS+ it is the minimum block size of the underlying device.
 230  *
 231  * The flags argument lets you disable group commit if you
 232  * want tighter guarantees on transactions (in exchange for
 233  * lower performance).
 234  *
 235  * The tbuffer_size is the size of the transaction buffer
 236  * used by the journal. If you specify zero, the journal code
 237  * will use a reasonable defaults.  The tbuffer_size should
 238  * be an integer multiple of the min_fs_block_size.
 239  *
 240  * Returns a valid journal pointer or NULL if one could not
 241  * be created.
 242  */
 243 journal *journal_create(struct vnode *jvp,
 244                                                 off_t         offset,
 245                                                 off_t         journal_size,
 246                                                 struct vnode *fsvp,
 247                                                 size_t        min_fs_block_size,
 248                                                 int32_t       flags,
 249                                                 int32_t       tbuffer_size,
 250                                                 void        (*flush)(void *arg),
 251                                                 void         *arg,
 252                                                 struct mount *fsmount);
 253
 254 /*
 255  * Call journal_open() when mounting an existing file system
 256  * that has a previously created journal.  It will take care
 257  * of validating the journal and replaying it if necessary.
 258  *
 259  * See journal_create() for a description of the arguments.
 260  *
 261  * Returns a valid journal pointer of NULL if it runs into
 262  * trouble reading/playing back the journal.
 263  */
 264 journal  *journal_open(struct vnode *jvp,
 265                                            off_t         offset,
 266                                            off_t         journal_size,
 267                                            struct vnode *fsvp,
 268                                            size_t        min_fs_block_size,
 269                                            int32_t       flags,
 270                                            int32_t       tbuffer_size,
 271                                            void        (*flush)(void *arg),
 272                                            void         *arg,
 273                                            struct mount *fsmount);
 274
 275 /*
 276  * Test whether the journal is clean or not.  This is intended
 277  * to be used when you're mounting read-only.  If the journal
 278  * is not clean for some reason then you should not mount the
 279  * volume as your data structures may be in an unknown state.
 280  */
 281 int journal_is_clean(struct vnode *jvp,
 282                      off_t         offset,
 283                      off_t         journal_size,
 284                      struct vnode *fsvp,
 285                      size_t        min_fs_block_size);
 286
 287
 288 /*
 289  * Call journal_close() just before your file system is unmounted.
 290  * It flushes any outstanding transactions and makes sure the
 291  * journal is in a consistent state.
 292  */
 293 void      journal_close(journal *journalp);
 294
 295 /*
 296  * flags for journal_create/open.  only can use
 297  * the low 16 bits for flags because internal
 298  * bits go in the high 16.
 299  */
 300 #define JOURNAL_NO_GROUP_COMMIT   0x00000001
 301 #define JOURNAL_RESET             0x00000002
 302
 303 /*
 304  * Transaction related functions.
 305  *
 306  * Before you start modifying file system meta data, you
 307  * should call journal_start_transaction().  Then before
 308  * you modify each block, call journal_modify_block_start()
 309  * and when you're done, journal_modify_block_end().  When
 310  * you've modified the last block as part of a transaction,
 311  * call journal_end_transaction() to commit the changes.
 312  *
 313  * If you decide to abort the modifications to a block you
 314  * should call journal_modify_block_abort().
 315  *
 316  * If as part of a transaction you need want to throw out
 317  * any previous copies of a block (because it got deleted)
 318  * then call journal_kill_block().  This will mark it so
 319  * that the journal does not play it back (effectively
 320  * dropping it).
 321  *
 322  * journal_trim_add_extent() marks a range of bytes on the device which should
 323  * be trimmed (invalidated, unmapped).  journal_trim_remove_extent() marks a
 324  * range of bytes which should no longer be trimmed.  Accumulated extents
 325  * will be trimmed when the transaction is flushed to the on-disk journal.
 326  */
 327 int   journal_start_transaction(journal *jnl);
 328 int   journal_modify_block_start(journal *jnl, struct buf *bp);
 329 int   journal_modify_block_abort(journal *jnl, struct buf *bp);
 330 int   journal_modify_block_end(journal *jnl, struct buf *bp, void (*func)(struct buf *bp, void *arg), void *arg);
 331 int   journal_kill_block(journal *jnl, struct buf *bp);
 332 #ifdef BSD_KERNEL_PRIVATE
 333 int   journal_trim_add_extent(journal *jnl, uint64_t offset, uint64_t length);
 334 int   journal_trim_remove_extent(journal *jnl, uint64_t offset, uint64_t length);
 335 void  journal_trim_set_callback(journal *jnl, jnl_trim_callback_t callback, void *arg);
 336 int   journal_trim_extent_overlap (journal *jnl, uint64_t offset, uint64_t length, uint64_t *end);
 337 /* Mark state in the journal that requests an immediate journal flush upon txn completion */
 338 int   journal_request_immediate_flush (journal *jnl);
 339 #endif
 340 int   journal_end_transaction(journal *jnl);
 341
 342 int   journal_active(journal *jnl);
 343
 344 typedef enum journal_flush_options {
 345         JOURNAL_WAIT_FOR_IO       = 0x01,   // Flush journal and metadata blocks, wait for async IO to complete.
 346         JOURNAL_FLUSH_FULL        = 0x02,   // Flush track cache to media
 347 } journal_flush_options_t;
 348
 349 int   journal_flush(journal *jnl, journal_flush_options_t options);
 350 void *journal_owner(journal *jnl);    // compare against current_thread()
 351 int   journal_uses_fua(journal *jnl);
 352 void  journal_lock(journal *jnl);
 353 void  journal_unlock(journal *jnl);
 354
 355
 356 /*
 357  * Relocate the journal.
 358  *
 359  * You provide the new starting offset and size for the journal. You may
 360  * optionally provide a new tbuffer_size; passing zero defaults to not
 361  * changing the tbuffer size except as needed to fit within the new journal
 362  * size.
 363  *
 364  * You must have already started a transaction. The transaction may contain
 365  * modified blocks (such as those needed to deallocate the old journal,
 366  * allocate the new journal, and update the location and size of the journal
 367  * in filesystem-private structures). Any transactions prior to the active
 368  * transaction will be flushed to the old journal. The new journal will be
 369  * initialized, and the blocks from the active transaction will be written to
 370  * the new journal. The caller will need to update the structures that
 371  * identify the location and size of the journal from the callback routine.
 372  */
 373 int journal_relocate(journal *jnl, off_t offset, off_t journal_size, int32_t tbuffer_size,
 374         errno_t (*callback)(void *), void *callback_arg);
 375
 376 uint32_t journal_current_txn(journal *jnl);
 377
 378 __END_DECLS
 379
 380 #endif /* __APPLE_API_UNSTABLE */
 381 #endif /* !_SYS_VFS_JOURNAL_H_ */