Loading...
1/*
2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33#include <linux/kernel.h>
34#include <linux/module.h>
35#include <linux/slab.h>
36#include <linux/delay.h>
37
38#include "iscsi_iser.h"
39
40#define ISCSI_ISER_MAX_CONN 8
41#define ISER_MAX_RX_CQ_LEN (ISER_QP_MAX_RECV_DTOS * ISCSI_ISER_MAX_CONN)
42#define ISER_MAX_TX_CQ_LEN (ISER_QP_MAX_REQ_DTOS * ISCSI_ISER_MAX_CONN)
43
44static void iser_cq_tasklet_fn(unsigned long data);
45static void iser_cq_callback(struct ib_cq *cq, void *cq_context);
46
47static void iser_cq_event_callback(struct ib_event *cause, void *context)
48{
49 iser_err("got cq event %d \n", cause->event);
50}
51
52static void iser_qp_event_callback(struct ib_event *cause, void *context)
53{
54 iser_err("got qp event %d\n",cause->event);
55}
56
57static void iser_event_handler(struct ib_event_handler *handler,
58 struct ib_event *event)
59{
60 iser_err("async event %d on device %s port %d\n", event->event,
61 event->device->name, event->element.port_num);
62}
63
64/**
65 * iser_create_device_ib_res - creates Protection Domain (PD), Completion
66 * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
67 * the adapator.
68 *
69 * returns 0 on success, -1 on failure
70 */
71static int iser_create_device_ib_res(struct iser_device *device)
72{
73 device->pd = ib_alloc_pd(device->ib_device);
74 if (IS_ERR(device->pd))
75 goto pd_err;
76
77 device->rx_cq = ib_create_cq(device->ib_device,
78 iser_cq_callback,
79 iser_cq_event_callback,
80 (void *)device,
81 ISER_MAX_RX_CQ_LEN, 0);
82 if (IS_ERR(device->rx_cq))
83 goto rx_cq_err;
84
85 device->tx_cq = ib_create_cq(device->ib_device,
86 NULL, iser_cq_event_callback,
87 (void *)device,
88 ISER_MAX_TX_CQ_LEN, 0);
89
90 if (IS_ERR(device->tx_cq))
91 goto tx_cq_err;
92
93 if (ib_req_notify_cq(device->rx_cq, IB_CQ_NEXT_COMP))
94 goto cq_arm_err;
95
96 tasklet_init(&device->cq_tasklet,
97 iser_cq_tasklet_fn,
98 (unsigned long)device);
99
100 device->mr = ib_get_dma_mr(device->pd, IB_ACCESS_LOCAL_WRITE |
101 IB_ACCESS_REMOTE_WRITE |
102 IB_ACCESS_REMOTE_READ);
103 if (IS_ERR(device->mr))
104 goto dma_mr_err;
105
106 INIT_IB_EVENT_HANDLER(&device->event_handler, device->ib_device,
107 iser_event_handler);
108 if (ib_register_event_handler(&device->event_handler))
109 goto handler_err;
110
111 return 0;
112
113handler_err:
114 ib_dereg_mr(device->mr);
115dma_mr_err:
116 tasklet_kill(&device->cq_tasklet);
117cq_arm_err:
118 ib_destroy_cq(device->tx_cq);
119tx_cq_err:
120 ib_destroy_cq(device->rx_cq);
121rx_cq_err:
122 ib_dealloc_pd(device->pd);
123pd_err:
124 iser_err("failed to allocate an IB resource\n");
125 return -1;
126}
127
128/**
129 * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
130 * CQ and PD created with the device associated with the adapator.
131 */
132static void iser_free_device_ib_res(struct iser_device *device)
133{
134 BUG_ON(device->mr == NULL);
135
136 tasklet_kill(&device->cq_tasklet);
137 (void)ib_unregister_event_handler(&device->event_handler);
138 (void)ib_dereg_mr(device->mr);
139 (void)ib_destroy_cq(device->tx_cq);
140 (void)ib_destroy_cq(device->rx_cq);
141 (void)ib_dealloc_pd(device->pd);
142
143 device->mr = NULL;
144 device->tx_cq = NULL;
145 device->rx_cq = NULL;
146 device->pd = NULL;
147}
148
149/**
150 * iser_create_ib_conn_res - Creates FMR pool and Queue-Pair (QP)
151 *
152 * returns 0 on success, -1 on failure
153 */
154static int iser_create_ib_conn_res(struct iser_conn *ib_conn)
155{
156 struct iser_device *device;
157 struct ib_qp_init_attr init_attr;
158 int ret = -ENOMEM;
159 struct ib_fmr_pool_param params;
160
161 BUG_ON(ib_conn->device == NULL);
162
163 device = ib_conn->device;
164
165 ib_conn->login_buf = kmalloc(ISER_RX_LOGIN_SIZE, GFP_KERNEL);
166 if (!ib_conn->login_buf)
167 goto out_err;
168
169 ib_conn->login_dma = ib_dma_map_single(ib_conn->device->ib_device,
170 (void *)ib_conn->login_buf, ISER_RX_LOGIN_SIZE,
171 DMA_FROM_DEVICE);
172
173 ib_conn->page_vec = kmalloc(sizeof(struct iser_page_vec) +
174 (sizeof(u64) * (ISCSI_ISER_SG_TABLESIZE +1)),
175 GFP_KERNEL);
176 if (!ib_conn->page_vec)
177 goto out_err;
178
179 ib_conn->page_vec->pages = (u64 *) (ib_conn->page_vec + 1);
180
181 params.page_shift = SHIFT_4K;
182 /* when the first/last SG element are not start/end *
183 * page aligned, the map whould be of N+1 pages */
184 params.max_pages_per_fmr = ISCSI_ISER_SG_TABLESIZE + 1;
185 /* make the pool size twice the max number of SCSI commands *
186 * the ML is expected to queue, watermark for unmap at 50% */
187 params.pool_size = ISCSI_DEF_XMIT_CMDS_MAX * 2;
188 params.dirty_watermark = ISCSI_DEF_XMIT_CMDS_MAX;
189 params.cache = 0;
190 params.flush_function = NULL;
191 params.access = (IB_ACCESS_LOCAL_WRITE |
192 IB_ACCESS_REMOTE_WRITE |
193 IB_ACCESS_REMOTE_READ);
194
195 ib_conn->fmr_pool = ib_create_fmr_pool(device->pd, ¶ms);
196 if (IS_ERR(ib_conn->fmr_pool)) {
197 ret = PTR_ERR(ib_conn->fmr_pool);
198 ib_conn->fmr_pool = NULL;
199 goto out_err;
200 }
201
202 memset(&init_attr, 0, sizeof init_attr);
203
204 init_attr.event_handler = iser_qp_event_callback;
205 init_attr.qp_context = (void *)ib_conn;
206 init_attr.send_cq = device->tx_cq;
207 init_attr.recv_cq = device->rx_cq;
208 init_attr.cap.max_send_wr = ISER_QP_MAX_REQ_DTOS;
209 init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS;
210 init_attr.cap.max_send_sge = 2;
211 init_attr.cap.max_recv_sge = 1;
212 init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
213 init_attr.qp_type = IB_QPT_RC;
214
215 ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
216 if (ret)
217 goto out_err;
218
219 ib_conn->qp = ib_conn->cma_id->qp;
220 iser_err("setting conn %p cma_id %p: fmr_pool %p qp %p\n",
221 ib_conn, ib_conn->cma_id,
222 ib_conn->fmr_pool, ib_conn->cma_id->qp);
223 return ret;
224
225out_err:
226 iser_err("unable to alloc mem or create resource, err %d\n", ret);
227 return ret;
228}
229
230/**
231 * releases the FMR pool, QP and CMA ID objects, returns 0 on success,
232 * -1 on failure
233 */
234static int iser_free_ib_conn_res(struct iser_conn *ib_conn, int can_destroy_id)
235{
236 BUG_ON(ib_conn == NULL);
237
238 iser_err("freeing conn %p cma_id %p fmr pool %p qp %p\n",
239 ib_conn, ib_conn->cma_id,
240 ib_conn->fmr_pool, ib_conn->qp);
241
242 /* qp is created only once both addr & route are resolved */
243 if (ib_conn->fmr_pool != NULL)
244 ib_destroy_fmr_pool(ib_conn->fmr_pool);
245
246 if (ib_conn->qp != NULL)
247 rdma_destroy_qp(ib_conn->cma_id);
248
249 /* if cma handler context, the caller acts s.t the cma destroy the id */
250 if (ib_conn->cma_id != NULL && can_destroy_id)
251 rdma_destroy_id(ib_conn->cma_id);
252
253 ib_conn->fmr_pool = NULL;
254 ib_conn->qp = NULL;
255 ib_conn->cma_id = NULL;
256 kfree(ib_conn->page_vec);
257
258 return 0;
259}
260
261/**
262 * based on the resolved device node GUID see if there already allocated
263 * device for this device. If there's no such, create one.
264 */
265static
266struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
267{
268 struct iser_device *device;
269
270 mutex_lock(&ig.device_list_mutex);
271
272 list_for_each_entry(device, &ig.device_list, ig_list)
273 /* find if there's a match using the node GUID */
274 if (device->ib_device->node_guid == cma_id->device->node_guid)
275 goto inc_refcnt;
276
277 device = kzalloc(sizeof *device, GFP_KERNEL);
278 if (device == NULL)
279 goto out;
280
281 /* assign this device to the device */
282 device->ib_device = cma_id->device;
283 /* init the device and link it into ig device list */
284 if (iser_create_device_ib_res(device)) {
285 kfree(device);
286 device = NULL;
287 goto out;
288 }
289 list_add(&device->ig_list, &ig.device_list);
290
291inc_refcnt:
292 device->refcount++;
293out:
294 mutex_unlock(&ig.device_list_mutex);
295 return device;
296}
297
298/* if there's no demand for this device, release it */
299static void iser_device_try_release(struct iser_device *device)
300{
301 mutex_lock(&ig.device_list_mutex);
302 device->refcount--;
303 iser_err("device %p refcount %d\n",device,device->refcount);
304 if (!device->refcount) {
305 iser_free_device_ib_res(device);
306 list_del(&device->ig_list);
307 kfree(device);
308 }
309 mutex_unlock(&ig.device_list_mutex);
310}
311
312static int iser_conn_state_comp_exch(struct iser_conn *ib_conn,
313 enum iser_ib_conn_state comp,
314 enum iser_ib_conn_state exch)
315{
316 int ret;
317
318 spin_lock_bh(&ib_conn->lock);
319 if ((ret = (ib_conn->state == comp)))
320 ib_conn->state = exch;
321 spin_unlock_bh(&ib_conn->lock);
322 return ret;
323}
324
325/**
326 * Frees all conn objects and deallocs conn descriptor
327 */
328static void iser_conn_release(struct iser_conn *ib_conn, int can_destroy_id)
329{
330 struct iser_device *device = ib_conn->device;
331
332 BUG_ON(ib_conn->state != ISER_CONN_DOWN);
333
334 mutex_lock(&ig.connlist_mutex);
335 list_del(&ib_conn->conn_list);
336 mutex_unlock(&ig.connlist_mutex);
337 iser_free_rx_descriptors(ib_conn);
338 iser_free_ib_conn_res(ib_conn, can_destroy_id);
339 ib_conn->device = NULL;
340 /* on EVENT_ADDR_ERROR there's no device yet for this conn */
341 if (device != NULL)
342 iser_device_try_release(device);
343 iscsi_destroy_endpoint(ib_conn->ep);
344}
345
346void iser_conn_get(struct iser_conn *ib_conn)
347{
348 atomic_inc(&ib_conn->refcount);
349}
350
351int iser_conn_put(struct iser_conn *ib_conn, int can_destroy_id)
352{
353 if (atomic_dec_and_test(&ib_conn->refcount)) {
354 iser_conn_release(ib_conn, can_destroy_id);
355 return 1;
356 }
357 return 0;
358}
359
360/**
361 * triggers start of the disconnect procedures and wait for them to be done
362 */
363void iser_conn_terminate(struct iser_conn *ib_conn)
364{
365 int err = 0;
366
367 /* change the ib conn state only if the conn is UP, however always call
368 * rdma_disconnect since this is the only way to cause the CMA to change
369 * the QP state to ERROR
370 */
371
372 iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP, ISER_CONN_TERMINATING);
373 err = rdma_disconnect(ib_conn->cma_id);
374 if (err)
375 iser_err("Failed to disconnect, conn: 0x%p err %d\n",
376 ib_conn,err);
377
378 wait_event_interruptible(ib_conn->wait,
379 ib_conn->state == ISER_CONN_DOWN);
380
381 iser_conn_put(ib_conn, 1); /* deref ib conn deallocate */
382}
383
384static int iser_connect_error(struct rdma_cm_id *cma_id)
385{
386 struct iser_conn *ib_conn;
387 ib_conn = (struct iser_conn *)cma_id->context;
388
389 ib_conn->state = ISER_CONN_DOWN;
390 wake_up_interruptible(&ib_conn->wait);
391 return iser_conn_put(ib_conn, 0); /* deref ib conn's cma id */
392}
393
394static int iser_addr_handler(struct rdma_cm_id *cma_id)
395{
396 struct iser_device *device;
397 struct iser_conn *ib_conn;
398 int ret;
399
400 device = iser_device_find_by_ib_device(cma_id);
401 if (!device) {
402 iser_err("device lookup/creation failed\n");
403 return iser_connect_error(cma_id);
404 }
405
406 ib_conn = (struct iser_conn *)cma_id->context;
407 ib_conn->device = device;
408
409 ret = rdma_resolve_route(cma_id, 1000);
410 if (ret) {
411 iser_err("resolve route failed: %d\n", ret);
412 return iser_connect_error(cma_id);
413 }
414
415 return 0;
416}
417
418static int iser_route_handler(struct rdma_cm_id *cma_id)
419{
420 struct rdma_conn_param conn_param;
421 int ret;
422
423 ret = iser_create_ib_conn_res((struct iser_conn *)cma_id->context);
424 if (ret)
425 goto failure;
426
427 memset(&conn_param, 0, sizeof conn_param);
428 conn_param.responder_resources = 4;
429 conn_param.initiator_depth = 1;
430 conn_param.retry_count = 7;
431 conn_param.rnr_retry_count = 6;
432
433 ret = rdma_connect(cma_id, &conn_param);
434 if (ret) {
435 iser_err("failure connecting: %d\n", ret);
436 goto failure;
437 }
438
439 return 0;
440failure:
441 return iser_connect_error(cma_id);
442}
443
444static void iser_connected_handler(struct rdma_cm_id *cma_id)
445{
446 struct iser_conn *ib_conn;
447
448 ib_conn = (struct iser_conn *)cma_id->context;
449 ib_conn->state = ISER_CONN_UP;
450 wake_up_interruptible(&ib_conn->wait);
451}
452
453static int iser_disconnected_handler(struct rdma_cm_id *cma_id)
454{
455 struct iser_conn *ib_conn;
456 int ret;
457
458 ib_conn = (struct iser_conn *)cma_id->context;
459
460 /* getting here when the state is UP means that the conn is being *
461 * terminated asynchronously from the iSCSI layer's perspective. */
462 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
463 ISER_CONN_TERMINATING))
464 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
465 ISCSI_ERR_CONN_FAILED);
466
467 /* Complete the termination process if no posts are pending */
468 if (ib_conn->post_recv_buf_count == 0 &&
469 (atomic_read(&ib_conn->post_send_buf_count) == 0)) {
470 ib_conn->state = ISER_CONN_DOWN;
471 wake_up_interruptible(&ib_conn->wait);
472 }
473
474 ret = iser_conn_put(ib_conn, 0); /* deref ib conn's cma id */
475 return ret;
476}
477
478static int iser_cma_handler(struct rdma_cm_id *cma_id, struct rdma_cm_event *event)
479{
480 int ret = 0;
481
482 iser_err("event %d status %d conn %p id %p\n",
483 event->event, event->status, cma_id->context, cma_id);
484
485 switch (event->event) {
486 case RDMA_CM_EVENT_ADDR_RESOLVED:
487 ret = iser_addr_handler(cma_id);
488 break;
489 case RDMA_CM_EVENT_ROUTE_RESOLVED:
490 ret = iser_route_handler(cma_id);
491 break;
492 case RDMA_CM_EVENT_ESTABLISHED:
493 iser_connected_handler(cma_id);
494 break;
495 case RDMA_CM_EVENT_ADDR_ERROR:
496 case RDMA_CM_EVENT_ROUTE_ERROR:
497 case RDMA_CM_EVENT_CONNECT_ERROR:
498 case RDMA_CM_EVENT_UNREACHABLE:
499 case RDMA_CM_EVENT_REJECTED:
500 ret = iser_connect_error(cma_id);
501 break;
502 case RDMA_CM_EVENT_DISCONNECTED:
503 case RDMA_CM_EVENT_DEVICE_REMOVAL:
504 case RDMA_CM_EVENT_ADDR_CHANGE:
505 ret = iser_disconnected_handler(cma_id);
506 break;
507 default:
508 iser_err("Unexpected RDMA CM event (%d)\n", event->event);
509 break;
510 }
511 return ret;
512}
513
514void iser_conn_init(struct iser_conn *ib_conn)
515{
516 ib_conn->state = ISER_CONN_INIT;
517 init_waitqueue_head(&ib_conn->wait);
518 ib_conn->post_recv_buf_count = 0;
519 atomic_set(&ib_conn->post_send_buf_count, 0);
520 atomic_set(&ib_conn->refcount, 1); /* ref ib conn allocation */
521 INIT_LIST_HEAD(&ib_conn->conn_list);
522 spin_lock_init(&ib_conn->lock);
523}
524
525 /**
526 * starts the process of connecting to the target
527 * sleeps until the connection is established or rejected
528 */
529int iser_connect(struct iser_conn *ib_conn,
530 struct sockaddr_in *src_addr,
531 struct sockaddr_in *dst_addr,
532 int non_blocking)
533{
534 struct sockaddr *src, *dst;
535 int err = 0;
536
537 sprintf(ib_conn->name, "%pI4:%d",
538 &dst_addr->sin_addr.s_addr, dst_addr->sin_port);
539
540 /* the device is known only --after-- address resolution */
541 ib_conn->device = NULL;
542
543 iser_err("connecting to: %pI4, port 0x%x\n",
544 &dst_addr->sin_addr, dst_addr->sin_port);
545
546 ib_conn->state = ISER_CONN_PENDING;
547
548 iser_conn_get(ib_conn); /* ref ib conn's cma id */
549 ib_conn->cma_id = rdma_create_id(iser_cma_handler,
550 (void *)ib_conn,
551 RDMA_PS_TCP, IB_QPT_RC);
552 if (IS_ERR(ib_conn->cma_id)) {
553 err = PTR_ERR(ib_conn->cma_id);
554 iser_err("rdma_create_id failed: %d\n", err);
555 goto id_failure;
556 }
557
558 src = (struct sockaddr *)src_addr;
559 dst = (struct sockaddr *)dst_addr;
560 err = rdma_resolve_addr(ib_conn->cma_id, src, dst, 1000);
561 if (err) {
562 iser_err("rdma_resolve_addr failed: %d\n", err);
563 goto addr_failure;
564 }
565
566 if (!non_blocking) {
567 wait_event_interruptible(ib_conn->wait,
568 (ib_conn->state != ISER_CONN_PENDING));
569
570 if (ib_conn->state != ISER_CONN_UP) {
571 err = -EIO;
572 goto connect_failure;
573 }
574 }
575
576 mutex_lock(&ig.connlist_mutex);
577 list_add(&ib_conn->conn_list, &ig.connlist);
578 mutex_unlock(&ig.connlist_mutex);
579 return 0;
580
581id_failure:
582 ib_conn->cma_id = NULL;
583addr_failure:
584 ib_conn->state = ISER_CONN_DOWN;
585connect_failure:
586 iser_conn_release(ib_conn, 1);
587 return err;
588}
589
590/**
591 * iser_reg_page_vec - Register physical memory
592 *
593 * returns: 0 on success, errno code on failure
594 */
595int iser_reg_page_vec(struct iser_conn *ib_conn,
596 struct iser_page_vec *page_vec,
597 struct iser_mem_reg *mem_reg)
598{
599 struct ib_pool_fmr *mem;
600 u64 io_addr;
601 u64 *page_list;
602 int status;
603
604 page_list = page_vec->pages;
605 io_addr = page_list[0];
606
607 mem = ib_fmr_pool_map_phys(ib_conn->fmr_pool,
608 page_list,
609 page_vec->length,
610 io_addr);
611
612 if (IS_ERR(mem)) {
613 status = (int)PTR_ERR(mem);
614 iser_err("ib_fmr_pool_map_phys failed: %d\n", status);
615 return status;
616 }
617
618 mem_reg->lkey = mem->fmr->lkey;
619 mem_reg->rkey = mem->fmr->rkey;
620 mem_reg->len = page_vec->length * SIZE_4K;
621 mem_reg->va = io_addr;
622 mem_reg->is_fmr = 1;
623 mem_reg->mem_h = (void *)mem;
624
625 mem_reg->va += page_vec->offset;
626 mem_reg->len = page_vec->data_size;
627
628 iser_dbg("PHYSICAL Mem.register, [PHYS p_array: 0x%p, sz: %d, "
629 "entry[0]: (0x%08lx,%ld)] -> "
630 "[lkey: 0x%08X mem_h: 0x%p va: 0x%08lX sz: %ld]\n",
631 page_vec, page_vec->length,
632 (unsigned long)page_vec->pages[0],
633 (unsigned long)page_vec->data_size,
634 (unsigned int)mem_reg->lkey, mem_reg->mem_h,
635 (unsigned long)mem_reg->va, (unsigned long)mem_reg->len);
636 return 0;
637}
638
639/**
640 * Unregister (previosuly registered) memory.
641 */
642void iser_unreg_mem(struct iser_mem_reg *reg)
643{
644 int ret;
645
646 iser_dbg("PHYSICAL Mem.Unregister mem_h %p\n",reg->mem_h);
647
648 ret = ib_fmr_pool_unmap((struct ib_pool_fmr *)reg->mem_h);
649 if (ret)
650 iser_err("ib_fmr_pool_unmap failed %d\n", ret);
651
652 reg->mem_h = NULL;
653}
654
655int iser_post_recvl(struct iser_conn *ib_conn)
656{
657 struct ib_recv_wr rx_wr, *rx_wr_failed;
658 struct ib_sge sge;
659 int ib_ret;
660
661 sge.addr = ib_conn->login_dma;
662 sge.length = ISER_RX_LOGIN_SIZE;
663 sge.lkey = ib_conn->device->mr->lkey;
664
665 rx_wr.wr_id = (unsigned long)ib_conn->login_buf;
666 rx_wr.sg_list = &sge;
667 rx_wr.num_sge = 1;
668 rx_wr.next = NULL;
669
670 ib_conn->post_recv_buf_count++;
671 ib_ret = ib_post_recv(ib_conn->qp, &rx_wr, &rx_wr_failed);
672 if (ib_ret) {
673 iser_err("ib_post_recv failed ret=%d\n", ib_ret);
674 ib_conn->post_recv_buf_count--;
675 }
676 return ib_ret;
677}
678
679int iser_post_recvm(struct iser_conn *ib_conn, int count)
680{
681 struct ib_recv_wr *rx_wr, *rx_wr_failed;
682 int i, ib_ret;
683 unsigned int my_rx_head = ib_conn->rx_desc_head;
684 struct iser_rx_desc *rx_desc;
685
686 for (rx_wr = ib_conn->rx_wr, i = 0; i < count; i++, rx_wr++) {
687 rx_desc = &ib_conn->rx_descs[my_rx_head];
688 rx_wr->wr_id = (unsigned long)rx_desc;
689 rx_wr->sg_list = &rx_desc->rx_sg;
690 rx_wr->num_sge = 1;
691 rx_wr->next = rx_wr + 1;
692 my_rx_head = (my_rx_head + 1) & (ISER_QP_MAX_RECV_DTOS - 1);
693 }
694
695 rx_wr--;
696 rx_wr->next = NULL; /* mark end of work requests list */
697
698 ib_conn->post_recv_buf_count += count;
699 ib_ret = ib_post_recv(ib_conn->qp, ib_conn->rx_wr, &rx_wr_failed);
700 if (ib_ret) {
701 iser_err("ib_post_recv failed ret=%d\n", ib_ret);
702 ib_conn->post_recv_buf_count -= count;
703 } else
704 ib_conn->rx_desc_head = my_rx_head;
705 return ib_ret;
706}
707
708
709/**
710 * iser_start_send - Initiate a Send DTO operation
711 *
712 * returns 0 on success, -1 on failure
713 */
714int iser_post_send(struct iser_conn *ib_conn, struct iser_tx_desc *tx_desc)
715{
716 int ib_ret;
717 struct ib_send_wr send_wr, *send_wr_failed;
718
719 ib_dma_sync_single_for_device(ib_conn->device->ib_device,
720 tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
721
722 send_wr.next = NULL;
723 send_wr.wr_id = (unsigned long)tx_desc;
724 send_wr.sg_list = tx_desc->tx_sg;
725 send_wr.num_sge = tx_desc->num_sge;
726 send_wr.opcode = IB_WR_SEND;
727 send_wr.send_flags = IB_SEND_SIGNALED;
728
729 atomic_inc(&ib_conn->post_send_buf_count);
730
731 ib_ret = ib_post_send(ib_conn->qp, &send_wr, &send_wr_failed);
732 if (ib_ret) {
733 iser_err("ib_post_send failed, ret:%d\n", ib_ret);
734 atomic_dec(&ib_conn->post_send_buf_count);
735 }
736 return ib_ret;
737}
738
739static void iser_handle_comp_error(struct iser_tx_desc *desc,
740 struct iser_conn *ib_conn)
741{
742 if (desc && desc->type == ISCSI_TX_DATAOUT)
743 kmem_cache_free(ig.desc_cache, desc);
744
745 if (ib_conn->post_recv_buf_count == 0 &&
746 atomic_read(&ib_conn->post_send_buf_count) == 0) {
747 /* getting here when the state is UP means that the conn is *
748 * being terminated asynchronously from the iSCSI layer's *
749 * perspective. */
750 if (iser_conn_state_comp_exch(ib_conn, ISER_CONN_UP,
751 ISER_CONN_TERMINATING))
752 iscsi_conn_failure(ib_conn->iser_conn->iscsi_conn,
753 ISCSI_ERR_CONN_FAILED);
754
755 /* no more non completed posts to the QP, complete the
756 * termination process w.o worrying on disconnect event */
757 ib_conn->state = ISER_CONN_DOWN;
758 wake_up_interruptible(&ib_conn->wait);
759 }
760}
761
762static int iser_drain_tx_cq(struct iser_device *device)
763{
764 struct ib_cq *cq = device->tx_cq;
765 struct ib_wc wc;
766 struct iser_tx_desc *tx_desc;
767 struct iser_conn *ib_conn;
768 int completed_tx = 0;
769
770 while (ib_poll_cq(cq, 1, &wc) == 1) {
771 tx_desc = (struct iser_tx_desc *) (unsigned long) wc.wr_id;
772 ib_conn = wc.qp->qp_context;
773 if (wc.status == IB_WC_SUCCESS) {
774 if (wc.opcode == IB_WC_SEND)
775 iser_snd_completion(tx_desc, ib_conn);
776 else
777 iser_err("expected opcode %d got %d\n",
778 IB_WC_SEND, wc.opcode);
779 } else {
780 iser_err("tx id %llx status %d vend_err %x\n",
781 wc.wr_id, wc.status, wc.vendor_err);
782 atomic_dec(&ib_conn->post_send_buf_count);
783 iser_handle_comp_error(tx_desc, ib_conn);
784 }
785 completed_tx++;
786 }
787 return completed_tx;
788}
789
790
791static void iser_cq_tasklet_fn(unsigned long data)
792{
793 struct iser_device *device = (struct iser_device *)data;
794 struct ib_cq *cq = device->rx_cq;
795 struct ib_wc wc;
796 struct iser_rx_desc *desc;
797 unsigned long xfer_len;
798 struct iser_conn *ib_conn;
799 int completed_tx, completed_rx;
800 completed_tx = completed_rx = 0;
801
802 while (ib_poll_cq(cq, 1, &wc) == 1) {
803 desc = (struct iser_rx_desc *) (unsigned long) wc.wr_id;
804 BUG_ON(desc == NULL);
805 ib_conn = wc.qp->qp_context;
806 if (wc.status == IB_WC_SUCCESS) {
807 if (wc.opcode == IB_WC_RECV) {
808 xfer_len = (unsigned long)wc.byte_len;
809 iser_rcv_completion(desc, xfer_len, ib_conn);
810 } else
811 iser_err("expected opcode %d got %d\n",
812 IB_WC_RECV, wc.opcode);
813 } else {
814 if (wc.status != IB_WC_WR_FLUSH_ERR)
815 iser_err("rx id %llx status %d vend_err %x\n",
816 wc.wr_id, wc.status, wc.vendor_err);
817 ib_conn->post_recv_buf_count--;
818 iser_handle_comp_error(NULL, ib_conn);
819 }
820 completed_rx++;
821 if (!(completed_rx & 63))
822 completed_tx += iser_drain_tx_cq(device);
823 }
824 /* #warning "it is assumed here that arming CQ only once its empty" *
825 * " would not cause interrupts to be missed" */
826 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
827
828 completed_tx += iser_drain_tx_cq(device);
829 iser_dbg("got %d rx %d tx completions\n", completed_rx, completed_tx);
830}
831
832static void iser_cq_callback(struct ib_cq *cq, void *cq_context)
833{
834 struct iser_device *device = (struct iser_device *)cq_context;
835
836 tasklet_schedule(&device->cq_tasklet);
837}
1/*
2 * Copyright (c) 2004, 2005, 2006 Voltaire, Inc. All rights reserved.
3 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
4 * Copyright (c) 2013-2014 Mellanox Technologies. All rights reserved.
5 *
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
24 *
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 * SOFTWARE.
33 */
34#include <linux/kernel.h>
35#include <linux/slab.h>
36#include <linux/delay.h>
37
38#include "iscsi_iser.h"
39
40static void iser_qp_event_callback(struct ib_event *cause, void *context)
41{
42 iser_err("qp event %s (%d)\n",
43 ib_event_msg(cause->event), cause->event);
44}
45
46static void iser_event_handler(struct ib_event_handler *handler,
47 struct ib_event *event)
48{
49 iser_err("async event %s (%d) on device %s port %d\n",
50 ib_event_msg(event->event), event->event,
51 dev_name(&event->device->dev), event->element.port_num);
52}
53
54/*
55 * iser_create_device_ib_res - creates Protection Domain (PD), Completion
56 * Queue (CQ), DMA Memory Region (DMA MR) with the device associated with
57 * the adaptor.
58 *
59 * Return: 0 on success, -1 on failure
60 */
61static int iser_create_device_ib_res(struct iser_device *device)
62{
63 struct ib_device *ib_dev = device->ib_device;
64
65 if (!(ib_dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS)) {
66 iser_err("IB device does not support memory registrations\n");
67 return -1;
68 }
69
70 device->pd = ib_alloc_pd(ib_dev,
71 iser_always_reg ? 0 : IB_PD_UNSAFE_GLOBAL_RKEY);
72 if (IS_ERR(device->pd))
73 goto pd_err;
74
75 INIT_IB_EVENT_HANDLER(&device->event_handler, ib_dev,
76 iser_event_handler);
77 ib_register_event_handler(&device->event_handler);
78 return 0;
79
80pd_err:
81 iser_err("failed to allocate an IB resource\n");
82 return -1;
83}
84
85/*
86 * iser_free_device_ib_res - destroy/dealloc/dereg the DMA MR,
87 * CQ and PD created with the device associated with the adaptor.
88 */
89static void iser_free_device_ib_res(struct iser_device *device)
90{
91 ib_unregister_event_handler(&device->event_handler);
92 ib_dealloc_pd(device->pd);
93
94 device->pd = NULL;
95}
96
97static struct iser_fr_desc *
98iser_create_fastreg_desc(struct iser_device *device,
99 struct ib_pd *pd,
100 bool pi_enable,
101 unsigned int size)
102{
103 struct iser_fr_desc *desc;
104 struct ib_device *ib_dev = device->ib_device;
105 enum ib_mr_type mr_type;
106 int ret;
107
108 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
109 if (!desc)
110 return ERR_PTR(-ENOMEM);
111
112 if (ib_dev->attrs.kernel_cap_flags & IBK_SG_GAPS_REG)
113 mr_type = IB_MR_TYPE_SG_GAPS;
114 else
115 mr_type = IB_MR_TYPE_MEM_REG;
116
117 desc->rsc.mr = ib_alloc_mr(pd, mr_type, size);
118 if (IS_ERR(desc->rsc.mr)) {
119 ret = PTR_ERR(desc->rsc.mr);
120 iser_err("Failed to allocate ib_fast_reg_mr err=%d\n", ret);
121 goto err_alloc_mr;
122 }
123
124 if (pi_enable) {
125 desc->rsc.sig_mr = ib_alloc_mr_integrity(pd, size, size);
126 if (IS_ERR(desc->rsc.sig_mr)) {
127 ret = PTR_ERR(desc->rsc.sig_mr);
128 iser_err("Failed to allocate sig_mr err=%d\n", ret);
129 goto err_alloc_mr_integrity;
130 }
131 }
132
133 return desc;
134
135err_alloc_mr_integrity:
136 ib_dereg_mr(desc->rsc.mr);
137err_alloc_mr:
138 kfree(desc);
139
140 return ERR_PTR(ret);
141}
142
143static void iser_destroy_fastreg_desc(struct iser_fr_desc *desc)
144{
145 struct iser_reg_resources *res = &desc->rsc;
146
147 ib_dereg_mr(res->mr);
148 if (res->sig_mr) {
149 ib_dereg_mr(res->sig_mr);
150 res->sig_mr = NULL;
151 }
152 kfree(desc);
153}
154
155/**
156 * iser_alloc_fastreg_pool - Creates pool of fast_reg descriptors
157 * for fast registration work requests.
158 * @ib_conn: connection RDMA resources
159 * @cmds_max: max number of SCSI commands for this connection
160 * @size: max number of pages per map request
161 *
162 * Return: 0 on success, or errno code on failure
163 */
164int iser_alloc_fastreg_pool(struct ib_conn *ib_conn,
165 unsigned cmds_max,
166 unsigned int size)
167{
168 struct iser_device *device = ib_conn->device;
169 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
170 struct iser_fr_desc *desc;
171 int i, ret;
172
173 INIT_LIST_HEAD(&fr_pool->list);
174 INIT_LIST_HEAD(&fr_pool->all_list);
175 spin_lock_init(&fr_pool->lock);
176 fr_pool->size = 0;
177 for (i = 0; i < cmds_max; i++) {
178 desc = iser_create_fastreg_desc(device, device->pd,
179 ib_conn->pi_support, size);
180 if (IS_ERR(desc)) {
181 ret = PTR_ERR(desc);
182 goto err;
183 }
184
185 list_add_tail(&desc->list, &fr_pool->list);
186 list_add_tail(&desc->all_list, &fr_pool->all_list);
187 fr_pool->size++;
188 }
189
190 return 0;
191
192err:
193 iser_free_fastreg_pool(ib_conn);
194 return ret;
195}
196
197/**
198 * iser_free_fastreg_pool - releases the pool of fast_reg descriptors
199 * @ib_conn: connection RDMA resources
200 */
201void iser_free_fastreg_pool(struct ib_conn *ib_conn)
202{
203 struct iser_fr_pool *fr_pool = &ib_conn->fr_pool;
204 struct iser_fr_desc *desc, *tmp;
205 int i = 0;
206
207 if (list_empty(&fr_pool->all_list))
208 return;
209
210 iser_info("freeing conn %p fr pool\n", ib_conn);
211
212 list_for_each_entry_safe(desc, tmp, &fr_pool->all_list, all_list) {
213 list_del(&desc->all_list);
214 iser_destroy_fastreg_desc(desc);
215 ++i;
216 }
217
218 if (i < fr_pool->size)
219 iser_warn("pool still has %d regions registered\n",
220 fr_pool->size - i);
221}
222
223/*
224 * iser_create_ib_conn_res - Queue-Pair (QP)
225 *
226 * Return: 0 on success, -1 on failure
227 */
228static int iser_create_ib_conn_res(struct ib_conn *ib_conn)
229{
230 struct iser_conn *iser_conn = to_iser_conn(ib_conn);
231 struct iser_device *device;
232 struct ib_device *ib_dev;
233 struct ib_qp_init_attr init_attr;
234 int ret = -ENOMEM;
235 unsigned int max_send_wr, cq_size;
236
237 BUG_ON(ib_conn->device == NULL);
238
239 device = ib_conn->device;
240 ib_dev = device->ib_device;
241
242 /* +1 for drain */
243 if (ib_conn->pi_support)
244 max_send_wr = ISER_QP_SIG_MAX_REQ_DTOS + 1;
245 else
246 max_send_wr = ISER_QP_MAX_REQ_DTOS + 1;
247 max_send_wr = min_t(unsigned int, max_send_wr,
248 (unsigned int)ib_dev->attrs.max_qp_wr);
249
250 cq_size = max_send_wr + ISER_QP_MAX_RECV_DTOS;
251 ib_conn->cq = ib_cq_pool_get(ib_dev, cq_size, -1, IB_POLL_SOFTIRQ);
252 if (IS_ERR(ib_conn->cq)) {
253 ret = PTR_ERR(ib_conn->cq);
254 goto cq_err;
255 }
256 ib_conn->cq_size = cq_size;
257
258 memset(&init_attr, 0, sizeof(init_attr));
259
260 init_attr.event_handler = iser_qp_event_callback;
261 init_attr.qp_context = (void *)ib_conn;
262 init_attr.send_cq = ib_conn->cq;
263 init_attr.recv_cq = ib_conn->cq;
264 /* +1 for drain */
265 init_attr.cap.max_recv_wr = ISER_QP_MAX_RECV_DTOS + 1;
266 init_attr.cap.max_send_sge = 2;
267 init_attr.cap.max_recv_sge = 1;
268 init_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
269 init_attr.qp_type = IB_QPT_RC;
270 init_attr.cap.max_send_wr = max_send_wr;
271 if (ib_conn->pi_support)
272 init_attr.create_flags |= IB_QP_CREATE_INTEGRITY_EN;
273 iser_conn->max_cmds = ISER_GET_MAX_XMIT_CMDS(max_send_wr - 1);
274
275 ret = rdma_create_qp(ib_conn->cma_id, device->pd, &init_attr);
276 if (ret)
277 goto out_err;
278
279 ib_conn->qp = ib_conn->cma_id->qp;
280 iser_info("setting conn %p cma_id %p qp %p max_send_wr %d\n", ib_conn,
281 ib_conn->cma_id, ib_conn->cma_id->qp, max_send_wr);
282 return ret;
283
284out_err:
285 ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size);
286cq_err:
287 iser_err("unable to alloc mem or create resource, err %d\n", ret);
288
289 return ret;
290}
291
292/*
293 * based on the resolved device node GUID see if there already allocated
294 * device for this device. If there's no such, create one.
295 */
296static
297struct iser_device *iser_device_find_by_ib_device(struct rdma_cm_id *cma_id)
298{
299 struct iser_device *device;
300
301 mutex_lock(&ig.device_list_mutex);
302
303 list_for_each_entry(device, &ig.device_list, ig_list)
304 /* find if there's a match using the node GUID */
305 if (device->ib_device->node_guid == cma_id->device->node_guid)
306 goto inc_refcnt;
307
308 device = kzalloc(sizeof *device, GFP_KERNEL);
309 if (!device)
310 goto out;
311
312 /* assign this device to the device */
313 device->ib_device = cma_id->device;
314 /* init the device and link it into ig device list */
315 if (iser_create_device_ib_res(device)) {
316 kfree(device);
317 device = NULL;
318 goto out;
319 }
320 list_add(&device->ig_list, &ig.device_list);
321
322inc_refcnt:
323 device->refcount++;
324out:
325 mutex_unlock(&ig.device_list_mutex);
326 return device;
327}
328
329/* if there's no demand for this device, release it */
330static void iser_device_try_release(struct iser_device *device)
331{
332 mutex_lock(&ig.device_list_mutex);
333 device->refcount--;
334 iser_info("device %p refcount %d\n", device, device->refcount);
335 if (!device->refcount) {
336 iser_free_device_ib_res(device);
337 list_del(&device->ig_list);
338 kfree(device);
339 }
340 mutex_unlock(&ig.device_list_mutex);
341}
342
343void iser_release_work(struct work_struct *work)
344{
345 struct iser_conn *iser_conn;
346
347 iser_conn = container_of(work, struct iser_conn, release_work);
348
349 /* Wait for conn_stop to complete */
350 wait_for_completion(&iser_conn->stop_completion);
351 /* Wait for IB resouces cleanup to complete */
352 wait_for_completion(&iser_conn->ib_completion);
353
354 mutex_lock(&iser_conn->state_mutex);
355 iser_conn->state = ISER_CONN_DOWN;
356 mutex_unlock(&iser_conn->state_mutex);
357
358 iser_conn_release(iser_conn);
359}
360
361/**
362 * iser_free_ib_conn_res - release IB related resources
363 * @iser_conn: iser connection struct
364 * @destroy: indicator if we need to try to release the
365 * iser device and memory regoins pool (only iscsi
366 * shutdown and DEVICE_REMOVAL will use this).
367 *
368 * This routine is called with the iser state mutex held
369 * so the cm_id removal is out of here. It is Safe to
370 * be invoked multiple times.
371 */
372static void iser_free_ib_conn_res(struct iser_conn *iser_conn, bool destroy)
373{
374 struct ib_conn *ib_conn = &iser_conn->ib_conn;
375 struct iser_device *device = ib_conn->device;
376
377 iser_info("freeing conn %p cma_id %p qp %p\n",
378 iser_conn, ib_conn->cma_id, ib_conn->qp);
379
380 if (ib_conn->qp) {
381 rdma_destroy_qp(ib_conn->cma_id);
382 ib_cq_pool_put(ib_conn->cq, ib_conn->cq_size);
383 ib_conn->qp = NULL;
384 }
385
386 if (destroy) {
387 if (iser_conn->rx_descs)
388 iser_free_rx_descriptors(iser_conn);
389
390 if (device) {
391 iser_device_try_release(device);
392 ib_conn->device = NULL;
393 }
394 }
395}
396
397/**
398 * iser_conn_release - Frees all conn objects and deallocs conn descriptor
399 * @iser_conn: iSER connection context
400 */
401void iser_conn_release(struct iser_conn *iser_conn)
402{
403 struct ib_conn *ib_conn = &iser_conn->ib_conn;
404
405 mutex_lock(&ig.connlist_mutex);
406 list_del(&iser_conn->conn_list);
407 mutex_unlock(&ig.connlist_mutex);
408
409 mutex_lock(&iser_conn->state_mutex);
410 /* In case we endup here without ep_disconnect being invoked. */
411 if (iser_conn->state != ISER_CONN_DOWN) {
412 iser_warn("iser conn %p state %d, expected state down.\n",
413 iser_conn, iser_conn->state);
414 iscsi_destroy_endpoint(iser_conn->ep);
415 iser_conn->state = ISER_CONN_DOWN;
416 }
417 /*
418 * In case we never got to bind stage, we still need to
419 * release IB resources (which is safe to call more than once).
420 */
421 iser_free_ib_conn_res(iser_conn, true);
422 mutex_unlock(&iser_conn->state_mutex);
423
424 if (ib_conn->cma_id) {
425 rdma_destroy_id(ib_conn->cma_id);
426 ib_conn->cma_id = NULL;
427 }
428
429 kfree(iser_conn);
430}
431
432/**
433 * iser_conn_terminate - triggers start of the disconnect procedures and
434 * waits for them to be done
435 * @iser_conn: iSER connection context
436 *
437 * Called with state mutex held
438 */
439int iser_conn_terminate(struct iser_conn *iser_conn)
440{
441 struct ib_conn *ib_conn = &iser_conn->ib_conn;
442 int err = 0;
443
444 lockdep_assert_held(&iser_conn->state_mutex);
445
446 /* terminate the iser conn only if the conn state is UP */
447 if (iser_conn->state != ISER_CONN_UP)
448 return 0;
449
450 iser_conn->state = ISER_CONN_TERMINATING;
451 iser_info("iser_conn %p state %d\n", iser_conn, iser_conn->state);
452
453 /* suspend queuing of new iscsi commands */
454 if (iser_conn->iscsi_conn)
455 iscsi_suspend_queue(iser_conn->iscsi_conn);
456
457 /*
458 * In case we didn't already clean up the cma_id (peer initiated
459 * a disconnection), we need to Cause the CMA to change the QP
460 * state to ERROR.
461 */
462 if (ib_conn->cma_id) {
463 err = rdma_disconnect(ib_conn->cma_id);
464 if (err)
465 iser_err("Failed to disconnect, conn: 0x%p err %d\n",
466 iser_conn, err);
467
468 /* block until all flush errors are consumed */
469 ib_drain_qp(ib_conn->qp);
470 }
471
472 return 1;
473}
474
475/*
476 * Called with state mutex held
477 */
478static void iser_connect_error(struct rdma_cm_id *cma_id)
479{
480 struct iser_conn *iser_conn = cma_id->context;
481
482 lockdep_assert_held(&iser_conn->state_mutex);
483
484 iser_conn->state = ISER_CONN_TERMINATING;
485}
486
487static void iser_calc_scsi_params(struct iser_conn *iser_conn,
488 unsigned int max_sectors)
489{
490 struct iser_device *device = iser_conn->ib_conn.device;
491 struct ib_device_attr *attr = &device->ib_device->attrs;
492 unsigned short sg_tablesize, sup_sg_tablesize;
493 unsigned short reserved_mr_pages;
494 u32 max_num_sg;
495
496 /*
497 * FRs without SG_GAPS can only map up to a (device) page per entry,
498 * but if the first entry is misaligned we'll end up using two entries
499 * (head and tail) for a single page worth data, so one additional
500 * entry is required.
501 */
502 if (attr->kernel_cap_flags & IBK_SG_GAPS_REG)
503 reserved_mr_pages = 0;
504 else
505 reserved_mr_pages = 1;
506
507 if (iser_conn->ib_conn.pi_support)
508 max_num_sg = attr->max_pi_fast_reg_page_list_len;
509 else
510 max_num_sg = attr->max_fast_reg_page_list_len;
511
512 sg_tablesize = DIV_ROUND_UP(max_sectors * SECTOR_SIZE, SZ_4K);
513 sup_sg_tablesize = min_t(uint, ISCSI_ISER_MAX_SG_TABLESIZE,
514 max_num_sg - reserved_mr_pages);
515 iser_conn->scsi_sg_tablesize = min(sg_tablesize, sup_sg_tablesize);
516 iser_conn->pages_per_mr =
517 iser_conn->scsi_sg_tablesize + reserved_mr_pages;
518}
519
520/*
521 * Called with state mutex held
522 */
523static void iser_addr_handler(struct rdma_cm_id *cma_id)
524{
525 struct iser_conn *iser_conn = cma_id->context;
526 struct iser_device *device;
527 struct ib_conn *ib_conn;
528 int ret;
529
530 lockdep_assert_held(&iser_conn->state_mutex);
531
532 if (iser_conn->state != ISER_CONN_PENDING)
533 /* bailout */
534 return;
535
536 ib_conn = &iser_conn->ib_conn;
537 device = iser_device_find_by_ib_device(cma_id);
538 if (!device) {
539 iser_err("device lookup/creation failed\n");
540 iser_connect_error(cma_id);
541 return;
542 }
543
544 ib_conn->device = device;
545
546 /* connection T10-PI support */
547 if (iser_pi_enable) {
548 if (!(device->ib_device->attrs.kernel_cap_flags &
549 IBK_INTEGRITY_HANDOVER)) {
550 iser_warn("T10-PI requested but not supported on %s, "
551 "continue without T10-PI\n",
552 dev_name(&ib_conn->device->ib_device->dev));
553 ib_conn->pi_support = false;
554 } else {
555 ib_conn->pi_support = true;
556 }
557 }
558
559 iser_calc_scsi_params(iser_conn, iser_max_sectors);
560
561 ret = rdma_resolve_route(cma_id, 1000);
562 if (ret) {
563 iser_err("resolve route failed: %d\n", ret);
564 iser_connect_error(cma_id);
565 return;
566 }
567}
568
569/*
570 * Called with state mutex held
571 */
572static void iser_route_handler(struct rdma_cm_id *cma_id)
573{
574 struct rdma_conn_param conn_param;
575 int ret;
576 struct iser_cm_hdr req_hdr;
577 struct iser_conn *iser_conn = cma_id->context;
578 struct ib_conn *ib_conn = &iser_conn->ib_conn;
579 struct ib_device *ib_dev = ib_conn->device->ib_device;
580
581 lockdep_assert_held(&iser_conn->state_mutex);
582
583 if (iser_conn->state != ISER_CONN_PENDING)
584 /* bailout */
585 return;
586
587 ret = iser_create_ib_conn_res(ib_conn);
588 if (ret)
589 goto failure;
590
591 memset(&conn_param, 0, sizeof conn_param);
592 conn_param.responder_resources = ib_dev->attrs.max_qp_rd_atom;
593 conn_param.initiator_depth = 1;
594 conn_param.retry_count = 7;
595 conn_param.rnr_retry_count = 6;
596
597 memset(&req_hdr, 0, sizeof(req_hdr));
598 req_hdr.flags = ISER_ZBVA_NOT_SUP;
599 if (!iser_always_reg)
600 req_hdr.flags |= ISER_SEND_W_INV_NOT_SUP;
601 conn_param.private_data = (void *)&req_hdr;
602 conn_param.private_data_len = sizeof(struct iser_cm_hdr);
603
604 ret = rdma_connect_locked(cma_id, &conn_param);
605 if (ret) {
606 iser_err("failure connecting: %d\n", ret);
607 goto failure;
608 }
609
610 return;
611failure:
612 iser_connect_error(cma_id);
613}
614
615/*
616 * Called with state mutex held
617 */
618static void iser_connected_handler(struct rdma_cm_id *cma_id,
619 const void *private_data)
620{
621 struct iser_conn *iser_conn = cma_id->context;
622 struct ib_qp_attr attr;
623 struct ib_qp_init_attr init_attr;
624
625 lockdep_assert_held(&iser_conn->state_mutex);
626
627 if (iser_conn->state != ISER_CONN_PENDING)
628 /* bailout */
629 return;
630
631 (void)ib_query_qp(cma_id->qp, &attr, ~0, &init_attr);
632 iser_info("remote qpn:%x my qpn:%x\n", attr.dest_qp_num, cma_id->qp->qp_num);
633
634 if (private_data) {
635 u8 flags = *(u8 *)private_data;
636
637 iser_conn->snd_w_inv = !(flags & ISER_SEND_W_INV_NOT_SUP);
638 }
639
640 iser_info("conn %p: negotiated %s invalidation\n",
641 iser_conn, iser_conn->snd_w_inv ? "remote" : "local");
642
643 iser_conn->state = ISER_CONN_UP;
644 complete(&iser_conn->up_completion);
645}
646
647/*
648 * Called with state mutex held
649 */
650static void iser_cleanup_handler(struct rdma_cm_id *cma_id,
651 bool destroy)
652{
653 struct iser_conn *iser_conn = cma_id->context;
654
655 lockdep_assert_held(&iser_conn->state_mutex);
656 /*
657 * We are not guaranteed that we visited disconnected_handler
658 * by now, call it here to be safe that we handle CM drep
659 * and flush errors.
660 */
661 if (iser_conn_terminate(iser_conn)) {
662 if (iser_conn->iscsi_conn)
663 iscsi_conn_failure(iser_conn->iscsi_conn,
664 ISCSI_ERR_CONN_FAILED);
665 else
666 iser_err("iscsi_iser connection isn't bound\n");
667 }
668 iser_free_ib_conn_res(iser_conn, destroy);
669 complete(&iser_conn->ib_completion);
670}
671
672static int iser_cma_handler(struct rdma_cm_id *cma_id,
673 struct rdma_cm_event *event)
674{
675 struct iser_conn *iser_conn;
676 int ret = 0;
677
678 iser_conn = cma_id->context;
679 iser_info("%s (%d): status %d conn %p id %p\n",
680 rdma_event_msg(event->event), event->event,
681 event->status, cma_id->context, cma_id);
682
683 mutex_lock(&iser_conn->state_mutex);
684 switch (event->event) {
685 case RDMA_CM_EVENT_ADDR_RESOLVED:
686 iser_addr_handler(cma_id);
687 break;
688 case RDMA_CM_EVENT_ROUTE_RESOLVED:
689 iser_route_handler(cma_id);
690 break;
691 case RDMA_CM_EVENT_ESTABLISHED:
692 iser_connected_handler(cma_id, event->param.conn.private_data);
693 break;
694 case RDMA_CM_EVENT_REJECTED:
695 iser_info("Connection rejected: %s\n",
696 rdma_reject_msg(cma_id, event->status));
697 fallthrough;
698 case RDMA_CM_EVENT_ADDR_ERROR:
699 case RDMA_CM_EVENT_ROUTE_ERROR:
700 case RDMA_CM_EVENT_CONNECT_ERROR:
701 case RDMA_CM_EVENT_UNREACHABLE:
702 iser_connect_error(cma_id);
703 break;
704 case RDMA_CM_EVENT_DISCONNECTED:
705 case RDMA_CM_EVENT_ADDR_CHANGE:
706 case RDMA_CM_EVENT_TIMEWAIT_EXIT:
707 iser_cleanup_handler(cma_id, false);
708 break;
709 case RDMA_CM_EVENT_DEVICE_REMOVAL:
710 /*
711 * we *must* destroy the device as we cannot rely
712 * on iscsid to be around to initiate error handling.
713 * also if we are not in state DOWN implicitly destroy
714 * the cma_id.
715 */
716 iser_cleanup_handler(cma_id, true);
717 if (iser_conn->state != ISER_CONN_DOWN) {
718 iser_conn->ib_conn.cma_id = NULL;
719 ret = 1;
720 }
721 break;
722 default:
723 iser_err("Unexpected RDMA CM event: %s (%d)\n",
724 rdma_event_msg(event->event), event->event);
725 break;
726 }
727 mutex_unlock(&iser_conn->state_mutex);
728
729 return ret;
730}
731
732void iser_conn_init(struct iser_conn *iser_conn)
733{
734 struct ib_conn *ib_conn = &iser_conn->ib_conn;
735
736 iser_conn->state = ISER_CONN_INIT;
737 init_completion(&iser_conn->stop_completion);
738 init_completion(&iser_conn->ib_completion);
739 init_completion(&iser_conn->up_completion);
740 INIT_LIST_HEAD(&iser_conn->conn_list);
741 mutex_init(&iser_conn->state_mutex);
742
743 ib_conn->reg_cqe.done = iser_reg_comp;
744}
745
746/*
747 * starts the process of connecting to the target
748 * sleeps until the connection is established or rejected
749 */
750int iser_connect(struct iser_conn *iser_conn, struct sockaddr *src_addr,
751 struct sockaddr *dst_addr, int non_blocking)
752{
753 struct ib_conn *ib_conn = &iser_conn->ib_conn;
754 int err = 0;
755
756 mutex_lock(&iser_conn->state_mutex);
757
758 sprintf(iser_conn->name, "%pISp", dst_addr);
759
760 iser_info("connecting to: %s\n", iser_conn->name);
761
762 /* the device is known only --after-- address resolution */
763 ib_conn->device = NULL;
764
765 iser_conn->state = ISER_CONN_PENDING;
766
767 ib_conn->cma_id = rdma_create_id(&init_net, iser_cma_handler,
768 iser_conn, RDMA_PS_TCP, IB_QPT_RC);
769 if (IS_ERR(ib_conn->cma_id)) {
770 err = PTR_ERR(ib_conn->cma_id);
771 iser_err("rdma_create_id failed: %d\n", err);
772 goto id_failure;
773 }
774
775 err = rdma_resolve_addr(ib_conn->cma_id, src_addr, dst_addr, 1000);
776 if (err) {
777 iser_err("rdma_resolve_addr failed: %d\n", err);
778 goto addr_failure;
779 }
780
781 if (!non_blocking) {
782 wait_for_completion_interruptible(&iser_conn->up_completion);
783
784 if (iser_conn->state != ISER_CONN_UP) {
785 err = -EIO;
786 goto connect_failure;
787 }
788 }
789 mutex_unlock(&iser_conn->state_mutex);
790
791 mutex_lock(&ig.connlist_mutex);
792 list_add(&iser_conn->conn_list, &ig.connlist);
793 mutex_unlock(&ig.connlist_mutex);
794 return 0;
795
796id_failure:
797 ib_conn->cma_id = NULL;
798addr_failure:
799 iser_conn->state = ISER_CONN_DOWN;
800connect_failure:
801 mutex_unlock(&iser_conn->state_mutex);
802 iser_conn_release(iser_conn);
803 return err;
804}
805
806int iser_post_recvl(struct iser_conn *iser_conn)
807{
808 struct ib_conn *ib_conn = &iser_conn->ib_conn;
809 struct iser_login_desc *desc = &iser_conn->login_desc;
810 struct ib_recv_wr wr;
811 int ret;
812
813 desc->sge.addr = desc->rsp_dma;
814 desc->sge.length = ISER_RX_LOGIN_SIZE;
815 desc->sge.lkey = ib_conn->device->pd->local_dma_lkey;
816
817 desc->cqe.done = iser_login_rsp;
818 wr.wr_cqe = &desc->cqe;
819 wr.sg_list = &desc->sge;
820 wr.num_sge = 1;
821 wr.next = NULL;
822
823 ret = ib_post_recv(ib_conn->qp, &wr, NULL);
824 if (unlikely(ret))
825 iser_err("ib_post_recv login failed ret=%d\n", ret);
826
827 return ret;
828}
829
830int iser_post_recvm(struct iser_conn *iser_conn, struct iser_rx_desc *rx_desc)
831{
832 struct ib_conn *ib_conn = &iser_conn->ib_conn;
833 struct ib_recv_wr wr;
834 int ret;
835
836 rx_desc->cqe.done = iser_task_rsp;
837 wr.wr_cqe = &rx_desc->cqe;
838 wr.sg_list = &rx_desc->rx_sg;
839 wr.num_sge = 1;
840 wr.next = NULL;
841
842 ret = ib_post_recv(ib_conn->qp, &wr, NULL);
843 if (unlikely(ret))
844 iser_err("ib_post_recv failed ret=%d\n", ret);
845
846 return ret;
847}
848
849
850/**
851 * iser_post_send - Initiate a Send DTO operation
852 * @ib_conn: connection RDMA resources
853 * @tx_desc: iSER TX descriptor
854 *
855 * Return: 0 on success, -1 on failure
856 */
857int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc)
858{
859 struct ib_send_wr *wr = &tx_desc->send_wr;
860 struct ib_send_wr *first_wr;
861 int ret;
862
863 ib_dma_sync_single_for_device(ib_conn->device->ib_device,
864 tx_desc->dma_addr, ISER_HEADERS_LEN,
865 DMA_TO_DEVICE);
866
867 wr->next = NULL;
868 wr->wr_cqe = &tx_desc->cqe;
869 wr->sg_list = tx_desc->tx_sg;
870 wr->num_sge = tx_desc->num_sge;
871 wr->opcode = IB_WR_SEND;
872 wr->send_flags = IB_SEND_SIGNALED;
873
874 if (tx_desc->inv_wr.next)
875 first_wr = &tx_desc->inv_wr;
876 else if (tx_desc->reg_wr.wr.next)
877 first_wr = &tx_desc->reg_wr.wr;
878 else
879 first_wr = wr;
880
881 ret = ib_post_send(ib_conn->qp, first_wr, NULL);
882 if (unlikely(ret))
883 iser_err("ib_post_send failed, ret:%d opcode:%d\n",
884 ret, wr->opcode);
885
886 return ret;
887}
888
889u8 iser_check_task_pi_status(struct iscsi_iser_task *iser_task,
890 enum iser_data_dir cmd_dir, sector_t *sector)
891{
892 struct iser_mem_reg *reg = &iser_task->rdma_reg[cmd_dir];
893 struct iser_fr_desc *desc = reg->desc;
894 unsigned long sector_size = iser_task->sc->device->sector_size;
895 struct ib_mr_status mr_status;
896 int ret;
897
898 if (desc && desc->sig_protected) {
899 desc->sig_protected = false;
900 ret = ib_check_mr_status(desc->rsc.sig_mr,
901 IB_MR_CHECK_SIG_STATUS, &mr_status);
902 if (ret) {
903 iser_err("ib_check_mr_status failed, ret %d\n", ret);
904 /* Not a lot we can do, return ambiguous guard error */
905 *sector = 0;
906 return 0x1;
907 }
908
909 if (mr_status.fail_status & IB_MR_CHECK_SIG_STATUS) {
910 sector_t sector_off = mr_status.sig_err.sig_err_offset;
911
912 sector_div(sector_off, sector_size + 8);
913 *sector = scsi_get_sector(iser_task->sc) + sector_off;
914
915 iser_err("PI error found type %d at sector %llx "
916 "expected %x vs actual %x\n",
917 mr_status.sig_err.err_type,
918 (unsigned long long)*sector,
919 mr_status.sig_err.expected,
920 mr_status.sig_err.actual);
921
922 switch (mr_status.sig_err.err_type) {
923 case IB_SIG_BAD_GUARD:
924 return 0x1;
925 case IB_SIG_BAD_REFTAG:
926 return 0x3;
927 case IB_SIG_BAD_APPTAG:
928 return 0x2;
929 }
930 }
931 }
932
933 return 0;
934}
935
936void iser_err_comp(struct ib_wc *wc, const char *type)
937{
938 if (wc->status != IB_WC_WR_FLUSH_ERR) {
939 struct iser_conn *iser_conn = to_iser_conn(wc->qp->qp_context);
940
941 iser_err("%s failure: %s (%d) vend_err %#x\n", type,
942 ib_wc_status_msg(wc->status), wc->status,
943 wc->vendor_err);
944
945 if (iser_conn->iscsi_conn)
946 iscsi_conn_failure(iser_conn->iscsi_conn,
947 ISCSI_ERR_CONN_FAILED);
948 } else {
949 iser_dbg("%s failure: %s (%d)\n", type,
950 ib_wc_status_msg(wc->status), wc->status);
951 }
952}