Audit Kernel Events, weekly report #3
gpf.kira at gmail.com
Mon Jun 14 13:28:02 UTC 2010
- Haven't done much this week due to university exams which will keep
me busy for another week; I doubt this is going to be a problem as
community bonding period was used for coding. This week I did some
more testing of my patches and was buffled by the following issue.
Although vn_fullpath(9) does not require any locks to be held before
calling it, holding locks may result in a deadlock situation? After
applying my early NFS patches, I would experience deadlocks when
vn_fullpath() would try to vn_lock() the vnode in question. The KPI
was called in the middle of a NFS pseudo-syscall, right after
obtaining the vnode in question.
That's why vn_fullpath() is now called at the end of each NFS
pseudo-syscall. This week, I altered NFS code so that it would get
called at the middle of each NFS RPC and now it would *not* deadlock.
I wonder if this was due to some bug that I previously introduced and
somehow fixed in the process.
Efstratios "GPF" Karatzas
More information about the soc-status