2.6.0-test3: setuid32(8) returns EAGAIN (WTF?!)

I just changed from 2.5.75 to 2.6.0-test3 and suddenly my imap server fails to start (it's dovecot). It wrote to syslog:

Aug 17 02:58:02 hellhound dovecot: Dovecot starting up
Aug 17 02:58:03 hellhound imap-login: setuid(8) failed: Resource temporarily unavailable
Aug 17 02:58:03 hellhound dovecot: Login process died too early - shutting down

So I strace -f it, and sure enough, here is what happens:

[init, fork, tzfile...]

8094  chroot("/var/run/dovecot//login") = 0
8094  chdir("/")                        = 0
8094  setuid32(0x8)                     = -1 EAGAIN (Resource temporarily unavailable)

Now this does not appear to be a valid return value for setuid32, and my understanding of POSIX and Susv3 is that dovecot is absolutely right in barfing at this.

Why is this happening? Please fix!

Felix

PS: Time for a brown paper bag bug-fix release, if you ask me.

In sys.c:set_user() we see

        if (atomic_read(&new_user->processes) >=
                                current->rlim[RLIMIT_NPROC].rlim_cur &&
                        new_user != &root_user) {
                free_uid(new_user);
                return -EAGAIN;
        }

which was added in patch-2.6.0-test2.
No doubt this causes your problem.

You might check what values these variables have for you.

Andries