Return-Path: Received: from merit-proxy02.merit.edu ([207.75.116.194]:53334 "EHLO merit-proxy02.merit.edu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752746Ab1GGQ0P (ORCPT ); Thu, 7 Jul 2011 12:26:15 -0400 Message-Id: <6882cf3d981a46649b0c04316b566c8d7bd30584.1310055433.git.rees@umich.edu> In-Reply-To: References: Subject: [PATCH 3/6] SQUASHME: pnfs: teach layoutcommit handle multiple segments To: Benny Halevy Cc: linux-nfs@vger.kernel.org, peter honeyman Date: Thu, 7 Jul 2011 12:26:13 -0400 From: Jim Rees Sender: linux-nfs-owner@vger.kernel.org List-ID: Content-Type: text/plain MIME-Version: 1.0 From: Peng Tao This rewrites commit 9658e7532d35faf0c09b8b34fd5808e5e58070ac in a way more reasonable. Layoutcommit takes a list of segments and last write offset is saved at inode level. Signed-off-by: Peng Tao --- fs/nfs/nfs4filelayout.c | 2 +- fs/nfs/nfs4proc.c | 8 +++++- fs/nfs/pnfs.c | 60 ++++++++++++++++++---------------------------- fs/nfs/pnfs.h | 4 ++- include/linux/nfs_xdr.h | 2 +- 5 files changed, 36 insertions(+), 40 deletions(-) diff --git a/fs/nfs/nfs4filelayout.c b/fs/nfs/nfs4filelayout.c index af9bf9e..6d7f937 100644 --- a/fs/nfs/nfs4filelayout.c +++ b/fs/nfs/nfs4filelayout.c @@ -170,7 +170,7 @@ filelayout_set_layoutcommit(struct nfs_write_data *wdata) pnfs_set_layoutcommit(wdata); dprintk("%s ionde %lu pls_end_pos %lu\n", __func__, wdata->inode->i_ino, - (unsigned long) wdata->lseg->pls_end_pos); + (unsigned long) NFS_I(wdata->inode)->layout->plh_lwb); } /* diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c index ace9d37..795033c5 100644 --- a/fs/nfs/nfs4proc.c +++ b/fs/nfs/nfs4proc.c @@ -5963,10 +5963,16 @@ nfs4_layoutcommit_done(struct rpc_task *task, void *calldata) static void nfs4_layoutcommit_release(void *calldata) { struct nfs4_layoutcommit_data *data = calldata; + struct pnfs_layout_segment *lseg, *tmp; pnfs_cleanup_layoutcommit(data->args.inode, data); /* Matched by references in pnfs_set_layoutcommit */ - put_lseg(data->lseg); + list_for_each_entry_safe(lseg, tmp, &data->lseg_list, pls_lc_list) { + list_del_init(&lseg->pls_lc_list); + if (test_and_clear_bit(NFS_LSEG_LAYOUTCOMMIT, + &lseg->pls_flags)) + put_lseg(lseg); + } put_rpccred(data->cred); kfree(data); } diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c index 1fdc8f7..30a0394 100644 --- a/fs/nfs/pnfs.c +++ b/fs/nfs/pnfs.c @@ -240,6 +240,7 @@ static void init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg) { INIT_LIST_HEAD(&lseg->pls_list); + INIT_LIST_HEAD(&lseg->pls_lc_list); atomic_set(&lseg->pls_refcount, 1); smp_mb(); set_bit(NFS_LSEG_VALID, &lseg->pls_flags); @@ -1273,25 +1274,18 @@ pnfs_try_to_read_data(struct nfs_read_data *rdata, } /* - * Currently there is only one (whole file) write lseg. + * There can be multiple RW segments. */ -static struct pnfs_layout_segment *pnfs_list_write_lseg(struct inode *inode) +static void pnfs_list_write_lseg(struct inode *inode, struct list_head *listp) { - struct pnfs_layout_segment *lseg, *rv = NULL; - loff_t max_pos = 0; + struct pnfs_layout_segment *lseg; list_for_each_entry(lseg, &NFS_I(inode)->layout->plh_segs, pls_list) { - if (lseg->pls_range.iomode == IOMODE_RW) { - if (max_pos < lseg->pls_end_pos) - max_pos = lseg->pls_end_pos; - if (test_and_clear_bit - (NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags)) - rv = lseg; - } + if (lseg->pls_range.iomode == IOMODE_RW && + test_bit(NFS_LSEG_LAYOUTCOMMIT, &lseg->pls_flags) && + list_empty(&lseg->pls_lc_list)) + list_add(&lseg->pls_lc_list, listp); } - rv->pls_end_pos = max_pos; - - return rv; } void @@ -1299,27 +1293,25 @@ pnfs_set_layoutcommit(struct nfs_write_data *wdata) { struct nfs_inode *nfsi = NFS_I(wdata->inode); loff_t end_pos = wdata->mds_offset + wdata->res.count; - loff_t isize = i_size_read(wdata->inode); bool mark_as_dirty = false; spin_lock(&nfsi->vfs_inode.i_lock); if (!test_and_set_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) { - /* references matched in nfs4_layoutcommit_release */ - get_lseg(wdata->lseg); - set_bit(NFS_LSEG_LAYOUTCOMMIT, &wdata->lseg->pls_flags); - wdata->lseg->pls_lc_cred = - get_rpccred(wdata->args.context->state->owner->so_cred); mark_as_dirty = true; + nfsi->layout->plh_lc_cred = + get_rpccred(wdata->args.context->state->owner->so_cred); dprintk("%s: Set layoutcommit for inode %lu ", __func__, wdata->inode->i_ino); } - if (end_pos > isize) - end_pos = isize; - if (end_pos > wdata->lseg->pls_end_pos) - wdata->lseg->pls_end_pos = end_pos; + if (!test_and_set_bit(NFS_LSEG_LAYOUTCOMMIT, &wdata->lseg->pls_flags)) { + /* references matched in nfs4_layoutcommit_release */ + get_lseg(wdata->lseg); + } + if (end_pos > nfsi->layout->plh_lwb) + nfsi->layout->plh_lwb = end_pos; spin_unlock(&nfsi->vfs_inode.i_lock); dprintk("%s: lseg %p end_pos %llu\n", - __func__, wdata->lseg, wdata->lseg->pls_end_pos); + __func__, wdata->lseg, nfsi->layout->plh_lwb); /* if pnfs_layoutcommit_inode() runs between inode locks, the next one * will be a noop because NFS_INO_LAYOUTCOMMIT will not be set */ @@ -1351,7 +1343,6 @@ pnfs_layoutcommit_inode(struct inode *inode, bool sync) { struct nfs4_layoutcommit_data *data; struct nfs_inode *nfsi = NFS_I(inode); - struct pnfs_layout_segment *lseg; struct rpc_cred *cred; loff_t end_pos; int status = 0; @@ -1375,23 +1366,20 @@ pnfs_layoutcommit_inode(struct inode *inode, bool sync) kfree(data); goto out; } - /* - * Currently only one (whole file) write lseg which is referenced - * in pnfs_set_layoutcommit and will be found. - */ - lseg = pnfs_list_write_lseg(inode); - end_pos = lseg->pls_end_pos; - cred = lseg->pls_lc_cred; - lseg->pls_end_pos = 0; - lseg->pls_lc_cred = NULL; + INIT_LIST_HEAD(&data->lseg_list); + pnfs_list_write_lseg(inode, &data->lseg_list); + + end_pos = nfsi->layout->plh_lwb; + cred = nfsi->layout->plh_lc_cred; + nfsi->layout->plh_lwb = 0; + nfsi->layout->plh_lc_cred = NULL; memcpy(&data->args.stateid.data, nfsi->layout->plh_stateid.data, sizeof(nfsi->layout->plh_stateid.data)); spin_unlock(&inode->i_lock); data->args.inode = inode; - data->lseg = lseg; data->cred = cred; nfs_fattr_init(&data->fattr); data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h index a0c856c..cfe332c 100644 --- a/fs/nfs/pnfs.h +++ b/fs/nfs/pnfs.h @@ -41,12 +41,12 @@ enum { struct pnfs_layout_segment { struct list_head pls_list; + struct list_head pls_lc_list; struct pnfs_layout_range pls_range; atomic_t pls_refcount; unsigned long pls_flags; struct pnfs_layout_hdr *pls_layout; struct rpc_cred *pls_lc_cred; /* LAYOUTCOMMIT credential */ - loff_t pls_end_pos; /* LAYOUTCOMMIT write end */ }; enum pnfs_try_status { @@ -132,6 +132,8 @@ struct pnfs_layout_hdr { unsigned long plh_block_lgets; /* block LAYOUTGET if >0 */ u32 plh_barrier; /* ignore lower seqids */ unsigned long plh_flags; + loff_t plh_lwb; /* last write byte for layoutcommit */ + struct rpc_cred *plh_lc_cred; /* layoutcommit cred */ struct inode *plh_inode; }; diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h index 065e941..27c12c7 100644 --- a/include/linux/nfs_xdr.h +++ b/include/linux/nfs_xdr.h @@ -275,7 +275,7 @@ struct nfs4_layoutcommit_res { struct nfs4_layoutcommit_data { struct rpc_task task; struct nfs_fattr fattr; - struct pnfs_layout_segment *lseg; + struct list_head lseg_list; struct rpc_cred *cred; struct nfs4_layoutcommit_args args; struct nfs4_layoutcommit_res res; -- 1.7.4.1