diff options
Diffstat (limited to 'net/sunrpc/xprtrdma')
| -rw-r--r-- | net/sunrpc/xprtrdma/fmr_ops.c | 28 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/frwr_ops.c | 143 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/physical_ops.c | 1 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/rpc_rdma.c | 108 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_backchannel.c | 17 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_marshal.c | 64 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 60 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_sendto.c | 196 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/svc_rdma_transport.c | 445 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/verbs.c | 204 | ||||
| -rw-r--r-- | net/sunrpc/xprtrdma/xprt_rdma.h | 14 |
11 files changed, 604 insertions, 676 deletions
diff --git a/net/sunrpc/xprtrdma/fmr_ops.c b/net/sunrpc/xprtrdma/fmr_ops.c index c14f3a4bf..b289e1065 100644 --- a/net/sunrpc/xprtrdma/fmr_ops.c +++ b/net/sunrpc/xprtrdma/fmr_ops.c @@ -80,13 +80,13 @@ fmr_op_init(struct rpcrdma_xprt *r_xprt) if (!r) goto out; - r->r.fmr.physaddrs = kmalloc(RPCRDMA_MAX_FMR_SGES * - sizeof(u64), GFP_KERNEL); - if (!r->r.fmr.physaddrs) + r->fmr.physaddrs = kmalloc(RPCRDMA_MAX_FMR_SGES * + sizeof(u64), GFP_KERNEL); + if (!r->fmr.physaddrs) goto out_free; - r->r.fmr.fmr = ib_alloc_fmr(pd, mr_access_flags, &fmr_attr); - if (IS_ERR(r->r.fmr.fmr)) + r->fmr.fmr = ib_alloc_fmr(pd, mr_access_flags, &fmr_attr); + if (IS_ERR(r->fmr.fmr)) goto out_fmr_err; list_add(&r->mw_list, &buf->rb_mws); @@ -95,9 +95,9 @@ fmr_op_init(struct rpcrdma_xprt *r_xprt) return 0; out_fmr_err: - rc = PTR_ERR(r->r.fmr.fmr); + rc = PTR_ERR(r->fmr.fmr); dprintk("RPC: %s: ib_alloc_fmr status %i\n", __func__, rc); - kfree(r->r.fmr.physaddrs); + kfree(r->fmr.physaddrs); out_free: kfree(r); out: @@ -109,7 +109,7 @@ __fmr_unmap(struct rpcrdma_mw *r) { LIST_HEAD(l); - list_add(&r->r.fmr.fmr->list, &l); + list_add(&r->fmr.fmr->list, &l); return ib_unmap_fmr(&l); } @@ -148,7 +148,7 @@ fmr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg, nsegs = RPCRDMA_MAX_FMR_SGES; for (i = 0; i < nsegs;) { rpcrdma_map_one(device, seg, direction); - mw->r.fmr.physaddrs[i] = seg->mr_dma; + mw->fmr.physaddrs[i] = seg->mr_dma; len += seg->mr_len; ++seg; ++i; @@ -158,13 +158,13 @@ fmr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg, break; } - rc = ib_map_phys_fmr(mw->r.fmr.fmr, mw->r.fmr.physaddrs, + rc = ib_map_phys_fmr(mw->fmr.fmr, mw->fmr.physaddrs, i, seg1->mr_dma); if (rc) goto out_maperr; seg1->rl_mw = mw; - seg1->mr_rkey = mw->r.fmr.fmr->rkey; + seg1->mr_rkey = mw->fmr.fmr->rkey; seg1->mr_base = seg1->mr_dma + pageoff; seg1->mr_nsegs = i; seg1->mr_len = len; @@ -219,7 +219,7 @@ fmr_op_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req) seg = &req->rl_segments[i]; mw = seg->rl_mw; - list_add(&mw->r.fmr.fmr->list, &unmap_list); + list_add(&mw->fmr.fmr->list, &unmap_list); i += seg->mr_nsegs; } @@ -281,9 +281,9 @@ fmr_op_destroy(struct rpcrdma_buffer *buf) while (!list_empty(&buf->rb_all)) { r = list_entry(buf->rb_all.next, struct rpcrdma_mw, mw_all); list_del(&r->mw_all); - kfree(r->r.fmr.physaddrs); + kfree(r->fmr.physaddrs); - rc = ib_dealloc_fmr(r->r.fmr.fmr); + rc = ib_dealloc_fmr(r->fmr.fmr); if (rc) dprintk("RPC: %s: ib_dealloc_fmr failed %i\n", __func__, rc); diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c index e16567389..c250924a9 100644 --- a/net/sunrpc/xprtrdma/frwr_ops.c +++ b/net/sunrpc/xprtrdma/frwr_ops.c @@ -109,20 +109,20 @@ static void __frwr_recovery_worker(struct work_struct *work) { struct rpcrdma_mw *r = container_of(work, struct rpcrdma_mw, - r.frmr.fr_work); - struct rpcrdma_xprt *r_xprt = r->r.frmr.fr_xprt; + frmr.fr_work); + struct rpcrdma_xprt *r_xprt = r->frmr.fr_xprt; unsigned int depth = r_xprt->rx_ia.ri_max_frmr_depth; struct ib_pd *pd = r_xprt->rx_ia.ri_pd; - if (ib_dereg_mr(r->r.frmr.fr_mr)) + if (ib_dereg_mr(r->frmr.fr_mr)) goto out_fail; - r->r.frmr.fr_mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, depth); - if (IS_ERR(r->r.frmr.fr_mr)) + r->frmr.fr_mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, depth); + if (IS_ERR(r->frmr.fr_mr)) goto out_fail; dprintk("RPC: %s: recovered FRMR %p\n", __func__, r); - r->r.frmr.fr_state = FRMR_IS_INVALID; + r->frmr.fr_state = FRMR_IS_INVALID; rpcrdma_put_mw(r_xprt, r); return; @@ -137,15 +137,15 @@ out_fail: static void __frwr_queue_recovery(struct rpcrdma_mw *r) { - INIT_WORK(&r->r.frmr.fr_work, __frwr_recovery_worker); - queue_work(frwr_recovery_wq, &r->r.frmr.fr_work); + INIT_WORK(&r->frmr.fr_work, __frwr_recovery_worker); + queue_work(frwr_recovery_wq, &r->frmr.fr_work); } static int __frwr_init(struct rpcrdma_mw *r, struct ib_pd *pd, struct ib_device *device, unsigned int depth) { - struct rpcrdma_frmr *f = &r->r.frmr; + struct rpcrdma_frmr *f = &r->frmr; int rc; f->fr_mr = ib_alloc_mr(pd, IB_MR_TYPE_MEM_REG, depth); @@ -158,6 +158,8 @@ __frwr_init(struct rpcrdma_mw *r, struct ib_pd *pd, struct ib_device *device, sg_init_table(f->sg, depth); + init_completion(&f->fr_linv_done); + return 0; out_mr_err: @@ -179,11 +181,11 @@ __frwr_release(struct rpcrdma_mw *r) { int rc; - rc = ib_dereg_mr(r->r.frmr.fr_mr); + rc = ib_dereg_mr(r->frmr.fr_mr); if (rc) dprintk("RPC: %s: ib_dereg_mr status %i\n", __func__, rc); - kfree(r->r.frmr.sg); + kfree(r->frmr.sg); } static int @@ -244,39 +246,76 @@ frwr_op_maxpages(struct rpcrdma_xprt *r_xprt) rpcrdma_max_segments(r_xprt) * ia->ri_max_frmr_depth); } -/* If FAST_REG or LOCAL_INV failed, indicate the frmr needs - * to be reset. +static void +__frwr_sendcompletion_flush(struct ib_wc *wc, struct rpcrdma_frmr *frmr, + const char *wr) +{ + frmr->fr_state = FRMR_IS_STALE; + if (wc->status != IB_WC_WR_FLUSH_ERR) + pr_err("rpcrdma: %s: %s (%u/0x%x)\n", + wr, ib_wc_status_msg(wc->status), + wc->status, wc->vendor_err); +} + +/** + * frwr_wc_fastreg - Invoked by RDMA provider for each polled FastReg WC + * @cq: completion queue (ignored) + * @wc: completed WR * - * WARNING: Only wr_id and status are reliable at this point */ static void -__frwr_sendcompletion_flush(struct ib_wc *wc, struct rpcrdma_mw *r) +frwr_wc_fastreg(struct ib_cq *cq, struct ib_wc *wc) { - if (likely(wc->status == IB_WC_SUCCESS)) - return; - - /* WARNING: Only wr_id and status are reliable at this point */ - r = (struct rpcrdma_mw *)(unsigned long)wc->wr_id; - if (wc->status == IB_WC_WR_FLUSH_ERR) - dprintk("RPC: %s: frmr %p flushed\n", __func__, r); - else - pr_warn("RPC: %s: frmr %p error, status %s (%d)\n", - __func__, r, ib_wc_status_msg(wc->status), wc->status); + struct rpcrdma_frmr *frmr; + struct ib_cqe *cqe; - r->r.frmr.fr_state = FRMR_IS_STALE; + /* WARNING: Only wr_cqe and status are reliable at this point */ + if (wc->status != IB_WC_SUCCESS) { + cqe = wc->wr_cqe; + frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe); + __frwr_sendcompletion_flush(wc, frmr, "fastreg"); + } } +/** + * frwr_wc_localinv - Invoked by RDMA provider for each polled LocalInv WC + * @cq: completion queue (ignored) + * @wc: completed WR + * + */ static void -frwr_sendcompletion(struct ib_wc *wc) +frwr_wc_localinv(struct ib_cq *cq, struct ib_wc *wc) { - struct rpcrdma_mw *r = (struct rpcrdma_mw *)(unsigned long)wc->wr_id; - struct rpcrdma_frmr *f = &r->r.frmr; + struct rpcrdma_frmr *frmr; + struct ib_cqe *cqe; - if (unlikely(wc->status != IB_WC_SUCCESS)) - __frwr_sendcompletion_flush(wc, r); + /* WARNING: Only wr_cqe and status are reliable at this point */ + if (wc->status != IB_WC_SUCCESS) { + cqe = wc->wr_cqe; + frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe); + __frwr_sendcompletion_flush(wc, frmr, "localinv"); + } +} - if (f->fr_waiter) - complete(&f->fr_linv_done); +/** + * frwr_wc_localinv - Invoked by RDMA provider for each polled LocalInv WC + * @cq: completion queue (ignored) + * @wc: completed WR + * + * Awaken anyone waiting for an MR to finish being fenced. + */ +static void +frwr_wc_localinv_wake(struct ib_cq *cq, struct ib_wc *wc) +{ + struct rpcrdma_frmr *frmr; + struct ib_cqe *cqe; + + /* WARNING: Only wr_cqe and status are reliable at this point */ + cqe = wc->wr_cqe; + frmr = container_of(cqe, struct rpcrdma_frmr, fr_cqe); + if (wc->status != IB_WC_SUCCESS) + __frwr_sendcompletion_flush(wc, frmr, "localinv"); + complete_all(&frmr->fr_linv_done); } static int @@ -313,8 +352,7 @@ frwr_op_init(struct rpcrdma_xprt *r_xprt) list_add(&r->mw_list, &buf->rb_mws); list_add(&r->mw_all, &buf->rb_all); - r->mw_sendcompletion = frwr_sendcompletion; - r->r.frmr.fr_xprt = r_xprt; + r->frmr.fr_xprt = r_xprt; } return 0; @@ -347,10 +385,9 @@ frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg, mw = rpcrdma_get_mw(r_xprt); if (!mw) return -ENOMEM; - } while (mw->r.frmr.fr_state != FRMR_IS_INVALID); - frmr = &mw->r.frmr; + } while (mw->frmr.fr_state != FRMR_IS_INVALID); + frmr = &mw->frmr; frmr->fr_state = FRMR_IS_VALID; - frmr->fr_waiter = false; mr = frmr->fr_mr; reg_wr = &frmr->fr_regwr; @@ -400,7 +437,8 @@ frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg, reg_wr->wr.next = NULL; reg_wr->wr.opcode = IB_WR_REG_MR; - reg_wr->wr.wr_id = (uintptr_t)mw; + frmr->fr_cqe.done = frwr_wc_fastreg; + reg_wr->wr.wr_cqe = &frmr->fr_cqe; reg_wr->wr.num_sge = 0; reg_wr->wr.send_flags = 0; reg_wr->mr = mr; @@ -434,15 +472,15 @@ static struct ib_send_wr * __frwr_prepare_linv_wr(struct rpcrdma_mr_seg *seg) { struct rpcrdma_mw *mw = seg->rl_mw; - struct rpcrdma_frmr *f = &mw->r.frmr; + struct rpcrdma_frmr *f = &mw->frmr; struct ib_send_wr *invalidate_wr; - f->fr_waiter = false; f->fr_state = FRMR_IS_INVALID; invalidate_wr = &f->fr_invwr; memset(invalidate_wr, 0, sizeof(*invalidate_wr)); - invalidate_wr->wr_id = (unsigned long)(void *)mw; + f->fr_cqe.done = frwr_wc_localinv; + invalidate_wr->wr_cqe = &f->fr_cqe; invalidate_wr->opcode = IB_WR_LOCAL_INV; invalidate_wr->ex.invalidate_rkey = f->fr_mr->rkey; @@ -455,7 +493,7 @@ __frwr_dma_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg, { struct ib_device *device = r_xprt->rx_ia.ri_device; struct rpcrdma_mw *mw = seg->rl_mw; - struct rpcrdma_frmr *f = &mw->r.frmr; + struct rpcrdma_frmr *f = &mw->frmr; seg->rl_mw = NULL; @@ -504,15 +542,15 @@ frwr_op_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req) i += seg->mr_nsegs; } - f = &seg->rl_mw->r.frmr; + f = &seg->rl_mw->frmr; /* Strong send queue ordering guarantees that when the * last WR in the chain completes, all WRs in the chain * are complete. */ f->fr_invwr.send_flags = IB_SEND_SIGNALED; - f->fr_waiter = true; - init_completion(&f->fr_linv_done); + f->fr_cqe.done = frwr_wc_localinv_wake; + reinit_completion(&f->fr_linv_done); INIT_CQCOUNT(&r_xprt->rx_ep); /* Transport disconnect drains the receive CQ before it @@ -520,14 +558,18 @@ frwr_op_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req) * unless ri_id->qp is a valid pointer. */ rc = ib_post_send(ia->ri_id->qp, invalidate_wrs, &bad_wr); - if (rc) + if (rc) { pr_warn("%s: ib_post_send failed %i\n", __func__, rc); + rdma_disconnect(ia->ri_id); + goto unmap; + } wait_for_completion(&f->fr_linv_done); /* ORDER: Now DMA unmap all of the req's MRs, and return * them to the free MW list. */ +unmap: for (i = 0, nchunks = req->rl_nchunks; nchunks; nchunks--) { seg = &req->rl_segments[i]; @@ -549,7 +591,7 @@ frwr_op_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg) struct rpcrdma_mr_seg *seg1 = seg; struct rpcrdma_ia *ia = &r_xprt->rx_ia; struct rpcrdma_mw *mw = seg1->rl_mw; - struct rpcrdma_frmr *frmr = &mw->r.frmr; + struct rpcrdma_frmr *frmr = &mw->frmr; struct ib_send_wr *invalidate_wr, *bad_wr; int rc, nsegs = seg->mr_nsegs; @@ -557,10 +599,11 @@ frwr_op_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg) seg1->rl_mw = NULL; frmr->fr_state = FRMR_IS_INVALID; - invalidate_wr = &mw->r.frmr.fr_invwr; + invalidate_wr = &mw->frmr.fr_invwr; memset(invalidate_wr, 0, sizeof(*invalidate_wr)); - invalidate_wr->wr_id = (uintptr_t)mw; + frmr->fr_cqe.done = frwr_wc_localinv; + invalidate_wr->wr_cqe = &frmr->fr_cqe; invalidate_wr->opcode = IB_WR_LOCAL_INV; invalidate_wr->ex.invalidate_rkey = frmr->fr_mr->rkey; DECR_CQCOUNT(&r_xprt->rx_ep); diff --git a/net/sunrpc/xprtrdma/physical_ops.c b/net/sunrpc/xprtrdma/physical_ops.c index dbb302ecf..481b9b6f4 100644 --- a/net/sunrpc/xprtrdma/physical_ops.c +++ b/net/sunrpc/xprtrdma/physical_ops.c @@ -68,7 +68,6 @@ physical_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg, rpcrdma_map_one(ia->ri_device, seg, rpcrdma_data_dir(writing)); seg->mr_rkey = ia->ri_dma_mr->rkey; seg->mr_base = seg->mr_dma; - seg->mr_nsegs = 1; return 1; } diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c index 0f28f2d74..888823bb6 100644 --- a/net/sunrpc/xprtrdma/rpc_rdma.c +++ b/net/sunrpc/xprtrdma/rpc_rdma.c @@ -132,6 +132,33 @@ rpcrdma_tail_pullup(struct xdr_buf *buf) return tlen; } +/* Split "vec" on page boundaries into segments. FMR registers pages, + * not a byte range. Other modes coalesce these segments into a single + * MR when they can. + */ +static int +rpcrdma_convert_kvec(struct kvec *vec, struct rpcrdma_mr_seg *seg, + int n, int nsegs) +{ + size_t page_offset; + u32 remaining; + char *base; + + base = vec->iov_base; + page_offset = offset_in_page(base); + remaining = vec->iov_len; + while (remaining && n < nsegs) { + seg[n].mr_page = NULL; + seg[n].mr_offset = base; + seg[n].mr_len = min_t(u32, PAGE_SIZE - page_offset, remaining); + remaining -= seg[n].mr_len; + base += seg[n].mr_len; + ++n; + page_offset = 0; + } + return n; +} + /* * Chunk assembly from upper layer xdr_buf. * @@ -150,11 +177,10 @@ rpcrdma_convert_iovs(struct xdr_buf *xdrbuf, unsigned int pos, int page_base; struct page **ppages; - if (pos == 0 && xdrbuf->head[0].iov_len) { - seg[n].mr_page = NULL; - seg[n].mr_offset = xdrbuf->head[0].iov_base; - seg[n].mr_len = xdrbuf->head[0].iov_len; - ++n; + if (pos == 0) { + n = rpcrdma_convert_kvec(&xdrbuf->head[0], seg, n, nsegs); + if (n == nsegs) + return -EIO; } len = xdrbuf->page_len; @@ -192,13 +218,9 @@ rpcrdma_convert_iovs(struct xdr_buf *xdrbuf, unsigned int pos, * xdr pad bytes, saving the server an RDMA operation. */ if (xdrbuf->tail[0].iov_len < 4 && xprt_rdma_pad_optimize) return n; + n = rpcrdma_convert_kvec(&xdrbuf->tail[0], seg, n, nsegs); if (n == nsegs) - /* Tail remains, but we're out of segments */ return -EIO; - seg[n].mr_page = NULL; - seg[n].mr_offset = xdrbuf->tail[0].iov_base; - seg[n].mr_len = xdrbuf->tail[0].iov_len; - ++n; } return n; @@ -773,20 +795,17 @@ rpcrdma_reply_handler(struct rpcrdma_rep *rep) struct rpcrdma_xprt *r_xprt = rep->rr_rxprt; struct rpc_xprt *xprt = &r_xprt->rx_xprt; __be32 *iptr; - int rdmalen, status; + int rdmalen, status, rmerr; unsigned long cwnd; - u32 credits; dprintk("RPC: %s: incoming rep %p\n", __func__, rep); if (rep->rr_len == RPCRDMA_BAD_LEN) goto out_badstatus; - if (rep->rr_len < RPCRDMA_HDRLEN_MIN) + if (rep->rr_len < RPCRDMA_HDRLEN_ERR) goto out_shortreply; headerp = rdmab_to_msg(rep->rr_rdmabuf); - if (headerp->rm_vers != rpcrdma_version) - goto out_badversion; #if defined(CONFIG_SUNRPC_BACKCHANNEL) if (rpcrdma_is_bcall(headerp)) goto out_bcall; @@ -809,15 +828,16 @@ rpcrdma_reply_handler(struct rpcrdma_rep *rep) */ list_del_init(&rqst->rq_list); spin_unlock_bh(&xprt->transport_lock); - dprintk("RPC: %s: reply 0x%p completes request 0x%p\n" - " RPC request 0x%p xid 0x%08x\n", - __func__, rep, req, rqst, - be32_to_cpu(headerp->rm_xid)); + dprintk("RPC: %s: reply %p completes request %p (xid 0x%08x)\n", + __func__, rep, req, be32_to_cpu(headerp->rm_xid)); /* from here on, the reply is no longer an orphan */ req->rl_reply = rep; xprt->reestablish_timeout = 0; + if (headerp->rm_vers != rpcrdma_version) + goto out_badversion; + /* check for expected message types */ /* The order of some of these tests is important. */ switch (headerp->rm_type) { @@ -878,6 +898,9 @@ rpcrdma_reply_handler(struct rpcrdma_rep *rep) status = rdmalen; break; + case rdma_error: + goto out_rdmaerr; + badheader: default: dprintk("%s: invalid rpcrdma reply header (type %d):" @@ -893,6 +916,7 @@ badheader: break; } +out: /* Invalidate and flush the data payloads before waking the * waiting application. This guarantees the memory region is * properly fenced from the server before the application @@ -903,15 +927,9 @@ badheader: if (req->rl_nchunks) r_xprt->rx_ia.ri_ops->ro_unmap_sync(r_xprt, req); - credits = be32_to_cpu(headerp->rm_credit); - if (credits == 0) - credits = 1; /* don't deadlock */ - else if (credits > r_xprt->rx_buf.rb_max_requests) - credits = r_xprt->rx_buf.rb_max_requests; - spin_lock_bh(&xprt->transport_lock); cwnd = xprt->cwnd; - xprt->cwnd = credits << RPC_CWNDSHIFT; + xprt->cwnd = atomic_read(&r_xprt->rx_buf.rb_credits) << RPC_CWNDSHIFT; if (xprt->cwnd > cwnd) xprt_release_rqst_cong(rqst->rq_task); @@ -935,13 +953,43 @@ out_bcall: return; #endif -out_shortreply: - dprintk("RPC: %s: short/invalid reply\n", __func__); - goto repost; - +/* If the incoming reply terminated a pending RPC, the next + * RPC call will post a replacement receive buffer as it is + * being marshaled. + */ out_badversion: dprintk("RPC: %s: invalid version %d\n", __func__, be32_to_cpu(headerp->rm_vers)); + status = -EIO; + r_xprt->rx_stats.bad_reply_count++; + goto out; + +out_rdmaerr: + rmerr = be32_to_cpu(headerp->rm_body.rm_error.rm_err); + switch (rmerr) { + case ERR_VERS: + pr_err("%s: server reports header version error (%u-%u)\n", + __func__, + be32_to_cpu(headerp->rm_body.rm_error.rm_vers_low), + be32_to_cpu(headerp->rm_body.rm_error.rm_vers_high)); + break; + case ERR_CHUNK: + pr_err("%s: server reports header decoding error\n", + __func__); + break; + default: + pr_err("%s: server reports unknown error %d\n", + __func__, rmerr); + } + status = -EREMOTEIO; + r_xprt->rx_stats.bad_reply_count++; + goto out; + +/* If no pending RPC transaction was matched, post a replacement + * receive buffer before returning. + */ +out_shortreply: + dprintk("RPC: %s: short/invalid reply\n", __func__); goto repost; out_nomatch: diff --git a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c index 65a7c232a..a2a7519b0 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c +++ b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c @@ -107,26 +107,18 @@ static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma, int ret; vec = svc_rdma_get_req_map(rdma); - ret = svc_rdma_map_xdr(rdma, sndbuf, vec); + ret = svc_rdma_map_xdr(rdma, sndbuf, vec, false); if (ret) goto out_err; - /* Post a recv buffer to handle the reply for this request. */ - ret = svc_rdma_post_recv(rdma, GFP_NOIO); - if (ret) { - pr_err("svcrdma: Failed to post bc receive buffer, err=%d.\n", - ret); - pr_err("svcrdma: closing transport %p.\n", rdma); - set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); - ret = -ENOTCONN; + ret = svc_rdma_repost_recv(rdma, GFP_NOIO); + if (ret) goto out_err; - } ctxt = svc_rdma_get_context(rdma); ctxt->pages[0] = virt_to_page(rqst->rq_buffer); ctxt->count = 1; - ctxt->wr_op = IB_WR_SEND; ctxt->direction = DMA_TO_DEVICE; ctxt->sge[0].lkey = rdma->sc_pd->local_dma_lkey; ctxt->sge[0].length = sndbuf->len; @@ -140,7 +132,8 @@ static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma, atomic_inc(&rdma->sc_dma_used); memset(&send_wr, 0, sizeof(send_wr)); - send_wr.wr_id = (unsigned long)ctxt; + ctxt->cqe.done = svc_rdma_wc_send; + send_wr.wr_cqe = &ctxt->cqe; send_wr.sg_list = ctxt->sge; send_wr.num_sge = 1; send_wr.opcode = IB_WR_SEND; diff --git a/net/sunrpc/xprtrdma/svc_rdma_marshal.c b/net/sunrpc/xprtrdma/svc_rdma_marshal.c index e2fca7617..765bca47c 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_marshal.c +++ b/net/sunrpc/xprtrdma/svc_rdma_marshal.c @@ -145,29 +145,44 @@ static __be32 *decode_reply_array(__be32 *va, __be32 *vaend) return (__be32 *)&ary->wc_array[nchunks]; } -int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req, - struct svc_rqst *rqstp) +int svc_rdma_xdr_decode_req(struct rpcrdma_msg *rmsgp, struct svc_rqst *rqstp) { - struct rpcrdma_msg *rmsgp = NULL; __be32 *va, *vaend; + unsigned int len; u32 hdr_len; - rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base; - /* Verify that there's enough bytes for header + something */ - if (rqstp->rq_arg.len <= RPCRDMA_HDRLEN_MIN) { + if (rqstp->rq_arg.len <= RPCRDMA_HDRLEN_ERR) { dprintk("svcrdma: header too short = %d\n", rqstp->rq_arg.len); return -EINVAL; } - if (rmsgp->rm_vers != rpcrdma_version) - return -ENOSYS; - - /* Pull in the extra for the padded case and bump our pointer */ - if (rmsgp->rm_type == rdma_msgp) { - int hdrlen; + if (rmsgp->rm_vers != rpcrdma_version) { + dprintk("%s: bad version %u\n", __func__, + be32_to_cpu(rmsgp->rm_vers)); + return -EPROTONOSUPPORT; + } + switch (be32_to_cpu(rmsgp->rm_type)) { + case RDMA_MSG: + case RDMA_NOMSG: + break; + + case RDMA_DONE: + /* Just drop it */ + dprintk("svcrdma: dropping RDMA_DONE message\n"); + return 0; + + case RDMA_ERROR: + /* Possible if this is a backchannel reply. + * XXX: We should cancel this XID, though. + */ + dprintk("svcrdma: dropping RDMA_ERROR message\n"); + return 0; + + case RDMA_MSGP: + /* Pull in the extra for the padded case, bump our pointer */ rmsgp->rm_body.rm_padded.rm_align = be32_to_cpu(rmsgp->rm_body.rm_padded.rm_align); rmsgp->rm_body.rm_padded.rm_thresh = @@ -175,11 +190,15 @@ int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req, va = &rmsgp->rm_body.rm_padded.rm_pempty[4]; rqstp->rq_arg.head[0].iov_base = va; - hdrlen = (u32)((unsigned long)va - (unsigned long)rmsgp); - rqstp->rq_arg.head[0].iov_len -= hdrlen; - if (hdrlen > rqstp->rq_arg.len) + len = (u32)((unsigned long)va - (unsigned long)rmsgp); + rqstp->rq_arg.head[0].iov_len -= len; + if (len > rqstp->rq_arg.len) return -EINVAL; - return hdrlen; + return len; + default: + dprintk("svcrdma: bad rdma procedure (%u)\n", + be32_to_cpu(rmsgp->rm_type)); + return -EINVAL; } /* The chunk list may contain either a read chunk list or a write @@ -188,20 +207,25 @@ int svc_rdma_xdr_decode_req(struct rpcrdma_msg **rdma_req, va = &rmsgp->rm_body.rm_chunks[0]; vaend = (__be32 *)((unsigned long)rmsgp + rqstp->rq_arg.len); va = decode_read_list(va, vaend); - if (!va) + if (!va) { + dprintk("svcrdma: failed to decode read list\n"); return -EINVAL; + } va = decode_write_list(va, vaend); - if (!va) + if (!va) { + dprintk("svcrdma: failed to decode write list\n"); return -EINVAL; + } va = decode_reply_array(va, vaend); - if (!va) + if (!va) { + dprintk("svcrdma: failed to decode reply chunk\n"); return -EINVAL; + } rqstp->rq_arg.head[0].iov_base = va; hdr_len = (unsigned long)va - (unsigned long)rmsgp; rqstp->rq_arg.head[0].iov_len -= hdr_len; - *rdma_req = rmsgp; return hdr_len; } diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c index c8b8a8b41..3b24a646e 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c @@ -180,9 +180,9 @@ int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt, clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags); memset(&read_wr, 0, sizeof(read_wr)); - read_wr.wr.wr_id = (unsigned long)ctxt; + ctxt->cqe.done = svc_rdma_wc_read; + read_wr.wr.wr_cqe = &ctxt->cqe; read_wr.wr.opcode = IB_WR_RDMA_READ; - ctxt->wr_op = read_wr.wr.opcode; read_wr.wr.send_flags = IB_SEND_SIGNALED; read_wr.rkey = rs_handle; read_wr.remote_addr = rs_offset; @@ -299,8 +299,9 @@ int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt, ctxt->read_hdr = head; /* Prepare REG WR */ + ctxt->reg_cqe.done = svc_rdma_wc_reg; + reg_wr.wr.wr_cqe = &ctxt->reg_cqe; reg_wr.wr.opcode = IB_WR_REG_MR; - reg_wr.wr.wr_id = 0; reg_wr.wr.send_flags = IB_SEND_SIGNALED; reg_wr.wr.num_sge = 0; reg_wr.mr = frmr->mr; @@ -310,6 +311,8 @@ int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt, /* Prepare RDMA_READ */ memset(&read_wr, 0, sizeof(read_wr)); + ctxt->cqe.done = svc_rdma_wc_read; + read_wr.wr.wr_cqe = &ctxt->cqe; read_wr.wr.send_flags = IB_SEND_SIGNALED; read_wr.rkey = rs_handle; read_wr.remote_addr = rs_offset; @@ -317,19 +320,18 @@ int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt, read_wr.wr.num_sge = 1; if (xprt->sc_dev_caps & SVCRDMA_DEVCAP_READ_W_INV) { read_wr.wr.opcode = IB_WR_RDMA_READ_WITH_INV; - read_wr.wr.wr_id = (unsigned long)ctxt; read_wr.wr.ex.invalidate_rkey = ctxt->frmr->mr->lkey; } else { read_wr.wr.opcode = IB_WR_RDMA_READ; read_wr.wr.next = &inv_wr; /* Prepare invalidate */ memset(&inv_wr, 0, sizeof(inv_wr)); - inv_wr.wr_id = (unsigned long)ctxt; + ctxt->inv_cqe.done = svc_rdma_wc_inv; + inv_wr.wr_cqe = &ctxt->inv_cqe; inv_wr.opcode = IB_WR_LOCAL_INV; inv_wr.send_flags = IB_SEND_SIGNALED | IB_SEND_FENCE; inv_wr.ex.invalidate_rkey = frmr->mr->lkey; } - ctxt->wr_op = read_wr.wr.opcode; /* Post the chain */ ret = svc_rdma_send(xprt, ®_wr.wr); @@ -612,7 +614,6 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp) struct svc_rdma_op_ctxt *ctxt = NULL; struct rpcrdma_msg *rmsgp; int ret = 0; - int len; dprintk("svcrdma: rqstp=%p\n", rqstp); @@ -642,8 +643,7 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp) * transport list */ if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) - goto close_out; - + goto defer; goto out; } dprintk("svcrdma: processing ctxt=%p on xprt=%p, rqstp=%p, status=%d\n", @@ -654,15 +654,13 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp) rdma_build_arg_xdr(rqstp, ctxt, ctxt->byte_len); /* Decode the RDMA header. */ - len = svc_rdma_xdr_decode_req(&rmsgp, rqstp); - rqstp->rq_xprt_hlen = len; - - /* If the request is invalid, reply with an error */ - if (len < 0) { - if (len == -ENOSYS) - svc_rdma_send_error(rdma_xprt, rmsgp, ERR_VERS); - goto close_out; - } + rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base; + ret = svc_rdma_xdr_decode_req(rmsgp, rqstp); + if (ret < 0) + goto out_err; + if (ret == 0) + goto out_drop; + rqstp->rq_xprt_hlen = ret; if (svc_rdma_is_backchannel_reply(xprt, rmsgp)) { ret = svc_rdma_handle_bc_reply(xprt->xpt_bc_xprt, rmsgp, @@ -698,26 +696,16 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp) svc_xprt_copy_addrs(rqstp, xprt); return ret; - close_out: - if (ctxt) - svc_rdma_put_context(ctxt, 1); - dprintk("svcrdma: transport %p is closing\n", xprt); - /* - * Set the close bit and enqueue it. svc_recv will see the - * close bit and call svc_xprt_delete - */ - set_bit(XPT_CLOSE, &xprt->xpt_flags); +out_err: + svc_rdma_send_error(rdma_xprt, rmsgp, ret); + svc_rdma_put_context(ctxt, 0); + return 0; + defer: return 0; +out_drop: + svc_rdma_put_context(ctxt, 1); repost: - ret = svc_rdma_post_recv(rdma_xprt, GFP_KERNEL); - if (ret) { - pr_err("svcrdma: could not post a receive buffer, err=%d.\n", - ret); - pr_err("svcrdma: closing transport %p.\n", rdma_xprt); - set_bit(XPT_CLOSE, &rdma_xprt->sc_xprt.xpt_flags); - ret = -ENOTCONN; - } - return ret; + return svc_rdma_repost_recv(rdma_xprt, GFP_KERNEL); } diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c index df57f3ce6..4f1b1c4f4 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c +++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c @@ -50,9 +50,15 @@ #define RPCDBG_FACILITY RPCDBG_SVCXPRT +static u32 xdr_padsize(u32 len) +{ + return (len & 3) ? (4 - (len & 3)) : 0; +} + int svc_rdma_map_xdr(struct svcxprt_rdma *xprt, struct xdr_buf *xdr, - struct svc_rdma_req_map *vec) + struct svc_rdma_req_map *vec, + bool write_chunk_present) { int sge_no; u32 sge_bytes; @@ -92,9 +98,20 @@ int svc_rdma_map_xdr(struct svcxprt_rdma *xprt, /* Tail SGE */ if (xdr->tail[0].iov_len) { - vec->sge[sge_no].iov_base = xdr->tail[0].iov_base; - vec->sge[sge_no].iov_len = xdr->tail[0].iov_len; - sge_no++; + unsigned char *base = xdr->tail[0].iov_base; + size_t len = xdr->tail[0].iov_len; + u32 xdr_pad = xdr_padsize(xdr->page_len); + + if (write_chunk_present && xdr_pad) { + base += xdr_pad; + len -= xdr_pad; + } + + if (len) { + vec->sge[sge_no].iov_base = base; + vec->sge[sge_no].iov_len = len; + sge_no++; + } } dprintk("svcrdma: %s: sge_no %d page_no %d " @@ -166,10 +183,10 @@ svc_rdma_get_write_array(struct rpcrdma_msg *rmsgp) * reply array is present */ static struct rpcrdma_write_array * -svc_rdma_get_reply_array(struct rpcrdma_msg *rmsgp) +svc_rdma_get_reply_array(struct rpcrdma_msg *rmsgp, + struct rpcrdma_write_array *wr_ary) { struct rpcrdma_read_chunk *rch; - struct rpcrdma_write_array *wr_ary; struct rpcrdma_write_array *rp_ary; /* XXX: Need to fix when reply chunk may occur with read list @@ -191,7 +208,6 @@ svc_rdma_get_reply_array(struct rpcrdma_msg *rmsgp) goto found_it; } - wr_ary = svc_rdma_get_write_array(rmsgp); if (wr_ary) { int chunk = be32_to_cpu(wr_ary->wc_nchunks); @@ -281,8 +297,8 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp, /* Prepare WRITE WR */ memset(&write_wr, 0, sizeof write_wr); - ctxt->wr_op = IB_WR_RDMA_WRITE; - write_wr.wr.wr_id = (unsigned long)ctxt; + ctxt->cqe.done = svc_rdma_wc_write; + write_wr.wr.wr_cqe = &ctxt->cqe; write_wr.wr.sg_list = &sge[0]; write_wr.wr.num_sge = sge_no; write_wr.wr.opcode = IB_WR_RDMA_WRITE; @@ -298,41 +314,37 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp, err: svc_rdma_unmap_dma(ctxt); svc_rdma_put_context(ctxt, 0); - /* Fatal error, close transport */ return -EIO; } +noinline static int send_write_chunks(struct svcxprt_rdma *xprt, - struct rpcrdma_msg *rdma_argp, + struct rpcrdma_write_array *wr_ary, struct rpcrdma_msg *rdma_resp, struct svc_rqst *rqstp, struct svc_rdma_req_map *vec) { - u32 xfer_len = rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len; + u32 xfer_len = rqstp->rq_res.page_len; int write_len; u32 xdr_off; int chunk_off; int chunk_no; int nchunks; - struct rpcrdma_write_array *arg_ary; struct rpcrdma_write_array *res_ary; int ret; - arg_ary = svc_rdma_get_write_array(rdma_argp); - if (!arg_ary) - return 0; res_ary = (struct rpcrdma_write_array *) &rdma_resp->rm_body.rm_chunks[1]; /* Write chunks start at the pagelist */ - nchunks = be32_to_cpu(arg_ary->wc_nchunks); + nchunks = be32_to_cpu(wr_ary->wc_nchunks); for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0; xfer_len && chunk_no < nchunks; chunk_no++) { struct rpcrdma_segment *arg_ch; u64 rs_offset; - arg_ch = &arg_ary->wc_array[chunk_no].wc_target; + arg_ch = &wr_ary->wc_array[chunk_no].wc_target; write_len = min(xfer_len, be32_to_cpu(arg_ch->rs_length)); /* Prepare the response chunk given the length actually @@ -350,11 +362,8 @@ static int send_write_chunks(struct svcxprt_rdma *xprt, xdr_off, write_len, vec); - if (ret <= 0) { - dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n", - ret); - return -EIO; - } + if (ret <= 0) + goto out_err; chunk_off += ret; xdr_off += ret; xfer_len -= ret; @@ -364,11 +373,16 @@ static int send_write_chunks(struct svcxprt_rdma *xprt, /* Update the req with the number of chunks actually used */ svc_rdma_xdr_encode_write_list(rdma_resp, chunk_no); - return rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len; + return rqstp->rq_res.page_len; + +out_err: + pr_err("svcrdma: failed to send write chunks, rc=%d\n", ret); + return -EIO; } +noinline static int send_reply_chunks(struct svcxprt_rdma *xprt, - struct rpcrdma_msg *rdma_argp, + struct rpcrdma_write_array *rp_ary, struct rpcrdma_msg *rdma_resp, struct svc_rqst *rqstp, struct svc_rdma_req_map *vec) @@ -380,25 +394,21 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt, int chunk_off; int nchunks; struct rpcrdma_segment *ch; - struct rpcrdma_write_array *arg_ary; struct rpcrdma_write_array *res_ary; int ret; - arg_ary = svc_rdma_get_reply_array(rdma_argp); - if (!arg_ary) - return 0; /* XXX: need to fix when reply lists occur with read-list and or * write-list */ res_ary = (struct rpcrdma_write_array *) &rdma_resp->rm_body.rm_chunks[2]; /* xdr offset starts at RPC message */ - nchunks = be32_to_cpu(arg_ary->wc_nchunks); + nchunks = be32_to_cpu(rp_ary->wc_nchunks); for (xdr_off = 0, chunk_no = 0; xfer_len && chunk_no < nchunks; chunk_no++) { u64 rs_offset; - ch = &arg_ary->wc_array[chunk_no].wc_target; + ch = &rp_ary->wc_array[chunk_no].wc_target; write_len = min(xfer_len, be32_to_cpu(ch->rs_length)); /* Prepare the reply chunk given the length actually @@ -415,11 +425,8 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt, xdr_off, write_len, vec); - if (ret <= 0) { - dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n", - ret); - return -EIO; - } + if (ret <= 0) + goto out_err; chunk_off += ret; xdr_off += ret; xfer_len -= ret; @@ -430,6 +437,10 @@ static int send_reply_chunks(struct svcxprt_rdma *xprt, svc_rdma_xdr_encode_reply_array(res_ary, chunk_no); return rqstp->rq_res.len; + +out_err: + pr_err("svcrdma: failed to send reply chunks, rc=%d\n", ret); + return -EIO; } /* This function prepares the portion of the RPCRDMA message to be @@ -464,13 +475,8 @@ static int send_reply(struct svcxprt_rdma *rdma, int pages; int ret; - /* Post a recv buffer to handle another request. */ - ret = svc_rdma_post_recv(rdma, GFP_KERNEL); + ret = svc_rdma_repost_recv(rdma, GFP_KERNEL); if (ret) { - printk(KERN_INFO - "svcrdma: could not post a receive buffer, err=%d." - "Closing transport %p.\n", ret, rdma); - set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); svc_rdma_put_context(ctxt, 0); return -ENOTCONN; } @@ -543,8 +549,8 @@ static int send_reply(struct svcxprt_rdma *rdma, goto err; } memset(&send_wr, 0, sizeof send_wr); - ctxt->wr_op = IB_WR_SEND; - send_wr.wr_id = (unsigned long)ctxt; + ctxt->cqe.done = svc_rdma_wc_send; + send_wr.wr_cqe = &ctxt->cqe; send_wr.sg_list = ctxt->sge; send_wr.num_sge = sge_no; send_wr.opcode = IB_WR_SEND; @@ -559,6 +565,7 @@ static int send_reply(struct svcxprt_rdma *rdma, err: svc_rdma_unmap_dma(ctxt); svc_rdma_put_context(ctxt, 1); + pr_err("svcrdma: failed to send reply, rc=%d\n", ret); return -EIO; } @@ -573,7 +580,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) container_of(xprt, struct svcxprt_rdma, sc_xprt); struct rpcrdma_msg *rdma_argp; struct rpcrdma_msg *rdma_resp; - struct rpcrdma_write_array *reply_ary; + struct rpcrdma_write_array *wr_ary, *rp_ary; enum rpcrdma_proc reply_type; int ret; int inline_bytes; @@ -587,12 +594,14 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) * places this at the start of page 0. */ rdma_argp = page_address(rqstp->rq_pages[0]); + wr_ary = svc_rdma_get_write_array(rdma_argp); + rp_ary = svc_rdma_get_reply_array(rdma_argp, wr_ary); /* Build an req vec for the XDR */ ctxt = svc_rdma_get_context(rdma); ctxt->direction = DMA_TO_DEVICE; vec = svc_rdma_get_req_map(rdma); - ret = svc_rdma_map_xdr(rdma, &rqstp->rq_res, vec); + ret = svc_rdma_map_xdr(rdma, &rqstp->rq_res, vec, wr_ary != NULL); if (ret) goto err0; inline_bytes = rqstp->rq_res.len; @@ -603,8 +612,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) if (!res_page) goto err0; rdma_resp = page_address(res_page); - reply_ary = svc_rdma_get_reply_array(rdma_argp); - if (reply_ary) + if (rp_ary) reply_type = RDMA_NOMSG; else reply_type = RDMA_MSG; @@ -612,27 +620,26 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) rdma_resp, reply_type); /* Send any write-chunk data and build resp write-list */ - ret = send_write_chunks(rdma, rdma_argp, rdma_resp, - rqstp, vec); - if (ret < 0) { - printk(KERN_ERR "svcrdma: failed to send write chunks, rc=%d\n", - ret); - goto err1; + if (wr_ary) { + ret = send_write_chunks(rdma, wr_ary, rdma_resp, rqstp, vec); + if (ret < 0) + goto err1; + inline_bytes -= ret + xdr_padsize(ret); } - inline_bytes -= ret; /* Send any reply-list data and update resp reply-list */ - ret = send_reply_chunks(rdma, rdma_argp, rdma_resp, - rqstp, vec); - if (ret < 0) { - printk(KERN_ERR "svcrdma: failed to send reply chunks, rc=%d\n", - ret); - goto err1; + if (rp_ary) { + ret = send_reply_chunks(rdma, rp_ary, rdma_resp, rqstp, vec); + if (ret < 0) + goto err1; + inline_bytes -= ret; } - inline_bytes -= ret; ret = send_reply(rdma, rqstp, res_page, rdma_resp, ctxt, vec, inline_bytes); + if (ret < 0) + goto err1; + svc_rdma_put_req_map(rdma, vec); dprintk("svcrdma: send_reply returns %d\n", ret); return ret; @@ -642,5 +649,68 @@ int svc_rdma_sendto(struct svc_rqst *rqstp) err0: svc_rdma_put_req_map(rdma, vec); svc_rdma_put_context(ctxt, 0); - return ret; + set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); + return -ENOTCONN; +} + +void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp, + int status) +{ + struct ib_send_wr err_wr; + struct page *p; + struct svc_rdma_op_ctxt *ctxt; + enum rpcrdma_errcode err; + __be32 *va; + int length; + int ret; + + ret = svc_rdma_repost_recv(xprt, GFP_KERNEL); + if (ret) + return; + + p = alloc_page(GFP_KERNEL); + if (!p) + return; + va = page_address(p); + + /* XDR encode an error reply */ + err = ERR_CHUNK; + if (status == -EPROTONOSUPPORT) + err = ERR_VERS; + length = svc_rdma_xdr_encode_error(xprt, rmsgp, err, va); + + ctxt = svc_rdma_get_context(xprt); + ctxt->direction = DMA_TO_DEVICE; + ctxt->count = 1; + ctxt->pages[0] = p; + + /* Prepare SGE for local address */ + ctxt->sge[0].lkey = xprt->sc_pd->local_dma_lkey; + ctxt->sge[0].length = length; + ctxt->sge[0].addr = ib_dma_map_page(xprt->sc_cm_id->device, + p, 0, length, DMA_TO_DEVICE); + if (ib_dma_mapping_error(xprt->sc_cm_id->device, ctxt->sge[0].addr)) { + dprintk("svcrdma: Error mapping buffer for protocol error\n"); + svc_rdma_put_context(ctxt, 1); + return; + } + atomic_inc(&xprt->sc_dma_used); + + /* Prepare SEND WR */ + memset(&err_wr, 0, sizeof(err_wr)); + ctxt->cqe.done = svc_rdma_wc_send; + err_wr.wr_cqe = &ctxt->cqe; + err_wr.sg_list = ctxt->sge; + err_wr.num_sge = 1; + err_wr.opcode = IB_WR_SEND; + err_wr.send_flags = IB_SEND_SIGNALED; + + /* Post It */ + ret = svc_rdma_send(xprt, &err_wr); + if (ret) { + dprintk("svcrdma: Error %d posting send for protocol error\n", + ret); + svc_rdma_unmap_dma(ctxt); + svc_rdma_put_context(ctxt, 1); + } } diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c index 5763825d0..90668969d 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_transport.c +++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c @@ -63,17 +63,10 @@ static struct svc_xprt *svc_rdma_create(struct svc_serv *serv, int flags); static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt); static void svc_rdma_release_rqst(struct svc_rqst *); -static void dto_tasklet_func(unsigned long data); 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); static int svc_rdma_secure_port(struct svc_rqst *); -static void rq_cq_reap(struct svcxprt_rdma *xprt); -static void sq_cq_reap(struct svcxprt_rdma *xprt); - -static DECLARE_TASKLET(dto_tasklet, dto_tasklet_func, 0UL); -static DEFINE_SPINLOCK(dto_lock); -static LIST_HEAD(dto_xprt_q); static struct svc_xprt_ops svc_rdma_ops = { .xpo_create = svc_rdma_create, @@ -352,15 +345,6 @@ static void svc_rdma_destroy_maps(struct svcxprt_rdma *xprt) } } -/* ib_cq event handler */ -static void cq_event_handler(struct ib_event *event, void *context) -{ - struct svc_xprt *xprt = context; - dprintk("svcrdma: received CQ event %s (%d), context=%p\n", - ib_event_msg(event->event), event->event, context); - set_bit(XPT_CLOSE, &xprt->xpt_flags); -} - /* QP event handler */ static void qp_event_handler(struct ib_event *event, void *context) { @@ -392,251 +376,171 @@ static void qp_event_handler(struct ib_event *event, void *context) } } -/* - * Data Transfer Operation Tasklet +/** + * svc_rdma_wc_receive - Invoked by RDMA provider for each polled Receive WC + * @cq: completion queue + * @wc: completed WR * - * Walks a list of transports with I/O pending, removing entries as - * they are added to the server's I/O pending list. Two bits indicate - * if SQ, RQ, or both have I/O pending. The dto_lock is an irqsave - * spinlock that serializes access to the transport list with the RQ - * and SQ interrupt handlers. */ -static void dto_tasklet_func(unsigned long data) +static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc) { - struct svcxprt_rdma *xprt; - unsigned long flags; + struct svcxprt_rdma *xprt = cq->cq_context; + struct ib_cqe *cqe = wc->wr_cqe; + struct svc_rdma_op_ctxt *ctxt; - spin_lock_irqsave(&dto_lock, flags); - while (!list_empty(&dto_xprt_q)) { - xprt = list_entry(dto_xprt_q.next, - struct svcxprt_rdma, sc_dto_q); - list_del_init(&xprt->sc_dto_q); - spin_unlock_irqrestore(&dto_lock, flags); + /* WARNING: Only wc->wr_cqe and wc->status are reliable */ + ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe); + ctxt->wc_status = wc->status; + svc_rdma_unmap_dma(ctxt); - rq_cq_reap(xprt); - sq_cq_reap(xprt); + if (wc->status != IB_WC_SUCCESS) + goto flushed; - svc_xprt_put(&xprt->sc_xprt); - spin_lock_irqsave(&dto_lock, flags); - } - spin_unlock_irqrestore(&dto_lock, flags); + /* All wc fields are now known to be valid */ + ctxt->byte_len = wc->byte_len; + spin_lock(&xprt->sc_rq_dto_lock); + list_add_tail(&ctxt->dto_q, &xprt->sc_rq_dto_q); + spin_unlock(&xprt->sc_rq_dto_lock); + + set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags); + if (test_bit(RDMAXPRT_CONN_PENDING, &xprt->sc_flags)) + goto out; + svc_xprt_enqueue(&xprt->sc_xprt); + goto out; + +flushed: + if (wc->status != IB_WC_WR_FLUSH_ERR) + pr_warn("svcrdma: receive: %s (%u/0x%x)\n", + ib_wc_status_msg(wc->status), + wc->status, wc->vendor_err); + set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags); + svc_rdma_put_context(ctxt, 1); + +out: + svc_xprt_put(&xprt->sc_xprt); } -/* - * Receive Queue Completion Handler - * - * Since an RQ completion handler is called on interrupt context, we - * need to defer the handling of the I/O to a tasklet - */ -static void rq_comp_handler(struct ib_cq *cq, void *cq_context) +static void svc_rdma_send_wc_common(struct svcxprt_rdma *xprt, + struct ib_wc *wc, + const char *opname) { - struct svcxprt_rdma *xprt = cq_context; - unsigned long flags; - - /* Guard against unconditional flush call for destroyed QP */ - if (atomic_read(&xprt->sc_xprt.xpt_ref.refcount)==0) - return; + if (wc->status != IB_WC_SUCCESS) + goto err; - /* - * Set the bit regardless of whether or not it's on the list - * because it may be on the list already due to an SQ - * completion. - */ - set_bit(RDMAXPRT_RQ_PENDING, &xprt->sc_flags); +out: + atomic_dec(&xprt->sc_sq_count); + wake_up(&xprt->sc_send_wait); + return; + +err: + set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags); + if (wc->status != IB_WC_WR_FLUSH_ERR) + pr_err("svcrdma: %s: %s (%u/0x%x)\n", + opname, ib_wc_status_msg(wc->status), + wc->status, wc->vendor_err); + goto out; +} - /* - * If this transport is not already on the DTO transport queue, - * add it - */ - spin_lock_irqsave(&dto_lock, flags); - if (list_empty(&xprt->sc_dto_q)) { - svc_xprt_get(&xprt->sc_xprt); - list_add_tail(&xprt->sc_dto_q, &dto_xprt_q); - } - spin_unlock_irqrestore(&dto_lock, flags); +static void svc_rdma_send_wc_common_put(struct ib_cq *cq, struct ib_wc *wc, + const char *opname) +{ + struct svcxprt_rdma *xprt = cq->cq_context; - /* Tasklet does all the work to avoid irqsave locks. */ - tasklet_schedule(&dto_tasklet); + svc_rdma_send_wc_common(xprt, wc, opname); + svc_xprt_put(&xprt->sc_xprt); } -/* - * rq_cq_reap - Process the RQ CQ. - * - * Take all completing WC off the CQE and enqueue the associated DTO - * context on the dto_q for the transport. +/** + * svc_rdma_wc_send - Invoked by RDMA provider for each polled Send WC + * @cq: completion queue + * @wc: completed WR * - * Note that caller must hold a transport reference. */ -static void rq_cq_reap(struct svcxprt_rdma *xprt) +void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc) { - int ret; - struct ib_wc wc; - struct svc_rdma_op_ctxt *ctxt = NULL; + struct ib_cqe *cqe = wc->wr_cqe; + struct svc_rdma_op_ctxt *ctxt; - if (!test_and_clear_bit(RDMAXPRT_RQ_PENDING, &xprt->sc_flags)) - return; + svc_rdma_send_wc_common_put(cq, wc, "send"); - ib_req_notify_cq(xprt->sc_rq_cq, IB_CQ_NEXT_COMP); - atomic_inc(&rdma_stat_rq_poll); + ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe); + svc_rdma_unmap_dma(ctxt); + svc_rdma_put_context(ctxt, 1); +} - while ((ret = ib_poll_cq(xprt->sc_rq_cq, 1, &wc)) > 0) { - ctxt = (struct svc_rdma_op_ctxt *)(unsigned long)wc.wr_id; - ctxt->wc_status = wc.status; - ctxt->byte_len = wc.byte_len; - svc_rdma_unmap_dma(ctxt); - if (wc.status != IB_WC_SUCCESS) { - /* Close the transport */ - dprintk("svcrdma: transport closing putting ctxt %p\n", ctxt); - set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags); - svc_rdma_put_context(ctxt, 1); - svc_xprt_put(&xprt->sc_xprt); - continue; - } - spin_lock_bh(&xprt->sc_rq_dto_lock); - list_add_tail(&ctxt->dto_q, &xprt->sc_rq_dto_q); - spin_unlock_bh(&xprt->sc_rq_dto_lock); - svc_xprt_put(&xprt->sc_xprt); - } +/** + * svc_rdma_wc_write - Invoked by RDMA provider for each polled Write WC + * @cq: completion queue + * @wc: completed WR + * + */ +void svc_rdma_wc_write(struct ib_cq *cq, struct ib_wc *wc) +{ + struct ib_cqe *cqe = wc->wr_cqe; + struct svc_rdma_op_ctxt *ctxt; - if (ctxt) - atomic_inc(&rdma_stat_rq_prod); + svc_rdma_send_wc_common_put(cq, wc, "write"); - set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags); - /* - * If data arrived before established event, - * don't enqueue. This defers RPC I/O until the - * RDMA connection is complete. - */ - if (!test_bit(RDMAXPRT_CONN_PENDING, &xprt->sc_flags)) - svc_xprt_enqueue(&xprt->sc_xprt); + ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe); + svc_rdma_unmap_dma(ctxt); + svc_rdma_put_context(ctxt, 0); } -/* - * Process a completion context +/** + * svc_rdma_wc_reg - Invoked by RDMA provider for each polled FASTREG WC + * @cq: completion queue + * @wc: completed WR + * */ -static void process_context(struct svcxprt_rdma *xprt, - struct svc_rdma_op_ctxt *ctxt) +void svc_rdma_wc_reg(struct ib_cq *cq, struct ib_wc *wc) { - struct svc_rdma_op_ctxt *read_hdr; - int free_pages = 0; - - svc_rdma_unmap_dma(ctxt); + svc_rdma_send_wc_common_put(cq, wc, "fastreg"); +} - switch (ctxt->wr_op) { - case IB_WR_SEND: - free_pages = 1; - break; +/** + * svc_rdma_wc_read - Invoked by RDMA provider for each polled Read WC + * @cq: completion queue + * @wc: completed WR + * + */ +void svc_rdma_wc_read(struct ib_cq *cq, struct ib_wc *wc) +{ + struct svcxprt_rdma *xprt = cq->cq_context; + struct ib_cqe *cqe = wc->wr_cqe; + struct svc_rdma_op_ctxt *ctxt; - case IB_WR_RDMA_WRITE: - break; + svc_rdma_send_wc_common(xprt, wc, "read"); - case IB_WR_RDMA_READ: - case IB_WR_RDMA_READ_WITH_INV: - svc_rdma_put_frmr(xprt, ctxt->frmr); + ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe); + svc_rdma_unmap_dma(ctxt); + svc_rdma_put_frmr(xprt, ctxt->frmr); - if (!test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) - break; + if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) { + struct svc_rdma_op_ctxt *read_hdr; read_hdr = ctxt->read_hdr; - svc_rdma_put_context(ctxt, 0); - - spin_lock_bh(&xprt->sc_rq_dto_lock); - set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags); + spin_lock(&xprt->sc_rq_dto_lock); list_add_tail(&read_hdr->dto_q, &xprt->sc_read_complete_q); - spin_unlock_bh(&xprt->sc_rq_dto_lock); - svc_xprt_enqueue(&xprt->sc_xprt); - return; + spin_unlock(&xprt->sc_rq_dto_lock); - default: - dprintk("svcrdma: unexpected completion opcode=%d\n", - ctxt->wr_op); - break; + set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags); + svc_xprt_enqueue(&xprt->sc_xprt); } - svc_rdma_put_context(ctxt, free_pages); + svc_rdma_put_context(ctxt, 0); + svc_xprt_put(&xprt->sc_xprt); } -/* - * Send Queue Completion Handler - potentially called on interrupt context. +/** + * svc_rdma_wc_inv - Invoked by RDMA provider for each polled LOCAL_INV WC + * @cq: completion queue + * @wc: completed WR * - * Note that caller must hold a transport reference. */ -static void sq_cq_reap(struct svcxprt_rdma *xprt) -{ - struct svc_rdma_op_ctxt *ctxt = NULL; - struct ib_wc wc_a[6]; - struct ib_wc *wc; - struct ib_cq *cq = xprt->sc_sq_cq; - int ret; - - memset(wc_a, 0, sizeof(wc_a)); - - if (!test_and_clear_bit(RDMAXPRT_SQ_PENDING, &xprt->sc_flags)) - return; - - ib_req_notify_cq(xprt->sc_sq_cq, IB_CQ_NEXT_COMP); - atomic_inc(&rdma_stat_sq_poll); - while ((ret = ib_poll_cq(cq, ARRAY_SIZE(wc_a), wc_a)) > 0) { - int i; - - for (i = 0; i < ret; i++) { - wc = &wc_a[i]; - if (wc->status != IB_WC_SUCCESS) { - dprintk("svcrdma: sq wc err status %s (%d)\n", - ib_wc_status_msg(wc->status), - wc->status); - - /* Close the transport */ - set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags); - } - - /* Decrement used SQ WR count */ - atomic_dec(&xprt->sc_sq_count); - wake_up(&xprt->sc_send_wait); - - ctxt = (struct svc_rdma_op_ctxt *) - (unsigned long)wc->wr_id; - if (ctxt) - process_context(xprt, ctxt); - - svc_xprt_put(&xprt->sc_xprt); - } - } - - if (ctxt) - atomic_inc(&rdma_stat_sq_prod); -} - -static void sq_comp_handler(struct ib_cq *cq, void *cq_context) +void svc_rdma_wc_inv(struct ib_cq *cq, struct ib_wc *wc) { - struct svcxprt_rdma *xprt = cq_context; - unsigned long flags; - - /* Guard against unconditional flush call for destroyed QP */ - if (atomic_read(&xprt->sc_xprt.xpt_ref.refcount)==0) - return; - - /* - * Set the bit regardless of whether or not it's on the list - * because it may be on the list already due to an RQ - * completion. - */ - set_bit(RDMAXPRT_SQ_PENDING, &xprt->sc_flags); - - /* - * If this transport is not already on the DTO transport queue, - * add it - */ - spin_lock_irqsave(&dto_lock, flags); - if (list_empty(&xprt->sc_dto_q)) { - svc_xprt_get(&xprt->sc_xprt); - list_add_tail(&xprt->sc_dto_q, &dto_xprt_q); - } - spin_unlock_irqrestore(&dto_lock, flags); - - /* Tasklet does all the work to avoid irqsave locks. */ - tasklet_schedule(&dto_tasklet); + svc_rdma_send_wc_common_put(cq, wc, "localInv"); } static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv, @@ -681,6 +585,7 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt, gfp_t flags) ctxt = svc_rdma_get_context(xprt); buflen = 0; ctxt->direction = DMA_FROM_DEVICE; + ctxt->cqe.done = svc_rdma_wc_receive; for (sge_no = 0; buflen < xprt->sc_max_req_size; sge_no++) { if (sge_no >= xprt->sc_max_sge) { pr_err("svcrdma: Too many sges (%d)\n", sge_no); @@ -705,7 +610,7 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt, gfp_t flags) recv_wr.next = NULL; recv_wr.sg_list = &ctxt->sge[0]; recv_wr.num_sge = ctxt->count; - recv_wr.wr_id = (u64)(unsigned long)ctxt; + recv_wr.wr_cqe = &ctxt->cqe; svc_xprt_get(&xprt->sc_xprt); ret = ib_post_recv(xprt->sc_qp, &recv_wr, &bad_recv_wr); @@ -722,6 +627,21 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt, gfp_t flags) return -ENOMEM; } +int svc_rdma_repost_recv(struct svcxprt_rdma *xprt, gfp_t flags) +{ + int ret = 0; + + ret = svc_rdma_post_recv(xprt, flags); + if (ret) { + pr_err("svcrdma: could not post a receive buffer, err=%d.\n", + ret); + pr_err("svcrdma: closing transport %p.\n", xprt); + set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags); + ret = -ENOTCONN; + } + return ret; +} + /* * This function handles the CONNECT_REQUEST event on a listening * endpoint. It is passed the cma_id for the _new_ connection. The context in @@ -1011,7 +931,6 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt) struct svcxprt_rdma *listen_rdma; struct svcxprt_rdma *newxprt = NULL; struct rdma_conn_param conn_param; - struct ib_cq_init_attr cq_attr = {}; struct ib_qp_init_attr qp_attr; struct ib_device *dev; unsigned int i; @@ -1069,22 +988,14 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt) dprintk("svcrdma: error creating PD for connect request\n"); goto errout; } - cq_attr.cqe = newxprt->sc_sq_depth; - newxprt->sc_sq_cq = ib_create_cq(dev, - sq_comp_handler, - cq_event_handler, - newxprt, - &cq_attr); + newxprt->sc_sq_cq = ib_alloc_cq(dev, newxprt, newxprt->sc_sq_depth, + 0, IB_POLL_SOFTIRQ); if (IS_ERR(newxprt->sc_sq_cq)) { dprintk("svcrdma: error creating SQ CQ for connect request\n"); goto errout; } - cq_attr.cqe = newxprt->sc_rq_depth; - newxprt->sc_rq_cq = ib_create_cq(dev, - rq_comp_handler, - cq_event_handler, - newxprt, - &cq_attr); + newxprt->sc_rq_cq = ib_alloc_cq(dev, newxprt, newxprt->sc_rq_depth, + 0, IB_POLL_SOFTIRQ); if (IS_ERR(newxprt->sc_rq_cq)) { dprintk("svcrdma: error creating RQ CQ for connect request\n"); goto errout; @@ -1173,13 +1084,6 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt) /* Swap out the handler */ newxprt->sc_cm_id->event_handler = rdma_cma_handler; - /* - * Arm the CQs for the SQ and RQ before accepting so we can't - * miss the first message - */ - ib_req_notify_cq(newxprt->sc_sq_cq, IB_CQ_NEXT_COMP); - ib_req_notify_cq(newxprt->sc_rq_cq, IB_CQ_NEXT_COMP); - /* Accept Connection */ set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags); memset(&conn_param, 0, sizeof conn_param); @@ -1319,10 +1223,10 @@ static void __svc_rdma_free(struct work_struct *work) ib_destroy_qp(rdma->sc_qp); if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq)) - ib_destroy_cq(rdma->sc_sq_cq); + ib_free_cq(rdma->sc_sq_cq); if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq)) - ib_destroy_cq(rdma->sc_rq_cq); + ib_free_cq(rdma->sc_rq_cq); if (rdma->sc_pd && !IS_ERR(rdma->sc_pd)) ib_dealloc_pd(rdma->sc_pd); @@ -1383,9 +1287,6 @@ int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr) spin_unlock_bh(&xprt->sc_lock); atomic_inc(&rdma_stat_sq_starve); - /* See if we can opportunistically reap SQ WR to make room */ - sq_cq_reap(xprt); - /* Wait until SQ WR available if SQ still full */ wait_event(xprt->sc_send_wait, atomic_read(&xprt->sc_sq_count) < @@ -1418,57 +1319,3 @@ int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr) } return ret; } - -void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp, - enum rpcrdma_errcode err) -{ - struct ib_send_wr err_wr; - struct page *p; - struct svc_rdma_op_ctxt *ctxt; - __be32 *va; - int length; - int ret; - - p = alloc_page(GFP_KERNEL); - if (!p) - return; - va = page_address(p); - - /* XDR encode error */ - length = svc_rdma_xdr_encode_error(xprt, rmsgp, err, va); - - ctxt = svc_rdma_get_context(xprt); - ctxt->direction = DMA_FROM_DEVICE; - ctxt->count = 1; - ctxt->pages[0] = p; - - /* Prepare SGE for local address */ - ctxt->sge[0].addr = ib_dma_map_page(xprt->sc_cm_id->device, - p, 0, length, DMA_FROM_DEVICE); - if (ib_dma_mapping_error(xprt->sc_cm_id->device, ctxt->sge[0].addr)) { - put_page(p); - svc_rdma_put_context(ctxt, 1); - return; - } - atomic_inc(&xprt->sc_dma_used); - ctxt->sge[0].lkey = xprt->sc_pd->local_dma_lkey; - ctxt->sge[0].length = length; - - /* Prepare SEND WR */ - memset(&err_wr, 0, sizeof err_wr); - ctxt->wr_op = IB_WR_SEND; - err_wr.wr_id = (unsigned long)ctxt; - err_wr.sg_list = ctxt->sge; - err_wr.num_sge = 1; - err_wr.opcode = IB_WR_SEND; - err_wr.send_flags = IB_SEND_SIGNALED; - - /* Post It */ - ret = svc_rdma_send(xprt, &err_wr); - if (ret) { - dprintk("svcrdma: Error %d posting send for protocol error\n", - ret); - svc_rdma_unmap_dma(ctxt); - svc_rdma_put_context(ctxt, 1); - } -} diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c index 878f1bfb1..f5ed9f982 100644 --- a/net/sunrpc/xprtrdma/verbs.c +++ b/net/sunrpc/xprtrdma/verbs.c @@ -112,89 +112,65 @@ rpcrdma_qp_async_error_upcall(struct ib_event *event, void *context) } } +/** + * rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC + * @cq: completion queue (ignored) + * @wc: completed WR + * + */ static void -rpcrdma_cq_async_error_upcall(struct ib_event *event, void *context) +rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc) { - struct rpcrdma_ep *ep = context; - - pr_err("RPC: %s: %s on device %s ep %p\n", - __func__, ib_event_msg(event->event), - event->device->name, context); - if (ep->rep_connected == 1) { - ep->rep_connected = -EIO; - rpcrdma_conn_func(ep); - wake_up_all(&ep->rep_connect_wait); - } + /* WARNING: Only wr_cqe and status are reliable at this point */ + if (wc->status != IB_WC_SUCCESS && wc->status != IB_WC_WR_FLUSH_ERR) + pr_err("rpcrdma: Send: %s (%u/0x%x)\n", + ib_wc_status_msg(wc->status), + wc->status, wc->vendor_err); } static void -rpcrdma_sendcq_process_wc(struct ib_wc *wc) +rpcrdma_receive_worker(struct work_struct *work) { - /* WARNING: Only wr_id and status are reliable at this point */ - if (wc->wr_id == RPCRDMA_IGNORE_COMPLETION) { - if (wc->status != IB_WC_SUCCESS && - wc->status != IB_WC_WR_FLUSH_ERR) - pr_err("RPC: %s: SEND: %s\n", - __func__, ib_wc_status_msg(wc->status)); - } else { - struct rpcrdma_mw *r; + struct rpcrdma_rep *rep = + container_of(work, struct rpcrdma_rep, rr_work); - r = (struct rpcrdma_mw *)(unsigned long)wc->wr_id; - r->mw_sendcompletion(wc); - } + rpcrdma_reply_handler(rep); } -/* The common case is a single send completion is waiting. By - * passing two WC entries to ib_poll_cq, a return code of 1 - * means there is exactly one WC waiting and no more. We don't - * have to invoke ib_poll_cq again to know that the CQ has been - * properly drained. +/* Perform basic sanity checking to avoid using garbage + * to update the credit grant value. */ static void -rpcrdma_sendcq_poll(struct ib_cq *cq) +rpcrdma_update_granted_credits(struct rpcrdma_rep *rep) { - struct ib_wc *pos, wcs[2]; - int count, rc; + struct rpcrdma_msg *rmsgp = rdmab_to_msg(rep->rr_rdmabuf); + struct rpcrdma_buffer *buffer = &rep->rr_rxprt->rx_buf; + u32 credits; - do { - pos = wcs; + if (rep->rr_len < RPCRDMA_HDRLEN_ERR) + return; - rc = ib_poll_cq(cq, ARRAY_SIZE(wcs), pos); - if (rc < 0) - break; + credits = be32_to_cpu(rmsgp->rm_credit); + if (credits == 0) + credits = 1; /* don't deadlock */ + else if (credits > buffer->rb_max_requests) + credits = buffer->rb_max_requests; - count = rc; - while (count-- > 0) - rpcrdma_sendcq_process_wc(pos++); - } while (rc == ARRAY_SIZE(wcs)); - return; + atomic_set(&buffer->rb_credits, credits); } -/* Handle provider send completion upcalls. +/** + * rpcrdma_receive_wc - Invoked by RDMA provider for each polled Receive WC + * @cq: completion queue (ignored) + * @wc: completed WR + * */ static void -rpcrdma_sendcq_upcall(struct ib_cq *cq, void *cq_context) +rpcrdma_receive_wc(struct ib_cq *cq, struct ib_wc *wc) { - do { - rpcrdma_sendcq_poll(cq); - } while (ib_req_notify_cq(cq, IB_CQ_NEXT_COMP | - IB_CQ_REPORT_MISSED_EVENTS) > 0); -} - -static void -rpcrdma_receive_worker(struct work_struct *work) -{ - struct rpcrdma_rep *rep = - container_of(work, struct rpcrdma_rep, rr_work); - - rpcrdma_reply_handler(rep); -} - -static void -rpcrdma_recvcq_process_wc(struct ib_wc *wc) -{ - struct rpcrdma_rep *rep = - (struct rpcrdma_rep *)(unsigned long)wc->wr_id; + struct ib_cqe *cqe = wc->wr_cqe; + struct rpcrdma_rep *rep = container_of(cqe, struct rpcrdma_rep, + rr_cqe); /* WARNING: Only wr_id and status are reliable at this point */ if (wc->status != IB_WC_SUCCESS) @@ -211,7 +187,8 @@ rpcrdma_recvcq_process_wc(struct ib_wc *wc) ib_dma_sync_single_for_cpu(rep->rr_device, rdmab_addr(rep->rr_rdmabuf), rep->rr_len, DMA_FROM_DEVICE); - prefetch(rdmab_to_msg(rep->rr_rdmabuf)); + + rpcrdma_update_granted_credits(rep); out_schedule: queue_work(rpcrdma_receive_wq, &rep->rr_work); @@ -219,57 +196,20 @@ out_schedule: out_fail: if (wc->status != IB_WC_WR_FLUSH_ERR) - pr_err("RPC: %s: rep %p: %s\n", - __func__, rep, ib_wc_status_msg(wc->status)); + pr_err("rpcrdma: Recv: %s (%u/0x%x)\n", + ib_wc_status_msg(wc->status), + wc->status, wc->vendor_err); rep->rr_len = RPCRDMA_BAD_LEN; goto out_schedule; } -/* The wc array is on stack: automatic memory is always CPU-local. - * - * struct ib_wc is 64 bytes, making the poll array potentially - * large. But this is at the bottom of the call chain. Further - * substantial work is done in another thread. - */ -static void -rpcrdma_recvcq_poll(struct ib_cq *cq) -{ - struct ib_wc *pos, wcs[4]; - int count, rc; - - do { - pos = wcs; - - rc = ib_poll_cq(cq, ARRAY_SIZE(wcs), pos); - if (rc < 0) - break; - - count = rc; - while (count-- > 0) - rpcrdma_recvcq_process_wc(pos++); - } while (rc == ARRAY_SIZE(wcs)); -} - -/* Handle provider receive completion upcalls. - */ -static void -rpcrdma_recvcq_upcall(struct ib_cq *cq, void *cq_context) -{ - do { - rpcrdma_recvcq_poll(cq); - } while (ib_req_notify_cq(cq, IB_CQ_NEXT_COMP | - IB_CQ_REPORT_MISSED_EVENTS) > 0); -} - static void rpcrdma_flush_cqs(struct rpcrdma_ep *ep) { struct ib_wc wc; while (ib_poll_cq(ep->rep_attr.recv_cq, 1, &wc) > 0) - rpcrdma_recvcq_process_wc(&wc); - while (ib_poll_cq(ep->rep_attr.send_cq, 1, &wc) > 0) - rpcrdma_sendcq_process_wc(&wc); + rpcrdma_receive_wc(NULL, &wc); } static int @@ -330,6 +270,7 @@ rpcrdma_conn_upcall(struct rdma_cm_id *id, struct rdma_cm_event *event) connected: dprintk("RPC: %s: %sconnected\n", __func__, connstate > 0 ? "" : "dis"); + atomic_set(&xprt->rx_buf.rb_credits, 1); ep->rep_connected = connstate; rpcrdma_conn_func(ep); wake_up_all(&ep->rep_connect_wait); @@ -560,9 +501,8 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, struct rpcrdma_create_data_internal *cdata) { struct ib_cq *sendcq, *recvcq; - struct ib_cq_init_attr cq_attr = {}; unsigned int max_qp_wr; - int rc, err; + int rc; if (ia->ri_device->attrs.max_sge < RPCRDMA_MAX_IOVS) { dprintk("RPC: %s: insufficient sge's available\n", @@ -614,9 +554,9 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, init_waitqueue_head(&ep->rep_connect_wait); INIT_DELAYED_WORK(&ep->rep_connect_worker, rpcrdma_connect_worker); - cq_attr.cqe = ep->rep_attr.cap.max_send_wr + 1; - sendcq = ib_create_cq(ia->ri_device, rpcrdma_sendcq_upcall, - rpcrdma_cq_async_error_upcall, NULL, &cq_attr); + sendcq = ib_alloc_cq(ia->ri_device, NULL, + ep->rep_attr.cap.max_send_wr + 1, + 0, IB_POLL_SOFTIRQ); if (IS_ERR(sendcq)) { rc = PTR_ERR(sendcq); dprintk("RPC: %s: failed to create send CQ: %i\n", @@ -624,16 +564,9 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, goto out1; } - rc = ib_req_notify_cq(sendcq, IB_CQ_NEXT_COMP); - if (rc) { - dprintk("RPC: %s: ib_req_notify_cq failed: %i\n", - __func__, rc); - goto out2; - } - - cq_attr.cqe = ep->rep_attr.cap.max_recv_wr + 1; - recvcq = ib_create_cq(ia->ri_device, rpcrdma_recvcq_upcall, - rpcrdma_cq_async_error_upcall, NULL, &cq_attr); + recvcq = ib_alloc_cq(ia->ri_device, NULL, + ep->rep_attr.cap.max_recv_wr + 1, + 0, IB_POLL_SOFTIRQ); if (IS_ERR(recvcq)) { rc = PTR_ERR(recvcq); dprintk("RPC: %s: failed to create recv CQ: %i\n", @@ -641,14 +574,6 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, goto out2; } - rc = ib_req_notify_cq(recvcq, IB_CQ_NEXT_COMP); - if (rc) { - dprintk("RPC: %s: ib_req_notify_cq failed: %i\n", - __func__, rc); - ib_destroy_cq(recvcq); - goto out2; - } - ep->rep_attr.send_cq = sendcq; ep->rep_attr.recv_cq = recvcq; @@ -673,10 +598,7 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia, return 0; out2: - err = ib_destroy_cq(sendcq); - if (err) - dprintk("RPC: %s: ib_destroy_cq returned %i\n", - __func__, err); + ib_free_cq(sendcq); out1: if (ia->ri_dma_mr) ib_dereg_mr(ia->ri_dma_mr); @@ -711,15 +633,8 @@ rpcrdma_ep_destroy(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia) ia->ri_id->qp = NULL; } - rc = ib_destroy_cq(ep->rep_attr.recv_cq); - if (rc) - dprintk("RPC: %s: ib_destroy_cq returned %i\n", - __func__, rc); - - rc = ib_destroy_cq(ep->rep_attr.send_cq); - if (rc) - dprintk("RPC: %s: ib_destroy_cq returned %i\n", - __func__, rc); + ib_free_cq(ep->rep_attr.recv_cq); + ib_free_cq(ep->rep_attr.send_cq); if (ia->ri_dma_mr) { rc = ib_dereg_mr(ia->ri_dma_mr); @@ -898,6 +813,7 @@ rpcrdma_create_req(struct rpcrdma_xprt *r_xprt) spin_lock(&buffer->rb_reqslock); list_add(&req->rl_all, &buffer->rb_allreqs); spin_unlock(&buffer->rb_reqslock); + req->rl_cqe.done = rpcrdma_wc_send; req->rl_buffer = &r_xprt->rx_buf; return req; } @@ -923,6 +839,7 @@ rpcrdma_create_rep(struct rpcrdma_xprt *r_xprt) } rep->rr_device = ia->ri_device; + rep->rr_cqe.done = rpcrdma_receive_wc; rep->rr_rxprt = r_xprt; INIT_WORK(&rep->rr_work, rpcrdma_receive_worker); return rep; @@ -943,6 +860,7 @@ rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt) buf->rb_max_requests = r_xprt->rx_data.max_requests; buf->rb_bc_srv_max_requests = 0; spin_lock_init(&buf->rb_lock); + atomic_set(&buf->rb_credits, 1); rc = ia->ri_ops->ro_init(r_xprt); if (rc) @@ -1259,7 +1177,7 @@ rpcrdma_ep_post(struct rpcrdma_ia *ia, } send_wr.next = NULL; - send_wr.wr_id = RPCRDMA_IGNORE_COMPLETION; + send_wr.wr_cqe = &req->rl_cqe; send_wr.sg_list = iov; send_wr.num_sge = req->rl_niovs; send_wr.opcode = IB_WR_SEND; @@ -1297,7 +1215,7 @@ rpcrdma_ep_post_recv(struct rpcrdma_ia *ia, int rc; recv_wr.next = NULL; - recv_wr.wr_id = (u64) (unsigned long) rep; + recv_wr.wr_cqe = &rep->rr_cqe; recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov; recv_wr.num_sge = 1; diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h index 38fe11b09..2ebc743cb 100644 --- a/net/sunrpc/xprtrdma/xprt_rdma.h +++ b/net/sunrpc/xprtrdma/xprt_rdma.h @@ -95,10 +95,6 @@ struct rpcrdma_ep { #define INIT_CQCOUNT(ep) atomic_set(&(ep)->rep_cqcount, (ep)->rep_cqinit) #define DECR_CQCOUNT(ep) atomic_sub_return(1, &(ep)->rep_cqcount) -/* Force completion handler to ignore the signal - */ -#define RPCRDMA_IGNORE_COMPLETION (0ULL) - /* Pre-allocate extra Work Requests for handling backward receives * and sends. This is a fixed value because the Work Queues are * allocated when the forward channel is set up. @@ -171,6 +167,7 @@ rdmab_to_msg(struct rpcrdma_regbuf *rb) struct rpcrdma_buffer; struct rpcrdma_rep { + struct ib_cqe rr_cqe; unsigned int rr_len; struct ib_device *rr_device; struct rpcrdma_xprt *rr_rxprt; @@ -204,11 +201,11 @@ struct rpcrdma_frmr { struct scatterlist *sg; int sg_nents; struct ib_mr *fr_mr; + struct ib_cqe fr_cqe; enum rpcrdma_frmr_state fr_state; + struct completion fr_linv_done; struct work_struct fr_work; struct rpcrdma_xprt *fr_xprt; - bool fr_waiter; - struct completion fr_linv_done;; union { struct ib_reg_wr fr_regwr; struct ib_send_wr fr_invwr; @@ -224,8 +221,7 @@ struct rpcrdma_mw { union { struct rpcrdma_fmr fmr; struct rpcrdma_frmr frmr; - } r; - void (*mw_sendcompletion)(struct ib_wc *); + }; struct list_head mw_list; struct list_head mw_all; }; @@ -281,6 +277,7 @@ struct rpcrdma_req { struct rpcrdma_regbuf *rl_sendbuf; struct rpcrdma_mr_seg rl_segments[RPCRDMA_MAX_SEGS]; + struct ib_cqe rl_cqe; struct list_head rl_all; bool rl_backchannel; }; @@ -311,6 +308,7 @@ struct rpcrdma_buffer { struct list_head rb_send_bufs; struct list_head rb_recv_bufs; u32 rb_max_requests; + atomic_t rb_credits; /* most recent credit grant */ u32 rb_bc_srv_max_requests; spinlock_t rb_reqslock; /* protect rb_allreqs */ |