2

I am trying to start a php (under apache) process (by calling the apache from a browser), that will survive shutting down the apache server (sudo service apache2 stop). Even when I make sure the created process has no parent (parent 1), and has its own session, still, somehow, the process is died when I stop the apache (or restart the apache)

I created a test.php file:

<?php
exec('setsid nohup sleep 1000 > /dev/null 2>/dev/null &');
?>

When running doing HTTP GET to this test.php, indeed we get an immediate OK response, and the process still lives. But, when we do:

sudo service apache2 stop

The sleep process dies. How can someone kill a process when the process doesn't belong to its group or session, and when the process is not a child?

danronmoon
  • 3,814
  • 5
  • 34
  • 56
Nathan B
  • 1,625
  • 1
  • 17
  • 15
  • 1
    _"I am trying to start a process, from php (+apache)"_ - Are you calling your script via the browser, through Apache? – M. Eriksson Aug 06 '18 at 11:20
  • That's right. Simple apache server, that invokes a php script that I wrote in the question. – Nathan B Aug 06 '18 at 11:21
  • Is cron an option? You could just check every minute if e.g. there are entries inside a DB table which then are processed. If it would be about files, imagine using along [this answer](https://unix.stackexchange.com/a/24955) – Tobias F. Aug 06 '18 at 11:27
  • Why cron? Why DB? I just want a simple process (orphan?) that will stay alive. As simple is that. I don't understand how apache has the right/privilege to kill the process. Maybe the apache saves a list of PIDs of spawed processes by the php, and during shutdown kill them one by one? – Nathan B Aug 06 '18 at 11:33
  • Try to create a process, fork it and kill the parent. There shouldn't be any connection to Apache then. I remember I used this method in a similar circumstance (can't remember exactly what). Basically you can tell which processes were spawned by Apache by tracking their `ppid`, but if the parent no longer exists you break the chain. – Havenard Aug 07 '18 at 06:20
  • @Havenard, can you please explain how to modify the code above to accomplish what you suggested? – Nathan B Aug 07 '18 at 08:09
  • @NadavB Well I did it in the process itself, because it was a program I made. I'm not sure how to achieve this other way. – Havenard Aug 07 '18 at 16:28
  • I faced similar issue in past, and end up using gearman. – Akshay Vanjare Aug 08 '22 at 11:55
  • You can create a cron job or a system service which runs in the background? – Dimitar Aug 10 '22 at 11:45
  • does [this](https://serverfault.com/questions/531641/when-how-does-apache-kill-a-child-process-that-you-spawn-in-php/531644#531644) answer your question? – Lety Aug 10 '22 at 15:04

6 Answers6

2

Apache will probably have a list of the spawned processes and kills them individually, and not as a group. In that case, all processes in the list will be kill(2)ed. But see below next paragraph for a possibility.

Look at the man pages for the kill(2) system call. In the ERRORS section, the only possibilities to fail are:

  • EINVAL, meaning an invalid signal number has been passed. Doesn't apply here.

  • ESRCH, the process (or process group) doesn't exist. Doesn't apply also.

  • EPERM, you don't have permission to send a signal. This applies here, but the only processes (and this has nothing to do with process hierarchies or parental relationships) you are allowed to send signals to are the processes that run with real/saved user id equals the effective user id of the sender process. So, as Apache has a registry of all the processes it launches, it is normal that it is able to kill the process.

Anyway, have you tried to create a process, from that process create a subprocess, and execute the setsid in the grandchild subprocess? That way, there's no chance for the Apache process to have it registered in the list of spawned processes. I have not tried that, but it could work.

From the FreeBSD kill(2) manual page:

For a process to have permission to send a signal to a process designated by pid, the user must be the super-user, or the real or saved user ID of the receiving process must match the real or effective user ID of the sending process. A single exception is the signal SIGCONT, which may always be sent to any process with the same session ID as the sender.

(emphasis is mine) in linux, it's almost the same, except

... (if the sending process) have the CAP_KILL capability in the user namespace of the target process...

but this doesn't apply here.

Luis Colorado
  • 10,974
  • 1
  • 16
  • 31
  • I've tried modifiying the php exec() command to run a .sh script that did the "setsid nohup sleep 1000" (2 levels), but I didn't try to create a process, that creates another subprocess and then create another subprocess with setsid. Are so many levels necessary? Aren't two levels enough? – Nathan B Aug 07 '18 at 08:19
1

I'm curious on why the apache server needs to be shut down, but im inclined to say you might try a different strategy altogether using a containerized solution. Exposing an nginx docker container that can trigger your php process would likely be more stable than using Apache, as the docker daemon always runs as root. I think it depends on the specific needs of your use case though, so explaining why you need to shut down Apache might get you better answers.

Isaac Ray
  • 1,351
  • 9
  • 17
  • There are many reasons for wanting to do this. Example: I have a page on my file server that shows me file system, ram and bandwidth in real time (with lag). If somebody is abusing my limited and non-free resources, I fight back by restarting the server until they stop. I'm planing to add a text_entry in the restart page to allow me to configure the seconds apache must wait to be started again. – Hatoru Hansou Sep 02 '22 at 03:38
0

List all active processes (e.g. ps aux). The processes, which were created by Apache run as user www-data.

When you stop Apache service, I suspect they will be terminated based on that or some internal list. apache2.service stop calls apachectl -k graceful-stop. This will SIGTERM all "child processes". Unfortunately, I was unable to find the exact place in the code. Maybe someone could do that to verify the hypothesis.

A solution:

Run the process as another user. How you go about that depends on your case. You will have to define some kind of interface.
For example, you could have another process listen on localhost or use a Unix domain socket. The same will be achieved by using gearman, as @Akshay Vanjare pointed out in the comments.
Your PHP script can then call to the interface.

No solutions:
  • It is a bad idea to control a daemon this way, see this answer.
  • As user www-data you cannot start a process as another password protected user, because su asks for pass and switching UID is not a good idea.
  • For testing, I set up a user with empty password and used su - NEW_USER -c "COMMAND" in PHP script. Do not do this in any system you care about. It is insecure. Very insecure. And you will have a new process every time the script is called. You would have to take care to kill it.
Further thoughts:
  • I also tried a few alternatives to "nohup" the command, such as daemonizing, fork(), disown etc. They did not work for me.

  • I did not try hard in the new process to ignore SIGTERM. maybe it is possible to solve the question that way, too.

  • To me, it makes sense for Apache to do (aggressive) clean up when it is stopped. It is behavior I would expect from a web browser, which has to handle a lot of children.

Paul Smith
  • 299
  • 2
  • 13
0

First about my answer, it's not a good idea to control dameon from the web.

Once this said, you can have a php script that write a flag or somthing like that either on a database, a file, redis, etc.

On the other side make a PHP script that you schedule with cron for looking for the flag. If found, the script can start a PHP daemon that will detach. Pay attention to running the cron script and the PHP daemon with secured user & rights.

But once again be carefull with security concern.

Alaindeseine
  • 3,260
  • 1
  • 11
  • 21
0

From the PHP developer perspective I don't think there is a way to achieve that.

Or you can implement a queue and worker to do your tasks/executions. and for that you can either use Redis or any other database.

https://laravel.com/docs/9.x/queues

0

You can write a service that then you can control with sudo service start|stop|restart

You can use visudo to edit sudoers file to allow www-data to use sudo to run a specific sh file, without being prompted by password. This sh file will contain the line service your_service start.

Example: I wanted to be able to restart apache2 from a client side ajax request.

Context: file server with no screen/keyboard attached. I have SSH setup, but I don't want to launch a client in my laptop. I protect the restart page with apache password protected directory, along with other administrative pages. For example: I have a page to restart/shutdown the whole system.

I did this:

Register a service to run with sudo service apache_restarter_service start

This service file calls a simple sh script, a two lines script: sleep 1 second and then call service apache2 restart.

Use visudo to edit sudoers file. Allow user www-data to use sudo to run a sh script that calls service apache_restarter_service start.

The service file is used only to start a process with root privileges, and it doesn't need to be long lived, so no need for Restart=always in service definition. This process cannot be killed by apache and apache won't even know about it. It will only know that the sh script returned 0.

Example of changes to sudoers file:

www-data ALL=(ALL) NOPASSWD:/opt/apache_restarter/from_php.sh

from_php.sh only calls service apache_restarter_service start. No need for sudo here, because your php script already used sudo when calling from_php.sh.

from_php.sh is needed here, because you don't want php to call service directly. Because a BUG in, or an ATTACK against, your script may do harm to your server. This way, we only authorized from_php.sh in sudoers.

The php script must call from_php.sh with sudo. You can use exec("sudo /opt/apache_restarter/from_php.sh"). An & at the end is not needed, because apache won't be restarted immediately.

Example of service definition file apache_restarter_service.service:

[Unit]
Description=Apache restarter one shot
After=multi-user.target
StartLimitIntervalSec=0

[Service]
Type=simple
Restart=no
User=root
ExecStart=/opt/apache_restarter/run.sh

[Install]
WantedBy=multi-user.target

To avoid the service to run once the system is restarter. No harm really, but also useless. You can have the service always disabled and enable it in from.php script before starting it. Or just tune the service file to not run one time at system start.

You can tune StartLimitIntervalSec=0 to your needs.

You can tune User=root to a safer, less privileged user. In my case I don't care but maybe you do. You may want to use www-data. I don't know why that may not work, but I didn't tried to use www-data so far.

You can write run.sh in many ways and to do different tasks. The simpler of which would be to simply wait 1 second and then restart apache with service apache2 restart.

While my interest is to be able to restart apache from client side, this also applies to spawn a process that won't be terminated when apache is stopped/restarted.

I also tried with nohup, &, setsid and whatnot. I will save people some time. Nothing works. Apache should be terminating all child processes as already suggested.


Update 2022-12-04

I got into the following situation: need to start a python3 web server into alternate port to provide file server statistics.

Why? This is the concept, if nobody interested in the statistics (iostat, ifstat, etc) then the stats_server.py is not even running (wasting RAM/CPU). So, having stats_server.py starting with system is overkill.

Once a user visit http://FILE_SERVER_IP/stats_launch.php, or from JavaScript fetch, the program stats_server.py must be launched and listen in http://FILE_SERVER_IP:ALTERNATE_PORT. You may also need to handle CORS headers to be able to fetch from the alternate port.

Also, it keep running and updating the stats with a second, or half second, interval, and saving to its own memory, so it can calculate averages or rates per second. Doing this server side gives more precision and less latency that just send the raw samples to client side.

This URL can be visited directly (you will see the output of iostat and ifstat, and values calculated by the python script, in plain text format), or user can load FILE_SERVER_IP/stats.html, where a fetch from JavaScript will periodically load that text plain output, and parse into nice HTML/CSS.

You may think that everything can be solved by fork or by using os.getpid(), os.getsid() and os.setsid() inside the python script, to detach it from the original session and avoid termination with parent process (search stackoverflow for use cases) and it will if your www-data user has the right permissions.

Otherwise I still recommend the visudo approach above. You still need to detach if you start your python file from a .sh file, the difference is that now you have the permissions to do so.

Hatoru Hansou
  • 317
  • 10
  • 22