From d0687be7c7ae21461da4438d5fd059b48487bfe1 Mon Sep 17 00:00:00 2001 From: Steve Wise Date: Fri, 3 Apr 2009 15:18:24 -0500 Subject: svcrdma: Fix dma map direction for rdma read targets The nfs server rdma transport was mapping rdma read target pages for TO_DEVICE instead of FROM_DEVICE. This causes data corruption on non cache-coherent systems if frmrs are used. Signed-off-by: Steve Wise Signed-off-by: J. Bruce Fields --- net/sunrpc/xprtrdma/svc_rdma_recvfrom.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'net/sunrpc') diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c index 629a287..42a6f9f 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c +++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c @@ -265,7 +265,7 @@ static int fast_reg_read_chunks(struct svcxprt_rdma *xprt, frmr->page_list->page_list[page_no] = ib_dma_map_single(xprt->sc_cm_id->device, page_address(rqstp->rq_arg.pages[page_no]), - PAGE_SIZE, DMA_TO_DEVICE); + PAGE_SIZE, DMA_FROM_DEVICE); if (ib_dma_mapping_error(xprt->sc_cm_id->device, frmr->page_list->page_list[page_no])) goto fatal_err; -- cgit v1.1 From f75e6745aa3084124ae1434fd7629853bdaf6798 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Tue, 21 Apr 2009 17:18:20 -0400 Subject: SUNRPC: Fix the problem of EADDRNOTAVAIL syslog floods on reconnect See http://bugzilla.kernel.org/show_bug.cgi?id=13034 If the port gets into a TIME_WAIT state, then we cannot reconnect without binding to a new port. Tested-by: Petr Vandrovec Tested-by: Jean Delvare Signed-off-by: Trond Myklebust Signed-off-by: Linus Torvalds --- net/sunrpc/xprt.c | 6 ++---- net/sunrpc/xprtsock.c | 26 +++++++++++++++++++++----- 2 files changed, 23 insertions(+), 9 deletions(-) (limited to 'net/sunrpc') diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c index a0bfe53..06ca058 100644 --- a/net/sunrpc/xprt.c +++ b/net/sunrpc/xprt.c @@ -672,10 +672,8 @@ xprt_init_autodisconnect(unsigned long data) if (test_and_set_bit(XPRT_LOCKED, &xprt->state)) goto out_abort; spin_unlock(&xprt->transport_lock); - if (xprt_connecting(xprt)) - xprt_release_write(xprt, NULL); - else - queue_work(rpciod_workqueue, &xprt->task_cleanup); + set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); + queue_work(rpciod_workqueue, &xprt->task_cleanup); return; out_abort: spin_unlock(&xprt->transport_lock); diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c index d40ff50..e185961 100644 --- a/net/sunrpc/xprtsock.c +++ b/net/sunrpc/xprtsock.c @@ -807,6 +807,9 @@ static void xs_reset_transport(struct sock_xprt *transport) * * This is used when all requests are complete; ie, no DRC state remains * on the server we want to save. + * + * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with + * xs_reset_transport() zeroing the socket from underneath a writer. */ static void xs_close(struct rpc_xprt *xprt) { @@ -824,6 +827,14 @@ static void xs_close(struct rpc_xprt *xprt) xprt_disconnect_done(xprt); } +static void xs_tcp_close(struct rpc_xprt *xprt) +{ + if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state)) + xs_close(xprt); + else + xs_tcp_shutdown(xprt); +} + /** * xs_destroy - prepare to shutdown a transport * @xprt: doomed transport @@ -1772,6 +1783,15 @@ static void xs_tcp_setup_socket(struct rpc_xprt *xprt, xprt, -status, xprt_connected(xprt), sock->sk->sk_state); switch (status) { + default: + printk("%s: connect returned unhandled error %d\n", + __func__, status); + case -EADDRNOTAVAIL: + /* We're probably in TIME_WAIT. Get rid of existing socket, + * and retry + */ + set_bit(XPRT_CONNECTION_CLOSE, &xprt->state); + xprt_force_disconnect(xprt); case -ECONNREFUSED: case -ECONNRESET: case -ENETUNREACH: @@ -1782,10 +1802,6 @@ static void xs_tcp_setup_socket(struct rpc_xprt *xprt, xprt_clear_connecting(xprt); return; } - /* get rid of existing socket, and retry */ - xs_tcp_shutdown(xprt); - printk("%s: connect returned unhandled error %d\n", - __func__, status); out_eagain: status = -EAGAIN; out: @@ -1994,7 +2010,7 @@ static struct rpc_xprt_ops xs_tcp_ops = { .buf_free = rpc_free, .send_request = xs_tcp_send_request, .set_retrans_timeout = xprt_set_retrans_timeout_def, - .close = xs_tcp_shutdown, + .close = xs_tcp_close, .destroy = xs_destroy, .print_stats = xs_tcp_print_stats, }; -- cgit v1.1 From 21515e46bc6a6279dd13f6c01898ada9720100a3 Mon Sep 17 00:00:00 2001 From: Steve Wise Date: Wed, 29 Apr 2009 14:14:00 -0500 Subject: svcrdma: clean up error paths. These fixes resolved crashes due to resource leak BUG_ON checks. The resource leaks were detected by introducing asynchronous transport errors. Signed-off-by: Steve Wise Signed-off-by: Tom Tucker Signed-off-by: J. Bruce Fields --- net/sunrpc/xprtrdma/svc_rdma_sendto.c | 3 +++ net/sunrpc/xprtrdma/svc_rdma_transport.c | 3 ++- 2 files changed, 5 insertions(+), 1 deletion(-) (limited to 'net/sunrpc') diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c index 6c26a67..8b510c5 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c +++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c @@ -183,6 +183,7 @@ static int fast_reg_xdr(struct svcxprt_rdma *xprt, fatal_err: printk("svcrdma: Error fast registering memory for xprt %p\n", xprt); + vec->frmr = NULL; svc_rdma_put_frmr(xprt, frmr); return -EIO; } @@ -516,6 +517,7 @@ static int send_reply(struct svcxprt_rdma *rdma, "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_frmr(rdma, vec->frmr); svc_rdma_put_context(ctxt, 0); return -ENOTCONN; } @@ -606,6 +608,7 @@ static int send_reply(struct svcxprt_rdma *rdma, return 0; err: + svc_rdma_unmap_dma(ctxt); svc_rdma_put_frmr(rdma, vec->frmr); svc_rdma_put_context(ctxt, 1); return -EIO; diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c index 3d810e7..4b0c2fa 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_transport.c +++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c @@ -520,8 +520,9 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt) svc_xprt_get(&xprt->sc_xprt); ret = ib_post_recv(xprt->sc_qp, &recv_wr, &bad_recv_wr); if (ret) { - svc_xprt_put(&xprt->sc_xprt); + svc_rdma_unmap_dma(ctxt); svc_rdma_put_context(ctxt, 1); + svc_xprt_put(&xprt->sc_xprt); } return ret; -- cgit v1.1 From 68743082b560067e3e93eab8b2568f238e486865 Mon Sep 17 00:00:00 2001 From: Vu Pham Date: Tue, 26 May 2009 14:51:00 -0400 Subject: XPRTRDMA: fix client rpcrdma FRMR registration on mlx4 devices mlx4/connectX FRMR requires local write enable together with remote rdma write enable. This fixes NFS/RDMA operation over the ConnectX Infiniband HCA in the default memreg mode. Signed-off-by: Vu Pham Signed-off-by: Tom Talpey Signed-off-by: Trond Myklebust --- net/sunrpc/xprtrdma/verbs.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'net/sunrpc') diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c index 3b21e0c..465aafc 100644 --- a/net/sunrpc/xprtrdma/verbs.c +++ b/net/sunrpc/xprtrdma/verbs.c @@ -1495,7 +1495,8 @@ rpcrdma_register_frmr_external(struct rpcrdma_mr_seg *seg, frmr_wr.wr.fast_reg.page_shift = PAGE_SHIFT; frmr_wr.wr.fast_reg.length = i << PAGE_SHIFT; frmr_wr.wr.fast_reg.access_flags = (writing ? - IB_ACCESS_REMOTE_WRITE : IB_ACCESS_REMOTE_READ); + IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE : + IB_ACCESS_REMOTE_READ); frmr_wr.wr.fast_reg.rkey = seg1->mr_chunk.rl_mw->r.frmr.fr_mr->rkey; DECR_CQCOUNT(&r_xprt->rx_ep); -- cgit v1.1 From 7f4218354fe312b327af06c3d8c95ed5f214c8ca Mon Sep 17 00:00:00 2001 From: "J. Bruce Fields" Date: Wed, 27 May 2009 18:51:06 -0400 Subject: nfsd: Revert "svcrpc: take advantage of tcp autotuning" This reverts commit 47a14ef1af48c696b214ac168f056ddc79793d0e "svcrpc: take advantage of tcp autotuning", which uncovered some further problems in the server rpc code, causing significant performance regressions in common cases. We will likely reinstate this patch after releasing 2.6.30 and applying some work on the underlying fixes to the problem (developed by Trond). Reported-by: Jeff Moyer Cc: Olga Kornievskaia Cc: Jim Rees Cc: Trond Myklebust Signed-off-by: J. Bruce Fields --- net/sunrpc/svcsock.c | 35 ++++++++++++++++++++++++++++------- 1 file changed, 28 insertions(+), 7 deletions(-) (limited to 'net/sunrpc') diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c index af31988..9d50423 100644 --- a/net/sunrpc/svcsock.c +++ b/net/sunrpc/svcsock.c @@ -345,6 +345,7 @@ static void svc_sock_setbufsize(struct socket *sock, unsigned int snd, lock_sock(sock->sk); sock->sk->sk_sndbuf = snd * 2; sock->sk->sk_rcvbuf = rcv * 2; + sock->sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; release_sock(sock->sk); #endif } @@ -796,6 +797,23 @@ static int svc_tcp_recvfrom(struct svc_rqst *rqstp) test_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags), test_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags)); + if (test_and_clear_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags)) + /* sndbuf needs to have room for one request + * per thread, otherwise we can stall even when the + * network isn't a bottleneck. + * + * We count all threads rather than threads in a + * particular pool, which provides an upper bound + * on the number of threads which will access the socket. + * + * rcvbuf just needs to be able to hold a few requests. + * Normally they will be removed from the queue + * as soon a a complete request arrives. + */ + svc_sock_setbufsize(svsk->sk_sock, + (serv->sv_nrthreads+3) * serv->sv_max_mesg, + 3 * serv->sv_max_mesg); + clear_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); /* Receive data. If we haven't got the record length yet, get @@ -1043,6 +1061,15 @@ static void svc_tcp_init(struct svc_sock *svsk, struct svc_serv *serv) tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF; + /* initialise setting must have enough space to + * receive and respond to one request. + * svc_tcp_recvfrom will re-adjust if necessary + */ + svc_sock_setbufsize(svsk->sk_sock, + 3 * svsk->sk_xprt.xpt_server->sv_max_mesg, + 3 * svsk->sk_xprt.xpt_server->sv_max_mesg); + + set_bit(XPT_CHNGBUF, &svsk->sk_xprt.xpt_flags); set_bit(XPT_DATA, &svsk->sk_xprt.xpt_flags); if (sk->sk_state != TCP_ESTABLISHED) set_bit(XPT_CLOSE, &svsk->sk_xprt.xpt_flags); @@ -1112,14 +1139,8 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv, /* Initialize the socket */ if (sock->type == SOCK_DGRAM) svc_udp_init(svsk, serv); - else { - /* initialise setting must have enough space to - * receive and respond to one request. - */ - svc_sock_setbufsize(svsk->sk_sock, 4 * serv->sv_max_mesg, - 4 * serv->sv_max_mesg); + else svc_tcp_init(svsk, serv); - } dprintk("svc: svc_setup_socket created %p (inet %p)\n", svsk, svsk->sk_sk); -- cgit v1.1 From 98779be861a05c4cb75bed916df72ec0cba8b53d Mon Sep 17 00:00:00 2001 From: Steve Wise Date: Thu, 14 May 2009 16:34:28 -0500 Subject: svcrdma: dma unmap the correct length for the RPCRDMA header page. The svcrdma module was incorrectly unmapping the RPCRDMA header page. On IBM pserver systems this causes a resource leak that results in running out of bus address space (10 cthon iterations will reproduce it). The code was mapping the full page but only unmapping the actual header length. The fix is to only map the header length. I also cleaned up the use of ib_dma_map_page() calls since the unmap logic always uses ib_dma_unmap_single(). I made these symmetrical. Signed-off-by: Steve Wise Signed-off-by: Tom Tucker Signed-off-by: J. Bruce Fields --- net/sunrpc/xprtrdma/svc_rdma_sendto.c | 12 ++++++------ net/sunrpc/xprtrdma/svc_rdma_transport.c | 10 +++++----- 2 files changed, 11 insertions(+), 11 deletions(-) (limited to 'net/sunrpc') diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c index 8b510c5..f11be72 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c +++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c @@ -128,7 +128,8 @@ static int fast_reg_xdr(struct svcxprt_rdma *xprt, page_bytes -= sge_bytes; frmr->page_list->page_list[page_no] = - ib_dma_map_page(xprt->sc_cm_id->device, page, 0, + ib_dma_map_single(xprt->sc_cm_id->device, + page_address(page), PAGE_SIZE, DMA_TO_DEVICE); if (ib_dma_mapping_error(xprt->sc_cm_id->device, frmr->page_list->page_list[page_no])) @@ -532,18 +533,17 @@ static int send_reply(struct svcxprt_rdma *rdma, clear_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags); /* Prepare the SGE for the RPCRDMA Header */ + ctxt->sge[0].lkey = rdma->sc_dma_lkey; + ctxt->sge[0].length = svc_rdma_xdr_get_reply_hdr_len(rdma_resp); ctxt->sge[0].addr = - ib_dma_map_page(rdma->sc_cm_id->device, - page, 0, PAGE_SIZE, DMA_TO_DEVICE); + ib_dma_map_single(rdma->sc_cm_id->device, page_address(page), + ctxt->sge[0].length, DMA_TO_DEVICE); if (ib_dma_mapping_error(rdma->sc_cm_id->device, ctxt->sge[0].addr)) goto err; atomic_inc(&rdma->sc_dma_used); ctxt->direction = DMA_TO_DEVICE; - ctxt->sge[0].length = svc_rdma_xdr_get_reply_hdr_len(rdma_resp); - ctxt->sge[0].lkey = rdma->sc_dma_lkey; - /* Determine how many of our SGE are to be transmitted */ for (sge_no = 1; byte_count && sge_no < vec->count; sge_no++) { sge_bytes = min_t(size_t, vec->sge[sge_no].iov_len, byte_count); diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c index 4b0c2fa..5151f9f 100644 --- a/net/sunrpc/xprtrdma/svc_rdma_transport.c +++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c @@ -500,8 +500,8 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt) BUG_ON(sge_no >= xprt->sc_max_sge); page = svc_rdma_get_page(); ctxt->pages[sge_no] = page; - pa = ib_dma_map_page(xprt->sc_cm_id->device, - page, 0, PAGE_SIZE, + pa = ib_dma_map_single(xprt->sc_cm_id->device, + page_address(page), PAGE_SIZE, DMA_FROM_DEVICE); if (ib_dma_mapping_error(xprt->sc_cm_id->device, pa)) goto err_put_ctxt; @@ -1315,8 +1315,8 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp, length = svc_rdma_xdr_encode_error(xprt, rmsgp, err, va); /* Prepare SGE for local address */ - sge.addr = ib_dma_map_page(xprt->sc_cm_id->device, - p, 0, PAGE_SIZE, DMA_FROM_DEVICE); + sge.addr = ib_dma_map_single(xprt->sc_cm_id->device, + page_address(p), PAGE_SIZE, DMA_FROM_DEVICE); if (ib_dma_mapping_error(xprt->sc_cm_id->device, sge.addr)) { put_page(p); return; @@ -1343,7 +1343,7 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp, if (ret) { dprintk("svcrdma: Error %d posting send for protocol error\n", ret); - ib_dma_unmap_page(xprt->sc_cm_id->device, + ib_dma_unmap_single(xprt->sc_cm_id->device, sge.addr, PAGE_SIZE, DMA_FROM_DEVICE); svc_rdma_put_context(ctxt, 1); -- cgit v1.1