src/patches/suse-2.6.27.31/patches.fixes/ocfs2-initialize-the-cluster-we-re-writing-to.patch

   1 From e7432675f8ca868a4af365759a8d4c3779a3d922 Mon Sep 17 00:00:00 2001
   2 From: Sunil Mushran <sunil.mushran@oracle.com>
   3 Date: Thu, 6 Aug 2009 16:12:58 -0700
   4 Subject: ocfs2: Initialize the cluster we're writing to in a non-sparse extend
   5 References: bnc#501563
   6
   7 In a non-sparse extend, we correctly allocate (and zero) the clusters between
   8 the old_i_size and pos, but we don't zero the portions of the cluster we're
   9 writing to outside of pos<->len.
  10
  11 It handles clustersize > pagesize and blocksize < pagesize.
  12
  13 [Cleaned up by Joel Becker.]
  14
  15 Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
  16 Signed-off-by: Joel Becker <joel.becker@oracle.com>
  17 Signed-off-by: Mark Fasheh <mfasheh@suse.com>
  18 ---
  19  fs/ocfs2/aops.c |   66 +++++++++++++++++++++++++++++++++++++++---------------
  20  1 files changed, 47 insertions(+), 19 deletions(-)
  21
  22 Index: linux-2.6.27-ocfs2-update-sle11/fs/ocfs2/aops.c
  23 ===================================================================
  24 --- linux-2.6.27-ocfs2-update-sle11.orig/fs/ocfs2/aops.c
  25 +++ linux-2.6.27-ocfs2-update-sle11/fs/ocfs2/aops.c
  26 @@ -909,18 +909,17 @@ struct ocfs2_write_cluster_desc {
  27          */
  28         unsigned        c_new;
  29         unsigned        c_unwritten;
  30 +       unsigned        c_needs_zero;
  31  };
  32
  33 -static inline int ocfs2_should_zero_cluster(struct ocfs2_write_cluster_desc *d)
  34 -{
  35 -       return d->c_new || d->c_unwritten;
  36 -}
  37 -
  38  struct ocfs2_write_ctxt {
  39         /* Logical cluster position / len of write */
  40         u32                             w_cpos;
  41         u32                             w_clen;
  42
  43 +       /* First cluster allocated in a nonsparse extend */
  44 +       u32                             w_first_new_cpos;
  45 +
  46         struct ocfs2_write_cluster_desc w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
  47
  48         /*
  49 @@ -998,6 +997,7 @@ static int ocfs2_alloc_write_ctxt(struct
  50                 return -ENOMEM;
  51
  52         wc->w_cpos = pos >> osb->s_clustersize_bits;
  53 +       wc->w_first_new_cpos = UINT_MAX;
  54         cend = (pos + len - 1) >> osb->s_clustersize_bits;
  55         wc->w_clen = cend - wc->w_cpos + 1;
  56         get_bh(di_bh);
  57 @@ -1239,20 +1239,18 @@ out:
  58   */
  59  static int ocfs2_write_cluster(struct address_space *mapping,
  60                                u32 phys, unsigned int unwritten,
  61 +                              unsigned int should_zero,
  62                                struct ocfs2_alloc_context *data_ac,
  63                                struct ocfs2_alloc_context *meta_ac,
  64                                struct ocfs2_write_ctxt *wc, u32 cpos,
  65                                loff_t user_pos, unsigned user_len)
  66  {
  67 -       int ret, i, new, should_zero = 0;
  68 +       int ret, i, new;
  69         u64 v_blkno, p_blkno;
  70         struct inode *inode = mapping->host;
  71         struct ocfs2_extent_tree et;
  72
  73         new = phys == 0 ? 1 : 0;
  74 -       if (new || unwritten)
  75 -               should_zero = 1;
  76 -
  77         if (new) {
  78                 u32 tmp_pos;
  79
  80 @@ -1363,7 +1361,9 @@ static int ocfs2_write_cluster_by_desc(s
  81                         local_len = osb->s_clustersize - cluster_off;
  82
  83                 ret = ocfs2_write_cluster(mapping, desc->c_phys,
  84 -                                         desc->c_unwritten, data_ac, meta_ac,
  85 +                                         desc->c_unwritten,
  86 +                                         desc->c_needs_zero,
  87 +                                         data_ac, meta_ac,
  88                                           wc, desc->c_cpos, pos, local_len);
  89                 if (ret) {
  90                         mlog_errno(ret);
  91 @@ -1413,14 +1413,14 @@ static void ocfs2_set_target_boundaries(
  92                  * newly allocated cluster.
  93                  */
  94                 desc = &wc->w_desc[0];
  95 -               if (ocfs2_should_zero_cluster(desc))
  96 +               if (desc->c_needs_zero)
  97                         ocfs2_figure_cluster_boundaries(osb,
  98                                                         desc->c_cpos,
  99                                                         &wc->w_target_from,
 100                                                         NULL);
 101
 102                 desc = &wc->w_desc[wc->w_clen - 1];
 103 -               if (ocfs2_should_zero_cluster(desc))
 104 +               if (desc->c_needs_zero)
 105                         ocfs2_figure_cluster_boundaries(osb,
 106                                                         desc->c_cpos,
 107                                                         NULL,
 108 @@ -1488,13 +1488,28 @@ static int ocfs2_populate_write_desc(str
 109                         phys++;
 110                 }
 111
 112 +               /*
 113 +                * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
 114 +                * file that got extended.  w_first_new_cpos tells us
 115 +                * where the newly allocated clusters are so we can
 116 +                * zero them.
 117 +                */
 118 +               if (desc->c_cpos >= wc->w_first_new_cpos) {
 119 +                       BUG_ON(phys == 0);
 120 +                       desc->c_needs_zero = 1;
 121 +               }
 122 +
 123                 desc->c_phys = phys;
 124                 if (phys == 0) {
 125                         desc->c_new = 1;
 126 +                       desc->c_needs_zero = 1;
 127                         *clusters_to_alloc = *clusters_to_alloc + 1;
 128                 }
 129 -               if (ext_flags & OCFS2_EXT_UNWRITTEN)
 130 +
 131 +               if (ext_flags & OCFS2_EXT_UNWRITTEN) {
 132                         desc->c_unwritten = 1;
 133 +                       desc->c_needs_zero = 1;
 134 +               }
 135
 136                 num_clusters--;
 137         }
 138 @@ -1654,10 +1669,13 @@ static int ocfs2_expand_nonsparse_inode(
 139         if (newsize <= i_size_read(inode))
 140                 return 0;
 141
 142 -       ret = ocfs2_extend_no_holes(inode, newsize, newsize - len);
 143 +       ret = ocfs2_extend_no_holes(inode, newsize, pos);
 144         if (ret)
 145                 mlog_errno(ret);
 146
 147 +       wc->w_first_new_cpos =
 148 +               ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
 149 +
 150         return ret;
 151  }
 152
 153 @@ -1666,7 +1684,7 @@ int ocfs2_write_begin_nolock(struct addr
 154                              struct page **pagep, void **fsdata,
 155                              struct buffer_head *di_bh, struct page *mmap_page)
 156  {
 157 -       int ret, credits = OCFS2_INODE_UPDATE_CREDITS;
 158 +       int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
 159         unsigned int clusters_to_alloc, extents_to_split;
 160         struct ocfs2_write_ctxt *wc;
 161         struct inode *inode = mapping->host;
 162 @@ -1744,8 +1762,19 @@ int ocfs2_write_begin_nolock(struct addr
 163
 164         }
 165
 166 -       ocfs2_set_target_boundaries(osb, wc, pos, len,
 167 -                                   clusters_to_alloc + extents_to_split);
 168 +       /*
 169 +        * We have to zero sparse allocated clusters, unwritten extent clusters,
 170 +        * and non-sparse clusters we just extended.  For non-sparse writes,
 171 +        * we know zeros will only be needed in the first and/or last cluster.
 172 +        */
 173 +       if (clusters_to_alloc || extents_to_split ||
 174 +           wc->w_desc[0].c_needs_zero ||
 175 +           wc->w_desc[wc->w_clen - 1].c_needs_zero)
 176 +               cluster_of_pages = 1;
 177 +       else
 178 +               cluster_of_pages = 0;
 179 +
 180 +       ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
 181
 182         handle = ocfs2_start_trans(osb, credits);
 183         if (IS_ERR(handle)) {
 184 @@ -1778,8 +1807,7 @@ int ocfs2_write_begin_nolock(struct addr
 185          * extent.
 186          */
 187         ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos,
 188 -                                        clusters_to_alloc + extents_to_split,
 189 -                                        mmap_page);
 190 +                                        cluster_of_pages, mmap_page);
 191         if (ret) {
 192                 mlog_errno(ret);
 193                 goto out_quota;