How mmap Turns Files into Memory: A Deep Dive into Virtual Memory

1. One‑sentence summary of mmap

mmap lets a process treat a portion of a file as if it were ordinary memory by mapping the file into physical memory and the process's virtual address space.

2. Virtual memory and virtual address space

Virtual address space is the set of addresses a process sees; it is a remapping of all physical addresses allocated (or to be allocated) to that process. The term “virtual memory” is often used interchangeably, though it is technically a space of addresses rather than real memory.

2.1 Principles of virtual space

2.1.1 Physical memory

Physical addresses are not necessarily contiguous; they include DRAM and I/O registers. Modern CPUs map I/O registers into physical memory via the PCI bus, allowing memory‑based I/O operations.

2.1.2 The three buses

Address bus – transfers addresses.

Data bus – transfers data.

Control bus – transfers commands.

The CPU sends a read command on the control bus, the address on the address bus, the MMU translates the virtual address, and the memory returns data via the data bus.

2.1.3 Virtual‑to‑physical address translation

Virtual addresses consist of a page (or segment) number and an offset. The MMU looks up the page/segment number in a page/segment table to obtain the corresponding physical page number, then combines it with the offset to form the physical address.

Storage methods

Page‑based storage: the page number indexes a page table to obtain a physical page number; physical address = page number × page size + offset.

Segment‑based storage: the segment number indexes a segment table to get the segment base address; physical address = segment base + offset.

Segment‑page hybrid combines both methods.

3. mmap mapping

From a coder’s perspective, mmap abstracts away the virtual address space and MMU details: the process simply reads and writes memory, and the OS handles the underlying file‑to‑memory mapping.

When a file is mmap‑ed, the process accesses memory directly; any modifications are reflected back to the disk.

The virtual space provides a contiguous address range.

Before access, the address points to unmapped pages.

When the process reads, the OS allocates physical pages (page‑fault handling).

The allocated pages are initially empty, then the file data is copied into them.