In looking at this further, I’m not seeing how this can be configured as documented and still work.

The sieve_pipe_socket_dir setting, as the documentation suggests, appears to be used by the plugin to construct a path to find the “script” (the socket in this case), and it is always relative to base_dir and must include the contents of sieve_pipe_socket_dir.

/* Try socket first */
if ( ext_config->socket_dir != NULL ) {
    path = t_strconcat(senv->user->set->base_dir, "/",
        ext_config->socket_dir, "/", program_name, NULL);

But nowhere in dovecot can I find a place where this subdir would be created at startup. The only subdirs that I can find created are all explicitly hardcoded in master-settings.c in a function called master_settings_do_fixes():

/* Make sure our permanent state directory exists */
if (mkdir_parents(set->state_dir, 0755) < 0 && errno != EEXIST)
	i_fatal("mkdir(%s) failed: %m", set->state_dir);

mkdir_login_dir(set, t_strconcat(set->base_dir, "/login", NULL));
mkdir_login_dir(set, t_strconcat(set->base_dir, "/token-login", NULL));

empty_dir = t_strconcat(set->base_dir, "/empty", NULL);
if (safe_mkdir(empty_dir, 0755, master_uid, getegid()) == 0) {
	i_warning("Corrected permissions for empty directory "
		  "%s", empty_dir);
}

Granted this is from a pretty quick browsing of the sources but I couldn’t find where subdirs specified in service listeners might be created at startup.

I’m working around this by defining sieve_pipe_socket_dir to the current directory (“.”), which allows the search path to resolve to base_dir, and find the socket in there. Then I just define the listener without the subdir. This seems to work fine and allows dovecot to start up.

Unless someone can reveal the magic incantation to use sieve_pipe_socket_dir to create an actual subdir, though, it seems like either a documentation issue or a bug. I’d argue the latter since the configuration variable isn’t actually useful.00

Garth

On Dec 17, 2017, at 4:51 PM, Garth Corral gcorral@abode.com wrote:

Hi, all

I’m new to the list but not to dovecot. I’ve been using it in a basic configuration for some time, but finally decided to tweak my deployed system to take advantage of some more dovecot features. In particular I’m trying to set up pigeonhole to implement spam retraining with imapsieve. All of this is with dovecot 2.2.31 (65cde28) and pigeonhole 0.4.19.

Before going any further let me start by saying that I have gotten all of this to work. It works when I can get dovecot to start up, that is. My configuration is pretty much straight from the docs, with a few tweaks for my particular needs. I’m trying to set up a pipe service using sieve-extprograms, and the relevant part of my config looks like this:

plugin {

sieve_pipe_input_eol = lf

sieve_pipe_socket_dir = sieve-pipe sieve_filter_socket_dir = sieve-filter sieve_execute_socket_dir = sieve-execute

sieve_pipe_bin_dir = /usr/local/libexec/dovecot/sieve-pipe sieve_filter_bin_dir = /usr/local/libexec/dovecot/sieve-filter sieve_execute_bin_dir = /usr/local/libexec/dovecot/sieve-execute }

service sieve-train-ham { executable = script /usr/local/libexec/dovecot/sieve-pipe/train-ham.sh

# Needs access to dspam config and lockfiles. user = dspam

# socket name is program-name in Sieve (without sieve-pipe/ prefix) unix_listener sieve-pipe/train-ham { } }

It’s my understanding from reading the docs that the sieve_pipe_socket_dir specifies a directory that is relative to base_dir, which is the default /var/run/dovecot in my case. The issue I’m having is that dovecot will not start, and spews the following errors:

dovecot: master: Error: bind(/var/run/dovecot/sieve-pipe/train-ham) failed: No such file or directory dovecot: master: Fatal: Failed to start listeners

Once dovecot fails startup, it leaves /var/run/dovecot around and if I manually create the sieve-pipe directory there it will start up, create the sockets there and everything works as intended subsequently. The problem, though, is that on normal shutdown all of /var/run/dovecot goes away and the at the next startup it fails to start again. Needless to say this isn’t great for unintended reboots, etc.

So, can anyone see anything obvious that I’m doing wrong? I’m just assuming that dovecot should create the needed subdir since I can’t find anything in the docs to suggest a way to create it otherwise. I’ve tried all I can think of at the moment to try to remedy this without success. I’m happy to provide additional details as needed to try to track this down.

Thanks in advance,

Garth

Op 12/18/2017 om 1:51 AM schreef Garth Corral:

Hi, all

I’m new to the list but not to dovecot. I’ve been using it in a basic configuration for some time, but finally decided to tweak my deployed system to take advantage of some more dovecot features. In particular I’m trying to set up pigeonhole to implement spam retraining with imapsieve. All of this is with dovecot 2.2.31 (65cde28) and pigeonhole 0.4.19.

Before going any further let me start by saying that I have gotten all of this to work. It works when I can get dovecot to start up, that is. My configuration is pretty much straight from the docs, with a few tweaks for my particular needs. I’m trying to set up a pipe service using sieve-extprograms, and the relevant part of my config looks like this:

plugin {

sieve_pipe_input_eol = lf

sieve_pipe_socket_dir = sieve-pipe sieve_filter_socket_dir = sieve-filter sieve_execute_socket_dir = sieve-execute

sieve_pipe_bin_dir = /usr/local/libexec/dovecot/sieve-pipe sieve_filter_bin_dir = /usr/local/libexec/dovecot/sieve-filter sieve_execute_bin_dir = /usr/local/libexec/dovecot/sieve-execute }

Define either *bin_dir or *socket_dir; not both. These are two alternative ways of running scripts, either using a direct fork() or through a script service, respectively.

https://wiki.dovecot.org/Pigeonhole/Sieve/Plugins/Extprograms

service sieve-train-ham { executable = script /usr/local/libexec/dovecot/sieve-pipe/train-ham.sh

# Needs access to dspam config and lockfiles. user = dspam

# socket name is program-name in Sieve (without sieve-pipe/ prefix) unix_listener sieve-pipe/train-ham { } }

It’s my understanding from reading the docs that the sieve_pipe_socket_dir specifies a directory that is relative to base_dir, which is the default /var/run/dovecot in my case. The issue I’m having is that dovecot will not start, and spews the following errors:

dovecot: master: Error: bind(/var/run/dovecot/sieve-pipe/train-ham) failed: No such file or directory dovecot: master: Fatal: Failed to start listeners

First of all, this directory does not strictly need to be relative to /var/run/dovecot. It can be an absolute path to somewhere else.

The relative directories are indeed not currently created implicitly. Hmm, we will discuss this internally.

Once dovecot fails startup, it leaves /var/run/dovecot around and if I manually create the sieve-pipe directory there it will start up, create the sockets there and everything works as intended subsequently. The problem, though, is that on normal shutdown all of /var/run/dovecot goes away and the at the next startup it fails to start again. Needless to say this isn’t great for unintended reboots, etc.

So, can anyone see anything obvious that I’m doing wrong? I’m just assuming that dovecot should create the needed subdir since I can’t find anything in the docs to suggest a way to create it otherwise. I’ve tried all I can think of at the moment to try to remedy this without success. I’m happy to provide additional details as needed to try to track this down.

Regards,

Stephan.