Return-Path: Received: from daytona.panasas.com ([67.152.220.89]:33252 "EHLO daytona.int.panasas.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1750920Ab0IMJuc (ORCPT ); Mon, 13 Sep 2010 05:50:32 -0400 Message-ID: <4C8DF3E5.4050108@panasas.com> Date: Mon, 13 Sep 2010 11:50:29 +0200 From: Boaz Harrosh To: Trond Myklebust CC: NFS list , Benny Halevy , Andy Adamson , Fred Isaman Subject: Re: [RFC 1/3] SQUASHME: Generalize the device cache so it can be used by all layouts References: <4C8D0759.6090201@panasas.com> <4C8D07D5.6010006@panasas.com> <1284320533.7880.131.camel@heimdal.trondhjem.org> In-Reply-To: <1284320533.7880.131.camel@heimdal.trondhjem.org> Content-Type: text/plain; charset=UTF-8 Sender: linux-nfs-owner@vger.kernel.org List-ID: MIME-Version: 1.0 On 09/12/2010 09:42 PM, Trond Myklebust wrote: > On Sun, 2010-09-12 at 19:03 +0200, Boaz Harrosh wrote: >> Current code thinks that there can be a single device_id per >> layout segment. Change that to assume no relations between segments >> and device_ids. It's now up to the layout-driver to make any relations. >> >> Files layout driver is fixed in next patch. >> >> Signed-off-by: Boaz Harrosh >> --- >> fs/nfs/pnfs.c | 20 +++++--------------- >> include/linux/nfs4_pnfs.h | 6 +----- >> 2 files changed, 6 insertions(+), 20 deletions(-) >> >> diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c >> index eb4e092..f26abc0 100644 >> --- a/fs/nfs/pnfs.c >> +++ b/fs/nfs/pnfs.c >> @@ -1892,26 +1892,16 @@ nfs4_init_deviceid_node(struct nfs4_deviceid *d) >> } >> EXPORT_SYMBOL(nfs4_init_deviceid_node); >> >> -/* Called from layoutdriver_io_operations->alloc_lseg */ >> -void >> -nfs4_set_layout_deviceid(struct pnfs_layout_segment *l, struct nfs4_deviceid *d) >> -{ >> - dprintk("%s [%d]\n", __func__, atomic_read(&d->de_kref.refcount)); >> - l->deviceid = d; >> -} >> -EXPORT_SYMBOL(nfs4_set_layout_deviceid); >> - >> /* Called from layoutdriver_io_operations->free_lseg */ >> void >> -nfs4_put_unset_layout_deviceid(struct pnfs_layout_segment *l, >> - struct nfs4_deviceid *d, >> - void (*free_callback)(struct kref *)) >> +nfs4_put_deviceid(struct nfs4_deviceid_cache *c, >> + struct nfs4_deviceid *d) >> { >> dprintk("%s [%d]\n", __func__, atomic_read(&d->de_kref.refcount)); >> - l->deviceid = NULL; >> - kref_put(&d->de_kref, free_callback); >> + kref_put(&d->de_kref, c->dc_free_callback); >> + /* Do we need to return the deviceid_cache ref */ >> } >> -EXPORT_SYMBOL(nfs4_put_unset_layout_deviceid); >> +EXPORT_SYMBOL(nfs4_put_deviceid); >> >> /* Find and reference a deviceid */ >> struct nfs4_deviceid * >> diff --git a/include/linux/nfs4_pnfs.h b/include/linux/nfs4_pnfs.h >> index 577cd2b..dc3410e 100644 >> --- a/include/linux/nfs4_pnfs.h >> +++ b/include/linux/nfs4_pnfs.h >> @@ -111,7 +111,6 @@ struct pnfs_layout_segment { >> struct kref kref; >> bool valid; >> struct pnfs_layout_hdr *layout; >> - struct nfs4_deviceid *deviceid; >> u8 ld_data[]; /* layout driver private data */ >> }; >> >> @@ -287,11 +286,8 @@ extern struct nfs4_deviceid *nfs4_find_get_deviceid( >> struct pnfs_deviceid *); >> extern struct nfs4_deviceid *nfs4_add_get_deviceid(struct nfs4_deviceid_cache *, >> struct nfs4_deviceid *); >> -extern void nfs4_set_layout_deviceid(struct pnfs_layout_segment *, >> +extern void nfs4_put_deviceid(struct nfs4_deviceid_cache *, >> struct nfs4_deviceid *); >> -extern void nfs4_put_unset_layout_deviceid(struct pnfs_layout_segment *, >> - struct nfs4_deviceid *, >> - void (*free_callback)(struct kref *)); >> extern void nfs4_delete_device(struct nfs4_deviceid_cache *, >> struct pnfs_deviceid *); >> > > This doesn't look like it needs to be part of the initial single-layout > submission. Why is it being labelled as a squashme? > > Trond No, this is a device_id cache per nfs-client it is part of the initial getdeviceinfo call. Many "single-layout" may refer to the same device_id therefor it is cached so the long getdeviceinfo call can be done only once. Cheers Boaz