From tux at mikebell.org Wed Dec 1 20:29:51 2004 From: tux at mikebell.org (tux at mikebell.org) Date: Wed, 1 Dec 2004 12:29:51 -0800 Subject: How does tux redirect the request to the secondary server? In-Reply-To: <20041130074617.59801.qmail@web15506.mail.cnb.yahoo.com> References: <20041130074617.59801.qmail@web15506.mail.cnb.yahoo.com> Message-ID: <20041201202951.GE1847@mikebell.org> On Tue, Nov 30, 2004 at 03:46:17PM +0800, shiner chen wrote: > does the secondary server send the response to client directly ? if tux does like that ,I want to know how does it do that! thanks ! It does. TUX is able to pass the connection on to the secondary server because it is in the kernel. There was some time a while ago about a syscall to allow userspace HTTP daemons to do the same thing, but I don't recall anything coming of it. From shiner_chen at yahoo.com.cn Thu Dec 2 01:33:38 2004 From: shiner_chen at yahoo.com.cn (shiner chen) Date: Thu, 2 Dec 2004 09:33:38 +0800 (CST) Subject: How does tux redirect the request to the secondary server? In-Reply-To: <20041201202951.GE1847@mikebell.org> Message-ID: <20041202013338.51925.qmail@web15504.mail.cnb.yahoo.com> Sorry, I dont get your answer's true point. I want to get your more detail explain. 1.Does the secondary server send the response to client directly? Or secondary server send the response to tux because tux is a proxy for dynamic request? 2.if secondary server send response to client directly ,how does the tux implement that? I think tux must modify the tcp/ip protocol stack code. is it right? tux at mikebell.org wrote: On Tue, Nov 30, 2004 at 03:46:17PM +0800, shiner chen wrote: > does the secondary server send the response to client directly ? if tux does like that ,I want to know how does it do that! thanks ! It does. TUX is able to pass the connection on to the secondary server because it is in the kernel. There was some time a while ago about a syscall to allow userspace HTTP daemons to do the same thing, but I don't recall anything coming of it. --------------------------------- Do You Yahoo!? ??????????????? -------------- next part -------------- An HTML attachment was scrubbed... URL: From sitz at onastick.net Thu Dec 2 02:08:26 2004 From: sitz at onastick.net (Noah) Date: Wed, 1 Dec 2004 21:08:26 -0500 Subject: How does tux redirect the request to the secondary server? In-Reply-To: <20041202013338.51925.qmail@web15504.mail.cnb.yahoo.com> References: <20041201202951.GE1847@mikebell.org> <20041202013338.51925.qmail@web15504.mail.cnb.yahoo.com> Message-ID: <20041202020826.GA21467@radu.onastick.net> On Thu, Dec 02, 2004 at 09:33:38AM +0800, shiner chen wrote: > Sorry, I dont get your answer's true point. I want to get your more detail explain. > > 1.Does the secondary server send the response to client directly? Or secondary server send the response to tux because tux is a proxy for dynamic request? The secondary server sends the response to the client directly. > 2.if secondary server send response to client directly ,how does the tux implement that? I think tux must modify the tcp/ip protocol stack code. is it right? Tux accepts the initial port 80 request, and makes a decision. If it decides to let the secondary server handle the request, it picks up the file descriptor which describes the network connection to the client and hands it off to the secondary server. The server sends the response to the client IP:port info in that file descriptor. --n -- dd of=/dev/fd0 if=/dev/flippy bs=1024 ^^^ Making Flippy Floppy From shiner_chen at yahoo.com.cn Thu Dec 2 07:15:23 2004 From: shiner_chen at yahoo.com.cn (shiner chen) Date: Thu, 2 Dec 2004 15:15:23 +0800 (CST) Subject: How does tux redirect the request to the secondary server? In-Reply-To: <20041202020826.GA21467@radu.onastick.net> Message-ID: <20041202071523.56816.qmail@web15506.mail.cnb.yahoo.com> Tux should get client's http request before it can decide whether secondary server should handle the req. So client must have a complete socket connection with tux. It is true that tux can insert a new socket into the accept queue of secondary server's listen socket. But the client just know the connection with tux. Secondary server is transparent for client. As a result, how client can receive the data from the secondary server? Yeah, we can use network hand-off or tcp splicing tech. to do that. But we must modify the tpc/ip protocol stack code. So is tux? Thanks much!! Noah wrote: On Thu, Dec 02, 2004 at 09:33:38AM +0800, shiner chen wrote: > Sorry, I dont get your answer's true point. I want to get your more detail explain. > > 1.Does the secondary server send the response to client directly? Or secondary server send the response to tux because tux is a proxy for dynamic request? The secondary server sends the response to the client directly. > 2.if secondary server send response to client directly ,how does the tux implement that? I think tux must modify the tcp/ip protocol stack code. is it right? Tux accepts the initial port 80 request, and makes a decision. If it decides to let the secondary server handle the request, it picks up the file descriptor which describes the network connection to the client and hands it off to the secondary server. The server sends the response to the client IP:port info in that file descriptor. --n -- dd of=/dev/fd0 if=/dev/flippy bs=1024 ^^^ Making Flippy Floppy _______________________________________________ tux-list mailing list tux-list at redhat.com https://www.redhat.com/mailman/listinfo/tux-list --------------------------------- Do You Yahoo!? ??????????????? -------------- next part -------------- An HTML attachment was scrubbed... URL: From max at expires-2004.workforce.de Thu Dec 2 08:57:40 2004 From: max at expires-2004.workforce.de (Max Helmet) Date: Thu, 2 Dec 2004 09:57:40 +0100 Subject: AW: How does tux redirect the request to the secondary server? In-Reply-To: <20041202071523.56816.qmail@web15506.mail.cnb.yahoo.com> Message-ID: > Yeah, we can use network hand-off or tcp splicing tech. > to do that. But we must modify the tpc/ip protocol stack > code. So is tux? Just have a look at the redirect_sock(*req, port) function in linux/net/tux/redirect.c: 1. It "looks up listening user-space socket" (->Apache) 2. "Requeue the 'old' socket as an accept-socket of the listening socket. This way we can shuffle a socket around. Since we've read the input data via the non-destructive MSG_PEEK, the secondary server can be used transparently." 3. Closes the request for tux as "it's now completely up to the secondary server to handle this request." The magic part here is the non-destructive reading of the request in read_request(*sock, *buf, max_size) via sock->sk->sk_prot->recvmsg(&iocb, sock->sk, &msg, max_size, MSG_DONTWAIT, MSG_PEEK, NULL); > Thanks much!! You welcome!!! Max From mingo at elte.hu Mon Dec 6 12:47:26 2004 From: mingo at elte.hu (Ingo Molnar) Date: Mon, 6 Dec 2004 13:47:26 +0100 Subject: [patch] tux3-2.6.10-rc3-A7 In-Reply-To: <20041010191517.GA6899@elte.hu> References: <20040419070756.GA15513@elte.hu> <20040909170618.GA27795@elte.hu> <20040910162028.GQ24408@krispykreme> <20040910193630.GB19488@elte.hu> <20040917101513.GA18752@elte.hu> <20041010191517.GA6899@elte.hu> Message-ID: <20041206124726.GA19273@elte.hu> i've uploaded the 2.6.10-rc3-A7 patch: http://redhat.com/~mingo/TUX-patches/tux3-2.6.10-rc3-A7 this is a fixes-only release. Changes since -A6: - do not try to redirect to IPv6 secondary sockets. (will reject the redirect instead.) This solves the IPv6/Apache crash. - fixed tuxstat related crash - make Tux work on selinux. - merged to 2.6.10-rc3 Ingo From jlirochon at cyanide-studio.com Mon Dec 6 13:37:20 2004 From: jlirochon at cyanide-studio.com (Julien Lirochon) Date: Mon, 06 Dec 2004 14:37:20 +0100 Subject: tux & symlinks Message-ID: <41B46090.40603@cyanide-studio.com> Hi there, It seems tux (A6) can't follow symlinks on my box. Is it a known problem ? Julien From hosoi-redhat at ryo.com Sun Dec 12 00:58:41 2004 From: hosoi-redhat at ryo.com (Ryosuke Hosoi) Date: Sat, 11 Dec 2004 16:58:41 -0800 (PST) Subject: tux & symlinks In-Reply-To: <41B46090.40603@cyanide-studio.com> References: <41B46090.40603@cyanide-studio.com> Message-ID: <20041211.165841.640925635.hosoi@ryo.com> Hi From: Julien Lirochon Subject: tux & symlinks Date: Mon, 06 Dec 2004 14:37:20 +0100 Message-ID: <41B46090.40603 at cyanide-studio.com> > It seems tux (A6) can't follow symlinks on my box. Is it a known problem ? I think tux can follow symlinks under the docroot only. -- Ryosuke Hosoi mailto:hosoi at ryo.com http://www.ryo.com/ From dpilon at dpilon.com Wed Dec 29 03:26:18 2004 From: dpilon at dpilon.com (Denis Pilon) Date: Tue, 28 Dec 2004 22:26:18 -0500 Subject: [patch] tux3-2.6.10-rc3-A7 unknown symbols In-Reply-To: <20041206124726.GA19273@elte.hu> References: <20040419070756.GA15513@elte.hu> <20040909170618.GA27795@elte.hu> <20040910162028.GQ24408@krispykreme> <20040910193630.GB19488@elte.hu> <20040917101513.GA18752@elte.hu> <20041010191517.GA6899@elte.hu> <20041206124726.GA19273@elte.hu> Message-ID: <41D223DA.4080903@dpilon.com> Hello All, I am getting unknown symbols when using this patch. Any recommendations ? WARNING: /lib/modules/2.6.10-40.caos/kernel/net/tux/tux.ko needs unknown symbol tcp_write_xmit WARNING: /lib/modules/2.6.10-40.caos/kernel/net/tux/tux.ko needs unknown symbol sock_alloc WARNING: /lib/modules/2.6.10-40.caos/kernel/net/tux/tux.ko needs unknown symbol set_fs_root WARNING: /lib/modules/2.6.10-40.caos/kernel/net/tux/tux.ko needs unknown symbol tcp_cwnd_application_limited WARNING: /lib/modules/2.6.10-40.caos/kernel/net/tux/tux.ko needs unknown symbol tcp_set_skb_tso_segs WARNING: /lib/modules/2.6.10-40.caos/kernel/net/tux/tux.ko needs unknown symbol tcp_v4_lookup_listener This is with released 2.6.10 and 2.6.10-rc3. DP Ingo Molnar wrote: >i've uploaded the 2.6.10-rc3-A7 patch: > > http://redhat.com/~mingo/TUX-patches/tux3-2.6.10-rc3-A7 > >this is a fixes-only release. Changes since -A6: > > - do not try to redirect to IPv6 secondary sockets. (will reject the > redirect instead.) This solves the IPv6/Apache crash. > > - fixed tuxstat related crash > > - make Tux work on selinux. > > - merged to 2.6.10-rc3 > > Ingo > >_______________________________________________ >tux-list mailing list >tux-list at redhat.com >https://www.redhat.com/mailman/listinfo/tux-list > >