Re: [PATCH] f2fs: compress: support zstd compress algorithm

[Chao Yu <yuchao0@huawei.com>]

On 2020/3/3 1:50, Eric Biggers wrote:
> On Fri, Feb 28, 2020 at 07:14:56PM +0800, Chao Yu wrote:
>> Add zstd compress algorithm support, use "compress_algorithm=zstd"
>> mountoption to enable it.
>>
>> Signed-off-by: Chao Yu <yuchao0@huawei.com>
>> ---
>>  Documentation/filesystems/f2fs.txt |   4 +-
>>  fs/f2fs/Kconfig                    |   9 ++
>>  fs/f2fs/compress.c                 | 151 +++++++++++++++++++++++++++++
>>  fs/f2fs/f2fs.h                     |   2 +
>>  fs/f2fs/super.c                    |   7 ++
>>  include/trace/events/f2fs.h        |   3 +-
>>  6 files changed, 173 insertions(+), 3 deletions(-)
>>
>> diff --git a/Documentation/filesystems/f2fs.txt b/Documentation/filesystems/f2fs.txt
>> index 4eb3e2ddd00e..b1a66cf0e967 100644
>> --- a/Documentation/filesystems/f2fs.txt
>> +++ b/Documentation/filesystems/f2fs.txt
>> @@ -235,8 +235,8 @@ checkpoint=%s[:%u[%]]     Set to "disable" to turn off checkpointing. Set to "en
>>                         hide up to all remaining free space. The actual space that
>>                         would be unusable can be viewed at /sys/fs/f2fs/<disk>/unusable
>>                         This space is reclaimed once checkpoint=enable.
>> -compress_algorithm=%s  Control compress algorithm, currently f2fs supports "lzo"
>> -                       and "lz4" algorithm.
>> +compress_algorithm=%s  Control compress algorithm, currently f2fs supports "lzo",
>> +                       "lz4" and "zstd" algorithm.
>>  compress_log_size=%u   Support configuring compress cluster size, the size will
>>                         be 4KB * (1 << %u), 16KB is minimum size, also it's
>>                         default size.
>> diff --git a/fs/f2fs/Kconfig b/fs/f2fs/Kconfig
>> index f0faada30f30..bb68d21e1f8c 100644
>> --- a/fs/f2fs/Kconfig
>> +++ b/fs/f2fs/Kconfig
>> @@ -118,3 +118,12 @@ config F2FS_FS_LZ4
>>  	default y
>>  	help
>>  	  Support LZ4 compress algorithm, if unsure, say Y.
>> +
>> +config F2FS_FS_ZSTD
>> +	bool "ZSTD compression support"
>> +	depends on F2FS_FS_COMPRESSION
>> +	select ZSTD_COMPRESS
>> +	select ZSTD_DECOMPRESS
>> +	default y
>> +	help
>> +	  Support ZSTD compress algorithm, if unsure, say Y.
>> diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
>> index bd3ea01db448..c8e1175eaf4e 100644
>> --- a/fs/f2fs/compress.c
>> +++ b/fs/f2fs/compress.c
>> @@ -11,6 +11,7 @@
>>  #include <linux/backing-dev.h>
>>  #include <linux/lzo.h>
>>  #include <linux/lz4.h>
>> +#include <linux/zstd.h>
>>  
>>  #include "f2fs.h"
>>  #include "node.h"
>> @@ -291,6 +292,151 @@ static const struct f2fs_compress_ops f2fs_lz4_ops = {
>>  };
>>  #endif
>>  
>> +#ifdef CONFIG_F2FS_FS_ZSTD
>> +#define F2FS_ZSTD_DEFAULT_CLEVEL	1
>> +
>> +static int zstd_init_compress_ctx(struct compress_ctx *cc)
>> +{
>> +	return 0;
>> +}
>> +
>> +static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
>> +{
>> +}
>> +
>> +static int zstd_compress_pages(struct compress_ctx *cc)
>> +{
>> +	ZSTD_parameters params;
>> +	ZSTD_CStream *stream;
>> +	ZSTD_inBuffer inbuf;
>> +	ZSTD_outBuffer outbuf;
>> +	void *workspace;
>> +	unsigned int workspace_size;
>> +	int src_size = cc->rlen;
>> +	int dst_size = src_size - PAGE_SIZE - COMPRESS_HEADER_SIZE;
>> +	int ret;
>> +
>> +	params = ZSTD_getParams(F2FS_ZSTD_DEFAULT_CLEVEL, src_size, 0);
>> +	workspace_size = ZSTD_CStreamWorkspaceBound(params.cParams);
>> +
>> +	workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
>> +					workspace_size, GFP_NOFS);
>> +	if (!workspace)
>> +		return -ENOMEM;
>> +
>> +	stream = ZSTD_initCStream(params, 0,
>> +					workspace, workspace_size);
> 
> Why is this allocating the memory for every compression operation, instead of
> ahead of time in ->init_compress_ctx()?

Actually, zstd decompression flow needs workspace like its compression flow,
however I realized that we don't have related callback interfaces in decompress
path, so I just add all steps of {,de}compression into .{,de}compress_pages()
functions.

In order to not break callback function fwk of compression, how about adding
{init,destroy}_decompress_ctx() callback interfaces, and relocating
initialization and destroy step into correct callback functions?

Thanks,

> 
> - Eric
> .
>