Re: [PATCH] remove useless MOD_{INC,DEC}_USE_COUNT from sunrpc

[Trond Myklebust]

>>>>> " " == Trond Myklebust <[e-mail]> writes:

     > There don't seem to be any checks and balances to prevent the
     > user from removing the sunrpc module immediately after
     > svc_create_thread() gets called.  I presume that the call to
     > lockd_up() in nfsd() will help (since that calls rpciod_up()
     > and hence MOD_INC_COUNT) but AFAICS there appears to be plenty
     > of possible races before we get to that point.

Sorry. I misread that code. The only nfsd problem goes the other way
round. nfsd() is not exported, and nfsd_svc() does not wait on the
threads to start, hence you have a race between nfsd_svc(), and the
MOD_INC_USE_COUNT inside the nfsd() threads.

Cheers,
  Trond