reiserfs_fs_sb.h

This is a ReiserFs file system driver for Windows NT/2000/XP/Vista.

/* Copyright 1996-2000 Hans Reiser, see reiserfs/README for licensing
 * and copyright details */

#ifndef _LINUX_REISER_FS_SB
#define _LINUX_REISER_FS_SB

#ifdef __KERNEL__
#include <linux/workqueue.h>
#include <linux/rwsem.h>
#endif

typedef enum {
  reiserfs_attrs_cleared	= 0x00000001,
} reiserfs_super_block_flags;

/* struct reiserfs_super_block accessors/mutators
 * since this is a disk structure, it will always be in 
 * little endian format. */
#define sb_block_count(sbp)         (le32_to_cpu((sbp)->s_v1.s_block_count))
#define set_sb_block_count(sbp,v)   ((sbp)->s_v1.s_block_count = cpu_to_le32(v))
#define sb_free_blocks(sbp)         (le32_to_cpu((sbp)->s_v1.s_free_blocks))
#define set_sb_free_blocks(sbp,v)   ((sbp)->s_v1.s_free_blocks = cpu_to_le32(v))
#define sb_root_block(sbp)          (le32_to_cpu((sbp)->s_v1.s_root_block))
#define set_sb_root_block(sbp,v)    ((sbp)->s_v1.s_root_block = cpu_to_le32(v))

#define sb_jp_journal_1st_block(sbp)  \
              (le32_to_cpu((sbp)->s_v1.s_journal.jp_journal_1st_block))
#define set_sb_jp_journal_1st_block(sbp,v) \
              ((sbp)->s_v1.s_journal.jp_journal_1st_block = cpu_to_le32(v))
#define sb_jp_journal_dev(sbp) \
              (le32_to_cpu((sbp)->s_v1.s_journal.jp_journal_dev))
#define set_sb_jp_journal_dev(sbp,v) \
              ((sbp)->s_v1.s_journal.jp_journal_dev = cpu_to_le32(v))
#define sb_jp_journal_size(sbp) \
              (le32_to_cpu((sbp)->s_v1.s_journal.jp_journal_size))
#define set_sb_jp_journal_size(sbp,v) \
              ((sbp)->s_v1.s_journal.jp_journal_size = cpu_to_le32(v))
#define sb_jp_journal_trans_max(sbp) \
              (le32_to_cpu((sbp)->s_v1.s_journal.jp_journal_trans_max))
#define set_sb_jp_journal_trans_max(sbp,v) \
              ((sbp)->s_v1.s_journal.jp_journal_trans_max = cpu_to_le32(v))
#define sb_jp_journal_magic(sbp) \
              (le32_to_cpu((sbp)->s_v1.s_journal.jp_journal_magic))
#define set_sb_jp_journal_magic(sbp,v) \
              ((sbp)->s_v1.s_journal.jp_journal_magic = cpu_to_le32(v))
#define sb_jp_journal_max_batch(sbp) \
              (le32_to_cpu((sbp)->s_v1.s_journal.jp_journal_max_batch))
#define set_sb_jp_journal_max_batch(sbp,v) \
              ((sbp)->s_v1.s_journal.jp_journal_max_batch = cpu_to_le32(v))
#define sb_jp_jourmal_max_commit_age(sbp) \
              (le32_to_cpu((sbp)->s_v1.s_journal.jp_journal_max_commit_age))
#define set_sb_jp_journal_max_commit_age(sbp,v) \
              ((sbp)->s_v1.s_journal.jp_journal_max_commit_age = cpu_to_le32(v))

#define sb_blocksize(sbp)          (le16_to_cpu((sbp)->s_v1.s_blocksize))
#define set_sb_blocksize(sbp,v)    ((sbp)->s_v1.s_blocksize = cpu_to_le16(v))
#define sb_oid_maxsize(sbp)        (le16_to_cpu((sbp)->s_v1.s_oid_maxsize))
#define set_sb_oid_maxsize(sbp,v)  ((sbp)->s_v1.s_oid_maxsize = cpu_to_le16(v))
#define sb_oid_cursize(sbp)        (le16_to_cpu((sbp)->s_v1.s_oid_cursize))
#define set_sb_oid_cursize(sbp,v)  ((sbp)->s_v1.s_oid_cursize = cpu_to_le16(v))
#define sb_umount_state(sbp)       (le16_to_cpu((sbp)->s_v1.s_umount_state))
#define set_sb_umount_state(sbp,v) ((sbp)->s_v1.s_umount_state = cpu_to_le16(v))
#define sb_fs_state(sbp)           (le16_to_cpu((sbp)->s_v1.s_fs_state))
#define set_sb_fs_state(sbp,v)     ((sbp)->s_v1.s_fs_state = cpu_to_le16(v)) 
#define sb_hash_function_code(sbp) \
              (le32_to_cpu((sbp)->s_v1.s_hash_function_code))
#define set_sb_hash_function_code(sbp,v) \
              ((sbp)->s_v1.s_hash_function_code = cpu_to_le32(v))
#define sb_tree_height(sbp)        (le16_to_cpu((sbp)->s_v1.s_tree_height))
#define set_sb_tree_height(sbp,v)  ((sbp)->s_v1.s_tree_height = cpu_to_le16(v))
#define sb_bmap_nr(sbp)            (le16_to_cpu((sbp)->s_v1.s_bmap_nr))
#define set_sb_bmap_nr(sbp,v)      ((sbp)->s_v1.s_bmap_nr = cpu_to_le16(v))
#define sb_version(sbp)            (le16_to_cpu((sbp)->s_v1.s_version))
#define set_sb_version(sbp,v)      ((sbp)->s_v1.s_version = cpu_to_le16(v))

#define sb_reserved_for_journal(sbp) \
              (le16_to_cpu((sbp)->s_v1.s_reserved_for_journal))
#define set_sb_reserved_for_journal(sbp,v) \
              ((sbp)->s_v1.s_reserved_for_journal = cpu_to_le16(v))

/* LOGGING -- */

/* These all interelate for performance.  
**
** If the journal block count is smaller than n transactions, you lose speed. 
** I don't know what n is yet, I'm guessing 8-16.
**
** typical transaction size depends on the application, how often fsync is
** called, and how many metadata blocks you dirty in a 30 second period.  
** The more small files (<16k) you use, the larger your transactions will
** be.
** 
** If your journal fills faster than dirty buffers get flushed to disk, it must flush them before allowing the journal
** to wrap, which slows things down.  If you need high speed meta data updates, the journal should be big enough
** to prevent wrapping before dirty meta blocks get to disk.
**
** If the batch max is smaller than the transaction max, you'll waste space at the end of the journal
** because journal_end sets the next transaction to start at 0 if the next transaction has any chance of wrapping.
**
** The large the batch max age, the better the speed, and the more meta data changes you'll lose after a crash.
**
*/

/* don't mess with these for a while */
				/* we have a node size define somewhere in reiserfs_fs.h. -Hans */
#define JOURNAL_BLOCK_SIZE  4096 /* BUG gotta get rid of this */
#define JOURNAL_MAX_CNODE   1500 /* max cnodes to allocate. */
#define JOURNAL_HASH_SIZE 8192   
#define JOURNAL_NUM_BITMAPS 5 /* number of copies of the bitmaps to have floating.  Must be >= 2 */

/* One of these for every block in every transaction
** Each one is in two hash tables.  First, a hash of the current transaction, and after journal_end, a
** hash of all the in memory transactions.
** next and prev are used by the current transaction (journal_hash).
** hnext and hprev are used by journal_list_hash.  If a block is in more than one transaction, the journal_list_hash
** links it in multiple times.  This allows flush_journal_list to remove just the cnode belonging
** to a given transaction.
*/
struct reiserfs_journal_cnode {
  struct buffer_head *bh ;		 /* real buffer head */
  struct super_block *sb ;		 /* dev of real buffer head */
  __u32 blocknr ;		 /* block number of real buffer head, == 0 when buffer on disk */
  long state ;
  struct reiserfs_journal_list *jlist ;  /* journal list this cnode lives in */
  struct reiserfs_journal_cnode *next ;  /* next in transaction list */
  struct reiserfs_journal_cnode *prev ;  /* prev in transaction list */
  struct reiserfs_journal_cnode *hprev ; /* prev in hash list */
  struct reiserfs_journal_cnode *hnext ; /* next in hash list */
};

struct reiserfs_bitmap_node {
  int id ;
  char *data ;
  struct list_head list ;
} ;

struct reiserfs_list_bitmap {
  struct reiserfs_journal_list *journal_list ;
  struct reiserfs_bitmap_node **bitmaps ;
} ;

/*
** one of these for each transaction.  The most important part here is the j_realblock.
** this list of cnodes is used to hash all the blocks in all the commits, to mark all the
** real buffer heads dirty once all the commits hit the disk,
** and to make sure every real block in a transaction is on disk before allowing the log area
** to be overwritten */
struct reiserfs_journal_list {
  unsigned long j_start ;
  unsigned long j_state;
  unsigned long j_len ;
  atomic_t j_nonzerolen ;
  atomic_t j_commit_left ;
  atomic_t j_older_commits_done ;      /* all commits older than this on disk*/
  struct semaphore j_commit_lock;
  unsigned long j_trans_id ;
  time_t j_timestamp ;
  struct reiserfs_list_bitmap *j_list_bitmap ;
  struct buffer_head *j_commit_bh ; /* commit buffer head */
  struct reiserfs_journal_cnode *j_realblock  ;
  struct reiserfs_journal_cnode *j_freedlist ; /* list of buffers that were freed during this trans.  free each of these on flush */
  /* time ordered list of all active transactions */
  struct list_head j_list;

  /* time ordered list of all transactions we haven't tried to flush yet */
  struct list_head j_working_list;

  /* list of tail conversion targets in need of flush before commit */
  struct list_head j_tail_bh_list;
  /* list of data=ordered buffers in need of flush before commit */
  struct list_head j_bh_list;
  int j_refcount;
} ;

struct reiserfs_journal {
  struct buffer_head ** j_ap_blocks ; /* journal blocks on disk */
  struct reiserfs_journal_cnode *j_last ; /* newest journal block */
  struct reiserfs_journal_cnode *j_first ; /*  oldest journal block.  start here for traverse */

  struct file         *j_dev_file;
  struct block_device *j_dev_bd;  
  int j_1st_reserved_block;     /* first block on s_dev of reserved area journal */        
	
  long j_state ;			
  unsigned long j_trans_id ;
  unsigned long j_mount_id ;
  unsigned long j_start ;             /* start of current waiting commit (index into j_ap_blocks) */
  unsigned long j_len ;               /* lenght of current waiting commit */
  unsigned long j_len_alloc ;         /* number of buffers requested by journal_begin() */
  atomic_t j_wcount ;            /* count of writers for current commit */
  unsigned long j_bcount ;            /* batch count. allows turning X transactions into 1 */
  unsigned long j_first_unflushed_offset ;  /* first unflushed transactions offset */
  unsigned long j_last_flush_trans_id ;    /* last fully flushed journal timestamp */
  struct buffer_head *j_header_bh ;   

  time_t j_trans_start_time ;         /* time this transaction started */
  struct semaphore j_lock;
  struct semaphore j_flush_sem;
  wait_queue_head_t j_join_wait ;    /* wait for current transaction to finish before starting new one */
  atomic_t j_jlock ;                       /* lock for j_join_wait */
  int j_list_bitmap_index ;	      /* number of next list bitmap to use */
  int j_must_wait ;		       /* no more journal begins allowed. MUST sleep on j_join_wait */
  int j_next_full_flush ;             /* next journal_end will flush all journal list */
  int j_next_async_flush ;             /* next journal_end will flush all async commits */

  int j_cnode_used ;	      /* number of cnodes on the used list */
  int j_cnode_free ;          /* number of cnodes on the free list */

  unsigned int j_trans_max ;           /* max number of blocks in a transaction.  */
  unsigned int j_max_batch ;           /* max number of blocks to batch into a trans */
  unsigned int j_max_commit_age ;      /* in seconds, how old can an async commit be */
  unsigned int j_max_trans_age ;       /* in seconds, how old can a transaction be */
  unsigned int j_default_max_commit_age ; /* the default for the max commit age */

  struct reiserfs_journal_cnode *j_cnode_free_list ;
  struct reiserfs_journal_cnode *j_cnode_free_orig ; /* orig pointer returned from vmalloc */

  struct reiserfs_journal_list *j_current_jl;
  int j_free_bitmap_nodes ;
  int j_used_bitmap_nodes ;

  int j_num_lists;      /* total number of active transactions */
  int j_num_work_lists; /* number that need attention from kreiserfsd */

  /* debugging to make sure things are flushed in order */
  int j_last_flush_id;

  /* debugging to make sure things are committed in order */
  int j_last_commit_id;

  struct list_head j_bitmap_nodes ;
  struct list_head j_dirty_buffers ;
  spinlock_t j_dirty_buffers_lock ; /* protects j_dirty_buffers */

  /* list of all active transactions */
  struct list_head j_journal_list;
  /* lists that haven't been touched by writeback attempts */
  struct list_head j_working_list;

  struct reiserfs_list_bitmap j_list_bitmap[JOURNAL_NUM_BITMAPS] ;	/* array of bitmaps to record the deleted blocks */
  struct reiserfs_journal_cnode *j_hash_table[JOURNAL_HASH_SIZE] ; 	    /* hash table for real buffer heads in current trans */ 
  struct reiserfs_journal_cnode *j_list_hash_table[JOURNAL_HASH_SIZE] ; /* hash table for all the real buffer heads in all 
  										the transactions */
  struct list_head j_prealloc_list;     /* list of inodes which have preallocated blocks */
  int j_persistent_trans;
  unsigned long j_max_trans_size ;
  unsigned long j_max_batch_size ;

  int j_errno;

  /* when flushing ordered buffers, throttle new ordered writers */
  struct work_struct j_work;
  atomic_t j_async_throttle;
};

enum journal_state_bits {
    J_WRITERS_BLOCKED = 1,   /* set when new writers not allowed */
    J_WRITERS_QUEUED,        /* set when log is full due to too many writers */
    J_ABORTED,               /* set when log is aborted */
};


#define JOURNAL_DESC_MAGIC "ReIsErLB" /* ick.  magic string to find desc blocks in the journal */

typedef __u32 (*hashf_t) (const signed char *, int);

struct reiserfs_bitmap_info
{