I just tried to install today's Rawhide (20060904) for s390 with hercules. Everything went fine until the installation process started telnet/sshd in order to continue the process. Here is the hercules output : [...] Starting telnetd and sshd to allow login over the network. INFO: trying to register non-static key. the code is fine but needs lockdep annotation. turning off the locking correctness validator. 002db688 015d399c 8002f736 8002f774 015d3950 015d3998 00000002 00000000 015d39f8 015d39b4 015d39b4 80016190 Call Trace: ( <0000000000016098>! show_trace+0x80/0xfc) <00000000000161a8>! show_stack+0x94/0xb4 <00000000000161f0>! dump_stack+0x28/0x38 <0000000000050f22>! __lock_acquire+0x97a/0xad0 <000000000005178c>! lock_acquire+0x64/0x8c <000000000022af30>! _spin_lock_irqsave+0x50/0x70 <0000000010892588>! ctc_tx+0x220/0xa28 ctc! <00000000001b1d92>! dev_hard_start_xmit+0x18e/0x23c <00000000001c40c4>! __qdisc_run+0xe8/0x20c <00000000001b4372>! dev_queue_xmit+0x132/0x28c <00000000001daca0>! ip_output+0x23c/0x2cc <00000000001da618>! ip_build_and_send_pkt+0x230/0x27c <00000000001ef58e>! tcp_v4_send_synack+0xc6/0x160 <00000000001f1fe4>! tcp_v4_conn_request+0x45c/0x554 <00000000001e8732>! tcp_rcv_state_process+0x6e/0xe94 <00000000001f0662>! tcp_v4_do_rcv+0x2ee/0x3c8 <00000000001f3448>! tcp_v4_rcv+0xaec/0xc30 <00000000001d4a54>! ip_local_deliver+0x1f4/0x33c <00000000001d478e>! ip_rcv+0x556/0x628 <00000000001b17c4>! netif_receive_skb+0x2bc/0x37c <00000000001b3574>! process_backlog+0xb4/0x1a0 <00000000001b3c2a>! net_rx_action+0xb2/0x1d4 <0000000000035b80>! __do_softirq+0x78/0x134 <000000000001e860>! do_softirq+0x7c/0xbc <0000000000035f76>! irq_exit+0x6a/0x84 <00000000001666fe>! do_IRQ+0x13e/0x170 <000000000001f094>! io_no_vtime+0x14/0x1a <000000000043e452>! stack_trace+0x33dba/0x80000 Connect now to 192.168.200.3 to start the installation. And telnet doesn't work : $ telnet 192.168.200.3 Trying 192.168.200.3... Connected to 192.168.200.3 (192.168.200.3). Escape character is '^]'. getaddrinfo: localhost: No such file or directory No address associated with hostname: Illegal seek Connection closed by foreign host. And ssh asks to save the key, but then "Connection closed by 192.168.200.3". Could this be an s390 specific network problem? Or something done wrong in the s390 installation process? To reproduce, simple follow the instructions found in the README.fedora file of the hercules package found in Extras.
It's something bizarre with telnet... I think it might be worth nuking telnet given how common ssh is at this point Any disagreement?
It used to still work a couple of months ago at most... And didn't someone (yourself?) mention on a fedora list in the past few days that the reason why openssh-server generated its keys upon first startup instead of when the package gets installed was because it made a huge difference for the install time on some s390 machines? I'm just asking because I'm not sure I see the real advantage of ssh over telnet in this case... unless the keys at that point are going to be kept for the installed system? That would be a nice feature.
re comment #2: it made a huge difference on our 31bit MP3000, but key generation is almost instantaneous on our zSeries machine. It's been a while since my last Hercules installation, how long does key generation take at the first boot after the installation ? btw: was this really an s390 installation or s390x ? re comment #1: telnetd aborts with a segfault (on s390x): [pid 495] mprotect(0x20000214000, 4096, PROT_READ) = 0 [pid 495] open("/etc/passwd", O_RDONLY) = 4 [pid 495] fcntl(4, F_GETFD) = 0 [pid 495] fcntl(4, F_SETFD, FD_CLOEXEC) = 0 [pid 495] fstat(4, {st_mode=S_IFREG|0644, st_size=128, ...}) = 0 [pid 495] mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x200001fd000 [pid 495] read(4, "root::0:0::/:/sbin/loader\ndebug::0:0:debug shell:/:/bin/bash\nsshd:x:74:74:Privilege-separated SSH:/var/empty/sshd:/sbin/nologin\n", 4096) = 128 [pid 495] --- SIGSEGV (Segmentation fault) @ 0 (0) --- Process 495 detached <... select resumed> ) = 1 (in [3]) --- SIGCHLD (Child exited) @ 0 (0) --- read(3, 0x80200fe8, 4000) = -1 EIO (Input/output error) I'm not sure if we should kick telnet, but have no strong objections either.
I'm not against nuking telnet... just trying to make sure it won't become an issue for RHEL5 users. I used hercules with the default configuration I've included in the Fedora Extras package : CPUMODEL 3090 and ARCHMODE ESA/390 then used the files from s390/os/images/ (and not s390x of that was the question). For anyone wanting to reproduce this without any actual s390 hardware, I've made sure that the hercules package comes in a "ready to run" shape. Simply install it and follow the instructions from /usr/share/doc/hercules-*/README.fedora (for FC devel, RHEL4 and RHL 7.2). You get your own mainframe in no time! I'm running it on a 64bit Xeon 3GHz and it's not fast, but not too slow either.
Thanks for the hercules info, Matthias. I'm still waiting on the zSeries Luggable from IBM. :) On nuking telnet... I'm all in favor of it. With the addition of IPv6 in to anaconda, binding telnet and/or expanding it to make it work with IPv6 is just unnecessarily annoying. I don't think the key generation time required by sshd will be that bad and users can deal with it. My two cents.
I've removed telnet login support for s390 installation from rawhide. Going forward, ssh will be the only supported method for connecting to the s390 installer instance. For the RHEL-5 bug, we fixed the telnet connection problem in 5.1. I am closing this as WONTFIX since I'm not doing any more telnet support for s390 installs in Fedora.