Re: [RFC] [PATCHv3 2/9] reiser4: block_alloc: add a "forward" parameter to reiser4_blocknr_hint to allocate blocks only in forward direction.

[Edward Shishkin <edward.shishkin@gmail.com>]

On 10/23/2014 10:58 PM, Ivan Shapovalov wrote:
> On Thursday 23 October 2014 at 16:37:00, Edward Shishkin wrote:	
>> On 10/23/2014 09:24 AM, Ivan Shapovalov wrote:
>>> On Tuesday 21 October 2014 at 17:50:23, Edward Shishkin wrote:	
>>>> [...]
>>>>
>>>> I suggest to call this bitfield "monotonic_forward"
>>> Isn't it an opposite of "backward"? Should we rename both?
>>
>> No, it isn't. And this is the reason of the rename.
>> monotonic_forward is to skip the second pass in alloc_blocks_forward()
> OK, will rename.
>
>>> BTW, in plugin/space/bitmap.c:1141 (around that line)
>>> in function alloc_blocks_forward()
>>> shouldn't the second scan be done with bitmap_alloc_backward(), as per the
>>> comment?
>>
>> I think that those comment means a jump in backward direction
>> (to the start) and one more pass with bitmap_alloc_forward().
>> Such a "compound, non-monotonic" forward...
> I don't intend to argue here, but doesn't that defeat the purpose of hint->blk?
> The first pass begins from the *closest* blocks. And if it fails, the second
> pass begins from the *most distant* blocks...
>
> ("closest" and "most distant" terms are used here with respect to hint->blk)


You don't like the block allocator?
Actually it was a subject of long investigations in ReiserFS (v3).
You can find a number of block allocators in ./fs/reiserfs/bitmap.c
I don't know which one went to reiser4..

Edward.