]> pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - fs/ext4/extents.c
Merge branch 'sg' of git://git.kernel.dk/linux-2.6-block
[linux-2.6-omap-h63xx.git] / fs / ext4 / extents.c
index ba25832a756c4cd5729eeacd284ecbe9c25defe3..85287742f2ae487562be3a8ac39e40774a5de9d3 100644 (file)
@@ -33,7 +33,7 @@
 #include <linux/fs.h>
 #include <linux/time.h>
 #include <linux/ext4_jbd2.h>
-#include <linux/jbd.h>
+#include <linux/jbd2.h>
 #include <linux/highuid.h>
 #include <linux/pagemap.h>
 #include <linux/quotaops.h>
@@ -52,7 +52,7 @@ static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
 {
        ext4_fsblk_t block;
 
-       block = le32_to_cpu(ex->ee_start);
+       block = le32_to_cpu(ex->ee_start_lo);
        block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
        return block;
 }
@@ -65,7 +65,7 @@ static ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
 {
        ext4_fsblk_t block;
 
-       block = le32_to_cpu(ix->ei_leaf);
+       block = le32_to_cpu(ix->ei_leaf_lo);
        block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
        return block;
 }
@@ -77,7 +77,7 @@ static ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
  */
 static void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
 {
-       ex->ee_start = cpu_to_le32((unsigned long) (pb & 0xffffffff));
+       ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
        ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
 }
 
@@ -88,40 +88,10 @@ static void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
  */
 static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
 {
-       ix->ei_leaf = cpu_to_le32((unsigned long) (pb & 0xffffffff));
+       ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
        ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
 }
 
-static int ext4_ext_check_header(const char *function, struct inode *inode,
-                               struct ext4_extent_header *eh)
-{
-       const char *error_msg = NULL;
-
-       if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
-               error_msg = "invalid magic";
-               goto corrupted;
-       }
-       if (unlikely(eh->eh_max == 0)) {
-               error_msg = "invalid eh_max";
-               goto corrupted;
-       }
-       if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
-               error_msg = "invalid eh_entries";
-               goto corrupted;
-       }
-       return 0;
-
-corrupted:
-       ext4_error(inode->i_sb, function,
-                       "bad header in inode #%lu: %s - magic %x, "
-                       "entries %u, max %u, depth %u",
-                       inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
-                       le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
-                       le16_to_cpu(eh->eh_depth));
-
-       return -EIO;
-}
-
 static handle_t *ext4_ext_journal_restart(handle_t *handle, int needed)
 {
        int err;
@@ -270,6 +240,70 @@ static int ext4_ext_space_root_idx(struct inode *inode)
        return size;
 }
 
+static int
+ext4_ext_max_entries(struct inode *inode, int depth)
+{
+       int max;
+
+       if (depth == ext_depth(inode)) {
+               if (depth == 0)
+                       max = ext4_ext_space_root(inode);
+               else
+                       max = ext4_ext_space_root_idx(inode);
+       } else {
+               if (depth == 0)
+                       max = ext4_ext_space_block(inode);
+               else
+                       max = ext4_ext_space_block_idx(inode);
+       }
+
+       return max;
+}
+
+static int __ext4_ext_check_header(const char *function, struct inode *inode,
+                                       struct ext4_extent_header *eh,
+                                       int depth)
+{
+       const char *error_msg;
+       int max = 0;
+
+       if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
+               error_msg = "invalid magic";
+               goto corrupted;
+       }
+       if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
+               error_msg = "unexpected eh_depth";
+               goto corrupted;
+       }
+       if (unlikely(eh->eh_max == 0)) {
+               error_msg = "invalid eh_max";
+               goto corrupted;
+       }
+       max = ext4_ext_max_entries(inode, depth);
+       if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
+               error_msg = "too large eh_max";
+               goto corrupted;
+       }
+       if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
+               error_msg = "invalid eh_entries";
+               goto corrupted;
+       }
+       return 0;
+
+corrupted:
+       ext4_error(inode->i_sb, function,
+                       "bad header in inode #%lu: %s - magic %x, "
+                       "entries %u, max %u(%u), depth %u(%u)",
+                       inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
+                       le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
+                       max, le16_to_cpu(eh->eh_depth), depth);
+
+       return -EIO;
+}
+
+#define ext4_ext_check_header(inode, eh, depth)        \
+       __ext4_ext_check_header(__FUNCTION__, inode, eh, depth)
+
 #ifdef EXT_DEBUG
 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
 {
@@ -330,6 +364,7 @@ static void ext4_ext_drop_refs(struct ext4_ext_path *path)
 /*
  * ext4_ext_binsearch_idx:
  * binary search for the closest index of the given block
+ * the header must be checked before calling this
  */
 static void
 ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int block)
@@ -337,27 +372,25 @@ ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int bloc
        struct ext4_extent_header *eh = path->p_hdr;
        struct ext4_extent_idx *r, *l, *m;
 
-       BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
-       BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
-       BUG_ON(le16_to_cpu(eh->eh_entries) <= 0);
 
        ext_debug("binsearch for %d(idx):  ", block);
 
        l = EXT_FIRST_INDEX(eh) + 1;
-       r = EXT_FIRST_INDEX(eh) + le16_to_cpu(eh->eh_entries) - 1;
+       r = EXT_LAST_INDEX(eh);
        while (l <= r) {
                m = l + (r - l) / 2;
                if (block < le32_to_cpu(m->ei_block))
                        r = m - 1;
                else
                        l = m + 1;
-               ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ei_block,
-                               m, m->ei_block, r, r->ei_block);
+               ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
+                               m, le32_to_cpu(m->ei_block),
+                               r, le32_to_cpu(r->ei_block));
        }
 
        path->p_idx = l - 1;
        ext_debug("  -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
-                 idx_block(path->p_idx));
+                 idx_pblock(path->p_idx));
 
 #ifdef CHECK_BINSEARCH
        {
@@ -389,6 +422,7 @@ ext4_ext_binsearch_idx(struct inode *inode, struct ext4_ext_path *path, int bloc
 /*
  * ext4_ext_binsearch:
  * binary search for closest extent of the given block
+ * the header must be checked before calling this
  */
 static void
 ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
@@ -396,9 +430,6 @@ ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
        struct ext4_extent_header *eh = path->p_hdr;
        struct ext4_extent *r, *l, *m;
 
-       BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
-       BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
-
        if (eh->eh_entries == 0) {
                /*
                 * this leaf is empty:
@@ -410,7 +441,7 @@ ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
        ext_debug("binsearch for %d:  ", block);
 
        l = EXT_FIRST_EXTENT(eh) + 1;
-       r = EXT_FIRST_EXTENT(eh) + le16_to_cpu(eh->eh_entries) - 1;
+       r = EXT_LAST_EXTENT(eh);
 
        while (l <= r) {
                m = l + (r - l) / 2;
@@ -418,8 +449,9 @@ ext4_ext_binsearch(struct inode *inode, struct ext4_ext_path *path, int block)
                        r = m - 1;
                else
                        l = m + 1;
-               ext_debug("%p(%u):%p(%u):%p(%u) ", l, l->ee_block,
-                               m, m->ee_block, r, r->ee_block);
+               ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
+                               m, le32_to_cpu(m->ee_block),
+                               r, le32_to_cpu(r->ee_block));
        }
 
        path->p_ext = l - 1;
@@ -469,11 +501,10 @@ ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
        short int depth, i, ppos = 0, alloc = 0;
 
        eh = ext_inode_hdr(inode);
-       BUG_ON(eh == NULL);
-       if (ext4_ext_check_header(__FUNCTION__, inode, eh))
+       depth = ext_depth(inode);
+       if (ext4_ext_check_header(inode, eh, depth))
                return ERR_PTR(-EIO);
 
-       i = depth = ext_depth(inode);
 
        /* account possible depth increase */
        if (!path) {
@@ -485,10 +516,12 @@ ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
        }
        path[0].p_hdr = eh;
 
+       i = depth;
        /* walk through the tree */
        while (i) {
                ext_debug("depth %d: num %d, max %d\n",
                          ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
+
                ext4_ext_binsearch_idx(inode, path + ppos, block);
                path[ppos].p_block = idx_pblock(path[ppos].p_idx);
                path[ppos].p_depth = i;
@@ -505,7 +538,7 @@ ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
                path[ppos].p_hdr = eh;
                i--;
 
-               if (ext4_ext_check_header(__FUNCTION__, inode, eh))
+               if (ext4_ext_check_header(inode, eh, i))
                        goto err;
        }
 
@@ -514,9 +547,6 @@ ext4_ext_find_extent(struct inode *inode, int block, struct ext4_ext_path *path)
        path[ppos].p_ext = NULL;
        path[ppos].p_idx = NULL;
 
-       if (ext4_ext_check_header(__FUNCTION__, inode, eh))
-               goto err;
-
        /* find extent */
        ext4_ext_binsearch(inode, path + ppos, block);
 
@@ -554,7 +584,7 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
                if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
                        len = (len - 1) * sizeof(struct ext4_extent_idx);
                        len = len < 0 ? 0 : len;
-                       ext_debug("insert new index %d after: %d. "
+                       ext_debug("insert new index %d after: %llu. "
                                        "move %d from 0x%p to 0x%p\n",
                                        logical, ptr, len,
                                        (curp->p_idx + 1), (curp->p_idx + 2));
@@ -565,7 +595,7 @@ static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
                /* insert before */
                len = len * sizeof(struct ext4_extent_idx);
                len = len < 0 ? 0 : len;
-               ext_debug("insert new index %d before: %d. "
+               ext_debug("insert new index %d before: %llu. "
                                "move %d from 0x%p to 0x%p\n",
                                logical, ptr, len,
                                curp->p_idx, (curp->p_idx + 1));
@@ -765,7 +795,7 @@ static int ext4_ext_split(handle_t *handle, struct inode *inode,
                BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
                                EXT_LAST_INDEX(path[i].p_hdr));
                while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
-                       ext_debug("%d: move %d:%d in new index %llu\n", i,
+                       ext_debug("%d: move %d:%llu in new index %llu\n", i,
                                        le32_to_cpu(path[i].p_idx->ei_block),
                                        idx_pblock(path[i].p_idx),
                                        newblock);
@@ -894,8 +924,13 @@ static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
        curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
        curp->p_hdr->eh_entries = cpu_to_le16(1);
        curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
-       /* FIXME: it works, but actually path[0] can be index */
-       curp->p_idx->ei_block = EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
+
+       if (path[0].p_hdr->eh_depth)
+               curp->p_idx->ei_block =
+                       EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
+       else
+               curp->p_idx->ei_block =
+                       EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
        ext4_idx_store_pblock(curp->p_idx, newblock);
 
        neh = ext_inode_hdr(inode);
@@ -1107,7 +1142,7 @@ static int
 ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
                                struct ext4_extent *ex2)
 {
-       unsigned short ext1_ee_len, ext2_ee_len;
+       unsigned short ext1_ee_len, ext2_ee_len, max_len;
 
        /*
         * Make sure that either both extents are uninitialized, or
@@ -1116,6 +1151,11 @@ ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
        if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
                return 0;
 
+       if (ext4_ext_is_uninitialized(ex1))
+               max_len = EXT_UNINIT_MAX_LEN;
+       else
+               max_len = EXT_INIT_MAX_LEN;
+
        ext1_ee_len = ext4_ext_get_actual_len(ex1);
        ext2_ee_len = ext4_ext_get_actual_len(ex2);
 
@@ -1128,7 +1168,7 @@ ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
         * as an RO_COMPAT feature, refuse to merge to extents if
         * this can result in the top bit of ee_len being set.
         */
-       if (ext1_ee_len + ext2_ee_len > EXT_MAX_LEN)
+       if (ext1_ee_len + ext2_ee_len > max_len)
                return 0;
 #ifdef AGGRESSIVE_TEST
        if (le16_to_cpu(ex1->ee_len) >= 4)
@@ -1140,6 +1180,53 @@ ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
        return 0;
 }
 
+/*
+ * This function tries to merge the "ex" extent to the next extent in the tree.
+ * It always tries to merge towards right. If you want to merge towards
+ * left, pass "ex - 1" as argument instead of "ex".
+ * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
+ * 1 if they got merged.
+ */
+int ext4_ext_try_to_merge(struct inode *inode,
+                         struct ext4_ext_path *path,
+                         struct ext4_extent *ex)
+{
+       struct ext4_extent_header *eh;
+       unsigned int depth, len;
+       int merge_done = 0;
+       int uninitialized = 0;
+
+       depth = ext_depth(inode);
+       BUG_ON(path[depth].p_hdr == NULL);
+       eh = path[depth].p_hdr;
+
+       while (ex < EXT_LAST_EXTENT(eh)) {
+               if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
+                       break;
+               /* merge with next extent! */
+               if (ext4_ext_is_uninitialized(ex))
+                       uninitialized = 1;
+               ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
+                               + ext4_ext_get_actual_len(ex + 1));
+               if (uninitialized)
+                       ext4_ext_mark_uninitialized(ex);
+
+               if (ex + 1 < EXT_LAST_EXTENT(eh)) {
+                       len = (EXT_LAST_EXTENT(eh) - ex - 1)
+                               * sizeof(struct ext4_extent);
+                       memmove(ex + 1, ex + 2, len);
+               }
+               eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries) - 1);
+               merge_done = 1;
+               WARN_ON(eh->eh_entries == 0);
+               if (!eh->eh_entries)
+                       ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
+                          "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
+       }
+
+       return merge_done;
+}
+
 /*
  * check if a portion of the "newext" extent overlaps with an
  * existing extent.
@@ -1322,31 +1409,12 @@ has_space:
        eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)+1);
        nearex = path[depth].p_ext;
        nearex->ee_block = newext->ee_block;
-       nearex->ee_start = newext->ee_start;
-       nearex->ee_start_hi = newext->ee_start_hi;
+       ext4_ext_store_pblock(nearex, ext_pblock(newext));
        nearex->ee_len = newext->ee_len;
 
 merge:
        /* try to merge extents to the right */
-       while (nearex < EXT_LAST_EXTENT(eh)) {
-               if (!ext4_can_extents_be_merged(inode, nearex, nearex + 1))
-                       break;
-               /* merge with next extent! */
-               if (ext4_ext_is_uninitialized(nearex))
-                       uninitialized = 1;
-               nearex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(nearex)
-                                       + ext4_ext_get_actual_len(nearex + 1));
-               if (uninitialized)
-                       ext4_ext_mark_uninitialized(nearex);
-
-               if (nearex + 1 < EXT_LAST_EXTENT(eh)) {
-                       len = (EXT_LAST_EXTENT(eh) - nearex - 1)
-                                       * sizeof(struct ext4_extent);
-                       memmove(nearex + 1, nearex + 2, len);
-               }
-               eh->eh_entries = cpu_to_le16(le16_to_cpu(eh->eh_entries)-1);
-               BUG_ON(eh->eh_entries == 0);
-       }
+       ext4_ext_try_to_merge(inode, path, nearex);
 
        /* try to merge extents to the left */
 
@@ -1475,7 +1543,7 @@ int ext4_ext_walk_space(struct inode *inode, unsigned long block,
 
 static void
 ext4_ext_put_in_cache(struct inode *inode, __u32 block,
-                       __u32 len, __u32 start, int type)
+                       __u32 len, ext4_fsblk_t start, int type)
 {
        struct ext4_ext_cache *cex;
        BUG_ON(len == 0);
@@ -1704,13 +1772,12 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
        unsigned uninitialized = 0;
        struct ext4_extent *ex;
 
+       /* the header must be checked already in ext4_ext_remove_space() */
        ext_debug("truncate since %lu in leaf\n", start);
        if (!path[depth].p_hdr)
                path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
        eh = path[depth].p_hdr;
        BUG_ON(eh == NULL);
-       BUG_ON(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max));
-       BUG_ON(eh->eh_magic != EXT4_EXT_MAGIC);
 
        /* find where to start removing */
        ex = EXT_LAST_EXTENT(eh);
@@ -1786,7 +1853,11 @@ ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
 
                ex->ee_block = cpu_to_le32(block);
                ex->ee_len = cpu_to_le16(num);
-               if (uninitialized)
+               /*
+                * Do not mark uninitialized if all the blocks in the
+                * extent have been removed.
+                */
+               if (uninitialized && num)
                        ext4_ext_mark_uninitialized(ex);
 
                err = ext4_ext_dirty(handle, inode, path + depth);
@@ -1860,7 +1931,7 @@ int ext4_ext_remove_space(struct inode *inode, unsigned long start)
                return -ENOMEM;
        }
        path[0].p_hdr = ext_inode_hdr(inode);
-       if (ext4_ext_check_header(__FUNCTION__, inode, path[0].p_hdr)) {
+       if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
                err = -EIO;
                goto out;
        }
@@ -1881,17 +1952,8 @@ int ext4_ext_remove_space(struct inode *inode, unsigned long start)
                if (!path[i].p_hdr) {
                        ext_debug("initialize header\n");
                        path[i].p_hdr = ext_block_hdr(path[i].p_bh);
-                       if (ext4_ext_check_header(__FUNCTION__, inode,
-                                                       path[i].p_hdr)) {
-                               err = -EIO;
-                               goto out;
-                       }
                }
 
-               BUG_ON(le16_to_cpu(path[i].p_hdr->eh_entries)
-                          > le16_to_cpu(path[i].p_hdr->eh_max));
-               BUG_ON(path[i].p_hdr->eh_magic != EXT4_EXT_MAGIC);
-
                if (!path[i].p_idx) {
                        /* this level hasn't been touched yet */
                        path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
@@ -1908,17 +1970,27 @@ int ext4_ext_remove_space(struct inode *inode, unsigned long start)
                                i, EXT_FIRST_INDEX(path[i].p_hdr),
                                path[i].p_idx);
                if (ext4_ext_more_to_rm(path + i)) {
+                       struct buffer_head *bh;
                        /* go to the next level */
                        ext_debug("move to level %d (block %llu)\n",
                                  i + 1, idx_pblock(path[i].p_idx));
                        memset(path + i + 1, 0, sizeof(*path));
-                       path[i+1].p_bh =
-                               sb_bread(sb, idx_pblock(path[i].p_idx));
-                       if (!path[i+1].p_bh) {
+                       bh = sb_bread(sb, idx_pblock(path[i].p_idx));
+                       if (!bh) {
                                /* should we reset i_size? */
                                err = -EIO;
                                break;
                        }
+                       if (WARN_ON(i + 1 > depth)) {
+                               err = -EIO;
+                               break;
+                       }
+                       if (ext4_ext_check_header(inode, ext_block_hdr(bh),
+                                                       depth - i - 1)) {
+                               err = -EIO;
+                               break;
+                       }
+                       path[i + 1].p_bh = bh;
 
                        /* save actual number of indexes since this
                         * number is changed at the next iteration */
@@ -2012,15 +2084,157 @@ void ext4_ext_release(struct super_block *sb)
 #endif
 }
 
+/*
+ * This function is called by ext4_ext_get_blocks() if someone tries to write
+ * to an uninitialized extent. It may result in splitting the uninitialized
+ * extent into multiple extents (upto three - one initialized and two
+ * uninitialized).
+ * There are three possibilities:
+ *   a> There is no split required: Entire extent should be initialized
+ *   b> Splits in two extents: Write is happening at either end of the extent
+ *   c> Splits in three extents: Somone is writing in middle of the extent
+ */
+int ext4_ext_convert_to_initialized(handle_t *handle, struct inode *inode,
+                                       struct ext4_ext_path *path,
+                                       ext4_fsblk_t iblock,
+                                       unsigned long max_blocks)
+{
+       struct ext4_extent *ex, newex;
+       struct ext4_extent *ex1 = NULL;
+       struct ext4_extent *ex2 = NULL;
+       struct ext4_extent *ex3 = NULL;
+       struct ext4_extent_header *eh;
+       unsigned int allocated, ee_block, ee_len, depth;
+       ext4_fsblk_t newblock;
+       int err = 0;
+       int ret = 0;
+
+       depth = ext_depth(inode);
+       eh = path[depth].p_hdr;
+       ex = path[depth].p_ext;
+       ee_block = le32_to_cpu(ex->ee_block);
+       ee_len = ext4_ext_get_actual_len(ex);
+       allocated = ee_len - (iblock - ee_block);
+       newblock = iblock - ee_block + ext_pblock(ex);
+       ex2 = ex;
+
+       /* ex1: ee_block to iblock - 1 : uninitialized */
+       if (iblock > ee_block) {
+               ex1 = ex;
+               ex1->ee_len = cpu_to_le16(iblock - ee_block);
+               ext4_ext_mark_uninitialized(ex1);
+               ex2 = &newex;
+       }
+       /*
+        * for sanity, update the length of the ex2 extent before
+        * we insert ex3, if ex1 is NULL. This is to avoid temporary
+        * overlap of blocks.
+        */
+       if (!ex1 && allocated > max_blocks)
+               ex2->ee_len = cpu_to_le16(max_blocks);
+       /* ex3: to ee_block + ee_len : uninitialised */
+       if (allocated > max_blocks) {
+               unsigned int newdepth;
+               ex3 = &newex;
+               ex3->ee_block = cpu_to_le32(iblock + max_blocks);
+               ext4_ext_store_pblock(ex3, newblock + max_blocks);
+               ex3->ee_len = cpu_to_le16(allocated - max_blocks);
+               ext4_ext_mark_uninitialized(ex3);
+               err = ext4_ext_insert_extent(handle, inode, path, ex3);
+               if (err)
+                       goto out;
+               /*
+                * The depth, and hence eh & ex might change
+                * as part of the insert above.
+                */
+               newdepth = ext_depth(inode);
+               if (newdepth != depth) {
+                       depth = newdepth;
+                       path = ext4_ext_find_extent(inode, iblock, NULL);
+                       if (IS_ERR(path)) {
+                               err = PTR_ERR(path);
+                               path = NULL;
+                               goto out;
+                       }
+                       eh = path[depth].p_hdr;
+                       ex = path[depth].p_ext;
+                       if (ex2 != &newex)
+                               ex2 = ex;
+               }
+               allocated = max_blocks;
+       }
+       /*
+        * If there was a change of depth as part of the
+        * insertion of ex3 above, we need to update the length
+        * of the ex1 extent again here
+        */
+       if (ex1 && ex1 != ex) {
+               ex1 = ex;
+               ex1->ee_len = cpu_to_le16(iblock - ee_block);
+               ext4_ext_mark_uninitialized(ex1);
+               ex2 = &newex;
+       }
+       /* ex2: iblock to iblock + maxblocks-1 : initialised */
+       ex2->ee_block = cpu_to_le32(iblock);
+       ext4_ext_store_pblock(ex2, newblock);
+       ex2->ee_len = cpu_to_le16(allocated);
+       if (ex2 != ex)
+               goto insert;
+       err = ext4_ext_get_access(handle, inode, path + depth);
+       if (err)
+               goto out;
+       /*
+        * New (initialized) extent starts from the first block
+        * in the current extent. i.e., ex2 == ex
+        * We have to see if it can be merged with the extent
+        * on the left.
+        */
+       if (ex2 > EXT_FIRST_EXTENT(eh)) {
+               /*
+                * To merge left, pass "ex2 - 1" to try_to_merge(),
+                * since it merges towards right _only_.
+                */
+               ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
+               if (ret) {
+                       err = ext4_ext_correct_indexes(handle, inode, path);
+                       if (err)
+                               goto out;
+                       depth = ext_depth(inode);
+                       ex2--;
+               }
+       }
+       /*
+        * Try to Merge towards right. This might be required
+        * only when the whole extent is being written to.
+        * i.e. ex2 == ex and ex3 == NULL.
+        */
+       if (!ex3) {
+               ret = ext4_ext_try_to_merge(inode, path, ex2);
+               if (ret) {
+                       err = ext4_ext_correct_indexes(handle, inode, path);
+                       if (err)
+                               goto out;
+               }
+       }
+       /* Mark modified extent as dirty */
+       err = ext4_ext_dirty(handle, inode, path + depth);
+       goto out;
+insert:
+       err = ext4_ext_insert_extent(handle, inode, path, &newex);
+out:
+       return err ? err : allocated;
+}
+
 int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
                        ext4_fsblk_t iblock,
                        unsigned long max_blocks, struct buffer_head *bh_result,
                        int create, int extend_disksize)
 {
        struct ext4_ext_path *path = NULL;
+       struct ext4_extent_header *eh;
        struct ext4_extent newex, *ex;
        ext4_fsblk_t goal, newblock;
-       int err = 0, depth;
+       int err = 0, depth, ret;
        unsigned long allocated = 0;
 
        __clear_bit(BH_New, &bh_result->b_state);
@@ -2033,8 +2247,10 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
        if (goal) {
                if (goal == EXT4_EXT_CACHE_GAP) {
                        if (!create) {
-                               /* block isn't allocated yet and
-                                * user doesn't want to allocate it */
+                               /*
+                                * block isn't allocated yet and
+                                * user doesn't want to allocate it
+                                */
                                goto out2;
                        }
                        /* we should allocate requested block */
@@ -2068,6 +2284,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
         * this is why assert can't be put in ext4_ext_find_extent()
         */
        BUG_ON(path[depth].p_ext == NULL && depth != 0);
+       eh = path[depth].p_hdr;
 
        ex = path[depth].p_ext;
        if (ex) {
@@ -2076,13 +2293,9 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
                unsigned short ee_len;
 
                /*
-                * Allow future support for preallocated extents to be added
-                * as an RO_COMPAT feature:
                 * Uninitialized extents are treated as holes, except that
-                * we avoid (fail) allocating new blocks during a write.
+                * we split out initialized portions during a write.
                 */
-               if (le16_to_cpu(ex->ee_len) > EXT_MAX_LEN)
-                       goto out2;
                ee_len = ext4_ext_get_actual_len(ex);
                /* if found extent covers block, simply return it */
                if (iblock >= ee_block && iblock < ee_block + ee_len) {
@@ -2091,12 +2304,27 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
                        allocated = ee_len - (iblock - ee_block);
                        ext_debug("%d fit into %lu:%d -> %llu\n", (int) iblock,
                                        ee_block, ee_len, newblock);
+
                        /* Do not put uninitialized extent in the cache */
-                       if (!ext4_ext_is_uninitialized(ex))
+                       if (!ext4_ext_is_uninitialized(ex)) {
                                ext4_ext_put_in_cache(inode, ee_block,
                                                        ee_len, ee_start,
                                                        EXT4_EXT_CACHE_EXTENT);
-                       goto out;
+                               goto out;
+                       }
+                       if (create == EXT4_CREATE_UNINITIALIZED_EXT)
+                               goto out;
+                       if (!create)
+                               goto out2;
+
+                       ret = ext4_ext_convert_to_initialized(handle, inode,
+                                                               path, iblock,
+                                                               max_blocks);
+                       if (ret <= 0)
+                               goto out2;
+                       else
+                               allocated = ret;
+                       goto outnew;
                }
        }
 
@@ -2105,8 +2333,10 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
         * we couldn't try to create block if create flag is zero
         */
        if (!create) {
-               /* put just found gap into cache to speed up
-                * subsequent requests */
+               /*
+                * put just found gap into cache to speed up
+                * subsequent requests
+                */
                ext4_ext_put_gap_in_cache(inode, path, iblock);
                goto out2;
        }
@@ -2120,6 +2350,19 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
        /* allocate new block */
        goal = ext4_ext_find_goal(inode, path, iblock);
 
+       /*
+        * See if request is beyond maximum number of blocks we can have in
+        * a single extent. For an initialized extent this limit is
+        * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
+        * EXT_UNINIT_MAX_LEN.
+        */
+       if (max_blocks > EXT_INIT_MAX_LEN &&
+           create != EXT4_CREATE_UNINITIALIZED_EXT)
+               max_blocks = EXT_INIT_MAX_LEN;
+       else if (max_blocks > EXT_UNINIT_MAX_LEN &&
+                create == EXT4_CREATE_UNINITIALIZED_EXT)
+               max_blocks = EXT_UNINIT_MAX_LEN;
+
        /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
        newex.ee_block = cpu_to_le32(iblock);
        newex.ee_len = cpu_to_le16(max_blocks);
@@ -2152,6 +2395,7 @@ int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
 
        /* previous routine could use block we allocated */
        newblock = ext_pblock(&newex);
+outnew:
        __set_bit(BH_New, &bh_result->b_state);
 
        /* Cache only when it is _not_ an uninitialized extent */
@@ -2221,7 +2465,8 @@ void ext4_ext_truncate(struct inode * inode, struct page *page)
        err = ext4_ext_remove_space(inode, last_block);
 
        /* In a multi-transaction truncate, we only make the final
-        * transaction synchronous. */
+        * transaction synchronous.
+        */
        if (IS_SYNC(inode))
                handle->h_sync = 1;