Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20190331064129.31702-1-brianvv.kernel@gmail.com> <74e3024f-6fde-a04d-4000-89c6e61acc26@iogearbox.net>
In-Reply-To: <74e3024f-6fde-a04d-4000-89c6e61acc26@iogearbox.net>
From:   =?UTF-8?Q?Maciej_=C5=BBenczykowski?= <maze@google.com>
Date:   Mon, 1 Apr 2019 14:37:59 -0700
Message-ID: <CANP3RGdf9RWj82S6zCK4QnkMmoUiPRmsqyKw6xiUepNweNNQjA@mail.gmail.com>
Subject: Re: [PATCH] bpf: do not start from first bucket if elem is not found
 on a htab
To:     Daniel Borkmann <daniel@iogearbox.net>
Cc:     Brian Vazquez <brianvv.kernel@gmail.com>,
        Brian Vazquez <brianvv@google.com>,
        Alexei Starovoitov <ast@kernel.org>,
        "David S . Miller" <davem@davemloft.net>,
        Kernel hackers <linux-kernel@vger.kernel.org>,
        Linux NetDev <netdev@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

> This breaks BPF selftest suite unfortunately:
>
> # ./test_maps
> test_maps: test_maps.c:114: test_hashmap: Assertion `bpf_map_get_next_key(fd, &key, &next_key) == 0 && (next_key == first_key)' failed.
> Aborted
>
> Some more background, situation is a bit tricky: pre 8fe45924387b ("bpf:
> map_get_next_key to return first key on NULL") there was no reliable way
> of getting to the start of a hash table, meaning it needed an 'invalid'
> key to return the first element for starting traversal which commit fixed
> by being able to pass in NULL. So some applications might still just rely
> on e.g. key of zero bytes (if guaranteed to not be used otherwise) to do
> just that. With this patch, we'd start out from anywhere in the hash table.

sdf@ just mentioned this internally as well... I think we simply will
need to add a flags field
with at least two options/flags:
- return error if key not found
- resume from bucket instead of beginning if key not found
and while we're at it:
- return concatenated key+value instead of just key to save an
immediate lookup on the returned key