sys/netgraph7/ng_tcpmss.c

   1 /*-
   2  * ng_tcpmss.c
   3  *
   4  * Copyright (c) 2004, Alexey Popov <lollypop@flexuser.ru>
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice unmodified, this list of conditions, and the following
  12  *    disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  * This software includes fragments of the following programs:
  30  *      tcpmssd         Ruslan Ermilov <ru@FreeBSD.org>
  31  *
  32  * $FreeBSD: src/sys/netgraph/ng_tcpmss.c,v 1.4 2007/01/15 05:01:31 glebius Exp $
  33  */
  34
  35 /*
  36  * This node is netgraph tool for workaround of PMTUD problem. It acts
  37  * like filter for IP packets. If configured, it reduces MSS of TCP SYN
  38  * packets.
  39  *
  40  * Configuration can be done by sending NGM_TCPMSS_CONFIG message. The
  41  * message sets filter for incoming packets on hook 'inHook'. Packet's
  42  * TCP MSS field is lowered to 'maxMSS' parameter and resulting packet
  43  * is sent to 'outHook'.
  44  *
  45  * XXX: statistics are updated not atomically, so they may broke on SMP.
  46  */
  47
  48 #include <sys/param.h>
  49 #include <sys/systm.h>
  50 #include <sys/errno.h>
  51 #include <sys/kernel.h>
  52 #include <sys/malloc.h>
  53 #include <sys/mbuf.h>
  54
  55 #include <netinet/in.h>
  56 #include <netinet/in_systm.h>
  57 #include <netinet/ip.h>
  58 #include <netinet/tcp.h>
  59
  60 #include <netgraph/ng_message.h>
  61 #include <netgraph/netgraph.h>
  62 #include <netgraph/ng_parse.h>
  63 #include <netgraph/ng_tcpmss.h>
  64
  65 /* Per hook info. */
  66 typedef struct {
  67         hook_p                          outHook;
  68         struct ng_tcpmss_hookstat       stats;
  69 } *hpriv_p;
  70
  71 /* Netgraph methods. */
  72 static ng_constructor_t ng_tcpmss_constructor;
  73 static ng_rcvmsg_t      ng_tcpmss_rcvmsg;
  74 static ng_newhook_t     ng_tcpmss_newhook;
  75 static ng_rcvdata_t     ng_tcpmss_rcvdata;
  76 static ng_disconnect_t  ng_tcpmss_disconnect;
  77
  78 static int correct_mss(struct tcphdr *, int, uint16_t, int);
  79
  80 /* Parse type for struct ng_tcpmss_hookstat. */
  81 static const struct ng_parse_struct_field ng_tcpmss_hookstat_type_fields[]
  82         = NG_TCPMSS_HOOKSTAT_INFO;
  83 static const struct ng_parse_type ng_tcpmss_hookstat_type = {
  84         &ng_parse_struct_type,
  85         &ng_tcpmss_hookstat_type_fields
  86 };
  87
  88 /* Parse type for struct ng_tcpmss_config. */
  89 static const struct ng_parse_struct_field ng_tcpmss_config_type_fields[]
  90         = NG_TCPMSS_CONFIG_INFO;
  91 static const struct ng_parse_type ng_tcpmss_config_type = {
  92         &ng_parse_struct_type,
  93         ng_tcpmss_config_type_fields
  94 };
  95
  96 /* List of commands and how to convert arguments to/from ASCII. */
  97 static const struct ng_cmdlist ng_tcpmss_cmds[] = {
  98         {
  99           NGM_TCPMSS_COOKIE,
 100           NGM_TCPMSS_GET_STATS,
 101           "getstats",
 102           &ng_parse_hookbuf_type,
 103           &ng_tcpmss_hookstat_type
 104         },
 105         {
 106           NGM_TCPMSS_COOKIE,
 107           NGM_TCPMSS_CLR_STATS,
 108           "clrstats",
 109           &ng_parse_hookbuf_type,
 110           NULL
 111         },
 112         {
 113           NGM_TCPMSS_COOKIE,
 114           NGM_TCPMSS_GETCLR_STATS,
 115           "getclrstats",
 116           &ng_parse_hookbuf_type,
 117           &ng_tcpmss_hookstat_type
 118         },
 119         {
 120           NGM_TCPMSS_COOKIE,
 121           NGM_TCPMSS_CONFIG,
 122           "config",
 123           &ng_tcpmss_config_type,
 124           NULL
 125         },
 126         { 0 }
 127 };
 128
 129 /* Netgraph type descriptor. */
 130 static struct ng_type ng_tcpmss_typestruct = {
 131         .version =      NG_ABI_VERSION,
 132         .name =         NG_TCPMSS_NODE_TYPE,
 133         .constructor =  ng_tcpmss_constructor,
 134         .rcvmsg =       ng_tcpmss_rcvmsg,
 135         .newhook =      ng_tcpmss_newhook,
 136         .rcvdata =      ng_tcpmss_rcvdata,
 137         .disconnect =   ng_tcpmss_disconnect,
 138         .cmdlist =      ng_tcpmss_cmds,
 139 };
 140
 141 NETGRAPH_INIT(tcpmss, &ng_tcpmss_typestruct);
 142
 143 #define ERROUT(x)       { error = (x); goto done; }
 144
 145 /*
 146  * Node constructor. No special actions required.
 147  */
 148 static int
 149 ng_tcpmss_constructor(node_p node)
 150 {
 151         return (0);
 152 }
 153
 154 /*
 155  * Add a hook. Any unique name is OK.
 156  */
 157 static int
 158 ng_tcpmss_newhook(node_p node, hook_p hook, const char *name)
 159 {
 160         hpriv_p priv;
 161
 162         MALLOC(priv, hpriv_p, sizeof(*priv), M_NETGRAPH, M_NOWAIT | M_ZERO);
 163         if (priv == NULL)
 164                 return (ENOMEM);
 165
 166         NG_HOOK_SET_PRIVATE(hook, priv);
 167
 168         return (0);
 169 }
 170
 171 /*
 172  * Receive a control message.
 173  */
 174 static int
 175 ng_tcpmss_rcvmsg
 176 (node_p node, item_p item, hook_p lasthook)
 177 {
 178         struct ng_mesg *msg, *resp = NULL;
 179         int error = 0;
 180
 181         NGI_GET_MSG(item, msg);
 182
 183         switch (msg->header.typecookie) {
 184         case NGM_TCPMSS_COOKIE:
 185                 switch (msg->header.cmd) {
 186                 case NGM_TCPMSS_GET_STATS:
 187                 case NGM_TCPMSS_CLR_STATS:
 188                 case NGM_TCPMSS_GETCLR_STATS:
 189                     {
 190                         hook_p hook;
 191                         hpriv_p priv;
 192
 193                         /* Check that message is long enough. */
 194                         if (msg->header.arglen != NG_HOOKSIZ)
 195                                 ERROUT(EINVAL);
 196
 197                         /* Find this hook. */
 198                         hook = ng_findhook(node, (char *)msg->data);
 199                         if (hook == NULL)
 200                                 ERROUT(ENOENT);
 201
 202                         priv = NG_HOOK_PRIVATE(hook);
 203
 204                         /* Create response. */
 205                         if (msg->header.cmd != NGM_TCPMSS_CLR_STATS) {
 206                                 NG_MKRESPONSE(resp, msg,
 207                                     sizeof(struct ng_tcpmss_hookstat), M_NOWAIT);
 208                                 if (resp == NULL)
 209                                         ERROUT(ENOMEM);
 210                                 bcopy(&priv->stats, resp->data,
 211                                     sizeof(struct ng_tcpmss_hookstat));
 212                         }
 213
 214                         if (msg->header.cmd != NGM_TCPMSS_GET_STATS)
 215                                 bzero(&priv->stats,
 216                                     sizeof(struct ng_tcpmss_hookstat));
 217                         break;
 218                     }
 219                 case NGM_TCPMSS_CONFIG:
 220                     {
 221                         struct ng_tcpmss_config *set;
 222                         hook_p in, out;
 223                         hpriv_p priv;
 224
 225                         /* Check that message is long enough. */
 226                         if (msg->header.arglen !=
 227                             sizeof(struct ng_tcpmss_config))
 228                                 ERROUT(EINVAL);
 229
 230                         set = (struct ng_tcpmss_config *)msg->data;
 231                         in = ng_findhook(node, set->inHook);
 232                         out = ng_findhook(node, set->outHook);
 233                         if (in == NULL || out == NULL)
 234                                 ERROUT(ENOENT);
 235
 236                         /* Configure MSS hack. */
 237                         priv = NG_HOOK_PRIVATE(in);
 238                         priv->outHook = out;
 239                         priv->stats.maxMSS = set->maxMSS;
 240
 241                         break;
 242                     }
 243                 default:
 244                         error = EINVAL;
 245                         break;
 246                 }
 247                 break;
 248         default:
 249                 error = EINVAL;
 250                 break;
 251         }
 252
 253 done:
 254         NG_RESPOND_MSG(error, node, item, resp);
 255         NG_FREE_MSG(msg);
 256
 257         return (error);
 258 }
 259
 260 /*
 261  * Receive data on a hook, and hack MSS.
 262  *
 263  */
 264 static int
 265 ng_tcpmss_rcvdata(hook_p hook, item_p item)
 266 {
 267         hpriv_p priv = NG_HOOK_PRIVATE(hook);
 268         struct mbuf *m = NULL;
 269         struct ip *ip;
 270         struct tcphdr *tcp;
 271         int iphlen, tcphlen, pktlen;
 272         int pullup_len = 0;
 273         int error = 0;
 274
 275         /* Drop packets if filter is not configured on this hook. */
 276         if (priv->outHook == NULL)
 277                 goto done;
 278
 279         NGI_GET_M(item, m);
 280
 281         /* Update stats on incoming hook. */
 282         pktlen = m->m_pkthdr.len;
 283         priv->stats.Octets += pktlen;
 284         priv->stats.Packets++;
 285
 286         /* Check whether we configured to fix MSS. */
 287         if (priv->stats.maxMSS == 0)
 288                 goto send;
 289
 290 #define M_CHECK(length) do {                                    \
 291         pullup_len += length;                                   \
 292         if ((m)->m_pkthdr.len < pullup_len)                     \
 293                 goto send;                                      \
 294         if ((m)->m_len < pullup_len &&                          \
 295            (((m) = m_pullup((m), pullup_len)) == NULL))         \
 296                 ERROUT(ENOBUFS);                                \
 297         } while (0)
 298
 299         /* Check mbuf packet size and arrange for IP header. */
 300         M_CHECK(sizeof(struct ip));
 301         ip = mtod(m, struct ip *);
 302
 303         /* Check IP version. */
 304         if (ip->ip_v != IPVERSION)
 305                 ERROUT(EINVAL);
 306
 307         /* Check IP header length. */
 308         iphlen = ip->ip_hl << 2;
 309         if (iphlen < sizeof(struct ip) || iphlen > pktlen )
 310                 ERROUT(EINVAL);
 311
 312         /* Check if it is TCP. */
 313         if (!(ip->ip_p == IPPROTO_TCP))
 314                 goto send;
 315
 316         /* Check mbuf packet size and arrange for IP+TCP header */
 317         M_CHECK(iphlen - sizeof(struct ip) + sizeof(struct tcphdr));
 318         ip = mtod(m, struct ip *);
 319         tcp = (struct tcphdr *)((caddr_t )ip + iphlen);
 320
 321         /* Check TCP header length. */
 322         tcphlen = tcp->th_off << 2;
 323         if (tcphlen < sizeof(struct tcphdr) || tcphlen > pktlen - iphlen)
 324                 ERROUT(EINVAL);
 325
 326         /* Check SYN packet and has options. */
 327         if (!(tcp->th_flags & TH_SYN) || tcphlen == sizeof(struct tcphdr))
 328                 goto send;
 329
 330         /* Update SYN stats. */
 331         priv->stats.SYNPkts++;
 332
 333         M_CHECK(tcphlen - sizeof(struct tcphdr));
 334         ip = mtod(m, struct ip *);
 335         tcp = (struct tcphdr *)((caddr_t )ip + iphlen);
 336
 337 #undef  M_CHECK
 338
 339         /* Fix MSS and update stats. */
 340         if (correct_mss(tcp, tcphlen, priv->stats.maxMSS,
 341             m->m_pkthdr.csum_flags))
 342                 priv->stats.FixedPkts++;
 343
 344 send:
 345         /* Deliver frame out destination hook. */
 346         NG_FWD_NEW_DATA(error, item, priv->outHook, m);
 347
 348         return (error);
 349
 350 done:
 351         NG_FREE_ITEM(item);
 352         NG_FREE_M(m);
 353
 354         return (error);
 355 }
 356
 357 /*
 358  * Hook disconnection.
 359  * We must check all hooks, since they may reference this one.
 360  */
 361 static int
 362 ng_tcpmss_disconnect(hook_p hook)
 363 {
 364         node_p node = NG_HOOK_NODE(hook);
 365         hook_p hook2;
 366
 367         LIST_FOREACH(hook2, &node->nd_hooks, hk_hooks) {
 368                 hpriv_p priv = NG_HOOK_PRIVATE(hook2);
 369
 370                 if (priv->outHook == hook)
 371                         priv->outHook = NULL;
 372         }
 373
 374         FREE(NG_HOOK_PRIVATE(hook), M_NETGRAPH);
 375
 376         if (NG_NODE_NUMHOOKS(NG_HOOK_NODE(hook)) == 0)
 377                 ng_rmnode_self(NG_HOOK_NODE(hook));
 378
 379         return (0);
 380 }
 381
 382 /*
 383  * Code from tcpmssd.
 384  */
 385
 386 /*-
 387  * The following macro is used to update an
 388  * internet checksum.  "acc" is a 32-bit
 389  * accumulation of all the changes to the
 390  * checksum (adding in old 16-bit words and
 391  * subtracting out new words), and "cksum"
 392  * is the checksum value to be updated.
 393  */
 394 #define TCPMSS_ADJUST_CHECKSUM(acc, cksum) do {         \
 395         acc += cksum;                                   \
 396         if (acc < 0) {                                  \
 397                 acc = -acc;                             \
 398                 acc = (acc >> 16) + (acc & 0xffff);     \
 399                 acc += acc >> 16;                       \
 400                 cksum = (u_short) ~acc;                 \
 401         } else {                                        \
 402                 acc = (acc >> 16) + (acc & 0xffff);     \
 403                 acc += acc >> 16;                       \
 404                 cksum = (u_short) acc;                  \
 405         }                                               \
 406 } while (0);
 407
 408 static int
 409 correct_mss(struct tcphdr *tc, int hlen, uint16_t maxmss, int flags)
 410 {
 411         int olen, optlen;
 412         u_char *opt;
 413         uint16_t *mss;
 414         int accumulate;
 415         int res = 0;
 416
 417         for (olen = hlen - sizeof(struct tcphdr), opt = (u_char *)(tc + 1);
 418              olen > 0; olen -= optlen, opt += optlen) {
 419                 if (*opt == TCPOPT_EOL)
 420                         break;
 421                 else if (*opt == TCPOPT_NOP)
 422                         optlen = 1;
 423                 else {
 424                         optlen = *(opt + 1);
 425                         if (optlen <= 0 || optlen > olen)
 426                                 break;
 427                         if (*opt == TCPOPT_MAXSEG) {
 428                                 if (optlen != TCPOLEN_MAXSEG)
 429                                         continue;
 430                                 mss = (uint16_t *)(opt + 2);
 431                                 if (ntohs(*mss) > maxmss) {
 432                                         accumulate = *mss;
 433                                         *mss = htons(maxmss);
 434                                         accumulate -= *mss;
 435                                         if ((flags & CSUM_TCP) == 0)
 436                                                 TCPMSS_ADJUST_CHECKSUM(accumulate, tc->th_sum);
 437                                         res = 1;
 438                                 }
 439                         }
 440                 }
 441         }
 442         return (res);
 443 }