lustre/kernel_patches/patches/extN-noread.patch

   1  fs/ext3/ialloc.c        |   47 +++++++++++++++++++++-
   2  fs/ext3/inode.c         |   99 ++++++++++++++++++++++++++++++++++++------------
   3  include/linux/ext3_fs.h |    2
   4  3 files changed, 122 insertions(+), 26 deletions(-)
   5
   6 --- linux-2.4.18-chaos52/fs/ext3/ialloc.c~extN-noread   2003-05-16 12:26:29.000000000 +0800
   7 +++ linux-2.4.18-chaos52-root/fs/ext3/ialloc.c  2003-05-16 12:26:31.000000000 +0800
   8 @@ -289,6 +289,37 @@ error_return:
   9  }
  10
  11  /*
  12 + * @block_group: block group of inode
  13 + * @offset: relative offset of inode within @block_group
  14 + *
  15 + * Check whether any of the inodes in this disk block are in use.
  16 + *
  17 + * Caller must be holding superblock lock (group/bitmap read lock in future).
  18 + */
  19 +int ext3_itable_block_used(struct super_block *sb, unsigned int block_group,
  20 +                          int offset)
  21 +{
  22 +       int bitmap_nr = load_inode_bitmap(sb, block_group);
  23 +       int inodes_per_block;
  24 +       unsigned long inum, iend;
  25 +       struct buffer_head *ibitmap;
  26 +
  27 +       if (bitmap_nr < 0)
  28 +               return 1;
  29 +
  30 +       inodes_per_block = sb->s_blocksize / EXT3_SB(sb)->s_inode_size;
  31 +       inum = offset & ~(inodes_per_block - 1);
  32 +       iend = inum + inodes_per_block;
  33 +       ibitmap = EXT3_SB(sb)->s_inode_bitmap[bitmap_nr];
  34 +       for (; inum < iend; inum++) {
  35 +               if (inum != offset && ext3_test_bit(inum, ibitmap->b_data))
  36 +                       return 1;
  37 +       }
  38 +
  39 +       return 0;
  40 +}
  41 +
  42 +/*
  43   * There are two policies for allocating an inode.  If the new inode is
  44   * a directory, then a forward search is made for a block group with both
  45   * free space and a low directory-to-inode ratio; if that fails, then of
  46 @@ -312,6 +343,7 @@ struct inode * ext3_new_inode (handle_t
  47         struct ext3_group_desc * gdp;
  48         struct ext3_group_desc * tmp;
  49         struct ext3_super_block * es;
  50 +       struct ext3_iloc iloc;
  51         int err = 0;
  52
  53         /* Cannot create files in a deleted directory */
  54 @@ -505,7 +537,7 @@ repeat:
  55         ei->i_prealloc_count = 0;
  56  #endif
  57         ei->i_block_group = i;
  58 -
  59 +
  60         if (ei->i_flags & EXT3_SYNC_FL)
  61                 inode->i_flags |= S_SYNC;
  62         if (IS_SYNC(inode))
  63 @@ -514,9 +546,18 @@ repeat:
  64         inode->i_generation = sbi->s_next_generation++;
  65
  66         ei->i_state = EXT3_STATE_NEW;
  67 -       err = ext3_mark_inode_dirty(handle, inode);
  68 +       err = ext3_get_inode_loc_new(inode, &iloc, 1);
  69         if (err) goto fail;
  70 -
  71 +       BUFFER_TRACE(iloc->bh, "get_write_access");
  72 +       err = ext3_journal_get_write_access(handle, iloc.bh);
  73 +       if (err) {
  74 +               brelse(iloc.bh);
  75 +               iloc.bh = NULL;
  76 +               goto fail;
  77 +       }
  78 +       err = ext3_mark_iloc_dirty(handle, inode, &iloc);
  79 +       if (err) goto fail;
  80 +
  81         unlock_super (sb);
  82         if(DQUOT_ALLOC_INODE(inode)) {
  83                 DQUOT_DROP(inode);
  84 --- linux-2.4.18-chaos52/fs/ext3/inode.c~extN-noread    2003-05-16 12:26:29.000000000 +0800
  85 +++ linux-2.4.18-chaos52-root/fs/ext3/inode.c   2003-05-16 12:27:06.000000000 +0800
  86 @@ -2013,21 +2013,26 @@ out_stop:
  87
  88 -/*
  89 - * ext3_get_inode_loc returns with an extra refcount against the
  90 - * inode's underlying buffer_head on success.
  91 - */
  92 +#define NUM_INODE_PREREAD      16
  93
  94 -int ext3_get_inode_loc (struct inode *inode, struct ext3_iloc *iloc)
  95 +/*
  96 + * ext3_get_inode_loc returns with an extra refcount against the inode's
  97 + * underlying buffer_head on success.  If this is for a new inode allocation
  98 + * (new is non-zero) then we may be able to optimize away the read if there
  99 + * are no other in-use inodes in this inode table block.  If we need to do
 100 + * a read, then read in a whole chunk of blocks to avoid blocking again soon
 101 + * if we are doing lots of creates/updates.
 102 + */
 103 +int ext3_get_inode_loc_new(struct inode *inode, struct ext3_iloc *iloc, int new)
 104  {
 105         struct super_block *sb = inode->i_sb;
 106         struct ext3_sb_info *sbi = EXT3_SB(sb);
 107 -       struct buffer_head *bh = 0;
 108 +       struct buffer_head *bh[NUM_INODE_PREREAD];
 109         unsigned long block;
 110         unsigned long block_group;
 111         unsigned long group_desc;
 112         unsigned long desc;
 113         unsigned long offset;
 114         struct ext3_group_desc * gdp;
 115 -
 116 +
 117         if ((inode->i_ino != EXT3_ROOT_INO &&
 118                 inode->i_ino != EXT3_JOURNAL_INO &&
 119                 inode->i_ino < EXT3_FIRST_INO(sb)) ||
 120 @@ -2042,38 +2047,86 @@ int ext3_get_inode_loc (struct inode *in
 121         }
 122         group_desc = block_group >> sbi->s_desc_per_block_bits;
 123         desc = block_group & (sbi->s_desc_per_block - 1);
 124 -       bh = sbi->s_group_desc[group_desc];
 125 -       if (!bh) {
 126 +       if (!sbi->s_group_desc[group_desc]) {
 127                 ext3_error(sb, __FUNCTION__, "Descriptor not loaded");
 128                 goto bad_inode;
 129         }
 130
 131 -       gdp = (struct ext3_group_desc *) bh->b_data;
 132 +       gdp = (struct ext3_group_desc *)(sbi->s_group_desc[group_desc]->b_data);
 133 +
 134         /*
 135          * Figure out the offset within the block group inode table
 136          */
 137 -       offset = ((inode->i_ino - 1) % sbi->s_inodes_per_group) *
 138 -               sbi->s_inode_size;
 139 +       offset = ((inode->i_ino - 1) % sbi->s_inodes_per_group);
 140 +
 141         block = le32_to_cpu(gdp[desc].bg_inode_table) +
 142 -               (offset >> EXT3_BLOCK_SIZE_BITS(sb));
 143 -       if (!(bh = sb_bread(sb, block))) {
 144 -               ext3_error (sb, __FUNCTION__,
 145 -                           "unable to read inode block - "
 146 -                           "inode=%lu, block=%lu", inode->i_ino, block);
 147 -               goto bad_inode;
 148 +               (offset * sbi->s_inode_size >> EXT3_BLOCK_SIZE_BITS(sb));
 149 +
 150 +       bh[0] = sb_getblk(sb, block);
 151 +       if (buffer_uptodate(bh[0]))
 152 +               goto done;
 153 +
 154 +       /* If we don't really need to read this block, and it isn't already
 155 +        * in memory, then we just zero it out.  Otherwise, we keep the
 156 +        * current block contents (deleted inode data) for posterity.
 157 +        */
 158 +       if (new && !ext3_itable_block_used(sb, block_group, offset)) {
 159 +               lock_buffer(bh[0]);
 160 +               memset(bh[0]->b_data, 0, bh[0]->b_size);
 161 +               mark_buffer_uptodate(bh[0], 1);
 162 +               unlock_buffer(bh[0]);
 163 +       } else {
 164 +               unsigned long block_end, itable_end;
 165 +               int count = 1;
 166 +
 167 +               itable_end = le32_to_cpu(gdp[desc].bg_inode_table) +
 168 +                               sbi->s_itb_per_group;
 169 +               block_end = block + NUM_INODE_PREREAD;
 170 +               if (block_end > itable_end)
 171 +                       block_end = itable_end;
 172 +
 173 +               for (++block; block < block_end; block++) {
 174 +                       bh[count] = sb_getblk(sb, block);
 175 +                       if (count && (buffer_uptodate(bh[count]) ||
 176 +                                     buffer_locked(bh[count]))) {
 177 +                               __brelse(bh[count]);
 178 +                       } else
 179 +                               count++;
 180 +               }
 181 +
 182 +               ll_rw_block(READ, count, bh);
 183 +
 184 +               /* Release all but the block we actually need (bh[0]) */
 185 +               while (--count > 0)
 186 +                       __brelse(bh[count]);
 187 +
 188 +               wait_on_buffer(bh[0]);
 189 +               if (!buffer_uptodate(bh[0])) {
 190 +                       ext3_error(sb, __FUNCTION__,
 191 +                                  "unable to read inode block - "
 192 +                                  "inode=%lu, block=%lu", inode->i_ino,
 193 +                                  bh[0]->b_blocknr);
 194 +                       goto bad_inode;
 195 +               }
 196         }
 197 -       offset &= (EXT3_BLOCK_SIZE(sb) - 1);
 198 + done:
 199 +       offset = (offset * sbi->s_inode_size) & (EXT3_BLOCK_SIZE(sb) - 1);
 200
 201 -       iloc->bh = bh;
 202 -       iloc->raw_inode = (struct ext3_inode *) (bh->b_data + offset);
 203 +       iloc->bh = bh[0];
 204 +       iloc->raw_inode = (struct ext3_inode *)(bh[0]->b_data + offset);
 205         iloc->block_group = block_group;
 206 -
 207 +
 208         return 0;
 209 -
 210 +
 211   bad_inode:
 212         return -EIO;
 213  }
 214
 215 +int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
 216 +{
 217 +       return ext3_get_inode_loc_new(inode, iloc, 0);
 218 +}
 219 +
 220  void ext3_read_inode(struct inode * inode)
 221  {
 222         struct ext3_iloc iloc;
 223 --- linux-2.4.18-chaos52/include/linux/ext3_fs.h~extN-noread    2003-05-16 12:26:29.000000000 +0800
 224 +++ linux-2.4.18-chaos52-root/include/linux/ext3_fs.h   2003-05-16 12:26:31.000000000 +0800
 225 @@ -640,6 +640,8 @@ extern int ext3_forget(handle_t *, int,
 226  extern struct buffer_head * ext3_getblk (handle_t *, struct inode *, long, int, int *);
 227  extern struct buffer_head * ext3_bread (handle_t *, struct inode *, int, int, int *);
 228
 229 +extern int ext3_itable_block_used(struct super_block *sb, unsigned int, int);
 230 +extern int ext3_get_inode_loc_new(struct inode *, struct ext3_iloc *, int);
 231  extern int  ext3_get_inode_loc (struct inode *, struct ext3_iloc *);
 232  extern void ext3_read_inode (struct inode *);
 233  extern void ext3_write_inode (struct inode *, int);
 234
 235 _