PPP for Linux Version 1.0.1 ============= based on ppp-2.1.2 June 1994 Michael Callahan callahan@maths.ox.ac.uk Al Longyear longyear@netcom.com Contents: INTRODUCTION CREDITS FUTURE PLANS INSTALLATION GENERAL NETWORK CONFIGURATION CONNECTING TO A PPP SERVER IF IT WORKS IF IT DOESN'T WORK IF IT STILL DOESN'T WORK (OR, BUG REPORTS) DYNAMIC ADDRESS ASSIGNMENT SETTING UP A MACHINE FOR INCOMING PPP CONNECTIONS ADDING MORE PPP CHANNELS CHANGES FROM LINUX PPP 0.1.x CONCLUSION INTRODUCTION This is a PPP driver for Linux. It has been used by many people and seems to be quite stable. It is capable of being used either as a 'client'--for connecting a Linux machine to a local Internet provider, for example--or as a 'server'--allowing a Linux machine with a modem and an Ethernet connection to the Internet to provide dial-in PPP links. (In fact, the PPP protocol does not make the distinction between client and server, but this is the way people often think about it.) The PPP protocol consists of two parts. One is a scheme for framing and encoding packets, the other is a series of protocols called LCP, IPCP, UPAP and CHAP, for negotiating link options and for authentication. This package similarly consists of two parts: a kernel module which handles PPP's low-level framing protocol, and a user-level program called pppd which implements PPP's negotiation protocols. The kernel module assembles/disassembles PPP frames, handles error detection, and forwards packets between the serial port and either the kernel network code or the user-level program pppd. IP packets go directly to the kernel network code. So once pppd has negotiated the link, it in practice lies completely dormant until you want to take the link down, when it negotiates a graceful disconnect. CREDITS I (MJC) wrote the original kernel driver from scratch. Laurence Culhane and Fred van Kempen's slip.c was priceless as a model (a perusal of the files will reveal that I often mimicked what slip.c did). Otherwise I just implemented what pppd needs, using RFC1331 as a guide. For the most part, the Linux driver provides the same interface as the free 386BSD and SunOS drivers. The exception is that Linux has no support for asynchronous I/O, so I hacked an ioctl into the PPP kernel module that provides a signal when packets appear and made pppd use this instead. Al Longyear ported version 2.0.4 of pppd (from the free package ppp-2.0.4) to Linux. He also provided several enhancements to both the kernel driver and the OS-independent part of pppd. His contributions to Linux PPP have been immense, and so this release is being distributed over both our names. The pppd program comes from the free distribution of PPP for Suns and 386BSD machines, maintained by Paul Mackerras. This package lists "thanks to" Brad Parker, Greg Christy, Drew D. Perkins, Rick Adams and Chris Torek. FUTURE PLANS The main missing feature is the ability to fire up a PPP connection automatically when a packet destined for the remote host is generated ("demand-dialing"). Work is progressing on this, but it involves some nontrivial design issues. INSTALLATION This version of PPP has been tested on 1.0.x (x=0..9) and 1.1.x (x=0..14) kernels. It will probably not work on kernels much earlier than this due to a change in the routing code. If you have an earlier kernel, please upgrade. joining the PPP channel of linux-activists: This isn't really part of installation, but if you DO use Linux PPP you should do this. Send a message with the line X-Mn-Admin: join PPP contained in the body to linux-activists-request@niksula.hut.fi You can send to the list by mailing to linux-activists@niksula.hut.fi and putting the line X-Mn-Key: PPP at the start of your message. The advantage of subscribing is that you'll be informed of updates and patches, and you'll be able to draw on the experience of many PPP users. If you have a problem, I may not be able to diagnose it, but someone else may have solved it already. Note also that I do not read the linux Usenet newsgroups regularly enough to catch any discussions of PPP; if you want to reach the PPP audience you should join the linux-activists channel. To leave the PPP mailing list :-(, send a message with the line X-Mn-Admin: leave PPP to linux-activists-request. YOU MUST USE THE EXACT SAME ACCOUNT WHICH YOU USED WHEN YOU JOINED. THIS MEANS THAT YOU NEED THE SAME ACCOUNT NAME, HOST NAME, AND SYSTEM NAME. kernel driver installation: This depends on the kernel version you're using. Since 1.1.14, Linux kernels have had built-in support for PPP. You'll be asked whether you want PPP when you run "make config". It's as easy as that. NEVER, NEVER, REPLACE THE FILES IN THE KERNEL WITH ANY FILE IN THIS PACKAGE. THE KERNEL IS ALWAYS MORE CURRENT THAN FILES IN THIS PACKAGE. In 1.1.13, PPP is there but the PPP line in config.in is commented out. If you have 1.1.13, you probably should just upgrade anyway. Kernel versions prior to 1.1.13 (including all 1.0.x kernels) have had (hidden) support for PPP in the kernel configuration setup for quite some time. Adding the PPP kernel driver is easy: 1) copy ppp.c from the linux subdirectory of the distribution to drivers/net and ppp.h to include/linux 2) uncomment the CONFIG_PPP line in config.in 3) if you are using 1.1.3 or earlier (including 1.0.x): uncomment the line in ppp.c that begins /* #define NET02D by removing the "/* " characters 4) in the top level of the kernel source make config make dep make Reboot with the new kernel. At startup, you should see something like this: PPP: version 0.2.7 (4 channels) NEW_TTY_DRIVERS OPTIMIZE_FLAGS TCP compression code copyright 1989 Regents of the University of California PPP line discipline registered. Don't worry if the words in capitals on the first line are slightly different. If you want more than 4 channels, see the section "ADDING MORE PPP CHANNELS" below. Now, try looking at the contents of /proc/net/dev. It should look something like this: Inter-| Receive | Transmit face |packets errs drop fifo frame|packets errs drop fifo colls carrier lo: 0 0 0 0 0 0 0 0 0 0 0 ppp0: 0 0 0 0 0 0 0 0 0 0 0 ppp1: 0 0 0 0 0 0 0 0 0 0 0 ppp2: 0 0 0 0 0 0 0 0 0 0 0 ppp3: 0 0 0 0 0 0 0 0 0 0 0 This indicates that the driver is successfully installed. (Of course, you should keep a kernel without PPP around, in case something goes wrong.) pppd installation: If you want to recompile pppd, go to the pppd subdirectory of the installation. Type ln -s Makefile.linux Makefile make There will be some warnings when compiling lcp.c, upap.c and chap.c but these are OK. If you want to recompile chat, consult README.linux in the chat directory. To install, type 'make install' in the chat and pppd directories. This will put chat and pppd binaries in /usr/etc and the pppd.8 manual page in /usr/man/man8. pppd needs to be run as root. You can either make it setuid root or just use it when you are root. 'make install' will try to install it setuid root. Making pppd setuid root is convenient for a single-user machine, but has security implications which you should investigate carefully before making it available on a multiuser machine. GENERAL NETWORK CONFIGURATION Since many people don't use the Linux networking code at all until they get a PPP link, this section describes generally what's needed to get things running. In principle none of this is special to PPP. For more details, you should consult the relevant Linux HOWTOs. If you already understand network setup, you can skip this section. The first file that requires attention is the rc script that does network configuration at boot time, called /etc/rc.net or /etc/rc.d/rc.net.{1,2} or something similar, depending on your Linux distribution. This file should 'ifconfig' the loopback interface lo, and should add an interface route for it. These lines might look something like this: $CONFIG lo 127.0.0.1 $ROUTE add loopback or /sbin/ifconfig lo 127.0.0.1 /sbin/route add 127.0.0.1 However, it should *not* config an ethernet card or install any other routes (unless you actually have an ethernet card, in which case I'll assume you know what to do). Many distributions will provide scripts that expect you to have an ethernet card. You also need to decide whether you want to allow incoming telnet/ftp/finger, etc. If so, you should have the rc startup script run the 'inetd' daemon. Next, you should set up /etc/hosts to have two lines. The first should just give the loopback or localhost address and the second should give your own host name and the IP address your PPP connection will use. For example: 127.0.0.1 loopback localhost # useful aliases 192.1.1.17 billpc.whitehouse.gov bill # my hostname where my IP address is 192.1.1.17 and my hostname is billpc.whitehouse.gov. (Not really, you understand.) If your PPP server does dynamic IP address assignment, give a guess as to an address you might get (see also "Dynamic Address Assignment" below). Finally, you need to configure the domain name system by putting appropriate lines in /etc/resolv.conf . It should look something like this: domain whitehouse.gov nameserver 192.1.2.1 nameserver 192.1.2.10 Assuming there are nameservers at 192.1.2.1 and 192.1.2.10, then when you get connected with PPP, you can reach hosts whose full names are 'hillarypc.whitehouse.gov' and 'chelseapc.whitehouse.gov' by the names 'hillarypc' and 'chelseapc'. You can probably find out the right domain name to use and the IP numbers of nameservers from whoever's providing your PPP link. CONNECTING TO A PPP SERVER To use PPP, you invoke the pppd program with appropriate options. Everything you need to know is contained in the pppd(8) manual page. However, it's useful to see some examples: Example 1: A simple dial-up connection. Here's a command for connecting to a PPP server by modem. pppd connect 'chat -v "" ATDT5551212 CONNECT "" ogin: ppp word: whitewater' \ /dev/cua1 38400 -detach debug crtscts modem defaultroute 192.1.1.17: Going through pppd's options in order: connect 'chat etc...' This gives a command to run to contact the PPP server. Here the supplied 'chat' program is used to dial a remote computer. The whole command is enclosed in single quotes because pppd expects a one-word argument for the 'connect' option. The options to 'chat' itself are: -v verbose mode; log what we do to syslog "" don't wait for any prompt, but instead... ATDT5551212 dial the modem, then CONNECT wait for answer "" send a return (null text followed by usual return) ogin: ppp word: whitewater log in. /dev/cua1 specify the callout serial port cua1 38400 specify baud rate -detach normally, pppd forks and puts itself in the background; this option prevents this debug log status in syslog crtscts use hardware flow control between computer and modem (at 38400 this is a must) modem indicate that this is a modem device; pppd will hang up the phone before and after making the call defaultroute once the PPP link is established, make it the default route; if you have a PPP link to the Internet this is probably what you want 192.1.1.17: this is a degenerate case of a general option of the form x.x.x.x:y.y.y.y . Here x.x.x.x is the local IP address and y.y.y.y is the IP address of the remote end of the PPP connection. If this option is not specified, or if just one side is specified, then x.x.x.x defaults to the IP address associated with the local machine's hostname (in /etc/hosts), and y.y.y.y is determined by the remote machine. So if this example had been taken from the fictional machine 'billpc', this option would actually be redundant. pppd will write error messages and debugging logs to the syslogd daemon using the facility name "local2". (Verbose output from chat is the same.) These messages may already be logged to the console or to a file like /usr/adm/messages; consult your /etc/syslog.conf file to see. If you want to make all pppd and chat messages go to the console, add the line local2.* /dev/console to syslog.conf; make sure to put one or more TAB characters between the two fields. Example 2: Connecting to PPP server over hard-wired link. This is a slightly more complicated example. This is the script I run to make my own PPP link, which is over a hard-wired Gandalf link to an Ultrix machine running Morningstar PPP. pppd connect /etc/ppp/ppp-connect defaultroute noipdefault debug \ kdebug 2 /dev/cua0 9600 Here /etc/ppp/ppp-connect is the following script: #! /bin/sh /etc/ppp/sendbreak chat -v -t60 "" \; "service :" blackice ogin: callahan word: PASSWORD \ black% "stty -echo; ppp" "Starting PPP now" && sleep 5 This sends a break to wake up my terminal server, sends a semicolon (which lets my terminal server do autobaud detection), then says we want the service "blackice". It logs in, waits for a shell prompt ("black%"), then starts PPP. The -t60 argument sets the timeout to a minute, since things here are sometimes very slow. The "&& sleep 5" causes the script to pause for 5 seconds, unless chat fails in which case it exits immediately. This is just to give the PPP server time to start (it's very slow). Also, the "stty -echo" turned out to be very important for me; without it, my pppd would sometimes start to send negotiation packets before the remote PPP server had time to turn off echoing. The negotiation packets would then get sent back to my local machine, be rejected (PPP is able to detect loopback) and pppd would fail before the remote PPP server even got going. The "stty -echo" command prevents this confusion. This kind of problem should only ever affect a *very* few people who connect to a PPP server that runs as a command on a slow Unix machine, but I wanted to mention it because it took me several frustrating hours to figure out. The pppd options are mostly familiar. Two that are new are "noipdefault" and "kdebug 2". "noipdefault" tells pppd to ask the remote end for the IP address to use; this is necessary if the PPP server implements dynamic IP address assignment as mine does (i.e., I don't know what address I'll get ahead of time). "kdebug 2" sets the kernel debugging level to 2, enabling slightly chattier messages from the ppp kernel code. Anyway, assuming your connection is working, you should see chat dial the modem, then perhaps some messages from pppd (depending on your syslog.conf setup), then some kernel messages like this: ppp: channel ppp0 mtu changed to 1500 ppp: channel ppp0 open ppp: channel ppp0 going up for IP packets! (These messages will only appear if you gave the option "kdebug 2" and have kern.info messages directed to the screen.) Simultaneously, pppd is also writing interesting things to /usr/adm/messages (or other log file, depending on syslog.conf). IF IT WORKS If you think you've got a connection, there are a number of things you can do to test it. First, type /sbin/ifconfig (ifconfig may live elsewhere, depending on your distribution.) This should show you all the network interfaces that are 'UP'. ppp0 should be one of them, and you should recognize the first IP address as your own and the "POINT-TO-POINT ADDR" as the address of your server. Here's what it looks like on my machine: lo Link encap Local Loopback inet addr 127.0.0.1 Bcast 127.255.255.255 Mask 255.0.0.0 UP LOOPBACK RUNNING MTU 2000 Metric 1 RX packets 0 errors 0 dropped 0 overrun 0 TX packets 0 errors 0 dropped 0 overrun 0 ppp0 Link encap Point-Point Protocol inet addr 192.76.32.2 P-t-P 129.67.1.165 Mask 255.255.255.0 UP POINTOPOINT RUNNING MTU 1500 Metric 1 RX packets 33 errors 0 dropped 0 overrun 0 TX packets 42 errors 0 dropped 0 overrun 0 Now, type ping z.z.z.z where z.z.z.z is the address of your server. This should work. Here's what it looks like for me: waddington:~$ ping 129.67.1.165 PING 129.67.1.165 (129.67.1.165): 56 data bytes 64 bytes from 129.67.1.165: icmp_seq=0 ttl=255 time=268 ms 64 bytes from 129.67.1.165: icmp_seq=1 ttl=255 time=247 ms 64 bytes from 129.67.1.165: icmp_seq=2 ttl=255 time=266 ms ^C --- 129.67.1.165 ping statistics --- 3 packets transmitted, 3 packets received, 0% packet loss round-trip min/avg/max = 247/260/268 ms waddington:~$ Try typing: netstat -nr This should show three routes, something like this: Kernel routing table Destination Gateway Genmask Flags Metric Ref Use Iface 129.67.1.165 0.0.0.0 255.255.255.255 UH 0 0 6 ppp0 127.0.0.0 0.0.0.0 255.0.0.0 U 0 0 0 lo 0.0.0.0 129.67.1.165 0.0.0.0 UG 0 0 6298 ppp0 If your output looks similar but doesn't have the destination 0.0.0.0 line (which refers to the default route used for connections), you may have run pppd without the 'defaultroute' option. At this point you can try telnetting/ftping/fingering whereever you want, bearing in mind that you'll have to use numeric IP addresses unless you've set up your /etc/resolv.conf correctly. IF IT DOESN'T WORK We have included a small FAQ of related problems in the linux directory. It is called "FAQ". If you don't seem to get a connection, the thing to do is to collect 'debug' output from pppd. To do this, make sure you run pppd with the 'debug' option, and put the following two lines in your /etc/syslog.conf file: local2.* /dev/console local2.* /usr/adm/ppplog This will cause pppd's messages to be written to the current virtual console and to the file /usr/adm/ppplog. Note that the left-hand field and the right-hand field must be separated by at least one TAB character. After modifying /etc/syslog.conf, you must execute the command 'kill -HUP ' where is the process ID of the currently running syslogd process to cause it to re-read the configuration file. Some messages to look for: - "pppd[NNN]: Connected..." means that the "connect" script has completed successfully. - "pppd[NNN]: sent [LCP ConfReq"... means that pppd has attempted to begin negotiation with the remote end. - "pppd[NNN]: recv [LCP ConfReq"... means that pppd has received a negotiation frame from the remote end. - "pppd[NNN]: ipcp up" means that pppd has reached the point where it believes the link is ready for IP traffic to travel across it. If you never see a "recv" message then there may be serious problems with your link. (For example, the link may not be passing all 8 bits.) If that's the case, it would be useful to collect a debug log which contains all the bytes being passed between your computer and the remote PPP server. To do this, alter your syslog.conf lines to look like this local2.*,kern.* /dev/console local2.*,kern.* /usr/adm/ppplog and HUP the syslog daemon as before. Then, run pppd with the option "kdebug 5". Whatever characters arrive over the PPP terminal line will appear in the debugging output. Occasionally you may see a message like ppp_toss: tossing frame, reason = 4 The PPP code is throwing away a packet ("frame") from the remote server because of a serial overrun. This means your CPU isn't able to read characters from the serial port as quickly as they arrive; the best solution is to get a 16550A serial chip, which gives the CPU some grace period. Reasons other than 4 indicate other kinds of serial errors, which should not occur. During the initial connection sequence, you may see one or more messages which indicate "bad fcs". This refers to a checksum error in a received PPP frame, and usually occurs at the start of a session when the peer system is sending some "text" messages, such as "hello this is the XYZ company". Messages of "bad fcs" once the link is established and the routes have been added are not normal and indicate transmssion errors or noise on the telephone line. IF IT STILL DOESN'T WORK (OR, BUG REPORTS) If you're still having difficulty, send the linux-activists PPP channel a bug report. It is extremely important to include as much information as possible; for example: - the version number of the kernel you are using - the version number of Linux PPP you are using - the exact command you use to start the PPP session - log output from a session run with the 'debug' option, captured using local2.*,kern.* in your syslog.conf file - the type of PPP peer that you are connecting to (eg, Xyzzy Corp terminal server, Morningstar PPP software, etc) - the kind of connection you use (modem, hardwired, etc...) DYNAMIC ADDRESS ASSIGNMENT You can use Linux PPP with a PPP server which assigns a different IP address every time you connect. You need to use the 'noipdefault' option to tell pppd to request the IP address from the remote host. Sometimes you may get an error message like "Cannot assign requested address" when you use a Linux client (for example, "talk"). This happens when the IP address given in /etc/hosts for our hostname differs from the IP address used by the PPP interface. The solution is to use ifconfig ppp0 to get the interface address and then edit /etc/hosts appropriately. SETTING UP A MACHINE FOR INCOMING PPP CONNECTIONS Suppose you want to permit another machine to call yours up and start a PPP session. This is possible using Linux PPP. Before doing anything else, you should make sure that your Linux machine is set up to answer incoming calls and provide login prompts over the serial line. You should verify that you can dial up, log in as a normal user, get a shell prompt, and issue some commands. This document does not discuss how to set up a Linux machine to answer incoming calls; for that, refer to the documentation that came with your distribution or the Serial HOWTO. Now, to provide PPP service, one way is to create an account named, say, 'ppp', with the login shell being a short script that starts pppd. For example, the /etc/passwd entry might look like this: ppp:(encrypted password):102:50:PPP client login:/tmp:/etc/ppp/ppplogin Here the file /etc/ppp/ppplogin would be an executable script containing something like: #!/bin/sh exec /usr/etc/pppd passive :192.1.2.23 Here we will insist that the remote machine use IP address 192.1.2.23, while the local PPP interface will use the IP address associated with this machine's hostname in /etc/hosts. The 'passive' option (which is not required) just means that pppd will try to open negotiations when it starts, but if it receives no reply it will just wait silently. This is appropriate if the remote end might take some time before it's ready to negotiate. (Note that the meaning of the 'passive' option changed between ppp-1.3 and ppp-2.0.) This setup is sufficient if you just want to connect two machines so that they can talk to one another. If you want to use Linux PPP to connect a single machine to an entire network, or to connect two networks together, then you need to arrange for packets to be routed from the networks to the PPP link. Setting up a link between networks is beyond the scope of this document; you should examine the routing options in the manual page for pppd carefully and find out about routed, etc. Let's consider just the first case. Suppose you have a Linux machine attached to an Ethernet, and you want to allow its PPP peer to be able to communicate with hosts on that Ethernet. To do this, you should have the remote machine use an IP address that would normally appear to be on the local Ethernet segment and you should give the 'proxyarp' option to pppd on the server. Suppose, for example, we have this setup: 192.1.2.23 192.1.2.17 +-----------+ PPP link +----------+ | chelseapc | ------------------- | billpc | +-----------+ +----------+ | Ethernet ----------------------------------- 192.1.2.x Here the PPP and Ethernet interfaces of billpc will have IP address 192.1.2.17. (It's OK for one or more PPP interfaces on a machine to share an IP address with an Ethernet interface.) There is an appropriate entry in /etc/passwd on billpc to allow chelseapc to call in, with the /etc/ppp/ppplogin script containing #!/bin/sh exec /usr/etc/pppd passive proxyarp :192.1.2.23 When the link comes up, pppd will enter a "proxy arp" entry for chelseapc into the arp table on billpc. What this means effectively is that billpc will pretend to the other machines on the 192.1.2.x Ethernet that its Ethernet interface is ALSO the interface for chelseapc (192.1.2.23) as well as billpc (192.1.2.17). In practice this means that chelseapc can communicate just as if it was directly connected to the Ethernet. ADDING MORE PPP CHANNELS By default, Linux PPP comes with 4 kernel channels, which means that at most 4 simultaneous PPP sessions are possible. If you desire more such sessions (for example if you are serving many dialup lines), you can easily reconfigure the kernel to add new channels. There are two steps. First you need to edit the kernel file drivers/net/Space.c . As distributed, it contains a section that looks like this: #if defined(CONFIG_PPP) extern int ppp_init(struct device *); static struct device ppp3_dev = { "ppp3", 0x0, 0x0, 0x0, 0x0, 3, 0, 0, 0, 0, NEXT_DEV, ppp_init, }; static struct device ppp2_dev = { "ppp2", 0x0, 0x0, 0x0, 0x0, 2, 0, 0, 0, 0, &ppp3_dev, ppp_init, }; static struct device ppp1_dev = { "ppp1", 0x0, 0x0, 0x0, 0x0, 1, 0, 0, 0, 0, &ppp2_dev, ppp_init, }; static struct device ppp0_dev = { "ppp0", 0x0, 0x0, 0x0, 0x0, 0, 0, 0, 0, 0, &ppp1_dev, ppp_init, }; #undef NEXT_DEV #define NEXT_DEV (&ppp0_dev) #endif /* PPP */ The pattern should be obvious. For more channels, you need to add more "static struct device pppN_dev" lines, changing the first, sixth and eleventh structure entries as appropriate. The highest numbered PPP device should have NEXT_DEV in its eleventh structure field, and you should change the ppp3_dev structure to have &ppp4_dev there instead. For example, to add 2 extra channels, you would have #if defined(CONFIG_PPP) extern int ppp_init(struct device *); static struct device ppp5_dev = { "ppp5", 0x0, 0x0, 0x0, 0x0, 5, 0, 0, 0, 0, NEXT_DEV, ppp_init, }; static struct device ppp4_dev = { "ppp4", 0x0, 0x0, 0x0, 0x0, 4, 0, 0, 0, 0, &ppp5_dev, ppp_init, }; static struct device ppp3_dev = { "ppp3", 0x0, 0x0, 0x0, 0x0, 3, 0, 0, 0, 0, &ppp4_dev, ppp_init, }; ... etc. Second, you need to change the line in ppp.h (in include/linux) to change the line that reads #define PPP_NRUNIT 4 to show the new number of channels; in our case it would become #define PPP_NRUNIT 6 Finally, recompile and reboot. The bootup message and the contents of /proc/net/dev should show the correct number of channels. CHANGES FROM LINUX PPP 0.1.x Linux PPP 0.1.x was based on the free PPP package PPP-1.3. Linux PPP 0.2.1 is based on PPP-2.0.4. There have been some changes to the pppd options along with significant enhancements. You should read "RELNOTES" in the pppd directory for a description of the changes. Also, some options which were added to PPP-1.3 for the Linux version have now changed names: 'defroute' is now 'defaultroute' 'kerndebug' is now 'kdebug' 'dropdtr' is now 'modem' In addition, it is now necessary to use the 'noipdefault' option if you want to get the local IP address from the remote PPP server. CONCLUSION Good luck! Michael