[PATCH v3 00/14] More layoutreturn nits

[PATCH v3 00/14] More layoutreturn nits

[Trond Myklebust]

Various non-critical bugfixes and cleanups targeting layoutreturn

v2: Add  CB_LAYOUTRECALL stateid verification fixups
v3: Add more layout stateid update and validity checks
    Fix layout barrier seqid updating
    Reordered for better bisectability
    Further cleanups


Trond Myklebust (14):
  pNFS: LAYOUTRETURN should only update the stateid if the layout is
    valid
  pNFS: Clear the layout return tracking on layout reinitialisation
  pNFS: Always update the layout stateid if NFS_LAYOUT_INVALID_STID is
    set
  pNFS: Always update the layout barrier seqid on LAYOUTGET
  pNFS: Fix CB_LAYOUTRECALL stateid verification
  pNFS: Ensure layoutreturn acts as a completion for layout callbacks
  pNFS: Do not set plh_return_seq for non-callback related layoutreturns
  NFS: pnfs_mark_matching_lsegs_return() should match the layout
    sequence id
  pNFS: Cleanup - don't open code pnfs_mark_layout_stateid_invalid()
  pNFS: Clear the layout metadata if the server changed the layout
    stateid
  pNFS: Remove redundant pnfs_mark_layout_returned_if_empty()
  pNFS: Remove redundant stateid invalidation
  pNFS: Cleanup - do layout segment initialisation in one place
  pNFS: Remove redundant smp_mb() from pnfs_init_lseg()

 fs/nfs/callback_proc.c |  64 ++++++++++++++-------
 fs/nfs/nfs42proc.c     |   3 +-
 fs/nfs/nfs4proc.c      |   6 +-
 fs/nfs/pnfs.c          | 151 +++++++++++++++++++++++++++++--------------------
 fs/nfs/pnfs.h          |  20 +++----
 5 files changed, 143 insertions(+), 101 deletions(-)

-- 
2.7.4

[PATCH v3 01/14] pNFS: LAYOUTRETURN should only update the stateid if the layout is valid

[Trond Myklebust]

If the layout was completely returned, then ignore the returned layout
stateid.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/nfs4proc.c | 2 +-
 fs/nfs/pnfs.h     | 5 +++++
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index ff416d0e24bc..dc50ba4d84ad 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -8130,7 +8130,7 @@ static void nfs4_layoutreturn_release(void *calldata)
 	pnfs_mark_matching_lsegs_invalid(lo, &freeme, &lrp->args.range,
 			be32_to_cpu(lrp->args.stateid.seqid));
 	pnfs_mark_layout_returned_if_empty(lo);
-	if (lrp->res.lrs_present)
+	if (lrp->res.lrs_present && pnfs_layout_is_valid(lo))
 		pnfs_set_layout_stateid(lo, &lrp->res.stateid, true);
 	pnfs_clear_layoutreturn_waitbit(lo);
 	spin_unlock(&lo->plh_inode->i_lock);
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index b21bd0bee784..2f4f26905c03 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -375,6 +375,11 @@ static inline bool nfs_have_layout(struct inode *inode)
 	return NFS_I(inode)->layout != NULL;
 }
 
+static inline bool pnfs_layout_is_valid(const struct pnfs_layout_hdr *lo)
+{
+	return test_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags) == 0;
+}
+
 static inline struct nfs4_deviceid_node *
 nfs4_get_deviceid(struct nfs4_deviceid_node *d)
 {
-- 
2.7.4

[PATCH v3 02/14] pNFS: Clear the layout return tracking on layout reinitialisation

[Trond Myklebust]

Ensure that we don't carry over layoutreturn info from a previous
incarnation of this layout.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 19 ++++++++++++++-----
 1 file changed, 14 insertions(+), 5 deletions(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 0fbe734cc38c..dcccded1aeed 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -873,15 +873,21 @@ void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
 	rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
 }
 
+static void
+pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
+{
+	lo->plh_return_iomode = 0;
+	lo->plh_return_seq = 0;
+	clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
+}
+
 static bool
 pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo)
 {
 	if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
 		return false;
-	lo->plh_return_iomode = 0;
-	lo->plh_return_seq = 0;
 	pnfs_get_layout_hdr(lo);
-	clear_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
+	pnfs_clear_layoutreturn_info(lo);
 	return true;
 }
 
@@ -1764,10 +1770,13 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
 		lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
 	}
 
-	clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
-
 	pnfs_get_lseg(lseg);
 	pnfs_layout_insert_lseg(lo, lseg, &free_me);
+	if (!pnfs_layout_is_valid(lo)) {
+		pnfs_clear_layoutreturn_info(lo);
+		clear_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
+	}
+
 
 	if (res->return_on_close)
 		set_bit(NFS_LSEG_ROC, &lseg->pls_flags);
-- 
2.7.4

[PATCH v3 03/14] pNFS: Always update the layout stateid if NFS_LAYOUT_INVALID_STID is set

[Trond Myklebust]

If the layout stateid is invalid, then pnfs_set_layout_stateid() must
always initialise it.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index dcccded1aeed..eef844785bc6 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -762,7 +762,7 @@ pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
 			bool update_barrier)
 {
 	u32 oldseq, newseq, new_barrier;
-	int empty = list_empty(&lo->plh_segs);
+	bool empty = !pnfs_layout_is_valid(lo);
 
 	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
 	newseq = be32_to_cpu(new->seqid);
-- 
2.7.4

[PATCH v3 04/14] pNFS: Always update the layout barrier seqid on LAYOUTGET

[Trond Myklebust]

Currently, pnfs_set_layout_stateid() will update the layout sequence
id barrier only if the stateid itself is newer than the current
layout stateid. However in a situation where multiple LAYOUTGET calls
and a LAYOUTRETURN raced, it is entirely possible for one of the
LAYOUTGET to set the current stateid to something newer than the
LAYOUTRETURN that needs to set the barrier.

The fix is to allow the "update_barrier" flag to force a check as to
whether or not the barrier needs to be updated.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 27 ++++++++++++++-------------
 1 file changed, 14 insertions(+), 13 deletions(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index eef844785bc6..85c3e7b47ddb 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -761,24 +761,25 @@ void
 pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo, const nfs4_stateid *new,
 			bool update_barrier)
 {
-	u32 oldseq, newseq, new_barrier;
-	bool empty = !pnfs_layout_is_valid(lo);
+	u32 oldseq, newseq, new_barrier = 0;
+	bool invalid = !pnfs_layout_is_valid(lo);
 
 	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
 	newseq = be32_to_cpu(new->seqid);
-	if (empty || pnfs_seqid_is_newer(newseq, oldseq)) {
+	if (invalid || pnfs_seqid_is_newer(newseq, oldseq)) {
 		nfs4_stateid_copy(&lo->plh_stateid, new);
-		if (update_barrier) {
-			new_barrier = be32_to_cpu(new->seqid);
-		} else {
-			/* Because of wraparound, we want to keep the barrier
-			 * "close" to the current seqids.
-			 */
-			new_barrier = newseq - atomic_read(&lo->plh_outstanding);
-		}
-		if (empty || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
-			lo->plh_barrier = new_barrier;
+		/*
+		 * Because of wraparound, we want to keep the barrier
+		 * "close" to the current seqids.
+		 */
+		new_barrier = newseq - atomic_read(&lo->plh_outstanding);
 	}
+	if (update_barrier)
+		new_barrier = be32_to_cpu(new->seqid);
+	else if (new_barrier == 0)
+		return;
+	if (invalid || pnfs_seqid_is_newer(new_barrier, lo->plh_barrier))
+		lo->plh_barrier = new_barrier;
 }
 
 static bool
-- 
2.7.4

[PATCH v3 05/14] pNFS: Fix CB_LAYOUTRECALL stateid verification

[Trond Myklebust]

We want to evaluate in this order:

If the client holds no layout for this inode, then return
NFS4ERR_NOMATCHING_LAYOUT; it probably forgot the layout.

If the client finds the inode among the list of layouts, but the corresponding
stateid has not yet been initialised, then return NFS4ERR_DELAY to ask the
server to retry once the outstanding LAYOUTGET is complete.

If the current layout stateid's "other" field does not match the recalled
stateid, return NFS4ERR_BAD_STATEID.

If already processing a layout recall with a newer stateid, return
NFS4ERR_OLD_STATEID. This can only happens for servers that are
non-compliant with the NFSv4.1 protocol.

If already processing a layout recall with an older stateid, return
NFS4ERR_DELAY to ask the server to retry once the outstanding
LAYOUTRETURN is complete. Again, this is technically incompliant with
the NFSv4.1 protocol.

If the current layout sequence id is newer than the recalled stateid's
sequence id, return NFS4ERR_OLD_STATEID. This too implies protocol
non-compliance.

If the current layout sequence id is older than the recalled stateid's
sequence id+1, return NFS4ERR_DELAY.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/callback_proc.c | 63 +++++++++++++++++++++++++++++++++++---------------
 1 file changed, 44 insertions(+), 19 deletions(-)

diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index aaa2e8d3df6f..837da8a02d35 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -119,27 +119,30 @@ out:
  * hashed by filehandle.
  */
 static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp,
-		struct nfs_fh *fh, nfs4_stateid *stateid)
+		struct nfs_fh *fh)
 {
 	struct nfs_server *server;
+	struct nfs_inode *nfsi;
 	struct inode *ino;
 	struct pnfs_layout_hdr *lo;
 
+restart:
 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
 		list_for_each_entry(lo, &server->layouts, plh_layouts) {
-			if (!nfs4_stateid_match_other(&lo->plh_stateid, stateid))
+			nfsi = NFS_I(lo->plh_inode);
+			if (nfs_compare_fh(fh, &nfsi->fh))
 				continue;
-			if (nfs_compare_fh(fh, &NFS_I(lo->plh_inode)->fh))
+			if (nfsi->layout != lo)
 				continue;
 			ino = igrab(lo->plh_inode);
 			if (!ino)
 				break;
 			spin_lock(&ino->i_lock);
 			/* Is this layout in the process of being freed? */
-			if (NFS_I(ino)->layout != lo) {
+			if (nfsi->layout != lo) {
 				spin_unlock(&ino->i_lock);
 				iput(ino);
-				break;
+				goto restart;
 			}
 			pnfs_get_layout_hdr(lo);
 			spin_unlock(&ino->i_lock);
@@ -151,13 +154,13 @@ static struct pnfs_layout_hdr * get_layout_by_fh_locked(struct nfs_client *clp,
 }
 
 static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp,
-		struct nfs_fh *fh, nfs4_stateid *stateid)
+		struct nfs_fh *fh)
 {
 	struct pnfs_layout_hdr *lo;
 
 	spin_lock(&clp->cl_lock);
 	rcu_read_lock();
-	lo = get_layout_by_fh_locked(clp, fh, stateid);
+	lo = get_layout_by_fh_locked(clp, fh);
 	rcu_read_unlock();
 	spin_unlock(&clp->cl_lock);
 
@@ -167,17 +170,39 @@ static struct pnfs_layout_hdr * get_layout_by_fh(struct nfs_client *clp,
 /*
  * Enforce RFC5661 section 12.5.5.2.1. (Layout Recall and Return Sequencing)
  */
-static bool pnfs_check_stateid_sequence(struct pnfs_layout_hdr *lo,
+static u32 pnfs_check_callback_stateid(struct pnfs_layout_hdr *lo,
 					const nfs4_stateid *new)
 {
 	u32 oldseq, newseq;
 
-	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
+	/* Is the stateid still not initialised? */
+	if (!pnfs_layout_is_valid(lo))
+		return NFS4ERR_DELAY;
+
+	/* Mismatched stateid? */
+	if (!nfs4_stateid_match_other(&lo->plh_stateid, new))
+		return NFS4ERR_BAD_STATEID;
+
 	newseq = be32_to_cpu(new->seqid);
+	/* Are we already in a layout recall situation? */
+	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags) &&
+	    lo->plh_return_seq != 0) {
+		if (newseq < lo->plh_return_seq)
+			return NFS4ERR_OLD_STATEID;
+		if (newseq > lo->plh_return_seq)
+			return NFS4ERR_DELAY;
+		goto out;
+	}
 
+	/* Check that the stateid matches what we think it should be. */
+	oldseq = be32_to_cpu(lo->plh_stateid.seqid);
 	if (newseq > oldseq + 1)
-		return false;
-	return true;
+		return NFS4ERR_DELAY;
+	/* Crazy server! */
+	if (newseq <= oldseq)
+		return NFS4ERR_OLD_STATEID;
+out:
+	return NFS_OK;
 }
 
 static u32 initiate_file_draining(struct nfs_client *clp,
@@ -188,7 +213,7 @@ static u32 initiate_file_draining(struct nfs_client *clp,
 	u32 rv = NFS4ERR_NOMATCHING_LAYOUT;
 	LIST_HEAD(free_me_list);
 
-	lo = get_layout_by_fh(clp, &args->cbl_fh, &args->cbl_stateid);
+	lo = get_layout_by_fh(clp, &args->cbl_fh);
 	if (!lo) {
 		trace_nfs4_cb_layoutrecall_file(clp, &args->cbl_fh, NULL,
 				&args->cbl_stateid, -rv);
@@ -196,18 +221,15 @@ static u32 initiate_file_draining(struct nfs_client *clp,
 	}
 
 	ino = lo->plh_inode;
+	pnfs_layoutcommit_inode(ino, false);
+
 
 	spin_lock(&ino->i_lock);
-	if (!pnfs_check_stateid_sequence(lo, &args->cbl_stateid)) {
-		rv = NFS4ERR_DELAY;
+	rv = pnfs_check_callback_stateid(lo, &args->cbl_stateid);
+	if (rv != NFS_OK)
 		goto unlock;
-	}
 	pnfs_set_layout_stateid(lo, &args->cbl_stateid, true);
-	spin_unlock(&ino->i_lock);
-
-	pnfs_layoutcommit_inode(ino, false);
 
-	spin_lock(&ino->i_lock);
 	/*
 	 * Enforce RFC5661 Section 12.5.5.2.1.5 (Bulk Recall and Return)
 	 */
@@ -223,6 +245,9 @@ static u32 initiate_file_draining(struct nfs_client *clp,
 		goto unlock;
 	}
 
+	/* Embrace your forgetfulness! */
+	rv = NFS4ERR_NOMATCHING_LAYOUT;
+
 	if (NFS_SERVER(ino)->pnfs_curr_ld->return_range) {
 		NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
 			&args->cbl_range);
-- 
2.7.4

[PATCH v3 06/14] pNFS: Ensure layoutreturn acts as a completion for layout callbacks

[Trond Myklebust]

When we return NFS_OK to the CB_LAYOUTRECALL, we are required to
send a layoutreturn that "completes" that layout recall request, using
the correct stateid.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 40 +++++++++++++++++++++++++---------------
 1 file changed, 25 insertions(+), 15 deletions(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 85c3e7b47ddb..878dc4b7085a 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -883,12 +883,28 @@ pnfs_clear_layoutreturn_info(struct pnfs_layout_hdr *lo)
 }
 
 static bool
-pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo)
+pnfs_prepare_layoutreturn(struct pnfs_layout_hdr *lo,
+		nfs4_stateid *stateid,
+		enum pnfs_iomode *iomode)
 {
 	if (test_and_set_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
 		return false;
 	pnfs_get_layout_hdr(lo);
-	pnfs_clear_layoutreturn_info(lo);
+	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags)) {
+		if (stateid != NULL) {
+			nfs4_stateid_copy(stateid, &lo->plh_stateid);
+			if (lo->plh_return_seq != 0)
+				stateid->seqid = cpu_to_be32(lo->plh_return_seq);
+		}
+		if (iomode != NULL)
+			*iomode = lo->plh_return_iomode;
+		pnfs_clear_layoutreturn_info(lo);
+		return true;
+	}
+	if (stateid != NULL)
+		nfs4_stateid_copy(stateid, &lo->plh_stateid);
+	if (iomode != NULL)
+		*iomode = IOMODE_ANY;
 	return true;
 }
 
@@ -956,10 +972,7 @@ static void pnfs_layoutreturn_before_put_layout_hdr(struct pnfs_layout_hdr *lo)
 		enum pnfs_iomode iomode;
 		bool send;
 
-		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
-		stateid.seqid = cpu_to_be32(lo->plh_return_seq);
-		iomode = lo->plh_return_iomode;
-		send = pnfs_prepare_layoutreturn(lo);
+		send = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
 		spin_unlock(&inode->i_lock);
 		if (send) {
 			/* Send an async layoutreturn so we dont deadlock */
@@ -996,7 +1009,6 @@ _pnfs_return_layout(struct inode *ino)
 		dprintk("NFS: %s no layout to return\n", __func__);
 		goto out;
 	}
-	nfs4_stateid_copy(&stateid, &nfsi->layout->plh_stateid);
 	/* Reference matched in nfs4_layoutreturn_release */
 	pnfs_get_layout_hdr(lo);
 	empty = list_empty(&lo->plh_segs);
@@ -1020,7 +1032,7 @@ _pnfs_return_layout(struct inode *ino)
 	}
 
 	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
-	send = pnfs_prepare_layoutreturn(lo);
+	send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
 	spin_unlock(&ino->i_lock);
 	pnfs_free_lseg_list(&tmp_list);
 	if (send)
@@ -1087,11 +1099,10 @@ bool pnfs_roc(struct inode *ino)
 			goto out_noroc;
 	}
 
-	nfs4_stateid_copy(&stateid, &lo->plh_stateid);
 	/* always send layoutreturn if being marked so */
-	if (test_and_clear_bit(NFS_LAYOUT_RETURN_REQUESTED,
-				   &lo->plh_flags))
-		layoutreturn = pnfs_prepare_layoutreturn(lo);
+	if (test_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags))
+		layoutreturn = pnfs_prepare_layoutreturn(lo,
+				&stateid, NULL);
 
 	list_for_each_entry_safe(lseg, tmp, &lo->plh_segs, pls_list)
 		/* If we are sending layoutreturn, invalidate all valid lsegs */
@@ -1874,10 +1885,9 @@ void pnfs_error_mark_layout_for_return(struct inode *inode,
 	if (!pnfs_mark_matching_lsegs_return(lo, &free_me,
 						&range, lseg->pls_seq)) {
 		nfs4_stateid stateid;
-		enum pnfs_iomode iomode = lo->plh_return_iomode;
+		enum pnfs_iomode iomode;
 
-		nfs4_stateid_copy(&stateid, &lo->plh_stateid);
-		return_now = pnfs_prepare_layoutreturn(lo);
+		return_now = pnfs_prepare_layoutreturn(lo, &stateid, &iomode);
 		spin_unlock(&inode->i_lock);
 		if (return_now)
 			pnfs_send_layoutreturn(lo, &stateid, iomode, false);
-- 
2.7.4

[PATCH v3 07/14] pNFS: Do not set plh_return_seq for non-callback related layoutreturns

[Trond Myklebust]

In cases where we need to send a layoutreturn in order to propagate
an error, we should not tie that to a specific layout stateid.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 878dc4b7085a..c57cbddca760 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1808,14 +1808,14 @@ static void
 pnfs_set_plh_return_info(struct pnfs_layout_hdr *lo, enum pnfs_iomode iomode,
 			 u32 seq)
 {
-	if (lo->plh_return_iomode == iomode)
-		return;
-	if (lo->plh_return_iomode != 0)
+	if (lo->plh_return_iomode != 0 && lo->plh_return_iomode != iomode)
 		iomode = IOMODE_ANY;
 	lo->plh_return_iomode = iomode;
 	set_bit(NFS_LAYOUT_RETURN_REQUESTED, &lo->plh_flags);
-	if (!lo->plh_return_seq || pnfs_seqid_is_newer(seq, lo->plh_return_seq))
+	if (seq != 0) {
+		WARN_ON_ONCE(lo->plh_return_seq != 0 && lo->plh_return_seq != seq);
 		lo->plh_return_seq = seq;
+	}
 }
 
 /**
@@ -1876,14 +1876,13 @@ void pnfs_error_mark_layout_for_return(struct inode *inode,
 	bool return_now = false;
 
 	spin_lock(&inode->i_lock);
-	pnfs_set_plh_return_info(lo, range.iomode, lseg->pls_seq);
+	pnfs_set_plh_return_info(lo, range.iomode, 0);
 	/*
 	 * mark all matching lsegs so that we are sure to have no live
 	 * segments at hand when sending layoutreturn. See pnfs_put_lseg()
 	 * for how it works.
 	 */
-	if (!pnfs_mark_matching_lsegs_return(lo, &free_me,
-						&range, lseg->pls_seq)) {
+	if (!pnfs_mark_matching_lsegs_return(lo, &free_me, &range, 0)) {
 		nfs4_stateid stateid;
 		enum pnfs_iomode iomode;
 
-- 
2.7.4

[PATCH v3 08/14] NFS: pnfs_mark_matching_lsegs_return() should match the layout sequence id

[Trond Myklebust]

When determining which layout segments to return, we do want
pnfs_mark_matching_lsegs_return to check that they match the layout
sequence id. This ensures that we don't waste time if the server
is replaying a layout recall that has already been satisfied.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 37 +++++++++++++++++++++++--------------
 1 file changed, 23 insertions(+), 14 deletions(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index c57cbddca760..52b2a4dfdcb0 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -486,15 +486,6 @@ pnfs_lseg_range_intersecting(const struct pnfs_layout_range *l1,
 	       (end2 == NFS4_MAX_UINT64 || end2 > start1);
 }
 
-static bool
-should_free_lseg(const struct pnfs_layout_range *lseg_range,
-		 const struct pnfs_layout_range *recall_range)
-{
-	return (recall_range->iomode == IOMODE_ANY ||
-		lseg_range->iomode == recall_range->iomode) &&
-	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
-}
-
 static bool pnfs_lseg_dec_and_remove_zero(struct pnfs_layout_segment *lseg,
 		struct list_head *tmp_list)
 {
@@ -533,6 +524,27 @@ static bool pnfs_seqid_is_newer(u32 s1, u32 s2)
 	return (s32)(s1 - s2) > 0;
 }
 
+static bool
+pnfs_should_free_range(const struct pnfs_layout_range *lseg_range,
+		 const struct pnfs_layout_range *recall_range)
+{
+	return (recall_range->iomode == IOMODE_ANY ||
+		lseg_range->iomode == recall_range->iomode) &&
+	       pnfs_lseg_range_intersecting(lseg_range, recall_range);
+}
+
+static bool
+pnfs_match_lseg_recall(const struct pnfs_layout_segment *lseg,
+		const struct pnfs_layout_range *recall_range,
+		u32 seq)
+{
+	if (seq != 0 && pnfs_seqid_is_newer(lseg->pls_seq, seq))
+		return false;
+	if (recall_range == NULL)
+		return true;
+	return pnfs_should_free_range(&lseg->pls_range, recall_range);
+}
+
 /**
  * pnfs_mark_matching_lsegs_invalid - tear down lsegs or mark them for later
  * @lo: layout header containing the lsegs
@@ -562,10 +574,7 @@ pnfs_mark_matching_lsegs_invalid(struct pnfs_layout_hdr *lo,
 	if (list_empty(&lo->plh_segs))
 		return 0;
 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
-		if (!recall_range ||
-		    should_free_lseg(&lseg->pls_range, recall_range)) {
-			if (seq && pnfs_seqid_is_newer(lseg->pls_seq, seq))
-				continue;
+		if (pnfs_match_lseg_recall(lseg, recall_range, seq)) {
 			dprintk("%s: freeing lseg %p iomode %d seq %u"
 				"offset %llu length %llu\n", __func__,
 				lseg, lseg->pls_range.iomode, lseg->pls_seq,
@@ -1845,7 +1854,7 @@ pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
 	assert_spin_locked(&lo->plh_inode->i_lock);
 
 	list_for_each_entry_safe(lseg, next, &lo->plh_segs, pls_list)
-		if (should_free_lseg(&lseg->pls_range, return_range)) {
+		if (pnfs_match_lseg_recall(lseg, return_range, seq)) {
 			dprintk("%s: marking lseg %p iomode %d "
 				"offset %llu length %llu\n", __func__,
 				lseg, lseg->pls_range.iomode,
-- 
2.7.4

[PATCH v3 09/14] pNFS: Cleanup - don't open code pnfs_mark_layout_stateid_invalid()

[Trond Myklebust]

Ensure nfs42_layoutstat_done() layoutget don't open code layout stateid
invalidation.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/nfs42proc.c | 3 +--
 fs/nfs/nfs4proc.c  | 3 +--
 fs/nfs/pnfs.c      | 2 +-
 fs/nfs/pnfs.h      | 2 ++
 4 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/fs/nfs/nfs42proc.c b/fs/nfs/nfs42proc.c
index aa03ed09ba06..6ea5ad6f0d44 100644
--- a/fs/nfs/nfs42proc.c
+++ b/fs/nfs/nfs42proc.c
@@ -336,8 +336,7 @@ nfs42_layoutstat_done(struct rpc_task *task, void *calldata)
 			 * Mark the bad layout state as invalid, then retry
 			 * with the current stateid.
 			 */
-			set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
-			pnfs_mark_matching_lsegs_invalid(lo, &head, NULL, 0);
+			pnfs_mark_layout_stateid_invalid(lo, &head);
 			spin_unlock(&inode->i_lock);
 			pnfs_free_lseg_list(&head);
 		} else
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index dc50ba4d84ad..8c453515d98c 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -7931,8 +7931,7 @@ nfs4_layoutget_handle_exception(struct rpc_task *task,
 			 * Mark the bad layout state as invalid, then retry
 			 * with the current stateid.
 			 */
-			set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
-			pnfs_mark_matching_lsegs_invalid(lo, &head, NULL, 0);
+			pnfs_mark_layout_stateid_invalid(lo, &head);
 			spin_unlock(&inode->i_lock);
 			pnfs_free_lseg_list(&head);
 			status = -EAGAIN;
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 52b2a4dfdcb0..1374fcd40484 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -259,7 +259,7 @@ pnfs_put_layout_hdr(struct pnfs_layout_hdr *lo)
  * is required.
  * Note that caller must hold inode->i_lock.
  */
-static int
+int
 pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
 		struct list_head *lseg_list)
 {
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index 2f4f26905c03..d71c9493693a 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -268,6 +268,8 @@ int pnfs_mark_matching_lsegs_return(struct pnfs_layout_hdr *lo,
 				struct list_head *tmp_list,
 				const struct pnfs_layout_range *recall_range,
 				u32 seq);
+int pnfs_mark_layout_stateid_invalid(struct pnfs_layout_hdr *lo,
+		struct list_head *lseg_list);
 bool pnfs_roc(struct inode *ino);
 void pnfs_roc_release(struct inode *ino);
 void pnfs_roc_set_barrier(struct inode *ino, u32 barrier);
-- 
2.7.4

[PATCH v3 10/14] pNFS: Clear the layout metadata if the server changed the layout stateid

[Trond Myklebust]

If the server changed the layout stateid's "other" field, then
we should treat the old layout as being completely gone. In that
case, we want to clear the metadata such as scheduled layoutreturns.

Do this by calling pnfs_mark_layout_stateid_invalid().

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 1374fcd40484..e16997da157f 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1785,7 +1785,7 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
 		 * inode invalid, and don't bother validating the stateid
 		 * sequence number.
 		 */
-		pnfs_mark_matching_lsegs_invalid(lo, &free_me, NULL, 0);
+		pnfs_mark_layout_stateid_invalid(lo, &free_me);
 
 		nfs4_stateid_copy(&lo->plh_stateid, &res->stateid);
 		lo->plh_barrier = be32_to_cpu(res->stateid.seqid);
-- 
2.7.4

[PATCH v3 11/14] pNFS: Remove redundant pnfs_mark_layout_returned_if_empty()

[Trond Myklebust]

That's already being taken care of in pnfs_layout_remove_lseg().

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/callback_proc.c |  1 -
 fs/nfs/nfs4proc.c      |  1 -
 fs/nfs/pnfs.c          |  1 -
 fs/nfs/pnfs.h          | 13 -------------
 4 files changed, 16 deletions(-)

diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index 837da8a02d35..c92a75e066a6 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -252,7 +252,6 @@ static u32 initiate_file_draining(struct nfs_client *clp,
 		NFS_SERVER(ino)->pnfs_curr_ld->return_range(lo,
 			&args->cbl_range);
 	}
-	pnfs_mark_layout_returned_if_empty(lo);
 unlock:
 	spin_unlock(&ino->i_lock);
 	pnfs_free_lseg_list(&free_me_list);
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 8c453515d98c..d38cc5f0282f 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -8128,7 +8128,6 @@ static void nfs4_layoutreturn_release(void *calldata)
 	spin_lock(&lo->plh_inode->i_lock);
 	pnfs_mark_matching_lsegs_invalid(lo, &freeme, &lrp->args.range,
 			be32_to_cpu(lrp->args.stateid.seqid));
-	pnfs_mark_layout_returned_if_empty(lo);
 	if (lrp->res.lrs_present && pnfs_layout_is_valid(lo))
 		pnfs_set_layout_stateid(lo, &lrp->res.stateid, true);
 	pnfs_clear_layoutreturn_waitbit(lo);
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index e16997da157f..49e952968ede 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1159,7 +1159,6 @@ void pnfs_roc_set_barrier(struct inode *ino, u32 barrier)
 
 	spin_lock(&ino->i_lock);
 	lo = NFS_I(ino)->layout;
-	pnfs_mark_layout_returned_if_empty(lo);
 	if (pnfs_seqid_is_newer(barrier, lo->plh_barrier))
 		lo->plh_barrier = barrier;
 	spin_unlock(&ino->i_lock);
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
index d71c9493693a..595648a3ad7a 100644
--- a/fs/nfs/pnfs.h
+++ b/fs/nfs/pnfs.h
@@ -552,19 +552,6 @@ pnfs_calc_offset_length(u64 offset, u64 end)
 	return 1 + end - offset;
 }
 
-/**
- * pnfs_mark_layout_returned_if_empty - marks the layout as returned
- * @lo: layout header
- *
- * Note: Caller must hold inode->i_lock
- */
-static inline void
-pnfs_mark_layout_returned_if_empty(struct pnfs_layout_hdr *lo)
-{
-	if (list_empty(&lo->plh_segs))
-		set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
-}
-
 static inline void
 pnfs_copy_range(struct pnfs_layout_range *dst,
 		const struct pnfs_layout_range *src)
-- 
2.7.4

[PATCH v3 12/14] pNFS: Remove redundant stateid invalidation

[Trond Myklebust]

The layout stateid will be invalidated once it holds no more layout
segments anyway.

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

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 49e952968ede..c9a684c128b8 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -1040,7 +1040,6 @@ _pnfs_return_layout(struct inode *ino)
 		goto out_put_layout_hdr;
 	}
 
-	set_bit(NFS_LAYOUT_INVALID_STID, &lo->plh_flags);
 	send = pnfs_prepare_layoutreturn(lo, &stateid, NULL);
 	spin_unlock(&ino->i_lock);
 	pnfs_free_lseg_list(&tmp_list);
-- 
2.7.4

[PATCH v3 13/14] pNFS: Cleanup - do layout segment initialisation in one place

[Trond Myklebust]

...instead of splitting the initialisation over init_lseg() and
pnfs_layout_process().

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
---
 fs/nfs/pnfs.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index c9a684c128b8..09b77a68422f 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -334,7 +334,9 @@ pnfs_layout_io_test_failed(struct pnfs_layout_hdr *lo, u32 iomode)
 }
 
 static void
-init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
+pnfs_init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg,
+		const struct pnfs_layout_range *range,
+		const nfs4_stateid *stateid)
 {
 	INIT_LIST_HEAD(&lseg->pls_list);
 	INIT_LIST_HEAD(&lseg->pls_lc_list);
@@ -342,6 +344,8 @@ init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
 	smp_mb();
 	set_bit(NFS_LSEG_VALID, &lseg->pls_flags);
 	lseg->pls_layout = lo;
+	lseg->pls_range = *range;
+	lseg->pls_seq = be32_to_cpu(stateid->seqid);
 }
 
 static void pnfs_free_lseg(struct pnfs_layout_segment *lseg)
@@ -1760,9 +1764,7 @@ pnfs_layout_process(struct nfs4_layoutget *lgp)
 		return lseg;
 	}
 
-	init_lseg(lo, lseg);
-	lseg->pls_range = res->range;
-	lseg->pls_seq = be32_to_cpu(res->stateid.seqid);
+	pnfs_init_lseg(lo, lseg, &res->range, &res->stateid);
 
 	spin_lock(&ino->i_lock);
 	if (pnfs_layoutgets_blocked(lo)) {
-- 
2.7.4

[PATCH v3 14/14] pNFS: Remove redundant smp_mb() from pnfs_init_lseg()

[Trond Myklebust]

It's not visible yet, and won't be until after we grab the inode->i_lock.

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

diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
index 09b77a68422f..1d47a845d77a 100644
--- a/fs/nfs/pnfs.c
+++ b/fs/nfs/pnfs.c
@@ -341,7 +341,6 @@ pnfs_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);
 	lseg->pls_layout = lo;
 	lseg->pls_range = *range;
-- 
2.7.4