filecoin探索之路:复制证明(一)_OMM:Victory Gem

一、复制证明简介

引用官方的解释就是:“InordertoregisterasectorwiththeFilecoinnetwork,thesectorhastobesealed.Sealingisacomputation-heavyprocessthatproducesauniquerepresentationofthedataintheformofaproof,calledProof-of-ReplicationorPoRep.”简单来说,复制证明就是在对扇区进行封装的过程中生成的扇区唯一标识。

复制证明要用到三种特殊参数:数据本身、执行密封的矿工参与者、特定矿工密封特定数据的时间。一旦其中的一个参数发生变化,那么得到的复制证明结果将会完全不同。换句话说,如果同一个矿工稍后试图密封相同的数据,那么这将导致不同的PoRep证明。

复制证明是一个很大的计算过程,接下来我将会分为两部分:P1、P2,从代码的形式给读者介绍复制证明的工作原理。

二、P1代码解析

在本次文章,我将主要介绍32GB封装的P1的过程。在此阶段,会发生PoRep的SDR编码和复制。

因为是第一次,我这里提一句,扇区的不同状态会触发miner不同的执行方法,1.16版本可以看externstorage-sealingfsm.go文件约460行代码内容,代码中记录了miner不同的状态以及触发方法。这里我只放P1状态的代码。

????????...

????????...

????????case?Packing:

????????????????return?m.handlePacking,?processed,?nil

????????case?GetTicket:

????????????????return?m.handleGetTicket,?processed,?nil

????????case?PreCommit1:

????????????????return?m.handlePreCommit1,?processed,?nil

????????case?PreCommit2:

????????????????return?m.handlePreCommit2,?processed,?nil

????????...

????????...

可以看到,PreCommit1调用的是handlePreCommit1方法,从下边可以看出,利用SealPreCommit1方法得到P1结果。

func?(m?*Sealing)?handlePreCommit1(ctx?statemachine.Context,?sector?SectorInfo)?error?{

????????...

????????...

????????pc1o,?err?:=?m.sealer.SealPreCommit1(sector.sealingCtx(ctx.Context()),?m.minerSector(sector.SectorType,?sector.SectorNumber),?sector.TicketValue,?sector.pieceInfos())

????????if?err?!=?nil?{

????????????????return?ctx.Send(SectorSealPreCommit1Failed{xerrors.Errorf("seal?pre?commit(1)?failed:?%w",?err。)

????????}

????????return?ctx.Send(SectorPreCommit1{

????????????????PreCommit1Out:?pc1o,

????????})

}

让我们深入看一下SealPreCommit1方法,这里我们最终调用的是:func(sb*Sealer)SealPreCommit1(...)方法。方法中有我们常常遇到的方法:AcquireSector(...)、Unpadded()。

AcquireSector方法是根据传入的类型与sectorID一起,组合成对应的path。

Uppadded方法是返回一个Piece的未填充大小,以字节为单位,计算公式是:s-(s/128)。有未填充大小,自然就有填充大小,填充大小的计算方法为Padded(),计算公式是:s+(s/127)

func?(sb?*Sealer)?SealPreCommit1(ctx?context.Context,?sector?storage.SectorRef,?ticket?abi.SealRandomness,?pieces?abi.PieceInfo)?(out?storage.PreCommit1Out,?err?error)?{

????????paths,?done,?err?:=?sb.sectors.AcquireSector(ctx,?sector,?storiface.FTUnsealed,?storiface.FTSealed|storiface.FTCache,?storiface.PathSealing)

????????if?err?!=?nil?{

????????????????return?nil,?xerrors.Errorf("acquiring?sector?paths:?%w",?err)

????????}

????????...

????????...

????????...

????????var?sum?abi.UnpaddedPieceSize

????????for?_,?piece?:=?range?pieces?{

????????????????sum?+=?piece.Size.Unpadded()

????????}

????????//?根据扇区证明类型获取扇区大小

????????ssize,?err?:=?sector.ProofType.SectorSize()

????????if?err?!=?nil?{

????????????????return?nil,?err

????????}

????????//?这里比较一次总piece大小和要求的扇区大小是否一致

????????ussize?:=?abi.PaddedPieceSize(ssize).Unpadded()

????????if?sum?!=?ussize?{

????????????????return?nil,?xerrors.Errorf("aggregated?piece?sizes?don't?match?sector?size:?%d?!=?%d?(%d)",?sum,?ussize,?int64(ussize-sum))

????????}

????????//?TODO:?context?cancellation?respect

????????p1o,?err?:=?ffi.SealPreCommitPhase1(

????????????????sector.ProofType,

????????????????paths.Cache,

????????????????paths.Unsealed,

????????????????paths.Sealed,

????????????????sector.ID.Number,

????????????????sector.ID.Miner,

????????????????ticket,

????????????????pieces,

????????)

????????...

????????...

}

接下来,一切准备就绪,我们将要开始我们的P1远游了,因为接下来的代码都不属于lotus,上面方法中我们可以看到ffi.SealPreCommitPhase1,ffi其实使用的是https://github.com/filecoin-project/filecoin-ffi库,我们通过这个库的如下方法,转入rust语言去实现P1。

func?SealPreCommitPhase1(registeredProof?RegisteredSealProof,?cacheDirPath?SliceRefUint8,?stagedSectorPath?SliceRefUint8,?sealedSectorPath?SliceRefUint8,?sectorId?uint64,?proverId?*ByteArray32,?ticket?*ByteArray32,?pieces?SliceRefPublicPieceInfo)?(byte,?error)?{

????????resp?:=?C.seal_pre_commit_phase1(registeredProof,?cacheDirPath,?stagedSectorPath,?sealedSectorPath,?C.uint64_t(sectorId),?proverId,?ticket,?pieces)

????????defer?resp.destroy()

????????if?err?:=?CheckErr(resp);?err?!=?nil?{

????????????????return?nil,?err

????????}

????????return?resp.value.copy(),?nil

}

C库其实就是ffi库自身的rust库,调用的方法如下所示:

fn?seal_pre_commit_phase1(

????registered_proof:?RegisteredSealProof,

????cache_dir_path:?c_slice::Ref<u8>,

????staged_sector_path:?c_slice::Ref<u8>,

????sealed_sector_path:?c_slice::Ref<u8>,

????sector_id:?u64,

????prover_id:?&,

????ticket:?&,

????pieces:?c_slice::Ref<PublicPieceInfo>,

)?->?repr_c::Box<SealPreCommitPhase1Response>?{

????catch_panic_response("seal_pre_commit_phase1",?||?{

????????let?public_pieces:?Vec<PieceInfo>?=?pieces.iter().map(Into::into).collect();

????????let?result?=?seal::seal_pre_commit_phase1(

????????????registered_proof.into(),

????????????as_path_buf(&cache_dir_path)?,

????????????as_path_buf(&staged_sector_path)?,

????????????as_path_buf(&sealed_sector_path)?,

????????????*prover_id,

????????????SectorId::from(sector_id),

????????????*ticket,

????????????&public_pieces,

????????)?;

????????let?result?=?serde_json::to_vec(&result)?;

????????Ok(result.into_boxed_slice().into())

????})

}

上面的seal库是:https://github.com/filecoin-project/rust-filecoin-proofs-api。在这个方法对应的文件中,我们可以看到很多方法都对应了一个*__inner方法。实际上seal_pre_commit_phase1只是做了个中转。我们可以直接看seal_pre_commit_phase1_inner方法

pub?fn?seal_pre_commit_phase1<R,?S,?T>(

????registered_proof:?RegisteredSealProof,

????cache_path:?R,

????in_path:?S,

????out_path:?T,

????prover_id:?ProverId,

????sector_id:?SectorId,

????ticket:?Ticket,

????piece_infos:?&,

)?->?Result<SealPreCommitPhase1Output>

where

????R:?AsRef<Path>,

????S:?AsRef<Path>,

????T:?AsRef<Path>,

{

????ensure!(

????????registered_proof.major_version()?==?1,

????????"unusupported?version"

????);

????with_shape!(

????????u64::from(registered_proof.sector_size()),

????????seal_pre_commit_phase1_inner,

????????registered_proof,

????????cache_path.as_ref(),

????????in_path.as_ref(),

????????out_path.as_ref(),

????????prover_id,

????????sector_id,

????????ticket,

????????piece_infos

????)

}

在inner方法中,filecoin_proofs_v1::seal_pre_commit_phase1,会调用证明子系统的实现部分。filecoin_proofs_v1使用的库是:https://github.com/filecoin-project/rust-fil-proofs。

fn?seal_pre_commit_phase1_inner<Tree:?'static?+?MerkleTreeTrait>(

????registered_proof:?RegisteredSealProof,

????cache_path:?&Path,

????in_path:?&Path,

????out_path:?&Path,

????prover_id:?ProverId,

????sector_id:?SectorId,

????ticket:?Ticket,

????piece_infos:?&,

)?->?Result<SealPreCommitPhase1Output>?{

????let?config?=?registered_proof.as_v1_config();

????let?output?=?filecoin_proofs_v1::seal_pre_commit_phase1::<_,?_,?_,?Tree>(

????????config,

????????cache_path,

????????in_path,

????????out_path,

????????prover_id,

????????sector_id,

????????ticket,

????????piece_infos,

????)?;

????let?filecoin_proofs_v1::types::SealPreCommitPhase1Output::<Tree>?{

????????labels,

????????config,

????????comm_d,

????}?=?output;

????Ok(SealPreCommitPhase1Output?{

????????registered_proof,

????????labels:?Labels::from_raw::<Tree>(registered_proof,?&labels)?,

????????config,

????????comm_d,

????})

}

filecoin_proofs_v1::seal_pre_commit_phase1方法就是真正实现P1的地方,我将会在这里详细讲解P1,使P1将在这里一一浮出水面。

pub?fn?seal_pre_commit_phase1<R,?S,?T,?Tree:?'static?+?MerkleTreeTrait>(

????porep_config:?PoRepConfig,

????cache_path:?R,

????in_path:?S,

????out_path:?T,

????prover_id:?ProverId,

????sector_id:?SectorId,

????ticket:?Ticket,

????piece_infos:?&,

)?->?Result<SealPreCommitPhase1Output<Tree>>

where

????R:?AsRef<Path>,

????S:?AsRef<Path>,

????T:?AsRef<Path>,

{

????info!("seal_pre_commit_phase1:start:?{:?}",?sector_id);

????//?Sanity?check?all?input?path?types.

????ensure!(

????????metadata(in_path.as_ref())?.is_file(),

????????"in_path?must?be?a?file"

????);

????ensure!(

????????metadata(out_path.as_ref())?.is_file(),

????????"out_path?must?be?a?file"

????);

????ensure!(

????????metadata(cache_path.as_ref())?.is_dir(),

????????"cache_path?must?be?a?directory"

????);

????let?sector_bytes?=?usize::from(PaddedBytesAmount::from(porep_config));

????fs::metadata(&in_path)

????????.with_context(||?format!("could?not?read?in_path={:?})",?in_path.as_ref().display()))?;

????fs::metadata(&out_path)

????????.with_context(||?format!("could?not?read?out_path={:?}",?out_path.as_ref().display()))?;

????//?Copy?unsealed?data?to?output?location,?where?it?will?be?sealed?in?place.

????fs::copy(&in_path,?&out_path).with_context(||?{

????????format!(

????????????"could?not?copy?in_path={:?}?to?out_path={:?}",

????????????in_path.as_ref().display(),

????????????out_path.as_ref().display()

????????)

????})?;

????let?f_data?=?OpenOptions::new()

????????.read(true)

????????.write(true)

????????.open(&out_path)

????????.with_context(||?format!("could?not?open?out_path={:?}",?out_path.as_ref().display()))?;

????//?Zero-pad?the?data?to?the?requested?size?by?extending?the?underlying?file?if?needed.

????f_data.set_len(sector_bytes?as?u64)?;

????let?data?=?unsafe?{

????????//?创建由文件支持的可写内存映射

????????MmapOptions::new()

????????????.map_mut(&f_data)

????????????.with_context(||?format!("could?not?mmap?out_path={:?}",?out_path.as_ref().display()))?

????};

????let?compound_setup_params?=?compound_proof::SetupParams?{

????????vanilla_params:?setup_params(

????????????PaddedBytesAmount::from(porep_config),

????????????usize::from(PoRepProofPartitions::from(porep_config)),

????????????porep_config.porep_id,

????????????porep_config.api_version,

????????)?,

????????partitions:?Some(usize::from(PoRepProofPartitions::from(porep_config))),

????????priority:?false,

????};

????//?利用param得到public_params,其vanilla_params.graph字段,就是构建出来的图的数据结构。

????let?compound_public_params?=?<StackedCompound<Tree,?DefaultPieceHasher>?as?CompoundProof<

????????StackedDrg<'_,?Tree,?DefaultPieceHasher>,

????????_,

????>>::setup(&compound_setup_params)?;

????trace!("building?merkle?tree?for?the?original?data");

????let?(config,?comm_d)?=?measure_op(Operation::CommD,?||?->?Result<_>?{

????????let?base_tree_size?=?get_base_tree_size::<DefaultBinaryTree>(porep_config.sector_size)?;

????????let?base_tree_leafs?=?get_base_tree_leafs::<DefaultBinaryTree>(base_tree_size)?;

????????ensure!(

????????????compound_public_params.vanilla_params.graph.size()?==?base_tree_leafs,

????????????"graph?size?and?leaf?size?don't?match"

????????);

????????trace!(

????????????"seal?phase?1:?sector_size?{},?base?tree?size?{},?base?tree?leafs?{}",

????????????u64::from(porep_config.sector_size),

????????????base_tree_size,

????????????base_tree_leafs,

????????);

????????let?mut?config?=?StoreConfig::new(

????????????cache_path.as_ref(),

????????????CacheKey::CommDTree.to_string(),

????????????default_rows_to_discard(base_tree_leafs,?BINARY_ARITY),

????????);

????????let?data_tree?=?create_base_merkle_tree::<BinaryMerkleTree<DefaultPieceHasher>>(

????????????Some(config.clone()),

????????????base_tree_leafs,

????????????&data,

????????)?;

????????drop(data);

????????config.size?=?Some(data_tree.len());

????????let?comm_d_root:?Fr?=?data_tree.root().into();

????????let?comm_d?=?commitment_from_fr(comm_d_root);

????????drop(data_tree);

????????Ok((config,?comm_d))

????})?;

????trace!("verifying?pieces");

????ensure!(

????????verify_pieces(&comm_d,?piece_infos,?porep_config.into())?,

????????"pieces?and?comm_d?do?not?match"

????);

????let?replica_id?=?generate_replica_id::<Tree::Hasher,?_>(

????????&prover_id,

????????sector_id.into(),

????????&ticket,

????????comm_d,

????????&porep_config.porep_id,

????);

????let?labels?=?StackedDrg::<Tree,?DefaultPieceHasher>::replicate_phase1(

????????&compound_public_params.vanilla_params,

????????&replica_id,

????????config.clone(),

????)?;

????let?out?=?SealPreCommitPhase1Output?{

????????labels,

????????config,

????????comm_d,

????};

????info!("seal_pre_commit_phase1:finish:?{:?}",?sector_id);

????Ok(out)

}

P1实现解释

上边seal_pre_commit_phase1的代码中,我们可以看到有三个path,这三个path分别对应:in_path->unsealedpath、out_path->sealedpath、cache_path->cachepath。代码会先去检查这三个path,他们两个是文件,一个是文件夹。

检查完path后我们可以看到fs::copy方法,它将unsealed文件拷贝到了sealed文件中,完成封装。

Copy完成后拿出sealed文件的数据,并利用.set_len()方法填充数据(或删减),使sealed数据达到证明类型配置规定的扇区大小。

setup_params()

setup_params()方法利用证明类型配置构建启动参数。这里传入的参数为:扇区大小、分区数、证明类型id、证明类型版本。分区数可看https://github.com/filecoin-project/rust-filecoin-proofs-api/blob/23ae2893741829bddc29d7211e06c914bab5423c/src/registry.rs中的partitions()方法,在对应https://github.com/filecoin-project/rust-fil-proofs/blob/ec2ef88a17ffed991b64dc8d96b30c36b275eca0/filecoin-proofs/src/constants.rs得到具体值。我分析以32GB扇区为主,因此分区数为10。另外三个就不讲了,跟分区数一样,都是从这两个文件得到的。

pub?fn?setup_params(

????sector_bytes:?PaddedBytesAmount,

????partitions:?usize,

????porep_id:?,

????api_version:?ApiVersion,

)?->?Result<stacked::SetupParams>?{

????//?得到挑战层数和最大挑战次数

????let?layer_challenges?=?select_challenges(

????????partitions,

????????*POREP_MINIMUM_CHALLENGES

????????????.read()

????????????.expect("POREP_MINIMUM_CHALLENGES?poisoned")

????????????.get(&u64::from(sector_bytes))

????????????.expect("unknown?sector?size")?as?usize,

????????*LAYERS

????????????.read()

????????????.expect("LAYERS?poisoned")

????????????.get(&u64::from(sector_bytes))

????????????.expect("unknown?sector?size"),

????);

????let?sector_bytes?=?u64::from(sector_bytes);

????ensure!(

????????sector_bytes?%?32?==?0,

????????"sector_bytes?({})?must?be?a?multiple?of?32",

????????sector_bytes,

????);

????let?nodes?=?(sector_bytes?/?32)?as?usize;????//?节点数,SDR共有11层,每一层的节点数量相当于1GiB的字节数量。

????let?degree?=?DRG_DEGREE;????//?用于所有?DRG?图的基础度数,?DRG_DEGREE=6。

????let?expansion_degree?=?EXP_DEGREE;?//大小是8,上一层中抽取的节点数量,用来计算当前层的节点数据

????Ok(stacked::SetupParams?{

????????nodes,

????????degree,

????????expansion_degree,

????????porep_id,

????????layer_challenges,

????????api_version,

????})

}

Merkletree和对应comm_d的生成

看完setup_params方法,让我们继续看seal_pre_commit_phase1中的compound_public_params参数,这里实际上set_up的时候,将compound_setup_params参数的值赋予进去,并增加了一个至关重要的vanilla_params.graph字段,就是构造出来的图的数据结构

??接下来我们可以看到seal_pre_commit_phase1方法的70行,在这一段代码用于生成markletree和comm_d

????let?(config,?comm_d)?=?measure_op(Operation::CommD,?||?->?Result<_>?{

????????let?base_tree_size?=?get_base_tree_size::<DefaultBinaryTree>(porep_config.sector_size)?;

????????let?base_tree_leafs?=?get_base_tree_leafs::<DefaultBinaryTree>(base_tree_size)?;

????????ensure!(

????????????compound_public_params.vanilla_params.graph.size()?==?base_tree_leafs,

????????????"graph?size?and?leaf?size?don't?match"

????????);

????????trace!(

????????????"seal?phase?1:?sector_size?{},?base?tree?size?{},?base?tree?leafs?{}",

????????????u64::from(porep_config.sector_size),

????????????base_tree_size,

????????????base_tree_leafs,

????????);

????????let?mut?config?=?StoreConfig::new(

????????????cache_path.as_ref(),

????????????CacheKey::CommDTree.to_string(),

????????????default_rows_to_discard(base_tree_leafs,?BINARY_ARITY),

????????);

????????//?创建默克尔树,根据其树根得到comm_d

????????let?data_tree?=?create_base_merkle_tree::<BinaryMerkleTree<DefaultPieceHasher>>(

????????????Some(config.clone()),

????????????base_tree_leafs,

????????????&data,

????????)?;

????????drop(data);

????????config.size?=?Some(data_tree.len());

????????let?comm_d_root:?Fr?=?data_tree.root().into();

????????let?comm_d?=?commitment_from_fr(comm_d_root);

????????drop(data_tree);

????????Ok((config,?comm_d))

????})?;

这里我们会生成treestoreconfig,然后利用config、base_tree_leafs、sealed填充数据,生成一个merkletree。得到了merkletree后就可以得到merkletree的根。再利用merkletree的根,通过commitment_from_fr算出comm_d。

生成副本id(replica_id)

当我们拿到了comm_d后,会利用verify_pieces方法验证一下comm_d,这个就不讲了,感兴趣的可以自己去看代码。

让我们看一下副本id是如何生成的

????let?replica_id?=?generate_replica_id::<Tree::Hasher,?_>(

????????&prover_id,

????????sector_id.into(),

????????&ticket,

????????comm_d,

????????&porep_config.porep_id,

????);

利用数据本身生成得到了comm_d,这里再加上矿工id、扇区id、ticket,证明类型id。就能得到replicaid值。

///?Generate?the?replica?id?as?expected?for?Stacked?DRG.

pub?fn?generate_replica_id<H:?Hasher,?T:?AsRef<>>(

????prover_id:?&,

????sector_id:?u64,

????ticket:?&,

????comm_d:?T,

????porep_seed:?&,

)?->?H::Domain?{

????//?以链式方式处理输入数据。

????let?hash?=?Sha256::new()

????????.chain_update(prover_id)

????????.chain_update(&sector_id.to_be_bytes())

????????.chain_update(ticket)

????????.chain_update(&comm_d)

????????.chain_update(porep_seed)

????????.finalize();

????bytes_into_fr_repr_safe(hash.as_ref()).into()????//通过将?le_bytes?的最重要的两位归零,将?32?字节的切片转换为?Fr::Repr。

}

生成labels

接下来就是P1最后的操作:生成labels。将public_params、复制id和treestoreconfig作为参数传入。

????pub?fn?replicate_phase1(

????????pp:?&'a?PublicParams<Tree>,

????????replica_id:?&<Tree::Hasher?as?Hasher>::Domain,

????????config:?StoreConfig,

????)?->?Result<Labels<Tree>>?{

????????info!("replicate_phase1");

????????let?labels?=?measure_op(Operation::EncodeWindowTimeAll,?||?{

????????????Self::generate_labels_for_encoding(&pp.graph,?&pp.layer_challenges,?replica_id,?config)

????????})?

????????.0;

????????Ok(labels)

????}

这里可以看到,代码提取出了public_params的.graph字段,就是构造出来的图的数据结构,和public_params中包含挑战层数和最大挑战次数的layer_challenges。

接下来看generate_labels_for_encoding。这里可以分为多核与单核进行SDR编码,创建labels。

????pub?fn?generate_labels_for_encoding(

????????graph:?&StackedBucketGraph<Tree::Hasher>,

????????layer_challenges:?&LayerChallenges,

????????replica_id:?&<Tree::Hasher?as?Hasher>::Domain,

????????config:?StoreConfig,

????)?->?Result<(Labels<Tree>,?Vec<LayerState>)>?{

????????let?mut?parent_cache?=?graph.parent_cache()?;

????????#

????????{

????????????if?SETTINGS.use_multicore_sdr?{

????????????????info!("multi?core?replication");

????????????????create_label::multi::create_labels_for_encoding(

????????????????????graph,

????????????????????&parent_cache,

????????????????????layer_challenges.layers(),

????????????????????replica_id,

????????????????????config,

????????????????)

????????????}?else?{

????????????????info!("single?core?replication");

????????????????create_label::single::create_labels_for_encoding(

????????????????????graph,

????????????????????&mut?parent_cache,

????????????????????layer_challenges.layers(),

????????????????????replica_id,

????????????????????config,

????????????????)

????????????}

????????}

????????#

????????{

????????????info!("single?core?replication");

????????????create_label::single::create_labels_for_encoding(

????????????????graph,

????????????????&mut?parent_cache,

????????????????layer_challenges.layers(),

????????????????replica_id,

????????????????config,

????????????)

????????}

????}

我将生成label的地址放在这里,想看的可以去看一下,这里就不细讲了。

多核:https://github.com/filecoin-project/rust-fil-proofs/blob/master/storage-proofs-porep/src/stacked/vanilla/create_label/multi.rs

单核:https://github.com/filecoin-project/rust-fil-proofs/blob/master/storage-proofs-porep/src/stacked/vanilla/create_label/single.rs

三、总结

其实rust语言我接触不多,开始的时候看得有点头痛,最后也是硬着头皮啃下来的。

如有大佬认为文章有不对的地方,欢迎纠正。

来源:金色财经

郑重声明: 本文版权归原作者所有, 转载文章仅为传播更多信息之目的, 如作者信息标记有误, 请第一时间联系我们修改或删除, 多谢。

金宝趣谈

[0:0ms0-5:765ms