Return-Path: Received: from daytona.panasas.com ([67.152.220.89]:6425 "EHLO daytona.panasas.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751740Ab1AXClW (ORCPT ); Sun, 23 Jan 2011 21:41:22 -0500 Message-ID: <4D3CE6D1.7000404@panasas.com> Date: Sun, 23 Jan 2011 21:41:21 -0500 From: Benny Halevy To: Fred Isaman CC: NFS list Subject: Re: [pnfs] [PATCH 1/2] pnfs: trim write count to wsize when retrying via MDS References: <4B804E57.4010507@panasas.com> <1266699969-21920-1-git-send-email-bhalevy@panasas.com> In-Reply-To: Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-nfs-owner@vger.kernel.org List-ID: MIME-Version: 1.0 On 2011-01-20 07:20, Fred Isaman wrote: > I'm going over the write code carefully preparing for wave4, and came > across some code I traced to this patch. > > Why is this needed? Why doesn't the pg_bsize check in > nfs_pageio_do_add_request() prevent this from ever being invoked? Apparently this got squashed in too early. It's actually needed only after "pnfs: Introduce pnfs_call_done and io done callbacks" which is post pnfs-submit. Benny > > Fred > > On Sat, Feb 20, 2010 at 4:06 PM, Benny Halevy wrote: >> We coalesce pages based on the strip size that may be bigger >> than the server can handle in one write (and even exceed the >> maximum RPC message size on the server) >> >> Signed-off-by: Benny Halevy >> --- >> fs/nfs/nfs4proc.c | 7 +++++++ >> fs/nfs/write.c | 8 ++++++++ >> include/linux/nfs_xdr.h | 1 + >> 3 files changed, 16 insertions(+), 0 deletions(-) >> >> diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c >> index 8132a96..93b1963 100644 >> --- a/fs/nfs/nfs4proc.c >> +++ b/fs/nfs/nfs4proc.c >> @@ -3180,6 +3180,13 @@ static int pnfs4_write_done(struct rpc_task *task, struct nfs_write_data *data) >> struct nfs_client *client = mds_svr->nfs_client; >> int status = task->tk_status >= 0 ? 0 : task->tk_status; >> >> + /* restore original count after retry? */ >> + if (data->pdata.orig_count) { >> + dprintk("%s: restoring original count %u\n", __func__, >> + data->pdata.orig_count); >> + data->args.count = data->pdata.orig_count; >> + } >> + >> if (data->pdata.pnfsflags & PNFS_NO_RPC) >> return 0; >> >> diff --git a/fs/nfs/write.c b/fs/nfs/write.c >> index 90ce78e..fee6f00 100644 >> --- a/fs/nfs/write.c >> +++ b/fs/nfs/write.c >> @@ -1108,6 +1108,14 @@ void nfs_write_prepare(struct rpc_task *task, void *calldata) >> #ifdef CONFIG_PNFS >> if (data->fldata.ds_nfs_client) >> clp = data->fldata.ds_nfs_client; >> + else if (data->args.count > NFS_SERVER(data->inode)->wsize) { >> + /* retrying via MDS? */ >> + data->pdata.orig_count = data->args.count; >> + data->args.count = NFS_SERVER(data->inode)->wsize; >> + dprintk("%s: trimmed count %u to wsize %u\n", __func__, >> + data->pdata.orig_count, data->args.count); >> + } else >> + data->pdata.orig_count = 0; >> #endif /* CONFIG_PNFS */ >> >> if (nfs4_setup_sequence(clp, &data->args.seq_args, >> diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h >> index efc4ca0..13ccd48 100644 >> --- a/include/linux/nfs_xdr.h >> +++ b/include/linux/nfs_xdr.h >> @@ -970,6 +970,7 @@ struct nfs_page; >> struct pnfs_call_data { >> struct pnfs_layout_segment *lseg; >> const struct rpc_call_ops *call_ops; >> + u32 orig_count; /* for retry via MDS */ >> int pnfs_error; >> u8 pnfsflags; >> u8 how; /* for FLUSH_STABLE */ >> -- >> 1.6.4.4 >> >> _______________________________________________ >> pNFS mailing list >> pNFS@linux-nfs.org >> http://linux-nfs.org/cgi-bin/mailman/listinfo/pnfs >>