include/linux/jbd.h

   1 /*
   2  * linux/include/linux/jbd.h
   3  *
   4  * Written by Stephen C. Tweedie <sct@redhat.com>
   5  *
   6  * Copyright 1998-2000 Red Hat, Inc --- All Rights Reserved
   7  *
   8  * This file is part of the Linux kernel and is made available under
   9  * the terms of the GNU General Public License, version 2, or at your
  10  * option, any later version, incorporated herein by reference.
  11  *
  12  * Definitions for transaction data structures for the buffer cache
  13  * filesystem journaling support.
  14  */
  15
  16 #ifndef _LINUX_JBD_H
  17 #define _LINUX_JBD_H
  18
  19 /* Allow this file to be included directly into e2fsprogs */
  20 #ifndef __KERNEL__
  21 #include "jfs_compat.h"
  22 #define JFS_DEBUG
  23 #define jfs_debug jbd_debug
  24 #else
  25
  26 #include <linux/types.h>
  27 #include <linux/buffer_head.h>
  28 #include <linux/journal-head.h>
  29 #include <linux/stddef.h>
  30 #include <linux/bit_spinlock.h>
  31 #include <linux/mutex.h>
  32 #include <linux/timer.h>
  33 #include <linux/lockdep.h>
  34
  35 #include <asm/semaphore.h>
  36 #endif
  37
  38 #define journal_oom_retry 1
  39
  40 /*
  41  * Define JBD_PARANIOD_IOFAIL to cause a kernel BUG() if ext3 finds
  42  * certain classes of error which can occur due to failed IOs.  Under
  43  * normal use we want ext3 to continue after such errors, because
  44  * hardware _can_ fail, but for debugging purposes when running tests on
  45  * known-good hardware we may want to trap these errors.
  46  */
  47 #undef JBD_PARANOID_IOFAIL
  48
  49 /*
  50  * The default maximum commit age, in seconds.
  51  */
  52 #define JBD_DEFAULT_MAX_COMMIT_AGE 5
  53
  54 #ifdef CONFIG_JBD_DEBUG
  55 /*
  56  * Define JBD_EXPENSIVE_CHECKING to enable more expensive internal
  57  * consistency checks.  By default we don't do this unless
  58  * CONFIG_JBD_DEBUG is on.
  59  */
  60 #define JBD_EXPENSIVE_CHECKING
  61 extern int journal_enable_debug;
  62
  63 #define jbd_debug(n, f, a...)                                           \
  64         do {                                                            \
  65                 if ((n) <= journal_enable_debug) {                      \
  66                         printk (KERN_DEBUG "(%s, %d): %s: ",            \
  67                                 __FILE__, __LINE__, __FUNCTION__);      \
  68                         printk (f, ## a);                               \
  69                 }                                                       \
  70         } while (0)
  71 #else
  72 #define jbd_debug(f, a...)      /**/
  73 #endif
  74
  75 extern void * __jbd_kmalloc (const char *where, size_t size, gfp_t flags, int retry);
  76 extern void * jbd_slab_alloc(size_t size, gfp_t flags);
  77 extern void jbd_slab_free(void *ptr, size_t size);
  78
  79 #define jbd_kmalloc(size, flags) \
  80         __jbd_kmalloc(__FUNCTION__, (size), (flags), journal_oom_retry)
  81 #define jbd_rep_kmalloc(size, flags) \
  82         __jbd_kmalloc(__FUNCTION__, (size), (flags), 1)
  83
  84 #define JFS_MIN_JOURNAL_BLOCKS 1024
  85
  86 #ifdef __KERNEL__
  87
  88 /**
  89  * typedef handle_t - The handle_t type represents a single atomic update being performed by some process.
  90  *
  91  * All filesystem modifications made by the process go
  92  * through this handle.  Recursive operations (such as quota operations)
  93  * are gathered into a single update.
  94  *
  95  * The buffer credits field is used to account for journaled buffers
  96  * being modified by the running process.  To ensure that there is
  97  * enough log space for all outstanding operations, we need to limit the
  98  * number of outstanding buffers possible at any time.  When the
  99  * operation completes, any buffer credits not used are credited back to
 100  * the transaction, so that at all times we know how many buffers the
 101  * outstanding updates on a transaction might possibly touch.
 102  *
 103  * This is an opaque datatype.
 104  **/
 105 typedef struct handle_s         handle_t;       /* Atomic operation type */
 106
 107
 108 /**
 109  * typedef journal_t - The journal_t maintains all of the journaling state information for a single filesystem.
 110  *
 111  * journal_t is linked to from the fs superblock structure.
 112  *
 113  * We use the journal_t to keep track of all outstanding transaction
 114  * activity on the filesystem, and to manage the state of the log
 115  * writing process.
 116  *
 117  * This is an opaque datatype.
 118  **/
 119 typedef struct journal_s        journal_t;      /* Journal control structure */
 120 #endif
 121
 122 /*
 123  * Internal structures used by the logging mechanism:
 124  */
 125
 126 #define JFS_MAGIC_NUMBER 0xc03b3998U /* The first 4 bytes of /dev/random! */
 127
 128 /*
 129  * On-disk structures
 130  */
 131
 132 /*
 133  * Descriptor block types:
 134  */
 135
 136 #define JFS_DESCRIPTOR_BLOCK    1
 137 #define JFS_COMMIT_BLOCK        2
 138 #define JFS_SUPERBLOCK_V1       3
 139 #define JFS_SUPERBLOCK_V2       4
 140 #define JFS_REVOKE_BLOCK        5
 141
 142 /*
 143  * Standard header for all descriptor blocks:
 144  */
 145 typedef struct journal_header_s
 146 {
 147         __be32          h_magic;
 148         __be32          h_blocktype;
 149         __be32          h_sequence;
 150 } journal_header_t;
 151
 152
 153 /*
 154  * The block tag: used to describe a single buffer in the journal
 155  */
 156 typedef struct journal_block_tag_s
 157 {
 158         __be32          t_blocknr;      /* The on-disk block number */
 159         __be32          t_flags;        /* See below */
 160 } journal_block_tag_t;
 161
 162 /*
 163  * The revoke descriptor: used on disk to describe a series of blocks to
 164  * be revoked from the log
 165  */
 166 typedef struct journal_revoke_header_s
 167 {
 168         journal_header_t r_header;
 169         __be32           r_count;       /* Count of bytes used in the block */
 170 } journal_revoke_header_t;
 171
 172
 173 /* Definitions for the journal tag flags word: */
 174 #define JFS_FLAG_ESCAPE         1       /* on-disk block is escaped */
 175 #define JFS_FLAG_SAME_UUID      2       /* block has same uuid as previous */
 176 #define JFS_FLAG_DELETED        4       /* block deleted by this transaction */
 177 #define JFS_FLAG_LAST_TAG       8       /* last tag in this descriptor block */
 178
 179
 180 /*
 181  * The journal superblock.  All fields are in big-endian byte order.
 182  */
 183 typedef struct journal_superblock_s
 184 {
 185 /* 0x0000 */
 186         journal_header_t s_header;
 187
 188 /* 0x000C */
 189         /* Static information describing the journal */
 190         __be32  s_blocksize;            /* journal device blocksize */
 191         __be32  s_maxlen;               /* total blocks in journal file */
 192         __be32  s_first;                /* first block of log information */
 193
 194 /* 0x0018 */
 195         /* Dynamic information describing the current state of the log */
 196         __be32  s_sequence;             /* first commit ID expected in log */
 197         __be32  s_start;                /* blocknr of start of log */
 198
 199 /* 0x0020 */
 200         /* Error value, as set by journal_abort(). */
 201         __be32  s_errno;
 202
 203 /* 0x0024 */
 204         /* Remaining fields are only valid in a version-2 superblock */
 205         __be32  s_feature_compat;       /* compatible feature set */
 206         __be32  s_feature_incompat;     /* incompatible feature set */
 207         __be32  s_feature_ro_compat;    /* readonly-compatible feature set */
 208 /* 0x0030 */
 209         __u8    s_uuid[16];             /* 128-bit uuid for journal */
 210
 211 /* 0x0040 */
 212         __be32  s_nr_users;             /* Nr of filesystems sharing log */
 213
 214         __be32  s_dynsuper;             /* Blocknr of dynamic superblock copy*/
 215
 216 /* 0x0048 */
 217         __be32  s_max_transaction;      /* Limit of journal blocks per trans.*/
 218         __be32  s_max_trans_data;       /* Limit of data blocks per trans. */
 219
 220 /* 0x0050 */
 221         __u32   s_padding[44];
 222
 223 /* 0x0100 */
 224         __u8    s_users[16*48];         /* ids of all fs'es sharing the log */
 225 /* 0x0400 */
 226 } journal_superblock_t;
 227
 228 #define JFS_HAS_COMPAT_FEATURE(j,mask)                                  \
 229         ((j)->j_format_version >= 2 &&                                  \
 230          ((j)->j_superblock->s_feature_compat & cpu_to_be32((mask))))
 231 #define JFS_HAS_RO_COMPAT_FEATURE(j,mask)                               \
 232         ((j)->j_format_version >= 2 &&                                  \
 233          ((j)->j_superblock->s_feature_ro_compat & cpu_to_be32((mask))))
 234 #define JFS_HAS_INCOMPAT_FEATURE(j,mask)                                \
 235         ((j)->j_format_version >= 2 &&                                  \
 236          ((j)->j_superblock->s_feature_incompat & cpu_to_be32((mask))))
 237
 238 #define JFS_FEATURE_INCOMPAT_REVOKE     0x00000001
 239
 240 /* Features known to this kernel version: */
 241 #define JFS_KNOWN_COMPAT_FEATURES       0
 242 #define JFS_KNOWN_ROCOMPAT_FEATURES     0
 243 #define JFS_KNOWN_INCOMPAT_FEATURES     JFS_FEATURE_INCOMPAT_REVOKE
 244
 245 #ifdef __KERNEL__
 246
 247 #include <linux/fs.h>
 248 #include <linux/sched.h>
 249
 250 #define JBD_ASSERTIONS
 251 #ifdef JBD_ASSERTIONS
 252 #define J_ASSERT(assert)                                                \
 253 do {                                                                    \
 254         if (!(assert)) {                                                \
 255                 printk (KERN_EMERG                                      \
 256                         "Assertion failure in %s() at %s:%d: \"%s\"\n", \
 257                         __FUNCTION__, __FILE__, __LINE__, # assert);    \
 258                 BUG();                                                  \
 259         }                                                               \
 260 } while (0)
 261
 262 #if defined(CONFIG_BUFFER_DEBUG)
 263 void buffer_assertion_failure(struct buffer_head *bh);
 264 #define J_ASSERT_BH(bh, expr)                                           \
 265         do {                                                            \
 266                 if (!(expr))                                            \
 267                         buffer_assertion_failure(bh);                   \
 268                 J_ASSERT(expr);                                         \
 269         } while (0)
 270 #define J_ASSERT_JH(jh, expr)   J_ASSERT_BH(jh2bh(jh), expr)
 271 #else
 272 #define J_ASSERT_BH(bh, expr)   J_ASSERT(expr)
 273 #define J_ASSERT_JH(jh, expr)   J_ASSERT(expr)
 274 #endif
 275
 276 #else
 277 #define J_ASSERT(assert)        do { } while (0)
 278 #endif          /* JBD_ASSERTIONS */
 279
 280 #if defined(JBD_PARANOID_IOFAIL)
 281 #define J_EXPECT(expr, why...)          J_ASSERT(expr)
 282 #define J_EXPECT_BH(bh, expr, why...)   J_ASSERT_BH(bh, expr)
 283 #define J_EXPECT_JH(jh, expr, why...)   J_ASSERT_JH(jh, expr)
 284 #else
 285 #define __journal_expect(expr, why...)                                       \
 286         ({                                                                   \
 287                 int val = (expr);                                            \
 288                 if (!val) {                                                  \
 289                         printk(KERN_ERR                                      \
 290                                 "EXT3-fs unexpected failure: %s;\n",# expr); \
 291                         printk(KERN_ERR why "\n");                           \
 292                 }                                                            \
 293                 val;                                                         \
 294         })
 295 #define J_EXPECT(expr, why...)          __journal_expect(expr, ## why)
 296 #define J_EXPECT_BH(bh, expr, why...)   __journal_expect(expr, ## why)
 297 #define J_EXPECT_JH(jh, expr, why...)   __journal_expect(expr, ## why)
 298 #endif
 299
 300 enum jbd_state_bits {
 301         BH_JBD                  /* Has an attached ext3 journal_head */
 302           = BH_PrivateStart,
 303         BH_JWrite,              /* Being written to log (@@@ DEBUGGING) */
 304         BH_Freed,               /* Has been freed (truncated) */
 305         BH_Revoked,             /* Has been revoked from the log */
 306         BH_RevokeValid,         /* Revoked flag is valid */
 307         BH_JBDDirty,            /* Is dirty but journaled */
 308         BH_State,               /* Pins most journal_head state */
 309         BH_JournalHead,         /* Pins bh->b_private and jh->b_bh */
 310         BH_Unshadow,            /* Dummy bit, for BJ_Shadow wakeup filtering */
 311 };
 312
 313 BUFFER_FNS(JBD, jbd)
 314 BUFFER_FNS(JWrite, jwrite)
 315 BUFFER_FNS(JBDDirty, jbddirty)
 316 TAS_BUFFER_FNS(JBDDirty, jbddirty)
 317 BUFFER_FNS(Revoked, revoked)
 318 TAS_BUFFER_FNS(Revoked, revoked)
 319 BUFFER_FNS(RevokeValid, revokevalid)
 320 TAS_BUFFER_FNS(RevokeValid, revokevalid)
 321 BUFFER_FNS(Freed, freed)
 322
 323 static inline struct buffer_head *jh2bh(struct journal_head *jh)
 324 {
 325         return jh->b_bh;
 326 }
 327
 328 static inline struct journal_head *bh2jh(struct buffer_head *bh)
 329 {
 330         return bh->b_private;
 331 }
 332
 333 static inline void jbd_lock_bh_state(struct buffer_head *bh)
 334 {
 335         bit_spin_lock(BH_State, &bh->b_state);
 336 }
 337
 338 static inline int jbd_trylock_bh_state(struct buffer_head *bh)
 339 {
 340         return bit_spin_trylock(BH_State, &bh->b_state);
 341 }
 342
 343 static inline int jbd_is_locked_bh_state(struct buffer_head *bh)
 344 {
 345         return bit_spin_is_locked(BH_State, &bh->b_state);
 346 }
 347
 348 static inline void jbd_unlock_bh_state(struct buffer_head *bh)
 349 {
 350         bit_spin_unlock(BH_State, &bh->b_state);
 351 }
 352
 353 static inline void jbd_lock_bh_journal_head(struct buffer_head *bh)
 354 {
 355         bit_spin_lock(BH_JournalHead, &bh->b_state);
 356 }
 357
 358 static inline void jbd_unlock_bh_journal_head(struct buffer_head *bh)
 359 {
 360         bit_spin_unlock(BH_JournalHead, &bh->b_state);
 361 }
 362
 363 struct jbd_revoke_table_s;
 364
 365 /**
 366  * struct handle_s - The handle_s type is the concrete type associated with
 367  *     handle_t.
 368  * @h_transaction: Which compound transaction is this update a part of?
 369  * @h_buffer_credits: Number of remaining buffers we are allowed to dirty.
 370  * @h_ref: Reference count on this handle
 371  * @h_err: Field for caller's use to track errors through large fs operations
 372  * @h_sync: flag for sync-on-close
 373  * @h_jdata: flag to force data journaling
 374  * @h_aborted: flag indicating fatal error on handle
 375  * @h_lockdep_map: lockdep info for debugging lock problems
 376  **/
 377
 378 /* Docbook can't yet cope with the bit fields, but will leave the documentation
 379  * in so it can be fixed later.
 380  */
 381
 382 struct handle_s
 383 {
 384         /* Which compound transaction is this update a part of? */
 385         transaction_t           *h_transaction;
 386
 387         /* Number of remaining buffers we are allowed to dirty: */
 388         int                     h_buffer_credits;
 389
 390         /* Reference count on this handle */
 391         int                     h_ref;
 392
 393         /* Field for caller's use to track errors through large fs */
 394         /* operations */
 395         int                     h_err;
 396
 397         /* Flags [no locking] */
 398         unsigned int    h_sync:         1;      /* sync-on-close */
 399         unsigned int    h_jdata:        1;      /* force data journaling */
 400         unsigned int    h_aborted:      1;      /* fatal error on handle */
 401
 402 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 403         struct lockdep_map      h_lockdep_map;
 404 #endif
 405 };
 406
 407
 408 /* The transaction_t type is the guts of the journaling mechanism.  It
 409  * tracks a compound transaction through its various states:
 410  *
 411  * RUNNING:     accepting new updates
 412  * LOCKED:      Updates still running but we don't accept new ones
 413  * RUNDOWN:     Updates are tidying up but have finished requesting
 414  *              new buffers to modify (state not used for now)
 415  * FLUSH:       All updates complete, but we are still writing to disk
 416  * COMMIT:      All data on disk, writing commit record
 417  * FINISHED:    We still have to keep the transaction for checkpointing.
 418  *
 419  * The transaction keeps track of all of the buffers modified by a
 420  * running transaction, and all of the buffers committed but not yet
 421  * flushed to home for finished transactions.
 422  */
 423
 424 /*
 425  * Lock ranking:
 426  *
 427  *    j_list_lock
 428  *      ->jbd_lock_bh_journal_head()    (This is "innermost")
 429  *
 430  *    j_state_lock
 431  *    ->jbd_lock_bh_state()
 432  *
 433  *    jbd_lock_bh_state()
 434  *    ->j_list_lock
 435  *
 436  *    j_state_lock
 437  *    ->t_handle_lock
 438  *
 439  *    j_state_lock
 440  *    ->j_list_lock                     (journal_unmap_buffer)
 441  *
 442  */
 443
 444 struct transaction_s
 445 {
 446         /* Pointer to the journal for this transaction. [no locking] */
 447         journal_t               *t_journal;
 448
 449         /* Sequence number for this transaction [no locking] */
 450         tid_t                   t_tid;
 451
 452         /*
 453          * Transaction's current state
 454          * [no locking - only kjournald alters this]
 455          * FIXME: needs barriers
 456          * KLUDGE: [use j_state_lock]
 457          */
 458         enum {
 459                 T_RUNNING,
 460                 T_LOCKED,
 461                 T_RUNDOWN,
 462                 T_FLUSH,
 463                 T_COMMIT,
 464                 T_FINISHED
 465         }                       t_state;
 466
 467         /*
 468          * Where in the log does this transaction's commit start? [no locking]
 469          */
 470         unsigned long           t_log_start;
 471
 472         /* Number of buffers on the t_buffers list [j_list_lock] */
 473         int                     t_nr_buffers;
 474
 475         /*
 476          * Doubly-linked circular list of all buffers reserved but not yet
 477          * modified by this transaction [j_list_lock]
 478          */
 479         struct journal_head     *t_reserved_list;
 480
 481         /*
 482          * Doubly-linked circular list of all buffers under writeout during
 483          * commit [j_list_lock]
 484          */
 485         struct journal_head     *t_locked_list;
 486
 487         /*
 488          * Doubly-linked circular list of all metadata buffers owned by this
 489          * transaction [j_list_lock]
 490          */
 491         struct journal_head     *t_buffers;
 492
 493         /*
 494          * Doubly-linked circular list of all data buffers still to be
 495          * flushed before this transaction can be committed [j_list_lock]
 496          */
 497         struct journal_head     *t_sync_datalist;
 498
 499         /*
 500          * Doubly-linked circular list of all forget buffers (superseded
 501          * buffers which we can un-checkpoint once this transaction commits)
 502          * [j_list_lock]
 503          */
 504         struct journal_head     *t_forget;
 505
 506         /*
 507          * Doubly-linked circular list of all buffers still to be flushed before
 508          * this transaction can be checkpointed. [j_list_lock]
 509          */
 510         struct journal_head     *t_checkpoint_list;
 511
 512         /*
 513          * Doubly-linked circular list of all buffers submitted for IO while
 514          * checkpointing. [j_list_lock]
 515          */
 516         struct journal_head     *t_checkpoint_io_list;
 517
 518         /*
 519          * Doubly-linked circular list of temporary buffers currently undergoing
 520          * IO in the log [j_list_lock]
 521          */
 522         struct journal_head     *t_iobuf_list;
 523
 524         /*
 525          * Doubly-linked circular list of metadata buffers being shadowed by log
 526          * IO.  The IO buffers on the iobuf list and the shadow buffers on this
 527          * list match each other one for one at all times. [j_list_lock]
 528          */
 529         struct journal_head     *t_shadow_list;
 530
 531         /*
 532          * Doubly-linked circular list of control buffers being written to the
 533          * log. [j_list_lock]
 534          */
 535         struct journal_head     *t_log_list;
 536
 537         /*
 538          * Protects info related to handles
 539          */
 540         spinlock_t              t_handle_lock;
 541
 542         /*
 543          * Number of outstanding updates running on this transaction
 544          * [t_handle_lock]
 545          */
 546         int                     t_updates;
 547
 548         /*
 549          * Number of buffers reserved for use by all handles in this transaction
 550          * handle but not yet modified. [t_handle_lock]
 551          */
 552         int                     t_outstanding_credits;
 553
 554         /*
 555          * Forward and backward links for the circular list of all transactions
 556          * awaiting checkpoint. [j_list_lock]
 557          */
 558         transaction_t           *t_cpnext, *t_cpprev;
 559
 560         /*
 561          * When will the transaction expire (become due for commit), in jiffies?
 562          * [no locking]
 563          */
 564         unsigned long           t_expires;
 565
 566         /*
 567          * How many handles used this transaction? [t_handle_lock]
 568          */
 569         int t_handle_count;
 570
 571 };
 572
 573 /**
 574  * struct journal_s - The journal_s type is the concrete type associated with
 575  *     journal_t.
 576  * @j_flags:  General journaling state flags
 577  * @j_errno:  Is there an outstanding uncleared error on the journal (from a
 578  *     prior abort)?
 579  * @j_sb_buffer: First part of superblock buffer
 580  * @j_superblock: Second part of superblock buffer
 581  * @j_format_version: Version of the superblock format
 582  * @j_state_lock: Protect the various scalars in the journal
 583  * @j_barrier_count:  Number of processes waiting to create a barrier lock
 584  * @j_barrier: The barrier lock itself
 585  * @j_running_transaction: The current running transaction..
 586  * @j_committing_transaction: the transaction we are pushing to disk
 587  * @j_checkpoint_transactions: a linked circular list of all transactions
 588  *  waiting for checkpointing
 589  * @j_wait_transaction_locked: Wait queue for waiting for a locked transaction
 590  *  to start committing, or for a barrier lock to be released
 591  * @j_wait_logspace: Wait queue for waiting for checkpointing to complete
 592  * @j_wait_done_commit: Wait queue for waiting for commit to complete
 593  * @j_wait_checkpoint:  Wait queue to trigger checkpointing
 594  * @j_wait_commit: Wait queue to trigger commit
 595  * @j_wait_updates: Wait queue to wait for updates to complete
 596  * @j_checkpoint_mutex: Mutex for locking against concurrent checkpoints
 597  * @j_head: Journal head - identifies the first unused block in the journal
 598  * @j_tail: Journal tail - identifies the oldest still-used block in the
 599  *  journal.
 600  * @j_free: Journal free - how many free blocks are there in the journal?
 601  * @j_first: The block number of the first usable block
 602  * @j_last: The block number one beyond the last usable block
 603  * @j_dev: Device where we store the journal
 604  * @j_blocksize: blocksize for the location where we store the journal.
 605  * @j_blk_offset: starting block offset for into the device where we store the
 606  *     journal
 607  * @j_fs_dev: Device which holds the client fs.  For internal journal this will
 608  *     be equal to j_dev
 609  * @j_maxlen: Total maximum capacity of the journal region on disk.
 610  * @j_list_lock: Protects the buffer lists and internal buffer state.
 611  * @j_inode: Optional inode where we store the journal.  If present, all journal
 612  *     block numbers are mapped into this inode via bmap().
 613  * @j_tail_sequence:  Sequence number of the oldest transaction in the log
 614  * @j_transaction_sequence: Sequence number of the next transaction to grant
 615  * @j_commit_sequence: Sequence number of the most recently committed
 616  *  transaction
 617  * @j_commit_request: Sequence number of the most recent transaction wanting
 618  *     commit
 619  * @j_uuid: Uuid of client object.
 620  * @j_task: Pointer to the current commit thread for this journal
 621  * @j_max_transaction_buffers:  Maximum number of metadata buffers to allow in a
 622  *     single compound commit transaction
 623  * @j_commit_interval: What is the maximum transaction lifetime before we begin
 624  *  a commit?
 625  * @j_commit_timer:  The timer used to wakeup the commit thread
 626  * @j_revoke_lock: Protect the revoke table
 627  * @j_revoke: The revoke table - maintains the list of revoked blocks in the
 628  *     current transaction.
 629  * @j_revoke_table: alternate revoke tables for j_revoke
 630  * @j_wbuf: array of buffer_heads for journal_commit_transaction
 631  * @j_wbufsize: maximum number of buffer_heads allowed in j_wbuf, the
 632  *      number that will fit in j_blocksize
 633  * @j_last_sync_writer: most recent pid which did a synchronous write
 634  * @j_private: An opaque pointer to fs-private information.
 635  */
 636
 637 struct journal_s
 638 {
 639         /* General journaling state flags [j_state_lock] */
 640         unsigned long           j_flags;
 641
 642         /*
 643          * Is there an outstanding uncleared error on the journal (from a prior
 644          * abort)? [j_state_lock]
 645          */
 646         int                     j_errno;
 647
 648         /* The superblock buffer */
 649         struct buffer_head      *j_sb_buffer;
 650         journal_superblock_t    *j_superblock;
 651
 652         /* Version of the superblock format */
 653         int                     j_format_version;
 654
 655         /*
 656          * Protect the various scalars in the journal
 657          */
 658         spinlock_t              j_state_lock;
 659
 660         /*
 661          * Number of processes waiting to create a barrier lock [j_state_lock]
 662          */
 663         int                     j_barrier_count;
 664
 665         /* The barrier lock itself */
 666         struct mutex            j_barrier;
 667
 668         /*
 669          * Transactions: The current running transaction...
 670          * [j_state_lock] [caller holding open handle]
 671          */
 672         transaction_t           *j_running_transaction;
 673
 674         /*
 675          * the transaction we are pushing to disk
 676          * [j_state_lock] [caller holding open handle]
 677          */
 678         transaction_t           *j_committing_transaction;
 679
 680         /*
 681          * ... and a linked circular list of all transactions waiting for
 682          * checkpointing. [j_list_lock]
 683          */
 684         transaction_t           *j_checkpoint_transactions;
 685
 686         /*
 687          * Wait queue for waiting for a locked transaction to start committing,
 688          * or for a barrier lock to be released
 689          */
 690         wait_queue_head_t       j_wait_transaction_locked;
 691
 692         /* Wait queue for waiting for checkpointing to complete */
 693         wait_queue_head_t       j_wait_logspace;
 694
 695         /* Wait queue for waiting for commit to complete */
 696         wait_queue_head_t       j_wait_done_commit;
 697
 698         /* Wait queue to trigger checkpointing */
 699         wait_queue_head_t       j_wait_checkpoint;
 700
 701         /* Wait queue to trigger commit */
 702         wait_queue_head_t       j_wait_commit;
 703
 704         /* Wait queue to wait for updates to complete */
 705         wait_queue_head_t       j_wait_updates;
 706
 707         /* Semaphore for locking against concurrent checkpoints */
 708         struct mutex            j_checkpoint_mutex;
 709
 710         /*
 711          * Journal head: identifies the first unused block in the journal.
 712          * [j_state_lock]
 713          */
 714         unsigned long           j_head;
 715
 716         /*
 717          * Journal tail: identifies the oldest still-used block in the journal.
 718          * [j_state_lock]
 719          */
 720         unsigned long           j_tail;
 721
 722         /*
 723          * Journal free: how many free blocks are there in the journal?
 724          * [j_state_lock]
 725          */
 726         unsigned long           j_free;
 727
 728         /*
 729          * Journal start and end: the block numbers of the first usable block
 730          * and one beyond the last usable block in the journal. [j_state_lock]
 731          */
 732         unsigned long           j_first;
 733         unsigned long           j_last;
 734
 735         /*
 736          * Device, blocksize and starting block offset for the location where we
 737          * store the journal.
 738          */
 739         struct block_device     *j_dev;
 740         int                     j_blocksize;
 741         unsigned long           j_blk_offset;
 742
 743         /*
 744          * Device which holds the client fs.  For internal journal this will be
 745          * equal to j_dev.
 746          */
 747         struct block_device     *j_fs_dev;
 748
 749         /* Total maximum capacity of the journal region on disk. */
 750         unsigned int            j_maxlen;
 751
 752         /*
 753          * Protects the buffer lists and internal buffer state.
 754          */
 755         spinlock_t              j_list_lock;
 756
 757         /* Optional inode where we store the journal.  If present, all */
 758         /* journal block numbers are mapped into this inode via */
 759         /* bmap(). */
 760         struct inode            *j_inode;
 761
 762         /*
 763          * Sequence number of the oldest transaction in the log [j_state_lock]
 764          */
 765         tid_t                   j_tail_sequence;
 766
 767         /*
 768          * Sequence number of the next transaction to grant [j_state_lock]
 769          */
 770         tid_t                   j_transaction_sequence;
 771
 772         /*
 773          * Sequence number of the most recently committed transaction
 774          * [j_state_lock].
 775          */
 776         tid_t                   j_commit_sequence;
 777
 778         /*
 779          * Sequence number of the most recent transaction wanting commit
 780          * [j_state_lock]
 781          */
 782         tid_t                   j_commit_request;
 783
 784         /*
 785          * Journal uuid: identifies the object (filesystem, LVM volume etc)
 786          * backed by this journal.  This will eventually be replaced by an array
 787          * of uuids, allowing us to index multiple devices within a single
 788          * journal and to perform atomic updates across them.
 789          */
 790         __u8                    j_uuid[16];
 791
 792         /* Pointer to the current commit thread for this journal */
 793         struct task_struct      *j_task;
 794
 795         /*
 796          * Maximum number of metadata buffers to allow in a single compound
 797          * commit transaction
 798          */
 799         int                     j_max_transaction_buffers;
 800
 801         /*
 802          * What is the maximum transaction lifetime before we begin a commit?
 803          */
 804         unsigned long           j_commit_interval;
 805
 806         /* The timer used to wakeup the commit thread: */
 807         struct timer_list       j_commit_timer;
 808
 809         /*
 810          * The revoke table: maintains the list of revoked blocks in the
 811          * current transaction.  [j_revoke_lock]
 812          */
 813         spinlock_t              j_revoke_lock;
 814         struct jbd_revoke_table_s *j_revoke;
 815         struct jbd_revoke_table_s *j_revoke_table[2];
 816
 817         /*
 818          * array of bhs for journal_commit_transaction
 819          */
 820         struct buffer_head      **j_wbuf;
 821         int                     j_wbufsize;
 822
 823         pid_t                   j_last_sync_writer;
 824
 825         /*
 826          * An opaque pointer to fs-private information.  ext3 puts its
 827          * superblock pointer here
 828          */
 829         void *j_private;
 830 };
 831
 832 /*
 833  * Journal flag definitions
 834  */
 835 #define JFS_UNMOUNT     0x001   /* Journal thread is being destroyed */
 836 #define JFS_ABORT       0x002   /* Journaling has been aborted for errors. */
 837 #define JFS_ACK_ERR     0x004   /* The errno in the sb has been acked */
 838 #define JFS_FLUSHED     0x008   /* The journal superblock has been flushed */
 839 #define JFS_LOADED      0x010   /* The journal superblock has been loaded */
 840 #define JFS_BARRIER     0x020   /* Use IDE barriers */
 841
 842 /*
 843  * Function declarations for the journaling transaction and buffer
 844  * management
 845  */
 846
 847 /* Filing buffers */
 848 extern void journal_unfile_buffer(journal_t *, struct journal_head *);
 849 extern void __journal_unfile_buffer(struct journal_head *);
 850 extern void __journal_refile_buffer(struct journal_head *);
 851 extern void journal_refile_buffer(journal_t *, struct journal_head *);
 852 extern void __journal_file_buffer(struct journal_head *, transaction_t *, int);
 853 extern void __journal_free_buffer(struct journal_head *bh);
 854 extern void journal_file_buffer(struct journal_head *, transaction_t *, int);
 855 extern void __journal_clean_data_list(transaction_t *transaction);
 856
 857 /* Log buffer allocation */
 858 extern struct journal_head * journal_get_descriptor_buffer(journal_t *);
 859 int journal_next_log_block(journal_t *, unsigned long *);
 860
 861 /* Commit management */
 862 extern void journal_commit_transaction(journal_t *);
 863
 864 /* Checkpoint list management */
 865 int __journal_clean_checkpoint_list(journal_t *journal);
 866 int __journal_remove_checkpoint(struct journal_head *);
 867 void __journal_insert_checkpoint(struct journal_head *, transaction_t *);
 868
 869 /* Buffer IO */
 870 extern int
 871 journal_write_metadata_buffer(transaction_t       *transaction,
 872                               struct journal_head  *jh_in,
 873                               struct journal_head **jh_out,
 874                               unsigned long        blocknr);
 875
 876 /* Transaction locking */
 877 extern void             __wait_on_journal (journal_t *);
 878
 879 /*
 880  * Journal locking.
 881  *
 882  * We need to lock the journal during transaction state changes so that nobody
 883  * ever tries to take a handle on the running transaction while we are in the
 884  * middle of moving it to the commit phase.  j_state_lock does this.
 885  *
 886  * Note that the locking is completely interrupt unsafe.  We never touch
 887  * journal structures from interrupts.
 888  */
 889
 890 static inline handle_t *journal_current_handle(void)
 891 {
 892         return current->journal_info;
 893 }
 894
 895 /* The journaling code user interface:
 896  *
 897  * Create and destroy handles
 898  * Register buffer modifications against the current transaction.
 899  */
 900
 901 extern handle_t *journal_start(journal_t *, int nblocks);
 902 extern int       journal_restart (handle_t *, int nblocks);
 903 extern int       journal_extend (handle_t *, int nblocks);
 904 extern int       journal_get_write_access(handle_t *, struct buffer_head *);
 905 extern int       journal_get_create_access (handle_t *, struct buffer_head *);
 906 extern int       journal_get_undo_access(handle_t *, struct buffer_head *);
 907 extern int       journal_dirty_data (handle_t *, struct buffer_head *);
 908 extern int       journal_dirty_metadata (handle_t *, struct buffer_head *);
 909 extern void      journal_release_buffer (handle_t *, struct buffer_head *);
 910 extern int       journal_forget (handle_t *, struct buffer_head *);
 911 extern void      journal_sync_buffer (struct buffer_head *);
 912 extern void      journal_invalidatepage(journal_t *,
 913                                 struct page *, unsigned long);
 914 extern int       journal_try_to_free_buffers(journal_t *, struct page *, gfp_t);
 915 extern int       journal_stop(handle_t *);
 916 extern int       journal_flush (journal_t *);
 917 extern void      journal_lock_updates (journal_t *);
 918 extern void      journal_unlock_updates (journal_t *);
 919
 920 extern journal_t * journal_init_dev(struct block_device *bdev,
 921                                 struct block_device *fs_dev,
 922                                 int start, int len, int bsize);
 923 extern journal_t * journal_init_inode (struct inode *);
 924 extern int         journal_update_format (journal_t *);
 925 extern int         journal_check_used_features
 926                    (journal_t *, unsigned long, unsigned long, unsigned long);
 927 extern int         journal_check_available_features
 928                    (journal_t *, unsigned long, unsigned long, unsigned long);
 929 extern int         journal_set_features
 930                    (journal_t *, unsigned long, unsigned long, unsigned long);
 931 extern int         journal_create     (journal_t *);
 932 extern int         journal_load       (journal_t *journal);
 933 extern void        journal_destroy    (journal_t *);
 934 extern int         journal_recover    (journal_t *journal);
 935 extern int         journal_wipe       (journal_t *, int);
 936 extern int         journal_skip_recovery        (journal_t *);
 937 extern void        journal_update_superblock    (journal_t *, int);
 938 extern void        __journal_abort_hard (journal_t *);
 939 extern void        journal_abort      (journal_t *, int);
 940 extern int         journal_errno      (journal_t *);
 941 extern void        journal_ack_err    (journal_t *);
 942 extern int         journal_clear_err  (journal_t *);
 943 extern int         journal_bmap(journal_t *, unsigned long, unsigned long *);
 944 extern int         journal_force_commit(journal_t *);
 945
 946 /*
 947  * journal_head management
 948  */
 949 struct journal_head *journal_add_journal_head(struct buffer_head *bh);
 950 struct journal_head *journal_grab_journal_head(struct buffer_head *bh);
 951 void journal_remove_journal_head(struct buffer_head *bh);
 952 void journal_put_journal_head(struct journal_head *jh);
 953
 954 /*
 955  * handle management
 956  */
 957 extern struct kmem_cache *jbd_handle_cache;
 958
 959 static inline handle_t *jbd_alloc_handle(gfp_t gfp_flags)
 960 {
 961         return kmem_cache_alloc(jbd_handle_cache, gfp_flags);
 962 }
 963
 964 static inline void jbd_free_handle(handle_t *handle)
 965 {
 966         kmem_cache_free(jbd_handle_cache, handle);
 967 }
 968
 969 /* Primary revoke support */
 970 #define JOURNAL_REVOKE_DEFAULT_HASH 256
 971 extern int         journal_init_revoke(journal_t *, int);
 972 extern void        journal_destroy_revoke_caches(void);
 973 extern int         journal_init_revoke_caches(void);
 974
 975 extern void        journal_destroy_revoke(journal_t *);
 976 extern int         journal_revoke (handle_t *,
 977                                 unsigned long, struct buffer_head *);
 978 extern int         journal_cancel_revoke(handle_t *, struct journal_head *);
 979 extern void        journal_write_revoke_records(journal_t *, transaction_t *);
 980
 981 /* Recovery revoke support */
 982 extern int      journal_set_revoke(journal_t *, unsigned long, tid_t);
 983 extern int      journal_test_revoke(journal_t *, unsigned long, tid_t);
 984 extern void     journal_clear_revoke(journal_t *);
 985 extern void     journal_switch_revoke_table(journal_t *journal);
 986
 987 /*
 988  * The log thread user interface:
 989  *
 990  * Request space in the current transaction, and force transaction commit
 991  * transitions on demand.
 992  */
 993
 994 int __log_space_left(journal_t *); /* Called with journal locked */
 995 int log_start_commit(journal_t *journal, tid_t tid);
 996 int __log_start_commit(journal_t *journal, tid_t tid);
 997 int journal_start_commit(journal_t *journal, tid_t *tid);
 998 int journal_force_commit_nested(journal_t *journal);
 999 int log_wait_commit(journal_t *journal, tid_t tid);
1000 int log_do_checkpoint(journal_t *journal);
1001
1002 void __log_wait_for_space(journal_t *journal);
1003 extern void     __journal_drop_transaction(journal_t *, transaction_t *);
1004 extern int      cleanup_journal_tail(journal_t *);
1005
1006 /* Debugging code only: */
1007
1008 #define jbd_ENOSYS() \
1009 do {                                                                       \
1010         printk (KERN_ERR "JBD unimplemented function %s\n", __FUNCTION__); \
1011         current->state = TASK_UNINTERRUPTIBLE;                             \
1012         schedule();                                                        \
1013 } while (1)
1014
1015 /*
1016  * is_journal_abort
1017  *
1018  * Simple test wrapper function to test the JFS_ABORT state flag.  This
1019  * bit, when set, indicates that we have had a fatal error somewhere,
1020  * either inside the journaling layer or indicated to us by the client
1021  * (eg. ext3), and that we and should not commit any further
1022  * transactions.
1023  */
1024
1025 static inline int is_journal_aborted(journal_t *journal)
1026 {
1027         return journal->j_flags & JFS_ABORT;
1028 }
1029
1030 static inline int is_handle_aborted(handle_t *handle)
1031 {
1032         if (handle->h_aborted)
1033                 return 1;
1034         return is_journal_aborted(handle->h_transaction->t_journal);
1035 }
1036
1037 static inline void journal_abort_handle(handle_t *handle)
1038 {
1039         handle->h_aborted = 1;
1040 }
1041
1042 #endif /* __KERNEL__   */
1043
1044 /* Comparison functions for transaction IDs: perform comparisons using
1045  * modulo arithmetic so that they work over sequence number wraps. */
1046
1047 static inline int tid_gt(tid_t x, tid_t y)
1048 {
1049         int difference = (x - y);
1050         return (difference > 0);
1051 }
1052
1053 static inline int tid_geq(tid_t x, tid_t y)
1054 {
1055         int difference = (x - y);
1056         return (difference >= 0);
1057 }
1058
1059 extern int journal_blocks_per_page(struct inode *inode);
1060
1061 /*
1062  * Return the minimum number of blocks which must be free in the journal
1063  * before a new transaction may be started.  Must be called under j_state_lock.
1064  */
1065 static inline int jbd_space_needed(journal_t *journal)
1066 {
1067         int nblocks = journal->j_max_transaction_buffers;
1068         if (journal->j_committing_transaction)
1069                 nblocks += journal->j_committing_transaction->
1070                                         t_outstanding_credits;
1071         return nblocks;
1072 }
1073
1074 /*
1075  * Definitions which augment the buffer_head layer
1076  */
1077
1078 /* journaling buffer types */
1079 #define BJ_None         0       /* Not journaled */
1080 #define BJ_SyncData     1       /* Normal data: flush before commit */
1081 #define BJ_Metadata     2       /* Normal journaled metadata */
1082 #define BJ_Forget       3       /* Buffer superseded by this transaction */
1083 #define BJ_IO           4       /* Buffer is for temporary IO use */
1084 #define BJ_Shadow       5       /* Buffer contents being shadowed to the log */
1085 #define BJ_LogCtl       6       /* Buffer contains log descriptors */
1086 #define BJ_Reserved     7       /* Buffer is reserved for access by journal */
1087 #define BJ_Locked       8       /* Locked for I/O during commit */
1088 #define BJ_Types        9
1089
1090 extern int jbd_blocks_per_page(struct inode *inode);
1091
1092 #ifdef __KERNEL__
1093
1094 #define buffer_trace_init(bh)   do {} while (0)
1095 #define print_buffer_fields(bh) do {} while (0)
1096 #define print_buffer_trace(bh)  do {} while (0)
1097 #define BUFFER_TRACE(bh, info)  do {} while (0)
1098 #define BUFFER_TRACE2(bh, bh2, info)    do {} while (0)
1099 #define JBUFFER_TRACE(jh, info) do {} while (0)
1100
1101 #endif  /* __KERNEL__ */
1102
1103 #endif  /* _LINUX_JBD_H */