lustre/kernel_patches/patches/ext3-o_direct-1-2.4.18-chaos.patch

   1
   2 --- linux-2.4.18/fs/ext3/inode.c~ext3-o_direct-1-2.4.18-chaos   Mon Jul 28 14:13:05 2003
   3 +++ linux-2.4.18-alexey/fs/ext3/inode.c Mon Jul 28 15:50:14 2003
   4 @@ -27,6 +27,7 @@
   5  #include <linux/ext3_jbd.h>
   6  #include <linux/jbd.h>
   7  #include <linux/locks.h>
   8 +#include <linux/iobuf.h>
   9  #include <linux/smp_lock.h>
  10  #include <linux/highuid.h>
  11  #include <linux/quotaops.h>
  12 @@ -733,9 +734,9 @@ err_out:
  13   * The BKL may not be held on entry here.  Be sure to take it early.
  14   */
  15
  16 -static int ext3_get_block_handle(handle_t *handle, struct inode *inode,
  17 -                                long iblock,
  18 -                                struct buffer_head *bh_result, int create)
  19 +static int
  20 +ext3_get_block_handle(handle_t *handle, struct inode *inode, long iblock,
  21 +               struct buffer_head *bh_result, int create, int extend_disksize)
  22  {
  23         int err = -EIO;
  24         int offsets[4];
  25 @@ -816,16 +817,18 @@ out:
  26         if (err)
  27                 goto cleanup;
  28
  29 -       new_size = inode->i_size;
  30 -       /*
  31 -        * This is not racy against ext3_truncate's modification of i_disksize
  32 -        * because VM/VFS ensures that the file cannot be extended while
  33 -        * truncate is in progress.  It is racy between multiple parallel
  34 -        * instances of get_block, but we have the BKL.
  35 -        */
  36 -       if (new_size > ei->i_disksize)
  37 -               ei->i_disksize = new_size;
  38 -
  39 +       if (extend_disksize) {
  40 +               /*
  41 +                * This is not racy against ext3_truncate's modification of
  42 +                * i_disksize because VM/VFS ensures that the file cannot be
  43 +                * extended while truncate is in progress.  It is racy between
  44 +                * multiple parallel instances of get_block, but we have BKL.
  45 +                */
  46 +               struct ext3_inode_info *ei = EXT3_I(inode);
  47 +               new_size = inode->i_size;
  48 +               if (new_size > ei->i_disksize)
  49 +                       ei->i_disksize = new_size;
  50 +       }
  51         bh_result->b_state |= (1UL << BH_New);
  52         goto got_it;
  53
  54 @@ -852,7 +855,38 @@ static int ext3_get_block(struct inode *
  55                 handle = ext3_journal_current_handle();
  56                 J_ASSERT(handle != 0);
  57         }
  58 -       ret = ext3_get_block_handle(handle, inode, iblock, bh_result, create);
  59 +       ret = ext3_get_block_handle(handle, inode, iblock,
  60 +                               bh_result, create, 1);
  61 +       return ret;
  62 +}
  63 +
  64 +#define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
  65 +
  66 +static int
  67 +ext3_direct_io_get_block(struct inode *inode, long iblock,
  68 +               struct buffer_head *bh_result, int create)
  69 +{
  70 +       handle_t *handle = journal_current_handle();
  71 +       int ret = 0;
  72 +
  73 +       lock_kernel();
  74 +       if (handle && handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
  75 +               /*
  76 +                * Getting low on buffer credits...
  77 +                */
  78 +               if (!ext3_journal_extend(handle, DIO_CREDITS)) {
  79 +                       /*
  80 +                        * Couldn't extend the transaction.  Start a new one
  81 +                        */
  82 +                       ret = ext3_journal_restart(handle, DIO_CREDITS);
  83 +               }
  84 +       }
  85 +       if (ret == 0)
  86 +               ret = ext3_get_block_handle(handle, inode, iblock,
  87 +                                       bh_result, create, 0);
  88 +       if (ret == 0)
  89 +               bh_result->b_size = (1 << inode->i_blkbits);
  90 +       unlock_kernel();
  91         return ret;
  92  }
  93
  94 @@ -870,7 +904,7 @@ struct buffer_head *ext3_getblk(handle_t
  95         dummy.b_state = 0;
  96         dummy.b_blocknr = -1000;
  97         buffer_trace_init(&dummy.b_history);
  98 -       *errp = ext3_get_block_handle(handle, inode, block, &dummy, create);
  99 +       *errp = ext3_get_block_handle(handle, inode, block, &dummy, create, 1);
 100         if (!*errp && buffer_mapped(&dummy)) {
 101                 struct buffer_head *bh;
 102                 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
 103 @@ -1380,6 +1414,67 @@ static int ext3_releasepage(struct page
 104         return journal_try_to_free_buffers(journal, page, wait);
 105  }
 106
 107 +static int
 108 +ext3_direct_IO(int rw, struct inode *inode, struct kiobuf *iobuf,
 109 +               unsigned long blocknr, int blocksize)
 110 +{
 111 +       struct ext3_inode_info *ei = EXT3_I(inode);
 112 +       handle_t *handle = NULL;
 113 +       int ret;
 114 +       int orphan = 0;
 115 +       loff_t offset = blocknr << inode->i_blkbits;    /* ugh */
 116 +       ssize_t count = iobuf->length;                  /* ditto */
 117 +
 118 +       if (rw == WRITE) {
 119 +               loff_t final_size = offset + count;
 120 +
 121 +               lock_kernel();
 122 +               handle = ext3_journal_start(inode, DIO_CREDITS);
 123 +               unlock_kernel();
 124 +               if (IS_ERR(handle)) {
 125 +                       ret = PTR_ERR(handle);
 126 +                       goto out;
 127 +               }
 128 +               if (final_size > inode->i_size) {
 129 +                       lock_kernel();
 130 +                       ret = ext3_orphan_add(handle, inode);
 131 +                       unlock_kernel();
 132 +                       if (ret)
 133 +                               goto out_stop;
 134 +                       orphan = 1;
 135 +                       ei->i_disksize = inode->i_size;
 136 +               }
 137 +       }
 138 +
 139 +       ret = generic_direct_IO(rw, inode, iobuf, blocknr,
 140 +                               blocksize, ext3_direct_io_get_block);
 141 +
 142 +out_stop:
 143 +       if (handle) {
 144 +               int err;
 145 +
 146 +               lock_kernel();
 147 +               if (orphan)
 148 +                       ext3_orphan_del(handle, inode);
 149 +               if (orphan && ret > 0) {
 150 +                       loff_t end = offset + ret;
 151 +                       if (end > inode->i_size) {
 152 +                               ei->i_disksize = end;
 153 +                               inode->i_size = end;
 154 +                               err = ext3_mark_inode_dirty(handle, inode);
 155 +                               if (!ret)
 156 +                                       ret = err;
 157 +                       }
 158 +               }
 159 +               err = ext3_journal_stop(handle, inode);
 160 +               if (ret == 0)
 161 +                       ret = err;
 162 +               unlock_kernel();
 163 +       }
 164 +out:
 165 +       return ret;
 166 +
 167 +}
 168
 169  struct address_space_operations ext3_aops = {
 170         readpage:       ext3_readpage,          /* BKL not held.  Don't need */
 171 @@ -1390,6 +1485,7 @@ struct address_space_operations ext3_aop
 172         bmap:           ext3_bmap,              /* BKL held */
 173         flushpage:      ext3_flushpage,         /* BKL not held.  Don't need */
 174         releasepage:    ext3_releasepage,       /* BKL not held.  Don't need */
 175 +       direct_IO:      ext3_direct_IO,         /* BKL not held.  Don't need */
 176  };
 177
 178  /*
 179 @@ -2987,7 +3083,7 @@ int ext3_prep_san_write(struct inode *in
 180         /* alloc blocks one by one */
 181         for (i = 0; i < nblocks; i++) {
 182                 ret = ext3_get_block_handle(handle, inode, blocks[i],
 183 -                                               &bh_tmp, 1);
 184 +                                               &bh_tmp, 1, 1);
 185                 if (ret)
 186                         break;
 187
 188 @@ -3047,7 +3143,7 @@ int ext3_map_inode_page(struct inode *in
 189                  if (blocks[i] != 0)
 190                          continue;
 191
 192 -                rc = ext3_get_block_handle(handle, inode, iblock, &bh, 1);
 193 +                rc = ext3_get_block_handle(handle, inode, iblock, &bh, 1, 1);
 194                  if (rc) {
 195                          printk(KERN_INFO "ext3_map_inode_page: error %d "
 196                                 "allocating block %ld\n", rc, iblock);
 197
 198 _