ext2: Move direct-io to use iomap

This patch converts ext2 direct-io path to iomap interface.
- This also takes care of DIO_SKIP_HOLES part in which we return -ENOTBLK
from ext2_iomap_begin(), in case if the write is done on a hole.
- This fallbacks to buffered-io in case of DIO_SKIP_HOLES or in case of
a partial write or if any error is detected in ext2_iomap_end().
We try to return -ENOTBLK in such cases.
- For any unaligned or extending DIO writes, we pass
IOMAP_DIO_FORCE_WAIT flag to ensure synchronous writes.
- For extending writes we set IOMAP_F_DIRTY in ext2_iomap_begin because
otherwise with dsync writes on devices that support FUA, generic_write_sync
won't be called and we might miss inode metadata updates.
- Since ext2 already now uses _nolock vartiant of sync write. Hence
there is no inode lock problem with iomap in this patch.
- ext2_iomap_ops are now being shared by DIO, DAX & fiemap path

Tested-by: Disha Goel <[e-mail]>
Reviewed-by: Christoph Hellwig <[e-mail]>
Signed-off-by: Ritesh Harjani (IBM) <[e-mail]>
Signed-off-by: Jan Kara <[e-mail]>
Message-Id: <610[e-mail]>

fs/ext2/ext2.h

···
 extern struct inode *ext2_iget (struct super_block *, unsigned long);
 extern int ext2_write_inode (struct inode *, struct writeback_control *);
 extern void ext2_evict_inode(struct inode *);
+void ext2_write_failed(struct address_space *mapping, loff_t to);
 extern int ext2_get_block(struct inode *, sector_t, struct buffer_head *, int);
 extern int ext2_setattr (struct mnt_idmap *, struct dentry *, struct iattr *);
 extern int ext2_getattr (struct mnt_idmap *, const struct path *,

fs/ext2/file.c

···
 	return ret;
 }
 
+static ssize_t ext2_dio_read_iter(struct kiocb *iocb, struct iov_iter *to)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_mapping->host;
+	ssize_t ret;
+
+	inode_lock_shared(inode);
+	ret = iomap_dio_rw(iocb, to, &ext2_iomap_ops, NULL, 0, NULL, 0);
+	inode_unlock_shared(inode);
+
+	return ret;
+}
+
+static int ext2_dio_write_end_io(struct kiocb *iocb, ssize_t size,
+				 int error, unsigned int flags)
+{
+	loff_t pos = iocb->ki_pos;
+	struct inode *inode = file_inode(iocb->ki_filp);
+
+	if (error)
+		goto out;
+
+	/*
+	 * If we are extending the file, we have to update i_size here before
+	 * page cache gets invalidated in iomap_dio_rw(). This prevents racing
+	 * buffered reads from zeroing out too much from page cache pages.
+	 * Note that all extending writes always happens synchronously with
+	 * inode lock held by ext2_dio_write_iter(). So it is safe to update
+	 * inode size here for extending file writes.
+	 */
+	pos += size;
+	if (pos > i_size_read(inode)) {
+		i_size_write(inode, pos);
+		mark_inode_dirty(inode);
+	}
+out:
+	return error;
+}
+
+static const struct iomap_dio_ops ext2_dio_write_ops = {
+	.end_io = ext2_dio_write_end_io,
+};
+
+static ssize_t ext2_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_mapping->host;
+	ssize_t ret;
+	unsigned int flags = 0;
+	unsigned long blocksize = inode->i_sb->s_blocksize;
+	loff_t offset = iocb->ki_pos;
+	loff_t count = iov_iter_count(from);
+
+	inode_lock(inode);
+	ret = generic_write_checks(iocb, from);
+	if (ret <= 0)
+		goto out_unlock;
+
+	ret = kiocb_modified(iocb);
+	if (ret)
+		goto out_unlock;
+
+	/* use IOMAP_DIO_FORCE_WAIT for unaligned or extending writes */
+	if (iocb->ki_pos + iov_iter_count(from) > i_size_read(inode) ||
+	   (!IS_ALIGNED(iocb->ki_pos | iov_iter_alignment(from), blocksize)))
+		flags |= IOMAP_DIO_FORCE_WAIT;
+
+	ret = iomap_dio_rw(iocb, from, &ext2_iomap_ops, &ext2_dio_write_ops,
+			   flags, NULL, 0);
+
+	/* ENOTBLK is magic return value for fallback to buffered-io */
+	if (ret == -ENOTBLK)
+		ret = 0;
+
+	if (ret < 0 && ret != -EIOCBQUEUED)
+		ext2_write_failed(inode->i_mapping, offset + count);
+
+	/* handle case for partial write and for fallback to buffered write */
+	if (ret >= 0 && iov_iter_count(from)) {
+		loff_t pos, endbyte;
+		ssize_t status;
+		int ret2;
+
+		iocb->ki_flags &= ~IOCB_DIRECT;
+		pos = iocb->ki_pos;
+		status = generic_perform_write(iocb, from);
+		if (unlikely(status < 0)) {
+			ret = status;
+			goto out_unlock;
+		}
+
+		iocb->ki_pos += status;
+		ret += status;
+		endbyte = pos + status - 1;
+		ret2 = filemap_write_and_wait_range(inode->i_mapping, pos,
+						    endbyte);
+		if (!ret2)
+			invalidate_mapping_pages(inode->i_mapping,
+						 pos >> PAGE_SHIFT,
+						 endbyte >> PAGE_SHIFT);
+		if (ret > 0)
+			generic_write_sync(iocb, ret);
+	}
+
+out_unlock:
+	inode_unlock(inode);
+	return ret;
+}
+
 static ssize_t ext2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 {
 #ifdef CONFIG_FS_DAX
 	if (IS_DAX(iocb->ki_filp->f_mapping->host))
 		return ext2_dax_read_iter(iocb, to);
 #endif
+	if (iocb->ki_flags & IOCB_DIRECT)
+		return ext2_dio_read_iter(iocb, to);
+
 	return generic_file_read_iter(iocb, to);
 }
 
···
 	if (IS_DAX(iocb->ki_filp->f_mapping->host))
 		return ext2_dax_write_iter(iocb, from);
 #endif
+	if (iocb->ki_flags & IOCB_DIRECT)
+		return ext2_dio_write_iter(iocb, from);
+
 	return generic_file_write_iter(iocb, from);
 }
 

fs/ext2/inode.c

···
 
 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
 
-static void ext2_write_failed(struct address_space *mapping, loff_t to)
+void ext2_write_failed(struct address_space *mapping, loff_t to)
 {
 	struct inode *inode = mapping->host;
 
···
 	bool new = false, boundary = false;
 	u32 bno;
 	int ret;
+	bool create = flags & IOMAP_WRITE;
+
+	/*
+	 * For writes that could fill holes inside i_size on a
+	 * DIO_SKIP_HOLES filesystem we forbid block creations: only
+	 * overwrites are permitted.
+	 */
+	if ((flags & IOMAP_DIRECT) &&
+	    (first_block << blkbits) < i_size_read(inode))
+		create = 0;
+
+	/*
+	 * Writes that span EOF might trigger an IO size update on completion,
+	 * so consider them to be dirty for the purposes of O_DSYNC even if
+	 * there is no other metadata changes pending or have been made here.
+	 */
+	if ((flags & IOMAP_WRITE) && offset + length > i_size_read(inode))
+		iomap->flags |= IOMAP_F_DIRTY;
 
 	ret = ext2_get_blocks(inode, first_block, max_blocks,
-			&bno, &new, &boundary, flags & IOMAP_WRITE);
+			&bno, &new, &boundary, create);
 	if (ret < 0)
 		return ret;
 
···
 		iomap->bdev = inode->i_sb->s_bdev;
 
 	if (ret == 0) {
+		/*
+		 * Switch to buffered-io for writing to holes in a non-extent
+		 * based filesystem to avoid stale data exposure problem.
+		 */
+		if (!create && (flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT))
+			return -ENOTBLK;
 		iomap->type = IOMAP_HOLE;
 		iomap->addr = IOMAP_NULL_ADDR;
 		iomap->length = 1 << blkbits;
···
 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
 		ssize_t written, unsigned flags, struct iomap *iomap)
 {
+	/*
+	 * Switch to buffered-io in case of any error.
+	 * Blocks allocated can be used by the buffered-io path.
+	 */
+	if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE) && written == 0)
+		return -ENOTBLK;
+
 	if (iomap->type == IOMAP_MAPPED &&
 	    written < length &&
 	    (flags & IOMAP_WRITE))
···
 	return generic_block_bmap(mapping,block,ext2_get_block);
 }
 
-static ssize_t
-ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
-{
-	struct file *file = iocb->ki_filp;
-	struct address_space *mapping = file->f_mapping;
-	struct inode *inode = mapping->host;
-	size_t count = iov_iter_count(iter);
-	loff_t offset = iocb->ki_pos;
-	ssize_t ret;
-
-	ret = blockdev_direct_IO(iocb, inode, iter, ext2_get_block);
-	if (ret < 0 && iov_iter_rw(iter) == WRITE)
-		ext2_write_failed(mapping, offset + count);
-	return ret;
-}
-
 static int
 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
 {
···
 	.write_begin		= ext2_write_begin,
 	.write_end		= ext2_write_end,
 	.bmap			= ext2_bmap,
-	.direct_IO		= ext2_direct_IO,
+	.direct_IO		= noop_direct_IO,
 	.writepages		= ext2_writepages,
 	.migrate_folio		= buffer_migrate_folio,
 	.is_partially_uptodate	= block_is_partially_uptodate,