Yes, /proc/[pid] handling is diferent and your solution is perfect just for
the other proc directories.

Informations under [pid] directories are about tasks (processes), and most
of this information is represented by task_struct structure (defined
in include/linux/sched.h).

I think one easy way to solve this problem is:

In include/linux/sched.h file
1 - add a new element in task_struct structure.
struct task_struct {
int my_value;
        .......

In  fs/proc/base.c file
1 - Create a new entry in static pid_entry tgid_base_stuff[] array.
e.g.  ONE("MY_FILE", S_IRUSR, proc_pid_my_file),

2 - Create a callback function that will be invoked when the new proc entry
is accessed (*task_struct is passed to this function).
static int proc_pid_my_file(struct seq_file *m, struct pid_namespace *ns,
 struct pid *pid, struct task_struct *task)
{
seq_printf(m, "%s %d\n", "here we go....", task->my_value);
 return 0;
}


Use the task_struct as usual:
struct task_struct *task;
 task = pid_task(find_vpid(1), PIDTYPE_PID);


I don't know if it is the best solution, but it seems it worked.
If anyone know a more easy or correct solution please let me know.

Thanks again,
Mauro Romano Trajber

On Wed, Jan 12, 2011 at 7:41 AM, Rajat Sharma [off-list ref] wrote:

yes you are right, it gives NULL for <pid> directories under proc but
I tried changing pid to something else, say /proc/sys and it works.
Looks like handling of pid directories is entirely different. I didn't
get time to explore on that, will have to dig more into this.

Rajat

On Wed, Jan 12, 2011 at 12:50 AM, Mauro Romano Trajber
[off-list ref] wrote:

quoted

Following your recommendations parent->pde always returns NULL. You know

why

quoted

?
// code...
err = kern_path("/proc/1/", LOOKUP_FOLLOW, &path);
if (err) {
return err;
}
struct proc_inode *parent = PROC_I(path.dentry->d_inode);
struct proc_dir_entry *parent_dir = parent->pde;
if (parent_dir == NULL) {
printk("parent_dir is NULL\n");
} else {
create_proc_entry("SOMEFile", 0644, parent_dir);
}

Mauro!
On Tue, Jan 11, 2011 at 4:39 AM, Rajat Sharma [off-list ref]

wrote:

quoted

quoted

Try this:
1. do a path_lookup for parent proc dir e.g. /proc/1234 and get its

inode.

quoted

quoted

2. get proc_inode structure for parent from vfs inode like this:
         sruct proc_inode *parent = PROC_I(inode).
   PROC_I is defined in proc_fs.h
3. get parent proc_dir_entry object:
         struct proc_dir_entry *parent_dir = parent->pde;
4. now you can call:
         create_proc_entry("SOME file", 0644, parent_dir);
5. or you can create a directory if you want:
         proc_mkdir("your dir", parent_dir);

hope this helps.

Rajat

On Tue, Jan 11, 2011 at 12:19 AM, Mauro Romano Trajber
[off-list ref] wrote:

quoted

I think I found:
static const struct pid_entry tgid_base_stuff[] at fs/proc/base.c has
all
/proc/[pid] entries.
But unfortunately it does not use create_proc_entry function, and I'm
trying
to create a new syscall that creates a new proc_entry for the caller
process.
Adding a new element in tgid_base_stuff[] makes the things more
complicated
than simply call a create_proc_entry function.
Is there another way to do it ?
Mauro Romano Trajber

On Mon, Jan 10, 2011 at 3:18 PM, Mauro Romano Trajber
[off-list ref]
wrote:

quoted

How can I create a new proc entry under /proc/[pid] ?
I using create_proc_entry("SOME_FILE", 0644, NULL /* here goes the

pid

quoted

quoted

quoted

quoted

proc entry */);
Is there any way to get PID directory as a parent proc entry ? How ?
Thanks,
Mauro

_______________________________________________
Kernelnewbies mailing list
Kernelnewbies at kernelnewbies.org
http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20110112/3722a578/attachment.html