ext4-delalloc-extents-48bit.patch

   1 [RFC] ext4-delayed-allocation.patch
   2
   3 Hi Alex,
   4
   5 I tested your patch on my system with a 20TB device, but some tests
   6 failed. Looking at the code, I saw that the support of 48-bit block
   7 number in extents is lacking.
   8
   9 I made some changes in the code (see the patch in attachment) and now
  10 all my tests are OK.
  11 The patch is not complete, I didn't update calls to wb_debug() which
  12 dump "ee_start".
  13
  14 Hope this helps.
  15
  16   Valérie
  17
  18 From: Valerie Clement <valerie.clement@bull.net>
  19
  20 ---
  21
  22  fs/ext4/writeback.c |   19 +++++++++++++------
  23  1 files changed, 13 insertions(+), 6 deletions(-)
  24
  25 diff --git a/fs/ext4/writeback.c b/fs/ext4/writeback.c
  26 index 8861500..20e1d2e 100644
  27 --- a/fs/ext4/writeback.c
  28 +++ b/fs/ext4/writeback.c
  29 @@ -274,7 +274,8 @@ static int ext4_wb_submit_extent(struct ext4_wb_control *wc, handle_t *handle,
  30         struct inode *inode = wc->mapping->host;
  31         int blkbits = inode->i_blkbits;
  32         struct page *page;
  33 -       unsigned long blk, off, len, remain;
  34 +       ext4_fsblk_t off;
  35 +       unsigned long blk, len, remain;
  36         unsigned long pstart, plen, prev;
  37         struct bio *bio = NULL;
  38         int nr_pages;
  39 @@ -332,6 +333,7 @@ alloc_new_bio:
  40                         nr_pages = (ex->ee_len - (blk - ex->ee_block));
  41                         nr_pages = (nr_pages >> (PAGE_CACHE_SHIFT - blkbits));
  42                         off = ex->ee_start + (blk - ex->ee_block);
  43 +                       off |= (ext4_fsblk_t) ex->ee_start_hi << 32;
  44                         bio = ext4_wb_bio_alloc(inode, off, nr_pages + 2);
  45                         if (bio == NULL)
  46                                 return -ENOMEM;
  47 @@ -377,7 +379,9 @@ ext4_wb_find_goal(struct inode *inode, struct ext4_ext_path *path,
  48
  49                 /* try to predict block placement */
  50                 if ((ex = path[depth].p_ext))
  51 -                       return ex->ee_start + (block - ex->ee_block);
  52 +                       return ((ex->ee_start
  53 +                               | ((ext4_fsblk_t) ex->ee_start_hi << 32))
  54 +                               + (block - ex->ee_block));
  55
  56                 /* it looks index is empty
  57                  * try to find starting from index itself */
  58 @@ -416,7 +420,8 @@ static int ext4_wb_handle_extent(struct inode *inode,
  59                                 (unsigned) ec->ec_block,
  60                                 (unsigned) ec->ec_len,
  61                                 (unsigned) ec->ec_start);
  62 -               nex.ee_start = ec->ec_start;
  63 +               nex.ee_start = ec->ec_start & 0xffffffff;
  64 +               nex.ee_start_hi = (ec->ec_start >> 32) & 0xffff;
  65                 nex.ee_block = ec->ec_block;
  66                 nex.ee_len = ec->ec_len;
  67                 err = ext4_wb_submit_extent(wc, NULL, &nex, 0);
  68 @@ -488,8 +493,8 @@ static int ext4_wb_handle_extent(struct inode *inode,
  69                         pblock, count, inode->i_ino, ec->ec_len);
  70
  71         /* insert new extent */
  72 -       nex.ee_start = pblock;
  73 -       nex.ee_start_hi = 0;
  74 +       nex.ee_start = pblock & 0xffffffff;
  75 +       nex.ee_start_hi = (pblock >> 32) & 0xffff;
  76         nex.ee_len = count;
  77         nex.ee_block = ec->ec_block;
  78         err = ext4_ext_insert_extent(handle, inode, path, &nex);
  79 @@ -520,7 +525,9 @@ static int ext4_wb_handle_extent(struct inode *inode,
  80         /* block have been allocated for data, so time to drop dirty
  81          * in correspondend buffer_heads to prevent corruptions */
  82         for (i = 0; i < nex.ee_len; i++)
  83 -               unmap_underlying_metadata(sb->s_bdev, nex.ee_start + i);
  84 +               unmap_underlying_metadata(sb->s_bdev,
  85 +                               ((ext4_fsblk_t) nex.ee_start_hi << 32)
  86 +                               + nex.ee_start + i);
  87
  88         /* correct on-disk inode size */
  89         if (nex.ee_len > 0) {