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