From fe238e601d2519f259103ab65caea3b077ed7b39 Mon Sep 17 00:00:00 2001 From: Dave Wysochanski Date: Fri, 1 Apr 2016 13:45:09 -0400 Subject: NFS: Save struct inode * inside nfs_commit_info to clarify usage of i_lock Commit ea2cf22 created nfs_commit_info and saved &inode->i_lock inside this NFS specific structure. This obscures the usage of i_lock. Instead, save struct inode * so later it's clear the spinlock taken is i_lock. Should be no functional change. Signed-off-by: Dave Wysochanski Signed-off-by: Anna Schumaker --- fs/nfs/filelayout/filelayout.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'fs/nfs/filelayout/filelayout.c') diff --git a/fs/nfs/filelayout/filelayout.c b/fs/nfs/filelayout/filelayout.c index 3384dc8e6683..3e50057eeadf 100644 --- a/fs/nfs/filelayout/filelayout.c +++ b/fs/nfs/filelayout/filelayout.c @@ -795,7 +795,7 @@ filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg, buckets[i].direct_verf.committed = NFS_INVALID_STABLE_HOW; } - spin_lock(cinfo->lock); + spin_lock(&cinfo->inode->i_lock); if (cinfo->ds->nbuckets >= size) goto out; for (i = 0; i < cinfo->ds->nbuckets; i++) { @@ -811,7 +811,7 @@ filelayout_alloc_commit_info(struct pnfs_layout_segment *lseg, swap(cinfo->ds->buckets, buckets); cinfo->ds->nbuckets = size; out: - spin_unlock(cinfo->lock); + spin_unlock(&cinfo->inode->i_lock); kfree(buckets); return 0; } -- cgit v1.2.3-55-g7522 From c7d73af2d249f0323f5cdb171a59497ce80011fb Mon Sep 17 00:00:00 2001 From: Tom Haynes Date: Wed, 25 May 2016 07:31:14 -0700 Subject: pnfs: pnfs_update_layout needs to consider if strict iomode checking is on As flexfiles has FF_FLAGS_NO_READ_IO, there is a need to generically support enforcing that a IOMODE_RW segment will not allow READ I/O. Signed-off-by: Tom Haynes Signed-off-by: Anna Schumaker --- fs/nfs/filelayout/filelayout.c | 2 ++ fs/nfs/flexfilelayout/flexfilelayout.c | 49 +++++++++++++++++++++++++--------- fs/nfs/pnfs.c | 34 ++++++++++++++--------- fs/nfs/pnfs.h | 1 + 4 files changed, 61 insertions(+), 25 deletions(-) (limited to 'fs/nfs/filelayout/filelayout.c') diff --git a/fs/nfs/filelayout/filelayout.c b/fs/nfs/filelayout/filelayout.c index 3e50057eeadf..aa59757389dc 100644 --- a/fs/nfs/filelayout/filelayout.c +++ b/fs/nfs/filelayout/filelayout.c @@ -890,6 +890,7 @@ filelayout_pg_init_read(struct nfs_pageio_descriptor *pgio, 0, NFS4_MAX_UINT64, IOMODE_READ, + false, GFP_KERNEL); if (IS_ERR(pgio->pg_lseg)) { pgio->pg_error = PTR_ERR(pgio->pg_lseg); @@ -915,6 +916,7 @@ filelayout_pg_init_write(struct nfs_pageio_descriptor *pgio, 0, NFS4_MAX_UINT64, IOMODE_RW, + false, GFP_NOFS); if (IS_ERR(pgio->pg_lseg)) { pgio->pg_error = PTR_ERR(pgio->pg_lseg); diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c index a7aeb7407252..0e8018bc9880 100644 --- a/fs/nfs/flexfilelayout/flexfilelayout.c +++ b/fs/nfs/flexfilelayout/flexfilelayout.c @@ -820,6 +820,36 @@ ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg, return NULL; } +static void +ff_layout_pg_get_read(struct nfs_pageio_descriptor *pgio, + struct nfs_page *req, + bool strict_iomode) +{ +retry_strict: + pnfs_put_lseg(pgio->pg_lseg); + pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, + req->wb_context, + 0, + NFS4_MAX_UINT64, + IOMODE_READ, + strict_iomode, + GFP_KERNEL); + if (IS_ERR(pgio->pg_lseg)) { + pgio->pg_error = PTR_ERR(pgio->pg_lseg); + pgio->pg_lseg = NULL; + } + + /* If we don't have checking, do get a IOMODE_RW + * segment, and the server wants to avoid READs + * there, then retry! + */ + if (pgio->pg_lseg && !strict_iomode && + ff_layout_avoid_read_on_rw(pgio->pg_lseg)) { + strict_iomode = true; + goto retry_strict; + } +} + static void ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req) @@ -830,19 +860,10 @@ ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio, int ds_idx; /* Use full layout for now */ - if (!pgio->pg_lseg || ff_layout_avoid_read_on_rw(pgio->pg_lseg)) { - pnfs_put_lseg(pgio->pg_lseg); - pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode, - req->wb_context, - 0, - NFS4_MAX_UINT64, - IOMODE_READ, - GFP_KERNEL); - if (IS_ERR(pgio->pg_lseg)) { - pgio->pg_error = PTR_ERR(pgio->pg_lseg); - pgio->pg_lseg = NULL; - } - } + if (!pgio->pg_lseg) + ff_layout_pg_get_read(pgio, req, false); + else if (ff_layout_avoid_read_on_rw(pgio->pg_lseg)) + ff_layout_pg_get_read(pgio, req, true); /* If no lseg, fall back to read through mds */ if (pgio->pg_lseg == NULL) @@ -894,6 +915,7 @@ ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio, 0, NFS4_MAX_UINT64, IOMODE_RW, + false, GFP_NOFS); if (IS_ERR(pgio->pg_lseg)) { pgio->pg_error = PTR_ERR(pgio->pg_lseg); @@ -952,6 +974,7 @@ ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio, 0, NFS4_MAX_UINT64, IOMODE_RW, + false, GFP_NOFS); if (IS_ERR(pgio->pg_lseg)) { pgio->pg_error = PTR_ERR(pgio->pg_lseg); diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c index 79ae3049608d..0c7e0d45a4de 100644 --- a/fs/nfs/pnfs.c +++ b/fs/nfs/pnfs.c @@ -1321,23 +1321,28 @@ out_existing: /* * iomode matching rules: - * iomode lseg match - * ----- ----- ----- - * ANY READ true - * ANY RW true - * RW READ false - * RW RW true - * READ READ true - * READ RW true + * iomode lseg strict match + * iomode + * ----- ----- ------ ----- + * ANY READ N/A true + * ANY RW N/A true + * RW READ N/A false + * RW RW N/A true + * READ READ N/A true + * READ RW true false + * READ RW false true */ static bool pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, - const struct pnfs_layout_range *range) + const struct pnfs_layout_range *range, + bool strict_iomode) { struct pnfs_layout_range range1; if ((range->iomode == IOMODE_RW && ls_range->iomode != IOMODE_RW) || + (range->iomode != ls_range->iomode && + strict_iomode == true) || !pnfs_lseg_range_intersecting(ls_range, range)) return 0; @@ -1352,7 +1357,8 @@ pnfs_lseg_range_match(const struct pnfs_layout_range *ls_range, */ static struct pnfs_layout_segment * pnfs_find_lseg(struct pnfs_layout_hdr *lo, - struct pnfs_layout_range *range) + struct pnfs_layout_range *range, + bool strict_iomode) { struct pnfs_layout_segment *lseg, *ret = NULL; @@ -1361,7 +1367,8 @@ pnfs_find_lseg(struct pnfs_layout_hdr *lo, list_for_each_entry(lseg, &lo->plh_segs, pls_list) { if (test_bit(NFS_LSEG_VALID, &lseg->pls_flags) && !test_bit(NFS_LSEG_LAYOUTRETURN, &lseg->pls_flags) && - pnfs_lseg_range_match(&lseg->pls_range, range)) { + pnfs_lseg_range_match(&lseg->pls_range, range, + strict_iomode)) { ret = pnfs_get_lseg(lseg); break; } @@ -1478,6 +1485,7 @@ pnfs_update_layout(struct inode *ino, loff_t pos, u64 count, enum pnfs_iomode iomode, + bool strict_iomode, gfp_t gfp_flags) { struct pnfs_layout_range arg = { @@ -1539,7 +1547,7 @@ lookup_again: goto out_unlock; } - lseg = pnfs_find_lseg(lo, &arg); + lseg = pnfs_find_lseg(lo, &arg, strict_iomode); if (lseg) { trace_pnfs_update_layout(ino, pos, count, iomode, lo, lseg, PNFS_UPDATE_LAYOUT_FOUND_CACHED); @@ -1883,6 +1891,7 @@ pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *r req_offset(req), rd_size, IOMODE_READ, + false, GFP_KERNEL); if (IS_ERR(pgio->pg_lseg)) { pgio->pg_error = PTR_ERR(pgio->pg_lseg); @@ -1907,6 +1916,7 @@ pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, req_offset(req), wb_size, IOMODE_RW, + false, GFP_NOFS); if (IS_ERR(pgio->pg_lseg)) { pgio->pg_error = PTR_ERR(pgio->pg_lseg); diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h index f9f3331bef49..b21bd0bee784 100644 --- a/fs/nfs/pnfs.h +++ b/fs/nfs/pnfs.h @@ -288,6 +288,7 @@ struct pnfs_layout_segment *pnfs_update_layout(struct inode *ino, loff_t pos, u64 count, enum pnfs_iomode iomode, + bool strict_iomode, gfp_t gfp_flags); void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo); -- cgit v1.2.3-55-g7522 From 73e6c5d854d3f7f75e8b46d3e54aeb5d83fe6b1f Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sun, 26 Jun 2016 12:27:25 -0400 Subject: pNFS/files: Fix layoutcommit after a commit to DS According to the errata https://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751 we should always send layout commit after a commit to DS. Fixes: bc7d4b8fd091 ("nfs/filelayout: set layoutcommit...") Signed-off-by: Trond Myklebust --- fs/nfs/filelayout/filelayout.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'fs/nfs/filelayout/filelayout.c') diff --git a/fs/nfs/filelayout/filelayout.c b/fs/nfs/filelayout/filelayout.c index aa59757389dc..b4c1407e8fe4 100644 --- a/fs/nfs/filelayout/filelayout.c +++ b/fs/nfs/filelayout/filelayout.c @@ -375,8 +375,7 @@ static int filelayout_commit_done_cb(struct rpc_task *task, return -EAGAIN; } - if (data->verf.committed == NFS_UNSTABLE) - pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb); + pnfs_set_layoutcommit(data->inode, data->lseg, data->lwb); return 0; } -- cgit v1.2.3-55-g7522 From 2e18d4d822ea9cc811ea26a880cf2ed47cbf8889 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sun, 26 Jun 2016 18:54:58 -0400 Subject: pNFS: Files and flexfiles always need to commit before layoutcommit So ensure that we mark the layout for commit once the write is done, and then ensure that the commit to ds is finished before sending layoutcommit. Note that by doing this, we're able to optimise away the commit for the case of servers that don't need layoutcommit in order to return updated attributes. Signed-off-by: Trond Myklebust --- fs/nfs/filelayout/filelayout.c | 9 ++++++--- fs/nfs/flexfilelayout/flexfilelayout.c | 7 +++++-- fs/nfs/nfs4xdr.c | 11 ++++++++--- fs/nfs/pnfs.c | 5 ++++- fs/nfs/pnfs_nfs.c | 7 +++++++ 5 files changed, 30 insertions(+), 9 deletions(-) (limited to 'fs/nfs/filelayout/filelayout.c') diff --git a/fs/nfs/filelayout/filelayout.c b/fs/nfs/filelayout/filelayout.c index b4c1407e8fe4..25bd91a6e088 100644 --- a/fs/nfs/filelayout/filelayout.c +++ b/fs/nfs/filelayout/filelayout.c @@ -255,13 +255,16 @@ static int filelayout_read_done_cb(struct rpc_task *task, static void filelayout_set_layoutcommit(struct nfs_pgio_header *hdr) { + loff_t end_offs = 0; if (FILELAYOUT_LSEG(hdr->lseg)->commit_through_mds || - hdr->res.verf->committed != NFS_DATA_SYNC) + hdr->res.verf->committed == NFS_FILE_SYNC) return; + if (hdr->res.verf->committed == NFS_DATA_SYNC) + end_offs = hdr->mds_offset + (loff_t)hdr->res.count; - pnfs_set_layoutcommit(hdr->inode, hdr->lseg, - hdr->mds_offset + hdr->res.count); + /* Note: if the write is unstable, don't set end_offs until commit */ + pnfs_set_layoutcommit(hdr->inode, hdr->lseg, end_offs); dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino, (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb); } diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c index 14f2ed3f1a5b..e6206eaf2bdf 100644 --- a/fs/nfs/flexfilelayout/flexfilelayout.c +++ b/fs/nfs/flexfilelayout/flexfilelayout.c @@ -1470,6 +1470,7 @@ static void ff_layout_read_release(void *data) static int ff_layout_write_done_cb(struct rpc_task *task, struct nfs_pgio_header *hdr) { + loff_t end_offs = 0; int err; trace_nfs4_pnfs_write(hdr, task->tk_status); @@ -1495,8 +1496,10 @@ static int ff_layout_write_done_cb(struct rpc_task *task, if (hdr->res.verf->committed == NFS_FILE_SYNC || hdr->res.verf->committed == NFS_DATA_SYNC) - ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, - hdr->mds_offset + (loff_t)hdr->res.count); + end_offs = hdr->mds_offset + (loff_t)hdr->res.count; + + /* Note: if the write is unstable, don't set end_offs until commit */ + ff_layout_set_layoutcommit(hdr->inode, hdr->lseg, end_offs); /* zero out fattr since we don't care DS attr at all */ hdr->fattr.valid = 0; diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c index 661e753fe1c9..7bd3a5c09d31 100644 --- a/fs/nfs/nfs4xdr.c +++ b/fs/nfs/nfs4xdr.c @@ -1985,9 +1985,14 @@ encode_layoutcommit(struct xdr_stream *xdr, p = xdr_encode_hyper(p, args->lastbytewritten + 1); /* length */ *p = cpu_to_be32(0); /* reclaim */ encode_nfs4_stateid(xdr, &args->stateid); - p = reserve_space(xdr, 20); - *p++ = cpu_to_be32(1); /* newoffset = TRUE */ - p = xdr_encode_hyper(p, args->lastbytewritten); + if (args->lastbytewritten != U64_MAX) { + p = reserve_space(xdr, 20); + *p++ = cpu_to_be32(1); /* newoffset = TRUE */ + p = xdr_encode_hyper(p, args->lastbytewritten); + } else { + p = reserve_space(xdr, 12); + *p++ = cpu_to_be32(0); /* newoffset = FALSE */ + } *p++ = cpu_to_be32(0); /* Never send time_modify_changed */ *p++ = cpu_to_be32(NFS_SERVER(args->inode)->pnfs_curr_ld->id);/* type */ diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c index 0c7e0d45a4de..62553182514e 100644 --- a/fs/nfs/pnfs.c +++ b/fs/nfs/pnfs.c @@ -2378,7 +2378,10 @@ pnfs_layoutcommit_inode(struct inode *inode, bool sync) nfs_fattr_init(&data->fattr); data->args.bitmask = NFS_SERVER(inode)->cache_consistency_bitmask; data->res.fattr = &data->fattr; - data->args.lastbytewritten = end_pos - 1; + if (end_pos != 0) + data->args.lastbytewritten = end_pos - 1; + else + data->args.lastbytewritten = U64_MAX; data->res.server = NFS_SERVER(inode); if (ld->prepare_layoutcommit) { diff --git a/fs/nfs/pnfs_nfs.c b/fs/nfs/pnfs_nfs.c index 0dfc476da3e1..0d10cc280a23 100644 --- a/fs/nfs/pnfs_nfs.c +++ b/fs/nfs/pnfs_nfs.c @@ -932,6 +932,13 @@ EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit); int pnfs_nfs_generic_sync(struct inode *inode, bool datasync) { + int ret; + + if (!pnfs_layoutcommit_outstanding(inode)) + return 0; + ret = nfs_commit_inode(inode, FLUSH_SYNC); + if (ret < 0) + return ret; if (datasync) return 0; return pnfs_layoutcommit_inode(inode, true); -- cgit v1.2.3-55-g7522 From e033fb51ebb2983ee17b4a1b96ccbaedb137d9e9 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Thu, 21 Jul 2016 09:43:43 -0400 Subject: pNFS/files: filelayout_write_done_cb must call nfs_writeback_update_inode() All write callbacks are required to call nfs_writeback_update_inode() upon success to ensure that file size changes are recorded, and the attribute cache is invalidated. Signed-off-by: Trond Myklebust --- fs/nfs/filelayout/filelayout.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'fs/nfs/filelayout/filelayout.c') diff --git a/fs/nfs/filelayout/filelayout.c b/fs/nfs/filelayout/filelayout.c index 25bd91a6e088..a3fc48ba4931 100644 --- a/fs/nfs/filelayout/filelayout.c +++ b/fs/nfs/filelayout/filelayout.c @@ -357,6 +357,12 @@ static int filelayout_write_done_cb(struct rpc_task *task, } filelayout_set_layoutcommit(hdr); + + /* zero out the fattr */ + hdr->fattr.valid = 0; + if (task->tk_status >= 0) + nfs_writeback_update_inode(hdr); + return 0; } -- cgit v1.2.3-55-g7522