[SSSD] [PATCHES] be_refresh: support users and groups

Pavel Březina pbrezina at redhat.com
Wed Mar 4 08:57:02 UTC 2015


On 03/03/2015 01:03 PM, Jakub Hrozek wrote:
> On Thu, Feb 19, 2015 at 12:18:13PM +0100, Pavel Březina wrote:
>> https://fedorahosted.org/sssd/ticket/2346
>>
>> Hi,
>> the first patch add support for subdomains, which was missing in current
>> implementation. It probably wasn't a big deal with netgroups but I think it
>> would be with users and groups.
>>
>> The majority of these patches refactors the current be_refresh to make
>> support of other object types than netgroups a lot easier and
>> straightforward.
>
> Code review so far, no testing was done. I'm just sending my notes so
> you can reply while I test the patches.
>
>>  From e744e3dbeab301c149accbff521cbc5232c6a48d Mon Sep 17 00:00:00 2001
>> From: =?UTF-8?q?Pavel=20B=C5=99ezina?= <pbrezina at redhat.com>
>> Date: Fri, 13 Feb 2015 13:16:04 +0100
>> Subject: [PATCH 1/6] be_refresh: refresh all domains in backend
>>
>> ---
>>   src/providers/dp_refresh.c        | 82 ++++++++++++++++++++++++---------------
>
> Can you also write a test for this module while we're changing it?

Well, there is a ticket for it. But the main problem is how do you want 
to write a meaningful test?

The be_refresh module consist of:

be_refresh_send (public)
- fetch expired objects (internal)
- refresh them (internal)

The only thing you can test is that sdap was actually called which isn't 
really worth the time IMHO. Or do you have other idea?

>>   src/providers/dp_refresh.h        |  1 +
>>   src/providers/ldap/ldap_common.h  |  1 +
>>   src/providers/ldap/sdap_refresh.c | 15 +++++--
>>   4 files changed, 64 insertions(+), 35 deletions(-)
>>
>> diff --git a/src/providers/dp_refresh.c b/src/providers/dp_refresh.c
>> index 817b6213ca47bba3fa34ce28fdcd1621d349b651..bd02d0cd99f9a061109f0c17797c6e018d602dc5 100644
>> --- a/src/providers/dp_refresh.c
>> +++ b/src/providers/dp_refresh.c
>> @@ -117,6 +117,7 @@ typedef errno_t
>>
>>
>>   struct be_refresh_cb {
>> +    const char *name;
>>       bool enabled;
>>       be_refresh_get_values_t get_values;
>>       be_refresh_send_t send_fn;
>> @@ -137,6 +138,7 @@ struct be_refresh_ctx *be_refresh_ctx_init(TALLOC_CTX *mem_ctx)
>>           return NULL;
>>       }
>>
>> +    ctx->callbacks[BE_REFRESH_TYPE_NETGROUPS].name = "netgroups";
>>       ctx->callbacks[BE_REFRESH_TYPE_NETGROUPS].get_values \
>>           = be_refresh_get_netgroups;
>
> In retrospective I wish we didn't use a pointer function here. A pointer
> function might make sense in cases where the function might be from a
> different provider for instance, but the sysdb expiration will be always
> read in the same manner. A switch-case here would be much simpler and
> allow for easier code browsing.

Yes... do you want me to write a patch?

> I also wonder if be_refresh_send needs the be_ctx parameter at all,
> since be_ctx is also part of the be_ptask structure?

It is signature for be_ptask send functions.

>
> I know the two remarks above are not strictly related to your patches,
> but I had to re-read the be_refresh code to make sure I understand it :-)
>
> The rest of the patch looks good to me.

I will wait until you finish testing, since the issues are below are 
trivial.

>
>>  From 85b2f150134e89672f764e7b267f7f47a285938f Mon Sep 17 00:00:00 2001
>> From: =?UTF-8?q?Pavel=20B=C5=99ezina?= <pbrezina at redhat.com>
>> Date: Fri, 13 Feb 2015 13:49:17 +0100
>> Subject: [PATCH 2/6] sdap_handle_acct_req_send: remove be_req
>>
>> be_req was used only as a talloc context for subreq. This memory context
>> was replace by state of the parent request which is more suitable for
>> tevent coding style.
>
> ACK. I traced the parent contexts all the way up and confirmed that in
> fact nothing changes, be_req is always used as the top-level state.
>
>>  From a92f1ffac23e90cfff848a05ef9611fa021b78a4 Mon Sep 17 00:00:00 2001
>> From: =?UTF-8?q?Pavel=20B=C5=99ezina?= <pbrezina at redhat.com>
>> Date: Mon, 16 Feb 2015 13:42:02 +0100
>> Subject: [PATCH 3/6] be_refresh: refactor netgroups refresh
>
> [...]
>
>> +    state->account_req = talloc_zero(state, struct be_acct_req);
>> +    if (state->account_req == NULL) {
>> +        ret = ENOMEM;
>> +        goto immediately;
>> +    }
>> +
>> +    state->account_req->entry_type = entry_type;
>> +    state->account_req->attr_type = BE_ATTR_ALL;
>
> I know that this attribute is currently not used really, but can you use
> BE_ATTR_CORE here instead? IIRC all the other invokers use _CORE as
> well, so using it here as well might make any future refactoring easier.
>
>>  From 4cff90263b247d8701de33ad1cf76db95698659a Mon Sep 17 00:00:00 2001
>> From: =?UTF-8?q?Pavel=20B=C5=99ezina?= <pbrezina at redhat.com>
>> Date: Thu, 19 Feb 2015 12:04:09 +0100
>> Subject: [PATCH 4/6] be_refresh: add sdap_refresh_init
>
> [...]
>
>> @@ -179,14 +179,11 @@ static int ldap_id_init_internal(struct be_ctx *bectx,
>>       }
>>
>>       /* setup periodical refresh of expired records */
>> -    ret = be_refresh_add_cb(bectx->refresh_ctx, BE_REFRESH_TYPE_NETGROUPS,
>> -                            sdap_refresh_netgroups_send,
>> -                            sdap_refresh_netgroups_recv,
>> -                            ctx);
>> -    if (ret != EOK && ret != EEXIST) {
>> -        DEBUG(SSSDBG_MINOR_FAILURE, "Periodical refresh of netgroups "
>> -              "will not work [%d]: %s\n", ret, strerror(ret));
>> -    }
>> +    ret = sdap_refresh_init(bectx->refresh_ctx, ctx);
>> +        if (ret != EOK && ret != EEXIST) {
>> +            DEBUG(SSSDBG_MINOR_FAILURE, "Periodical refresh "
>> +                  "will not work [%d]: %s\n", ret, strerror(ret));
>> +        }
>
> Wrong indent, otherwise patch LGTM.
>
>>  From 42efd926f41c4691758e5871fd3a3330802e7f41 Mon Sep 17 00:00:00 2001
>> From: =?UTF-8?q?Pavel=20B=C5=99ezina?= <pbrezina at redhat.com>
>> Date: Thu, 19 Feb 2015 12:00:52 +0100
>> Subject: [PATCH 5/6] be_refresh: support users
>>
>> Resolves:
>> https://fedorahosted.org/sssd/ticket/2346
>> ---
>>   src/db/sysdb.c                    |  7 +++++++
>>   src/db/sysdb.h                    |  2 ++
>>   src/providers/dp_refresh.c        | 23 +++++++++++++++++++++++
>>   src/providers/dp_refresh.h        |  1 +
>>   src/providers/ldap/sdap_refresh.c | 29 +++++++++++++++++++++++++++++
>>   5 files changed, 62 insertions(+)
>>
>> diff --git a/src/db/sysdb.c b/src/db/sysdb.c
>> index 61a2240016b5cb77e6fbbc3286fd1a194c5a0b48..2bb4a41aa4a9e6201ac27ac8d9a1803c1fb5c43e 100644
>> --- a/src/db/sysdb.c
>> +++ b/src/db/sysdb.c
>> @@ -172,6 +172,13 @@ struct ldb_dn *sysdb_user_dn(TALLOC_CTX *mem_ctx, struct sss_domain_info *dom,
>>       return dn;
>>   }
>>
>> +struct ldb_dn *sysdb_user_base_dn(TALLOC_CTX *mem_ctx,
>> +                                  struct sss_domain_info *dom)
>> +{
>> +    return ldb_dn_new_fmt(mem_ctx, dom->sysdb->ldb,
>> +                          SYSDB_TMPL_USER_BASE, dom->name);
>> +}
>> +
>
> I don't like adding a convenience function and then only using it on one
> place. Either use ldb_dn_new_fmt() directly or convert the uses of
> ldb_dn_new_fmt to this new convenience function.
>
>
>>  From 20bcc5371af8e429b46b0289ef9fd517c1e932c4 Mon Sep 17 00:00:00 2001
>> From: =?UTF-8?q?Pavel=20B=C5=99ezina?= <pbrezina at redhat.com>
>> Date: Thu, 19 Feb 2015 12:10:23 +0100
>> Subject: [PATCH 6/6] be_refresh: support groups
>>
>> Resolves:
>> https://fedorahosted.org/sssd/ticket/2346
>
> Same comment about the convenience function, otherwise LGTM.
>
> CI passed: http://sssd-ci.duckdns.org/logs/job/8/57/summary.html
>
> Did you test the refresh with really large (hundreds at least, better
> thousands of entries) directories?
> _______________________________________________
> sssd-devel mailing list
> sssd-devel at lists.fedorahosted.org
> https://lists.fedorahosted.org/mailman/listinfo/sssd-devel
>




More information about the sssd-devel mailing list