switch ext3 to inode->i_acl
[safe/jmp/linux-2.6] / fs / ext3 / inode.c
index 0775354..5f51fed 100644 (file)
@@ -36,6 +36,8 @@
 #include <linux/mpage.h>
 #include <linux/uio.h>
 #include <linux/bio.h>
+#include <linux/fiemap.h>
+#include <linux/namei.h>
 #include "xattr.h"
 #include "acl.h"
 
@@ -95,7 +97,7 @@ int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
        BUFFER_TRACE(bh, "call ext3_journal_revoke");
        err = ext3_journal_revoke(handle, blocknr, bh);
        if (err)
-               ext3_abort(inode->i_sb, __FUNCTION__,
+               ext3_abort(inode->i_sb, __func__,
                           "error %d when attempting revoke", err);
        BUFFER_TRACE(bh, "exit");
        return err;
@@ -392,7 +394,7 @@ no_block:
  *     @inode: owner
  *     @ind: descriptor of indirect block.
  *
- *     This function returns the prefered place for block allocation.
+ *     This function returns the preferred place for block allocation.
  *     It is used when heuristic for sequential allocation fails.
  *     Rules are:
  *       + if there is a block to the left of our position - allocate near it.
@@ -436,19 +438,17 @@ static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
 }
 
 /**
- *     ext3_find_goal - find a prefered place for allocation.
+ *     ext3_find_goal - find a preferred place for allocation.
  *     @inode: owner
  *     @block:  block we want
- *     @chain:  chain of indirect blocks
  *     @partial: pointer to the last triple within a chain
- *     @goal:  place to store the result.
  *
- *     Normally this function find the prefered place for block allocation,
- *     stores it in *@goal and returns zero.
+ *     Normally this function find the preferred place for block allocation,
+ *     returns it.
  */
 
 static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
-               Indirect chain[4], Indirect *partial)
+                                  Indirect *partial)
 {
        struct ext3_block_alloc_info *block_i;
 
@@ -820,7 +820,7 @@ int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
                while (count < maxblocks && count <= blocks_to_boundary) {
                        ext3_fsblk_t blk;
 
-                       if (!verify_chain(chain, partial)) {
+                       if (!verify_chain(chain, chain + depth - 1)) {
                                /*
                                 * Indirect block might be removed by
                                 * truncate while we were reading it.
@@ -884,7 +884,7 @@ int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
        if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
                ext3_init_block_alloc_info(inode);
 
-       goal = ext3_find_goal(inode, iblock, chain, partial);
+       goal = ext3_find_goal(inode, iblock, partial);
 
        /* the number of blocks need to allocate for [d,t]indirect blocks */
        indirect_blks = (chain + depth) - partial - 1;
@@ -941,58 +941,55 @@ out:
        return err;
 }
 
-#define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
+/* Maximum number of blocks we map for direct IO at once. */
+#define DIO_MAX_BLOCKS 4096
+/*
+ * Number of credits we need for writing DIO_MAX_BLOCKS:
+ * We need sb + group descriptor + bitmap + inode -> 4
+ * For B blocks with A block pointers per block we need:
+ * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
+ * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
+ */
+#define DIO_CREDITS 25
 
 static int ext3_get_block(struct inode *inode, sector_t iblock,
                        struct buffer_head *bh_result, int create)
 {
        handle_t *handle = ext3_journal_current_handle();
-       int ret = 0;
+       int ret = 0, started = 0;
        unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
 
-       if (!create)
-               goto get_block;         /* A read */
-
-       if (max_blocks == 1)
-               goto get_block;         /* A single block get */
-
-       if (handle->h_transaction->t_state == T_LOCKED) {
-               /*
-                * Huge direct-io writes can hold off commits for long
-                * periods of time.  Let this commit run.
-                */
-               ext3_journal_stop(handle);
-               handle = ext3_journal_start(inode, DIO_CREDITS);
-               if (IS_ERR(handle))
+       if (create && !handle) {        /* Direct IO write... */
+               if (max_blocks > DIO_MAX_BLOCKS)
+                       max_blocks = DIO_MAX_BLOCKS;
+               handle = ext3_journal_start(inode, DIO_CREDITS +
+                               2 * EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb));
+               if (IS_ERR(handle)) {
                        ret = PTR_ERR(handle);
-               goto get_block;
-       }
-
-       if (handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
-               /*
-                * Getting low on buffer credits...
-                */
-               ret = ext3_journal_extend(handle, DIO_CREDITS);
-               if (ret > 0) {
-                       /*
-                        * Couldn't extend the transaction.  Start a new one.
-                        */
-                       ret = ext3_journal_restart(handle, DIO_CREDITS);
+                       goto out;
                }
+               started = 1;
        }
 
-get_block:
-       if (ret == 0) {
-               ret = ext3_get_blocks_handle(handle, inode, iblock,
+       ret = ext3_get_blocks_handle(handle, inode, iblock,
                                        max_blocks, bh_result, create, 0);
-               if (ret > 0) {
-                       bh_result->b_size = (ret << inode->i_blkbits);
-                       ret = 0;
-               }
+       if (ret > 0) {
+               bh_result->b_size = (ret << inode->i_blkbits);
+               ret = 0;
        }
+       if (started)
+               ext3_journal_stop(handle);
+out:
        return ret;
 }
 
+int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
+               u64 start, u64 len)
+{
+       return generic_block_fiemap(inode, fieinfo, start, len,
+                                   ext3_get_block);
+}
+
 /*
  * `handle' can be NULL if create is zero
  */
@@ -1152,19 +1149,22 @@ static int ext3_write_begin(struct file *file, struct address_space *mapping,
                                struct page **pagep, void **fsdata)
 {
        struct inode *inode = mapping->host;
-       int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
+       int ret;
        handle_t *handle;
        int retries = 0;
        struct page *page;
        pgoff_t index;
        unsigned from, to;
+       /* Reserve one block more for addition to orphan list in case
+        * we allocate blocks but write fails for some reason */
+       int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
 
        index = pos >> PAGE_CACHE_SHIFT;
        from = pos & (PAGE_CACHE_SIZE - 1);
        to = from + len;
 
 retry:
-       page = __grab_cache_page(mapping, index);
+       page = grab_cache_page_write_begin(mapping, index, flags);
        if (!page)
                return -ENOMEM;
        *pagep = page;
@@ -1187,9 +1187,21 @@ retry:
        }
 write_begin_failed:
        if (ret) {
+               /*
+                * 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.
+                *
+                * Add inode to orphan list in case we crash before truncate
+                * finishes.
+                */
+               if (pos + len > inode->i_size)
+                       ext3_orphan_add(handle, inode);
                ext3_journal_stop(handle);
                unlock_page(page);
                page_cache_release(page);
+               if (pos + len > inode->i_size)
+                       vmtruncate(inode, inode->i_size);
        }
        if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
                goto retry;
@@ -1202,11 +1214,23 @@ int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
 {
        int err = journal_dirty_data(handle, bh);
        if (err)
-               ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
+               ext3_journal_abort_handle(__func__, __func__,
                                                bh, handle, err);
        return err;
 }
 
+/* For ordered writepage and write_end functions */
+static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
+{
+       /*
+        * Write could have mapped the buffer but it didn't copy the data in
+        * yet. So avoid filing such buffer into a transaction.
+        */
+       if (buffer_mapped(bh) && buffer_uptodate(bh))
+               return ext3_journal_dirty_data(handle, bh);
+       return 0;
+}
+
 /* For write_end() in data=journal mode */
 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
 {
@@ -1217,26 +1241,20 @@ static int write_end_fn(handle_t *handle, struct buffer_head *bh)
 }
 
 /*
- * Generic write_end handler for ordered and writeback ext3 journal modes.
- * We can't use generic_write_end, because that unlocks the page and we need to
- * unlock the page after ext3_journal_stop, but ext3_journal_stop must run
- * after block_write_end.
+ * This is nasty and subtle: ext3_write_begin() could have allocated blocks
+ * for the whole page but later we failed to copy the data in. Update inode
+ * size according to what we managed to copy. The rest is going to be
+ * truncated in write_end function.
  */
-static int ext3_generic_write_end(struct file *file,
-                               struct address_space *mapping,
-                               loff_t pos, unsigned len, unsigned copied,
-                               struct page *page, void *fsdata)
+static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
 {
-       struct inode *inode = file->f_mapping->host;
-
-       copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
-
-       if (pos+copied > inode->i_size) {
-               i_size_write(inode, pos+copied);
+       /* What matters to us is i_disksize. We don't write i_size anywhere */
+       if (pos + copied > inode->i_size)
+               i_size_write(inode, pos + copied);
+       if (pos + copied > EXT3_I(inode)->i_disksize) {
+               EXT3_I(inode)->i_disksize = pos + copied;
                mark_inode_dirty(inode);
        }
-
-       return copied;
 }
 
 /*
@@ -1256,34 +1274,29 @@ static int ext3_ordered_write_end(struct file *file,
        unsigned from, to;
        int ret = 0, ret2;
 
-       from = pos & (PAGE_CACHE_SIZE - 1);
-       to = from + len;
+       copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 
+       from = pos & (PAGE_CACHE_SIZE - 1);
+       to = from + copied;
        ret = walk_page_buffers(handle, page_buffers(page),
-               from, to, NULL, ext3_journal_dirty_data);
+               from, to, NULL, journal_dirty_data_fn);
 
-       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 > EXT3_I(inode)->i_disksize)
-                       EXT3_I(inode)->i_disksize = new_i_size;
-               copied = ext3_generic_write_end(file, mapping, pos, len, copied,
-                                                       page, fsdata);
-               if (copied < 0)
-                       ret = copied;
-       }
+       if (ret == 0)
+               update_file_sizes(inode, pos, copied);
+       /*
+        * There may be allocated blocks outside of i_size because
+        * we failed to copy some data. Prepare for truncate.
+        */
+       if (pos + len > inode->i_size)
+               ext3_orphan_add(handle, inode);
        ret2 = ext3_journal_stop(handle);
        if (!ret)
                ret = ret2;
        unlock_page(page);
        page_cache_release(page);
 
+       if (pos + len > inode->i_size)
+               vmtruncate(inode, inode->i_size);
        return ret ? ret : copied;
 }
 
@@ -1294,24 +1307,22 @@ static int ext3_writeback_write_end(struct file *file,
 {
        handle_t *handle = ext3_journal_current_handle();
        struct inode *inode = file->f_mapping->host;
-       int ret = 0, ret2;
-       loff_t new_i_size;
-
-       new_i_size = pos + copied;
-       if (new_i_size > EXT3_I(inode)->i_disksize)
-               EXT3_I(inode)->i_disksize = new_i_size;
-
-       copied = ext3_generic_write_end(file, mapping, pos, len, copied,
-                                                       page, fsdata);
-       if (copied < 0)
-               ret = copied;
+       int ret;
 
-       ret2 = ext3_journal_stop(handle);
-       if (!ret)
-               ret = ret2;
+       copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
+       update_file_sizes(inode, pos, copied);
+       /*
+        * There may be allocated blocks outside of i_size because
+        * we failed to copy some data. Prepare for truncate.
+        */
+       if (pos + len > inode->i_size)
+               ext3_orphan_add(handle, inode);
+       ret = ext3_journal_stop(handle);
        unlock_page(page);
        page_cache_release(page);
 
+       if (pos + len > inode->i_size)
+               vmtruncate(inode, inode->i_size);
        return ret ? ret : copied;
 }
 
@@ -1332,15 +1343,23 @@ static int ext3_journalled_write_end(struct file *file,
        if (copied < len) {
                if (!PageUptodate(page))
                        copied = 0;
-               page_zero_new_buffers(page, from+copied, to);
+               page_zero_new_buffers(page, from + copied, to);
+               to = from + copied;
        }
 
        ret = walk_page_buffers(handle, page_buffers(page), from,
                                to, &partial, write_end_fn);
        if (!partial)
                SetPageUptodate(page);
-       if (pos+copied > inode->i_size)
-               i_size_write(inode, pos+copied);
+
+       if (pos + copied > inode->i_size)
+               i_size_write(inode, pos + copied);
+       /*
+        * There may be allocated blocks outside of i_size because
+        * we failed to copy some data. Prepare for truncate.
+        */
+       if (pos + len > inode->i_size)
+               ext3_orphan_add(handle, inode);
        EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
        if (inode->i_size > EXT3_I(inode)->i_disksize) {
                EXT3_I(inode)->i_disksize = inode->i_size;
@@ -1355,6 +1374,8 @@ static int ext3_journalled_write_end(struct file *file,
        unlock_page(page);
        page_cache_release(page);
 
+       if (pos + len > inode->i_size)
+               vmtruncate(inode, inode->i_size);
        return ret ? ret : copied;
 }
 
@@ -1422,11 +1443,9 @@ static int bput_one(handle_t *handle, struct buffer_head *bh)
        return 0;
 }
 
-static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
+static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
 {
-       if (buffer_mapped(bh))
-               return ext3_journal_dirty_data(handle, bh);
-       return 0;
+       return !buffer_mapped(bh);
 }
 
 /*
@@ -1499,6 +1518,19 @@ static int ext3_ordered_writepage(struct page *page,
        if (ext3_journal_current_handle())
                goto out_fail;
 
+       if (!page_has_buffers(page)) {
+               create_empty_buffers(page, inode->i_sb->s_blocksize,
+                               (1 << BH_Dirty)|(1 << BH_Uptodate));
+               page_bufs = page_buffers(page);
+       } else {
+               page_bufs = page_buffers(page);
+               if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
+                                      NULL, buffer_unmapped)) {
+                       /* Provide NULL get_block() to catch bugs if buffers
+                        * weren't really mapped */
+                       return block_write_full_page(page, NULL, wbc);
+               }
+       }
        handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
 
        if (IS_ERR(handle)) {
@@ -1506,11 +1538,6 @@ static int ext3_ordered_writepage(struct page *page,
                goto out_fail;
        }
 
-       if (!page_has_buffers(page)) {
-               create_empty_buffers(page, inode->i_sb->s_blocksize,
-                               (1 << BH_Dirty)|(1 << BH_Uptodate));
-       }
-       page_bufs = page_buffers(page);
        walk_page_buffers(handle, page_bufs, 0,
                        PAGE_CACHE_SIZE, NULL, bget_one);
 
@@ -1558,6 +1585,15 @@ static int ext3_writeback_writepage(struct page *page,
        if (ext3_journal_current_handle())
                goto out_fail;
 
+       if (page_has_buffers(page)) {
+               if (!walk_page_buffers(NULL, page_buffers(page), 0,
+                                     PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
+                       /* Provide NULL get_block() to catch bugs if buffers
+                        * weren't really mapped */
+                       return block_write_full_page(page, NULL, wbc);
+               }
+       }
+
        handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
        if (IS_ERR(handle)) {
                ret = PTR_ERR(handle);
@@ -1680,7 +1716,8 @@ static int ext3_releasepage(struct page *page, gfp_t wait)
  * if the machine crashes during the write.
  *
  * If the O_DIRECT write is intantiating holes inside i_size and the machine
- * crashes then stale disk data _may_ be exposed inside the file.
+ * crashes then stale disk data _may_ be exposed inside the file. But current
+ * VFS code falls back into buffered path in that case so we are safe.
  */
 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
                        const struct iovec *iov, loff_t offset,
@@ -1689,7 +1726,7 @@ static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
        struct file *file = iocb->ki_filp;
        struct inode *inode = file->f_mapping->host;
        struct ext3_inode_info *ei = EXT3_I(inode);
-       handle_t *handle = NULL;
+       handle_t *handle;
        ssize_t ret;
        int orphan = 0;
        size_t count = iov_length(iov, nr_segs);
@@ -1697,17 +1734,21 @@ static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
        if (rw == WRITE) {
                loff_t final_size = offset + count;
 
-               handle = ext3_journal_start(inode, DIO_CREDITS);
-               if (IS_ERR(handle)) {
-                       ret = PTR_ERR(handle);
-                       goto out;
-               }
                if (final_size > inode->i_size) {
+                       /* Credits for sb + inode write */
+                       handle = ext3_journal_start(inode, 2);
+                       if (IS_ERR(handle)) {
+                               ret = PTR_ERR(handle);
+                               goto out;
+                       }
                        ret = ext3_orphan_add(handle, inode);
-                       if (ret)
-                               goto out_stop;
+                       if (ret) {
+                               ext3_journal_stop(handle);
+                               goto out;
+                       }
                        orphan = 1;
                        ei->i_disksize = inode->i_size;
+                       ext3_journal_stop(handle);
                }
        }
 
@@ -1715,18 +1756,21 @@ static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
                                 offset, nr_segs,
                                 ext3_get_block, NULL);
 
-       /*
-        * Reacquire the handle: ext3_get_block() can restart the transaction
-        */
-       handle = ext3_journal_current_handle();
-
-out_stop:
-       if (handle) {
+       if (orphan) {
                int err;
 
-               if (orphan && inode->i_nlink)
+               /* Credits for sb + inode write */
+               handle = ext3_journal_start(inode, 2);
+               if (IS_ERR(handle)) {
+                       /* This is really bad luck. We've written the data
+                        * but cannot extend i_size. Bail out and pretend
+                        * the write failed... */
+                       ret = PTR_ERR(handle);
+                       goto out;
+               }
+               if (inode->i_nlink)
                        ext3_orphan_del(handle, inode);
-               if (orphan && ret > 0) {
+               if (ret > 0) {
                        loff_t end = offset + ret;
                        if (end > inode->i_size) {
                                ei->i_disksize = end;
@@ -1769,44 +1813,47 @@ static int ext3_journalled_set_page_dirty(struct page *page)
 }
 
 static const struct address_space_operations ext3_ordered_aops = {
-       .readpage       = ext3_readpage,
-       .readpages      = ext3_readpages,
-       .writepage      = ext3_ordered_writepage,
-       .sync_page      = block_sync_page,
-       .write_begin    = ext3_write_begin,
-       .write_end      = ext3_ordered_write_end,
-       .bmap           = ext3_bmap,
-       .invalidatepage = ext3_invalidatepage,
-       .releasepage    = ext3_releasepage,
-       .direct_IO      = ext3_direct_IO,
-       .migratepage    = buffer_migrate_page,
+       .readpage               = ext3_readpage,
+       .readpages              = ext3_readpages,
+       .writepage              = ext3_ordered_writepage,
+       .sync_page              = block_sync_page,
+       .write_begin            = ext3_write_begin,
+       .write_end              = ext3_ordered_write_end,
+       .bmap                   = ext3_bmap,
+       .invalidatepage         = ext3_invalidatepage,
+       .releasepage            = ext3_releasepage,
+       .direct_IO              = ext3_direct_IO,
+       .migratepage            = buffer_migrate_page,
+       .is_partially_uptodate  = block_is_partially_uptodate,
 };
 
 static const struct address_space_operations ext3_writeback_aops = {
-       .readpage       = ext3_readpage,
-       .readpages      = ext3_readpages,
-       .writepage      = ext3_writeback_writepage,
-       .sync_page      = block_sync_page,
-       .write_begin    = ext3_write_begin,
-       .write_end      = ext3_writeback_write_end,
-       .bmap           = ext3_bmap,
-       .invalidatepage = ext3_invalidatepage,
-       .releasepage    = ext3_releasepage,
-       .direct_IO      = ext3_direct_IO,
-       .migratepage    = buffer_migrate_page,
+       .readpage               = ext3_readpage,
+       .readpages              = ext3_readpages,
+       .writepage              = ext3_writeback_writepage,
+       .sync_page              = block_sync_page,
+       .write_begin            = ext3_write_begin,
+       .write_end              = ext3_writeback_write_end,
+       .bmap                   = ext3_bmap,
+       .invalidatepage         = ext3_invalidatepage,
+       .releasepage            = ext3_releasepage,
+       .direct_IO              = ext3_direct_IO,
+       .migratepage            = buffer_migrate_page,
+       .is_partially_uptodate  = block_is_partially_uptodate,
 };
 
 static const struct address_space_operations ext3_journalled_aops = {
-       .readpage       = ext3_readpage,
-       .readpages      = ext3_readpages,
-       .writepage      = ext3_journalled_writepage,
-       .sync_page      = block_sync_page,
-       .write_begin    = ext3_write_begin,
-       .write_end      = ext3_journalled_write_end,
-       .set_page_dirty = ext3_journalled_set_page_dirty,
-       .bmap           = ext3_bmap,
-       .invalidatepage = ext3_invalidatepage,
-       .releasepage    = ext3_releasepage,
+       .readpage               = ext3_readpage,
+       .readpages              = ext3_readpages,
+       .writepage              = ext3_journalled_writepage,
+       .sync_page              = block_sync_page,
+       .write_begin            = ext3_write_begin,
+       .write_end              = ext3_journalled_write_end,
+       .set_page_dirty         = ext3_journalled_set_page_dirty,
+       .bmap                   = ext3_bmap,
+       .invalidatepage         = ext3_invalidatepage,
+       .releasepage            = ext3_releasepage,
+       .is_partially_uptodate  = block_is_partially_uptodate,
 };
 
 void ext3_set_aops(struct inode *inode)
@@ -2129,7 +2176,21 @@ static void ext3_free_data(handle_t *handle, struct inode *inode,
 
        if (this_bh) {
                BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
-               ext3_journal_dirty_metadata(handle, this_bh);
+
+               /*
+                * The buffer head should have an attached journal head at this
+                * point. However, if the data is corrupted and an indirect
+                * block pointed to itself, it would have been detached when
+                * the block was cleared. Check for this instead of OOPSing.
+                */
+               if (bh2jh(this_bh))
+                       ext3_journal_dirty_metadata(handle, this_bh);
+               else
+                       ext3_error(inode->i_sb, "ext3_free_data",
+                                  "circular indirect block detected, "
+                                  "inode=%lu, block=%llu",
+                                  inode->i_ino,
+                                  (unsigned long long)this_bh->b_blocknr);
        }
 }
 
@@ -2255,6 +2316,19 @@ static void ext3_free_branches(handle_t *handle, struct inode *inode,
        }
 }
 
+int ext3_can_truncate(struct inode *inode)
+{
+       if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
+               return 0;
+       if (S_ISREG(inode->i_mode))
+               return 1;
+       if (S_ISDIR(inode->i_mode))
+               return 1;
+       if (S_ISLNK(inode->i_mode))
+               return !ext3_inode_is_fast_symlink(inode);
+       return 0;
+}
+
 /*
  * ext3_truncate()
  *
@@ -2299,13 +2373,11 @@ void ext3_truncate(struct inode *inode)
        unsigned blocksize = inode->i_sb->s_blocksize;
        struct page *page;
 
-       if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
-           S_ISLNK(inode->i_mode)))
-               return;
-       if (ext3_inode_is_fast_symlink(inode))
-               return;
-       if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
-               return;
+       if (!ext3_can_truncate(inode))
+               goto out_notrans;
+
+       if (inode->i_size == 0 && ext3_should_writeback_data(inode))
+               ei->i_state |= EXT3_STATE_FLUSH_ON_CLOSE;
 
        /*
         * We have to lock the EOF page here, because lock_page() nests
@@ -2318,7 +2390,7 @@ void ext3_truncate(struct inode *inode)
                page = grab_cache_page(mapping,
                                inode->i_size >> PAGE_CACHE_SHIFT);
                if (!page)
-                       return;
+                       goto out_notrans;
        }
 
        handle = start_transaction(inode);
@@ -2329,7 +2401,7 @@ void ext3_truncate(struct inode *inode)
                        unlock_page(page);
                        page_cache_release(page);
                }
-               return;         /* AKPM: return what? */
+               goto out_notrans;
        }
 
        last_block = (inode->i_size + blocksize-1)
@@ -2453,16 +2525,23 @@ out_stop:
                ext3_orphan_del(handle, inode);
 
        ext3_journal_stop(handle);
+       return;
+out_notrans:
+       /*
+        * Delete the inode from orphan list so that it doesn't stay there
+        * forever and trigger assertion on umount.
+        */
+       if (inode->i_nlink)
+               ext3_orphan_del(NULL, inode);
 }
 
 static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
                unsigned long ino, struct ext3_iloc *iloc)
 {
-       unsigned long desc, group_desc, block_group;
+       unsigned long block_group;
        unsigned long offset;
        ext3_fsblk_t block;
-       struct buffer_head *bh;
-       struct ext3_group_desc * gdp;
+       struct ext3_group_desc *gdp;
 
        if (!ext3_valid_inum(sb, ino)) {
                /*
@@ -2474,27 +2553,15 @@ static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
        }
 
        block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
-       if (block_group >= EXT3_SB(sb)->s_groups_count) {
-               ext3_error(sb,"ext3_get_inode_block","group >= groups count");
+       gdp = ext3_get_group_desc(sb, block_group, NULL);
+       if (!gdp)
                return 0;
-       }
-       smp_rmb();
-       group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
-       desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
-       bh = EXT3_SB(sb)->s_group_desc[group_desc];
-       if (!bh) {
-               ext3_error (sb, "ext3_get_inode_block",
-                           "Descriptor not loaded");
-               return 0;
-       }
-
-       gdp = (struct ext3_group_desc *)bh->b_data;
        /*
         * Figure out the offset within the block group inode table
         */
        offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
                EXT3_INODE_SIZE(sb);
-       block = le32_to_cpu(gdp[desc].bg_inode_table) +
+       block = le32_to_cpu(gdp->bg_inode_table) +
                (offset >> EXT3_BLOCK_SIZE_BITS(sb));
 
        iloc->block_group = block_group;
@@ -2528,6 +2595,16 @@ static int __ext3_get_inode_loc(struct inode *inode,
        }
        if (!buffer_uptodate(bh)) {
                lock_buffer(bh);
+
+               /*
+                * If the buffer has the write error flag, we have failed
+                * to write out another inode in the same block.  In this
+                * case, we don't have to read the block because we may
+                * read the old inode data successfully.
+                */
+               if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
+                       set_buffer_uptodate(bh);
+
                if (buffer_uptodate(bh)) {
                        /* someone brought it uptodate while we waited */
                        unlock_buffer(bh);
@@ -2658,21 +2735,27 @@ void ext3_get_inode_flags(struct ext3_inode_info *ei)
                ei->i_flags |= EXT3_DIRSYNC_FL;
 }
 
-void ext3_read_inode(struct inode * inode)
+struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
 {
        struct ext3_iloc iloc;
        struct ext3_inode *raw_inode;
-       struct ext3_inode_info *ei = EXT3_I(inode);
+       struct ext3_inode_info *ei;
        struct buffer_head *bh;
+       struct inode *inode;
+       long ret;
        int block;
 
-#ifdef CONFIG_EXT3_FS_POSIX_ACL
-       ei->i_acl = EXT3_ACL_NOT_CACHED;
-       ei->i_default_acl = EXT3_ACL_NOT_CACHED;
-#endif
+       inode = iget_locked(sb, ino);
+       if (!inode)
+               return ERR_PTR(-ENOMEM);
+       if (!(inode->i_state & I_NEW))
+               return inode;
+
+       ei = EXT3_I(inode);
        ei->i_block_alloc_info = NULL;
 
-       if (__ext3_get_inode_loc(inode, &iloc, 0))
+       ret = __ext3_get_inode_loc(inode, &iloc, 0);
+       if (ret < 0)
                goto bad_inode;
        bh = iloc.bh;
        raw_inode = ext3_raw_inode(&iloc);
@@ -2703,6 +2786,7 @@ void ext3_read_inode(struct inode * inode)
                    !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
                        /* this inode is deleted */
                        brelse (bh);
+                       ret = -ESTALE;
                        goto bad_inode;
                }
                /* The only unlinked inodes we let through here have
@@ -2746,6 +2830,7 @@ void ext3_read_inode(struct inode * inode)
                if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
                    EXT3_INODE_SIZE(inode->i_sb)) {
                        brelse (bh);
+                       ret = -EIO;
                        goto bad_inode;
                }
                if (ei->i_extra_isize == 0) {
@@ -2770,9 +2855,11 @@ void ext3_read_inode(struct inode * inode)
                inode->i_op = &ext3_dir_inode_operations;
                inode->i_fop = &ext3_dir_operations;
        } else if (S_ISLNK(inode->i_mode)) {
-               if (ext3_inode_is_fast_symlink(inode))
+               if (ext3_inode_is_fast_symlink(inode)) {
                        inode->i_op = &ext3_fast_symlink_inode_operations;
-               else {
+                       nd_terminate_link(ei->i_data, inode->i_size,
+                               sizeof(ei->i_data) - 1);
+               } else {
                        inode->i_op = &ext3_symlink_inode_operations;
                        ext3_set_aops(inode);
                }
@@ -2787,11 +2874,12 @@ void ext3_read_inode(struct inode * inode)
        }
        brelse (iloc.bh);
        ext3_set_inode_flags(inode);
-       return;
+       unlock_new_inode(inode);
+       return inode;
 
 bad_inode:
-       make_bad_inode(inode);
-       return;
+       iget_failed(inode);
+       return ERR_PTR(ret);
 }
 
 /*
@@ -2876,7 +2964,6 @@ static int ext3_do_update_inode(handle_t *handle,
                                ext3_update_dynamic_rev(sb);
                                EXT3_SET_RO_COMPAT_FEATURE(sb,
                                        EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
-                               sb->s_dirt = 1;
                                handle->h_sync = 1;
                                err = ext3_journal_dirty_metadata(handle,
                                                EXT3_SB(sb)->s_sbh);
@@ -3004,7 +3091,7 @@ int ext3_setattr(struct dentry *dentry, struct iattr *attr)
                        error = PTR_ERR(handle);
                        goto err_out;
                }
-               error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
+               error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
                if (error) {
                        ext3_journal_stop(handle);
                        return error;
@@ -3039,12 +3126,6 @@ int ext3_setattr(struct dentry *dentry, struct iattr *attr)
 
        rc = inode_setattr(inode, attr);
 
-       /* If inode_setattr's call to ext3_truncate failed to get a
-        * transaction handle at all, we need to clean up the in-core
-        * orphan list manually. */
-       if (inode->i_nlink)
-               ext3_orphan_del(NULL, inode);
-
        if (!rc && (ia_valid & ATTR_MODE))
                rc = ext3_acl_chmod(inode);
 
@@ -3095,7 +3176,7 @@ static int ext3_writepage_trans_blocks(struct inode *inode)
                ret = 2 * (bpp + indirects) + 2;
 
 #ifdef CONFIG_QUOTA
-       /* We know that structure was already allocated during DQUOT_INIT so
+       /* We know that structure was already allocated during vfs_dq_init so
         * we will be updating only the data blocks + inodes */
        ret += 2*EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb);
 #endif
@@ -3186,7 +3267,7 @@ int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
  * i_size has been changed by generic_commit_write() and we thus need
  * to include the updated inode in the current transaction.
  *
- * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
+ * Also, vfs_dq_alloc_space() will always dirty the inode when blocks
  * are allocated to the file.
  *
  * If the inode is marked synchronous, we don't honour that here - doing
@@ -3205,7 +3286,7 @@ void ext3_dirty_inode(struct inode *inode)
                current_handle->h_transaction != handle->h_transaction) {
                /* This task has a transaction open against a different fs */
                printk(KERN_EMERG "%s: transactions do not match!\n",
-                      __FUNCTION__);
+                      __func__);
        } else {
                jbd_debug(5, "marking dirty.  outer handle=%p\n",
                                current_handle);