Date:   Fri, 8 May 2020 15:51:54 +0900
From:   Daeho Jeong <daeho43@...il.com>
To:     Chao Yu <yuchao0@...wei.com>
Cc:     jaegeuk@...nel.org, Daeho Jeong <daehojeong@...gle.com>,
        linux-kernel@...r.kernel.org,
        linux-f2fs-devel@...ts.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH] f2fs: compress: fix zstd data corruption

Great, thanks for checking~! :)


2020년 5월 8일 (금) 오후 3:42, Chao Yu <yuchao0@...wei.com>님이 작성:
>
> On 2020/5/8 11:30, Daeho Jeong wrote:
> > I am a little bit confused.
> >
> > In compress_log=2 (4 pages),
> >
> > Every compression algorithm will set the cc->nr_cpages to 5 pages like below.
> >
> >         max_len = COMPRESS_HEADER_SIZE + cc->clen;
> >         cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
> >
> >         cc->cpages = f2fs_kzalloc(sbi, sizeof(struct page *) *
> >                                         cc->nr_cpages, GFP_NOFS);
> >
> > And call cops->compress_pages(cc) and the returned length of the compressed data will be set to cc->clen for every case.
> > And if the cc->clen is larger than max_len, we will give up compression.
> >
> >         ret = cops->compress_pages(cc);
> >         if (ret)
> >                 goto out_vunmap_cbuf;
> >
> >         max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
> >
> >         if (cc->clen > max_len) {
> >                 ret = -EAGAIN;
> >                 goto out_vunmap_cbuf;
> >         }
> >
> > So, with your patch, we will just use 3 pages for ZSTD and 5 pages for LZO and LZ4 now.
> > My question was whether it is also possible to decrease the compression buffer size for LZO and LZ4 to 3 pages like ZSTD case.
> > I was just curious about that. :)
> I guess we can change LZ4 as we did for ZSTD case, since it supports partially
> compression:
>
> - lz4_compress_pages
>  - LZ4_compress_default
>   - LZ4_compress_fast
>    - LZ4_compress_fast_extState
>     if (maxOutputSize < LZ4_COMPRESSBOUND(inputSize))
>      - LZ4_compress_generic(..., limitedOutput, ...)
>       - if (outputLimited && boundary_check_condition) return 0;
>
> And for LZO case, it looks we have to keep to allocate 5 pages for worst
> compression case as it doesn't support partially compression as I checked.
>
> Thanks,
>
> >
> >
> > 2020년 5월 8일 (금) 오전 11:48, Chao Yu <yuchao0@...wei.com <mailto:yuchao0@...wei.com>>님이 작성:
> >
> >     Hi Daeho,
> >
> >     On 2020/5/8 9:28, Daeho Jeong wrote:
> >     > Hi Chao,
> >     >
> >     > IIUC, you are trying not to use ZSTD_compressBound() to save the memory
> >     > space. Am I right?
> >     >
> >     > Then, how about LZ4_compressBound() for LZ4 and lzo1x_worst_compress() for
> >     > LZO?
> >
> >     Oops, it looks those limits were wrongly used...
> >
> >     #define LZ4_COMPRESSBOUND(isize)        (\
> >             (unsigned int)(isize) > (unsigned int)LZ4_MAX_INPUT_SIZE \
> >             ? 0 \
> >             : (isize) + ((isize)/255) + 16)
> >
> >     #define lzo1x_worst_compress(x) ((x) + ((x) / 16) + 64 + 3 + 2)
> >
> >     Newly calculated boundary size is larger than target buffer size.
> >
> >     However comments on LZ4_compress_default() said:
> >
> >     ...
> >      * @maxOutputSize: full or partial size of buffer 'dest'
> >      *      which must be already allocated
> >     ...
> >     int LZ4_compress_default(const char *source, char *dest, int inputSize,
> >             int maxOutputSize, void *wrkmem);
> >
> >     And @out_len in lzo1x_1_compress() was passed as an output parameter to
> >     pass length of data that compressor compressed into @out buffer.
> >
> >     Let me know if I missed sth.
> >
> >     Thannks,
> >
> >     > Could we save more memory space for these two cases like ZSTD?
> >     > As you know, we are using 5 pages compression buffer for LZ4 and LZO in
> >     > compress_log_size=2,
> >     > and if the compressed data doesn't fit in 3 pages, it returns -EAGAIN to
> >     > give up compressing that one.
> >     >
> >     > Thanks,
> >     >
> >     > 2020년 5월 8일 (금) 오전 10:17, Chao Yu <yuchao0@...wei.com <mailto:yuchao0@...wei.com>>님이 작성:
> >     >
> >     >> During zstd compression, ZSTD_endStream() may return non-zero value
> >     >> because distination buffer is full, but there is still compressed data
> >     >> remained in intermediate buffer, it means that zstd algorithm can not
> >     >> save at last one block space, let's just writeback raw data instead of
> >     >> compressed one, this can fix data corruption when decompressing
> >     >> incomplete stored compression data.
> >     >>
> >     >> Signed-off-by: Daeho Jeong <daehojeong@...gle.com <mailto:daehojeong@...gle.com>>
> >     >> Signed-off-by: Chao Yu <yuchao0@...wei.com <mailto:yuchao0@...wei.com>>
> >     >> ---
> >     >>  fs/f2fs/compress.c | 7 +++++++
> >     >>  1 file changed, 7 insertions(+)
> >     >>
> >     >> diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
> >     >> index c22cc0d37369..5e4947250262 100644
> >     >> --- a/fs/f2fs/compress.c
> >     >> +++ b/fs/f2fs/compress.c
> >     >> @@ -358,6 +358,13 @@ static int zstd_compress_pages(struct compress_ctx
> >     >> *cc)
> >     >>                 return -EIO;
> >     >>         }
> >     >>
> >     >> +       /*
> >     >> +        * there is compressed data remained in intermediate buffer due to
> >     >> +        * no more space in cbuf.cdata
> >     >> +        */
> >     >> +       if (ret)
> >     >> +               return -EAGAIN;
> >     >> +
> >     >>         cc->clen = outbuf.pos;
> >     >>         return 0;
> >     >>  }
> >     >> --
> >     >> 2.18.0.rc1
> >     >>
> >     >>
> >     >>
> >     >> _______________________________________________
> >     >> Linux-f2fs-devel mailing list
> >     >> Linux-f2fs-devel@...ts.sourceforge.net <mailto:Linux-f2fs-devel@...ts.sourceforge.net>
> >     >> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
> >     >>
> >     >
> >

Powered by blists - more mailing lists