/proc

Summary by Bruce Alan Wynn

We've all known that it was there for quite some time, but how many of us ever even looked at the /proc filesystem?

In this presentation, Stephen described the wealth of process-specific information that can be found in the /proc filesystem. He explained why traditional UNIX tools may report strange statistics on objects in this filesystem-as Stephen says, "stat() tells lies."

The reason for these seeming inconsistencies is the nature of the /proc filesystem and its files. Although you can read the directory for the names of the files, the actual file "contents" are generated dynamically when you open the file for reading. Hence, subsequent commands are actually accessing different "files."

The Linux /proc filesystem is currently read-only. Therefore, system administrators can only use it to examine process statistics, but not change them. Tools that access these files display process information in textual form for ease of human consumption.

But what if you want to change these things and tune your system?

Stephen also spoke at length about using the sysctl() system call. Using sysctl(), some process information is writable, allowing tuning. Because sysctl() does not need to convert the process information into human-readable form (it is stored in well defined structures), access to the process information is faster than using the /proc filesystem directly.

In conclusion, Stephen pointed out that Linux 2.1 is continuing to develop the sysctl() interface by migrating the ability to configure sysctl() at compile time.

Originally published in ;login: Vol. 22, No.2, April 1997.