[PATCH v2 0/8] allow for offlining/removing xprt via sysfs

[PATCH v2 0/8] allow for offlining/removing xprt via sysfs

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

In this patch series, I propose to add ability to offline a transport
so that it's no longer available for tasks with ability to either
bring it back online (if this transport recovers) or it can be removed
all together. When a transport is begin removed, all the moveable tasks
that have been assigned to this transport but didn't complete will be
retried on a different transport. Note, that if a transport has stuck,
un-moveable tasks, it can't be removed.

in v2 of this patch series:
-- able to online an offlined transport
-- for the getattr, to mark it moveable, use presense of session instead
of the minorversion
-- added things to the xprt_info displayed via sysfs: (1) whether or not
this is a main transport, in hopes to prevent to offline/remove a main
transport. (2) display source port of the TCP transport, as it makes it
easier to make transports to the network trace. (2) display xprt's
queuelen which keeps track of tasks that have been assigned to this
xprt.

Olga Kornievskaia (8):
  SUNRPC mark the first transport
  SUNRPC display xprt's main value in sysfs's xprt_info
  SUNRPC query transport's source port
  SUNRPC for TCP display xprt's source port in sysfs xprt_info
  SUNRPC: take a xprt offline using sysfs
  NFSv4.1 identify and mark RPC tasks that can move between transports
  sunrpc: display xprt's queuelen of assigned tasks via sysfs
  sunrpc: remove an offlined xprt using sysfs

 fs/nfs/nfs4proc.c               | 38 +++++++++++--
 fs/nfs/pagelist.c               |  8 ++-
 fs/nfs/write.c                  |  6 +-
 include/linux/sunrpc/sched.h    |  2 +
 include/linux/sunrpc/xprt.h     |  3 +
 include/linux/sunrpc/xprtsock.h |  1 +
 net/sunrpc/clnt.c               | 25 +++++++++
 net/sunrpc/sysfs.c              | 98 ++++++++++++++++++++++++++++++---
 net/sunrpc/sysfs.h              |  1 +
 net/sunrpc/xprtmultipath.c      |  6 +-
 net/sunrpc/xprtsock.c           |  7 +++
 11 files changed, 177 insertions(+), 18 deletions(-)

-- 
2.27.0

[PATCH v2 1/8] SUNRPC mark the first transport

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

When an RPC client gets created it's first transport is special
and should be marked a main transport.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
---
 include/linux/sunrpc/xprt.h | 1 +
 net/sunrpc/clnt.c           | 1 +
 2 files changed, 2 insertions(+)

diff --git a/include/linux/sunrpc/xprt.h b/include/linux/sunrpc/xprt.h
index 13a4eaf385cf..692e5946c029 100644
--- a/include/linux/sunrpc/xprt.h
+++ b/include/linux/sunrpc/xprt.h
@@ -293,6 +293,7 @@ struct rpc_xprt {
 	struct rcu_head		rcu;
 	const struct xprt_class	*xprt_class;
 	struct rpc_sysfs_xprt	*xprt_sysfs;
+	bool			main; /*mark if this is the 1st transport */
 };
 
 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 4d8fd9d9c264..7ca946567e13 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -412,6 +412,7 @@ static struct rpc_clnt * rpc_new_client(const struct rpc_create_args *args,
 	}
 
 	rpc_clnt_set_transport(clnt, xprt, timeout);
+	xprt->main = true;
 	xprt_iter_init(&clnt->cl_xpi, xps);
 	xprt_switch_put(xps);
 
-- 
2.27.0

[PATCH v2 2/8] SUNRPC display xprt's main value in sysfs's xprt_info

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

Display in sysfs in the information about the xprt if this is a
main transport or not.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
---
 net/sunrpc/sysfs.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index 45c0cf18360d..6ef5469fe998 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -103,10 +103,10 @@ static ssize_t rpc_sysfs_xprt_info_show(struct kobject *kobj,
 	ret = sprintf(buf, "last_used=%lu\ncur_cong=%lu\ncong_win=%lu\n"
 		       "max_num_slots=%u\nmin_num_slots=%u\nnum_reqs=%u\n"
 		       "binding_q_len=%u\nsending_q_len=%u\npending_q_len=%u\n"
-		       "backlog_q_len=%u\n", xprt->last_used, xprt->cong,
-		       xprt->cwnd, xprt->max_reqs, xprt->min_reqs,
+		       "backlog_q_len=%u\nmain_xprt=%d\n", xprt->last_used,
+		       xprt->cong, xprt->cwnd, xprt->max_reqs, xprt->min_reqs,
 		       xprt->num_reqs, xprt->binding.qlen, xprt->sending.qlen,
-		       xprt->pending.qlen, xprt->backlog.qlen);
+		       xprt->pending.qlen, xprt->backlog.qlen, xprt->main);
 	xprt_put(xprt);
 	return ret + 1;
 }
-- 
2.27.0

[PATCH v2 3/8] SUNRPC query transport's source port

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

Provide ability to query transport's source port.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
---
 include/linux/sunrpc/xprtsock.h | 1 +
 net/sunrpc/xprtsock.c           | 7 +++++++
 2 files changed, 8 insertions(+)

diff --git a/include/linux/sunrpc/xprtsock.h b/include/linux/sunrpc/xprtsock.h
index 3c1423ee74b4..8c2a712cb242 100644
--- a/include/linux/sunrpc/xprtsock.h
+++ b/include/linux/sunrpc/xprtsock.h
@@ -10,6 +10,7 @@
 
 int		init_socket_xprt(void);
 void		cleanup_socket_xprt(void);
+unsigned short	get_srcport(struct rpc_xprt *);
 
 #define RPC_MIN_RESVPORT	(1U)
 #define RPC_MAX_RESVPORT	(65535U)
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index 2ad4d0df45fe..4611845ec1eb 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -1653,6 +1653,13 @@ static int xs_get_srcport(struct sock_xprt *transport)
 	return port;
 }
 
+unsigned short get_srcport(struct rpc_xprt *xprt)
+{
+	struct sock_xprt *sock = container_of(xprt, struct sock_xprt, xprt);
+	return sock->srcport;
+}
+EXPORT_SYMBOL(get_srcport);
+
 static unsigned short xs_next_srcport(struct sock_xprt *transport, unsigned short port)
 {
 	if (transport->srcport != 0)
-- 
2.27.0

[PATCH v2 4/8] SUNRPC for TCP display xprt's source port in sysfs xprt_info

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

Using TCP connection's source port it is useful to match connections
seen on the network traces to the xprts used by the linux nfs client.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
---
 net/sunrpc/sysfs.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index 6ef5469fe998..816f543d4237 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -5,6 +5,7 @@
 #include <linux/sunrpc/clnt.h>
 #include <linux/kobject.h>
 #include <linux/sunrpc/addr.h>
+#include <linux/sunrpc/xprtsock.h>
 
 #include "sysfs.h"
 
@@ -103,10 +104,13 @@ static ssize_t rpc_sysfs_xprt_info_show(struct kobject *kobj,
 	ret = sprintf(buf, "last_used=%lu\ncur_cong=%lu\ncong_win=%lu\n"
 		       "max_num_slots=%u\nmin_num_slots=%u\nnum_reqs=%u\n"
 		       "binding_q_len=%u\nsending_q_len=%u\npending_q_len=%u\n"
-		       "backlog_q_len=%u\nmain_xprt=%d\n", xprt->last_used,
-		       xprt->cong, xprt->cwnd, xprt->max_reqs, xprt->min_reqs,
-		       xprt->num_reqs, xprt->binding.qlen, xprt->sending.qlen,
-		       xprt->pending.qlen, xprt->backlog.qlen, xprt->main);
+		       "backlog_q_len=%u\nmain_xprt=%d\nsrc_port=%u\n",
+		       xprt->last_used, xprt->cong, xprt->cwnd, xprt->max_reqs,
+		       xprt->min_reqs, xprt->num_reqs, xprt->binding.qlen,
+		       xprt->sending.qlen, xprt->pending.qlen,
+		       xprt->backlog.qlen, xprt->main,
+		       (xprt->xprt_class->ident == XPRT_TRANSPORT_TCP) ?
+		       get_srcport(xprt) : 0);
 	xprt_put(xprt);
 	return ret + 1;
 }
-- 
2.27.0

[PATCH v2 5/8] SUNRPC: take a xprt offline using sysfs

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

Using sysfs's xprt_state attribute, mark a particular transport offline.
It will not be picked during the round-robin selection. It's not allowed
to take the main (1st created transport associated with the rpc_client)
offline. Also bring a transport back online via sysfs by writing "online"
and that would allow for this transport to be picked during the round-
robin selection.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
---
 include/linux/sunrpc/xprt.h |  1 +
 net/sunrpc/sysfs.c          | 66 ++++++++++++++++++++++++++++++++++---
 net/sunrpc/sysfs.h          |  1 +
 net/sunrpc/xprtmultipath.c  |  6 ++--
 4 files changed, 68 insertions(+), 6 deletions(-)

diff --git a/include/linux/sunrpc/xprt.h b/include/linux/sunrpc/xprt.h
index 692e5946c029..b8ed7fa1b4ca 100644
--- a/include/linux/sunrpc/xprt.h
+++ b/include/linux/sunrpc/xprt.h
@@ -427,6 +427,7 @@ void			xprt_release_write(struct rpc_xprt *, struct rpc_task *);
 #define XPRT_BOUND		(4)
 #define XPRT_BINDING		(5)
 #define XPRT_CLOSING		(6)
+#define XPRT_OFFLINE		(7)
 #define XPRT_CONGESTED		(9)
 #define XPRT_CWND_WAIT		(10)
 #define XPRT_WRITE_SPACE	(11)
diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index 816f543d4237..e66888cc4c14 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -68,6 +68,15 @@ rpc_sysfs_xprt_kobj_get_xprt(struct kobject *kobj)
 	return xprt_get(x->xprt);
 }
 
+static inline struct rpc_xprt_switch *
+rpc_sysfs_xprt_kobj_get_xprt_switch(struct kobject *kobj)
+{
+	struct rpc_sysfs_xprt *x = container_of(kobj,
+		struct rpc_sysfs_xprt, kobject);
+
+	return xprt_switch_get(x->xprt_switch);
+}
+
 static inline struct rpc_xprt_switch *
 rpc_sysfs_xprt_switch_kobj_get_xprt(struct kobject *kobj)
 {
@@ -122,7 +131,7 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 	struct rpc_xprt *xprt = rpc_sysfs_xprt_kobj_get_xprt(kobj);
 	ssize_t ret;
 	int locked, connected, connecting, close_wait, bound, binding,
-	    closing, congested, cwnd_wait, write_space;
+	    closing, congested, cwnd_wait, write_space, offline;
 
 	if (!xprt)
 		return 0;
@@ -140,8 +149,9 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 		congested = test_bit(XPRT_CONGESTED, &xprt->state);
 		cwnd_wait = test_bit(XPRT_CWND_WAIT, &xprt->state);
 		write_space = test_bit(XPRT_WRITE_SPACE, &xprt->state);
+		offline = test_bit(XPRT_OFFLINE, &xprt->state);
 
-		ret = sprintf(buf, "state=%s %s %s %s %s %s %s %s %s %s\n",
+		ret = sprintf(buf, "state=%s %s %s %s %s %s %s %s %s %s %s\n",
 			      locked ? "LOCKED" : "",
 			      connected ? "CONNECTED" : "",
 			      connecting ? "CONNECTING" : "",
@@ -151,7 +161,8 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 			      closing ? "CLOSING" : "",
 			      congested ? "CONGESTED" : "",
 			      cwnd_wait ? "CWND_WAIT" : "",
-			      write_space ? "WRITE_SPACE" : "");
+			      write_space ? "WRITE_SPACE" : "",
+			      offline ? "OFFLINE" : "");
 	}
 
 	xprt_put(xprt);
@@ -235,6 +246,52 @@ static ssize_t rpc_sysfs_xprt_dstaddr_store(struct kobject *kobj,
 	goto out;
 }
 
+static ssize_t rpc_sysfs_xprt_state_change(struct kobject *kobj,
+					   struct kobj_attribute *attr,
+					   const char *buf, size_t count)
+{
+	struct rpc_xprt *xprt = rpc_sysfs_xprt_kobj_get_xprt(kobj);
+	int offline = 0, online = 0;
+	struct rpc_xprt_switch *xps = rpc_sysfs_xprt_kobj_get_xprt_switch(kobj);
+
+	if (!xprt)
+		return 0;
+
+	if (!strncmp(buf, "offline", 7))
+		offline = 1;
+	else if (!strncmp(buf, "online", 6))
+		online = 1;
+	else
+		return -EINVAL;
+
+	if (wait_on_bit_lock(&xprt->state, XPRT_LOCKED, TASK_KILLABLE)) {
+		count = -EINTR;
+		goto out_put;
+	}
+	if (xprt->main) {
+		count = -EINVAL;
+		goto release_tasks;
+	}
+	if (offline) {
+		set_bit(XPRT_OFFLINE, &xprt->state);
+		spin_lock(&xps->xps_lock);
+		xps->xps_nactive--;
+		spin_unlock(&xps->xps_lock);
+	} else if (online) {
+		clear_bit(XPRT_OFFLINE, &xprt->state);
+		spin_lock(&xps->xps_lock);
+		xps->xps_nactive++;
+		spin_unlock(&xps->xps_lock);
+	}
+
+release_tasks:
+	xprt_release_write(xprt, NULL);
+out_put:
+	xprt_put(xprt);
+	xprt_switch_put(xps);
+	return count;
+}
+
 int rpc_sysfs_init(void)
 {
 	rpc_sunrpc_kset = kset_create_and_add("sunrpc", NULL, kernel_kobj);
@@ -305,7 +362,7 @@ static struct kobj_attribute rpc_sysfs_xprt_info = __ATTR(xprt_info,
 	0444, rpc_sysfs_xprt_info_show, NULL);
 
 static struct kobj_attribute rpc_sysfs_xprt_change_state = __ATTR(xprt_state,
-	0644, rpc_sysfs_xprt_state_show, NULL);
+	0644, rpc_sysfs_xprt_state_show, rpc_sysfs_xprt_state_change);
 
 static struct attribute *rpc_sysfs_xprt_attrs[] = {
 	&rpc_sysfs_xprt_dstaddr.attr,
@@ -468,6 +525,7 @@ void rpc_sysfs_xprt_setup(struct rpc_xprt_switch *xprt_switch,
 	if (rpc_xprt) {
 		xprt->xprt_sysfs = rpc_xprt;
 		rpc_xprt->xprt = xprt;
+		rpc_xprt->xprt_switch = xprt_switch;
 		kobject_uevent(&rpc_xprt->kobject, KOBJ_ADD);
 	}
 }
diff --git a/net/sunrpc/sysfs.h b/net/sunrpc/sysfs.h
index ff10451de6fa..6620cebd1037 100644
--- a/net/sunrpc/sysfs.h
+++ b/net/sunrpc/sysfs.h
@@ -22,6 +22,7 @@ struct rpc_sysfs_xprt_switch {
 struct rpc_sysfs_xprt {
 	struct kobject kobject;
 	struct rpc_xprt *xprt;
+	struct rpc_xprt_switch *xprt_switch;
 };
 
 int rpc_sysfs_init(void);
diff --git a/net/sunrpc/xprtmultipath.c b/net/sunrpc/xprtmultipath.c
index 584349c8cad4..7d40cdf81274 100644
--- a/net/sunrpc/xprtmultipath.c
+++ b/net/sunrpc/xprtmultipath.c
@@ -65,7 +65,8 @@ static void xprt_switch_remove_xprt_locked(struct rpc_xprt_switch *xps,
 {
 	if (unlikely(xprt == NULL))
 		return;
-	xps->xps_nactive--;
+	if (!test_bit(XPRT_OFFLINE, &xprt->state))
+		xps->xps_nactive--;
 	xps->xps_nxprts--;
 	if (xps->xps_nxprts == 0)
 		xps->xps_net = NULL;
@@ -231,7 +232,8 @@ void xprt_iter_default_rewind(struct rpc_xprt_iter *xpi)
 static
 bool xprt_is_active(const struct rpc_xprt *xprt)
 {
-	return kref_read(&xprt->kref) != 0;
+	return (kref_read(&xprt->kref) != 0 &&
+		!test_bit(XPRT_OFFLINE, &xprt->state));
 }
 
 static
-- 
2.27.0

[PATCH v2 6/8] NFSv4.1 identify and mark RPC tasks that can move between transports

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

In preparation for when we can re-try a task on a different transport,
identify and mark such RPC tasks as moveable. Only 4.1+ operarations can
be re-tried on a different transport.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
---
 fs/nfs/nfs4proc.c            | 38 +++++++++++++++++++++++++++++++-----
 fs/nfs/pagelist.c            |  8 ++++++--
 fs/nfs/write.c               |  6 +++++-
 include/linux/sunrpc/sched.h |  2 ++
 4 files changed, 46 insertions(+), 8 deletions(-)

diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index e653654c10bc..d3ee3700c9dd 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -1155,7 +1155,11 @@ static int nfs4_call_sync_sequence(struct rpc_clnt *clnt,
 				   struct nfs4_sequence_args *args,
 				   struct nfs4_sequence_res *res)
 {
-	return nfs4_do_call_sync(clnt, server, msg, args, res, 0);
+	unsigned short task_flags = 0;
+
+	if (server->nfs_client->cl_minorversion)
+		task_flags = RPC_TASK_MOVEABLE;
+	return nfs4_do_call_sync(clnt, server, msg, args, res, task_flags);
 }
 
 
@@ -2569,6 +2573,9 @@ static int nfs4_run_open_task(struct nfs4_opendata *data,
 	};
 	int status;
 
+	if (server->nfs_client->cl_minorversion)
+		task_setup_data.flags |= RPC_TASK_MOVEABLE;
+
 	kref_get(&data->kref);
 	data->rpc_done = false;
 	data->rpc_status = 0;
@@ -3749,6 +3756,9 @@ int nfs4_do_close(struct nfs4_state *state, gfp_t gfp_mask, int wait)
 	};
 	int status = -ENOMEM;
 
+	if (server->nfs_client->cl_minorversion)
+		task_setup_data.flags |= RPC_TASK_MOVEABLE;
+
 	nfs4_state_protect(server->nfs_client, NFS_SP4_MACH_CRED_CLEANUP,
 		&task_setup_data.rpc_client, &msg);
 
@@ -4188,6 +4198,9 @@ static int _nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle,
 	};
 	unsigned short task_flags = 0;
 
+	if (nfs4_has_session(server->nfs_client))
+		task_flags = RPC_TASK_MOVEABLE;
+
 	/* Is this is an attribute revalidation, subject to softreval? */
 	if (inode && (server->flags & NFS_MOUNT_SOFTREVAL))
 		task_flags |= RPC_TASK_TIMEOUT;
@@ -4307,6 +4320,9 @@ static int _nfs4_proc_lookup(struct rpc_clnt *clnt, struct inode *dir,
 	};
 	unsigned short task_flags = 0;
 
+	if (server->nfs_client->cl_minorversion)
+		task_flags = RPC_TASK_MOVEABLE;
+
 	/* Is this is an attribute revalidation, subject to softreval? */
 	if (nfs_lookup_is_soft_revalidate(dentry))
 		task_flags |= RPC_TASK_TIMEOUT;
@@ -6538,7 +6554,7 @@ static int _nfs4_proc_delegreturn(struct inode *inode, const struct cred *cred,
 		.rpc_client = server->client,
 		.rpc_message = &msg,
 		.callback_ops = &nfs4_delegreturn_ops,
-		.flags = RPC_TASK_ASYNC | RPC_TASK_TIMEOUT,
+		.flags = RPC_TASK_ASYNC | RPC_TASK_TIMEOUT | RPC_TASK_MOVEABLE,
 	};
 	int status = 0;
 
@@ -6856,6 +6872,11 @@ static struct rpc_task *nfs4_do_unlck(struct file_lock *fl,
 		.workqueue = nfsiod_workqueue,
 		.flags = RPC_TASK_ASYNC,
 	};
+	struct nfs_client *client =
+		NFS_SERVER(lsp->ls_state->inode)->nfs_client;
+
+	if (client->cl_minorversion)
+		task_setup_data.flags |= RPC_TASK_MOVEABLE;
 
 	nfs4_state_protect(NFS_SERVER(lsp->ls_state->inode)->nfs_client,
 		NFS_SP4_MACH_CRED_CLEANUP, &task_setup_data.rpc_client, &msg);
@@ -7130,6 +7151,10 @@ static int _nfs4_do_setlk(struct nfs4_state *state, int cmd, struct file_lock *f
 		.flags = RPC_TASK_ASYNC | RPC_TASK_CRED_NOREF,
 	};
 	int ret;
+	struct nfs_client *client = NFS_SERVER(state->inode)->nfs_client;
+
+	if (client->cl_minorversion)
+		task_setup_data.flags |= RPC_TASK_MOVEABLE;
 
 	dprintk("%s: begin!\n", __func__);
 	data = nfs4_alloc_lockdata(fl, nfs_file_open_context(fl->fl_file),
@@ -9186,7 +9211,7 @@ static struct rpc_task *_nfs41_proc_sequence(struct nfs_client *clp,
 		.rpc_client = clp->cl_rpcclient,
 		.rpc_message = &msg,
 		.callback_ops = &nfs41_sequence_ops,
-		.flags = RPC_TASK_ASYNC | RPC_TASK_TIMEOUT,
+		.flags = RPC_TASK_ASYNC | RPC_TASK_TIMEOUT | RPC_TASK_MOVEABLE,
 	};
 	struct rpc_task *ret;
 
@@ -9509,7 +9534,8 @@ nfs4_proc_layoutget(struct nfs4_layoutget *lgp, long *timeout)
 		.rpc_message = &msg,
 		.callback_ops = &nfs4_layoutget_call_ops,
 		.callback_data = lgp,
-		.flags = RPC_TASK_ASYNC | RPC_TASK_CRED_NOREF,
+		.flags = RPC_TASK_ASYNC | RPC_TASK_CRED_NOREF |
+			 RPC_TASK_MOVEABLE,
 	};
 	struct pnfs_layout_segment *lseg = NULL;
 	struct nfs4_exception exception = {
@@ -9650,6 +9676,7 @@ int nfs4_proc_layoutreturn(struct nfs4_layoutreturn *lrp, bool sync)
 		.rpc_message = &msg,
 		.callback_ops = &nfs4_layoutreturn_call_ops,
 		.callback_data = lrp,
+		.flags = RPC_TASK_MOVEABLE,
 	};
 	int status = 0;
 
@@ -9804,6 +9831,7 @@ nfs4_proc_layoutcommit(struct nfs4_layoutcommit_data *data, bool sync)
 		.rpc_message = &msg,
 		.callback_ops = &nfs4_layoutcommit_ops,
 		.callback_data = data,
+		.flags = RPC_TASK_MOVEABLE,
 	};
 	struct rpc_task *task;
 	int status = 0;
@@ -10131,7 +10159,7 @@ static int nfs41_free_stateid(struct nfs_server *server,
 		.rpc_client = server->client,
 		.rpc_message = &msg,
 		.callback_ops = &nfs41_free_stateid_ops,
-		.flags = RPC_TASK_ASYNC,
+		.flags = RPC_TASK_ASYNC | RPC_TASK_MOVEABLE,
 	};
 	struct nfs_free_stateid_data *data;
 	struct rpc_task *task;
diff --git a/fs/nfs/pagelist.c b/fs/nfs/pagelist.c
index cf9cc62ec48e..cc232d1f16f2 100644
--- a/fs/nfs/pagelist.c
+++ b/fs/nfs/pagelist.c
@@ -954,6 +954,7 @@ static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
 {
 	struct nfs_pgio_header *hdr;
 	int ret;
+	unsigned short task_flags = 0;
 
 	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
 	if (!hdr) {
@@ -962,14 +963,17 @@ static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
 	}
 	nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
 	ret = nfs_generic_pgio(desc, hdr);
-	if (ret == 0)
+	if (ret == 0) {
+		if (NFS_SERVER(hdr->inode)->nfs_client->cl_minorversion)
+			task_flags = RPC_TASK_MOVEABLE;
 		ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
 					hdr,
 					hdr->cred,
 					NFS_PROTO(hdr->inode),
 					desc->pg_rpc_callops,
 					desc->pg_ioflags,
-					RPC_TASK_CRED_NOREF);
+					RPC_TASK_CRED_NOREF | task_flags);
+	}
 	return ret;
 }
 
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 3bf82178166a..eae9bf114041 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -1810,6 +1810,7 @@ nfs_commit_list(struct inode *inode, struct list_head *head, int how,
 		struct nfs_commit_info *cinfo)
 {
 	struct nfs_commit_data	*data;
+	unsigned short task_flags = 0;
 
 	/* another commit raced with us */
 	if (list_empty(head))
@@ -1820,8 +1821,11 @@ nfs_commit_list(struct inode *inode, struct list_head *head, int how,
 	/* Set up the argument struct */
 	nfs_init_commit(data, head, NULL, cinfo);
 	atomic_inc(&cinfo->mds->rpcs_out);
+	if (NFS_SERVER(inode)->nfs_client->cl_minorversion)
+		task_flags = RPC_TASK_MOVEABLE;
 	return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
-				   data->mds_ops, how, RPC_TASK_CRED_NOREF);
+				   data->mds_ops, how,
+				   RPC_TASK_CRED_NOREF | task_flags);
 }
 
 /*
diff --git a/include/linux/sunrpc/sched.h b/include/linux/sunrpc/sched.h
index df696efdd675..a237b8dbf608 100644
--- a/include/linux/sunrpc/sched.h
+++ b/include/linux/sunrpc/sched.h
@@ -121,6 +121,7 @@ struct rpc_task_setup {
  */
 #define RPC_TASK_ASYNC		0x0001		/* is an async task */
 #define RPC_TASK_SWAPPER	0x0002		/* is swapping in/out */
+#define RPC_TASK_MOVEABLE	0x0004		/* nfs4.1+ rpc tasks */
 #define RPC_TASK_NULLCREDS	0x0010		/* Use AUTH_NULL credential */
 #define RPC_CALL_MAJORSEEN	0x0020		/* major timeout seen */
 #define RPC_TASK_ROOTCREDS	0x0040		/* force root creds */
@@ -139,6 +140,7 @@ struct rpc_task_setup {
 #define RPC_IS_SOFT(t)		((t)->tk_flags & (RPC_TASK_SOFT|RPC_TASK_TIMEOUT))
 #define RPC_IS_SOFTCONN(t)	((t)->tk_flags & RPC_TASK_SOFTCONN)
 #define RPC_WAS_SENT(t)		((t)->tk_flags & RPC_TASK_SENT)
+#define RPC_IS_MOVEABLE(t)	((t)->tk_flags & RPC_TASK_MOVEABLE)
 
 #define RPC_TASK_RUNNING	0
 #define RPC_TASK_QUEUED		1
-- 
2.27.0

[PATCH v2 7/8] sunrpc: display xprt's queuelen of assigned tasks via sysfs

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

Once a task grabs a trasnport it's reflected in the queuelen of
the rpc_xprt structure. Add display of that value in the xprt's
info file in sysfs.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
---
 net/sunrpc/sysfs.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index e66888cc4c14..754bd010ea29 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -113,13 +113,15 @@ static ssize_t rpc_sysfs_xprt_info_show(struct kobject *kobj,
 	ret = sprintf(buf, "last_used=%lu\ncur_cong=%lu\ncong_win=%lu\n"
 		       "max_num_slots=%u\nmin_num_slots=%u\nnum_reqs=%u\n"
 		       "binding_q_len=%u\nsending_q_len=%u\npending_q_len=%u\n"
-		       "backlog_q_len=%u\nmain_xprt=%d\nsrc_port=%u\n",
+		       "backlog_q_len=%u\nmain_xprt=%d\nsrc_port=%u\n"
+		       "tasks_queuelen=%ld\n",
 		       xprt->last_used, xprt->cong, xprt->cwnd, xprt->max_reqs,
 		       xprt->min_reqs, xprt->num_reqs, xprt->binding.qlen,
 		       xprt->sending.qlen, xprt->pending.qlen,
 		       xprt->backlog.qlen, xprt->main,
 		       (xprt->xprt_class->ident == XPRT_TRANSPORT_TCP) ?
-		       get_srcport(xprt) : 0);
+		       get_srcport(xprt) : 0,
+		       atomic_long_read(&xprt->queuelen));
 	xprt_put(xprt);
 	return ret + 1;
 }
-- 
2.27.0

[PATCH v2 8/8] sunrpc: remove an offlined xprt using sysfs

[Olga Kornievskaia]

From: Olga Kornievskaia <kolga@netapp.com>

Once a transport has been put offline, this transport can be also
removed from the list of transports. Any tasks that have been stuck
on this transport would find the next available active transport
and be re-tried. This transport would be removed from the xprt_switch
list and freed.

Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
---
 include/linux/sunrpc/xprt.h |  1 +
 net/sunrpc/clnt.c           | 24 ++++++++++++++++++++++++
 net/sunrpc/sysfs.c          | 26 ++++++++++++++++++++++----
 3 files changed, 47 insertions(+), 4 deletions(-)

diff --git a/include/linux/sunrpc/xprt.h b/include/linux/sunrpc/xprt.h
index b8ed7fa1b4ca..c8c39f22d3b1 100644
--- a/include/linux/sunrpc/xprt.h
+++ b/include/linux/sunrpc/xprt.h
@@ -428,6 +428,7 @@ void			xprt_release_write(struct rpc_xprt *, struct rpc_task *);
 #define XPRT_BINDING		(5)
 #define XPRT_CLOSING		(6)
 #define XPRT_OFFLINE		(7)
+#define XPRT_REMOVE		(8)
 #define XPRT_CONGESTED		(9)
 #define XPRT_CWND_WAIT		(10)
 #define XPRT_WRITE_SPACE	(11)
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 7ca946567e13..f8ec29b06d0d 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -2106,6 +2106,30 @@ call_connect_status(struct rpc_task *task)
 	case -ENOTCONN:
 	case -EAGAIN:
 	case -ETIMEDOUT:
+		if (!(task->tk_flags & RPC_TASK_NO_ROUND_ROBIN) &&
+		    (task->tk_flags & RPC_TASK_MOVEABLE) &&
+		    test_bit(XPRT_REMOVE, &xprt->state)) {
+			struct rpc_xprt *saved = task->tk_xprt;
+			struct rpc_xprt_switch *xps;
+
+			rcu_read_lock();
+			xps = xprt_switch_get(rcu_dereference(clnt->cl_xpi.xpi_xpswitch));
+			rcu_read_unlock();
+			if (xps->xps_nxprts > 1) {
+				long value;
+
+				xprt_release(task);
+				value = atomic_long_dec_return(&xprt->queuelen);
+				if (value == 0)
+					rpc_xprt_switch_remove_xprt(xps, saved);
+				xprt_put(saved);
+				task->tk_xprt = NULL;
+				task->tk_action = call_start;
+			}
+			xprt_switch_put(xps);
+			if (!task->tk_xprt)
+				return;
+		}
 		goto out_retry;
 	case -ENOBUFS:
 		rpc_delay(task, HZ >> 2);
diff --git a/net/sunrpc/sysfs.c b/net/sunrpc/sysfs.c
index 754bd010ea29..34b60db5321f 100644
--- a/net/sunrpc/sysfs.c
+++ b/net/sunrpc/sysfs.c
@@ -133,7 +133,7 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 	struct rpc_xprt *xprt = rpc_sysfs_xprt_kobj_get_xprt(kobj);
 	ssize_t ret;
 	int locked, connected, connecting, close_wait, bound, binding,
-	    closing, congested, cwnd_wait, write_space, offline;
+	    closing, congested, cwnd_wait, write_space, offline, remove;
 
 	if (!xprt)
 		return 0;
@@ -152,8 +152,9 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 		cwnd_wait = test_bit(XPRT_CWND_WAIT, &xprt->state);
 		write_space = test_bit(XPRT_WRITE_SPACE, &xprt->state);
 		offline = test_bit(XPRT_OFFLINE, &xprt->state);
+		remove = test_bit(XPRT_REMOVE, &xprt->state);
 
-		ret = sprintf(buf, "state=%s %s %s %s %s %s %s %s %s %s %s\n",
+		ret = sprintf(buf, "state=%s %s %s %s %s %s %s %s %s %s %s %s\n",
 			      locked ? "LOCKED" : "",
 			      connected ? "CONNECTED" : "",
 			      connecting ? "CONNECTING" : "",
@@ -164,7 +165,8 @@ static ssize_t rpc_sysfs_xprt_state_show(struct kobject *kobj,
 			      congested ? "CONGESTED" : "",
 			      cwnd_wait ? "CWND_WAIT" : "",
 			      write_space ? "WRITE_SPACE" : "",
-			      offline ? "OFFLINE" : "");
+			      offline ? "OFFLINE" : "",
+			      remove ? "REMOVE" : "");
 	}
 
 	xprt_put(xprt);
@@ -253,7 +255,7 @@ static ssize_t rpc_sysfs_xprt_state_change(struct kobject *kobj,
 					   const char *buf, size_t count)
 {
 	struct rpc_xprt *xprt = rpc_sysfs_xprt_kobj_get_xprt(kobj);
-	int offline = 0, online = 0;
+	int offline = 0, online = 0, remove = 0;
 	struct rpc_xprt_switch *xps = rpc_sysfs_xprt_kobj_get_xprt_switch(kobj);
 
 	if (!xprt)
@@ -263,6 +265,8 @@ static ssize_t rpc_sysfs_xprt_state_change(struct kobject *kobj,
 		offline = 1;
 	else if (!strncmp(buf, "online", 6))
 		online = 1;
+	else if (!strncmp(buf, "remove", 6))
+		remove = 1;
 	else
 		return -EINVAL;
 
@@ -284,6 +288,20 @@ static ssize_t rpc_sysfs_xprt_state_change(struct kobject *kobj,
 		spin_lock(&xps->xps_lock);
 		xps->xps_nactive++;
 		spin_unlock(&xps->xps_lock);
+	} else if (remove) {
+		if (test_bit(XPRT_OFFLINE, &xprt->state)) {
+			set_bit(XPRT_REMOVE, &xprt->state);
+			xprt_force_disconnect(xprt);
+			if (test_bit(XPRT_CONNECTED, &xprt->state)) {
+				if (!xprt->sending.qlen &&
+				    !xprt->pending.qlen &&
+				    !xprt->backlog.qlen &&
+				    !atomic_long_read(&xprt->queuelen))
+					rpc_xprt_switch_remove_xprt(xps, xprt);
+			}
+		} else {
+			count = -EINVAL;
+		}
 	}
 
 release_tasks:
-- 
2.27.0