autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Guillaume Rousse]

autofs 5.0.1 isn't able to retrieve ldap master map automatically, as it
did in 4.1.x:
Aug 24 16:33:17 alceste automount[23385]: lookup_init: lookup(ldap):
failed to get query dn

When giving it an explicit dn as MASTER_MAP_NAME, everything is fine,
however.
-- 
Guillaume Rousse
Projet Estime, INRIA
Domaine de Voluceau
Rocquencourt - B.P. 105
78153 Le Chesnay Cedex - France

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Thu, 24 Aug 2006, Guillaume Rousse wrote:

> autofs 5.0.1 isn't able to retrieve ldap master map automatically, as it
> did in 4.1.x:
> Aug 24 16:33:17 alceste automount[23385]: lookup_init: lookup(ldap):
> failed to get query dn
> 
> When giving it an explicit dn as MASTER_MAP_NAME, everything is fine,
> however.

Some more info would be good.

Ian

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Guillaume Rousse]

Ian Kent wrote:
> On Thu, 24 Aug 2006, Guillaume Rousse wrote:
> 
>> autofs 5.0.1 isn't able to retrieve ldap master map automatically, as it
>> did in 4.1.x:
>> Aug 24 16:33:17 alceste automount[23385]: lookup_init: lookup(ldap):
>> failed to get query dn
>>
>> When giving it an explicit dn as MASTER_MAP_NAME, everything is fine,
>> however.
> 
> Some more info would be good.
Unfortunatly, that's all I get even with maximum debug output configured:
--verbose and --debug flags for automount
LOGGING="debug" in configuration file

I use the following LDAP attributes:
MAP_OBJECT_CLASS="automountMap"
ENTRY_OBJECT_CLASS="automount"
MAP_ATTRIBUTE="ou"
ENTRY_ATTRIBUTE="cn"
VALUE_ATTRIBUTE="automountInformation"

I'll try to run it in the debugger, but I have to fix my debug package
first.
-- 
Guillaume Rousse
Projet Estime, INRIA
Domaine de Voluceau
Rocquencourt - B.P. 105
78153 Le Chesnay Cedex - France

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Fri, 25 Aug 2006, Guillaume Rousse wrote:

> Ian Kent wrote:
> > On Thu, 24 Aug 2006, Guillaume Rousse wrote:
> > 
> >> autofs 5.0.1 isn't able to retrieve ldap master map automatically, as it
> >> did in 4.1.x:
> >> Aug 24 16:33:17 alceste automount[23385]: lookup_init: lookup(ldap):
> >> failed to get query dn
> >>
> >> When giving it an explicit dn as MASTER_MAP_NAME, everything is fine,
> >> however.
> > 
> > Some more info would be good.
> Unfortunatly, that's all I get even with maximum debug output configured:
> --verbose and --debug flags for automount
> LOGGING="debug" in configuration file
> 
> I use the following LDAP attributes:
> MAP_OBJECT_CLASS="automountMap"
> ENTRY_OBJECT_CLASS="automount"
> MAP_ATTRIBUTE="ou"
> ENTRY_ATTRIBUTE="cn"
> VALUE_ATTRIBUTE="automountInformation"
> 
> I'll try to run it in the debugger, but I have to fix my debug package
> first.

Can you give a couple of examples of you LDAP entries please.

Ian

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Guillaume Rousse]

Ian Kent wrote:
>> I'll try to run it in the debugger, but I have to fix my debug package
>> first.
gdbdoesn't complain anymore about missing symbols, but despite my
attempts, I've been unable to set a working break point on faulty
function (get_query_dn).

> Can you give a couple of examples of you LDAP entries please.
Here it is:

# autofs, village.inria.fr
dn: ou=autofs,dc=village,dc=inria,dc=fr
ou: autofs
objectClass: top
objectClass: organizationalUnit

# auto.master, autofs, village.inria.fr
dn: ou=auto.master,ou=autofs,dc=village,dc=inria,dc=fr
objectClass: top
objectClass: automountMap
ou: auto.master

# /home/beaune, auto.master, autofs, village.inria.fr
dn: cn=/home/beaune,ou=auto.master,ou=autofs,dc=village,dc=inria,dc=fr
objectClass: top
objectClass: automount
cn: /home/beaune
automountInformation:
ldap:ou=auto.home.beaune,ou=autofs,dc=village,dc=inria,d
 c=fr

# /home/yquem, auto.master, autofs, village.inria.fr
dn: cn=/home/yquem,ou=auto.master,ou=autofs,dc=village,dc=inria,dc=fr
objectClass: top
objectClass: automount
cn: /home/yquem
automountInformation:
ldap:ou=auto.home.yquem,ou=autofs,dc=village,dc=inria,dc
 =fr

# /net/yquem, auto.master, autofs, village.inria.fr
dn: cn=/net/yquem,ou=auto.master,ou=autofs,dc=village,dc=inria,dc=fr
objectClass: top
objectClass: automount
cn: /net/yquem
automountInformation:
ldap:ou=auto.net.yquem,ou=autofs,dc=village,dc=inria,dc=
 fr

etc...
-- 
Guillaume Rousse
Projet Estime, INRIA
Domaine de Voluceau
Rocquencourt - B.P. 105
78153 Le Chesnay Cedex - France

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Fri, 25 Aug 2006, Guillaume Rousse wrote:

> Ian Kent wrote:
> >> I'll try to run it in the debugger, but I have to fix my debug package
> >> first.
> gdbdoesn't complain anymore about missing symbols, but despite my
> attempts, I've been unable to set a working break point on faulty
> function (get_query_dn).

Is your LDAP client able to query this stuff with only the information 
used in the autofs queries (the ones you see in the debug output)?

That's usually where I get my error info from.

Ian

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Guillaume Rousse]

Ian Kent wrote:
> On Fri, 25 Aug 2006, Guillaume Rousse wrote:
> 
>> Ian Kent wrote:
>>>> I'll try to run it in the debugger, but I have to fix my debug package
>>>> first.
>> gdbdoesn't complain anymore about missing symbols, but despite my
>> attempts, I've been unable to set a working break point on faulty
>> function (get_query_dn).
> 
> Is your LDAP client able to query this stuff with only the information 
> used in the autofs queries (the ones you see in the debug output)?
Not sure to understand you here. I get nothing else in my debug output
than: failed to get query dn

Using informations from the config file, however, such as the automount
map class, I can retrieve all those maps easily from ldapsearch:
[rousse@alceste ~]$ ldapsearch -x objectClass=automountMap
# extended LDIF
#
# LDAPv3
# base <> with scope subtree
# filter: objectClass=automountMap
# requesting: ALL
#

# auto.master, autofs, village.inria.fr
dn: ou=auto.master,ou=autofs,dc=village,dc=inria,dc=fr
objectClass: top
objectClass: automountMap
ou: auto.master

# auto.home.beaune, autofs, village.inria.fr
dn: ou=auto.home.beaune,ou=autofs,dc=village,dc=inria,dc=fr
objectClass: top
objectClass: automountMap
ou: auto.home.beaune
...
# search result
search: 2
result: 0 Success

# numResponses: 7
# numEntries: 6

-- 
Guillaume Rousse
Projet Estime, INRIA
Domaine de Voluceau
Rocquencourt - B.P. 105
78153 Le Chesnay Cedex - France

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Fri, 2006-08-25 at 14:49 +0200, Guillaume Rousse wrote:
> Ian Kent wrote:
> > On Fri, 25 Aug 2006, Guillaume Rousse wrote:
> > 
> >> Ian Kent wrote:
> >>>> I'll try to run it in the debugger, but I have to fix my debug package
> >>>> first.
> >> gdbdoesn't complain anymore about missing symbols, but despite my
> >> attempts, I've been unable to set a working break point on faulty
> >> function (get_query_dn).
> > 
> > Is your LDAP client able to query this stuff with only the information 
> > used in the autofs queries (the ones you see in the debug output)?
> Not sure to understand you here. I get nothing else in my debug output
> than: failed to get query dn

Have you setup syslog.conf to log daemon.* to somewhere?
You won't get much at all unless syslog is sending it some where.

Mmmm .. I see I don't debug print the query in get_query_dn.

But looking at the code, if there is a map name the filter would be
something like:

(&(objectclass=<map_obj_class>)(<map_attr>=<map name>))

so I think for your defaults that would be

(&(objectclass=automountMap)(ou=auto.master))

It gets the base dn from the output of this query.
If that doesn't work nothing else does, but you already know that.

Ian

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Fri, 2006-08-25 at 14:25 +0200, Guillaume Rousse wrote:
> Ian Kent wrote:
> >> I'll try to run it in the debugger, but I have to fix my debug package
> >> first.
> gdbdoesn't complain anymore about missing symbols, but despite my
> attempts, I've been unable to set a working break point on faulty
> function (get_query_dn).
> 
> > Can you give a couple of examples of you LDAP entries please.
> Here it is:

This looks OK.
I wonder what could be causing this unusual problem.

Ian

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Fri, 2006-08-25 at 21:18 +0800, Ian Kent wrote:
> On Fri, 2006-08-25 at 14:49 +0200, Guillaume Rousse wrote:
> > Ian Kent wrote:
> > > On Fri, 25 Aug 2006, Guillaume Rousse wrote:
> > > 
> > >> Ian Kent wrote:
> > >>>> I'll try to run it in the debugger, but I have to fix my debug package
> > >>>> first.
> > >> gdbdoesn't complain anymore about missing symbols, but despite my
> > >> attempts, I've been unable to set a working break point on faulty
> > >> function (get_query_dn).
> > > 
> > > Is your LDAP client able to query this stuff with only the information 
> > > used in the autofs queries (the ones you see in the debug output)?
> > Not sure to understand you here. I get nothing else in my debug output
> > than: failed to get query dn
> 
> Have you setup syslog.conf to log daemon.* to somewhere?
> You won't get much at all unless syslog is sending it some where.
> 
> Mmmm .. I see I don't debug print the query in get_query_dn.
> 
> But looking at the code, if there is a map name the filter would be
> something like:
> 
> (&(objectclass=<map_obj_class>)(<map_attr>=<map name>))
> 
> so I think for your defaults that would be
> 
> (&(objectclass=automountMap)(ou=auto.master))
> 
> It gets the base dn from the output of this query.
> If that doesn't work nothing else does, but you already know that.

Oh .. btw .. to get quick info you can just stick an error print in
like:

error(LOGOPT_ANY, "query %s", query);

and it should show up in the log regardless of the debug setting.

I find getting a gdb session going is a hassle much of the time.

Ian

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Piete.Brooks--autofs]

> OK, I found the issue: when all you have is mapname, and you have to use
> default server and base, you should use LDAP_SCOPE_SUBTREE scope for
> your query, not LDAP_SCOPE_ONELEVEL

Why?

As I said before, it works for me as it is. I see no need to change it.

I've just had my wrist slapped for suggesting doing three queries for the 
three "common" schemas, so changing from a one level search (which should be 
very fast as I'd not expect much at the top level) to searching the WHOLE tree 
(which I expect to be several orders of magnitude slower) would seem not to be 
a good idea.


If you do the search manually, does it work?
Are the two programmes using the same conf file (e.g. std /etc/ldap.conf vs 
/etc/openldap/ldap.conf)?

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Guillaume Rousse]

Ian Kent wrote:
> On Fri, 2006-08-25 at 21:18 +0800, Ian Kent wrote:
>> On Fri, 2006-08-25 at 14:49 +0200, Guillaume Rousse wrote:
>>> Ian Kent wrote:
>>>> On Fri, 25 Aug 2006, Guillaume Rousse wrote:
>>>>
>>>>> Ian Kent wrote:
>>>>>>> I'll try to run it in the debugger, but I have to fix my debug package
>>>>>>> first.
>>>>> gdbdoesn't complain anymore about missing symbols, but despite my
>>>>> attempts, I've been unable to set a working break point on faulty
>>>>> function (get_query_dn).
>>>> Is your LDAP client able to query this stuff with only the information 
>>>> used in the autofs queries (the ones you see in the debug output)?
>>> Not sure to understand you here. I get nothing else in my debug output
>>> than: failed to get query dn
>> Have you setup syslog.conf to log daemon.* to somewhere?
>> You won't get much at all unless syslog is sending it some where.
>>
>> Mmmm .. I see I don't debug print the query in get_query_dn.
>>
>> But looking at the code, if there is a map name the filter would be
>> something like:
>>
>> (&(objectclass=<map_obj_class>)(<map_attr>=<map name>))
>>
>> so I think for your defaults that would be
>>
>> (&(objectclass=automountMap)(ou=auto.master))
>>
>> It gets the base dn from the output of this query.
>> If that doesn't work nothing else does, but you already know that.
> 
> Oh .. btw .. to get quick info you can just stick an error print in
> like:
> 
> error(LOGOPT_ANY, "query %s", query);
> 
> and it should show up in the log regardless of the debug setting.
OK, I found the issue: when all you have is mapname, and you have to use
default server and base, you should use LDAP_SCOPE_SUBTREE scope for
your query, not LDAP_SCOPE_ONELEVEL.

Here is a patch that fixes the problem:
diff -Naur autofs-5.0.1/modules/lookup_ldap.c
autofs-5.0.1-fix-ldap-lookup/modules/lookup_ldap.c
--- autofs-5.0.1/modules/lookup_ldap.c  2006-07-13 04:11:39.000000000 -0400
+++ autofs-5.0.1-fix-ldap-lookup/modules/lookup_ldap.c  2006-08-25
11:19:59.000000000 -0400
@@ -802,7 +802,7 @@
                              MODPREFIX "error forming query string");
                        return 0;
                }
-               scope = LDAP_SCOPE_ONELEVEL;
+               scope = LDAP_SCOPE_SUBTREE;
        } else {
                if (sprintf(query, "(objectclass=%s)", class) >= l) {
                        debug(LOGOPT_NONE,


> I find getting a gdb session going is a hassle much of the time.
It depends on your setup. I usually build my packages on mandriva
cluster, and install then on my own machines for testing them, rather
than directly building and installing locally.
-- 
Guillaume Rousse
Projet Estime, INRIA
Domaine de Voluceau
Rocquencourt - B.P. 105
78153 Le Chesnay Cedex - France

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Guillaume Rousse]

Piete.Brooks--autofs@cl.cam.ac.uk wrote:
>> OK, I found the issue: when all you have is mapname, and you have to use
>> default server and base, you should use LDAP_SCOPE_SUBTREE scope for
>> your query, not LDAP_SCOPE_ONELEVEL
> 
> Why?
Because using LDAP_SCOPE_ONELEVEL assume your master map is directly
beyond your default base, which is likely to be wrong.

> As I said before, it works for me as it is. I see no need to change it.
> 
> I've just had my wrist slapped for suggesting doing three queries for the 
> three "common" schemas, so changing from a one level search (which should be 
> very fast as I'd not expect much at the top level) to searching the WHOLE tree 
> (which I expect to be several orders of magnitude slower) would seem not to be 
> a good idea.
Only once, for the initial master map lookup, and only if no explicit
map name has been given.

> 
> If you do the search manually, does it work?
Yes, because ldapsearch default's scope is sub.

> Are the two programmes using the same conf file (e.g. std /etc/ldap.conf vs 
> /etc/openldap/ldap.conf)?
of course. ldap client use ldap library, as does automount.
-- 
Guillaume Rousse
Projet Estime, INRIA
Domaine de Voluceau
Rocquencourt - B.P. 105
78153 Le Chesnay Cedex - France

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Fri, 2006-08-25 at 18:04 +0200, Guillaume Rousse wrote:
> Piete.Brooks--autofs@cl.cam.ac.uk wrote:
> >> OK, I found the issue: when all you have is mapname, and you have to use
> >> default server and base, you should use LDAP_SCOPE_SUBTREE scope for
> >> your query, not LDAP_SCOPE_ONELEVEL
> > 
> > Why?
> Because using LDAP_SCOPE_ONELEVEL assume your master map is directly
> beyond your default base, which is likely to be wrong.

Ha, of course, and I thought that would be more efficient.
Pity it just doesn't work.

I'll fix it.

Ian

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Fri, 2006-08-25 at 16:21 +0100, Piete.Brooks--autofs@cl.cam.ac.uk
wrote:
> I've just had my wrist slapped for suggesting doing three queries for the 
> three "common" schemas, so changing from a one level search (which should be 

I didn't mean to be nasty.
It's just that it seems like every time I do something in response to
requests someone else then wants the opposite.

Actually, using the multiple schema we were seeing up to "nine" queries
per lookup during a mount failure situation which is really bad for a
busy server. It's now one or two so it's much better.

Ian

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Piete.Brooks--autofs]

> I didn't mean to be nasty.

Not taken as such. Sorry -- I should scatter a few more ":-)"'s around :-))

> It's just that it seems like every time I do something in response to
> requests someone else then wants the opposite.

That's where judgement kicks in.
When you have equal numbers objecting to both options, you've got it right :-))

> Actually, using the multiple schema we were seeing up to "nine" queries
> per lookup during a mount failure situation

Ahh -- I was thinking of the multiple schema only during initial discovery.
I was assuming that like the mount_bind code, at the start of day it would 
look at its environment and decide how to do things.
If not told what schema to use, it would try the pre defined (3 at the moment) 
schema until one of them worked, and then latch onto that one.
(e.g. default is the null string, and if it finds the string null, call a 
discovery procedure).

As such this would not effect a mount failure -- it would only be done once at 
startup (and when it gets a HUP etc)


Would it be OTT to add an option to say whether the initial lookup of 
"auto.master" was single level or recursive?
[ I really don't understand why recursive is needed if auto.master is at the 
top level :-( ]

Re: autofs 5.0.1-rc1 still doesn't find master map in ldap automatically

[Ian Kent]

On Sat, 26 Aug 2006, Piete.Brooks--autofs@cl.cam.ac.uk wrote:

> > I didn't mean to be nasty.
> 
> Not taken as such. Sorry -- I should scatter a few more ":-)"'s around :-))
> 
> > It's just that it seems like every time I do something in response to
> > requests someone else then wants the opposite.
> 
> That's where judgement kicks in.
> When you have equal numbers objecting to both options, you've got it right :-))
> 
> > Actually, using the multiple schema we were seeing up to "nine" queries
> > per lookup during a mount failure situation
> 
> Ahh -- I was thinking of the multiple schema only during initial discovery.
> I was assuming that like the mount_bind code, at the start of day it would 
> look at its environment and decide how to do things.
> If not told what schema to use, it would try the pre defined (3 at the moment) 
> schema until one of them worked, and then latch onto that one.
> (e.g. default is the null string, and if it finds the string null, call a 
> discovery procedure).
> 
> As such this would not effect a mount failure -- it would only be done once at 
> startup (and when it gets a HUP etc)

Yep. I guessed that was what you were getting at after I posted.

It's a good idea however I'd rather not change that part of the 
code for this release becuase there's already been major changes in that 
module so I'll let that settle. I'll reconsider it for the next release.

> 
> 
> Would it be OTT to add an option to say whether the initial lookup of 
> "auto.master" was single level or recursive?
> [ I really don't understand why recursive is needed if auto.master is at the 
> top level :-( ]

Guillaume has his autofs stuff at a level below the top by the look of it.

I'm reluctant to add an option as if there's choice then people will 
always be asking questions about it. I like to keep the configuration to a 
minimum.

I'm not clear on the performance implications of this yet so I'll change 
it and wait till someone can give feedback on cost. LDAP servers seem to 
cope fairly well with this type of query anyway (default option for 
ldapsearch) and it's only done at startup, to get the base for future 
searches.

Ian