1 /* 2 * Copyright (c) 2015 Oracle. All rights reserved. 3 * 4 * Support for backward direction RPCs on RPC/RDMA (server-side). 5 */ 6 7 #include <linux/sunrpc/svc_rdma.h> 8 #include "xprt_rdma.h" 9 10 #define RPCDBG_FACILITY RPCDBG_SVCXPRT 11 12 #undef SVCRDMA_BACKCHANNEL_DEBUG 13 14 int svc_rdma_handle_bc_reply(struct rpc_xprt *xprt, struct rpcrdma_msg *rmsgp, 15 struct xdr_buf *rcvbuf) 16 { 17 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); 18 struct kvec *dst, *src = &rcvbuf->head[0]; 19 struct rpc_rqst *req; 20 unsigned long cwnd; 21 u32 credits; 22 size_t len; 23 __be32 xid; 24 __be32 *p; 25 int ret; 26 27 p = (__be32 *)src->iov_base; 28 len = src->iov_len; 29 xid = rmsgp->rm_xid; 30 31 #ifdef SVCRDMA_BACKCHANNEL_DEBUG 32 pr_info("%s: xid=%08x, length=%zu\n", 33 __func__, be32_to_cpu(xid), len); 34 pr_info("%s: RPC/RDMA: %*ph\n", 35 __func__, (int)RPCRDMA_HDRLEN_MIN, rmsgp); 36 pr_info("%s: RPC: %*ph\n", 37 __func__, (int)len, p); 38 #endif 39 40 ret = -EAGAIN; 41 if (src->iov_len < 24) 42 goto out_shortreply; 43 44 spin_lock_bh(&xprt->transport_lock); 45 req = xprt_lookup_rqst(xprt, xid); 46 if (!req) 47 goto out_notfound; 48 49 dst = &req->rq_private_buf.head[0]; 50 memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf)); 51 if (dst->iov_len < len) 52 goto out_unlock; 53 memcpy(dst->iov_base, p, len); 54 55 credits = be32_to_cpu(rmsgp->rm_credit); 56 if (credits == 0) 57 credits = 1; /* don't deadlock */ 58 else if (credits > r_xprt->rx_buf.rb_bc_max_requests) 59 credits = r_xprt->rx_buf.rb_bc_max_requests; 60 61 cwnd = xprt->cwnd; 62 xprt->cwnd = credits << RPC_CWNDSHIFT; 63 if (xprt->cwnd > cwnd) 64 xprt_release_rqst_cong(req->rq_task); 65 66 ret = 0; 67 xprt_complete_rqst(req->rq_task, rcvbuf->len); 68 rcvbuf->len = 0; 69 70 out_unlock: 71 spin_unlock_bh(&xprt->transport_lock); 72 out: 73 return ret; 74 75 out_shortreply: 76 dprintk("svcrdma: short bc reply: xprt=%p, len=%zu\n", 77 xprt, src->iov_len); 78 goto out; 79 80 out_notfound: 81 dprintk("svcrdma: unrecognized bc reply: xprt=%p, xid=%08x\n", 82 xprt, be32_to_cpu(xid)); 83 84 goto out_unlock; 85 } 86 87 /* Send a backwards direction RPC call. 88 * 89 * Caller holds the connection's mutex and has already marshaled 90 * the RPC/RDMA request. 91 * 92 * This is similar to svc_rdma_reply, but takes an rpc_rqst 93 * instead, does not support chunks, and avoids blocking memory 94 * allocation. 95 * 96 * XXX: There is still an opportunity to block in svc_rdma_send() 97 * if there are no SQ entries to post the Send. This may occur if 98 * the adapter has a small maximum SQ depth. 99 */ 100 static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma, 101 struct rpc_rqst *rqst) 102 { 103 struct xdr_buf *sndbuf = &rqst->rq_snd_buf; 104 struct svc_rdma_op_ctxt *ctxt; 105 struct svc_rdma_req_map *vec; 106 struct ib_send_wr send_wr; 107 int ret; 108 109 vec = svc_rdma_get_req_map(rdma); 110 ret = svc_rdma_map_xdr(rdma, sndbuf, vec, false); 111 if (ret) 112 goto out_err; 113 114 ret = svc_rdma_repost_recv(rdma, GFP_NOIO); 115 if (ret) 116 goto out_err; 117 118 ctxt = svc_rdma_get_context(rdma); 119 ctxt->pages[0] = virt_to_page(rqst->rq_buffer); 120 ctxt->count = 1; 121 122 ctxt->direction = DMA_TO_DEVICE; 123 ctxt->sge[0].lkey = rdma->sc_pd->local_dma_lkey; 124 ctxt->sge[0].length = sndbuf->len; 125 ctxt->sge[0].addr = 126 ib_dma_map_page(rdma->sc_cm_id->device, ctxt->pages[0], 0, 127 sndbuf->len, DMA_TO_DEVICE); 128 if (ib_dma_mapping_error(rdma->sc_cm_id->device, ctxt->sge[0].addr)) { 129 ret = -EIO; 130 goto out_unmap; 131 } 132 atomic_inc(&rdma->sc_dma_used); 133 134 memset(&send_wr, 0, sizeof(send_wr)); 135 ctxt->cqe.done = svc_rdma_wc_send; 136 send_wr.wr_cqe = &ctxt->cqe; 137 send_wr.sg_list = ctxt->sge; 138 send_wr.num_sge = 1; 139 send_wr.opcode = IB_WR_SEND; 140 send_wr.send_flags = IB_SEND_SIGNALED; 141 142 ret = svc_rdma_send(rdma, &send_wr); 143 if (ret) { 144 ret = -EIO; 145 goto out_unmap; 146 } 147 148 out_err: 149 svc_rdma_put_req_map(rdma, vec); 150 dprintk("svcrdma: %s returns %d\n", __func__, ret); 151 return ret; 152 153 out_unmap: 154 svc_rdma_unmap_dma(ctxt); 155 svc_rdma_put_context(ctxt, 1); 156 goto out_err; 157 } 158 159 /* Server-side transport endpoint wants a whole page for its send 160 * buffer. The client RPC code constructs the RPC header in this 161 * buffer before it invokes ->send_request. 162 * 163 * Returns NULL if there was a temporary allocation failure. 164 */ 165 static void * 166 xprt_rdma_bc_allocate(struct rpc_task *task, size_t size) 167 { 168 struct rpc_rqst *rqst = task->tk_rqstp; 169 struct svc_xprt *sxprt = rqst->rq_xprt->bc_xprt; 170 struct svcxprt_rdma *rdma; 171 struct page *page; 172 173 rdma = container_of(sxprt, struct svcxprt_rdma, sc_xprt); 174 175 /* Prevent an infinite loop: try to make this case work */ 176 if (size > PAGE_SIZE) 177 WARN_ONCE(1, "svcrdma: large bc buffer request (size %zu)\n", 178 size); 179 180 page = alloc_page(RPCRDMA_DEF_GFP); 181 if (!page) 182 return NULL; 183 184 return page_address(page); 185 } 186 187 static void 188 xprt_rdma_bc_free(void *buffer) 189 { 190 /* No-op: ctxt and page have already been freed. */ 191 } 192 193 static int 194 rpcrdma_bc_send_request(struct svcxprt_rdma *rdma, struct rpc_rqst *rqst) 195 { 196 struct rpc_xprt *xprt = rqst->rq_xprt; 197 struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt); 198 struct rpcrdma_msg *headerp = (struct rpcrdma_msg *)rqst->rq_buffer; 199 int rc; 200 201 /* Space in the send buffer for an RPC/RDMA header is reserved 202 * via xprt->tsh_size. 203 */ 204 headerp->rm_xid = rqst->rq_xid; 205 headerp->rm_vers = rpcrdma_version; 206 headerp->rm_credit = cpu_to_be32(r_xprt->rx_buf.rb_bc_max_requests); 207 headerp->rm_type = rdma_msg; 208 headerp->rm_body.rm_chunks[0] = xdr_zero; 209 headerp->rm_body.rm_chunks[1] = xdr_zero; 210 headerp->rm_body.rm_chunks[2] = xdr_zero; 211 212 #ifdef SVCRDMA_BACKCHANNEL_DEBUG 213 pr_info("%s: %*ph\n", __func__, 64, rqst->rq_buffer); 214 #endif 215 216 rc = svc_rdma_bc_sendto(rdma, rqst); 217 if (rc) 218 goto drop_connection; 219 return rc; 220 221 drop_connection: 222 dprintk("svcrdma: failed to send bc call\n"); 223 xprt_disconnect_done(xprt); 224 return -ENOTCONN; 225 } 226 227 /* Send an RPC call on the passive end of a transport 228 * connection. 229 */ 230 static int 231 xprt_rdma_bc_send_request(struct rpc_task *task) 232 { 233 struct rpc_rqst *rqst = task->tk_rqstp; 234 struct svc_xprt *sxprt = rqst->rq_xprt->bc_xprt; 235 struct svcxprt_rdma *rdma; 236 int ret; 237 238 dprintk("svcrdma: sending bc call with xid: %08x\n", 239 be32_to_cpu(rqst->rq_xid)); 240 241 if (!mutex_trylock(&sxprt->xpt_mutex)) { 242 rpc_sleep_on(&sxprt->xpt_bc_pending, task, NULL); 243 if (!mutex_trylock(&sxprt->xpt_mutex)) 244 return -EAGAIN; 245 rpc_wake_up_queued_task(&sxprt->xpt_bc_pending, task); 246 } 247 248 ret = -ENOTCONN; 249 rdma = container_of(sxprt, struct svcxprt_rdma, sc_xprt); 250 if (!test_bit(XPT_DEAD, &sxprt->xpt_flags)) 251 ret = rpcrdma_bc_send_request(rdma, rqst); 252 253 mutex_unlock(&sxprt->xpt_mutex); 254 255 if (ret < 0) 256 return ret; 257 return 0; 258 } 259 260 static void 261 xprt_rdma_bc_close(struct rpc_xprt *xprt) 262 { 263 dprintk("svcrdma: %s: xprt %p\n", __func__, xprt); 264 } 265 266 static void 267 xprt_rdma_bc_put(struct rpc_xprt *xprt) 268 { 269 dprintk("svcrdma: %s: xprt %p\n", __func__, xprt); 270 271 xprt_free(xprt); 272 module_put(THIS_MODULE); 273 } 274 275 static struct rpc_xprt_ops xprt_rdma_bc_procs = { 276 .reserve_xprt = xprt_reserve_xprt_cong, 277 .release_xprt = xprt_release_xprt_cong, 278 .alloc_slot = xprt_alloc_slot, 279 .release_request = xprt_release_rqst_cong, 280 .buf_alloc = xprt_rdma_bc_allocate, 281 .buf_free = xprt_rdma_bc_free, 282 .send_request = xprt_rdma_bc_send_request, 283 .set_retrans_timeout = xprt_set_retrans_timeout_def, 284 .close = xprt_rdma_bc_close, 285 .destroy = xprt_rdma_bc_put, 286 .print_stats = xprt_rdma_print_stats 287 }; 288 289 static const struct rpc_timeout xprt_rdma_bc_timeout = { 290 .to_initval = 60 * HZ, 291 .to_maxval = 60 * HZ, 292 }; 293 294 /* It shouldn't matter if the number of backchannel session slots 295 * doesn't match the number of RPC/RDMA credits. That just means 296 * one or the other will have extra slots that aren't used. 297 */ 298 static struct rpc_xprt * 299 xprt_setup_rdma_bc(struct xprt_create *args) 300 { 301 struct rpc_xprt *xprt; 302 struct rpcrdma_xprt *new_xprt; 303 304 if (args->addrlen > sizeof(xprt->addr)) { 305 dprintk("RPC: %s: address too large\n", __func__); 306 return ERR_PTR(-EBADF); 307 } 308 309 xprt = xprt_alloc(args->net, sizeof(*new_xprt), 310 RPCRDMA_MAX_BC_REQUESTS, 311 RPCRDMA_MAX_BC_REQUESTS); 312 if (!xprt) { 313 dprintk("RPC: %s: couldn't allocate rpc_xprt\n", 314 __func__); 315 return ERR_PTR(-ENOMEM); 316 } 317 318 xprt->timeout = &xprt_rdma_bc_timeout; 319 xprt_set_bound(xprt); 320 xprt_set_connected(xprt); 321 xprt->bind_timeout = RPCRDMA_BIND_TO; 322 xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO; 323 xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO; 324 325 xprt->prot = XPRT_TRANSPORT_BC_RDMA; 326 xprt->tsh_size = RPCRDMA_HDRLEN_MIN / sizeof(__be32); 327 xprt->ops = &xprt_rdma_bc_procs; 328 329 memcpy(&xprt->addr, args->dstaddr, args->addrlen); 330 xprt->addrlen = args->addrlen; 331 xprt_rdma_format_addresses(xprt, (struct sockaddr *)&xprt->addr); 332 xprt->resvport = 0; 333 334 xprt->max_payload = xprt_rdma_max_inline_read; 335 336 new_xprt = rpcx_to_rdmax(xprt); 337 new_xprt->rx_buf.rb_bc_max_requests = xprt->max_reqs; 338 339 xprt_get(xprt); 340 args->bc_xprt->xpt_bc_xprt = xprt; 341 xprt->bc_xprt = args->bc_xprt; 342 343 if (!try_module_get(THIS_MODULE)) 344 goto out_fail; 345 346 /* Final put for backchannel xprt is in __svc_rdma_free */ 347 xprt_get(xprt); 348 return xprt; 349 350 out_fail: 351 xprt_rdma_free_addresses(xprt); 352 args->bc_xprt->xpt_bc_xprt = NULL; 353 xprt_put(xprt); 354 xprt_free(xprt); 355 return ERR_PTR(-EINVAL); 356 } 357 358 struct xprt_class xprt_rdma_bc = { 359 .list = LIST_HEAD_INIT(xprt_rdma_bc.list), 360 .name = "rdma backchannel", 361 .owner = THIS_MODULE, 362 .ident = XPRT_TRANSPORT_BC_RDMA, 363 .setup = xprt_setup_rdma_bc, 364 }; 365