[Ocfs2-devel] what is global bitmap, block group, suballocator, etc

[Ocfs2-devel] what is global bitmap, block group, suballocator, etc

[Coly Li]

Hi list,

There are some ocfs2 terminologies that I don't understand very well. After checking the ocfs2 wiki and kernel/tools
code, I am still not sure whether I understand them correctly.

- Global bitmap
  What's the usage of global bitmap? Which kind of resource allocation will set global bitmap?

- Cluster group
  It seems cluster group is similar to block group of ext3/4. Does each node of the cluster has its own data/metadata
bitmap inode in every cluster group ?

- Block group
  From Tiger's slide, I don't find block group. But from fs/ocfs2 code, I see block group. Is it same to cluster group ?

- Suballocator & Suballocator slot
  I don't find related document explains what is suballocator and how it works. I am checking the code now, but not
clear to me yet.

I know all the answers are in the code, I spent some time on code reading already, if there is some hints, I can do it
faster :-)

Thanks in advance.
-- 
Coly Li
SuSE Labs

[Ocfs2-devel] what is global bitmap, block group, suballocator, etc

[Tao Ma]

Hi Coly,

Coly Li wrote:
> Hi list,
> 
> There are some ocfs2 terminologies that I don't understand very well. After checking the ocfs2 wiki and kernel/tools
> code, I am still not sure whether I understand them correctly.
> 
> - Global bitmap
>   What's the usage of global bitmap? Which kind of resource allocation will set global bitmap?
global bitmap is a system file which indicates the usage of clusters in 
  the whole volume. It is a chain file. So it has a ocfs2_chain_list in 
its ocfs2_dinode.id2. And the chain_list is a chain of many groups. A 
group is a range of clusters and the 1st block of a group has a bitmap 
to indicate which cluster(or block, cpg is used here) is allocated or not.
> 
> - Cluster group
>   It seems cluster group is similar to block group of ext3/4. Does each node of the cluster has its own data/metadata
> bitmap inode in every cluster group ?
See group above.
> 
> - Block group
>   From Tiger's slide, I don't find block group. But from fs/ocfs2 code, I see block group. Is it same to cluster group ?
you mean ocfs2_block_group_alloc? Actually we allocates clusters from 
the allocators(see ocfs2_claim_clusters) and then call 
ocfs2_block_group_fill to fill it into different system 
file(inode_alloc, extent_alloc).
> 
> - Suballocator & Suballocator slot
>   I don't find related document explains what is suballocator and how it works. I am checking the code now, but not
> clear to me yet.
suballocator is used so that when a node try to get some resource, it 
will try its own suballocator first. If it can't get the resource it 
wants, we have to lock the global_bitmap(there are some cases we can try 
local_alloc first and if it fails, then try global_bitmap) and ask from 
that(it will need a global lock which needs more time). We we the 
suballocator grows, it will request a large hunk from the global bitmap 
so that the next attempt can success in suballocator. There are 
inode_alloc for inode, extent_alloc for extent, and local_alloc(it can 
be used by extent_alloc allocation or cluster allocation).

Regards,
Tao

[Ocfs2-devel] what is global bitmap, block group, suballocator, etc

[Coly Li]

Hi Tao,

Thanks for your reply. I spent some time to read the code and understand :)

On 2009?11?25? 15:59, Tao Ma Wrote:
> Hi Coly,
> 
> Coly Li wrote:
>> Hi list,
>>
>> There are some ocfs2 terminologies that I don't understand very well.
>> After checking the ocfs2 wiki and kernel/tools
>> code, I am still not sure whether I understand them correctly.
>>
>> - Global bitmap
>>   What's the usage of global bitmap? Which kind of resource allocation
>> will set global bitmap?
> global bitmap is a system file which indicates the usage of clusters in
>  the whole volume. It is a chain file. So it has a ocfs2_chain_list in
> its ocfs2_dinode.id2. And the chain_list is a chain of many groups. A
> group is a range of clusters and the 1st block of a group has a bitmap
> to indicate which cluster(or block, cpg is used here) is allocated or not.

What is a chain file ? Is it a file not for regular file data storage (a.k.a dinode->i_size == 0), but to record
allocation chains of an ocfs2 volume (a.k.a dinode->id2.i_chain) ?

Another question is, for each node (slot), should it have only 1 chain file ? If yes, I guess the chain number limit is
ocfs2_chain_recs_per_inode().


-- 
Coly Li
SuSE Labs

[Ocfs2-devel] what is global bitmap, block group, suballocator, etc

[Tao Ma]

Coly Li wrote:
> Hi Tao,
> 
> Thanks for your reply. I spent some time to read the code and understand :)
> 
> On 2009?11?25? 15:59, Tao Ma Wrote:
>> Hi Coly,
>>
>> Coly Li wrote:
>>> Hi list,
>>>
>>> There are some ocfs2 terminologies that I don't understand very well.
>>> After checking the ocfs2 wiki and kernel/tools
>>> code, I am still not sure whether I understand them correctly.
>>>
>>> - Global bitmap
>>>   What's the usage of global bitmap? Which kind of resource allocation
>>> will set global bitmap?
>> global bitmap is a system file which indicates the usage of clusters in
>>  the whole volume. It is a chain file. So it has a ocfs2_chain_list in
>> its ocfs2_dinode.id2. And the chain_list is a chain of many groups. A
>> group is a range of clusters and the 1st block of a group has a bitmap
>> to indicate which cluster(or block, cpg is used here) is allocated or not.
> 
> What is a chain file ? Is it a file not for regular file data storage (a.k.a dinode->i_size == 0), but to record
> allocation chains of an ocfs2 volume (a.k.a dinode->id2.i_chain) ?
yeah, it is used in system file only. So when you allocate some clusters 
from the global_bitmap, you add it to your own chain file according to 
the usage.
> 
> Another question is, for each node (slot), should it have only 1 chain file ? If yes, I guess the chain number limit is
> ocfs2_chain_recs_per_inode().
No, you have different system files for different usage.
Please check the flag of the system file, it should have OCFS2_CHAIN_FL 
set. You can also investigate it by debugfs.ocfs2.
e.g
echo 'stat //inode_alloc:0000'|debugfs.ocfs2 /dev/sdx
you can see "Chain" in the "Flags:" field.

So normally we have inode_alloc:000X for inode allocation, 
extent_alloc:000X for metadata allocation,

Regards,
Tao