bsd/netinet/in_pcblist.c

   1 /*
   2  * Copyright (c) 2010-2012 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. The rights granted to you under the License
  10  * may not be used to create, or enable the creation or redistribution of,
  11  * unlawful or unlicensed copies of an Apple operating system, or to
  12  * circumvent, violate, or enable the circumvention or violation of, any
  13  * terms of an Apple operating system software license agreement.
  14  *
  15  * Please obtain a copy of the License at
  16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
  17  *
  18  * The Original Code and all software distributed under the License are
  19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  23  * Please see the License for the specific language governing rights and
  24  * limitations under the License.
  25  *
  26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  27  */
  28 /*
  29  * Copyright (c) 1982, 1986, 1990, 1993
  30  *      The Regents of the University of California.  All rights reserved.
  31  *
  32  * Redistribution and use in source and binary forms, with or without
  33  * modification, are permitted provided that the following conditions
  34  * are met:
  35  * 1. Redistributions of source code must retain the above copyright
  36  *    notice, this list of conditions and the following disclaimer.
  37  * 2. Redistributions in binary form must reproduce the above copyright
  38  *    notice, this list of conditions and the following disclaimer in the
  39  *    documentation and/or other materials provided with the distribution.
  40  * 3. All advertising materials mentioning features or use of this software
  41  *    must display the following acknowledgement:
  42  *      This product includes software developed by the University of
  43  *      California, Berkeley and its contributors.
  44  * 4. Neither the name of the University nor the names of its contributors
  45  *    may be used to endorse or promote products derived from this software
  46  *    without specific prior written permission.
  47  *
  48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  58  * SUCH DAMAGE.
  59  */
  60
  61 #include <sys/types.h>
  62 #include <sys/malloc.h>
  63 #include <sys/socket.h>
  64 #include <sys/socketvar.h>
  65 #include <sys/protosw.h>
  66 #include <sys/domain.h>
  67 #include <sys/kernel.h>
  68 #include <sys/sysctl.h>
  69 #include <sys/dtrace.h>
  70 #include <sys/kauth.h>
  71
  72 #include <net/route.h>
  73 #include <net/if_var.h>
  74
  75 #include <netinet/in.h>
  76 #include <netinet/in_pcb.h>
  77 #include <netinet/ip_var.h>
  78
  79 #include <netinet/udp.h>
  80 #include <netinet/udp_var.h>
  81
  82 #include <netinet/tcp.h>
  83 #include <netinet/tcp_fsm.h>
  84 #include <netinet/tcp_seq.h>
  85 #include <netinet/tcp_timer.h>
  86 #include <netinet/tcp_var.h>
  87
  88 #ifndef ROUNDUP64
  89 #define ROUNDUP64(x) P2ROUNDUP((x), sizeof(u_int64_t))
  90 #endif
  91
  92 #ifndef ADVANCE64
  93 #define ADVANCE64(p, n) (void*)((char *)(p) + ROUNDUP64(n))
  94 #endif
  95
  96
  97 void sotoxsocket_n(struct socket *, struct xsocket_n *);
  98 void sbtoxsockbuf_n(struct sockbuf *, struct xsockbuf_n *);
  99 void sbtoxsockstat_n(struct socket *, struct xsockstat_n *);
 100 void inpcb_to_xinpcb_n(struct inpcb *, struct xinpcb_n *);
 101 void tcpcb_to_xtcpcb_n(struct tcpcb *, struct xtcpcb_n *);
 102
 103 __private_extern__ void
 104 sotoxsocket_n(struct socket *so, struct xsocket_n *xso)
 105 {
 106         xso->xso_len = sizeof(struct xsocket_n);
 107         xso->xso_kind = XSO_SOCKET;
 108
 109         if (so != NULL) {
 110                 xso->xso_so = (u_int64_t)(uintptr_t)so;
 111                 xso->so_type = so->so_type;
 112                 xso->so_options = so->so_options;
 113                 xso->so_linger = so->so_linger;
 114                 xso->so_state = so->so_state;
 115                 xso->so_pcb = (u_int64_t)(uintptr_t)so->so_pcb;
 116                 if (so->so_proto) {
 117                         xso->xso_protocol = so->so_proto->pr_protocol;
 118                         xso->xso_family = so->so_proto->pr_domain->dom_family;
 119                 } else {
 120                         xso->xso_protocol = xso->xso_family = 0;
 121                 }
 122                 xso->so_qlen = so->so_qlen;
 123                 xso->so_incqlen = so->so_incqlen;
 124                 xso->so_qlimit = so->so_qlimit;
 125                 xso->so_timeo = so->so_timeo;
 126                 xso->so_error = so->so_error;
 127                 xso->so_pgid = so->so_pgid;
 128                 xso->so_oobmark = so->so_oobmark;
 129                 xso->so_uid = kauth_cred_getuid(so->so_cred);
 130         }
 131 }
 132
 133 __private_extern__ void
 134 sbtoxsockbuf_n(struct sockbuf *sb, struct xsockbuf_n *xsb)
 135 {
 136         xsb->xsb_len = sizeof(struct xsockbuf_n);
 137         xsb->xsb_kind = (sb->sb_flags & SB_RECV) ? XSO_RCVBUF : XSO_SNDBUF;
 138
 139         if (sb != NULL) {
 140                 xsb->sb_cc = sb->sb_cc;
 141                 xsb->sb_hiwat = sb->sb_hiwat;
 142                 xsb->sb_mbcnt = sb->sb_mbcnt;
 143                 xsb->sb_mbmax = sb->sb_mbmax;
 144                 xsb->sb_lowat = sb->sb_lowat;
 145                 xsb->sb_flags = sb->sb_flags;
 146                 xsb->sb_timeo = (short)
 147                         (sb->sb_timeo.tv_sec * hz) + sb->sb_timeo.tv_usec / tick;
 148                 if (xsb->sb_timeo == 0 && sb->sb_timeo.tv_usec != 0)
 149                         xsb->sb_timeo = 1;
 150         }
 151 }
 152
 153 __private_extern__ void
 154 sbtoxsockstat_n(struct socket *so, struct xsockstat_n *xst)
 155 {
 156         int i;
 157
 158         xst->xst_len = sizeof(struct xsockstat_n);
 159         xst->xst_kind = XSO_STATS;
 160
 161         for (i = 0; i < SO_TC_STATS_MAX; i++) {
 162                 xst->xst_tc_stats[i].rxpackets = so->so_tc_stats[i].rxpackets;
 163                 xst->xst_tc_stats[i].rxbytes = so->so_tc_stats[i].rxbytes;
 164                 xst->xst_tc_stats[i].txpackets = so->so_tc_stats[i].txpackets;
 165                 xst->xst_tc_stats[i].txbytes = so->so_tc_stats[i].txbytes;
 166         }
 167 }
 168
 169 __private_extern__ void
 170 inpcb_to_xinpcb_n(struct inpcb *inp, struct xinpcb_n *xinp)
 171 {
 172         xinp->xi_len = sizeof(struct xinpcb_n);
 173         xinp->xi_kind = XSO_INPCB;
 174         xinp->xi_inpp = (u_int64_t)(uintptr_t)inp;
 175         xinp->inp_fport = inp->inp_fport;
 176         xinp->inp_lport = inp->inp_lport;
 177         xinp->inp_ppcb = (u_int64_t)(uintptr_t)inp->inp_ppcb;
 178         xinp->inp_gencnt = inp->inp_gencnt;
 179         xinp->inp_flags = inp->inp_flags;
 180         xinp->inp_flow = inp->inp_flow;
 181         xinp->inp_vflag = inp->inp_vflag;
 182         xinp->inp_ip_ttl = inp->inp_ip_ttl;
 183         xinp->inp_ip_p = inp->inp_ip_p;
 184         xinp->inp_dependfaddr.inp6_foreign = inp->inp_dependfaddr.inp6_foreign;
 185         xinp->inp_dependladdr.inp6_local = inp->inp_dependladdr.inp6_local;
 186         xinp->inp_depend4.inp4_ip_tos = inp->inp_depend4.inp4_ip_tos;
 187         xinp->inp_depend6.inp6_hlim = inp->inp_depend6.inp6_hlim;
 188         xinp->inp_depend6.inp6_cksum = inp->inp_depend6.inp6_cksum;
 189         xinp->inp_depend6.inp6_ifindex = inp->inp_depend6.inp6_ifindex;
 190         xinp->inp_depend6.inp6_hops = inp->inp_depend6.inp6_hops;
 191         xinp->inp_flowhash = inp->inp_flowhash;
 192 }
 193
 194 __private_extern__ void
 195 tcpcb_to_xtcpcb_n(struct tcpcb *tp, struct xtcpcb_n *xt)
 196 {
 197         int i;
 198
 199         xt->xt_len = sizeof(struct xtcpcb_n);
 200         xt->xt_kind = XSO_TCPCB;
 201
 202         xt->t_segq = (u_int32_t)(uintptr_t)tp->t_segq.lh_first;
 203         xt->t_dupacks = tp->t_dupacks;
 204         for (i = 0; i < TCPT_NTIMERS_EXT; i++)
 205                 xt->t_timer[i] = tp->t_timer[i];
 206         xt->t_state = tp->t_state;
 207         xt->t_flags = tp->t_flags;
 208         xt->t_force = tp->t_force;
 209         xt->snd_una = tp->snd_una;
 210         xt->snd_max = tp->snd_max;
 211         xt->snd_nxt = tp->snd_nxt;
 212         xt->snd_up = tp->snd_up;
 213         xt->snd_wl1 = tp->snd_wl1;
 214         xt->snd_wl2 = tp->snd_wl2;
 215         xt->iss = tp->iss;
 216         xt->irs = tp->irs;
 217         xt->rcv_nxt = tp->rcv_nxt;
 218         xt->rcv_adv = tp->rcv_adv;
 219         xt->rcv_wnd = tp->rcv_wnd;
 220         xt->rcv_up = tp->rcv_up;
 221         xt->snd_wnd = tp->snd_wnd;
 222         xt->snd_cwnd = tp->snd_cwnd;
 223         xt->snd_ssthresh = tp->snd_ssthresh;
 224         xt->t_maxopd = tp->t_maxopd;
 225         xt->t_rcvtime = tp->t_rcvtime;
 226         xt->t_starttime = tp->t_starttime;
 227         xt->t_rtttime = tp->t_rtttime;
 228         xt->t_rtseq = tp->t_rtseq;
 229         xt->t_rxtcur = tp->t_rxtcur;
 230         xt->t_maxseg = tp->t_maxseg;
 231         xt->t_srtt = tp->t_srtt;
 232         xt->t_rttvar = tp->t_rttvar;
 233         xt->t_rxtshift = tp->t_rxtshift;
 234         xt->t_rttmin = tp->t_rttmin;
 235         xt->t_rttupdated = tp->t_rttupdated;
 236         xt->max_sndwnd = tp->max_sndwnd;
 237         xt->t_softerror = tp->t_softerror;
 238         xt->t_oobflags = tp->t_oobflags;
 239         xt->t_iobc = tp->t_iobc;
 240         xt->snd_scale = tp->snd_scale;
 241         xt->rcv_scale = tp->rcv_scale;
 242         xt->request_r_scale = tp->request_r_scale;
 243         xt->requested_s_scale = tp->requested_s_scale;
 244         xt->ts_recent = tp->ts_recent;
 245         xt->ts_recent_age = tp->ts_recent_age;
 246         xt->last_ack_sent = tp->last_ack_sent;
 247         xt->cc_send = tp->cc_send;
 248         xt->cc_recv = tp->cc_recv;
 249         xt->snd_recover = tp->snd_recover;
 250         xt->snd_cwnd_prev = tp->snd_cwnd_prev;
 251         xt->snd_ssthresh_prev = tp->snd_ssthresh_prev;
 252         xt->t_badrxtwin = tp->t_badrxtwin;
 253 }
 254
 255 __private_extern__ int
 256 get_pcblist_n(short proto, struct sysctl_req *req, struct inpcbinfo *pcbinfo)
 257 {
 258         int error = 0;
 259         int i, n;
 260         struct inpcb *inp, **inp_list = NULL;
 261         inp_gen_t gencnt;
 262         struct xinpgen xig;
 263         void *buf = NULL;
 264         size_t item_size = ROUNDUP64(sizeof(struct xinpcb_n)) +
 265                                                 ROUNDUP64(sizeof(struct xsocket_n)) +
 266                                                 2 * ROUNDUP64(sizeof(struct xsockbuf_n)) +
 267                                                 ROUNDUP64(sizeof(struct xsockstat_n));
 268
 269         if (proto == IPPROTO_TCP)
 270                 item_size += ROUNDUP64(sizeof(struct xtcpcb_n));
 271
 272         /*
 273          * The process of preparing the PCB list is too time-consuming and
 274          * resource-intensive to repeat twice on every request.
 275          */
 276         lck_rw_lock_exclusive(pcbinfo->mtx);
 277         if (req->oldptr == USER_ADDR_NULL) {
 278                         n = pcbinfo->ipi_count;
 279                         req->oldidx = 2 * (sizeof xig)
 280                                         + (n + n/8) * item_size;
 281                         goto done;
 282         }
 283
 284         if (req->newptr != USER_ADDR_NULL) {
 285                         error = EPERM;
 286                         goto done;
 287         }
 288
 289         /*
 290          * OK, now we're committed to doing something.
 291          */
 292         gencnt = pcbinfo->ipi_gencnt;
 293         n = pcbinfo->ipi_count;
 294
 295         bzero(&xig, sizeof(xig));
 296         xig.xig_len = sizeof xig;
 297         xig.xig_count = n;
 298         xig.xig_gen = gencnt;
 299         xig.xig_sogen = so_gencnt;
 300         error = SYSCTL_OUT(req, &xig, sizeof xig);
 301         if (error) {
 302                         goto done;
 303         }
 304     /*
 305      * We are done if there is no pcb
 306      */
 307     if (n == 0) {
 308                         goto done;
 309     }
 310
 311         buf = _MALLOC(item_size, M_TEMP, M_WAITOK);
 312         if (buf == 0) {
 313                         error = ENOMEM;
 314                         goto done;
 315         }
 316
 317         inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK);
 318         if (inp_list == 0) {
 319                         error = ENOMEM;
 320                         goto done;
 321         }
 322
 323         for (inp = pcbinfo->listhead->lh_first, i = 0; inp && i < n;
 324                  inp = inp->inp_list.le_next) {
 325                         if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD)
 326                                         inp_list[i++] = inp;
 327         }
 328         n = i;
 329
 330         error = 0;
 331         for (i = 0; i < n; i++) {
 332                 inp = inp_list[i];
 333                 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) {
 334                         struct xinpcb_n *xi = (struct xinpcb_n *)buf;
 335                         struct xsocket_n *xso = (struct xsocket_n *)ADVANCE64(xi, sizeof(*xi));
 336                         struct xsockbuf_n *xsbrcv = (struct xsockbuf_n *)ADVANCE64(xso, sizeof(*xso));
 337                         struct xsockbuf_n *xsbsnd = (struct xsockbuf_n *)ADVANCE64(xsbrcv, sizeof(*xsbrcv));
 338                         struct xsockstat_n *xsostats = (struct xsockstat_n *)ADVANCE64(xsbsnd, sizeof(*xsbsnd));
 339
 340                         bzero(buf, item_size);
 341
 342                         inpcb_to_xinpcb_n(inp, xi);
 343                         sotoxsocket_n(inp->inp_socket, xso);
 344                         sbtoxsockbuf_n(inp->inp_socket ? &inp->inp_socket->so_rcv : NULL, xsbrcv);
 345                         sbtoxsockbuf_n(inp->inp_socket ? &inp->inp_socket->so_snd : NULL, xsbsnd);
 346                         sbtoxsockstat_n(inp->inp_socket, xsostats);
 347                         if (proto == IPPROTO_TCP) {
 348                                 struct  xtcpcb_n *xt = (struct xtcpcb_n *)ADVANCE64(xsostats, sizeof(*xsostats));
 349
 350                                 /*
 351                                  * inp->inp_ppcb, can only be NULL on
 352                                  * an initialization race window.
 353                                  * No need to lock.
 354                                  */
 355                                 if (inp->inp_ppcb == NULL)
 356                                         continue;
 357
 358                                 tcpcb_to_xtcpcb_n((struct tcpcb *)inp->inp_ppcb, xt);
 359                         }
 360                         error = SYSCTL_OUT(req, buf, item_size);
 361                 }
 362         }
 363         if (!error) {
 364                 /*
 365                  * Give the user an updated idea of our state.
 366                  * If the generation differs from what we told
 367                  * her before, she knows that something happened
 368                  * while we were processing this request, and it
 369                  * might be necessary to retry.
 370                  */
 371                 bzero(&xig, sizeof(xig));
 372                 xig.xig_len = sizeof xig;
 373                 xig.xig_gen = pcbinfo->ipi_gencnt;
 374                 xig.xig_sogen = so_gencnt;
 375                 xig.xig_count = pcbinfo->ipi_count;
 376                 error = SYSCTL_OUT(req, &xig, sizeof xig);
 377         }
 378 done:
 379         lck_rw_done(pcbinfo->mtx);
 380         if (inp_list)
 381                 FREE(inp_list, M_TEMP);
 382         if (buf)
 383                 FREE(buf, M_TEMP);
 384         return error;
 385 }
 386
 387 __private_extern__ void
 388 inpcb_get_ports_used(unsigned int ifindex, uint8_t *bitfield, struct inpcbinfo *pcbinfo)
 389 {
 390         lck_rw_lock_shared(pcbinfo->mtx);
 391
 392         struct inpcb *inp;
 393         inp_gen_t       gencnt = pcbinfo->ipi_gencnt;
 394         for (inp = LIST_FIRST(pcbinfo->listhead); inp; inp = LIST_NEXT(inp, inp_list)) {
 395                 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD &&
 396                         (ifindex == 0 || inp->inp_last_outifp == NULL || ifindex == inp->inp_last_outifp->if_index)) {
 397                         uint16_t port = ntohs(inp->inp_lport);
 398                         bitfield[port / 8] |= 1 << (port & 0x7);
 399                 }
 400         }
 401
 402         lck_rw_done(pcbinfo->mtx);
 403 }
 404
 405 __private_extern__ uint32_t
 406 inpcb_count_opportunistic(unsigned int ifindex, struct inpcbinfo *pcbinfo,
 407     u_int32_t flags)
 408 {
 409         uint32_t opportunistic = 0;
 410
 411         lck_rw_lock_shared(pcbinfo->mtx);
 412
 413         struct inpcb *inp;
 414         inp_gen_t       gencnt = pcbinfo->ipi_gencnt;
 415         for (inp = LIST_FIRST(pcbinfo->listhead);
 416                 inp; inp = LIST_NEXT(inp, inp_list)) {
 417                 if (inp->inp_gencnt <= gencnt &&
 418                     inp->inp_state != INPCB_STATE_DEAD &&
 419                     inp->inp_socket != NULL &&
 420                     so_get_opportunistic(inp->inp_socket) &&
 421                     inp->inp_last_outifp != NULL &&
 422                     ifindex == inp->inp_last_outifp->if_index) {
 423                         opportunistic++;
 424                         struct socket *so = inp->inp_socket;
 425                         if ((flags & INPCB_OPPORTUNISTIC_SETCMD) &&
 426                             (so->so_state & SS_ISCONNECTED)) {
 427                                 socket_lock(so, 1);
 428                                 if (flags & INPCB_OPPORTUNISTIC_THROTTLEON) {
 429                                         so->so_flags |= SOF_SUSPENDED;
 430                                         soevent(so,
 431                                             (SO_FILT_HINT_LOCKED |
 432                                             SO_FILT_HINT_SUSPEND));
 433                                 } else {
 434                                         so->so_flags &= ~(SOF_SUSPENDED);
 435                                         soevent(so,
 436                                             (SO_FILT_HINT_LOCKED |
 437                                             SO_FILT_HINT_RESUME));
 438                                 }
 439                                 SOTHROTTLELOG(("throttle[%d]: so %p [%d,%d] "
 440                                     "%s\n", so->last_pid, so, INP_SOCKAF(so),
 441                                     INP_SOCKTYPE(so),
 442                                     (so->so_flags & SOF_SUSPENDED) ?
 443                                     "SUSPENDED" : "RESUMED"));
 444                                 socket_unlock(so, 1);
 445                         }
 446                 }
 447         }
 448
 449         lck_rw_done(pcbinfo->mtx);
 450
 451         return (opportunistic);
 452 }