On 12/10, 徐瑞斌 wrote:
> Hi, Jaegeuk,
>
> I comment here the patch your provided (3 parts, since the patch contains 3
> fixes):
> 1. + dn->data_blkaddr = blkaddr;
> ret = reserve_new_block(sbi, &dn->data_blkaddr, &sum, type, 0);
>
> We cannot assign dn->data_blkaddr here. The old one is to be used in
> reserve_new_block() function. Also, reserve_new_block() function actually
> will set dn->data_blkaddr to blkaddr in the end.
This tries to avoid deleting the block address used in the previous offset.
Otherwise, we'll see wrong i_blocks.
>
> 2. Added condition "n < (1 << c.sldc_cc.log_cluster_size) * BLOCK_SZ"
>
> The semantic meaning of the whole if statement is to say:
> When the compression fail (ret != 0) or the original read size is
> smaller than the compressed size plus (the minimum block saved (specified
> by the user) x block size), we will not do compression but just write the
> data as is.
This is missing the last block having < 4Kb.
>
> The right hand side (RHS) of your added condition is exactly the read size,
> i.e. the cluster size. That means the condition is always false except the
> read of the last part of the file, when the file size is not exactly the
> multiple of the cluster size. That means we will never try to compress the
> last part of the file (when the last part is not a multiple of the cluster
> size)
>
> IMHO, the original implementation should be correct.
>
> 3. node_blk->i.i_blocks += cpu_to_le64(cblocks);
>
> I am not quite sure of the i_blocks count. Did you mean that when the file
> is mutable, meaning that the file reserves some blocks for future write,
> we will add count to i_blocks to mark the block as a used block by the
> file, right? I thought we only need to increment the allocated count...
Should add it.
>
> Regards,
> Robin Hsu 徐瑞斌
>
>
> On Thu, Dec 10, 2020 at 4:42 PM Chao Yu <[email protected]> wrote:
>
> > On 2020/12/8 16:15, Robin Hsu wrote:
> > > From: Robin Hsu <[email protected]>
> > >
> > > Add F2FS compression support for sload
> > > * Support file extension filter, either default-accept or default-deny
> > > policy
> > > * Support choice of compression algorithm, LZO (version 2) or LZ4
> > > (default)
> > > * Support custom log of cluster size
> > > * Support minimum number of compressed blocks per cluster (default 1).
> > > A cluster will not be compressed if the number can not be met.
> > > * suuport -r (read-only) option
> >
> > Could you please update manual as well?
> >
> > > +
> > > + /* sldc: sload compression support */
> >
> > Personally, I don't like the naming method of adding "sldc_" prefix... :(
> >
> > > + bool sldc_en;
> > > + bool sldc_use_allow_list; /* default false to use the deny list */
> > > + struct compress_ctx sldc_cc;
> > > + u8 sldc_ca; /* compress algorithm: 0 = LZO, 1 = LZ4 */
> > > + compress_ops *sldc_compr;
> > > + enum filter_policy sldc_policy;
> > > + /* max_cppc can used to specify minimum compression rate */
> > > + unsigned int sldc_min_cbpc; /* min compressed pages per cluster */
> > > + bool sldc_got_opt;
> > > + bool sldc_immutable;
> > > + struct ext_tbl_op *sldc_ef; /* extension filter */
> >
> > The variables name like sldc_en, sldc_ca, min_cbpc, sldc_ef makes
> > developers
> > hard to understand w/o comments, and also there is no comments for several
> > variable like sldc_en, sldc_cc...
> >
> > Could you please improve the naming like f2fs-tools style?
> >
> > Thanks,
> >
Chao and Robin,
I refactored the patch to modify the names and structures.
Please take a look at this.
https://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs-tools.git/commit/?h=dev&id=5cd8e5fcc79611c0692f18c7e7e13d6b4742d3c0
On 12/10, Jaegeuk Kim wrote:
> On 12/10, 徐瑞斌 wrote:
> > Hi, Jaegeuk,
> >
> > I comment here the patch your provided (3 parts, since the patch contains 3
> > fixes):
> > 1. + dn->data_blkaddr = blkaddr;
> > ret = reserve_new_block(sbi, &dn->data_blkaddr, &sum, type, 0);
> >
> > We cannot assign dn->data_blkaddr here. The old one is to be used in
> > reserve_new_block() function. Also, reserve_new_block() function actually
> > will set dn->data_blkaddr to blkaddr in the end.
>
> This tries to avoid deleting the block address used in the previous offset.
> Otherwise, we'll see wrong i_blocks.
>
> >
> > 2. Added condition "n < (1 << c.sldc_cc.log_cluster_size) * BLOCK_SZ"
> >
> > The semantic meaning of the whole if statement is to say:
> > When the compression fail (ret != 0) or the original read size is
> > smaller than the compressed size plus (the minimum block saved (specified
> > by the user) x block size), we will not do compression but just write the
> > data as is.
>
> This is missing the last block having < 4Kb.
>
> >
> > The right hand side (RHS) of your added condition is exactly the read size,
> > i.e. the cluster size. That means the condition is always false except the
> > read of the last part of the file, when the file size is not exactly the
> > multiple of the cluster size. That means we will never try to compress the
> > last part of the file (when the last part is not a multiple of the cluster
> > size)
> >
> > IMHO, the original implementation should be correct.
> >
> > 3. node_blk->i.i_blocks += cpu_to_le64(cblocks);
> >
> > I am not quite sure of the i_blocks count. Did you mean that when the file
> > is mutable, meaning that the file reserves some blocks for future write,
> > we will add count to i_blocks to mark the block as a used block by the
> > file, right? I thought we only need to increment the allocated count...
>
> Should add it.
>
> >
> > Regards,
> > Robin Hsu 徐瑞斌
> >
> >
> > On Thu, Dec 10, 2020 at 4:42 PM Chao Yu <[email protected]> wrote:
> >
> > > On 2020/12/8 16:15, Robin Hsu wrote:
> > > > From: Robin Hsu <[email protected]>
> > > >
> > > > Add F2FS compression support for sload
> > > > * Support file extension filter, either default-accept or default-deny
> > > > policy
> > > > * Support choice of compression algorithm, LZO (version 2) or LZ4
> > > > (default)
> > > > * Support custom log of cluster size
> > > > * Support minimum number of compressed blocks per cluster (default 1).
> > > > A cluster will not be compressed if the number can not be met.
> > > > * suuport -r (read-only) option
> > >
> > > Could you please update manual as well?
> > >
> > > > +
> > > > + /* sldc: sload compression support */
> > >
> > > Personally, I don't like the naming method of adding "sldc_" prefix... :(
> > >
> > > > + bool sldc_en;
> > > > + bool sldc_use_allow_list; /* default false to use the deny list */
> > > > + struct compress_ctx sldc_cc;
> > > > + u8 sldc_ca; /* compress algorithm: 0 = LZO, 1 = LZ4 */
> > > > + compress_ops *sldc_compr;
> > > > + enum filter_policy sldc_policy;
> > > > + /* max_cppc can used to specify minimum compression rate */
> > > > + unsigned int sldc_min_cbpc; /* min compressed pages per cluster */
> > > > + bool sldc_got_opt;
> > > > + bool sldc_immutable;
> > > > + struct ext_tbl_op *sldc_ef; /* extension filter */
> > >
> > > The variables name like sldc_en, sldc_ca, min_cbpc, sldc_ef makes
> > > developers
> > > hard to understand w/o comments, and also there is no comments for several
> > > variable like sldc_en, sldc_cc...
> > >
> > > Could you please improve the naming like f2fs-tools style?
> > >
> > > Thanks,
> > >
>
>
> _______________________________________________
> Linux-f2fs-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel