Age | Commit message (Collapse) | Author | Files | Lines |
|
This set of functions had a few important issues:
1. the base_dn was always NULL, as the base array was never actually used
to construct any DN. This means each function searched the whole database
multiple times.
It would try to remove SYSDB_USN from all database entries 3 times.
Then it would try to find non updated entries another 3 times and delete
them, arguably find empty results the last 2 times.
2. Remove use of sysdb_private.h, that header is *PRIVATE* which means it
should not be used anywhere but within sysdb. Do this by using existing
functions instead of using ldb calls directly. This is important to keep
sysdb as conistent and self-contained as possible.
|
|
In the AD case, deployments sometimes add groups as parents of the
primary GID group. These groups are then returned during initgroups
in the tokenGroups attribute and member/memberof links are established
between the user and the group. However, any update of these groups
would remove the links, so a sequence of calls: id -G user; id user; id
-G user would return different group memberships.
The downside of this approach is that the user is returned as a group
member during getgrgid call as well.
|
|
The code would call sysdb_attrs_add_uint32 which added another UID or GID
to the ID=0 we already downloaded from LDAP (0 is the default value) when
ID-mapping an entry. This led to funky behaviour later on when we wanted
to process the ID.
|
|
|
|
https://fedorahosted.org/sssd/ticket/1736
When there are no rules during first refresh, we don't have valid
USN value. We use 0 in this case, but it turned out that OpenLDAP
takes it as invalid time format (if modifyTimestamp is used instead
of USN) and thus returns no records.
Now we don't include USN/modifyTimestamp attribute in the filter
if such situasion occurs.
|
|
Use correct name for _done() function from the caller.
Remove unneded initializzations to NULL for a lot of variables hat are going to
be assigned as the first thing done in the functions.
|
|
|
|
Added new parameter to split_on_separator that allows to skip
empty values.
The whole function was rewritten. Unit test case was added to
check the new implementation.
https://fedorahosted.org/sssd/ticket/1484
|
|
If we are unable to delete some rule from cache we print a debug
message and ignore the error. Thus we should set ret to EOK after
the for loop otherwise we return whether the last rule was deleted
successfully or not.
This also removes compilation warning that ret may be used
uninitialized (when we don't go inside the loop at all).
|
|
switch statements should always have a default section. In this
particular case gcc gave a "'send_fn' may be used uninitialized in this
function" warning.
|
|
If the Data Provider receives a request for the auto.master map, it
passes on a flag to let the actual provider let know he should
invalidate the existing maps
|
|
https://fedorahosted.org/sssd/ticket/1689
Add a online callback if the first full refresh fails due to the
provider beeing offline so we can perform the refresh as soon as
possible.
|
|
https://fedorahosted.org/sssd/ticket/1689
If the first full refresh of sudo rules fails because the data provider
is offline, we will schedule another one in 2, 4, ... minutes.
|
|
https://fedorahosted.org/sssd/ticket/1689
|
|
Reduces amount of code duplication.
|
|
https://fedorahosted.org/sssd/ticket/1699
We want to continue with the next server on all errors, not only
on ETIMEDOUT.
This particullar ticket was dealing with ECONNREFUSED.
|
|
https://fedorahosted.org/sssd/ticket/1682
The problem was in following code:
if (ret != EOK || state->dp_error != DP_ERR_OK || state->error != EOK) {
tevent_req_error(req, ret);
return;
}
In situation when data provider error occurs (e.g. when offline),
ret == EOK but dp_error != DP_ERR_OK and we take the true branch.
This results in calling tevent_req_error(req, EOK).
Unfortunately, with EOK tevent_req_error only returns false, but
does not trigger callback and this tevent request hangs forever,
because no tevent_req_done(req) is called.
|
|
|
|
https://fedorahosted.org/sssd/ticket/1701
|
|
|
|
https://fedorahosted.org/sssd/ticket/1695
|
|
Avoids hardcoding magic numbers everywhere and self documents why a
mask is being applied.
|
|
|
|
https://fedorahosted.org/sssd/ticket/1638
If pwd_exp_warning == 0, expiry warning should be printed if it is
returned by server.
If pwd_exp_warning > 0, expiry warning should be printed only if
the password will expire in time <= pwd_exp_warning.
ppolicy->expiry contains period in seconds after which the password
expires. Not the exact timestamp. Thus we should not add 'now' to
pwd_exp_warning.
|
|
When converting built-in SID to unix GID/UID a confusing debug
message about the failed conversion was printed. This patch special
cases these built-in objects.
https://fedorahosted.org/sssd/ticket/1593
|
|
initialized variable, was causing build warning
|
|
https://fedorahosted.org/sssd/ticket/1612
This patch changes the handling of ghost attributes when saving the
actual user entry. Instead of always linking all groups that contained
the ghost attribute with the new user entry, the original member
attributes are now saved in the group object and the user entry is only
linked with its direct parents.
As the member attribute is compared against the originalDN of the user,
if either the originalDN or the originalMember attributes are missing,
the user object is linked with all the groups as a fallback.
The original member attributes are only saved if the LDAP schema
supports nesting.
|
|
|
|
The check is too restrictive as the select_principal_from_keytab can
return something else than user requested right now.
Consider that user query for host/myserver@EXAMPLE.COM, then the
select_principal_from_keytab function will return "myserver" in primary and
"EXAMPLE.COM" in realm. So the caller needs to add logic to also break
down the principal to get rid of the host/ part. The heuristics would
simply get too complex.
select_principal_from_keytab will error out anyway if there's no
suitable principal at all.
|
|
The AD and IPA initialization functions shared the same code. This patch
moves the code into a common initialization function.
|
|
Currently the sysdb context is pointed to the subdomain subtree
containing user the user to be checked at the beginning of a HBAC
request. As a result all HBAC rules and related data is save in the
subdomain tree as well. But since the HBAC rules of the configured
domain apply to all users it is sufficient to save them once in the
subtree of the configured domain.
Since most of the sysdb operations during a HBAC request are related to
the HBAC rules and related data this patch does not change the default
sysdb context but only create a special context to look up subdomain
users.
|
|
|
|
The element being reallocated is part of the "group_attrs" array, not
attrs.
|
|
https://fedorahosted.org/sssd/ticket/1647
A logic bug in the LDAP provider causes an attempt to allocate a zero-length
array for group members while processing an empty group. The allocation
would return NULL and saving the empty group would fail.
|
|
Allocating temporary context on NULL helps vind memory leaks with
valgrind and avoid growing memory over time by allocating on a
long-lived context.
|
|
|
|
The connections request was terminated before setting the expiry timeout
in case no authentication was set.
https://fedorahosted.org/sssd/ticket/1649
|
|
|
|
https://fedorahosted.org/sssd/ticket/1640
Normal rules requires that sudoHost attribute is present. But this
attribute is not mandatory for a special rule named cn=defaults.
This patch modifies filter so that we store even rules that doesn't
have sudoHost attribute specified. SUDO will then decide whether it
is allowed or not.
|
|
https://fedorahosted.org/sssd/ticket/1376
|
|
Currently it is only checked if an expired group still has members of
the local domain. If not, the group is delete from the cache. With this
patch the whole cache, i.e. including subdomains, is searched for
members.
|
|
https://fedorahosted.org/sssd/ticket/1482
When we add fullname to user_attrs, then sysdb_add_basic_user()
will set fullname to gecos when it initially creates the user
object in the cache, but it will be overwritten in the same
transaction when sysdb_store_user() adds all the user_attrs.
|
|
|
|
The ldb_val's length parameter should not include the terminating NULL.
This was causing funky behaviour as the users were saved as binary
attributes.
https://fedorahosted.org/sssd/ticket/1614
|
|
https://fedorahosted.org/sssd/ticket/1495
|
|
The ldap_child would return a NULL ccache but the error code would still
indicate success.
https://fedorahosted.org/sssd/ticket/1594
|
|
We should test both ret and (dp_error, errno) pair.
|
|
https://fedorahosted.org/sssd/ticket/1596
In case that LDAP server contains zero sudo rules, the full refresh
completes succussfully and stores current USN value (= 0). But then
smart refresh will fail because it takes USN=0 as invalid value.
|
|
https://fedorahosted.org/sssd/ticket/1581
If the namingContext attribute had no values or multiple values, then
our code would dereference a NULL pointer.
|
|
|