[SSSD] Design Discussion: ID Mapping calls for the NSS responder

Hi, I tried to extract some common requirements of #1534 'Integrate SSSD with CIFS client' and #1557 'Use the Global Catalog in SSSD for the AD provider' in the https://fedorahosted.org/sssd/wiki/DesignDocs/NSSResponderIDMappingCalls design page. Currently there is no specific ticket for this, I think it can be opened if the design is accepted. For your convenience the content can be found below as well. bye, Sumit == ID Mapping calls for the NSS responder == Related tickets: * [https://fedorahosted.org/sssd/ticket/1534 RFE Integrate SSSD with CIFS client] * [https://fedorahosted.org/sssd/ticket/1557 RFE Use the Global Catalog in SSSD for the AD provider] * [https://fedorahosted.org/sssd/ticket/1559 RFE Use the getpwnam()/getgrnam() interface as a gateway to resolve SID to Names] Related design documents: * [https://fedorahosted.org/sssd/wiki/DesignDocs/IntegrateSSSDWithCIFSClient Integrate SSSD with CIFS Client] * [https://fedorahosted.org/sssd/wiki/DesignDocs/GlobalCatalogLookups Global Catalog Lookups in SSSD] === Problem Statement === When SSSD is used in environments with AD, either as a member of the AD domain or as a member of a trusted IPA domain, it has to map users and groups managed by AD to a POSIX ID. The AD user and groups are identified by * a name which may change * a unique SID which never changes, i.e. new SID == new object Applications interacting tightly with users and groups from AD domains e.g. * samba * cifs-client * FreeIPA need to know which SID relates to which POSIX ID or name. Mapping a SID to a user or group would be possible with the current interfaces as described in ticket #1559. But getting a SID for a user or a group would be at least hard and ugly if not impossible. I think the solution proposed in ticket #1559 is a good and useful shortcut. But by making the *bySID lookups also available via a new calls applications will have a more reliable interface, e.g. by allowing more specific error codes (see details below). === Overview of the solution === === Implementation details === The NSS responder will be extended with four new calls: {{{ SSS_NSS_GETSIDBYNAME = 0x0111, /**< Takes a zero terminated fully qualified name and returns the zero terminated string representation of the SID of the object with the given name. */ SSS_NSS_GETSIDBYID = 0x0112, /**< Takes an unsigned 32bit integer (POSIX ID) and returns the zero terminated string representation of the SID of the object with the given ID. */ SSS_NSS_GETNAMEBYSID = 0x0113, /**< Takes the zero terminated string representation of a SID and returns the zero terminated fully qualified name of the related object. */ SSS_NSS_GETIDBYSID = 0x0114, /**< Takes the zero terminated string representation of a SID and returns and returns the POSIX ID of the related object as unsigned 32bit integer value and another unsigned 32bit integer value indicating the type (unknown, user, group, both) of the object. */ }}} Alternatively SSS_NSS_GETSIDBYID and SSS_NSS_GETIDBYSID could be implemented to take an return arrays SIDs and POSIX IDs respectively as the related cifs_idmap API calls (see [https://fedorahosted.org/sssd/wiki/DesignDocs/IntegrateSSSDWithCIFSClient Integrate SSSD with CIFS Client]). I took the approach mentioned above because it better matches the other NSS responder calls and additionally I do not like the implicit required ordering in the input and output array. After receiving the request the NSS responder will first check if it can create the response from cached data. If not the request is forwarded to the providers. For the *byName and *byID calls the corresponding existing interface can be used. For the *bySID call a new call must be added to the providers. Currently only the IPA and AD provider will support the new calls. If the provider cannot handle the requests it will return an appropriate error code which it return to the client via the NSS responder. Additionally on the provider side it must be ensured that the string representation of the SID is saved in the cache. It looks that this is already the case for the AD provider. But I think this is currently not the case for the IPA provider for both IPA and trusted users. Also the PAC responder should add the SID to the cache object. The sid2name extended operation on the FreeIPA server should get a new request type and corresponding response types so that the SID is returned with the user and group data. (A ticket for this should be opened for FreeIPA if this design is approved.) ==== C-API ==== The C-API for those calls will be made available in a new library libsss_nss_idmap (other names are welcome). In contrast to libnss_sss loaded via glibc into client programs the new library can be linked with other programs. The new calls will be declared in a header sss_nss_idmap.h and can have reasonable return values to make error detection an reporting easier for the clients using the new API. {{{ /** * @brief Find SID by fully qualified name * * @param[in] fq_name Fully qualified name of a user or a group * @param[out] sid String representation of the SID of the requested user or group, must be freed by the caller * * @return * - 0 (EOK): success, sid contains the requested SID * - ENOENT: requested object was not found in the domain extracted from the given name * - ENETUNREACH: SSSD does not know how to handle the domain extracted from the given name * - ENOSYS: this call is not supported by the configured provider * - EINVAL: input cannot be parsed * - EIO: remote servers cannot be reached * - EFAULT: any other error */ int sss_nss_getsidbyname(const char *fq_name, char **sid); /** * @brief Find SID by a POSIX UID or GID * * @param[in] id POSIX UID or GID * @param[out] sid String representation of the SID of the requested user or group, must be freed by the caller * * @return * - see #sss_nss_getsidbyname */ int sss_nss_getsidbyid(uint32_t id, char **sid); /** * @brief Return the fully qualified name for the given SID * * @param[in] sid String representation of the SID * @param[out] fq_name Fully qualified name of a user or a group, must be freed by the caller * * @return * - see #sss_nss_getsidbyname */ int sss_nss_getnamebysid(const char *sid, char **fq_name); /** * @brief Return the POSIX ID for the given SID * * @param[in] sid String representation of the SID * @param[out] id POSIX ID related to the SID * @param[out] id_type Type of the object related to the ID * * @return * - see #sss_nss_getsidbyname */ int sss_nss_getidbysid(const char *sid, uint32_t *id, enum id_type id_type); }}} Currently I do not see a strong requirement to allow different kind of memory allocators (e.g. talloc). I think it is not necessary to add special set of return values/error code but the standard ones are sufficient. Maybe ENETUNREACH it indicate that SSSD could not find the right domain for the request could be replaced by a better one. (EDOM would be a candidate, but imo it should be reserved for mathematical operations.) ==== Python bindings ==== To allow easy usage of the new calls by the FreeIPA python framework, python binding would be useful. ==== Lookup utility ==== A small utility sss_idmap (other names are welcome) will be added which offers access to the new calls via libsss_nss_idmap. This utility will make testing easier and might help user and administrators as well. {{{ # sss_idmap --help Usage: sss_idmap [OPTION...] -n, --name-to-sid=NAME Converts name to sid -s, --sid-to-name=SID Converts sid to name -S, --sid-to-id=SID Converts sid to POSIX ID -i, --id-to-sid=ID Converts POSIX ID to sid Help options: -?, --help Show this help message --usage Display brief usage message }}} === How to test === The sss_idmap utility or the python bindings can be used to create tests, e.g. {{{ # sss_idmap --sid-to-name=S-1-5-21-111111-222222-333333-1234 DOM\user # sss_idmap --name-to-sid=DOM\user S-1-5-21-111111-222222-333333-1234 # sss_idmap --sid-to-name=abcdefg Usage: sss_idmap ....... Invalid SID }}} === Author(s) === Sumit Bose <sbose(a)redhat.com&gt;