From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from mail-yx1-f66.google.com (mail-yx1-f66.google.com [74.125.224.66]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 8C4592BEFE8 for ; Mon, 18 May 2026 14:54:31 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=74.125.224.66 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1779116073; cv=none; b=RS3tdCbFjwXUnd8mlpJmDky2iQrscOPR41r8p1U5yylLFknj3krh3HoYyorDz3ELWnTlgf+C+k6xSGr0Bkuc2/ap8o8IH1J71Jy+KANtPNlxE8EcMLlrGYkHsftca9ppyLcey6RQyun6FbPkINotjJukFroDAlFHdk/ViNubJLw= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1779116073; c=relaxed/simple; bh=0hNpcKFFPUeeY8mvNc8d+Furck1M+6yn4y5f78ZljG4=; h=Date:From:To:Cc:Subject:Message-ID:References:MIME-Version: Content-Type:Content-Disposition:In-Reply-To; b=b9BSIa36hfa+dwrxATCjFzxMsGe+RK69RY3V1Ud1msphZqMrnjvjBOsdpOf4/sHIT22YYOk4D3f/V3z57ihh0WH41CbDWvsw1JEQcPadEvGFMBtpx4yqxQ1sdNfwuE2o5PGbaCbVKEr7pj6O8AJvrzPn+/UJqb2L7u2L1vanqBE= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=gmail.com; spf=pass smtp.mailfrom=gmail.com; dkim=pass (2048-bit key) header.d=gmail.com header.i=@gmail.com header.b=mLCFccEt; arc=none smtp.client-ip=74.125.224.66 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=gmail.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=gmail.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=gmail.com header.i=@gmail.com header.b="mLCFccEt" Received: by mail-yx1-f66.google.com with SMTP id 956f58d0204a3-65c52bb5dd7so2321521d50.2 for ; Mon, 18 May 2026 07:54:31 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20251104; t=1779116071; x=1779720871; darn=vger.kernel.org; h=in-reply-to:content-transfer-encoding:content-disposition :mime-version:references:message-id:subject:cc:to:from:date:from:to :cc:subject:date:message-id:reply-to; bh=c0dfQ7Wr8JZMajT0PBZM7edBivQd6TDRFDvTIsxG1qg=; b=mLCFccEtbfVYAyhGY6YYh7xqeUg1akP7dse6zMWJCu2L7YFFjJAqXYbmv7yDVbjdZ6 1ARZDVF9kLxQoVcJD9BzNvv38S7ZDRg4J03b439PGLoN77J56D2DG9GDjpQXPw9m7xb8 baKkW4Bzkrq4qniWEBH/IuA+INSXvCP8XzXWJOR8F8jkEdlCiEYzAI6DWxBrbRoqCDLC +cZS+VDQ05dqREcWckwmrnvuWZ0NcUSoSgVnOAbGmZyLodIAebNS6N6DgU4LnFkD9ppI uKYPi8ooDCmmtkHe/A9w+AIw1Y+9QPY0fOnKP7WW4xc7QfgaOSBkr2Dymd/cZZBDw8Yh DxFA== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20251104; t=1779116071; x=1779720871; h=in-reply-to:content-transfer-encoding:content-disposition :mime-version:references:message-id:subject:cc:to:from:date:x-gm-gg :x-gm-message-state:from:to:cc:subject:date:message-id:reply-to; bh=c0dfQ7Wr8JZMajT0PBZM7edBivQd6TDRFDvTIsxG1qg=; b=F3ZCaYRV0YypfCQ9TQjyLCaJ7uThraykzgs61dDkwOx0JK6C2rzYqt+DgHgUC9Ja4A 9N7Pgr7qUIug0xuFmHguWTzvxVnbW/AljbUwIQQqTNhNSGzm7QQ+9WBDc+ta0Hq6f4Js 0rOeLHpl3zMwM4NfAUOf5eoMc9dOL8RIfY+nWa/LA5ADTCID7NZA5OLkKwFKq5mvsuUB By2VgKarbseE2BkNnQIJPVjO+5Fg8sEouoXntOLCxrF527AxZVJB6BWmWxTLCNk/4beQ 96sGrDc+9wk/BJNCBcck53PjRT80zvRZWWnoEZQuTGmxDnVbmcAXPhSAHGn8jZ7J6mpd +Izw== X-Forwarded-Encrypted: i=1; AFNElJ8U4MqhKQl35ompWPaYJhi+EnvtfoFRx9JFJ1gp/V8ywwsyfpW2ouGym/sgn0ERr9mXKjsqSuA=@vger.kernel.org X-Gm-Message-State: AOJu0YzZS6i5X6lI0yGz7J/Iz0uIyFgGt2KjH9YaiM8+RAghP8bZz/qY 6zPUAhvpf8U9qeuhK/fFeP/Z3ngJKodI+h60joLextB3sNPXZ+4dFpjUPD3e0beq X-Gm-Gg: Acq92OEHdlWzZVr6w9N8cOxZw7MBKNxYAwKI6NIuPVvG2WbIFCwu0c6URtSmWNo1LG8 bGf+DetrMdPrkuTpOjzWDE6F6AdZ8kWNIFt3IYtwnkBddvNjL0LA9nLkvW/O/sIADS8JYQ0pAO8 0F5dfyGTZbL2sLJ9u8B9krb/GlIuaqi4Ty8XV5phoE/NSDjJYqHcASUIM6TEXEVJ6VQHsfdiss+ c+H1Q3pevfI7HPsKD0tokv7Zloa86rut1PgD14Hu3Y8kAPQtkx/GyN1CI6z9XwcpeNscwjRlMF3 rBeGorsvfefqMmdJvglzLicCottdOusoVC003bHhLPgWj3KqyCjgSj5nwVZ8pSigjX3XyvekNF9 HdyahoiJa20wCmXq9u1JKPNrB20yE9ENjpskIFHVTN+OfWmjkZBqeh4PmGu6Llt5Yqvsvl3yZsx DrSidVQvxD+XY6gzD9 X-Received: by 2002:a05:690e:c4c:b0:65e:3df3:1c30 with SMTP id 956f58d0204a3-65e3df3218fmr8960255d50.7.1779116070571; Mon, 18 May 2026 07:54:30 -0700 (PDT) Received: from localhost ([2a03:2880:2ff:5e::]) by smtp.gmail.com with ESMTPSA id 956f58d0204a3-65e0db0aa51sm6466044d50.12.2026.05.18.07.54.29 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 18 May 2026 07:54:30 -0700 (PDT) Date: Mon, 18 May 2026 07:54:29 -0700 From: Stanislav Fomichev To: Jason Xing Cc: davem@davemloft.net, edumazet@google.com, kuba@kernel.org, pabeni@redhat.com, bjorn@kernel.org, magnus.karlsson@intel.com, maciej.fijalkowski@intel.com, jonathan.lemon@gmail.com, sdf@fomichev.me, ast@kernel.org, daniel@iogearbox.net, hawk@kernel.org, john.fastabend@gmail.com, horms@kernel.org, andrew+netdev@lunn.ch, bpf@vger.kernel.org, netdev@vger.kernel.org, Jason Xing Subject: Re: [PATCH net 3/4] xsk: drain continuation descs after overflow in xsk_build_skb() Message-ID: References: <20260510012310.88570-1-kerneljasonxing@gmail.com> <20260510012310.88570-4-kerneljasonxing@gmail.com> Precedence: bulk X-Mailing-List: netdev@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: On 05/16, Jason Xing wrote: > On Fri, May 15, 2026 at 11:41 PM Stanislav Fomichev > wrote: > > > > On 05/15, Jason Xing wrote: > > > On Fri, May 15, 2026 at 8:29 AM Stanislav Fomichev wrote: > > > > > > > > On 05/14, Jason Xing wrote: > > > > > On Thu, May 14, 2026 at 12:27 AM Stanislav Fomichev > > > > > wrote: > > > > > > > > > > > > On 05/10, Jason Xing wrote: > > > > > > > From: Jason Xing > > > > > > > > > > > > > > When a multi-buffer packet exceeds MAX_SKB_FRAGS and triggers -EOVERFLOW, > > > > > > > only the current descriptor is released from the TX ring. The remaining > > > > > > > continuation descriptors of the same packet stay in the ring. Since > > > > > > > xs->skb is set to NULL after the drop, the TX loop picks up these > > > > > > > leftover frags and misinterprets each one as the beginning of a new > > > > > > > packet, corrupting the packet stream. > > > > > > > > > > > > > > Fix this by adding a drain_cont flag to xdp_sock. When overflow occurs > > > > > > > and the dropped descriptor has XDP_PKT_CONTD set, the flag is raised. > > > > > > > The main TX loop in __xsk_generic_xmit() then handles continuation > > > > > > > descriptors one at a time: each gets a normal CQ reservation (with > > > > > > > backpressure), its address is submitted to the completion queue, and > > > > > > > the descriptor is released from the TX ring. When the last fragment > > > > > > > (without XDP_PKT_CONTD) is processed, the flag is cleared and the > > > > > > > function returns -EOVERFLOW so the next call starts with a fresh > > > > > > > budget for normal packets. > > > > > > > > > > > > > > This reuses the existing CQ backpressure and budget mechanisms, so if > > > > > > > the CQ is full the function returns -EAGAIN and userspace drains the > > > > > > > CQ before retrying. Zero buffer leakage, zero packet stream corruption. > > > > > > > > > > > > > > Closes: https://lore.kernel.org/all/20260425041726.85FB3C2BCB2@smtp.kernel.org/ > > > > > > > Fixes: cf24f5a5feea ("xsk: add support for AF_XDP multi-buffer on Tx path") > > > > > > > Signed-off-by: Jason Xing > > > > > > > --- > > > > > > > include/net/xdp_sock.h | 1 + > > > > > > > net/xdp/xsk.c | 22 ++++++++++++++++++++++ > > > > > > > 2 files changed, 23 insertions(+) > > > > > > > > > > > > > > diff --git a/include/net/xdp_sock.h b/include/net/xdp_sock.h > > > > > > > index 23e8861e8b25..1958d19d9925 100644 > > > > > > > --- a/include/net/xdp_sock.h > > > > > > > +++ b/include/net/xdp_sock.h > > > > > > > @@ -80,6 +80,7 @@ struct xdp_sock { > > > > > > > * call of __xsk_generic_xmit(). > > > > > > > */ > > > > > > > struct sk_buff *skb; > > > > > > > + bool drain_cont; > > > > > > > > > > > > > > struct list_head map_list; > > > > > > > /* Protects map_list */ > > > > > > > diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c > > > > > > > index 3f1e590c855d..232dd7126905 100644 > > > > > > > --- a/net/xdp/xsk.c > > > > > > > +++ b/net/xdp/xsk.c > > > > > > > @@ -936,6 +936,8 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs, > > > > > > > xs->tx->invalid_descs++; > > > > > > > } > > > > > > > xskq_cons_release(xs->tx); > > > > > > > + if (xp_mb_desc(desc)) > > > > > > > + xs->drain_cont = true; > > > > > > > } else { > > > > > > > /* Let application retry */ > > > > > > > xsk_cq_cancel_locked(xs->pool, 1); > > > > > > > @@ -982,6 +984,26 @@ static int __xsk_generic_xmit(struct sock *sk) > > > > > > > goto out; > > > > > > > } > > > > > > > > > > > > > > + if (unlikely(xs->drain_cont)) { > > > > > > > + unsigned long flags; > > > > > > > + u32 idx; > > > > > > > + > > > > > > > > > > > > [..] > > > > > > > > > > > > > + spin_lock_irqsave(&xs->pool->cq_prod_lock, flags); > > > > > > > + idx = xskq_get_prod(xs->pool->cq); > > > > > > > + xskq_prod_write_addr(xs->pool->cq, idx, desc.addr); > > > > > > > + xskq_prod_submit_n(xs->pool->cq, 1); > > > > > > > + spin_unlock_irqrestore(&xs->pool->cq_prod_lock, flags); > > > > > > > > > > > > Not sure I understand why you want this if you're still marking the desc > > > > > > invalid. > > > > > > > > > > The key point is that as long as we read the desc from txq, the desc > > > > > should be either put back to txq again or publish it in the cq (for > > > > > application to keep track of this) at this point. Or else, the > > > > > application will lose track of this desc, which breaks the whole > > > > > logic. > > > > > > > > Yes, makes sense! > > > > > > > > > > > + > > > > > > > + xs->tx->invalid_descs++; > > > > > > > + xskq_cons_release(xs->tx); > > > > > > > + if (!xp_mb_desc(&desc)) { > > > > > > > + xs->drain_cont = false; > > > > > > > + err = -EOVERFLOW; > > > > > > > + goto out; > > > > > > > + } > > > > > > > > > > > > I also don't understand why you want to return -EOVERFLOW again? Why not > > > > > > (quietly) swallow these invalid xp_mb_desc from the previous packet and move > > > > > > on? > > > > > > > > > > This particular desc is really one of overflow cases, right? We should > > > > > warn users to handle this with this error code. > > > > > > > > Right, but didn't we already return -EOVERFLOW to the user once? > > > > The part where you set drain_count=true will -EOVERFLOW. Then this > > > > remainder will also return -EOVERFLOW? > > > > > > Right, my intention is to alert users twice under this kind of > > > circumstance. We silently drain the remaining portion of the skb the > > > second time, which looks a bit strange, doesn't it? > > > > > > > > > > > But let's maybe step back, what's the expectation on the user > > > > when we return -EOVERFLOW? In theory, the user can re-submit new > > > > shorter packet (at the current prod index), right? And then your > > > > !xp_mb_desc logic will break/swallow/-EOVERFLOW it? > > > > > > People would be aware of the skb that is too big to handle when > > > receiving two times of overflow warning. To put it in a simpler way, > > > the second time in xmit path only handling the remaining is enough, no > > > more descs that belong to another skb should be taken care of. > > > > > > If the user is able to quickly react to this case, I think the whole > > > skb should be re-put into the txq again instead of adding the > > > remaining part. I'm not so sure if I interpret the "break" correctly > > > here. > > > > Let's say the user puts a packet with too many descriptors. We find that > > mid-way and return -EOVERFLOW. What is user supposed to do with that error? > > In my mind, the user should drain TX ring completely and start posting > > shorter packets. You seem to be trying to handle the case where the user > > leaves the remainder of the previously EOVERFLOW'd packet in place, why? > > Good point. I think now I understand what your opinion is here: > diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c > index 3b604776f09f..72d05a1fb0b5 100644 > --- a/net/xdp/xsk.c > +++ b/net/xdp/xsk.c > @@ -996,11 +996,8 @@ static int __xsk_generic_xmit(struct sock *sk) > > xs->tx->invalid_descs++; > xskq_cons_release(xs->tx); > - if (!xp_mb_desc(&desc)) { > + if (!xp_mb_desc(&desc)) > xs->drain_cont = false; > - err = -EOVERFLOW; > - goto out; > - } > continue; > } > > Then as long as we have enough budget, we silently handle the whole > big skb and then continue with the rest of the skbs in txq. Users will > be alerted by the counter of invalid_descs. I think that is what > you're asking, right? Yes, this resolves one part. But the larger question still stands: why do this tracking at all? Why can't we clarify in the documentation that the user is responsible for flushing tx queue completely after EOVERFLOW? What am I missing?