ip/tcp_metrics.c

   1 /*
   2  * tcp_metrics.c        "ip tcp_metrics/tcpmetrics"
   3  *
   4  *              This program is free software; you can redistribute it and/or
   5  *              modify it under the terms of the GNU General Public License
   6  *              version 2 as published by the Free Software Foundation;
   7  *
   8  * Authors:     Julian Anastasov <ja@ssi.bg>, August 2012
   9  */
  10
  11 #include <stdio.h>
  12 #include <stdlib.h>
  13 #include <string.h>
  14 #include <unistd.h>
  15 #include <errno.h>
  16 #include <sys/types.h>
  17 #include <sys/socket.h>
  18 #include <arpa/inet.h>
  19 #include <sys/ioctl.h>
  20 #include <linux/if.h>
  21
  22 #include <linux/genetlink.h>
  23 #include <linux/tcp_metrics.h>
  24
  25 #include "utils.h"
  26 #include "ip_common.h"
  27 #include "libgenl.h"
  28
  29 static void usage(void)
  30 {
  31         fprintf(stderr, "Usage: ip tcp_metrics/tcpmetrics { COMMAND | help }\n");
  32         fprintf(stderr, "       ip tcp_metrics { show | flush } SELECTOR\n");
  33         fprintf(stderr, "       ip tcp_metrics delete [ address ] ADDRESS\n");
  34         fprintf(stderr, "SELECTOR := [ [ address ] PREFIX ]\n");
  35         exit(-1);
  36 }
  37
  38 /* netlink socket */
  39 static struct rtnl_handle grth = { .fd = -1 };
  40 static int genl_family = -1;
  41
  42 #define TCPM_REQUEST(_req, _bufsiz, _cmd, _flags) \
  43         GENL_REQUEST(_req, _bufsiz, genl_family, 0, \
  44                      TCP_METRICS_GENL_VERSION, _cmd, _flags)
  45
  46 #define CMD_LIST        0x0001  /* list, lst, show              */
  47 #define CMD_DEL         0x0002  /* delete, remove               */
  48 #define CMD_FLUSH       0x0004  /* flush                        */
  49
  50 static struct {
  51         char    *name;
  52         int     code;
  53 } cmds[] = {
  54         {       "list",         CMD_LIST        },
  55         {       "lst",          CMD_LIST        },
  56         {       "show",         CMD_LIST        },
  57         {       "delete",       CMD_DEL         },
  58         {       "remove",       CMD_DEL         },
  59         {       "flush",        CMD_FLUSH       },
  60 };
  61
  62 static char *metric_name[TCP_METRIC_MAX + 1] = {
  63         [TCP_METRIC_RTT]                = "rtt",
  64         [TCP_METRIC_RTTVAR]             = "rttvar",
  65         [TCP_METRIC_SSTHRESH]           = "ssthresh",
  66         [TCP_METRIC_CWND]               = "cwnd",
  67         [TCP_METRIC_REORDERING]         = "reordering",
  68 };
  69
  70 static struct
  71 {
  72         int flushed;
  73         char *flushb;
  74         int flushp;
  75         int flushe;
  76         int cmd;
  77         inet_prefix daddr;
  78         inet_prefix saddr;
  79 } f;
  80
  81 static int flush_update(void)
  82 {
  83         if (rtnl_send_check(&grth, f.flushb, f.flushp) < 0) {
  84                 perror("Failed to send flush request\n");
  85                 return -1;
  86         }
  87         f.flushp = 0;
  88         return 0;
  89 }
  90
  91 static int process_msg(const struct sockaddr_nl *who, struct nlmsghdr *n,
  92                        void *arg)
  93 {
  94         FILE *fp = (FILE *) arg;
  95         struct genlmsghdr *ghdr;
  96         struct rtattr *attrs[TCP_METRICS_ATTR_MAX + 1], *a;
  97         int len = n->nlmsg_len;
  98         inet_prefix daddr, saddr;
  99         int i, atype, stype;
 100
 101         if (n->nlmsg_type != genl_family)
 102                 return -1;
 103
 104         len -= NLMSG_LENGTH(GENL_HDRLEN);
 105         if (len < 0)
 106                 return -1;
 107
 108         ghdr = NLMSG_DATA(n);
 109         if (ghdr->cmd != TCP_METRICS_CMD_GET)
 110                 return 0;
 111
 112         parse_rtattr(attrs, TCP_METRICS_ATTR_MAX, (void *) ghdr + GENL_HDRLEN,
 113                      len);
 114
 115         if (attrs[TCP_METRICS_ATTR_ADDR_IPV4]) {
 116                 if (f.daddr.family && f.daddr.family != AF_INET)
 117                         return 0;
 118                 a = attrs[TCP_METRICS_ATTR_ADDR_IPV4];
 119                 daddr.family = AF_INET;
 120                 atype = TCP_METRICS_ATTR_ADDR_IPV4;
 121         } else if (attrs[TCP_METRICS_ATTR_ADDR_IPV6]) {
 122                 if (f.daddr.family && f.daddr.family != AF_INET6)
 123                         return 0;
 124                 a = attrs[TCP_METRICS_ATTR_ADDR_IPV6];
 125                 daddr.family = AF_INET6;
 126                 atype = TCP_METRICS_ATTR_ADDR_IPV6;
 127         } else {
 128                 return 0;
 129         }
 130
 131         if (get_addr_rta(&daddr, a, daddr.family))
 132                 return 0;
 133
 134         if (f.daddr.family && f.daddr.bitlen >= 0 &&
 135             inet_addr_match(&daddr, &f.daddr, f.daddr.bitlen))
 136                 return 0;
 137
 138         if (attrs[TCP_METRICS_ATTR_SADDR_IPV4]) {
 139                 if (f.saddr.family && f.saddr.family != AF_INET)
 140                         return 0;
 141                 a = attrs[TCP_METRICS_ATTR_SADDR_IPV4];
 142                 saddr.family = AF_INET;
 143                 stype = TCP_METRICS_ATTR_SADDR_IPV4;
 144         } else if (attrs[TCP_METRICS_ATTR_SADDR_IPV6]) {
 145                 if (f.saddr.family && f.saddr.family != AF_INET6)
 146                         return 0;
 147                 a = attrs[TCP_METRICS_ATTR_SADDR_IPV6];
 148                 saddr.family = AF_INET6;
 149                 stype = TCP_METRICS_ATTR_SADDR_IPV6;
 150         } else {
 151                 saddr.family = AF_UNSPEC;
 152                 stype = 0;
 153         }
 154
 155         /* Only get/check for the source-address if the kernel supports it. */
 156         if (saddr.family) {
 157                 if (get_addr_rta(&saddr, a, saddr.family))
 158                         return 0;
 159
 160                 if (f.saddr.family && f.saddr.bitlen >= 0 &&
 161                     inet_addr_match(&saddr, &f.saddr, f.saddr.bitlen))
 162                         return 0;
 163         }
 164
 165         if (f.flushb) {
 166                 struct nlmsghdr *fn;
 167
 168                 TCPM_REQUEST(req2, 128, TCP_METRICS_CMD_DEL, NLM_F_REQUEST);
 169
 170                 addattr_l(&req2.n, sizeof(req2), atype, daddr.data,
 171                           daddr.bytelen);
 172                 if (saddr.family)
 173                         addattr_l(&req2.n, sizeof(req2), stype, saddr.data,
 174                                   saddr.bytelen);
 175
 176                 if (NLMSG_ALIGN(f.flushp) + req2.n.nlmsg_len > f.flushe) {
 177                         if (flush_update())
 178                                 return -1;
 179                 }
 180                 fn = (struct nlmsghdr *) (f.flushb + NLMSG_ALIGN(f.flushp));
 181                 memcpy(fn, &req2.n, req2.n.nlmsg_len);
 182                 fn->nlmsg_seq = ++grth.seq;
 183                 f.flushp = (((char *) fn) + req2.n.nlmsg_len) - f.flushb;
 184                 f.flushed++;
 185                 if (show_stats < 2)
 186                         return 0;
 187         }
 188
 189         if (f.cmd & (CMD_DEL | CMD_FLUSH))
 190                 fprintf(fp, "Deleted ");
 191
 192         fprintf(fp, "%s",
 193                 format_host(daddr.family, daddr.bytelen, daddr.data));
 194
 195         a = attrs[TCP_METRICS_ATTR_AGE];
 196         if (a) {
 197                 unsigned long long val = rta_getattr_u64(a);
 198
 199                 fprintf(fp, " age %llu.%03llusec",
 200                         val / 1000, val % 1000);
 201         }
 202
 203         a = attrs[TCP_METRICS_ATTR_TW_TS_STAMP];
 204         if (a) {
 205                 __s32 val = (__s32) rta_getattr_u32(a);
 206                 __u32 tsval;
 207
 208                 a = attrs[TCP_METRICS_ATTR_TW_TSVAL];
 209                 tsval = a ? rta_getattr_u32(a) : 0;
 210                 fprintf(fp, " tw_ts %u/%dsec ago", tsval, val);
 211         }
 212
 213         a = attrs[TCP_METRICS_ATTR_VALS];
 214         if (a) {
 215                 struct rtattr *m[TCP_METRIC_MAX + 1 + 1];
 216                 unsigned long rtt = 0, rttvar = 0;
 217
 218                 parse_rtattr_nested(m, TCP_METRIC_MAX + 1, a);
 219
 220                 for (i = 0; i < TCP_METRIC_MAX + 1; i++) {
 221                         unsigned long val;
 222
 223                         a = m[i + 1];
 224                         if (!a)
 225                                 continue;
 226                         if (i != TCP_METRIC_RTT &&
 227                             i != TCP_METRIC_RTT_US &&
 228                             i != TCP_METRIC_RTTVAR &&
 229                             i != TCP_METRIC_RTTVAR_US) {
 230                                 if (metric_name[i])
 231                                         fprintf(fp, " %s ", metric_name[i]);
 232                                 else
 233                                         fprintf(fp, " metric_%d ", i);
 234                         }
 235                         val = rta_getattr_u32(a);
 236                         switch (i) {
 237                         case TCP_METRIC_RTT:
 238                                 if (!rtt)
 239                                         rtt = (val * 1000UL) >> 3;
 240                                 break;
 241                         case TCP_METRIC_RTTVAR:
 242                                 if (!rttvar)
 243                                         rttvar = (val * 1000UL) >> 2;
 244                                 break;
 245                         case TCP_METRIC_RTT_US:
 246                                 rtt = val >> 3;
 247                                 break;
 248                         case TCP_METRIC_RTTVAR_US:
 249                                 rttvar = val >> 2;
 250                                 break;
 251                         case TCP_METRIC_SSTHRESH:
 252                         case TCP_METRIC_CWND:
 253                         case TCP_METRIC_REORDERING:
 254                         default:
 255                                 fprintf(fp, "%lu", val);
 256                                 break;
 257                         }
 258                 }
 259                 if (rtt)
 260                         fprintf(fp, " rtt %luus", rtt);
 261                 if (rttvar)
 262                         fprintf(fp, " rttvar %luus", rttvar);
 263         }
 264
 265         a = attrs[TCP_METRICS_ATTR_FOPEN_MSS];
 266         if (a)
 267                 fprintf(fp, " fo_mss %u", rta_getattr_u16(a));
 268
 269         a = attrs[TCP_METRICS_ATTR_FOPEN_SYN_DROPS];
 270         if (a) {
 271                 __u16 syn_loss = rta_getattr_u16(a);
 272                 unsigned long long ts;
 273
 274                 a = attrs[TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS];
 275                 ts = a ? rta_getattr_u64(a) : 0;
 276
 277                 fprintf(fp, " fo_syn_drops %u/%llu.%03llusec ago",
 278                         syn_loss, ts / 1000, ts % 1000);
 279         }
 280
 281         a = attrs[TCP_METRICS_ATTR_FOPEN_COOKIE];
 282         if (a) {
 283                 char cookie[32 + 1];
 284                 unsigned char *ptr = RTA_DATA(a);
 285                 int i, max = RTA_PAYLOAD(a);
 286
 287                 if (max > 16)
 288                         max = 16;
 289                 cookie[0] = 0;
 290                 for (i = 0; i < max; i++)
 291                         sprintf(cookie + i + i, "%02x", ptr[i]);
 292                 fprintf(fp, " fo_cookie %s", cookie);
 293         }
 294
 295         if (saddr.family) {
 296                 fprintf(fp, " source %s",
 297                         format_host(saddr.family, saddr.bytelen, saddr.data));
 298         }
 299
 300         fprintf(fp, "\n");
 301
 302         fflush(fp);
 303         return 0;
 304 }
 305
 306 static int tcpm_do_cmd(int cmd, int argc, char **argv)
 307 {
 308         TCPM_REQUEST(req, 1024, TCP_METRICS_CMD_GET, NLM_F_REQUEST);
 309         struct nlmsghdr *answer;
 310         int atype = -1, stype = -1;
 311         int ack;
 312
 313         memset(&f, 0, sizeof(f));
 314         f.daddr.bitlen = -1;
 315         f.daddr.family = preferred_family;
 316         f.saddr.bitlen = -1;
 317         f.saddr.family = preferred_family;
 318
 319         switch (preferred_family) {
 320         case AF_UNSPEC:
 321         case AF_INET:
 322         case AF_INET6:
 323                 break;
 324         default:
 325                 fprintf(stderr, "Unsupported protocol family: %d\n", preferred_family);
 326                 return -1;
 327         }
 328
 329         for (; argc > 0; argc--, argv++) {
 330                 if (strcmp(*argv, "src") == 0 ||
 331                     strcmp(*argv, "source") == 0) {
 332                         char *who = *argv;
 333
 334                         NEXT_ARG();
 335                         if (matches(*argv, "help") == 0)
 336                                 usage();
 337                         if (f.saddr.bitlen >= 0)
 338                                 duparg2(who, *argv);
 339
 340                         get_prefix(&f.saddr, *argv, preferred_family);
 341                         if (f.saddr.bytelen && f.saddr.bytelen * 8 == f.saddr.bitlen) {
 342                                 if (f.saddr.family == AF_INET)
 343                                         stype = TCP_METRICS_ATTR_SADDR_IPV4;
 344                                 else if (f.saddr.family == AF_INET6)
 345                                         stype = TCP_METRICS_ATTR_SADDR_IPV6;
 346                         }
 347
 348                         if (stype < 0) {
 349                                 fprintf(stderr, "Error: a specific IP address is expected rather than \"%s\"\n",
 350                                         *argv);
 351                                 return -1;
 352                         }
 353                 } else {
 354                         char *who = "address";
 355
 356                         if (strcmp(*argv, "addr") == 0 ||
 357                             strcmp(*argv, "address") == 0) {
 358                                 who = *argv;
 359                                 NEXT_ARG();
 360                         }
 361                         if (matches(*argv, "help") == 0)
 362                                 usage();
 363                         if (f.daddr.bitlen >= 0)
 364                                 duparg2(who, *argv);
 365
 366                         get_prefix(&f.daddr, *argv, preferred_family);
 367                         if (f.daddr.bytelen && f.daddr.bytelen * 8 == f.daddr.bitlen) {
 368                                 if (f.daddr.family == AF_INET)
 369                                         atype = TCP_METRICS_ATTR_ADDR_IPV4;
 370                                 else if (f.daddr.family == AF_INET6)
 371                                         atype = TCP_METRICS_ATTR_ADDR_IPV6;
 372                         }
 373                         if ((CMD_DEL & cmd) && atype < 0) {
 374                                 fprintf(stderr, "Error: a specific IP address is expected rather than \"%s\"\n",
 375                                         *argv);
 376                                 return -1;
 377                         }
 378                 }
 379                 argc--; argv++;
 380         }
 381
 382         if (cmd == CMD_DEL && atype < 0)
 383                 missarg("address");
 384
 385         /* flush for exact address ? Single del */
 386         if (cmd == CMD_FLUSH && atype >= 0)
 387                 cmd = CMD_DEL;
 388
 389         /* flush for all addresses ? Single del without address */
 390         if (cmd == CMD_FLUSH && f.daddr.bitlen <= 0 &&
 391             f.saddr.bitlen <= 0 && preferred_family == AF_UNSPEC) {
 392                 cmd = CMD_DEL;
 393                 req.g.cmd = TCP_METRICS_CMD_DEL;
 394                 ack = 1;
 395         } else if (cmd == CMD_DEL) {
 396                 req.g.cmd = TCP_METRICS_CMD_DEL;
 397                 ack = 1;
 398         } else {        /* CMD_FLUSH, CMD_LIST */
 399                 ack = 0;
 400         }
 401
 402         if (genl_init_handle(&grth, TCP_METRICS_GENL_NAME, &genl_family))
 403                 exit(1);
 404         req.n.nlmsg_type = genl_family;
 405
 406         if (!(cmd & CMD_FLUSH) && (atype >= 0 || (cmd & CMD_DEL))) {
 407                 if (ack)
 408                         req.n.nlmsg_flags |= NLM_F_ACK;
 409                 if (atype >= 0)
 410                         addattr_l(&req.n, sizeof(req), atype, &f.daddr.data,
 411                                   f.daddr.bytelen);
 412                 if (stype >= 0)
 413                         addattr_l(&req.n, sizeof(req), stype, &f.saddr.data,
 414                                   f.saddr.bytelen);
 415         } else {
 416                 req.n.nlmsg_flags |= NLM_F_DUMP;
 417         }
 418
 419         f.cmd = cmd;
 420         if (cmd & CMD_FLUSH) {
 421                 int round = 0;
 422                 char flushb[4096-512];
 423
 424                 f.flushb = flushb;
 425                 f.flushp = 0;
 426                 f.flushe = sizeof(flushb);
 427
 428                 for (;;) {
 429                         req.n.nlmsg_seq = grth.dump = ++grth.seq;
 430                         if (rtnl_send(&grth, &req, req.n.nlmsg_len) < 0) {
 431                                 perror("Failed to send flush request");
 432                                 exit(1);
 433                         }
 434                         f.flushed = 0;
 435                         if (rtnl_dump_filter(&grth, process_msg, stdout) < 0) {
 436                                 fprintf(stderr, "Flush terminated\n");
 437                                 exit(1);
 438                         }
 439                         if (f.flushed == 0) {
 440                                 if (round == 0) {
 441                                         fprintf(stderr, "Nothing to flush.\n");
 442                                 } else if (show_stats)
 443                                         printf("*** Flush is complete after %d round%s ***\n",
 444                                                round, round > 1 ? "s" : "");
 445                                 fflush(stdout);
 446                                 return 0;
 447                         }
 448                         round++;
 449                         if (flush_update() < 0)
 450                                 exit(1);
 451                         if (show_stats) {
 452                                 printf("\n*** Round %d, deleting %d entries ***\n",
 453                                        round, f.flushed);
 454                                 fflush(stdout);
 455                         }
 456                 }
 457                 return 0;
 458         }
 459
 460         if (ack) {
 461                 if (rtnl_talk(&grth, &req.n, NULL) < 0)
 462                         return -2;
 463         } else if (atype >= 0) {
 464                 if (rtnl_talk(&grth, &req.n, &answer) < 0)
 465                         return -2;
 466                 if (process_msg(NULL, answer, stdout) < 0) {
 467                         fprintf(stderr, "Dump terminated\n");
 468                         exit(1);
 469                 }
 470                 free(answer);
 471         } else {
 472                 req.n.nlmsg_seq = grth.dump = ++grth.seq;
 473                 if (rtnl_send(&grth, &req, req.n.nlmsg_len) < 0) {
 474                         perror("Failed to send dump request");
 475                         exit(1);
 476                 }
 477
 478                 if (rtnl_dump_filter(&grth, process_msg, stdout) < 0) {
 479                         fprintf(stderr, "Dump terminated\n");
 480                         exit(1);
 481                 }
 482         }
 483         return 0;
 484 }
 485
 486 int do_tcp_metrics(int argc, char **argv)
 487 {
 488         int i;
 489
 490         if (argc < 1)
 491                 return tcpm_do_cmd(CMD_LIST, 0, NULL);
 492         for (i = 0; i < ARRAY_SIZE(cmds); i++) {
 493                 if (matches(argv[0], cmds[i].name) == 0)
 494                         return tcpm_do_cmd(cmds[i].code, argc-1, argv+1);
 495         }
 496         if (matches(argv[0], "help") == 0)
 497                 usage();
 498
 499         fprintf(stderr, "Command \"%s\" is unknown, try \"ip tcp_metrics help\".\n",
 500                         *argv);
 501         exit(-1);
 502 }