[PATCH 00/20] NFSv4.1: Add support for session dynamic slot management

[PATCH 00/20] NFSv4.1: Add support for session dynamic slot management

[Trond Myklebust]

This patch series adds support for session dynamic slot management
in accordance with Section 2.10.6.1 of RFC5661.

By tracking the sr_target_highest_slotid and sr_highest_slotid in the
server's replies to our SEQUENCE operations, we can figure out when
it is safe to increase the number of RPC requests that go in parallel
to the server, and when we should decrease that number.


Trond Myklebust (20):
  NFSv4.1: Don't confuse CREATE_SESSION arguments and results
  NFSv4.1: Adjust CREATE_SESSION arguments when mounting a new
    filesystem
  NFSv4.1: We must bump the clientid sequence number after
    CREATE_SESSION
  NFSv4.1: nfs4_alloc_slots doesn't need zeroing
  NFSv4.1: clean up nfs4_recall_slot to use nfs4_alloc_slots
  NFSv4.1: Shrink struct nfs4_sequence_res by moving sr_renewal_time
  NFSv4.1: Shrink struct nfs4_sequence_res by moving the session
    pointer
  NFSv4.1: Label each entry in the session slot tables with its slot
    number
  NFSv4.1: Simplify struct nfs4_sequence_args too
  NFSv4.1: Simplify slot allocation
  NFSv4.1: Clean up nfs4_free_slot
  NFSv4.1: Ensure that the client tracks the server
    target_highest_slotid
  NFSv4.1: Reset the sequence number for slots that have been
    deallocated
  NFSv4.1: Fix nfs4_callback_recallslot to work with dynamic slot
    allocation
  NFSv4.1: Don't confuse target_highest_slotid and max_slots in
    cb_recall_slot
  NFSv4.1: Allow the server to recall all but one slot
  NFSv4.1: Support dynamic resizing of the session slot table
  NFSv4.1: Allow SEQUENCE to resize the slot table on the fly
  NFSv4.1: Remove the state manager code to resize the slot table
  NFSv4.1: CB_RECALL_SLOT must schedule a sequence op after updating
    targets

 fs/nfs/callback.h         |   2 +-
 fs/nfs/callback_proc.c    |  15 +--
 fs/nfs/callback_xdr.c     |   2 +-
 fs/nfs/nfs4_fs.h          |   6 +-
 fs/nfs/nfs4proc.c         | 316 ++++++++++++++++++++++++++++++++--------------
 fs/nfs/nfs4state.c        |  52 +-------
 fs/nfs/nfs4xdr.c          |  37 +++---
 include/linux/nfs_fs_sb.h |  11 +-
 include/linux/nfs_xdr.h   |  12 +-
 9 files changed, 275 insertions(+), 178 deletions(-)

-- 
1.7.11.7

[PATCH 01/20] NFSv4.1: Don't confuse CREATE_SESSION arguments and results

[Trond Myklebust]

Don't store the target request and response sizes in the same
variables used to store the server's replies to those targets.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c         | 30 +++++++++++++++++-------------
 include/linux/nfs_fs_sb.h |  3 +++
 2 files changed, 20 insertions(+), 13 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index a32d953..3e572dc 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -5807,8 +5807,8 @@ void nfs4_destroy_session(struct nfs4_session *session)
 static void nfs4_init_channel_attrs(struct nfs41_create_session_args *args)
 {
 	struct nfs4_session *session = args->client->cl_session;
-	unsigned int mxrqst_sz = session->fc_attrs.max_rqst_sz,
-		     mxresp_sz = session->fc_attrs.max_resp_sz;
+	unsigned int mxrqst_sz = session->fc_target_max_rqst_sz,
+		     mxresp_sz = session->fc_target_max_resp_sz;
 
 	if (mxrqst_sz == 0)
 		mxrqst_sz = NFS_MAX_FILE_IO_SIZE;
@@ -6015,24 +6015,28 @@ int nfs4_init_session(struct nfs_server *server)
 {
 	struct nfs_client *clp = server->nfs_client;
 	struct nfs4_session *session;
-	unsigned int rsize, wsize;
+	unsigned int target_max_rqst_sz = NFS_MAX_FILE_IO_SIZE;
+	unsigned int target_max_resp_sz = NFS_MAX_FILE_IO_SIZE;
 
 	if (!nfs4_has_session(clp))
 		return 0;
 
+	if (server->rsize != 0)
+		target_max_resp_sz = server->rsize;
+	target_max_resp_sz += nfs41_maxread_overhead;
+
+	if (server->wsize != 0)
+		target_max_rqst_sz = server->wsize;
+	target_max_rqst_sz += nfs41_maxwrite_overhead;
+
 	session = clp->cl_session;
 	spin_lock(&clp->cl_lock);
 	if (test_and_clear_bit(NFS4_SESSION_INITING, &session->session_state)) {
-
-		rsize = server->rsize;
-		if (rsize == 0)
-			rsize = NFS_MAX_FILE_IO_SIZE;
-		wsize = server->wsize;
-		if (wsize == 0)
-			wsize = NFS_MAX_FILE_IO_SIZE;
-
-		session->fc_attrs.max_rqst_sz = wsize + nfs41_maxwrite_overhead;
-		session->fc_attrs.max_resp_sz = rsize + nfs41_maxread_overhead;
+		/* Initialise targets and channel attributes */
+		session->fc_target_max_rqst_sz = target_max_rqst_sz;
+		session->fc_attrs.max_rqst_sz = target_max_rqst_sz;
+		session->fc_target_max_resp_sz = target_max_resp_sz;
+		session->fc_attrs.max_resp_sz = target_max_resp_sz;
 	}
 	spin_unlock(&clp->cl_lock);
 
diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h
index a9e76ee..97c8f91 100644
--- a/include/linux/nfs_fs_sb.h
+++ b/include/linux/nfs_fs_sb.h
@@ -242,6 +242,9 @@ struct nfs4_session {
 	struct nfs4_channel_attrs	bc_attrs;
 	struct nfs4_slot_table		bc_slot_table;
 	struct nfs_client		*clp;
+	/* Create session arguments */
+	unsigned int			fc_target_max_rqst_sz;
+	unsigned int			fc_target_max_resp_sz;
 };
 
 #endif /* CONFIG_NFS_V4 */
-- 
1.7.11.7

[PATCH 02/20] NFSv4.1: Adjust CREATE_SESSION arguments when mounting a new filesystem

[Trond Myklebust]

If we're mounting a new filesystem, ensure that the session has negotiated
large enough request and reply sizes to match the wsize and rsize mount
arguments.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 3e572dc..ee82cdd 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -6037,9 +6037,22 @@ int nfs4_init_session(struct nfs_server *server)
 		session->fc_attrs.max_rqst_sz = target_max_rqst_sz;
 		session->fc_target_max_resp_sz = target_max_resp_sz;
 		session->fc_attrs.max_resp_sz = target_max_resp_sz;
+	} else {
+		/* Just adjust the targets */
+		if (target_max_rqst_sz > session->fc_target_max_rqst_sz) {
+			session->fc_target_max_rqst_sz = target_max_rqst_sz;
+			set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
+		}
+		if (target_max_resp_sz > session->fc_target_max_resp_sz) {
+			session->fc_target_max_resp_sz = target_max_resp_sz;
+			set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
+		}
 	}
 	spin_unlock(&clp->cl_lock);
 
+	if (test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
+		nfs4_schedule_lease_recovery(clp);
+
 	return nfs41_check_session_ready(clp);
 }
 
-- 
1.7.11.7

[PATCH 03/20] NFSv4.1: We must bump the clientid sequence number after CREATE_SESSION

[Trond Myklebust]

We must always bump the clientid sequence number after a successful
call to CREATE_SESSION on the server. The result of
nfs4_verify_channel_attrs() is irrelevant to that requirement.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index ee82cdd..1ac339b 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -5917,10 +5917,9 @@ static int _nfs4_proc_create_session(struct nfs_client *clp,
 
 	status = rpc_call_sync(session->clp->cl_rpcclient, &msg, RPC_TASK_TIMEOUT);
 
-	if (!status)
+	if (!status) {
 		/* Verify the session's negotiated channel_attrs values */
 		status = nfs4_verify_channel_attrs(&args, session);
-	if (!status) {
 		/* Increment the clientid slot sequence id */
 		clp->cl_seqid++;
 	}
-- 
1.7.11.7

[PATCH 04/20] NFSv4.1: nfs4_alloc_slots doesn't need zeroing

[Trond Myklebust]

All that memory is going to be initialised to non-zero by
nfs4_add_and_init_slots anyway.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 1ac339b..0402ebb 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -5658,7 +5658,7 @@ int nfs4_proc_get_lease_time(struct nfs_client *clp, struct nfs_fsinfo *fsinfo)
 
 static struct nfs4_slot *nfs4_alloc_slots(u32 max_slots, gfp_t gfp_flags)
 {
-	return kcalloc(max_slots, sizeof(struct nfs4_slot), gfp_flags);
+	return kmalloc_array(max_slots, sizeof(struct nfs4_slot), gfp_flags);
 }
 
 static void nfs4_add_and_init_slots(struct nfs4_slot_table *tbl,
-- 
1.7.11.7

[PATCH 05/20] NFSv4.1: clean up nfs4_recall_slot to use nfs4_alloc_slots

[Trond Myklebust]

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4_fs.h   | 2 ++
 fs/nfs/nfs4proc.c  | 2 +-
 fs/nfs/nfs4state.c | 3 +--
 3 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index a525fde..36880b9 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -258,6 +258,8 @@ extern int nfs4_proc_get_lease_time(struct nfs_client *clp,
 extern int nfs4_proc_layoutcommit(struct nfs4_layoutcommit_data *data,
 				  bool sync);
 
+extern struct nfs4_slot *nfs4_alloc_slots(u32 max_slots, gfp_t gfp_flags);
+
 static inline bool
 is_ds_only_client(struct nfs_client *clp)
 {
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 0402ebb..5e5cc5a 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -5656,7 +5656,7 @@ int nfs4_proc_get_lease_time(struct nfs_client *clp, struct nfs_fsinfo *fsinfo)
 	return status;
 }
 
-static struct nfs4_slot *nfs4_alloc_slots(u32 max_slots, gfp_t gfp_flags)
+struct nfs4_slot *nfs4_alloc_slots(u32 max_slots, gfp_t gfp_flags)
 {
 	return kmalloc_array(max_slots, sizeof(struct nfs4_slot), gfp_flags);
 }
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index f3d1bc4..96fcbb9 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -2033,8 +2033,7 @@ static int nfs4_recall_slot(struct nfs_client *clp)
 		return 0;
 	nfs4_begin_drain_session(clp);
 	fc_tbl = &clp->cl_session->fc_slot_table;
-	new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
-		      GFP_NOFS);
+	new = nfs4_alloc_slots(fc_tbl->target_max_slots, GFP_NOFS);
         if (!new)
 		return -ENOMEM;
 
-- 
1.7.11.7

[PATCH 06/20] NFSv4.1: Shrink struct nfs4_sequence_res by moving sr_renewal_time

[Trond Myklebust]

Store the renewal time inside the session slot instead.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c       | 15 +++++++++------
 include/linux/nfs_xdr.h |  2 +-
 2 files changed, 10 insertions(+), 7 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 5e5cc5a..14b3974 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -486,6 +486,7 @@ static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
 
 static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
 {
+	struct nfs4_slot *slot;
 	unsigned long timestamp;
 	struct nfs_client *clp;
 
@@ -502,12 +503,14 @@ static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *
 	if (!RPC_WAS_SENT(task))
 		goto out;
 
+	slot = res->sr_slot;
+
 	/* Check the SEQUENCE operation status */
 	switch (res->sr_status) {
 	case 0:
 		/* Update the slot's sequence and clientid lease timer */
-		++res->sr_slot->seq_nr;
-		timestamp = res->sr_renewal_time;
+		++slot->seq_nr;
+		timestamp = slot->renewal_time;
 		clp = res->sr_session->clp;
 		do_renew_lease(clp, timestamp);
 		/* Check sequence flags */
@@ -521,12 +524,12 @@ static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *
 		 */
 		dprintk("%s: slot=%td seq=%d: Operation in progress\n",
 			__func__,
-			res->sr_slot - res->sr_session->fc_slot_table.slots,
-			res->sr_slot->seq_nr);
+			slot - res->sr_session->fc_slot_table.slots,
+			slot->seq_nr);
 		goto out_retry;
 	default:
 		/* Just update the slot sequence no. */
-		++res->sr_slot->seq_nr;
+		++slot->seq_nr;
 	}
 out:
 	/* The session may be reset by one of the error handlers. */
@@ -637,6 +640,7 @@ int nfs41_setup_sequence(struct nfs4_session *session,
 
 	rpc_task_set_priority(task, RPC_PRIORITY_NORMAL);
 	slot = tbl->slots + slotid;
+	slot->renewal_time = jiffies;
 	args->sa_session = session;
 	args->sa_slotid = slotid;
 
@@ -644,7 +648,6 @@ int nfs41_setup_sequence(struct nfs4_session *session,
 
 	res->sr_session = session;
 	res->sr_slot = slot;
-	res->sr_renewal_time = jiffies;
 	res->sr_status_flags = 0;
 	/*
 	 * sr_status is only set in decode_sequence, and so will remain
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index a73ea89..9cb1c63 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -187,6 +187,7 @@ struct nfs4_channel_attrs {
 
 /* nfs41 sessions slot seqid */
 struct nfs4_slot {
+	unsigned long		renewal_time;
 	u32		 	seq_nr;
 };
 
@@ -200,7 +201,6 @@ struct nfs4_sequence_res {
 	struct nfs4_session	*sr_session;
 	struct nfs4_slot	*sr_slot;	/* slot used to send request */
 	int			sr_status;	/* sequence operation status */
-	unsigned long		sr_renewal_time;
 	u32			sr_status_flags;
 };
 
-- 
1.7.11.7

[PATCH 07/20] NFSv4.1: Shrink struct nfs4_sequence_res by moving the session pointer

[Trond Myklebust]

Move the session pointer into the slot table, then have struct nfs4_slot
point to that slot table.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4_fs.h          |  3 ++-
 fs/nfs/nfs4proc.c         | 34 ++++++++++++++++++++++++----------
 fs/nfs/nfs4state.c        |  2 +-
 fs/nfs/nfs4xdr.c          |  8 +++++---
 include/linux/nfs_fs_sb.h |  1 +
 include/linux/nfs_xdr.h   |  2 +-
 6 files changed, 34 insertions(+), 16 deletions(-)

diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index 36880b9..42c5869 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -258,7 +258,8 @@ extern int nfs4_proc_get_lease_time(struct nfs_client *clp,
 extern int nfs4_proc_layoutcommit(struct nfs4_layoutcommit_data *data,
 				  bool sync);
 
-extern struct nfs4_slot *nfs4_alloc_slots(u32 max_slots, gfp_t gfp_flags);
+extern struct nfs4_slot *nfs4_alloc_slots(struct nfs4_slot_table *table,
+		u32 max_slots, gfp_t gfp_flags);
 
 static inline bool
 is_ds_only_client(struct nfs_client *clp)
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 14b3974..f8cb60b 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -467,25 +467,29 @@ void nfs4_check_drain_bc_complete(struct nfs4_session *ses)
 
 static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
 {
+	struct nfs4_session *session;
 	struct nfs4_slot_table *tbl;
 
-	tbl = &res->sr_session->fc_slot_table;
 	if (!res->sr_slot) {
 		/* just wake up the next guy waiting since
 		 * we may have not consumed a slot after all */
 		dprintk("%s: No slot\n", __func__);
 		return;
 	}
+	tbl = res->sr_slot->table;
+	session = tbl->session;
+	tbl = &session->fc_slot_table;
 
 	spin_lock(&tbl->slot_tbl_lock);
 	nfs4_free_slot(tbl, res->sr_slot - tbl->slots);
-	nfs4_check_drain_fc_complete(res->sr_session);
+	nfs4_check_drain_fc_complete(session);
 	spin_unlock(&tbl->slot_tbl_lock);
 	res->sr_slot = NULL;
 }
 
 static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
 {
+	struct nfs4_session *session;
 	struct nfs4_slot *slot;
 	unsigned long timestamp;
 	struct nfs_client *clp;
@@ -504,6 +508,7 @@ static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *
 		goto out;
 
 	slot = res->sr_slot;
+	session = slot->table->session;
 
 	/* Check the SEQUENCE operation status */
 	switch (res->sr_status) {
@@ -511,7 +516,7 @@ static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *
 		/* Update the slot's sequence and clientid lease timer */
 		++slot->seq_nr;
 		timestamp = slot->renewal_time;
-		clp = res->sr_session->clp;
+		clp = session->clp;
 		do_renew_lease(clp, timestamp);
 		/* Check sequence flags */
 		if (res->sr_status_flags != 0)
@@ -524,7 +529,7 @@ static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *
 		 */
 		dprintk("%s: slot=%td seq=%d: Operation in progress\n",
 			__func__,
-			slot - res->sr_session->fc_slot_table.slots,
+			slot - session->fc_slot_table.slots,
 			slot->seq_nr);
 		goto out_retry;
 	default:
@@ -546,7 +551,7 @@ out_retry:
 static int nfs4_sequence_done(struct rpc_task *task,
 			       struct nfs4_sequence_res *res)
 {
-	if (res->sr_session == NULL)
+	if (res->sr_slot == NULL)
 		return 1;
 	return nfs41_sequence_done(task, res);
 }
@@ -591,7 +596,6 @@ static void nfs41_init_sequence(struct nfs4_sequence_args *args,
 	args->sa_cache_this = 0;
 	if (cache_reply)
 		args->sa_cache_this = 1;
-	res->sr_session = NULL;
 	res->sr_slot = NULL;
 }
 
@@ -646,7 +650,6 @@ int nfs41_setup_sequence(struct nfs4_session *session,
 
 	dprintk("<-- %s slotid=%d seqid=%d\n", __func__, slotid, slot->seq_nr);
 
-	res->sr_session = session;
 	res->sr_slot = slot;
 	res->sr_status_flags = 0;
 	/*
@@ -5659,9 +5662,18 @@ int nfs4_proc_get_lease_time(struct nfs_client *clp, struct nfs_fsinfo *fsinfo)
 	return status;
 }
 
-struct nfs4_slot *nfs4_alloc_slots(u32 max_slots, gfp_t gfp_flags)
+struct nfs4_slot *nfs4_alloc_slots(struct nfs4_slot_table *table,
+		u32 max_slots, gfp_t gfp_flags)
 {
-	return kmalloc_array(max_slots, sizeof(struct nfs4_slot), gfp_flags);
+	struct nfs4_slot *tbl;
+	u32 i;
+
+	tbl = kmalloc_array(max_slots, sizeof(*tbl), gfp_flags);
+	if (tbl != NULL) {
+		for (i = 0; i < max_slots; i++)
+			tbl[i].table = table;
+	}
+	return tbl;
 }
 
 static void nfs4_add_and_init_slots(struct nfs4_slot_table *tbl,
@@ -5699,7 +5711,7 @@ static int nfs4_realloc_slot_table(struct nfs4_slot_table *tbl, u32 max_reqs,
 
 	/* Does the newly negotiated max_reqs match the existing slot table? */
 	if (max_reqs != tbl->max_slots) {
-		new = nfs4_alloc_slots(max_reqs, GFP_NOFS);
+		new = nfs4_alloc_slots(tbl, max_reqs, GFP_NOFS);
 		if (!new)
 			goto out;
 	}
@@ -5738,11 +5750,13 @@ static int nfs4_setup_session_slot_tables(struct nfs4_session *ses)
 	dprintk("--> %s\n", __func__);
 	/* Fore channel */
 	tbl = &ses->fc_slot_table;
+	tbl->session = ses;
 	status = nfs4_realloc_slot_table(tbl, ses->fc_attrs.max_reqs, 1);
 	if (status) /* -ENOMEM */
 		return status;
 	/* Back channel */
 	tbl = &ses->bc_slot_table;
+	tbl->session = ses;
 	status = nfs4_realloc_slot_table(tbl, ses->bc_attrs.max_reqs, 0);
 	if (status && tbl->slots == NULL)
 		/* Fore and back channel share a connection so get
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 96fcbb9..9495789 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -2033,7 +2033,7 @@ static int nfs4_recall_slot(struct nfs_client *clp)
 		return 0;
 	nfs4_begin_drain_session(clp);
 	fc_tbl = &clp->cl_session->fc_slot_table;
-	new = nfs4_alloc_slots(fc_tbl->target_max_slots, GFP_NOFS);
+	new = nfs4_alloc_slots(fc_tbl, fc_tbl->target_max_slots, GFP_NOFS);
         if (!new)
 		return -ENOMEM;
 
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 672d9b0..4126f05 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -5507,12 +5507,13 @@ static int decode_sequence(struct xdr_stream *xdr,
 			   struct rpc_rqst *rqstp)
 {
 #if defined(CONFIG_NFS_V4_1)
+	struct nfs4_session *session;
 	struct nfs4_sessionid id;
 	u32 dummy;
 	int status;
 	__be32 *p;
 
-	if (!res->sr_session)
+	if (res->sr_slot == NULL)
 		return 0;
 
 	status = decode_op_hdr(xdr, OP_SEQUENCE);
@@ -5526,8 +5527,9 @@ static int decode_sequence(struct xdr_stream *xdr,
 	 * sequence number, the server is looney tunes.
 	 */
 	status = -EREMOTEIO;
+	session = res->sr_slot->table->session;
 
-	if (memcmp(id.data, res->sr_session->sess_id.data,
+	if (memcmp(id.data, session->sess_id.data,
 		   NFS4_MAX_SESSIONID_LEN)) {
 		dprintk("%s Invalid session id\n", __func__);
 		goto out_err;
@@ -5545,7 +5547,7 @@ static int decode_sequence(struct xdr_stream *xdr,
 	}
 	/* slot id */
 	dummy = be32_to_cpup(p++);
-	if (dummy != res->sr_slot - res->sr_session->fc_slot_table.slots) {
+	if (dummy != res->sr_slot - session->fc_slot_table.slots) {
 		dprintk("%s Invalid slot id\n", __func__);
 		goto out_err;
 	}
diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h
index 97c8f91..b041287 100644
--- a/include/linux/nfs_fs_sb.h
+++ b/include/linux/nfs_fs_sb.h
@@ -209,6 +209,7 @@ struct nfs_server {
 /* Sessions */
 #define SLOT_TABLE_SZ DIV_ROUND_UP(NFS4_MAX_SLOT_TABLE, 8*sizeof(long))
 struct nfs4_slot_table {
+	struct nfs4_session *session;		/* Parent session */
 	struct nfs4_slot *slots;		/* seqid per slot */
 	unsigned long   used_slots[SLOT_TABLE_SZ]; /* used/unused bitmap */
 	spinlock_t	slot_tbl_lock;
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index 9cb1c63..0fd88ab 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -187,6 +187,7 @@ struct nfs4_channel_attrs {
 
 /* nfs41 sessions slot seqid */
 struct nfs4_slot {
+	struct nfs4_slot_table	*table;
 	unsigned long		renewal_time;
 	u32		 	seq_nr;
 };
@@ -198,7 +199,6 @@ struct nfs4_sequence_args {
 };
 
 struct nfs4_sequence_res {
-	struct nfs4_session	*sr_session;
 	struct nfs4_slot	*sr_slot;	/* slot used to send request */
 	int			sr_status;	/* sequence operation status */
 	u32			sr_status_flags;
-- 
1.7.11.7

[PATCH 08/20] NFSv4.1: Label each entry in the session slot tables with its slot number

[Trond Myklebust]

Instead of doing slot table pointer gymnastics every time we want to
know which slot we're using.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c       | 12 +++++++-----
 fs/nfs/nfs4xdr.c        |  2 +-
 include/linux/nfs_xdr.h |  1 +
 3 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index f8cb60b..4f2d058 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -527,9 +527,9 @@ static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *
 		 * returned NFS4ERR_DELAY as per Section 2.10.6.2
 		 * of RFC5661.
 		 */
-		dprintk("%s: slot=%td seq=%d: Operation in progress\n",
+		dprintk("%s: slot=%u seq=%u: Operation in progress\n",
 			__func__,
-			slot - session->fc_slot_table.slots,
+			slot->slot_nr,
 			slot->seq_nr);
 		goto out_retry;
 	default:
@@ -672,9 +672,9 @@ int nfs4_setup_sequence(const struct nfs_server *server,
 	if (session == NULL)
 		goto out;
 
-	dprintk("--> %s clp %p session %p sr_slot %td\n",
+	dprintk("--> %s clp %p session %p sr_slot %d\n",
 		__func__, session->clp, session, res->sr_slot ?
-			res->sr_slot - session->fc_slot_table.slots : -1);
+			res->sr_slot->slot_nr : -1);
 
 	ret = nfs41_setup_sequence(session, args, res, task);
 out:
@@ -5670,8 +5670,10 @@ struct nfs4_slot *nfs4_alloc_slots(struct nfs4_slot_table *table,
 
 	tbl = kmalloc_array(max_slots, sizeof(*tbl), gfp_flags);
 	if (tbl != NULL) {
-		for (i = 0; i < max_slots; i++)
+		for (i = 0; i < max_slots; i++) {
 			tbl[i].table = table;
+			tbl[i].slot_nr = i;
+		}
 	}
 	return tbl;
 }
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 4126f05..50bac70 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -5547,7 +5547,7 @@ static int decode_sequence(struct xdr_stream *xdr,
 	}
 	/* slot id */
 	dummy = be32_to_cpup(p++);
-	if (dummy != res->sr_slot - session->fc_slot_table.slots) {
+	if (dummy != res->sr_slot->slot_nr) {
 		dprintk("%s Invalid slot id\n", __func__);
 		goto out_err;
 	}
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index 0fd88ab..9c9b76c 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -189,6 +189,7 @@ struct nfs4_channel_attrs {
 struct nfs4_slot {
 	struct nfs4_slot_table	*table;
 	unsigned long		renewal_time;
+	u32			slot_nr;
 	u32		 	seq_nr;
 };
 
-- 
1.7.11.7

[PATCH 09/20] NFSv4.1: Simplify struct nfs4_sequence_args too

[Trond Myklebust]

Replace the session pointer + slotid with a pointer to the
allocated slot.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c       |  6 +++---
 fs/nfs/nfs4xdr.c        | 21 ++++++++++-----------
 include/linux/nfs_xdr.h |  3 +--
 3 files changed, 14 insertions(+), 16 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 4f2d058..7c2b037 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -592,7 +592,7 @@ out:
 static void nfs41_init_sequence(struct nfs4_sequence_args *args,
 		struct nfs4_sequence_res *res, int cache_reply)
 {
-	args->sa_session = NULL;
+	args->sa_slot = NULL;
 	args->sa_cache_this = 0;
 	if (cache_reply)
 		args->sa_cache_this = 1;
@@ -645,8 +645,8 @@ int nfs41_setup_sequence(struct nfs4_session *session,
 	rpc_task_set_priority(task, RPC_PRIORITY_NORMAL);
 	slot = tbl->slots + slotid;
 	slot->renewal_time = jiffies;
-	args->sa_session = session;
-	args->sa_slotid = slotid;
+
+	args->sa_slot = slot;
 
 	dprintk("<-- %s slotid=%d seqid=%d\n", __func__, slotid, slot->seq_nr);
 
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 50bac70..27b0fec 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -1833,18 +1833,16 @@ static void encode_sequence(struct xdr_stream *xdr,
 			    struct compound_hdr *hdr)
 {
 #if defined(CONFIG_NFS_V4_1)
-	struct nfs4_session *session = args->sa_session;
+	struct nfs4_session *session;
 	struct nfs4_slot_table *tp;
-	struct nfs4_slot *slot;
+	struct nfs4_slot *slot = args->sa_slot;
 	__be32 *p;
 
-	if (!session)
+	if (slot == NULL)
 		return;
 
-	tp = &session->fc_slot_table;
-
-	WARN_ON(args->sa_slotid == NFS4_MAX_SLOT_TABLE);
-	slot = tp->slots + args->sa_slotid;
+	tp = slot->table;
+	session = tp->session;
 
 	encode_op_hdr(xdr, OP_SEQUENCE, decode_sequence_maxsz, hdr);
 
@@ -1858,12 +1856,12 @@ static void encode_sequence(struct xdr_stream *xdr,
 		((u32 *)session->sess_id.data)[1],
 		((u32 *)session->sess_id.data)[2],
 		((u32 *)session->sess_id.data)[3],
-		slot->seq_nr, args->sa_slotid,
+		slot->seq_nr, slot->slot_nr,
 		tp->highest_used_slotid, args->sa_cache_this);
 	p = reserve_space(xdr, NFS4_MAX_SESSIONID_LEN + 16);
 	p = xdr_encode_opaque_fixed(p, session->sess_id.data, NFS4_MAX_SESSIONID_LEN);
 	*p++ = cpu_to_be32(slot->seq_nr);
-	*p++ = cpu_to_be32(args->sa_slotid);
+	*p++ = cpu_to_be32(slot->slot_nr);
 	*p++ = cpu_to_be32(tp->highest_used_slotid);
 	*p = cpu_to_be32(args->sa_cache_this);
 #endif /* CONFIG_NFS_V4_1 */
@@ -2025,8 +2023,9 @@ static void encode_free_stateid(struct xdr_stream *xdr,
 static u32 nfs4_xdr_minorversion(const struct nfs4_sequence_args *args)
 {
 #if defined(CONFIG_NFS_V4_1)
-	if (args->sa_session)
-		return args->sa_session->clp->cl_mvops->minor_version;
+
+	if (args->sa_slot)
+		return args->sa_slot->table->session->clp->cl_mvops->minor_version;
 #endif /* CONFIG_NFS_V4_1 */
 	return 0;
 }
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index 9c9b76c..deb31bb 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -194,8 +194,7 @@ struct nfs4_slot {
 };
 
 struct nfs4_sequence_args {
-	struct nfs4_session	*sa_session;
-	u32			sa_slotid;
+	struct nfs4_slot	*sa_slot;
 	u8			sa_cache_this;
 };
 
-- 
1.7.11.7

[PATCH 10/20] NFSv4.1: Simplify slot allocation

[Trond Myklebust]

Clean up the NFSv4.1 slot allocation by replacing nfs_find_slot() with
a function nfs_alloc_slot() that returns a pointer to the nfs4_slot
instead of an offset into the slot table.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c | 29 ++++++++++++++---------------
 1 file changed, 14 insertions(+), 15 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 7c2b037..59cef54 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -557,20 +557,18 @@ static int nfs4_sequence_done(struct rpc_task *task,
 }
 
 /*
- * nfs4_find_slot - efficiently look for a free slot
+ * nfs4_alloc_slot - efficiently look for a free slot
  *
- * nfs4_find_slot looks for an unset bit in the used_slots bitmap.
+ * nfs4_alloc_slot looks for an unset bit in the used_slots bitmap.
  * If found, we mark the slot as used, update the highest_used_slotid,
  * and respectively set up the sequence operation args.
- * The slot number is returned if found, or NFS4_NO_SLOT otherwise.
  *
  * Note: must be called with under the slot_tbl_lock.
  */
-static u32
-nfs4_find_slot(struct nfs4_slot_table *tbl)
+static struct nfs4_slot *nfs4_alloc_slot(struct nfs4_slot_table *tbl)
 {
+	struct nfs4_slot *ret = NULL;
 	u32 slotid;
-	u32 ret_id = NFS4_NO_SLOT;
 
 	dprintk("--> %s used_slots=%04lx highest_used=%u max_slots=%u\n",
 		__func__, tbl->used_slots[0], tbl->highest_used_slotid,
@@ -582,11 +580,14 @@ nfs4_find_slot(struct nfs4_slot_table *tbl)
 	if (slotid > tbl->highest_used_slotid ||
 			tbl->highest_used_slotid == NFS4_NO_SLOT)
 		tbl->highest_used_slotid = slotid;
-	ret_id = slotid;
+	ret = &tbl->slots[slotid];
+	ret->renewal_time = jiffies;
+
 out:
 	dprintk("<-- %s used_slots=%04lx highest_used=%d slotid=%d \n",
-		__func__, tbl->used_slots[0], tbl->highest_used_slotid, ret_id);
-	return ret_id;
+		__func__, tbl->used_slots[0], tbl->highest_used_slotid,
+		ret ? ret->slot_nr : -1);
+	return ret;
 }
 
 static void nfs41_init_sequence(struct nfs4_sequence_args *args,
@@ -606,7 +607,6 @@ int nfs41_setup_sequence(struct nfs4_session *session,
 {
 	struct nfs4_slot *slot;
 	struct nfs4_slot_table *tbl;
-	u32 slotid;
 
 	dprintk("--> %s\n", __func__);
 	/* slot already allocated? */
@@ -633,8 +633,8 @@ int nfs41_setup_sequence(struct nfs4_session *session,
 		return -EAGAIN;
 	}
 
-	slotid = nfs4_find_slot(tbl);
-	if (slotid == NFS4_NO_SLOT) {
+	slot = nfs4_alloc_slot(tbl);
+	if (slot == NULL) {
 		rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
 		spin_unlock(&tbl->slot_tbl_lock);
 		dprintk("<-- %s: no free slots\n", __func__);
@@ -643,12 +643,11 @@ int nfs41_setup_sequence(struct nfs4_session *session,
 	spin_unlock(&tbl->slot_tbl_lock);
 
 	rpc_task_set_priority(task, RPC_PRIORITY_NORMAL);
-	slot = tbl->slots + slotid;
-	slot->renewal_time = jiffies;
 
 	args->sa_slot = slot;
 
-	dprintk("<-- %s slotid=%d seqid=%d\n", __func__, slotid, slot->seq_nr);
+	dprintk("<-- %s slotid=%d seqid=%d\n", __func__,
+			slot->slot_nr, slot->seq_nr);
 
 	res->sr_slot = slot;
 	res->sr_status_flags = 0;
-- 
1.7.11.7

[PATCH 11/20] NFSv4.1: Clean up nfs4_free_slot

[Trond Myklebust]

Change the argument to take the pointer to the slot, instead of
just the slotid.

We know that the new value of highest_used_slot must be less than
the current value. No need to scan the whole table.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 59cef54..14389f2 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -412,16 +412,18 @@ static void renew_lease(const struct nfs_server *server, unsigned long timestamp
  * Must be called while holding tbl->slot_tbl_lock
  */
 static void
-nfs4_free_slot(struct nfs4_slot_table *tbl, u32 slotid)
+nfs4_free_slot(struct nfs4_slot_table *tbl, struct nfs4_slot *slot)
 {
+	u32 slotid = slot->slot_nr;
+
 	/* clear used bit in bitmap */
 	__clear_bit(slotid, tbl->used_slots);
 
 	/* update highest_used_slotid when it is freed */
 	if (slotid == tbl->highest_used_slotid) {
-		slotid = find_last_bit(tbl->used_slots, tbl->max_slots);
-		if (slotid < tbl->max_slots)
-			tbl->highest_used_slotid = slotid;
+		u32 new_max = find_last_bit(tbl->used_slots, slotid);
+		if (new_max < slotid)
+			tbl->highest_used_slotid = new_max;
 		else
 			tbl->highest_used_slotid = NFS4_NO_SLOT;
 	}
@@ -481,7 +483,7 @@ static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
 	tbl = &session->fc_slot_table;
 
 	spin_lock(&tbl->slot_tbl_lock);
-	nfs4_free_slot(tbl, res->sr_slot - tbl->slots);
+	nfs4_free_slot(tbl, res->sr_slot);
 	nfs4_check_drain_fc_complete(session);
 	spin_unlock(&tbl->slot_tbl_lock);
 	res->sr_slot = NULL;
-- 
1.7.11.7

[PATCH 12/20] NFSv4.1: Ensure that the client tracks the server target_highest_slotid

[Trond Myklebust]

Dynamic slot allocation in NFSv4.1 depends on the client being able to
track the server's target value for the highest slotid in the
slot table.  See the reference in Section 2.10.6.1 of RFC5661.

To avoid ordering problems in the case where 2 SEQUENCE replies contain
conflicting updates to this target value, we also introduce a generation
counter, to track whether or not an RPC containing a SEQUENCE operation
was launched before or after the last update.

Also rename the nfs4_slot_table target_max_slots field to
'target_highest_slotid' to avoid confusion with a slot
table size or number of slots.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c         | 25 +++++++++++++++++++++++++
 fs/nfs/nfs4state.c        |  7 +++----
 fs/nfs/nfs4xdr.c          |  4 ++--
 include/linux/nfs_fs_sb.h |  5 +++--
 include/linux/nfs_xdr.h   |  2 ++
 5 files changed, 35 insertions(+), 8 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 14389f2..8333260 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -489,6 +489,28 @@ static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
 	res->sr_slot = NULL;
 }
 
+/* Update the client's idea of target_highest_slotid */
+static void nfs41_set_target_slotid_locked(struct nfs4_slot_table *tbl,
+		u32 target_highest_slotid)
+{
+	if (tbl->target_highest_slotid == target_highest_slotid)
+		return;
+	tbl->target_highest_slotid = target_highest_slotid;
+	tbl->generation++;
+}
+
+static void nfs41_update_target_slotid(struct nfs4_slot_table *tbl,
+		struct nfs4_slot *slot,
+		struct nfs4_sequence_res *res)
+{
+	spin_lock(&tbl->slot_tbl_lock);
+	if (tbl->generation != slot->generation)
+		goto out;
+	nfs41_set_target_slotid_locked(tbl, res->sr_target_highest_slotid);
+out:
+	spin_unlock(&tbl->slot_tbl_lock);
+}
+
 static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
 {
 	struct nfs4_session *session;
@@ -523,6 +545,7 @@ static int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *
 		/* Check sequence flags */
 		if (res->sr_status_flags != 0)
 			nfs4_schedule_lease_recovery(clp);
+		nfs41_update_target_slotid(slot->table, slot, res);
 		break;
 	case -NFS4ERR_DELAY:
 		/* The server detected a resend of the RPC call and
@@ -584,6 +607,7 @@ static struct nfs4_slot *nfs4_alloc_slot(struct nfs4_slot_table *tbl)
 		tbl->highest_used_slotid = slotid;
 	ret = &tbl->slots[slotid];
 	ret->renewal_time = jiffies;
+	ret->generation = tbl->generation;
 
 out:
 	dprintk("<-- %s used_slots=%04lx highest_used=%d slotid=%d \n",
@@ -5694,6 +5718,7 @@ static void nfs4_add_and_init_slots(struct nfs4_slot_table *tbl,
 		tbl->max_slots = max_slots;
 	}
 	tbl->highest_used_slotid = NFS4_NO_SLOT;
+	tbl->target_highest_slotid = max_slots - 1;
 	for (i = 0; i < tbl->max_slots; i++)
 		tbl->slots[i].seq_nr = ivalue;
 	spin_unlock(&tbl->slot_tbl_lock);
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 9495789..842cb8c 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -2033,17 +2033,16 @@ static int nfs4_recall_slot(struct nfs_client *clp)
 		return 0;
 	nfs4_begin_drain_session(clp);
 	fc_tbl = &clp->cl_session->fc_slot_table;
-	new = nfs4_alloc_slots(fc_tbl, fc_tbl->target_max_slots, GFP_NOFS);
+	new = nfs4_alloc_slots(fc_tbl, fc_tbl->target_highest_slotid + 1, GFP_NOFS);
         if (!new)
 		return -ENOMEM;
 
 	spin_lock(&fc_tbl->slot_tbl_lock);
-	for (i = 0; i < fc_tbl->target_max_slots; i++)
+	for (i = 0; i <= fc_tbl->target_highest_slotid; i++)
 		new[i].seq_nr = fc_tbl->slots[i].seq_nr;
 	old = fc_tbl->slots;
 	fc_tbl->slots = new;
-	fc_tbl->max_slots = fc_tbl->target_max_slots;
-	fc_tbl->target_max_slots = 0;
+	fc_tbl->max_slots = fc_tbl->target_highest_slotid + 1;
 	clp->cl_session->fc_attrs.max_reqs = fc_tbl->max_slots;
 	spin_unlock(&fc_tbl->slot_tbl_lock);
 
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 27b0fec..05d34f1 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -5552,8 +5552,8 @@ static int decode_sequence(struct xdr_stream *xdr,
 	}
 	/* highest slot id - currently not processed */
 	dummy = be32_to_cpup(p++);
-	/* target highest slot id - currently not processed */
-	dummy = be32_to_cpup(p++);
+	/* target highest slot id */
+	res->sr_target_highest_slotid = be32_to_cpup(p++);
 	/* result flags */
 	res->sr_status_flags = be32_to_cpup(p);
 	status = 0;
diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h
index b041287..57d4069 100644
--- a/include/linux/nfs_fs_sb.h
+++ b/include/linux/nfs_fs_sb.h
@@ -217,8 +217,9 @@ struct nfs4_slot_table {
 	u32		max_slots;		/* # slots in table */
 	u32		highest_used_slotid;	/* sent to server on each SEQ.
 						 * op for dynamic resizing */
-	u32		target_max_slots;	/* Set by CB_RECALL_SLOT as
-						 * the new max_slots */
+	u32		target_highest_slotid;	/* Server max_slot target */
+	unsigned long	generation;		/* Generation counter for
+						   target_highest_slotid */
 	struct completion complete;
 };
 
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index deb31bb..08c47db 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -188,6 +188,7 @@ struct nfs4_channel_attrs {
 /* nfs41 sessions slot seqid */
 struct nfs4_slot {
 	struct nfs4_slot_table	*table;
+	unsigned long		generation;
 	unsigned long		renewal_time;
 	u32			slot_nr;
 	u32		 	seq_nr;
@@ -202,6 +203,7 @@ struct nfs4_sequence_res {
 	struct nfs4_slot	*sr_slot;	/* slot used to send request */
 	int			sr_status;	/* sequence operation status */
 	u32			sr_status_flags;
+	u32			sr_target_highest_slotid;
 };
 
 struct nfs4_get_lease_time_args {
-- 
1.7.11.7

[PATCH 13/20] NFSv4.1: Reset the sequence number for slots that have been deallocated

[Trond Myklebust]

When the server tells us that it is dynamically resizing the session
replay cache, we should reset the sequence number for those slots
that have been deallocated.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c         | 18 ++++++++++++++++++
 fs/nfs/nfs4xdr.c          |  4 ++--
 include/linux/nfs_fs_sb.h |  1 +
 include/linux/nfs_xdr.h   |  1 +
 4 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 8333260..fc58ad8 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -499,6 +499,22 @@ static void nfs41_set_target_slotid_locked(struct nfs4_slot_table *tbl,
 	tbl->generation++;
 }
 
+static void nfs41_set_server_slotid_locked(struct nfs4_slot_table *tbl,
+		u32 highest_slotid)
+{
+	unsigned int max_slotid, i;
+
+	if (tbl->server_highest_slotid == highest_slotid)
+		return;
+	if (tbl->highest_used_slotid > highest_slotid)
+		return;
+	max_slotid = min(tbl->max_slots - 1, highest_slotid);
+	/* Reset the seq_nr for deallocated slots */
+	for (i = tbl->server_highest_slotid + 1; i <= max_slotid; i++)
+		tbl->slots[i].seq_nr = 1;
+	tbl->server_highest_slotid = highest_slotid;
+}
+
 static void nfs41_update_target_slotid(struct nfs4_slot_table *tbl,
 		struct nfs4_slot *slot,
 		struct nfs4_sequence_res *res)
@@ -506,6 +522,7 @@ static void nfs41_update_target_slotid(struct nfs4_slot_table *tbl,
 	spin_lock(&tbl->slot_tbl_lock);
 	if (tbl->generation != slot->generation)
 		goto out;
+	nfs41_set_server_slotid_locked(tbl, res->sr_highest_slotid);
 	nfs41_set_target_slotid_locked(tbl, res->sr_target_highest_slotid);
 out:
 	spin_unlock(&tbl->slot_tbl_lock);
@@ -5719,6 +5736,7 @@ static void nfs4_add_and_init_slots(struct nfs4_slot_table *tbl,
 	}
 	tbl->highest_used_slotid = NFS4_NO_SLOT;
 	tbl->target_highest_slotid = max_slots - 1;
+	tbl->server_highest_slotid = max_slots - 1;
 	for (i = 0; i < tbl->max_slots; i++)
 		tbl->slots[i].seq_nr = ivalue;
 	spin_unlock(&tbl->slot_tbl_lock);
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 05d34f1..a67040f 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -5550,8 +5550,8 @@ static int decode_sequence(struct xdr_stream *xdr,
 		dprintk("%s Invalid slot id\n", __func__);
 		goto out_err;
 	}
-	/* highest slot id - currently not processed */
-	dummy = be32_to_cpup(p++);
+	/* highest slot id */
+	res->sr_highest_slotid = be32_to_cpup(p++);
 	/* target highest slot id */
 	res->sr_target_highest_slotid = be32_to_cpup(p++);
 	/* result flags */
diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h
index 57d4069..646e64b 100644
--- a/include/linux/nfs_fs_sb.h
+++ b/include/linux/nfs_fs_sb.h
@@ -218,6 +218,7 @@ struct nfs4_slot_table {
 	u32		highest_used_slotid;	/* sent to server on each SEQ.
 						 * op for dynamic resizing */
 	u32		target_highest_slotid;	/* Server max_slot target */
+	u32		server_highest_slotid;	/* Server highest slotid */
 	unsigned long	generation;		/* Generation counter for
 						   target_highest_slotid */
 	struct completion complete;
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index 08c47db..3ddb08f 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -203,6 +203,7 @@ struct nfs4_sequence_res {
 	struct nfs4_slot	*sr_slot;	/* slot used to send request */
 	int			sr_status;	/* sequence operation status */
 	u32			sr_status_flags;
+	u32			sr_highest_slotid;
 	u32			sr_target_highest_slotid;
 };
 
-- 
1.7.11.7

[PATCH 14/20] NFSv4.1: Fix nfs4_callback_recallslot to work with dynamic slot allocation

[Trond Myklebust]

Ensure that the NFSv4.1 CB_RECALL_SLOT callback updates the slot table
target max slotid safely.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/callback_proc.c | 2 +-
 fs/nfs/nfs4_fs.h       | 2 ++
 fs/nfs/nfs4proc.c      | 8 ++++++++
 3 files changed, 11 insertions(+), 1 deletion(-)

diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index 0be08b9..15b9879 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -576,7 +576,7 @@ __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 	if (args->crsa_target_max_slots == fc_tbl->max_slots)
 		goto out;
 
-	fc_tbl->target_max_slots = args->crsa_target_max_slots;
+	nfs41_set_target_slotid(fc_tbl, args->crsa_target_max_slots);
 	nfs41_handle_recall_slot(cps->clp);
 out:
 	dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index 42c5869..5d4e82b 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -260,6 +260,8 @@ extern int nfs4_proc_layoutcommit(struct nfs4_layoutcommit_data *data,
 
 extern struct nfs4_slot *nfs4_alloc_slots(struct nfs4_slot_table *table,
 		u32 max_slots, gfp_t gfp_flags);
+extern void nfs41_set_target_slotid(struct nfs4_slot_table *tbl,
+		u32 target_highest_slotid);
 
 static inline bool
 is_ds_only_client(struct nfs_client *clp)
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index fc58ad8..731a6fd 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -499,6 +499,14 @@ static void nfs41_set_target_slotid_locked(struct nfs4_slot_table *tbl,
 	tbl->generation++;
 }
 
+void nfs41_set_target_slotid(struct nfs4_slot_table *tbl,
+		u32 target_highest_slotid)
+{
+	spin_lock(&tbl->slot_tbl_lock);
+	nfs41_set_target_slotid_locked(tbl, target_highest_slotid);
+	spin_unlock(&tbl->slot_tbl_lock);
+}
+
 static void nfs41_set_server_slotid_locked(struct nfs4_slot_table *tbl,
 		u32 highest_slotid)
 {
-- 
1.7.11.7

[PATCH 15/20] NFSv4.1: Don't confuse target_highest_slotid and max_slots in cb_recall_slot

[Trond Myklebust]

Don't confuse the table size and the target_highest_slotid...

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/callback.h      |  2 +-
 fs/nfs/callback_proc.c | 12 +++++-------
 fs/nfs/callback_xdr.c  |  2 +-
 3 files changed, 7 insertions(+), 9 deletions(-)

diff --git a/fs/nfs/callback.h b/fs/nfs/callback.h
index 4251c2a..e75631e 100644
--- a/fs/nfs/callback.h
+++ b/fs/nfs/callback.h
@@ -142,7 +142,7 @@ extern __be32 nfs4_callback_recallany(struct cb_recallanyargs *args,
 
 struct cb_recallslotargs {
 	struct sockaddr	*crsa_addr;
-	uint32_t	crsa_target_max_slots;
+	uint32_t	crsa_target_highest_slotid;
 };
 extern __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args,
 					 void *dummy,
diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index 15b9879..ed0b446 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -561,22 +561,20 @@ __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 	if (!cps->clp) /* set in cb_sequence */
 		goto out;
 
-	dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target max slots %d\n",
+	dprintk_rcu("NFS: CB_RECALL_SLOT request from %s target highest slotid %d\n",
 		rpc_peeraddr2str(cps->clp->cl_rpcclient, RPC_DISPLAY_ADDR),
-		args->crsa_target_max_slots);
+		args->crsa_target_highest_slotid);
 
 	fc_tbl = &cps->clp->cl_session->fc_slot_table;
 
 	status = htonl(NFS4ERR_BAD_HIGH_SLOT);
-	if (args->crsa_target_max_slots > fc_tbl->max_slots ||
-	    args->crsa_target_max_slots < 1)
+	if (args->crsa_target_highest_slotid >= fc_tbl->max_slots ||
+	    args->crsa_target_highest_slotid < 1)
 		goto out;
 
 	status = htonl(NFS4_OK);
-	if (args->crsa_target_max_slots == fc_tbl->max_slots)
-		goto out;
 
-	nfs41_set_target_slotid(fc_tbl, args->crsa_target_max_slots);
+	nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
 	nfs41_handle_recall_slot(cps->clp);
 out:
 	dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
diff --git a/fs/nfs/callback_xdr.c b/fs/nfs/callback_xdr.c
index 742ff4f..81e8c7d 100644
--- a/fs/nfs/callback_xdr.c
+++ b/fs/nfs/callback_xdr.c
@@ -520,7 +520,7 @@ static __be32 decode_recallslot_args(struct svc_rqst *rqstp,
 	p = read_buf(xdr, 4);
 	if (unlikely(p == NULL))
 		return htonl(NFS4ERR_BADXDR);
-	args->crsa_target_max_slots = ntohl(*p++);
+	args->crsa_target_highest_slotid = ntohl(*p++);
 	return 0;
 }
 
-- 
1.7.11.7

[PATCH 16/20] NFSv4.1: Allow the server to recall all but one slot

[Trond Myklebust]

If the server wants to leave us with only one slot, or it wants
to "shrink" our slot table to something larger than we have now,
then so be it.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/callback_proc.c | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index ed0b446..a0546ec 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -567,11 +567,6 @@ __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 
 	fc_tbl = &cps->clp->cl_session->fc_slot_table;
 
-	status = htonl(NFS4ERR_BAD_HIGH_SLOT);
-	if (args->crsa_target_highest_slotid >= fc_tbl->max_slots ||
-	    args->crsa_target_highest_slotid < 1)
-		goto out;
-
 	status = htonl(NFS4_OK);
 
 	nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
-- 
1.7.11.7

[PATCH 17/20] NFSv4.1: Support dynamic resizing of the session slot table

[Trond Myklebust]

Allow the server to control the size of the session slot table
by adjusting the value of sr_target_max_slots in the reply to the
SEQUENCE operation.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4proc.c         | 12 ++++++++++--
 fs/nfs/nfs4state.c        |  6 +++---
 include/linux/nfs_fs_sb.h |  1 +
 3 files changed, 14 insertions(+), 5 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 731a6fd..5641380 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -493,10 +493,17 @@ static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
 static void nfs41_set_target_slotid_locked(struct nfs4_slot_table *tbl,
 		u32 target_highest_slotid)
 {
+	unsigned int max_slotid, i;
+
 	if (tbl->target_highest_slotid == target_highest_slotid)
 		return;
 	tbl->target_highest_slotid = target_highest_slotid;
 	tbl->generation++;
+
+	max_slotid = min(tbl->max_slots - 1, tbl->target_highest_slotid);
+	for (i = tbl->max_slotid + 1; i <= max_slotid; i++)
+		rpc_wake_up_next(&tbl->slot_tbl_waitq);
+	tbl->max_slotid = max_slotid;
 }
 
 void nfs41_set_target_slotid(struct nfs4_slot_table *tbl,
@@ -623,8 +630,8 @@ static struct nfs4_slot *nfs4_alloc_slot(struct nfs4_slot_table *tbl)
 	dprintk("--> %s used_slots=%04lx highest_used=%u max_slots=%u\n",
 		__func__, tbl->used_slots[0], tbl->highest_used_slotid,
 		tbl->max_slots);
-	slotid = find_first_zero_bit(tbl->used_slots, tbl->max_slots);
-	if (slotid >= tbl->max_slots)
+	slotid = find_first_zero_bit(tbl->used_slots, tbl->max_slotid + 1);
+	if (slotid > tbl->max_slotid)
 		goto out;
 	__set_bit(slotid, tbl->used_slots);
 	if (slotid > tbl->highest_used_slotid ||
@@ -5745,6 +5752,7 @@ static void nfs4_add_and_init_slots(struct nfs4_slot_table *tbl,
 	tbl->highest_used_slotid = NFS4_NO_SLOT;
 	tbl->target_highest_slotid = max_slots - 1;
 	tbl->server_highest_slotid = max_slots - 1;
+	tbl->max_slotid = max_slots - 1;
 	for (i = 0; i < tbl->max_slots; i++)
 		tbl->slots[i].seq_nr = ivalue;
 	spin_unlock(&tbl->slot_tbl_lock);
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 842cb8c..1b7fa73 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -254,15 +254,14 @@ static void nfs4_end_drain_session(struct nfs_client *clp)
 {
 	struct nfs4_session *ses = clp->cl_session;
 	struct nfs4_slot_table *tbl;
-	int max_slots;
+	unsigned int i;
 
 	if (ses == NULL)
 		return;
 	tbl = &ses->fc_slot_table;
 	if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
 		spin_lock(&tbl->slot_tbl_lock);
-		max_slots = tbl->max_slots;
-		while (max_slots--) {
+		for (i = 0; i <= tbl->max_slotid; i++) {
 			if (rpc_wake_up_first(&tbl->slot_tbl_waitq,
 						nfs4_set_task_privileged,
 						NULL) == NULL)
@@ -2043,6 +2042,7 @@ static int nfs4_recall_slot(struct nfs_client *clp)
 	old = fc_tbl->slots;
 	fc_tbl->slots = new;
 	fc_tbl->max_slots = fc_tbl->target_highest_slotid + 1;
+	fc_tbl->max_slotid = fc_tbl->target_highest_slotid;
 	clp->cl_session->fc_attrs.max_reqs = fc_tbl->max_slots;
 	spin_unlock(&fc_tbl->slot_tbl_lock);
 
diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h
index 646e64b..3071550 100644
--- a/include/linux/nfs_fs_sb.h
+++ b/include/linux/nfs_fs_sb.h
@@ -215,6 +215,7 @@ struct nfs4_slot_table {
 	spinlock_t	slot_tbl_lock;
 	struct rpc_wait_queue	slot_tbl_waitq;	/* allocators may wait here */
 	u32		max_slots;		/* # slots in table */
+	u32		max_slotid;		/* Max allowed slotid value */
 	u32		highest_used_slotid;	/* sent to server on each SEQ.
 						 * op for dynamic resizing */
 	u32		target_highest_slotid;	/* Server max_slot target */
-- 
1.7.11.7

[PATCH 18/20] NFSv4.1: Allow SEQUENCE to resize the slot table on the fly

[Trond Myklebust]

Instead of an array of slots, use a singly linked list of slots that
can be dynamically appended to or shrunk.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/nfs4_fs.h        |   4 +-
 fs/nfs/nfs4proc.c       | 174 +++++++++++++++++++++++++++++++-----------------
 fs/nfs/nfs4state.c      |  22 ++----
 include/linux/nfs_xdr.h |   1 +
 4 files changed, 121 insertions(+), 80 deletions(-)

diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index 5d4e82b..856bc49 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -258,10 +258,10 @@ extern int nfs4_proc_get_lease_time(struct nfs_client *clp,
 extern int nfs4_proc_layoutcommit(struct nfs4_layoutcommit_data *data,
 				  bool sync);
 
-extern struct nfs4_slot *nfs4_alloc_slots(struct nfs4_slot_table *table,
-		u32 max_slots, gfp_t gfp_flags);
 extern void nfs41_set_target_slotid(struct nfs4_slot_table *tbl,
 		u32 target_highest_slotid);
+extern int nfs4_resize_slot_table(struct nfs4_slot_table *tbl,
+		u32 max_reqs, u32 ivalue);
 
 static inline bool
 is_ds_only_client(struct nfs_client *clp)
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 5641380..adedcfa 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -397,6 +397,27 @@ static void renew_lease(const struct nfs_server *server, unsigned long timestamp
 #if defined(CONFIG_NFS_V4_1)
 
 /*
+ * nfs4_shrink_slot_table - free retired slots from the slot table
+ */
+static void nfs4_shrink_slot_table(struct nfs4_slot_table  *tbl, u32 newsize)
+{
+	struct nfs4_slot **p;
+	if (newsize >= tbl->max_slots)
+		return;
+
+	p = &tbl->slots;
+	while (newsize--)
+		p = &(*p)->next;
+	while (*p) {
+		struct nfs4_slot *slot = *p;
+
+		*p = slot->next;
+		kfree(slot);
+		tbl->max_slots--;
+	}
+}
+
+/*
  * nfs4_free_slot - free a slot and efficiently update slot table.
  *
  * freeing a slot is trivially done by clearing its respective bit
@@ -500,7 +521,7 @@ static void nfs41_set_target_slotid_locked(struct nfs4_slot_table *tbl,
 	tbl->target_highest_slotid = target_highest_slotid;
 	tbl->generation++;
 
-	max_slotid = min(tbl->max_slots - 1, tbl->target_highest_slotid);
+	max_slotid = min(NFS4_MAX_SLOT_TABLE - 1, tbl->target_highest_slotid);
 	for (i = tbl->max_slotid + 1; i <= max_slotid; i++)
 		rpc_wake_up_next(&tbl->slot_tbl_waitq);
 	tbl->max_slotid = max_slotid;
@@ -517,16 +538,12 @@ void nfs41_set_target_slotid(struct nfs4_slot_table *tbl,
 static void nfs41_set_server_slotid_locked(struct nfs4_slot_table *tbl,
 		u32 highest_slotid)
 {
-	unsigned int max_slotid, i;
-
 	if (tbl->server_highest_slotid == highest_slotid)
 		return;
 	if (tbl->highest_used_slotid > highest_slotid)
 		return;
-	max_slotid = min(tbl->max_slots - 1, highest_slotid);
-	/* Reset the seq_nr for deallocated slots */
-	for (i = tbl->server_highest_slotid + 1; i <= max_slotid; i++)
-		tbl->slots[i].seq_nr = 1;
+	/* Deallocate slots */
+	nfs4_shrink_slot_table(tbl, highest_slotid + 1);
 	tbl->server_highest_slotid = highest_slotid;
 }
 
@@ -613,6 +630,42 @@ static int nfs4_sequence_done(struct rpc_task *task,
 	return nfs41_sequence_done(task, res);
 }
 
+static struct nfs4_slot *nfs4_new_slot(struct nfs4_slot_table  *tbl,
+		u32 slotid, u32 seq_init, gfp_t gfp_mask)
+{
+	struct nfs4_slot *slot;
+
+	slot = kzalloc(sizeof(*slot), gfp_mask);
+	if (slot) {
+		slot->table = tbl;
+		slot->slot_nr = slotid;
+		slot->seq_nr = seq_init;
+	}
+	return slot;
+}
+
+static struct nfs4_slot *nfs4_find_or_create_slot(struct nfs4_slot_table  *tbl,
+		u32 slotid, u32 seq_init, gfp_t gfp_mask)
+{
+	struct nfs4_slot **p, *slot;
+
+	p = &tbl->slots;
+	for (;;) {
+		if (*p == NULL) {
+			*p = nfs4_new_slot(tbl, tbl->max_slots,
+					seq_init, gfp_mask);
+			if (*p == NULL)
+				break;
+			tbl->max_slots++;
+		}
+		slot = *p;
+		if (slot->slot_nr == slotid)
+			return slot;
+		p = &slot->next;
+	}
+	return NULL;
+}
+
 /*
  * nfs4_alloc_slot - efficiently look for a free slot
  *
@@ -629,15 +682,17 @@ static struct nfs4_slot *nfs4_alloc_slot(struct nfs4_slot_table *tbl)
 
 	dprintk("--> %s used_slots=%04lx highest_used=%u max_slots=%u\n",
 		__func__, tbl->used_slots[0], tbl->highest_used_slotid,
-		tbl->max_slots);
+		tbl->max_slotid + 1);
 	slotid = find_first_zero_bit(tbl->used_slots, tbl->max_slotid + 1);
 	if (slotid > tbl->max_slotid)
 		goto out;
+	ret = nfs4_find_or_create_slot(tbl, slotid, 1, GFP_NOWAIT);
+	if (ret == NULL)
+		goto out;
 	__set_bit(slotid, tbl->used_slots);
 	if (slotid > tbl->highest_used_slotid ||
 			tbl->highest_used_slotid == NFS4_NO_SLOT)
 		tbl->highest_used_slotid = slotid;
-	ret = &tbl->slots[slotid];
 	ret->renewal_time = jiffies;
 	ret->generation = tbl->generation;
 
@@ -5719,67 +5774,56 @@ int nfs4_proc_get_lease_time(struct nfs_client *clp, struct nfs_fsinfo *fsinfo)
 	return status;
 }
 
-struct nfs4_slot *nfs4_alloc_slots(struct nfs4_slot_table *table,
-		u32 max_slots, gfp_t gfp_flags)
+static int nfs4_grow_slot_table(struct nfs4_slot_table *tbl,
+		 u32 max_reqs, u32 ivalue)
 {
-	struct nfs4_slot *tbl;
-	u32 i;
-
-	tbl = kmalloc_array(max_slots, sizeof(*tbl), gfp_flags);
-	if (tbl != NULL) {
-		for (i = 0; i < max_slots; i++) {
-			tbl[i].table = table;
-			tbl[i].slot_nr = i;
-		}
-	}
-	return tbl;
+	if (max_reqs <= tbl->max_slots)
+		return 0;
+	if (nfs4_find_or_create_slot(tbl, max_reqs - 1, ivalue, GFP_NOFS))
+		return 0;
+	return -ENOMEM;
 }
 
-static void nfs4_add_and_init_slots(struct nfs4_slot_table *tbl,
-		struct nfs4_slot *new,
-		u32 max_slots,
+static void nfs4_reset_slot_table(struct nfs4_slot_table *tbl,
+		u32 server_highest_slotid,
 		u32 ivalue)
 {
-	struct nfs4_slot *old = NULL;
-	u32 i;
+	struct nfs4_slot **p;
 
-	spin_lock(&tbl->slot_tbl_lock);
-	if (new) {
-		old = tbl->slots;
-		tbl->slots = new;
-		tbl->max_slots = max_slots;
+	nfs4_shrink_slot_table(tbl, server_highest_slotid + 1);
+	p = &tbl->slots;
+	while (*p) {
+		(*p)->seq_nr = ivalue;
+		p = &(*p)->next;
 	}
 	tbl->highest_used_slotid = NFS4_NO_SLOT;
-	tbl->target_highest_slotid = max_slots - 1;
-	tbl->server_highest_slotid = max_slots - 1;
-	tbl->max_slotid = max_slots - 1;
-	for (i = 0; i < tbl->max_slots; i++)
-		tbl->slots[i].seq_nr = ivalue;
-	spin_unlock(&tbl->slot_tbl_lock);
-	kfree(old);
+	tbl->target_highest_slotid = server_highest_slotid;
+	tbl->server_highest_slotid = server_highest_slotid;
+	tbl->max_slotid = server_highest_slotid;
 }
 
 /*
  * (re)Initialise a slot table
  */
-static int nfs4_realloc_slot_table(struct nfs4_slot_table *tbl, u32 max_reqs,
-				 u32 ivalue)
+static int nfs4_realloc_slot_table(struct nfs4_slot_table *tbl,
+		u32 max_reqs, u32 ivalue)
 {
-	struct nfs4_slot *new = NULL;
-	int ret = -ENOMEM;
+	int ret;
 
 	dprintk("--> %s: max_reqs=%u, tbl->max_slots %d\n", __func__,
 		max_reqs, tbl->max_slots);
 
-	/* Does the newly negotiated max_reqs match the existing slot table? */
-	if (max_reqs != tbl->max_slots) {
-		new = nfs4_alloc_slots(tbl, max_reqs, GFP_NOFS);
-		if (!new)
-			goto out;
-	}
-	ret = 0;
+	if (max_reqs > NFS4_MAX_SLOT_TABLE)
+		max_reqs = NFS4_MAX_SLOT_TABLE;
+
+	ret = nfs4_grow_slot_table(tbl, max_reqs, ivalue);
+	if (ret)
+		goto out;
+
+	spin_lock(&tbl->slot_tbl_lock);
+	nfs4_reset_slot_table(tbl, max_reqs - 1, ivalue);
+	spin_unlock(&tbl->slot_tbl_lock);
 
-	nfs4_add_and_init_slots(tbl, new, max_reqs, ivalue);
 	dprintk("%s: tbl=%p slots=%p max_slots=%d\n", __func__,
 		tbl, tbl->slots, tbl->max_slots);
 out:
@@ -5787,18 +5831,28 @@ out:
 	return ret;
 }
 
+int nfs4_resize_slot_table(struct nfs4_slot_table *tbl,
+		 u32 max_reqs, u32 ivalue)
+{
+	int ret;
+
+	if (max_reqs > NFS4_MAX_SLOT_TABLE)
+		max_reqs = NFS4_MAX_SLOT_TABLE;
+	ret = nfs4_grow_slot_table(tbl, max_reqs, ivalue);
+	if (ret)
+		return ret;
+	spin_lock(&tbl->slot_tbl_lock);
+	nfs4_shrink_slot_table(tbl, max_reqs);
+	tbl->max_slotid = max_reqs - 1;
+	spin_unlock(&tbl->slot_tbl_lock);
+	return 0;
+}
+
 /* Destroy the slot table */
 static void nfs4_destroy_slot_tables(struct nfs4_session *session)
 {
-	if (session->fc_slot_table.slots != NULL) {
-		kfree(session->fc_slot_table.slots);
-		session->fc_slot_table.slots = NULL;
-	}
-	if (session->bc_slot_table.slots != NULL) {
-		kfree(session->bc_slot_table.slots);
-		session->bc_slot_table.slots = NULL;
-	}
-	return;
+	nfs4_shrink_slot_table(&session->fc_slot_table, 0);
+	nfs4_shrink_slot_table(&session->bc_slot_table, 0);
 }
 
 /*
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 1b7fa73..c14b2c7 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -2025,29 +2025,15 @@ out:
 static int nfs4_recall_slot(struct nfs_client *clp)
 {
 	struct nfs4_slot_table *fc_tbl;
-	struct nfs4_slot *new, *old;
-	int i;
+	u32 new_size;
 
 	if (!nfs4_has_session(clp))
 		return 0;
 	nfs4_begin_drain_session(clp);
-	fc_tbl = &clp->cl_session->fc_slot_table;
-	new = nfs4_alloc_slots(fc_tbl, fc_tbl->target_highest_slotid + 1, GFP_NOFS);
-        if (!new)
-		return -ENOMEM;
 
-	spin_lock(&fc_tbl->slot_tbl_lock);
-	for (i = 0; i <= fc_tbl->target_highest_slotid; i++)
-		new[i].seq_nr = fc_tbl->slots[i].seq_nr;
-	old = fc_tbl->slots;
-	fc_tbl->slots = new;
-	fc_tbl->max_slots = fc_tbl->target_highest_slotid + 1;
-	fc_tbl->max_slotid = fc_tbl->target_highest_slotid;
-	clp->cl_session->fc_attrs.max_reqs = fc_tbl->max_slots;
-	spin_unlock(&fc_tbl->slot_tbl_lock);
-
-	kfree(old);
-	return 0;
+	fc_tbl = &clp->cl_session->fc_slot_table;
+	new_size = fc_tbl->server_highest_slotid + 1;
+	return nfs4_resize_slot_table(fc_tbl, new_size, 1);
 }
 
 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index 3ddb08f..44d256f 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -188,6 +188,7 @@ struct nfs4_channel_attrs {
 /* nfs41 sessions slot seqid */
 struct nfs4_slot {
 	struct nfs4_slot_table	*table;
+	struct nfs4_slot	*next;
 	unsigned long		generation;
 	unsigned long		renewal_time;
 	u32			slot_nr;
-- 
1.7.11.7

[PATCH 19/20] NFSv4.1: Remove the state manager code to resize the slot table

[Trond Myklebust]

The state manager no longer needs any special machinery to stop the
session flow and resize the slot table. It is all done on the fly by
the SEQUENCE op code now.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/callback_proc.c |  1 -
 fs/nfs/nfs4_fs.h       |  4 ----
 fs/nfs/nfs4proc.c      | 17 -----------------
 fs/nfs/nfs4state.c     | 33 ---------------------------------
 4 files changed, 55 deletions(-)

diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index a0546ec..8610bd1 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -570,7 +570,6 @@ __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 	status = htonl(NFS4_OK);
 
 	nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
-	nfs41_handle_recall_slot(cps->clp);
 out:
 	dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 	return status;
diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index 856bc49..fa1a055 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -21,7 +21,6 @@ enum nfs4_client_state {
 	NFS4CLNT_RECLAIM_NOGRACE,
 	NFS4CLNT_DELEGRETURN,
 	NFS4CLNT_SESSION_RESET,
-	NFS4CLNT_RECALL_SLOT,
 	NFS4CLNT_LEASE_CONFIRM,
 	NFS4CLNT_SERVER_SCOPE_MISMATCH,
 	NFS4CLNT_PURGE_STATE,
@@ -260,8 +259,6 @@ extern int nfs4_proc_layoutcommit(struct nfs4_layoutcommit_data *data,
 
 extern void nfs41_set_target_slotid(struct nfs4_slot_table *tbl,
 		u32 target_highest_slotid);
-extern int nfs4_resize_slot_table(struct nfs4_slot_table *tbl,
-		u32 max_reqs, u32 ivalue);
 
 static inline bool
 is_ds_only_client(struct nfs_client *clp)
@@ -358,7 +355,6 @@ extern void nfs4_schedule_state_manager(struct nfs_client *);
 extern void nfs4_schedule_path_down_recovery(struct nfs_client *clp);
 extern void nfs4_schedule_stateid_recovery(const struct nfs_server *, struct nfs4_state *);
 extern void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags);
-extern void nfs41_handle_recall_slot(struct nfs_client *clp);
 extern void nfs41_handle_server_scope(struct nfs_client *,
 				      struct nfs41_server_scope **);
 extern void nfs4_put_lock_state(struct nfs4_lock_state *lsp);
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index adedcfa..ec2b034 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -5831,23 +5831,6 @@ out:
 	return ret;
 }
 
-int nfs4_resize_slot_table(struct nfs4_slot_table *tbl,
-		 u32 max_reqs, u32 ivalue)
-{
-	int ret;
-
-	if (max_reqs > NFS4_MAX_SLOT_TABLE)
-		max_reqs = NFS4_MAX_SLOT_TABLE;
-	ret = nfs4_grow_slot_table(tbl, max_reqs, ivalue);
-	if (ret)
-		return ret;
-	spin_lock(&tbl->slot_tbl_lock);
-	nfs4_shrink_slot_table(tbl, max_reqs);
-	tbl->max_slotid = max_reqs - 1;
-	spin_unlock(&tbl->slot_tbl_lock);
-	return 0;
-}
-
 /* Destroy the slot table */
 static void nfs4_destroy_slot_tables(struct nfs4_session *session)
 {
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index c14b2c7..3940cd4 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -302,7 +302,6 @@ static void nfs41_finish_session_reset(struct nfs_client *clp)
 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
 	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
 	/* create_session negotiated new slot table */
-	clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
 	nfs41_setup_state_renewal(clp);
 }
@@ -1905,14 +1904,6 @@ void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
 }
 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
 
-void nfs41_handle_recall_slot(struct nfs_client *clp)
-{
-	set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
-	dprintk("%s: scheduling slot recall for server %s\n", __func__,
-			clp->cl_hostname);
-	nfs4_schedule_state_manager(clp);
-}
-
 static void nfs4_reset_all_state(struct nfs_client *clp)
 {
 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
@@ -2022,20 +2013,6 @@ out:
 	return status;
 }
 
-static int nfs4_recall_slot(struct nfs_client *clp)
-{
-	struct nfs4_slot_table *fc_tbl;
-	u32 new_size;
-
-	if (!nfs4_has_session(clp))
-		return 0;
-	nfs4_begin_drain_session(clp);
-
-	fc_tbl = &clp->cl_session->fc_slot_table;
-	new_size = fc_tbl->server_highest_slotid + 1;
-	return nfs4_resize_slot_table(fc_tbl, new_size, 1);
-}
-
 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
 {
 	struct rpc_cred *cred;
@@ -2066,7 +2043,6 @@ static int nfs4_bind_conn_to_session(struct nfs_client *clp)
 #else /* CONFIG_NFS_V4_1 */
 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
 static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
-static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
 
 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
 {
@@ -2126,15 +2102,6 @@ static void nfs4_state_manager(struct nfs_client *clp)
 			continue;
 		}
 
-		/* Recall session slots */
-		if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)) {
-			section = "recall slot";
-			status = nfs4_recall_slot(clp);
-			if (status < 0)
-				goto out_error;
-			continue;
-		}
-
 		/* First recover reboot state... */
 		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
 			section = "reclaim reboot";
-- 
1.7.11.7

[PATCH 20/20] NFSv4.1: CB_RECALL_SLOT must schedule a sequence op after updating targets

[Trond Myklebust]

RFC5661 requires us to make sure that the server knows we've updated
our slot table size by sending at least one SEQUENCE op containing the
new 'highest_slotid' value.
We can do so using the 'CHECK_LEASE' functionality of the state
manager.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
---
 fs/nfs/callback_proc.c | 1 +
 fs/nfs/nfs4_fs.h       | 1 +
 fs/nfs/nfs4state.c     | 7 +++++++
 3 files changed, 9 insertions(+)

diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index 8610bd1..f99faad 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -570,6 +570,7 @@ __be32 nfs4_callback_recallslot(struct cb_recallslotargs *args, void *dummy,
 	status = htonl(NFS4_OK);
 
 	nfs41_set_target_slotid(fc_tbl, args->crsa_target_highest_slotid);
+	nfs41_server_notify_target_slotid_update(cps->clp);
 out:
 	dprintk("%s: exit with status = %d\n", __func__, ntohl(status));
 	return status;
diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index fa1a055..0a109ec 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -334,6 +334,7 @@ struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp);
 int nfs41_discover_server_trunking(struct nfs_client *clp,
 			struct nfs_client **, struct rpc_cred *);
 extern void nfs4_schedule_session_recovery(struct nfs4_session *, int);
+extern void nfs41_server_notify_target_slotid_update(struct nfs_client *clp);
 #else
 static inline void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
 {
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 3940cd4..55a3e8c 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -1904,6 +1904,13 @@ void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
 }
 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
 
+void nfs41_server_notify_target_slotid_update(struct nfs_client *clp)
+{
+	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
+	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
+	nfs4_schedule_state_manager(clp);
+}
+
 static void nfs4_reset_all_state(struct nfs_client *clp)
 {
 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
-- 
1.7.11.7