xref: /freebsd/sys/dev/iscsi/icl_soft.c (revision 8a272653d9fbd9fc37691c9aad6a05089b4ecb4d)
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 struct icl_soft_pdu {
67 	struct icl_pdu	 ip;
68 
69 	/* soft specific stuff goes here. */
70 	u_int		 ref_cnt;
71 	icl_pdu_cb	 cb;
72 	int		 error;
73 };
74 
75 SYSCTL_NODE(_kern_icl, OID_AUTO, soft, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
76     "Software iSCSI");
77 static int coalesce = 1;
78 SYSCTL_INT(_kern_icl_soft, OID_AUTO, coalesce, CTLFLAG_RWTUN,
79     &coalesce, 0, "Try to coalesce PDUs before sending");
80 static int partial_receive_len = 256 * 1024;
81 SYSCTL_INT(_kern_icl_soft, OID_AUTO, partial_receive_len, CTLFLAG_RWTUN,
82     &partial_receive_len, 0, "Minimum read size for partially received "
83     "data segment");
84 static int max_data_segment_length = 256 * 1024;
85 SYSCTL_INT(_kern_icl_soft, OID_AUTO, max_data_segment_length, CTLFLAG_RWTUN,
86     &max_data_segment_length, 0, "Maximum data segment length");
87 static int first_burst_length = 1024 * 1024;
88 SYSCTL_INT(_kern_icl_soft, OID_AUTO, first_burst_length, CTLFLAG_RWTUN,
89     &first_burst_length, 0, "First burst length");
90 static int max_burst_length = 1024 * 1024;
91 SYSCTL_INT(_kern_icl_soft, OID_AUTO, max_burst_length, CTLFLAG_RWTUN,
92     &max_burst_length, 0, "Maximum burst length");
93 static int sendspace = 1536 * 1024;
94 SYSCTL_INT(_kern_icl_soft, OID_AUTO, sendspace, CTLFLAG_RWTUN,
95     &sendspace, 0, "Default send socket buffer size");
96 static int recvspace = 1536 * 1024;
97 SYSCTL_INT(_kern_icl_soft, OID_AUTO, recvspace, CTLFLAG_RWTUN,
98     &recvspace, 0, "Default receive socket buffer size");
99 
100 static MALLOC_DEFINE(M_ICL_SOFT, "icl_soft", "iSCSI software backend");
101 static uma_zone_t icl_soft_pdu_zone;
102 
103 static volatile u_int	icl_ncons;
104 
105 #define ICL_CONN_LOCK(X)		mtx_lock(X->ic_lock)
106 #define ICL_CONN_UNLOCK(X)		mtx_unlock(X->ic_lock)
107 #define ICL_CONN_LOCK_ASSERT(X)		mtx_assert(X->ic_lock, MA_OWNED)
108 #define ICL_CONN_LOCK_ASSERT_NOT(X)	mtx_assert(X->ic_lock, MA_NOTOWNED)
109 
110 STAILQ_HEAD(icl_pdu_stailq, icl_pdu);
111 
112 static icl_conn_new_pdu_t	icl_soft_conn_new_pdu;
113 static icl_conn_pdu_free_t	icl_soft_conn_pdu_free;
114 static icl_conn_pdu_data_segment_length_t
115 				    icl_soft_conn_pdu_data_segment_length;
116 static icl_conn_pdu_append_data_t	icl_soft_conn_pdu_append_data;
117 static icl_conn_pdu_get_data_t	icl_soft_conn_pdu_get_data;
118 static icl_conn_pdu_queue_t	icl_soft_conn_pdu_queue;
119 static icl_conn_pdu_queue_cb_t	icl_soft_conn_pdu_queue_cb;
120 static icl_conn_handoff_t	icl_soft_conn_handoff;
121 static icl_conn_free_t		icl_soft_conn_free;
122 static icl_conn_close_t		icl_soft_conn_close;
123 static icl_conn_task_setup_t	icl_soft_conn_task_setup;
124 static icl_conn_task_done_t	icl_soft_conn_task_done;
125 static icl_conn_transfer_setup_t	icl_soft_conn_transfer_setup;
126 static icl_conn_transfer_done_t	icl_soft_conn_transfer_done;
127 #ifdef ICL_KERNEL_PROXY
128 static icl_conn_connect_t	icl_soft_conn_connect;
129 #endif
130 
131 static kobj_method_t icl_soft_methods[] = {
132 	KOBJMETHOD(icl_conn_new_pdu, icl_soft_conn_new_pdu),
133 	KOBJMETHOD(icl_conn_pdu_free, icl_soft_conn_pdu_free),
134 	KOBJMETHOD(icl_conn_pdu_data_segment_length,
135 	    icl_soft_conn_pdu_data_segment_length),
136 	KOBJMETHOD(icl_conn_pdu_append_data, icl_soft_conn_pdu_append_data),
137 	KOBJMETHOD(icl_conn_pdu_get_data, icl_soft_conn_pdu_get_data),
138 	KOBJMETHOD(icl_conn_pdu_queue, icl_soft_conn_pdu_queue),
139 	KOBJMETHOD(icl_conn_pdu_queue_cb, icl_soft_conn_pdu_queue_cb),
140 	KOBJMETHOD(icl_conn_handoff, icl_soft_conn_handoff),
141 	KOBJMETHOD(icl_conn_free, icl_soft_conn_free),
142 	KOBJMETHOD(icl_conn_close, icl_soft_conn_close),
143 	KOBJMETHOD(icl_conn_task_setup, icl_soft_conn_task_setup),
144 	KOBJMETHOD(icl_conn_task_done, icl_soft_conn_task_done),
145 	KOBJMETHOD(icl_conn_transfer_setup, icl_soft_conn_transfer_setup),
146 	KOBJMETHOD(icl_conn_transfer_done, icl_soft_conn_transfer_done),
147 #ifdef ICL_KERNEL_PROXY
148 	KOBJMETHOD(icl_conn_connect, icl_soft_conn_connect),
149 #endif
150 	{ 0, 0 }
151 };
152 
153 DEFINE_CLASS(icl_soft, icl_soft_methods, sizeof(struct icl_conn));
154 
155 static void
156 icl_conn_fail(struct icl_conn *ic)
157 {
158 	if (ic->ic_socket == NULL)
159 		return;
160 
161 	/*
162 	 * XXX
163 	 */
164 	ic->ic_socket->so_error = EDOOFUS;
165 	(ic->ic_error)(ic);
166 }
167 
168 static void
169 icl_soft_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip)
170 {
171 	struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip;
172 
173 	KASSERT(isp->ref_cnt == 0, ("freeing active PDU"));
174 	m_freem(ip->ip_bhs_mbuf);
175 	m_freem(ip->ip_ahs_mbuf);
176 	m_freem(ip->ip_data_mbuf);
177 	uma_zfree(icl_soft_pdu_zone, isp);
178 #ifdef DIAGNOSTIC
179 	refcount_release(&ic->ic_outstanding_pdus);
180 #endif
181 }
182 
183 static void
184 icl_soft_pdu_call_cb(struct icl_pdu *ip)
185 {
186 	struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip;
187 
188 	if (isp->cb != NULL)
189 		isp->cb(ip, isp->error);
190 #ifdef DIAGNOSTIC
191 	refcount_release(&ip->ip_conn->ic_outstanding_pdus);
192 #endif
193 	uma_zfree(icl_soft_pdu_zone, isp);
194 }
195 
196 static void
197 icl_soft_pdu_done(struct icl_pdu *ip, int error)
198 {
199 	struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip;
200 
201 	if (error != 0)
202 		isp->error = error;
203 
204 	m_freem(ip->ip_bhs_mbuf);
205 	ip->ip_bhs_mbuf = NULL;
206 	m_freem(ip->ip_ahs_mbuf);
207 	ip->ip_ahs_mbuf = NULL;
208 	m_freem(ip->ip_data_mbuf);
209 	ip->ip_data_mbuf = NULL;
210 
211 	if (atomic_fetchadd_int(&isp->ref_cnt, -1) == 1)
212 		icl_soft_pdu_call_cb(ip);
213 }
214 
215 static void
216 icl_soft_mbuf_done(struct mbuf *mb)
217 {
218 	struct icl_soft_pdu *isp = (struct icl_soft_pdu *)mb->m_ext.ext_arg1;
219 
220 	icl_soft_pdu_call_cb(&isp->ip);
221 }
222 
223 /*
224  * Allocate icl_pdu with empty BHS to fill up by the caller.
225  */
226 struct icl_pdu *
227 icl_soft_conn_new_pdu(struct icl_conn *ic, int flags)
228 {
229 	struct icl_soft_pdu *isp;
230 	struct icl_pdu *ip;
231 
232 #ifdef DIAGNOSTIC
233 	refcount_acquire(&ic->ic_outstanding_pdus);
234 #endif
235 	isp = uma_zalloc(icl_soft_pdu_zone, flags | M_ZERO);
236 	if (isp == NULL) {
237 		ICL_WARN("failed to allocate soft PDU");
238 #ifdef DIAGNOSTIC
239 		refcount_release(&ic->ic_outstanding_pdus);
240 #endif
241 		return (NULL);
242 	}
243 	ip = &isp->ip;
244 	ip->ip_conn = ic;
245 
246 	CTASSERT(sizeof(struct iscsi_bhs) <= MHLEN);
247 	ip->ip_bhs_mbuf = m_gethdr(flags, MT_DATA);
248 	if (ip->ip_bhs_mbuf == NULL) {
249 		ICL_WARN("failed to allocate BHS mbuf");
250 		icl_soft_conn_pdu_free(ic, ip);
251 		return (NULL);
252 	}
253 	ip->ip_bhs = mtod(ip->ip_bhs_mbuf, struct iscsi_bhs *);
254 	memset(ip->ip_bhs, 0, sizeof(struct iscsi_bhs));
255 	ip->ip_bhs_mbuf->m_len = sizeof(struct iscsi_bhs);
256 
257 	return (ip);
258 }
259 
260 static int
261 icl_pdu_ahs_length(const struct icl_pdu *request)
262 {
263 
264 	return (request->ip_bhs->bhs_total_ahs_len * 4);
265 }
266 
267 static size_t
268 icl_pdu_data_segment_length(const struct icl_pdu *request)
269 {
270 	uint32_t len = 0;
271 
272 	len += request->ip_bhs->bhs_data_segment_len[0];
273 	len <<= 8;
274 	len += request->ip_bhs->bhs_data_segment_len[1];
275 	len <<= 8;
276 	len += request->ip_bhs->bhs_data_segment_len[2];
277 
278 	return (len);
279 }
280 
281 size_t
282 icl_soft_conn_pdu_data_segment_length(struct icl_conn *ic,
283     const struct icl_pdu *request)
284 {
285 
286 	return (icl_pdu_data_segment_length(request));
287 }
288 
289 static void
290 icl_pdu_set_data_segment_length(struct icl_pdu *response, uint32_t len)
291 {
292 
293 	response->ip_bhs->bhs_data_segment_len[2] = len;
294 	response->ip_bhs->bhs_data_segment_len[1] = len >> 8;
295 	response->ip_bhs->bhs_data_segment_len[0] = len >> 16;
296 }
297 
298 static size_t
299 icl_pdu_padding(const struct icl_pdu *ip)
300 {
301 
302 	if ((ip->ip_data_len % 4) != 0)
303 		return (4 - (ip->ip_data_len % 4));
304 
305 	return (0);
306 }
307 
308 static size_t
309 icl_pdu_size(const struct icl_pdu *response)
310 {
311 	size_t len;
312 
313 	KASSERT(response->ip_ahs_len == 0, ("responding with AHS"));
314 
315 	len = sizeof(struct iscsi_bhs) + response->ip_data_len +
316 	    icl_pdu_padding(response);
317 	if (response->ip_conn->ic_header_crc32c)
318 		len += ISCSI_HEADER_DIGEST_SIZE;
319 	if (response->ip_data_len != 0 && response->ip_conn->ic_data_crc32c)
320 		len += ISCSI_DATA_DIGEST_SIZE;
321 
322 	return (len);
323 }
324 
325 static void
326 icl_soft_receive_buf(struct mbuf **r, size_t *rs, void *buf, size_t s)
327 {
328 
329 	m_copydata(*r, 0, s, buf);
330 	m_adj(*r, s);
331 	while ((*r) != NULL && (*r)->m_len == 0)
332 		*r = m_free(*r);
333 	*rs -= s;
334 }
335 
336 static void
337 icl_pdu_receive_ahs(struct icl_pdu *request, struct mbuf **r, size_t *rs)
338 {
339 
340 	request->ip_ahs_len = icl_pdu_ahs_length(request);
341 	if (request->ip_ahs_len == 0)
342 		return;
343 
344 	request->ip_ahs_mbuf = *r;
345 	*r = m_split(request->ip_ahs_mbuf, request->ip_ahs_len, M_WAITOK);
346 	*rs -= request->ip_ahs_len;
347 }
348 
349 static uint32_t
350 icl_mbuf_to_crc32c(const struct mbuf *m0)
351 {
352 	uint32_t digest = 0xffffffff;
353 	const struct mbuf *m;
354 
355 	for (m = m0; m != NULL; m = m->m_next)
356 		digest = calculate_crc32c(digest,
357 		    mtod(m, const void *), m->m_len);
358 
359 	digest = digest ^ 0xffffffff;
360 
361 	return (digest);
362 }
363 
364 static int
365 icl_pdu_check_header_digest(struct icl_pdu *request, struct mbuf **r, size_t *rs)
366 {
367 	uint32_t received_digest, valid_digest;
368 
369 	if (request->ip_conn->ic_header_crc32c == false)
370 		return (0);
371 
372 	CTASSERT(sizeof(received_digest) == ISCSI_HEADER_DIGEST_SIZE);
373 	icl_soft_receive_buf(r, rs, &received_digest, ISCSI_HEADER_DIGEST_SIZE);
374 
375 	/* Temporary attach AHS to BHS to calculate header digest. */
376 	request->ip_bhs_mbuf->m_next = request->ip_ahs_mbuf;
377 	valid_digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
378 	request->ip_bhs_mbuf->m_next = NULL;
379 	if (received_digest != valid_digest) {
380 		ICL_WARN("header digest check failed; got 0x%x, "
381 		    "should be 0x%x", received_digest, valid_digest);
382 		return (-1);
383 	}
384 
385 	return (0);
386 }
387 
388 /*
389  * Return the number of bytes that should be waiting in the receive socket
390  * before icl_pdu_receive_data_segment() gets called.
391  */
392 static size_t
393 icl_pdu_data_segment_receive_len(const struct icl_pdu *request)
394 {
395 	size_t len;
396 
397 	len = icl_pdu_data_segment_length(request);
398 	if (len == 0)
399 		return (0);
400 
401 	/*
402 	 * Account for the parts of data segment already read from
403 	 * the socket buffer.
404 	 */
405 	KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
406 	len -= request->ip_data_len;
407 
408 	/*
409 	 * Don't always wait for the full data segment to be delivered
410 	 * to the socket; this might badly affect performance due to
411 	 * TCP window scaling.
412 	 */
413 	if (len > partial_receive_len) {
414 #if 0
415 		ICL_DEBUG("need %zd bytes of data, limiting to %zd",
416 		    len, partial_receive_len));
417 #endif
418 		len = partial_receive_len;
419 
420 		return (len);
421 	}
422 
423 	/*
424 	 * Account for padding.  Note that due to the way code is written,
425 	 * the icl_pdu_receive_data_segment() must always receive padding
426 	 * along with the last part of data segment, because it would be
427 	 * impossible to tell whether we've already received the full data
428 	 * segment including padding, or without it.
429 	 */
430 	if ((len % 4) != 0)
431 		len += 4 - (len % 4);
432 
433 #if 0
434 	ICL_DEBUG("need %zd bytes of data", len));
435 #endif
436 
437 	return (len);
438 }
439 
440 static int
441 icl_pdu_receive_data_segment(struct icl_pdu *request, struct mbuf **r,
442     size_t *rs, bool *more_neededp)
443 {
444 	struct icl_conn *ic;
445 	size_t len, padding = 0;
446 	struct mbuf *m;
447 
448 	ic = request->ip_conn;
449 
450 	*more_neededp = false;
451 	ic->ic_receive_len = 0;
452 
453 	len = icl_pdu_data_segment_length(request);
454 	if (len == 0)
455 		return (0);
456 
457 	if ((len % 4) != 0)
458 		padding = 4 - (len % 4);
459 
460 	/*
461 	 * Account for already received parts of data segment.
462 	 */
463 	KASSERT(len > request->ip_data_len, ("len <= request->ip_data_len"));
464 	len -= request->ip_data_len;
465 
466 	if (len + padding > *rs) {
467 		/*
468 		 * Not enough data in the socket buffer.  Receive as much
469 		 * as we can.  Don't receive padding, since, obviously, it's
470 		 * not the end of data segment yet.
471 		 */
472 #if 0
473 		ICL_DEBUG("limited from %zd to %zd",
474 		    len + padding, *rs - padding));
475 #endif
476 		len = *rs - padding;
477 		*more_neededp = true;
478 		padding = 0;
479 	}
480 
481 	/*
482 	 * Must not try to receive padding without at least one byte
483 	 * of actual data segment.
484 	 */
485 	if (len > 0) {
486 		m = *r;
487 		*r = m_split(m, len + padding, M_WAITOK);
488 		*rs -= len + padding;
489 
490 		if (request->ip_data_mbuf == NULL)
491 			request->ip_data_mbuf = m;
492 		else
493 			m_cat(request->ip_data_mbuf, m);
494 
495 		request->ip_data_len += len;
496 	} else
497 		ICL_DEBUG("len 0");
498 
499 	if (*more_neededp)
500 		ic->ic_receive_len =
501 		    icl_pdu_data_segment_receive_len(request);
502 
503 	return (0);
504 }
505 
506 static int
507 icl_pdu_check_data_digest(struct icl_pdu *request, struct mbuf **r, size_t *rs)
508 {
509 	uint32_t received_digest, valid_digest;
510 
511 	if (request->ip_conn->ic_data_crc32c == false)
512 		return (0);
513 
514 	if (request->ip_data_len == 0)
515 		return (0);
516 
517 	CTASSERT(sizeof(received_digest) == ISCSI_DATA_DIGEST_SIZE);
518 	icl_soft_receive_buf(r, rs, &received_digest, ISCSI_DATA_DIGEST_SIZE);
519 
520 	/*
521 	 * Note that ip_data_mbuf also contains padding; since digest
522 	 * calculation is supposed to include that, we iterate over
523 	 * the entire ip_data_mbuf chain, not just ip_data_len bytes of it.
524 	 */
525 	valid_digest = icl_mbuf_to_crc32c(request->ip_data_mbuf);
526 	if (received_digest != valid_digest) {
527 		ICL_WARN("data digest check failed; got 0x%x, "
528 		    "should be 0x%x", received_digest, valid_digest);
529 		return (-1);
530 	}
531 
532 	return (0);
533 }
534 
535 /*
536  * Somewhat contrary to the name, this attempts to receive only one
537  * "part" of PDU at a time; call it repeatedly until it returns non-NULL.
538  */
539 static struct icl_pdu *
540 icl_conn_receive_pdu(struct icl_conn *ic, struct mbuf **r, size_t *rs)
541 {
542 	struct icl_pdu *request;
543 	size_t len;
544 	int error = 0;
545 	bool more_needed;
546 
547 	if (ic->ic_receive_state == ICL_CONN_STATE_BHS) {
548 		KASSERT(ic->ic_receive_pdu == NULL,
549 		    ("ic->ic_receive_pdu != NULL"));
550 		request = icl_soft_conn_new_pdu(ic, M_NOWAIT);
551 		if (request == NULL) {
552 			ICL_DEBUG("failed to allocate PDU; "
553 			    "dropping connection");
554 			icl_conn_fail(ic);
555 			return (NULL);
556 		}
557 		ic->ic_receive_pdu = request;
558 	} else {
559 		KASSERT(ic->ic_receive_pdu != NULL,
560 		    ("ic->ic_receive_pdu == NULL"));
561 		request = ic->ic_receive_pdu;
562 	}
563 
564 	switch (ic->ic_receive_state) {
565 	case ICL_CONN_STATE_BHS:
566 		//ICL_DEBUG("receiving BHS");
567 		icl_soft_receive_buf(r, rs, request->ip_bhs,
568 		    sizeof(struct iscsi_bhs));
569 
570 		/*
571 		 * We don't enforce any limit for AHS length;
572 		 * its length is stored in 8 bit field.
573 		 */
574 
575 		len = icl_pdu_data_segment_length(request);
576 		if (len > ic->ic_max_data_segment_length) {
577 			ICL_WARN("received data segment "
578 			    "length %zd is larger than negotiated; "
579 			    "dropping connection", len);
580 			error = EINVAL;
581 			break;
582 		}
583 
584 		ic->ic_receive_state = ICL_CONN_STATE_AHS;
585 		ic->ic_receive_len = icl_pdu_ahs_length(request);
586 		break;
587 
588 	case ICL_CONN_STATE_AHS:
589 		//ICL_DEBUG("receiving AHS");
590 		icl_pdu_receive_ahs(request, r, rs);
591 		ic->ic_receive_state = ICL_CONN_STATE_HEADER_DIGEST;
592 		if (ic->ic_header_crc32c == false)
593 			ic->ic_receive_len = 0;
594 		else
595 			ic->ic_receive_len = ISCSI_HEADER_DIGEST_SIZE;
596 		break;
597 
598 	case ICL_CONN_STATE_HEADER_DIGEST:
599 		//ICL_DEBUG("receiving header digest");
600 		error = icl_pdu_check_header_digest(request, r, rs);
601 		if (error != 0) {
602 			ICL_DEBUG("header digest failed; "
603 			    "dropping connection");
604 			break;
605 		}
606 
607 		ic->ic_receive_state = ICL_CONN_STATE_DATA;
608 		ic->ic_receive_len =
609 		    icl_pdu_data_segment_receive_len(request);
610 		break;
611 
612 	case ICL_CONN_STATE_DATA:
613 		//ICL_DEBUG("receiving data segment");
614 		error = icl_pdu_receive_data_segment(request, r, rs,
615 		    &more_needed);
616 		if (error != 0) {
617 			ICL_DEBUG("failed to receive data segment;"
618 			    "dropping connection");
619 			break;
620 		}
621 
622 		if (more_needed)
623 			break;
624 
625 		ic->ic_receive_state = ICL_CONN_STATE_DATA_DIGEST;
626 		if (request->ip_data_len == 0 || ic->ic_data_crc32c == false)
627 			ic->ic_receive_len = 0;
628 		else
629 			ic->ic_receive_len = ISCSI_DATA_DIGEST_SIZE;
630 		break;
631 
632 	case ICL_CONN_STATE_DATA_DIGEST:
633 		//ICL_DEBUG("receiving data digest");
634 		error = icl_pdu_check_data_digest(request, r, rs);
635 		if (error != 0) {
636 			ICL_DEBUG("data digest failed; "
637 			    "dropping connection");
638 			break;
639 		}
640 
641 		/*
642 		 * We've received complete PDU; reset the receive state machine
643 		 * and return the PDU.
644 		 */
645 		ic->ic_receive_state = ICL_CONN_STATE_BHS;
646 		ic->ic_receive_len = sizeof(struct iscsi_bhs);
647 		ic->ic_receive_pdu = NULL;
648 		return (request);
649 
650 	default:
651 		panic("invalid ic_receive_state %d\n", ic->ic_receive_state);
652 	}
653 
654 	if (error != 0) {
655 		/*
656 		 * Don't free the PDU; it's pointed to by ic->ic_receive_pdu
657 		 * and will get freed in icl_soft_conn_close().
658 		 */
659 		icl_conn_fail(ic);
660 	}
661 
662 	return (NULL);
663 }
664 
665 static void
666 icl_conn_receive_pdus(struct icl_conn *ic, struct mbuf **r, size_t *rs)
667 {
668 	struct icl_pdu *response;
669 
670 	for (;;) {
671 		if (ic->ic_disconnecting)
672 			return;
673 
674 		/*
675 		 * Loop until we have a complete PDU or there is not enough
676 		 * data in the socket buffer.
677 		 */
678 		if (*rs < ic->ic_receive_len) {
679 #if 0
680 			ICL_DEBUG("not enough data; have %zd, need %zd",
681 			    *rs, ic->ic_receive_len);
682 #endif
683 			return;
684 		}
685 
686 		response = icl_conn_receive_pdu(ic, r, rs);
687 		if (response == NULL)
688 			continue;
689 
690 		if (response->ip_ahs_len > 0) {
691 			ICL_WARN("received PDU with unsupported "
692 			    "AHS; opcode 0x%x; dropping connection",
693 			    response->ip_bhs->bhs_opcode);
694 			icl_soft_conn_pdu_free(ic, response);
695 			icl_conn_fail(ic);
696 			return;
697 		}
698 
699 		(ic->ic_receive)(response);
700 	}
701 }
702 
703 static void
704 icl_receive_thread(void *arg)
705 {
706 	struct icl_conn *ic;
707 	size_t available, read = 0;
708 	struct socket *so;
709 	struct mbuf *m, *r = NULL;
710 	struct uio uio;
711 	int error, flags;
712 
713 	ic = arg;
714 	so = ic->ic_socket;
715 
716 	for (;;) {
717 		SOCKBUF_LOCK(&so->so_rcv);
718 		if (ic->ic_disconnecting) {
719 			SOCKBUF_UNLOCK(&so->so_rcv);
720 			break;
721 		}
722 
723 		/*
724 		 * Set the low watermark, to be checked by
725 		 * soreadable() in icl_soupcall_receive()
726 		 * to avoid unnecessary wakeups until there
727 		 * is enough data received to read the PDU.
728 		 */
729 		available = sbavail(&so->so_rcv);
730 		if (read + available < ic->ic_receive_len) {
731 			so->so_rcv.sb_lowat = ic->ic_receive_len - read;
732 			cv_wait(&ic->ic_receive_cv, &so->so_rcv.sb_mtx);
733 			so->so_rcv.sb_lowat = so->so_rcv.sb_hiwat + 1;
734 			available = sbavail(&so->so_rcv);
735 		}
736 		SOCKBUF_UNLOCK(&so->so_rcv);
737 
738 		if (available == 0) {
739 			if (so->so_error != 0) {
740 				ICL_DEBUG("connection error %d; "
741 				    "dropping connection", so->so_error);
742 				icl_conn_fail(ic);
743 				break;
744 			}
745 			continue;
746 		}
747 
748 		memset(&uio, 0, sizeof(uio));
749 		uio.uio_resid = available;
750 		flags = MSG_DONTWAIT;
751 		error = soreceive(so, NULL, &uio, &m, NULL, &flags);
752 		if (error != 0) {
753 			ICL_DEBUG("soreceive error %d", error);
754 			break;
755 		}
756 		if (uio.uio_resid != 0) {
757 			m_freem(m);
758 			ICL_DEBUG("short read");
759 			break;
760 		}
761 		if (r)
762 			m_cat(r, m);
763 		else
764 			r = m;
765 		read += available;
766 
767 		icl_conn_receive_pdus(ic, &r, &read);
768 	}
769 
770 	if (r)
771 		m_freem(r);
772 
773 	ICL_CONN_LOCK(ic);
774 	ic->ic_receive_running = false;
775 	cv_signal(&ic->ic_send_cv);
776 	ICL_CONN_UNLOCK(ic);
777 	kthread_exit();
778 }
779 
780 static int
781 icl_soupcall_receive(struct socket *so, void *arg, int waitflag)
782 {
783 	struct icl_conn *ic;
784 
785 	if (!soreadable(so))
786 		return (SU_OK);
787 
788 	ic = arg;
789 	cv_signal(&ic->ic_receive_cv);
790 	return (SU_OK);
791 }
792 
793 static int
794 icl_pdu_finalize(struct icl_pdu *request)
795 {
796 	size_t padding, pdu_len;
797 	uint32_t digest, zero = 0;
798 	int ok;
799 	struct icl_conn *ic;
800 
801 	ic = request->ip_conn;
802 
803 	icl_pdu_set_data_segment_length(request, request->ip_data_len);
804 
805 	pdu_len = icl_pdu_size(request);
806 
807 	if (ic->ic_header_crc32c) {
808 		digest = icl_mbuf_to_crc32c(request->ip_bhs_mbuf);
809 		ok = m_append(request->ip_bhs_mbuf, sizeof(digest),
810 		    (void *)&digest);
811 		if (ok != 1) {
812 			ICL_WARN("failed to append header digest");
813 			return (1);
814 		}
815 	}
816 
817 	if (request->ip_data_len != 0) {
818 		padding = icl_pdu_padding(request);
819 		if (padding > 0) {
820 			ok = m_append(request->ip_data_mbuf, padding,
821 			    (void *)&zero);
822 			if (ok != 1) {
823 				ICL_WARN("failed to append padding");
824 				return (1);
825 			}
826 		}
827 
828 		if (ic->ic_data_crc32c) {
829 			digest = icl_mbuf_to_crc32c(request->ip_data_mbuf);
830 
831 			ok = m_append(request->ip_data_mbuf, sizeof(digest),
832 			    (void *)&digest);
833 			if (ok != 1) {
834 				ICL_WARN("failed to append data digest");
835 				return (1);
836 			}
837 		}
838 
839 		m_cat(request->ip_bhs_mbuf, request->ip_data_mbuf);
840 		request->ip_data_mbuf = NULL;
841 	}
842 
843 	request->ip_bhs_mbuf->m_pkthdr.len = pdu_len;
844 
845 	return (0);
846 }
847 
848 static void
849 icl_conn_send_pdus(struct icl_conn *ic, struct icl_pdu_stailq *queue)
850 {
851 	struct icl_pdu *request, *request2;
852 	struct mbuf *m;
853 	struct socket *so;
854 	long available, size, size2;
855 	int coalesced, error;
856 
857 	ICL_CONN_LOCK_ASSERT_NOT(ic);
858 
859 	so = ic->ic_socket;
860 
861 	SOCKBUF_LOCK(&so->so_snd);
862 	/*
863 	 * Check how much space do we have for transmit.  We can't just
864 	 * call sosend() and retry when we get EWOULDBLOCK or EMSGSIZE,
865 	 * as it always frees the mbuf chain passed to it, even in case
866 	 * of error.
867 	 */
868 	available = sbspace(&so->so_snd);
869 
870 	/*
871 	 * Notify the socket upcall that we don't need wakeups
872 	 * for the time being.
873 	 */
874 	so->so_snd.sb_lowat = so->so_snd.sb_hiwat + 1;
875 	SOCKBUF_UNLOCK(&so->so_snd);
876 
877 	while (!STAILQ_EMPTY(queue)) {
878 		request = STAILQ_FIRST(queue);
879 		size = icl_pdu_size(request);
880 		if (available < size) {
881 			/*
882 			 * Set the low watermark, to be checked by
883 			 * sowriteable() in icl_soupcall_send()
884 			 * to avoid unnecessary wakeups until there
885 			 * is enough space for the PDU to fit.
886 			 */
887 			SOCKBUF_LOCK(&so->so_snd);
888 			available = sbspace(&so->so_snd);
889 			if (available < size) {
890 #if 1
891 				ICL_DEBUG("no space to send; "
892 				    "have %ld, need %ld",
893 				    available, size);
894 #endif
895 				so->so_snd.sb_lowat = max(size,
896 				    so->so_snd.sb_hiwat / 8);
897 				SOCKBUF_UNLOCK(&so->so_snd);
898 				return;
899 			}
900 			SOCKBUF_UNLOCK(&so->so_snd);
901 		}
902 		STAILQ_REMOVE_HEAD(queue, ip_next);
903 		error = icl_pdu_finalize(request);
904 		if (error != 0) {
905 			ICL_DEBUG("failed to finalize PDU; "
906 			    "dropping connection");
907 			icl_soft_pdu_done(request, EIO);
908 			icl_conn_fail(ic);
909 			return;
910 		}
911 		if (coalesce) {
912 			m = request->ip_bhs_mbuf;
913 			for (coalesced = 1; ; coalesced++) {
914 				request2 = STAILQ_FIRST(queue);
915 				if (request2 == NULL)
916 					break;
917 				size2 = icl_pdu_size(request2);
918 				if (available < size + size2)
919 					break;
920 				STAILQ_REMOVE_HEAD(queue, ip_next);
921 				error = icl_pdu_finalize(request2);
922 				if (error != 0) {
923 					ICL_DEBUG("failed to finalize PDU; "
924 					    "dropping connection");
925 					icl_soft_pdu_done(request, EIO);
926 					icl_soft_pdu_done(request2, EIO);
927 					icl_conn_fail(ic);
928 					return;
929 				}
930 				while (m->m_next)
931 					m = m->m_next;
932 				m_cat(m, request2->ip_bhs_mbuf);
933 				request2->ip_bhs_mbuf = NULL;
934 				request->ip_bhs_mbuf->m_pkthdr.len += size2;
935 				size += size2;
936 				icl_soft_pdu_done(request2, 0);
937 			}
938 #if 0
939 			if (coalesced > 1) {
940 				ICL_DEBUG("coalesced %d PDUs into %ld bytes",
941 				    coalesced, size);
942 			}
943 #endif
944 		}
945 		available -= size;
946 		error = sosend(so, NULL, NULL, request->ip_bhs_mbuf,
947 		    NULL, MSG_DONTWAIT, curthread);
948 		request->ip_bhs_mbuf = NULL; /* Sosend consumes the mbuf. */
949 		if (error != 0) {
950 			ICL_DEBUG("failed to send PDU, error %d; "
951 			    "dropping connection", error);
952 			icl_soft_pdu_done(request, error);
953 			icl_conn_fail(ic);
954 			return;
955 		}
956 		icl_soft_pdu_done(request, 0);
957 	}
958 }
959 
960 static void
961 icl_send_thread(void *arg)
962 {
963 	struct icl_conn *ic;
964 	struct icl_pdu_stailq queue;
965 
966 	ic = arg;
967 
968 	STAILQ_INIT(&queue);
969 
970 	ICL_CONN_LOCK(ic);
971 	for (;;) {
972 		for (;;) {
973 			/*
974 			 * Populate the local queue from the main one.
975 			 * This way the icl_conn_send_pdus() can go through
976 			 * all the queued PDUs without holding any locks.
977 			 */
978 			STAILQ_CONCAT(&queue, &ic->ic_to_send);
979 
980 			ic->ic_check_send_space = false;
981 			ICL_CONN_UNLOCK(ic);
982 			icl_conn_send_pdus(ic, &queue);
983 			ICL_CONN_LOCK(ic);
984 
985 			/*
986 			 * The icl_soupcall_send() was called since the last
987 			 * call to sbspace(); go around;
988 			 */
989 			if (ic->ic_check_send_space)
990 				continue;
991 
992 			/*
993 			 * Local queue is empty, but we still have PDUs
994 			 * in the main one; go around.
995 			 */
996 			if (STAILQ_EMPTY(&queue) &&
997 			    !STAILQ_EMPTY(&ic->ic_to_send))
998 				continue;
999 
1000 			/*
1001 			 * There might be some stuff in the local queue,
1002 			 * which didn't get sent due to not having enough send
1003 			 * space.  Wait for socket upcall.
1004 			 */
1005 			break;
1006 		}
1007 
1008 		if (ic->ic_disconnecting) {
1009 			//ICL_DEBUG("terminating");
1010 			break;
1011 		}
1012 
1013 		cv_wait(&ic->ic_send_cv, ic->ic_lock);
1014 	}
1015 
1016 	/*
1017 	 * We're exiting; move PDUs back to the main queue, so they can
1018 	 * get freed properly.  At this point ordering doesn't matter.
1019 	 */
1020 	STAILQ_CONCAT(&ic->ic_to_send, &queue);
1021 
1022 	ic->ic_send_running = false;
1023 	cv_signal(&ic->ic_send_cv);
1024 	ICL_CONN_UNLOCK(ic);
1025 	kthread_exit();
1026 }
1027 
1028 static int
1029 icl_soupcall_send(struct socket *so, void *arg, int waitflag)
1030 {
1031 	struct icl_conn *ic;
1032 
1033 	if (!sowriteable(so))
1034 		return (SU_OK);
1035 
1036 	ic = arg;
1037 
1038 	ICL_CONN_LOCK(ic);
1039 	ic->ic_check_send_space = true;
1040 	ICL_CONN_UNLOCK(ic);
1041 
1042 	cv_signal(&ic->ic_send_cv);
1043 
1044 	return (SU_OK);
1045 }
1046 
1047 static int
1048 icl_soft_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *request,
1049     const void *addr, size_t len, int flags)
1050 {
1051 	struct icl_soft_pdu *isp = (struct icl_soft_pdu *)request;
1052 	struct mbuf *mb, *newmb;
1053 	size_t copylen, off = 0;
1054 
1055 	KASSERT(len > 0, ("len == 0"));
1056 
1057 	if (flags & ICL_NOCOPY) {
1058 		newmb = m_get(flags & ~ICL_NOCOPY, MT_DATA);
1059 		if (newmb == NULL) {
1060 			ICL_WARN("failed to allocate mbuf");
1061 			return (ENOMEM);
1062 		}
1063 
1064 		newmb->m_flags |= M_RDONLY;
1065 		m_extaddref(newmb, __DECONST(char *, addr), len, &isp->ref_cnt,
1066 		    icl_soft_mbuf_done, isp, NULL);
1067 		newmb->m_len = len;
1068 	} else {
1069 		newmb = m_getm2(NULL, len, flags, MT_DATA, 0);
1070 		if (newmb == NULL) {
1071 			ICL_WARN("failed to allocate mbuf for %zd bytes", len);
1072 			return (ENOMEM);
1073 		}
1074 
1075 		for (mb = newmb; mb != NULL; mb = mb->m_next) {
1076 			copylen = min(M_TRAILINGSPACE(mb), len - off);
1077 			memcpy(mtod(mb, char *), (const char *)addr + off, copylen);
1078 			mb->m_len = copylen;
1079 			off += copylen;
1080 		}
1081 		KASSERT(off == len, ("%s: off != len", __func__));
1082 	}
1083 
1084 	if (request->ip_data_mbuf == NULL) {
1085 		request->ip_data_mbuf = newmb;
1086 		request->ip_data_len = len;
1087 	} else {
1088 		m_cat(request->ip_data_mbuf, newmb);
1089 		request->ip_data_len += len;
1090 	}
1091 
1092 	return (0);
1093 }
1094 
1095 void
1096 icl_soft_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip,
1097     size_t off, void *addr, size_t len)
1098 {
1099 
1100 	m_copydata(ip->ip_data_mbuf, off, len, addr);
1101 }
1102 
1103 static void
1104 icl_soft_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip)
1105 {
1106 
1107 	icl_soft_conn_pdu_queue_cb(ic, ip, NULL);
1108 }
1109 
1110 static void
1111 icl_soft_conn_pdu_queue_cb(struct icl_conn *ic, struct icl_pdu *ip,
1112     icl_pdu_cb cb)
1113 {
1114 	struct icl_soft_pdu *isp = (struct icl_soft_pdu *)ip;
1115 
1116 	ICL_CONN_LOCK_ASSERT(ic);
1117 	isp->ref_cnt++;
1118 	isp->cb = cb;
1119 
1120 	if (ic->ic_disconnecting || ic->ic_socket == NULL) {
1121 		ICL_DEBUG("icl_pdu_queue on closed connection");
1122 		icl_soft_pdu_done(ip, ENOTCONN);
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 static struct icl_conn *
1141 icl_soft_new_conn(const char *name, struct mtx *lock)
1142 {
1143 	struct icl_conn *ic;
1144 
1145 	refcount_acquire(&icl_ncons);
1146 
1147 	ic = (struct icl_conn *)kobj_create(&icl_soft_class, M_ICL_SOFT, M_WAITOK | M_ZERO);
1148 
1149 	STAILQ_INIT(&ic->ic_to_send);
1150 	ic->ic_lock = lock;
1151 	cv_init(&ic->ic_send_cv, "icl_tx");
1152 	cv_init(&ic->ic_receive_cv, "icl_rx");
1153 #ifdef DIAGNOSTIC
1154 	refcount_init(&ic->ic_outstanding_pdus, 0);
1155 #endif
1156 	ic->ic_max_data_segment_length = max_data_segment_length;
1157 	ic->ic_name = name;
1158 	ic->ic_offload = "None";
1159 	ic->ic_unmapped = false;
1160 
1161 	return (ic);
1162 }
1163 
1164 void
1165 icl_soft_conn_free(struct icl_conn *ic)
1166 {
1167 
1168 #ifdef DIAGNOSTIC
1169 	KASSERT(ic->ic_outstanding_pdus == 0,
1170 	    ("destroying session with %d outstanding PDUs",
1171 	     ic->ic_outstanding_pdus));
1172 #endif
1173 	cv_destroy(&ic->ic_send_cv);
1174 	cv_destroy(&ic->ic_receive_cv);
1175 	kobj_delete((struct kobj *)ic, M_ICL_SOFT);
1176 	refcount_release(&icl_ncons);
1177 }
1178 
1179 static int
1180 icl_conn_start(struct icl_conn *ic)
1181 {
1182 	size_t minspace;
1183 	struct sockopt opt;
1184 	int error, one = 1;
1185 
1186 	ICL_CONN_LOCK(ic);
1187 
1188 	/*
1189 	 * XXX: Ugly hack.
1190 	 */
1191 	if (ic->ic_socket == NULL) {
1192 		ICL_CONN_UNLOCK(ic);
1193 		return (EINVAL);
1194 	}
1195 
1196 	ic->ic_receive_state = ICL_CONN_STATE_BHS;
1197 	ic->ic_receive_len = sizeof(struct iscsi_bhs);
1198 	ic->ic_disconnecting = false;
1199 
1200 	ICL_CONN_UNLOCK(ic);
1201 
1202 	/*
1203 	 * For sendspace, this is required because the current code cannot
1204 	 * send a PDU in pieces; thus, the minimum buffer size is equal
1205 	 * to the maximum PDU size.  "+4" is to account for possible padding.
1206 	 */
1207 	minspace = sizeof(struct iscsi_bhs) + ic->ic_max_data_segment_length +
1208 	    ISCSI_HEADER_DIGEST_SIZE + ISCSI_DATA_DIGEST_SIZE + 4;
1209 	if (sendspace < minspace) {
1210 		ICL_WARN("kern.icl.sendspace too low; must be at least %zd",
1211 		    minspace);
1212 		sendspace = minspace;
1213 	}
1214 	if (recvspace < minspace) {
1215 		ICL_WARN("kern.icl.recvspace too low; must be at least %zd",
1216 		    minspace);
1217 		recvspace = minspace;
1218 	}
1219 
1220 	error = soreserve(ic->ic_socket, sendspace, recvspace);
1221 	if (error != 0) {
1222 		ICL_WARN("soreserve failed with error %d", error);
1223 		icl_soft_conn_close(ic);
1224 		return (error);
1225 	}
1226 	ic->ic_socket->so_snd.sb_flags |= SB_AUTOSIZE;
1227 	ic->ic_socket->so_rcv.sb_flags |= SB_AUTOSIZE;
1228 
1229 	/*
1230 	 * Disable Nagle.
1231 	 */
1232 	bzero(&opt, sizeof(opt));
1233 	opt.sopt_dir = SOPT_SET;
1234 	opt.sopt_level = IPPROTO_TCP;
1235 	opt.sopt_name = TCP_NODELAY;
1236 	opt.sopt_val = &one;
1237 	opt.sopt_valsize = sizeof(one);
1238 	error = sosetopt(ic->ic_socket, &opt);
1239 	if (error != 0) {
1240 		ICL_WARN("disabling TCP_NODELAY failed with error %d", error);
1241 		icl_soft_conn_close(ic);
1242 		return (error);
1243 	}
1244 
1245 	/*
1246 	 * Register socket upcall, to get notified about incoming PDUs
1247 	 * and free space to send outgoing ones.
1248 	 */
1249 	SOCKBUF_LOCK(&ic->ic_socket->so_snd);
1250 	soupcall_set(ic->ic_socket, SO_SND, icl_soupcall_send, ic);
1251 	SOCKBUF_UNLOCK(&ic->ic_socket->so_snd);
1252 	SOCKBUF_LOCK(&ic->ic_socket->so_rcv);
1253 	soupcall_set(ic->ic_socket, SO_RCV, icl_soupcall_receive, ic);
1254 	SOCKBUF_UNLOCK(&ic->ic_socket->so_rcv);
1255 
1256 	/*
1257 	 * Start threads.
1258 	 */
1259 	ICL_CONN_LOCK(ic);
1260 	ic->ic_send_running = ic->ic_receive_running = true;
1261 	ICL_CONN_UNLOCK(ic);
1262 	error = kthread_add(icl_send_thread, ic, NULL, NULL, 0, 0, "%stx",
1263 	    ic->ic_name);
1264 	if (error != 0) {
1265 		ICL_WARN("kthread_add(9) failed with error %d", error);
1266 		ICL_CONN_LOCK(ic);
1267 		ic->ic_send_running = ic->ic_receive_running = false;
1268 		cv_signal(&ic->ic_send_cv);
1269 		ICL_CONN_UNLOCK(ic);
1270 		icl_soft_conn_close(ic);
1271 		return (error);
1272 	}
1273 	error = kthread_add(icl_receive_thread, ic, NULL, NULL, 0, 0, "%srx",
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_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 
1285 	return (0);
1286 }
1287 
1288 int
1289 icl_soft_conn_handoff(struct icl_conn *ic, int fd)
1290 {
1291 	struct file *fp;
1292 	struct socket *so;
1293 	cap_rights_t rights;
1294 	int error;
1295 
1296 	ICL_CONN_LOCK_ASSERT_NOT(ic);
1297 
1298 #ifdef ICL_KERNEL_PROXY
1299 	/*
1300 	 * We're transitioning to Full Feature phase, and we don't
1301 	 * really care.
1302 	 */
1303 	if (fd == 0) {
1304 		ICL_CONN_LOCK(ic);
1305 		if (ic->ic_socket == NULL) {
1306 			ICL_CONN_UNLOCK(ic);
1307 			ICL_WARN("proxy handoff without connect");
1308 			return (EINVAL);
1309 		}
1310 		ICL_CONN_UNLOCK(ic);
1311 		return (0);
1312 	}
1313 #endif
1314 
1315 	/*
1316 	 * Steal the socket from userland.
1317 	 */
1318 	error = fget(curthread, fd,
1319 	    cap_rights_init_one(&rights, CAP_SOCK_CLIENT), &fp);
1320 	if (error != 0)
1321 		return (error);
1322 	if (fp->f_type != DTYPE_SOCKET) {
1323 		fdrop(fp, curthread);
1324 		return (EINVAL);
1325 	}
1326 	so = fp->f_data;
1327 	if (so->so_type != SOCK_STREAM) {
1328 		fdrop(fp, curthread);
1329 		return (EINVAL);
1330 	}
1331 
1332 	ICL_CONN_LOCK(ic);
1333 
1334 	if (ic->ic_socket != NULL) {
1335 		ICL_CONN_UNLOCK(ic);
1336 		fdrop(fp, curthread);
1337 		return (EBUSY);
1338 	}
1339 
1340 	ic->ic_socket = fp->f_data;
1341 	fp->f_ops = &badfileops;
1342 	fp->f_data = NULL;
1343 	fdrop(fp, curthread);
1344 	ICL_CONN_UNLOCK(ic);
1345 
1346 	error = icl_conn_start(ic);
1347 
1348 	return (error);
1349 }
1350 
1351 void
1352 icl_soft_conn_close(struct icl_conn *ic)
1353 {
1354 	struct icl_pdu *pdu;
1355 	struct socket *so;
1356 
1357 	/*
1358 	 * Wake up the threads, so they can properly terminate.
1359 	 * Receive thread sleeps on so->so_rcv lock, send on ic->ic_lock.
1360 	 */
1361 	ICL_CONN_LOCK(ic);
1362 	if (!ic->ic_disconnecting) {
1363 		so = ic->ic_socket;
1364 		if (so)
1365 			SOCKBUF_LOCK(&so->so_rcv);
1366 		ic->ic_disconnecting = true;
1367 		if (so)
1368 			SOCKBUF_UNLOCK(&so->so_rcv);
1369 	}
1370 	while (ic->ic_receive_running || ic->ic_send_running) {
1371 		cv_signal(&ic->ic_receive_cv);
1372 		cv_signal(&ic->ic_send_cv);
1373 		cv_wait(&ic->ic_send_cv, ic->ic_lock);
1374 	}
1375 
1376 	/* Some other thread could close the connection same time. */
1377 	so = ic->ic_socket;
1378 	if (so == NULL) {
1379 		ICL_CONN_UNLOCK(ic);
1380 		return;
1381 	}
1382 	ic->ic_socket = NULL;
1383 
1384 	/*
1385 	 * Deregister socket upcalls.
1386 	 */
1387 	ICL_CONN_UNLOCK(ic);
1388 	SOCKBUF_LOCK(&so->so_snd);
1389 	if (so->so_snd.sb_upcall != NULL)
1390 		soupcall_clear(so, SO_SND);
1391 	SOCKBUF_UNLOCK(&so->so_snd);
1392 	SOCKBUF_LOCK(&so->so_rcv);
1393 	if (so->so_rcv.sb_upcall != NULL)
1394 		soupcall_clear(so, SO_RCV);
1395 	SOCKBUF_UNLOCK(&so->so_rcv);
1396 	soclose(so);
1397 	ICL_CONN_LOCK(ic);
1398 
1399 	if (ic->ic_receive_pdu != NULL) {
1400 		//ICL_DEBUG("freeing partially received PDU");
1401 		icl_soft_conn_pdu_free(ic, ic->ic_receive_pdu);
1402 		ic->ic_receive_pdu = NULL;
1403 	}
1404 
1405 	/*
1406 	 * Remove any outstanding PDUs from the send queue.
1407 	 */
1408 	while (!STAILQ_EMPTY(&ic->ic_to_send)) {
1409 		pdu = STAILQ_FIRST(&ic->ic_to_send);
1410 		STAILQ_REMOVE_HEAD(&ic->ic_to_send, ip_next);
1411 		icl_soft_pdu_done(pdu, ENOTCONN);
1412 	}
1413 
1414 	KASSERT(STAILQ_EMPTY(&ic->ic_to_send),
1415 	    ("destroying session with non-empty send queue"));
1416 	ICL_CONN_UNLOCK(ic);
1417 }
1418 
1419 int
1420 icl_soft_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip,
1421     struct ccb_scsiio *csio, uint32_t *task_tagp, void **prvp)
1422 {
1423 
1424 	return (0);
1425 }
1426 
1427 void
1428 icl_soft_conn_task_done(struct icl_conn *ic, void *prv)
1429 {
1430 }
1431 
1432 int
1433 icl_soft_conn_transfer_setup(struct icl_conn *ic, union ctl_io *io,
1434     uint32_t *transfer_tag, void **prvp)
1435 {
1436 
1437 	return (0);
1438 }
1439 
1440 void
1441 icl_soft_conn_transfer_done(struct icl_conn *ic, void *prv)
1442 {
1443 }
1444 
1445 static int
1446 icl_soft_limits(struct icl_drv_limits *idl)
1447 {
1448 
1449 	idl->idl_max_recv_data_segment_length = max_data_segment_length;
1450 	idl->idl_max_send_data_segment_length = max_data_segment_length;
1451 	idl->idl_max_burst_length = max_burst_length;
1452 	idl->idl_first_burst_length = first_burst_length;
1453 
1454 	return (0);
1455 }
1456 
1457 #ifdef ICL_KERNEL_PROXY
1458 int
1459 icl_soft_conn_connect(struct icl_conn *ic, int domain, int socktype,
1460     int protocol, struct sockaddr *from_sa, struct sockaddr *to_sa)
1461 {
1462 
1463 	return (icl_soft_proxy_connect(ic, domain, socktype, protocol,
1464 	    from_sa, to_sa));
1465 }
1466 
1467 int
1468 icl_soft_handoff_sock(struct icl_conn *ic, struct socket *so)
1469 {
1470 	int error;
1471 
1472 	ICL_CONN_LOCK_ASSERT_NOT(ic);
1473 
1474 	if (so->so_type != SOCK_STREAM)
1475 		return (EINVAL);
1476 
1477 	ICL_CONN_LOCK(ic);
1478 	if (ic->ic_socket != NULL) {
1479 		ICL_CONN_UNLOCK(ic);
1480 		return (EBUSY);
1481 	}
1482 	ic->ic_socket = so;
1483 	ICL_CONN_UNLOCK(ic);
1484 
1485 	error = icl_conn_start(ic);
1486 
1487 	return (error);
1488 }
1489 #endif /* ICL_KERNEL_PROXY */
1490 
1491 static int
1492 icl_soft_load(void)
1493 {
1494 	int error;
1495 
1496 	icl_soft_pdu_zone = uma_zcreate("icl_soft_pdu",
1497 	    sizeof(struct icl_soft_pdu), NULL, NULL, NULL, NULL,
1498 	    UMA_ALIGN_PTR, 0);
1499 	refcount_init(&icl_ncons, 0);
1500 
1501 	/*
1502 	 * The reason we call this "none" is that to the user,
1503 	 * it's known as "offload driver"; "offload driver: soft"
1504 	 * doesn't make much sense.
1505 	 */
1506 	error = icl_register("none", false, 0,
1507 	    icl_soft_limits, icl_soft_new_conn);
1508 	KASSERT(error == 0, ("failed to register"));
1509 
1510 #if defined(ICL_KERNEL_PROXY) && 0
1511 	/*
1512 	 * Debugging aid for kernel proxy functionality.
1513 	 */
1514 	error = icl_register("proxytest", true, 0,
1515 	    icl_soft_limits, icl_soft_new_conn);
1516 	KASSERT(error == 0, ("failed to register"));
1517 #endif
1518 
1519 	return (error);
1520 }
1521 
1522 static int
1523 icl_soft_unload(void)
1524 {
1525 
1526 	if (icl_ncons != 0)
1527 		return (EBUSY);
1528 
1529 	icl_unregister("none", false);
1530 #if defined(ICL_KERNEL_PROXY) && 0
1531 	icl_unregister("proxytest", true);
1532 #endif
1533 
1534 	uma_zdestroy(icl_soft_pdu_zone);
1535 
1536 	return (0);
1537 }
1538 
1539 static int
1540 icl_soft_modevent(module_t mod, int what, void *arg)
1541 {
1542 
1543 	switch (what) {
1544 	case MOD_LOAD:
1545 		return (icl_soft_load());
1546 	case MOD_UNLOAD:
1547 		return (icl_soft_unload());
1548 	default:
1549 		return (EINVAL);
1550 	}
1551 }
1552 
1553 moduledata_t icl_soft_data = {
1554 	"icl_soft",
1555 	icl_soft_modevent,
1556 	0
1557 };
1558 
1559 DECLARE_MODULE(icl_soft, icl_soft_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
1560 MODULE_DEPEND(icl_soft, icl, 1, 1, 1);
1561 MODULE_VERSION(icl_soft, 1);
1562