This patch corrects a punch hole bug found by xfstests
when the block size is set to 1k. Test 127 runs longer
before it fails, but that appears to be a separate bug.
This bug happens because the punch hole code only zeros
out non block aligned blocks, and then releases the pages
for data that is page aligned. This means that if the
blocks are smaller than a page, then the blocks contained
in the non page aligned regions (but still block aligned)
are left unzeroed and mapped.
This patch adds a new ext4_unmap_partial_page_buffers routine
that unmapps the block aligned buffers in a page that are
contained in a specified range.
Signed-off-by: Allison Henderson <achender@xxxxxxxxxxxxxxxxxx>
---
v1 -> v2
Added EXT4_BLOCK_ZERO_DISCARD_BUFFER flag
v2 -> v3
Moved code out of ext4_zero_block_page_range and in
to new ext4_unmap_page_range function
v3 -> v4
Renamed ext4_unmap_page_range to ext4_unmap_partial_page_buffers
Moved ext4_unmap_partial_page_buffers from inode.c to extents.c
Corrected comments for non block/page aligned handling
Added checks to avoid unnecessary page unmaps
Removed unneeded journaling and mapping from new routine
:100644 100644 4d73e11... a946023... M fs/ext4/extents.c
fs/ext4/extents.c | 142 +++++++++++++++++++++++++++++++++++++++++++++++++++--
1 files changed, 138 insertions(+), 4 deletions(-)
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 4d73e11..a946023 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -4137,6 +4137,107 @@ static int ext4_xattr_fiemap(struct inode *inode,
}
/*
+ * ext4_unmap_partial_page_buffers()
+ * Unmaps a page range of length 'length' starting from offset
+ * 'from'. The range to be unmaped must be contained with in
+ * one page. If the specified range exceeds the end of the page
+ * it will be shortened to end of the page that cooresponds to
+ * 'from'. Only block aligned buffers will be unmapped and unblock
+ * aligned buffers are skipped
+ */
+static int ext4_unmap_partial_page_buffers(handle_t *handle,
+ struct address_space *mapping, loff_t from, loff_t length)
+{
+ ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
+ unsigned int offset = from & (PAGE_CACHE_SIZE-1);
+ unsigned int blocksize, max, pos;
+ unsigned int end_of_block, range_to_unmap;
+ ext4_lblk_t iblock;
+ struct inode *inode = mapping->host;
+ struct buffer_head *bh;
+ struct page *page;
+ int err = 0;
+
+ page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
+ mapping_gfp_mask(mapping) & ~__GFP_FS);
+ if (!page)
+ return -EINVAL;
+
+ blocksize = inode->i_sb->s_blocksize;
+ max = PAGE_CACHE_SIZE - offset;
+
+ /*
+ * correct length if it does not fall between
+ * 'from' and the end of the page
+ */
+ if (length > max || length < 0)
+ length = max;
+
+ iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
+
+ if (!page_has_buffers(page))
+ goto unlock;
+
+ /* Find the buffer that contains "offset" */
+ bh = page_buffers(page);
+ pos = blocksize;
+ while (offset >= pos) {
+ bh = bh->b_this_page;
+ iblock++;
+ pos += blocksize;
+ }
+
+ pos = offset;
+ while (pos < offset + length) {
+ err = 0;
+
+ /* The length of space left to zero */
+ range_to_unmap = offset + length - pos;
+
+ /* The length of space until the end of the block */
+ end_of_block = blocksize - (pos & (blocksize-1));
+
+ /* Do not unmap past end of block */
+ if (range_to_unmap > end_of_block)
+ range_to_unmap = end_of_block;
+
+ if (buffer_freed(bh)) {
+ BUFFER_TRACE(bh, "freed: skip");
+ goto next;
+ }
+
+ if (!buffer_mapped(bh)) {
+ BUFFER_TRACE(bh, "unmapped: skip");
+ goto next;
+ }
+
+ /* If the range is not block aligned, skip */
+ if (range_to_unmap != blocksize)
+ goto next;
+
+ clear_buffer_dirty(bh);
+ bh->b_bdev = NULL;
+ clear_buffer_mapped(bh);
+ clear_buffer_req(bh);
+ clear_buffer_new(bh);
+ clear_buffer_delay(bh);
+ clear_buffer_unwritten(bh);
+ clear_buffer_uptodate(bh);
+ ClearPageUptodate(page);
+
+ BUFFER_TRACE(bh, "buffer unmapped");
+next:
+ bh = bh->b_this_page;
+ iblock++;
+ pos += range_to_unmap;
+ }
+unlock:
+ unlock_page(page);
+ page_cache_release(page);
+ return err;
+}
+
+/*
* ext4_ext_punch_hole
*
* Punches a hole of "length" bytes in a file starting
@@ -4157,7 +4258,7 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
struct address_space *mapping = inode->i_mapping;
struct ext4_map_blocks map;
handle_t *handle;
- loff_t first_block_offset, last_block_offset, block_len;
+ loff_t first_block_offset, last_block_offset, block_len, page_len;
loff_t first_page, last_page, first_page_offset, last_page_offset;
int ret, credits, blocks_released, err = 0;
@@ -4206,9 +4307,9 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
goto out;
/*
- * Now we need to zero out the un block aligned data.
- * If the file is smaller than a block, just
- * zero out the middle
+ * Now we need to zero out the non-block-aligned data.
+ * If the file space being truncated is smaller than
+ * than a block, just zero out the middle
*/
if (first_block > last_block)
ext4_block_zero_page_range(handle, mapping, offset, length);
@@ -4227,6 +4328,39 @@ int ext4_ext_punch_hole(struct file *file, loff_t offset, loff_t length)
}
}
+ /*
+ * Now we need to unmap the non-page-aligned buffers.
+ * If the block size is smaller than the page size
+ * and the file space being truncated is not
+ * page aligned, then unmap the buffers
+ */
+ if (inode->i_sb->s_blocksize < PAGE_CACHE_SIZE &&
+ !((offset % PAGE_CACHE_SIZE == 0) &&
+ (length % PAGE_CACHE_SIZE == 0))) {
+
+ /*
+ * If the file space being truncated is smaller
+ * than a page, just unmap the middle
+ */
+ if (first_page > last_page) {
+ ext4_unmap_partial_page_buffers(handle,
+ mapping, offset, length);
+ } else {
+ /* unmap page buffers before the first aligned page */
+ page_len = first_page_offset - offset;
+ if (page_len > 0)
+ ext4_unmap_partial_page_buffers(handle,
+ mapping, offset, page_len);
+
+ /* unmap the page buffers after the last aligned page */
+ page_len = offset + length - last_page_offset;
+ if (page_len > 0) {
+ ext4_unmap_partial_page_buffers(handle,
+ mapping, last_page_offset, page_len);
+ }
+ }
+ }
+
/* If there are no blocks to remove, return now */
if (first_block >= last_block)
goto out;
--
1.7.1
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
