sys/dev/iscsi/icl_soft.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2012 The FreeBSD Foundation
   5  * All rights reserved.
   6  *
   7  * This software was developed by Edward Tomasz Napierala under sponsorship
   8  * from the FreeBSD Foundation.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  */
  32
  33 /*
  34  * Software implementation of iSCSI Common Layer kobj(9) interface.
  35  */
  36
  37 #include <sys/cdefs.h>
  38 __FBSDID("$FreeBSD$");
  39
  40 #include <sys/param.h>
  41 #include <sys/capsicum.h>
  42 #include <sys/condvar.h>
  43 #include <sys/conf.h>
  44 #include <sys/file.h>
  45 #include <sys/kernel.h>
  46 #include <sys/kthread.h>
  47 #include <sys/lock.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/mutex.h>
  50 #include <sys/module.h>
  51 #include <sys/protosw.h>
  52 #include <sys/socket.h>
  53 #include <sys/socketvar.h>
  54 #include <sys/sysctl.h>
  55 #include <sys/systm.h>
  56 #include <sys/sx.h>
  57 #include <sys/uio.h>
  58 #include <vm/uma.h>
  59 #include <netinet/in.h>
  60 #include <netinet/tcp.h>
  61
  62 #include <dev/iscsi/icl.h>
  63 #include <dev/iscsi/iscsi_proto.h>
  64 #include <icl_conn_if.h>
  65
  66 static int coalesce = 1;
  67 SYSCTL_INT(_kern_icl, OID_AUTO, coalesce, CTLFLAG_RWTUN,
  68     &coalesce, 0, "Try to coalesce PDUs before sending");
  69 static int partial_receive_len = 128 * 1024;
  70 SYSCTL_INT(_kern_icl, OID_AUTO, partial_receive_len, CTLFLAG_RWTUN,
  71     &partial_receive_len, 0, "Minimum read size for partially received "
  72     "data segment");
  73 static int sendspace = 1048576;
  74 SYSCTL_INT(_kern_icl, OID_AUTO, sendspace, CTLFLAG_RWTUN,
  75     &sendspace, 0, "Default send socket buffer size");
  76 static int recvspace = 1048576;
  77 SYSCTL_INT(_kern_icl, OID_AUTO, recvspace, CTLFLAG_RWTUN,
  78     &recvspace, 0, "Default receive socket buffer size");
  79
  80 static MALLOC_DEFINE(M_ICL_SOFT, "icl_soft", "iSCSI software backend");
  81 static uma_zone_t icl_pdu_zone;
  82
  83 static volatile u_int   icl_ncons;
  84
  85 #define ICL_CONN_LOCK(X)                mtx_lock(X->ic_lock)
  86 #define ICL_CONN_UNLOCK(X)              mtx_unlock(X->ic_lock)
  87 #define ICL_CONN_LOCK_ASSERT(X)         mtx_assert(X->ic_lock, MA_OWNED)
  88 #define ICL_CONN_LOCK_ASSERT_NOT(X)     mtx_assert(X->ic_lock, MA_NOTOWNED)
  89
  90 STAILQ_HEAD(icl_pdu_stailq, icl_pdu);
  91
  92 static icl_conn_new_pdu_t       icl_soft_conn_new_pdu;
  93 static icl_conn_pdu_free_t      icl_soft_conn_pdu_free;
  94 static icl_conn_pdu_data_segment_length_t
  95                                     icl_soft_conn_pdu_data_segment_length;
  96 static icl_conn_pdu_append_data_t       icl_soft_conn_pdu_append_data;
  97 static icl_conn_pdu_get_data_t  icl_soft_conn_pdu_get_data;
  98 static icl_conn_pdu_queue_t     icl_soft_conn_pdu_queue;
  99 static icl_conn_handoff_t       icl_soft_conn_handoff;
 100 static icl_conn_free_t          icl_soft_conn_free;
 101 static icl_conn_close_t         icl_soft_conn_close;
 102 static icl_conn_task_setup_t    icl_soft_conn_task_setup;
 103 static icl_conn_task_done_t     icl_soft_conn_task_done;
 104 static icl_conn_transfer_setup_t        icl_soft_conn_transfer_setup;
 105 static icl_conn_transfer_done_t icl_soft_conn_transfer_done;
 106 #ifdef ICL_KERNEL_PROXY
 107 static icl_conn_connect_t       icl_soft_conn_connect;
 108 #endif
 109
 110 static kobj_method_t icl_soft_methods[] = {
 111         KOBJMETHOD(icl_conn_new_pdu, icl_soft_conn_new_pdu),
 112         KOBJMETHOD(icl_conn_pdu_free, icl_soft_conn_pdu_free),
 113         KOBJMETHOD(icl_conn_pdu_data_segment_length,
 114             icl_soft_conn_pdu_data_segment_length),
 115         KOBJMETHOD(icl_conn_pdu_append_data, icl_soft_conn_pdu_append_data),
 116         KOBJMETHOD(icl_conn_pdu_get_data, icl_soft_conn_pdu_get_data),
 117         KOBJMETHOD(icl_conn_pdu_queue, icl_soft_conn_pdu_queue),
 118         KOBJMETHOD(icl_conn_handoff, icl_soft_conn_handoff),
 119         KOBJMETHOD(icl_conn_free, icl_soft_conn_free),
 120         KOBJMETHOD(icl_conn_close, icl_soft_conn_close),
 121         KOBJMETHOD(icl_conn_task_setup, icl_soft_conn_task_setup),
 122         KOBJMETHOD(icl_conn_task_done, icl_soft_conn_task_done),
 123         KOBJMETHOD(icl_conn_transfer_setup, icl_soft_conn_transfer_setup),
 124         KOBJMETHOD(icl_conn_transfer_done, icl_soft_conn_transfer_done),
 125 #ifdef ICL_KERNEL_PROXY
 126         KOBJMETHOD(icl_conn_connect, icl_soft_conn_connect),
 127 #endif
 128         { 0, 0 }
 129 };
 130
 131 DEFINE_CLASS(icl_soft, icl_soft_methods, sizeof(struct icl_conn));
 132
 133 static void
 134 icl_conn_fail(struct icl_conn *ic)
 135 {
 136         if (ic->ic_socket == NULL)
 137                 return;
 138
 139         /*
 140          * XXX
 141          */
 142         ic->ic_socket->so_error = EDOOFUS;
 143         (ic->ic_error)(ic);
 144 }
 145
 146 static struct mbuf *
 147 icl_conn_receive(struct icl_conn *ic, size_t len)
 148 {
 149         struct uio uio;
 150         struct socket *so;
 151         struct mbuf *m;
 152         int error, flags;
 153
 154         so = ic->ic_socket;
 155
 156         memset(&uio, 0, sizeof(uio));
 157         uio.uio_resid = len;
 158
 159         flags = MSG_DONTWAIT;
 160         error = soreceive(so, NULL, &uio, &m, NULL, &flags);
 161         if (error != 0) {
 162                 ICL_DEBUG("soreceive error %d", error);
 163                 return (NULL);
 164         }
 165         if (uio.uio_resid != 0) {
 166                 m_freem(m);
 167                 ICL_DEBUG("short read");
 168                 return (NULL);
 169         }
 170
 171         return (m);
 172 }
 173
 174 static int
 175 icl_conn_receive_buf(struct icl_conn *ic, void *buf, size_t len)
 176 {
 177         struct iovec iov[1];
 178         struct uio uio;
 179         struct socket *so;
 180         int error, flags;
 181
 182         so = ic->ic_socket;
 183
 184         memset(&uio, 0, sizeof(uio));
 185         iov[0].iov_base = buf;
 186         iov[0].iov_len = len;
 187         uio.uio_iov = iov;
 188         uio.uio_iovcnt = 1;
 189         uio.uio_offset = 0;
 190         uio.uio_resid = len;
 191         uio.uio_segflg = UIO_SYSSPACE;
 192         uio.uio_rw = UIO_READ;
 193
 194         flags = MSG_DONTWAIT;
 195         error = soreceive(so, NULL, &uio, NULL, NULL, &flags);
 196         if (error != 0) {
 197                 ICL_DEBUG("soreceive error %d", error);
 198                 return (-1);
 199         }
 200         if (uio.uio_resid != 0) {
 201                 ICL_DEBUG("short read");
 202                 return (-1);
 203         }
 204
 205         return (0);
 206 }
 207
 208 static void
 209 icl_soft_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip)
 210 {
 211
 212         m_freem(ip->ip_bhs_mbuf);
 213         m_freem(ip->ip_ahs_mbuf);
 214         m_freem(ip->ip_data_mbuf);
 215         uma_zfree(icl_pdu_zone, ip);
 216 #ifdef DIAGNOSTIC
 217         refcount_release(&ic->ic_outstanding_pdus);
 218 #endif
 219 }
 220
 221 /*
 222  * Allocate icl_pdu with empty BHS to fill up by the caller.
 223  */
 224 struct icl_pdu *
 225 icl_soft_conn_new_pdu(struct icl_conn *ic, int flags)
 226 {
 227         struct icl_pdu *ip;
 228
 229 #ifdef DIAGNOSTIC
 230         refcount_acquire(&ic->ic_outstanding_pdus);
 231 #endif
 232         ip = uma_zalloc(icl_pdu_zone, flags | M_ZERO);
 233         if (ip == NULL) {
 234                 ICL_WARN("failed to allocate %zd bytes", sizeof(*ip));
 235 #ifdef DIAGNOSTIC
 236                 refcount_release(&ic->ic_outstanding_pdus);
 237 #endif
 238                 return (NULL);
 239         }
 240         ip->ip_conn = ic;
 241
 242         CTASSERT(sizeof(struct iscsi_bhs) <= MHLEN);
 243         ip->ip_bhs_mbuf = m_gethdr(flags, MT_DATA);
 244         if (ip->ip_bhs_mbuf == NULL) {
 245                 ICL_WARN("failed to allocate BHS mbuf");
 246                 icl_soft_conn_pdu_free(ic, ip);
 247                 return (NULL);
 248         }
 249         ip->ip_bhs = mtod(ip->ip_bhs_mbuf, struct iscsi_bhs *);
 250         memset(ip->ip_bhs, 0, sizeof(struct iscsi_bhs));
 251         ip->ip_bhs_mbuf->m_len = sizeof(struct iscsi_bhs);
 252
 253         return (ip);
 254 }
 255
 256 static int
 257 icl_pdu_ahs_length(const struct icl_pdu *request)
 258 {
 259
 260         return (request->ip_bhs->bhs_total_ahs_len * 4);
 261 }
 262
 263 static size_t
 264 icl_pdu_data_segment_length(const struct icl_pdu *request)
 265 {
 266         uint32_t len = 0;
 267
 268         len += request->ip_bhs->bhs_data_segment_len[0];
 269         len <<= 8;
 270         len += request->ip_bhs->bhs_data_segment_len[1];
 271         len <<= 8;
 272         len += request->ip_bhs->bhs_data_segment_len[2];
 273
 274         return (len);
 275 }
 276
 277 size_t
 278 icl_soft_conn_pdu_data_segment_length(struct icl_conn *ic,
 279     const struct icl_pdu *request)
 280 {
 281
 282         return (icl_pdu_data_segment_length(request));
 283 }
 284
 285 static void
 286 icl_pdu_set_data_segment_length(struct icl_pdu *response, uint32_t len)
 287 {
 288
 289         response->ip_bhs->bhs_data_segment_len[2] = len;
 290         response->ip_bhs->bhs_data_segment_len[1] = len >> 8;
 291         response->ip_bhs->bhs_data_segment_len[0] = len >> 16;
 292 }
 293
 294 static size_t
 295 icl_pdu_padding(const struct icl_pdu *ip)
 296 {
 297
 298         if ((ip->ip_data_len % 4) != 0)
 299                 return (4 - (ip->ip_data_len % 4));
 300
 301         return (0);
 302 }
 303
 304 static size_t
 305 icl_pdu_size(const struct icl_pdu *response)
 306 {
 307         size_t len;
 308
 309         KASSERT(response->ip_ahs_len == 0, ("responding with AHS"));
 310
 311         len = sizeof(struct iscsi_bhs) + response->ip_data_len +
 312             icl_pdu_padding(response);
 313         if (response->ip_conn->ic_header_crc32c)
 314                 len += ISCSI_HEADER_DIGEST_SIZE;
 315         if (response->ip_data_len != 0 && response->ip_conn->ic_data_crc32c)
 316                 len += ISCSI_DATA_DIGEST_SIZE;
 317
 318         return (len);
 319 }
 320
 321 static int
 322 icl_pdu_receive_bhs(struct icl_pdu *request, size_t *availablep)
 323 {
 324
 325         if (icl_conn_receive_buf(request->ip_conn,
 326             request->ip_bhs, sizeof(struct iscsi_bhs))) {
 327                 ICL_DEBUG("failed to receive BHS");
 328                 return (-1);
 329         }
 330
 331         *availablep -= sizeof(struct iscsi_bhs);
 332         return (0);
 333 }
 334
 335 static int
 336 icl_pdu_receive_ahs(struct icl_pdu *request, size_t *availablep)
 337 {
 338
 339         request->ip_ahs_len = icl_pdu_ahs_length(request);
 340         if (request->ip_ahs_len == 0)
 341                 return (0);
 342
 343         request->ip_ahs_mbuf = icl_conn_receive(request->ip_conn,
 344             request->ip_ahs_len);
 345         if (request->ip_ahs_mbuf == NULL) {
 346                 ICL_DEBUG("failed to receive AHS");
 347                 return (-1);
 348         }
 349
 350         *availablep -= request->ip_ahs_len;
 351         return (0);
 352 }
 353
 354 static uint32_t
 355 icl_mbuf_to_crc32c(const struct mbuf *m0)
 356 {
 357         uint32_t digest = 0xffffffff;
 358         const struct mbuf *m;
 359
 360         for (m = m0; m != NULL; m = m->m_next)
 361                 digest = calculate_crc32c(digest,
 362                     mtod(m, const void *), m->m_len);
 363
 364         digest = digest ^ 0xffffffff;
 365
 366         return (digest);
 367 }
 368
 369 static int
 370 icl_pdu_check_header_digest(struct icl_pdu *request, size_t *availablep)
 371 {
 372         uint32_t received_digest, valid_digest;
 373
 374         if (request->ip_conn->ic_header_crc32c == false)
 375                 return (0);
 376
 377         CTASSERT(sizeof(received_digest) == ISCSI_HEADER_DIGEST_SIZE);
 378         if (icl_conn_receive_buf(request->ip_conn,
 379             &received_digest, ISCSI_HEADER_DIGEST_SIZE)) {
 380                 ICL_DEBUG("failed to receive header digest");
 381                 return (-1);
 382         }
 383         *availablep -= ISCSI_HEADER_DIGEST_SIZE;
 384
 385         /* Temporary attach AHS to BHS to calculate header digest. */
 386         request->ip_bhs_mbuf->m_next = request->ip_ahs_mbuf;
 387         valid_digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
 388         request->ip_bhs_mbuf->m_next = NULL;
 389         if (received_digest != valid_digest) {
 390                 ICL_WARN("header digest check failed; got 0x%x, "
 391                     "should be 0x%x", received_digest, valid_digest);
 392                 return (-1);
 393         }
 394
 395         return (0);
 396 }
 397
 398 /*
 399  * Return the number of bytes that should be waiting in the receive socket
 400  * before icl_pdu_receive_data_segment() gets called.
 401  */
 402 static size_t
 403 icl_pdu_data_segment_receive_len(const struct icl_pdu *request)
 404 {
 405         size_t len;
 406
 407         len = icl_pdu_data_segment_length(request);
 408         if (len == 0)
 409                 return (0);
 410
 411         /*
 412          * Account for the parts of data segment already read from
 413          * the socket buffer.
 414          */
 415         KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
 416         len -= request->ip_data_len;
 417
 418         /*
 419          * Don't always wait for the full data segment to be delivered
 420          * to the socket; this might badly affect performance due to
 421          * TCP window scaling.
 422          */
 423         if (len > partial_receive_len) {
 424 #if 0
 425                 ICL_DEBUG("need %zd bytes of data, limiting to %zd",
 426                     len, partial_receive_len));
 427 #endif
 428                 len = partial_receive_len;
 429
 430                 return (len);
 431         }
 432
 433         /*
 434          * Account for padding.  Note that due to the way code is written,
 435          * the icl_pdu_receive_data_segment() must always receive padding
 436          * along with the last part of data segment, because it would be
 437          * impossible to tell whether we've already received the full data
 438          * segment including padding, or without it.
 439          */
 440         if ((len % 4) != 0)
 441                 len += 4 - (len % 4);
 442
 443 #if 0
 444         ICL_DEBUG("need %zd bytes of data", len));
 445 #endif
 446
 447         return (len);
 448 }
 449
 450 static int
 451 icl_pdu_receive_data_segment(struct icl_pdu *request,
 452     size_t *availablep, bool *more_neededp)
 453 {
 454         struct icl_conn *ic;
 455         size_t len, padding = 0;
 456         struct mbuf *m;
 457
 458         ic = request->ip_conn;
 459
 460         *more_neededp = false;
 461         ic->ic_receive_len = 0;
 462
 463         len = icl_pdu_data_segment_length(request);
 464         if (len == 0)
 465                 return (0);
 466
 467         if ((len % 4) != 0)
 468                 padding = 4 - (len % 4);
 469
 470         /*
 471          * Account for already received parts of data segment.
 472          */
 473         KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
 474         len -= request->ip_data_len;
 475
 476         if (len + padding > *availablep) {
 477                 /*
 478                  * Not enough data in the socket buffer.  Receive as much
 479                  * as we can.  Don't receive padding, since, obviously, it's
 480                  * not the end of data segment yet.
 481                  */
 482 #if 0
 483                 ICL_DEBUG("limited from %zd to %zd",
 484                     len + padding, *availablep - padding));
 485 #endif
 486                 len = *availablep - padding;
 487                 *more_neededp = true;
 488                 padding = 0;
 489         }
 490
 491         /*
 492          * Must not try to receive padding without at least one byte
 493          * of actual data segment.
 494          */
 495         if (len > 0) {
 496                 m = icl_conn_receive(request->ip_conn, len + padding);
 497                 if (m == NULL) {
 498                         ICL_DEBUG("failed to receive data segment");
 499                         return (-1);
 500                 }
 501
 502                 if (request->ip_data_mbuf == NULL)
 503                         request->ip_data_mbuf = m;
 504                 else
 505                         m_cat(request->ip_data_mbuf, m);
 506
 507                 request->ip_data_len += len;
 508                 *availablep -= len + padding;
 509         } else
 510                 ICL_DEBUG("len 0");
 511
 512         if (*more_neededp)
 513                 ic->ic_receive_len =
 514                     icl_pdu_data_segment_receive_len(request);
 515
 516         return (0);
 517 }
 518
 519 static int
 520 icl_pdu_check_data_digest(struct icl_pdu *request, size_t *availablep)
 521 {
 522         uint32_t received_digest, valid_digest;
 523
 524         if (request->ip_conn->ic_data_crc32c == false)
 525                 return (0);
 526
 527         if (request->ip_data_len == 0)
 528                 return (0);
 529
 530         CTASSERT(sizeof(received_digest) == ISCSI_DATA_DIGEST_SIZE);
 531         if (icl_conn_receive_buf(request->ip_conn,
 532             &received_digest, ISCSI_DATA_DIGEST_SIZE)) {
 533                 ICL_DEBUG("failed to receive data digest");
 534                 return (-1);
 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_soft_conn_new_pdu(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_soft_conn_pdu_free(ic, 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         for (;;) {
 771                 if (ic->ic_disconnecting) {
 772                         //ICL_DEBUG("terminating");
 773                         break;
 774                 }
 775
 776                 /*
 777                  * Set the low watermark, to be checked by
 778                  * soreadable() in icl_soupcall_receive()
 779                  * to avoid unnecessary wakeups until there
 780                  * is enough data received to read the PDU.
 781                  */
 782                 SOCKBUF_LOCK(&so->so_rcv);
 783                 available = sbavail(&so->so_rcv);
 784                 if (available < ic->ic_receive_len) {
 785                         so->so_rcv.sb_lowat = ic->ic_receive_len;
 786                         cv_wait(&ic->ic_receive_cv, &so->so_rcv.sb_mtx);
 787                 } else
 788                         so->so_rcv.sb_lowat = so->so_rcv.sb_hiwat + 1;
 789                 SOCKBUF_UNLOCK(&so->so_rcv);
 790
 791                 icl_conn_receive_pdus(ic, available);
 792         }
 793
 794         ICL_CONN_LOCK(ic);
 795         ic->ic_receive_running = false;
 796         cv_signal(&ic->ic_send_cv);
 797         ICL_CONN_UNLOCK(ic);
 798         kthread_exit();
 799 }
 800
 801 static int
 802 icl_soupcall_receive(struct socket *so, void *arg, int waitflag)
 803 {
 804         struct icl_conn *ic;
 805
 806         if (!soreadable(so))
 807                 return (SU_OK);
 808
 809         ic = arg;
 810         cv_signal(&ic->ic_receive_cv);
 811         return (SU_OK);
 812 }
 813
 814 static int
 815 icl_pdu_finalize(struct icl_pdu *request)
 816 {
 817         size_t padding, pdu_len;
 818         uint32_t digest, zero = 0;
 819         int ok;
 820         struct icl_conn *ic;
 821
 822         ic = request->ip_conn;
 823
 824         icl_pdu_set_data_segment_length(request, request->ip_data_len);
 825
 826         pdu_len = icl_pdu_size(request);
 827
 828         if (ic->ic_header_crc32c) {
 829                 digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
 830                 ok = m_append(request->ip_bhs_mbuf, sizeof(digest),
 831                     (void *)&digest);
 832                 if (ok != 1) {
 833                         ICL_WARN("failed to append header digest");
 834                         return (1);
 835                 }
 836         }
 837
 838         if (request->ip_data_len != 0) {
 839                 padding = icl_pdu_padding(request);
 840                 if (padding > 0) {
 841                         ok = m_append(request->ip_data_mbuf, padding,
 842                             (void *)&zero);
 843                         if (ok != 1) {
 844                                 ICL_WARN("failed to append padding");
 845                                 return (1);
 846                         }
 847                 }
 848
 849                 if (ic->ic_data_crc32c) {
 850                         digest = icl_mbuf_to_crc32c(request->ip_data_mbuf);
 851
 852                         ok = m_append(request->ip_data_mbuf, sizeof(digest),
 853                             (void *)&digest);
 854                         if (ok != 1) {
 855                                 ICL_WARN("failed to append data digest");
 856                                 return (1);
 857                         }
 858                 }
 859
 860                 m_cat(request->ip_bhs_mbuf, request->ip_data_mbuf);
 861                 request->ip_data_mbuf = NULL;
 862         }
 863
 864         request->ip_bhs_mbuf->m_pkthdr.len = pdu_len;
 865
 866         return (0);
 867 }
 868
 869 static void
 870 icl_conn_send_pdus(struct icl_conn *ic, struct icl_pdu_stailq *queue)
 871 {
 872         struct icl_pdu *request, *request2;
 873         struct socket *so;
 874         long available, size, size2;
 875         int coalesced, error;
 876
 877         ICL_CONN_LOCK_ASSERT_NOT(ic);
 878
 879         so = ic->ic_socket;
 880
 881         SOCKBUF_LOCK(&so->so_snd);
 882         /*
 883          * Check how much space do we have for transmit.  We can't just
 884          * call sosend() and retry when we get EWOULDBLOCK or EMSGSIZE,
 885          * as it always frees the mbuf chain passed to it, even in case
 886          * of error.
 887          */
 888         available = sbspace(&so->so_snd);
 889
 890         /*
 891          * Notify the socket upcall that we don't need wakeups
 892          * for the time being.
 893          */
 894         so->so_snd.sb_lowat = so->so_snd.sb_hiwat + 1;
 895         SOCKBUF_UNLOCK(&so->so_snd);
 896
 897         while (!STAILQ_EMPTY(queue)) {
 898                 request = STAILQ_FIRST(queue);
 899                 size = icl_pdu_size(request);
 900                 if (available < size) {
 901
 902                         /*
 903                          * Set the low watermark, to be checked by
 904                          * sowriteable() in icl_soupcall_send()
 905                          * to avoid unnecessary wakeups until there
 906                          * is enough space for the PDU to fit.
 907                          */
 908                         SOCKBUF_LOCK(&so->so_snd);
 909                         available = sbspace(&so->so_snd);
 910                         if (available < size) {
 911 #if 1
 912                                 ICL_DEBUG("no space to send; "
 913                                     "have %ld, need %ld",
 914                                     available, size);
 915 #endif
 916                                 so->so_snd.sb_lowat = size;
 917                                 SOCKBUF_UNLOCK(&so->so_snd);
 918                                 return;
 919                         }
 920                         SOCKBUF_UNLOCK(&so->so_snd);
 921                 }
 922                 STAILQ_REMOVE_HEAD(queue, ip_next);
 923                 error = icl_pdu_finalize(request);
 924                 if (error != 0) {
 925                         ICL_DEBUG("failed to finalize PDU; "
 926                             "dropping connection");
 927                         icl_soft_conn_pdu_free(ic, request);
 928                         icl_conn_fail(ic);
 929                         return;
 930                 }
 931                 if (coalesce) {
 932                         coalesced = 1;
 933                         for (;;) {
 934                                 request2 = STAILQ_FIRST(queue);
 935                                 if (request2 == NULL)
 936                                         break;
 937                                 size2 = icl_pdu_size(request2);
 938                                 if (available < size + size2)
 939                                         break;
 940                                 STAILQ_REMOVE_HEAD(queue, ip_next);
 941                                 error = icl_pdu_finalize(request2);
 942                                 if (error != 0) {
 943                                         ICL_DEBUG("failed to finalize PDU; "
 944                                             "dropping connection");
 945                                         icl_soft_conn_pdu_free(ic, request);
 946                                         icl_soft_conn_pdu_free(ic, request2);
 947                                         icl_conn_fail(ic);
 948                                         return;
 949                                 }
 950                                 m_cat(request->ip_bhs_mbuf, request2->ip_bhs_mbuf);
 951                                 request2->ip_bhs_mbuf = NULL;
 952                                 request->ip_bhs_mbuf->m_pkthdr.len += size2;
 953                                 size += size2;
 954                                 STAILQ_REMOVE_AFTER(queue, request, ip_next);
 955                                 icl_soft_conn_pdu_free(ic, request2);
 956                                 coalesced++;
 957                         }
 958 #if 0
 959                         if (coalesced > 1) {
 960                                 ICL_DEBUG("coalesced %d PDUs into %ld bytes",
 961                                     coalesced, size);
 962                         }
 963 #endif
 964                 }
 965                 available -= size;
 966                 error = sosend(so, NULL, NULL, request->ip_bhs_mbuf,
 967                     NULL, MSG_DONTWAIT, curthread);
 968                 request->ip_bhs_mbuf = NULL; /* Sosend consumes the mbuf. */
 969                 if (error != 0) {
 970                         ICL_DEBUG("failed to send PDU, error %d; "
 971                             "dropping connection", error);
 972                         icl_soft_conn_pdu_free(ic, request);
 973                         icl_conn_fail(ic);
 974                         return;
 975                 }
 976                 icl_soft_conn_pdu_free(ic, request);
 977         }
 978 }
 979
 980 static void
 981 icl_send_thread(void *arg)
 982 {
 983         struct icl_conn *ic;
 984         struct icl_pdu_stailq queue;
 985
 986         ic = arg;
 987
 988         STAILQ_INIT(&queue);
 989
 990         ICL_CONN_LOCK(ic);
 991         for (;;) {
 992                 for (;;) {
 993                         /*
 994                          * If the local queue is empty, populate it from
 995                          * the main one.  This way the icl_conn_send_pdus()
 996                          * can go through all the queued PDUs without holding
 997                          * any locks.
 998                          */
 999                         if (STAILQ_EMPTY(&queue))
1000                                 STAILQ_SWAP(&ic->ic_to_send, &queue, icl_pdu);
1001
1002                         ic->ic_check_send_space = false;
1003                         ICL_CONN_UNLOCK(ic);
1004                         icl_conn_send_pdus(ic, &queue);
1005                         ICL_CONN_LOCK(ic);
1006
1007                         /*
1008                          * The icl_soupcall_send() was called since the last
1009                          * call to sbspace(); go around;
1010                          */
1011                         if (ic->ic_check_send_space)
1012                                 continue;
1013
1014                         /*
1015                          * Local queue is empty, but we still have PDUs
1016                          * in the main one; go around.
1017                          */
1018                         if (STAILQ_EMPTY(&queue) &&
1019                             !STAILQ_EMPTY(&ic->ic_to_send))
1020                                 continue;
1021
1022                         /*
1023                          * There might be some stuff in the local queue,
1024                          * which didn't get sent due to not having enough send
1025                          * space.  Wait for socket upcall.
1026                          */
1027                         break;
1028                 }
1029
1030                 if (ic->ic_disconnecting) {
1031                         //ICL_DEBUG("terminating");
1032                         break;
1033                 }
1034
1035                 cv_wait(&ic->ic_send_cv, ic->ic_lock);
1036         }
1037
1038         /*
1039          * We're exiting; move PDUs back to the main queue, so they can
1040          * get freed properly.  At this point ordering doesn't matter.
1041          */
1042         STAILQ_CONCAT(&ic->ic_to_send, &queue);
1043
1044         ic->ic_send_running = false;
1045         cv_signal(&ic->ic_send_cv);
1046         ICL_CONN_UNLOCK(ic);
1047         kthread_exit();
1048 }
1049
1050 static int
1051 icl_soupcall_send(struct socket *so, void *arg, int waitflag)
1052 {
1053         struct icl_conn *ic;
1054
1055         if (!sowriteable(so))
1056                 return (SU_OK);
1057
1058         ic = arg;
1059
1060         ICL_CONN_LOCK(ic);
1061         ic->ic_check_send_space = true;
1062         ICL_CONN_UNLOCK(ic);
1063
1064         cv_signal(&ic->ic_send_cv);
1065
1066         return (SU_OK);
1067 }
1068
1069 static int
1070 icl_soft_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *request,
1071     const void *addr, size_t len, int flags)
1072 {
1073         struct mbuf *mb, *newmb;
1074         size_t copylen, off = 0;
1075
1076         KASSERT(len > 0, ("len == 0"));
1077
1078         newmb = m_getm2(NULL, len, flags, MT_DATA, 0);
1079         if (newmb == NULL) {
1080                 ICL_WARN("failed to allocate mbuf for %zd bytes", len);
1081                 return (ENOMEM);
1082         }
1083
1084         for (mb = newmb; mb != NULL; mb = mb->m_next) {
1085                 copylen = min(M_TRAILINGSPACE(mb), len - off);
1086                 memcpy(mtod(mb, char *), (const char *)addr + off, copylen);
1087                 mb->m_len = copylen;
1088                 off += copylen;
1089         }
1090         KASSERT(off == len, ("%s: off != len", __func__));
1091
1092         if (request->ip_data_mbuf == NULL) {
1093                 request->ip_data_mbuf = newmb;
1094                 request->ip_data_len = len;
1095         } else {
1096                 m_cat(request->ip_data_mbuf, newmb);
1097                 request->ip_data_len += len;
1098         }
1099
1100         return (0);
1101 }
1102
1103 void
1104 icl_soft_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip,
1105     size_t off, void *addr, size_t len)
1106 {
1107
1108         m_copydata(ip->ip_data_mbuf, off, len, addr);
1109 }
1110
1111 static void
1112 icl_pdu_queue(struct icl_pdu *ip)
1113 {
1114         struct icl_conn *ic;
1115
1116         ic = ip->ip_conn;
1117
1118         ICL_CONN_LOCK_ASSERT(ic);
1119
1120         if (ic->ic_disconnecting || ic->ic_socket == NULL) {
1121                 ICL_DEBUG("icl_pdu_queue on closed connection");
1122                 icl_soft_conn_pdu_free(ic, ip);
1123                 return;
1124         }
1125
1126         if (!STAILQ_EMPTY(&ic->ic_to_send)) {
1127                 STAILQ_INSERT_TAIL(&ic->ic_to_send, ip, ip_next);
1128                 /*
1129                  * If the queue is not empty, someone else had already
1130                  * signaled the send thread; no need to do that again,
1131                  * just return.
1132                  */
1133                 return;
1134         }
1135
1136         STAILQ_INSERT_TAIL(&ic->ic_to_send, ip, ip_next);
1137         cv_signal(&ic->ic_send_cv);
1138 }
1139
1140 void
1141 icl_soft_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip)
1142 {
1143
1144         icl_pdu_queue(ip);
1145 }
1146
1147 static struct icl_conn *
1148 icl_soft_new_conn(const char *name, struct mtx *lock)
1149 {
1150         struct icl_conn *ic;
1151
1152         refcount_acquire(&icl_ncons);
1153
1154         ic = (struct icl_conn *)kobj_create(&icl_soft_class, M_ICL_SOFT, M_WAITOK | M_ZERO);
1155
1156         STAILQ_INIT(&ic->ic_to_send);
1157         ic->ic_lock = lock;
1158         cv_init(&ic->ic_send_cv, "icl_tx");
1159         cv_init(&ic->ic_receive_cv, "icl_rx");
1160 #ifdef DIAGNOSTIC
1161         refcount_init(&ic->ic_outstanding_pdus, 0);
1162 #endif
1163         ic->ic_max_data_segment_length = ICL_MAX_DATA_SEGMENT_LENGTH;
1164         ic->ic_name = name;
1165         ic->ic_offload = "None";
1166         ic->ic_unmapped = false;
1167
1168         return (ic);
1169 }
1170
1171 void
1172 icl_soft_conn_free(struct icl_conn *ic)
1173 {
1174
1175 #ifdef DIAGNOSTIC
1176         KASSERT(ic->ic_outstanding_pdus == 0,
1177             ("destroying session with %d outstanding PDUs",
1178              ic->ic_outstanding_pdus));
1179 #endif
1180         cv_destroy(&ic->ic_send_cv);
1181         cv_destroy(&ic->ic_receive_cv);
1182         kobj_delete((struct kobj *)ic, M_ICL_SOFT);
1183         refcount_release(&icl_ncons);
1184 }
1185
1186 static int
1187 icl_conn_start(struct icl_conn *ic)
1188 {
1189         size_t minspace;
1190         struct sockopt opt;
1191         int error, one = 1;
1192
1193         ICL_CONN_LOCK(ic);
1194
1195         /*
1196          * XXX: Ugly hack.
1197          */
1198         if (ic->ic_socket == NULL) {
1199                 ICL_CONN_UNLOCK(ic);
1200                 return (EINVAL);
1201         }
1202
1203         ic->ic_receive_state = ICL_CONN_STATE_BHS;
1204         ic->ic_receive_len = sizeof(struct iscsi_bhs);
1205         ic->ic_disconnecting = false;
1206
1207         ICL_CONN_UNLOCK(ic);
1208
1209         /*
1210          * For sendspace, this is required because the current code cannot
1211          * send a PDU in pieces; thus, the minimum buffer size is equal
1212          * to the maximum PDU size.  "+4" is to account for possible padding.
1213          *
1214          * What we should actually do here is to use autoscaling, but set
1215          * some minimal buffer size to "minspace".  I don't know a way to do
1216          * that, though.
1217          */
1218         minspace = sizeof(struct iscsi_bhs) + ic->ic_max_data_segment_length +
1219             ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE + 4;
1220         if (sendspace < minspace) {
1221                 ICL_WARN("kern.icl.sendspace too low; must be at least %zd",
1222                     minspace);
1223                 sendspace = minspace;
1224         }
1225         if (recvspace < minspace) {
1226                 ICL_WARN("kern.icl.recvspace too low; must be at least %zd",
1227                     minspace);
1228                 recvspace = minspace;
1229         }
1230
1231         error = soreserve(ic->ic_socket, sendspace, recvspace);
1232         if (error != 0) {
1233                 ICL_WARN("soreserve failed with error %d", error);
1234                 icl_soft_conn_close(ic);
1235                 return (error);
1236         }
1237         ic->ic_socket->so_snd.sb_flags |= SB_AUTOSIZE;
1238         ic->ic_socket->so_rcv.sb_flags |= SB_AUTOSIZE;
1239
1240         /*
1241          * Disable Nagle.
1242          */
1243         bzero(&opt, sizeof(opt));
1244         opt.sopt_dir = SOPT_SET;
1245         opt.sopt_level = IPPROTO_TCP;
1246         opt.sopt_name = TCP_NODELAY;
1247         opt.sopt_val = &one;
1248         opt.sopt_valsize = sizeof(one);
1249         error = sosetopt(ic->ic_socket, &opt);
1250         if (error != 0) {
1251                 ICL_WARN("disabling TCP_NODELAY failed with error %d", error);
1252                 icl_soft_conn_close(ic);
1253                 return (error);
1254         }
1255
1256         /*
1257          * Register socket upcall, to get notified about incoming PDUs
1258          * and free space to send outgoing ones.
1259          */
1260         SOCKBUF_LOCK(&ic->ic_socket->so_snd);
1261         soupcall_set(ic->ic_socket, SO_SND, icl_soupcall_send, ic);
1262         SOCKBUF_UNLOCK(&ic->ic_socket->so_snd);
1263         SOCKBUF_LOCK(&ic->ic_socket->so_rcv);
1264         soupcall_set(ic->ic_socket, SO_RCV, icl_soupcall_receive, ic);
1265         SOCKBUF_UNLOCK(&ic->ic_socket->so_rcv);
1266
1267         /*
1268          * Start threads.
1269          */
1270         ICL_CONN_LOCK(ic);
1271         ic->ic_send_running = ic->ic_receive_running = true;
1272         ICL_CONN_UNLOCK(ic);
1273         error = kthread_add(icl_send_thread, ic, NULL, NULL, 0, 0, "%stx",
1274             ic->ic_name);
1275         if (error != 0) {
1276                 ICL_WARN("kthread_add(9) failed with error %d", error);
1277                 ICL_CONN_LOCK(ic);
1278                 ic->ic_send_running = ic->ic_receive_running = false;
1279                 cv_signal(&ic->ic_send_cv);
1280                 ICL_CONN_UNLOCK(ic);
1281                 icl_soft_conn_close(ic);
1282                 return (error);
1283         }
1284         error = kthread_add(icl_receive_thread, ic, NULL, NULL, 0, 0, "%srx",
1285             ic->ic_name);
1286         if (error != 0) {
1287                 ICL_WARN("kthread_add(9) failed with error %d", error);
1288                 ICL_CONN_LOCK(ic);
1289                 ic->ic_receive_running = false;
1290                 cv_signal(&ic->ic_send_cv);
1291                 ICL_CONN_UNLOCK(ic);
1292                 icl_soft_conn_close(ic);
1293                 return (error);
1294         }
1295
1296         return (0);
1297 }
1298
1299 int
1300 icl_soft_conn_handoff(struct icl_conn *ic, int fd)
1301 {
1302         struct file *fp;
1303         struct socket *so;
1304         cap_rights_t rights;
1305         int error;
1306
1307         ICL_CONN_LOCK_ASSERT_NOT(ic);
1308
1309 #ifdef ICL_KERNEL_PROXY
1310         /*
1311          * We're transitioning to Full Feature phase, and we don't
1312          * really care.
1313          */
1314         if (fd == 0) {
1315                 ICL_CONN_LOCK(ic);
1316                 if (ic->ic_socket == NULL) {
1317                         ICL_CONN_UNLOCK(ic);
1318                         ICL_WARN("proxy handoff without connect");
1319                         return (EINVAL);
1320                 }
1321                 ICL_CONN_UNLOCK(ic);
1322                 return (0);
1323         }
1324 #endif
1325
1326         /*
1327          * Steal the socket from userland.
1328          */
1329         error = fget(curthread, fd,
1330             cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1331         if (error != 0)
1332                 return (error);
1333         if (fp->f_type != DTYPE_SOCKET) {
1334                 fdrop(fp, curthread);
1335                 return (EINVAL);
1336         }
1337         so = fp->f_data;
1338         if (so->so_type != SOCK_STREAM) {
1339                 fdrop(fp, curthread);
1340                 return (EINVAL);
1341         }
1342
1343         ICL_CONN_LOCK(ic);
1344
1345         if (ic->ic_socket != NULL) {
1346                 ICL_CONN_UNLOCK(ic);
1347                 fdrop(fp, curthread);
1348                 return (EBUSY);
1349         }
1350
1351         ic->ic_socket = fp->f_data;
1352         fp->f_ops = &badfileops;
1353         fp->f_data = NULL;
1354         fdrop(fp, curthread);
1355         ICL_CONN_UNLOCK(ic);
1356
1357         error = icl_conn_start(ic);
1358
1359         return (error);
1360 }
1361
1362 void
1363 icl_soft_conn_close(struct icl_conn *ic)
1364 {
1365         struct icl_pdu *pdu;
1366         struct socket *so;
1367
1368         ICL_CONN_LOCK(ic);
1369
1370         /*
1371          * Wake up the threads, so they can properly terminate.
1372          */
1373         ic->ic_disconnecting = true;
1374         while (ic->ic_receive_running || ic->ic_send_running) {
1375                 cv_signal(&ic->ic_receive_cv);
1376                 cv_signal(&ic->ic_send_cv);
1377                 cv_wait(&ic->ic_send_cv, ic->ic_lock);
1378         }
1379
1380         /* Some other thread could close the connection same time. */
1381         so = ic->ic_socket;
1382         if (so == NULL) {
1383                 ICL_CONN_UNLOCK(ic);
1384                 return;
1385         }
1386         ic->ic_socket = NULL;
1387
1388         /*
1389          * Deregister socket upcalls.
1390          */
1391         ICL_CONN_UNLOCK(ic);
1392         SOCKBUF_LOCK(&so->so_snd);
1393         if (so->so_snd.sb_upcall != NULL)
1394                 soupcall_clear(so, SO_SND);
1395         SOCKBUF_UNLOCK(&so->so_snd);
1396         SOCKBUF_LOCK(&so->so_rcv);
1397         if (so->so_rcv.sb_upcall != NULL)
1398                 soupcall_clear(so, SO_RCV);
1399         SOCKBUF_UNLOCK(&so->so_rcv);
1400         soclose(so);
1401         ICL_CONN_LOCK(ic);
1402
1403         if (ic->ic_receive_pdu != NULL) {
1404                 //ICL_DEBUG("freeing partially received PDU");
1405                 icl_soft_conn_pdu_free(ic, ic->ic_receive_pdu);
1406                 ic->ic_receive_pdu = NULL;
1407         }
1408
1409         /*
1410          * Remove any outstanding PDUs from the send queue.
1411          */
1412         while (!STAILQ_EMPTY(&ic->ic_to_send)) {
1413                 pdu = STAILQ_FIRST(&ic->ic_to_send);
1414                 STAILQ_REMOVE_HEAD(&ic->ic_to_send, ip_next);
1415                 icl_soft_conn_pdu_free(ic, pdu);
1416         }
1417
1418         KASSERT(STAILQ_EMPTY(&ic->ic_to_send),
1419             ("destroying session with non-empty send queue"));
1420         ICL_CONN_UNLOCK(ic);
1421 }
1422
1423 int
1424 icl_soft_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip,
1425     struct ccb_scsiio *csio, uint32_t *task_tagp, void **prvp)
1426 {
1427
1428         return (0);
1429 }
1430
1431 void
1432 icl_soft_conn_task_done(struct icl_conn *ic, void *prv)
1433 {
1434 }
1435
1436 int
1437 icl_soft_conn_transfer_setup(struct icl_conn *ic, union ctl_io *io,
1438     uint32_t *transfer_tag, void **prvp)
1439 {
1440
1441         return (0);
1442 }
1443
1444 void
1445 icl_soft_conn_transfer_done(struct icl_conn *ic, void *prv)
1446 {
1447 }
1448
1449 static int
1450 icl_soft_limits(struct icl_drv_limits *idl)
1451 {
1452
1453         idl->idl_max_recv_data_segment_length = 128 * 1024;
1454         idl->idl_max_send_data_segment_length = 128 * 1024;
1455         idl->idl_max_burst_length = 262144;
1456         idl->idl_first_burst_length = 65536;
1457
1458         return (0);
1459 }
1460
1461 #ifdef ICL_KERNEL_PROXY
1462 int
1463 icl_soft_conn_connect(struct icl_conn *ic, int domain, int socktype,
1464     int protocol, struct sockaddr *from_sa, struct sockaddr *to_sa)
1465 {
1466
1467         return (icl_soft_proxy_connect(ic, domain, socktype, protocol,
1468             from_sa, to_sa));
1469 }
1470
1471 int
1472 icl_soft_handoff_sock(struct icl_conn *ic, struct socket *so)
1473 {
1474         int error;
1475
1476         ICL_CONN_LOCK_ASSERT_NOT(ic);
1477
1478         if (so->so_type != SOCK_STREAM)
1479                 return (EINVAL);
1480
1481         ICL_CONN_LOCK(ic);
1482         if (ic->ic_socket != NULL) {
1483                 ICL_CONN_UNLOCK(ic);
1484                 return (EBUSY);
1485         }
1486         ic->ic_socket = so;
1487         ICL_CONN_UNLOCK(ic);
1488
1489         error = icl_conn_start(ic);
1490
1491         return (error);
1492 }
1493 #endif /* ICL_KERNEL_PROXY */
1494
1495 static int
1496 icl_soft_load(void)
1497 {
1498         int error;
1499
1500         icl_pdu_zone = uma_zcreate("icl_pdu",
1501             sizeof(struct icl_pdu), NULL, NULL, NULL, NULL,
1502             UMA_ALIGN_PTR, 0);
1503         refcount_init(&icl_ncons, 0);
1504
1505         /*
1506          * The reason we call this "none" is that to the user,
1507          * it's known as "offload driver"; "offload driver: soft"
1508          * doesn't make much sense.
1509          */
1510         error = icl_register("none", false, 0,
1511             icl_soft_limits, icl_soft_new_conn);
1512         KASSERT(error == 0, ("failed to register"));
1513
1514 #if defined(ICL_KERNEL_PROXY) && 0
1515         /*
1516          * Debugging aid for kernel proxy functionality.
1517          */
1518         error = icl_register("proxytest", true, 0,
1519             icl_soft_limits, icl_soft_new_conn);
1520         KASSERT(error == 0, ("failed to register"));
1521 #endif
1522
1523         return (error);
1524 }
1525
1526 static int
1527 icl_soft_unload(void)
1528 {
1529
1530         if (icl_ncons != 0)
1531                 return (EBUSY);
1532
1533         icl_unregister("none", false);
1534 #if defined(ICL_KERNEL_PROXY) && 0
1535         icl_unregister("proxytest", true);
1536 #endif
1537
1538         uma_zdestroy(icl_pdu_zone);
1539
1540         return (0);
1541 }
1542
1543 static int
1544 icl_soft_modevent(module_t mod, int what, void *arg)
1545 {
1546
1547         switch (what) {
1548         case MOD_LOAD:
1549                 return (icl_soft_load());
1550         case MOD_UNLOAD:
1551                 return (icl_soft_unload());
1552         default:
1553                 return (EINVAL);
1554         }
1555 }
1556
1557 moduledata_t icl_soft_data = {
1558         "icl_soft",
1559         icl_soft_modevent,
1560         0
1561 };
1562
1563 DECLARE_MODULE(icl_soft, icl_soft_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
1564 MODULE_DEPEND(icl_soft, icl, 1, 1, 1);
1565 MODULE_VERSION(icl_soft, 1);