diff options
Diffstat (limited to 'net/sunrpc/xprtrdma')
-rw-r--r-- | net/sunrpc/xprtrdma/backchannel.c | 2 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/frwr_ops.c | 13 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/rpc_rdma.c | 14 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_backchannel.c | 3 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 23 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_rw.c | 61 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_sendto.c | 69 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_transport.c | 6 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/verbs.c | 15 | ||||
-rw-r--r-- | net/sunrpc/xprtrdma/xprt_rdma.h | 1 |
10 files changed, 144 insertions, 63 deletions
diff --git a/net/sunrpc/xprtrdma/backchannel.c b/net/sunrpc/xprtrdma/backchannel.c index ce986591f213..63b000824654 100644 --- a/net/sunrpc/xprtrdma/backchannel.c +++ b/net/sunrpc/xprtrdma/backchannel.c @@ -158,6 +158,7 @@ void xprt_rdma_bc_free_rqst(struct rpc_rqst *rqst) spin_lock(&xprt->bc_pa_lock); list_add_tail(&rqst->rq_bc_pa_list, &xprt->bc_pa_list); spin_unlock(&xprt->bc_pa_lock); + xprt_put(xprt); } static struct rpc_rqst *rpcrdma_bc_rqst_get(struct rpcrdma_xprt *r_xprt) @@ -254,6 +255,7 @@ void rpcrdma_bc_receive_call(struct rpcrdma_xprt *r_xprt, /* Queue rqst for ULP's callback service */ bc_serv = xprt->bc_serv; + xprt_get(xprt); spin_lock(&bc_serv->sv_cb_lock); list_add(&rqst->rq_bc_list, &bc_serv->sv_cb_list); spin_unlock(&bc_serv->sv_cb_lock); diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c index 794ba4ca0994..7b1694c0a77e 100644 --- a/net/sunrpc/xprtrdma/frwr_ops.c +++ b/net/sunrpc/xprtrdma/frwr_ops.c @@ -381,8 +381,8 @@ struct rpcrdma_mr_seg *frwr_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_frwr *frwr; struct rpcrdma_mr *mr; struct ib_mr *ibmr; + int i, n, dma_nents; struct ib_reg_wr *reg_wr; - int i, n; u8 key; mr = NULL; @@ -417,15 +417,16 @@ struct rpcrdma_mr_seg *frwr_map(struct rpcrdma_xprt *r_xprt, break; } mr->mr_dir = rpcrdma_data_dir(writing); + mr->mr_nents = i; - mr->mr_nents = - ib_dma_map_sg(ia->ri_id->device, mr->mr_sg, i, mr->mr_dir); - if (!mr->mr_nents) + dma_nents = ib_dma_map_sg(ia->ri_id->device, mr->mr_sg, mr->mr_nents, + mr->mr_dir); + if (!dma_nents) goto out_dmamap_err; ibmr = frwr->fr_mr; - n = ib_map_mr_sg(ibmr, mr->mr_sg, mr->mr_nents, NULL, PAGE_SIZE); - if (unlikely(n != mr->mr_nents)) + n = ib_map_mr_sg(ibmr, mr->mr_sg, dma_nents, NULL, PAGE_SIZE); + if (n != dma_nents) goto out_mapmr_err; ibmr->iova &= 0x00000000ffffffff; diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c index 85115a2e2639..77a911d14015 100644 --- a/net/sunrpc/xprtrdma/rpc_rdma.c +++ b/net/sunrpc/xprtrdma/rpc_rdma.c @@ -71,7 +71,7 @@ static unsigned int rpcrdma_max_call_header_size(unsigned int maxsegs) size = RPCRDMA_HDRLEN_MIN; /* Maximum Read list size */ - size = maxsegs * rpcrdma_readchunk_maxsz * sizeof(__be32); + size += maxsegs * rpcrdma_readchunk_maxsz * sizeof(__be32); /* Minimal Read chunk size */ size += sizeof(__be32); /* segment count */ @@ -96,7 +96,7 @@ static unsigned int rpcrdma_max_reply_header_size(unsigned int maxsegs) size = RPCRDMA_HDRLEN_MIN; /* Maximum Write list size */ - size = sizeof(__be32); /* segment count */ + size += sizeof(__be32); /* segment count */ size += maxsegs * rpcrdma_segment_maxsz * sizeof(__be32); size += sizeof(__be32); /* list discriminator */ @@ -1219,8 +1219,7 @@ rpcrdma_decode_error(struct rpcrdma_xprt *r_xprt, struct rpcrdma_rep *rep, be32_to_cpup(p), be32_to_cpu(rep->rr_xid)); } - r_xprt->rx_stats.bad_reply_count++; - return -EREMOTEIO; + return -EIO; } /* Perform XID lookup, reconstruction of the RPC reply, and @@ -1259,13 +1258,11 @@ out: spin_unlock(&xprt->queue_lock); return; -/* If the incoming reply terminated a pending RPC, the next - * RPC call will post a replacement receive buffer as it is - * being marshaled. - */ out_badheader: trace_xprtrdma_reply_hdr(rep); r_xprt->rx_stats.bad_reply_count++; + rqst->rq_task->tk_status = status; + status = 0; goto out; } @@ -1365,6 +1362,7 @@ void rpcrdma_reply_handler(struct rpcrdma_rep *rep) xprt->cwnd = credits << RPC_CWNDSHIFT; spin_unlock_bh(&xprt->transport_lock); } + rpcrdma_post_recvs(r_xprt, false); req = rpcr_to_rdmar(rqst); if (req->rl_reply) { diff --git a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c index bed57d8b5c19..6436b65f11fb 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c +++ b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c @@ -195,6 +195,7 @@ rpcrdma_bc_send_request(struct svcxprt_rdma *rdma, struct rpc_rqst *rqst) pr_info("%s: %*ph\n", __func__, 64, rqst->rq_buffer); #endif + rqst->rq_xtime = ktime_get(); rc = svc_rdma_bc_sendto(rdma, rqst, ctxt); if (rc) { svc_rdma_send_ctxt_put(rdma, ctxt); @@ -241,6 +242,8 @@ static void xprt_rdma_bc_close(struct rpc_xprt *xprt) { dprintk("svcrdma: %s: xprt %p\n", __func__, xprt); + + xprt_disconnect_done(xprt); xprt->cwnd = RPC_CWNDSHIFT; } diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c index 65e2fb9aac65..6566620912d6 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c @@ -193,6 +193,7 @@ svc_rdma_recv_ctxt_get(struct svcxprt_rdma *rdma) out: ctxt->rc_page_count = 0; + ctxt->rc_read_payload_length = 0; return ctxt; out_empty: @@ -226,6 +227,26 @@ void svc_rdma_recv_ctxt_put(struct svcxprt_rdma *rdma, svc_rdma_recv_ctxt_destroy(rdma, ctxt); } +/** + * svc_rdma_release_rqst - Release transport-specific per-rqst resources + * @rqstp: svc_rqst being released + * + * Ensure that the recv_ctxt is released whether or not a Reply + * was sent. For example, the client could close the connection, + * or svc_process could drop an RPC, before the Reply is sent. + */ +void svc_rdma_release_rqst(struct svc_rqst *rqstp) +{ + struct svc_rdma_recv_ctxt *ctxt = rqstp->rq_xprt_ctxt; + struct svc_xprt *xprt = rqstp->rq_xprt; + struct svcxprt_rdma *rdma = + container_of(xprt, struct svcxprt_rdma, sc_xprt); + + rqstp->rq_xprt_ctxt = NULL; + if (ctxt) + svc_rdma_recv_ctxt_put(rdma, ctxt); +} + static int __svc_rdma_post_recv(struct svcxprt_rdma *rdma, struct svc_rdma_recv_ctxt *ctxt) { @@ -760,6 +781,8 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp) __be32 *p; int ret; + rqstp->rq_xprt_ctxt = NULL; + spin_lock(&rdma_xprt->sc_rq_dto_lock); ctxt = svc_rdma_next_recv_ctxt(&rdma_xprt->sc_read_complete_q); if (ctxt) { diff --git a/net/sunrpc/xprtrdma/svc_rdma_rw.c b/net/sunrpc/xprtrdma/svc_rdma_rw.c index 2121c9b4d275..04ec2a83af69 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_rw.c +++ b/net/sunrpc/xprtrdma/svc_rdma_rw.c @@ -322,8 +322,6 @@ static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc) if (atomic_sub_return(cc->cc_sqecount, &rdma->sc_sq_avail) > 0) { ret = ib_post_send(rdma->sc_qp, first_wr, &bad_wr); - trace_svcrdma_post_rw(&cc->cc_cqe, - cc->cc_sqecount, ret); if (ret) break; return 0; @@ -336,6 +334,7 @@ static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc) trace_svcrdma_sq_retry(rdma); } while (1); + trace_svcrdma_sq_post_err(rdma, ret); set_bit(XPT_CLOSE, &xprt->xpt_flags); /* If even one was posted, there will be a completion. */ @@ -481,18 +480,19 @@ static int svc_rdma_send_xdr_kvec(struct svc_rdma_write_info *info, vec->iov_len); } -/* Send an xdr_buf's page list by itself. A Write chunk is - * just the page list. a Reply chunk is the head, page list, - * and tail. This function is shared between the two types - * of chunk. +/* Send an xdr_buf's page list by itself. A Write chunk is just + * the page list. A Reply chunk is @xdr's head, page list, and + * tail. This function is shared between the two types of chunk. */ static int svc_rdma_send_xdr_pagelist(struct svc_rdma_write_info *info, - struct xdr_buf *xdr) + struct xdr_buf *xdr, + unsigned int offset, + unsigned long length) { info->wi_xdr = xdr; - info->wi_next_off = 0; + info->wi_next_off = offset - xdr->head[0].iov_len; return svc_rdma_build_writes(info, svc_rdma_pagelist_to_sg, - xdr->page_len); + length); } /** @@ -500,6 +500,8 @@ static int svc_rdma_send_xdr_pagelist(struct svc_rdma_write_info *info, * @rdma: controlling RDMA transport * @wr_ch: Write chunk provided by client * @xdr: xdr_buf containing the data payload + * @offset: payload's byte offset in @xdr + * @length: size of payload, in bytes * * Returns a non-negative number of bytes the chunk consumed, or * %-E2BIG if the payload was larger than the Write chunk, @@ -509,19 +511,20 @@ static int svc_rdma_send_xdr_pagelist(struct svc_rdma_write_info *info, * %-EIO if rdma_rw initialization failed (DMA mapping, etc). */ int svc_rdma_send_write_chunk(struct svcxprt_rdma *rdma, __be32 *wr_ch, - struct xdr_buf *xdr) + struct xdr_buf *xdr, + unsigned int offset, unsigned long length) { struct svc_rdma_write_info *info; int ret; - if (!xdr->page_len) + if (!length) return 0; info = svc_rdma_write_info_alloc(rdma, wr_ch); if (!info) return -ENOMEM; - ret = svc_rdma_send_xdr_pagelist(info, xdr); + ret = svc_rdma_send_xdr_pagelist(info, xdr, offset, length); if (ret < 0) goto out_err; @@ -530,7 +533,7 @@ int svc_rdma_send_write_chunk(struct svcxprt_rdma *rdma, __be32 *wr_ch, goto out_err; trace_svcrdma_encode_write(xdr->page_len); - return xdr->page_len; + return length; out_err: svc_rdma_write_info_free(info); @@ -570,7 +573,9 @@ int svc_rdma_send_reply_chunk(struct svcxprt_rdma *rdma, __be32 *rp_ch, * client did not provide Write chunks. */ if (!writelist && xdr->page_len) { - ret = svc_rdma_send_xdr_pagelist(info, xdr); + ret = svc_rdma_send_xdr_pagelist(info, xdr, + xdr->head[0].iov_len, + xdr->page_len); if (ret < 0) goto out_err; consumed += xdr->page_len; @@ -671,7 +676,6 @@ static int svc_rdma_build_read_chunk(struct svc_rqst *rqstp, struct svc_rdma_read_info *info, __be32 *p) { - unsigned int i; int ret; ret = -EINVAL; @@ -694,12 +698,6 @@ static int svc_rdma_build_read_chunk(struct svc_rqst *rqstp, info->ri_chunklen += rs_length; } - /* Pages under I/O have been copied to head->rc_pages. - * Prevent their premature release by svc_xprt_release() . - */ - for (i = 0; i < info->ri_readctxt->rc_page_count; i++) - rqstp->rq_pages[i] = NULL; - return ret; } @@ -794,6 +792,26 @@ out: return ret; } +/* Pages under I/O have been copied to head->rc_pages. Ensure they + * are not released by svc_xprt_release() until the I/O is complete. + * + * This has to be done after all Read WRs are constructed to properly + * handle a page that is part of I/O on behalf of two different RDMA + * segments. + * + * Do this only if I/O has been posted. Otherwise, we do indeed want + * svc_xprt_release() to clean things up properly. + */ +static void svc_rdma_save_io_pages(struct svc_rqst *rqstp, + const unsigned int start, + const unsigned int num_pages) +{ + unsigned int i; + + for (i = start; i < num_pages + start; i++) + rqstp->rq_pages[i] = NULL; +} + /** * svc_rdma_recv_read_chunk - Pull a Read chunk from the client * @rdma: controlling RDMA transport @@ -847,6 +865,7 @@ int svc_rdma_recv_read_chunk(struct svcxprt_rdma *rdma, struct svc_rqst *rqstp, ret = svc_rdma_post_chunk_ctxt(&info->ri_cc); if (ret < 0) goto out_err; + svc_rdma_save_io_pages(rqstp, 0, head->rc_page_count); return 0; out_err: diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c index 6fdba72f89f4..217106c66a13 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c +++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c @@ -306,15 +306,17 @@ int svc_rdma_send(struct svcxprt_rdma *rdma, struct ib_send_wr *wr) } svc_xprt_get(&rdma->sc_xprt); + trace_svcrdma_post_send(wr); ret = ib_post_send(rdma->sc_qp, wr, NULL); - trace_svcrdma_post_send(wr, ret); - if (ret) { - set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); - svc_xprt_put(&rdma->sc_xprt); - wake_up(&rdma->sc_send_wait); - } - break; + if (ret) + break; + return 0; } + + trace_svcrdma_sq_post_err(rdma, ret); + set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); + svc_xprt_put(&rdma->sc_xprt); + wake_up(&rdma->sc_send_wait); return ret; } @@ -854,7 +856,18 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) if (wr_lst) { /* XXX: Presume the client sent only one Write chunk */ - ret = svc_rdma_send_write_chunk(rdma, wr_lst, xdr); + unsigned long offset; + unsigned int length; + + if (rctxt->rc_read_payload_length) { + offset = rctxt->rc_read_payload_offset; + length = rctxt->rc_read_payload_length; + } else { + offset = xdr->head[0].iov_len; + length = xdr->page_len; + } + ret = svc_rdma_send_write_chunk(rdma, wr_lst, xdr, offset, + length); if (ret < 0) goto err2; svc_rdma_xdr_encode_write_list(rdma_resp, wr_lst, ret); @@ -871,12 +884,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) wr_lst, rp_ch); if (ret < 0) goto err1; - ret = 0; - -out: - rqstp->rq_xprt_ctxt = NULL; - svc_rdma_recv_ctxt_put(rdma, rctxt); - return ret; + return 0; err2: if (ret != -E2BIG && ret != -EINVAL) @@ -885,14 +893,39 @@ out: ret = svc_rdma_send_error_msg(rdma, sctxt, rqstp); if (ret < 0) goto err1; - ret = 0; - goto out; + return 0; err1: svc_rdma_send_ctxt_put(rdma, sctxt); err0: trace_svcrdma_send_failed(rqstp, ret); set_bit(XPT_CLOSE, &xprt->xpt_flags); - ret = -ENOTCONN; - goto out; + return -ENOTCONN; +} + +/** + * svc_rdma_read_payload - special processing for a READ payload + * @rqstp: svc_rqst to operate on + * @offset: payload's byte offset in @xdr + * @length: size of payload, in bytes + * + * Returns zero on success. + * + * For the moment, just record the xdr_buf location of the READ + * payload. svc_rdma_sendto will use that location later when + * we actually send the payload. + */ +int svc_rdma_read_payload(struct svc_rqst *rqstp, unsigned int offset, + unsigned int length) +{ + struct svc_rdma_recv_ctxt *rctxt = rqstp->rq_xprt_ctxt; + + /* XXX: Just one READ payload slot for now, since our + * transport implementation currently supports only one + * Write chunk. + */ + rctxt->rc_read_payload_offset = offset; + rctxt->rc_read_payload_length = length; + + return 0; } diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c index 0004535c0188..acb7c558e4fd 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_transport.c +++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c @@ -71,7 +71,6 @@ static struct svc_xprt *svc_rdma_create(struct svc_serv *serv, struct sockaddr *sa, int salen, int flags); static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt); -static void svc_rdma_release_rqst(struct svc_rqst *); static void svc_rdma_detach(struct svc_xprt *xprt); static void svc_rdma_free(struct svc_xprt *xprt); static int svc_rdma_has_wspace(struct svc_xprt *xprt); @@ -82,6 +81,7 @@ static const struct svc_xprt_ops svc_rdma_ops = { .xpo_create = svc_rdma_create, .xpo_recvfrom = svc_rdma_recvfrom, .xpo_sendto = svc_rdma_sendto, + .xpo_read_payload = svc_rdma_read_payload, .xpo_release_rqst = svc_rdma_release_rqst, .xpo_detach = svc_rdma_detach, .xpo_free = svc_rdma_free, @@ -559,10 +559,6 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt) return NULL; } -static void svc_rdma_release_rqst(struct svc_rqst *rqstp) -{ -} - /* * When connected, an svc_xprt has at least two references: * diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c index 3f67c395845d..638285d6602d 100644 --- a/net/sunrpc/xprtrdma/verbs.c +++ b/net/sunrpc/xprtrdma/verbs.c @@ -81,7 +81,6 @@ rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction, gfp_t flags); static void rpcrdma_regbuf_dma_unmap(struct rpcrdma_regbuf *rb); static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb); -static void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp); /* Wait for outstanding transport work to finish. ib_drain_qp * handles the drains in the wrong order for us, so open code @@ -169,7 +168,6 @@ rpcrdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc) rdmab_addr(rep->rr_rdmabuf), wc->byte_len, DMA_FROM_DEVICE); - rpcrdma_post_recvs(r_xprt, false); rpcrdma_reply_handler(rep); return; @@ -246,6 +244,7 @@ rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event) ia->ri_id->device->name, rpcrdma_addrstr(r_xprt), rpcrdma_portstr(r_xprt)); #endif + init_completion(&ia->ri_remove_done); set_bit(RPCRDMA_IAF_REMOVING, &ia->ri_flags); ep->rep_connected = -ENODEV; xprt_force_disconnect(xprt); @@ -300,7 +299,6 @@ rpcrdma_create_id(struct rpcrdma_xprt *xprt, struct rpcrdma_ia *ia) trace_xprtrdma_conn_start(xprt); init_completion(&ia->ri_done); - init_completion(&ia->ri_remove_done); id = rdma_create_id(xprt->rx_xprt.xprt_net, rpcrdma_cm_event_handler, xprt, RDMA_PS_TCP, IB_QPT_RC); @@ -611,6 +609,7 @@ static int rpcrdma_ep_recreate_xprt(struct rpcrdma_xprt *r_xprt, struct ib_qp_init_attr *qp_init_attr) { struct rpcrdma_ia *ia = &r_xprt->rx_ia; + struct rpcrdma_ep *ep = &r_xprt->rx_ep; int rc, err; trace_xprtrdma_reinsert(r_xprt); @@ -625,6 +624,7 @@ static int rpcrdma_ep_recreate_xprt(struct rpcrdma_xprt *r_xprt, pr_err("rpcrdma: rpcrdma_ep_create returned %d\n", err); goto out2; } + memcpy(qp_init_attr, &ep->rep_attr, sizeof(*qp_init_attr)); rc = -ENETUNREACH; err = rdma_create_qp(ia->ri_id, ia->ri_pd, qp_init_attr); @@ -1502,8 +1502,13 @@ rpcrdma_ep_post(struct rpcrdma_ia *ia, return 0; } -static void -rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp) +/** + * rpcrdma_post_recvs - Refill the Receive Queue + * @r_xprt: controlling transport instance + * @temp: mark Receive buffers to be deleted after use + * + */ +void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp) { struct rpcrdma_buffer *buf = &r_xprt->rx_buf; struct rpcrdma_ep *ep = &r_xprt->rx_ep; diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h index d1e0749bcbc4..6554f89ebaa1 100644 --- a/net/sunrpc/xprtrdma/xprt_rdma.h +++ b/net/sunrpc/xprtrdma/xprt_rdma.h @@ -496,6 +496,7 @@ void rpcrdma_ep_disconnect(struct rpcrdma_ep *, struct rpcrdma_ia *); int rpcrdma_ep_post(struct rpcrdma_ia *, struct rpcrdma_ep *, struct rpcrdma_req *); +void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp); /* * Buffer calls - xprtrdma/verbs.c |