Re: [patch] POSIX advisory mode lock panic fix by Dfly
:Joerg Sonnenberger <joerg@britannica.bec.de> scribbled:
:> OK, here is the updated version. I moved the actual handling into lockf.h,
:> only the rlimit check is left in resource.c and the call to lf_count_adjust.
:>
:> Matt, what do you think?
:>
:> Joerg
:
:This looks like what I had finished last night and is actually almost
:totally identical except for a couple of variable names ^_^. Nice work.
:
:--
:Kind regards,
:
:Devon H. O'Dell
Well, I would say that we are almost there.
I'm a bit confused about lf_count_adjust() ... why is it changing
the newowner pointer? Maybe you meant newowner->ui_posixlocks +=
instead of newowner += ?
lf_count_increment() is still entanged with the resource check.
lf_count_increment() should be integrated with or called from
the allocation code, and the resource check should be done
separately. lf_count_decrement() should be integrated with the
structure freeing code. When you have the increment/decrement
disconnected the way it is in the patch, chances are very high that
bugs will creep in.
-Matt
Matthew Dillon
<dillon@backplane.com>
討論串 (同標題文章)
完整討論串 (本文為第 10 之 17 篇)：