Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <595e0169-fe87-0477-fbe4-ff89387605df@linaro.org>
Date:   Mon, 18 Jul 2022 12:40:53 +0100
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.11.0
Subject: Re: [PATCH -next] nvmem: core: Fix memleak in nvmem_register()
Content-Language: en-US
To:     Greg KH <gregkh@linuxfoundation.org>,
        Gaosheng Cui <cuigaosheng1@huawei.com>
Cc:     linux-kernel@vger.kernel.org, gongruiqi1@huawei.com,
        wangweiyang2@huawei.com
References: <20220716075352.2622809-1-cuigaosheng1@huawei.com>
 <YtJ1mthCP+4laqOn@kroah.com>
From:   Srinivas Kandagatla <srinivas.kandagatla@linaro.org>
In-Reply-To: <YtJ1mthCP+4laqOn@kroah.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 16/07/2022 09:23, Greg KH wrote:
> On Sat, Jul 16, 2022 at 03:53:52PM +0800, Gaosheng Cui wrote:
>> dev_set_name will alloc memory for nvmem->dev.kobj.name in
>> nvmem_register, when nvmem_validate_keepouts failed, nvmem's
>> memory will be freed and return, but nobody will free memory
>> for nvmem->dev.kobj.name, there will be memleak, so using
>> kfree_const(nvmem->dev.kobj.name) to fix it.
>>
>> Fixes: de0534df9347 ("nvmem: core: fix error handling while validating keepout regions")
>> Signed-off-by: Gaosheng Cui <cuigaosheng1@huawei.com>
>> ---
>>   drivers/nvmem/core.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
>> index 1e3c754efd0d..6c75c9afa5f3 100644
>> --- a/drivers/nvmem/core.c
>> +++ b/drivers/nvmem/core.c
>> @@ -833,6 +833,7 @@ struct nvmem_device *nvmem_register(const struct nvmem_config *config)
>>   		rval = nvmem_validate_keepouts(nvmem);
>>   		if (rval) {
>>   			ida_free(&nvmem_ida, nvmem->id);
>> +			kfree_const(nvmem->dev.kobj.name);
>>   			kfree(nvmem);
> 
> Something is really wrong, you should never be touching the name pointer
> of a kobject directly like this.  Also the device structure itself
> should be cleaning up the memory, not a kfree.  So this feels wrong...

The problem is that at this point we have not device_intialize() the dev 
yet, so there is no refcount setup for kobject yet.

I see two drivers (drivers/nvme/core.c and drivers/base/node.c)  doing 
something similar.

Most common (99.99%) usage pattern is

dev_set_name(...)
ret = device_register(..);
if (ret < 0)
	put_device(..)

but in this particular case we have:

dev_set_name(...)
..
ret = do_some_error_checks()
if (ret < 0)
	return ret;
...
ret = device_register(..);
if (ret < 0)
	put_device(..)

One solution is to follow the most used pattern and move the checks 
after device register and let device core take care of freeing this 
kobj.name as part of its cleanup routine.

--srini

> 
> greg k-h