sys/kern/uipc_msg.c

   1 /*
   2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
   3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
   4  *
   5  * This code is derived from software contributed to The DragonFly Project
   6  * by Jeffrey M. Hsu.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following 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  * 3. Neither the name of The DragonFly Project nor the names of its
  17  *    contributors may be used to endorse or promote products derived
  18  *    from this software without specific, prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  * $DragonFly: src/sys/kern/uipc_msg.c,v 1.26 2008/10/27 02:56:30 sephe Exp $
  34  */
  35
  36 #include <sys/param.h>
  37 #include <sys/systm.h>
  38 #include <sys/kernel.h>
  39 #include <sys/msgport.h>
  40 #include <sys/protosw.h>
  41 #include <sys/socket.h>
  42 #include <sys/socketvar.h>
  43 #include <sys/socketops.h>
  44 #include <sys/thread.h>
  45 #include <sys/thread2.h>
  46 #include <sys/msgport2.h>
  47 #include <vm/pmap.h>
  48 #include <net/netmsg2.h>
  49
  50 #include <net/netisr.h>
  51 #include <net/netmsg.h>
  52
  53 /*
  54  * Abort a socket and free it.  Called from soabort() only.  soabort()
  55  * got a ref on the socket which we must free on reply.
  56  */
  57 void
  58 so_pru_abort(struct socket *so)
  59 {
  60         struct netmsg_pru_abort msg;
  61
  62         netmsg_init(&msg.base, so, &curthread->td_msgport,
  63                     0, so->so_proto->pr_usrreqs->pru_abort);
  64         (void)lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
  65         sofree(msg.base.nm_so);
  66 }
  67
  68 /*
  69  * Abort a socket and free it, asynchronously.  Called from
  70  * soaborta() only.  soaborta() got a ref on the socket which we must
  71  * free on reply.
  72  */
  73 void
  74 so_pru_aborta(struct socket *so)
  75 {
  76         struct netmsg_pru_abort *msg;
  77
  78         msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_WAITOK | M_ZERO);
  79         netmsg_init(&msg->base, so, &netisr_afree_free_so_rport,
  80                     0, so->so_proto->pr_usrreqs->pru_abort);
  81         lwkt_sendmsg(so->so_port, &msg->base.lmsg);
  82 }
  83
  84 /*
  85  * Abort a socket and free it.  Called from soabort_oncpu() only.
  86  * Caller must make sure that the current CPU is inpcb's owner CPU.
  87  */
  88 void
  89 so_pru_abort_oncpu(struct socket *so)
  90 {
  91         struct netmsg_pru_abort msg;
  92         netisr_fn_t func = so->so_proto->pr_usrreqs->pru_abort;
  93
  94         netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
  95         msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
  96         msg.base.lmsg.ms_flags |= MSGF_SYNC;
  97         func((netmsg_t)&msg);
  98         KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
  99         sofree(msg.base.nm_so);
 100 }
 101
 102 /*
 103  * WARNING!  Synchronous call from user context
 104  */
 105 int
 106 so_pru_accept_direct(struct socket *so, struct sockaddr **nam)
 107 {
 108         struct netmsg_pru_accept msg;
 109         netisr_fn_t func = so->so_proto->pr_usrreqs->pru_accept;
 110
 111         netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
 112         msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
 113         msg.base.lmsg.ms_flags |= MSGF_SYNC;
 114         msg.nm_nam = nam;
 115         func((netmsg_t)&msg);
 116         KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
 117         return(msg.base.lmsg.ms_error);
 118 }
 119
 120 int
 121 so_pru_attach(struct socket *so, int proto, struct pru_attach_info *ai)
 122 {
 123         struct netmsg_pru_attach msg;
 124         int error;
 125
 126         netmsg_init(&msg.base, so, &curthread->td_msgport,
 127                     0, so->so_proto->pr_usrreqs->pru_attach);
 128         msg.nm_proto = proto;
 129         msg.nm_ai = ai;
 130         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 131         return (error);
 132 }
 133
 134 int
 135 so_pru_attach_direct(struct socket *so, int proto, struct pru_attach_info *ai)
 136 {
 137         struct netmsg_pru_attach msg;
 138         netisr_fn_t func = so->so_proto->pr_usrreqs->pru_attach;
 139
 140         netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
 141         msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
 142         msg.base.lmsg.ms_flags |= MSGF_SYNC;
 143         msg.nm_proto = proto;
 144         msg.nm_ai = ai;
 145         func((netmsg_t)&msg);
 146         KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
 147         return(msg.base.lmsg.ms_error);
 148 }
 149
 150 /*
 151  * NOTE: If the target port changes the bind operation will deal with it.
 152  */
 153 int
 154 so_pru_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
 155 {
 156         struct netmsg_pru_bind msg;
 157         int error;
 158
 159         netmsg_init(&msg.base, so, &curthread->td_msgport,
 160                     0, so->so_proto->pr_usrreqs->pru_bind);
 161         msg.nm_nam = nam;
 162         msg.nm_td = td;         /* used only for prison_ip() */
 163         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 164         return (error);
 165 }
 166
 167 int
 168 so_pru_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 169 {
 170         struct netmsg_pru_connect msg;
 171         int error;
 172
 173         netmsg_init(&msg.base, so, &curthread->td_msgport,
 174                     0, so->so_proto->pr_usrreqs->pru_connect);
 175         msg.nm_nam = nam;
 176         msg.nm_td = td;
 177         msg.nm_m = NULL;
 178         msg.nm_flags = 0;
 179         msg.nm_reconnect = 0;
 180         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 181         return (error);
 182 }
 183
 184 int
 185 so_pru_connect2(struct socket *so1, struct socket *so2)
 186 {
 187         struct netmsg_pru_connect2 msg;
 188         int error;
 189
 190         netmsg_init(&msg.base, so1, &curthread->td_msgport,
 191                     0, so1->so_proto->pr_usrreqs->pru_connect2);
 192         msg.nm_so1 = so1;
 193         msg.nm_so2 = so2;
 194         error = lwkt_domsg(so1->so_port, &msg.base.lmsg, 0);
 195         return (error);
 196 }
 197
 198 /*
 199  * WARNING!  Synchronous call from user context.  Control function may do
 200  *           copyin/copyout.
 201  */
 202 int
 203 so_pru_control_direct(struct socket *so, u_long cmd, caddr_t data,
 204                       struct ifnet *ifp)
 205 {
 206         struct netmsg_pru_control msg;
 207         netisr_fn_t func = so->so_proto->pr_usrreqs->pru_control;
 208
 209         netmsg_init(&msg.base, so, &netisr_adone_rport, 0, func);
 210         msg.base.lmsg.ms_flags &= ~(MSGF_REPLY | MSGF_DONE);
 211         msg.base.lmsg.ms_flags |= MSGF_SYNC;
 212         msg.nm_cmd = cmd;
 213         msg.nm_data = data;
 214         msg.nm_ifp = ifp;
 215         msg.nm_td = curthread;
 216         func((netmsg_t)&msg);
 217         KKASSERT(msg.base.lmsg.ms_flags & MSGF_DONE);
 218         return(msg.base.lmsg.ms_error);
 219 }
 220
 221 int
 222 so_pru_detach(struct socket *so)
 223 {
 224         struct netmsg_pru_detach msg;
 225         int error;
 226
 227         netmsg_init(&msg.base, so, &curthread->td_msgport,
 228                     0, so->so_proto->pr_usrreqs->pru_detach);
 229         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 230         return (error);
 231 }
 232
 233 int
 234 so_pru_disconnect(struct socket *so)
 235 {
 236         struct netmsg_pru_disconnect msg;
 237         int error;
 238
 239         netmsg_init(&msg.base, so, &curthread->td_msgport,
 240                     0, so->so_proto->pr_usrreqs->pru_disconnect);
 241         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 242         return (error);
 243 }
 244
 245 int
 246 so_pru_listen(struct socket *so, struct thread *td)
 247 {
 248         struct netmsg_pru_listen msg;
 249         int error;
 250
 251         netmsg_init(&msg.base, so, &curthread->td_msgport,
 252                     0, so->so_proto->pr_usrreqs->pru_listen);
 253         msg.nm_td = td;         /* used only for prison_ip() XXX JH */
 254         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 255         return (error);
 256 }
 257
 258 int
 259 so_pru_peeraddr(struct socket *so, struct sockaddr **nam)
 260 {
 261         struct netmsg_pru_peeraddr msg;
 262         int error;
 263
 264         netmsg_init(&msg.base, so, &curthread->td_msgport,
 265                     0, so->so_proto->pr_usrreqs->pru_peeraddr);
 266         msg.nm_nam = nam;
 267         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 268         return (error);
 269 }
 270
 271 int
 272 so_pru_rcvd(struct socket *so, int flags)
 273 {
 274         struct netmsg_pru_rcvd msg;
 275         int error;
 276
 277         netmsg_init(&msg.base, so, &curthread->td_msgport,
 278                     0, so->so_proto->pr_usrreqs->pru_rcvd);
 279         msg.nm_flags = flags;
 280         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 281         return (error);
 282 }
 283
 284 int
 285 so_pru_rcvoob(struct socket *so, struct mbuf *m, int flags)
 286 {
 287         struct netmsg_pru_rcvoob msg;
 288         int error;
 289
 290         netmsg_init(&msg.base, so, &curthread->td_msgport,
 291                     0, so->so_proto->pr_usrreqs->pru_rcvoob);
 292         msg.nm_m = m;
 293         msg.nm_flags = flags;
 294         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 295         return (error);
 296 }
 297
 298 /*
 299  * NOTE: If the target port changes the implied connect will deal with it.
 300  */
 301 int
 302 so_pru_send(struct socket *so, int flags, struct mbuf *m,
 303             struct sockaddr *addr, struct mbuf *control, struct thread *td)
 304 {
 305         struct netmsg_pru_send msg;
 306         int error;
 307
 308         netmsg_init(&msg.base, so, &curthread->td_msgport,
 309                     0, so->so_proto->pr_usrreqs->pru_send);
 310         msg.nm_flags = flags;
 311         msg.nm_m = m;
 312         msg.nm_addr = addr;
 313         msg.nm_control = control;
 314         msg.nm_td = td;
 315         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 316         return (error);
 317 }
 318
 319 int
 320 so_pru_sense(struct socket *so, struct stat *sb)
 321 {
 322         struct netmsg_pru_sense msg;
 323         int error;
 324
 325         netmsg_init(&msg.base, so, &curthread->td_msgport,
 326                     0, so->so_proto->pr_usrreqs->pru_sense);
 327         msg.nm_stat = sb;
 328         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 329         return (error);
 330 }
 331
 332 int
 333 so_pru_shutdown(struct socket *so)
 334 {
 335         struct netmsg_pru_shutdown msg;
 336         int error;
 337
 338         netmsg_init(&msg.base, so, &curthread->td_msgport,
 339                     0, so->so_proto->pr_usrreqs->pru_shutdown);
 340         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 341         return (error);
 342 }
 343
 344 int
 345 so_pru_sockaddr(struct socket *so, struct sockaddr **nam)
 346 {
 347         struct netmsg_pru_sockaddr msg;
 348         int error;
 349
 350         netmsg_init(&msg.base, so, &curthread->td_msgport,
 351                     0, so->so_proto->pr_usrreqs->pru_sockaddr);
 352         msg.nm_nam = nam;
 353         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 354         return (error);
 355 }
 356
 357 int
 358 so_pr_ctloutput(struct socket *so, struct sockopt *sopt)
 359 {
 360         struct netmsg_pr_ctloutput msg;
 361         int error;
 362
 363         KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
 364         netmsg_init(&msg.base, so, &curthread->td_msgport,
 365                     0, so->so_proto->pr_ctloutput);
 366         msg.nm_sopt = sopt;
 367         error = lwkt_domsg(so->so_port, &msg.base.lmsg, 0);
 368         return (error);
 369 }
 370
 371 /*
 372  * Protocol control input, typically via icmp.
 373  *
 374  * If the protocol pr_ctlport is not NULL we call it to figure out the
 375  * protocol port.  If NULL is returned we can just return, otherwise
 376  * we issue a netmsg to call pr_ctlinput in the proper thread.
 377  *
 378  * This must be done synchronously as arg and/or extra may point to
 379  * temporary data.
 380  */
 381 void
 382 so_pru_ctlinput(struct protosw *pr, int cmd, struct sockaddr *arg, void *extra)
 383 {
 384         struct netmsg_pru_ctlinput msg;
 385         lwkt_port_t port;
 386
 387         if (pr->pr_ctlport == NULL)
 388                 return;
 389         KKASSERT(pr->pr_ctlinput != NULL);
 390         port = pr->pr_ctlport(cmd, arg, extra);
 391         if (port == NULL)
 392                 return;
 393         netmsg_init(&msg.base, NULL, &curthread->td_msgport,
 394                     0, pr->pr_ctlinput);
 395         msg.nm_cmd = cmd;
 396         msg.nm_arg = arg;
 397         msg.nm_extra = extra;
 398         lwkt_domsg(port, &msg.base.lmsg, 0);
 399 }
 400
 401 /*
 402  * If we convert all the protosw pr_ functions for all the protocols
 403  * to take a message directly, this layer can go away.  For the moment
 404  * our dispatcher ignores the return value, but since we are handling
 405  * the replymsg ourselves we return EASYNC by convention.
 406  */
 407
 408 /*
 409  * Handle a predicate event request.  This function is only called once
 410  * when the predicate message queueing request is received.
 411  */
 412 void
 413 netmsg_so_notify(netmsg_t msg)
 414 {
 415         struct signalsockbuf *ssb;
 416
 417         ssb = (msg->notify.nm_etype & NM_REVENT) ?
 418                         &msg->base.nm_so->so_rcv :
 419                         &msg->base.nm_so->so_snd;
 420
 421         /*
 422          * Reply immediately if the event has occured, otherwise queue the
 423          * request.
 424          */
 425         if (msg->notify.nm_predicate(&msg->notify)) {
 426                 lwkt_replymsg(&msg->base.lmsg,
 427                               msg->base.lmsg.ms_error);
 428         } else {
 429                 lwkt_gettoken(&kq_token);
 430                 TAILQ_INSERT_TAIL(&ssb->ssb_kq.ki_mlist, &msg->notify, nm_list);
 431                 atomic_set_int(&ssb->ssb_flags, SSB_MEVENT);
 432                 lwkt_reltoken(&kq_token);
 433         }
 434 }
 435
 436 /*
 437  * Called by doio when trying to abort a netmsg_so_notify message.
 438  * Unlike the other functions this one is dispatched directly by
 439  * the LWKT subsystem, so it takes a lwkt_msg_t as an argument.
 440  *
 441  * The original message, lmsg, is under the control of the caller and
 442  * will not be destroyed until we return so we can safely reference it
 443  * in our synchronous abort request.
 444  *
 445  * This part of the abort request occurs on the originating cpu which
 446  * means we may race the message flags and the original message may
 447  * not even have been processed by the target cpu yet.
 448  */
 449 void
 450 netmsg_so_notify_doabort(lwkt_msg_t lmsg)
 451 {
 452         struct netmsg_so_notify_abort msg;
 453
 454         if ((lmsg->ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
 455                 netmsg_init(&msg.base, NULL, &curthread->td_msgport,
 456                             0, netmsg_so_notify_abort);
 457                 msg.nm_notifymsg = (void *)lmsg;
 458                 lwkt_domsg(lmsg->ms_target_port, &msg.base.lmsg, 0);
 459         }
 460 }
 461
 462 /*
 463  * Predicate requests can be aborted.  This function is only called once
 464  * and will interlock against processing/reply races (since such races
 465  * occur on the same thread that controls the port where the abort is
 466  * requeued).
 467  *
 468  * This part of the abort request occurs on the target cpu.  The message
 469  * flags must be tested again in case the test that we did on the
 470  * originating cpu raced.  Since messages are handled in sequence, the
 471  * original message will have already been handled by the loop and either
 472  * replied to or queued.
 473  *
 474  * We really only need to interlock with MSGF_REPLY (a bit that is set on
 475  * our cpu when we reply).  Note that MSGF_DONE is not set until the
 476  * reply reaches the originating cpu.  Test both bits anyway.
 477  */
 478 void
 479 netmsg_so_notify_abort(netmsg_t msg)
 480 {
 481         struct netmsg_so_notify_abort *abrtmsg = &msg->notify_abort;
 482         struct netmsg_so_notify *nmsg = abrtmsg->nm_notifymsg;
 483         struct signalsockbuf *ssb;
 484
 485         /*
 486          * The original notify message is not destroyed until after the
 487          * abort request is returned, so we can check its state.
 488          */
 489         if ((nmsg->base.lmsg.ms_flags & (MSGF_DONE | MSGF_REPLY)) == 0) {
 490                 ssb = (nmsg->nm_etype & NM_REVENT) ?
 491                                 &nmsg->base.nm_so->so_rcv :
 492                                 &nmsg->base.nm_so->so_snd;
 493                 lwkt_gettoken(&kq_token);
 494                 TAILQ_REMOVE(&ssb->ssb_kq.ki_mlist, nmsg, nm_list);
 495                 lwkt_reltoken(&kq_token);
 496                 lwkt_replymsg(&nmsg->base.lmsg, EINTR);
 497         }
 498
 499         /*
 500          * Reply to the abort message
 501          */
 502         lwkt_replymsg(&abrtmsg->base.lmsg, 0);
 503 }