Re: [PATCH v4 5/5] fs: unicode: Add utf8 module and a unicode layer

From: Shreeya Patel <shreeya.patel@collabora.com>
To: Eric Biggers <ebiggers@kernel.org>
Cc: tytso@mit.edu, adilger.kernel@dilger.ca, jaegeuk@kernel.org,
	chao@kernel.org, krisman@collabora.com, drosen@google.com,
	yuchao0@huawei.com, linux-ext4@vger.kernel.org,
	linux-kernel@vger.kernel.org,
	linux-f2fs-devel@lists.sourceforge.net,
	linux-fsdevel@vger.kernel.org, kernel@collabora.com,
	andre.almeida@collabora.com
Subject: Re: [PATCH v4 5/5] fs: unicode: Add utf8 module and a unicode layer
Date: Fri, 26 Mar 2021 01:56:00 +0530	[thread overview]
Message-ID: <2db48ab8-1297-e044-dcec-6c8b8875fdb0@collabora.com> (raw)
In-Reply-To: <YFznIVf/F68oEuC6@sol.localdomain>

On 26/03/21 1:10 am, Eric Biggers wrote:
> On Thu, Mar 25, 2021 at 05:38:11AM +0530, Shreeya Patel wrote:
>> Also, indirect calls using function pointers are easily exploitable by
>> speculative execution attacks, hence use static_call() in unicode.h and
>> unicode-core.c files inorder to prevent these attacks by making direct
>> calls and also to improve the performance of function pointers.
> I don't think you need to worry about avoiding indirect calls to prevent
> speculative execution attacks.  That's what the mitigations like Retpoline are
> for.  Instead my concern was just that indirect calls are *slow*, especially
> when those mitigations are enabled.  Some of the casefolding operations are
> called a lot (e.g., repeatedly during path resolution), and it would be
> desirable to avoid adding more overhead there.
>
>> diff --git a/fs/unicode/Kconfig b/fs/unicode/Kconfig
>> index 2c27b9a5cd6c..2961b0206b4d 100644
>> --- a/fs/unicode/Kconfig
>> +++ b/fs/unicode/Kconfig
>> @@ -8,7 +8,16 @@ config UNICODE
>>   	  Say Y here to enable UTF-8 NFD normalization and NFD+CF casefolding
>>   	  support.
>>   
>> +# UTF-8 encoding can be compiled as a module using UNICODE_UTF8 option.
>> +# Having UTF-8 encoding as a module will avoid carrying large
>> +# database table present in utf8data.h_shipped into the kernel
>> +# by being able to load it only when it is required by the filesystem.
>> +config UNICODE_UTF8
>> +	tristate "UTF-8 module"
>> +	depends on UNICODE
>> +	default m
>> +
> The help for UNICODE still says that it enables UTF-8 support.  But now there is
> a separate option that people will need to remember to enable.
>
> Please document each of these options properly.
>
> Perhaps EXT4_FS and F2FS_FS just should select UNICODE_UTF8 if UNICODE, so that
> UNICODE_UTF8 doesn't have to be a user-selectable symbol?

It is not a user-selectable symbol. It depends on UNICODE and if someone 
enables it,
by default UNICODE_UTF8 will be enabled as a module.

>> +DEFINE_STATIC_CALL(validate, unicode_validate_static_call);
>> +EXPORT_STATIC_CALL(validate);
> Global symbols can't have generic names like "validate".  Please add an
> appropriate prefix like "unicode_".
>
> Also, the thing called "unicode_validate_static_call" isn't actually a static
> call as the name suggests, but rather the default function used by the static
> call.  It should be called something like unicode_validate_default.
>
> Likewise for all the others.

Thanks for your reviews, I'll make the change suggested by you in v5.

>
> - Eric
>