2 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the BSD-type
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
14 * Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
17 * Redistributions in binary form must reproduce the above
18 * copyright notice, this list of conditions and the following
19 * disclaimer in the documentation and/or other materials provided
20 * with the distribution.
22 * Neither the name of the Network Appliance, Inc. nor the names of
23 * its contributors may be used to endorse or promote products
24 * derived from this software without specific prior written
27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
30 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
31 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
32 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
33 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
34 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
35 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
36 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
37 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
43 * Encapsulates the major functions managing:
50 #include <linux/interrupt.h>
51 #include <linux/slab.h>
52 #include <linux/prefetch.h>
53 #include <linux/sunrpc/addr.h>
54 #include <linux/sunrpc/svc_rdma.h>
55 #include <asm/bitops.h>
56 #include <linux/module.h> /* try_module_get()/module_put() */
58 #include "xprt_rdma.h"
64 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
65 # define RPCDBG_FACILITY RPCDBG_TRANS
72 static struct workqueue_struct *rpcrdma_receive_wq;
75 rpcrdma_alloc_wq(void)
77 struct workqueue_struct *recv_wq;
79 recv_wq = alloc_workqueue("xprtrdma_receive",
80 WQ_MEM_RECLAIM | WQ_UNBOUND | WQ_HIGHPRI,
85 rpcrdma_receive_wq = recv_wq;
90 rpcrdma_destroy_wq(void)
92 struct workqueue_struct *wq;
94 if (rpcrdma_receive_wq) {
95 wq = rpcrdma_receive_wq;
96 rpcrdma_receive_wq = NULL;
97 destroy_workqueue(wq);
102 rpcrdma_qp_async_error_upcall(struct ib_event *event, void *context)
104 struct rpcrdma_ep *ep = context;
106 pr_err("RPC: %s: %s on device %s ep %p\n",
107 __func__, ib_event_msg(event->event),
108 event->device->name, context);
109 if (ep->rep_connected == 1) {
110 ep->rep_connected = -EIO;
111 rpcrdma_conn_func(ep);
112 wake_up_all(&ep->rep_connect_wait);
117 * rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC
118 * @cq: completion queue (ignored)
123 rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
125 /* WARNING: Only wr_cqe and status are reliable at this point */
126 if (wc->status != IB_WC_SUCCESS && wc->status != IB_WC_WR_FLUSH_ERR)
127 pr_err("rpcrdma: Send: %s (%u/0x%x)\n",
128 ib_wc_status_msg(wc->status),
129 wc->status, wc->vendor_err);
132 /* Perform basic sanity checking to avoid using garbage
133 * to update the credit grant value.
136 rpcrdma_update_granted_credits(struct rpcrdma_rep *rep)
138 struct rpcrdma_msg *rmsgp = rdmab_to_msg(rep->rr_rdmabuf);
139 struct rpcrdma_buffer *buffer = &rep->rr_rxprt->rx_buf;
142 if (rep->rr_len < RPCRDMA_HDRLEN_ERR)
145 credits = be32_to_cpu(rmsgp->rm_credit);
147 credits = 1; /* don't deadlock */
148 else if (credits > buffer->rb_max_requests)
149 credits = buffer->rb_max_requests;
151 atomic_set(&buffer->rb_credits, credits);
155 * rpcrdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
156 * @cq: completion queue (ignored)
161 rpcrdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
163 struct ib_cqe *cqe = wc->wr_cqe;
164 struct rpcrdma_rep *rep = container_of(cqe, struct rpcrdma_rep,
167 /* WARNING: Only wr_id and status are reliable at this point */
168 if (wc->status != IB_WC_SUCCESS)
171 /* status == SUCCESS means all fields in wc are trustworthy */
172 if (wc->opcode != IB_WC_RECV)
175 dprintk("RPC: %s: rep %p opcode 'recv', length %u: success\n",
176 __func__, rep, wc->byte_len);
178 rep->rr_len = wc->byte_len;
179 rep->rr_wc_flags = wc->wc_flags;
180 rep->rr_inv_rkey = wc->ex.invalidate_rkey;
182 ib_dma_sync_single_for_cpu(rep->rr_device,
183 rdmab_addr(rep->rr_rdmabuf),
184 rep->rr_len, DMA_FROM_DEVICE);
186 rpcrdma_update_granted_credits(rep);
189 queue_work(rpcrdma_receive_wq, &rep->rr_work);
193 if (wc->status != IB_WC_WR_FLUSH_ERR)
194 pr_err("rpcrdma: Recv: %s (%u/0x%x)\n",
195 ib_wc_status_msg(wc->status),
196 wc->status, wc->vendor_err);
197 rep->rr_len = RPCRDMA_BAD_LEN;
202 rpcrdma_update_connect_private(struct rpcrdma_xprt *r_xprt,
203 struct rdma_conn_param *param)
205 struct rpcrdma_create_data_internal *cdata = &r_xprt->rx_data;
206 const struct rpcrdma_connect_private *pmsg = param->private_data;
207 unsigned int rsize, wsize;
209 /* Default settings for RPC-over-RDMA Version One */
210 r_xprt->rx_ia.ri_reminv_expected = false;
211 rsize = RPCRDMA_V1_DEF_INLINE_SIZE;
212 wsize = RPCRDMA_V1_DEF_INLINE_SIZE;
215 pmsg->cp_magic == rpcrdma_cmp_magic &&
216 pmsg->cp_version == RPCRDMA_CMP_VERSION) {
217 r_xprt->rx_ia.ri_reminv_expected = true;
218 rsize = rpcrdma_decode_buffer_size(pmsg->cp_send_size);
219 wsize = rpcrdma_decode_buffer_size(pmsg->cp_recv_size);
222 if (rsize < cdata->inline_rsize)
223 cdata->inline_rsize = rsize;
224 if (wsize < cdata->inline_wsize)
225 cdata->inline_wsize = wsize;
226 pr_info("rpcrdma: max send %u, max recv %u\n",
227 cdata->inline_wsize, cdata->inline_rsize);
228 rpcrdma_set_max_header_sizes(r_xprt);
232 rpcrdma_conn_upcall(struct rdma_cm_id *id, struct rdma_cm_event *event)
234 struct rpcrdma_xprt *xprt = id->context;
235 struct rpcrdma_ia *ia = &xprt->rx_ia;
236 struct rpcrdma_ep *ep = &xprt->rx_ep;
237 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
238 struct sockaddr *sap = (struct sockaddr *)&ep->rep_remote_addr;
240 struct ib_qp_attr *attr = &ia->ri_qp_attr;
241 struct ib_qp_init_attr *iattr = &ia->ri_qp_init_attr;
244 switch (event->event) {
245 case RDMA_CM_EVENT_ADDR_RESOLVED:
246 case RDMA_CM_EVENT_ROUTE_RESOLVED:
248 complete(&ia->ri_done);
250 case RDMA_CM_EVENT_ADDR_ERROR:
251 ia->ri_async_rc = -EHOSTUNREACH;
252 dprintk("RPC: %s: CM address resolution error, ep 0x%p\n",
254 complete(&ia->ri_done);
256 case RDMA_CM_EVENT_ROUTE_ERROR:
257 ia->ri_async_rc = -ENETUNREACH;
258 dprintk("RPC: %s: CM route resolution error, ep 0x%p\n",
260 complete(&ia->ri_done);
262 case RDMA_CM_EVENT_ESTABLISHED:
264 ib_query_qp(ia->ri_id->qp, attr,
265 IB_QP_MAX_QP_RD_ATOMIC | IB_QP_MAX_DEST_RD_ATOMIC,
267 dprintk("RPC: %s: %d responder resources"
269 __func__, attr->max_dest_rd_atomic,
270 attr->max_rd_atomic);
271 rpcrdma_update_connect_private(xprt, &event->param.conn);
273 case RDMA_CM_EVENT_CONNECT_ERROR:
274 connstate = -ENOTCONN;
276 case RDMA_CM_EVENT_UNREACHABLE:
277 connstate = -ENETDOWN;
279 case RDMA_CM_EVENT_REJECTED:
280 connstate = -ECONNREFUSED;
282 case RDMA_CM_EVENT_DISCONNECTED:
283 connstate = -ECONNABORTED;
285 case RDMA_CM_EVENT_DEVICE_REMOVAL:
288 dprintk("RPC: %s: %sconnected\n",
289 __func__, connstate > 0 ? "" : "dis");
290 atomic_set(&xprt->rx_buf.rb_credits, 1);
291 ep->rep_connected = connstate;
292 rpcrdma_conn_func(ep);
293 wake_up_all(&ep->rep_connect_wait);
296 dprintk("RPC: %s: %pIS:%u (ep 0x%p): %s\n",
297 __func__, sap, rpc_get_port(sap), ep,
298 rdma_event_msg(event->event));
302 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
303 if (connstate == 1) {
304 int ird = attr->max_dest_rd_atomic;
305 int tird = ep->rep_remote_cma.responder_resources;
307 pr_info("rpcrdma: connection to %pIS:%u on %s, memreg '%s', %d credits, %d responders%s\n",
308 sap, rpc_get_port(sap),
310 ia->ri_ops->ro_displayname,
311 xprt->rx_buf.rb_max_requests,
312 ird, ird < 4 && ird < tird / 2 ? " (low!)" : "");
313 } else if (connstate < 0) {
314 pr_info("rpcrdma: connection to %pIS:%u closed (%d)\n",
315 sap, rpc_get_port(sap), connstate);
322 static void rpcrdma_destroy_id(struct rdma_cm_id *id)
325 module_put(id->device->owner);
330 static struct rdma_cm_id *
331 rpcrdma_create_id(struct rpcrdma_xprt *xprt,
332 struct rpcrdma_ia *ia, struct sockaddr *addr)
334 struct rdma_cm_id *id;
337 init_completion(&ia->ri_done);
339 id = rdma_create_id(&init_net, rpcrdma_conn_upcall, xprt, RDMA_PS_TCP,
343 dprintk("RPC: %s: rdma_create_id() failed %i\n",
348 ia->ri_async_rc = -ETIMEDOUT;
349 rc = rdma_resolve_addr(id, NULL, addr, RDMA_RESOLVE_TIMEOUT);
351 dprintk("RPC: %s: rdma_resolve_addr() failed %i\n",
355 wait_for_completion_interruptible_timeout(&ia->ri_done,
356 msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1);
359 * Until xprtrdma supports DEVICE_REMOVAL, the provider must
360 * be pinned while there are active NFS/RDMA mounts to prevent
361 * hangs and crashes at umount time.
363 if (!ia->ri_async_rc && !try_module_get(id->device->owner)) {
364 dprintk("RPC: %s: Failed to get device module\n",
366 ia->ri_async_rc = -ENODEV;
368 rc = ia->ri_async_rc;
372 ia->ri_async_rc = -ETIMEDOUT;
373 rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT);
375 dprintk("RPC: %s: rdma_resolve_route() failed %i\n",
379 wait_for_completion_interruptible_timeout(&ia->ri_done,
380 msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1);
381 rc = ia->ri_async_rc;
387 module_put(id->device->owner);
394 * Exported functions.
398 * Open and initialize an Interface Adapter.
399 * o initializes fields of struct rpcrdma_ia, including
400 * interface and provider attributes and protection zone.
403 rpcrdma_ia_open(struct rpcrdma_xprt *xprt, struct sockaddr *addr, int memreg)
405 struct rpcrdma_ia *ia = &xprt->rx_ia;
408 ia->ri_id = rpcrdma_create_id(xprt, ia, addr);
409 if (IS_ERR(ia->ri_id)) {
410 rc = PTR_ERR(ia->ri_id);
413 ia->ri_device = ia->ri_id->device;
415 ia->ri_pd = ib_alloc_pd(ia->ri_device, 0);
416 if (IS_ERR(ia->ri_pd)) {
417 rc = PTR_ERR(ia->ri_pd);
418 pr_err("rpcrdma: ib_alloc_pd() returned %d\n", rc);
424 if (frwr_is_supported(ia)) {
425 ia->ri_ops = &rpcrdma_frwr_memreg_ops;
429 case RPCRDMA_MTHCAFMR:
430 if (fmr_is_supported(ia)) {
431 ia->ri_ops = &rpcrdma_fmr_memreg_ops;
436 pr_err("rpcrdma: Unsupported memory registration mode: %d\n",
445 ib_dealloc_pd(ia->ri_pd);
448 rpcrdma_destroy_id(ia->ri_id);
455 * Clean up/close an IA.
456 * o if event handles and PD have been initialized, free them.
460 rpcrdma_ia_close(struct rpcrdma_ia *ia)
462 dprintk("RPC: %s: entering\n", __func__);
463 if (ia->ri_id != NULL && !IS_ERR(ia->ri_id)) {
465 rdma_destroy_qp(ia->ri_id);
466 rpcrdma_destroy_id(ia->ri_id);
470 /* If the pd is still busy, xprtrdma missed freeing a resource */
471 if (ia->ri_pd && !IS_ERR(ia->ri_pd))
472 ib_dealloc_pd(ia->ri_pd);
476 * Create unconnected endpoint.
479 rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
480 struct rpcrdma_create_data_internal *cdata)
482 struct rpcrdma_connect_private *pmsg = &ep->rep_cm_private;
483 struct ib_cq *sendcq, *recvcq;
484 unsigned int max_qp_wr;
487 if (ia->ri_device->attrs.max_sge < RPCRDMA_MAX_SEND_SGES) {
488 dprintk("RPC: %s: insufficient sge's available\n",
493 if (ia->ri_device->attrs.max_qp_wr <= RPCRDMA_BACKWARD_WRS) {
494 dprintk("RPC: %s: insufficient wqe's available\n",
498 max_qp_wr = ia->ri_device->attrs.max_qp_wr - RPCRDMA_BACKWARD_WRS - 1;
500 /* check provider's send/recv wr limits */
501 if (cdata->max_requests > max_qp_wr)
502 cdata->max_requests = max_qp_wr;
504 ep->rep_attr.event_handler = rpcrdma_qp_async_error_upcall;
505 ep->rep_attr.qp_context = ep;
506 ep->rep_attr.srq = NULL;
507 ep->rep_attr.cap.max_send_wr = cdata->max_requests;
508 ep->rep_attr.cap.max_send_wr += RPCRDMA_BACKWARD_WRS;
509 ep->rep_attr.cap.max_send_wr += 1; /* drain cqe */
510 rc = ia->ri_ops->ro_open(ia, ep, cdata);
513 ep->rep_attr.cap.max_recv_wr = cdata->max_requests;
514 ep->rep_attr.cap.max_recv_wr += RPCRDMA_BACKWARD_WRS;
515 ep->rep_attr.cap.max_recv_wr += 1; /* drain cqe */
516 ep->rep_attr.cap.max_send_sge = RPCRDMA_MAX_SEND_SGES;
517 ep->rep_attr.cap.max_recv_sge = 1;
518 ep->rep_attr.cap.max_inline_data = 0;
519 ep->rep_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
520 ep->rep_attr.qp_type = IB_QPT_RC;
521 ep->rep_attr.port_num = ~0;
523 dprintk("RPC: %s: requested max: dtos: send %d recv %d; "
524 "iovs: send %d recv %d\n",
526 ep->rep_attr.cap.max_send_wr,
527 ep->rep_attr.cap.max_recv_wr,
528 ep->rep_attr.cap.max_send_sge,
529 ep->rep_attr.cap.max_recv_sge);
531 /* set trigger for requesting send completion */
532 ep->rep_cqinit = ep->rep_attr.cap.max_send_wr/2 - 1;
533 if (ep->rep_cqinit <= 2)
534 ep->rep_cqinit = 0; /* always signal? */
536 init_waitqueue_head(&ep->rep_connect_wait);
537 INIT_DELAYED_WORK(&ep->rep_connect_worker, rpcrdma_connect_worker);
539 sendcq = ib_alloc_cq(ia->ri_device, NULL,
540 ep->rep_attr.cap.max_send_wr + 1,
542 if (IS_ERR(sendcq)) {
543 rc = PTR_ERR(sendcq);
544 dprintk("RPC: %s: failed to create send CQ: %i\n",
549 recvcq = ib_alloc_cq(ia->ri_device, NULL,
550 ep->rep_attr.cap.max_recv_wr + 1,
552 if (IS_ERR(recvcq)) {
553 rc = PTR_ERR(recvcq);
554 dprintk("RPC: %s: failed to create recv CQ: %i\n",
559 ep->rep_attr.send_cq = sendcq;
560 ep->rep_attr.recv_cq = recvcq;
562 /* Initialize cma parameters */
563 memset(&ep->rep_remote_cma, 0, sizeof(ep->rep_remote_cma));
565 /* Prepare RDMA-CM private message */
566 pmsg->cp_magic = rpcrdma_cmp_magic;
567 pmsg->cp_version = RPCRDMA_CMP_VERSION;
568 pmsg->cp_flags |= ia->ri_ops->ro_send_w_inv_ok;
569 pmsg->cp_send_size = rpcrdma_encode_buffer_size(cdata->inline_wsize);
570 pmsg->cp_recv_size = rpcrdma_encode_buffer_size(cdata->inline_rsize);
571 ep->rep_remote_cma.private_data = pmsg;
572 ep->rep_remote_cma.private_data_len = sizeof(*pmsg);
574 /* Client offers RDMA Read but does not initiate */
575 ep->rep_remote_cma.initiator_depth = 0;
576 if (ia->ri_device->attrs.max_qp_rd_atom > 32) /* arbitrary but <= 255 */
577 ep->rep_remote_cma.responder_resources = 32;
579 ep->rep_remote_cma.responder_resources =
580 ia->ri_device->attrs.max_qp_rd_atom;
582 /* Limit transport retries so client can detect server
583 * GID changes quickly. RPC layer handles re-establishing
584 * transport connection and retransmission.
586 ep->rep_remote_cma.retry_count = 6;
588 /* RPC-over-RDMA handles its own flow control. In addition,
589 * make all RNR NAKs visible so we know that RPC-over-RDMA
590 * flow control is working correctly (no NAKs should be seen).
592 ep->rep_remote_cma.flow_control = 0;
593 ep->rep_remote_cma.rnr_retry_count = 0;
606 * Disconnect and destroy endpoint. After this, the only
607 * valid operations on the ep are to free it (if dynamically
608 * allocated) or re-create it.
611 rpcrdma_ep_destroy(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
613 dprintk("RPC: %s: entering, connected is %d\n",
614 __func__, ep->rep_connected);
616 cancel_delayed_work_sync(&ep->rep_connect_worker);
619 rpcrdma_ep_disconnect(ep, ia);
620 rdma_destroy_qp(ia->ri_id);
621 ia->ri_id->qp = NULL;
624 ib_free_cq(ep->rep_attr.recv_cq);
625 ib_free_cq(ep->rep_attr.send_cq);
629 * Connect unconnected endpoint.
632 rpcrdma_ep_connect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
634 struct rdma_cm_id *id, *old;
638 if (ep->rep_connected != 0) {
639 struct rpcrdma_xprt *xprt;
641 dprintk("RPC: %s: reconnecting...\n", __func__);
643 rpcrdma_ep_disconnect(ep, ia);
645 xprt = container_of(ia, struct rpcrdma_xprt, rx_ia);
646 id = rpcrdma_create_id(xprt, ia,
647 (struct sockaddr *)&xprt->rx_data.addr);
652 /* TEMP TEMP TEMP - fail if new device:
653 * Deregister/remarshal *all* requests!
654 * Close and recreate adapter, pd, etc!
655 * Re-determine all attributes still sane!
656 * More stuff I haven't thought of!
659 if (ia->ri_device != id->device) {
660 printk("RPC: %s: can't reconnect on "
661 "different device!\n", __func__);
662 rpcrdma_destroy_id(id);
667 rc = rdma_create_qp(id, ia->ri_pd, &ep->rep_attr);
669 dprintk("RPC: %s: rdma_create_qp failed %i\n",
671 rpcrdma_destroy_id(id);
679 rdma_destroy_qp(old);
680 rpcrdma_destroy_id(old);
682 dprintk("RPC: %s: connecting...\n", __func__);
683 rc = rdma_create_qp(ia->ri_id, ia->ri_pd, &ep->rep_attr);
685 dprintk("RPC: %s: rdma_create_qp failed %i\n",
687 /* do not update ep->rep_connected */
692 ep->rep_connected = 0;
694 rc = rdma_connect(ia->ri_id, &ep->rep_remote_cma);
696 dprintk("RPC: %s: rdma_connect() failed with %i\n",
701 wait_event_interruptible(ep->rep_connect_wait, ep->rep_connected != 0);
704 * Check state. A non-peer reject indicates no listener
705 * (ECONNREFUSED), which may be a transient state. All
706 * others indicate a transport condition which has already
707 * undergone a best-effort.
709 if (ep->rep_connected == -ECONNREFUSED &&
710 ++retry_count <= RDMA_CONNECT_RETRY_MAX) {
711 dprintk("RPC: %s: non-peer_reject, retry\n", __func__);
714 if (ep->rep_connected <= 0) {
715 /* Sometimes, the only way to reliably connect to remote
716 * CMs is to use same nonzero values for ORD and IRD. */
717 if (retry_count++ <= RDMA_CONNECT_RETRY_MAX + 1 &&
718 (ep->rep_remote_cma.responder_resources == 0 ||
719 ep->rep_remote_cma.initiator_depth !=
720 ep->rep_remote_cma.responder_resources)) {
721 if (ep->rep_remote_cma.responder_resources == 0)
722 ep->rep_remote_cma.responder_resources = 1;
723 ep->rep_remote_cma.initiator_depth =
724 ep->rep_remote_cma.responder_resources;
727 rc = ep->rep_connected;
729 struct rpcrdma_xprt *r_xprt;
732 dprintk("RPC: %s: connected\n", __func__);
734 r_xprt = container_of(ia, struct rpcrdma_xprt, rx_ia);
735 extras = r_xprt->rx_buf.rb_bc_srv_max_requests;
738 rc = rpcrdma_ep_post_extra_recv(r_xprt, extras);
740 pr_warn("%s: rpcrdma_ep_post_extra_recv: %i\n",
749 ep->rep_connected = rc;
754 * rpcrdma_ep_disconnect
756 * This is separate from destroy to facilitate the ability
757 * to reconnect without recreating the endpoint.
759 * This call is not reentrant, and must not be made in parallel
760 * on the same endpoint.
763 rpcrdma_ep_disconnect(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia)
767 rc = rdma_disconnect(ia->ri_id);
769 /* returns without wait if not connected */
770 wait_event_interruptible(ep->rep_connect_wait,
771 ep->rep_connected != 1);
772 dprintk("RPC: %s: after wait, %sconnected\n", __func__,
773 (ep->rep_connected == 1) ? "still " : "dis");
775 dprintk("RPC: %s: rdma_disconnect %i\n", __func__, rc);
776 ep->rep_connected = rc;
779 ib_drain_qp(ia->ri_id->qp);
783 rpcrdma_mr_recovery_worker(struct work_struct *work)
785 struct rpcrdma_buffer *buf = container_of(work, struct rpcrdma_buffer,
786 rb_recovery_worker.work);
787 struct rpcrdma_mw *mw;
789 spin_lock(&buf->rb_recovery_lock);
790 while (!list_empty(&buf->rb_stale_mrs)) {
791 mw = list_first_entry(&buf->rb_stale_mrs,
792 struct rpcrdma_mw, mw_list);
793 list_del_init(&mw->mw_list);
794 spin_unlock(&buf->rb_recovery_lock);
796 dprintk("RPC: %s: recovering MR %p\n", __func__, mw);
797 mw->mw_xprt->rx_ia.ri_ops->ro_recover_mr(mw);
799 spin_lock(&buf->rb_recovery_lock);
801 spin_unlock(&buf->rb_recovery_lock);
805 rpcrdma_defer_mr_recovery(struct rpcrdma_mw *mw)
807 struct rpcrdma_xprt *r_xprt = mw->mw_xprt;
808 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
810 spin_lock(&buf->rb_recovery_lock);
811 list_add(&mw->mw_list, &buf->rb_stale_mrs);
812 spin_unlock(&buf->rb_recovery_lock);
814 schedule_delayed_work(&buf->rb_recovery_worker, 0);
818 rpcrdma_create_mrs(struct rpcrdma_xprt *r_xprt)
820 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
821 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
826 for (count = 0; count < 32; count++) {
827 struct rpcrdma_mw *mw;
830 mw = kzalloc(sizeof(*mw), GFP_KERNEL);
834 rc = ia->ri_ops->ro_init_mr(ia, mw);
840 mw->mw_xprt = r_xprt;
842 list_add(&mw->mw_list, &free);
843 list_add(&mw->mw_all, &all);
846 spin_lock(&buf->rb_mwlock);
847 list_splice(&free, &buf->rb_mws);
848 list_splice(&all, &buf->rb_all);
849 r_xprt->rx_stats.mrs_allocated += count;
850 spin_unlock(&buf->rb_mwlock);
852 dprintk("RPC: %s: created %u MRs\n", __func__, count);
856 rpcrdma_mr_refresh_worker(struct work_struct *work)
858 struct rpcrdma_buffer *buf = container_of(work, struct rpcrdma_buffer,
859 rb_refresh_worker.work);
860 struct rpcrdma_xprt *r_xprt = container_of(buf, struct rpcrdma_xprt,
863 rpcrdma_create_mrs(r_xprt);
867 rpcrdma_create_req(struct rpcrdma_xprt *r_xprt)
869 struct rpcrdma_buffer *buffer = &r_xprt->rx_buf;
870 struct rpcrdma_req *req;
872 req = kzalloc(sizeof(*req), GFP_KERNEL);
874 return ERR_PTR(-ENOMEM);
876 INIT_LIST_HEAD(&req->rl_free);
877 spin_lock(&buffer->rb_reqslock);
878 list_add(&req->rl_all, &buffer->rb_allreqs);
879 spin_unlock(&buffer->rb_reqslock);
880 req->rl_cqe.done = rpcrdma_wc_send;
881 req->rl_buffer = &r_xprt->rx_buf;
882 INIT_LIST_HEAD(&req->rl_registered);
883 req->rl_send_wr.next = NULL;
884 req->rl_send_wr.wr_cqe = &req->rl_cqe;
885 req->rl_send_wr.sg_list = req->rl_send_sge;
886 req->rl_send_wr.opcode = IB_WR_SEND;
891 rpcrdma_create_rep(struct rpcrdma_xprt *r_xprt)
893 struct rpcrdma_create_data_internal *cdata = &r_xprt->rx_data;
894 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
895 struct rpcrdma_rep *rep;
899 rep = kzalloc(sizeof(*rep), GFP_KERNEL);
903 rep->rr_rdmabuf = rpcrdma_alloc_regbuf(cdata->inline_rsize,
904 DMA_FROM_DEVICE, GFP_KERNEL);
905 if (IS_ERR(rep->rr_rdmabuf)) {
906 rc = PTR_ERR(rep->rr_rdmabuf);
910 rep->rr_device = ia->ri_device;
911 rep->rr_cqe.done = rpcrdma_wc_receive;
912 rep->rr_rxprt = r_xprt;
913 INIT_WORK(&rep->rr_work, rpcrdma_reply_handler);
914 rep->rr_recv_wr.next = NULL;
915 rep->rr_recv_wr.wr_cqe = &rep->rr_cqe;
916 rep->rr_recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov;
917 rep->rr_recv_wr.num_sge = 1;
927 rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt)
929 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
932 buf->rb_max_requests = r_xprt->rx_data.max_requests;
933 buf->rb_bc_srv_max_requests = 0;
934 atomic_set(&buf->rb_credits, 1);
935 spin_lock_init(&buf->rb_mwlock);
936 spin_lock_init(&buf->rb_lock);
937 spin_lock_init(&buf->rb_recovery_lock);
938 INIT_LIST_HEAD(&buf->rb_mws);
939 INIT_LIST_HEAD(&buf->rb_all);
940 INIT_LIST_HEAD(&buf->rb_stale_mrs);
941 INIT_DELAYED_WORK(&buf->rb_refresh_worker,
942 rpcrdma_mr_refresh_worker);
943 INIT_DELAYED_WORK(&buf->rb_recovery_worker,
944 rpcrdma_mr_recovery_worker);
946 rpcrdma_create_mrs(r_xprt);
948 INIT_LIST_HEAD(&buf->rb_send_bufs);
949 INIT_LIST_HEAD(&buf->rb_allreqs);
950 spin_lock_init(&buf->rb_reqslock);
951 for (i = 0; i < buf->rb_max_requests; i++) {
952 struct rpcrdma_req *req;
954 req = rpcrdma_create_req(r_xprt);
956 dprintk("RPC: %s: request buffer %d alloc"
957 " failed\n", __func__, i);
961 req->rl_backchannel = false;
962 list_add(&req->rl_free, &buf->rb_send_bufs);
965 INIT_LIST_HEAD(&buf->rb_recv_bufs);
966 for (i = 0; i < buf->rb_max_requests + RPCRDMA_MAX_BC_REQUESTS; i++) {
967 struct rpcrdma_rep *rep;
969 rep = rpcrdma_create_rep(r_xprt);
971 dprintk("RPC: %s: reply buffer %d alloc failed\n",
976 list_add(&rep->rr_list, &buf->rb_recv_bufs);
981 rpcrdma_buffer_destroy(buf);
985 static struct rpcrdma_req *
986 rpcrdma_buffer_get_req_locked(struct rpcrdma_buffer *buf)
988 struct rpcrdma_req *req;
990 req = list_first_entry(&buf->rb_send_bufs,
991 struct rpcrdma_req, rl_free);
992 list_del(&req->rl_free);
996 static struct rpcrdma_rep *
997 rpcrdma_buffer_get_rep_locked(struct rpcrdma_buffer *buf)
999 struct rpcrdma_rep *rep;
1001 rep = list_first_entry(&buf->rb_recv_bufs,
1002 struct rpcrdma_rep, rr_list);
1003 list_del(&rep->rr_list);
1008 rpcrdma_destroy_rep(struct rpcrdma_rep *rep)
1010 rpcrdma_free_regbuf(rep->rr_rdmabuf);
1015 rpcrdma_destroy_req(struct rpcrdma_req *req)
1017 rpcrdma_free_regbuf(req->rl_recvbuf);
1018 rpcrdma_free_regbuf(req->rl_sendbuf);
1019 rpcrdma_free_regbuf(req->rl_rdmabuf);
1024 rpcrdma_destroy_mrs(struct rpcrdma_buffer *buf)
1026 struct rpcrdma_xprt *r_xprt = container_of(buf, struct rpcrdma_xprt,
1028 struct rpcrdma_ia *ia = rdmab_to_ia(buf);
1029 struct rpcrdma_mw *mw;
1033 spin_lock(&buf->rb_mwlock);
1034 while (!list_empty(&buf->rb_all)) {
1035 mw = list_entry(buf->rb_all.next, struct rpcrdma_mw, mw_all);
1036 list_del(&mw->mw_all);
1038 spin_unlock(&buf->rb_mwlock);
1039 ia->ri_ops->ro_release_mr(mw);
1041 spin_lock(&buf->rb_mwlock);
1043 spin_unlock(&buf->rb_mwlock);
1044 r_xprt->rx_stats.mrs_allocated = 0;
1046 dprintk("RPC: %s: released %u MRs\n", __func__, count);
1050 rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf)
1052 cancel_delayed_work_sync(&buf->rb_recovery_worker);
1054 while (!list_empty(&buf->rb_recv_bufs)) {
1055 struct rpcrdma_rep *rep;
1057 rep = rpcrdma_buffer_get_rep_locked(buf);
1058 rpcrdma_destroy_rep(rep);
1060 buf->rb_send_count = 0;
1062 spin_lock(&buf->rb_reqslock);
1063 while (!list_empty(&buf->rb_allreqs)) {
1064 struct rpcrdma_req *req;
1066 req = list_first_entry(&buf->rb_allreqs,
1067 struct rpcrdma_req, rl_all);
1068 list_del(&req->rl_all);
1070 spin_unlock(&buf->rb_reqslock);
1071 rpcrdma_destroy_req(req);
1072 spin_lock(&buf->rb_reqslock);
1074 spin_unlock(&buf->rb_reqslock);
1075 buf->rb_recv_count = 0;
1077 rpcrdma_destroy_mrs(buf);
1081 rpcrdma_get_mw(struct rpcrdma_xprt *r_xprt)
1083 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1084 struct rpcrdma_mw *mw = NULL;
1086 spin_lock(&buf->rb_mwlock);
1087 if (!list_empty(&buf->rb_mws)) {
1088 mw = list_first_entry(&buf->rb_mws,
1089 struct rpcrdma_mw, mw_list);
1090 list_del_init(&mw->mw_list);
1092 spin_unlock(&buf->rb_mwlock);
1099 dprintk("RPC: %s: no MWs available\n", __func__);
1100 schedule_delayed_work(&buf->rb_refresh_worker, 0);
1102 /* Allow the reply handler and refresh worker to run */
1109 rpcrdma_put_mw(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mw *mw)
1111 struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
1113 spin_lock(&buf->rb_mwlock);
1114 list_add_tail(&mw->mw_list, &buf->rb_mws);
1115 spin_unlock(&buf->rb_mwlock);
1118 static struct rpcrdma_rep *
1119 rpcrdma_buffer_get_rep(struct rpcrdma_buffer *buffers)
1121 /* If an RPC previously completed without a reply (say, a
1122 * credential problem or a soft timeout occurs) then hold off
1123 * on supplying more Receive buffers until the number of new
1124 * pending RPCs catches up to the number of posted Receives.
1126 if (unlikely(buffers->rb_send_count < buffers->rb_recv_count))
1129 if (unlikely(list_empty(&buffers->rb_recv_bufs)))
1131 buffers->rb_recv_count++;
1132 return rpcrdma_buffer_get_rep_locked(buffers);
1136 * Get a set of request/reply buffers.
1138 * Reply buffer (if available) is attached to send buffer upon return.
1140 struct rpcrdma_req *
1141 rpcrdma_buffer_get(struct rpcrdma_buffer *buffers)
1143 struct rpcrdma_req *req;
1145 spin_lock(&buffers->rb_lock);
1146 if (list_empty(&buffers->rb_send_bufs))
1148 buffers->rb_send_count++;
1149 req = rpcrdma_buffer_get_req_locked(buffers);
1150 req->rl_reply = rpcrdma_buffer_get_rep(buffers);
1151 spin_unlock(&buffers->rb_lock);
1155 spin_unlock(&buffers->rb_lock);
1156 pr_warn("RPC: %s: out of request buffers\n", __func__);
1161 * Put request/reply buffers back into pool.
1162 * Pre-decrement counter/array index.
1165 rpcrdma_buffer_put(struct rpcrdma_req *req)
1167 struct rpcrdma_buffer *buffers = req->rl_buffer;
1168 struct rpcrdma_rep *rep = req->rl_reply;
1170 req->rl_send_wr.num_sge = 0;
1171 req->rl_reply = NULL;
1173 spin_lock(&buffers->rb_lock);
1174 buffers->rb_send_count--;
1175 list_add_tail(&req->rl_free, &buffers->rb_send_bufs);
1177 buffers->rb_recv_count--;
1178 list_add_tail(&rep->rr_list, &buffers->rb_recv_bufs);
1180 spin_unlock(&buffers->rb_lock);
1184 * Recover reply buffers from pool.
1185 * This happens when recovering from disconnect.
1188 rpcrdma_recv_buffer_get(struct rpcrdma_req *req)
1190 struct rpcrdma_buffer *buffers = req->rl_buffer;
1192 spin_lock(&buffers->rb_lock);
1193 req->rl_reply = rpcrdma_buffer_get_rep(buffers);
1194 spin_unlock(&buffers->rb_lock);
1198 * Put reply buffers back into pool when not attached to
1199 * request. This happens in error conditions.
1202 rpcrdma_recv_buffer_put(struct rpcrdma_rep *rep)
1204 struct rpcrdma_buffer *buffers = &rep->rr_rxprt->rx_buf;
1206 spin_lock(&buffers->rb_lock);
1207 buffers->rb_recv_count--;
1208 list_add_tail(&rep->rr_list, &buffers->rb_recv_bufs);
1209 spin_unlock(&buffers->rb_lock);
1213 * rpcrdma_alloc_regbuf - allocate and DMA-map memory for SEND/RECV buffers
1214 * @size: size of buffer to be allocated, in bytes
1215 * @direction: direction of data movement
1218 * Returns an ERR_PTR, or a pointer to a regbuf, a buffer that
1219 * can be persistently DMA-mapped for I/O.
1221 * xprtrdma uses a regbuf for posting an outgoing RDMA SEND, or for
1222 * receiving the payload of RDMA RECV operations. During Long Calls
1223 * or Replies they may be registered externally via ro_map.
1225 struct rpcrdma_regbuf *
1226 rpcrdma_alloc_regbuf(size_t size, enum dma_data_direction direction,
1229 struct rpcrdma_regbuf *rb;
1231 rb = kmalloc(sizeof(*rb) + size, flags);
1233 return ERR_PTR(-ENOMEM);
1235 rb->rg_device = NULL;
1236 rb->rg_direction = direction;
1237 rb->rg_iov.length = size;
1243 * __rpcrdma_map_regbuf - DMA-map a regbuf
1244 * @ia: controlling rpcrdma_ia
1245 * @rb: regbuf to be mapped
1248 __rpcrdma_dma_map_regbuf(struct rpcrdma_ia *ia, struct rpcrdma_regbuf *rb)
1250 if (rb->rg_direction == DMA_NONE)
1253 rb->rg_iov.addr = ib_dma_map_single(ia->ri_device,
1254 (void *)rb->rg_base,
1257 if (ib_dma_mapping_error(ia->ri_device, rdmab_addr(rb)))
1260 rb->rg_device = ia->ri_device;
1261 rb->rg_iov.lkey = ia->ri_pd->local_dma_lkey;
1266 rpcrdma_dma_unmap_regbuf(struct rpcrdma_regbuf *rb)
1268 if (!rpcrdma_regbuf_is_mapped(rb))
1271 ib_dma_unmap_single(rb->rg_device, rdmab_addr(rb),
1272 rdmab_length(rb), rb->rg_direction);
1273 rb->rg_device = NULL;
1277 * rpcrdma_free_regbuf - deregister and free registered buffer
1278 * @rb: regbuf to be deregistered and freed
1281 rpcrdma_free_regbuf(struct rpcrdma_regbuf *rb)
1286 rpcrdma_dma_unmap_regbuf(rb);
1291 * Prepost any receive buffer, then post send.
1293 * Receive buffer is donated to hardware, reclaimed upon recv completion.
1296 rpcrdma_ep_post(struct rpcrdma_ia *ia,
1297 struct rpcrdma_ep *ep,
1298 struct rpcrdma_req *req)
1300 struct ib_send_wr *send_wr = &req->rl_send_wr;
1301 struct ib_send_wr *send_wr_fail;
1304 if (req->rl_reply) {
1305 rc = rpcrdma_ep_post_recv(ia, req->rl_reply);
1308 req->rl_reply = NULL;
1311 dprintk("RPC: %s: posting %d s/g entries\n",
1312 __func__, send_wr->num_sge);
1314 if (DECR_CQCOUNT(ep) > 0)
1315 send_wr->send_flags = 0;
1316 else { /* Provider must take a send completion every now and then */
1318 send_wr->send_flags = IB_SEND_SIGNALED;
1321 rc = ib_post_send(ia->ri_id->qp, send_wr, &send_wr_fail);
1323 goto out_postsend_err;
1327 pr_err("rpcrdma: RDMA Send ib_post_send returned %i\n", rc);
1332 rpcrdma_ep_post_recv(struct rpcrdma_ia *ia,
1333 struct rpcrdma_rep *rep)
1335 struct ib_recv_wr *recv_wr_fail;
1338 if (!rpcrdma_dma_map_regbuf(ia, rep->rr_rdmabuf))
1340 rc = ib_post_recv(ia->ri_id->qp, &rep->rr_recv_wr, &recv_wr_fail);
1346 pr_err("rpcrdma: failed to DMA map the Receive buffer\n");
1350 pr_err("rpcrdma: ib_post_recv returned %i\n", rc);
1355 * rpcrdma_ep_post_extra_recv - Post buffers for incoming backchannel requests
1356 * @r_xprt: transport associated with these backchannel resources
1357 * @min_reqs: minimum number of incoming requests expected
1359 * Returns zero if all requested buffers were posted, or a negative errno.
1362 rpcrdma_ep_post_extra_recv(struct rpcrdma_xprt *r_xprt, unsigned int count)
1364 struct rpcrdma_buffer *buffers = &r_xprt->rx_buf;
1365 struct rpcrdma_ia *ia = &r_xprt->rx_ia;
1366 struct rpcrdma_rep *rep;
1370 spin_lock(&buffers->rb_lock);
1371 if (list_empty(&buffers->rb_recv_bufs))
1373 rep = rpcrdma_buffer_get_rep_locked(buffers);
1374 spin_unlock(&buffers->rb_lock);
1376 rc = rpcrdma_ep_post_recv(ia, rep);
1384 spin_unlock(&buffers->rb_lock);
1385 pr_warn("%s: no extra receive buffers\n", __func__);
1389 rpcrdma_recv_buffer_put(rep);