IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
This replaces ctdb_killtcp(), which did the initialisation inside a
loop. The new logic is inverted, making it more natural.
The variable containing all the state is called "state" in
anticipation of the next commit that will convert this to a tevent_req
computation. This will mean less churn.
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
One less use of trbt_tree_t. The code is easier to read and is
significantly smaller.
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
This and the supporting functions duplicate functionality (parse_ip()
and parse_ip_port()) from common/system_util.c. The old functions
will be removed at a later time.
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
The current code is ambiguous in its use of src and dst. This allows
new code to use server and client for clarity.
Signed-off-by: Amitay Isaacs <amitay@gmail.com>
Reviewed-by: Martin Schwenke <martin@meltin.net>
Guenther
Signed-off-by: Guenther Deschner <gd@samba.org>
Pair-Programmed-With: Jose A. Rivera <jarrpa@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Ralph Böhme <slow@samba.org>
Autobuild-Date(master): Tue Sep 19 09:36:40 CEST 2017 on sn-devel-144
It looks like commit 070f24bc9d was incomplete because it doesn't
match on the trailing fullstop and newline.
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Autobuild-User(master): Martin Schwenke <martins@samba.org>
Autobuild-Date(master): Tue Sep 19 05:32:22 CEST 2017 on sn-devel-144
This reverts commit 19318d2835.
With this commit, a shutdown that occurs while the startup event is
running can cause an abort because the startup callback will try to
decrease the run state from SHUTDOWN to RUNNING.
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
Currently we have tests that check we can resolve object conflicts, but
these don't test anything related to conflicting linked attributes.
This patch adds some basic tests that checks that Samba can resolve
conflicting linked attributes.
This highlights some problems with Samba, as the following tests
currently fail:
- test_conflict_single_valued_link: Samba currently can't resolve a
conflicting targets for a single-valued linked attribute - the
replication exits with an error.
- test_link_deletion_conflict: If 2 DCs add the same linked attribute,
currently when they resolve this conflict the RMD_VERSION for the
linked attribute incorrectly gets incremented. This means the version
numbers get out of step and subsequent changes to the linked attribute
can be dropped/ignored.
- test_full_sync_link_conflict: fails for the same reason as above.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
Autobuild-User(master): Garming Sam <garming@samba.org>
Autobuild-Date(master): Mon Sep 18 09:56:41 CEST 2017 on sn-devel-144
The max_links calculation didn't work particularly well if max_links was
set to a value lower than max_objects.
As soon as repl_chunk->object_count exceeded repl_chunk->max_links, the
chunk would be deemed full, even if there was only one link to send (or
even worse, no links to send). For example, if max_objects=100 and
max_links=10, then it would send back chunks of 10 objects (or 9 objects
and 1 link).
I believe the historic reason this logic exists is to avoid overfilling
the response message. It's hard to tell what the appropriate limit would
be because the total message size would depend on how many attributes
each object has.
I couldn't think of logic that would be suitable for all cases. I toyed
with the idea of working out a percentage of how full the message is.
However, adjusting the max_links doesn't really make sense when the
settings are small enough, e.g. max_objects=100 and max_links=100 is
never going to overfill the message, so there's no reason to alter the
values.
In the end I went with:
- If the user is using non-default values, just use those.
- In the default value case, just use the historic calculation
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12972
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
We display warnings if a target object is missing but it's still OK to
continue the replication. Currently we need to check the target twice -
once to verify it when we first receive it, and once when we actually
commit it (we can't skip the 2nd check altogether because in the join
case, they could occur quite far apart).
One annoying side-effect is we get the same warning message coming out
twice in these special cases.
In the cases where we're checking the dsdb_repl_flags, we can actually
just bypass the verification checks for the target object (if it doesn't
exist we still continue anyway). This may save us a tiny bit of
unnecessary work.
For cross-partition links, we can limit logging these warnings to when
the objects are actually being committed. This avoids spurious warnings
in the join case (i.e. we receive the link before we receive the target
object's partition, but we have received all partitions by the time we
actually commit the objects).
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12972
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
While running the selftests, I noticed a case where DC replication
unexpectedly sends a linked attribute for a deleted object (created in
the drs.ridalloc_exop tests). The problem is due to the
msDS-NC-Replica-Locations attribute, which is a (known) one-way link.
Because it is a one-way link, when the test demotes the DC and deletes
the link target, there is no backlink to delete the link from the source
object.
After much debate and head-scratching, we decided that there wasn't an
ideal way to resolve this problem. Any automated intervention could
potentially do the wrong thing, especially if the link spans partitions.
Running dbcheck will find this problem and is able to fix it (providing
the deleted object is still a tombstone). So the recommendation is to
run dbcheck on your DCs every 6 months (or more frequently if using a
lower tombstone lifetime setting).
However, it does highlight a problem with the current GET_TGT
implementation. If the tombstone object had been expunged and you
upgraded to 4.8, then you would be stuck - replication would fail
because the target object can't be resolved, even with GET_TGT, and
dbcheck would not be able to fix the hanging link. The solution is to
not fail the replication for an unknown target if GET_TGT has already
been set (i.e. the dsdb_repl_flags contains
DSDB_REPL_FLAG_TARGETS_UPTODATE).
It's debatable whether we should add a hanging link in this case or
ignore/drop the link. Some cases to consider:
- If you're talking to a DC that still sends all the links last, you
could still get object deletion between processing the source object's
links and sending the target (GET_TGT just restarts the replication
cycle from scratch). Adding a hanging link in this case would be
incorrect and would add spurious information to the DB.
- Suppose there's a bug in Samba that incorrectly results in an object
disappearing. If other DCs then remove any links that pointed to that
object, it makes recovering from the problem harder. However, simply
ignoring the link shouldn't result in data loss, i.e. replication won't
remove the existing link information from other DCs. Data loss in this
case would only occur if a new DC were brought online, or if it were a
new link that was affected.
Based on this, I think ignoring the link does the least harm.
This problem also highlights that we should really be using the same
logic in both the unknown target and the deleted target cases.
Combining the logic and moving it into a common
replmd_allow_missing_target() function fixes the problem. (This also has
the side-effect of fixing another logic flaw - in the deleted object
case we would unnecessarily retry with GET_TGT if the target object was
in another partition. This is pointless work, because GET_TGT won't
resolve the target).
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12972
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
A source object can potentially link to thousands of target objects.
We have to be careful not to overfill the GetNCChanges response message
with more data than it's possible to send. We also don't want the client
to timeout while we're busy checking the linked attributes. The GET_TGT
support added so far is fairly dumb - this patch extends it to better
handle larger numbers of links.
To do so, this extends the repl_chunk usage so that it also works out if
the current chunk is full of links. Now as soon as the chunk is full of
either links or objects, we stop and send it back.
These changes now mean that we need to also check:
- that all the links for the last source object in the previous chunk
have been sent, before we move on and send the next object. This only
takes effect when immediate_link_sync is configured. It also means
that a chunk in the middle of the replication cycle can now contain
only links, and no objects.
- when GET_TGT is used, we only send back the links that we've verified
the target object for. i.e. if we stop checking links because we timed
out, we only send back the links whose targets were checked.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
To prepare GET_TGT to deal with a large number of links better, there
is now a 'repl_chunk' struct to help keep track of all the factors
relating to the current chunk of replication data (i.e. how many
objects/links we can send and how many we've already processed). This
means we can have a consistent way of working out whether the current
chunk is full (whether that be due to objects, links, or just too much
time taken).
This patch should not alter functionality. This is just a refactor to
add the basic framework, which will be used in the next patch.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Signed-off-by: Garming Sam <garming@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Add a test where a source object links to multiple different targets.
First we do the replication without GET_TGT and check that the server
can handle sending a chunk containing only links (in the middle of the
replication). Then we repeat the replication forcing GET_TGT to be used.
To avoid having to create 1500 objects/links, I've lowered the 'max
link sync' setting on the vampire_dc testenv to 250.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
Samba would drop linked attributes that span partitions if it didn't
know about the target object. This patch adds a test that exposes the
problem.
I've re-used the code from the previous re-animation test to do this.
I've also added a very basic DcConnection helper class that basically
stores the connection state information the drs_base.py uses for
replication. This allows us to switch the DC we want to replicate from
easily. This approach could potentially be retro-fitted to some of the
existing test cases, as it allows us to test both the DRS client code
and server code at the same time.
Note this test case relates to the code change for commit
fae5df891c.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
BUG: https://bugzilla.samba.org/show_bug.cgi?id=12972
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
Reading between the lines, this scenario seems to be the main reason
that Microsoft added the GET_TGT flag. MS AD can handle getting links
for unknown targets OK, but if it receives links for a deleted/recycled
target then it would tend to drop the received links. Samba client also
used to drop the links if talking to a Microsoft DC (or a Samba server
with GET_TGT support).
The specific scenario is the client side already knows about a deleted
object. That object is then re-animated and used as the target for a
linked attribute. *Then* the target object gets updated again so it gets
sent in a later replication chunk to the linked attribute, i.e. the
client receives the link before it learns that the target object has
been re-animated.
In this test we're interested in particular at how the client behaves
when it receives a linked attribute for a deleted object. (It *should*
retry with GET_TGT to make sure the target is up-to-date. However, it
was just dropping the linked attribute).
To exercise the client-side, we disable replication, setup the
links/objects on one DC the way we want them, then force a replication
to the second DC. We then check that when we query each DC, they both
tell us about the links/objects we're expecting (i.e. no links got
lost).
Note that this wasn't a problem with older versions of Samba-to-Samba
because sending the links last guaranteed that the target objects were
always up-to-date.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
This adds basic DRS_GET_TGT support. If the GET_TGT flag is specified
then the server will use the object cache to store the objects it sends
back. If the target object for a linked attribute is not in the cache
(i.e. it has not been sent already), then it is added to the response
message.
Note that large numbers of linked attributes will not be handled well
yet - the server could potentially try to send more than will fit in a
single repsonse message.
Also note that the client can sometimes set the GET_TGT flag even if the
server is still sending the links last. In this case, we know the client
supports GET_TGT so it's safe to send the links interleaved with the
source objects (the alternative of fetching the target objects but not
sending the links until last doesn't really make any sense).
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
Add tests that delete the source and target objects for linked
attributes in the middle of a replication cycle.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
We already check that when we use GET_ANC that we still only receive a
single object when EXOP_REPL_OBJ is used. This extends the test to also
check that only a single object is returned when GET_TGT is used.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
The code has to handle needing GET_ANC and GET_TGT in combination, i.e.
where we fetch the target object for the linked attribute and the target
object's parent is unknown as well. This patch adds a test case to
exercise this code path.
The second part of this test exercises GET_ANC/GET_TGT for an
incremental replication, where the objects are getting filtered by an
uptodateness-vector/HWM.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
We have identified a case where the Samba server can send linked
attributes but not the target object. In this case, the Samba DRS client
would hit the "Failed to re-resolve GUID" case in replmd and silently
discard the linked attribute.
However, Samba will resend the linked attribute in the next cycle
(because its USN is still higher than the committed HWM), so it should
recover OK. On older releases, this may have caused problems if the
first error resulting in a hanging link (which might mean the second
time it's processed it still fails to be added).
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
test_repl_get_tgt:
- Adds 2 sets of objects
- Links one set to the other
- Changes the order so the target object comes last in the
replication (which means the client has to use GET_TGT)
- Checks that when GET_TGT is used that we have received all target
objects we need to resolve the linked attibutes
- Checks that we expect to receive the linked attributes *before*
the last chunk is sent (by default, Samba sends all the links at
the end, so this fails)
- Checks that we eventually receive all expected objects, and all
links we receive match what is expected
test_repl_get_tgt_chain:
This adds the linked attributes in a more complicated chain. We add
300 objects, but the links for 100 objects will point to a linked
chain of 200 objects.
This was mainly to determine whether or not Windows follows the
target object (i.e. whether it sends all the links for the target
object as well). It turns out Windows maintains its own linked
attribute DB, so it sends the links based on USN.
Note that the 2 testenvs fail for different reasons. promoted_dc fails
because it is sending all the linked attributes last. vampire_dc fails
because it doesn't support GET_TGT yet, so it sends the link before the
peer knows about the target object.
Note that to test against vampire_dc (rather than the ad_dc_ntvfs DC),
we need to send the GetNCChanges requests to DC2 instead of DC1.
I've left the DC numbering scheme as is, but I've addeed a test_ldb_dc
handle to drs_base.py - it defaults to DC1, but tests can override it
easily and still have everything work.
While running the new tests through autobuild, I noticed an intermittent
LDAP_ENTRY_ALREADY_EXISTS failure in the test setup(). This appears to
be due to a timing issue in the background replication between the
multiple testenvs. Adding some randomness so that the test base OU is
unique seems to avoid the problem.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
The existing code never passed the more_flags parameter into the
actual getNCChanges request, i.e. _getnc_req10(). This meant the
existing GET_TGT tests effectively did nothing.
Passing the flag through properly means we have to now change the tests
as the DNs returned by Windows now include any target objects in the
linked attributes. These tests now fail against Samba (because it
doesn't support GET_TGT yet).
Also added comments to the tests to help explain what they are actually
doing.
Note that Samba and Windows can return the objects in different orders,
due to significant differences in their underlying DB implementations
(Windows stores links in a separate DB, so sends links ordered strictly
by USN, whereas Samba sends links based on the USN of the source
object). To make the test a fair comparison between Windows and Samba,
we need to use dn_ordered=False.
Signed-off-by: Tim Beale <timbeale@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Garming Sam <garming@catalyst.net.nz>
Set the process group in the samba daemon, the --no-process-group option
allows this to be disabled. The no-process-group option needs to be
disabled in self test.
Signed-off-by: Gary Lockyer <gary@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Ralph Böhme <slow@samba.org>
Autobuild-Date(master): Mon Sep 18 04:39:50 CEST 2017 on sn-devel-144
Free the fde in the event handler to prevent the event triggering again
While not strictly necessary in this case, this code serves as an
example of the usage of tfork.
Bug: https://bugzilla.samba.org/show_bug.cgi?id=13037
Signed-off-by: Gary Lockyer <gary@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Ralph Böhme <slow@samba.org>
Autobuild-Date(master): Sat Sep 16 23:50:27 CEST 2017 on sn-devel-144
Make closing of the event_fd the global responsibility of the
parent process if it called tfork_event_fd().
Bug: https://bugzilla.samba.org/show_bug.cgi?id=13037
Signed-off-by: Ralph Boehme <slow@samba.org>
Reviewed-by: Gary Lockyer <gary@catalyst.net.nz>
Reviewed-by: Stefan Metzmacher <metze@samba.org>
The previous design relied on only calling close() of the status pipe.
We now write a single 0 byte to the status FD as well as closing it in the
parent process. Both of these operations typically trigger a read
event on the other end of the FD, held in the waiter process (the child).
The child process blocks on the status FD, until it becomes readable.
However if there is a sibling process that was launched after the waiter
process they also will hold the status FD open and the status FD would,
until this change, never become readable to the waiter process (the child).
This caused the waiter process (child) not to exit and the parent process
to hang in tfork_status() while expecting the waitpid() to return.
That is, file descriptors are essentially global variables copied
to children in the process tree. The last child that (unwittingly) holds
the file descriptor open is the one that needs to trigger the close() this
code previously depended on.
Without this change, there is no notification of process death until
all these unrelated children exit for their own reasons.
We can write up to 4K (PIPE_BUF) into this pipe before blocking,
but we only write one byte. Additionally sys_write() refuses to block.
Bug: https://bugzilla.samba.org/show_bug.cgi?id=13037
Signed-off-by: Gary Lockyer <gary@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
Add tests to ensure that:
- The event_fd becomes readable once the worker process has terminated
- That the event_fd is not closed by the tfork code.
- If this is done in tevent code and the event fde has not been
freed, "Bad talloc magic value - " errors can result.
- That the status call does not block if the parent process launches
more than one child process.
- The status file descriptor for a child is passed to the
subsequent children. These processes hold the FD open, so that
closing the fd does not make the read end go readable, and the
process calling status blocks.
Bug: https://bugzilla.samba.org/show_bug.cgi?id=13037
Signed-off-by: Gary Lockyer <gary@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Reviewed-by: Ralph Boehme <slow@samba.org>
Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
Autobuild-User(master): Andrew Bartlett <abartlet@samba.org>
Autobuild-Date(master): Sat Sep 16 12:28:17 CEST 2017 on sn-devel-144
IMHO a full talloc_stackframe is overkill for the one allocation that is left
here.
Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>
