[PATCH] fs: fix libfs data leak

simple_prepare_write leaks uninitialised kernel data.  This happens because
the it leaves an uninitialised "hole" over the part of the page that the
write is expected to go to.  This is fine, but it then marks the page
uptodate, which means a concurrent read can come in and copy the
uninitialised memory into userspace before it written to.

Fix it by simply marking it uptodate in simple_commit_write instead, after
the hole has been filled in.  This could theoretically break an fs that
uses simple_prepare_write and not simple_commit_write, and that relies on
the incorrect simple_prepare_write behaviour.  Luckily, none of those
exists in the tree.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/Documentation/filesystems/vfs.txt b/Documentation/filesystems/vfs.txt
index 7737bfd03cf8481e97717a849fc6c83887b7a94e..ea271f2d395400b82cd8023db74fed76d878ce06 100644 (file)
--- a/Documentation/filesystems/vfs.txt
+++ b/Documentation/filesystems/vfs.txt
@@ -617,6 +617,11 @@ struct address_space_operations {
        In this case the prepare_write will be retried one the lock is
        regained.
 
+       Note: the page _must not_ be marked uptodate in this function
+       (or anywhere else) unless it actually is uptodate right now. As
+       soon as a page is marked uptodate, it is possible for a concurrent
+       read(2) to copy it to userspace.
+
   commit_write: If prepare_write succeeds, new data will be copied
         into the page and then commit_write will be called.  It will
         typically update the size of the file (if appropriate) and

diff --git a/fs/libfs.c b/fs/libfs.c
index 7d487047dbb8814ea1d5f0a01c9e30c630d9a322..cf79196535ec5fc277f2dc9f190a3cc7e551dcd6 100644 (file)
--- a/fs/libfs.c
+++ b/fs/libfs.c
@@ -335,17 +335,18 @@ int simple_prepare_write(struct file *file, struct page *page,
                        flush_dcache_page(page);
                        kunmap_atomic(kaddr, KM_USER0);
                }
-               SetPageUptodate(page);
        }
        return 0;
 }
 
 int simple_commit_write(struct file *file, struct page *page,
-                       unsigned offset, unsigned to)
+                       unsigned from, unsigned to)
 {
        struct inode *inode = page->mapping->host;
        loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
 
+       if (!PageUptodate(page))
+               SetPageUptodate(page);
        /*
         * No need to use i_size_read() here, the i_size
         * cannot change under us because we hold the i_mutex.