Implemented new block allocator for reiserfs, new tail policy for smaller tails.

New block allocator contains code from Alexander Zarochencev, Jeff Mahoney and Oleg Drokin.

fs/reiserfs/fix_node.c

@@ -795,8 +795,8 @@ static int  get_empty_nodes(
   else /* If we have enough already then there is nothing to do. */
     return CARRY_ON;
 
-  if ( reiserfs_new_blocknrs (p_s_tb->transaction_handle, a_n_blocknrs,
-			      PATH_PLAST_BUFFER(p_s_tb->tb_path)->b_blocknr, n_amount_needed) == NO_DISK_SPACE )
+  if ( reiserfs_new_form_blocknrs (p_s_tb, a_n_blocknrs,
+                                   n_amount_needed) == NO_DISK_SPACE )
     return NO_DISK_SPACE;
 
   /* for each blocknumber we just got, get a buffer and stick it on FEB */

fs/reiserfs/hashes.c

@@ -19,6 +19,7 @@
 //
 
 #include <asm/types.h>
+#include <asm/page.h>
 
 
 
@@ -57,7 +58,6 @@ u32 keyed_hash(const signed char *msg, int len)
 	u32 pad;
 	int i;
  
-
 	//	assert(len >= 0 && len < 256);
 
 	pad = (u32)len | ((u32)len << 8);
@@ -92,7 +92,7 @@ u32 keyed_hash(const signed char *msg, int len)
 	{
 	    	//assert(len < 16);
 		if (len >= 16)
-		    *(int *)0 = 0;
+		    BUG();
 
 		a = (u32)msg[ 0]      |
 		    (u32)msg[ 1] << 8 |
@@ -118,7 +118,7 @@ u32 keyed_hash(const signed char *msg, int len)
 	{
 	    	//assert(len < 12);
 		if (len >= 12)
-		    *(int *)0 = 0;
+		    BUG();
 		a = (u32)msg[ 0]      |
 		    (u32)msg[ 1] << 8 |
 		    (u32)msg[ 2] << 16|
@@ -139,7 +139,7 @@ u32 keyed_hash(const signed char *msg, int len)
 	{
 	    	//assert(len < 8);
 		if (len >= 8)
-		    *(int *)0 = 0;
+		    BUG();
 		a = (u32)msg[ 0]      |
 		    (u32)msg[ 1] << 8 |
 		    (u32)msg[ 2] << 16|
@@ -156,7 +156,7 @@ u32 keyed_hash(const signed char *msg, int len)
 	{
 	    	//assert(len < 4);
 		if (len >= 4)
-		    *(int *)0 = 0;
+		    BUG();
 		a = b = c = d = pad;
 		for(i = 0; i < len; i++)
 		{

fs/reiserfs/inode.c

@@ -156,33 +156,6 @@ static inline void fix_tail_page_for_writing(struct page *page) {
     }
 }
 
-
-
-
-/* we need to allocate a block for new unformatted node.  Try to figure out
-   what point in bitmap reiserfs_new_blocknrs should start from. */
-static b_blocknr_t find_tag (struct buffer_head * bh, struct item_head * ih,
-			     __u32 * item, int pos_in_item)
-{
-  __u32 block ;
-  if (!is_indirect_le_ih (ih))
-	 /* something more complicated could be here */
-	 return bh->b_blocknr;
-
-  /* for indirect item: go to left and look for the first non-hole entry in
-	  the indirect item */
-  if (pos_in_item == I_UNFM_NUM (ih))
-	 pos_in_item --;
-  while (pos_in_item >= 0) {
-	 block = get_block_num(item, pos_in_item) ;
-	 if (block)
-		return block ;
-	 pos_in_item --;
-  }
-  return bh->b_blocknr;
-}
-
-
 /* reiserfs_get_block does not need to allocate a block only if it has been
    done already or non-hole position has been found in the indirect item */
 static inline int allocation_needed (int retval, b_blocknr_t allocated, 
@@ -341,10 +314,10 @@ static int _get_block_create_0 (struct inode * inode, long block,
     ** kmap schedules
     */
     if (!p) {
-    p = (char *)kmap(bh_result->b_page) ;
-    if (fs_changed (fs_gen, inode->i_sb) && item_moved (&tmp_ih, &path)) {
-        goto research;
-    }
+	p = (char *)kmap(bh_result->b_page) ;
+	if (fs_changed (fs_gen, inode->i_sb) && item_moved (&tmp_ih, &path)) {
+	    goto research;
+	}
     }
     p += offset ;
     memset (p, 0, inode->i_sb->s_blocksize);
@@ -505,24 +478,24 @@ static int convert_tail_for_hole(struct inode *inode,
 }
 
 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
+			   long block,
                            struct inode *inode, 
 			   b_blocknr_t *allocated_block_nr, 
-			   unsigned long tag,
+			   struct path * path,
 			   int flags) {
   
 #ifdef REISERFS_PREALLOCATE
     if (!(flags & GET_BLOCK_NO_ISEM)) {
-        return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr, tag);
+	return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr, path, block);
     }
 #endif
-    return reiserfs_new_unf_blocknrs (th, allocated_block_nr, tag);
+    return reiserfs_new_unf_blocknrs (th, allocated_block_nr, path, block);
 }
 
 int reiserfs_get_block (struct inode * inode, sector_t block,
 			struct buffer_head * bh_result, int create)
 {
     int repeat, retval;
-    unsigned long tag;
     b_blocknr_t allocated_block_nr = 0;// b_blocknr_t is unsigned long
     INITIALIZE_PATH(path);
     int pos_in_item;
@@ -601,7 +574,6 @@ int reiserfs_get_block (struct inode * inode, sector_t block,
 
     if (allocation_needed (retval, allocated_block_nr, ih, item, pos_in_item)) {
 	/* we have to allocate block for the unformatted node */
-	tag = find_tag (bh, ih, item, pos_in_item);
 	if (!transaction_started) {
 	    pathrelse(&path) ;
 	    journal_begin(&th, inode->i_sb, jbegin_count) ;
@@ -610,7 +582,7 @@ int reiserfs_get_block (struct inode * inode, sector_t block,
 	    goto research ;
 	}
 
-	repeat = _allocate_block(&th, inode, &allocated_block_nr, tag, create);
+	repeat = _allocate_block(&th, block, inode, &allocated_block_nr, &path, create);
 
 	if (repeat == NO_DISK_SPACE) {
 	    /* restart the transaction to give the journal a chance to free
@@ -618,7 +590,7 @@ int reiserfs_get_block (struct inode * inode, sector_t block,
 	    ** research if we succeed on the second try
 	    */
 	    restart_transaction(&th, inode, &path) ; 
-	    repeat = _allocate_block(&th, inode,&allocated_block_nr,tag,create);
+	    repeat = _allocate_block(&th, block, inode, &allocated_block_nr, NULL, create);
 
 	    if (repeat != NO_DISK_SPACE) {
 		goto research ;
@@ -769,8 +741,8 @@ int reiserfs_get_block (struct inode * inode, sector_t block,
 	    add_to_flushlist(inode, unbh) ;
 
 	    /* mark it dirty now to prevent commit_write from adding
-	    ** this buffer to the inode's dirty buffer list
-	    */
+	     ** this buffer to the inode's dirty buffer list
+	     */
 		/*
 		 * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
 		 * It's still atomic, but it sets the page dirty too,
@@ -778,7 +750,7 @@ int reiserfs_get_block (struct inode * inode, sector_t block,
 		 * VM (which was also the case with __mark_buffer_dirty())
 		 */
 	    mark_buffer_dirty(unbh) ;
-		  
+
 	    //inode->i_blocks += inode->i_sb->s_blocksize / 512;
 	    //mark_tail_converted (inode);
 	} else {
@@ -1591,6 +1563,10 @@ int reiserfs_new_inode (struct reiserfs_transaction_handle *th,
 	set_inode_sd_version (inode, STAT_DATA_V2);
     
     /* insert the stat data into the tree */
+#ifdef DISPLACE_NEW_PACKING_LOCALITIES
+    if (REISERFS_I(dir)->new_packing_locality)
+	th->displace_new_blocks = 1;
+#endif
     retval = reiserfs_insert_item (th, &path_to_key, &key, &ih, (char *)(&sd));
     if (retval) {
 	err = retval;
@@ -1598,6 +1574,10 @@ int reiserfs_new_inode (struct reiserfs_transaction_handle *th,
 	goto out_bad_inode;
     }
 
+#ifdef DISPLACE_NEW_PACKING_LOCALITIES
+    if (!th->displace_new_blocks)
+	REISERFS_I(dir)->new_packing_locality = 0;
+#endif
     if (S_ISDIR(mode)) {
 	/* insert item with "." and ".." */
 	retval = reiserfs_new_directory (th, &ih, &path_to_key, dir);
@@ -1773,16 +1753,16 @@ void reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps) {
     reiserfs_update_inode_transaction(p_s_inode) ;
     windex = push_journal_writer("reiserfs_vfs_truncate_file") ;
     if (update_timestamps)
-           /* we are doing real truncate: if the system crashes before the last
-              transaction of truncating gets committed - on reboot the file
-              either appears truncated properly or not truncated at all */
-           add_save_link (&th, p_s_inode, 1);
+	    /* we are doing real truncate: if the system crashes before the last
+	       transaction of truncating gets committed - on reboot the file
+	       either appears truncated properly or not truncated at all */
+	add_save_link (&th, p_s_inode, 1);
     reiserfs_do_truncate (&th, p_s_inode, page, update_timestamps) ;
     pop_journal_writer(windex) ;
     journal_end(&th, p_s_inode->i_sb,  JOURNAL_PER_BALANCE_CNT * 2 + 1 ) ;
 
     if (update_timestamps)
-       remove_save_link (p_s_inode, 1/* truncate */);
+	remove_save_link (p_s_inode, 1/* truncate */);
 
     if (page) {
         length = offset & (blocksize - 1) ;

fs/reiserfs/journal.c

@@ -510,14 +510,12 @@ int dump_journal_writers(void) {
 **
 */
 int reiserfs_in_journal(struct super_block *p_s_sb,
-                        unsigned long bl, int search_all, 
+                        int bmap_nr, int bit_nr, int search_all, 
 			unsigned long *next_zero_bit) {
   struct reiserfs_journal_cnode *cn ;
   struct reiserfs_list_bitmap *jb ;
   int i ;
-  int bmap_nr = bl / (p_s_sb->s_blocksize << 3) ;
-  int bit_nr = bl % (p_s_sb->s_blocksize << 3) ;
-  int tmp_bit ;
+  unsigned long bl;
 
   *next_zero_bit = 0 ; /* always start this at zero. */
 
@@ -537,15 +535,15 @@ int reiserfs_in_journal(struct super_block *p_s_sb,
       jb = SB_JOURNAL(p_s_sb)->j_list_bitmap + i ;
       if (jb->journal_list && jb->bitmaps[bmap_nr] &&
           test_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data)) {
-	tmp_bit = find_next_zero_bit((unsigned long *)
+	*next_zero_bit = find_next_zero_bit((unsigned long *)
 	                             (jb->bitmaps[bmap_nr]->data),
 	                             p_s_sb->s_blocksize << 3, bit_nr+1) ; 
-	*next_zero_bit = bmap_nr * (p_s_sb->s_blocksize << 3) + tmp_bit ;
 	return 1 ;
       }
     }
   }
 
+  bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
   /* is it in any old transactions? */
   if (search_all && (cn = get_journal_hash_dev(p_s_sb, SB_JOURNAL(p_s_sb)->j_list_hash_table, bl))) {
     return 1; 
@@ -1820,7 +1818,8 @@ static void reiserfs_journal_commit_task_func(struct reiserfs_journal_commit_tas
   jl = SB_JOURNAL_LIST(ct->p_s_sb) + ct->jindex ;
 
   flush_commit_list(ct->p_s_sb, SB_JOURNAL_LIST(ct->p_s_sb) + ct->jindex, 1) ; 
-  if (jl->j_len > 0 && atomic_read(&(jl->j_nonzerolen)) > 0 && 
+
+  if (jl->j_len > 0 && atomic_read(&(jl->j_nonzerolen)) > 0 &&
       atomic_read(&(jl->j_commit_left)) == 0) {
     kupdate_one_transaction(ct->p_s_sb, jl) ;
   }

fs/reiserfs/namei.c

@@ -661,6 +661,10 @@ static int reiserfs_mkdir (struct inode * dir, struct dentry *dentry, int mode)
     struct reiserfs_transaction_handle th ;
     int jbegin_count = JOURNAL_PER_BALANCE_CNT * 3; 
 
+#ifdef DISPLACE_NEW_PACKING_LOCALITIES
+    /* set flag that new packing locality created and new blocks for the content     * of that directory are not displaced yet */
+    REISERFS_I(dir)->new_packing_locality = 1;
+#endif
     mode = S_IFDIR | mode;
     if (!(inode = new_inode(dir->i_sb))) {
 	return -ENOMEM ;

fs/reiserfs/procfs.c

@@ -111,7 +111,7 @@ int reiserfs_global_version_in_proc( char *buffer, char **start, off_t offset,
 #define SF( x ) ( r -> x )
 #define SFP( x ) SF( s_proc_info_data.x )
 #define SFPL( x ) SFP( x[ level ] )
-#define SFPF( x ) SFP( find_forward.x )
+#define SFPF( x ) SFP( scan_bitmap.x )
 #define SFPJ( x ) SFP( journal.x )
 
 #define D2C( x ) le16_to_cpu( x )
@@ -184,7 +184,7 @@ int reiserfs_super_in_proc( char *buffer, char **start, off_t offset,
 			reiserfs_no_unhashed_relocation( sb ) ? "NO_UNHASHED_RELOCATION " : "",
 			reiserfs_hashed_relocation( sb ) ? "UNHASHED_RELOCATION " : "",
 			reiserfs_test4( sb ) ? "TEST4 " : "",
-			dont_have_tails( sb ) ? "NO_TAILS " : "TAILS ",
+			have_large_tails( sb ) ? "TAILS " : have_small_tails(sb)?"SMALL_TAILS ":"NO_TAILS ",
 			replay_only( sb ) ? "REPLAY_ONLY " : "",
 			reiserfs_dont_log( sb ) ? "DONT_LOG " : "LOG ",
 			convert_reiserfs( sb ) ? "CONV " : "",
@@ -314,27 +314,30 @@ int reiserfs_bitmap_in_proc( char *buffer, char **start, off_t offset,
 	r = REISERFS_SB(sb);
 
 	len += sprintf( &buffer[ len ], "free_block: %lu\n"
-			"find_forward:"
-			"         wait"
-			"         bmap"
-			"        retry"
-			" journal_hint"
-			"  journal_out"
+			"  scan_bitmap:"
+			"          wait"
+			"          bmap"
+			"         retry"
+			"        stolen"
+			"  journal_hint"
+			"journal_nohint"
 			"\n"
-			" %12lu"
-			" %12lu"
-			" %12lu"
-			" %12lu"
-			" %12lu"
-			" %12lu"
+			" %14lu"
+			" %14lu"
+			" %14lu"
+			" %14lu"
+			" %14lu"
+			" %14lu"
+			" %14lu"
 			"\n",
 			SFP( free_block ),
 			SFPF( call ), 
 			SFPF( wait ), 
 			SFPF( bmap ),
 			SFPF( retry ),
+			SFPF( stolen ),
 			SFPF( in_journal_hint ),
-			SFPF( in_journal_out ) );
+			SFPF( in_journal_nohint ) );
 
 	procinfo_epilogue( sb );
 	return reiserfs_proc_tail( len, buffer, start, offset, count, eof );

fs/reiserfs/resize.c

@@ -19,7 +19,8 @@
 int reiserfs_resize (struct super_block * s, unsigned long block_count_new)
 {
 	struct reiserfs_super_block * sb;
-	struct buffer_head ** bitmap, * bh;
+        struct reiserfs_bitmap_info *bitmap;
+	struct buffer_head * bh;
 	struct reiserfs_transaction_handle th;
 	unsigned int bmap_nr_new, bmap_nr;
 	unsigned int block_r_new, block_r;
@@ -103,26 +104,29 @@ int reiserfs_resize (struct super_block * s, unsigned long block_count_new)
 	
 	    /* allocate additional bitmap blocks, reallocate array of bitmap
 	     * block pointers */
-	    bitmap = reiserfs_kmalloc(sizeof(struct buffer_head *) * bmap_nr_new, GFP_KERNEL, s);
+	    bitmap = vmalloc(sizeof(struct reiserfs_bitmap_info) * bmap_nr_new);
 	    if (!bitmap) {
 		printk("reiserfs_resize: unable to allocate memory.\n");
 		return -ENOMEM;
 	    }
+	    memset (bitmap, 0, sizeof (struct reiserfs_bitmap_info) * SB_BMAP_NR(s));
 	    for (i = 0; i < bmap_nr; i++)
 		bitmap[i] = SB_AP_BITMAP(s)[i];
 	    for (i = bmap_nr; i < bmap_nr_new; i++) {
-		bitmap[i] = sb_getblk(s, i * s->s_blocksize * 8);
-		memset(bitmap[i]->b_data, 0, sb_blocksize(sb));
-		reiserfs_test_and_set_le_bit(0, bitmap[i]->b_data);
-
-		mark_buffer_dirty(bitmap[i]) ;
-		set_buffer_uptodate(bitmap[i]);
-		ll_rw_block(WRITE, 1, bitmap + i);
-		wait_on_buffer(bitmap[i]);
+		bitmap[i].bh = sb_getblk(s, i * s->s_blocksize * 8);
+		memset(bitmap[i].bh->b_data, 0, sb_blocksize(sb));
+		reiserfs_test_and_set_le_bit(0, bitmap[i].bh->b_data);
+
+		mark_buffer_dirty(bitmap[i].bh) ;
+		set_buffer_uptodate(bitmap[i].bh);
+		ll_rw_block(WRITE, 1, &bitmap[i].bh);
+		wait_on_buffer(bitmap[i].bh);
+		// update bitmap_info stuff
+		bitmap[i].first_zero_hint=1;
+		bitmap[i].free_count = sb_blocksize(sb) * 8 - 1;
 	    }	
 	    /* free old bitmap blocks array */
-	    reiserfs_kfree(SB_AP_BITMAP(s), 
-			   sizeof(struct buffer_head *) * bmap_nr, s);
+	    vfree(SB_AP_BITMAP(s));
 	    SB_AP_BITMAP(s) = bitmap;
 	}
 	
@@ -130,18 +134,26 @@ int reiserfs_resize (struct super_block * s, unsigned long block_count_new)
 	journal_begin(&th, s, 10);
 
 	/* correct last bitmap blocks in old and new disk layout */
-	reiserfs_prepare_for_journal(s, SB_AP_BITMAP(s)[bmap_nr - 1], 1);
+	reiserfs_prepare_for_journal(s, SB_AP_BITMAP(s)[bmap_nr - 1].bh, 1);
 	for (i = block_r; i < s->s_blocksize * 8; i++)
 	    reiserfs_test_and_clear_le_bit(i, 
-					   SB_AP_BITMAP(s)[bmap_nr - 1]->b_data);
-	journal_mark_dirty(&th, s, SB_AP_BITMAP(s)[bmap_nr - 1]);
+					   SB_AP_BITMAP(s)[bmap_nr - 1].bh->b_data);
+	SB_AP_BITMAP(s)[bmap_nr - 1].free_count += s->s_blocksize * 8 - block_r;
+	if ( !SB_AP_BITMAP(s)[bmap_nr - 1].first_zero_hint)
+	    SB_AP_BITMAP(s)[bmap_nr - 1].first_zero_hint = block_r;
+
+	journal_mark_dirty(&th, s, SB_AP_BITMAP(s)[bmap_nr - 1].bh);
 
-	reiserfs_prepare_for_journal(s, SB_AP_BITMAP(s)[bmap_nr_new - 1], 1);
+	reiserfs_prepare_for_journal(s, SB_AP_BITMAP(s)[bmap_nr_new - 1].bh, 1);
 	for (i = block_r_new; i < s->s_blocksize * 8; i++)
 	    reiserfs_test_and_set_le_bit(i,
-					 SB_AP_BITMAP(s)[bmap_nr_new - 1]->b_data);
-	journal_mark_dirty(&th, s, SB_AP_BITMAP(s)[bmap_nr_new - 1]);
+					 SB_AP_BITMAP(s)[bmap_nr_new - 1].bh->b_data);
+	journal_mark_dirty(&th, s, SB_AP_BITMAP(s)[bmap_nr_new - 1].bh);
  
+	SB_AP_BITMAP(s)[bmap_nr_new - 1].free_count -= s->s_blocksize * 8 - block_r_new;
+	/* Extreme case where last bitmap is the only valid block in itself. */
+	if ( !SB_AP_BITMAP(s)[bmap_nr_new - 1].free_count )
+	    SB_AP_BITMAP(s)[bmap_nr_new - 1].first_zero_hint = 0;
  	/* update super */
 	reiserfs_prepare_for_journal(s, SB_BUFFER_WITH_SB(s), 1) ;
 	free_blocks = SB_FREE_BLOCKS(s);

fs/reiserfs/stree.c

@@ -1620,9 +1620,9 @@ int reiserfs_cut_from_item (struct reiserfs_transaction_handle *th,
     
     do_balance(&s_cut_balance, NULL, NULL, c_mode);
     if ( n_is_inode_locked ) {
-	/* we've done an indirect->direct conversion.  when the data block 
-	** was freed, it was removed from the list of blocks that must 
-	** be flushed before the transaction commits, so we don't need to 
+	/* we've done an indirect->direct conversion.  when the data block
+	** was freed, it was removed from the list of blocks that must
+	** be flushed before the transaction commits, so we don't need to
 	** deal with it here.
 	*/
 	REISERFS_I(p_s_inode)->i_flags &= ~i_pack_on_close_mask ;
@@ -1813,6 +1813,9 @@ int reiserfs_paste_into_item (struct reiserfs_transaction_handle *th,
     int                 retval;
 
     init_tb_struct(th, &s_paste_balance, th->t_super, p_s_search_path, n_pasted_size);
+#ifdef DISPLACE_NEW_PACKING_LOCALITIES
+    s_paste_balance.key = p_s_key->on_disk_key;
+#endif
     
     while ( (retval = fix_nodes(M_PASTE, &s_paste_balance, NULL, p_c_body)) == REPEAT_SEARCH ) {
 	/* file system changed while we were in the fix_nodes */
@@ -1823,7 +1826,7 @@ int reiserfs_paste_into_item (struct reiserfs_transaction_handle *th,
 	    goto error_out ;
 	}
 	if (retval == POSITION_FOUND) {
-	    reiserfs_warning ("PAP-5710: reiserfs_paste_into_item: entry or pasted byte (%K) exists", p_s_key);
+	    reiserfs_warning ("PAP-5710: reiserfs_paste_into_item: entry or pasted byte (%K) exists\n", p_s_key);
 	    retval = -EEXIST ;
 	    goto error_out ;
 	}
@@ -1858,6 +1861,9 @@ int reiserfs_insert_item(struct reiserfs_transaction_handle *th,
     int                 retval;
 
     init_tb_struct(th, &s_ins_balance, th->t_super, p_s_path, IH_SIZE + ih_item_len(p_s_ih));
+#ifdef DISPLACE_NEW_PACKING_LOCALITIES
+    s_ins_balance.key = key->on_disk_key;
+#endif
 
     /*
     if (p_c_body == 0)

include/linux/reiserfs_fs.h

@@ -59,7 +59,8 @@
 #define USE_INODE_GENERATION_COUNTER
 
 #define REISERFS_PREALLOCATE
-#define PREALLOCATION_SIZE 8
+#define DISPLACE_NEW_PACKING_LOCALITIES
+#define PREALLOCATION_SIZE 9
 
 /* n must be power of 2 */
 #define _ROUND_UP(x,n) (((x)+(n)-1u) & ~((n)-1u))
@@ -326,7 +327,7 @@ static inline struct reiserfs_sb_info *REISERFS_SB(const struct super_block *sb)
    time cost for a 4 block file and saves an amount of space that is
    less significant as a percentage of space, or so goes the hypothesis.
    -Hans */
-#define STORE_TAIL_IN_UNFM(n_file_size,n_tail_size,n_block_size) \
+#define STORE_TAIL_IN_UNFM_S1(n_file_size,n_tail_size,n_block_size) \
 (\
   (!(n_tail_size)) || \
   (((n_tail_size) > MAX_DIRECT_ITEM_LEN(n_block_size)) || \
@@ -339,6 +340,18 @@ static inline struct reiserfs_sb_info *REISERFS_SB(const struct super_block *sb)
      ( (n_tail_size) >=   (MAX_DIRECT_ITEM_LEN(n_block_size) * 3)/4) ) ) \
 )
 
+/* Another strategy for tails, this one means only create a tail if all the
+   file would fit into one DIRECT item.
+   Primary intention for this one is to increase performance by decreasing
+   seeking.
+*/   
+#define STORE_TAIL_IN_UNFM_S2(n_file_size,n_tail_size,n_block_size) \
+(\
+  (!(n_tail_size)) || \
+  (((n_file_size) > MAX_DIRECT_ITEM_LEN(n_block_size)) ) \
+)
+
+
 
 /*
  * values for s_umount_state field
@@ -1433,6 +1446,10 @@ struct tree_balance
 
   int fs_gen;                  /* saved value of `reiserfs_generation' counter
 			          see FILESYSTEM_CHANGED() macro in reiserfs_fs.h */
+#ifdef DISPLACE_NEW_PACKING_LOCALITIES
+  struct key  key;	      /* key pointer, to pass to block allocator or
+				 another low-level subsystem */
+#endif
 } ;
 
 /* These are modes of balancing */
@@ -1673,7 +1690,7 @@ int journal_mark_freed(struct reiserfs_transaction_handle *, struct super_block
 int push_journal_writer(char *w) ;
 int pop_journal_writer(int windex) ;
 int journal_transaction_should_end(struct reiserfs_transaction_handle *, int) ;
-int reiserfs_in_journal(struct super_block *p_s_sb, unsigned long bl, int searchall, unsigned long *next) ;
+int reiserfs_in_journal(struct super_block *p_s_sb, int bmap_nr, int bit_nr, int searchall, unsigned long *next) ;
 int journal_begin(struct reiserfs_transaction_handle *, struct super_block *p_s_sb, unsigned long) ;
 void flush_async_commits(struct super_block *p_s_sb) ;
 
@@ -1818,8 +1835,8 @@ void reiserfs_do_truncate (struct reiserfs_transaction_handle *th,
 #define file_size(inode) ((inode)->i_size)
 #define tail_size(inode) (file_size (inode) & (i_block_size (inode) - 1))
 
-#define tail_has_to_be_packed(inode) (!dont_have_tails ((inode)->i_sb) &&\
-!STORE_TAIL_IN_UNFM(file_size (inode), tail_size(inode), i_block_size (inode)))
+#define tail_has_to_be_packed(inode) (have_large_tails ((inode)->i_sb)?\
+!STORE_TAIL_IN_UNFM_S1(file_size (inode), tail_size(inode), inode->i_sb->s_blocksize):have_small_tails ((inode)->i_sb)?!STORE_TAIL_IN_UNFM_S2(file_size (inode), tail_size(inode), inode->i_sb->s_blocksize):0 )
 
 void padd_item (char * item, int total_length, int length);
 
@@ -2015,22 +2032,87 @@ void make_empty_node (struct buffer_info *);
 struct buffer_head * get_FEB (struct tree_balance *);
 
 /* bitmap.c */
+
+/* structure contains hints for block allocator, and it is a container for
+ * arguments, such as node, search path, transaction_handle, etc. */
+ struct __reiserfs_blocknr_hint {
+     struct inode * inode;		/* inode passed to allocator, if we allocate unf. nodes */
+     long block;			/* file offset, in blocks */
+     struct key key;
+     struct path * path;		/* search path, used by allocator to deternine search_start by
+					 * various ways */
+     struct reiserfs_transaction_handle * th; /* transaction handle is needed to log super blocks and
+					       * bitmap blocks changes  */
+     b_blocknr_t beg, end;
+     b_blocknr_t search_start;		/* a field used to transfer search start value (block number)
+					 * between different block allocator procedures
+					 * (determine_search_start() and others) */
+    int prealloc_size;			/* is set in determine_prealloc_size() function, used by underlayed
+					 * function that do actual allocation */
+
+    int formatted_node:1;		/* the allocator uses different polices for getting disk space for
+					 * formatted/unformatted blocks with/without preallocation */
+    int preallocate:1;
+};
+
+typedef struct __reiserfs_blocknr_hint reiserfs_blocknr_hint_t;
+
+int reiserfs_parse_alloc_options (struct super_block *, char *);
 int is_reusable (struct super_block * s, unsigned long block, int bit_value);
 void reiserfs_free_block (struct reiserfs_transaction_handle *th, unsigned long);
-int reiserfs_new_blocknrs (struct reiserfs_transaction_handle *th,
-			   unsigned long * pblocknrs, unsigned long start_from, int amount_needed);
-int reiserfs_new_unf_blocknrs (struct reiserfs_transaction_handle *th,
-			       unsigned long * pblocknr, unsigned long start_from);
+int reiserfs_allocate_blocknrs(reiserfs_blocknr_hint_t *, b_blocknr_t * , int, int);
+extern inline int reiserfs_new_form_blocknrs (struct tree_balance * tb,
+					      b_blocknr_t *new_blocknrs, int amount_needed)
+{
+    reiserfs_blocknr_hint_t hint = {
+	th:tb->transaction_handle,
+	path: tb->tb_path,
+	inode: NULL,
+	key: tb->key,
+	block: 0,
+	formatted_node:1
+    };
+    return reiserfs_allocate_blocknrs(&hint, new_blocknrs, amount_needed, 0);
+}
+
+extern inline int reiserfs_new_unf_blocknrs (struct reiserfs_transaction_handle *th,
+					     b_blocknr_t *new_blocknrs,
+					     struct path * path, long block)
+{
+    reiserfs_blocknr_hint_t hint = {
+	th: th,
+	path: path,
+	inode: NULL,
+	block: block,
+	formatted_node: 0,
+	preallocate: 0
+    };
+    return reiserfs_allocate_blocknrs(&hint, new_blocknrs, 1, 0);
+}
+
 #ifdef REISERFS_PREALLOCATE
-int reiserfs_new_unf_blocknrs2 (struct reiserfs_transaction_handle *th, 
-				struct inode * inode,
-				unsigned long * pblocknr, 
-				unsigned long start_from);
+extern inline int reiserfs_new_unf_blocknrs2(struct reiserfs_transaction_handle *th,
+					     struct inode * inode,
+					     b_blocknr_t *new_blocknrs,
+					     struct path * path, long block)
+{
+    reiserfs_blocknr_hint_t hint = {
+	th: th,
+	path: path,
+	inode: inode,
+	block: block,
+	formatted_node: 0,
+	preallocate: 1
+    };
+    return reiserfs_allocate_blocknrs(&hint, new_blocknrs, 1, 0);
+}
 
 void reiserfs_discard_prealloc (struct reiserfs_transaction_handle *th, 
 				struct inode * inode);
 void reiserfs_discard_all_prealloc (struct reiserfs_transaction_handle *th);
 #endif
+void reiserfs_claim_blocks_to_be_allocated( struct super_block *sb, int blocks);
+void reiserfs_release_claimed_blocks( struct super_block *sb, int blocks);
 
 /* hashes.c */
 __u32 keyed_hash (const signed char *msg, int len);

include/linux/reiserfs_fs_i.h

@@ -37,6 +37,10 @@ struct reiserfs_inode_info {
     struct list_head i_prealloc_list; /* per-transaction list of inodes which
                                        * have preallocated blocks */
 
+    int new_packing_locality:1;	      /* new_packig_locality is created; new blocks
+				       * for the contents of this directory should be
+				       * displaced */
+
     /* we use these for fsync or O_SYNC to decide which transaction
     ** needs to be committed in order for this inode to be properly
     ** flushed */

include/linux/reiserfs_fs_sb.h

@@ -160,7 +160,10 @@ struct reiserfs_transaction_handle {
   int t_blocks_allocated ;      /* number of blocks this writer allocated */
   unsigned long t_trans_id ;    /* sanity check, equals the current trans id */
   struct super_block *t_super ; /* super for this FS when journal_begin was 
-                                   called. */
+                                   called. saves calls to reiserfs_get_super */
+  int displace_new_blocks:1;	/* if new block allocation occurres, that block
+				   should be displaced from others */
+
 } ;
 
 /*
@@ -254,6 +257,14 @@ struct reiserfs_journal {
 
 typedef __u32 (*hashf_t) (const signed char *, int);
 
+struct reiserfs_bitmap_info
+{
+    // FIXME: Won't work with block sizes > 8K
+    __u16  first_zero_hint;
+    __u16  free_count;
+    struct buffer_head *bh; /* the actual bitmap */
+};
+
 struct proc_dir_entry;
 
 #if defined( CONFIG_PROC_FS ) && defined( CONFIG_REISERFS_PROC_INFO )
@@ -298,14 +309,15 @@ typedef struct reiserfs_proc_info_data
   stat_cnt_t need_r_neighbor[ 5 ];
 
   stat_cnt_t free_block;
-  struct __find_forward_stats {
+  struct __scan_bitmap_stats {
 	stat_cnt_t call;
 	stat_cnt_t wait;
 	stat_cnt_t bmap;
 	stat_cnt_t retry;
 	stat_cnt_t in_journal_hint;
-	stat_cnt_t in_journal_out;
-  } find_forward;
+	stat_cnt_t in_journal_nohint;
+	stat_cnt_t stolen;
+  } scan_bitmap;
   struct __journal_stats {
 	stat_cnt_t in_journal;
 	stat_cnt_t in_journal_bitmap;
@@ -335,7 +347,7 @@ struct reiserfs_sb_info
 				/* both the comment and the choice of
                                    name are unclear for s_rs -Hans */
     struct reiserfs_super_block * s_rs;           /* Pointer to the super block in the buffer */
-    struct buffer_head ** s_ap_bitmap;       /* array of buffers, holding block bitmap */
+    struct reiserfs_bitmap_info * s_ap_bitmap;
     struct reiserfs_journal *s_journal ;		/* pointer to journal information */
     unsigned short s_mount_state;                 /* reiserfs state (valid, invalid) */
   
@@ -348,6 +360,16 @@ struct reiserfs_sb_info
                                    here (currently - NOTAIL, NOLOG,
                                    REPLAYONLY) */
 
+    struct {			/* This is a structure that describes block allocator options */
+	unsigned long bits;	/* Bitfield for enable/disable kind of options */
+	unsigned long large_file_size; /* size started from which we consider file to be a large one(in blocks) */
+	int border;		/* percentage of disk, border takes */
+	int preallocmin;	/* Minimal file size (in blocks) starting from which we do preallocations */
+	int preallocsize;	/* Number of blocks we try to prealloc when file
+				   reaches preallocmin size (in blocks) or
+				   prealloc_list is empty. */
+    } s_alloc_options;
+
 				/* Comment? -Hans */
     wait_queue_head_t s_wait;
 				/* To be obsoleted soon by per buffer seals.. -Hans */
@@ -374,6 +396,7 @@ struct reiserfs_sb_info
     int s_is_unlinked_ok;
     reiserfs_proc_info_data_t s_proc_info_data;
     struct proc_dir_entry *procdir;
+    int reserved_blocks; /* amount of blocks reserved for further allocations */
 };
 
 /* Definitions of reiserfs on-disk properties: */
@@ -381,7 +404,8 @@ struct reiserfs_sb_info
 #define REISERFS_3_6 1
 
 /* Mount options */
-#define NOTAIL 0  /* -o notail: no tails will be created in a session */
+#define REISERFS_LARGETAIL 0  /* large tails will be created in a session */
+#define REISERFS_SMALLTAIL 17  /* small (for files less than block size) tails will be created in a session */
 #define REPLAYONLY 3 /* replay journal and return 0. Use by fsck */
 #define REISERFS_NOLOG 4      /* -o nolog: turn journalling off */
 #define REISERFS_CONVERT 5    /* -o conv: causes conversion of old
@@ -429,7 +453,8 @@ struct reiserfs_sb_info
 #define reiserfs_hashed_relocation(s) (REISERFS_SB(s)->s_mount_opt & (1 << REISERFS_HASHED_RELOCATION))
 #define reiserfs_test4(s) (REISERFS_SB(s)->s_mount_opt & (1 << REISERFS_TEST4))
 
-#define dont_have_tails(s) (REISERFS_SB(s)->s_mount_opt & (1 << NOTAIL))
+#define have_large_tails(s) (REISERFS_SB(s)->s_mount_opt & (1 << REISERFS_LARGETAIL))
+#define have_small_tails(s) (REISERFS_SB(s)->s_mount_opt & (1 << REISERFS_SMALLTAIL))
 #define replay_only(s) (REISERFS_SB(s)->s_mount_opt & (1 << REPLAYONLY))
 #define reiserfs_dont_log(s) (REISERFS_SB(s)->s_mount_opt & (1 << REISERFS_NOLOG))
 #define old_format_only(s) (REISERFS_SB(s)->s_properties & (1 << REISERFS_3_5))