Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: f2fs_convert_inline_inode causing rebalance based on random
 uninitialized value in dn.node_changed
To:     Jaegeuk Kim <jaegeuk@kernel.org>
Cc:     Changman Lee <cm224.lee@samsung.com>,
        Chao Yu <chao2.yu@samsung.com>,
        linux-f2fs-devel@lists.sourceforge.net,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
References: <9fcca081-9a60-8ae3-5cac-d8aa38c38ff2@canonical.com>
 <YD/nFt6Gswnyogfa@google.com>
From:   Colin Ian King <colin.king@canonical.com>
Message-ID: <9b586bbb-bb94-6fdf-c9a4-9415dbc6d8d0@canonical.com>
Date:   Wed, 3 Mar 2021 22:59:27 +0000
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.8.0
MIME-Version: 1.0
In-Reply-To: <YD/nFt6Gswnyogfa@google.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 03/03/2021 19:44, Jaegeuk Kim wrote:
> On 03/02, Colin Ian King wrote:
>> 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@kernel.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?
> 
> Thank you for the report. Yes, it seems that's a right call and we need to
> check the error to decide calling f2fs_balance_fs() in line 222, since
> set_new_dnode() is used to set all the fields in dnode_of_data. So, if you
> don't mind, could you please post a patch?

Just to clarify, just setting dn.node_changes to false is enough?

I'm not entirely sure what you meant when you wrote "and we need to
check the error to decide calling f2fs_balance_fs() in line 222".

Colin

> 
> Thanks,
> 
>>
>> Colin
>>