ext4: Add BUG_ON debugging checks to noalloc_get_block_write()
[safe/jmp/linux-2.6] / fs / ext4 / inode.c
index 8dd3d5d..e6113c3 100644 (file)
@@ -371,6 +371,34 @@ static int ext4_block_to_path(struct inode *inode,
        return n;
 }
 
+static int __ext4_check_blockref(const char *function, struct inode *inode,
+                                __le32 *p, unsigned int max) {
+
+       unsigned int maxblocks = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es);
+       __le32 *bref = p;
+       while (bref < p+max) {
+               if (unlikely(le32_to_cpu(*bref) >= maxblocks)) {
+                       ext4_error(inode->i_sb, function,
+                                  "block reference %u >= max (%u) "
+                                  "in inode #%lu, offset=%d",
+                                  le32_to_cpu(*bref), maxblocks,
+                                  inode->i_ino, (int)(bref-p));
+                       return -EIO;
+               }
+               bref++;
+       }
+       return 0;
+}
+
+
+#define ext4_check_indirect_blockref(inode, bh)                         \
+        __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data,  \
+                             EXT4_ADDR_PER_BLOCK((inode)->i_sb))
+
+#define ext4_check_inode_blockref(inode)                                \
+        __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data,   \
+                             EXT4_NDIR_BLOCKS)
+
 /**
  *     ext4_get_branch - read the chain of indirect blocks leading to data
  *     @inode: inode in question
@@ -415,9 +443,22 @@ static Indirect *ext4_get_branch(struct inode *inode, int depth,
        if (!p->key)
                goto no_block;
        while (--depth) {
-               bh = sb_bread(sb, le32_to_cpu(p->key));
-               if (!bh)
+               bh = sb_getblk(sb, le32_to_cpu(p->key));
+               if (unlikely(!bh))
                        goto failure;
+                  
+               if (!bh_uptodate_or_lock(bh)) {
+                       if (bh_submit_read(bh) < 0) {
+                               put_bh(bh);
+                               goto failure;
+                       }
+                       /* validate block references */
+                       if (ext4_check_indirect_blockref(inode, bh)) {
+                               put_bh(bh);
+                               goto failure;
+                       }
+               }
+               
                add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
                /* Reader: end */
                if (!p->key)
@@ -851,6 +892,10 @@ err_out:
 }
 
 /*
+ * The ext4_ind_get_blocks() function handles non-extents inodes
+ * (i.e., using the traditional indirect/double-indirect i_blocks
+ * scheme) for ext4_get_blocks().
+ *
  * Allocation strategy is simple: if we have to allocate something, we will
  * have to go the whole way to leaf. So let's do it before attaching anything
  * to tree, set linkage between the newborn blocks, write them if sync is
@@ -868,15 +913,16 @@ err_out:
  * return = 0, if plain lookup failed.
  * return < 0, error case.
  *
- *
- * Need to be called with
- * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
- * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
+ * The ext4_ind_get_blocks() function should be called with
+ * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
+ * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
+ * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
+ * blocks.
  */
-static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
+static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
                                  ext4_lblk_t iblock, unsigned int maxblocks,
                                  struct buffer_head *bh_result,
-                                 int create, int extend_disksize)
+                                 int flags)
 {
        int err = -EIO;
        ext4_lblk_t offsets[4];
@@ -893,7 +939,7 @@ static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
 
 
        J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
-       J_ASSERT(handle != NULL || create == 0);
+       J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
        depth = ext4_block_to_path(inode, iblock, offsets,
                                        &blocks_to_boundary);
 
@@ -922,7 +968,7 @@ static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
        }
 
        /* Next simple case - plain lookup or failed read of indirect block */
-       if (!create || err == -EIO)
+       if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
                goto cleanup;
 
        /*
@@ -961,7 +1007,7 @@ static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
         * protect it if you're about to implement concurrent
         * ext4_get_block() -bzzz
        */
-       if (!err && extend_disksize) {
+       if (!err && (flags & EXT4_GET_BLOCKS_EXTEND_DISKSIZE)) {
                disksize = ((loff_t) iblock + count) << inode->i_blkbits;
                if (disksize > i_size_read(inode))
                        disksize = i_size_read(inode);
@@ -1067,13 +1113,20 @@ static void ext4_da_update_reserve_space(struct inode *inode, int used)
        /*
         * free those over-booking quota for metadata blocks
         */
-
        if (mdb_free)
                vfs_dq_release_reservation_block(inode, mdb_free);
+
+       /*
+        * If we have done all the pending block allocations and if
+        * there aren't any writers on the inode, we can discard the
+        * inode's preallocations.
+        */
+       if (!total && (atomic_read(&inode->i_writecount) == 0))
+               ext4_discard_preallocations(inode);
 }
 
 /*
- * The ext4_get_blocks_wrap() function try to look up the requested blocks,
+ * The ext4_get_blocks() function tries to look up the requested blocks,
  * and returns if the blocks are already mapped.
  *
  * Otherwise it takes the write lock of the i_data_sem and allocate blocks
@@ -1081,7 +1134,7 @@ static void ext4_da_update_reserve_space(struct inode *inode, int used)
  * mapped.
  *
  * If file type is extents based, it will call ext4_ext_get_blocks(),
- * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
+ * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
  * based files
  *
  * On success, it returns the number of blocks being mapped or allocate.
@@ -1094,30 +1147,31 @@ static void ext4_da_update_reserve_space(struct inode *inode, int used)
  *
  * It returns the error in case of allocation failure.
  */
-int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
-                       unsigned int max_blocks, struct buffer_head *bh,
-                       int create, int extend_disksize, int flag)
+int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
+                   unsigned int max_blocks, struct buffer_head *bh,
+                   int flags)
 {
        int retval;
 
        clear_buffer_mapped(bh);
+       clear_buffer_unwritten(bh);
 
        /*
-        * Try to see if we can get  the block without requesting
-        * for new file system block.
+        * Try to see if we can get the block without requesting a new
+        * file system block.
         */
        down_read((&EXT4_I(inode)->i_data_sem));
        if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
                retval =  ext4_ext_get_blocks(handle, inode, block, max_blocks,
-                               bh, 0, 0);
+                               bh, 0);
        } else {
-               retval = ext4_get_blocks_handle(handle,
-                               inode, block, max_blocks, bh, 0, 0);
+               retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
+                                            bh, 0);
        }
        up_read((&EXT4_I(inode)->i_data_sem));
 
        /* If it is only a block(s) look up */
-       if (!create)
+       if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
                return retval;
 
        /*
@@ -1131,6 +1185,18 @@ int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
                return retval;
 
        /*
+        * When we call get_blocks without the create flag, the
+        * BH_Unwritten flag could have gotten set if the blocks
+        * requested were part of a uninitialized extent.  We need to
+        * clear this flag now that we are committed to convert all or
+        * part of the uninitialized extent to be an initialized
+        * extent.  This is because we need to avoid the combination
+        * of BH_Unwritten and BH_Mapped flags being simultaneously
+        * set on the buffer_head.
+        */
+       clear_buffer_unwritten(bh);
+
+       /*
         * New blocks allocate and/or writing to uninitialized extent
         * will possibly result in updating i_data, so we take
         * the write lock of i_data_sem, and call get_blocks()
@@ -1144,7 +1210,7 @@ int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
         * let the underlying get_block() function know to
         * avoid double accounting
         */
-       if (flag)
+       if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
                EXT4_I(inode)->i_delalloc_reserved_flag = 1;
        /*
         * We need to check for EXT4 here because migrate
@@ -1152,10 +1218,10 @@ int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
         */
        if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
                retval =  ext4_ext_get_blocks(handle, inode, block, max_blocks,
-                               bh, create, extend_disksize);
+                                             bh, flags);
        } else {
-               retval = ext4_get_blocks_handle(handle, inode, block,
-                               max_blocks, bh, create, extend_disksize);
+               retval = ext4_ind_get_blocks(handle, inode, block,
+                                            max_blocks, bh, flags);
 
                if (retval > 0 && buffer_new(bh)) {
                        /*
@@ -1168,7 +1234,7 @@ int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
                }
        }
 
-       if (flag) {
+       if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
                EXT4_I(inode)->i_delalloc_reserved_flag = 0;
                /*
                 * Update reserved blocks/metadata blocks
@@ -1207,8 +1273,8 @@ int ext4_get_block(struct inode *inode, sector_t iblock,
                started = 1;
        }
 
-       ret = ext4_get_blocks_wrap(handle, inode, iblock,
-                                       max_blocks, bh_result, create, 0, 0);
+       ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
+                             create ? EXT4_GET_BLOCKS_CREATE : 0);
        if (ret > 0) {
                bh_result->b_size = (ret << inode->i_blkbits);
                ret = 0;
@@ -1227,17 +1293,19 @@ struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
 {
        struct buffer_head dummy;
        int fatal = 0, err;
+       int flags = EXT4_GET_BLOCKS_EXTEND_DISKSIZE;
 
        J_ASSERT(handle != NULL || create == 0);
 
        dummy.b_state = 0;
        dummy.b_blocknr = -1000;
        buffer_trace_init(&dummy.b_history);
-       err = ext4_get_blocks_wrap(handle, inode, block, 1,
-                                       &dummy, create, 1, 0);
+       if (create)
+               flags |= EXT4_GET_BLOCKS_CREATE;
+       err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
        /*
-        * ext4_get_blocks_handle() returns number of blocks
-        * mapped. 0 in case of a HOLE.
+        * ext4_get_blocks() returns number of blocks mapped. 0 in
+        * case of a HOLE.
         */
        if (err > 0) {
                if (err > 1)
@@ -1791,7 +1859,7 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd)
  * @logical - first logical block to start assignment with
  *
  * the function goes through all passed space and put actual disk
- * block numbers into buffer heads, dropping BH_Delay
+ * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
  */
 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
                                 struct buffer_head *exbh)
@@ -1841,16 +1909,24 @@ static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
                        do {
                                if (cur_logical >= logical + blocks)
                                        break;
-                               if (buffer_delay(bh)) {
-                                       bh->b_blocknr = pblock;
-                                       clear_buffer_delay(bh);
-                                       bh->b_bdev = inode->i_sb->s_bdev;
-                               } else if (buffer_unwritten(bh)) {
-                                       bh->b_blocknr = pblock;
-                                       clear_buffer_unwritten(bh);
-                                       set_buffer_mapped(bh);
-                                       set_buffer_new(bh);
-                                       bh->b_bdev = inode->i_sb->s_bdev;
+
+                               if (buffer_delay(bh) ||
+                                               buffer_unwritten(bh)) {
+
+                                       BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
+
+                                       if (buffer_delay(bh)) {
+                                               clear_buffer_delay(bh);
+                                               bh->b_blocknr = pblock;
+                                       } else {
+                                               /*
+                                                * unwritten already should have
+                                                * blocknr assigned. Verify that
+                                                */
+                                               clear_buffer_unwritten(bh);
+                                               BUG_ON(bh->b_blocknr != pblock);
+                                       }
+
                                } else if (buffer_mapped(bh))
                                        BUG_ON(bh->b_blocknr != pblock);
 
@@ -1929,9 +2005,14 @@ static void ext4_print_free_blocks(struct inode *inode)
        return;
 }
 
-#define                EXT4_DELALLOC_RSVED     1
+/*
+ * This function is used by mpage_da_map_blocks().  We separate it out
+ * as a separate function just to make life easier, and because
+ * mpage_da_map_blocks() used to be a generic function that took a
+ * get_block_t.
+ */
 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
-                                  struct buffer_head *bh_result, int create)
+                                  struct buffer_head *bh_result)
 {
        int ret;
        unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
@@ -1940,8 +2021,9 @@ static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
 
        handle = ext4_journal_current_handle();
        BUG_ON(!handle);
-       ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
-                                  bh_result, create, 0, EXT4_DELALLOC_RSVED);
+       ret = ext4_get_blocks(handle, inode, iblock, max_blocks,
+                             bh_result, EXT4_GET_BLOCKS_CREATE|
+                             EXT4_GET_BLOCKS_DELALLOC_RESERVE);
        if (ret <= 0)
                return ret;
 
@@ -1960,8 +2042,8 @@ static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
 
        /*
         * Update on-disk size along with block allocation we don't
-        * use 'extend_disksize' as size may change within already
-        * allocated block -bzzz
+        * use EXT4_GET_BLOCKS_EXTEND_DISKSIZE as size may change
+        * within already allocated block -bzzz
         */
        disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
        if (disksize > i_size_read(inode))
@@ -1992,9 +2074,24 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
         * We consider only non-mapped and non-allocated blocks
         */
        if ((mpd->b_state  & (1 << BH_Mapped)) &&
-           !(mpd->b_state & (1 << BH_Delay)))
+               !(mpd->b_state & (1 << BH_Delay)) &&
+               !(mpd->b_state & (1 << BH_Unwritten)))
                return 0;
-       new.b_state = mpd->b_state;
+       /*
+        * We need to make sure the BH_Delay flag is passed down to
+        * ext4_da_get_block_write(), since it calls ext4_get_blocks()
+        * with the EXT4_GET_BLOCKS_DELALLOC_RESERVE flag.  This flag
+        * causes ext4_get_blocks() to call
+        * ext4_da_update_reserve_space() if the passed buffer head
+        * has the BH_Delay flag set.  In the future, once we clean up
+        * the interfaces to ext4_get_blocks(), we should pass in a
+        * separate flag which requests that the delayed allocation
+        * statistics should be updated, instead of depending on the
+        * state information getting passed down via the map_bh's
+        * state bitmasks plus the magic
+        * EXT4_GET_BLOCKS_DELALLOC_RESERVE flag.
+        */
+       new.b_state = mpd->b_state & (1 << BH_Delay);
        new.b_blocknr = 0;
        new.b_size = mpd->b_size;
        next = mpd->b_blocknr;
@@ -2005,7 +2102,7 @@ static int mpage_da_map_blocks(struct mpage_da_data *mpd)
        if (!new.b_size)
                return 0;
 
-       err = ext4_da_get_block_write(mpd->inode, next, &new, 1);
+       err = ext4_da_get_block_write(mpd->inode, next, &new);
        if (err) {
                /*
                 * If get block returns with error we simply
@@ -2131,6 +2228,17 @@ flush_it:
        return;
 }
 
+static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
+{
+       /*
+        * unmapped buffer is possible for holes.
+        * delay buffer is possible with delayed allocation.
+        * We also need to consider unwritten buffer as unmapped.
+        */
+       return (!buffer_mapped(bh) || buffer_delay(bh) ||
+                               buffer_unwritten(bh)) && buffer_dirty(bh);
+}
+
 /*
  * __mpage_da_writepage - finds extent of pages and blocks
  *
@@ -2215,8 +2323,7 @@ static int __mpage_da_writepage(struct page *page,
                         * Otherwise we won't make progress
                         * with the page in ext4_da_writepage
                         */
-                       if (buffer_dirty(bh) &&
-                           (!buffer_mapped(bh) || buffer_delay(bh))) {
+                       if (ext4_bh_unmapped_or_delay(NULL, bh)) {
                                mpage_add_bh_to_extent(mpd, logical,
                                                       bh->b_size,
                                                       bh->b_state);
@@ -2242,13 +2349,25 @@ static int __mpage_da_writepage(struct page *page,
 }
 
 /*
- * this is a special callback for ->write_begin() only
- * it's intention is to return mapped block or reserve space
+ * This is a special get_blocks_t callback which is used by
+ * ext4_da_write_begin().  It will either return mapped block or
+ * reserve space for a single block.
+ *
+ * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
+ * We also have b_blocknr = -1 and b_bdev initialized properly
+ *
+ * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
+ * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
+ * initialized properly.
  */
 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
                                  struct buffer_head *bh_result, int create)
 {
        int ret = 0;
+       sector_t invalid_block = ~((sector_t) 0xffff);
+
+       if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
+               invalid_block = ~0;
 
        BUG_ON(create == 0);
        BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
@@ -2258,7 +2377,7 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
         * preallocated blocks are unmapped but should treated
         * the same as allocated blocks.
         */
-       ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1,  bh_result, 0, 0, 0);
+       ret = ext4_get_blocks(NULL, inode, iblock, 1,  bh_result, 0);
        if ((ret == 0) && !buffer_delay(bh_result)) {
                /* the block isn't (pre)allocated yet, let's reserve space */
                /*
@@ -2270,38 +2389,58 @@ static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
                        /* not enough space to reserve */
                        return ret;
 
-               map_bh(bh_result, inode->i_sb, 0);
+               map_bh(bh_result, inode->i_sb, invalid_block);
                set_buffer_new(bh_result);
                set_buffer_delay(bh_result);
        } else if (ret > 0) {
                bh_result->b_size = (ret << inode->i_blkbits);
+               if (buffer_unwritten(bh_result)) {
+                       /* A delayed write to unwritten bh should
+                        * be marked new and mapped.  Mapped ensures
+                        * that we don't do get_block multiple times
+                        * when we write to the same offset and new
+                        * ensures that we do proper zero out for
+                        * partial write.
+                        */
+                       set_buffer_new(bh_result);
+                       set_buffer_mapped(bh_result);
+               }
                ret = 0;
        }
 
        return ret;
 }
 
-static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
-{
-       /*
-        * unmapped buffer is possible for holes.
-        * delay buffer is possible with delayed allocation
-        */
-       return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
-}
-
-static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
+/*
+ * This function is used as a standard get_block_t calback function
+ * when there is no desire to allocate any blocks.  It is used as a
+ * callback function for block_prepare_write(), nobh_writepage(), and
+ * block_write_full_page().  These functions should only try to map a
+ * single block at a time.
+ *
+ * Since this function doesn't do block allocations even if the caller
+ * requests it by passing in create=1, it is critically important that
+ * any caller checks to make sure that any buffer heads are returned
+ * by this function are either all already mapped or marked for
+ * delayed allocation before calling nobh_writepage() or
+ * block_write_full_page().  Otherwise, b_blocknr could be left
+ * unitialized, and the page write functions will be taken by
+ * surprise.
+ */
+static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
                                   struct buffer_head *bh_result, int create)
 {
        int ret = 0;
        unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
 
+       BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
+
        /*
         * we don't want to do block allocation in writepage
         * so call get_block_wrap with create = 0
         */
-       ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
-                                  bh_result, 0, 0, 0);
+       ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
+       BUG_ON(create && ret == 0);
        if (ret > 0) {
                bh_result->b_size = (ret << inode->i_blkbits);
                ret = 0;
@@ -2310,10 +2449,11 @@ static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
 }
 
 /*
- * get called vi ext4_da_writepages after taking page lock (have journal handle)
- * get called via journal_submit_inode_data_buffers (no journal handle)
- * get called via shrink_page_list via pdflush (no journal handle)
- * or grab_page_cache when doing write_begin (have journal handle)
+ * This function can get called via...
+ *   - ext4_da_writepages after taking page lock (have journal handle)
+ *   - journal_submit_inode_data_buffers (no journal handle)
+ *   - shrink_page_list via pdflush (no journal handle)
+ *   - grab_page_cache when doing write_begin (have journal handle)
  */
 static int ext4_da_writepage(struct page *page,
                                struct writeback_control *wbc)
@@ -2364,7 +2504,7 @@ static int ext4_da_writepage(struct page *page,
                 * do block allocation here.
                 */
                ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
-                                               ext4_normal_get_block_write);
+                                         noalloc_get_block_write);
                if (!ret) {
                        page_bufs = page_buffers(page);
                        /* check whether all are mapped and non delay */
@@ -2389,11 +2529,10 @@ static int ext4_da_writepage(struct page *page,
        }
 
        if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
-               ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
+               ret = nobh_writepage(page, noalloc_get_block_write, wbc);
        else
-               ret = block_write_full_page(page,
-                                               ext4_normal_get_block_write,
-                                               wbc);
+               ret = block_write_full_page(page, noalloc_get_block_write,
+                                           wbc);
 
        return ret;
 }
@@ -2705,7 +2844,7 @@ retry:
        *pagep = page;
 
        ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
-                                                       ext4_da_get_block_prep);
+                               ext4_da_get_block_prep);
        if (ret < 0) {
                unlock_page(page);
                ext4_journal_stop(handle);
@@ -2743,7 +2882,7 @@ static int ext4_da_should_update_i_disksize(struct page *page,
        for (i = 0; i < idx; i++)
                bh = bh->b_this_page;
 
-       if (!buffer_mapped(bh) || (buffer_delay(bh)))
+       if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
                return 0;
        return 1;
 }
@@ -3013,12 +3152,10 @@ static int __ext4_normal_writepage(struct page *page,
        struct inode *inode = page->mapping->host;
 
        if (test_opt(inode->i_sb, NOBH))
-               return nobh_writepage(page,
-                                       ext4_normal_get_block_write, wbc);
+               return nobh_writepage(page, noalloc_get_block_write, wbc);
        else
-               return block_write_full_page(page,
-                                               ext4_normal_get_block_write,
-                                               wbc);
+               return block_write_full_page(page, noalloc_get_block_write,
+                                            wbc);
 }
 
 static int ext4_normal_writepage(struct page *page,
@@ -3070,7 +3207,7 @@ static int __ext4_journalled_writepage(struct page *page,
        int err;
 
        ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
-                                       ext4_normal_get_block_write);
+                                 noalloc_get_block_write);
        if (ret != 0)
                goto out_unlock;
 
@@ -3155,9 +3292,8 @@ static int ext4_journalled_writepage(struct page *page,
                 * really know unless we go poke around in the buffer_heads.
                 * But block_write_full_page will do the right thing.
                 */
-               return block_write_full_page(page,
-                                               ext4_normal_get_block_write,
-                                               wbc);
+               return block_write_full_page(page, noalloc_get_block_write,
+                                            wbc);
        }
 no_write:
        redirty_page_for_writepage(wbc, page);
@@ -3901,6 +4037,9 @@ void ext4_truncate(struct inode *inode)
        if (!ext4_can_truncate(inode))
                return;
 
+       if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
+               ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
+
        if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
                ext4_ext_truncate(inode);
                return;
@@ -4143,12 +4282,7 @@ make_io:
                        unsigned num;
 
                        table = ext4_inode_table(sb, gdp);
-                       /* Make sure s_inode_readahead_blks is a power of 2 */
-                       while (EXT4_SB(sb)->s_inode_readahead_blks &
-                              (EXT4_SB(sb)->s_inode_readahead_blks-1))
-                               EXT4_SB(sb)->s_inode_readahead_blks = 
-                                  (EXT4_SB(sb)->s_inode_readahead_blks &
-                                   (EXT4_SB(sb)->s_inode_readahead_blks-1));
+                       /* s_inode_readahead_blks is always a power of 2 */
                        b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
                        if (table > b)
                                b = table;
@@ -4311,11 +4445,9 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
        ei->i_flags = le32_to_cpu(raw_inode->i_flags);
        inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
        ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
-       if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
-           cpu_to_le32(EXT4_OS_HURD)) {
+       if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
                ei->i_file_acl |=
                        ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
-       }
        inode->i_size = ext4_isize(raw_inode);
        ei->i_disksize = inode->i_size;
        inode->i_generation = le32_to_cpu(raw_inode->i_generation);
@@ -4363,14 +4495,32 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
                        (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
        }
 
-       if (ei->i_flags & EXT4_EXTENTS_FL) {
-               /* Validate extent which is part of inode */
-               ret = ext4_ext_check_inode(inode);
-               if (ret) {
-                       brelse(bh);
-                       goto bad_inode;
-               }
-
+       ret = 0;
+       if (ei->i_file_acl &&
+           ((ei->i_file_acl < 
+             (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
+              EXT4_SB(sb)->s_gdb_count)) ||
+            (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
+               ext4_error(sb, __func__,
+                          "bad extended attribute block %llu in inode #%lu",
+                          ei->i_file_acl, inode->i_ino);
+               ret = -EIO;
+               goto bad_inode;
+       } else if (ei->i_flags & EXT4_EXTENTS_FL) {
+               if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
+                   (S_ISLNK(inode->i_mode) &&
+                    !ext4_inode_is_fast_symlink(inode)))
+                       /* Validate extent which is part of inode */
+                       ret = ext4_ext_check_inode(inode);
+       } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
+                  (S_ISLNK(inode->i_mode) &&
+                   !ext4_inode_is_fast_symlink(inode))) {
+               /* Validate block references which are part of inode */
+               ret = ext4_check_inode_blockref(inode);
+       }
+       if (ret) {
+               brelse(bh);
+               goto bad_inode;
        }
 
        if (S_ISREG(inode->i_mode)) {
@@ -4389,7 +4539,8 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
                        inode->i_op = &ext4_symlink_inode_operations;
                        ext4_set_aops(inode);
                }
-       } else {
+       } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
+             S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
                inode->i_op = &ext4_special_inode_operations;
                if (raw_inode->i_block[0])
                        init_special_inode(inode, inode->i_mode,
@@ -4397,6 +4548,13 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
                else
                        init_special_inode(inode, inode->i_mode,
                           new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
+       } else {
+               brelse(bh);
+               ret = -EIO;
+               ext4_error(inode->i_sb, __func__, 
+                          "bogus i_mode (%o) for inode=%lu",
+                          inode->i_mode, inode->i_ino);
+               goto bad_inode;
        }
        brelse(iloc.bh);
        ext4_set_inode_flags(inode);
@@ -4836,7 +4994,8 @@ static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
  */
 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 {
-       int groups, gdpblocks;
+       ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
+       int gdpblocks;
        int idxblocks;
        int ret = 0;
 
@@ -4863,8 +5022,8 @@ int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
                groups += nrblocks;
 
        gdpblocks = groups;
-       if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
-               groups = EXT4_SB(inode->i_sb)->s_groups_count;
+       if (groups > ngroups)
+               groups = ngroups;
        if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
                gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
 
@@ -4904,7 +5063,7 @@ int ext4_writepage_trans_blocks(struct inode *inode)
  * Calculate the journal credits for a chunk of data modification.
  *
  * This is called from DIO, fallocate or whoever calling
- * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
+ * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
  *
  * journal buffers for data blocks are not included here, as DIO
  * and fallocate do no need to journal data buffers.
@@ -5190,8 +5349,9 @@ static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
        return !buffer_mapped(bh);
 }
 
-int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
+int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
+       struct page *page = vmf->page;
        loff_t size;
        unsigned long len;
        int ret = -EINVAL;
@@ -5243,6 +5403,8 @@ int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
                goto out_unlock;
        ret = 0;
 out_unlock:
+       if (ret)
+               ret = VM_FAULT_SIGBUS;
        up_read(&inode->i_alloc_sem);
        return ret;
 }