On Mon, Mar 2, 2015 at 5:29 PM, Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx> wrote: > On Mon, Mar 2, 2015 at 5:01 PM, Olga Kornievskaia <aglo@xxxxxxxxx> wrote: >> On Mon, Mar 2, 2015 at 3:53 PM, Olga Kornievskaia <aglo@xxxxxxxxx> wrote: >>> On Mon, Mar 2, 2015 at 3:47 PM, Trond Myklebust >>> <trond.myklebust@xxxxxxxxxxxxxxx> wrote: >>>> Hi Olga, >>>> >>>> On Mon, Mar 2, 2015 at 3:15 PM, Olga Kornievskaia <aglo@xxxxxxxxx> wrote: >>>>> Hi folks, >>>>> >>>>> I'm experiencing that CLOSE uses a delegation stateid instead of the >>>>> open stateid which I think is what the spec says. Server replies with >>>>> BAD_STATEID. >>>>> >>>>> Is this a bug or did I misread the spec? Thanks. >>>>> >>>> >>>> That would be a client bug. Do you have a reproducer? >>> >>> Yep. Just cat a (2nd) file after mount (i.e.., a file needs to have a >>> delegation). A CLOSE will use a delegation stateid. Problem is seenl >>> on the network trace. It will also leads to failure on unmount with >>> CLIENTID_BUSY because there is still an open state that client never >>> released. Please note that both "cat" and "unmount" will "succeed" >>> from the user's perspective. Thus, unless testing also looks at the >>> network trace, this failure will never be caught. >> >> Anna pointed me at the commit 566fcec60. It seems to be that's what >> broke it as it removed the use of openstateid for the stateid arg. But >> I really don't understand the necessity of the patch. CLOSE must >> always use the openstateid. Therefore it doesn't need to worry that >> stateid is changed from openstateid to delegation stateid or locking >> stateid. It should just use the openstateid as it did before. >> > > Doh! Yes, the change to ->stateid is wrong. I must have been on borken > autopilot... > > The reason for the patch itself is to ensure the seqid hasn't changed. > It has nothing to do with delegation stateids or locking. > Can either one of you please send a patch to fix up 566fcec60? Please > note that we also need to change the comparison in nfs4_close_done to > match the copy in nfs4_close_prepare. > Something like works (btw this is on top of nfs-for-next which also has that problem): After 566fcec60 the client uses the "current stateid" from the nfs4_state structure to close a file. This could potentially contain a delegation stateid, which is disallowed by the protocol and causes servers to return NFS4ERR_BAD_STATEID. This patch restores the (correct) behavior of sending the open stateid to close a file. Reported-by: Olga Kornievskaia <kolga@xxxxxxxxxx> Fixes: 566fcec60 (NFSv4: Fix an atomicity problem in CLOSE) Signed-off-by: Anna Schumaker <Anna.Schumaker@xxxxxxxxxx> diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index a211daf..732526e 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -2655,7 +2655,7 @@ static void nfs4_close_done(struct rpc_task *task, void *d case -NFS4ERR_BAD_STATEID: case -NFS4ERR_EXPIRED: if (!nfs4_stateid_match(&calldata->arg.stateid, - &state->stateid)) { + &state->open_stateid)) { rpc_restart_call_prepare(task); goto out_release; } @@ -2691,7 +2691,7 @@ static void nfs4_close_prepare(struct rpc_task *task, void is_rdwr = test_bit(NFS_O_RDWR_STATE, &state->flags); is_rdonly = test_bit(NFS_O_RDONLY_STATE, &state->flags); is_wronly = test_bit(NFS_O_WRONLY_STATE, &state->flags); - nfs4_stateid_copy(&calldata->arg.stateid, &state->stateid); + nfs4_stateid_copy(&calldata->arg.stateid, &state->open_stateid); /* Calculate the change in open mode */ calldata->arg.fmode = 0; if (state->n_rdwr == 0) { > -- > Trond Myklebust > Linux NFS client maintainer, PrimaryData > trond.myklebust@xxxxxxxxxxxxxxx -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html