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