Re: [PATCH] f2fs: Add ckpt_thread_task sysfs node

Re: [PATCH] f2fs: Add ckpt_thread_task sysfs node

[zhaoyuenan]

Hi Chao,

On 5/22/26 10:50, Chao Yu wrote:
> Hi Yuenan,
> 
> Thanks for your contribution!
> 
> If there is a method to get pid via userspace, I prefer to do it in userspace,
> rather than implementing and maintaining an duplicated one in kernel.
> 
> Or do you have a strong reason to do this in kernel?

The main reason is that retrieving this PID from userspace is currently
quite complex and fragile.

Previously, userspace tools had to rely on pipelines like:
pgrep f2fs_ckpt-$(ls -l /dev/block/by-name/userdata | awk '{print $5$6}' | tr ',' ':')

This approach has a few drawbacks:
1. It requires parsing major/minor numbers and executing multiple commands,
   which is inefficient for simple monitoring tools.
2. On some older kernels, the thread name can be truncated due to TASK_COMM_LEN
   limitations, causing 'pgrep' to fail to match the full 'f2fs_ckpt-X:Y' string.

Exposing the PID via a sysfs node provides a deterministic, lightweight,
and reliable way (just a simple 'cat') to fetch it without any userspace
guessing or complex parsing.

Does this address your concern?

Thanks,
Yuenan

Re: [f2fs-dev] [PATCH] f2fs: Add ckpt_thread_task sysfs node

[zhaoyuenan]

Hi Chao,

On 5/22/26 10:50, Chao Yu wrote:
> Hi Yuenan,
> 
> Thanks for your contribution!
> 
> If there is a method to get pid via userspace, I prefer to do it in userspace,
> rather than implementing and maintaining an duplicated one in kernel.
> 
> Or do you have a strong reason to do this in kernel?

The main reason is that retrieving this PID from userspace is currently
quite complex and fragile.

Previously, userspace tools had to rely on pipelines like:
pgrep f2fs_ckpt-$(ls -l /dev/block/by-name/userdata | awk '{print $5$6}' | tr ',' ':')

This approach has a few drawbacks:
1. It requires parsing major/minor numbers and executing multiple commands,
   which is inefficient for simple monitoring tools.
2. On some older kernels, the thread name can be truncated due to TASK_COMM_LEN
   limitations, causing 'pgrep' to fail to match the full 'f2fs_ckpt-X:Y' string.

Exposing the PID via a sysfs node provides a deterministic, lightweight,
and reliable way (just a simple 'cat') to fetch it without any userspace
guessing or complex parsing.

Does this address your concern?

Thanks,
Yuenan


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Re: [PATCH] f2fs: Add ckpt_thread_task sysfs node

[Chao Yu]

Hi Yuenan,

On 5/22/26 21:56, zhaoyuenan wrote:
> Hi Chao,
> 
> On 5/22/26 10:50, Chao Yu wrote:
>> Hi Yuenan,
>>
>> Thanks for your contribution!
>>
>> If there is a method to get pid via userspace, I prefer to do it in userspace,
>> rather than implementing and maintaining an duplicated one in kernel.
>>
>> Or do you have a strong reason to do this in kernel?
> 
> The main reason is that retrieving this PID from userspace is currently
> quite complex and fragile.
> 
> Previously, userspace tools had to rely on pipelines like:
> pgrep f2fs_ckpt-$(ls -l /dev/block/by-name/userdata | awk '{print $5$6}' | tr ',' ':')
> 
> This approach has a few drawbacks:
> 1. It requires parsing major/minor numbers and executing multiple commands,
>    which is inefficient for simple monitoring tools.

I don't see why there is any performance concern in your scenario.

> 2. On some older kernels, the thread name can be truncated due to TASK_COMM_LEN
>    limitations, causing 'pgrep' to fail to match the full 'f2fs_ckpt-X:Y' string.

I don't think this method has scalability, is it better to backport below patch? so
that you can trace all kthreads and set them to target cgroup.

https://lore.kernel.org/all/20211120112850.46047-1-laoar.shao@gmail.com

Thanks,

> 
> Exposing the PID via a sysfs node provides a deterministic, lightweight,
> and reliable way (just a simple 'cat') to fetch it without any userspace
> guessing or complex parsing.
> 
> Does this address your concern?
> 
> Thanks,
> Yuenan

Re: [f2fs-dev] [PATCH] f2fs: Add ckpt_thread_task sysfs node

[Chao Yu via Linux-f2fs-devel]

Hi Yuenan,

On 5/22/26 21:56, zhaoyuenan wrote:
> Hi Chao,
> 
> On 5/22/26 10:50, Chao Yu wrote:
>> Hi Yuenan,
>>
>> Thanks for your contribution!
>>
>> If there is a method to get pid via userspace, I prefer to do it in userspace,
>> rather than implementing and maintaining an duplicated one in kernel.
>>
>> Or do you have a strong reason to do this in kernel?
> 
> The main reason is that retrieving this PID from userspace is currently
> quite complex and fragile.
> 
> Previously, userspace tools had to rely on pipelines like:
> pgrep f2fs_ckpt-$(ls -l /dev/block/by-name/userdata | awk '{print $5$6}' | tr ',' ':')
> 
> This approach has a few drawbacks:
> 1. It requires parsing major/minor numbers and executing multiple commands,
>    which is inefficient for simple monitoring tools.

I don't see why there is any performance concern in your scenario.

> 2. On some older kernels, the thread name can be truncated due to TASK_COMM_LEN
>    limitations, causing 'pgrep' to fail to match the full 'f2fs_ckpt-X:Y' string.

I don't think this method has scalability, is it better to backport below patch? so
that you can trace all kthreads and set them to target cgroup.

https://lore.kernel.org/all/20211120112850.46047-1-laoar.shao@gmail.com

Thanks,

> 
> Exposing the PID via a sysfs node provides a deterministic, lightweight,
> and reliable way (just a simple 'cat') to fetch it without any userspace
> guessing or complex parsing.
> 
> Does this address your concern?
> 
> Thanks,
> Yuenan



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel