sys/dev/iscsi/icl_soft.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  * Software implementation of iSCSI Common Layer kobj(9) interface.
  33  */
  34
  35 #include <sys/cdefs.h>
  36 __FBSDID("$FreeBSD$");
  37
  38 #include <sys/param.h>
  39 #include <sys/capsicum.h>
  40 #include <sys/condvar.h>
  41 #include <sys/conf.h>
  42 #include <sys/file.h>
  43 #include <sys/kernel.h>
  44 #include <sys/kthread.h>
  45 #include <sys/lock.h>
  46 #include <sys/mbuf.h>
  47 #include <sys/mutex.h>
  48 #include <sys/module.h>
  49 #include <sys/protosw.h>
  50 #include <sys/socket.h>
  51 #include <sys/socketvar.h>
  52 #include <sys/sysctl.h>
  53 #include <sys/systm.h>
  54 #include <sys/sx.h>
  55 #include <sys/uio.h>
  56 #include <vm/uma.h>
  57 #include <netinet/in.h>
  58 #include <netinet/tcp.h>
  59
  60 #include <dev/iscsi/icl.h>
  61 #include <dev/iscsi/iscsi_proto.h>
  62 #include <icl_conn_if.h>
  63
  64 static int coalesce = 1;
  65 SYSCTL_INT(_kern_icl, OID_AUTO, coalesce, CTLFLAG_RWTUN,
  66     &coalesce, 0, "Try to coalesce PDUs before sending");
  67 static int partial_receive_len = 128 * 1024;
  68 SYSCTL_INT(_kern_icl, OID_AUTO, partial_receive_len, CTLFLAG_RWTUN,
  69     &partial_receive_len, 0, "Minimum read size for partially received "
  70     "data segment");
  71 static int sendspace = 1048576;
  72 SYSCTL_INT(_kern_icl, OID_AUTO, sendspace, CTLFLAG_RWTUN,
  73     &sendspace, 0, "Default send socket buffer size");
  74 static int recvspace = 1048576;
  75 SYSCTL_INT(_kern_icl, OID_AUTO, recvspace, CTLFLAG_RWTUN,
  76     &recvspace, 0, "Default receive socket buffer size");
  77
  78 static MALLOC_DEFINE(M_ICL_SOFT, "icl_soft", "iSCSI software backend");
  79 static uma_zone_t icl_pdu_zone;
  80
  81 static volatile u_int   icl_ncons;
  82
  83 #define ICL_CONN_LOCK(X)                mtx_lock(X->ic_lock)
  84 #define ICL_CONN_UNLOCK(X)              mtx_unlock(X->ic_lock)
  85 #define ICL_CONN_LOCK_ASSERT(X)         mtx_assert(X->ic_lock, MA_OWNED)
  86 #define ICL_CONN_LOCK_ASSERT_NOT(X)     mtx_assert(X->ic_lock, MA_NOTOWNED)
  87
  88 STAILQ_HEAD(icl_pdu_stailq, icl_pdu);
  89
  90 static icl_conn_new_pdu_t       icl_soft_conn_new_pdu;
  91 static icl_conn_pdu_free_t      icl_soft_conn_pdu_free;
  92 static icl_conn_pdu_data_segment_length_t
  93                                     icl_soft_conn_pdu_data_segment_length;
  94 static icl_conn_pdu_append_data_t       icl_soft_conn_pdu_append_data;
  95 static icl_conn_pdu_get_data_t  icl_soft_conn_pdu_get_data;
  96 static icl_conn_pdu_queue_t     icl_soft_conn_pdu_queue;
  97 static icl_conn_handoff_t       icl_soft_conn_handoff;
  98 static icl_conn_free_t          icl_soft_conn_free;
  99 static icl_conn_close_t         icl_soft_conn_close;
 100 static icl_conn_connected_t     icl_soft_conn_connected;
 101
 102 static kobj_method_t icl_soft_methods[] = {
 103         KOBJMETHOD(icl_conn_new_pdu, icl_soft_conn_new_pdu),
 104         KOBJMETHOD(icl_conn_pdu_free, icl_soft_conn_pdu_free),
 105         KOBJMETHOD(icl_conn_pdu_data_segment_length,
 106             icl_soft_conn_pdu_data_segment_length),
 107         KOBJMETHOD(icl_conn_pdu_append_data, icl_soft_conn_pdu_append_data),
 108         KOBJMETHOD(icl_conn_pdu_get_data, icl_soft_conn_pdu_get_data),
 109         KOBJMETHOD(icl_conn_pdu_queue, icl_soft_conn_pdu_queue),
 110         KOBJMETHOD(icl_conn_handoff, icl_soft_conn_handoff),
 111         KOBJMETHOD(icl_conn_free, icl_soft_conn_free),
 112         KOBJMETHOD(icl_conn_close, icl_soft_conn_close),
 113         KOBJMETHOD(icl_conn_connected, icl_soft_conn_connected),
 114         { 0, 0 }
 115 };
 116
 117 DEFINE_CLASS(icl_soft, icl_soft_methods, sizeof(struct icl_conn));
 118
 119 static void
 120 icl_conn_fail(struct icl_conn *ic)
 121 {
 122         if (ic->ic_socket == NULL)
 123                 return;
 124
 125         /*
 126          * XXX
 127          */
 128         ic->ic_socket->so_error = EDOOFUS;
 129         (ic->ic_error)(ic);
 130 }
 131
 132 static struct mbuf *
 133 icl_conn_receive(struct icl_conn *ic, size_t len)
 134 {
 135         struct uio uio;
 136         struct socket *so;
 137         struct mbuf *m;
 138         int error, flags;
 139
 140         so = ic->ic_socket;
 141
 142         memset(&uio, 0, sizeof(uio));
 143         uio.uio_resid = len;
 144
 145         flags = MSG_DONTWAIT;
 146         error = soreceive(so, NULL, &uio, &m, NULL, &flags);
 147         if (error != 0) {
 148                 ICL_DEBUG("soreceive error %d", error);
 149                 return (NULL);
 150         }
 151         if (uio.uio_resid != 0) {
 152                 m_freem(m);
 153                 ICL_DEBUG("short read");
 154                 return (NULL);
 155         }
 156
 157         return (m);
 158 }
 159
 160 static struct icl_pdu *
 161 icl_pdu_new_empty(struct icl_conn *ic, int flags)
 162 {
 163         struct icl_pdu *ip;
 164
 165 #ifdef DIAGNOSTIC
 166         refcount_acquire(&ic->ic_outstanding_pdus);
 167 #endif
 168         ip = uma_zalloc(icl_pdu_zone, flags | M_ZERO);
 169         if (ip == NULL) {
 170                 ICL_WARN("failed to allocate %zd bytes", sizeof(*ip));
 171 #ifdef DIAGNOSTIC
 172                 refcount_release(&ic->ic_outstanding_pdus);
 173 #endif
 174                 return (NULL);
 175         }
 176
 177         ip->ip_conn = ic;
 178
 179         return (ip);
 180 }
 181
 182 static void
 183 icl_pdu_free(struct icl_pdu *ip)
 184 {
 185         struct icl_conn *ic;
 186
 187         ic = ip->ip_conn;
 188
 189         m_freem(ip->ip_bhs_mbuf);
 190         m_freem(ip->ip_ahs_mbuf);
 191         m_freem(ip->ip_data_mbuf);
 192         uma_zfree(icl_pdu_zone, ip);
 193 #ifdef DIAGNOSTIC
 194         refcount_release(&ic->ic_outstanding_pdus);
 195 #endif
 196 }
 197
 198 void
 199 icl_soft_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip)
 200 {
 201
 202         icl_pdu_free(ip);
 203 }
 204
 205 /*
 206  * Allocate icl_pdu with empty BHS to fill up by the caller.
 207  */
 208 struct icl_pdu *
 209 icl_soft_conn_new_pdu(struct icl_conn *ic, int flags)
 210 {
 211         struct icl_pdu *ip;
 212
 213         ip = icl_pdu_new_empty(ic, flags);
 214         if (ip == NULL)
 215                 return (NULL);
 216
 217         ip->ip_bhs_mbuf = m_getm2(NULL, sizeof(struct iscsi_bhs),
 218             flags, MT_DATA, M_PKTHDR);
 219         if (ip->ip_bhs_mbuf == NULL) {
 220                 ICL_WARN("failed to allocate %zd bytes", sizeof(*ip));
 221                 icl_pdu_free(ip);
 222                 return (NULL);
 223         }
 224         ip->ip_bhs = mtod(ip->ip_bhs_mbuf, struct iscsi_bhs *);
 225         memset(ip->ip_bhs, 0, sizeof(struct iscsi_bhs));
 226         ip->ip_bhs_mbuf->m_len = sizeof(struct iscsi_bhs);
 227
 228         return (ip);
 229 }
 230
 231 static int
 232 icl_pdu_ahs_length(const struct icl_pdu *request)
 233 {
 234
 235         return (request->ip_bhs->bhs_total_ahs_len * 4);
 236 }
 237
 238 static size_t
 239 icl_pdu_data_segment_length(const struct icl_pdu *request)
 240 {
 241         uint32_t len = 0;
 242
 243         len += request->ip_bhs->bhs_data_segment_len[0];
 244         len <<= 8;
 245         len += request->ip_bhs->bhs_data_segment_len[1];
 246         len <<= 8;
 247         len += request->ip_bhs->bhs_data_segment_len[2];
 248
 249         return (len);
 250 }
 251
 252 size_t
 253 icl_soft_conn_pdu_data_segment_length(struct icl_conn *ic,
 254     const struct icl_pdu *request)
 255 {
 256
 257         return (icl_pdu_data_segment_length(request));
 258 }
 259
 260 static void
 261 icl_pdu_set_data_segment_length(struct icl_pdu *response, uint32_t len)
 262 {
 263
 264         response->ip_bhs->bhs_data_segment_len[2] = len;
 265         response->ip_bhs->bhs_data_segment_len[1] = len >> 8;
 266         response->ip_bhs->bhs_data_segment_len[0] = len >> 16;
 267 }
 268
 269 static size_t
 270 icl_pdu_padding(const struct icl_pdu *ip)
 271 {
 272
 273         if ((ip->ip_data_len % 4) != 0)
 274                 return (4 - (ip->ip_data_len % 4));
 275
 276         return (0);
 277 }
 278
 279 static size_t
 280 icl_pdu_size(const struct icl_pdu *response)
 281 {
 282         size_t len;
 283
 284         KASSERT(response->ip_ahs_len == 0, ("responding with AHS"));
 285
 286         len = sizeof(struct iscsi_bhs) + response->ip_data_len +
 287             icl_pdu_padding(response);
 288         if (response->ip_conn->ic_header_crc32c)
 289                 len += ISCSI_HEADER_DIGEST_SIZE;
 290         if (response->ip_data_len != 0 && response->ip_conn->ic_data_crc32c)
 291                 len += ISCSI_DATA_DIGEST_SIZE;
 292
 293         return (len);
 294 }
 295
 296 static int
 297 icl_pdu_receive_bhs(struct icl_pdu *request, size_t *availablep)
 298 {
 299         struct mbuf *m;
 300
 301         m = icl_conn_receive(request->ip_conn, sizeof(struct iscsi_bhs));
 302         if (m == NULL) {
 303                 ICL_DEBUG("failed to receive BHS");
 304                 return (-1);
 305         }
 306
 307         request->ip_bhs_mbuf = m_pullup(m, sizeof(struct iscsi_bhs));
 308         if (request->ip_bhs_mbuf == NULL) {
 309                 ICL_WARN("m_pullup failed");
 310                 return (-1);
 311         }
 312         request->ip_bhs = mtod(request->ip_bhs_mbuf, struct iscsi_bhs *);
 313
 314         /*
 315          * XXX: For architectures with strict alignment requirements
 316          *      we may need to allocate ip_bhs and copy the data into it.
 317          *      For some reason, though, not doing this doesn't seem
 318          *      to cause problems; tested on sparc64.
 319          */
 320
 321         *availablep -= sizeof(struct iscsi_bhs);
 322         return (0);
 323 }
 324
 325 static int
 326 icl_pdu_receive_ahs(struct icl_pdu *request, size_t *availablep)
 327 {
 328
 329         request->ip_ahs_len = icl_pdu_ahs_length(request);
 330         if (request->ip_ahs_len == 0)
 331                 return (0);
 332
 333         request->ip_ahs_mbuf = icl_conn_receive(request->ip_conn,
 334             request->ip_ahs_len);
 335         if (request->ip_ahs_mbuf == NULL) {
 336                 ICL_DEBUG("failed to receive AHS");
 337                 return (-1);
 338         }
 339
 340         *availablep -= request->ip_ahs_len;
 341         return (0);
 342 }
 343
 344 static uint32_t
 345 icl_mbuf_to_crc32c(const struct mbuf *m0)
 346 {
 347         uint32_t digest = 0xffffffff;
 348         const struct mbuf *m;
 349
 350         for (m = m0; m != NULL; m = m->m_next)
 351                 digest = calculate_crc32c(digest,
 352                     mtod(m, const void *), m->m_len);
 353
 354         digest = digest ^ 0xffffffff;
 355
 356         return (digest);
 357 }
 358
 359 static int
 360 icl_pdu_check_header_digest(struct icl_pdu *request, size_t *availablep)
 361 {
 362         struct mbuf *m;
 363         uint32_t received_digest, valid_digest;
 364
 365         if (request->ip_conn->ic_header_crc32c == false)
 366                 return (0);
 367
 368         m = icl_conn_receive(request->ip_conn, ISCSI_HEADER_DIGEST_SIZE);
 369         if (m == NULL) {
 370                 ICL_DEBUG("failed to receive header digest");
 371                 return (-1);
 372         }
 373
 374         CTASSERT(sizeof(received_digest) == ISCSI_HEADER_DIGEST_SIZE);
 375         m_copydata(m, 0, ISCSI_HEADER_DIGEST_SIZE, (void *)&received_digest);
 376         m_freem(m);
 377
 378         *availablep -= ISCSI_HEADER_DIGEST_SIZE;
 379
 380         /*
 381          * XXX: Handle AHS.
 382          */
 383         valid_digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
 384         if (received_digest != valid_digest) {
 385                 ICL_WARN("header digest check failed; got 0x%x, "
 386                     "should be 0x%x", received_digest, valid_digest);
 387                 return (-1);
 388         }
 389
 390         return (0);
 391 }
 392
 393 /*
 394  * Return the number of bytes that should be waiting in the receive socket
 395  * before icl_pdu_receive_data_segment() gets called.
 396  */
 397 static size_t
 398 icl_pdu_data_segment_receive_len(const struct icl_pdu *request)
 399 {
 400         size_t len;
 401
 402         len = icl_pdu_data_segment_length(request);
 403         if (len == 0)
 404                 return (0);
 405
 406         /*
 407          * Account for the parts of data segment already read from
 408          * the socket buffer.
 409          */
 410         KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
 411         len -= request->ip_data_len;
 412
 413         /*
 414          * Don't always wait for the full data segment to be delivered
 415          * to the socket; this might badly affect performance due to
 416          * TCP window scaling.
 417          */
 418         if (len > partial_receive_len) {
 419 #if 0
 420                 ICL_DEBUG("need %zd bytes of data, limiting to %zd",
 421                     len, partial_receive_len));
 422 #endif
 423                 len = partial_receive_len;
 424
 425                 return (len);
 426         }
 427
 428         /*
 429          * Account for padding.  Note that due to the way code is written,
 430          * the icl_pdu_receive_data_segment() must always receive padding
 431          * along with the last part of data segment, because it would be
 432          * impossible to tell whether we've already received the full data
 433          * segment including padding, or without it.
 434          */
 435         if ((len % 4) != 0)
 436                 len += 4 - (len % 4);
 437
 438 #if 0
 439         ICL_DEBUG("need %zd bytes of data", len));
 440 #endif
 441
 442         return (len);
 443 }
 444
 445 static int
 446 icl_pdu_receive_data_segment(struct icl_pdu *request,
 447     size_t *availablep, bool *more_neededp)
 448 {
 449         struct icl_conn *ic;
 450         size_t len, padding = 0;
 451         struct mbuf *m;
 452
 453         ic = request->ip_conn;
 454
 455         *more_neededp = false;
 456         ic->ic_receive_len = 0;
 457
 458         len = icl_pdu_data_segment_length(request);
 459         if (len == 0)
 460                 return (0);
 461
 462         if ((len % 4) != 0)
 463                 padding = 4 - (len % 4);
 464
 465         /*
 466          * Account for already received parts of data segment.
 467          */
 468         KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
 469         len -= request->ip_data_len;
 470
 471         if (len + padding > *availablep) {
 472                 /*
 473                  * Not enough data in the socket buffer.  Receive as much
 474                  * as we can.  Don't receive padding, since, obviously, it's
 475                  * not the end of data segment yet.
 476                  */
 477 #if 0
 478                 ICL_DEBUG("limited from %zd to %zd",
 479                     len + padding, *availablep - padding));
 480 #endif
 481                 len = *availablep - padding;
 482                 *more_neededp = true;
 483                 padding = 0;
 484         }
 485
 486         /*
 487          * Must not try to receive padding without at least one byte
 488          * of actual data segment.
 489          */
 490         if (len > 0) {
 491                 m = icl_conn_receive(request->ip_conn, len + padding);
 492                 if (m == NULL) {
 493                         ICL_DEBUG("failed to receive data segment");
 494                         return (-1);
 495                 }
 496
 497                 if (request->ip_data_mbuf == NULL)
 498                         request->ip_data_mbuf = m;
 499                 else
 500                         m_cat(request->ip_data_mbuf, m);
 501
 502                 request->ip_data_len += len;
 503                 *availablep -= len + padding;
 504         } else
 505                 ICL_DEBUG("len 0");
 506
 507         if (*more_neededp)
 508                 ic->ic_receive_len =
 509                     icl_pdu_data_segment_receive_len(request);
 510
 511         return (0);
 512 }
 513
 514 static int
 515 icl_pdu_check_data_digest(struct icl_pdu *request, size_t *availablep)
 516 {
 517         struct mbuf *m;
 518         uint32_t received_digest, valid_digest;
 519
 520         if (request->ip_conn->ic_data_crc32c == false)
 521                 return (0);
 522
 523         if (request->ip_data_len == 0)
 524                 return (0);
 525
 526         m = icl_conn_receive(request->ip_conn, ISCSI_DATA_DIGEST_SIZE);
 527         if (m == NULL) {
 528                 ICL_DEBUG("failed to receive data digest");
 529                 return (-1);
 530         }
 531
 532         CTASSERT(sizeof(received_digest) == ISCSI_DATA_DIGEST_SIZE);
 533         m_copydata(m, 0, ISCSI_DATA_DIGEST_SIZE, (void *)&received_digest);
 534         m_freem(m);
 535
 536         *availablep -= ISCSI_DATA_DIGEST_SIZE;
 537
 538         /*
 539          * Note that ip_data_mbuf also contains padding; since digest
 540          * calculation is supposed to include that, we iterate over
 541          * the entire ip_data_mbuf chain, not just ip_data_len bytes of it.
 542          */
 543         valid_digest = icl_mbuf_to_crc32c(request->ip_data_mbuf);
 544         if (received_digest != valid_digest) {
 545                 ICL_WARN("data digest check failed; got 0x%x, "
 546                     "should be 0x%x", received_digest, valid_digest);
 547                 return (-1);
 548         }
 549
 550         return (0);
 551 }
 552
 553 /*
 554  * Somewhat contrary to the name, this attempts to receive only one
 555  * "part" of PDU at a time; call it repeatedly until it returns non-NULL.
 556  */
 557 static struct icl_pdu *
 558 icl_conn_receive_pdu(struct icl_conn *ic, size_t *availablep)
 559 {
 560         struct icl_pdu *request;
 561         struct socket *so;
 562         size_t len;
 563         int error;
 564         bool more_needed;
 565
 566         so = ic->ic_socket;
 567
 568         if (ic->ic_receive_state == ICL_CONN_STATE_BHS) {
 569                 KASSERT(ic->ic_receive_pdu == NULL,
 570                     ("ic->ic_receive_pdu != NULL"));
 571                 request = icl_pdu_new_empty(ic, M_NOWAIT);
 572                 if (request == NULL) {
 573                         ICL_DEBUG("failed to allocate PDU; "
 574                             "dropping connection");
 575                         icl_conn_fail(ic);
 576                         return (NULL);
 577                 }
 578                 ic->ic_receive_pdu = request;
 579         } else {
 580                 KASSERT(ic->ic_receive_pdu != NULL,
 581                     ("ic->ic_receive_pdu == NULL"));
 582                 request = ic->ic_receive_pdu;
 583         }
 584
 585         if (*availablep < ic->ic_receive_len) {
 586 #if 0
 587                 ICL_DEBUG("not enough data; need %zd, "
 588                     "have %zd", ic->ic_receive_len, *availablep);
 589 #endif
 590                 return (NULL);
 591         }
 592
 593         switch (ic->ic_receive_state) {
 594         case ICL_CONN_STATE_BHS:
 595                 //ICL_DEBUG("receiving BHS");
 596                 error = icl_pdu_receive_bhs(request, availablep);
 597                 if (error != 0) {
 598                         ICL_DEBUG("failed to receive BHS; "
 599                             "dropping connection");
 600                         break;
 601                 }
 602
 603                 /*
 604                  * We don't enforce any limit for AHS length;
 605                  * its length is stored in 8 bit field.
 606                  */
 607
 608                 len = icl_pdu_data_segment_length(request);
 609                 if (len > ic->ic_max_data_segment_length) {
 610                         ICL_WARN("received data segment "
 611                             "length %zd is larger than negotiated "
 612                             "MaxDataSegmentLength %zd; "
 613                             "dropping connection",
 614                             len, ic->ic_max_data_segment_length);
 615                         error = EINVAL;
 616                         break;
 617                 }
 618
 619                 ic->ic_receive_state = ICL_CONN_STATE_AHS;
 620                 ic->ic_receive_len = icl_pdu_ahs_length(request);
 621                 break;
 622
 623         case ICL_CONN_STATE_AHS:
 624                 //ICL_DEBUG("receiving AHS");
 625                 error = icl_pdu_receive_ahs(request, availablep);
 626                 if (error != 0) {
 627                         ICL_DEBUG("failed to receive AHS; "
 628                             "dropping connection");
 629                         break;
 630                 }
 631                 ic->ic_receive_state = ICL_CONN_STATE_HEADER_DIGEST;
 632                 if (ic->ic_header_crc32c == false)
 633                         ic->ic_receive_len = 0;
 634                 else
 635                         ic->ic_receive_len = ISCSI_HEADER_DIGEST_SIZE;
 636                 break;
 637
 638         case ICL_CONN_STATE_HEADER_DIGEST:
 639                 //ICL_DEBUG("receiving header digest");
 640                 error = icl_pdu_check_header_digest(request, availablep);
 641                 if (error != 0) {
 642                         ICL_DEBUG("header digest failed; "
 643                             "dropping connection");
 644                         break;
 645                 }
 646
 647                 ic->ic_receive_state = ICL_CONN_STATE_DATA;
 648                 ic->ic_receive_len =
 649                     icl_pdu_data_segment_receive_len(request);
 650                 break;
 651
 652         case ICL_CONN_STATE_DATA:
 653                 //ICL_DEBUG("receiving data segment");
 654                 error = icl_pdu_receive_data_segment(request, availablep,
 655                     &more_needed);
 656                 if (error != 0) {
 657                         ICL_DEBUG("failed to receive data segment;"
 658                             "dropping connection");
 659                         break;
 660                 }
 661
 662                 if (more_needed)
 663                         break;
 664
 665                 ic->ic_receive_state = ICL_CONN_STATE_DATA_DIGEST;
 666                 if (request->ip_data_len == 0 || ic->ic_data_crc32c == false)
 667                         ic->ic_receive_len = 0;
 668                 else
 669                         ic->ic_receive_len = ISCSI_DATA_DIGEST_SIZE;
 670                 break;
 671
 672         case ICL_CONN_STATE_DATA_DIGEST:
 673                 //ICL_DEBUG("receiving data digest");
 674                 error = icl_pdu_check_data_digest(request, availablep);
 675                 if (error != 0) {
 676                         ICL_DEBUG("data digest failed; "
 677                             "dropping connection");
 678                         break;
 679                 }
 680
 681                 /*
 682                  * We've received complete PDU; reset the receive state machine
 683                  * and return the PDU.
 684                  */
 685                 ic->ic_receive_state = ICL_CONN_STATE_BHS;
 686                 ic->ic_receive_len = sizeof(struct iscsi_bhs);
 687                 ic->ic_receive_pdu = NULL;
 688                 return (request);
 689
 690         default:
 691                 panic("invalid ic_receive_state %d\n", ic->ic_receive_state);
 692         }
 693
 694         if (error != 0) {
 695                 /*
 696                  * Don't free the PDU; it's pointed to by ic->ic_receive_pdu
 697                  * and will get freed in icl_soft_conn_close().
 698                  */
 699                 icl_conn_fail(ic);
 700         }
 701
 702         return (NULL);
 703 }
 704
 705 static void
 706 icl_conn_receive_pdus(struct icl_conn *ic, size_t available)
 707 {
 708         struct icl_pdu *response;
 709         struct socket *so;
 710
 711         so = ic->ic_socket;
 712
 713         /*
 714          * This can never happen; we're careful to only mess with ic->ic_socket
 715          * pointer when the send/receive threads are not running.
 716          */
 717         KASSERT(so != NULL, ("NULL socket"));
 718
 719         for (;;) {
 720                 if (ic->ic_disconnecting)
 721                         return;
 722
 723                 if (so->so_error != 0) {
 724                         ICL_DEBUG("connection error %d; "
 725                             "dropping connection", so->so_error);
 726                         icl_conn_fail(ic);
 727                         return;
 728                 }
 729
 730                 /*
 731                  * Loop until we have a complete PDU or there is not enough
 732                  * data in the socket buffer.
 733                  */
 734                 if (available < ic->ic_receive_len) {
 735 #if 0
 736                         ICL_DEBUG("not enough data; have %zd, "
 737                             "need %zd", available,
 738                             ic->ic_receive_len);
 739 #endif
 740                         return;
 741                 }
 742
 743                 response = icl_conn_receive_pdu(ic, &available);
 744                 if (response == NULL)
 745                         continue;
 746
 747                 if (response->ip_ahs_len > 0) {
 748                         ICL_WARN("received PDU with unsupported "
 749                             "AHS; opcode 0x%x; dropping connection",
 750                             response->ip_bhs->bhs_opcode);
 751                         icl_pdu_free(response);
 752                         icl_conn_fail(ic);
 753                         return;
 754                 }
 755
 756                 (ic->ic_receive)(response);
 757         }
 758 }
 759
 760 static void
 761 icl_receive_thread(void *arg)
 762 {
 763         struct icl_conn *ic;
 764         size_t available;
 765         struct socket *so;
 766
 767         ic = arg;
 768         so = ic->ic_socket;
 769
 770         ICL_CONN_LOCK(ic);
 771         ic->ic_receive_running = true;
 772         ICL_CONN_UNLOCK(ic);
 773
 774         for (;;) {
 775                 if (ic->ic_disconnecting) {
 776                         //ICL_DEBUG("terminating");
 777                         break;
 778                 }
 779
 780                 /*
 781                  * Set the low watermark, to be checked by
 782                  * soreadable() in icl_soupcall_receive()
 783                  * to avoid unneccessary wakeups until there
 784                  * is enough data received to read the PDU.
 785                  */
 786                 SOCKBUF_LOCK(&so->so_rcv);
 787                 available = sbavail(&so->so_rcv);
 788                 if (available < ic->ic_receive_len) {
 789                         so->so_rcv.sb_lowat = ic->ic_receive_len;
 790                         cv_wait(&ic->ic_receive_cv, &so->so_rcv.sb_mtx);
 791                 } else
 792                         so->so_rcv.sb_lowat = so->so_rcv.sb_hiwat + 1;
 793                 SOCKBUF_UNLOCK(&so->so_rcv);
 794
 795                 icl_conn_receive_pdus(ic, available);
 796         }
 797
 798         ICL_CONN_LOCK(ic);
 799         ic->ic_receive_running = false;
 800         cv_signal(&ic->ic_send_cv);
 801         ICL_CONN_UNLOCK(ic);
 802         kthread_exit();
 803 }
 804
 805 static int
 806 icl_soupcall_receive(struct socket *so, void *arg, int waitflag)
 807 {
 808         struct icl_conn *ic;
 809
 810         if (!soreadable(so))
 811                 return (SU_OK);
 812
 813         ic = arg;
 814         cv_signal(&ic->ic_receive_cv);
 815         return (SU_OK);
 816 }
 817
 818 static int
 819 icl_pdu_finalize(struct icl_pdu *request)
 820 {
 821         size_t padding, pdu_len;
 822         uint32_t digest, zero = 0;
 823         int ok;
 824         struct icl_conn *ic;
 825
 826         ic = request->ip_conn;
 827
 828         icl_pdu_set_data_segment_length(request, request->ip_data_len);
 829
 830         pdu_len = icl_pdu_size(request);
 831
 832         if (ic->ic_header_crc32c) {
 833                 digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
 834                 ok = m_append(request->ip_bhs_mbuf, sizeof(digest),
 835                     (void *)&digest);
 836                 if (ok != 1) {
 837                         ICL_WARN("failed to append header digest");
 838                         return (1);
 839                 }
 840         }
 841
 842         if (request->ip_data_len != 0) {
 843                 padding = icl_pdu_padding(request);
 844                 if (padding > 0) {
 845                         ok = m_append(request->ip_data_mbuf, padding,
 846                             (void *)&zero);
 847                         if (ok != 1) {
 848                                 ICL_WARN("failed to append padding");
 849                                 return (1);
 850                         }
 851                 }
 852
 853                 if (ic->ic_data_crc32c) {
 854                         digest = icl_mbuf_to_crc32c(request->ip_data_mbuf);
 855
 856                         ok = m_append(request->ip_data_mbuf, sizeof(digest),
 857                             (void *)&digest);
 858                         if (ok != 1) {
 859                                 ICL_WARN("failed to append data digest");
 860                                 return (1);
 861                         }
 862                 }
 863
 864                 m_cat(request->ip_bhs_mbuf, request->ip_data_mbuf);
 865                 request->ip_data_mbuf = NULL;
 866         }
 867
 868         request->ip_bhs_mbuf->m_pkthdr.len = pdu_len;
 869
 870         return (0);
 871 }
 872
 873 static void
 874 icl_conn_send_pdus(struct icl_conn *ic, struct icl_pdu_stailq *queue)
 875 {
 876         struct icl_pdu *request, *request2;
 877         struct socket *so;
 878         size_t available, size, size2;
 879         int coalesced, error;
 880
 881         ICL_CONN_LOCK_ASSERT_NOT(ic);
 882
 883         so = ic->ic_socket;
 884
 885         SOCKBUF_LOCK(&so->so_snd);
 886         /*
 887          * Check how much space do we have for transmit.  We can't just
 888          * call sosend() and retry when we get EWOULDBLOCK or EMSGSIZE,
 889          * as it always frees the mbuf chain passed to it, even in case
 890          * of error.
 891          */
 892         available = sbspace(&so->so_snd);
 893
 894         /*
 895          * Notify the socket upcall that we don't need wakeups
 896          * for the time being.
 897          */
 898         so->so_snd.sb_lowat = so->so_snd.sb_hiwat + 1;
 899         SOCKBUF_UNLOCK(&so->so_snd);
 900
 901         while (!STAILQ_EMPTY(queue)) {
 902                 request = STAILQ_FIRST(queue);
 903                 size = icl_pdu_size(request);
 904                 if (available < size) {
 905
 906                         /*
 907                          * Set the low watermark, to be checked by
 908                          * sowriteable() in icl_soupcall_send()
 909                          * to avoid unneccessary wakeups until there
 910                          * is enough space for the PDU to fit.
 911                          */
 912                         SOCKBUF_LOCK(&so->so_snd);
 913                         available = sbspace(&so->so_snd);
 914                         if (available < size) {
 915 #if 1
 916                                 ICL_DEBUG("no space to send; "
 917                                     "have %zd, need %zd",
 918                                     available, size);
 919 #endif
 920                                 so->so_snd.sb_lowat = size;
 921                                 SOCKBUF_UNLOCK(&so->so_snd);
 922                                 return;
 923                         }
 924                         SOCKBUF_UNLOCK(&so->so_snd);
 925                 }
 926                 STAILQ_REMOVE_HEAD(queue, ip_next);
 927                 error = icl_pdu_finalize(request);
 928                 if (error != 0) {
 929                         ICL_DEBUG("failed to finalize PDU; "
 930                             "dropping connection");
 931                         icl_conn_fail(ic);
 932                         icl_pdu_free(request);
 933                         return;
 934                 }
 935                 if (coalesce) {
 936                         coalesced = 1;
 937                         for (;;) {
 938                                 request2 = STAILQ_FIRST(queue);
 939                                 if (request2 == NULL)
 940                                         break;
 941                                 size2 = icl_pdu_size(request2);
 942                                 if (available < size + size2)
 943                                         break;
 944                                 STAILQ_REMOVE_HEAD(queue, ip_next);
 945                                 error = icl_pdu_finalize(request2);
 946                                 if (error != 0) {
 947                                         ICL_DEBUG("failed to finalize PDU; "
 948                                             "dropping connection");
 949                                         icl_conn_fail(ic);
 950                                         icl_pdu_free(request);
 951                                         icl_pdu_free(request2);
 952                                         return;
 953                                 }
 954                                 m_cat(request->ip_bhs_mbuf, request2->ip_bhs_mbuf);
 955                                 request2->ip_bhs_mbuf = NULL;
 956                                 request->ip_bhs_mbuf->m_pkthdr.len += size2;
 957                                 size += size2;
 958                                 STAILQ_REMOVE_AFTER(queue, request, ip_next);
 959                                 icl_pdu_free(request2);
 960                                 coalesced++;
 961                         }
 962 #if 0
 963                         if (coalesced > 1) {
 964                                 ICL_DEBUG("coalesced %d PDUs into %zd bytes",
 965                                     coalesced, size);
 966                         }
 967 #endif
 968                 }
 969                 available -= size;
 970                 error = sosend(so, NULL, NULL, request->ip_bhs_mbuf,
 971                     NULL, MSG_DONTWAIT, curthread);
 972                 request->ip_bhs_mbuf = NULL; /* Sosend consumes the mbuf. */
 973                 if (error != 0) {
 974                         ICL_DEBUG("failed to send PDU, error %d; "
 975                             "dropping connection", error);
 976                         icl_conn_fail(ic);
 977                         icl_pdu_free(request);
 978                         return;
 979                 }
 980                 icl_pdu_free(request);
 981         }
 982 }
 983
 984 static void
 985 icl_send_thread(void *arg)
 986 {
 987         struct icl_conn *ic;
 988         struct icl_pdu_stailq queue;
 989
 990         ic = arg;
 991
 992         STAILQ_INIT(&queue);
 993
 994         ICL_CONN_LOCK(ic);
 995         ic->ic_send_running = true;
 996
 997         for (;;) {
 998                 for (;;) {
 999                         /*
1000                          * If the local queue is empty, populate it from
1001                          * the main one.  This way the icl_conn_send_pdus()
1002                          * can go through all the queued PDUs without holding
1003                          * any locks.
1004                          */
1005                         if (STAILQ_EMPTY(&queue))
1006                                 STAILQ_SWAP(&ic->ic_to_send, &queue, icl_pdu);
1007
1008                         ic->ic_check_send_space = false;
1009                         ICL_CONN_UNLOCK(ic);
1010                         icl_conn_send_pdus(ic, &queue);
1011                         ICL_CONN_LOCK(ic);
1012
1013                         /*
1014                          * The icl_soupcall_send() was called since the last
1015                          * call to sbspace(); go around;
1016                          */
1017                         if (ic->ic_check_send_space)
1018                                 continue;
1019
1020                         /*
1021                          * Local queue is empty, but we still have PDUs
1022                          * in the main one; go around.
1023                          */
1024                         if (STAILQ_EMPTY(&queue) &&
1025                             !STAILQ_EMPTY(&ic->ic_to_send))
1026                                 continue;
1027
1028                         /*
1029                          * There might be some stuff in the local queue,
1030                          * which didn't get sent due to not having enough send
1031                          * space.  Wait for socket upcall.
1032                          */
1033                         break;
1034                 }
1035
1036                 if (ic->ic_disconnecting) {
1037                         //ICL_DEBUG("terminating");
1038                         break;
1039                 }
1040
1041                 cv_wait(&ic->ic_send_cv, ic->ic_lock);
1042         }
1043
1044         /*
1045          * We're exiting; move PDUs back to the main queue, so they can
1046          * get freed properly.  At this point ordering doesn't matter.
1047          */
1048         STAILQ_CONCAT(&ic->ic_to_send, &queue);
1049
1050         ic->ic_send_running = false;
1051         cv_signal(&ic->ic_send_cv);
1052         ICL_CONN_UNLOCK(ic);
1053         kthread_exit();
1054 }
1055
1056 static int
1057 icl_soupcall_send(struct socket *so, void *arg, int waitflag)
1058 {
1059         struct icl_conn *ic;
1060
1061         if (!sowriteable(so))
1062                 return (SU_OK);
1063
1064         ic = arg;
1065
1066         ICL_CONN_LOCK(ic);
1067         ic->ic_check_send_space = true;
1068         ICL_CONN_UNLOCK(ic);
1069
1070         cv_signal(&ic->ic_send_cv);
1071
1072         return (SU_OK);
1073 }
1074
1075 static int
1076 icl_pdu_append_data(struct icl_pdu *request, const void *addr, size_t len,
1077     int flags)
1078 {
1079         struct mbuf *mb, *newmb;
1080         size_t copylen, off = 0;
1081
1082         KASSERT(len > 0, ("len == 0"));
1083
1084         newmb = m_getm2(NULL, len, flags, MT_DATA, M_PKTHDR);
1085         if (newmb == NULL) {
1086                 ICL_WARN("failed to allocate mbuf for %zd bytes", len);
1087                 return (ENOMEM);
1088         }
1089
1090         for (mb = newmb; mb != NULL; mb = mb->m_next) {
1091                 copylen = min(M_TRAILINGSPACE(mb), len - off);
1092                 memcpy(mtod(mb, char *), (const char *)addr + off, copylen);
1093                 mb->m_len = copylen;
1094                 off += copylen;
1095         }
1096         KASSERT(off == len, ("%s: off != len", __func__));
1097
1098         if (request->ip_data_mbuf == NULL) {
1099                 request->ip_data_mbuf = newmb;
1100                 request->ip_data_len = len;
1101         } else {
1102                 m_cat(request->ip_data_mbuf, newmb);
1103                 request->ip_data_len += len;
1104         }
1105
1106         return (0);
1107 }
1108
1109 int
1110 icl_soft_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *request,
1111     const void *addr, size_t len, int flags)
1112 {
1113
1114         return (icl_pdu_append_data(request, addr, len, flags));
1115 }
1116
1117 static void
1118 icl_pdu_get_data(struct icl_pdu *ip, size_t off, void *addr, size_t len)
1119 {
1120
1121         m_copydata(ip->ip_data_mbuf, off, len, addr);
1122 }
1123
1124 void
1125 icl_soft_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip,
1126     size_t off, void *addr, size_t len)
1127 {
1128
1129         return (icl_pdu_get_data(ip, off, addr, len));
1130 }
1131
1132 static void
1133 icl_pdu_queue(struct icl_pdu *ip)
1134 {
1135         struct icl_conn *ic;
1136
1137         ic = ip->ip_conn;
1138
1139         ICL_CONN_LOCK_ASSERT(ic);
1140
1141         if (ic->ic_disconnecting || ic->ic_socket == NULL) {
1142                 ICL_DEBUG("icl_pdu_queue on closed connection");
1143                 icl_pdu_free(ip);
1144                 return;
1145         }
1146
1147         if (!STAILQ_EMPTY(&ic->ic_to_send)) {
1148                 STAILQ_INSERT_TAIL(&ic->ic_to_send, ip, ip_next);
1149                 /*
1150                  * If the queue is not empty, someone else had already
1151                  * signaled the send thread; no need to do that again,
1152                  * just return.
1153                  */
1154                 return;
1155         }
1156
1157         STAILQ_INSERT_TAIL(&ic->ic_to_send, ip, ip_next);
1158         cv_signal(&ic->ic_send_cv);
1159 }
1160
1161 void
1162 icl_soft_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip)
1163 {
1164
1165         icl_pdu_queue(ip);
1166 }
1167
1168 static struct icl_conn *
1169 icl_soft_new_conn(const char *name, struct mtx *lock)
1170 {
1171         struct icl_conn *ic;
1172
1173         refcount_acquire(&icl_ncons);
1174
1175         ic = (struct icl_conn *)kobj_create(&icl_soft_class, M_ICL_SOFT, M_WAITOK | M_ZERO);
1176
1177         STAILQ_INIT(&ic->ic_to_send);
1178         ic->ic_lock = lock;
1179         cv_init(&ic->ic_send_cv, "icl_tx");
1180         cv_init(&ic->ic_receive_cv, "icl_rx");
1181 #ifdef DIAGNOSTIC
1182         refcount_init(&ic->ic_outstanding_pdus, 0);
1183 #endif
1184         ic->ic_max_data_segment_length = ICL_MAX_DATA_SEGMENT_LENGTH;
1185         ic->ic_name = name;
1186         ic->ic_offload = "None";
1187
1188         return (ic);
1189 }
1190
1191 void
1192 icl_soft_conn_free(struct icl_conn *ic)
1193 {
1194
1195         cv_destroy(&ic->ic_send_cv);
1196         cv_destroy(&ic->ic_receive_cv);
1197         kobj_delete((struct kobj *)ic, M_ICL_SOFT);
1198         refcount_release(&icl_ncons);
1199 }
1200
1201 static int
1202 icl_conn_start(struct icl_conn *ic)
1203 {
1204         size_t minspace;
1205         struct sockopt opt;
1206         int error, one = 1;
1207
1208         ICL_CONN_LOCK(ic);
1209
1210         /*
1211          * XXX: Ugly hack.
1212          */
1213         if (ic->ic_socket == NULL) {
1214                 ICL_CONN_UNLOCK(ic);
1215                 return (EINVAL);
1216         }
1217
1218         ic->ic_receive_state = ICL_CONN_STATE_BHS;
1219         ic->ic_receive_len = sizeof(struct iscsi_bhs);
1220         ic->ic_disconnecting = false;
1221
1222         ICL_CONN_UNLOCK(ic);
1223
1224         /*
1225          * For sendspace, this is required because the current code cannot
1226          * send a PDU in pieces; thus, the minimum buffer size is equal
1227          * to the maximum PDU size.  "+4" is to account for possible padding.
1228          *
1229          * What we should actually do here is to use autoscaling, but set
1230          * some minimal buffer size to "minspace".  I don't know a way to do
1231          * that, though.
1232          */
1233         minspace = sizeof(struct iscsi_bhs) + ic->ic_max_data_segment_length +
1234             ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE + 4;
1235         if (sendspace < minspace) {
1236                 ICL_WARN("kern.icl.sendspace too low; must be at least %zd",
1237                     minspace);
1238                 sendspace = minspace;
1239         }
1240         if (recvspace < minspace) {
1241                 ICL_WARN("kern.icl.recvspace too low; must be at least %zd",
1242                     minspace);
1243                 recvspace = minspace;
1244         }
1245
1246         error = soreserve(ic->ic_socket, sendspace, recvspace);
1247         if (error != 0) {
1248                 ICL_WARN("soreserve failed with error %d", error);
1249                 icl_soft_conn_close(ic);
1250                 return (error);
1251         }
1252         ic->ic_socket->so_snd.sb_flags |= SB_AUTOSIZE;
1253         ic->ic_socket->so_rcv.sb_flags |= SB_AUTOSIZE;
1254
1255         /*
1256          * Disable Nagle.
1257          */
1258         bzero(&opt, sizeof(opt));
1259         opt.sopt_dir = SOPT_SET;
1260         opt.sopt_level = IPPROTO_TCP;
1261         opt.sopt_name = TCP_NODELAY;
1262         opt.sopt_val = &one;
1263         opt.sopt_valsize = sizeof(one);
1264         error = sosetopt(ic->ic_socket, &opt);
1265         if (error != 0) {
1266                 ICL_WARN("disabling TCP_NODELAY failed with error %d", error);
1267                 icl_soft_conn_close(ic);
1268                 return (error);
1269         }
1270
1271         /*
1272          * Start threads.
1273          */
1274         error = kthread_add(icl_send_thread, ic, NULL, NULL, 0, 0, "%stx",
1275             ic->ic_name);
1276         if (error != 0) {
1277                 ICL_WARN("kthread_add(9) failed with error %d", error);
1278                 icl_soft_conn_close(ic);
1279                 return (error);
1280         }
1281
1282         error = kthread_add(icl_receive_thread, ic, NULL, NULL, 0, 0, "%srx",
1283             ic->ic_name);
1284         if (error != 0) {
1285                 ICL_WARN("kthread_add(9) failed with error %d", error);
1286                 icl_soft_conn_close(ic);
1287                 return (error);
1288         }
1289
1290         /*
1291          * Register socket upcall, to get notified about incoming PDUs
1292          * and free space to send outgoing ones.
1293          */
1294         SOCKBUF_LOCK(&ic->ic_socket->so_snd);
1295         soupcall_set(ic->ic_socket, SO_SND, icl_soupcall_send, ic);
1296         SOCKBUF_UNLOCK(&ic->ic_socket->so_snd);
1297         SOCKBUF_LOCK(&ic->ic_socket->so_rcv);
1298         soupcall_set(ic->ic_socket, SO_RCV, icl_soupcall_receive, ic);
1299         SOCKBUF_UNLOCK(&ic->ic_socket->so_rcv);
1300
1301         return (0);
1302 }
1303
1304 int
1305 icl_soft_conn_handoff(struct icl_conn *ic, int fd)
1306 {
1307         struct file *fp;
1308         struct socket *so;
1309         cap_rights_t rights;
1310         int error;
1311
1312         ICL_CONN_LOCK_ASSERT_NOT(ic);
1313
1314         /*
1315          * Steal the socket from userland.
1316          */
1317         error = fget(curthread, fd,
1318             cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1319         if (error != 0)
1320                 return (error);
1321         if (fp->f_type != DTYPE_SOCKET) {
1322                 fdrop(fp, curthread);
1323                 return (EINVAL);
1324         }
1325         so = fp->f_data;
1326         if (so->so_type != SOCK_STREAM) {
1327                 fdrop(fp, curthread);
1328                 return (EINVAL);
1329         }
1330
1331         ICL_CONN_LOCK(ic);
1332
1333         if (ic->ic_socket != NULL) {
1334                 ICL_CONN_UNLOCK(ic);
1335                 fdrop(fp, curthread);
1336                 return (EBUSY);
1337         }
1338
1339         ic->ic_socket = fp->f_data;
1340         fp->f_ops = &badfileops;
1341         fp->f_data = NULL;
1342         fdrop(fp, curthread);
1343         ICL_CONN_UNLOCK(ic);
1344
1345         error = icl_conn_start(ic);
1346
1347         return (error);
1348 }
1349
1350 void
1351 icl_soft_conn_close(struct icl_conn *ic)
1352 {
1353         struct icl_pdu *pdu;
1354
1355         ICL_CONN_LOCK_ASSERT_NOT(ic);
1356
1357         ICL_CONN_LOCK(ic);
1358         if (ic->ic_socket == NULL) {
1359                 ICL_CONN_UNLOCK(ic);
1360                 return;
1361         }
1362
1363         /*
1364          * Deregister socket upcalls.
1365          */
1366         ICL_CONN_UNLOCK(ic);
1367         SOCKBUF_LOCK(&ic->ic_socket->so_snd);
1368         if (ic->ic_socket->so_snd.sb_upcall != NULL)
1369                 soupcall_clear(ic->ic_socket, SO_SND);
1370         SOCKBUF_UNLOCK(&ic->ic_socket->so_snd);
1371         SOCKBUF_LOCK(&ic->ic_socket->so_rcv);
1372         if (ic->ic_socket->so_rcv.sb_upcall != NULL)
1373                 soupcall_clear(ic->ic_socket, SO_RCV);
1374         SOCKBUF_UNLOCK(&ic->ic_socket->so_rcv);
1375         ICL_CONN_LOCK(ic);
1376
1377         ic->ic_disconnecting = true;
1378
1379         /*
1380          * Wake up the threads, so they can properly terminate.
1381          */
1382         while (ic->ic_receive_running || ic->ic_send_running) {
1383                 //ICL_DEBUG("waiting for send/receive threads to terminate");
1384                 cv_signal(&ic->ic_receive_cv);
1385                 cv_signal(&ic->ic_send_cv);
1386                 cv_wait(&ic->ic_send_cv, ic->ic_lock);
1387         }
1388         //ICL_DEBUG("send/receive threads terminated");
1389
1390         ICL_CONN_UNLOCK(ic);
1391         soclose(ic->ic_socket);
1392         ICL_CONN_LOCK(ic);
1393         ic->ic_socket = NULL;
1394
1395         if (ic->ic_receive_pdu != NULL) {
1396                 //ICL_DEBUG("freeing partially received PDU");
1397                 icl_pdu_free(ic->ic_receive_pdu);
1398                 ic->ic_receive_pdu = NULL;
1399         }
1400
1401         /*
1402          * Remove any outstanding PDUs from the send queue.
1403          */
1404         while (!STAILQ_EMPTY(&ic->ic_to_send)) {
1405                 pdu = STAILQ_FIRST(&ic->ic_to_send);
1406                 STAILQ_REMOVE_HEAD(&ic->ic_to_send, ip_next);
1407                 icl_pdu_free(pdu);
1408         }
1409
1410         KASSERT(STAILQ_EMPTY(&ic->ic_to_send),
1411             ("destroying session with non-empty send queue"));
1412 #ifdef DIAGNOSTIC
1413         KASSERT(ic->ic_outstanding_pdus == 0,
1414             ("destroying session with %d outstanding PDUs",
1415              ic->ic_outstanding_pdus));
1416 #endif
1417         ICL_CONN_UNLOCK(ic);
1418 }
1419
1420 bool
1421 icl_soft_conn_connected(struct icl_conn *ic)
1422 {
1423         ICL_CONN_LOCK_ASSERT_NOT(ic);
1424
1425         ICL_CONN_LOCK(ic);
1426         if (ic->ic_socket == NULL) {
1427                 ICL_CONN_UNLOCK(ic);
1428                 return (false);
1429         }
1430         if (ic->ic_socket->so_error != 0) {
1431                 ICL_CONN_UNLOCK(ic);
1432                 return (false);
1433         }
1434         ICL_CONN_UNLOCK(ic);
1435         return (true);
1436 }
1437
1438 static int
1439 icl_soft_limits(size_t *limitp)
1440 {
1441
1442         *limitp = 128 * 1024;
1443
1444         return (0);
1445 }
1446
1447 #ifdef ICL_KERNEL_PROXY
1448 int
1449 icl_conn_handoff_sock(struct icl_conn *ic, struct socket *so)
1450 {
1451         int error;
1452
1453         ICL_CONN_LOCK_ASSERT_NOT(ic);
1454
1455         if (so->so_type != SOCK_STREAM)
1456                 return (EINVAL);
1457
1458         ICL_CONN_LOCK(ic);
1459         if (ic->ic_socket != NULL) {
1460                 ICL_CONN_UNLOCK(ic);
1461                 return (EBUSY);
1462         }
1463         ic->ic_socket = so;
1464         ICL_CONN_UNLOCK(ic);
1465
1466         error = icl_conn_start(ic);
1467
1468         return (error);
1469 }
1470 #endif /* ICL_KERNEL_PROXY */
1471
1472 static int
1473 icl_soft_load(void)
1474 {
1475         int error;
1476
1477         icl_pdu_zone = uma_zcreate("icl_pdu",
1478             sizeof(struct icl_pdu), NULL, NULL, NULL, NULL,
1479             UMA_ALIGN_PTR, 0);
1480         refcount_init(&icl_ncons, 0);
1481
1482         /*
1483          * The reason we call this "none" is that to the user,
1484          * it's known as "offload driver"; "offload driver: soft"
1485          * doesn't make much sense.
1486          */
1487         error = icl_register("none", 0, icl_soft_limits, icl_soft_new_conn);
1488         KASSERT(error == 0, ("failed to register"));
1489
1490         return (error);
1491 }
1492
1493 static int
1494 icl_soft_unload(void)
1495 {
1496
1497         if (icl_ncons != 0)
1498                 return (EBUSY);
1499
1500         icl_unregister("none");
1501
1502         uma_zdestroy(icl_pdu_zone);
1503
1504         return (0);
1505 }
1506
1507 static int
1508 icl_soft_modevent(module_t mod, int what, void *arg)
1509 {
1510
1511         switch (what) {
1512         case MOD_LOAD:
1513                 return (icl_soft_load());
1514         case MOD_UNLOAD:
1515                 return (icl_soft_unload());
1516         default:
1517                 return (EINVAL);
1518         }
1519 }
1520
1521 moduledata_t icl_soft_data = {
1522         "icl_soft",
1523         icl_soft_modevent,
1524         0
1525 };
1526
1527 DECLARE_MODULE(icl_soft, icl_soft_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
1528 MODULE_DEPEND(icl_soft, icl, 1, 1, 1);
1529 MODULE_VERSION(icl_soft, 1);