curiosity between bread and UFS_BALLOC on struct buf
Scott Long
scottl at samsco.org
Mon Oct 2 23:51:42 PDT 2006
Ok, that makes more sense. Yes, you need to use getblk(). Once you've
modified the bp->b_data, you need to set appropriate flags on the buf
so the the VM system knows that buffer is dirty and needs to be flushed
to disk. Although B_DIRTY looks like the obvious flag, it apparently
only has meaning to ext2fs. What you need to do is clear B_INVAL if it
is set. You might also need to twiddle B_CACHE, but I can't recall for
sure anymore.
Scott
Shane Adams wrote:
> Hey thanks for the response! I should of mentioned that I am hacking on my own filesystem, using UFS/FFS as a basis. So I'm actually below VOP_*. I did find in my digging throught the code that UFS_BALLOC ultimatly for my code path gets the bufer from a call to getblk.
>
> From the documentation I've read I would think that I could then immediatly bwrite on that buffer, which I attempt, but my change application specific changes to the buffer do not persist. I have a couple of anciallary functions that do the bread/bwrite and those work perfectly.
>
> *shrug*
>
> Hopefully I'll nail it!
>
> Cheers,
> Shane
>
>
> ----- Original Message ----
> From: Scott Long <scottl at samsco.org>
> To: Shane Adams <adamsch1 at yahoo.com>
> Cc: freebsd-fs at freebsd.org
> Sent: Monday, October 2, 2006 8:31:44 PM
> Subject: Re: curiosity between bread and UFS_BALLOC on struct buf
>
>
>
> On Mon, 2 Oct 2006, Shane Adams wrote:
>
>> Hey all,
>
>> I've been puzzling with a bit of code. I'm creating a file from within
>> a kernel module, that works fine, but I've got a question if perhaps I'm
>> missing something:
>>
>> I use UFS_BALLOC to allocate some space. That function has a struct buf
>> thats returned. What I would expect is that I could manipulate the
>> bp->b_data field then do a bwrite and have my changes saved. Doesnt
>> workthat way. Only way I can get it to work is
>>
>> 1 UFS_BALLOC( ... )
>>
>> 2 bwrite( .. )
>>
>> 3 bread( ... )
>>
>> 4 /* Make changes to what I read */
>>
>> 5 bwrite()
>>
>> Seems like I could get rid of, 2 and 3. I'm not sure why I have to
>> bwrite the buffer returned from UFS_BALLOC. I dug around and there are
>> a few examples of where ffs/ufs manipulate the buffer returned from
>> UFS_BALLOC so I am guessing I'm missing a flag or something.
>>
>> Hope the question is clear enough!
>
> You shouldn't be groping around in UFS internals like this. Using
> VOP_WRITE will probably serve you much better. Look at something like
> kern_ktrace.c to see how file writing in the kernel is done.
>
> Scott
>
>
>
>
More information about the freebsd-fs
mailing list