sys/dev/iscsi/icl.c

   1 /*-
   2  * Copyright (c) 2012 The FreeBSD Foundation
   3  * All rights reserved.
   4  *
   5  * This software was developed by Edward Tomasz Napierala under sponsorship
   6  * from the FreeBSD Foundation.
   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  *
  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  */
  30
  31 /*
  32  * iSCSI Common Layer.  It's used by both the initiator and target to send
  33  * and receive iSCSI PDUs.
  34  */
  35
  36 #include <sys/cdefs.h>
  37 __FBSDID("$FreeBSD$");
  38
  39 #include <sys/param.h>
  40 #include <sys/capability.h>
  41 #include <sys/condvar.h>
  42 #include <sys/conf.h>
  43 #include <sys/file.h>
  44 #include <sys/kernel.h>
  45 #include <sys/kthread.h>
  46 #include <sys/lock.h>
  47 #include <sys/mbuf.h>
  48 #include <sys/mutex.h>
  49 #include <sys/module.h>
  50 #include <sys/protosw.h>
  51 #include <sys/socket.h>
  52 #include <sys/socketvar.h>
  53 #include <sys/sysctl.h>
  54 #include <sys/systm.h>
  55 #include <sys/sx.h>
  56 #include <sys/uio.h>
  57 #include <vm/uma.h>
  58 #include <netinet/in.h>
  59 #include <netinet/tcp.h>
  60
  61 #include <dev/iscsi/icl.h>
  62 #include <dev/iscsi/iscsi_proto.h>
  63
  64 SYSCTL_NODE(_kern, OID_AUTO, icl, CTLFLAG_RD, 0, "iSCSI Common Layer");
  65 static int debug = 1;
  66 TUNABLE_INT("kern.icl.debug", &debug);
  67 SYSCTL_INT(_kern_icl, OID_AUTO, debug, CTLFLAG_RWTUN,
  68     &debug, 0, "Enable debug messages");
  69 static int coalesce = 1;
  70 TUNABLE_INT("kern.icl.coalesce", &coalesce);
  71 SYSCTL_INT(_kern_icl, OID_AUTO, coalesce, CTLFLAG_RWTUN,
  72     &coalesce, 0, "Try to coalesce PDUs before sending");
  73 static int partial_receive_len = 128 * 1024;
  74 TUNABLE_INT("kern.icl.partial_receive_len", &partial_receive_len);
  75 SYSCTL_INT(_kern_icl, OID_AUTO, partial_receive_len, CTLFLAG_RWTUN,
  76     &partial_receive_len, 0, "Minimum read size for partially received "
  77     "data segment");
  78 static int sendspace = 1048576;
  79 TUNABLE_INT("kern.icl.sendspace", &sendspace);
  80 SYSCTL_INT(_kern_icl, OID_AUTO, sendspace, CTLFLAG_RWTUN,
  81     &sendspace, 0, "Default send socket buffer size");
  82 static int recvspace = 1048576;
  83 TUNABLE_INT("kern.icl.recvspace", &recvspace);
  84 SYSCTL_INT(_kern_icl, OID_AUTO, recvspace, CTLFLAG_RWTUN,
  85     &recvspace, 0, "Default receive socket buffer size");
  86
  87 static uma_zone_t icl_conn_zone;
  88 static uma_zone_t icl_pdu_zone;
  89
  90 static volatile u_int   icl_ncons;
  91
  92 #define ICL_DEBUG(X, ...)                                               \
  93         do {                                                            \
  94                 if (debug > 1)                                          \
  95                         printf("%s: " X "\n", __func__, ## __VA_ARGS__);\
  96         } while (0)
  97
  98 #define ICL_WARN(X, ...)                                                \
  99         do {                                                            \
 100                 if (debug > 0) {                                        \
 101                         printf("WARNING: %s: " X "\n",                  \
 102                             __func__, ## __VA_ARGS__);                  \
 103                 }                                                       \
 104         } while (0)
 105
 106 #define ICL_CONN_LOCK(X)                mtx_lock(X->ic_lock)
 107 #define ICL_CONN_UNLOCK(X)              mtx_unlock(X->ic_lock)
 108 #define ICL_CONN_LOCK_ASSERT(X)         mtx_assert(X->ic_lock, MA_OWNED)
 109 #define ICL_CONN_LOCK_ASSERT_NOT(X)     mtx_assert(X->ic_lock, MA_NOTOWNED)
 110
 111 STAILQ_HEAD(icl_pdu_stailq, icl_pdu);
 112
 113 static void
 114 icl_conn_fail(struct icl_conn *ic)
 115 {
 116         if (ic->ic_socket == NULL)
 117                 return;
 118
 119         /*
 120          * XXX
 121          */
 122         ic->ic_socket->so_error = EDOOFUS;
 123         (ic->ic_error)(ic);
 124 }
 125
 126 static struct mbuf *
 127 icl_conn_receive(struct icl_conn *ic, size_t len)
 128 {
 129         struct uio uio;
 130         struct socket *so;
 131         struct mbuf *m;
 132         int error, flags;
 133
 134         so = ic->ic_socket;
 135
 136         memset(&uio, 0, sizeof(uio));
 137         uio.uio_resid = len;
 138
 139         flags = MSG_DONTWAIT;
 140         error = soreceive(so, NULL, &uio, &m, NULL, &flags);
 141         if (error != 0) {
 142                 ICL_DEBUG("soreceive error %d", error);
 143                 return (NULL);
 144         }
 145         if (uio.uio_resid != 0) {
 146                 m_freem(m);
 147                 ICL_DEBUG("short read");
 148                 return (NULL);
 149         }
 150
 151         return (m);
 152 }
 153
 154 static struct icl_pdu *
 155 icl_pdu_new(struct icl_conn *ic, int flags)
 156 {
 157         struct icl_pdu *ip;
 158
 159 #ifdef DIAGNOSTIC
 160         refcount_acquire(&ic->ic_outstanding_pdus);
 161 #endif
 162         ip = uma_zalloc(icl_pdu_zone, flags | M_ZERO);
 163         if (ip == NULL) {
 164                 ICL_WARN("failed to allocate %zd bytes", sizeof(*ip));
 165 #ifdef DIAGNOSTIC
 166                 refcount_release(&ic->ic_outstanding_pdus);
 167 #endif
 168                 return (NULL);
 169         }
 170
 171         ip->ip_conn = ic;
 172
 173         return (ip);
 174 }
 175
 176 void
 177 icl_pdu_free(struct icl_pdu *ip)
 178 {
 179         struct icl_conn *ic;
 180
 181         ic = ip->ip_conn;
 182
 183         m_freem(ip->ip_bhs_mbuf);
 184         m_freem(ip->ip_ahs_mbuf);
 185         m_freem(ip->ip_data_mbuf);
 186         uma_zfree(icl_pdu_zone, ip);
 187 #ifdef DIAGNOSTIC
 188         refcount_release(&ic->ic_outstanding_pdus);
 189 #endif
 190 }
 191
 192 /*
 193  * Allocate icl_pdu with empty BHS to fill up by the caller.
 194  */
 195 struct icl_pdu *
 196 icl_pdu_new_bhs(struct icl_conn *ic, int flags)
 197 {
 198         struct icl_pdu *ip;
 199
 200         ip = icl_pdu_new(ic, flags);
 201         if (ip == NULL)
 202                 return (NULL);
 203
 204         ip->ip_bhs_mbuf = m_getm2(NULL, sizeof(struct iscsi_bhs),
 205             flags, MT_DATA, M_PKTHDR);
 206         if (ip->ip_bhs_mbuf == NULL) {
 207                 ICL_WARN("failed to allocate %zd bytes", sizeof(*ip));
 208                 icl_pdu_free(ip);
 209                 return (NULL);
 210         }
 211         ip->ip_bhs = mtod(ip->ip_bhs_mbuf, struct iscsi_bhs *);
 212         memset(ip->ip_bhs, 0, sizeof(struct iscsi_bhs));
 213         ip->ip_bhs_mbuf->m_len = sizeof(struct iscsi_bhs);
 214
 215         return (ip);
 216 }
 217
 218 static int
 219 icl_pdu_ahs_length(const struct icl_pdu *request)
 220 {
 221
 222         return (request->ip_bhs->bhs_total_ahs_len * 4);
 223 }
 224
 225 size_t
 226 icl_pdu_data_segment_length(const struct icl_pdu *request)
 227 {
 228         uint32_t len = 0;
 229
 230         len += request->ip_bhs->bhs_data_segment_len[0];
 231         len <<= 8;
 232         len += request->ip_bhs->bhs_data_segment_len[1];
 233         len <<= 8;
 234         len += request->ip_bhs->bhs_data_segment_len[2];
 235
 236         return (len);
 237 }
 238
 239 static void
 240 icl_pdu_set_data_segment_length(struct icl_pdu *response, uint32_t len)
 241 {
 242
 243         response->ip_bhs->bhs_data_segment_len[2] = len;
 244         response->ip_bhs->bhs_data_segment_len[1] = len >> 8;
 245         response->ip_bhs->bhs_data_segment_len[0] = len >> 16;
 246 }
 247
 248 static size_t
 249 icl_pdu_padding(const struct icl_pdu *ip)
 250 {
 251
 252         if ((ip->ip_data_len % 4) != 0)
 253                 return (4 - (ip->ip_data_len % 4));
 254
 255         return (0);
 256 }
 257
 258 static size_t
 259 icl_pdu_size(const struct icl_pdu *response)
 260 {
 261         size_t len;
 262
 263         KASSERT(response->ip_ahs_len == 0, ("responding with AHS"));
 264
 265         len = sizeof(struct iscsi_bhs) + response->ip_data_len +
 266             icl_pdu_padding(response);
 267         if (response->ip_conn->ic_header_crc32c)
 268                 len += ISCSI_HEADER_DIGEST_SIZE;
 269         if (response->ip_data_len != 0 && response->ip_conn->ic_data_crc32c)
 270                 len += ISCSI_DATA_DIGEST_SIZE;
 271
 272         return (len);
 273 }
 274
 275 static int
 276 icl_pdu_receive_bhs(struct icl_pdu *request, size_t *availablep)
 277 {
 278         struct mbuf *m;
 279
 280         m = icl_conn_receive(request->ip_conn, sizeof(struct iscsi_bhs));
 281         if (m == NULL) {
 282                 ICL_DEBUG("failed to receive BHS");
 283                 return (-1);
 284         }
 285
 286         request->ip_bhs_mbuf = m_pullup(m, sizeof(struct iscsi_bhs));
 287         if (request->ip_bhs_mbuf == NULL) {
 288                 ICL_WARN("m_pullup failed");
 289                 return (-1);
 290         }
 291         request->ip_bhs = mtod(request->ip_bhs_mbuf, struct iscsi_bhs *);
 292
 293         /*
 294          * XXX: For architectures with strict alignment requirements
 295          *      we may need to allocate ip_bhs and copy the data into it.
 296          *      For some reason, though, not doing this doesn't seem
 297          *      to cause problems; tested on sparc64.
 298          */
 299
 300         *availablep -= sizeof(struct iscsi_bhs);
 301         return (0);
 302 }
 303
 304 static int
 305 icl_pdu_receive_ahs(struct icl_pdu *request, size_t *availablep)
 306 {
 307
 308         request->ip_ahs_len = icl_pdu_ahs_length(request);
 309         if (request->ip_ahs_len == 0)
 310                 return (0);
 311
 312         request->ip_ahs_mbuf = icl_conn_receive(request->ip_conn,
 313             request->ip_ahs_len);
 314         if (request->ip_ahs_mbuf == NULL) {
 315                 ICL_DEBUG("failed to receive AHS");
 316                 return (-1);
 317         }
 318
 319         *availablep -= request->ip_ahs_len;
 320         return (0);
 321 }
 322
 323 static uint32_t
 324 icl_mbuf_to_crc32c(const struct mbuf *m0)
 325 {
 326         uint32_t digest = 0xffffffff;
 327         const struct mbuf *m;
 328
 329         for (m = m0; m != NULL; m = m->m_next)
 330                 digest = calculate_crc32c(digest,
 331                     mtod(m, const void *), m->m_len);
 332
 333         digest = digest ^ 0xffffffff;
 334
 335         return (digest);
 336 }
 337
 338 static int
 339 icl_pdu_check_header_digest(struct icl_pdu *request, size_t *availablep)
 340 {
 341         struct mbuf *m;
 342         uint32_t received_digest, valid_digest;
 343
 344         if (request->ip_conn->ic_header_crc32c == false)
 345                 return (0);
 346
 347         m = icl_conn_receive(request->ip_conn, ISCSI_HEADER_DIGEST_SIZE);
 348         if (m == NULL) {
 349                 ICL_DEBUG("failed to receive header digest");
 350                 return (-1);
 351         }
 352
 353         CTASSERT(sizeof(received_digest) == ISCSI_HEADER_DIGEST_SIZE);
 354         m_copydata(m, 0, ISCSI_HEADER_DIGEST_SIZE, (void *)&received_digest);
 355         m_freem(m);
 356
 357         *availablep -= ISCSI_HEADER_DIGEST_SIZE;
 358
 359         /*
 360          * XXX: Handle AHS.
 361          */
 362         valid_digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
 363         if (received_digest != valid_digest) {
 364                 ICL_WARN("header digest check failed; got 0x%x, "
 365                     "should be 0x%x", received_digest, valid_digest);
 366                 return (-1);
 367         }
 368
 369         return (0);
 370 }
 371
 372 /*
 373  * Return the number of bytes that should be waiting in the receive socket
 374  * before icl_pdu_receive_data_segment() gets called.
 375  */
 376 static size_t
 377 icl_pdu_data_segment_receive_len(const struct icl_pdu *request)
 378 {
 379         size_t len;
 380
 381         len = icl_pdu_data_segment_length(request);
 382         if (len == 0)
 383                 return (0);
 384
 385         /*
 386          * Account for the parts of data segment already read from
 387          * the socket buffer.
 388          */
 389         KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
 390         len -= request->ip_data_len;
 391
 392         /*
 393          * Don't always wait for the full data segment to be delivered
 394          * to the socket; this might badly affect performance due to
 395          * TCP window scaling.
 396          */
 397         if (len > partial_receive_len) {
 398 #if 0
 399                 ICL_DEBUG("need %zd bytes of data, limiting to %zd",
 400                     len, partial_receive_len));
 401 #endif
 402                 len = partial_receive_len;
 403
 404                 return (len);
 405         }
 406
 407         /*
 408          * Account for padding.  Note that due to the way code is written,
 409          * the icl_pdu_receive_data_segment() must always receive padding
 410          * along with the last part of data segment, because it would be
 411          * impossible to tell whether we've already received the full data
 412          * segment including padding, or without it.
 413          */
 414         if ((len % 4) != 0)
 415                 len += 4 - (len % 4);
 416
 417 #if 0
 418         ICL_DEBUG("need %zd bytes of data", len));
 419 #endif
 420
 421         return (len);
 422 }
 423
 424 static int
 425 icl_pdu_receive_data_segment(struct icl_pdu *request,
 426     size_t *availablep, bool *more_neededp)
 427 {
 428         struct icl_conn *ic;
 429         size_t len, padding = 0;
 430         struct mbuf *m;
 431
 432         ic = request->ip_conn;
 433
 434         *more_neededp = false;
 435         ic->ic_receive_len = 0;
 436
 437         len = icl_pdu_data_segment_length(request);
 438         if (len == 0)
 439                 return (0);
 440
 441         if ((len % 4) != 0)
 442                 padding = 4 - (len % 4);
 443
 444         /*
 445          * Account for already received parts of data segment.
 446          */
 447         KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
 448         len -= request->ip_data_len;
 449
 450         if (len + padding > *availablep) {
 451                 /*
 452                  * Not enough data in the socket buffer.  Receive as much
 453                  * as we can.  Don't receive padding, since, obviously, it's
 454                  * not the end of data segment yet.
 455                  */
 456 #if 0
 457                 ICL_DEBUG("limited from %zd to %zd",
 458                     len + padding, *availablep - padding));
 459 #endif
 460                 len = *availablep - padding;
 461                 *more_neededp = true;
 462                 padding = 0;
 463         }
 464
 465         /*
 466          * Must not try to receive padding without at least one byte
 467          * of actual data segment.
 468          */
 469         if (len > 0) {
 470                 m = icl_conn_receive(request->ip_conn, len + padding);
 471                 if (m == NULL) {
 472                         ICL_DEBUG("failed to receive data segment");
 473                         return (-1);
 474                 }
 475
 476                 if (request->ip_data_mbuf == NULL)
 477                         request->ip_data_mbuf = m;
 478                 else
 479                         m_cat(request->ip_data_mbuf, m);
 480
 481                 request->ip_data_len += len;
 482                 *availablep -= len + padding;
 483         } else
 484                 ICL_DEBUG("len 0");
 485
 486         if (*more_neededp)
 487                 ic->ic_receive_len =
 488                     icl_pdu_data_segment_receive_len(request);
 489
 490         return (0);
 491 }
 492
 493 static int
 494 icl_pdu_check_data_digest(struct icl_pdu *request, size_t *availablep)
 495 {
 496         struct mbuf *m;
 497         uint32_t received_digest, valid_digest;
 498
 499         if (request->ip_conn->ic_data_crc32c == false)
 500                 return (0);
 501
 502         if (request->ip_data_len == 0)
 503                 return (0);
 504
 505         m = icl_conn_receive(request->ip_conn, ISCSI_DATA_DIGEST_SIZE);
 506         if (m == NULL) {
 507                 ICL_DEBUG("failed to receive data digest");
 508                 return (-1);
 509         }
 510
 511         CTASSERT(sizeof(received_digest) == ISCSI_DATA_DIGEST_SIZE);
 512         m_copydata(m, 0, ISCSI_DATA_DIGEST_SIZE, (void *)&received_digest);
 513         m_freem(m);
 514
 515         *availablep -= ISCSI_DATA_DIGEST_SIZE;
 516
 517         /*
 518          * Note that ip_data_mbuf also contains padding; since digest
 519          * calculation is supposed to include that, we iterate over
 520          * the entire ip_data_mbuf chain, not just ip_data_len bytes of it.
 521          */
 522         valid_digest = icl_mbuf_to_crc32c(request->ip_data_mbuf);
 523         if (received_digest != valid_digest) {
 524                 ICL_WARN("data digest check failed; got 0x%x, "
 525                     "should be 0x%x", received_digest, valid_digest);
 526                 return (-1);
 527         }
 528
 529         return (0);
 530 }
 531
 532 /*
 533  * Somewhat contrary to the name, this attempts to receive only one
 534  * "part" of PDU at a time; call it repeatedly until it returns non-NULL.
 535  */
 536 static struct icl_pdu *
 537 icl_conn_receive_pdu(struct icl_conn *ic, size_t *availablep)
 538 {
 539         struct icl_pdu *request;
 540         struct socket *so;
 541         size_t len;
 542         int error;
 543         bool more_needed;
 544
 545         so = ic->ic_socket;
 546
 547         if (ic->ic_receive_state == ICL_CONN_STATE_BHS) {
 548                 KASSERT(ic->ic_receive_pdu == NULL,
 549                     ("ic->ic_receive_pdu != NULL"));
 550                 request = icl_pdu_new(ic, M_NOWAIT);
 551                 if (request == NULL) {
 552                         ICL_DEBUG("failed to allocate PDU; "
 553                             "dropping connection");
 554                         icl_conn_fail(ic);
 555                         return (NULL);
 556                 }
 557                 ic->ic_receive_pdu = request;
 558         } else {
 559                 KASSERT(ic->ic_receive_pdu != NULL,
 560                     ("ic->ic_receive_pdu == NULL"));
 561                 request = ic->ic_receive_pdu;
 562         }
 563
 564         if (*availablep < ic->ic_receive_len) {
 565 #if 0
 566                 ICL_DEBUG("not enough data; need %zd, "
 567                     "have %zd", ic->ic_receive_len, *availablep);
 568 #endif
 569                 return (NULL);
 570         }
 571
 572         switch (ic->ic_receive_state) {
 573         case ICL_CONN_STATE_BHS:
 574                 //ICL_DEBUG("receiving BHS");
 575                 error = icl_pdu_receive_bhs(request, availablep);
 576                 if (error != 0) {
 577                         ICL_DEBUG("failed to receive BHS; "
 578                             "dropping connection");
 579                         break;
 580                 }
 581
 582                 /*
 583                  * We don't enforce any limit for AHS length;
 584                  * its length is stored in 8 bit field.
 585                  */
 586
 587                 len = icl_pdu_data_segment_length(request);
 588                 if (len > ic->ic_max_data_segment_length) {
 589                         ICL_WARN("received data segment "
 590                             "length %zd is larger than negotiated "
 591                             "MaxDataSegmentLength %zd; "
 592                             "dropping connection",
 593                             len, ic->ic_max_data_segment_length);
 594                         error = EINVAL;
 595                         break;
 596                 }
 597
 598                 ic->ic_receive_state = ICL_CONN_STATE_AHS;
 599                 ic->ic_receive_len = icl_pdu_ahs_length(request);
 600                 break;
 601
 602         case ICL_CONN_STATE_AHS:
 603                 //ICL_DEBUG("receiving AHS");
 604                 error = icl_pdu_receive_ahs(request, availablep);
 605                 if (error != 0) {
 606                         ICL_DEBUG("failed to receive AHS; "
 607                             "dropping connection");
 608                         break;
 609                 }
 610                 ic->ic_receive_state = ICL_CONN_STATE_HEADER_DIGEST;
 611                 if (ic->ic_header_crc32c == false)
 612                         ic->ic_receive_len = 0;
 613                 else
 614                         ic->ic_receive_len = ISCSI_HEADER_DIGEST_SIZE;
 615                 break;
 616
 617         case ICL_CONN_STATE_HEADER_DIGEST:
 618                 //ICL_DEBUG("receiving header digest");
 619                 error = icl_pdu_check_header_digest(request, availablep);
 620                 if (error != 0) {
 621                         ICL_DEBUG("header digest failed; "
 622                             "dropping connection");
 623                         break;
 624                 }
 625
 626                 ic->ic_receive_state = ICL_CONN_STATE_DATA;
 627                 ic->ic_receive_len =
 628                     icl_pdu_data_segment_receive_len(request);
 629                 break;
 630
 631         case ICL_CONN_STATE_DATA:
 632                 //ICL_DEBUG("receiving data segment");
 633                 error = icl_pdu_receive_data_segment(request, availablep,
 634                     &more_needed);
 635                 if (error != 0) {
 636                         ICL_DEBUG("failed to receive data segment;"
 637                             "dropping connection");
 638                         break;
 639                 }
 640
 641                 if (more_needed)
 642                         break;
 643
 644                 ic->ic_receive_state = ICL_CONN_STATE_DATA_DIGEST;
 645                 if (request->ip_data_len == 0 || ic->ic_data_crc32c == false)
 646                         ic->ic_receive_len = 0;
 647                 else
 648                         ic->ic_receive_len = ISCSI_DATA_DIGEST_SIZE;
 649                 break;
 650
 651         case ICL_CONN_STATE_DATA_DIGEST:
 652                 //ICL_DEBUG("receiving data digest");
 653                 error = icl_pdu_check_data_digest(request, availablep);
 654                 if (error != 0) {
 655                         ICL_DEBUG("data digest failed; "
 656                             "dropping connection");
 657                         break;
 658                 }
 659
 660                 /*
 661                  * We've received complete PDU; reset the receive state machine
 662                  * and return the PDU.
 663                  */
 664                 ic->ic_receive_state = ICL_CONN_STATE_BHS;
 665                 ic->ic_receive_len = sizeof(struct iscsi_bhs);
 666                 ic->ic_receive_pdu = NULL;
 667                 return (request);
 668
 669         default:
 670                 panic("invalid ic_receive_state %d\n", ic->ic_receive_state);
 671         }
 672
 673         if (error != 0) {
 674                 /*
 675                  * Don't free the PDU; it's pointed to by ic->ic_receive_pdu
 676                  * and will get freed in icl_conn_close().
 677                  */
 678                 icl_conn_fail(ic);
 679         }
 680
 681         return (NULL);
 682 }
 683
 684 static void
 685 icl_conn_receive_pdus(struct icl_conn *ic, size_t available)
 686 {
 687         struct icl_pdu *response;
 688         struct socket *so;
 689
 690         so = ic->ic_socket;
 691
 692         /*
 693          * This can never happen; we're careful to only mess with ic->ic_socket
 694          * pointer when the send/receive threads are not running.
 695          */
 696         KASSERT(so != NULL, ("NULL socket"));
 697
 698         for (;;) {
 699                 if (ic->ic_disconnecting)
 700                         return;
 701
 702                 if (so->so_error != 0) {
 703                         ICL_DEBUG("connection error %d; "
 704                             "dropping connection", so->so_error);
 705                         icl_conn_fail(ic);
 706                         return;
 707                 }
 708
 709                 /*
 710                  * Loop until we have a complete PDU or there is not enough
 711                  * data in the socket buffer.
 712                  */
 713                 if (available < ic->ic_receive_len) {
 714 #if 0
 715                         ICL_DEBUG("not enough data; have %zd, "
 716                             "need %zd", available,
 717                             ic->ic_receive_len);
 718 #endif
 719                         return;
 720                 }
 721
 722                 response = icl_conn_receive_pdu(ic, &available);
 723                 if (response == NULL)
 724                         continue;
 725
 726                 if (response->ip_ahs_len > 0) {
 727                         ICL_WARN("received PDU with unsupported "
 728                             "AHS; opcode 0x%x; dropping connection",
 729                             response->ip_bhs->bhs_opcode);
 730                         icl_pdu_free(response);
 731                         icl_conn_fail(ic);
 732                         return;
 733                 }
 734
 735                 (ic->ic_receive)(response);
 736         }
 737 }
 738
 739 static void
 740 icl_receive_thread(void *arg)
 741 {
 742         struct icl_conn *ic;
 743         size_t available;
 744         struct socket *so;
 745
 746         ic = arg;
 747         so = ic->ic_socket;
 748
 749         ICL_CONN_LOCK(ic);
 750         ic->ic_receive_running = true;
 751         ICL_CONN_UNLOCK(ic);
 752
 753         for (;;) {
 754                 if (ic->ic_disconnecting) {
 755                         //ICL_DEBUG("terminating");
 756                         break;
 757                 }
 758
 759                 /*
 760                  * Set the low watermark, to be checked by
 761                  * soreadable() in icl_soupcall_receive()
 762                  * to avoid unneccessary wakeups until there
 763                  * is enough data received to read the PDU.
 764                  */
 765                 SOCKBUF_LOCK(&so->so_rcv);
 766                 available = so->so_rcv.sb_cc;
 767                 if (available < ic->ic_receive_len) {
 768                         so->so_rcv.sb_lowat = ic->ic_receive_len;
 769                         cv_wait(&ic->ic_receive_cv, &so->so_rcv.sb_mtx);
 770                 } else
 771                         so->so_rcv.sb_lowat = so->so_rcv.sb_hiwat + 1;
 772                 SOCKBUF_UNLOCK(&so->so_rcv);
 773
 774                 icl_conn_receive_pdus(ic, available);
 775         }
 776
 777         ICL_CONN_LOCK(ic);
 778         ic->ic_receive_running = false;
 779         ICL_CONN_UNLOCK(ic);
 780         kthread_exit();
 781 }
 782
 783 static int
 784 icl_soupcall_receive(struct socket *so, void *arg, int waitflag)
 785 {
 786         struct icl_conn *ic;
 787
 788         if (!soreadable(so))
 789                 return (SU_OK);
 790
 791         ic = arg;
 792         cv_signal(&ic->ic_receive_cv);
 793         return (SU_OK);
 794 }
 795
 796 static int
 797 icl_pdu_finalize(struct icl_pdu *request)
 798 {
 799         size_t padding, pdu_len;
 800         uint32_t digest, zero = 0;
 801         int ok;
 802         struct icl_conn *ic;
 803
 804         ic = request->ip_conn;
 805
 806         icl_pdu_set_data_segment_length(request, request->ip_data_len);
 807
 808         pdu_len = icl_pdu_size(request);
 809
 810         if (ic->ic_header_crc32c) {
 811                 digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
 812                 ok = m_append(request->ip_bhs_mbuf, sizeof(digest),
 813                     (void *)&digest);
 814                 if (ok != 1) {
 815                         ICL_WARN("failed to append header digest");
 816                         return (1);
 817                 }
 818         }
 819
 820         if (request->ip_data_len != 0) {
 821                 padding = icl_pdu_padding(request);
 822                 if (padding > 0) {
 823                         ok = m_append(request->ip_data_mbuf, padding,
 824                             (void *)&zero);
 825                         if (ok != 1) {
 826                                 ICL_WARN("failed to append padding");
 827                                 return (1);
 828                         }
 829                 }
 830
 831                 if (ic->ic_data_crc32c) {
 832                         digest = icl_mbuf_to_crc32c(request->ip_data_mbuf);
 833
 834                         ok = m_append(request->ip_data_mbuf, sizeof(digest),
 835                             (void *)&digest);
 836                         if (ok != 1) {
 837                                 ICL_WARN("failed to append data digest");
 838                                 return (1);
 839                         }
 840                 }
 841
 842                 m_cat(request->ip_bhs_mbuf, request->ip_data_mbuf);
 843                 request->ip_data_mbuf = NULL;
 844         }
 845
 846         request->ip_bhs_mbuf->m_pkthdr.len = pdu_len;
 847
 848         return (0);
 849 }
 850
 851 static void
 852 icl_conn_send_pdus(struct icl_conn *ic, struct icl_pdu_stailq *queue)
 853 {
 854         struct icl_pdu *request, *request2;
 855         struct socket *so;
 856         size_t available, size, size2;
 857         int coalesced, error;
 858
 859         ICL_CONN_LOCK_ASSERT_NOT(ic);
 860
 861         so = ic->ic_socket;
 862
 863         SOCKBUF_LOCK(&so->so_snd);
 864         /*
 865          * Check how much space do we have for transmit.  We can't just
 866          * call sosend() and retry when we get EWOULDBLOCK or EMSGSIZE,
 867          * as it always frees the mbuf chain passed to it, even in case
 868          * of error.
 869          */
 870         available = sbspace(&so->so_snd);
 871
 872         /*
 873          * Notify the socket upcall that we don't need wakeups
 874          * for the time being.
 875          */
 876         so->so_snd.sb_lowat = so->so_snd.sb_hiwat + 1;
 877         SOCKBUF_UNLOCK(&so->so_snd);
 878
 879         while (!STAILQ_EMPTY(queue)) {
 880                 if (ic->ic_disconnecting)
 881                         return;
 882                 request = STAILQ_FIRST(queue);
 883                 size = icl_pdu_size(request);
 884                 if (available < size) {
 885
 886                         /*
 887                          * Set the low watermark, to be checked by
 888                          * sowriteable() in icl_soupcall_send()
 889                          * to avoid unneccessary wakeups until there
 890                          * is enough space for the PDU to fit.
 891                          */
 892                         SOCKBUF_LOCK(&so->so_snd);
 893                         available = sbspace(&so->so_snd);
 894                         if (available < size) {
 895 #if 1
 896                                 ICL_DEBUG("no space to send; "
 897                                     "have %zd, need %zd",
 898                                     available, size);
 899 #endif
 900                                 so->so_snd.sb_lowat = size;
 901                                 SOCKBUF_UNLOCK(&so->so_snd);
 902                                 return;
 903                         }
 904                         SOCKBUF_UNLOCK(&so->so_snd);
 905                 }
 906                 STAILQ_REMOVE_HEAD(queue, ip_next);
 907                 error = icl_pdu_finalize(request);
 908                 if (error != 0) {
 909                         ICL_DEBUG("failed to finalize PDU; "
 910                             "dropping connection");
 911                         icl_conn_fail(ic);
 912                         icl_pdu_free(request);
 913                         return;
 914                 }
 915                 if (coalesce) {
 916                         coalesced = 1;
 917                         for (;;) {
 918                                 request2 = STAILQ_FIRST(queue);
 919                                 if (request2 == NULL)
 920                                         break;
 921                                 size2 = icl_pdu_size(request2);
 922                                 if (available < size + size2)
 923                                         break;
 924                                 STAILQ_REMOVE_HEAD(queue, ip_next);
 925                                 error = icl_pdu_finalize(request2);
 926                                 if (error != 0) {
 927                                         ICL_DEBUG("failed to finalize PDU; "
 928                                             "dropping connection");
 929                                         icl_conn_fail(ic);
 930                                         icl_pdu_free(request);
 931                                         icl_pdu_free(request2);
 932                                         return;
 933                                 }
 934                                 m_cat(request->ip_bhs_mbuf, request2->ip_bhs_mbuf);
 935                                 request2->ip_bhs_mbuf = NULL;
 936                                 request->ip_bhs_mbuf->m_pkthdr.len += size2;
 937                                 size += size2;
 938                                 STAILQ_REMOVE_AFTER(queue, request, ip_next);
 939                                 icl_pdu_free(request2);
 940                                 coalesced++;
 941                         }
 942 #if 0
 943                         if (coalesced > 1) {
 944                                 ICL_DEBUG("coalesced %d PDUs into %zd bytes",
 945                                     coalesced, size);
 946                         }
 947 #endif
 948                 }
 949                 available -= size;
 950                 error = sosend(so, NULL, NULL, request->ip_bhs_mbuf,
 951                     NULL, MSG_DONTWAIT, curthread);
 952                 request->ip_bhs_mbuf = NULL; /* Sosend consumes the mbuf. */
 953                 if (error != 0) {
 954                         ICL_DEBUG("failed to send PDU, error %d; "
 955                             "dropping connection", error);
 956                         icl_conn_fail(ic);
 957                         icl_pdu_free(request);
 958                         return;
 959                 }
 960                 icl_pdu_free(request);
 961         }
 962 }
 963
 964 static void
 965 icl_send_thread(void *arg)
 966 {
 967         struct icl_conn *ic;
 968         struct icl_pdu_stailq queue;
 969
 970         ic = arg;
 971
 972         STAILQ_INIT(&queue);
 973
 974         ICL_CONN_LOCK(ic);
 975         ic->ic_send_running = true;
 976
 977         for (;;) {
 978                 if (ic->ic_disconnecting) {
 979                         //ICL_DEBUG("terminating");
 980                         break;
 981                 }
 982
 983                 for (;;) {
 984                         /*
 985                          * If the local queue is empty, populate it from
 986                          * the main one.  This way the icl_conn_send_pdus()
 987                          * can go through all the queued PDUs without holding
 988                          * any locks.
 989                          */
 990                         if (STAILQ_EMPTY(&queue))
 991                                 STAILQ_SWAP(&ic->ic_to_send, &queue, icl_pdu);
 992
 993                         ic->ic_check_send_space = false;
 994                         ICL_CONN_UNLOCK(ic);
 995                         icl_conn_send_pdus(ic, &queue);
 996                         ICL_CONN_LOCK(ic);
 997
 998                         /*
 999                          * The icl_soupcall_send() was called since the last
1000                          * call to sbspace(); go around;
1001                          */
1002                         if (ic->ic_check_send_space)
1003                                 continue;
1004
1005                         /*
1006                          * Local queue is empty, but we still have PDUs
1007                          * in the main one; go around.
1008                          */
1009                         if (STAILQ_EMPTY(&queue) &&
1010                             !STAILQ_EMPTY(&ic->ic_to_send))
1011                                 continue;
1012
1013                         /*
1014                          * There might be some stuff in the local queue,
1015                          * which didn't get sent due to not having enough send
1016                          * space.  Wait for socket upcall.
1017                          */
1018                         break;
1019                 }
1020
1021                 cv_wait(&ic->ic_send_cv, ic->ic_lock);
1022         }
1023
1024         /*
1025          * We're exiting; move PDUs back to the main queue, so they can
1026          * get freed properly.  At this point ordering doesn't matter.
1027          */
1028         STAILQ_CONCAT(&ic->ic_to_send, &queue);
1029
1030         ic->ic_send_running = false;
1031         ICL_CONN_UNLOCK(ic);
1032         kthread_exit();
1033 }
1034
1035 static int
1036 icl_soupcall_send(struct socket *so, void *arg, int waitflag)
1037 {
1038         struct icl_conn *ic;
1039
1040         if (!sowriteable(so))
1041                 return (SU_OK);
1042
1043         ic = arg;
1044
1045         ICL_CONN_LOCK(ic);
1046         ic->ic_check_send_space = true;
1047         ICL_CONN_UNLOCK(ic);
1048
1049         cv_signal(&ic->ic_send_cv);
1050
1051         return (SU_OK);
1052 }
1053
1054 int
1055 icl_pdu_append_data(struct icl_pdu *request, const void *addr, size_t len,
1056     int flags)
1057 {
1058         struct mbuf *mb, *newmb;
1059         size_t copylen, off = 0;
1060
1061         KASSERT(len > 0, ("len == 0"));
1062
1063         newmb = m_getm2(NULL, len, flags, MT_DATA, M_PKTHDR);
1064         if (newmb == NULL) {
1065                 ICL_WARN("failed to allocate mbuf for %zd bytes", len);
1066                 return (ENOMEM);
1067         }
1068
1069         for (mb = newmb; mb != NULL; mb = mb->m_next) {
1070                 copylen = min(M_TRAILINGSPACE(mb), len - off);
1071                 memcpy(mtod(mb, char *), (const char *)addr + off, copylen);
1072                 mb->m_len = copylen;
1073                 off += copylen;
1074         }
1075         KASSERT(off == len, ("%s: off != len", __func__));
1076
1077         if (request->ip_data_mbuf == NULL) {
1078                 request->ip_data_mbuf = newmb;
1079                 request->ip_data_len = len;
1080         } else {
1081                 m_cat(request->ip_data_mbuf, newmb);
1082                 request->ip_data_len += len;
1083         }
1084
1085         return (0);
1086 }
1087
1088 void
1089 icl_pdu_get_data(struct icl_pdu *ip, size_t off, void *addr, size_t len)
1090 {
1091
1092         m_copydata(ip->ip_data_mbuf, off, len, addr);
1093 }
1094
1095 void
1096 icl_pdu_queue(struct icl_pdu *ip)
1097 {
1098         struct icl_conn *ic;
1099
1100         ic = ip->ip_conn;
1101
1102         ICL_CONN_LOCK_ASSERT(ic);
1103
1104         if (ic->ic_disconnecting || ic->ic_socket == NULL) {
1105                 ICL_DEBUG("icl_pdu_queue on closed connection");
1106                 icl_pdu_free(ip);
1107                 return;
1108         }
1109
1110         if (!STAILQ_EMPTY(&ic->ic_to_send)) {
1111                 STAILQ_INSERT_TAIL(&ic->ic_to_send, ip, ip_next);
1112                 /*
1113                  * If the queue is not empty, someone else had already
1114                  * signaled the send thread; no need to do that again,
1115                  * just return.
1116                  */
1117                 return;
1118         }
1119
1120         STAILQ_INSERT_TAIL(&ic->ic_to_send, ip, ip_next);
1121         cv_signal(&ic->ic_send_cv);
1122 }
1123
1124 struct icl_conn *
1125 icl_conn_new(const char *name, struct mtx *lock)
1126 {
1127         struct icl_conn *ic;
1128
1129         refcount_acquire(&icl_ncons);
1130
1131         ic = uma_zalloc(icl_conn_zone, M_WAITOK | M_ZERO);
1132
1133         STAILQ_INIT(&ic->ic_to_send);
1134         ic->ic_lock = lock;
1135         cv_init(&ic->ic_send_cv, "icl_tx");
1136         cv_init(&ic->ic_receive_cv, "icl_rx");
1137 #ifdef DIAGNOSTIC
1138         refcount_init(&ic->ic_outstanding_pdus, 0);
1139 #endif
1140         ic->ic_max_data_segment_length = ICL_MAX_DATA_SEGMENT_LENGTH;
1141         ic->ic_name = name;
1142
1143         return (ic);
1144 }
1145
1146 void
1147 icl_conn_free(struct icl_conn *ic)
1148 {
1149
1150         cv_destroy(&ic->ic_send_cv);
1151         cv_destroy(&ic->ic_receive_cv);
1152         uma_zfree(icl_conn_zone, ic);
1153         refcount_release(&icl_ncons);
1154 }
1155
1156 static int
1157 icl_conn_start(struct icl_conn *ic)
1158 {
1159         size_t minspace;
1160         struct sockopt opt;
1161         int error, one = 1;
1162
1163         ICL_CONN_LOCK(ic);
1164
1165         /*
1166          * XXX: Ugly hack.
1167          */
1168         if (ic->ic_socket == NULL) {
1169                 ICL_CONN_UNLOCK(ic);
1170                 return (EINVAL);
1171         }
1172
1173         ic->ic_receive_state = ICL_CONN_STATE_BHS;
1174         ic->ic_receive_len = sizeof(struct iscsi_bhs);
1175         ic->ic_disconnecting = false;
1176
1177         ICL_CONN_UNLOCK(ic);
1178
1179         /*
1180          * For sendspace, this is required because the current code cannot
1181          * send a PDU in pieces; thus, the minimum buffer size is equal
1182          * to the maximum PDU size.  "+4" is to account for possible padding.
1183          *
1184          * What we should actually do here is to use autoscaling, but set
1185          * some minimal buffer size to "minspace".  I don't know a way to do
1186          * that, though.
1187          */
1188         minspace = sizeof(struct iscsi_bhs) + ic->ic_max_data_segment_length +
1189             ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE + 4;
1190         if (sendspace < minspace) {
1191                 ICL_WARN("kern.icl.sendspace too low; must be at least %zd",
1192                     minspace);
1193                 sendspace = minspace;
1194         }
1195         if (recvspace < minspace) {
1196                 ICL_WARN("kern.icl.recvspace too low; must be at least %zd",
1197                     minspace);
1198                 recvspace = minspace;
1199         }
1200
1201         error = soreserve(ic->ic_socket, sendspace, recvspace);
1202         if (error != 0) {
1203                 ICL_WARN("soreserve failed with error %d", error);
1204                 icl_conn_close(ic);
1205                 return (error);
1206         }
1207
1208         /*
1209          * Disable Nagle.
1210          */
1211         bzero(&opt, sizeof(opt));
1212         opt.sopt_dir = SOPT_SET;
1213         opt.sopt_level = IPPROTO_TCP;
1214         opt.sopt_name = TCP_NODELAY;
1215         opt.sopt_val = &one;
1216         opt.sopt_valsize = sizeof(one);
1217         error = sosetopt(ic->ic_socket, &opt);
1218         if (error != 0) {
1219                 ICL_WARN("disabling TCP_NODELAY failed with error %d", error);
1220                 icl_conn_close(ic);
1221                 return (error);
1222         }
1223
1224         /*
1225          * Start threads.
1226          */
1227         error = kthread_add(icl_send_thread, ic, NULL, NULL, 0, 0, "%stx",
1228             ic->ic_name);
1229         if (error != 0) {
1230                 ICL_WARN("kthread_add(9) failed with error %d", error);
1231                 icl_conn_close(ic);
1232                 return (error);
1233         }
1234
1235         error = kthread_add(icl_receive_thread, ic, NULL, NULL, 0, 0, "%srx",
1236             ic->ic_name);
1237         if (error != 0) {
1238                 ICL_WARN("kthread_add(9) failed with error %d", error);
1239                 icl_conn_close(ic);
1240                 return (error);
1241         }
1242
1243         /*
1244          * Register socket upcall, to get notified about incoming PDUs
1245          * and free space to send outgoing ones.
1246          */
1247         SOCKBUF_LOCK(&ic->ic_socket->so_snd);
1248         soupcall_set(ic->ic_socket, SO_SND, icl_soupcall_send, ic);
1249         SOCKBUF_UNLOCK(&ic->ic_socket->so_snd);
1250         SOCKBUF_LOCK(&ic->ic_socket->so_rcv);
1251         soupcall_set(ic->ic_socket, SO_RCV, icl_soupcall_receive, ic);
1252         SOCKBUF_UNLOCK(&ic->ic_socket->so_rcv);
1253
1254         return (0);
1255 }
1256
1257 int
1258 icl_conn_handoff(struct icl_conn *ic, int fd)
1259 {
1260         struct file *fp;
1261         struct socket *so;
1262         cap_rights_t rights;
1263         int error;
1264
1265         ICL_CONN_LOCK_ASSERT_NOT(ic);
1266
1267         /*
1268          * Steal the socket from userland.
1269          */
1270         error = fget(curthread, fd,
1271             cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1272         if (error != 0)
1273                 return (error);
1274         if (fp->f_type != DTYPE_SOCKET) {
1275                 fdrop(fp, curthread);
1276                 return (EINVAL);
1277         }
1278         so = fp->f_data;
1279         if (so->so_type != SOCK_STREAM) {
1280                 fdrop(fp, curthread);
1281                 return (EINVAL);
1282         }
1283
1284         ICL_CONN_LOCK(ic);
1285
1286         if (ic->ic_socket != NULL) {
1287                 ICL_CONN_UNLOCK(ic);
1288                 fdrop(fp, curthread);
1289                 return (EBUSY);
1290         }
1291
1292         ic->ic_socket = fp->f_data;
1293         fp->f_ops = &badfileops;
1294         fp->f_data = NULL;
1295         fdrop(fp, curthread);
1296         ICL_CONN_UNLOCK(ic);
1297
1298         error = icl_conn_start(ic);
1299
1300         return (error);
1301 }
1302
1303 void
1304 icl_conn_shutdown(struct icl_conn *ic)
1305 {
1306         ICL_CONN_LOCK_ASSERT_NOT(ic);
1307
1308         ICL_CONN_LOCK(ic);
1309         if (ic->ic_socket == NULL) {
1310                 ICL_CONN_UNLOCK(ic);
1311                 return;
1312         }
1313         ICL_CONN_UNLOCK(ic);
1314
1315         soshutdown(ic->ic_socket, SHUT_RDWR);
1316 }
1317
1318 void
1319 icl_conn_close(struct icl_conn *ic)
1320 {
1321         struct icl_pdu *pdu;
1322
1323         ICL_CONN_LOCK_ASSERT_NOT(ic);
1324
1325         ICL_CONN_LOCK(ic);
1326         if (ic->ic_socket == NULL) {
1327                 ICL_CONN_UNLOCK(ic);
1328                 return;
1329         }
1330
1331         /*
1332          * Deregister socket upcalls.
1333          */
1334         ICL_CONN_UNLOCK(ic);
1335         SOCKBUF_LOCK(&ic->ic_socket->so_snd);
1336         if (ic->ic_socket->so_snd.sb_upcall != NULL)
1337                 soupcall_clear(ic->ic_socket, SO_SND);
1338         SOCKBUF_UNLOCK(&ic->ic_socket->so_snd);
1339         SOCKBUF_LOCK(&ic->ic_socket->so_rcv);
1340         if (ic->ic_socket->so_rcv.sb_upcall != NULL)
1341                 soupcall_clear(ic->ic_socket, SO_RCV);
1342         SOCKBUF_UNLOCK(&ic->ic_socket->so_rcv);
1343         ICL_CONN_LOCK(ic);
1344
1345         ic->ic_disconnecting = true;
1346
1347         /*
1348          * Wake up the threads, so they can properly terminate.
1349          */
1350         cv_signal(&ic->ic_receive_cv);
1351         cv_signal(&ic->ic_send_cv);
1352         while (ic->ic_receive_running || ic->ic_send_running) {
1353                 //ICL_DEBUG("waiting for send/receive threads to terminate");
1354                 ICL_CONN_UNLOCK(ic);
1355                 cv_signal(&ic->ic_receive_cv);
1356                 cv_signal(&ic->ic_send_cv);
1357                 pause("icl_close", 1 * hz);
1358                 ICL_CONN_LOCK(ic);
1359         }
1360         //ICL_DEBUG("send/receive threads terminated");
1361
1362         ICL_CONN_UNLOCK(ic);
1363         soclose(ic->ic_socket);
1364         ICL_CONN_LOCK(ic);
1365         ic->ic_socket = NULL;
1366
1367         if (ic->ic_receive_pdu != NULL) {
1368                 //ICL_DEBUG("freeing partially received PDU");
1369                 icl_pdu_free(ic->ic_receive_pdu);
1370                 ic->ic_receive_pdu = NULL;
1371         }
1372
1373         /*
1374          * Remove any outstanding PDUs from the send queue.
1375          */
1376         while (!STAILQ_EMPTY(&ic->ic_to_send)) {
1377                 pdu = STAILQ_FIRST(&ic->ic_to_send);
1378                 STAILQ_REMOVE_HEAD(&ic->ic_to_send, ip_next);
1379                 icl_pdu_free(pdu);
1380         }
1381
1382         KASSERT(STAILQ_EMPTY(&ic->ic_to_send),
1383             ("destroying session with non-empty send queue"));
1384 #ifdef DIAGNOSTIC
1385         KASSERT(ic->ic_outstanding_pdus == 0,
1386             ("destroying session with %d outstanding PDUs",
1387              ic->ic_outstanding_pdus));
1388 #endif
1389         ICL_CONN_UNLOCK(ic);
1390 }
1391
1392 bool
1393 icl_conn_connected(struct icl_conn *ic)
1394 {
1395         ICL_CONN_LOCK_ASSERT_NOT(ic);
1396
1397         ICL_CONN_LOCK(ic);
1398         if (ic->ic_socket == NULL) {
1399                 ICL_CONN_UNLOCK(ic);
1400                 return (false);
1401         }
1402         if (ic->ic_socket->so_error != 0) {
1403                 ICL_CONN_UNLOCK(ic);
1404                 return (false);
1405         }
1406         ICL_CONN_UNLOCK(ic);
1407         return (true);
1408 }
1409
1410 #ifdef ICL_KERNEL_PROXY
1411 int
1412 icl_conn_handoff_sock(struct icl_conn *ic, struct socket *so)
1413 {
1414         int error;
1415
1416         ICL_CONN_LOCK_ASSERT_NOT(ic);
1417
1418         if (so->so_type != SOCK_STREAM)
1419                 return (EINVAL);
1420
1421         ICL_CONN_LOCK(ic);
1422         if (ic->ic_socket != NULL) {
1423                 ICL_CONN_UNLOCK(ic);
1424                 return (EBUSY);
1425         }
1426         ic->ic_socket = so;
1427         ICL_CONN_UNLOCK(ic);
1428
1429         error = icl_conn_start(ic);
1430
1431         return (error);
1432 }
1433 #endif /* ICL_KERNEL_PROXY */
1434
1435 static int
1436 icl_unload(void)
1437 {
1438
1439         if (icl_ncons != 0)
1440                 return (EBUSY);
1441
1442         uma_zdestroy(icl_conn_zone);
1443         uma_zdestroy(icl_pdu_zone);
1444
1445         return (0);
1446 }
1447
1448 static void
1449 icl_load(void)
1450 {
1451
1452         icl_conn_zone = uma_zcreate("icl_conn",
1453             sizeof(struct icl_conn), NULL, NULL, NULL, NULL,
1454             UMA_ALIGN_PTR, 0);
1455         icl_pdu_zone = uma_zcreate("icl_pdu",
1456             sizeof(struct icl_pdu), NULL, NULL, NULL, NULL,
1457             UMA_ALIGN_PTR, 0);
1458
1459         refcount_init(&icl_ncons, 0);
1460 }
1461
1462 static int
1463 icl_modevent(module_t mod, int what, void *arg)
1464 {
1465
1466         switch (what) {
1467         case MOD_LOAD:
1468                 icl_load();
1469                 return (0);
1470         case MOD_UNLOAD:
1471                 return (icl_unload());
1472         default:
1473                 return (EINVAL);
1474         }
1475 }
1476
1477 moduledata_t icl_data = {
1478         "icl",
1479         icl_modevent,
1480         0
1481 };
1482
1483 DECLARE_MODULE(icl, icl_data, SI_SUB_DRIVERS, SI_ORDER_FIRST);
1484 MODULE_VERSION(icl, 1);