Sun Apr 24 16:23:49 2011 UTC ()
Pull up following revision(s) (requested by tls in ticket #1600):
	sys/nfs/nfs_socket.c: revision 1.189
As suggested by matt@: change socket buffer reservations for NFS send/receive
to 3 times max RPC size rather than 2 times.  Avoids nasty TCP stalls observed
at Panix.  Will require increase to sbmax via sysctl for those running really
huge NFS rsize/wsize (>64K).


(riz)
diff -r1.173.4.8 -r1.173.4.9 src/sys/nfs/nfs_socket.c
cvs diff -r1.173.4.8 -r1.173.4.9 src/sys/nfs/nfs_socket.c (expand / switch to unified diff)

--- src/sys/nfs/nfs_socket.c 2011/03/29 19:47:37 1.173.4.8
+++ src/sys/nfs/nfs_socket.c 2011/04/24 16:23:49 1.173.4.9
@@ -1,14 +1,14 @@ @@ -1,14 +1,14 @@
1/* $NetBSD: nfs_socket.c,v 1.173.4.8 2011/03/29 19:47:37 riz Exp $ */ 1/* $NetBSD: nfs_socket.c,v 1.173.4.9 2011/04/24 16:23:49 riz Exp $ */
2 2
3/* 3/*
4 * Copyright (c) 1989, 1991, 1993, 1995 4 * Copyright (c) 1989, 1991, 1993, 1995
5 * The Regents of the University of California. All rights reserved. 5 * The Regents of the University of California. All rights reserved.
6 * 6 *
7 * This code is derived from software contributed to Berkeley by 7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph. 8 * Rick Macklem at The University of Guelph.
9 * 9 *
10 * Redistribution and use in source and binary forms, with or without 10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions 11 * modification, are permitted provided that the following conditions
12 * are met: 12 * are met:
13 * 1. Redistributions of source code must retain the above copyright 13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer. 14 * notice, this list of conditions and the following disclaimer.
@@ -29,27 +29,27 @@ @@ -29,27 +29,27 @@
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE. 32 * SUCH DAMAGE.
33 * 33 *
34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95 34 * @(#)nfs_socket.c 8.5 (Berkeley) 3/30/95
35 */ 35 */
36 36
37/* 37/*
38 * Socket operations for use by nfs 38 * Socket operations for use by nfs
39 */ 39 */
40 40
41#include <sys/cdefs.h> 41#include <sys/cdefs.h>
42__KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.173.4.8 2011/03/29 19:47:37 riz Exp $"); 42__KERNEL_RCSID(0, "$NetBSD: nfs_socket.c,v 1.173.4.9 2011/04/24 16:23:49 riz Exp $");
43 43
44#include "fs_nfs.h" 44#include "fs_nfs.h"
45#include "opt_nfs.h" 45#include "opt_nfs.h"
46#include "opt_nfsserver.h" 46#include "opt_nfsserver.h"
47#include "opt_mbuftrace.h" 47#include "opt_mbuftrace.h"
48#include "opt_inet.h" 48#include "opt_inet.h"
49 49
50#include <sys/param.h> 50#include <sys/param.h>
51#include <sys/systm.h> 51#include <sys/systm.h>
52#include <sys/evcnt.h> 52#include <sys/evcnt.h>
53#include <sys/callout.h> 53#include <sys/callout.h>
54#include <sys/proc.h> 54#include <sys/proc.h>
55#include <sys/mount.h> 55#include <sys/mount.h>
@@ -311,51 +311,51 @@ nfs_connect(nmp, rep, l) @@ -311,51 +311,51 @@ nfs_connect(nmp, rep, l)
311 } 311 }
312 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { 312 if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) {
313 so->so_rcv.sb_timeo = (5 * hz); 313 so->so_rcv.sb_timeo = (5 * hz);
314 so->so_snd.sb_timeo = (5 * hz); 314 so->so_snd.sb_timeo = (5 * hz);
315 } else { 315 } else {
316 /* 316 /*
317 * enable receive timeout to detect server crash and reconnect. 317 * enable receive timeout to detect server crash and reconnect.
318 * otherwise, we can be stuck in soreceive forever. 318 * otherwise, we can be stuck in soreceive forever.
319 */ 319 */
320 so->so_rcv.sb_timeo = (5 * hz); 320 so->so_rcv.sb_timeo = (5 * hz);
321 so->so_snd.sb_timeo = 0; 321 so->so_snd.sb_timeo = 0;
322 } 322 }
323 if (nmp->nm_sotype == SOCK_DGRAM) { 323 if (nmp->nm_sotype == SOCK_DGRAM) {
324 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 324 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 3;
325 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 325 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
326 NFS_MAXPKTHDR) * 2; 326 NFS_MAXPKTHDR) * 2;
327 } else if (nmp->nm_sotype == SOCK_SEQPACKET) { 327 } else if (nmp->nm_sotype == SOCK_SEQPACKET) {
328 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 2; 328 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 3;
329 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) + 329 rcvreserve = (max(nmp->nm_rsize, nmp->nm_readdirsize) +
330 NFS_MAXPKTHDR) * 2; 330 NFS_MAXPKTHDR) * 3;
331 } else { 331 } else {
332 sounlock(so); 332 sounlock(so);
333 if (nmp->nm_sotype != SOCK_STREAM) 333 if (nmp->nm_sotype != SOCK_STREAM)
334 panic("nfscon sotype"); 334 panic("nfscon sotype");
335 if (so->so_proto->pr_flags & PR_CONNREQUIRED) { 335 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
336 val = 1; 336 val = 1;
337 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val, 337 so_setsockopt(NULL, so, SOL_SOCKET, SO_KEEPALIVE, &val,
338 sizeof(val)); 338 sizeof(val));
339 } 339 }
340 if (so->so_proto->pr_protocol == IPPROTO_TCP) { 340 if (so->so_proto->pr_protocol == IPPROTO_TCP) {
341 val = 1; 341 val = 1;
342 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val, 342 so_setsockopt(NULL, so, IPPROTO_TCP, TCP_NODELAY, &val,
343 sizeof(val)); 343 sizeof(val));
344 } 344 }
345 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + 345 sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR +
346 sizeof (u_int32_t)) * 2; 346 sizeof (u_int32_t)) * 3;
347 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + 347 rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR +
348 sizeof (u_int32_t)) * 2; 348 sizeof (u_int32_t)) * 3;
349 solock(so); 349 solock(so);
350 } 350 }
351 error = soreserve(so, sndreserve, rcvreserve); 351 error = soreserve(so, sndreserve, rcvreserve);
352 if (error) { 352 if (error) {
353 sounlock(so); 353 sounlock(so);
354 goto bad; 354 goto bad;
355 } 355 }
356 so->so_rcv.sb_flags |= SB_NOINTR; 356 so->so_rcv.sb_flags |= SB_NOINTR;
357 so->so_snd.sb_flags |= SB_NOINTR; 357 so->so_snd.sb_flags |= SB_NOINTR;
358 sounlock(so); 358 sounlock(so);
359 359
360 /* Initialize other non-zero congestion variables */ 360 /* Initialize other non-zero congestion variables */
361 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] = 361 nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = nmp->nm_srtt[3] =