sys/dev/iscsi/icl_soft.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2012 The FreeBSD Foundation
   5  *
   6  * This software was developed by Edward Tomasz Napierala under sponsorship
   7  * from the FreeBSD Foundation.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  * 1. Redistributions of source code must retain the above copyright
  13  *    notice, this list of conditions and the following disclaimer.
  14  * 2. Redistributions in binary form must reproduce the above copyright
  15  *    notice, this list of conditions and the following disclaimer in the
  16  *    documentation and/or other materials provided with the distribution.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  28  * SUCH DAMAGE.
  29  *
  30  */
  31
  32 /*
  33  * Software implementation of iSCSI Common Layer kobj(9) interface.
  34  */
  35
  36 #include <sys/cdefs.h>
  37 __FBSDID("$FreeBSD$");
  38
  39 #include <sys/param.h>
  40 #include <sys/capsicum.h>
  41 #include <sys/condvar.h>
  42 #include <sys/conf.h>
  43 #include <sys/gsb_crc32.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 #define ICL_CONN_STATE_BHS              1
  67 #define ICL_CONN_STATE_AHS              2
  68 #define ICL_CONN_STATE_HEADER_DIGEST    3
  69 #define ICL_CONN_STATE_DATA             4
  70 #define ICL_CONN_STATE_DATA_DIGEST      5
  71
  72 struct icl_soft_conn {
  73         struct icl_conn  ic;
  74
  75         /* soft specific stuff goes here. */
  76         STAILQ_HEAD(, icl_pdu) to_send;
  77         struct cv        send_cv;
  78         struct cv        receive_cv;
  79         struct icl_pdu  *receive_pdu;
  80         size_t           receive_len;
  81         int              receive_state;
  82         bool             receive_running;
  83         bool             check_send_space;
  84         bool             send_running;
  85 };
  86
  87 struct icl_soft_pdu {
  88         struct icl_pdu   ip;
  89
  90         /* soft specific stuff goes here. */
  91         u_int            ref_cnt;
  92         icl_pdu_cb       cb;
  93         int              error;
  94 };
  95
  96 SYSCTL_NODE(_kern_icl, OID_AUTO, soft, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
  97     "Software iSCSI");
  98 static int coalesce = 1;
  99 SYSCTL_INT(_kern_icl_soft, OID_AUTO, coalesce, CTLFLAG_RWTUN,
 100     &coalesce, 0, "Try to coalesce PDUs before sending");
 101 static int partial_receive_len = 256 * 1024;
 102 SYSCTL_INT(_kern_icl_soft, OID_AUTO, partial_receive_len, CTLFLAG_RWTUN,
 103     &partial_receive_len, 0, "Minimum read size for partially received "
 104     "data segment");
 105 static int max_data_segment_length = 256 * 1024;
 106 SYSCTL_INT(_kern_icl_soft, OID_AUTO, max_data_segment_length, CTLFLAG_RWTUN,
 107     &max_data_segment_length, 0, "Maximum data segment length");
 108 static int first_burst_length = 1024 * 1024;
 109 SYSCTL_INT(_kern_icl_soft, OID_AUTO, first_burst_length, CTLFLAG_RWTUN,
 110     &first_burst_length, 0, "First burst length");
 111 static int max_burst_length = 1024 * 1024;
 112 SYSCTL_INT(_kern_icl_soft, OID_AUTO, max_burst_length, CTLFLAG_RWTUN,
 113     &max_burst_length, 0, "Maximum burst length");
 114 static int sendspace = 1536 * 1024;
 115 SYSCTL_INT(_kern_icl_soft, OID_AUTO, sendspace, CTLFLAG_RWTUN,
 116     &sendspace, 0, "Default send socket buffer size");
 117 static int recvspace = 1536 * 1024;
 118 SYSCTL_INT(_kern_icl_soft, OID_AUTO, recvspace, CTLFLAG_RWTUN,
 119     &recvspace, 0, "Default receive socket buffer size");
 120
 121 static MALLOC_DEFINE(M_ICL_SOFT, "icl_soft", "iSCSI software backend");
 122 static uma_zone_t icl_soft_pdu_zone;
 123
 124 static volatile u_int   icl_ncons;
 125
 126 #define ICL_CONN_LOCK(X)                mtx_lock(X->ic_lock)
 127 #define ICL_CONN_UNLOCK(X)              mtx_unlock(X->ic_lock)
 128 #define ICL_CONN_LOCK_ASSERT(X)         mtx_assert(X->ic_lock, MA_OWNED)
 129 #define ICL_CONN_LOCK_ASSERT_NOT(X)     mtx_assert(X->ic_lock, MA_NOTOWNED)
 130
 131 STAILQ_HEAD(icl_pdu_stailq, icl_pdu);
 132
 133 static icl_conn_new_pdu_t       icl_soft_conn_new_pdu;
 134 static icl_conn_pdu_free_t      icl_soft_conn_pdu_free;
 135 static icl_conn_pdu_data_segment_length_t
 136                                     icl_soft_conn_pdu_data_segment_length;
 137 static icl_conn_pdu_append_data_t       icl_soft_conn_pdu_append_data;
 138 static icl_conn_pdu_get_data_t  icl_soft_conn_pdu_get_data;
 139 static icl_conn_pdu_queue_t     icl_soft_conn_pdu_queue;
 140 static icl_conn_pdu_queue_cb_t  icl_soft_conn_pdu_queue_cb;
 141 static icl_conn_handoff_t       icl_soft_conn_handoff;
 142 static icl_conn_free_t          icl_soft_conn_free;
 143 static icl_conn_close_t         icl_soft_conn_close;
 144 static icl_conn_task_setup_t    icl_soft_conn_task_setup;
 145 static icl_conn_task_done_t     icl_soft_conn_task_done;
 146 static icl_conn_transfer_setup_t        icl_soft_conn_transfer_setup;
 147 static icl_conn_transfer_done_t icl_soft_conn_transfer_done;
 148 #ifdef ICL_KERNEL_PROXY
 149 static icl_conn_connect_t       icl_soft_conn_connect;
 150 #endif
 151
 152 static kobj_method_t icl_soft_methods[] = {
 153         KOBJMETHOD(icl_conn_new_pdu, icl_soft_conn_new_pdu),
 154         KOBJMETHOD(icl_conn_pdu_free, icl_soft_conn_pdu_free),
 155         KOBJMETHOD(icl_conn_pdu_data_segment_length,
 156             icl_soft_conn_pdu_data_segment_length),
 157         KOBJMETHOD(icl_conn_pdu_append_data, icl_soft_conn_pdu_append_data),
 158         KOBJMETHOD(icl_conn_pdu_get_data, icl_soft_conn_pdu_get_data),
 159         KOBJMETHOD(icl_conn_pdu_queue, icl_soft_conn_pdu_queue),
 160         KOBJMETHOD(icl_conn_pdu_queue_cb, icl_soft_conn_pdu_queue_cb),
 161         KOBJMETHOD(icl_conn_handoff, icl_soft_conn_handoff),
 162         KOBJMETHOD(icl_conn_free, icl_soft_conn_free),
 163         KOBJMETHOD(icl_conn_close, icl_soft_conn_close),
 164         KOBJMETHOD(icl_conn_task_setup, icl_soft_conn_task_setup),
 165         KOBJMETHOD(icl_conn_task_done, icl_soft_conn_task_done),
 166         KOBJMETHOD(icl_conn_transfer_setup, icl_soft_conn_transfer_setup),
 167         KOBJMETHOD(icl_conn_transfer_done, icl_soft_conn_transfer_done),
 168 #ifdef ICL_KERNEL_PROXY
 169         KOBJMETHOD(icl_conn_connect, icl_soft_conn_connect),
 170 #endif
 171         { 0, 0 }
 172 };
 173
 174 DEFINE_CLASS(icl_soft, icl_soft_methods, sizeof(struct icl_soft_conn));
 175
 176 static void
 177 icl_conn_fail(struct icl_conn *ic)
 178 {
 179         if (ic->ic_socket == NULL)
 180                 return;
 181
 182         /*
 183          * XXX
 184          */
 185         ic->ic_socket->so_error = EDOOFUS;
 186         (ic->ic_error)(ic);
 187 }
 188
 189 static void
 190 icl_soft_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip)
 191 {
 192         struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip;
 193
 194         KASSERT(isp->ref_cnt == 0, ("freeing active PDU"));
 195         m_freem(ip->ip_bhs_mbuf);
 196         m_freem(ip->ip_ahs_mbuf);
 197         m_freem(ip->ip_data_mbuf);
 198         uma_zfree(icl_soft_pdu_zone, isp);
 199 #ifdef DIAGNOSTIC
 200         refcount_release(&ic->ic_outstanding_pdus);
 201 #endif
 202 }
 203
 204 static void
 205 icl_soft_pdu_call_cb(struct icl_pdu *ip)
 206 {
 207         struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip;
 208
 209         if (isp->cb != NULL)
 210                 isp->cb(ip, isp->error);
 211 #ifdef DIAGNOSTIC
 212         refcount_release(&ip->ip_conn->ic_outstanding_pdus);
 213 #endif
 214         uma_zfree(icl_soft_pdu_zone, isp);
 215 }
 216
 217 static void
 218 icl_soft_pdu_done(struct icl_pdu *ip, int error)
 219 {
 220         struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip;
 221
 222         if (error != 0)
 223                 isp->error = error;
 224
 225         m_freem(ip->ip_bhs_mbuf);
 226         ip->ip_bhs_mbuf = NULL;
 227         m_freem(ip->ip_ahs_mbuf);
 228         ip->ip_ahs_mbuf = NULL;
 229         m_freem(ip->ip_data_mbuf);
 230         ip->ip_data_mbuf = NULL;
 231
 232         if (atomic_fetchadd_int(&isp->ref_cnt, -1) == 1)
 233                 icl_soft_pdu_call_cb(ip);
 234 }
 235
 236 static void
 237 icl_soft_mbuf_done(struct mbuf *mb)
 238 {
 239         struct icl_soft_pdu *isp = (struct icl_soft_pdu *)mb->m_ext.ext_arg1;
 240
 241         icl_soft_pdu_call_cb(&isp->ip);
 242 }
 243
 244 /*
 245  * Allocate icl_pdu with empty BHS to fill up by the caller.
 246  */
 247 struct icl_pdu *
 248 icl_soft_conn_new_pdu(struct icl_conn *ic, int flags)
 249 {
 250         struct icl_soft_pdu *isp;
 251         struct icl_pdu *ip;
 252
 253 #ifdef DIAGNOSTIC
 254         refcount_acquire(&ic->ic_outstanding_pdus);
 255 #endif
 256         isp = uma_zalloc(icl_soft_pdu_zone, flags | M_ZERO);
 257         if (isp == NULL) {
 258                 ICL_WARN("failed to allocate soft PDU");
 259 #ifdef DIAGNOSTIC
 260                 refcount_release(&ic->ic_outstanding_pdus);
 261 #endif
 262                 return (NULL);
 263         }
 264         ip = &isp->ip;
 265         ip->ip_conn = ic;
 266
 267         CTASSERT(sizeof(struct iscsi_bhs) <= MHLEN);
 268         ip->ip_bhs_mbuf = m_gethdr(flags, MT_DATA);
 269         if (ip->ip_bhs_mbuf == NULL) {
 270                 ICL_WARN("failed to allocate BHS mbuf");
 271                 icl_soft_conn_pdu_free(ic, ip);
 272                 return (NULL);
 273         }
 274         ip->ip_bhs = mtod(ip->ip_bhs_mbuf, struct iscsi_bhs *);
 275         memset(ip->ip_bhs, 0, sizeof(struct iscsi_bhs));
 276         ip->ip_bhs_mbuf->m_len = sizeof(struct iscsi_bhs);
 277
 278         return (ip);
 279 }
 280
 281 static int
 282 icl_pdu_ahs_length(const struct icl_pdu *request)
 283 {
 284
 285         return (request->ip_bhs->bhs_total_ahs_len * 4);
 286 }
 287
 288 static size_t
 289 icl_pdu_data_segment_length(const struct icl_pdu *request)
 290 {
 291         uint32_t len = 0;
 292
 293         len += request->ip_bhs->bhs_data_segment_len[0];
 294         len <<= 8;
 295         len += request->ip_bhs->bhs_data_segment_len[1];
 296         len <<= 8;
 297         len += request->ip_bhs->bhs_data_segment_len[2];
 298
 299         return (len);
 300 }
 301
 302 size_t
 303 icl_soft_conn_pdu_data_segment_length(struct icl_conn *ic,
 304     const struct icl_pdu *request)
 305 {
 306
 307         return (icl_pdu_data_segment_length(request));
 308 }
 309
 310 static void
 311 icl_pdu_set_data_segment_length(struct icl_pdu *response, uint32_t len)
 312 {
 313
 314         response->ip_bhs->bhs_data_segment_len[2] = len;
 315         response->ip_bhs->bhs_data_segment_len[1] = len >> 8;
 316         response->ip_bhs->bhs_data_segment_len[0] = len >> 16;
 317 }
 318
 319 static size_t
 320 icl_pdu_padding(const struct icl_pdu *ip)
 321 {
 322
 323         if ((ip->ip_data_len % 4) != 0)
 324                 return (4 - (ip->ip_data_len % 4));
 325
 326         return (0);
 327 }
 328
 329 static size_t
 330 icl_pdu_size(const struct icl_pdu *response)
 331 {
 332         size_t len;
 333
 334         KASSERT(response->ip_ahs_len == 0, ("responding with AHS"));
 335
 336         len = sizeof(struct iscsi_bhs) + response->ip_data_len +
 337             icl_pdu_padding(response);
 338         if (response->ip_conn->ic_header_crc32c)
 339                 len += ISCSI_HEADER_DIGEST_SIZE;
 340         if (response->ip_data_len != 0 && response->ip_conn->ic_data_crc32c)
 341                 len += ISCSI_DATA_DIGEST_SIZE;
 342
 343         return (len);
 344 }
 345
 346 static void
 347 icl_soft_receive_buf(struct mbuf **r, size_t *rs, void *buf, size_t s)
 348 {
 349
 350         m_copydata(*r, 0, s, buf);
 351         m_adj(*r, s);
 352         while ((*r) != NULL && (*r)->m_len == 0)
 353                 *r = m_free(*r);
 354         *rs -= s;
 355 }
 356
 357 static void
 358 icl_pdu_receive_ahs(struct icl_pdu *request, struct mbuf **r, size_t *rs)
 359 {
 360
 361         request->ip_ahs_len = icl_pdu_ahs_length(request);
 362         if (request->ip_ahs_len == 0)
 363                 return;
 364
 365         request->ip_ahs_mbuf = *r;
 366         *r = m_split(request->ip_ahs_mbuf, request->ip_ahs_len, M_WAITOK);
 367         *rs -= request->ip_ahs_len;
 368 }
 369
 370 static uint32_t
 371 icl_mbuf_to_crc32c(const struct mbuf *m0)
 372 {
 373         uint32_t digest = 0xffffffff;
 374         const struct mbuf *m;
 375
 376         for (m = m0; m != NULL; m = m->m_next)
 377                 digest = calculate_crc32c(digest,
 378                     mtod(m, const void *), m->m_len);
 379
 380         digest = digest ^ 0xffffffff;
 381
 382         return (digest);
 383 }
 384
 385 static int
 386 icl_pdu_check_header_digest(struct icl_pdu *request, struct mbuf **r, size_t *rs)
 387 {
 388         uint32_t received_digest, valid_digest;
 389
 390         if (request->ip_conn->ic_header_crc32c == false)
 391                 return (0);
 392
 393         CTASSERT(sizeof(received_digest) == ISCSI_HEADER_DIGEST_SIZE);
 394         icl_soft_receive_buf(r, rs, &received_digest, ISCSI_HEADER_DIGEST_SIZE);
 395
 396         /* Temporary attach AHS to BHS to calculate header digest. */
 397         request->ip_bhs_mbuf->m_next = request->ip_ahs_mbuf;
 398         valid_digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
 399         request->ip_bhs_mbuf->m_next = NULL;
 400         if (received_digest != valid_digest) {
 401                 ICL_WARN("header digest check failed; got 0x%x, "
 402                     "should be 0x%x", received_digest, valid_digest);
 403                 return (-1);
 404         }
 405
 406         return (0);
 407 }
 408
 409 /*
 410  * Return the number of bytes that should be waiting in the receive socket
 411  * before icl_pdu_receive_data_segment() gets called.
 412  */
 413 static size_t
 414 icl_pdu_data_segment_receive_len(const struct icl_pdu *request)
 415 {
 416         size_t len;
 417
 418         len = icl_pdu_data_segment_length(request);
 419         if (len == 0)
 420                 return (0);
 421
 422         /*
 423          * Account for the parts of data segment already read from
 424          * the socket buffer.
 425          */
 426         KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
 427         len -= request->ip_data_len;
 428
 429         /*
 430          * Don't always wait for the full data segment to be delivered
 431          * to the socket; this might badly affect performance due to
 432          * TCP window scaling.
 433          */
 434         if (len > partial_receive_len) {
 435 #if 0
 436                 ICL_DEBUG("need %zd bytes of data, limiting to %zd",
 437                     len, partial_receive_len));
 438 #endif
 439                 len = partial_receive_len;
 440
 441                 return (len);
 442         }
 443
 444         /*
 445          * Account for padding.  Note that due to the way code is written,
 446          * the icl_pdu_receive_data_segment() must always receive padding
 447          * along with the last part of data segment, because it would be
 448          * impossible to tell whether we've already received the full data
 449          * segment including padding, or without it.
 450          */
 451         if ((len % 4) != 0)
 452                 len += 4 - (len % 4);
 453
 454 #if 0
 455         ICL_DEBUG("need %zd bytes of data", len));
 456 #endif
 457
 458         return (len);
 459 }
 460
 461 static int
 462 icl_pdu_receive_data_segment(struct icl_pdu *request, struct mbuf **r,
 463     size_t *rs, bool *more_neededp)
 464 {
 465         struct icl_soft_conn *isc;
 466         size_t len, padding = 0;
 467         struct mbuf *m;
 468
 469         isc = (struct icl_soft_conn *)request->ip_conn;
 470
 471         *more_neededp = false;
 472         isc->receive_len = 0;
 473
 474         len = icl_pdu_data_segment_length(request);
 475         if (len == 0)
 476                 return (0);
 477
 478         if ((len % 4) != 0)
 479                 padding = 4 - (len % 4);
 480
 481         /*
 482          * Account for already received parts of data segment.
 483          */
 484         KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
 485         len -= request->ip_data_len;
 486
 487         if (len + padding > *rs) {
 488                 /*
 489                  * Not enough data in the socket buffer.  Receive as much
 490                  * as we can.  Don't receive padding, since, obviously, it's
 491                  * not the end of data segment yet.
 492                  */
 493 #if 0
 494                 ICL_DEBUG("limited from %zd to %zd",
 495                     len + padding, *rs - padding));
 496 #endif
 497                 len = *rs - padding;
 498                 *more_neededp = true;
 499                 padding = 0;
 500         }
 501
 502         /*
 503          * Must not try to receive padding without at least one byte
 504          * of actual data segment.
 505          */
 506         if (len > 0) {
 507                 m = *r;
 508                 *r = m_split(m, len + padding, M_WAITOK);
 509                 *rs -= len + padding;
 510
 511                 if (request->ip_data_mbuf == NULL)
 512                         request->ip_data_mbuf = m;
 513                 else
 514                         m_cat(request->ip_data_mbuf, m);
 515
 516                 request->ip_data_len += len;
 517         } else
 518                 ICL_DEBUG("len 0");
 519
 520         if (*more_neededp)
 521                 isc->receive_len = icl_pdu_data_segment_receive_len(request);
 522
 523         return (0);
 524 }
 525
 526 static int
 527 icl_pdu_check_data_digest(struct icl_pdu *request, struct mbuf **r, size_t *rs)
 528 {
 529         uint32_t received_digest, valid_digest;
 530
 531         if (request->ip_conn->ic_data_crc32c == false)
 532                 return (0);
 533
 534         if (request->ip_data_len == 0)
 535                 return (0);
 536
 537         CTASSERT(sizeof(received_digest) == ISCSI_DATA_DIGEST_SIZE);
 538         icl_soft_receive_buf(r, rs, &received_digest, ISCSI_DATA_DIGEST_SIZE);
 539
 540         /*
 541          * Note that ip_data_mbuf also contains padding; since digest
 542          * calculation is supposed to include that, we iterate over
 543          * the entire ip_data_mbuf chain, not just ip_data_len bytes of it.
 544          */
 545         valid_digest = icl_mbuf_to_crc32c(request->ip_data_mbuf);
 546         if (received_digest != valid_digest) {
 547                 ICL_WARN("data digest check failed; got 0x%x, "
 548                     "should be 0x%x", received_digest, valid_digest);
 549                 return (-1);
 550         }
 551
 552         return (0);
 553 }
 554
 555 /*
 556  * Somewhat contrary to the name, this attempts to receive only one
 557  * "part" of PDU at a time; call it repeatedly until it returns non-NULL.
 558  */
 559 static struct icl_pdu *
 560 icl_conn_receive_pdu(struct icl_soft_conn *isc, struct mbuf **r, size_t *rs)
 561 {
 562         struct icl_conn *ic = &isc->ic;
 563         struct icl_pdu *request;
 564         size_t len;
 565         int error = 0;
 566         bool more_needed;
 567
 568         if (isc->receive_state == ICL_CONN_STATE_BHS) {
 569                 KASSERT(isc->receive_pdu == NULL,
 570                     ("isc->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                 isc->receive_pdu = request;
 579         } else {
 580                 KASSERT(isc->receive_pdu != NULL,
 581                     ("isc->receive_pdu == NULL"));
 582                 request = isc->receive_pdu;
 583         }
 584
 585         switch (isc->receive_state) {
 586         case ICL_CONN_STATE_BHS:
 587                 //ICL_DEBUG("receiving BHS");
 588                 icl_soft_receive_buf(r, rs, request->ip_bhs,
 589                     sizeof(struct iscsi_bhs));
 590
 591                 /*
 592                  * We don't enforce any limit for AHS length;
 593                  * its length is stored in 8 bit field.
 594                  */
 595
 596                 len = icl_pdu_data_segment_length(request);
 597                 if (len > ic->ic_max_recv_data_segment_length) {
 598                         ICL_WARN("received data segment "
 599                             "length %zd is larger than negotiated; "
 600                             "dropping connection", len);
 601                         error = EINVAL;
 602                         break;
 603                 }
 604
 605                 isc->receive_state = ICL_CONN_STATE_AHS;
 606                 isc->receive_len = icl_pdu_ahs_length(request);
 607                 break;
 608
 609         case ICL_CONN_STATE_AHS:
 610                 //ICL_DEBUG("receiving AHS");
 611                 icl_pdu_receive_ahs(request, r, rs);
 612                 isc->receive_state = ICL_CONN_STATE_HEADER_DIGEST;
 613                 if (ic->ic_header_crc32c == false)
 614                         isc->receive_len = 0;
 615                 else
 616                         isc->receive_len = ISCSI_HEADER_DIGEST_SIZE;
 617                 break;
 618
 619         case ICL_CONN_STATE_HEADER_DIGEST:
 620                 //ICL_DEBUG("receiving header digest");
 621                 error = icl_pdu_check_header_digest(request, r, rs);
 622                 if (error != 0) {
 623                         ICL_DEBUG("header digest failed; "
 624                             "dropping connection");
 625                         break;
 626                 }
 627
 628                 isc->receive_state = ICL_CONN_STATE_DATA;
 629                 isc->receive_len = icl_pdu_data_segment_receive_len(request);
 630                 break;
 631
 632         case ICL_CONN_STATE_DATA:
 633                 //ICL_DEBUG("receiving data segment");
 634                 error = icl_pdu_receive_data_segment(request, r, rs,
 635                     &more_needed);
 636                 if (error != 0) {
 637                         ICL_DEBUG("failed to receive data segment;"
 638                             "dropping connection");
 639                         break;
 640                 }
 641
 642                 if (more_needed)
 643                         break;
 644
 645                 isc->receive_state = ICL_CONN_STATE_DATA_DIGEST;
 646                 if (request->ip_data_len == 0 || ic->ic_data_crc32c == false)
 647                         isc->receive_len = 0;
 648                 else
 649                         isc->receive_len = ISCSI_DATA_DIGEST_SIZE;
 650                 break;
 651
 652         case ICL_CONN_STATE_DATA_DIGEST:
 653                 //ICL_DEBUG("receiving data digest");
 654                 error = icl_pdu_check_data_digest(request, r, rs);
 655                 if (error != 0) {
 656                         ICL_DEBUG("data digest failed; "
 657                             "dropping connection");
 658                         break;
 659                 }
 660
 661                 /*
 662                  * We've received complete PDU; reset the receive state machine
 663                  * and return the PDU.
 664                  */
 665                 isc->receive_state = ICL_CONN_STATE_BHS;
 666                 isc->receive_len = sizeof(struct iscsi_bhs);
 667                 isc->receive_pdu = NULL;
 668                 return (request);
 669
 670         default:
 671                 panic("invalid receive_state %d\n", isc->receive_state);
 672         }
 673
 674         if (error != 0) {
 675                 /*
 676                  * Don't free the PDU; it's pointed to by isc->receive_pdu
 677                  * and will get freed in icl_soft_conn_close().
 678                  */
 679                 icl_conn_fail(ic);
 680         }
 681
 682         return (NULL);
 683 }
 684
 685 static void
 686 icl_conn_receive_pdus(struct icl_soft_conn *isc, struct mbuf **r, size_t *rs)
 687 {
 688         struct icl_conn *ic = &isc->ic;
 689         struct icl_pdu *response;
 690
 691         for (;;) {
 692                 if (ic->ic_disconnecting)
 693                         return;
 694
 695                 /*
 696                  * Loop until we have a complete PDU or there is not enough
 697                  * data in the socket buffer.
 698                  */
 699                 if (*rs < isc->receive_len) {
 700 #if 0
 701                         ICL_DEBUG("not enough data; have %zd, need %zd",
 702                             *rs, isc->receive_len);
 703 #endif
 704                         return;
 705                 }
 706
 707                 response = icl_conn_receive_pdu(isc, r, rs);
 708                 if (response == NULL)
 709                         continue;
 710
 711                 if (response->ip_ahs_len > 0) {
 712                         ICL_WARN("received PDU with unsupported "
 713                             "AHS; opcode 0x%x; dropping connection",
 714                             response->ip_bhs->bhs_opcode);
 715                         icl_soft_conn_pdu_free(ic, response);
 716                         icl_conn_fail(ic);
 717                         return;
 718                 }
 719
 720                 (ic->ic_receive)(response);
 721         }
 722 }
 723
 724 static void
 725 icl_receive_thread(void *arg)
 726 {
 727         struct icl_soft_conn *isc = arg;
 728         struct icl_conn *ic = &isc->ic;
 729         size_t available, read = 0;
 730         struct socket *so;
 731         struct mbuf *m, *r = NULL;
 732         struct uio uio;
 733         int error, flags;
 734
 735         so = ic->ic_socket;
 736
 737         for (;;) {
 738                 SOCKBUF_LOCK(&so->so_rcv);
 739                 if (ic->ic_disconnecting) {
 740                         SOCKBUF_UNLOCK(&so->so_rcv);
 741                         break;
 742                 }
 743
 744                 /*
 745                  * Set the low watermark, to be checked by
 746                  * soreadable() in icl_soupcall_receive()
 747                  * to avoid unnecessary wakeups until there
 748                  * is enough data received to read the PDU.
 749                  */
 750                 available = sbavail(&so->so_rcv);
 751                 if (read + available < isc->receive_len) {
 752                         so->so_rcv.sb_lowat = isc->receive_len - read;
 753                         cv_wait(&isc->receive_cv, SOCKBUF_MTX(&so->so_rcv));
 754                         so->so_rcv.sb_lowat = so->so_rcv.sb_hiwat + 1;
 755                         available = sbavail(&so->so_rcv);
 756                 }
 757                 SOCKBUF_UNLOCK(&so->so_rcv);
 758
 759                 if (available == 0) {
 760                         if (so->so_error != 0) {
 761                                 ICL_DEBUG("connection error %d; "
 762                                     "dropping connection", so->so_error);
 763                                 icl_conn_fail(ic);
 764                                 break;
 765                         }
 766                         continue;
 767                 }
 768
 769                 memset(&uio, 0, sizeof(uio));
 770                 uio.uio_resid = available;
 771                 flags = MSG_DONTWAIT;
 772                 error = soreceive(so, NULL, &uio, &m, NULL, &flags);
 773                 if (error != 0) {
 774                         ICL_DEBUG("soreceive error %d", error);
 775                         break;
 776                 }
 777                 if (uio.uio_resid != 0) {
 778                         m_freem(m);
 779                         ICL_DEBUG("short read");
 780                         break;
 781                 }
 782                 if (r)
 783                         m_cat(r, m);
 784                 else
 785                         r = m;
 786                 read += available;
 787
 788                 icl_conn_receive_pdus(isc, &r, &read);
 789         }
 790
 791         if (r)
 792                 m_freem(r);
 793
 794         ICL_CONN_LOCK(ic);
 795         isc->receive_running = false;
 796         cv_signal(&isc->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_soft_conn *isc;
 805
 806         if (!soreadable(so))
 807                 return (SU_OK);
 808
 809         isc = arg;
 810         cv_signal(&isc->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_soft_conn *isc, struct icl_pdu_stailq *queue)
 871 {
 872         struct icl_conn *ic = &isc->ic;
 873         struct icl_pdu *request, *request2;
 874         struct mbuf *m;
 875         struct socket *so;
 876         long available, size, size2;
 877         int coalesced, error;
 878
 879         ICL_CONN_LOCK_ASSERT_NOT(ic);
 880
 881         so = ic->ic_socket;
 882
 883         SOCKBUF_LOCK(&so->so_snd);
 884         /*
 885          * Check how much space do we have for transmit.  We can't just
 886          * call sosend() and retry when we get EWOULDBLOCK or EMSGSIZE,
 887          * as it always frees the mbuf chain passed to it, even in case
 888          * of error.
 889          */
 890         available = sbspace(&so->so_snd);
 891         isc->check_send_space = false;
 892
 893         /*
 894          * Notify the socket upcall that we don't need wakeups
 895          * for the time being.
 896          */
 897         so->so_snd.sb_lowat = so->so_snd.sb_hiwat + 1;
 898         SOCKBUF_UNLOCK(&so->so_snd);
 899
 900         while (!STAILQ_EMPTY(queue)) {
 901                 request = STAILQ_FIRST(queue);
 902                 size = icl_pdu_size(request);
 903                 if (available < size) {
 904                         /*
 905                          * Set the low watermark, to be checked by
 906                          * sowriteable() in icl_soupcall_send()
 907                          * to avoid unnecessary wakeups until there
 908                          * is enough space for the PDU to fit.
 909                          */
 910                         SOCKBUF_LOCK(&so->so_snd);
 911                         available = sbspace(&so->so_snd);
 912                         if (available < size) {
 913 #if 1
 914                                 ICL_DEBUG("no space to send; "
 915                                     "have %ld, need %ld",
 916                                     available, size);
 917 #endif
 918                                 so->so_snd.sb_lowat = max(size,
 919                                     so->so_snd.sb_hiwat / 8);
 920                                 SOCKBUF_UNLOCK(&so->so_snd);
 921                                 return;
 922                         }
 923                         SOCKBUF_UNLOCK(&so->so_snd);
 924                 }
 925                 STAILQ_REMOVE_HEAD(queue, ip_next);
 926                 error = icl_pdu_finalize(request);
 927                 if (error != 0) {
 928                         ICL_DEBUG("failed to finalize PDU; "
 929                             "dropping connection");
 930                         icl_soft_pdu_done(request, EIO);
 931                         icl_conn_fail(ic);
 932                         return;
 933                 }
 934                 if (coalesce) {
 935                         m = request->ip_bhs_mbuf;
 936                         for (coalesced = 1; ; coalesced++) {
 937                                 request2 = STAILQ_FIRST(queue);
 938                                 if (request2 == NULL)
 939                                         break;
 940                                 size2 = icl_pdu_size(request2);
 941                                 if (available < size + size2)
 942                                         break;
 943                                 STAILQ_REMOVE_HEAD(queue, ip_next);
 944                                 error = icl_pdu_finalize(request2);
 945                                 if (error != 0) {
 946                                         ICL_DEBUG("failed to finalize PDU; "
 947                                             "dropping connection");
 948                                         icl_soft_pdu_done(request, EIO);
 949                                         icl_soft_pdu_done(request2, EIO);
 950                                         icl_conn_fail(ic);
 951                                         return;
 952                                 }
 953                                 while (m->m_next)
 954                                         m = m->m_next;
 955                                 m_cat(m, request2->ip_bhs_mbuf);
 956                                 request2->ip_bhs_mbuf = NULL;
 957                                 request->ip_bhs_mbuf->m_pkthdr.len += size2;
 958                                 size += size2;
 959                                 icl_soft_pdu_done(request2, 0);
 960                         }
 961 #if 0
 962                         if (coalesced > 1) {
 963                                 ICL_DEBUG("coalesced %d PDUs into %ld bytes",
 964                                     coalesced, size);
 965                         }
 966 #endif
 967                 }
 968                 available -= size;
 969                 error = sosend(so, NULL, NULL, request->ip_bhs_mbuf,
 970                     NULL, MSG_DONTWAIT, curthread);
 971                 request->ip_bhs_mbuf = NULL; /* Sosend consumes the mbuf. */
 972                 if (error != 0) {
 973                         ICL_DEBUG("failed to send PDU, error %d; "
 974                             "dropping connection", error);
 975                         icl_soft_pdu_done(request, error);
 976                         icl_conn_fail(ic);
 977                         return;
 978                 }
 979                 icl_soft_pdu_done(request, 0);
 980         }
 981 }
 982
 983 static void
 984 icl_send_thread(void *arg)
 985 {
 986         struct icl_soft_conn *isc;
 987         struct icl_conn *ic;
 988         struct icl_pdu_stailq queue;
 989
 990         isc = arg;
 991         ic = &isc->ic;
 992
 993         STAILQ_INIT(&queue);
 994
 995         ICL_CONN_LOCK(ic);
 996         for (;;) {
 997                 for (;;) {
 998                         /*
 999                          * Populate the local queue from the main one.
1000                          * This way the icl_conn_send_pdus() can go through
1001                          * all the queued PDUs without holding any locks.
1002                          */
1003                         if (STAILQ_EMPTY(&queue) || isc->check_send_space)
1004                                 STAILQ_CONCAT(&queue, &isc->to_send);
1005
1006                         ICL_CONN_UNLOCK(ic);
1007                         icl_conn_send_pdus(isc, &queue);
1008                         ICL_CONN_LOCK(ic);
1009
1010                         /*
1011                          * The icl_soupcall_send() was called since the last
1012                          * call to sbspace(); go around;
1013                          */
1014                         if (isc->check_send_space)
1015                                 continue;
1016
1017                         /*
1018                          * Local queue is empty, but we still have PDUs
1019                          * in the main one; go around.
1020                          */
1021                         if (STAILQ_EMPTY(&queue) &&
1022                             !STAILQ_EMPTY(&isc->to_send))
1023                                 continue;
1024
1025                         /*
1026                          * There might be some stuff in the local queue,
1027                          * which didn't get sent due to not having enough send
1028                          * space.  Wait for socket upcall.
1029                          */
1030                         break;
1031                 }
1032
1033                 if (ic->ic_disconnecting) {
1034                         //ICL_DEBUG("terminating");
1035                         break;
1036                 }
1037
1038                 cv_wait(&isc->send_cv, ic->ic_lock);
1039         }
1040
1041         /*
1042          * We're exiting; move PDUs back to the main queue, so they can
1043          * get freed properly.  At this point ordering doesn't matter.
1044          */
1045         STAILQ_CONCAT(&isc->to_send, &queue);
1046
1047         isc->send_running = false;
1048         cv_signal(&isc->send_cv);
1049         ICL_CONN_UNLOCK(ic);
1050         kthread_exit();
1051 }
1052
1053 static int
1054 icl_soupcall_send(struct socket *so, void *arg, int waitflag)
1055 {
1056         struct icl_soft_conn *isc;
1057         struct icl_conn *ic;
1058
1059         if (!sowriteable(so))
1060                 return (SU_OK);
1061
1062         isc = arg;
1063         ic = &isc->ic;
1064
1065         ICL_CONN_LOCK(ic);
1066         isc->check_send_space = true;
1067         ICL_CONN_UNLOCK(ic);
1068
1069         cv_signal(&isc->send_cv);
1070
1071         return (SU_OK);
1072 }
1073
1074 static int
1075 icl_soft_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *request,
1076     const void *addr, size_t len, int flags)
1077 {
1078         struct icl_soft_pdu *isp = (struct icl_soft_pdu *)request;
1079         struct mbuf *mb, *newmb;
1080         size_t copylen, off = 0;
1081
1082         KASSERT(len > 0, ("len == 0"));
1083
1084         if (flags & ICL_NOCOPY) {
1085                 newmb = m_get(flags & ~ICL_NOCOPY, MT_DATA);
1086                 if (newmb == NULL) {
1087                         ICL_WARN("failed to allocate mbuf");
1088                         return (ENOMEM);
1089                 }
1090
1091                 newmb->m_flags |= M_RDONLY;
1092                 m_extaddref(newmb, __DECONST(char *, addr), len, &isp->ref_cnt,
1093                     icl_soft_mbuf_done, isp, NULL);
1094                 newmb->m_len = len;
1095         } else {
1096                 newmb = m_getm2(NULL, len, flags, MT_DATA, 0);
1097                 if (newmb == NULL) {
1098                         ICL_WARN("failed to allocate mbuf for %zd bytes", len);
1099                         return (ENOMEM);
1100                 }
1101
1102                 for (mb = newmb; mb != NULL; mb = mb->m_next) {
1103                         copylen = min(M_TRAILINGSPACE(mb), len - off);
1104                         memcpy(mtod(mb, char *), (const char *)addr + off, copylen);
1105                         mb->m_len = copylen;
1106                         off += copylen;
1107                 }
1108                 KASSERT(off == len, ("%s: off != len", __func__));
1109         }
1110
1111         if (request->ip_data_mbuf == NULL) {
1112                 request->ip_data_mbuf = newmb;
1113                 request->ip_data_len = len;
1114         } else {
1115                 m_cat(request->ip_data_mbuf, newmb);
1116                 request->ip_data_len += len;
1117         }
1118
1119         return (0);
1120 }
1121
1122 void
1123 icl_soft_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip,
1124     size_t off, void *addr, size_t len)
1125 {
1126
1127         m_copydata(ip->ip_data_mbuf, off, len, addr);
1128 }
1129
1130 static void
1131 icl_soft_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip)
1132 {
1133
1134         icl_soft_conn_pdu_queue_cb(ic, ip, NULL);
1135 }
1136
1137 static void
1138 icl_soft_conn_pdu_queue_cb(struct icl_conn *ic, struct icl_pdu *ip,
1139     icl_pdu_cb cb)
1140 {
1141         struct icl_soft_conn *isc = (struct icl_soft_conn *)ic;
1142         struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip;
1143
1144         ICL_CONN_LOCK_ASSERT(ic);
1145         isp->ref_cnt++;
1146         isp->cb = cb;
1147
1148         if (ic->ic_disconnecting || ic->ic_socket == NULL) {
1149                 ICL_DEBUG("icl_pdu_queue on closed connection");
1150                 icl_soft_pdu_done(ip, ENOTCONN);
1151                 return;
1152         }
1153
1154         if (!STAILQ_EMPTY(&isc->to_send)) {
1155                 STAILQ_INSERT_TAIL(&isc->to_send, ip, ip_next);
1156                 /*
1157                  * If the queue is not empty, someone else had already
1158                  * signaled the send thread; no need to do that again,
1159                  * just return.
1160                  */
1161                 return;
1162         }
1163
1164         STAILQ_INSERT_TAIL(&isc->to_send, ip, ip_next);
1165         cv_signal(&isc->send_cv);
1166 }
1167
1168 static struct icl_conn *
1169 icl_soft_new_conn(const char *name, struct mtx *lock)
1170 {
1171         struct icl_soft_conn *isc;
1172         struct icl_conn *ic;
1173
1174         refcount_acquire(&icl_ncons);
1175
1176         isc = (struct icl_soft_conn *)kobj_create(&icl_soft_class, M_ICL_SOFT,
1177             M_WAITOK | M_ZERO);
1178
1179         STAILQ_INIT(&isc->to_send);
1180         cv_init(&isc->send_cv, "icl_tx");
1181         cv_init(&isc->receive_cv, "icl_rx");
1182
1183         ic = &isc->ic;
1184         ic->ic_lock = lock;
1185 #ifdef DIAGNOSTIC
1186         refcount_init(&ic->ic_outstanding_pdus, 0);
1187 #endif
1188         ic->ic_name = name;
1189         ic->ic_offload = "None";
1190         ic->ic_unmapped = false;
1191
1192         return (ic);
1193 }
1194
1195 void
1196 icl_soft_conn_free(struct icl_conn *ic)
1197 {
1198         struct icl_soft_conn *isc = (struct icl_soft_conn *)ic;
1199
1200 #ifdef DIAGNOSTIC
1201         KASSERT(ic->ic_outstanding_pdus == 0,
1202             ("destroying session with %d outstanding PDUs",
1203              ic->ic_outstanding_pdus));
1204 #endif
1205         cv_destroy(&isc->send_cv);
1206         cv_destroy(&isc->receive_cv);
1207         kobj_delete((struct kobj *)isc, M_ICL_SOFT);
1208         refcount_release(&icl_ncons);
1209 }
1210
1211 static int
1212 icl_conn_start(struct icl_conn *ic)
1213 {
1214         struct icl_soft_conn *isc = (struct icl_soft_conn *)ic;
1215         size_t minspace;
1216         struct sockopt opt;
1217         int error, one = 1;
1218
1219         ICL_CONN_LOCK(ic);
1220
1221         /*
1222          * XXX: Ugly hack.
1223          */
1224         if (ic->ic_socket == NULL) {
1225                 ICL_CONN_UNLOCK(ic);
1226                 return (EINVAL);
1227         }
1228
1229         isc->receive_state = ICL_CONN_STATE_BHS;
1230         isc->receive_len = sizeof(struct iscsi_bhs);
1231         ic->ic_disconnecting = false;
1232
1233         ICL_CONN_UNLOCK(ic);
1234
1235         /*
1236          * For sendspace, this is required because the current code cannot
1237          * send a PDU in pieces; thus, the minimum buffer size is equal
1238          * to the maximum PDU size.  "+4" is to account for possible padding.
1239          */
1240         minspace = sizeof(struct iscsi_bhs) +
1241             ic->ic_max_send_data_segment_length +
1242             ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE + 4;
1243         if (sendspace < minspace) {
1244                 ICL_WARN("kern.icl.sendspace too low; must be at least %zd",
1245                     minspace);
1246                 sendspace = minspace;
1247         }
1248         minspace = sizeof(struct iscsi_bhs) +
1249             ic->ic_max_recv_data_segment_length +
1250             ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE + 4;
1251         if (recvspace < minspace) {
1252                 ICL_WARN("kern.icl.recvspace too low; must be at least %zd",
1253                     minspace);
1254                 recvspace = minspace;
1255         }
1256
1257         error = soreserve(ic->ic_socket, sendspace, recvspace);
1258         if (error != 0) {
1259                 ICL_WARN("soreserve failed with error %d", error);
1260                 icl_soft_conn_close(ic);
1261                 return (error);
1262         }
1263         ic->ic_socket->so_snd.sb_flags |= SB_AUTOSIZE;
1264         ic->ic_socket->so_rcv.sb_flags |= SB_AUTOSIZE;
1265
1266         /*
1267          * Disable Nagle.
1268          */
1269         bzero(&opt, sizeof(opt));
1270         opt.sopt_dir = SOPT_SET;
1271         opt.sopt_level = IPPROTO_TCP;
1272         opt.sopt_name = TCP_NODELAY;
1273         opt.sopt_val = &one;
1274         opt.sopt_valsize = sizeof(one);
1275         error = sosetopt(ic->ic_socket, &opt);
1276         if (error != 0) {
1277                 ICL_WARN("disabling TCP_NODELAY failed with error %d", error);
1278                 icl_soft_conn_close(ic);
1279                 return (error);
1280         }
1281
1282         /*
1283          * Register socket upcall, to get notified about incoming PDUs
1284          * and free space to send outgoing ones.
1285          */
1286         SOCKBUF_LOCK(&ic->ic_socket->so_snd);
1287         soupcall_set(ic->ic_socket, SO_SND, icl_soupcall_send, isc);
1288         SOCKBUF_UNLOCK(&ic->ic_socket->so_snd);
1289         SOCKBUF_LOCK(&ic->ic_socket->so_rcv);
1290         soupcall_set(ic->ic_socket, SO_RCV, icl_soupcall_receive, isc);
1291         SOCKBUF_UNLOCK(&ic->ic_socket->so_rcv);
1292
1293         /*
1294          * Start threads.
1295          */
1296         ICL_CONN_LOCK(ic);
1297         isc->send_running = isc->receive_running = true;
1298         ICL_CONN_UNLOCK(ic);
1299         error = kthread_add(icl_send_thread, ic, NULL, NULL, 0, 0, "%stx",
1300             ic->ic_name);
1301         if (error != 0) {
1302                 ICL_WARN("kthread_add(9) failed with error %d", error);
1303                 ICL_CONN_LOCK(ic);
1304                 isc->send_running = isc->receive_running = false;
1305                 cv_signal(&isc->send_cv);
1306                 ICL_CONN_UNLOCK(ic);
1307                 icl_soft_conn_close(ic);
1308                 return (error);
1309         }
1310         error = kthread_add(icl_receive_thread, ic, NULL, NULL, 0, 0, "%srx",
1311             ic->ic_name);
1312         if (error != 0) {
1313                 ICL_WARN("kthread_add(9) failed with error %d", error);
1314                 ICL_CONN_LOCK(ic);
1315                 isc->receive_running = false;
1316                 cv_signal(&isc->send_cv);
1317                 ICL_CONN_UNLOCK(ic);
1318                 icl_soft_conn_close(ic);
1319                 return (error);
1320         }
1321
1322         return (0);
1323 }
1324
1325 int
1326 icl_soft_conn_handoff(struct icl_conn *ic, int fd)
1327 {
1328         struct file *fp;
1329         struct socket *so;
1330         cap_rights_t rights;
1331         int error;
1332
1333         ICL_CONN_LOCK_ASSERT_NOT(ic);
1334
1335 #ifdef ICL_KERNEL_PROXY
1336         /*
1337          * We're transitioning to Full Feature phase, and we don't
1338          * really care.
1339          */
1340         if (fd == 0) {
1341                 ICL_CONN_LOCK(ic);
1342                 if (ic->ic_socket == NULL) {
1343                         ICL_CONN_UNLOCK(ic);
1344                         ICL_WARN("proxy handoff without connect");
1345                         return (EINVAL);
1346                 }
1347                 ICL_CONN_UNLOCK(ic);
1348                 return (0);
1349         }
1350 #endif
1351
1352         /*
1353          * Steal the socket from userland.
1354          */
1355         error = fget(curthread, fd,
1356             cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp);
1357         if (error != 0)
1358                 return (error);
1359         if (fp->f_type != DTYPE_SOCKET) {
1360                 fdrop(fp, curthread);
1361                 return (EINVAL);
1362         }
1363         so = fp->f_data;
1364         if (so->so_type != SOCK_STREAM) {
1365                 fdrop(fp, curthread);
1366                 return (EINVAL);
1367         }
1368
1369         ICL_CONN_LOCK(ic);
1370
1371         if (ic->ic_socket != NULL) {
1372                 ICL_CONN_UNLOCK(ic);
1373                 fdrop(fp, curthread);
1374                 return (EBUSY);
1375         }
1376
1377         ic->ic_socket = fp->f_data;
1378         fp->f_ops = &badfileops;
1379         fp->f_data = NULL;
1380         fdrop(fp, curthread);
1381         ICL_CONN_UNLOCK(ic);
1382
1383         error = icl_conn_start(ic);
1384
1385         return (error);
1386 }
1387
1388 void
1389 icl_soft_conn_close(struct icl_conn *ic)
1390 {
1391         struct icl_soft_conn *isc = (struct icl_soft_conn *)ic;
1392         struct icl_pdu *pdu;
1393         struct socket *so;
1394
1395         /*
1396          * Wake up the threads, so they can properly terminate.
1397          * Receive thread sleeps on so->so_rcv lock, send on ic->ic_lock.
1398          */
1399         ICL_CONN_LOCK(ic);
1400         if (!ic->ic_disconnecting) {
1401                 so = ic->ic_socket;
1402                 if (so)
1403                         SOCKBUF_LOCK(&so->so_rcv);
1404                 ic->ic_disconnecting = true;
1405                 if (so)
1406                         SOCKBUF_UNLOCK(&so->so_rcv);
1407         }
1408         while (isc->receive_running || isc->send_running) {
1409                 cv_signal(&isc->receive_cv);
1410                 cv_signal(&isc->send_cv);
1411                 cv_wait(&isc->send_cv, ic->ic_lock);
1412         }
1413
1414         /* Some other thread could close the connection same time. */
1415         so = ic->ic_socket;
1416         if (so == NULL) {
1417                 ICL_CONN_UNLOCK(ic);
1418                 return;
1419         }
1420         ic->ic_socket = NULL;
1421
1422         /*
1423          * Deregister socket upcalls.
1424          */
1425         ICL_CONN_UNLOCK(ic);
1426         SOCKBUF_LOCK(&so->so_snd);
1427         if (so->so_snd.sb_upcall != NULL)
1428                 soupcall_clear(so, SO_SND);
1429         SOCKBUF_UNLOCK(&so->so_snd);
1430         SOCKBUF_LOCK(&so->so_rcv);
1431         if (so->so_rcv.sb_upcall != NULL)
1432                 soupcall_clear(so, SO_RCV);
1433         SOCKBUF_UNLOCK(&so->so_rcv);
1434         soclose(so);
1435         ICL_CONN_LOCK(ic);
1436
1437         if (isc->receive_pdu != NULL) {
1438                 //ICL_DEBUG("freeing partially received PDU");
1439                 icl_soft_conn_pdu_free(ic, isc->receive_pdu);
1440                 isc->receive_pdu = NULL;
1441         }
1442
1443         /*
1444          * Remove any outstanding PDUs from the send queue.
1445          */
1446         while (!STAILQ_EMPTY(&isc->to_send)) {
1447                 pdu = STAILQ_FIRST(&isc->to_send);
1448                 STAILQ_REMOVE_HEAD(&isc->to_send, ip_next);
1449                 icl_soft_pdu_done(pdu, ENOTCONN);
1450         }
1451
1452         KASSERT(STAILQ_EMPTY(&isc->to_send),
1453             ("destroying session with non-empty send queue"));
1454         ICL_CONN_UNLOCK(ic);
1455 }
1456
1457 int
1458 icl_soft_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip,
1459     struct ccb_scsiio *csio, uint32_t *task_tagp, void **prvp)
1460 {
1461
1462         return (0);
1463 }
1464
1465 void
1466 icl_soft_conn_task_done(struct icl_conn *ic, void *prv)
1467 {
1468 }
1469
1470 int
1471 icl_soft_conn_transfer_setup(struct icl_conn *ic, union ctl_io *io,
1472     uint32_t *transfer_tag, void **prvp)
1473 {
1474
1475         return (0);
1476 }
1477
1478 void
1479 icl_soft_conn_transfer_done(struct icl_conn *ic, void *prv)
1480 {
1481 }
1482
1483 static int
1484 icl_soft_limits(struct icl_drv_limits *idl)
1485 {
1486
1487         idl->idl_max_recv_data_segment_length = max_data_segment_length;
1488         idl->idl_max_send_data_segment_length = max_data_segment_length;
1489         idl->idl_max_burst_length = max_burst_length;
1490         idl->idl_first_burst_length = first_burst_length;
1491
1492         return (0);
1493 }
1494
1495 #ifdef ICL_KERNEL_PROXY
1496 int
1497 icl_soft_conn_connect(struct icl_conn *ic, int domain, int socktype,
1498     int protocol, struct sockaddr *from_sa, struct sockaddr *to_sa)
1499 {
1500
1501         return (icl_soft_proxy_connect(ic, domain, socktype, protocol,
1502             from_sa, to_sa));
1503 }
1504
1505 int
1506 icl_soft_handoff_sock(struct icl_conn *ic, struct socket *so)
1507 {
1508         int error;
1509
1510         ICL_CONN_LOCK_ASSERT_NOT(ic);
1511
1512         if (so->so_type != SOCK_STREAM)
1513                 return (EINVAL);
1514
1515         ICL_CONN_LOCK(ic);
1516         if (ic->ic_socket != NULL) {
1517                 ICL_CONN_UNLOCK(ic);
1518                 return (EBUSY);
1519         }
1520         ic->ic_socket = so;
1521         ICL_CONN_UNLOCK(ic);
1522
1523         error = icl_conn_start(ic);
1524
1525         return (error);
1526 }
1527 #endif /* ICL_KERNEL_PROXY */
1528
1529 static int
1530 icl_soft_load(void)
1531 {
1532         int error;
1533
1534         icl_soft_pdu_zone = uma_zcreate("icl_soft_pdu",
1535             sizeof(struct icl_soft_pdu), NULL, NULL, NULL, NULL,
1536             UMA_ALIGN_PTR, 0);
1537         refcount_init(&icl_ncons, 0);
1538
1539         /*
1540          * The reason we call this "none" is that to the user,
1541          * it's known as "offload driver"; "offload driver: soft"
1542          * doesn't make much sense.
1543          */
1544         error = icl_register("none", false, 0,
1545             icl_soft_limits, icl_soft_new_conn);
1546         KASSERT(error == 0, ("failed to register"));
1547
1548 #if defined(ICL_KERNEL_PROXY) && 0
1549         /*
1550          * Debugging aid for kernel proxy functionality.
1551          */
1552         error = icl_register("proxytest", true, 0,
1553             icl_soft_limits, icl_soft_new_conn);
1554         KASSERT(error == 0, ("failed to register"));
1555 #endif
1556
1557         return (error);
1558 }
1559
1560 static int
1561 icl_soft_unload(void)
1562 {
1563
1564         if (icl_ncons != 0)
1565                 return (EBUSY);
1566
1567         icl_unregister("none", false);
1568 #if defined(ICL_KERNEL_PROXY) && 0
1569         icl_unregister("proxytest", true);
1570 #endif
1571
1572         uma_zdestroy(icl_soft_pdu_zone);
1573
1574         return (0);
1575 }
1576
1577 static int
1578 icl_soft_modevent(module_t mod, int what, void *arg)
1579 {
1580
1581         switch (what) {
1582         case MOD_LOAD:
1583                 return (icl_soft_load());
1584         case MOD_UNLOAD:
1585                 return (icl_soft_unload());
1586         default:
1587                 return (EINVAL);
1588         }
1589 }
1590
1591 moduledata_t icl_soft_data = {
1592         "icl_soft",
1593         icl_soft_modevent,
1594         0
1595 };
1596
1597 DECLARE_MODULE(icl_soft, icl_soft_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
1598 MODULE_DEPEND(icl_soft, icl, 1, 1, 1);
1599 MODULE_VERSION(icl_soft, 1);