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: <1cec95d5-5cd4-fbf9-754b-e6a1229d45c3@huaweicloud.com>
Date:   Mon, 22 May 2023 16:58:25 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.10.0
Subject: Re: [PATCH RFC v2] Randomized slab caches for kmalloc()
Content-Language: en-US
To:     Hyeonggon Yoo <42.hyeyoo@gmail.com>
Cc:     linux-mm@kvack.org, linux-kernel@vger.kernel.org,
        linux-hardening@vger.kernel.org,
        Alexander Lobakin <aleksander.lobakin@intel.com>,
        kasan-dev@googlegroups.com, Wang Weiyang <wangweiyang2@huawei.com>,
        Xiu Jianfeng <xiujianfeng@huawei.com>,
        Vlastimil Babka <vbabka@suse.cz>,
        Christoph Lameter <cl@linux.com>,
        David Rientjes <rientjes@google.com>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Pekka Enberg <penberg@kernel.org>,
        Kees Cook <keescook@chromium.org>,
        Paul Moore <paul@paul-moore.com>,
        James Morris <jmorris@namei.org>,
        "Serge E. Hallyn" <serge@hallyn.com>,
        "Gustavo A. R. Silva" <gustavoars@kernel.org>,
        Gong Ruiqi <gongruiqi1@huawei.com>
References: <20230508075507.1720950-1-gongruiqi1@huawei.com>
 <CAB=+i9QxWL6ENDz_r1jPbiZsTUj1EE3u-j0uP6y_MxFSM9RerQ@mail.gmail.com>
 <5f5a858a-7017-5424-0fa0-db3b79e5d95e@huawei.com>
 <CAB=+i9R0GZiau7PKDSGdCOijPH1TVqA3rJ5tQLejJpoR55h6dg@mail.gmail.com>
 <19707cc6-fa5e-9835-f709-bc8568e4c9cd@huawei.com>
 <CAB=+i9T-iqtMZw8y7SxkaFBtiXA93YwFFEtQyGynBsorud1+_Q@mail.gmail.com>
From:   "GONG, Ruiqi" <gongruiqi@huaweicloud.com>
In-Reply-To: <CAB=+i9T-iqtMZw8y7SxkaFBtiXA93YwFFEtQyGynBsorud1+_Q@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 2023/05/22 16:03, Hyeonggon Yoo wrote:
> On Mon, May 22, 2023 at 4:35 PM Gong Ruiqi <gongruiqi1@huawei.com> wrote:
>> On 2023/05/17 6:35, Hyeonggon Yoo wrote:
> [...]
>>>>>> +#ifdef CONFIG_RANDOM_KMALLOC_CACHES
>>>>>> +# define SLAB_RANDOMSLAB       ((slab_flags_t __force)0x01000000U)
>>>>>> +#else
>>>>>> +# define SLAB_RANDOMSLAB       0
>>>>>> +#endif
>>>
>>> There is already the SLAB_KMALLOC flag that indicates if a cache is a
>>> kmalloc cache. I think that would be enough for preventing merging
>>> kmalloc caches?
>>
>> After digging into the code of slab merging (e.g. slab_unmergeable(),
>> find_mergeable(), SLAB_NEVER_MERGE, SLAB_MERGE_SAME etc), I haven't
>> found an existing mechanism that prevents normal kmalloc caches with
>> SLAB_KMALLOC from being merged with other slab caches. Maybe I missed
>> something?
>>
>> While SLAB_RANDOMSLAB, unlike SLAB_KMALLOC, is added into
>> SLAB_NEVER_MERGE, which explicitly indicates the no-merge policy.
> 
> I mean, why not make slab_unmergable()/find_mergeable() not to merge kmalloc
> caches when CONFIG_RANDOM_KMALLOC_CACHES is enabled, instead of a new flag?
> 
> Something like this:
> 
> diff --git a/mm/slab_common.c b/mm/slab_common.c
> index 607249785c07..13ac08e3e6a0 100644
> --- a/mm/slab_common.c
> +++ b/mm/slab_common.c
> @@ -140,6 +140,9 @@ int slab_unmergeable(struct kmem_cache *s)
>   if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
>   return 1;
> 
> + if (IS_ENALBED(CONFIG_RANDOM_KMALLOC_CACHES) && (flags & SLAB_KMALLOC))
> + return 1;
> +
>   if (s->ctor)
>   return 1;
> 
> @@ -176,6 +179,9 @@ struct kmem_cache *find_mergeable(unsigned int
> size, unsigned int align,
>   if (flags & SLAB_NEVER_MERGE)
>   return NULL;
> 
> + if (IS_ENALBED(CONFIG_RANDOM_KMALLOC_CACHES) && (flags & SLAB_KMALLOC))
> + return NULL;
> +
>   list_for_each_entry_reverse(s, &slab_caches, list) {
>   if (slab_unmergeable(s))
>   continue;

Ah I see. My concern is that it would affect not only normal kmalloc
caches, but kmalloc_{dma,cgroup,rcl} as well: since they were all marked
with SLAB_KMALLOC when being created, this code could potentially change
their mergeablity. I think it's better not to influence those irrelevant
caches.