On Wed, 1 Aug 2012 15:37:57 +0200
Pavel Shilovsky <piastryyy@xxxxxxxxx> wrote:
> 2012/7/31 Jeff Layton <jlayton@xxxxxxxxx>:
> > On Mon, 30 Jul 2012 21:17:53 -0400
> > Jeff Layton <jlayton@xxxxxxxxxx> wrote:
> >
> >> On Mon, 30 Jul 2012 23:11:10 +0200
> >> Pavel Shilovsky <piastryyy@xxxxxxxxx> wrote:
> >>
> >> > 2012/7/29 Jeff Layton <jlayton@xxxxxxxxxx>:
> >> > > On Fri, 27 Jul 2012 10:05:32 +0400
> >> > > Pavel Shilovsky <piastryyy@xxxxxxxxx> wrote:
> >> > >
> >> > >> 2012/7/27 Jeff Layton <jlayton@xxxxxxxxxx>:
> >> > >> > On Fri, 27 Jul 2012 03:57:44 +0400
> >> > >> > Pavel Shilovsky <piastryyy@xxxxxxxxx> wrote:
> >> > >> >
> >> > >> >> 2012/7/25 Jeff Layton <jlayton@xxxxxxxxxx>:
> >> > >> >> > We want to send SMBs as "atomically" as possible. Prior to sending any
> >> > >> >> > data on the socket, cork it to make sure that no non-full frames go
> >> > >> >> > out. Afterward, uncork it to make sure all of the data gets pushed out
> >> > >> >> > to the wire.
> >> > >> >> >
> >> > >> >> > Note that this more or less renders the socket=TCP_NODELAY mount option
> >> > >> >> > obsolete. When TCP_CORK and TCP_NODELAY are used on the same socket,
> >> > >> >> > TCP_NODELAY is essentially ignored.
> >> > >> >> >
> >> > >> >> > Acked-by: Pavel Shilovsky <pshilovsky@xxxxxxxxx>
> >> > >> >> > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
> >> > >> >> > ---
> >> > >> >> > fs/cifs/connect.c | 4 ++++
> >> > >> >> > fs/cifs/transport.c | 12 ++++++++++++
> >> > >> >> > 2 files changed, 16 insertions(+)
> >> > >> >> >
> >> > >> >> > diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
> >> > >> >> > index 6df6fa1..a828a8c 100644
> >> > >> >> > --- a/fs/cifs/connect.c
> >> > >> >> > +++ b/fs/cifs/connect.c
> >> > >> >> > @@ -1676,6 +1676,10 @@ cifs_parse_mount_options(const char *mountdata, const char *devname,
> >> > >> >> > if (string == NULL)
> >> > >> >> > goto out_nomem;
> >> > >> >> >
> >> > >> >> > + /*
> >> > >> >> > + * FIXME: since we now cork/uncork the socket while
> >> > >> >> > + * sending, should we deprecate this option?
> >> > >> >> > + */
> >> > >> >> > if (strnicmp(string, "TCP_NODELAY", 11) == 0)
> >> > >> >> > vol->sockopt_tcp_nodelay = 1;
> >> > >> >> > break;
> >> > >> >> > diff --git a/fs/cifs/transport.c b/fs/cifs/transport.c
> >> > >> >> > index d93f15d..a3e58b2 100644
> >> > >> >> > --- a/fs/cifs/transport.c
> >> > >> >> > +++ b/fs/cifs/transport.c
> >> > >> >> > @@ -27,6 +27,7 @@
> >> > >> >> > #include <linux/net.h>
> >> > >> >> > #include <linux/delay.h>
> >> > >> >> > #include <linux/freezer.h>
> >> > >> >> > +#include <linux/tcp.h>
> >> > >> >> > #include <asm/uaccess.h>
> >> > >> >> > #include <asm/processor.h>
> >> > >> >> > #include <linux/mempool.h>
> >> > >> >> > @@ -247,12 +248,23 @@ smb_send_rqst(struct TCP_Server_Info *server, struct smb_rqst *rqst)
> >> > >> >> > int n_vec = rqst->rq_nvec;
> >> > >> >> > unsigned int smb_buf_length = get_rfc1002_length(iov[0].iov_base);
> >> > >> >> > size_t total_len;
> >> > >> >> > + struct socket *ssocket = server->ssocket;
> >> > >> >> > + int val = 1;
> >> > >> >> >
> >> > >> >> > cFYI(1, "Sending smb: smb_len=%u", smb_buf_length);
> >> > >> >> > dump_smb(iov[0].iov_base, iov[0].iov_len);
> >> > >> >> >
> >> > >> >> > + /* cork the socket */
> >> > >> >> > + kernel_setsockopt(ssocket, SOL_TCP, TCP_CORK,
> >> > >> >> > + (char *)&val, sizeof(val));
> >> > >> >> > +
> >> > >> >> > rc = smb_send_kvec(server, iov, n_vec, &total_len);
> >> > >> >> >
> >> > >> >> > + /* uncork it */
> >> > >> >> > + val = 0;
> >> > >> >> > + kernel_setsockopt(ssocket, SOL_TCP, TCP_CORK,
> >> > >> >> > + (char *)&val, sizeof(val));
> >> > >> >> > +
> >> > >> >> > if ((total_len > 0) && (total_len != smb_buf_length + 4)) {
> >> > >> >> > cFYI(1, "partial send (wanted=%u sent=%zu): terminating "
> >> > >> >> > "session", smb_buf_length + 4, total_len);
> >> > >> >> > --
> >> > >> >> > 1.7.11.2
> >> > >> >> >
> >> > >> >> > --
> >> > >> >> > To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
> >> > >> >> > the body of a message to majordomo@xxxxxxxxxxxxxxx
> >> > >> >> > More majordomo info at http://vger.kernel.org/majordomo-info.html
> >> > >> >>
> >> > >> >> I tested it with SMB2 against Windows 7 server. When iosize is 64K
> >> > >> >> everything is ok but when we increase iosize to 1M (by using
> >> > >> >> multicredit requests) and the server loses the network connection and
> >> > >> >> only reboot helps.
> >> > >> >>
> >> > >> >> Also if I commented corking/uncorking the socket - everything is ok. I
> >> > >> >> think this change needs some more investigation (how does it deals
> >> > >> >> with 1M iosize on Samba, etc?)
> >> > >> >>
> >> > >> >
> >> > >> > Hmm, haven't seen that with a 1M iosize with smb1 against samba.
> >> > >> >
> >> > >> > I'll see if I can reproduce it.
> >> > >> >
> >> > >> > --
> >> > >> > Jeff Layton <jlayton@xxxxxxxxxx>
> >> > >>
> >> > >> Forgot to mentioned how I reproduce it - dbench with 5 clients.
> >> > >>
> >> > >
> >> > > Ok, I've built a cifs.ko from your smb2-dev-cifs-3.6 branch. Here are my mount options:
> >> > >
> >> > > //win7.poochiereds.net/export /mnt/win7 cifs rw,relatime,vers=2.1,sec=ntlmsspi,cache=strict,unc=\\win7.poochiereds.net\export,username=testuser,domain=WIN7,uid=0,noforceuid,gid=0,noforcegid,addr=192.168.1.32,file_mode=0755,dir_mode=0755,nounix,serverino,rsize=1048576,wsize=1048576,actimeo=1 0 0
> >> >
> >> > I don't set rsize and wsize explicitly but I don't think it's related.
> >> > On what connection did you test it? I use 100Mbit LAN.
> >> >
> >>
> >> The clients and servers are both KVM guests. I'll give it a go over a
> >> physical network tomorrow.
> >>
> >
> > Ok, ran it with client as a KVM guest and the server as win7 running on
> > bare metal over a gigE network. It still ran just fine.
> >
> > So what are the symptoms that you see here? Does dbench just hang? If
> > so, could you collect /proc/<pid>/stack from the hung process(es)?
> > Maybe that would tell us what's going on...
>
> Windows 7 server doesn't response to packets after some time running
> dbench. Also, I even can't ping google.com from this Windows machine.
> It seems that everything ok with dbench and Linux machine.
>
> So, it looks like Windows server problem on my configuration but of
> course seems very strage. I will try this patch with Samba server
> further.
>
> This patch doesn't break things with Windows untill we use multicredit
> requests (more than 64K, that are not targeted to 3.6 kernel). But I
> am going to target multicredit requests feature for 3.7. May be we
> should make cork/nodelay switchable? Or just merge the patchset
> without this patch for 3.6 and delay this patch for 3.7 - we will have
> much time to investigate this strange behavior?
>
> Thoughts?
>
If the server isn't responding then that seems like something is broken
on the server. Maybe you have a broken network driver? Do you have any
captures?
I don't think we should ship w/o the TCP_CORK code unless there is
clear evidence that it's a problem on the Linux end. We really don't
want the server sending half-baked SMBs, and that's much more likely to
occur if you don't cork the socket prior to sending.
--
Jeff Layton <jlayton@xxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
