RE: [PATCH bpf-next] bpf, test_run: fix alignment problem in bpf_test_init()

[John Fastabend <john.fastabend@gmail.com>]

Zhengchao Shao wrote:
> The problem reported by syz is as follows:
> BUG: KASAN: slab-out-of-bounds in __build_skb_around+0x230/0x330
> Write of size 32 at addr ffff88807ec6b2c0 by task bpf_repo/6711
> Call Trace:
> <TASK>
> dump_stack_lvl+0x8e/0xd1
> print_report+0x155/0x454
> kasan_report+0xba/0x1f0
> kasan_check_range+0x35/0x1b0
> memset+0x20/0x40
> __build_skb_around+0x230/0x330
> build_skb+0x4c/0x260
> bpf_prog_test_run_skb+0x2fc/0x1ce0
> __sys_bpf+0x1798/0x4b60
> __x64_sys_bpf+0x75/0xb0
> do_syscall_64+0x35/0x80
> entry_SYSCALL_64_after_hwframe+0x46/0xb0
> </TASK>
> 
> Allocated by task 6711:
> kasan_save_stack+0x1e/0x40
> kasan_set_track+0x21/0x30
> __kasan_kmalloc+0xa1/0xb0
> __kmalloc+0x4e/0xb0
> bpf_test_init.isra.0+0x77/0x100
> bpf_prog_test_run_skb+0x219/0x1ce0
> __sys_bpf+0x1798/0x4b60
> __x64_sys_bpf+0x75/0xb0
> do_syscall_64+0x35/0x80
> entry_SYSCALL_64_after_hwframe+0x46/0xb0
> 
> The process is as follows:
> bpf_prog_test_run_skb()
> 	bpf_test_init()
> 		data = kzalloc()	//The length of input is 576.
> 					//The actual allocated memory
> 					//size is 1024.
> 	build_skb()
> 		__build_skb_around()
> 			size = ksize(data)//size = 1024
> 			size -= SKB_DATA_ALIGN(
> 					sizeof(struct skb_shared_info));
> 					//size = 704
> 			skb_set_end_offset(skb, size);
> 			shinfo = skb_shinfo(skb);//shinfo = data + 704
> 			memset(shinfo...)	//Write out of bounds
> 
> In bpf_test_init(), the accessible space allocated to data is 576 bytes,
> and the memory allocated to data is 1024 bytes. In __build_skb_around(),
> shinfo indicates the offset of 704 bytes of data, which triggers the issue
> of writing out of bounds.
> 
> Fixes: 1cf1cae963c2 ("bpf: introduce BPF_PROG_TEST_RUN command")
> Reported-by: syzbot+fda18eaa8c12534ccb3b@syzkaller.appspotmail.com
> Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
> ---
>  net/bpf/test_run.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c
> index fcb3e6c5e03c..fbd5337b8f68 100644
> --- a/net/bpf/test_run.c
> +++ b/net/bpf/test_run.c
> @@ -766,6 +766,8 @@ static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
>  			   u32 size, u32 headroom, u32 tailroom)
>  {
>  	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
> +	unsigned int true_size;
> +	void *true_data;
>  	void *data;
>  
>  	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
> @@ -779,6 +781,14 @@ static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
>  	if (!data)
>  		return ERR_PTR(-ENOMEM);
>  
> +	true_size = ksize(data);
> +	if (size + headroom + tailroom < true_size) {
> +		true_data = krealloc(data, true_size, GFP_USER | __GFP_ZERO);

This comes from a kzalloc, should we zero realloc'd memory as well?

> +			if (!true_data)
> +				return ERR_PTR(-ENOMEM);

I think its worth fixing the extra tab here.

> +		data = true_data;
> +	}
> +
>  	if (copy_from_user(data + headroom, data_in, user_size)) {
>  		kfree(data);
>  		return ERR_PTR(-EFAULT);
> -- 
> 2.17.1
>