xref: /third_party/lwip/0056-fix-tso-small-packet-drop-in-kernel-server.patch

From abeef0770f76cd0eff8e5c6e50de0b280079d7f0 Mon Sep 17 00:00:00 2001
From: jiangheng <jiangheng14@huawei.com>
Date: Mon, 13 Mar 2023 19:25:42 +0800
Subject: [PATCH] fix tso small packet drop in kernel server

---
 src/core/tcp_out.c     | 254 +++++++++++++++++++++--------------------
 src/include/lwipopts.h |   2 +
 2 files changed, 130 insertions(+), 126 deletions(-)

diff --git a/src/core/tcp_out.c b/src/core/tcp_out.c
index 8a0d653..b1c317d 100644
--- a/src/core/tcp_out.c
+++ b/src/core/tcp_out.c
@@ -1312,60 +1312,33 @@ tcp_build_wnd_scale_option(u32_t *opts)
 #endif
 
 #if GAZELLE_ENABLE
-static struct tcp_seg *tcp_output_over(struct tcp_pcb *pcb, struct tcp_seg *seg, struct tcp_seg *useg)
-{
-  if (TCP_TCPLEN(seg) > 0) {
-    seg->next = NULL;
-    if (useg == NULL) {
-      pcb->unacked = seg;
-      pcb->last_unacked = seg;
-      useg = seg;
-    } else {
-      if (TCP_SEQ_LT(lwip_ntohl(seg->tcphdr->seqno), lwip_ntohl(useg->tcphdr->seqno))) {
-        /* add segment to before tail of unacked list, keeping the list sorted */
-        struct tcp_seg **cur_seg = &(pcb->unacked);
-        while (*cur_seg &&
-              TCP_SEQ_LT(lwip_ntohl((*cur_seg)->tcphdr->seqno), lwip_ntohl(seg->tcphdr->seqno))) {
-          cur_seg = &((*cur_seg)->next );
-        }
-        seg->next = (*cur_seg);
-        (*cur_seg) = seg;
-      } else {
-        /* add segment to tail of unacked list */
-        useg->next = seg;
-        useg = seg;
-        pcb->last_unacked = seg;
-      }
-    }
-  } else {
-    tcp_seg_free(seg);
-  }
-
-  return useg;
-}
-static err_t tcp_output_seg(struct tcp_pcb *pcb, struct tcp_seg *seg, struct netif *netif, u32_t snd_nxt)
-{
-  if (pcb->state != SYN_SENT) {
-    TCPH_SET_FLAG(seg->tcphdr, TCP_ACK);
-  }
-
-  err_t err = tcp_output_segment(seg, pcb, netif);
-  if (err != ERR_OK) {
-    /* segment could not be sent, for whatever reason */
-    tcp_set_flags(pcb, TF_NAGLEMEMERR);
-    return err;
-  }
-
-  if (pcb->state != SYN_SENT) {
-    tcp_clear_flags(pcb, TF_ACK_DELAY | TF_ACK_NOW);
-  }
-
-  if (TCP_SEQ_LT(pcb->snd_nxt, snd_nxt)) {
-      pcb->snd_nxt = snd_nxt;
-  }
-
-  return ERR_OK;
-}
+u32_t start_seqno = 0;
+#define TCP_INIT_SEGMENT(tem_seg, _pcb, _p, _hdrflags, _seqno, _optflags) \
+do { \
+  struct tcp_seg *_seg = tem_seg; \
+  u8_t _optlen; \
+  rte_prefetch2(_p); \
+ \
+  _optlen = LWIP_TCP_OPT_LENGTH_SEGMENT(_optflags, _pcb); \
+  _seg->flags = _optflags; \
+  _seg->next = NULL; \
+  _seg->p = _p; \
+  _seg->len = _p->tot_len - _optlen; \
+  /* build TCP header */ \
+  pbuf_add_header(_p, TCP_HLEN); \
+  _seg->tcphdr = (struct tcp_hdr *)_seg->p->payload; \
+  _seg->tcphdr->src = lwip_htons(_pcb->local_port); \
+  _seg->tcphdr->dest = lwip_htons(_pcb->remote_port); \
+ /* _seg->tcphdr->src = lwip_htons(_pcb->local_port); \ */ \
+ /* _seg->tcphdr->dest = lwip_htons(_pcb->remote_port); \ */ \
+  _seg->tcphdr->seqno = lwip_htonl(_seqno); \
+                                          \
+  if (start_seqno == 0) {\
+      start_seqno = _seqno; \
+  } \
+  TCPH_HDRLEN_FLAGS_SET(_seg->tcphdr, (5 + _optlen / 4), _hdrflags); \
+  _seg->tcphdr->urgp = 0; \
+} while(0)
 #endif
 /**
  * @ingroup tcp_raw
@@ -1471,97 +1444,127 @@ tcp_output(struct tcp_pcb *pcb)
   pcb->persist_backoff = 0;
 
   /* useg should point to last segment on unacked queue */
-  useg = pcb->last_unacked;
+  useg = pcb->unacked;
+  if (useg != NULL) {
+    for (; useg->next != NULL; useg = useg->next);
+  }
 
   /* data available and window allows it to be sent? */
-
-  u32_t send_len = 0;
 #if GAZELLE_ENABLE
   if ((get_eth_params_tx_ol() & DEV_TX_OFFLOAD_TCP_TSO) && pcb->need_tso_send) {
-    while(seg && send_len < 0xffff) {
-      /**
-       * 1) walk unsent queue, find all seg witch wait to send. chain buf in these segs.
-       * 2) create new segment, send and free new segment.
-       * 3) update snd_nxt, unacked queue, and unsent queue
-       */
-      struct tcp_seg *start_seg = seg;
-      struct pbuf *first_pbuf = NULL;
-      struct pbuf *pre_pbuf = NULL;
-      u8_t pbuf_chain_len = 0;
-      u32_t next_seqno = lwip_ntohl(seg->tcphdr->seqno);
-      while (seg != NULL && pbuf_chain_len < GAZELLE_TCP_MAX_PBUF_CHAIN_LEN) {
+     uint16_t send_pkt = 0;
+
+     do {
+        struct tcp_seg * start_seg = seg;
+        struct pbuf *new_pbuf = NULL;
+
+        struct pbuf *tmp_pbuf = NULL;
         u32_t seg_seqno = lwip_ntohl(seg->tcphdr->seqno);
-        if (seg_seqno - pcb->lastack + seg->len > wnd) {
-          if (first_pbuf)
-            break;
-          else
-            goto output_done;
+        u32_t last_seg_seqno = seg_seqno;
+
+        struct tcp_seg *last_seg = NULL;
+        u16_t last_seg_len = 0;
+        u8_t pbuf_chain_len = 0;
+        while (seg != NULL && seg_seqno - pcb->lastack + seg->len <= wnd && pbuf_chain_len < GAZELLE_TCP_MAX_PBUF_CHAIN_LEN) {
+            if (last_seg_len != 0 && (last_seg_len + seg->len < 1460) && seg->len < GAZELLE_TCP_MIN_TSO_SEG_LEN) {
+                break;
+            }
+
+            if ((tcp_do_output_nagle(pcb) == 0) &&
+                ((pcb->flags & (TF_NAGLEMEMERR | TF_FIN)) == 0)) {
+                break;
+            }
+            if (last_seg_seqno + last_seg_len == seg_seqno) {
+                pbuf_remove_header(seg->p, seg->p->tot_len - seg->len);
+                if (new_pbuf == NULL) {
+                    new_pbuf = seg->p;
+                    tmp_pbuf = new_pbuf;
+                } else {
+                    new_pbuf->tot_len += seg->p->len;
+                    tmp_pbuf->next = seg->p;
+                    tmp_pbuf = tmp_pbuf->next;
+                }
+            } else {
+                break;
+            }
+
+            last_seg = seg;
+            last_seg_len = seg->len;
+            last_seg_seqno = seg_seqno;
+            seg = seg->next;
+            seg_seqno = (seg != NULL) ? lwip_ntohl(seg->tcphdr->seqno) : seg_seqno;
+            pbuf_chain_len++;
         }
 
-        if ((tcp_do_output_nagle(pcb) == 0) && ((pcb->flags & (TF_NAGLEMEMERR | TF_FIN)) == 0)) {
-          if (first_pbuf)
-            break;
-          else
-            goto output_done;
+        // tcp_do_output_nagle, break
+        if (new_pbuf == NULL) {
+            goto end_loop;
         }
 
-        if (seg->len < TCP_MSS || next_seqno != seg_seqno || pbuf_chain_len >= GAZELLE_TCP_MAX_PBUF_CHAIN_LEN) {
-          break;
-        }
-        if (first_pbuf == NULL && (seg->next == NULL || seg->next->len < TCP_MSS)) {
-          break;
-        }
+        struct tcp_seg new_seg;
+        TCP_INIT_SEGMENT(&new_seg, pcb, new_pbuf, 0, lwip_ntohl(start_seg->tcphdr->seqno), 0);
 
-        pbuf_remove_header(seg->p, seg->p->tot_len - seg->len);
-        if (first_pbuf == NULL) {
-          first_pbuf = seg->p;
-        } else {
-          first_pbuf->tot_len += seg->p->len;
-          pre_pbuf->next = seg->p;
+        if (pcb->state != SYN_SENT) {
+              TCPH_SET_FLAG(new_seg.tcphdr, TCP_ACK);
         }
 
-        send_len += seg->len;
-        pre_pbuf = seg->p;
-        next_seqno = seg_seqno + TCP_TCPLEN(seg);
-        seg = seg->next;
-        pcb->unsent = seg;
-        pbuf_chain_len++;
-      }
-
-      if (first_pbuf == NULL) {
-        err = tcp_output_seg(pcb, seg, netif, next_seqno + seg->len);
+        err = tcp_output_segment(&new_seg, pcb, netif);
         if (err != ERR_OK) {
-          if (pcb->unsent == NULL)
-            pcb->last_unsent = NULL;
-	  pcb->need_tso_send = 0;
-          return err;
+            /* segment could not be sent, for whatever reason */
+            tcp_set_flags(pcb, TF_NAGLEMEMERR);
+            return err;
         }
-        pcb->unsent = seg->next;
-        useg = tcp_output_over(pcb, seg, useg);
-        seg = pcb->unsent;
-        continue;
-      }
-
-      struct tcp_seg new_seg;
-      tcp_init_segment(&new_seg, pcb, first_pbuf, 0, lwip_ntohl(start_seg->tcphdr->seqno), 0);
 
-      err = tcp_output_seg(pcb, &new_seg, netif, next_seqno);
+        pcb->unsent = last_seg->next;
+        if (pcb->state != SYN_SENT) {
+            tcp_clear_flags(pcb, TF_ACK_DELAY | TF_ACK_NOW);
+        }
 
-      for (u32_t i = 0; i < pbuf_chain_len; i++) {
-        struct tcp_seg *next_seg = start_seg->next;
-        start_seg->p->next = NULL;
-        useg = tcp_output_over(pcb, start_seg, useg);
-        start_seg = next_seg;
-      }
+        snd_nxt = last_seg_seqno + TCP_TCPLEN(last_seg);
+        if (TCP_SEQ_LT(pcb->snd_nxt, snd_nxt)) {
+            pcb->snd_nxt = snd_nxt;
+        }
 
-      pbuf_remove_header(new_seg.p, new_seg.p->tot_len - new_seg.len - TCPH_HDRLEN_BYTES(new_seg.tcphdr));
-      new_seg.p->tot_len = new_seg.p->len;
-    }
-  pcb->need_tso_send = 0;
+        pbuf_remove_header(new_seg.p, new_seg.p->tot_len - new_seg.len - TCP_HLEN);
+        new_seg.p->tot_len = new_seg.p->len;
+
+        for (int start = pbuf_chain_len; start > 0; start--) {
+            struct tcp_seg *tmp_seg = start_seg;
+            start_seg = start_seg->next;
+            tmp_seg->p->next = NULL;
+            if (TCP_TCPLEN(tmp_seg) > 0) {
+                tmp_seg->next = NULL;
+                if (pcb->unacked == NULL) {
+                    pcb->unacked = tmp_seg;
+                    useg = tmp_seg;
+                } else {
+                    if (TCP_SEQ_LT(lwip_ntohl(tmp_seg->tcphdr->seqno), lwip_ntohl(useg->tcphdr->seqno))) {
+                        /* add segment to before tail of unacked list, keeping the list sorted */
+                        struct tcp_seg **cur_seg = &(pcb->unacked);
+                        while (*cur_seg &&
+                            TCP_SEQ_LT(lwip_ntohl((*cur_seg)->tcphdr->seqno), lwip_ntohl(tmp_seg->tcphdr->seqno)))     {
+                            cur_seg = &((*cur_seg)->next );
+                        }
+                        tmp_seg->next = (*cur_seg);
+                        (*cur_seg) = tmp_seg;
+                    } else {
+                        /* add segment to tail of unacked list */
+                        useg->next = tmp_seg;
+                        useg = useg->next;
+                    }
+                }
+            } else {
+              tcp_seg_free(tmp_seg);
+            }
+        }
+     } while(seg != NULL && lwip_ntohl(seg->tcphdr->seqno) - pcb->lastack + seg->len <= wnd && send_pkt++ < 10);
+end_loop:
+    pcb->need_tso_send = 0;
   } else
 #endif
 {
-  while (seg != NULL && send_len < 0xffff &&
+  uint16_t send_pkt = 0;
+  while (seg != NULL && send_pkt++ < 10 &&
          lwip_ntohl(seg->tcphdr->seqno) - pcb->lastack + seg->len <= wnd) {
     LWIP_ASSERT("RST not expected here!",
                 (TCPH_FLAGS(seg->tcphdr) & TCP_RST) == 0);
@@ -1576,7 +1579,6 @@ tcp_output(struct tcp_pcb *pcb)
         ((pcb->flags & (TF_NAGLEMEMERR | TF_FIN)) == 0)) {
       break;
     }
-    send_len += seg->len;
 #if TCP_CWND_DEBUG
     LWIP_DEBUGF(TCP_CWND_DEBUG, ("tcp_output: snd_wnd %"TCPWNDSIZE_F", cwnd %"TCPWNDSIZE_F", wnd %"U32_F", effwnd %"U32_F", seq %"U32_F", ack %"U32_F", i %"S16_F"\n",
                                  pcb->snd_wnd, pcb->cwnd, wnd,
diff --git a/src/include/lwipopts.h b/src/include/lwipopts.h
index 742b4a9..0d2a6d9 100644
--- a/src/include/lwipopts.h
+++ b/src/include/lwipopts.h
@@ -55,6 +55,8 @@
 
 #define GAZELLE_TCP_MAX_PBUF_CHAIN_LEN 40
 
+#define GAZELLE_TCP_MIN_TSO_SEG_LEN 256
+
 /*
    ----------------------------------
    ---------- NIC offloads ----------
-- 
2.33.0