[V2 0/3] FF_FLAGS_NO_READ_IO

[V2 0/3] FF_FLAGS_NO_READ_IO

[Tom Haynes]

Hi Anna,

These are patched against your linux-next branch.

They are available as my linux-next branch at
git://git.linux-nfs.org/projects/loghyr/linux-nfs.git

Tom Haynes (3):
  nfs/flexfiles: Helper function to detect FF_FLAGS_NO_READ_IO
  nfs/flexfiles: Use the layout segment for reading unless it a
    IOMODE_RW and reading is disabled
  pnfs: pnfs_update_layout needs to consider if strict iomode checking
    is on

 fs/nfs/filelayout/filelayout.c            |  2 ++
 fs/nfs/flexfilelayout/flexfilelayout.c    | 50 +++++++++++++++++++++++--------
 fs/nfs/flexfilelayout/flexfilelayout.h    | 11 ++++++-
 fs/nfs/flexfilelayout/flexfilelayoutdev.c |  6 ++++
 fs/nfs/pnfs.c                             | 34 +++++++++++++--------
 fs/nfs/pnfs.h                             |  1 +
 6 files changed, 78 insertions(+), 26 deletions(-)

-- 
1.8.3.1

[V2 1/3] nfs/flexfiles: Helper function to detect FF_FLAGS_NO_READ_IO

[Tom Haynes]

The mds can inform the client not to use the IOMODE_RW layout
segment for doing READs. I.e., it is basically a
IOMODE_WRITE layout segment.

It would do this to not interfere with the WRITEs.

Signed-off-by: Tom Haynes <loghyr@primarydata.com>
---
 fs/nfs/flexfilelayout/flexfilelayout.h    | 11 ++++++++++-
 fs/nfs/flexfilelayout/flexfilelayoutdev.c |  6 ++++++
 2 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/fs/nfs/flexfilelayout/flexfilelayout.h b/fs/nfs/flexfilelayout/flexfilelayout.h
index b540581..1bcdb15 100644
--- a/fs/nfs/flexfilelayout/flexfilelayout.h
+++ b/fs/nfs/flexfilelayout/flexfilelayout.h
@@ -10,7 +10,8 @@
 #define FS_NFS_NFS4FLEXFILELAYOUT_H
 
 #define FF_FLAGS_NO_LAYOUTCOMMIT 1
-#define FF_FLAGS_NO_IO_THRU_MDS 2
+#define FF_FLAGS_NO_IO_THRU_MDS  2
+#define FF_FLAGS_NO_READ_IO      4
 
 #include "../pnfs.h"
 
@@ -153,6 +154,12 @@ ff_layout_no_fallback_to_mds(struct pnfs_layout_segment *lseg)
 }
 
 static inline bool
+ff_layout_no_read_on_rw(struct pnfs_layout_segment *lseg)
+{
+	return FF_LAYOUT_LSEG(lseg)->flags & FF_FLAGS_NO_READ_IO;
+}
+
+static inline bool
 ff_layout_test_devid_unavailable(struct nfs4_deviceid_node *node)
 {
 	return nfs4_test_deviceid_unavailable(node);
@@ -192,4 +199,6 @@ struct rpc_cred *ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg,
 				       u32 ds_idx, struct rpc_cred *mdscred);
 bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg);
 bool ff_layout_avoid_mds_available_ds(struct pnfs_layout_segment *lseg);
+bool ff_layout_avoid_read_on_rw(struct pnfs_layout_segment *lseg);
+
 #endif /* FS_NFS_NFS4FLEXFILELAYOUT_H */
diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
index 35d84d0..0aa36be 100644
--- a/fs/nfs/flexfilelayout/flexfilelayoutdev.c
+++ b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
@@ -557,6 +557,12 @@ bool ff_layout_avoid_mds_available_ds(struct pnfs_layout_segment *lseg)
 	       ff_layout_has_available_ds(lseg);
 }
 
+bool ff_layout_avoid_read_on_rw(struct pnfs_layout_segment *lseg)
+{
+	return lseg->pls_range.iomode == IOMODE_RW &&
+	       ff_layout_no_read_on_rw(lseg);
+}
+
 module_param(dataserver_retrans, uint, 0644);
 MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
 			"retries a request before it attempts further "
-- 
1.8.3.1

[V2 2/3] nfs/flexfiles: Use the layout segment for reading unless it a IOMODE_RW and reading is disabled

[Tom Haynes]

Signed-off-by: Tom Haynes <loghyr@primarydata.com>
---
 fs/nfs/flexfilelayout/flexfilelayout.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c
index 40bccb2..a7aeb74 100644
--- a/fs/nfs/flexfilelayout/flexfilelayout.c
+++ b/fs/nfs/flexfilelayout/flexfilelayout.c
@@ -830,7 +830,8 @@ ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 	int ds_idx;
 
 	/* Use full layout for now */
-	if (!pgio->pg_lseg) {
+	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,
@@ -840,9 +841,9 @@ ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
 		if (IS_ERR(pgio->pg_lseg)) {
 			pgio->pg_error = PTR_ERR(pgio->pg_lseg);
 			pgio->pg_lseg = NULL;
-			return;
 		}
 	}
+
 	/* If no lseg, fall back to read through mds */
 	if (pgio->pg_lseg == NULL)
 		goto out_mds;
-- 
1.8.3.1

[V2 3/3] pnfs: pnfs_update_layout needs to consider if strict iomode checking is on

[Tom Haynes]

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 <loghyr@primarydata.com>
---
 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(-)

diff --git a/fs/nfs/filelayout/filelayout.c b/fs/nfs/filelayout/filelayout.c
index 3e50057..aa59757 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 a7aeb74..0e8018b 100644
--- a/fs/nfs/flexfilelayout/flexfilelayout.c
+++ b/fs/nfs/flexfilelayout/flexfilelayout.c
@@ -821,6 +821,36 @@ ff_layout_choose_best_ds_for_read(struct pnfs_layout_segment *lseg,
 }
 
 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 79ae304..0c7e0d4 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 f9f3331..b21bd0b 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);
 
-- 
1.8.3.1