xref: /freebsd/sys/dev/iser/icl_iser.c (revision 3cbb4cc200f8a0ad7ed08233425ea54524a21f1c)
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 | CTLFLAG_MPSAFE, 0,
30     "iSER module");
31 int iser_debug = 0;
32 SYSCTL_INT(_kern_iser, OID_AUTO, debug, CTLFLAG_RWTUN,
33     &iser_debug, 0, "Enable iser debug messages");
34 
35 static MALLOC_DEFINE(M_ICL_ISER, "icl_iser", "iSCSI iser backend");
36 static uma_zone_t icl_pdu_zone;
37 
38 static volatile u_int	icl_iser_ncons;
39 struct iser_global ig;
40 
41 static void iser_conn_release(struct icl_conn *ic);
42 
43 static icl_conn_new_pdu_t	iser_conn_new_pdu;
44 static icl_conn_pdu_free_t	iser_conn_pdu_free;
45 static icl_conn_pdu_data_segment_length_t iser_conn_pdu_data_segment_length;
46 static icl_conn_pdu_append_data_t	iser_conn_pdu_append_data;
47 static icl_conn_pdu_queue_t	iser_conn_pdu_queue;
48 static icl_conn_handoff_t	iser_conn_handoff;
49 static icl_conn_free_t		iser_conn_free;
50 static icl_conn_close_t		iser_conn_close;
51 static icl_conn_connect_t	iser_conn_connect;
52 static icl_conn_task_setup_t	iser_conn_task_setup;
53 static icl_conn_task_done_t	iser_conn_task_done;
54 static icl_conn_pdu_get_data_t	iser_conn_pdu_get_data;
55 
56 static kobj_method_t icl_iser_methods[] = {
57 	KOBJMETHOD(icl_conn_new_pdu, iser_conn_new_pdu),
58 	KOBJMETHOD(icl_conn_pdu_free, iser_conn_pdu_free),
59 	KOBJMETHOD(icl_conn_pdu_data_segment_length, iser_conn_pdu_data_segment_length),
60 	KOBJMETHOD(icl_conn_pdu_append_data, iser_conn_pdu_append_data),
61 	KOBJMETHOD(icl_conn_pdu_queue, iser_conn_pdu_queue),
62 	KOBJMETHOD(icl_conn_handoff, iser_conn_handoff),
63 	KOBJMETHOD(icl_conn_free, iser_conn_free),
64 	KOBJMETHOD(icl_conn_close, iser_conn_close),
65 	KOBJMETHOD(icl_conn_connect, iser_conn_connect),
66 	KOBJMETHOD(icl_conn_task_setup, iser_conn_task_setup),
67 	KOBJMETHOD(icl_conn_task_done, iser_conn_task_done),
68 	KOBJMETHOD(icl_conn_pdu_get_data, iser_conn_pdu_get_data),
69 	{ 0, 0 }
70 };
71 
72 DEFINE_CLASS(icl_iser, icl_iser_methods, sizeof(struct iser_conn));
73 
74 /**
75  * iser_initialize_headers() - Initialize task headers
76  * @pdu:       iser pdu
77  * @iser_conn:    iser connection
78  *
79  * Notes:
80  * This routine may race with iser teardown flow for scsi
81  * error handling TMFs. So for TMF we should acquire the
82  * state mutex to avoid dereferencing the IB device which
83  * may have already been terminated (racing teardown sequence).
84  */
85 int
86 iser_initialize_headers(struct icl_iser_pdu *pdu, struct iser_conn *iser_conn)
87 {
88 	struct iser_tx_desc *tx_desc = &pdu->desc;
89 	struct iser_device *device = iser_conn->ib_conn.device;
90 	u64 dma_addr;
91 	int ret = 0;
92 
93 	dma_addr = ib_dma_map_single(device->ib_device, (void *)tx_desc,
94 				ISER_HEADERS_LEN, DMA_TO_DEVICE);
95 	if (ib_dma_mapping_error(device->ib_device, dma_addr)) {
96 		ret = -ENOMEM;
97 		goto out;
98 	}
99 
100 	tx_desc->mapped = true;
101 	tx_desc->dma_addr = dma_addr;
102 	tx_desc->tx_sg[0].addr   = tx_desc->dma_addr;
103 	tx_desc->tx_sg[0].length = ISER_HEADERS_LEN;
104 	tx_desc->tx_sg[0].lkey   = device->mr->lkey;
105 
106 out:
107 
108 	return (ret);
109 }
110 
111 int
112 iser_conn_pdu_append_data(struct icl_conn *ic, struct icl_pdu *request,
113 			  const void *addr, size_t len, int flags)
114 {
115 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
116 
117 	if (request->ip_bhs->bhs_opcode & ISCSI_BHS_OPCODE_LOGIN_REQUEST ||
118 	    request->ip_bhs->bhs_opcode & ISCSI_BHS_OPCODE_TEXT_REQUEST) {
119 		ISER_DBG("copy to login buff");
120 		memcpy(iser_conn->login_req_buf, addr, len);
121 		request->ip_data_len = len;
122 	}
123 
124 	return (0);
125 }
126 
127 void
128 iser_conn_pdu_get_data(struct icl_conn *ic, struct icl_pdu *ip,
129 		       size_t off, void *addr, size_t len)
130 {
131 	/* If we have a receive data, copy it to upper layer buffer */
132 	if (ip->ip_data_mbuf)
133 		memcpy(addr, ip->ip_data_mbuf + off, len);
134 }
135 
136 /*
137  * Allocate icl_pdu with empty BHS to fill up by the caller.
138  */
139 struct icl_pdu *
140 iser_new_pdu(struct icl_conn *ic, int flags)
141 {
142 	struct icl_iser_pdu *iser_pdu;
143 	struct icl_pdu *ip;
144 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
145 
146 	iser_pdu = uma_zalloc(icl_pdu_zone, flags | M_ZERO);
147 	if (iser_pdu == NULL) {
148 		ISER_WARN("failed to allocate %zd bytes", sizeof(*iser_pdu));
149 		return (NULL);
150 	}
151 
152 	iser_pdu->iser_conn = iser_conn;
153 	ip = &iser_pdu->icl_pdu;
154 	ip->ip_conn = ic;
155 	ip->ip_bhs = &iser_pdu->desc.iscsi_header;
156 
157 	return (ip);
158 }
159 
160 struct icl_pdu *
161 iser_conn_new_pdu(struct icl_conn *ic, int flags)
162 {
163 	return (iser_new_pdu(ic, flags));
164 }
165 
166 void
167 iser_pdu_free(struct icl_conn *ic, struct icl_pdu *ip)
168 {
169 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
170 
171 	uma_zfree(icl_pdu_zone, iser_pdu);
172 }
173 
174 size_t
175 iser_conn_pdu_data_segment_length(struct icl_conn *ic,
176 				  const struct icl_pdu *request)
177 {
178 	uint32_t len = 0;
179 
180 	len += request->ip_bhs->bhs_data_segment_len[0];
181 	len <<= 8;
182 	len += request->ip_bhs->bhs_data_segment_len[1];
183 	len <<= 8;
184 	len += request->ip_bhs->bhs_data_segment_len[2];
185 
186 	return (len);
187 }
188 
189 void
190 iser_conn_pdu_free(struct icl_conn *ic, struct icl_pdu *ip)
191 {
192 	iser_pdu_free(ic, ip);
193 }
194 
195 static bool
196 is_control_opcode(uint8_t opcode)
197 {
198 	bool is_control = false;
199 
200 	switch (opcode & ISCSI_OPCODE_MASK) {
201 		case ISCSI_BHS_OPCODE_NOP_OUT:
202 		case ISCSI_BHS_OPCODE_LOGIN_REQUEST:
203 		case ISCSI_BHS_OPCODE_LOGOUT_REQUEST:
204 		case ISCSI_BHS_OPCODE_TEXT_REQUEST:
205 			is_control = true;
206 			break;
207 		case ISCSI_BHS_OPCODE_SCSI_COMMAND:
208 			is_control = false;
209 			break;
210 		default:
211 			ISER_ERR("unknown opcode %d", opcode);
212 	}
213 
214 	return (is_control);
215 }
216 
217 void
218 iser_conn_pdu_queue(struct icl_conn *ic, struct icl_pdu *ip)
219 {
220 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
221 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
222 	int ret;
223 
224 	if (iser_conn->state != ISER_CONN_UP)
225 		return;
226 
227 	ret = iser_initialize_headers(iser_pdu, iser_conn);
228 	if (ret) {
229 		ISER_ERR("Failed to map TX descriptor pdu %p", iser_pdu);
230 		return;
231 	}
232 
233 	if (is_control_opcode(ip->ip_bhs->bhs_opcode)) {
234 		ret = iser_send_control(iser_conn, iser_pdu);
235 		if (unlikely(ret))
236 			ISER_ERR("Failed to send control pdu %p", iser_pdu);
237 	} else {
238 		ret = iser_send_command(iser_conn, iser_pdu);
239 		if (unlikely(ret))
240 			ISER_ERR("Failed to send command pdu %p", iser_pdu);
241 	}
242 }
243 
244 static struct icl_conn *
245 iser_new_conn(const char *name, struct mtx *lock)
246 {
247 	struct iser_conn *iser_conn;
248 	struct icl_conn *ic;
249 
250 	refcount_acquire(&icl_iser_ncons);
251 
252 	iser_conn = (struct iser_conn *)kobj_create(&icl_iser_class, M_ICL_ISER, M_WAITOK | M_ZERO);
253 	if (!iser_conn) {
254 		ISER_ERR("failed to allocate iser conn");
255 		refcount_release(&icl_iser_ncons);
256 		return (NULL);
257 	}
258 
259 	cv_init(&iser_conn->up_cv, "iser_cv");
260 	sx_init(&iser_conn->state_mutex, "iser_conn_state_mutex");
261 	mtx_init(&iser_conn->ib_conn.beacon.flush_lock, "iser_flush_lock", NULL, MTX_DEF);
262 	cv_init(&iser_conn->ib_conn.beacon.flush_cv, "flush_cv");
263 	mtx_init(&iser_conn->ib_conn.lock, "iser_lock", NULL, MTX_DEF);
264 
265 	ic = &iser_conn->icl_conn;
266 	ic->ic_lock = lock;
267 	ic->ic_name = name;
268 	ic->ic_offload = strdup("iser", M_TEMP);
269 	ic->ic_iser = true;
270 	ic->ic_unmapped = true;
271 
272 	return (ic);
273 }
274 
275 void
276 iser_conn_free(struct icl_conn *ic)
277 {
278 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
279 
280 	iser_conn_release(ic);
281 	mtx_destroy(&iser_conn->ib_conn.lock);
282 	cv_destroy(&iser_conn->ib_conn.beacon.flush_cv);
283 	mtx_destroy(&iser_conn->ib_conn.beacon.flush_lock);
284 	sx_destroy(&iser_conn->state_mutex);
285 	cv_destroy(&iser_conn->up_cv);
286 	kobj_delete((struct kobj *)iser_conn, M_ICL_ISER);
287 	refcount_release(&icl_iser_ncons);
288 }
289 
290 int
291 iser_conn_handoff(struct icl_conn *ic, int fd)
292 {
293 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
294 	int error = 0;
295 
296 	sx_xlock(&iser_conn->state_mutex);
297 	if (iser_conn->state != ISER_CONN_UP) {
298 		error = EINVAL;
299 		ISER_ERR("iser_conn %p state is %d, teardown started\n",
300 			 iser_conn, iser_conn->state);
301 		goto out;
302 	}
303 
304 	error = iser_alloc_rx_descriptors(iser_conn, ic->ic_maxtags);
305 	if (error)
306 		goto out;
307 
308 	error = iser_post_recvm(iser_conn, iser_conn->min_posted_rx);
309 	if (error)
310 		goto post_error;
311 
312 	iser_conn->handoff_done = true;
313 
314 	sx_xunlock(&iser_conn->state_mutex);
315 	return (error);
316 
317 post_error:
318 	iser_free_rx_descriptors(iser_conn);
319 out:
320 	sx_xunlock(&iser_conn->state_mutex);
321 	return (error);
322 
323 }
324 
325 /**
326  * Frees all conn objects
327  */
328 static void
329 iser_conn_release(struct icl_conn *ic)
330 {
331 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
332 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
333 	struct iser_conn *curr, *tmp;
334 
335 	mtx_lock(&ig.connlist_mutex);
336 	/*
337 	 * Search for iser connection in global list.
338 	 * It may not be there in case of failure in connection establishment
339 	 * stage.
340 	 */
341 	list_for_each_entry_safe(curr, tmp, &ig.connlist, conn_list) {
342 		if (iser_conn == curr) {
343 			ISER_WARN("found iser_conn %p", iser_conn);
344 			list_del(&iser_conn->conn_list);
345 		}
346 	}
347 	mtx_unlock(&ig.connlist_mutex);
348 
349 	/*
350 	 * In case we reconnecting or removing session, we need to
351 	 * release IB resources (which is safe to call more than once).
352 	 */
353 	sx_xlock(&iser_conn->state_mutex);
354 	iser_free_ib_conn_res(iser_conn, true);
355 	sx_xunlock(&iser_conn->state_mutex);
356 
357 	if (ib_conn->cma_id != NULL) {
358 		rdma_destroy_id(ib_conn->cma_id);
359 		ib_conn->cma_id = NULL;
360 	}
361 
362 }
363 
364 void
365 iser_conn_close(struct icl_conn *ic)
366 {
367 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
368 
369 	ISER_INFO("closing conn %p", iser_conn);
370 
371 	sx_xlock(&iser_conn->state_mutex);
372 	/*
373 	 * In case iser connection is waiting on conditional variable
374 	 * (state PENDING) and we try to close it before connection establishment,
375 	 * we need to signal it to continue releasing connection properly.
376 	 */
377 	if (!iser_conn_terminate(iser_conn) && iser_conn->state == ISER_CONN_PENDING)
378 		cv_signal(&iser_conn->up_cv);
379 	sx_xunlock(&iser_conn->state_mutex);
380 
381 }
382 
383 int
384 iser_conn_connect(struct icl_conn *ic, int domain, int socktype,
385 		int protocol, struct sockaddr *from_sa, struct sockaddr *to_sa)
386 {
387 	struct iser_conn *iser_conn = icl_to_iser_conn(ic);
388 	struct ib_conn *ib_conn = &iser_conn->ib_conn;
389 	int err = 0;
390 
391 	iser_conn_release(ic);
392 
393 	sx_xlock(&iser_conn->state_mutex);
394 	 /* the device is known only --after-- address resolution */
395 	ib_conn->device = NULL;
396 	iser_conn->handoff_done = false;
397 
398 	iser_conn->state = ISER_CONN_PENDING;
399 
400 	ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler, (void *)iser_conn,
401 			RDMA_PS_TCP, IB_QPT_RC);
402 	if (IS_ERR(ib_conn->cma_id)) {
403 		err = -PTR_ERR(ib_conn->cma_id);
404 		ISER_ERR("rdma_create_id failed: %d", err);
405 		goto id_failure;
406 	}
407 
408 	err = rdma_resolve_addr(ib_conn->cma_id, from_sa, to_sa, 1000);
409 	if (err) {
410 		ISER_ERR("rdma_resolve_addr failed: %d", err);
411 		if (err < 0)
412 			err = -err;
413 		goto addr_failure;
414 	}
415 
416 	ISER_DBG("before cv_wait: %p", iser_conn);
417 	cv_wait(&iser_conn->up_cv, &iser_conn->state_mutex);
418 	ISER_DBG("after cv_wait: %p", iser_conn);
419 
420 	if (iser_conn->state != ISER_CONN_UP) {
421 		err = EIO;
422 		goto addr_failure;
423 	}
424 
425 	err = iser_alloc_login_buf(iser_conn);
426 	if (err)
427 		goto addr_failure;
428 	sx_xunlock(&iser_conn->state_mutex);
429 
430 	mtx_lock(&ig.connlist_mutex);
431 	list_add(&iser_conn->conn_list, &ig.connlist);
432 	mtx_unlock(&ig.connlist_mutex);
433 
434 	return (0);
435 
436 id_failure:
437 	ib_conn->cma_id = NULL;
438 addr_failure:
439 	sx_xunlock(&iser_conn->state_mutex);
440 	return (err);
441 }
442 
443 int
444 iser_conn_task_setup(struct icl_conn *ic, struct icl_pdu *ip,
445 		     struct ccb_scsiio *csio,
446 		     uint32_t *task_tagp, void **prvp)
447 {
448 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
449 
450 	*prvp = ip;
451 	iser_pdu->csio = csio;
452 
453 	return (0);
454 }
455 
456 void
457 iser_conn_task_done(struct icl_conn *ic, void *prv)
458 {
459 	struct icl_pdu *ip = prv;
460 	struct icl_iser_pdu *iser_pdu = icl_to_iser_pdu(ip);
461 	struct iser_device *device = iser_pdu->iser_conn->ib_conn.device;
462 	struct iser_tx_desc *tx_desc = &iser_pdu->desc;
463 
464 	if (iser_pdu->dir[ISER_DIR_IN]) {
465 		iser_unreg_rdma_mem(iser_pdu, ISER_DIR_IN);
466 		iser_dma_unmap_task_data(iser_pdu,
467 					 &iser_pdu->data[ISER_DIR_IN],
468 					 DMA_FROM_DEVICE);
469 	}
470 
471 	if (iser_pdu->dir[ISER_DIR_OUT]) {
472 		iser_unreg_rdma_mem(iser_pdu, ISER_DIR_OUT);
473 		iser_dma_unmap_task_data(iser_pdu,
474 					 &iser_pdu->data[ISER_DIR_OUT],
475 					 DMA_TO_DEVICE);
476 	}
477 
478 	if (likely(tx_desc->mapped)) {
479 		ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
480 				    ISER_HEADERS_LEN, DMA_TO_DEVICE);
481 		tx_desc->mapped = false;
482 	}
483 
484 	iser_pdu_free(ic, ip);
485 }
486 
487 static int
488 iser_limits(struct icl_drv_limits *idl)
489 {
490 
491 	idl->idl_max_recv_data_segment_length = 128 * 1024;
492 	idl->idl_max_send_data_segment_length = 128 * 1024;
493 	idl->idl_max_burst_length = 262144;
494 	idl->idl_first_burst_length = 65536;
495 
496 	return (0);
497 }
498 
499 static int
500 icl_iser_load(void)
501 {
502 	int error;
503 
504 	ISER_DBG("Starting iSER datamover...");
505 
506 	icl_pdu_zone = uma_zcreate("icl_iser_pdu", sizeof(struct icl_iser_pdu),
507 				   NULL, NULL, NULL, NULL,
508 				   UMA_ALIGN_PTR, 0);
509 	/* FIXME: Check rc */
510 
511 	refcount_init(&icl_iser_ncons, 0);
512 
513 	error = icl_register("iser", true, 0, iser_limits, iser_new_conn);
514 	KASSERT(error == 0, ("failed to register iser"));
515 
516 	memset(&ig, 0, sizeof(struct iser_global));
517 
518 	/* device init is called only after the first addr resolution */
519 	sx_init(&ig.device_list_mutex,  "global_device_lock");
520 	INIT_LIST_HEAD(&ig.device_list);
521 	mtx_init(&ig.connlist_mutex, "iser_global_conn_lock", NULL, MTX_DEF);
522 	INIT_LIST_HEAD(&ig.connlist);
523 	sx_init(&ig.close_conns_mutex,  "global_close_conns_lock");
524 
525 	return (error);
526 }
527 
528 static int
529 icl_iser_unload(void)
530 {
531 	ISER_DBG("Removing iSER datamover...");
532 
533 	if (icl_iser_ncons != 0)
534 		return (EBUSY);
535 
536 	sx_destroy(&ig.close_conns_mutex);
537 	mtx_destroy(&ig.connlist_mutex);
538 	sx_destroy(&ig.device_list_mutex);
539 
540 	icl_unregister("iser", true);
541 
542 	uma_zdestroy(icl_pdu_zone);
543 
544 	return (0);
545 }
546 
547 static int
548 icl_iser_modevent(module_t mod, int what, void *arg)
549 {
550 	switch (what) {
551 	case MOD_LOAD:
552 		return (icl_iser_load());
553 	case MOD_UNLOAD:
554 		return (icl_iser_unload());
555 	default:
556 		return (EINVAL);
557 	}
558 }
559 
560 moduledata_t icl_iser_data = {
561 	.name = "icl_iser",
562 	.evhand = icl_iser_modevent,
563 	.priv = 0
564 };
565 
566 DECLARE_MODULE(icl_iser, icl_iser_data, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);
567 MODULE_DEPEND(icl_iser, icl, 1, 1, 1);
568 MODULE_DEPEND(icl_iser, ibcore, 1, 1, 1);
569 MODULE_DEPEND(icl_iser, linuxkpi, 1, 1, 1);
570 MODULE_VERSION(icl_iser, 1);
571