From owner-svn-src-stable-10@freebsd.org Tue Jul 4 01:28:18 2017 Return-Path: Delivered-To: svn-src-stable-10@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id D1A969F3B70; Tue, 4 Jul 2017 01:28:18 +0000 (UTC) (envelope-from rmacklem@FreeBSD.org) Received: from repo.freebsd.org (repo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id AD7B564604; Tue, 4 Jul 2017 01:28:18 +0000 (UTC) (envelope-from rmacklem@FreeBSD.org) Received: from repo.freebsd.org ([127.0.1.37]) by repo.freebsd.org (8.15.2/8.15.2) with ESMTP id v641SHjK016543; Tue, 4 Jul 2017 01:28:17 GMT (envelope-from rmacklem@FreeBSD.org) Received: (from rmacklem@localhost) by repo.freebsd.org (8.15.2/8.15.2/Submit) id v641SHCx016540; Tue, 4 Jul 2017 01:28:17 GMT (envelope-from rmacklem@FreeBSD.org) Message-Id: <201707040128.v641SHCx016540@repo.freebsd.org> X-Authentication-Warning: repo.freebsd.org: rmacklem set sender to rmacklem@FreeBSD.org using -f From: Rick Macklem Date: Tue, 4 Jul 2017 01:28:17 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-10@freebsd.org Subject: svn commit: r320637 - in stable/10/sys: fs/nfs fs/nfsclient sys X-SVN-Group: stable-10 X-SVN-Commit-Author: rmacklem X-SVN-Commit-Paths: in stable/10/sys: fs/nfs fs/nfsclient sys X-SVN-Commit-Revision: 320637 X-SVN-Commit-Repository: base MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-stable-10@freebsd.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: SVN commit messages for only the 10-stable src tree List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 04 Jul 2017 01:28:18 -0000 Author: rmacklem Date: Tue Jul 4 01:28:17 2017 New Revision: 320637 URL: https://svnweb.freebsd.org/changeset/base/320637 Log: MFC: r320062, r320070, r320126 This is a partial merge of only the NFS changes and not the maxbcachebuf tunable. The NFS client changes make the code handle different I/O sizes more correctly. However, with the limit at 64K, they are not actually necessary. This MFC is mainly being done so that subsequent MFCs to the NFS code will merge easily. Modified: stable/10/sys/fs/nfs/nfs_commonkrpc.c stable/10/sys/fs/nfsclient/nfs_clrpcops.c stable/10/sys/sys/param.h Directory Properties: stable/10/ (props changed) Modified: stable/10/sys/fs/nfs/nfs_commonkrpc.c ============================================================================== --- stable/10/sys/fs/nfs/nfs_commonkrpc.c Tue Jul 4 01:23:36 2017 (r320636) +++ stable/10/sys/fs/nfs/nfs_commonkrpc.c Tue Jul 4 01:28:17 2017 (r320637) @@ -162,7 +162,7 @@ newnfs_connect(struct nfsmount *nmp, struct nfssockreq struct ucred *cred, NFSPROC_T *p, int callback_retry_mult) { int rcvreserve, sndreserve; - int pktscale; + int pktscale, pktscalesav; struct sockaddr *saddr; struct ucred *origcred; CLIENT *client; @@ -211,6 +211,7 @@ newnfs_connect(struct nfsmount *nmp, struct nfssockreq pktscale = 2; if (pktscale > 64) pktscale = 64; + pktscalesav = pktscale; /* * soreserve() can fail if sb_max is too small, so shrink pktscale * and try again if there is an error. @@ -229,8 +230,12 @@ newnfs_connect(struct nfsmount *nmp, struct nfssockreq goto out; } do { - if (error != 0 && pktscale > 2) + if (error != 0 && pktscale > 2) { + if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM && + pktscale == pktscalesav) + printf("Consider increasing kern.ipc.maxsockbuf\n"); pktscale--; + } if (nrp->nr_sotype == SOCK_DGRAM) { if (nmp != NULL) { sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) * @@ -244,15 +249,19 @@ newnfs_connect(struct nfsmount *nmp, struct nfssockreq if (nrp->nr_sotype != SOCK_STREAM) panic("nfscon sotype"); if (nmp != NULL) { - sndreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR + + sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR + sizeof (u_int32_t)) * pktscale; - rcvreserve = (NFS_MAXBSIZE + NFS_MAXPKTHDR + + rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR + sizeof (u_int32_t)) * pktscale; } else { sndreserve = rcvreserve = 1024 * pktscale; } } error = soreserve(so, sndreserve, rcvreserve); + if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM && + pktscale <= 2) + printf("Must increase kern.ipc.maxsockbuf or reduce" + " rsize, wsize\n"); } while (error != 0 && pktscale > 2); soclose(so); if (error) { Modified: stable/10/sys/fs/nfsclient/nfs_clrpcops.c ============================================================================== --- stable/10/sys/fs/nfsclient/nfs_clrpcops.c Tue Jul 4 01:23:36 2017 (r320636) +++ stable/10/sys/fs/nfsclient/nfs_clrpcops.c Tue Jul 4 01:28:17 2017 (r320637) @@ -4612,7 +4612,7 @@ nfsrpc_createsession(struct nfsmount *nmp, struct nfsc struct nfssockreq *nrp, uint32_t sequenceid, int mds, struct ucred *cred, NFSPROC_T *p) { - uint32_t crflags, *tl; + uint32_t crflags, maxval, *tl; struct nfsrv_descript nfsd; struct nfsrv_descript *nd = &nfsd; int error, irdcnt; @@ -4630,8 +4630,8 @@ nfsrpc_createsession(struct nfsmount *nmp, struct nfsc /* Fill in fore channel attributes. */ NFSM_BUILD(tl, uint32_t *, 7 * NFSX_UNSIGNED); *tl++ = 0; /* Header pad size */ - *tl++ = txdr_unsigned(100000); /* Max request size */ - *tl++ = txdr_unsigned(100000); /* Max response size */ + *tl++ = txdr_unsigned(nmp->nm_wsize + NFS_MAXXDR);/* Max request size */ + *tl++ = txdr_unsigned(nmp->nm_rsize + NFS_MAXXDR);/* Max reply size */ *tl++ = txdr_unsigned(4096); /* Max response size cached */ *tl++ = txdr_unsigned(20); /* Max operations */ *tl++ = txdr_unsigned(64); /* Max slots */ @@ -4678,7 +4678,26 @@ nfsrpc_createsession(struct nfsmount *nmp, struct nfsc /* Get the fore channel slot count. */ NFSM_DISSECT(tl, uint32_t *, 7 * NFSX_UNSIGNED); - tl += 3; /* Skip the other counts. */ + tl++; /* Skip the header pad size. */ + + /* Make sure nm_wsize is small enough. */ + maxval = fxdr_unsigned(uint32_t, *tl++); + while (maxval < nmp->nm_wsize + NFS_MAXXDR) { + if (nmp->nm_wsize > 8096) + nmp->nm_wsize /= 2; + else + break; + } + + /* Make sure nm_rsize is small enough. */ + maxval = fxdr_unsigned(uint32_t, *tl++); + while (maxval < nmp->nm_rsize + NFS_MAXXDR) { + if (nmp->nm_rsize > 8096) + nmp->nm_rsize /= 2; + else + break; + } + sep->nfsess_maxcache = fxdr_unsigned(int, *tl++); tl++; sep->nfsess_foreslots = fxdr_unsigned(uint16_t, *tl++); Modified: stable/10/sys/sys/param.h ============================================================================== --- stable/10/sys/sys/param.h Tue Jul 4 01:23:36 2017 (r320636) +++ stable/10/sys/sys/param.h Tue Jul 4 01:28:17 2017 (r320637) @@ -242,9 +242,7 @@ * Filesystems can of course request smaller chunks. Actual * backing memory uses a chunk size of a page (PAGE_SIZE). * The default value here can be overridden on a per-architecture - * basis by defining it in . This should - * probably be done to increase its value, when MAXBCACHEBUF is - * defined as a larger value in . + * basis by defining it in . * * If you make BKVASIZE too small you risk seriously fragmenting * the buffer KVM map which may slow things down a bit. If you