Doing a zero-copy move of data from a kernel buffer to hard disk

Doing a zero-copy move of data from a kernel buffer to hard disk

[Neeraj Kumar]

I am trying to move data from a buffer in kernel space into the hard
disk without having to incur any additional copies from kernel buffer to
user buffers or any other kernel buffers. Any ideas/suggestions would be
most helpful.

The use case is basically a demux driver which collects data into a
demux buffer in kernel space and this buffer has to be emptied
periodically by copying the contents into a FUSE-based partition on the
disk. As the buffer gets full, a user process is signalled which then
determines the sector numbers on the disk the contents need to be copied
to.

I was hoping to mmap the above demux kernel buffer into user address
space and issue a write system call to the raw partition device. But
from what I can see, the this data is being cached by the kernel on its
way to the Hard Disk driver. And so I am assuming that involves
additional copies by the linux kernel.

At this point I am wondering if there is any other mechansim to do this
without involving additional copies by the kernel. I realize this is an
unsual usage scenario for non-embedded environments, but I would
appreciate any feedback on possible options.

BTW - I have tried using O_DIRECT when opening the raw partition, but
the subsequent write call fails if the buffer being passed is the
mmapped buffer.


Thanx!



      

Re: Doing a zero-copy move of data from a kernel buffer to hard disk

[Robert Hancock]

On 09/20/2010 02:12 PM, Neeraj Kumar wrote:
> I am trying to move data from a buffer in kernel space into the hard
> disk without having to incur any additional copies from kernel buffer to
> user buffers or any other kernel buffers. Any ideas/suggestions would be
> most helpful.
>
> The use case is basically a demux driver which collects data into a
> demux buffer in kernel space and this buffer has to be emptied
> periodically by copying the contents into a FUSE-based partition on the
> disk. As the buffer gets full, a user process is signalled which then
> determines the sector numbers on the disk the contents need to be copied
> to.
>
> I was hoping to mmap the above demux kernel buffer into user address
> space and issue a write system call to the raw partition device. But
> from what I can see, the this data is being cached by the kernel on its
> way to the Hard Disk driver. And so I am assuming that involves
> additional copies by the linux kernel.
>
> At this point I am wondering if there is any other mechansim to do this
> without involving additional copies by the kernel. I realize this is an
> unsual usage scenario for non-embedded environments, but I would
> appreciate any feedback on possible options.
>
> BTW - I have tried using O_DIRECT when opening the raw partition, but
> the subsequent write call fails if the buffer being passed is the
> mmapped buffer.

O_DIRECT would be the right way to do it - I think you'd need to figure 
out why that write is failing. Keep in mind you need to observe the 
alignment restrictions of O_DIRECT (man 2 open).

Re: Doing a zero-copy move of data from a kernel buffer to hard disk

[Neeraj Kumar]

>On 09/20/2010 02:12 PM, Neeraj Kumar wrote:
>> I am trying to move data from a buffer in kernel space into the hard
>> disk without having to incur any additional copies from kernel buffer to
> >user buffers or any other kernel buffers. Any ideas/suggestions would be
> >most helpful.
>>
>>The use case is basically a demux driver which collects data into a
>> demux buffer in kernel space and this buffer has to be emptied
>>periodically by copying the contents into a FUSE-based partition on the
> disk. As the buffer gets full, a user process is signalled which then
>> determines the sector numbers on the disk the contents need to be copied
>> to.
>>
>> I was hoping to mmap the above demux kernel buffer into user address
>> space and issue a write system call to the raw partition device. But
>> from what I can see, the this data is being cached by the kernel on its
>> way to the Hard Disk driver. And so I am assuming that involves
>> additional copies by the linux kernel.
>>
>> At this point I am wondering if there is any other mechansim to do this
>> without involving additional copies by the kernel. I realize this is an
>> unsual usage scenario for non-embedded environments, but I would
>> appreciate any feedback on possible options.
>>
>> BTW - I have tried using O_DIRECT when opening the raw partition, but
>> the subsequent write call fails if the buffer being passed is the
>> mmapped buffer.

>O_DIRECT would be the right way to do it - I think you'd need to figure 
>out why that write is failing. Keep in mind you need to observe the 
>alignment restrictions of O_DIRECT (man 2 open).


I was able to figure out that O_DIRECT can not work with memory mapped 
pages. In the kernel, get_user_pages() returns error if user provided pages are 
mapped kernel pages.

Thanks
Neeraj



      

Re: Doing a zero-copy move of data from a kernel buffer to hard disk

[Miklos Szeredi]

On Mon, 20 Sep 2010, Neeraj Kumar wrote:
> I am trying to move data from a buffer in kernel space into the hard
> disk without having to incur any additional copies from kernel buffer to
> user buffers or any other kernel buffers. Any ideas/suggestions would be
> most helpful.
> 
> The use case is basically a demux driver which collects data into a
> demux buffer in kernel space and this buffer has to be emptied
> periodically by copying the contents into a FUSE-based partition on the
> disk. As the buffer gets full, a user process is signalled which then
> determines the sector numbers on the disk the contents need to be copied
> to.
> 
> I was hoping to mmap the above demux kernel buffer into user address
> space and issue a write system call to the raw partition device. But
> from what I can see, the this data is being cached by the kernel on its
> way to the Hard Disk driver. And so I am assuming that involves
> additional copies by the linux kernel.

Isn't this functionality already served by relayfs?

> At this point I am wondering if there is any other mechansim to do this
> without involving additional copies by the kernel. I realize this is an
> unsual usage scenario for non-embedded environments, but I would
> appreciate any feedback on possible options.
> 
> BTW - I have tried using O_DIRECT when opening the raw partition, but
> the subsequent write call fails if the buffer being passed is the
> mmapped buffer.

The other zero-copy  mechanism is splice(2).

Thanks,
Miklos

Re: Doing a zero-copy move of data from a kernel buffer to hard disk

[Neeraj Kumar]

On Mon, 20 Sep 2010, Neeraj Kumar wrote:
 >> I am trying to move data from a buffer in kernel space into the hard
 >> disk without having to incur any additional copies from kernel buffer to
 >> user buffers or any other kernel buffers. Any ideas/suggestions would be
 >> most helpful.
 >> 
 >> The use case is basically a demux driver which collects data into a
 >> demux buffer in kernel space and this buffer has to be emptied
 >> periodically by copying the contents into a FUSE-based partition on the
 >> disk. As the buffer gets full, a user process is signalled which then
 >> determines the sector numbers on the disk the contents need to be copied
 >> to.
 >> 
 >> I was hoping to mmap the above demux kernel buffer into user address
 >> space and issue a write system call to the raw partition device. But
 >> from what I can see, the this data is being cached by the kernel on its
 >> way to the Hard Disk driver. And so I am assuming that involves
 >> additional copies by the linux kernel.

>Isn't this functionality already served by relayfs?

I could not figure out how relayfs could help me.

 >> At this point I am wondering if there is any other mechansim to do this
 >> without involving additional copies by the kernel. I realize this is an
 >> unsual usage scenario for non-embedded environments, but I would
 >> appreciate any feedback on possible options.
 >> 
 >> BTW - I have tried using O_DIRECT when opening the raw partition, but
 >> the subsequent write call fails if the buffer being passed is the
 >> mmapped buffer.

>The other zero-copy  mechanism is splice(2).

I looked at the implementation of splice(). It looks promising. I have 2 more 
questions related to it:
1. SPLICE_F_MOVE which is needed for zero copy movement of pages is defined as 
no op in the kernel code.
    Does splice still tries for zero copy movement?

2. I may need to implement splice_read() and splice_write() for the demux 
driver.  I did not find any other
 driver making use of these interfaces in the kernel except file system. Is 
there any sample code available
which could help me to understand/implement splice() interfaces in a better way?


Thanks,
Miklos