xback:

More details as I'm currently digging into this issue:

/sbin/sysupgrade finally makes following call:

ubus call system sysupgrade '{
"prefix": "/tmp/root",
"path": "/tmp/upgrade/5.0.0.17_imx6.ubi",
"command": ". /lib/functions.sh; include /lib/upgrade; do_upgrade_stage2"

When this call is issued and there is no kernel console available, the board reboots immediately.

More details later on .. hopefully ..

xback:

Issue identified:

The wiki page [1] states that the ventana kernel prints can be disabled using following u-boot command:

setenv console
saveenv

In this case, I would expect the following line checking the kernel boot using dmesg:
[ 0.000000] Kernel command line: ubi0:ubi ubi.mtd=2 rootfstype=squashfs,ubifs

But I'm seeing this:
[ 0.000000] Kernel command line: console=,115200 ubi0:ubi ubi.mtd=2 rootfstype=squashfs,ubifs

I've also tried this:

setenv console null
saveenv

Which results in this one:
[ 0.000000] Kernel command line: console=null,115200 ubi0:ubi ubi.mtd=2 rootfstype=squashfs,ubifs

Practically, these faulty strings cause sysupgrade to reboot instantly as soon as the upgrade starts. (100% repro rate)
When removing the faulty console= line from the kernel bootargs, sysupgrade works perfectly again.

I've reported this back to Gateworks Corp.

[1]
http://trac.gateworks.com/wiki/silenceconsole#DisableKernelserialconsole

TB:

I have the same problem, is there any solution?

dpprog:

The problem with this code is:
procd/upgraded/upgraded.c:

static void sysupgrade(char *path, char *command)
{
char *args[] = { "/lib/upgrade/stage2", path, command, NULL };

upgrade_proc.cb = upgrade_proc_cb;
upgrade_proc.pid = fork();
if (upgrade_proc.pid < 0) {
	fprintf(stderr, "Failed to fork sysupgrade\n");
	return;
}

if (!upgrade_proc.pid) {
	/* Child */
	execvp(args[0], args);
	fprintf(stderr, "Failed to exec sysupgrade\n");
	_exit(-1);
}

uloop_process_add(&upgrade_proc);
uloop_run();

}
execvp() returned -1. errno: 2

dpprog:

With the console turned off, the process /sbin/upgraded will start up with a working directory /dev/

TB:

Dmitry, thank you for the information.
If we do not have physical TTYs, we need to enable virtual.
In my case, configuration for the kernel;

CONFIG_CMDLINE="rootfstype=squashfs,jffs2 noinitrd console=tty0 console=null"
CONFIG_TTY=y
CONFIG_VT=y
CONFIG_VT_CONSOLE=y
CONFIG_CONSOLE_TRANSLATIONS=y
CONFIG_VT_HW_CONSOLE_BINDING=n
CONFIG_VGA_CONSOLE=n


After this procedure, the update proceeds correctly.

abatyiev:

Hello everyone,

I've got hit by same bug on ath79. Sysupgrade almost immediately goes into reboot without changing firmware.

Apparently, it is a bug in procd.

There is state.c file with following function:

static void set_stdio(const char* tty)
{
if (chdir("/dev") ||
!freopen(tty, "r", stdin) ||
!freopen(tty, "w", stdout) ||
!freopen(tty, "w", stderr) ||
chdir("/"))
ERROR("failed to set stdio: %m\n");
else
fcntl(STDERR_FILENO, F_SETFL, fcntl(STDERR_FILENO, F_GETFL) | O_NONBLOCK);
}


As you can see, if there is an error, the chdir("/") function is not called. That means, that during sysupgrade the cwd of procd process would be "/dev" (you can check that by "ls -l /proc/1/cwd").

Such arragement breaks /sbin/upgraded process (that is exec'ed into from procd), when it tries to chroot("."), because instead of chrooting into "/", it instead chroots into "/dev" and fails to find/exec "/lib/upgrade/stage2".

Steps to reproduce:

1. /dev/console should be broken (access to it should yield "No such device" error)
2. In logs, there should be "user.err kernel: [ 8.256052] procd: failed to set stdio: No such device"
3. "ls -l /proc/1/cwd" should point to "/dev"
4. "sysupgrade new.bin" should immediately reboot the device without changing the firmware

Quick and dirty fix (i.e. "I want to upgrade my firmware somehow"):

mv /sbin/udevtrigger /sbin/udevtrigger.old
cat > /sbin/udevtrigger <<<EOF
#!/bin/sh

EOF


Proper fix should insert "chdir("/");" into sysupgrade.c file of procd just before chroot.

Too old.