udf: Fix races with i_size changes during readpage

__udf_adinicb_readpage() uses i_size several times. When truncate
changes i_size while the function is running, it can observe several
different values and thus e.g. expose uninitialized parts of page to
userspace. Also use i_size_read() in the function since it does not hold
inode_lock. Since i_size is guaranteed to be small, this cannot really
cause any issues even on 32-bit archs but let's be careful.

CC: stable@vger.kernel.org
Fixes: 9c2fc0de1a6e638fe58c354a463f544f42a90a09
Signed-off-by: Jan Kara <jack@suse.cz>

diff --git a/fs/udf/file.c b/fs/udf/file.c
index f5eb2d5b3bac86d91eaaddddbec6237dbbadfa38..e06d2c15749a50027757eef97c667e2352e29714 100644 (file)
--- a/fs/udf/file.c
+++ b/fs/udf/file.c
@@ -43,10 +43,15 @@ static void __udf_adinicb_readpage(struct page *page)
        struct inode *inode = page->mapping->host;
        char *kaddr;
        struct udf_inode_info *iinfo = UDF_I(inode);
+       loff_t isize = i_size_read(inode);
 
+       /*
+        * We have to be careful here as truncate can change i_size under us.
+        * So just sample it once and use the same value everywhere.
+        */
        kaddr = kmap_atomic(page);
-       memcpy(kaddr, iinfo->i_ext.i_data + iinfo->i_lenEAttr, inode->i_size);
-       memset(kaddr + inode->i_size, 0, PAGE_SIZE - inode->i_size);
+       memcpy(kaddr, iinfo->i_ext.i_data + iinfo->i_lenEAttr, isize);
+       memset(kaddr + isize, 0, PAGE_SIZE - isize);
        flush_dcache_page(page);
        SetPageUptodate(page);
        kunmap_atomic(kaddr);