linux-kernel - f2fs_convert_inline_inode causing rebalance based on random uninitialized value in dn.node

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <9fcca081-9a60-8ae3-5cac-d8aa38c38ff2@canonical.com>
Date:   Tue, 2 Mar 2021 20:07:08 +0000
From:   Colin Ian King <colin.king@...onical.com>
To:     Jaegeuk Kim <jaegeuk@...nel.org>,
        Changman Lee <cm224.lee@...sung.com>,
        Chao Yu <chao2.yu@...sung.com>,
        linux-f2fs-devel@...ts.sourceforge.net
Cc:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: f2fs_convert_inline_inode causing rebalance based on random
 uninitialized value in dn.node_changed

Hi,

Static analysis on linux-next detected a potential uninitialized
variable dn.node_changed that does not get set when a call to
f2fs_get_node_page() fails.  This uninitialized value gets used in the
call to f2fs_balance_fs() that may or not may not balances dirty node
and dentry pages depending on the uninitialized state of the variable.

I believe the issue was introduced by commit:

commit 2a3407607028f7c780f1c20faa4e922bf631d340
Author: Jaegeuk Kim <jaegeuk@...nel.org>
Date:   Tue Dec 22 13:23:35 2015 -0800

    f2fs: call f2fs_balance_fs only when node was changed

The analysis is a follows:

184 int f2fs_convert_inline_inode(struct inode *inode)
185 {
186        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);

   1. var_decl: Declaring variable dn without initializer.

187        struct dnode_of_data dn;

   NOTE dn is not initialized here.

188        struct page *ipage, *page;
189        int err = 0;
190

   2. Condition !f2fs_has_inline_data(inode), taking false branch.
   3. Condition f2fs_hw_is_readonly(sbi), taking false branch.
   4. Condition f2fs_readonly(sbi->sb), taking false branch.

191        if (!f2fs_has_inline_data(inode) ||
192                        f2fs_hw_is_readonly(sbi) ||
f2fs_readonly(sbi->sb))
193                return 0;
194
195        err = dquot_initialize(inode);

   5. Condition err, taking false branch.

196        if (err)
197                return err;
198
199        page = f2fs_grab_cache_page(inode->i_mapping, 0, false);

   6. Condition !page, taking false branch.

200        if (!page)
201                return -ENOMEM;
202
203        f2fs_lock_op(sbi);
204
205        ipage = f2fs_get_node_page(sbi, inode->i_ino);

   7. Condition IS_ERR(ipage), taking true branch.

206        if (IS_ERR(ipage)) {
207                err = PTR_ERR(ipage);

   8. Jumping to label out.

208                goto out;
209        }
210

   NOTE: set_new_dnode memset's dn so sets the flag to false, but we
don't get to this memset if IS_ERR(ipage) above is true.

211        set_new_dnode(&dn, inode, ipage, ipage, 0);
212
213        if (f2fs_has_inline_data(inode))
214                err = f2fs_convert_inline_page(&dn, page);
215
216        f2fs_put_dnode(&dn);
217 out:
218        f2fs_unlock_op(sbi);
219
220        f2fs_put_page(page, 1);
221

Uninitialized scalar variable:

   9. uninit_use_in_call: Using uninitialized value dn.node_changed when
calling f2fs_balance_fs.

222        f2fs_balance_fs(sbi, dn.node_changed);
223
224        return err;
225 }

I think a suitable fix will be to set dn.node_changed to false on in
line 207-208 but I'm concerned if I'm missing something subtle to the
rebalancing if I do this.

Comments?

Colin