Add some more documentation
[openwrt/svn-archive/archive.git] / docs / adding.tex
1 Linux is now one of the most widespread operating system for embedded devices due
2 to its openess as well as the wide variety of platforms it can run on. Many
3 manufacturer actually use it in firmware you can find on many devices: DVB-T
4 decoders, routers, print servers, DVD players ... Most of the time the stock
5 firmware is not really open to the consumer, even if it uses open source software.
6
7 You might be interested in running a Linux based firmware for your router for
8 various reasons: extending the use of a network protocol (such as IPv6), having
9 new features, new piece of software inside, or for security reasons. A fully
10 open-source firmware is de-facto needed for such applications, since you want to
11 be free to use this or that version of a particular reason, be able to correct a
12 particular bug. Few manufacturers do ship their routers with a Sample Development Kit,
13 that would allow you to create your own and custom firmware and most of the time,
14 when they do, you will most likely not be able to complete the firmware creation process.
15
16 This is one of the reasons why OpenWrt and other firmware exists: providing a
17 version independent, and tools independent firmware, that can be run on various
18 platforms, known to be running Linux originaly.
19
20 \subsection{Which Operating System does this device run?}
21
22 There is a lot of methods to ensure your device is running Linux. Some of them do
23 need your router to be unscrewed and open, some can be done by probing the device
24 using its external network interfaces.
25
26 \subsubsection{Operating System fingerprinting and port scanning}
27
28 A large bunch of tools over the Internet exists in order to let you do OS
29 fingerprinting, we will show here an example using \textbf{nmap}:
30
31 \begin{Verbatim}
32 nmap -P0 -O <IP address>
33 Starting Nmap 4.20 ( http://insecure.org ) at 2007-01-08 11:05 CET
34 Interesting ports on 192.168.2.1:
35 Not shown: 1693 closed ports
36 PORT STATE SERVICE
37 22/tcp open ssh
38 23/tcp open telnet
39 53/tcp open domain
40 80/tcp open http
41 MAC Address: 00:13:xx:xx:xx:xx (Cisco-Linksys)
42 Device type: broadband router
43 Running: Linksys embedded
44 OS details: Linksys WRT54GS v4 running OpenWrt w/Linux kernel 2.4.30
45 Network Distance: 1 hop
46 \end{Verbatim}
47
48 nmap is able to report whether your device uses a Linux TCP/IP stack, and if so,
49 will show you which Linux kernel version is probably runs. This report is quite
50 reliable and it can make the distinction between BSD and Linux TCP/IP stacks and others.
51
52 Using the same tool, you can also do port scanning and service version discovery.
53 For instance, the following command will report which IP-based services are running
54 on the device, and which version of the service is being used:
55
56 \begin{verbatim}
57 nmap -P0 -sV <IP address>
58 Starting Nmap 4.20 ( http://insecure.org ) at 2007-01-08 11:06 CET
59 Interesting ports on 192.168.2.1:
60 Not shown: 1693 closed ports
61 PORT STATE SERVICE VERSION
62 22/tcp open ssh Dropbear sshd 0.48 (protocol 2.0)
63 23/tcp open telnet Busybox telnetd
64 53/tcp open domain ISC Bind dnsmasq-2.35
65 80/tcp open http OpenWrt BusyBox httpd
66 MAC Address: 00:13:xx:xx:xx:xx (Cisco-Linksys)
67 Service Info: Device: WAP
68 \end{verbatim}
69
70 The web server version, if identified, can be determining in knowing the Operating
71 System. For instance, the \textbf{BOA} web server is typical from devices running
72 an open-source Unix or Unix-like.
73
74 \subsubsection{Wireless Communications Fingerprinting}
75
76 Although this method is not really known and widespread, using a wireless scanner
77 to discover which OS your router or Access Point run can be used. We do not have
78 a clear example of how this could be achieved, but you will have to monitor raw
79 802.11 frames and compare them to a very similar device running a Linux based firmware.
80
81 \subsubsection{Web server security exploits}
82
83 The Linksys WRT54G was originally hacked by using a "ping bug" discovered in the
84 web interface. This tip has not been fixed for months by Linksys, allowing people
85 to enable the "boot\_wait" helper process via the web interface. Many web servers
86 used in firmwares are open source web server, thus allowing the code to be audited
87 to find an exploit. Once you know the web server version that runs on your device,
88 by using \textbf{nmap -sV} or so, you might be interested in using exploits to reach
89 shell access on your device.
90
91 \subsubsection{Native Telnet/SSH access}
92
93 Some firmwares might have restricted or unrestricted Telnet/SSH access, if so,
94 try to log in with the web interface login/password and see if you can type in
95 some commands. This is actually the case for some Broadcom BCM963xx based firmwares
96 such as the one in Neuf/Cegetel ISP routers, Club-Internet ISP CI-Box and many
97 others. Some commands, like \textbf{cat} might be left here and be used to
98 determine the Linux kernel version.
99
100 \subsubsection{Analysing a binary firmware image}
101
102 You are very likely to find a firmware binary image on the manufacturer website,
103 even if your device runs a proprietary operating system. If so, you can download
104 it and use an hexadecimal editor to find printable words such as \textbf{vmlinux},
105 \textbf{linux}, \textbf{ramdisk}, \textbf{mtd} and others.
106
107 Some Unix tools like \textbf{hexdump} or \textbf{strings} can be used to analyse
108 the firmware. Below there is an example with a binary firmware found other the Internet:
109
110 \begin{verbatim}
111 hexdump -C <binary image.extension> | less (more)
112 00000000 46 49 52 45 32 2e 35 2e 30 00 00 00 00 00 00 00 |FIRE2.5.0.......|
113 00000010 00 00 00 00 31 2e 30 2e 30 00 00 00 00 00 00 00 |....1.0.0.......|
114 00000020 00 00 00 00 00 00 00 38 00 43 36 29 00 0a e6 dc |.......8.C6)..??|
115 00000030 54 49 44 45 92 89 54 66 1f 8b 08 08 f8 10 68 42 |TIDE..Tf....?.hB|
116 00000040 02 03 72 61 6d 64 69 73 6b 00 ec 7d 09 bc d5 d3 |..ramdisk.?}.???|
117 00000050 da ff f3 9b f7 39 7b ef 73 f6 19 3b 53 67 ea 44 |???.?9{?s?.;Sg?D|
118 \end{verbatim}
119
120 Scroll over the firmware to find printable words that can be significant.
121
122 \subsubsection{Amount of flash memory}
123
124 Linux can hardly fit in a 2MB flash device, once you have opened the device and
125 located the flash chip, try to find its characteristics on the Internet. If
126 your flash chip is a 2MB or less device, your device is most likely to run a
127 proprietary OS such as WindRiver VxWorks, or a custom manufacturer OS like Zyxel ZynOS.
128
129 OpenWrt does not currently run on devices which have 2MB or less of flash memory.
130 This limitation will probably not be worked around since those devices are most
131 of the time micro-routers, or Wireless Access Points, which are not the main
132 OpenWrt target.
133
134 \subsubsection{Pluging a serial port}
135
136 By using a serial port and a level shifter, you may reach the console that is being shown by the device
137 for debugging or flashing purposes. By analysing the output of this device, you can
138 easily notice if the device uses a Linux kenrel or something different.
139
140 \subsection{Finding and using the manufacturer SDK}
141
142 Once you are sure your device run a Linux based firmware, you will be able to start
143 hacking on it. If the manufacturer respected the GPL, it will have released a Sample
144 Development Kit with the device.
145
146 \subsubsection{GPL violations}
147
148 Some manufacturers do release a Linux based binary firmware, with no sources at all.
149 The first step before doing anything is to read the license coming with your device,
150 then write them about this lack of Open Source code. If the manufacturer answers
151 you they do not have to release a SDK containing Open Source software, then we
152 recommend you get in touch with the gpl-violations.org community.
153
154 You will find below a sample letter that can be sent to the manufacturer:
155
156 \begin{verse}
157 Miss, Mister,
158
159 I am using a <device name>, and I cannot find neither on your website nor on the
160 CD-ROM the open source software used to build or modify the firmware.
161
162 In conformance to the GPL license, you have to release the following sources:
163
164 \begin{itemize}
165 \item complete toolchain that made the kernel and applications be compiled (gcc, binutils, libc)
166 \item tools to build a custom firmware (mksquashfs, mkcramfs ...)
167 \item kernel sources with patches to make it run on this specific hardware, this does not include binary drivers
168 \end{itemize}
169
170 Thank you very much in advance for your answer.
171
172 Best regards, <your name>
173 \end{verse}
174
175 \subsubsection{Using the SDK}
176
177 Once the SDK is available, you are most likely not to be able to build a complete
178 or functional firmware using it, but parts of it, like only the kernel, or only
179 the root filesystem. Most manufacturers do not really care releasing a tool that
180 do work every time you uncompress and use it.
181
182 You should anyway be able to use the following components:
183
184 \begin{itemize}
185 \item kernel sources with more or less functional patches for your hardware
186 \item binary drivers linked or to be linked with the shipped kernel version
187 \item packages of the toolchain used to compile the whole firmware: gcc, binutils, libc or uClibc
188 \item binary tools to create a valid firmware image
189 \end{itemize}
190
191 Your work can be divided into the following tasks:
192
193 \begin{itemize}
194 \item create a clean patch of the hardware specific part of the linux kernel
195 \item spot potential kernel GPL violations especially on netfilter and USB stack stuff
196 \item make the binary drivers work, until there are open source drivers
197 \item use standard a GNU toolchain to make working executables
198 \item understand and write open source tools to generate a valid firmware image
199 \end{itemize}
200
201 \subsubsection{Creating a hardware specific kernel patch}
202
203 Most of the time, the kernel source that comes along with the SDK is not really
204 clean, and is not a standard Linux version, it also has architecture specific
205 fixes backported from the \textbf{CVS} or the \textbf{git} repository of the
206 kernel development trees. Anyway, some parts can be easily isolated and used as
207 a good start to make a vanilla kernel work your hardware.
208
209 Some directories are very likely to have local modifications needed to make your
210 hardware be recognized and used under Linux. First of all, you need to find out
211 the linux kernel version that is used by your hardware, this can be found by
212 editing the \textbf{linux/Makefile} file.
213
214 \begin{verbatim}
215 head -5 linux-2.x.x/Makefile
216 VERSION = 2
217 PATCHLEVEL = x
218 SUBLEVEL = y
219 EXTRAVERSION = z
220 NAME=A fancy name
221 \end{verbatim}
222
223 So now, you know that you have to download a standard kernel tarball at
224 \textbf{kernel.org} that matches the version being used by your hardware.
225
226 Then you can create a \textbf{diff} file between the two trees, especially for the
227 following directories:
228
229 \begin{verbatim}
230 diff -urN linux-2.x.x/arch/<sub architecture> linux-2.x.x-modified/arch/<sub architecture> > 01-architecture.patch
231 diff -urN linux-2.x.x/include/ linux-2.x.x-modified/include > 02-includes.patch
232 diff -urN linux-2.x.x/drivers/ linux-2.x.x-modified/drivers > 03-drivers.patch
233 \end{verbatim}
234
235 This will constitute a basic set of three patches that are very likely to contain
236 any needed modifications that has been made to the stock Linux kernel to run on
237 your specific device. Of course, the content produced by the \textbf{diff -urN}
238 may not always be relevant, so that you have to clean up those patches to only
239 let the "must have" code into them.
240
241 The fist patch will contain all the code that is needed by the board to be
242 initialized at startup, as well as processor detection and other boot time
243 specific fixes.
244
245 The second patch will contain all useful definitions for that board: adresses,
246 kernel granularity, redefinitions, processor family and features ...
247
248 The third patch may contain drivers for: serial console, ethernet NIC, wireless
249 NIC, USB NIC ... Most of the time this patch contains nothing else than "glue"
250 code that has been added to make the binary driver work with the Linux kernel.
251 This code might not be useful if you plan on writing drivers from scratch for
252 this hardware.
253
254 \subsubsection{Using the device bootloader}
255
256 The bootloader is the first program that is started right after your device has
257 been powered on. This program, can be more or less sophisticated, some do let you
258 do network booting, USB mass storage booting ... The bootloader is device and
259 architeture specific, some bootloaders were designed to be universal such as
260 RedBoot or U-Boot so that you can meet those loaders on totally different
261 platforms and expect them to behave the same way.
262
263 If your device runs a proprietary operating system, you are very likely to deal
264 with a proprietary boot loader as well. This may not always be a limitation,
265 some proprietary bootloaders can even have source code available (i.e : Broadcom CFE).
266
267 According to the bootloader features, hacking on the device will be more or less
268 easier. It is very probable that the bootloader, even exotic and rare, has a
269 documentation somewhere over the Internet. In order to know what will be possible
270 with your bootloader and the way you are going to hack the device, look over the
271 following features :
272
273 \begin{itemize}
274 \item does the bootloader allow net booting via bootp/DHCP/NFS or tftp
275 \item does the bootloader accept loading ELF binaries ?
276 \item does the bootloader have a kernel/firmware size limitation ?
277 \item does the bootloader expect a firmware format to be loaded with ?
278 \item are the loaded files executed from RAM or flash ?
279 \end{itemize}
280
281 Net booting is something very convenient, because you will only have to set up network
282 booting servers on your development station, and keep the original firmware on the device
283 till you are sure you can replace it. This also prevents your device from being flashed,
284 and potentially bricked every time you want to test a modification on the kernel/filesystem.
285
286 If your device needs to be flashed every time you load a firmware, the bootlader might
287 only accept a specific firmware format to be loaded, so that you will have to
288 understand the firmware format as well.
289
290 \subsubsection{Making binary drivers work}
291
292 As we have explained before, manufacturers do release binary drivers in their GPL
293 tarball. When those drivers are statically linked into the kernel, they become GPL
294 as well, fortunately or unfortunately, most of the drivers are not statically linked.
295 This anyway lets you a chance to dynamically link the driver with the current kernel
296 version, and try to make them work together.
297
298 This is one of the most tricky and grey part of the fully open source projects.
299 Some drivers require few modifications to be working with your custom kernel,
300 because they worked with an earlier kernel, and few modifications have been made
301 to the kernel in-between those versions. This is for instance the case with the
302 binary driver of the Broadcom BCM43xx Wireless Chipsets, where only few differences
303 were made to the network interface structures.
304
305 Some general principles can be applied no matter which kernel version is used in
306 order to make binary drivers work with your custom kernel:
307
308 \begin{itemize}
309 \item turn on kernel debugging features such as:
310 \begin{itemize}
311 \item CONFIG\_DEBUG\_KERNEL
312 \item CONFIG\_DETECT\_SOFTLOCKUP
313 \item CONFIG\_DEBUG\_KOBJECT
314 \item CONFIG\_KALLSYMS
315 \item CONFIG\_KALLSYMS\_ALL
316 \end{itemize}
317 \item link binary drivers when possible to the current kernel version
318 \item try to load those binary drivers
319 \item catch the lockups and understand them
320 \end{itemize}
321
322 Most of the time, loading binary drivers will fail, and generate a kernel oops.
323 You can know the last symbol the binary drivers attempted to use, and see in the
324 kernel headers file, if you do not have to move some structures field before or
325 after that symbol in order to keep compatibily with both the binary driver and
326 the stock kernel drivers.
327
328 \subsubsection{Understanding the firmware format}
329
330 You might want to understand the firmware format, even if you are not yet capable
331 of running a custom firmware on your device, because this is sometimes a blocking
332 part of the flashing process.
333
334 A firmare format is most of the time composed of the following fields:
335
336 \begin{itemize}
337 \item header, containing a firmare version and additional fields: Vendor, Hardware version ...
338 \item CRC32 checksum on either the whole file or just part of it
339 \item Binary and/or compressed kernel image
340 \item Binary and/or compressed root filesystem image
341 \item potential garbage
342 \end{itemize}
343
344 Once you have figured out how the firmware format is partitioned, you will have
345 to write your own tool that produces valid firmare binaries. One thing to be very
346 careful here is the endianness of either the machine that produces the binary
347 firmware and the device that will be flashed using this binary firmware.
348
349 \subsubsection{Writing a flash map driver}
350
351 The flash map driver has an important role in making your custom firmware work
352 because it is responsible of mapping the correct flash regions and associated
353 rights to specific parts of the system such as: bootloader, kernel, user filesystem.
354
355 Writing your own flash map driver is not really a hard task once you know how your
356 firmware image and flash is structured. You will find below a commented example
357 that covers the case of the device where the bootloader can pass to the kernel its partition plan.
358
359 First of all, you need to make your flash map driver be visible in the kernel
360 configuration options, this can be done by editing the file
361 \textbf{linux/drivers/mtd/maps/Kconfig}:
362
363 \begin{verbatim}
364 config MTD_DEVICE_FLASH
365 tristate "Device Flash device"
366 depends on ARCHITECTURE && DEVICE
367 help
368 Flash memory access on DEVICE boards. Currently only works with
369 Bootloader Foo and Bootloader Bar.
370 \end{verbatim}
371
372 Then add your source file to the \textbf{linux/drivers/mtd/maps/Makefile}, so
373 that it will be compiled along with the kernel.
374
375 \begin{verbatim}
376 obj-\$(CONFIG_MTD_DEVICE_FLASH) += device-flash.o
377 \end{verbatim}
378
379 You can then write the kernel driver itself, by creating a
380 \textbf{linux/drivers/mtd/maps/device-flash.c} C source file.
381
382 \begin{verbatim}
383 // Includes that are required for the flash map driver to know of the prototypes:
384 #include <asm/io.h>
385 #include <linux/init.h>
386 #include <linux/kernel.h>
387 #include <linux/mtd/map.h>
388 #include <linux/mtd/mtd.h>
389 #include <linux/mtd/partitions.h>
390 #include <linux/vmalloc.h>
391
392 // Put some flash map definitions here:
393 #define WINDOW_ADDR 0x1FC00000 /* Real address of the flash */
394 #define WINDOW_SIZE 0x400000 /* Size of flash */
395 #define BUSWIDTH 2 /* Buswidth */
396
397 static void __exit device_mtd_cleanup(void);
398
399 static struct mtd_info *device_mtd_info;
400
401 static struct map_info devicd_map = {
402 .name = "device",
403 .size = WINDOW_SIZE,
404 .bankwidth = BUSWIDTH,
405 .phys = WINDOW_ADDR,
406 };
407
408 static int __init device_mtd_init(void)
409 {
410 // Display that we found a flash map device
411 printk("device: 0x\%08x at 0x\%08x\n", WINDOW_SIZE, WINDOW_ADDR);
412 // Remap the device address to a kernel address
413 device_map.virt = ioremap(WINDOW_ADDR, WINDOW_SIZE);
414
415 // If impossible to remap, exit with the EIO error
416 if (!device_map.virt) {
417 printk("device: Failed to ioremap\n");
418 return -EIO;
419 }
420
421 // Initlialise the device map
422 simple_map_init(&device_map);
423
424 /* MTD informations are closely linked to the flash map device
425 you might also use "jedec_probe" "amd_probe" or "intel_probe" */
426 device_mtd_info = do_map_probe("cfi_probe", &device_map);
427
428 if (device_mtd_info) {
429 device_mtd_info->owner = THIS_MODULE;
430
431 int parsed_nr_parts = 0;
432
433 // We try here to use the partition schema provided by the bootloader specific code
434 if (parsed_nr_parts == 0) {
435 int ret = parse_bootloader_partitions(device_mtd_info, &parsed_parts, 0);
436 if (ret > 0) {
437 part_type = "BootLoader";
438 parsed_nr_parts = ret;
439 }
440 }
441
442 add_mtd_partitions(devicd_mtd_info, parsed_parts, parsed_nr_parts);
443
444 return 0;
445 }
446 iounmap(device_map.virt);
447
448 return -ENXIO;
449 }
450
451 // This function will make the driver clean up the MTD device mapping
452 static void __exit device_mtd_cleanup(void)
453 {
454 // If we found a MTD device before
455 if (device_mtd_info) {
456 // Delete every partitions
457 del_mtd_partitions(device_mtd_info);
458 // Delete the associated map
459 map_destroy(device_mtd_info);
460 }
461
462 // If the virtual address is already in use
463 if (device_map.virt) {
464 // Unmap the physical address to a kernel space address
465 iounmap(device_map.virt);
466 // Reset the structure field
467 device_map.virt = 0;
468 }
469 }
470
471
472 // Macros that indicate which function is called on loading/unloading the module
473 module_init(device_mtd_init);
474 module_exit(device_mtd_cleanup);
475
476
477 // Macros defining licence and author, parameters can be defined here too.
478 MODULE_LICENSE("GPL");
479 MODULE_AUTHOR("Me, myself and I <memyselfandi@domain.tld");
480 \end{verbatim}