Btrfs: Add O_DIRECT read and write (writes == buffered + cache flush)
This adds basic O_DIRECT read and write support. In the write case, we
just do a normal buffered write followed by a cache flush. O_DIRECT +
O_SYNC are required to trigger metadata syncs.
In the read case, there is a basic btrfs_get_block call for use by
the generic O_DIRECT code. This does honor multi-volume mapping rules
but it skips all checksumming.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 5fa4504..5b93864 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -796,8 +796,6 @@
PAGE_CACHE_SIZE / (sizeof(struct page *)));
pinned[0] = NULL;
pinned[1] = NULL;
- if (file->f_flags & O_DIRECT)
- return -EINVAL;
pos = *ppos;
start_pos = pos;
@@ -909,6 +907,15 @@
start_pos, num_written);
if (err < 0)
num_written = err;
+ } else if (num_written > 0 && (file->f_flags & O_DIRECT)) {
+ do_sync_mapping_range(inode->i_mapping, start_pos,
+ start_pos + num_written - 1,
+ SYNC_FILE_RANGE_WRITE |
+ SYNC_FILE_RANGE_WAIT_AFTER);
+
+ invalidate_mapping_pages(inode->i_mapping,
+ start_pos >> PAGE_CACHE_SHIFT,
+ (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
}
current->backing_dev_info = NULL;
return num_written ? num_written : err;
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 48f1d1b..32e778f 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2265,6 +2265,70 @@
return em;
}
+static int btrfs_get_block(struct inode *inode, sector_t iblock,
+ struct buffer_head *bh_result, int create)
+{
+ struct extent_map *em;
+ u64 start = (u64)iblock << inode->i_blkbits;
+ struct btrfs_multi_bio *multi = NULL;
+ struct btrfs_root *root = BTRFS_I(inode)->root;
+ u64 len;
+ u64 logical;
+ u64 map_length;
+ int ret = 0;
+
+ em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
+
+ if (!em || IS_ERR(em))
+ goto out;
+
+ if (em->start > start || em->start + em->len <= start)
+ goto out;
+
+ if (em->block_start == EXTENT_MAP_INLINE) {
+ ret = -EINVAL;
+ goto out;
+ }
+
+ if (em->block_start == EXTENT_MAP_HOLE ||
+ em->block_start == EXTENT_MAP_DELALLOC) {
+ goto out;
+ }
+
+ len = em->start + em->len - start;
+ len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
+
+ logical = start - em->start;
+ logical = em->block_start + logical;
+
+ map_length = len;
+ ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
+ logical, &map_length, &multi, 0);
+ BUG_ON(ret);
+ bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
+ bh_result->b_size = min(map_length, len);
+ bh_result->b_bdev = multi->stripes[0].dev->bdev;
+ set_buffer_mapped(bh_result);
+ kfree(multi);
+out:
+ free_extent_map(em);
+ return ret;
+}
+
+static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
+ const struct iovec *iov, loff_t offset,
+ unsigned long nr_segs)
+{
+ struct file *file = iocb->ki_filp;
+ struct inode *inode = file->f_mapping->host;
+
+ if (rw == WRITE)
+ return -EINVAL;
+
+ return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
+ offset, nr_segs, btrfs_get_block, NULL);
+}
+
static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
{
return extent_bmap(mapping, iblock, btrfs_get_extent);
@@ -3136,6 +3200,7 @@
btrfs_throttle(root);
return err;
}
+
static int btrfs_permission(struct inode *inode, int mask,
struct nameidata *nd)
{
@@ -3193,6 +3258,7 @@
.readpages = btrfs_readpages,
.sync_page = block_sync_page,
.bmap = btrfs_bmap,
+ .direct_IO = btrfs_direct_IO,
.invalidatepage = btrfs_invalidatepage,
.releasepage = btrfs_releasepage,
.set_page_dirty = __set_page_dirty_nobuffers,