[PATCH 1/4] NFSv4.1/pnfs Improve the packing of struct pnfs_layout_hdr

[PATCH 1/4] NFSv4.1/pnfs Improve the packing of struct pnfs_layout_hdr

[Trond Myklebust]

Eliminate a couple of holes in the structure, and move the 2 atomics
into the same cacheline.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index d3979dd1037a..4df87ef3dccc 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -183,15 +183,15 @@ struct pnfs_layoutdriver_type {
 
 struct pnfs_layout_hdr {
 	atomic_t		plh_refcount;
+	atomic_t		plh_outstanding; /* number of RPCs out */
 	struct list_head	plh_layouts;   /* other client layouts */
 	struct list_head	plh_bulk_destroy;
 	struct list_head	plh_segs;      /* layout segments list */
-	nfs4_stateid		plh_stateid;
-	atomic_t		plh_outstanding; /* number of RPCs out */
 	unsigned long		plh_block_lgets; /* block LAYOUTGET if >0 */
-	u32			plh_barrier; /* ignore lower seqids */
 	unsigned long		plh_retry_timestamp;
 	unsigned long		plh_flags;
+	nfs4_stateid		plh_stateid;
+	u32			plh_barrier; /* ignore lower seqids */
 	enum pnfs_iomode	plh_return_iomode;
 	loff_t			plh_lwb; /* last write byte for layoutcommit */
 	struct rpc_cred		*plh_lc_cred; /* layoutcommit cred */
-- 
2.4.3

[PATCH 2/4] NFSv4.1/pnfs: Add sanity check for the layout range returned by the server

[Trond Myklebust]

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 3530bb703214..68cc4b169769 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1561,6 +1561,26 @@ out_unlock:
 }
 EXPORT_SYMBOL_GPL(pnfs_update_layout);
 
+static bool
+pnfs_sanity_check_layout_range(struct pnfs_layout_range *range)
+{
+	switch (range->iomode) {
+	case IOMODE_READ:
+	case IOMODE_RW:
+		break;
+	default:
+		return false;
+	}
+	if (range->offset == NFS4_MAX_UINT64)
+		return false;
+	if (range->length == 0)
+		return false;
+	if (range->length != NFS4_MAX_UINT64 &&
+	    range->length > NFS4_MAX_UINT64 - range->offset)
+		return false;
+	return true;
+}
+
 struct pnfs_layout_segment *
 pnfs_layout_process(struct nfs4_layoutget *lgp)
 {
@@ -1569,7 +1589,10 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
 	struct pnfs_layout_segment *lseg;
 	struct inode *ino = lo->plh_inode;
 	LIST_HEAD(free_me);
-	int status = 0;
+	int status = -EINVAL;
+
+	if (!pnfs_sanity_check_layout_range(&res->range))
+		goto out;
 
 	/* Inject layout blob into I/O device driver */
 	lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res, lgp->gfp_flags);
-- 
2.4.3

[PATCH 3/4] NFSv4.1/pnfs: Allow pNFS device drivers to customise layout segment insertion

[Trond Myklebust]

This is needed in order to allow merging of contiguous layout segments,
and also to correct the ordering of layouts for those device drivers that
don't necessarily want to place the read-write layouts first.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 59 ++++++++++++++++++++++++++++++++++++++++++++++++++---------
 fs/nfs/pnfs.h | 11 +++++++++++
 2 files changed, 61 insertions(+), 9 deletions(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 68cc4b169769..914c1daf08df 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1188,16 +1188,41 @@ pnfs_lseg_range_cmp(const struct pnfs_layout_range *l1,
 	return (int)(l1->iomode == IOMODE_READ) - (int)(l2->iomode == IOMODE_READ);
 }
 
-static void
-pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
-		   struct pnfs_layout_segment *lseg)
+static bool
+pnfs_lseg_range_is_after(const struct pnfs_layout_range *l1,
+		const struct pnfs_layout_range *l2)
+{
+	return pnfs_lseg_range_cmp(l1, l2) > 0;
+}
+
+static bool
+pnfs_lseg_no_merge(struct pnfs_layout_segment *lseg,
+		struct pnfs_layout_segment *old)
+{
+	return false;
+}
+
+void
+pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+		   struct pnfs_layout_segment *lseg,
+		   bool (*is_after)(const struct pnfs_layout_range *,
+			   const struct pnfs_layout_range *),
+		   bool (*do_merge)(struct pnfs_layout_segment *,
+			   struct pnfs_layout_segment *),
+		   struct list_head *free_me)
 {
-	struct pnfs_layout_segment *lp;
+	struct pnfs_layout_segment *lp, *tmp;
 
 	dprintk("%s:Begin\n", __func__);
 
-	list_for_each_entry(lp, &lo->plh_segs, pls_list) {
-		if (pnfs_lseg_range_cmp(&lseg->pls_range, &lp->pls_range) > 0)
+	list_for_each_entry_safe(lp, tmp, &lo->plh_segs, pls_list) {
+		if (test_bit(NFS_LSEG_VALID, &lp->pls_flags) == 0)
+			continue;
+		if (do_merge(lseg, lp)) {
+			mark_lseg_invalid(lp, free_me);
+			continue;
+		}
+		if (is_after(&lseg->pls_range, &lp->pls_range))
 			continue;
 		list_add_tail(&lseg->pls_list, &lp->pls_list);
 		dprintk("%s: inserted lseg %p "
@@ -1219,6 +1244,24 @@ out:
 
 	dprintk("%s:Return\n", __func__);
 }
+EXPORT_SYMBOL_GPL(pnfs_generic_layout_insert_lseg);
+
+static void
+pnfs_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+		   struct pnfs_layout_segment *lseg,
+		   struct list_head *free_me)
+{
+	struct inode *inode = lo->plh_inode;
+	struct pnfs_layoutdriver_type *ld = NFS_SERVER(inode)->pnfs_curr_ld;
+
+	if (ld->add_lseg != NULL)
+		ld->add_lseg(lo, lseg, free_me);
+	else
+		pnfs_generic_layout_insert_lseg(lo, lseg,
+				pnfs_lseg_range_is_after,
+				pnfs_lseg_no_merge,
+				free_me);
+}
 
 static struct pnfs_layout_hdr *
 alloc_init_layout_hdr(struct inode *ino,
@@ -1311,8 +1354,6 @@ pnfs_find_lseg(struct pnfs_layout_hdr *lo,
 			ret = pnfs_get_lseg(lseg);
 			break;
 		}
-		if (lseg->pls_range.offset > range->offset)
-			break;
 	}
 
 	dprintk("%s:Return lseg %p ref %d\n",
@@ -1637,7 +1678,7 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
 	clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
 
 	pnfs_get_lseg(lseg);
-	pnfs_layout_insert_lseg(lo, lseg);
+	pnfs_layout_insert_lseg(lo, lseg, &free_me);
 
 	if (res->return_on_close)
 		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index 4df87ef3dccc..869069d8b996 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -128,6 +128,9 @@ struct pnfs_layoutdriver_type {
 
 	struct pnfs_layout_segment * (*alloc_lseg) (struct pnfs_layout_hdr *layoutid, struct nfs4_layoutget_res *lgr, gfp_t gfp_flags);
 	void (*free_lseg) (struct pnfs_layout_segment *lseg);
+	void (*add_lseg) (struct pnfs_layout_hdr *layoutid,
+			struct pnfs_layout_segment *lseg,
+			struct list_head *free_me);
 
 	void (*return_range) (struct pnfs_layout_hdr *lo,
 			      struct pnfs_layout_range *range);
@@ -285,6 +288,14 @@ struct pnfs_layout_segment *pnfs_update_layout(struct inode *ino,
 					       gfp_t gfp_flags);
 void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo);
 
+void pnfs_generic_layout_insert_lseg(struct pnfs_layout_hdr *lo,
+		   struct pnfs_layout_segment *lseg,
+		   bool (*is_after)(const struct pnfs_layout_range *lseg_range,
+			   const struct pnfs_layout_range *old),
+		   bool (*do_merge)(struct pnfs_layout_segment *lseg,
+			   struct pnfs_layout_segment *old),
+		   struct list_head *free_me);
+
 void nfs4_deviceid_mark_client_invalid(struct nfs_client *clp);
 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *);
 int pnfs_write_done_resend_to_mds(struct nfs_pgio_header *);
-- 
2.4.3

[PATCH 4/4] NFSv4.1/flexfiles: Allow coalescing of new layout segments and existing ones

[Trond Myklebust]

In order to ensure atomicity of updates, we merge the old layout segments
into the new ones, and then invalidate the old ones.

Also ensure that we order the list of layout segments so that
RO segments are preferred over RW.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/flexfilelayout/flexfilelayout.c | 60 ++++++++++++++++++++++++++++++++++
 fs/nfs/pnfs.h                          | 16 +++++++++
 2 files changed, 76 insertions(+)

diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c
index 7fefa8ad9578..4ec624cfcf8b 100644
--- a/fs/nfs/flexfilelayout/flexfilelayout.c
+++ b/fs/nfs/flexfilelayout/flexfilelayout.c
@@ -265,6 +265,65 @@ static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
 	}
 }
 
+static bool
+ff_lseg_range_is_after(const struct pnfs_layout_range *l1,
+		const struct pnfs_layout_range *l2)
+{
+	u64 end1, end2;
+
+	if (l1->iomode != l2->iomode)
+		return l1->iomode != IOMODE_READ;
+	end1 = pnfs_calc_offset_end(l1->offset, l1->length);
+	end2 = pnfs_calc_offset_end(l2->offset, l2->length);
+	if (end1 < l2->offset)
+		return false;
+	if (end2 < l1->offset)
+		return true;
+	return l2->offset <= l1->offset;
+}
+
+static bool
+ff_lseg_merge(struct pnfs_layout_segment *new,
+		struct pnfs_layout_segment *old)
+{
+	u64 new_end, old_end;
+
+	if (new->pls_range.iomode != old->pls_range.iomode)
+		return false;
+	old_end = pnfs_calc_offset_end(old->pls_range.offset,
+			old->pls_range.length);
+	if (old_end < new->pls_range.offset)
+		return false;
+	new_end = pnfs_calc_offset_end(new->pls_range.offset,
+			new->pls_range.length);
+	if (new_end < old->pls_range.offset)
+		return false;
+
+	/* Mergeable: copy info from 'old' to 'new' */
+	if (new_end < old_end)
+		new_end = old_end;
+	if (new->pls_range.offset < old->pls_range.offset)
+		new->pls_range.offset = old->pls_range.offset;
+	new->pls_range.length = pnfs_calc_offset_length(new->pls_range.offset,
+			new_end);
+	if (test_bit(NFS_LSEG_ROC, &old->pls_flags))
+		set_bit(NFS_LSEG_ROC, &new->pls_flags);
+	if (test_bit(NFS_LSEG_LAYOUTRETURN, &old->pls_flags))
+		set_bit(NFS_LSEG_LAYOUTRETURN, &new->pls_flags);
+	return true;
+}
+
+static void
+ff_layout_add_lseg(struct pnfs_layout_hdr *lo,
+		struct pnfs_layout_segment *lseg,
+		struct list_head *free_me)
+{
+	pnfs_generic_layout_insert_lseg(lo, lseg,
+			ff_lseg_range_is_after,
+			ff_lseg_merge,
+			free_me);
+}
+
 static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
 {
 	int i, j;
@@ -2046,6 +2105,7 @@ static struct pnfs_layoutdriver_type flexfilelayout_type = {
 	.free_layout_hdr	= ff_layout_free_layout_hdr,
 	.alloc_lseg		= ff_layout_alloc_lseg,
 	.free_lseg		= ff_layout_free_lseg,
+	.add_lseg		= ff_layout_add_lseg,
 	.pg_read_ops		= &ff_layout_pg_read_ops,
 	.pg_write_ops		= &ff_layout_pg_write_ops,
 	.get_ds_info		= ff_layout_get_ds_info,
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index 869069d8b996..78c9351ff117 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -539,6 +539,22 @@ pnfs_use_threshold(struct nfs4_threshold **dst, struct nfs4_threshold *src,
 					nfss->pnfs_curr_ld->id == src->l_type);
 }
 
+static inline u64
+pnfs_calc_offset_end(u64 offset, u64 len)
+{
+	if (len == NFS4_MAX_UINT64 || len >= NFS4_MAX_UINT64 - offset)
+		return NFS4_MAX_UINT64;
+	return offset + len - 1;
+}
+
+static inline u64
+pnfs_calc_offset_length(u64 offset, u64 end)
+{
+	if (end == NFS4_MAX_UINT64 || end <= offset)
+		return NFS4_MAX_UINT64;
+	return 1 + end - offset;
+}
+
 extern unsigned int layoutstats_timer;
 
 #ifdef NFS_DEBUG
-- 
2.4.3