xref: /freebsd/sys/dev/iser/icl_iser.c (revision 95c4a39244184b2df3cb34b842859a57facc284b)
1 /* $FreeBSD$ */
2 /*-
3  * Copyright (c) 2015, Mellanox Technologies, Inc. All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include "icl_iser.h"
28 
29 SYSCTL_NODE(_kern, OID_AUTO, iser, CTLFLAG_RW, 0, "iSER module");
30 int iser_debug = 0;
31 SYSCTL_INT(_kern_iser, OID_AUTO, debug, CTLFLAG_RWTUN,
32     &iser_debug, 0, "Enable iser debug messages");
33 
34 static MALLOC_DEFINE(M_ICL_ISER, "icl_iser", "iSCSI iser backend");
35 static uma_zone_t icl_pdu_zone;
36 
37 static volatile u_int	icl_iser_ncons;
38 struct iser_global ig;
39 
40 static void iser_conn_release(struct icl_conn *ic);
41 
42 static icl_conn_new_pdu_t	iser_conn_new_pdu;
43 static icl_conn_pdu_free_t	iser_conn_pdu_free;
44 static icl_conn_pdu_data_segment_length_t iser_conn_pdu_data_segment_length;
45 static icl_conn_pdu_append_data_t	iser_conn_pdu_append_data;
46 static icl_conn_pdu_queue_t	iser_conn_pdu_queue;
47 static icl_conn_handoff_t	iser_conn_handoff;
48 static icl_conn_free_t		iser_conn_free;
49 static icl_conn_close_t		iser_conn_close;
50 static icl_conn_connect_t	iser_conn_connect;
51 static icl_conn_task_setup_t	iser_conn_task_setup;
52 static icl_conn_task_done_t	iser_conn_task_done;
53 static icl_conn_pdu_get_data_t	iser_conn_pdu_get_data;
54 
55 static kobj_method_t icl_iser_methods[] = {
56 	KOBJMETHOD(icl_conn_new_pdu, iser_conn_new_pdu),
57 	KOBJMETHOD(icl_conn_pdu_free, iser_conn_pdu_free),
58 	KOBJMETHOD(icl_conn_pdu_data_segment_length, iser_conn_pdu_data_segment_length),
59 	KOBJMETHOD(icl_conn_pdu_append_data, iser_conn_pdu_append_data),
60 	KOBJMETHOD(icl_conn_pdu_queue, iser_conn_pdu_queue),
61 	KOBJMETHOD(icl_conn_handoff, iser_conn_handoff),
62 	KOBJMETHOD(icl_conn_free, iser_conn_free),
63 	KOBJMETHOD(icl_conn_close, iser_conn_close),
64 	KOBJMETHOD(icl_conn_connect, iser_conn_connect),
65 	KOBJMETHOD(icl_conn_task_setup, iser_conn_task_setup),
66 	KOBJMETHOD(icl_conn_task_done, iser_conn_task_done),
67 	KOBJMETHOD(icl_conn_pdu_get_data, iser_conn_pdu_get_data),
68 	{ 0, 0 }
69 };
70 
71 DEFINE_CLASS(icl_iser, icl_iser_methods, sizeof(struct iser_conn));
72 
73 /**
74  * iser_initialize_headers() - Initialize task headers
75  * @pdu:       iser pdu
76  * @iser_conn:    iser connection
77  *
78  * Notes:
79  * This routine may race with iser teardown flow for scsi
80  * error handling TMFs. So for TMF we should acquire the
81  * state mutex to avoid dereferencing the IB device which
82  * may have already been terminated (racing teardown sequence).
83  */
84 int
85 iser_initialize_headers(struct icl_iser_pdu *pdu, struct iser_conn *iser_conn)
86 {
87 	struct iser_tx_desc *tx_desc = &pdu->desc;
88 	struct iser_device *device = iser_conn->ib_conn.device;
89 	u64 dma_addr;
90 	int ret = 0;
91 
92 	dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
93 				ISER_HEADERS_LEN, DMA_TO_DEVICE);
94 	if (ib_dma_mapping_error(device->ib_device, dma_addr)) {
95 		ret = -ENOMEM;
96 		goto out;
97 	}
98 
99 	tx_desc->mapped = true;
100 	tx_desc->dma_addr = dma_addr;
101 	tx_desc->tx_sg[0].addr   = tx_desc->dma_addr;
102 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
103 	tx_desc->tx_sg[0].lkey   = device->mr->lkey;
104 
105 out:
106 
107 	return (ret);
108 }
109 
110 int
111 iser_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *request,
112 			  const void *addr, size_t len, int flags)
113 {
114 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
115 
116 	if (request->ip_bhs->bhs_opcode & ISCSI_BHS_OPCODE_LOGIN_REQUEST ||
117 	    request->ip_bhs->bhs_opcode & ISCSI_BHS_OPCODE_TEXT_REQUEST) {
118 		ISER_DBG("copy to login buff");
119 		memcpy(iser_conn->login_req_buf, addr, len);
120 		request->ip_data_len = len;
121 	}
122 
123 	return (0);
124 }
125 
126 void
127 iser_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip,
128 		       size_t off, void *addr, size_t len)
129 {
130 	/* If we have a receive data, copy it to upper layer buffer */
131 	if (ip->ip_data_mbuf)
132 		memcpy(addr, ip->ip_data_mbuf + off, len);
133 }
134 
135 /*
136  * Allocate icl_pdu with empty BHS to fill up by the caller.
137  */
138 struct icl_pdu *
139 iser_new_pdu(struct icl_conn *ic, int flags)
140 {
141 	struct icl_iser_pdu *iser_pdu;
142 	struct icl_pdu *ip;
143 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
144 
145 	iser_pdu = uma_zalloc(icl_pdu_zone, flags | M_ZERO);
146 	if (iser_pdu == NULL) {
147 		ISER_WARN("failed to allocate %zd bytes", sizeof(*iser_pdu));
148 		return (NULL);
149 	}
150 
151 	iser_pdu->iser_conn = iser_conn;
152 	ip = &iser_pdu->icl_pdu;
153 	ip->ip_conn = ic;
154 	ip->ip_bhs = &iser_pdu->desc.iscsi_header;
155 
156 	return (ip);
157 }
158 
159 struct icl_pdu *
160 iser_conn_new_pdu(struct icl_conn *ic, int flags)
161 {
162 	return (iser_new_pdu(ic, flags));
163 }
164 
165 void
166 iser_pdu_free(struct icl_conn *ic, struct icl_pdu *ip)
167 {
168 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
169 
170 	uma_zfree(icl_pdu_zone, iser_pdu);
171 }
172 
173 size_t
174 iser_conn_pdu_data_segment_length(struct icl_conn *ic,
175 				  const struct icl_pdu *request)
176 {
177 	uint32_t len = 0;
178 
179 	len += request->ip_bhs->bhs_data_segment_len[0];
180 	len <<= 8;
181 	len += request->ip_bhs->bhs_data_segment_len[1];
182 	len <<= 8;
183 	len += request->ip_bhs->bhs_data_segment_len[2];
184 
185 	return (len);
186 }
187 
188 void
189 iser_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip)
190 {
191 	iser_pdu_free(ic, ip);
192 }
193 
194 static bool
195 is_control_opcode(uint8_t opcode)
196 {
197 	bool is_control = false;
198 
199 	switch (opcode & ISCSI_OPCODE_MASK) {
200 		case ISCSI_BHS_OPCODE_NOP_OUT:
201 		case ISCSI_BHS_OPCODE_LOGIN_REQUEST:
202 		case ISCSI_BHS_OPCODE_LOGOUT_REQUEST:
203 		case ISCSI_BHS_OPCODE_TEXT_REQUEST:
204 			is_control = true;
205 			break;
206 		case ISCSI_BHS_OPCODE_SCSI_COMMAND:
207 			is_control = false;
208 			break;
209 		default:
210 			ISER_ERR("unknown opcode %d", opcode);
211 	}
212 
213 	return (is_control);
214 }
215 
216 void
217 iser_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip)
218 {
219 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
220 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
221 	int ret;
222 
223 	ret = iser_initialize_headers(iser_pdu, iser_conn);
224 	if (ret) {
225 		ISER_ERR("Failed to map TX descriptor pdu %p", iser_pdu);
226 		return;
227 	}
228 
229 	if (is_control_opcode(ip->ip_bhs->bhs_opcode)) {
230 		ret = iser_send_control(iser_conn, iser_pdu);
231 		if (unlikely(ret))
232 			ISER_ERR("Failed to send control pdu %p", iser_pdu);
233 	} else {
234 		ret = iser_send_command(iser_conn, iser_pdu);
235 		if (unlikely(ret))
236 			ISER_ERR("Failed to send command pdu %p", iser_pdu);
237 	}
238 }
239 
240 static struct icl_conn *
241 iser_new_conn(const char *name, struct mtx *lock)
242 {
243 	struct iser_conn *iser_conn;
244 	struct icl_conn *ic;
245 
246 	refcount_acquire(&icl_iser_ncons);
247 
248 	iser_conn = (struct iser_conn *)kobj_create(&icl_iser_class, M_ICL_ISER, M_WAITOK | M_ZERO);
249 	if (!iser_conn) {
250 		ISER_ERR("failed to allocate iser conn");
251 		refcount_release(&icl_iser_ncons);
252 		return (NULL);
253 	}
254 
255 	cv_init(&iser_conn->up_cv, "iser_cv");
256 	sx_init(&iser_conn->state_mutex, "iser_conn_state_mutex");
257 	mtx_init(&iser_conn->ib_conn.beacon.flush_lock, "flush_lock", NULL, MTX_DEF);
258 	cv_init(&iser_conn->ib_conn.beacon.flush_cv, "flush_cv");
259 	mtx_init(&iser_conn->ib_conn.lock, "lock", NULL, MTX_DEF);
260 
261 	ic = &iser_conn->icl_conn;
262 	ic->ic_lock = lock;
263 	ic->ic_name = name;
264 	ic->ic_offload = strdup("iser", M_TEMP);
265 	ic->ic_iser = true;
266 	ic->ic_unmapped = true;
267 
268 	return (ic);
269 }
270 
271 void
272 iser_conn_free(struct icl_conn *ic)
273 {
274 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
275 
276 	iser_conn_release(ic);
277 	cv_destroy(&iser_conn->ib_conn.beacon.flush_cv);
278 	mtx_destroy(&iser_conn->ib_conn.beacon.flush_lock);
279 	sx_destroy(&iser_conn->state_mutex);
280 	cv_destroy(&iser_conn->up_cv);
281 	kobj_delete((struct kobj *)iser_conn, M_ICL_ISER);
282 	refcount_release(&icl_iser_ncons);
283 }
284 
285 int
286 iser_conn_handoff(struct icl_conn *ic, int fd)
287 {
288 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
289 	int error = 0;
290 
291 	sx_xlock(&iser_conn->state_mutex);
292 	if (iser_conn->state != ISER_CONN_UP) {
293 		error = EINVAL;
294 		ISER_ERR("iser_conn %p state is %d, teardown started\n",
295 			 iser_conn, iser_conn->state);
296 		goto out;
297 	}
298 
299 	error = iser_alloc_rx_descriptors(iser_conn, ic->ic_maxtags);
300 	if (error)
301 		goto out;
302 
303 	error = iser_post_recvm(iser_conn, iser_conn->min_posted_rx);
304 	if (error)
305 		goto post_error;
306 
307 	iser_conn->handoff_done = true;
308 
309 	sx_xunlock(&iser_conn->state_mutex);
310 	return (error);
311 
312 post_error:
313 	iser_free_rx_descriptors(iser_conn);
314 out:
315 	sx_xunlock(&iser_conn->state_mutex);
316 	return (error);
317 
318 }
319 
320 /**
321  * Frees all conn objects
322  */
323 static void
324 iser_conn_release(struct icl_conn *ic)
325 {
326 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
327 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
328 	struct iser_conn *curr, *tmp;
329 
330 	mtx_lock(&ig.connlist_mutex);
331 	/*
332 	 * Search for iser connection in global list.
333 	 * It may not be there in case of failure in connection establishment
334 	 * stage.
335 	 */
336 	list_for_each_entry_safe(curr, tmp, &ig.connlist, conn_list) {
337 		if (iser_conn == curr) {
338 			ISER_WARN("found iser_conn %p", iser_conn);
339 			list_del(&iser_conn->conn_list);
340 		}
341 	}
342 	mtx_unlock(&ig.connlist_mutex);
343 
344 	/*
345 	 * In case we reconnecting or removing session, we need to
346 	 * release IB resources (which is safe to call more than once).
347 	 */
348 	sx_xlock(&iser_conn->state_mutex);
349 	iser_free_ib_conn_res(iser_conn, true);
350 	sx_xunlock(&iser_conn->state_mutex);
351 
352 	if (ib_conn->cma_id != NULL) {
353 		rdma_destroy_id(ib_conn->cma_id);
354 		ib_conn->cma_id = NULL;
355 	}
356 
357 }
358 
359 void
360 iser_conn_close(struct icl_conn *ic)
361 {
362 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
363 
364 	ISER_INFO("closing conn %p", iser_conn);
365 
366 	sx_xlock(&iser_conn->state_mutex);
367 	/*
368 	 * In case iser connection is waiting on conditional variable
369 	 * (state PENDING) and we try to close it before connection establishment,
370 	 * we need to signal it to continue releasing connection properly.
371 	 */
372 	if (!iser_conn_terminate(iser_conn) && iser_conn->state == ISER_CONN_PENDING)
373 		cv_signal(&iser_conn->up_cv);
374 	sx_xunlock(&iser_conn->state_mutex);
375 
376 }
377 
378 int
379 iser_conn_connect(struct icl_conn *ic, int domain, int socktype,
380 		int protocol, struct sockaddr *from_sa, struct sockaddr *to_sa)
381 {
382 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
383 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
384 	int err = 0;
385 
386 	iser_conn_release(ic);
387 
388 	sx_xlock(&iser_conn->state_mutex);
389 	 /* the device is known only --after-- address resolution */
390 	ib_conn->device = NULL;
391 	iser_conn->handoff_done = false;
392 
393 	iser_conn->state = ISER_CONN_PENDING;
394 
395 	ib_conn->cma_id = rdma_create_id(iser_cma_handler, (void *)iser_conn,
396 			RDMA_PS_TCP, IB_QPT_RC);
397 	if (IS_ERR(ib_conn->cma_id)) {
398 		err = -PTR_ERR(ib_conn->cma_id);
399 		ISER_ERR("rdma_create_id failed: %d", err);
400 		goto id_failure;
401 	}
402 
403 	err = rdma_resolve_addr(ib_conn->cma_id, from_sa, to_sa, 1000);
404 	if (err) {
405 		ISER_ERR("rdma_resolve_addr failed: %d", err);
406 		if (err < 0)
407 			err = -err;
408 		goto addr_failure;
409 	}
410 
411 	ISER_DBG("before cv_wait: %p", iser_conn);
412 	cv_wait(&iser_conn->up_cv, &iser_conn->state_mutex);
413 	ISER_DBG("after cv_wait: %p", iser_conn);
414 
415 	if (iser_conn->state != ISER_CONN_UP) {
416 		err = EIO;
417 		goto addr_failure;
418 	}
419 
420 	err = iser_alloc_login_buf(iser_conn);
421 	if (err)
422 		goto addr_failure;
423 	sx_xunlock(&iser_conn->state_mutex);
424 
425 	mtx_lock(&ig.connlist_mutex);
426 	list_add(&iser_conn->conn_list, &ig.connlist);
427 	mtx_unlock(&ig.connlist_mutex);
428 
429 	return (0);
430 
431 id_failure:
432 	ib_conn->cma_id = NULL;
433 addr_failure:
434 	sx_xunlock(&iser_conn->state_mutex);
435 	return (err);
436 }
437 
438 int
439 iser_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip,
440 		     struct ccb_scsiio *csio,
441 		     uint32_t *task_tagp, void **prvp)
442 {
443 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
444 
445 	*prvp = ip;
446 	iser_pdu->csio = csio;
447 
448 	return (0);
449 }
450 
451 void
452 iser_conn_task_done(struct icl_conn *ic, void *prv)
453 {
454 	struct icl_pdu *ip = prv;
455 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
456 	struct iser_device *device = iser_pdu->iser_conn->ib_conn.device;
457 	struct iser_tx_desc *tx_desc = &iser_pdu->desc;
458 
459 	if (iser_pdu->dir[ISER_DIR_IN]) {
460 		iser_unreg_rdma_mem(iser_pdu, ISER_DIR_IN);
461 		iser_dma_unmap_task_data(iser_pdu,
462 					 &iser_pdu->data[ISER_DIR_IN],
463 					 DMA_FROM_DEVICE);
464 	}
465 
466 	if (iser_pdu->dir[ISER_DIR_OUT]) {
467 		iser_unreg_rdma_mem(iser_pdu, ISER_DIR_OUT);
468 		iser_dma_unmap_task_data(iser_pdu,
469 					 &iser_pdu->data[ISER_DIR_OUT],
470 					 DMA_TO_DEVICE);
471 	}
472 
473 	if (likely(tx_desc->mapped)) {
474 		ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
475 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
476 		tx_desc->mapped = false;
477 	}
478 
479 	iser_pdu_free(ic, ip);
480 }
481 
482 static int
483 iser_limits(size_t *limitp)
484 {
485 	*limitp = 128 * 1024;
486 
487 	return (0);
488 }
489 
490 static int
491 icl_iser_load(void)
492 {
493 	int error;
494 
495 	ISER_DBG("Starting iSER datamover...");
496 
497 	icl_pdu_zone = uma_zcreate("icl_iser_pdu", sizeof(struct icl_iser_pdu),
498 				   NULL, NULL, NULL, NULL,
499 				   UMA_ALIGN_PTR, 0);
500 	/* FIXME: Check rc */
501 
502 	refcount_init(&icl_iser_ncons, 0);
503 
504 	error = icl_register("iser", true, 0, iser_limits, iser_new_conn);
505 	KASSERT(error == 0, ("failed to register iser"));
506 
507 	memset(&ig, 0, sizeof(struct iser_global));
508 
509 	/* device init is called only after the first addr resolution */
510 	sx_init(&ig.device_list_mutex,  "global_device_lock");
511 	INIT_LIST_HEAD(&ig.device_list);
512 	mtx_init(&ig.connlist_mutex, "global_conn_lock", NULL, MTX_DEF);
513 	INIT_LIST_HEAD(&ig.connlist);
514 	sx_init(&ig.close_conns_mutex,  "global_close_conns_lock");
515 
516 	return (error);
517 }
518 
519 static int
520 icl_iser_unload(void)
521 {
522 	ISER_DBG("Removing iSER datamover...");
523 
524 	if (icl_iser_ncons != 0)
525 		return (EBUSY);
526 
527 	sx_destroy(&ig.close_conns_mutex);
528 	mtx_destroy(&ig.connlist_mutex);
529 	sx_destroy(&ig.device_list_mutex);
530 
531 	icl_unregister("iser", true);
532 
533 	uma_zdestroy(icl_pdu_zone);
534 
535 	return (0);
536 }
537 
538 static int
539 icl_iser_modevent(module_t mod, int what, void *arg)
540 {
541 	switch (what) {
542 	case MOD_LOAD:
543 		return (icl_iser_load());
544 	case MOD_UNLOAD:
545 		return (icl_iser_unload());
546 	default:
547 		return (EINVAL);
548 	}
549 }
550 
551 moduledata_t icl_iser_data = {
552 	.name = "icl_iser",
553 	.evhand = icl_iser_modevent,
554 	.priv = 0
555 };
556 
557 DECLARE_MODULE(icl_iser, icl_iser_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
558 MODULE_DEPEND(icl_iser, icl, 1, 1, 1);
559 MODULE_DEPEND(icl_iser, ibcore, 1, 1, 1);
560 MODULE_DEPEND(icl_iser, linuxkpi, 1, 1, 1);
561 MODULE_VERSION(icl_iser, 1);
562