xref: /freebsd/sys/dev/iser/icl_iser.c (revision 148a8da8)
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 	if (iser_conn->state != ISER_CONN_UP)
224 		return;
225 
226 	ret = iser_initialize_headers(iser_pdu, iser_conn);
227 	if (ret) {
228 		ISER_ERR("Failed to map TX descriptor pdu %p", iser_pdu);
229 		return;
230 	}
231 
232 	if (is_control_opcode(ip->ip_bhs->bhs_opcode)) {
233 		ret = iser_send_control(iser_conn, iser_pdu);
234 		if (unlikely(ret))
235 			ISER_ERR("Failed to send control pdu %p", iser_pdu);
236 	} else {
237 		ret = iser_send_command(iser_conn, iser_pdu);
238 		if (unlikely(ret))
239 			ISER_ERR("Failed to send command pdu %p", iser_pdu);
240 	}
241 }
242 
243 static struct icl_conn *
244 iser_new_conn(const char *name, struct mtx *lock)
245 {
246 	struct iser_conn *iser_conn;
247 	struct icl_conn *ic;
248 
249 	refcount_acquire(&icl_iser_ncons);
250 
251 	iser_conn = (struct iser_conn *)kobj_create(&icl_iser_class, M_ICL_ISER, M_WAITOK | M_ZERO);
252 	if (!iser_conn) {
253 		ISER_ERR("failed to allocate iser conn");
254 		refcount_release(&icl_iser_ncons);
255 		return (NULL);
256 	}
257 
258 	cv_init(&iser_conn->up_cv, "iser_cv");
259 	sx_init(&iser_conn->state_mutex, "iser_conn_state_mutex");
260 	mtx_init(&iser_conn->ib_conn.beacon.flush_lock, "iser_flush_lock", NULL, MTX_DEF);
261 	cv_init(&iser_conn->ib_conn.beacon.flush_cv, "flush_cv");
262 	mtx_init(&iser_conn->ib_conn.lock, "iser_lock", NULL, MTX_DEF);
263 
264 	ic = &iser_conn->icl_conn;
265 	ic->ic_lock = lock;
266 	ic->ic_name = name;
267 	ic->ic_offload = strdup("iser", M_TEMP);
268 	ic->ic_iser = true;
269 	ic->ic_unmapped = true;
270 
271 	return (ic);
272 }
273 
274 void
275 iser_conn_free(struct icl_conn *ic)
276 {
277 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
278 
279 	iser_conn_release(ic);
280 	mtx_destroy(&iser_conn->ib_conn.lock);
281 	cv_destroy(&iser_conn->ib_conn.beacon.flush_cv);
282 	mtx_destroy(&iser_conn->ib_conn.beacon.flush_lock);
283 	sx_destroy(&iser_conn->state_mutex);
284 	cv_destroy(&iser_conn->up_cv);
285 	kobj_delete((struct kobj *)iser_conn, M_ICL_ISER);
286 	refcount_release(&icl_iser_ncons);
287 }
288 
289 int
290 iser_conn_handoff(struct icl_conn *ic, int fd)
291 {
292 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
293 	int error = 0;
294 
295 	sx_xlock(&iser_conn->state_mutex);
296 	if (iser_conn->state != ISER_CONN_UP) {
297 		error = EINVAL;
298 		ISER_ERR("iser_conn %p state is %d, teardown started\n",
299 			 iser_conn, iser_conn->state);
300 		goto out;
301 	}
302 
303 	error = iser_alloc_rx_descriptors(iser_conn, ic->ic_maxtags);
304 	if (error)
305 		goto out;
306 
307 	error = iser_post_recvm(iser_conn, iser_conn->min_posted_rx);
308 	if (error)
309 		goto post_error;
310 
311 	iser_conn->handoff_done = true;
312 
313 	sx_xunlock(&iser_conn->state_mutex);
314 	return (error);
315 
316 post_error:
317 	iser_free_rx_descriptors(iser_conn);
318 out:
319 	sx_xunlock(&iser_conn->state_mutex);
320 	return (error);
321 
322 }
323 
324 /**
325  * Frees all conn objects
326  */
327 static void
328 iser_conn_release(struct icl_conn *ic)
329 {
330 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
331 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
332 	struct iser_conn *curr, *tmp;
333 
334 	mtx_lock(&ig.connlist_mutex);
335 	/*
336 	 * Search for iser connection in global list.
337 	 * It may not be there in case of failure in connection establishment
338 	 * stage.
339 	 */
340 	list_for_each_entry_safe(curr, tmp, &ig.connlist, conn_list) {
341 		if (iser_conn == curr) {
342 			ISER_WARN("found iser_conn %p", iser_conn);
343 			list_del(&iser_conn->conn_list);
344 		}
345 	}
346 	mtx_unlock(&ig.connlist_mutex);
347 
348 	/*
349 	 * In case we reconnecting or removing session, we need to
350 	 * release IB resources (which is safe to call more than once).
351 	 */
352 	sx_xlock(&iser_conn->state_mutex);
353 	iser_free_ib_conn_res(iser_conn, true);
354 	sx_xunlock(&iser_conn->state_mutex);
355 
356 	if (ib_conn->cma_id != NULL) {
357 		rdma_destroy_id(ib_conn->cma_id);
358 		ib_conn->cma_id = NULL;
359 	}
360 
361 }
362 
363 void
364 iser_conn_close(struct icl_conn *ic)
365 {
366 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
367 
368 	ISER_INFO("closing conn %p", iser_conn);
369 
370 	sx_xlock(&iser_conn->state_mutex);
371 	/*
372 	 * In case iser connection is waiting on conditional variable
373 	 * (state PENDING) and we try to close it before connection establishment,
374 	 * we need to signal it to continue releasing connection properly.
375 	 */
376 	if (!iser_conn_terminate(iser_conn) && iser_conn->state == ISER_CONN_PENDING)
377 		cv_signal(&iser_conn->up_cv);
378 	sx_xunlock(&iser_conn->state_mutex);
379 
380 }
381 
382 int
383 iser_conn_connect(struct icl_conn *ic, int domain, int socktype,
384 		int protocol, struct sockaddr *from_sa, struct sockaddr *to_sa)
385 {
386 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
387 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
388 	int err = 0;
389 
390 	iser_conn_release(ic);
391 
392 	sx_xlock(&iser_conn->state_mutex);
393 	 /* the device is known only --after-- address resolution */
394 	ib_conn->device = NULL;
395 	iser_conn->handoff_done = false;
396 
397 	iser_conn->state = ISER_CONN_PENDING;
398 
399 	ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler, (void *)iser_conn,
400 			RDMA_PS_TCP, IB_QPT_RC);
401 	if (IS_ERR(ib_conn->cma_id)) {
402 		err = -PTR_ERR(ib_conn->cma_id);
403 		ISER_ERR("rdma_create_id failed: %d", err);
404 		goto id_failure;
405 	}
406 
407 	err = rdma_resolve_addr(ib_conn->cma_id, from_sa, to_sa, 1000);
408 	if (err) {
409 		ISER_ERR("rdma_resolve_addr failed: %d", err);
410 		if (err < 0)
411 			err = -err;
412 		goto addr_failure;
413 	}
414 
415 	ISER_DBG("before cv_wait: %p", iser_conn);
416 	cv_wait(&iser_conn->up_cv, &iser_conn->state_mutex);
417 	ISER_DBG("after cv_wait: %p", iser_conn);
418 
419 	if (iser_conn->state != ISER_CONN_UP) {
420 		err = EIO;
421 		goto addr_failure;
422 	}
423 
424 	err = iser_alloc_login_buf(iser_conn);
425 	if (err)
426 		goto addr_failure;
427 	sx_xunlock(&iser_conn->state_mutex);
428 
429 	mtx_lock(&ig.connlist_mutex);
430 	list_add(&iser_conn->conn_list, &ig.connlist);
431 	mtx_unlock(&ig.connlist_mutex);
432 
433 	return (0);
434 
435 id_failure:
436 	ib_conn->cma_id = NULL;
437 addr_failure:
438 	sx_xunlock(&iser_conn->state_mutex);
439 	return (err);
440 }
441 
442 int
443 iser_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip,
444 		     struct ccb_scsiio *csio,
445 		     uint32_t *task_tagp, void **prvp)
446 {
447 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
448 
449 	*prvp = ip;
450 	iser_pdu->csio = csio;
451 
452 	return (0);
453 }
454 
455 void
456 iser_conn_task_done(struct icl_conn *ic, void *prv)
457 {
458 	struct icl_pdu *ip = prv;
459 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
460 	struct iser_device *device = iser_pdu->iser_conn->ib_conn.device;
461 	struct iser_tx_desc *tx_desc = &iser_pdu->desc;
462 
463 	if (iser_pdu->dir[ISER_DIR_IN]) {
464 		iser_unreg_rdma_mem(iser_pdu, ISER_DIR_IN);
465 		iser_dma_unmap_task_data(iser_pdu,
466 					 &iser_pdu->data[ISER_DIR_IN],
467 					 DMA_FROM_DEVICE);
468 	}
469 
470 	if (iser_pdu->dir[ISER_DIR_OUT]) {
471 		iser_unreg_rdma_mem(iser_pdu, ISER_DIR_OUT);
472 		iser_dma_unmap_task_data(iser_pdu,
473 					 &iser_pdu->data[ISER_DIR_OUT],
474 					 DMA_TO_DEVICE);
475 	}
476 
477 	if (likely(tx_desc->mapped)) {
478 		ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
479 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
480 		tx_desc->mapped = false;
481 	}
482 
483 	iser_pdu_free(ic, ip);
484 }
485 
486 static int
487 iser_limits(struct icl_drv_limits *idl)
488 {
489 
490 	idl->idl_max_recv_data_segment_length = 128 * 1024;
491 	idl->idl_max_send_data_segment_length = 128 * 1024;
492 	idl->idl_max_burst_length = 262144;
493 	idl->idl_first_burst_length = 65536;
494 
495 	return (0);
496 }
497 
498 static int
499 icl_iser_load(void)
500 {
501 	int error;
502 
503 	ISER_DBG("Starting iSER datamover...");
504 
505 	icl_pdu_zone = uma_zcreate("icl_iser_pdu", sizeof(struct icl_iser_pdu),
506 				   NULL, NULL, NULL, NULL,
507 				   UMA_ALIGN_PTR, 0);
508 	/* FIXME: Check rc */
509 
510 	refcount_init(&icl_iser_ncons, 0);
511 
512 	error = icl_register("iser", true, 0, iser_limits, iser_new_conn);
513 	KASSERT(error == 0, ("failed to register iser"));
514 
515 	memset(&ig, 0, sizeof(struct iser_global));
516 
517 	/* device init is called only after the first addr resolution */
518 	sx_init(&ig.device_list_mutex,  "global_device_lock");
519 	INIT_LIST_HEAD(&ig.device_list);
520 	mtx_init(&ig.connlist_mutex, "iser_global_conn_lock", NULL, MTX_DEF);
521 	INIT_LIST_HEAD(&ig.connlist);
522 	sx_init(&ig.close_conns_mutex,  "global_close_conns_lock");
523 
524 	return (error);
525 }
526 
527 static int
528 icl_iser_unload(void)
529 {
530 	ISER_DBG("Removing iSER datamover...");
531 
532 	if (icl_iser_ncons != 0)
533 		return (EBUSY);
534 
535 	sx_destroy(&ig.close_conns_mutex);
536 	mtx_destroy(&ig.connlist_mutex);
537 	sx_destroy(&ig.device_list_mutex);
538 
539 	icl_unregister("iser", true);
540 
541 	uma_zdestroy(icl_pdu_zone);
542 
543 	return (0);
544 }
545 
546 static int
547 icl_iser_modevent(module_t mod, int what, void *arg)
548 {
549 	switch (what) {
550 	case MOD_LOAD:
551 		return (icl_iser_load());
552 	case MOD_UNLOAD:
553 		return (icl_iser_unload());
554 	default:
555 		return (EINVAL);
556 	}
557 }
558 
559 moduledata_t icl_iser_data = {
560 	.name = "icl_iser",
561 	.evhand = icl_iser_modevent,
562 	.priv = 0
563 };
564 
565 DECLARE_MODULE(icl_iser, icl_iser_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
566 MODULE_DEPEND(icl_iser, icl, 1, 1, 1);
567 MODULE_DEPEND(icl_iser, ibcore, 1, 1, 1);
568 MODULE_DEPEND(icl_iser, linuxkpi, 1, 1, 1);
569 MODULE_VERSION(icl_iser, 1);
570