While attempting to extract the SSSD's HBAC rule evaluation into a common library, I discovered several shortcomings. Some were related to extracting the library, others were related to the rule evaluation itself.
Related to extracting the library, I needed for the code to be able to identify and resolve without having direct access to LDAP or the LDB. This meant I could not rely on the memberOf objects to be valid.
As for the rule evaluation, I discovered that we were only caching for offline use those remote hosts that had previously attempted to log in. This meant that if the IPA server was unreachable, valid users would be unable to log in from some valid machines.
After careful consideration (and a few false starts), I decided that a total rewrite was necessary. So here it is.
The major changes are these:
1) We have moved to always performing a full enumeration of known IPA hosts and HBAC services. This actually results in a net reduction in round-trips to the LDAP server, which was a primary bottleneck. Additionally, this means that we also have a full understanding when offline as to whether a remote host is permitted access or not.
2) I've added a configurable timeout for hbac rule refreshes to the LDAP server. Previously, if we were online, EVERY pam_account() request would result in at least one LDAP round-trip. By default, we'll only refresh the HBAC rules every five seconds now.
3) I broke the acquisition and processing of the rules, hosts and services into separate C files, as ipa_access.c was getting to be much to large to manage.
The overall view of the code (when online) is this: A request comes in, we call to LDAP to get, in order, the set of known hosts, the set of known PAM services and then finally the set of *enabled* rules with the current ipa_hostname as the targethost.
And now, the patches:
Patch 0001: The new standalone evaluator library. It's very simple at the moment (but it will become much more complex in the near future when timerule evaluation is added). Jakub, Rob and I agreed on the interface and Jakub is creating Python bindings for it.
Patch 0002: This is the meat of the new IPA access backend. All of the helper functions necessary for looking up, storing and converting the hosts, rules and services are in this patch.
Patch 0003: This is not really a patch as much as it is a trash can. I split this from Patch 0004 to make the latter more reviewable. Patch 0003 contains only the deletions of the old access provider implementation.
Patch 0004: Adds the new IPA access provider, taking advantage of the helper functions from Patch 0002.
Patch 0005: Adds a refresh timeout to reduce round-trips to IPA while online.
On 06/07/2011 04:59 AM, Stephen Gallagher wrote:
While attempting to extract the SSSD's HBAC rule evaluation into a common library, I discovered several shortcomings. Some were related to extracting the library, others were related to the rule evaluation itself.
Related to extracting the library, I needed for the code to be able to identify and resolve without having direct access to LDAP or the LDB. This meant I could not rely on the memberOf objects to be valid.
As for the rule evaluation, I discovered that we were only caching for offline use those remote hosts that had previously attempted to log in. This meant that if the IPA server was unreachable, valid users would be unable to log in from some valid machines.
After careful consideration (and a few false starts), I decided that a total rewrite was necessary. So here it is.
The major changes are these:
- We have moved to always performing a full enumeration of known IPA
hosts and HBAC services. This actually results in a net reduction in round-trips to the LDAP server, which was a primary bottleneck. Additionally, this means that we also have a full understanding when offline as to whether a remote host is permitted access or not.
- I've added a configurable timeout for hbac rule refreshes to the LDAP
server. Previously, if we were online, EVERY pam_account() request would result in at least one LDAP round-trip. By default, we'll only refresh the HBAC rules every five seconds now.
- I broke the acquisition and processing of the rules, hosts and
services into separate C files, as ipa_access.c was getting to be much to large to manage.
The overall view of the code (when online) is this: A request comes in, we call to LDAP to get, in order, the set of known hosts, the set of known PAM services and then finally the set of *enabled* rules with the current ipa_hostname as the targethost.
And now, the patches:
Patch 0001: The new standalone evaluator library. It's very simple at the moment (but it will become much more complex in the near future when timerule evaluation is added). Jakub, Rob and I agreed on the interface and Jakub is creating Python bindings for it.
I'll rebase the bindings on top of these patches and send to the list.
Patch 0002: This is the meat of the new IPA access backend. All of the helper functions necessary for looking up, storing and converting the hosts, rules and services are in this patch.
Patch 0003: This is not really a patch as much as it is a trash can. I split this from Patch 0004 to make the latter more reviewable. Patch 0003 contains only the deletions of the old access provider implementation.
Patch 0004: Adds the new IPA access provider, taking advantage of the helper functions from Patch 0002.
Patch 0005: Adds a refresh timeout to reduce round-trips to IPA while online.
Nack, msgs2attrs_array() is not defined:
src/providers/ipa/ipa_access.c: In function 'hbac_get_cached_rules': src/providers/ipa/ipa_access.c:589:5: error: implicit declaration of function 'msgs2attrs_array' [-Werror=implicit-function-declaration] cc1: some warnings being treated as errors
On Tue, 2011-06-07 at 10:41 +0200, Jakub Hrozek wrote:
On 06/07/2011 04:59 AM, Stephen Gallagher wrote:
While attempting to extract the SSSD's HBAC rule evaluation into a common library, I discovered several shortcomings. Some were related to extracting the library, others were related to the rule evaluation itself.
Related to extracting the library, I needed for the code to be able to identify and resolve without having direct access to LDAP or the LDB. This meant I could not rely on the memberOf objects to be valid.
As for the rule evaluation, I discovered that we were only caching for offline use those remote hosts that had previously attempted to log in. This meant that if the IPA server was unreachable, valid users would be unable to log in from some valid machines.
After careful consideration (and a few false starts), I decided that a total rewrite was necessary. So here it is.
The major changes are these:
- We have moved to always performing a full enumeration of known IPA
hosts and HBAC services. This actually results in a net reduction in round-trips to the LDAP server, which was a primary bottleneck. Additionally, this means that we also have a full understanding when offline as to whether a remote host is permitted access or not.
- I've added a configurable timeout for hbac rule refreshes to the LDAP
server. Previously, if we were online, EVERY pam_account() request would result in at least one LDAP round-trip. By default, we'll only refresh the HBAC rules every five seconds now.
- I broke the acquisition and processing of the rules, hosts and
services into separate C files, as ipa_access.c was getting to be much to large to manage.
The overall view of the code (when online) is this: A request comes in, we call to LDAP to get, in order, the set of known hosts, the set of known PAM services and then finally the set of *enabled* rules with the current ipa_hostname as the targethost.
And now, the patches:
Patch 0001: The new standalone evaluator library. It's very simple at the moment (but it will become much more complex in the near future when timerule evaluation is added). Jakub, Rob and I agreed on the interface and Jakub is creating Python bindings for it.
I'll rebase the bindings on top of these patches and send to the list.
Patch 0002: This is the meat of the new IPA access backend. All of the helper functions necessary for looking up, storing and converting the hosts, rules and services are in this patch.
Patch 0003: This is not really a patch as much as it is a trash can. I split this from Patch 0004 to make the latter more reviewable. Patch 0003 contains only the deletions of the old access provider implementation.
Patch 0004: Adds the new IPA access provider, taking advantage of the helper functions from Patch 0002.
Patch 0005: Adds a refresh timeout to reduce round-trips to IPA while online.
Nack, msgs2attrs_array() is not defined:
src/providers/ipa/ipa_access.c: In function 'hbac_get_cached_rules': src/providers/ipa/ipa_access.c:589:5: error: implicit declaration of function 'msgs2attrs_array' [-Werror=implicit-function-declaration] cc1: some warnings being treated as errors
Whoops, I forgot I reordered some of my patches. I missed two at the beginning.
Patch 0001: Retrieve the group name as well as the GID during sysdb_initgroups() (used later for getting the user memberships of a user for evaluating the rules)
Patch 0002: Add a helper function to convert a list of ldb_messages to sysdb_attrs.
The other patches are the same as the previous email, but numbered +2.
On Tue, 2011-06-07 at 08:11 -0400, Stephen Gallagher wrote:
On Tue, 2011-06-07 at 10:41 +0200, Jakub Hrozek wrote:
On 06/07/2011 04:59 AM, Stephen Gallagher wrote:
While attempting to extract the SSSD's HBAC rule evaluation into a common library, I discovered several shortcomings. Some were related to extracting the library, others were related to the rule evaluation itself.
Related to extracting the library, I needed for the code to be able to identify and resolve without having direct access to LDAP or the LDB. This meant I could not rely on the memberOf objects to be valid.
As for the rule evaluation, I discovered that we were only caching for offline use those remote hosts that had previously attempted to log in. This meant that if the IPA server was unreachable, valid users would be unable to log in from some valid machines.
After careful consideration (and a few false starts), I decided that a total rewrite was necessary. So here it is.
The major changes are these:
- We have moved to always performing a full enumeration of known IPA
hosts and HBAC services. This actually results in a net reduction in round-trips to the LDAP server, which was a primary bottleneck. Additionally, this means that we also have a full understanding when offline as to whether a remote host is permitted access or not.
- I've added a configurable timeout for hbac rule refreshes to the LDAP
server. Previously, if we were online, EVERY pam_account() request would result in at least one LDAP round-trip. By default, we'll only refresh the HBAC rules every five seconds now.
- I broke the acquisition and processing of the rules, hosts and
services into separate C files, as ipa_access.c was getting to be much to large to manage.
The overall view of the code (when online) is this: A request comes in, we call to LDAP to get, in order, the set of known hosts, the set of known PAM services and then finally the set of *enabled* rules with the current ipa_hostname as the targethost.
And now, the patches:
Patch 0001: The new standalone evaluator library. It's very simple at the moment (but it will become much more complex in the near future when timerule evaluation is added). Jakub, Rob and I agreed on the interface and Jakub is creating Python bindings for it.
I'll rebase the bindings on top of these patches and send to the list.
Patch 0002: This is the meat of the new IPA access backend. All of the helper functions necessary for looking up, storing and converting the hosts, rules and services are in this patch.
Patch 0003: This is not really a patch as much as it is a trash can. I split this from Patch 0004 to make the latter more reviewable. Patch 0003 contains only the deletions of the old access provider implementation.
Patch 0004: Adds the new IPA access provider, taking advantage of the helper functions from Patch 0002.
Patch 0005: Adds a refresh timeout to reduce round-trips to IPA while online.
Nack, msgs2attrs_array() is not defined:
src/providers/ipa/ipa_access.c: In function 'hbac_get_cached_rules': src/providers/ipa/ipa_access.c:589:5: error: implicit declaration of function 'msgs2attrs_array' [-Werror=implicit-function-declaration] cc1: some warnings being treated as errors
Whoops, I forgot I reordered some of my patches. I missed two at the beginning.
Patch 0001: Retrieve the group name as well as the GID during sysdb_initgroups() (used later for getting the user memberships of a user for evaluating the rules)
Patch 0002: Add a helper function to convert a list of ldb_messages to sysdb_attrs.
The other patches are the same as the previous email, but numbered +2.
This is a generic NACK due to a concern that seem is not properly considered here.
As can be seen by other patches on the list we are going to support handling MS-PACs. In MS-PACs we do not have group names, only SIDs that we can transalte back to GIDs and eventually names. But at login names may not be available unless a cached group entry is found.
In this case the HBAC evaluator need to be very careful if DENY rules are present (I hate deny rules, they are evil :-), and we are in offline mode, not able to resolve incomplete groups.
If the user is member of any unresolved groups that have fake names, we must flag it, and during HBAC evaluation if we encounter any DENY rule that can't resolve the group it refers to we need to virtually match it and throw access denied at the user. This is because one of the unresolved groups could be the group in DENY rule, and because we are not able to resolve it we must consider the worst case (a match).
If all DENY rules can reference complete cached groups, or the user is not member of any "fake" group the rule can be evaluated as UNMATCHED if that's the case as we can be confident the group referenced by the rule is not a group the user is member of.
Simo.
On Tue, 2011-06-07 at 09:03 -0400, Simo Sorce wrote:
On Tue, 2011-06-07 at 08:11 -0400, Stephen Gallagher wrote:
On Tue, 2011-06-07 at 10:41 +0200, Jakub Hrozek wrote:
On 06/07/2011 04:59 AM, Stephen Gallagher wrote:
While attempting to extract the SSSD's HBAC rule evaluation into a common library, I discovered several shortcomings. Some were related to extracting the library, others were related to the rule evaluation itself.
Related to extracting the library, I needed for the code to be able to identify and resolve without having direct access to LDAP or the LDB. This meant I could not rely on the memberOf objects to be valid.
As for the rule evaluation, I discovered that we were only caching for offline use those remote hosts that had previously attempted to log in. This meant that if the IPA server was unreachable, valid users would be unable to log in from some valid machines.
After careful consideration (and a few false starts), I decided that a total rewrite was necessary. So here it is.
The major changes are these:
- We have moved to always performing a full enumeration of known IPA
hosts and HBAC services. This actually results in a net reduction in round-trips to the LDAP server, which was a primary bottleneck. Additionally, this means that we also have a full understanding when offline as to whether a remote host is permitted access or not.
- I've added a configurable timeout for hbac rule refreshes to the LDAP
server. Previously, if we were online, EVERY pam_account() request would result in at least one LDAP round-trip. By default, we'll only refresh the HBAC rules every five seconds now.
- I broke the acquisition and processing of the rules, hosts and
services into separate C files, as ipa_access.c was getting to be much to large to manage.
The overall view of the code (when online) is this: A request comes in, we call to LDAP to get, in order, the set of known hosts, the set of known PAM services and then finally the set of *enabled* rules with the current ipa_hostname as the targethost.
And now, the patches:
Patch 0001: The new standalone evaluator library. It's very simple at the moment (but it will become much more complex in the near future when timerule evaluation is added). Jakub, Rob and I agreed on the interface and Jakub is creating Python bindings for it.
I'll rebase the bindings on top of these patches and send to the list.
Patch 0002: This is the meat of the new IPA access backend. All of the helper functions necessary for looking up, storing and converting the hosts, rules and services are in this patch.
Patch 0003: This is not really a patch as much as it is a trash can. I split this from Patch 0004 to make the latter more reviewable. Patch 0003 contains only the deletions of the old access provider implementation.
Patch 0004: Adds the new IPA access provider, taking advantage of the helper functions from Patch 0002.
Patch 0005: Adds a refresh timeout to reduce round-trips to IPA while online.
Nack, msgs2attrs_array() is not defined:
src/providers/ipa/ipa_access.c: In function 'hbac_get_cached_rules': src/providers/ipa/ipa_access.c:589:5: error: implicit declaration of function 'msgs2attrs_array' [-Werror=implicit-function-declaration] cc1: some warnings being treated as errors
Whoops, I forgot I reordered some of my patches. I missed two at the beginning.
Patch 0001: Retrieve the group name as well as the GID during sysdb_initgroups() (used later for getting the user memberships of a user for evaluating the rules)
Patch 0002: Add a helper function to convert a list of ldb_messages to sysdb_attrs.
The other patches are the same as the previous email, but numbered +2.
This is a generic NACK due to a concern that seem is not properly considered here.
As can be seen by other patches on the list we are going to support handling MS-PACs. In MS-PACs we do not have group names, only SIDs that we can transalte back to GIDs and eventually names. But at login names may not be available unless a cached group entry is found.
In this case the HBAC evaluator need to be very careful if DENY rules are present (I hate deny rules, they are evil :-), and we are in offline mode, not able to resolve incomplete groups.
If the user is member of any unresolved groups that have fake names, we must flag it, and during HBAC evaluation if we encounter any DENY rule that can't resolve the group it refers to we need to virtually match it and throw access denied at the user. This is because one of the unresolved groups could be the group in DENY rule, and because we are not able to resolve it we must consider the worst case (a match).
If all DENY rules can reference complete cached groups, or the user is not member of any "fake" group the rule can be evaluated as UNMATCHED if that's the case as we can be confident the group referenced by the rule is not a group the user is member of.
Some more data after an IRC discussion with Stephen.
The current library have no way to directly consult the user database, so it has no way to know if a group exist and is resolvable, and therefore not one of the unresolved groups of the users.
This would force us to DENY access even though we have enough data outside of the evaluation library to rule the DENY rule as unmatching.
In order to solve this problem the library should define a callback handler that can be set by the caller so that the library can ask for user/group information.
This way on deny rules if the group is not found and the user is marked to have unresolved groups, the callback can be called and it can be verified if this group otherwise exists or if we have to consider it as potentially one of the unresolved groups of the user.
If the callback is limited to this use only then callers that can assert fully resolved groups for the user from the get-go, can pass in NULL for the callback. Callers that also do not care about getting a DENY where it may have not matched can also do so as a NULL will be interpreted as group not found result causing a match in the deny operation evaluation.
Simo.
On 06/07/2011 02:11 PM, Stephen Gallagher wrote:
On Tue, 2011-06-07 at 10:41 +0200, Jakub Hrozek wrote:
On 06/07/2011 04:59 AM, Stephen Gallagher wrote:
While attempting to extract the SSSD's HBAC rule evaluation into a common library, I discovered several shortcomings. Some were related to extracting the library, others were related to the rule evaluation itself.
Related to extracting the library, I needed for the code to be able to identify and resolve without having direct access to LDAP or the LDB. This meant I could not rely on the memberOf objects to be valid.
As for the rule evaluation, I discovered that we were only caching for offline use those remote hosts that had previously attempted to log in. This meant that if the IPA server was unreachable, valid users would be unable to log in from some valid machines.
After careful consideration (and a few false starts), I decided that a total rewrite was necessary. So here it is.
The major changes are these:
- We have moved to always performing a full enumeration of known IPA
hosts and HBAC services. This actually results in a net reduction in round-trips to the LDAP server, which was a primary bottleneck. Additionally, this means that we also have a full understanding when offline as to whether a remote host is permitted access or not.
- I've added a configurable timeout for hbac rule refreshes to the LDAP
server. Previously, if we were online, EVERY pam_account() request would result in at least one LDAP round-trip. By default, we'll only refresh the HBAC rules every five seconds now.
- I broke the acquisition and processing of the rules, hosts and
services into separate C files, as ipa_access.c was getting to be much to large to manage.
The overall view of the code (when online) is this: A request comes in, we call to LDAP to get, in order, the set of known hosts, the set of known PAM services and then finally the set of *enabled* rules with the current ipa_hostname as the targethost.
And now, the patches:
Patch 0001: The new standalone evaluator library. It's very simple at the moment (but it will become much more complex in the near future when timerule evaluation is added). Jakub, Rob and I agreed on the interface and Jakub is creating Python bindings for it.
I'll rebase the bindings on top of these patches and send to the list.
Patch 0002: This is the meat of the new IPA access backend. All of the helper functions necessary for looking up, storing and converting the hosts, rules and services are in this patch.
Patch 0003: This is not really a patch as much as it is a trash can. I split this from Patch 0004 to make the latter more reviewable. Patch 0003 contains only the deletions of the old access provider implementation.
Patch 0004: Adds the new IPA access provider, taking advantage of the helper functions from Patch 0002.
Patch 0005: Adds a refresh timeout to reduce round-trips to IPA while online.
Nack, msgs2attrs_array() is not defined:
src/providers/ipa/ipa_access.c: In function 'hbac_get_cached_rules': src/providers/ipa/ipa_access.c:589:5: error: implicit declaration of function 'msgs2attrs_array' [-Werror=implicit-function-declaration] cc1: some warnings being treated as errors
Whoops, I forgot I reordered some of my patches. I missed two at the beginning.
I know the patches are going to change, but at least some of the comments below would apply to the next iteration as well.
Patch 0001: Retrieve the group name as well as the GID during sysdb_initgroups() (used later for getting the user memberships of a user for evaluating the rules)
Ack provided that the patch is still valid in the PAC context - that wasn't clear to me from the IRC discussion between you and Simo.
Patch 0002: Add a helper function to convert a list of ldb_messages to sysdb_attrs.
Ack
The other patches are the same as the previous email, but numbered +2.
Patch 3:
Nack,
in sssd.spec.in, "%defattr(-,root,root-)" is missing a coma between the third and fourth paramater.
Is it necessary to provide an explicit Requires: on libtalloc? Usually RPM is good in guessing the dependencies automatically. This might be moot if you're getting rid of talloc as mentioned on IRC.
I'm curious about using size_t for loop counter in hbac_evaluate_element() - it's certainly not wrong, but why not simply use unsigned long?
Patch 4: In overall, the code reads really good.
Why do you talloc_steal the items of the array sdap_get_generic_recv() returns? The memory hierarchy seems OK to me in sdap_get_generic_* functions.
What is the rationale of ever passing delete_subdir=false to ipa_hbac_save_list()? When you're sure you're just priming the sysdb? I don't see it ever used in the code.
In ipa_hbac_save_list() the, error message "IPA_UNIQUE_ID not found" should probably say "naming attribute not found", it does not have to be IPA unique ID.
In hbac_service_attrs_to_rule() a comment says "/* First check if this is a specific host */". I think it should say "..specific service".
In ipa_hbac_rule_info_send(), please use IPA_HBAC_RULE instead of hardcoding ipaHBACRule.
nitpick - in hbac_ctx_to_eval_request() the block after "if (pd->rhost == NULL || pd->rhost[0] == '\0') {" is indented one level too deep.
A comment in hbac_eval_host_element() says "/* Find the service groups */", it should probably say "host groups".
Patch 5: Ack
Patch 6: just nitpicking, but can you add a blank line between variable declarations and the ipa_hbac_rule_info_recv() call in hbac_sysdb_save()?
Patch 7: Nack, the new option belongs to SSSDConfigAPI.py, too.
New patches attached with some significant changes. After much discussion, we decided that, rather than implement a facility for identifying "unresolved groups" for the user, we would instead perform a group lookup during the initial pull-down of the HBAC rules. This way, we know that all groups that are relevant to HBAC rules are available on the system. So if a user is a member of a group that is not resolvable, we don't have to worry that it might match a DENY rule.
On Fri, 2011-06-10 at 21:31 +0200, Jakub Hrozek wrote:
I know the patches are going to change, but at least some of the comments below would apply to the next iteration as well.
Patch 0001: Retrieve the group name as well as the GID during sysdb_initgroups() (used later for getting the user memberships of a user for evaluating the rules)
Ack provided that the patch is still valid in the PAC context - that wasn't clear to me from the IRC discussion between you and Simo.
This patch is unchanged.
Patch 0002: Add a helper function to convert a list of ldb_messages to sysdb_attrs.
Ack
This patch is unchanged.
The other patches are the same as the previous email, but numbered +2.
Patch 3:
Nack,
in sssd.spec.in, "%defattr(-,root,root-)" is missing a coma between the third and fourth paramater.
Good catch. Fixed.
Is it necessary to provide an explicit Requires: on libtalloc? Usually RPM is good in guessing the dependencies automatically. This might be moot if you're getting rid of talloc as mentioned on IRC.
I've removed the requirement on talloc in the evaluator, so it's also been removed from the spec file. (But you're right, the explicit Requires: would have been unnecessary).
I'm curious about using size_t for loop counter in hbac_evaluate_element() - it's certainly not wrong, but why not simply use unsigned long?
It's just a habit I've gotten into when dealing with loops based on an array index.
Patch 4: In overall, the code reads really good.
Why do you talloc_steal the items of the array sdap_get_generic_recv() returns? The memory hierarchy seems OK to me in sdap_get_generic_* functions.
That was in the old code and I didn't want to rock the boat, but you're right. That's wasteful. If the hierarchy is ever wrong here it should be a bug. Fixed.
What is the rationale of ever passing delete_subdir=false to ipa_hbac_save_list()? When you're sure you're just priming the sysdb? I don't see it ever used in the code.
ipa_hbac_save_list() was copied verbatim from hbac_save_list() (Removed in Patch 5). I didn't change it at all. I'd rather leave it in for the future. At some point we'll probably want to optimize this lookup to use LastUSN to just get the delta of changes to the rules, hosts and services, rather than enumerating them fully.
In ipa_hbac_save_list() the, error message "IPA_UNIQUE_ID not found" should probably say "naming attribute not found", it does not have to be IPA unique ID.
Fixed.
In hbac_service_attrs_to_rule() a comment says "/* First check if this is a specific host */". I think it should say "..specific service".
Copy-paste error. Fixed.
In ipa_hbac_rule_info_send(), please use IPA_HBAC_RULE instead of hardcoding ipaHBACRule.
Fixed.
nitpick - in hbac_ctx_to_eval_request() the block after "if (pd->rhost == NULL || pd->rhost[0] == '\0') {" is indented one level too deep.
Fixed.
A comment in hbac_eval_host_element() says "/* Find the service groups */", it should probably say "host groups".
Copy-paste error. Fixed.
Patch 5: Ack
Patch is unchanged.
Patch 6: just nitpicking, but can you add a blank line between variable declarations and the ipa_hbac_rule_info_recv() call in hbac_sysdb_save()?
This code was rewritten in patch 8. It should be fine now.
Patch 7: Nack, the new option belongs to SSSDConfigAPI.py, too.
Added.
New Patches: I added these as new patches to make the review easier.
Patch 0008: Adds a new routine to update any expired or missing group entries in the cache that are relevant to the HBAC rules. I'm "cheating" a little bit here for a significant performance increase. Because the originalDN attribute is a DN, it cannot be made into a search filter normally, so in order to look these up it would require a series of BASE searches. Then I'd need to gather the results and write a new cache update routine. What I did instead was abuse the fact that IPA has a well-known construct for usergroups in its tree. I parse the originalDN to see if it's a group and extract the group name. I then use this to construct a single search request with all of the missing or expired groups and send it through sdap_get_groups_send(), which has the advantage of already being designed to handle multiple (and recursive) groups and cluster them for a single transaction to save them. Admittedly, this will need to be modified if we ever opt to support the HBAC provider against non-IPA LDAP servers, but for now the performance gains are enormous.
Patch 0009: While fixing these, I realized that if there were no applicable rules for the system, it was returning PAM_SYSTEM_ERR. It makes more sense to return PAM_PERM_DENIED and purge any existing rules from the cache.
On Fri, 2011-06-24 at 09:54 -0400, Stephen Gallagher wrote:
New patches attached with some significant changes. After much discussion, we decided that, rather than implement a facility for identifying "unresolved groups" for the user, we would instead perform a group lookup during the initial pull-down of the HBAC rules. This way, we know that all groups that are relevant to HBAC rules are available on the system. So if a user is a member of a group that is not resolvable, we don't have to worry that it might match a DENY rule.
NACK
New patches fail to take in account non-posix groups.
Simo.
On Fri, 2011-06-24 at 11:15 -0400, Simo Sorce wrote:
On Fri, 2011-06-24 at 09:54 -0400, Stephen Gallagher wrote:
New patches attached with some significant changes. After much discussion, we decided that, rather than implement a facility for identifying "unresolved groups" for the user, we would instead perform a group lookup during the initial pull-down of the HBAC rules. This way, we know that all groups that are relevant to HBAC rules are available on the system. So if a user is a member of a group that is not resolvable, we don't have to worry that it might match a DENY rule.
NACK
New patches fail to take in account non-posix groups.
I've added three new patches to account for the non-POSIX groups. I've also removed the old patch 0001, as with these new changes it is no longer necessary.
Patch 0009 and 0010 make no functional changes, they just reorganize the code a little bit to make 0012 easier.
Patch 0012: This makes two changes. 1) Change the construction of the eval request to use ipa_get_groupname() and the originalMemberOf attribute to create the list of groups, rather than calling initgroups.
2) In hbac_user_attrs_to_rule(), if the member is not detected as a POSIX user or group, check to see if it matches with ipa_get_groupname(), indicating that it is a non-POSIX group. There is no need to detect non-POSIX users here, as they wouldn't be eligible to log in anyway.
On 06/24/2011 02:24 PM, Stephen Gallagher wrote:
On Fri, 2011-06-24 at 11:15 -0400, Simo Sorce wrote:
On Fri, 2011-06-24 at 09:54 -0400, Stephen Gallagher wrote:
New patches attached with some significant changes. After much discussion, we decided that, rather than implement a facility for identifying "unresolved groups" for the user, we would instead perform a group lookup during the initial pull-down of the HBAC rules. This way, we know that all groups that are relevant to HBAC rules are available on the system. So if a user is a member of a group that is not resolvable, we don't have to worry that it might match a DENY rule.
NACK
New patches fail to take in account non-posix groups.
I've added three new patches to account for the non-POSIX groups. I've also removed the old patch 0001, as with these new changes it is no longer necessary.
Patch 0009 and 0010 make no functional changes, they just reorganize the code a little bit to make 0012 easier.
Patch 0012: This makes two changes.
- Change the construction of the eval request to use
ipa_get_groupname() and the originalMemberOf attribute to create the list of groups, rather than calling initgroups.
- In hbac_user_attrs_to_rule(), if the member is not detected as a
POSIX user or group, check to see if it matches with ipa_get_groupname(), indicating that it is a non-POSIX group. There is no need to detect non-POSIX users here, as they wouldn't be eligible to log in anyway.
0002: ACK
0003: NACK The defattr in "%files -n libipa_hbac-devel" is missing a comma before the last parameter, it says "%defattr(-,root,root-)".
hbac_free_error(): I wonder if it makes sense to add an explicit "if (!error) return". I think it is expected that most free functions accept NULL as a noop and it is quite useful for goto-cleanup cases.
0004: ACK
0005: ACK
0006: ACK
0007: ACK, altough the description of ipa_hbac_refresh was added to config/SSSDConfig.py in patch 0008. Feel free to fix or not, I don't think this should be a nack as the patchset will be applied as a whole anyway.
For the rest of the patches - the code seems to be OK sans the one little issue below. I will defer the "architectural" review to Simo as the original comments were based on his PAC development experience.
0008: NACK Please free tmp_ctx in the "immediate:" label in ipa_hbac_update_groups_send()
The "/* Not found as a user; Check Groups*/" seems to be misleading, the code never checks users.
0009: ACK
0010: ACK
0011: ACK
0012: ACK
On Mon, 2011-06-27 at 18:31 -0400, Jakub Hrozek wrote:
0002: ACK
0003: NACK The defattr in "%files -n libipa_hbac-devel" is missing a comma before the last parameter, it says "%defattr(-,root,root-)".
Whoops, I could have sworn I fixed that...
hbac_free_error(): I wonder if it makes sense to add an explicit "if (!error) return". I think it is expected that most free functions accept NULL as a noop and it is quite useful for goto-cleanup cases.
Yeah, that was an oversight. I converted it from talloc_free() (while eliminating the talloc requirement) and didn't think about the NULL case.
0004: ACK
0005: ACK
0006: ACK
0007: ACK, altough the description of ipa_hbac_refresh was added to config/SSSDConfig.py in patch 0008. Feel free to fix or not, I don't think this should be a nack as the patchset will be applied as a whole anyway.
Oops. Looks like I squashed that into the wrong patch. Fixed.
For the rest of the patches - the code seems to be OK sans the one little issue below. I will defer the "architectural" review to Simo as the original comments were based on his PAC development experience.
0008: NACK Please free tmp_ctx in the "immediate:" label in ipa_hbac_update_groups_send()
Thanks, I missed that.
The "/* Not found as a user; Check Groups*/" seems to be misleading, the code never checks users.
That was a relic from before I decided to live with IPA-isms. Originally I was doing a sysdb_search_user() to eliminate users from consideration. I've removed this comment.
0009: ACK
0010: ACK
0011: ACK
0012: ACK
Updated patches attached.
On 06/28/2011 08:14 AM, Stephen Gallagher wrote:
On Mon, 2011-06-27 at 18:31 -0400, Jakub Hrozek wrote:
0002: ACK
0003: NACK The defattr in "%files -n libipa_hbac-devel" is missing a comma before the last parameter, it says "%defattr(-,root,root-)".
Whoops, I could have sworn I fixed that...
hbac_free_error(): I wonder if it makes sense to add an explicit "if (!error) return". I think it is expected that most free functions accept NULL as a noop and it is quite useful for goto-cleanup cases.
Yeah, that was an oversight. I converted it from talloc_free() (while eliminating the talloc requirement) and didn't think about the NULL case.
ACK
0004: ACK
0005: ACK
0006: ACK
0007: ACK, altough the description of ipa_hbac_refresh was added to config/SSSDConfig.py in patch 0008. Feel free to fix or not, I don't think this should be a nack as the patchset will be applied as a whole anyway.
Oops. Looks like I squashed that into the wrong patch. Fixed.
ACK
For the rest of the patches - the code seems to be OK sans the one little issue below. I will defer the "architectural" review to Simo as the original comments were based on his PAC development experience.
0008: NACK Please free tmp_ctx in the "immediate:" label in ipa_hbac_update_groups_send()
Thanks, I missed that.
The "/* Not found as a user; Check Groups*/" seems to be misleading, the code never checks users.
That was a relic from before I decided to live with IPA-isms. Originally I was doing a sysdb_search_user() to eliminate users from consideration. I've removed this comment.
ACK
0009: ACK
0010: ACK
0011: ACK
0012: ACK
Updated patches attached.
I also tried applying the python bindings on top of the latest patches and running the testsuite - it passed. So unless there are any specific requirements on changing the bindings, I think they are ready for review as well.
On Tue, Jun 28, 2011 at 08:14:57AM -0400, Stephen Gallagher wrote:
On Mon, 2011-06-27 at 18:31 -0400, Jakub Hrozek wrote:
0002: ACK
0003: NACK The defattr in "%files -n libipa_hbac-devel" is missing a comma before the last parameter, it says "%defattr(-,root,root-)".
Whoops, I could have sworn I fixed that...
hbac_free_error(): I wonder if it makes sense to add an explicit "if (!error) return". I think it is expected that most free functions accept NULL as a noop and it is quite useful for goto-cleanup cases.
Yeah, that was an oversight. I converted it from talloc_free() (while eliminating the talloc requirement) and didn't think about the NULL case.
0004: ACK
NACK
+ /* Get the source host */ + if (pd->rhost == NULL || pd->rhost[0] == '\0') { + /* If we haven't been passed an rhost, we + * have to assume it's coming from the + * target host + */
this assumption is wrong, if rhost is missing we cannot assume anything about the remote host. See http://www.kernel.org/pub/linux/libs/pam/Linux-PAM-html/adg-security-user-id... for details.
bye, Sumit
0005: ACK
0006: ACK
0007: ACK, altough the description of ipa_hbac_refresh was added to config/SSSDConfig.py in patch 0008. Feel free to fix or not, I don't think this should be a nack as the patchset will be applied as a whole anyway.
Oops. Looks like I squashed that into the wrong patch. Fixed.
For the rest of the patches - the code seems to be OK sans the one little issue below. I will defer the "architectural" review to Simo as the original comments were based on his PAC development experience.
0008: NACK Please free tmp_ctx in the "immediate:" label in ipa_hbac_update_groups_send()
Thanks, I missed that.
The "/* Not found as a user; Check Groups*/" seems to be misleading, the code never checks users.
That was a relic from before I decided to live with IPA-isms. Originally I was doing a sysdb_search_user() to eliminate users from consideration. I've removed this comment.
0009: ACK
0010: ACK
0011: ACK
0012: ACK
Updated patches attached.
So many changes have been made since the original pass that I have revisited the layout of my patches and have squashed many of them together. With these patches, we are fully compatible with the change to disable DENY rules in FreeIPA.
Patch 0001: Add helper function msgs2attrs_array Unchanged from previous versions
Patch 0002: Add HBAC evaluator and tests The evaluator library has changed its interface slightly. It now always returns an hbac_info object that contains the name of the rule that matched (if one did).
Patch 0003: Add helper functions for looking up HBAC rule components I squashed about six revisions of this work into this patch. This patch does all the work of retrieving the rule components from LDAP and converting them into the evaluator format.
Patch 0004: Remove old HBAC implementation Unchanged from previous
Patch 0005: Add new HBAC lookup and evaluation routines Like patch 0003, this squashes in all of the assorted fixes that have gone in throughout this process. These are the changes necessary to use patches 0002 and 0003 in the access provider.
Patch 0006: Add ipa_hbac_refresh option Mostly unchanged, except I fixed a typo in the manpage
Patch 0007: Add ipa_hbac_treat_deny_as option By default, we will treat the presence of any DENY rule as denying all users. This option will allow the admin to explicitly ignore DENY rules during a transitional period.
Patch 0008: Treat NULL or empty rhost as unknown Previously, we were assuming this meant it was coming from the localhost, but this is not a safe assumption. We will now treat it as unknown and it will fail to match any rule that requires a specified srchost or group of srchosts.
I left patch 0008 as a separate patch just for clarity, since it was reasonably self-contained. And also so it was clear that I'd addressed Sumit's concern.
On Wed, 2011-06-29 at 14:34 +0200, Sumit Bose wrote:
0004: ACK
NACK
- /* Get the source host */
- if (pd->rhost == NULL || pd->rhost[0] == '\0') {
/* If we haven't been passed an rhost, we* have to assume it's coming from the* target host*/this assumption is wrong, if rhost is missing we cannot assume anything about the remote host. See http://www.kernel.org/pub/linux/libs/pam/Linux-PAM-html/adg-security-user-id... for details.
And now with patches attached...
On Fri, 2011-07-01 at 16:42 -0400, Stephen Gallagher wrote:
So many changes have been made since the original pass that I have revisited the layout of my patches and have squashed many of them together. With these patches, we are fully compatible with the change to disable DENY rules in FreeIPA.
Patch 0001: Add helper function msgs2attrs_array Unchanged from previous versions
Patch 0002: Add HBAC evaluator and tests The evaluator library has changed its interface slightly. It now always returns an hbac_info object that contains the name of the rule that matched (if one did).
Patch 0003: Add helper functions for looking up HBAC rule components I squashed about six revisions of this work into this patch. This patch does all the work of retrieving the rule components from LDAP and converting them into the evaluator format.
Patch 0004: Remove old HBAC implementation Unchanged from previous
Patch 0005: Add new HBAC lookup and evaluation routines Like patch 0003, this squashes in all of the assorted fixes that have gone in throughout this process. These are the changes necessary to use patches 0002 and 0003 in the access provider.
Patch 0006: Add ipa_hbac_refresh option Mostly unchanged, except I fixed a typo in the manpage
Patch 0007: Add ipa_hbac_treat_deny_as option By default, we will treat the presence of any DENY rule as denying all users. This option will allow the admin to explicitly ignore DENY rules during a transitional period.
Patch 0008: Treat NULL or empty rhost as unknown Previously, we were assuming this meant it was coming from the localhost, but this is not a safe assumption. We will now treat it as unknown and it will fail to match any rule that requires a specified srchost or group of srchosts.
I left patch 0008 as a separate patch just for clarity, since it was reasonably self-contained. And also so it was clear that I'd addressed Sumit's concern.
On Wed, 2011-06-29 at 14:34 +0200, Sumit Bose wrote:
0004: ACK
NACK
- /* Get the source host */
- if (pd->rhost == NULL || pd->rhost[0] == '\0') {
/* If we haven't been passed an rhost, we* have to assume it's coming from the* target host*/this assumption is wrong, if rhost is missing we cannot assume anything about the remote host. See http://www.kernel.org/pub/linux/libs/pam/Linux-PAM-html/adg-security-user-id... for details.
sssd-devel mailing list sssd-devel@lists.fedorahosted.org https://fedorahosted.org/mailman/listinfo/sssd-devel
On 07/01/2011 04:44 PM, Stephen Gallagher wrote:
And now with patches attached...
On Fri, 2011-07-01 at 16:42 -0400, Stephen Gallagher wrote:
So many changes have been made since the original pass that I have revisited the layout of my patches and have squashed many of them together. With these patches, we are fully compatible with the change to disable DENY rules in FreeIPA.
Patch 0001: Add helper function msgs2attrs_array Unchanged from previous versions
Patch 0002: Add HBAC evaluator and tests The evaluator library has changed its interface slightly. It now always returns an hbac_info object that contains the name of the rule that matched (if one did).
I only had time to review these two patches as the bindings are built on top of them -- sorry.
0001 - unchanged, ack 0002 - ack, but please also fix the docstring of hbac_evaluate() before pushing (error vs. info)
On Mon, Jul 04, 2011 at 07:27:17PM -0400, Jakub Hrozek wrote:
On 07/01/2011 04:44 PM, Stephen Gallagher wrote:
And now with patches attached...
On Fri, 2011-07-01 at 16:42 -0400, Stephen Gallagher wrote:
So many changes have been made since the original pass that I have revisited the layout of my patches and have squashed many of them together. With these patches, we are fully compatible with the change to disable DENY rules in FreeIPA.
Patch 0001: Add helper function msgs2attrs_array Unchanged from previous versions
Patch 0002: Add HBAC evaluator and tests The evaluator library has changed its interface slightly. It now always returns an hbac_info object that contains the name of the rule that matched (if one did).
I only had time to review these two patches as the bindings are built on top of them -- sorry.
0001 - unchanged, ack 0002 - ack, but please also fix the docstring of hbac_evaluate() before pushing (error vs. info)
Hi,
0002 also adds an empty line at the end of configure.ac.
0005-Add-new-HBAC-lookup-and-evaluation-routines.patch: + base_dn = sysdb_custom_subtree_dn(sysdb, NULL, + domain->name, + HBAC_RULES_SUBDIR); + if (base_dn == NULL) { + ipa_access_reply(hbac_ctx, PAM_SYSTEM_ERR); + return; + }
sysdb_custom_subtree_dn() always fails, because ldb_dn_new_fmt() does not like a NULL memory context, additionally base_dn is never freed.
0006-Add-ipa_hbac_refresh-option.patch: I would prefer to save a reference to the ipa_access_ctx in hbac_ctx in ipa_access_handler() and use this later on. But this is just cosmetics.
0007-Add-ipa_hbac_treat_deny_as-option.patch + <para> + <emphasis>DENY_ALL</emphasis>: If any HBAC rules + are detected, all users will be denied access. + </para>
I think you meant 'If any HBAC deny rules ...'
Although it is not the default I think it would make sense to have a level 5 or above debug message that tells that deny rules are ignored and not loaded from the server when the rules are evaluated.
I've done some testing and all the logic is working as expected.
bye, Sumit
sssd-devel mailing list sssd-devel@lists.fedorahosted.org https://fedorahosted.org/mailman/listinfo/sssd-devel
On Fri, 2011-07-08 at 14:59 +0200, Sumit Bose wrote:
On Mon, Jul 04, 2011 at 07:27:17PM -0400, Jakub Hrozek wrote:
On 07/01/2011 04:44 PM, Stephen Gallagher wrote:
And now with patches attached...
On Fri, 2011-07-01 at 16:42 -0400, Stephen Gallagher wrote:
So many changes have been made since the original pass that I have revisited the layout of my patches and have squashed many of them together. With these patches, we are fully compatible with the change to disable DENY rules in FreeIPA.
Patch 0001: Add helper function msgs2attrs_array Unchanged from previous versions
Patch 0002: Add HBAC evaluator and tests The evaluator library has changed its interface slightly. It now always returns an hbac_info object that contains the name of the rule that matched (if one did).
I only had time to review these two patches as the bindings are built on top of them -- sorry.
0001 - unchanged, ack 0002 - ack, but please also fix the docstring of hbac_evaluate() before pushing (error vs. info)
Fixed
Hi,
0002 also adds an empty line at the end of configure.ac.
I didn't see this in my patch.
0005-Add-new-HBAC-lookup-and-evaluation-routines.patch:
base_dn = sysdb_custom_subtree_dn(sysdb, NULL,domain->name,HBAC_RULES_SUBDIR);if (base_dn == NULL) {ipa_access_reply(hbac_ctx, PAM_SYSTEM_ERR);return;}sysdb_custom_subtree_dn() always fails, because ldb_dn_new_fmt() does not like a NULL memory context, additionally base_dn is never freed.
Thanks, I added a tmp_ctx here and make sure everything gets freed.
0006-Add-ipa_hbac_refresh-option.patch: I would prefer to save a reference to the ipa_access_ctx in hbac_ctx in ipa_access_handler() and use this later on. But this is just cosmetics.
Done.
0007-Add-ipa_hbac_treat_deny_as-option.patch
<para><emphasis>DENY_ALL</emphasis>: If any HBAC rulesare detected, all users will be denied access.</para>I think you meant 'If any HBAC deny rules ...'
Yup, good catch.
Although it is not the default I think it would make sense to have a level 5 or above debug message that tells that deny rules are ignored and not loaded from the server when the rules are evaluated.
I decided not to bother with this. If this has been set, they're aware of it.
I've done some testing and all the logic is working as expected.
New patches attached. Thanks for the review!
On Fri, Jul 08, 2011 at 12:34:44PM -0400, Stephen Gallagher wrote:
On Fri, 2011-07-08 at 14:59 +0200, Sumit Bose wrote:
On Mon, Jul 04, 2011 at 07:27:17PM -0400, Jakub Hrozek wrote:
On 07/01/2011 04:44 PM, Stephen Gallagher wrote:
And now with patches attached...
On Fri, 2011-07-01 at 16:42 -0400, Stephen Gallagher wrote:
So many changes have been made since the original pass that I have revisited the layout of my patches and have squashed many of them together. With these patches, we are fully compatible with the change to disable DENY rules in FreeIPA.
Patch 0001: Add helper function msgs2attrs_array Unchanged from previous versions
Patch 0002: Add HBAC evaluator and tests The evaluator library has changed its interface slightly. It now always returns an hbac_info object that contains the name of the rule that matched (if one did).
I only had time to review these two patches as the bindings are built on top of them -- sorry.
0001 - unchanged, ack 0002 - ack, but please also fix the docstring of hbac_evaluate() before pushing (error vs. info)
Fixed
Hi,
0002 also adds an empty line at the end of configure.ac.
I didn't see this in my patch.
ah, sorry it is not configure.ac, but ipa_hbac.pc.in, line 558 in your patch. If you find it, you can remove it before pushing, otherwise it is fine as well.
0005-Add-new-HBAC-lookup-and-evaluation-routines.patch:
base_dn = sysdb_custom_subtree_dn(sysdb, NULL,domain->name,HBAC_RULES_SUBDIR);if (base_dn == NULL) {ipa_access_reply(hbac_ctx, PAM_SYSTEM_ERR);return;}sysdb_custom_subtree_dn() always fails, because ldb_dn_new_fmt() does not like a NULL memory context, additionally base_dn is never freed.
Thanks, I added a tmp_ctx here and make sure everything gets freed.
0006-Add-ipa_hbac_refresh-option.patch: I would prefer to save a reference to the ipa_access_ctx in hbac_ctx in ipa_access_handler() and use this later on. But this is just cosmetics.
Done.
0007-Add-ipa_hbac_treat_deny_as-option.patch
<para><emphasis>DENY_ALL</emphasis>: If any HBAC rulesare detected, all users will be denied access.</para>I think you meant 'If any HBAC deny rules ...'
Yup, good catch.
Although it is not the default I think it would make sense to have a level 5 or above debug message that tells that deny rules are ignored and not loaded from the server when the rules are evaluated.
I decided not to bother with this. If this has been set, they're aware of it.
I've done some testing and all the logic is working as expected.
New patches attached. Thanks for the review!
ACK to all, just push it :-)
bye, Sumit
On Fri, 2011-07-08 at 20:31 +0200, Sumit Bose wrote:
On Fri, Jul 08, 2011 at 12:34:44PM -0400, Stephen Gallagher wrote:
On Fri, 2011-07-08 at 14:59 +0200, Sumit Bose wrote:
On Mon, Jul 04, 2011 at 07:27:17PM -0400, Jakub Hrozek wrote:
On 07/01/2011 04:44 PM, Stephen Gallagher wrote:
And now with patches attached...
On Fri, 2011-07-01 at 16:42 -0400, Stephen Gallagher wrote:
So many changes have been made since the original pass that I have revisited the layout of my patches and have squashed many of them together. With these patches, we are fully compatible with the change to disable DENY rules in FreeIPA.
Patch 0001: Add helper function msgs2attrs_array Unchanged from previous versions
Patch 0002: Add HBAC evaluator and tests The evaluator library has changed its interface slightly. It now always returns an hbac_info object that contains the name of the rule that matched (if one did).
I only had time to review these two patches as the bindings are built on top of them -- sorry.
0001 - unchanged, ack 0002 - ack, but please also fix the docstring of hbac_evaluate() before pushing (error vs. info)
Fixed
Hi,
0002 also adds an empty line at the end of configure.ac.
I didn't see this in my patch.
ah, sorry it is not configure.ac, but ipa_hbac.pc.in, line 558 in your patch. If you find it, you can remove it before pushing, otherwise it is fine as well.
0005-Add-new-HBAC-lookup-and-evaluation-routines.patch:
base_dn = sysdb_custom_subtree_dn(sysdb, NULL,domain->name,HBAC_RULES_SUBDIR);if (base_dn == NULL) {ipa_access_reply(hbac_ctx, PAM_SYSTEM_ERR);return;}sysdb_custom_subtree_dn() always fails, because ldb_dn_new_fmt() does not like a NULL memory context, additionally base_dn is never freed.
Thanks, I added a tmp_ctx here and make sure everything gets freed.
0006-Add-ipa_hbac_refresh-option.patch: I would prefer to save a reference to the ipa_access_ctx in hbac_ctx in ipa_access_handler() and use this later on. But this is just cosmetics.
Done.
0007-Add-ipa_hbac_treat_deny_as-option.patch
<para><emphasis>DENY_ALL</emphasis>: If any HBAC rulesare detected, all users will be denied access.</para>I think you meant 'If any HBAC deny rules ...'
Yup, good catch.
Although it is not the default I think it would make sense to have a level 5 or above debug message that tells that deny rules are ignored and not loaded from the server when the rules are evaluated.
I decided not to bother with this. If this has been set, they're aware of it.
I've done some testing and all the logic is working as expected.
New patches attached. Thanks for the review!
ACK to all, just push it :-)
Fixed the extra line and pushed to master.
sssd-devel@lists.fedorahosted.org