Going a bit further with the topics of my previous post. ;-)
As hinted in the log excerpt, it could well be that Dovecot invokes
getpwuid(3), directly or indirectly:
> dovecot[97622] <Info>: master: Dovecot v2.1.14 starting up (core dumps
> disabled)
> [...]
> dovecot[97624] <Debug>: pop3(user1): Debug: Namespace : Using
> permissions from /_Data/Mailstores/100016/mboxes: mode=0700 gid=-1
> com.apple.launchd[1] (com.apple.launchd.peruser.100016[97633]) <Error>:
> getpwuid("100016") failed
> com.apple.launchd[1] (com.apple.launchd.peruser.100016[97633])
> <Notice>: Job failed to exec(3). Setting up event to tell us when to try
> again: 3: No such process
> com.apple.launchd[1] (com.apple.launchd.peruser.100016[97633])
> <Notice>: Job failed to exec(3) for weird reason: 3
In this precise case (opening a first time pop connection), this seems to be an
indirect call to getpwuid(3). Anyway, keeping pseudomaster.plist and
pseudomaster.c unchanged, I quickly modified pseudochild.c so that it now calls
getpwuid(3).
When loading pseudomaster.plist for the first time, this is what get written in
the logs:
ALMba.local pseudomaster[66978] <Notice>: Master: started.
ALMba.local pseudomaster[66980] <Notice>: Child: forked.
ALMba.local pseudochild[66980] <Notice>: Pseudochild: started.
ALMba com.apple.launchd[1] (com.apple.launchd.peruser.100018[66981])
<Error>: getpwuid("100018") failed
ALMba com.apple.launchd[1] (com.apple.launchd.peruser.100018[66981])
<Notice>: Job failed to exec(3). Setting up event to tell us when to try again:
3: No such process
ALMba com.apple.launchd[1] (com.apple.launchd.peruser.100018[66981])
<Notice>: Job failed to exec(3) for weird reason: 3
and process pseudochild is just hanging.
Note the process number 66981, which is not pseudochild's one (66980), as if
there were an attempt to spawn a subprocess. And this is a very elusive one;
for example, no way to catch it with execsnoop or similar tools. Even launchd
appears somewhat lost.
Subsequent unloads/reloads of pseudomaster.plist always end with a hanging
pseudochild process, yet without those com.apple.launchd.peruser messages
anymore.
In fact, in order to get those messages back, one has to remove the job bearing
label "com.apple.launchd.peruser.100018".
And this is without mentioning the directories
/var/log/com.apple.launchd.peruser.100018 and
/var/db/launchd/com.apple.launchd.peruser.100018 created under such
circumstances; moreover, those directories seem to persist across reboots...
So, all of this looks quite similar to the problems encountered with Dovecot;
the fact that the stack of the hung pop process ends around a gethostbyname
call could thus just be a red herring.
On the other hand, Dovecot manages to go a bit further than my pseudochild.c
code: with an uid/gid pair such as 100018/20, Dovecot's pop3 process doesn't
hang and performs without a glitch, while pseudochild desperately insists on
entering in a stuck state.
Currently, I'm with a wtf? mood...
I would really appreciate some hints, some explanations for all those
phenomenona...
TIA,
Axel
pseudochild.c
=============
#include <syslog.h>
#include <unistd.h>
#include <sys/errno.h>
#include <sys/types.h>
#include <pwd.h>
#include <uuid/uuid.h>
int main( int argc, const char * argv[])
{
struct passwd * pw;
gid_t gidset[1];
uid_t uid = 100018;
gid_t gid = 100018;
gidset[0] = gid;
syslog(LOG_NOTICE|LOG_MAIL, "Pseudochild: started.");
if (setgid(gid) != 0)
{
syslog(LOG_NOTICE|LOG_MAIL, "Pseudochild: setgid() failed.");
_exit(1);
}
if (setgroups(1, gidset) != 0)
{
syslog(LOG_NOTICE|LOG_MAIL, "Pseudochild: setgroups() failed.");
_exit(1);
}
if (setuid(uid) != 0)
{
syslog(LOG_NOTICE|LOG_MAIL, "Pseudochild: setuid() failed.");
_exit(1);
}
errno = 0;
pw = getpwuid(uid);
if ( pw != NULL )
{
syslog(LOG_NOTICE|LOG_MAIL, "Pseudochild: getpwuid()
succeeded");
_exit(0);
}
else
{
syslog(LOG_NOTICE|LOG_MAIL, "Pseudochild: getpwuid() failed
with rc %i", errno);
_exit(1);
}
}
_______________________________________________
launchd-dev mailing list
launchd-dev@lists.macosforge.org
https://lists.macosforge.org/mailman/listinfo/launchd-dev
