View Issue Details
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0001373||libextractor||libextractor main library||public||2008-06-22 04:19||2012-09-09 02:36|
|Product Version||current SVN|
|Target Version||1.0.0||Fixed in Version||1.0.0|
|Summary||0001373: module handling in LE is not thread-safe|
|Description||Recently, I added a multithread test case to LE. Running it shows crashes in seemingly random places. This was observed on OS X. I tracked the problem to module loading/unloading, and reading the libtool docs shows the reason: libtool (libltdl) module loading is not thread-safe, and LE uses libtool functions without protection. I see no way around this other than adding locking around libtool functions (or documentating that all module handling is not thread-safe.) I suppose this applies to GNUnet as well. |
|Tags||No tags attached.|
link to relevant docs:
Well, in GNUnet we have already the strong assumption that all loading/unloading is done in single-threaded mode during startup/shutdown. Everything else will break tons of stuff.
For LE -- well, the assumption is that obviously the user does not load/unload on the same handle while also calling extract on that handle (that would be invalid). If the problem is that two threads cannot call, say, load-default-extractors at the same time, well, then we have to add some locking.
Yes. The problem is that calling any of the functions that call libltdl functions (eg. lt_dlopenex) unprotected is not thread-safe. For instance, EXTRACTOR_loadDefaultLibraries is not thread-safe.
Of course, sharing a handle between threads would be asking for trouble anyway, and I'm not talking about any weird programming practices here. Unless..unless an EXTRACTOR_KeywordList would also be invalid after freeing the EXTRACTOR_ExtractorList that extracted it, because I'm freeing the extractors before the keywordlist in my testcase.
Sharing a handle between threads is *legal*, as long as all of the threads use it "read only" (reading == extracting, writing == manipulating plugin set).
And no, the keyword list is independent of the extractor list (and so freeing them in any order is fine).
So yes, overall this sounds like we may want to add locking around the ltdl functions (and only around those functions).
||Fixed in SVN 7306. Tested on Mac OS X/ppc and Ubuntu 8.04/x86.|
||Documentation needs an update also.|
||Are you going to do the update?|
||Yeah, I can do it.|
||Took a stab at updating the docs -- the texinfo doc was the only place where reentrancy was mentioned, so, I updated only that. SVN 7311.|
||Should be fixed in SVN 7311.|
|2008-06-22 04:19||holin||New Issue|
|2008-06-22 07:32||holin||Note Added: 0003520|
|2008-06-22 09:16||Christian Grothoff||Note Added: 0003521|
|2008-06-22 09:27||holin||Note Added: 0003522|
|2008-06-22 10:00||Christian Grothoff||Note Added: 0003525|
|2008-06-22 10:54||holin||Status||new => assigned|
|2008-06-22 10:54||holin||Assigned To||=> holin|
|2008-06-22 12:49||holin||Note Added: 0003530|
|2008-06-22 12:50||holin||Note Added: 0003531|
|2008-06-22 22:49||Christian Grothoff||Note Added: 0003533|
|2008-06-23 04:26||holin||Note Added: 0003536|
|2008-06-23 10:09||holin||Note Added: 0003537|
|2008-06-23 10:10||holin||Status||assigned => resolved|
|2008-06-23 10:10||holin||Fixed in Version||=> current SVN|
|2008-06-23 10:10||holin||Resolution||open => fixed|
|2008-06-23 10:10||holin||Note Added: 0003538|
|2008-07-13 19:40||Christian Grothoff||Status||resolved => closed|
|2012-09-09 02:36||Christian Grothoff||Fixed in Version||current SVN => 1.0.0|
|2012-09-09 02:36||Christian Grothoff||Target Version||=> 1.0.0|