[ Previous | Next | Contents | Home | Search ]
AIX Version 4.3 Kernel Extensions and Device Support Programming Concepts

Understanding Logical Volumes and Bad Blocks

The physical layer of the LVDD initiates all bad-block processing and isolates all of the decision making from the physical disk device driver. This happens so the physical disk device driver does not need to handle mirroring, which is the duplication of data transparent to the user.

Relocating Bad Blocks

The physical layer of the logical volume device driver (LVDD) checks each physical request to see if there are any known software-relocated bad blocks in the request. The LVDD determines if a request contains known software-relocated bad blocks by hashing the physical address. Then a hash chain of the LVDD defects directory is searched to see if any bad-block entries are in the address range of the request.

If bad blocks exist in a physical request, the request is split into pieces. The first piece contains any blocks up to the relocated block. The second piece contains the relocated block (the relocated address is specified in the bad-block entry) of the defects directory. The third piece contains any blocks after the relocated block to the end of the request or to the next relocated block. These separate pieces are processed sequentially until the entire request has been satisfied.

Once the I/O for the first of the separate pieces has completed, the iodone kernel service calls the LVDD physical layer's termination routine (specified in the b_done field of the buf structure). The termination routine initiates I/O for the second piece of the original request (containing the relocated block), and then for the third piece. When the entire physical operation is completed, the appropriate scheduler's policy routine (in the second layer of the LVDD) is called to start the next phase of the logical operation.

Detecting and Correcting Bad Blocks

If a logical volume is mirrored, a newly detected bad block is fixed by relocating that block. A good mirror is read and then the block is relocated using data from the good mirror. With mirroring, the user does not need to know when bad blocks are found. However, the physical disk device driver does log permanent I/O errors so the user can determine the rate of media surface errors.

When a bad block is detected during I/O, the physical disk device driver sets the error fields in the buf structure to indicate that there was a media surface error. The physical layer of the LVDD then initiates any bad-block processing that must be done.

If the operation was a nonmirrored read, the block is not relocated because the data in the relocated block is not initialized until a write is performed to the block. To support this delayed relocation, an entry for the bad block is put into the LVDD defects directory and into the bad-block directory on disk. These entries contain no relocated block address and the status for the block is set to indicate that relocation is desired.

On each I/O request, the physical layer checks whether there are any bad blocks in the request. If the request is a write and contains a block that is in a relocation-desired state, the request is sent to the physical disk device driver with safe hardware relocation requested. If the request is a read, a read of the known defective block is attempted.

If the operation was a read operation in a mirrored LP, a request to read one of the other mirrors is initiated. If the second read is successful, then the read is turned into a write request and the physical disk device driver is called with safe hardware relocation specified to fix the bad mirror.

If the hardware relocation fails or the device does not support safe hardware relocation, the physical layer of the LVDD attempts software relocation. At the end of each volume is a reserved area used by the LVDD as a pool of relocation blocks. When a bad block is detected and the disk device driver is unable to relocate the block, the LVDD picks the next unused block in the relocation pool and writes to this new location. A new entry is added to the LVDD defects directory in memory (and to the bad-block directory on disk) that maps the bad-block address to the new relocation block address. Any subsequent I/O requests to the bad-block address are routed to the relocation address.

Attention: Formatting a fixed disk deletes any data that may be on the disk. Format a fixed disk only when absolutely necessary and preferably after backing up all data on the disk.

If you need to format a fixed disk completely (including reinitializing any bad blocks), use the formatting function supplied by the diag command. (The diag command typically, but not necessarily, writes over all data on a fixed disk. Refer to the documentation that comes with the fixed disk to determine the effect of formatting with the diag command.)

Related Information

The buf structure.

The iodone kernel service.

Serial DASD Subsystem Device Driver in AIX Technical Reference: Kernel and Subsystems Volume 1.

The lvdd special file.

The write subroutine, readx subroutine.

The bread kernel service, bwrite block I/O kernel services.

[ Previous | Next | Contents | Home | Search ]