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: [PATCH 167/190] Revert "gdrom: fix a memory leak bug"
To:     Peter Rosin <peda@axentia.se>, Greg KH <gregkh@linuxfoundation.org>
Cc:     linux-kernel@vger.kernel.org
References: <20210421130105.1226686-168-gregkh@linuxfoundation.org>
 <23ffd3c5-0095-6a95-8c30-2b6f67377ce7@axentia.se>
 <14bdfbee-b8b9-d8ac-de34-b6d852d1485b@kernel.dk> <YIgLIscB1I8MxkWP@kroah.com>
 <5a55e8d4-e9b6-1854-51bc-afacd3ca097c@axentia.se>
From:   Jens Axboe <axboe@kernel.dk>
Message-ID: <e2ee4904-b13b-fd75-7066-d28e5a7ca711@kernel.dk>
Date:   Tue, 27 Apr 2021 08:39:15 -0600
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.10.0
MIME-Version: 1.0
In-Reply-To: <5a55e8d4-e9b6-1854-51bc-afacd3ca097c@axentia.se>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 4/27/21 8:03 AM, Peter Rosin wrote:
> On 2021-04-27 15:01, Greg KH wrote:
>> On Fri, Apr 23, 2021 at 08:20:30AM -0600, Jens Axboe wrote:
>>> On 4/22/21 3:29 PM, Peter Rosin wrote:
>>>>> This reverts commit 093c48213ee37c3c3ff1cf5ac1aa2a9d8bc66017.
>>>>
>>>> The reverted patch looks fishy.
>>>>
>>>> gc.cd_info is kzalloc:ed on probe. In case probe fails after this allocation, the
>>>> memory is kfree:d but the variable is NOT zeroed out.
>>>>
>>>> AFAICT, the above leads to a double-free on exit by the added line.
>>>>
>>>> I believe gd.cd_info should be kfree:d on remove instead.
>>>>
>>>> However, might not gc.toc also be kfree:d twice for similar reasons?
>>>>
>>>> I could easily be mistaken.
>>>
>>> >From taking a quick look the other day, that's my conclusion too. I
>>> don't think the patch is correct, but I don't think the surrounding code
>>> is correct right now either.
>>
>> Thanks for the review from both of you, I'll keep this commit in the
>> tree.
> Err, which commit is "this" and what tree are you keeping it in? I
> think you mean that you are keeping the revert in your tree with
> reverts, and not that you mean that we should keep the original
> commit in Linus' tree.
> 
> In any case, I'd think that the original memory leak is somewhat
> better than the introduced double-free and therefore the revert
> should be done.

It should probably look like the below, though I doubt it matters
since only one device is supported anyway. As long as the free
happens post unregister, it likely won't make a difference. But
it is cleaner and easier to verify, and should double device support
ever be introduced, the existing code is buggy.

But given that, I don't think we should keep the revert patch.

diff --git a/drivers/cdrom/gdrom.c b/drivers/cdrom/gdrom.c
index 9874fc1c815b..02d369881165 100644
--- a/drivers/cdrom/gdrom.c
+++ b/drivers/cdrom/gdrom.c
@@ -831,6 +831,8 @@ static int remove_gdrom(struct platform_device *devptr)
 	if (gdrom_major)
 		unregister_blkdev(gdrom_major, GDROM_DEV_NAME);
 	unregister_cdrom(gd.cd_info);
+	kfree(gd.toc);
+	kfree(gd.cd_info);
 
 	return 0;
 }
@@ -862,8 +864,6 @@ static void __exit exit_gdrom(void)
 {
 	platform_device_unregister(pd);
 	platform_driver_unregister(&gdrom_driver);
-	kfree(gd.toc);
-	kfree(gd.cd_info);
 }
 
 module_init(init_gdrom);

-- 
Jens Axboe