// SPDX-License-Identifier: GPL-2.0

  /*

   * Write ahead logging implementation copyright Chris Mason 2000
   *
   * The background commits make this code very interrelated, and
   * overly complex.  I need to rethink things a bit....The major players:
   *
   * journal_begin -- call with the number of blocks you expect to log.
   *                  If the current transaction is too
   *		    old, it will block until the current transaction is
   *		    finished, and then start a new one.
   *		    Usually, your transaction will get joined in with
   *                  previous ones for speed.
   *
   * journal_join  -- same as journal_begin, but won't block on the current
   *                  transaction regardless of age.  Don't ever call
   *                  this.  Ever.  There are only two places it should be
   *                  called from, and they are both inside this file.
   *
   * journal_mark_dirty -- adds blocks into this transaction.  clears any flags
   *                       that might make them get sent to disk
   *                       and then marks them BH_JDirty.  Puts the buffer head
   *                       into the current transaction hash.
   *
   * journal_end -- if the current transaction is batchable, it does nothing
   *                   otherwise, it could do an async/synchronous commit, or
   *                   a full flush of all log and real blocks in the
   *                   transaction.
   *
   * flush_old_commits -- if the current transaction is too old, it is ended and
   *                      commit blocks are sent to disk.  Forces commit blocks
   *                      to disk for all backgrounded commits that have been
   *                      around too long.
   *		     -- Note, if you call this as an immediate flush from
   *		        from within kupdate, it will ignore the immediate flag
   */

  #include <linux/time.h>

  #include <linux/semaphore.h>

  #include <linux/vmalloc.h>

  #include "reiserfs.h"

  #include <linux/kernel.h>
  #include <linux/errno.h>
  #include <linux/fcntl.h>
  #include <linux/stat.h>
  #include <linux/string.h>

  #include <linux/buffer_head.h>
  #include <linux/workqueue.h>
  #include <linux/writeback.h>
  #include <linux/blkdev.h>

  #include <linux/backing-dev.h>

  #include <linux/uaccess.h>

  #include <linux/slab.h>

  /* gets a struct reiserfs_journal_list * from a list head */
  #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
                                 j_list))
  #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
                                 j_working_list))

  /* must be correct to keep the desc and commit structs at 4k */
  #define JOURNAL_TRANS_HALF 1018

  #define BUFNR 64		/*read ahead */

  
  /* cnode stat bits.  Move these into reiserfs_fs.h */

  /* this block was freed, and can't be written.  */
  #define BLOCK_FREED 2
  /* this block was freed during this transaction, and can't be written */
  #define BLOCK_FREED_HOLDER 3

  /* used in flush_journal_list */
  #define BLOCK_NEEDS_FLUSH 4

  #define BLOCK_DIRTIED 5

  /* journal list state bits */
  #define LIST_TOUCHED 1
  #define LIST_DIRTY   2

  #define LIST_COMMIT_PENDING  4	/* someone will commit this list */

  
  /* flags for do_journal_end */
  #define FLUSH_ALL   1		/* flush commit and real blocks */
  #define COMMIT_NOW  2		/* end and commit this transaction */

  #define WAIT        4		/* wait for the log blocks to hit the disk */

  static int do_journal_end(struct reiserfs_transaction_handle *, int flags);

  static int flush_journal_list(struct super_block *s,
  			      struct reiserfs_journal_list *jl, int flushall);
  static int flush_commit_list(struct super_block *s,
  			     struct reiserfs_journal_list *jl, int flushall);
  static int can_dirty(struct reiserfs_journal_cnode *cn);
  static int journal_join(struct reiserfs_transaction_handle *th,

  			struct super_block *sb);

  static void release_journal_dev(struct super_block *super,

  			       struct reiserfs_journal *journal);

  static int dirty_one_transaction(struct super_block *s,

  				 struct reiserfs_journal_list *jl);

  static void flush_async_commits(struct work_struct *work);

  static void queue_log_writer(struct super_block *s);
  
  /* values for join in do_journal_begin_r */
  enum {

  	JBEGIN_REG = 0,		/* regular journal begin */

  	/* join the running transaction if at all possible */
  	JBEGIN_JOIN = 1,
  	/* called from cleanup code, ignores aborted flag */
  	JBEGIN_ABORT = 2,

  };
  
  static int do_journal_begin_r(struct reiserfs_transaction_handle *th,

  			      struct super_block *sb,

  			      unsigned long nblocks, int join);

  static void init_journal_hash(struct super_block *sb)

  {

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	memset(journal->j_hash_table, 0,
  	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));

  }
  
  /*

   * clears BH_Dirty and sticks the buffer on the clean list.  Called because
   * I can't allow refile_buffer to make schedule happen after I've freed a
   * block.  Look at remove_from_transaction and journal_mark_freed for
   * more details.
   */

  static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
  {
  	if (bh) {
  		clear_buffer_dirty(bh);
  		clear_buffer_journal_test(bh);
  	}
  	return 0;

  }

  static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block

  							 *sb)

  {
  	struct reiserfs_bitmap_node *bn;
  	static int id;

  	bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);

  	if (!bn) {
  		return NULL;
  	}

  	bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);

  	if (!bn->data) {

  		kfree(bn);

  		return NULL;
  	}
  	bn->id = id++;

  	INIT_LIST_HEAD(&bn->list);
  	return bn;
  }

  static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)

  {

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	struct reiserfs_bitmap_node *bn = NULL;
  	struct list_head *entry = journal->j_bitmap_nodes.next;
  
  	journal->j_used_bitmap_nodes++;

  repeat:

  
  	if (entry != &journal->j_bitmap_nodes) {
  		bn = list_entry(entry, struct reiserfs_bitmap_node, list);
  		list_del(entry);

  		memset(bn->data, 0, sb->s_blocksize);

  		journal->j_free_bitmap_nodes--;
  		return bn;
  	}

  	bn = allocate_bitmap_node(sb);

  	if (!bn) {
  		yield();
  		goto repeat;
  	}
  	return bn;

  }

  static inline void free_bitmap_node(struct super_block *sb,

  				    struct reiserfs_bitmap_node *bn)
  {

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	journal->j_used_bitmap_nodes--;
  	if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {

  		kfree(bn->data);
  		kfree(bn);

  	} else {
  		list_add(&bn->list, &journal->j_bitmap_nodes);
  		journal->j_free_bitmap_nodes++;
  	}
  }

  static void allocate_bitmap_nodes(struct super_block *sb)

  {
  	int i;

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	struct reiserfs_bitmap_node *bn = NULL;
  	for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {

  		bn = allocate_bitmap_node(sb);

  		if (bn) {
  			list_add(&bn->list, &journal->j_bitmap_nodes);
  			journal->j_free_bitmap_nodes++;
  		} else {

  			/* this is ok, we'll try again when more are needed */
  			break;

  		}
  	}

  }

  static int set_bit_in_list_bitmap(struct super_block *sb,

  				  b_blocknr_t block,

  				  struct reiserfs_list_bitmap *jb)
  {

  	unsigned int bmap_nr = block / (sb->s_blocksize << 3);
  	unsigned int bit_nr = block % (sb->s_blocksize << 3);

  	if (!jb->bitmaps[bmap_nr]) {

  		jb->bitmaps[bmap_nr] = get_bitmap_node(sb);

  	}
  	set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
  	return 0;

  }

  static void cleanup_bitmap_list(struct super_block *sb,

  				struct reiserfs_list_bitmap *jb)
  {
  	int i;
  	if (jb->bitmaps == NULL)
  		return;

  	for (i = 0; i < reiserfs_bmap_count(sb); i++) {

  		if (jb->bitmaps[i]) {

  			free_bitmap_node(sb, jb->bitmaps[i]);

  			jb->bitmaps[i] = NULL;
  		}
  	}

  }
  
  /*

   * only call this on FS unmount.
   */

  static int free_list_bitmaps(struct super_block *sb,

  			     struct reiserfs_list_bitmap *jb_array)
  {
  	int i;
  	struct reiserfs_list_bitmap *jb;
  	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
  		jb = jb_array + i;
  		jb->journal_list = NULL;

  		cleanup_bitmap_list(sb, jb);

  		vfree(jb->bitmaps);
  		jb->bitmaps = NULL;
  	}
  	return 0;
  }

  static int free_bitmap_nodes(struct super_block *sb)

  {

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	struct list_head *next = journal->j_bitmap_nodes.next;
  	struct reiserfs_bitmap_node *bn;
  
  	while (next != &journal->j_bitmap_nodes) {
  		bn = list_entry(next, struct reiserfs_bitmap_node, list);
  		list_del(next);

  		kfree(bn->data);
  		kfree(bn);

  		next = journal->j_bitmap_nodes.next;
  		journal->j_free_bitmap_nodes--;
  	}
  
  	return 0;

  }
  
  /*

   * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
   * jb_array is the array to be filled in.
   */

  int reiserfs_allocate_list_bitmaps(struct super_block *sb,

  				   struct reiserfs_list_bitmap *jb_array,

  				   unsigned int bmap_nr)

  {
  	int i;
  	int failed = 0;
  	struct reiserfs_list_bitmap *jb;
  	int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
  
  	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
  		jb = jb_array + i;
  		jb->journal_list = NULL;

  		jb->bitmaps = vzalloc(mem);

  		if (!jb->bitmaps) {

  			reiserfs_warning(sb, "clm-2000", "unable to "

  					 "allocate bitmaps for journal lists");

  			failed = 1;
  			break;
  		}

  	}
  	if (failed) {

  		free_list_bitmaps(sb, jb_array);

  		return -1;
  	}
  	return 0;

  }
  
  /*

   * find an available list bitmap.  If you can't find one, flush a commit list
   * and try again
   */

  static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,

  						    struct reiserfs_journal_list
  						    *jl)
  {
  	int i, j;

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	struct reiserfs_list_bitmap *jb = NULL;
  
  	for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
  		i = journal->j_list_bitmap_index;
  		journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
  		jb = journal->j_list_bitmap + i;
  		if (journal->j_list_bitmap[i].journal_list) {

  			flush_commit_list(sb,

  					  journal->j_list_bitmap[i].
  					  journal_list, 1);
  			if (!journal->j_list_bitmap[i].journal_list) {
  				break;
  			}
  		} else {
  			break;
  		}
  	}

  	/* double check to make sure if flushed correctly */
  	if (jb->journal_list)

  		return NULL;

  	jb->journal_list = jl;
  	return jb;

  }

  /*

   * allocates a new chunk of X nodes, and links them all together as a list.
   * Uses the cnode->next and cnode->prev pointers
   * returns NULL on failure
   */

  static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
  {
  	struct reiserfs_journal_cnode *head;
  	int i;
  	if (num_cnodes <= 0) {
  		return NULL;
  	}

  	head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));

  	if (!head) {
  		return NULL;
  	}

  	head[0].prev = NULL;
  	head[0].next = head + 1;
  	for (i = 1; i < num_cnodes; i++) {
  		head[i].prev = head + (i - 1);
  		head[i].next = head + (i + 1);	/* if last one, overwrite it after the if */
  	}
  	head[num_cnodes - 1].next = NULL;
  	return head;

  }

  /* pulls a cnode off the free list, or returns NULL on failure */

  static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)

  {
  	struct reiserfs_journal_cnode *cn;

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	reiserfs_check_lock_depth(sb, "get_cnode");

  
  	if (journal->j_cnode_free <= 0) {
  		return NULL;
  	}
  	journal->j_cnode_used++;
  	journal->j_cnode_free--;
  	cn = journal->j_cnode_free_list;
  	if (!cn) {
  		return cn;
  	}
  	if (cn->next) {
  		cn->next->prev = NULL;
  	}
  	journal->j_cnode_free_list = cn->next;
  	memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
  	return cn;

  }
  
  /*

   * returns a cnode to the free list
   */

  static void free_cnode(struct super_block *sb,

  		       struct reiserfs_journal_cnode *cn)
  {

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	reiserfs_check_lock_depth(sb, "free_cnode");

  	journal->j_cnode_used--;
  	journal->j_cnode_free++;
  	/* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
  	cn->next = journal->j_cnode_free_list;
  	if (journal->j_cnode_free_list) {
  		journal->j_cnode_free_list->prev = cn;
  	}
  	cn->prev = NULL;	/* not needed with the memset, but I might kill the memset, and forget to do this */
  	journal->j_cnode_free_list = cn;

  }

  static void clear_prepared_bits(struct buffer_head *bh)
  {
  	clear_buffer_journal_prepared(bh);
  	clear_buffer_journal_restore_dirty(bh);

  }

  /*
   * return a cnode with same dev, block number and size in table,
   * or null if not found
   */

  static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
  								  super_block
  								  *sb,
  								  struct
  								  reiserfs_journal_cnode
  								  **table,
  								  long bl)

  {

  	struct reiserfs_journal_cnode *cn;
  	cn = journal_hash(table, sb, bl);
  	while (cn) {
  		if (cn->blocknr == bl && cn->sb == sb)
  			return cn;
  		cn = cn->hnext;
  	}
  	return (struct reiserfs_journal_cnode *)0;

  }
  
  /*

   * this actually means 'can this block be reallocated yet?'.  If you set
   * search_all, a block can only be allocated if it is not in the current
   * transaction, was not freed by the current transaction, and has no chance
   * of ever being overwritten by a replay after crashing.
   *
   * If you don't set search_all, a block can only be allocated if it is not
   * in the current transaction.  Since deleting a block removes it from the
   * current transaction, this case should never happen.  If you don't set
   * search_all, make sure you never write the block without logging it.
   *
   * next_zero_bit is a suggestion about the next block to try for find_forward.
   * when bl is rejected because it is set in a journal list bitmap, we search
   * for the next zero bit in the bitmap that rejected bl.  Then, we return
   * that through next_zero_bit for find_forward to try.
   *
   * Just because we return something in next_zero_bit does not mean we won't
   * reject it on the next call to reiserfs_in_journal
   */

  int reiserfs_in_journal(struct super_block *sb,

  			unsigned int bmap_nr, int bit_nr, int search_all,

  			b_blocknr_t * next_zero_bit)
  {

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	struct reiserfs_journal_cnode *cn;
  	struct reiserfs_list_bitmap *jb;
  	int i;
  	unsigned long bl;
  
  	*next_zero_bit = 0;	/* always start this at zero. */

  	PROC_INFO_INC(sb, journal.in_journal);

  	/*
  	 * If we aren't doing a search_all, this is a metablock, and it
  	 * will be logged before use.  if we crash before the transaction
  	 * that freed it commits,  this transaction won't have committed
  	 * either, and the block will never be written

  	 */
  	if (search_all) {
  		for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {

  			PROC_INFO_INC(sb, journal.in_journal_bitmap);

  			jb = journal->j_list_bitmap + i;
  			if (jb->journal_list && jb->bitmaps[bmap_nr] &&
  			    test_bit(bit_nr,
  				     (unsigned long *)jb->bitmaps[bmap_nr]->
  				     data)) {
  				*next_zero_bit =
  				    find_next_zero_bit((unsigned long *)
  						       (jb->bitmaps[bmap_nr]->
  							data),

  						       sb->s_blocksize << 3,

  						       bit_nr + 1);
  				return 1;
  			}
  		}
  	}

  	bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;

  	/* is it in any old transactions? */
  	if (search_all
  	    && (cn =

  		get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {

  		return 1;
  	}
  
  	/* is it in the current transaction.  This should never happen */

  	if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {

  		BUG();
  		return 1;
  	}

  	PROC_INFO_INC(sb, journal.in_journal_reusable);

  	/* safe for reuse */
  	return 0;

  }

  /* insert cn into table */

  static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
  				       struct reiserfs_journal_cnode *cn)
  {
  	struct reiserfs_journal_cnode *cn_orig;

  	cn_orig = journal_hash(table, cn->sb, cn->blocknr);
  	cn->hnext = cn_orig;
  	cn->hprev = NULL;
  	if (cn_orig) {
  		cn_orig->hprev = cn;
  	}
  	journal_hash(table, cn->sb, cn->blocknr) = cn;

  }
  
  /* lock the current transaction */

  static inline void lock_journal(struct super_block *sb)

  {

  	PROC_INFO_INC(sb, journal.lock_journal);

  
  	reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);

  }
  
  /* unlock the current transaction */

  static inline void unlock_journal(struct super_block *sb)

  {

  	mutex_unlock(&SB_JOURNAL(sb)->j_mutex);

  }
  
  static inline void get_journal_list(struct reiserfs_journal_list *jl)
  {

  	jl->j_refcount++;

  }
  
  static inline void put_journal_list(struct super_block *s,

  				    struct reiserfs_journal_list *jl)

  {

  	if (jl->j_refcount < 1) {

  		reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",

  			       jl->j_trans_id, jl->j_refcount);
  	}
  	if (--jl->j_refcount == 0)

  		kfree(jl);

  }
  
  /*

   * this used to be much more involved, and I'm keeping it just in case
   * things get ugly again.  it gets called by flush_commit_list, and
   * cleans up any data stored about blocks freed during a transaction.
   */

  static void cleanup_freed_for_journal_list(struct super_block *sb,

  					   struct reiserfs_journal_list *jl)
  {

  	struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
  	if (jb) {

  		cleanup_bitmap_list(sb, jb);

  	}
  	jl->j_list_bitmap->journal_list = NULL;
  	jl->j_list_bitmap = NULL;

  }
  
  static int journal_list_still_alive(struct super_block *s,

  				    unsigned int trans_id)

  {
  	struct reiserfs_journal *journal = SB_JOURNAL(s);
  	struct list_head *entry = &journal->j_journal_list;
  	struct reiserfs_journal_list *jl;
  
  	if (!list_empty(entry)) {
  		jl = JOURNAL_LIST_ENTRY(entry->next);
  		if (jl->j_trans_id <= trans_id) {
  			return 1;
  		}
  	}
  	return 0;
  }

  /*
   * If page->mapping was null, we failed to truncate this page for
   * some reason.  Most likely because it was truncated after being
   * logged via data=journal.
   *
   * This does a check to see if the buffer belongs to one of these
   * lost pages before doing the final put_bh.  If page->mapping was
   * null, it tries to free buffers on the page, which should make the

   * final put_page drop the page from the lru.

   */
  static void release_buffer_page(struct buffer_head *bh)
  {
  	struct page *page = bh->b_page;

  	if (!page->mapping && trylock_page(page)) {

  		get_page(page);

  		put_bh(bh);
  		if (!page->mapping)
  			try_to_free_buffers(page);
  		unlock_page(page);

  		put_page(page);

  	} else {
  		put_bh(bh);
  	}
  }

  static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
  {

  	if (buffer_journaled(bh)) {

  		reiserfs_warning(NULL, "clm-2084",

  				 "pinned buffer %lu:%pg sent to disk",
  				 bh->b_blocknr, bh->b_bdev);

  	}
  	if (uptodate)
  		set_buffer_uptodate(bh);
  	else
  		clear_buffer_uptodate(bh);

  	unlock_buffer(bh);

  	release_buffer_page(bh);

  }
  
  static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
  {
  	if (uptodate)
  		set_buffer_uptodate(bh);
  	else
  		clear_buffer_uptodate(bh);
  	unlock_buffer(bh);
  	put_bh(bh);
  }
  
  static void submit_logged_buffer(struct buffer_head *bh)
  {
  	get_bh(bh);
  	bh->b_end_io = reiserfs_end_buffer_io_sync;
  	clear_buffer_journal_new(bh);
  	clear_buffer_dirty(bh);
  	if (!test_clear_buffer_journal_test(bh))
  		BUG();
  	if (!buffer_uptodate(bh))
  		BUG();

  	submit_bh(REQ_OP_WRITE, 0, bh);

  }
  
  static void submit_ordered_buffer(struct buffer_head *bh)
  {
  	get_bh(bh);
  	bh->b_end_io = reiserfs_end_ordered_io;
  	clear_buffer_dirty(bh);
  	if (!buffer_uptodate(bh))
  		BUG();

  	submit_bh(REQ_OP_WRITE, 0, bh);

  }

  #define CHUNK_SIZE 32
  struct buffer_chunk {

  	struct buffer_head *bh[CHUNK_SIZE];
  	int nr;

  };

  static void write_chunk(struct buffer_chunk *chunk)
  {
  	int i;

  	for (i = 0; i < chunk->nr; i++) {
  		submit_logged_buffer(chunk->bh[i]);
  	}
  	chunk->nr = 0;

  }

  static void write_ordered_chunk(struct buffer_chunk *chunk)
  {
  	int i;

  	for (i = 0; i < chunk->nr; i++) {
  		submit_ordered_buffer(chunk->bh[i]);
  	}
  	chunk->nr = 0;

  }
  
  static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,

  			spinlock_t * lock, void (fn) (struct buffer_chunk *))

  {

  	int ret = 0;

  	BUG_ON(chunk->nr >= CHUNK_SIZE);

  	chunk->bh[chunk->nr++] = bh;
  	if (chunk->nr >= CHUNK_SIZE) {
  		ret = 1;

  		if (lock) {

  			spin_unlock(lock);

  			fn(chunk);

  			spin_lock(lock);

  		} else {
  			fn(chunk);
  		}

  	}
  	return ret;

  }

  static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);

  static struct reiserfs_jh *alloc_jh(void)
  {
  	struct reiserfs_jh *jh;
  	while (1) {
  		jh = kmalloc(sizeof(*jh), GFP_NOFS);
  		if (jh) {
  			atomic_inc(&nr_reiserfs_jh);
  			return jh;
  		}
  		yield();

  	}

  }
  
  /*
   * we want to free the jh when the buffer has been written
   * and waited on
   */

  void reiserfs_free_jh(struct buffer_head *bh)
  {
  	struct reiserfs_jh *jh;
  
  	jh = bh->b_private;
  	if (jh) {
  		bh->b_private = NULL;
  		jh->bh = NULL;
  		list_del_init(&jh->list);
  		kfree(jh);
  		if (atomic_read(&nr_reiserfs_jh) <= 0)
  			BUG();
  		atomic_dec(&nr_reiserfs_jh);
  		put_bh(bh);
  	}

  }
  
  static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,

  			   int tail)

  {

  	struct reiserfs_jh *jh;

  	if (bh->b_private) {
  		spin_lock(&j->j_dirty_buffers_lock);
  		if (!bh->b_private) {
  			spin_unlock(&j->j_dirty_buffers_lock);
  			goto no_jh;
  		}
  		jh = bh->b_private;
  		list_del_init(&jh->list);
  	} else {

  no_jh:

  		get_bh(bh);
  		jh = alloc_jh();
  		spin_lock(&j->j_dirty_buffers_lock);

  		/*
  		 * buffer must be locked for __add_jh, should be able to have

  		 * two adds at the same time
  		 */

  		BUG_ON(bh->b_private);

  		jh->bh = bh;
  		bh->b_private = jh;

  	}

  	jh->jl = j->j_current_jl;
  	if (tail)
  		list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
  	else {
  		list_add_tail(&jh->list, &jh->jl->j_bh_list);
  	}
  	spin_unlock(&j->j_dirty_buffers_lock);
  	return 0;

  }

  int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
  {
  	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);

  }

  int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
  {
  	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);

  }
  
  #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)

  static int write_ordered_buffers(spinlock_t * lock,

  				 struct reiserfs_journal *j,

  				 struct reiserfs_journal_list *jl,

  				 struct list_head *list)
  {

  	struct buffer_head *bh;
  	struct reiserfs_jh *jh;
  	int ret = j->j_errno;
  	struct buffer_chunk chunk;
  	struct list_head tmp;
  	INIT_LIST_HEAD(&tmp);
  
  	chunk.nr = 0;
  	spin_lock(lock);
  	while (!list_empty(list)) {
  		jh = JH_ENTRY(list->next);
  		bh = jh->bh;
  		get_bh(bh);

  		if (!trylock_buffer(bh)) {

  			if (!buffer_dirty(bh)) {

  				list_move(&jh->list, &tmp);

  				goto loop_next;
  			}
  			spin_unlock(lock);
  			if (chunk.nr)
  				write_ordered_chunk(&chunk);
  			wait_on_buffer(bh);
  			cond_resched();
  			spin_lock(lock);
  			goto loop_next;
  		}

  		/*
  		 * in theory, dirty non-uptodate buffers should never get here,

  		 * but the upper layer io error paths still have a few quirks.
  		 * Handle them here as gracefully as we can
  		 */
  		if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
  			clear_buffer_dirty(bh);
  			ret = -EIO;
  		}

  		if (buffer_dirty(bh)) {

  			list_move(&jh->list, &tmp);

  			add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
  		} else {
  			reiserfs_free_jh(bh);
  			unlock_buffer(bh);
  		}

  loop_next:

  		put_bh(bh);
  		cond_resched_lock(lock);
  	}
  	if (chunk.nr) {
  		spin_unlock(lock);

  		write_ordered_chunk(&chunk);

  		spin_lock(lock);

  	}

  	while (!list_empty(&tmp)) {
  		jh = JH_ENTRY(tmp.prev);
  		bh = jh->bh;
  		get_bh(bh);
  		reiserfs_free_jh(bh);
  
  		if (buffer_locked(bh)) {
  			spin_unlock(lock);
  			wait_on_buffer(bh);
  			spin_lock(lock);
  		}
  		if (!buffer_uptodate(bh)) {
  			ret = -EIO;
  		}

  		/*
  		 * ugly interaction with invalidatepage here.
  		 * reiserfs_invalidate_page will pin any buffer that has a
  		 * valid journal head from an older transaction.  If someone
  		 * else sets our buffer dirty after we write it in the first
  		 * loop, and then someone truncates the page away, nobody
  		 * will ever write the buffer. We're safe if we write the
  		 * page one last time after freeing the journal header.

  		 */
  		if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
  			spin_unlock(lock);

  			ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);

  			spin_lock(lock);
  		}

  		put_bh(bh);
  		cond_resched_lock(lock);

  	}

  	spin_unlock(lock);
  	return ret;
  }

  static int flush_older_commits(struct super_block *s,
  			       struct reiserfs_journal_list *jl)
  {
  	struct reiserfs_journal *journal = SB_JOURNAL(s);
  	struct reiserfs_journal_list *other_jl;
  	struct reiserfs_journal_list *first_jl;
  	struct list_head *entry;

  	unsigned int trans_id = jl->j_trans_id;
  	unsigned int other_trans_id;
  	unsigned int first_trans_id;

  find_first:

  	/*
  	 * first we walk backwards to find the oldest uncommitted transation
  	 */
  	first_jl = jl;
  	entry = jl->j_list.prev;
  	while (1) {
  		other_jl = JOURNAL_LIST_ENTRY(entry);
  		if (entry == &journal->j_journal_list ||
  		    atomic_read(&other_jl->j_older_commits_done))
  			break;

  		first_jl = other_jl;
  		entry = other_jl->j_list.prev;
  	}

  	/* if we didn't find any older uncommitted transactions, return now */
  	if (first_jl == jl) {
  		return 0;
  	}

  	first_trans_id = first_jl->j_trans_id;
  
  	entry = &first_jl->j_list;
  	while (1) {
  		other_jl = JOURNAL_LIST_ENTRY(entry);
  		other_trans_id = other_jl->j_trans_id;
  
  		if (other_trans_id < trans_id) {
  			if (atomic_read(&other_jl->j_commit_left) != 0) {
  				flush_commit_list(s, other_jl, 0);
  
  				/* list we were called with is gone, return */
  				if (!journal_list_still_alive(s, trans_id))
  					return 1;

  				/*
  				 * the one we just flushed is gone, this means
  				 * all older lists are also gone, so first_jl
  				 * is no longer valid either.  Go back to the
  				 * beginning.

  				 */
  				if (!journal_list_still_alive
  				    (s, other_trans_id)) {
  					goto find_first;
  				}
  			}
  			entry = entry->next;
  			if (entry == &journal->j_journal_list)
  				return 0;
  		} else {
  			return 0;

  		}

  	}

  	return 0;

  }

  
  static int reiserfs_async_progress_wait(struct super_block *s)

  {

  	struct reiserfs_journal *j = SB_JOURNAL(s);

  
  	if (atomic_read(&j->j_async_throttle)) {

  		int depth;
  
  		depth = reiserfs_write_unlock_nested(s);

  		congestion_wait(BLK_RW_ASYNC, HZ / 10);

  		reiserfs_write_lock_nested(s, depth);

  	}

  	return 0;

  }
  
  /*

   * if this journal list still has commit blocks unflushed, send them to disk.
   *
   * log areas must be flushed in order (transaction 2 can't commit before
   * transaction 1) Before the commit block can by written, every other log
   * block must be safely on disk
   */

  static int flush_commit_list(struct super_block *s,
  			     struct reiserfs_journal_list *jl, int flushall)
  {
  	int i;

  	b_blocknr_t bn;

  	struct buffer_head *tbh = NULL;

  	unsigned int trans_id = jl->j_trans_id;

  	struct reiserfs_journal *journal = SB_JOURNAL(s);

  	int retval = 0;

  	int write_len;

  	int depth;

  
  	reiserfs_check_lock_depth(s, "flush_commit_list");
  
  	if (atomic_read(&jl->j_older_commits_done)) {
  		return 0;
  	}

  	/*
  	 * before we can put our commit blocks on disk, we have to make
  	 * sure everyone older than us is on disk too

  	 */
  	BUG_ON(jl->j_len <= 0);
  	BUG_ON(trans_id == journal->j_trans_id);
  
  	get_journal_list(jl);
  	if (flushall) {
  		if (flush_older_commits(s, jl) == 1) {

  			/*
  			 * list disappeared during flush_older_commits.
  			 * return
  			 */

  			goto put_jl;
  		}
  	}
  
  	/* make sure nobody is trying to flush this one at the same time */

  	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);

  	if (!journal_list_still_alive(s, trans_id)) {

  		mutex_unlock(&jl->j_commit_mutex);

  		goto put_jl;
  	}
  	BUG_ON(jl->j_trans_id == 0);
  
  	/* this commit is done, exit */

  	if (atomic_read(&jl->j_commit_left) <= 0) {

  		if (flushall) {

  			atomic_set(&jl->j_older_commits_done, 1);

  		}

  		mutex_unlock(&jl->j_commit_mutex);

  		goto put_jl;
  	}
  
  	if (!list_empty(&jl->j_bh_list)) {

  		int ret;

  
  		/*
  		 * We might sleep in numerous places inside
  		 * write_ordered_buffers. Relax the write lock.
  		 */

  		depth = reiserfs_write_unlock_nested(s);

  		ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
  					    journal, jl, &jl->j_bh_list);
  		if (ret < 0 && retval == 0)
  			retval = ret;

  		reiserfs_write_lock_nested(s, depth);

  	}
  	BUG_ON(!list_empty(&jl->j_bh_list));
  	/*
  	 * for the description block and all the log blocks, submit any buffers

  	 * that haven't already reached the disk.  Try to write at least 256
  	 * log blocks. later on, we will only wait on blocks that correspond
  	 * to this transaction, but while we're unplugging we might as well
  	 * get a chunk of data on there.

  	 */
  	atomic_inc(&journal->j_async_throttle);

  	write_len = jl->j_len + 1;
  	if (write_len < 256)
  		write_len = 256;
  	for (i = 0 ; i < write_len ; i++) {

  		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
  		    SB_ONDISK_JOURNAL_SIZE(s);
  		tbh = journal_find_get_block(s, bn);

  		if (tbh) {

  			if (buffer_dirty(tbh)) {

  		            depth = reiserfs_write_unlock_nested(s);

  			    ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);

  			    reiserfs_write_lock_nested(s, depth);

  			}

  			put_bh(tbh) ;
  		}

  	}
  	atomic_dec(&journal->j_async_throttle);

  	for (i = 0; i < (jl->j_len + 1); i++) {
  		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
  		    (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
  		tbh = journal_find_get_block(s, bn);

  		depth = reiserfs_write_unlock_nested(s);
  		__wait_on_buffer(tbh);
  		reiserfs_write_lock_nested(s, depth);

  		/*
  		 * since we're using ll_rw_blk above, it might have skipped
  		 * over a locked buffer.  Double check here
  		 */

  		/* redundant, sync_dirty_buffer() checks */
  		if (buffer_dirty(tbh)) {

  			depth = reiserfs_write_unlock_nested(s);

  			sync_dirty_buffer(tbh);

  			reiserfs_write_lock_nested(s, depth);

  		}

  		if (unlikely(!buffer_uptodate(tbh))) {

  #ifdef CONFIG_REISERFS_CHECK

  			reiserfs_warning(s, "journal-601",
  					 "buffer write failed");

  #endif

  			retval = -EIO;
  		}

  		/* once for journal_find_get_block */
  		put_bh(tbh);
  		/* once due to original getblk in do_journal_end */
  		put_bh(tbh);

  		atomic_dec(&jl->j_commit_left);

  	}

  	BUG_ON(atomic_read(&jl->j_commit_left) != 1);

  	/*
  	 * If there was a write error in the journal - we can't commit

  	 * this transaction - it will be invalid and, if successful,
  	 * will just end up propagating the write error out to

  	 * the file system.
  	 */

  	if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
  		if (buffer_dirty(jl->j_commit_bh))
  			BUG();
  		mark_buffer_dirty(jl->j_commit_bh) ;

  		depth = reiserfs_write_unlock_nested(s);

  		if (reiserfs_barrier_flush(s))

  			__sync_dirty_buffer(jl->j_commit_bh,

  					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);

  		else
  			sync_dirty_buffer(jl->j_commit_bh);

  		reiserfs_write_lock_nested(s, depth);

  	}

  	/*
  	 * If there was a write error in the journal - we can't commit this

  	 * transaction - it will be invalid and, if successful, will just end

  	 * up propagating the write error out to the filesystem.
  	 */

  	if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {

  #ifdef CONFIG_REISERFS_CHECK

  		reiserfs_warning(s, "journal-615", "buffer write failed");

  #endif

  		retval = -EIO;
  	}
  	bforget(jl->j_commit_bh);
  	if (journal->j_last_commit_id != 0 &&
  	    (jl->j_trans_id - journal->j_last_commit_id) != 1) {

  		reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",

  				 journal->j_last_commit_id, jl->j_trans_id);
  	}
  	journal->j_last_commit_id = jl->j_trans_id;

  	/*
  	 * now, every commit block is on the disk.  It is safe to allow
  	 * blocks freed during this transaction to be reallocated
  	 */

  	cleanup_freed_for_journal_list(s, jl);
  
  	retval = retval ? retval : journal->j_errno;
  
  	/* mark the metadata dirty */
  	if (!retval)
  		dirty_one_transaction(s, jl);

  	atomic_dec(&jl->j_commit_left);

  
  	if (flushall) {

  		atomic_set(&jl->j_older_commits_done, 1);

  	}

  	mutex_unlock(&jl->j_commit_mutex);

  put_jl:

  	put_journal_list(s, jl);
  
  	if (retval)
  		reiserfs_abort(s, retval, "Journal write error in %s",

  			       __func__);

  	return retval;

  }
  
  /*

   * flush_journal_list frequently needs to find a newer transaction for a
   * given block.  This does that, or returns NULL if it can't find anything
   */

  static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
  							  reiserfs_journal_cnode
  							  *cn)
  {
  	struct super_block *sb = cn->sb;
  	b_blocknr_t blocknr = cn->blocknr;

  	cn = cn->hprev;
  	while (cn) {
  		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
  			return cn->jlist;
  		}
  		cn = cn->hprev;
  	}
  	return NULL;

  }

  static void remove_journal_hash(struct super_block *,
  				struct reiserfs_journal_cnode **,
  				struct reiserfs_journal_list *, unsigned long,
  				int);

  
  /*

   * once all the real blocks have been flushed, it is safe to remove them
   * from the journal list for this transaction.  Aside from freeing the
   * cnode, this also allows the block to be reallocated for data blocks
   * if it had been deleted.
   */

  static void remove_all_from_journal_list(struct super_block *sb,

  					 struct reiserfs_journal_list *jl,
  					 int debug)
  {

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	struct reiserfs_journal_cnode *cn, *last;
  	cn = jl->j_realblock;

  	/*
  	 * which is better, to lock once around the whole loop, or
  	 * to lock for each call to remove_journal_hash?

  	 */
  	while (cn) {
  		if (cn->blocknr != 0) {
  			if (debug) {

  				reiserfs_warning(sb, "reiserfs-2201",

  						 "block %u, bh is %d, state %ld",
  						 cn->blocknr, cn->bh ? 1 : 0,
  						 cn->state);
  			}
  			cn->state = 0;

  			remove_journal_hash(sb, journal->j_list_hash_table,

  					    jl, cn->blocknr, 1);
  		}
  		last = cn;
  		cn = cn->next;

  		free_cnode(sb, last);

  	}
  	jl->j_realblock = NULL;

  }
  
  /*

   * if this timestamp is greater than the timestamp we wrote last to the
   * header block, write it to the header block.  once this is done, I can
   * safely say the log area for this transaction won't ever be replayed,
   * and I can start releasing blocks in this transaction for reuse as data
   * blocks.  called by flush_journal_list, before it calls
   * remove_all_from_journal_list
   */

  static int _update_journal_header_block(struct super_block *sb,

  					unsigned long offset,

  					unsigned int trans_id)

  {
  	struct reiserfs_journal_header *jh;

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	int depth;

  	if (reiserfs_is_journal_aborted(journal))
  		return -EIO;

  	if (trans_id >= journal->j_last_flush_trans_id) {
  		if (buffer_locked((journal->j_header_bh))) {

  			depth = reiserfs_write_unlock_nested(sb);
  			__wait_on_buffer(journal->j_header_bh);
  			reiserfs_write_lock_nested(sb, depth);

  			if (unlikely(!buffer_uptodate(journal->j_header_bh))) {

  #ifdef CONFIG_REISERFS_CHECK

  				reiserfs_warning(sb, "journal-699",

  						 "buffer write failed");

  #endif

  				return -EIO;
  			}
  		}
  		journal->j_last_flush_trans_id = trans_id;
  		journal->j_first_unflushed_offset = offset;
  		jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
  							b_data);
  		jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
  		jh->j_first_unflushed_offset = cpu_to_le32(offset);
  		jh->j_mount_id = cpu_to_le32(journal->j_mount_id);

  		set_buffer_dirty(journal->j_header_bh);

  		depth = reiserfs_write_unlock_nested(sb);

  
  		if (reiserfs_barrier_flush(sb))

  			__sync_dirty_buffer(journal->j_header_bh,

  					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);

  		else

  			sync_dirty_buffer(journal->j_header_bh);

  		reiserfs_write_lock_nested(sb, depth);

  		if (!buffer_uptodate(journal->j_header_bh)) {

  			reiserfs_warning(sb, "journal-837",

  					 "IO error during journal replay");

  			return -EIO;
  		}
  	}
  	return 0;
  }

  static int update_journal_header_block(struct super_block *sb,

  				       unsigned long offset,

  				       unsigned int trans_id)

  {

  	return _update_journal_header_block(sb, offset, trans_id);

  }

  /*
  ** flush any and all journal lists older than you are

  ** can only be called from flush_journal_list
  */

  static int flush_older_journal_lists(struct super_block *sb,

  				     struct reiserfs_journal_list *jl)
  {
  	struct list_head *entry;
  	struct reiserfs_journal_list *other_jl;

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	unsigned int trans_id = jl->j_trans_id;

  	/*
  	 * we know we are the only ones flushing things, no extra race

  	 * protection is required.
  	 */

  restart:

  	entry = journal->j_journal_list.next;
  	/* Did we wrap? */
  	if (entry == &journal->j_journal_list)
  		return 0;
  	other_jl = JOURNAL_LIST_ENTRY(entry);
  	if (other_jl->j_trans_id < trans_id) {
  		BUG_ON(other_jl->j_refcount <= 0);
  		/* do not flush all */

  		flush_journal_list(sb, other_jl, 0);

  
  		/* other_jl is now deleted from the list */
  		goto restart;
  	}
  	return 0;

  }
  
  static void del_from_work_list(struct super_block *s,

  			       struct reiserfs_journal_list *jl)
  {
  	struct reiserfs_journal *journal = SB_JOURNAL(s);
  	if (!list_empty(&jl->j_working_list)) {
  		list_del_init(&jl->j_working_list);
  		journal->j_num_work_lists--;
  	}

  }

  /*
   * flush a journal list, both commit and real blocks
   *
   * always set flushall to 1, unless you are calling from inside
   * flush_journal_list
   *
   * IMPORTANT.  This can only be called while there are no journal writers,
   * and the journal is locked.  That means it can only be called from
   * do_journal_end, or by journal_release
   */

  static int flush_journal_list(struct super_block *s,
  			      struct reiserfs_journal_list *jl, int flushall)

  {

  	struct reiserfs_journal_list *pjl;
  	struct reiserfs_journal_cnode *cn, *last;
  	int count;
  	int was_jwait = 0;
  	int was_dirty = 0;
  	struct buffer_head *saved_bh;
  	unsigned long j_len_saved = jl->j_len;
  	struct reiserfs_journal *journal = SB_JOURNAL(s);
  	int err = 0;

  	int depth;

  
  	BUG_ON(j_len_saved <= 0);
  
  	if (atomic_read(&journal->j_wcount) != 0) {

  		reiserfs_warning(s, "clm-2048", "called with wcount %d",

  				 atomic_read(&journal->j_wcount));
  	}

  	/* if flushall == 0, the lock is already held */
  	if (flushall) {

  		reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);

  	} else if (mutex_trylock(&journal->j_flush_mutex)) {

  		BUG();
  	}

  	count = 0;
  	if (j_len_saved > journal->j_trans_max) {

  		reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",

  			       j_len_saved, jl->j_trans_id);
  		return 0;
  	}

  	/* if all the work is already done, get out of here */

  	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
  	    atomic_read(&jl->j_commit_left) <= 0) {

  		goto flush_older_and_return;
  	}

  	/*
  	 * start by putting the commit list on disk.  This will also flush
  	 * the commit lists of any olders transactions

  	 */
  	flush_commit_list(s, jl, 1);
  
  	if (!(jl->j_state & LIST_DIRTY)
  	    && !reiserfs_is_journal_aborted(journal))
  		BUG();
  
  	/* are we done now? */

  	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
  	    atomic_read(&jl->j_commit_left) <= 0) {

  		goto flush_older_and_return;
  	}

  	/*
  	 * loop through each cnode, see if we need to write it,
  	 * or wait on a more recent transaction, or just ignore it

  	 */

  	if (atomic_read(&journal->j_wcount) != 0) {

  		reiserfs_panic(s, "journal-844", "journal list is flushing, "
  			       "wcount is not 0");

  	}
  	cn = jl->j_realblock;
  	while (cn) {
  		was_jwait = 0;
  		was_dirty = 0;
  		saved_bh = NULL;
  		/* blocknr of 0 is no longer in the hash, ignore it */
  		if (cn->blocknr == 0) {
  			goto free_cnode;
  		}

  		/*
  		 * This transaction failed commit.
  		 * Don't write out to the disk
  		 */

  		if (!(jl->j_state & LIST_DIRTY))
  			goto free_cnode;
  
  		pjl = find_newer_jl_for_cn(cn);

  		/*
  		 * the order is important here.  We check pjl to make sure we
  		 * don't clear BH_JDirty_wait if we aren't the one writing this
  		 * block to disk

  		 */
  		if (!pjl && cn->bh) {
  			saved_bh = cn->bh;

  			/*
  			 * we do this to make sure nobody releases the
  			 * buffer while we are working with it

  			 */
  			get_bh(saved_bh);
  
  			if (buffer_journal_dirty(saved_bh)) {
  				BUG_ON(!can_dirty(cn));
  				was_jwait = 1;
  				was_dirty = 1;
  			} else if (can_dirty(cn)) {

  				/*
  				 * everything with !pjl && jwait
  				 * should be writable
  				 */

  				BUG();
  			}
  		}

  		/*
  		 * if someone has this block in a newer transaction, just make
  		 * sure they are committed, and don't try writing it to disk

  		 */
  		if (pjl) {
  			if (atomic_read(&pjl->j_commit_left))
  				flush_commit_list(s, pjl, 1);
  			goto free_cnode;
  		}

  		/*
  		 * bh == NULL when the block got to disk on its own, OR,
  		 * the block got freed in a future transaction

  		 */
  		if (saved_bh == NULL) {
  			goto free_cnode;
  		}

  		/*
  		 * this should never happen.  kupdate_one_transaction has
  		 * this list locked while it works, so we should never see a
  		 * buffer here that is not marked JDirty_wait

  		 */
  		if ((!was_jwait) && !buffer_locked(saved_bh)) {

  			reiserfs_warning(s, "journal-813",
  					 "BAD! buffer %llu %cdirty %cjwait, "

  					 "not in a newer transaction",

  					 (unsigned long long)saved_bh->
  					 b_blocknr, was_dirty ? ' ' : '!',
  					 was_jwait ? ' ' : '!');
  		}
  		if (was_dirty) {

  			/*
  			 * we inc again because saved_bh gets decremented
  			 * at free_cnode
  			 */

  			get_bh(saved_bh);
  			set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
  			lock_buffer(saved_bh);
  			BUG_ON(cn->blocknr != saved_bh->b_blocknr);
  			if (buffer_dirty(saved_bh))
  				submit_logged_buffer(saved_bh);
  			else
  				unlock_buffer(saved_bh);
  			count++;
  		} else {

  			reiserfs_warning(s, "clm-2082",
  					 "Unable to flush buffer %llu in %s",

  					 (unsigned long long)saved_bh->

  					 b_blocknr, __func__);

  		}

  free_cnode:

  		last = cn;
  		cn = cn->next;
  		if (saved_bh) {

  			/*
  			 * we incremented this to keep others from
  			 * taking the buffer head away
  			 */

  			put_bh(saved_bh);

  			if (atomic_read(&saved_bh->b_count) < 0) {

  				reiserfs_warning(s, "journal-945",
  						 "saved_bh->b_count < 0");

  			}
  		}
  	}
  	if (count > 0) {
  		cn = jl->j_realblock;
  		while (cn) {
  			if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
  				if (!cn->bh) {

  					reiserfs_panic(s, "journal-1011",
  						       "cn->bh is NULL");

  				}

  				depth = reiserfs_write_unlock_nested(s);
  				__wait_on_buffer(cn->bh);
  				reiserfs_write_lock_nested(s, depth);

  				if (!cn->bh) {

  					reiserfs_panic(s, "journal-1012",
  						       "cn->bh is NULL");

  				}
  				if (unlikely(!buffer_uptodate(cn->bh))) {
  #ifdef CONFIG_REISERFS_CHECK

  					reiserfs_warning(s, "journal-949",
  							 "buffer write failed");

  #endif
  					err = -EIO;
  				}

  				/*
  				 * note, we must clear the JDirty_wait bit
  				 * after the up to date check, otherwise we
  				 * race against our flushpage routine

  				 */
  				BUG_ON(!test_clear_buffer_journal_dirty
  				       (cn->bh));

  				/* drop one ref for us */

  				put_bh(cn->bh);

  				/* drop one ref for journal_mark_dirty */
  				release_buffer_page(cn->bh);

  			}
  			cn = cn->next;
  		}
  	}
  
  	if (err)
  		reiserfs_abort(s, -EIO,
  			       "Write error while pushing transaction to disk in %s",

  			       __func__);

  flush_older_and_return:

  	/*
  	 * before we can update the journal header block, we _must_ flush all
  	 * real blocks from all older transactions to disk.  This is because
  	 * once the header block is updated, this transaction will not be
  	 * replayed after a crash

  	 */
  	if (flushall) {
  		flush_older_journal_lists(s, jl);
  	}
  
  	err = journal->j_errno;

  	/*
  	 * before we can remove everything from the hash tables for this
  	 * transaction, we must make sure it can never be replayed
  	 *
  	 * since we are only called from do_journal_end, we know for sure there
  	 * are no allocations going on while we are flushing journal lists.  So,
  	 * we only need to update the journal header block for the last list
  	 * being flushed

  	 */
  	if (!err && flushall) {
  		err =
  		    update_journal_header_block(s,
  						(jl->j_start + jl->j_len +
  						 2) % SB_ONDISK_JOURNAL_SIZE(s),
  						jl->j_trans_id);
  		if (err)
  			reiserfs_abort(s, -EIO,
  				       "Write error while updating journal header in %s",

  				       __func__);

  	}
  	remove_all_from_journal_list(s, jl, 0);
  	list_del_init(&jl->j_list);
  	journal->j_num_lists--;
  	del_from_work_list(s, jl);
  
  	if (journal->j_last_flush_id != 0 &&
  	    (jl->j_trans_id - journal->j_last_flush_id) != 1) {

  		reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",

  				 journal->j_last_flush_id, jl->j_trans_id);
  	}
  	journal->j_last_flush_id = jl->j_trans_id;

  	/*
  	 * not strictly required since we are freeing the list, but it should

  	 * help find code using dead lists later on
  	 */
  	jl->j_len = 0;

  	atomic_set(&jl->j_nonzerolen, 0);

  	jl->j_start = 0;
  	jl->j_realblock = NULL;
  	jl->j_commit_bh = NULL;
  	jl->j_trans_id = 0;
  	jl->j_state = 0;
  	put_journal_list(s, jl);
  	if (flushall)

  		mutex_unlock(&journal->j_flush_mutex);

  	return err;
  }
  
  static int write_one_transaction(struct super_block *s,
  				 struct reiserfs_journal_list *jl,
  				 struct buffer_chunk *chunk)
  {
  	struct reiserfs_journal_cnode *cn;
  	int ret = 0;
  
  	jl->j_state |= LIST_TOUCHED;
  	del_from_work_list(s, jl);
  	if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
  		return 0;
  	}
  
  	cn = jl->j_realblock;
  	while (cn) {

  		/*
  		 * if the blocknr == 0, this has been cleared from the hash,
  		 * skip it

  		 */
  		if (cn->blocknr == 0) {
  			goto next;
  		}
  		if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
  			struct buffer_head *tmp_bh;

  			/*
  			 * we can race against journal_mark_freed when we try

  			 * to lock_buffer(cn->bh), so we have to inc the buffer
  			 * count, and recheck things after locking
  			 */
  			tmp_bh = cn->bh;
  			get_bh(tmp_bh);
  			lock_buffer(tmp_bh);
  			if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
  				if (!buffer_journal_dirty(tmp_bh) ||
  				    buffer_journal_prepared(tmp_bh))
  					BUG();
  				add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
  				ret++;
  			} else {
  				/* note, cn->bh might be null now */
  				unlock_buffer(tmp_bh);
  			}
  			put_bh(tmp_bh);
  		}

  next:

  		cn = cn->next;
  		cond_resched();
  	}
  	return ret;
  }
  
  /* used by flush_commit_list */
  static int dirty_one_transaction(struct super_block *s,
  				 struct reiserfs_journal_list *jl)
  {
  	struct reiserfs_journal_cnode *cn;
  	struct reiserfs_journal_list *pjl;
  	int ret = 0;
  
  	jl->j_state |= LIST_DIRTY;
  	cn = jl->j_realblock;
  	while (cn) {

  		/*
  		 * look for a more recent transaction that logged this
  		 * buffer.  Only the most recent transaction with a buffer in
  		 * it is allowed to send that buffer to disk

  		 */
  		pjl = find_newer_jl_for_cn(cn);
  		if (!pjl && cn->blocknr && cn->bh
  		    && buffer_journal_dirty(cn->bh)) {
  			BUG_ON(!can_dirty(cn));

  			/*
  			 * if the buffer is prepared, it will either be logged

  			 * or restored.  If restored, we need to make sure
  			 * it actually gets marked dirty
  			 */
  			clear_buffer_journal_new(cn->bh);
  			if (buffer_journal_prepared(cn->bh)) {
  				set_buffer_journal_restore_dirty(cn->bh);
  			} else {
  				set_buffer_journal_test(cn->bh);
  				mark_buffer_dirty(cn->bh);
  			}
  		}
  		cn = cn->next;
  	}
  	return ret;
  }
  
  static int kupdate_transactions(struct super_block *s,
  				struct reiserfs_journal_list *jl,
  				struct reiserfs_journal_list **next_jl,

  				unsigned int *next_trans_id,

  				int num_blocks, int num_trans)
  {
  	int ret = 0;
  	int written = 0;
  	int transactions_flushed = 0;

  	unsigned int orig_trans_id = jl->j_trans_id;

  	struct buffer_chunk chunk;
  	struct list_head *entry;
  	struct reiserfs_journal *journal = SB_JOURNAL(s);
  	chunk.nr = 0;

  	reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);

  	if (!journal_list_still_alive(s, orig_trans_id)) {
  		goto done;
  	}

  	/*
  	 * we've got j_flush_mutex held, nobody is going to delete any

  	 * of these lists out from underneath us
  	 */
  	while ((num_trans && transactions_flushed < num_trans) ||
  	       (!num_trans && written < num_blocks)) {
  
  		if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
  		    atomic_read(&jl->j_commit_left)
  		    || !(jl->j_state & LIST_DIRTY)) {
  			del_from_work_list(s, jl);
  			break;
  		}
  		ret = write_one_transaction(s, jl, &chunk);
  
  		if (ret < 0)
  			goto done;
  		transactions_flushed++;
  		written += ret;
  		entry = jl->j_list.next;
  
  		/* did we wrap? */
  		if (entry == &journal->j_journal_list) {
  			break;
  		}
  		jl = JOURNAL_LIST_ENTRY(entry);
  
  		/* don't bother with older transactions */
  		if (jl->j_trans_id <= orig_trans_id)
  			break;
  	}
  	if (chunk.nr) {
  		write_chunk(&chunk);
  	}

  done:

  	mutex_unlock(&journal->j_flush_mutex);

  	return ret;
  }

  /*
   * for o_sync and fsync heavy applications, they tend to use
   * all the journa list slots with tiny transactions.  These
   * trigger lots and lots of calls to update the header block, which
   * adds seeks and slows things down.
   *
   * This function tries to clear out a large chunk of the journal lists
   * at once, which makes everything faster since only the newest journal
   * list updates the header block
   */

  static int flush_used_journal_lists(struct super_block *s,

  				    struct reiserfs_journal_list *jl)
  {
  	unsigned long len = 0;
  	unsigned long cur_len;
  	int ret;
  	int i;
  	int limit = 256;
  	struct reiserfs_journal_list *tjl;
  	struct reiserfs_journal_list *flush_jl;

  	unsigned int trans_id;

  	struct reiserfs_journal *journal = SB_JOURNAL(s);
  
  	flush_jl = tjl = jl;
  
  	/* in data logging mode, try harder to flush a lot of blocks */
  	if (reiserfs_data_log(s))
  		limit = 1024;
  	/* flush for 256 transactions or limit blocks, whichever comes first */
  	for (i = 0; i < 256 && len < limit; i++) {
  		if (atomic_read(&tjl->j_commit_left) ||
  		    tjl->j_trans_id < jl->j_trans_id) {
  			break;
  		}
  		cur_len = atomic_read(&tjl->j_nonzerolen);
  		if (cur_len > 0) {
  			tjl->j_state &= ~LIST_TOUCHED;
  		}
  		len += cur_len;
  		flush_jl = tjl;
  		if (tjl->j_list.next == &journal->j_journal_list)
  			break;
  		tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
  	}

  	get_journal_list(jl);
  	get_journal_list(flush_jl);

  
  	/*
  	 * try to find a group of blocks we can flush across all the
  	 * transactions, but only bother if we've actually spanned
  	 * across multiple lists

  	 */
  	if (flush_jl != jl) {
  		ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
  	}
  	flush_journal_list(s, flush_jl, 1);

  	put_journal_list(s, flush_jl);
  	put_journal_list(s, jl);

  	return 0;

  }
  
  /*

   * removes any nodes in table with name block and dev as bh.
   * only touchs the hnext and hprev pointers.
   */

  void remove_journal_hash(struct super_block *sb,

  			 struct reiserfs_journal_cnode **table,
  			 struct reiserfs_journal_list *jl,
  			 unsigned long block, int remove_freed)
  {
  	struct reiserfs_journal_cnode *cur;
  	struct reiserfs_journal_cnode **head;
  
  	head = &(journal_hash(table, sb, block));
  	if (!head) {
  		return;
  	}
  	cur = *head;
  	while (cur) {
  		if (cur->blocknr == block && cur->sb == sb
  		    && (jl == NULL || jl == cur->jlist)
  		    && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
  			if (cur->hnext) {
  				cur->hnext->hprev = cur->hprev;
  			}
  			if (cur->hprev) {
  				cur->hprev->hnext = cur->hnext;
  			} else {
  				*head = cur->hnext;
  			}
  			cur->blocknr = 0;
  			cur->sb = NULL;
  			cur->state = 0;

  			/*
  			 * anybody who clears the cur->bh will also
  			 * dec the nonzerolen
  			 */
  			if (cur->bh && cur->jlist)

  				atomic_dec(&cur->jlist->j_nonzerolen);

  			cur->bh = NULL;
  			cur->jlist = NULL;
  		}
  		cur = cur->hnext;
  	}
  }

  static void free_journal_ram(struct super_block *sb)

  {

  	struct reiserfs_journal *journal = SB_JOURNAL(sb);

  	kfree(journal->j_current_jl);

  	journal->j_num_lists--;
  
  	vfree(journal->j_cnode_free_orig);

  	free_list_bitmaps(sb, journal->j_list_bitmap);
  	free_bitmap_nodes(sb);	/* must be after free_list_bitmaps */

  	if (journal->j_header_bh) {
  		brelse(journal->j_header_bh);
  	}

  	/*
  	 * j_header_bh is on the journal dev, make sure
  	 * not to release the journal dev until we brelse j_header_bh

  	 */

  	release_journal_dev(sb, journal);

  	vfree(journal);

  }
  
  /*

   * call on unmount.  Only set error to 1 if you haven't made your way out
   * of read_super() yet.  Any other caller must keep error at 0.
   */

  static int do_journal_release(struct reiserfs_transaction_handle *th,

  			      struct super_block *sb, int error)

  {
  	struct reiserfs_transaction_handle myth;
  	int flushed = 0;