From: Eric Dumazet edumazet@google.com
[ Upstream commit cd171461b90a2d2cf230943df60d580174633718 ]
tcp_rcv_state_process() must tweak tp->advmss for TS enabled flows before the call to tcp_init_transfer() / tcp_init_buffer_space().
Otherwise tp->rcvq_space.space is off by 120 bytes (TCP_INIT_CWND * TCPOLEN_TSTAMP_ALIGNED).
Signed-off-by: Eric Dumazet edumazet@google.com Reviewed-by: Wei Wang weiwan@google.com Link: https://patch.msgid.link/20250513193919.1089692-7-edumazet@google.com Signed-off-by: Jakub Kicinski kuba@kernel.org Signed-off-by: Sasha Levin sashal@kernel.org ---
**YES** This commit should be backported to stable kernel trees. ## Detailed Analysis ### Nature of the Fix This commit addresses a **subtle but significant bug** in the TCP receive buffer space initialization for passive (server-side) connections when TCP timestamps are enabled. The fix is a simple 2-line reordering that moves the `advmss` adjustment for timestamp options **before** the call to `tcp_init_transfer()`. ### The Problem 1. **Root Cause**: In `tcp_rcv_state_process()` for `TCP_SYN_RECV` state, the `tp->advmss` reduction for timestamp options (`TCPOLEN_TSTAMP_ALIGNED = 12 bytes`) was happening **after** `tcp_init_transfer()` was called. 2. **Impact**: Since `tcp_init_transfer()` calls `tcp_init_buffer_space()`, which initializes `tp->rcvq_space.space` using the formula: ```c tp->rcvq_space.space = min3(tp->rcv_ssthresh, tp->rcv_wnd, (u32)TCP_INIT_CWND linux tp->advmss); ``` The calculation was using an **unadjusted `advmss` value**, leading to a 120-byte overestimate: - `TCP_INIT_CWND (10) × TCPOLEN_TSTAMP_ALIGNED (12) = 120 bytes` 3. **Consequence**: The `rcvq_space.space` field is critical for TCP receive buffer auto-tuning in `tcp_rcv_space_adjust()`, and this miscalculation could lead to suboptimal buffer management and performance issues. ### Why This Should Be Backported #### ✅ **Bug Fix Criteria Met**: 1. **Clear Bug**: This fixes a real initialization ordering bug that affects TCP performance 2. **User Impact**: Affects all passive TCP connections with timestamp options enabled (very common) 3. **Minimal Risk**: The fix is a simple 2-line reordering with no functional changes 4. **Contained Scope**: Only affects the initialization path in `tcp_rcv_state_process()` #### ✅ **Follows Stable Tree Rules**: 1. **Important**: TCP receive buffer tuning affects network performance for most connections 2. **Small & Contained**: The change moves just 2 lines of existing code 3. **No Regression Risk**: The fix corrects an obvious ordering error without introducing new logic 4. **No Architectural Changes**: No new features or major changes to TCP stack #### ✅ **Comparison with Similar Backported Commits**: The provided reference commits show a pattern of TCP receive buffer and `rcvq_space` related fixes being consistently backported: - Integer overflow fixes in `tcp_rcv_space_adjust()` ✅ **Backported** - Data race fixes for `sysctl_tcp_moderate_rcvbuf` ✅ **Backported** - MPTCP `rcv_space_init` fixes ✅ **Backported** - Timestamp handling in `tcp_rcv_space_adjust()` ✅ **Backported** This fix follows the exact same pattern: a small, surgical fix to TCP receive buffer initialization logic. ### Technical Verification Looking at the code change in `/home/sasha/linux/net/ipv4/tcp_input.c:6872-6873`, the fix simply moves: ```c if (tp->rx_opt.tstamp_ok) tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; ``` from **after** `tcp_init_transfer()` to **before** it. This ensures that when `tcp_init_buffer_space()` is called within `tcp_init_transfer()`, it uses the correct timestamp- adjusted `advmss` value. The change is **safe, targeted, and addresses a clear functional bug** that affects TCP performance for a large class of connections. It meets all criteria for stable tree backporting.
net/ipv4/tcp_input.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index 410a5b4189ea7..3757d35040a25 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -6625,6 +6625,9 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb) if (!tp->srtt_us) tcp_synack_rtt_meas(sk, req);
+ if (tp->rx_opt.tstamp_ok) + tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; + if (req) { tcp_rcv_synrecv_state_fastopen(sk); } else { @@ -6649,9 +6652,6 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb) tp->snd_wnd = ntohs(th->window) << tp->rx_opt.snd_wscale; tcp_init_wl(tp, TCP_SKB_CB(skb)->seq);
- if (tp->rx_opt.tstamp_ok) - tp->advmss -= TCPOLEN_TSTAMP_ALIGNED; - if (!inet_csk(sk)->icsk_ca_ops->cong_control) tcp_update_pacing_rate(sk);