Exploring procfs

In summary, all linux's processes can be found in /proc folder. This folder is of procfs type which, like I said before, is a virtual filesystem and most of its file descriptors point to in-memory data. This is why if you run a ls /proc -l you'll notice that most files and folders are of size 0.

ls -l /proc
dr-xr-xr-x.  9 root           root                         0 Sep 25 22:10 1
dr-xr-xr-x.  9 root           root                         0 Oct  1 10:38 10
dr-xr-xr-x.  9 root           root                         0 Oct  1 12:46 101
dr-xr-xr-x.  9 root           root                         0 Oct  1 12:46 102
...

Inside '/proc' there is one folder for each process running with its pid as name. So I opened one of the folders to see what I could learn about a running process just by reading these filed.

ls -l /proc/<pid>
total 0
dr-xr-xr-x. 2 binpipe binpipe 0 Sep 28 23:15 attr
-rw-r--r--. 1 root   root   0 Oct  1 10:46 autogroup
-r--------. 1 root   root   0 Oct  1 10:46 auxv
-r--r--r--. 1 root   root   0 Sep 28 23:15 cgroup
--w-------. 1 root   root   0 Oct  1 10:46 clear_refs
-r--r--r--. 1 root   root   0 Sep 28 22:41 cmdline
-rw-r--r--. 1 root   root   0 Oct  1 10:46 comm
-rw-r--r--. 1 root   root   0 Oct  1 10:46 coredump_filter
...

Ok, now I have a bunch of files like autogroup, gid_map and maps that I have no idea what they're for. A good starting point would be checking for their documentation. But why on earth shouldn't I just open them?

So I started looping through the files one by one and most of them were completely unreadable for me, until I ran into the golden pot:

cat /proc/<pid>/status
Name:	chrome
State:	S (sleeping)
Tgid:	3054
Ngid:	0
Pid:	3054
PPid:	2934
TracerPid:	0
Uid:	1000	1000	1000	1000
Gid:	1000	1000	1000	1000
FDSize:	64
Groups:	10 1000 1001
VmPeak:	 1305996 kB
VmSize:	 1232520 kB
...

This is great! Finally something human readable. It contains general data about the process, like its state, memory usage and owner. But is this all I need?

Not satisfied with '/proc' file exploration, I decided to run ps against strace to see if it's accessing any of the files I found.

1	strace -o ./strace_log ps aux

Strace returns all system calls executed by a program. So I filter strace result by 'open' system call and as I suspected (maybe I didn't) the files being open by operating system were the same I first checked:

cat ./strace_log | grep open
[...]
open("/proc/1/stat", O_RDONLY) = 6
open("/proc/1/status", O_RDONLY) = 6
[...]
open("/proc/2/stat", O_RDONLY) = 6
open("/proc/2/status", O_RDONLY) = 6
open("/proc/2/cmdline", O_RDONLY) = 6
open("/proc/3/stat", O_RDONLY) = 6
open("/proc/3/status", O_RDONLY) = 6
open("/proc/3/cmdline", O_RDONLY) = 6
[...]

Ok, so we have stat, status and cmdline files to check, now all we need to do is to parse this and extract what we need.

The code

The implementation turned out to be fairly simple and it comes down to reading files and display its content in an organized matter.

Process data structure

We want to display our data in a tabular way; where each process is a record on this table. Let's take the following class as one of our table records:

class ProcessData
  attr_reader :pid
  attr_reader :name
  attr_reader :user
  attr_reader :state
  attr_reader :rss

  def initialize pid, name, user, state, rss
    @pid = pid
    @name = parse_name name
    @user = user
    @state = state
    @rss = rss
  end
end

Finding Pid's for running processes

Take into account what we know so far:

/proc folder contains sub-folders with all processes
All process folders have their pid as name

So gathering a list of all current pids should be easy:

def get_current_pids
  pids = []
  Dir.foreach("/proc") { |d|
    if is_process_folder?(d)
      pids.push(d)
    end
  }
  return pids
end

In order to be a valid process folder it must fulfill two requirements:

It's a folder (duh?)
It's name contains only number (this is why we have to cast folder name to int)

1
2
3

def is_process_folder? folder
  File.directory?("/proc/#{folder}") and (folder.to_i != 0)
end

Extracting process data

Now that we know every pid in the system we should create a method that exposes data from /proc/<pid>/status for any of them.

But first, lets analyze the file.

cat /proc/<pid>/status
Name:	chrome
State:	S (sleeping)
...
Uid:	1000	1000	1000	1000

This file is organized in the following way: Key:\t[values]. This means that for every piece of data in this file we can follow this same pattern to extract it. However, some lines will have an individual value and others will have a list of values (like Uid)

def get_process_data pid
  proc_data = {}
  File.open("/proc/#{pid}/status") { |file|
    begin
      while line = file.readline
        data = line.strip.split("\t")
        key = data.delete_at(0).downcase
        proc_data[key] = data
      end
      file.close
    rescue EOFError
      file.close
    end
  }
  return proc_data
end

The method above results in the following structure:

get_process_data 2917
=> {"name:"=>["chrome"],
 "state:"=>["S (sleeping)"],
 "tgid:"=>["2917"],
 "ngid:"=>["0"],
 "pid:"=>["2917"],
 "ppid:"=>["1"],
 "tracerpid:"=>["0"],
 "uid:"=>["1000", "1000", "1000", "1000"],
 ...

Reading user data

User uid and name association is kept in /etc/passwd file, so in order to show the correct username we must also read this file and parse it.

For the sake of simplicity, let's just read the whole file and save it in a Hash with key as Uid and value as name.

def get_users
  users = {}
  File.open("/etc/passwd", "r") { |file|
    begin
      while line = file.readline
        data = line.strip.split(":")
        users[data[2]] = data[0]
      end
      file.close
    rescue EOFError
      file.close
    end
  }
  return users
end

Creating process records

So far we have found the pids in the system, read the status file and extracted the data. What we have to do now is to filter and organize this data into a single record that will be presented to the user.

current_processes = filesystem.get_current_pids

current_processes.each { |p|
  process = create_process p
  puts "#{process.name}\t#{process.user}\t#{process.state}\t#{process.command}\t#{process.rss}"
}

def create_process pid
  data = get_process_data pid

  name = data["name:"][0]
  user_id = data["uid:"][0]
  state = data["state:"][0]

  if data["vmrss:"] != nil
    rss = data["vmrss:"][0]
  end

  user = get_users[user_id]

  return ProcessData.new(pid, name, user, state, rss)
end

The reason why we get VMRss value is because we want to check resident memory values, this means, only what's stored in the physical memory and not what's sitting in our disk.

Extra (formatting)

You can format ProcessData text in a tabular way to get a prettier output.

format="%6s\t%-15s\t%-10s\t%-10s\t%-10s\n"
printf(format, "PID", "NAME", "USER", "STATE", "MEMORY")
printf(format, "------", "---------------", "----------", "----------", "----------")
current_processes.each { |p|
  process = create_process p
  printf(format, process.pid, process.name, process.user, process.state, process.rss)
}

Result:

PID	NAME           	USER      	STATE     	MEMORY    
------	---------------	----------	----------	----------
   1	systemd        	root      	S (sleeping)	    8444 kB
   2	kthreadd       	root      	S (sleeping)	          
   3	ksoftirqd/0    	root      	S (sleeping)	          
   ...

Conclusion

There is a lot of information that you can find under /proc folder. This post only covers basic data like name, state and resident memory. But if you dig deep into those files you will find a lot more, like memory mapping and CPU usage.

BINPIPE | blog

Search Posts on Binpipe Blog

Understanding /proc