UNIX Filesystems: Evolution, Design, and Implementation

Traditional UNIX Locking Primitives

This section examines the earlier uni-processor (UP) UNIX synchronization primitives starting with 5th Edition UNIX and going up to SVR4.0. Over this twenty-year time period, the implementation stayed remarkably similar. As noted in his book Lions Commentary on UNIX 6th Edition-with Source Code [LION96], John Lions notes that the early mechanisms for handling critical sections of code were “totally inappropriate in a multi-processor system.”

As mentioned earlier, in UP UNIX implementations, the kernel needed to protect data structures in the case when a process went to sleep or when handling interrupts. The reasons a process might sleep include: waiting for I/O, waiting for a lock owned by another process, or giving up the CPU to another process after using up its timeslice.

If a process needs to access some resource such as a buffer cache buffer that is currently in use, it will issue a sleep() call specifying the address of the resource it requires. A swtch() call is made to relinquish control of the CPU, allowing another process to run. For example, to wait on a busy buffer, the following code sequence is made:

if (bp->b_flags & B_BUSY) {
    bp->b_flags |= B_WANTED;
    sleep(bp, PRIBIO);
}

The address of the structure on which the process is waiting (called the wait channel) is stored in the p_wchan field of the proc structure. The priority argument passed to sleep() will be described in more detail later in the chapter. Note for now though that ...