Merge cam.ac.uk:/rain/usr/src/bkntfs-tng-2.5 into cam.ac.uk:/usr/src/tng

fs/ntfs/ChangeLog

@@ -8,7 +8,7 @@ ToDo:
 	  functions need to clone block_read_full_page and modify it to cope
 	  with the significance of the different attribute sizes.
 	  Still need to go through:
-		aops.c, attrib.c, compress.c, dir.c, mft.c
+		aops.c, attrib.c, compress.c, dir.c
 	- Find and fix bugs.
 	- W.r.t. s_maxbytes still need to be careful on reading/truncating as
 	  there are dragons lurking in the details, e.g. read_inode() currently
@@ -33,12 +33,19 @@ ToDo:
 	  in between. But now we have a semaphore so are ok. Update
 	  load_attribute_list to use the semaphore so can be called outside
 	  read_inode, too. Apply the same optimization where desired.
+	- Optimize mft_readpage() to not do i/o on buffer heads beyond
+	  initialized_size, just zero the buffer heads instead. Question: How
+	  to setup the buffer heads so they point to the on disk location
+	  correctly (after all they are allocated) but are not read from disk?
 
 tng-0.0.9 - Work in progress
 
 	- Add kill_super, just keeping up with the vfs changes in the kernel.
 	- Repeat some changes from tng-0.0.8 that somehow got lost on the way
 	  from the CVS import into BitKeeper.
+	- Begin to implement proper handling of allocated_size vs
+	  initialized_size vs data_size (i.e. i_size). Done are
+	  mft.c::ntfs_mft_readpage() and aops.c::end_buffer_read_index_async().
 
 tng-0.0.8 - 08/03/2002 - Now using BitKeeper, http://linux-ntfs.bkbits.net/
 

fs/ntfs/aops.c

@@ -431,12 +431,46 @@ void end_buffer_read_index_async(struct buffer_head *bh, int uptodate)
 	unsigned long flags;
 	struct buffer_head *tmp;
 	struct page *page;
+	ntfs_inode *ni;
 
 	mark_buffer_uptodate(bh, uptodate);
 
-	/* This is a temporary buffer used for page I/O. */
 	page = bh->b_page;
-	if (!uptodate)
+	ni = NTFS_I(page->mapping->host);
+	if (likely(uptodate)) {
+		/*
+		 * The below code is very cpu intensive so we add an extra
+		 * check to ensure it is only run when processing buffer heads
+		 * in pages reaching beyond the initialized data size.
+		 */
+		if ((page->index + 1) << PAGE_CACHE_SHIFT >
+				ni->initialized_size) {
+			s64 file_ofs;
+			char *addr;
+			int page_ofs;
+
+			addr = kmap_atomic(page, KM_BIO_IRQ);
+			BUG_ON(bh->b_data < addr);
+			BUG_ON(bh->b_data + bh->b_size > addr +
+					PAGE_CACHE_SIZE);
+			page_ofs = bh->b_data - addr;
+			file_ofs = (page->index << PAGE_CACHE_SHIFT) + page_ofs;
+			/* Check for the current buffer head overflowing. */
+			if (file_ofs + bh->b_size > ni->initialized_size) {
+				int bh_ofs = 0;
+				if (file_ofs < ni->initialized_size) {
+					bh_ofs = ni->initialized_size -
+							file_ofs;
+					BUG_ON(bh_ofs < 0);
+					BUG_ON(bh_ofs >= bh->b_size);
+				}
+				memset(bh->b_data + bh_ofs, 0,
+						bh->b_size - bh_ofs);
+				flush_dcache_page(page);
+			}
+			kunmap_atomic(addr, KM_BIO_IRQ);
+		}
+	} else
 		SetPageError(page);
 	/*
 	 * Be _very_ careful from here on. Bad things can happen if
@@ -470,7 +504,6 @@ void end_buffer_read_index_async(struct buffer_head *bh, int uptodate)
 		char *addr;
 		unsigned int i, recs, nr_err = 0;
 		u32 rec_size;
-		ntfs_inode *ni = NTFS_I(page->mapping->host);
 
 		addr = kmap_atomic(page, KM_BIO_IRQ);
 		rec_size = ni->_IDM(index_block_size);
@@ -487,6 +520,7 @@ void end_buffer_read_index_async(struct buffer_head *bh, int uptodate)
 					PAGE_CACHE_SHIFT >>
 					ni->_IDM(index_block_size_bits)) + i));
 		}
+		flush_dcache_page(page);
 		kunmap_atomic(addr, KM_BIO_IRQ);
 		if (!nr_err && recs)
 			SetPageUptodate(page);

fs/ntfs/mft.c

@@ -148,7 +148,17 @@ int format_mft_record(ntfs_inode *ni, MFT_RECORD *mft_rec)
  *
  * Note, we only setup asynchronous I/O on the page and return. I/O completion
  * is signalled via our asynchronous I/O completion handler
- * end_buffer_read_index_async().
+ * end_buffer_read_index_async(). We also take care of the nitty gritty
+ * details towards the end of the file and zero out non-initialized regions.
+ *
+ * TODO:/FIXME: The current implementation is simple but wasteful as we perform
+ * actual i/o from disk for all data up to allocated size completely ignoring
+ * the fact that initialized size, and data size for that matter, may well be
+ * lower and hence there is no point in reading them in. We can just zero the
+ * page range, which is what is currently done in our async i/o completion
+ * handler anyway once the read from disk completes. However, I am not sure how
+ * to setup the buffer heads in that case, so for now we do the pointless i/o.
+ * Any help with this would be appreciated...
  */
 static int ntfs_mft_readpage(struct file *file, struct page *page)
 {
@@ -160,7 +170,7 @@ static int ntfs_mft_readpage(struct file *file, struct page *page)
 	ntfs_volume *vol;
 	struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
 	sector_t iblock, lblock;
-	unsigned int blocksize, blocks,vcn_ofs;
+	unsigned int blocksize, blocks, vcn_ofs;
 	int i, nr;
 	unsigned char blocksize_bits;