[PATCH] per process I/O statistics for userspace

[PATCH] per process I/O statistics for userspace

[Levent Serinol]

with following patch, userspace processes/utilities will be able to
access per process I/O statistics. for example, top like utilites can
use this information.


--- linux-2.6.13/fs/proc/array.c.org    2005-08-29 02:41:01.000000000 +0300
+++ linux-2.6.13/fs/proc/array.c        2005-09-12 10:22:55.000000000 +0300
@@ -408,7 +408,7 @@ static int do_task_stat(struct task_stru

        res = sprintf(buffer,"%d (%s) %c %d %d %d %d %d %lu %lu \
 %lu %lu %lu %lu %lu %ld %ld %ld %ld %d %ld %llu %lu %ld %lu %lu %lu %lu %lu \
-%lu %lu %lu %lu %lu %lu %lu %lu %d %d %lu %lu\n",
+%lu %lu %lu %lu %lu %lu %lu %lu %d %d %lu %lu %llu %llu\n",
                task->pid,
                tcomm,
                state,
@@ -453,7 +453,9 @@ static int do_task_stat(struct task_stru
                task->exit_signal,
                task_cpu(task),
                task->rt_priority,
-               task->policy);
+               task->policy,
+               task->rchar,
+               task->wchar);
        if(mm)
                mmput(mm);
        return res;
--
Signed-off-by: Levent Serinol <lserinol@gmail.com>

Re: [PATCH] per process I/O statistics for userspace

[Pavel Machek]

Hi!

> with following patch, userspace processes/utilities will be able to
> access per process I/O statistics. for example, top like utilites can
> use this information.

Nice, but should not this perhaps go into the other file? Adding more
integers into long line does not seem nice...
									Pavel

> 
> --- linux-2.6.13/fs/proc/array.c.org    2005-08-29 02:41:01.000000000 +0300
> +++ linux-2.6.13/fs/proc/array.c        2005-09-12 10:22:55.000000000 +0300
> @@ -408,7 +408,7 @@ static int do_task_stat(struct task_stru
> 
>         res = sprintf(buffer,"%d (%s) %c %d %d %d %d %d %lu %lu \
>  %lu %lu %lu %lu %lu %ld %ld %ld %ld %d %ld %llu %lu %ld %lu %lu %lu %lu %lu \
> -%lu %lu %lu %lu %lu %lu %lu %lu %d %d %lu %lu\n",
> +%lu %lu %lu %lu %lu %lu %lu %lu %d %d %lu %lu %llu %llu\n",
>                 task->pid,
>                 tcomm,
>                 state,
> @@ -453,7 +453,9 @@ static int do_task_stat(struct task_stru
>                 task->exit_signal,
>                 task_cpu(task),
>                 task->rt_priority,
> -               task->policy);
> +               task->policy,
> +               task->rchar,
> +               task->wchar);
>         if(mm)
>                 mmput(mm);
>         return res;
> --
> Signed-off-by: Levent Serinol <lserinol@gmail.com>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

-- 
if you have sharp zaurus hardware you don't need... you know my address

Re: [PATCH] per process I/O statistics for userspace

[Levent Serinol]

Hi,

 You're right. Also, this information differs from other ones. So,
putting it into another file makes sense.
  Here is the better one which puts statistics in a file called iostat.

================================================================
--- linux-2.6.13/fs/proc/array.c.org    2005-08-29 02:41:01.000000000 +0300
+++ linux-2.6.13/fs/proc/array.c        2005-09-14 22:05:16.000000000 +0300
@@ -482,3 +482,8 @@ int proc_pid_statm(struct task_struct *t
        return sprintf(buffer,"%d %d %d %d %d %d %d\n",
                       size, resident, shared, text, lib, data, 0);
 }
+
+int proc_pid_iostat(struct task_struct *task, char *buffer)
+{
+       return sprintf(buffer,"%llu %llu\n",task->rchar,task->wchar);
+}
--- linux-2.6.13/fs/proc/base.c.org     2005-08-29 02:41:01.000000000 +0300
+++ linux-2.6.13/fs/proc/base.c 2005-09-14 22:04:03.000000000 +0300
@@ -86,6 +86,7 @@ enum pid_directory_inos {
        PROC_TGID_FD_DIR,
        PROC_TGID_OOM_SCORE,
        PROC_TGID_OOM_ADJUST,
+       PROC_TGID_IOSTAT,
        PROC_TID_INO,
        PROC_TID_STATUS,
        PROC_TID_MEM,
@@ -123,6 +124,7 @@ enum pid_directory_inos {
        PROC_TID_FD_DIR = 0x8000,       /* 0x8000-0xffff */
        PROC_TID_OOM_SCORE,
        PROC_TID_OOM_ADJUST,
+       PROC_TID_IOSTAT,
 };

  struct pid_entry {
@@ -169,6 +171,7 @@ static struct pid_entry tgid_base_stuff[
  #ifdef CONFIG_AUDITSYSCALL
        E(PROC_TGID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
  #endif
+       E(PROC_TGID_IOSTAT,    "iostat",   S_IFREG|S_IRUGO),
        {0,0,NULL,0}
 };
  static struct pid_entry tid_base_stuff[] = {
@@ -205,6 +208,7 @@ static struct pid_entry tid_base_stuff[]
  #ifdef CONFIG_AUDITSYSCALL
        E(PROC_TID_LOGINUID, "loginuid", S_IFREG|S_IWUSR|S_IRUGO),
  #endif
+       E(PROC_TID_IOSTAT,     "iostat",   S_IFREG|S_IRUGO),
        {0,0,NULL,0}
 };

@@ -1596,6 +1600,11 @@ static struct dentry *proc_pident_lookup
                        inode->i_fop = &proc_loginuid_operations;
                        break;
  #endif
+               case PROC_TID_IOSTAT:
+               case PROC_TGID_IOSTAT:
+                       inode->i_fop = &proc_info_file_operations;
+                       ei->op.proc_read = proc_pid_iostat;
+                       break;
                default:
                        printk("procfs: impossible type (%d)",p->type);
                        iput(inode);
--- linux-2.6.13/fs/proc/internal.h.org 2005-09-14 22:07:07.000000000 +0300
+++ linux-2.6.13/fs/proc/internal.h     2005-09-14 22:06:41.000000000 +0300
@@ -36,6 +36,7 @@ extern int proc_tid_stat(struct task_str
  extern int proc_tgid_stat(struct task_struct *, char *);
  extern int proc_pid_status(struct task_struct *, char *);
  extern int proc_pid_statm(struct task_struct *, char *);
+extern int proc_pid_iostat(struct task_struct *, char *);

  static inline struct task_struct *proc_task(struct inode *inode)
 {
=================================================================

On 9/14/05, Pavel Machek <pavel@ucw.cz> wrote:
> Hi!
> 
> > with following patch, userspace processes/utilities will be able to
> > access per process I/O statistics. for example, top like utilites can
> > use this information.
> 
> Nice, but should not this perhaps go into the other file? Adding more
> integers into long line does not seem nice...
>                                                                         Pavel
> 
> >
> > --- linux-2.6.13/fs/proc/array.c.org    2005-08-29 02:41:01.000000000 +0300
> > +++ linux-2.6.13/fs/proc/array.c        2005-09-12 10:22:55.000000000 +0300
> > @@ -408,7 +408,7 @@ static int do_task_stat(struct task_stru
> >
> >         res = sprintf(buffer,"%d (%s) %c %d %d %d %d %d %lu %lu \
> >  %lu %lu %lu %lu %lu %ld %ld %ld %ld %d %ld %llu %lu %ld %lu %lu %lu %lu %lu \
> > -%lu %lu %lu %lu %lu %lu %lu %lu %d %d %lu %lu\n",
> > +%lu %lu %lu %lu %lu %lu %lu %lu %d %d %lu %lu %llu %llu\n",
> >                 task->pid,
> >                 tcomm,
> >                 state,
> > @@ -453,7 +453,9 @@ static int do_task_stat(struct task_stru
> >                 task->exit_signal,
> >                 task_cpu(task),
> >                 task->rt_priority,
> > -               task->policy);
> > +               task->policy,
> > +               task->rchar,
> > +               task->wchar);
> >         if(mm)
> >                 mmput(mm);
> >         return res;
> > --
> > Signed-off-by: Levent Serinol <lserinol@gmail.com>
> > -
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at  http://www.tux.org/lkml/
> 
> --
> if you have sharp zaurus hardware you don't need... you know my address
> 


-- 

Stay out of the road, if you want to grow old. 
~ Pink Floyd ~.

Re: [PATCH] per process I/O statistics for userspace

[Andrew Morton]

Levent Serinol <lserinol@gmail.com> wrote:
>
> +int proc_pid_iostat(struct task_struct *task, char *buffer)
>  +{
>  +       return sprintf(buffer,"%llu %llu\n",task->rchar,task->wchar);
>  +}

Those fields have been sitting there unused for months.  I'd like to hear
what the system accounting guys are intending to do with them before
proceeding, please.

Re: [PATCH] per process I/O statistics for userspace

[Jay Lan]

Andrew,

These data are used in CSA Accounting today and i believe ELSA
intends to use them also.

Since PAGG is not in the tree, SGI have been offering our customers
PAGG/JOB/CSA kernel code as either kernel patches at oss.sgi.com
site or out-of-tree kernel modules today.

SGI intends to submit another solution for PAGG/JOB soon.

Once the future of PAGG/JOB becomes clear, we will know better how
to integrate CSA and BSD accounting. We are not sitting on it.

Thanks,
  - jay


Andrew Morton wrote:
> Levent Serinol <lserinol@gmail.com> wrote:
> 
>>+int proc_pid_iostat(struct task_struct *task, char *buffer)
>> +{
>> +       return sprintf(buffer,"%llu %llu\n",task->rchar,task->wchar);
>> +}
> 
> 
> Those fields have been sitting there unused for months.  I'd like to hear
> what the system accounting guys are intending to do with them before
> proceeding, please.
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: [PATCH] per process I/O statistics for userspace

[Levent Serinol]

Hi Andrew,

 What's your last decision about the patch ?

On 9/15/05, Jay Lan <jlan@engr.sgi.com> wrote:
> Andrew,
> 
> These data are used in CSA Accounting today and i believe ELSA
> intends to use them also.
> 
> Since PAGG is not in the tree, SGI have been offering our customers
> PAGG/JOB/CSA kernel code as either kernel patches at oss.sgi.com
> site or out-of-tree kernel modules today.
> 
> SGI intends to submit another solution for PAGG/JOB soon.
> 
> Once the future of PAGG/JOB becomes clear, we will know better how
> to integrate CSA and BSD accounting. We are not sitting on it.
> 
> Thanks,
>   - jay
> 
> 
> Andrew Morton wrote:
> > Levent Serinol <lserinol@gmail.com> wrote:
> >
> >>+int proc_pid_iostat(struct task_struct *task, char *buffer)
> >> +{
> >> +       return sprintf(buffer,"%llu %llu\n",task->rchar,task->wchar);
> >> +}
> >
> >
> > Those fields have been sitting there unused for months.  I'd like to hear
> > what the system accounting guys are intending to do with them before
> > proceeding, please.
> > -
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at  http://www.tux.org/lkml/
> 
> 


-- 

Stay out of the road, if you want to grow old. 
~ Pink Floyd ~.

Re: [PATCH] per process I/O statistics for userspace

[Andrew Morton]

Levent Serinol <lserinol@gmail.com> wrote:
>
> Hi Andrew,
> 

You inspired me.  Please see
http://www.zip.com.au/~akpm/linux/patches/stuff/top-posting.txt

>  What's your last decision about the patch ?

To wait and see what the system accounting guys come up with.  If we export
this info in /proc then it'll need to remain exported for ever.  So if/when
the system accounting people export it by other means, the info will be
duplicated.