1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2015-2018 Oracle. All rights reserved.
4 *
5 * Support for backward direction RPCs on RPC/RDMA (server-side).
6 */
7
8 #include <linux/sunrpc/svc_rdma.h>
9
10 #include "xprt_rdma.h"
11 #include <trace/events/rpcrdma.h>
12
13 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
14
15 #undef SVCRDMA_BACKCHANNEL_DEBUG
16
17 /**
18 * svc_rdma_handle_bc_reply - Process incoming backchannel reply
19 * @xprt: controlling backchannel transport
20 * @rdma_resp: pointer to incoming transport header
21 * @rcvbuf: XDR buffer into which to decode the reply
22 *
23 * Returns:
24 * %0 if @rcvbuf is filled in, xprt_complete_rqst called,
25 * %-EAGAIN if server should call ->recvfrom again.
26 */
svc_rdma_handle_bc_reply(struct rpc_xprt * xprt,__be32 * rdma_resp,struct xdr_buf * rcvbuf)27 int svc_rdma_handle_bc_reply(struct rpc_xprt *xprt, __be32 *rdma_resp,
28 struct xdr_buf *rcvbuf)
29 {
30 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
31 struct kvec *dst, *src = &rcvbuf->head[0];
32 struct rpc_rqst *req;
33 u32 credits;
34 size_t len;
35 __be32 xid;
36 __be32 *p;
37 int ret;
38
39 p = (__be32 *)src->iov_base;
40 len = src->iov_len;
41 xid = *rdma_resp;
42
43 #ifdef SVCRDMA_BACKCHANNEL_DEBUG
44 pr_info("%s: xid=%08x, length=%zu\n",
45 __func__, be32_to_cpu(xid), len);
46 pr_info("%s: RPC/RDMA: %*ph\n",
47 __func__, (int)RPCRDMA_HDRLEN_MIN, rdma_resp);
48 pr_info("%s: RPC: %*ph\n",
49 __func__, (int)len, p);
50 #endif
51
52 ret = -EAGAIN;
53 if (src->iov_len < 24)
54 goto out_shortreply;
55
56 spin_lock(&xprt->queue_lock);
57 req = xprt_lookup_rqst(xprt, xid);
58 if (!req)
59 goto out_notfound;
60
61 dst = &req->rq_private_buf.head[0];
62 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
63 if (dst->iov_len < len)
64 goto out_unlock;
65 memcpy(dst->iov_base, p, len);
66 xprt_pin_rqst(req);
67 spin_unlock(&xprt->queue_lock);
68
69 credits = be32_to_cpup(rdma_resp + 2);
70 if (credits == 0)
71 credits = 1; /* don't deadlock */
72 else if (credits > r_xprt->rx_buf.rb_bc_max_requests)
73 credits = r_xprt->rx_buf.rb_bc_max_requests;
74
75 spin_lock(&xprt->transport_lock);
76 xprt->cwnd = credits << RPC_CWNDSHIFT;
77 spin_unlock(&xprt->transport_lock);
78
79 spin_lock(&xprt->queue_lock);
80 ret = 0;
81 xprt_complete_rqst(req->rq_task, rcvbuf->len);
82 xprt_unpin_rqst(req);
83 rcvbuf->len = 0;
84
85 out_unlock:
86 spin_unlock(&xprt->queue_lock);
87 out:
88 return ret;
89
90 out_shortreply:
91 dprintk("svcrdma: short bc reply: xprt=%p, len=%zu\n",
92 xprt, src->iov_len);
93 goto out;
94
95 out_notfound:
96 dprintk("svcrdma: unrecognized bc reply: xprt=%p, xid=%08x\n",
97 xprt, be32_to_cpu(xid));
98 goto out_unlock;
99 }
100
101 /* Send a backwards direction RPC call.
102 *
103 * Caller holds the connection's mutex and has already marshaled
104 * the RPC/RDMA request.
105 *
106 * This is similar to svc_rdma_send_reply_msg, but takes a struct
107 * rpc_rqst instead, does not support chunks, and avoids blocking
108 * memory allocation.
109 *
110 * XXX: There is still an opportunity to block in svc_rdma_send()
111 * if there are no SQ entries to post the Send. This may occur if
112 * the adapter has a small maximum SQ depth.
113 */
svc_rdma_bc_sendto(struct svcxprt_rdma * rdma,struct rpc_rqst * rqst,struct svc_rdma_send_ctxt * ctxt)114 static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma,
115 struct rpc_rqst *rqst,
116 struct svc_rdma_send_ctxt *ctxt)
117 {
118 int ret;
119
120 ret = svc_rdma_map_reply_msg(rdma, ctxt, &rqst->rq_snd_buf, NULL);
121 if (ret < 0)
122 return -EIO;
123
124 /* Bump page refcnt so Send completion doesn't release
125 * the rq_buffer before all retransmits are complete.
126 */
127 get_page(virt_to_page(rqst->rq_buffer));
128 ctxt->sc_send_wr.opcode = IB_WR_SEND;
129 return svc_rdma_send(rdma, &ctxt->sc_send_wr);
130 }
131
132 /* Server-side transport endpoint wants a whole page for its send
133 * buffer. The client RPC code constructs the RPC header in this
134 * buffer before it invokes ->send_request.
135 */
136 static int
xprt_rdma_bc_allocate(struct rpc_task * task)137 xprt_rdma_bc_allocate(struct rpc_task *task)
138 {
139 struct rpc_rqst *rqst = task->tk_rqstp;
140 size_t size = rqst->rq_callsize;
141 struct page *page;
142
143 if (size > PAGE_SIZE) {
144 WARN_ONCE(1, "svcrdma: large bc buffer request (size %zu)\n",
145 size);
146 return -EINVAL;
147 }
148
149 page = alloc_page(RPCRDMA_DEF_GFP);
150 if (!page)
151 return -ENOMEM;
152 rqst->rq_buffer = page_address(page);
153
154 rqst->rq_rbuffer = kmalloc(rqst->rq_rcvsize, RPCRDMA_DEF_GFP);
155 if (!rqst->rq_rbuffer) {
156 put_page(page);
157 return -ENOMEM;
158 }
159 return 0;
160 }
161
162 static void
xprt_rdma_bc_free(struct rpc_task * task)163 xprt_rdma_bc_free(struct rpc_task *task)
164 {
165 struct rpc_rqst *rqst = task->tk_rqstp;
166
167 put_page(virt_to_page(rqst->rq_buffer));
168 kfree(rqst->rq_rbuffer);
169 }
170
171 static int
rpcrdma_bc_send_request(struct svcxprt_rdma * rdma,struct rpc_rqst * rqst)172 rpcrdma_bc_send_request(struct svcxprt_rdma *rdma, struct rpc_rqst *rqst)
173 {
174 struct rpc_xprt *xprt = rqst->rq_xprt;
175 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
176 struct svc_rdma_send_ctxt *ctxt;
177 __be32 *p;
178 int rc;
179
180 ctxt = svc_rdma_send_ctxt_get(rdma);
181 if (!ctxt)
182 goto drop_connection;
183
184 p = ctxt->sc_xprt_buf;
185 *p++ = rqst->rq_xid;
186 *p++ = rpcrdma_version;
187 *p++ = cpu_to_be32(r_xprt->rx_buf.rb_bc_max_requests);
188 *p++ = rdma_msg;
189 *p++ = xdr_zero;
190 *p++ = xdr_zero;
191 *p = xdr_zero;
192 svc_rdma_sync_reply_hdr(rdma, ctxt, RPCRDMA_HDRLEN_MIN);
193
194 #ifdef SVCRDMA_BACKCHANNEL_DEBUG
195 pr_info("%s: %*ph\n", __func__, 64, rqst->rq_buffer);
196 #endif
197
198 rqst->rq_xtime = ktime_get();
199 rc = svc_rdma_bc_sendto(rdma, rqst, ctxt);
200 if (rc) {
201 svc_rdma_send_ctxt_put(rdma, ctxt);
202 goto drop_connection;
203 }
204 return 0;
205
206 drop_connection:
207 dprintk("svcrdma: failed to send bc call\n");
208 return -ENOTCONN;
209 }
210
211 /* Send an RPC call on the passive end of a transport
212 * connection.
213 */
214 static int
xprt_rdma_bc_send_request(struct rpc_rqst * rqst)215 xprt_rdma_bc_send_request(struct rpc_rqst *rqst)
216 {
217 struct svc_xprt *sxprt = rqst->rq_xprt->bc_xprt;
218 struct svcxprt_rdma *rdma;
219 int ret;
220
221 dprintk("svcrdma: sending bc call with xid: %08x\n",
222 be32_to_cpu(rqst->rq_xid));
223
224 mutex_lock(&sxprt->xpt_mutex);
225
226 ret = -ENOTCONN;
227 rdma = container_of(sxprt, struct svcxprt_rdma, sc_xprt);
228 if (!test_bit(XPT_DEAD, &sxprt->xpt_flags)) {
229 ret = rpcrdma_bc_send_request(rdma, rqst);
230 if (ret == -ENOTCONN)
231 svc_close_xprt(sxprt);
232 }
233
234 mutex_unlock(&sxprt->xpt_mutex);
235
236 if (ret < 0)
237 return ret;
238 return 0;
239 }
240
241 static void
xprt_rdma_bc_close(struct rpc_xprt * xprt)242 xprt_rdma_bc_close(struct rpc_xprt *xprt)
243 {
244 dprintk("svcrdma: %s: xprt %p\n", __func__, xprt);
245 xprt->cwnd = RPC_CWNDSHIFT;
246 }
247
248 static void
xprt_rdma_bc_put(struct rpc_xprt * xprt)249 xprt_rdma_bc_put(struct rpc_xprt *xprt)
250 {
251 dprintk("svcrdma: %s: xprt %p\n", __func__, xprt);
252
253 xprt_free(xprt);
254 }
255
256 static const struct rpc_xprt_ops xprt_rdma_bc_procs = {
257 .reserve_xprt = xprt_reserve_xprt_cong,
258 .release_xprt = xprt_release_xprt_cong,
259 .alloc_slot = xprt_alloc_slot,
260 .free_slot = xprt_free_slot,
261 .release_request = xprt_release_rqst_cong,
262 .buf_alloc = xprt_rdma_bc_allocate,
263 .buf_free = xprt_rdma_bc_free,
264 .send_request = xprt_rdma_bc_send_request,
265 .wait_for_reply_request = xprt_wait_for_reply_request_def,
266 .close = xprt_rdma_bc_close,
267 .destroy = xprt_rdma_bc_put,
268 .print_stats = xprt_rdma_print_stats
269 };
270
271 static const struct rpc_timeout xprt_rdma_bc_timeout = {
272 .to_initval = 60 * HZ,
273 .to_maxval = 60 * HZ,
274 };
275
276 /* It shouldn't matter if the number of backchannel session slots
277 * doesn't match the number of RPC/RDMA credits. That just means
278 * one or the other will have extra slots that aren't used.
279 */
280 static struct rpc_xprt *
xprt_setup_rdma_bc(struct xprt_create * args)281 xprt_setup_rdma_bc(struct xprt_create *args)
282 {
283 struct rpc_xprt *xprt;
284 struct rpcrdma_xprt *new_xprt;
285
286 if (args->addrlen > sizeof(xprt->addr)) {
287 dprintk("RPC: %s: address too large\n", __func__);
288 return ERR_PTR(-EBADF);
289 }
290
291 xprt = xprt_alloc(args->net, sizeof(*new_xprt),
292 RPCRDMA_MAX_BC_REQUESTS,
293 RPCRDMA_MAX_BC_REQUESTS);
294 if (!xprt) {
295 dprintk("RPC: %s: couldn't allocate rpc_xprt\n",
296 __func__);
297 return ERR_PTR(-ENOMEM);
298 }
299
300 xprt->timeout = &xprt_rdma_bc_timeout;
301 xprt_set_bound(xprt);
302 xprt_set_connected(xprt);
303 xprt->bind_timeout = RPCRDMA_BIND_TO;
304 xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
305 xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO;
306
307 xprt->prot = XPRT_TRANSPORT_BC_RDMA;
308 xprt->ops = &xprt_rdma_bc_procs;
309
310 memcpy(&xprt->addr, args->dstaddr, args->addrlen);
311 xprt->addrlen = args->addrlen;
312 xprt_rdma_format_addresses(xprt, (struct sockaddr *)&xprt->addr);
313 xprt->resvport = 0;
314
315 xprt->max_payload = xprt_rdma_max_inline_read;
316
317 new_xprt = rpcx_to_rdmax(xprt);
318 new_xprt->rx_buf.rb_bc_max_requests = xprt->max_reqs;
319
320 xprt_get(xprt);
321 args->bc_xprt->xpt_bc_xprt = xprt;
322 xprt->bc_xprt = args->bc_xprt;
323
324 /* Final put for backchannel xprt is in __svc_rdma_free */
325 xprt_get(xprt);
326 return xprt;
327 }
328
329 struct xprt_class xprt_rdma_bc = {
330 .list = LIST_HEAD_INIT(xprt_rdma_bc.list),
331 .name = "rdma backchannel",
332 .owner = THIS_MODULE,
333 .ident = XPRT_TRANSPORT_BC_RDMA,
334 .setup = xprt_setup_rdma_bc,
335 };
336