My mount attempts aren't working.
I turned on nfsd debugging and got...
Sep 3 17:34:00 rowlf rpc.mountd: authenticated mount request from gonzo.int.wil
.cx:787 for /pub (/pub)
Sep 3 17:34:00 rowlf kernel: found domain gonzo.int.wil.cx
Sep 3 17:34:00 rowlf kernel: found fsidtype 0
Sep 3 17:34:00 rowlf kernel: found fsid length 8
Sep 3 17:34:00 rowlf kernel: Path seems to be </pub>
Sep 3 17:34:00 rowlf kernel: Found the path /pub
Sep 3 17:34:00 rowlf kernel: And found export
Sep 3 17:34:00 rowlf kernel: nfsd: exp_rootfh(/pub [e0000040f69e1a80] gonzo.int
.wil.cx:sdb3/195064)
Sep 3 17:34:00 rowlf kernel: nfsd: fh_compose(exp 08:13/195064 //pub, ino=19506
4)
Sep 3 17:34:01 rowlf kernel: nfsd_dispatch: vers 2 proc 1
Sep 3 17:34:01 rowlf kernel: nfsd: GETATTR 32: 00000001 13000800 0002f9f8 0000
0000 00000000 00000000
Sep 3 17:34:01 rowlf kernel: nfsd: fh_verify(32: 00000001 13000800 0002f9f8 000
00000 00000000 00000000)
Then I added full rpc debugging, and got this:
Sep 3 17:56:26 rowlf kernel: svc: socket e0000040fe7b6f80(inet e0000040fe0f6480), count=112, busy=0
Sep 3 17:56:26 rowlf kernel: svc: socket e0000040fe0f6480 served by daemon e0000040425e6900
Sep 3 17:56:26 rowlf kernel: svc: server e0000040425e6900, socket e0000040fe7b6f80, inuse=1
Sep 3 17:56:26 rowlf kernel: svc: socket e0000040fe0f6480 served by daemon e0000040425e7500
Sep 3 17:56:26 rowlf kernel: svc: got len=104
Sep 3 17:56:26 rowlf kernel: svc: svc_authenticate (1)
Sep 3 17:56:26 rowlf kernel: svc: socket e0000040fe0f6480 busy, not enqueued
Sep 3 17:56:26 rowlf kernel: svc: calling dispatcher
Sep 3 17:56:26 rowlf kernel: nfsd_dispatch: vers 2 proc 1
Sep 3 17:56:26 rowlf kernel: nfsd: GETATTR 32: 00000001 13000800 0002f9f8 00000000 00000000 00000000
Sep 3 17:56:26 rowlf kernel: nfsd: fh_verify(32: 00000001 13000800 0002f9f8 00000000 00000000 00000000)
Sep 3 17:56:26 rowlf kernel: svc: service e0000040425e6900, releasing skb e0000040fe5ddf80
Sep 3 17:56:26 rowlf kernel: svc: socket e0000040fe7b6f80 sendto([e00000404258c000 96... ], 96) = 0 (addr 2501a8c0)
Sep 3 17:56:26 rowlf kernel: svc: socket e0000040fe0f6480 busy, not enqueued
Sep 3 17:56:26 rowlf kernel: svc: server e0000040425e6900 waiting for data (to = 3686400)
Sep 3 17:56:26 rowlf kernel: svc: server e0000040425e7500, socket e0000040fe7b6f80, inuse=1
Sep 3 17:56:26 rowlf kernel: svc: got len=-11
tcpdump confirms the same thing -- packets go in, but they don't come out.
--
"It's not Hollywood. War is real, war is primarily not about defeat or
victory, it is about death. I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
On Wednesday September 3, [email protected] wrote:
>
> My mount attempts aren't working.
>
...
>
> Then I added full rpc debugging, and got this:
>
...
> Sep 3 17:56:26 rowlf kernel: svc: socket e0000040fe7b6f80 sendto([e00000404258c000 96... ], 96) = 0 (addr 2501a8c0)
...
>
> tcpdump confirms the same thing -- packets go in, but they don't come out.
>
Very helpful details.
It looks like a simple case of 32-bit parochialism :-)
Does this patch help?
Thanks,
NeilBrown
diff ./net/sunrpc/svcsock.c~current~ ./net/sunrpc/svcsock.c
--- ./net/sunrpc/svcsock.c~current~ 2003-09-05 13:47:33.000000000 +1000
+++ ./net/sunrpc/svcsock.c 2003-09-05 13:48:42.000000000 +1000
@@ -376,8 +376,8 @@ svc_sendto(struct svc_rqst *rqstp, struc
msg.msg_flags = MSG_MORE;
msg.msg_control = &cm;
- msg.msg_controllen = sizeof(cm);
- cm.cmh.cmsg_len = sizeof(cm);
+ msg.msg_controllen = CMSG_LEN(sizeof(cm));
+ cm.cmh.cmsg_len = CMSG_LEN(sizeof(cm));
cm.cmh.cmsg_level = SOL_IP;
cm.cmh.cmsg_type = IP_PKTINFO;
cm.pki.ipi_ifindex = 0;
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
On Fri, Sep 05, 2003 at 01:53:27PM +1000, Neil Brown wrote:
> Very helpful details.
> It looks like a simple case of 32-bit parochialism :-)
>
> Does this patch help?
Nope. Applied patch, rebooted with it, here's the full details again
(don't think anything changed, but just in case):
Sep 5 07:52:06 rowlf kernel: svc: socket e00000404404a680(inet e000004043d22480
), count=112, busy=0
Sep 5 07:52:06 rowlf kernel: svc: socket e000004043d22480 served by daemon e000
0040ea866900
Sep 5 07:52:06 rowlf kernel: svc: server e0000040ea866900, socket e00000404404a
680, inuse=1
Sep 5 07:52:06 rowlf kernel: svc: socket e000004043d22480 served by daemon e000
0040ea867500
Sep 5 07:52:06 rowlf kernel: svc: got len=104
Sep 5 07:52:06 rowlf kernel: svc: svc_authenticate (1)
Sep 5 07:52:06 rowlf kernel: svc: socket e000004043d22480 busy, not enqueued
Sep 5 07:52:06 rowlf kernel: svc: calling dispatcher
Sep 5 07:52:06 rowlf kernel: nfsd_dispatch: vers 2 proc 1
Sep 5 07:52:06 rowlf kernel: nfsd: GETATTR 32: 00000001 13000800 0002f9f8 0000
0000 00000000 00000000
Sep 5 07:52:06 rowlf kernel: nfsd: fh_verify(32: 00000001 13000800 0002f9f8 000
00000 00000000 00000000)
Sep 5 07:52:06 rowlf kernel: nfsd: storing reply in cache
Sep 5 07:52:06 rowlf kernel: svc: service e0000040ea866900, releasing skb e0000
040ff163d80
Sep 5 07:52:06 rowlf kernel: svc: socket e00000404404a680 sendto([e0000040ea30c
000 96... ], 96) = 0 (addr 2501a8c0)
Sep 5 07:52:06 rowlf kernel: svc: socket e000004043d22480 busy, not enqueued
Sep 5 07:52:06 rowlf kernel: svc: server e0000040ea866900 waiting for data (to
= 3686400)
Sep 5 07:52:06 rowlf kernel: svc: server e0000040ea867500, socket e00000404404a
680, inuse=1
Sep 5 07:52:06 rowlf kernel: svc: got len=-11
Sep 5 07:52:06 rowlf kernel: svc: server e0000040ea867500 waiting for data (to
= 3686400)
--
"It's not Hollywood. War is real, war is primarily not about defeat or
victory, it is about death. I've seen thousands and thousands of dead bodies.
Do you think I want to have an academic debate on this subject?" -- Robert Fisk
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
On Friday September 5, [email protected] wrote:
> On Fri, Sep 05, 2003 at 01:53:27PM +1000, Neil Brown wrote:
> > Very helpful details.
> > It looks like a simple case of 32-bit parochialism :-)
> >
> > Does this patch help?
>
> Nope. Applied patch, rebooted with it, here's the full details again
> (don't think anything changed, but just in case):
Ok, I looked more closely at the .h files etc and I think the
following is correct (instead of the previous patch).
(i.e. sizeof(cm.pki) instead of sizeof(cm)).
NeilBrown
diff ./net/sunrpc/svcsock.c~current~ ./net/sunrpc/svcsock.c
--- ./net/sunrpc/svcsock.c~current~ 2003-09-08 14:00:16.000000000 +1000
+++ ./net/sunrpc/svcsock.c 2003-09-08 14:00:11.000000000 +1000
@@ -376,8 +376,8 @@ svc_sendto(struct svc_rqst *rqstp, struc
msg.msg_flags = MSG_MORE;
msg.msg_control = &cm;
- msg.msg_controllen = sizeof(cm);
- cm.cmh.cmsg_len = sizeof(cm);
+ msg.msg_controllen = CMSG_LEN(sizeof(cm.pki));
+ cm.cmh.cmsg_len = CMSG_LEN(sizeof(cm.pki));
cm.cmh.cmsg_level = SOL_IP;
cm.cmh.cmsg_type = IP_PKTINFO;
cm.pki.ipi_ifindex = 0;
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
>>>>> " " == Neil Brown <[email protected]> writes:
> On Friday September 5, [email protected] wrote:
>> On Fri, Sep 05, 2003 at 01:53:27PM +1000, Neil Brown wrote:
>> > Very helpful details. It looks like a simple case of 32-bit
>> > parochialism :-)
>> >
>> > Does this patch help?
>>
>> Nope. Applied patch, rebooted with it, here's the full details
>> again (don't think anything changed, but just in case):
> Ok, I looked more closely at the .h files etc and I think the
> following is correct (instead of the previous patch).
> (i.e. sizeof(cm.pki) instead of sizeof(cm)).
Nope see the code in ip_sockglue.c + the macros in socket.h....
AFAICS the control messages have wierd alignment requirements.
Shouldn't it rather be something like the following?
Cheers,
Trond
--- linux-2.6.0-up/net/sunrpc/svcsock.c.orig 2003-08-30 23:35:33.000000000 -0400
+++ linux-2.6.0-up/net/sunrpc/svcsock.c 2003-09-08 00:39:07.000000000 -0400
@@ -352,9 +352,9 @@
struct svc_sock *svsk = rqstp->rq_sock;
struct socket *sock = svsk->sk_sock;
int slen;
- struct { struct cmsghdr cmh;
- struct in_pktinfo pki;
- } cm;
+ char *buffer[CMSG_SPACE(sizeof(struct in_pktinfo))];
+ struct cmsghdr *cmh = (struct cmsghdr *)buffer;
+ struct in_pktinfo *pki = (struct in_pktinfo *)CMSG_DATA(cmh);
int len = 0;
int result;
int size;
@@ -374,13 +374,13 @@
msg.msg_iovlen = 0;
msg.msg_flags = MSG_MORE;
- msg.msg_control = &cm;
- msg.msg_controllen = sizeof(cm);
- cm.cmh.cmsg_len = sizeof(cm);
- cm.cmh.cmsg_level = SOL_IP;
- cm.cmh.cmsg_type = IP_PKTINFO;
- cm.pki.ipi_ifindex = 0;
- cm.pki.ipi_spec_dst.s_addr = rqstp->rq_daddr;
+ msg.msg_control = cmh;
+ msg.msg_controllen = sizeof(buffer);
+ cmh->cmsg_len = CMSG_LEN(sizeof(*pki));
+ cmh->cmsg_level = SOL_IP;
+ cmh->cmsg_type = IP_PKTINFO;
+ pki->ipi_ifindex = 0;
+ pki->ipi_spec_dst.s_addr = rqstp->rq_daddr;
if (sock_sendmsg(sock, &msg, 0) < 0)
goto out;
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
On September 8, [email protected] wrote:
> >>>>> " " == Neil Brown <[email protected]> writes:
>
> > On Friday September 5, [email protected] wrote:
> >> On Fri, Sep 05, 2003 at 01:53:27PM +1000, Neil Brown wrote:
> >> > Very helpful details. It looks like a simple case of 32-bit
> >> > parochialism :-)
> >> >
> >> > Does this patch help?
> >>
> >> Nope. Applied patch, rebooted with it, here's the full details
> >> again (don't think anything changed, but just in case):
>
> > Ok, I looked more closely at the .h files etc and I think the
> > following is correct (instead of the previous patch).
>
> > (i.e. sizeof(cm.pki) instead of sizeof(cm)).
>
> Nope see the code in ip_sockglue.c + the macros in socket.h....
> AFAICS the control messages have wierd alignment requirements.
>
> Shouldn't it rather be something like the following?
Probably it should, but that felt ugly, and the alignment requirements
are "round up to multiple of size of long", and I believe that just
putting them together in a struct would do that. But I guess it isn't
100% guaranteed so it is safer to use your code.
Once I get confirmation from Matthew that something actually works for
him, I'll send an appropriate change to Linus.
thanks.
NeilBrown
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs