[PATCH v2] erofs-utils: introduce the I/O manager

[Gao Xiang]

(+ Cc linux-erofs)

On 2024/6/5 11:32, Hongzhen Luo wrote:
> 
> 在 2024/6/5 10:55, Gao Xiang 写道:
>>
>>
>> On 2024/6/4 20:10, Hongzhen Luo wrote:
>>> Introduce the I/O manager to provide a more flexible way to specify
>>> the virtual storage.
>>>
>>> Signed-off-by: Hongzhen Luo <hongzhen at linux.alibaba.com>
>>> ---
>>> v2: Updated the handling of `.dev_close` being unimplemented in `erofs_dev_close` and so on.
>>> v1: https://lore.kernel.org/all/20240604093556.3883585-1-hongzhen@linux.alibaba.com/
>>
>> I'm unable to apply this patch.
>>
> Sorry about that, I overlooked developing based on the latest branch. I will send a v3 soon.
>>> ---
>>
>> ...
>>
>>>   +struct erofs_io_manager {
>>> +    int (*dev_open)(struct erofs_sb_info *sbi, const char *devname);
>>
>>     int (*open)(...);
>>
>>> +    int (*dev_open_ro)(struct erofs_sb_info *sbi, const char *dev);
>>
>>     int (*open_ro)(...);
>>
>>> +    int (*blob_open_ro)(struct erofs_sb_info *sbi, const char *dev);
>>
>>
>>     int (*blob_open_ro)(...);
>>
>>> +    int (*dev_read)(struct erofs_sb_info *sbi, int device_id,
>>> +        void *buf, u64 offset, size_t len);
>>
>>     int (*read)(...);
>>
>>> +    int (*dev_write)(struct erofs_sb_info *sbi, const void *buf,
>>> +        u64 offset, size_t len);
>>
>>     int (*write)(...);
>>
>>> +    int (*dev_fillzero)(struct erofs_sb_info *sbi, u64 offset,
>>> +        size_t len, bool padding);
>>     int (*fillzero)(...);
>>
>>> +    int (*dev_fsync)(struct erofs_sb_info *sbi);
>>
>>     int (*fsync)(...);
>>
>>> +    int (*dev_resize)(struct erofs_sb_info *sbi, erofs_blk_t nblocks);
>>
>>     int (*resize)(...);
>>
>>> +    void (*dev_close)(struct erofs_sb_info *sbi);
>>
>>     int (*close)(...);
>>
>>> +    void (*blob_closeall)(struct erofs_sb_info *sbi);
>>
>>     int (*blob_closeall)(...);
>>
>>> +    void *private_data;
>>
>> What does this mean?
>>
> Interface implementers can use it to store the data they wish to save. For example, in C++, this pointer can be used to store an object pointer, thereby accessing the data associated with the object.
> 
> Another option is to remove this pointer and let the implementer wrap the I/O manager into another object. Which strategy to adopt needs to be discussed.

I don't see how private_data is used in this patch and it's
unrelated to C++.  If it's unused, please remove it entirely.

Thanks,
Gao Xiang