Re: [PATCH bpf-next 2/2] selftests/bpf: Add tests for bpf_get_dentry_xattr

[Song Liu <songliubraving@meta.com>]

Hi Christian, 

Thanks a lot for your detailed explanation! We will revisit the design 
based on these comments and suggestions. 

One more question about a potential new kfunc bpf_get_inode_xattr(): 
Should it take dentry as input? IOW, should it look like:

__bpf_kfunc int bpf_get_inode_xattr(struct dentry *dentry, const char *name__str,
                                    struct bpf_dynptr *value_p)
{
        struct bpf_dynptr_kern *value_ptr = (struct bpf_dynptr_kern *)value_p;
        u32 value_len;
        void *value;
        int ret;

        if (strncmp(name__str, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN))
                return -EPERM;

        value_len = __bpf_dynptr_size(value_ptr);
        value = __bpf_dynptr_data_rw(value_ptr, value_len);
        if (!value)
                return -EINVAL;

        ret = inode_permission(&nop_mnt_idmap, dentry->d_inode, MAY_READ);
        if (ret)
                return ret;
        return __vfs_getxattr(dentry, dentry->d_inode, name__str, value, value_len);
}


I am asking because many security_inode_* hooks actually taking dentry as 
argument. So it makes sense to use dentry for kfuncs. Maybe we should
call it bpf_get_dentry_xattr, which is actually the same kfunc in this
set (1/2)?

Thanks,
Song



> On Jul 29, 2024, at 6:46 AM, Christian Brauner <brauner@kernel.org> wrote:
[...]
>>> Imho, what you're doing belongs into inode_permission() not into
>>> security_file_open(). That's already too late and it's somewhat clear
>>> from the example you're using that you're essentially doing permission
>>> checking during path lookup.
>> 
>> I am not sure I follow the suggestion to implement this with 
>> security_inode_permission()? Could you please share more details about
>> this idea?

[...]