On Wed 20-06-18 16:15:03, Ross Zwisler wrote:
> diff --git a/fs/ext4/truncate.h b/fs/ext4/truncate.h
> index 0cb13badf473..a3b78241e9f6 100644
> --- a/fs/ext4/truncate.h
> +++ b/fs/ext4/truncate.h
> @@ -12,6 +12,7 @@
> static inline void ext4_truncate_failed_write(struct inode *inode)
> {
> down_write(&EXT4_I(inode)->i_mmap_sem);
> + ext4_break_layouts(inode);
> truncate_inode_pages(inode->i_mapping, inode->i_size);
> ext4_truncate(inode);
> up_write(&EXT4_I(inode)->i_mmap_sem);
One comment here: I don't think ext4_break_layouts() is necessary here.
ext4_truncate_failed_write() exists to truncate blocks beyond EOF. As such
these blocks could never be even mmaped, let alone pinned by GUP. Maybe
just add a comment that blocks we are truncating here were never visible to
userspace and so we don't need ext4_break_layouts() protection.
Honza
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR
