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