Linux Tomorrow

Under Linux, everything is managed as a file; even devices are accessed as files (in the /dev directory). Although you might think that "normal" files are either text or binary (or possibly device or pipe files), the /proc directory contains a stranger type: virtual files. These files are listed, but don't actually exist on disk; the operating system creates them on the fly if you try to read them.

Most virtual files always have a current timestamp, which indicates that they are constantly being kept up to date. The /proc directory itself is created every time you boot your box. You need to work as root to be able to examine the whole directory; some of the files (such as the process-related ones) are owned by the user who launched it. Although almost all the files are read-only, a few writable ones (notably in /proc/sys) allow you to change kernel parameters. (Of course, you must be careful if you do this.)

/proc directory organization

The /proc directory is organized in virtual directories and subdirectories, and it groups files by similar topic. Working as root, the ls /code/* command brings up something like this:

1     2432  3340  3715  3762  5441  815        devices      modules
129   2474  3358  3716  3764  5445  acpi       diskstats    mounts
1290  248   3413  3717  3812  5459  asound     dma          mtrr
133   2486  3435  3718  3813  5479  bus        execdomains  partitions
1420  2489  3439  3728  3814  557   dri        fb           self
165   276   3450  3731  39    5842  driver     filesystems  slabinfo
166   280   36    3733  3973  5854  fs         interrupts   splash
2     2812  3602  3734  4     6     ide        iomem        stat
2267  3     3603  3735  40    6381  irq        ioports      swaps
2268  326   3614  3737  4083  6558  net        kallsyms     sysrq-trigger
2282  327   3696  3739  4868  6561  scsi       kcore        timer_list
2285  3284  3697  3742  4873  6961  sys        keys         timer_stats
2295  329   3700  3744  4878  7206  sysvipc    key-users    uptime
2335  3295  3701  3745  5     7207  tty        kmsg         version
2400  330   3706  3747  5109  7222  buddyinfo  loadavg      vmcore
2401  3318  3709  3749  5112  7225  cmdline    locks        vmstat
2427  3329  3710  3751  541   7244  config.gz  meminfo      zoneinfo
2428  3336  3714  3753  5440  752   cpuinfo    misc

The numbered directories (more on them later) correspond to each running process; a special self symlink points to the current process. Some virtual files provide hardware information, such as /proc/cpuinfo, /proc/meminfo, and /proc/interrupts. Others give file-related info, such as /proc/filesystems or /proc/partitions. The files under /proc/sys are related to kernel configuration parameters, as we'll see.

The cat /proc/meminfo command might bring up something like this:

# cat /proc/meminfo
MemTotal:       483488 kB
MemFree:          9348 kB
Buffers:          6796 kB
Cached:         168292 kB
...several lines snipped...

If you try the top or free commands, you might recognize some of these numbers. In fact, several well-known utilities access the /proc directory to get their information. For example, if you want to know what kernel you're running, you might try uname -srv, or go to the source and type cat /proc/version. Some other interesting files include:

• /proc/apm: Provides information on Advanced Power Management, if it's installed.
• /proc/acpi: A similar directory that offers plenty of data on the more modern Advanced Configuration and Power Interface. For example, to see if your laptop is connected to the AC power, you can use cat /proc/acpi/ac_adapter/AC/state to get either "on line" or "off line."
• /proc/cmdline: Shows the parameters that were passed to the kernel at boot time. In my case, it contains root=/dev/disk/by-id/scsi-SATA_FUJITSU_MHS2040_NLA5T3314DW3-part3 vga=0x317 resume=/dev/sda2 splash=silent PROFILE=QuintaWiFi, which tells me which partition is the root of the filesystem, which VGA mode to use, and more. The last parameter has to do with openSUSE's System Configuration Profile Management.
• /proc/cpuinfo: Provides data on the processor of your box. For example, in my laptop, cat /proc/cpuinfo gets me a listing that starts with:

processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 8
model name      : Mobile AMD Athlon(tm) XP 2200+
stepping        : 1
cpu MHz         : 927.549
cache size      : 256 KB

This shows that I have only one processor, numbered 0, of the 80686 family (the 6 in cpu family goes as the middle digit): an AMD Athlon XP, running at less than 1GHz.

• /proc/loadav: A related file that shows the average load on the processor; its information includes CPU usage in the last minute, last five minutes, and last 10 minutes, as well as the number of currently running processes.
• /proc/stat: Also gives statistics, but goes back to the last boot.
• 
  
• /proc/uptime: A short file that has only two numbers: how many seconds your box has been up, and how many seconds it has been idle.
• /proc/devices: Displays all currently configured and loaded character and block devices. /proc/ide and /proc/scsi provide data on IDE and SCSI devices.
• /proc/ioports: Shows you information about the regions used for I/O communication with those devices.
• /proc/dma: Shows the Direct Memory Access channels in use.
• /proc/filesystems: Shows which filesystem types are supported by your kernel. A portion of this file might look like this:

nodev   sysfs
nodev   rootfs
nodev   bdev
nodev   proc
nodev   cpuset
...some lines snipped...
nodev   ramfs
nodev   hugetlbfs
nodev   mqueue
      ext3
nodev   usbfs
      ext2
nodev   autofs

The first column shows whether the filesystem is mounted on a block device. In my case, I have partitions configured with ext2 and ext3 mounted.

• /proc/mounts: Shows all the mounts used by your machine (its output looks much like /etc/mtab). Similarly, /proc/partitions and /proc/swaps show all partitions and swap space.
• 
  
• /proc/fs: If you're exporting filesystems with NFS, this directory has among its many subdirectories and files /proc/fs/nfsd/exports, which shows the file system that are being shared and their permissions.
• /proc/net: You can't beat this for network information. Describing each file in this directory would require too much space, but it includes /dev (each network device), several iptables (firewall) related files, net and socket statistics, wireless information, and more.

There are also several RAM-related files. I've already mentioned /proc/meminfo, but you've also got /proc/iomem, which shows you how RAM memory is used in your box, and /proc/kcore, which represents the physical RAM of your box. Unlike most other virtual files, /proc/kcore shows a size that's equal to your RAM plus a small overhead. (Don't try to cat this file, because its contents are binary and will mess up your screen.) Finally, there are many hardware-related files and directories, such as /proc/interrupts and /proc/irq, /proc/pci (all PCI devices), /proc/bus, and so on, but they include very specific information, which most users won't need.

What's in a process?

As I said, the numerical named directories represent all running processes. When a process ends, its /proc directory disappears automatically. If you check any of these directories while they exist, you will find plenty of files, such as:

attr             cpuset   fdinfo    mountstats  stat
auxv             cwd      loginuid  oom_adj     statm
clear_refs       environ  maps      oom_score   status
cmdline          exe      mem       root        task
coredump_filter  fd       mounts    smaps       wchan

Let's take a look at the principal files:

• cmdline: Contains the command that started the process, with all its parameters.
• cwd: A symlink to the current working directory (CWD) for the process; exe links to the process executable, and root links to its root directory.
• environ: Shows all environment variables for the process.
• fd: Contains all file descriptors for a process, showing which files or devices it is using.
• maps, statm, and mem: Deal with the memory in use by the process.
• stat and status: Provide information about the status of the process, but the latter is far clearer than the former.

These files provide several script programming challenges. For example, if you want to hunt for zombie processes, you could scan all numbered directories and check whether "(Z) Zombie" appears in the /status file. I once needed to check whether a certain program was running; I did a scan and looked at the /cmdline files instead, searching for the desired string. (You can also do this by working with the output of the ps command, but that's not the point here.) And if you want to program a better-looking top, all the needed information is right at your fingertips.

Tweaking the system: /proc/sys

/proc/sys not only provides information about the system, it also allows you to change kernel parameters on the fly, and enable or disable features. (Of course, this could prove harmful to your system -- consider yourself warned!)

To determine whether you can configure a file or if it's just read-only, use ls -ld; if a file has the "W" attribute, it means you may use it to configure the kernel somehow. For example, ls -ld /proc/kernel/* starts like this:

dr-xr-xr-x 0 root root 0 2008-01-26 00:49 pty
dr-xr-xr-x 0 root root 0 2008-01-26 00:49 random
-rw-r--r-- 1 root root 0 2008-01-26 00:49 acct
-rw-r--r-- 1 root root 0 2008-01-26 00:49 acpi_video_flags
-rw-r--r-- 1 root root 0 2008-01-26 00:49 audit_argv_kb
-r--r--r-- 1 root root 0 2008-01-26 00:49 bootloader_type
-rw------- 1 root root 0 2008-01-26 00:49 cad_pid
-rw------- 1 root root 0 2008-01-26 00:49 cap-bound

You can see that bootloader_type isn't meant to be changed, but other files are. To change a file, use something like echo 10 >/proc/sys/vm/swappiness. This particular example would allow you to tune the virtual memory paging performance. By the way, these changes are only temporary, and their effects will disappear when you reboot your system; use sysctl and the /etc/sysctl.conf file to effect more permanent changes.

Let's take a high-level look at the /proc/sys directories:

• debug: Has (surprise!) debugging information. This is good if you're into kernel development.
• dev: Provides parameters for specific devices on your system; for example, check the /dev/cdrom directory.
• fs: Offers data on every possible aspect of the filesystem.
• kernel: Lets you affect the kernel configuration and operation directly.
• net: Lets you control network-related matters. Be careful, because messing with this can make you lose connectivity!
• vm: Deals with the VM subsystem.

Conclusion

The /proc special directory provides full detailed information about the inner workings of Linux and lets you fine-tune many aspects of its configuration. If you spend some time learning all the possibilities of this directory, you'll be able to get a more perfect Linux box. And isn't that something we all want?