Design for storage centre

[bluemoon]

Hi,

How would you design the storage center?
Here is mine, I would like to have your review, see if I missed anything.
network storage and cache will be added on top of this.


I end up having four entities:
- IO Driver: Collection of block-size IO functions
- FS Driver: FS parsing facility
- IO Device: A physical device entity
- Drive: A partition, "drive letter".


IO driver:
Block-size IO, example are Ramdisk, ATA.

Code: Select all

typedef struct IODRIVER_S {
    uint32_t  name;
    void*     opaque;

    int (*init)  (struct IODRIVER_S* driver);
    int (*deinit)(struct IODRIVER_S* driver);
    int (*open)  (struct IODRIVER_S* driver, struct IODEVICE_S* device, const char* resource, int option);
    int (*close) (struct IODRIVER_S* driver, struct IODEVICE_S* device);
    int (*ioctl) (struct IODRIVER_S* driver, struct IODEVICE_S* device, int req, ...);
    int (*read)  (struct IODRIVER_S* driver, struct IODEVICE_S* device, void* buf, off_t block, int nblock);
    int (*write) (struct IODRIVER_S* driver, struct IODEVICE_S* device, void* buf, off_t block, int nblock);
} IODRIVER;

FS Driver:

Code: Select all

typedef struct FSDRIVER_S {
    uint32_t     name;
    unsigned int flags;
    void*        opaque;

    int (*probe)   (struct FSDRIVER_S* fs, IODEVICE* device, uint8_t* vbr);
    int (*open)    (struct FSDRIVER_S* fs, IODEVICE* device, FS_FILE* file, const char* path);
    int (*close)   (struct FSDRIVER_S* fs, IODEVICE* device, FS_FILE* file);
    int (*read)    (struct FSDRIVER_S* fs, IODEVICE* device, FS_FILE* file, void* buf, off_t offset, size_t nbytes);
    int (*write)   (struct FSDRIVER_S* fs, IODEVICE* device, FS_FILE* file, void* buf, off_t offset, size_t nbytes);
    int (*opendir) (struct FSDRIVER_S* fs, IODEVICE* device, FS_DIR *dir, const char* path);
    int (*readdir) (struct FSDRIVER_S* fs, IODEVICE* device, FS_DIR *dir, unsigned int index, FS_DIRENT* dirent );
    int (*closedir)(struct FSDRIVER_S* fs, IODEVICE* device, FS_DIR *dir);
    
    // int (*fstat) (struct FSDRIVER* fs, VFS_FILE* file, char* buf);
} FSDRIVER;

iodevice (physical disk) - block IO, work on sector numbers
- name (to identify which device of same type)
- driver:io
- driver-specific information used by io driver


drive - provide unified file operations
- name (use with mount)
- iodevice
- partition (LBA start, size)
- fs driver
- driver-specific information used by fs driver


The initialization sequence is:
- Register *all* IO driver and FS driver, I'm not doing on-demand to add extra complexity
- get a device type and token from device list (eg ATA:0, RAMDISK:2M~4M)
- Resolve an IO driver by device type
- Create an IO Device instant, the device can be accessed by sector IO now.
- Loop all partition for (handle multiple partition or single partition device):
1) Load VBR
2) Pass the VBR to each FS driver by calling fs->probe
3) fs->probe return true if such partition is supported
4) Create a Drive instant.

[Brendan]

Hi,

I like the idea of implementing things like partitions and archives as file systems.

For example, the device "/dev/hda" might contain an "MBR partitions" file system, and that file system might contain 4 files (hda0, hda1, hda2, hda3). One of these files might contain a FAT32 file system. The FAT32 file system might have a file "foo/bar.zip". The "foo/bar.zip" archive is a file system too.

Basically (for this example) you could do "cat foo/bar.zip/hello.txt" to display the file "hello.txt" from within the archive, from within the FAT32 file system, from within a partition, from the "/dev/hda" device.


Cheers,

Brendan