lustre/kernel_patches/patches/ext3-o_direct-1.patch

   1 --- 2.4.20-pre2/fs/ext3/inode.c~ext3-o_direct   Thu Aug 15 01:11:02 2002
   2 +++ 2.4.20-pre2-akpm/fs/ext3/inode.c    Thu Aug 15 01:25:55 2002
   3 @@ -27,6 +27,7 @@
   4  #include <linux/ext3_jbd.h>
   5  #include <linux/jbd.h>
   6  #include <linux/locks.h>
   7 +#include <linux/iobuf.h>
   8  #include <linux/smp_lock.h>
   9  #include <linux/highuid.h>
  10  #include <linux/quotaops.h>
  11 @@ -716,9 +717,9 @@ err_out:
  12   * The BKL may not be held on entry here.  Be sure to take it early.
  13   */
  14
  15 -static int ext3_get_block_handle(handle_t *handle, struct inode *inode,
  16 -                                long iblock,
  17 -                                struct buffer_head *bh_result, int create)
  18 +static int
  19 +ext3_get_block_handle(handle_t *handle, struct inode *inode, long iblock,
  20 +               struct buffer_head *bh_result, int create, int extend_disksize)
  21  {
  22         int err = -EIO;
  23         int offsets[4];
  24 @@ -798,16 +799,18 @@ out:
  25         if (err)
  26                 goto cleanup;
  27
  28 -       new_size = inode->i_size;
  29 -       /*
  30 -        * This is not racy against ext3_truncate's modification of i_disksize
  31 -        * because VM/VFS ensures that the file cannot be extended while
  32 -        * truncate is in progress.  It is racy between multiple parallel
  33 -        * instances of get_block, but we have the BKL.
  34 -        */
  35 -       if (new_size > inode->u.ext3_i.i_disksize)
  36 -               inode->u.ext3_i.i_disksize = new_size;
  37 -
  38 +       if (extend_disksize) {
  39 +               /*
  40 +                * This is not racy against ext3_truncate's modification of
  41 +                * i_disksize because VM/VFS ensures that the file cannot be
  42 +                * extended while truncate is in progress.  It is racy between
  43 +                * multiple parallel instances of get_block, but we have BKL.
  44 +                */
  45 +               struct ext3_inode_info *ei = EXT3_I(inode);
  46 +               new_size = inode->i_size;
  47 +               if (new_size > ei->i_disksize)
  48 +                       ei->i_disksize = new_size;
  49 +       }
  50         bh_result->b_state |= (1UL << BH_New);
  51         goto got_it;
  52
  53 @@ -834,7 +837,38 @@ static int ext3_get_block(struct inode *
  54                 handle = ext3_journal_current_handle();
  55                 J_ASSERT(handle != 0);
  56         }
  57 -       ret = ext3_get_block_handle(handle, inode, iblock, bh_result, create);
  58 +       ret = ext3_get_block_handle(handle, inode, iblock,
  59 +                               bh_result, create, 1);
  60 +       return ret;
  61 +}
  62 +
  63 +#define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
  64 +
  65 +static int
  66 +ext3_direct_io_get_block(struct inode *inode, long iblock,
  67 +               struct buffer_head *bh_result, int create)
  68 +{
  69 +       handle_t *handle = journal_current_handle();
  70 +       int ret = 0;
  71 +
  72 +       lock_kernel();
  73 +       if (handle && handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
  74 +               /*
  75 +                * Getting low on buffer credits...
  76 +                */
  77 +               if (!ext3_journal_extend(handle, DIO_CREDITS)) {
  78 +                       /*
  79 +                        * Couldn't extend the transaction.  Start a new one
  80 +                        */
  81 +                       ret = ext3_journal_restart(handle, DIO_CREDITS);
  82 +               }
  83 +       }
  84 +       if (ret == 0)
  85 +               ret = ext3_get_block_handle(handle, inode, iblock,
  86 +                                       bh_result, create, 0);
  87 +       if (ret == 0)
  88 +               bh_result->b_size = (1 << inode->i_blkbits);
  89 +       unlock_kernel();
  90         return ret;
  91  }
  92
  93 @@ -852,7 +886,7 @@ struct buffer_head *ext3_getblk(handle_t
  94         dummy.b_state = 0;
  95         dummy.b_blocknr = -1000;
  96         buffer_trace_init(&dummy.b_history);
  97 -       *errp = ext3_get_block_handle(handle, inode, block, &dummy, create);
  98 +       *errp = ext3_get_block_handle(handle, inode, block, &dummy, create, 1);
  99         if (!*errp && buffer_mapped(&dummy)) {
 100                 struct buffer_head *bh;
 101                 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
 102 @@ -1349,6 +1383,67 @@ static int ext3_releasepage(struct page
 103         return journal_try_to_free_buffers(journal, page, wait);
 104  }
 105
 106 +static int
 107 +ext3_direct_IO(int rw, struct inode *inode, struct kiobuf *iobuf,
 108 +               unsigned long blocknr, int blocksize)
 109 +{
 110 +       struct ext3_inode_info *ei = EXT3_I(inode);
 111 +       handle_t *handle = NULL;
 112 +       int ret;
 113 +       int orphan = 0;
 114 +       loff_t offset = blocknr << inode->i_blkbits;    /* ugh */
 115 +       ssize_t count = iobuf->length;                  /* ditto */
 116 +
 117 +       if (rw == WRITE) {
 118 +               loff_t final_size = offset + count;
 119 +
 120 +               lock_kernel();
 121 +               handle = ext3_journal_start(inode, DIO_CREDITS);
 122 +               unlock_kernel();
 123 +               if (IS_ERR(handle)) {
 124 +                       ret = PTR_ERR(handle);
 125 +                       goto out;
 126 +               }
 127 +               if (final_size > inode->i_size) {
 128 +                       lock_kernel();
 129 +                       ret = ext3_orphan_add(handle, inode);
 130 +                       unlock_kernel();
 131 +                       if (ret)
 132 +                               goto out_stop;
 133 +                       orphan = 1;
 134 +                       ei->i_disksize = inode->i_size;
 135 +               }
 136 +       }
 137 +
 138 +       ret = generic_direct_IO(rw, inode, iobuf, blocknr,
 139 +                               blocksize, ext3_direct_io_get_block);
 140 +
 141 +out_stop:
 142 +       if (handle) {
 143 +               int err;
 144 +
 145 +               lock_kernel();
 146 +               if (orphan)
 147 +                       ext3_orphan_del(handle, inode);
 148 +               if (orphan && ret > 0) {
 149 +                       loff_t end = offset + ret;
 150 +                       if (end > inode->i_size) {
 151 +                               ei->i_disksize = end;
 152 +                               inode->i_size = end;
 153 +                               err = ext3_mark_inode_dirty(handle, inode);
 154 +                               if (!ret)
 155 +                                       ret = err;
 156 +                       }
 157 +               }
 158 +               err = ext3_journal_stop(handle, inode);
 159 +               if (ret == 0)
 160 +                       ret = err;
 161 +               unlock_kernel();
 162 +       }
 163 +out:
 164 +       return ret;
 165 +
 166 +}
 167
 168  struct address_space_operations ext3_aops = {
 169         readpage:       ext3_readpage,          /* BKL not held.  Don't need */
 170 @@ -1359,6 +1454,7 @@ struct address_space_operations ext3_aop
 171         bmap:           ext3_bmap,              /* BKL held */
 172         flushpage:      ext3_flushpage,         /* BKL not held.  Don't need */
 173         releasepage:    ext3_releasepage,       /* BKL not held.  Don't need */
 174 +       direct_IO:      ext3_direct_IO,         /* BKL not held.  Don't need */
 175  };
 176
 177  /*