nfsd4: remove some dprintk's
[safe/jmp/linux-2.6] / fs / ext4 / inode.c
index b567e71..c7fed5b 100644 (file)
@@ -34,6 +34,7 @@
 #include <linux/writeback.h>
 #include <linux/pagevec.h>
 #include <linux/mpage.h>
+#include <linux/namei.h>
 #include <linux/uio.h>
 #include <linux/bio.h>
 #include "ext4_jbd2.h"
 static inline int ext4_begin_ordered_truncate(struct inode *inode,
                                              loff_t new_size)
 {
-       return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode)->jinode,
-                                                  new_size);
+       return jbd2_journal_begin_ordered_truncate(
+                                       EXT4_SB(inode->i_sb)->s_journal,
+                                       &EXT4_I(inode)->jinode,
+                                       new_size);
 }
 
 static void ext4_invalidatepage(struct page *page, unsigned long offset);
@@ -71,12 +74,17 @@ static int ext4_inode_is_fast_symlink(struct inode *inode)
  * "bh" may be NULL: a metadata block may have been freed from memory
  * but there may still be a record of it in the journal, and that record
  * still needs to be revoked.
+ *
+ * If the handle isn't valid we're not journaling so there's nothing to do.
  */
 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
                        struct buffer_head *bh, ext4_fsblk_t blocknr)
 {
        int err;
 
+       if (!ext4_handle_valid(handle))
+               return 0;
+
        might_sleep();
 
        BUFFER_TRACE(bh, "enter");
@@ -169,7 +177,9 @@ static handle_t *start_transaction(struct inode *inode)
  */
 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
 {
-       if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
+       if (!ext4_handle_valid(handle))
+               return 0;
+       if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
                return 0;
        if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
                return 0;
@@ -183,6 +193,7 @@ static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
  */
 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
 {
+       BUG_ON(EXT4_JOURNAL(inode) == NULL);
        jbd_debug(2, "restarting handle %p\n", handle);
        return ext4_journal_restart(handle, blocks_for_truncate(inode));
 }
@@ -215,7 +226,7 @@ void ext4_delete_inode(struct inode *inode)
        }
 
        if (IS_SYNC(inode))
-               handle->h_sync = 1;
+               ext4_handle_sync(handle);
        inode->i_size = 0;
        err = ext4_mark_inode_dirty(handle, inode);
        if (err) {
@@ -232,7 +243,7 @@ void ext4_delete_inode(struct inode *inode)
         * enough credits left in the handle to remove the inode from
         * the orphan list and set the dtime field.
         */
-       if (handle->h_buffer_credits < 3) {
+       if (!ext4_handle_has_enough_credits(handle, 3)) {
                err = ext4_journal_extend(handle, 3);
                if (err > 0)
                        err = ext4_journal_restart(handle, 3);
@@ -351,9 +362,9 @@ static int ext4_block_to_path(struct inode *inode,
                final = ptrs;
        } else {
                ext4_warning(inode->i_sb, "ext4_block_to_path",
-                               "block %lu > max",
+                               "block %lu > max in inode %lu",
                                i_block + direct_blocks +
-                               indirect_blocks + double_blocks);
+                               indirect_blocks + double_blocks, inode->i_ino);
        }
        if (boundary)
                *boundary = final - 1 - (i_block & (ptrs - 1));
@@ -486,18 +497,9 @@ static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
                Indirect *partial)
 {
-       struct ext4_block_alloc_info *block_i;
-
-       block_i =  EXT4_I(inode)->i_block_alloc_info;
-
        /*
-        * try the heuristic for sequential allocation,
-        * failing that at least try to get decent locality.
+        * XXX need to get goal block from mballoc's data structures
         */
-       if (block_i && (block == block_i->last_alloc_logical_block + 1)
-               && (block_i->last_alloc_physical_block != 0)) {
-               return block_i->last_alloc_physical_block + 1;
-       }
 
        return ext4_find_near(inode, partial);
 }
@@ -514,10 +516,10 @@ static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
  *     return the total number of blocks to be allocate, including the
  *     direct and indirect blocks.
  */
-static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
+static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
                int blocks_to_boundary)
 {
-       unsigned long count = 0;
+       unsigned int count = 0;
 
        /*
         * Simple case, [t,d]Indirect block(s) has not allocated yet
@@ -555,6 +557,7 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
                                int indirect_blks, int blks,
                                ext4_fsblk_t new_blocks[4], int *err)
 {
+       struct ext4_allocation_request ar;
        int target, i;
        unsigned long count = 0, blk_allocated = 0;
        int index = 0;
@@ -603,10 +606,17 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
        if (!target)
                goto allocated;
        /* Now allocate data blocks */
-       count = target;
-       /* allocating blocks for data blocks */
-       current_block = ext4_new_blocks(handle, inode, iblock,
-                                               goal, &count, err);
+       memset(&ar, 0, sizeof(ar));
+       ar.inode = inode;
+       ar.goal = goal;
+       ar.len = target;
+       ar.logical = iblock;
+       if (S_ISREG(inode->i_mode))
+               /* enable in-core preallocation only for regular files */
+               ar.flags = EXT4_MB_HINT_DATA;
+
+       current_block = ext4_mb_new_blocks(handle, &ar, err);
+
        if (*err && (target == blks)) {
                /*
                 * if the allocation failed and we didn't allocate
@@ -622,7 +632,7 @@ static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
                 */
                        new_blocks[index] = current_block;
                }
-               blk_allocated += count;
+               blk_allocated += ar.len;
        }
 allocated:
        /* total number of blocks allocated for direct blocks */
@@ -717,8 +727,8 @@ static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
                set_buffer_uptodate(bh);
                unlock_buffer(bh);
 
-               BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
-               err = ext4_journal_dirty_metadata(handle, bh);
+               BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
+               err = ext4_handle_dirty_metadata(handle, inode, bh);
                if (err)
                        goto failed;
        }
@@ -757,10 +767,8 @@ static int ext4_splice_branch(handle_t *handle, struct inode *inode,
 {
        int i;
        int err = 0;
-       struct ext4_block_alloc_info *block_i;
        ext4_fsblk_t current_block;
 
-       block_i = EXT4_I(inode)->i_block_alloc_info;
        /*
         * If we're splicing into a [td]indirect block (as opposed to the
         * inode) then we need to get write access to the [td]indirect block
@@ -786,17 +794,6 @@ static int ext4_splice_branch(handle_t *handle, struct inode *inode,
                        *(where->p + i) = cpu_to_le32(current_block++);
        }
 
-       /*
-        * update the most recently allocated logical & physical block
-        * in i_block_alloc_info, to assist find the proper goal block for next
-        * allocation
-        */
-       if (block_i) {
-               block_i->last_alloc_logical_block = block + blks - 1;
-               block_i->last_alloc_physical_block =
-                               le32_to_cpu(where[num].key) + blks - 1;
-       }
-
        /* We are done with atomic stuff, now do the rest of housekeeping */
 
        inode->i_ctime = ext4_current_time(inode);
@@ -813,8 +810,8 @@ static int ext4_splice_branch(handle_t *handle, struct inode *inode,
                 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
                 */
                jbd_debug(5, "splicing indirect only\n");
-               BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
-               err = ext4_journal_dirty_metadata(handle, where->bh);
+               BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
+               err = ext4_handle_dirty_metadata(handle, inode, where->bh);
                if (err)
                        goto err_out;
        } else {
@@ -861,10 +858,10 @@ err_out:
  * 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)
  */
-int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
-               ext4_lblk_t iblock, unsigned long maxblocks,
-               struct buffer_head *bh_result,
-               int create, int extend_disksize)
+static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
+                                 ext4_lblk_t iblock, unsigned int maxblocks,
+                                 struct buffer_head *bh_result,
+                                 int create, int extend_disksize)
 {
        int err = -EIO;
        ext4_lblk_t offsets[4];
@@ -914,12 +911,8 @@ int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
                goto cleanup;
 
        /*
-        * Okay, we need to do block allocation.  Lazily initialize the block
-        * allocation info here if necessary
+        * Okay, we need to do block allocation.
        */
-       if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
-               ext4_init_block_alloc_info(inode);
-
        goal = ext4_find_goal(inode, iblock, partial);
 
        /* the number of blocks need to allocate for [d,t]indirect blocks */
@@ -1070,7 +1063,7 @@ 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 long max_blocks, struct buffer_head *bh,
+                       unsigned int max_blocks, struct buffer_head *bh,
                        int create, int extend_disksize, int flag)
 {
        int retval;
@@ -1161,8 +1154,8 @@ int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
 /* Maximum number of blocks we map for direct IO at once. */
 #define DIO_MAX_BLOCKS 4096
 
-static int ext4_get_block(struct inode *inode, sector_t iblock,
-                       struct buffer_head *bh_result, int create)
+int ext4_get_block(struct inode *inode, sector_t iblock,
+                  struct buffer_head *bh_result, int create)
 {
        handle_t *handle = ext4_journal_current_handle();
        int ret = 0, started = 0;
@@ -1246,8 +1239,8 @@ struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
                                set_buffer_uptodate(bh);
                        }
                        unlock_buffer(bh);
-                       BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
-                       err = ext4_journal_dirty_metadata(handle, bh);
+                       BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
+                       err = ext4_handle_dirty_metadata(handle, inode, bh);
                        if (!fatal)
                                fatal = err;
                } else {
@@ -1360,6 +1353,10 @@ static int ext4_write_begin(struct file *file, struct address_space *mapping,
        pgoff_t index;
        unsigned from, to;
 
+       trace_mark(ext4_write_begin,
+                  "dev %s ino %lu pos %llu len %u flags %u",
+                  inode->i_sb->s_id, inode->i_ino,
+                  (unsigned long long) pos, len, flags);
        index = pos >> PAGE_CACHE_SHIFT;
        from = pos & (PAGE_CACHE_SIZE - 1);
        to = from + len;
@@ -1371,7 +1368,11 @@ retry:
                goto out;
        }
 
-       page = __grab_cache_page(mapping, index);
+       /* We cannot recurse into the filesystem as the transaction is already
+        * started */
+       flags |= AOP_FLAG_NOFS;
+
+       page = grab_cache_page_write_begin(mapping, index, flags);
        if (!page) {
                ext4_journal_stop(handle);
                ret = -ENOMEM;
@@ -1380,7 +1381,7 @@ retry:
        *pagep = page;
 
        ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
-                                                       ext4_get_block);
+                               ext4_get_block);
 
        if (!ret && ext4_should_journal_data(inode)) {
                ret = walk_page_buffers(handle, page_buffers(page),
@@ -1391,6 +1392,13 @@ retry:
                unlock_page(page);
                ext4_journal_stop(handle);
                page_cache_release(page);
+               /*
+                * block_write_begin may have instantiated a few blocks
+                * outside i_size.  Trim these off again. Don't need
+                * i_size_read because we hold i_mutex.
+                */
+               if (pos + len > inode->i_size)
+                       vmtruncate(inode, inode->i_size);
        }
 
        if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
@@ -1405,7 +1413,7 @@ static int write_end_fn(handle_t *handle, struct buffer_head *bh)
        if (!buffer_mapped(bh) || buffer_freed(bh))
                return 0;
        set_buffer_uptodate(bh);
-       return ext4_journal_dirty_metadata(handle, bh);
+       return ext4_handle_dirty_metadata(handle, NULL, bh);
 }
 
 /*
@@ -1424,19 +1432,25 @@ static int ext4_ordered_write_end(struct file *file,
        struct inode *inode = mapping->host;
        int ret = 0, ret2;
 
+       trace_mark(ext4_ordered_write_end,
+                  "dev %s ino %lu pos %llu len %u copied %u",
+                  inode->i_sb->s_id, inode->i_ino,
+                  (unsigned long long) pos, len, copied);
        ret = ext4_jbd2_file_inode(handle, inode);
 
        if (ret == 0) {
-               /*
-                * generic_write_end() will run mark_inode_dirty() if i_size
-                * changes.  So let's piggyback the i_disksize mark_inode_dirty
-                * into that.
-                */
                loff_t new_i_size;
 
                new_i_size = pos + copied;
-               if (new_i_size > EXT4_I(inode)->i_disksize)
-                       EXT4_I(inode)->i_disksize = new_i_size;
+               if (new_i_size > EXT4_I(inode)->i_disksize) {
+                       ext4_update_i_disksize(inode, new_i_size);
+                       /* We need to mark inode dirty even if
+                        * new_i_size is less that inode->i_size
+                        * bu greater than i_disksize.(hint delalloc)
+                        */
+                       ext4_mark_inode_dirty(handle, inode);
+               }
+
                ret2 = generic_write_end(file, mapping, pos, len, copied,
                                                        page, fsdata);
                copied = ret2;
@@ -1460,9 +1474,19 @@ static int ext4_writeback_write_end(struct file *file,
        int ret = 0, ret2;
        loff_t new_i_size;
 
+       trace_mark(ext4_writeback_write_end,
+                  "dev %s ino %lu pos %llu len %u copied %u",
+                  inode->i_sb->s_id, inode->i_ino,
+                  (unsigned long long) pos, len, copied);
        new_i_size = pos + copied;
-       if (new_i_size > EXT4_I(inode)->i_disksize)
-               EXT4_I(inode)->i_disksize = new_i_size;
+       if (new_i_size > EXT4_I(inode)->i_disksize) {
+               ext4_update_i_disksize(inode, new_i_size);
+               /* We need to mark inode dirty even if
+                * new_i_size is less that inode->i_size
+                * bu greater than i_disksize.(hint delalloc)
+                */
+               ext4_mark_inode_dirty(handle, inode);
+       }
 
        ret2 = generic_write_end(file, mapping, pos, len, copied,
                                                        page, fsdata);
@@ -1487,7 +1511,12 @@ static int ext4_journalled_write_end(struct file *file,
        int ret = 0, ret2;
        int partial = 0;
        unsigned from, to;
+       loff_t new_i_size;
 
+       trace_mark(ext4_journalled_write_end,
+                  "dev %s ino %lu pos %llu len %u copied %u",
+                  inode->i_sb->s_id, inode->i_ino,
+                  (unsigned long long) pos, len, copied);
        from = pos & (PAGE_CACHE_SIZE - 1);
        to = from + len;
 
@@ -1501,11 +1530,12 @@ static int ext4_journalled_write_end(struct file *file,
                                to, &partial, write_end_fn);
        if (!partial)
                SetPageUptodate(page);
-       if (pos+copied > inode->i_size)
+       new_i_size = pos + copied;
+       if (new_i_size > inode->i_size)
                i_size_write(inode, pos+copied);
        EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
-       if (inode->i_size > EXT4_I(inode)->i_disksize) {
-               EXT4_I(inode)->i_disksize = inode->i_size;
+       if (new_i_size > EXT4_I(inode)->i_disksize) {
+               ext4_update_i_disksize(inode, new_i_size);
                ret2 = ext4_mark_inode_dirty(handle, inode);
                if (!ret)
                        ret = ret2;
@@ -1633,7 +1663,8 @@ struct mpage_da_data {
        get_block_t *get_block;
        struct writeback_control *wbc;
        int io_done;
-       long pages_written;
+       int pages_written;
+       int retval;
 };
 
 /*
@@ -1652,18 +1683,25 @@ struct mpage_da_data {
  */
 static int mpage_da_submit_io(struct mpage_da_data *mpd)
 {
-       struct address_space *mapping = mpd->inode->i_mapping;
-       int ret = 0, err, nr_pages, i;
-       unsigned long index, end;
+       long pages_skipped;
        struct pagevec pvec;
+       unsigned long index, end;
+       int ret = 0, err, nr_pages, i;
+       struct inode *inode = mpd->inode;
+       struct address_space *mapping = inode->i_mapping;
 
        BUG_ON(mpd->next_page <= mpd->first_page);
-       pagevec_init(&pvec, 0);
+       /*
+        * We need to start from the first_page to the next_page - 1
+        * to make sure we also write the mapped dirty buffer_heads.
+        * If we look at mpd->lbh.b_blocknr we would only be looking
+        * at the currently mapped buffer_heads.
+        */
        index = mpd->first_page;
        end = mpd->next_page - 1;
 
+       pagevec_init(&pvec, 0);
        while (index <= end) {
-               /* XXX: optimize tail */
                nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
                if (nr_pages == 0)
                        break;
@@ -1675,8 +1713,16 @@ static int mpage_da_submit_io(struct mpage_da_data *mpd)
                                break;
                        index++;
 
+                       BUG_ON(!PageLocked(page));
+                       BUG_ON(PageWriteback(page));
+
+                       pages_skipped = mpd->wbc->pages_skipped;
                        err = mapping->a_ops->writepage(page, mpd->wbc);
-                       if (!err)
+                       if (!err && (pages_skipped == mpd->wbc->pages_skipped))
+                               /*
+                                * have successfully written the page
+                                * without skipping the same
+                                */
                                mpd->pages_written++;
                        /*
                         * In error case, we have to continue because
@@ -1820,6 +1866,24 @@ static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
        return;
 }
 
+static void ext4_print_free_blocks(struct inode *inode)
+{
+       struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+       printk(KERN_EMERG "Total free blocks count %lld\n",
+                       ext4_count_free_blocks(inode->i_sb));
+       printk(KERN_EMERG "Free/Dirty block details\n");
+       printk(KERN_EMERG "free_blocks=%lld\n",
+                       (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
+       printk(KERN_EMERG "dirty_blocks=%lld\n",
+                       (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
+       printk(KERN_EMERG "Block reservation details\n");
+       printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
+                       EXT4_I(inode)->i_reserved_data_blocks);
+       printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
+                       EXT4_I(inode)->i_reserved_meta_blocks);
+       return;
+}
+
 /*
  * mpage_da_map_blocks - go through given space
  *
@@ -1834,7 +1898,7 @@ static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
        int err = 0;
        struct buffer_head new;
        struct buffer_head *lbh = &mpd->lbh;
-       sector_t next = lbh->b_blocknr;
+       sector_t next;
 
        /*
         * We consider only non-mapped and non-allocated blocks
@@ -1844,6 +1908,7 @@ static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
        new.b_state = lbh->b_state;
        new.b_blocknr = 0;
        new.b_size = lbh->b_size;
+       next = lbh->b_blocknr;
        /*
         * If we didn't accumulate anything
         * to write simply return
@@ -1860,6 +1925,13 @@ static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
                 */
                if (err == -EAGAIN)
                        return 0;
+
+               if (err == -ENOSPC &&
+                               ext4_count_free_blocks(mpd->inode->i_sb)) {
+                       mpd->retval = err;
+                       return 0;
+               }
+
                /*
                 * get block failure will cause us
                 * to loop in writepages. Because
@@ -1877,8 +1949,7 @@ static int  mpage_da_map_blocks(struct mpage_da_data *mpd)
                printk(KERN_EMERG "This should not happen.!! "
                                        "Data will be lost\n");
                if (err == -ENOSPC) {
-                       printk(KERN_CRIT "Total free blocks count %lld\n",
-                               ext4_count_free_blocks(mpd->inode->i_sb));
+                       ext4_print_free_blocks(mpd->inode);
                }
                /* invlaidate all the pages */
                ext4_da_block_invalidatepages(mpd, next,
@@ -2058,11 +2129,29 @@ static int __mpage_da_writepage(struct page *page,
                bh = head;
                do {
                        BUG_ON(buffer_locked(bh));
+                       /*
+                        * We need to try to allocate
+                        * unmapped blocks in the same page.
+                        * Otherwise we won't make progress
+                        * with the page in ext4_da_writepage
+                        */
                        if (buffer_dirty(bh) &&
                                (!buffer_mapped(bh) || buffer_delay(bh))) {
                                mpage_add_bh_to_extent(mpd, logical, bh);
                                if (mpd->io_done)
                                        return MPAGE_DA_EXTENT_TAIL;
+                       } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
+                               /*
+                                * mapped dirty buffer. We need to update
+                                * the b_state because we look at
+                                * b_state in mpage_da_map_blocks. We don't
+                                * update b_size because if we find an
+                                * unmapped buffer_head later we need to
+                                * use the b_state flag of that buffer_head.
+                                */
+                               if (mpd->lbh.b_size == 0)
+                                       mpd->lbh.b_state =
+                                               bh->b_state & BH_FLAGS;
                        }
                        logical++;
                } while ((bh = bh->b_this_page) != head);
@@ -2085,39 +2174,34 @@ static int __mpage_da_writepage(struct page *page,
  */
 static int mpage_da_writepages(struct address_space *mapping,
                               struct writeback_control *wbc,
-                              get_block_t get_block)
+                              struct mpage_da_data *mpd)
 {
-       struct mpage_da_data mpd;
-       long to_write;
        int ret;
 
-       if (!get_block)
+       if (!mpd->get_block)
                return generic_writepages(mapping, wbc);
 
-       mpd.wbc = wbc;
-       mpd.inode = mapping->host;
-       mpd.lbh.b_size = 0;
-       mpd.lbh.b_state = 0;
-       mpd.lbh.b_blocknr = 0;
-       mpd.first_page = 0;
-       mpd.next_page = 0;
-       mpd.get_block = get_block;
-       mpd.io_done = 0;
-       mpd.pages_written = 0;
-
-       to_write = wbc->nr_to_write;
-
-       ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
+       mpd->lbh.b_size = 0;
+       mpd->lbh.b_state = 0;
+       mpd->lbh.b_blocknr = 0;
+       mpd->first_page = 0;
+       mpd->next_page = 0;
+       mpd->io_done = 0;
+       mpd->pages_written = 0;
+       mpd->retval = 0;
 
+       ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
        /*
         * Handle last extent of pages
         */
-       if (!mpd.io_done && mpd.next_page != mpd.first_page) {
-               if (mpage_da_map_blocks(&mpd) == 0)
-                       mpage_da_submit_io(&mpd);
-       }
+       if (!mpd->io_done && mpd->next_page != mpd->first_page) {
+               if (mpage_da_map_blocks(mpd) == 0)
+                       mpage_da_submit_io(mpd);
 
-       wbc->nr_to_write = to_write - mpd.pages_written;
+               mpd->io_done = 1;
+               ret = MPAGE_DA_EXTENT_TAIL;
+       }
+       wbc->nr_to_write -= mpd->pages_written;
        return ret;
 }
 
@@ -2197,18 +2281,9 @@ static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
                if (disksize > i_size_read(inode))
                        disksize = i_size_read(inode);
                if (disksize > EXT4_I(inode)->i_disksize) {
-                       /*
-                        * XXX: replace with spinlock if seen contended -bzzz
-                        */
-                       down_write(&EXT4_I(inode)->i_data_sem);
-                       if (disksize > EXT4_I(inode)->i_disksize)
-                               EXT4_I(inode)->i_disksize = disksize;
-                       up_write(&EXT4_I(inode)->i_data_sem);
-
-                       if (EXT4_I(inode)->i_disksize == disksize) {
-                               ret = ext4_mark_inode_dirty(handle, inode);
-                               return ret;
-                       }
+                       ext4_update_i_disksize(inode, disksize);
+                       ret = ext4_mark_inode_dirty(handle, inode);
+                       return ret;
                }
                ret = 0;
        }
@@ -2254,10 +2329,13 @@ static int ext4_da_writepage(struct page *page,
 {
        int ret = 0;
        loff_t size;
-       unsigned long len;
+       unsigned int len;
        struct buffer_head *page_bufs;
        struct inode *inode = page->mapping->host;
 
+       trace_mark(ext4_da_writepage,
+                  "dev %s ino %lu page_index %lu",
+                  inode->i_sb->s_id, inode->i_ino, page->index);
        size = i_size_read(inode);
        if (page->index == size >> PAGE_CACHE_SHIFT)
                len = size & ~PAGE_CACHE_MASK;
@@ -2315,6 +2393,8 @@ static int ext4_da_writepage(struct page *page,
                        unlock_page(page);
                        return 0;
                }
+               /* now mark the buffer_heads as dirty and uptodate */
+               block_commit_write(page, 0, PAGE_CACHE_SIZE);
        }
 
        if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
@@ -2355,13 +2435,32 @@ static int ext4_da_writepages_trans_blocks(struct inode *inode)
 static int ext4_da_writepages(struct address_space *mapping,
                              struct writeback_control *wbc)
 {
+       pgoff_t index;
+       int range_whole = 0;
        handle_t *handle = NULL;
-       loff_t range_start = 0;
+       struct mpage_da_data mpd;
        struct inode *inode = mapping->host;
+       int no_nrwrite_index_update;
+       int pages_written = 0;
+       long pages_skipped;
+       int range_cyclic, cycled = 1, io_done = 0;
        int needed_blocks, ret = 0, nr_to_writebump = 0;
-       long to_write, pages_skipped = 0;
        struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
 
+       trace_mark(ext4_da_writepages,
+                  "dev %s ino %lu nr_t_write %ld "
+                  "pages_skipped %ld range_start %llu "
+                  "range_end %llu nonblocking %d "
+                  "for_kupdate %d for_reclaim %d "
+                  "for_writepages %d range_cyclic %d",
+                  inode->i_sb->s_id, inode->i_ino,
+                  wbc->nr_to_write, wbc->pages_skipped,
+                  (unsigned long long) wbc->range_start,
+                  (unsigned long long) wbc->range_end,
+                  wbc->nonblocking, wbc->for_kupdate,
+                  wbc->for_reclaim, wbc->for_writepages,
+                  wbc->range_cyclic);
+
        /*
         * No pages to write? This is mainly a kludge to avoid starting
         * a transaction for special inodes like journal inode on last iput()
@@ -2369,6 +2468,20 @@ static int ext4_da_writepages(struct address_space *mapping,
         */
        if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
                return 0;
+
+       /*
+        * If the filesystem has aborted, it is read-only, so return
+        * right away instead of dumping stack traces later on that
+        * will obscure the real source of the problem.  We test
+        * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
+        * the latter could be true if the filesystem is mounted
+        * read-only, and in that case, ext4_da_writepages should
+        * *never* be called, so if that ever happens, we would want
+        * the stack trace.
+        */
+       if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
+               return -EROFS;
+
        /*
         * Make sure nr_to_write is >= sbi->s_mb_stream_request
         * This make sure small files blocks are allocated in
@@ -2379,20 +2492,33 @@ static int ext4_da_writepages(struct address_space *mapping,
                nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
                wbc->nr_to_write = sbi->s_mb_stream_request;
        }
+       if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
+               range_whole = 1;
+
+       range_cyclic = wbc->range_cyclic;
+       if (wbc->range_cyclic) {
+               index = mapping->writeback_index;
+               if (index)
+                       cycled = 0;
+               wbc->range_start = index << PAGE_CACHE_SHIFT;
+               wbc->range_end  = LLONG_MAX;
+               wbc->range_cyclic = 0;
+       } else
+               index = wbc->range_start >> PAGE_CACHE_SHIFT;
 
-       if (!wbc->range_cyclic)
-               /*
-                * If range_cyclic is not set force range_cont
-                * and save the old writeback_index
-                */
-               wbc->range_cont = 1;
+       mpd.wbc = wbc;
+       mpd.inode = mapping->host;
 
-       range_start =  wbc->range_start;
+       /*
+        * we don't want write_cache_pages to update
+        * nr_to_write and writeback_index
+        */
+       no_nrwrite_index_update = wbc->no_nrwrite_index_update;
+       wbc->no_nrwrite_index_update = 1;
        pages_skipped = wbc->pages_skipped;
 
-restart_loop:
-       to_write = wbc->nr_to_write;
-       while (!ret && to_write > 0) {
+retry:
+       while (!ret && wbc->nr_to_write > 0) {
 
                /*
                 * we  insert one extent at a time. So we need
@@ -2407,48 +2533,76 @@ restart_loop:
                handle = ext4_journal_start(inode, needed_blocks);
                if (IS_ERR(handle)) {
                        ret = PTR_ERR(handle);
-                       printk(KERN_EMERG "%s: jbd2_start: "
+                       printk(KERN_CRIT "%s: jbd2_start: "
                               "%ld pages, ino %lu; err %d\n", __func__,
                                wbc->nr_to_write, inode->i_ino, ret);
                        dump_stack();
                        goto out_writepages;
                }
+               mpd.get_block = ext4_da_get_block_write;
+               ret = mpage_da_writepages(mapping, wbc, &mpd);
 
-               to_write -= wbc->nr_to_write;
-               ret = mpage_da_writepages(mapping, wbc,
-                                         ext4_da_get_block_write);
                ext4_journal_stop(handle);
-               if (ret == MPAGE_DA_EXTENT_TAIL) {
+
+               if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
+                       /* commit the transaction which would
+                        * free blocks released in the transaction
+                        * and try again
+                        */
+                       jbd2_journal_force_commit_nested(sbi->s_journal);
+                       wbc->pages_skipped = pages_skipped;
+                       ret = 0;
+               } else if (ret == MPAGE_DA_EXTENT_TAIL) {
                        /*
                         * got one extent now try with
                         * rest of the pages
                         */
-                       to_write += wbc->nr_to_write;
+                       pages_written += mpd.pages_written;
+                       wbc->pages_skipped = pages_skipped;
                        ret = 0;
-               } else if (wbc->nr_to_write) {
+                       io_done = 1;
+               } else if (wbc->nr_to_write)
                        /*
                         * There is no more writeout needed
                         * or we requested for a noblocking writeout
                         * and we found the device congested
                         */
-                       to_write += wbc->nr_to_write;
                        break;
-               }
-               wbc->nr_to_write = to_write;
        }
-
-       if (wbc->range_cont && (pages_skipped != wbc->pages_skipped)) {
-               /* We skipped pages in this loop */
-               wbc->range_start = range_start;
-               wbc->nr_to_write = to_write +
-                               wbc->pages_skipped - pages_skipped;
-               wbc->pages_skipped = pages_skipped;
-               goto restart_loop;
+       if (!io_done && !cycled) {
+               cycled = 1;
+               index = 0;
+               wbc->range_start = index << PAGE_CACHE_SHIFT;
+               wbc->range_end  = mapping->writeback_index - 1;
+               goto retry;
        }
+       if (pages_skipped != wbc->pages_skipped)
+               printk(KERN_EMERG "This should not happen leaving %s "
+                               "with nr_to_write = %ld ret = %d\n",
+                               __func__, wbc->nr_to_write, ret);
+
+       /* Update index */
+       index += pages_written;
+       wbc->range_cyclic = range_cyclic;
+       if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
+               /*
+                * set the writeback_index so that range_cyclic
+                * mode will write it back later
+                */
+               mapping->writeback_index = index;
 
 out_writepages:
-       wbc->nr_to_write = to_write - nr_to_writebump;
-       wbc->range_start = range_start;
+       if (!no_nrwrite_index_update)
+               wbc->no_nrwrite_index_update = 0;
+       wbc->nr_to_write -= nr_to_writebump;
+       trace_mark(ext4_da_writepage_result,
+                  "dev %s ino %lu ret %d pages_written %d "
+                  "pages_skipped %ld congestion %d "
+                  "more_io %d no_nrwrite_index_update %d",
+                  inode->i_sb->s_id, inode->i_ino, ret,
+                  pages_written, wbc->pages_skipped,
+                  wbc->encountered_congestion, wbc->more_io,
+                  wbc->no_nrwrite_index_update);
        return ret;
 }
 
@@ -2461,7 +2615,7 @@ static int ext4_nonda_switch(struct super_block *sb)
        /*
         * switch to non delalloc mode if we are running low
         * on free block. The free block accounting via percpu
-        * counters can get slightly wrong with FBC_BATCH getting
+        * counters can get slightly wrong with percpu_counter_batch getting
         * accumulated on each CPU without updating global counters
         * Delalloc need an accurate free block accounting. So switch
         * to non delalloc when we are near to error range.
@@ -2500,6 +2654,11 @@ static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
                                        len, flags, pagep, fsdata);
        }
        *fsdata = (void *)0;
+
+       trace_mark(ext4_da_write_begin,
+                  "dev %s ino %lu pos %llu len %u flags %u",
+                  inode->i_sb->s_id, inode->i_ino,
+                  (unsigned long long) pos, len, flags);
 retry:
        /*
         * With delayed allocation, we don't log the i_disksize update
@@ -2512,8 +2671,11 @@ retry:
                ret = PTR_ERR(handle);
                goto out;
        }
+       /* We cannot recurse into the filesystem as the transaction is already
+        * started */
+       flags |= AOP_FLAG_NOFS;
 
-       page = __grab_cache_page(mapping, index);
+       page = grab_cache_page_write_begin(mapping, index, flags);
        if (!page) {
                ext4_journal_stop(handle);
                ret = -ENOMEM;
@@ -2527,6 +2689,13 @@ retry:
                unlock_page(page);
                ext4_journal_stop(handle);
                page_cache_release(page);
+               /*
+                * block_write_begin may have instantiated a few blocks
+                * outside i_size.  Trim these off again. Don't need
+                * i_size_read because we hold i_mutex.
+                */
+               if (pos + len > inode->i_size)
+                       vmtruncate(inode, inode->i_size);
        }
 
        if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
@@ -2582,6 +2751,10 @@ static int ext4_da_write_end(struct file *file,
                }
        }
 
+       trace_mark(ext4_da_write_end,
+                  "dev %s ino %lu pos %llu len %u copied %u",
+                  inode->i_sb->s_id, inode->i_ino,
+                  (unsigned long long) pos, len, copied);
        start = pos & (PAGE_CACHE_SIZE - 1);
        end = start + copied - 1;
 
@@ -2607,6 +2780,11 @@ static int ext4_da_write_end(struct file *file,
                                EXT4_I(inode)->i_disksize = new_i_size;
                        }
                        up_write(&EXT4_I(inode)->i_data_sem);
+                       /* We need to mark inode dirty even if
+                        * new_i_size is less that inode->i_size
+                        * bu greater than i_disksize.(hint delalloc)
+                        */
+                       ext4_mark_inode_dirty(handle, inode);
                }
        }
        ret2 = generic_write_end(file, mapping, pos, len, copied,
@@ -2669,7 +2847,7 @@ static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
                filemap_write_and_wait(mapping);
        }
 
-       if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
+       if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
                /*
                 * This is a REALLY heavyweight approach, but the use of
                 * bmap on dirty files is expected to be extremely rare:
@@ -2787,6 +2965,9 @@ static int ext4_normal_writepage(struct page *page,
        loff_t size = i_size_read(inode);
        loff_t len;
 
+       trace_mark(ext4_normal_writepage,
+                  "dev %s ino %lu page_index %lu",
+                  inode->i_sb->s_id, inode->i_ino, page->index);
        J_ASSERT(PageLocked(page));
        if (page->index == size >> PAGE_CACHE_SHIFT)
                len = size & ~PAGE_CACHE_MASK;
@@ -2872,6 +3053,9 @@ static int ext4_journalled_writepage(struct page *page,
        loff_t size = i_size_read(inode);
        loff_t len;
 
+       trace_mark(ext4_journalled_writepage,
+                  "dev %s ino %lu page_index %lu",
+                  inode->i_sb->s_id, inode->i_ino, page->index);
        J_ASSERT(PageLocked(page));
        if (page->index == size >> PAGE_CACHE_SHIFT)
                len = size & ~PAGE_CACHE_MASK;
@@ -2940,7 +3124,10 @@ static void ext4_invalidatepage(struct page *page, unsigned long offset)
        if (offset == 0)
                ClearPageChecked(page);
 
-       jbd2_journal_invalidatepage(journal, page, offset);
+       if (journal)
+               jbd2_journal_invalidatepage(journal, page, offset);
+       else
+               block_invalidatepage(page, offset);
 }
 
 static int ext4_releasepage(struct page *page, gfp_t wait)
@@ -2950,7 +3137,10 @@ static int ext4_releasepage(struct page *page, gfp_t wait)
        WARN_ON(PageChecked(page));
        if (!page_has_buffers(page))
                return 0;
-       return jbd2_journal_try_to_free_buffers(journal, page, wait);
+       if (journal)
+               return jbd2_journal_try_to_free_buffers(journal, page, wait);
+       else
+               return try_to_free_buffers(page);
 }
 
 /*
@@ -3222,7 +3412,7 @@ int ext4_block_truncate_page(handle_t *handle,
 
        err = 0;
        if (ext4_should_journal_data(inode)) {
-               err = ext4_journal_dirty_metadata(handle, bh);
+               err = ext4_handle_dirty_metadata(handle, inode, bh);
        } else {
                if (ext4_should_order_data(inode))
                        err = ext4_jbd2_file_inode(handle, inode);
@@ -3346,8 +3536,8 @@ static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
        __le32 *p;
        if (try_to_extend_transaction(handle, inode)) {
                if (bh) {
-                       BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
-                       ext4_journal_dirty_metadata(handle, bh);
+                       BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
+                       ext4_handle_dirty_metadata(handle, inode, bh);
                }
                ext4_mark_inode_dirty(handle, inode);
                ext4_journal_test_restart(handle, inode);
@@ -3447,7 +3637,7 @@ static void ext4_free_data(handle_t *handle, struct inode *inode,
                                  count, block_to_free_p, p);
 
        if (this_bh) {
-               BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
+               BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
 
                /*
                 * The buffer head should have an attached journal head at this
@@ -3455,8 +3645,8 @@ static void ext4_free_data(handle_t *handle, struct inode *inode,
                 * block pointed to itself, it would have been detached when
                 * the block was cleared. Check for this instead of OOPSing.
                 */
-               if (bh2jh(this_bh))
-                       ext4_journal_dirty_metadata(handle, this_bh);
+               if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
+                       ext4_handle_dirty_metadata(handle, inode, this_bh);
                else
                        ext4_error(inode->i_sb, __func__,
                                   "circular indirect block detected, "
@@ -3486,7 +3676,7 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
        ext4_fsblk_t nr;
        __le32 *p;
 
-       if (is_handle_aborted(handle))
+       if (ext4_handle_is_aborted(handle))
                return;
 
        if (depth--) {
@@ -3556,7 +3746,7 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
                         * will merely complain about releasing a free block,
                         * rather than leaking blocks.
                         */
-                       if (is_handle_aborted(handle))
+                       if (ext4_handle_is_aborted(handle))
                                return;
                        if (try_to_extend_transaction(handle, inode)) {
                                ext4_mark_inode_dirty(handle, inode);
@@ -3575,9 +3765,10 @@ static void ext4_free_branches(handle_t *handle, struct inode *inode,
                                                                   parent_bh)){
                                        *p = 0;
                                        BUFFER_TRACE(parent_bh,
-                                       "call ext4_journal_dirty_metadata");
-                                       ext4_journal_dirty_metadata(handle,
-                                                                   parent_bh);
+                                       "call ext4_handle_dirty_metadata");
+                                       ext4_handle_dirty_metadata(handle,
+                                                                  inode,
+                                                                  parent_bh);
                                }
                        }
                }
@@ -3685,7 +3876,7 @@ void ext4_truncate(struct inode *inode)
         */
        down_write(&ei->i_data_sem);
 
-       ext4_discard_reservation(inode);
+       ext4_discard_preallocations(inode);
 
        /*
         * The orphan list entry will now protect us from any crash which
@@ -3765,7 +3956,7 @@ do_indirects:
         * synchronous
         */
        if (IS_SYNC(inode))
-               handle->h_sync = 1;
+               ext4_handle_sync(handle);
 out_stop:
        /*
         * If this was a simple ftruncate(), and the file will remain alive
@@ -3780,41 +3971,6 @@ out_stop:
        ext4_journal_stop(handle);
 }
 
-static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
-               unsigned long ino, struct ext4_iloc *iloc)
-{
-       ext4_group_t block_group;
-       unsigned long offset;
-       ext4_fsblk_t block;
-       struct ext4_group_desc *gdp;
-
-       if (!ext4_valid_inum(sb, ino)) {
-               /*
-                * This error is already checked for in namei.c unless we are
-                * looking at an NFS filehandle, in which case no error
-                * report is needed
-                */
-               return 0;
-       }
-
-       block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
-       gdp = ext4_get_group_desc(sb, block_group, NULL);
-       if (!gdp)
-               return 0;
-
-       /*
-        * Figure out the offset within the block group inode table
-        */
-       offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
-               EXT4_INODE_SIZE(sb);
-       block = ext4_inode_table(sb, gdp) +
-               (offset >> EXT4_BLOCK_SIZE_BITS(sb));
-
-       iloc->block_group = block_group;
-       iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
-       return block;
-}
-
 /*
  * ext4_get_inode_loc returns with an extra refcount against the inode's
  * underlying buffer_head on success. If 'in_mem' is true, we have all
@@ -3824,19 +3980,35 @@ static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
 static int __ext4_get_inode_loc(struct inode *inode,
                                struct ext4_iloc *iloc, int in_mem)
 {
-       ext4_fsblk_t block;
-       struct buffer_head *bh;
+       struct ext4_group_desc  *gdp;
+       struct buffer_head      *bh;
+       struct super_block      *sb = inode->i_sb;
+       ext4_fsblk_t            block;
+       int                     inodes_per_block, inode_offset;
+
+       iloc->bh = NULL;
+       if (!ext4_valid_inum(sb, inode->i_ino))
+               return -EIO;
 
-       block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
-       if (!block)
+       iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
+       gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
+       if (!gdp)
                return -EIO;
 
-       bh = sb_getblk(inode->i_sb, block);
+       /*
+        * Figure out the offset within the block group inode table
+        */
+       inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
+       inode_offset = ((inode->i_ino - 1) %
+                       EXT4_INODES_PER_GROUP(sb));
+       block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
+       iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
+
+       bh = sb_getblk(sb, block);
        if (!bh) {
-               ext4_error (inode->i_sb, "ext4_get_inode_loc",
-                               "unable to read inode block - "
-                               "inode=%lu, block=%llu",
-                                inode->i_ino, block);
+               ext4_error(sb, "ext4_get_inode_loc", "unable to read "
+                          "inode block - inode=%lu, block=%llu",
+                          inode->i_ino, block);
                return -EIO;
        }
        if (!buffer_uptodate(bh)) {
@@ -3864,28 +4036,12 @@ static int __ext4_get_inode_loc(struct inode *inode,
                 */
                if (in_mem) {
                        struct buffer_head *bitmap_bh;
-                       struct ext4_group_desc *desc;
-                       int inodes_per_buffer;
-                       int inode_offset, i;
-                       ext4_group_t block_group;
-                       int start;
-
-                       block_group = (inode->i_ino - 1) /
-                                       EXT4_INODES_PER_GROUP(inode->i_sb);
-                       inodes_per_buffer = bh->b_size /
-                               EXT4_INODE_SIZE(inode->i_sb);
-                       inode_offset = ((inode->i_ino - 1) %
-                                       EXT4_INODES_PER_GROUP(inode->i_sb));
-                       start = inode_offset & ~(inodes_per_buffer - 1);
+                       int i, start;
 
-                       /* Is the inode bitmap in cache? */
-                       desc = ext4_get_group_desc(inode->i_sb,
-                                               block_group, NULL);
-                       if (!desc)
-                               goto make_io;
+                       start = inode_offset & ~(inodes_per_block - 1);
 
-                       bitmap_bh = sb_getblk(inode->i_sb,
-                               ext4_inode_bitmap(inode->i_sb, desc));
+                       /* Is the inode bitmap in cache? */
+                       bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
                        if (!bitmap_bh)
                                goto make_io;
 
@@ -3898,14 +4054,14 @@ static int __ext4_get_inode_loc(struct inode *inode,
                                brelse(bitmap_bh);
                                goto make_io;
                        }
-                       for (i = start; i < start + inodes_per_buffer; i++) {
+                       for (i = start; i < start + inodes_per_block; i++) {
                                if (i == inode_offset)
                                        continue;
                                if (ext4_test_bit(i, bitmap_bh->b_data))
                                        break;
                        }
                        brelse(bitmap_bh);
-                       if (i == start + inodes_per_buffer) {
+                       if (i == start + inodes_per_block) {
                                /* all other inodes are free, so skip I/O */
                                memset(bh->b_data, 0, bh->b_size);
                                set_buffer_uptodate(bh);
@@ -3916,6 +4072,36 @@ static int __ext4_get_inode_loc(struct inode *inode,
 
 make_io:
                /*
+                * If we need to do any I/O, try to pre-readahead extra
+                * blocks from the inode table.
+                */
+               if (EXT4_SB(sb)->s_inode_readahead_blks) {
+                       ext4_fsblk_t b, end, table;
+                       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));
+                       b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
+                       if (table > b)
+                               b = table;
+                       end = b + EXT4_SB(sb)->s_inode_readahead_blks;
+                       num = EXT4_INODES_PER_GROUP(sb);
+                       if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
+                                      EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
+                               num -= ext4_itable_unused_count(sb, gdp);
+                       table += num / inodes_per_block;
+                       if (end > table)
+                               end = table;
+                       while (b <= end)
+                               sb_breadahead(sb, b++);
+               }
+
+               /*
                 * There are other valid inodes in the buffer, this inode
                 * has in-inode xattrs, or we don't have this inode in memory.
                 * Read the block from disk.
@@ -3925,10 +4111,9 @@ make_io:
                submit_bh(READ_META, bh);
                wait_on_buffer(bh);
                if (!buffer_uptodate(bh)) {
-                       ext4_error(inode->i_sb, "ext4_get_inode_loc",
-                                       "unable to read inode block - "
-                                       "inode=%lu, block=%llu",
-                                       inode->i_ino, block);
+                       ext4_error(sb, __func__,
+                                  "unable to read inode block - inode=%lu, "
+                                  "block=%llu", inode->i_ino, block);
                        brelse(bh);
                        return -EIO;
                }
@@ -4020,11 +4205,10 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
                return inode;
 
        ei = EXT4_I(inode);
-#ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
+#ifdef CONFIG_EXT4_FS_POSIX_ACL
        ei->i_acl = EXT4_ACL_NOT_CACHED;
        ei->i_default_acl = EXT4_ACL_NOT_CACHED;
 #endif
-       ei->i_block_alloc_info = NULL;
 
        ret = __ext4_get_inode_loc(inode, &iloc, 0);
        if (ret < 0)
@@ -4123,9 +4307,11 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
                inode->i_op = &ext4_dir_inode_operations;
                inode->i_fop = &ext4_dir_operations;
        } else if (S_ISLNK(inode->i_mode)) {
-               if (ext4_inode_is_fast_symlink(inode))
+               if (ext4_inode_is_fast_symlink(inode)) {
                        inode->i_op = &ext4_fast_symlink_inode_operations;
-               else {
+                       nd_terminate_link(ei->i_data, inode->i_size,
+                               sizeof(ei->i_data) - 1);
+               } else {
                        inode->i_op = &ext4_symlink_inode_operations;
                        ext4_set_aops(inode);
                }
@@ -4155,7 +4341,6 @@ static int ext4_inode_blocks_set(handle_t *handle,
        struct inode *inode = &(ei->vfs_inode);
        u64 i_blocks = inode->i_blocks;
        struct super_block *sb = inode->i_sb;
-       int err = 0;
 
        if (i_blocks <= ~0U) {
                /*
@@ -4165,36 +4350,27 @@ static int ext4_inode_blocks_set(handle_t *handle,
                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
                raw_inode->i_blocks_high = 0;
                ei->i_flags &= ~EXT4_HUGE_FILE_FL;
-       } else if (i_blocks <= 0xffffffffffffULL) {
+               return 0;
+       }
+       if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
+               return -EFBIG;
+
+       if (i_blocks <= 0xffffffffffffULL) {
                /*
                 * i_blocks can be represented in a 48 bit variable
                 * as multiple of 512 bytes
                 */
-               err = ext4_update_rocompat_feature(handle, sb,
-                                           EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
-               if (err)
-                       goto  err_out;
-               /* i_block is stored in the split  48 bit fields */
                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
                raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
                ei->i_flags &= ~EXT4_HUGE_FILE_FL;
        } else {
-               /*
-                * i_blocks should be represented in a 48 bit variable
-                * as multiple of  file system block size
-                */
-               err = ext4_update_rocompat_feature(handle, sb,
-                                           EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
-               if (err)
-                       goto  err_out;
                ei->i_flags |= EXT4_HUGE_FILE_FL;
                /* i_block is stored in file system block size */
                i_blocks = i_blocks >> (inode->i_blkbits - 9);
                raw_inode->i_blocks_lo   = cpu_to_le32(i_blocks);
                raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
        }
-err_out:
-       return err;
+       return 0;
 }
 
 /*
@@ -4279,8 +4455,8 @@ static int ext4_do_update_inode(handle_t *handle,
                        EXT4_SET_RO_COMPAT_FEATURE(sb,
                                        EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
                        sb->s_dirt = 1;
-                       handle->h_sync = 1;
-                       err = ext4_journal_dirty_metadata(handle,
+                       ext4_handle_sync(handle);
+                       err = ext4_handle_dirty_metadata(handle, inode,
                                        EXT4_SB(sb)->s_sbh);
                }
        }
@@ -4307,9 +4483,8 @@ static int ext4_do_update_inode(handle_t *handle,
                raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
        }
 
-
-       BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
-       rc = ext4_journal_dirty_metadata(handle, bh);
+       BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
+       rc = ext4_handle_dirty_metadata(handle, inode, bh);
        if (!err)
                err = rc;
        ei->i_state &= ~EXT4_STATE_NEW;
@@ -4372,6 +4547,25 @@ int ext4_write_inode(struct inode *inode, int wait)
        return ext4_force_commit(inode->i_sb);
 }
 
+int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
+{
+       int err = 0;
+
+       mark_buffer_dirty(bh);
+       if (inode && inode_needs_sync(inode)) {
+               sync_dirty_buffer(bh);
+               if (buffer_req(bh) && !buffer_uptodate(bh)) {
+                       ext4_error(inode->i_sb, __func__,
+                                  "IO error syncing inode, "
+                                  "inode=%lu, block=%llu",
+                                  inode->i_ino,
+                                  (unsigned long long)bh->b_blocknr);
+                       err = -EIO;
+               }
+       }
+       return err;
+}
+
 /*
  * ext4_setattr()
  *
@@ -4551,9 +4745,10 @@ static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
 {
        if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
-               return ext4_indirect_trans_blocks(inode, nrblocks, 0);
-       return ext4_ext_index_trans_blocks(inode, nrblocks, 0);
+               return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
+       return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
 }
+
 /*
  * Account for index blocks, block groups bitmaps and block group
  * descriptor blocks if modify datablocks and index blocks
@@ -4675,16 +4870,15 @@ int
 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
                         struct ext4_iloc *iloc)
 {
-       int err = 0;
-       if (handle) {
-               err = ext4_get_inode_loc(inode, iloc);
-               if (!err) {
-                       BUFFER_TRACE(iloc->bh, "get_write_access");
-                       err = ext4_journal_get_write_access(handle, iloc->bh);
-                       if (err) {
-                               brelse(iloc->bh);
-                               iloc->bh = NULL;
-                       }
+       int err;
+
+       err = ext4_get_inode_loc(inode, iloc);
+       if (!err) {
+               BUFFER_TRACE(iloc->bh, "get_write_access");
+               err = ext4_journal_get_write_access(handle, iloc->bh);
+               if (err) {
+                       brelse(iloc->bh);
+                       iloc->bh = NULL;
                }
        }
        ext4_std_error(inode->i_sb, err);
@@ -4756,7 +4950,8 @@ int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
 
        might_sleep();
        err = ext4_reserve_inode_write(handle, inode, &iloc);
-       if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
+       if (ext4_handle_valid(handle) &&
+           EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
            !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
                /*
                 * We need extra buffer credits since we may write into EA block
@@ -4808,6 +5003,11 @@ void ext4_dirty_inode(struct inode *inode)
        handle_t *current_handle = ext4_journal_current_handle();
        handle_t *handle;
 
+       if (!ext4_handle_valid(current_handle)) {
+               ext4_mark_inode_dirty(current_handle, inode);
+               return;
+       }
+
        handle = ext4_journal_start(inode, 2);
        if (IS_ERR(handle))
                goto out;
@@ -4845,8 +5045,9 @@ static int ext4_pin_inode(handle_t *handle, struct inode *inode)
                        BUFFER_TRACE(iloc.bh, "get_write_access");
                        err = jbd2_journal_get_write_access(handle, iloc.bh);
                        if (!err)
-                               err = ext4_journal_dirty_metadata(handle,
-                                                                 iloc.bh);
+                               err = ext4_handle_dirty_metadata(handle,
+                                                                inode,
+                                                                iloc.bh);
                        brelse(iloc.bh);
                }
        }
@@ -4872,6 +5073,8 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
         */
 
        journal = EXT4_JOURNAL(inode);
+       if (!journal)
+               return 0;
        if (is_journal_aborted(journal))
                return -EROFS;
 
@@ -4901,7 +5104,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
                return PTR_ERR(handle);
 
        err = ext4_mark_inode_dirty(handle, inode);
-       handle->h_sync = 1;
+       ext4_handle_sync(handle);
        ext4_journal_stop(handle);
        ext4_std_error(inode->i_sb, err);