[oi-dev] [developer] [Samba] gencache.tdb: device busy

Sun Apr 21 12:28:48 UTC 2019

You could check whether the mutex already looks initialized to us using a
debugger or logging statements added to the code.

If that's the case, we can think again about relaxing the check I added in
9959.

On Sun, Apr 21, 2019, 08:20 Rouven WEILER <Rouven_Weiler at gmx.net> wrote:

> Any chance to get to know what happens?
>
> Am 21.04.2019 um 14:12 schrieb Richard Lowe <richlowe at richlowe.net>:
>
> It's possible there's a bug.  It's also possible you're getting really
> unlucky and the uninitialized mutex looks valid because the magic is valid.
>
> On Sun, Apr 21, 2019, 06:10 Rouven WEILER <Rouven_Weiler at gmx.net> wrote:
>
>> Dear all,
>>
>> trying to make samba work and getting rid of my error message
>>
>> [2019/04/01 09:51:41.112485,  0] ../../lib/tdb_wrap/tdb_wrap.c:64(tdb_wrap_log)
>>  tdb(/var/samba/lock/gencache.tdb): tdb_open_ex: tdb_mutex_init failed for /var/samba/lock/gencache.tdb: Device busy
>>
>>
>> I followed the info of Youzhong. It definitely works. So thankx.
>> BUT: The OS (OpenIndiana; state of April,19th) is patched against the
>> illumos issue 9959
>>
>> So there might be some mechanism (maybe MUTEX_MAGIC) that avoids the
>> initialization of the mutex before init.
>> I am really not deep in all that pthread stuff…
>>
>> Is there the possiblity that there is a bug in the code base of issue
>> 9959?
>>
>>
>> Best regards
>>
>> Rouven
>>
>>
>> Am 18.04.2019 um 19:20 schrieb Robert Mustacchi <rm at joyent.com>:
>>
>> On 4/18/19 10:12 , Youzhong Yang wrote:
>>
>> Little bit confused. The page  https://www.openindiana.org/download/
>> shows
>> latest ISO is dated 2018.10, but 9959 fix was submitted on 11/26/2018, I
>> guess Hipster doesn't have it.
>>
>>
>> The ISO is just the install image. Generally, after using that to
>> bootstrap, then folks will update to latest bits via IPS. As a result,
>> the ISO isn't updated (and generally doesn't have to be) quite as often.
>>
>> Robert
>>
>> On Thu, Apr 18, 2019 at 12:38 PM Rouven WEILER <Rouven_Weiler at gmx.net>
>> wrote:
>>
>> On my system the issue is solved and the sources are patched accordingly.
>> Thus, the issue seems to be somewhere else.
>>
>> Am 18.04.2019 um 18:27 schrieb Rouven WEILER <Rouven_Weiler at gmx.net>:
>>
>> concerning openindiana I am using the most recent one. I right now asked
>> if the issue is solved in the latest version.
>>
>> another interesting thing is that older samba versions do not throw EBUSY,
>> although on the same underlying system.
>>
>>
>>
>> Am 18.04.2019 um 18:07 schrieb Youzhong Yang <youzhong at gmail.com>:
>>
>> That is interesting. I am curious about how pthread_mutexattr_init ends up
>> returning EBUSY. What verison of Openindiana are you running? is there a
>> link pointing to the fix that it has for 9959?
>>
>> On Thu, Apr 18, 2019 at 11:53 AM Rouven WEILER <Rouven_Weiler at gmx.net>
>> wrote:
>>
>> I understand and see that it works, but as written below
>> pthread_mutexattr_init throws EBUSY, not pthread_mutex_init.
>>
>> Also Openindiana seems to include the patch for issue 9959 mentioned. I
>> will clarify that to make sure.
>>
>> I just want to make sure that the patch you suggest aims for the issue
>> and is not a workaround.
>>
>>
>> Am 18.04.2019 um 16:09 schrieb Youzhong Yang <youzhong at gmail.com>:
>>
>> I reproduced the issue on a smartos version which does not have the fix
>> for  https://www.illumos.org/issues/9959.
>>
>> The pthread_mutex_init() at line 590 returns EBUSY:
>> ret = pthread_mutex_init(chain, &ma);
>>
>>
>> https://git.samba.org/?p=samba.git;a=blob;f=lib/tdb/common/mutex.c;h=8a122d5d6a2ec321bce25cf06062acc808f1e126;hb=HEAD#l590
>>
>>
>> Then I applied the following patch to lib/tdb/common/mutex.c:
>>
>> # git grep -B 1 -n pthread_mutex_init lib/tdb/common/mutex.c
>> lib/tdb/common/mutex.c-590-             bzero(chain,
>> sizeof(pthread_mutex_t));
>> lib/tdb/common/mutex.c:591:             ret = pthread_mutex_init(chain,
>> &ma);
>> --
>> lib/tdb/common/mutex.c-599-     bzero(&m->allrecord_mutex,
>> sizeof(pthread_mutex_t));
>> lib/tdb/common/mutex.c:600:     ret =
>> pthread_mutex_init(&m->allrecord_mutex, &ma);
>> --
>> lib/tdb/common/mutex.c-693-     bzero(&m, sizeof(pthread_mutex_t));
>> lib/tdb/common/mutex.c:694:     ret = pthread_mutex_init(&m, &ma);
>> --
>> lib/tdb/common/mutex.c-891-     bzero(ptr, sizeof(pthread_mutex_t));
>> lib/tdb/common/mutex.c:892:     ret = pthread_mutex_init(ptr, &ma);
>>
>> It works like a charm. Fair enough?
>>
>> So again I recommend either of the following two:
>> - apply 9959 fix to your OS image, or use a newer version which has 9959
>> fix
>> - bzero pthread_mutex_t struct before calling pthread_mutex_init().
>>
>> In my environment I applied bzero patch to all pthread_mutex_init() calls
>> in the entire samba code base, you know, who knows what is going to
>> happen.
>>
>>
>>
>> On Thu, Apr 18, 2019 at 4:29 AM Rouven WEILER <Rouven_Weiler at gmx.net>
>> wrote:
>>
>> That sounds reasonable!
>> Could you help me to dtrace that in a feasible way? Up to now I am not
>> used to dtrace. Except knowing what is and that its powerful.
>>
>> Am 18.04.2019 um 09:29 schrieb Jason King <jason.brian.king at gmail.com>:
>>
>> It appears that the robust mutexes are mmapped from the .tdb file.
>> EBUSY means that the mutex has already been intialized.  It also appears
>> that if the TDB_CLEAR_IF_FIRST flag is _not_ set when opening the tdb, it
>> will try to initialize the mutexes.  I’m wondering if the tdb file gets
>> created, the memory image of the initialized mutexes is getting saved,
>> then
>> when it’s opened again, if that flag is not set, it tries to re-initialize
>> the mutexes and fails (because they are already constructed) legitimately
>> returning EBUSY.  If the samba code is in fact trying to reinitialize an
>> already initialized mutex (dtrace might be able to help determine this —
>> at
>> this point it’s merely a supposition), that would explain the error.
>>
>>
>>
>>
>> From: Rouven WEILER <rouven_weiler at gmx.net> <rouven_weiler at gmx.net>
>> Reply: illumos-developer <developer at lists.illumos.org>
>> <developer at lists.illumos.org>
>> Date: April 18, 2019 at 12:48:48 AM
>> To: Youzhong Yang <youzhong at gmail.com> <youzhong at gmail.com>
>> Cc: Robert Mustacchi <rm at joyent.com> <rm at joyent.com>, Toomas Soome via
>> illumos-developer <developer at lists.illumos.org>
>> <developer at lists.illumos.org>
>> Subject:  [developer] Aw: Re: [Samba] gencache.tdb: device busy
>>
>> You are totally right. That's the same I got from the man pages.
>>
>> The responsible code lines of samba also seems to bear no issue and also
>> this issue arises on samba version 4.10.x not on earlier ones.
>> In addition, there seems to be only a problem with
>> illumos/openindiana...
>> That's why I do not know how to get that fixed. Up to now I hadn't time
>> to dtrace.
>>
>> Only thing I got so far: When the mutexattr is already set the return
>> value will result on undefined behaviour.
>> But why only after version 4.10.x on openindiana/illumos?
>>
>> Just for completion here the codelines of samba:
>>
>> https://git.samba.org/?p=samba.git;a=blob;f=lib/tdb/common/mutex.c;h=8a122d5d6a2ec321bce25cf06062acc808f1e126;hb=HEAD#l570
>>
>>
>>
>> Gesendet: Mittwoch, 17. April 2019 um 19:19 Uhr
>> Von: "Youzhong Yang" <youzhong at gmail.com>
>> An: "Rouven WEILER" <Rouven_Weiler at gmx.net>
>> Cc: "Robert Mustacchi" <rm at joyent.com>, "Toomas Soome via
>> illumos-developer" <developer at lists.illumos.org>
>> Betreff: Re: [Samba] gencache.tdb: device busy
>>
>> Unless I am totally wrong, pthread_mutexattr_init() shouldn't return
>> EBUSY:
>>
>>
>> http://src.illumos.org/source/xref/illumos-gate/usr/src/lib/libc/port/threads/pthr_mutex.c#37
>>
>>
>> On Tue, Apr 16, 2019 at 2:45 PM Rouven WEILER <Rouven_Weiler at gmx.net[
>> mailto:Rouven_Weiler at gmx.net <Rouven_Weiler at gmx.net>]> wrote:Hi Robert
>> & Youzhong,
>>
>> I got a response from the Samba-Devs.
>> Would the issue mentioned by Youzhong also hot pthread_mutexattr_t?
>>
>> Concerning dtrace I hadn’t the time yet but I will check for that.
>> Is there maybe a way to trace all forks of smbd? Usually the issue
>> arises after the fork…
>>
>> FIND BELOW THE STATEMENT FROM samba.org[http://samba.org]:
>>
>> This doesn't look exactly the same. In your
>> log below you say:
>>
>> ret = pthread_mutexattr_init(&ma); in line 570
>> I got errorno 16 --> device busy.
>>
>> But the bug report is about needing the value
>> passed to pthread_mutex_init() needing to be
>> zero, not the value passed to pthread_mutexattr_init().
>>
>> Can you just try adding '= {0}' to all the
>> places where we declare:
>>
>> pthread_mutexattr_t ma
>>
>> (i.e. make it say)
>>
>> pthread_mutexattr_t ma = {0};
>>
>> and see if it makes your problem go away ?
>>
>> END OF STATEMENT
>>
>>
>> Anfang der weitergeleiteten Nachricht:
>>
>> Von: Jeremy Allison via samba <samba at lists.samba.org[
>>
>> mailto:samba at lists.samba.org <samba at lists.samba.org>]>
>>
>> Betreff: Aw: [Samba] gencache.tdb: device busy
>> Datum: 16. April 2019 um 02:07:05 MESZ
>> An: Rouven WEILER <Rouven_Weiler at gmx.net[mailto:Rouven_Weiler at gmx.net
>>
>> <Rouven_Weiler at gmx.net>]>
>>
>> Kopie: samba at lists.samba.org[mailto:samba at lists.samba.org
>>
>> <samba at lists.samba.org>]
>>
>> Antwort an: Jeremy Allison <jra at samba.org[mailto:jra at samba.org
>>
>> <jra at samba.org>]>
>>
>>
>> On Fri, Apr 12, 2019 at 09:46:04AM +0200, Rouven WEILER via samba
>>
>> wrote:
>>
>>
>>
>> Hi Jeremy,
>>
>> I got some info on that topic from the illumos devs:
>>
>>
>> It's a sporadic issue, you're lucky enough to not encounter it on
>>
>> 4.9.5.
>>
>>
>> I confirmed in 4.10.2, it happens:
>>
>> winbindd.log:  tdb(/tmw-nas-3p/samba/var/lock/gencache.tdb):
>>
>> tdb_open_ex: tdb_mutex_init failed for
>> /tmw-nas-3p/samba/var/lock/gencache.tdb: Device busy
>>
>>
>> So either apply OS fix, or apply a software fix (zeroing
>>
>> pthread_mutex_t before pthread_mutex_init()).
>>
>>
>> This all should be based on the following issue:
>>
>>
>> https://www.illumos.org/issues/9959[https://www.illumos.org/issues/9959]
>>
>>
>> As far as I understand the issue making an software fix in samba for
>>
>> the OS seems more reasonable since the problem can occur sporadic and thus
>> again...
>>
>> Could you tell me how to help to fix that bug?
>>
>>
>> This doesn't look exactly the same. In your
>> log below you say:
>>
>> ret = pthread_mutexattr_init(&ma); in line 570
>> I got errorno 16 --> device busy.
>>
>> But the bug report is about needing the value
>> passed to pthread_mutex_init() needing to be
>> zero, not the value passed to pthread_mutexattr_init().
>>
>> Can you just try adding '= {0}' to all the
>> places where we declare:
>>
>> pthread_mutexattr_t ma
>>
>> (i.e. make it say)
>>
>> pthread_mutexattr_t ma = {0};
>>
>> and see if it makes your problem go away ?
>>
>>
>> Gesendet: Dienstag, 02. April 2019 um 18:59 Uhr
>> Von: "Jeremy Allison" <jra at samba.org[mailto:jra at samba.org
>>
>> <jra at samba.org>]>
>>
>> An: "Rouven WEILER" <Rouven_Weiler at gmx.net[
>>
>> mailto:Rouven_Weiler at gmx.net <Rouven_Weiler at gmx.net>]>
>>
>> Cc: samba at lists.samba.org[mailto:samba at lists.samba.org
>>
>> <samba at lists.samba.org>]
>>
>> Betreff: Re: Re: [Samba] gencache.tdb: device busy
>> On Tue, Apr 02, 2019 at 01:24:19PM +0200, Rouven WEILER wrote:
>>
>>
>>
>> I added LOG statements to mutex.c printing the strerror from the
>>
>> return values.
>>
>>
>> From
>>
>> ret = pthread_mutexattr_init(&ma); in line 570
>>
>> I got errorno 16 --> device busy.
>>
>> But due the manual of pthread_mutex_attr_init() there should only be
>>
>> ENOMEM and EINVAL...
>>
>>
>> Doing gdb smbd -> run
>> Doe not reveal anything, but I never used gdb for debugging
>>
>> anything. I just gave it a shot...
>>
>>
>> Sorry, looks a little like the libc/pthread_mutex support
>> on Solaris is broken. You might want to raise this on a
>> Solaris/Illumos forum.
>>
>> --
>> To unsubscribe from this list go to the following URL and read the
>> instructions:
>>
>>
>> https://lists.samba.org/mailman/options/samba[https://lists.samba.org/mailman/options/samba]
>>
>>
>> --
>> To unsubscribe from this list go to the following URL and read the
>> instructions:
>>
>>
>> https://lists.samba.org/mailman/options/samba[https://lists.samba.org/mailman/options/samba]
>>
>>
>> ------------------------------------------
>> illumos: illumos-developer
>> Permalink:
>>
>> https://illumos.topicbox.com/groups/developer/T38b7c9b3f92fa699-M99a2afeb19714648c051ab5f
>> Delivery options:
>> https://illumos.topicbox.com/groups/developer/subscription
>>
>> *illumos <https://illumos.topicbox.com/latest>* / illumos-developer /
>>
>> see discussions <https://illumos.topicbox.com/groups/developer> +
>> participants <https://illumos.topicbox.com/groups/developer/members> +
>> delivery
>> options <https://illumos.topicbox.com/groups/developer/subscription>
>> Permalink
>> <
>> https://illumos.topicbox.com/groups/developer/T38b7c9b3f92fa699-M2372380a78d4d5078f031f0c
>> >
>>
>>
>>
>> ------------------------------------------
>> illumos: illumos-developer
>> Permalink:
>> https://illumos.topicbox.com/groups/developer/T38b7c9b3f92fa699-M1048b12f29cda459a4de0893
>> Delivery options:
>> https://illumos.topicbox.com/groups/developer/subscription
>>
>>
>> _______________________________________________
>> oi-dev mailing list
>> oi-dev at openindiana.org
>> https://openindiana.org/mailman/listinfo/oi-dev
>>
> _______________________________________________
> oi-dev mailing list
> oi-dev at openindiana.org
> https://openindiana.org/mailman/listinfo/oi-dev
>
> _______________________________________________
> oi-dev mailing list
> oi-dev at openindiana.org
> https://openindiana.org/mailman/listinfo/oi-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openindiana.org/pipermail/oi-dev/attachments/20190421/0166519e/attachment-0005.html>