7 files changed, 105 insertions, 28 deletions

diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
index 0db6a1f724e1..2e3898316468 100644
--- a/fs/ocfs2/alloc.c
+++ b/fs/ocfs2/alloc.c
@@ -3726,6 +3726,7 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
                         u32 cpos,
                         u64 start_blk,
                         u32 new_clusters,
+                        u8 flags,
                         struct ocfs2_alloc_context *meta_ac)
 {
         int status;
@@ -3749,6 +3750,7 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
         rec.e_cpos = cpu_to_le32(cpos);
         rec.e_blkno = cpu_to_le64(start_blk);
         rec.e_leaf_clusters = cpu_to_le16(new_clusters);
+        rec.e_flags = flags;
 
         status = ocfs2_figure_insert_type(inode, fe_bh, &last_eb_bh, &rec,
                                           &insert);
diff --git a/fs/ocfs2/alloc.h b/fs/ocfs2/alloc.h
index d3acf45225c2..e3284f3eb6b9 100644
--- a/fs/ocfs2/alloc.h
+++ b/fs/ocfs2/alloc.h
@@ -34,6 +34,7 @@ int ocfs2_insert_extent(struct ocfs2_super *osb,
                         u32 cpos,
                         u64 start_blk,
                         u32 new_clusters,
+                        u8 flags,
                         struct ocfs2_alloc_context *meta_ac);
 struct ocfs2_cached_dealloc_ctxt;
 int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *di_bh,
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
index 8af923316d22..ec8b606b30e1 100644
--- a/fs/ocfs2/aops.c
+++ b/fs/ocfs2/aops.c
@@ -1136,7 +1136,7 @@ static int ocfs2_write_cluster(struct address_space *mapping,
                  */
                 tmp_pos = cpos;
                 ret = ocfs2_do_extend_allocation(OCFS2_SB(inode->i_sb), inode,
-                                                 &tmp_pos, 1, wc->w_di_bh,
+                                                 &tmp_pos, 1, 0, wc->w_di_bh,
                                                  wc->w_handle, data_ac,
                                                  meta_ac, NULL);
                 /*
diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
index c441ef1f2bad..0d5fdde959c8 100644
--- a/fs/ocfs2/dir.c
+++ b/fs/ocfs2/dir.c
@@ -368,7 +368,7 @@ int ocfs2_do_extend_dir(struct super_block *sb,
                 u32 offset = OCFS2_I(dir)->ip_clusters;
 
                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
-                                                    1, parent_fe_bh, handle,
+                                                    1, 0, parent_fe_bh, handle,
                                                     data_ac, meta_ac, NULL);
                 BUG_ON(status == -EAGAIN);
                 if (status < 0) {
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
index 6745086da6fd..3e21ad9a6dde 100644
--- a/fs/ocfs2/file.c
+++ b/fs/ocfs2/file.c
@@ -425,6 +425,7 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
                                struct inode *inode,
                                u32 *logical_offset,
                                u32 clusters_to_add,
+                               int mark_unwritten,
                                struct buffer_head *fe_bh,
                                handle_t *handle,
                                struct ocfs2_alloc_context *data_ac,
@@ -437,9 +438,13 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
         enum ocfs2_alloc_restarted reason = RESTART_NONE;
         u32 bit_off, num_bits;
         u64 block;
+        u8 flags = 0;
 
         BUG_ON(!clusters_to_add);
 
+        if (mark_unwritten)
+                flags = OCFS2_EXT_UNWRITTEN;
+
         free_extents = ocfs2_num_free_extents(osb, inode, fe);
         if (free_extents < 0) {
                 status = free_extents;
@@ -489,7 +494,7 @@ int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
         status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
                                      *logical_offset, block, num_bits,
-                                     meta_ac);
+                                     flags, meta_ac);
         if (status < 0) {
                 mlog_errno(status);
                 goto leave;
@@ -522,9 +527,11 @@ leave:
  * For a given allocation, determine which allocators will need to be
  * accessed, and lock them, reserving the appropriate number of bits.
  *
- * Called from ocfs2_extend_allocation() for file systems which don't
- * support holes, and from ocfs2_write() for file systems which
- * understand sparse inodes.
+ * Sparse file systems call this from ocfs2_write_begin_nolock()
+ * and ocfs2_allocate_unwritten_extents().
+ *
+ * File systems which don't support holes call this from
+ * ocfs2_extend_allocation().
  */
 int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
                           u32 clusters_to_add, u32 extents_to_split,
@@ -595,14 +602,13 @@ out:
         return ret;
 }
 
-static int ocfs2_extend_allocation(struct inode *inode,
-                                   u32 clusters_to_add)
+static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
+                                     u32 clusters_to_add, int mark_unwritten)
 {
         int status = 0;
         int restart_func = 0;
-        int drop_alloc_sem = 0;
         int credits;
-        u32 prev_clusters, logical_start;
+        u32 prev_clusters;
         struct buffer_head *bh = NULL;
         struct ocfs2_dinode *fe = NULL;
         handle_t *handle = NULL;
@@ -617,7 +623,7 @@ static int ocfs2_extend_allocation(struct inode *inode,
          * This function only exists for file systems which don't
          * support holes.
          */
-        BUG_ON(ocfs2_sparse_alloc(osb));
+        BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
 
         status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
                                   OCFS2_BH_CACHED, inode);
@@ -633,18 +639,9 @@ static int ocfs2_extend_allocation(struct inode *inode,
                 goto leave;
         }
 
-        logical_start = OCFS2_I(inode)->ip_clusters;
-
 restart_all:
         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
 
-        /* blocks peope in read/write from reading our allocation
-         * until we're done changing it. We depend on i_mutex to block
-         * other extend/truncate calls while we're here. Ordering wrt
-         * start_trans is important here -- always do it before! */
-        down_write(&OCFS2_I(inode)->ip_alloc_sem);
-        drop_alloc_sem = 1;
-
         status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
                                        &meta_ac);
         if (status) {
@@ -678,6 +675,7 @@ restarted_transaction:
                                             inode,
                                             &logical_start,
                                             clusters_to_add,
+                                            mark_unwritten,
                                             bh,
                                             handle,
                                             data_ac,
@@ -730,10 +728,6 @@ restarted_transaction:
              OCFS2_I(inode)->ip_clusters, i_size_read(inode));
 
 leave:
-        if (drop_alloc_sem) {
-                up_write(&OCFS2_I(inode)->ip_alloc_sem);
-                drop_alloc_sem = 0;
-        }
         if (handle) {
                 ocfs2_commit_trans(osb, handle);
                 handle = NULL;
@@ -759,6 +753,25 @@ leave:
         return status;
 }
 
+static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
+                                   u32 clusters_to_add, int mark_unwritten)
+{
+        int ret;
+
+        /*
+         * The alloc sem blocks peope in read/write from reading our
+         * allocation until we're done changing it. We depend on
+         * i_mutex to block other extend/truncate calls while we're
+         * here.
+         */
+        down_write(&OCFS2_I(inode)->ip_alloc_sem);
+        ret = __ocfs2_extend_allocation(inode, logical_start, clusters_to_add,
+                                        mark_unwritten);
+        up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+        return ret;
+}
+
 /* Some parts of this taken from generic_cont_expand, which turned out
  * to be too fragile to do exactly what we need without us having to
  * worry about recursive locking in ->prepare_write() and
@@ -900,7 +913,9 @@ static int ocfs2_extend_file(struct inode *inode,
         }
 
         if (clusters_to_add) {
-                ret = ocfs2_extend_allocation(inode, clusters_to_add);
+                ret = ocfs2_extend_allocation(inode,
+                                              OCFS2_I(inode)->ip_clusters,
+                                              clusters_to_add, 0);
                 if (ret < 0) {
                         mlog_errno(ret);
                         goto out_unlock;
@@ -1176,6 +1191,64 @@ out:
         return ret;
 }
 
+/*
+ * Allocate enough extents to cover the region starting at byte offset
+ * start for len bytes. Existing extents are skipped, any extents
+ * added are marked as "unwritten".
+ */
+static int ocfs2_allocate_unwritten_extents(struct inode *inode,
+                                            u64 start, u64 len)
+{
+        int ret;
+        u32 cpos, phys_cpos, clusters, alloc_size;
+
+        /*
+         * We consider both start and len to be inclusive.
+         */
+        cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+        clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
+        clusters -= cpos;
+
+        while (clusters) {
+                ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
+                                         &alloc_size, NULL);
+                if (ret) {
+                        mlog_errno(ret);
+                        goto out;
+                }
+
+                /*
+                 * Hole or existing extent len can be arbitrary, so
+                 * cap it to our own allocation request.
+                 */
+                if (alloc_size > clusters)
+                        alloc_size = clusters;
+
+                if (phys_cpos) {
+                        /*
+                         * We already have an allocation at this
+                         * region so we can safely skip it.
+                         */
+                        goto next;
+                }
+
+                ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
+                if (ret) {
+                        if (ret != -ENOSPC)
+                                mlog_errno(ret);
+                        goto out;
+                }
+
+next:
+                cpos += alloc_size;
+                clusters -= alloc_size;
+        }
+
+        ret = 0;
+out:
+        return ret;
+}
+
 static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
                                          loff_t *ppos,
                                          size_t count,
diff --git a/fs/ocfs2/file.h b/fs/ocfs2/file.h
index 54df3c4bd2fd..79115c92dc30 100644
--- a/fs/ocfs2/file.h
+++ b/fs/ocfs2/file.h
@@ -39,13 +39,14 @@ enum ocfs2_alloc_restarted {
 };
 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
                                struct inode *inode,
-                               u32 *cluster_start,
+                               u32 *logical_offset,
                                u32 clusters_to_add,
+                               int mark_unwritten,
                                struct buffer_head *fe_bh,
                                handle_t *handle,
                                struct ocfs2_alloc_context *data_ac,
                                struct ocfs2_alloc_context *meta_ac,
-                               enum ocfs2_alloc_restarted *reason);
+                               enum ocfs2_alloc_restarted *reason_ret);
 int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
                           u32 clusters_to_add, u32 extents_to_split,
                           struct ocfs2_alloc_context **data_ac,
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
index 36289e6295ce..d430fdab16e9 100644
--- a/fs/ocfs2/namei.c
+++ b/fs/ocfs2/namei.c
@@ -1674,7 +1674,7 @@ static int ocfs2_symlink(struct inode *dir,
                 u32 offset = 0;
 
                 inode->i_op = &ocfs2_symlink_inode_operations;
-                status = ocfs2_do_extend_allocation(osb, inode, &offset, 1,
+                status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
                                                     new_fe_bh,
                                                     handle, data_ac, NULL,
                                                     NULL);