[Q][SOLVED]Writing into ramdisk from init.rc - Android Q&A, Help & Troubleshooting

EDIT: The solution is in post #6.
Hi,
I'm trying to make a kernel modification which would create a file on boot before any partition is mounted. I wrote a simple shell script which tries to create a file in the root directory. The problem is I can't write any file to the / directory from a program or a script called by exec from init.rc. If I run dmesg, I can't find any indication of a problem there. I tried to modify the script to write to the /data directory on post-fs, and that worked, but that is not what I want.
I can't find out where the problem is. Do you have any ideas? Thanks.
simplified init.rc snippet:
Code:
on fs
# some work
exec /sbin/busybox sh /test.sh
# some other work
mount_all /fstab.grouper
# and another work
/test.sh
Code:
#!/sbin/busybox sh
echo "test" > /test

---

/ mounts as ro pretty early...
try to remount it rw before making files, or change the part of init.rc that mounts it ro.

---

frantisek.nesveda said:
Hi,
I'm trying to make a kernel modification which would create a file on boot before any partition is mounted. I wrote a simple shell script which tries to create a file in the root directory. The problem is I can't write any file to the / directory from a program or a script called by exec from init.rc. If I run dmesg, I can't find any indication of a problem there. I tried to modify the script to write to the /data directory on post-fs, and that worked, but that is not what I want.
I can't find out where the problem is. Do you have any ideas? Thanks.
simplified init.rc snippet:
Code:
on fs
# some work
exec /sbin/busybox sh /test.sh
# some other work
mount_all /fstab.grouper
# and another work
/test.sh
Code:
#!/sbin/busybox sh
echo "test" > /test
Click to expand...
Click to collapse
I am having the same problem, my script is not called. Did you find the solution?

---

Khaon said:
I am having the same problem, my script is not called. Did you find the solution?
Click to expand...
Click to collapse
Not yet. But recently I got back to the project where I would use it, so maybe I can take a look again and see if I am any wiser now. I'll try to solve it this evening or tomorrow, I'll be reinstalling my tablet so I'll have a good oportunity. I'll post the results here.

---

frantisek.nesveda said:
Not yet. But recently I got back to the project where I would use it, so maybe I can take a look again and see if I am any wiser now. I'll try to solve it this evening or tomorrow, I'll be reinstalling my tablet so I'll have a good oportunity. I'll post the results here.
Click to expand...
Click to collapse
That would be great, the only thing I could see is upon booting its permissions are read-only. And chmod 0777 doesn't change its permissions(rootfs is rw)

---

Khaon said:
That would be great, the only thing I could see is upon booting its permissions are read-only. And chmod 0777 doesn't change its permissions(rootfs is rw)
Click to expand...
Click to collapse
OK, I found the cause of the problem and the solution.
The implementation of the exec command in Android init scripts is missing from the init binary by default - the source looks like this:
Code:
int do_exec(int nargs, char **args)
{
return -1;
}
So there are two options - either write the do_exec function yourself, or use a trick to execute your programs.
Writing it yourself would be probably cleaner, but it would mean compiling the init binary again and not everyone wants to do that, so I'll be using the trick.
The trick is to register your program as a service and then invoke it via the start command.
Example init.rc:
Code:
# service invocation - where you need it
# warning - this will not wait for the finish of the program
start my_awesome_program
.
.
.
# service declaration - along with other declarations at the end of the file
service my_awesome_program /sbin/my_awesome_program
disabled
oneshot
If you want to implement the do_exec function and use the exec command, you can use for example CyanogenMod's implementation.
There may be a third option, but I haven't tested that - you can copy the init binary from a kernel which has the exec command working - I suppose all CM kernels should, but I haven't tried.
It's funny how when you already know the solution, you can find so much info, but when you don't, nothing comes up since you are searching wrong.

---

frantisek.nesveda said:
OK, I found the cause of the problem and the solution.
The implementation of the exec command in Android init scripts is missing from the init binary by default - the source looks like this:
Code:
int do_exec(int nargs, char **args)
{
return -1;
}
So there are two options - either write the do_exec function yourself, or use a trick to execute your programs.
Writing it yourself would be probably cleaner, but it would mean compiling the init binary again and not everyone wants to do that, so I'll be using the trick.
The trick is to register your program as a service and then invoke it via the start command.
Example init.rc:
Code:
# service invocation - where you need it
start my_awesome_program
.
.
.
# service declaration - along with other declarations at the end of the file
service my_awesome_program /sbin/my_awesome_program
disabled
oneshot
If you want to implement the do_exec function and use the exec command, you can use for example CyanogenMod's implementation.
There may be a third option, but I haven't tested that - you can copy the init binary from a kernel which has the exec command working - I suppose all CM kernels should, but I haven't tried.
It's funny how when you already know the solution, you can find so much info, but when you don't, nothing comes up since you are searching wrong.
Click to expand...
Click to collapse
Oh my, many thanks! So stupid exec always return ;D, wouldn't have thought one sec on checking the implementation of those functions.

---

Khaon said:
Oh my, many thanks! So stupid exec always return ;D, wouldn't have thought one sec on checking the implementation of those functions.
Click to expand...
Click to collapse
A small warning - starting the program as a service will not wait for the program finish, so its results will not be available immediately after the start call. I learnt this the hard way after a few hours of debugging just now.
I added this info to my previous post too.

---

frantisek.nesveda said:
A small warning - starting the program as a service will not wait for the program finish, so its results will not be available immediately after the start call. I learnt this the hard way after a few hours of debugging just now.
I added this info to my previous post too.
Click to expand...
Click to collapse
I always use
Code:
class core
user root
in addition to oneshot.
Haven't you com accross SElinux restrictions using busybox? Because I do I need to set permissions.
I have this issue