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