I ported these changes to illumos and ran ztest. It fails on the 2nd call to zrl_destroy() in dnode_hold_impl(). I believe this change is incorrect; the handle is still potentially in use here. We simply failed to make the handle point to our dnode.
--matt On Thu, Jul 3, 2014 at 10:36 AM, Matthew Ahrens <[email protected]> wrote: > Thanks for pointing these out. I will port this commit to illumos. Look > for an illumos review request shortly. > > --matt > > > On Thu, Jul 3, 2014 at 1:31 AM, Jorgen Lundman <[email protected]> > wrote: > >> >> Hello list, >> >> Apple made the mutex type in Darwin be of opaque type, so we have to call >> the API to allocate a new mutex. We then discovered that our total active >> number of mutex allocations are ever increasing. >> >> I wrote a framework around the SPL mutex API and dump the leaked nodes >> after module unload, and found a few places to fix up. I have noticed that >> some of these mutex_destroy calls are also missing in ZOL and FreeBSD. I >> did not check IllumOS (sorry). >> >> Of course, on most ZFS platforms, the mutex "data" are housed inside >> various structures and get freed automatically. (Making mutex_destroy >> somewhat optional) but perhaps upstream are still interested in this >> patch, >> in the name of code "correctness". >> >> https://github.com/openzfsonosx/zfs/commit/2e30c792597a >> >> Lund >> >> -- >> Jorgen Lundman | <[email protected]> >> Unix Administrator | +81 (0)3 -5456-2687 ext 1017 (work) >> Shibuya-ku, Tokyo | +81 (0)90-5578-8500 (cell) >> Japan | +81 (0)3 -3375-1767 (home) >> _______________________________________________ >> developer mailing list >> [email protected] >> http://lists.open-zfs.org/mailman/listinfo/developer >> > >
_______________________________________________ developer mailing list [email protected] http://lists.open-zfs.org/mailman/listinfo/developer
