Re: [PATCH] affs: add kunmap in error path

[Al Viro]

On Tue, Feb 02, 2016 at 02:24:36AM +0000, Al Viro wrote:
> On Mon, Feb 01, 2016 at 08:55:17PM -0500, Insu Yun wrote:
> > When error occurs, it needs to unmap page.
> 
> Point, but...  I'm not sure there's any point in keeping it mapped all
> along.  After all, what we are doing is a bunch of memcpy() into the
> parts of page, so kmap_atomic()/kunmap_atomic() just around memcpy
> would do just fine.

IOW, something like this (completely untested):

diff --git a/fs/affs/file.c b/fs/affs/file.c
index 0548c53..22fc7c8 100644
--- a/fs/affs/file.c
+++ b/fs/affs/file.c
@@ -511,8 +511,6 @@ affs_do_readpage_ofs(struct page *page, unsigned to)
        pr_debug("%s(%lu, %ld, 0, %d)\n", __func__, inode->i_ino,
                 page->index, to);
        BUG_ON(to > PAGE_CACHE_SIZE);
-       kmap(page);
-       data = page_address(page);
        bsize = AFFS_SB(sb)->s_data_blksize;
        tmp = page->index << PAGE_CACHE_SHIFT;
        bidx = tmp / bsize;
@@ -524,14 +522,15 @@ affs_do_readpage_ofs(struct page *page, unsigned to)
                        return PTR_ERR(bh);
                tmp = min(bsize - boff, to - pos);
                BUG_ON(pos + tmp > to || tmp > bsize);
+               data = kmap_atomic(page);
                memcpy(data + pos, AFFS_DATA(bh) + boff, tmp);
+               kunmap_atomic(data);
                affs_brelse(bh);
                bidx++;
                pos += tmp;
                boff = 0;
        }
        flush_dcache_page(page);
-       kunmap(page);
        return 0;
 }