From: mingming cao Subject: Re: Proposal draft for data checksumming for ext4 Date: Mon, 14 Apr 2014 17:26:29 -0700 Message-ID: <534C7CB5.9010303@oracle.com> References: <20140320175950.GJ9070@birch.djwong.org> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: "Darrick J. Wong" , linux-ext4@vger.kernel.org, "Theodore Ts'o" To: =?UTF-8?B?THVrw6HFoSBDemVybmVy?= Return-path: Received: from aserp1040.oracle.com ([141.146.126.69]:39858 "EHLO aserp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750768AbaDOAYo (ORCPT ); Mon, 14 Apr 2014 20:24:44 -0400 In-Reply-To: Sender: linux-ext4-owner@vger.kernel.org List-ID: On 03/23/2014 06:59 PM, Luk=C3=A1=C5=A1 Czerner wrote: > On Thu, 20 Mar 2014, Darrick J. Wong wrote: > >> Date: Thu, 20 Mar 2014 10:59:50 -0700 >> From: Darrick J. Wong >> To: Luk=C3=A1=C5=A1 Czerner >> Cc: linux-ext4@vger.kernel.org, Theodore Ts'o >> Subject: Re: Proposal draft for data checksumming for ext4 >> >> On Thu, Mar 20, 2014 at 05:40:06PM +0100, Luk=C3=A1=C5=A1 Czerner wr= ote: >>> Hi all, >>> >>> I've started thinking about implementing data checksumming for ext4= file >>> system. This is not meant to be a formal proposal or a definitive d= esign >>> description since I am not that far yet, but just a few ideas to st= art >>> the discussion and trying to figure out what the best design for da= ta >>> checksumming in ext4 might be. >>> >>> >>> >>> Data checksumming for ext4 >>> Version 0.1 >>> March 20, 2014 >>> >>> >>> Goal >>> =3D=3D=3D=3D >>> >>> The goal is to implement data checksumming for ext4 file system in = order >>> to improve data integrity and increase protection against silent da= ta >>> corruption while maintaining reasonable performance and usability o= f the >>> file system. >>> >>> While data checksums can be certainly used in different ways, for e= xample >>> data deduplication this proposal is very much focused on data integ= rity. >>> >>> >>> Checksum function >>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D >>> >>> By default I plan to use crc32c checksum, but I do not see a reason= why not >>> not to be able to support different checksum function. Also by defa= ult the >>> checksum size should be 32 bits, but the plan is to make the format >>> flexible enough to be able to support different checksum sizes. >> >> Were you thinking of allowing the use of different functions f= or data and >> metadata checksums? > > Hi Darrick, > > I have not, but I think that this would be very easy to do if we can > agree that it's good to have. > > >> >>> Checksumming and Validating >>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D >>> >>> On write checksums on the data blocks need to be computed right bef= ore its >>> bio is submitted and written out as metadata to its position (see b= ellow) >>> after the bio completes (similarly as we do unwritten extent conver= sion >>> today). >>> >>> Similarly on read checksums needs to be computed after the bio comp= letes >>> and compared with the stored values to verify that the data is inta= ct. >>> >>> All of this should be done using workqueues (Concurrency Managed >>> Workqueues) so we do not block the other operations and to spread t= he >>> checksum computation and comparison across CPUs. One wq for reads a= nd one >>> for writes. Specific setup of the wq such as priority, or concurren= cy limits >>> should be decided later based on the performance evaluation. >>> >>> While we already have ext4 infrastructure to submit bios in >>> fs/ext4/page-io.c where the entry point is ext4_bio_write_page() we= would >>> need the same for reads to be able to provide ext4 specific hooks f= or >>> io completion. >>> >>> >>> Where to store the checksums >>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= =3D=3D=3D=3D=3D >>> >>> While the problems above are pretty straightforward when it comes t= o the >>> design, actually storing and retrieving the data checksums from to/= from >>> the ext4 format requires much more thought to be efficient enough a= nd play >>> nicely with the overall ext4 design while trying not to be too intr= usive. >>> >>> I came up with several ideas about where to store and how to access= data >>> checksums. While some of the ideas might not be the most viable opt= ions, >>> it's still interesting to think about the advantages and disadvanta= ges of >>> each particular solution. >>> >>> a) Static layout >>> ---------------- >>> >>> This scheme fits perfectly into the ext4 design. Checksum blocks >>> would be preallocated the same way as we do with inode tables for e= xample. >>> Each block group should have it's own contiguous region of checksum= blocks >>> to be able to store checksums for bocks from entire block group it = belongs >>> to. Each checksum block would contain header including checksum of = the >>> checksum block. >>> >>> We still have unused 4 Bytes in the ext4_group_desc structure, so s= toring >>> a block number for the checksum table should not be a problem. >> >> What if you have a 64bit filesystem? Do you have some strategy in m= ind to work >> around that? What about the snapshot exclusion bitmap field? Afaic= t that >> never went in, so perhaps that field could be reused? > > Yes we can use the exclusion bitmap field. I think that would not be > a problem. We could also use addressing from the start of the block > group and keep the checksum table in the block group. > >> >>> Finding a checksum location of each block in the block group should= be done >>> in O(1) time, which is very good. Other advantage is a locality wit= h the >>> data blocks in question since both resides in the same block group. >>> >>> Big disadvantage is the fact that this solution is not very flexibi= le which >>> comes from the fact that the location of "checksum table" is static= ally >>> located at a precise position in the file system at mkfs time. >> >> Having a big dumb block of checksums would be easier to prefetch fro= m disk for >> fsck and kernel driver, rather than having to dig through some tree = structure. >> (More on that below) > > I agree, it is also much more robust solution than having a tree. > >> >>> There are also other problems we should be concerned with. Ext4 fil= e system >>> does have support for metadata checksumming so all the metadata doe= s have >>> its own checksum. While we can avoid unnecessarily checksuming inod= es, group >>> descriptors and basicall all statically positioned metadata, we sti= ll have >>> dynamically allocated metadata blocks such as extent blocks. These = block >>> do not have to be checksummed but we would still have space reserve= d in the >>> checksum table. >> >> Don't forget directory blocks--they (should) have checksums too, so = you can >> skip those. >> >> I wonder, could we use this table to store backrefs too? It would m= ake the >> table considerably larger, but then we could (potentially) reconstru= ct broken >> extent trees. > > Definitely, that is one thing I did not discussed here, but I'd like > to have the checksum blocks self descriptive so we can alway know > where it belongs and who is the owner. So yes, having a backrefs is > really good idea. > Hello Lukas, Darrick, I just read this documents, quite interesting. I wonder if this was=20 discussed at ext4 workshop in Napa and any agreement has reached as to=20 how to store the data checksumming? Have you thought about store extent checksumming/backrefs rather than=20 per data block? If we are interested in adding backrefs, then I wonder if we could use=20 this table to store reflink counters as well? I understand only data=20 blocks referenced by multiple files need this, but this is a O(1)=20 faster solution than per-inode refcount tree, and we are thinking of=20 tracking data checksums/backrefs anyway, and the refcounters could be=20 checksummed at the same time. Mingming >> >>> I think that we should be able to make this feature without introdu= cing any >>> incompatibility, but it would make more sense to make it RO compati= ble only >>> so we can preserve the checksums. But that's up to the implementati= on. >> >> I think you'd have to have it be rocompat, otherwise you could write= data with >> an old kernel and a new kernel would freak out. > > Yes, I think that we could make it not freak out, but we would loose > the checksums, so for that I think that having this rocompat will > probably make more sense. > > Thanks! > -Lukas > >> >>> b) Special inode >>> ---------------- >>> >>> This is very "lazy" solution and should not be difficult to impleme= nt. The >>> idea is to have a special inode which would store the checksum bloc= ks in >>> it's own data blocks. >>> >>> The big disadvantage is that we would have to walk the extent tree = twice for >>> each read, or write. There is not much to say about this solution o= ther than >>> again we can make this feature without introducing any incompatibil= ity, but >>> it would probably make more sense to make it RO compatible to prese= rve the >>> checksums. >>> >>> c) Per inode checksum b-tree >>> ---------------------------- >>> >>> See d) >>> >>> d) Per block group checksum b-tree >>> ---------------------------------- >>> >>> Those two schemes are very similar in that both would store checksu= m in a >>> b-tree with a block number (we could use logical block number in pe= r inode >>> tree) as a key. Obviously finding a checksum would be in logarithmi= c time, >>> while the size of the tree would be possibly much bigger in the per= -inode >>> case. In per block group case we will have much smaller boundary of >>> number of checksum blocks stored. >>> >>> This and the fact that we would have to have at least one checksum = block >>> per inode (which would be wasteful in the case of small files) is m= aking per >>> block group solution much more viable. However the major disadvanta= ge of >>> per block group solution is that the checksum tree would create a s= ource of >>> contention when reading/writing from/to a different inodes in the s= ame block >>> group. This might be mitigated by having a worker thread per a rang= e of block >>> groups - but it might still be a bottleneck. >>> >>> Again we still have 4 Bytes in ext4_group_desc to store the pointer= to the >>> root of the tree. While the ext4_inode structure have 4Bytes of >>> i_obso_faddr but that's not enough. So we would have to figure out = where to >>> store it - we could possibly abuse i_block to store it along with t= he extent >>> nodes. >> >> I think(?) your purpose in using either a special inode or a btree t= o store the >> checksums is to avoid wasting checksum blocks on things that are alr= eady >> checksummed? I'm not sure that we'd save enough space to justify th= e extra >> processing. >> >> --D >> >>> File system scrub >>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D >>> >>> While this is certainly a feature which we want to have in both use= rspace >>> e2fsprogs and kernel I do not have any design notes at this stage. >>> >>> >>> >>> >>> I am sure that there are other possibilities and variants of those = design >>> ideas, but I think that this should be enough to have a discussion = started. >>> As I is not I think that the most viable option is d) that is, per = block >>> group checksum tree, which gives us enough flexibility while not be= ing too >>> complex solution. >>> >>> I'll try to update this description as it will be getting more conc= rete >>> structure and I hope that we will have some productive discussion a= bout >>> this at LSF. >>> >>> Thanks! >>> -Lukas >>> -- >>> To unsubscribe from this list: send the line "unsubscribe linux-ext= 4" in >>> the body of a message to majordomo@vger.kernel.org >>> More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" i= n the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html