Customizing Asus AM200g - V. firmware modification

3.7.2008 01:45 | Přečteno: 8578× | Linux | | poslední úprava: 3.7.2008 01:53

I would not dare to upload a firmware built from scratch, but what about only slightly modified one? For me, it would suffice to just add a call to some script on the flash drive that would be launched at boot. From there I'd be able to do all I need (ATM).
The WL-600g GPL Makefile contains the following default sequence of commands in its buildimage section:

./buildFS
$(HOSTTOOLS_DIR)/mksquashfs $(TARGET_FS) $(PROFILE_DIR)/rootfs.img -noappend -be -lzma -no-fragments -noI
$(OBJCOPY) -O binary vmlinux vmlinux.bin
$(HOSTTOOLS_DIR)/cmplzma -k -2 vmlinux vmlinux.bin vmlinux.lz;\
$(HOSTTOOLS_DIR)/bcmImageBuilder --output $(CFE_FS_KERNEL_IMAGE_NAME) --chip $(BRCM_CHIP) --board $(BRCM_BOARD_ID) \
--productname $(CUSTOMER_PRODUCT_NAME) --blocksize $(BRCM_FLASHBLK_SIZE) --cfefile $(CFE_FILE) --rootfsfile rootfs.img --kernelfile vmlinux.lz --include-cfe;

Let's keep hoping that WL-600g is the same as AM-200g, try to reverse the process, do the modifications and put together again. bcmImageBuilder should just put together the filesystem file, kernel, cfe file (st like bios - see OpenWRT - Everything you need to know about broadcom hardware (Part 1)) and some checksums. There is a utility done by OpenWRT people that should be a replacement of bcmImageBuilder. Here. From there I learned the structure of the header, let's repeat it here for the sake of completeness: The header takes the first 256 bytes

/* Image component */
	struct imagecomp {
	        uint8_t                 address[12];    /* Address of this component as ASCII */
	        uint8_t                 len[10];        /* Length of this component as ASCII */
	};
	
	/* Image tag */
	struct imagetag {
	        uint8_t                 tagver[4];      /*   0 -   3: Version of the tag as ASCII (2) */
	        uint8_t                 sig1[20];       /*   4 -  23: BCM_MAGIC_1 */
	        uint8_t                 sig2[14];       /*  24 -  37: BCM_MAGIC_2 */
	        uint8_t                 chipid[6];      /*  38 -  43: Chip id as ASCII (6345) */
	        uint8_t                 boardid[16];    /*  44 -  59: Board id as ASCII (96345GW2, etc...) */
	        uint8_t                 bigendian[2];   /*  60 -  61: "1" for big endian, "0" for little endian */
	        uint8_t                 imagelen[10];   /*  62 -  71: The length of all data that follows */
	        struct imagecomp        cfe;            /*  72 -  93: The offset and length of CFE */
	        struct imagecomp        rootfs;         /*  94 - 115: The offset and length of the root file system */
	        struct imagecomp        kernel;         /* 116 - 137: The offset and length of the kernel */
	        uint8_t                 dualimage[2];   /* 138 - 139: use "0" here */
	        uint8_t                 inactive[2];    /* 140 - 141: use "0" here */
	        uint8_t                 reserved1[74];  /* 142 - 215: reserved */
	        uint32_t                imagecrc;       /* 216 - 219: crc of the images (net byte order) */
	        uint8_t                 reserved2[16];  /* 220 - 235: reserved */
	        uint32_t                headercrc;      /* 236 - 239: crc starting from sig1 until headercrc (net byte order) */
	        uint8_t                 reserved3[16];  /* 240 - 255: reserved */
	};

Let's decompose it for the firmware from Joyce:

tagver="36 00 00 00" = "6..."
sig1="42 72 6f 61 64 63 6f 6d 20 43 6f 72 70 6f 72 61 74 69 6f 00" = "Broadcom Corporatio."
sig2="76 65 72 2e 20 32 2e 30 00 00 00 00 00 00" = "ver. 2.0......"
chipid="36 33 34 38 00 00" = "6348.."
boardid="39 36 33 34 38 47 57 2d 31 31 00 00 00 00 00 00" = "96348GW-11......"
bigendian="31 00" = "1"
imagelen="33 38 32 32 33 35 36 00 00 00" = "3822356..." #indeed 3822356 + 256 = 3822612 (the size of the firmware file)
cfe="33 32 31 37 30 33 31 31 36 38 00 00 36 32 31 37 32 00 00 00 00 00" = "32170311..62172....." # this starts to be tricky ... 
## 62172 is clearly the cfe size, but the cfe itself is slightly different from the files provided in the WL-600g GPL package.
rootfs="33 32 68 31 37 30 39 36 39 36 30 00 00 33 31 33 37 35 33 36 00 00 00" = "3217096960..3137536..." #fs size = 3137536
kernel="33 32 32 30 32 33 34 34 39 36 00 00 36 32 32 36 34 38 00 00 00 00" = "3220234496..622648...." # kernel size = 622648
## 3137536+622648+62172+256=3822612 - all fits so far!
dualimage="00 00"
inactive="00 00"
reserved="00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
          00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
          00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
          00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
          00 00 00 00 00 00 00 00 00 00"
imagecrc="c9 ea a8 50"
reserved2=""

The reserved2 part is different from WL-600g - after some experimentation, I found out that the first four bytes are CRC of the fs image and the second four bytes are CRC of the kernel, but was not able to figure out the CRC algorithm.

to extract cfe:
dd bs=1 if=JOYCE_IAD_U2_B_306063520_cfe_fs_kernel of=cfe skip=256 count=62172
to extract fs:
dd bs=1 if=JOYCE_IAD_U2_B_306063520_cfe_fs_kernel of=fs skip=62428 count=3137536
to extract kernel:
dd bs=1 if=JOYCE_IAD_U2_B_306063520_cfe_fs_kernel of=kernel skip=3199964 count=622648

By calling the Asus' bcmImageBuilder command on the extracted files, I got a firmware identical (verified with md5sum) to the original:

../../Dokumenty/GPL_AM604G_ForRussia/GENERIC_6348_WLAN_A34_3-06-02-01/hostTools/bcmImageBuilder --output bcmimage --chip 6348 --board 96348GW-11 --blocksize 64 --cfefile cfe --rootfsfile fs --kernelfile kernel --include-cfe

So, let's proceed further...

$ file fs
fs: Squashfs filesystem, big endian, version 2.0, 3136535 bytes, 423 inodes, blocksize: 65536 bytes, created: Mon May 28 08:35:59 2007

the file command confirms the fs file is indeed a squashfs image. Let's extract it:

$ unsquashfs -s fs
Reading a different endian SQUASHFS filesystem on fs
Can't find a SQUASHFS superblock on fs

This is already reported in Ubuntu, I reported it to the Mandriva guys as well. The hand-compiled squashfs 3.3 seemed to work just fine, but it was not able to deal with the Broadcom specific lzma compression. I tried to mount the fs file on the router, but unfortunately mount was not compiled with support for mounting loop devices

# mount -o ro -o loop /var/usb/usb_2/fs /var/usb/a
mount: Mounting /var/usb/usb_2/fs on /var/usb/a failed: Block device required

After some retries I found out about the excellent OpenBox4 people hacking another Broadcom based ADSL router. They developed nb4-unsquash - a tool to extract the content of Broadcom-lzma-compressed squashfs.

But I still did not trust it entirely. I found out I can mount the internal flash and that's enough for me. First I made a backup of the original fs:

# mkdir /var/usb/a
# mount -o ro /dev/mtdblock0 /var/usb/a
# /var/usb/usb_1/bin/tar cvf /var/usb/usb_2/firmwarefs.tar /var/usb/a/

Copied the filesystem to my PC:

on-my-PC $ mkdir /tmp/fs
on-my-PC $ netcat -l -p 9000 | dd of=/tmp/fs/fs.tar
on-router# /var/usb/usb_1/bin/dd if=/var/usb/usb_2/firmwarefs.tar |/var/usb/usb_1/bin/mipsel-linux-netcat 192.168.1.64 9000

Unpacked (as root because the contents are special files):

$ su
# cd /tmp/fs
# tar xvf fs.tar

Before doing any modifications I tried recreating the squashfs image:

# /home/hajma/Dokumenty/GPL_AM604G_ForRussia/GENERIC_6348_WLAN_A34_3-06-02-01/hostTools/mksquashfs /tmp/fs/var/usb/a rootfs.img -noappend -be -lzma -no-fragments -noI 

and verified with the nb4-unsquash that it looks similar to the original one:

# nb4-unsquash -d new rootfs.img
# nb4-unsquash -d old /tmp/joycefs 
# ls -laR old/ > /tmp/oldfslist
# ls -laR new/ > /tmp/newfslist
# kdiff3 /tmp/oldfslist /tmp/newfslist

I saw that all except some modification times was identical, so I put the recreated fs together with the old kernel and cfe:

# /home/hajma/Dokumenty/GPL_AM604G_ForRussia/GENERIC_6348_WLAN_A34_3-06-02-01/hostTools/bcmImageBuilder --output bcmimage --chip 6348 --board 96348GW-11 --blocksize 64 --cfefile cfe --rootfsfile rootfs.img --kernelfile kernel --include-cfe

Then I took a deep breath and flashed the router with the recreated firmware ... ... ... ... ... ... And it worked!

Unfortunately it was not possible to simply find a place in the boot sequence where everything is already done and flash is mounted. The /etc/profile, launched by the line ::respawn:-/bin/sh of /etc/inittab, calls /bin/cfm, which takes care of all the ADSL settings and mounting the flash - but it stays loaded and the script does not go on. So after several retries I found a solution - added another line to the /etc/inittab:

$ diff -u inittab.old inittab
--- inittab.old 2008-07-03 00:31:44.000000000 +0200
+++ inittab     2008-07-03 00:31:52.000000000 +0200
@@ -1,5 +1,6 @@
 ::sysinit:/etc/init.d/rcS
 ::respawn:-/bin/sh
+::respawn:-/bin/myscript.sh
 tty2::askfirst:-/bin/sh
 ::ctrlaltdel:/bin/umount -a -r

that will launch a script that waits until the flash drive is mounted and then passes the command to script residing on the flash:

$ cat bin/myscript.sh
#!/bin/sh
while [ "`echo /var/usb/usb_*`" = "/var/usb/usb_*" ]; do
  ping 127.0.0.1 # the shell does not have the sleep builtin, hence using ping as a substitute
done

ping -c 20 127.0.0.1 # wait 20 seconds after the usb is mounted just to be sure load is not too high

/bin/sh /var/usb/usb_2/myscript.sh > /var/usb/usb_2/myscript.log 2>&1

Then I created the new image:

# /home/hajma/Dokumenty/GPL_AM604G_ForRussia/GENERIC_6348_WLAN_A34_3-06-02-01/hostTools/mksquashfs /tmp/fs/var/usb/a rootfs.img -noappend -be -lzma -no-fragments -noI 

Combined with the rest of the lot:

# /home/hajma/Dokumenty/GPL_AM604G_ForRussia/GENERIC_6348_WLAN_A34_3-06-02-01/hostTools/bcmImageBuilder --output bcmimage --chip 6348 --board 96348GW-11 --blocksize 64 --cfefile cfe --rootfsfile rootfs.img --kernelfile kernel --include-cfe

and flashed. Works like a charm. I put the firmware to the OpenWRT page on AM-200g. direct link

In the next part I'll try to fix the dhcp.

FYI, this is the content of the /var/usb/usb_2/myscript.sh file, launching the jabber daemon:

#!/bin/sh
/bin/iptables -I INPUT -p tcp -i ppp_8_48_1 --dport 5222 -m state --state NEW -j ACCEPT
/bin/iptables -I INPUT -p tcp -i ppp_8_48_1 --dport 5223 -m state --state NEW -j ACCEPT
/bin/iptables -I INPUT -p tcp -i ppp_8_48_1 --dport 5269 -m state --state NEW -j ACCEPT
/var/usb/usb_1/bin/jabberd -c /var/usb/usb_2/jabberd/jabber.xml

FYI, the steps to cross-compile netcat and coreutils (for the tar and dd commands):

$ wget ftp://heanet.dl.sourceforge.net/n/ne/netcat/netcat-0.7.1.tar.bz2
$ tar xvf netcat-0.7.1.tar.bz2
$ cd netcat-0.7.1
$ ./configure --build=i686-pc-linux-gnu --host=mipsel-linux --target=mipsel-linux --prefix=/var/usb/usb_1
$ make
$ make install

$ cd ..
$ wget http://ftp.gnu.org/gnu/coreutils/coreutils-6.9.tar.bz2
$ tar xvf coreutils-6.9.tar.bz2 
$ cd coreutils-6.9
$ ./configure --build=i686-pc-linux-gnu --host=mipsel-linux --target=mipsel-linux --prefix=/var/usb/usb_1
$ make
$ make install

Customizing Asus AM200g - IV. jabberd configuration
Customizing Asus AM200g - III. jabberd and vim
Customizing Asus AM200g - II. Preparing the environment
Customizing Asus AM200g - I. PREPARATION
       

Komentáře

[Viliam Púčik]

Pisat blog na ceskom serveri v anglictine so "zdovodnenim", ze tak bude tvoj clanok pristupnejsi dalsim necesky(/neslovensky) hovoriacim ludom podla mna neobstoji.

Bud:
a) Si to mohol uploadnut na nejaky english server/wiki ...
b) ... a cesky "original" blognut na abclinuxu
c) ... zvysovat si ego mozes aj inym sposobom

Inak nemam, co sa tyka obsahu tvojho clanku, vyhrady.

[hajma]

0. Nevím jestli jsi to zaregistroval, ale na internetu je, od doby co před patnácti lety vymysleli vyhledávací servery, srdečně jedno na jakém serveru se jaký článek povaluje.
1. Proč bych to měl uploadovat někam jinam? Žádný jiný server mi neposkytne tolik komfortu jako abclinuxu.cz, i když je pravda, že o linuxu už toho tady není tolik, co bejvalo. Jestli ti vadím, tak si trhni.
2. Kvůli tobě to budu dělat dvakrát, si uhádl...
3. To každopádně můžu. Nicméně mi osobně nepřipadá, že bych si zrovna tímhle honil ego. Pokud ti přijde, že zápiskem v angličtině si někdo může honit ego, měl bys asi ve škole začít víc makat.

[Viliam Púčik]

Vobec som nemal zaujem ta urazat a ospravedlnujem sa, ak to tak vyznelo. Samozrejme, ze mozes pisat v akomkolvek jazyku kdekolvek chces.

0. Z pohladu vyhladavaca - ano, z pohladu citatela - da sa polemizovat (napr.: ceskych clankov na linux.com nie je az tak vela)

1.0 Islo len o radu, tvoj english clanok by sa skor hodil na medzinarodnejsie orientovany server.

1.1 Nevadis a ani tvoj blog, vadi mi len jeho jazykova mutacia na abclinuxu.cz.

2. ;)

3. ;)

[JRD McLAREN]

No ale aby sme sa trosku pohli... (aj ked po skoro dvoch rokoch) Dodam postrehy, kezde ma zase chytlo upraovat firmware.

Asus vydal source code k AM-200g a je k stiahnutiu tu: http://support.asus.com/download/download.aspx?product=7&model=AM200g&SLanguage=en-us&os=8

samozrejme aj nove k AM-604g su pouzitelne...

Co sa tyka vytvarania a skusania "vlastnorucne vyrobeneho firmware", je ovela bezpecnejsie aj ked mozno viac pracnejsie, skusat to bez nutnosti flashovania do routru. Treba k tomu TFTP server a SERIOVY KABEL. Bootloader routra vie natiahnut image fs priamo z tftp.

[jege]

Полный дебил - написать статью английски на чешском сервере.