Re: [f2fs-dev] [PATCH] f2fs: compress: fix zstd data corruption

[Chao Yu]

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.
> 
> Â Â Â Â max_len = COMPRESS_HEADER_SIZE + cc->clen;
> Â Â Â Â cc->nr_cpages = DIV_ROUND_UP(max_len, PAGE_SIZE);
> 
> Â Â Â Â cc->cpages = f2fs_kzalloc(sbi, sizeof(struct page *) *
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â 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.
> 
> Â Â Â Â ret = cops->compress_pages(cc);
> Â Â Â Â if (ret)
> Â Â Â Â Â Â Â Â goto out_vunmap_cbuf;
> 
> Â Â Â Â max_len = PAGE_SIZE * (cc->cluster_size - 1) - COMPRESS_HEADER_SIZE;
> 
> Â Â Â Â if (cc->clen > max_len) {
> Â Â Â Â Â Â Â Â ret = -EAGAIN;
> Â Â Â Â Â Â Â Â goto out_vunmap_cbuf;
> Â Â Â Â }
> 
> 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 decreaseÂthe 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,

> 
> 
> 2020ë 5ì 8ì (ê) ìì 11:48, Chao Yu <yuchao0@xxxxxxxxxx <mailto:yuchao0@xxxxxxxxxx>>ëì ìì:
> 
>     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)Â Â Â Â (\
>     Â Â Â Â (unsigned int)(isize) > (unsigned int)LZ4_MAX_INPUT_SIZE \
>     Â Â Â Â ? 0 \
>     Â Â Â Â : (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:
> 
>     ...
>     Â* @maxOutputSize: full or partial size of buffer 'dest'
>     Â*Â Â Â which must be already allocated
>     ...
>     int LZ4_compress_default(const char *source, char *dest, int inputSize,
>     Â Â Â Â 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,
>     >
>     > 2020ë 5ì 8ì (ê) ìì 10:17, Chao Yu <yuchao0@xxxxxxxxxx <mailto:yuchao0@xxxxxxxxxx>>ëì ìì:
>     >
>     >> 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@xxxxxxxxxx <mailto:daehojeong@xxxxxxxxxx>>
>     >> Signed-off-by: Chao Yu <yuchao0@xxxxxxxxxx <mailto:yuchao0@xxxxxxxxxx>>
>     >> ---
>     >>Â fs/f2fs/compress.c | 7 +++++++
>     >>Â 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)
>     >>Â Â Â Â Â Â Â Â Âreturn -EIO;
>     >>Â Â Â Â Â}
>     >>
>     >> +Â Â Â Â/*
>     >> +Â Â Â Â * there is compressed data remained in intermediate buffer due to
>     >> +Â Â Â Â * no more space in cbuf.cdata
>     >> +Â Â Â Â */
>     >> +Â Â Â Âif (ret)
>     >> +Â Â Â Â Â Â Â Âreturn -EAGAIN;
>     >> +
>     >>Â Â Â Â Âcc->clen = outbuf.pos;
>     >>Â Â Â Â Âreturn 0;
>     >>Â }
>     >> --
>     >> 2.18.0.rc1
>     >>
>     >>
>     >>
>     >> _______________________________________________
>     >> Linux-f2fs-devel mailing list
>     >> Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx <mailto:Linux-f2fs-devel@xxxxxxxxxxxxxxxxxxxxx>
>     >> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
>     >>
>     >
>