Re: [PATCH] USB: gadget: f_mass_storage: fix in fsg_common_init() error recovery

["Michał Nazarewicz" <m.nazarewicz@samsung.com>]

>> fsg_common_init() as a first argument takes a pointer to a fsg_common
>> structure which, if not NULL, is reused and it is then assumed that
>> caller is responsible far maintaining allocation and deallocation of
>> this structure.
>>
>> The idea is that one can do:
>>
>> static struct fsg_common fsg_common;
>> /* ... */
>> fsg_common_init(&fsg_common, cdev, fsg_config);
>>
>> or allocate fsg_common structure as a part of a larger structure.  In such
>> cases kfree() cannot be called on the object.  Which case we are dealing
>> with is indicated by the common->free_storage_on_release flag.
>> fsg_common_release() consults it and either calls or does not call krfee().

On Wed, 12 May 2010 11:28:31 +0200, Viral Mehta <Viral.Mehta@lntinfotech.com> wrote:
> Oh, makes sense. All instances has fsg_common_init(0, ...) and so i just missed...

I'm preparing a some patches for g_multi which change that.

Also, a change for g_mass_storage which uses a static memory would be like 10 line
patch which I may submit later.

> But, still the central idea was, why should we go to whole error_release path which
> really does removing device file and closing luns and etc.
> However, it will not make any difference since curlun->nluns will be zero and so
> there will be no loop in fsg_common_releas() function.

The way I see it, it does not matter that much -- it's error recovery so we assume
it's unlikely to happen and as such speed optimisation is not really needed here --
it's better to optimise for space and minimise the number of possible paths.

> Apart from that, IMHO,
> 2894         kfree(common->luns);
> should crash in case if your error path is followed.
>
> common->luns is allocd
> 2712         /* Create the LUNs, open their backing files, and register the
> 2713          * LUN devices in sysfs. */
> 2714         curlun = kzalloc(nluns * sizeof *curlun, GFP_KERNEL);
> 2715         if (!curlun) {
> 2716                 kfree(common);
> 2717                 return ERR_PTR(-ENOMEM);
> 2718         }

Yes, that's why the submitted patch changes that to "rc = -ENOMEM;
goto errer_release;".

-- 
Best regards,                                        _     _
| Humble Liege of Serenely Enlightened Majesty of  o' \,=./ `o
| Computer Science,  Michał "mina86" Nazarewicz       (o o)
+----[mina86*mina86.com]---[mina86*jabber.org]----ooO--(_)--Ooo--