From Files to Sectors
The filesystem understands files, directories, and inodes. The physical disk understands only one thing: numbered sectors (or Logical Block Addresses, LBAs). The Block Layer is the translation engine between these two worlds.
When the filesystem needs to write data, it tells the Block Layer: "Write these memory pages to these disk blocks." The Block Layer packages this request into a kernel data structure called a bio (Block I/O). The bio struct is the fundamental unit of I/O in the Linux kernel. It contains:
bi_bdev: Which block device to target.bi_iter.bi_sector: The starting LBA (sector address) on the device.bi_opf: The operation flags — read, write, flush, discard.bi_io_vec: A list of memory page segments containing the data (a scatter-gather list).
Multiple bio structs can be chained together for large transfers. The Block Layer then hands these to the I/O scheduler for optimization before passing them to the device driver.
Merging: The First Optimization
Before scheduling, the Block Layer performs request merging. If three separate 4 KB writes arrive for consecutive sectors (LBA 100, LBA 101, LBA 102), the Block Layer merges them into a single 12 KB request. This reduces the number of commands sent to the hardware, which reduces overhead — especially on HDDs where each command has a fixed setup cost.
Before merging: After merging:
bio(LBA=100, 4KB) ─┐
bio(LBA=101, 4KB) ─┤──► bio(LBA=100, 12KB)
bio(LBA=102, 4KB) ─┘
Key insight: The Block Layer also handles plugging. When a process starts issuing I/O, the kernel "plugs" the queue — holding requests briefly to accumulate more, then "unplugs" to flush them all at once. This is analogous to letting a bathtub fill before draining rather than draining each drop individually.