March 2008 - launchd-dev - lists.macosforge.org

OnDemand USB Software
by Mike Fischer 11 Jul '08

11 Jul '08

Hi! I'm trying to evaluate if it is feasable to have an on-demand daemon depend on the presence of a certain usb device. I.e. I'd like the daemon to launch when the USB-device is plugged in. So far, looking at tn2083 and the man pages, I don't see any way to have a depency on USB or IOKit in general. Am I missing something here? If there is no direct support would it make sense to have a very small, lightweight daemon always running that registers for IOKit IOServiceAddMatchingNotification notifications and creates a semaphore file that can be used to trigger the launch of the real daemon for example? BTW: This would be mainly for Leopard, although Tiger would be interesting too. Thanks! Mike -- Mike Fischer Softwareentwicklung, EDV-Beratung Schulung, Vertrieb Web: <http://homepage.mac.com/mike_fischer/index.html> Note: I read this list in digest mode! Send me a private copy for faster responses.

3 2

launchd (257) not building
by Dave MacLachlan 06 May '08

06 May '08

> Hey Ryan... >> error: CoreFoundation/CFPriv.h: > No such file or directory I'm running into problems getting launchd > to compile as well, but at least for this part you can grab these > headers from > http://src.macosforge.org/Roots/9C31/CF.root.tar.gz Hope that helps > somewhat. I'm still looking for > bootfiles.h Further follow up, I haven't found bootfiles.h, but from what I can glean: #define kBootRootActiveKey "bootroot-active" #define SO_EXECPATH 0x1085 (could somebody please correct me if those values are wrong) but I can't find quarantine.h anywhere, and radar is currently down so I can't log a bug against it. I'm trying to build up liblaunch, so I can build up a debug version of libSystem so I can verify what appears to be a bug in libc_debug which is occurring when I run my app against the _debug frameworks that DTS is shipping. Talk about yak shaving. Cheers, Dave

5 8

Re: [launchd-dev] OnDemand USB Software
by Mike Fischer 01 Apr '08

01 Apr '08

Hi Quinn, sorry for the delay, somehow I missed your answer (probably because the mailing list digest only arrived two days ago, 9 days after you sent the answer ;-). Am 27.03.2008 um 16:12 schrieb Quinn <eskimo1(a)apple.com>: > At 23:23 +0100 17/3/08, Mike Fischer wrote: >> So far, looking at tn2083 and the man pages, I don't see any way to >> have a depency on USB or IOKit in general. Am I missing something >> here? > > No, you're not missing something. This is a much-requested feature > that we're tracking as <rdar://problem/4551362>. OK. If we go ahead with this project I'll probably file a (duplicate) bug to raise the priority. >> If there is no direct support would it make sense to have a very >> small, lightweight daemon always running that registers for IOKit >> IOServiceAddMatchingNotification notifications and creates a >> semaphore file that can be used to trigger the launch of the real >> daemon for example? > > That's a fine way to do it. I probably wouldn't use a semaphore > file, because it persists in the file system. However, there are > numerous other options available to you. If I was doing this, I'd > probably have your main daemon register a UNIX domain socket and have > your lightweight daemon connect to that socket to trigger a launch. OK. The semaphore file was just the first trigger I could think of that would work with launchd. The UNIX domain socket is probably a better idea. If I understand this correctly I would not be using the socket for actual communication, just for triggering the launch of the "server", i.e. the real (on demand) daemon? So I'd include something like this in the real daemon's launchd.plist: <key>Sockets</key> <dict> <key>Listeners</key> <dict> <key>SockNodeName</key> <string>/tmp/MYSOCKETNAME</string> </dict> </dict> Would I need to add the SockFamily or SockProtocol keys and if so what would be the correct values for a UNIX domain socket? Anything else I'd need to add? In the real daemon I'd do a socket(2), bind(2), accept(2) (the listen (2) is already done by launchd I gather)? Or do I need this at all? And in the USB-Watcher daemon I'd do a socket(2), connect(2), which would trigger the launch of the real daemon right? BTW: The launchd.plist(5) man page contains references to a launch(3) man page (10.5.2, Xcode 3.0 as well as 10.4.11, Xcode 2.4.1). Unfortunatly the launch(3) man page does not seem to exist. Filed as <rdar://5830171>. I saw /usr/include/launch.h and the SampleD sample code but these don't really explain what is going on and how to properly check-in with launchd. Any hints on where to get more information? >> BTW: This would be mainly for Leopard, although Tiger would be >> interesting too. > > This approach would work just fine on 10.4.x (x >= 6 if you use UNIX > domain sockets). Good to know. Now I need to discuss these options with my customer and see if we can do this without completely changing their existing cross platform code. Thanks! Your answer is very much appreciated. Mike -- Mike Fischer Softwareentwicklung, EDV-Beratung Schulung, Vertrieb Web: <http://homepage.mac.com/mike_fischer/index.html> Note: I read this list in digest mode! Send me a private copy for faster responses.

2 1

How to on demand launch for configurable port number?
by Mike Fischer 01 Apr '08

01 Apr '08

Hi! Let's say I have a daemon that listens on a TCP port. The exact port number can be configured, i.e. it is not fixed. It might even change when the user changes the configuration, albeit not very often. Actually this daemon is a HTTP server that serves status information about a service. Thus it will not be used most of the time. Always running the server would be a waste of resources IMHO and I want avoid that by making it an on-demand service. Is it possible to make launchd launch the daemon on demand when a connection request comes in? If so how would I specify this in the launchd.plist file, taking into account the varying TCP port number? Currently I'd immagine that I need to 'launchctl stop' the service, modify the associated launchd.plist to reflect the currently configured port and then 'launchctl start' the service. Unless of course there is a better way that doesn't involve changing the launchd.plist? Not sure if this is relevant but the daemon does not use Bonjour at the moment to advertise its service. That could be added though (and might be a good idea for other reasons). I tried to find similar services, for example the Apache Web Server. But while it uses configurable port numbers it does not launch on demand unfortunatly. So I can't use it as an example of how to do this. Thanks! Mike -- Mike Fischer Softwareentwicklung, EDV-Beratung Schulung, Vertrieb Web: <http://homepage.mac.com/mike_fischer/index.html> Note: I read this list in digest mode! Send me a private copy for faster responses.

2 1

Obtaining the value of SSH_AUTH_SOCK in an entry?
by Chris Cleeland 27 Mar '08

27 Mar '08

Hi, I'm trying to set up an entry for launchd that will fire off ssh on-demand to create a tunnel. I would like for the ssh to get authentication information from ssh-agent, but I'm not sure if launchd children inherit SSH_AUTH_SOCK in the environment, or if I somehow have to designate that it gets it. In the pre-leopard days when I used SSHKeyChain.app, the SSH_AUTH_SOCK value was constant, and I could just set SSH_AUTH_SOCK in an <EnvironmentVariables> declaration in the entry. Thanks, -cj

3 3

Stray Proccess Log message
by Geoff Franks 18 Mar '08

18 Mar '08

I have a previously-cronned job that I'm moving to launchd on my new 10.5 server. It runs every 10 minutes, and causes the following entry to show up in the syslog: Feb 21 15:43:45 server1 com.apple.launchd[1] (edu.buffalo.hwi.raidutil[43977]): Stray process with PGID equal to this dead job: PID 43983 PPID 1 sendmail The script the launchd job runs calls sendmail a the end of it, so my guess is that the script finishes, while sendmail is still running, and then launchd has to clean up after the script exits, so tries to kill the sendmail process. However, my emails are all going through and seemingly working properly. Additionally, I specified the AbandonProccessGrop <true/> key in the plist so it wouldn't try cleaning up the other processes spawned. Shouldn't this take care of the log message? Geoff Franks Sr. Systems Administrator Hauptman Woodward Institute

4 9

Lowering the bar to entry
by Dave MacLachlan 17 Mar '08

17 Mar '08

Hey guys, Would it be possible to add an intro page similar to the one for DarwinBuild http://trac.macosforge.org/projects/darwinbuild/ That shows quickly: a) how to find the mailing list b) how to download the source c) and a readme describing how to build? specifically how to build the libraries vs all of launchd? There is a blog entry that covers some of this: http://launchd.macosforge.org/?p=2 but it isn't extra easy to find. Cheers, Dave

1 0

launchd (257) not building
by Dave MacLachlan 16 Mar '08

16 Mar '08

Hey Ryan... > error: CoreFoundation/CFPriv.h: No such file or directory I'm running into problems getting launchd to compile as well, but at least for this part you can grab these headers from http://src.macosforge.org/Roots/9C31/CF.root.tar.gz Hope that helps somewhat. I'm still looking for bootfiles.h Cheers, Dave

1 0

My Launch Agent is Persisting Beyond Logout
by Mac QA 16 Mar '08

16 Mar '08

Hi, I have a custom process that is invoked when a user logs in via having a launchd config file located in /Library/Launch Agents. The config file looks something like this: <key>KeepAlive</key> <true/> <key>Label</key> <string>com.company.mylaunchagent</string> <key>OnDemand</key> <false/> <key>ProgramArguments</key> <array> <string>/mylaunchagent</string> </array> <key>RunAtLoad</key> <true/> The mylaunchagent is invoked just fine. If I do a fast user switch to login another user then a second instance is invoked for the second user. All seems good. However, when the second user logs out the second instance of mylaunchagent continues to stay running even though the user logs out! The Apple Multiple User Environments document [1] makes two statements regarding this. First, "If a user process daemonizes itself prior to logout, it can live past the end of the user logout and prolong the existence of the login session." And Second, "Per-user services are shut down automatically and are not given the chance to abort the logout procedure." The mylaunchagent process is always a background process just invoked by launchd, so given the first statement I guess that is why it is persisting logout. However, given the second statement, since this is a per-user service as launchagents are by definition, I don't understand why it isn't shut down automatically on logout. Frankly, those two statements seem contradictory. [1] Apple Multiple User Environments http://developer.apple.com/documentation/MacOSX/Conceptual/BPMultipleUsers/… Can anyone point me towards how to make sure my launch agent processes get properly killed when a user logs out? I appreciate any tips. Thanks!

5 7

Directory Access launching?
by Levi Brown 12 Mar '08

12 Mar '08

Hello all, I have an interesting situation which I thought I might be able to utilize launchd to help me solve, but I could use some expert advice. The idea is pretty simple: I use gnupg. gnupg is used by services I use. gnupg stores all its information in ~/.gnupg unless the gnupg executable is called with a flag to tell it to do otherwise. Since I do not have control of how the executable is referenced by these services, the ~/.gnupg directory is my only choice. Stored inside the ~/.gnupg directory is the public and private keys for signing, en/decrypting, configuration, etc. which is sensitive information and could easily be accessed should my machine be stolen, for instance. I'd like to store the contents of this directory on an encrypted disk image which is automatically attached and detached based on need. So, ideally, there would be a way to have a particular directory be monitored for access, and once accessed, a script could be called to mount the image. After a period of inactivity to this path, the volume could be detached. automount and autofs are intended to do this kind of operation for network volumes an even provide a executable map facility which gives one a hook into the need to mount a directory based on access, however this is "old school" unix compared to the launchd train of thought, and I've not been able to get a solution working with this approach. Is there some configuration key like PathState, or NetworkState which could be used to achieve this goal? Other suggestions? Thanks, Levi

2 1