Merge pull request #928 from ywc689/toa-enhancement

ipvs: toa enhancements

patch/dcdn-toa.patch

@@ -1,32 +1,31 @@
-From 55e8e5da2b4b0893d36cb3f621bedf9833c4ea50 Mon Sep 17 00:00:00 2001
+From cee6889685240558ebea795615539b7289070842 Mon Sep 17 00:00:00 2001
 From: wangyetong <[email protected]>
 Date: Thu, 14 Sep 2023 15:33:42 +0800
 Subject: [PATCH] added dcdn toa
 
 ---
- include/ipvs/conn.h        |  5 +++++
+ include/ipvs/conn.h        |  4 ++++
  include/ipvs/proto_tcp.h   |  2 ++
- src/ipvs/ip_vs_proto_tcp.c | 54 +++++++++++++++++++++++++++++++++++++++++++++-
+ src/ipvs/ip_vs_proto_tcp.c | 55 +++++++++++++++++++++++++++++++++++++++++++++-
  3 files changed, 60 insertions(+), 1 deletion(-)
 
 diff --git a/include/ipvs/conn.h b/include/ipvs/conn.h
-index fa0bdeb..88dcb44 100644
+index 843721e..78fb0ba 100644
 --- a/include/ipvs/conn.h
 +++ b/include/ipvs/conn.h
-@@ -166,6 +166,11 @@ struct dp_vs_conn {
-     /* flag for gfwip */
-     bool outwall;
+@@ -167,6 +167,10 @@ struct dp_vs_conn {
+     /* connection redirect in fnat/snat/nat modes */
+     struct dp_vs_redirect  *redirect;
 
 +    /* dcdn toa found or not */
 +    bool dcdn_found;
 +    /* dcdn toa address */
 +    struct in_addr    dcdn_addr;
-+
  } __rte_cache_aligned;
 
  /* for syn-proxy to save all ack packet in conn before rs's syn-ack arrives */
 diff --git a/include/ipvs/proto_tcp.h b/include/ipvs/proto_tcp.h
-index 9f5162a..41d5646 100644
+index 3d1515a..f0cf50c 100644
 --- a/include/ipvs/proto_tcp.h
 +++ b/include/ipvs/proto_tcp.h
 @@ -28,6 +28,7 @@ enum {
@@ -46,11 +45,11 @@ index 9f5162a..41d5646 100644
  #define TCP_OLEN_TSTAMP_ALIGNED     12
  #define TCP_OLEN_SACK_BASE          2
 diff --git a/src/ipvs/ip_vs_proto_tcp.c b/src/ipvs/ip_vs_proto_tcp.c
-index cbb7cb2..2cd889a 100644
+index 6acbbca..5b185fa 100644
 --- a/src/ipvs/ip_vs_proto_tcp.c
 +++ b/src/ipvs/ip_vs_proto_tcp.c
-@@ -305,6 +305,43 @@ static void tcp_in_remove_ts(struct tcphdr *tcph)
-     }
+@@ -441,6 +441,43 @@ static int tcp_in_add_proxy_proto(struct dp_vs_conn *conn, struct rte_mbuf *mbuf
+     return proxy_proto_insert(&ppinfo, conn, mbuf, tcph, hdr_shift);
  }
 
 +/* check dcdn toa option */
@@ -90,10 +89,10 @@ index cbb7cb2..2cd889a 100644
 +    return EDPVS_NOTEXIST;
 +}
 +
- static inline int tcp_in_add_toa(struct dp_vs_conn *conn, struct rte_mbuf *mbuf,
+ static int tcp_in_add_toa(struct dp_vs_conn *conn, struct rte_mbuf *mbuf,
                            struct tcphdr *tcph)
  {
-@@ -382,7 +419,10 @@ static inline int tcp_in_add_toa(struct dp_vs_conn *conn, struct rte_mbuf *mbuf,
+@@ -518,7 +555,10 @@ static int tcp_in_add_toa(struct dp_vs_conn *conn, struct rte_mbuf *mbuf,
 
      if (conn->af == AF_INET) {
          struct tcpopt_ip4_addr *toa_ip4 = (struct tcpopt_ip4_addr *)(tcph + 1);
@@ -105,21 +104,18 @@ index cbb7cb2..2cd889a 100644
      }
      else {
          struct tcpopt_ip6_addr *toa_ip6 = (struct tcpopt_ip6_addr *)(tcph + 1);
-@@ -694,9 +734,13 @@ static int tcp_fnat_in_handler(struct dp_vs_proto *proto,
-                         struct dp_vs_conn *conn, struct rte_mbuf *mbuf)
- {
-     struct tcphdr *th;
+@@ -842,6 +882,10 @@ static int tcp_fnat_in_handler(struct dp_vs_proto *proto,
+     int af;             /* outbound af */
+     int iphdrlen;
+     int err, pp_hdr_shift = 0;
 +    struct in_addr dcdn_addr;
-     /* af/mbuf may be changed for nat64 which in af is ipv6 and out is ipv4 */
-     int af = tuplehash_out(conn).af;
-     int iphdrlen = ((AF_INET6 == af) ? ip6_hdrlen(mbuf): ip4_hdrlen(mbuf));
 +#ifdef CONFIG_DPVS_IPVS_DEBUG
 +    char dcdn_buf[64];
 +#endif
 
-     if (mbuf_may_pull(mbuf, iphdrlen + sizeof(*th)) != 0)
-         return EDPVS_INVPKT;
-@@ -720,6 +764,14 @@ static int tcp_fnat_in_handler(struct dp_vs_proto *proto,
+     af = tuplehash_out(conn).af;
+     iphdrlen = ((AF_INET6 == af) ? ip6_hdrlen(mbuf): ip4_hdrlen(mbuf));
+@@ -866,6 +910,15 @@ static int tcp_fnat_in_handler(struct dp_vs_proto *proto,
      if (th->syn && !th->ack) {
          tcp_in_remove_ts(th);
          tcp_in_init_seq(conn, mbuf, th);
@@ -131,9 +127,10 @@ index cbb7cb2..2cd889a 100644
 +            RTE_LOG(DEBUG, IPVS, "get dcdn toa addr %s\n", dcdn_buf);
 +#endif
 +        }
-         tcp_in_add_toa(conn, mbuf, th);
++        tcp_in_add_toa(conn, mbuf, th);
      }
 
+     /* Add toa/proxy_protocol to the first data packet */
 -- 
 1.8.3.1
 

src/ipvs/ip_vs_proto_tcp.c

@@ -306,46 +306,129 @@ static void tcp_in_remove_ts(struct tcphdr *tcph)
     }
 }
 
-/* use NOP option to replace TCP_OLEN_IP4_ADDR and TCP_OLEN_IP6_ADDR opt */
-static void tcp_in_remove_toa(struct tcphdr *tcph, int af)
+/*
+ * Remove NOP and TOA options preset in the mbuf and compact option space.
+ * If still no enough space, trim more options except for the protected ones.
+ *
+ * Return the trimmed length on success, otherwise dpvs error num on failure.
+ * */
+static int tcp_in_prune_options(int af, int reqlen, struct rte_mbuf *mbuf, struct tcphdr *tcph)
 {
-    unsigned char *ptr;
-    int len, i;
-    uint32_t tcp_opt_len = af == AF_INET ? TCP_OLEN_IP4_ADDR : TCP_OLEN_IP6_ADDR;
+    unsigned char *ptr, *fast, *slow;
+    const unsigned char *l3hdr, *payload;
+    int i, optlen;
+    unsigned int pruned;
+    uint8_t opcode, opsize;
+    uint64_t opts_protected;
+    const uint8_t opts_maxlen[64] = {
+        [2] = 4,  [3] = 3,   [4] = 2,
+        [8] = 10, [30] = 40, [34] = 18
+    };
 
     ptr = (unsigned char *)(tcph + 1);
-    len = (tcph->doff << 2) - sizeof(struct tcphdr);
+    fast = slow = ptr;
+    optlen = (tcph->doff << 2) - sizeof(struct tcphdr);
+    payload = ptr + optlen;
 
-    while (len > 0) {
-        int opcode = *ptr++;
-        int opsize;
+    if (optlen < reqlen)    /* make no sense to do anything */
+        return 0;
 
+    while (optlen > 0) {
+        opcode = *ptr++;
         switch (opcode) {
         case TCP_OPT_EOL:
-            return;
+            goto fini;
         case TCP_OPT_NOP:
-            len--;
+            fast++;
+            optlen--;
             continue;
         default:
             opsize = *ptr++;
-            if (opsize < 2)    /* silly options */
-                return;
-            if (opsize > len)
-                return;    /* partial options */
-            if ((opcode == TCP_OPT_ADDR) && (opsize == tcp_opt_len)) {
-                for (i = 0; i < tcp_opt_len; i++) {
-                    *(ptr - 2 + i) = TCP_OPT_NOP;
+            if (opsize < 2)             /* silly options */
+                goto fini;
+            if (opsize > optlen)        /* partial options */
+                goto fini;
+            if (opcode == TCP_OPT_ADDR) {
+                fast += opsize;
+            } else {
+                for (i = 0; i < opsize; i++) {
+                    if (slow != fast)
+                        *slow = *fast;
+                    slow++;
+                    fast++;
                 }
-                /* DON'T RETURN
-                 * keep search other TCP_OPT_ADDR ,and clear them.
-                 * See https://github.com/iqiyi/dpvs/pull/925 for more detail. */
             }
 
             ptr += opsize - 2;
-            len -= opsize;
+            optlen -= opsize;
             break;
         }
     }
+
+fini:
+    pruned = payload - slow;
+    if (pruned < reqlen) {
+        /* further trim the options, the tcp functionality relies on unprotected
+         * options may get hurt, refer to:
+         *   https://www.iana.org/assignments/tcp-parameters/tcp-parameters.xhtml
+         *   #tcp-parameters-1
+         * */
+        ptr = slow;
+        slow = fast = (unsigned char *)(tcph + 1);
+        if (tcph->syn)
+            opts_protected = (1ULL << 2) | (1ULL << 3) | (1ULL << 4)    /* MSS, WS, SACKP */
+                | (1ULL << 8) | (1ULL << 30) | (1ULL << 34);            /* TS, MPTCP, TFO */
+        else
+            opts_protected = (1ULL << 8);    /* TS, drop SACK, MPTCP DSS/REMOVE_ADDR */
+        while (fast < ptr) {
+            opcode = *fast;
+            opsize = *(fast + 1);
+            if (opcode < 64 && ((1ULL << opcode) & opts_protected)
+                    && (opsize <= opts_maxlen[opcode])) {
+                for (i = 0; i < opsize; i++)
+                    *slow++ = *fast++;
+                opts_protected ^= (1ULL << opcode);
+            } else {
+                fast += opsize;
+                pruned += opsize;
+                if (pruned >= reqlen) {
+                    while (fast < ptr)
+                        *slow++ = *fast++;
+                    break;
+                }
+            }
+        }
+        pruned = payload - slow;
+    }
+    if (pruned > 0) {
+        while (pruned & 0x3) {  /* 4-bytes alignment for tcp options */
+            *slow++ = 0;
+            pruned--;
+        }
+        if (!pruned)
+            return 0;
+        /* trim the packet */
+        l3hdr = rte_pktmbuf_mtod(mbuf, void *);
+        if (unlikely(mbuf_may_pull(mbuf, mbuf->pkt_len) != 0)) {
+            memset(slow, 0, pruned);
+            return EDPVS_INVPKT;
+        }
+        if (unlikely(payload - l3hdr > mbuf->pkt_len)) {
+            memset(slow, 0, pruned);
+            return EDPVS_INVPKT;
+        }
+        memmove(slow, payload, mbuf->pkt_len - (payload - l3hdr));
+        rte_pktmbuf_trim(mbuf, pruned);
+        tcph->doff -= (pruned >> 2);
+        if (af == AF_INET)
+            ((struct rte_ipv4_hdr *)l3hdr)->total_length =
+                htons(ntohs(((struct rte_ipv4_hdr *)l3hdr)->total_length) - pruned);
+        else
+            ((struct rte_ipv6_hdr *)l3hdr)->payload_len =
+                htons(ntohs(((struct rte_ipv6_hdr *)l3hdr)->payload_len) - pruned);
+        return pruned;
+    }
+    return 0;
 }
 
 static int tcp_in_add_proxy_proto(struct dp_vs_conn *conn, struct rte_mbuf *mbuf,
@@ -797,15 +880,12 @@ static int tcp_fnat_in_handler(struct dp_vs_proto *proto,
                         struct dp_vs_conn *conn, struct rte_mbuf *mbuf)
 {
     struct tcphdr *th;
-    /* af/mbuf may be changed for nat64 which in af is ipv6 and out is ipv4 */
-    int iaf, oaf;
-    int iphdrlen;
+    int af;             /* outbound af */
+    int iphdrlen, toalen;
     int err, pp_hdr_shift = 0;
 
-    iaf = tuplehash_in(conn).af;
-    oaf = tuplehash_out(conn).af;
-
-    iphdrlen = ((AF_INET6 == oaf) ? ip6_hdrlen(mbuf): ip4_hdrlen(mbuf));
+    af = tuplehash_out(conn).af;
+    iphdrlen = ((AF_INET6 == af) ? ip6_hdrlen(mbuf): ip4_hdrlen(mbuf));
 
     if (mbuf_may_pull(mbuf, iphdrlen + sizeof(*th)) != 0)
         return EDPVS_INVPKT;
@@ -819,41 +899,49 @@ static int tcp_fnat_in_handler(struct dp_vs_proto *proto,
 
     /*
      * for SYN packet
-     * 1. remove tcp timestamp option
-     *    laddress for different client have diff timestamp.
-     * 2. save original TCP sequence for seq-adjust later.
-     *    since TCP option will be change.
-     * 3. add TOA option
-     *    so that RS with TOA module can get real client IP.
+     * 1. remove tcp timestamp option,
+     *    laddrs for different clients have diff timestamp.
+     * 2. save original TCP sequence for seq-adjust later
+     *    since TCP option will be changed.
      */
     if (th->syn && !th->ack) {
         tcp_in_remove_ts(th);
-
         tcp_in_init_seq(conn, mbuf, th);
-        if (PROXY_PROTOCOL_V1 != PROXY_PROTOCOL_VERSION(conn->pp_version)
-                && PROXY_PROTOCOL_V2 != PROXY_PROTOCOL_VERSION(conn->pp_version)) {
-            if (unlikely(tcp_in_add_toa(conn, mbuf, th) != EDPVS_OK)) {
-                tcp_in_remove_toa(th, iaf);
-            }
-        }
     }
 
-    /* add toa/proxy_proto to first data packet */
+    /* Add toa/proxy_protocol to the first data packet */
     if (ntohl(th->ack_seq) == conn->fnat_seq.fdata_seq
             && !th->syn && !th->rst /*&& !th->fin*/) {
         if (PROXY_PROTOCOL_V2 == PROXY_PROTOCOL_VERSION(conn->pp_version)
                 || PROXY_PROTOCOL_V1 == PROXY_PROTOCOL_VERSION(conn->pp_version)) {
             if (conn->fnat_seq.isn - conn->fnat_seq.delta + 1 == ntohl(th->seq)) {
-                /* avoid inserting repetitive ppdata when the first rs ack delayed */
+                /* avoid inserting repetitive proxy protocol data
+                 * when the first rs ack is delayed */
                 err = tcp_in_add_proxy_proto(conn, mbuf, th, iphdrlen, &pp_hdr_shift);
                 if (unlikely(EDPVS_OK != err))
                     RTE_LOG(INFO, IPVS, "%s: insert proxy protocol fail -- %s\n",
                             __func__, dpvs_strerror(err));
                 th = ((void *)th) + pp_hdr_shift;
             }
-        } else {
-            if (unlikely(tcp_in_add_toa(conn, mbuf, th) != EDPVS_OK)) {
-                tcp_in_remove_toa(th, iaf);
+        } else { /* use toa */
+            err = tcp_in_add_toa(conn, mbuf, th);
+            if (unlikely(EDPVS_OK != err)) {
+                toalen = tuplehash_in(conn).af == AF_INET ? TCP_OLEN_IP4_ADDR : TCP_OLEN_IP6_ADDR;
+                if (tcp_in_prune_options(af, toalen, mbuf, th) >= toalen
+                        && (EDPVS_NOROOM == err || EDPVS_FRAG == err)) {
+                    err = tcp_in_add_toa(conn, mbuf, th);
+                }
+                if (EDPVS_OK != err) {
+                    char caddrbuf[64], vaddrbuf[64], laddrbuf[64], daddrbuf[64];
+                    const char *caddr, *vaddr, *laddr, *daddr;
+                    caddr = inet_ntop(conn->af, &conn->caddr, caddrbuf, sizeof(caddrbuf)) ? caddrbuf : "::";
+                    vaddr = inet_ntop(conn->af, &conn->vaddr, vaddrbuf, sizeof(vaddrbuf)) ? vaddrbuf : "::";
+                    laddr = inet_ntop(af, &conn->laddr, laddrbuf, sizeof(laddrbuf)) ? laddrbuf : "::";
+                    daddr = inet_ntop(af, &conn->daddr, daddrbuf, sizeof(daddrbuf)) ? daddrbuf : "::";
+                    RTE_LOG(WARNING, IPVS, "TOA add failed(%s): [%s]:%d -> [%s]:%d; [%s]:%d -> [%s]:%d\n",
+                            dpvs_strerror(err), caddr, htons(conn->cport), vaddr, htons(conn->vport),
+                            laddr, htons(conn->lport), daddr, htons(conn->dport));
+                }
             }
         }
     }
@@ -864,7 +952,7 @@ static int tcp_fnat_in_handler(struct dp_vs_proto *proto,
     th->source  = conn->lport;
     th->dest    = conn->dport;
 
-    return tcp_send_csum(oaf, iphdrlen, th, conn, mbuf, conn->in_dev);
+    return tcp_send_csum(af, iphdrlen, th, conn, mbuf, conn->in_dev);
 }
 
 static int tcp_fnat_out_handler(struct dp_vs_proto *proto,