romfs: fix nommu map length to keep inside filesystem

Checks introduced in commit 4991e7251 ("romfs: do not use
mtd->get_unmapped_area directly") re-introduce problems fixed in the earlier
commit 2b4b2482e ("romfs: fix romfs_get_unmapped_area() argument check").

If a flat binary app is located at the end of a romfs, its page aligned
length may be outside of the romfs filesystem. The flat binary loader, via
nommu do_mmap_pgoff(), page aligns the length it is mmaping. So simple
offset+size checks will fail - returning EINVAL.

We can truncate the length to keep it inside the romfs filesystem, and that
also keeps the call to mtd_get_unmapped_area() happy.

Are there any side effects to truncating the size here though?

Signed-off-by: Greg Ungerer <gerg@uclinux.org>

diff --git a/fs/romfs/mmap-nommu.c b/fs/romfs/mmap-nommu.c
index e1a7779dd3cb18e72276569f8d2b0ce77e53a42c..f373bde8f545da481ba0a7caa873271dd599b30d 100644 (file)
--- a/fs/romfs/mmap-nommu.c
+++ b/fs/romfs/mmap-nommu.c
@@ -49,8 +49,11 @@ static unsigned long romfs_get_unmapped_area(struct file *file,
                return (unsigned long) -EINVAL;
 
        offset += ROMFS_I(inode)->i_dataoffset;
-       if (offset > mtd->size - len)
+       if (offset >= mtd->size)
                return (unsigned long) -EINVAL;
+       /* the mapping mustn't extend beyond the EOF */
+       if ((offset + len) > mtd->size)
+               len = mtd->size - offset;
 
        ret = mtd_get_unmapped_area(mtd, len, offset, flags);
        if (ret == -EOPNOTSUPP)